Conjecture Atlas

Proof source

Relative Bedford–Taylor theory on Berkovich spaces Fourth research record: original-test functoriality, faithful energy forms, capacities, and limits of relative solvability

Complete proof text in the unsigned reading edition.

Download sourceBrowse research records

Reading edition of the complete source note. Original labels and explicit equation tags are retained; references use source titles where compiled numbering is unavailable. Current result status is recorded separately in the project register.

Contents of this source
  1. Part I. Frozen R1–R3 mathematical dependencies
  2. Part II. Fourth-round proofs and counterexamples

Abstract

This continuation retains the arbitrary fixed-background bounded Bedford–Taylor calculus proved in the preceding records. It develops uniform approximation of tangentially compatible original piecewise CLD tests, reconstructs finite etale traces from finitely many Newton polygons, and proves preservation of the bounded potential curvature kernel under disc projections and arbitrary analytic morphisms over the prescribed original field. Exact spreading of original tests from algebraic parameter fibers, combined with the known full Gauss-fiber formula, further gives scalar kernel transfer to finite extensions of finite polyradius Gauss residue fields. A separate exact finite-coefficient descent reduces the remaining scalar question to residue fields of finite-dimensional algebraic parameter points, without solving those general fibers. In remaining degree one, a bounded continuous psh function provides a faithful curvature measure with exactly the coefficient-null ideal of the background. The associated energy form is regular, strongly local and coercive. Its Poisson problem has unique energy solutions; its capacity controls relative BT outer capacity in one direction, and complete pole sets give an orthogonal form decomposition. In every remaining degree, complete pole carriers yield qualified capacity additivity, while finite tester pasting and a constrained variational retraction extend the relative theory. A smooth nonpolar example shows that a bounded classical subsolution does not suffice for solvability in the global bounded-row psh completion; an explicit non-psh energy solution isolates the obstruction. General potential quotients, arbitrary-background raw capacity regularity, and unrestricted completed-psh Dirichlet existence remain open or false in the precise scopes recorded here.

Scope and reading guide

The current comparison across all eleven Bedford–Taylor directions is Section “Fourth-round correspondence and current proof boundaries”. The exact original-literature theorem comparison remains Section “Theorem-by-theorem comparison with the classical theory”. Part I retains the R1–R3 mathematical sources verbatim, with their original section numbering. Historical open statements are superseded only by the qualified later theorem explicitly identified in Part II. The early projection and analytic-graph gates recorded within the R4 proof-development sections are resolved, in their stated scopes, by Sections “Preservation of the bounded potential kernel by disc projection” and “Functoriality of the bounded potential kernel over the original field”. Section “Gauss residue fields: full fibers and exact test spreading” then treats the stated Gauss extensions; earlier narrower scope paragraphs are retained as part of the dependency record.

For the new proofs, start with Section “Functoriality of the bounded potential kernel over the original field” for the same-field analytic kernel theorem, Section “A faithful curvature measure and a coercive relative energy form” for the faithful reference measure and linear energy theory, Section “A one-way comparison with the regular energy capacity” for the one-way capacity comparison, and Section “Fourth-round correspondence and current proof boundaries” for their limitations. The smooth nonpolar obstruction and its radial energy solution have separate full proofs in Part II. Original compact smooth CLD tests and the original absolute value are retained throughout.

R and B are authorized internal inputs in their actual hypotheses; they have not been subjected to a new comprehensive proof audit. A2’s same-model geometric adelic nefness theorem is already complete under its stated global assumptions. An arbitrary positive separately closed background is allowed in the fixed-background results. The faithful Hilbert energy construction has remaining degree one; it applies in higher degree only after fixing the other mixed slots. Every boundary or global defining-row hypothesis is part of its individual theorem.

The provenance labels INTERNAL_INPUT, EXTERNAL_KNOWN, DIRECT_COROLLARY, and NEW_PROVED are distinct from the logical statuses PROVED, CONDITIONAL, OPEN, and FALSE. A failed argument is labeled FAILED_ATTEMPT. Here NEW_PROVED means a complete argument supplied within this project relative to its declared inputs; it does not assert worldwide priority, external refereeing, or formal verification. Independent internal readings and exact source versions are recorded in the accompanying archive.

No general pullback of a current, wedge of two arbitrary positive currents, automatic extension of kernel identities beyond the specified finite and Gauss field extensions, or equality on all GK δ\delta-tests is asserted. Raw BT capacity, its open outer regularization, and energy capacity are kept distinct. Energy representatives and global bounded psh defining rows are different solution classes.

The archive records elapsed working intervals and the complete continuation state. Parallel branches are not added to elapsed time; pure proof time was not separately instrumented. This record makes no claim that its present continuation lasted eight hours.

Part I. Frozen R1–R3 mathematical dependencies

Scope, inputs, and conventions

Fix a finite extension K/QpK/\mathbf Q_p with its given absolute value. Let XX be a good, Hausdorff, boundaryless, quasi-smooth KK-analytic space of pure dimension n≥1n\geq1, and let U⊂XU\subset X be an ordinary open set. We use only the original Chambert–Loir–Ducros sheaf Aa,bA^{a,b}, its compactly supported smooth tests, and its original calibration. We assume 0≤p≤n0\leq p\leq n. A current of bidegree (p,p)(p,p) acts on Acn−p,n−pA_c^{n-p,n-p}. We write D=d′d′′D=d'd''.

The class Cb(U)\mathcal C_b(U) consists of the actual real-valued, locally bounded, upper-semicontinuous classical psh functions of B, Definition 2.2: pullback to every analytic curve after every complete isometric extension is subharmonic. No function is changed on a null or polar set. A norm in this report always means the actual pointwise bound ∥u∥V,∞=sup⁡x∈V∣u(x)∣,\|u\|_{V,\infty}=\sup_{x\in V}|u(x)|, even where the traditional notation L∞(V)L^\infty(V) is used. There is no unspecified reference volume measure behind this notation.

The accepted internal results

The user has authorized the following theorems as established inputs. They have been checked for scope and dependencies, but their proofs have not been re-audited in this project.

R.

Continuous CLD-positive potentials and adelic mixed measures, 10 November 2026, 19 pages. Theorem 2.9 gives local uniform smooth positive approximation of finite continuous original-current-positive potentials on smooth quasi-projective algebraic analytifications over a pp-adic completion. Lemma 3.1 gives finite coefficient Radon control for the original positive cone and stability under a fixed continuous multiplier. Lemma 3.2 smooths pure Fubini–Study metrics on their original line. Proposition A.1 gives regularized maxima with exact inactive branches. Theorem 2.10 identifies positivity of the prescribed Yuan–Zhang integrable adelic mixed measure under its stated hypotheses.

B.

Locally bounded classical plurisubharmonic functions and mixed currents, 10 November 2026, 32 pages. Theorems 2.5–2.7 identify Cb\mathcal C_b with the local decreasing PL class, and construct strongly positive mixed currents and weighted joint decreasing continuity with background [1][1]. Definition 2.4 and Lemma 7.2 specify actual-space coefficient Radon measures and Borel multiplication. Theorem 4.4, Lemma D.2, and Proposition 7.1 provide the common projective line, frame, ordinary neighborhood, and decreasing pure Fubini–Study approximants for a finite family of decreasing sequences. Lemma D.4 supplies ordinary algebraic neighborhoods without changing KK. These common-domain inputs are needed in Lemma “A common domain for smooth majorants”; separate local regularizations with shrinking domains would not suffice.

A2.

A non-Archimedean criterion for geometric adelic nefness, revised 12 October 2026, 31 pages. Theorem 1.1 is the completed same-model geometric nefness theorem under its number-field, integrable adelic line, and full Cartier-boundary hypotheses. It preserves the open line and its prescribed identifications. This is an accepted global application, not an open problem in the present project. It is not used to prove the local relative BT theorems below.

Any use of the broader analytic setting passes through the local algebraic neighborhoods already supplied by B. A finite extension of Qp\mathbf Q_p can be realized as a number-field completion; the use of R does not change the prescribed absolute value. No argument below makes a new base change or renormalizes integration.

Three positivity cones

On a finite original moment presentation, strongly positive forms are generated by decomposable squares. Positive forms are represented by positive semidefinite symmetric coefficient tensors. Weakly positive forms are nonnegative on decomposable vectors. These cones need not coincide in middle degree. A positive current in this report is nonnegative on positive complementary original forms, as in R, Convention 2.2. A strongly positive current is nonnegative on weakly positive complementary forms, as in B, Convention 2.3. Thus the arbitrary background considered here is not assumed strongly positive.

We use the finite-matrix domination from R, not a decomposition of every symmetric test into decomposable squares. The latter would lose a middle-degree annihilator unless one had strong positivity and used the separate argument in B, Lemma 5.1. Wedge with a positive (1,1)(1,1)-form preserves the original positive cone. More generally, wedge with a strongly positive form preserves it; these are the only cone-preserving wedge operations used in the proofs.

Meaning of the result labels
Label Meaning in this record
EXTERNAL_KNOWN A verified statement of a cited external source.
INTERNAL_INPUT A theorem accepted from the supplied manuscripts.
DIRECT_COROLLARY A consequence obtained directly from named accepted inputs or preceding proved results.
NEW_PROVED A statement with a complete proof in this record, relative to the adopted inputs; this does not claim bibliographic priority or external peer review.
CONDITIONAL A stated implication whose extra hypothesis is not established here.
OPEN An unresolved assertion in the declared scope.
FALSE An assertion ruled out by a specified counterexample.
FAILED_ATTEMPT A proof route with an identified invalid or missing step.

The separate analogue-status table uses PROVED, CONDITIONAL, OPEN, and FALSE. Here PROVED includes the user’s accepted internal inputs and proofs derived from them. It does not mean that the input manuscripts have been independently validated by this research run.

What is not being constructed

The product has an arbitrary fixed positive closed background and bounded classical psh potentials. It is not a wedge product of two entirely arbitrary positive currents. It is not the full GK δ\delta-form algebra, and its current identities do not quantify over all δ\delta-test forms. Later applications to Green forms or Beilinson–Bloch heights do not enter the proof of the present calculus.

Ordered products and relative finite-test estimates

Throughout this section, K/QpK/\mathbf Q_p is finite with its prescribed absolute value. The space XX is good, Hausdorff, boundaryless and quasi-smooth, of pure dimension nn. All open sets are ordinary open sets, and all forms are original CLD forms. We write D=d′d′′D=d'd''. The word positive means dual positivity for the original positive form cone used in R, Convention 2.2. It does not mean strong positivity. The background SS is a symmetric positive (p,p)(p,p)-current satisfying d′S=d′′S=0d'S=d''S=0. All L∞L^\infty bounds below mean actual pointwise suprema on the stated sets; no ambient measure is used to discard exceptional values.

The precise order-zero input

Lemma (Positive controls on a finite test stage). Let WW be an ordinary open set, let L⊂WL\subset W be compact, and fix a finite original moment presentation of a smooth (q,q)(q,q)-form η\eta near LL. After replacing η\eta by a compactly supported extension agreeing with it near LL, there are positive forms Θ1,…,ΘN∈Acq,q(W)\Theta_1,\ldots,\Theta_N\in A_c^{q,q}(W) and a constant CηC_\eta such that every positive (n−q,n−q)(n-q,n-q)-current TT satisfies

∣μT,η∣(L)≤Cη∑a=1N⟨T,Θa⟩.|\mu_{T,\eta}|(L) \leq C_\eta\sum_{a=1}^N\langle T,\Theta_a\rangle.

The same controls work for all TT. For forms on one fixed finite presentation, with supports in one fixed compact set, CηC_\eta is bounded by a constant times the supremum norm of the coefficient matrices.

Proof. This is the local domination argument of R, Lemma 3.1. We recall the details that matter for subsequent estimates. A finite smooth partition of unity reduces the assertion to one moment presentation with real coordinates t1,…,trt_1,\ldots,t_r. Put β=∑a=1rd′ta∧d′′ta,Ωq=βqq!,\beta=\sum_{a=1}^r d't_a\wedge d''t_a, \qquad \Omega_q=\frac{\beta^q}{q!}, with Ω0=1\Omega_0=1. Choose a nonnegative compact smooth cutoff ρ\rho, equal to one near the relevant support. The symmetric coefficient matrix of Ωq\Omega_q is positive definite. Hence, with a constant CηC_\eta controlled by the coefficient supremum norm, Cη∥h∥∞ρΩq±hηsym≥0C_\eta\|h\|_\infty\rho\Omega_q\pm h\eta_{\rm sym}\geq0 for every compact smooth scalar hh. Antisymmetric forms and their negatives belong to the null part of this positivity convention; a positive current annihilates them. Therefore ∣⟨T,hη⟩∣≤Cη∥h∥∞⟨T,ρΩq⟩.|\langle T,h\eta\rangle| \leq C_\eta\|h\|_\infty\langle T,\rho\Omega_q\rangle. If hh uses further moment coordinates, pull the displayed cone inequalities to a common refinement. Positivity is preserved, so the constant does not depend on those additional scalar coordinates.

Original smooth scalars are uniformly dense on compact sets. Riesz representation thus gives the coefficient Radon measure and the stated total-variation bound. Summing over the finite partition gives (eq:finite-positive-controls). The local algebraic neighborhoods in B, Proposition 7.1 and Lemma D.4, let us apply R’s argument on the analytic spaces considered here. ◻

Status and dependence.

Lemma “Positive controls on a finite test stage” is DIRECT_COROLLARY, from R, Lemma 3.1 and B’s local algebraic transport. In particular it must not be presented as a new positivity theorem. Strong positivity, and B’s decomposable-square annihilator in middle degree, are not used in this argument.

The scalar coefficient relations hold as identities of Radon measures: test each relation after multiplication by every original smooth scalar and use uniqueness of the representing measure. Consequently, for every locally bounded Borel function aa, ⟨aT,η⟩=∫a dμT,η\langle aT,\eta\rangle=\int a\,d\mu_{T,\eta} defines an intrinsic coefficient-Radon current. It commutes with restriction and with common refinements. The proof is the measure uniqueness argument of B, Lemma 7.2, applied to the original-positive coefficient measures supplied by R. It does not replace aa by an almost-everywhere representative chosen using another measure.

One multiplication and ordered iteration

Proposition (One bounded potential over an arbitrary background). For every u∈Cb(W)u\in\mathcal C_b(W) the current D(uS)D(uS) is symmetric, positive, coefficient Radon, and separately d′d'- and d′′d''-closed. If aja_j are locally uniformly bounded Borel functions converging pointwise to aa, then ajS→aSa_jS\to aS on every original compact test. In particular, D(ujS)⟶D(uS)D(u_jS)\longrightarrow D(uS) whenever uj→uu_j\to u pointwise and the uju_j have a common local bound. In the last assertion the uju_j need not be monotone.

Proof. The convergence assertions are dominated convergence for the fixed finite coefficient measures of SS, followed by continuity of distributional differentiation. For a continuous classical psh function vv, the accepted local approximation results of R and B give smooth psh functions vm→vv_m\to v locally uniformly on a fixed local neighborhood. The identity D(vmS)=Dvm∧SD(v_mS)=Dv_m\wedge S and preservation of the original positive cone under wedge with a positive (1,1)(1,1)-form show positivity. Passing to the limit proves positivity of D(vS)D(vS). For general uu, use B, Theorem 2.5 to obtain continuous semipositive PL functions uj↓uu_j\downarrow u on a fixed local neighborhood. They are uniformly bounded on each compact subdomain by their first term and the locally bounded limit. Dominated convergence for the fixed coefficient measures of SS gives D(ujS)→D(uS)D(u_jS)\to D(uS). The cone of positive currents is weakly closed.

Closedness follows from d′D=d′′D=0d'D=d''D=0, and symmetry follows from the CLD involution. Lemma “Positive controls on a finite test stage” supplies the coefficient-Radon property. A finite localization proves the assertions on WW. ◻

Corollary (Canonical ordered products). For 0≤k≤n−p0\leq k\leq n-p, the recursion TS,0=S,TS,k(u1,…,uk)=D(u1TS,k−1(u2,…,uk))T_{S,0}=S,\qquad T_{S,k}(u_1,\ldots,u_k) =D\bigl(u_1T_{S,k-1}(u_2,\ldots,u_k)\bigr) defines a positive, symmetric-as-a-current, coefficient-Radon, separately closed current for every ordered tuple ui∈Cb(W)u_i\in\mathcal C_b(W). The recursion is intrinsic, multilinear on the cone, positively homogeneous, unchanged by adding a constant in a differentiated slot, and compatible with ordinary restriction and gluing.

Proof. Apply Proposition “One bounded potential over an arbitrary background” inductively. Borel multiplication and distributional differentiation are intrinsic and local. Their linearity proves additivity and positive homogeneity by induction. A constant in the outer slot contributes D(cT)=0D(cT)=0 because the inner current is separately closed; the assertion for other slots follows inductively. On overlaps, equal coefficient currents have equal Borel multiples, so the recursion glues. ◻

Both preceding statements are DIRECT_COROLLARY. At this point “symmetric-as-a-current” refers only to the CLD involution. It does not assert invariance under permutations of two discontinuous differentiated potentials. Similarly, intrinsic recursion does not yet assert convergence along every simultaneous regularization of all slots. These are separate analytical assertions.

Lemma (Continuous slots). Let u∈Cb(W)u\in\mathcal C_b(W) and let vv be continuous and classical psh. Then D(uD(vS))=D(vD(uS)).D\bigl(uD(vS)\bigr)=D\bigl(vD(uS)\bigr).

Proof. First take both potentials continuous. Smooth local regularization, R, Lemma 3.1, and the coefficient mass bounds show that joint locally uniform convergence of continuous potentials passes through the recursion. Explicitly, if Qj→QQ_j\to Q are positive currents and aj→aa_j\to a locally uniformly, with aa continuous, then ajQj−aQ=(aj−a)Qj+a(Qj−Q)⟶0.a_jQ_j-aQ=(a_j-a)Q_j+a(Q_j-Q)\longrightarrow0. The first term is controlled by the locally uniform coefficient mass bound, and the second by R, Lemma 3.1. Smooth commutativity therefore passes to continuous potentials over the fixed arbitrary background.

Choose continuous psh uj↓uu_j\downarrow u on one fixed local neighborhood. The left side of D(ujD(vS))=D(vD(ujS))D\bigl(u_jD(vS)\bigr)=D\bigl(vD(u_jS)\bigr) converges by dominated convergence against the fixed coefficient measures of D(vS)D(vS). On the right, D(ujS)→D(uS)D(u_jS)\to D(uS) by Proposition “One bounded potential over an arbitrary background”; multiplication by the fixed continuous function vv passes to the limit by R, Lemma 3.1. Differentiate and localize. ◻

This is again DIRECT_COROLLARY. No simultaneously varying discontinuous multiplier is used in its proof.

Relative Chern–Levine–Nirenberg bounds

Theorem (Finite-test relative CLN estimate). Fix K0⋐V⋐WK_0\Subset V\Subset W, an integer k≤n−pk\leq n-p, and a finite original moment presentation of a (q,q)(q,q)-test η\eta near K0K_0, where q=n−p−kq=n-p-k. There are positive compact tests θ1,…,θN∈Acn−p,n−p(V)\theta_1,\ldots,\theta_N\in A_c^{n-p,n-p}(V) and a constant CC, determined by these fixed data, such that for every positive separately closed background SS and every tuple u1,…,uk∈Cb(W)u_1,\ldots,u_k\in\mathcal C_b(W),

∣μTS,k(u1,…,uk),η∣(K0)≤C∏i=1k∥ui∥L∞(V)∑ℓ=1N⟨S,θℓ⟩.\bigl|\mu_{T_{S,k}(u_1,\ldots,u_k),\eta}\bigr|(K_0) \leq C\prod_{i=1}^k\|u_i\|_{L^\infty(V)} \sum_{\ell=1}^N\langle S,\theta_\ell\rangle.

One may replace each norm by osc⁡V(ui)\operatorname{osc}_V(u_i). The empty product is one when k=0k=0.

For a locally bounded Borel weight aa, the same right side multiplied by ∥a∥L∞(V)\|a\|_{L^\infty(V)} bounds ∣μaTS,k,η∣(K0)|\mu_{aT_{S,k},\eta}|(K_0). The controls and constants are independent of the potentials and of every regularization parameter. On a fixed finite test stage, the constant is bounded by a fixed constant times the coefficient supremum norm of η\eta.

Proof. Choose nested compact neighborhoods between K0K_0 and VV, with enough room for k+1k+1 cutoff choices. Lemma “Positive controls on a finite test stage” first bounds the left side by finitely many evaluations C0∑a⟨TS,k,Θa(0)⟩,Θa(0)∈Acq,q(V),Θa(0)≥0.C_0\sum_a\langle T_{S,k},\Theta_a^{(0)}\rangle, \qquad \Theta_a^{(0)}\in A_c^{q,q}(V),\quad \Theta_a^{(0)}\geq0. All these controls are chosen before the potentials. For every one of them, the defining recursion gives ⟨TS,k,Θa(0)⟩=∫u1 dμTS,k−1(u2,…,uk),DΘa(0).\langle T_{S,k},\Theta_a^{(0)}\rangle =\int u_1\,d\mu_{T_{S,k-1}(u_2,\ldots,u_k),D\Theta_a^{(0)}}. Apply Lemma “Positive controls on a finite test stage” to the fixed smooth test DΘa(0)D\Theta_a^{(0)}, using the next compact neighborhood. We obtain positive tests Θab(1)\Theta_{ab}^{(1)} of degree (q+1,q+1)(q+1,q+1) and constants independent of the potentials such that ⟨TS,k,Θa(0)⟩≤Ca∥u1∥L∞(V)∑b⟨TS,k−1,Θab(1)⟩.\langle T_{S,k},\Theta_a^{(0)}\rangle \leq C_a\|u_1\|_{L^\infty(V)} \sum_b\langle T_{S,k-1},\Theta_{ab}^{(1)}\rangle. Repeat this operation kk times. At each stage only a finite collection of fixed smooth compact tests is differentiated and controlled. The final forms have degree (q+k,q+k)=(n−p,n−p)(q+k,q+k)=(n-p,n-p), are positive, and are supported in VV. Absorb the finitely many constants into CC and enumerate the final controls as θℓ\theta_\ell. This proves (eq:relative-cln-product).

Adding a constant to any differentiated potential leaves the ordered product unchanged. Subtracting inf⁡Vui\inf_Vu_i therefore replaces the norm by its oscillation. Local boundedness and V‾⊂W\overline V\subset W make these extrema finite. The weighted estimate follows from ∣aμ∣≤∥a∥∞∣μ∣|a\mu|\leq\|a\|_\infty|\mu|.

For the last assertion, choose the initial cutoff controls using the fixed stage and use the matrix norm only in its first domination constant. All later differentiated controls are then fixed. No derivative of η\eta enters a varying constant. This proves the stated uniformity. ◻

Status.

The estimate is a relative extension needed in this project, with a complete proof above. Under the user’s stipulated inputs its logical status is DIRECT_COROLLARY: it follows by finite iteration of R’s fixed-test domination and the ordered recursion. It should not be advertised as a new independent analytical mechanism.

Charts and refinements.

The auxiliary forms need not come from a single ambient Euclidean ball, and no strictly psh exhaustion is assumed. Each domination is an inequality of intrinsic original CLD forms. Pulling it to a common moment refinement preserves positivity; R’s estimate already permits scalar coefficients using additional coordinates. Different choices of finite charts can give different numerical constants and controls, but they bound the same intrinsic coefficient measures. The proof therefore establishes compatibility of the estimate without asserting a chart-independent numerical norm.

What the estimate does not prove by itself.

A uniform total-variation bound does not allow a discontinuous multiplier to pass through weakly convergent coefficient measures. Consequently (eq:relative-cln-product), by itself, proves neither exchange of two discontinuous slots nor weighted joint decreasing continuity. A further comparison, energy, or capacity argument is required. In particular, one cannot replace the fixed continuous multiplier in R, Lemma 3.1 by an arbitrary bounded psh function merely by invoking this bound.

A common collar and weighted monotone convergence

The results of this section use the internal inputs listed in Section “Scope, inputs, and conventions”. In particular, they do not assert an independent proof of the regularization theorems in R and B. All currents and test forms belong to the original CLD calculus. Positivity means positivity in the cone used in R, not strong positivity. Set D=d′d′′D=d'd''.

We first record the exact local consequence of the internal approximation results that is needed below. A smooth psh function in this section has a positive original (1,1)(1,1) curvature form.

Lemma (A common domain for smooth majorants). Let x∈Ux\in U, and let ui,j↓uiu_{i,j}\downarrow u_i be finitely many decreasing families in Cb(U)\mathcal C_b(U) with locally bounded limits. After shrinking around xx, there is one ordinary neighborhood WW, independent of i,ji,j, with the following property. Each of the functions a=ui,ja=u_{i,j} and a=uia=u_i has smooth psh majorants aνa_\nu on WW such that aν≥a,aν(x′)⟶a(x′)(x′∈W).a_\nu\geq a,\qquad a_\nu(x')\longrightarrow a(x')\quad(x'\in W). The majorants are bounded uniformly in ν\nu on each fixed compact subset after a finite prefix is discarded. If a≤Ma\leq M on a compact neighborhood C⋐WC\Subset W, its majorants may be chosen with aν≤M+1a_\nu\leq M+1 on CC. The index discarded may depend on the function aa.

Proof. Use B, Proposition 7.1, to work on one ordinary algebraic neighborhood, with the original field and calibration. Apply B, Lemma D.2, to the entire finite family of sequences, before choosing approximants. It gives one projective variety, one ample line, one frame on WW, and bounded classical metrics whose weights on WW are the given functions plus one fixed constant. By B, Theorem 4.4, every such metric is a decreasing limit of pure Fubini–Study metrics on this same line. Their weights pνp_\nu are continuous, satisfy pν≥ap_\nu\geq a, and decrease pointwise to aa after the fixed constant is removed.

Smooth each pure Fubini–Study metric uniformly, using R, Lemma 3.2 and Proposition A.1. If the uniform error is at most ϵν↓0\epsilon_\nu\downarrow0, add ϵν\epsilon_\nu to the smoothed weight. The resulting smooth psh function aνa_\nu satisfies pν≤aν≤pν+2ϵν.p_\nu\leq a_\nu\leq p_\nu+2\epsilon_\nu. This proves pointwise convergence and the majorant property. On CC, the increasing open sets {pν<M+1/2}\{p_\nu<M+1/2\} cover CC whenever a≤Ma\leq M. Compactness gives a common index on CC. Discard this prefix and require 2ϵν<1/22\epsilon_\nu<1/2. The same argument with any bound for aa gives local uniform boundedness of the row. No monotonicity of the smoothed row is asserted or needed. ◻

Lemma (A smooth psh collar). Under the assumptions of Lemma “A common domain for smooth majorants”, there are compact polynomial norm boxes x∈Int⁡C1,C1⊂Int⁡Cmid,Cmid⊂Int⁡C2⋐W,x\in\operatorname{Int}C_1,\qquad C_1\subset\operatorname{Int}C_{\mathrm{mid}},\qquad C_{\mathrm{mid}}\subset\operatorname{Int}C_2\Subset W, a smooth psh function bb on a neighborhood of C2C_2, and a nonnegative original smooth cutoff χ\chi compactly supported in Ω=Int⁡C2\Omega=\operatorname{Int}C_2, with the following properties. After adding one constant cc to all the original functions, put vi,j=max⁡{ui,j+c,b},vi=max⁡{ui+c,b}.v_{i,j}=\max\{u_{i,j}+c,b\},\qquad v_i=\max\{u_i+c,b\}. Then:

  1. vi,j,vi∈Cb(Ω)v_{i,j},v_i\in\mathcal C_b(\Omega) and vi,j↓viv_{i,j}\downarrow v_i;

  2. vi,j=ui,j+cv_{i,j}=u_{i,j}+c and vi=ui+cv_i=u_i+c on a neighborhood of C1C_1;

  3. χ=1\chi=1 on a neighborhood of CmidC_{\mathrm{mid}};

  4. all vi,j,viv_{i,j},v_i equal bb on one fixed neighborhood of supp⁡d′χ∪supp⁡d′′χ\operatorname{supp}d'\chi\cup\operatorname{supp}d''\chi;

  5. each patched function vv has smooth psh majorants vν≥vv_\nu\geq v converging pointwise to vv, uniformly bounded on supp⁡χ\operatorname{supp}\chi, and equal to bb on that same collar.

The majorant rows in (5) need not be decreasing.

Proof. Choose the boxes by the construction in B, Lemma D.2. In particular, lower-bound polynomials have no zero near C2C_2. Choose thresholds strictly between the defining thresholds for C1C_1 and CmidC_{\mathrm{mid}}. The finite maximum b0=max⁡{0,log⁡∣PaNa/ca∣,log⁡∣dl/QlMl∣}b_0=\max\{0,\log|P_a^{N_a}/c_a|, \log|d_l/Q_l^{M_l}|\} is zero on a neighborhood of C1C_1 and at least some δ>0\delta>0 on a neighborhood in C2C_2 of C2∖Int⁡CmidC_2\setminus\operatorname{Int}C_{\mathrm{mid}}. All its branches other than the zero branch are logarithms of analytic units where they are active. A branch with a zero is strictly inactive near that zero.

Apply the regularized maximum of R, Proposition A.1, with smoothing radius smaller than all the fixed gaps. The result b~\widetilde b is an original smooth psh function: zero branches at polynomial zeros are deleted by the exact inactivity property. On a neighborhood of C1C_1, only the constant zero branch is active, so b~=0\widetilde b=0. On the outer collar, b~≥δ\widetilde b\geq\delta.

The endpoint bounds give a constant cc and a finite MM such that 1≤ui+c≤ui,j+c≤Mon a neighborhood of C2.1\leq u_i+c\leq u_{i,j}+c\leq M \quad\hbox{on a neighborhood of }C_2. Multiply b~\widetilde b by a positive constant so that the resulting bb is greater than M+3M+3 on the outer collar. Choose χ\chi, with 0≤χ≤10\leq\chi\leq1, equal to one near CmidC_{\mathrm{mid}} and compactly supported in Ω\Omega. Smooth cutoff existence and a finite localization are part of the original CLD test calculus. Its derivative support lies where the fixed outer-collar inequality holds.

Classical psh functions are stable under maxima and addition of constants. This proves (1)–(4). To prove (5), apply Lemma “A common domain for smooth majorants” to the unpatched function a=ui,j+ca=u_{i,j}+c or ui+cu_i+c. Discard a prefix so that its smooth psh majorants aνa_\nu are at most M+1M+1 on C2C_2. Define vν=rmax⁡ϵν(aν,b),ϵν↓0,v_\nu=\operatorname{rmax}_{\epsilon_\nu}(a_\nu,b), \qquad \epsilon_\nu\downarrow0, with all radii less than the fixed collar gap. The regularized maximum majorizes the ordinary maximum and has error at most ϵν\epsilon_\nu. It is exactly bb on the collar because the other branch is inactive there. These facts give all the assertions in (5). ◻

Remark. Neither lemma supplies a strictly psh exhaustion of an arbitrary analytic domain. The collar is constructed on a chosen small norm box. The closed positive form used to detect a test below is chosen separately and may be degenerate in directions not occurring in that test.

Lemma (Green exchange on a common collar). Let QQ be a symmetric positive d′d'- and d′′d''-closed current of bidegree (n−1,n−1)(n-1,n-1) on Ω\Omega. Let χ∈Ac0,0(Ω)\chi\in A_c^{0,0}(\Omega). Suppose that aa is smooth, gg is locally bounded Borel, and both equal one smooth function bb on a neighborhood of supp⁡d′χ∪supp⁡d′′χ\operatorname{supp}d'\chi\cup\operatorname{supp}d''\chi. Whenever D(gQ)D(gQ) has coefficient Radon type, one has ⟨aD(gQ),χ⟩=⟨gD(aQ),χ⟩.\langle aD(gQ),\chi\rangle =\langle gD(aQ),\chi\rangle. In this statement positivity of aa and gg is not required.

Proof. Since QQ is closed and aa is smooth, D(aQ)=Da∧QD(aQ)=Da\wedge Q. The definition of the distributional derivative gives

⟨aD(gQ)−gD(aQ),χ⟩=⟨gQ,D(χa)−χDa⟩=⟨Q,g{aDχ+d′χ∧d′′a+d′a∧d′′χ}⟩.\begin{aligned} \langle aD(gQ)-gD(aQ),\chi\rangle &=\langle gQ,D(\chi a)-\chi Da\rangle\\ &=\langle Q,g\{aD\chi+d'\chi\wedge d''a +d'a\wedge d''\chi\}\rangle. \end{aligned}

Every term in braces is supported in the collar. Substituting a=g=ba=g=b there makes the last expression ⟨bQ,D(χb)−χDb⟩=0.\langle bQ,D(\chi b)-\chi Db\rangle=0. The final equality follows by applying the same distributional identity to the smooth pair (b,b)(b,b). The substitutions are valid identities of coefficient measures. No derivative of the Borel function gg is used. ◻

Lemma (Monotonicity of the ordered trace integral). Let SS have bidegree (p,p)(p,p), let 0≤k≤n−p0\leq k\leq n-p, and let α\alpha be a smooth, positive, d′d'- and d′′d''-closed form of bidegree (n−p−k,n−p−k)(n-p-k,n-p-k) such that wedging it with a positive current preserves positivity. Let χ≥0\chi\geq0 be compactly supported in Ω\Omega.

For 0≤i≤k0\leq i\leq k, suppose fi,gi∈Cb(Ω)f_i,g_i\in\mathcal C_b(\Omega), fi≤gif_i\leq g_i, and all these functions equal the same smooth psh function bb near the derivative support of χ\chi. Assume that each gig_i has the smooth majorants with this exact collar supplied by Lemma “A smooth psh collar”. Define, using the ordered products, J(f0;f1,…,fk)=⟨f0TS,k(f1,…,fk),χα⟩.J(f_0;f_1,\ldots,f_k) =\langle f_0T_{S,k}(f_1,\ldots,f_k),\chi\alpha\rangle. Then J(f0;f1,…,fk)≤J(g0;g1,…,gk).J(f_0;f_1,\ldots,f_k) \leq J(g_0;g_1,\ldots,g_k). Only existence and positivity of the ordered products are used here.

Proof. The case k=0k=0 is measure monotonicity. For k>0k>0, fix smooth psh majorants Gi≥giG_i\geq g_i with the common collar. First replace the weight f0f_0 by G0G_0, using positivity of TS,k(f1,…,fk)∧αT_{S,k}(f_1,\ldots,f_k)\wedge\alpha. Apply Lemma “Green exchange on a common collar” with Q=TS,k−1(f2,…,fk)∧αQ=T_{S,k-1}(f_2,\ldots,f_k)\wedge\alpha to move G0G_0 into a differentiated position and f1f_1 into the weight. The resulting current DG0∧QDG_0\wedge Q is positive and closed. Replace the weight f1f_1 by G1G_1, and repeat. At the end one obtains J(f0;f1,…,fk)≤⟨Gk DG0∧⋯∧DGk−1∧S,χα⟩.J(f_0;f_1,\ldots,f_k) \leq\langle G_k\,DG_0\wedge\cdots\wedge DG_{k-1}\wedge S, \chi\alpha\rangle. A final application of the same Green identity, together with commutation of smooth even-degree factors, makes the right-hand side J(G0;G1,…,Gk)J(G_0;G_1,\ldots,G_k).

Now let the index of GkG_k tend to infinity, keeping every other index fixed. Bounded convergence against the fixed coefficient measures of SS implies D(GkS)→D(gkS)D(G_kS)\to D(g_kS). The outer factors and the weight are smooth and fixed, so this convergence can be tested directly. Next let the index of Gk−1G_{k-1} tend to infinity. The background D(gkS)D(g_kS) is now fixed, so the same argument applies. Continue in the order k,k−1,…,1k,k-1,\ldots,1. Finally, with the ordered product fixed, let the weight G0G_0 tend to g0g_0 by bounded convergence. The result is exactly J(g0;g1,…,gk)J(g_0;g_1,\ldots,g_k) in its original order. At no step are two discontinuous limits interchanged. ◻

Lemma (An upper bound for decreasing weights). Let Pj,PP_j,P be positive coefficient Radon currents of one bidegree, with Pj→PP_j\to P on all original compact smooth tests. Let aj↓aa_j\downarrow a be locally uniformly bounded functions. Suppose that, near the support of a fixed positive compact test Θ\Theta, every fixed ala_l has bounded continuous majorants converging pointwise to ala_l. Then lim sup⁡j→∞⟨ajPj,Θ⟩≤⟨aP,Θ⟩.\limsup_{j\to\infty}\langle a_jP_j,\Theta\rangle \leq\langle aP,\Theta\rangle.

Proof. For j≥lj\geq l and a continuous majorant al,ν≥ala_{l,\nu}\geq a_l, positivity gives ⟨ajPj,Θ⟩≤⟨al,νPj,Θ⟩.\langle a_jP_j,\Theta\rangle \leq\langle a_{l,\nu}P_j,\Theta\rangle. Keep l,νl,\nu fixed and pass to the limit in jj, using R, Lemma 3.1. Bounded convergence against the fixed positive Radon measure μP,Θ\mu_{P,\Theta} first allows ν→∞\nu\to\infty and then l→∞l\to\infty. ◻

Theorem (Weighted decreasing continuity for a fixed arbitrary background). Assume the internal inputs R and B. Let SS be any symmetric positive original CLD current of bidegree (p,p)(p,p) on UU, separately d′d'- and d′′d''-closed. Let 0≤k≤n−p0\leq k\leq n-p. If ui,j↓uiu_{i,j}\downarrow u_i in Cb(U)\mathcal C_b(U) for 0≤i≤k0\leq i\leq k, with locally bounded limits, then u0,jTS,k(u1,j,…,uk,j)⟶u0TS,k(u1,…,uk)u_{0,j}T_{S,k}(u_{1,j},\ldots,u_{k,j}) \longrightarrow u_0T_{S,k}(u_1,\ldots,u_k) on every original compact smooth complementary test form. The ordered unweighted products have the corresponding joint decreasing continuity. The background SS is fixed throughout.

Proof. Write WkW_k for the weighted assertion with kk differentiated slots. Bounded convergence for the fixed coefficient measures of SS proves W0W_0. Distributional differentiation shows that Wk−1W_{k-1} implies joint unweighted convergence in degree kk. We prove WkW_k assuming this unweighted convergence.

The statement is local on compact test supports. Use a finite original smooth partition to reduce to one test η\eta supported in a small norm box C1C_1 as in Lemma “A smooth psh collar”. Include in the chosen ordinary neighborhood all the unit coordinates needed to present η\eta. Let these coordinates be t1,…,tNt_1,\ldots,t_N, and set β=∑a=1Nd′ta∧d′′ta,r=n−p−k,α=βrr!.\beta=\sum_{a=1}^N d't_a\wedge d''t_a, \qquad r=n-p-k,\qquad \alpha=\frac{\beta^r}{r!}. For r=0r=0 set α=1\alpha=1. The form α\alpha is closed and strongly positive. In the fixed finite coefficient space it is a positive definite trace form. After replacing η\eta by its symmetric part, there is C>0C>0 such that Θ±=Cχα±η\Theta_\pm=C\chi\alpha\pm\eta are positive original compact tests. This is precisely the matrix domination used in R, Lemma 3.1; additional scalar coordinates are handled by pullback to a common refinement. No definiteness in all possible moment directions is required.

Patch all the potentials to vi,j,viv_{i,j},v_i by Lemma “A smooth psh collar”, and put Pj=TS,k(v1,j,…,vk,j),P=TS,k(v1,…,vk).P_j=T_{S,k}(v_{1,j},\ldots,v_{k,j}),\qquad P=T_{S,k}(v_1,\ldots,v_k). The induction hypothesis gives Pj→PP_j\to P. Applying Lemma “Monotonicity of the ordered trace integral” to the ordered tuples (v0,…,vk)≤(v0,j,…,vk,j)(v_0,\ldots,v_k)\leq(v_{0,j},\ldots,v_{k,j}) yields ⟨v0P,χα⟩≤⟨v0,jPj,χα⟩.\langle v_0P,\chi\alpha\rangle \leq\langle v_{0,j}P_j,\chi\alpha\rangle. Lemma “An upper bound for decreasing weights” gives the reverse limsup inequality. Consequently

⟨v0,jPj,χα⟩⟶⟨v0P,χα⟩.\langle v_{0,j}P_j,\chi\alpha\rangle \longrightarrow\langle v_0P,\chi\alpha\rangle.

Apply Lemma “An upper bound for decreasing weights” to Θ+\Theta_+ and Θ−\Theta_-. Together with (eq:trace-limit), the two resulting inequalities give respectively the upper and lower bounds for ⟨v0,jPj,η⟩\langle v_{0,j}P_j,\eta\rangle. Thus this pairing converges to ⟨v0P,η⟩\langle v_0P,\eta\rangle.

On a neighborhood of supp⁡η\operatorname{supp}\eta, the patched potentials are ui,j+cu_{i,j}+c and ui+cu_i+c. Ordinary locality of the recursion identifies the differentiated products there with the original products, since constants in differentiated slots vanish. Remove the constant cc from the weight by the unweighted degree-kk convergence already established. Finite summation over the test partition proves WkW_k.

This induction proves all WkW_k and all allowed unweighted assertions. It uses neither capacity nor energy convergence, nor any interchange of two discontinuous approximation parameters. ◻

Corollary (Symmetry and the first-stage calculus). Under the hypotheses of Theorem “Weighted decreasing continuity for a fixed arbitrary background”, the currents TS,kT_{S,k} are symmetric in the differentiated arguments, positive, symmetric as currents, and separately d′d'- and d′′d''-closed. They are additive and positively homogeneous in each potential, commute with ordinary restriction and gluing, and are independent of the regularization used to compute them. In particular, D(uD(vS))=D(vD(uS))(u,v∈Cb(U), p≤n−2).D\bigl(uD(vS)\bigr)=D\bigl(vD(uS)\bigr) \qquad(u,v\in\mathcal C_b(U),\ p\leq n-2).

Proof. All assertions except permutation symmetry were proved for the ordered recursion. Choose, locally on one fixed domain, continuous semipositive PL approximants decreasing to each potential, as in B. Continuous arbitrary-background products are symmetric, by uniform smooth approximation and fixed-continuous-multiplier convergence. Theorem “Weighted decreasing continuity for a fixed arbitrary background”, with a constant weight, passes this identity to the given potentials. The recursion identifies the result independently of the approximants and on overlaps. ◻

Corollary (Top-degree convergence as measures). If p+k=np+k=n, the weighted convergence in Theorem “Weighted decreasing continuity for a fixed arbitrary background” holds against every compactly supported continuous scalar function, locally as signed Radon measures.

Proof. Unweighted convergence against one nonnegative smooth cutoff equal to one near a prescribed compact set bounds the masses of the positive measures there. The weights have a common actual pointwise bound on a slightly larger compact neighborhood. Thus the weighted variations are uniformly bounded. Uniform density of original smooth scalar functions, with a fixed cutoff, extends convergence to continuous scalar tests. ◻

Remark (What independence means here). The construction is intrinsic to the given potentials and to SS. Independence of regularization does not assert that D(uS)D(uS) depends only on the ambient current D[u]D[u] for every possible pair of potentials. That stronger quotient-by-potentials statement is distinct from the claims proved here and from the special Bedford–Taylor-current condition in CLD 2025, Definition 8.6.2. Equality on all original tests also does not assert equality on all δ\delta-tests.

Continuous functions of several monotone weights

This section records a DIRECT_COROLLARY of Theorem “Weighted decreasing continuity for a fixed arbitrary background”. It is useful when a localization factor is a continuous function of several psh potentials, although that factor need not itself be psh.

Lemma (Increasing convex compositions). Let h∈Cb(V)h\in\mathcal C_b(V), and suppose that its values lie in a compact interval II. If φ:I→R\varphi:I\to\mathbf R is continuous, convex, and nondecreasing, then φ∘h∈Cb(V)\varphi\circ h\in\mathcal C_b(V). In particular, if h≥0h\geq0, every positive integral power of hh is classical psh. If hj↓hh_j\downarrow h have values in II, then φ(hj)↓φ(h)\varphi(h_j)\downarrow\varphi(h).

Proof. Interpolate φ\varphi linearly on successively refined dyadic partitions of II. The resulting functions φm\varphi_m are convex and nondecreasing, are finite maxima of affine functions with nonnegative slopes, and decrease uniformly to φ\varphi. Each φm(h)\varphi_m(h) is therefore a finite maximum of nonnegative multiples of hh plus constants. The defining classical class is stable under these operations. Its decreasing-limit property, with the local boundedness of the limit, gives φ(h)∈Cb(V)\varphi(h)\in\mathcal C_b(V). The last assertion follows from continuity and monotonicity of φ\varphi. ◻

Proposition (Functional weighted convergence). Let SS be a fixed background as in Theorem “Weighted decreasing continuity for a fixed arbitrary background”, and let 0≤k≤n−p0\leq k\leq n-p. Suppose that ui,j↓ui(1≤i≤k),ha,j↓ha(1≤a≤s)u_{i,j}\downarrow u_i\quad(1\leq i\leq k), \qquad h_{a,j}\downarrow h_a\quad(1\leq a\leq s) are finite families in Cb(U)\mathcal C_b(U) with locally bounded limits. Set Pj=TS,k(u1,j,…,uk,j),P=TS,k(u1,…,uk).P_j=T_{S,k}(u_{1,j},\ldots,u_{k,j}),\qquad P=T_{S,k}(u_1,\ldots,u_k). For every continuous function F:Rs→RF:\mathbf R^s\to\mathbf R, F(h1,j,…,hs,j)Pj⟶F(h1,…,hs)PF(h_{1,j},\ldots,h_{s,j})P_j \longrightarrow F(h_1,\ldots,h_s)P on every original compact smooth complementary test. The factors F(h1,j,…,hs,j)F(h_{1,j},\ldots,h_{s,j}) are the actual Borel functions, and no psh assertion about those factors is required.

Proof. Fix a compact test η\eta and a relatively compact neighborhood VV of its support. The endpoint bounds provide one compact interval for all values of each ha,jh_{a,j} and hah_a on VV. Add fixed constants so that the shifted functions xa,j=ha,j+cax_{a,j}=h_{a,j}+c_a and xa=ha+cax_a=h_a+c_a are nonnegative there. Absorb these shifts into FF.

We first prove convergence for a polynomial in the xax_a. Every monomial of total degree d≥1d\geq1 can be regarded as a product z1⋯zdz_1\cdots z_d in which variables may repeat. The polarization identity z1⋯zd=1d!∑A⊂{1,…,d}(−1)d−∣A∣(∑a∈Aza)dz_1\cdots z_d =\frac1{d!}\sum_{A\subset\{1,\ldots,d\}} (-1)^{d-|A|}\left(\sum_{a\in A}z_a\right)^d expresses it as a linear combination of powers of sums with nonnegative coefficients. Each nonzero sum of shifted psh functions is nonnegative and classical psh. By Lemma “Increasing convex compositions”, its ddth power is also classical psh and decreases along the given sequences. The weighted theorem therefore gives convergence for every term in the displayed identity. Constants are covered by its constant-weight case. Finite linearity proves the assertion for every polynomial.

The unweighted convergence Pj→PP_j\to P, together with R, Lemma 3.1, supplies a finite bound sup⁡j∣μPj,η∣(supp⁡η)+∣μP,η∣(supp⁡η)<∞.\sup_j|\mu_{P_j,\eta}|(\operatorname{supp}\eta) +|\mu_{P,\eta}|(\operatorname{supp}\eta)<\infty. Approximate FF uniformly by polynomials on the common compact rectangle of shifted values. An approximation error of at most ϵ\epsilon changes each pairing by at most ϵ\epsilon times this fixed variation bound. First pass to the limit in jj for the polynomial, and then let ϵ↓0\epsilon\downarrow0. This proves the assertion on η\eta. ◻

Remark. The proposition permits, for example, bounded continuous cutoffs of uj−vju_j-v_j and products of finitely many bounded psh weights. It does not permit replacing FF by a discontinuous function without an additional argument. In applications to strict maximum sets, one first uses continuous scalar cutoffs and only afterwards takes a bounded pointwise limit against the already fixed coefficient measures.

A sufficient condition for varying backgrounds

The main theorem fixes the background. The following separate result gives a sufficient additional hypothesis under which a background may vary. It is an elementary DIRECT_COROLLARY of the coefficient-Radon domination argument and the ordered construction; it does not assert unrestricted varying-background stability.

Lemma (Order domination and Borel multipliers). Let Pj,P,RP_j,P,R be original positive currents of one bidegree on UU. Suppose that Pj→PP_j\to P on all original compact smooth tests and that 0≤Pj≤CR0\leq P_j\leq C R for one constant C<∞C<\infty. Let aja_j be Borel functions converging pointwise to aa, with one common bound on each compact subset. Then ajPj⟶aPa_jP_j\longrightarrow aP on every original compact smooth test. Closedness of these currents is not required in this lemma.

Proof. Weak closure of the positive cone gives 0≤P≤CR0\leq P\leq CR. Fix a compact test η\eta. The finite positive control construction in R, Lemma 3.1 supplies a finite positive Radon measure νη\nu_\eta, depending only on RR, CC and the test, such that

∣μPj,η∣≤νη,∣μP,η∣≤νη.|\mu_{P_j,\eta}|\leq\nu_\eta, \qquad |\mu_{P,\eta}|\leq\nu_\eta.

Here these are inequalities of measures, not only bounds on their total masses. Indeed, on each moment piece choose a positive compact control Θ\Theta with AΘ±η≥0A\Theta\pm\eta\geq0. For every nonnegative smooth scalar hh, ∣⟨Pj,hη⟩∣≤A⟨Pj,hΘ⟩≤AC⟨R,hΘ⟩.|\langle P_j,h\eta\rangle| \leq A\langle P_j,h\Theta\rangle \leq AC\langle R,h\Theta\rangle. Radon uniqueness and scalar density give domination by ACμR,ΘAC\mu_{R,\Theta}. Sum over the finite pieces to obtain (eq:fixed-measure-domination).

For a fixed bounded Borel function bb, approximate bb in L1(νη)L^1(\nu_\eta) by continuous functions on a compact neighborhood of the test support. Such approximation is available for a finite Radon measure; cutoffs keep the functions on that neighborhood. R, Lemma 3.1 gives convergence with each fixed continuous multiplier, and (eq:fixed-measure-domination) bounds the approximation error uniformly in jj. It follows that ∫b dμPj,η⟶∫b dμP,η.\int b\,d\mu_{P_j,\eta}\longrightarrow \int b\,d\mu_{P,\eta}. Finally, ∣∫(aj−a) dμPj,η∣≤∫∣aj−a∣ dνη⟶0\left|\int(a_j-a)\,d\mu_{P_j,\eta}\right| \leq\int|a_j-a|\,d\nu_\eta\longrightarrow0 by bounded convergence for the fixed measure νη\nu_\eta. Apply the fixed-multiplier assertion to b=ab=a and combine the two limits. ◻

Proposition (Order-dominated varying backgrounds). Let R0,Sj,SR_0,S_j,S be symmetric positive separately closed original (p,p)(p,p)-currents on UU. Assume Sj⟶S,0≤Sj≤CR0S_j\longrightarrow S, \qquad 0\leq S_j\leq C R_0 on UU, with one finite constant CC. Fix u1,…,uk∈Cb(U)u_1,\ldots,u_k\in\mathcal C_b(U), where 0≤k≤n−p0\leq k\leq n-p. Then TSj,k(u1,…,uk)⟶TS,k(u1,…,uk).T_{S_j,k}(u_1,\ldots,u_k) \longrightarrow T_{S,k}(u_1,\ldots,u_k). Moreover, for every locally uniformly bounded sequence of Borel functions aj→aa_j\to a pointwise, ajTSj,k(u1,…,uk)⟶aTS,k(u1,…,uk).a_jT_{S_j,k}(u_1,\ldots,u_k) \longrightarrow aT_{S,k}(u_1,\ldots,u_k). All convergence is on the full original compact smooth test sheaf. The differentiated potential tuple is fixed.

Proof. For a fixed potential tuple the ordered construction is linear in the background. Positivity therefore gives

0≤TSj,r(uk−r+1,…,uk)≤CTR0,r(uk−r+1,…,uk)0\leq T_{S_j,r}(u_{k-r+1},\ldots,u_k) \leq C T_{R_0,r}(u_{k-r+1},\ldots,u_k)

for every 0≤r≤k0\leq r\leq k: the difference is the ordered product with the positive closed background CR0−SjCR_0-S_j.

Induct on rr. The case r=0r=0 is the assumed weak convergence. Assume convergence at degree r−1r-1. Its currents are dominated by the one fixed positive current on the right of (eq:product-order-domination). Apply Lemma “Order domination and Borel multipliers” with the fixed multiplier uk−r+1u_{k-r+1}, and then apply DD. This gives convergence at degree rr by the recursion. At r=kr=k, apply the same lemma once more, now with the varying Borel weights aja_j, to obtain the final assertion. ◻

Remark. The hypothesis Sj≤CR0S_j\leq CR_0 supplies domination by fixed coefficient measures. It is stronger than a uniform bound on local masses. The conclusion permits discontinuous Borel weights precisely because this additional domination is available. No varying differentiated tuple, and no assertion for arbitrary weakly convergent backgrounds, is included.

Continuous changes of potential

Lemma (Continuous harmonic multipliers). Assume the local regularization theorem of R. Let SS be an allowed positive separately closed original current of bidegree (p,p)(p,p), with p≤n−1p\leq n-1. If hh is finite continuous and D[h]=0D[h]=0 as an ambient original current, then D(hS)=0.D(hS)=0. Consequently, if u,vu,v are finite continuous and D[u]=D[v]D[u]=D[v], then D(uS)=D(vS)D(uS)=D(vS). This conclusion concerns continuous potentials.

Proof. Work on the ordinary local algebraic neighborhoods supplied by B, Proposition 7.1. Both hh and −h-h have positive ambient curvature currents, since those currents vanish. By R, Theorem 2.9, each is locally uniformly approximable by smooth functions with positive curvature forms. For a smooth approximant hmh_m, the current D(hmS)=Dhm∧SD(h_mS)=Dh_m\wedge S is positive. Locally uniform convergence passes through multiplication by the fixed coefficient-Radon current SS and then through DD. Hence D(hS)D(hS) is positive. Apply the same argument to −h-h: the current −D(hS)-D(hS) is positive as well. The original positive cone of currents is pointed, by the finite-test domination in R, Lemma 3.1, so D(hS)=0D(hS)=0. Take h=u−vh=u-v for the final assertion and localize. No classical-psh assertion about hh is needed. ◻

This is a DIRECT_COROLLARY of R and the fixed- background one-potential construction. Applying it to the positive closed current formed by the remaining slots also proves potential independence for a continuous slot in a mixed product. It does not establish the same statement for an arbitrary discontinuous difference u−vu-v with D[u−v]=0D[u-v]=0.

A limited quotient-by-potentials statement

The main theorem constructs products of specified, actual pointwise potentials. The assertion that D(uS)D(uS) depends only on the ambient current D[u]D[u] is stronger. The following cases do follow from the available inputs. They are DIRECT_COROLLARY statements, not an assertion of this stronger property for every background.

Lemma (Ambient mixed backgrounds). Let u,v∈Cb(U)u,v\in\mathcal C_b(U) satisfy D[u]=D[v]D[u]=D[v] on all original compact smooth tests. Let 0≤p≤n−10\leq p\leq n-1, and suppose locally that S=∑ℓ=1NcℓT[1],p(aℓ,1,…,aℓ,p),aℓ,i∈Cb(U),cℓ∈R.S=\sum_{\ell=1}^{N}c_\ell T_{[1],p}(a_{\ell,1},\ldots,a_{\ell,p}), \qquad a_{\ell,i}\in\mathcal C_b(U),\quad c_\ell\in\mathbf R. Then D(uS)=D(vS).D(uS)=D(vS). The conclusion also holds for a locally finite sum of this kind. In particular, it applies when such a current is an allowed positive closed background.

Proof. It suffices to consider one summand. By the symmetry already present in B, Theorem 2.6,

D(uT[1],p(a1,…,ap))=T[1],p+1(a1,…,ap,u)=D(a1D(a2⋯D(apD[u])⋯ )).\begin{aligned} D\bigl(uT_{[1],p}(a_1,\ldots,a_p)\bigr) &=T_{[1],p+1}(a_1,\ldots,a_p,u)\\ &=D\bigl(a_1D(a_2\cdots D(a_pD[u])\cdots)\bigr). \end{aligned}

Replace the innermost current D[u]D[u] by D[v]D[v]. At every subsequent step, equality of coefficient-Radon currents implies equality after multiplication by the same actual Borel function, by B, Lemma 7.2. Distributional differentiation preserves that equality. Reversing the preceding symmetry gives D(vS)D(vS). Finite linearity and localization prove the stated extensions. For p=0p=0, the assertion is the original equality D[u]=D[v]D[u]=D[v]. ◻

Lemma (Smooth closed backgrounds). Under the same hypothesis on u,vu,v, let S=[α]S=[\alpha], where α\alpha is an original smooth (p,p)(p,p)-form satisfying d′α=d′′α=0d'\alpha=d''\alpha=0. Then D(uS)=D(vS)D(uS)=D(vS).

Proof. Set h=u−vh=u-v. The given local boundedness makes all products below well defined. Borel multiplication of the integration current agrees with smooth-form multiplication: h[α]=α∧[h].h[\alpha]=\alpha\wedge[h]. The distributional Leibniz rule and the two closedness assumptions give D(hS)=D(α∧[h])=α∧D[h]=0.D(hS)=D(\alpha\wedge[h])=\alpha\wedge D[h]=0. This is an equality on the entire original compact smooth test sheaf. ◻

Proposition (Passage through a dominated weak closure). Fix u,v∈Cb(U)u,v\in\mathcal C_b(U) with D[u]=D[v]D[u]=D[v]. Let Sj,S,R0S_j,S,R_0 be allowed positive separately closed backgrounds of degree (p,p)(p,p), with Sj⟶S,0≤Sj≤CR0S_j\longrightarrow S,\qquad 0\leq S_j\leq C R_0 for one finite constant CC. Suppose that D(uSj)=D(vSj)D(uS_j)=D(vS_j) for every jj, for example by either preceding lemma. Then D(uS)=D(vS).D(uS)=D(vS).

Proof. The function h=u−vh=u-v is an actual locally bounded Borel function. Lemma “Order domination and Borel multipliers”, with the fixed multiplier hh, gives hSj→hShS_j\to hS on all original compact smooth tests. Apply DD to this convergence. Every D(hSj)D(hS_j) is zero by hypothesis, so D(hS)=0D(hS)=0. ◻

Remark (The remaining scope question). These arguments give neither a representation of an arbitrary original-positive closed current by ambient bounded-potential products, nor a dominated approximation of every such current by the backgrounds above. They also do not show that a locally bounded difference h=u−vh=u-v satisfying D[h]=0D[h]=0 is a continuous function at its actual point values. This original-current kernel regularity question remains open in this investigation. The previously proved continuous-harmonic multiplier lemma applies if that additional continuity is known.

Relative energy and convergence in capacity

Throughout this section, the fixed-background calculus of Theorem “Weighted decreasing continuity for a fixed arbitrary background” is available. Its proof uses no energy or capacity assertion from this section. All measures below are measures on the actual Berkovich space, and all multipliers use their given point values. We continue to use the original positive cone of R.

A relative energy form

Lemma (Regularization on a prescribed smaller norm box). Work in one ordinary algebraic neighborhood supplied by B. Let Cin⋐Int⁡CoutC_{\mathrm{in}}\Subset\operatorname{Int}C_{\mathrm{out}} be compact polynomial norm boxes, and let finitely many bounded classical psh functions be defined on a neighborhood of CoutC_{\mathrm{out}}. They admit smooth decreasing psh approximants on one neighborhood of the entire CinC_{\mathrm{in}}. Thus the smaller domain need not be chosen separately at each point of a prescribed compact test support.

Proof. This is the compact-box form of the construction in B, Lemma D.2, followed by its Theorem 4.4. We spell out the distinction from applying local regularization separately at every point. Insert a middle norm box and construct the finite polynomial maximum bb used in that lemma: it is zero on CinC_{\mathrm{in}} and is at least a fixed δ>0\delta>0 on the outer collar. Homogenizing its branches and adding inactive global generating sections realizes bb as the weight of one pure-Fubini–Study metric qq, in a frame nonzero near CoutC_{\mathrm{out}}. After adding one common constant, all given functions lie in [1,M][1,M] there. Choose mδ>M+1m\delta>M+1 and give each extended metric the local weight max⁡{a+c,mb}\max\{a+c,mb\}, setting it equal to mqmq outside the middle box. These prescriptions agree on their overlap. They define bounded classical metrics on one projective variety and one ample line, with weights exactly a+ca+c on a neighborhood of CinC_{\mathrm{in}}.

Theorem 4.4 of B gives decreasing global pure-Fubini–Study approximants for each metric. Their smooth global approximations from R, Lemma 3.2, restrict to the same frame near the whole CinC_{\mathrm{in}}. If the uniform smoothing error is at most ϵm=4−m\epsilon_m=4^{-m}, adding 2ϵm2\epsilon_m gives a decreasing smooth row: its values lie between pm+ϵmp_m+\epsilon_m and pm+3ϵmp_m+3\epsilon_m, while pm+1≤pmp_{m+1}\leq p_m. These particular smooth approximants are classical psh. Indeed, after any complete field extension and pullback to an analytic curve, their active branches are logarithms of norms of analytic units, hence harmonic functions. The regularized maximum is convex and nondecreasing in each variable, so its composition with these branches is subharmonic, by the curve chain rule in B, Lemma B.1. A section vanishing identically or at a point gives an inactive branch nearby, which is deleted. Positive scaling and addition of constants preserve this property. Thus the rows used below belong to Cb\mathcal C_b; no general identification of original-form positivity with classical plurisubharmonicity is being assumed. The same convex, coordinatewise nondecreasing composition argument applies to the additional regularized maxima with the smooth collar below. Remove the common constant cc to finish. No extension of a background current is used in this argument. ◻

Let QQ be a symmetric positive, separately d′d'- and d′′d''-closed current of bidegree (n−1,n−1)(n-1,n-1) on an ordinary open Ω\Omega. Let Db(Ω)\mathcal D_b(\Omega) be the real vector space generated locally by Cb(Ω)\mathcal C_b(\Omega). If f,g∈Db(Ω)f,g\in\mathcal D_b(\Omega), define

ΓQ(f,g)=12{D(fgQ)−fD(gQ)−gD(fQ)},ΓQ(f)=ΓQ(f,f).\Gamma_Q(f,g) =\frac12\bigl\{D(fgQ)-fD(gQ)-gD(fQ)\bigr\}, \qquad \Gamma_Q(f)=\Gamma_Q(f,f).

Here D(fQ)D(fQ) is a signed coefficient Radon current, by linearity of the ordered one-potential construction. The first term in (eq:relative-gamma) is initially a distribution. The following lemma proves that the full expression is a measure.

Lemma (Relative energy measures). For f,g∈Db(Ω)f,g\in\mathcal D_b(\Omega), ΓQ(f,g)\Gamma_Q(f,g) is a signed Radon measure and ΓQ(f)\Gamma_Q(f) is nonnegative. For every nonnegative compact smooth χ\chi, one has

∣∫χ dΓQ(f,g)∣2≤(∫χ dΓQ(f))(∫χ dΓQ(g)).\left|\int\chi\,d\Gamma_Q(f,g)\right|^2 \leq\left(\int\chi\,d\Gamma_Q(f)\right) \left(\int\chi\,d\Gamma_Q(g)\right).

Suppose, for the following integration formula, that f=a−bf=a-b, where a,b∈Cba,b\in\mathcal C_b are functions on one common norm-box domain and agree outside a compact LL contained in a strictly smaller norm box. Suppose also that the psh summands of gg are defined on that larger domain. This includes the common-collar differences used below. Then

∫f D(gQ)=∫g D(fQ)=−∫ΓQ(f,g),∫ΓQ(f)=−∫f D(fQ).\int f\,D(gQ)=\int g\,D(fQ) =-\int\Gamma_Q(f,g), \qquad \int\Gamma_Q(f)=-\int f\,D(fQ).

The integrals in (eq:relative-green-energy) are supported on a fixed compact neighborhood of supp⁡f\operatorname{supp}f.

Proof. First suppose that all functions are smooth. Then ΓQ(f,g)=12(d′f∧d′′g+d′g∧d′′f)∧Q.\Gamma_Q(f,g) =\frac12(d'f\wedge d''g+d'g\wedge d''f)\wedge Q. Symmetry of QQ makes the pairing symmetric. Since d′f∧d′′fd'f\wedge d''f is a strongly positive (1,1)(1,1)-form, the measure ΓQ(f)\Gamma_Q(f) is positive. Positivity of ΓQ(f+tg)\Gamma_Q(f+t g) for every real tt proves (eq:relative-energy-CS) by the discriminant inequality. Stokes proves (eq:relative-green-energy).

We explain carefully the approximation used to pass these identities to bounded potentials. On a common smaller domain, let pm↓ap_m\downarrow a be the continuous pure-Fubini–Study weights supplied by B for a given a∈Cba\in\mathcal C_b. Smooth pmp_m uniformly with error at most ϵm=4−m\epsilon_m=4^{-m}, and call the result sms_m. Then am=sm+2ϵm,pm+ϵm≤am≤pm+3ϵm,am+1<am.a_m=s_m+2\epsilon_m, \qquad p_m+\epsilon_m\leq a_m\leq p_m+3\epsilon_m, \qquad a_{m+1}<a_m. Thus ama_m are smooth psh functions decreasing pointwise to aa. The same construction is made for finitely many psh summands.

If an exact common collar is required, apply a regularized maximum with the fixed smooth collar function from Lemma “A smooth psh collar”. Its smoothing radii may decrease to zero. For completeness, the regularized maximum is nondecreasing in the smoothing radius: for fixed input, its expression as the integral of a maximum of affine functions of the radius is convex, and symmetry of the mollifier makes it even. Hence it is nondecreasing for nonnegative radius. The patched smooth rows remain decreasing and equal the prescribed collar function exactly.

Write f=∑ciaif=\sum c_i a_i and g=∑dlblg=\sum d_l b_l, with classical psh ai,bla_i,b_l, and use such simultaneous smooth rows. Products of bounded functions converge pointwise and are uniformly bounded on each compact set. Therefore D(fmgmQ)→D(fgQ)D(f_mg_mQ)\to D(fgQ) by bounded convergence against the fixed coefficient measures of QQ. Each term in fmD(gmQ)f_mD(g_mQ) is a constant multiple of ai,mD(bl,mQ)a_{i,m}D(b_{l,m}Q), so it converges by Theorem “Weighted decreasing continuity for a fixed arbitrary background” with one differentiated slot. The same holds with ff and gg exchanged. Consequently ΓQ(fm,gm)⟶ΓQ(f,g)\Gamma_Q(f_m,g_m)\longrightarrow\Gamma_Q(f,g) as distributions. A positive distribution in top degree is a positive Radon measure. Apply this to f+tgf+t g, and polarize, to obtain the measure assertion and (eq:relative-energy-CS).

It remains to justify compact-support integration by parts without assuming differentiability of ff. Write f=a−bf=a-b as in the statement. Choose several nested norm boxes between LL and the boundary of the given domain. First apply Lemma “Regularization on a prescribed smaller norm box” with a smaller box still containing all the boxes to be used below. This gives smooth decreasing rows for a,ba,b and all psh summands of gg on a neighborhood of that entire box. This step is essential: applying a pointwise regularization theorem separately would not produce a domain containing the full support of ff.

Inside this common regularization domain, choose a second smooth psh collar which is zero near LL and dominates the bounded shifted values of aa and bb on an outer collar. Patch their smooth rows by a regularized maximum with this collar, with radii decreasing to zero. The limiting patched difference is still exactly ff: on LL the collar is inactive, and outside LL the two limiting functions agree. The smooth differences fmf_m have support in one fixed compact set because their two rows have the same exact outer collar. The smooth rows of the psh summands of gg are already defined near this compact set. No approximation or global extension of the fixed QQ is involved.

For these smooth rows, insert one compact smooth cutoff equal to one on the common support of fmf_m. Stokes gives the two Green equalities. Every term is a finite sum of weighted one-potential products, so Theorem “Weighted decreasing continuity for a fixed arbitrary background” passes it to the limit. The energy terms converge by the already proved distributional formula and positivity, with the same cutoff. All resulting multipliers are the Borel multipliers in (eq:relative-gamma); no exceptional-set representative is chosen. This local integration formula is the only compact-support version used below. ◻

Remark (A noncircular estimate before joint continuity). There is a useful estimate that precedes the preceding lemma. Suppose aj↓aa_j\downarrow a are smooth psh functions, uniformly bounded near a compact set, and all equal one smooth function outside a fixed compact L⋐ΩL\Subset\Omega. Choose χ≥0\chi\geq0, compactly supported, with χ=1\chi=1 near LL, and put Eχ(a)=∫χ d′a∧d′′a∧Q.E_\chi(a)=\int\chi\,d'a\wedge d''a\wedge Q. For j<lj<l, g=aj−al≥0g=a_j-a_l\geq0 has compact support in LL, and smooth integration by parts gives

Eχ(al)−Eχ(aj)=Eχ(g)+2∫g D(ajQ)≥Eχ(g).E_\chi(a_l)-E_\chi(a_j) =E_\chi(g)+2\int g\,D(a_jQ)\geq E_\chi(g).

The identity Eχ(aj)=12⟨Q,aj2Dχ⟩−⟨D(ajQ),χaj⟩E_\chi(a_j)=\frac12\langle Q,a_j^2D\chi\rangle -\langle D(a_jQ),\chi a_j\rangle and the relative CLN estimate bound these energies uniformly. They therefore increase to a finite limit, and the gradients are Cauchy in the local Hilbert seminorm induced by QQ. In fact the smooth energy measures are Cauchy in total variation: ∥χΓQ(aj)−χΓQ(al)∥TV≤Eχ(aj−al)1/2(Eχ(aj)1/2+Eχ(al)1/2).\|\chi\Gamma_Q(a_j)-\chi\Gamma_Q(a_l)\|_{\mathrm{TV}} \leq E_\chi(a_j-a_l)^{1/2} \bigl(E_\chi(a_j)^{1/2}+E_\chi(a_l)^{1/2}\bigr). To see this, polarize the difference and use the measure-valued Cauchy–Schwarz inequality. That inequality follows by writing the positive 2×22\times2 matrix of energy measures relative to the sum of its two diagonal measures, applying the pointwise discriminant inequality to its densities, and then ordinary Cauchy–Schwarz. This statement uses only the one-potential construction and relative CLN. We do not use it to prove Theorem “Weighted decreasing continuity for a fixed arbitrary background”; that theorem has the independent trace-squeeze proof.

A localized relative capacity

Fix a positive closed background SS of bidegree (p,p)(p,p), and put q=n−pq=n-p. Work on a relatively compact ordinary moment domain Ω\Omega on which a smooth closed strongly positive (1,1)(1,1)-form β\beta is defined. A typical choice is the trace form in the proof of Theorem “Weighted decreasing continuity for a fixed arbitrary background”. No strict positivity of β\beta in all analytic directions is assumed. For 1≤k≤q1\leq k\leq q and a Borel subset E⊂ΩE\subset\Omega, define

Cap⁡S,β,k(E;Ω)=sup⁡v∈Cb(Ω)0≤v≤1∫ETS,k(v,…,v)∧βq−k.\operatorname{Cap}_{S,\beta,k}(E;\Omega) =\sup_{\substack{v\in\mathcal C_b(\Omega)\\0\leq v\leq1}} \int_ET_{S,k}(v,\ldots,v)\wedge\beta^{q-k}.

For k=qk=q, β\beta drops out, and we write Cap⁡S(E;Ω)\operatorname{Cap}_S(E;\Omega). In particular, S=[1]S=[1] gives the usual candidate non-Archimedean Bedford–Taylor capacity in top degree. We make no nondegeneracy assertion for an arbitrary SS or Ω\Omega.

Proposition (Elementary properties and mixed domination). The capacity in (eq:relative-capacity) is monotone and countably subadditive on Borel sets, finite on compact subsets of Ω\Omega, and continuous from below on increasing Borel sets. Restriction to a smaller domain can only increase it. If a1,…,ak∈Cb(Ω)a_1,\ldots,a_k\in\mathcal C_b(\Omega) satisfy 0≤ai≤Mi0\leq a_i\leq M_i, then

∫ETS,k(a1,…,ak)∧βq−k≤kkk!(∏i=1kMi)Cap⁡S,β,k(E;Ω).\int_ET_{S,k}(a_1,\ldots,a_k)\wedge\beta^{q-k} \leq\frac{k^k}{k!}\left(\prod_{i=1}^kM_i\right) \operatorname{Cap}_{S,\beta,k}(E;\Omega).

If some Mi=0M_i=0, the left side vanishes. Constants may first be subtracted from the aia_i, so bounds on their oscillations suffice.

Proof. The first assertions follow from the same properties of positive measures and the supremum in (eq:relative-capacity); for an increasing sequence, the two suprema over the sequence index and the admissible functions commute. Compact finiteness is the uniform relative CLN bound. Domain monotonicity follows by restricting every admissible function on the larger domain.

If all Mi>0M_i>0, set vi=ai/Miv_i=a_i/M_i and v=k−1∑iviv=k^{-1}\sum_i v_i. Then 0≤v≤10\leq v\leq1. Symmetry, multilinearity, and positivity show that the expansion of TS,k(v,…,v)T_{S,k}(v,\ldots,v) contains k!k−kTS,k(v1,…,vk)k!k^{-k}T_{S,k}(v_1,\ldots,v_k), with all other terms positive. This gives (eq:mixed-capacity-domination). ◻

Theorem (A quantitative relative capacity estimate). For this theorem, take Ω\Omega to be the ordinary interior of one polynomial norm box in an algebraic neighborhood, and assume that the compact set LL below is contained in a strictly smaller nested norm box. Let uj,u∈Cb(Ω)u_j,u\in\mathcal C_b(\Omega) satisfy uj↓uu_j\downarrow u. Suppose that there are such a compact L⋐ΩL\Subset\Omega and one smooth psh function hh such that uj=u=hu_j=u=h on Ω∖L\Omega\setminus L, and that the functions have a common bound on a neighborhood of LL. Set gj=uj−u≥0g_j=u_j-u\geq0. For 1≤k≤q1\leq k\leq q, there is a finite constant CC, independent of jj, vv, and δ>0\delta>0, such that

Cap⁡S,β,k({gj>δ};Ω)≤Cδ(∫gj TS,k(u,…,u)∧βq−k)2−k.\operatorname{Cap}_{S,\beta,k}(\{g_j>\delta\};\Omega) \leq\frac{C}{\delta} \left(\int g_j\, T_{S,k}(u,\ldots,u)\wedge\beta^{q-k}\right)^{2^{-k}}.

The constant depends on the compact collar, β\beta, a bound for uu on a slightly larger compact set, and finitely many positive cutoff evaluations of SS. Consequently uj→uu_j\to u in this relative capacity.

Proof. Choose a nonnegative compact smooth χ\chi equal to one near LL. Fix an admissible vv, and abbreviate products using the symmetric calculus already proved. For 0≤r≤k0\leq r\leq k, set Ir=∫gj (Dv)r(Du)k−r∧S∧βq−k.I_r=\int g_j\,(Dv)^r(Du)^{k-r}\wedge S\wedge\beta^{q-k}. These numbers are finite and nonnegative. For r≥1r\geq1, put Qr=(Dv)r−1(Du)k−r∧S∧βq−k.Q_r=(Dv)^{r-1}(Du)^{k-r}\wedge S\wedge\beta^{q-k}. This is an arbitrary allowed positive closed current of bidegree (n−1,n−1)(n-1,n-1); the energy lemma therefore applies to it. Since gjg_j has compact support in LL, Green’s identity and Cauchy–Schwarz give

Ir=−∫ΓQr(gj,v),Ir2≤(∫ΓQr(gj))(∫χ ΓQr(v)).\begin{aligned} I_r &= -\int\Gamma_{Q_r}(g_j,v),\notag\\ I_r^2 &\leq\left(\int\Gamma_{Q_r}(g_j)\right) \left(\int\chi\,\Gamma_{Q_r}(v)\right). \end{aligned}

The first factor satisfies the one-sided estimate

0≤∫ΓQr(gj)=−∫gjD(gjQr)=∫gjD(uQr)−∫gjD(ujQr)≤Ir−1.\begin{aligned} 0\leq\int\Gamma_{Q_r}(g_j) &= -\int g_jD(g_jQ_r)\notag\\ &=\int g_jD(uQ_r)-\int g_jD(u_jQ_r) \leq I_{r-1}. \end{aligned}

The final inequality uses gj≥0g_j\geq0 and positivity of D(ujQr)D(u_jQ_r). Thus there is no need to estimate a signed curvature by exchanging two limits.

For the second factor, the energy identity gives ∫χ ΓQr(v)=12⟨Qr,v2Dχ⟩−⟨D(vQr),χv⟩.\int\chi\,\Gamma_{Q_r}(v) =\frac12\langle Q_r,v^2D\chi\rangle -\langle D(vQ_r),\chi v\rangle. It is nonnegative. Since 0≤v≤10\leq v\leq1, it is bounded above by the absolute coefficient estimate for the first term. Apply the relative CLN bound to QrQ_r, whose differentiated arguments consist of r−1r-1 copies of vv and k−rk-r copies of the fixed bounded uu. The smooth β\beta factors are fixed. Hence 0≤∫χ ΓQr(v)≤Ar,0\leq\int\chi\,\Gamma_{Q_r}(v)\leq A_r, where ArA_r is finite and independent of vv and jj. Combining this with (eq:capacity-energy-step)–(eq:capacity-one-sided) and iterating yields Ik≤Ak1/2Ak−11/4⋯A11/2kI01/2k.I_k\leq A_k^{1/2}A_{k-1}^{1/4}\cdots A_1^{1/2^k}I_0^{1/2^k}. On {gj>δ}\{g_j>\delta\}, the inequality 1≤gj/δ1\leq g_j/\delta gives (eq:capacity-rate) after taking the supremum over vv.

Finally, I0I_0 is the integral of gj↓0g_j\downarrow0 against the fixed locally finite positive measure TS,k(u,…,u)∧βq−kT_{S,k}(u,\ldots,u)\wedge\beta^{q-k}. The functions gjg_j have a common bound and fixed compact support. Bounded convergence gives I0→0I_0\to0. This last step does not use a claim about varying backgrounds. ◻

Corollary (Local convergence in capacity and quasicontinuity). On a sufficiently small ordinary norm-box neighborhood, let V0⋐ΩV_0\Subset\Omega be a smaller core and let uj↓uu_j\downarrow u in Cb\mathcal C_b on a neighborhood of Ω‾\overline\Omega. Then, for every δ>0\delta>0, Cap⁡S,β,k(V0∩{uj−u>δ};Ω)⟶0.\operatorname{Cap}_{S,\beta,k} (V_0\cap\{u_j-u>\delta\};\Omega)\longrightarrow0. Every u∈Cbu\in\mathcal C_b is locally quasicontinuous for this capacity: for every ϵ>0\epsilon>0, there is an ordinary open G⊂ΩG\subset\Omega with Cap⁡S,β,k(G;Ω)<ϵ\operatorname{Cap}_{S,\beta,k}(G;\Omega)<\epsilon such that u∣V0∖Gu|_{V_0\setminus G} is continuous in the subspace topology. For k=qk=q, this statement uses the intrinsic relative capacity Cap⁡S\operatorname{Cap}_S and needs no trace form.

Proof. Patch the entire decreasing family by the common collar of Lemma “A smooth psh collar”. The patched functions agree with the original ones plus a fixed constant on a neighborhood of V0V_0, and their differences have support in one fixed compact set. Apply Theorem “A quantitative relative capacity estimate” to them.

For quasicontinuity, take continuous psh approximants decreasing to the patched uu, with the same collar. Choose indices jlj_l such that Cap⁡S,β,k({ujl−u>2−l};Ω)<ϵ2−l−1.\operatorname{Cap}_{S,\beta,k} (\{u_{j_l}-u>2^{-l}\};\Omega) <\epsilon 2^{-l-1}. Each exceptional set is ordinary open: ujlu_{j_l} is continuous and uu is upper semicontinuous, so ujl−uu_{j_l}-u is lower semicontinuous. Their union GG is open and has capacity less than ϵ\epsilon by countable subadditivity. On Ω∖G\Omega\setminus G, the chosen subsequence converges uniformly to the patched uu. Its restriction to V0V_0 is the original uu up to the fixed constant. ◻

Remark (Scope of the capacity conclusions). The preceding results concern the Borel capacity defined by the supremum of the constructed measures. They do not establish outer regularity, continuity from above on compact sets, Choquet capacitability, relative extremal functions, or an identification of capacity-zero sets with ambient pluripolar sets. Degeneracy is allowed. In particular, an arbitrary SS can be supported on an ambient pluripolar set, and the relative measures can charge that set.

The relative energy lemma is a direct consequence of the first-stage calculus. The quantitative estimate (eq:capacity-rate) and its local quasicontinuity consequence provide an additional general relative theorem. Their proof depends on the already proved fixed-background symmetry, so they are not used retrospectively in the proof of that symmetry.

Stability under local convergence in capacity

This section is a DIRECT_COROLLARY of the relative calculus, mixed domination, and local quasicontinuity proved above. In particular, it is not used in establishing any of those assertions. The background SS remains fixed. Put q=n−pq=n-p and assume q≥1q\geq1.

Lemma (Top capacity controls each finite coefficient stage). Let η\eta be an original compact smooth test of bidegree (q−r,q−r)(q-r,q-r), 0≤r≤q0\leq r\leq q, presented by unit moment coordinates t1,…,tNt_1,\ldots,t_N on an ordinary domain Ω\Omega. Suppose these coordinates are bounded on Ω\Omega. Let a1,…,ar∈Cb(Ω)a_1,\ldots,a_r\in\mathcal C_b(\Omega) have actual bounds ∣ai∣≤M|a_i|\leq M. There is a constant CηC_\eta, independent of these potentials, such that, for every Borel set E⊂supp⁡ηE\subset\operatorname{supp}\eta, ∣μTS,r(a1,…,ar),η∣(E)≤CηCap⁡S(E;Ω).|\mu_{T_{S,r}(a_1,\ldots,a_r),\eta}|(E) \leq C_\eta\operatorname{Cap}_S(E;\Omega). The constant may depend on MM, the test, and the bounded moment coordinates. For r=0r=0 the tuple is empty.

Proof. Set ρ=12∑a=1Nta2\rho=\frac12\sum_{a=1}^Nt_a^2 and β=Dρ=∑ad′ta∧d′′ta\beta=D\rho=\sum_a d't_a\wedge d''t_a. The function ρ\rho is bounded, smooth and classical psh: after every complete field extension and curve pullback, each unit logarithm is harmonic, and this convex function of their values is subharmonic. Finite coefficient-matrix domination, as in Lemma “Positive controls on a finite test stage”, gives an inequality of measures on the test support, ∣μTS,r(a1,…,ar),η∣≤C TS,r(a1,…,ar)∧βq−r.|\mu_{T_{S,r}(a_1,\ldots,a_r),\eta}| \leq C\,T_{S,r}(a_1,\ldots,a_r)\wedge\beta^{q-r}. This is a measure inequality, not merely a total-mass bound: multiply the positive test inequalities by every nonnegative smooth scalar and use Radon uniqueness. Smooth factors commute with the recursion, so the measure on the right is TS,q(a1,…,ar,ρ,…,ρ⏟q−r).T_{S,q}(a_1,\ldots,a_r, \underbrace{\rho,\ldots,\rho}_{q-r}). Shift the aia_i by MM, and apply the mixed domination inequality (eq:mixed-capacity-domination) in degree qq. The bounds 2M2M and sup⁡Ωρ\sup_\Omega\rho give the required constant. A zero bound contributes a constant differentiated slot, whose product vanishes. No positivity of DρD\rho outside the test’s moment directions is needed. ◻

Lemma (A capacity-controlled multiplier limit). Let Pj→PP_j\to P be positive currents of one fixed bidegree on Ω\Omega, with convergence on all original compact smooth complementary tests. Fix such a compact test η\eta supported in L⋐ΩL\Subset\Omega. Assume that the coefficient measures for PjP_j and PP satisfy ∣μPj,η∣(E), ∣μP,η∣(E)≤CηCap⁡S(E;Ω)(E⊂L Borel).|\mu_{P_j,\eta}|(E),\ |\mu_{P,\eta}|(E) \leq C_\eta\operatorname{Cap}_S(E;\Omega) \qquad(E\subset L\text{ Borel}). Let aj,aa_j,a be real-valued Borel functions, uniformly bounded near LL, with Cap⁡S(L∩{∣aj−a∣>δ};Ω)⟶0(δ>0).\operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};\Omega) \longrightarrow0\qquad(\delta>0). Assume that aa is continuous outside ordinary open sets of arbitrarily small Cap⁡S( ⋅ ;Ω)\operatorname{Cap}_S(\,\cdot\,;\Omega) on a compact neighborhood of LL. Then ⟨ajPj,η⟩⟶⟨aP,η⟩.\langle a_jP_j,\eta\rangle\longrightarrow\langle aP,\eta\rangle.

Proof. Weak convergence of the positive currents bounds their coefficient variations on LL by one number AA. If ∣aj∣,∣a∣≤M|a_j|,|a|\leq M, then ∣∫(aj−a) dμPj,η∣≤δA+2MCηCap⁡S(L∩{∣aj−a∣>δ};Ω).\left|\int(a_j-a)\,d\mu_{P_j,\eta}\right| \leq \delta A+2MC_\eta \operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};\Omega). First let j→∞j\to\infty, and then let δ↓0\delta\downarrow0.

To handle the fixed multiplier aa, choose an ordinary open exceptional set GG of capacity less than ϵ\epsilon. On a compact neighborhood HH of LL, the restriction of aa to H∖GH\setminus G is continuous. The space HH is compact Hausdorff, hence normal. Tietze extension and a cutoff give a continuous function aϵa_\epsilon near LL, bounded by MM, which equals aa on L∖GL\setminus G. Thus ∣∫(a−aϵ) dμPj,η∣,∣∫(a−aϵ) dμP,η∣≤2MCηϵ.\left|\int(a-a_\epsilon)\,d\mu_{P_j,\eta}\right|, \left|\int(a-a_\epsilon)\,d\mu_{P,\eta}\right| \leq 2MC_\eta\epsilon. For fixed aϵa_\epsilon, R, Lemma 3.1 gives convergence. Let ϵ↓0\epsilon\downarrow0 and combine the two estimates. Neither outer regularity of the capacity nor a choice of a different representative for aa is used. ◻

Definition (Local convergence in the relative capacity). We say that real-valued Borel functions aja_j on UU converge to aa locally in SS-capacity if, for every relatively compact polynomial norm-box domain Ω\Omega whose closure is contained in an ordinary algebraic neighborhood W⊂UW\subset U, every compact L⋐ΩL\Subset\Omega, and every δ>0\delta>0, Cap⁡S(L∩{∣aj−a∣>δ};Ω)⟶0.\operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};\Omega) \longrightarrow0. Within any prescribed finite unit-moment presentation one may first choose such a neighborhood WW, and then choose boxes with closure contained in WW. The quantifier covers all these boxes, including their further shrinks. We do not identify this condition with convergence for a single larger-domain capacity: such an identification would require an additional comparison between capacities of domains.

Theorem (Weighted stability in local relative capacity). Let 0≤k≤q0\leq k\leq q. For 0≤i≤k0\leq i\leq k, let ui,j,ui∈Cb(U)u_{i,j},u_i\in\mathcal C_b(U) have one actual uniform bound on every compact subset of UU, independent of jj. Suppose ui,j→uiu_{i,j}\to u_i locally in SS-capacity in the sense of Definition “Local convergence in the relative capacity”. Then u0,jTS,k(u1,j,…,uk,j)⟶u0TS,k(u1,…,uk)u_{0,j}T_{S,k}(u_{1,j},\ldots,u_{k,j}) \longrightarrow u_0T_{S,k}(u_1,\ldots,u_k) on every original compact smooth complementary test. No monotonicity is required. Unweighted products have the same stability.

Proof. Localize a test to finitely many relatively compact norm boxes on which its unit moment coordinates are defined and bounded. The assumed local uniform bounds apply on their closures. On each piece, Lemma “Top capacity controls each finite coefficient stage” controls the coefficient measures of every degree-rr product, 0≤r≤k0\leq r\leq k, uniformly in jj, and also controls its proposed limit.

Each uiu_i is locally quasicontinuous by Corollary “Local convergence in capacity and quasicontinuity”. If that corollary is applied on still smaller boxes, domain monotonicity gives Cap⁡S(G;Ω)≤Cap⁡S(G;Ω′)\operatorname{Cap}_S(G;\Omega)\leq \operatorname{Cap}_S(G;\Omega') for G⊂Ω′⊂ΩG\subset\Omega'\subset\Omega. A finite cover and countable subadditivity therefore give the ordinary open exceptional set required near the test support, with arbitrarily small capacity measured in Ω\Omega. Choose a compact neighborhood of the test support inside the union of the smaller cores; continuity glues on its complement of that exceptional set. This supplies the compact-neighborhood hypothesis of Lemma “A capacity-controlled multiplier limit”.

Now induct on the number of differentiated slots. In degree zero the background is fixed, and Lemma “A capacity-controlled multiplier limit” applies with Pj=P=SP_j=P=S. At each later degree, weighted convergence at the preceding degree passes through DD and gives unweighted convergence of the new products. Apply Lemma “A capacity-controlled multiplier limit” with weight u0,ju_{0,j} to obtain the weighted assertion at this degree. The capacity controls and quasicontinuity just verified supply all its hypotheses. Sum the finitely many localized tests. ◻

Remark. Theorem “Weighted stability in local relative capacity” strengthens the mode of convergence available once the monotone theory has been established. It neither gives compactness of arbitrary bounded psh families in this capacity, nor proves that every weak limit of their Monge–Ampère measures comes from a bounded psh function. Those are different assertions. The result does not change the fixed-background requirement.

The remaining compact-regularity obstruction

This section records a precise reduction, not a proof of outer regularity for the full Bedford–Taylor capacity. The initial capacity estimate concerns decreasing potential sequences. The later nonmonotone stability theorem assumes convergence in capacity; it does not extract a convergent subsequence from an arbitrary normalized family. Compact continuity of the capacity instead permits a different maximizing potential at every index. This distinction is essential.

A measure-family reduction

Proposition (Weak closure and compact outer regularity). Let LL be a compact metrizable space and let M\mathcal M be a nonempty family of positive Radon measures on LL satisfying sup⁡μ∈Mμ(L)<∞.\sup_{\mu\in\mathcal M}\mu(L)<\infty. For a Borel set E⊂LE\subset L, put c(E)=sup⁡μ∈Mμ(E),c∗(K)=inf⁡G⊃KG open in Lc(G)(K⊂L compact).c(E)=\sup_{\mu\in\mathcal M}\mu(E), \qquad c^*(K)=\inf_{\substack{G\supset K\\G\text{ open in }L}}c(G) \quad(K\subset L\text{ compact}). Let M‾ w\overline{\mathcal M}^{\,w} be the weak closure in the space of finite measures. Then:

  1. cc is inner regular on Borel sets: c(E)=sup⁡K⊂EK compactc(K).c(E)=\sup_{\substack{K\subset E\\K\text{ compact}}}c(K).

  2. For every compact K⊂LK\subset L,

    c∗(K)=sup⁡μ∈M‾ wμ(K).c^*(K)=\sup_{\mu\in\overline{\mathcal M}^{\,w}}\mu(K).

  3. The following conditions are equivalent:

    1. c(Kj)↓c(K)c(K_j)\downarrow c(K) whenever Kj↓KK_j\downarrow K are compact;

    2. c∗(K)=c(K)c^*(K)=c(K) for every compact KK;

    3. every μ∈M‾ w\mu\in\overline{\mathcal M}^{\,w} satisfies μ(K)≤c(K)\mu(K)\leq c(K) for every compact KK.

Proof. For the first assertion, Radon inner regularity gives sup⁡μ∈Mμ(E)=sup⁡μ∈Msup⁡K⋐Eμ(K)=sup⁡K⋐Esup⁡μ∈Mμ(K).\sup_{\mu\in\mathcal M}\mu(E) =\sup_{\mu\in\mathcal M} \sup_{K\Subset E}\mu(K) =\sup_{K\Subset E}\sup_{\mu\in\mathcal M}\mu(K). This interchanges two suprema, which is legitimate. It does not interchange a supremum with an infimum over open neighborhoods.

The mass bound makes M‾ w\overline{\mathcal M}^{\,w} compact. Suppose μj∈M\mu_j\in\mathcal M converges weakly to μ\mu. For an open G⊃KG\supset K, the open-set inequality for weak convergence yields μ(K)≤μ(G)≤lim inf⁡jμj(G)≤c(G).\mu(K)\leq\mu(G)\leq\liminf_j\mu_j(G)\leq c(G). Taking the infimum over GG proves that the right side of (eq:compact-capacity-closure) is at most the left side.

Choose decreasing open neighborhoods GjG_j of KK, with compact closures FjF_j, such that Fj↓KF_j\downarrow K and the GjG_j form a neighborhood basis of KK. Choose μj∈M\mu_j\in\mathcal M with μj(Gj)≥c(Gj)−1/j.\mu_j(G_j)\geq c(G_j)-1/j. After passing to a subsequence, μj\mu_j converges weakly to some μ∈M‾ w\mu\in\overline{\mathcal M}^{\,w}. For each fixed ll, all sufficiently large jj satisfy Gj⊂FlG_j\subset F_l, so the closed-set inequality gives lim⁡jc(Gj)≤lim sup⁡jμj(Fl)≤μ(Fl).\lim_j c(G_j) \leq\limsup_j\mu_j(F_l)\leq\mu(F_l). Let l→∞l\to\infty. The measure is finite, hence μ(Fl)↓μ(K)\mu(F_l)\downarrow\mu(K), while lim⁡jc(Gj)=c∗(K)\lim_jc(G_j)=c^*(K). This proves (eq:compact-capacity-closure).

Condition (a) implies (b) by applying it to compact neighborhoods Fj↓KF_j\downarrow K and using c(K)≤c∗(K)≤c(Fj)c(K)\leq c^*(K)\leq c(F_j). Conversely, assume (b) and let Kj↓KK_j\downarrow K. Given ϵ>0\epsilon>0, choose an open G⊃KG\supset K with c(G)≤c(K)+ϵc(G)\leq c(K)+\epsilon. The nested compact sets KjK_j eventually lie in GG; otherwise their intersections with the closed complement of GG would have a common point. Thus lim⁡jc(Kj)≤c(K)+ϵ\lim_jc(K_j)\leq c(K)+\epsilon. The reverse inequality is monotonicity. Finally, equivalence with (c) follows from (eq:compact-capacity-closure). ◻

Status. All assertions of the proposition are DIRECT_COROLLARY elementary reductions from Radon regularity and weak compactness of bounded positive measure families. They locate the missing analytic assertion and are not counted as new BT theorems.

Application to the constructed relative capacity.

Fix its domain Ω\Omega, its background SS, and its trace form β\beta, and write μv=TS,k(v,…,v)∧βn−p−k,v∈Cb(Ω),0≤v≤1.\mu_v=T_{S,k}(v,\ldots,v)\wedge\beta^{n-p-k}, \qquad v\in\mathcal C_b(\Omega),\quad0\leq v\leq1. Choose an ordinary open core W⋐ΩW\Subset\Omega lying in one smaller compact norm box, and an original smooth cutoff χ\chi, with 0≤χ≤10\leq\chi\leq1, supported inside that box, and χ=1\chi=1 on a neighborhood of W‾\overline W. The measures Mχ={χμv:0≤v≤1}\mathcal M_\chi=\{\chi\mu_v:0\leq v\leq1\} have uniformly bounded total mass by relative CLN and can be viewed on one compact metrizable norm-box neighborhood of supp⁡χ\operatorname{supp}\chi. Their capacity on subsets of WW is exactly Cap⁡S,β,k( ⋅ ;Ω)\operatorname{Cap}_{S,\beta,k}(\,\cdot\,;\Omega). Proposition “Weak closure and compact outer regularity” therefore identifies the precise remaining local assertion.

Weak-closure domination (open). If 0≤vj≤10\leq v_j\leq1 are classical psh functions on the fixed domain Ω\Omega, and χμvj→ν\chi\mu_{v_j}\to\nu weakly, then, for every compact K⊂WK\subset W, ν(K)≤Cap⁡S,β,k(K;Ω).\nu(K)\leq \operatorname{Cap}_{S,\beta,k}(K;\Omega).

This condition is weaker than representing ν∣W\nu|_W as the Monge–Ampère measure of one admissible potential. It also retains the original capacity domain: representing the limit by a bounded psh function only on a smaller domain would bound it by that larger domain-restricted capacity, which is insufficient. If weak-closure domination is proved, compact continuity from above and outer regularity on compact sets follow immediately. Outer regularity on all Borel sets and a capacitability theorem still need their own argument; neither is assumed here.

Cases where the reduction closes

Proposition (Finite families and uniformly compact potential families).

  1. If M\mathcal M is finite, its capacity cc is outer regular on every Borel set.

  2. If M\mathcal M is weakly compact, its capacity is continuous from above on compact sets and outer regular on compact sets.

  3. In the relative Bedford–Taylor setting, let F\mathcal F be a compact family, for local uniform convergence, of continuous classical psh functions on Ω\Omega with values in [0,1][0,1]. The capacity obtained by restricting the supremum to v∈Fv\in\mathcal F has the two compact-set regularity properties on every relatively compact core.

Proof. For (1), write M={μ1,…,μN}\mathcal M=\{\mu_1,\ldots,\mu_N\}. Given a Borel EE and ϵ>0\epsilon>0, choose open Gi⊃EG_i\supset E with μi(Gi)≤μi(E)+ϵ\mu_i(G_i)\leq\mu_i(E)+\epsilon. The finite intersection G=⋂iGiG=\bigcap_iG_i is open and satisfies c(G)≤c(E)+ϵc(G)\leq c(E)+\epsilon. Assertion (2) follows directly from Proposition “Weak closure and compact outer regularity”.

For (3), uniform stability of the continuous fixed-background products makes v↦χμvv\mapsto\chi\mu_v continuous from local uniform convergence to weak convergence of finite measures on the compact cutoff support. Its image is compact, so (2) applies. These are restricted-family results. No local-uniform compactness of the entire normalized classical psh class is asserted. ◻

Why the direct extensions fail
Radon regularity is not uniform family regularity.

On L=[0,1]L=[0,1], take A=Q∩(0,1],M={δx:x∈A}.A=\mathbf Q\cap(0,1],\qquad \mathcal M=\{\delta_x:x\in A\}. The set AA is dense in LL, but 0∉A0\notin A. Every measure is Radon and has mass one. Nevertheless, c({0})=0,c([0,1/m])=1,c∗({0})=1.c(\{0\})=0,\qquad c([0,1/m])=1,\qquad c^*(\{0\})=1. The weak closure contains δ0\delta_0, for example as the limit of δ1/j\delta_{1/j}, exactly accounting for the defect in (eq:compact-capacity-closure). Thus positivity, Radon regularity, uniform mass bounds, countable subadditivity, and continuity from below do not by themselves imply compact continuity from above. This is an abstract obstruction to a proposed proof, not a counterexample to the full non-Archimedean BT capacity.

Finite maxima do not supply an admissible upper envelope.

On the ordinary unit disc in AK1,an\mathbf A_K^{1,\mathrm{an}}, the bounded continuous classical psh functions aj=max⁡{−1,j−1log⁡∣T∣}a_j=\max\{-1,j^{-1}\log|T|\} increase pointwise. Their limit is zero at every point with T≠0T\ne0 and is −1-1 at the rigid zero. It is not upper semicontinuous. Thus closure under finite maxima and bounded decreasing limits does not justify taking the pointwise supremum of a bounded family. An upper-semicontinuous envelope theorem, with its all-extension curve condition, remains an additional assertion.

Potential order does not order curvature measures.

Even if a tail envelope is available, vj≤wjv_j\leq w_j does not imply μvj≤μwj\mu_{v_j}\leq\mu_{w_j}. On an annulus crossing the unit circle, a suitable positive rescaling of max⁡{0,log⁡∣T∣}\max\{0,\log|T|\} lies in [0,1][0,1] and has a nonzero positive Laplacian, whereas the larger constant function 11 has zero Laplacian. An envelope route therefore also needs a contact or extremal-measure comparison; pointwise domination alone is insufficient.

Status.

The measure-family reduction and the restricted-family propositions have complete proofs above. They isolate the weak-closure domination lemma as the next analytic target. For the full normalized Cb(Ω)\mathcal C_b(\Omega) family, that lemma, compact continuity from above, and outer regularity remain OPEN. The failed arguments recorded here are not used elsewhere in the report.

Maximum locality and relative comparison

This section uses the fixed-background mixed-current theorem proved above. Thus SS is a fixed symmetric positive original CLD current of bidegree (p,p)(p,p), with d′S=d′′S=0d'S=d''S=0, and all potentials belong to Cb\mathcal C_b. Positivity means the cone used in R, not strong positivity. No varying-background convergence is used.

For a coefficient-Radon current TT and a Borel set EE, write 1ET\mathbf 1_E T for its Borel multiple. In particular, μ1ET,η=1EμT,η\mu_{\mathbf 1_E T,\eta}=\mathbf 1_E\mu_{T,\eta} for every original compact smooth complementary test η\eta. This notation does not assert that EE is ordinarily open, nor that 1ET\mathbf 1_E T is closed.

Theorem (Maximum locality for a fixed background). Let 1≤k≤n−p1\leq k\leq n-p, let u,v,w2,…,wk∈Cb(V)u,v,w_2,\ldots,w_k\in\mathcal C_b(V), and put M=max⁡(u,v)M=\max(u,v) and E={u>v}E=\{u>v\}. Then 1ETS,k(M,w2,…,wk)=1ETS,k(u,w2,…,wk).(ML)\mathbf 1_E T_{S,k}(M,w_2,\ldots,w_k) =\mathbf 1_E T_{S,k}(u,w_2,\ldots,w_k). \tag{ML} The equality holds as coefficient-Radon currents on the original CLD test sheaf. The corresponding identity on {v>u}\{v>u\} holds with vv in the first differentiated slot.

Status. NEW_PROVED consequence of the fixed-background theorem. Its [1][1]-background case was already a DIRECT_COROLLARY of B; see below.

Proof. The assertion is local on ordinary open sets. By B, Theorem 2.5, choose on one ordinary neighborhood simultaneous decreasing continuous semipositive PL approximants uj↓u,vj↓v,wi,j↓wi.u_j\downarrow u,\qquad v_j\downarrow v,\qquad w_{i,j}\downarrow w_i. Finite intersections of the fixed neighborhoods suffice. Set Mj=max⁡(uj,vj),hj=Mj−vj,h=M−v=(u−v)+.M_j=\max(u_j,v_j),\qquad h_j=M_j-v_j, \qquad h=M-v=(u-v)_+. Both MjM_j and vjv_j decrease to locally bounded classical psh functions. The functions hjh_j themselves need not be monotone; no monotonicity of hjh_j is asserted or used.

Write Aj=TS,k(Mj,w2,j,…,wk,j),Bj=TS,k(uj,w2,j,…,wk,j).A_j=T_{S,k}(M_j,w_{2,j},\ldots,w_{k,j}),\qquad B_j=T_{S,k}(u_j,w_{2,j},\ldots,w_{k,j}). The set Ej={uj>vj}E_j=\{u_j>v_j\} is ordinarily open. Ordinary locality gives Aj=BjA_j=B_j there, while hj=0h_j=0 on its complement. Hence, as coefficient-Radon currents, hjAj=hjBj.(1)h_j A_j=h_j B_j. \tag{1} Apply weighted joint decreasing continuity to each of MjAj,vjAj,MjBj,vjBjM_j A_j,v_j A_j,M_j B_j,v_j B_j. Subtracting the resulting limits in (1) gives h(A−B)=0,A=TS,k(M,w2,…,wk),B=TS,k(u,w2,…,wk).(2)h(A-B)=0, \quad A=T_{S,k}(M,w_2,\ldots,w_k),\quad B=T_{S,k}(u,w_2,\ldots,w_k). \tag{2} This passage uses two admissible psh weights, not a convergence theorem for an arbitrary changing Borel weight.

For an original compact test η\eta, multiply (2) by arbitrary original smooth scalar functions. Uniqueness of its Radon coefficient measure gives hμA−B,η=0h\mu_{A-B,\eta}=0. On Em={h≥1/m}E_m=\{h\geq 1/m\}, the Borel function 1/h1/h is bounded. Therefore 1EmμA−B,η=0\mathbf 1_{E_m}\mu_{A-B,\eta}=0. Since E=⋃m≥1EmE=\bigcup_{m\geq1}E_m, it follows that 1EμA−B,η=0\mathbf 1_E\mu_{A-B,\eta}=0. This proves (ML) on every original test, and the local identities glue. Interchanging uu and vv proves the final assertion. ◻

Theorem (Relative comparison with a compact boundary gap). Let m=n−p≥1m=n-p\geq1. For a∈Cb(V)a\in\mathcal C_b(V), put μa=TS,m(a,…,a).\mu_a=T_{S,m}(a,\ldots,a). Let u,v∈Cb(V)u,v\in\mathcal C_b(V). Assume that for every ε>0\varepsilon>0, {u<v−ε}‾ is compact and contained in V.(BG)\overline{\{u<v-\varepsilon\}}\ \text{is compact and contained in }V. \tag{BG} The closure in (BG) is the ordinary topological closure of a Borel set. Then ∫{u<v}dμv ≤ ∫{u<v}dμu.(CP)\int_{\{u<v\}}d\mu_v\ \leq\ \int_{\{u<v\}}d\mu_u. \tag{CP} The integrals are allowed to be infinite. If V⋐UV\Subset U and lim inf⁡x→∂V(u(x)−v(x))≥0\liminf_{x\to\partial V}(u(x)-v(x))\geq0, condition (BG) holds.

Status. NEW_PROVED consequence of the fixed-background theorem. It requires no strictly psh exhaustion and no capacity theorem.

Proof. Fix ε>0\varepsilon>0 and let w=max⁡(u,v−ε)w=\max(u,v-\varepsilon). Form the positive closed current Q=∑j=0m−1TS,m−1(w,…,w⏟j,u,…,u⏟m−1−j).Q=\sum_{j=0}^{m-1} T_{S,m-1}(\underbrace{w,\ldots,w}_{j}, \underbrace{u,\ldots,u}_{m-1-j}). For m=1m=1, this means Q=SQ=S. Multilinearity, symmetry, and the recursive definition give the telescoping identity μw−μu=D((w−u)Q).(3)\mu_w-\mu_u=D\bigl((w-u)Q\bigr). \tag{3} Here w−uw-u is a bounded Borel function on its compact support; the identity is obtained by subtracting two well-defined Borel multiples. It does not require w−uw-u to be psh.

By (BG), the support of (w−u)Q(w-u)Q is contained in a compact subset of VV. Choose an original smooth compact cutoff χ\chi equal to one near that support. Equation (3) and distributional differentiation give ⟨μw−μu,χ⟩=⟨(w−u)Q,Dχ⟩=0.(4)\langle\mu_w-\mu_u,\chi\rangle =\langle(w-u)Q,D\chi\rangle=0. \tag{4} The signed measure μw−μu\mu_w-\mu_u has compact support, since the two potentials agree on the ordinary complement of {u<v−ε}‾\overline{\{u<v-\varepsilon\}}. Thus (4) is its total signed mass. No subtraction of two possibly infinite total masses on VV is made.

Set Aε={u<v−ε},Bε={u=v−ε},Cε={u>v−ε}.A_\varepsilon=\{u<v-\varepsilon\},\qquad B_\varepsilon=\{u=v-\varepsilon\},\qquad C_\varepsilon=\{u>v-\varepsilon\}. The set Aε∪BεA_\varepsilon\cup B_\varepsilon is contained in the compact set {u<v−ε/2}‾\overline{\{u<v-\varepsilon/2\}}, so every measure below has finite mass on it. Maximum locality and symmetry give 1Aεμw=1Aεμv,1Cεμw=1Cεμu.\mathbf 1_{A_\varepsilon}\mu_w =\mathbf 1_{A_\varepsilon}\mu_v, \qquad \mathbf 1_{C_\varepsilon}\mu_w =\mathbf 1_{C_\varepsilon}\mu_u. For the first equality, constants in differentiated slots have zero curvature. Applying (4) and partitioning into these three Borel sets yields μv(Aε)+μw(Bε)=μu(Aε∪Bε).\mu_v(A_\varepsilon)+\mu_w(B_\varepsilon) =\mu_u(A_\varepsilon\cup B_\varepsilon). Positivity therefore implies μv({u<v−ε})≤μu({u≤v−ε}).(5)\mu_v(\{u<v-\varepsilon\}) \leq\mu_u(\{u\leq v-\varepsilon\}). \tag{5} Along any positive sequence ε↓0\varepsilon\downarrow0, both sets in (5) increase to {u<v}\{u<v\}. Continuity from below of positive measures proves (CP).

For the final assertion, the boundary hypothesis means that every boundary point has a neighborhood in which u−v>−εu-v>-\varepsilon. Compactness of ∂V\partial V gives a finite collar of this kind. Its complement in V‾\overline V is compact and contained in VV, proving (BG). ◻

What comparison does not prove.

The theorem does not by itself imply pointwise domination or uniqueness for an arbitrary background. Section “Domination and uniqueness modulo capacity zero” proves precise conclusions modulo capacity zero. For instance, S=0S=0 makes both Monge–Ampère measures zero for all potentials. More generally, relative products can only detect the support and the curvature directions of SS. Domination requires an additional mechanism that detects a strict inequality of potentials; the complex proof uses strictly psh perturbations for this purpose.

The part already implied by B

Proposition (Maximum locality for the previously constructed products). On every analytic space permitted by B, identity (ML) for B’s products Tk=T[1],kT_k=T_{[1],k} already follows from B, Theorems 2.5–2.7, without the fixed-background extension proved in this report. The same is true when the background is itself one of B’s mixed currents.

Status. DIRECT_COROLLARY.

Proof. Repeat the proof of Theorem “Maximum locality for a fixed background” with S=[1]S=[1]. Every input of that proof is already supplied by B: Theorem 2.5 supplies decreasing continuous regularizations on a fixed ordinary neighborhood; Theorem 2.6 supplies ordinary locality and coefficient-Radon products; Theorem 2.7 supplies weighted joint decreasing continuity on all original complementary tests. The last Borel division argument uses the Borel multiplication from B, Definition 2.4 and Lemma 5.1. Thus the proof does not use any theorem first established in this report. If the background is a B mixed current, absorb its fixed potentials into the differentiated slots and use B’s symmetry and weighted continuity in those additional slots. ◻

Domination and uniqueness modulo capacity zero

There is a precise consequence of comparison that does not require a nondegeneracy theorem. It separates the proved measure-theoretic conclusion from the additional condition needed for pointwise order. The results in this section are DIRECT_COROLLARY of Theorem “Relative comparison with a compact boundary gap” and the positive mixed calculus.

Let SS have bidegree (p,p)(p,p) and put q=n−p≥1q=n-p\geq1. On an ordinary open VV, use the top-degree capacity Cap⁡S(E;V)=sup⁡h∈Cb(V)0≤h≤1∫ETS,q(h,…,h),E⊂V Borel.\operatorname{Cap}_S(E;V) =\sup_{\substack{h\in\mathcal C_b(V)\\0\leq h\leq1}} \int_E T_{S,q}(h,\ldots,h), \qquad E\subset V\text{ Borel}. This definition makes sense without a moment frame. Relative CLN gives compact finiteness, and countable subadditivity is unchanged. Write μa=TS,q(a,…,a)\mu_a=T_{S,q}(a,\ldots,a). For two potentials a,ba,b, the compact-gap condition means {a<b−δ}‾⋐Vfor every δ>0,\overline{\{a<b-\delta\}}\Subset V \qquad\text{for every }\delta>0, where the closure is the ordinary topological closure.

Proposition (Domination modulo capacity zero). Let u,v∈Cb(V)u,v\in\mathcal C_b(V) satisfy the compact-gap condition in the order (u,v)(u,v). If μu({u<v})=0\mu_u(\{u<v\})=0, then Cap⁡S({u<v};V)=0.\operatorname{Cap}_S(\{u<v\};V)=0. This does not assert that the strict inequality set is empty.

Proof. Fix δ>0\delta>0, 0<ϵ<δ0<\epsilon<\delta, and any admissible h∈Cb(V)h\in\mathcal C_b(V) with 0≤h≤10\leq h\leq1. Put vδ,ϵ=v+ϵ(h−1)−δ,E={u<vδ,ϵ}.v_{\delta,\epsilon}=v+\epsilon(h-1)-\delta, \qquad E=\{u<v_{\delta,\epsilon}\}. Then {u<v−2δ}⊂E⊂{u<v−δ}.\{u<v-2\delta\}\subset E\subset\{u<v-\delta\}. The pair (u,vδ,ϵ)(u,v_{\delta,\epsilon}) satisfies the compact-gap condition. Comparison and positivity of the mixed expansion give 0≤ϵq∫{u<v−2δ}μh≤∫Eμvδ,ϵ≤∫Eμu=0.0\leq\epsilon^q\int_{\{u<v-2\delta\}}\mu_h \leq\int_E\mu_{v_{\delta,\epsilon}} \leq\int_E\mu_u=0. Constants disappear in differentiated slots, which explains the ϵqμh\epsilon^q\mu_h term. Since hh was arbitrary, the capacity of {u<v−2δ}\{u<v-2\delta\} is zero. Take δ\delta through a sequence decreasing to zero and use countable subadditivity. All sets are Borel; none is assumed ordinary open. ◻

Proposition (Uniqueness modulo capacity zero). Let u,v∈Cb(V)u,v\in\mathcal C_b(V) satisfy the compact-gap condition in both orders. If μu=μv\mu_u=\mu_v as Radon measures, then Cap⁡S({u≠v};V)=0.\operatorname{Cap}_S(\{u\ne v\};V)=0.

Proof. Fix δ>0\delta>0, 0<ϵ<δ0<\epsilon<\delta, and an admissible hh as above. Set uδ,ϵ=u+ϵ(h−1)−δu_{\delta,\epsilon}=u+\epsilon(h-1)-\delta and E={v<uδ,ϵ}E=\{v<u_{\delta,\epsilon}\}. Then {v<u−2δ}⊂E⊂{v<u−δ}⋐V,\{v<u-2\delta\}\subset E\subset\{v<u-\delta\}\Subset V, where the last compact containment refers to its closure. Comparison for (v,uδ,ϵ)(v,u_{\delta,\epsilon}) and positivity yield ∫Eμu+ϵq∫Eμh≤∫Eμuδ,ϵ≤∫Eμv=∫Eμu.\int_E\mu_u+\epsilon^q\int_E\mu_h \leq\int_E\mu_{u_{\delta,\epsilon}} \leq\int_E\mu_v=\int_E\mu_u. The term being cancelled is finite, because EE has compact closure in VV. Hence μh({v<u−2δ})=0\mu_h(\{v<u-2\delta\})=0. Taking the supremum over hh and then a countable union in δ\delta gives capacity zero for {v<u}\{v<u\}. Interchange uu and vv and take a union. ◻

Corollary (A conditional pointwise conclusion). Assume, in addition, the following property for (S,V)(S,V): every nonempty strict gap {a<b−δ}\{a<b-\delta\}, for a,b∈Cb(V)a,b\in\mathcal C_b(V) satisfying the compact-gap condition and δ>0\delta>0, has positive Cap⁡S( ⋅ ;V)\operatorname{Cap}_S(\,\cdot\,;V). Then Proposition “Domination modulo capacity zero” gives u≥vu\geq v pointwise, and Proposition “Uniqueness modulo capacity zero” gives u=vu=v pointwise.

Proof. If a prohibited pointwise inequality held at a point, it would hold there with some positive margin δ\delta. Its strict gap is a subset of the capacity-zero set already proved, contradicting the additional property. ◻

The last corollary has status CONDITIONAL. The condition on strict gaps is a new research obligation for any proposed class of domains and backgrounds; positivity on nonempty ordinary opens alone does not identify it. It fails for S=0S=0, and may fail away from the support of a nonzero background. The propositions retain their literal meaning even when the capacity degenerates.

Counterexamples to unrestricted extensions

All three examples use the original field KK, with its prescribed absolute value. They concern statements stronger than the fixed-background theorem.

A bounded discontinuous classical psh function over KK

Proposition (A sequence of branch wells). There exist an ordinary open disc V⊂AK1,anV\subset\mathbf A_K^{1,\mathrm{an}}, a function w∈Cb(V)w\in\mathcal C_b(V), a Gauss point ξ∈V\xi\in V, and rigid points xi∈Vx_i\in V such that xi⟶ξ,w(ξ)=0,w(xi)=−1.x_i\longrightarrow\xi,\qquad w(\xi)=0,\qquad w(x_i)=-1. The function ww is the pointwise decreasing limit of continuous classical psh functions on one fixed disc, uniformly bounded there.

Proof. Let K~\widetilde K be the finite residue field. Choose pairwise distinct monic irreducible polynomials P~i∈K~[T]\widetilde P_i\in\widetilde K[T], and lift them to monic polynomials Pi∈K∘[T]P_i\in K^\circ[T]. Write di=deg⁡Pid_i=\deg P_i and ai=2−ia_i=2^{-i}, with indices starting at i=1i=1. Thus ∑iai=1\sum_i a_i=1. Choose R>1R>1, put C=log⁡RC=\log R, and work on V={x:∣T(x)∣<R}.V=\{x:|T(x)|<R\}. For example one may take R=∣π∣−1R=|\pi|^{-1}; no normalization of the given absolute value is made. Set fi(x)=aidimax⁡ ⁣{log⁡∣Pi(x)∣,−diai}.(7)f_i(x)=\frac{a_i}{d_i} \max\!\left\{\log|P_i(x)|,-\frac{d_i}{a_i}\right\}. \tag{7} At a zero of PiP_i, the first entry is interpreted as −∞-\infty. Each fif_i is continuous and classical psh: logarithms of analytic functions are classical psh, and this class is closed under maxima with constants and positive scalar multiples (Gubler–Rabinoff, Proposition 4.2).

On {∣T∣≤1}\{|T|\leq1\}, one has −1≤fi≤0-1\leq f_i\leq0. Moreover, at any point of this closed disc at most one fif_i is nonzero. Indeed, for i≠ji\ne j, coprimality of the reductions gives integral polynomials A,B,HA,B,H with 1=APi+BPj+πH.1=A P_i+B P_j+\pi H. If both ∣Pi(x)∣<1|P_i(x)|<1 and ∣Pj(x)∣<1|P_j(x)|<1, evaluation at a point with ∣T(x)∣≤1|T(x)|\leq1 contradicts the non-Archimedean triangle inequality. Consequently every finite sum of the fif_i is bounded below by −1-1 on the closed unit disc.

If 1<∣T(x)∣<R1<|T(x)|<R, monicity and integrality give ∣Pi(x)∣=∣T(x)∣di,fi(x)=ailog⁡∣T(x)∣.|P_i(x)|=|T(x)|^{d_i},\qquad f_i(x)=a_i\log|T(x)|. In particular, fi≤aiCf_i\leq a_i C everywhere on VV. Define wN=C+∑i=1N(fi−aiC)=∑i=1Nfi+C∑i>Nai.(8)w_N=C+\sum_{i=1}^N(f_i-a_i C) =\sum_{i=1}^N f_i+C\sum_{i>N}a_i. \tag{8} The functions wNw_N are continuous classical psh on the same open disc, and wN+1−wN=fN+1−aN+1C≤0,−1≤wN≤C.w_{N+1}-w_N=f_{N+1}-a_{N+1}C\leq0, \qquad -1\leq w_N\leq C. Thus w=lim⁡NwNw=\lim_N w_N is locally bounded classical psh, by the decreasing limit property in Gubler–Rabinoff, Proposition 4.2(4).

Let ξ\xi be the Gauss point of the closed unit disc. Since ∣Pi(ξ)∣=1|P_i(\xi)|=1, equations (7)–(8) give w(ξ)=0w(\xi)=0. Let xix_i be the rigid Berkovich point defined by PiP_i. The irreducibility of P~i\widetilde P_i implies irreducibility of PiP_i over KK. The point xix_i is a point of the original KK-analytic line; using its finite residue extension to describe its seminorm is not a base change of the analytic space. At this point, fi(xi)=−1f_i(x_i)=-1, while fj(xi)=0f_j(x_i)=0 for j≠ij\ne i. Therefore w(xi)=−1w(x_i)=-1.

It remains to verify the convergence of the points. For a nonzero F∈K[T]F\in K[T], multiply by a scalar so that its Gauss norm equals one. Its reduction is then a nonzero polynomial over K~\widetilde K. Only finitely many distinct irreducibles P~i\widetilde P_i divide that reduction. For every other ii, evaluation at xix_i is a unit, so ∣F(xi)∣=∥F∥Gauss|F(x_i)|=\|F\|_{\mathrm{Gauss}}. Hence ∣F(xi)∣→∣F(ξ)∣|F(x_i)|\to|F(\xi)| for every polynomial FF. This is precisely convergence to ξ\xi in the topology of the Berkovich affine line. It follows in particular that ww is discontinuous at ξ\xi. ◻

Corollary (Failure for varying backgrounds). Weak convergence of positive closed backgrounds does not, by itself, imply weighted convergence for a fixed locally bounded classical psh weight. This fails already with no differentiated potential.

Status of the unrestricted assertion. FALSE.

Proof. On the curve in Proposition “A sequence of branch wells”, take Si=δxiS_i=\delta_{x_i} and S=δξS=\delta_\xi. These are symmetric positive top-degree original currents, and are separately d′d'- and d′′d''-closed for degree reasons. The convergence xi→ξx_i\to\xi gives Si→SS_i\to S on every compact original smooth scalar test. Nevertheless, wSi=−δxi⟶−δξ,wS=0.wS_i=-\delta_{x_i}\longrightarrow-\delta_\xi, \qquad wS=0. Thus the failure occurs for p=n=1p=n=1 and k=0k=0, with a constant sequence of admissible weights. It does not involve changing a function on a null set, changing the base field, or changing the test sheaf. ◻

Remark (A strict inequality set need not be ordinary open). The function ww above satisfies w(ξ)=0w(\xi)=0, w(xi)=−1w(x_i)=-1, and xi→ξx_i\to\xi. Thus {w>−1/2}\{w>-1/2\} contains ξ\xi but contains no xix_i, and is not an ordinary open neighborhood of ξ\xi. Both ww and the constant −1/2-1/2 are bounded classical psh functions. This checks explicitly the distinction used in maximum locality.

A nonzero relative product carried by an ambient pluripolar set

Proposition (Concentration on a smooth divisor). For every m≥1m\geq1, there are a smooth (m+1)(m+1)-dimensional ordinary analytic open VV, a nonzero positive closed (1,1)(1,1)-current SS, and bounded continuous classical psh functions u1,…,umu_1,\ldots,u_m such that TS,m(u1,…,um)T_{S,m}(u_1,\ldots,u_m) is a nonzero positive measure supported on an ambient pluripolar set.

Status of universal ambient non-concentration. FALSE, even for continuous potentials and an integration-current background.

Proof. Choose 0<r<1<R0<r<1<R and R0>0R_0>0. With coordinates z0,z1,…,zmz_0,z_1,\ldots,z_m, put V={∣z0∣<R0}×∏i=1m{r<∣zi∣<R},Y={z0=0}∩V.V=\{|z_0|<R_0\}\times \prod_{i=1}^m\{r<|z_i|<R\}, \qquad Y=\{z_0=0\}\cap V. These are ordinary opens in the relevant smooth algebraic analytifications. Let ι:Y↪V\iota:Y\hookrightarrow V be the closed embedding and take S=ι∗[1]YS=\iota_*[1]_Y. This is a symmetric positive closed original CLD current. The set YY is ambient pluripolar, since the nontrivial classical psh function log⁡∣z0∣\log|z_0| equals −∞-\infty precisely there.

Set ui=max⁡{0,log⁡∣zi∣},1≤i≤m.u_i=\max\{0,\log|z_i|\},\qquad 1\leq i\leq m. Each uiu_i is bounded and continuous on VV. Pushforward commutes with distributional DD, and continuous multiplication satisfies the projection formula. Recursion therefore gives TS,m(u1,…,um)=ι∗T[1],mY(u1∣Y,…,um∣Y).(9)T_{S,m}(u_1,\ldots,u_m) =\iota_* T^Y_{[1],m}(u_1|_Y,\ldots,u_m|_Y). \tag{9} The right side can be calculated on (PK1)m(\mathbf P^1_K)^m. For the standard model (PK∘1)m(\mathbf P^1_{K^\circ})^m, take the pullback of O(1)\mathcal O(1) from each of its mm factors, with its standard model metric. In the affine frames, their weights are the displayed uiu_i. The special fiber is irreducible and reduced, and the mixed intersection degree of these mm line bundles is one. The model measure formula, together with R, Lemma 3.3, identifies their original CLD mixed measure with the Dirac mass at the product Gauss point ξY\xi_Y. This point lies in the displayed polyannulus. Restricting to YY and using (9) gives TS,m(u1,…,um)=δι(ξY).T_{S,m}(u_1,\ldots,u_m)=\delta_{\iota(\xi_Y)}. In particular it has positive mass on YY. ◻

Implication for capacity.

An ambient pluripolar set need not be negligible for a relative capacity. Indeed, in this example a positive rescaling of u1+⋯+umu_1+\cdots+u_m lies between zero and one on VV, and its mm-fold curvature relative to SS has a positive contribution on YY. Any non-concentration theorem must impose a condition on the background, or formulate polarity intrinsically on its support.

The unqualified capacity can vanish identically

Proposition (Compact-domain degeneration). Let XX be a connected compact boundaryless KK-analytic space on which the classical maximum principle holds, for example X=PKn,anX=\mathbf P_K^{n,\mathrm{an}} with n≥1n\geq1. Every finite locally bounded classical psh function on XX is constant. Hence sup⁡u∈Cb(X)0≤u≤1∫E(Du)n=0\sup_{\substack{u\in\mathcal C_b(X)\\0\leq u\leq1}} \int_E (Du)^n=0 for every Borel set E⊂XE\subset X.

Status of nondegeneracy on every ordinary domain. FALSE. The calculation itself is a DIRECT_COROLLARY of the classical maximum principle.

Proof. An upper-semicontinuous real function on a compact space attains its maximum MM. Its maximum set F={u=M}F=\{u=M\} is nonempty and closed. Every point of FF is interior because XX is boundaryless. Gubler–Rabinoff, Theorem 6.1, makes uu locally constant at such a point. Thus FF is also open, and connectedness gives F=XF=X. The curvature of a constant vanishes, proving the capacity assertion. ◻

The entire space XX is not pluripolar when the definition excludes identically −∞-\infty local witnesses. Thus an unqualified equivalence between zero capacity and pluripolarity also fails for this capacity. Two different constants have the same zero Monge–Ampère measure; with empty boundary, no Dirichlet uniqueness follows without a normalization or additional geometric data.

Theorem-by-theorem comparison with the classical theory

Scope and source conventions

The comparison below separates the statement of a theorem from the category in which it is proved. All complex statements use their authors’ normalization of ddcdd^c; replacing it by a positive constant multiple changes numerical constants, not the conclusions listed here. All statements proved in this report use D=d′d′′D=d'd'', the prescribed absolute value on the finite extension K/QpK/\mathbf Q_p, and the original compactly supported CLD smooth test forms. No renormalization of the valuation is implicit.

The following source versions have been read and pinned: BT76 and BT82 are the original journal articles; Dem12 is the June 21, 2012 version; CLD12 is arXiv:1204.6277v1; CLD25 is arXiv:1204.6277v2; GR25 is arXiv:2506.13548v2; Wu26 is arXiv:2609.27610v1; GK17 is the arXiv:1406.7637v3 text of the subsequently published article; BGJK21 is arXiv:2102.07392v1. The theorem numbers below refer to these versions. The source register records URLs, hashes, and verification notes. Page numbers are printed page numbers, which agree with PDF page numbers in the quoted classical and CLD files.

The 2025 test-category issue.

CLD25 develops Chapter 8 on its tropical spaces and subsequently applies that formalism to analytic spaces; it also develops harmonic tropicalizations and additional differential structures. Its theorems are recorded as EXTERNAL_KNOWN in their own stated categories. An arbitrary functional on the original CLD12 test sheaf is not automatically a current on an enlarged test sheaf. Accordingly, no CLD25 assertion is used here to extend an arbitrary original current SS without a proved comparison of the relevant current categories. The proofs in this report instead use the original-category inputs R and B. Products arising from common smooth or model data can be compared by their actual definitions; this does not prove an extension theorem for arbitrary currents.

Internal provenance.

R, Theorem 2.9, provides continuous-potential regularization in its stated algebraic setting, with R, Lemma 3.1, supplying the original positive-cone coefficient-Radon and continuous-multiplier control. B, Theorems 2.5–2.7, provide classical-psh regularization and the entire bounded-potential calculus with background [1][1], including weighted joint decreasing continuity. These are INTERNAL_INPUT, not new results of this round. A2, Theorem 1.1, is already an established input under its number-field, integrable-adelic, fixed-boundary hypotheses. It is not an open direction in the present comparison.

Status matrix

Each status below refers to the precise assertion in its cell. In particular, an OPEN entry is not a claim that every special case is unknown in the literature, and a FALSE entry concerns the explicitly unrestricted version. Local capacity and energy refinements proved elsewhere in this report have their own hypotheses; they are not silently identified with a nondegenerate global Bedford–Taylor capacity.

Classical target

Established non-Archimedean baseline

Assessment in this project

1

Locally bounded psh functions and decreasing regularization

GR25 gives the classical curve-test class; B 2.5 identifies it with its decreasing semipositive PL regularization class.

PROVED / INTERNAL_INPUT. Continuous original-current positivity is covered by R in R’s scope; unrestricted distributional positivity is not silently equated with classical psh.

2

Positive closed currents and mixed products

CLD12 treats approachable continuous potentials; CLD25 treats continuous potentials on arbitrary backgrounds in its category; B 2.6 treats [1][1].

Arbitrary fixed original positive closed SS, bounded classical psh potentials: PROVED / NEW_PROVED in the fixed-background theorem. Ordered existence alone is a DIRECT_COROLLARY of R/B.

3

Chern–Levine–Nirenberg estimates

CLD12 5.6.2 gives the smooth estimate; CLD25 8.4.1 gives a relative smooth estimate in its category.

Original coefficient-Radon relative mass control for arbitrary fixed SS: PROVED / DIRECT_COROLLARY. No global strictly psh form is assumed.

4

Symmetry and weighted monotone convergence

B 2.6–2.7 gives the full [1][1]-background result; CLD25 8.7.2 concerns local uniform convergence of continuous potentials.

Fixed arbitrary original SS: PROVED / NEW_PROVED. Unrestricted varying-background weighted stability: FALSE, Corollary “Failure for varying backgrounds”.

5

Locality and maximum principles

Ordinary locality is in B; strict-set maximum locality is already a DIRECT_COROLLARY of B 2.5–2.7. GR25 6.1 gives the classical psh maximum principle.

The extension of Borel maximum locality to arbitrary fixed SS: PROVED / NEW_PROVED, Theorem “Maximum locality for a fixed background”. The set {u>v}\{u>v\} is Borel; ordinary openness is not asserted.

6

Bedford–Taylor capacities

The mixed measures supply candidate set functions; neither a formal supremum nor CLN alone proves nondegeneracy or full Choquet continuity.

The stated Borel axioms and mixed domination are PROVED; the norm-box estimate in Theorem “A quantitative relative capacity estimate” is PROVED / NEW_PROVED. Outer regularity, continuity from above on compact sets, and Choquet capacitability remain OPEN. Nondegeneracy on every analytic domain: FALSE, Proposition “Compact-domain degeneration”. A complete intrinsic domain classification remains OPEN.

7

Quasicontinuity and convergence in capacity

No unrestricted such theorem follows just from weak monotone convergence. The complex energy proof supplies an exact template, with a dependency warning explained below.

Local bounded convergence in capacity and quasicontinuity on smaller cores of sufficiently small norm boxes: PROVED / NEW_PROVED, Corollary “Local convergence in capacity and quasicontinuity”. They use the stated Borel capacity and ordinary open exceptional sets; the decreasing family is defined near the closure of the box. With actual local uniform bounds, convergence in top relative capacity on all prescribed local norm boxes gives nonmonotone weighted stability: PROVED / DIRECT_COROLLARY, Theorem “Weighted stability in local relative capacity”. Full global/unbounded analogues and outer regularity remain OPEN.

8

Pluripolar and negligible sets

CLD25 16.9.2–4 gives integrability and non-concentration for continuous approachable potentials against [1][1]/smooth tests, with hypotheses on the polar witness.

Universal ambient non-concentration for arbitrary SS: FALSE, Proposition “Concentration on a smooth divisor”. The full zero-capacity/polar/ negligible equivalence is OPEN in the intended domain class.

9

Comparison and domination

The compact-gap comparison for [1][1] and B mixed backgrounds is a DIRECT_COROLLARY of B. Wu26 4.15–4.16 gives separate global projective results with its stated hypotheses.

Extension of compact-gap comparison to arbitrary fixed SS: PROVED / NEW_PROVED, Theorem “Relative comparison with a compact boundary gap”. Domination and uniqueness modulo capacity zero are PROVED / DIRECT_COROLLARY, Propositions “Domination modulo capacity zero”–“Uniqueness modulo capacity zero”. Pointwise conclusions require strict-gap capacity detection and are CONDITIONAL; unrestricted arbitrary-background pointwise domination remains FALSE.

10

Relative extremal functions

Global envelope theorems in Wu26 have residue-characteristic and global geometric hypotheses; they do not give the required local theorem over finite extensions of Qp\mathbf Q_p.

Local envelope regularity, equilibrium support, and capacity/mass identities in a suitable domain class: OPEN. Proposed statements under additional barrier/envelope assumptions are CONDITIONAL.

11

Monge–Ampère and Dirichlet problems

CLD supplies the continuous operator and model comparisons; BGJK21 solves specified invariant toric/abelian equations.

The bounded [1][1] operator is PROVED / INTERNAL_INPUT (B); its fixed arbitrary-background extension is also proved here. Appropriate general local Dirichlet existence and uniqueness are OPEN. Unqualified existence/uniqueness on every domain is FALSE.

1. Locally bounded plurisubharmonic functions
Classical statement and location.

Dem12, Chapter I, Definition 5.1 and Theorems 5.4–5.8, pp. 39–40, define psh functions by upper semicontinuity and subharmonicity on complex lines, prove decreasing-limit stability, decreasing smooth convolution regularization on smaller opens, convex increasing composition, and the distributional curvature characterization. In Theorem 5.5 the function is not identically −∞-\infty on any component, and the regularization u∗ρεu*\rho_\varepsilon is defined on Ωε\Omega_\varepsilon, not on an arbitrary unchanged domain. The upper-envelope theorem is I.5.7; its proof uses Choquet’s lemma I.4.23 and the submean inequality. [Dem12, Ch. I, pp. 39–40]

Non-Archimedean comparison.

GR25, Definition 4.1 and Proposition 4.2, pp. 21–22, use pullbacks to analytic curves after every complete valued extension. Their basic properties include maxima, positive linear combinations, logarithms of analytic functions, and decreasing nets. Theorem 4.3, pp. 22–23, relates classical psh PL functions to semipositivity under its good/strict/nontrivial valuation hypotheses. B, Theorem 2.5, is the stronger accepted input needed here: actual pointwise locally bounded classical psh functions admit decreasing continuous semipositive PL approximations on a fixed ordinary neighborhood over the prescribed field.[GR25, §4]

CLD12, Definition 5.5.1, p. 84, uses a distributional definition for continuous functions. Definition 5.6.3 and Remark 5.6.4, p. 87, introduce local uniform psh approachability and explicitly identify the then-missing implication from continuous current positivity to approachability. R, Theorem 2.9, resolves the stated continuous problem in R’s setting. CLD25, Definition 8.1.1, p. 195, instead defines a sheaf by closure under decreasing sequences; Definition 8.1.7 and Proposition 8.1.11, pp. 196–198, specify its continuous class. Theorem names or similar notation do not identify these classes with classical psh without a bridge. [CLD12, §§5.5–5.6][CLD25, §8.1]

Required change.

The present theory fixes B’s actual pointwise class Cb\mathcal C_b. There is no replacement of a function by an almost-everywhere representative: an arbitrary background can charge sets invisible to ambient volume.

2. Positive currents and the recursive product
Classical statement and proof ingredients.

Dem12, III.1.14, p. 133, gives measure coefficients and trace control for positive currents; III.1.23, p. 136, gives local weak compactness under mass bounds. III.3.2, p. 145, states that for locally bounded psh uu and arbitrary positive closed TT, ddc(uT)dd^c(uT) is positive closed. Its proof uses decreasing smooth regularization, bounded convergence for the fixed coefficient measures of TT, and continuity of distributional differentiation. Iteration gives an ordered product before symmetry is known. BT82, equation (2.1), p. 4, records the recursive definition for the [1][1]-background products.[Dem12, III.1.14, III.1.23, III.3.2] [BT82, §2, p. 4]

Non-Archimedean comparison.

CLD12, Definition 5.4.1, pp. 81–82, distinguishes the three positivity cones. Proposition 5.4.6 and Corollary 5.4.7, pp. 83–84, treat top positive currents as Radon measures and continuous multiplication. Corollaries 5.6.5–5.6.6, p. 88, construct approachable continuous mixed products. Paragraph 5.6.8, p. 89, explicitly discusses D(uT)D(uT) for bounded Borel uu and explains that positivity needs appropriate decreasing psh regularization; it is not a completed locally bounded theory. [CLD12, §§5.4, 5.6]

CLD25, Lemma 8.5.5, p. 207, and §8.7.1, p. 209, give continuous recursive products on arbitrary positive closed backgrounds in that formalism. Its Definition 8.6.2, p. 208, calls a current “Bedford–Taylor” if the resulting operation depends only on the curvature of its potential. That extra property is different from the fixed-potential recursion asked for here and is not assumed for our arbitrary SS. [CLD25, §§8.5–8.7]

R’s coefficient-Radon theorem concerns the original positive cone, whereas B’s [1][1]-products are strongly positive. These assertions are not interchangeable in middle degree. The arbitrary-background ordered construction uses R’s exact cone, B’s regularization, bounded convergence against fixed coefficient measures, and continuous-stage positivity. Thus ordered existence is correctly labeled DIRECT_COROLLARY; its later joint continuity and symmetry require further analysis.

3. Relative Chern–Levine–Nirenberg estimates
Classical statement.

For compact L⊂K∘L\subset K^\circ on a complex manifold and fixed positive closed TT, Dem12, III.3.3, p. 146, proves ∥ddcu1∧⋯∧ddcuq∧T∥L≤CK,L∏i=1q∥ui∥L∞(K) ∥T∥K.\|dd^cu_1\wedge\cdots\wedge dd^cu_q\wedge T\|_L \leq C_{K,L}\prod_{i=1}^q\|u_i\|_{L^\infty(K)}\,\|T\|_K. The mass seminorm is the sum of absolute variations of local coefficient measures; its equivalence to a positive trace is explained on p. 145. The proof iterates a cutoff integration by parts and domination of the cutoff’s Hessian by a local positive trace. BT82, Theorem 2.10, p. 10, also records mass, bounded-weight, and energy estimates. [Dem12, III.3.3][BT82, Theorem 2.10]

Non-Archimedean comparison and direct consequence.

CLD12, Proposition 5.6.2, pp. 86–87, proves its smooth estimate. CLD25, Proposition 8.4.1, pp. 202–203, gives a smooth relative version; Proposition 8.4.3, pp. 203–205, gives an additional weighted estimate under a smooth psh defining-function hypothesis. These are not statements that every analytic open has a strictly psh exhaustion. [CLD12, Proposition 5.6.2][CLD25, §8.4]

For the original arbitrary SS, Theorem “Finite-test relative CLN estimate” provides a finite family of positive original tests controlling each requested coefficient test, uniform constants after differentiation and cutoff, and compatibility with tropical refinements. Its status is DIRECT_COROLLARY: finite iteration of R’s positive coefficient control and the ordered recursion supplies the estimate. It is not counted as a new analytic breakthrough. The result controls arbitrary original complementary tests, not just positive scalar tests against top-degree measures.

4. Symmetry and weighted joint decreasing continuity
Classical statement.

BT82, Theorem 2.1 and Corollary 2.2, p. 4, give joint decreasing continuity and symmetry for the [1][1]-products. Theorem 2.4, pp. 7–9, includes a simultaneously decreasing locally bounded psh weight. For arbitrary fixed positive closed TT, Dem12, III.3.7, pp. 147–149, proves u0,j ddcu1,j∧⋯∧ddcuq,j∧T⟶u0 ddcu1∧⋯∧ddcuq∧T,u_{0,j}\,dd^cu_{1,j}\wedge\cdots\wedge dd^cu_{q,j}\wedge T \longrightarrow u_0\,dd^cu_1\wedge\cdots\wedge dd^cu_q\wedge T, when all ui,j↓uiu_{i,j}\downarrow u_i and the limits are locally bounded. III.3.10, p. 149, then deduces symmetry.[BT82, Theorems 2.1, 2.4] [Dem12, III.3.7–3.10]

The decisive classical mechanism.

Dem12 first max-patches every potential to one smooth psh function in a common boundary collar. CLN yields compactness. Lemma III.3.9, p. 148, gives a one-sided inequality for a weak limit of decreasing upper-semicontinuous multipliers times varying positive measures. Integration by parts moves one smooth regularization at a time; ordered limits then give the reverse inequality for the total positive trace. Trace positivity eliminates the defect. This is more than formal integration by parts, and it is not valid to exchange two discontinuous limits without the trace argument. The present proof supplies a test-adapted original-CLD replacement for the required collar and trace.

Baseline and forbidden strengthening.

B, Theorem 2.7, already proves the whole weighted assertion for background [1][1]. CLD25, Proposition 8.7.2, pp. 209–210, instead proves local uniform convergence of continuous potentials with weakly varying backgrounds; Proposition 8.7.4 gives continuous symmetry. The change from uniform to decreasing convergence is substantive. The branch-well counterexample in Corollary “Failure for varying backgrounds” shows why one must keep the background fixed in the bounded discontinuous theorem. [CLD25, Propositions 8.7.2, 8.7.4]

5. Ordinary locality, maximum locality, and maximum principles
Exact locality assertion.

For u,v,w2,…,wku,v,w_2,\ldots,w_k locally bounded psh, the desired maximum identity is 1{u>v} ddcmax⁡(u,v)∧ddcw2∧⋯∧T=1{u>v} ddcu∧ddcw2∧⋯∧T.\mathbf1_{\{u>v\}}\, dd^c\max(u,v)\wedge dd^cw_2\wedge\cdots\wedge T =\mathbf1_{\{u>v\}}\, dd^cu\wedge dd^cw_2\wedge\cdots\wedge T. It follows directly from the verified weighted theorem BT82, Theorem 2.4, by the continuous-approximation and weighted-difference argument proved in Theorem “Maximum locality for a fixed background”. We do not assign it an invented numbered location in BT82. Ordinary-open locality follows immediately from recursion; Borel strict-set locality is a separate statement.

Wu26, Theorem 4.11, p. 27, supplies local maximum locality for its top-degree products. Part (1) in the pinned v1 has MA(v)\mathrm{MA}(v) where the right-hand side on {u>v}\{u>v\} must be MA(u)\mathrm{MA}(u); part (2) has the correct orientation. This textual issue is recorded, not used as a premise. B 2.5–2.7 alone already imply the [1][1] maximum identity by the proof above, including all complementary tests.[Wu26, Theorem 4.11]

The classical local maximum principle for psh functions has a distinct non-Archimedean counterpart: GR25, Theorem 6.1, pp. 30–31, states local constancy at an interior local maximum on the connected situation of that theorem; Theorem 6.4, pp. 33–34, gives a Shilov-boundary maximum principle on affinoids. Neither assertion identifies ordinary opens with sets defined by inequalities of discontinuous psh functions. [GR25, §6]

6. Capacities and their elementary axioms
Classical definition and exact axioms.

BT82, Definition 3.1, p. 11, defines for compact F⊂ΩF\subset\Omega C(F,Ω)=sup⁡0<u<1∫F(ddcu)n,C(E,Ω)=sup⁡F⊂EF compactC(F,Ω).\begin{aligned} C(F,\Omega)&=\sup_{0<u<1}\int_F(dd^cu)^n,\\ C(E,\Omega)&=\sup_{\substack{F\subset E\\F\text{ compact}}}C(F,\Omega). \end{aligned} For Borel EE, the latter agrees with the supremum of the integrals over EE. Proposition 3.2 gives monotonicity, contravariance in the domain, countable subadditivity, and continuity from below for Borel sets. Section 6, p. 22, defines the outer capacity C∗(E)=inf⁡G⊃EG openC(G)C^*(E)=\inf_{\substack{G\supset E\\G\text{ open}}}C(G); Proposition 6.1 gives its elementary properties. [BT82, Definition 3.1, Proposition 3.2, §6]

Theorem 8.2, p. 29, proves the full Choquet axioms on strongly pseudoconvex domains, including continuity along decreasing compact sets and along arbitrary increasing sets. The Choquet theorem stated there then gives capacitability of analytic sets. These last assertions are not formal consequences of defining a supremum over measures. [BT82, Theorem 8.2 and the following Choquet theorem]

Established non-Archimedean properties and remaining requirements.

Proposition “Elementary properties and mixed domination” proves monotonicity, countable subadditivity on Borel sets, compact finiteness, continuity from below on increasing Borel sets, domain contravariance, and mixed-measure domination for Cap⁡S,β,k\operatorname{Cap}_{S,\beta,k}. It uses the constructed positive measures, symmetry, and relative CLN. For this Borel capacity, equality with its outer envelope remains OPEN. Defining a separate outer capacity does not establish that equality or transfer the proved estimates to it. Continuity from above on compact sets and Choquet capacitability also remain OPEN. Nondegeneracy requires enough admissible psh curvature on the chosen domain; it fails identically on a connected proper space for the untwisted capacity, by Proposition “Compact-domain degeneration”. No theorem in the checked CLD12 §§5.4–5.7 or CLD25 Chapter 8 supplies the full bounded-discontinuous relative capacity package on all ordinary domains. A useful local norm-box capacity is a more limited statement and must retain its localization data.

7. Quasicontinuity and convergence in capacity
Classical theorem and dependency.

BT82, Proposition 3.3, p. 12, gives the energy Cauchy–Schwarz inequality. Theorem 3.4, pp. 12–13, proves capacity convergence for decreasing potentials in the common-collar class P∗(Ω,F,ρ,A,B)P^*(\Omega,F,\rho,A,B), defined in (2.2), pp. 4–5, by agreement with Aρ+BA\rho+B off a compact set. The proof repeatedly applies Cauchy–Schwarz and ends by using the already proved weighted Theorem 2.4; see p. 13. Consequently it cannot be imported to prove that weighted theorem in a circle. Theorem 3.5, pp. 13–14, gives quasicontinuity. Its unbounded case uses the sublevel-capacity estimate Theorem 6.10, p. 26. [BT82, Proposition 3.3; Theorems 3.4, 3.5, 6.10]

Theorem 7.2, pp. 26–27, gives quasi-uniform convergence for monotone psh sequences converging almost everywhere to a psh limit. Theorem 7.4, p. 27, gives a stronger mixed continuity theorem with all but one sequence monotone. These require the intervening negligible-set and capacity theory; they are not corollaries of fixed-background weak decreasing continuity alone.[BT82, Theorems 7.2, 7.4]

Transfer achieved in this round and its scope.

After weighted continuity and symmetry have been established independently, Theorem “A quantitative relative capacity estimate” proves the uniform estimate over the relative capacity tests on a polynomial norm box. Its hypotheses include a common smooth psh collar and a compact support for the differences inside a strictly smaller nested box. Corollary “Local convergence in capacity and quasicontinuity” then proves convergence in capacity on a smaller core for a decreasing Cb\mathcal C_b family defined near the closure of the box, and local bounded quasicontinuity with ordinary open exceptional sets. These are PROVED / NEW_PROVED relative results. They concern the defined Borel capacity; no outer-regularity or nondegeneracy assertion is implicit. For k=n−pk=n-p, the trace form β\beta drops out. A general sequence of Borel multipliers cannot replace the admissible psh differences. Unbounded truncation estimates, global capacity nondegeneracy criteria, and the full monotone-a.e. theorem remain OPEN.

Nonmonotone stability now available.

Theorem “Weighted stability in local relative capacity” is a DIRECT_COROLLARY of this established theory. Its hypotheses are actual bounds uniform in the sequence index on every compact set, and convergence of every potential and the psh weight in Cap⁡S( ⋅ ;Ω)\operatorname{Cap}_S(\,\cdot\,;\Omega) on every relatively compact polynomial norm-box domain Ω\Omega whose closure lies in an ordinary algebraic neighborhood in UU. The quantified family includes boxes inside any prescribed finite moment presentation. It proves weighted joint convergence on all original compact smooth complementary tests without monotonicity. It does not assert equivalence with convergence for one larger-domain capacity. On each finite presentation, bounded ρ=12∑ta2\rho=\tfrac12\sum t_a^2 fills the missing differentiated slots and controls lower-degree coefficient variations by the top relative capacity. Quasicontinuity then supplies the fixed-multiplier limit. These arguments use neither outer regularity nor a capacity compactness theorem, and are not used in proving the earlier monotone or capacity estimates.

8. Pluripolar sets and negligible sets
Classical statements.

BT82, Proposition 5.1, p. 18, proves that a negligible set {h<h∗}\{h<h^*\}, where hh is a locally upper bounded psh envelope, has capacity zero. Theorem 6.9, pp. 25–26, identifies outer-capacity zero sets with pluripolar sets on strongly pseudoconvex domains; Corollary 6.6 and Proposition 6.3 supply the equivalent extremal-function condition. Theorem 7.1, p. 26, concludes that negligible sets are pluripolar. Theorem 8.3, p. 29, characterizes pluripolarity of analytic sets by their compact subsets.[BT82, §§5–8]

What CLD already proves.

CLD25, Theorem 8.12.1 and Corollary 8.12.3, pp. 223–224, give integrability and polar non-concentration for continuous locally approachable potentials, assuming suitable smooth psh compact sublevels and an approachable locally integrable polar witness. In the analytic setting, Proposition 16.9.1 and Theorem 16.9.2, p. 466, and Corollaries 16.9.3–16.9.4, p. 467, give the corresponding conclusions for Du1⋯Dum∧αDu_1\cdots Du_m\wedge\alpha, where α\alpha is a smooth complementary form. In particular they prove the indicated measures do not charge lower-dimensional Zariski-closed sets. [CLD25, §8.12, §16.9]

These theorems are not an assertion for an arbitrary positive closed background. Proposition “Concentration on a smooth divisor” gives a positive integration-current background whose nonzero differentiated relative product is entirely carried by an ambient polar divisor. Thus the correct relative theory must impose a condition on SS, use a background-relative notion of negligible set, or make a statement on the support of SS. Even for [1][1], extending the full zero-capacity/polar/negligible equivalence to the intended bounded theory requires envelope and unbounded-witness analysis not supplied by the first-stage calculus.

9. Comparison, domination, and uniqueness
Classical theorem.

BT82, Theorem 4.1, pp. 14–16, proves for bounded psh u,vu,v on a bounded open Ω⊂Cn\Omega\subset\mathbf C^n, with lim inf⁡z→∂Ω(u−v)≥0\liminf_{z\to\partial\Omega}(u-v)\geq0, that ∫{u<v}(ddcv)n≤∫{u<v}(ddcu)n.\int_{\{u<v\}}(dd^cv)^n\leq \int_{\{u<v\}}(dd^cu)^n. Corollaries 4.4–4.5, p. 17, give uniqueness with common boundary data and the domination principle under the stated zero-mass hypothesis. The printed proof of uniqueness uses a strictly psh perturbation and the fact that a nonempty strict inequality set of psh functions has positive ambient volume. This detection step is additional to comparison. [BT82, Theorem 4.1, Corollaries 4.4–4.5]

Non-Archimedean statement now available.

Theorem “Relative comparison with a compact boundary gap” proves the relative inequality for arbitrary fixed original SS when every {u<v−ε}‾\overline{\{u<v-\varepsilon\}} is compactly contained in the domain. Its proof uses Borel maximum locality and the compact-support identity TS,m(wm)−TS,m(um)=D ⁣((w−u)∑j=0m−1TS,m−1(wj,um−1−j)),w=max⁡(u,v−ε).T_{S,m}(w^m)-T_{S,m}(u^m) =D\!\left((w-u)\sum_{j=0}^{m-1} T_{S,m-1}(w^j,u^{m-1-j})\right), \qquad w=\max(u,v-\varepsilon). There is no subtraction of two infinite total masses. However, S=0S=0, or a background missing directions or regions, prevents comparison from implying domination. Wu26, Theorems 4.15–4.16, pp. 29–30, gives separate global projective comparison/domination results with its polarization and envelope-property hypotheses; those are not the arbitrary-background local assertion.[Wu26, Theorems 4.15–4.16]

Comparison already implied by B.

For S=[1]S=[1], the same compact-gap comparison proof uses only B’s symmetry, recursion, coefficient-Radon multiplication, and the maximum locality already derived from B 2.5–2.7. Thus this case is DIRECT_COROLLARY, as is the case where SS is already a B mixed current, by absorbing its fixed potentials into the differentiated slots. The NEW_PROVED assertion is the extension to arbitrary fixed original positive closed backgrounds.

The measure-theoretic domination conclusion.

Proposition “Domination modulo capacity zero” now proves, under the same compact-gap condition in the order (u,v)(u,v), that μu({u<v})=0⟹Cap⁡S({u<v};V)=0.\mu_u(\{u<v\})=0 \quad\Longrightarrow\quad \operatorname{Cap}_S(\{u<v\};V)=0. If μu=μv\mu_u=\mu_v and the compact-gap condition holds in both orders, Proposition “Uniqueness modulo capacity zero” proves Cap⁡S({u≠v};V)=0\operatorname{Cap}_S(\{u\ne v\};V)=0. These are DIRECT_COROLLARY statements for the defined Borel capacity; they do not identify a capacity-zero set with the empty set or an ambient pluripolar set. The additional property that every nonempty strict gap of such potentials has positive relative capacity gives the pointwise conclusions in Corollary “A conditional pointwise conclusion”, which is explicitly CONDITIONAL. Establishing that detection property for a useful domain/background class remains OPEN; positivity of capacity on ordinary open sets alone is insufficient.

10. Relative extremal functions and equilibrium measures
Classical statement.

For E⊂ΩE\subset\Omega, set hE=sup⁡{v∈PSH⁡(Ω):v≤0, v∣E≤−1}.h_E=\sup\{v\in\operatorname{PSH}(\Omega):v\leq0,\ v|_E\leq-1\}. BT82, Proposition 5.3, p. 19, proves for compact EE in a strictly pseudoconvex domain that (ddchE∗)n(dd^ch_E^*)^n is supported on EE and that its total mass equals C(E,Ω)C(E,\Omega). The envelope is psh after upper regularization; the proof also needs increasing convergence (Proposition 5.2), negligible sets, comparison, and local homogeneous Dirichlet replacements. The simultaneous induction is explicit on pp. 19–22. Merely writing the envelope does not prove its equilibrium identity.[BT82, §5, especially Proposition 5.3]

Actual non-Archimedean frontier.

One must first specify ordinary domains, admissible boundary barriers, and a class for which upper regularized envelopes stay classical psh. Then one must prove maximality off the obstacle and identify its Monge–Ampère measure. Wu26’s principal envelope theorem, Theorem 7.1, p. 51, concerns geometrically unibranch projective varieties (in particular normal varieties) over discretely or trivially valued fields of equicharacteristic zero; the residue characteristic excludes the finite extensions of Qp\mathbf Q_p fixed here. B’s accepted local regularization remains available but does not by itself prove the required obstacle-envelope theorem. [Wu26, Theorem 7.1]

11. Monge–Ampère operators and Dirichlet problems
Original classical theorem.

BT76, Theorem D, p. 4, states: on a bounded strictly pseudoconvex Ω⊂Cn\Omega\subset\mathbf C^n, for continuous boundary data φ∈C(∂Ω)\varphi\in C(\partial\Omega) and nonnegative f∈C(Ω‾)f\in C(\overline\Omega), there is a unique u∈PSH⁡(Ω)∩C(Ω‾)u\in\operatorname{PSH}(\Omega)\cap C(\overline\Omega) with the prescribed boundary data and (ddcu)n=f(dd^cu)^n=f, using the paper’s volume-form convention. Its existence proof is in §8, pp. 39–43, and uses the regularity result Theorem C on balls plus spherical modifications. The theorem does not assert solvability for every positive Radon measure.[BT76, Theorems C–D and §8]

Non-Archimedean consequence and remaining problem.

B already supplies the bounded [1][1]-background operator. The present first-stage theorem extends its calculus to arbitrary fixed original backgrounds, and the comparison theorem proves the relative compact-gap inequality; that is necessary but not sufficient for existence or uniqueness. One still needs a domain geometry, barriers, envelope regularity, and a measure class compatible with the actual capacity. BGJK21 proves invariant toric and abelian Monge–Ampère existence results, with specified global geometry and data; it is a model for selected mechanisms, not an arbitrary local Dirichlet theorem. A proper connected untwisted space already shows why empty-boundary uniqueness fails for this formulation. [BGJK21, §§6–8]

Compatibility with tropical, δ\delta-form, and adelic theories

BGJK21, Theorem 3.1.4, pp. 13–15, shows that locally bounded tropical psh functions on its toric partial compactifications are finite and continuous. Theorem 4.1.2, pp. 20–21, gives arbitrary positive closed Lagerberg-background products, with local uniform convergence of the potentials and weak convergence of the background. Theorem 6.1.1, p. 33, compares the corresponding tropical and non-Archimedean continuous products. Its bounded class is therefore not a source of the missing discontinuous-potential argument on general Berkovich spaces. [BGJK21, Theorems 3.1.4, 4.1.2, 6.1.1]

GK17, Proposition 4.15, constructs the differential algebra of δ\delta-forms; its Poincaré–Lelong theorem is Theorem 7.2, and Theorem 10.5 identifies formal-metric mixed measures with Chambert–Loir measures. These statements include nonpositive, nonclosed, and nondiagonal objects beyond the present BT calculus. Agreement of two ordinary currents on the original smooth tests does not imply agreement on every δ\delta-test. On common semipositive PL/model data, ordinary-current agreement is supplied by the continuous-stage identification, R’s model-measure comparison, and recursion; a broader δ\delta-test comparison remains a separate theorem.[GK17, Proposition 4.15, Theorems 7.2, 10.5]

R, Theorems 2.9–2.10 and Lemma 3.3, already supply the stated continuous regularization, positivity, and model/adelic mixed-measure comparisons. A2, Theorem 1.1, already obtains same-model geometric adelic nefness from one-place original-current positivity for the specified integrable arithmetic adelic rational line. Its fixed compactification, full Cartier boundary, prescribed open-line isomorphisms, and number-field hypotheses remain part of that result. None is removed by the local BT extension, and none needs to be proved again in this project.

Dependency discipline.

The useful classical order is regularization and coefficient measures⟹ ordered products and CLN⟹ weighted monotone continuity and symmetry⟹ energy/capacity estimates and maximum locality.\begin{aligned} &\text{regularization and coefficient measures}\\ &\quad\Longrightarrow\ \text{ordered products and CLN}\\ &\quad\Longrightarrow\ \text{weighted monotone continuity and symmetry}\\ &\quad\Longrightarrow\ \text{energy/capacity estimates and maximum locality}. \end{aligned} Extremal functions, pluripolar equivalences, and Dirichlet existence require further inputs. The comparison proof given here follows directly from maximum locality and compact-support mass conservation and does not assume those later inputs. This separation prevents capacity from being used circularly to justify the weighted theorem on which its classical proof already depends.

Minimal remaining problems and the next research step

The fixed-background bounded-potential calculus is proved relative to the accepted inputs. The following questions are not hidden hypotheses of that proof. They concern genuinely later parts of the proposed theory, or a stronger notion of independence than the one requested for the recursive product.

Compact regularity, then envelopes

The first local analytic target is the weak-closure domination statement in Section “The remaining compact-regularity obstruction”. Its quantifiers are important: the normalized potentials live on one fixed domain, their measures need only converge weakly, and the resulting bound must use the capacity of that same domain. It suffices for compact outer regularity and compact continuity from above. It is weaker than a compactness-and-representation theorem for the potentials, and is therefore the more economical initial target.

The next envelope problem is to specify a class of ordinary domains and prove that, for each compact obstacle EE, the function hE∗=(sup⁡{v∈Cb(V):v≤0, v∣E≤−1})∗h_E^*=\left(\sup\{v\in\mathcal C_b(V):v\leq0,\ v|_E\leq-1\}\right)^* belongs to the intended classical psh class, with all complete extensions in its curve test. The upper regularization can change values on EE. One must therefore prove, rather than assume, that the obstacle condition is retained in the required pointwise or capacity sense. Maximality away from EE, support of its equilibrium measure, and a formula relating that measure to capacity are separate steps. The finite-max counterexample in the preceding section explains why a formal supremum is insufficient.

Neither compact regularity alone nor local bounded quasicontinuity has been used to assert Borel outer regularity, negligible-set theorems, or Choquet capacitability. Each of these remains OPEN in the full stated relative setting.

An exact sufficient condition for polar non-concentration

Proposition (A conditional truncation criterion). Let q=n−p≥1q=n-p\geq1, let K0⋐VK_0\Subset V, and let φ:V→[−∞,0]\varphi:V\to[-\infty,0] be a Borel function such that φN=max⁡{φ,−N}∈Cb(V)\varphi_N=\max\{\varphi,-N\}\in\mathcal C_b(V) for every integer N≥1N\geq1. Suppose that one number A<∞A<\infty satisfies

sup⁡N≥1 supv∈Cb(V)0≤v≤1∫K0(−φN)TS,q(v,…,v)≤A.\sup_{N\geq1}\ sup_{\substack{v\in\mathcal C_b(V)\\0\leq v\leq1}} \int_{K_0}(-\varphi_N)T_{S,q}(v,\ldots,v)\leq A.

Then, for every t>0t>0, Cap⁡S(K0∩{φ<−t};V)≤A/t,Cap⁡S(K0∩{φ=−∞};V)=0.\operatorname{Cap}_S(K_0\cap\{\varphi<-t\};V)\leq A/t, \qquad \operatorname{Cap}_S(K_0\cap\{\varphi=-\infty\};V)=0. Every mixed relative measure of bounded potentials whose oscillations are bounded on VV gives zero mass to the latter set.

Proof. Choose an integer N≥tN\geq t. On {φ<−t}\{\varphi<-t\} one has −φN≥t-\varphi_N\geq t. For each admissible vv, Markov’s inequality and (eq:uniform-polar-weight) give t∫K0∩{φ<−t}TS,q(v,…,v)≤A.t\int_{K_0\cap\{\varphi<-t\}}T_{S,q}(v,\ldots,v)\leq A. Take the supremum over vv. The pole set is contained in every sublevel set, so its capacity is at most A/tA/t for every t>0t>0. Let t→∞t\to\infty. The mixed-measure assertion follows from (eq:mixed-capacity-domination), after shifting and normalizing the bounded potentials. ◻

Status. CONDITIONAL. The implication is proved, but the uniform weighted bound is not. The bounded CLN estimate alone gives a bound growing with NN and cannot supply (eq:uniform-polar-weight). A plausible next analytic problem is to derive such a bound from integrability of φ\varphi against a specified finite collection of coefficient controls of SS, together with an appropriate local barrier. The exact sufficient assumptions must be proved. In the integration-current counterexample supported on a polar divisor, −φN=N-\varphi_N=N on the relative mass; the left side grows linearly. Thus an unrestricted assertion for all SS is false, not merely unproved.

From capacity-zero conclusions to pointwise order

Section “Domination and uniqueness modulo capacity zero” proves domination and uniqueness modulo zero capacity. The remaining pointwise problem is to determine for which pairs (S,V)(S,V) the capacity detects every nonempty strict psh gap in Corollary “A conditional pointwise conclusion”. Positivity on nonempty ordinary open sets is not automatically the same assertion: differences of discontinuous psh functions need not have ordinary open strict-sign sets. An appropriate relative notion may be needed if SS is carried by a proper analytic subset.

The local stability theorem in Section “Stability under local convergence in capacity” now handles nonmonotone sequences that converge in the specified local capacity. It does not prove precompactness of arbitrary bounded psh families, nor replace the domain quantifier by a single larger-domain capacity. Those would require additional theorems.

Dirichlet problems and compatibility

A local Dirichlet theorem must begin by choosing the domain class and the boundary on which data are prescribed. Ordinary topological, Shilov, and projective model boundaries serve different purposes. For a proposed domain class, the next concrete tasks are to construct barriers, prove the required envelope property, identify measures controlled by the relative capacity, and establish pointwise uniqueness under a verified detection condition. General existence for arbitrary positive Radon data is not assumed.

There is also an elementary obstruction to a domain-free exact Monge–Ampère existence statement. On a proper boundaryless space, the positive top product has ∫XTS,q(u,…,u)=⟨D(uTS,q−1(u,…,u)),1⟩=0\int_X T_{S,q}(u,\ldots,u) =\left\langle D\bigl(uT_{S,q-1}(u,\ldots,u)\bigr),1\right\rangle =0 for q≥1q\geq1, since 11 is a compactly supported smooth test and D1=0D1=0. Thus no nonzero positive measure is such an exact product there. A curvature class or a suitable local boundary geometry is substantive additional data.

Ordinary-current compatibility on the already identified continuous PL/model overlap is retained. Extending identities to all δ\delta-tests requires its own test-space and multiplication theorem. The full GK algebra includes objects outside the positive closed diagonal setting. A2’s same-model geometric adelic nefness theorem is already established and is not included among the open problems.

Failed routes retained for continuation
  1. FAILED_ATTEMPT: replace fixed-continuous-multiplier convergence by bounded-measurable-multiplier convergence using only uniform masses. The branch-well example disproves the resulting varying-background assertion.

  2. FAILED_ATTEMPT: exchange two discontinuous limits in the smooth Green formula. The proved replacement is the ordered one-parameter trace argument; the order of limits is explicit.

  3. FAILED_ATTEMPT: prove weighted monotone continuity using a capacity convergence theorem whose proof already invokes weighted monotone continuity. The proof in Section “A common collar and weighted monotone convergence” does not use capacity or the later energy measures.

  4. FAILED_ATTEMPT: use a globally strictly positive Euclidean trace on every analytic domain. The replacement is a polynomial norm-box collar and a trace chosen for each finite test presentation.

  5. FAILED_ATTEMPT: represent every symmetric middle-degree test by decomposable positive squares. The proof instead uses R’s positive coefficient-matrix cone and respects its distinction from strong positivity.

  6. FAILED_ATTEMPT: pass compact-support energy identities through pointwise local approximations or multiply a difference of psh functions by an arbitrary smooth partition. The latter need not remain a difference of psh functions. The corrected proof first regularizes on a prescribed compact box and then uses a second collar that leaves the original difference unchanged.

  7. FAILED_ATTEMPT: infer outer regularity of a supremum of measures from the individual Radon properties. The Dirac-family example shows the exact failure; weak-closure domination is the remaining analytic statement.

  8. FAILED_ATTEMPT: treat {u>v}\{u>v\} as an ordinary open set or infer ambient pluripolar non-concentration for every SS. The explicit examples in Section “Counterexamples to unrestricted extensions” rule out these assertions.

First attack fixed-domain weak-closure domination, keeping track of where an envelope/contact statement is actually needed. In parallel, test the exact uniform truncation estimate (eq:uniform-polar-weight) under an explicit SS-integrability hypothesis. Then establish a domain and background criterion that detects strict psh gaps, so that the proved comparison and capacity-zero uniqueness become pointwise statements. Only after those steps should an obstacle-envelope or Dirichlet existence theorem be attempted. The archived main proof and inputs should be reused; no general re-audit of R, B, or A2 is part of this plan.

Upper envelopes supplied by the internal projective theory

Status: DIRECT_COROLLARY. The following assertions use B, Theorems C.6 and 4.4, as accepted inputs. They do not require another proof of envelope continuity or of the internal contact theorem. The first record did not use the full force of these inputs in discussing local envelopes. This section corrects that omission; it is not counted as an independent new regularization theorem.

Lemma (A bounded classical competitor lies below the continuous envelope). Let FF be one of the fields allowed in B, let Y/FY/F be smooth, projective and geometrically integral, and let LL be ample. If rr is a bounded classical metric on LL, ff is a continuous metric on LL, and r≤fr\leq f, then r≤PLfr\leq P_L f at every point.

Proof. By B, Theorem 4.4(2), there are continuous model-semipositive metrics rj↓rr_j\downarrow r on the same identified line. Fix ϵ>0\epsilon>0. The open sets {rj<f+ϵ}\{r_j<f+\epsilon\} increase and cover the compact space YanY^{\mathrm{an}}. Consequently one of them is all of YanY^{\mathrm{an}}. The metric rj−ϵr_j-\epsilon is a continuous model-semipositive minorant of ff and is therefore at most PLfP_Lf. Thus r≤PLf+ϵr\leq P_Lf+\epsilon. Letting ϵ↓0\epsilon\downarrow0 proves the assertion with the actual pointwise representatives. Real translations preserve continuous model-semipositivity; equivalently they can be obtained as uniform limits of the permitted rational valuation translations. No comparison of Monge–Ampère measures is used. ◻

Proposition (Envelopes of bounded upper-semicontinuous obstacles). In the preceding setting, assume YanY^{\mathrm{an}} is metrizable, as it is for the finite pp-adic fields and the finite unit-Gauss towers used here. Let ff be an upper-semicontinuous metric whose difference from one continuous metric is bounded. There is a greatest bounded classical metric PLbf≤fP_L^{\mathrm{b}}f\leq f. If fj↓ff_j\downarrow f are continuous metrics, with a common bounded range relative to a fixed continuous reference metric, then PLbf=lim⁡jPLfj.P_L^{\mathrm{b}}f=\lim_j P_L f_j. In particular, this limit is independent of the continuous majorants.

Proof. Choose a continuous model-semipositive reference metric aa. A bounded upper-semicontinuous function on a compact metric space is the decreasing limit of uniformly bounded continuous functions: for a compatible metric dd, the functions x⟼sup⁡y((f−a)(y)−jd(x,y))x\longmapsto\sup_y\bigl((f-a)(y)-j d(x,y)\bigr) give one such choice. Add aa to obtain fjf_j. By B, Theorem C.6, pj=PLfjp_j=P_L f_j is continuous model-semipositive. The sequence decreases. For a sufficiently large constant MM, the semipositive metric a−Ma-M lies below every fjf_j; thus a−M≤pj≤fja-M\leq p_j\leq f_j, and p=lim⁡jpjp=\lim_jp_j is bounded and classical by decreasing closure. Moreover p≤fp\leq f.

If r≤fr\leq f is any bounded classical metric, then r≤fjr\leq f_j for every jj. Lemma “A bounded classical competitor lies below the continuous envelope” gives r≤pjr\leq p_j; hence r≤pr\leq p. Thus pp is the greatest permitted minorant, which also proves independence. The metrizability used here follows, for these fields, from a countable dense coefficient subfield and polynomial seminorm inequalities with rational radii on a finite affine cover. It does not posit metrizability for arbitrary complete non-Archimedean fields. ◻

Corollary (Upper envelopes of bounded projective families). Let (rα)α∈A(r_\alpha)_{\alpha\in A} be a nonempty family of bounded classical metrics on LL, uniformly bounded relative to one continuous reference metric. Write h=sup⁡αrαh=\sup_\alpha r_\alpha, with its upper semicontinuous regularization h∗h^*. Then h∗h^* is a bounded classical metric on LL.

Proof. The regularization is taken in local frames; changes of frame add a continuous function and therefore commute with this operation. Put p=PLb(h∗)p=P_L^{\mathrm b}(h^*). Every rαr_\alpha is a bounded classical minorant of h∗h^*, so h≤ph\leq p. Since pp is upper semicontinuous, h∗≤ph^*\leq p. The defining obstacle inequality gives p≤h∗p\leq h^*. Thus p=h∗p=h^*. ◻

Theorem (Local bounded upper-envelope theorem). Let UU have the fixed analytic hypotheses of the first record. Let (uα)α∈A(u_\alpha)_{\alpha\in A} be a nonempty family in Cb(U)\mathcal C_b(U) that is locally uniformly bounded above. Then (sup⁡α∈Auα)∗∈Cb(U).\left(\sup_{\alpha\in A}u_\alpha\right)^*\in\mathcal C_b(U). The regularization is the ordinary upper-semicontinuous regularization on the actual Berkovich space. The conclusion includes the complete field-extension and curve tests in the definition of classical psh.

Proof. Work near an arbitrary point. By B, Lemmas D.4–D.5 and the compactification used in Proposition 4.5, work on an ordinary open of a smooth projective geometrically integral variety over an allowed finite pp-adic constant field. Choose the nested compact polynomial norm boxes C1⊂Int⁡Cmid⊂Cmid⊂Int⁡C2⊂UC_1\subset\operatorname{Int}C_{\rm mid} \subset C_{\rm mid}\subset\operatorname{Int}C_2\subset U used in B, Lemma D.2.

Fix one member uα0u_{\alpha_0}. It has a finite lower bound on C2C_2. Choose MM with uα0≥−Mu_{\alpha_0}\geq-M there and replace each uαu_\alpha on a neighborhood of C2C_2 by vα=max⁡{uα,−M}v_\alpha=\max\{u_\alpha,-M\}. These functions have one common two-sided bound on C2C_2, and sup⁡αvα=sup⁡αuαon C2.\sup_\alpha v_\alpha=\sup_\alpha u_\alpha\quad\hbox{on }C_2. The collar construction in the proof of B, Lemma D.2 depends only on these bounds and the chosen boxes. Explicitly, its pure-FS metric qq has a weight bb that is zero on C1C_1 and bounded below by δ>0\delta>0 on the outer collar. Choose a single translation cc and one integer mm with 0<vα+c≤M′0<v_\alpha+c\leq M' and mδ>M′+1m\delta>M'+1 for every α\alpha. On the same ample line and in the same frame, put Rα=max⁡{vα+c,mb}on Int⁡C2,Rα=mqoff Cmid.R_\alpha=\max\{v_\alpha+c,mb\} \quad\hbox{on }\operatorname{Int}C_2, \qquad R_\alpha=mq\quad\hbox{off }C_{\rm mid}. The prescriptions agree on the ordinary overlap. Maximum closure and locality give bounded classical metrics; their differences from mqmq all lie in [0,M′][0,M']. This argument applies to an arbitrary index set: the finiteness assumption in the statement of the cited decreasing family lemma is replaced here by the displayed common bound, and no simultaneous regularization of this family is being presumed.

Corollary “Upper envelopes of bounded projective families” applies to these metrics. On Int⁡C1\operatorname{Int}C_1 their weights are vα+cv_\alpha+c. Upper regularization is local on ordinary open sets; subtracting cc therefore gives the asserted function near the point. All metrics in this argument are classical in the original sense, and restrictions preserve all its field-extension and curve tests. Classicality is local, so the conclusion holds on UU. ◻

Remark (What this result does not assert). The theorem does not say that the unregularized supremum is upper semicontinuous, or that its exceptional set has zero capacity for an arbitrary background. The latter assertion is false, even when the family is an increasing sequence of continuous bounded potentials; see Section “Analytic supports and a relative-envelope obstruction”.

Quasi-open sets and the weak closure of bounded BT families

The fixed field, original CLD test sheaf, positivity cone, and actual point-value convention are those of the frozen first-round report. We use its fixed-background calculus, relative coefficient CLN bounds, local capacity convergence, and local quasicontinuity as inputs. We do not assume outer regularity on arbitrary Borel sets or a theorem about upper envelopes. The capacity domain is fixed throughout each argument.

Fix an ordinary norm-box domain Ω\Omega, a symmetric positive closed background SS of bidegree (p,p)(p,p), and put q=n−p≥1q=n-p\geq1. Write C(E)=Cap⁡S(E;Ω)=sup⁡v∈Cb(Ω)0≤v≤1∫ETS,q(v,…,v)C(E)=\operatorname{Cap}_S(E;\Omega) =\sup_{\substack{v\in\mathcal C_b(\Omega)\\0\leq v\leq1}} \int_E T_{S,q}(v,\ldots,v) for Borel E⊂ΩE\subset\Omega. The same arguments apply to the localized Cap⁡S,β,k\operatorname{Cap}_{S,\beta,k} capacities of the first round. Define its outer regularization by C∗(E)=inf⁡G⊃EG ordinary open in ΩC(G).C^*(E)=\inf_{\substack{G\supset E\\G\text{ ordinary open in }\Omega}} C(G). This is a definition; equality with C(E)C(E) is not presumed.

Outer regularity on the proved class of quasi-open sets

Definition. A Borel set E⊂ΩE\subset\Omega is CC-quasi-open if for every ϵ>0\epsilon>0 there is an ordinary open G⊂ΩG\subset\Omega such that C(G)<ϵC(G)<\epsilon and E∪GE\cup G is ordinary open.

Proposition (Quasi-open sets are outer regular). Every CC-quasi-open Borel set satisfies C∗(E)=C(E)C^*(E)=C(E). Countable unions and finite intersections of CC-quasi-open sets are CC-quasi-open.

Let V⋐ΩV\Subset\Omega be an ordinary open core on which real-valued functions a,ba,b are CC-quasicontinuous in the sense proved in the first round. Then V∩{a>b},V∩{∣a−b∣>δ}(δ>0)V\cap\{a>b\},\qquad V\cap\{|a-b|>\delta\}\quad(\delta>0) are CC-quasi-open, and hence outer regular for CC. In particular this applies to locally bounded classical psh functions on the permitted local cores.

Proof. Monotonicity gives C(E)≤C∗(E)C(E)\leq C^*(E). If E∪GE\cup G is open and C(G)<ϵC(G)<\epsilon, countable subadditivity gives C∗(E)≤C(E∪G)≤C(E)+ϵ.C^*(E)\leq C(E\cup G)\leq C(E)+\epsilon. Letting ϵ↓0\epsilon\downarrow0 proves the first assertion. For a countable family EiE_i, choose exceptional opens GiG_i with C(Gi)<ϵ2−i−1C(G_i)<\epsilon2^{-i-1} and take G=⋃iGiG=\bigcup_iG_i. Then ⋃iEi∪G\bigcup_iE_i\cup G is open. For a finite intersection, choose such a common GG and use (⋂i=1NEi)∪G=⋂i=1N(Ei∪G).\left(\bigcap_{i=1}^N E_i\right)\cup G =\bigcap_{i=1}^N(E_i\cup G).

For the final assertion, choose an ordinary open GG of capacity less than ϵ\epsilon such that both aa and bb are continuous on V∖GV\setminus G. Their strict inequality set there is relatively open in V∖GV\setminus G. Thus there is an ordinary open O⊂VO\subset V such that (V∩{a>b})∖G=O∖G.(V\cap\{a>b\})\setminus G=O\setminus G. Consequently (V∩{a>b})∪G=O∪G(V\cap\{a>b\})\cup G=O\cup G is ordinary open. The proof for ∣a−b∣>δ|a-b|>\delta is identical. Notice that the argument does not declare the original strict inequality sets ordinarily open. ◻

Corollary (Outer-small exceptional sets). If a local strict inequality set between two bounded classical psh functions has zero capacity, it has ordinary open neighborhoods with arbitrarily small capacity. Thus the exceptional sets in the first-round domination and uniqueness statements modulo capacity zero may also be taken to have outer capacity zero.

Proof. Apply Proposition “Quasi-open sets are outer regular” to each strict inequality set. Equality of two functions fails on the union of the two strict inequality sets; use countable subadditivity. ◻

What weak limits already preserve

Lemma (Open-set and quasi-open-set bounds survive weak limits). Let νj\nu_j be positive Radon measures on a relatively compact ordinary region W⊂ΩW\subset\Omega, converging locally weakly to ν\nu. Suppose that for every ordinary open G⊂WG\subset W, νj(G)≤AC(G)\nu_j(G)\leq A C(G) with one finite constant AA. Then ν(G)≤AC(G)\nu(G)\leq A C(G) for such opens, and ν(E)≤AC(E)\nu(E)\leq A C(E) for every CC-quasi-open E⊂WE\subset W whose outer neighborhoods are taken inside WW.

The analogous statement holds for total variations of signed Radon measures converging locally weakly and satisfying the same uniform open-set bounds.

Proof. For positive measures, the weak-convergence inequality on opens gives ν(G)≤lim inf⁡jνj(G)\nu(G)\leq\liminf_j\nu_j(G). For signed measures, use ∣ν∣(G)=sup⁡{∣∫f dν∣:f∈Cc(G), ∣f∣≤1}≤lim inf⁡j∣νj∣(G).|\nu|(G) =\sup\left\{\left|\int f\,d\nu\right|: f\in C_c(G),\ |f|\leq1\right\} \leq\liminf_j|\nu_j|(G). The local form follows by exhaustion with compactly supported continuous tests. For a Borel E⊂WE\subset W, Radon outer regularity of the individual limiting measure gives ∣ν∣(E)≤Ainf⁡G⊃EC(G)=AC∗(E),|\nu|(E)\leq A\inf_{G\supset E} C(G)=A C^*(E), with the neighborhood infimum restricted to WW. Proposition “Quasi-open sets are outer regular” identifies this with AC(E)A C(E) for the stated quasi-open sets. ◻

Corollary (Partial weak-closure domination). Let 0≤vj≤10\leq v_j\leq1 be classical psh functions on the fixed domain Ω\Omega, with no monotonicity assumption. If TS,q(vj,…,vj)⟶νT_{S,q}(v_j,\ldots,v_j)\longrightarrow\nu locally weakly, then ν(E)≤C(E)\nu(E)\leq C(E) on every local CC-quasi-open set. In particular this holds on the strict inequality sets of Proposition “Quasi-open sets are outer regular”. The assertion applies to all members of the local weak closure of the normalized BT measure family, including limits for which no potential representation is known.

Proof. Every measure in the original family satisfies the capacity bound on all Borel sets by the definition of CC. Apply Lemma “Open-set and quasi-open-set bounds survive weak limits” locally. Positive smooth cutoffs equal to one on a prescribed core give finite uniformly bounded measures, so the same argument applies to the weak closure there. ◻

A multiplier theorem for a capacity-controlled weak closure

Theorem (Quasicontinuous multipliers through weak closure). Let Pj→PP_j\to P be positive coefficient-Radon original CLD currents of one bidegree, with convergence on all original compact smooth tests. Fix a compact test η\eta. Suppose that on a compact neighborhood HH of its support there is one finite constant AηA_\eta such that

∣μPj,η∣(G)≤AηC(G)|\mu_{P_j,\eta}|(G)\leq A_\eta C(G)

for every ordinary open GG in a fixed slightly larger region. Let aa be bounded and CC-quasicontinuous on that region. Then ⟨aPj,η⟩⟶⟨aP,η⟩.\langle aP_j,\eta\rangle\longrightarrow\langle aP,\eta\rangle. No Borel capacity bound for PP is required as an extra hypothesis.

More generally, suppose aj,aa_j,a are uniformly bounded, CC-quasicontinuous there, and aj→aa_j\to a locally in CC-capacity on every compact subset of the region. Then ⟨ajPj,η⟩⟶⟨aP,η⟩.\langle a_jP_j,\eta\rangle\longrightarrow\langle aP,\eta\rangle. All multipliers are their actual Borel functions.

Proof. The positive-current coefficient theorem implies weak convergence of the signed coefficient measures against fixed continuous compact scalar tests. Their total variations are locally uniformly bounded. Lemma “Open-set and quasi-open-set bounds survive weak limits” passes (eq:r2-open-coefficient-bound) to μP,η\mu_{P,\eta} on opens.

Let MM bound ∣a∣|a| on HH. For ϵ>0\epsilon>0, choose an ordinary open GG with C(G)<ϵC(G)<\epsilon such that aa is continuous on H∖GH\setminus G. By Tietze extension on the compact Hausdorff neighborhood HH, followed by a cutoff equal to one near supp⁡η\operatorname{supp}\eta, there is a continuous compact scalar aϵa_\epsilon, with ∣aϵ∣≤M|a_\epsilon|\leq M, equal to aa on supp⁡η∖G\operatorname{supp}\eta\setminus G. Therefore ∣∫(a−aϵ) dμPj,η∣+∣∫(a−aϵ) dμP,η∣≤4MAηϵ.\left|\int(a-a_\epsilon)\,d\mu_{P_j,\eta}\right| +\left|\int(a-a_\epsilon)\,d\mu_{P,\eta}\right| \leq4M A_\eta\epsilon. The integrals with aϵa_\epsilon converge by the fixed-continuous multiplier theorem. Let ϵ↓0\epsilon\downarrow0.

For the varying-weight assertion, choose an ordinary open core VV with supp⁡η⊂V⋐H\operatorname{supp}\eta\subset V\Subset H. The Borel set Ej,δ=V∩{∣aj−a∣>δ}E_{j,\delta}=V\cap\{|a_j-a|>\delta\} is CC-quasi-open by Proposition “Quasi-open sets are outer regular”. Individual Radon outer regularity and the open-set bound imply ∣μPj,η∣(Ej,δ)≤AηC(Ej,δ).|\mu_{P_j,\eta}|(E_{j,\delta}) \leq A_\eta C(E_{j,\delta}). The latter capacity tends to zero, because it is bounded by the capacity of the corresponding set on the larger compact HH. If NN bounds the coefficient variations on supp⁡η\operatorname{supp}\eta, then ∣∫(aj−a) dμPj,η∣≤Nδ+2MAηC(Ej,δ).\left|\int(a_j-a)\,d\mu_{P_j,\eta}\right| \leq N\delta+2M A_\eta C(E_{j,\delta}). First let j→∞j\to\infty and then δ↓0\delta\downarrow0. Combine this estimate with the fixed-weight assertion just proved. ◻

Corollary (Fixed additional BT slots on a weak closure). On a fixed local norm-box domain, let Pj=TS,r(v1,j,…,vr,j),0≤vi,j≤1,P_j=T_{S,r}(v_{1,j},\ldots,v_{r,j}),\qquad0\leq v_{i,j}\leq1, and suppose Pj→PP_j\to P on original compact smooth tests. The generating potentials need not converge and need not be monotone. For every fixed bounded classical psh function uu, uPj⟶uP,D(uPj)⟶D(uP)uP_j\longrightarrow uP, \qquad D(uP_j)\longrightarrow D(uP) on all original compact tests, whenever the indicated degrees are allowed. More generally, any fixed tuple of additional bounded classical psh differentiated slots, and a fixed bounded classical psh weight, passes through this weak convergence.

Proof. The coefficient-capacity domination proved in the first round gives (eq:r2-open-coefficient-bound), uniformly in jj, on each compact original test stage. More precisely, first divide a given compact test into finitely many moment presentations and choose the local norm boxes required by that domination statement. The capacities used for these coefficient estimates have those fixed local domains. A domain may depend on the test presentation, but never on jj; this localization makes no comparison assertion between capacities on different domains. The restrictions of all vi,jv_{i,j} remain in [0,1][0,1] there. The given uu is locally quasicontinuous by the first-round theorem. Apply Theorem “Quasicontinuous multipliers through weak closure”, and then use continuity of distributional DD.

For iteration, after one additional slot the approximating currents remain bounded mixed BT currents with the same original background SS. The additional fixed potential has a finite bound on the slightly larger local domain. Positive rescaling and addition of constants put it into a normalized bounded range; the first-round mixed coefficient-capacity estimate therefore remains uniform. The limit is the corresponding positive closed ordered product with background PP. Repeat one slot at a time and finally apply the multiplier theorem to the fixed weight.

This is a capacity-controlled varying-background theorem, not an assertion for arbitrary weakly convergent positive backgrounds. The branch-well/Dirac counterexample in the first round lacks the uniform capacity control required here. ◻

The unresolved compact-set step

The preceding arguments do not establish ν(K)≤C(K)(K compact)\nu(K)\leq C(K) \qquad(K\text{ compact}) for an arbitrary weak limit of normalized BT measures. A compact set is ordinarily closed; it need not be quasi-open. The theorem above does not turn its characteristic function into a quasicontinuous multiplier. The remaining lemma is precisely weak-closure domination on compact sets from the first-round reduction, with the original capacity domain unchanged.

Two proposed stronger shortcuts fail. First, individual quasicontinuity gives a small exceptional open for each potential; even a common exceptional open for a countable family supplies continuity of each member, not equicontinuity or compactness of the normalized family. Second, replacing a sequence by finite or tail maxima does not order its curvature measures. Moreover the tail supremum need not be upper semicontinuous. No upper-envelope or contact theorem is assumed in this section.

The new proved assertions concern quasi-open sets and weak-closure calculus under capacity bounds. Full compact continuity, full Borel outer regularity, and general weak-closure domination for an arbitrary background remain OPEN. For the ambient background on the later exhaustion domains, the extremal arguments in Sections “Ambient extremals on domains with a bounded negative exhaustion” and “Polar witnesses and capacitability from ambient extremals” prove the corresponding compact and relatively compact Borel assertions. They are additional arguments, not consequences of weak convergence alone.

A bounded capacitary closure and its BT calculus

This section introduces an auxiliary class for a specific purpose: an increasing bounded sequence of classical psh functions has an actual Borel limit which need not be upper semicontinuous, and its ambient upper regularization can change values on a set detected by SS. We construct the calculus on limits in outer relative capacity without making either of those replacements. The original class Cb\mathcal C_b and all of its pointwise theorems remain as in the first record. The new class is relative to the fixed current SS.

Put q=n−p≥1q=n-p\geq1. On an ordinary norm-box domain Ω\Omega, write CΩ(E)=Cap⁡S(E;Ω)C_\Omega(E)=\operatorname{Cap}_{S}(E;\Omega) and let CΩ∗C_\Omega^* denote its outer regularization, defined in Section “Quasi-open sets and the weak closure of bounded BT families”. All sets occurring as function level sets below are Borel. The outer capacity is also defined on arbitrary subsets.

Lemma (Elementary outer-capacity facts). The function CΩ∗C_\Omega^* is monotone and countably subadditive, and CΩ∗(G)=CΩ(G)C_\Omega^*(G)=C_\Omega(G) on ordinary open sets. If Ω′⊂Ω\Omega'\subset\Omega is ordinary open and E⊂Ω′E\subset\Omega', then CΩ∗(E)≤CΩ′∗(E).C_\Omega^*(E)\leq C_{\Omega'}^*(E). For decreasing compact sets Kj↓KK_j\downarrow K contained in Ω\Omega, one has CΩ∗(Kj)↓CΩ∗(K).C_\Omega^*(K_j)\downarrow C_\Omega^*(K). The last assertion is about the outer regularization; it does not establish CΩ∗(K)=CΩ(K)C_\Omega^*(K)=C_\Omega(K).

Proof. For countable subadditivity choose open covers whose capacities approximate the respective infima, with errors summing to ϵ\epsilon, and take their union. Monotonicity and the assertion on opens follow directly from the definition. For domain monotonicity, an open G⊂Ω′G\subset\Omega' is also open in Ω\Omega, and every normalized potential on Ω\Omega restricts to an admissible potential on Ω′\Omega'. Thus CΩ(G)≤CΩ′(G)C_\Omega(G)\leq C_{\Omega'}(G), and taking infima proves the claim. Finally, every open neighborhood GG of KK contains all sufficiently large KjK_j, by compactness of K1K_1. Hence lim⁡jCΩ∗(Kj)≤CΩ(G)\lim_j C_\Omega^*(K_j)\leq C_\Omega(G). Take the infimum over GG and use monotonicity for the reverse inequality. ◻

Definition (Local outer-capacity convergence). Real Borel functions aja_j converge to aa locally in outer SS-capacity if, for every norm-box domain Ω\Omega whose compact closure lies in an allowed ordinary algebraic neighborhood, every compact H⋐ΩH\Subset\Omega, and every δ>0\delta>0, CΩ∗(H∩{∣aj−a∣>δ})⟶0.C_\Omega^*(H\cap\{|a_j-a|>\delta\})\longrightarrow0. The Cauchy condition is the corresponding limit with aj−ala_j-a_l, as both indices tend to infinity. We quantify over all these domains, not just one larger domain. Two functions are locally equal quasi-everywhere if the outer capacity of their disagreement set on each such compact HH is zero.

Lemma (Quasicontinuity of capacity limits). Let aja_j be real, locally bounded, locally SS-quasicontinuous functions, and let aa be a locally bounded Borel function. If aj→aa_j\to a locally in outer SS-capacity, then aa is locally SS-quasicontinuous. The conclusion requires neither compact outer regularity of CΩC_\Omega nor nondegeneracy of the capacity.

Proof. Fix H⋐ΩH\Subset\Omega and a slightly larger compact neighborhood H′H' in Ω\Omega. On H′H' choose a subsequence ajla_{j_l} such that CΩ∗(H′∩{∣ajl−a∣>2−l})<ϵ2−l−3.C_\Omega^*(H'\cap\{|a_{j_l}-a|>2^{-l}\}) <\epsilon2^{-l-3}. Choose ordinary open covers GlG_l of these sets with CΩ(Gl)<ϵ2−l−2C_\Omega(G_l)<\epsilon2^{-l-2}. Local quasicontinuity and a finite cover of H′H' give ordinary open sets FlF_l, with the same capacity bound, such that ajla_{j_l} is continuous on H′∖FlH'\setminus F_l. Domain monotonicity allows all these estimates to be measured in the fixed domain Ω\Omega. The union G=⋃l(Gl∪Fl)G=\bigcup_l(G_l\cup F_l) has capacity less than ϵ\epsilon. On H′∖GH'\setminus G the functions ajla_{j_l} are continuous and converge uniformly to aa. Their uniform limit is continuous there. Restriction to HH proves the assertion. ◻

Definition (Bounded capacitary closure). For an ordinary algebraic norm-box domain WW and M<∞M<\infty, let ASM(W)\mathcal A^M_S(W) consist of the Borel functions aa with ∣a∣≤M|a|\leq M which are limits, locally in outer SS-capacity on WW, of sequences aj∈Cb(W),∣aj∣≤Mon W.a_j\in\mathcal C_b(W),\qquad |a_j|\leq M\quad\hbox{on }W. We retain actual Borel representatives but may also take the quotient by local quasi-everywhere equality. Let AS,b(U)\mathcal A_{S,b}(U) be the local class: near each point a member restricts to some ASM(W)\mathcal A^M_S(W) as above. The particular bound and approximating domain may depend on the neighborhood.

Proposition (Existence of capacitary limits). On a norm-box domain WW over the fixed finite pp-adic field, every sequence aj∈Cb(W)a_j\in\mathcal C_b(W) with ∣aj∣≤M|a_j|\leq M which is locally Cauchy in SS-capacity has a Borel limit a∈ASM(W)a\in\mathcal A^M_S(W) locally in outer SS-capacity. It is unique up to local quasi-everywhere equality.

The quotient of ASM(W)\mathcal A^M_S(W) is complete for this local capacity convergence. Addition, nonnegative scalar multiplication, and finite maxima preserve the local class, with the corresponding bounds. Every member is locally quasicontinuous.

Proof. The space WW has a countable basis of norm-box neighborhoods: use a countable dense coefficient subfield on an affine algebraic chart and polynomial seminorm inequalities with rational thresholds. Choose countably many pairs Vl⋐Ωl⋐WV_l\Subset\Omega_l\Subset W of ordinary opens, with Ωl\Omega_l permitted norm boxes, cofinal in the following sense: whenever H⋐Ω⊂WH\Subset\Omega\subset W is a permitted compact/domain pair, finitely many VlV_l cover HH and their corresponding Ωl\Omega_l lie in Ω\Omega. The VlV_l can be chosen to have compact closure in Ωl\Omega_l.

Each gap Vl∩{∣aj−ak∣>δ}V_l\cap\{|a_j-a_k|>\delta\} is quasi-open for CΩlC_{\Omega_l} by the proved quasicontinuity of Cb\mathcal C_b and Proposition “Quasi-open sets are outer regular”. Its outer capacity is therefore its capacity. The assumed Cauchy condition on V‾l\overline V_l lets us choose a subsequence ajma_{j_m} such that, for all l≤ml\leq m, CΩl∗(Vl∩{∣ajm+1−ajm∣>2−m})<2−m.C_{\Omega_l}^*(V_l\cap \{|a_{j_{m+1}}-a_{j_m}|>2^{-m}\})<2^{-m}. For each pair (l,m)(l,m) with l≤ml\leq m, choose an ordinary open cover Gl,m⊂ΩlG_{l,m}\subset\Omega_l of that set with capacity less than 21−m2^{1-m}. For fixed ll the tails Gl(N)=⋃m≥NGl,mG_l^{(N)}=\bigcup_{m\geq N}G_{l,m}, N≥lN\geq l, satisfy CΩl(Gl(N))≤22−N.C_{\Omega_l}(G_l^{(N)})\leq 2^{2-N}. Outside their intersection the subsequence is eventually pointwise Cauchy on VlV_l, since ∑m2−m<∞\sum_m2^{-m}<\infty. Define the Borel function a=lim sup⁡majma=\limsup_m a_{j_m} at every point; it lies in [−M,M][-M,M]. Outside Gl(N)G_l^{(N)}, ∣a−ajN∣≤21−Non Vl.|a-a_{j_N}|\leq 2^{1-N}\quad\hbox{on }V_l. This proves convergence of the subsequence in outer capacity on each VlV_l. The Cauchy condition and the triangle inequality for level sets extend convergence to the entire sequence. Cofinality, finite subadditivity and domain monotonicity then give the stated convergence on every permitted compact/domain pair. Indeed, a cover inside each Ωl⊂Ω\Omega_l\subset\Omega is an ordinary open cover in Ω\Omega with no greater capacity.

Two limits have disagreement sets covered by the countable union of their positive level gaps; the level-set triangle inequality gives zero outer capacity for each such gap. This proves uniqueness. Lemma “Quasicontinuity of capacity limits” gives quasicontinuity.

For completeness, start with a Cauchy sequence in ASM(W)\mathcal A^M_S(W). For its jjth member choose a defining classical approximant within 2−j2^{-j} in capacity, at threshold 2−j2^{-j}, on the first jj compact/domain pairs of a cofinal enumeration. Such a choice is possible by its definition, and every chosen approximant has the same bound MM. The resulting classical sequence is Cauchy on every pair. The preceding construction gives its limit, which is also the limit of the original sequence by the level-set triangle inequality. Finally, sums, nonnegative multiples and maxima of defining classical sequences converge to the corresponding pointwise operations: use the inequalities ∣(a+b)−(c+d)∣≤∣a−c∣+∣b−d∣,∣max⁡(a,b)−max⁡(c,d)∣≤max⁡(∣a−c∣,∣b−d∣).|(a+b)-(c+d)|\leq|a-c|+|b-d|, \qquad |\max(a,b)-\max(c,d)|\leq\max(|a-c|,|b-d|). The classical class has these closure properties. Localizing proves the assertions for AS,b\mathcal A_{S,b}. ◻

Theorem (BT calculus on the bounded capacitary closure). Let SS be the arbitrary fixed original-positive separately closed background. For 0≤k≤q0\leq k\leq q and a1,…,ak∈AS,b(U)a_1,\ldots,a_k\in\mathcal A_{S,b}(U), the recursion TS,0=S,TS,k(a1,…,ak)=D(a1TS,k−1(a2,…,ak))\mathcal T_{S,0}=S,\qquad \mathcal T_{S,k}(a_1,\ldots,a_k) =D\bigl(a_1\mathcal T_{S,k-1}(a_2,\ldots,a_k)\bigr) defines positive, symmetric, separately closed coefficient-Radon currents. It agrees with the original BT product on Cb\mathcal C_b, is independent of defining approximants and of local quasi-everywhere changes of representatives, and is symmetric in all differentiated slots. It is additive and positively homogeneous in those slots and respects restriction and gluing.

If ai,j∈Cb(U)a_{i,j}\in\mathcal C_b(U), 0≤i≤k0\leq i\leq k, have actual uniform bounds on compact subsets and converge locally in outer SS-capacity to locally bounded Borel functions aia_i, then

a0,jTS,k(a1,j,…,ak,j)⟶a0TS,k(a1,…,ak)a_{0,j}T_{S,k}(a_{1,j},\ldots,a_{k,j}) \longrightarrow a_0\mathcal T_{S,k}(a_1,\ldots,a_k)

on all original compact smooth complementary tests. The same stability holds for sequences in the capacitary closure if, locally, their defining approximants admit one common bound.

Proof. Work on a compact test stage, localizing into finitely many smaller norm boxes Ω\Omega where the test’s unit moment coordinates are defined and bounded. R1 finite coefficient domination gives, for every bounded classical approximating product PjP_j and each fixed test η\eta on this stage,

∣μPj,η∣(E)≤CηCΩ(E)(E Borel),|\mu_{P_j,\eta}|(E)\leq C_\eta C_\Omega(E) \quad(E\text{ Borel}),

with a constant depending only on the common bounds and the test stage. In degree zero this holds for Pj=SP_j=S itself, by filling the complementary degree with the bounded quadratic moment potential. If positive currents PjP_j converge to PP, their coefficient measures converge weakly against compact continuous scalars by R. Lower semicontinuity of total variation on opens and individual Radon outer regularity imply

∣μP,η∣(E)≤CηCΩ∗(E)(E Borel).|\mu_{P,\eta}|(E)\leq C_\eta C_\Omega^*(E) \quad(E\text{ Borel}).

This is precisely the weak-closure argument of Lemma “Open-set and quasi-open-set bounds survive weak limits”; it does not replace CΩ∗C_\Omega^* by CΩC_\Omega on compact sets.

We induct on kk. The case k=0k=0 follows from Borel multiplication of SS, the coefficient bound, and the multiplier theorem “Quasicontinuous multipliers through weak closure”. Limits aia_i are quasicontinuous by Lemma “Quasicontinuity of capacity limits”. Suppose the assertions through degree k−1k-1 are known. Weighted convergence there, followed by distributional DD, gives convergence of the degree-kk products to the displayed recursive expression. A weak limit of positive symmetric closed currents retains these properties. The coefficient-Radon theorem of R applies to it, and (eq:r2-limit-coefficient-outer) supplies its outer-capacity control. Apply the multiplier theorem once more to get (eq:r2-completed-weighted-limit) in degree kk. Its varying-weight proof also follows directly from ∣∫(a0,j−a0) dμPj,η∣≤δA+2MCηCΩ∗(H∩{∣a0,j−a0∣>δ}),\left|\int(a_{0,j}-a_0)\,d\mu_{P_j,\eta}\right| \leq\delta A+2M C_\eta C_\Omega^*(H\cap\{|a_{0,j}-a_0|>\delta\}), and continuous extension of a0a_0 off an ordinary exceptional open of arbitrarily small capacity. Here AA bounds coefficient variation on the compact test support and MM bounds the weights.

The recursion uses the actual Borel values of aia_i and therefore does not contain any approximation choice. On overlaps the same recursion gives the same current, proving gluing. Formula (eq:r2-limit-coefficient-outer) shows that changing a slot on a locally outer-capacity-zero set leaves every relevant Borel product unchanged; induction proves representative independence. Symmetry, addition and positive homogeneity follow by taking limits of the corresponding identities for classical defining sequences, with simultaneous finite-slot approximation.

For sequences already in the closure, their coefficient measures obey (eq:r2-limit-coefficient-outer), with uniform constants when the defining bounds are uniform. The same multiplier argument and induction apply. All localizations are over finitely many actual original test stages, so the conclusion holds for every original compact complementary test, including tests not coming from one initially chosen moment map. ◻

Status and role.

The analytic extension theorem is recorded as NEW_PROVED, relative to the R1 fixed-background theory and the R2 weak-closure multiplier lemma. The outer-capacity and completion lemmas are DIRECT_COROLLARY measure-capacity arguments with proofs included. The construction is useful even when CΩC_\Omega is degenerate. It does not assert that every member of AS,b\mathcal A_{S,b} has a classical psh representative on the ambient space, or that outer capacity equals the original Borel capacity. In particular it does not solve the unresolved compact weak-closure domination problem by definition. For the ambient background alone, Corollary “The unique classical representative” later identifies each equivalence class with a unique classical representative; the arbitrary-background distinction remains.

Increasing sequences and relative Aubin–Mabuchi energy

This section uses the frozen first-round calculus and its common-collar Green identity. Let SS be any allowed positive closed original current of bidegree (p,p)(p,p), put q=n−p≥1q=n-p\geq1, and fix one ordinary polynomial norm-box domain Ω\Omega. The capacity is always C(E)=Cap⁡S(E;Ω)C(E)=\operatorname{Cap}_S(E;\Omega), with this same domain.

Choose a bounded smooth classical psh function bb on Ω\Omega and a compact L⋐ΩL\Subset\Omega, contained in a strictly smaller norm box. Let Pb\mathcal P_b denote the locally bounded classical psh functions uu which equal bb on Ω∖L\Omega\setminus L. All families below have a common bound on a fixed compact neighborhood of LL, hence on all of Ω\Omega because of the common collar. Such families arise from the common-collar construction; its maximum operation also preserves increasing sequences.

A compactly supported relative energy

Definition. For u∈Pbu\in\mathcal P_b, define Eb(u)=1q+1∑i=0q∫Ω(u−b)(Du)i∧(Db)q−i∧S.\mathcal E_b(u)=\frac1{q+1}\sum_{i=0}^q \int_\Omega (u-b)(Du)^i\wedge(Db)^{q-i}\wedge S. Every integral is a Borel integral supported in LL. Products use the proved fixed-background calculus; in particular this definition does not require a smooth background or a strict psh exhaustion.

Theorem (Relative energy difference formula). For u,v∈Pbu,v\in\mathcal P_b,

Eb(v)−Eb(u)=1q+1∑i=0q∫Ω(v−u)(Dv)i∧(Du)q−i∧S.\mathcal E_b(v)-\mathcal E_b(u) =\frac1{q+1}\sum_{i=0}^q \int_\Omega(v-u)(Dv)^i\wedge(Du)^{q-i}\wedge S.

Consequently u≤vu\leq v implies Eb(u)≤Eb(v)\mathcal E_b(u)\leq\mathcal E_b(v), and in that case

∫Ω(v−u)(Du)q∧S≤(q+1)(Eb(v)−Eb(u)).\int_\Omega(v-u)(Du)^q\wedge S \leq(q+1)\bigl(\mathcal E_b(v)-\mathcal E_b(u)\bigr).

On a uniformly bounded subset of Pb\mathcal P_b, the energies are uniformly bounded in absolute value.

Proof. Put g=v−ug=v-u and ut=(1−t)u+tvu_t=(1-t)u+tv, for 0≤t≤10\leq t\leq1. All these functions have the same collar. Multilinearity makes F(t)=Eb(ut)F(t)=\mathcal E_b(u_t) a finite polynomial in tt; differentiating it is therefore an algebraic operation on finitely many finite integrals. Write αt=Dut\alpha_t=Du_t and β=Db\beta=Db. The derivative is

(q+1)F′(t)=∑i=0q∫g αtiβq−iS+∑i=1qi∫(ut−b)Dg αti−1βq−iS.\begin{aligned} (q+1)F'(t) &=\sum_{i=0}^q\int g\,\alpha_t^i\beta^{q-i}S\\ &\quad+\sum_{i=1}^q i\int(u_t-b)Dg\, \alpha_t^{i-1}\beta^{q-i}S. \end{aligned}

In each term of the second sum, the background αti−1βq−iS\alpha_t^{i-1}\beta^{q-i}S is positive closed of bidegree (n−1,n−1)(n-1,n-1), and both differences have compact support in the common norm box. The first-round Green identity gives ∫(ut−b)Dg αti−1βq−iS=∫g(αt−β)αti−1βq−iS.\int(u_t-b)Dg\,\alpha_t^{i-1}\beta^{q-i}S =\int g(\alpha_t-\beta)\alpha_t^{i-1}\beta^{q-i}S. The resulting sums telescope, leaving F′(t)=∫g αtqS.F'(t)=\int g\,\alpha_t^qS. Expand αtq=((1−t)Du+tDv)q\alpha_t^q=((1-t)Du+tDv)^q and integrate from zero to one. The identity (qi)∫01ti(1−t)q−i dt=1q+1\binom qi\int_0^1t^i(1-t)^{q-i}\,dt=\frac1{q+1} proves (eq:r2-energy-difference).

If g≥0g\geq0, every summand is nonnegative. Retaining the term i=0i=0 gives (eq:r2-lower-energy-control). Finally, the relative CLN estimate bounds the masses of all mixed products on LL uniformly when the potentials have one common bound. Multiplying by the bounded function u−bu-b gives the asserted energy bound. ◻

Lemma (The pairwise capacity bound). On a uniformly bounded subset of Pb\mathcal P_b, there is a constant AA such that u≤vu\leq v implies, for every δ>0\delta>0,

C({v−u>δ})≤Aδ((q+1)(Eb(v)−Eb(u)))2−q.C(\{v-u>\delta\}) \leq\frac A\delta \left((q+1)\bigl(\mathcal E_b(v)-\mathcal E_b(u)\bigr) \right)^{2^{-q}}.

The same estimate holds with C∗C^* on the left.

Proof. The proof of the first-round quantitative capacity estimate applies to any single pair u≤vu\leq v with compactly supported difference. Indeed its iteration uses only g=v−u≥0g=v-u\geq0, Green’s identity, Cauchy–Schwarz, positivity of D(vQ)D(vQ), and uniform CLN bounds. Before its final limit step it gives C({v−u>δ})≤Aδ(∫(v−u)(Du)qS)2−q.C(\{v-u>\delta\}) \leq\frac A\delta\left(\int(v-u)(Du)^qS\right)^{2^{-q}}. Apply (eq:r2-lower-energy-control). The strict inequality set is quasi-open by the preceding section, so its outer capacity equals its Borel capacity. This proves the last assertion without assuming compact outer regularity. ◻

Increasing limits converge to their actual pointwise limit

Theorem (Increasing sequences are Cauchy in outer capacity). Let uj∈Pbu_j\in\mathcal P_b be increasing and uniformly bounded on the fixed compact neighborhood. Put u∞(x)=sup⁡juj(x),e∞=sup⁡jEb(uj).u_\infty(x)=\sup_j u_j(x),\qquad e_\infty=\sup_j\mathcal E_b(u_j). Then e∞e_\infty is finite. For every δ>0\delta>0,

C∗({u∞−uj>δ})≤Aδ((q+1)(e∞−Eb(uj)))2−q.C^*(\{u_\infty-u_j>\delta\}) \leq\frac A\delta \left((q+1)(e_\infty-\mathcal E_b(u_j))\right)^{2^{-q}}.

The actual pointwise limit u∞u_\infty is locally CC-quasicontinuous on each ordinary core. In particular uj→u∞u_j\to u_\infty in outer capacity on the common difference support, with the original domain Ω\Omega unchanged.

The theorem does not assert that u∞u_\infty is upper semicontinuous, classically psh, or equal outside a capacity-zero set to its upper semicontinuous regularization.

Proof. Theorem “Relative energy difference formula” makes the energies increasing and uniformly bounded. For l≥jl\geq j, apply Lemma “The pairwise capacity bound” to (uj,ul)(u_j,u_l). Its right side is bounded by the right side of (eq:r2-increasing-actual-rate), independently of ll. The strict gap sets increase with ll and their union is {u∞−uj>δ}\{u_\infty-u_j>\delta\}. Continuity from below of the Borel capacity, already proved in the first round, gives the stated bound with CC on the left. Each strict gap set is quasi-open, and countable unions of quasi-open sets are quasi-open. Therefore the union has outer capacity equal to its Borel capacity. This proves (eq:r2-increasing-actual-rate) directly, before any regularity of u∞u_\infty is known.

Here is a direct quasicontinuity argument, with no assumption about upper envelopes. Fix an ordinary core VV and ϵ>0\epsilon>0. Choose a subsequence ujlu_{j_l} so late that the pairwise estimate guarantees C∗({ujl+1−ujl>2−l})<ϵ2−l−3.C^*(\{u_{j_{l+1}}-u_{j_l}>2^{-l}\}) <\epsilon2^{-l-3}. This choice can be made successively from the bound involving e∞−Eb(ujl)e_\infty-\mathcal E_b(u_{j_l}). Choose ordinary open covers GlG_l of these gap sets with C(Gl)<ϵ2−l−2C(G_l)<\epsilon2^{-l-2}. Also use first-round quasicontinuity to choose ordinary opens HlH_l with C(Hl)<ϵ2−l−2C(H_l)<\epsilon2^{-l-2} such that ujlu_{j_l} is continuous on V∖HlV\setminus H_l. The union G=⋃l(Gl∪Hl)G=\bigcup_l(G_l\cup H_l) has capacity less than ϵ\epsilon. On V∖GV\setminus G, the continuous functions ujlu_{j_l} have consecutive increments bounded by 2−l2^{-l}. They therefore converge uniformly there. Their pointwise limit is the specified function u∞u_\infty, so this function is continuous on V∖GV\setminus G.

The convergence assertion now follows from (eq:r2-increasing-actual-rate). ◻

Corollary (Local increasing convergence without a prescribed collar). Let uj∈Cb(U)u_j\in\mathcal C_b(U) increase pointwise and be uniformly bounded on every compact subset of UU. Set u∞=sup⁡juju_\infty=\sup_j u_j, with its actual Borel values. For every permitted norm-box domain Ω⋐U\Omega\Subset U, every compact H⋐ΩH\Subset\Omega, and every δ>0\delta>0, Cap⁡S∗(H∩{u∞−uj>δ};Ω)⟶0.\operatorname{Cap}_S^*(H\cap\{u_\infty-u_j>\delta\};\Omega) \longrightarrow0. Thus the convergence is local outer-capacity convergence in the all-domain sense of Definition “Local outer-capacity convergence”. The limit is locally quasicontinuous and belongs to AS,b(U)\mathcal A_{S,b}(U).

Proof. Fix (H,Ω)(H,\Omega). Cover HH by finitely many ordinary smaller cores VaV_a contained in norm-box domains Ωa⋐Ω\Omega_a\Subset\Omega. On each such domain use the common-collar construction on the whole sequence. Its construction requires a common two-sided bound on a slightly larger box; it does not use the direction of monotonicity. After one fixed shift cac_a, it gives a bounded smooth classical psh barrier bab_a and vj,a=max⁡{uj+ca,ba},v_{j,a}=\max\{u_j+c_a,b_a\}, equal to uj+cau_j+c_a on VaV_a, and equal to bab_a outside a fixed compact subset of Ωa\Omega_a. The maximum preserves increasing order. Its pointwise supremum equals u∞+cau_\infty+c_a on VaV_a. Theorem “Increasing sequences are Cauchy in outer capacity” therefore gives Cap⁡S∗(Va∩{u∞−uj>δ};Ωa)⟶0.\operatorname{Cap}_S^*\bigl( V_a\cap\{u_\infty-u_j>\delta\};\Omega_a\bigr) \longrightarrow0. Domain monotonicity of outer capacity, with Ωa⊂Ω\Omega_a\subset\Omega, and finite subadditivity imply the asserted estimate on HH. This repeats the collar construction inside each requested domain; convergence for one larger capacity domain has not been substituted for convergence on its smaller subdomains.

The actual limit is a locally bounded Borel function. Applying the quasicontinuity lemma for outer-capacity limits gives local quasicontinuity. On each relatively compact neighborhood the original sequence itself is uniformly bounded and gives the defining approximation for AS,b\mathcal A_{S,b}. ◻

Corollary (Weighted increasing limits with actual representatives). Let 0≤k≤q0\leq k\leq q, and suppose that ui,j∈Cb(U)u_{i,j}\in\mathcal C_b(U), 0≤i≤k0\leq i\leq k, increase with jj and have a common bound on every compact subset for each fixed slot. Write ui=sup⁡jui,ju_i=\sup_j u_{i,j} with actual Borel values. Then u0,jTS,k(u1,j,…,uk,j)⟶u0TS,k(u1,…,uk)u_{0,j}T_{S,k}(u_{1,j},\ldots,u_{k,j}) \longrightarrow u_0\mathcal T_{S,k}(u_1,\ldots,u_k) on every original compact smooth complementary test form, where T\mathcal T is the calculus on the bounded capacitary closure. No ambient upper regularization of any uiu_i is made.

Proof. Apply Corollary “Local increasing convergence without a prescribed collar” in each slot, followed by Theorem “BT calculus on the bounded capacitary closure”. These are successive uses: the energy estimate and the local increasing convergence were proved using only the frozen classical-potential calculus, before applying its extension to the capacitary closure. ◻

The precise upper-envelope question

Suppose an upper-envelope theorem supplies a classical psh function U=(sup⁡juj)∗U=(\sup_j u_j)^* on the same local domain. Because all uju_j equal the same smooth bb outside LL, so does UU there. The following formulation separates the available energy argument from the remaining analytic step.

Proposition (The energy no-jump criterion). Under these hypotheses, the following are equivalent:

  1. Eb(U)=sup⁡jEb(uj)\mathcal E_b(U)=\sup_j\mathcal E_b(u_j);

  2. uj→Uu_j\to U locally in the relative capacity, on a core containing the common compact difference support;

  3. U=u∞U=u_\infty outside a set of relative capacity zero on that core.

The zero set in (3) has outer capacity zero, since both functions are quasicontinuous on the core.

Proof. For (1)⇒\Rightarrow(2), apply Lemma “The pairwise capacity bound” to (uj,U)(u_j,U); the energy difference tends to zero. For (2)⇒\Rightarrow(1), use (eq:r2-energy-difference) with (uj,U)(u_j,U). Every measure on its right is a mixed top-degree measure of uniformly bounded psh potentials on the same Ω\Omega. After translating and rescaling the potentials, the first-round mixed-capacity bound gives one uniform bound by a constant times C(E)C(E) on every Borel set EE. Their masses on LL are uniformly bounded by CLN. Split each integral of U−ujU-u_j into the sets where this gap is at most δ\delta and greater than δ\delta. The first part is bounded by a uniform mass times δ\delta, and the second by a uniform bound times C({U−uj>δ}∩L)C(\{U-u_j>\delta\}\cap L). First let j→∞j\to\infty, then δ↓0\delta\downarrow0. This proves energy convergence without changing the capacity domain or invoking a stronger convergence on smaller domains.

Theorem “Increasing sequences are Cauchy in outer capacity” gives capacity convergence to the actual limit u∞u_\infty. If (2) holds, subadditivity applied to the two error-level sets gives C({∣U−u∞∣>δ})=0C(\{|U-u_\infty|>\delta\})=0 for every δ>0\delta>0 on a slightly smaller core. Taking their countable union proves (3). Conversely, (3), subadditivity, and the already proved convergence to u∞u_\infty give (2). Quasicontinuity of the two functions makes the strict error-level sets quasi-open, so their outer capacities also vanish. ◻

For arbitrary backgrounds SS, the upper-envelope identification cannot be assumed: an ambient envelope can raise the limit on a subvariety which carries positive relative capacity. For S=[1]S=[1], the criterion isolates a step which is proved later by the ordinary divisorial barrier in Theorem “Increasing envelopes have no ambient capacity jump”. Continuity and orthogonality for a continuous obstacle in B, Theorem C.6, and the upper contact statement in B, Theorem 4.4, do not by their statements identify that energy. The later barrier proof supplies the additional argument; it was not inserted into the criterion’s proof.

Relative capacitary upper envelopes

The upper envelope constructed here belongs to the bounded capacitary closure. Its order is the order modulo relative outer-capacity-zero sets. The construction does not replace it by an ambient upper semicontinuous regularization. In particular it continues to apply when the background is carried by an ambient pluripolar subvariety.

One common collar and one capacity domain

Use the notation of Section “Increasing sequences and relative Aubin–Mabuchi energy”: q=n−p≥1q=n-p\geq1, a fixed norm box Ω\Omega, a bounded smooth classical psh collar bb, and the class Pb\mathcal P_b with one compact difference support L⋐ΩL\Subset\Omega. Write CΩ∗=Cap⁡S∗( ⋅ ;Ω)C_\Omega^*=\operatorname{Cap}_S^*(\,\cdot\,;\Omega). For Borel functions define a⪯Ωda\preceq_\Omega d to mean CΩ∗({a>d})=0C_\Omega^*(\{a>d\})=0.

Theorem (A capacitary envelope from a maximizing sequence). Let F⊂Pb\mathcal F\subset\mathcal P_b be nonempty and uniformly bounded. Let F∨\mathcal F^\vee be its closure under nonempty finite maxima, and set e∗=sup⁡v∈F∨Eb(v).e_* = \sup_{v\in\mathcal F^\vee}\mathcal E_b(v). There is an increasing sequence uj∈F∨u_j\in\mathcal F^\vee such that Eb(uj)↑e∗\mathcal E_b(u_j)\uparrow e_*. Its actual pointwise limit H=sup⁡jujH=\sup_j u_j belongs to AS,b(Ω)\mathcal A_{S,b}(\Omega) and has the following properties:

  1. For each v∈Fv\in\mathcal F, one has v⪯ΩHv\preceq_\Omega H.

  2. If a Borel function hh satisfies v⪯Ωhv\preceq_\Omega h for every v∈Fv\in\mathcal F, then H⪯ΩhH\preceq_\Omega h.

  3. The equivalence class of HH modulo CΩ∗C_\Omega^*-zero sets is independent of the maximizing sequence.

Moreover H=bH=b outside LL, and the extended BT calculus gives Eb(H)=e∗\mathcal E_b(H)=e_*, with the same finite-sum definition of the energy.

Proof. The energy is uniformly bounded on F∨\mathcal F^\vee by the common bounds and CLN. Choose vj∈F∨v_j\in\mathcal F^\vee with Eb(vj)>e∗−2−j\mathcal E_b(v_j)>e_*-2^{-j}, and put uj=max⁡(v1,…,vj)u_j=\max(v_1,\ldots,v_j). Monotonicity of the energy gives e∗−2−j<Eb(vj)≤Eb(uj)≤e∗.e_*-2^{-j}<\mathcal E_b(v_j)\leq\mathcal E_b(u_j)\leq e_*. The preceding increasing convergence theorem, including its all-domain local version, applies to HH and gives membership in the capacitary closure.

Fix v∈Fv\in\mathcal F and put wj=max⁡(uj,v)w_j=\max(u_j,v). These functions are in F∨\mathcal F^\vee, and 0≤Eb(wj)−Eb(uj)≤e∗−Eb(uj)⟶0.0\leq\mathcal E_b(w_j)-\mathcal E_b(u_j) \leq e_*-\mathcal E_b(u_j)\longrightarrow0. For every δ>0\delta>0, the pairwise energy estimate implies CΩ∗({v>H+δ})≤CΩ∗({wj−uj>δ})⟶0.C_\Omega^*(\{v>H+\delta\}) \leq C_\Omega^*(\{w_j-u_j>\delta\})\longrightarrow0. Taking the countable union over δ=1/m\delta=1/m proves (1).

Every uju_j is a finite maximum of members of F\mathcal F. If hh is as in (2), then uj≤hu_j\leq h off a finite union of outer-capacity-zero sets. Taking the countable union of these exceptional sets over jj gives H⪯ΩhH\preceq_\Omega h. Applying (2) to two constructed upper bounds proves (3).

Finally, uj−bu_j-b and H−bH-b vanish outside LL. Choose a smooth compact cutoff equal to one near LL. Apply weighted increasing BT convergence separately to the two weights uju_j and bb in each summand of Eb(uj)\mathcal E_b(u_j). The cutoff makes this a finite collection of original compact smooth tests. Their limits are precisely the summands of Eb(H)\mathcal E_b(H), proving the last identity. ◻

The exceptional set in (1) is allowed to depend on vv. For an uncountable family, (1) does not assert that the pointwise supremum of the entire family is bounded by HH outside one common exceptional set.

A countable energy construction for the local order

The preceding theorem uses one fixed capacity domain. The following construction enforces the estimates in every smaller domain, without assuming equivalence of capacities on different domains.

Let WW be an allowed ordinary norm-box domain over the fixed finite pp-adic field. Define a⪯S,locda\preceq_{S,\mathrm{loc}}d when, for every permitted H0⋐Ω⊂WH_0\Subset\Omega\subset W, Cap⁡S∗(H0∩{a>d};Ω)=0.\operatorname{Cap}_S^*(H_0\cap\{a>d\};\Omega)=0.

Theorem (Local relative capacitary envelope). Let F⊂Cb(W)\mathcal F\subset\mathcal C_b(W) be nonempty and uniformly bounded above on each compact subset of WW. There is an increasing sequence uj∈F∨u_j\in\mathcal F^\vee whose actual Borel supremum HH belongs to AS,b(W)\mathcal A_{S,b}(W) and is the least upper bound of F\mathcal F for ⪯S,loc\preceq_{S,\mathrm{loc}}: v⪯S,locH(v∈F),[ v⪯S,loch for all v∈F ]⟹H⪯S,loch.v\preceq_{S,\mathrm{loc}}H\quad(v\in\mathcal F), \qquad \bigl[\ v\preceq_{S,\mathrm{loc}}h\ \text{for all }v\in\mathcal F\ \bigr] \Longrightarrow H\preceq_{S,\mathrm{loc}}h. Its local quasi-everywhere class is independent of all choices.

Proof. Fix v0∈Fv_0\in\mathcal F and first replace the family by {max⁡(v,v0):v∈F}\{\max(v,v_0):v\in\mathcal F\}. This family is uniformly bounded on every compact subset, since v0v_0 supplies a common lower bound. It has exactly the same local quasi-everywhere upper bounds as the original family, and every finite maximum of its members is a finite maximum of original members. We may therefore assume the two-sided bounds in the rest of the construction.

Choose countably many pairs Va⋐Ωa⋐WV_a\Subset\Omega_a\Subset W cofinal for compact/domain pairs, as in Proposition “Existence of capacitary limits”. Thus every H0⋐Ω⊂WH_0\Subset\Omega\subset W is covered by finitely many VaV_a whose corresponding Ωa\Omega_a lie in Ω\Omega. The common bound of F∨\mathcal F^\vee on a slightly larger box lets us choose, for each aa, one fixed collar operation Pa(v)=max⁡{v+ca,ba}on Ωa.P_a(v)=\max\{v+c_a,b_a\}\quad\text{on }\Omega_a. It equals v+cav+c_a on VaV_a and the bounded smooth psh function bab_a outside one fixed compact subset of Ωa\Omega_a. Its construction is uniform over F∨\mathcal F^\vee, and Pa(max⁡(v,w))=max⁡(Pa(v),Pa(w)).P_a(\max(v,w))=\max(P_a(v),P_a(w)). Write Ea(v)=Eba(Pa(v))E_a(v)=\mathcal E_{b_a}(P_a(v)). The relative energy theorem gives a finite Ba≥1B_a\geq1 with ∣Ea(v)∣≤Ba|E_a(v)|\leq B_a on F∨\mathcal F^\vee. Define Φ(v)=∑a=1∞2−aEa(v)+Ba2Ba+1.\Phi(v)=\sum_{a=1}^{\infty}2^{-a} \frac{E_a(v)+B_a}{2B_a+1}. The series converges uniformly, takes values in [0,1][0,1], and is monotone for pointwise order.

Put e=sup⁡v∈F∨Φ(v)e=\sup_{v\in\mathcal F^\vee}\Phi(v). Choose vj∈F∨v_j\in\mathcal F^\vee with Φ(vj)>e−2−j\Phi(v_j)>e-2^{-j} and set uj=max⁡(v1,…,vj)u_j=\max(v_1,\ldots,v_j). Then Φ(uj)↑e\Phi(u_j)\uparrow e. The functions uju_j are locally uniformly bounded, so their actual limit HH belongs to AS,b(W)\mathcal A_{S,b}(W) by Corollary “Local increasing convergence without a prescribed collar”.

Fix v∈Fv\in\mathcal F and set wj=max⁡(uj,v)w_j=\max(u_j,v). For each aa the difference Ea(wj)−Ea(uj)E_a(w_j)-E_a(u_j) is nonnegative. Consequently 0≤Ea(wj)−Ea(uj)≤2a(2Ba+1)(Φ(wj)−Φ(uj))≤2a(2Ba+1)(e−Φ(uj))⟶0.0\leq E_a(w_j)-E_a(u_j) \leq 2^a(2B_a+1)\bigl(\Phi(w_j)-\Phi(u_j)\bigr) \leq 2^a(2B_a+1)(e-\Phi(u_j))\longrightarrow0. For δ>0\delta>0, the pairwise energy estimate in Ωa\Omega_a gives Cap⁡S∗(Va∩{v>H+δ};Ωa)≤Cap⁡S∗({Pa(wj)−Pa(uj)>δ};Ωa)⟶0.\operatorname{Cap}_S^*\bigl(V_a\cap\{v>H+\delta\};\Omega_a\bigr) \leq\operatorname{Cap}_S^*\bigl( \{P_a(w_j)-P_a(u_j)>\delta\};\Omega_a\bigr) \longrightarrow0. For a requested pair H0⋐ΩH_0\Subset\Omega, take the finite cofinal cover of H0H_0. Domain monotonicity and finite subadditivity transfer the zero bounds to Ω\Omega. Taking the countable union over δ=1/m\delta=1/m proves v⪯S,locHv\preceq_{S,\mathrm{loc}}H.

The least-upper-bound assertion follows from countability, exactly as in Theorem “A capacitary envelope from a maximizing sequence”: each uju_j uses finitely many members of F\mathcal F, and there are only countably many jj. The union of the corresponding local outer-capacity-zero sets remains locally outer-capacity zero. Uniqueness in the local ordered quotient follows. ◻

Corollary (Countable cofinal families). If F\mathcal F has a pointwise cofinal countable subfamily, its pointwise supremum is Borel and agrees locally quasi-everywhere with HH. This includes every countable family.

Proof. Let vm∈Fv_m\in\mathcal F be pointwise cofinal. Then sup⁡F=sup⁡mvm\sup\mathcal F=\sup_m v_m, and vm⪯S,locHv_m\preceq_{S,\mathrm{loc}}H for all mm. Countable subadditivity gives sup⁡F⪯S,locH\sup\mathcal F\preceq_{S,\mathrm{loc}}H. Conversely every uju_j is a finite maximum of members of F\mathcal F, so H≤sup⁡FH\leq\sup\mathcal F pointwise. ◻

Obstacle families and the scope of the result

Let F\mathcal F be as in the local theorem, and suppose every member satisfies the actual pointwise condition v≤ψv\leq\psi on a specified set EE. Nonempty finite maxima preserve this condition. The constructed representative H=sup⁡jujH=\sup_j u_j therefore satisfies H≤ψH\leq\psi at every point of EE. This gives an extremal object in the capacitary closure which retains a prescribed pointwise obstacle while respecting the background SS.

These envelope constructions are recorded as NEW_PROVED consequences of the relative energy estimates and the completed calculus. They do not assert that HH is classically psh, that H=(sup⁡F)∗H=(\sup\mathcal F)^*, or that its Monge–Ampère measure is supported on a contact set. They do not establish compact outer regularity or weak-closure domination for the original Borel capacity for an arbitrary background. For the ambient background, the later negligible-set theorem identifies HH with the ordinary usc envelope quasi-everywhere, and the exhaustion-domain extremal theorems supply the stated equilibrium and capacity formulas. These later conclusions use additional arguments.

Local comparison with the capacity of a fixed affine domain

Fix a smooth integral affine KK-variety YY of dimension n≥1n\geq1 and an ordinary open set

V⊂Yan,V‾⋐Yan.(D)V\subset Y^{\mathrm{an}},\qquad \overline V\Subset Y^{\mathrm{an}}. \tag{D}

Let SS be any symmetric original-positive, separately d′d'- and d′′d''-closed current of bidegree (p,p)(p,p) on VV, and put q=n−p≥1q=n-p\geq1. Restrictions of SS to smaller ordinary domains are understood in the notation Cap⁡S( ⋅ ;Ω)\operatorname{Cap}_S(\,\cdot\,;\Omega). All functions retain their actual point values and the all-extensions classical definition.

The purpose of the construction is to compare admissible functions on different domains without assuming that they extend unchanged. An explicit smooth pluriharmonic correction on the inner core is sufficient, because that correction disappears from every differentiated BT slot over an arbitrary closed background.

The effect of a change of frame

Lemma (Smooth pluriharmonic invariance). Let O⊂VO\subset V be ordinary open. Let ai,bi∈Cb(O)a_i,b_i\in\mathcal C_b(O) satisfy ai=bi+ψia_i=b_i+\psi_i, where ψi\psi_i is original smooth and Dψi=0D\psi_i=0. Then, for 0≤k≤q0\leq k\leq q, TS,k(a1,…,ak)=TS,k(b1,…,bk)on O.T_{S,k}(a_1,\ldots,a_k)=T_{S,k}(b_1,\ldots,b_k) \quad\text{on }O. In particular this applies when each ψi\psi_i is a real constant plus a real multiple of the logarithm of the absolute value of an analytic unit.

Proof. For every separately closed current QQ the original smooth product rule gives D(ψiQ)=Dψi∧Q=0.D(\psi_i Q)=D\psi_i\wedge Q=0. All cross terms vanish because d′Q=d′′Q=0d'Q=d''Q=0. This is an identity against all original compact smooth complementary tests. The result now follows inductively from the ordered recursion: first replace the inner k−1k-1 slots by induction and then use the displayed identity for the outer slot. Only multiplication by an original smooth function is used in this last step. A unit logarithm is original smooth and has zero DD, as it is linear in its unit moment coordinate. Its positive and negative are classical psh after every complete extension. ◻

This DIRECT_COROLLARY of the ordered calculus is much narrower than a claim that arbitrary-background products depend only on the ambient currents D[ai]D[a_i] for arbitrary potentials. Only the specified smooth zero-curvature differences are used below.

Uniform extension from one inner norm box

Choose a closed affine embedding Y↪AKNY\hookrightarrow\mathbf A^N_K. Let ZZ be the normalization of its projective closure, let A=ν∗OPN(1),A=\nu^*\mathcal O_{\mathbf P^N}(1), and write e∞=ν∗X0e_\infty=\nu^*X_0 for its fixed frame on YY. The line AA is ample, and YY identifies with the inverse image of the standard affine chart because it is already normal. We do not assume that ZZ is smooth. No current or BT product on its singular locus will be used.

The polynomial norm boxes below are those in the constructive part of B, Lemmas D.1–D.2. They use one finite list of polynomial upper and lower bounds, with strict slack between the successive thresholds and with upper bounds on the affine coordinates. In particular the closed boxes are compact. They may always be chosen locally inside a prescribed ordinary neighborhood.

Lemma (A uniform extension up to one fixed frame term). Let Ω⊂V\Omega\subset V be ordinary open, and fix nested polynomial norm boxes C1⊂Int⁡Cmid,Cmid⊂Int⁡C2,C2⋐Ω,C_1\subset\operatorname{Int}C_{\mathrm{mid}},\qquad C_{\mathrm{mid}}\subset\operatorname{Int}C_2,\qquad C_2\Subset\Omega, with the strict-threshold properties just described. Put W=Int⁡C1W=\operatorname{Int}C_1. There are a number A0>0A_0>0 and one original smooth pluriharmonic function ψ\psi on a neighborhood of C1C_1, depending on the boxes and VV but not on vv or SS, such that every v∈Cb(Ω)v\in\mathcal C_b(\Omega) with 0≤v≤10\leq v\leq1 admits v~∈Cb(V)\widetilde v\in\mathcal C_b(V) satisfying

0≤v~≤1 on V,v~=A0−1v+ψ on a neighborhood of C1.(E)0\leq\widetilde v\leq1\text{ on }V, \qquad \widetilde v=A_0^{-1}v+\psi\text{ on a neighborhood of }C_1. \tag{E}

Consequently, for every k≤qk\leq q,

TS,k(v~,…,v~)=A0−kTS,k(v,…,v)on W.(P)T_{S,k}(\widetilde v,\ldots,\widetilde v) =A_0^{-k}T_{S,k}(v,\ldots,v)\quad\text{on }W. \tag{P}

Proof. We give the geometric construction in full to avoid invoking either a smooth compactification or B, Theorem 4.4.

Choose intermediate thresholds between those defining C1C_1 and CmidC_{\mathrm{mid}}. By clearing rational powers, the polynomial collar can be written b=max⁡{0,log⁡∣PaNa/ca∣,log⁡∣dl/QlMl∣},ca,dl∈K×.b=\max\{0,\log|P_a^{N_a}/c_a|, \log|d_l/Q_l^{M_l}|\}, \qquad c_a,d_l\in K^\times. Every denominator QlQ_l is nonzero on a neighborhood of C2C_2. The chosen strict gaps give b=0 near C1,b≥δ>0 on C2∖Int⁡Cmid.b=0\text{ near }C_1, \qquad b\geq\delta>0\text{ on }C_2\setminus\operatorname{Int} C_{\mathrm{mid}}. The function is finite and continuous on a neighborhood of C2C_2. It is classical psh there: each nonconstant branch is a logarithm of an analytic function or of an analytic unit; the constant branch makes branches with a zero inactive in a neighborhood of that zero. These assertions persist after every complete extension.

Clear the denominators and homogenize in the fixed projective embedding. After taking a common sufficiently high degree m0m_0, all displayed branches are ratios sa/s0s_a/s_0 of global sections of Am0A^{m_0}. Here s0s_0 is a product of homogenized denominator polynomials and a power of X0X_0, pulled back to ZZ, and is nowhere zero on a neighborhood of C2C_2. Include the branch s0/s0=1s_0/s_0=1. Append small nonzero KK-scalar multiples of a finite global generating tuple of Am0A^{m_0}. The scalars can be chosen so that all their ratios to s0s_0 have absolute value strictly less than one on C2C_2, by compactness. They do not change the maximum bb there, and the full tuple generates the line everywhere on ZZ.

Its pure Fubini–Study metric q0q_0 on Am0A^{m_0} has weight bb in the frame s0s_0 near C2C_2. This metric is classical on ZZ in the literal sense that every local frame weight is classical psh after every complete extension. Indeed, pullback along any analytic curve gives the finite maximum of logarithms of analytic functions with no common zero. This argument uses generation and the sheaf and maximum properties, and needs no smoothness of ZZ.

Fix c>0c>0, put M=c+1M=c+1, and choose an integer m>0m>0 with mδ>M+1m\delta>M+1. Set d=mm0,L=Ad,e=s0m,Q=mq0.d=mm_0,\qquad L=A^d,\qquad e=s_0^m,\qquad Q=mq_0. All these objects are independent of vv. Define a metric RvR_v on LL using the two ordinary opens Int⁡C2\operatorname{Int}C_2 and Zan∖CmidZ^{\mathrm{an}}\setminus C_{\mathrm{mid}}: its ee-weight on the first is max⁡{v+c,mb},\max\{v+c,mb\}, and on the second it equals QQ. The opens cover ZanZ^{\mathrm{an}} because Cmid⊂Int⁡C2C_{\mathrm{mid}}\subset\operatorname{Int}C_2. On their overlap the inequality mb≥mδ>M≥v+cmb\geq m\delta>M\geq v+c makes the two prescriptions identical. They therefore glue as metrics on the same identified line. Their weights are all-extensions classical by restriction, maxima, and locality on curves. This is an explicit gluing construction on the normal variety, not an application of a global approximation or envelope theorem there.

The metric difference is a globally defined bounded Borel function and satisfies 0≤Rv−Q≤M.0\leq R_v-Q\leq M. On Int⁡C2\operatorname{Int}C_2 this is the identity Rv−Q=max⁡{v+c−mb,0}R_v-Q=\max\{v+c-mb,0\} and the bound follows from b≥0b\geq0. On the other open the difference is zero.

Now use the fixed frame e∞de_\infty^d on all of YY. Let wv=wt⁡e∞d(Rv),Q∞=wt⁡e∞d(Q).w_v=\operatorname{wt}_{e_\infty^d}(R_v),\qquad Q_\infty=\operatorname{wt}_{e_\infty^d}(Q). The function wvw_v is classical psh on YanY^{\mathrm{an}}, and Q∞Q_\infty is continuous there. In particular they are bounded on VV, uniformly in vv, because Q∞≤wv≤Q∞+MandV‾⋐Yan.Q_\infty\leq w_v\leq Q_\infty+M \quad\text{and}\quad\overline V\Subset Y^{\mathrm{an}}. Write a=min⁡V‾Q∞,A0=1+osc⁡V‾(Q∞)+M,v~=wv−aA0∣V.a=\min_{\overline V}Q_\infty, \qquad A_0=1+\operatorname{osc}_{\overline V}(Q_\infty)+M, \qquad \widetilde v=\frac{w_v-a}{A_0}\Big|_V. This gives 0≤v~≤10\leq\widetilde v\leq1 on the original domain VV. The actual bounds and A0A_0 are independent of the competitor.

Near C1C_1 we have b=0b=0 and v+c>0v+c>0, so the ee-weight of RvR_v is exactly v+cv+c. The ratio f=s0me∞df=\frac{s_0^m}{e_\infty^d} is an algebraic unit near C2C_2. Since e=fe∞de=f e_\infty^d, the metric transition law gives wv=v+c+log⁡∣f∣near C1.w_v=v+c+\log|f|\quad\text{near }C_1. Thus (E) holds with the fixed function ψ=A0−1(c+log⁡∣f∣−a)\psi=A_0^{-1}(c+\log|f|-a). It is original smooth pluriharmonic. Positive homogeneity and Lemma “Smooth pluriharmonic invariance” prove (P), with the original background and on all original tests. The normal projective variety was used only to construct the metric and its frame; every current in this final step is on the smooth open YY. ◻

The lemma is recorded as NEW_PROVED in the stated uniform framed-affine scope, by the constructive collar mechanism of B, Lemma D.2. Its proof does not use that lemma’s smooth projective ambient hypothesis, a resolution of singularities, or B, Theorem 4.4. The following capacity consequences are DIRECT_COROLLARY.

Borel and outer capacities on different domains

For arbitrary subsets of a domain OO, write Cap⁡S∗(E;O)=inf⁡G⊃EG ordinary open in OCap⁡S(G;O).\operatorname{Cap}_S^*(E;O) =\inf_{\substack{G\supset E\\G\text{ ordinary open in }O}} \operatorname{Cap}_S(G;O). This is the outer capacity by definition; no assertion that it agrees with the original Borel capacity is used here.

Theorem (Uniform local comparison with a fixed domain). Assume (D). For every ordinary Ω⊂V\Omega\subset V and every ordinary WW with W‾⋐Ω\overline W\Subset\Omega, there is a finite constant C=C(Y,V,Ω,W,q)>0C=C(Y,V,\Omega,W,q)>0, independent of SS, such that

Cap⁡S(E;V)≤Cap⁡S(E;Ω)≤CCap⁡S(E;V)(E⊂W Borel).(C)\operatorname{Cap}_S(E;V) \leq\operatorname{Cap}_S(E;\Omega) \leq C\operatorname{Cap}_S(E;V) \quad(E\subset W\text{ Borel}). \tag{C}

For arbitrary E⊂WE\subset W one also has

Cap⁡S∗(E;V)≤Cap⁡S∗(E;Ω)≤CCap⁡S∗(E;V).(C*)\operatorname{Cap}_S^*(E;V) \leq\operatorname{Cap}_S^*(E;\Omega) \leq C\operatorname{Cap}_S^*(E;V). \tag{C*}

For a single inner core in Lemma “A uniform extension up to one fixed frame term”, one may take C=A0qC=A_0^q.

Proof. The first inequality of (C) is restriction of competitors: a normalized function on VV remains normalized and classical on Ω\Omega, and its product restricts there by ordinary locality.

First suppose that WW is the core of one nested-box construction. For an arbitrary competitor 0≤v≤10\leq v\leq1 on Ω\Omega, the extension lemma and equality of actual Radon measures on WW give ∫ETS,q(v,…,v)=A0q∫ETS,q(v~,…,v~)≤A0qCap⁡S(E;V).\int_E T_{S,q}(v,\ldots,v) =A_0^q\int_E T_{S,q}(\widetilde v,\ldots,\widetilde v) \leq A_0^q\operatorname{Cap}_S(E;V). Taking the supremum over vv proves the second inequality for that core. The constant is uniform over all competitors and all allowed backgrounds.

For general WW, cover the compact W‾\overline W by finitely many inner cores W1,…,WNW_1,\ldots,W_N of nested polynomial norm boxes whose outer boxes lie in Ω\Omega. Let AiA_i be the constants from their extension lemmas. The same competitor vv on Ω\Omega can be extended separately on each core. Positivity and a finite cover give ∫ETS,q(v,…,v)≤∑i=1N∫E∩WiTS,q(v,…,v)≤∑i=1NAiqCap⁡S(E∩Wi;V)≤(∑i=1NAiq)Cap⁡S(E;V).\begin{split} \int_E T_{S,q}(v,\ldots,v) &\leq\sum_{i=1}^N\int_{E\cap W_i}T_{S,q}(v,\ldots,v)\\ &\leq\sum_{i=1}^NA_i^q\operatorname{Cap}_S(E\cap W_i;V)\\ &\leq\left(\sum_{i=1}^NA_i^q\right) \operatorname{Cap}_S(E;V). \end{split} Take the supremum. Thus C=∑iAiqC=\sum_iA_i^q works. This finite-cover step avoids assuming that an arbitrary compact core lies in one strictly smaller box of a previously chosen presentation. In particular, no strict-inequality description of an ordinary interior is silently imposed.

For outer capacities, every ordinary open of Ω\Omega is ordinary open in VV, so restriction and the first Borel inequality give the first inequality of (C*). Conversely, if G⊂VG\subset V is ordinary open and contains EE, then G∩WG\cap W is an ordinary open of Ω\Omega containing EE. The already proved Borel comparison and monotonicity give Cap⁡S∗(E;Ω)≤Cap⁡S(G∩W;Ω)≤CCap⁡S(G∩W;V)≤CCap⁡S(G;V).\operatorname{Cap}_S^*(E;\Omega) \leq\operatorname{Cap}_S(G\cap W;\Omega) \leq C\operatorname{Cap}_S(G\cap W;V) \leq C\operatorname{Cap}_S(G;V). Take the infimum over such GG. This argument invokes the definition of outer capacity and no outer-regularity theorem. ◻

The endpoint p=np=n can be included by defining the degree-zero capacity to be S(E)S(E); domain restriction then makes both Borel capacities identical, with constant one.

Equivalent local convergence and null sets

Corollary (One fixed capacity domain suffices locally). Assume (D), and let aj,aa_j,a be real Borel functions on VV. The following are equivalent:

  1. For every compact L⋐VL\Subset V and every δ>0\delta>0, Cap⁡S(L∩{∣aj−a∣>δ};V)⟶0.\operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};V) \longrightarrow0.

  2. The functions converge in the all-domain local SS-capacity sense of Definition “Local convergence in the relative capacity”, restricted to VV.

The same equivalence holds with the outer capacities throughout. No uniform boundedness or classicality of the functions aj,aa_j,a is needed for this equivalence of convergence conditions.

Proof. Suppose (1) holds. Given a permitted relatively compact domain Ω⊂V\Omega\subset V and a compact L⋐ΩL\Subset\Omega, choose an ordinary WW with L⊂WL\subset W and W‾⋐Ω\overline W\Subset\Omega. The second inequality of (C), with this fixed geometry, implies the required convergence using Ω\Omega as admissibility domain.

Conversely, cover a prescribed compact L⋐VL\Subset V by finitely many ordinary WiW_i with compact closure in permitted small norm-box domains Ωi⋐V\Omega_i\Subset V. Domain monotonicity and finite subadditivity give Cap⁡S(L∩{∣aj−a∣>δ};V)≤∑iCap⁡S(W‾i∩{∣aj−a∣>δ};Ωi).\begin{split} \operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};V) &\leq\sum_i\operatorname{Cap}_S (\overline W_i\cap\{|a_j-a|>\delta\};\Omega_i). \end{split} Each summand tends to zero by (2). The same proof uses (C*) and subadditivity of outer capacity for the outer version. ◻

Corollary (Equivalence of local null classes). For a Borel set E⊂VE\subset V, its capacity with domain VV is zero if and only if Cap⁡S(E∩L;Ω)=0\operatorname{Cap}_S(E\cap L;\Omega)=0 for every permitted small domain Ω⊂V\Omega\subset V and compact L⋐ΩL\Subset\Omega. For arbitrary subsets EE, the same assertion holds with outer capacities.

Proof. One direction is the comparison theorem. For the converse, VV is second countable in the present finite pp-adic affine setting. Choose countably many ordinary cores WiW_i covering VV, with W‾i⋐Ωi\overline W_i\Subset\Omega_i for permitted small norm-box domains Ωi⋐V\Omega_i\Subset V. Monotonicity under restriction and countable subadditivity give Cap⁡S(E;V)≤∑iCap⁡S(E∩W‾i;Ωi)=0.\operatorname{Cap}_S(E;V) \leq\sum_i\operatorname{Cap}_S(E\cap\overline W_i;\Omega_i)=0. The outer-capacity proof is identical. No ambient upper-semicontinuous representative is substituted in this argument. ◻

These statements identify the local convergence and local null-set notions on the qualified fixed domain VV. They do not assert uniform constants as the core approaches the boundary of VV, convergence in capacity on the entire noncompact domain from local convergence, or capacity nondegeneracy. They remain valid for S=0S=0 and backgrounds supported on a proper analytic subvariety. They do not apply to an arbitrary proper ambient space with only constant untwisted global psh competitors. Every comparison is between ordinary domains; the sets being measured may be arbitrary Borel sets, including non-open strict gaps.

Analytic supports and a relative-envelope obstruction

Proposition (BT products and capacity on an analytic retract). Let i:Y↪Ui:Y\hookrightarrow U be a closed analytic embedding of allowed pure-dimensional quasi-smooth spaces, of dimensions mm and nn. Let RR be a symmetric original-positive separately closed current of bidegree (a,a)(a,a) on YY, with 0≤a≤m0\leq a\leq m, and put S=i∗R,p=n−m+a,q=m−a=n−p.S=i_*R,\qquad p=n-m+a,\qquad q=m-a=n-p. For 0≤k≤q0\leq k\leq q and u1,…,uk∈Cb(U)u_1,\ldots,u_k\in\mathcal C_b(U),

TS,kU(u1,…,uk)=i∗TR,kY(u1∘i,…,uk∘i)T^U_{S,k}(u_1,\ldots,u_k) =i_*T^Y_{R,k}(u_1\circ i,\ldots,u_k\circ i)

on all original compact smooth complementary tests. If there is an analytic retraction r:U→Yr:U\to Y with r∘i=idr\circ i=\mathrm{id}, then, for every Borel E⊂UE\subset U,

Cap⁡S(E;U)=Cap⁡R(i−1E;Y).\operatorname{Cap}_{S}(E;U) =\operatorname{Cap}_{R}(i^{-1}E;Y).

The corresponding equality holds for outer capacities of arbitrary subsets. In degree zero the capacity is simply the measure of the set.

Status: DIRECT_COROLLARY. This is a functorial consequence of the fixed-background recursion, not a restriction of that recursion to integration-current backgrounds.

Proof. The map ii is proper. Its pullback preserves the original positive test cone and commutes with the two differentials; hence SS has the stated positivity and closedness. For any coefficient-Radon current PP on YY and locally bounded Borel function ff on UU, f i∗P=i∗((f∘i)P).f\,i_*P=i_*((f\circ i)P). Indeed, the coefficient measure of i∗Pi_*P against a test η\eta is the pushforward of the coefficient measure of PP against i∗ηi^*\eta. This follows first on smooth scalar tests and then on Borel functions by Radon uniqueness. Apply this identity and commutation of DD with proper pushforward inductively through the recursion. Analytic pullback preserves classical psh, so every restricted slot is allowed. This proves (eq:r2-retract-product).

Every normalized potential on UU restricts to a normalized potential on YY, giving the inequality ≤\leq in (eq:r2-retract-capacity). Conversely every normalized vv on YY extends to v∘rv\circ r on UU, still classical and between zero and one. Equation (eq:r2-retract-product) gives the opposite inequality. For outer capacities, restricting any ordinary open neighborhood of EE gives one of i−1Ei^{-1}E. Conversely, if H⊂YH\subset Y is open and contains i−1Ei^{-1}E, choose an open G0⊂UG_0\subset U with i−1G0=Hi^{-1}G_0=H. The open set G0∪(U∖i(Y))G_0\cup(U\setminus i(Y)) contains EE and has capacity Cap⁡R(H;Y)\operatorname{Cap}_{R}(H;Y). Taking infima proves the assertion. ◻

Proposition (Ambient negligible sets can have positive relative capacity). For every m≥1m\geq1, there is an allowed (m+1)(m+1)-dimensional domain UU, a nonzero positive closed background SS of bidegree (1,1)(1,1), and an increasing sequence aj∈Cb(U)∩C0(U)a_j\in\mathcal C_b(U)\cap C^0(U), with −1≤aj≤0-1\leq a_j\leq0, such that, writing h=sup⁡jajh=\sup_j a_j, h∗=0,Cap⁡S({h<h∗};U)>0.h^*=0,\qquad \operatorname{Cap}_{S}(\{h<h^*\};U)>0. There is even a compact singleton K⊂UK\subset U of positive relative capacity for which the ambient relative extremal function hK=sup⁡{v∈Cb(U):v≤0, v∣K≤−1}h_K=\sup\{v\in\mathcal C_b(U):v\leq0,\ v|_K\leq-1\} satisfies hK∗=0h_K^*=0.

Status of the assertions refuted: FALSE. The construction refutes both universal relative negligibility of ambient envelope exceptions and the unqualified formula Cap⁡S(K;U)=∫UTS,m((hK∗)m)\operatorname{Cap}_S(K;U)=\int_U T_{S,m}((h_K^*)^m).

Proof. Choose 0<r<1<R0<r<1<R and set U={∣z0∣<1}×∏i=1m{r<∣zi∣<R},Y={z0=0},S=i∗[1]Y.U=\{|z_0|<1\}\times\prod_{i=1}^m\{r<|z_i|<R\}, \qquad Y=\{z_0=0\},\qquad S=i_*[1]_Y. The projection onto the last mm coordinates is an analytic retraction. Put aj=max⁡{−1,j−1log⁡∣z0∣}.a_j=\max\{-1,j^{-1}\log|z_0|\}. The logarithm is interpreted as −∞-\infty on YY. The maximum is continuous and classical, it equals −1-1 on YY, and it increases to zero at every point outside YY. Hence h=−1h=-1 on YY and h=0h=0 outside YY.

The complement of YY is dense in UU. For an explicit verification at y∈Yy\in Y, use the Gauss seminorm in the z0z_0 variable over the completed residue field H(y)\mathcal H(y), with radius s>0s>0. It defines a point ysy_s of the original KK-analytic product with ∣z0(ys)∣=s|z_0(y_s)|=s. As s↓0s\downarrow0, evaluations of polynomials tend to their values at yy, and so do local analytic functions by uniform approximation on a compact product neighborhood. Thus ys→yy_s\to y. Consequently h∗=0h^*=0.

Let ξY\xi_Y be the product Gauss point of YY and K={i(ξY)}K=\{i(\xi_Y)\}. For ui=max⁡{0,log⁡∣zi∣}u_i=\max\{0,\log|z_i|\}, the standard model metric on (PK∘1)m(\mathbf P^1_{K^\circ})^m gives TS,m(u1,…,um)=δi(ξY).T_{S,m}(u_1,\ldots,u_m)=\delta_{i(\xi_Y)}. This is exactly the mixed model calculation of the first record, Proposition “Concentration on a smooth divisor”, with the original absolute value and calibration. Put M=mlog⁡RM=m\log R and v=M−1∑iuiv=M^{-1}\sum_i u_i. Then 0≤v≤10\leq v\leq1 on UU and positivity and multilinearity give TS,m(vm)(K)≥m!(mlog⁡R)m>0.T_{S,m}(v^m)(K)\geq \frac{m!}{(m\log R)^m}>0. In particular both KK and Y={h<h∗}Y=\{h<h^*\} have positive relative capacity.

Every aja_j is a competitor for hKh_K. Thus hK=0h_K=0 off YY, while hK≤0h_K\leq0 everywhere. Density again gives hK∗=0h_K^*=0. The constant −1-1 is a competitor and every competitor is at most −1-1 on KK, so hK=−1h_K=-1 on KK itself. All differentiated slots of hK∗=0h_K^*=0 vanish, whereas the displayed capacity of KK is positive. This proves the claimed failure without an appeal to an unspecified pluripolar exceptional set. ◻

A correct replacement in this setting.

For a background supported on an analytic retract, the intrinsic capacity and extremal problem on YY are the appropriate comparison objects, by Proposition “BT products and capacity on an analytic retract”. Ambient upper regularization can erase values on YY that the background detects. For a general positive closed current its support need not be an analytic space; this observation does not by itself define an intrinsic psh envelope on that support. Such a construction remains a separate analytic problem.

Detection of strict gaps and actual point values

All fields, currents, and test forms retain the conventions of the first round. In particular, K/QpK/\mathbf Q_p is finite with its prescribed absolute value, D=d′d′′D=d'd'', and Cb\mathcal C_b denotes the actual pointwise, locally bounded, all-extensions classical psh class. No representative is altered on a null set. This section supplies a sufficient geometric condition for the strict-gap hypothesis in Corollary “A conditional pointwise conclusion”. The condition is proved for S=[1]S=[1] on the domains specified below, and for an explicit family of positive closed backgrounds. It is not asserted for every positive closed background or every ordinary domain.

Shilov boundaries determine pointwise order

Definition (A family of affinoid neighborhoods). Let VV be an ordinary open in a good strictly analytic space. A family A\mathscr A of strict affinoid domains contained in VV is called an affinoid neighborhood basis if, whenever x∈O⊂Vx\in O\subset V and OO is ordinary open, some A∈AA\in\mathscr A satisfies x∈Int⁡V(A)⊂A⊂O.x\in\operatorname{Int}_V(A)\subset A\subset O. Write Γ(A)\Gamma(A) for its finite Shilov boundary and put D=⋃A∈AΓ(A)\mathscr D=\bigcup_{A\in\mathscr A}\Gamma(A).

The external input used here is [GR25, Theorem 6.4, pp. 33–34]: a classical psh function on an affinoid space attains its maximum at a Shilov point. Its definition of classical psh uses curve pullback after every complete extension. The theorem applies to the restriction of a function in Cb(V)\mathcal C_b(V) to each AA; it does not require that the function be continuous. The following lemma is a DIRECT_COROLLARY of that theorem and upper semicontinuity.

Lemma (A strict gap contains a Shilov point). Let aa be a finite classical psh function on VV, let bb be a finite upper-semicontinuous function on VV, and let δ≥0\delta\geq0. If a(x)>b(x)+δa(x)>b(x)+\delta, then every sufficiently small ordinary neighborhood of xx contains a point z∈Dz\in\mathscr D with a(z)>b(z)+δa(z)>b(z)+\delta. Consequently, a≤b on D⟹a≤b on V.a\leq b\text{ on }\mathscr D\quad\Longrightarrow\quad a\leq b \text{ on }V. In particular, two members of Cb(V)\mathcal C_b(V) that agree on D\mathscr D agree at every point of VV.

Proof. Choose ϵ>0\epsilon>0 with b(x)+δ+2ϵ<a(x)b(x)+\delta+2\epsilon<a(x). Upper semicontinuity gives an ordinary neighborhood OO of xx on which b<b(x)+ϵb<b(x)+\epsilon. Choose A∈AA\in\mathscr A with x∈Int⁡V(A)x\in\operatorname{Int}_V(A) and A⊂OA\subset O. The affinoid maximum principle gives z∈Γ(A)z\in\Gamma(A) such that a(z)=max⁡Aa≥a(x)a(z)=\max_A a\geq a(x). Therefore a(z)≥a(x)>b(x)+δ+2ϵ>b(z)+δ.a(z)\geq a(x)>b(x)+\delta+2\epsilon>b(z)+\delta. The same construction can be made inside any prescribed ordinary neighborhood of xx, by intersecting it with OO. This proves the first assertion. If pointwise order failed, apply it with δ=0\delta=0 to contradict order on D\mathscr D. Apply that implication in both orders to obtain equality. The strict-gap set itself has nowhere been treated as ordinary open. ◻

For example, for a fixed x∈Vx\in V the same maximum principle and upper semicontinuity give the recovery formula a(x)=inf⁡A∈Ax∈Int⁡V(A)max⁡z∈Γ(A)a(z).a(x)=\inf_{\substack{A\in\mathscr A\\ x\in\operatorname{Int}_V(A)}} \max_{z\in\Gamma(A)}a(z). The infimum is over the members of the full neighborhood basis. This is a pointwise statement, rather than an assertion about equality classes modulo any measure.

An atom at a prescribed divisorial point

For the fixed-domain result, assume that Y/KY/K is a smooth integral affine variety of dimension n≥1n\geq1 and that

V⊂Yanis ordinary open withV‾⋐Yan.(G)V\subset Y^{\mathrm{an}} \quad\text{is ordinary open with}\quad \overline V\Subset Y^{\mathrm{an}}. \tag{G}

The closure here is the ordinary topological closure. More generally, the argument applies when VV has compact closure inside an algebraic open of a normal projective variety on which a fixed ample line bundle has a specified nowhere-vanishing frame, and that open is smooth. No convexity, exhaustion, or regularity of the boundary of VV is assumed. Small norm boxes in the algebraic neighborhoods supplied by B satisfy (G), after choosing one affine chart.

We use the following external facts with their original field: [GM19, Proposition A.3, p. 365; Corollary A.6, p. 367; Proposition A.9, p. 368] identify Shilov points of strict affinoids with divisorial points, show locality of divisoriality, and realize divisorial points on algebraic K∘K^\circ-models of a proper variety. The ample extension theorem is [GM19, Proposition 4.11, p. 353]. The ordinary CLD model-measure formula is [CLD12, Theorem 6.9.3 and Proposition 6.9.7]; its generically smooth and stable-field hypotheses apply here. For smooth projective varieties the same normalization is explicitly recorded in R, Lemma 3.3, and B, equation (4.1). We only use the ordinary-current measure, not equality on all δ\delta-test forms.

Lemma (A bounded model tester on the original domain). Assume (G). For every divisorial point z∈Vz\in V there exists a continuous semipositive real PL function hz∈Cb(V)h_z\in\mathcal C_b(V) such that 0≤hz≤1on V,(Dhz)n({z})>0.0\leq h_z\leq1\quad\text{on }V, \qquad (Dh_z)^n(\{z\})>0. The function is defined and bounded on a neighborhood of V‾\overline V. If a larger relatively compact ordinary V+V^+ with V‾⊂V+⋐Yan\overline V\subset V^+\Subset Y^{\mathrm{an}} is fixed in advance, one can arrange 0≤hz≤10\leq h_z\leq1 on this same V+V^+ for every zz.

Proof. Choose a closed embedding Y↪AKNY\hookrightarrow\mathbf A^N_K and take the normalization ZZ of its closure in PKN\mathbf P^N_K. Because YY is normal, the inverse image of the standard affine chart in ZZ is identified with YY. If ν:Z→Y‾\nu:Z\to\overline Y is the normalization, then L=ν∗OPN(1)L=\nu^*\mathcal O_{\mathbf P^N}(1) is ample, and the pullback of the homogeneous coordinate X0X_0 gives a fixed frame ee of LL on YY. Thus no modification of the open line or its frame has been made.

The point zz is divisorial in ZanZ^{\mathrm{an}}. By the cited model realization theorem there is an algebraic K∘K^\circ-model Z0\mathscr Z_0 of ZZ on which zz is associated with an irreducible component of the special fiber. Normalize this model. Normalization is finite in the present excellent discretely valued situation, and the center selected by zz is still a special-fiber component. Apply the ample extension theorem to obtain a dominating model Z\mathscr Z and an ample line bundle L\mathscr L with a specified identification L∣Z≃L⊗m,m>0.\mathscr L|_Z\simeq L^{\otimes m},\qquad m>0. Normalize once more if necessary and pull back L\mathscr L; finite pullback preserves ampleness. A dominating model of a normal model preserves each codimension-one divisorial valuation represented on the latter. Indeed, after localizing the target at the generic point of its component one has a discrete valuation ring, and the proper birational morphism is an isomorphism at that generic point. Hence zz is still associated with a special-fiber component EE of Z\mathscr Z.

For clarity, the ample extension used here does not require resolution of mixed-characteristic models. One may first choose a projective model carrying an ample extension of a power of LL, take a vertical blowup dominating Z0\mathscr Z_0, and tensor a sufficiently high pullback power with its relatively ample tautological line. The center of the blowup lies in the special fiber, so the tautological line has trivial generic fiber. This produces an ample extension of the stated power of LL, before normalization. This is an existence of a dominating model, not a claim about an arbitrary previously fixed model.

The model metric of L\mathscr L gives the weight ϕ=−log⁡∥e⊗m∥Lon Yan.\phi=-\log\|e^{\otimes m}\|_{\mathscr L} \quad\text{on }Y^{\mathrm{an}}. It is continuous and semipositive PL. In particular it is classical psh after all complete extensions by [GR25, Theorem 4.3, pp. 22–23]. Equivalently, replace L\mathscr L by a positive power that is generated by model sections. Its weight is then a pure Fubini–Study maximum in the same frame; the inactive-branch rule handles zeros of the sections. This also shows directly that all functions used are defined over the original KK.

On a normal algebraic model the original CLD model formula gives the atom (Dϕ)n({z})=bEdeg⁡K~,E((L∣E)n)>0,(D\phi)^n(\{z\}) =b_E\deg_{\widetilde K,E}\bigl((\mathscr L|_E)^n\bigr)>0, where bEb_E is the special-fiber multiplicity and the structural degree retains residue-field degrees. Ampleness makes the displayed degree strictly positive. This formula is interpreted in the original calibration; neither the absolute value nor the integration current is rescaled. The continuous BT product on the smooth open YY is this ordinary CLD measure by the accepted continuous compatibility of R and B. If a power was used to obtain generating sections, its tensor degree is retained in this formula and in the frame, or divided out in every slot.

Fix V+V^+ as in the statement, taking one if none was prescribed. The continuous function ϕ\phi is bounded on V+‾\overline {V^+}. Choose real numbers A<BA<B with A≤ϕ≤BA\leq\phi\leq B there, and set hz=(ϕ−A)/(B−A)h_z=(\phi-A)/(B-A). A positive constant multiple and a constant translate preserve classicality and semipositive real PL structure. They multiply the displayed atom by (B−A)−n>0(B-A)^{-n}>0. This proves all assertions on the fixed domain VV. ◻

The lemma is a DIRECT_COROLLARY of the cited model geometry, model-measure formula, and the accepted continuous compatibility. It does not solve a prescribed Dirac Monge–Ampère equation: other positive atoms of (Dhz)n(Dh_z)^n are allowed.

Theorem (Strict-gap detection for the ambient capacity). Assume (G). If a,b∈Cb(V)a,b\in\mathcal C_b(V) and {a>b+δ}\{a>b+\delta\} is nonempty for some δ≥0\delta\geq0, then Cap⁡[1]({a>b+δ};V)>0.\operatorname{Cap}_{[1]}(\{a>b+\delta\};V)>0. Here the supremum defining capacity uses functions on the original domain VV. In particular the strict-gap hypothesis of Corollary “A conditional pointwise conclusion” holds for S=[1]S=[1] on VV.

Proof. Choose a strict affinoid neighborhood basis in VV. By Lemma “A strict gap contains a Shilov point”, the gap contains a Shilov point zz of one of these affinoids. Such a point is divisorial in YanY^{\mathrm{an}} by the cited locality of divisorial points. Choose the tester hzh_z from Lemma “A bounded model tester on the original domain”. The gap is Borel, since both potentials are upper semicontinuous. Thus Cap⁡[1]({a>b+δ};V)≥∫{a>b+δ}(Dhz)n≥(Dhz)n({z})>0.\operatorname{Cap}_{[1]}(\{a>b+\delta\};V) \geq\int_{\{a>b+\delta\}}(Dh_z)^n \geq(Dh_z)^n(\{z\})>0. Every tester in this inequality is defined on VV. There is no use of an implication from positive capacity on a smaller domain to positive capacity on VV. ◻

This is a proved consequence of the two lemmas, independently of any outer regularity or identification of polar sets. In the project ledger it is a DIRECT_COROLLARY of those inputs and the definition of capacity.

Corollary (Pointwise ambient domination and uniqueness). Assume (G) and let u,v∈Cb(V)u,v\in\mathcal C_b(V). If {u<v−ϵ}‾⋐V(ϵ>0),(Du)n({u<v})=0,\overline{\{u<v-\epsilon\}}\Subset V\quad(\epsilon>0), \qquad (Du)^n(\{u<v\})=0, then u≥vu\geq v at every point of VV. If the compact-gap condition holds in both orders and (Du)n=(Dv)n(Du)^n=(Dv)^n, then u=vu=v at every point.

Proof. Apply Proposition “Domination modulo capacity zero” or Proposition “Uniqueness modulo capacity zero”, respectively, and then Theorem “Strict-gap detection for the ambient capacity” to any prohibited strict inequality with a positive margin. This uses only Borel restrictions of the measures. ◻

Faithfulness of original trace currents

Proposition (Actual classical representatives are determined by their original currents). Let UU have the local algebraic neighborhoods supplied by B, and let u,v∈Cb(U)u,v\in\mathcal C_b(U). Then [u]≤[v] as original CLD (0,0)-currents⟺u≤v pointwise on U.[u]\leq[v]\text{ as original CLD }(0,0)\text{-currents} \quad\Longleftrightarrow\quad u\leq v\text{ pointwise on }U. On the left, [v]−[u][v]-[u] is nonnegative on every positive original compactly supported (n,n)(n,n)-form. In particular [u]=[v][u]=[v] implies actual pointwise equality. This proposition uses the fixed-background weighted monotone theorem of the first round; it makes no assertion of stability for arbitrary varying backgrounds.

Proof. Pointwise order plainly implies current order. For the converse it is enough to work in a domain satisfying (G), inside an algebraic neighborhood of each point. Fix a divisorial point zz there, and choose its model tester hh by Lemma “A bounded model tester on the original domain”. On a fixed smaller ordinary neighborhood WW of zz, the smooth approximation theorem of R gives smooth psh functions converging uniformly to hh. The pure-FS realization in the proof of the lemma allows this also to be obtained directly from R, Proposition A.1. After taking a subsequence sjs_j with ∥sj−h∥W≤4−j\|s_j-h\|_W\leq4^{-j}, set hj=sj+2⋅4−j.h_j=s_j+2\cdot4^{-j}. Then hj↓hh_j\downarrow h on the same WW, since h+4−j≤hj≤h+3⋅4−jh+4^{-j}\leq h_j\leq h+3\cdot4^{-j}. Each hjh_j is smooth and belongs to the all-extensions classical class: locally it is a convex, coordinatewise nondecreasing smooth function of logarithms of analytic units, with zero-section branches inactive. Its curvature is an original positive (1,1)(1,1)-form.

For every nonnegative original smooth scalar χ∈Ac0,0(W)\chi\in A_c^{0,0}(W), the form χ(Dhj)n\chi(Dh_j)^n is positive and compactly supported. The assumed current order therefore gives ⟨(v−u)(Dhj)n,χ⟩≥0.\big\langle (v-u)(Dh_j)^n,\chi\big\rangle\geq0. Apply Theorem “Weighted decreasing continuity for a fixed arbitrary background” with fixed background [1][1], all differentiated slots equal to hjh_j, and the weight equal to the constant sequence uu or vv. It yields (v−u)(Dhj)n⟶(v−u)(Dh)n(v-u)(Dh_j)^n\longrightarrow(v-u)(Dh)^n on all original scalar tests. The limit is therefore a positive Radon measure. Its atom at zz is (v(z)−u(z))(Dh)n({z})(v(z)-u(z))(Dh)^n(\{z\}), and the second factor is strictly positive. Hence u(z)≤v(z)u(z)\leq v(z).

This holds at every point in the Shilov determining set of the local domain. Lemma “A strict gap contains a Shilov point” gives u≤vu\leq v throughout that domain, and these domains cover UU by B. This proves the converse. Equality follows by interchanging uu and vv. ◻

Thus equality almost everywhere for all local original trace measures does determine the actual classical representatives: it implies equality of their (0,0)(0,0)-currents, to which the proposition applies. The essential passage to an atomic tester uses weighted convergence of a decreasing potential with fixed background [1][1]. Weak convergence of arbitrary background measures would not justify it. The proposition is a DIRECT_COROLLARY of that theorem, the atomic testers, and the Shilov maximum principle.

A precise condition on an arbitrary background

Let SS be an original symmetric positive separately d′d'- and d′′d''-closed current of bidegree (p,p)(p,p) on VV, with q=n−p≥1q=n-p\geq1. An explicit sufficient condition is the following: there is an affinoid neighborhood basis A\mathscr A such that, for each z∈D=⋃A∈AΓ(A)z\in\mathscr D=\bigcup_{A\in\mathscr A}\Gamma(A), there is hz∈Cb(V)h_z\in\mathcal C_b(V) satisfying

0≤hz≤1,TS,q(hz,…,hz)({z})>0.(A)0\leq h_z\leq1, \qquad T_{S,q}(h_z,\ldots,h_z)(\{z\})>0. \tag{A}

The condition is stated in actual Borel measures on VV, not only in one tropical image.

Proposition (Pointwise conclusions under atomic detection). If (A) holds, every nonempty strict gap of two members of Cb(V)\mathcal C_b(V) has positive Cap⁡S( ⋅ ;V)\operatorname{Cap}_S(\,\cdot\,;V). Consequently the pointwise domination and uniqueness conclusions of Corollary “Pointwise ambient domination and uniqueness” hold with (Da)n(Da)^n replaced by TS,q(a,…,a)T_{S,q}(a,\ldots,a).

Proof. Lemma “A strict gap contains a Shilov point” supplies a point zz of the gap in D\mathscr D. Its tester in (A) assigns positive measure to the gap. The definition of capacity and the first-round domination and uniqueness propositions finish the proof. ◻

For a general SS this is CONDITIONAL: the condition is not deduced from positivity, nonzero mass, or full support alone. The next construction shows that it is nonvacuous in every allowed degree.

Proposition (Backgrounds with detection in every degree). Assume (G). There is a sequence of normalized continuous semipositive PL functions hjh_j on a fixed neighborhood of V‾\overline V such that, for every 0≤p<n0\leq p<n, the series Sp=∑j≥12−jT[1],p(hj,…,hj)S_p=\sum_{j\geq1}2^{-j}T_{[1],p}(h_j,\ldots,h_j) defines a symmetric positive separately closed original current on VV satisfying (A). Here T[1],0=[1]T_{[1],0}=[1]. More generally, any positive closed background SS of the same degree with S−ϵSpS-\epsilon S_p positive for some ϵ>0\epsilon>0 satisfies that condition.

The measure σ=∑j≥12−j(Dhj)n\sigma=\sum_{j\geq1}2^{-j}(Dh_j)^n is a finite positive Radon measure on VV with full support. If two members of Cb(V)\mathcal C_b(V) are ordered σ\sigma-almost everywhere, they are ordered at every point of VV. In particular this one fixed measure determines their actual representatives.

Proof. The space YanY^{\mathrm{an}} is second countable in the present setting. Here is the elementary reason needed for this assertion. A finite extension of Qp\mathbf Q_p has a countable dense subfield. On each bounded coordinate region, polynomials with coefficients in that subfield approximate arbitrary KK-polynomials uniformly. Their absolute-value functions therefore determine the analytic topology, and inverse images of intervals with rational endpoints give a countable basis. Passing to a closed affine subvariety and an ordinary open preserves second countability.

For each member OO of a countable ordinary basis, cover OO by interiors of strict affinoid domains contained in OO. Good strict analyticity supplies the cover, and second countability makes OO Lindelöf, so a countable subcover suffices. The resulting countable family is an affinoid neighborhood basis. Its Shilov determining set D={z1,z2,…}\mathscr D=\{z_1,z_2,\ldots\} is countable. Fix V‾⊂V+⋐Yan\overline V\subset V^+\Subset Y^{\mathrm{an}} and choose the testers hj=hzjh_j=h_{z_j} of Lemma “A bounded model tester on the original domain”, normalized on this one V+V^+.

Relative CLN with background [1][1] bounds the coefficient variations of T[1],p(hj,…,hj)T_{[1],p}(h_j,\ldots,h_j) uniformly in jj on every fixed compact finite original presentation in V+V^+. The bounds use only 0≤hj≤10\leq h_j\leq1 there. Thus the positive series defining SpS_p converges absolutely in each such coefficient measure. The resulting measures satisfy the original relations and refinement compatibilities, and so define an original current. Positivity and symmetry pass term by term. Closedness follows by applying d′d' or d′′d'' to a fixed compact smooth test and summing the absolutely convergent pairings. The same estimate on the compact set V‾⊂V+\overline V\subset V^+ shows that σ(V)<∞\sigma(V)<\infty.

Fix jj and let Rj=Sp−2−jT[1],p(hj,…,hj)R_j=S_p-2^{-j}T_{[1],p}(h_j,\ldots,h_j). The current RjR_j is positive and separately closed. Finite additivity in the background follows directly from the defining recursion and linearity of coefficient Borel multiplication. Therefore, for q=n−pq=n-p, TSp,q(hj,…,hj)=2−jTT[1],p(hj,…,hj),q(hj,…,hj)+TRj,q(hj,…,hj)≥2−j(Dhj)n.\begin{split} T_{S_p,q}(h_j,\ldots,h_j) &=2^{-j}T_{T_{[1],p}(h_j,\ldots,h_j),q}(h_j,\ldots,h_j) +T_{R_j,q}(h_j,\ldots,h_j)\\ &\geq 2^{-j}(Dh_j)^n. \end{split} The first term is the indicated nn-fold product by its literal ordered recursion. In particular it has a positive atom at zjz_j. This proves (A) for SpS_p. If S−ϵSpS-\epsilon S_p is positive, the same finite decomposition proves it for SS. No assertion of continuity under arbitrary varying backgrounds is used to pass from the series to these products.

Finally, σ({zj})>0\sigma(\{z_j\})>0 for every jj. The set D\mathscr D is dense because every nonempty ordinary open contains one of the affinoids in the basis and hence a Shilov point. Thus σ\sigma has full support. Almost-everywhere order forces order at each of its positive atoms, and Lemma “A strict gap contains a Shilov point” then gives order everywhere. The conclusion uses these particular determining atoms, not merely the assertion that σ\sigma has full support. ◻

This construction and its proof are recorded as NEW_PROVED relative to the adopted inputs, without a claim of bibliographic priority. They exhibit backgrounds to which the conditional criterion applies. They do not identify all backgrounds with the desired property.

Limits of the conclusion and excluded arguments
  1. The domain in (G) is essential to the fixed-domain proof. On a connected proper positive-dimensional analytic space, the global maximum principle makes every finite global classical psh function constant, so the untwisted ambient capacity defined using such global functions is zero. Distinct constants then give nonempty capacity-zero strict gaps, while their Monge–Ampère measures both vanish. The first-round statements modulo capacity zero remain true; a pointwise uniqueness assertion on a proper space needs a normalization. No domain-independent detection theorem is claimed.

  2. A positive closed background may vanish off a proper analytic subset. In an ordinary ball disjoint from that subset, every product with this background vanishes by restriction. Nonempty strict gaps there have zero relative capacity. In particular ambient positivity does not furnish condition (A). The ambient-polar concentration example of the first round is fully compatible with the present results.

  3. Positive capacity on all nonempty ordinary opens, or full support of one trace measure, is not substituted for strict-gap detection. A gap of two upper-semicontinuous functions need not be ordinary open. Lemma “A strict gap contains a Shilov point” is the step that actually relates its pointwise content to detectable points.

  4. Local positivity of capacity on a smaller domain does not by itself prove positivity for the capacity of a larger prescribed domain: admissible competitors must extend to the latter. The ample frame and bounded closure in Lemma “A bounded model tester on the original domain” were used precisely to construct the tester on that larger fixed domain.

  5. An unconditional pointwise domination or uniqueness theorem for arbitrary SS is FALSE. For example, take S=0S=0 on the open unit disc, u=max⁡{log⁡∣T∣,−1}u=\max\{\log|T|,-1\}, and v=0v=0. Both functions are continuous bounded classical psh; their relative Monge–Ampère measures vanish. For 0<ϵ<10<\epsilon<1, the ordinary closure of {u<v−ϵ}\{u<v-\epsilon\} is contained in the compact closed disc of radius e−ϵ<1e^{-\epsilon}<1; for ϵ≥1\epsilon\geq1 the set is empty. The reverse gaps are empty. Thus the compact-gap hypotheses hold in both orders, but u<vu<v throughout the domain. What remains OPEN here is finding necessary conditions, or weaker useful sufficient conditions, for backgrounds not covered by (A). The atomic condition may be stronger than necessary. No implication from an unspecified smooth strict-positivity condition to detection is asserted.

A connected example with a nonzero background

The failure of unconditional pointwise domination is not limited to the zero current or disconnected spaces.

Example (A nonzero connected-domain counterexample). Let n≥2n\geq2, let R>1R>1, and let U={ ∣zi∣<R:0≤i<n }⊂(AKn)an,Y={z0=0}∩U,S=[Y].U=\{\,|z_i|<R:0\leq i<n\,\}\subset(\mathbf A_K^n)^{\mathrm{an}}, \qquad Y=\{z_0=0\}\cap U, \qquad S=[Y]. The open polydisc UU is connected and SS is a nonzero symmetric positive separately closed original current of bidegree (1,1)(1,1). Choose 0<r<10<r<1 and 0<c<log⁡(1/r)0<c<\log(1/r), and set b=max⁡{0,log⁡(∣z0−1∣/r),log⁡(∣z1∣/r),…,log⁡(∣zn−1∣/r)},u=b,v=max⁡{c,b}.b=\max\bigl\{0,\log(|z_0-1|/r), \log(|z_1|/r),\ldots, \log(|z_{n-1}|/r)\bigr\}, \qquad u=b,\quad v=\max\{c,b\}. Then u,vu,v are continuous bounded classical psh on UU. They satisfy the compact-gap condition in both orders and TS,n−1(u,…,u)=TS,n−1(v,…,v),TS,n−1(u,…,u)({u<v})=0,T_{S,n-1}(u,\ldots,u)=T_{S,n-1}(v,\ldots,v), \qquad T_{S,n-1}(u,\ldots,u)(\{u<v\})=0, but u<vu<v on a nonempty relatively compact ordinary polydisc. Consequently unconditional pointwise relative domination and uniqueness are FALSE even for a nonzero background on a connected domain satisfying (G).

Proof. Truncation by the constant zero makes the displayed maxima finite and continuous across the zero sets of the coordinates. Logarithms of analytic functions are classical psh after all complete extensions, and finite maxima and positive constant translations preserve that class. The coordinate bounds on UU give actual uniform upper bounds for bb; it is bounded below by zero.

Put ρ=rec<1\rho=re^c<1. Then {u<v}={b<c}={ ∣z0−1∣<ρ,∣zi∣<ρ (1≤i<n) }.\{u<v\}=\{b<c\} =\{\,|z_0-1|<\rho, |z_i|<\rho\ (1\leq i<n)\,\}. This polydisc is nonempty. Its closure is contained in the closed polydisc with the same center and radius. That closed polydisc is compact and lies in UU, since ∣z0∣=1<R|z_0|=1<R on it and ∣zi∣<1<R|z_i|<1<R for i≥1i\geq1. It is disjoint from YY. More precisely, for 0<ϵ<c0<\epsilon<c the gap {u<v−ϵ}\{u<v-\epsilon\} is the same polydisc with radius rec−ϵ<1re^{c-\epsilon}<1, while for ϵ≥c\epsilon\geq c it is empty. The reverse gaps are empty because v≥uv\geq u. These facts check every compact-gap quantifier.

On the ordinary neighborhood {∣z0∣<1}∩U\{|z_0|<1\}\cap U of YY, the non-Archimedean triangle inequality gives ∣z0−1∣=1|z_0-1|=1. Hence b≥log⁡(1/r)>cb\geq\log(1/r)>c there and u=vu=v there. Every recursively defined product with background [Y][Y] is supported on YY: Borel multiplication does not enlarge the support, and distributional differentiation does not enlarge it. The products for uu and vv are equal on this neighborhood by ordinary locality, and vanish outside YY; hence they are equal on UU. Their common measure assigns zero mass to the displayed gap because it is disjoint from YY. No base change and no modified calibration enter the example. ◻

Detection by backgrounds constructed from smooth factors

Atomic detection by SS-products is a sufficient condition, but the proof can also use a controlled family of smooth factors approaching the ambient atomic testers. The following construction makes no claim about whether its resulting measures have atoms.

Continue to assume (G). Use the determining set D={zj}\mathscr D=\{z_j\} and the continuous model testers hjh_j in Proposition “Backgrounds with detection in every degree”, normalized on the fixed V+V^+. For each jj choose smooth classical psh functions gj,mg_{j,m} on YanY^{\mathrm{an}} such that

gj,m↓hj,hj+4−m≤gj,m≤hj+3⋅4−m.(S)g_{j,m}\downarrow h_j,\qquad h_j+4^{-m}\leq g_{j,m}\leq h_j+3\cdot4^{-m}. \tag{S}

Here hjh_j denotes its original model weight with the same constant normalization, also defined on YanY^{\mathrm{an}}. These approximants exist on the whole required domain: after taking a generated power, the model metric is a finite pure-FS maximum of algebraic section ratios in its specified frame. The smooth regularized maxima of R, Proposition A.1 have uniformly bounded error everywhere; zeros of sections are inactive locally. Apply the constant-translation construction of R, Lemma A.2 to that global uniform error. The functions remain all-extensions classical by the convex unit-logarithm description. Set fj,m=12gj,m.f_{j,m}=\tfrac12 g_{j,m}. Thus 0≤fj,m≤10\leq f_{j,m}\leq1 on V+V^+, and fj,m↓hj/2f_{j,m}\downarrow h_j/2 on the same domain.

Proposition (A sufficient condition using smooth factors). Let 0≤p<n0\leq p<n and q=n−pq=n-p. Suppose that a symmetric positive separately closed original (p,p)(p,p)-current SS on VV has the following property: for every j,mj,m there is a constant aj,m>0a_{j,m}>0 such that

S−aj,m(Dfj,m)p is a positive current on V.(F)S-a_{j,m}(Df_{j,m})^p\text{ is a positive current on }V. \tag{F}

As usual, the right-hand factor is interpreted as (Dfj,m)p∧[1](Df_{j,m})^p\wedge[1], and equals [1][1] when p=0p=0. Then every nonempty strict gap of two members of Cb(V)\mathcal C_b(V) has positive Cap⁡S( ⋅ ;V)\operatorname{Cap}_S(\,\cdot\,;V). The pointwise relative domination and uniqueness conclusions hold under the first-round compact-gap hypotheses.

Proof. Finite additivity in the background and positivity give TS,q(fj,m,…,fj,m)≥aj,m(Dfj,m)n.T_{S,q}(f_{j,m},\ldots,f_{j,m}) \geq a_{j,m}(Df_{j,m})^n. Let u,v∈Cb(V)u,v\in\mathcal C_b(V) and δ≥0\delta\geq0, and suppose that E={u>v+δ}E=\{u>v+\delta\} has SS-capacity zero. Since each fj,mf_{j,m} is an admissible capacity tester, (Dfj,m)n(E)=0(Df_{j,m})^n(E)=0 for every j,mj,m. Put w=max⁡{u,v+δ}∈Cb(V)w=\max\{u,v+\delta\}\in\mathcal C_b(V). The nonnegative Borel function w−v−δw-v-\delta is zero off EE, so (w−v−δ)(Dfj,m)n=0.(w-v-\delta)(Df_{j,m})^n=0. Fix jj. Theorem “Weighted decreasing continuity for a fixed arbitrary background”, applied with fixed background [1][1] and differentiated slots fj,m↓hj/2f_{j,m}\downarrow h_j/2, passes this identity to the limit. The two nonconstant weights ww and vv are constant sequences in that theorem; the scalar constant δ\delta uses its unweighted case. Thus (w−v−δ)(D(hj/2))n=0.(w-v-\delta)(D(h_j/2))^n=0. All statements are local on compact original tests, so local boundedness of u,v,wu,v,w suffices. The limiting measure has a positive atom at zjz_j, which forces w(zj)−v(zj)−δ=0w(z_j)-v(z_j)-\delta=0, or u(zj)≤v(zj)+δu(z_j)\leq v(z_j)+\delta. Lemma “A strict gap contains a Shilov point” then gives u≤v+δu\leq v+\delta everywhere on VV. Hence EE was empty. The pointwise comparison consequences follow from the first-round results modulo capacity zero. ◻

Corollary (Explicit backgrounds and a faithful smooth trace measure). For every 0≤p<n0\leq p<n, the series S^p=∑j,m≥12−j−m(Dfj,m)p\widehat S_p=\sum_{j,m\geq1}2^{-j-m}(Df_{j,m})^p defines a positive separately closed original current on VV that satisfies (F). So does any background S≥ϵS^pS\geq\epsilon\widehat S_p with ϵ>0\epsilon>0. Moreover, σ^=∑j,m≥12−j−m(Dfj,m)n\widehat\sigma=\sum_{j,m\geq1}2^{-j-m}(Df_{j,m})^n is a finite positive Radon measure of full support on VV, formed from original smooth top-form trace measures. For u,v∈Cb(V)u,v\in\mathcal C_b(V), u≤vσ^-almost everywhere⟺u≤veverywhere on V.u\leq v\quad\widehat\sigma\text{-almost everywhere} \quad\Longleftrightarrow\quad u\leq v\quad\text{everywhere on }V.

Proof. Uniform coefficient bounds and convergence of the positive series follow from relative CLN, exactly as in Proposition “Backgrounds with detection in every degree”, since all fj,mf_{j,m} lie in [0,1][0,1] on the fixed V+V^+. Removing one summand leaves a positive separately closed current, which proves (F) with aj,m=2−j−ma_{j,m}=2^{-j-m}. The same argument applies after multiplication by ϵ\epsilon and addition of a positive background.

If u≤vu\leq v almost everywhere for σ^\widehat\sigma, the positive coefficients in its defining sum imply (Dfj,m)n({u>v})=0(Df_{j,m})^n(\{u>v\})=0 for every j,mj,m. Apply the weighted limit argument in the preceding proof with δ=0\delta=0. It gives pointwise order on D\mathscr D, hence everywhere. The reverse implication is immediate.

Finally take a nonempty ordinary open O⊂VO\subset V, choose zj∈O∩Dz_j\in O\cap\mathscr D, and choose a nonnegative original smooth cutoff supported in OO and equal to one near zjz_j. Unweighted monotone convergence gives ∫χ(Dfj,m)n⟶2−n∫χ(Dhj)n>0.\int\chi(Df_{j,m})^n \longrightarrow 2^{-n}\int\chi(Dh_j)^n>0. Some summand assigns positive mass to OO, and consequently σ^(O)>0\widehat\sigma(O)>0. This proves full support without asserting any atomic or non-atomic description of these smooth-factor products. ◻

The counterexample above and this construction are NEW_PROVED relative to the adopted inputs. The implication under condition (F) is a DIRECT_COROLLARY of fixed-background weighted convergence and the established Shilov detectors. These are sufficient conditions; a necessary characterization of pointwise-faithful backgrounds is not proved here.

A relative weighted L1L^1 Chern–Levine–Nirenberg estimate

The background is fixed throughout. Let XX be an allowed analytic space over the prescribed finite extension K/QpK/\mathbf Q_p, of pure dimension nn, and let SS be a symmetric, original-positive, separately d′d'- and d′′d''-closed current of bidegree (p,p)(p,p). Put q=n−pq=n-p and D=d′d′′D=d'd''. All products, coefficient measures and tests below belong to the original CLD calculus. No extension of SS to the additional test forms in CLD 2025 is used.

We prove a relative version of the weighted-form argument in CLD 2025, Proposition 8.4.3. The passage to bounded classical potentials uses the fixed-background calculus proved in R1. The result is recorded as NEW_PROVED in R2, with this method and these dependencies stated explicitly; no claim of literature priority is made.

The positive cutoff forms

Let OO be an ordinary open set and let ρ\rho be an original smooth psh function on OO, meaning that A=DρA=D\rho is a positive (1,1)(1,1)-form. Suppose that V={ρ<0}⋐O.V=\{\rho<0\}\Subset O. Write t=(−ρ)+t=(-\rho)_+ and B=d′ρ∧d′′ρB=d'\rho\wedge d''\rho. Fix a=4a=4. For 0≤k≤q0\leq k\leq q, put

βk=tk+aAq−k+(k+a)tk+a−1B∧Aq−k−1,\beta_k=t^{k+a}A^{q-k} +(k+a)t^{k+a-1}B\wedge A^{q-k-1},

where the second term is omitted for k=qk=q. The forms are extended by zero off VV. They have compact support in V‾\overline V, are strongly positive, and have continuous coefficients. In fact their coefficients are of class C2C^2 on each original moment presentation: the smallest power occurring is t3t^3.

The coefficient-Radon property of SS defines βk∧S\beta_k\wedge S and its products with smooth forms. Integration by parts with these C2C^2 cutoff forms is legitimate. To check this within the original test category, approximate their one-variable coefficient functions in C2C^2 by smooth functions of ρ\rho, and multiply by one fixed original smooth cutoff equal to one near V‾\overline V. The approximants have one compact support. The coefficient measures permit passage in every term involving at most two derivatives. The approximating forms need not be positive; positivity of the limiting forms follows directly from (eq:r2-cutoff-forms).

On VV, set b=k+ab=k+a, r=q−kr=q-k, and ck={2,k<q,1,k=q.c_k=\begin{cases}2,&k<q,\\1,&k=q.\end{cases} For 1≤k≤q1\leq k\leq q, direct differentiation gives

d′′βk=−ckb tb−1d′′ρ∧Ar,\begin{aligned} d''\beta_k &=-c_k b\,t^{b-1}d''\rho\wedge A^r, \end{aligned}

Dβk=ckb{−tb−1Ar+1+(b−1)tb−2B∧Ar}.\begin{aligned} D\beta_k &=c_k b\{-t^{b-1}A^{r+1} +(b-1)t^{b-2}B\wedge A^r\}. \end{aligned}

The endpoint k=qk=q is distinguished because βq=tq+a\beta_q=t^{q+a} has no second summand. The following proof uses both cases explicitly.

Lemma (Smooth relative weighted estimate). Assume q≥1q\geq1. Let F≤0F\leq0 and u1,…,uku_1,\ldots,u_k be original smooth psh functions on a neighborhood of V‾\overline V, where 0≤k≤q0\leq k\leq q. Put Mi=sup⁡V∣ui∣M_i=\sup_V|u_i|. Then

∫(−F) βk∧Du1∧⋯∧Duk∧S≤C(k,a)∏i=1kMi∫(−F) β0∧S,\int(-F)\,\beta_k\wedge Du_1\wedge\cdots\wedge Du_k\wedge S \leq C(k,a)\prod_{i=1}^kM_i \int(-F)\,\beta_0\wedge S,

where C(0,a)=1C(0,a)=1 and

C(k,a)=∏i=1k4e(a+i)(a+i+1)a+i−1.C(k,a)=\prod_{i=1}^{k} \frac{4e(a+i)(a+i+1)}{a+i-1}.

The constants do not depend on SS, FF, or the chosen functions.

Proof. All integrals have compact support. Write Ψk−1=Du1∧⋯∧Duk−1\Psi_{k-1}=Du_1\wedge\cdots\wedge Du_{k-1} and denote the left-hand side of (eq:r2-smooth-l1) by JkJ_k. We first estimate the last slot when its function uu is strictly positive on a neighborhood of V‾\overline V. Twice integrating by parts against the closed current Ψk−1∧S\Psi_{k-1}\wedge S gives

Jk=∫{u(FDβk−DF∧βk)+2F d′u∧d′′βk}∧Ψk−1∧S.J_k =\int\bigl\{u(FD\beta_k-DF\wedge\beta_k) +2F\,d'u\wedge d''\beta_k\bigr\} \wedge\Psi_{k-1}\wedge S.

Here and below cross terms can be replaced by their symmetric parts because the remaining current is symmetric. For example, the two cross terms in D(Fβk)D(F\beta_k) have equal pairings. Stokes applied to Fu d′′βk∧Ψk−1∧SFu\,d''\beta_k\wedge\Psi_{k-1}\wedge S yields (eq:r2-integration-identity), with the displayed signs.

Put w=−F≥0w=-F\geq0 and Γ(f,g)=12(d′f∧d′′g+d′g∧d′′f),Γ(f)=d′f∧d′′f.\Gamma(f,g)=\tfrac12(d'f\wedge d''g+d'g\wedge d''f), \qquad \Gamma(f)=d'f\wedge d''f. For every positive scalar hh, the positive-square identity gives 2Γ(u,ρ)≤h−1Γ(u)+hB.2\Gamma(u,\rho)\leq h^{-1}\Gamma(u)+hB. Apply this pointwise on VV with h=ckbu/th=c_kbu/t. Equations (eq:r2-dbeta)–(eq:r2-integration-identity), and positivity of uDF∧βkuDF\wedge\beta_k, give

Jk≤∫wu{−Dβk+ck2b2tb−2B∧Ar}∧Ψk−1∧S+∫w tbu−1Γ(u)∧Ar∧Ψk−1∧S.\begin{aligned} J_k &\leq\int wu\{-D\beta_k+c_k^2b^2t^{b-2}B\wedge A^r\} \wedge\Psi_{k-1}\wedge S\\ &\hspace{1cm} +\int w\,t^b u^{-1}\Gamma(u)\wedge A^r \wedge\Psi_{k-1}\wedge S. \end{aligned}

The form in braces is ckb tb−1Ar+1+ckb((ck−1)b+1)tb−2B∧Ar.c_kb\,t^{b-1}A^{r+1} +c_kb\bigl((c_k-1)b+1\bigr)t^{b-2}B\wedge A^r. Since ckc_k is either 11 or 22, it is bounded above by Ckβk−1,Ck=2b(b+1)b−1.C_k\beta_{k-1},\qquad C_k=\frac{2b(b+1)}{b-1}. Also tbAr≤βkt^bA^r\leq\beta_k. Wedging each displayed inequality with the other strongly positive factors and then with the original-positive current SS preserves it. We obtain

Jk(F;u1,…,uk−1,u)≤Ck∥u∥VJk−1+∫w βk∧Ψk−1∧u−1Γ(u)∧S.J_k(F;u_1,\ldots,u_{k-1},u) \leq C_k\|u\|_{V}J_{k-1} +\int w\,\beta_k\wedge\Psi_{k-1}\wedge u^{-1}\Gamma(u)\wedge S.

Now let uku_k have arbitrary sign and set M=sup⁡V∣uk∣M=\sup_V|u_k|. If M=0M=0, its curvature vanishes on VV and there is nothing to prove. Otherwise put z=exp⁡(B0uk)z=\exp(B_0u_k), where B0>0B_0>0. Then Dz=z−1Γ(z)+B0z Duk≥z−1Γ(z)+B0e−B0MDuk.Dz=z^{-1}\Gamma(z)+B_0z\,Du_k \geq z^{-1}\Gamma(z)+B_0e^{-B_0M}Du_k. Apply (eq:r2-positive-slot) to zz. The energy term cancels with the identical term in this lower bound for DzDz, leaving B0e−B0MJk≤CkeB0MJk−1.B_0e^{-B_0M}J_k\leq C_ke^{B_0M}J_{k-1}. Choose B0=(2M)−1B_0=(2M)^{-1}. This gives Jk≤4e(a+k)(a+k+1)a+k−1 MkJk−1.J_k\leq\frac{4e(a+k)(a+k+1)}{a+k-1}\,M_kJ_{k-1}. Induction from J0J_0 proves the assertion. ◻

Bounded potentials and negative psh witnesses

We now work in a common algebraic norm-box neighborhood of the kind used in R1. The compact set V‾\overline V lies in a prescribed smaller norm box, and all potentials are defined on a larger one. Thus the compact-box regularization lemma of R1 gives smooth decreasing classical psh approximants on a neighborhood of the entire V‾\overline V. This domain condition is only local geometry; it places no restriction on the positive closed current SS.

For a negative extended-real classical psh function ϕ:O→[−∞,0]\phi:O\to[-\infty,0], put ϕN=max⁡{ϕ,−N}∈Cb(O),AS,ρ(ϕ)=∫(−ϕ) β0∧S∈[0,∞].\phi_N=\max\{\phi,-N\}\in\mathcal C_b(O), \qquad A_{S,\rho}(\phi)=\int(-\phi)\,\beta_0\wedge S\in[0,\infty]. This is an integral of the actual Borel function against one specified positive Radon measure. It is not an ambient-volume integrability condition.

Theorem (Uniform truncation estimate over an arbitrary background). Assume q≥1q\geq1 and AS,ρ(ϕ)<∞A_{S,\rho}(\phi)<\infty. Let K⋐VK\Subset V and let u1,…,uq∈Cb(O)u_1,\ldots,u_q\in\mathcal C_b(O) have pointwise bounds ∣ui∣≤Mi|u_i|\leq M_i on a fixed neighborhood of V‾\overline V. Then, with δK=min⁡K(−ρ)>0\delta_K=\min_K(-\rho)>0,

sup⁡N≥1∫K(−ϕN)TS,q(u1,…,uq)≤δK−(q+a)C(q,a)∏i=1qMi AS,ρ(ϕ).\sup_{N\geq1}\int_K(-\phi_N) T_{S,q}(u_1,\ldots,u_q) \leq \delta_K^{-(q+a)} C(q,a) \prod_{i=1}^qM_i\,A_{S,\rho}(\phi).

In particular, ϕ\phi is integrable on KK for every such mixed measure, and the same bound holds with −ϕ-\phi in place of −ϕN-\phi_N.

Proof. First fix NN and ϵ>0\epsilon>0. The bounded classical psh function F=ϕN−ϵF=\phi_N-\epsilon is at most −ϵ-\epsilon. Choose a smooth decreasing row Fj↓FF_j\downarrow F on one neighborhood of V‾\overline V. After a prefix is discarded, Fj≤−ϵ/2F_j\leq-\epsilon/2 there. This follows from compactness applied to the increasing open sets where the continuous majorants are less than −ϵ/2-\epsilon/2.

Simultaneously choose smooth decreasing classical psh rows ui,j↓uiu_{i,j}\downarrow u_i on the same neighborhood. Given ε>0\varepsilon>0, a prefix makes ∣ui,j∣≤Mi+ε|u_{i,j}|\leq M_i+\varepsilon on V‾\overline V. The lower bound follows from ui,j≥uiu_{i,j}\geq u_i; the upper bound is the same compactness argument. All choices are made for this fixed NN. No regularization theorem for the unbounded function ϕ\phi is assumed.

Apply Lemma “Smooth relative weighted estimate” with k=qk=q. The R1 weighted decreasing theorem, in top degree against continuous compact scalar tests, passes the left side to ∫(−ϕN+ϵ) βq TS,q(u1,…,uq).\int(-\phi_N+\epsilon)\,\beta_q\, T_{S,q}(u_1,\ldots,u_q). On the right, bounded convergence is against the fixed measure β0∧S\beta_0\wedge S. Let ε↓0\varepsilon\downarrow0 and then ϵ↓0\epsilon\downarrow0. We obtain ∫(−ϕN) βq TS,q(u1,…,uq)≤C(q,a)∏iMi∫(−ϕN) β0∧S.\int(-\phi_N)\,\beta_q\, T_{S,q}(u_1,\ldots,u_q) \leq C(q,a)\prod_iM_i\int(-\phi_N)\,\beta_0\wedge S. Because βq=tq+a≥δKq+a\beta_q=t^{q+a}\geq\delta_K^{q+a} on KK and −ϕN≤−ϕ-\phi_N\leq-\phi, this is (eq:r2-uniform-truncation). Finally −ϕN↑−ϕ-\phi_N\uparrow-\phi. Monotone convergence for each fixed mixed measure proves the integrability assertion and its bound. ◻

Corollary (Uniformity over normalized potentials). Under the same hypotheses, sup⁡N≥1 sup⁡v∈Cb(O)0≤v≤1∫K(−ϕN) TS,q(v,…,v)≤δK−(q+a)C(q,a)AS,ρ(ϕ).\sup_{N\geq1}\ \sup_{\substack{v\in\mathcal C_b(O)\\0\leq v\leq1}} \int_K(-\phi_N)\,T_{S,q}(v,\ldots,v) \leq \delta_K^{-(q+a)}C(q,a)A_{S,\rho}(\phi). The same assertion holds with different normalized potentials in the differentiated slots.

Corollary (A precise relative polar conclusion). For the capacity with the fixed admissibility domain OO, Cap⁡S(E;O)=sup⁡v∈Cb(O)0≤v≤1∫ETS,q(v,…,v),\operatorname{Cap}_{S}(E;O) =\sup_{\substack{v\in\mathcal C_b(O)\\0\leq v\leq1}} \int_E T_{S,q}(v,\ldots,v), one has, for every N>0N>0, Cap⁡S(K∩{ϕ≤−N};O)≤δK−(q+a)C(q,a)NAS,ρ(ϕ).\operatorname{Cap}_{S}\bigl(K\cap\{\phi\leq-N\};O\bigr) \leq \frac{\delta_K^{-(q+a)}C(q,a)}{N} A_{S,\rho}(\phi). Hence K∩{ϕ=−∞}K\cap\{\phi=-\infty\} has zero relative capacity and zero mass for every bounded mixed measure in Theorem “Uniform truncation estimate over an arbitrary background”.

Proof. On the indicated set, −ϕ≥N-\phi\geq N. Apply the integrability bound in Theorem “Uniform truncation estimate over an arbitrary background” and take the supremum. The pole set is contained in every such level set. Its zero mass also follows directly from the proved integrability of ϕ\phi. ◻

Corollary (Outer capacity and quasicontinuity of an integrable witness). Keep the hypotheses of Theorem “Uniform truncation estimate over an arbitrary background”, and let WW be ordinary open with W‾⊂K⋐V\overline W\subset K\Subset V. Define outer capacity by Cap⁡S∗(E;O)=inf⁡G⊂O ordinary openE⊂GCap⁡S(G;O).\operatorname{Cap}_S^*(E;O) =\inf_{\substack{G\subset O\text{ ordinary open}\\E\subset G}} \operatorname{Cap}_S(G;O). Write CK=δK−(q+a)C(q,a)AS,ρ(ϕ)C_K=\delta_K^{-(q+a)}C(q,a)A_{S,\rho}(\phi). Then, for N>0N>0,

Cap⁡S(W∩{ϕ<−N};O)≤CK/N.\operatorname{Cap}_S(W\cap\{\phi<-N\};O)\leq C_K/N.

Consequently W∩{ϕ=−∞}W\cap\{\phi=-\infty\} has outer capacity zero. The truncations ϕN\phi_N converge to ϕ\phi locally in outer capacity: for every δ>0\delta>0, Cap⁡S∗(W∩{ϕN−ϕ>δ};O)≤CK/N⟶0,\operatorname{Cap}_S^* (W\cap\{\phi_N-\phi>\delta\};O)\leq C_K/N \longrightarrow0, where the difference is +∞+\infty at a pole of ϕ\phi. Finally, for every ϵ>0\epsilon>0 there is an ordinary open G⊂OG\subset O with Cap⁡S(G;O)<ϵ\operatorname{Cap}_S(G;O)<\epsilon such that ϕ∣W∖G\phi|_{W\setminus G} is finite and continuous. These are DIRECT_COROLLARY assertions; no outer-regularity theorem for arbitrary Borel sets is assumed.

Proof. The function ϕ\phi is upper semicontinuous, so HN=W∩{ϕ<−N}H_N=W\cap\{\phi<-N\} is ordinary open. It is contained in K∩{ϕ≤−N}K\cap\{\phi\leq-N\}, so Corollary “A precise relative polar conclusion” proves (eq:r2-open-sublevel-bound). Each HNH_N contains the pole set in WW, giving its zero outer capacity. Since ϕN=ϕ\phi_N=\phi off {ϕ<−N}\{\phi<-N\}, the same ordinary open set covers every displayed truncation exceptional set.

For the last assertion choose NN with CK/N<ϵ/2C_K/N<\epsilon/2. The bounded classical function ϕN\phi_N is quasicontinuous by R1. Precisely, cover W‾\overline W by finitely many smaller norm-box cores in OO and apply the bounded local quasicontinuity theorem on each larger box with a total error less than ϵ/2\epsilon/2. Restriction of admissible functions gives Cap⁡S(E;O)≤Cap⁡S(E;Ω)\operatorname{Cap}_S(E;O)\leq\operatorname{Cap}_S(E;\Omega) for sets in any of these boxes Ω⊂O\Omega\subset O. Thus the union GN⊂OG_N\subset O of their ordinary open exceptional sets has Cap⁡S(GN;O)<ϵ/2\operatorname{Cap}_S(G_N;O)<\epsilon/2, and ϕN\phi_N is continuous on W∖GNW\setminus G_N. Put G=GN∪HNG=G_N\cup H_N. Countable subadditivity bounds its capacity by ϵ\epsilon. On W∖GW\setminus G one has ϕ=ϕN≥−N\phi=\phi_N\geq-N, proving the assertion in the subspace topology. ◻

Intrinsic local hypotheses and scope

One sufficient hypothesis, independent of a particular barrier, is local coefficient integrability of ϕ\phi with respect to SS:

∫∣ϕ∣ dμS,Θ<∞for every positive original compact smooth (q,q)-test Θ.\int|\phi|\,d\mu_{S,\Theta}<\infty \quad\hbox{for every positive original compact smooth $(q,q)$-test $\Theta$.}

The finite positive controls of R imply that this is equivalent to integrability against all local coefficient variations. They also dominate the continuous-coefficient compact form β0\beta_0 by finitely many positive smooth controls. Therefore (eq:r2-coefficient-integrability) implies AS,ρ(ϕ)<∞A_{S,\rho}(\phi)<\infty for every barrier used above.

The R1 collar construction supplies the required barriers locally: take a smooth psh function bb which is zero on the inner norm box and large on an outer collar, and put ρ=b−c\rho=b-c with c>0c>0 between the two levels. Then {ρ<0}⋐O\{\rho<0\}\Subset O and contains the chosen core. No strict positivity of DρD\rho is required. A finite covering gives the stated integrability and mass bounds on arbitrary compact subsets for which (eq:r2-coefficient-integrability) holds.

Lower-degree coefficient estimates follow by inserting bounded smooth quadratic moment potentials into the remaining slots and using R’s finite-test domination. Thus the same hypothesis makes ϕTS,k(u1,…,uk)\phi T_{S,k}(u_1,\ldots,u_k) a coefficient-Radon current for every bounded tuple and every k≤qk\leq q.

For q=0q=0, the asserted uniform truncation bound is simply integrability of ϕ\phi against the fixed top-degree measure SS; no differentiated slot is present. Finally, the capacity conclusion keeps its admissibility domain OO. To obtain the corresponding statement on a smaller domain, choose its collar and verify its integrability condition there; monotonicity under domain restriction does not transfer the displayed upper bound automatically.

The hypothesis on ϕ\phi and SS is essential to the scope of this result. If S=[Y]S=[Y] and ϕ\phi is identically −∞-\infty on YY, it is generally not satisfied. Nothing here asserts that arbitrary relative products avoid every ambient pluripolar set. Nor does the uniform bound alone establish weighted convergence when both an unbounded witness and its background measure vary.

Restriction of a positive closed current to a complete pole set

Let SS be a symmetric original-positive, separately d′d'- and d′′d''-closed current of bidegree (p,p)(p,p) on an allowed ordinary open UU, and put q=n−pq=n-p. Let ϕ:U→[−∞,∞)\phi:U\to[-\infty,\infty) be an extended-real classical psh function, and set P={ϕ=−∞}.P=\{\phi=-\infty\}. All restrictions in this section mean Borel multiplication of the actual coefficient-Radon measures. In particular, they are not restrictions to a tropical image. The proof uses the bounded capacitary closure developed in this record and the original CLD positive cone of R throughout.

A first-order estimate on a fixed test presentation

For a bounded classical psh function aa, define the distribution

ΓS(a)=12D(a2S)−aD(aS).\Gamma_S(a)=\tfrac12D(a^2S)-aD(aS).

It is a positive coefficient-Radon current of bidegree (p+1,p+1)(p+1,p+1). Indeed, locally add a constant so that a≥0a\geq0 and use a decreasing smooth classical psh row. The square of each term is classical psh and decreases to a2a^2. The first-round weighted theorem passes (eq:r2-relative-gamma-current) to the limit of the positive currents d′aj∧d′′aj∧Sd'a_j\wedge d''a_j\wedge S. Adding the constant does not change (eq:r2-relative-gamma-current), since SS is separately closed. Positivity and R’s coefficient-Radon theorem give the assertion. Wedging with a fixed closed smooth trace form recovers the first-round relative energy measures.

Lemma (First-order relative Cauchy–Schwarz estimate). Suppose q≥1q\geq1. Localize an original compact smooth test η\eta of bidegree (q−1,q)(q-1,q) to one finite moment presentation with coordinates t1,…,trt_1,\ldots,t_r. Let β=∑a=1rd′ta∧d′′ta,\beta=\sum_{a=1}^r d't_a\wedge d''t_a, and choose a nonnegative compact original smooth cutoff χ\chi equal to one near supp⁡η\operatorname{supp}\eta. There is a constant Cη<∞C_\eta<\infty, independent of SS and of the bounded classical psh function aa, such that

∣⟨d′(aS),η⟩∣2≤Cη⟨ΓS(a),χβq−1⟩⟨S,χβq⟩.\bigl|\langle d'(aS),\eta\rangle\bigr|^2 \leq C_\eta \langle\Gamma_S(a),\chi\beta^{q-1}\rangle \langle S,\chi\beta^q\rangle.

The same assertion holds for d′′d'' and tests of bidegree (q,q−1)(q,q-1). The constant remains valid when the moment presentation is refined to include additional coordinates used by aa or its approximants.

Proof. First let aa be smooth and write ξ=d′a\xi=d'a. The finite expansion of η\eta has terms fI,Jd′tI∧d′′tJf_{I,J}d't_I\wedge d''t_J, where ∣I∣=q−1|I|=q-1 and ∣J∣=q|J|=q. Under the standard symmetric-matrix convention for original (q,q)(q,q)-forms, the symmetric part of ξ∧η\xi\wedge\eta is a finite sum, with fixed signs, of the symmetric tensors between AI=ξ∧d′tI,BJ=d′tJ.A_I=\xi\wedge d't_I,\qquad B_J=d't_J. For two arbitrary qq-covectors A,BA,B, the square (ϵA±ϵ−1/2B)2(\sqrt\epsilon A\pm\epsilon^{-1/2}B)^2 is in the original positive cone. The associated matrix Young inequality therefore bounds either sign of their symmetric cross tensor by ϵA2+ϵ−1B2\epsilon A^2+\epsilon^{-1}B^2, up to a fixed harmless factor. Apply it to the finitely many terms and use the fixed bounds of the coefficients fI,Jf_{I,J} on the test support. Summing the AI2A_I^2 terms gives a fixed multiple of d′a∧d′′a∧βq−1d'a\wedge d''a\wedge\beta^{q-1}; summing the BJ2B_J^2 terms is bounded by a fixed multiple of βq\beta^q. After rescaling ϵ\epsilon, we obtain, for every ϵ>0\epsilon>0, ±Sym⁡(d′a∧η)≤ϵχ d′a∧d′′a∧βq−1+C′ ⁣ηϵχβq.\pm\operatorname{Sym}(d'a\wedge\eta) \leq \epsilon\chi\,d'a\wedge d''a\wedge\beta^{q-1} +\frac{C'\!_\eta}{\epsilon}\chi\beta^q. Outside the test support the left side is zero. The cutoff makes the right side a compact positive test. No strict positivity in additional analytic directions is used: AIA_I contains the full covector d′ad'a, and only its remaining q−1q-1 factors come from the fixed frame. The same matrix-square argument works on every common refinement, with the same finite index sets and constants.

The current SS annihilates the antisymmetric component, by the original-positive coefficient argument in R. Evaluation and separate closedness give ∣⟨d′(aS),η⟩∣≤ϵE+ϵ−1C′ ⁣ηM,E=⟨ΓS(a),χβq−1⟩,M=⟨S,χβq⟩.|\langle d'(aS),\eta\rangle| \leq\epsilon E+\epsilon^{-1}C'\!_\eta M, \quad E=\langle\Gamma_S(a),\chi\beta^{q-1}\rangle, \quad M=\langle S,\chi\beta^q\rangle. Optimizing proves (eq:r2-first-order-CS); if EE or MM is zero, take a limit in ϵ\epsilon to obtain the same conclusion.

For bounded classical aa, choose a smooth decreasing row on a fixed neighborhood of the compact test stage, as in R1. Multiplication against the fixed coefficient measures of SS gives ajS→aSa_jS\to aS. The weighted first-round theorem gives ΓS(aj)→ΓS(a)\Gamma_S(a_j)\to\Gamma_S(a), as explained above. Pass to the limit in the smooth inequality. The involution JJ, or the identical calculation with primed factors interchanged, proves the d′′d'' assertion. A finite partition handles every original compact test, not only those in one initially fixed chart. ◻

The idempotent identity

Lemma (Cubic chain identity in the bounded capacitary closure). Let 0≤a≤M0\leq a\leq M be a member of AS,b\mathcal A_{S,b}, locally with a common bound for its defining approximants. Then

D(a3S)=3aD(a2S)−3a2D(aS).D(a^3S)=3aD(a^2S)-3a^2D(aS).

The square and cube belong to the same local capacitary closure. Moreover (eq:r2-relative-gamma-current) is a positive coefficient-Radon current, and it is continuous under bounded outer-capacity convergence of classical approximants.

Proof. For a smooth function, the usual product rule gives D(a2S)=2aD(aS)+2d′a∧d′′a∧S,D(a^2S)=2aD(aS)+2d'a\wedge d''a\wedge S, D(a3S)=3a2D(aS)+6a d′a∧d′′a∧S,D(a^3S)=3a^2D(aS)+6a\,d'a\wedge d''a\wedge S, which imply (eq:r2-cubic-chain). For bounded nonnegative classical aa, choose decreasing smooth nonnegative classical approximants. Their squares and cubes are again classical psh and decrease to the corresponding powers. Pass the two weighted one-slot terms in (eq:r2-cubic-chain) by the first-round weighted theorem; pass the left side distributionally against the fixed SS.

For a nonnegative member of the capacitary closure, replace its classical defining approximants by their maxima with zero. The approximants remain uniformly bounded and converge in outer capacity. Squaring and cubing are Lipschitz on that bounded range, so their powers also converge in outer capacity. They are classical psh by the preceding paragraph and decreasing closure, hence define a2,a3a^2,a^3 in the capacitary class. The completed weighted theorem now passes every term of the classical identity. The same argument passes 12D(aj2S)−ajD(ajS)\tfrac12D(a_j^2S)-a_jD(a_jS) to (eq:r2-relative-gamma-current). Positivity passes from the classical energy currents, and R gives coefficient-Radon type. ◻

Theorem (Positive closed restriction to a complete pole set). For the arbitrary background SS and the complete pole set PP specified above, both currents SP=1PS,SPc=1U∖PSS_P=\mathbf1_P S,\qquad S_{P^c}=\mathbf1_{U\setminus P}S are symmetric, original-positive, and separately d′d'- and d′′d''-closed. For every 0≤k≤q0\leq k\leq q and u1,…,uk∈Cb(U)u_1,\ldots,u_k\in\mathcal C_b(U), one has

1PTS,k(u1,…,uk)=TSP,k(u1,…,uk),\mathbf1_P T_{S,k}(u_1,\ldots,u_k) =T_{S_P,k}(u_1,\ldots,u_k),

and the analogous identity for PcP^c, on every original compact smooth complementary test form.

Status: NEW_PROVED, relative to R1 and the proved R2 capacitary calculus. This is a restriction theorem for a current already defined on UU. It is not an extension theorem for a current initially defined only on U∖PU\setminus P.

Proof. Positivity, symmetry and coefficient-Radon type of both Borel restrictions are immediate. In top degree q=0q=0, separate closedness is automatic and there are no differentiated slots. Suppose therefore q≥1q\geq1.

Work locally and subtract a constant from ϕ\phi so that ϕ≤0\phi\leq0 on the neighborhood. Set wN=1+max⁡{ϕ,−N}N,N≥1.w_N=1+\frac{\max\{\phi,-N\}}{N},\qquad N\geq1. Then 0≤wN≤10\leq w_N\leq1, the functions are classical psh, and they increase pointwise to the actual Borel function w=1Pcw=\mathbf1_{P^c}. The local increasing-convergence theorem puts ww in AS,b\mathcal A_{S,b} and gives outer-capacity convergence of wNw_N to ww. Its square and cube equal itself. The cubic identity therefore gives D(wS)=3wD(wS)−3wD(wS)=0,ΓS(w)=12D(wS)−wD(wS)=0.D(wS)=3wD(wS)-3wD(wS)=0, \qquad \Gamma_S(w)=\tfrac12D(wS)-wD(wS)=0. Completed weighted convergence gives wNS⟶wS,ΓS(wN)⟶0.w_NS\longrightarrow wS, \qquad \Gamma_S(w_N)\longrightarrow0. Fix an arbitrary first-order original compact test and localize it to the presentations of Lemma “First-order relative Cauchy–Schwarz estimate”. The fixed factor ⟨S,χβq⟩\langle S,\chi\beta^q\rangle is finite, whereas the energy factor tends to zero. The lemma implies d′(wNS)→0d'(w_NS)\to0 on that test. Distributional continuity also gives d′(wNS)→d′(wS)d'(w_NS)\to d'(wS), so d′(wS)=0d'(wS)=0. The same proof gives d′′(wS)=0d''(wS)=0. Subtracting from the separately closed SS proves closedness of SPS_P. These local conclusions glue because the Borel restrictions have already been defined globally.

For (eq:r2-polar-restriction-BT), first take k=1k=1 and a smooth classical psh function uu. Separate closedness just proved and smooth Borel multiplication give D(uSP)=Du∧SP=1P(Du∧S)=1PD(uS).D(uS_P)=Du\wedge S_P =\mathbf1_P(Du\wedge S)=\mathbf1_PD(uS). For general u∈Cbu\in\mathcal C_b, use a decreasing smooth classical row uj↓uu_j\downarrow u on a fixed compact norm-box stage. It has one common bound on the slightly larger domain. The left side converges by the fixed-background first-round theorem for SPS_P. The currents D(ujS)D(u_jS) converge to D(uS)D(uS) and satisfy uniform coefficient bounds by the top relative capacity of that same domain. The multiplier 1P=1−w\mathbf1_P=1-w is quasicontinuous for this capacity because w∈AS,bw\in\mathcal A_{S,b}. The R2 quasicontinuous multiplier theorem therefore passes the right side to 1PD(uS)\mathbf1_PD(uS). This proves the one-slot identity on all original tests without asserting stability for arbitrary varying backgrounds.

For several slots, apply the one-slot identity to the arbitrary positive closed background Q=TS,k−1(u2,…,uk)Q=T_{S,k-1}(u_2,\ldots,u_k). The restriction theorem already applies to this QQ and the same classical pole set. Induction gives 1PQ=TSP,k−1(u2,…,uk),\mathbf1_PQ=T_{S_P,k-1}(u_2,\ldots,u_k), and hence 1PD(u1Q)=D(u11PQ)=TSP,k(u1,…,uk).\mathbf1_PD(u_1Q)=D(u_1\mathbf1_PQ) =T_{S_P,k}(u_1,\ldots,u_k). The complement follows in the same way, or by finite linearity in the decomposition S=SP+SPcS=S_P+S_{P^c}. ◻

Corollary (Exact preservation of pole-set mass). If 1PS=0\mathbf1_PS=0, then 1PTS,k(u1,…,uk)=0\mathbf1_PT_{S,k}(u_1,\ldots,u_k)=0 for every bounded classical tuple. This conclusion needs no integrability estimate for the original witness. Conversely, the theorem allows nonzero SPS_P; it does not assert that arbitrary relative products avoid ambient pole sets.

Proof. Apply (eq:r2-polar-restriction-BT) with SP=0S_P=0. ◻

The ambient vanishing needs a separate argument.

For S=[1]S=[1], the theorem shows that 1P[1]\mathbf1_P[1] is a positive separately closed original current of degree zero. This calculation alone does not imply its vanishing. Under the connectedness and nontriviality assumptions on the witness, Theorem “Vanishing of ambient smooth traces on a classical pole set” proves that vanishing by an independent argument using divisorial points and bounded model approximations. No distributional elliptic regularity is invoked. CLD 2025, Remark 7.8.3, explicitly distinguishes the equation D[u]=0D[u]=0 from a continuity theorem on its general tropical spaces, and Proposition 8.1.12 retains local integrability as an extra condition in its psh class. The proof in the cited ambient trace theorem supplies the additional analytic input, rather than silently interpreting a closed degree-zero density as a continuous pluriharmonic function.

Divisorial points are not poles of a nontrivial classical psh function

The field, ordinary topology, and all-extensions classical definition remain fixed. The geometric lemma below uses strict affinoid domains inside the algebraic neighborhoods supplied by B. Its normalization step is essential: a formal model with only one irreducible special-fiber component need not have a unique Shilov point unless the corresponding generic-point fiber is controlled.

Lemma (A domain with one prescribed Shilov point). Let UU be an ordinary open in an allowed good, Hausdorff, boundaryless, quasi-smooth, pure nn-dimensional KK-analytic space, with n≥1n\geq1. If z∈Uz\in U is divisorial, there exists a strict affinoid domain A⊂UA\subset U such that Γ(A)={z},Int⁡U(A)≠∅.\Gamma(A)=\{z\},\qquad \operatorname{Int}_U(A)\ne\varnothing. The lemma does not assert that zz itself belongs to this ordinary interior. In fact it belongs to the boundary of AA.

Proof. Work in a strict algebraic neighborhood of zz, as permitted by B, and choose a strict affinoid neighborhood C⊂UC\subset U of zz. Divisoriality is unchanged on passing to CC, by [GM19, Corollary A.6, p. 367]. By the definition of a divisorial point, equivalently [GM19, Proposition A.5, pp. 366–367], there is an admissible formal K∘K^\circ-model C\mathfrak C of CC such that the reduction of zz is the generic point of a special-fiber component.

We may and do make this formal model normal. Indeed, CC is normal because it is quasi-smooth. The formal affine rings are topologically of finite type over the excellent complete discrete valuation ring K∘K^\circ. Their normalizations in their reduced generic fibers are finite, and glue to a normal admissible formal model. The generic fiber remains CC, since it is already normal. The reduction of zz on this new model is still generic in the special fiber: it lies over the former generic point under a finite integral map, and a prime above a minimal prime over the uniformizer is again minimal over that uniformizer, by incomparability. Denote the new generic point by η\eta and its component by EE.

Remove the other irreducible components of the special fiber and choose a formal affine open neighborhood A\mathfrak A of η\eta inside the remaining formal open. Such a choice is possible because the formal model is locally noetherian and η\eta lies on no other component; one first works in an affine neighborhood and then avoids its finitely many other components. The special fiber of A\mathfrak A has exactly one irreducible component, an open part of EE. Set A=AηA=\mathfrak A_\eta, its generic fiber. Formal open immersions induce analytic-domain inclusions on generic fibers, so AA is a strict affinoid domain of CC, hence of UU, and it contains zz.

The model A\mathfrak A is normal and its generic fiber is reduced. The uniqueness of the point over its special-fiber generic point is the normal-model assertion in [CLD12, Proposition 6.9.7, pp. 119–120]. It can also be read directly at the height-one local ring of the normal model: the resulting discrete valuation, normalized by the prescribed absolute value of the uniformizer, determines that point. By [GM19, Proposition A.3, p. 365], the points reducing to generic points of this formal special fiber are exactly the Shilov boundary of AA. Hence Γ(A)={z}\Gamma(A)=\{z\}.

We check the ordinary-interior assertion separately. The special fiber of A\mathfrak A has a closed point ξ\xi, and the reduction map is surjective, so some y∈Ay\in A reduces to ξ\xi. The reduced closure of ξ\xi is a finite scheme over K~\widetilde K, hence is proper. The formal reduction criterion [CLD12, Lemma 6.5.1, p. 105] therefore gives y∈Int⁡(A/K)y\in\operatorname{Int}(A/K). Equivalently, one can use the analytic Nullstellensatz and the fact that rigid points of a strict analytic space over a nontrivially valued field do not belong to its analytic boundary; see [CLD12, Remark 4.3.9, p. 67].

For the analytic-domain inclusion A↪UA\hookrightarrow U, relative interior is ordinary topological interior. Since UU is boundaryless and locally separated, the composition formula gives Int⁡U(A)=Int⁡(A/U)=Int⁡(A/K);\operatorname{Int}_U(A) =\operatorname{Int}(A/U)=\operatorname{Int}(A/K); these are [GR25, §2.11(3)–(4), pp. 7–8]. Thus the ordinary interior is nonempty. Finally AA is pure of dimension nn, and its Shilov points belong to its analytic boundary by [GR25, Lemma 6.2, p. 31]. The same identification of interiors shows that z∉Int⁡U(A)z\notin\operatorname{Int}_U(A), as asserted. ◻

The geometric lemma is a DIRECT_COROLLARY of the cited formal-model and boundary facts, with the needed normalization spelled out. It makes no field extension. Rigid or reduction points with a finite residue-field extension, when used in the proof, are points of the original KK-analytic space.

Remark (Why one component alone is insufficient). Over the same discrete valuation ring with uniformizer π\pi, the formal scheme Spf⁡(K∘⟨T⟩/(T2−π2))\operatorname{Spf}\bigl(K^\circ\langle T\rangle/(T^2-\pi^2)\bigr) has only one irreducible special-fiber component, but its generic fiber consists of two rigid points, T=πT=\pi and T=−πT=-\pi, and its Shilov boundary consists of both. The characteristic of KK is zero, so these points are distinct even when the residue characteristic is two. This formal scheme is not normal. Thus replacing the normalization paragraph by a claim that “one component implies one Shilov point” would be an invalid argument.

Proposition (A nontrivial pole set has empty ordinary interior). Let UU satisfy the allowed analytic hypotheses and be connected. Let ϕ:U→R∪{−∞}\phi:U\to\mathbf R\cup\{-\infty\} be classical psh and not identically −∞-\infty. Then P={ϕ=−∞}P=\{\phi=-\infty\} has empty ordinary interior in UU.

Proof. For integers N≥1N\geq1 define hN=max⁡{ϕ,−N}N.h_N=\frac{\max\{\phi,-N\}}{N}. These are members of Cb(U)\mathcal C_b(U) and are bounded below by −1-1. Upper semicontinuity of ϕ\phi, whose values do not include +∞+\infty, gives a finite upper bound MM on a neighborhood of each point. On such a neighborhood, hN≤max⁡{M,0}N.h_N\leq \frac{\max\{M,0\}}{N}. In particular the family is locally uniformly bounded above, although it need not be increasing. Apply the bounded upper-envelope theorem, Theorem “Local bounded upper-envelope theorem”, to each tail and put gj=(sup⁡N≥jhN)∗∈Cb(U).g_j=\left(\sup_{N\geq j}h_N\right)^*\in\mathcal C_b(U). The star is ordinary upper-semicontinuous regularization. The tail envelopes decrease, and on the same neighborhood the estimates give −1≤gj≤max⁡{M,0}j.-1\leq g_j\leq\frac{\max\{M,0\}}j. Consequently g=lim⁡jgjg=\lim_j g_j is in Cb(U)\mathcal C_b(U) by decreasing closure of the classical class, and −1≤g≤0-1\leq g\leq0 everywhere. At every point where ϕ\phi is finite, hNh_N tends to zero, so every tail supremum, and hence every gjg_j, is at least zero there. Thus g=0g=0 at that point. Since ϕ\phi is not identically −∞-\infty, gg attains its maximum zero somewhere on UU.

The set {g=0}\{g=0\} is closed by upper semicontinuity and open by the local maximum principle [GR25, Theorem 6.1, pp. 30–31], because all points of UU are interior points. It is nonempty, so connectedness implies g≡0g\equiv0. If PP contained a nonempty ordinary open OO, every hNh_N and every tail supremum would equal −1-1 on OO. Locality of ordinary upper regularization would give gj=−1g_j=-1 on OO for all jj, hence g=−1g=-1 there, a contradiction. ◻

Theorem (No divisorial poles). Under the assumptions of Proposition “A nontrivial pole set has empty ordinary interior”, the value ϕ(z)\phi(z) is finite at every divisorial point z∈Uz\in U.

Proof. Suppose instead that ϕ(z)=−∞\phi(z)=-\infty at such a point. Choose A⊂UA\subset U from Lemma “A domain with one prescribed Shilov point”. For every N≥1N\geq1, the finite classical truncation ϕN=max⁡{ϕ,−N}\phi_N=\max\{\phi,-N\} is bounded on the compact domain AA. It takes its maximum on AA at a Shilov point, by [GR25, Theorem 6.4, pp. 33–34]. Since the unique Shilov point is zz and ϕN(z)=−N\phi_N(z)=-N, it follows that ϕN≡−N\phi_N\equiv-N on AA. Letting N→∞N\to\infty gives ϕ≡−∞\phi\equiv-\infty on AA. Its nonempty ordinary interior is therefore contained in PP, contrary to Proposition “A nontrivial pole set has empty ordinary interior”. ◻

Both pole statements are DIRECT_COROLLARY of the bounded upper-envelope theorem already obtained from the accepted internal inputs, the external maximum principles, and the geometric lemma. They do not use any capacity-zero theorem for pole sets, a trace nonconcentration assertion, or a relative increasing-continuity statement. Thus they may be used as inputs to those subsequent questions without a capacity argument being fed back into itself.

The conclusion concerns divisorial points in the ambient dimension. For example, a proper analytic subvariety can still be a pole set and can carry an integration-current background. Divisorial points of that smaller subvariety are not ambient divisorial points. This theorem therefore does not assert nonconcentration on ambient polar sets for arbitrary positive closed backgrounds.

Ambient pole sets and all original trace measures

This section combines the absence of divisorial poles from Theorem “No divisorial poles” with the proved local increasing-capacity theorem. Density of divisorial points alone would not justify the conclusion. The passage from model atoms to a smooth trace uses the quasicontinuity of the actual pole indicator and a uniform bound on one fixed capacity domain.

Theorem (Vanishing of ambient smooth traces on a classical pole set). Let UU be a connected allowed ordinary open and let ϕ:U→[−∞,∞)\phi:U\to[-\infty,\infty) be classical psh and not identically −∞-\infty. Put P={ϕ=−∞}P=\{\phi=-\infty\}. For every original compact smooth (n,n)(n,n)-test η\eta, one has ∣μ[1],η∣(P)=0.|\mu_{[1],\eta}|(P)=0. Equivalently, 1P[1]=0\mathbf1_P[1]=0 as an original coefficient-Radon current. No local L1L^1 assertion about the original function ϕ\phi is assumed in this theorem.

Status: NEW_PROVED, relative to the stated internal inputs and the preceding R2 theorems.

Proof. Use a finite partition of the test and ordinary algebraic localization from B. Work on one moment presentation, with unit coordinates t1,…,trt_1,\ldots,t_r defined on a fixed larger compact norm-box neighborhood. Choose a smaller norm-box domain Ω⋐U\Omega\Subset U containing the localized test support, on which the coordinates are bounded. Set ρ=12∑a=1rta2,β=Dρ=∑a=1rd′ta∧d′′ta.\rho=\tfrac12\sum_{a=1}^r t_a^2, \qquad \beta=D\rho=\sum_{a=1}^r d't_a\wedge d''t_a. The function ρ\rho is smooth and classical psh. Indeed its pullback to every curve after every complete field extension is a convex function of the harmonic logarithms of analytic units; the curve chain rule gives nonnegative Laplacian. The finite matrix domination of R supplies a cutoff χ\chi and a finite constant CηC_\eta such that

∣μ[1],η∣≤Cη χβn∧[1].|\mu_{[1],\eta}|\leq C_\eta\,\chi\beta^n\wedge[1].

Any antisymmetric component is annihilated, as in R. It therefore suffices to show that the positive smooth trace μ=βn∧[1]\mu=\beta^n\wedge[1] vanishes on PP near the test support.

Apply the common local projective globalization from B, Lemma D.2, to ρ\rho, and then its projective pure-Fubini–Study regularization theorem. More explicitly, this is the constructive globalization part of the first-round Lemma “Regularization on a prescribed smaller norm box”, before its final smoothing step; it applies on the entire prescribed smaller norm box. In the fixed line frame, after removing the common additive constant, this gives ρj↓ρ\rho_j\downarrow\rho on a neighborhood of the entire prescribed compact box. The functions ρj\rho_j are restrictions of rational pure-Fubini–Study metric weights on one projective variety and one identified ample line. They are bounded uniformly on the fixed Ω\Omega: the first row and the continuous limit bound the whole decreasing row on its compact closure. Choose M>0M>0 with ∣ρj∣≤M|\rho_j|\leq M there for every jj.

The positive measures μj=T[1],n(ρj,…,ρj)\mu_j=T_{[1],n}(\rho_j,\ldots,\rho_j) are restrictions of the corresponding calibrated model measures. Each is a finite sum of atoms at divisorial points. Ordinary algebraic localization, including the possible finite constant field in B, Lemma D.5, preserves this assertion; the calibration may acquire its stated fixed degree factor, which does not affect the support or zero-mass conclusion. By Theorem “No divisorial poles”, none of these points belongs to PP. Consequently 1Pμj=0.\mathbf1_P\mu_j=0. The fixed-background decreasing theorem gives μj→μ\mu_j\to\mu on all original scalar tests, and hence weakly on compact continuous scalar tests.

These weak limits have a uniform bound on the same admissibility domain. In fact vj=(ρj+M)/(2M)v_j=(\rho_j+M)/(2M) belongs to Cb(Ω)\mathcal C_b(\Omega) and is between zero and one, so for every Borel E⊂ΩE\subset\Omega, μj(E)≤(2M)nCap⁡[1](E;Ω).\mu_j(E)\leq(2M)^n\operatorname{Cap}_{[1]}(E;\Omega). Subtract a constant from ϕ\phi so that it is nonpositive on a neighborhood of Ω‾\overline\Omega, and use wN=1+N−1max⁡{ϕ,−N}↑1Pc.w_N=1+N^{-1}\max\{\phi,-N\}\uparrow\mathbf1_{P^c}. The local increasing theorem gives quasicontinuity of this actual limit for the same capacity. Thus 1P\mathbf1_P is a bounded quasicontinuous multiplier. The R2 weak-closure multiplier theorem applies to μj→μ\mu_j\to\mu and yields 1Pμ=lim⁡j1Pμj=0\mathbf1_P\mu =\lim_j\mathbf1_P\mu_j=0 on the compact core. This passage, rather than density of the model points, is the essential step. Equation (eq:r2-ambient-trace-domination) now proves the assertion for the localized test. The finite partition proves it for every original compact smooth (n,n)(n,n)-form. ◻

Corollary (Ambient bounded BT products avoid classical pole sets). Under the same hypotheses, for every bounded classical tuple and every 0≤k≤n0\leq k\leq n, 1PT[1],k(u1,…,uk)=0.\mathbf1_P T_{[1],k}(u_1,\ldots,u_k)=0. Every coefficient measure of these currents has total variation zero on PP. In particular every bounded mixed Monge–Ampère measure gives PP zero mass.

Proof. The theorem gives 1P[1]=0\mathbf1_P[1]=0. Apply the exact restriction identity of Theorem “Positive closed restriction to a complete pole set”. Equality of currents and uniqueness of their coefficient-Radon measures give the coefficient assertion. ◻

Corollary (Local outer-capacity zero of ambient pluripolar sets). For every permitted norm-box domain Ω⋐U\Omega\Subset U and compact H⋐ΩH\Subset\Omega, Cap⁡[1]∗(H∩P;Ω)=0.\operatorname{Cap}_{[1]}^*(H\cap P;\Omega)=0. More generally, every locally pluripolar subset has local outer capacity zero, where locally pluripolar means locally contained in the pole set of a classical psh function not identically −∞-\infty on its connected witness domain. This statement does not itself assert the converse. The converse and qualified pole patching on exhaustion domains are proved later in Section “Polar witnesses and capacitability from ambient extremals”.

Proof. Every normalized bounded potential on Ω\Omega gives a measure that vanishes on PP, by the preceding corollary applied locally. Thus the Borel capacity of W∩PW\cap P is zero for an ordinary open core H⊂W⋐ΩH\subset W\Subset\Omega. The pole indicator is quasicontinuous on WW by the increasing argument used above. Hence W∩P=W∩{1Pc<1/2}W\cap P=W\cap\{\mathbf1_{P^c}<1/2\} is quasi-open. The proved outer regularity of quasi-open sets gives outer capacity zero, and monotonicity gives the assertion on HH. For a locally pluripolar set, cover its intersection with a prescribed compact core by smaller open witness cores whose larger witness domains lie in Ω\Omega. Second countability in the fixed finite pp-adic setting gives a countable subcover. Apply the local conclusion in each and use countable subadditivity together with the correct direction of capacity domain monotonicity. ◻

Corollary (Quasicontinuity of extended classical psh functions). Every extended-real classical psh function that is not identically −∞-\infty on any connected component is locally finite and continuous outside ordinary open sets of arbitrarily small ambient outer BT capacity. Its lower truncations converge locally in outer capacity. No L1L^1 bound on the unattenuated witness is asserted by this conclusion.

Proof. Work on a compact core, subtract a local upper bound, and thus write the function as a negative witness ϕ\phi. Choose the collar trace ν=β0∧[1]\nu=\beta_0\wedge[1] of the weighted theorem. The ambient trace theorem and finite smooth domination of this continuous-coefficient trace give ν(P)=0\nu(P)=0. Lemma “An integrable witness with the same pole set” supplies ψ=−ϑ(−ϕ),ϑ(t)=∑j≥1min⁡{t/Lj,1},∫(−ψ) dν<∞.\psi=-\vartheta(-\phi),\qquad \vartheta(t)=\sum_{j\geq1}\min\{t/L_j,1\}, \qquad \int(-\psi)\,d\nu<\infty. The function ϑ\vartheta is increasing and tends to infinity: for any integer mm, all of its first mm summands equal one once t≥max⁡j≤mLjt\geq\max_{j\leq m}L_j. For every N>0N>0, {ϕ<−N}⊂{ψ≤−ϑ(N)}.\{\phi<-N\}\subset\{\psi\leq-\vartheta(N)\}. The weighted theorem applied to ψ\psi therefore bounds the capacity of the ordinary open sublevel set on the core by C/ϑ(N)C/\vartheta(N), which tends to zero. The original truncation max⁡{ϕ,−N}\max\{\phi,-N\} agrees with ϕ\phi outside that open set. This proves local outer-capacity convergence. Finally combine one such truncation’s bounded quasicontinuity with the small ordinary open sublevel set, exactly as in Corollary “Outer capacity and quasicontinuity of an integrable witness”. The original function is finite and continuous on the complement. ◻

Attenuated witnesses and an obstruction to exponential tails

The notation and original test category are those of Section “A relative weighted L1L^1 Chern–Levine–Nirenberg estimate”. The first observation weakens the hypothesis needed for a pole-set conclusion; it does not assert integrability of the original witness under that weaker hypothesis.

Lemma (An integrable witness with the same pole set). Let ϕ:O→[−∞,0]\phi:O\to[-\infty,0] be classical psh, and let ν\nu be a finite positive Radon measure on OO. If ν({ϕ=−∞})=0,\nu(\{\phi=-\infty\})=0, there is a negative classical psh function ψ\psi on the same ordinary open set such that {ψ=−∞}={ϕ=−∞},∫(−ψ) dν<∞.\{\psi=-\infty\}=\{\phi=-\infty\},\qquad \int(-\psi)\,d\nu<\infty. More precisely, one can choose Lj≥2jL_j\geq2^j, tending increasingly to infinity, and set

ψ=∑j≥1max⁡{ϕ,−Lj}Lj,∫(−ψ) dν≤1.\psi=\sum_{j\geq1}\frac{\max\{\phi,-L_j\}}{L_j}, \qquad \int(-\psi)\,d\nu\leq1.

Status: DIRECT_COROLLARY. This is an elementary measure-theoretic attenuation combined with decreasing closure of extended-real classical psh functions (Gubler–Rabinoff, Proposition 4.2(3)–(4)).

Proof. Put t=−ϕt=-\phi. The function tt is finite ν\nu-almost everywhere. For L→∞L\to\infty, the functions min⁡(t,L)/L\min(t,L)/L are bounded by one and converge to zero ν\nu-almost everywhere. Since ν\nu is finite, bounded convergence permits the choice of increasing Lj≥2jL_j\geq2^j such that ∫min⁡(t,Lj)Lj dν≤2−j.\int\frac{\min(t,L_j)}{L_j}\,d\nu\leq2^{-j}. Every finite partial sum in (eq:r2-attenuated-witness) is a negative bounded classical psh function. The partial sums decrease, so their extended-real limit is classical psh. At a point where t<∞t<\infty, the absolute sum is bounded by t∑jLj−1<∞.t\sum_j L_j^{-1}<\infty. At a point where t=∞t=\infty, each summand is −1-1, so the limit is −∞-\infty. This proves equality of the actual pole sets, including the complete-extension classicality supplied by decreasing closure. Finally monotone convergence gives ∫(−ψ) dν=∑j∫min⁡(t,Lj)Lj dν≤1.\int(-\psi)\,d\nu =\sum_j\int\frac{\min(t,L_j)}{L_j}\,d\nu\leq1. ◻

Corollary (A pole-set criterion using one background trace). In the geometry of Theorem “Uniform truncation estimate over an arbitrary background”, let ν=β0∧S\nu=\beta_0\wedge S and let WW be ordinary open with W‾⋐V\overline W\Subset V. If (β0∧S)({ϕ=−∞})=0,(\beta_0\wedge S)(\{\phi=-\infty\})=0, then W∩{ϕ=−∞}W\cap\{\phi=-\infty\} has zero outer capacity for the fixed admissibility domain OO. Every bounded mixed top-degree measure over SS gives this set zero mass. No L1L^1 assumption on the original ϕ\phi is necessary for these conclusions.

Status: DIRECT_COROLLARY.

Proof. The compactly supported positive trace ν\nu is finite. Apply Lemma “An integrable witness with the same pole set”, then apply Theorem “Uniform truncation estimate over an arbitrary background” and Corollary “Outer capacity and quasicontinuity of an integrable witness” to ψ\psi. The pole sets coincide. All domains remain the originally chosen ones. For a mixed tuple that is only locally bounded, take its finite bounds on a neighborhood of the compact core, as in the weighted theorem. ◻

Corollary (Preservation of non-concentration on a specified pole set). Let P={ϕ=−∞}P=\{\phi=-\infty\} for a negative classical psh function ϕ\phi on UU. Suppose that every coefficient measure of SS has total variation zero on PP. Then every coefficient measure of TS,k(u1,…,uk)T_{S,k}(u_1,\ldots,u_k) has total variation zero on PP, for ui∈Cb(U)u_i\in\mathcal C_b(U) and 0≤k≤q0\leq k\leq q. In particular, a background whose coefficient measures avoid all such pole sets retains this property after every bounded BT product.

Status: DIRECT_COROLLARY.

Proof. Work locally on nested norm boxes with a collar as in the weighted theorem. The hypothesis and finite positive controls imply (β0∧S)(P)=0(\beta_0\wedge S)(P)=0, including for the continuous-coefficient cutoff trace. The previous corollary makes every bounded top-degree mixed measure over SS vanish on PP in a smaller core.

For a coefficient of a lower-degree product, localize its original test to a finite moment presentation, and let β=Dρ0\beta=D\rho_0 be the quadratic trace form of that presentation, where ρ0\rho_0 is smooth psh and bounded on the larger box. The finite-test domination from R bounds the variation of that coefficient on the core by a constant multiple of TS,k(u1,…,uk)∧βq−k=TS,q(u1,…,uk,ρ0,…,ρ0⏟q−k).T_{S,k}(u_1,\ldots,u_k)\wedge\beta^{q-k} =T_{S,q}(u_1,\ldots,u_k, \underbrace{\rho_0,\ldots,\rho_0}_{q-k}). The equality is compatibility with smooth slots, already proved in R1. This mixed measure vanishes on PP by the preceding paragraph. Finite local presentations and ordinary restriction give the assertion for every original coefficient. The case k=0k=0 is the hypothesis itself. ◻

The ambient specialization.

For S=[1]S=[1], the criterion reduces the general polar-set question to zero mass of pole sets for the specified original smooth volume traces. That input is proved separately in Theorem “Vanishing of ambient smooth traces on a classical pole set”, using divisorial non-poles, bounded model approximations and the quasicontinuous multiplier theorem. Neither the attenuation argument nor that trace theorem asserts local L1L^1 integrability of the original unattenuated witness. Complex distributional ellipticity is not used. The criterion also makes clear why an arbitrary background concentrated on the pole set is excluded.

Proposition (No uniform exponential improvement over arbitrary backgrounds). Over the original finite extension K/QpK/\mathbf Q_p, there are a two-dimensional allowed ordinary open set UU, a bounded continuous classical psh function ff, a background S=D(f[1])S=D(f[1]), a negative classical psh witness ϕ\phi, and a compact K0⋐UK_0\Subset U with the following properties. The background trace integrability condition of Theorem “Uniform truncation estimate over an arbitrary background” holds for every fixed collar used below, but there is a normalized continuous classical psh vv such that ∫K0exp⁡(−aϕ) TS,1(v)=∞for every a>0.\int_{K_0}\exp(-a\phi)\,T_{S,1}(v)=\infty \quad\text{for every }a>0. Moreover the relative capacities of the deep sublevels have a polynomial lower bound. Consequently the weighted L1L^1 theorem cannot be strengthened, under its stated hypotheses, to an exponential tail bound uniform over normalized potentials.

Status of the exponential assertion refuted: FALSE.

Proof. Choose a uniformizer π\pi and put λ=−log⁡∣π∣>0\lambda=-\log|\pi|>0. Choose 0<r<1<R0<r<1<R and set U={∣z0∣<1}×{r<∣z1∣<R}.U=\{|z_0|<1\}\times\{r<|z_1|<R\}. For j≥1j\geq1, put fj=max⁡{log⁡∣z0∣,−jλ},f=∑j≥1j−4fj,ϕ=log⁡∣z0∣,v=max⁡{0,log⁡∣z1∣}log⁡R.f_j=\max\{\log|z_0|,-j\lambda\},\qquad f=\sum_{j\geq1}j^{-4}f_j,\qquad \phi=\log|z_0|,\qquad v=\frac{\max\{0,\log|z_1|\}}{\log R}. Each fjf_j is continuous classical psh and takes values in [−jλ,0][-j\lambda,0]. The series for ff converges uniformly on UU because ∑j−3<∞\sum j^{-3}<\infty. Its partial sums decrease, so ff is bounded continuous classical psh. The current S=D(f[1])S=D(f[1]) is symmetric, original-positive and separately closed. Writing Sj=D(fj[1])S_j=D(f_j[1]), linearity and convergence give S=∑j≥1j−4Sj.S=\sum_{j\geq1}j^{-4}S_j. The positive series converges locally as a coefficient-Radon current. Indeed, relative CLN bounds its mass against each fixed positive cutoff by a constant times ∑j−4(1+jλ)<∞\sum j^{-4}(1+j\lambda)<\infty.

The support of SjS_j lies in {∣z0∣=∣π∣j}\{|z_0|=|\pi|^j\}: on one side fjf_j is constant and on the other it is the logarithm of the norm of an analytic unit. Thus −ϕ=jλ-\phi=j\lambda on this support. For a fixed positive collar trace β0\beta_0, finite-test domination and CLN give a constant CρC_\rho, independent of jj, such that ∫β0∧Sj≤Cρ(1+jλ).\int\beta_0\wedge S_j\leq C_\rho(1+j\lambda). The same is valid for its continuous coefficients by positive smooth controls. Equality of the positive coefficient measures of the series and monotone convergence imply AS,ρ(ϕ)=∑j≥1j−4 jλ∫β0∧Sj≤Cρ∑j≥1j−4jλ(1+jλ)<∞.A_{S,\rho}(\phi) =\sum_{j\geq1}j^{-4}\,j\lambda \int\beta_0\wedge S_j \leq C_\rho\sum_{j\geq1}j^{-4}j\lambda(1+j\lambda)<\infty.

Let xjx_j be the product Gauss point with radii (∣π∣j,1)(|\pi|^j,1). The usual model metric of O(1)\mathcal O(1) on the first projective line, after the coordinate rescaling z0↦z0/πjz_0\mapsto z_0/\pi^j, has affine weight fj+jλf_j+j\lambda. Together with the standard model metric in z1z_1, the mixed model intersection has degree one. The original calibrated model-measure identity therefore gives TSj,1(v)=(log⁡R)−1δxj.T_{S_j,1}(v)=(\log R)^{-1}\delta_{x_j}. Fixed continuous multiplication by vv and distributional DD pass the positive partial sums to the limit, yielding

TS,1(v)=(log⁡R)−1∑j≥1j−4δxj.T_{S,1}(v)=(\log R)^{-1} \sum_{j\geq1}j^{-4}\delta_{x_j}.

Choose r<r0<1<R0<Rr<r_0<1<R_0<R and take K0={∣z0∣≤∣π∣}×{r0≤∣z1∣≤R0}.K_0=\{|z_0|\leq|\pi|\} \times\{r_0\leq|z_1|\leq R_0\}. This is compact in UU, contains every xjx_j, and lies in local norm boxes admitting the required collar. Since ϕ(xj)=−jλ\phi(x_j)=-j\lambda, equation (eq:r2-polynomial-tail-measure) gives ∫K0(−ϕ) TS,1(v)=λlog⁡R∑j≥1j−3<∞,∫K0e−aϕ TS,1(v)=1log⁡R∑j≥1j−4eajλ=∞.\int_{K_0}(-\phi)\,T_{S,1}(v) =\frac{\lambda}{\log R}\sum_{j\geq1}j^{-3}<\infty, \qquad \int_{K_0}e^{-a\phi}\,T_{S,1}(v) =\frac1{\log R}\sum_{j\geq1}j^{-4}e^{aj\lambda}=\infty. The potential vv is between zero and one on all of UU. Hence, for every N>0N>0, Cap⁡S(K0∩{ϕ<−N};U)≥1log⁡R∑j>N/λj−4≥13log⁡R(1+N/λ)−3.\operatorname{Cap}_S(K_0\cap\{\phi<-N\};U) \geq\frac1{\log R}\sum_{j>N/\lambda}j^{-4} \geq\frac1{3\log R}(1+N/\lambda)^{-3}. This cannot be bounded above by Ae−aNA e^{-aN} with fixed finite AA and a>0a>0. All powers of π\pi, all Gauss points, and all model metrics used here are defined over the original field with its prescribed absolute value. ◻

Proposition (A uniform first moment does not give uniform integrability). Even for S=[1]S=[1] on a smooth analytic curve, a fixed negative psh witness satisfying the collar integrability condition need not commute with weak limits of normalized bounded BT measures. There can be positive measures μj→0\mu_j\to0 with ∫K0(−ϕ) dμj=1for all j,(−ϕ)μj⟶δ0≠0\int_{K_0}(-\phi)\,d\mu_j=1\quad\text{for all }j, \qquad (-\phi)\mu_j\longrightarrow\delta_0\ne0 on one fixed compact K0K_0 containing their supports.

Status of the asserted unbounded-weight stability: FALSE.

Proof. On U={∣z∣<1}⊂AK1,anU=\{|z|<1\}\subset\mathbf A^{1,\mathrm{an}}_K, set ϕ=log⁡∣z∣\phi=\log|z|, with value −∞-\infty at 00, and put vj=1+max⁡{log⁡∣z∣,−jλ}jλ,λ=−log⁡∣π∣.v_j=1+\frac{\max\{\log|z|,-j\lambda\}}{j\lambda}, \qquad \lambda=-\log|\pi|. These are continuous classical psh functions with 0≤vj≤10\leq v_j\leq1. Let ξj\xi_j be the Gauss point of radius ∣π∣j|\pi|^j. The calibrated one-variable model calculation gives μj=D(vj[1])=(jλ)−1δξj,−ϕ(ξj)=jλ.\mu_j=D(v_j[1])=(j\lambda)^{-1}\delta_{\xi_j}, \qquad -\phi(\xi_j)=j\lambda. All points ξj\xi_j lie in the fixed compact disc K0={∣z∣≤∣π∣}⋐UK_0=\{|z|\leq|\pi|\}\Subset U and tend to 00. Thus μj→0\mu_j\to0 even in total variation, whereas (−ϕ)μj=δξj→δ0(-\phi)\mu_j=\delta_{\xi_j}\to\delta_0 as Radon measures, and hence on every original compact smooth scalar test. The witness is finite μj\mu_j-almost everywhere. Its first moment is exactly one, but for every M>0M>0, sup⁡j∫K0∩{−ϕ>M}(−ϕ) dμj=1.\sup_j\int_{K_0\cap\{-\phi>M\}}(-\phi)\,d\mu_j=1. This is the precise failure of uniform integrability.

The background trace hypothesis can be checked directly. Choose a radial smooth psh collar ρ\rho that is constant on a disc containing K0K_0, with {ρ<0}⋐U\{\rho<0\}\Subset U still containing that disc. Such a collar is a convex smooth regularized maximum of zero and an affine function of log⁡∣z∣\log|z|, minus a constant. For q=1q=1, its trace β0=(−ρ)+4Dρ+4(−ρ)+3d′ρ∧d′′ρ\beta_0=(-\rho)_+^4D\rho+ 4(-\rho)_+^3d'\rho\wedge d''\rho is supported in a compact annulus away from 00. The function ϕ\phi is bounded there, so ∫(−ϕ)β0∧[1]<∞\int(-\phi)\beta_0\wedge[1]<\infty.

The sequence vjv_j increases. Its actual pointwise supremum is one off 00 and zero at 00, so it is not upper semicontinuous. It belongs to the bounded capacitary closure by Corollary “Local increasing convergence without a prescribed collar”. No claim is made that this supremum is a classical psh limit. The bounded-weight calculus is consistent with μj→0\mu_j\to0; the failure concerns only the attempted extension to the fixed unbounded multiplier. ◻

Quasicontinuous obstacles and local maximality

This section concerns the ambient background [1][1]. It uses the continuous-obstacle envelope and orthogonality theorem of B, Theorem C.6, without changing its hypotheses. A separate limit argument extends orthogonality to a class of bounded discontinuous obstacles. It then supplies the local equivalence between maximality and vanishing Monge–Ampère measure. No relative obstacle orthogonality for an arbitrary background is asserted.

Lemma (Continuous majorants of a quasicontinuous function). Let fj↓ff_j\downarrow f be continuous real functions on an ordinary neighborhood of a compact HH, with ff bounded and CΩC_\Omega-quasicontinuous on a compact neighborhood of HH. Then CΩ∗(H∩{fj−f>δ})⟶0(δ>0).C_\Omega^*(H\cap\{f_j-f>\delta\})\longrightarrow0 \qquad(\delta>0). Here Ω\Omega is the fixed local capacity domain containing that neighborhood.

Proof. For ϵ>0\epsilon>0, choose an ordinary open GG of capacity less than ϵ\epsilon such that ff is continuous on the chosen compact neighborhood minus GG. On this compact set, the continuous functions fjf_j decrease to a continuous function. Dini’s theorem gives uniform convergence there. Thus the indicated level set is contained in GG for every sufficiently large jj, and its outer capacity is at most CΩ(G)<ϵC_\Omega(G)<\epsilon. ◻

Theorem (Orthogonality for bounded quasicontinuous usc obstacles). Let Y/FY/F be smooth projective geometrically integral over a finite extension F/QpF/\mathbf Q_p, with its prescribed absolute value, and let LL be ample. Let ff be a bounded upper-semicontinuous metric on LL, relative to a continuous reference metric. Assume that its local weights are locally quasicontinuous for the ambient capacities Cap⁡[1]\operatorname{Cap}_{[1]} on the permitted local norm boxes. Put p=PLbf,p=P_L^{\mathrm b} f, the greatest bounded classical minorant from Proposition “Envelopes of bounded upper-semicontinuous obstacles”. Then

∫Yan(f−p)c1(L,p)n=0.\int_{Y^{\mathrm{an}}}(f-p)c_1(L,p)^n=0.

In particular this applies to f=r+gf=r+g, where rr is a bounded classical metric and gg is any continuous real function on YanY^{\mathrm{an}}.

Status: NEW_PROVED. The new step is the extension to these discontinuous obstacles; the continuous-obstacle theorem is an INTERNAL_INPUT.

Proof. Choose uniformly bounded continuous metrics fj↓ff_j\downarrow f. Let pj=PLfjp_j=P_L f_j. The accepted envelope theorem makes each pjp_j continuous model-semipositive and gives ∫(fj−pj)c1(L,pj)n=0.\int(f_j-p_j)c_1(L,p_j)^n=0. Proposition “Envelopes of bounded upper-semicontinuous obstacles” gives pj↓pp_j\downarrow p, with a common bound relative to a continuous semipositive reference metric. Write μj=c1(L,pj)n\mu_j=c_1(L,p_j)^n and μ=c1(L,p)n\mu=c_1(L,p)^n. The fixed-background monotone theorem, in local frames, gives μj→μ\mu_j\to\mu on all original scalar tests. The measures have the same finite total mass deg⁡F(Ln)\deg_F(L^n), by continuous compatibility and passage to the decreasing limit.

Fix a local compact test stage. In its frame the weights of all pjp_j have uniform bounds on a slightly larger box. The R1 coefficient-capacity estimate therefore controls μj\mu_j uniformly by that fixed box’s ambient capacity. By Lemma “Continuous majorants of a quasicontinuous function”, the local weights of fjf_j converge to those of ff in outer capacity. These weights are quasicontinuous, as is their limit. The multiplier theorem “Quasicontinuous multipliers through weak closure” yields convergence of the integrals with multiplier fjf_j. The weighted decreasing theorem yields convergence with multiplier pjp_j. Subtract the two convergences. A finite original smooth partition of unity on the compact projective space makes the differences of local weights equal to the global scalar functions fj−pjf_j-p_j and f−pf-p. Summing gives (eq:r2-qc-orthogonality).

Finally, bounded classical local weights are quasicontinuous by R1, and addition of a continuous function preserves quasicontinuity and upper semicontinuity. Thus r+gr+g satisfies all the stated assumptions. No monotone-convergence theorem for arbitrary discontinuous weights was used: their convergence was proved using Dini’s theorem off ordinary small-capacity opens. ◻

Definition (Local maximality). A function u∈Cb(U)u\in\mathcal C_b(U) is locally maximal if every point has an ordinary neighborhood VV with compact closure in a smooth affine algebraic chart such that the following holds. Whenever v∈Cb(V)v\in\mathcal C_b(V) and v=uv=u on V∖KV\setminus K for some compact K⋐VK\Subset V, one has v≤uv\leq u on VV.

Equivalently one may use competitors whose strict positive gap over uu has compact closure in VV: replace such a competitor by its maximum with uu. This definition is local and does not use the whole of a proper analytic space as a boundaryless comparison domain.

Theorem (Maximality and the homogeneous equation). Let u∈Cb(U)u\in\mathcal C_b(U) on any allowed good, Hausdorff, boundaryless, quasi-smooth pure-dimensional space over the fixed KK. Then u is locally maximal⟺(Du)n=0 on U.u\text{ is locally maximal} \quad\Longleftrightarrow\quad (Du)^n=0\text{ on }U. The equality is that of the ordinary original-CLD Radon measure. No continuity assumption on uu is imposed.

Status: NEW_PROVED, relative to the adopted inputs and the preceding orthogonality extension.

Proof. Assume first that (Du)n=0(Du)^n=0. On a relatively compact framed affine neighborhood VV, a competitor as in Definition “Local maximality” satisfies every compact-gap condition of Corollary “Pointwise ambient domination and uniqueness”. Its strict gap has zero (Du)n(Du)^n-mass. That corollary gives u≥vu\geq v pointwise. Such neighborhoods exist by B, so uu is locally maximal.

Conversely, fix a point where maximality holds on VV. The common-collar construction of B, Lemma D.2, realizes u+cu+c, on a smaller ordinary neighborhood W⋐VW\Subset V, as the weight of a bounded classical metric rr on an ample line LL over a smooth projective variety. Its frame is fixed on WW. Choose any nonnegative original smooth scalar χ\chi compactly supported in WW; it extends by zero to the projective space. For ϵ>0\epsilon>0, let f=r+ϵχ,p=PLbf.f=r+\epsilon\chi,\qquad p=P_L^{\mathrm b}f. The metric rr is a permitted minorant, so r≤p≤r+ϵχr\leq p\leq r+\epsilon\chi. In particular p=rp=r off the compact support of χ\chi. Its weight on WW, minus cc, glues with uu outside that support to give a member of Cb(V)\mathcal C_b(V) which equals uu off a compact subset of VV. Maximality forces this function to be at most uu. Therefore p=rp=r everywhere, also outside WW by the obstacle inequalities.

The obstacle f=r+ϵχf=r+\epsilon\chi satisfies Theorem “Orthogonality for bounded quasicontinuous usc obstacles”. Its conclusion is 0=∫(f−p)c1(L,p)n=ϵ∫χ c1(L,r)n=ϵ∫Wχ (Du)n.0=\int(f-p)c_1(L,p)^n =\epsilon\int\chi\,c_1(L,r)^n =\epsilon\int_W\chi\,(Du)^n. This holds for every such χ\chi, so positivity and Radon uniqueness give (Du)n=0(Du)^n=0 on WW. Cover UU by these neighborhoods. If the local algebraic chart uses the finite constant-field convention in B, Lemma D.5, its original calibration changes by a positive constant; the vanishing assertion is unchanged. Thus it holds for the prescribed KK and its original absolute value. ◻

Remark (Scope for Dirichlet theory). The theorem identifies bounded solutions of the homogeneous equation with a local maximality property. It does not prove that a Perron envelope attains arbitrary prescribed boundary values, that every ordinary domain admits boundary barriers, or that a nonzero prescribed measure admits a solution. Those are separate Dirichlet existence and boundary-regularity questions. For an arbitrary background SS, even pointwise domination can fail, as the connected nonzero-background counterexample in this record shows; the ambient theorem above cannot be transferred to that setting without further hypotheses.

A homogeneous Dirichlet theorem with explicit peak barriers

The results in this section concern the ambient background [1][1]. The geometric hypothesis is stated separately: no assertion is made that an arbitrary Berkovich analytic domain has these barriers. The solution will be bounded and classical psh, with the stated ordinary boundary limits. Interior continuity is not part of the conclusion.

Let VV be a connected ordinary open set whose closure V‾\overline V is compact and is contained in a smooth affine algebraic chart. Write ∂V=V‾∖V\partial V=\overline V\setminus V for its ordinary topological boundary, and assume that this set is nonempty. All closures and limits in this section refer to the ordinary Berkovich topology.

Definition (Continuous psh peak barriers). The domain VV has continuous psh peak barriers if for each ξ∈∂V\xi\in\partial V there is a function βξ∈Cb(V)∩C(V‾)\beta_\xi\in \mathcal C_b(V)\cap C(\overline V) such that βξ(ξ)=0\beta_\xi(\xi)=0 and βξ(x)<0\beta_\xi(x)<0 for every x∈V‾∖{ξ}x\in\overline V\setminus\{\xi\}.

Lemma (Boundary maximum principle). Suppose that h∈Cb(V)h\in\mathcal C_b(V) is bounded above and lim sup⁡V∋x→ξh(x)≤0(ξ∈∂V).\limsup_{V\ni x\to\xi}h(x)\leq0 \qquad(\xi\in\partial V). Then h≤0h\leq0 on VV.

Proof. If h(x0)>0h(x_0)>0, choose 0<a<h(x0)0<a<h(x_0). For each boundary point the limsup condition gives a neighborhood on which h<ah<a in VV. The union of these neighborhoods contains the compact boundary. Thus {h≥a}\{h\geq a\} is contained in a compact subset of VV; it is closed there by upper semicontinuity. The bounded usc function hh attains its global positive maximum at a point of VV. The strong maximum principle for classical psh functions, [GR25, Theorem 6.1], makes it constant on the connected domain VV. A positive constant contradicts the boundary condition at the nonempty boundary. ◻

Theorem (Homogeneous Dirichlet problem under peak barriers). Assume that VV has the barriers in Definition “Continuous psh peak barriers”. For every g∈C(∂V)g\in C(\partial V) there is a unique bounded classical psh function HgH_g on VV such that (DHg)n=0on V,lim⁡V∋x→ξHg(x)=g(ξ)(ξ∈∂V).(DH_g)^n=0\quad\hbox{on }V,\qquad \lim_{V\ni x\to\xi}H_g(x)=g(\xi) \quad(\xi\in\partial V). The measure equation is tested against all original compactly supported smooth CLD scalar tests. Moreover min⁡∂Vg≤Hg≤max⁡∂Vg,∥Hg1−Hg2∥L∞(V)≤∥g1−g2∥L∞(∂V).\min_{\partial V}g\leq H_g\leq\max_{\partial V}g, \qquad \|H_{g_1}-H_{g_2}\|_{L^\infty(V)} \leq\|g_1-g_2\|_{L^\infty(\partial V)} . In particular the solution is monotone with respect to its boundary values.

Status: NEW_PROVED under the displayed barrier hypothesis. The corresponding assertion for general ordinary analytic domains remains OPEN; no existence of barriers in general dimension is supplied here.

Proof. Put m=min⁡∂Vgm=\min_{\partial V}g and M=max⁡∂VgM=\max_{\partial V}g, and let Fg\mathcal F_g consist of all v∈Cb(V)v\in\mathcal C_b(V) satisfying m≤v≤M,lim sup⁡V∋x→ηv(x)≤g(η)(η∈∂V).m\leq v\leq M,\qquad \limsup_{V\ni x\to\eta}v(x)\leq g(\eta) \quad(\eta\in\partial V). This family contains the constant mm. Let H=sup⁡v∈FgvH=\sup_{v\in\mathcal F_g}v, pointwise, and let H∗H^* denote its ordinary upper-semicontinuous regularization on VV. Theorem “Local bounded upper-envelope theorem” gives H∗∈Cb(V)H^*\in\mathcal C_b(V) and m≤H∗≤Mm\leq H^*\leq M.

Fix ξ∈∂V\xi\in\partial V and ϵ>0\epsilon>0. Continuity of gg gives a boundary neighborhood of ξ\xi on which ∣g−g(ξ)∣<ϵ|g-g(\xi)|<\epsilon. On the compact complement, βξ\beta_\xi has a strictly negative maximum. Consequently one can choose a finite A>0A>0 such that, on all of ∂V\partial V, b−=g(ξ)−ϵ+Aβξ≤g,b+=g(ξ)+ϵ−Aβξ≥g.b_-=g(\xi)-\epsilon+A\beta_\xi\leq g,\qquad b_+=g(\xi)+\epsilon-A\beta_\xi\geq g . The function b−b_- is classical psh on VV, continuous on the closure, and at most MM. Thus max⁡(m,b−)∈Fg\max(m,b_-)\in\mathcal F_g, and H∗≥b−H^*\geq b_-. For every v∈Fgv\in\mathcal F_g, the function v−b+=v+Aβξ−g(ξ)−ϵv-b_+=v+A\beta_\xi-g(\xi)-\epsilon is bounded classical psh and has boundary limsup at most zero. Lemma “Boundary maximum principle” implies v≤b+v\leq b_+. Since b+b_+ is continuous, also H∗≤b+H^*\leq b_+. Taking limits at ξ\xi in g(ξ)−ϵ+Aβξ≤H∗≤g(ξ)+ϵ−Aβξg(\xi)-\epsilon+A\beta_\xi \leq H^* \leq g(\xi)+\epsilon-A\beta_\xi and then letting ϵ\epsilon decrease to zero proves H∗(x)→g(ξ)H^*(x)\to g(\xi) as x→ξx\to\xi from VV. Hence H∗∈FgH^*\in\mathcal F_g, so H=H∗H=H^* pointwise.

It remains to prove the equation. Let W⋐VW\Subset V be an ordinary neighborhood, and let v∈Cb(W)v\in\mathcal C_b(W) agree with HH outside a compact set K⋐WK\Subset W. The function equal to max⁡(v,H)\max(v,H) on WW and to HH off KK is well defined and classical psh by ordinary gluing and maximum stability. Call it ww. It is globally bounded, since its modification is confined to KK, and it has boundary limits gg. Lemma “Boundary maximum principle”, applied to w−Mw-M, gives w≤Mw\leq M; also w≥H≥mw\geq H\geq m. Therefore w∈Fgw\in\mathcal F_g and w≤Hw\leq H by the definition of HH. It follows that v≤Hv\leq H. This proves local maximality of HH, and Theorem “Maximality and the homogeneous equation” yields (DH)n=0(DH)^n=0.

If H1,H2H_1,H_2 are two bounded classical solutions with the same boundary values, then for each ϵ>0\epsilon>0 the set {H1<H2−ϵ}\{H_1<H_2-\epsilon\} has compact closure in VV. Indeed, the ordinary boundary limits give a neighborhood of each boundary point on which this set is empty, and compactness of the boundary gives a common compact core. The analogous assertion holds after interchanging the functions. Since their Monge–Ampère measures vanish, pointwise ambient domination (Corollary “Pointwise ambient domination and uniqueness”) gives both H1≥H2H_1\geq H_2 and H2≥H1H_2\geq H_1.

More generally set δ=∥g1−g2∥L∞(∂V)\delta=\|g_1-g_2\|_{L^\infty(\partial V)}. The pair Hg2+δ,Hg1H_{g_2}+\delta,H_{g_1} satisfies the required one-sided compact-gap condition and has zero measures, so domination gives Hg1≤Hg2+δH_{g_1}\leq H_{g_2}+\delta. Reversing the indices proves the norm estimate. If g1≤g2g_1\leq g_2, apply this comparison directly with no shift to obtain Hg1≤Hg2H_{g_1}\leq H_{g_2}. ◻

Remark (What the theorem does not settle). A classification or construction of suitable domains in arbitrary dimension is a separate geometric problem. The proof uses continuous peak barriers to control the ordinary boundary and the discontinuous-obstacle theorem to identify maximality with the homogeneous equation. It supplies neither interior continuity for HgH_g nor existence for an arbitrary prescribed nonzero measure. For an arbitrary positive closed background SS, the pointwise domination step would require an additional gap-detection hypothesis.

A bounded family which is not precompact in capacity

This example uses the fixed ambient background S=[1]S=[1]. All potentials, including their pointwise limit, are continuous classical psh. It rules out replacing the capacity-convergence hypothesis by bounded pointwise convergence, and rules out a capacity-precompactness argument for the whole normalized test family. It is not a counterexample to weak-closure domination of that family’s measures.

A moving residue branch over the original field

Let π\pi be a uniformizer of the fixed finite extension K/QpK/\mathbf Q_p, and retain its original absolute value. Put λ=−log⁡∣π∣>0,V={x∈AK1,an:∣T(x)∣<∣π∣−1}.\lambda=-\log|\pi|>0,\qquad V=\{x\in\mathbf A_K^{1,\mathrm{an}}:|T(x)|<|\pi|^{-1}\}. Choose distinct monic irreducible polynomials P‾i∈K~[T]\overline P_i\in\widetilde K[T], of degrees di≥2d_i\geq2, and monic lifts Pi∈K∘[T]P_i\in K^\circ[T]. Define hi=1dimax⁡{log⁡∣Pi(T)∣,−diλ},ϕi=λ−1hi,ψi=1+ϕi2.h_i=\frac1{d_i}\max\{\log|P_i(T)|,-d_i\lambda\}, \qquad \phi_i=\lambda^{-1}h_i,\qquad \psi_i=\frac{1+\phi_i}{2}. These are continuous classical psh functions. In fact hih_i is the local weight of the pure Fubini–Study metric on OP1(1)\mathcal O_{\mathbf P^1}(1) specified, after tensoring by did_i, by the two sections Pihom(X,Z),πdiZdi.P_i^{\mathrm{hom}}(X,Z),\qquad \pi^{d_i}Z^{d_i}. They have no common zero. The factors di−1d_i^{-1} and λ−1\lambda^{-1} above are factors in the potential, not a renormalization of the field or of the integration current.

Let ζ\zeta be the unit Gauss point. Let ξi\xi_i be the unique Shilov point of the closed residue disc Bi={∣Pi(T)∣≤∣π∣di}.B_i=\{|P_i(T)|\leq|\pi|^{d_i}\}. More explicitly, PiP_i has did_i simple Hensel roots in the unramified extension of degree did_i. Over that extension, BiB_i is the disjoint union of the closed discs of radius ∣π∣di|\pi|^{d_i} about those roots. Their conjugate Gauss points restrict to the same original KK-point ξi\xi_i. The residue field of its branch has degree did_i.

Lemma (The branch weights and their measures). On VV one has 0≤ψi<10\leq\psi_i<1 and Dhi=δξi,Dψi=12λδξi.D h_i=\delta_{\xi_i},\qquad D\psi_i=\frac1{2\lambda}\delta_{\xi_i}. Moreover ξi→ζ\xi_i\to\zeta, and pointwise on VV ψi⟶ψ=12+12λmax⁡{0,log⁡∣T∣}.\psi_i\longrightarrow \psi=\frac12+\frac1{2\lambda}\max\{0,\log|T|\}. The limit is continuous classical psh, and Dψ=12λδζ.D\psi=\frac1{2\lambda}\delta_\zeta. At the indicated points, ψi(ξi)=0,ψj(ξi)=12 (j≠i),ψ(ξi)=ψ(ζ)=12.\psi_i(\xi_i)=0,\qquad \psi_j(\xi_i)=\frac12\ (j\ne i),\qquad \psi(\xi_i)=\psi(\zeta)=\frac12.

Proof. Irreducibility of the separable reduction gives the unramified root extension and the described discs by Hensel’s lemma. The closed KK-domain BiB_i is the closed disc over this unramified residue extension, viewed as a KK-analytic domain; it has one Shilov point. On its interior hi=−λh_i=-\lambda. Off BiB_i, the function is di−1log⁡∣Pi∣d_i^{-1}\log|P_i|, which is harmonic there because PiP_i has no zero there. Thus the positive curvature measure of the displayed model metric is supported at ξi\xi_i. Its total mass is deg⁡KOP1(1)=1\deg_K\mathcal O_{\mathbf P^1}(1)=1 in the original model formula, so it is δξi\delta_{\xi_i}. Equivalently, the radial slope di−1d_i^{-1} at that branch is multiplied by its residue degree did_i; these two factors cancel. This is the same normalization, not a degree-one assertion about the residue field. Scaling gives the formula for DψiD\psi_i.

For ∣T∣≤1|T|\leq1, all ∣Pi∣≤1|P_i|\leq1, so −λ≤hi≤0-\lambda\leq h_i\leq0. For ∣T∣>1|T|>1, monicity and integral coefficients give ∣Pi(T)∣=∣T∣di|P_i(T)|=|T|^{d_i}, hence hi=log⁡∣T∣h_i=\log|T|. The asserted normalization on VV follows.

At a point of the closed unit disc, at most one of the numbers ∣Pi∣|P_i| can be strictly less than one. Indeed, two distinct irreducible reductions are coprime; a lifted Bezout identity would otherwise have left side of norm less than one and right side of norm one. Therefore hi(x)→0h_i(x)\to0 at every such point. Outside that disc all hih_i equal log⁡∣T∣\log|T|. This proves the pointwise convergence. The formula for DψD\psi is the one-variable canonical model formula at the unit Gauss point. The same coprimality argument gives all the stated point values.

Finally, take a nonzero F∈K[T]F\in K[T] and divide by a coefficient of maximum norm so that its reduction is a nonzero polynomial. Only finitely many of the P‾i\overline P_i divide that reduction. For all other ii, ∣F(ξi)∣=max⁡ν∣coeff⁡ν(F)∣=∣F(ζ)∣.|F(\xi_i)|=\max_\nu|\operatorname{coeff}_\nu(F)| =|F(\zeta)|. Polynomial seminorms define the Berkovich topology, so ξi→ζ\xi_i\to\zeta. ◻

The counterexample in every dimension

Let Ω=Vn\Omega=V^n, an ordinary open in AKn,an\mathbf A_K^{n,\mathrm{an}}. For 2≤l≤n2\leq l\leq n put bl(Tl)=λ−1max⁡{0,log⁡∣Tl∣},ui=1n(ψi(T1)+∑l=2nbl(Tl)).b_l(T_l)=\lambda^{-1}\max\{0,\log|T_l|\}, \qquad u_i=\frac1n\left(\psi_i(T_1)+\sum_{l=2}^n b_l(T_l)\right). For n=1n=1 the sum is empty. Then 0≤ui<10\leq u_i<1 on Ω\Omega, and uiu_i converges pointwise to the continuous classical psh function u=1n(ψ(T1)+∑l=2nbl(Tl)).u=\frac1n\left(\psi(T_1)+\sum_{l=2}^n b_l(T_l)\right). Let ziz_i be the product Gauss point with first coordinate ξi\xi_i and the remaining coordinates unit Gauss points, and let zz be the all-unit Gauss point. Explicitly, its seminorm on polynomials in the remaining coordinates is the maximum of the coefficient seminorms at ξi\xi_i. This description shows that zi→zz_i\to z.

Proposition (Failure of capacity precompactness and pointwise weighted stability). Set cn=n!2nnλn>0.c_n=\frac{n!}{2n^n\lambda^n}>0. For the fixed ambient background [1][1], (Dui)n=cnδzi,(Du)n=cnδz.(Du_i)^n=c_n\delta_{z_i},\qquad (Du)^n=c_n\delta_z. For i≠ji\ne j, Cap⁡[1]({∣ui−uj∣>14n};Ω)≥cn.\operatorname{Cap}_{[1]} \left(\left\{|u_i-u_j|>\frac1{4n}\right\};\Omega\right) \geq c_n. Consequently the normalized continuous classical family (ui)(u_i) has no capacity-Cauchy subsequence. The same lower bound holds for the error sets with uju_j replaced by the pointwise limit uu. This failure also holds for local capacity convergence in the all-domain sense: one compact subset of one fixed smaller domain already detects every pairwise error.

Although (Dui)n→(Du)n(Du_i)^n\to(Du)^n weakly, the weighted products fail to converge to the product of the pointwise limits: ui(Dui)n=0for every i,u(Du)n=cn2nδz≠0.u_i(Du_i)^n=0\quad\text{for every }i,\qquad u(Du)^n=\frac{c_n}{2n}\delta_z\ne0.

Proof. The mixed product with one factor from each coordinate is Dψi(T1)∧Db2(T2)∧⋯∧Dbn(Tn)=12λnδzi.D\psi_i(T_1)\wedge Db_2(T_2)\wedge\cdots\wedge Db_n(T_n) =\frac1{2\lambda^n}\delta_{z_i}. This is the product model formula: the mixed intersection number of the nn distinct pullbacks of OP1(1)\mathcal O_{\mathbf P^1}(1) is one. The first coordinate retains the residue-degree cancellation proved in the lemma. There is no factorial in this mixed product. A term with a repeated coordinate vanishes; it is the product of two curvatures pulled back from the same curve. Equivalently, the relevant positive product on (P1)n(\mathbf P^1)^n has zero mixed intersection number and hence zero mass. Expanding the nnth power of the sum defining uiu_i therefore contributes exactly n!n! equal mixed terms and the factor n−nn^{-n}. This gives cnδzic_n\delta_{z_i}; the same calculation gives the formula for uu.

At ziz_i one has ui(zi)=0,uj(zi)=u(zi)=12n(j≠i).u_i(z_i)=0,\qquad u_j(z_i)=u(z_i)=\frac1{2n}\quad(j\ne i). The error set thus contains ziz_i. The function uiu_i itself is admissible in the capacity supremum, so its Monge–Ampère mass cnc_n at ziz_i proves both lower bounds. They hold for every pair of distinct indices, excluding every capacity-Cauchy subsequence.

For the local assertion, put K∗={z}∪{zi:i≥1},Ω0=Int⁡{ ∣Tl∣≤∣π∣−1/2 for 1≤l≤n }.K_*=\{z\}\cup\{z_i:i\geq1\},\qquad \Omega_0=\operatorname{Int} \{\,|T_l|\leq|\pi|^{-1/2}\text{ for }1\leq l\leq n\,\}. The convergence zi→zz_i\to z makes K∗K_* compact, and K∗⋐Ω0⋐ΩK_*\Subset\Omega_0\Subset\Omega. All uiu_i remain normalized admissible tests after restriction to Ω0\Omega_0. Therefore Cap⁡[1](K∗∩{∣ui−uj∣>14n};Ω0)≥cn(i≠j),\operatorname{Cap}_{[1]} \left(K_*\cap\left\{|u_i-u_j|>\frac1{4n}\right\};\Omega_0\right) \geq c_n\qquad(i\ne j), and the same estimate holds with uju_j replaced by uu. Thus passing to local capacities does not restore a capacity-Cauchy subsequence.

Weak convergence of the unweighted measures follows from zi→zz_i\to z. The displayed point values give the weighted identities, with u(z)=1/(2n)u(z)=1/(2n). These are different ordinary Radon measures and are distinguished by an original smooth compact scalar cutoff which equals one near zz. ◻

Status and excluded route.

The claims of capacity precompactness of the whole normalized family, and of weighted continuity under bounded pointwise convergence, are FALSE; this is an explicit counterexample in every dimension. Each individual ingredient uses the established continuous/model calculus. The example does not contradict monotone continuity, convergence in capacity, the completion theorem, or weak-closure domination: its unweighted weak limit is itself the Monge–Ampère measure of an admissible continuous potential.

Divisorial determination of the ambient capacitary order

The statement here is an additional consequence of the model approximation and controlled-multiplier arguments already proved. It gives the pointwise reduction used to solve the increasing-envelope problem in the following section. Density of divisorial points by itself would not suffice for any assertion about discontinuous functions.

Proposition (Divisorial test for local quasi-everywhere order). Let VV be an ordinary open with compact closure in a smooth affine algebraic chart. Let a,ba,b be locally bounded real Borel functions on VV, locally quasicontinuous for the ambient capacities on all permitted local norm boxes. Then the following conditions are equivalent:

  1. a≤ba\leq b locally outside sets of outer Cap⁡[1]\operatorname{Cap}_{[1]}-capacity zero;

  2. a(z)≤b(z)a(z)\leq b(z) for every ambient divisorial point z∈Vz\in V.

In particular the local ambient capacitary equivalence class of a bounded quasicontinuous function is determined by its actual values at ambient divisorial points.

Status: DIRECT_COROLLARY. The inputs are the atomic testers, the whole-box model regularization from B, and the controlled quasicontinuous multiplier theorem; a new regularization theorem is not asserted.

Proof. For (i)⇒\Rightarrow(ii), a divisorial point has positive singleton capacity on every sufficiently small framed affine capacity domain containing it, by Lemma “A bounded model tester on the original domain”. It therefore cannot lie in a local outer-capacity-zero exceptional set.

For the converse fix a permitted capacity domain Ω⊂V\Omega\subset V and an ordinary open W⋐ΩW\Subset\Omega on a compact neighborhood of whose closure aa and bb are bounded and Cap⁡[1]( ⋅ ;Ω)\operatorname{Cap}_{[1]}(\,\cdot\,;\Omega)-quasicontinuous. Let H⊂WH\subset W be compact. Set g=(a−b)+g=(a-b)_+. This is a bounded nonnegative quasicontinuous Borel function near W‾\overline W, and it vanishes at every ambient divisorial point there.

Let v∈Cb(Ω)v\in\mathcal C_b(\Omega) with 0≤v≤10\leq v\leq1. Cover HH by finitely many smaller cores HαH_\alpha and choose nonnegative original smooth scalar cutoffs χα\chi_\alpha, supported in slightly larger norm boxes and equal to one on HαH_\alpha. The whole-box regularization from B gives, on each such larger box, a uniformly bounded decreasing sequence of continuous semipositive rational pure Fubini–Study weights vα,j↓vv_{\alpha,j}\downarrow v. Here rational pure-FS is the model class specified in B, Theorem 4.4(2); arbitrary real weights in a Fubini–Study presentation are not needed. Their Monge–Ampère measures μα,j=(Dvα,j)n\mu_{\alpha,j}=(Dv_{\alpha,j})^n are restrictions of model measures and hence are supported at ambient divisorial points. If the construction shifts vv by a fixed constant, subtract that constant; curvature and the argument are unchanged. Consequently ∫χαg dμα,j=0.\int \chi_\alpha g\,d\mu_{\alpha,j}=0 . The measures converge weakly to (Dv)n(Dv)^n and, on the support of χα\chi_\alpha, have a common bound by the capacity of one fixed intermediate norm box. The function gg is quasicontinuous for this same box. Theorem “Quasicontinuous multipliers through weak closure” therefore gives ∫χαg (Dv)n=0.\int \chi_\alpha g\,(Dv)^n=0 . Only a fixed bounded quasicontinuous multiplier has been passed through this controlled family of measures. No arbitrary Borel-multiplier continuity is used.

It follows that (Dv)n(H∩{g>δ})=0(Dv)^n(H\cap\{g>\delta\})=0 for every δ>0\delta>0. Since this holds for every normalized vv on Ω\Omega, the raw capacity of this set is zero. More usefully, the argument applies to every compact core in WW. Exhausting WW by compact sets and using continuity from below of the raw capacity shows that W∩{g>δ}W\cap\{g>\delta\} has raw capacity zero relative to Ω\Omega. This set is quasiopen relative to Ω\Omega: off an arbitrarily small ordinary exceptional open, gg is continuous on a neighborhood of W‾\overline W, and WW is open. Proposition “Quasi-open sets are outer regular” thus gives zero outer capacity as well. Taking the countable union over positive rational δ\delta proves (i). The all-domain local formulation follows by repeating the argument in every requested norm box, or by the framed-domain comparison Theorem “Uniform local comparison with a fixed domain”. ◻

Corollary (The exact divisorial form of the no-jump problem). Let uj∈Cb(U)u_j\in\mathcal C_b(U) increase and be locally uniformly bounded. Put h=sup⁡juj,Uh=h∗.h=\sup_j u_j,\qquad U_h=h^*. The preceding results give local ambient quasicontinuity of hh, and Uh∈Cb(U)U_h\in\mathcal C_b(U). The assertion h=Uhlocally Cap⁡[1]∗-quasi-everywhereh=U_h\quad\hbox{locally } \operatorname{Cap}_{[1]}^*\hbox{-quasi-everywhere} is equivalent to the assertion sup⁡juj(z)=lim sup⁡x→zsup⁡juj(x)for every ambient divisorial z.\sup_j u_j(z) =\limsup_{x\to z}\sup_j u_j(x) \quad\hbox{for every ambient divisorial }z. When the sequence has a common compact collar, these conditions are also equivalent to the energy no-jump condition proved in Proposition “The energy no-jump criterion”.

Proof. The actual increasing-limit theorem gives the required quasicontinuity of hh; the bounded upper-envelope theorem gives that of UhU_h. Apply the proposition to the pair, using h≤Uhh\leq U_h. For the energy equivalence, the earlier common-collar criterion first gives quasi-everywhere equality for its fixed capacity domain. The atomic testers on that same domain force equality at each divisorial point. The present proposition then upgrades this to the all-domain local quasi-everywhere assertion. Conversely that local assertion implies the fixed-domain one on the compact support of the common-collar difference by a finite cover and domain monotonicity. Thus it implies the earlier energy criterion as well. ◻

Remark (Why Shilov control alone did not close the problem). For a fixed psh function, the Shilov maximum principle controls its values on an affinoid by its Shilov boundary. The one-Shilov domain constructed at a divisorial point in this record need not contain that point in its ordinary interior. Thus that domain alone does not control the limsup over all ordinary directions appearing in the displayed equality. A uniform upper-control statement for bounded psh families at divisorial points is needed in addition to this reduction. It is supplied by Theorem “Uniform upper control at a divisorial point”; the resulting no-jump conclusion is Theorem “Increasing envelopes have no ambient capacity jump”.

Divisorial barriers and the absence of an upper-envelope jump

The argument below supplies the missing ordinary-neighborhood control at a divisorial point. It uses a local model potential whose Monge–Ampère measure has only one atom on that neighborhood. The potential need not vanish only at that point: its zero set is an affinoid domain. The proof uses the already established pointwise ambient domination principle, not an assertion that negligible sets are pluripolar.

All model constructions are over the fixed field KK, with its original absolute value and integration calibration. Write R=K∘R=K^\circ. Throughout this section n≥1n\geq1.

Lemma (An algebraic affine model inside a prescribed neighborhood). Let Y/KY/K be smooth integral affine of dimension nn, let O⊂YanO\subset Y^{\mathrm{an}} be ordinary open, and let z∈Oz\in O be divisorial. There is a normal, integral, flat, finite-type RR-algebra BB with the following properties. Its generic fiber YB=Spec⁡(B⊗RK)Y_B=\operatorname{Spec}(B\otimes_RK) is a smooth affine algebraic open in YY. The strict affinoid generic fiber A=(Spf⁡B^)η⊂YBanA=\bigl(\operatorname{Spf}\widehat B\bigr)_\eta \subset Y_B^{\mathrm{an}} of its uniformizer-adic completion is contained in OO and contains zz. The special fiber of Spec⁡B\operatorname{Spec}B has exactly one irreducible component, and its generic point is the reduction of zz. In particular Γ(A)={z}\Gamma(A)=\{z\}.

Proof. Choose a compact polynomial norm box CC with z∈Int⁡(C),C⊂O.z\in\operatorname{Int}(C),\qquad C\subset O. After clearing rational powers in its thresholds, write its finitely many defining inequalities as ∣ai∣≤1(1≤i≤s)|a_i|\leq1\quad(1\leq i\leq s) on an affine algebraic open Y0⊂YY_0\subset Y. Here the aia_i are regular on Y0Y_0: an upper constraint uses a ratio FN/cF^N/c, and a lower constraint uses d/GMd/G^M after removing G=0G=0. Bounds on affine coordinates are included, so the displayed locus is the compact box CC. Such thresholds can be chosen strictly around the values at zz, since rational powers of ∣K×∣|K^\times| are dense in R>0\mathbf R_{>0}. In particular ∣ai(z)∣≤1|a_i(z)|\leq1.

Take a normal projective compactification Y‾\overline Y of YY. Proposition A.9 of [GM19] realizes zz on a proper algebraic RR-model of Y‾\overline Y. Normalize that model. Normalization is finite because the model is of finite type over the excellent DVR RR; it is an isomorphism on the normal generic fiber. The center of zz is still the generic point ηE\eta_E of a special-fiber component EE. Equivalently, one can use the strict transform of the component on a normal dominating model. Its height-one local ring is the valuation ring of the same divisorial valuation.

For each ii, the inequality ∣ai(z)∣≤1|a_i(z)|\leq1 says exactly that the rational function aia_i belongs to the DVR OY,ηE\mathcal O_{\mathscr Y,\eta_E}. It is therefore regular on a Zariski neighborhood of ηE\eta_E. Intersect these finitely many neighborhoods. Remove the other special-fiber components. Also remove the closure of the generic boundary Y‾∖Y0\overline Y\setminus Y_0. This last removal does not remove ηE\eta_E. Indeed, the ideal II of the schematic closure is uniformizer-saturated. At the DVR OY,ηE\mathcal O_{\mathscr Y,\eta_E} its localization is nonzero, since the boundary does not contain the generic point of Y‾\overline Y. Any nonzero ideal of that DVR contains a power of the uniformizer of RR. Saturation then forces IηE=OY,ηEI_{\eta_E}=\mathcal O_{\mathscr Y,\eta_E}, so the closure does not contain ηE\eta_E.

Choose an affine open Spec⁡B\operatorname{Spec}B containing ηE\eta_E inside the resulting open. Then BB is normal, flat and finite type over RR, its special fiber has only the component coming from EE, and B⊗RKB\otimes_RK defines a smooth affine open YBY_B of Y0Y_0. All aia_i belong to BB. Thus ∣ai∣≤1|a_i|\leq1 on the generic fiber of Spf⁡B^\operatorname{Spf}\widehat B, so A⊂CA\subset C. It contains zz because its reduction is ηE∈Spec⁡B\eta_E\in\operatorname{Spec}B.

Normality supplies the unique point above this special-fiber generic point, by [CLD12, Proposition 6.9.7]; the completion has the corresponding relatively normal formal model, as RR is excellent. The affine formal-model description of the Shilov boundary in [GM19, Proposition A.3] now gives Γ(A)={z}\Gamma(A)=\{z\}. No generic modification has changed Y0Y_0: the only normalization above is an isomorphism on the already normal generic compactification. In particular A⊂OA\subset O is an actual domain of the originally specified analytic space. ◻

Lemma (A local model barrier with a single divisorial atom). Under the hypotheses of Lemma “An algebraic affine model inside a prescribed neighborhood”, there are a smooth affine algebraic open YB⊂YY_B\subset Y, a continuous semipositive PL function ρ∈Cb(YBan)\rho\in\mathcal C_b(Y_B^{\mathrm{an}}), and a real number ϵ>0\epsilon>0 such that ρ≥0,ρ(z)=0,Ω={ρ<ϵ},{ρ≤2ϵ}⋐O∩YBan,\rho\geq0,\qquad \rho(z)=0,\qquad \Omega=\{\rho<\epsilon\},\qquad \{\rho\leq2\epsilon\}\Subset O\cap Y_B^{\mathrm{an}}, and, on the ordinary open neighborhood Ω\Omega of zz, (Dρ)n=czδz(cz>0).(D\rho)^n=c_z\delta_z\qquad(c_z>0). Every sublevel {ρ≤t}\{\rho\leq t\} is compact in YBanY_B^{\mathrm{an}}. All statements about the measure are ordinary CLD-current statements, with the accepted continuous BT compatibility.

Proof. Use BB from the preceding lemma and choose finitely many RR-algebra generators g1,…,gNg_1,\ldots,g_N. Their closed embedding Spec⁡B↪ARN\operatorname{Spec}B\hookrightarrow\mathbf A_R^N identifies its formal generic fiber with A={∣gi∣≤1 for every i}⊂YBan.A=\{|g_i|\leq1\text{ for every }i\}\subset Y_B^{\mathrm{an}}. Define ρ=log⁡max⁡{1,∣g1∣,…,∣gN∣}.\rho=\log\max\{1,|g_1|,\ldots,|g_N|\}. The function is finite, continuous and classical after all complete extensions: pullback to a curve is a finite maximum of logarithms of analytic functions and the constant zero. It is semipositive PL. Its zero set is exactly AA. Closed sublevels are intersections of the closed affine variety YBanY_B^{\mathrm{an}} with compact closed polydiscs, hence are compact.

Take the projective closure of Spec⁡B\operatorname{Spec}B in PRN\mathbf P_R^N, and normalize it. Denote this normal projective flat model by Z\mathscr Z. Its normalization is finite, and its standard affine chart X0≠0X_0\ne0 is still Spec⁡B\operatorname{Spec}B because BB is normal. Let L=ν∗OPRN(1).\mathscr L=\nu^*\mathcal O_{\mathbf P_R^N}(1). It is ample and generated, and its model weight in the frame X0X_0 on YBY_B is exactly ρ\rho, with gi=Xi/X0g_i=X_i/X_0. The generic fiber is normal and generically smooth over the characteristic-zero field KK. The original CLD model formula therefore applies, since the discretely valued field KK is stable; see [CLD12, Theorem 6.9.3, §6.9.6 and Proposition 6.9.7]. It gives the finite measure c1(L)n=∑FbFdeg⁡K~,F((L∣F)n) δzF.c_1(\mathscr L)^n =\sum_F b_F\deg_{\widetilde K,F} ((\mathscr L|_F)^n)\,\delta_{z_F}. Here FF runs over special-fiber components, bF>0b_F>0 is its multiplicity, and the degree retains its structural residue-field degree. Ampleness makes every displayed coefficient positive. The points zFz_F lie in the analytification of every nonempty generic algebraic open, so ρ(zF)\rho(z_F) is finite.

We determine exactly which of these atoms have height zero. At the generic point of FF, choose a local frame ee of L\mathscr L. The sections Xi/eX_i/e are regular in the corresponding DVR, and at least one is a unit because the sections generate L\mathscr L. Consequently ρ(zF)=log⁡max⁡i∣Xi/X0∣(zF)=−log⁡∣X0/e∣(zF).\rho(z_F) =\log\max_i|X_i/X_0|(z_F) =-\log|X_0/e|(z_F). This equals zero precisely when X0/eX_0/e is a unit, that is, when the generic point of FF belongs to the affine chart X0≠0X_0\ne0. That chart has exactly one special-fiber component. Its associated point is zz. Thus zz is the only height-zero atom, and every other atom has strictly positive height.

Choose 2ϵ2\epsilon smaller than the finitely many positive heights, if there are any. Shrink ϵ\epsilon further so that {ρ≤2ϵ}⊂O\{\rho\leq2\epsilon\}\subset O: the compact sublevels decrease to A⊂OA\subset O, so compactness gives this inclusion. They are compact subsets of the open O∩YBanO\cap Y_B^{\mathrm{an}}, as required. The restriction of the displayed model measure to Ω={ρ<ϵ}\Omega=\{\rho<\epsilon\} is czδzc_z\delta_z, where cz=bEdeg⁡K~,E((L∣E)n)>0.c_z=b_E\deg_{\widetilde K,E}((\mathscr L|_E)^n)>0. On the smooth open YBY_B its model curvature measure is the ordinary continuous BT product by the accepted compatibility from R and B. This proves the claimed identity against all original compact smooth scalar tests. No equation with a prescribed mass has been solved, and no change of field or normalization has been made. ◻

Theorem (Uniform upper control at a divisorial point). Use ρ,ϵ,Ω\rho,\epsilon,\Omega from Lemma “A local model barrier with a single divisorial atom”. Given real numbers m≤Mm\leq M, every u∈Cb(Ω)u\in\mathcal C_b(\Omega) satisfying u≤M on Ω,u(z)≥mu\leq M\text{ on }\Omega,\qquad u(z)\geq m obeys u(x)≤u(z)+Aρ(x)(x∈Ω),A=2(M−m+1)ϵ.u(x)\leq u(z)+A\rho(x)\quad(x\in\Omega), \qquad A=\frac{2(M-m+1)}\epsilon. In particular, a family with these same bounds is uniformly upper semicontinuous at zz in the ordinary topology: for each δ>0\delta>0 the ordinary neighborhood Ω∩{Aρ<δ}\Omega\cap\{A\rho<\delta\} works simultaneously for every member of the family.

Proof. Put v=u(z)+Aρv=u(z)+A\rho. This is continuous classical psh and (Dv)n=Anczδz.(Dv)^n=A^n c_z\delta_z. On {ρ≥ϵ/2}∩Ω\{\rho\geq\epsilon/2\}\cap\Omega one has v≥m+Aϵ/2=M+1>u.v\geq m+A\epsilon/2=M+1>u. Thus for every δ>0\delta>0, {v<u−δ}‾⊂{ρ≤ϵ/2}⋐Ω.\overline{\{v<u-\delta\}} \subset\{\rho\leq\epsilon/2\}\Subset\Omega. This verifies the compact-gap condition at every part of the ordinary boundary, without treating the discontinuous strict gap as an ordinary open set. At zz one has v(z)=u(z)v(z)=u(z), so (Dv)n({v<u})=0.(Dv)^n(\{v<u\})=0. The domain Ω\Omega has compact closure in the smooth affine chart YBY_B, and therefore satisfies the framed affine hypothesis of Corollary “Pointwise ambient domination and uniqueness”. Apply that proved pointwise domination principle to the pair (v,u)(v,u). It gives v≥uv\geq u everywhere. Since ρ\rho is continuous and vanishes at zz, the simultaneous-neighborhood assertion follows. ◻

The two geometric lemmas are direct consequences of model realization, normality and the model-measure formula, with the neighborhood construction made explicit. The uniform upper control is recorded as NEW_PROVED relative to the previous ambient domination theorem. Its proof uses neither the upper-envelope identification nor a polar-set theorem.

Corollary (Upper regularization preserves divisorial values). Let UU have the allowed analytic hypotheses. Let F⊂Cb(U)\mathcal F\subset\mathcal C_b(U) be nonempty and locally uniformly bounded above, and put h=sup⁡u∈Fu,Uh=h∗.h=\sup_{u\in\mathcal F}u,\qquad U_h=h^*. Then Uh(z)=h(z)at every ambient divisorial z∈U.U_h(z)=h(z)\qquad\text{at every ambient divisorial }z\in U. The family can be uncountable; no measurability of hh is used in this assertion.

Proof. Fix zz and pass to an ordinary smooth affine algebraic neighborhood, using the accepted B, Lemma D.4. Choose u0∈Fu_0\in\mathcal F and replace every uu by max⁡(u,u0)\max(u,u_0). The supremum is unchanged. The new family is uniformly bounded above and below on a compact neighborhood of zz, since u0u_0 is locally bounded. Apply the preceding theorem on a smaller Ω\Omega, with common constants m,M,Am,M,A and the fixed barrier ρ\rho. Taking the supremum gives h(x)≤h(z)+Aρ(x)(x∈Ω).h(x)\leq h(z)+A\rho(x)\quad(x\in\Omega). Ordinary upper regularization at zz, where ρ(z)=0\rho(z)=0, yields Uh(z)≤h(z)U_h(z)\leq h(z). The reverse inequality holds by definition. All algebraic charts and barriers here are over the original KK; no geometric-integrality field change is needed for this step. ◻

Theorem (Increasing envelopes have no ambient capacity jump). Let uj∈Cb(U)u_j\in\mathcal C_b(U) increase and be locally uniformly bounded above. Set h=sup⁡jujh=\sup_j u_j and Uh=h∗U_h=h^*. Then Uh∈Cb(U),h=Uhlocally outside sets of outer Cap⁡[1]-capacity zero.U_h\in\mathcal C_b(U),\qquad h=U_h\quad\text{locally outside sets of outer } \operatorname{Cap}_{[1]}\text{-capacity zero}. Moreover uj→Uhu_j\to U_h locally in outer ambient capacity on every permitted norm-box domain. If the sequence has one common compact collar, its ambient energy satisfies Eb(Uh)=sup⁡jEb(uj).\mathcal E_b(U_h)=\sup_j\mathcal E_b(u_j).

Proof. The established bounded upper-envelope theorem gives Uh∈Cb(U)U_h\in\mathcal C_b(U). Actual increasing convergence gives local ambient quasicontinuity of the Borel function hh, and quasicontinuity of UhU_h is already known. The preceding corollary identifies their values at every ambient divisorial point. Proposition “Divisorial test for local quasi-everywhere order” therefore gives the asserted local outer-capacity-zero equality. Combine it with actual increasing convergence to hh and subadditivity of outer capacity to obtain convergence to UhU_h. The common-collar energy assertion is then Proposition “The energy no-jump criterion”, for the ambient background [1][1]. ◻

Theorem (Negligible upper-envelope sets have ambient outer capacity zero). Let F⊂Cb(U)\mathcal F\subset\mathcal C_b(U) be nonempty and locally uniformly bounded above. Write h=sup⁡u∈Fu,Uh=h∗.h=\sup_{u\in\mathcal F}u,\qquad U_h=h^*. Then the set {h<Uh}\{h<U_h\} has locally zero outer Cap⁡[1]\operatorname{Cap}_{[1]}-capacity, whether or not the raw supremum hh is Borel. Locally, a countable family of finite maxima of members of F\mathcal F has an increasing actual supremum HH with H≤h≤Uh,H=Uhlocally ambient-quasi-everywhere.H\leq h\leq U_h, \qquad H=U_h\quad\text{locally ambient-quasi-everywhere}.

Proof. Work on one permitted local norm-box domain WW. The local capacitary-envelope theorem, with background [1][1], supplies an increasing sequence of finite maxima of members of F∣W\mathcal F|_W whose actual Borel supremum HH belongs to A[1],b(W)\mathcal A_{[1],b}(W). It satisfies H≤hH\leq h pointwise, and every u∈Fu\in\mathcal F satisfies u≤Hu\leq H locally ambient-quasi-everywhere.

Fix an ambient divisorial z∈Wz\in W. Its singleton has positive capacity on each sufficiently small framed affine domain containing it. Thus the exceptional set for each individual inequality u≤Hu\leq H cannot contain zz. Consequently u(z)≤H(z)u(z)\leq H(z) for every u∈Fu\in\mathcal F, even though their exceptional sets may depend on uu. Taking the supremum and using H≤hH\leq h gives H(z)=h(z)=Uh(z),H(z)=h(z)=U_h(z), where the last equality is Corollary “Upper regularization preserves divisorial values”.

Both HH and UhU_h are locally bounded, Borel and locally ambient-quasicontinuous. Proposition “Divisorial test for local quasi-everywhere order” gives H=UhH=U_h locally outside a set of outer capacity zero. The pointwise sandwich implies {h<Uh}⊂{H<Uh}.\{h<U_h\}\subset\{H<U_h\}. Outer capacity is defined for arbitrary sets and is monotone, so no Borel measurability of the left side is needed. This proves the theorem locally on every requested domain. ◻

Scope and dependencies.

The upper-envelope conclusions in this section are NEW_PROVED consequences of the model barrier and the previously proved ambient calculus. They concern the ambient capacity Cap⁡[1]\operatorname{Cap}_{[1]}; the assertion for unrestricted positive closed backgrounds remains false, as the subvariety-supported example already shows. Zero ambient outer capacity has been proved here. A converse identification of all zero-capacity sets with pluripolar sets is not used or asserted.

The proof order is model realization and the original model formula, then the established pointwise ambient domination, then uniform divisorial upper control. The quasicontinuous divisorial criterion and actual increasing convergence are used only after that control is proved. Thus no assertion about negligible sets is inserted into its own proof.

Classical representatives of the ambient capacitary closure

The completion retains actual Borel functions. The result here identifies their classes modulo ambient outer-capacity-zero sets; it does not identify all of their actual point values. It is specific to the ambient background [1][1].

Corollary (The unique classical representative). Let UU have the allowed analytic hypotheses. Every a∈A[1],b(U)a\in\mathcal A_{[1],b}(U) has a unique function ra∈Cb(U)r_a\in\mathcal C_b(U) such that a=ralocally outside sets of outer Cap⁡[1]-capacity zero.a=r_a\quad\text{locally outside sets of outer } \operatorname{Cap}_{[1]}\text{-capacity zero}. At each ambient divisorial point zz one has ra(z)=a(z)r_a(z)=a(z). If a∈A[1]M(W)a\in\mathcal A^M_{[1]}(W) on a norm-box domain WW, then ∣ra∣≤M|r_a|\leq M there. For any defining approximants aj∈Cb(W)a_j\in\mathcal C_b(W) with ∣aj∣≤M|a_j|\leq M, one formula is ra=lim⁡m→∞(sup⁡j≥maj)∗.r_a=\lim_{m\to\infty} \left(\sup_{j\geq m}a_j\right)^*. The right side is a decreasing limit of bounded classical psh functions, and the final actual classical function is independent of the approximants.

Status: DIRECT_COROLLARY. The inputs are divisorial no-jump, the divisorial-capacitary order criterion, and the already proved existence of capacitary limits.

Proof. First work on a defining domain WW with a∈A[1]M(W)a\in\mathcal A^M_{[1]}(W) and approximants aja_j as stated. Set vm=(sup⁡j≥maj)∗.v_m=\left(\sup_{j\geq m}a_j\right)^*. The bounded upper-envelope theorem gives vm∈Cb(W)v_m\in\mathcal C_b(W), and −M≤vm≤M-M\leq v_m\leq M. The tails decrease, so vmv_m decreases to a bounded classical psh function rr with the same bound.

Fix an ambient divisorial point z∈Wz\in W and choose a small framed affine norm-box capacity domain containing zz. Its singleton has positive capacity by the atomic tester lemma. Local outer-capacity convergence aj→aa_j\to a, applied to the compact singleton {z}\{z\}, therefore implies aj(z)→a(z)a_j(z)\to a(z). Otherwise one fixed positive error threshold would put that singleton into infinitely many error sets, contradicting their capacities tending to zero. Corollary “Upper regularization preserves divisorial values”, applied to each tail, gives vm(z)=sup⁡j≥maj(z).v_m(z)=\sup_{j\geq m}a_j(z). It follows that r(z)=a(z)r(z)=a(z) at every ambient divisorial point. The function aa is locally quasicontinuous by the completion theorem; rr is locally quasicontinuous as a bounded classical potential. Proposition “Divisorial test for local quasi-everywhere order” therefore gives r=ar=a locally ambient-quasi-everywhere.

For uniqueness, suppose r,s∈Cb(W)r,s\in\mathcal C_b(W) have this property. Positive singleton capacity forces r(z)=s(z)r(z)=s(z) at every ambient divisorial point. Shilov boundaries of strict affinoid neighborhoods consist of such points. The actual pointwise order lemma, Lemma “A strict gap contains a Shilov point”, applied in both orders, gives r=sr=s at every point of WW. This also proves independence of the defining approximants.

For a general local member a∈A[1],b(U)a\in\mathcal A_{[1],b}(U), perform the construction on each defining neighborhood. On an ordinary overlap the two resulting classical functions are both locally quasi-everywhere equal to aa, hence are actually equal by the uniqueness just proved. They glue to ra∈Cb(U)r_a\in\mathcal C_b(U) by the classical sheaf property, which retains the all-extensions curve tests. Its divisorial values and uniqueness are local and hence hold on UU. ◻

Remark (The representative need not be the usc regularization of the Borel function). Take a connected ordinary polydisc W⊂AKn,anW\subset\mathbf A_K^{n,\mathrm{an}}, n≥1n\geq1, containing the KK-rational origin oo, and put a=1{o}a=\mathbf1_{\{o\}}. The singleton is locally pluripolar: the classical function log⁡∣T1∣\log|T_1| is not identically −∞-\infty and has a pole there. The already proved ambient polar theorem therefore makes {o}\{o\} locally outer-capacity zero. Consequently the constant zero sequence converges to aa in local outer capacity, so a∈A[1],b(W)a\in\mathcal A_{[1],b}(W) and ra=0r_a=0.

The point oo is closed, so its indicator is upper semicontinuous and a∗=aa^*=a. This function is not classical psh: it attains the strict value one at an interior point and is zero elsewhere, contradicting the classical strong maximum principle on the connected domain. Thus rar_a is not in general a∗a^*. For the actual supremum of an increasing classical sequence, in contrast, the proved no-jump theorem does identify its classical representative with that supremum’s usc regularization. This example uses only the ambient polar-zero theorem; it does not depend on a converse from capacity zero to pluripolarity.

Corollary (The completed ambient calculus uses the classical representative). For 0≤k≤n0\leq k\leq n and a0,…,ak∈A[1],b(U)a_0,\ldots,a_k\in\mathcal A_{[1],b}(U), T[1],k(a1,…,ak)=T[1],k(ra1,…,rak),\mathcal T_{[1],k}(a_1,\ldots,a_k) =T_{[1],k}(r_{a_1},\ldots,r_{a_k}), and a0T[1],k(a1,…,ak)=ra0T[1],k(ra1,…,rak).a_0\mathcal T_{[1],k}(a_1,\ldots,a_k) =r_{a_0}T_{[1],k}(r_{a_1},\ldots,r_{a_k}). These are equalities of coefficient-Radon currents on every original complementary smooth test form. A uniformly bounded classical sequence that is locally Cauchy in ambient capacity thus converges locally in outer capacity to a unique actual classical function, with the established joint weighted BT continuity whenever all the slots have the stated bounds.

Proof. The completed calculus is independent of changes on local outer-capacity-zero sets and agrees with the classical calculus on Cb\mathcal C_b, by Theorem “BT calculus on the bounded capacitary closure”. This gives the unweighted identity. The same theorem’s coefficient bounds by outer capacity make every coefficient measure of this product vanish on the local disagreement set of a0a_0 and ra0r_{a_0}, proving the weighted identity as a Borel multiplication identity. A local capacity-Cauchy sequence has a limit in the completed class; the first corollary identifies its unique classical representative, and zero-capacity changes of the limit preserve convergence. The weighted convergence assertion is then the already established completed calculus theorem. ◻

No corresponding conclusion for an arbitrary background SS is claimed. For example, a background supported on an ambient pluripolar subvariety can distinguish the actual increasing limit from its ambient upper regularization. The relative capacitary closure remains necessary in that setting.

A Dirac Dirichlet problem on the constructed divisorial domains

Corollary (Constant boundary data and one divisorial atom). Let z,ρ,ϵ,Ωz,\rho,\epsilon,\Omega and czc_z be supplied by Lemma “A local model barrier with a single divisorial atom”. For every real m>0m>0 and b∈Rb\in\mathbf R, the function um,b=b+(mcz)1/n(ρ−ϵ)u_{m,b}=b+\left(\frac m{c_z}\right)^{1/n} (\rho-\epsilon) is continuous on a neighborhood of Ω‾\overline\Omega, belongs to Cb(Ω)\mathcal C_b(\Omega), and satisfies (Dum,b)n=mδzon Ω,um,b=bon ∂Ω.(Du_{m,b})^n=m\delta_z\quad\text{on }\Omega, \qquad u_{m,b}=b\quad\text{on }\partial\Omega. Here the closure and boundary are ordinary ones in the smooth affine chart YBanY_B^{\mathrm{an}}. It is the unique bounded classical solution with this measure and the boundary condition lim⁡x→yx∈Ωw(x)=b(y∈∂Ω).\lim_{\substack{x\to y\\x\in\Omega}}w(x)=b \quad(y\in\partial\Omega). In particular uniqueness holds among classical solutions continuous on Ω‾\overline\Omega with boundary value bb.

Status: DIRECT_COROLLARY. The domain and its barrier have been constructed together. This corollary concerns that family of domains and a single divisorial atom. It makes no prescribed-domain assertion for general atomic or diffuse measures.

Proof. Positive scaling and translation preserve classicality, and the proved homogeneity of the BT product gives (Dum,b)n=mcz(Dρ)n=mδz.(Du_{m,b})^n=\frac m{c_z}(D\rho)^n=m\delta_z. Continuity is inherited from ρ\rho. Since Ω={ρ<ϵ}\Omega=\{\rho<\epsilon\} and ρ\rho is continuous, each point of its ordinary boundary satisfies ρ=ϵ\rho=\epsilon. This proves the boundary value.

Suppose ww is another solution with the stated boundary limit. Both ww and um,bu_{m,b} have the same finite boundary limit at every point. For each δ>0\delta>0, every boundary point therefore has an ordinary neighborhood on which their difference has absolute value less than δ\delta, at points of Ω\Omega. The boundary is compact, being a closed subset of the compact Ω‾\overline\Omega. A finite union of these neighborhoods covers it. Consequently both strict-gap sets with margin δ\delta have closure compactly contained in Ω\Omega. The equality of the Monge–Ampère measures and Corollary “Pointwise ambient domination and uniqueness” now give w=um,bw=u_{m,b} at every point. ◻

Ambient extremals on domains with a bounded negative exhaustion

This section concerns S=[1]S=[1]. The domain hypothesis is explicit and weaker than a strictly psh exhaustion. Its conclusions use the proved divisorial no-jump theorem; none is claimed here for an arbitrary positive closed background.

Definition (The exhaustion domains used in this section). Let Y/KY/K be smooth affine of dimension n≥1n\geq1. An exhaustion domain is a connected ordinary open V⋐YanV\Subset Y^{\mathrm{an}} with nonempty ordinary boundary and a function ρ∈Cb(V)∩C(V‾),ρ<0 on V,ρ=0 on ∂V.\rho\in\mathcal C_b(V)\cap C(\overline V),\qquad \rho<0\text{ on }V,\qquad \rho=0\text{ on }\partial V. In particular ρ\rho is bounded and {ρ≤−ϵ}⋐V\{\rho\leq-\epsilon\}\Subset V for every ϵ>0\epsilon>0. No positivity stronger than classical psh is required of ρ\rho, and no equation for (Dρ)n(D\rho)^n is assumed.

Lemma (Local supply of exhaustion domains). Every point in an allowed ordinary algebraic neighborhood has arbitrarily small exhaustion domains of the preceding type. This does not assert that every ordinary analytic domain is an exhaustion domain.

Proof. Inside a prescribed ordinary neighborhood OO, choose a compact rational polynomial norm box CC containing the point in its ordinary interior. After passing to an affine algebraic open Y0Y_0 and clearing powers in the rational thresholds, write C={∣ai∣≤1 for 1≤i≤N}.C=\{|a_i|\leq1\text{ for }1\leq i\leq N\}. The regular functions aia_i include the required bounds on affine coordinates and on inverses used for lower constraints. Consequently each locus {∣ai∣≤et for all i}\{|a_i|\leq e^t\text{ for all }i\}, with t≥0t\geq0, is compact in Y0anY_0^{\mathrm{an}}. Equivalently, one can include scaled powers of a set of affine algebra generators. Put ρ0=log⁡max⁡(1,∣a1∣,…,∣aN∣).\rho_0=\log\max(1,|a_1|,\ldots,|a_N|). This is continuous classical psh by the finite maximum rule after complete extensions and curve pullbacks. Its compact sublevels decrease to CC as t↓0t\downarrow0. Choose a small positive rational valuation threshold ϵ\epsilon such that {ρ0≤2ϵ}⊂O\{\rho_0\leq2\epsilon\}\subset O.

Analytic spaces are locally connected, as recalled in [GR25] from Berkovich’s local connectedness theorem. Let VV be the connected component containing the specified point of {ρ0<ϵ}\{\rho_0<\epsilon\}. It is ordinary open, and its compact closure is contained in {ρ0≤ϵ}\{\rho_0\leq\epsilon\}. Since a connected component is relatively closed in that ordinary open, its boundary is contained in {ρ0=ϵ}\{\rho_0=\epsilon\}. Thus ρ=ρ0−ϵ\rho=\rho_0-\epsilon is the required bounded continuous negative exhaustion.

For completeness the boundary is nonempty. Otherwise VV would be a compact open-and-closed subdomain of the strict affinoid {ρ0≤2ϵ}\{\rho_0\leq2\epsilon\}, hence an affinoid component. It is pure of positive dimension and therefore has a Shilov point on its analytic boundary, by [GR25, Lemma 6.2]. This contradicts its being an ordinary open in the boundaryless space Y0anY_0^{\mathrm{an}} and the interior-composition rule [GR25, §2.11(3)–(4)]. ◻

The lemma is a DIRECT_COROLLARY of the ordinary norm-box geometry and classical maximum calculus already used in the record. It does not supply strict positivity or peak barriers at every boundary point.

Fix an exhaustion domain VV for the rest of this section and keep the capacity domain fixed: C(E)=Cap⁡[1](E;V),C∗(E)=CV∗(E).C(E)=\operatorname{Cap}_{[1]}(E;V),\qquad C^*(E)=C_V^*(E). For any subset E⊂VE\subset V, define FE={v∈Cb(V):−1≤v≤0, v=−1 on E},hE=sup⁡v∈FEv,UE=hE∗.\mathcal F_E=\{v\in\mathcal C_b(V):-1\leq v\leq0,\ v=-1\text{ on }E\}, \qquad h_E=\sup_{v\in\mathcal F_E}v,\qquad U_E=h_E^*. The constraints in FE\mathcal F_E are actual pointwise constraints, including when EE is not Borel. The family contains the constant −1-1. When E=∅E=\varnothing, UE=0U_E=0.

Lemma (Basic extremal properties and a countable selection). One has UE∈Cb(V)U_E\in\mathcal C_b(V) and −1≤UE≤0-1\leq U_E\leq0. If E⊂FE\subset F, then UF≤UEU_F\leq U_E pointwise. The exceptional subset E∩{UE>−1}E\cap\{U_E>-1\} has locally zero outer ambient capacity. If E⋐VE\Subset V, then UEU_E has ordinary boundary limit zero, and the exceptional subset has C∗C^*-capacity zero in the fixed domain. There is an increasing sequence uj∈FEu_j\in\mathcal F_E with (sup⁡juj)∗=UE,uj⟶UElocally in outer ambient capacity.\left(\sup_j u_j\right)^*=U_E,\qquad u_j\longrightarrow U_E \quad\text{locally in outer ambient capacity}.

Proof. The bounded upper-envelope theorem gives classicality. Monotonicity follows from inclusion of the admissible families. Since hE=−1h_E=-1 on EE, the exceptional subset is contained in {hE<UE}\{h_E<U_E\}. The arbitrary-family negligible-set theorem gives its local outer-capacity zero, without requiring that hEh_E or EE be Borel. On a relatively compact EE, a finite local cover and domain monotonicity give zero for the fixed domain VV.

If E⋐VE\Subset V, choose c>0c>0 such that cρ≤−1c\rho\leq-1 on E‾\overline E. The function max⁡(−1,cρ)\max(-1,c\rho) belongs to FE\mathcal F_E. It tends to zero at every ordinary boundary point, and max⁡(−1,cρ)≤UE≤0.\max(-1,c\rho)\leq U_E\leq0. This proves the boundary assertion. The same lower barrier works simultaneously for all EE contained in one fixed compact subset of VV.

Here is an elementary countable selection, independent of Choquet’s capacitability theorem. The finite pp-adic field makes VV second countable. For a countable ordinary basis (Bl)(B_l), let Ml=sup⁡{v(x):v∈FE, x∈Bl}.M_l=\sup\{v(x):v\in\mathcal F_E,\ x\in B_l\}. For each l,ml,m choose vl,m∈FEv_{l,m}\in\mathcal F_E and xl,m∈Blx_{l,m}\in B_l with vl,m(xl,m)>Ml−1/mv_{l,m}(x_{l,m})>M_l-1/m. Enumerate these functions and take their successive finite maxima uju_j. They remain in FE\mathcal F_E. For a=sup⁡juja=\sup_j u_j, one has sup⁡Bla=Ml=sup⁡BlhE\sup_{B_l}a=M_l=\sup_{B_l}h_E for each ll. The neighborhood formula for upper regularization therefore gives a∗=hE∗=UEa^*=h_E^*=U_E. The proved increasing no-jump theorem then gives the claimed local outer-capacity convergence. ◻

Lemma (Order of total Monge–Ampère masses). Let u,v∈Cb(V)u,v\in\mathcal C_b(V) be bounded, with u≤v≤0u\leq v\leq0 and ordinary boundary limits zero. Then ∫V(Dv)n≤∫V(Du)n.\int_V(Dv)^n\leq\int_V(Du)^n. The assertion allows infinite masses and contains no subtraction of infinities.

Proof. If v≡0v\equiv0 the assertion is immediate. Otherwise the strong maximum principle on the connected domain gives v<0v<0 everywhere. For 0<t<10<t<1, put w=(1−t)vw=(1-t)v. Then w>v≥uw>v\geq u on all of VV. The boundary limits imply {u<w−ϵ}‾⋐V(ϵ>0).\overline{\{u<w-\epsilon\}}\Subset V\qquad(\epsilon>0). Indeed each such gap is absent on a neighborhood of the compact ordinary boundary. The comparison principle gives (1−t)n∫V(Dv)n=∫{u<w}(Dw)n≤∫{u<w}(Du)n=∫V(Du)n.(1-t)^n\int_V(Dv)^n =\int_{\{u<w\}}(Dw)^n \leq\int_{\{u<w\}}(Du)^n =\int_V(Du)^n. Let t↓0t\downarrow0. ◻

Theorem (The compact extremal and its equilibrium mass). For every compact K⋐VK\Subset V, the function UKU_K has ordinary boundary limit zero and is locally maximal on V∖KV\setminus K. Its measure μK=(DUK)n\mu_K=(DU_K)^n is supported in KK, is finite, and satisfies

C(K)=μK(K)=μK(V).C(K)=\mu_K(K)=\mu_K(V).

The equality UK=−1U_K=-1 on KK is asserted outside a C∗C^*-zero subset; it is not asserted at every point of KK.

Proof. The boundary statement and the zero-capacity exceptional set are in Lemma “Basic extremal properties and a countable selection”. Choose there an increasing sequence uj∈FKu_j\in\mathcal F_K whose upper-regularized supremum is UKU_K.

We first prove that the measure vanishes off KK. Work in a small ordinary W⋐V∖KW\Subset V\setminus K on which the constructive common projective globalization applies, and fix a nonnegative original smooth scalar χ\chi compactly supported in WW. Globalize the whole bounded family uju_j on one identified ample line, obtaining bounded classical metrics RjR_j whose weights on a neighborhood of supp⁡χ\operatorname{supp}\chi are uj+cu_j+c with one fixed cc. The construction is a maximum with one common collar, so the metrics increase and have one common bound. Extend χ\chi by zero to the projective space. For a fixed ϵ>0\epsilon>0, put pj=PLb(Rj+ϵχ).p_j=P_L^{\mathrm b}(R_j+\epsilon\chi). The exact obstacle inequalities give Rj≤pj≤Rj+ϵχR_j\leq p_j\leq R_j+\epsilon\chi, hence pj=Rjp_j=R_j off supp⁡χ\operatorname{supp}\chi. Envelope monotonicity makes pjp_j increasing.

Its local weight minus cc glues with uju_j off that support to a function wj∈Cb(V)w_j\in\mathcal C_b(V). This modification is compact and uniformly bounded. It satisfies wj≥uj≥−1w_j\geq u_j\geq-1, and has boundary limsup at most zero. The boundary maximum principle gives wj≤0w_j\leq0. Since the modification is away from KK, the constraint wj=−1w_j=-1 on KK holds at every point. Therefore wj∈FKw_j\in\mathcal F_K, and uj≤wj≤hK,(sup⁡jwj)∗=UK.u_j\leq w_j\leq h_K,\qquad \left(\sup_jw_j\right)^*=U_K. Both increasing rows converge locally in outer capacity to UKU_K by the no-jump theorem.

The proved quasicontinuous-obstacle orthogonality gives, in local weights, 0=∫W(uj+ϵχ−wj)(Dwj)n.0=\int_W(u_j+\epsilon\chi-w_j)(Dw_j)^n . The integrand vanishes off the fixed compact support of χ\chi. Insert a fixed smooth cutoff equal to one near that support. Weighted increasing BT convergence sends the terms with weights uju_j and wjw_j to the same limit, while weak convergence handles the fixed continuous weight χ\chi. Consequently ϵ∫Wχ(DUK)n=0.\epsilon\int_W\chi(DU_K)^n=0. If the globalization uses B’s finite constant-field convention, its local calibration multiplies this zero identity by a positive constant, exactly as in the maximality proof. The conclusion is for the original KK-measure. Such WW cover V∖KV\setminus K, and all original nonnegative compact scalar tests are allowed there. Thus μK\mu_K is zero on V∖KV\setminus K. It is finite by local CLN on KK, and local maximality follows from the proved homogeneous-equation criterion.

This argument deliberately lifts the ACTUAL admissible functions uju_j. Directly modifying UKU_K and treating a quasi-everywhere obstacle inequality as a pointwise admissibility condition would not be justified.

Since UK+1U_K+1 is a normalized test on the original VV, μK(K)≤C(K)\mu_K(K)\leq C(K). For the reverse inequality take v∈Cb(V)v\in\mathcal C_b(V) with 0≤v≤10\leq v\leq1 and 0<t<10<t<1, and set w=(1−t)v−1+t/2.w=(1-t)v-1+t/2. It lies between −1+t/2-1+t/2 and −t/2-t/2. On KK outside the exceptional zero-capacity set, UK=−1<wU_K=-1<w. That exceptional Borel set has zero (Dv)n(Dv)^n-mass, because (Dv)n(E)≤C(E)≤C∗(E)(Dv)^n(E)\leq C(E)\leq C^*(E). The boundary limit UK=0U_K=0 gives the compact-gap condition for (UK,w)(U_K,w). Comparison and the established support give (1−t)n∫K(Dv)n≤∫{UK<w}(Dw)n≤∫{UK<w}μK≤μK(K).(1-t)^n\int_K(Dv)^n \leq\int_{\{U_K<w\}}(Dw)^n \leq\int_{\{U_K<w\}}\mu_K \leq\mu_K(K). Take the supremum over normalized vv and let t↓0t\downarrow0. This proves (eq:r2-compact-capacity-mass). ◻

Theorem (Compact continuity and compact outer regularity). If Kj⋐VK_j\Subset V are decreasing compact sets with K=⋂jKjK=\bigcap_jK_j, then C(Kj)↓C(K),( sup⁡jUKj )∗=UK.C(K_j)\downarrow C(K),\qquad (\,\sup_j U_{K_j}\,)^*=U_K. Moreover UKj→UKU_{K_j}\to U_K locally in outer ambient capacity and their Monge–Ampère measures converge weakly on all original compact scalar tests. Every compact K⋐VK\Subset V satisfies C∗(K)=C(K).C^*(K)=C(K). All these capacities use the same original domain VV.

Proof. The case of an empty intersection is eventually empty and is immediate by compactness. Otherwise put U=(sup⁡jUKj)∗U=(\sup_jU_{K_j})^*. Monotonicity gives U≤UKU\leq U_K. For v∈FKv\in\mathcal F_K and ϵ>0\epsilon>0, upper semicontinuity makes Oϵ={v<−1+ϵ}O_\epsilon=\{v<-1+\epsilon\} an ordinary open neighborhood of KK. Nested compactness gives Kj⊂OϵK_j\subset O_\epsilon for all sufficiently large jj. The function vϵ=max⁡(−1,v−ϵ)v_\epsilon=\max(-1,v-\epsilon) then belongs to FKj\mathcal F_{K_j}. Hence vϵ≤UKj≤Uv_\epsilon\leq U_{K_j}\leq U. Letting ϵ↓0\epsilon\downarrow0 gives v≤Uv\leq U. Taking the supremum and upper regularization gives UK≤UU_K\leq U, proving equality.

The increasing no-jump and BT theorems now give the stated capacity and measure convergence. Every measure involved is supported in the one compact set K1K_1. Testing with a compact smooth cutoff equal to one near K1K_1 therefore gives convergence of their total masses. Formula (eq:r2-compact-capacity-mass) proves compact continuity of the raw capacity.

Choose decreasing compact neighborhoods Lj⋐VL_j\Subset V of KK with intersection KK, using local compactness and metrizability. Since K⊂Int⁡LjK\subset\operatorname{Int}L_j, C(K)≤C∗(K)≤C(Int⁡Lj)≤C(Lj)⟶C(K).C(K)\leq C^*(K) \leq C(\operatorname{Int}L_j) \leq C(L_j)\longrightarrow C(K). This proves compact outer regularity without interchanging an infimum of neighborhoods with a supremum of individual Radon measures. ◻

Theorem (Open-set extremals and their total mass). Let G⋐VG\Subset V be ordinary open. Then UG=−1U_G=-1 at every point of GG, UGU_G has ordinary boundary limit zero on ∂V\partial V, and supp⁡(DUG)n⊂∂VG,∫V(DUG)n=C(G).\operatorname{supp}(DU_G)^n\subset\partial_VG, \qquad \int_V(DU_G)^n=C(G). Here ∂VG=G‾∖G\partial_VG=\overline G\setminus G is the ordinary relative boundary. In particular the mass formula is a TOTAL-mass formula; (DUG)n(G)=0(DU_G)^n(G)=0. If Km⋐GK_m\Subset G increase with G=⋃mInt⁡KmG=\bigcup_m\operatorname{Int}K_m, then UKm↓UGU_{K_m}\downarrow U_G pointwise.

Proof. The family constraint makes hG=−1h_G=-1 on the ordinary open GG, so its upper regularization is also exactly −1-1 there. The exhaustion barrier from Lemma “Basic extremal properties and a countable selection”, using G‾\overline G, gives the boundary limit zero.

Choose the indicated compact exhaustion and put U=lim⁡mUKmU=\lim_m U_{K_m}. This is a bounded decreasing limit of classical psh functions and is therefore classical. For each x∈Gx\in G, one has x∈Int⁡Kmx\in\operatorname{Int}K_m for some mm. On that interior UKm=−1U_{K_m}=-1 exactly, by the same open-set regularization observation. Thus U=−1U=-1 on GG, and U∈FGU\in\mathcal F_G. Conversely FG⊂FKm\mathcal F_G\subset\mathcal F_{K_m} for every mm, so UG≤UKmU_G\leq U_{K_m} and hence UG≤UU_G\leq U. The admissibility of UU gives the reverse inequality.

The decreasing BT theorem gives convergence of the measures. They are all supported in the fixed compact G‾\overline G. A cutoff equal to one near it gives convergence of total masses. Since the raw capacity is continuous from below, ∫V(DUG)n=lim⁡m∫V(DUKm)n=lim⁡mC(Km)=C(G).\int_V(DU_G)^n =\lim_m\int_V(DU_{K_m})^n =\lim_m C(K_m)=C(G). The limit measure is supported in G‾\overline G. Ordinary locality and the actual equality UG=−1U_G=-1 on GG make the measure zero on GG, proving the boundary support. ◻

Status and scope.

The compact and open extremal mass formulas and compact continuity are NEW_PROVED on the displayed exhaustion domains, relative to the reviewed ambient no-jump and obstacle-orthogonality theorems. The results do not identify every arbitrary ordinary domain as such a domain, do not assert a strictly psh exhaustion, and do not transfer these mass formulas to an arbitrary background SS. Compact outer regularity is proved here for the raw ambient capacity. Full Borel outer regularity and Choquet capacitability require the additional steps stated in the subsequent section.

Polar witnesses and capacitability from ambient extremals

Status: NEW_PROVED, relative to the adopted inputs and the preceding reviewed R2 theorems. The witness construction, arbitrary-set extremal mass formula, and ordinary-open proof of increasing continuity are supplied below. The Choquet and patching consequences are DIRECT_COROLLARY. The two theorem groups (N) and (E) are separated to display dependencies; both have been proved in the preceding sections. Neither a general Josefson theorem nor Borel outer regularity is an input.

Domain and the two theorem inputs

Let VV be a connected ordinary open with compact closure in a smooth affine algebraic chart over the fixed finite extension K/QpK/\mathbf Q_p. Assume that its ordinary boundary is nonempty and that there is ρ∈Cb(V)∩C(V‾),ρ<0 on V,ρ=0 on ∂V.\rho\in\mathcal C_b(V)\cap C(\overline V),\qquad \rho<0\text{ on }V,\qquad \rho=0\text{ on }\partial V. Thus each negative sublevel of ρ\rho has compact closure in VV. All capacities in this section use the ambient background [1][1] and the fixed admissibility domain VV. Write C(E)=Cap⁡[1](E;V),C∗(E)=inf⁡G⊃EG ordinary open in VC(G).C(E)=\operatorname{Cap}_{[1]}(E;V),\qquad C^*(E)=\inf_{\substack{G\supset E\\G\text{ ordinary open in }V}}C(G). The first notation is initially for Borel sets; the second is defined for arbitrary subsets. All Monge–Ampère measures and their weak limits use the original CLD scalar test functions.

For any subset A⊂VA\subset V, define FA={v∈Cb(V):−1≤v≤0, v=−1 on A},hA=sup⁡v∈FAv,UA=hA∗.\mathcal F_A=\{v\in\mathcal C_b(V):-1\leq v\leq0,\ v=-1\text{ on }A\}, \qquad h_A=\sup_{v\in\mathcal F_A}v,\qquad U_A=h_A^*. The ordinary upper regularization UAU_A is bounded classical by Theorem “Local bounded upper-envelope theorem”. The conditions v=−1v=-1 and v≤−1v\leq-1 are identical for this normalized family. The raw supremum is never substituted for its regularization without a stated argument.

The two theorem groups are recorded separately to make the dependency order explicit:

  1. For every locally uniformly upper-bounded family of bounded classical psh functions, the difference between its actual supremum and its ordinary usc regularization is contained locally in a Borel set of ambient outer capacity zero. In particular a locally uniformly bounded increasing sequence has its actual supremum equal to its regularization outside such a set, and their values agree at every ambient divisorial point. Local and fixed-domain null classes are identified only through Corollary “Equivalence of local null classes”. These are Theorem “Negligible upper-envelope sets have ambient outer capacity zero”, Theorem “Increasing envelopes have no ambient capacity jump”, and Corollary “Upper regularization preserves divisorial values”.

  2. If G⋐VG\Subset V is ordinary open, then UG=−1U_G=-1 at every point of GG, UGU_G has ordinary boundary limit zero, supp⁡(DUG)n⊂G‾\operatorname{supp}(DU_G)^n\subset\overline G, and ∫V(DUG)n=C(G).\int_V(DU_G)^n=C(G). For a compact K⋐VK\Subset V, the corresponding compact extremal identity is ∫V(DUK)n=C(K).\int_V(DU_K)^n=C(K). The open-set identity concerns total mass. It does not assert ∫G(DUG)n=C(G)\int_G(DU_G)^n=C(G): the function is constant on GG. These are Theorems “Open-set extremals and their total mass” and “The compact extremal and its equilibrium mass”, with the domain and boundary properties of Lemma “Basic extremal properties and a countable selection”.

For every fixed A⊂W⋐VA\subset W\Subset V, choose a>0a>0 sufficiently large that aρ≤−1a\rho\leq-1 on W‾\overline W. Then b=max⁡(−1,aρ)∈FA,b≤UA≤0.b=\max(-1,a\rho)\in\mathcal F_A, \qquad b\leq U_A\leq0. This gives the ordinary boundary limit zero of UAU_A, uniformly for all such AA. In particular {UA<−1+η}⋐V\{U_A<-1+\eta\}\Subset V for every 0<η<10<\eta<1, with a common compact bound for fixed WW and η\eta. The capacity of WW is finite by relative CLN on a slightly larger compact stage.

We also use the total-mass order of Lemma “Order of total Monge–Ampère masses”: if u≤v≤0u\leq v\leq0 are bounded classical functions with ordinary boundary limits zero, then ∫V(Dv)n≤∫V(Du)n\int_V(Dv)^n\leq\int_V(Du)^n, including infinite masses.

A nontrivial pole witness for an outer-capacity-zero set

Theorem (A witness on the specified exhaustion domain). Let E⊂W⋐VE\subset W\Subset V be an arbitrary subset with C∗(E)=0C^*(E)=0. For every fixed ambient divisorial point z∈Vz\in V there is a nonpositive extended classical psh function ϕ\phi on VV such that E⊂{ϕ=−∞},−1≤ϕ(z)≤0.E\subset\{\phi=-\infty\},\qquad -1\leq\phi(z)\leq0. In particular the witness is not identically −∞-\infty.

Proof. Choose an ordinary W′W' with W‾⊂W′⋐V\overline W\subset W'\Subset V. From C∗(E)=0C^*(E)=0 choose ordinary opens Oj⊃EO_j\supset E with C(Oj)<2−jC(O_j)<2^{-j} and set Gj=W′∩⋂i=1jOi,uj=UGj.G_j=W'\cap\bigcap_{i=1}^jO_i,\qquad u_j=U_{G_j}. The GjG_j decrease, while −1≤uj↑h≤0-1\leq u_j\uparrow h\leq0. They share a lower barrier b=max⁡(−1,aρ)b=\max(-1,a\rho) as above. Their measures are supported in the one compact set W‾′\overline W', and their total masses tend to zero by (E).

Put U=h∗∈Cb(V)U=h^*\in\mathcal C_b(V). By (N), h=Uh=U locally outside sets of outer capacity zero. The actual increasing-limit theorem “Local increasing convergence without a prescribed collar”, the completed calculus “BT calculus on the bounded capacitary closure”, and independence under such changes of representatives give (Duj)n⟶(DU)n(Du_j)^n\longrightarrow(DU)^n on all original compact scalar tests. Their vanishing total masses imply (DU)n=0(DU)^n=0. Also b≤U≤0b\leq U\leq0, so UU has ordinary boundary limit zero. Pointwise ambient domination “Pointwise ambient domination and uniqueness”, applied to UU and 00, now gives U=0U=0 on VV. Indeed all sets {U<−ϵ}\{U<-\epsilon\} are relatively compact by the lower barrier, and (DU)n(DU)^n vanishes on the whole strict gap.

By the divisorial part of (N), uj(z)↑0u_j(z)\uparrow0. Choose a strictly increasing sequence jlj_l such that ujl(z)>−2−lu_{j_l}(z)>-2^{-l}. The finite sums ϕm=∑l=1mujl\phi_m=\sum_{l=1}^m u_{j_l} are nonpositive classical psh and decrease with mm. Their pointwise limit ϕ\phi is extended classical psh by the classical decreasing closure, and ϕ(z)≥−∑l2−l=−1\phi(z)\geq-\sum_l2^{-l}=-1. For x∈Ex\in E, every GjG_j contains xx and the open-set extremal satisfies uj(x)=−1u_j(x)=-1 exactly. Thus ϕm(x)=−m\phi_m(x)=-m and ϕ(x)=−∞\phi(x)=-\infty. No integrability of the original set or a global pole-patching theorem has been used. ◻

Corollary (The local zero-set equivalence on this domain). On the same exhaustion domain, for every arbitrary E⊂W⋐VE\subset W\Subset V, C∗(E)=0⟺E⊂{ϕ=−∞} for some nonpositive classical ϕ on V, ϕ≢−∞.C^*(E)=0 \quad\Longleftrightarrow\quad E\subset\{\phi=-\infty\} \text{ for some nonpositive classical }\phi\text{ on }V, \ \phi\not\equiv-\infty. The implication from right to left is the already proved ambient polar theorem, not a new nonconcentration hypothesis.

Proof. The forward implication is the theorem. For the reverse implication, Corollary “Local outer-capacity zero of ambient pluripolar sets” gives local outer capacity zero on every compact/domain pair. The fixed-domain comparison “Equivalence of local null classes” gives C∗(E)=0C^*(E)=0. ◻

Definition (Local negligibility in the ordinary topology). Let EE be an arbitrary subset of an allowed ordinary open UU. It is locally negligible if every x∈Ux\in U has an ordinary neighborhood O⊂UO\subset U and a nonempty locally uniformly upper-bounded family F⊂Cb(O)\mathcal F\subset\mathcal C_b(O) such that E∩O⊂{sup⁡v∈Fv<(sup⁡v∈Fv)∗}.E\cap O\subset \left\{\sup_{v\in\mathcal F}v< \left(\sup_{v\in\mathcal F}v\right)^*\right\}. It is locally increasing-negligible if the family can be chosen as one increasing locally uniformly bounded sequence in Cb(O)\mathcal C_b(O). Local pluripolarity means that every x∈Ux\in U has a connected ordinary neighborhood O⊂UO\subset U and an extended classical psh function ϕ\phi on OO, not identically −∞-\infty there, with E∩O⊂{ϕ=−∞}E\cap O\subset\{\phi=-\infty\}. Local outer-capacity zero means that Cap⁡[1]∗(E∩H;Ω)=0\operatorname{Cap}_{[1]}^*(E\cap H;\Omega)=0 for every permitted ordinary norm-box domain Ω⊂U\Omega\subset U and every compact H⋐ΩH\Subset\Omega.

Corollary (Local polar, negligible, and outer-zero sets). On every allowed ordinary open UU, the following properties of an arbitrary subset E⊂UE\subset U are equivalent:

  1. EE is locally pluripolar;

  2. EE has local ambient outer-capacity zero;

  3. EE is locally increasing-negligible;

  4. EE is locally negligible.

These are local witness assertions. A single global psh witness on UU is not asserted.

Proof. The implication (i)⇒\Rightarrow(ii) is Corollary “Local outer-capacity zero of ambient pluripolar sets”. For (ii)⇒\Rightarrow(i), fix x∈Ux\in U and use Lemma “Local supply of exhaustion domains” in an ordinary algebraic neighborhood to choose an exhaustion domain VV containing xx. Choose a connected ordinary WW with x∈W⋐Vx\in W\Subset V and an ambient divisorial point z∈Wz\in W. The local null-class comparison gives Cap⁡[1]∗(E∩W;V)=0\operatorname{Cap}_{[1]}^*(E\cap W;V)=0. Theorem “A witness on the specified exhaustion domain” supplies a nonpositive classical witness ϕ\phi on VV, finite at zz, whose pole set contains E∩WE\cap W. Its restriction to WW is a nontrivial classical witness, as required.

For (i)⇒\Rightarrow(iii), shrink a connected witness neighborhood if necessary and subtract a finite upper bound so that ϕ≤0\phi\leq0. The pole set P={ϕ=−∞}P=\{\phi=-\infty\} has empty ordinary interior by Proposition “A nontrivial pole set has empty ordinary interior”; hence the witness remains nontrivial on every smaller connected ordinary neighborhood. The actual bounded classical functions wN=1+max⁡(ϕ,−N)Nw_N=1+\frac{\max(\phi,-N)}{N} increase to 1Pc\mathbf1_{P^c}. Because PP has empty ordinary interior, the ordinary usc regularization of this actual limit is the constant one. Its discrepancy set is exactly PP, which contains the specified part of EE. This proves increasing negligibility. The implication (iii)⇒\Rightarrow(iv) is immediate.

Finally (iv)⇒\Rightarrow(ii) follows from Theorem “Negligible upper-envelope sets have ambient outer capacity zero”. To express it on a prescribed compact/domain pair (H,Ω)(H,\Omega), restrict the local witness families to smaller permitted domains contained in Ω\Omega. Choose countably many ordinary cores covering E∩HE\cap H; such a choice exists in the locally second-countable finite pp-adic setting. Each core intersection has outer capacity zero on its own domain and therefore on Ω\Omega by domain monotonicity. Countable subadditivity gives the required zero capacity on HH. No uncountable union of exceptional sets is taken. ◻

Corollary (Pole patching on one exhaustion domain). On the same connected exhaustion domain VV, let E⊂VE\subset V have local ambient outer-capacity zero. In particular this applies to a locally pluripolar EE. Then there is one nonpositive extended classical psh function ϕ\phi on VV, not identically −∞-\infty, such that E⊂{ϕ=−∞}E\subset\{\phi=-\infty\}. No relative compactness of EE itself is required for this conclusion.

Provenance: DIRECT_COROLLARY. The domain VV and its exhaustion hypothesis are retained.

Proof. Choose ordinary Wm⋐VW_m\Subset V increasing to VV and fix one ambient divisorial point z∈Vz\in V. Put Em=E∩WmE_m=E\cap W_m. The fixed-domain null-class comparison gives C∗(Em)=0C^*(E_m)=0. The witness theorem gives nonpositive classical ϕm\phi_m on VV with Em⊂{ϕm=−∞}E_m\subset\{\phi_m=-\infty\} and ϕm(z)≥−1\phi_m(z)\geq-1. The series ϕ=∑m≥12−mϕm\phi=\sum_{m\geq1}2^{-m}\phi_m is a decreasing limit of nonpositive classical partial sums and is therefore extended classical. It satisfies ϕ(z)≥−1\phi(z)\geq-1, so it is nontrivial. Every point of EE belongs to some EmE_m, and the corresponding positive summand forces ϕ=−∞\phi=-\infty there. ◻

Remark (Why the domain qualification is essential). Untwisted pole globalization on every analytic space is FALSE. For example, the rational point 00 of PK1,an\mathbf P^{1,\mathrm{an}}_K is locally pluripolar, with local witness log⁡∣t∣\log|t|. A nontrivial global extended classical psh function on this proper connected space attains a finite maximum and is constant by the strong maximum principle [GR25, Theorem 6.1]. It therefore has no pole. The preceding corollary is globalization on the one specified exhaustion domain; it is not that false unrestricted assertion, nor a theorem about globally twisted psh functions.

An outer-capacity formula for arbitrary relatively compact sets

Theorem (The extremal total-mass formula). For every arbitrary A⊂W⋐VA\subset W\Subset V, C∗(A)=∫V(DUA)n<∞.C^*(A)=\int_V(DU_A)^n<\infty. This assertion neither assumes Borel measurability of AA nor identifies the raw capacity with its outer regularization beforehand.

Proof. If G⋐VG\Subset V is ordinary open and contains AA, then FG⊂FA\mathcal F_G\subset\mathcal F_A, hence UG≤UA≤0U_G\leq U_A\leq0. Both functions have boundary limit zero. The mass-order lemma and (E) imply ∫V(DUA)n≤∫V(DUG)n=C(G).\int_V(DU_A)^n\leq\int_V(DU_G)^n=C(G). Taking the infimum over such GG gives ∫V(DUA)n≤C∗(A)\int_V(DU_A)^n\leq C^*(A). Relatively compact opens suffice here: intersect any open cover with a fixed relatively compact open neighborhood of W‾\overline W. This also proves finiteness.

Since hA=−1h_A=-1 on AA, input (N) shows that A∩{UA>−1}⊂NA\cap\{U_A>-1\}\subset N for some set N⊂WN\subset W with C∗(N)=0C^*(N)=0. We may use a Borel outer-zero cover of this defect if AA itself is not Borel. Fix 0<η<10<\eta<1 and put Gη={UA<−1+η},vη=max⁡(−1,UA1−η).G_\eta=\{U_A<-1+\eta\},\qquad v_\eta=\max\left(-1,\frac{U_A}{1-\eta}\right). The set GηG_\eta is ordinary open and relatively compact, because UAU_A is usc and bounded below by the common exhaustion barrier. The function vηv_\eta is classical, has boundary value zero, belongs to [−1,0][-1,0], and equals −1-1 on GηG_\eta. Consequently UA1−η≤vη≤UGη.\frac{U_A}{1-\eta}\leq v_\eta\leq U_{G_\eta}. Twice applying the mass-order lemma gives C(Gη)=∫V(DUGη)n≤∫V(Dvη)n≤(1−η)−n∫V(DUA)n.C(G_\eta)=\int_V(DU_{G_\eta})^n \leq\int_V(Dv_\eta)^n \leq(1-\eta)^{-n}\int_V(DU_A)^n. Finally A⊂Gη∪NA\subset G_\eta\cup N. Monotonicity and subadditivity of outer capacity, and its equality with raw capacity on ordinary open sets, imply C∗(A)≤C(Gη)≤(1−η)−n∫V(DUA)n.C^*(A)\leq C(G_\eta) \leq(1-\eta)^{-n}\int_V(DU_A)^n. Let η\eta decrease to zero. The two inequalities prove the formula. ◻

The missing continuity axiom and the Choquet conclusion

Theorem (Continuity from below on arbitrary subsets of a core). If Aj⊂Aj+1⊂W⋐VA_j\subset A_{j+1}\subset W\Subset V are arbitrary subsets and A=⋃jAjA=\bigcup_jA_j, then C∗(A)=lim⁡jC∗(Aj).C^*(A)=\lim_j C^*(A_j).

Proof. Write Uj=UAjU_j=U_{A_j}; these bounded classical functions decrease with jj. By (N), each defect Aj∩{Uj>−1}A_j\cap\{U_j>-1\} is contained in an outer-zero set NjN_j. The countable union of the NjN_j has outer capacity zero. For ϵ>0\epsilon>0, choose one ordinary open set GG containing that union with C(G)<ϵC(G)<\epsilon.

Fix 0<η<10<\eta<1 and put Oj=G∪{Uj<−1+η}.O_j=G\cup\{U_j<-1+\eta\}. These are increasing ordinary open sets, since the UjU_j decrease. Each OjO_j contains AjA_j. The truncation estimate in the preceding proof and its mass formula give C(Oj)≤ϵ+(1−η)−nC∗(Aj).C(O_j)\leq\epsilon+ (1-\eta)^{-n}C^*(A_j). The raw capacity is continuous from below on Borel sets, directly from its definition as a supremum of Radon measures. Applying this only to the increasing open sets OjO_j yields C∗(A)≤C(⋃jOj)=lim⁡jC(Oj)≤ϵ+(1−η)−nlim⁡jC∗(Aj).C^*(A)\leq C\Bigl(\bigcup_jO_j\Bigr) =\lim_j C(O_j) \leq\epsilon+(1-\eta)^{-n}\lim_j C^*(A_j). Let ϵ\epsilon and then η\eta decrease to zero. Monotonicity supplies the reverse inequality.

This is the ordinary-open argument of the classical capacitability proof, [BT82, §8, Theorem 8.2], with its analytic hypotheses proved separately here. An infinite intersection of open covers is not declared open, and no infimum/supremum interchange is made. ◻

Corollary (Local Choquet capacitability and Borel outer regularity). Fix an ordinary core W⋐VW\Subset V. The set function A⟼C∗(A),A⊂W,A\longmapsto C^*(A),\qquad A\subset W, is a finite monotone countably subadditive Choquet capacity: it is continuous from below on arbitrary increasing sets and from above on decreasing compact sets. Every Suslin subset A⊂WA\subset W (analytic in the descriptive-set-theoretic sense) is capacitable: C∗(A)=sup⁡K⊂AK compactC(K).C^*(A)=\sup_{\substack{K\subset A\\K\text{ compact}}}C(K). In particular every Borel A⊂WA\subset W satisfies C(A)=C∗(A).C(A)=C^*(A).

Proof. Monotonicity and countable subadditivity are already proved for outer capacity. Compact continuity from above is Lemma “Elementary outer-capacity facts”; increasing continuity is the preceding theorem. Finiteness follows by covering W‾\overline W with one relatively compact ordinary open and applying CLN to its closure. The core WW is a locally compact second-countable metrizable space in the fixed finite pp-adic algebraic setting. The Choquet capacitability theorem, in the form recalled in [BT82, §8, following Theorem 8.2], therefore gives C∗(A)=sup⁡K⊂A compactC∗(K)C^*(A)=\sup_{K\subset A\text{ compact}}C^*(K) for such Suslin AA. This terminology concerns analytic sets of the Polish space WW, not analytic subspaces over the base field KK. For compact KK, the mass formula of the preceding theorem and the compact part of (E) give C∗(K)=C(K)C^*(K)=C(K).

For Borel AA, raw inner regularity is elementary: C(A)=sup⁡vsup⁡K⊂A compact∫K(Dv)n=sup⁡K⊂A compactC(K).C(A)=\sup_{v}\sup_{K\subset A\text{ compact}} \int_K(Dv)^n =\sup_{K\subset A\text{ compact}}C(K). Only two suprema are interchanged in this equality. Combining it with capacitability gives C(A)=C∗(A)C(A)=C^*(A). ◻

Scope and dependency order.

The capacity-zero witness uses only the open extremal identity, ambient no-jump, the proved increasing calculus, and pointwise ambient domination. The arbitrary-set mass formula and Choquet continuity do not use that witness theorem: their only null-set input is (N). Thus the argument does not prove negligible sets polar by first assuming that conclusion. The capacitability statement is on each fixed relatively compact core of this explicitly qualified domain. It does not assert a global Josefson theorem on an arbitrary analytic space, or discard the negative-exhaustion-domain hypothesis.

A distinction in abstract capacity theory.

For a Borel capacity cc, its Borel-envelope extension inf⁡{c(B):A⊂B, B Borel}\inf\{c(B): A\subset B,\ B\text{ Borel}\} is different from the ordinary-open extension used here. The familiar increasing-set argument using ⋂k≥jGk\bigcap_{k\geq j}G_k yields Borel sets, not open ones. Raw Borel inner regularity and raw Borel continuity from below therefore do not, by that argument, supply the missing continuity axiom for the ordinary-open extension. Theorem “Continuity from below on arbitrary subsets of a core” supplies the required axiom by actual ordinary sublevel sets of extremal functions.

Second-round theorem and status increment

This section updates, rather than replaces, the source-by-source comparison in Section “Theorem-by-theorem comparison with the classical theory”. The first-round statements remain frozen with their original hypotheses. An earlier OPEN entry is closed only by the particular second-round theorem cited below. No claim of literature priority is encoded by NEW_PROVED: that label records a proof supplied in this project, relative to its accepted inputs.

Two independent status axes and the fixed category

The labels EXTERNAL_KNOWN, INTERNAL_INPUT, DIRECT_COROLLARY, and NEW_PROVED record provenance. The labels PROVED, CONDITIONAL, OPEN, and FALSE assess a precisely qualified proposed analogy. A theorem proved under an explicit additional geometric hypothesis is PROVED with that hypothesis; the corresponding general-domain analogy is CONDITIONAL or OPEN. A counterexample to an unrestricted assertion does not change the status of its qualified replacement.

All local assertions retain the finite extension K/QpK/\mathbf Q_p, its prescribed absolute value, and the allowed good, Hausdorff, boundaryless, quasi-smooth pure-dimensional analytic spaces. Unless a narrower theorem states otherwise, SS is an arbitrary fixed symmetric current of bidegree (p,p)(p,p), positive in the original CLD cone and separately d′d'- and d′′d''-closed. Products and current equalities are tested on all original compactly supported smooth complementary CLD forms. Neither a change of positivity cone nor an enlargement of the test sheaf is implicit. Put q=n−pq=n-p; the relative top-capacity statements below use q≥1q\geq1.

R and B remain INTERNAL_INPUT. In particular, the bounded classical class and its [1][1]-background calculus are not second-round discoveries. The completed geometric adelic nefness theorem A2 also remains an INTERNAL_INPUT under its own number-field, integrability, and boundary hypotheses. The original classical, CLD, GR, and GK theorems recorded in Section “Theorem-by-theorem comparison with the classical theory” remain EXTERNAL_KNOWN in their stated categories.

Increment for the eleven Bedford–Taylor directions

Target

Qualified second-round increment

Remaining qualification or obstruction

1

Locally bounded psh functions

PROVED / DIRECT_COROLLARY: upper regularizations of locally uniformly upper-bounded classical families are classical, by Theorem “Local bounded upper-envelope theorem”; the projective bounded usc obstacle envelope is Proposition “Envelopes of bounded upper-semicontinuous obstacles”. PROVED / DIRECT_COROLLARY: original trace-current order determines the actual bounded classical representative, Proposition “Actual classical representatives are determined by their original currents”.

The new capacitary class is relative to SS and retains Borel representatives. For S=[1]S=[1], every such class now has a unique actual classical representative modulo local outer-capacity zero, Corollary “The unique classical representative” (DIRECT_COROLLARY). Actual Borel values on the exceptional set are not identified. No corresponding assertion for arbitrary SS is made.

2

Positive closed currents and mixed products

The arbitrary-fixed-SS bounded classical calculus is already PROVED in R1. Theorem “BT calculus on the bounded capacitary closure” extends it to AS,b\mathcal A_{S,b}, with well-defined Borel recursion, quasi-everywhere independence, symmetry, and restriction/gluing: PROVED / NEW_PROVED.

This does not define the wedge of two arbitrary positive closed currents. Nor does it settle dependence only on the currents D[ui]D[u_i]: the general discontinuous potential-quotient assertion remains OPEN.

3

Chern–Levine–Nirenberg estimates

PROVED / NEW_PROVED: the weighted L1L^1 estimate and uniform truncation estimate of Theorem “Uniform truncation estimate over an arbitrary background”, under the specified collar-trace integrability condition. The deep-sublevel outer-capacity estimate and integrable-witness consequences are DIRECT_COROLLARY, Corollaries “Uniformity over normalized potentials” and “Outer capacity and quasicontinuity of an integrable witness”.

The integrability hypothesis concerns an explicit trace of the fixed SS; it is not automatic for arbitrary SS. Uniform exponential tails and unbounded-weight stability from a uniform first moment alone are FALSE, Propositions “No uniform exponential improvement over arbitrary backgrounds” and “A uniform first moment does not give uniform integrability”.

4

Symmetry and weighted monotone convergence

R1 already proves symmetry and weighted joint decreasing continuity for actual bounded classical limits and fixed SS. PROVED / NEW_PROVED: bounded increasing sequences converge in outer relative capacity to their actual Borel supremum, Theorem “Increasing sequences are Cauchy in outer capacity” and Corollary “Local increasing convergence without a prescribed collar”; weighted increasing continuity in the completed class is Corollary “Weighted increasing limits with actual representatives”. The ambient no-jump theorem “Increasing envelopes have no ambient capacity jump” identifies this supremum with its usc regularization locally quasi-everywhere: PROVED / NEW_PROVED. Theorem “Quasicontinuous multipliers through weak closure” proves the qualified quasicontinuous multiplier passage through weak closure.

Ambient no-jump and its common-collar energy equality are closed questions in this record. Replacing the actual supremum by its ambient usc regularization remains FALSE without qualification for arbitrary SS. Unrestricted varying-SS weighted stability remains FALSE.

5

Locality and maximum principles

R1 Borel maximum locality and ordinary restriction/locality remain PROVED. Theorem “Positive closed restriction to a complete pole set” proves that 1PS\mathbf1_P S and 1PcS\mathbf1_{P^c}S are separately closed for a classical pole set PP, and that BT products commute with this restriction: PROVED / NEW_PROVED. Theorem “Maximality and the homogeneous equation” proves local classical maximality equivalent to (Du)n=0(Du)^n=0 for the ambient background.

Strict inequality sets of discontinuous potentials are Borel and locally quasi-open, not automatically ordinary open. The polar restriction theorem starts with a current defined on the entire domain; it is not a general extension theorem for a current defined only off the pole set.

6

Bedford–Taylor capacities

PROVED / DIRECT_COROLLARY: outer regularity on quasi-open sets, Proposition “Quasi-open sets are outer regular”, and the elementary outer-capacity properties in Lemma “Elementary outer-capacity facts”. PROVED / NEW_PROVED: two-sided local comparison with one fixed relatively compact affine capacity domain, Theorem “Uniform local comparison with a fixed domain”. For the ambient capacity on an exhaustion domain, compact raw outer regularity and compact continuity are Theorem “Compact continuity and compact outer regularity”; outer-capacity continuity for arbitrary increasing subsets of a fixed relatively compact core is Theorem “Continuity from below on arbitrary subsets of a core”. Choquet capacitability and Borel outer regularity on that core are Corollary “Local Choquet capacitability and Borel outer regularity”.

Raw compact outer regularity and the corresponding Choquet theory for arbitrary relative backgrounds SS remain OPEN. The ambient result retains the exhaustion-domain and fixed-core scope. General-domain nondegeneracy remains FALSE; the R1 proper-space example still applies.

7

Quasicontinuity and capacity convergence

R1 bounded quasicontinuity is retained. Theorems “Increasing sequences are Cauchy in outer capacity” and “BT calculus on the bounded capacitary closure” give the increasing and capacitary-completion extensions. Corollary “Quasicontinuity of extended classical psh functions” gives local finite quasicontinuity and lower-truncation outer-capacity convergence for every nontrivial extended classical psh function: PROVED / DIRECT_COROLLARY of the new polar-trace and weighted estimates.

Capacity precompactness of the full normalized continuous psh family, and weighted BT stability under bounded pointwise convergence alone, are FALSE, Proposition “Failure of capacity precompactness and pointwise weighted stability”. This counterexample does not refute compact-set domination by the weak closure of the admissible measure family.

8

Pluripolar and negligible sets

PROVED / NEW_PROVED: every original smooth ambient trace vanishes on a nontrivial classical pole set, Theorem “Vanishing of ambient smooth traces on a classical pole set”. DIRECT_COROLLARY: all ambient bounded BT coefficients avoid such sets, and every locally pluripolar subset has local outer capacity zero, Corollaries “Ambient bounded BT products avoid classical pole sets” and “Local outer-capacity zero of ambient pluripolar sets”. Theorem “A witness on the specified exhaustion domain” constructs a nontrivial witness for an outer-zero set. The resulting local equivalence of pluripolar, outer-zero, and negligible sets is Corollary “Local polar, negligible, and outer-zero sets”. Pole patching on one exhaustion domain is Corollary “Pole patching on one exhaustion domain”.

These local ambient equivalences are PROVED. Unqualified untwisted pole globalization on every analytic space is FALSE, already on PK1,an\mathbf P^{1,\mathrm{an}}_K. Arbitrary relative backgrounds may charge ambient pluripolar sets; their mass is governed by Theorem “Positive closed restriction to a complete pole set”, not forced to vanish.

9

Comparison and domination

The R1 compact-gap comparison theorem for arbitrary fixed SS remains PROVED. PROVED / DIRECT_COROLLARY: ambient strict-gap detection and pointwise domination/uniqueness on the stated relatively compact framed affine domains, Theorem “Strict-gap detection for the ambient capacity” and Corollary “Pointwise ambient domination and uniqueness”. Relative atomic and smooth-factor detection criteria and explicit faithful backgrounds are proved in Propositions “Pointwise conclusions under atomic detection” and “A sufficient condition using smooth factors”.

Pointwise domination for arbitrary SS remains CONDITIONAL on detection; the unrestricted version is FALSE even for nonzero SS on a connected domain, Example “A nonzero connected-domain counterexample”. Capacity-modulo uniqueness and actual pointwise uniqueness are distinct assertions.

10

Relative extremal functions

PROVED / NEW_PROVED: least upper bounds in the local SS-quasi-everywhere order, represented in AS,b\mathcal A_{S,b}, Theorem “Local relative capacitary envelope”. Ambient bounded usc obstacle envelopes follow directly from B; orthogonality for bounded quasicontinuous usc projective obstacles is Theorem “Orthogonality for bounded quasicontinuous usc obstacles”, PROVED / NEW_PROVED. For the ambient background on an exhaustion domain, compact/open extremal equilibrium formulas are Theorems “The compact extremal and its equilibrium mass” and “Open-set extremals and their total mass”; arbitrary relatively compact sets satisfy the outer-capacity total-mass formula “The extremal total-mass formula”.

The ambient and relative categories remain distinct. General relative equilibrium-measure/contact-set formulas and their capacity identities remain OPEN; unrestricted replacement of an SS-capacitary envelope by the ambient envelope is FALSE, Proposition “Ambient negligible sets can have positive relative capacity”.

11

Monge–Ampère and Dirichlet theory

PROVED / NEW_PROVED: the ambient local equivalence between maximality and the homogeneous equation, Theorem “Maximality and the homogeneous equation”. The explicit peak-barrier theorem “Homogeneous Dirichlet problem under peak barriers” gives a bounded classical solution of (DHg)n=0(DH_g)^n=0, ordinary boundary values, uniqueness, and the boundary sup-norm stability estimate under Definition “Continuous psh peak barriers”. Corollary “Constant boundary data and one divisorial atom” gives an explicit continuous solution and uniqueness for a single positive divisorial atom and constant boundary data on the domains constructed with its barrier: PROVED / DIRECT_COROLLARY.

As a general-domain analogy the Dirichlet conclusion is CONDITIONAL on those explicit barriers. Their general geometric construction, interior continuity for the Perron solution, and existence for a general prescribed nonzero measure remain OPEN. Bounded negative exhaustion domains are explicitly defined and locally supplied; they are not silently assumed to have peak barriers at every boundary point.

Scope distinctions that survive the second round
Fixed backgrounds and controlled varying backgrounds.

The primary calculus keeps SS fixed. The order-dominated replacement in Proposition “Order-dominated varying backgrounds” allows 0≤Sj≤CR00\leq S_j\leq C R_0, Sj→SS_j\to S, with one fixed positive closed R0R_0 and a fixed differentiated tuple; its multiplier hypotheses must also be retained. The second-round weak-closure multiplier theorem instead requires uniform coefficient bounds by a fixed local capacity and quasicontinuous weights. Neither statement removes the counterexample in Corollary “Failure for varying backgrounds”.

Actual representatives and ambient upper regularization.

An actual increasing supremum is now an element of the bounded capacitary closure, and its BT products are established there. For the ambient background, Theorem “Increasing envelopes have no ambient capacity jump” identifies it with its ordinary usc regularization locally outside outer-zero sets and proves the common-collar energy no-jump equality. More generally, Corollary “The unique classical representative” assigns each a∈A[1],ba\in\mathcal A_{[1],b} a unique classical representative rar_a modulo such sets. For an arbitrary Borel representative aa, rar_a need not equal a∗a^*; arbitrary modifications on polar sets are allowed in that capacitary class. This distinction does not affect the actual increasing-supremum theorem. Theorem “Local relative capacitary envelope” obtains a least upper bound without taking an uncountable union of exceptional sets: each competitor has its own quasi-everywhere comparison, and a countable energy construction determines the resulting class. For an arbitrary relative background SS, the capacitary representative cannot in general be replaced by the ambient usc envelope.

Raw capacity and outer capacity.

For arbitrary SS, the identity C∗(E)=C(E)C^*(E)=C(E) is proved for quasi-open EE; full raw compact outer regularity remains open. For S=[1]S=[1] on an exhaustion domain, it is now proved for compact sets and for every Borel subset of each fixed relatively compact core. The ordinary-open proof of Theorem “Continuity from below on arbitrary subsets of a core” supplies the additional continuity axiom needed for the local Choquet theorem; compact continuity of C∗C^* alone would not supply that proof. The raw-capacity closure criterion in Proposition “Weak closure and compact outer regularity” is therefore resolved for the stated ambient compact stages but remains an exact target for arbitrary relative SS. The explicit potential-noncompactness example has an admissible unweighted weak limit and does not refute compact-set domination by the admissible measure family’s weak closure.

Polar sets and arbitrary support.

The general ambient pole-trace problem is now PROVED; it is no longer a missing hypothesis in the ambient nonconcentration result. The proof uses the absence of divisorial poles, whole-box model approximation, and the capacity-controlled multiplier passage, as recorded in Theorem “Vanishing of ambient smooth traces on a classical pole set”. For arbitrary SS, the exact statement is 1PTS,k=T1PS,k\mathbf1_P T_{S,k}=T_{\mathbf1_PS,k} from Theorem “Positive closed restriction to a complete pole set”. In particular, a background already carried by PP is not made nonpolar by applying bounded BT operators. Ambient outer-zero sets now admit local nontrivial pole witnesses, and the local pluripolar/negligible/outer-zero classes agree by Corollary “Local polar, negligible, and outer-zero sets”. There is also pole patching on a single connected exhaustion domain. A global untwisted version for all analytic spaces is false on proper connected spaces; these local and qualified global statements are not conflated.

Potential quotients.

Proposition “Actual classical representatives are determined by their original currents” concerns [u][u] and [v][v], not merely D[u]D[u] and D[v]D[v]. General independence under replacing a differentiated potential by another actual bounded classical potential with the same DD-current remains OPEN. The continuous harmonic-difference result “Continuous harmonic multipliers”, the ambient-mixed and smooth background cases “Ambient mixed backgrounds” and “Smooth closed backgrounds”, and the dominated-closure case “Passage through a dominated weak closure” remain proved partial answers. None is silently promoted to the general assertion.

Comparison with other differential calculi.

Agreement on ordinary CLD tests with compatible model and Chambert–Loir measures is retained where the accepted R/B comparison applies. Equality on every δ\delta-test form is a different statement. The report does not extend arbitrary original currents to an enlarged CLD25 test sheaf, or claim the nonpositive, nonclosed, or nondiagonal parts of the GK calculus. The completed A2 adelic theorem is an input to subsequent geometric applications, not an unresolved item in this local analytic program.

Three minimal targets for the next round

The former ambient no-jump and local polar-witness targets are now proved. The following remaining targets are not assumed anywhere in the second-round arguments.

  1. Compact domination for the raw capacity’s measure closure for arbitrary relative backgrounds. For an arbitrary allowed fixed SS, on one fixed permitted capacity domain and compact test stage, prove or refute ν(K)≤Cap⁡S(K;Ω)for every compact K\nu(K)\leq\operatorname{Cap}_S(K;\Omega) \quad\text{for every compact }K for every weak limit ν\nu of normalized top BT measures (with the localization conventions of Proposition “Weak closure and compact outer regularity”). It is enough to resolve this precise closure statement; a capacity-convergent subsequence of all normalized potentials is not available. A counterexample must violate the displayed inequality, rather than only potential precompactness or weighted pointwise continuity. The ambient result on exhaustion domains is already closed and is not assigned again as new work.

  2. The general potential-quotient lemma. Given actual u,v∈Cb(U)u,v\in\mathcal C_b(U) with D[u]=D[v]D[u]=D[v] as original currents, establish or refute D((u−v)S)=0D\bigl((u-v)S\bigr)=0 for every allowed fixed positive closed background SS. The continuous-difference and specified-background cases are already proved. Equality of the original currents [u][u] and [v][v], and the unique ambient capacitary representative, do not by themselves settle this statement about equality of their DD-currents.

  3. Nonzero data on one prescribed domain. Starting with an explicitly qualified domain and boundary barriers, construct and control solutions for a sum of at least two prescribed positive divisorial atoms on that same domain. The single-atom result “Constant boundary data and one divisorial atom” constructs its domain together with the atom and barrier. It does not provide this next fixed-domain step; adding its potentials produces mixed terms. A passage from finite atomic data to more general measures would then require uniform potential bounds, boundary control, and a proved capacity stability estimate for that family of solutions.

The homogeneous peak-barrier theorem yields bounded classical solutions with ordinary boundary limits; interior continuity has not been proved there. General prescribed nonzero measures and general ordinary domains still require additional geometric and analytic hypotheses. The relative equilibrium/contact problem and comparison on enlarged δ\delta-test categories remain separate subsequent tasks.

Excluded routes and the conclusions they actually exclude
  1. FALSE: arbitrary varying-background weighted stability, even at order zero; see Corollary “Failure for varying backgrounds”. Fixed-background monotonicity and the qualified order-dominated replacement survive.

  2. FALSE: relative nonconcentration on every ambient pluripolar set, or universal relative negligibility of the discrepancy between an actual supremum and its ambient usc envelope; see Proposition “Concentration on a smooth divisor” and Proposition “Ambient negligible sets can have positive relative capacity”. The valid replacement is support restriction and the SS-capacitary envelope.

  3. FALSE: pointwise domination solely from relative measure inequalities for an arbitrary nonzero SS on a connected domain; see Example “A nonzero connected-domain counterexample”. Relative gap detection is a real additional condition.

  4. FALSE: nondegeneracy of the untwisted capacity on every analytic domain, including proper ones; see Proposition “Compact-domain degeneration”. Local framed domains and projective metric capacities must be stated separately.

  5. FALSE: compactness in capacity of the whole normalized continuous psh family, or weighted BT continuity under bounded pointwise convergence; see Proposition “Failure of capacity precompactness and pointwise weighted stability”. OPEN: the compact raw-capacity weak-closure inequality for arbitrary relative SS. The latter is not contradicted by this example; the qualified ambient compact inequality is now proved.

  6. FALSE: a uniform first-moment estimate implies uniform integrability of a fixed unbounded multiplier, or universal exponential tail control for arbitrary SS; see Propositions “A uniform first moment does not give uniform integrability” and “No uniform exponential improvement over arbitrary backgrounds”. The proved weighted estimate is used with its actual trace-integrability and truncation conclusions.

  7. FAILED_ATTEMPT: deriving raw outer regularity by interchanging an infimum over ordinary neighborhoods with a supremum over a measure family. Individual Radon outer regularity does not justify that interchange. For arbitrary SS, the quasi-open argument and weak-closure characterization remain the available replacements. For the ambient exhaustion-domain capacity, the extremal mass formula and increasing ordinary sublevels supply the additional proof.

  8. FALSE: untwisted pole globalization on every analytic space. The rational point of PK1,an\mathbf P^{1,\mathrm{an}}_K in Section “Polar witnesses and capacitability from ambient extremals” is locally pluripolar, while nontrivial global classical psh functions there are constant. Corollary “Pole patching on one exhaustion domain” proves the qualified replacement on one connected exhaustion domain.

  9. FAILED_ATTEMPT: inferring smooth-trace nullity merely from density of divisorial points, or passing an indicator through weak convergence without coefficient control. The successful polar proof additionally uses quasicontinuity of the actual indicator and uniform same-domain capacity estimates.

  10. FAILED_ATTEMPT: swapping two discontinuous moving limits by formal integration by parts, using the desired joint continuity to prove its own energy estimate, or treating an uncountable union of quasi-everywhere exceptional sets as negligible. The established R1 trace argument and R2 countable energy-envelope construction avoid these inferences.

The second-round record therefore closes several analytic gaps while retaining explicit frontier statements. It does not label the full non-Archimedean Bedford–Taylor program complete.

Mixed energy controls capacity in every remaining degree

The fixed background in this section is arbitrary: SS is symmetric, original-positive, of bidegree (p,p)(p,p), and separately closed, and q=n−p≥1q=n-p\geq1. The original field, ordinary topology, and all original CLD tests are retained. The estimates resolve the varying-tester energy obstruction mentioned in Section “Relative compact regularity: a linear-degree reduction and a contact obstruction”. They do not identify nonlinear weak limits of the original unaveraged measures.

Expressions such as (Da)i(Db)jS(Da)^i(Db)^jS abbreviate the already constructed mixed products over the fixed SS. An energy expression always means ΓQ(f,g)\Gamma_Q(f,g) for the specified mixed background QQ of bidegree (n−1,n−1)(n-1,n-1). It never means a wedge of an arbitrary energy current with another current.

A common-collar energy distance

Let Ω\Omega be a sufficiently small ordinary norm-box domain and L⋐ΩL\Subset\Omega a compact set in a strictly smaller norm box. Fix a smooth classical psh function hh with semipositive original curvature form. Consider u,v∈Cb(Ω)u,v\in\mathcal C_b(\Omega) with u=v=hon Ω∖L.u=v=h\quad\hbox{on }\Omega\setminus L. There is one fixed bound for u,v,hu,v,h near LL. Put f=u−vf=u-v and

IS(u,v)=∫f((Dv)qS−(Du)qS).I_S(u,v)=\int f\bigl((Dv)^qS-(Du)^qS\bigr).

This integral is finite and has compact support.

Lemma (The positive energy identity). For the actual point values of the potentials one has IS(u,v)=∑a=0q−1∫Γ(Du)a(Dv)q−1−aS(f)≥0.I_S(u,v)=\sum_{a=0}^{q-1} \int\Gamma_{(Du)^a(Dv)^{q-1-a}S}(f)\geq0.

Proof. Symmetry and multilinearity of the proved mixed calculus give (Du)qS−(Dv)qS=D(f∑a=0q−1(Du)a(Dv)q−1−aS).(Du)^qS-(Dv)^qS =D\left(f\sum_{a=0}^{q-1}(Du)^a(Dv)^{q-1-a}S\right). Apply −∫fD(fQ)=∫ΓQ(f)-\int fD(fQ)=\int\Gamma_Q(f) to each positive separately closed summand. No weakly varying arbitrary background is used. ◻

Theorem (An all-degree mixed-energy capacity estimate). Under these common-collar assumptions, for every δ>0\delta>0,

Cap⁡S∗({∣u−v∣>δ};Ω)≤Cδ−2IS(u,v) 2−q.\operatorname{Cap}_S^*(\{|u-v|>\delta\};\Omega) \leq C\delta^{-2} I_S(u,v)^{\,2^{-q}}.

The finite constant depends on the fixed compact collar, the common bound for the potentials, qq, and finitely many positive coefficient evaluations of SS. It is independent of the normalized capacity tester and all regularization indices. Restriction to a smaller compact core gives the corresponding cutoff version.

Proof. Fix w∈Cb(Ω)w\in\mathcal C_b(\Omega) with 0≤w≤10\leq w\leq1. For 0≤r≤q−10\leq r\leq q-1 put Er(w)=∫Γ(Dw)r(D(u+v))q−1−rS(f).E_r(w)=\int\Gamma_{(Dw)^r(D(u+v))^{q-1-r}S}(f). These are nonnegative integrals on the same compact support, since f=0f=0 off LL. Expansion and the positive energy identity give

E0(w)≤2q−1IS(u,v).E_0(w)\leq2^{q-1}I_S(u,v).

For r≥1r\geq1 write Q=(Dw)r−1(D(u+v))q−1−rSQ=(Dw)^{r-1}(D(u+v))^{q-1-r}S, of bidegree (n−2,n−2)(n-2,n-2). First suppose that the potentials are smooth. Stokes for the compactly supported first-order expression gives

Er(w)=∫d′f∧d′′w∧(Du−Dv)∧Q.E_r(w)=\int d'f\wedge d''w\wedge(Du-Dv)\wedge Q.

For the sign, apply d′d' to d′f∧d′′f∧d′′w∧Qd'f\wedge d''f\wedge d''w\wedge Q; the result is −d′f∧Df∧d′′w∧Q+d′f∧d′′f∧Dw∧Q-d'f\wedge Df\wedge d''w\wedge Q+ d'f\wedge d''f\wedge Dw\wedge Q.

Choose a fixed nonnegative compact smooth cutoff κ\kappa equal to one near LL. Symmetry and Cauchy–Schwarz, separately for the positive backgrounds Du QDu\,Q and Dv QDv\,Q, imply

Er(w)≤(∫ΓDu Q(f))1/2(∫κ ΓDu Q(w))1/2+(∫ΓDv Q(f))1/2(∫κ ΓDv Q(w))1/2.\begin{aligned} E_r(w) &\leq\left(\int\Gamma_{Du\,Q}(f)\right)^{1/2} \left(\int\kappa\,\Gamma_{Du\,Q}(w)\right)^{1/2}\\ &\quad+\left(\int\Gamma_{Dv\,Q}(f)\right)^{1/2} \left(\int\kappa\,\Gamma_{Dv\,Q}(w)\right)^{1/2}. \end{aligned}

The first factors squared sum to Er−1(w)E_{r-1}(w). The second factors squared have a uniform finite sum: for either R=Du QR=Du\,Q or R=Dv QR=Dv\,Q the cutoff identity is ∫κ ΓR(w)=12⟨R,w2Dκ⟩−∫κwD(wR).\int\kappa\,\Gamma_R(w) =\tfrac12\langle R,w^2D\kappa\rangle-\int\kappa wD(wR). The subtracted term is nonnegative. The relative finite-test CLN estimate bounds the first term for the fixed form DκD\kappa. All slots of RR are bounded copies of u,v,wu,v,w or their sum. Thus, for a fixed ArA_r independent of ww,

Er(w)≤ArEr−1(w)1/2.E_r(w)\leq A_r E_{r-1}(w)^{1/2}.

For q=1q=1 this recurrence is empty.

The compact Green identity for f2f^2 now gives ∫f2(Dw)qS=−2∫f Γ(Dw)q−1S(f,w).\int f^2(Dw)^qS =-2\int f\,\Gamma_{(Dw)^{q-1}S}(f,w). Cauchy–Schwarz with the same compact cutoff and another uniform CLN bound yield

∫f2(Dw)qS≤AEq−1(w)1/2.\int f^2(Dw)^qS\leq A E_{q-1}(w)^{1/2}.

The common bound for ff is absorbed in AA. Iteration proves

∫f2(Dw)qS≤CIS(u,v) 2−q.\int f^2(Dw)^qS\leq C I_S(u,v)^{\,2^{-q}}.

For bounded potentials, use the protected prescribed-box construction for simultaneous smooth decreasing rows. If necessary, first modify u,vu,v outside a neighborhood of LL to a common auxiliary psh collar, preserving them up to the same constant near LL and preserving their difference ff everywhere. Their rows then have an exact common smooth collar outside a fixed larger compact set L′⋐ΩL'\Subset\Omega. This does not assert that arbitrary smoothing preserves the originally prescribed hh. All limiting weighted and energy expressions in the argument are supported in LL, so the auxiliary modification leaves them unchanged. Choose all cutoffs for the larger fixed stage. Take a simultaneous smooth decreasing row for ww there. After discarding a prefix the latter lies in [0,2][0,2], so all constants remain uniform. Every background in the preceding formulas is an explicit mixed product over the same SS. Expand each energy as ΓR(a,b)=12{D(abR)−aD(bR)−bD(aR)}.\Gamma_R(a,b)=\tfrac12\{D(abR)-aD(bR)-bD(aR)\}. Its scalar factors are polynomials in finitely many decreasing psh rows. The functional-weighted joint monotone theorem passes every term to the asserted bounded-potential expression on a fixed cutoff. The differences retain fixed compact support. This proves the identities and inequalities without using capacity continuity to justify the energy calculation.

Chebyshev and the supremum over ww give the asserted raw capacity bound. The strict gap set is relatively quasi-open by quasicontinuity of u,vu,v. Proposition “Quasi-open sets are outer regular” therefore identifies its raw and outer capacities. No such set was assumed ordinarily open. ◻

Concavity selects simultaneous convex blocks

For the fixed collar define ES,h(u)=1q+1∑j=0q∫(u−h)(Du)j(Dh)q−jS.\mathcal E_{S,h}(u) =\frac1{q+1}\sum_{j=0}^q \int(u-h)(Du)^j(Dh)^{q-j}S. All nonzero weighted integrands lie on the fixed compact core. No nondegeneracy assumption on DhDh is imposed.

Lemma (Quantitative midpoint concavity). The energy is concave and

ES,h ⁣(u+v2)−ES,h(u)+ES,h(v)2≥cqIS(u,v),cq=q4q+1.\mathcal E_{S,h}\!\left(\frac{u+v}{2}\right) -\frac{\mathcal E_{S,h}(u)+\mathcal E_{S,h}(v)}2 \geq c_q I_S(u,v),\qquad c_q=\frac q{4^{q+1}}.

For ut=(1−t)u+tvu_t=(1-t)u+tv, ddtES,h(ut)=∫(v−u)(Dut)qS,d2dt2ES,h(ut)=−q∫Γ(Dut)q−1S(v−u).\frac{d}{dt}\mathcal E_{S,h}(u_t)=\int(v-u)(Du_t)^qS, \qquad \frac{d^2}{dt^2}\mathcal E_{S,h}(u_t) =-q\int\Gamma_{(Du_t)^{q-1}S}(v-u).

Proof. The energy is a polynomial in tt. In its derivative, compact Green exchange replaces ∫(ut−h)D(v−u)(Dut)j−1(Dh)q−jS\int(u_t-h)D(v-u)(Du_t)^{j-1}(Dh)^{q-j}S by ∫(v−u)(Dut−Dh)(Dut)j−1(Dh)q−jS\int(v-u)(Du_t-Dh)(Du_t)^{j-1}(Dh)^{q-j}S. The finite sum telescopes to the first derivative. Differentiating again and using compact Green gives the second derivative and concavity. These are identities in the established bounded mixed calculus; smooth common-collar rows give the same proof.

The one-variable midpoint Green formula expresses the gap as q2∫01min⁡(t,1−t)(∫Γ(Dut)q−1S(u−v))dt.\frac q2\int_0^1\min(t,1-t) \left(\int\Gamma_{(Du_t)^{q-1}S}(u-v)\right)dt. On [1/4,3/4][1/4,3/4], expansion into positive mixed energy measures gives ∫Γ(Dut)q−1S(u−v)≥4−(q−1)∫Γ(D(u+v))q−1S(u−v)≥4−(q−1)IS(u,v).\int\Gamma_{(Du_t)^{q-1}S}(u-v) \geq4^{-(q-1)}\int\Gamma_{(D(u+v))^{q-1}S}(u-v) \geq4^{-(q-1)}I_S(u,v). The weight is at least 1/41/4 on this interval of length 1/21/2. This yields the displayed constant. ◻

Corollary (An energy criterion for capacity convergence). For common-collar u,vu,v put JS,h(u,v)=ES,h(u)−ES,h(v)+∫(v−u)(Du)qS.J_{S,h}(u,v)=\mathcal E_{S,h}(u)-\mathcal E_{S,h}(v) +\int(v-u)(Du)^qS. Then IS(u,v)q+1≤JS,h(u,v)≤qIS(u,v)q+1.\frac{I_S(u,v)}{q+1} \leq J_{S,h}(u,v) \leq\frac{qI_S(u,v)}{q+1}. For a uniformly bounded common-collar sequence uju_j and uu, the following conditions are equivalent:

  1. IS(uj,u)→0I_S(u_j,u)\to0;

  2. uj→uu_j\to u locally in outer relative capacity on all permitted compact/domain pairs;

  3. ES,h(uj)→ES,h(u)\mathcal E_{S,h}(u_j)\to\mathcal E_{S,h}(u) and ∫(uj−u)(Du)qS→0\int(u_j-u)(Du)^qS\to0.

In particular, for an actual pointwise convergent sequence with these bounds and collar, convergence of its energies suffices for capacity convergence.

Proof. Taylor’s formula and the preceding second derivative give JS,h(u,v)=∑a=0q−1q−aq+1∫Γ(Dv)a(Du)q−1−aS(v−u).J_{S,h}(u,v)= \sum_{a=0}^{q-1}\frac{q-a}{q+1} \int\Gamma_{(Dv)^a(Du)^{q-1-a}S}(v-u). Indeed the coefficient is q(q−1a)∫01ta(1−t)q−adt=(q−a)/(q+1)q\binom{q-1}{a}\int_0^1t^a(1-t)^{q-a}dt=(q-a)/(q+1). Each measure is positive, proving the two bounds. Condition (3) therefore implies (1). The all-degree estimate gives convergence on the fixed capacity domain Ω\Omega from (1), and Corollary “One fixed capacity domain suffices locally” gives the all-domain local assertion (2).

Conversely, the completed weighted capacity-continuity theorem applies to every finite term in the energy, with a compact cutoff equal to one near LL. It gives both limits in (3), and also the defining limit for II. The common support and local bounds justify all these compact pairings. Under actual pointwise convergence, the integral against the fixed measure (Du)qS(Du)^qS tends to zero by bounded convergence on LL. This proves the last assertion. This corollary is DIRECT_COROLLARY of the preceding estimates and the established completed calculus. ◻

Theorem (Convex-block compactness in every positive remaining degree). Let Ω\Omega be an ordinary open in one allowed smooth affine algebraic neighborhood, and let SS be the fixed background above. Every sequence uj∈Cb(Ω)u_j\in\mathcal C_b(\Omega) uniformly bounded on each compact neighborhood admits finite tail convex combinations vN∈conv⁡{uj:j≥N}v_N\in\operatorname{conv}\{u_j:j\geq N\} and an actual Borel a∈AS,b(Ω)a\in\mathcal A_{S,b}(\Omega) such that vN→av_N\to a locally in outer relative capacity on every permitted compact/domain pair.

For an ordinary Ω\Omega as in this theorem, use the same definition of AS[0,1](Ω)\mathcal A_S^{[0,1]}(\Omega) as in Definition def:r3-normalized-completion: its normalized approximating functions must be defined on the entire Ω\Omega. If 0≤uj≤10\leq u_j\leq1 on that entire original Ω\Omega, then 0≤vN≤10\leq v_N\leq1 there and one may take a∈AS[0,1](Ω)a\in\mathcal A_S^{[0,1]}(\Omega). Its normalized approximants are defined on that same domain. The completed weighted mixed calculus applies to these convex blocks and their limit.

Proof. Choose a countable cofinal family of norm-box stages Vl⋐Ωl⋐ΩV_l\Subset\Omega_l\Subset\Omega as in the R2 completion theorem. Every permitted compact/domain pair is covered by finitely many VlV_l whose Ωl\Omega_l lie inside its prescribed capacity domain.

For each ll, the common bounds on the original sequence and the smooth collar construction give clc_l, a smooth psh collar hlh_l, and a fixed compact Ll⋐ΩlL_l\Subset\Omega_l, such that uj,l=max⁡(uj+cl,hl)u_{j,l}=\max(u_j+c_l,h_l) equals uj+clu_j+c_l near V‾l\overline V_l and equals hlh_l off LlL_l. These functions have a common bound near LlL_l. The construction uses the uniform bound only; no monotonicity of uju_j is required.

Put Uj=(uj,l)l≥1U_j=(u_{j,l})_{l\geq1}. A finite convex vector W=∑jbjUjW=\sum_jb_jU_j has components Wl=∑jbjuj,lW_l=\sum_jb_ju_{j,l} with the same collar hlh_l. On VlV_l it agrees with the global convex combination ∑jbjuj+cl\sum_jb_ju_j+c_l. These componentwise convex combinations are retained; the nonlinear patching operation is not applied a second time to the global sum.

Write El=ES,hl\mathcal E_l=\mathcal E_{S,h_l}. Relative CLN gives Bl<∞B_l<\infty such that ∣El(Wl)∣≤Bl|\mathcal E_l(W_l)|\leq B_l for all these convex vectors. The functional F(W)=∑l≥12−l1+BlEl(Wl)F(W)=\sum_{l\geq1}\frac{2^{-l}}{1+B_l}\mathcal E_l(W_l) is bounded and concave. The series converges uniformly and absolutely on the convex hull. Let mN=sup⁡{F(W):W∈conv⁡(Uj:j≥N)}.m_N=\sup\{F(W):W\in\operatorname{conv}(U_j:j\geq N)\}. Then mN↓m∞m_N\downarrow m_\infty with finite limit. By the definition of supremum, choose a finite vector WNW_N in the indicated tail with F(WN)≥mN−ϵNF(W_N)\geq m_N-\epsilon_N, where ϵN=2−N\epsilon_N=2^{-N}. No compactness of the convex hull or existence of a maximizing vector is assumed.

For M≥NM\geq N the midpoint of WN,WMW_N,W_M belongs to the NN-tail. Its concavity gap is nonnegative and at most mN−mM+ϵN+ϵM2⟶0(N,M→∞).\frac{m_N-m_M+\epsilon_N+\epsilon_M}{2}\longrightarrow0 \quad(N,M\to\infty). Every summand of the gap is nonnegative. The midpoint estimate consequently gives, for each fixed ll, IS(WN,l,WM,l)⟶0.I_S(W_{N,l},W_{M,l})\longrightarrow0. The all-degree estimate makes the component sequences Cauchy in capacity on their fixed stages.

Use the same finite weights defining WNW_N to put vN=∑jbN,jujv_N=\sum_jb_{N,j}u_j on all of Ω\Omega. It is in the required tail convex hull and equals WN,l−clW_{N,l}-c_l on VlV_l. Thus, for every ll and every δ>0\delta>0, Cap⁡S∗(Vl∩{∣vN−vM∣>δ};Ωl)⟶0.\operatorname{Cap}_S^* (V_l\cap\{|v_N-v_M|>\delta\};\Omega_l)\longrightarrow0. Finite subadditivity and domain monotonicity give this Cauchy property on every prescribed compact/domain pair. Run the summable-subsequence construction of the R2 completion theorem on this one global countable cofinal list: at its mmth step make the first mm capacity errors summable. The pointwise limsup of that single subsequence is one actual Borel function on all of Ω\Omega. It is the local capacity limit on every stage, and preserves the common local bounds and the interval [0,1][0,1] when present. The approximants remain the global vNv_N. This proves all assertions. Weighted mixed convergence is then an application of the already proved completed calculus. ◻

Corollary (Actual bounded pointwise limits have a completed calculus). In the preceding setting, suppose uj∈Cb(Ω)u_j\in\mathcal C_b(\Omega) are locally uniformly bounded and converge at every point to uu. Then the actual Borel function uu belongs to AS,b(Ω)\mathcal A_{S,b}(\Omega). It has local relative quasicontinuity and the completed weighted BT calculus. There are finite tail convex combinations of the given sequence converging locally in outer SS-capacity to these same point values modulo relative outer-null sets.

This does not assert weighted or higher-degree unweighted BT convergence for the original unaveraged uju_j.

Proof. Choose vNv_N and its capacity limit aa from the theorem. At every fixed point, every finite convex combination of the NNth tail tends to uu as N→∞N\to\infty, so vN→uv_N\to u pointwise. The summable-subsequence construction for the known capacity convergence gives a subsequence converging pointwise to aa off locally relative-outer-null sets on the same global cofinal list. Thus u=au=a outside these sets. Subadditivity shows that vN→uv_N\to u in capacity on each exact compact/domain pair. This is membership in the completed class with the given Borel values. The remaining assertions are the completed-calculus and quasicontinuity theorems, and are DIRECT_COROLLARY. ◻

The nonlinear distinction.

For q=1q=1, linearity identifies a weak limit of D(ujS)D(u_jS) with D(aS)D(aS), as in Theorem “Linear-degree convex-block compactness”. For q>1q>1 the measure of a convex combination contains mixed terms. The new theorem does not identify a weak limit of (Duj)qS(Du_j)^qS with (Da)qS(Da)^qS. It provides capacity compactness after convex averaging, not capacity precompactness of the original sequence; the latter remains false even for the ambient background and continuous potentials.

Status and dependence.

The all-degree energy-capacity estimate and the simultaneous concave-tail selection argument are NEW_PROVED. The positive energy identity and the Aubin–Mabuchi variation formulas are DIRECT_COROLLARY of the bounded mixed and common-collar Green calculus. No compact outer regularity, strictly psh exhaustion, arbitrary-current wedge, or arbitrary varying-background stability is used.

Strong psh-dual topology for arbitrary fixed backgrounds

Keep the fixed finite extension K/QpK/\mathbf Q_p, its original absolute value, and the allowed analytic spaces. Let SS be a fixed symmetric original-positive, separately d′d'- and d′′d''-closed current of bidegree (p,p)(p,p), and put q=n−p≥1q=n-p\geq1. All identities below use the full original compact smooth test sheaf. Write μu=(Du)qS\mu_u=(Du)^qS for u∈Cb(U)u\in\mathcal C_b(U). For a compactly supported finite signed measure σ\sigma set ∥σ∥psh,U=sup⁡a∈Cb(U)0≤a≤1∣∫a dσ∣,dpsh,U(ν,ν′)=∥ν−ν′∥psh,U.\|\sigma\|_{\mathrm{psh},U} =\sup_{\substack{a\in\mathcal C_b(U)\\0\leq a\leq1}} \left|\int a\,d\sigma\right|, \qquad d_{\mathrm{psh},U}(\nu,\nu') =\|\nu-\nu'\|_{\mathrm{psh},U}. This is a seminorm; no faithfulness assertion is implicit.

The common-collar estimate

Let Ω\Omega be a prescribed ordinary norm-box domain. Let L⋐ΩL\Subset\Omega lie in a strictly smaller norm box, and let hh be smooth classical psh with semipositive original curvature form. Suppose u,v∈Cb(Ω),∣u∣,∣v∣≤B,u=v=hon Ω∖L.u,v\in\mathcal C_b(\Omega),\qquad |u|,|v|\leq B,\qquad u=v=h\quad\hbox{on }\Omega\setminus L. Choose original smooth 0≤χ≤10\leq\chi\leq1 of compact support, equal to one on a neighborhood of LL. Locality shows that μu−μv\mu_u-\mu_v is supported in LL. Thus dS(u,v):=∥μu−μv∥psh,Ω=dpsh,Ω(χμu,χμv)\mathfrak d_S(u,v) :=\|\mu_u-\mu_v\|_{\mathrm{psh},\Omega} =d_{\mathrm{psh},\Omega}(\chi\mu_u,\chi\mu_v) does not depend on the cutoff. Put IS(u,v)=∫(u−v)(μv−μu)I_S(u,v)=\int(u-v)(\mu_v-\mu_u).

Theorem (Energy and strong psh-dual distance). Under these hypotheses there is a finite constant CC, depending only on the fixed compact stages, the common bound, qq, and finitely many positive coefficient evaluations of SS, such that

dS(u,v)≤CIS(u,v)1/2,IS(u,v)≤4B dS(u,v).\mathfrak d_S(u,v)\leq C I_S(u,v)^{1/2}, \qquad I_S(u,v)\leq4B\,\mathfrak d_S(u,v).

The constant is independent of the normalized tester and of regularization indices.

Status: NEW_PROVED for the uniform forward estimate; the reverse estimate is a DIRECT_COROLLARY of normalization and positivity of ISI_S.

Proof. Set f=u−vf=u-v and Q=∑b=0q−1(Du)b(Dv)q−1−bS.Q=\sum_{b=0}^{q-1}(Du)^b(Dv)^{q-1-b}S . Each summand is an already defined positive separately closed current of bidegree (n−1,n−1)(n-1,n-1). Symmetry, telescoping, and the compact Green identity give D(fQ)=μu−μv,∫ΓQ(f)=IS(u,v)≥0.D(fQ)=\mu_u-\mu_v,\qquad \int\Gamma_Q(f)=I_S(u,v)\geq0 . For a normalized actual classical tester aa, the protected common-collar approximation and the bounded Green identity yield ∫a(μu−μv)=−∫χ ΓQ(a,f).\int a(\mu_u-\mu_v)=-\int\chi\,\Gamma_Q(a,f). There is no cutoff error: ΓQ(χ,f)=0\Gamma_Q(\chi,f)=0, since χ\chi is constant near LL and f=0f=0 off LL. This identity is legitimate for bounded, possibly discontinuous a,u,va,u,v by the fixed-background energy calculus; one first uses smooth rows on the prescribed box and then its weighted convergence. No varying arbitrary background is passed to a limit.

For 0≤a≤10\leq a\leq1, the chain identity and positivity give 0≤∫χΓQ(a)=12⟨a2Q,Dχ⟩−∫χa D(aQ)≤12∣μQ,Dχ∣(supp⁡χ).0\leq\int\chi\Gamma_Q(a) =\tfrac12\langle a^2Q,D\chi\rangle -\int\chi a\,D(aQ) \leq\tfrac12|\mu_{Q,D\chi}|(\operatorname{supp}\chi). Relative CLN bounds the last expression uniformly in u,v,au,v,a. Cauchy–Schwarz therefore bounds the preceding pairing by CIS(u,v)1/2C I_S(u,v)^{1/2}. Taking the supremum proves the first inequality. If B>0B>0, the functions (u+B)/(2B)(u+B)/(2B) and (v+B)/(2B)(v+B)/(2B) are normalized testers. Their difference is f/(2B)f/(2B), so IS(u,v)≤4BdS(u,v)I_S(u,v)\leq4B\mathfrak d_S(u,v). The case B=0B=0 is trivial. ◻

Corollary (Equivalent relative topologies). For uniformly bounded common-collar potentials, convergence of IS(uj,u)I_S(u_j,u) to zero, convergence of dS(uj,u)\mathfrak d_S(u_j,u) to zero, and local outer SS-capacity convergence on all permitted compact/domain pairs are equivalent. The corresponding Cauchy conditions are equivalent as well. These statements concern relative capacity classes, not equality of actual values at capacity-zero points.

Proof. Combine Theorem “Energy and strong psh-dual distance” with Theorem “An all-degree mixed-energy capacity estimate” and Corollary “An energy criterion for capacity convergence”. For completeness, the capacity-Cauchy implication for pairs follows directly from uniform coefficient-capacity control: IS(uj,uk)≤∫L∣uj−uk∣(μuj+μuk)≤Cδ+C′Cap⁡S∗(L∩{∣uj−uk∣>δ};Ω).I_S(u_j,u_k) \leq\int_L|u_j-u_k|(\mu_{u_j}+\mu_{u_k}) \leq C\delta+ C'\operatorname{Cap}_S^*(L\cap\{|u_j-u_k|>\delta\};\Omega). First let j,kj,k tend to infinity, then δ↓0\delta\downarrow0. Reverse framed domain comparison supplies the all-domain local form of fixed-domain capacity convergence. ◻

Localized strong convergence without a collar

Theorem (Capacity convergence implies strong measure convergence). Let uj,u∈Cb(U)u_j,u\in\mathcal C_b(U) have one common actual bound on every compact subset. Suppose uj→uu_j\to u locally in outer SS-capacity, on all permitted compact/domain pairs. For every original smooth compact scalar 0≤χ≤10\leq\chi\leq1, dpsh,U(χμuj,χμu)⟶0.d_{\mathrm{psh},U}(\chi\mu_{u_j},\chi\mu_u)\longrightarrow0. Neither continuity nor monotonicity of the potentials is required.

Status: NEW_PROVED.

Proof. A finite nonnegative smooth partition reduces the cutoff to one prescribed regularization box. On this box choose original compact cutoffs 0≤η≤κ≤10\leq\eta\leq\kappa\leq1, with η=1\eta=1 near supp⁡χ\operatorname{supp}\chi and κ=1\kappa=1 near supp⁡η\operatorname{supp}\eta. Put fj=uj−u,Qj=∑b=0q−1(Duj)b(Du)q−1−bS,ej=∫η ΓQj(fj).f_j=u_j-u,\qquad Q_j=\sum_{b=0}^{q-1}(Du_j)^b(Du)^{q-1-b}S,\qquad e_j=\int\eta\,\Gamma_{Q_j}(f_j). All these expressions use the established bounded energy calculus.

First let aa be a smooth classical psh function with 0≤a≤20\leq a\leq2 on a neighborhood of the cutoff stages. Stokes gives

∫aχ D(fjQj)=−∫χ ΓQj(a,fj)−∫a ΓQj(χ,fj).\int a\chi\,D(f_jQ_j) =-\int\chi\,\Gamma_{Q_j}(a,f_j) -\int a\,\Gamma_{Q_j}(\chi,f_j).

Here is a precise limiting justification. Fix jj and the background QjQ_j first, and smooth only uj,uu_j,u downward on the prescribed compact box. The smooth identity passes by distributional and energy convergence against the fixed smooth tests χ\chi and κa\kappa a. A smooth cutoff is locally a difference of two smooth classical functions after adding a sufficiently large convex quadratic in the unit moment coordinates. The established bilinear energy identity therefore applies. This passage does not pair a changing nonsmooth multiplier with a changing coefficient measure.

The chain identity, a≥0a\geq0, and relative CLN show ∫η ΓQj(a)≤∫κ ΓQj(a)≤2∣μQj,Dκ∣(supp⁡κ)≤C.\int\eta\,\Gamma_{Q_j}(a) \leq\int\kappa\,\Gamma_{Q_j}(a) \leq2|\mu_{Q_j,D\kappa}|(\operatorname{supp}\kappa)\leq C. The energy of the fixed smooth χ\chi is uniformly bounded by the same coefficient estimates. Apply the weighted Cauchy–Schwarz inequality to each term in (eq:r3-local-strong-green); a≤2a\leq2 controls the second. One obtains

∣∫aχ d(μuj−μu)∣≤Cχej1/2.\left|\int a\chi\,d(\mu_{u_j}-\mu_u)\right| \leq C_\chi e_j^{1/2}.

The constant is independent of this tester.

Now let a∈Cb(U)a\in\mathcal C_b(U) be any actual normalized tester. Its restriction has a smooth decreasing classical row on the fixed compact stages; late terms lie between zero and two there. For fixed jj, bounded convergence against the fixed finite signed measure χ(μuj−μu)\chi(\mu_{u_j}-\mu_u) passes (eq:r3-local-strong-bound) to aa. No energy limit involving this tester is needed. The same bound holds uniformly for every normalized actual tester on UU.

It remains to prove ej→0e_j\to0. The energy expansion is ej=12⟨fj2Qj,Dη⟩−∫ηfj d(μuj−μu).e_j=\tfrac12\langle f_j^2Q_j,D\eta\rangle -\int\eta f_j\,d(\mu_{u_j}-\mu_u). All coefficient measures in this formula belong to explicit mixed products over the one fixed SS. Formula (eq:r2-approx-coefficient-cap), with the common local bounds, gives uniform mass and capacity control on the fixed stages. Splitting into ∣fj∣≤δ|f_j|\leq\delta and ∣fj∣>δ|f_j|>\delta bounds the first term in absolute value by Cδ2+C′Cap⁡S∗({∣fj∣>δ}∩H;Ω)C\delta^2+C'\operatorname{Cap}_S^*(\{|f_j|>\delta\}\cap H;\Omega) and the second by Cδ+C′Cap⁡S∗({∣fj∣>δ}∩H;Ω)C\delta+C'\operatorname{Cap}_S^*(\{|f_j|>\delta\}\cap H;\Omega), for a fixed compact stage HH. Let j→∞j\to\infty and then δ↓0\delta\downarrow0. Thus ej→0e_j\to0. Taking the supremum in (eq:r3-local-strong-bound) and summing the finitely many cutoff pieces proves the theorem. ◻

Corollary (Bounded monotone and mixed strong convergence). The localized psh-dual monotone convergence theorem “Uniform psh-weight convergence for decreasing rows” holds for locally uniformly bounded actual classical decreasing rows, without requiring that their terms be continuous. More generally, mixed top-degree products whose finitely many slots converge locally in outer SS-capacity with common bounds converge in the localized psh-dual seminorm.

Proof. The proved decreasing convergence in relative capacity and Theorem “Capacity convergence implies strong measure convergence” give the first assertion. For the second use finite polarization, q!TS,q(a1,…,aq)=∑J⊂{1,…,q}(−1)q−∣J∣(D∑i∈Jai)qS.q!T_{S,q}(a_1,\ldots,a_q) =\sum_{J\subset\{1,\ldots,q\}}(-1)^{q-|J|} \left(D\sum_{i\in J}a_i\right)^qS. Every sum has common local bounds and the required capacity convergence. The triangle inequality finishes the proof. ◻

Scope.

The background may charge ambient pluripolar sets. The theorem does not replace relative quasi-everywhere equality by pointwise equality, solve the general potential-quotient problem, or allow an arbitrary changing background SjS_j. For q=0q=0 the differentiated measure is the fixed background itself, so the corresponding convergence assertion is immediate.

Strong convergence to relative completed potentials

Keep the arbitrary fixed background SS and the original psh-dual seminorm of Section “Strong psh-dual topology for arbitrary fixed backgrounds”. For a∈AS,b(U)a\in\mathcal A_{S,b}(U), write μa=TS,q(a,…,a),q=n−p≥1.\mathcal\mu_a=\mathcal T_{S,q}(a,\ldots,a),\qquad q=n-p\geq1. This section does not assume that aa has an ambient classical representative.

Corollary (Strong convergence to the completed measure). Let uj∈Cb(U)u_j\in\mathcal C_b(U) have common actual bounds on compact subsets and converge locally in outer SS-capacity, on all permitted compact/domain pairs, to a locally bounded Borel function aa. Then a∈AS,b(U)a\in\mathcal A_{S,b}(U) and, for every original smooth compact scalar 0≤χ≤10\leq\chi\leq1,

dpsh,U(χ(Duj)qS,χμa)⟶0.d_{\mathrm{psh},U} \bigl(\chi(Du_j)^qS,\chi\mathcal\mu_a\bigr) \longrightarrow0.

The limit uses the actual completed relative potential, not its ambient usc regularization. In particular, whenever the proved bounded relative-capacity completion supplies such a limit to a capacity-Cauchy sequence, its localized measures are Cauchy and converge in this strong seminorm as well.

Status: DIRECT_COROLLARY of the pair estimate in Theorem “Capacity convergence implies strong measure convergence” and the completed weighted BT calculus.

Proof. Capacity convergence implies the corresponding capacity Cauchy condition. Apply the proof of Theorem “Capacity convergence implies strong measure convergence” to a pair uj,uku_j,u_k. On each of its finitely many cutoff stages put fj,k=uj−uk,Qj,k=∑b=0q−1(Duj)b(Duk)q−1−bS.f_{j,k}=u_j-u_k,\qquad Q_{j,k}=\sum_{b=0}^{q-1}(Du_j)^b(Du_k)^{q-1-b}S. The constants are uniform for both indices, and the proof gives dpsh,U(χ(Duj)qS,χ(Duk)qS)≤C(∫η ΓQj,k(fj,k))1/2d_{\mathrm{psh},U}\bigl(\chi(Du_j)^qS,\chi(Du_k)^qS\bigr) \leq C\left(\int\eta\,\Gamma_{Q_{j,k}}(f_{j,k}) \right)^{1/2} on a single stage; for a general cutoff one sums finitely many such bounds. The energy expansion and uniform coefficient-capacity estimates in that proof show that every displayed energy tends to zero as j,k→∞j,k\to\infty. This step involves only classical potentials and explicit mixed backgrounds over the one fixed SS. No energy measure of an unproved Borel representative is introduced.

The completed weighted calculus already supplies μa\mathcal\mu_a as a positive Radon measure. For each fixed normalized tester b∈Cb(U)b\in\mathcal C_b(U) it gives ∫bχ(Duk)qS⟶∫bχ dμa.\int b\chi(Du_k)^qS\longrightarrow\int b\chi\,d\mathcal\mu_a. Indeed bb is simply a constant sequence in the weighted slot. Given ϵ>0\epsilon>0, choose NN so that the preceding pair seminorm is at most ϵ\epsilon for all j,k≥Nj,k\geq N. For fixed j≥Nj\geq N and fixed bb, let k→∞k\to\infty to obtain ∣∫bχ d((Duj)qS−μa)∣≤ϵ.\left|\int b\chi\,d\bigl((Du_j)^qS-\mathcal\mu_a\bigr)\right| \leq\epsilon. The bound is independent of bb. Taking its supremum only now proves (eq:r3-completed-strong-dual). Thus the argument does not interchange a weak limit with an uncontrolled supremum. ◻

Corollary (Stability within the bounded completed class). Suppose aj,a∈AS,b(U)a_j,a\in\mathcal A_{S,b}(U) converge locally in outer SS-capacity and, locally, their defining classical approximants have one common actual bound. Then dpsh,U(χμaj,χμa)⟶0d_{\mathrm{psh},U}(\chi\mathcal\mu_{a_j}, \chi\mathcal\mu_a)\longrightarrow0 for every cutoff as above. The corresponding statement holds for mixed top-degree products with finitely many convergent slots and common defining bounds.

Proof. Work first on one of the fixed local cutoff stages. For every jj, choose a sufficiently late classical defining approximant vjv_j of aja_j so that its capacity error tends to zero on the finitely many larger compact stages used by the local estimate, and its localized measure differs from that of aja_j in psh-dual seminorm by at most 1/j1/j. Such a choice is possible by Corollary “Strong convergence to the completed measure”, applied to the defining sequence for this fixed jj. The common defining bound keeps all constants uniform. A countable diagonal, if necessary, also makes vj→av_j\to a in the local all-domain capacity sense. The same corollary applied to vjv_j gives its strong convergence to μa\mathcal\mu_a. The triangle inequality proves the assertion. Local tester estimates apply to restrictions of normalized testers on UU, and a finite partition of the compact cutoff gives the stated global seminorm bound. Finally use the finite polarization identity of Corollary “Bounded monotone and mixed strong convergence” for the mixed statement. No classical representative of the completed limits is required. ◻

Scope.

This strengthens stability of the already defined relative completed operator. It does not make the ambient classical class closed under these limits, and it does not add an arbitrary varying-background convergence theorem. The explicit completed solution in Proposition “An actual completed solution” is one application where an ambient classical zero-boundary solution is impossible.

A canonical pluripolar decomposition of positive closed currents

The background in this section is arbitrary. The purpose of the decomposition is to locate the part on which replacing an actual increasing limit by its ambient upper regularization can lose information. The construction uses actual coefficient measures, not an unspecified ambient volume measure.

We work over the prescribed finite extension K/QpK/\mathbf Q_p. A current is positive in the original CLD cone, is symmetric, and is separately d′d'- and d′′d''-closed. All tests are original compactly supported smooth CLD forms. A current TT is said to be carried by a Borel set EE if 1ET=T\mathbf1_E T=T. This is a coefficient-measure assertion; it does not assert that the topological support of TT is contained in a nonclosed EE.

A finite measure detecting all coefficient-null sets

Lemma (A countable family of coefficient controls). Let UU be an ordinary open in the analytification of a smooth integral affine KK-variety of dimension nn. Fix 0≤p≤n0\leq p\leq n. There is a countable family of positive original forms Θl∈Acn−p,n−p(U)\Theta_l\in A_c^{n-p,n-p}(U) with this property: for every original compact smooth complementary test η\eta, there are finitely many indices l1,…,lNl_1,\ldots,l_N and a finite constant CηC_\eta such that every original-positive (p,p)(p,p)-current TT satisfies the inequality of Radon measures

∣μT,η∣≤Cη∑j=1NμT,Θlj.(1)|\mu_{T,\eta}| \leq C_\eta\sum_{j=1}^N\mu_{T,\Theta_{l_j}}. \tag{1}

The controls depend on U,pU,p, but not on TT.

Proof. Choose a countable dense subfield K0⊂KK_0\subset K containing the finitely many coefficients needed to define the affine variety and its fixed affine embedding. Such a field exists since KK is finite over Qp\mathbf Q_p and is separable as a topological field. There are countably many rational functions with coefficients in K0K_0 and countably many finite lists of these functions.

Every finite analytic unit presentation can locally be replaced by one of these lists without changing any logarithmic absolute value. Indeed B, Lemma D.1, first replaces the analytic units on a compact affinoid neighborhood by rational functions over KK. Write their numerators and denominators in the fixed affine coordinates. The numerators and denominators are units on that compact neighborhood. Approximate their finitely many coefficients by elements of K0K_0 closely enough that the uniform errors are smaller than the positive minima of the original absolute values. The ultrametric inequality preserves each numerator and denominator norm, and hence each rational function norm, at every point of the neighborhood. Their original smooth logarithmic differentials also agree on its ordinary interior. This is an exact replacement of the moment coordinates, not a limiting identification of currents.

For each finite rational list, let OO be its algebraic unit locus intersected with UU, and write its logarithmic coordinates as t1,…,tst_1,\ldots,t_s. The space OO is locally compact and second countable, so choose a countable compact exhaustion with interiors covering OO, and nonnegative original smooth compact cutoffs χj\chi_j equal to one near these compact sets. Put r=n−pr=n-p and include the forms χj1r!(∑i=1sd′ti∧d′′ti)r\chi_j\frac1{r!} \left(\sum_{i=1}^s d't_i\wedge d''t_i\right)^r in the countable family, extending them by zero off OO. For r=0r=0 the parenthesized factor is one. Original smooth cutoffs and compact exhaustions are available by the same ordinary topology and partition-of-unity inputs as in R1.

Now divide a given compact test η\eta into finitely many pieces on neighborhoods where the exact rational replacements apply. Each piece has one of the listed moment presentations. On its fixed compact support, its symmetric coefficient matrix is bounded by a multiple of the positive definite matrix of the displayed power. Choose χj=1\chi_j=1 near that support. R’s finite-test cone domination, recalled in Lemma “Positive controls on a finite test stage”, then gives (1). More explicitly, its cone inequalities remain valid after multiplication by any original smooth scalar on a common refinement; Riesz uniqueness therefore makes the domination an inequality of coefficient measures. Positive currents annihilate the antisymmetric null part. Summing the finitely many pieces proves the assertion. ◻

Corollary (A faithful finite control measure). For every such positive current SS, there is a finite positive Radon measure λS\lambda_S on UU such that, for every Borel E⊂UE\subset U,

λS(E)=0⟺1ES=0.(2)\lambda_S(E)=0 \quad\Longleftrightarrow\quad \mathbf1_E S=0. \tag{2}

The measure need not be canonical; its null ideal is canonical.

Proof. Set μl=μS,Θl\mu_l=\mu_{S,\Theta_l}. Each is a finite positive Radon measure, since Θl\Theta_l has compact support. Define λS=∑l≥12−l1+μl(U) μl.\lambda_S=\sum_{l\geq1} \frac{2^{-l}}{1+\mu_l(U)}\,\mu_l. Its total mass is at most one. It is Radon: finite initial sums are Radon, and the tails have uniformly small total mass, so compact inner approximation passes to the sum. If λS(E)=0\lambda_S(E)=0, then μl(E)=0\mu_l(E)=0 for every ll; the preceding lemma makes every coefficient restriction 1EμS,η\mathbf1_E\mu_{S,\eta} vanish. Conversely, 1ES=0\mathbf1_E S=0 makes all μl(E)\mu_l(E) vanish. This proves (2). ◻

These two control statements are DIRECT_COROLLARY of the original coefficient positivity in R, exact moment replacement in B, and separability of the prescribed field. No globally strictly positive form is used. The countable family may contain arbitrarily many different moment presentations.

The maximal pluripolar part

Call SS non-pluripolar if 1ES=0\mathbf1_E S=0 for every Borel locally pluripolar set EE. This term refers to all its coefficient measures, not only to one selected top-degree trace.

Theorem (Pluripolar decomposition on an exhaustion domain). Let VV be a connected exhaustion domain of Definition “The exhaustion domains used in this section”, of positive dimension. Every allowed positive closed current SS on VV has a unique decomposition

S=Spp+Snp(3)S=S_{\rm pp}+S_{\rm np} \tag{3}

into positive symmetric separately closed currents such that SppS_{\rm pp} is carried by a complete pole set in VV and SnpS_{\rm np} is non-pluripolar. There is a nonpositive, nontrivial extended classical psh function ϕ\phi on VV, with Q={ϕ=−∞}Q=\{\phi=-\infty\}, for which

Spp=1QS,Snp=1V∖QS.(4)S_{\rm pp}=\mathbf1_QS, \qquad S_{\rm np}=\mathbf1_{V\setminus Q}S. \tag{4}

The two currents are independent of the control measure and the chosen witness. They commute with ordinary restriction, addition of positive closed currents, and nonnegative scaling.

Proof. Use the finite control measure λS\lambda_S above. Let P\mathscr P consist of the complete pole sets of nonpositive, nontrivial extended classical psh functions on VV, and put m=sup⁡P∈PλS(P)≤λS(V).m=\sup_{P\in\mathscr P}\lambda_S(P)\leq\lambda_S(V). The empty set belongs to P\mathscr P. Choose Pj∈PP_j\in\mathscr P with λS(Pj)→m\lambda_S(P_j)\to m, and choose nonpositive witnesses ϕj\phi_j. Fix one ambient divisorial point z∈Vz\in V. By the R2 no-divisorial-poles theorem, every ϕj(z)\phi_j(z) is finite. Choose positive bjb_j such that bj∣ϕj(z)∣≤2−jb_j|\phi_j(z)|\leq2^{-j}. The decreasing sum ϕ=∑j≥1bjϕj\phi=\sum_{j\geq1}b_j\phi_j is extended classical psh and is nontrivial, since ϕ(z)≥−1\phi(z)\geq-1. This is the negative-series construction already used in the R2 pole-witness theorem. Its pole set QQ contains every PjP_j, so m≤λS(Q)≤mm\leq\lambda_S(Q)\leq m. In particular the supremum is attained by this one complete pole set.

Let E⊂VE\subset V be Borel and locally pluripolar. The R2 same-domain pole-patching theorem provides a member of P\mathscr P containing EE. Applying the same negative sum construction to that witness and ϕ\phi gives R∈PR\in\mathscr P containing E∪QE\cup Q. Maximality gives λS(E∖Q)≤λS(R)−λS(Q)=0.\lambda_S(E\setminus Q) \leq\lambda_S(R)-\lambda_S(Q)=0. By (2), 1E∖QS=0\mathbf1_{E\setminus Q}S=0. Thus the second current in (4) is non-pluripolar. The R2 pole-restriction theorem makes both displayed restrictions positive and separately closed. This proves existence, including the actual coefficient interpretation.

For uniqueness, suppose S=A+BS=A+B is another such decomposition, with AA carried by a locally pluripolar Borel set PP and BB non-pluripolar. Positivity gives 0≤A≤S0\leq A\leq S. Since 1P∖QS=0\mathbf1_{P\setminus Q}S=0, positivity of coefficient controls gives 1P∖QA=0\mathbf1_{P\setminus Q}A=0. Hence 1QA=A\mathbf1_Q A=A. Also 1QB=0\mathbf1_QB=0. Restricting S=A+BS=A+B to QQ now gives A=SppA=S_{\rm pp} and B=SnpB=S_{\rm np}.

These properties are local. In particular restriction of a non-pluripolar current to an ordinary open W⊂VW\subset V is still non-pluripolar. To check this precisely, exhaust WW by compact subsets: a locally pluripolar set intersected with one of these compacts is locally pluripolar in VV, and coefficient measures pass to the countable union. Uniqueness therefore identifies the restricted decomposition with the one constructed on a smaller exhaustion domain. For sums, the sum of the non-pluripolar parts is non-pluripolar, and the sum of the pluripolar parts is carried by the union of their two pole sets, contained in one complete pole set by the negative-sum construction. Uniqueness proves additivity. Scaling is immediate. ◻

Corollary (Local decomposition on the allowed analytic spaces). On any allowed analytic space, the preceding local decompositions glue to a canonical decomposition S=Spp+SnpS=S_{\rm pp}+S_{\rm np}. Locally, SppS_{\rm pp} is carried by a complete pole set; SnpS_{\rm np} gives zero mass to every locally pluripolar Borel set in all coefficient measures. No global countability assumption on the analytic space is required for this local assertion.

Proof. Use B’s ordinary algebraic neighborhoods and the R2 local basis of exhaustion domains. The uniqueness just proved identifies the decompositions on every smaller such domain in an overlap. Currents and their positivity and separate closedness are local for ordinary opens, so the pieces glue. In dimension zero the only locally pluripolar set is empty; the assertion there is Spp=0S_{\rm pp}=0, Snp=SS_{\rm np}=S. ◻

Compatibility with the bounded Bedford–Taylor calculus

Theorem (Decomposition commutes with bounded mixed products). For every 0≤k≤n−p0\leq k\leq n-p and u1,…,uk∈Cb(U)u_1,\ldots,u_k\in\mathcal C_b(U), (TS,k(u1,…,uk))pp=TSpp,k(u1,…,uk),\bigl(T_{S,k}(u_1,\ldots,u_k)\bigr)_{\rm pp} =T_{S_{\rm pp},k}(u_1,\ldots,u_k), (TS,k(u1,…,uk))np=TSnp,k(u1,…,uk).\bigl(T_{S,k}(u_1,\ldots,u_k)\bigr)_{\rm np} =T_{S_{\rm np},k}(u_1,\ldots,u_k). These are current equalities against all original tests. In particular all mixed products over a non-pluripolar background have non-pluripolar coefficient measures.

Proof. Work on an exhaustion neighborhood where Spp=1QSS_{\rm pp}=\mathbf1_QS as above. Linearity of the background in the recursive construction gives TS,k=TSpp,k+TSnp,k.T_{S,k}=T_{S_{\rm pp},k}+T_{S_{\rm np},k}. The R2 pole-restriction identity gives 1QTSpp,k=TSpp,k\mathbf1_QT_{S_{\rm pp},k}=T_{S_{\rm pp},k}. For any locally pluripolar Borel set EE, choose a complete pole set PP containing it on this neighborhood. Since 1PSnp=0\mathbf1_PS_{\rm np}=0, the same identity gives 1PTSnp,k=T1PSnp,k=0.\mathbf1_PT_{S_{\rm np},k} =T_{\mathbf1_PS_{\rm np},k}=0. Thus the two terms have exactly the required polar and non-pluripolar properties. Uniqueness of the decomposition, applied to the resulting positive closed current, identifies them with its canonical parts. Locality proves the assertion on UU. ◻

Status and limits.

The decomposition and its BT compatibility are recorded as NEW_PROVED relative to the proved R2 pole restriction and pole-patching theorems. Original smooth positive backgrounds and bounded ambient BT backgrounds are non-pluripolar by R2’s ambient trace theorem. Subvariety integration backgrounds can instead be entirely pluripolar. This is not an analytic-stratum or Siu decomposition: no analyticity of a pole set is asserted. Nor is the capacity of a sum asserted to be the sum of the capacities, since the same test potential enters both summands.

Zero extension across a closed complete pluripolar set

The restriction theorem of R2 starts with a current on the whole ordinary open set. Here the current is initially defined only on the complement of the exceptional set. Its closedness across that set is part of the conclusion. The proof adapts the classical El Mir argument, but the finite-mass hypothesis and the first-order estimate are formulated for all original CLD coefficient tests.

The base field, analytic category, and original positivity cone are unchanged. A closed set P⊂UP\subset U is locally complete pluripolar here if, near each of its points, it is exactly the pole set of a nontrivial extended classical psh function. On a smaller neighborhood such a witness can be taken nonpositive. No assertion below replaces this equality by containment in an unspecified larger pole set.

The mass condition and a first-order estimate

Let SS be a symmetric original-positive (p,p)(p,p)-current on U∖PU\setminus P, and put q=n−pq=n-p. For an original compact smooth (q,q)(q,q)-form η\eta on UU, denote by μS,η\mu_{S,\eta} the coefficient measure on U∖PU\setminus P obtained by restricting the form there. This notation does not require that its restricted support be compact in U∖PU\setminus P.

We say that SS has locally finite coefficient mass across PP if, locally on UU,

∣μS,η∣(U∖P)<∞for every η∈Acq,q(U).|\mu_{S,\eta}|(U\setminus P)<\infty \quad\hbox{for every }\eta\in A_c^{q,q}(U).

Equivalently, it suffices to require this for the finite positive controls of each original test stage. In particular the condition includes controls for the coordinates of compact cutoffs. It is not a finite-mass assumption for just one initially selected tropical trace.

Lemma (Coefficientwise extension and weighted first-order control). Under (eq:r3-finite-mass-across-pole), the coefficientwise zero extension ⟨S~,η⟩=∫U∖PdμS,η\langle\widetilde S,\eta\rangle =\int_{U\setminus P}d\mu_{S,\eta} is an original-positive symmetric coefficient-Radon current on UU, with 1PS~=0\mathbf1_P\widetilde S=0.

Suppose q≥1q\geq1. Fix a moment presentation of an original (q−1,q)(q-1,q)-test α\alpha near its compact support, let t1,…,tst_1,\ldots,t_s be its moment coordinates, and put β=∑id′ti∧d′′ti.\beta=\sum_i d't_i\wedge d''t_i. Let χ≥0\chi\geq0 be an original compact cutoff equal to one near supp⁡α\operatorname{supp}\alpha, with support in that moment presentation’s ordinary domain. For every original smooth real ff and real Borel bb for which the displayed integrals are finite,

∣∫U∖Pb d′f∧α∧S∣2≤Cα(∫U∖Pχ d′f∧d′′f∧βq−1∧S)(∫U∖Pχb2βq∧S).\left|\int_{U\setminus P}b\,d'f\wedge\alpha\wedge S\right|^2 \leq C_\alpha \left(\int_{U\setminus P}\chi\,d'f\wedge d''f \wedge\beta^{q-1}\wedge S\right) \left(\int_{U\setminus P}\chi b^2\beta^q\wedge S\right).

There is the analogous inequality for (q,q−1)(q,q-1)-tests and d′′fd''f. The constant depends on the fixed test presentation and its bounded coefficients, and not on additional moment coordinates used by ff or by bounded approximations to bb. For a test with several presentations one uses a finite partition and the corresponding finite sum of these controls.

Proof. Extend each coefficient measure by zero on the closed set PP. The local finite-mass assumption makes these extensions Radon. Their linear relations and common-refinement identities hold off PP and both sides have zero restriction to PP, so the same identities hold on UU. On each fixed test stage, the positive controls in Lemma “Positive controls on a finite test stage” give an order-zero bound with finite right hand side. Hence the extended functionals are continuous in the original test topology and define a current. Positivity and symmetry pass through zero extension. This proves the first assertion without using either closedness or a Bedford–Taylor product.

For the second assertion, expand α\alpha as a finite sum of bounded smooth coefficients times d′tI∧d′′tJd't_I\wedge d''t_J, where ∣I∣=q−1|I|=q-1 and ∣J∣=q|J|=q. Use the standard Lagerberg signs for squares. On a common refinement, set AI=d′f∧d′tI,BJ=b d′tJ.A_I=d'f\wedge d't_I,\qquad B_J=b\,d't_J. The coefficient matrix of SS defines a positive semidefinite bilinear measure on these qq-vectors: the positivity needed here is duality with the original positive form cone. The two-by-two square inequality, followed by the scalar Cauchy–Schwarz inequality for its densities, bounds each mixed term by the square root of ∫χ [AI]2∧Stimes∫χ [BJ]2∧S.\int\chi\, [A_I]^2\wedge S \quad\hbox{times}\quad \int\chi\, [B_J]^2\wedge S. Here [AI]2[A_I]^2 denotes the positive square with that standard sign. Bounded coefficients of α\alpha contribute a fixed factor. Summing the first squares is bounded by a fixed multiple of d′f∧d′′f∧βq−1d'f\wedge d''f\wedge\beta^{q-1}, and summing the second squares is bounded by a fixed multiple of b2βqb^2\beta^q. The finite-sum Cauchy–Schwarz inequality gives (eq:r3-weighted-first-order-control).

One may first take bb smooth and then use coefficient-Radon measure identities and bounded approximation, or apply the same two-by-two inequality to Radon densities directly. Pullback to additional moment coordinates preserves the original positive cone. Thus the bounds use the fixed coordinates of α\alpha but impose no common fixed presentation on the varying ff. The d′′d'' assertion follows by the CLD involution. A finite original partition proves the last assertion. ◻

Smooth cutoffs adapted to the pole set

Lemma (Bounded psh cutoffs with a zero neighborhood). Work on a sufficiently small ordinary algebraic neighborhood, and let HH be a compact norm-box core in a larger such neighborhood. Suppose ϕ≤0\phi\leq0 is an extended classical psh function there and P={ϕ=−∞}P=\{\phi=-\infty\} is closed. On one fixed ordinary neighborhood of HH there are original smooth classical psh functions wNw_N, with semipositive original (1,1)(1,1)-forms DwNDw_N, such that 0≤wN≤1+N−2on H,wN=0near P∩H,0\leq w_N\leq1+N^{-2}\quad\hbox{on }H,\qquad w_N=0\quad\hbox{near }P\cap H, and wN(x)→1w_N(x)\to1 for every x∈H∖Px\in H\setminus P. No monotonicity in NN is asserted or needed.

Proof. The truncation ϕN=max⁡(ϕ,−2N)\phi_N=\max(\phi,-2N) is bounded classical psh on the larger neighborhood. By the prescribed-box regularization in Lemma “Regularization on a prescribed smaller norm box”, it has smooth classical psh approximants with semipositive original curvature forms decreasing to ϕN\phi_N on a fixed smaller neighborhood containing HH. The smaller geometric box can be chosen independently of NN; the projective line and regularization index need not be independent of NN.

Choose one approximant ψN≥ϕN\psi_N\geq\phi_N such that ψN<N−1on H,ψN<−3N/2on P∩H.\psi_N<N^{-1}\quad\hbox{on }H,\qquad \psi_N<-3N/2\quad\hbox{on }P\cap H. Both choices follow from compactness and the decreasing continuous rows: their limits are at most zero on HH and equal −2N-2N on P∩HP\cap H. The second inequality holds on an ordinary neighborhood of P∩HP\cap H.

Let FF be the convolution of t↦max⁡(t,0)t\mapsto\max(t,0) with an even nonnegative smooth probability kernel supported in (−1/4,1/4)(-1/4,1/4). Then FF is nonnegative, smooth, convex and nondecreasing, equals zero for t≤−1/4t\leq-1/4, and equals tt for t≥1/4t\geq1/4. Put wN=F(1+ψN/N).w_N=F(1+\psi_N/N). Convex nondecreasing composition preserves the classical psh property, by the all-extensions curve definition; the result is also an original smooth function. Moreover, DwN=F′(1+ψN/N)DψN/N+F′′(1+ψN/N)d′ψN∧d′′ψN/N2Dw_N=F'(1+\psi_N/N)D\psi_N/N+ F''(1+\psi_N/N)d'\psi_N\wedge d''\psi_N/N^2 is a semipositive original form. The stated upper bound follows by monotonicity of FF, and the negative inequality on ψN\psi_N gives the zero neighborhood. If ϕ(x)\phi(x) is finite, then, for all large NN, ψN(x)≥ϕ(x)\psi_N(x)\geq\phi(x) and 1+ϕ(x)/N>1/41+\phi(x)/N>1/4. Hence 1+ϕ(x)/N≤wN(x)≤1+N−2.1+\phi(x)/N\leq w_N(x)\leq1+N^{-2}. This proves the pointwise limit. No diagonal approximation at all points is involved. ◻

Theorem (A Skoda–El Mir zero-extension theorem). Let PP be a closed locally complete pluripolar subset of an allowed ordinary open UU. Let SS be a symmetric original-positive (p,p)(p,p)-current on U∖PU\setminus P, separately d′d'- and d′′d''-closed there. Assume (eq:r3-finite-mass-across-pole). Then its coefficientwise zero extension S~\widetilde S to UU is separately d′d'- and d′′d''-closed. All assertions are on all original compact smooth tests.

Proof. Lemma “Coefficientwise extension and weighted first-order control” constructs S~\widetilde S and proves its positivity, symmetry and coefficient-Radon property. If p=np=n, separate closedness follows from degree. Suppose q=n−p≥1q=n-p\geq1. Closedness is local for ordinary opens. Near a point of PP, choose a nonpositive complete pole witness and nested norm boxes as in Lemma “Bounded psh cutoffs with a zero neighborhood”. A finite partition reduces each compact test to this situation and, within it, to one fixed moment presentation.

Fix an original (q−1,q)(q-1,q)-test α\alpha. Choose a nonnegative compact original smooth χ\chi, equal to one near its support, with support in a compact core HH on which the cutoffs wNw_N are defined and within the ordinary domain of that fixed moment presentation. Put β=∑id′ti∧d′′ti\beta=\sum_i d't_i\wedge d''t_i for the coordinates of α\alpha. The positive current S∧βq−1S\wedge\beta^{q-1} is separately closed on the complement of PP. The function wNw_N is zero on an ordinary neighborhood of P∩HP\cap H, so every term below can be integrated by parts on that complement. One can insert a further compact cutoff away from PP equal to one on the relevant supports; its differentiated terms multiply zero functions.

The smooth identity D(wN2)=2wNDwN+2d′wN∧d′′wND(w_N^2)=2w_NDw_N+2d'w_N\wedge d''w_N gives

EN:=∫U∖Pχ d′wN∧d′′wN∧βq−1∧S=12∫U∖PwN2Dχ∧βq−1∧S−∫U∖PχwNDwN∧βq−1∧S≤2 ∣μS,Dχ∧βq−1∣(supp⁡χ)=:C<∞.\begin{aligned} E_N &:=\int_{U\setminus P} \chi\,d'w_N\wedge d''w_N\wedge\beta^{q-1}\wedge S\\ &=\frac12\int_{U\setminus P} w_N^2D\chi\wedge\beta^{q-1}\wedge S -\int_{U\setminus P} \chi w_NDw_N\wedge\beta^{q-1}\wedge S\\ &\leq 2\,|\mu_{S,D\chi\wedge\beta^{q-1}}| (\operatorname{supp}\chi) =:C<\infty . \end{aligned}

Here 0≤wN≤20\leq w_N\leq2 on the fixed compact set, and the last term being subtracted is nonnegative. The finite constant concerns the one fixed smooth form Dχ∧βq−1D\chi\wedge\beta^{q-1}, including the moment coordinates of χ\chi. This step has used closedness only where SS was originally given.

Set aN=2wN−wN2=1−(1−wN)2.a_N=2w_N-w_N^2=1-(1-w_N)^2. On HH one has 0≤aN≤10\leq a_N\leq1; it is zero near P∩HP\cap H, and tends to one at every point of H∖PH\setminus P. Consequently aNαa_N\alpha is a valid compact test on U∖PU\setminus P. Separate closedness there gives, with the harmless sign determined by total degree, ∣∫U∖PaNd′α∧S∣=2∣∫U∖P(1−wN)d′wN∧α∧S∣.\left|\int_{U\setminus P}a_Nd'\alpha\wedge S\right| =2\left|\int_{U\setminus P} (1-w_N)d'w_N\wedge\alpha\wedge S\right|. By (eq:r3-weighted-first-order-control), its square is at most 4CαEN∫U∖Pχ(1−wN)2βq∧S.4C_\alpha E_N \int_{U\setminus P}\chi(1-w_N)^2\beta^q\wedge S . The latter integral tends to zero by dominated convergence for the fixed finite coefficient measure χβq∧S~\chi\beta^q\wedge\widetilde S. Its restriction to PP is zero by construction. Uniform boundedness of ENE_N therefore makes the preceding error tend to zero. Dominated convergence for the fixed signed coefficient measure of d′αd'\alpha identifies the left hand side in the limit with ∣⟨S~,d′α⟩∣|\langle\widetilde S,d'\alpha\rangle|. This proves d′S~=0d'\widetilde S=0. The d′′d'' estimate and the same argument prove d′′S~=0d''\widetilde S=0.

Each localization and inequality uses original forms and preserves common-refinement identities. No convergence statement for a varying background current or an unbounded multiplier was invoked. ◻

Minimality and extension of bounded products

Corollary (The least positive extension). If RR is any symmetric original-positive separately closed extension of SS to UU, then R=S~+A,R=\widetilde S+A, where AA is positive, symmetric, separately closed and carried by PP. Conversely, adding any such AA gives an extension. Thus S~\widetilde S is the least positive extension. It is the unique extension whose coefficient measures give zero mass to PP.

Proof. Coefficientwise, the difference equals 1PR\mathbf1_PR, which is positive and symmetric. The theorem proves that S~\widetilde S is closed, so the difference is separately closed by subtraction. It is carried by PP. The converse and uniqueness follow immediately from the coefficient restrictions. No assertion is made that a positive current carried by PP must have an analytic-support description. ◻

Corollary (Bounded mixed products extend without extra mass). Let u1,…,uk∈Cb(U)u_1,\ldots,u_k\in\mathcal C_b(U), where 0≤k≤n−p0\leq k\leq n-p. The current TS,k(u1∣U∖P,…,uk∣U∖P)T_{S,k}(u_1|_{U\setminus P},\ldots, u_k|_{U\setminus P}) has locally finite coefficient mass across PP, and its coefficientwise zero extension is TS~,k(u1,…,uk).T_{\widetilde S,k}(u_1,\ldots,u_k). This current gives zero coefficient mass to PP. The same coefficient equality holds after multiplication by any locally bounded Borel weight on UU.

Proof. The theorem makes S~\widetilde S an admissible fixed background on UU. Its bounded mixed products exist and have locally finite coefficients by the established relative calculus and relative CLN estimate. Ordinary restriction identifies them with the asserted products on U∖PU\setminus P. The R2 pluripolar restriction identity gives 1PTS~,k(u1,…,uk)=T1PS~,k(u1,…,uk)=0.\mathbf1_P T_{\widetilde S,k}(u_1,\ldots,u_k) =T_{\mathbf1_P\widetilde S,k}(u_1,\ldots,u_k)=0. These two coefficient assertions identify the product with the zero extension and prove the mass claim. Actual Borel multiplication preserves the equality. The potentials in this corollary are required on UU; it does not assert a removable-singularity theorem for arbitrary potentials given only off PP. ◻

Status and dependencies.

The zero-extension theorem is NEW_PROVED in the prescribed non-Archimedean setting. The proof uses R’s original coefficient cone and B’s prescribed-box regularization. Its cutoff mechanism is the classical Skoda–El Mir mechanism; no novelty is claimed for that classical argument. The mass and first-order lemma makes the required transfer to all original tests explicit. The last two corollaries are DIRECT_COROLLARY, the last also using the R2 pole-restriction identity. Mere support avoidance in CLD’s zero-extension convention does not prove the finite-mass extension theorem stated here.

Two sharp scope tests for the zero-extension theorem

These examples test two hypotheses of Theorem “A Skoda–El Mir zero-extension theorem”. They do not concern the previously proved restriction theorem for a current already defined on the whole space. The field and its absolute value remain the prescribed finite pp-adic ones; every form below is an original CLD form.

Proposition (Closed pluripolar containment is insufficient). There are a closed pluripolar set EE in a smooth two-dimensional ordinary analytic domain UU and an original-positive symmetric (1,1)(1,1)-current SS on U∖EU\setminus E such that:

  1. SS is separately d′d'- and d′′d''-closed;

  2. all its original coefficient masses are locally finite across EE;

  3. its coefficientwise zero extension to UU is not closed.

The exceptional set is not locally complete pluripolar. Thus replacing closed locally complete pluripolar by closed pluripolar in Theorem “A Skoda–El Mir zero-extension theorem” is FALSE.

Proof. Let π\pi be a uniformizer and λ=−log⁡∣π∣>0\lambda=-\log|\pi|>0. Use coordinates (t,z)(t,z) on Gm,K×AK1\mathbf G_{m,K}\times\mathbf A^1_K and put s=log⁡∣t∣,U={−λ<s<λ, ∣z∣<1},Y=U∩{z=0},E=Y∩{s=0}.s=\log|t|,\qquad U=\{-\lambda<s<\lambda,\ |z|<1\},\qquad Y=U\cap\{z=0\},\qquad E=Y\cap\{s=0\}. The set UU is an ordinary open in the smooth affine surface, and YY is its closed smooth analytic curve. No normalization of the prescribed absolute value has been changed. The set EE is closed in UU. It is pluripolar because E⊂Y={log⁡∣z∣=−∞},E\subset Y=\{\log|z|=-\infty\}, and log⁡∣z∣\log|z| is a nontrivial extended classical psh function on UU.

On U∖EU\setminus E, define the current of integration over the component Z=Y∩{s>0}.Z=Y\cap\{s>0\}. Equivalently, this current is [Y][Y] locally where s>0s>0 and is zero locally where s<0s<0 or z≠0z\ne0. These neighborhoods cover U∖EU\setminus E, so it is a well-defined symmetric original-positive separately closed current SS. There is no boundary point of ZZ inside U∖EU\setminus E. Its zero extension is the actual Borel multiplication T=1{s>0}[Y]on U.T=\mathbf1_{\{s>0\}}[Y]\quad\hbox{on }U. For every original compact smooth (1,1)(1,1)-test η\eta, ∣μS,η∣(U∖E)≤∣μ[Y],η∣(U)<∞.|\mu_{S,\eta}|(U\setminus E) \leq|\mu_{[Y],\eta}|(U)<\infty. Thus the full coefficient-mass condition holds, including all test presentations, not just the radial one used below.

Choose f∈Cc∞((−λ,λ))f\in C_c^\infty((-\lambda,\lambda)) with f(0)=1f(0)=1. Choose an original smooth scalar cutoff κ(z)\kappa(z) on {∣z∣<1}\{|z|<1\}, compactly supported there and equal to one on a neighborhood of z=0z=0. One explicit construction takes a smooth function of log⁡∣z∣\log|z| on z≠0z\ne0, equal to one for log⁡∣z∣≤−λ\log|z|\leq-\lambda and zero for log⁡∣z∣≥−λ/2\log|z|\geq-\lambda/2, and gives it value one at zero. It is original smooth because it is constant on an ordinary neighborhood of zero. Now put α=κ(z)f(s)d′′s∈Ac0,1(U).\alpha=\kappa(z)f(s)d''s\in A_c^{0,1}(U). Its support is compact: it is contained in the product of a closed smaller zz-disc and a closed smaller tt-annulus. Pullback to YY gives d′α∣Y=f′(s)d′s∧d′′s.d'\alpha|_Y=f'(s)d's\wedge d''s. The original calibration of the identity tropicalization of Gm\mathbf G_m has multiplicity one. Hence the defining CLD integral, equivalently its one-dimensional Lebesgue formula, gives

⟨T,d′α⟩=∫{s>0}⊂Yf′(s)d′s∧d′′s=∫0λf′(s) ds=−1.\langle T,d'\alpha\rangle =\int_{\{s>0\}\subset Y}f'(s)d's\wedge d''s =\int_0^\lambda f'(s)\,ds=-1.

Here ff vanishes near λ\lambda. Thus d′T≠0d'T\ne0. Symmetry, or the same computation with a (1,0)(1,0)-test, also gives d′′T≠0d''T\ne0. The nonzero boundary functional is an original test functional; no additional δ\delta-test is being introduced.

For completeness, the local-completeness failure can be verified independently of the extension theorem. Let x∈Yx\in Y be the unit Gauss point of the tt-annulus. Suppose on an ordinary neighborhood WW of xx that E∩W={ϕ=−∞}E\cap W=\{\phi=-\infty\} for an extended classical psh function ϕ\phi. Choose a connected ordinary neighborhood CC of xx in Y∩WY\cap W. Such neighborhoods exist by local connectedness of analytic curves; the same fact is used in [GR25, proof of Proposition 3.5(3)]. The restriction ϕ∣C\phi|_C is extended classical psh.

The map sending rr to the radius-ere^r Gauss point is continuous. Since CC is a neighborhood of xx, it contains points of this map with sufficiently small r>0r>0, which are outside EE. Thus ϕ∣C\phi|_C is not identically −∞-\infty. But xx is a divisorial point of this smooth curve and ϕ(x)=−∞\phi(x)=-\infty, contradicting the proved Theorem “No divisorial poles” applied on CC. This proves that EE is not locally complete pluripolar. ◻

Interpretation.

The proposition is an explicit NEW_PROVED counterexample within the allowed original analytic category. Its purpose is the necessity of the complete-pole hypothesis for this extension statement. It is not a counterexample to the extension theorem with all its hypotheses. The current has already finite mass; strengthening only the mass bound cannot repair the false version.

Example (The local mass hypothesis cannot be omitted). Let U={∣t∣<1}⊂AK1,anU=\{|t|<1\}\subset\mathbf A_K^{1,\mathrm{an}} and P={0}={log⁡∣t∣=−∞}P=\{0\}=\{\log|t|=-\infty\}. For j≥1j\geq1 let xj=ζ0,∣π∣jx_j=\zeta_{0,|\pi|^j} be the corresponding Gauss point and set μ=∑j=1∞δxjon U∖P.\mu=\sum_{j=1}^{\infty}\delta_{x_j} \quad\hbox{on }U\setminus P. The points xjx_j tend to zero. Every compact subset of U∖PU\setminus P has a positive minimum for ∣t∣|t|, hence meets only finitely many of the xjx_j. Thus μ\mu is a positive Radon measure there, and defines an original top-degree (1,1)(1,1)-current; separate closedness is automatic in that degree. The closed set PP is a complete pole set.

Any positive Radon extension to UU would give infinite mass to the compact closed disc {∣t∣≤∣π∣}\{|t|\leq|\pi|\}, since that disc contains every xjx_j. This is impossible. Therefore no positive coefficient-Radon extension exists. This elementary example has remaining degree q=0q=0; it tests deletion of the mass hypothesis from the all-bidegrees theorem and is not presented as a new remaining-degree-one obstruction.

Potential quotients: normal traces and the remaining regularity gate

Fix the original absolute value on a finite extension K/QpK/\mathbf Q_p. All currents, positive cones, and compactly supported tests in this section are original CLD objects. The functions in Cb\mathcal C_b retain their actual values and the all-complete-extensions classical definition. For the normal-coordinate statements, the spaces are ordinary opens in smooth algebraic analytifications over this same KK.

The question is whether

D[u]=D[v],u,v∈Cb(U)⟹D((u−v)S)=0D[u]=D[v],\quad u,v\in\mathcal C_b(U) \quad\Longrightarrow\quad D\bigl((u-v)S\bigr)=0

for every symmetric original-positive, separately d′d'- and d′′d''-closed current SS of bidegree (p,p)(p,p), p<np<n. The full implication remains OPEN. The results below give additional backgrounds and a precise sufficient regularity statement. They do not replace the quantifier over arbitrary SS by a restricted one in the main research question.

Bounded classical weights through a normal Gauss kernel

Let BB be an ordinary open in a smooth dd-dimensional KK-variety, let Ds={∣t∣<s}\mathbb D_s=\{|t|<s\}, and write W=B×KDs,π:W⟶B,i:B⟶WW=B\times_K\mathbb D_s,\qquad \pi:W\longrightarrow B,\qquad i:B\longrightarrow W for the projection and zero section. For a∈Ba\in B and 0<r<s0<r<s, let σr(a)\sigma_r(a) be the image in WW of the radius-rr Gauss point of the H(a)\mathscr H(a)-disc over aa.

Choose a smooth convex function ψ:R→R\psi:\mathbf R\to\mathbf R which equals 00 on (−∞,−ϵ](-\infty,-\epsilon] and equals the identity on [ϵ,∞)[\epsilon,\infty), where ϵ>0\epsilon>0 is fixed. Put bN(t)=−N+ψ(log⁡∣t∣+N),βN=DbN.b_N(t)=-N+\psi(\log|t|+N),\qquad \beta_N=Db_N. The function is extended by the constant −N-N at t=0t=0. For sufficiently large NN, βN\beta_N is a smooth strongly positive closed form supported in e−N−ϵ≤∣t∣≤e−N+ϵ<se^{-N-\epsilon}\leq|t|\leq e^{-N+\epsilon}<s.

Lemma (The normal trace for a bounded classical weight). For u∈Cb(W)u\in\mathcal C_b(W) and every original η∈Acd,d(B)\eta\in A_c^{d,d}(B),

lim⁡N→∞∫Wu βN∧π∗η=∫B(u∘i)η.\lim_{N\to\infty}\int_W u\,\beta_N\wedge\pi^*\eta =\int_B (u\circ i)\eta.

The same conclusion holds for any finite real linear combination of such uu. All integrals use the prescribed Borel representatives.

Proof. Let μη\mu_\eta be the signed Radon measure of the smooth top form η\eta on BB. The product calibration and Borel coarea give

∫Wu βN∧π∗η=∫B(∫Ru(σex(a))ψ′′(x+N) dx)dμη(a).\int_W u\,\beta_N\wedge\pi^*\eta =\int_B\left(\int_{\mathbf R} u\bigl(\sigma_{e^x}(a)\bigr)\psi''(x+N)\,dx\right) d\mu_\eta(a).

Here the inner integral is restricted to the support of ψ′′(x+N)\psi''(x+N), which lies below log⁡s\log s. For clarity, this is coarea for smooth forms, with a Borel scalar coefficient, rather than an external product of arbitrary currents. On a finite moment presentation of the base top form, keep its calibrated skeleton cells separate. Adjoining the coordinate tt gives their products with the radial interval; integration on a product cell is the base calibrated measure times dxdx. The normal coefficient is ψ′′(x+N)\psi''(x+N). Borel Fubini proves (eq:r3-normal-coarea) for bounded Borel coefficients, exactly as in the cellwise argument of B, Lemma A.3. A finite partition and signed linearity recover every η\eta. Any structural degree of the base is already in μη\mu_\eta; the normal factor has mass one, since ∫ψ′′(x+N) dx=1\int\psi''(x+N)\,dx=1. The Gauss lift is continuous in (a,r)(a,r) on an affinoid base patch: for a polynomial in tt its value is ∣∑ifiti∣(σr(a))=max⁡i∣fi(a)∣ri|\sum_i f_it^i|(\sigma_r(a))=\max_i|f_i(a)|r^i, and uniform approximation gives the assertion for analytic functions on a compact radial range. Thus the Borel integrand in (eq:r3-normal-coarea) is evaluated on actual analytic points and is measurable.

Fix a∈Ba\in B. Pull uu back to the H(a)\mathscr H(a)-disc over aa. This is a locally bounded subharmonic function by the all-extensions classical hypothesis. Its maximum on a closed radius-rr disc occurs at the Gauss point, so u(i(a))≤u(σr(a)).u(i(a))\leq u(\sigma_r(a)). As r↓0r\downarrow0, σr(a)→i(a)\sigma_r(a)\to i(a) in the ordinary topology. Upper semicontinuity gives the opposite limiting upper bound. Therefore

u(σr(a))⟶u(i(a)).u(\sigma_r(a))\longrightarrow u(i(a)).

This argument concerns only the pullback of the given function to a curve. It does not assume base extension of a current equality.

Choose s0<ss_0<s and a compact base neighborhood of supp⁡η\operatorname{supp}\eta. Its product with the closed radius-s0s_0 disc is contained in a finite union of compact analytic domains in WW. Local boundedness of uu gives one bound there. For large NN, every normal radius in (eq:r3-normal-coarea) is at most s0s_0. The inner measures are probability measures supported at radii tending uniformly to zero. Equation (eq:r3-radial-actual-limit) and dominated convergence against the fixed finite measure ∣μη∣|\mu_\eta| prove (eq:r3-normal-trace). Finite real linearity proves the last assertion. ◻

This is a short NEW_PROVED extension of the continuous normal-kernel argument in R, Lemma 5.1. The additional input is the actual radial limit supplied by classical subharmonicity; weak convergence of βN\beta_N alone would not justify the weighted limit.

Proposition (Restriction of a bounded difference with positive curvature). Let f=u−vf=u-v, with u,v∈Cb(W)u,v\in\mathcal C_b(W). If D[f]D[f] is an original-positive current on WW, then D[f∘i]D[f\circ i] is an original-positive current on BB. If D[u]=D[v]D[u]=D[v], then D[u∘i]=D[v∘i]D[u\circ i]=D[v\circ i] on every original compact smooth complementary test on BB.

Consequently, if j:Z↪Xj:Z\hookrightarrow X is a smooth closed immersion of smooth KK-varieties and u,v∈Cb(U)u,v\in\mathcal C_b(U) on an ordinary U⊂XanU\subset X^{\mathrm{an}} satisfy D[u]=D[v]D[u]=D[v], then

D[u∣Zan∩U]=D[v∣Zan∩U].D[u|_{Z^{\mathrm{an}}\cap U}]=D[v|_{Z^{\mathrm{an}}\cap U}].

Proof. The assertion has no curvature content when d=0d=0. For d≥1d\geq1, take a positive original α∈Acd−1,d−1(B)\alpha\in A_c^{d-1,d-1}(B). The form βN∧π∗α\beta_N\wedge\pi^*\alpha is positive, smooth, and compactly supported in WW. Separate closedness of βN\beta_N and distributional integration by parts give 0≤⟨D[f],βN∧π∗α⟩=∫Wf βN∧π∗(Dα).0\leq\langle D[f],\beta_N\wedge\pi^*\alpha\rangle =\int_W f\,\beta_N\wedge\pi^*(D\alpha). The normal trace lemma, applied to uu and vv separately, gives the limit ⟨D[f∘i],α⟩\langle D[f\circ i],\alpha\rangle. This proves positivity. In the equality case the same computation starts with zero and applies to every original test α\alpha, without a positivity restriction; it proves the full current equality.

By R, Lemma 5.2, each point of the smooth immersion has one ordinary analytic product neighborhood with a normal polydisc over an ordinary neighborhood in ZanZ^{\mathrm{an}}. Apply the zero-section result successively to its normal coordinates. At every step the restricted functions remain actual locally bounded classical psh functions by their definition. Locality gives (eq:r3-subvariety-restriction). No equality on tests after an arbitrary extension of KK has been used or asserted. ◻

Corollary (Backgrounds carried by a smooth subvariety). In the last setting, put h=u−vh=u-v. Then D(h[Zan∩U])=0.D(h[Z^{\mathrm{an}}\cap U])=0. More generally the same conclusion holds for S=j∗QS=j_*Q when, locally on Zan∩UZ^{\mathrm{an}}\cap U, QQ is a finite linear combination of bounded-potential ambient BT products on ZZ, or is a smooth separately closed form current. The pushforward is taken only in the degrees where it is defined. In particular this includes every positive closed background of either displayed kind.

Proof. Borel multiplication and the proper local pushforward satisfy h j∗Q=j∗((h∣Z)Q)h\,j_*Q=j_*((h|_Z)Q) on all original tests, with the original calibration. The restricted curvature equality is (eq:r3-subvariety-restriction). For ambient bounded-potential products on ZZ, the symmetry argument of Lemma “Ambient mixed backgrounds” puts that curvature in the innermost differentiated slot; for a smooth closed form, use the distributional Leibniz rule. Thus D((h∣Z)Q)=0D((h|_Z)Q)=0. Pushforward commutes with DD, proving the claim. The first assertion is the case Q=[1]Q=[1]. ◻

The corollary is a DIRECT_COROLLARY of the normal trace and the previously proved restricted quotient statements. The relative background [Z][Z] is allowed to charge the ambient pluripolar set ZZ; ambient nonconcentration has not been imposed.

A further closure class supplied by R2

Corollary (Bounded ambient BT weak closures). Let u,v∈Cb(U)u,v\in\mathcal C_b(U) satisfy D[u]=D[v]D[u]=D[v]. Suppose that locally on each fixed norm-box domain one has Sj=T[1],p(a1,j,…,ap,j)⟶S,0≤ai,j≤1,0≤p<n,S_j=T_{[1],p}(a_{1,j},\ldots,a_{p,j})\longrightarrow S, \qquad 0\leq a_{i,j}\leq1, \qquad 0\leq p<n, on all original compact smooth tests. Then D((u−v)S)=0D((u-v)S)=0. No monotonicity or convergence of the generating potentials is required.

Proof. For every jj, ambient BT symmetry gives D(uSj)=D(vSj)D(uS_j)=D(vS_j). Corollary “Fixed additional BT slots on a weak closure”, applied separately to the fixed weights uu and vv, gives uSj→uSuS_j\to uS and vSj→vSvS_j\to vS. Distributional differentiation passes the equality to the limit. The local argument glues. ◻

This is a DIRECT_COROLLARY of R2, not a regularization theorem for arbitrary currents. The uniform bounded-potential family provides the capacity control in that result. Neither an arbitrary positive closed current nor a current supported on an ambient polar set has been shown to admit such an approximation.

Continuous representatives: a precise sufficient gate

Lemma (A continuous current representative is the actual difference). Let u,v∈Cb(U)u,v\in\mathcal C_b(U) and let HH be a finite continuous function on UU. If [u]−[v]=[H][u]-[v]=[H] on every original compact smooth top-degree test, then u−v=Hu-v=H at every point of UU. No classical-psh hypothesis on HH is required.

Proof. Work in a framed affine ordinary neighborhood as in Section “Detection of strict gaps and actual point values”. At a prescribed divisorial point zz, choose its continuous model tester aa with (Da)n({z})>0(Da)^n(\{z\})>0. On one smaller neighborhood, its pure-FS construction gives smooth classical psh aj↓aa_j\downarrow a. For every original compact smooth scalar χ\chi, ∫χ(u−v−H)(Daj)n=0.\int \chi(u-v-H)(Da_j)^n=0. The fixed-background weighted decreasing theorem passes the uu and vv terms to their actual Borel limits. The HH term passes by continuity of the fixed multiplier and weak convergence of (Daj)n(Da_j)^n. Hence (u−v−H)(Da)n=0(u-v-H)(Da)^n=0 as a signed Radon measure. Its atom at zz gives u(z)−v(z)=H(z)u(z)-v(z)=H(z).

This holds at every divisorial point in the Shilov determining family of the neighborhood. The function v+Hv+H is finite and upper semicontinuous, although it has not been declared psh. Lemma “A strict gap contains a Shilov point”, applied to the classical function uu and the upper-semicontinuous function v+Hv+H, gives u≤v+Hu\leq v+H everywhere. Apply the same lemma to vv and u−Hu-H for the reverse inequality. Locality finishes the proof. ◻

The lemma is a DIRECT_COROLLARY of weighted convergence, the divisorial testers, and Shilov determination. It is a recovery statement after a continuous representative has been supplied.

Proposition (A sufficient local regularity statement). Assume the following statement on the permitted analytic spaces:

For u,v∈Cb(U)u,v\in\mathcal C_b(U) with D[u]=D[v]D[u]=D[v], every point has an ordinary neighborhood WW and a finite continuous HWH_W such that [u]−[v]=[HW][u]-[v]=[H_W] on WW.

Then (eq:r3-general-potential-quotient) holds for every allowed positive separately closed background SS.

Proof. Lemma “A continuous current representative is the actual difference” gives u−v=HWu-v=H_W at the actual points of WW. Distributional differentiation gives D[HW]=0D[H_W]=0. The continuous-harmonic multiplier lemma from R1, based on R, Theorem 2.9, then gives D(HWS)=0D(H_WS)=0 on WW. Indeed HWH_W and −HW-H_W are locally uniform limits of smooth functions with positive curvature forms; multiplying those forms by SS, passing the uniform limit through its coefficient-Radon measures, and using the pointed positive cone gives both signs of the required zero current. These local equalities glue. ◻

The proposition is CONDITIONAL; the displayed existence of a continuous representative remains OPEN. This is strictly a sufficient reduction, not a claimed equivalence. The known faithfulness of u↦[u]u\mapsto[u] does not construct HWH_W. Likewise u−vu-v is not a psh slot merely because both uu and vv are psh.

Remark (The external harmonicity statements have different scopes). In CLD’s 2025 revision, Section 7.8 begins with a boundaryless tropical space of level nn. Definition 7.8.2 defines its harmonic sheaf using locally real linear combinations of paralinear harmonic functions. Remark 7.8.3 discusses both the possible gap between that sheaf and continuous functions with D[u]=0D[u]=0, and the unresolved continuity, up to current-equivalence, of locally integrable functions with D[u]=0D[u]=0 in that tropical setting. It is not an established theorem about the present bounded differences of classical psh functions over KK, and it is not used as a proof that our specific gate is equivalent to that broader problem.

Similarly, GR, Definition 7.1 and Proposition 7.2, make a function pluriharmonic when it and its negative are classically psh; its continuity then follows from the two upper-semicontinuity conditions. The current equality D[u]=D[v]D[u]=D[v] has not been shown to put either sign of u−vu-v in that class. These are source comparisons, not substitute regularity proofs.

A false extension to arbitrary ambient-capacitary representatives

Example (Ambient current zero does not imply a relative multiplier zero). Let R>1R>1 and U={∣z∣<R, ∣t∣<R}⊂(AK2)an,Y={z=0}∩U,q(t)=max⁡(0,log⁡∣t∣).U=\{|z|<R,\ |t|<R\}\subset(\mathbf A^2_K)^{\mathrm{an}}, \qquad Y=\{z=0\}\cap U, \qquad q(t)=\max(0,\log|t|). Define the actual Borel function f(x)={q(t(x)),x∈Y,0,x∉Y.f(x)=\begin{cases}q(t(x)),&x\in Y,\\0,&x\notin Y.\end{cases} It is bounded and upper semicontinuous: YY is closed, qq is continuous and nonnegative, and the outside value is zero. It is locally ambient-capacity-quasicontinuous, since it vanishes off the analytic polar set YY, which has local outer ambient capacity zero. It represents the zero class in the ambient capacitary completion.

Every original smooth top-degree trace measure on UU gives zero mass to YY. Hence [f]=0[f]=0 and D[f]=0D[f]=0. Nevertheless, with the original positive separately closed background S=[Y]S=[Y], D(fS)=i∗D[q]=δ(0,ζ0,1)≠0,D(fS)=i_*D[q]=\delta_{(0,\zeta_{0,1})}\ne0, where ζ0,1\zeta_{0,1} is the unit Gauss point of the tt-disc. The atom has mass one in the original KK calibration, by the one-variable model formula for max⁡(0,log⁡∣t∣)\max(0,\log|t|).

This does not refute (eq:r3-general-potential-quotient): ff cannot be a difference a−ba-b of two members of Cb(U)\mathcal C_b(U). Otherwise [a]=[b][a]=[b] and R2 current-faithfulness would force a=ba=b at every point, whereas ff is nonzero on part of YY. The example instead proves FALSE the shortcut which replaces the actual difference hypothesis by boundedness, upper semicontinuity, and ambient quasicontinuity. Its ingredients are direct consequences of the already proved polar and model measure facts; it is not counted as a new BT construction.

What the normal trace does not supply

The normal-kernel proof gives restriction along a smooth immersion over the original field. It does not establish that an equality of original currents pulls back to every complete field extension, or to an arbitrary analytic curve map after such an extension. Nor does it approximate an arbitrary positive closed current by smooth-subvariety integration currents with multiplier control. The bounded-difference restriction statement and the continuous representative gate are therefore kept separate. The unresolved step in (eq:r3-general-potential-quotient) is not hidden in an unproved change of test category or in an exchange of two discontinuous limits.

The projection route for potential quotients: a precise test gate

This section supplements Section “Potential quotients: normal traces and the remaining regularity gate”. It does not resolve its unrestricted potential-independence question. All spaces, analytic products, and absolute values are over the original finite extension K/QpK/\mathbf Q_p. All tests are original CLD smooth tests. Let UU be an allowed ordinary smooth nn-dimensional algebraic analytic neighborhood, n≥1n\geq1, let Ds\mathbb D_s be an ordinary disc, and put W=U×KDsW=U\times_K\mathbb D_s, with projection π\pi. Suppose u,v∈Cb(U)u,v\in\mathcal C_b(U) and D[u]=D[v]D[u]=D[v], and set f=u−vf=u-v. The functions u∘π,v∘πu\circ\pi,v\circ\pi are classical by the all-extensions definition. Pulling back the equality of currents is a separate question.

Proposition (A sufficient projected-test criterion). For α∈Acn,n(W)\alpha\in A_c^{n,n}(W) define the compactly supported ordinary current Pα=π∗[α]P_\alpha=\pi_*[\alpha] of bidegree (n−1,n−1)(n-1,n-1). The pushforward is defined because the projection is proper on the support of α\alpha. The current να=DPα=π∗[Dα]\nu_\alpha=DP_\alpha=\pi_*[D\alpha] is a finite signed Radon measure, and

⟨D[f∘π],α⟩=∫Uf dνα.\langle D[f\circ\pi],\alpha\rangle =\int_U f\,d\nu_\alpha.

Suppose, as a compact current on UU, that Pα=∑ℓ=1Ncℓ[ηℓ],ηℓ∈Acn−1,n−1(U),P_\alpha=\sum_{\ell=1}^N c_\ell[\eta_\ell],\qquad \eta_\ell\in A_c^{n-1,n-1}(U), where the cℓc_\ell are continuous and bounded on the relevant compact supports. Then ⟨D[f∘π],α⟩=0\langle D[f\circ\pi],\alpha\rangle=0.

Status: DIRECT_COROLLARY. The conclusion for every α\alpha is CONDITIONAL on the displayed representation, or an equally strong pairing-continuity statement, for every projected test.

Proof. The first identity is the definition of DD on the locally bounded Borel function f∘πf\circ\pi, followed by pushforward of the fixed signed measure DαD\alpha. It uses no pullback of nonsmooth currents. Use a finite original smooth partition on the compact support to reduce to a neighborhood on which the accepted regularization supplies smooth psh uj↓uu_j\downarrow u and vj↓vv_j\downarrow v, with uniform local bounds. Write fj=uj−vjf_j=u_j-v_j. Since DujDu_j and DvjDv_j converge weakly to the same positive current, the fixed-continuous-multiplier theorem of R gives ∑ℓ∫UcℓDfj∧ηℓ⟶0.\sum_\ell\int_U c_\ell Df_j\wedge\eta_\ell\longrightarrow0. For the smooth functions fjf_j, ordinary smooth pullback and Stokes identify this sum with ⟨Pα,Dfj⟩=∫Ufj dνα.\langle P_\alpha,Df_j\rangle =\int_U f_j\,d\nu_\alpha. Bounded convergence against the one fixed finite signed measure να\nu_\alpha gives the asserted zero limit. If a representation is initially stated only near the support, multiply it by a compact cutoff equal to one there before using it as an identity of compact currents. The finite partition reduction preserves continuity of the coefficients and the degrees of the test forms. ◻

Lemma (Separated radial tests). Let I⋐(−∞,log⁡s)I\Subset(-\infty,\log s), x=log⁡∣t∣x=\log|t|, and let γ\gamma be a compactly supported Lagerberg form of bidegree (r,s′)(r,s') on II, where r,s′∈{0,1}r,s'\in\{0,1\}. If η∈Acn−r,n−s′(U)\eta\in A_c^{n-r,n-s'}(U), then the form α=π∗η∧γ(x),\alpha=\pi^*\eta\wedge\gamma(x), extended by zero off the annulus, satisfies ⟨D[f∘π],α⟩=0\langle D[f\circ\pi],\alpha\rangle=0. The same holds for finite sums, limits in the original compact test topology, and analytic changes of centre t′=t−a(x)t'=t-a(x) for which aa is analytic and ∣a∣<s|a|<s on a neighborhood of the base support.

Status: DIRECT_COROLLARY.

Proof. Use separate moment coordinates on the base and the radial annulus. On each calibrated product cell the integration formula is iterated integration in the base coordinates and the radial coordinate. Consequently it is valid also for the bounded Borel multiplier pulled back from ff: first approximate a bounded Borel function by simple functions in the resulting fixed coefficient measures.

If r=s′=1r=s'=1, the only contribution is the base pairing ⟨D[f],η⟩\langle D[f],\eta\rangle times the radial integral of γ\gamma, and is zero. If r=s′=0r=s'=0, the only possible contribution contains DγD\gamma and its integral is zero by one-dimensional Stokes. In the remaining two bidegrees the possible cross term has radial factor d′γd'\gamma or d′′γd''\gamma, again with zero integral. Degree considerations eliminate the other terms. This proves the claim without differentiating ff. Finite sums and convergence in the test topology preserve the equality by the definition of a current. Under the stated bound on aa, translation is an analytic automorphism of the disc over the relevant base neighborhood and preserves π\pi; applying the same computation in t′t' proves the last assertion. No assertion that these tests generate all original tests on the analytic product is made. ◻

Example (The analytic product is not the topological product). The assertion that products of functions of the two factors are dense in all original smooth functions on the Berkovich analytic product is FALSE. Choose a uniformizer ϖ\varpi, and work in an ordinary bidisc of radius R>1R>1 with coordinates (x,t)(x,t). Let z0z_0 be the two-variable unit Gauss point. Let z1z_1 be the image of the two-variable unit Gauss point in (X,S)(X,S) under x=X,t=X+ϖS.x=X,\qquad t=X+\varpi S. Both points project to the unit Gauss point in each factor. Indeed, for a polynomial PP, the Gauss norm of P(X+ϖS)P(X+\varpi S) is at most the maximum coefficient norm of PP, and its coefficient of S0S^0 is P(X)P(X), which gives the opposite inequality. However, ∣t−x∣(z0)=1,∣t−x∣(z1)=∣ϖ∣.|t-x|(z_0)=1,\qquad |t-x|(z_1)=|\varpi|. Thus the original smooth function log⁡∣t−x∣\log|t-x| on the ordinary nonvanishing locus of t−xt-x separates the two points. Multiplying by an original compact cutoff equal to one near both gives a compact smooth separating function. Every finite sum of products of functions from the two factors, and every uniform limit of such sums on the compact set {z0,z1}\{z_0,z_1\}, takes equal values at the two points. Equivalently, the nonzero top-degree current δz0−δz1\delta_{z_0}-\delta_{z_1} annihilates every such tensor test.

The example disproves this density shortcut, not the desired projection-kernel implication. It also does not contradict the translated radial tests, which can involve t−xt-x.

What the published projection results supply.

CLD’s 2025 manuscript, Section 9.1, constructs the product of its topological tropical spaces. Its factor-generated refinements in Section 9.2 (in particular Corollary 9.2.4), and the tensor-current construction and differential identity in 9.4.4 and 9.4.8, have that product as their domain. The preceding example prevents their factor-density conclusion from being substituted for the missing assertion on the analytic fiber product.

Mihatsch, δ\delta-Forms on Lubin–Tate spaces, arXiv:2112.10018v4, Theorem 3.23, printed page 58, provides flat refined skeletons for locally dominant analytic morphisms. Definition 2.18, printed page 22, is formulated with piecewise smooth fiber integrals. Example 2.45(3), printed pages 41–42, also illustrates that a balanced pushforward need not be a δ\delta-form. It is an abstract sheaf-level warning, not a counterexample for the present disc projection. These results do not state the continuous-coefficient original-test representation in Proposition “A sufficient projected-test criterion”. Gubler–Künnemann, Section 6.10, concerns proper pushforward of δ\delta-currents and likewise does not prove that representation.

Status: FAILED_ATTEMPT. The inference “flat after refinement, hence an admissible projected original smooth test” is not justified. The missing statement is pairing continuity for the projected test through cell changes and refinements, in exactly the original sheaf.

The conditional graph consequence.

Suppose the projection-kernel implication has been established for iterated disc products. Let CC be an ordinary open in a smooth algebraic KK-curve, with a closed immersion ι:C↪Dm\iota:C\hookrightarrow\mathbb D^m, and let g:C→Ug:C\to U be an analytic map. Assume explicitly that the graph (g,ι)(g,\iota) is locally in the smooth algebraic closed-immersion scope of the already proved normal-trace restriction theorem; for example, both maps may be induced by algebraic maps with an algebraic closed embedding for ι\iota. This hypothesis is stronger than merely requiring an arbitrary analytic closed immersion. Its graph is then a permitted smooth closed immersion into U×KDmU\times_K\mathbb D^m. Applying that normal-trace theorem to the pulled-back kernel identity would give D[f∘g]=0D[f\circ g]=0. This is a CONDITIONAL same-KK consequence. It neither supplies the kernel identity nor changes the ground field; it therefore does not by itself verify the all-complete-extensions curve test in Gubler–Rabinoff, Definition 4.1.

Proposition (The positive square of a kernel difference). Under D[u]=D[v]D[u]=D[v], the established bounded energy calculus gives

Γ[1](f)=12D[f2]−fD[f]=12D[f2]≥0.\Gamma_{[1]}(f) =\tfrac12D[f^2]-fD[f] =\tfrac12D[f^2]\geq0.

For every fixed smooth positive closed (1,1)(1,1)-form β\beta and every nonnegative compact smooth χ\chi on its domain, 0≤⟨Γ[1](f),χβn−1⟩=12∫f2Dχ∧βn−1≤12∥f∥∞,supp⁡Dχ2∫∣Dχ∧βn−1∣.0\leq\langle\Gamma_{[1]}(f),\chi\beta^{n-1}\rangle =\tfrac12\int f^2D\chi\wedge\beta^{n-1} \leq\tfrac12\|f\|_{\infty,\operatorname{supp}D\chi}^2 \int|D\chi\wedge\beta^{n-1}|.

Status: DIRECT_COROLLARY.

Proof. On a regularization neighborhood use smooth decreasing psh rows uj,vju_j,v_j and fj=uj−vjf_j=u_j-v_j. The smooth currents d′fj∧d′′fjd'f_j\wedge d''f_j are positive. Expand fjDfjf_jDf_j into its four weighted psh terms; the established fixed-background weighted monotone theorem gives the limit fD[f]fD[f]. Bounded convergence gives D[fj2]→D[f2]D[f_j^2]\to D[f^2]. The smooth square identity thus gives the first formula and its positivity, which glue by locality. Pairing with χβn−1\chi\beta^{n-1} and using that β\beta is separately closed gives the second formula and the displayed elementary bound. ◻

This estimate is not zero energy. On a relatively compact annulus, f=log⁡∣t∣f=\log|t|, u=f+Cu=f+C and v=Cv=C are bounded classical potentials and D[f]=0D[f]=0, whereas Γ[1](f)=d′log⁡∣t∣∧d′′log⁡∣t∣\Gamma_{[1]}(f)=d'\log|t|\wedge d''\log|t| has positive radial integral. Nor is the Borel function f2f^2 shown continuous by the positive-square identity. Consequently R’s continuous-potential input cannot be applied to it without a separate regularity proof. CLD 2025, Remark 7.8.3, is in its enlarged tropical test category; Gubler–Rabinoff 2025, Definition 7.1 and Proposition 7.2, concern functions whose two signs satisfy the classical psh conditions. Neither statement turns the original current equation D[f]=0D[f]=0 alone into that hypothesis.

Remaining gate.

The unrestricted implication D[u]=D[v]⇒D((u−v)S)=0D[u]=D[v]\Rightarrow D((u-v)S)=0 for an arbitrary allowed positive closed background remains OPEN. The proved projected-test criterion, the radial cases, and the positive-square identity are retained; neither the tensor-density shortcut nor a formal passage to another test sheaf closes the gap.

Relative compact regularity: a linear-degree reduction and a contact obstruction

The field, actual classical class Cb\mathcal C_b, original CLD tests, and original-positive cone remain those of the first two records. The background is fixed throughout. No equality on additional δ\delta-tests is asserted. The target of this section is the sharp inequality

ν(K)≤Cap⁡S(K;Ω)(RC)\nu(K)\leq\operatorname{Cap}_S(K;\Omega) \tag{RC}

for weak limits of normalized top relative BT measures on one fixed domain. We do not prove (RC) for an arbitrary SS. We prove a representation and an exact reduction when the remaining degree is one, and give counterexamples to a proposed contact-measure proof in every intermediate degree.

External comparison.

For complex positive closed currents, Dabbek–Elkhadhra define the analogous capacity in [DE06, Definition 2.1, p. 471]. Their list of open problems on p. 472 explicitly asks for continuity on decreasing compact sets, equality with a contact-constrained capacity, and attainment of the capacity of a compact set by one potential. This is an EXTERNAL_KNOWN statement about that paper; it is not a claim that all these questions remain open in the present literature. In particular, that paper cannot be cited as already proving the compact outer regularity sought here.

An energy estimate for one remaining degree

For this subsection let SS have bidegree (n−1,n−1)(n-1,n-1). It is arbitrary within the fixed original-positive, symmetric, separately closed class. Thus CΩ′(E)=Cap⁡S(E;Ω′)=sup⁡w∈Cb(Ω′)0≤w≤1D(wS)(E).C_{\Omega'}(E)=\operatorname{Cap}_S(E;\Omega') =\sup_{\substack{w\in\mathcal C_b(\Omega')\\0\leq w\leq1}} D(wS)(E). The relative energy measure ΓS(f,g)\Gamma_S(f,g) for differences and real linear combinations of bounded classical potentials is the one in Lemma “Relative energy measures”. For an original smooth real scalar χ\chi we also use ΓS(χ)=d′χ∧d′′χ∧S.\Gamma_S(\chi)=d'\chi\wedge d''\chi\wedge S. This is a positive measure: the (1,1)(1,1)-factor is strongly positive, so pairing it with the specified original-positive SS is legitimate. The cutoff is not being assumed psh.

Choose permitted local stages V⋐Ω′⋐Ω,V\Subset\Omega'\Subset\Omega, with VV ordinary open, and original smooth cutoffs 0≤χ,κ≤10\leq\chi,\kappa\leq1 compactly supported in Ω′\Omega' such that χ=1\chi=1 near V‾\overline V and κ=1\kappa=1 near supp⁡χ\operatorname{supp}\chi. The cutoffs may be chosen with support inside a smaller compact norm box, so prescribed-domain regularization applies near their entire supports. Put

Nχ(f)2=∫χ dΓS(f)+∫f2 dΓS(χ).N_\chi(f)^2 =\int\chi\,d\Gamma_S(f) +\int f^2\,d\Gamma_S(\chi).

This is the square of a positive semidefinite quadratic seminorm on the real span of bounded classical potentials. The associated bilinear form is obtained by polarization.

Lemma (A fixed-energy bound for relative capacity). For the preceding stage there is a finite constant Aχ,κ,SA_{\chi,\kappa,S} such that, whenever a,b∈Cb(Ω)a,b\in\mathcal C_b(\Omega) satisfy 0≤a,b≤10\leq a,b\leq1, f=a−bf=a-b, and δ>0\delta>0, one has

∫χf2 D(wS)≤Aχ,κ,SNχ(f)(w∈Cb(Ω′), 0≤w≤1),\begin{aligned} \int\chi f^2\,D(wS) &\leq A_{\chi,\kappa,S} N_\chi(f) &&(w\in\mathcal C_b(\Omega'),\ 0\leq w\leq1), \end{aligned}

CΩ′∗(V∩{∣f∣>δ})≤Aχ,κ,Sδ−2Nχ(f).\begin{aligned} C_{\Omega'}^*(V\cap\{|f|>\delta\}) &\leq A_{\chi,\kappa,S}\delta^{-2}N_\chi(f). \end{aligned}

The same bound holds with any compact subset of VV in place of VV. The constant is independent of a,b,wa,b,w and of all regularization indices.

Proof. First take smooth a,b,wa,b,w. For compactly supported expressions, Stokes and symmetry of SS give

∫χf2D(wS)=−2∫χf dΓS(f,w)−∫f2 dΓS(χ,w).\int\chi f^2D(wS) =-2\int\chi f\,d\Gamma_S(f,w) -\int f^2\,d\Gamma_S(\chi,w).

There is no boundary term. The positive 2×22\times2 matrices of energy measures give their measure-valued Cauchy–Schwarz inequality, also after bounded Borel multiplication. Since χ≤κ\chi\leq\kappa and κ=1\kappa=1 near the support of dχd\chi, it follows that

∣∫χf dΓS(f,w)∣≤∥f∥∞(∫χ dΓS(f))1/2(∫κ dΓS(w))1/2,∣∫f2 dΓS(χ,w)∣≤∥f∥∞(∫f2 dΓS(χ))1/2(∫κ dΓS(w))1/2.\begin{aligned} \left|\int\chi f\,d\Gamma_S(f,w)\right| &\leq \|f\|_\infty \left(\int\chi\,d\Gamma_S(f)\right)^{1/2} \left(\int\kappa\,d\Gamma_S(w)\right)^{1/2},\\ \left|\int f^2\,d\Gamma_S(\chi,w)\right| &\leq \|f\|_\infty \left(\int f^2\,d\Gamma_S(\chi)\right)^{1/2} \left(\int\kappa\,d\Gamma_S(w)\right)^{1/2}. \end{aligned}

All norms in this proof can be taken on the fixed compact neighborhood of the cutoff supports.

The energy of a normalized nonnegative tester is uniformly bounded. Indeed, the defining identity for Γ\Gamma gives ∫κ dΓS(w)=12⟨S,w2Dκ⟩−∫κw D(wS).\int\kappa\,d\Gamma_S(w) =\frac12\langle S,w^2D\kappa\rangle -\int\kappa w\,D(wS). The last integral is nonnegative. The first is bounded in absolute value by the variation of the fixed coefficient measure μS,Dκ\mu_{S,D\kappa} times ∥w∥∞2/2\|w\|_\infty^2/2. This measure is finite on the compact stage by the accepted coefficient-Radon property. In particular one finite bound Mκ,SM_{\kappa,S} works for all 0≤w≤20\leq w\leq2 near the stage. Combining the preceding inequalities yields, in the smooth case, ∫χf2D(wS)≤5Mκ,S ∥f∥∞Nχ(f).\int\chi f^2D(wS) \leq \sqrt{5M_{\kappa,S}}\,\|f\|_\infty N_\chi(f).

For general bounded classical a,b,wa,b,w, use the simultaneous smooth decreasing rows of Lemma “Regularization on a prescribed smaller norm box” on one neighborhood of the whole compact stage. After discarding a prefix, each row lies in [0,2][0,2] there. To see the upper bound, the continuous approximants decrease to a function at most one; their open sublevel sets below two form an increasing cover of the compact stage. A finite subcover gives the uniform index. The smooth differences therefore have absolute value at most two.

Proposition “Functional weighted convergence”, applied to F(x,y)=(x−y)2F(x,y)=(x-y)^2, passes the left side to ∫χf2D(wS)\int\chi f^2D(wS). The first summand of Nχ(f)2N_\chi(f)^2 converges by the distributional formula for Γ\Gamma and the weighted fixed-background theorem. The second converges by bounded convergence against the fixed measure ΓS(χ)\Gamma_S(\chi). Passing the smooth inequality proves (eq:r3-energy-square-bound), for example with Aχ,κ,S=25Mκ,SA_{\chi,\kappa,S}=2\sqrt{5M_{\kappa,S}}. This argument uses no continuity theorem with a varying background.

Since χ=1\chi=1 on VV, Chebyshev’s inequality and the supremum over normalized ww give the estimate for the raw capacity of V∩{∣f∣>δ}V\cap\{|f|>\delta\}. Both aa and bb are relatively quasicontinuous. That strict level set on the ordinary core VV is therefore quasi-open, and Proposition “Quasi-open sets are outer regular” identifies its raw and outer capacities. This proves (eq:r3-energy-capacity-bound). Monotonicity gives the compact version. In particular no strict level set has been silently treated as ordinarily open. ◻

Status.

Lemma “A fixed-energy bound for relative capacity” is NEW_PROVED. Its inputs are the established original-cone energy measures, weighted monotone calculus, coefficient-Radon control, and quasicontinuity. It needs no global strictly psh exhaustion and no compact outer regularity.

Convex blocks give actual capacitary limits

The following class preserves the normalization on the original domain; it is narrower than merely asking that an arbitrary member of AS1(Ω)\mathcal A_S^1(\Omega) take values in [0,1][0,1].

Definition. Let Ω\Omega be a permitted ordinary algebraic norm-box domain. Write AS[0,1](Ω)\mathcal A_S^{[0,1]}(\Omega) for the Borel functions a:Ω→[0,1]a:\Omega\to[0,1] admitting a sequence aj∈Cb(Ω),0≤aj≤1 on Ω,a_j\in\mathcal C_b(\Omega),\qquad 0\leq a_j\leq1\text{ on }\Omega, converging to aa locally in outer SS-capacity in the all-domain sense of Definition “Local outer-capacity convergence”. Actual Borel representatives are retained. Its BT calculus is the restriction of Theorem “BT calculus on the bounded capacitary closure”.

Theorem (Linear-degree convex-block compactness). Let SS have bidegree (n−1,n−1)(n-1,n-1) on the fixed norm-box domain Ω\Omega. Given any sequence uj∈Cb(Ω)u_j\in\mathcal C_b(\Omega) with 0≤uj≤10\leq u_j\leq1, there exist a subsequence (ujl)(u_{j_l}), finite convex combinations vm∈conv⁡{ujl:l≥m},v_m\in\operatorname{conv}\{u_{j_l}:l\geq m\}, and a∈AS[0,1](Ω)a\in\mathcal A_S^{[0,1]}(\Omega) such that vm→av_m\to a locally in outer SS-capacity on every permitted compact/domain pair in Ω\Omega.

If in addition D(ujS)→νD(u_jS)\to\nu locally weakly on all original compact smooth scalar tests, then

ν=D(aS),D(vmS)⟶D(aS),vmD(vmS)⟶aD(aS).\nu=D(aS),\qquad D(v_mS)\longrightarrow D(aS),\qquad v_mD(v_mS)\longrightarrow aD(aS).

The last convergence is for the convex blocks; it is not a weighted convergence assertion about the original sequence.

Proof. Choose the countable cofinal pairs Vl⋐Ωl⋐ΩV_l\Subset\Omega_l\Subset\Omega used in Proposition “Existence of capacitary limits”, and cutoffs χl,κl\chi_l,\kappa_l as in the preceding lemma. Let V\mathscr V be the real span of all normalized classical potentials on Ω\Omega. For each ll, write Nl=NχlN_l=N_{\chi_l}. Positivity of Γ\Gamma and the elementary L2L^2 term in (eq:r3-cutoff-energy-norm) show that NlN_l is a Hilbert seminorm on V\mathscr V.

There is a finite BlB_l such that Nl(u)2≤Bl(u∈Cb(Ω), 0≤u≤1).N_l(u)^2\leq B_l \quad(u\in\mathcal C_b(\Omega),\ 0\leq u\leq1). For the first summand use the cutoff energy identity from the preceding proof with χl\chi_l in place of κ\kappa; for the second use u2≤1u^2\leq1 and the finite mass of ΓS(χl)\Gamma_S(\chi_l). Define

∥f∥H2=∑l=1∞2−l1+BlNl(f)2.\|f\|_{\mathcal H}^2 =\sum_{l=1}^{\infty} \frac{2^{-l}}{1+B_l}N_l(f)^2.

This is finite on V\mathscr V: for f=∑i=1rciuif=\sum_{i=1}^r c_i u_i, the seminorm triangle inequality gives Nl(f)2≤(∑i∣ci∣)2BlN_l(f)^2\leq(\sum_i|c_i|)^2 B_l. Quotient by its null space and complete to obtain a real Hilbert space H\mathcal H. Each normalized uju_j has norm at most one. This construction makes no claim that a Hilbert vector is already a pointwise function or that a separately chosen Dirichlet form is closable in a trace L2L^2 space.

The closed span of the sequence in H\mathcal H is separable. Weak sequential compactness of its unit ball gives a subsequence ujlu_{j_l} converging weakly to h∈Hh\in\mathcal H. For every mm, hh belongs to the norm closure of the convex hull of the tail {ujl:l≥m}\{u_{j_l}:l\geq m\}. Here is the needed elementary Hilbert argument. If hh were outside that closed convex set and pp its nearest point, then ⟨h−p,v−p⟩≤0for every v in the set.\langle h-p,v-p\rangle\leq0 \quad\text{for every }v\text{ in the set}. Passing to the weak limit along its tail would give ∥h−p∥2≤0\|h-p\|^2\leq0, a contradiction. Consequently choose finite convex combinations vmv_m as stated with ∥vm−h∥H<2−m\|v_m-h\|_{\mathcal H}<2^{-m}.

The vmv_m are actual classical potentials on the whole original Ω\Omega, still in [0,1][0,1]. For each fixed ll, Nl(vm−vr)⟶0(m,r→∞).N_l(v_m-v_r)\longrightarrow0\quad(m,r\to\infty). Lemma “A fixed-energy bound for relative capacity” makes the sequence Cauchy in outer capacity on VlV_l with capacity domain Ωl\Omega_l. For any permitted H⋐Ω′⊂ΩH\Subset\Omega'\subset\Omega, finitely many such VlV_l cover HH, with Ωl⊂Ω′\Omega_l\subset\Omega'. Finite subadditivity and domain monotonicity therefore give the Cauchy property on this exact pair (H,Ω′)(H,\Omega'). Thus it holds on all permitted pairs, not only for one larger capacity domain.

Proposition “Existence of capacitary limits” now supplies an actual Borel limit locally in outer capacity. Its construction takes a pointwise limsup of a subsequence of the vmv_m, so the resulting representative has values in [0,1][0,1] at every point. It is in AS[0,1](Ω)\mathcal A_S^{[0,1]}(\Omega) by the very same normalized approximants. This step supplies the pointwise representative which the abstract Hilbert argument alone would not supply.

Suppose finally that D(ujS)→νD(u_jS)\to\nu. Linearity in one differentiated slot makes D(vmS)D(v_mS) the corresponding finite convex combination of tail measures. For each fixed original test its scalar pairings converge to the pairing with ν\nu. The completed calculus instead identifies their limit as D(aS)D(aS), hence ν=D(aS)\nu=D(aS). Its weighted capacity-continuity theorem also gives the last limit in (eq:r3-linear-weak-representation). All conclusions use the fixed SS and all original tests. ◻

Status.

Theorem “Linear-degree convex-block compactness” is NEW_PROVED. It holds in every ambient dimension, but its remaining degree is specifically one. Convex combinations are essential: the theorem does not contradict the normalized family without a capacity-Cauchy subsequence in Proposition “Failure of capacity precompactness and pointwise weighted stability”.

What the representation does and does not settle

Define, for Borel E⊂ΩE\subset\Omega, CA(E;Ω)=sup⁡a∈AS[0,1](Ω)D(aS)(E).C_{\mathcal A}(E;\Omega) =\sup_{a\in\mathcal A_S^{[0,1]}(\Omega)}D(aS)(E). There is no enlargement of the capacity domain in this definition.

Corollary (The outer capacity on compact sets is the completed capacity). In the degree-one setting of Theorem “Linear-degree convex-block compactness”,

CΩ(E)≤CA(E;Ω)≤CΩ∗(E)(E Borel),\begin{aligned} C_\Omega(E)&\leq C_{\mathcal A}(E;\Omega) \leq C_\Omega^*(E) &&(E\text{ Borel}),\end{aligned}

CA(K;Ω)=CΩ∗(K)(K⋐Ω compact).\begin{aligned} C_{\mathcal A}(K;\Omega)&=C_\Omega^*(K) &&(K\Subset\Omega\text{ compact}). \end{aligned}

Consequently raw compact outer regularity for the original classical capacity is equivalent here to the no-gain assertion

sup⁡a∈AS[0,1](Ω)D(aS)(K)≤sup⁡u∈Cb(Ω)0≤u≤1D(uS)(K)(K⋐Ω).\sup_{a\in\mathcal A_S^{[0,1]}(\Omega)}D(aS)(K) \leq \sup_{\substack{u\in\mathcal C_b(\Omega)\\0\leq u\leq1}}D(uS)(K) \quad(K\Subset\Omega).

This last assertion is not proved for an arbitrary SS.

Proof. The first inequality of (eq:r3-completed-cap-sandwich) follows from constant approximating sequences. If aa is in the normalized completion, choose its normalized approximants aja_j on Ω\Omega. The completed calculus and local mass bounds give local weak convergence of D(ajS)D(a_jS) to D(aS)D(aS), including against compact continuous scalar tests. For every ordinary open G⊂ΩG\subset\Omega, the open-set inequality implies D(aS)(G)≤lim inf⁡jD(ajS)(G)≤CΩ(G).D(aS)(G)\leq\liminf_jD(a_jS)(G)\leq C_\Omega(G). Taking open neighborhoods of a Borel EE and then taking the supremum over aa proves the second inequality. This argument does not use the unavailable compact-set inequality for weak limits.

Fix compact K⋐ΩK\Subset\Omega. Choose an ordinary neighborhood WW of KK and a compactly supported original smooth cutoff 0≤ζ≤10\leq\zeta\leq1 equal to one near W‾\overline W. Put Mζ={ζD(uS):u∈Cb(Ω), 0≤u≤1}.\mathcal M_\zeta =\{\zeta D(uS):u\in\mathcal C_b(\Omega),\ 0\leq u\leq1\}. On a compact metrizable neighborhood of the cutoff support this is a uniformly mass-bounded family. Its capacity on subsets of WW is exactly CΩC_\Omega; open neighborhoods of KK may be restricted to WW. Proposition “Weak closure and compact outer regularity” therefore gives CΩ∗(K)=sup⁡ν~∈Mζ‾ wν~(K).C_\Omega^*(K) =\sup_{\widetilde\nu\in \overline{\mathcal M_\zeta}^{\,w}} \widetilde\nu(K).

Take a sequence ζD(ujS)\zeta D(u_jS) converging to one such ν~\widetilde\nu. Relative CLN bounds D(ujS)D(u_jS) on every compact subset of Ω\Omega. By a countable compact exhaustion and diagonal weak compactness, pass to a subsequence for which D(ujS)D(u_jS) converges locally weakly on all of Ω\Omega to a positive Radon measure ν\nu. More explicitly, use compact cutoffs equal to one on successive compact neighborhoods; the finite weak limits agree on their overlapping interiors and hence define ν\nu. The already convergent cutoff sequence then satisfies ν~=ζν\widetilde\nu=\zeta\nu.

Theorem “Linear-degree convex-block compactness” represents ν=D(aS)\nu=D(aS) for some a∈AS[0,1](Ω)a\in\mathcal A_S^{[0,1]}(\Omega). Since ζ=1\zeta=1 on KK, ν~(K)=D(aS)(K)≤CA(K;Ω).\widetilde\nu(K)=D(aS)(K)\leq C_{\mathcal A}(K;\Omega). Take the supremum over the weak closure and combine with (eq:r3-completed-cap-sandwich) to obtain (eq:r3-completed-compact-cap). Finally, the equivalence with (eq:r3-completion-no-gain) is immediate from that equality and the inclusion of classical potentials. ◻

Scope and minimal remaining gap.

The corollary is a DIRECT_COROLLARY of the new convex-block theorem and the proved measure-family reduction. It replaces a weak-limit question by the exact completed-class inequality (eq:r3-completion-no-gain). It does not replace that inequality by a constant-factor estimate or a smaller-domain capacity. For an arbitrary SS, an actual completed representative has not been shown to have a classical representative on the same domain. Ambient upper regularization can change values on a set charged by SS, as the support examples of Section “Analytic supports and a relative-envelope obstruction” show. The ambient S=[1]S=[1] representative theorem is therefore not an input here.

The non-pluripolar-background case closes in degree one

The next two assertions allow every remaining degree q=n−p≥1q=n-p\geq1. We say that a fixed positive closed background SS is locally non-pluripolar if the total variations of all its coefficient-Radon measures give zero mass to every locally pluripolar Borel set. Equivalently, on each ordinary connected witness domain one has 1PS=0\mathbf1_P S=0 whenever P={ϕ=−∞}P=\{\phi=-\infty\} and ϕ\phi is extended classical psh and not identically −∞-\infty. The equivalence follows by the definition of local pluripolarity, countable local covers in the present finite pp-adic setting, and coefficient-Radon multiplication. This is an explicit extra hypothesis, not a property of all positive closed currents.

Lemma (Ambient polar sets are relatively outer-null under nonconcentration). Let SS be locally non-pluripolar. Every locally pluripolar set has locally zero outer Cap⁡S\operatorname{Cap}_S-capacity, on every permitted compact/domain pair.

Proof. First work where ϕ≤0\phi\leq0 is a nontrivial classical witness and put P={ϕ=−∞}P=\{\phi=-\infty\}. By the hypothesis and Theorem “Positive closed restriction to a complete pole set”, 1PTS,q(v,…,v)=T1PS,q(v,…,v)=0\mathbf1_P T_{S,q}(v,\ldots,v) =T_{\mathbf1_PS,q}(v,\ldots,v)=0 for every normalized classical tester on this same domain. Thus the raw capacity of PP is zero there.

It remains essential to obtain an outer-zero assertion. The bounded increasing classical sequence wN=1+max⁡(ϕ,−N)Nw_N=1+\frac{\max(\phi,-N)}{N} has actual limit w=1Pcw=\mathbf1_{P^c}. Corollary “Local increasing convergence without a prescribed collar” and the completion theorem put ww in AS,b\mathcal A_{S,b} and make it locally relatively quasicontinuous. On an ordinary relatively compact core VV, P∩V=V∩{w<1/2}P\cap V=V\cap\{w<1/2\} is quasi-open. Proposition “Quasi-open sets are outer regular” therefore turns its raw zero capacity into outer zero capacity. This argument does not assume compact outer regularity.

Finally, fix an arbitrary permitted pair H⋐ΩH\Subset\Omega. Cover the part of the locally pluripolar set in HH by countably many ordinary cores with witness domains contained in Ω\Omega. On each core the preceding argument applies. Domain monotonicity and countable subadditivity of the outer capacity give zero capacity on the original pair. Subsets need not be Borel, since outer capacity is monotone. ◻

Proposition (Classical representatives for a non-pluripolar background). Let SS be locally non-pluripolar on the fixed norm-box domain Ω\Omega, in any remaining degree q≥1q\geq1. For every a∈AS[0,1](Ω)a\in\mathcal A_S^{[0,1]}(\Omega) there is r∈Cb(Ω)r\in\mathcal C_b(\Omega), 0≤r≤10\leq r\leq1 on Ω\Omega, such that a=ra=r locally outside sets of outer SS-capacity zero. The representative is not asserted to be unique.

For any finite tuple of such completed potentials, any choices of these representatives give the same weighted and unweighted BT currents on all original compact smooth complementary tests. In particular the normalized completed class does not enlarge the capacity of any Borel set in this case.

Proof. Choose normalized defining approximants aja_j on the entire Ω\Omega. The summable-subsequence construction in Proposition “Existence of capacitary limits”, now applied to their known capacity limit aa, gives a subsequence ajla_{j_l} converging pointwise to aa outside a locally relative-outer-null set NN. For clarity, on the first ll members of a countable cofinal list of compact/domain pairs require the capacity of {∣ajl−a∣>2−l}\{|a_{j_l}-a|>2^{-l}\} to be at most 2−l2^{-l}. Open covers and the countable subadditivity bound on their limsup sets give the asserted NN on every pair.

Set hm=sup⁡l≥majl,rm=hm∗.h_m=\sup_{l\geq m}a_{j_l},\qquad r_m=h_m^*. The bounded upper-envelope theorem, Theorem “Local bounded upper-envelope theorem”, gives rm∈Cb(Ω)r_m\in\mathcal C_b(\Omega) with 0≤rm≤10\leq r_m\leq1. The sequence decreases, so r=lim⁡mrmr=\lim_m r_m is classical on the same domain with the same normalization. For each mm, the discrepancy set Em={hm<rm}E_m=\{h_m<r_m\} is locally pluripolar by Theorem “Negligible upper-envelope sets have ambient outer capacity zero” and Corollary “Local polar, negligible, and outer-zero sets”. Lemma “Ambient polar sets are relatively outer-null under nonconcentration” makes EmE_m locally relative-outer-null. Outside N∪⋃mEmN\cup\bigcup_mE_m one has r=lim⁡mhm=lim sup⁡lajl=a.r=\lim_m h_m=\limsup_l a_{j_l}=a. This proves the asserted equality modulo local relative outer capacity zero, without replacing any of the given Borel point values in the original completed class.

The completed calculus is independent of precisely these changes of representatives, including the weight slot and the degree-zero background multiplication, by Theorem “BT calculus on the bounded capacitary closure”. It therefore identifies each completed product with the product of the chosen classical representatives. In particular, if aa is normalized, its measure TS,q(a,…,a)T_{S,q}(a,\ldots,a) is the measure of a normalized classical rr on the original Ω\Omega. Taking suprema over Borel sets gives equality of the two capacity families. Uniqueness is intentionally absent: for S=0S=0 all functions are relatively quasi-everywhere equal. ◻

Status. The preceding lemma and representative proposition are DIRECT_COROLLARY consequences of the proved R2 polar-restriction, envelope, negligible-set, and completion theorems. They retain the stated nonconcentration hypothesis.

Theorem (Sharp compact regularity for a non-pluripolar degree-one background). Let SS be an arbitrary fixed locally non-pluripolar, symmetric original-positive separately closed current of bidegree (n−1,n−1)(n-1,n-1) on the permitted fixed norm-box domain Ω\Omega. Every locally weak limit of D(ujS),uj∈Cb(Ω),0≤uj≤1,D(u_jS),\qquad u_j\in\mathcal C_b(\Omega),\quad0\leq u_j\leq1, is D(rS)D(rS) for some r∈Cb(Ω)r\in\mathcal C_b(\Omega) with 0≤r≤10\leq r\leq1. Consequently, for every compact K⋐ΩK\Subset\Omega, Cap⁡S∗(K;Ω)=Cap⁡S(K;Ω).\operatorname{Cap}_S^*(K;\Omega) =\operatorname{Cap}_S(K;\Omega). The raw capacity is continuous on decreasing compact subsets of Ω\Omega. Equivalently, every localized weak-closure measure satisfies the sharp bound (RC) on the core where its cutoff is one.

Proof. Theorem “Linear-degree convex-block compactness” represents any such weak limit by a normalized completed potential on Ω\Omega. Proposition “Classical representatives for a non-pluripolar background” replaces its class by a normalized classical potential on this same domain, preserving the current. The compact assertion follows either from Corollary “The outer capacity on compact sets is the completed capacity” or directly from Proposition “Weak closure and compact outer regularity”. That proposition also gives continuity from above on compact sets. Its localized version is exactly the final assertion. ◻

Status: NEW_PROVED. The conclusion holds in all ambient dimensions under the explicit non-pluripolarity hypothesis and with remaining degree one. The representative proposition alone does not give weak-closure representation for q>1q>1.

Why an ambient contact measure is not a relative one

We first give a compact-obstacle example in remaining degree one. It applies to all classical competitors, not merely toric ones.

Proposition (An ambient compact extremal with relative mass off the obstacle). Let n≥2n\geq2, choose a uniformizer π\pi of the original KK, put λ=−log⁡∣π∣>0\lambda=-\log|\pi|>0, and work on the analytic torus X=(Gm,Kn)anX=(\mathbf G_{m,K}^n)^{\mathrm{an}}. Set ti=log⁡∣Ti∣,θi=d′ti∧d′′ti,β=∑i=1nθi,S=[βn−1].t_i=\log|T_i|,\qquad \theta_i=d't_i\wedge d''t_i,\qquad \beta=\sum_{i=1}^n\theta_i, \qquad S=[\beta^{n-1}]. For 0<a<λ0<a<\lambda define the ordinary domain and its compact core Ω={max⁡i∣ti∣<λ},Ka={max⁡i∣ti∣≤a},\Omega=\{\max_i|t_i|<\lambda\},\qquad K_a=\{\max_i|t_i|\leq a\}, and the continuous classical psh function H=max⁡{−1,max⁡i∣ti∣−λλ−a}.H=\max\left\{-1, \frac{\max_i|t_i|-\lambda}{\lambda-a}\right\}. Then HH is the actual pointwise supremum of all v∈Cb(Ω)v\in\mathcal C_b(\Omega) satisfying v≤0v\leq0 on Ω\Omega and v≤−1v\leq-1 on KaK_a. Nevertheless D(HS)(Ω∖Ka)>0.D(HS)(\Omega\setminus K_a)>0. Thus the ambient compact extremal does not in general have a relative Monge–Ampère measure supported on its obstacle.

Proof. The domain is an ordinary open with compact closure in the smooth affine torus, and KaK_a is compact in it. If a strict norm-box core is desired, take, for example, a=λ/2a=\lambda/2 and express its bounds using Ti2T_i^2 and the original value ∣π∣|\pi|; no valuation is rescaled. Logarithms of analytic units are harmonic after every complete extension and curve pullback. Finite maxima of their affine combinations and constants are classical psh. This proves that H∈Cb(Ω)H\in\mathcal C_b(\Omega), that −1≤H<0-1\leq H<0, that H=−1H=-1 on KaK_a, and that HH has ordinary boundary value zero.

We claim (DH)n=0(DH)^n=0 off KaK_a. Near any point there the constant branch −1-1 is inactive. Among the branches tit_i and −ti-t_i of the maximum, at most one sign for each ii is active, since the maximum is strictly positive. Delete inactive branches on a neighborhood. If the remaining affine functions are ℓ1,…,ℓm\ell_1,\ldots,\ell_m, then m≤nm\leq n, and max⁡(ℓ1,…,ℓm)=ℓm+max⁡(ℓ1−ℓm,…,ℓm−1−ℓm,0).\max(\ell_1,\ldots,\ell_m) =\ell_m+\max(\ell_1-\ell_m,\ldots, \ell_{m-1}-\ell_m,0). The differences are logarithms of analytic units. Smooth the finite maximum by translation-invariant convex regularized maxima. Their curvature factors through at most m−1m-1 such moment coordinates, so its nnth exterior power vanishes identically as an original form. The uniform continuous fixed-background calculus passes this equality to HH. This proves the claim without a regularity assertion for arbitrary psh functions on the skeleton.

Let vv be an admissible competitor. For every ϵ>0\epsilon>0, the set {H<v−ϵ}\{H<v-\epsilon\} is contained in {H<−ϵ}\{H<-\epsilon\}, whose closure is compact in Ω\Omega. Also {H<v}\{H<v\} is disjoint from KaK_a, because H=−1H=-1 and v≤−1v\leq-1 there. Thus (DH)n({H<v})=0(DH)^n(\{H<v\})=0. Corollary “Pointwise ambient domination and uniqueness”, on this relatively compact affine domain, gives v≤Hv\leq H at every point. Since HH itself is admissible, it is the stated actual supremum. This comparison includes all non-toric competitors.

To compute relative mass outside KaK_a, choose a tropical open rectangle meeting t1=t2=r,a<r<λ,∣ti∣<r (i≥3),t_1=t_2=r,\qquad a<r<\lambda, \qquad |t_i|<r\ (i\geq3), small enough that the only potentially maximal branches are t1,t2t_1,t_2, and that their maximum is greater than aa. There H=t2−λ+max⁡(0,t1−t2)λ−a.H=\frac{t_2-\lambda+\max(0,t_1-t_2)}{\lambda-a}. The one-variable distributional second derivative of max⁡(0,s)\max(0,s) is δ0\delta_0. One may equivalently obtain the following formula by smooth convex regularization and the original CLD tropical integration formula. For a nonnegative original smooth tester ψ∘trop⁡\psi\circ\operatorname{trop} supported in that rectangle, ∫(ψ∘trop⁡) D(HS)=2(n−1)!λ−a∫ψ(t2,t2,t3,…,tn) dt2 dt3⋯dtn.\int(\psi\circ\operatorname{trop})\,D(HS) =\frac{2(n-1)!}{\lambda-a} \int\psi(t_2,t_2,t_3,\ldots,t_n) \,dt_2\,dt_3\cdots dt_n. Indeed, the square of the gradient of s=t1−t2s=t_1-t_2 in the diagonal form β\beta is two, and d′s∧d′′s∧βn−1=2(n−1)!θ1∧⋯∧θnd's\wedge d''s\wedge\beta^{n-1} =2(n-1)!\theta_1\wedge\cdots\wedge\theta_n. Choose ψ\psi positive at a point of the indicated diagonal. The displayed integral is strictly positive and is supported outside KaK_a, proving the assertion. ◻

There is also a smooth common-collar obstruction in every intermediate remaining degree. It directly tests the relative energy envelope rather than only the ambient compact extremal.

Proposition (A relative energy maximum need not satisfy contact orthogonality). On the same torus and domain, let 1≤q<n1\leq q<n and put S=[βn−q],b=∑i=1qti2.S=[\beta^{n-q}],\qquad b=\sum_{i=1}^q t_i^2. Choose nonzero φ∈Cc∞((−λ,λ)n)\varphi\in C_c^\infty((-\lambda,\lambda)^n), φ≥0\varphi\geq0, and let χ=φ∘trop⁡\chi=\varphi\circ\operatorname{trop}, L=supp⁡χL=\operatorname{supp}\chi, and f=b+χf=b+\chi. Consider F={v∈Cb(Ω):b−1≤v≤f,v=b on Ω∖L}.\mathcal F=\{v\in\mathcal C_b(\Omega): b-1\leq v\leq f,\quad v=b\text{ on }\Omega\setminus L\}. The greatest actual member and the relative capacitary energy envelope of this uniformly bounded common-collar family are bb (the latter modulo local relative capacity zero). However

∫Ω(f−b)TS,q(b,…,b)=2qq!(n−q)!∫Rnφ(t) dt>0.\int_\Omega(f-b)T_{S,q}(b,\ldots,b) =2^q q!(n-q)!\int_{\mathbf R^n}\varphi(t)\,dt>0.

In particular, relative energy maximization over ambient classical psh competitors does not imply the expected contact orthogonality, even for a smooth strongly positive background.

Proof. The smooth function bb is classical psh: it is a convex smooth function of harmonic logarithms of units, after every complete extension and curve pullback. It is bounded on Ω\Omega and Db=2∑i=1qθi,(Db)n=0,TS,q(b,…,b)=2qq!(n−q)!θ1∧⋯∧θn.Db=2\sum_{i=1}^q\theta_i, \qquad (Db)^n=0, \qquad T_{S,q}(b,\ldots,b) =2^q q!(n-q)!\theta_1\wedge\cdots\wedge\theta_n. For any v∈Fv\in\mathcal F, the closures of all positive-margin sets {b<v−ϵ}\{b<v-\epsilon\} lie in the fixed compact LL. Pointwise ambient domination again gives v≤bv\leq b on all of Ω\Omega. Since b∈Fb\in\mathcal F, it is the greatest member. Relative Aubin–Mabuchi energy is monotone on this common-collar class, so Eb(v)≤Eb(b)=0\mathcal E_b(v)\leq\mathcal E_b(b)=0. The constant sequence bb is a maximizing sequence in Theorem “A capacitary envelope from a maximizing sequence”, identifying its capacitary envelope with bb.

The original CLD integration formula for the identity tropical chart of the torus has multiplicity one and integrates θ1∧⋯∧θn\theta_1\wedge\cdots\wedge\theta_n as Lebesgue measure with its positive orientation. Substitution gives (eq:r3-failed-relative-orthogonality). Its strict positivity uses only the chosen nonzero nonnegative smooth φ\varphi. All products in this example are ordinary original currents. ◻

Status of the counterexamples.

The two counterexamples are NEW_PROVED; the contact or orthogonality implications they disprove have status FALSE. Neither is a counterexample to (RC). The second concerns a continuous obstacle and a common-collar family; it does not identify every possible relative compact-capacity extremal. The smooth backgrounds in these examples are locally non-pluripolar, by the ambient smooth-trace polar-zero theorem. Thus the degree-one contact failure coexists with the compact outer regularity proved above; it is a failure of the proposed proof mechanism, rather than a failure of that regularity.

The remaining nonlinear weak-limit obstruction

For q>1q>1, applying the same square integration-by-parts argument to ∫χf2(Dw)q∧S\int\chi f^2(Dw)^q\wedge S uses the energy of ff relative to Qw=(Dw)q−1∧S.Q_w=(Dw)^{q-1}\wedge S. This is a positive closed background of bidegree (n−1,n−1)(n-1,n-1), but it changes with the capacity tester ww. The direct Hilbert argument above therefore does not by itself give a fixed seminorm controlling every tester: convergence in countably many selected seminorms would not justify that supremum. The independent mixed-energy recurrence in Theorem “An all-degree mixed-energy capacity estimate” resolves this uniform-control problem. Together with quantified concavity, it gives the all-degree convex-block theorem “Convex-block compactness in every positive remaining degree”. Thus tester-dependent energy is no longer an open obstruction in this record.

The separate nonlinear weak-limit problem remains. For q>1q>1, the MA measure of a convex combination contains mixed terms; it is not the corresponding convex combination of the original MA measures. The new capacity-convergent convex blocks do not therefore identify an arbitrary weak limit of those original measures with the MA measure of their completed limit.

The current status of arbitrary-background compact outer regularity remains OPEN. The precise reusable advances are the fixed-energy capacity estimate, normalized convex-block representation in degree one, its exact completed-capacity reduction, and sharp compact regularity under the additional non-pluripolarity hypothesis in that degree. The stronger convex-block compactness result for every remaining degree is proved separately in Section “Mixed energy controls capacity in every remaining degree”. The contact counterexamples rule out the ambient contact and relative-energy orthogonality routes above.

A contact inequality for every fixed positive background

Let SS be any allowed symmetric original-positive separately closed (p,p)(p,p)-current on an ordinary open set UU, and put q=n−p≥1q=n-p\geq1. Write μu=TS,q(u,…,u)\mu_u=T_{S,q}(u,\ldots,u) for actual u∈Cb(U)u\in\mathcal C_b(U). Restrictions below are Borel restrictions of the coefficient-Radon top measures.

Proposition (Relative maximum and contact inequalities). For u,v∈Cb(U)u,v\in\mathcal C_b(U),

μmax⁡(u,v)≥1{u≥v}μu+1{u<v}μv.\mu_{\max(u,v)} \geq{\bf1}_{\{u\geq v\}}\mu_u +{\bf1}_{\{u<v\}}\mu_v.

In particular, if u≤hu\leq h,

1{u=h}μu≤1{u=h}μh.{\bf1}_{\{u=h\}}\mu_u \leq{\bf1}_{\{u=h\}}\mu_h.

Status: DIRECT_COROLLARY of the proved Borel maximum locality and decreasing BT continuity.

Proof. Put wj=max⁡(u+1/j,v)w_j=\max(u+1/j,v). This sequence decreases to w=max⁡(u,v)w=\max(u,v). On the Borel set E={u≥v}E=\{u\geq v\} the strict inequality u+1/j>vu+1/j>v holds, so Borel maximum locality in each differentiated slot and invariance under constants give

1Eμwj=1Eμu.{\bf1}_E\mu_{w_j}={\bf1}_E\mu_u.

The top measures converge weakly against compact continuous functions, by the original-current decreasing theorem and the coefficient-Radon convergence result. For a compact K⊂EK\subset E, local Portmanteau therefore gives μw(K)≥lim sup⁡jμwj(K)=μu(K).\mu_w(K)\geq\limsup_j\mu_{w_j}(K)=\mu_u(K). For an arbitrary Borel subset of EE, inner regularity of the fixed Radon measure μu\mu_u extends this inequality from its compact subsets. Work first in a relatively compact stage and then exhaust if necessary. This step does not require EE to be closed or ordinary open. On the complementary Borel set {u<v}\{u<v\}, strict Borel maximum locality gives μw=μv\mu_w=\mu_v. The two disjoint restrictions prove (eq:r3-relative-max-inequality). If u≤hu\leq h, take v=hv=h, so that w=hw=h and {u≥h}={u=h}\{u\geq h\}=\{u=h\}, giving (eq:r3-relative-contact-inequality). ◻

Corollary (Zero mass on contact with a maximal potential). If u≤hu\leq h are actual bounded classical potentials and TS,q(h,…,h)=0T_{S,q}(h,\ldots,h)=0, then 1{u=h}TS,q(u,…,u)=0{\bf1}_{\{u=h\}}T_{S,q}(u,\ldots,u)=0.

The sets in this statement need only be Borel. No arbitrary varying-background theorem, capacity regularity assertion, or exchange of discontinuous limits is used.

Nonzero atomic Dirichlet problems on one fixed domain

Fix a finite extension K/QpK/\mathbf Q_p with its original absolute value. Let Y/KY/K be smooth integral affine of pure dimension n≥1n\geq1. Throughout this section the domain is fixed in advance: V⋐YanV\Subset Y^{\mathrm{an}} is connected and ordinary open, has nonempty ordinary boundary, and admits

ρ∈Cb(V)∩C(V‾),ρ<0 on V,ρ=0 on ∂V.(H)\rho\in\mathcal C_b(V)\cap C(\overline V),\qquad \rho<0\text{ on }V,\qquad \rho=0\text{ on }\partial V. \tag{H}

Closures and boundary limits are ordinary ones in YanY^{\mathrm{an}}. The points and masses below are chosen after VV has been fixed. We use D=d′d′′D=d'd'' and the ambient background [1][1], with the original CLD calibration. All weak equalities and limits are tested against the full original compactly supported smooth scalar class.

The inputs from the preceding records are the bounded mixed calculus, ordinary upper-envelope closure, divisorial no-jump, local maximality equivalence (Du)n=0(Du)^n=0, the boundary maximum principle, and pointwise ambient comparison, domination and uniqueness. These inputs retain the all-extensions classical definition. The projective input B, Theorems C.6 and 4.4, has already been used in proving those local results; no new projective or residue-characteristic hypothesis is inserted here.

Point obstacles and their actual envelopes

Fix pairwise distinct ambient divisorial points Z={z1,…,zr}⊂VZ=\{z_1,\ldots,z_r\}\subset V, where r≥1r\geq1. For a=(a1,…,ar)∈[0,∞)ra=(a_1,\ldots,a_r)\in[0,\infty)^r put A(a)=max⁡iaiA(a)=\max_i a_i and Fa={v∈Cb(V):−A(a)≤v≤0, v(zi)≤−ai for every i},Ua=(sup⁡v∈Fav)∗.\mathcal F_a=\{v\in\mathcal C_b(V):-A(a)\leq v\leq0, \ v(z_i)\leq-a_i\text{ for every }i\}, \qquad U_a=\left(\sup_{v\in\mathcal F_a}v\right)^*. The constant −A(a)-A(a) makes this family nonempty.

Proposition (Actual weighted point envelopes). For every aa the function UaU_a belongs to Fa\mathcal F_a and Ua=sup⁡v∈FavU_a=\sup_{v\in\mathcal F_a}v at every point. It is also the greatest member of Cb(V)\mathcal C_b(V) with v≤0v\leq0 and v(zi)≤−aiv(z_i)\leq-a_i, even without imposing the lower bound in the definition of Fa\mathcal F_a. It has ordinary boundary limit zero.

For all a,b∈[0,∞)ra,b\in[0,\infty)^r and t≥0t\geq0,

∥Ua−Ub∥L∞(V)≤∥a−b∥∞,Uta=tUa.(1)\|U_a-U_b\|_{L^\infty(V)}\leq\|a-b\|_\infty, \qquad U_{ta}=tU_a. \tag{1}

For every fixed M<∞M<\infty, all parameters a∈[0,M]ra\in[0,M]^r have one common continuous lower barrier with zero boundary limit.

Proof. Upper-envelope closure gives Ua∈Cb(V)U_a\in\mathcal C_b(V) and −A(a)≤Ua≤0-A(a)\leq U_a\leq0. Divisorial no-jump gives the crucial actual constraint Ua(zi)=sup⁡v∈Fav(zi)≤−ai.U_a(z_i)=\sup_{v\in\mathcal F_a}v(z_i)\leq-a_i. Thus UaU_a itself is in Fa\mathcal F_a. The pointwise supremum is consequently at least UaU_a, and is at most UaU_a by the definition of upper regularization. This proves the equality and the greatest-member assertion for the displayed family. If v∈Cb(V)v\in\mathcal C_b(V) is any other nonpositive admissible function, then max⁡(v,−A(a))∈Fa\max(v,-A(a))\in\mathcal F_a and dominates vv. This proves the assertion without the artificial lower bound.

Set dZ=min⁡i(−ρ(zi))>0.d_Z=\min_i(-\rho(z_i))>0. For M>0M>0 the continuous classical function bM=max⁡(−M,MdZρ)b_M=\max\left(-M,\frac{M}{d_Z}\rho\right) equals −M-M at every ziz_i and has boundary limit zero. For a∈[0,M]ra\in[0,M]^r, the greatest-member characterization gives bM≤Ua≤0b_M\leq U_a\leq0. This proves the common barrier and the boundary limit. When a=0a=0, the family consists of zero and U0=0U_0=0.

Let δ=∥a−b∥∞\delta=\|a-b\|_\infty. The function w=max⁡(−A(b),Ua−δ)w=\max(-A(b),U_a-\delta) lies between −A(b)-A(b) and zero. At each ziz_i, both branches are at most −bi-b_i, because bi≤ai+δb_i\leq a_i+\delta. Hence w∈Fbw\in\mathcal F_b and Ub≥Ua−δU_b\geq U_a-\delta. Interchanging the parameters proves the first inequality in (1). For t>0t>0, multiplication by tt is a bijection between the two admissible families, and commutes with ordinary upper regularization. The case t=0t=0 has already been noted. ◻

Proposition (A continuous homogeneous map of obstacle depths to masses). There are nonnegative real functions ci(a)c_i(a) such that

(DUa)n=∑i=1rci(a)δzi.(2)(DU_a)^n=\sum_{i=1}^r c_i(a)\delta_{z_i}. \tag{2}

Each cic_i is locally Lipschitz on [0,∞)r[0,\infty)^r, satisfies ci(ta)=tnci(a)c_i(ta)=t^n c_i(a), and obeys

ai=0 ⟹ ci(a)=0.(3)a_i=0\ \Longrightarrow\ c_i(a)=0. \tag{3}

For every a≠0a\ne0 the total mass m(a)=∑ici(a)m(a)=\sum_i c_i(a) is strictly positive.

Proof. Write Z+(a)={zi:ai>0}Z_+(a)=\{z_i:a_i>0\}. We first prove that UaU_a is locally maximal off Z+(a)Z_+(a). Take an ordinary W⋐V∖Z+(a)W\Subset V\setminus Z_+(a) and a competitor v∈Cb(W)v\in\mathcal C_b(W) which equals UaU_a outside a compact L⋐WL\Subset W. Glue max⁡(v,Ua)\max(v,U_a) on WW to UaU_a off LL. The resulting function w∈Cb(V)w\in\mathcal C_b(V) is bounded above: only the compact set LL has changed. It has boundary limit zero, so the boundary maximum principle gives w≤0w\leq0. Moreover w≥Ua≥−A(a)w\geq U_a\geq-A(a). At each positive-depth constraint it equals UaU_a; each zero-depth constraint is automatic from w≤0w\leq0. Thus w∈Faw\in\mathcal F_a. The actual greatest-member property forces w≤Uaw\leq U_a, and hence v≤Uav\leq U_a. This is local maximality.

The proved homogeneous-equation criterion now gives (DUa)n=0(DU_a)^n=0 on V∖Z+(a)V\setminus Z_+(a). This positive Radon measure is supported on a finite compact set, so it is finite and has the form (2). Its support also proves (3). Homogeneity follows from (1) and the proved BT homogeneity.

Here is a direct continuity estimate, with no varying-background assertion. Choose original smooth compact scalar cutoffs χi\chi_i which equal one near ziz_i and zero near every other point of ZZ. Such cutoffs exist in the original CLD smooth sheaf. For a,b∈[0,M]ra,b\in[0,M]^r, the bounded mixed calculus gives (DUa)n−(DUb)n=D((Ua−Ub)∑j=0n−1(DUa)j∧(DUb)n−1−j).(DU_a)^n-(DU_b)^n =D\left((U_a-U_b) \sum_{j=0}^{n-1}(DU_a)^j\wedge(DU_b)^{n-1-j}\right). Pair with χi\chi_i. The coefficient-Radon estimates for the fixed smooth test DχiD\chi_i and relative CLN, on a compact neighborhood of its support, give a constant Ci,MC_{i,M} with ∣ci(a)−ci(b)∣≤Ci,M∥Ua−Ub∥L∞(V)≤Ci,M∥a−b∥∞.\begin{split} |c_i(a)-c_i(b)| &\leq C_{i,M}\|U_a-U_b\|_{L^\infty(V)}\\ &\leq C_{i,M}\|a-b\|_\infty. \end{split} The constant depends on the fixed test stage, M,nM,n and the domain geometry, not on a,ba,b. Borel multiplication by the bounded difference is legitimate for these coefficient-Radon currents. When n=1n=1, the sum contains the single background [1][1] and the same argument applies. Alternatively, uniform convergence in (1), the proved BT stability, and the fixed finite support give continuity of every coefficient directly. No mass can escape to the boundary because the support is always contained in ZZ.

Finally suppose a≠0a\ne0 and m(a)=0m(a)=0. Then (DUa)n=0(DU_a)^n=0. The common lower barrier makes every set {Ua<−δ}\{U_a<-\delta\} relatively compact in VV. Pointwise ambient domination applied to (Ua,0)(U_a,0) gives Ua≥0U_a\geq0. This contradicts Ua(zi)≤−ai<0U_a(z_i)\leq-a_i<0 for an index with ai>0a_i>0. Therefore m(a)>0m(a)>0. ◻

A finite-dimensional surjectivity argument

Lemma (Surjectivity of a face-preserving simplex map). Let Δr−1={a∈[0,∞)r:∑iai=1}.\Delta^{r-1}=\{a\in[0,\infty)^r:\textstyle\sum_i a_i=1\}. If a continuous F:Δr−1→Δr−1F:\Delta^{r-1}\to\Delta^{r-1} satisfies ai=0⇒Fi(a)=0a_i=0\Rightarrow F_i(a)=0, then FF is surjective.

Proof. First let yy be in the relative interior of the simplex. Put di(a)=(yi−Fi(a))+,σ(a)=∑idi(a),Ti(a)=ai+di(a)1+σ(a).d_i(a)=(y_i-F_i(a))_+, \qquad \sigma(a)=\sum_i d_i(a), \qquad T_i(a)=\frac{a_i+d_i(a)}{1+\sigma(a)}. This is a continuous self-map of the simplex. Brouwer’s fixed-point theorem, used here as an external topological input [Brouwer11], gives a fixed point aa. Its equations are di(a)=σ(a)ai.d_i(a)=\sigma(a)a_i. If ai=0a_i=0, the face condition gives Fi(a)=0F_i(a)=0, whereas yi>0y_i>0 gives di(a)>0d_i(a)>0, a contradiction. Thus all ai>0a_i>0. If σ(a)>0\sigma(a)>0, every di(a)>0d_i(a)>0, so yi>Fi(a)y_i>F_i(a) for every ii, contradicting that both vectors have sum one. Hence σ(a)=0\sigma(a)=0. Then yi≤Fi(a)y_i\leq F_i(a) for every ii, and equality of their sums gives F(a)=yF(a)=y.

For a boundary target, restrict to its support face. The face condition makes that face invariant, and the preceding argument applies to its relative interior. A one-point simplex is immediate. This proves surjectivity. ◻

For r=2r=2 only the intermediate value theorem is needed: the normalized first mass along a=(t,1−t)a=(t,1-t) is continuous, equals zero at t=0t=0, and equals one at t=1t=1.

Theorem (Finite atomic Dirichlet problem on the fixed domain). Assume (H). Let z1,…,zrz_1,\ldots,z_r be distinct ambient divisorial points of VV, let mi≥0m_i\geq0, and let b∈Rb\in\mathbf R. There exists a unique bounded classical psh function u∈Cb(V)u\in\mathcal C_b(V) such that

(Du)n=∑i=1rmiδzi,lim⁡x→ξx∈Vu(x)=b(ξ∈∂V).(4)(Du)^n=\sum_{i=1}^r m_i\delta_{z_i},\qquad \lim_{\substack{x\to\xi\\x\in V}}u(x)=b \quad(\xi\in\partial V). \tag{4}

The equality of measures has the original calibration. For a fixed finite set ZZ, there is a finite constant CZC_Z, depending on VV and ZZ but not on the masses or bb, such that

∥u−b∥L∞(V)≤CZ(∑imi)1/n.(5)\|u-b\|_{L^\infty(V)} \leq C_Z\left(\sum_i m_i\right)^{1/n}. \tag{5}

With V,ZV,Z fixed, the solution depends continuously in the uniform norm on the nonnegative mass vector and on bb. No continuity of uu at arbitrary interior points is asserted by this theorem alone.

Proof. It is enough first to take b=0b=0. If all masses are zero, take u=0u=0. Otherwise let Mμ=∑imi>0M_\mu=\sum_i m_i>0 and yi=mi/Mμy_i=m_i/M_\mu. On Δr−1\Delta^{r-1} define Fi(a)=ci(a)m(a).F_i(a)=\frac{c_i(a)}{m(a)}. The denominator is positive; the map is continuous and preserves faces by Proposition “A continuous homogeneous map of obstacle depths to masses”. Choose a∈Δr−1a\in\Delta^{r-1} with F(a)=yF(a)=y, using the simplex lemma. Put t=(Mμm(a))1/n,u=tUa.t=\left(\frac{M_\mu}{m(a)}\right)^{1/n},\qquad u=tU_a. Homogeneity gives every prescribed coefficient mim_i, and the boundary limit is zero. Adding bb proves existence in (4). The same previously fixed domain has been used throughout; no domains adapted to individual atoms have been glued together.

For uniqueness, two solutions have equal ordinary boundary limits. For every positive margin, both strict-gap closures are therefore compact in VV: cover the compact ordinary boundary by finitely many neighborhoods on which the difference has smaller absolute value. Pointwise ambient uniqueness gives equality at every point.

Since m(a)m(a) is continuous and positive on the compact simplex, mZ=min⁡Δr−1m(a)>0m_Z=\min_{\Delta^{r-1}}m(a)>0. Also ∥Ua∥∞≤A(a)≤1\|U_a\|_\infty\leq A(a)\leq1 there. Thus CZ=mZ−1/nC_Z=m_Z^{-1/n} proves (5).

To prove stability, let the nonnegative mass vectors tend to a vector μ\mu. If their total masses tend to zero, (5) gives uniform convergence to zero. Otherwise represent their zero-boundary solutions as tjUajt_jU_{a_j} with aj∈Δr−1a_j\in\Delta^{r-1} as above. The parameters tjt_j are bounded. From any subsequence one can extract a further subsequence with aj→aa_j\to a and tj→tt_j\to t. The envelope Lipschitz estimate gives uniform convergence to tUatU_a. The atomic mass map and the common boundary barrier show that this limit solves the limiting problem. Uniqueness identifies it with the specified solution. All subsequential limits therefore agree, which proves uniform convergence of the entire sequence. Translations handle the varying boundary constant. ◻

The analytic construction and the simplex argument are recorded as NEW_PROVED, relative to the preceding R1/R2 theorems. Brouwer’s theorem is EXTERNAL_KNOWN. The construction does not sum single-atom solutions. Its coefficients ci(a)c_i(a) come from the full nonlinear BT measure, so every mixed term is included before prescribing the masses.

Order and a summable countable atomic class

Lemma (Monge–Ampère order reverses zero-boundary solution order). Let u1,u2∈Cb(V)u_1,u_2\in\mathcal C_b(V) be bounded with boundary limits zero. If (Du1)n≥(Du2)n(Du_1)^n\geq(Du_2)^n as Radon measures, then u1≤u2u_1\leq u_2 at every point. The measures need not have finite total mass on VV.

Proof. Write μi=(Dui)n\mu_i=(Du_i)^n. For t,δ>0t,\delta>0 put w=(1+t)u1−δ,E={u2<w}.w=(1+t)u_1-\delta,\qquad E=\{u_2<w\}. The boundary limit of u2−wu_2-w is δ\delta, so EE itself has compact closure in VV. The compact-gap comparison principle gives (1+t)nμ1(E)≤μ2(E)≤μ1(E).(1+t)^n\mu_1(E)\leq\mu_2(E)\leq\mu_1(E). The masses being compared are finite on this compact set. Hence μ1(E)=μ2(E)=0\mu_1(E)=\mu_2(E)=0. Pointwise ambient domination for (u2,w)(u_2,w) gives u2≥wu_2\geq w everywhere. Let δ↓0\delta\downarrow0 and then t↓0t\downarrow0. The extra constant shift is what permits the cancellation without assuming a finite total mass on the whole domain. ◻

Write G(μ)G(\mu) for a zero-boundary solution when one has been constructed. Positive homogeneity and uniqueness give G(sμ)=s1/nG(μ)G(s\mu)=s^{1/n}G(\mu) for s≥0s\geq0. If the relevant solutions exist, positivity of all mixed BT terms gives

G(μ)+G(ν)≤G(μ+ν)≤min⁡{G(μ),G(ν)}.(6)G(\mu)+G(\nu) \leq G(\mu+\nu) \leq\min\{G(\mu),G(\nu)\}. \tag{6}

Indeed the full mixed expansion makes the Monge–Ampère measure of the sum at least μ+ν\mu+\nu, so the lemma gives the lower inequality. The two measure-order comparisons give the upper inequalities. This explains precisely why adding single-atom solutions is a useful subsolution but does not compute the solution in dimension greater than one.

Corollary (A countable atomic Dirichlet class). Let zi∈Vz_i\in V be pairwise distinct ambient divisorial points and mi>0m_i>0, for i≥1i\geq1. Let gi=G(δzi)g_i=G(\delta_{z_i}) be their single-atom zero-boundary solutions on this same fixed domain. Suppose

∑i=1∞mi1/n∥gi∥L∞(V)<∞.(7)\sum_{i=1}^{\infty}m_i^{1/n}\|g_i\|_{L^\infty(V)}<\infty. \tag{7}

Then μ=∑imiδzi\mu=\sum_i m_i\delta_{z_i} is a locally finite Radon measure on VV. It has a unique bounded classical zero-boundary solution. The solutions for its finite partial sums decrease uniformly to that solution, with the explicit error bound

0≤uN−u≤∑i>Nmi1/n∥gi∥L∞(V).(8)0\leq u_N-u\leq \sum_{i>N}m_i^{1/n}\|g_i\|_{L^\infty(V)}. \tag{8}

Their weighted BT measures converge on every original compact smooth test. Adding a constant gives the same conclusion for any constant boundary value.

Proof. Set μN=∑i=1Nmiδzi\mu_N=\sum_{i=1}^N m_i\delta_{z_i} and uN=G(μN)u_N=G(\mu_N), using the finite atomic theorem. The measure order lemma makes uNu_N decrease. The mixed sandwich gives ∑i=1Nmi1/ngi≤uN≤0.\sum_{i=1}^N m_i^{1/n}g_i\leq u_N\leq0. Condition (7) makes the series s=∑imi1/ngis=\sum_i m_i^{1/n}g_i uniformly convergent and bounded on VV. Its finite sums decrease and are classical, so ss is classical. Its boundary limit is zero: first make the uniform tail small, then use the boundary limits of the finite initial sum. Thus s≤uN↓u≤0s\leq u_N\downarrow u\leq0 gives a bounded classical limit with zero boundary limit. In fact, repeated use of the mixed sandwich gives, for every N,M≥1N,M\geq1, uN+∑i=N+1N+Mmi1/ngi≤uN+M≤uN.u_N+\sum_{i=N+1}^{N+M}m_i^{1/n}g_i \leq u_{N+M}\leq u_N. Let M→∞M\to\infty. This proves (8), and hence uniform convergence.

The fixed-background decreasing BT theorem gives (DuN)n→(Du)n(Du_N)^n\to(Du)^n. For a nonnegative original smooth compact scalar test χ\chi it follows that ∑imiχ(zi)=lim⁡N∫χ dμN=∫χ(Du)n<∞.\sum_i m_i\chi(z_i) =\lim_N\int\chi\,d\mu_N =\int\chi(Du)^n<\infty. Choose such a cutoff at least one near any prescribed compact subset of VV. This proves local finiteness of the atomic sum; on this locally compact metrizable space it is the corresponding Radon measure. The same identity gives (Du)n=μ(Du)^n=\mu on all original tests. The weighted decreasing theorem also gives uN(DuN)n→u(Du)nu_N(Du_N)^n\to u(Du)^n locally. Uniqueness follows from the two compact-gap conditions and pointwise ambient uniqueness, without a global total-mass assumption. ◻

For the singleton, the R2 compact-extremal formula identifies gz=Cap⁡[1]({z};V)−1/nU{z},∥gz∥∞=Cap⁡[1]({z};V)−1/n.g_z=\operatorname{Cap}_{[1]}(\{z\};V)^{-1/n}U_{\{z\}}, \qquad \|g_z\|_\infty =\operatorname{Cap}_{[1]}(\{z\};V)^{-1/n}. The point constraint is exact at the divisorial point, so U{z}(z)=−1U_{\{z\}}(z)=-1. Thus (7) is equivalently the explicit condition ∑i(miCap⁡[1]({zi};V))1/n<∞.\sum_i \left(\frac{m_i} {\operatorname{Cap}_{[1]}(\{z_i\};V)}\right)^{1/n}<\infty. The order, sandwich and countable-atomic conclusions are DIRECT_COROLLARY of the finite atomic theorem and the proved positive mixed calculus, with their proofs given above. No arbitrary diffuse-measure existence or stability under moving supports is asserted here.

Corollary (A quantitative bound for atomic changes of mass). Let μ=∑imiδzi\mu=\sum_i m_i\delta_{z_i} and ν=∑iniδzi\nu=\sum_i n_i\delta_{z_i} be locally finite nonnegative atomic measures on one finite or countable set of distinct divisorial points in VV. Suppose bounded classical zero-boundary solutions uμ,uνu_\mu,u_\nu exist. For κi=Cap⁡[1]({zi};V)>0,d+=∑mi>ni(mi−niκi)1/n,d−=∑ni>mi(ni−miκi)1/n,\kappa_i=\operatorname{Cap}_{[1]}(\{z_i\};V)>0, \qquad d_+=\sum_{m_i>n_i}\left(\frac{m_i-n_i}{\kappa_i}\right)^{1/n}, \quad d_-=\sum_{n_i>m_i}\left(\frac{n_i-m_i}{\kappa_i}\right)^{1/n}, assume d++d−<∞d_++d_-<\infty. Then

−d+≤uμ−uν≤d−,∥uμ−uν∥L∞(V)≤max⁡(d+,d−)≤∑i(∣mi−ni∣κi)1/n.(9)-d_+\leq u_\mu-u_\nu\leq d_-, \qquad \|u_\mu-u_\nu\|_{L^\infty(V)} \leq\max(d_+,d_-) \leq\sum_i\left(\frac{|m_i-n_i|}{\kappa_i}\right)^{1/n}. \tag{9}

In particular this applies to the finite atomic solutions and to any two countable solutions constructed under (7) on their combined support.

Proof. Put α=(μ−ν)+\alpha=(\mu-\nu)_+ and β=(ν−μ)+\beta=(\nu-\mu)_+, with the positive and negative parts taken coefficient by coefficient. The summability hypothesis and the preceding corollary produce zero-boundary solutions G(α),G(β)G(\alpha),G(\beta) with −d+≤G(α)≤0,−d−≤G(β)≤0.-d_+\leq G(\alpha)\leq0, \qquad -d_-\leq G(\beta)\leq0. Positivity of every mixed BT term gives (D(uν+G(α)))n≥ν+α≥μ.(D(u_\nu+G(\alpha)))^n\geq\nu+\alpha\geq\mu. The measure-order lemma therefore yields uν+G(α)≤uμu_\nu+G(\alpha)\leq u_\mu. Interchanging μ,ν\mu,\nu gives uμ+G(β)≤uνu_\mu+G(\beta)\leq u_\nu. The displayed bounds on G(α),G(β)G(\alpha),G(\beta) prove the conclusion. This proof does not need the existence of a separate solution for μ+ν\mu+\nu. ◻

On the extendible-exhaustion domains of Theorem “Continuity of the finite-point Dirichlet envelopes”, every finite partial solution extends continuously to V‾\overline V. The uniform tail bound consequently makes every countable solution satisfying (7) continuous on V‾\overline V as well. This last statement uses that additional domain hypothesis; the uniform estimate itself holds on every domain fixed at the start of this section.

Contact and unique depth parameters for positive atomic data

Keep the fixed domain VV, the distinct divisorial points z1,…,zrz_1,\ldots,z_r, and the actual envelopes UaU_a from Section “Nonzero atomic Dirichlet problems on one fixed domain”. The statements below are DIRECT_COROLLARY of that construction and local maximality. They do not require interior continuity of UaU_a.

Proposition (Complementarity at a finite point obstacle). For every a∈[0,∞)ra\in[0,\infty)^r and every index ii, ci(a)>0⟹Ua(zi)=−ai.c_i(a)>0\quad\Longrightarrow\quad U_a(z_i)=-a_i. Equivalently, the atomic measure of UaU_a is supported on those point constraints at which equality holds.

Proof. Suppose that Ua(zi)<−aiU_a(z_i)<-a_i, and write δ=−ai−Ua(zi)>0\delta=-a_i-U_a(z_i)>0. Choose an ordinary W⋐VW\Subset V containing ziz_i and no other zjz_j. Let v∈Cb(W)v\in\mathcal C_b(W) agree with UaU_a off a compact subset L⋐WL\Subset W. The function max⁡(v,Ua)−Ua\max(v,U_a)-U_a vanishes off LL and is bounded there, so it has a finite upper bound M≥0M\geq0. Choose 0<t≤10<t\leq1 with tM<δtM<\delta. On WW put w=(1−t)Ua+tmax⁡(v,Ua),w=(1-t)U_a+t\max(v,U_a), and set w=Uaw=U_a off LL. The formulas agree on the ordinary overlap. Positive linear combinations, the maximum rule, and ordinary gluing show that ww is classical psh. It is bounded, has ordinary boundary limit zero, and satisfies w≥Uaw\geq U_a. The boundary maximum principle gives w≤0w\leq0. At ziz_i one has w(zi)≤Ua(zi)+tM<−ai;w(z_i)\leq U_a(z_i)+tM<-a_i; at every other constraint point w=Uaw=U_a. Thus w∈Faw\in\mathcal F_a, using w≥Ua≥−A(a)w\geq U_a\geq-A(a) for the lower bound. The actual greatest-member property implies w≤Uaw\leq U_a. Since t>0t>0, it follows that v≤Uav\leq U_a.

This proves local maximality of UaU_a on WW. The proved maximality criterion gives (DUa)n∣W=0(DU_a)^n|_W=0, so ci(a)=0c_i(a)=0. The contrapositive is the assertion. Only actual constraints were used; no quasi-everywhere constraint was substituted. ◻

Corollary (Unique depth coordinates for strictly positive masses). Let A={a∈[0,∞)r:ci(a)>0 for all i}.\mathcal A=\{a\in[0,\infty)^r:c_i(a)>0\text{ for all }i\}. Then A\mathcal A is an open subset of (0,∞)r(0,\infty)^r, and the full mass map c:A⟶(0,∞)r,a⟼(c1(a),…,cr(a)),c:\mathcal A\longrightarrow(0,\infty)^r, \qquad a\longmapsto(c_1(a),\ldots,c_r(a)), is a homeomorphism. Its inverse is the actual point-depth map (m1,…,mr)⟼(−G(∑imiδzi)(zj))j=1r.(m_1,\ldots,m_r)\longmapsto \bigl(-G(\textstyle\sum_i m_i\delta_{z_i})(z_j)\bigr)_{j=1}^r. Consequently the normalized mass map FF restricts to a homeomorphism from A∩Δr−1\mathcal A\cap\Delta^{r-1} onto the relative interior of Δr−1\Delta^{r-1}. For a target y∈relint⁡Δr−1y\in\operatorname{relint}\Delta^{r-1} its unique normalized depth vector is ai(y)=−G(∑jyjδzj)(zi)∑ℓ=1r−G(∑jyjδzj)(zℓ).a_i(y)= \frac{-G(\sum_jy_j\delta_{z_j})(z_i)} {\sum_{\ell=1}^r-G(\sum_jy_j\delta_{z_j})(z_\ell)}.

Proof. Continuity of the coefficients makes A\mathcal A open, and face preservation puts it inside (0,∞)r(0,\infty)^r. Given a strictly positive mass vector, the finite atomic theorem constructs its unique solution as tUa=UtatU_a=U_{ta} for some t>0t>0 and a∈Δr−1a\in\Delta^{r-1}. Its mass vector is exactly the target, so ta∈Ata\in\mathcal A. Complementarity gives −Uta(zi)=tai-U_{ta}(z_i)=ta_i for every ii. This proves surjectivity and the displayed inverse formula.

If c(a)=c(b)c(a)=c(b) for a,b∈Aa,b\in\mathcal A, actual Dirichlet uniqueness gives Ua=UbU_a=U_b. Complementarity then gives ai=−Ua(zi)=−Ub(zi)=bia_i=-U_a(z_i)=-U_b(z_i)=b_i for every ii. Thus the map is injective. The inverse is continuous because the fixed-support solution map is continuous in the uniform norm, and evaluation at a fixed point is continuous for that norm. The mass map itself is continuous by its proved local Lipschitz estimate.

For normalized masses, scaling the unique solution scales all its depths by the same positive factor. Dividing those depths by their sum gives the displayed point of the simplex. The sum is positive because the full depth vector is in A\mathcal A. The formula is continuous, and homogeneity shows it is the inverse of the restricted normalized map. ◻

This does not assert that the mass map is injective on the whole nonnegative cone. A constraint with strict slack has zero mass, and its prescribed depth need not be recoverable from the solution. The inverse theorem is restricted to the region on which every prescribed atom has positive mass.

Continuous envelopes with divisorial point constraints

This section uses the continuous ample-envelope theorem of B and the divisorial no-jump theorem proved in R2. The first result is a DIRECT_COROLLARY of those inputs. The local application also uses the explicit order-preserving common-collar construction. No continuity theorem for general homogeneous Dirichlet solutions is assumed or obtained.

A continuous insertion argument

Lemma (Continuous insertion with a uniform error). Let ZZ be a compact Hausdorff space. Let aa be a bounded upper semicontinuous real function, and let ff be a bounded lower semicontinuous real function with a≤fa\leq f. For every ϵ>0\epsilon>0 there is a continuous function gg such that a−ϵ≤g≤f.a-\epsilon\leq g\leq f.

Proof. For each x∈Zx\in Z, choose a neighborhood OxO_x on which a(y)<a(x)+ϵ/4a(y)<a(x)+\epsilon/4 and f(y)>f(x)−ϵ/4f(y)>f(x)-\epsilon/4. The constant cx=a(x)−ϵ/2c_x=a(x)-\epsilon/2 then satisfies a(y)−ϵ<cx<f(y)a(y)-\epsilon<c_x<f(y) on OxO_x. Take a finite subcover and a continuous partition of unity subordinate to it. The same partition-weighted sum of the constants cxc_x is the required gg. ◻

Let Y/KY/K be smooth, projective and geometrically integral, and let LL be ample. All metric inequalities below are inequalities of local weights; their differences are globally defined real functions. The fixed field and its absolute value are unchanged.

Theorem (A lower semicontinuous obstacle with an attained envelope). Let ff be a lower semicontinuous metric on LL whose difference from a continuous reference metric is bounded. Put hf=sup⁡{r:r is a bounded classical metric on L,r≤f},Pf=hf∗.h_f=\sup\{r:r\text{ is a bounded classical metric on }L, r\leq f\}, \qquad P_f=h_f^*. If Pf≤fP_f\leq f at every point, then hf=Pfh_f=P_f and PfP_f is continuous. More precisely, for every ϵ>0\epsilon>0 there is a continuous obstacle gϵ≤fg_\epsilon\leq f such that Pf−ϵ≤PLgϵ≤Pf.P_f-\epsilon\leq P_Lg_\epsilon\leq P_f.

Proof. Choose a continuous semipositive reference metric r0r_0. A sufficiently negative translate of r0r_0 is a competitor, and every competitor is bounded above by r0+Cr_0+C for one constant CC. Replacing competitors by their maxima with one such negative translate gives a uniformly bounded family without changing the supremum. Corollary “Upper envelopes of bounded projective families” therefore makes PfP_f a bounded classical metric. Under the stated hypothesis it is itself a competitor. Thus hf=Pfh_f=P_f pointwise.

Apply Lemma “Continuous insertion with a uniform error” to Pf−r0P_f-r_0 and f−r0f-r_0, and add r0r_0 back. This gives a continuous metric gϵg_\epsilon with Pf−ϵ≤gϵ≤fP_f-\epsilon\leq g_\epsilon\leq f. By B, Theorem C.6, PLgϵP_Lg_\epsilon is continuous and semipositive. The actual-competitor comparison Lemma “A bounded classical competitor lies below the continuous envelope” gives Pf−ϵ≤PLgϵP_f-\epsilon\leq P_Lg_\epsilon. On the other hand, PLgϵ≤gϵ≤fP_Lg_\epsilon\leq g_\epsilon\leq f, so it is a competitor for hfh_f and is at most PfP_f. These uniform inequalities prove continuity of PfP_f. ◻

Corollary (Discontinuities confined to divisorial points). In the preceding theorem, suppose that every discontinuity point of ff is ambient divisorial. Then Pf≤fP_f\leq f pointwise, and the conclusion of the theorem holds.

Proof. At a continuity point of ff, upper regularization of hf≤fh_f\leq f gives Pf≤fP_f\leq f. At a divisorial point zz, the local divisorial no-jump result, Corollary “Upper regularization preserves divisorial values”, gives Pf(z)=hf(z)≤f(z)P_f(z)=h_f(z)\leq f(z). These two cases cover the space. ◻

Corollary (Finite divisorial obstacles). Let FF be a continuous semipositive metric on LL, let z1,…,zNz_1,\ldots,z_N be distinct divisorial points, and let ai≥0a_i\geq0. The greatest bounded classical metric subject to r≤F,r(zi)≤F(zi)−ai(1≤i≤N)r\leq F,\qquad r(z_i)\leq F(z_i)-a_i\quad(1\leq i\leq N) exists and is continuous. Denote it by PaP_a. Then F−max⁡iai≤Pa≤F,∥Pa−Pb∥∞≤max⁡i∣ai−bi∣.F-\max_i a_i\leq P_a\leq F, \qquad \|P_a-P_b\|_\infty\leq\max_i|a_i-b_i|.

Proof. The obstacle fa=F−∑iai1{zi}f_a=F-\sum_i a_i\mathbf1_{\{z_i\}} is lower semicontinuous and is continuous off the finite divisorial set. The preceding corollary applies. The displayed lower bound is an admissible continuous metric. If δ=max⁡i∣ai−bi∣\delta=\max_i|a_i-b_i|, then Pa−δP_a-\delta is admissible for the bb-constraints, whence Pa−δ≤PbP_a-\delta\leq P_b. Interchange aa and bb. ◻

The lower semicontinuity hypothesis is deliberate. The point obstacle has downward jumps and is generally not upper semicontinuous. It must not be inserted into the upper semicontinuous-obstacle theorem without the preceding argument.

An application on domains with an extendible exhaustion

Lemma (A framed smooth compactification of an affine variety). A smooth geometrically integral affine variety Y/KY/K has a smooth projective compactification Y‾\overline Y, an ample line AA on Y‾\overline Y, and a section ss with Y‾∖div⁡(s)=Y\overline Y\setminus\operatorname{div}(s)=Y. After taking a power, AA is very ample and ss is one coordinate section of its projective embedding.

Proof. Embed YY in affine space, normalize its projective closure, and pull back the hyperplane line and its section at infinity. Normalization is finite and is an isomorphism on the smooth open YY. The pulled-back line is ample, and the nonvanishing locus of its section is exactly YY.

Use the characteristic-zero projective resolution preserving YY already invoked in B, Proposition 4.5. The ample-boundary property can be retained at each blowup in this resolution. Indeed, suppose HH is ample and sHs_H has nonvanishing locus YY, and let π:Z′→Z\pi:Z'\to Z be a blowup of an ideal II whose cosupport is in the boundary. Write IOZ′=OZ′(−E)I\mathcal O_{Z'}=\mathcal O_{Z'}(-E). For sufficiently large NN, the line H′=π∗HN⊗OZ′(−E)H'=\pi^*H^N\otimes\mathcal O_{Z'}(-E) is ample. Increasing NN if necessary, the section sHNs_H^N belongs locally to IHNI H^N: this follows from the containment of the cosupport in div⁡(sH)\operatorname{div}(s_H) and the Noetherian radical criterion, using finitely many affine charts. The induced section of H′H' has divisor Nπ∗div⁡(sH)−EN\pi^*\operatorname{div}(s_H)-E. For still larger NN this divisor is effective and has exactly the full boundary as its support. Thus the property persists. The same finite-pullback argument handles a normalization if one is inserted. The resulting smooth projective compactification has the required ample line and section. A sufficiently high very ample power, with the powered section included in a basis, gives the last assertion. ◻

Let Y/KY/K now be smooth, geometrically integral and affine. Suppose that ρ\rho is a continuous classical psh function on YanY^{\mathrm{an}} and that VV is a connected component of {ρ<0}\{\rho<0\} with V⋐Yan,∅≠∂V⊂{ρ=0}.V\Subset Y^{\mathrm{an}},\qquad \varnothing\ne\partial V\subset\{\rho=0\}. This includes the norm-box exhaustion domains constructed from log⁡max⁡(1,∣g1∣,…,∣gN∣)−ϵ\log\max(1,|g_1|,\ldots,|g_N|)-\epsilon in R2. It is stronger than merely having an exhaustion defined on VV.

For distinct divisorial points zi∈Vz_i\in V and ai≥0a_i\geq0, let Ua=sup⁡{u∈Cb(V):u is bounded, u≤0, u(zi)≤−ai for every i}.U_a=\sup\{u\in\mathcal C_b(V):u\text{ is bounded},\ u\leq0, \ u(z_i)\leq-a_i\text{ for every }i\}. The divisorial no-jump theorem makes this actual supremum classical and admissible, as in the finite-atomic construction.

Theorem (Continuity of the finite-point Dirichlet envelopes). Under the displayed extendible-exhaustion hypotheses, UaU_a is continuous on V‾\overline V, with value zero on ∂V\partial V. Consequently the finite divisorial-atomic Dirichlet solutions with constant boundary values constructed in this round are continuous on V‾\overline V for these domains.

Proof. Choose C>0C>0 with Cρ(zi)≤−aiC\rho(z_i)\leq-a_i for every ii. The bounded function Cρ∣VC\rho|_V is a competitor. Hence Cρ≤Ua≤0on V.C\rho\leq U_a\leq0\quad\hbox{on }V. These inequalities already give the claimed boundary values.

For δ>0\delta>0, define a function on YanY^{\mathrm{an}} by wδ={max⁡(Ua,Cρ+δ),on V,Cρ+δ,off V.w_\delta= \begin{cases} \max(U_a,C\rho+\delta),&\text{on }V,\\ C\rho+\delta,&\text{off }V. \end{cases} Near every point of ∂V\partial V, one has Cρ+δ>0≥UaC\rho+\delta>0\geq U_a on the part lying in VV. Thus the two prescriptions agree on an ordinary neighborhood of the boundary. Ordinary gluing and the maximum rule show that wδw_\delta is classical psh. As δ↓0\delta\downarrow0, these functions decrease locally boundedly to the classical psh function w={Ua,on V,Cρ,off V.w= \begin{cases}U_a,&\text{on }V,\\ C\rho,&\text{off }V. \end{cases} Moreover w≤g:=max⁡(0,Cρ)w\leq g:=\max(0,C\rho), and gg is continuous classical psh on YanY^{\mathrm{an}}.

Use the framed compactification of Lemma “A framed smooth compactification of an affine variety”. Its one affine coordinate chart contains the entire YY. Large upper bounds on its finitely many affine coordinates give nested compact norm boxes whose innermost ordinary interior contains V‾\overline V; all outer boxes remain in YanY^{\mathrm{an}}. The functions ww and gg are defined on a neighborhood of the outer box and have common bounds there. Apply the explicit common-collar construction from B, Lemma D.2, in the order-preserving form used in Theorem “Local bounded upper-envelope theorem”. For clarity, in one fixed frame that construction is G(f)=max⁡(f+c,mb)\mathcal G(f)=\max(f+c,mb) on the outer box and the same fixed metric off its middle box. Here b=0b=0 on the inner box, the common translation cc makes f+c>0f+c>0 there, and mm forces equality with the fixed metric on the outer collar. It gives bounded classical metrics R=G(w)≤F=G(g)R=\mathcal G(w)\leq F=\mathcal G(g) on one ample line. The metric FF is continuous and classical. B, Theorem 4.4, gives decreasing continuous model-semipositive metrics converging to FF; compactness and Dini’s theorem make this convergence uniform. Thus FF is a continuous semipositive metric, as required by the projective finite-point corollary. Their weights on VV are respectively Ua+cU_a+c and cc, since g=0g=0 on VV.

Let PaP_a be the continuous projective finite-point envelope of Corollary “Finite divisorial obstacles”, with upper obstacle FF and depths aia_i. The metric RR is a competitor, so R≤PaR\leq P_a. Conversely, the local weight Pa−cP_a-c on VV is bounded classical, is at most zero, and satisfies the actual point constraints. It is therefore at most UaU_a. On VV this gives Ua+c=R≤Pa≤Ua+c.U_a+c=R\leq P_a\leq U_a+c. Thus Ua=Pa−cU_a=P_a-c on VV, proving interior continuity. Combined with the previously established ordinary boundary limits, this proves continuity on V‾\overline V. ◻

Scope.

The projective envelope statement is a direct consequence of B and R2. The local theorem is a DIRECT_COROLLARY after the extension and common-collar construction above. Its extra exhaustion hypothesis is explicit. This argument neither proves interior continuity for the general peak-barrier Perron solution of R2 nor replaces arbitrary boundary values by constant ones.

Non-atomic Dirichlet data on fixed Gauss skeleton boxes

Fix the field, the smooth integral affine Y/KY/K, and the connected ordinary exhaustion domain V⋐YanV\Subset Y^{\mathrm{an}} of (H). In particular n=dim⁡Y≥1n=\dim Y\geq1, ρ<0\rho<0 on VV, and ρ\rho is continuous on V‾\overline V with ordinary boundary value zero. The domain remains fixed throughout. All Monge–Ampère measures use [1][1], the original absolute value, and the original CLD tests and calibration.

The new step is a compactness mechanism for a specified class of supports. It proves weak-measure stability in that class; it does not assume such stability. The main existence theorem is recorded as NEW_PROVED, relative to the finite atomic construction and the preceding R1/R2 results.

The support geometry and its function estimates

For a bounded open rectangular box I⊂RnI\subset\mathbf R^n, put A(I)={x∈(Gm,Kn)an:(log⁡∣T1(x)∣,…,log⁡∣Tn(x)∣)∈I}.\mathbb A(I)= \{x\in(\mathbf G_{m,K}^n)^{\mathrm{an}}: (\log|T_1(x)|,\ldots,\log|T_n(x)|)\in I\}. Its Gauss section σ:I→A(I)\sigma:I\to\mathbb A(I) is characterized by ∣∑m∈ZncmTm∣(σ(x))=max⁡m∣cm∣em⋅x\left|\sum_{m\in\mathbf Z^n}c_mT^m\right|(\sigma(x)) =\max_m |c_m|e^{m\cdot x} for Laurent polynomials. Suppose that, for 1≤ℓ≤L1\leq\ell\leq L, there are KK-analytic isomorphisms onto ordinary opens Φℓ:A(Iℓ)→∼Ωℓ⊂V\Phi_\ell:\mathbb A(I_\ell)\xrightarrow{\sim}\Omega_\ell \subset V and compact rectangular boxes Jℓ⋐IℓJ_\ell\Subset I_\ell, each with nonempty Euclidean interior. Set

Kℓ=Φℓ(σ(Jℓ)),K=⋃ℓ=1LKℓ⋐V.(S)K_\ell=\Phi_\ell(\sigma(J_\ell)),\qquad K=\bigcup_{\ell=1}^L K_\ell\Subset V. \tag{S}

The larger full polyannulus, and the positive margin between its box and JℓJ_\ell, are part of the hypotheses. A compact subset merely called a skeleton is not being substituted for these data. The measure may be supported on any closed subset of this fixed KK; the approximating divisorial points below are allowed to lie anywhere in KK. Thus an irrational-radius face need not itself contain a dense set of divisorial points.

Lemma (Uniform continuity on an inner Gauss box). If u∈Cb(A(I))u\in\mathcal C_b(\mathbb A(I)) and −B≤u≤0-B\leq u\leq0, then Fu(x)=u(σ(x))F_u(x)=u(\sigma(x)) is convex in each coordinate separately. For J⋐IJ\Subset I as above there is a constant CI,JC_{I,J}, independent of uu and BB, such that ∣Fu(x)−Fu(y)∣≤CI,JB∥x−y∥1(x,y∈J).|F_u(x)-F_u(y)|\leq C_{I,J}B\|x-y\|_1 \quad(x,y\in J). In particular a uniformly bounded family of classical functions restricts to an equicontinuous family on KK in (S).

Proof. Fix a coordinate ii and fix all the other real coordinates xjx_j. Let F/KF/K be the completed residue field of the weighted Gauss point with these n−1n-1 coordinates. For n=1n=1, take F=KF=K. The corresponding coordinate elements cj∈F×c_j\in F^\times have ∣cj∣=exj|c_j|=e^{x_j} and the Gauss independence property. The map from the FF-annulus with log⁡∣t∣∈Ii\log|t|\in I_i given by Ti=t,Tj=cj(j≠i)T_i=t,\qquad T_j=c_j\quad(j\ne i) lands in A(I)F\mathbb A(I)_F. At its radius-ese^s Gauss point, the image in the original KK-space is σ(x1,…,s,…,xn)\sigma(x_1,\ldots,s,\ldots,x_n). This follows directly by grouping a Laurent polynomial according to its exponent of TiT_i and applying the Gauss norm first to the remaining coefficients.

The all-complete-extensions classical hypothesis makes the pullback of uu classical subharmonic on that annulus. Let a<ba<b be any two logarithmic radii strictly inside IiI_i. Subtract the affine function of log⁡∣t∣\log|t| interpolating its values at the two Gauss endpoints. This affine function is pluriharmonic, so the difference is still classical psh. Its values at the two Shilov points of the closed subannulus are zero. The affinoid maximum principle [GR25, Theorem 6.4] makes the difference nonpositive. This is exactly the convexity inequality for the iith coordinate of FuF_u.

A real convex function with range in [−B,0][-B,0] on an interval has uniformly bounded secant slopes on every inner interval: comparison with secants to the outer endpoints bounds their absolute values by the oscillation divided by the margin. If δ>0\delta>0 is the smallest coordinate margin from JJ to the endpoints of II, the bound 2B/δ2B/\delta is sufficient. Change one coordinate at a time inside JJ and add these estimates. This proves the displayed inequality.

Each Gauss section is continuous and is a homeomorphism onto its image, since its coordinates recover xx. Pullback by Φℓ\Phi_\ell preserves the classical definition. Apply the estimate on each of the finitely many boxes. All auxiliary complete extensions above were used only for the curve tests of functions, not for any asserted base change of currents. ◻

Lemma (A uniform singleton capacity on the original domain). There is cK>0c_K>0, depending on the data in (S) and on VV, such that Cap⁡[1]({z};V)≥cK(z∈K).\operatorname{Cap}_{[1]}(\{z\};V)\geq c_K \quad(z\in K). Moreover KK has a dense set of ambient divisorial points over the original field KK.

Proof. First work on A(Iℓ)\mathbb A(I_\ell). Choose Aℓ>0A_\ell>0 with Aℓ>∑i=1n(sup⁡(Iℓ)i−inf⁡(Jℓ)i).A_\ell>\sum_{i=1}^n (\sup (I_\ell)_i-\inf(J_\ell)_i). For a∈Jℓa\in J_\ell the continuous classical function ha=1Aℓ∑i=1nmax⁡(0,log⁡∣Ti∣−ai)h_a=\frac1{A_\ell}\sum_{i=1}^n \max(0,\log|T_i|-a_i) lies in [0,1][0,1] on the same A(Iℓ)\mathbb A(I_\ell), independently of aa. Its ordinary Monge–Ampère measure is

(Dha)n=n!Aℓnδσ(a).(10)(Dh_a)^n=\frac{n!}{A_\ell^n}\delta_{\sigma(a)}. \tag{10}

Here is a check of that normalization on all original tests. Replace each maximum by a smooth convex regularization γϵ(log⁡∣Ti∣−ai)\gamma_\epsilon(\log|T_i|-a_i), with second derivative a nonnegative approximate identity of mass one. The ordinary product calibration on the split torus gives, for every original compact smooth scalar test χ\chi, ∫A(Iℓ)χ⋀i=1nDγϵ(log⁡∣Ti∣−ai)=∫Iℓχ(σ(x))∏i=1nγϵ′′(xi−ai) dx.\int_{\mathbb A(I_\ell)}\chi \bigwedge_{i=1}^nD\gamma_\epsilon(\log|T_i|-a_i) =\int_{I_\ell}\chi(\sigma(x)) \prod_{i=1}^n\gamma_\epsilon''(x_i-a_i)\,dx. To check the formula when χ\chi requires additional moments, adjoin those moments to the coordinate map and subdivide its tropicalization. On cells whose projection to the coordinate box has rank less than nn, the displayed top form vanishes. On a full-rank cell the canonical calibration is the degree over the coordinate Gauss skeleton. The coordinate map here is the identity inclusion into the split torus, so its degree is one and its point over xx is precisely σ(x)\sigma(x). The construction of the canonical calibration and its compatibility with projections [CLD12, §3.5.1, Lemma 3.5.2, and §§3.7–3.8] therefore give the formula with the actual value χ(σ(x))\chi(\sigma(x)), also after every required refinement. For small ϵ\epsilon its support lies in IℓI_\ell, and the right side tends to χ(σ(a))\chi(\sigma(a)). Locally uniform BT convergence gives the corresponding mixed product of the maxima. Repeated coordinate factors vanish already for the smooth rank-one forms. Expansion of the nnth power leaves precisely n!n! distinct-coordinate terms, proving (10). No finite extension or change in the integration current was made.

Choose an intermediate open box Jℓ+J_\ell^+ with Jℓ⋐Jℓ+⋐IℓJ_\ell\Subset J_\ell^+\Subset I_\ell and put Wℓ=Φℓ(A(Jℓ+))W_\ell=\Phi_\ell(\mathbb A(J_\ell^+)). Its closure is compact in Ωℓ\Omega_\ell. The reverse direction of Theorem “Uniform local comparison with a fixed domain” provides a fixed Cℓ<∞C_\ell<\infty such that Cap⁡[1](E;Ωℓ)≤CℓCap⁡[1](E;V)(E⊂Wℓ).\operatorname{Cap}_{[1]}(E;\Omega_\ell) \leq C_\ell\operatorname{Cap}_{[1]}(E;V) \quad(E\subset W_\ell). The isomorphism Φℓ\Phi_\ell has degree one and preserves the original products and tests. Apply the last inequality to a singleton and use (10). One may take cK=min⁡ℓn!CℓAℓn>0.c_K=\min_\ell\frac{n!}{C_\ell A_\ell^n}>0. This lower bound uses the proved reverse collar comparison; ordinary domain monotonicity alone has the opposite direction.

Write λ=−log⁡∣π∣>0\lambda=-\log|\pi|>0 for a uniformizer of the fixed field. The points a∈Jℓ∩(λQ)na\in J_\ell\cap(\lambda\mathbf Q)^n are dense in JℓJ_\ell. Their Gauss points are the unique Shilov points of strict fixed-radius polyannuli: rational valuation thresholds are expressed by finitely many integral power inequalities over KK. They are therefore divisorial, and remain so on ordinary neighborhoods and under the analytic isomorphism, by [GM19, Proposition A.3 and Corollary A.6]. This gives a dense divisorial subset of KK, without extending the ground field used for the data or currents. ◻

From skeletal compactness to uniform compactness of solutions

Lemma (A common support determines the uniform distance). Let u,v∈Cb(V)u,v\in\mathcal C_b(V) be bounded, with ordinary boundary limit zero, and suppose both (Du)n(Du)^n and (Dv)n(Dv)^n are supported on a compact subset H⋐VH\Subset V. Then ∥u−v∥L∞(V)=sup⁡H∣u−v∣.\|u-v\|_{L^\infty(V)}=\sup_H|u-v|.

Proof. Put b=sup⁡H∣u−v∣b=\sup_H|u-v|. For every ϵ>b\epsilon>b, the Borel gap E={u<v−ϵ}E=\{u<v-\epsilon\} is disjoint from HH, so (Du)n(E)=0(Du)^n(E)=0. The boundary values make its closure compact in VV, and give every compact-gap condition required for pointwise ambient domination. That principle gives u≥v−ϵu\geq v-\epsilon on VV. Interchange u,vu,v and let ϵ↓b\epsilon\downarrow b. The reverse norm inequality follows from H⊂VH\subset V. No strict-gap set was treated as an ordinary open set. ◻

Lemma (Uniform bounds for finite atomic data in KK). Let ν\nu be a finite nonnegative sum of divisorial atoms in KK, and let uνu_\nu be its zero-boundary solution from Theorem “Finite atomic Dirichlet problem on the fixed domain”. Put M=ν(K)M=\nu(K) and dK=min⁡K(−ρ)>0,bK=max⁡(−1,ρ/dK).d_K=\min_K(-\rho)>0, \qquad b_K=\max(-1,\rho/d_K). Then

∥uν∥∞≤(M/cK)1/n,(M/cK)1/nbK≤uν≤0.(11)\|u_\nu\|_\infty\leq(M/c_K)^{1/n}, \qquad (M/c_K)^{1/n}b_K\leq u_\nu\leq0. \tag{11}

In particular all such solutions with bounded total mass have one common ordinary boundary barrier.

Proof. The assertion is immediate for M=0M=0. Otherwise discard zero-mass points. The finite atomic construction writes uν=tUau_\nu=tU_a, with t>0t>0 and a∈Δr−1a\in\Delta^{r-1}. Put A=max⁡iai>0A=\max_i a_i>0 and choose ii with ai=Aa_i=A. The defining lower bound and the actual constraint give Ua(zi)=−AU_a(z_i)=-A. Consequently B=∥uν∥∞=tAB=\|u_\nu\|_\infty=tA. The function uν/Bu_\nu/B lies in [−1,0][-1,0] and equals −1-1 at ziz_i. It is a competitor for the singleton extremal, so uν/B≤U{zi}≤0u_\nu/B\leq U_{\{z_i\}}\leq0. Both functions have ordinary boundary limit zero. The exact total-mass order Lemma “Order of total Monge–Ampère masses” and the compact-extremal formula give cK≤Cap⁡[1]({zi};V)=∫V(DU{zi})n≤∫V(D(uν/B))n=MBn.c_K\leq\operatorname{Cap}_{[1]}(\{z_i\};V) =\int_V(DU_{\{z_i\}})^n \leq\int_V(D(u_\nu/B))^n=\frac{M}{B^n}. All masses here are finite; the cited order theorem in fact also permits infinite masses without cancellation. This proves the norm bound.

At every constraint point zj∈Kz_j\in K, the function bKb_K equals −1-1, which is at most −aj/A-a_j/A. The greatest-member description of Ua/AU_{a/A} therefore gives bK≤Ua/A=uν/Bb_K\leq U_{a/A}=u_\nu/B. Since bK≤0b_K\leq0 and B≤(M/cK)1/nB\leq(M/c_K)^{1/n}, the second bound follows. The same bKb_K works for every choice of the finite support in KK. ◻

Arbitrary finite measures on the prescribed skeleton

Theorem (A non-atomic fixed-domain Dirichlet class). Assume (S). For every finite nonnegative Radon measure μ\mu supported on KK and every b∈Rb\in\mathbf R, there is a unique bounded actual classical psh function u∈Cb(V)u\in\mathcal C_b(V) with (Du)n=μ,lim⁡x→ξx∈Vu(x)=b(ξ∈∂V).(Du)^n=\mu, \qquad\lim_{\substack{x\to\xi\\x\in V}}u(x)=b \quad(\xi\in\partial V). The bounds (11) hold for u−bu-b with M=μ(K)M=\mu(K).

If μj→μ\mu_j\to\mu weakly as finite measures on this same compact KK, and bj→bb_j\to b, then their solutions converge uniformly on VV. Their weighted measures also satisfy uj(Duj)n⟶u(Du)nu_j(Du_j)^n\longrightarrow u(Du)^n against every compact continuous scalar test, hence against every original compact smooth scalar test.

If VV additionally satisfies the extendible-exhaustion hypotheses of Theorem “Continuity of the finite-point Dirichlet envelopes”, all these solutions are continuous on V‾\overline V.

Proof. First take b=0b=0. The compact space KK is metrizable; it is a finite union of compact Gauss boxes. Its dense divisorial subset permits finite atomic measures μj\mu_j supported in KK, with μj(K)=μ(K)\mu_j(K)=\mu(K), that converge weakly to μ\mu. For example take finite Borel partitions of successively smaller diameter and replace each part’s mass by a nearby divisorial point in KK. Uniform continuity of functions in C(K)C(K) verifies the asserted weak convergence.

Let uju_j be the finite atomic solutions. Their uniform bound and common boundary barrier are (11). Lemma “Uniform continuity on an inner Gauss box” makes their restrictions equicontinuous on each KℓK_\ell. Arzelà–Ascoli and a finite diagonal extraction give a subsequence uniformly Cauchy on all of KK. Lemma “A common support determines the uniform distance” then makes this subsequence uniformly Cauchy on the entire original domain VV. Denote its uniform limit by uu.

Uniform limits here are classical. Explicitly, take a further subsequence with uniform errors at most 4−j4^{-j}; adding 2⋅4−j2\cdot4^{-j} makes it decrease to uu. The decreasing closure of the classical class applies. The common finite bound makes uu locally bounded, and the common bKb_K barrier gives its ordinary zero boundary limit.

Uniform convergence implies local capacity convergence on every prescribed small norm box. The proved fixed-background BT stability, Theorem “Weighted stability in local relative capacity”, therefore gives (Duj)n→(Du)n(Du_j)^n\to(Du)^n on all original compact tests. But (Duj)n=μj→μ(Du_j)^n=\mu_j\to\mu on those tests. The associated Radon measures are determined by the original smooth scalar tests, so (Du)n=μ(Du)^n=\mu. This passage uses uniform convergence of the solutions obtained above, not an assumed stability under weak convergence of their right-hand sides. The bounds pass to the uniform limit. Pointwise ambient uniqueness, with the two boundary compact-gap conditions, gives uniqueness among all bounded classical solutions.

Now let arbitrary μj→μ\mu_j\to\mu on KK. Their total masses are bounded. The solutions just constructed satisfy the same uniform bounds, boundary barrier, and skeletal equicontinuity. From any subsequence, the preceding compactness and support argument extracts a further subsequence converging uniformly on VV. Uniform BT stability identifies its measure with μ\mu, and uniqueness identifies its limit with uu. Thus the entire sequence converges uniformly.

Every bounded classical function is continuous on KK, by Lemma “Uniform continuity on an inner Gauss box”. For a fixed continuous compact scalar χ\chi, write ∫χuj dμj−∫χu dμ=∫Kχ(uj−u) dμj+∫Kχu d(μj−μ).\int\chi u_j\,d\mu_j-\int\chi u\,d\mu =\int_K\chi(u_j-u)\,d\mu_j +\int_K\chi u\,d(\mu_j-\mu). The first term tends to zero by the uniform convergence and mass bound; the second tends to zero because χu\chi u is continuous on KK. This proves the stated weighted limit with the actual representatives. Adding the constants bjb_j proves all assertions for varying constant boundary values.

Under the additional extendible-exhaustion hypotheses, every finite atomic approximant extends continuously to V‾\overline V, with value zero on the boundary. Uniform convergence on VV is also uniform on V‾\overline V after those extensions. Their limit is therefore continuous on V‾\overline V. ◻

For fixed MM, the solutions with μ≥0\mu\geq0, support in KK, and μ(K)≤M\mu(K)\leq M form a compact set for the uniform norm: the finite positive measures of bounded mass on compact KK are weakly compact, and the theorem gives a continuous solution map. This is compactness of this prescribed-measure class, not compactness of every uniformly bounded psh family.

Example (Non-atomic data in every dimension). Let R>1R>1 and fix V={max⁡i∣Ti∣<R}⊂(AKn)an,ρ=log⁡max⁡(1,∣T1∣,…,∣Tn∣)−log⁡R.V=\{\max_i|T_i|<R\}\subset(\mathbf A_K^n)^{\mathrm{an}}, \qquad \rho=\log\max(1,|T_1|,\ldots,|T_n|)-\log R. This is an extendible-exhaustion domain. Fix finite a<b<log⁡Ra<b<\log R and let K=σ([a,b]n)K=\sigma([a,b]^n). Every finite positive Radon measure on this KK has a unique bounded classical Dirichlet solution with any constant boundary value, continuous on V‾\overline V. For instance μ=σ∗(f(x) dx),0≤f∈L1([a,b]n),\mu=\sigma_*(f(x)\,dx),\qquad 0\leq f\in L^1([a,b]^n), is non-atomic and nonzero when f≢0f\not\equiv0 in L1L^1. No equation on a new domain adapted to its mass distribution has been used. The local testers in (10) are already bounded on this whole VV, so with A=n(log⁡R−a)A=n(\log R-a) one can use cK=n!/Anc_K=n!/A^n in the uniform norm bound.

The remaining diffuse-data scope.

The argument has not proved existence for every compactly supported measure, or for every measure dominated by (Dw)n(Dw)^n for a continuous bounded classical ww. Outside the stated Gauss-box geometry, the uniform skeletal continuity needed for Arzelà–Ascoli has not been supplied. The earlier moving-branch counterexample rules out replacing that step by unrestricted compactness of bounded psh families. Nor may an atomization of a non-atomic measure be assumed to remain dominated by that measure: a positive point mass already contradicts such domination on its singleton. The present proof obtains its testing continuity and uniform compactness from the specified support geometry instead of assuming a general variational compactness theorem.

Moving atoms and failure of weak-data Dirichlet stability

Let π\pi be a uniformizer, r=∣π∣r=|\pi|, λ=−log⁡r\lambda=-\log r, R=r−1R=r^{-1}, and Ω={∣Tl∣<R:1≤l≤n}\Omega=\{|T_l|<R:1\leq l\leq n\}. Choose pairwise distinct monic irreducibles P‾i\overline P_i of degrees di≥2d_i\geq2 over the residue field, and monic integral lifts PiP_i. Let ξi\xi_i be the Shilov point of {∣Pi(T1)∣≤rdi}\{|P_i(T_1)|\leq r^{d_i}\} in the original analytic affine line. It is a single original KK-analytic point of residual degree did_i. After the unramified splitting extension, it consists of the conjugate Gauss points of discs centered at the roots of PiP_i and of radius rdir^{d_i}.

Let ziz_i be the product Gauss point with first factor ξi\xi_i and remaining coordinate radii rr, and let zz be the product Gauss point with first radius one and remaining radii rr. Let yy have all coordinate radii one. The distinct residue branches give ξi→ζ0,1\xi_i\to\zeta_{0,1}, hence zi→zz_i\to z. For completeness, a fixed polynomial has only finitely many irreducible residue factors; its norm at ξi\xi_i eventually equals its unit Gauss norm. Uniform polynomial approximation on compact polydiscs gives the asserted convergence for the product points.

Define

gi=max⁡{di−1log⁡∣Pi(T1)∣−λ, log⁡∣T2∣−λ,…,log⁡∣Tn∣−λ, −2λ}.g_i=\max\left\{d_i^{-1}\log|P_i(T_1)|-\lambda,\, \log|T_2|-\lambda,\ldots,\log|T_n|-\lambda,\,-2\lambda\right\}.

The list of remaining coordinates is empty when n=1n=1.

Theorem (Explicit zero-boundary weak instability). The functions gig_i are continuous classical psh on Ω‾\overline\Omega, take values in [−2λ,0][-2\lambda,0], have ordinary boundary value zero, and satisfy (Dgi)n=δzi.(Dg_i)^n=\delta_{z_i}. They converge pointwise, indeed eventually at each point, to g=max⁡{0,log⁡∣T1∣,…,log⁡∣Tn∣}−λ,(Dg)n=δy.g=\max\{0,\log|T_1|,\ldots,\log|T_n|\}-\lambda,\qquad (Dg)^n=\delta_y. Put B=2(n−1)/nλB=2^{(n-1)/n}\lambda and H=max⁡{Bλ(log⁡∣T1∣−λ), B2λ(log⁡∣Tl∣−λ) (2≤l≤n), −B}.H=\max\left\{ \frac B\lambda(\log|T_1|-\lambda),\, \frac B{2\lambda}(\log|T_l|-\lambda)\ (2\leq l\leq n),\, -B\right\}. Then HH is the continuous zero-boundary solution of (DH)n=δz(DH)^n=\delta_z, but ∥gi−H∥∞≥2λ−B>0.\|g_i-H\|_\infty\geq2\lambda-B>0. For n≥2n\geq2, z≠yz\ne y and g(z)=−λ≠−B=H(z)g(z)=-\lambda\ne-B=H(z). Thus weak convergence of the data on a fixed compact support does not imply uniform or capacity convergence of the solutions; in dimension at least two it does not even imply pointwise convergence to the solution of the limit datum.

Status: NEW_PROVED counterexample; the corresponding unrestricted weak-data stability assertion is FALSE.

Proof. Finite maxima of logarithms with a finite floor are continuous classical after every complete extension. At an ordinary boundary point, some ∣Tl∣=R|T_l|=R. If l=1l=1, monicity and integrality give ∣Pi(T1)∣=∣T1∣di|P_i(T_1)|=|T_1|^{d_i}; otherwise the displayed llth term is zero. All other terms are nonpositive on the closed polydisc. This proves the boundary and boundedness assertions.

We compute the atom over the original field. The separable irreducible reduction splits in the unramified extension of degree did_i. Its roots are units, and distinct roots have distance one. Near a conjugate root α\alpha, log⁡∣Pi(T1)∣=log⁡∣T1−α∣\log|P_i(T_1)|=\log|T_1-\alpha| on the corresponding open residue disc, since every other factor is a unit of norm one. The tie of all terms in (eq:r3-moving-atomic-potentials) occurs at radii rdi,r,…,rr^{d_i},r,\ldots,r. The slope simplex has volume 1/(din!)1/(d_i n!). Original torus integration contributes n!n!, and passage to the original degree-did_i point contributes the factor did_i. Its original mass is exactly one.

There is a useful independent check that no other mass was omitted. On PKn\mathbf P_K^n, homogenize the sections πdiPi(T1),πdiTldi (2≤l≤n),π2di.\pi^{d_i}P_i(T_1),\quad \pi^{d_i}T_l^{d_i}\ (2\leq l\leq n),\quad \pi^{2d_i}. They are base-point free sections of O(di)\mathcal O(d_i): on the affine chart the last is nonzero; at infinity their leading terms are the coordinate did_ith powers. Their pure-FS weight divided by did_i is the displayed gig_i in this frame. Its positive ordinary top measure has total mass deg⁡(O(1)n)=1\deg(\mathcal O(1)^n)=1, by the established ordinary-current model compatibility. The original atom already has mass one. This proves (Dgi)n=δzi(Dg_i)^n=\delta_{z_i} on all original tests, without assuming a pullback formula for an arbitrary delta form.

If ∣T1∣>1|T_1|>1, monicity gives gi=gg_i=g. If ∣T1∣≤1|T_1|\leq1, at most one of the distinct residue polynomials can have norm below one. For every other index the first logarithmic term is −λ-\lambda, again giving gi=gg_i=g. Thus the equality is eventual at every point, and always 0≤g−gi≤λ.0\leq g-g_i\leq\lambda. The slope simplex for gg at yy is the standard unit simplex; elsewhere the active slope differences have rank below nn. Original tropical calibration gives (Dg)n=δy(Dg)^n=\delta_y. For HH, the floor ties precisely at zz. The slopes are B/λB/\lambda in the first coordinate and B/(2λ)B/(2\lambda) in the others, so its top mass is Bn2n−1λn=1.\frac{B^n}{2^{n-1}\lambda^n}=1. The same rank calculation excludes further mass. Real slopes are justified by smooth convex regularization and the proved bounded continuity, so no rationality of B/λB/\lambda is needed. The ordinary boundary values are zero. Actual zero-boundary uniqueness identifies gig_i and HH as the respective solutions. At ziz_i, one has gi=−2λg_i=-2\lambda and H=−BH=-B, proving the uniform lower bound.

Finally set K∗={z,z1,z2,…}K_*=\{z,z_1,z_2,\ldots\} and Ω0={∣Tl∣<r−1/2 for all l}⋐Ω\Omega_0=\{|T_l|<r^{-1/2}\text{ for all }l\}\Subset\Omega. This is a fixed compact set in Ω0\Omega_0. The tester 1+gi/(2λ)1+g_i/(2\lambda), restricted to Ω0\Omega_0, is normalized and has mass (2λ)−n(2\lambda)^{-n} at ziz_i. Therefore any strict error set on K∗K_* containing ziz_i has outer capacity at least this constant on Ω0\Omega_0. Taking an error threshold less than 2λ−B2\lambda-B rules out local capacity convergence to HH. ◻

Corollary (Convex blocks do not preserve the original weak limit). For n≥2n\geq2 the example has tail convex blocks converging uniformly to gg, whose top measures converge to δy\delta_y, whereas the unaveraged top measures converge weakly to δz≠δy\delta_z\ne\delta_y.

Proof. The open residue branches Oi={∣Pi(T1)∣<1}∩ΩO_i=\{|P_i(T_1)|<1\}\cap\Omega are pairwise disjoint. The difference g−gig-g_i vanishes off OiO_i and is at most λ\lambda. Thus for any NN distinct indices, 0≤g−1N∑gi≤λN0\leq g-\frac1N\sum g_i\leq\frac{\lambda}{N} uniformly. Uniform BT continuity gives the measure limit δy\delta_y. The original measures have weak limit δz\delta_z. This explicitly excludes deducing a nonlinear unaveraged weak-limit representation solely from capacity compactness of convex blocks. ◻

Additional scope checks.

The weighted measures gi(Dgi)ng_i(Dg_i)^n tend to −2λδz-2\lambda\delta_z. For distinct i,ji,j, gj(zi)=−λg_j(z_i)=-\lambda, so the same fixed compact capacity testers rule out a capacity-Cauchy subsequence. The compact K∗K_* is not contained in finitely many fixed full Gauss boxes on which all these bounded potentials have a common interior Lipschitz estimate: such equicontinuity would contradict zi→zz_i\to z and gi(zi)−gi(z)=−λg_i(z_i)-g_i(z)=-\lambda. There is no contradiction with sharp weak-closure capacity domination, since δz\delta_z is itself the measure of HH. Nor is there a common locally bounded Monge–Ampere subsolution dominating all δzi\delta_{z_i}: its locally finite measure would have mass at least one at infinitely many points of K∗K_*.

A bounded-subsolution theorem on extendible domains

The final Dirichlet theorem in this section has ambient background [1][1]. Its preliminary uniform weighted estimate holds over every fixed allowed positive closed background. The topology of the data is stronger than ordinary weak convergence, which is insufficient even for the explicit moving-atom examples.

We keep the original finite field extension K/QpK/\mathbf Q_p, its absolute value, and all original CLD compact smooth tests. Positive backgrounds are symmetric, original-positive, and separately d′d'- and d′′d''-closed. Actual bounded classical psh functions are used throughout. Products and energies mean the already established fixed-background BT products, not wedges of arbitrary currents.

Uniformity over bounded psh weights

For finite signed Radon measures carried by a fixed compact subset of an ordinary domain OO, define ∥σ∥psh,O=sup⁡a∈Cb(O)0≤a≤1∣∫a dσ∣,dpsh,O(μ,ν)=∥μ−ν∥psh,O.\|\sigma\|_{\mathrm{psh},O} =\sup_{\substack{a\in\mathcal C_b(O)\\0\leq a\leq1}} \left|\int a\,d\sigma\right|, \qquad d_{\mathrm{psh},O}(\mu,\nu)=\|\mu-\nu\|_{\mathrm{psh},O}. The bound 0≤a≤10\leq a\leq1 is on the same full domain OO.

Theorem (Uniform psh-weight convergence for decreasing rows). Let SS be a fixed positive background of bidegree (p,p)(p,p) on an allowed ordinary domain OO, and put q=n−p≥1q=n-p\geq1. Suppose uj∈Cb(O)∩C0(O)u_j\in\mathcal C_b(O)\cap C^0(O) decrease to u∈Cb(O)u\in\mathcal C_b(O). If ff is original smooth with compact support in OO and 0≤f≤10\leq f\leq1, then sup⁡a∈Cb(O)0≤a≤1∣∫af(TS,q(ujq)−TS,q(uq))∣⟶0.\sup_{\substack{a\in\mathcal C_b(O)\\0\leq a\leq1}} \left|\int af\bigl(T_{S,q}(u_j^q)-T_{S,q}(u^q)\bigr)\right| \longrightarrow0. Here ujqu_j^q denotes qq repeated slots, not a scalar power. All constants used in the proof depend only on fixed compact stages, local bounds, and finitely many positive evaluations of SS.

Status: NEW_PROVED.

Proof. Split ff by a finite original smooth nonnegative partition subordinate to smaller norm boxes covering its support. It suffices to prove the assertion for one summand, still denoted ff, supported strictly inside one such box. Choose an original nonnegative compact cutoff χ\chi equal to one near its support. All potentials are uniformly bounded on a fixed larger compact stage. For j≤kj\leq k put gjk=uj−uk,Qjk=∑r=0q−1(Duj)r(Duk)q−1−rS,μj=(Duj)qS.g_{jk}=u_j-u_k,\quad Q_{jk}=\sum_{r=0}^{q-1}(Du_j)^r(Du_k)^{q-1-r}S,\quad \mu_j=(Du_j)^qS. Every summand of QjkQ_{jk} is a positive separately closed current of bidegree (n−1,n−1)(n-1,n-1), and the proved mixed calculus gives D(gjkQjk)=μj−μk.D(g_{jk}Q_{jk})=\mu_j-\mu_k. Let ejk=∫χ ΓQjk(gjk)≥0.e_{jk}=\int\chi\,\Gamma_{Q_{jk}}(g_{jk})\geq0. For an original smooth psh tester aa with 0≤a≤20\leq a\leq2 on the fixed stage, compact Green and the product rule give ∫af(μj−μk)=−∫f ΓQjk(a,gjk)−∫a ΓQjk(f,gjk).\int af(\mu_j-\mu_k) =-\int f\,\Gamma_{Q_{jk}}(a,g_{jk}) -\int a\,\Gamma_{Q_{jk}}(f,g_{jk}). The pairing involving the original smooth ff has its usual smooth first factor; no differentiated arbitrary Borel function is used. Cauchy–Schwarz gives

∣∫af(μj−μk)∣≤Aejk1/2,\left|\int af(\mu_j-\mu_k)\right|\leq A e_{jk}^{1/2},

with AA independent of a,j,ka,j,k. Indeed the required factors other than ejke_{jk} are bounded by ∫χ ΓQjk(a),∫χ ΓQjk(f).\int\chi\,\Gamma_{Q_{jk}}(a),\qquad \int\chi\,\Gamma_{Q_{jk}}(f). For the first, the cutoff identity is ∫χ ΓQjk(a)=12⟨Qjk,a2Dχ⟩−∫χaD(aQjk)≤12∣⟨Qjk,a2Dχ⟩∣.\int\chi\,\Gamma_{Q_{jk}}(a) =\tfrac12\langle Q_{jk},a^2D\chi\rangle -\int\chi aD(aQ_{jk}) \leq\tfrac12\left|\langle Q_{jk},a^2D\chi\rangle\right|. The subtracted measure is nonnegative. The finite-test CLN estimate bounds the last expression uniformly, using a2≤4a^2\leq4 and the fixed local bounds for the slots of QjkQ_{jk}. The second factor is bounded by the same CLN theorem against the fixed original form χ d′f∧d′′f\chi\,d'f\wedge d''f. The product rule and Cauchy–Schwarz can first be applied to smooth approximants of uj,uku_j,u_k and then passed to their limits by functional weighted monotone convergence. Every background occurring in this passage is an explicit mixed product over the same SS.

For an actual a∈Cb(O)a\in\mathcal C_b(O) with 0≤a≤10\leq a\leq1, take a smooth decreasing majorizing row on the prescribed box. After a prefix it is between zero and two on the compact stage. For fixed j,kj,k, dominated convergence against the fixed finite measures fμj,fμkf\mu_j,f\mu_k passes (eq:r3-uniform-weight-energy) to aa. The bound AA has not changed. This avoids introducing a first-order product with a simultaneously varying discontinuous tester.

The energy has the exact scalar expression

ejk=12⟨Qjk,gjk2Dχ⟩−∫χgjk(μj−μk).e_{jk} =\tfrac12\langle Q_{jk},g_{jk}^2D\chi\rangle -\int\chi g_{jk}(\mu_j-\mu_k).

Fix jj and let k→∞k\to\infty. Functional weighted joint monotone convergence applies to every term after expanding the scalar polynomials in uj,uku_j,u_k. It gives a finite nonnegative limit eje_j. Letting j→∞j\to\infty in that resulting expression gives ej→0e_j\to0, because its scalar difference is now uj−u→0u_j-u\to0 and the same functional weighted theorem applies to the mixed tuples (uj,u)(u_j,u). The constant row uu is an allowed decreasing row.

For each fixed bounded psh tester aa, weighted monotone continuity gives aμk→aμa\mu_k\to a\mu, where μ=(Du)qS\mu=(Du)^qS, against the test ff. Pass k→∞k\to\infty in (eq:r3-uniform-weight-energy); then ∣∫af(μj−μ)∣≤Aej1/2for every normalized a.\left|\int af(\mu_j-\mu)\right|\leq A e_j^{1/2} \quad\text{for every normalized }a. Take the supremum only after this fixed-tester limit. The right side tends to zero. Summing the finitely many initial localized estimates proves the assertion on OO. No capacity convergence was used to prove the uniform weighted estimate. ◻

Lemma (The psh dual distance detects weak convergence). Let VV be relatively compact in a smooth affine algebraic analytification and fix a compact L⊂VL\subset V. On finite positive Radon measures carried by LL, dpsh,Vd_{\mathrm{psh},V} is a metric. Convergence in this metric implies ordinary weak convergence of these measures.

Proof. Constants are testers, so convergence controls total masses. The differences of bounded continuous classical psh functions on VV form a real vector lattice: for a,b,c,da,b,c,d in that cone, max⁡(a−b,c−d)=max⁡(a+d,c+b)−(b+d).\max(a-b,c-d)=\max(a+d,c+b)-(b+d). This lattice contains constants and separates the points of LL. For the separation assertion, an affine polynomial PP whose absolute values differ at two Berkovich points yields, for a suitable finite cc, the bounded continuous classical function max⁡(log⁡∣P∣,c)\max(\log|P|,c), still separating them. It is bounded above on V‾\overline V. Affine polynomial seminorms separate the points of the affine analytification. The real lattice Stone–Weierstrass theorem therefore gives uniform density of this lattice in C(L)C(L). Every bounded psh function in its definition can be shifted and positively rescaled into [0,1][0,1] on VV. The metric consequently controls each fixed lattice test. Uniform approximation and the bounded masses give convergence against every continuous test on LL. The same argument with distance zero proves equality of the measures. Symmetry and the triangle inequality are immediate. ◻

The domain and the common exterior

For the remainder of the section, YY is a smooth geometrically integral affine KK-variety and ρ\rho is continuous classical psh on all of YanY^{\mathrm{an}}. Let VV be a connected ordinary component of {ρ<0}\{\rho<0\} with compact closure in YanY^{\mathrm{an}}, nonempty ordinary boundary, and ρ=0\rho=0 on that boundary. This is the extendible exhaustion-domain hypothesis. No strict psh exhaustion is assumed. Write MA⁡(v)=(Dv)n\operatorname{MA}(v)=(Dv)^n on VV.

Lemma (Weak boundary order for a subsolution). Let a,b∈Cb(V)a,b\in\mathcal C_b(V) be bounded. Suppose bb has ordinary boundary limit zero, lim sup⁡x→∂Va(x)≤0\limsup_{x\to\partial V}a(x)\leq0, and MA⁡(a)≥MA⁡(b)\operatorname{MA}(a)\geq\operatorname{MA}(b). Then a≤ba\leq b at every actual point of VV.

Proof. The boundary maximum principle gives a,b≤0a,b\leq0. Fix t,δ>0t,\delta>0 and put w=(1+t)a−δw=(1+t)a-\delta. The strict gap set E={b<w}E=\{b<w\} is relatively compact: the boundary limsup of w−bw-b is at most −δ-\delta, and the boundary is compact. It need not be ordinary open. The proved comparison theorem gives (1+t)nMA⁡(b)(E)≤MA⁡(w)(E)≤MA⁡(b)(E).(1+t)^n\operatorname{MA}(b)(E) \leq\operatorname{MA}(w)(E) \leq\operatorname{MA}(b)(E). These masses are finite, so MA⁡(b)(E)=0\operatorname{MA}(b)(E)=0. Pointwise ambient domination with the strict boundary gap yields w≤bw\leq b. Let δ↓0\delta\downarrow0 and then t↓0t\downarrow0. ◻

Lemma (A common lower barrier and exterior). Fix a compact L⋐VL\Subset V and B>0B>0. Every bounded classical zero-boundary solution vv with −B≤v≤0,MA⁡(v) carried by L-B\leq v\leq0,\qquad\operatorname{MA}(v)\text{ carried by }L satisfies v≥Cρv\geq C\rho on VV, where CC depends only on B,L,ρB,L,\rho. It has a classical extension v~\widetilde v to YanY^{\mathrm{an}}, equal to vv on VV and to CρC\rho on Yan∖VY^{\mathrm{an}}\setminus V. For the whole family one can make a common prescribed-box collar extension WvW_v such that, on a fixed neighborhood of V‾\overline V, Wv=v~+cW_v=\widetilde v+c for one constant cc, and all WvW_v equal one fixed smooth classical collar outside one larger compact set. In particular Wv−Wv′W_v-W_{v'} vanishes at every point outside VV, including every point of ∂V\partial V.

Proof. If LL is empty, domination gives v=0v=0 and the assertion is immediate. Otherwise choose CC with Cmin⁡L(−ρ)≥BC\min_L(-\rho)\geq B. The function b=max⁡(−B,Cρ)b=\max(-B,C\rho) has zero boundary value and satisfies b≤vb\leq v on LL. The measure MA⁡(v)\operatorname{MA}(v) therefore vanishes on {v<b}\{v<b\}. Pointwise domination gives b≤vb\leq v, and in particular Cρ≤vC\rho\leq v.

For ε>0\varepsilon>0, use max⁡(v,Cρ+ε)\max(v,C\rho+\varepsilon) inside VV and Cρ+εC\rho+\varepsilon outside it. Near every boundary point the second branch is active, by the ordinary boundary limit of vv. The ordinary gluing theorem makes this a global classical function. Let ε↓0\varepsilon\downarrow0. Decreasing closure gives precisely the claimed actual extension, including its value zero on the boundary. The family has uniform bounds on every fixed compact neighborhood of V‾\overline V: inside VV use BB, and outside it all functions equal CρC\rho.

Apply the constructive common-collar part of the prescribed-box regularization lemma (the B extension step) to this one uniformly bounded family. Choose the inner compact stage to contain V‾\overline V in its interior and the outer barrier above the common bound on the outer collar. The same shift and maximum construction preserve every v~+c\widetilde v+c near V‾\overline V and give the same smooth outer collar. Outside VV all inputs were already identical, so the same pointwise construction keeps them identical there. This proves the final assertion. The algebraic compactification used for this collar does not change the field or the original current test category. ◻

Theorem (Compact-support stability in the psh dual metric). On the specified extendible VV, let vjv_j be bounded actual classical zero-boundary solutions, with one common bound −B≤vj≤0-B\leq v_j\leq0. Suppose all μj=MA⁡(vj)\mu_j=\operatorname{MA}(v_j) are carried by one compact L⋐VL\Subset V and form a Cauchy sequence for dpsh,Vd_{\mathrm{psh},V}. Then there is a unique bounded actual classical zero-boundary vv such that vj⟶vin all-domain local outer ambient capacity,dpsh,V(μj,MA⁡(v))⟶0.v_j\longrightarrow v\quad\text{in all-domain local outer ambient capacity},\qquad d_{\mathrm{psh},V}(\mu_j,\operatorname{MA}(v))\longrightarrow0. Its measure is carried by LL. In particular, if dpsh,V(μj,μ)→0d_{\mathrm{psh},V}(\mu_j,\mu)\to0 for a finite positive measure μ\mu carried by LL, then MA⁡(v)=μ\operatorname{MA}(v)=\mu.

Status: NEW_PROVED.

Proof. The case B=0B=0 is immediate. Use the common exterior and collar from Lemma “A common lower barrier and exterior”. All Wj=WvjW_j=W_{v_j} are uniformly bounded on one fixed collar domain Ω\Omega. Their differences vanish outside VV, including its boundary. Hence the positive mixed energy has the exact expression I[1](Wj,Wk)=∫V(vj−vk)(μk−μj)≤2B dpsh,V(μj,μk).\begin{split} I_{[1]}(W_j,W_k) &=\int_V(v_j-v_k)(\mu_k-\mu_j)\\ &\leq2B\,d_{\mathrm{psh},V}(\mu_j,\mu_k). \end{split} For the bound, write vj−vk=B(aj−ak)v_j-v_k=B(a_j-a_k) with aj=1+vj/Ba_j=1+v_j/B, so both aj,aka_j,a_k are normalized psh testers on VV. No equality of the extension measures on ∂V\partial V is asserted: their actual multiplier Wj−WkW_j-W_k is zero there, which is the reason no boundary mass occurs in the identity.

The all-degree estimate of Theorem “An all-degree mixed-energy capacity estimate” makes WjW_j Cauchy in outer ambient capacity on Ω\Omega. Domain comparison gives all-domain local convergence and the completed calculus supplies a limit in A[1],b\mathcal A_{[1],b}. The proved ambient representative theorem supplies its unique actual classical representative. More explicitly, take a summably capacity-convergent subsequence and use its decreasing usc-regularized tail suprema. Near V‾\overline V the continuous barriers Cρ+c≤Wj≤max⁡(0,Cρ)+cC\rho+c\leq W_j\leq\max(0,C\rho)+c pass through this construction at actual points. Subtracting cc on VV gives a bounded classical representative vv satisfying Cρ≤v≤0C\rho\leq v\leq0. Thus it has ordinary boundary limit zero. The Cauchy property identifies the same limit for the full sequence.

The completed weighted calculus gives weak convergence of MA⁡(vj)\operatorname{MA}(v_j) to MA⁡(v)\operatorname{MA}(v) on every original compact test in VV. These measures are supported in LL, so the limit is carried by LL and is finite. For every fixed normalized actual psh tester aa on VV, insert an original smooth cutoff equal to one near LL. Weighted capacity continuity then gives ∫a dμk⟶∫a dMA⁡(v).\int a\,d\mu_k\longrightarrow \int a\,d\operatorname{MA}(v). For fixed jj, pass k→∞k\to\infty in the defining dual inequality, and only afterwards take the supremum in aa. This yields dpsh,V(μj,MA⁡(v))≤lim sup⁡k→∞dpsh,V(μj,μk)⟶0.d_{\mathrm{psh},V}(\mu_j,\operatorname{MA}(v)) \leq\limsup_{k\to\infty}d_{\mathrm{psh},V}(\mu_j,\mu_k) \longrightarrow0. This proves strong convergence without assuming in advance that the measure metric is complete. It identifies any prescribed metric limit by Lemma “The psh dual distance detects weak convergence”. Uniqueness of the actual zero-boundary solution is the proved ambient comparison and domination theorem. ◻

Strong atomic approximation from a subsolution

Theorem (The bounded-subsolution criterion). On the specified extendible domain VV, let μ\mu be a finite positive Radon measure with compact support in VV. The following are equivalent:

  1. There is a bounded actual classical uu on VV with ordinary boundary limit zero and (Du)n=μ(Du)^n=\mu.

  2. There is a bounded actual classical ww on VV with (Dw)n≥μ(Dw)^n\geq\mu.

The solution in the first assertion is unique. No boundary condition is required of the subsolution in the second assertion. This theorem asserts a bounded classical solution; it does not assert interior continuity from the bounded-subsolution hypothesis alone.

Status: NEW_PROVED.

Proof. Necessity is immediate. For sufficiency the zero measure has solution zero, so suppose μ≠0\mu\ne0, and let K=supp⁡μK=\operatorname{supp}\mu. Subtract a constant from ww so that −M≤w≤0-M\leq w\leq0. Choose CC so large that Cρ<−M−1C\rho<-M-1 on an ordinary neighborhood of KK. Then b=max⁡(w,Cρ)b=\max(w,C\rho) is bounded classical, equals ww on that neighborhood, and has ordinary boundary value zero. Ordinary locality and positivity give (Db)n≥μ(Db)^n\geq\mu on all of VV: on the neighborhood of KK this follows from the subsolution, and outside KK the datum is zero. This uses no presumed ordinary openness of a comparison set involving ww; the displayed strict constant bound supplies the required open neighborhood.

The same epsilon-maximum construction as in the preceding lemma extends bb to a global classical function b~\widetilde b, equal to CρC\rho off VV. We now use the accepted constructive input from B, with the entire compact V‾\overline V protected. Extend the local potential to one semipositive metric on a projective algebraic model in a fixed frame, and take its decreasing pure Fubini–Study rows. In that frame, after the common constant or unit-log correction, their local potentials bjb_j are continuous classical on a fixed neighborhood of V‾\overline V and satisfy bj↓b~.b_j\downarrow\widetilde b. Their ordinary top-degree measures are finite positive sums of Dirac masses at divisorial points of the original KK-analytic space. This last fact is the accepted semipositive PL/model measure compatibility applied to these particular algebraic pure-Fubini–Study rows. It is not a claim that every locally defined PL function has globally finite atomic curvature.

On the compact boundary ∂V\partial V, the continuous functions bjb_j decrease to the continuous zero function. Dini’s theorem on that compact set gives εj=max⁡∂Vbj↓0.\varepsilon_j=\max_{\partial V}b_j\downarrow0. The maximum principle gives bj≤εjb_j\leq\varepsilon_j on VV. Also bj≥bb_j\geq b there. Only the boundary restriction is used in this Dini argument; continuity of bb in the interior is not assumed.

Put ν=(Db)n\nu=(Db)^n. Since μ≤ν\mu\leq\nu, Radon–Nikodym gives μ=fν\mu=f\nu for a Borel 0≤f≤10\leq f\leq1, chosen zero outside KK. Choose a compact L⋐VL\Subset V with K⊂L∘K\subset L^{\circ}. There exist original smooth flf_l with common compact support in LL, 0≤fl≤10\leq f_l\leq1, and ∫∣fl−f∣ dν⟶0.\int|f_l-f|\,d\nu\longrightarrow0. Here only the finite measure on a compact collar is involved. Radon regularity first approximates a bounded Borel function by continuous compactly supported functions; original smooth partitions and smooth approximation on finitely many moment presentations, followed by a smooth truncation into [0,1][0,1] and a fixed support cutoff, give the asserted original smooth functions. Thus neither a changed test sheaf nor an infinite-volume density assertion is needed.

For each fixed ll, Theorem “Uniform psh-weight convergence for decreasing rows”, with S=[1]S=[1], gives dpsh,V(fl(Dbj)n,flν)⟶0(j→∞).d_{\mathrm{psh},V}\bigl(f_l(Db_j)^n,f_l\nu\bigr) \longrightarrow0\quad(j\to\infty). Choose strictly increasing j(l)j(l) so that this distance is at most 2−l2^{-l}, and set μl=fl(Dbj(l))n.\mu_l=f_l(Db_{j(l)})^n. These are finite positive divisorial atomic measures carried by the same compact LL, and dpsh,V(μl,μ)≤2−l+∫∣fl−f∣ dν⟶0.d_{\mathrm{psh},V}(\mu_l,\mu) \leq2^{-l}+\int|f_l-f|\,d\nu\longrightarrow0. This is the needed strong approximation, not merely weak approximation by atomic measures. Also μl≤(Dbj(l))n\mu_l\leq(Db_{j(l)})^n on all of VV.

The finite-divisorial-atom Dirichlet theorem on this same exhaustion domain gives bounded actual zero-boundary classical solutions vlv_l with MA⁡(vl)=μl\operatorname{MA}(v_l)=\mu_l. Apply Lemma “Weak boundary order for a subsolution” with a=bj(l)−εj(l)a=b_{j(l)}-\varepsilon_{j(l)} and b=vlb=v_l. Its hypotheses hold by the preceding measure domination and boundary maximum. Hence 0≥vl≥bj(l)−εj(l)≥b−ε1.0\geq v_l\geq b_{j(l)}-\varepsilon_{j(l)} \geq b-\varepsilon_1. This is one common global bound, independent of the atoms and their masses. Theorem “Compact-support stability in the psh dual metric” now gives a bounded actual classical zero-boundary uu with MA⁡(u)=μ\operatorname{MA}(u)=\mu. It also supplies capacity convergence of the constructed solutions. Actual pointwise uniqueness follows from the ambient domination theorem. ◻

Scope of stability and of the algebraic input.

The compact support, common uniform bound, and extendible exhaustion are used explicitly in the common-exterior energy argument. The construction retains the original field and invokes B only for its accepted algebraic collar and pure-Fubini–Study approximation, with the already known original-current/model-measure comparison. It does not derive arbitrary-background Dirichlet existence, ordinary weak-data stability, or continuity under merely bounded subsolutions. The preliminary uniform weighted theorem, in contrast, is valid for arbitrary fixed SS and all its permitted remaining degrees.

For the moving-atom measures δzj⇀δz\delta_{z_j}\rightharpoonup\delta_z of the explicit counterexample, the normalized tester 1+gj/(2λ)1+g_j/(2\lambda) has values zero at zjz_j and 1/21/2 at zz. Consequently their psh dual distances are at least 1/21/2. The strong topology used here therefore excludes that actual weak convergence defect.

The fixed-domain Dirichlet problem under power capacity growth

Throughout this section K/QpK/\mathbf Q_p is the prescribed finite extension with its original absolute value. Let Y/KY/K be smooth, geometrically integral and affine, of dimension n≥1n\geq1. Let ρ\rho be a continuous classical psh function on all of YanY^{\mathrm{an}}. Fix a connected component VV of {ρ<0}\{\rho<0\} such that V⋐Yan,∅≠∂V⊂{ρ=0}.V\Subset Y^{\mathrm{an}},\qquad \varnothing\ne\partial V\subset\{\rho=0\}. All boundaries are ordinary boundaries. Write C(E)=Cap⁡[1](E;V)C(E)=\operatorname{Cap}_{[1]}(E;V). The background in this Dirichlet section is [1][1], not an arbitrary relative current. The currents and their convergence use all original CLD tests; the classical functions retain actual values and all extension tests.

Lemma (An affine projective collar for a fixed lower obstacle). Fix B>0B>0 and put FB={u∈Cb(V):Bρ≤u≤0}.\mathcal F_B=\{u\in\mathcal C_b(V):B\rho\leq u\leq0\}. There is a smooth projective compactification ZZ of YY, an ample line LL on ZZ, a fixed frame on an ordinary neighborhood of V‾\overline V, and an affine order-preserving map u↦Uu\mapsto U from FB\mathcal F_B to bounded classical metrics on LL with the following properties. In that frame the weight of UU is u+cu+c on VV, with one constant cc independent of uu. All the metrics are equal on Zan∖VZ^{\mathrm{an}}\setminus V. They have common bounds relative to a continuous reference metric, and the metrics FF and LBL_B associated with 00 and BρB\rho are continuous. One has LB≤U≤FL_B\leq U\leq F.

Conversely, any bounded classical metric PP with LB≤P≤FL_B\leq P\leq F and P=FP=F outside VV corresponds in this frame to a member of FB\mathcal F_B. The affine family has energy E(U)=1n+1∑j=0n∫Zan(U−F)c1(L,U)jc1(L,F)n−j.\mathcal E(U)=\frac1{n+1}\sum_{j=0}^n \int_{Z^{\mathrm{an}}}(U-F)c_1(L,U)^j c_1(L,F)^{n-j}. It satisfies the mixed energy identities, midpoint inequality and local energy-to-capacity estimates already proved.

Proof. For u∈FBu\in\mathcal F_B and ϵ>0\epsilon>0, glue max⁡(u,Bρ+ϵ)\max(u,B\rho+\epsilon) on VV to Bρ+ϵB\rho+\epsilon outside VV. Near every ordinary boundary point the latter function is positive while u≤0u\leq0, so these prescriptions agree on an ordinary neighborhood. The glued functions are classical and decrease locally boundedly as ϵ↓0\epsilon\downarrow0 to u~=u on V,u~=Bρ off V.\widetilde u=u\text{ on }V,\qquad \widetilde u=B\rho\text{ off }V. The lower bound guarantees the actual first equality. This gives a classical function on YanY^{\mathrm{an}}. The map u↦u~u\mapsto\widetilde u is affine and order preserving. Its upper member is the continuous function equal to zero on VV and to BρB\rho off VV; its lower member is the continuous function BρB\rho.

Use the framed smooth compactification Lemma “A framed smooth compactification of an affine variety”, with one frame whose nonvanishing locus contains all of YY. Choose nested compact norm boxes in that frame, the innermost containing V‾\overline V in its ordinary interior. On the outer box the family u~\widetilde u has one common bound. The constructive collar of B, Lemma D.2, in the form used in Theorem “Local bounded upper-envelope theorem”, gives on one ample power the weights U=max⁡(u~+c,mb)U=\max(\widetilde u+c,mb) on the outer box, and a fixed metric outside the middle box. Here b=0b=0 on the inner box, cc makes every u~+c\widetilde u+c positive there, and the integer mm makes the fixed branch strictly dominate on the outer collar. Since the inputs all agree outside VV, the outputs all agree there. On VV the maximum is inactive and its weight is u+cu+c. Thus this particular family remains affine, even though the collar operation on arbitrary inputs is a maximum. The upper and lower members are continuous, and ordinary restriction proves the converse assertion.

Global mixed Green exchange gives the stated energy identities; differences of the metrics vanish outside the fixed compact V‾\overline V. Equivalently one can compute in this frame on a larger norm box. When applying the smooth-collar formulation of Theorem “An all-degree mixed-energy capacity estimate”, use its prescribed-box auxiliary collar reduction: preserve the weights near V‾\overline V, up to a common constant, and replace only their already common exterior by a protected smooth collar. Their difference and all energy integrands supported by that difference are unchanged. The constants are uniform for the present bounded family. This does not assert that the original BρB\rho is smooth. ◻

Lemma (A constrained maximizer). Let H⋐VH\Subset V be compact, and let μ\mu be a finite positive Radon measure carried by HH, with μ(E)≤AC(E)1+α(E⊂H Borel),A,α>0.\mu(E)\leq A C(E)^{1+\alpha}\quad (E\subset H\text{ Borel}),\qquad A,\alpha>0. For every B>0B>0, the functional F(U)=E(U)−∫Hu dμ\mathcal F(U)=\mathcal E(U)-\int_H u\,d\mu on the preceding affine family attains its supremum.

Proof. The common bounds, the fixed projective intersection mass and finiteness of μ\mu bound the functional above and below. Choose a maximizing sequence UjU_j, with local weights uj+cu_j+c on VV. The family is convex. The integral term is affine, so its midpoint gap cancels, and Lemma “Quantitative midpoint concavity” gives 0≤cnI(Uj,Uk)≤sup⁡F−F(Uj)+F(Uk)2⟶0(j,k→∞).0\leq c_n I(U_j,U_k) \leq\sup\mathcal F- \frac{\mathcal F(U_j)+\mathcal F(U_k)}2 \longrightarrow0\quad(j,k\to\infty). The energy-to-capacity estimate on the fixed surrounding norm box therefore makes the weights capacity Cauchy on every smaller compact stage. The reverse framed domain comparison, Corollary “One fixed capacity domain suffices locally”, upgrades this to all-domain local capacity Cauchy before taking any completion. The ambient completion and unique classical representative, Corollary “The unique classical representative”, produce a bounded classical limit metric UU.

Its actual bounds LB≤U≤FL_B\leq U\leq F and its equality to FF outside VV follow, for example, from the decreasing tail-usc formula for that representative: the two bounding metrics are continuous, so both bounds pass through the tail regularization and its decreasing limit. Thus the limit is an admissible member of the same affine family. The completed weighted calculus on a finite cover of V‾\overline V gives E(Uj)→E(U)\mathcal E(U_j)\to\mathcal E(U); outside VV all weighted differences in this formula vanish.

Let M=μ(V)M=\mu(V) and let LL be a bound for ∣uj∣,∣u∣|u_j|,|u|. For δ>0\delta>0, set Ej=H∩{∣uj−u∣>δ}E_j=H\cap\{|u_j-u|>\delta\}. Local capacity convergence and a finite cover of HH give C∗(Ej)→0C^*(E_j)\to0 on the fixed domain VV. Hence ∫H∣uj−u∣ dμ≤δM+2LA C∗(Ej)1+α.\int_H|u_j-u|\,d\mu \leq\delta M+2LA\,C^*(E_j)^{1+\alpha}. First let j→∞j\to\infty, then δ↓0\delta\downarrow0. This proves continuity of the measure term, so the admissible limit attains the supremum. No weak convergence theorem for arbitrary varying Monge–Ampère measures was used. ◻

Lemma (The upper variation). If μ≠0\mu\ne0 and UU is the constrained maximizer, with local function uu, then u<0u<0 everywhere on VV and (Du)n≤μ.(Du)^n\leq\mu. In particular its Monge–Ampère measure is carried by HH.

Proof. The zero function is not a maximizer. Indeed the competitor tBρtB\rho, 0<t≤10<t\leq1, is admissible. The reference metric FF has constant weight on VV, and all differences are zero off VV. Consequently its energy is E(UtBρ)=tn+1n+1∫VBρ (BDρ)n.\mathcal E(U_{tB\rho})= \frac{t^{n+1}}{n+1}\int_V B\rho\,(BD\rho)^n. This is a finite constant times tn+1t^{n+1}. The change in the measure term is −t∫HBρ dμ>0-t\int_H B\rho\,d\mu>0, of first order in tt. Thus u≠0u\ne0. The classical strong maximum principle on the connected domain makes the nonpositive uu strictly negative.

Let ϕ≥0\phi\geq0 be an original compact smooth scalar test in VV, extended by zero to ZZ. Since uu is usc and negative, its maximum on supp⁡ϕ\operatorname{supp}\phi is strictly negative. For small t>0t>0, u+tϕ≤0u+t\phi\leq0 there. Let Pt=PLb(U+tϕ).P_t=P_L^{\mathrm b}(U+t\phi). The bounded-usc obstacle theorem supplies this actual greatest classical minorant, and U≤Pt≤U+tϕU\leq P_t\leq U+t\phi. It is equal to UU off the support of ϕ\phi, and is an admissible member of the same constrained family. Put Δt=Pt−U\Delta_t=P_t-U. Quasicontinuous-obstacle orthogonality, Theorem “Orthogonality for bounded quasicontinuous usc obstacles”, gives ∫Δtc1(L,Pt)n=t∫ϕc1(L,Pt)n.\int\Delta_t c_1(L,P_t)^n =t\int\phi c_1(L,P_t)^n. Energy concavity and 0≤Δt≤tϕ0\leq\Delta_t\leq t\phi yield ∫ϕc1(L,Pt)n≤E(Pt)−E(U)t≤∫ϕc1(L,U)n.\int\phi c_1(L,P_t)^n \leq\frac{\mathcal E(P_t)-\mathcal E(U)}t \leq\int\phi c_1(L,U)^n. Uniform convergence Pt→UP_t\to U and the proved bounded capacity continuity identify both limits with ∫Vϕ(Du)n\int_V\phi(Du)^n. Maximality of F\mathcal F gives E(Pt)−E(U)≤∫HΔt dμ≤t∫Hϕ dμ.\mathcal E(P_t)-\mathcal E(U) \leq\int_H\Delta_t\,d\mu \leq t\int_H\phi\,d\mu. Thus (Du)n≤μ(Du)^n\leq\mu on all nonnegative original scalar tests. The coefficient-Radon identification turns this into the claimed measure inequality. All perturbations have support inside VV; no boundary mass is discarded in these calculations. ◻

Lemma (An estimate independent of the lower obstacle). Let v∈Cb(V)v\in\mathcal C_b(V) be bounded, nonpositive, with zero ordinary boundary limit, and suppose (Dv)n≤μ(Dv)^n\leq\mu. For M=μ(V)>0M=\mu(V)>0 put S=(1+21−2−α)(AMα)1/[n(1+α)].S=\left(1+\frac2{1-2^{-\alpha}}\right) (AM^\alpha)^{1/[n(1+\alpha)]}. Then v≥−Sv\geq-S. If its measure is carried by HH, then with dH=min⁡H(−ρ)>0d_H=\min_H(-\rho)>0 one has v≥max⁡{−S,(S/dH)ρ}.v\geq\max\{-S,(S/d_H)\rho\}.

Proof. For s>0s>0 put f(s)=C({v<−s})1/nf(s)=C(\{v<-s\})^{1/n}. These sets have compact closure because of the ordinary boundary limit. Given a tester a∈Cb(V)a\in\mathcal C_b(V), 0≤a≤10\leq a\leq1, compare vv with w=−s−t+taw=-s-t+ta. One has {v<−s−t}⊂{v<w}⊂{v<−s}.\{v<-s-t\}\subset\{v<w\}\subset\{v<-s\}. The middle gap has compact closure. Comparison gives tn∫{v<−s−t}(Da)n≤∫{v<w}(Dw)n≤(Dv)n({v<w})≤μ({v<−s}).t^n\int_{\{v<-s-t\}}(Da)^n \leq\int_{\{v<w\}}(Dw)^n \leq(Dv)^n(\{v<w\})\leq\mu(\{v<-s\}). Taking the supremum and using the measure bound on H∩{v<−s}H\cap\{v<-s\} yields tf(s+t)≤A1/nf(s)1+α.t f(s+t)\leq A^{1/n} f(s)^{1+\alpha}. The same estimate without the capacity bound, followed by s↓0s\downarrow0 and increasing-set continuity of raw capacity, gives tf(t)≤M1/nt f(t)\leq M^{1/n}.

Set T=(AMα)1/[n(1+α)]T=(AM^\alpha)^{1/[n(1+\alpha)]} and start with s0=Ts_0=T. As long as f(sj)>0f(s_j)>0, take tj=2A1/nf(sj)αt_j=2A^{1/n}f(s_j)^\alpha and sj+1=sj+tjs_{j+1}=s_j+t_j. Then f(sj+1)≤f(sj)/2f(s_{j+1})\leq f(s_j)/2, and the initial estimate gives tj≤2T2−jαt_j\leq2T2^{-j\alpha}. Their sum is at most 2T/(1−2−α)2T/(1-2^{-\alpha}). If the iteration stops, all deeper sublevels already have zero capacity; otherwise f(S)=0f(S)=0 by monotonicity and the geometric decay. The strict sublevel {v<−S}\{v<-S\} is ordinary open because vv is usc. Ambient capacity nondegeneracy on the fixed framed domain makes it empty.

The bounded classical barrier b=max⁡(−S,(S/dH)ρ)b=\max(-S,(S/d_H)\rho) equals −S-S on HH and has zero boundary limit. Thus (Dv)n({v<b})=0(Dv)^n(\{v<b\})=0 when the measure is carried by HH. Pointwise domination, applied first with the positive margin b−δb-\delta and then δ↓0\delta\downarrow0, gives v≥bv\geq b. No strict gap between two discontinuous potentials has been called ordinary open. ◻

Theorem (Power-capacity Dirichlet existence). Let μ\mu be a finite positive Radon measure carried by a compact H⋐VH\Subset V, and assume the power bound in Lemma “A constrained maximizer”. There is a unique bounded u∈Cb(V)u\in\mathcal C_b(V) with zero ordinary boundary limit and (Du)n=μ.(Du)^n=\mu. For M>0M>0 it satisfies max⁡(−S,(S/dH)ρ)≤u≤0\max(-S,(S/d_H)\rho)\leq u\leq0, with SS as above. For M=0M=0 the solution is zero. All equalities hold on the full original test sheaf. Continuity on V‾\overline V is proved later by the one-sided Dini estimate.

Status: NEW_PROVED.

Proof. Assume M>0M>0 and choose B>S/dHB>S/d_H. The constrained maximizer exists, and its upper variation gives (Du)n≤μ(Du)^n\leq\mu. The preceding lemma gives u≥(S/dH)ρu\geq(S/d_H)\rho. Therefore u−Bρ≥(B−S/dH)(−ρ)>0at every point of V.u-B\rho\geq(B-S/d_H)(-\rho)>0 \quad\text{at every point of }V. This is a uniform positive margin on every compact perturbation support, and releases the artificial lower obstacle there.

Let ϕ≥0\phi\geq0 be original compact smooth in VV. For sufficiently small t>0t>0, the global lower metric satisfies LB≤U−tϕL_B\leq U-t\phi. Put Pt=PLb(U−tϕ)P_t=P_L^{\mathrm b}(U-t\phi). Since LBL_B is a competitor and U−t∥ϕ∥∞U-t\|\phi\|_\infty is also a global classical minorant, LB≤Pt≤U−tϕ≤U,U−t∥ϕ∥∞≤Pt.L_B\leq P_t\leq U-t\phi\leq U, \qquad U-t\|\phi\|_\infty\leq P_t. Outside VV the lower metric equals UU, so Pt=UP_t=U there. Thus PtP_t is admissible. Write Δt=U−Pt\Delta_t=U-P_t. The obstacle orthogonality identity is now ∫Δtc1(L,Pt)n=t∫ϕc1(L,Pt)n.\int\Delta_t c_1(L,P_t)^n =t\int\phi c_1(L,P_t)^n. Since Δt≥tϕ\Delta_t\geq t\phi, concavity gives ∫ϕc1(L,U)n≤E(U)−E(Pt)t≤∫ϕc1(L,Pt)n.\int\phi c_1(L,U)^n \leq\frac{\mathcal E(U)-\mathcal E(P_t)}t \leq\int\phi c_1(L,P_t)^n. Uniform convergence again shows that the middle expression tends to ∫Vϕ(Du)n\int_V\phi(Du)^n. Maximality gives the opposite measure inequality: E(U)−E(Pt)≥∫HΔt dμ≥t∫Hϕ dμ.\mathcal E(U)-\mathcal E(P_t) \geq\int_H\Delta_t\,d\mu \geq t\int_H\phi\,d\mu. Hence (Du)n≥μ(Du)^n\geq\mu, proving equality. The original sandwich Bρ≤u≤0B\rho\leq u\leq0 gives the ordinary zero boundary limit, and the a priori lemma gives the stated sharper bounds.

If M=0M=0, take u=0u=0. For uniqueness, if u,vu,v have the same measure and zero boundary values, compare vv with (1+ϵ)u−δ(1+\epsilon)u-\delta on their compact strict gap. The comparison inequality and the common measure force that gap to have zero measure; ambient pointwise domination makes it empty. Let δ↓0\delta\downarrow0 and then ϵ↓0\epsilon\downarrow0 and repeat in the other order. All cancellations occur on compact gaps of finite mass. This proves uniqueness without a weak-data stability assumption. ◻

Dini capacity moduli for the fixed-domain Dirichlet problem

Retain the extendible domain (Y,ρ,V)(Y,\rho,V) and the original ambient background [1][1] of Section “The fixed-domain Dirichlet problem under power capacity growth”. Write C(E)=Cap⁡[1](E;V)C(E)=\operatorname{Cap}_{[1]}(E;V). All functions retain their actual classical values over the prescribed field.

Definition (Dini capacity modulus). A finite nondecreasing function ω:[0,∞)→[0,∞)\omega:[0,\infty)\to[0,\infty) with ω(0)=0\omega(0)=0 is a Dini capacity modulus if

∫0aω(t) dtt<∞\int_0^a\omega(t)\,\frac{dt}{t}<\infty

for some a>0a>0. No continuity or strict positivity is imposed.

Monotonicity implies measurability. The condition forces ω(t)→0\omega(t)\to0 at zero, and then holds for every finite positive upper endpoint, since the integrand is bounded on compact intervals away from zero.

Theorem (Dirichlet solvability under a Dini modulus). Let H⋐VH\Subset V be compact, and let μ\mu be a finite positive Radon measure carried by HH. Suppose

μ(E)≤C(E)ω(C(E)1/n)n(E⊂H Borel)\mu(E)\leq C(E)\omega(C(E)^{1/n})^n \qquad(E\subset H\text{ Borel})

for a Dini capacity modulus. There is a unique bounded u∈Cb(V)u\in\mathcal C_b(V) with ordinary boundary limit zero and (Du)n=μ(Du)^n=\mu. For M=μ(V)>0M=\mu(V)>0, put

Sω(s)=s+2∑j≥0ω(2−jM1/n/s),S∗=inf⁡s>0Sω(s).S_\omega(s)=s+2\sum_{j\geq0}\omega(2^{-j}M^{1/n}/s), \qquad S_* =\inf_{s>0}S_\omega(s).

Each series is finite. With dH=min⁡H(−ρ)d_H=\min_H(-\rho), one has

max⁡{−S∗,(S∗/dH)ρ}≤u≤0.\max\{-S_*,(S_*/d_H)\rho\}\leq u\leq0.

For M=0M=0, u=0u=0. The infimum defining S∗S_* need not be attained.

Status: DIRECT_COROLLARY of the preceding variational construction, with the scalar iteration below. Continuity is established separately in Theorem “Continuity on the ordinary closure”.

Lemma (Dini sublevel iteration). If v∈Cb(V)v\in\mathcal C_b(V) is bounded, nonpositive, has zero boundary limit, and (Dv)n≤μ(Dv)^n\leq\mu, then ∥v∥∞≤Sω(s)\|v\|_\infty\leq S_\omega(s) for every s>0s>0. If its measure is carried by HH, it also satisfies v≥max⁡(−Sω(s),(Sω(s)/dH)ρ)v\geq\max(-S_\omega(s),(S_\omega(s)/d_H)\rho).

Proof. Set f(s)=C({v<−s})1/nf(s)=C(\{v<-s\})^{1/n}. The compact-gap comparison used in Lemma “An estimate independent of the lower obstacle”, with normalized testers on this same domain, gives

tf(s+t)≤f(s)ω(f(s)),tf(t)≤M1/n(s,t>0).t f(s+t)\leq f(s)\omega(f(s)),\qquad t f(t)\leq M^{1/n}\quad(s,t>0).

For the first inequality apply the measure bound on H∩{v<−s}H\cap\{v<-s\} and use monotonicity of c↦cω(c1/n)nc\mapsto c\omega(c^{1/n})^n. The second follows by increasing continuity of raw capacity as the shift ss decreases to zero.

Starting at s0>0s_0>0, while f(sj)>0f(s_j)>0, take tj=2ω(f(sj))t_j=2\omega(f(s_j)) and sj+1=sj+tjs_{j+1}=s_j+t_j. If ω(f(sj))=0\omega(f(s_j))=0, the first inequality makes f(sj+t)=0f(s_j+t)=0 for every t>0t>0. Increasing-set continuity then gives f(sj)=0f(s_j)=0, so the iteration has already stopped. Otherwise f(sj+1)≤12f(sj),tj≤2ω(2−jM1/n/s0).f(s_{j+1})\leq\tfrac12 f(s_j),\qquad t_j\leq2\omega(2^{-j}M^{1/n}/s_0). The sum is finite because

∑j≥0ω(2−ja)≤ω(a)+1log⁡2∫0aω(t) dtt.\sum_{j\geq0}\omega(2^{-j}a) \leq\omega(a)+\frac1{\log2}\int_0^a\omega(t)\,\frac{dt}{t}.

Indeed the integral over [2−ja,2−j+1a][2^{-j}a,2^{-j+1}a] is at least (log⁡2)ω(2−ja)(\log2)\omega(2^{-j}a) for j≥1j\geq1. Monotonicity of ff now gives f(Sω(s0))=0f(S_\omega(s_0))=0, whether the iteration terminates or continues indefinitely. The strict sublevel {v<−Sω(s0)}\{v<-S_\omega(s_0)\} is an ordinary open set, since vv is usc. Ambient capacity nondegeneracy makes it empty. The barrier b=max⁡(−Sω(s0),(Sω(s0)/dH)ρ)b=\max(-S_\omega(s_0),(S_\omega(s_0)/d_H)\rho) equals −Sω(s0)-S_\omega(s_0) on HH. Thus v≥bv\geq b on the carrying set of its measure, and compact-gap pointwise domination gives v≥bv\geq b throughout VV. ◻

Proof of the theorem. For M>0M>0, fix s0>0s_0>0 and choose B>Sω(s0)/dHB>S_\omega(s_0)/d_H. Use the class FB\mathcal F_B, its common affine collar and energy from Lemma “An affine projective collar for a fixed lower obstacle”. The power bound in the constrained-maximizer argument was used for continuity of integration against μ\mu under bounded capacity convergence on HH. Here, if ∣vj∣,∣v∣≤L|v_j|,|v|\leq L and cj=C(H∩{∣vj−v∣>δ})c_j=C(H\cap\{|v_j-v|>\delta\}), then ∫H∣vj−v∣ dμ≤δM+2Lcjω(cj1/n)n.\int_H|v_j-v|\,d\mu \leq\delta M+2L c_j\omega(c_j^{1/n})^n. The last term tends to zero when cj→0c_j\to0, with value zero at cj=0c_j=0. The same midpoint argument and ambient representative therefore give a maximizer of the energy minus ∫v dμ\int v\,d\mu. The test tBρtB\rho again proves that it is not the zero function, and the classical strong maximum principle makes it strictly negative.

The positive obstacle variation of Lemma “The upper variation” yields (DuB)n≤μ(Du_B)^n\leq\mu. The preceding lemma makes uB≥(Sω(s0)/dH)ρu_B\geq(S_\omega(s_0)/d_H)\rho. The strict choice of BB leaves a positive lower-obstacle margin on each compact perturbation support. The negative obstacle variation in Theorem “Power-capacity Dirichlet existence” therefore applies without any change of domain or envelope input and gives (DuB)n≥μ(Du_B)^n\geq\mu. These variations use the already proved ambient quasicontinuous-obstacle orthogonality, not an arbitrary-relative orthogonality assertion. Ambient pointwise comparison gives uniqueness.

Apply the a priori lemma to this one solution for every s>0s>0. Infimizing gives u≥−S∗u\geq-S_*. Along any sequence of parameters with Sω(s)→S∗S_\omega(s)\to S_*, the barrier inequalities give the other bound in (eq:r3-dini-solution-bound). No minimum is needed. ◻

Remark. For ω(t)=1[1,∞)(t)\omega(t)=\mathbf1_{[1,\infty)}(t) and M=1M=1, Sω(s)=sS_\omega(s)=s when s>1s>1, whereas Sω(s)≥s+2S_\omega(s)\geq s+2 when 0<s≤10<s\leq1. Thus its infimum is 11 and is not attained.

Corollary (Power specialization). For ω(t)=A1/ntα\omega(t)=A^{1/n}t^\alpha, A,α>0A,\alpha>0, the measure condition is μ(E)≤AC(E)1+α\mu(E)\leq A C(E)^{1+\alpha}. For M>0M>0, ∥u∥∞≤(1+α)α−α/(1+α)(21−2−α)1/(1+α)(AMα)1/[n(1+α)].\|u\|_\infty\leq (1+\alpha)\alpha^{-\alpha/(1+\alpha)} \left(\frac2{1-2^{-\alpha}}\right)^{1/(1+\alpha)} (AM^\alpha)^{1/[n(1+\alpha)]}.

Proof. Sum the geometric series to obtain Sω(s)=s+Qs−αS_\omega(s)=s+Qs^{-\alpha}, where Q=2A1/nMα/n/(1−2−α)Q=2A^{1/n}M^{\alpha/n}/(1-2^{-\alpha}), and minimize at s=(αQ)1/(1+α)s=(\alpha Q)^{1/(1+\alpha)}. ◻

Corollary (Logarithmic modulus). For L>0L>0 and β>1\beta>1, the sufficient condition is

μ(E)≤LnC(E)[log⁡(e+C(E)−1/n)]nβ.\mu(E)\leq \frac{L^n C(E)}{[\log(e+C(E)^{-1/n})]^{n\beta}}.

For M>0M>0, s≥eM1/ns\geq eM^{1/n} and ℓ=log⁡(s/M1/n)\ell=\log(s/M^{1/n}),

∥u∥∞≤s+2L(ℓ−β+ℓ1−β(β−1)log⁡2).\|u\|_\infty\leq s+2L\left(\ell^{-\beta} +\frac{\ell^{1-\beta}}{(\beta-1)\log2}\right).

Proof. Use ω(t)=L[log⁡(e+1/t)]−β\omega(t)=L[\log(e+1/t)]^{-\beta} for t>0t>0 and ω(0)=0\omega(0)=0. Its Dini integral converges precisely for β>1\beta>1. For a≤e−1a\leq e^{-1}, bound ω(a)\omega(a) by L(log⁡(1/a))−βL(\log(1/a))^{-\beta} and the Dini integral by L(log⁡(1/a))1−β/(β−1)L(\log(1/a))^{1-\beta}/(\beta-1). Insert these bounds into (eq:r3-dini-dyadic-bound) with a=M1/n/sa=M^{1/n}/s. ◻

It is enough to prescribe the modulus and the measure condition at capacities with C(E)1/n<aC(E)^{1/n}<a. Extend it for t≥at\geq a by max⁡{ω(a),M1/n/a}\max\{\omega(a),M^{1/n}/a\}. The remaining sets have mass at most MM, so the extended condition holds. This does not assert that every individual datum under a non-Dini bound is unsolvable; the separate radial construction concerns a uniform guarantee.

One-sided capacity stability and continuity of Dini solutions

Here the background is [1][1]. All functions are actual bounded classical psh functions over the fixed original field, and all currents use the original CLD tests. On a relatively compact ordinary open VV in a permitted smooth affine neighborhood, write C(E)=Cap⁡[1](E;V)C(E)=\operatorname{Cap}_{[1]}(E;V). For a Dini modulus put Φω(b)=2∑j≥0ω(2−jb),Φω(0)=0.\Phi_\omega(b)=2\sum_{j\geq0}\omega(2^{-j}b),\qquad \Phi_\omega(0)=0. The dyadic integral comparison proves

Φω(b)≤2ω(b)+2log⁡2∫0bω(r) drr,Φω(b)⟶0(b↓0).\Phi_\omega(b)\leq2\omega(b)+\frac2{\log2} \int_0^b\omega(r)\,\frac{dr}{r},\qquad \Phi_\omega(b)\longrightarrow0\quad(b\downarrow0).

Theorem (One-sided support-capacity estimate). Let u,v∈Cb(V)u,v\in\mathcal C_b(V) be bounded and suppose {u<v−δ}‾⋐V(δ>0).\overline{\{u<v-\delta\}}\Subset V\qquad(\delta>0). Suppose μ=(Du)n\mu=(Du)^n is carried by a compact H⋐VH\Subset V and

μ(E)≤C(E)ω(C(E)1/n)n(E⊂H Borel).\mu(E)\leq C(E)\omega(C(E)^{1/n})^n \quad(E\subset H\text{ Borel}).

Then for every s>0s>0,

sup⁡V(v−u)≤s+Φω(C(H∩{u<v−s})1/n).\sup_V(v-u)\leq s+\Phi_\omega\bigl( C(H\cap\{u<v-s\})^{1/n}\bigr).

In particular, if vjv_j satisfies the same gap condition and converges to uu in outer capacity on HH, with capacity domain VV, then lim sup⁡jsup⁡V(vj−u)≤0\limsup_j\sup_V(v_j-u)\leq0.

Status: NEW_PROVED.

Proof. Put Es={u<v−s}E_s=\{u<v-s\} and f(s)=C(H∩Es)1/nf(s)=C(H\cap E_s)^{1/n}. It is finite by CLN on HH. For any normalized classical tester 0≤b≤10\leq b\leq1, set w=v−s−t+tbw=v-s-t+tb. Then Es+t⊂{u<w}⊂Es.E_{s+t}\subset\{u<w\}\subset E_s. The middle set has compact closure. Positivity of the mixed expansion and compact-gap comparison give tn∫Es+t(Db)n≤∫{u<w}(Dw)n≤μ({u<w})≤μ(H∩Es).t^n\int_{E_{s+t}}(Db)^n \leq\int_{\{u<w\}}(Dw)^n \leq\mu(\{u<w\})\leq\mu(H\cap E_s). Taking the supremum over bb gives the stronger full-gap bound

tnC(Es+t)≤f(s)nω(f(s))n,tf(s+t)≤f(s)ω(f(s)).t^n C(E_{s+t})\leq f(s)^n\omega(f(s))^n, \qquad t f(s+t)\leq f(s)\omega(f(s)).

Starting at ss, repeat the halving iteration of Lemma “Dini sublevel iteration”, with increments 2ω(f(sj))2\omega(f(s_j)). The total increment is at most Φω(f(s))\Phi_\omega(f(s)). A zero increment with f(sj)>0f(s_j)>0 is impossible: the recurrence and increasing-set continuity give f(sj)=0f(s_j)=0 in that case. Hence, for L=s+Φω(f(s))L=s+\Phi_\omega(f(s)), one has f(L)=0f(L)=0.

The measure condition gives μ(EL)=μ(H∩EL)=0\mu(E_L)=\mu(H\cap E_L)=0. The strict gap ELE_L is only Borel; it is not called ordinary open. Apply the proved ambient pointwise domination theorem to uu and v−Lv-L. Its gap has compact closure, and the measure of uu gives it no mass, so u≥v−Lu\geq v-L at every point. This proves (eq:r3-one-sided-dini-bound). The convergence statement follows first for any fixed s>0s>0 from (eq:r3-stability-dyadic-tail), and then by s↓0s\downarrow0. ◻

Theorem (Continuity on the ordinary closure). On the extendible exhaustion domains of Theorem “Dirichlet solvability under a Dini modulus”, its solution extends to a continuous function on V‾\overline V with value zero on ∂V\partial V. In particular this holds for power-capacity data.

Proof. The proved compact-support barrier gives u≥Cρu\geq C\rho for a finite CC. The common-exterior extension lemma produces a classical function UU on a neighborhood of V‾\overline V, equal to uu on VV and to CρC\rho on the exterior. The function is zero on the ordinary boundary. This is a bounded classical extension, not yet an assertion of interior continuity.

B’s regularization provides continuous classical wj↓Uw_j\downarrow U on one neighborhood of V‾\overline V, with a common bound there. Since U=0U=0 on the compact boundary, Dini’s lemma applied only to the continuous restrictions on ∂V\partial V gives ϵj=max⁡∂Vwj↓0.\epsilon_j=\max_{\partial V}w_j\downarrow0. Put vj=wj−ϵjv_j=w_j-\epsilon_j on VV. The boundary upper limit of vj−uv_j-u is at most zero. Thus each positive-margin gap has compact closure. Decreasing relative capacity convergence of the wjw_j, and the fixed-domain comparison on the compact HH, give vj→uv_j\to u in outer ambient capacity there. The preceding theorem implies lim sup⁡jsup⁡V(vj−u)≤0\limsup_j\sup_V(v_j-u)\leq0. Since wj≥uw_j\geq u, it follows that wj→uw_j\to u uniformly on VV. On the boundary the errors are at most ϵj\epsilon_j. Hence the continuous wj∣V‾w_j|_{\overline V} converge uniformly to the extension of uu by zero, proving continuity on the whole ordinary closure. ◻

Theorem (Strong-data uniform stability for power growth). Fix an extendible (Y,ρ,V)(Y,\rho,V), a compact H⋐VH\Subset V, B>0B>0, and A,α>0A,\alpha>0. Suppose u,v∈Cb(V)u,v\in\mathcal C_b(V) have ordinary boundary limit zero, −B≤u,v≤0-B\leq u,v\leq0, and their measures μ,ν\mu,\nu are carried by HH and obey μ(E),ν(E)≤AC(E)1+α(E⊂H Borel).\mu(E),\nu(E)\leq A C(E)^{1+\alpha} \qquad(E\subset H\text{ Borel}). With d=dpsh,V(μ,ν)=sup⁡b∈Cb(V)0≤b≤1∣∫b d(μ−ν)∣,d=d_{\mathrm{psh},V}(\mu,\nu) =\sup_{\substack{b\in\mathcal C_b(V)\\0\leq b\leq1}} \left|\int b\,d(\mu-\nu)\right|, there is a finite constant depending only on the fixed data and the specified collar and finite CLN controls such that

∥u−v∥∞≤C0dα/[2n(n+2α)].\|u-v\|_\infty\leq C_0 d^{\alpha/[2^n(n+2\alpha)]}.

No optimality of this exponent is asserted.

Proof. The common compact support and bound give, by Lemma “A common lower barrier and exterior”, one larger domain Ω\Omega and classical extensions U,V′U,V' agreeing with a fixed collar function outside VV. Their mixed energy is I=∫H(u−v) d(ν−μ)≤2Bd.I=\int_H(u-v)\,d(\nu-\mu)\leq2B d. Indeed (u+B)/B(u+B)/B and (v+B)/B(v+B)/B are normalized testers on VV; the two constant terms cancel in their difference. The extensions have one bound on the chosen larger compact stage. Apply Theorem “An all-degree mixed-energy capacity estimate”, then reverse domain comparison only on H⋐VH\Subset V, to obtain

C(H∩{∣u−v∣>s})≤C1s−2d2−n.C(H\cap\{|u-v|>s\})\leq C_1s^{-2}d^{2^{-n}}.

The reverse comparison has been used on a compact core, not at the boundary of VV.

For the power modulus, Φω(b)=2A1/nbα/(1−2−α)\Phi_\omega(b)= 2A^{1/n}b^\alpha/(1-2^{-\alpha}). Apply the one-sided estimate in both orders and use (eq:r3-dual-gap-capacity): ∥u−v∥∞≤s+C2s−2α/ndα/(n2n).\|u-v\|_\infty\leq s+C_2s^{-2\alpha/n}d^{\alpha/(n2^n)}. For 0<d≤10<d\leq1 take s=dα/[2n(n+2α)]s=d^{\alpha/[2^n(n+2\alpha)]}. For d=0d=0 let s↓0s\downarrow0 in the gap estimate and domination, or use that the dual distance determines compact measures and ambient uniqueness. For d>1d>1, the uniform bound 2B2B is absorbed in the constant. This proves the claimed exponent. ◻

Corollary (Strong-data stability under a common Dini modulus). Under the same domain, compact support and uniform potential bound, replace the two power bounds by one common Dini bound. If dpsh,V(μj,μ)→0d_{\mathrm{psh},V}(\mu_j,\mu)\to0, the corresponding bounded zero-boundary solutions converge uniformly on V‾\overline V.

Proof. The energy argument still gives (eq:r3-dual-gap-capacity). For a fixed s>0s>0, the capacity on the right tends to zero. Apply the two one-sided Dini estimates and then (eq:r3-stability-dyadic-tail); finally let ss decrease to zero. The boundary values agree. This is a DIRECT_COROLLARY, not a weak-measure stability theorem. ◻

Nonconstant boundary data with a continuous maximal lift

Keep the original field and the extendible domain (Y,ρ,V)(Y,\rho,V) of Section “The fixed-domain Dirichlet problem under power capacity growth”. Let hh be a continuous classical psh function on all of YanY^{\mathrm{an}} and suppose (Dh)n=0on V.(Dh)^n=0\quad\text{on }V. Write g=h∣∂Vg=h|_{\partial V}. This is a hypothesis on a specified classical lift of the boundary function; it is not an assertion that every continuous boundary function admits such a lift.

Theorem (Dini data with a continuous maximal boundary lift). Let K0⋐VK_0\Subset V be compact and let μ\mu be a finite positive Radon measure carried by K0K_0. Suppose, for a Dini modulus ω\omega as in Definition “Dini capacity modulus”, that μ(E)≤C(E)ω(C(E)1/n)n,E⊂K0 Borel,C(E)=Cap⁡[1](E;V).\mu(E)\leq C(E)\omega(C(E)^{1/n})^n, \qquad E\subset K_0\text{ Borel},\qquad C(E)=\operatorname{Cap}_{[1]}(E;V). There is a unique bounded classical function uu on VV, continuous on V‾\overline V, with boundary value gg and (Du)n=μ(Du)^n=\mu. If M=μ(V)>0M=\mu(V)>0, set d=min⁡K0(−ρ),Sω(s)=s+2∑j≥0ω(2−jM1/n/s),S∗=inf⁡s>0Sω(s).d=\min_{K_0}(-\rho),\qquad S_\omega(s)=s+2\sum_{j\geq0}\omega(2^{-j}M^{1/n}/s), \qquad S_* =\inf_{s>0}S_\omega(s). Then the actual pointwise bounds are h+max⁡{−S∗,(S∗/d)ρ}≤u≤h.h+\max\{-S_*,(S_*/d)\rho\}\leq u\leq h. If M=0M=0, the solution is u=hu=h. The uniqueness assertion holds among all bounded classical solutions with the stated ordinary boundary limits, without an assumed interior continuity.

Status: NEW_PROVED for this explicit nonconstant-boundary extension. The mixed calculus, contact inequality, projective obstacle theorems and scalar Dini iteration are previously established inputs.

Proof. Assume M>0M>0. Fix s0>0s_0>0, put S=Sω(s0)S=S_\omega(s_0), and choose B>S/dB>S/d. The constrained family is FBh={v∈Cb(V):h+Bρ≤v≤h}.\mathcal F_B^h=\{v\in\mathcal C_b(V):h+B\rho\leq v\leq h\}. Its members already have the ordinary boundary limit gg.

The common affine extension. For ϵ>0\epsilon>0 glue max⁡(v,h+Bρ+ϵ)\max(v,h+B\rho+\epsilon) on VV to h+Bρ+ϵh+B\rho+\epsilon off VV. Near each boundary point the second branch is larger than hh, hence larger than vv. The prescriptions agree on an ordinary neighborhood, and their decreasing limit is classical, equal to vv on VV and to h+Bρh+B\rho outside. All members agree outside VV and have one bound on every chosen surrounding compact box. The proof of Lemma “An affine projective collar for a fixed lower obstacle” therefore gives a single ample line on a smooth projective compactification, and metrics UvU_v with weight v+cv+c on a neighborhood of V‾\overline V. The map remains affine, since its maximum collar is inactive on VV and all inputs are identical outside VV. Its continuous lower and upper metrics LB,FL_B,F have weights h+Bρ+ch+B\rho+c and h+ch+c on VV, and agree off VV. Use the global mixed energy EF\mathcal E_F relative to FF.

Existence of a constrained maximizer. Maximize F(Uv)=EF(Uv)−∫K0v dμ.\mathcal F(U_v)=\mathcal E_F(U_v)-\int_{K_0}v\,d\mu. The common bounds and projective intersection mass bound this functional. The midpoint argument of Lemma “A constrained maximizer” applies without a change: the integral is affine, the midpoint energy gap controls II, and the prescribed-box estimate followed by reverse framed domain comparison makes a maximizing sequence locally capacity Cauchy on every permitted local domain. The ambient completion has a unique actual classical representative. The continuous upper and lower metrics pass through its tail-usc formula, so this limit is admissible and equals the common exterior metric off VV. Weighted capacity continuity gives continuity of the energy. For the measure term, if ∣vj∣,∣v∣≤L|v_j|,|v|\leq L on K0K_0 and cj=C∗(K0∩{∣vj−v∣>δ})c_j=C^*(K_0\cap\{|v_j-v|>\delta\}), then ∫K0∣vj−v∣ dμ≤δM+2Lcjω(cj1/n)n⟶δM.\int_{K_0}|v_j-v|\,d\mu \leq\delta M+2L c_j\omega(c_j^{1/n})^n\longrightarrow\delta M. Letting δ↓0\delta\downarrow0 proves the required continuity. Thus a constrained maximizer UU exists, with local function uu.

The positive variation and the contact set. The set O={u<h}O=\{u<h\} is ordinary open because u−hu-h is usc and hh is continuous. For a nonnegative original smooth ϕ\phi with compact support in OO, choose t>0t>0 small enough that u+tϕ≤hu+t\phi\leq h there. Extend ϕ\phi by zero to the projective space and set Pt=PLb(U+tϕ)P_t=P_L^{\mathrm b}(U+t\phi). The actual usc-obstacle envelope satisfies U≤Pt≤U+tϕU\leq P_t\leq U+t\phi, is unchanged off the perturbation support, and is admissible. Orthogonality and energy concavity give, exactly with these global metrics, ∫ϕc1(L,Pt)n≤EF(Pt)−EF(U)t≤∫ϕc1(L,U)n.\int\phi c_1(L,P_t)^n \leq\frac{\mathcal E_F(P_t)-\mathcal E_F(U)}t \leq\int\phi c_1(L,U)^n. Uniform convergence of PtP_t to UU gives the common limit ∫ϕ(Du)n\int\phi(Du)^n. Maximality bounds the middle expression by ∫ϕ dμ\int\phi\,d\mu. Thus (Du)n≤μ(Du)^n\leq\mu on OO.

On the Borel contact set A={u=h}A=\{u=h\} the proved contact inequality, Corollary “Zero mass on contact with a maximal potential”, gives 1A(Du)n≤1A(Dh)n=0.\mathbf1_A(Du)^n\leq\mathbf1_A(Dh)^n=0. Combining the two disjoint pieces yields (Du)n≤μ(Du)^n\leq\mu on all of VV. In particular this measure is carried by K0K_0. There is no use of a strong maximum principle for u−hu-h, which is not asserted to be psh.

The estimate for the difference without differentiating it. Put f(s)=C({u<h−s})1/nf(s)=C(\{u<h-s\})^{1/n}. For a normalized classical tester 0≤a≤10\leq a\leq1 on VV, compare uu with the classical function w=h−s−t+ta.w=h-s-t+ta. The mixed expansion is positive and gives (Dw)n≥tn(Da)n(Dw)^n\geq t^n(Da)^n. Moreover {u<h−s−t}⊂{u<w}⊂{u<h−s}.\{u<h-s-t\}\subset\{u<w\}\subset\{u<h-s\}. All positive-margin gaps have compact closure by the common boundary values. Comparison and (Du)n≤μ(Du)^n\leq\mu therefore give tf(s+t)≤f(s)ω(f(s)),tf(t)≤M1/n.t f(s+t)\leq f(s)\omega(f(s)),\qquad t f(t)\leq M^{1/n}. The second estimate follows by decreasing the auxiliary positive shift to zero and using increasing-set continuity of raw capacity. Apply the scalar halving argument from Lemma “Dini sublevel iteration”: starting at s0s_0, the increments 2ω(f(sj))2\omega(f(s_j)) make ff decrease by factors of two, and their sum is bounded by 2∑j≥0ω(2−jM1/n/s0)2\sum_{j\geq0}\omega(2^{-j}M^{1/n}/s_0). If the modulus vanishes at a positive value encountered, the recurrence makes all deeper capacities zero and the iteration stops. In either case f(S)=0f(S)=0. The set {u<h−S}\{u<h-S\} is ordinary open, so ambient capacity nondegeneracy makes it empty. Thus u≥h−Su\geq h-S. This argument has used the psh functions u,h,w,au,h,w,a themselves; it has not treated u−hu-h as a psh potential.

The classical barrier b=h+max⁡{−S,(S/d)ρ}b=h+\max\{-S,(S/d)\rho\} equals h−Sh-S on K0K_0, so b≤ub\leq u there. Since (Du)n(Du)^n is carried by K0K_0, the strict gap {u<b}\{u<b\} has zero (Du)n(Du)^n-mass. Pointwise domination applied first to b−δb-\delta and then with δ↓0\delta\downarrow0 gives b≤ub\leq u everywhere. In particular u−(h+Bρ)≥(B−S/d)(−ρ)>0on V.u-(h+B\rho)\geq(B-S/d)(-\rho)>0\quad\text{on }V.

Release of the lower obstacle. For an arbitrary nonnegative original compact smooth ϕ\phi in VV, this strict margin permits LB≤U−tϕL_B\leq U-t\phi for all sufficiently small t>0t>0. Let Pt=PLb(U−tϕ)P_t=P_L^{\mathrm b}(U-t\phi). The lower metric and the global constant shift give LB≤Pt≤U−tϕ≤U,U−t∥ϕ∥∞≤Pt.L_B\leq P_t\leq U-t\phi\leq U, \qquad U-t\|\phi\|_\infty\leq P_t. Outside VV the lower metric equals UU, so Pt=UP_t=U there and PtP_t is admissible. Orthogonality and concavity, now in the negative direction, give ∫ϕc1(L,U)n≤EF(U)−EF(Pt)t≤∫ϕc1(L,Pt)n.\int\phi c_1(L,U)^n \leq\frac{\mathcal E_F(U)-\mathcal E_F(P_t)}t \leq\int\phi c_1(L,P_t)^n. For clarity, orthogonality reads ∫(U−Pt)c1(L,Pt)n=t∫ϕc1(L,Pt)n\int(U-P_t)c_1(L,P_t)^n=t\int\phi c_1(L,P_t)^n; the lower inequality uses U−Pt≥tϕU-P_t\geq t\phi. The common limit is ∫ϕ(Du)n\int\phi(Du)^n. Maximality gives EF(U)−EF(Pt)≥∫K0(U−Pt) dμ≥t∫K0ϕ dμ.\mathcal E_F(U)-\mathcal E_F(P_t) \geq\int_{K_0}(U-P_t)\,d\mu \geq t\int_{K_0}\phi\,d\mu. Thus (Du)n≥μ(Du)^n\geq\mu, and equality follows. All these energy differences are global expressions on one line; no boundary mass is inferred to vanish merely by restriction.

The construction gives one solution. Repeat only the a priori estimate for that solution for each s>0s>0 and take a sequence with Sω(s)→S∗S_\omega(s)\to S_*. This gives the two displayed bounds with S∗S_*; no minimizing parameter is required.

Continuity and uniqueness. The constructed global metric admits bounded continuous classical decreasing regularizations on one neighborhood of V‾\overline V by B. In the fixed frame write their local functions as wj↓uw_j\downarrow u on VV; on the boundary their limit is hh. Compactness of the boundary and Dini’s lemma give ϵj=max⁡∂V(wj−h)↓0.\epsilon_j=\max_{\partial V}(w_j-h)\downarrow0. The functions vj=wj−ϵjv_j=w_j-\epsilon_j have boundary upper limit at most gg. Their positive-margin gaps above uu have compact closure. Decreasing capacity convergence and domain comparison give vj→uv_j\to u in outer capacity on K0K_0, with domain VV. Theorem “One-sided support-capacity estimate” applies; its gap hypothesis does not require constant boundary values. It gives lim sup⁡jsup⁡V(vj−u)≤0\limsup_j\sup_V(v_j-u)\leq0. Since wj≥uw_j\geq u and ϵj→0\epsilon_j\to0, the wjw_j converge uniformly on VV, and also on its boundary, to the extension of uu by gg. This is the required continuity on V‾\overline V.

If two bounded classical solutions have the same ordinary boundary limits gg, the gaps of one above the other with any positive margin have compact closure. Compact-gap comparison and ambient pointwise domination give equality. One can verify this with the scaling argument after subtracting one sufficiently large constant from both functions: their boundary values then are strictly negative, so the comparison of bb with (1+ϵ)a−δ(1+\epsilon)a-\delta has compact gap. Equality of their measures and their finite mass on that compact gap force the gap to have zero measure, and domination removes it. Let δ,ϵ↓0\delta,\epsilon\downarrow0 and interchange a,ba,b. For M=0M=0, hh itself solves the problem and the same uniqueness argument applies. This completes the proof. ◻

Corollary (An upper lift with curvature dominated by the datum). The preceding theorem remains valid if the assumption (Dh)n=0(Dh)^n=0 is replaced by (Dh)n≤μon V.(Dh)^n\leq\mu\quad\text{on }V. All other hypotheses, the ordinary boundary values, continuity, uniqueness and estimates are unchanged.

Status: DIRECT_COROLLARY.

Proof. The contact inequality now gives 1A(Du)n≤1A(Dh)n≤1Aμ\mathbf1_A(Du)^n\leq\mathbf1_A(Dh)^n\leq\mathbf1_A\mu. Combined with the positive variation on {u<h}\{u<h\} it still gives (Du)n≤μ(Du)^n\leq\mu everywhere. Every subsequent step uses that inequality, the positivity of mixed terms, and the fixed boundary lift, and is unchanged. If μ=0\mu=0, the new hypothesis also forces (Dh)n=0(Dh)^n=0. The direction of the curvature inequality is substantive: this is an upper lift of smaller curvature than the prescribed datum, not an arbitrary continuous extension. ◻

Example (A nonconstant maximal lift in every dimension at least two). On Y=AKnY=\mathbf A_K^n, n≥2n\geq2, take h=max⁡(0,log⁡∣T1∣)h=\max(0,\log|T_1|) and ρ=max⁡{−1,log⁡∣T1∣−log⁡R,…,log⁡∣Tn∣−log⁡R},R>1.\rho=\max\{-1,\log|T_1|-\log R,\ldots, \log|T_n|-\log R\},\qquad R>1. These are continuous classical psh functions, and V={∣Ti∣<R for every i}V=\{|T_i|<R\text{ for every }i\} is an extendible domain of the required type. The curvature DhDh is nonzero, but its top power is zero because hh depends on one coordinate only and n≥2n\geq2. Its boundary restriction is nonconstant: it is zero at a boundary point with T1=0T_1=0 and ∣T2∣=R|T_2|=R, and is log⁡R\log R where ∣T1∣=R|T_1|=R. Thus the theorem includes explicit nonconstant boundary data with a nonconstant classical lift.

Exhaustion of subsolution-dominated data

Keep the extendible domain (Y,ρ,V)(Y,\rho,V) of Section “The fixed-domain Dirichlet problem under power capacity growth”, the original field and absolute value, and the ambient background [1][1]. Boundary limits in this section are ordinary boundary limits at every point of ∂V\partial V. The measures need only be locally finite.

Theorem (The bounded subsolution criterion without compact support). Let μ\mu be a locally finite positive Radon measure on VV. There exists a bounded u∈Cb(V)u\in\mathcal C_b(V) with zero ordinary boundary limit and (Du)n=μ(Du)^n=\mu if and only if there exists a bounded w∈Cb(V)w\in\mathcal C_b(V) with zero ordinary boundary limit and (Dw)n≥μ.(Dw)^n\geq\mu. The solution is unique and satisfies w≤u≤0w\leq u\leq0 for each such subsolution ww. The measure μ\mu is not required to have finite total mass or compact support. A constant boundary value is handled by adding that constant to the solution and subsolution.

Status: DIRECT_COROLLARY. The inputs are the proved compact-data bounded-subsolution theorem, the fixed-background decreasing BT theorem, and ambient pointwise comparison. The matching boundary condition on ww is essential to this exhaustion argument.

Proof. The reverse implication follows by taking w=uw=u. For the forward construction choose nested compact sets Hj⋐VH_j\Subset V with Hj⊂Hj+1∘H_j\subset H_{j+1}^{\circ} and ⋃jHj∘=V\bigcup_jH_j^{\circ}=V, and put μj=1Hjμ\mu_j=\mathbf1_{H_j}\mu. These are finite compactly carried measures and μj↑μ\mu_j\uparrow\mu. The same bounded ww is a subsolution for each μj\mu_j. Theorem “The bounded-subsolution criterion” supplies bounded classical uju_j with zero boundary limit and (Duj)n=μj(Du_j)^n=\mu_j. The weak-boundary order lemma “Weak boundary order for a subsolution” gives w≤uj+1≤uj≤0.w\leq u_{j+1}\leq u_j\leq0. Consequently uju_j decreases to an actual bounded classical function uu, with w≤u≤0w\leq u\leq0. The boundary sandwich implies that uu has zero limit at every ordinary boundary point. The decreasing BT theorem gives (Duj)n→(Du)n(Du_j)^n\to(Du)^n on every original compact smooth scalar test. If such a test has support HH, then H⊂HjH\subset H_j for every sufficiently large jj, so its pairing with μj\mu_j equals its pairing with μ\mu. Thus (Du)n=μ(Du)^n=\mu. No passage of total mass through the boundary is used.

For completeness, the comparison argument for uniqueness uses only finite mass on compact gaps, even if μ(V)=∞\mu(V)=\infty. Let a,ba,b be two bounded zero-boundary classical functions with (Da)n≥(Db)n(Da)^n\geq(Db)^n. For ϵ,δ>0\epsilon,\delta>0 the strict gap E={b<(1+ϵ)a−δ}E=\{b<(1+\epsilon)a-\delta\} has compact closure in VV by their boundary limits. The compact-gap comparison theorem gives (1+ϵ)n(Da)n(E)≤(Db)n(E)≤(Da)n(E).(1+\epsilon)^n(Da)^n(E)\leq(Db)^n(E)\leq(Da)^n(E). All three quantities are finite on this compact gap, so both measures vanish on EE. Ambient pointwise domination, with lower function bb, implies b≥(1+ϵ)a−δb\geq(1+\epsilon)a-\delta on VV. Let δ↓0\delta\downarrow0 and then ϵ↓0\epsilon\downarrow0 to obtain a≤ba\leq b. Equality of the measures gives equality of the functions by applying the argument in both orders. The same argument verifies the order used in the exhaustion, and uses strict gaps as Borel sets, without asserting that they are ordinary open. Translation by a constant proves the final assertion. ◻

Corollary (Mixed data dominated by bounded potentials). Let u1,…,un∈Cb(V)u_1,\ldots,u_n\in\mathcal C_b(V) be globally bounded and let ff be a Borel function with 0≤f≤C<∞0\leq f\leq C<\infty. Set μ=f Du1∧⋯∧Dun.\mu=f\,Du_1\wedge\cdots\wedge Du_n. If ff is compactly supported, this measure has a unique bounded zero-boundary classical solution. No boundary condition on the uiu_i is needed in this compactly supported case. If instead every uiu_i has zero ordinary boundary limit, the same conclusion holds without compact support of ff and without a finite-total-mass assumption.

Status: DIRECT_COROLLARY.

Proof. If C=0C=0, take the zero solution. Otherwise put w=(Cn!)1/n∑i=1nui.w=\left(\frac{C}{n!}\right)^{1/n}\sum_{i=1}^n u_i. This is a bounded classical function. Positivity of all mixed terms in the expansion and the coefficient n!n! of the product with all nn different factors give (Dw)n=Cn!(∑iDui)n≥C Du1∧⋯∧Dun≥μ.(Dw)^n =\frac{C}{n!}\left(\sum_i Du_i\right)^n \geq C\,Du_1\wedge\cdots\wedge Du_n\geq\mu. The mixed measure is locally finite by the local CLN theorem, so bounded Borel multiplication defines a locally finite Radon measure. For compactly supported ff, apply Theorem “The bounded-subsolution criterion”, which requires only a bounded subsolution in this case. If all the uiu_i have zero boundary limit, then so does ww, and the preceding theorem applies to the possibly noncompact data. This is an ambient diagonal Dirichlet conclusion; it makes no arbitrary-background relative Dirichlet existence assertion. ◻

Uniform power capacity growth does not imply weak-data stability

Keep Ω,λ,gi,H,zi,z,K∗,Ω0\Omega,\lambda,g_i,H,z_i,z,K_*,\Omega_0 from Section “Moving atoms and failure of weak-data Dirichlet stability”. Write μi=δzi\mu_i=\delta_{z_i} and μ=δz\mu=\delta_z.

Corollary (Weak instability under every fixed power bound). For each fixed α>0\alpha>0 there is one constant Aα=(2λ)n(1+α)A_\alpha=(2\lambda)^{n(1+\alpha)} such that all Borel E⊂ΩE\subset\Omega satisfy

μi(E), μ(E)≤AαCap⁡[1](E;Ω)1+αfor every i.\mu_i(E),\ \mu(E) \leq A_\alpha\operatorname{Cap}_{[1]}(E;\Omega)^{1+\alpha} \quad\hbox{for every }i.

Nevertheless the corresponding continuous bounded zero-boundary solutions gig_i do not converge uniformly or locally in outer capacity to HH, and have no local capacity-Cauchy subsequence. For n≥2n\geq2 they do not converge pointwise to HH. Moreover dpsh,Ω(μi,μ)≥12for every i.d_{\mathrm{psh},\Omega}(\mu_i,\mu)\geq\tfrac12 \quad\hbox{for every }i .

Status: DIRECT_COROLLARY of the explicit moving-atom construction. The assertion that weak data convergence, common compact support, common zero boundary values, and a uniform power capacity bound force capacity or uniform solution convergence is FALSE.

Proof. The functions bi=1+gi/(2λ)b_i=1+g_i/(2\lambda) and b=1+H/(2λ)b=1+H/(2\lambda) are normalized classical psh on Ω\Omega, because B<2λB<2\lambda. Their top measures are (2λ)−nδzi(2\lambda)^{-n}\delta_{z_i} and (2λ)−nδz(2\lambda)^{-n}\delta_z. Hence every set containing the respective atom has capacity at least (2λ)−n(2\lambda)^{-n}. This proves (eq:r3-common-power-capacity-bound); sets without that atom have zero measure. The convention at infinite capacity is harmless.

Let 0<ϵ<2λ−B0<\epsilon<2\lambda-B. The set K∗∩{∣gi−H∣>ϵ}K_*\cap\{|g_i-H|>\epsilon\} contains ziz_i. The same normalized tester restricted to the smaller ordinary domain Ω0\Omega_0 shows that its raw, and thus its outer, capacity on that fixed domain is at least (2λ)−n(2\lambda)^{-n}. This is failure of local all-domain capacity convergence, not merely an assertion on the original large domain. For i≠ji\ne j, ∣gi−gj∣(zi)=λ|g_i-g_j|(z_i)=\lambda; a fixed threshold less than λ\lambda gives the same lower bound for every pair and rules out a capacity-Cauchy subsequence. Uniform and, for n≥2n\geq2, pointwise failures were explicitly computed in the preceding theorem. Finally bi(zi)=0b_i(z_i)=0 and bi(z)=1/2b_i(z)=1/2, so this one normalized tester proves the psh-dual lower bound. ◻

This is consistent with the proved capacity-growth existence theorems and their a priori bounds. It excludes ordinary weak data convergence as a substitute for the strong psh-dual topology in the corresponding stability assertion.

Linear capacity domination does not ensure a bounded solution

Fix r=∣π∣r=|\pi|, λ=−log⁡r>0\lambda=-\log r>0, R=r−1R=r^{-1}, and the ordinary polydisc Ω={∣Tl∣<R}l=1n\Omega=\{|T_l|<R\}_{l=1}^n. Set tj=2jλ,rj=Re−tj=r 2j−1,s=max⁡llog⁡∣Tl∣−log⁡R.t_j=2^j\lambda,\qquad r_j=R e^{-t_j}=r^{\,2^j-1},\qquad s=\max_l\log|T_l|-\log R . Let zjz_j be the diagonal radius-rjr_j Gauss point, let Bj={∣Tl∣≤rj}l=1nB_j=\{|T_l|\leq r_j\}_{l=1}^n, and let oo be the rigid origin. Thus zj→oz_j\to o. Write Kj={o,zj,zj+1,…}K_j=\{o,z_j,z_{j+1},\ldots\}.

Lemma (Exact radial compact capacities). For every jj, Uj=max⁡(−1,s/tj),(DUj)n=tj−nδzj,U_j=\max(-1,s/t_j),\qquad (DU_j)^n=t_j^{-n}\delta_{z_j}, and, on the same fixed domain Ω\Omega, Cap⁡[1]({zj};Ω)=Cap⁡[1](Kj;Ω)=Cap⁡[1](Bj;Ω)=tj−n.\operatorname{Cap}_{[1]}(\{z_j\};\Omega) =\operatorname{Cap}_{[1]}(K_j;\Omega) =\operatorname{Cap}_{[1]}(B_j;\Omega)=t_j^{-n}.

Proof. The ordinary torus slope simplex has volume tj−n/n!t_j^{-n}/n!; the original factor n!n! gives the measure identity, with no mass on lower-rank strata or coordinate zero loci. The function 1+Uj1+U_j is a normalized tester, giving the singleton lower bound. For an arbitrary normalized a∈Cb(Ω)a\in\mathcal C_b(\Omega) and ϵ>0\epsilon>0, compare v=(1+ϵ)Ujv=(1+\epsilon)U_j with a−1a-1. The compact set BjB_j lies in {v<a−1}\{v<a-1\}. For every δ>0\delta>0 the closure of {v<a−1−δ}\{v<a-1-\delta\} is compactly contained in Ω\Omega, since it lies in {v<−δ}\{v<-\delta\}. Thus the proved comparison principle gives ∫Bj(Da)n≤∫{v<a−1}(Da)n≤∫{v<a−1}(Dv)n≤(1+ϵ)ntj−n.\int_{B_j}(Da)^n \leq\int_{\{v<a-1\}}(Da)^n \leq\int_{\{v<a-1\}}(Dv)^n \leq(1+\epsilon)^n t_j^{-n}. Let ϵ↓0\epsilon\downarrow0, then take the supremum over aa. The inclusions {zj}⊂Kj⊂Bj\{z_j\}\subset K_j\subset B_j give all three equalities. No outer regularity interchange is used. ◻

Theorem (A finite nonpolar linearly dominated datum). The measure

μ=∑j=1∞mjδzj,mj=tj−n−tj+1−n>0\mu=\sum_{j=1}^\infty m_j\delta_{z_j},\qquad m_j=t_j^{-n}-t_{j+1}^{-n}>0

is finite, supported in K1⋐ΩK_1\Subset\Omega, and gives no mass to locally pluripolar sets. It satisfies, for every Borel EE,

μ(E)≤Cap⁡[1](E;Ω).\mu(E)\leq\operatorname{Cap}_{[1]}(E;\Omega).

There is no locally bounded actual classical psh solution with ordinary boundary limit zero and (Du)n=μ(Du)^n=\mu. For every α>0\alpha>0 there is no finite constant AA for which μ(E)≤ACap⁡[1](E;Ω)1+α\mu(E)\leq A\operatorname{Cap}_{[1]}(E;\Omega)^{1+\alpha} holds for all Borel EE.

Status: NEW_PROVED counterexample. The assertion that finite nonpolar data and linear capacity domination suffice for bounded zero-boundary solvability is FALSE in every dimension.

Proof. The total mass is t1−nt_1^{-n}. Every zjz_j is divisorial and has positive singleton capacity, so an ambient locally polar set contains none of these points. The origin has zero mass. If EE contains any atoms, let jj be its least atom index. Then μ(E)≤∑l≥jml=tj−n=Cap⁡[1]({zj};Ω)≤Cap⁡[1](E;Ω).\mu(E)\leq\sum_{l\geq j}m_l=t_j^{-n} =\operatorname{Cap}_{[1]}(\{z_j\};\Omega) \leq\operatorname{Cap}_{[1]}(E;\Omega). Otherwise μ(E)=0\mu(E)=0. Moreover μ(Kj)=Cap⁡[1](Kj;Ω)=tj−n\mu(K_j)=\operatorname{Cap}_{[1]}(K_j;\Omega)=t_j^{-n}, which disproves every uniform higher power bound.

We compute the finite partial solutions, rather than assume weak stability for their data. Put Aj(N)=(∑k=jNmk)1/n(1≤j≤N),AN+1(N)=0A_j^{(N)}=\left(\sum_{k=j}^N m_k\right)^{1/n} \quad(1\leq j\leq N),\qquad A_{N+1}^{(N)}=0 and

uN=∑j=1N(Aj(N)−Aj+1(N))max⁡(s,−tj).u_N=\sum_{j=1}^N (A_j^{(N)}-A_{j+1}^{(N)})\max(s,-t_j).

It is continuous classical, has ordinary zero boundary values, and is bounded for each fixed NN. Its scalar radial slope between successive breaks is Aj(N)A_j^{(N)}. At zjz_j the centered local tropical function, up to an additive constant, is max⁡{Ax1,…,Axn,Bx1,…,Bxn},A=Aj(N),B=Aj+1(N).\max\{A x_1,\ldots,A x_n,B x_1,\ldots,B x_n\}, \qquad A=A_j^{(N)},\quad B=A_{j+1}^{(N)}. The convex hull of the slopes is {ξl≥0, B≤∑lξl≤A}\{\xi_l\geq0,\ B\leq\sum_l\xi_l\leq A\}, with volume (An−Bn)/n!(A^n-B^n)/n!. Its original top mass is therefore An−Bn=mjA^n-B^n=m_j.

Here the calibration can be verified directly by smoothing the finite maximum with log-sum-exp. Its Hessian is positive definite when the slope hull has interior. For any interior slope target ξ\xi, the smooth function minus ξ⋅x\xi\cdot x is coercive and strictly convex, so its gradient is a diffeomorphism onto that interior. Change of variables gives the displayed volume. The original coordinate tropical integration has multiplicity one and factor n!n!. Scaling the smoothing parameter concentrates this mass at the original Gauss point. At other strata the active slope differences have rank below nn; coordinate zero branches are locally inactive. Smooth original integration under refinements and bounded BT convergence therefore prove, on all original compact tests,

(DuN)n=∑j=1Nmjδzj.(Du_N)^n=\sum_{j=1}^N m_j\delta_{z_j}.

This computation applies to any positive finite weights at these radial points and will be used below with other weights.

With t0=0t_0=0, summation by parts gives the actual origin value −uN(o)=∑j=1NAj(N)(tj−tj−1).-u_N(o)=\sum_{j=1}^N A_j^{(N)}(t_j-t_{j-1}). Since Aj(N)≥mj1/n=(1−2−n)1/n/tjA_j^{(N)}\geq m_j^{1/n} =(1-2^{-n})^{1/n}/t_j, it follows that −uN(o)≥(1−2−n)1/n(1+N−12)⟶∞.-u_N(o)\geq(1-2^{-n})^{1/n} \left(1+\frac{N-1}{2}\right)\longrightarrow\infty. If uu were a locally bounded zero-boundary solution for μ\mu, measure order and the ordinary boundary limits would imply u≤uNu\leq u_N for every NN, because (Du)n≥(DuN)n(Du)^n\geq(Du_N)^n. This contradicts the finite actual value u(o)u(o). The comparison is valid on compact gaps, so it does not require subtracting infinite total masses near the boundary. ◻

A non-Dini modulus obstruction in every dimension

Use the same fixed ordinary polydisc, radial points, and scales tj=2jλt_j=2^j\lambda as in Section “Linear capacity domination does not ensure a bounded solution”. Write qj=tj−1q_j=t_j^{-1} and Cj=tj−nC_j=t_j^{-n}. For a finite nondecreasing function ω:(0,∞)→(0,∞)\omega:(0,\infty)\to(0,\infty) set F(x)=x ω(x1/n)n(0<x<∞),F(0)=0,F(∞)=∞.F(x)=x\,\omega(x^{1/n})^n\quad(0<x<\infty),\qquad F(0)=0,\qquad F(\infty)=\infty.

Theorem (Failure of the universal guarantee outside the Dini range). Suppose

∫0aω(t) dtt=∞for some a>0.\int_0^a\omega(t)\,\frac{dt}{t}=\infty \quad\hbox{for some }a>0.

There is a finite nonpolar positive measure μ\mu supported in the fixed compact K1={o,z1,z2,…}⋐ΩK_1=\{o,z_1,z_2,\ldots\}\Subset\Omega such that every Borel E⊂ΩE\subset\Omega satisfies

μ(E)≤F(Cap⁡[1](E;Ω)),\mu(E)\leq F\bigl(\operatorname{Cap}_{[1]}(E;\Omega)\bigr),

but μ\mu has no locally bounded actual classical zero-boundary Monge–Ampere solution.

Status: NEW_PROVED counterexample, using the already calibrated radial finite solutions. The uniform existence guarantee for this prescribed non-Dini modulus is FALSE; this is not an if-and-only-if criterion for an individual measure.

Proof. Set Wj=Cjω(qj)n,mj=Wj−Wj+1,μ=∑j=1∞mjδzj.W_j=C_j\omega(q_j)^n,\qquad m_j=W_j-W_{j+1},\qquad \mu=\sum_{j=1}^{\infty}m_j\delta_{z_j}. Monotonicity of ω\omega and qj+1=qj/2q_{j+1}=q_j/2 imply

Wj+1≤2−nWj,mj≥(1−2−n)Wj>0.W_{j+1}\leq2^{-n}W_j,\qquad m_j\geq(1-2^{-n})W_j>0.

Since ω(qj)≤ω(q1)<∞\omega(q_j)\leq\omega(q_1)<\infty, Wj→0W_j\to0. The total mass is W1W_1, and its tail from index jj is WjW_j. As before it is nonpolar: all its atoms are divisorial, and the origin has no mass.

If a Borel set contains atoms, take its least atom index jj. Lemma “Exact radial compact capacities” and monotonicity of FF give μ(E)≤Wj=F(Cj)=F(Cap⁡[1]({zj};Ω))≤F(Cap⁡[1](E;Ω)).\mu(E)\leq W_j=F(C_j) =F(\operatorname{Cap}_{[1]}(\{z_j\};\Omega)) \leq F(\operatorname{Cap}_{[1]}(E;\Omega)). Sets containing no atom have zero mass. This proves the growth bound, including the cases of zero or infinite capacity.

For the first NN weights use the explicit finite solution (eq:r3-radial-partial-solution), now with Aj(N)=(∑k=jNmk)1/nA_j^{(N)}=(\sum_{k=j}^Nm_k)^{1/n}. The calibrated identity (eq:r3-radial-partial-measure) holds for these weights as well. Its origin depth obeys −uN(o)=∑j=1NAj(N)(tj−tj−1)≥(1−2−n)1/n(ω(q1)+12∑j=2Nω(qj)).\begin{split} -u_N(o) &=\sum_{j=1}^N A_j^{(N)}(t_j-t_{j-1})\\ &\geq(1-2^{-n})^{1/n} \left(\omega(q_1)+\frac12\sum_{j=2}^N\omega(q_j)\right). \end{split} For every jj, monotonicity gives (log⁡2)ω(qj+1)≤∫qj+1qjω(t) dtt≤(log⁡2)ω(qj).(\log2)\omega(q_{j+1}) \leq\int_{q_{j+1}}^{q_j}\omega(t)\,\frac{dt}{t} \leq(\log2)\omega(q_j). The integral over any compact interval bounded away from zero is finite. Hypothesis (eq:r3-nondini-hypothesis) therefore forces ∑jω(qj)=∞\sum_j\omega(q_j)=\infty. The depths tend to infinity. Any bounded zero-boundary solution for μ\mu would lie below every finite partial solution by measure order, contradicting its finite actual value at the origin. ◻

Corollary (The logarithmic threshold). For 0<β≤10<\beta\leq1, there is such a measure with μ(E)≤Cap⁡[1](E;Ω)[log⁡(e+Cap⁡[1](E;Ω)−1/n)]nβ\mu(E)\leq \frac{\operatorname{Cap}_{[1]}(E;\Omega)} {\bigl[\log(e+\operatorname{Cap}_{[1]}(E;\Omega)^{-1/n})\bigr]^{n\beta}} at finite positive capacity, yet with no bounded zero-boundary solution. At zero and infinite capacity use the preceding conventions.

Proof. Take ω(t)=[log⁡(e+1/t)]−β\omega(t)=[\log(e+1/t)]^{-\beta}. It is finite, positive, and nondecreasing, and its Dini integral diverges exactly for 0<β≤10<\beta\leq1 in this range. ◻

Allowing a finite nonnegative nondecreasing modulus does not enlarge the non-Dini case: if it vanished at any positive argument, it would vanish on the whole smaller interval and its Dini integral there would be finite. Combined with the proved Dini sufficiency theorem, the construction gives the sharp universal-modulus threshold in the stated fixed-domain, compact-support class. It makes no assertion that every individual datum outside the Dini range fails.

The sharp universal capacity-modulus criterion on a polydisc

Fix n≥1n\geq1, the original finite extension K/QpK/\mathbf Q_p, a uniformizer π\pi, and the same polydisc as in the radial obstruction: R=∣π∣−1,Ω={∣Tl∣<R:1≤l≤n},C(E)=Cap⁡[1](E;Ω).R=|\pi|^{-1},\qquad \Omega=\{|T_l|<R:1\leq l\leq n\},\qquad C(E)=\operatorname{Cap}_{[1]}(E;\Omega). Let ω:[0,∞)→[0,∞)\omega:[0,\infty)\to[0,\infty) be finite, nondecreasing, and zero at zero. Put F(c)=cω(c1/n)nF(c)=c\omega(c^{1/n})^n for finite c>0c>0, with F(0)=0F(0)=0 and F(+∞)=+∞F(+\infty)=+\infty.

Corollary (A universal rather than individual equivalence). The following assertions are equivalent.

  1. ∫0aω(t) dt/t<∞\int_0^a\omega(t)\,dt/t<\infty for some a>0a>0.

  2. Every finite positive Radon measure μ\mu carried by a compact subset of Ω\Omega and satisfying μ(E)≤F(C(E))\mu(E)\leq F(C(E)) for every Borel E⊂ΩE\subset\Omega has a bounded actual classical zero-boundary solution of (Du)n=μ(Du)^n=\mu.

  3. Every measure in (2) has such a solution continuous on Ω‾\overline\Omega.

Such bounded zero-boundary solutions are unique. If (1) fails, a counterexample can be chosen to be a sum of positive atoms at original divisorial points tending to the rigid origin, with no mass at the origin or on any ambient locally pluripolar set.

Status: DIRECT_COROLLARY of the proved sufficient Dini theorem, its continuity improvement, and the non-Dini radial construction. The quantifier is over all measures with the specified bound, not over one individual measure.

Proof. The function ρ=max⁡{−1,log⁡∣T1∣−log⁡R,…,log⁡∣Tn∣−log⁡R}\rho=\max\{-1,\log|T_1|-\log R,\ldots, \log|T_n|-\log R\} is continuous classical on the whole affine space and makes Ω={ρ<0}\Omega=\{\rho<0\} an extendible exhaustion domain. Therefore (1) implies (3) by Theorems “Dirichlet solvability under a Dini modulus” and “Continuity on the ordinary closure”. Compactness of the ordinary closure makes (3) imply (2).

If (1) fails, ω(t)>0\omega(t)>0 for every t>0t>0. Otherwise ω(t0)=0\omega(t_0)=0 would make it vanish on (0,t0](0,t_0], and finite monotonicity would make its Dini integral finite on every compact interval away from zero. Thus the restriction of ω\omega to (0,∞)(0,\infty) satisfies the hypotheses of Theorem “Failure of the universal guarantee outside the Dini range”. Its explicit measure on this same polydisc contradicts (2), and has the stated nonpolar atomic form. Uniqueness is ambient pointwise comparison. ◻

The zero measure, for example, satisfies every such growth bound and has the zero solution. Thus failure of the Dini condition does not forbid solutions for all individual data. The assertion also does not extend to arbitrary relative backgrounds with ambient classical solutions, as Theorem “A bounded matching subsolution does not suffice” shows.

A nonzero polar background with no classical representative

Let V={∣x∣<1, ∣y∣<1}⊂(AK2)an,Y1={x=0}∩V,Y2={y=0}∩V,V=\{|x|<1,\ |y|<1\}\subset(\mathbf A_K^2)^{\mathrm{an}}, \qquad Y_1=\{x=0\}\cap V,\quad Y_2=\{y=0\}\cap V, and S=[Y1]+[Y2]S=[Y_1]+[Y_2]. This is a nonzero symmetric original-positive separately closed (1,1)(1,1)-current on a connected smooth space. The remaining degree is one. All capacities retain normalized ambient actual classical testers.

Proposition (Incompatible branch values in a completed class). There is a∈AS[0,1](V)a\in\mathcal A_S^{[0,1]}(V) which is an actual increasing limit of continuous normalized classical functions, but no actual classical psh function is equal to aa locally outside an outer SS-capacity-zero set. One can take aj=max⁡{0,1+j−1log⁡∣x∣}↑a,a=1{x≠0}.a_j=\max\{0,1+j^{-1}\log|x|\}\uparrow a, \qquad a=\mathbf1_{\{x\ne0\}}. Its completed relative curvature is D(aS)=0D(aS)=0.

Status: NEW_PROVED counterexample; the unrestricted classical-representative assertion is FALSE.

Proof. Each aja_j is continuous classical and lies in [0,1][0,1]. The finite floor removes the logarithmic singularity. Because log⁡∣x∣<0\log|x|<0 on VV, the rows increase to the displayed actual Borel limit. The proved increasing-capacity theorem places it in the normalized relative completion on every permitted local compact/domain pair.

Let ξi,r\xi_{i,r} be the centered Gauss point of radius 0<r<10<r<1 on YiY_i. Each singleton has strictly positive relative capacity. For example, with ℓr=−log⁡r\ell_r=-\log r, the normalized tester br(y)=max⁡{0,1+ℓr−1log⁡∣y∣}b_r(y)=\max\{0,1+\ell_r^{-1}\log|y|\} has D(brS)=ℓr−1δξ1,rD(b_rS)=\ell_r^{-1}\delta_{\xi_{1,r}}; its restriction to Y2Y_2 is zero. Exchanging x,yx,y gives the other singleton. These are ordinary original-field curve calibrations. The same positive atomic mass can be tested on a smaller local domain containing the given point, so a local outer-null exceptional set cannot contain any ξi,r\xi_{i,r}.

If a classical ww represented the relative class, it would satisfy w(ξ1,r)=0w(\xi_{1,r})=0 and w(ξ2,r)=1w(\xi_{2,r})=1 for every rr. Restrict ww to the first analytic axis. The classical maximum theorem on its closed radius-rr disc gives w(o)≤w(ξ1,r)=0w(o)\leq w(\xi_{1,r})=0. On the other axis, ξ2,r→o\xi_{2,r}\to o as r↓0r\downarrow0. Upper semicontinuity gives w(o)≥1w(o)\geq1. This contradiction proves nonrepresentability without assuming that ww is continuous.

On Y1Y_1 every row is zero; on Y2Y_2 its Laplacian is the atom j−1δξ2,e−jj^{-1}\delta_{\xi_{2,e^{-j}}}. Hence D(ajS)→0D(a_jS)\to0 on all original compact tests. The completed increasing calculus identifies the limit with D(aS)D(aS), proving the last assertion. ◻

This is not a counterexample to compact capacity regularity: its completed curvature is zero. Nor is aa a difference of two locally bounded classical functions. For any such function hh, the maximum theorem and upper semicontinuity imply h(ξi,r)→h(o)h(\xi_{i,r})\to h(o) on each axis. A difference must therefore have the same limiting value on both axes, whereas aa has limits zero and one. Thus this example does not refute the still open potential quotient for differences of classical potentials.

A relative Dirichlet obstruction in every positive codimension

Fix integers p,q≥1p,q\geq1, put n=p+qn=p+q, and let V={∣T1∣<1,…,∣Tn∣<1}⊂(AKn)an.V=\{|T_1|<1,\ldots,|T_n|<1\}\subset(\mathbf A_K^n)^{\rm an}. Choose a uniformizer π\pi, and write r=∣π∣r=|\pi| and λ=−log⁡r>0\lambda=-\log r>0. The actual function G=max⁡{log⁡∣T1∣,…,log⁡∣Tn∣,log⁡r}G=\max\{\log|T_1|,\ldots,\log|T_n|,\log r\} is bounded continuous classical psh on VV, lies between −λ-\lambda and zero, and has ordinary boundary limit zero. Let Z1={T1=Tq+2=⋯=Tn=0}∩V,Z2={T2=Tq+2=⋯=Tn=0}∩V,\begin{split} Z_1&=\{T_1=T_{q+2}=\cdots=T_n=0\}\cap V,\\ Z_2&=\{T_2=T_{q+2}=\cdots=T_n=0\}\cap V , \end{split} with the additional equations absent when p=1p=1. These are smooth qq-dimensional coordinate polydiscs. Write ii:Zi↪Vi_i:Z_i\hookrightarrow V and S=[Z1]+[Z2]S=[Z_1]+[Z_2]. This is a nonzero symmetric original-positive separately closed (p,p)(p,p)-current with connected support.

Let ziz_i be the Gauss point of the radius-rr closed polydisc in the qq free coordinates of ZiZ_i. Both are points of the original KK-analytic space. They lie outside Z1∩Z2Z_1\cap Z_2 and are divisorial intrinsically on ZiZ_i. Put μ=δz1+2qδz2.\mu=\delta_{z_1}+2^q\delta_{z_2}. All statements below concern all original compact smooth tests.

Theorem (A bounded matching subsolution does not suffice). There is no u∈Cb(V)u\in\mathcal C_b(V) with ordinary boundary limit zero and TS,q(u,…,u)=μ.T_{S,q}(u,\ldots,u)=\mu. Nevertheless the bounded continuous classical function w=2Gw=2G has the same boundary values and satisfies

TS,q(w,…,w)=2qδz1+2qδz2≥μ.T_{S,q}(w,\ldots,w) =2^q\delta_{z_1}+2^q\delta_{z_2}\geq\mu.

For every α>0\alpha>0, all Borel E⊂VE\subset V satisfy

μ(E)≤AαCap⁡S(E;V)1+α,Aα=(1+2q)λq(1+α).\mu(E)\leq A_\alpha\operatorname{Cap}_S(E;V)^{1+\alpha}, \qquad A_\alpha=(1+2^q)\lambda^{q(1+\alpha)}.

Status: NEW_PROVED counterexample. Thus the arbitrary-background classical bounded-subsolution Dirichlet assertion is FALSE, even with this power capacity bound in every remaining degree q≥1q\geq1.

Proof. On ZiZ_i, let gi=G∣Zig_i=G|_{Z_i}. The original tropical calibration of the radial maximum gives

(Dgi)q=δzi.(Dg_i)^q=\delta_{z_i}.

Indeed near the radius-rr Gauss point its slope polytope is conv⁡(0,e1,…,eq)\operatorname{conv}(0,e_1,\ldots,e_q) of volume 1/q!1/q!; the original CLD top-product normalization is q!q! and the coordinate tropical multiplicity is one. Elsewhere the slope differences have rank below qq. One can smooth the maximum and pass by the proved bounded BT continuity, so the equality is an ordinary-current equality on all original tests.

For a hypothetical uu, analytic restriction preserves the actual classical class. Integration-current compatibility and background linearity give

TS,q(u,…,u)=i1∗(D(u∣Z1))q+i2∗(D(u∣Z2))q.T_{S,q}(u,\ldots,u) =i_{1*}(D(u|_{Z_1}))^q+i_{2*}(D(u|_{Z_2}))^q .

Both summands are positive and carried by their respective coordinate subspaces. Since z1∉Z2z_1\notin Z_2 and z2∉Z1z_2\notin Z_1, equality with μ\mu forces the intrinsic measures to be δz1\delta_{z_1} and 2qδz22^q\delta_{z_2}, respectively. In particular there is no mass on the intersection: a positive summand cannot cancel mass there.

The ordinary boundary of each coordinate polydisc lies in the ordinary boundary of VV, so both restrictions have boundary limit zero. The proved intrinsic ambient uniqueness theorem, which determines actual bounded classical potentials, and (eq:r3-relative-branch-measure) imply u∣Z1=g1,u∣Z2=2g2.u|_{Z_1}=g_1,\qquad u|_{Z_2}=2g_2 . At the common rigid origin these values are −λ-\lambda and −2λ-2\lambda, a contradiction.

Formula (eq:r3-relative-matching-subsolution) follows from the same integration identity and positive homogeneity. Finally b=1+G/λb=1+G/\lambda is an actual normalized tester on VV and TS,q(b,…,b)=λ−q(δz1+δz2).T_{S,q}(b,\ldots,b) =\lambda^{-q}(\delta_{z_1}+\delta_{z_2}). If EE contains an atom of μ\mu, its capacity is at least λ−q\lambda^{-q}, whereas μ(E)≤1+2q\mu(E)\leq1+2^q. This gives (eq:r3-relative-power-datum); if it contains no atom the inequality is immediate. ◻

Proposition (An actual completed solution). Define, with the actual value log⁡0=−∞\log0=-\infty, aj=max⁡{2G, G+j−1log⁡∣T2∣}.a_j=\max\{2G,\ G+j^{-1}\log|T_2|\}. These are bounded continuous classical psh functions, increase pointwise, and have boundary limit zero. Their actual limit is a={G,T2≠0,2G,T2=0.a= \begin{cases} G,&T_2\ne0,\\ 2G,&T_2=0. \end{cases} It belongs to the bounded relative capacitary completion AS,b(V)\mathcal A_{S,b}(V), has boundary limit zero, and satisfies

TS,q(a,…,a)=μ.\mathcal T_{S,q}(a,\ldots,a)=\mu.

The sequence converges locally in outer SS-capacity on all permitted compact/domain pairs. In contrast, a∗=Ga^*=G and TS,q(a∗,…,a∗)=δz1+δz2≠μT_{S,q}(a^*,\ldots,a^*)=\delta_{z_1}+\delta_{z_2}\ne\mu.

Status: DIRECT_COROLLARY of the actual increasing-limit calculus, integration compatibility, and intrinsic ambient polar zero, applied to this explicit construction.

Proof. The maximum with the bounded floor handles the logarithmic pole. It is classical by truncating the logarithm and decreasing to the bounded maximum. It is continuous at the zero locus because the logarithmic branch is inactive on a neighborhood of each such point. Since log⁡∣T2∣≤0\log|T_2|\leq0 on VV, the sequence increases and 2G≤aj≤G≤0.2G\leq a_j\leq G\leq0. These bounds give its boundary limit and the stated actual pointwise limit. The relative increasing-limit theorem gives membership in AS,b\mathcal A_{S,b} and local all-domain capacity convergence.

To identify the measure, restrict only the classical approximants. On Z2Z_2, every restriction is 2g22g_2. On Z1Z_1 their actual increasing limit equals g1g_1 off the proper intrinsic analytic divisor Z1∩Z2Z_1\cap Z_2, and equals 2g12g_1 on that divisor. The divisor has intrinsic ambient outer capacity zero. Hence the intrinsic completed class and its measure are those of g1g_1. Intrinsic increasing continuity gives (D(aj∣Z1))q⟶δz1,(D(aj∣Z2))q=2qδz2.(D(a_j|_{Z_1}))^q\longrightarrow\delta_{z_1},\qquad (D(a_j|_{Z_2}))^q=2^q\delta_{z_2}. Apply (eq:r3-relative-axis-splitting) to each classical aja_j. Pushforward against all original tests, followed by the relative completed limit, proves (eq:r3-relative-completed-solution). No pullback theorem for an arbitrary completed Borel representative is assumed.

The complement of {T2=0}\{T_2=0\} is dense: normal Gauss points of positive radius approach every point of that coordinate hyperplane. Continuity of GG and a≤Ga\leq G therefore give a∗=Ga^*=G. The last measure formula follows from (eq:r3-relative-branch-measure). ◻

Scope.

This example works for every positive codimension pp and every remaining degree qq. It does not refute the ambient S=[1]S=[1] Dirichlet theorem. It supplies a particular completed solution, not a general arbitrary-background completed Dirichlet theorem, and it does not define a wedge of two arbitrary currents.

Total-variation limits may require a completed solution

Use p,q,V,Zi,S,G,aj,μp,q,V,Z_i,S,G,a_j,\mu from Section “A relative Dirichlet obstruction in every positive codimension”. Let ζj∈Z1\zeta_j\in Z_1 be the Gauss point whose free coordinate T2T_2 has radius rjr^j and whose other q−1q-1 free coordinates have radius rr. Thus ζ1=z1\zeta_1=z_1.

Proposition (The exact measures of the approximants). For every integer j≥1j\geq1 one has the ordinary measure identity

TS,q(aj,…,aj)=μ+2q−1jδζj.T_{S,q}(a_j,\ldots,a_j) =\mu+\frac{2^q-1}{j}\delta_{\zeta_j}.

Consequently

∥TS,q(ajq)−μ∥TV=dpsh,V(TS,q(ajq),μ)=2q−1j.\|T_{S,q}(a_j^q)-\mu\|_{\rm TV} =d_{\mathrm{psh},V}(T_{S,q}(a_j^q),\mu) =\frac{2^q-1}{j}.

All these measures have one common compact support in VV.

Status: NEW_PROVED for the exact formula. The convergence and closure obstruction below are DIRECT_COROLLARY.

Proof. On Z2Z_2, aj=2g2a_j=2g_2 and its intrinsic top measure is 2qδz22^q\delta_{z_2}. On Z1Z_1 write t1=T2t_1=T_2 and tl=Tl+1t_l=T_{l+1} for 2≤l≤q2\leq l\leq q. On its torus let sl=log⁡∣tl∣<0s_l=\log|t_l|<0 and set h=max⁡(−λ,s2,…,sq),g=max⁡(s1,h),kj=max⁡(s1/j,h).h=\max(-\lambda,s_2,\ldots,s_q),\quad g=\max(s_1,h),\quad k_j=\max(s_1/j,h). For q=1q=1, h=−λh=-\lambda. Since s1≤s1/js_1\leq s_1/j, a direct three-case comparison with hh gives

aj∣Z1=max⁡(2g,g+s1/j)=g+kj.a_j|_{Z_1}=\max(2g,g+s_1/j)=g+k_j.

The equality extends over the coordinate zero loci by continuity. For j=1j=1 this is 2g2g, giving 2qδz12^q\delta_{z_1}, exactly the asserted formula when ζ1=z1\zeta_1=z_1.

Suppose j>1j>1. At z1z_1, all sl=−λs_l=-\lambda, and kj=s1/jk_j=s_1/j is affine near this tropical point. Hence the mass there is the mass of gg, namely one. At ζj\zeta_j the tropical coordinates are s1=−jλs_1=-j\lambda and sl=−λs_l=-\lambda for l≥2l\geq2. In centered coordinates x1=s1+jλx_1=s_1+j\lambda, xl=sl+λx_l=s_l+\lambda one has g+kj=−2λ+max⁡(0,x2,…,xq)+max⁡(0,x1/j,x2,…,xq).g+k_j=-2\lambda+ \max(0,x_2,\ldots,x_q) +\max(0,x_1/j,x_2,\ldots,x_q). The slope polytope at this vertex is the Minkowski sum

Pj=conv⁡(0,e2,…,eq)+conv⁡(0,e1/j,e2,…,eq).P_j=\operatorname{conv}(0,e_2,\ldots,e_q) +\operatorname{conv}(0,e_1/j,e_2,\ldots,e_q).

For q=1q=1 it is the interval [0,1/j][0,1/j]. In general its slice at ξ1=t\xi_1=t, 0≤t≤1/j0\leq t\leq1/j, is ξ2,…,ξq≥0,ξ2+⋯+ξq≤2−jt.\xi_2,\ldots,\xi_q\geq0,\qquad \xi_2+\cdots+\xi_q\leq2-jt. Indeed a point of the second simplex with first coordinate tt has remaining coordinate sum at most 1−jt1-jt, and the first simplex contributes at most one; conversely any nonnegative vector with the indicated total can be split between these two bounds. Thus vol⁡q(Pj)=∫01/j(2−jt)q−1(q−1)! dt=2q−1jq!.\operatorname{vol}_q(P_j) =\int_0^{1/j}\frac{(2-jt)^{q-1}}{(q-1)!}\,dt =\frac{2^q-1}{j q!}.

Here the use of the original torus normalization can be checked without an unproved delta pullback. Smooth each maximum by its log-sum-exp regularization. The sum of the two regularizations is the log-sum-exp for all sums of their slope vectors. Its Hessian is positive definite, since the slopes affinely span Rq\mathbf R^q. Its gradient maps Rq\mathbf R^q diffeomorphically onto int⁡Pj\operatorname{int}P_j: for an interior target ξ\xi, the regularization minus ξ⋅x\xi\cdot x is strictly convex and coercive, and hence has a unique critical point. The change of variables formula integrates its Hessian determinant to vol⁡q(Pj)\operatorname{vol}_q(P_j). The original CLD top wedge on the coordinate tropical chart has the factor q!q! and multiplicity one. Rescaling the regularization parameter concentrates the mass at the displayed Gauss point. The smooth local computation, the original tropical integration formula under refinements, and bounded BT continuity identify the limit on every original compact smooth test. Its mass is therefore (2q−1)/j(2^q-1)/j.

For completeness there is no other top-degree mass. Away from the ties among s1s_1, s1/js_1/j, and hh, both terms are affine, or one is affine and the other depends on fewer than qq directions, or both depend only on s2,…,sqs_2,\ldots,s_q. At s1=h<s1/js_1=h<s_1/j, the second term is affine, so the only possible top vertex is the radial vertex z1z_1 already counted. At s1<h=s1/js_1<h=s_1/j with h>−λh>-\lambda, the constant floor is inactive. Every active total gradient satisfies jξ1+ξ2+⋯+ξq=2j\xi_1+\xi_2+\cdots+\xi_q=2; their differences have rank at most q−1q-1. If h=−λh=-\lambda but fewer than q−1q-1 of the remaining coordinates equal the floor, fewer than qq directions are active. The only remaining case is ζj\zeta_j. These rank conclusions are also read from the smooth regularizations and the same bounded limit. Near a zero of some tlt_l, its logarithmic branch is inactive; near t1=0t_1=0 both maxima equal hh. Thus there is no additional mass on coordinate zero loci.

The intrinsic result on Z1Z_1 is δz1+(2q−1)j−1δζj\delta_{z_1}+(2^q-1)j^{-1}\delta_{\zeta_j}. Integration-current compatibility and linearity in SS prove (eq:r3-relative-dirichlet-tv-formula) on VV. The extra measure is positive; its total variation equals its total mass, and the normalized constant tester one attains the same psh-dual value. Finally ζj\zeta_j tends to the point of Z1∩Z2Z_1\cap Z_2 with t1=0t_1=0 and the remaining coordinates of radius rr. This proves common compact support, including when q=1q=1, where the limit is the rigid origin. ◻

Corollary (Failure of classical closure in total variation). For this fixed background, top measures of uniformly bounded continuous classical potentials with common ordinary zero boundary values need not form a closed family, even under total-variation convergence on a fixed compact support. The same failure holds for strong psh-dual convergence. The limit measure has a continuous matching subsolution and satisfies every fixed power capacity growth condition in (eq:r3-relative-power-datum), but has no ambient classical zero-boundary solution. Its actual relative completed solution is aa from Proposition “An actual completed solution”.

Proof. Use (eq:r3-relative-tv-and-dual) and Theorem “A bounded matching subsolution does not suffice”. The common bounds and boundary values of the aja_j, their capacity convergence to aa, and the completed equation were proved in the cited proposition. ◻

Constructive gluing on finitely many analytic carriers

This construction gives completed relative potentials with independently prescribed branch operators. The hypotheses explicitly use local outer capacity zero; raw capacity zero is not silently substituted for it.

Proposition (Logarithmic separation of finitely many carriers). Let UU be an allowed ordinary open set and let S1,…,SNS_1,\ldots,S_N be symmetric original-positive separately closed currents of one bidegree (p,p)(p,p). Put S=∑iSiS=\sum_iS_i and q=n−p≥1q=n-p\geq1. Suppose that SiS_i is carried by a closed set AiA_i, and there are analytic functions fif_i on UU, with ∣fi∣≤1|f_i|\leq1, such that Ai⊂Fi:={fi=0}.A_i\subset F_i:=\{f_i=0\}. Assume each log⁡∣fi∣\log|f_i| is an allowed extended classical psh function, and, for i≠li\ne l, assume that FlF_l has locally zero outer SiS_i-capacity on all permitted compact/domain pairs.

Let h1,…,hN,w∈Cb(U)h_1,\ldots,h_N,w\in\mathcal C_b(U) satisfy w≤hiw\leq h_i on AiA_i. Define ψi=∑l≠ilog⁡∣fl∣,aj=max⁡{w,h1+j−1ψ1,…,hN+j−1ψN}.\psi_i=\sum_{l\ne i}\log|f_l|, \qquad a_j=\max\{w,h_1+j^{-1}\psi_1,\ldots, h_N+j^{-1}\psi_N\}. For N=1N=1 the empty sum is zero. Then aja_j is classical, locally uniformly bounded, and increases to an actual Borel function a∈AS,b(U)a\in\mathcal A_{S,b}(U). For every 0≤k≤q0\leq k\leq q,

TS,k(a,…,a)=∑i=1NTSi,k(hi,…,hi)\mathcal T_{S,k}(a,\ldots,a) =\sum_{i=1}^N T_{S_i,k}(h_i,\ldots,h_i)

on all original compact smooth complementary tests. If the hih_i and ww are continuous, each aja_j is continuous as well.

Status: DIRECT_COROLLARY of the proved increasing-limit and completed-calculus theorems, with an explicit gluing construction. It is not a wedge product of two arbitrary currents.

Proof. The functions ψi\psi_i are extended classical psh and nonpositive. Taking their sums with hih_i, and then the finite maximum with the locally bounded floor ww, gives an actual locally bounded classical function. Moreover w≤aj≤max⁡(w,h1,…,hN),w\leq a_j\leq\max(w,h_1,\ldots,h_N), so the bounds are uniform on each compact set, and aja_j increases with jj. The relative increasing-limit theorem gives all-domain local outer-capacity convergence to the actual limit aa, both for SS and separately for each fixed SiS_i. Thus aa lies in every completed class required below.

At a point of Ai∖⋃l≠iFlA_i\setminus\bigcup_{l\ne i}F_l, the penalty ψi\psi_i is finite. Every other branch penalty contains log⁡∣fi∣=−∞\log|f_i|=-\infty and is inactive. Hence a=max⁡(w,hi)=hion Ai∖⋃l≠iFl.a=\max(w,h_i)=h_i \quad\hbox{on }A_i\setminus\bigcup_{l\ne i}F_l. The ordinary open set U∖AiU\setminus A_i has zero local outer SiS_i-capacity: the background and every one of its BT products are zero there by ordinary restriction, and the set is itself an open zero-capacity neighborhood. The assumed outer-null sets FlF_l, l≠il\ne i, therefore account for all possible discrepancies between aa and hih_i in the SiS_i-class. Finite subadditivity gives a=hia=h_i locally outside an outer SiS_i-capacity-zero set. The completed calculus is invariant under exactly these changes, so TSi,k(a,…,a)=TSi,k(hi,…,hi).\mathcal T_{S_i,k}(a,\ldots,a) =T_{S_i,k}(h_i,\ldots,h_i).

For every classical approximant aja_j, the original recursion is linear in its background: TS,k(aj,…,aj)=∑i=1NTSi,k(aj,…,aj).T_{S,k}(a_j,\ldots,a_j) =\sum_{i=1}^N T_{S_i,k}(a_j,\ldots,a_j). Pass to the completed limits along this common increasing sequence and use the preceding identities. This proves (eq:r3-finite-carrier-glued-operator); for k=0k=0 it is the identity S=∑iSiS=\sum_iS_i.

If w,hiw,h_i are continuous, the branches are continuous off their logarithmic zero sets. Near a point where a penalty is −∞-\infty, that branch lies below the continuous finite floor on a sufficiently small neighborhood, because the other summands are locally bounded above. It can therefore be deleted locally. The remaining finite maximum is continuous, proving the last assertion. ◻

Corollary (Completed Dirichlet gluing on coordinate branches). Let VV be the open unit polydisc in (AKn)an(\mathbf A_K^n)^{\mathrm{an}}, and let Z1,…,ZNZ_1,\ldots,Z_N be distinct coordinate subspaces of one dimension q≥1q\geq1 and positive codimension p=n−qp=n-q. Put S=∑i[Zi]S=\sum_i[Z_i], with all ZiZ_i restricted to VV. For each ii, suppose a finite positive Radon measure νi\nu_i on ZiZ_i has a bounded continuous classical solution hih_i with intrinsic ordinary boundary value zero, and suppose (Dhi)q=νi,CiGi≤hi≤0(Dh_i)^q=\nu_i,\qquad C_iG_i\leq h_i\leq0 for some finite Ci≥0C_i\geq0, where, for one fixed 0<r<10<r<1, G=max⁡(log⁡∣T1∣,…,log⁡∣Tn∣,log⁡r),Gi=G∣Zi.G=\max(\log|T_1|,\ldots,\log|T_n|,\log r), \qquad G_i=G|_{Z_i}. Then the relative completed class contains an actual bounded solution with ambient ordinary boundary limit zero of TS,q(a,…,a)=∑i(ii)∗νi.\mathcal T_{S,q}(a,\ldots,a) =\sum_i(i_i)_*\nu_i. It is an increasing limit of uniformly bounded continuous ambient classical zero-boundary potentials. No matching of the intrinsic values of hih_i at intersections is required.

In particular, this conclusion applies when each νi\nu_i has compact support in ZiZ_i and satisfies an intrinsic Dini capacity bound on that coordinate polydisc.

Proof. Let πi:V→Zi\pi_i:V\to Z_i be the coordinate retraction and put Hi=hi∘πiH_i=h_i\circ\pi_i. These are bounded continuous classical psh functions on VV, and Hi≤0H_i\leq0. Choose C≥max⁡(1,C1,…,CN)C\geq\max(1,C_1,\ldots,C_N) and w=CGw=CG. Then w∣Zi≤hi=Hi∣Ziw|_{Z_i}\leq h_i=H_i|_{Z_i}.

For every ii one can choose a linear function fif_i which vanishes on ZiZ_i and is not identically zero on any ZlZ_l with l≠il\ne i. Indeed, the linear forms vanishing on ZiZ_i form a finite-dimensional KK-space. For each distinct same-dimensional coordinate space ZlZ_l, the forms also vanishing there form a proper subspace. An infinite field is not a finite union of such proper subspaces. Scale the chosen form so that its coefficients are integral; then ∣fi∣≤1|f_i|\leq1 on VV.

For i≠li\ne l, the locus {fl=0}∩Zi\{f_l=0\}\cap Z_i is a proper analytic subset of the smooth ZiZ_i. It has zero local outer ambient capacity there. For each compact subset of VV, its intersection with this locus consequently has outer capacity zero with capacity domain ZiZ_i, by finite localization and domain monotonicity. Retraction compatibility Proposition “BT products and capacity on an analytic retract” gives outer [Zi][Z_i]-capacity zero on that compact set with domain VV. The fixed-domain null-class comparison Corollary “Equivalence of local null classes” gives the same local all-domain assertion. Thus the preceding proposition applies with carrier Ai=ZiA_i=Z_i, branch function HiH_i, and floor CGCG. It constructs continuous ambient classical aja_j and an actual completed limit with the required operator, by integration-current compatibility.

Since all penalties are nonpositive, CG≤aj≤0,CG≤a≤0.CG\leq a_j\leq0,\qquad CG\leq a\leq0. The continuous floor CGCG tends to zero at every ordinary boundary point of VV, proving the ambient boundary limits. It also gives a uniform global bound. This argument does not require the projected branch functions HiH_i to vanish on every ambient boundary face.

Finally the intrinsic unit polydiscs have the continuous global classical exhaustions GiG_i. The proved Dini Dirichlet theorem and its continuity improvement supply hih_i, and their compact-support barrier estimate supplies the finite constants CiC_i. This verifies the last assertion using only the already established ambient theory on each intrinsic branch. ◻

Scope.

The construction is finite and its carrier intersections satisfy an explicit relative outer-null condition. It does not identify ambient pluripolar sets with relative null sets for an arbitrary background. It also does not supply an ambient classical representative when branch values are incompatible. The examples of Sections “A relative Dirichlet obstruction in every positive codimension” and “Total-variation limits may require a completed solution” show why the completed target class is necessary even under favorable intrinsic data hypotheses.

Capacity additivity on finitely many analytic carriers

This section gives an all-degree class of backgrounds for which raw relative capacity is outer regular on compact sets. The argument uses separate ambient testers on the individual carriers and glues them into one normalized ambient classical potential. It does not replace an actual relative representative by an ambient usc envelope.

Work over the fixed finite extension K/QpK/\mathbf Q_p, with its original absolute value. Let Ω\Omega be an ordinary open subset of an allowed good, Hausdorff, boundaryless, quasi-smooth, pure nn-dimensional KK-analytic space. Fix 0≤p<n0\leq p<n and q=n−pq=n-p. All backgrounds below are symmetric, original CLD-positive, and separately d′d'- and d′′d''-closed currents of bidegree (p,p)(p,p). All current identities are against the full original compact smooth test sheaf. For such a background RR, write μR,h=TR,q(h,…,h),CR(E)=sup⁡h∈Cb(Ω)0≤h≤1μR,h(E)\mu_{R,h}=T_{R,q}(h,\ldots,h),\qquad C_R(E)=\sup_{\substack{h\in\mathcal C_b(\Omega)\\0\leq h\leq1}} \mu_{R,h}(E) for Borel E⊂ΩE\subset\Omega, and CR∗(A)=inf⁡G⊃AG ordinary open in ΩCR(G)C_R^*(A)=\inf_{\substack{G\supset A\\G\text{ ordinary open in }\Omega}} C_R(G) for arbitrary A⊂ΩA\subset\Omega. These are capacities on the one fixed domain Ω\Omega. Infinite values are permitted unless a relatively compact core is specified.

Theorem (Finite analytic-carrier additivity). Let S1,…,SmS_1,\ldots,S_m be backgrounds as above. Suppose there are analytic functions f1,…,fmf_1,\ldots,f_m on Ω\Omega such that ∣fi∣≤1,Ai={fi=0},|f_i|\leq1,\qquad A_i=\{f_i=0\}, and every coefficient measure of SiS_i is carried by AiA_i. Assume, for all distinct i,li,l, that

CSi(Al)=0.C_{S_i}(A_l)=0.

Put S=∑i=1mSiS=\sum_{i=1}^m S_i. Then every Borel E⊂ΩE\subset\Omega satisfies

CS(E)=∑i=1mCSi(E).C_S(E)=\sum_{i=1}^m C_{S_i}(E).

For every arbitrary subset A⊂ΩA\subset\Omega one also has

CS∗(A)=∑i=1mCSi∗(A).C_S^*(A)=\sum_{i=1}^m C_{S_i}^*(A).

Status: NEW_PROVED. This label applies to the simultaneous testing assertion (eq:r3-carrier-raw-additivity).

The outer identity (eq:r3-carrier-outer-additivity) is a DIRECT_COROLLARY. The carrier condition is a condition on actual coefficient measures, not an identification of an abstract current support with an intrinsic analytic space.

Proof. Linearity of the fixed-background recursion gives, for every normalized hh, μS,h=∑i=1mμSi,h.\mu_{S,h}=\sum_{i=1}^m\mu_{S_i,h}. Hence CS(E)≤∑iCSi(E)C_S(E)\leq\sum_iC_{S_i}(E). We prove the reverse inequality by constructing one tester for any chosen finite tuple of testers.

Fix hi∈Cb(Ω)h_i\in\mathcal C_b(\Omega) with 0≤hi≤10\leq h_i\leq1, and fix 0<δ<1/20<\delta<1/2. Give analytic logarithms their actual value −∞-\infty on their zero sets. Put ψi=∑l≠ilog⁡∣fl∣≤0,bε,δ=max⁡{0, δ+(1−2δ)hi+εψi:1≤i≤m},ε>0.\psi_i=\sum_{l\ne i}\log|f_l|\leq0, \qquad b_{\varepsilon,\delta} =\max\bigl\{0,\, \delta+(1-2\delta)h_i+\varepsilon\psi_i: 1\leq i\leq m\bigr\},\qquad\varepsilon>0. An empty sum is zero. This function belongs to Cb(Ω)\mathcal C_b(\Omega) and satisfies

0≤bε,δ≤1−δ<1.0\leq b_{\varepsilon,\delta}\leq1-\delta<1.

Here membership can be checked entirely with bounded classical functions: replace each log⁡∣fl∣\log|f_l| by max⁡{log⁡∣fl∣,−N}\max\{\log|f_l|,-N\}, take the same finite maximum, and let N→∞N\to\infty. The resulting bounded classical functions decrease to bε,δb_{\varepsilon,\delta} and have the common lower bound zero. The accepted decreasing closure of the actual classical class gives the assertion. Truncated analytic logarithms, positive linear combinations, and finite maxima are classical after every complete extension, so this argument uses the required classical class.

For R>0R>0 set Ai,R=Ai∩⋂l≠i{fl≠0}∩{ψi>−R}.A_{i,R}=A_i\cap\bigcap_{l\ne i}\{f_l\ne0\} \cap\{\psi_i>-R\}. Choose εR<δ/2\varepsilon R<\delta/2 and consider the ordinary open set

Ni=⋂l≠i{fl≠0}∩{ψi>−R}∩{∣fi∣<exp⁡(−1/ε)}.N_i= \bigcap_{l\ne i}\{f_l\ne0\} \cap\{\psi_i>-R\} \cap\{|f_i|<\exp(-1/\varepsilon)\}.

The finite sum ψi\psi_i is continuous on the first intersection, so this is indeed ordinary open, and it contains Ai,RA_{i,R}. On NiN_i the iith branch of the maximum is greater than δ−εR>δ/2\delta-\varepsilon R>\delta/2. If l≠il\ne i, all logarithms are nonpositive and ψl≤log⁡∣fi∣\psi_l\leq\log|f_i|. The llth branch is therefore less than δ+(1−2δ)⋅1−1=−δ.\delta+(1-2\delta)\cdot1-1=-\delta. Consequently, as an equality of actual functions on this ordinary open neighborhood,

bε,δ=δ+(1−2δ)hi+εψion Ni.b_{\varepsilon,\delta} =\delta+(1-2\delta)h_i+\varepsilon\psi_i \quad\text{on }N_i.

The bounds 0≤hi≤10\leq h_i\leq1 are what make this an ordinary-open argument even when the hih_i are discontinuous. No set defined by comparing two discontinuous psh functions is declared open.

Every flf_l, l≠il\ne i, is an analytic unit on NiN_i. Thus ψi\psi_i is original smooth and Dψi=0D\psi_i=0 there. Ordinary locality, positive homogeneity, and the smooth pluriharmonic invariance of Lemma “Smooth pluriharmonic invariance” imply

μSi,bε,δ∣Ni=(1−2δ)qμSi,hi∣Ni.\mu_{S_i,b_{\varepsilon,\delta}}|_{N_i} =(1-2\delta)^q\mu_{S_i,h_i}|_{N_i}.

This uses only unit logarithms and the original smooth product rule over separately closed tail currents. It does not assume potential independence for an arbitrary difference whose ambient current has zero DD.

The measure μSi,hi\mu_{S_i,h_i} is carried by AiA_i: the background vanishes on the ordinary open complement of AiA_i, and restriction and recursion give vanishing of its products there. Moreover (eq:r3-carrier-cross-null) gives μSi,hi(Al)=0(l≠i).\mu_{S_i,h_i}(A_l)=0\quad(l\ne i). The increasing union of Ai,RA_{i,R}, as R→∞R\to\infty, is Ai∖⋃l≠iAlA_i\setminus\bigcup_{l\ne i}A_l, which therefore carries the whole measure. Positivity, background linearity, (eq:r3-carrier-pasted-normalized), and (eq:r3-carrier-local-measure) yield, for every Borel EE, CS(E)≥μS,bε,δ(E)≥(1−2δ)q∑i=1mμSi,hi(E∩Ai,R).\begin{split} C_S(E) &\geq\mu_{S,b_{\varepsilon,\delta}}(E)\\ &\geq(1-2\delta)^q \sum_{i=1}^m\mu_{S_i,h_i}(E\cap A_{i,R}). \end{split} The current identities on the ordinary neighborhoods imply the displayed identities on their Borel subsets by uniqueness of the coefficient-Radon measures. For each RR choose a permitted ε\varepsilon; no convergence theorem for these changing testers is needed. Monotone convergence on the right as R→∞R\to\infty, followed by δ↓0\delta\downarrow0, gives CS(E)≥∑i=1mμSi,hi(E).C_S(E)\geq\sum_{i=1}^m\mu_{S_i,h_i}(E). Taking independent suprema over the finite tuple (h1,…,hm)(h_1,\ldots,h_m) proves (eq:r3-carrier-raw-additivity), including infinite values.

Apply the raw identity to ordinary open sets. Taking the infimum over one common open neighborhood of AA first gives CS∗(A)≥∑iCSi∗(A)C_S^*(A)\geq\sum_iC_{S_i}^*(A). If the right side is finite, choose, for any η>0\eta>0, ordinary opens Gi⊃AG_i\supset A with CSi(Gi)≤CSi∗(A)+η/m.C_{S_i}(G_i)\leq C_{S_i}^*(A)+\eta/m. The finite intersection G=⋂iGiG=\bigcap_iG_i is ordinary open, contains AA, and satisfies CS(G)=∑iCSi(G)≤∑iCSi∗(A)+η.C_S(G)=\sum_iC_{S_i}(G) \leq\sum_iC_{S_i}^*(A)+\eta. Let η↓0\eta\downarrow0. If the right side is infinite the earlier lower bound already proves equality. This proves (eq:r3-carrier-outer-additivity). ◻

Corollary (Transfer of regularity through finitely many carriers). Under Theorem “Finite analytic-carrier additivity”, fix an ordinary core W⋐ΩW\Subset\Omega. If each CSiC_{S_i} is Borel outer regular on WW, the same is true of CSC_S. If each CSi∗C_{S_i}^* on subsets of WW is a finite Choquet capacity, the same is true of CS∗C_S^*. If, in addition, each such capacity has compact inner capacitability on Suslin subsets of WW, so does CS∗C_S^*.

Status: DIRECT_COROLLARY.

Proof. Borel outer regularity follows from the two additive formulas. Finiteness, monotonicity, countable subadditivity, increasing-set continuity, and decreasing-compact continuity are preserved by a finite sum. For compact inner capacitability, approximate each of the finitely many summands from below by a compact Ki⊂AK_i\subset A. The finite union of the KiK_i is a compact subset of AA and simultaneously gives the required lower approximation for the sum. ◻

Corollary (The crossing coordinate supports have regular capacity). In the notation of Theorem “A bounded matching subsolution does not suffice”, let S=[Z1]+[Z2]S=[Z_1]+[Z_2] on the ordinary unit polydisc VV. For every Borel E⊂VE\subset V one has the exact same-domain identity

Cap⁡S(E;V)=Cap⁡[1]Z1(i1−1E;Z1)+Cap⁡[1]Z2(i2−1E;Z2).\operatorname{Cap}_S(E;V) =\operatorname{Cap}_{[1]_{Z_1}}(i_1^{-1}E;Z_1) +\operatorname{Cap}_{[1]_{Z_2}}(i_2^{-1}E;Z_2).

The corresponding equality holds for outer capacities of arbitrary subsets. On every fixed ordinary core W⋐VW\Subset V, the relative capacity is Borel outer regular, and its outer capacity is a finite Choquet capacity with compact inner capacitability on Suslin subsets of WW.

Status: DIRECT_COROLLARY.

Proof. Take f1=T1f_1=T_1 and f2=T2f_2=T_2. The additional equations defining ZiZ_i when p>1p>1 cause no problem: the theorem only asks that [Zi][Z_i] be carried by the corresponding zero set, not that this zero set equal ZiZ_i. The restrictions of f2f_2 to Z1Z_1 and of f1f_1 to Z2Z_2 are nonzero coordinate functions. Their zero sets are proper intrinsic analytic divisors, whose intrinsic ambient capacity is zero. This follows from the proved ambient polar theorem, Corollary “Local outer-capacity zero of ambient pluripolar sets”, on a countable relatively compact exhaustion of these polydiscs; every normalized tester on the full polydisc restricts to each local domain. Thus the full raw intrinsic capacities of the divisors are zero.

Each coordinate projection V→ZiV\to Z_i is an analytic retraction. Proposition “BT products and capacity on an analytic retract” identifies raw and outer relative capacities for [Zi][Z_i] with the respective intrinsic capacities. It verifies (eq:r3-carrier-cross-null) and, together with Theorem “Finite analytic-carrier additivity”, proves (eq:r3-crossing-capacity-sum) and its outer version.

For fixed W⋐VW\Subset V, its inverse image ii−1Wi_i^{-1}W is an ordinary relatively compact core in the intrinsic smooth polydisc ZiZ_i. Apply the ambient exhaustion-domain result Corollary “Local Choquet capacitability and Borel outer regularity” on each such intrinsic core, and then Corollary “Transfer of regularity through finitely many carriers”. Pullback of a decreasing compact family to the closed subspace ZiZ_i is again a decreasing compact family, and inverse image preserves increasing unions. For Suslin sets, intersection with the closed coordinate support remains Suslin; compact inner approximants on the two supports have compact images in WW. These observations verify every stated core property for the ambient relative capacity. ◻

In particular, for this background and any compact H⊂VH\subset V, Cap⁡S(H;V)=Cap⁡S∗(H;V).\operatorname{Cap}_S(H;V)=\operatorname{Cap}_S^*(H;V). If 0≤χ≤10\leq\chi\leq1 is original smooth with compact support and χ=1\chi=1 near HH, every ordinary weak limit ν\nu of the finite measures χ μS,h\chi\,\mu_{S,h}, 0≤h≤10\leq h\leq1, satisfies ν(H)≤Cap⁡S(H;V).\nu(H)\leq\operatorname{Cap}_S(H;V). Indeed, for an ordinary open neighborhood GG of HH contained in {χ=1}∘\{\chi=1\}^{\circ}, the weak open-set inequality gives ν(H)≤ν(G)≤Cap⁡S(G;V)\nu(H)\leq\nu(G)\leq\operatorname{Cap}_S(G;V); take the infimum over such GG and use compact outer regularity. The local CLN estimate gives the required uniform finite mass for this compactly localized family.

Scope.

This is an all-qq positive result for an explicit class of backgrounds carried by ambient polar analytic sets. It coexists with the failure of an ambient classical relative Dirichlet solution in Theorem “A bounded matching subsolution does not suffice” and with the change of measure under ambient usc regularization in Proposition “An actual completed solution”. Thus good raw capacity regularity does not by itself supply an ambient classical representative for every relative completed potential. The proof supplies neither compact outer regularity for an arbitrary fixed SS nor a countable-carrier additivity theorem. Its finite pasting and pairwise relative-capacity-null hypothesis are substantive.

Order approximation of backgrounds and countable carrier sums

All capacities below use the same normalized actual classical testers on the fixed ordinary domain Ω\Omega and the original CLD test sheaf. Fix q=n−p≥1q=n-p\geq1. A positive background means a symmetric, original-positive, separately d′d'- and d′′d''-closed current of bidegree (p,p)(p,p).

Proposition (Uniform capacity control under order approximation). Let SS and SNS_N be positive backgrounds on Ω\Omega such that 0≤SN≤S,SN⟶S0\leq S_N\leq S,\qquad S_N\longrightarrow S as currents against every original compact smooth complementary test. The sequence need not be increasing. Fix a compact set K0K_0 and ordinary opens satisfying K0⊂W,W‾⊂W1⋐Ω.K_0\subset W,\qquad \overline W\subset W_1\Subset\Omega. There are a constant CC and finitely many original positive compact smooth complementary test forms θℓ\theta_\ell, supported in W1W_1, which depend only on these compact stages, the degree, and the chosen finite CLN presentations, such that eN=C∑ℓ⟨S−SN,θℓ⟩⟶0e_N=C\sum_\ell\langle S-S_N,\theta_\ell\rangle\longrightarrow0 and

0≤CS(E)−CSN(E)≤eNfor every Borel E⊂K0,\begin{aligned} 0\leq C_S(E)-C_{S_N}(E)&\leq e_N &&\text{for every Borel }E\subset K_0,\end{aligned}

0≤CS∗(A)−CSN∗(A)≤eNfor every subset A⊂K0.\begin{aligned} 0\leq C_S^*(A)-C_{S_N}^*(A)&\leq e_N &&\text{for every subset }A\subset K_0. \end{aligned}

All capacities in these formulas are finite. Consequently Borel outer regularity on K0K_0 passes from all the CSNC_{S_N} to CSC_S. The properties of being a finite Choquet capacity and of compact inner capacitability on Suslin subsets of K0K_0 likewise pass from the corresponding outer capacities to CS∗C_S^*.

Status: DIRECT_COROLLARY.

Proof. Put RN=S−SNR_N=S-S_N. This is a positive background. The finite-test relative CLN estimate, with the normalized bound 0≤h≤10\leq h\leq1, gives

CRN(W)≤C∑ℓ⟨RN,θℓ⟩=eN.C_{R_N}(W) \leq C\sum_\ell\langle R_N,\theta_\ell\rangle=e_N.

The forms and the constant are fixed before choosing NN or the background: the background occurs in this estimate only through its displayed positive pairings. One obtains the estimate on WW by a finite covering of W‾\overline W and the CLN cutoffs supported in W1W_1. Weak convergence on the original test sheaf proves eN→0e_N\to0. The same estimate for SS proves finiteness on WW.

For every normalized tester hh, background linearity gives the identity of positive Radon measures μS,h=μSN,h+μRN,h.\mu_{S,h}=\mu_{S_N,h}+\mu_{R_N,h}. Consequently every Borel F⊂WF\subset W satisfies CSN(F)≤CS(F)≤CSN(F)+CRN(W)≤CSN(F)+eN.C_{S_N}(F)\leq C_S(F) \leq C_{S_N}(F)+C_{R_N}(W) \leq C_{S_N}(F)+e_N. This proves (eq:r3-order-raw).

For A⊂K0A\subset K_0, the ordinary opens containing AA and contained in WW are cofinal for each outer-capacity infimum: replace an arbitrary open neighborhood GG by G∩WG\cap W. Applying the last inequalities to these ordinary opens and taking infima proves (eq:r3-order-outer). There is no exchange of an infimum with an infinite sum and no infinite intersection of open sets.

Suppose every CSNC_{S_N} is Borel outer regular on K0K_0. For Borel E⊂K0E\subset K_0 the two estimates give CS(E)≤CS∗(E)≤CSN∗(E)+eN=CSN(E)+eN≤CS(E)+eN.C_S(E)\leq C_S^*(E) \leq C_{S_N}^*(E)+e_N =C_{S_N}(E)+e_N\leq C_S(E)+e_N. Let N→∞N\to\infty. This proves Borel outer regularity.

Outer capacities are monotone and countably subadditive. Uniform convergence in (eq:r3-order-outer) preserves increasing-set continuity and decreasing-compact continuity. Explicitly, if Aj↑A⊂K0A_j\uparrow A\subset K_0, then for every NN, CS∗(A)≤CSN∗(A)+eN=lim⁡jCSN∗(Aj)+eN≤lim⁡jCS∗(Aj)+eN;C_S^*(A)\leq C_{S_N}^*(A)+e_N =\lim_j C_{S_N}^*(A_j)+e_N \leq\lim_j C_S^*(A_j)+e_N; letting N→∞N\to\infty proves continuity from below. For decreasing compacts Lj↓L⊂K0L_j\downarrow L\subset K_0, the analogous inequality lim⁡jCS∗(Lj)≤lim⁡jCSN∗(Lj)+eN=CSN∗(L)+eN≤CS∗(L)+eN\lim_j C_S^*(L_j) \leq\lim_j C_{S_N}^*(L_j)+e_N =C_{S_N}^*(L)+e_N\leq C_S^*(L)+e_N proves continuity from above. Finiteness was already proved.

Finally, if A⊂K0A\subset K_0 is Suslin and each approximating outer capacity has compact inner capacitability, then CS∗(A)≤CSN∗(A)+eN=sup⁡L⊂A compactCSN∗(L)+eN≤sup⁡L⊂A compactCS∗(L)+eN.C_S^*(A)\leq C_{S_N}^*(A)+e_N =\sup_{L\subset A\text{ compact}}C_{S_N}^*(L)+e_N \leq\sup_{L\subset A\text{ compact}}C_S^*(L)+e_N. Let N→∞N\to\infty and use monotonicity for the reverse inequality. If compact outer regularity also holds, the final compact terms can be written as raw capacities. This proves all the assertions. ◻

Corollary (Countably many separated analytic carriers). Let (Si)i≥1(S_i)_{i\geq1} be positive backgrounds and suppose the partial sums PN=∑i=1NSiP_N=\sum_{i=1}^N S_i converge weakly, on every original compact test, to a locally coefficient-Radon positive background SS. Assume that there are analytic functions fif_i on Ω\Omega with ∣fi∣≤1|f_i|\leq1 such that SiS_i is carried by {fi=0}\{f_i=0\} and CSi({fl=0};Ω)=0(i≠l).C_{S_i}(\{f_l=0\};\Omega)=0\qquad(i\ne l). On every compact K0⊂ΩK_0\subset\Omega one then has

CS(E;Ω)=∑i≥1CSi(E;Ω)for Borel E⊂K0,CS∗(A;Ω)=∑i≥1CSi∗(A;Ω)for arbitrary A⊂K0.\begin{aligned} C_S(E;\Omega)&=\sum_{i\geq1}C_{S_i}(E;\Omega) &&\text{for Borel }E\subset K_0,\\ C_S^*(A;\Omega)&=\sum_{i\geq1}C_{S_i}^*(A;\Omega) &&\text{for arbitrary }A\subset K_0. \end{aligned}

The series are finite and their tails tend to zero uniformly over the stated subsets of K0K_0. If each component capacity has Borel outer regularity and the stated Choquet properties on such compact cores, the same is true of SS.

Status: DIRECT_COROLLARY.

Proof. Positivity of the series implies 0≤PN≤S0\leq P_N\leq S: for fixed NN, S−PNS-P_N is the weak limit of its positive tails. It is separately closed. Apply Proposition “Uniform capacity control under order approximation” to PNP_N. For every finite NN, Theorem “Finite analytic-carrier additivity” identifies the raw and outer capacities of PNP_N with the corresponding finite sums. The uniform errors of Proposition “Uniform capacity control under order approximation” then give exactly the two asserted series identities, together with uniform convergence of their tails. Finite-component regularity passes first to each PNP_N and then to SS by the proposition. ◻

Scope.

The additional step for a countable family is the uniform CLN tail estimate, not an infinite logarithmic maximum or an infinite intersection of ordinary open neighborhoods. The order-approximation result does not assert that an arbitrary positive closed background admits approximants with already regular capacity. In particular it does not close the remaining unrestricted arbitrary-background compact outer-regularity problem.

What weak compactness does and does not give for outer capacity

The following abstract reduction isolates a remaining relative-capacity gate. Its counterexample concerns a family of ordinary Radon measures; it is not a realization by non-Archimedean BT measures over one fixed background. In particular it does not refute arbitrary-background compact outer regularity.

Let XX be a compact metrizable space and let M\mathcal M be a nonempty family of finite positive Radon measures with sup⁡μ∈Mμ(X)<∞\sup_{\mu\in\mathcal M}\mu(X)<\infty. Put N=M‾ w\mathcal N=\overline{\mathcal M}^{\,w}, a weakly compact family. For Borel EE and arbitrary A⊂XA\subset X, respectively, define c(E)=sup⁡μ∈Mμ(E),co(A)=inf⁡G⊃AG openc(G),c^(A)=sup⁡ν∈Nν∗(A).c(E)=\sup_{\mu\in\mathcal M}\mu(E),\qquad c^{\mathrm o}(A)=\inf_{\substack{G\supset A\\G\text{ open}}}c(G), \qquad \widehat c(A)=\sup_{\nu\in\mathcal N}\nu^*(A). Here ν∗\nu^* is the Radon outer measure, equivalently its open-cover outer regularization. The open outer capacity coc^{\mathrm o} must not be confused with a Borel-envelope extension.

Proposition (The weakly closed measure-family capacity). The function c^\widehat c is finite, monotone, countably subadditive, continuous from below on arbitrary increasing subsets, and continuous from above on decreasing compact sets. Moreover

c^(G)=c(G)for open G,\begin{aligned} \widehat c(G)&=c(G)&&\text{for open }G,\end{aligned}

c^(K)=co(K)for compact K.\begin{aligned} \widehat c(K)&=c^{\mathrm o}(K)&&\text{for compact }K. \end{aligned}

For Borel EE one has c(E)≤c^(E)≤co(E),c^(E)=sup⁡K⊂E compactco(K).c(E)\leq\widehat c(E)\leq c^{\mathrm o}(E),\qquad \widehat c(E)=\sup_{K\subset E\text{ compact}}c^{\mathrm o}(K). The latter compact inner-capacitability identity also holds for Suslin subsets, with c^\widehat c defined above on arbitrary subsets.

Status: DIRECT_COROLLARY of elementary Radon regularity, weak compactness, and the Choquet capacitability theorem. This is not a new analytic estimate for BT products.

Proof. Finiteness, monotonicity, and countable subadditivity follow directly from those of the individual finite Radon outer measures. We spell out continuity from below on arbitrary, possibly nonmeasurable sets. For a fixed finite Radon measure ν\nu and Aj↑AA_j\uparrow A, choose Borel hulls Bj⊃AjB_j\supset A_j with ν(Bj)=ν∗(Aj)\nu(B_j)=\nu^*(A_j), and put Hj=⋂k≥jBkH_j=\bigcap_{k\geq j}B_k. Then Aj⊂HjA_j\subset H_j, the Borel sets HjH_j increase, and ν(Hj)=ν∗(Aj)\nu(H_j)=\nu^*(A_j). Thus ν∗(A)≤ν ⁣(⋃jHj)=lim⁡jν∗(Aj)≤ν∗(A).\nu^*(A)\leq\nu\!\left(\bigcup_jH_j\right) =\lim_j\nu^*(A_j)\leq\nu^*(A). Taking suprema commutes with this increasing supremum, so c^(Aj)↑c^(A)\widehat c(A_j)\uparrow\widehat c(A).

Let Kj↓KK_j\downarrow K be compact. Choose νj∈N\nu_j\in\mathcal N with νj(Kj)≥c^(Kj)−1/j\nu_j(K_j)\geq\widehat c(K_j)-1/j and pass to a weakly convergent subsequence with limit ν∈N\nu\in\mathcal N. For every fixed ll, eventually Kj⊂KlK_j\subset K_l, and the closed-set weak inequality gives lim⁡jc^(Kj)≤lim sup⁡jνj(Kl)≤ν(Kl).\lim_j\widehat c(K_j)\leq\limsup_j\nu_j(K_l)\leq\nu(K_l). Let l→∞l\to\infty. The result is at most ν(K)≤c^(K)\nu(K)\leq\widehat c(K); monotonicity supplies the reverse inequality.

If ν∈N\nu\in\mathcal N, take μj∈M\mu_j\in\mathcal M converging weakly to it. For open GG, the open-set weak inequality gives ν(G)≤lim inf⁡jμj(G)≤c(G)\nu(G)\leq\liminf_j\mu_j(G)\leq c(G). The inclusion M⊂N\mathcal M\subset\mathcal N gives (eq:r3-closure-open). It follows at once that c^(K)≤co(K)\widehat c(K)\leq c^{\mathrm o}(K). For the reverse inequality let Gj={x:dist⁡(x,K)<1/j}G_j=\{x:\operatorname{dist}(x,K)<1/j\} and choose μj∈M\mu_j\in\mathcal M with μj(Gj)≥c(Gj)−1/j\mu_j(G_j)\geq c(G_j)-1/j. After passing to a weak limit ν∈N\nu\in\mathcal N, compare GjG_j with each fixed compact neighborhood Fl={x:dist⁡(x,K)≤1/l}F_l=\{x:\operatorname{dist}(x,K)\leq1/l\}. The same closed-set inequality yields co(K)=lim⁡jc(Gj)≤ν(Fl).c^{\mathrm o}(K)=\lim_jc(G_j)\leq\nu(F_l). Let l→∞l\to\infty to get co(K)≤ν(K)c^{\mathrm o}(K)\leq\nu(K), proving (eq:r3-closure-compact). Empty sets are immediate.

The two inequalities for Borel EE follow from the inclusion of families and from open covers. For Borel inner regularity, only two suprema are interchanged: c^(E)=sup⁡ν∈Nsup⁡K⊂E compactν(K)=sup⁡K⊂E compactc^(K).\widehat c(E) =\sup_{\nu\in\mathcal N}\sup_{K\subset E\text{ compact}}\nu(K) =\sup_{K\subset E\text{ compact}}\widehat c(K). For Suslin sets the established finite Choquet axioms imply the same inner-capacitability formula by the Choquet theorem, in the classical form recalled in [BT82, §8]. Suslin means analytic in descriptive set theory, not an analytic subspace over KK. ◻

Proposition (The exact missing outer-regularity implication). The ordinary-open extension coc^{\mathrm o} is finite, monotone, countably subadditive, outer regular by definition, and continuous from above on decreasing compact sets. The following are equivalent:

  1. coc^{\mathrm o} is continuous from below on all increasing subsets;

  2. co(E)=c^(E)c^{\mathrm o}(E)=\widehat c(E) for every Borel EE.

Under either condition the equality also holds for every Suslin set. In contrast, the sharp compact weak-closure bound ν(K)≤c(K)(ν∈N, K compact)\nu(K)\leq c(K)\quad(\nu\in\mathcal N,\ K\text{ compact}) implies c(E)=c^(E)c(E)=\widehat c(E) for Borel EE, but by itself does not imply c(E)=co(E)c(E)=c^{\mathrm o}(E) for Borel EE.

Proof. The elementary outer-capacity properties follow by taking open covers; for countable subadditivity choose errors summing to a prescribed ε\varepsilon. If Kj↓KK_j\downarrow K are compact, every open neighborhood of KK contains some KjK_j, which proves compact continuity from above.

Under the first condition, coc^{\mathrm o} is itself a finite Choquet capacity. Its compact inner-capacitability on Borel sets, together with (eq:r3-closure-compact), gives co(E)=sup⁡K⊂E compactco(K)=c^(E).c^{\mathrm o}(E)=\sup_{K\subset E\text{ compact}}c^{\mathrm o}(K) =\widehat c(E). Conversely suppose the second condition. For arbitrary Aj↑AA_j\uparrow A, choose Borel sets Bj⊃AjB_j\supset A_j with co(Bj)=co(Aj)c^{\mathrm o}(B_j)=c^{\mathrm o}(A_j). Such hulls are obtained by taking countable intersections of nearly minimizing open covers; they are Borel and are not asserted open. Put Hj=⋂k≥jBkH_j=\bigcap_{k\geq j}B_k and H=⋃jHjH=\bigcup_jH_j. Then Aj⊂HjA_j\subset H_j, co(Hj)=co(Aj)c^{\mathrm o}(H_j)=c^{\mathrm o}(A_j), and A⊂HA\subset H. All Hj,HH_j,H are Borel. Therefore co(A)≤co(H)=c^(H)=lim⁡jc^(Hj)=lim⁡jco(Aj).c^{\mathrm o}(A)\leq c^{\mathrm o}(H)=\widehat c(H) =\lim_j\widehat c(H_j)=\lim_jc^{\mathrm o}(A_j). The reverse inequality is monotonicity. Suslin equality follows from the same Choquet inner-capacitability formulas.

Finally the sharp compact bound and (eq:r3-closure-compact) give c(K)=c^(K)c(K)=\widehat c(K) on compacts. Both raw Borel capacities have compact inner regularity, so they agree on Borel sets. The last assertion is proved false by the following explicit example. ◻

Example (A weakly compact family with a Borel outer-capacity jump). Take X=[0,1]2X=[0,1]^2 and M={μx:x∈[0,1]},μx=δx⊗L1∣[0,1].\mathcal M=\{\mu_x:x\in[0,1]\},\qquad \mu_x=\delta_x\otimes\mathcal L^1|_{[0,1]}. This family is weakly compact, and its raw capacity is c(E)=sup⁡xL1(Ex)c(E)=\sup_x\mathcal L^1(E_x) for Borel EE. Define E=({0}×[0,12]) ∪ ⋃j≥1({1/j}×(12,1]).E=\bigl(\{0\}\times[0,\tfrac12]\bigr) \ \cup\ \bigcup_{j\geq1} \bigl(\{1/j\}\times(\tfrac12,1]\bigr). Then c(E)=c^(E)=12,co(E)=1.c(E)=\widehat c(E)=\tfrac12, \qquad c^{\mathrm o}(E)=1. The set EE is both FσF_\sigma and GδG_\delta. There are compact Kl↑EK_l\uparrow E with co(Kl)=1/2c^{\mathrm o}(K_l)=1/2 for every ll. The same example remains valid after replacing M\mathcal M by its weakly compact convex hull.

Status: FALSE for the proposed abstract implication “weakly compact measure family implies ordinary-open Borel outer regularity.” Its use to close the BT capacity problem is a FAILED_ATTEMPT.

Proof. For every continuous function on the compact square, its fiber integral depends continuously on xx. Thus x↦μxx\mapsto\mu_x is weakly continuous and its image is compact. All nonempty fibers of EE have length 1/21/2, proving the raw and weakly closed values.

Let GG be any relatively open subset of the square containing EE. It contains the compact set {0}×[0,1/2]\{0\}\times[0,1/2], so for some η>0\eta>0 it contains [0,η)×[0,1/2+η)[0,\eta)\times[0,1/2+\eta). For sufficiently large jj, the remainder {1/j}×(1/2,1]\{1/j\}\times(1/2,1] is already contained in EE. Hence the entire fiber at 1/j1/j is contained in GG, and c(G)=1c(G)=1. This proves co(E)=1c^{\mathrm o}(E)=1.

Put Kl=({0}×[0,12])∪⋃1≤j≤l({1/j}×[12+1/(l+2),1]).K_l=\bigl(\{0\}\times[0,\tfrac12]\bigr) \cup\bigcup_{1\leq j\leq l} \bigl(\{1/j\}\times[\tfrac12+1/(l+2),1]\bigr). These are increasing compacts with union EE. Their raw capacities are 1/21/2, and (eq:r3-closure-compact) gives co(Kl)=1/2c^{\mathrm o}(K_l)=1/2. This is the claimed failure of continuity from below, even along an increasing sequence of compact sets.

The displayed union makes EE an FσF_\sigma. Let D={0}∪{1/j:j≥1}D=\{0\}\cup\{1/j:j\geq1\}, a closed subset of [0,1][0,1]. The complement of EE is the union of ([0,1]∖D)×[0,1],{0}×(12,1],⋃j≥1{1/j}×[0,12].([0,1]\setminus D)\times[0,1],\qquad \{0\}\times(\tfrac12,1],\qquad \bigcup_{j\geq1}\{1/j\}\times[0,\tfrac12]. Each is FσF_\sigma, so EE is also GδG_\delta.

The weakly compact convex hull is {τ⊗L1:τ a probability on [0,1]}\{\tau\otimes\mathcal L^1:\tau\text{ a probability on }[0,1]\}. For Borel EE, integration of the fiber lengths against τ\tau has supremum exactly their pointwise supremum, since Dirac probabilities are allowed. Thus the raw capacity on all Borel sets, and hence its ordinary-open extension, is unchanged. ◻

The family need not even give a strongly subadditive capacity. For example, let F1={0,1}×[0,1/2]F_1=\{0,1\}\times[0,1/2] and F2=({0}×[1/2,1])∪({1}×[0,1/2])F_2=(\{0\}\times[1/2,1])\cup(\{1\}\times[0,1/2]). Then c(F1)=c(F2)=1/2c(F_1)=c(F_2)=1/2, whereas c(F1∪F2)=1c(F_1\cup F_2)=1 and c(F1∩F2)=1/2c(F_1\cap F_2)=1/2.

Application and limitation for the present project.

For a compactly localized family of normalized BT measures, the relative CLN estimate supplies the finite mass hypothesis. Weak closure then gives precisely the identities on opens and compacts proved above. It does not identify the resulting inner Choquet capacity with the original ordinary-open outer capacity on Borel sets. The abstract example is not asserted to satisfy any BT maximum or locality structure, and is not a counterexample to an arbitrary-SS theorem. A BT-specific approximation, contact, or separation argument is still required for the unrestricted problem. The analytic-carrier results in this round provide such arguments for a concrete class; they do not rely on the disproved shortcut.

Current theorem correspondence and the advance beyond R2

Section “Theorem-by-theorem comparison with the classical theory” gives the original theorem-by-theorem comparison with BT76/82, Demailly, CLD12/25, GK, GR, and Wu. Section “Second-round theorem and status increment” records the frozen second-round state. This section states the present assessment. An old open entry is closed only with the hypotheses of the specific theorem cited below.

The original finite extension K/QpK/\mathbf Q_p, its absolute value, and all original compact smooth complementary CLD tests are retained. An arbitrary background means a fixed symmetric current in the original positive cone, separately d′d'- and d′′d''-closed. Put q=n−pq=n-p. The all-degree relative energy and top-capacity statements have q≥1q\geq1. They do not construct the wedge of two arbitrary positive closed currents.

What was already complete before this continuation

R and B are INTERNAL_INPUT; their classical regularization, original coefficient-Radon control and background-[1][1] mixed calculus are not discoveries of this round. A2’s geometric adelic nefness theorem is also complete under its stated global hypotheses.

R1 already established the seven requested fixed-background bounded BT properties, including symmetry and weighted joint decreasing continuity. Its finite-test relative CLN estimate is a DIRECT_COROLLARY of R/B. R2 supplied the completed class AS,b\mathcal A_{S,b}, actual increasing limits, polar restriction, fixed-domain capacity comparison, ambient classical representatives, the qualified ambient polar/negligible/capacity-zero equivalence, ambient extremal mass formulas and local Choquet theory. Those results remain frozen dependencies.

The eleven directions after the third round

Direction

Current qualified result

Remaining limitation

1

Locally bounded psh

PROVED: B’s actual classical class and R2’s ambient representative theorem. Proposition “Incompatible branch values in a completed class” gives a nonzero polar background whose completion has no classical representative.

The unrestricted arbitrary-SS representative assertion is FALSE, even on a connected smooth bidisc.

2

Positive closed currents and products

PROVED: the arbitrary-fixed-SS bounded calculus and its completion. New locally finite coefficient-mass zero extension across a closed locally complete pole set is Theorem “A Skoda–El Mir zero-extension theorem”.

The arbitrary-background potential quotient remains OPEN. Normal traces settle smooth algebraic-immersion and specified weak-closure backgrounds, not every positive closed current.

3

CLN estimates

The finite-test mass estimate remains a DIRECT_COROLLARY. Theorem “An all-degree mixed-energy capacity estimate” adds a genuinely relative all-degree energy-to-capacity bound.

Constants retain the finite local controls, common bound and collar hypotheses. No global strictly psh form is assumed.

4

Symmetry and weighted continuity

The requested decreasing theorem was already proved. New results give simultaneous tail-convex-block compactness and strong local measure convergence in the normalized-psh dual distance, including completed limits.

Nonlinear weak limits need not survive convexification. Unrestricted varying-background weighted stability remains FALSE.

5

Locality and maxima

Ordinary locality, Borel maximum locality and ambient maximality are retained. Proposition “Relative maximum and contact inequalities” adds the contact measure inequality as a direct corollary.

Strict gap sets of discontinuous potentials are not automatically ordinary open. Closed polar sets need not satisfy the locally complete hypothesis needed for zero extension.

6

BT capacities

Ambient core Choquet theory is retained. New results treat nonpolar remaining degree one, separated analytic carriers in every degree, and order limits with a uniform CLN tail. See Theorems “Sharp compact regularity for a non-pluripolar degree-one background” and “Finite analytic-carrier additivity”, and Proposition “Uniform capacity control under order approximation”.

For entirely arbitrary SS, sharp compact regularity and the further Borel outer-regularity gate remain OPEN. Abstract weak compactness alone does not supply them.

7

Quasicontinuity and capacity convergence

Bounded relative quasicontinuity and monotone capacity convergence are retained. The common-collar energy, psh-dual distance and relative capacity determine the same bounded topology.

The full normalized psh family is not capacity precompact. The limit class may be AS,b\mathcal A_{S,b} without an ambient classical representative.

8

Pluripolar and negligible sets

The qualified ambient equivalences are retained. Section “A canonical pluripolar decomposition of positive closed currents” gives a canonical polar/nonpolar decomposition with its countable coefficient-control hypothesis.

Relative measures may charge ambient polar analytic sets. A current carried by a pole set need not have topological support contained in that nonclosed set.

9

Comparison and domination

Fixed-SS compact-gap comparison and q.e. domination are retained. Theorem “One-sided support-capacity estimate” gives one-sided ambient stability; power bounds yield a quantitative uniform estimate for strong convergence of data.

Ordinary weak-data convergence is insufficient, even under one uniform power-capacity bound. General relative pointwise domination still requires detection of its capacity ambiguity.

10

Extremal functions

Ambient obstacle orthogonality and capacity/mass formulas are retained. On the prescribed exhaustion domains, weighted divisorial obstacles give a continuous mass map, a finite-data existence proof, and active-set complementarity.

The ambient extremal-contact formula is false for general relative capacity, including specified smooth backgrounds. No arbitrary-SS orthogonality theorem is used in the existence proofs.

11

Dirichlet problems

PROVED: ambient compactly supported data with a bounded subsolution, and compactly supported Dini-capacity data on extendible domains; the latter solutions are continuous. Specified maximal boundary lifts give nonconstant boundary versions. A matching bounded zero-boundary subsolution permits noncompact data.

An arbitrary-relative classical existence theorem from a continuous matching subsolution is FALSE in every p,q≥1p,q\geq1. Completed branch gluing supplies a qualified replacement; general domains, boundary data and completed equations remain open.

Principal new proofs and their dependencies
Relative energy and strong topology.

For bounded potentials with one common smooth classical collar value, or with the proved protected reduction to such a collar, IS(u,v)=∫(u−v)(MA⁡S(v)−MA⁡S(u))≥0.I_S(u,v)=\int (u-v) \bigl(\operatorname{MA}_S(v)-\operatorname{MA}_S(u)\bigr) \geq0. The mixed carré-du-champ identity and repeated Cauchy–Schwarz transfer give Cap⁡S∗({∣u−v∣>δ};Ω)≤Cδ−2IS(u,v)2−q.\operatorname{Cap}_S^*(\{|u-v|>\delta\};\Omega) \leq C\delta^{-2} I_S(u,v)^{2^{-q}}. The midpoint energy gap yields Cauchy tail convex blocks. The normalized-psh dual distance satisfies dS(u,v)≤CIS(u,v)1/2d_S(u,v)\leq C I_S(u,v)^{1/2} and IS(u,v)≤4BdS(u,v)I_S(u,v)\leq4B d_S(u,v). The localized strong-convergence proof retains the cutoff derivative term. For each pair of potentials it first fixes the already constructed mixed background, and only then regularizes. This prevents a circular use of an unproved varying Borel energy multiplier.

Relative capacity on analytic carriers.

For SiS_i carried by {fi=0}\{f_i=0\}, ∣fi∣≤1|f_i|\leq1, and pairwise relative-null cross carriers, normalized testers hih_i are combined by bε,δ=max⁡(0,δ+(1−2δ)hi+ε∑l≠ilog⁡∣fl∣:1≤i≤m).b_{\varepsilon,\delta} =\max\left(0, \delta+(1-2\delta)h_i+\varepsilon \sum_{l\ne i}\log|f_l|:1\leq i\leq m\right). Explicit uniform margins make one branch active on an ordinary open neighborhood of the relevant part of its carrier, even when hih_i is discontinuous. There the logarithms are smooth units with zero curvature. This proves exact raw capacity additivity on all Borel sets and outer additivity on arbitrary subsets. For countable sums, the extra ingredient is the uniform CLN estimate for the positive tail background, not an infinite maximum or an infinite open intersection. These are actual normalized testers on the same domain.

Ambient nonzero Dirichlet existence.

An extendible exhaustion gives a common exterior for the energy. The constrained energy maximum is attained using the midpoint capacity estimate and the ambient representative theorem. Positive obstacle variation gives one measure inequality. The sublevel capacity iteration gives a bound independent of the initially imposed floor. The floor is then inactive on every compact perturbation, and negative variation yields the other inequality. The obstacle orthogonality is the already proved ambient theorem based on B; no arbitrary-relative orthogonality is assumed.

For μ(E)≤C(E)ω(C(E)1/n)n\mu(E)\leq C(E)\omega(C(E)^{1/n})^n, the sufficient scalar condition is ∫0aω(t) dt/t<∞\int_0^a\omega(t)\,dt/t<\infty. Its dyadic estimate gives an explicit bound and one-sided stability. Uniform convergence of B’s regularizing rows is then proved; it is not assumed from the start. Dini’s elementary uniform-convergence lemma is used only on the already continuous ordinary boundary. The radial measure in Theorem “Failure of the universal guarantee outside the Dini range” proves the converse for a universal guarantee on a fixed polydisc. It does not classify each individual measure satisfying a weaker bound.

A necessary change in the relative solution class.

For two intersecting coordinate qq-planes and S=[Z1]+[Z2]S=[Z_1]+[Z_2], the datum δz1+2qδz2\delta_{z_1}+2^q\delta_{z_2} forces the two intrinsic zero-boundary solutions to take incompatible values at their common origin. Yet a continuous matching subsolution and every fixed positive power-capacity bound are available. This proves failure of the proposed ambient classical solution class. The explicit increasing completed solution has continuous approximants with MA⁡S(aj)=μ+2q−1jδζj.\operatorname{MA}_S(a_j) =\mu+\frac{2^q-1}{j}\delta_{\zeta_j}. Thus the classical relative Dirichlet image is not even closed under total variation with one common compact support and potential bound. The completed gluing theorem handles independent intrinsic branch solutions, without requiring incompatible intersection values to agree. The same crossing background has regular capacity by analytic-carrier additivity. Capacity regularity and classical representability are therefore separate issues.

Additional original-source pins

El Mir’s extension theorem is Theorem II.1, printed p. 15, with proof on pp. 15–20; the restriction corollary II.5 is on pp. 21–22. Demailly III.2.3, pp. 139–140, is the corresponding complex template. The CLD12 support/dimension lemma 3.2.5 and CLD25 7.1.12–13 have their particular hypotheses; they are not silently substituted for the arbitrary-original-current extension proved here.[ElMir84, II.1, II.5][Dem12, III.2.3]

Kołodziej 1996, Theorem 1 and Proposition 2, pp. 13–15, give the capacity-sublevel template. Its Theorem 2, p. 16, has additional local-capacity and modulus hypotheses for existence. In the 1998 paper, Lemma 2.3.1 is on pp. 81–83; Theorem B, p. 96, supplies bounds and continuity for an existing solution. Theorem C, p. 97, with proof in §3.4, pp. 103–111, is the bounded-subsolution existence theorem, with the boundary qualification stated on p. 111. The present non-Archimedean existence argument is supplied in full, rather than inferred from the name of either classical theorem. [Kol96, Theorems 1–2][Kol98, Theorems B–C]

Dabbek–Elkhadhra 2006, Definition 2.1 and Proposition 2.2, p. 471, with questions on p. 472, provide a precise relative-capacity template. The follow-up source register is a scoped search, not a claim that those classical questions are globally open today. [DE06, §2]

The projection discussion checks the actual topological-product category in CLD25 §9.1 and its tensor-density results. Mihatsch Theorem 3.23, printed p. 58, gives refined flat skeleton maps; Definition 2.18 gives piecewise smooth fibre integrals, and Example 2.45(3), pp. 41–42, distinguishes balanced pushforwards from δ\delta-forms. These do not establish the original projected-test continuity required by the potential-quotient argument. [CLD25, §9.1–9.4][Mih24, 2.18, 2.45(3), 3.23]

The smallest unresolved gates
Q: potential quotients for arbitrary original backgrounds.

For u,v∈Cb(U)u,v\in\mathcal C_b(U) with D[u]=D[v]D[u]=D[v] on all original tests, prove D((u−v)S)=0D((u-v)S)=0 for every allowed SS. A continuous representative of the bounded classical difference would suffice. Normal traces settle smooth algebraic immersions over the original field. The projection route has an explicit continuous-coefficient sufficient criterion, but no theorem covering every original projected test. An analytic fibre product is not the topological product of its factor spaces, and a same-field graph argument is not the all-extensions curve test.

C1: sharp compact control for a completely arbitrary SS.

For the fixed localized family of normalized relative BT measures, show that every weak-closure measure ν\nu satisfies ν(K)≤Cap⁡S(K;Ω)\nu(K)\leq\operatorname{Cap}_S(K;\Omega) on compact cores. The all-degree convex-block result does not preserve the original nonlinear weak limit. The analytic-carrier and order-approximation theorems do not assert that every positive closed current admits approximants of those types.

C2: Borel outer regularity beyond compact control.

Establish increasing-set continuity of open outer relative capacity, or the equivalent Borel identification with the weak-closure capacity in Proposition “The exact missing outer-regularity implication”. The explicit vertical-fibre example has a convex weakly compact measure family and sharp compact control, but fails this outer-regularity step. It is an abstract counterexample to that shortcut, not an actual non-Archimedean BT counterexample.

D: the appropriate relative completed Dirichlet problem.

The unrestricted ambient classical assertion is false. The replacement must specify a completed solution class, a boundary condition which survives its limits, and a justified uniqueness statement. The finite analytic-branch construction gives a concrete model, not the general existence theorem. For the ambient problem, broader domains and arbitrary continuous boundary data still require proved barriers or an appropriate homogeneous solution.

Comparison with the other calculi.

An equality on all GK δ\delta-tests needs its own comparison theorem. The full nonpositive, nonclosed and off-diagonal GK calculus is not a consequence of the positive BT construction. A2 remains a completed geometric input. Strong Green forms and general heights remain later applications and do not replace any of Q, C1, C2 or D.

The next continuation should work on these specific gates and test each proposed mechanism against the moving-atom, radial-modulus, intersecting-support, vertical-fibre and analytic-product examples. Reproving the accepted bounded calculus or auditing R/B/A2 afresh would not address the remaining obstacles.

Part II. Fourth-round proofs and counterexamples

Compatible piecewise smooth tests for the potential-quotient problem

The full question remains whether D[u]=D[v],u,v∈Cb(U)⟹D((u−v)S)=0D[u]=D[v],\quad u,v\in\mathcal C_b(U) \quad\Longrightarrow\quad D((u-v)S)=0 for every allowed fixed positive separately closed current SS. The results in this section address the projected-test obstruction identified in R3. They enlarge the admissible test class in a quantitative way; they do not assume that all projected tests belong to that enlarged class.

The field is the prescribed finite extension K/QpK/\mathbf Q_p, with its original absolute value. All smooth forms, current equalities and coefficient measures are original CLD objects. Positivity is the original positive semidefinite coefficient cone of R, not an assumption of strong positivity. Let UU be an allowed ordinary open, with a fixed finite moment presentation t=(t1,…,tr)t=(t_1,\ldots,t_r) on a neighborhood of a compact set HH. Each tit_i is the logarithm of the norm of an analytic unit. Enlarge this presentation once to contain every coordinate used by the finitely many forms and cutoffs below, and put β=∑i=1rd′ti∧d′′ti.\beta=\sum_{i=1}^r d't_i\wedge d''t_i. All currents T,TjT,T_j below are symmetric. Let TT have bidegree (p,p)(p,p) and put q=n−pq=n-p.

A uniform estimate near an affine moment wall

Lemma (Normal mass in a shrinking wall tube). Assume q≥1q\geq1. Let ℓ=a0+∑iaiti\ell=a_0+\sum_i a_i t_i be a nonconstant affine expression in the moment variables. There are finitely many original positive compact smooth test forms Θb\Theta_b and a constant CC, fixed independently of TT and 0<ϵ≤10<\epsilon\leq1, such that every original-positive separately closed TT satisfies

(d′ℓ∧d′′ℓ∧βq−1∧T)(H∩{∣ℓ∣≤ϵ})≤Cϵ∑b⟨T,Θb⟩.\bigl(d'\ell\wedge d''\ell\wedge\beta^{q-1}\wedge T\bigr) (H\cap\{|\ell|\leq\epsilon\}) \leq C\epsilon\sum_b\langle T,\Theta_b\rangle.

In particular the displayed positive measure gives zero mass to H∩{ℓ=0}H\cap\{\ell=0\}.

Status: NEW_PROVED. The inputs are the original coefficient-Radon domination and the smooth product rule over a separately closed current.

Proof. Choose a nonnegative compact original smooth cutoff χ\chi equal to one on a neighborhood of HH. Put gϵ=ℓ2+ϵ2.g_\epsilon=\sqrt{\ell^2+\epsilon^2}. Since Dℓ=0D\ell=0, one has Dgϵ=ϵ2(ℓ2+ϵ2)3/2d′ℓ∧d′′ℓ≥123/2ϵ1{∣ℓ∣≤ϵ}d′ℓ∧d′′ℓ.Dg_\epsilon =\frac{\epsilon^2}{(\ell^2+\epsilon^2)^{3/2}} d'\ell\wedge d''\ell \geq\frac1{2^{3/2}\epsilon} \mathbf1_{\{|\ell|\leq\epsilon\}} d'\ell\wedge d''\ell. The inequality is an inequality of positive coefficient measures after wedging with T∧βq−1T\wedge\beta^{q-1}. Separate closedness and compact Stokes give ⟨Dgϵ∧T,χβq−1⟩=⟨T,gϵDχ∧βq−1⟩.\langle Dg_\epsilon\wedge T,\chi\beta^{q-1}\rangle =\langle T,g_\epsilon D\chi\wedge\beta^{q-1}\rangle. The left side is nonnegative. The functions gϵg_\epsilon have one finite bound on supp⁡χ\operatorname{supp}\chi for 0<ϵ≤10<\epsilon\leq1. Apply the fixed finite positive controls to the one smooth form Dχ∧βq−1D\chi\wedge\beta^{q-1}; Borel coefficient multiplication bounds the right side by C∑b⟨T,Θb⟩C\sum_b\langle T,\Theta_b\rangle. Multiplication by 23/2ϵ2^{3/2}\epsilon proves the estimate. Decreasing the tube to the wall proves the final assertion by continuity of the fixed locally finite measure. Constant affine expressions require no estimate: their normal covector is zero. ◻

Lemma (Variation of a normal-factor coefficient). Under the preceding assumptions, let ξ\xi and ζ\zeta be forms of bidegrees (q−1,q)(q-1,q) and (q,q−1)(q,q-1), respectively, whose Borel coefficients on the fixed presentation have a fixed bound on HH. For the finite coefficient measures defined by that presentation,

∣μT,d′ℓ∧ξ∣(H∩{∣ℓ∣≤ϵ})+∣μT,d′′ℓ∧ζ∣(H∩{∣ℓ∣≤ϵ})≤Cϵ1/2∑b⟨T,Θb⟩.\begin{aligned} &|\mu_{T,d'\ell\wedge\xi}|(H\cap\{|\ell|\leq\epsilon\}) +|\mu_{T,d''\ell\wedge\zeta}|(H\cap\{|\ell|\leq\epsilon\})\\ &\hspace{25mm}\leq C\epsilon^{1/2} \sum_b\langle T,\Theta_b\rangle. \end{aligned}

The controls can be chosen for any prescribed finite list of such bounded-coefficient forms. Their coefficient measures vanish on the wall. All constants remain valid on common refinements of this fixed presentation.

Proof. This is the finite-matrix Cauchy–Schwarz argument, now with a Borel indicator of the tube. A coefficient of d′ℓ∧ξd'\ell\wedge\xi is a symmetric cross tensor between a qq-covector containing d′ℓd'\ell and one of the fixed coordinate qq-covectors. The original positive cone contains every positive semidefinite square, including nondecomposable ones. Matrix Young inequality, summed over the fixed finite index sets, therefore gives ∣μT,d′ℓ∧ξ∣(E)≤C((d′ℓ∧d′′ℓ∧βq−1∧T)(E))1/2((βq∧T)(E))1/2|\mu_{T,d'\ell\wedge\xi}|(E) \leq C \bigl((d'\ell\wedge d''\ell\wedge\beta^{q-1}\wedge T)(E)\bigr)^{1/2} \bigl((\beta^q\wedge T)(E)\bigr)^{1/2} for Borel E⊂HE\subset H. To obtain total variation, multiply the coefficients also by an arbitrary bounded Borel scalar of norm at most one before the same matrix inequality and take the supremum. Antisymmetric components are annihilated by the original-positive convention. The identical argument with the primed factors interchanged treats the other term. The first factor is O(ϵ1/2)O(\epsilon^{1/2}) by the preceding lemma, and the second is uniformly controlled by the fixed positive trace. Absorb both controls into their finite sum. The matrix inequalities pull back to every common refinement, so no new constant is needed for additional scalar test coordinates. At ϵ=0\epsilon=0 use decreasing continuity of the variation measures. ◻

Piecewise smooth tests with tangential matching

Definition (The finite original piecewise test class). On an open subset of Rr\mathbf R^r, fix a finite affine hyperplane arrangement. A compatible piecewise smooth (q,q)(q,q)-form is a collection of smooth forms on the closed chambers, smooth on neighborhoods of those chambers, whose restrictions to the tangent space of each common face agree. Only finitely many chambers meet the compact stages considered. Pull it back by the original unit moment map tt and multiply by a fixed original compact smooth scalar cutoff. Finite sums of such expressions are called finite original piecewise tests here.

On a face, choose the ambient tensor supplied by any incident chamber when representing such a test by Borel coefficients. The next theorem shows that its pairing with a positive separately closed current is independent of that choice. This definition does not include an arbitrary piecewise form on an abstract skeleton that has not been represented by an original moment map.

Theorem (Uniform approximation of compatible piecewise tests). Let η\eta be a finite original piecewise test of bidegree (q,q)(q,q). There are original compact smooth forms ηϵ\eta_\epsilon, on one fixed compact support stage, and finitely many fixed positive controls Θb\Theta_b such that for every original-positive separately closed (p,p)(p,p)-current TT,

∣⟨T,ηϵ⟩−⟨T,η⟩∣≤Cϵ1/2∑b⟨T,Θb⟩(0<ϵ≤ϵ0).|\langle T,\eta_\epsilon\rangle-\langle T,\eta\rangle| \leq C\epsilon^{1/2}\sum_b\langle T,\Theta_b\rangle \qquad(0<\epsilon\leq\epsilon_0).

The Borel-coefficient pairing on the right is independent of the chamber representatives on every face. Consequently, if Tj→TT_j\to T weakly on all original compact smooth tests and all these currents are original-positive and separately closed, then ⟨Tj,η⟩⟶⟨T,η⟩.\langle T_j,\eta\rangle\longrightarrow\langle T,\eta\rangle. For q=0q=0, the conclusion instead follows from uniform approximation of continuous piecewise smooth scalar functions; no wall-normal estimate for top-degree measures is asserted.

Status: NEW_PROVED.

Proof. It suffices to work in one fixed presentation with one compact scalar cutoff. Enlarge the compact image stage slightly so that every chamber coefficient and its first derivatives are bounded there. Choose a standard smooth compactly supported convolution kernel of radius ϵ\epsilon in Rr\mathbf R^r, convolve the Borel coefficient tensors, and then pull back and multiply by the fixed cutoff. This gives the forms ηϵ\eta_\epsilon with fixed compact support.

We give the error estimate explicitly. If there is just one wall ℓ=0\ell=0, choose a linear coordinate system with first coordinate ℓ\ell. Tangential matching and the ordinary Hadamard lemma give for the difference of the two smooth branches η+−η−=ℓθ+d′ℓ∧ξ+d′′ℓ∧ζ,\eta_+-\eta_- =\ell\theta+d'\ell\wedge\xi+d''\ell\wedge\zeta, with bounded smooth coefficients on the compact stage. The tangential part therefore has size O(ϵ)O(\epsilon) inside the tube, while the other two parts are controlled by Lemma “Variation of a normal-factor coefficient”.

For finitely many walls the same decomposition of the error follows without imposing a normal-crossing condition. Consider two points x,yx,y with ∣x−y∣≤ϵ|x-y|\leq\epsilon and the line segment joining them. For a generic segment, it crosses each affine wall at most once and avoids intersections of distinct walls. Along each chamber segment the change of the coefficients is bounded by a constant times that segment’s length. Their total is O(ϵ)O(\epsilon). At a crossing of the wall ℓb=0\ell_b=0, the jump has zero restriction to its tangent hyperplane. Linear algebra therefore writes that jump as d′ℓb∧ξb+d′′ℓb∧ζbd'\ell_b\wedge\xi_b+d''\ell_b\wedge\zeta_b, with uniformly bounded coefficients. A fixed linear right inverse on this finite-dimensional normal ideal supplies these bounds. A crossing can occur only when ∣ℓb(x)∣≤Cbϵ|\ell_b(x)|\leq C_b\epsilon.

If xx lies on walls, first move it an arbitrarily small distance into its chosen incident chamber and take the limit. The same finite bound holds; any initial changes of incident chamber are normal-ideal terms for walls through xx. Degenerate crossing segments can likewise be obtained by limits of generic ones. All choices may be made by the finite chamber ordering, so the resulting bounded coefficients are Borel. Averaging over yy in the convolution gives the pointwise coefficient decomposition ηϵ−η=ϵθϵ+∑b(d′ℓb∧ξb,ϵ+d′′ℓb∧ζb,ϵ),\eta_\epsilon-\eta =\epsilon\theta_\epsilon+ \sum_b\bigl(d'\ell_b\wedge\xi_{b,\epsilon} +d''\ell_b\wedge\zeta_{b,\epsilon}\bigr), where the coefficients have one bound and each normal-factor term is supported in ∣ℓb∣≤Cbϵ|\ell_b|\leq C_b\epsilon. The fixed compact cutoff is included in these coefficients. Finite coefficient domination controls the first term by O(ϵ)O(\epsilon), and Lemma “Variation of a normal-factor coefficient” controls the remaining finite sum by O(ϵ1/2)O(\epsilon^{1/2}). This proves (eq:r4-piecewise-uniform-pairing).

The difference between any two ambient representatives on a face lies in the ideal generated by the normals of the walls containing that face. The last assertion of Lemma “Variation of a normal-factor coefficient” makes its pairing zero, proving independence there. In degree zero, face matching is ordinary continuity and convolution converges uniformly, so R’s coefficient estimate suffices.

Finally weak convergence makes every finite positive control ⟨Tj,Θb⟩\langle T_j,\Theta_b\rangle bounded. Fix ϵ\epsilon, pass to the weak limit with the original smooth ηϵ\eta_\epsilon, and then let ϵ↓0\epsilon\downarrow0 in the two uniform error bounds. This proves the stated convergence. A finite localization proves the theorem for every test in the defined class. Neither a strong-positivity assumption nor a change of original test sheaf was used. ◻

The resulting improvement of the projection gate

Proposition (Piecewise projected tests annihilate the kernel). Let π:U×KDs→U\pi:U\times_K\mathbb D_s\to U, dim⁡U=n≥1\dim U=n\geq1, and let u,v∈Cb(U)u,v\in\mathcal C_b(U) satisfy D[u]=D[v]D[u]=D[v]. For α∈Acn,n(U×KDs)\alpha\in A_c^{n,n}(U\times_K\mathbb D_s), put Pα=π∗[α]P_\alpha=\pi_*[\alpha]. Suppose, locally on a finite compact cover, that PαP_\alpha is the ordinary current of a finite original piecewise (n−1,n−1)(n-1,n-1)-test as in Definition “The finite original piecewise test class”. Then ⟨D[(u−v)∘π],α⟩=0.\langle D[(u-v)\circ\pi],\alpha\rangle=0.

Status: DIRECT_COROLLARY of the new test-continuity theorem and the fixed-measure reduction in R3. The universal assertion for all projected tests is still CONDITIONAL on the stated geometric representation.

Proof. The fixed signed measure να=DPα=π∗[Dα]\nu_\alpha=DP_\alpha=\pi_*[D\alpha] is finite. Choose simultaneous locally bounded smooth decreasing psh rows uj↓uu_j\downarrow u, vj↓vv_j\downarrow v on each compact stage. Their positive separately closed currents Duj,DvjDu_j,Dv_j converge to the same current. Theorem “Uniform approximation of compatible piecewise tests” therefore gives ⟨Pα,D(uj−vj)⟩⟶0.\langle P_\alpha,D(u_j-v_j)\rangle\longrightarrow0. Smooth Stokes identifies this with ∫(uj−vj) dνα\int(u_j-v_j)\,d\nu_\alpha. Bounded convergence against this one fixed signed measure gives ∫(u−v) dνα=0\int(u-v)\,d\nu_\alpha=0. The original projection identity then gives the assertion. Finite original smooth partitions handle the stated finite local cover. No pullback of an arbitrary original current has been presumed. ◻

Example (Tangential matching cannot be omitted). On the original two-dimensional torus take an ordinary product annulus with coordinates x=log⁡∣T1∣x=\log|T_1|, y=log⁡∣T2∣y=\log|T_2| in (−2,2)(-2,2). Put cj=(−1)j/jc_j=(-1)^j/j and aj=max⁡(x,cj),Tj=D[aj],T=D[max⁡(x,0)].a_j=\max(x,c_j),\qquad T_j=D[a_j],\qquad T=D[\max(x,0)]. The functions are continuous classical psh and converge uniformly, so Tj→TT_j\to T as positive separately closed original currents. Choose nonnegative smooth compact scalar functions ρ,σ\rho,\sigma in (−1,1)(-1,1) with ρ(0)>0\rho(0)>0 and ∫σ(y) dy>0\int\sigma(y)\,dy>0. The Borel test η=1{x>0}ρ(x)σ(y) d′y∧d′′y\eta=\mathbf1_{\{x>0\}}\rho(x)\sigma(y)\,d'y\wedge d''y has a tangential jump at x=0x=0. The original torus calibration gives ⟨Tj,η⟩=1{cj>0}ρ(cj)∫σ(y) dy.\langle T_j,\eta\rangle =\mathbf1_{\{c_j>0\}}\rho(c_j)\int\sigma(y)\,dy. This pairing has two distinct subsequential limits. Thus arbitrary bounded piecewise coefficient tests do not satisfy the theorem. Every point and current is over the original KK; real radii define Berkovich Gauss points and do not require a field extension. This is a counterexample to the enlarged-test shortcut, not to the potential-quotient conjecture.

What remains unproved.

Mihatsch’s δ\delta-Forms on Lubin–Tate spaces, arXiv:2112.10018v4, Section 2.2 defines compatible piecewise smooth forms, and Definition 2.18 and Theorem 3.23 provide piecewise smooth fiber integration on suitably refined flat skeletons. Example 2.45(3) explicitly warns that a flat pushforward need not descend to the specified smaller sheaf of forms. The theorem above handles original moment-piecewise tests, but does not prove that the skeletonwise pushforward PαP_\alpha has such a common original representation. Establishing that descent, with the original calibration and compatibility under further refinements, is the remaining geometric gate for this route.

Even after that projection gate, the already proved normal-trace argument applies only to graphs lying in its smooth algebraic closed-immersion scope over the original field. It does not automatically supply arbitrary analytic graph restrictions after every complete field extension. Nor has a continuous representative of every bounded classical difference in the original DD-kernel been constructed here. The unrestricted arbitrary-background potential quotient is therefore still OPEN.

A projection class with an explicit original piecewise formula

This section verifies a concrete subclass of the geometric hypothesis in Proposition “Piecewise projected tests annihilate the kernel”. The relative coordinate functions are split over the original base. No finite cover, scalar extension, or altered normalization is used.

Let UU be an allowed smooth nn-dimensional ordinary algebraic analytic neighborhood over the fixed KK, with n≥1n\geq1. Let a1,…,ama_1,\ldots,a_m, m≥1m\geq1, be analytic functions on UU, such that ai−aja_i-a_j is an analytic unit whenever i≠ji\ne j. Let t=(t1,…,tr)t=(t_1,\ldots,t_r) be an original unit moment map on UU, enlarged to include bij=log⁡∣ai−aj∣(i≠j).b_{ij}=\log|a_i-a_j|\quad(i\ne j). Thus each bijb_{ij} is a coordinate, or a fixed affine expression in these coordinates. Put bii=−∞b_{ii}=-\infty as a notation, not as a smooth coordinate. On the complement of the graphs of the aia_i in U×KA1,anU\times_K\mathbf A^{1,\mathrm{an}}, with vertical coordinate TT, consider the original moment map Φ(x,T)=(t(x),log⁡∣T−a1(x)∣,…,log⁡∣T−am(x)∣).\Phi(x,T)=\bigl(t(x),\log|T-a_1(x)|,\ldots, \log|T-a_m(x)|\bigr).

A split-moment test is a finite sum of forms

α=(χ∘π)Φ∗ω,χ∈Ac0,0(U),ω smooth of bidegree (n,n),\alpha=(\chi\circ\pi)\Phi^*\omega, \qquad \chi\in A_c^{0,0}(U), \qquad\omega\text{ smooth of bidegree }(n,n),

where the coefficient support of ω\omega is contained in one fixed bounded box in the last mm coordinates. One may also use a compact cutoff in the first rr real coordinates. All coefficients are defined on a real open neighborhood of the relevant compact stages and extended with such cutoffs. The form is extended by zero across the graphs. This extension is original smooth: near each graph its own logarithmic coordinate leaves the lower end of the support box. The base support is compact, the functions aia_i are bounded there, and the vertical norm is bounded on the support. Thus α\alpha has compact support in the actual analytic product.

Theorem (The split-moment projection formula). For every split-moment test α\alpha, the ordinary current π∗[α]\pi_*[\alpha] is represented by a finite original piecewise smooth (n−1,n−1)(n-1,n-1)-test in the sense of Definition “The finite original piecewise test class”. The representation has no additional current supported on base walls. It is compatible with arbitrary further original moment refinements on the base. Consequently, if u,v∈Cb(U)u,v\in\mathcal C_b(U) and D[u]=D[v]D[u]=D[v], then ⟨D[(u−v)∘π],α⟩=0\langle D[(u-v)\circ\pi],\alpha\rangle=0 for every split-moment test. The same assertion applies to a split-moment test supported in U×KDsU\times_K\mathbb D_s and then viewed on that ordinary disc product.

Status: NEW_PROVED for the explicit finite-original presentation. The resulting kernel equality is a DIRECT_COROLLARY. The weighted description of a split line-fiber skeleton used below is an EXTERNAL_KNOWN input, also verified directly by the Gauss norm and product calibration in the proof.

Proof. It suffices to treat one summand of (eq:r4-split-test). At a base point xx, write σi,s(x)\sigma_{i,s}(x) for the image of the radius-ese^s Gauss point centred at ai(x)a_i(x) in the H(x)\mathscr H(x)-line. The original point lies in the analytic fiber over xx. The Gauss norm gives the exact identities

log⁡∣T−ak∣(σi,s(x))=max⁡{s,bik(x)},σi,s(x)=σj,s(x)⟺s≥bij(x).\log|T-a_k|(\sigma_{i,s}(x)) =\max\{s,b_{ik}(x)\},\qquad \sigma_{i,s}(x)=\sigma_{j,s}(x) \Longleftrightarrow s\geq b_{ij}(x).

For k=ik=i the maximum is simply ss.

The part of the line-fiber skeleton seen by the functions T−aiT-a_i is the union of these rays. A point off their union lies in a complementary disc on which all their logarithmic norms are constant, so it contributes no vertical differential direction. Count each common ray only once by assigning it to its first centre. With L1=+∞,Li=min⁡j<ibij(i≥2),L_1=+\infty,\qquad L_i=\min_{j<i}b_{ij}\quad(i\geq2), the iith assigned ray has parameter range s<Li(x)s<L_i(x). Indeed it has not yet met any earlier ray exactly when s<bij(x)s<b_{ij}(x) for every j<ij<i. Endpoints have one-dimensional measure zero. Each noncollapsed ray has the standard normal weight dsds, with no residue-degree factor: its centre is H(x)\mathscr H(x)-rational. Any structural degree of a base skeleton cell is already in the base calibration.

Here is a formula in one finite real moment space. Define the continuous piecewise affine map Fi(t,s)=(t,max⁡(s,bi1(t)),…,max⁡(s,bim(t))).F_i(t,s)= \bigl(t,\max(s,b_{i1}(t)),\ldots, \max(s,b_{im}(t))\bigr). Let Is\mathcal I_s denote contraction with the normal top coefficient in the product calibration: for a form of bidegree (n,n)(n,n) it extracts its d′s∧d′′sd's\wedge d''s coefficient as a form of bidegree (n−1,n−1)(n-1,n-1) in the base variables, with the usual product-integration sign convention. Put

η(t)=∑i=1m∫−∞Li(t)Is(Fi∗ω)(t,s) ds.\eta(t)=\sum_{i=1}^m \int_{-\infty}^{L_i(t)}\mathcal I_s(F_i^*\omega)(t,s)\,ds.

For i=1i=1 the upper endpoint is +∞+\infty. This is a finite ordinary integral: the iith vertical coordinate of FiF_i is ss, so the bounded support of ω\omega in that coordinate supplies one fixed compact ss-interval for every term, including real tt in a neighborhood of the compact base moment image. In particular the formula is defined in that full real neighborhood, not just on its tropical image.

We verify that η\eta is a compatible piecewise smooth form there. Subdivide by the finitely many hyperplanes ordering the affine functions bijb_{ij} and the fixed vertical support endpoints. On each chamber, LiL_i is one fixed affine function, and the breakpoints of the maxima in FiF_i have a fixed order. Splitting the integral at these breakpoints expresses every coefficient as a finite sum of integrals of smooth functions with affine endpoints. The coefficients are therefore smooth up to that closed chamber, with smooth extensions obtained from the same signed-integral formulas. They and their first derivatives are bounded on the chosen compact stage.

For matching on a common base face, fix tangent vectors to that face. Except for the finitely many limiting values of ss at a breakpoint, the maps FiF_i and their derivatives in those tangent vectors have the same limiting values from the adjacent chambers. This follows directly from continuity of the maximum: where two affine expressions agree on the face, their tangential derivatives agree there. The same is true of the endpoint LiL_i. The integrands are uniformly bounded on the fixed compact ss-interval. Dominated convergence of these tangential coefficients therefore proves that the two restrictions of η\eta to the face agree. When an interval collapses its integral tends to zero; no term is retained on that lower dimensional base face. Repeating the argument on intersections of faces proves the full compatibility required in Definition “The finite original piecewise test class”.

It remains to prove the ordinary current identity, including its test scope. Let γ\gamma be any original compact smooth (1,1)(1,1)-test on UU, and refine the base moment presentation to include all its coordinates and a finite partition. On each calibrated full-dimensional base cell, adjoining T−aiT-a_i gives the above Gauss rays with the product calibration. The original integration formula and Borel Fubini on these finite cell presentations give ∫U×A1α∧π∗γ=∫Uχ t∗η∧γ.\int_{U\times\mathbf A^1}\alpha\wedge\pi^*\gamma =\int_U\chi\,t^*\eta\wedge\gamma. One can equivalently read this product description in Mihatsch, arXiv:2112.10018v4, proof of Proposition 3.21, equations (3.27)–(3.31) and its final weight verification. In the present split case the centres and all pairwise distances are already defined on the original base, so none of the preliminary field or finite-cover reductions in that general proposition is used. Under a refinement, the formulas (eq:r4-split-Gauss-coordinates) are unchanged and only the base calibrated cells subdivide. Their weights sum to the original weight. This verifies the displayed identity for all original tests, not only for tests in the first chosen chart.

The identity is exactly π∗[α]=[χt∗η]\pi_*[\alpha]=[\chi t^*\eta]. It also excludes any extra ordinary current on a collapsed base cell: the identity has already been checked on the entire original complementary test sheaf. Geometrically, a full-dimensional source cell here has one relative direction and full base dimension; a cell over a base face has total dimension at most that face’s dimension plus one and contributes no additional top integral.

The projected test is thus in the finite original piecewise class. Proposition “Piecewise projected tests annihilate the kernel” proves the kernel equality. If the original support is in the disc product, extension by zero into the affine-line product is original smooth and the same argument applies. ◻

Corollary (Split polynomial and rational vertical moments). The theorem also applies when the finitely many vertical units in a test presentation have the form gℓ(x,T)=bℓ(x)∏i=1m(T−ai(x))eℓi,bℓ∈OU×,eℓi∈Z,g_\ell(x,T)=b_\ell(x) \prod_{i=1}^m(T-a_i(x))^{e_{\ell i}}, \qquad b_\ell\in\mathcal O_U^\times,\quad e_{\ell i}\in\mathbf Z, and the compact test admits the indicated split refinement. The centres and their pairwise differences have the same assumptions as in the theorem. Strictly smaller analytic remainders may be discarded only on an ordinary neighborhood where they preserve the absolute values of every chosen unit.

Proof. Adjoin the base units bℓb_\ell and all the linear factors to the moment presentation. Then each log⁡∣gℓ∣\log|g_\ell| is an affine linear combination of the enlarged original logarithms, so the test becomes a split-moment test. For a permitted remainder, ultrametricity gives equality of the two logarithmic functions on the stated ordinary neighborhood, hence equality of their original smooth differentials and of the represented test there. This is a DIRECT_COROLLARY; it does not assert a splitting of arbitrary polynomials on the original base. ◻

The remaining descent problem.

General original units in the vertical variable need not have this split presentation over the original base. A finite flat cover or an extension used to split them cannot simply pull back D[u]=D[v]D[u]=D[v], since precisely that original-current functoriality is unproved. Nor may pairwise collision loci be removed without showing that the projected test extends with the required compatibility there. The next geometric question is therefore descent of the displayed piecewise formula through finite flat base presentations, or an independent original-base formula for nonsplit vertical units. The theorem settles a concrete multiple-centre projection class and does not claim that it generates all original tests on the analytic product.

Finite etale trace in the original piecewise test class

Throughout this section the ground field is the original finite extension K/QpK/\mathbf Q_p, with its original absolute value. A finite etale map is used only to compute the pushforward of forms and their currents. No pullback of a general current is presupposed.

Lemma (Finite tomography of unordered vectors). Fix integers d,m≥1d,m\geq1 and let A={0,…,d}mA=\{0,\ldots,d\}^m. For an unordered list Z=(z1,…,zd)Z=(z_1,\ldots,z_d) of vectors in Rm\mathbf R^m, including repetitions, record the ordered list of a⋅zla\cdot z_l, 1≤l≤d1\leq l\leq d, for every a∈Aa\in A. These records determine ZZ uniquely. The set of consistent records is a finite union of polyhedral sets. On a finite polyhedral subdivision of that set, the vectors can be listed as affine linear functions of the records. If two such listings meet on a face, they agree there as unordered lists.

Proof. The records give the polynomial in XX PZ(X,a)=∏l=1d(X−a⋅zl)P_Z(X,a)=\prod_{l=1}^d(X-a\cdot z_l) for every aa in the grid. Each coefficient has degree at most dd in each coordinate of aa. Successive one-variable polynomial interpolation on {0,…,d}\{0,\ldots,d\} determines PZ(X,a)P_Z(X,a) as a polynomial in X,a1,…,amX,a_1,\ldots,a_m. Unique factorization into its monic linear factors X−a⋅zlX-a\cdot z_l determines the unordered list, with multiplicities.

Here is the finite polyhedral assertion explicitly. The records for the coordinate vectors ei∈Ae_i\in A list the iith coordinates of all zlz_l. Fix their first coordinate in its given order. Each possible joint listing is obtained by permuting the other m−1m-1 coordinate lists; there are only finitely many choices. For each choice and each aa, also choose a permutation matching the projected coordinates with the recorded ordered list for aa. Its validity is a finite system of linear equalities, together with the ordering inequalities on the records. The consistent set is the finite union of these polyhedra. A common subdivision makes one valid listing affine on each cell. The uniqueness just proved gives the last assertion. ◻

Lemma (Extension from a finite polyhedral locus). Let C⊂RNC\subset\mathbf R^N be a closed finite polyhedral set. Suppose a form τ\tau on CC is smooth up to each cell of a finite subdivision and has matching tangential restrictions on common faces. For every compact L⊂CL\subset C, there is a real open neighborhood of LL on which τ\tau extends to a finite compatible piecewise smooth form. On compact stages the extended coefficients and their first derivatives are bounded.

Proof. Choose a large closed polyhedral box containing LL in its interior and triangulate it together with CC and the given subdivision. After barycentric subdivision, the subcomplex contained in CC is full: the vertices of any simplex which belong to that subcomplex span a face of that simplex lying in CC. Let λv\lambda_v be the continuous piecewise affine hat functions of this triangulation and put q(x)=∑v∈Cλv(x),R(x)=∑v∈Cλv(x)vq(x).q(x)=\sum_{v\in C}\lambda_v(x),\qquad R(x)=\frac{\sum_{v\in C}\lambda_v(x)v}{q(x)}. On q>0q>0 the point R(x)R(x) belongs to the face spanned by the CC-vertices of the simplex containing xx. Thus RR maps to CC and is the identity on CC. It is continuous and piecewise smooth. Work in the interior of the box with q>1/2q>1/2, which is an open neighborhood of LL. On each closed cell the denominator is bounded away from zero. Pull back the smooth form on the corresponding CC-face by RR. These pullbacks have smooth extensions near their closed cells and match tangentially, because RR and its tangential derivatives agree on common faces. Their pullbacks agree with τ\tau on CC. A further finite hyperplane subdivision puts the extension in the form of Definition “The finite original piecewise test class”. All asserted bounds follow on compact stages from the finite construction and the lower bound on qq. ◻

Theorem (Original finite etale trace). Let p:Z→Up:Z\to U be a finite etale map of smooth good Hausdorff boundaryless KK-analytic spaces of pure dimension nn. The ordinary pushforward of a compact original smooth form, or of a finite original compatible piecewise smooth form, is locally represented by finite original compatible piecewise smooth forms on UU. Their ordinary currents agree with the pushforward on every original smooth complementary test. The representation is compatible with arbitrary further original moment refinements. In particular it introduces no additional current supported on a base wall.

Status: NEW_PROVED. The finite-flat tropical degree formula used to verify the ordinary-current identity is EXTERNAL_KNOWN. The joint-norm reconstruction and the finite-original descent are proved here.

Proof. All assertions are local on the base. Fix x∈Ux\in U. The fiber p−1(x)p^{-1}(x) is finite. Choose pairwise disjoint ordinary neighborhoods of its points on each of which the given form has one finite moment presentation. Properness of pp permits shrinking the base so that its entire preimage is contained in their union. The resulting finitely many open-and-closed pieces are finite etale over the base. Shrink again so that their ranks are constant. Treat each piece separately and sum. We may therefore assume that pp has constant degree dd and that the given form is ω(log⁡∣h1∣,…,log⁡∣hm∣)\omega(\log|h_1|,\ldots,\log|h_m|) for analytic units h1,…,hmh_1,\ldots,h_m on the whole preimage, with ω\omega a smooth or finite compatible piecewise smooth superform. Finitely many initial moment presentations may be combined by adjoining their units. Work on a relatively compact base neighborhood whose preimage closure lies in this presentation. All hih_i and their inverses are uniformly bounded there.

For each a∈A={0,…,d}ma\in A=\{0,\ldots,d\}^m, form the unit ha=∏ihiaih^a=\prod_i h_i^{a_i} and its characteristic polynomial for the finite locally free algebra p∗OZp_*\mathcal O_Z: Qa(T)=Td+ca,1Td−1+⋯+ca,d.Q_a(T)=T^d+c_{a,1}T^{d-1}+\cdots+c_{a,d}. Its coefficients are analytic on the original base, and its constant coefficient is a unit. At a base point yy, the roots over an algebraic closure of H(y)\mathscr H(y) are the values of hah^a at its dd geometric sheets, counted with multiplicity. Every embedding of a finite extension of the complete field H(y)\mathscr H(y) preserves its absolute value. Thus repeated geometric sheets at one Berkovich fiber point contribute exactly its finite local degree.

We only use these algebraic roots to compute their norms. Gauss norm multiplicativity gives, for every real ss,

max⁡0≤k≤d(log⁡∣ca,k(y)∣+(d−k)s)=∑l=1dmax⁡(s,ra,l(y)),ca,0=1,\max_{0\leq k\leq d} \bigl(\log|c_{a,k}(y)|+(d-k)s\bigr) =\sum_{l=1}^d\max(s,r_{a,l}(y)), \qquad c_{a,0}=1,

where ra,lr_{a,l} are the log norms of these roots. The changes of slope on the right determine their ordered list. Equivalently, this is the elementary Newton polygon formula. Its ordered roots are continuous piecewise affine functions of the finite coefficient log norms: a choice of polygon vertices gives affine slopes, and the finitely many possible choices are specified by affine inequalities.

There is no need to use the logarithm of a coefficient vanishing at xx. Choose B>0B>0 so that every root log norm of every QaQ_a on the chosen neighborhood belongs to [−B,B][-B,B]. The valuations of all lower-hull vertices are then at most dBdB. If ca,k(x)=0c_{a,k}(x)=0, shrink the base so that ∣ca,k∣<exp⁡(−dB−1)|c_{a,k}|<\exp(-dB-1); such a coefficient lies strictly above the lower Newton hull and may be omitted throughout this neighborhood. All remaining nonzero coefficients at xx are analytic units after a further shrinking. The leading and constant coefficients are retained. Consequently finitely many original analytic unit moments t=(log⁡∣ca,k∣)a,k retainedt=(\log|c_{a,k}|)_{a,k\text{ retained}} determine the ordered root records r(t)r(t) by finite continuous piecewise affine formulas on a full real moment space.

Let CC be the subset of that real space where these records are jointly consistent in the sense of Lemma “Finite tomography of unordered vectors”. It is a closed finite polyhedral set after a common subdivision, and it contains the actual base moment image. On each of its cells choose an affine listing z1(t),…,zd(t)z_1(t),\ldots,z_d(t) and put

τC=∑l=1dzl∗ω.\tau_C=\sum_{l=1}^d z_l^*\omega.

This does not require the individual sheets to be analytic sections. It is an expression in their reconstructed norm vectors.

The forms in (eq:r4-unordered-form-trace) match tangentially on every common face. Indeed on that face the two vector lists agree as unordered lists. Subdivide further by their finitely many coordinate-equality hyperplanes. On the relative interior of each resulting face a fixed permutation identifies the affine vector maps. They therefore have identical tangential derivatives on that face. If some vectors coincide identically there, their tangential maps and derivatives coincide as well. Summing the pullbacks proves the assertion. If ω\omega was piecewise smooth, also subdivide by the inverse images of its finitely many walls; its own tangential compatibility gives the same conclusion.

Lemma “Extension from a finite polyhedral locus” extends τC\tau_C near each compact base moment image to a finite compatible piecewise smooth form τ\tau on a full real neighborhood. Its pullback t∗τt^*\tau is therefore a finite original piecewise form. This extension step is what distinguishes the claim from an assertion only about forms on an abstract skeleton.

We verify the current identity, which also makes the choice of extension immaterial. Take any original smooth compact complementary test γ\gamma on the base. Refine its moment presentation to include all the retained coefficients and use a finite partition on the support. The finite-flat tropical degree formula identifies source calibrated full-dimensional cells over every base cell with their finite fiber weights. Those weights are exactly the multiplicities of the geometric sheets used above. The source unit log norms on those cells are the affine vector listings zlz_l; grouping identical listings adds their multiplicities. Thus the original integration formula gives

∫Zω(log⁡∣h∣)∧p∗γ=∫Ut∗τ∧γ.\int_Z \omega(\log|h|)\wedge p^*\gamma =\int_U t^*\tau\wedge\gamma.

For precision, the finite-flat degree input is Mihatsch, arXiv:2112.10018v4, Lemma 3.15; its use after adjoining source units is Proposition 3.19 and its proof, especially equation (3.20). The characteristic-polynomial coefficients ensure that adjoining those units does not add graded transcendence degree. Indeed at any actual root θ\theta with ∣log⁡∣θ∣∣≤B|\log|\theta||\leq B, an omitted coefficient satisfies log⁡∣ca,kθ−k∣<−dB−1+kB≤−1.\log|c_{a,k}\theta^{-k}| <-dB-1+kB\leq-1. Its monomial is strictly smaller than the leading monomial θd\theta^d, hence cannot be a maximal term of the initial characteristic relation. The nontrivial initial relation uses only retained unit coefficients and still proves graded algebraicity. This explicitly covers coefficients vanishing at the original base point. The same argument works after every additional base refinement. The equality concerns calibrated weighted piecewise linear spaces; we do not use an equality of their sheaves of linear functions, which Remark 3.20 explicitly distinguishes.

On every refined base cell the moment map takes values in CC and its tangent space maps to the tangent space of the appropriate face of CC. Consequently the pullback of the extension agrees with τC\tau_C there. This justifies using (eq:r4-unordered-form-trace) in the full original integration formula, regardless of how it was extended off CC. For piecewise forms the same cell integrals apply; face choices are immaterial by the original normal-factor wall estimate. This proves (eq:r4-finite-trace-current-identity) on every original test, and hence the stated ordinary-current identity. It rules out any additional wall-supported current.

Finally cover the compact projected support by finitely many base neighborhoods just constructed and multiply by an original partition of unity. Finite sums preserve the asserted class and its tangential compatibility. This proves the theorem. ◻

Corollary (Finite etale preservation of the bounded potential kernel). In the preceding setting assume n≥1n\geq1, let u,v∈Cb(U)u,v\in\mathcal C_b(U), and suppose D[u]=D[v]D[u]=D[v] on all original tests. Then D[u∘p]=D[v∘p]D[u\circ p]=D[v\circ p] on all original tests of ZZ. This is a statement about the specified bounded potentials, not a definition of pullback for arbitrary currents.

Status: DIRECT_COROLLARY.

Proof. For an original compact test α\alpha of bidegree (n−1,n−1)(n-1,n-1) on ZZ, the theorem represents Pα=p∗[α]P_\alpha=p_*[\alpha] by finite original compatible piecewise tests. The signed Radon measure ν=DPα=p∗D[α]\nu=D P_\alpha=p_*D[\alpha] is fixed. Take local bounded smooth decreasing psh rows uj↓uu_j\downarrow u and vj↓vv_j\downarrow v on neighborhoods of the projected support, using a finite original partition if needed. Their positive closed curvatures converge to the same current. Theorem “Uniform approximation of compatible piecewise tests” gives ⟨Pα,D(uj−vj)⟩⟶0.\langle P_\alpha,D(u_j-v_j)\rangle\longrightarrow0. Compact Stokes identifies the left side with ∫(uj−vj) dν\int(u_j-v_j)\,d\nu. Fixed-measure dominated convergence identifies its limit with ∫(u−v) dν\int(u-v)\,d\nu, which is ⟨D[(u−v)∘p],α⟩\langle D[(u-v)\circ p],\alpha\rangle by the definition of pushforward. This proves the assertion. The compositions are actual classical functions by their curve definition. ◻

What this does and does not settle.

This theorem supplies finite etale descent in the original piecewise category and makes finite etale kernel pullback a proved consequence of that descent. Its proof does not appeal to a pullback of an arbitrary original current. It does not by itself prove the corresponding statement for arbitrary complete extensions, for an analytic map of positive relative dimension, or for arbitrary analytic graph immersions. These remaining scopes must still be proved separately before the general background quotient problem can be declared solved.

Preservation of the bounded potential kernel by disc projection

We retain the original field KK and absolute value. Let YY be a smooth integral affine KK-variety of dimension n≥1n\geq1, let U⊂YanU\subset Y^{\rm an} be an ordinary open set in the scope of B, and let Ds⊂AK1,an\mathbb D_s\subset\mathbf A_K^{1,\rm an} be an ordinary open disc. All tests and currents in this section are original CLD tests and currents. Write π:U×KDs→U\pi:U\times_K\mathbb D_s\to U.

Lemma (Rational moment localization of an original test). An original compact smooth form on an ordinary open subset of (Y×KA1)an(Y\times_K\mathbf A^1)^{\rm an} is a finite sum of forms which have the following description. On a neighborhood of their support their unit moments are rational functions in K(Y)(T)K(Y)(T), and their compact scalar localization is a smooth function of finitely many truncated logarithms of global regular functions in K[Y][T]K[Y][T]. Off the zeros of finitely many nonzero polynomials in that ring, each summand is one global rational moment expression, extended by zero outside its compact localization. All equalities are actual pointwise equalities of norms and of original smooth forms.

Proof. Let LL be the compact support. Around each point choose an original unit presentation and a smaller compact neighborhood inside it. B, Lemma D.1, replaces its finitely many analytic units on a slightly larger compact neighborhood by rational functions with exactly the same pointwise absolute values. The replacement is on the original space and field. Equality of logarithmic functions on the ordinary interior also gives equality of their original smooth differentials.

The ordinary topology of an affine analytification has a basis of finite norm conditions on its global regular functions. Adding upper bounds on fixed affine coordinates makes the closed norm boxes compact. Within the preceding presentation choose nested such boxes and a scalar bump equal to one on the inner box and supported on the outer box. Explicitly it is a finite product of smooth functions of log⁡∣G∣\log|G| for regular GG, constant for sufficiently negative arguments whenever the box includes G=0G=0. The constant value extends smoothly across that zero locus. These are original smooth compact functions on the whole affine analytification.

Choose finitely many such bumps bib_i whose regions of value one cover LL. Put b=∑ibib=\sum_i b_i, and choose a smooth real function RR which equals 1/t1/t for t≥1/2t\geq1/2 and is zero for t≤1/4t\leq1/4. Then χi=biR(b)\chi_i=b_iR(b) have compact support inside their respective presentations and ∑iχi=1\sum_i\chi_i=1 near LL. Their scalar expressions involve only the finitely many regular functions used in the norm boxes. Multiplying the form by these χi\chi_i gives the required finite sum.

Write every rational unit as a quotient of polynomials in K[Y][T]K[Y][T]. Its numerator and denominator are nonvanishing near the compact support of the corresponding χi\chi_i. Extend the smooth real coefficient forms from a neighborhood of that compact moment image by a real cutoff. The expression χiω(log⁡∣r1∣,…,log⁡∣rm∣)\chi_i\omega(\log|r_1|,\ldots,\log|r_m|) then defines the given summand globally on the unit locus of the rational functions, and is zero near its complement because the scalar support is compactly contained in that unit locus. After deleting the zero loci of the regular functions used in the bumps, their truncated logarithms are also ordinary unit moments. This proves the last assertion. Identically zero regular functions contribute only constant scalars and may be omitted from the finite list. ◻

Lemma (A common finite etale splitting away from a base divisor). Let P1,…,PN∈K[Y][T]P_1,\ldots,P_N\in K[Y][T] be nonzero. There is a nonzero Δ∈K[Y]\Delta\in K[Y] and a finite etale surjective map p:Z→Y0=Y∖{Δ=0}p:Z\to Y_0=Y\setminus\{\Delta=0\} of constant positive degree such that on ZZ Pl(x,T)=bl(x)∏i=1m(T−ai(x))eli,bl∈OZ×,eli≥0.P_l(x,T)=b_l(x)\prod_{i=1}^m(T-a_i(x))^{e_{li}}, \qquad b_l\in\mathcal O_Z^\times,\quad e_{li}\geq0. The aia_i are regular and pairwise distinct, with each ai−aja_i-a_j a unit. One may adjoin P0=TP_0=T so that m≥1m\geq1.

Proof. Work over the fraction field K(Y)K(Y). Remove repeated factors from the product of the PlP_l and make the resulting polynomial monic. Characteristic zero makes it separable. A finite normal splitting field is a finite separable extension of K(Y)K(Y). The normalization of the affine variety in that extension is finite. After deleting the nonzero leading coefficients, the finitely many denominators just introduced, and the discriminant of the monic squarefree polynomial, its integral splitting algebra is finite etale. Equivalently, restrict the normalization to this etale open. The roots are integral there and hence regular on the normalization. Their pairwise differences are units because their product, up to sign and square, is the invertible discriminant. The original factors retain their multiplicities, and their leading coefficients are units. The finite separable extension gives a constant positive rank on this integral open. Clearing all deleted denominators gives one nonzero Δ∈K[Y]\Delta\in K[Y]. No alteration of the ground field is used. ◻

Theorem (The original disc-projection kernel theorem). Let u,v∈Cb(U)u,v\in\mathcal C_b(U) and suppose D[u]=D[v]D[u]=D[v] on every original compact complementary test on UU. Then D[u∘π]=D[v∘π]D[u\circ\pi]=D[v\circ\pi] on every original compact complementary test on U×KDsU\times_K\mathbb D_s. The equality holds for the actual locally bounded classical pullbacks over the original KK. It does not define the pullback of an arbitrary positive closed current under π\pi.

Status: NEW_PROVED. Its inputs beyond R and B are the already proved R2 ambient coefficient nonpolarity, the finite etale original trace theorem, and the split projection formula of this round.

Proof. Put f=u−vf=u-v and Q=D[u∘π]−D[v∘π].Q=D[u\circ\pi]-D[v\circ\pi]. Composition preserves actual classicality by its all-extension curve definition. Thus the two terms are positive closed ambient bounded Bedford–Taylor currents. Every one of their coefficient variations vanishes on analytic zero loci, by the proved ambient coefficient-polar theorem. The same is true of the signed coefficient variations of QQ.

Fix an arbitrary original compact test α\alpha of bidegree (n,n)(n,n) on the product, where n=dim⁡Yn=\dim Y. Use Lemma “Rational moment localization of an original test” and linearity to treat one compactly localized rational moment summand. Let P1,…,PNP_1,\ldots,P_N contain all polynomial numerators and denominators of its rational moments and all regular functions used by its scalar norm-box localization. Adjoin TT. Apply Lemma “A common finite etale splitting away from a base divisor” to this fixed finite list, obtaining Δ\Delta and p:Z→Y0p:Z\to Y_0.

Let H(x,T)=Δ(x)∏l=0NPl(x,T),P0=T.H(x,T)=\Delta(x)\prod_{l=0}^N P_l(x,T), \qquad P_0=T. This is a nonzero regular function on the integral affine variety Y×A1Y\times\mathbf A^1. Choose smooth functions 0≤θj≤10\leq\theta_j\leq1 on R\mathbf R with θj(s)=0\theta_j(s)=0 for s≤−j−1s\leq-j-1, θj(s)=1\theta_j(s)=1 for s≥−js\geq-j, and θj(s)→1\theta_j(s)\to1 for every finite ss. The function ζj=θj(log⁡∣H∣)\zeta_j=\theta_j(\log|H|) extends by zero at H=0H=0 and is original smooth. The form αj=ζjα\alpha_j=\zeta_j\alpha has compact support where every chosen PlP_l and Δ\Delta is nonzero. It is a global finite rational unit moment expression there. It need not have a bound independent of jj in any smooth norm; no such bound will be used.

Let U0=U∩Y0anU_0=U\cap Y_0^{\rm an} and restrict the finite etale map to ZU=p−1(U0)Z_U=p^{-1}(U_0). Theorem “Original finite etale trace” and Corollary “Finite etale preservation of the bounded potential kernel” give D[u∘p]=D[v∘p]on ZU.D[u\circ p]=D[v\circ p]\quad\hbox{on }Z_U. Let q=p×id⁡q=p\times\operatorname{id} on the corresponding disc products. Its pullback q∗αjq^*\alpha_j is original smooth and compactly supported. The polynomial factorization expresses all its vertical unit moments as affine linear combinations of the base unit logs and the logs log⁡∣T−ai∣\log|T-a_i|. The support avoids every graph T=aiT=a_i. Because that support is compact, all these linear-factor logs have finite upper and lower bounds there. Multiplying their real coefficient expression by smooth cutoffs equal to one near those ranges puts its coefficient support inside a bounded box in all the linear-factor coordinates. An original base cutoff equal to one near the projected support may likewise be inserted. It is therefore exactly a split-moment test in the sense of (eq:r4-split-test), on the original finite etale base ZUZ_U. Its extension into the affine-line product remains smooth and compactly supported. The split projection theorem gives

⟨D[f∘p∘πZ],q∗αj⟩=0.\bigl\langle D[f\circ p\circ\pi_Z],q^*\alpha_j\bigr\rangle=0.

We now descend this numerical equality without invoking a pullback of QQ. If dd is the finite etale degree, the ordinary finite-flat integration formula for smooth top forms gives the equality of signed Radon measures q∗[q∗(Dαj)]=d[Dαj].q_*[q^*(D\alpha_j)]=d[D\alpha_j]. The equality holds on every original compact smooth scalar test and hence on bounded Borel multipliers by Radon uniqueness. The function f∘πf\circ\pi is Borel and bounded on a neighborhood of the fixed projected support. The definition of a distributional derivative and the smooth pullback rule D(q∗αj)=q∗(Dαj)D(q^*\alpha_j)=q^*(D\alpha_j) therefore imply ⟨D[f∘p∘πZ],q∗αj⟩=d⟨D[f∘π],αj⟩=d⟨Q,αj⟩.\bigl\langle D[f\circ p\circ\pi_Z],q^*\alpha_j\bigr\rangle =d\langle D[f\circ\pi],\alpha_j\rangle =d\langle Q,\alpha_j\rangle. Equation (eq:r4-split-upstairs-zero) proves ⟨Q,αj⟩=0\langle Q,\alpha_j\rangle=0 for each jj.

Finally 0≤ζj≤10\leq\zeta_j\leq1 converges to one away from the fixed analytic set {H=0}\{H=0\}. Every coefficient variation of QQ vanishes on that set. Fixed-measure dominated convergence on the compact support of α\alpha gives ⟨Q,α⟩=lim⁡j⟨Q,ζjα⟩=0.\langle Q,\alpha\rangle =\lim_j\langle Q,\zeta_j\alpha\rangle=0. There is no simultaneously changing coefficient measure in this last step. Summing the finitely many original localized summands proves the theorem for the original arbitrary test. ◻

Corollary (Finite-dimensional ordinary polydisc factors). Under the same hypotheses, the kernel equality is preserved by projection from UU times any finite-dimensional ordinary polydisc over KK.

Proof. Apply Theorem “The original disc-projection kernel theorem” successively. Each intermediate space is an ordinary open in the smooth integral affine analytification Y×ArY\times\mathbf A^r, and the pulled-back potentials remain actual locally bounded classical. This is a DIRECT_COROLLARY. ◻

Remaining scope of the background quotient question.

The new theorem resolves the original-test disc and polydisc projection gate in the stated algebraic-neighborhood scope. Combining it with the already proved same-field smooth algebraic normal-trace theorem applies only when the graph immersion lies in that theorem’s established algebraic scope. It does not silently enlarge that scope to arbitrary analytic graphs or arbitrary complete field extensions. Nor does it yet prove that every original distributional DD-kernel difference has a continuous representative. The full statement D[u]=D[v]⇒D((u−v)S)=0D[u]=D[v]\Rightarrow D((u-v)S)=0 for arbitrary SS therefore remains open beyond the separately established background classes until those additional steps are supplied.

Functoriality of the bounded potential kernel over the original field

The ground field throughout the main theorem is the prescribed finite extension K/QpK/\mathbf Q_p with its original absolute value. An allowed space is good, Hausdorff, boundaryless and quasi-smooth, as in B. Currents and tests are the original CLD objects. Only pullbacks of specified functions, and pullbacks of original smooth test forms, occur below. There is no definition of pullback for an arbitrary current.

Lemma (Kernel preservation by a product with an allowed space). Let X,ZX,Z be allowed spaces over KK, and let u,v∈Cb(X)u,v\in\mathcal C_b(X) satisfy D[u]=D[v]D[u]=D[v]. For the projection pX:X×KZ→Xp_X:X\times_K Z\to X, D[u∘pX]=D[v∘pX]D[u\circ p_X]=D[v\circ p_X] on every original compact complementary test.

Status: DIRECT_COROLLARY of the proved polydisc kernel and the established same-field algebraic normal-trace restriction.

Proof. Work locally around an arbitrary point of the analytic product. Its two images have ordinary algebraic neighborhoods by B, Lemma D.4. Their product is an ordinary open neighborhood of the chosen product point. We may therefore replace XX and ZZ by ordinary opens in smooth affine varieties X0,Z0X_0,Z_0 over the same KK. Restrict to the relevant integral components. Choose a closed algebraic immersion Z0↪AMZ_0\hookrightarrow\mathbf A^M. Corollary “Finite-dimensional ordinary polydisc factors”, on increasing ordinary polydiscs in AM\mathbf A^M, proves the equality on X×AM,anX\times\mathbf A^{M,\rm an}. The algebraic immersion X0×Z0↪X0×AMX_0\times Z_0\hookrightarrow X_0\times\mathbf A^M is a smooth closed immersion of smooth varieties. Apply Proposition “Restriction of a bounded difference with positive curvature” to the two actual classical pullbacks and then restrict to X×ZX\times Z. This proves the assertion locally at every product point. The degree-zero case for XX is immediate, since a finite function on a zero-dimensional smooth space is locally constant. ◻

Lemma (An ordinary rectangular neighborhood of a zero section). Let CC be a good Hausdorff analytic space, let c∈Cc\in C, and let OO be an ordinary open neighborhood of (c,0)(c,0) in C×KAN,anC\times_K\mathbf A^{N,\rm an}. After shrinking an ordinary base neighborhood C1C_1 of cc, there is s>0s>0 with C1×KDsN⊂O.C_1\times_K\mathbb D_s^N\subset O. If the zero section over a compact affinoid neighborhood BB is already contained in OO, one may choose C1=Int⁡CBC_1=\operatorname{Int}_C B after making ss sufficiently small.

Proof. The inverse image of OO under the zero section is ordinary open and contains cc. Choose an affinoid neighborhood BB of cc contained in this inverse image. Fix any R>0R>0. The actual analytic domains B×KD‾rN(0<r≤R)B\times_K\overline{\mathbb D}_r^N\qquad(0<r\leq R) are compact, are nested as rr decreases, and their intersection is the zero section over BB. The last statement follows because a multiplicative seminorm with all normal coordinates of norm zero factors through their common zero ideal. If none of these compact tubes were contained in OO, their intersections with the closed complement of OO inside the fixed radius-RR compact tube would have the finite intersection property. Compactness would give a point in the zero section outside OO, a contradiction. Thus a sufficiently small closed tube lies in OO; its base interior times the open disc is the required ordinary product neighborhood. This argument is about analytic products, not a replacement by Cartesian products of their sets of points. ◻

Theorem (Same-field analytic functoriality). Let F:Z→XF:Z\to X be an arbitrary analytic morphism of allowed KK-analytic spaces. If u,v∈Cb(X)u,v\in\mathcal C_b(X) satisfy D[u]=D[v]D[u]=D[v] on all original tests, then D[u∘F]=D[v∘F]D[u\circ F]=D[v\circ F] on all original tests of ZZ. The functions are their actual locally bounded classical pullbacks.

Status: NEW_PROVED relative to the disc-projection result of this round and the accepted algebraic tubular-neighborhood input R, Lemma 5.2. The graph in this theorem is allowed to be analytic. Its restriction is proved by an explicit translation to a zero section, not by extending the algebraic-immersion hypothesis of the old restriction theorem without proof.

Proof. The assertion is local at a point z∈Zz\in Z. If the target is zero-dimensional near F(z)F(z), the two functions there are locally constant and the conclusion is immediate. Assume from now on that its local dimension is positive. By B, Lemma D.4, choose ordinary algebraic neighborhoods of zz and of F(z)F(z); shrink the former so that its image lies in the latter. Thus the source is an ordinary open in a smooth affine KK-variety and the target is an ordinary open in a smooth affine KK-variety YY. Choose a closed algebraic immersion Y↪ANY\hookrightarrow\mathbf A^N.

R, Lemma 5.2, gives an ordinary neighborhood VV of F(z)F(z) in YanY^{\rm an}, an ordinary ambient neighborhood W⊂AN,anW\subset\mathbf A^{N,\rm an}, and an analytic isomorphism Φ:V×KDsr→∼W\Phi:V\times_K\mathbb D_s^r\xrightarrow{\sim}W whose zero section is the given algebraic immersion. Here rr is its codimension. We may shrink VV inside the target open and correspondingly restrict WW. Let RW=pr⁡V∘Φ−1:W→VR_W=\operatorname{pr}_V\circ\Phi^{-1}:W\to V. The functions u~=u∘RW,v~=v∘RW\widetilde u=u\circ R_W,\qquad \widetilde v=v\circ R_W are actual locally bounded classical functions on WW, since all-extension classicality is preserved by analytic composition. They restrict to u,vu,v on the embedded VV. The polydisc kernel theorem, followed by the analytic isomorphism Φ\Phi, gives D[u~]=D[v~]D[\widetilde u]=D[\widetilde v] on WW. An analytic isomorphism identifies the original test sheaves and their calibrated integrals, so this transfer does not invoke a general current pullback.

Shrink the source to an ordinary neighborhood CC of zz such that F(C)⊂VF(C)\subset V. Regard FF as an analytic map C→ANC\to\mathbf A^N. By Lemma “Kernel preservation by a product with an allowed space”, the two functions u~(x)\widetilde u(x) and v~(x)\widetilde v(x) on C×WC\times W have equal original curvature currents. The analytic automorphism A:C×AN,an⟶C×AN,an,A(c,t)=(c,F(c)+t)A:C\times\mathbf A^{N,\rm an}\longrightarrow C\times\mathbf A^{N,\rm an},\qquad A(c,t)=(c,F(c)+t) has inverse (c,x)↦(c,x−F(c))(c,x)\mapsto(c,x-F(c)). Let O=A−1(C×W)O=A^{-1}(C\times W), an ordinary open containing the zero section over CC. On OO put U(c,t)=u~(F(c)+t),V′(c,t)=v~(F(c)+t).U(c,t)=\widetilde u(F(c)+t),\qquad V'(c,t)=\widetilde v(F(c)+t). These are actual classical functions and D[U]=D[V′]D[U]=D[V'] on every original test, by the same isomorphism rule.

Lemma “An ordinary rectangular neighborhood of a zero section” supplies an ordinary neighborhood C1C_1 of zz and a radius a>0a>0 such that C1×DaN⊂OC_1\times\mathbb D_a^N\subset O. Restrict the equality there. Apply the established zero-section normal-trace lemma successively in its NN normal coordinates. Its base at each stage is an ordinary open in a smooth algebraic analytification; its potentials remain actual locally bounded classical functions. The conclusion is D[U∣t=0]=D[V′∣t=0]on C1.D[U|_{t=0}]=D[V'|_{t=0}] \quad\hbox{on }C_1. At the zero section these functions are exactly u∘Fu\circ F and v∘Fv\circ F. This proves the claimed local equality and hence the theorem. A zero-dimensional source has no curvature tests and requires no additional argument. ◻

Corollary (Smooth analytic immersion backgrounds). Let j:Z↪Xj:Z\hookrightarrow X be a smooth closed analytic immersion of allowed spaces over this same KK. If D[u]=D[v]D[u]=D[v], then D((u−v)[Z])=0D((u-v)[Z])=0 where the local pushforward is defined. More generally the same is true for S=j∗QS=j_*Q when QQ is locally a finite linear combination of bounded ambient BT products on ZZ, a smooth separately closed form current, or belongs to the bounded ambient BT weak-closure class already covered in R3.

Proof. The theorem gives D[u∘j]=D[v∘j]D[u\circ j]=D[v\circ j] on ZZ. Apply the corresponding R3 quotient statement there and push forward using Borel multiplication and the ordinary current pushforward, which commute with DD in this local closed-immersion setting. This is a DIRECT_COROLLARY. The ambient current may be carried by the analytic polar subspace ZZ; no nonpolarity assumption is imposed on it. ◻

The complete-extension gate remains distinct.

The theorem applies to arbitrary analytic morphisms over the original KK, not merely algebraic maps. It does not view a curve over an arbitrary complete extension as a finite-type KK-analytic curve. Such a change would be invalid in general. The original equality has not yet been transferred to all those extension spaces. Consequently this section does not deduce that both signs of u−vu-v are classically psh, does not assert a continuous representative for every kernel difference, and does not declare the arbitrary-background quotient solved.

The finite-extension part of the functoriality question

Corollary (Finite extensions with their inherited calibration). Let K′/KK'/K be a finite extension, endowed with the unique extension of the prescribed absolute value, and let XX be an allowed KK-analytic space. If u,v∈Cb(X)u,v\in\mathcal C_b(X) and D[u]=D[v]D[u]=D[v] on all original KK-tests, then D[u∘πK′/K]=D[v∘πK′/K]D[u\circ\pi_{K'/K}]=D[v\circ\pi_{K'/K}] on all original K′K'-tests of XK′X_{K'}. More generally, for any analytic morphism F:Z→XK′F:Z\to X_{K'} from an allowed K′K'-analytic space, D[u∘πK′/K∘F]=D[v∘πK′/K∘F]D[u\circ\pi_{K'/K}\circ F] =D[v\circ\pi_{K'/K}\circ F] on all its original K′K'-tests.

Status: DIRECT_COROLLARY. Only a finite scalar extension is involved, and the absolute value is not renormalized. This does not settle the corresponding assertion for arbitrary complete extensions.

Proof. Viewed over KK, the projection XK′→XX_{K'}\to X is finite etale. Its source remains good, Hausdorff, boundaryless and quasi-smooth over KK. Thus Corollary “Finite etale preservation of the bounded potential kernel” gives the first equality on the original KK-test sheaf of this same analytic space.

There is no hidden change of test functions. Viewing a K′K'-analytic space over KK changes neither its analytic functions nor their absolute values. Its original unit moment presentations, and hence its original smooth form sheaf, are identical. Its calibrated top integration currents obey [1]K=[K′:K] [1]K′.[1]_K=[K':K]\,[1]_{K'}. One checks this on any calibrated full-dimensional moment cell: the finite degree of its structural map to the original KK-torus is the degree of the same map to the K′K'-torus multiplied by [K′:K][K':K]. Refinement and a finite partition give the displayed identity on every original smooth top test. This is the finite-constant calibration rule of B, Lemma D.5; it does not replace the given absolute value. The same factor applies to integration with a locally bounded Borel coefficient, and distributional differentiation preserves it. Dividing the KK-current equality by this positive constant proves the equality on all original K′K'-tests.

For the second assertion, view ZZ over KK. Since K′/KK'/K is finite separable, this again is an allowed KK-analytic space. Apply Theorem “Same-field analytic functoriality” to the KK-analytic composite Z→XK′→XZ\to X_{K'}\to X. The identical calibration argument on ZZ gives the asserted K′K' equality. ◻

In particular all analytic curve maps over finite extensions are covered at the level of these original current equalities. An analytic curve over a general complete extension need not be an allowed finite-type KK-analytic space; the proof does not treat it as one. The unrestricted complete-extension gate in the classical definition therefore remains separate.

Gauss residue fields: full fibers and exact test spreading

The field KK remains the prescribed finite extension of Qp\mathbf Q_p, with its original absolute value. Every use of the bounded BT calculus in this section takes place on an allowed space over this KK. No version of R or B over a general complete extension is assumed. For a positive real polyradius r=(r1,…,rs)\mathbf r=(r_1,\ldots,r_s), write ζr∈(Gms)an,Kr=H(ζr)=K(z1,…,zs)^r\zeta_{\mathbf r}\in(\mathbf G_m^s)^{\rm an},\qquad K_{\mathbf r}=\mathscr H(\zeta_{\mathbf r}) =\widehat{K(z_1,\ldots,z_s)}_{\mathbf r} for the Gauss point and its completed residue field. The number of parameters ss is finite. Radii are arbitrary positive real numbers.

The fixed full-fiber formula

Let XX be an allowed space of pure dimension nn, and let AA be an ordinary open neighborhood of ζr\zeta_{\mathbf r} in the parameter torus. Denote the projections of X×KAX\times_K A by qq and pp, and put b=log⁡max⁡(1,∣z1∣/r1,…,∣zs∣/rs),Br=(D[b∘p])s.b=\log\max(1,|z_1|/r_1,\ldots,|z_s|/r_s),\qquad B_{\mathbf r}=(D[b\circ p])^s. The function b∘pb\circ p is continuous classical psh. The actual fiber of pp over ζr\zeta_{\mathbf r} is the whole KrK_{\mathbf r}-analytic space XKrX_{K_{\mathbf r}}. Write jj for its identification with the corresponding closed subset of X×KAX\times_K A.

Proposition (The calibrated full Gauss fiber). For every original compact smooth form η\eta of bidegree (n,n)(n,n) on X×KAX\times_K A,

⟨Br,η⟩=∫XKrj∗η.\langle B_{\mathbf r},\eta\rangle =\int_{X_{K_{\mathbf r}}}j^*\eta.

The integral on the right uses the original KrK_{\mathbf r}-calibration. Its coefficient is exactly one, with no additional factorial. Moreover, for every locally bounded Borel function ww on the product,

⟨wBr,η⟩=∫XKr(w∘j)j∗η.\langle wB_{\mathbf r},\eta\rangle =\int_{X_{K_{\mathbf r}}}(w\circ j)j^*\eta.

Here pullback of a smooth test means restriction of its analytic unit moments to the fiber. No pullback of an arbitrary current is used.

Status: EXTERNAL_KNOWN for (eq:r4-full-gauss-fiber), and DIRECT_COROLLARY for its fixed Borel-weighted version.

Proof. The unweighted identity is precisely [CLD25, Proposition 18.6.1 and Corollary 18.6.2, pp. 532–534], applied to the analytic functions z1,…,zsz_1,\ldots,z_s on the product, with the parameter denoted tt in that corollary equal to one. Proposition 18.6.1 explicitly defines the right-hand current by integration on the full completed-residue-field analytic fiber. Corollary 18.6.2 computes the coefficient to be one.

This use of the 2025 statement concerns common original data only. For example, locally smooth convex approximations to max⁡(0,t1−log⁡r1,…,ts−log⁡rs)\max(0,t_1-\log r_1,\ldots,t_s-\log r_s) in the original unit moment coordinates identify its curvature products, on every original test, with the continuous-potential products used here. Both use the same calibrated smooth integration. This comparison does not extend an arbitrary original current to the larger 2025 test sheaf.

Fix η\eta. Apply the identity also to χη\chi\eta, where χ\chi is an arbitrary original compact smooth scalar function. Uniqueness of Radon measures gives μBr,η=j∗μj∗η.\mu_{B_{\mathbf r},\eta}=j_*\mu_{j^*\eta}. Indeed original smooth scalar functions determine Radon measures on the actual Hausdorff analytic space. The right-hand measure is Radon: the fiber is closed and its intersection with the compact support of η\eta is compact. Integration of a fixed locally bounded Borel function in this signed-measure identity proves (eq:r4-borel-gauss-fiber). There is no varying-measure limit in this step. ◻

Exact spreading of every original fiber test

Lemma (Exact spreading from an algebraic parameter fiber). Let TT be a smooth affine KK-variety, let a∈Tana\in T^{\rm an}, and let F=H(a)F=\mathscr H(a). Let XX be an ordinary open in the analytification of a smooth affine KK-variety YY. For every original compact smooth form β\beta on XFX_F, there are an ordinary open neighborhood A⊂TanA\subset T^{\rm an} of aa and an original compact smooth form α\alpha of the same bidegree on X×KAX\times_K A such that ja∗α=β.j_a^*\alpha=\beta. This is equality of original forms, not only approximation of their integrals. Consequently ja∗Dα=Dβj_a^*D\alpha=D\beta. The same assertion holds locally for an arbitrary allowed XX over KK, after a finite partition subordinate to its ordinary algebraic neighborhoods.

Status: NEW_PROVED as a test-spreading lemma. Its proof uses affinoid density and the original definition of smooth forms, not a regularization theorem over FF.

Proof. We give the support and zero-locus details, since mere density of K(T)K(T) in a residue field would not by itself prove this assertion. The image in XX of supp⁡β\operatorname{supp}\beta is compact. Choose an original smooth compact cutoff χX\chi_X equal to one near this image. Its support is covered by finitely many compact polynomial norm domains in YanY^{\rm an}, all contained in XX. Their scalar extensions are compact; in particular all the finitely many affine coordinates used below have uniform bounds on their union, denoted DFD_F. Finite localization permits working on one such stage at a time.

Around the compact support of β\beta, choose finitely many compact polynomial or rational norm boxes in YFanY_F^{\rm an} whose slightly larger boxes remain in original unit-moment presentations for β\beta. Such boxes form a neighborhood basis: affine-coordinate bounds first give an affinoid domain, and finitely many strict norm inequalities give smaller rational domains. Their algebras are the completed algebras obtained from F[Y]F[Y] by the finitely many rational localizations specifying the boxes. Thus fractions of elements of F[Y]F[Y], with the prescribed denominators inverted, are uniformly dense in their analytic functions.

On a slightly larger box, approximate each analytic unit hh occurring in the presentation by a rational function P/QP/Q with error strictly less than the positive minimum of ∣h∣|h|. The non-Archimedean triangle inequality then gives ∣P/Q∣=∣h∣|P/Q|=|h| throughout that box. In particular the logarithms, and all their original smooth differentials, agree on its ordinary interior. This replaces the finitely many analytic unit moments by rational FF-moments exactly where the local term is to be supported.

Choose localizing bumps by smooth profiles of the finitely many polynomial norms defining the boxes. When a polynomial may vanish, the profile is constant below a fixed positive norm threshold; this defines an original smooth scalar function across its zero locus. Include extra lower cutoffs for every numerator and denominator of a rational unit moment. They are zero below a positive threshold, so that all those polynomials are units on a neighborhood of the support of the local term. Upper norm cutoffs give compact supports. If b1,…,bNb_1,\ldots,b_N are such bumps with ∑bi≥1\sum b_i\geq1 near supp⁡β\operatorname{supp}\beta, use χi=biR(∑bi),\chi_i=b_i R\Bigl(\sum b_i\Bigr), where RR is smooth, equals 1/t1/t for t≥1/2t\geq1/2, and is zero for t≤1/4t\leq1/4. This is a partition equal to one near the required support. Extend the real coefficient forms of the moment presentations with a real smooth cutoff around the compact moment images. We have now written β\beta as a finite sum of localized rational-moment forms, globally extended by zero, with the lower cutoffs just described. Only finitely many coefficients in FF occur in the polynomials used for these moments and bumps.

The evaluations at aa of rational functions on TT whose denominators do not vanish at aa are dense in FF. In fact their images contain the fraction field of the coordinate ring modulo the prime kernel of aa, and FF is its completion. Approximate all the preceding finite coefficients by such evaluations. Bounds on the affine coordinates on DFD_F make the resulting polynomial approximations uniform on the whole of DFD_F.

There are only finitely many positive cutoff thresholds. Choose the coefficient approximations with errors smaller than all of them, and smaller than the unit lower bounds on the active supports. For clarity, if PP is one of the polynomials and P~\widetilde P its approximation, then ∥P−P~∥DF<δ⟹{∣P∣=∣P~∣,∣P∣≥δ,∣P∣,∣P~∣<δ,∣P∣<δ.\|P-\widetilde P\|_{D_F}<\delta \quad\Longrightarrow\quad \begin{cases} |P|=|\widetilde P|,& |P|\geq\delta,\\ |P|,|\widetilde P|<\delta,& |P|<\delta. \end{cases} Every profile used at a possible zero of PP is constant on the second region. Its value is therefore unchanged everywhere on DFD_F, including the zero locus. Every unit moment is protected by a lower cutoff above this threshold, so its logarithm is unchanged on an ordinary neighborhood of the active support. Hence the localized differential forms, not merely their scalar values, are unchanged. In particular the localized expression is still zero on precisely the same part of the fiber outside its support.

The approximating rational functions on TT are analytic on an ordinary neighborhood AA of aa after removing their finitely many denominator zero loci. Their coefficients define analytic polynomials on Y×KAY\times_K A. Use the same moment expressions and norm-profile cutoffs there. Lower cutoffs enforce the required unit loci, and profiles constant near zero give original smoothness across the remaining polynomial zero loci. Multiply by χX∘q\chi_X\circ q and by an original compact smooth cutoff on AA equal to one near aa. The resulting form α\alpha has compact support: it is contained in a finite union of actual products of compact affinoid domains. Its fiber restriction is exactly β\beta by the preceding uniform norm and profile equalities. Restriction of unit moments commutes with d′d' and d′′d'', giving the assertion about DD.

For a general allowed XX, apply B, Lemma D.4, only over the original field KK to the compact image of the support. A finite original partition reduces to the preceding construction on ordinary affine algebraic neighborhoods. Intersect the finitely many parameter neighborhoods and add the resulting forms. ◻

The neighborhood AA in this lemma is an ordinary neighborhood. Even when TT is a torus and aa is its Gauss point, AA need not be an entire annulus or a polyradius box: a rational coefficient may have zeros of its denominator at the same radius, away from the Gauss point. Removing those zeros is legitimate and is needed.

The kernel over a finite polyradius Gauss field

Theorem (Gauss residue-field preservation of the potential kernel). Let u,v∈Cb(X)u,v\in\mathcal C_b(X) on an allowed KK-analytic space and suppose D[u]=D[v]D[u]=D[v] on every original compact complementary test. For every finite positive real polyradius r\mathbf r and the scalar projection π:XKr→X\pi:X_{K_{\mathbf r}}\to X, D[u∘π]=D[v∘π]D[u\circ\pi]=D[v\circ\pi] on every original KrK_{\mathbf r}-test. The pullbacks are the actual locally bounded classical functions.

Status: DIRECT_COROLLARY of the preceding exact spreading lemma, the external full-fiber identity, and the same-field product kernel already proved in this round.

Proof. The assertion in dimension zero is immediate. Assume n≥1n\geq1 and write f=u−vf=u-v. On X×KAX\times_K A, the original-field product result Lemma “Kernel preservation by a product with an allowed space” gives D[u∘q]=D[v∘q].D[u\circ q]=D[v\circ q]. The background BrB_{\mathbf r} is an ambient BT product. Therefore the already established ambient mixed-background independence, Lemma “Ambient mixed backgrounds”, gives

D((f∘q)Br)=0.D\bigl((f\circ q) B_{\mathbf r}\bigr)=0.

This step uses only the proved ambient-product special case of the potential quotient; it does not invoke the open arbitrary-background assertion.

Let β\beta be any original compact KrK_{\mathbf r}-smooth form of bidegree (n−1,n−1)(n-1,n-1) on XKrX_{K_{\mathbf r}}. Apply Lemma “Exact spreading from an algebraic parameter fiber”, with parameter T=GmsT=\mathbf G_m^s and a=ζra=\zeta_{\mathbf r}, to obtain AA and α\alpha with j∗α=βj^*\alpha=\beta. Test (eq:r4-gauss-weighted-kernel) on α\alpha and use the fixed Borel-weighted fiber identity on DαD\alpha. It gives 0=⟨(f∘q)Br,Dα⟩=∫XKr(f∘π)Dβ=⟨D[f∘π],β⟩.0=\langle(f\circ q)B_{\mathbf r},D\alpha\rangle =\int_{X_{K_{\mathbf r}}}(f\circ\pi)D\beta =\langle D[f\circ\pi],\beta\rangle. All the BT operations in this argument occurred on the original KK-space X×KAX\times_K A. The assertion for ss parameters was proved in one step; no induction applies R or B over an intermediate infinite extension. ◻

Finite extensions of these Gauss fields

Lemma (An algebraic etale realization of a finite fiber field). Let E=KrE=K_{\mathbf r} and let L/EL/E be finite. There are a Zariski open T0⊂GmsT_0\subset\mathbf G_m^s containing ζr\zeta_{\mathbf r} in its analytification and a finite etale map T′→T0T'\to T_0 of degree e=[L:E]e=[L:E] such that the analytic fiber over ζr\zeta_{\mathbf r} is a single point a′a' with completed residue field isometric to LL. The spaces and the map are defined over the original KK.

Status: EXTERNAL_KNOWN elementary Hensel–Krasner realization; the following proof fixes its precise field and calibration scope.

Proof. Characteristic zero makes L/EL/E separable. Choose θ\theta with L=E(θ)L=E(\theta), and let PP be its monic minimal polynomial, of degree ee. Approximate its coefficients by elements of the dense field K(z1,…,zs)K(z_1,\ldots,z_s). For a sufficiently close monic polynomial QQ of the same degree, Hensel’s argument in the complete field LL gives a root θ′∈L\theta'\in L arbitrarily close to θ\theta. Explicitly, on a sufficiently small closed ball about θ\theta the map t⟼t−Q(t)/Q′(θ)t\longmapsto t-Q(t)/Q'(\theta) is a strict contraction into that ball: the coefficient errors make Q(θ)Q(\theta) sufficiently small, while Q′(θ)Q'(\theta) stays close to the nonzero number P′(θ)P'(\theta) and the higher Taylor terms are dominated by the linear term. Completeness gives its fixed point.

Choose the closeness smaller than the minimum distance between θ\theta and its distinct EE-conjugates. Then E(θ′)=LE(\theta')=L. For completeness, if an automorphism over E(θ′)E(\theta') of an algebraic closure moved θ\theta, invariance of the unique extended absolute value would put both θ\theta and that conjugate at the same, strictly smaller distance from θ′\theta' than their mutual distance. The ultrametric inequality contradicts this. Thus θ∈E(θ′)\theta\in E(\theta'), and the reverse inclusion follows from θ′∈L\theta'\in L. Consequently QQ is irreducible over EE, has nonzero discriminant, and is also irreducible over K(z1,…,zs)K(z_1,\ldots,z_s).

Clear the finitely many coefficient denominators and invert the discriminant on an affine Zariski open T0T_0 of the parameter torus. Every nonzero rational function has nonzero value at its Gauss point, so ζr∈T0an\zeta_{\mathbf r}\in T_0^{\rm an}. The monic algebra O(T0)[t]/(Q(t))\mathcal O(T_0)[t]/(Q(t)) defines the required finite etale KK-variety T′T'. Its completed fiber algebra at the Gauss point is E[t]/(Q)≃LE[t]/(Q)\simeq L, a field. The extension of the complete absolute value is unique, so this fiber is precisely one analytic point with residue field LL. ◻

Corollary (Finite extensions after a Gauss extension). Under the hypotheses of Theorem “Gauss residue-field preservation of the potential kernel”, let LL be any finite extension of KrK_{\mathbf r}, with the inherited absolute value. Then D[u∘πL/K]=D[v∘πL/K]D[u\circ\pi_{L/K}]=D[v\circ\pi_{L/K}] on all original LL-tests of XLX_L.

Status: DIRECT_COROLLARY. The following argument returns to an original-KK parameter space; it does not apply the finite-KK theorem over KrK_{\mathbf r}.

Proof. Use the realization T′→T0T'\to T_0 just proved, and an ordinary neighborhood A′A' of its point a′a'. On the original-KK space X×KA′X\times_K A' pull back the parameter function b=log⁡max⁡(1,∣zi∣/ri)b=\log\max(1,|z_i|/r_i). Denote its ss-fold curvature product by B′B'. The external full-fiber formula gives the integration current on the fiber of the parameter map to T0T_0 at ζr\zeta_{\mathbf r}. That fiber is XLX_L viewed as an E=KrE=K_{\mathbf r}-analytic space. Finite-constant calibration gives

B′=e j∗′[1]XL,L,e=[L:E].B'=e\,j'_*[1]_{X_L,L},\qquad e=[L:E].

Indeed on an LL-analytic space viewed over EE its integration current is ee times the LL-integration current; see [CLD25, Remark 14.4.3, p. 410]. This is not the different pushforward normalization for scalar base change. Equality (eq:r4-finite-gauss-fiber-calibration) also holds after a fixed locally bounded Borel multiplication, by the coefficient-Radon argument used in Proposition “The calibrated full Gauss fiber”.

The same-field product kernel on X×KA′X\times_K A' and ambient BT independence give D((f∘q)B′)=0D((f\circ q)B')=0, with f=u−vf=u-v. Given any original LL-test β\beta on XLX_L, use exact spreading with parameter variety T′T' and point a′a' to obtain a compact original KK-test α\alpha restricting to β\beta. Pairing the last equality with α\alpha gives 0=e∫XL(f∘πL/K)Dβ.0=e\int_{X_L}(f\circ\pi_{L/K})D\beta. Division by the positive integer ee proves the result. All regularization, product, and potential-independence inputs were again used only over the original finite field KK. ◻

A failed shortcut and the remaining gate

Example (A single Gauss section misses correlated fiber points). Take X=Ax1,anX=\mathbf A^{1,\rm an}_x and a parameter zz of radius one. Put w=x−zw=x-z. Choose χ∈Cc∞(R)\chi\in C_c^\infty(\mathbf R) with χ(0)=1\chi(0)=1, and choose a nonzero nonnegative ψ∈Cc∞((−∞,0))\psi\in C_c^\infty(( -\infty,0)). On the unit locus of zwzw put α=χ(log⁡∣z∣)ψ(log⁡∣w∣)d′log⁡∣w∣∧d′′log⁡∣w∣,\alpha=\chi(\log|z|)\psi(\log|w|) d'\log|w|\wedge d''\log|w|, and extend it by zero. This is an original smooth compact form: its support has both zz and ww bounded above and away from zero, and x=z+wx=z+w is bounded there. The algebraic coordinate change (x,z)↦(w,z)(x,z)\mapsto(w,z) and unit torus calibration give ⟨Dmax⁡(0,log⁡∣z∣),α⟩=∫Rψ(t) dt>0.\bigl\langle D\max(0,\log|z|),\alpha\bigr\rangle =\int_{\mathbf R}\psi(t)\,dt>0. In contrast, the independent radius-one Gauss lift over any a∈Xa\in X satisfies ∣x−z∣=max⁡(∣x(a)∣,1)≥1.|x-z|=\max(|x(a)|,1)\geq1. Thus α\alpha vanishes on that entire independent Gauss section. The full fiber over the parameter Gauss point contains points with ∣x−z∣<1|x-z|<1 and cannot be replaced by this section. The shortcut is FALSE; the full-fiber identity used above retains precisely these additional points.

The statements proved here concern finite extensions of finite polyradius Gauss residue fields, including a single such field. They do not assert kernel preservation for an arbitrary complete extension of KK, or for every analytic morphism defined over such an extension. In particular an arbitrary extension-analytic curve is not silently viewed as an allowed finite-type KK-analytic space. Passing from this explicitly treated family of fields to all the complete extensions in the definition of classical psh remains an OPEN gate. Although the exact spreading lemma also applies at non-Gauss algebraic parameter points, no representation of their general residue-field fibers by the bounded ambient BT backgrounds used above has been proved. Density of Gauss points is not a substitute for that missing fixed-fiber statement. The arbitrary-background potential quotient therefore remains OPEN at this stage.

Reduction of scalar kernel functoriality to finite parameter fields

This section isolates the remaining scalar-extension question. It does not identify a general parameter fiber with a bounded ambient BT product. The base field KK and its absolute value remain fixed.

Lemma (Exact finite-coefficient descent of original tests). Let L/KL/K be any complete extension, with its prescribed extending absolute value, and let XX be an allowed KK-analytic space. Given finitely many original compact smooth forms β1,…,βN\beta_1,\ldots,\beta_N on XLX_L, there exist a finite tuple c=(c1,…,cm)c=(c_1,\ldots,c_m) of elements of LL, a closed complete subfield E=K(c1,…,cm)‾⊂L,E=\overline{K(c_1,\ldots,c_m)}\subset L, and original compact smooth forms ηi\eta_i on XEX_E, of the same bidegrees as the βi\beta_i, such that βi=p∗ηi,p:XL⟶XE.\beta_i=p^*\eta_i,\qquad p:X_L\longrightarrow X_E. The equality is exact. The field EE is isometric to H(a)\mathscr H(a) for the point a∈AKm,ana\in\mathbf A_K^{m,\rm an} defined by evaluating its coordinates at c1,…,cmc_1,\ldots,c_m. In particular p∗Dηi=Dβip^*D\eta_i=D\beta_i.

Status: NEW_PROVED finite-coefficient descent. The analytic mechanism is the rational localization already proved in Lemma “Exact spreading from an algebraic parameter fiber”; the present argument extracts its finite coefficient field inside a given LL.

Proof. The union of the images in XX of the test supports is compact. Over the original field KK, choose a finite original smooth partition on this compact set, subordinate to ordinary affine algebraic neighborhoods supplied by B, Lemma D.4. Pulling it back to XLX_L reduces to one original test in an ordinary open X⊂YanX\subset Y^{\rm an}, with YY smooth affine over KK. Choose an original KK-smooth compact cutoff χX\chi_X equal to one near the image of its support. Cover supp⁡χX\operatorname{supp}\chi_X by finitely many compact polynomial or rational norm domains Dℓ⊂XD_\ell\subset X. Their extensions (Dℓ)L(D_\ell)_L are compact affinoid domains and provide bounds on every affine coordinate.

Apply the rational-localization part of Lemma “Exact spreading from an algebraic parameter fiber”, now with the complete field denoted FF there equal to LL. Here this means its explicit affinoid-density argument, not an invocation of R or B over LL. On finitely many slightly enlarged norm boxes, approximate each analytic unit in the test presentations by a rational expression with coefficients in LL, with error smaller than its positive minimum norm. This preserves its logarithm exactly on the box. Use smooth norm-profile bumps to localize there, with the following two protections: profiles involving possible polynomial zeros are constant below a positive threshold, and every actual unit numerator or denominator has an additional lower cutoff. The latter makes each localized term identically zero near its non-unit locus. Extend the real coefficient forms with real smooth cutoffs and add the finitely many localized terms. Multiplied by χX\chi_X, the result is exactly the original test everywhere on XLX_L, including outside its support. These are precisely the support and zero-locus steps justified in the cited lemma.

Only finitely many elements of LL are now used as coefficients of the rational moments and of the polynomials specifying their cutoffs. Include all of them, for all charts and all βi\beta_i, in the tuple cc. A KK-analytic function used by χX\chi_X needs no new coefficient: it has its scalar pullback over every intermediate field. The finite real cutoff thresholds and the real smooth coefficient functions are also unchanged by scalar extension.

Define ηi\eta_i over EE by the very same finite localized formulas. There is no new coefficient approximation in this step. The lower cutoffs ensure that every rational unit moment is defined wherever its localized term can be nonzero, over EE as well as over LL; profiles constant near zero ensure original smoothness across all other zero loci. Thus these formulas give genuine original forms, extended by zero on the stated ordinary opens. Their supports are contained in the inverse image of supp⁡χX\operatorname{supp}\chi_X, which is a closed subset of the finite union of compact affinoid domains (Dℓ)E(D_\ell)_E. They consequently have compact support on XEX_E. This proves compactness using scalar extensions of affinoid domains, without replacing analytic products by Cartesian products of point sets.

Scalar pullback of each formula recovers exactly the rationalized LL-test. This proves p∗ηi=βip^*\eta_i=\beta_i on each piece; extend by zero and add the pieces to obtain the global assertion. All finitely many coefficient lists can be united before taking EE.

Finally the seminorm P↦∣P(c1,…,cm)∣P\mapsto|P(c_1,\ldots,c_m)| on K[T1,…,Tm]K[T_1,\ldots,T_m] defines the point aa. Its prime-kernel quotient has fraction field K(c1,…,cm)K(c_1,\ldots,c_m), and completing inside the complete field LL gives exactly EE. This identifies EE isometrically with H(a)\mathscr H(a). Smooth pullback commutes with d′d' and d′′d'' by the original unit-moment definition, giving the last assertion. ◻

Proposition (The fixed Borel integration identity under scalar extension). Let E⊂LE\subset L be a complete extension, let ZZ be a good Hausdorff pure nn-dimensional EE-analytic space, and let p:ZL→Zp:Z_L\to Z be scalar base change. If η\eta is an original compact smooth (n,n)(n,n)-form and ff is a locally bounded Borel function on ZZ, then ∫ZL(f∘p)p∗η=∫Zfη.\int_{Z_L}(f\circ p)p^*\eta=\int_Z f\eta. There is no degree factor in this formula.

Status: EXTERNAL_KNOWN for the measure identity, and DIRECT_COROLLARY for its fixed Borel weight.

Proof. The canonical scalar-base-change calibration gives p∗μp∗η=μη;p_*\mu_{p^*\eta}=\mu_\eta; this is [CLD25, Proposition 15.3.1, p. 425], restricted to the common original test category. Apply the definition of pushforward to the fixed Borel function ff. Compact supports and local boundedness make the integrals finite. Equivalently one can first localize by a compact cutoff equal to one on the test support. This is an identity of fixed signed Radon measures; it makes no continuity assertion for changing measures or changing Borel weights. Its factor one is distinct from the finite-constant calibration on the same analytic space viewed over a smaller field. ◻

Theorem (Conditional reduction to algebraic parameter residue fields). Fix u,v∈Cb(X)u,v\in\mathcal C_b(X) on an allowed KK-analytic space, with D[u]=D[v]D[u]=D[v] on all original KK-tests. Assume the following precise additional property:

For every finite mm and every a∈AKm,ana\in\mathbf A_K^{m,\rm an}, the actual scalar pullbacks to XH(a)X_{\mathscr H(a)} satisfy D[uH(a)]=D[vH(a)]D[u_{\mathscr H(a)}]=D[v_{\mathscr H(a)}] on all original H(a)\mathscr H(a)-tests.

Then for every complete extension L/KL/K the scalar pullbacks satisfy D[uL]=D[vL]D[u_L]=D[v_L] on all original LL-tests. Thus scalar kernel transfer for all algebraic parameter residue fields is equivalent to scalar kernel transfer for all complete extensions, with these fixed X,u,vX,u,v.

Status: CONDITIONAL for the scalar kernel conclusion; the reduction itself follows from the two proved statements above. The additional property has not been proved for general non-Gauss parameter points.

Proof. Dimension zero is immediate. Otherwise, take any original compact LL-test β\beta of bidegree (n−1,n−1)(n-1,n-1). By Lemma “Exact finite-coefficient descent of original tests”, it equals p∗ηp^*\eta for an original compact test on XEX_E, where E≃H(a)E\simeq\mathscr H(a) for a finite-dimensional algebraic parameter point. Put f=u−vf=u-v and denote its actual scalar pullbacks by fEf_E and fLf_L. They are locally bounded Borel functions and fL=fE∘pf_L=f_E\circ p. Using the fixed Borel integration identity gives ⟨D[fL],β⟩=∫XLfLDβ=∫XL(fE∘p)p∗(Dη)=∫XEfEDη=⟨D[fE],η⟩=0.\begin{aligned} \langle D[f_L],\beta\rangle &=\int_{X_L}f_L D\beta =\int_{X_L}(f_E\circ p)p^*(D\eta)\\ &=\int_{X_E}f_E D\eta =\langle D[f_E],\eta\rangle=0. \end{aligned} The last equality is exactly the additional hypothesis. Since β\beta was arbitrary, the asserted original-current equality holds. Conversely the all-complete-extension assertion applies in particular to every H(a)\mathscr H(a), so gives that hypothesis. No BT operation or regularization theorem over EE or LL enters this reduction. ◻

The unresolved geometric gate is therefore already visible at residue fields of finite-dimensional algebraic parameter points. The Gauss and finite-over-Gauss cases proved earlier supply only part of this family. Exact descent of tests does not represent a general parameter fiber by a bounded ambient BT current, and density of Gauss points does not supply such a representation. This reduction asserts scalar base-change equalities only. It does not assert functoriality under every analytic morphism over an arbitrary extension, or turn the still open arbitrary-background potential quotient into a proved statement.

Two simultaneous-pasting mechanisms for relative capacity

The unrestricted ordinary-open Borel outer regularity problem remains open. We prove two precise statements that use the BT structure beyond weak compactness of a measure family. The first extends separation by analytic zero sets to complete pole sets of extended classical psh functions. The second pastes uniformly close normalized testers on finitely many separated compact pieces, using a bounded global classical compensator. The cost of the latter construction depends on the separating cutoffs; this dependence cannot be discarded.

The field is the fixed finite extension K/QpK/\mathbf Q_p with its original absolute value. All backgrounds are symmetric, original CLD-positive, and separately d′d'- and d′′d''-closed. Every current identity uses all original compact smooth complementary tests. Write q=n−p≥1q=n-p\geq1 and μS,h=TS,q(h,…,h),CS(E;Ω)=sup⁡h∈Cb(Ω)0≤h≤1μS,h(E).\mu_{S,h}=T_{S,q}(h,\ldots,h),\qquad C_S(E;\Omega)=\sup_{\substack{h\in\mathcal C_b(\Omega)\\0\leq h\leq1}} \mu_{S,h}(E). The raw capacity is defined on Borel sets; CS∗C_S^* always denotes its outer regularization by ordinary open neighborhoods in the same specified domain. Bounds on potentials mean actual pointwise bounds.

Separation by complete pole sets

Theorem (Finite pole-carrier additivity). Let Ω\Omega be an allowed ordinary open set, and let S1,…,SmS_1,\ldots,S_m have the same bidegree (p,p)(p,p). Suppose ϕi≤0\phi_i\leq0 are extended classical psh functions on Ω\Omega, with Pi={ϕi=−∞}P_i=\{\phi_i=-\infty\}, such that 1PiSi=Si,CSi(Pl;Ω)=0(i≠l).\mathbf1_{P_i}S_i=S_i, \qquad C_{S_i}(P_l;\Omega)=0\quad(i\ne l). The first equality concerns every coefficient measure. The pole sets need not be closed, analytic, or principal. Put S=∑iSiS=\sum_iS_i. For every Borel E⊂ΩE\subset\Omega and every arbitrary subset A⊂ΩA\subset\Omega, respectively, one has

CS(E;Ω)=∑iCSi(E;Ω),\begin{aligned} C_S(E;\Omega)&=\sum_i C_{S_i}(E;\Omega), \end{aligned}

CS∗(A;Ω)=∑iCSi∗(A;Ω).\begin{aligned} C_S^*(A;\Omega)&=\sum_i C_{S_i}^*(A;\Omega). \end{aligned}

Infinite values are allowed.

Status: NEW_PROVED for the extension to general complete pole carriers. The outer identity is a DIRECT_COROLLARY of the raw identity. Analytic unit-log invariance is not used in this extension.

Proof. Fix normalized actual hi∈Cb(Ω)h_i\in\mathcal C_b(\Omega) and 0<δ<1/20<\delta<1/2. For ϵ>0\epsilon>0 put ψi=∑l≠iϕl,bi=max⁡{0,δ+(1−2δ)hi+ϵψi},b=max⁡ibi.\psi_i=\sum_{l\ne i}\phi_l,\qquad b_i=\max\{0,\delta+(1-2\delta)h_i+\epsilon\psi_i\}, \qquad b=\max_i b_i. The empty sum is zero. Every bib_i and bb is an actual member of Cb(Ω)\mathcal C_b(\Omega), between zero and 1−δ1-\delta. Indeed, replace each ϕl\phi_l by ϕl,N=max⁡(ϕl,−N)\phi_{l,N}=\max(\phi_l,-N). The resulting bounded classical maxima decrease to the displayed functions and have the common lower bound zero. This uses only bounded decreasing closure, not an unbounded BT operator.

For R>0R>0 let Ai,R=Pi∩{ψi>−R},A_{i,R}=P_i\cap\{\psi_i>-R\}, and choose ϵR<δ/2\epsilon R<\delta/2. On Ai,RA_{i,R} the iith branch is greater than δ/2\delta/2. All other blb_l vanish there, since their penalties contain ϕi=−∞\phi_i=-\infty. Strict Borel maximum locality for the bounded functions b1,…,bmb_1,\ldots,b_m therefore gives

1Ai,RμSi,b=1Ai,RμSi,bi.\mathbf1_{A_{i,R}}\mu_{S_i,b} =\mathbf1_{A_{i,R}}\mu_{S_i,b_i}.

Neither Ai,RA_{i,R} nor {ψi>−R}\{\psi_i>-R\} is asserted ordinarily open.

We next prove the required measure lower bound for this branch. Set ci,N=δ+(1−2δ)hi+ϵ∑l≠iϕl,N,bi,N=max⁡(0,ci,N).c_{i,N}=\delta+(1-2\delta)h_i +\epsilon\sum_{l\ne i}\phi_{l,N}, \qquad b_{i,N}=\max(0,c_{i,N}). All these functions are bounded classical. Since truncation raises the penalties, ci,N>δ/2c_{i,N}>\delta/2 on Ai,RA_{i,R} for every NN. Borel maximum locality and positivity of the mixed expansion give 1Ai,RμSi,bi,N=1Ai,RμSi,ci,N≥(1−2δ)q1Ai,RμSi,hi.\mathbf1_{A_{i,R}}\mu_{S_i,b_{i,N}} =\mathbf1_{A_{i,R}}\mu_{S_i,c_{i,N}} \geq(1-2\delta)^q\mathbf1_{A_{i,R}}\mu_{S_i,h_i}. This is the point where positivity replaces the zero-curvature identity used for analytic logarithms in R3.

The sequence bi,N↓bib_{i,N}\downarrow b_i has a bounded limit. For a compact L⊂Ai,RL\subset A_{i,R}, decreasing BT continuity and the local compact-set weak inequality imply μSi,bi(L)≥lim sup⁡NμSi,bi,N(L)≥(1−2δ)qμSi,hi(L).\mu_{S_i,b_i}(L) \geq\limsup_N\mu_{S_i,b_{i,N}}(L) \geq(1-2\delta)^q\mu_{S_i,h_i}(L). One may insert a fixed original smooth cutoff equal to one near LL to apply the finite-measure weak inequality. Radon inner regularity now extends the inequality from compact subsets to every Borel subset of Ai,RA_{i,R}. Thus no unrestricted Borel restriction is passed through a varying-measure limit.

The pole restriction identity (eq:r2-polar-restriction-BT) shows that every μSi,hi\mu_{S_i,h_i} is carried by PiP_i, even when PiP_i is not closed. The cross-capacity hypothesis makes it vanish on PlP_l, l≠il\ne i. Consequently the increasing sets Ai,RA_{i,R} capture its entire mass as R→∞R\to\infty. By positivity and background linearity, CS(E;Ω)≥(1−2δ)q∑iμSi,hi(E∩Ai,R).C_S(E;\Omega) \geq(1-2\delta)^q\sum_i\mu_{S_i,h_i}(E\cap A_{i,R}). For each RR use a new admissible ϵ\epsilon; no convergence of these changing testers is invoked. Let R→∞R\to\infty, then δ↓0\delta\downarrow0, and take the independent finite suprema in the hih_i. This gives the lower bound in (eq:r4-pole-raw-sum). The upper bound follows immediately from μS,h=∑iμSi,h\mu_{S,h}=\sum_i\mu_{S_i,h}.

For the outer formula, one common open neighborhood first gives CS∗(A)≥∑iCSi∗(A)C_S^*(A)\geq\sum_iC_{S_i}^*(A). If that sum is finite, choose separate open neighborhoods nearly minimizing each summand and intersect the finitely many neighborhoods. The raw formula and monotonicity give the reverse inequality. If a summand is infinite, the earlier lower bound already proves equality. All domains remain Ω\Omega throughout. ◻

Corollary (A coefficient condition and countable sums). The cross-capacity condition of the theorem holds if instead 1PlSi=0(i≠l).\mathbf1_{P_l}S_i=0\qquad(i\ne l). Suppose a countable family satisfies these hypotheses pairwise, and its positive partial sums converge on all original compact tests to a locally coefficient-Radon positive closed background SS. Then the two capacity sum formulas hold on every fixed compact core, with a countable sum. If the component capacities are Borel outer regular, respectively finite Choquet capacities, on the required cores, these properties pass to SS.

Status: DIRECT_COROLLARY.

Proof. For the first assertion, pole restriction commutation gives 1PlμSi,h=μ1PlSi,h=0\mathbf1_{P_l}\mu_{S_i,h}=\mu_{\mathbf1_{P_l}S_i,h}=0 for every normalized hh. For the countable assertion apply the finite theorem to each partial sum and then use Proposition “Uniform capacity control under order approximation” on a compact core K0⊂W⋐ΩK_0\subset W\Subset\Omega. Its fixed-geometry CLN estimate makes the capacity of the positive current tail on WW tend to zero uniformly over subsets of K0K_0, both for raw and outer capacities. This proves the countable identities and transfers the stated regularity properties. No infinite maximum or infinite intersection of ordinary open sets is used. ◻

A global compensator for finitely many original cutoffs

In the rest of this section let YY be a smooth affine KK-variety and let Ω⋐Yan\Omega\Subset Y^{\mathrm{an}} be ordinary open. No globally strict psh exhaustion is assumed. This affine scope is used for global regular numerators and denominators in the next construction.

Lemma (A bounded classical compensator). For any finite family of original real compact smooth functions χ1,…,χm\chi_1,\ldots,\chi_m on Ω\Omega, there is a nonnegative bounded original smooth classical psh function ρ\rho on Ω\Omega such that ρ+2χi\rho+2\chi_i and ρ−2χi\rho-2\chi_i are classical psh for every ii. The function ρ\rho can be chosen as a finite positive sum of functions F(log⁡∣a∣+A),a∈K[Y],F(\log|a|+A),\qquad a\in K[Y], with finite real AA, where FF is one fixed smooth convex nondecreasing function that vanishes on (−∞,−1](-\infty,-1] and has F′′=2F''=2 on [0,∞)[0,\infty). Each displayed function is extended by zero at the zeros of aa. The compensator depends on the cutoffs and compact presentations, not on any positive current.

Status: DIRECT_COROLLARY of exact rational moment replacement in B and the classical curve chain rule, with the following explicit global construction.

Proof. Cover the union of the cutoff supports by finitely many smaller ordinary neighborhoods with compact closure, on each of which all needed functions have smooth unit-moment presentations. Use B’s exact rational-unit replacement, as recorded in Lemma “A countable family of coefficient controls”, to write these moments as log⁡∣a/b∣\log|a/b|, with a,b∈K[Y]a,b\in K[Y] nonvanishing on the respective compact stages. Replacement preserves their norms exactly on those neighborhoods, so it preserves the original smooth functions and their logarithmic differentials there.

Choose a nonnegative smooth bump of integral one supported in (−1,0)(-1,0), and integrate it twice from −∞-\infty to obtain a smooth convex nondecreasing ζ\zeta with ζ=0\zeta=0 on (−∞,−1](-\infty,-1] and ζ′=1\zeta'=1 on [0,∞)[0,\infty). Put F=ζ2F=\zeta^2. Then FF has all the stated properties. For every selected numerator or denominator choose AA such that ∣a∣>e−A|a|>e^{-A} on its compact stage, and set Fa=F(log⁡∣a∣+A),F_a=F(\log|a|+A), extended by zero near the zeros of aa. This is globally original smooth and classical psh on YanY^{\mathrm{an}}: off the zero set it is a smooth convex nondecreasing function of a unit logarithm, and it is identically zero on an ordinary neighborhood of the zero set. The classical assertion holds after every complete extension by the curve chain rule. On the chosen unit stage, DFa=2 d′log⁡∣a∣∧d′′log⁡∣a∣.DF_a=2\,d'\log|a|\wedge d''\log|a|.

On each presentation use the logarithms of its numerators and denominators as independent real coordinates. The old rational moments are their linear differences. A sufficiently large positive multiple of the sum of their squares has Hessian dominating the positive and negative of twice each cutoff Hessian on a neighborhood of the compact moment image. Equivalently one can use (dta−dtb)2≤2(dta2+dtb2)(dt_a-dt_b)^2\leq2(dt_a^2+dt_b^2). Choose a finite positive sum of the FaF_a providing this domination for each of the finitely many presentations, and call the sum ρ\rho.

On any one smaller neighborhood, its selected summands together with ±2χi\pm2\chi_i form a locally convex smooth function of harmonic unit logarithms. Their pullbacks to every complete-extension curve are subharmonic by the chain rule. The other summands of ρ\rho are already classical. This proves classicality of ρ±2χi\rho\pm2\chi_i on the neighborhood. Off the cutoff support the assertion is just classicality of ρ\rho, so the local statements glue. This argument does not invoke an unproved general implication from original-current positivity to classicality.

Every FaF_a is nonnegative and is bounded above on the compact Ω‾\overline\Omega, since aa is globally regular. Thus ρ\rho is nonnegative, original smooth and bounded on Ω\Omega, as required. ◻

Theorem (Pasting uniformly close normalized testers). Fix pairwise disjoint compact sets K1,…,Km⊂ΩK_1,\ldots,K_m\subset\Omega. There is a finite AKA_K, depending on chosen separating original smooth cutoffs and the fixed affine domain, with the following property. Let 0<δ≤10<\delta\leq1 and let ui∈Cb(Ω)u_i\in\mathcal C_b(\Omega) satisfy 0≤ui≤10\leq u_i\leq1 and ∣ui(x)−uj(x)∣≤δ(x∈Ω, 1≤i,j≤m).|u_i(x)-u_j(x)|\leq\delta \quad(x\in\Omega,\ 1\leq i,j\leq m). For every allowed fixed background SS there is one normalized v∈Cb(Ω)v\in\mathcal C_b(\Omega) such that for every choice of Borel Ei⊂KiE_i\subset K_i,

μS,v ⁣(⋃iEi)≥(1+AKδ)−q∑iμS,ui(Ei).\mu_{S,v}\!\left(\bigcup_iE_i\right) \geq(1+A_K\delta)^{-q}\sum_i\mu_{S,u_i}(E_i).

The same vv works for all the subsets EiE_i. Its construction and AKA_K are independent of SS. As δ→0\delta\to0, if all uiu_i converge uniformly to one function, the constructed vv has the same uniform limit.

Status: NEW_PROVED.

Proof. Choose 0≤χi≤10\leq\chi_i\leq1 original smooth with pairwise disjoint compact supports, each equal to one on an ordinary neighborhood NiN_i of KiK_i. The disjoint compact sets admit such choices by ordinary smooth partitions of unity. Let ρ\rho be the preceding compensator, put M=sup⁡ΩρM=\sup_\Omega\rho, and set ψi=ρ+2χi,AK=M+2,w=max⁡i(ui+δψi),v=w1+AKδ.\psi_i=\rho+2\chi_i,\quad A_K=M+2,\quad w=\max_i(u_i+\delta\psi_i),\quad v=\frac{w}{1+A_K\delta}. Every ψi\psi_i is classical and lies in [0,AK][0,A_K], so ww is classical and 0≤w≤1+AKδ0\leq w\leq1+A_K\delta. Thus vv is normalized. On NiN_i, one has χi=1\chi_i=1 and χj=0\chi_j=0 for j≠ij\ne i. Consequently (ui+δψi)−(uj+δψj)≥−δ+2δ=δ>0.(u_i+\delta\psi_i)-(u_j+\delta\psi_j) \geq-\delta+2\delta=\delta>0. The iith branch is therefore uniquely active on this prescribed ordinary open neighborhood, despite possible discontinuity of the uiu_i. Ordinary maximum locality and positivity of the mixed expansion give there μS,v=(1+AKδ)−q(D(ui+δψi))qS≥(1+AKδ)−qμS,ui.\mu_{S,v} =(1+A_K\delta)^{-q}(D(u_i+\delta\psi_i))^qS \geq(1+A_K\delta)^{-q}\mu_{S,u_i}. The disjointness of the EiE_i proves (eq:r4-close-tester-pasting). All current assertions are identities or inequalities on the full original test sheaf before their Radon restrictions are taken. Finally the correction and normalizing factor differ uniformly from zero and one by O(AKδ)O(A_K\delta), which proves the uniform-limit assertion. ◻

Proposition (Pasting outside a small capacity error). Keep the domain, disjoint compact sets, cutoffs and constant AKA_K of Theorem “Pasting uniformly close normalized testers”. Let u1,…,umu_1,\ldots,u_m be any normalized classical testers; no uniform closeness is assumed. For 0<δ≤10<\delta\leq1 put Bi={max⁡l(ul−ui)≥2δ},v=max⁡i(ui+δ(ρ+2χi))1+AKδ.B_i=\left\{\max_l(u_l-u_i)\geq2\delta\right\},\qquad v=\frac{\max_i(u_i+\delta(\rho+2\chi_i))}{1+A_K\delta}. Then vv is normalized classical and, for every Borel Ei⊂KiE_i\subset K_i,

μS,v ⁣(⋃iEi)≥(1+AKδ)−q∑i(μS,ui(Ei)−CS∗(Ki∩Bi;Ω)).\mu_{S,v}\!\left(\bigcup_iE_i\right) \geq(1+A_K\delta)^{-q} \sum_i\left(\mu_{S,u_i}(E_i) -C_S^*(K_i\cap B_i;\Omega)\right).

Status: DIRECT_COROLLARY of the same explicit maximum construction and strict Borel locality.

Proof. The bounds and classicality of vv did not use closeness of the uiu_i. On Ki∖BiK_i\setminus B_i its iith branch is strictly larger than every other branch. Strict Borel maximum locality and mixed positivity give the previous measure lower bound on that Borel set. Since uiu_i is normalized on the entire domain, μS,ui(Ei∩Bi)≤CS(Ki∩Bi;Ω)≤CS∗(Ki∩Bi;Ω).\mu_{S,u_i}(E_i\cap B_i) \leq C_S(K_i\cap B_i;\Omega) \leq C_S^*(K_i\cap B_i;\Omega). Subtract this possible loss and sum over the disjoint EiE_i. All capacities and test functions have the same domain Ω\Omega. ◻

Corollary (Simultaneous extraction near a capacity limit). Let KiK_i be fixed disjoint compacts and Ei⊂KiE_i\subset K_i Borel. For each ii, let ui,ju_{i,j} be normalized classical testers. Suppose all these rows converge to one actual Borel function aa in local outer SS-capacity with domain Ω\Omega on ⋃iKi\bigcup_iK_i. Then CS ⁣(⋃iEi;Ω)≥∑ilim sup⁡jμS,ui,j(Ei).C_S\!\left(\bigcup_iE_i;\Omega\right) \geq\sum_i\limsup_j\mu_{S,u_{i,j}}(E_i).

Proof. Choose a subsequence separately for each of the finitely many rows so that its mass on its own EiE_i tends to its limsup. Relabel these selected rows with the same index. Each still converges in capacity to aa on every KlK_l. For fixed δ>0\delta>0, Ki∩Bi,j⊂⋃l(Ki∩{∣ul,j−a∣≥δ}) ∪ (Ki∩{∣ui,j−a∣≥δ}).K_i\cap B_{i,j}\subset \bigcup_l\bigl(K_i\cap\{|u_{l,j}-a|\geq\delta\}\bigr) \ \cup\ \bigl(K_i\cap\{|u_{i,j}-a|\geq\delta\}\bigr). Finite subadditivity makes all error capacities tend to zero. Apply Proposition “Pasting outside a small capacity error”, first let j→∞j\to\infty with δ\delta fixed, and then let δ↓0\delta\downarrow0. CLN bounds all displayed masses on these fixed compact sets. No convergence on the arbitrary Borel EiE_i is assumed or deduced from weak convergence of measures. ◻

Corollary (Simultaneous mass extraction near one uniform limit). Let KiK_i be fixed disjoint compacts and Ei⊂KiE_i\subset K_i Borel. For each ii, let ui,ju_{i,j} be normalized classical testers converging uniformly on Ω\Omega to the same function. Then CS ⁣(⋃iEi;Ω)≥∑ilim sup⁡jμS,ui,j(Ei).C_S\!\left(\bigcup_iE_i;\Omega\right) \geq\sum_i\limsup_j\mu_{S,u_{i,j}}(E_i).

Proof. For each of the finitely many indices choose a subsequence attaining its limsup; all resulting rows still have the same uniform limit. Their maximal pairwise distance tends to zero. Apply the theorem to each finite row and let its index tend to infinity. CLN makes all masses on these fixed compacts finite and uniformly bounded. ◻

The full BT tester class repairs the vertical-fiber subfamily

Example (An actual toric subfamily and its missing testers). On the original two-dimensional split torus take Ω={−2<x=log⁡∣T1∣<2, −2<y=log⁡∣T2∣<2},S=[d′x∧d′′x+d′y∧d′′y].\Omega=\{-2<x=\log|T_1|<2,\ -2<y=\log|T_2|<2\},\qquad S=[d'x\wedge d''x+d'y\wedge d''y]. For 0≤t≤10\leq t\leq1 define the continuous normalized classical tester ut=∣x−t∣2+t.u_t=\frac{|x-t|}{2+t}. Let σ\sigma denote the coordinate Gauss section. The ordinary tropical calibration, valid on all original tests, gives D(utS)=22+t σ∗(δt⊗dy).D(u_tS)=\frac{2}{2+t}\, \sigma_*(\delta_t\otimes dy). Consider the Borel set E=σ ⁣({0}×[0,12] ∪ ⋃j≥1{1/j}×(12,1]).E=\sigma\!\left( \{0\}\times[0,\tfrac12] \ \cup\ \bigcup_{j\geq1}\{1/j\}\times(\tfrac12,1] \right). For the measure subfamily generated only by the utu_t, the raw capacity of EE is 1/21/2 and its ordinary-open outer capacity is 11. For the full relative BT capacity one instead has CS(E;Ω)≥1.C_S(E;\Omega)\geq1. Thus this embedded vertical-fiber subfamily is not a counterexample to Borel outer regularity of the full BT capacity.

Proof. The current formula follows by smoothing the convex absolute value: its one-variable second derivative has mass 2/(2+t)2/(2+t), and its wedge with SS keeps only the d′y∧d′′yd'y\wedge d''y term. Original product calibration on the split torus has degree one; the same common-refinement argument used in (10) identifies the actual scalar test value on the Gauss section. Bounded BT convergence gives the displayed ordinary-current equality.

Every nonempty fiber of EE has length 1/21/2, and the coefficient 2/(2+t)2/(2+t) is at most one, with equality at zero. This proves the subfamily raw value. Any ordinary open set containing EE restricts to an open set on the Gauss box containing its displayed subset. Compactness of the lower half of the zero fiber forces the entire 1/j1/j fiber over y∈[0,1]y\in[0,1] into that open set for large jj. Their coefficients tend to one, so the outer value is at least one. Conversely use open covers lying over −η<y<1+η-\eta<y<1+\eta; every subfamily measure of such a cover is at most 1+2η1+2\eta. Letting η↓0\eta\downarrow0 gives the upper bound.

Fix 0<η<1/20<\eta<1/2. The compacts K1=σ({0}×[0,12]),K2=σ ⁣(({0}∪{1/j:j≥1})×[12+η,1])K_1=\sigma(\{0\}\times[0,\tfrac12]),\qquad K_2=\sigma\!\left((\{0\}\cup\{1/j:j\geq1\}) \times[\tfrac12+\eta,1]\right) are disjoint and lie strictly inside Ω\Omega. Take Ei=E∩KiE_i=E\cap K_i. One has ∥ut−u0∥∞≤t\|u_t-u_0\|_\infty\leq t. Apply Corollary “Simultaneous mass extraction near one uniform limit” to the constant row u0u_0 for E1E_1 and the row u1/ju_{1/j} for E2E_2. It gives CS(E;Ω)≥12+(12−η)=1−η.C_S(E;\Omega)\geq\tfrac12+(\tfrac12-\eta)=1-\eta. Let η↓0\eta\downarrow0. The added testers are actual normalized classical maxima with small positive psh corrections; the proof does not replace the capacity by the weak closure of the displayed subfamily. No exact upper bound for the full capacity is claimed. ◻

A necessary dependence and the unresolved extraction gate

The separating-cutoff cost in Theorem “Pasting uniformly close normalized testers” cannot be bounded solely by the ambient domain. On the annulus −2<log⁡∣T∣<2-2<\log|T|<2, choose an original smooth 0≤χϵ≤10\leq\chi_\epsilon\leq1 whose values at the Gauss radii 11 and eϵe^\epsilon are zero and one, respectively. If 0≤ρ≤M0\leq\rho\leq M and ρ+2χϵ\rho+2\chi_\epsilon are classical, their restrictions to the inner Gauss interval are convex and have Lipschitz bounds at most MM and M+2M+2, respectively. These are the elementary inner secant bounds from the proof of Lemma “Uniform continuity on an inner Gauss box”. For 0<ϵ<10<\epsilon<1, 2=2∣χϵ(eϵ)−χϵ(1)∣≤(2M+2)ϵ,M≥ϵ−1−1.2=2|\chi_\epsilon(e^\epsilon)-\chi_\epsilon(1)| \leq(2M+2)\epsilon, \qquad M\geq\epsilon^{-1}-1. Thus a uniform separation-independent compensator estimate is FALSE. Here values at radii mean values at their original Gauss points; no field extension is used.

For an arbitrary Borel EE in a fixed compact core, a sufficient next step would be the following quantitative extraction statement: from near-maximizers on ordinary neighborhoods of EE, obtain finitely many pairwise disjoint compact pieces KiK_i, normalized testers uiu_i with pairwise uniform distance at most δ\delta, and ∑iμS,ui(E∩Ki)≥CS∗(E;Ω)−η,AKδ⟶0(η↓0).\sum_i\mu_{S,u_i}(E\cap K_i) \geq C_S^*(E;\Omega)-\eta, \qquad A_K\delta\longrightarrow0\quad(\eta\downarrow0). The proved pasting inequality would then give CS(E)=CS∗(E)C_S(E)=C_S^*(E). This extraction statement is OPEN. Neither abstract weak compactness nor the R3 convex-block theorem supplies it: the latter does not preserve the unaveraged nonlinear measures, and the cutoff cost can diverge when the pieces move together.

The pole-carrier theorem supplies a separate sufficient mechanism, including nonanalytic pole supports, but it requires the stated mutual null conditions. It does not separate an arbitrary non-pluripolar summand from a nested pluripolar summand at no cost. No previously frozen R3 theorem is changed or contradicted here.

A complete strong image for a prescribed relative collar class

The background is arbitrary and fixed. We retain the prescribed finite extension K/QpK/\mathbf Q_p, its absolute value, the original positive cone, and all original compact smooth complementary CLD tests. Let Ω\Omega be an ordinary algebraic norm-box domain relatively compact in a smooth affine analytification of pure dimension nn. Let SS be symmetric, original-positive, separately d′d'- and d′′d''-closed, of bidegree (p,p)(p,p), and put q=n−p≥1q=n-p\geq1.

Fix a compact L⋐ΩL\Subset\Omega contained in a strictly smaller norm box. Fix a smooth classical psh function hh on a neighborhood of Ω‾\overline\Omega, with semipositive original curvature form, and a number B>0B>0 such that ∣h∣≤B|h|\leq B on Ω\Omega. Define Ph,LB={u∈Cb(Ω):∣u∣≤B, u=h on Ω∖L}.\mathscr P_{h,L}^{B} =\{u\in\mathcal C_b(\Omega): |u|\leq B,\ u=h\text{ on }\Omega\setminus L\}. The exact equality here is an equality of actual point values. In particular the family is nonempty and convex.

Definition (The completed admissible collar class). Let DS,h,LB\mathscr D_{S,h,L}^{B} be the set of local relative quasi-everywhere classes having a bounded Borel representative aa which is a limit of a sequence um∈Ph,LBu_m\in\mathscr P_{h,L}^{B} locally in outer SS-capacity on every permitted compact/domain pair in Ω\Omega. The bound and the collar are the same for all members of each defining sequence and for the entire class. We use representatives with ∣a∣≤B|a|\leq B and a=ha=h on Ω∖L\Omega\setminus L.

Such representatives always exist: the summably Cauchy subsequence construction in the bounded capacity-completion theorem uses the actual limsup of defining rows, which has both stated properties. This is a closure of a specified global family on Ω\Omega. It is not identified with every locally defined member of the sheaf AS,b\mathcal A_{S,b} having an informal boundary value.

For a∈DS,h,LBa\in\mathscr D_{S,h,L}^{B} write μa=TS,q(a,…,a)\mu_a=\mathcal T_{S,q}(a,\ldots,a). The completed calculus already defines this positive Radon measure. On the ordinary open Ω∖L\Omega\setminus L, it equals μh\mu_h by locality. Thus μa−μb\mu_a-\mu_b is a finite signed measure carried by LL, even when the individual measures have infinite total mass on Ω\Omega. Put dS(a,b)=sup⁡v∈Cb(Ω)0≤v≤1∣∫v d(μa−μb)∣,IS(a,b)=∫(a−b) d(μb−μa).d_S(a,b)= \sup_{\substack{v\in\mathcal C_b(\Omega)\\0\leq v\leq1}} \left|\int v\,d(\mu_a-\mu_b)\right|, \qquad I_S(a,b)=\int(a-b)\,d(\mu_b-\mu_a). The latter integral is finite and independent of the representatives: the completed weighted calculus and coefficient outer-capacity control annihilate their local relative-null changes.

Lemma (The energy inequalities pass to the completed class). There is a finite constant depending only on SS, the fixed finite CLN controls, the compact collar stages, BB and qq such that every a,b∈DS,h,LBa,b\in\mathscr D_{S,h,L}^{B} satisfy

0≤IS(a,b),dS(a,b)≤CIS(a,b)1/2,IS(a,b)≤4B dS(a,b),\begin{aligned} 0\leq I_S(a,b),\qquad d_S(a,b)&\leq C I_S(a,b)^{1/2}, &I_S(a,b)&\leq4B\,d_S(a,b),\end{aligned}

Cap⁡S∗({∣a−b∣>δ};Ω)≤Cδ−2IS(a,b)2−q(δ>0).\begin{aligned} \operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega) &\leq C\delta^{-2}I_S(a,b)^{2^{-q}} &&(\delta>0). \end{aligned}

The gap sets in the second line lie in LL for the chosen representatives. In particular

Cap⁡S∗({∣a−b∣>δ};Ω)≤Cδ−2(4B dS(a,b))2−q.\operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega) \leq C\delta^{-2} \bigl(4B\,d_S(a,b)\bigr)^{2^{-q}} .

Status: DIRECT_COROLLARY of the proved R3 energy inequalities, the completed weighted calculus, and completed strong convergence. No new energy of an arbitrary Borel difference is assumed.

Proof. Choose defining rows um,vmu_m,v_m in the same Ph,LB\mathscr P_{h,L}^{B} for a,ba,b. Let χ\chi be original smooth, compactly supported in Ω\Omega, and equal to one near LL. The completed weighted theorem gives ∫χ(um−vm)(μvm−μum)⟶∫χ(a−b)(μb−μa)=IS(a,b).\int\chi(u_m-v_m)(\mu_{v_m}-\mu_{u_m}) \longrightarrow \int\chi(a-b)(\mu_b-\mu_a)=I_S(a,b). Each integral on the left is the ordinary compact-collar energy. R3 strong convergence to the completed measures gives ∥μum−μa∥psh,Ω→0,∥μvm−μb∥psh,Ω→0.\|\mu_{u_m}-\mu_a\|_{\mathrm{psh},\Omega}\to0,\qquad \|\mu_{v_m}-\mu_b\|_{\mathrm{psh},\Omega}\to0. Indeed the measure differences are carried by LL, so insertion of χ\chi does not change them. Consequently dS(um,vm)→dS(a,b)d_S(u_m,v_m)\to d_S(a,b). Passing the original two energy inequalities to the limit proves (eq:r4-completed-energy-dual). In particular the reverse inequality has not treated (a+B)/(2B)(a+B)/(2B) as a classical tester; it was proved before passing to the completed limit.

Fix 0<2ϵ<δ0<2\epsilon<\delta. The gap {∣a−b∣>δ}\{|a-b|>\delta\} is contained in the union of L∩{∣a−um∣>ϵ},L∩{∣b−vm∣>ϵ},{∣um−vm∣>δ−2ϵ}.L\cap\{|a-u_m|>\epsilon\},\quad L\cap\{|b-v_m|>\epsilon\},\quad \{|u_m-v_m|>\delta-2\epsilon\}. The first two sets have outer capacity tending to zero. Apply the original energy-capacity bound to the third, then let m→∞m\to\infty and ϵ↓0\epsilon\downarrow0. Outer subadditivity is sufficient; no continuity from below or compact identification of raw and outer capacity is used. This proves (eq:r4-completed-energy-capacity), hence the last estimate. ◻

Proposition (Capacity completeness with an actual collar). The class DS,h,LB\mathscr D_{S,h,L}^{B} is complete for local outer SS-capacity convergence modulo local relative quasi-everywhere equality. Every limit has an actual bounded representative equal to hh on Ω∖L\Omega\setminus L. Therefore its ordinary boundary limit on ∂Ω\partial\Omega is h∣∂Ωh|_{\partial\Omega}. If a fixed continuous lower and upper bound are imposed on all defining rows, they can be retained by an actual limit representative as well.

Proof. Use the countably cofinal compact/domain pairs of the R2 completion theorem. For a capacity-Cauchy sequence aja_j in this class, choose a defining uj∈Ph,LBu_j\in\mathscr P_{h,L}^{B} whose capacity error from aja_j, at threshold 2−j2^{-j}, is at most 2−j2^{-j} on the first jj pairs. The level-set triangle inequality makes (uj)(u_j) capacity Cauchy on every pair. The existing bounded completion theorem gives a Borel limit. Since all uju_j belong to this one admissible family, the limit belongs to DS,h,LB\mathscr D_{S,h,L}^{B}. Its actual limsup representative from the summable subsequence is bounded by BB and equals hh outside LL. The same triangle inequality identifies it as the limit of aja_j and proves uniqueness modulo the stated relative-null sets.

A compact subset of Ω\Omega cannot meet a sufficiently small neighborhood of a prescribed point of ∂Ω\partial\Omega. Thus this representative equals the continuous hh near every ordinary boundary point, proving the boundary assertion. Any fixed pointwise bounds on the defining rows pass to the same actual limsup. This last assertion concerns the selected representative; arbitrary changes on null sets need not preserve ordinary boundary limits. ◻

Theorem (A closed complete strong measure image). Let Mh,L\mathfrak M_{h,L} be the positive Radon measures on Ω\Omega which equal μh\mu_h on Ω∖L\Omega\setminus L, equipped with dpsh,Ω(ν,ν′)=sup⁡v∈Cb(Ω)0≤v≤1∣∫v d(ν−ν′)∣.d_{\mathrm{psh},\Omega}(\nu,\nu') =\sup_{\substack{v\in\mathcal C_b(\Omega)\\0\leq v\leq1}} \left|\int v\,d(\nu-\nu')\right|. This is a metric on Mh,L\mathfrak M_{h,L}. The image IS,h,LB={μa:a∈DS,h,LB}\mathfrak I_{S,h,L}^{B} =\{\mu_a:a\in\mathscr D_{S,h,L}^{B}\} is complete for this metric and hence closed in Mh,L\mathfrak M_{h,L}. Every measure in the image has a unique solution class in DS,h,LB\mathscr D_{S,h,L}^{B}. More precisely, if μaj\mu_{a_j} is Cauchy in this metric, there is a unique relative quasi-everywhere class aa in the same admissible completion such that aj→a locally in outer S-capacity,dpsh,Ω(μaj,μa)→0.a_j\to a\text{ locally in outer }S\text{-capacity}, \qquad d_{\mathrm{psh},\Omega}(\mu_{a_j},\mu_a)\to0. If the data have a prescribed metric limit ν∈Mh,L\nu\in\mathfrak M_{h,L}, then μa=ν\mu_a=\nu. No ambient classical representative is asserted.

Status: DIRECT_COROLLARY of the preceding completed estimates and R2 completeness. This is a solution-image theorem for the specified admissible class, not an existence criterion for arbitrary dominated measures.

Proof. First verify that the displayed seminorm separates the measures. Their difference is a finite signed Radon measure supported on LL. Bounded continuous classical psh functions on Ω\Omega include constants and suitable finite truncations of logarithms of affine regular functions. Such functions separate the points of the compact set LL. Their differences form a real vector lattice, since max⁡(a−b,c−d)=max⁡(a+d,c+b)−(b+d).\max(a-b,c-d)=\max(a+d,c+b)-(b+d). The lattice Stone–Weierstrass theorem gives density in C(L)C(L). Every bounded psh function in this lattice construction can be shifted and positively scaled into [0,1][0,1] on Ω\Omega. Consequently distance zero annihilates a dense set of continuous tests and hence annihilates the signed measure. This proves the metric assertion.

Suppose the image measures are metric Cauchy. Estimate (eq:r4-data-capacity-inverse) makes the aja_j Cauchy in capacity on the fixed domain Ω\Omega. Reverse framed domain comparison on compact cores gives the required all-domain local Cauchy condition. Proposition “Capacity completeness with an actual collar” produces aa in the same class. The completed strong convergence theorem gives convergence of χμaj\chi\mu_{a_j} to χμa\chi\mu_a in the same psh-dual seminorm, for a compact χ=1\chi=1 near LL. Outside LL the measures are already equal to μh\mu_h. Thus the asserted global difference seminorm tends to zero. This proves completeness without presuming that Mh,L\mathfrak M_{h,L} itself is complete.

If μa=μb\mu_a=\mu_b, then dS(a,b)=0d_S(a,b)=0 and the energy-capacity estimate gives zero outer capacity to every positive gap. Countable subadditivity and fixed-domain null-class comparison show that a=ba=b locally relative quasi-everywhere. The converse is the invariance of the completed calculus. Thus the solution class is unique. A prescribed metric limit agrees with μa\mu_a by the metric separation just proved. Completeness implies closedness, or the same argument applies directly to any convergent sequence of image measures. ◻

Scope of the boundary and of existence.

All defining rows are defined on the same full Ω\Omega, obey one common bound, and agree with one actual collar. This is stronger than merely sharing a boundary limit. The theorem neither supplies a classical representative nor proves a bounded-subsolution existence theorem for arbitrary SS. For q=1q=1 the admissible family and its completion are convex and the measure map is linear, so this particular closed image is convex as well. For higher degree no convexity of its nonlinear image is asserted. The arguments use no arbitrary varying-background continuity and no relative obstacle orthogonality.

A closed interior Dirichlet image with fixed actual exterior values

Retain the arbitrary fixed SS, degree q=n−p≥1q=n-p\geq1, norm-box domain Ω\Omega, smooth outer collar hh, compact LL and bound B>0B>0 of Section “A complete strong image for a prescribed relative collar class”. Choose an ordinary open VV with V‾⊂L⋐Ω\overline V\subset L\Subset\Omega, and a compact K⋐VK\Subset V. Let ee be a fixed bounded actual function on Ω∖V\Omega\setminus V, equal to hh off LL. In particular ee is specified at every point of ∂V\partial V.

Fix continuous functions ℓ,r\ell,r on V‾\overline V with ℓ≤r\ell\leq r and ℓ∣∂V=r∣∂V=e∣∂V=g.\ell|_{\partial V}=r|_{\partial V} =e|_{\partial V}=g. Assume that the following class is nonempty: Pbd={u∈Ph,LB:u=e on Ω∖V, ℓ≤u≤r on V}.\mathscr P_{\rm bd} =\{u\in\mathscr P_{h,L}^{B}: u=e\text{ on }\Omega\setminus V,\ \ell\leq u\leq r\text{ on }V\}. All equalities and inequalities in this definition are actual. No assertion of nonemptiness for arbitrary data or barriers is implicit in the notation.

Let Dbd\mathscr D_{\rm bd} be the closure of this global family in the all-domain local outer SS-capacity sense, modulo local relative quasi-everywhere equality. Its representatives are selected by the actual limsup construction from defining rows. Thus they satisfy ∣a∣≤B|a|\leq B, a=ea=e at every point of Ω∖V\Omega\setminus V, and ℓ≤a≤r\ell\leq a\leq r on VV. In particular every such selected representative has ordinary boundary value gg. Define Dbd,K={a∈Dbd:νa:=1Vμa is carried by K},μa=TS,q(a,…,a).\mathscr D_{{\rm bd},K} =\{a\in\mathscr D_{\rm bd}: \nu_a:=\mathbf1_V\mu_a \text{ is carried by }K\}, \qquad \mu_a=\mathcal T_{S,q}(a,\ldots,a). The interior measure νa\nu_a is finite. We do not require μa(∂V)=0\mu_a(\partial V)=0, or require the boundary restrictions of μa\mu_a to agree for different members.

Lemma (The boundary part disappears from the energy multiplier). For a,b∈Dbd,Ka,b\in\mathscr D_{{\rm bd},K}, the completed collar energy satisfies

IS(a,b)=∫K(a−b) d(νb−νa),0≤IS(a,b)≤4B dpsh,V(νa,νb),I_S(a,b)=\int_K(a-b)\,d(\nu_b-\nu_a), \qquad 0\leq I_S(a,b) \leq4B\,d_{\mathrm{psh},V}(\nu_a,\nu_b),

where dpsh,V(ν,ν′)=sup⁡v∈Cb(V)0≤v≤1∣∫Kv d(ν−ν′)∣.d_{\mathrm{psh},V}(\nu,\nu') =\sup_{\substack{v\in\mathcal C_b(V)\\0\leq v\leq1}} \left|\int_K v\,d(\nu-\nu')\right|. Consequently

Cap⁡S∗({∣a−b∣>δ};Ω)≤Cδ−2(4B dpsh,V(νa,νb))2−q.\operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega) \leq C\delta^{-2} \bigl(4B\,d_{\mathrm{psh},V}(\nu_a,\nu_b)\bigr)^{2^{-q}}.

The constants have the same fixed dependencies as the outer-collar energy-capacity estimate.

Status: DIRECT_COROLLARY.

Proof. For the selected representatives a−ba-b is exactly zero on Ω∖V\Omega\setminus V, including every point of ∂V\partial V. The difference μb−μa\mu_b-\mu_a is a finite signed measure on the fixed compact LL. Borel multiplication therefore gives (a−b)(μb−μa)=1V(a−b)(μb−μa)=(a−b)(νb−νa),(a-b)(\mu_b-\mu_a) =\mathbf1_V(a-b)(\mu_b-\mu_a) =(a-b)(\nu_b-\nu_a), as equality of finite signed measures. This is not a statement that the boundary measures themselves vanish. The completed weighted calculus makes each product invariant under relative-null modifications, but the displayed cancellation uses the chosen actual equality on the boundary, not an unsupported representative convention. Integration proves the first formula. Nonnegativity follows from Lemma “The energy inequalities pass to the completed class”.

Choose global defining rows um,vm∈Pbdu_m,v_m\in\mathscr P_{\rm bd} for a,ba,b. On the compact KK, each of the fixed measures νa,νb\nu_a,\nu_b is uniformly absolutely continuous with respect to the relative outer capacity with domain VV. Indeed the rows restricted to VV have the common bound BB; normalizing them into [0,1][0,1] and using the completed coefficient outer-capacity bound yields νc(E)≤(2B)qCap⁡S∗(E;V),E⊂K Borel,c=a,b.\nu_c(E)\leq (2B)^q \operatorname{Cap}_S^*(E;V), \qquad E\subset K\text{ Borel},\quad c=a,b. This conclusion can also be obtained from the open-set weak-limit inequality followed by the outer regularity of each fixed Radon measure. It uses no compact equality of raw and outer BT capacity. Capacity convergence and the uniform bounds therefore imply ∫K(um−vm) d(νb−νa)⟶∫K(a−b) d(νb−νa).\int_K(u_m-v_m)\,d(\nu_b-\nu_a) \longrightarrow\int_K(a-b)\,d(\nu_b-\nu_a). Both (um+B)/(2B)(u_m+B)/(2B) and (vm+B)/(2B)(v_m+B)/(2B) are legitimate normalized classical testers on VV. Their constant terms cancel in the difference, so the absolute value of each integral on the left is at most 4B dpsh,V(νa,νb)4B\,d_{\mathrm{psh},V}(\nu_a,\nu_b). Passing to the limit proves (eq:r4-interior-energy). Apply the already proved completed energy-capacity estimate to obtain (eq:r4-interior-inverse-capacity). ◻

Theorem (Completeness and closedness of the interior image). Let MK+(V)\mathfrak M_K^+(V) be the finite positive Radon measures carried by KK, equipped with dpsh,Vd_{\mathrm{psh},V}. This is a metric space. The image Ibd,K={νa:a∈Dbd,K}\mathfrak I_{{\rm bd},K} =\{\nu_a:a\in\mathscr D_{{\rm bd},K}\} is complete, and hence closed, in that metric space. Every datum in this image has a unique local relative quasi-everywhere solution class in Dbd,K\mathscr D_{{\rm bd},K}.

More precisely, if νaj\nu_{a_j} is Cauchy in this metric, then for a unique class a∈Dbd,Ka\in\mathscr D_{{\rm bd},K}, aj⟶alocally in outer S-capacity on every permitted pair in Ω,dpsh,V(νaj,νa)⟶0.a_j\longrightarrow a \quad\text{locally in outer }S\text{-capacity on every permitted pair in }\Omega, \qquad d_{\mathrm{psh},V}(\nu_{a_j},\nu_a)\longrightarrow0. The selected limit representative retains the actual exterior ee, the two continuous barriers, and the ordinary boundary value gg. If the given data already converge in the metric to ν∈MK+(V)\nu\in\mathfrak M_K^+(V), then νa=ν\nu_a=\nu.

Status: DIRECT_COROLLARY of the completed energy estimate, bounded capacity completeness, and completed localized strong convergence.

Proof. Bounded continuous classical functions on VV obtained by truncating logarithms of affine regular functions separate points of KK. Their differences form a vector lattice containing constants. The same lattice Stone–Weierstrass argument as in Theorem “A closed complete strong measure image” shows that dpsh,Vd_{\mathrm{psh},V} separates finite signed measures carried by KK. It is finite because every tester lies between zero and one. Thus it is a metric.

The class Dbd\mathscr D_{\rm bd} is capacity complete. For a Cauchy sequence in this class, choose a diagonal sequence of its defining rows in the one family Pbd\mathscr P_{\rm bd}, controlling the first jj compact/domain pairs at step jj. The existing capacity completion then produces a limit defined by those rows. Their actual limsup preserves ee at all points outside VV, the bound BB, and the barriers. Since the barriers are continuous and agree at each boundary point, they give the asserted ordinary boundary limit. This is the argument of Proposition “Capacity completeness with an actual collar” applied to the more restricted defining family.

For the sequence in the theorem, (eq:r4-interior-inverse-capacity) first gives capacity Cauchyness on Ω\Omega. Reverse framed domain comparison on compact cores gives the all-domain local condition. Hence there is a∈Dbda\in\mathscr D_{\rm bd} with the stated capacity convergence. Completed weak convergence on VV shows that μa\mu_a vanishes on the ordinary open V∖KV\setminus K: all the positive measures μaj\mu_{a_j} vanish there. Thus a∈Dbd,Ka\in\mathscr D_{{\rm bd},K}.

Choose an original compact smooth χ\chi on VV equal to one near KK. Then χμaj=νaj\chi\mu_{a_j}=\nu_{a_j} and χμa=νa\chi\mu_a=\nu_a. Completed localized strong convergence with the fixed background S∣VS|_V gives sup⁡v∈Cb(V)0≤v≤1∣∫vχ d(μaj−μa)∣⟶0.\sup_{\substack{v\in\mathcal C_b(V)\\0\leq v\leq1}} \left|\int v\chi\,d(\mu_{a_j}-\mu_a)\right|\longrightarrow0. This is exactly the required metric convergence of the interior data. In particular variable boundary measures of the global extensions never enter this final passage: the cutoff has support strictly inside VV. This proves completeness. It also identifies any prescribed metric limit by separation.

If νa=νb\nu_a=\nu_b, then (eq:r4-interior-energy) gives IS(a,b)=0I_S(a,b)=0. The completed energy-capacity estimate makes every positive gap outer-null, and the countable union over rational thresholds gives local relative quasi-everywhere equality. Conversely that equality preserves the completed operator. This proves the asserted uniqueness of classes. ◻

Scope.

The theorem retains a single global defining bound and a specified exterior extension. It does not assert that a general datum dominated by a relative Monge–Ampere measure belongs to this closed image. The smooth nonpolar degree-one counterexample of Theorem “Nonpolar bounded-subsolution failure” rules out such an unrestricted inference even for a natural global bounded completion. Nor is pointwise uniqueness or an ambient classical representative asserted here.

A relative variational retraction and its exact Euler inequality

Retain the arbitrary fixed SS, q≥1q\geq1, norm-box domain Ω\Omega, compact LL, smooth classical collar hh, common bound B>0B>0, and completed class D=DS,h,LB\mathscr D=\mathscr D_{S,h,L}^{B} of Section “A complete strong image for a prescribed relative collar class”. Every class has global defining rows with that bound and that collar. Equality and inequalities between completed potentials are understood locally relative quasi-everywhere unless actual representatives are specified.

The purpose of this section is to replace an unjustified relative obstacle orthogonality assertion by an exact constrained variational statement. Its output need not solve a prescribed Monge–Ampere equation. The failure of an unrestricted existence assertion is illustrated by Theorem “Nonpolar bounded-subsolution failure”.

Definition (Admissible data and admissible convex constraints). Let NS,h,L\mathfrak N_{S,h,L} consist of the measures ν∈Mh,L\nu\in\mathfrak M_{h,L} whose restriction to LL is uniformly absolutely continuous at zero with respect to the fixed outer capacity: for every ϵ>0\epsilon>0 there is δ>0\delta>0 such that E⊂L Borel,Cap⁡S∗(E;Ω)<δ⟹ν(E)<ϵ.E\subset L\text{ Borel},\quad \operatorname{Cap}_S^*(E;\Omega)<\delta \quad\Longrightarrow\quad \nu(E)<\epsilon . In particular ν\nu annihilates the relevant relative-null sets. Fix a nonempty convex subset K⊂D\mathscr K\subset\mathscr D which is closed for local outer-capacity convergence.

All image measures μa\mu_a, a∈Da\in\mathscr D, belong to NS,h,L\mathfrak N_{S,h,L}. Indeed the completed coefficient estimate (eq:r2-limit-coefficient-outer), with a fixed compact scalar equal to one near LL, gives μa(E)≤CCap⁡S∗(E;Ω)\mu_a(E)\leq C\operatorname{Cap}_S^*(E;\Omega) for E⊂LE\subset L. The common defining bound makes this control uniform on the whole class.

An example of K\mathscr K is a nonempty family defined by bounded Borel obstacles ℓ≤a≤r\ell\leq a\leq r quasi-everywhere. It is convex. It is capacity closed because, for δ>0\delta>0, the set {a>r+δ}\{a>r+\delta\} is contained in {∣a−aj∣>δ}∪{aj>r}\{|a-a_j|>\delta\}\cup\{a_j>r\} whenever aj→aa_j\to a; take outer capacities on each compact stage, then take a countable union of positive gaps. The lower inequality is identical. This does not assert that the pointwise obstacle envelope is classical or has any contact measure property.

Define the completed compact-collar energy by E(a)=1q+1∑r=0q∫(a−h) TS,q(a,…,a⏟r,h,…,h⏟q−r).\mathcal E(a)=\frac1{q+1}\sum_{r=0}^q \int (a-h)\, \mathcal T_{S,q}(\underbrace{a,\ldots,a}_{r}, \underbrace{h,\ldots,h}_{q-r}). Each weighted integral is supported in LL. For ν∈NS,h,L\nu\in\mathfrak N_{S,h,L} put Fν(a)=E(a)−∫L(a−h) dν.\mathcal F_\nu(a)=\mathcal E(a) -\int_L(a-h)\,d\nu.

Lemma (Completed energy and datum pairings). The energy and these pairings are finite and well defined on relative classes. They are continuous under local outer-capacity convergence in D\mathscr D, for each fixed admissible ν\nu. The energy is concave, and for at=(1−t)a+tba_t=(1-t)a+tb,

ddtE(at)=∫L(b−a) dμat,\begin{aligned} \frac{d}{dt}\mathcal E(a_t) &=\int_L(b-a)\,d\mu_{a_t},\end{aligned}

E((a+b)/2)−12E(a)−12E(b)≥cqIS(a,b),cq=q4q+1.\begin{aligned} \mathcal E((a+b)/2) -\tfrac12\mathcal E(a)-\tfrac12\mathcal E(b) &\geq c_q I_S(a,b),\qquad c_q=\frac{q}{4^{q+1}}. \end{aligned}

For ν,ν′∈NS,h,L\nu,\nu'\in\mathfrak N_{S,h,L} and a,b∈Da,b\in\mathscr D,

∣∫L(a−b) d(ν−ν′)∣≤4B dpsh,Ω(ν,ν′).\left|\int_L(a-b)\,d(\nu-\nu')\right| \leq4B\,d_{\mathrm{psh},\Omega}(\nu,\nu').

Proof. Relative CLN and ∣a−h∣≤2B|a-h|\leq2B give a uniform finite bound for every term of the energy, by passage from the defining classical rows. Their weighted completed convergence proves continuity of the energy. It also proves independence of relative-null changes, including in the weight slot. The data term is finite because ν(L)<∞\nu(L)<\infty. Its independence follows from the defining absolute continuity.

If aj→aa_j\to a in capacity, split LL into ∣aj−a∣≤ϵ|a_j-a|\leq\epsilon and its complementary gap. Then ∫L∣aj−a∣ dν≤ϵν(L)+2B ν(L∩{∣aj−a∣>ϵ})⟶ϵν(L).\int_L|a_j-a|\,d\nu \leq\epsilon\nu(L)+2B\,\nu(L\cap\{|a_j-a|>\epsilon\}) \longrightarrow\epsilon\nu(L). The fixed-domain capacity of the gap tends to zero, by the all-domain hypothesis and finite localization. Let ϵ↓0\epsilon\downarrow0. This proves continuity of the pairing.

For defining rows um,vmu_m,v_m of a,ba,b, the energy along (1−t)um+tvm(1-t)u_m+tv_m is a polynomial of degree at most q+1q+1. The finite coefficients converge by the completed weighted calculus. Passing the original finite-polynomial derivative identity therefore gives (eq:r4-energy-variation). Passing concavity and the R3 midpoint inequality, together with the convergence of IS(um,vm)I_S(u_m,v_m), gives (eq:r4-energy-midpoint). No differentiation of an uncontrolled weak limit is involved.

For the last estimate, the globally defined functions (um+B)/(2B)(u_m+B)/(2B) and (vm+B)/(2B)(v_m+B)/(2B) are normalized classical testers on Ω\Omega. Hence ∣∫L(um−vm) d(ν−ν′)∣≤4B dpsh,Ω(ν,ν′).\left|\int_L(u_m-v_m)\,d(\nu-\nu')\right| \leq4B\,d_{\mathrm{psh},\Omega}(\nu,\nu'). For these fixed two data measures, the capacity continuity of their integrals just proved permits m→∞m\to\infty. This yields (eq:r4-completed-data-pairing). The completed functions themselves were not declared admissible classical testers in the definition of the metric. ◻

Theorem (The constrained relative variational problem). For every ν∈NS,h,L\nu\in\mathfrak N_{S,h,L}, the functional Fν\mathcal F_\nu has a unique maximizing relative class aν∈Ka_\nu\in\mathscr K. It is characterized by

∫L(b−aν) d(μaν−ν)≤0(b∈K).\int_L(b-a_\nu)\,d(\mu_{a_\nu}-\nu)\leq0 \qquad(b\in\mathscr K).

Every maximizing sequence converges to this class locally in outer SS-capacity.

Status: NEW_PROVED for this constrained completed variational construction over arbitrary fixed SS. Its inputs are the already proved energy and completion theorems; it is not a new CLN estimate or a relative contact-set theorem.

Proof. There is a uniform finite bound on ∣E(a)∣|\mathcal E(a)| for a∈Da\in\mathscr D, and ∣∫L(a−h)dν∣≤2Bν(L)|\int_L(a-h)d\nu|\leq2B\nu(L). Thus the supremum mν=sup⁡KFνm_\nu=\sup_{\mathscr K}\mathcal F_\nu is finite, and K≠∅\mathscr K\ne\varnothing gives a maximizing sequence aja_j. Convexity makes (aj+ak)/2(a_j+a_k)/2 admissible. The measure term is affine, so (eq:r4-energy-midpoint) gives 0≤cqIS(aj,ak)≤mν−12Fν(aj)−12Fν(ak)⟶0.0\leq c_q I_S(a_j,a_k) \leq m_\nu-\tfrac12\mathcal F_\nu(a_j) -\tfrac12\mathcal F_\nu(a_k) \longrightarrow0. The sign is essential: the concave midpoint value lies above the arithmetic mean and below the supremum. The completed energy-capacity estimate makes this sequence capacity Cauchy, first on the fixed domain and then on every smaller domain by reverse framed comparison. The collar completion is complete, and the capacity-closed constraint gives a limit aν∈Ka_\nu\in\mathscr K. Lemma “Completed energy and datum pairings” passes the objective to its limit, proving attainment.

If a,ba,b are two maximizers, the same midpoint bound gives IS(a,b)=0I_S(a,b)=0, so the completed energy-capacity estimate gives a=ba=b relative quasi-everywhere. The preceding Cauchy argument and uniqueness apply to every maximizing sequence.

For b∈Kb\in\mathscr K, the segment aν+t(b−aν)a_\nu+t(b-a_\nu) stays admissible for 0≤t≤10\leq t\leq1. Its one-sided derivative at zero is nonpositive at a maximum. Use (eq:r4-energy-variation) to obtain (eq:r4-variational-inequality). Conversely, concavity gives Fν(b)−Fν(a)≤∫L(b−a) d(μa−ν).\mathcal F_\nu(b)-\mathcal F_\nu(a) \leq\int_L(b-a)\,d(\mu_a-\nu). If all the displayed inequalities hold, this is nonpositive for every b∈Kb\in\mathscr K, proving maximality and the stated characterization. ◻

Theorem (Strong stability and a retraction onto the image). For ν,ν′∈NS,h,L\nu,\nu'\in\mathfrak N_{S,h,L}, put a=aνa=a_\nu, b=aν′b=a_{\nu'}. Then

IS(a,b)≤∫L(a−b) d(ν′−ν)≤4B dpsh,Ω(ν,ν′),\begin{aligned} I_S(a,b) &\leq\int_L(a-b)\,d(\nu'-\nu) \leq4B\,d_{\mathrm{psh},\Omega}(\nu,\nu'), \end{aligned}

dpsh,Ω(μa,μb)≤C(4B dpsh,Ω(ν,ν′))1/2,\begin{aligned} d_{\mathrm{psh},\Omega}(\mu_a,\mu_b) &\leq C\bigl(4B\,d_{\mathrm{psh},\Omega}(\nu,\nu')\bigr)^{1/2}, \end{aligned}

Cap⁡S∗({∣a−b∣>δ};Ω)≤Cδ−2(4B dpsh,Ω(ν,ν′))2−q.\begin{aligned} \operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega) &\leq C\delta^{-2} \bigl(4B\,d_{\mathrm{psh},\Omega}(\nu,\nu')\bigr)^{2^{-q}} . \end{aligned}

The map RK:NS,h,L⟶IK:={μc:c∈K},RK(ν)=μaν,R_{\mathscr K}:\mathfrak N_{S,h,L}\longrightarrow \mathfrak I_{\mathscr K}:=\{\mu_c:c\in\mathscr K\}, \qquad R_{\mathscr K}(\nu)=\mu_{a_\nu}, is a retraction: it is the identity on IK\mathfrak I_{\mathscr K}. This image is complete and closed in Mh,L\mathfrak M_{h,L}. A datum has a solution in K\mathscr K exactly when it is fixed by RKR_{\mathscr K}, and that solution class is unique.

Status: NEW_PROVED for the variational retraction and its stability; the closed-image assertion is the preceding collar theorem applied to a capacity-closed subset.

Proof. Use (eq:r4-variational-inequality) for the competitor bb at the maximizer aa, and for the competitor aa at bb: ∫(b−a) d(μa−ν)≤0,∫(a−b) d(μb−ν′)≤0.\int(b-a)\,d(\mu_a-\nu)\leq0,\qquad \int(a-b)\,d(\mu_b-\nu')\leq0. Adding and rearranging yields the first inequality in (eq:r4-variational-monotonicity), since IS(a,b)=∫(a−b)d(μb−μa)I_S(a,b)=\int(a-b)d(\mu_b-\mu_a). The datum-pairing lemma proves the second inequality. The completed energy-dual and energy-capacity estimates give the next two bounds. Their constants depend on the fixed admissible family, not on moduli of absolute continuity chosen separately for the two data measures.

For c∈Kc\in\mathscr K and ν=μc\nu=\mu_c, concavity gives Fν(b)≤Fν(c)\mathcal F_\nu(b)\leq\mathcal F_\nu(c) for every b∈Kb\in\mathscr K. Uniqueness of the maximizer implies aν=ca_\nu=c, so RK(ν)=νR_{\mathscr K}(\nu)=\nu. Every image measure is admissible data by the coefficient outer-capacity estimate stated above, making this a retraction with its asserted domain and codomain.

For completeness of the image, a metric-Cauchy sequence μcj\mu_{c_j} makes cjc_j capacity Cauchy by the collar estimate. Its limit lies in D\mathscr D by completeness and in K\mathscr K by capacity closedness. Completed strong convergence gives the image limit. The metric separates the ambient measures, so this complete image is closed. Finally, the retraction identity proves the fixed-point criterion, and the injectivity of the collar measure map gives uniqueness modulo relative-null sets. ◻

Corollary (What the Euler inequality actually permits). Suppose a bounded Borel direction ξ\xi, zero outside LL, satisfies aν+tξ∈Ka_\nu+t\xi\in\mathscr K and aν−tξ∈Ka_\nu-t\xi\in\mathscr K for some t>0t>0. Then ∫Lξ d(μaν−ν)=0.\int_L\xi\,d(\mu_{a_\nu}-\nu)=0.

Proof. Use the two stated competitors in (eq:r4-variational-inequality) and divide by tt. ◻

This is equality only in directions known to remain inside the completed psh constraint. A pointwise gap below an obstacle does not by itself make arbitrary smooth positive and negative perturbations admissible. Consequently the variational inequality does not imply a contact-support formula or μaν=ν\mu_{a_\nu}=\nu for arbitrary data. Those would require an additional theorem supplying the missing feasible directions. This precise distinction is compatible with the smooth nonpolar counterexample and avoids the false general ambient-relative contact mechanism.

A smooth nonpolar background can still obstruct completed solvability

This counterexample has remaining degree one, in every ambient dimension n≥2n\geq2. It rules out a general bounded-subsolution criterion even for the global bounded capacitary closure of classical potentials. The global defining-row hypothesis is stated explicitly; no assertion identifying it with the entire locally defined sheaf AS,b\mathcal A_{S,b} is needed.

Let Y=(Gm,Kn)anY=(\mathbf G_{m,K}^{n})^{\rm an} over the original finite extension K/QpK/\mathbf Q_p, and put ti=log⁡∣Ti∣,t=(t1,…,tn),B={x∈Rn:∣x∣<1},V=t−1(B).t_i=\log|T_i|,\qquad t=(t_1,\ldots,t_n),\qquad B=\{x\in\mathbf R^n:|x|<1\},\qquad V=t^{-1}(B). The closure of VV is compact in the torus: it is contained in the compact polyannulus −1≤ti≤1-1\leq t_i\leq1. Its ordinary boundary is contained in {∣t∣=1}\{|t|=1\}. The function ρ=∣t∣2−1\rho=|t|^2-1 is smooth continuous classical psh on the torus, negative on VV, and zero on its ordinary boundary. It is an exhaustion there, since each negative sublevel is compact in VV. Write β=∑i=1nd′ti∧d′′ti,S=[βn−1].\beta=\sum_{i=1}^n d't_i\wedge d''t_i,\qquad S=[\beta^{n-1}]. The brackets indicate the ordinary current of this smooth form. This is a nonzero symmetric original-positive separately closed current of bidegree (n−1,n−1)(n-1,n-1).

Let σ:Rn→Y\sigma:\mathbf R^n\to Y be the original Gauss section, characterized on Laurent polynomials by ∣∑mcmTm∣σ(x)=max⁡m∣cm∣em⋅x.\left|\sum_m c_mT^m\right|_{\sigma(x)} =\max_m |c_m|e^{m\cdot x}. Choose a nonzero real function f∈Cc∞(B)f\in C_c^\infty(B) with 0≤f≤10\leq f\leq1, and define the positive ordinary top measure μ=f(t) βn.\mu=f(t)\,\beta^n . All equalities below concern original CLD currents and tests.

Lemma (Original torus integration and separate convexity). On the standard torus one has the equality of Radon measures

βn=n! σ∗(dx),\beta^n=n!\,\sigma_*(dx),

where dxdx is Lebesgue measure. For any actual classical psh uu on VV, its Gauss restriction U(x)=u(σ(x))U(x)=u(\sigma(x)) is convex in each coordinate separately on every real box contained in BB. If ∣uj∣≤M|u_j|\leq M on VV, the restrictions UjU_j are uniformly bounded and uniformly Lipschitz on every smaller real box compactly contained in BB.

Proof. The original tropical integration formula on the coordinate torus has tropical multiplicity one. The expansion of the top power contains n!n! copies of the oriented coordinate volume form, which proves (eq:r4-torus-volume). This is an ordinary-current statement: upon refining an original test presentation, the full-rank cells projecting to the coordinate tropicalization carry the Gauss-section integral with the same calibration, whereas cells of smaller projection rank contribute zero to the displayed top form. Equivalently, the ordinary coordinate-torus integration formula may be tested first against arbitrary original smooth compact scalars and then extended by Radon uniqueness. This is the same original torus calibration used in the proved Gauss-box results.

For clarity, the separate convexity uses the actual classical condition after complete extensions. Fix all coordinates but the iith, and take the completed residue field of their (n−1)(n-1)-dimensional Gauss point. Their coordinate images are constants in that field with the prescribed absolute values and Gauss independence. The analytic curve with Ti=zT_i=z and the remaining TlT_l equal to these constants maps the entire annulus corresponding to an admissible interval into VV after that complete extension. Its Gauss point of radius ese^s maps back to σ(x1,…,s,…,xn)\sigma(x_1,\ldots,s,\ldots,x_n).

The pullback of uu is subharmonic by the classical definition. On any closed subannulus subtract the affine function of log⁡∣z∣\log|z| interpolating its two endpoint Gauss values. This is a harmonic unit logarithm, and the subharmonic maximum theorem with the two Shilov endpoints makes the difference nonpositive. This is exactly the convexity inequality. It uses a full annulus, not merely an ordinary neighborhood of its Gauss point. The argument is also Lemma “Uniform continuity on an inner Gauss box”.

A one-dimensional convex function bounded by MM in absolute value has slopes of absolute value at most 2M/δ2M/\delta on an interval at distance δ\delta from the endpoints of a larger interval. Apply this to each coordinate on nested real boxes. Changing one coordinate at a time gives the uniform local Lipschitz bound for the ℓ1\ell^1 distance. ◻

Definition (Global bounded defining rows on VV). A bounded Borel function aa belongs to GS,b(V)\mathscr G_{S,b}(V) if, for some finite MM, there are uj∈Cb(V)u_j\in\mathcal C_b(V) with ∣uj∣≤M|u_j|\leq M on the whole VV such that uj→au_j\to a locally in outer SS-capacity on every permitted compact/domain pair in VV. The representative aa is required to be bounded as an actual function.

This is the bounded global closure with precisely the same convergence used in R2. Its members belong locally to AS,b\mathcal A_{S,b}, and for remaining degree one their completed operator is D(aS)D(aS). Restrictions of globally bounded defining rows from a larger collar domain are included.

Lemma (Traces of global completed potentials). If a∈GS,b(V)a\in\mathscr G_{S,b}(V), the Borel function A(x)=a(σ(x))A(x)=a(\sigma(x)) has, up to Lebesgue-almost-everywhere equality, a continuous separately convex representative UU on BB. If aa has ordinary boundary limit zero on VV, then UU has limit zero on the real sphere ∂B\partial B.

Proof. The global function v=∣t∣2v=|t|^2 is a normalized classical tester on VV and satisfies D(vS)=2βn.D(vS)=2\beta^n. Consequently, for every Borel E⊂VE\subset V,

2βn(E)≤Cap⁡S(E;V)≤Cap⁡S∗(E;V).2\beta^n(E)\leq\operatorname{Cap}_S(E;V) \leq\operatorname{Cap}_S^*(E;V).

For a real compact F⋐BF\Subset B, the subset t−1(F)t^{-1}(F) is compact in VV. Local all-domain capacity convergence and a finite covering by permitted compact norm-box stages imply convergence in outer capacity on this compact set with capacity domain VV, by domain monotonicity and subadditivity.

Take a common bound for aa and its defining row. Split t−1(F)t^{-1}(F) into the gap ∣uj−a∣>ϵ|u_j-a|>\epsilon and its complement. Formula (eq:r4-capacity-controls-volume) and (eq:r4-torus-volume) give ∫F∣uj(σ(x))−A(x)∣ dx⟶0.\int_F|u_j(\sigma(x))-A(x)|\,dx\longrightarrow0. More explicitly, the integral is at most ϵ vol⁡(F)\epsilon\,\operatorname{vol}(F) plus a fixed constant times the outer capacity of that gap; let j→∞j\to\infty and then ϵ↓0\epsilon\downarrow0.

The preceding lemma gives local equicontinuity and boundedness of the restrictions Uj=uj∘σU_j=u_j\circ\sigma. Arzela–Ascoli on an exhaustion of BB by compact sets, followed by a diagonal subsequence, gives local uniform convergence to a continuous separately convex UU. The Lloc1L^1_{\rm loc} limit just identified makes U=AU=A almost everywhere.

Every σ(y)\sigma(y) for y∈∂By\in\partial B is an ordinary boundary point of VV, since it is approached through the continuous Gauss section from BB and has ∣t∣=1|t|=1. If aa has zero ordinary boundary limit, compactness of this Gauss sphere implies that for every ϵ>0\epsilon>0 there is rϵ<1r_\epsilon<1 such that ∣A(x)∣<ϵ(rϵ<∣x∣<1).|A(x)|<\epsilon\quad(r_\epsilon<|x|<1). Otherwise a sequence tending to the sphere, and a convergent subsequence of its Gauss images, contradict the boundary limit. Since U=AU=A almost everywhere and UU is continuous, the same collar satisfies ∣U∣≤ϵ|U|\leq\epsilon. This proves the last claim. It does not assume that an arbitrary relative classical representative automatically inherits boundary values. ◻

Theorem (Nonpolar bounded-subsolution failure). The current SS above has no coefficient mass on any ambient locally pluripolar set. The measure μ\mu is nonzero, finite, nonatomic and compactly supported in VV. The smooth bounded classical function w=12(∣t∣2−1)w=\tfrac12(|t|^2-1) has ordinary boundary value zero and satisfies D(wS)=βn≥μ.D(wS)=\beta^n\geq\mu . Nevertheless there is no a∈GS,b(V)a\in\mathscr G_{S,b}(V) with ordinary boundary limit zero and D(aS)=μ.D(aS)=\mu . In particular there is no bounded actual classical solution. Thus a general bounded-subsolution criterion in this globally defined bounded relative completion is FALSE, even for a smooth nonpolar background and compact nonatomic data.

Status: NEW_PROVED counterexample.

Proof. The smooth potential ψ=∣t∣2/2\psi=|t|^2/2 has Dψ=βD\psi=\beta and is bounded on VV. Hence SS is the ambient bounded BT product (Dψ)n−1(D\psi)^{n-1}. Ambient coefficient-capacity control and the proved ambient polar outer-zero theorem imply that every coefficient variation of SS gives zero mass to locally pluripolar sets. This proves the precise nonpolar assertion; it is stronger than an assertion about only one top trace. Formula (eq:r4-torus-volume) gives μ=n! σ∗(f(x) dx).\mu=n!\,\sigma_*(f(x)\,dx). It is therefore finite, nonatomic, nonzero, and carried by the compact σ(supp⁡f)⋐V\sigma(\operatorname{supp}f)\Subset V. The displayed subsolution calculation follows from Dρ=2βD\rho=2\beta.

Suppose the asserted solution aa exists, and let UU be the continuous separately convex representative of its Gauss trace from Lemma “Traces of global completed potentials”. For φ∈Cc∞(B)\varphi\in C_c^\infty(B), φ(t)\varphi(t) is an original smooth compact test on VV, because inverse images of compact tropical sets are compact in the torus. The original smooth form identity is Dφ(t)∧βn−1=Δφ(t)n βn.D\varphi(t)\wedge\beta^{n-1} =\frac{\Delta\varphi(t)}{n}\,\beta^n . The equation D(aS)=μD(aS)=\mu, Borel multiplication of the coefficient-Radon current, and (eq:r4-torus-volume) consequently give (n−1)!∫BU Δφ dx=n!∫Bfφ dx.(n-1)!\int_B U\,\Delta\varphi\,dx =n!\int_B f\varphi\,dx. Thus

ΔU=nfas real distributions on B.\Delta U=n f \quad\text{as real distributions on }B .

Only legitimate original tests were used; equality on their toric subclass is a necessary consequence of the proposed full-current equation.

Choose 0<r0<10<r_0<1 with supp⁡f⊂{∣x∣<r0}\operatorname{supp}f\subset\{|x|<r_0\}. On the connected annulus A={r0<∣x∣<1}A=\{r_0<|x|<1\}, separate convexity makes each distribution ∂i2U\partial_i^2U positive. This follows directly by one-dimensional convexity and Fubini on small real boxes. Their sum is zero by (eq:r4-trace-poisson); hence each ∂i2U=0\partial_i^2U=0 there.

On every real box in AA, the continuous function UU is therefore affine in each coordinate, and successive linear interpolation gives one multiaffine polynomial on that box. Polynomials from overlapping boxes agree on their open intersection, and thus agree identically. Since AA is connected when n≥2n\geq2, a chain of overlapping boxes gives a single multiaffine polynomial PP on the whole annulus. The boundary assertion of the preceding lemma gives P=0P=0 on the real unit sphere. Write P(x′,xn)=P0(x′)+xnP1(x′)P(x',x_n)=P_0(x')+x_nP_1(x'). For every ∣x′∣<1|x'|<1, the two sphere points xn=±1−∣x′∣2x_n=\pm\sqrt{1-|x'|^2} imply P0(x′)=P1(x′)=0P_0(x')=P_1(x')=0. Both polynomials vanish identically, so P=0P=0.

It follows that U=0U=0 on the outer annulus and is compactly supported inside BB. Choose η∈Cc∞(B)\eta\in C_c^\infty(B) equal to one on {∣x∣≤r0}\{|x|\leq r_0\}. Its derivatives are supported where U=0U=0, and (eq:r4-trace-poisson) yields 0=∫BU Δη dx=n∫Bf dx>0.0=\int_B U\,\Delta\eta\,dx=n\int_B f\,dx>0. This contradiction proves the theorem. ◻

Precise scope of the negative result.

The global rows on the whole torus-preimage domain are essential in the trace-convexity argument. A classical potential defined only on an ordinary neighborhood of a Gauss point need not have a convex restriction to a preassigned skeleton. For example, −log⁡∣T−1∣-\log|T-1| is harmonic on a neighborhood avoiding T=1T=1, whereas on the usual Gauss line its restriction is −max⁡(t,0)-\max(t,0), which is concave. The proof therefore makes no unjustified passage from local representatives of the sheaf AS,b\mathcal A_{S,b} to global defining rows.

The example is also compatible with the closed-image theorem “A closed complete strong measure image”. That theorem closes the image of one bounded admissible family in a strong data topology; it does not assert that every measure dominated by a member belongs to that image. Here 0≤μ≤D(wS)0\leq\mu\leq D(wS) does not supply even a globally completed psh solution with the specified zero boundary limit. A larger, non-psh energy solution space may therefore be needed for a linear relative Poisson problem. No claim that such a solution is psh is made.

A faithful curvature measure and a coercive relative energy form

Fix the original finite extension K/QpK/\mathbf Q_p and its absolute value. Let Ω\Omega be an ordinary algebraic norm box relatively compact in the analytification of a smooth integral affine KK-variety YY of dimension n≥1n\geq1. Let SS be an arbitrary symmetric original CLD-positive current of bidegree (n−1,n−1)(n-1,n-1) on Ω\Omega, separately closed under d′d' and d′′d''. Every current and coefficient measure below uses the original smooth CLD test sheaf. The background need not be strongly positive, smooth, or supported on a subvariety.

The purpose is to construct an actual reference measure and close the energy form. This fills the trace and core hypotheses left conditional in the separate R3 degree-one research note. It does not identify a Dirichlet-form capacity with raw relative Bedford–Taylor capacity.

For real original smooth functions set ΓS(f,g)=12(d′f∧d′′g+d′g∧d′′f)∧S,ΓS(f)=ΓS(f,f).\Gamma_S(f,g)=\tfrac12(d'f\wedge d''g+d'g\wedge d''f)\wedge S, \qquad \Gamma_S(f)=\Gamma_S(f,f). These are signed, respectively positive, Radon measures, by the original positive cone and coefficient-Radon theorem. Mixed energies are interpreted with this symmetric convention.

A countable smooth curvature construction

Lemma (A global smooth compensator for a finite moment list). Let H⋐ΩH\Subset\Omega and let ti=log⁡∣ri∣t_i=\log|r_i|, 1≤i≤m1\leq i\leq m, be a finite rational unit presentation on a neighborhood of HH. There is an original smooth continuous classical psh function aa on YanY^{\rm an}, normalized so that 0≤a≤10\leq a\leq1 on Ω\Omega, and a finite N>0N>0, such that

Da ≥ N−1∑id′ti∧d′′tinear H,d′a∧d′′a ≤ 2a Daon Ω.Da\ \geq\ N^{-1}\sum_i d't_i\wedge d''t_i \quad\hbox{near }H, \qquad d'a\wedge d''a\ \leq\ 2a\,Da\quad\hbox{on }\Omega.

Both inequalities are inequalities of positive (1,1)(1,1)-forms on common original moment presentations.

Proof. Write ri=bi/cir_i=b_i/c_i, where bi,cib_i,c_i are regular functions on the affine variety YY and neither vanishes on a neighborhood of HH. Choose A>0A>0 so that log⁡∣bi∣+A\log|b_i|+A and log⁡∣ci∣+A\log|c_i|+A are positive on a slightly smaller neighborhood of HH. Fix a nonnegative smooth function κ\kappa supported in (−1,0)(-1,0) with integral one, and put ζ(s)=∫−∞s(s−r)κ(r) dr,F(s)=ζ(s)2.\zeta(s)=\int_{-\infty}^{s}(s-r)\kappa(r)\,dr, \qquad F(s)=\zeta(s)^2. Then ζ\zeta is smooth, convex and nondecreasing, is zero for s≤−1s\leq-1, and satisfies ζ′=1\zeta'=1 for s≥0s\geq0. Consequently F≥0F\geq0, F′≥0F'\geq0, F′′≥0F''\geq0, F′′=2(s≥0),(F′)2≤2FF′′everywhere.F''=2\quad(s\geq0),\qquad (F')^2\leq2FF''\quad\hbox{everywhere}. For a regular function bb, the function F(log⁡∣b∣+A)F(\log|b|+A) is extended by zero where b=0b=0. It is original smooth: it is a smooth function of the unit logarithm on the nonvanishing locus and is identically zero on an ordinary neighborhood of the zero locus. It is classical psh after every complete isometric extension, by convex nondecreasing composition on curves, or directly by its nonnegative one-variable second derivative on each unit locus. The constant branch handles zeros and identically zero pullbacks.

Let gg be the sum of these expressions for all bi,cib_i,c_i. Near HH, the identity Dlog⁡∣bi∣=Dlog⁡∣ci∣=0D\log|b_i|=D\log|c_i|=0 gives Dg=2∑i(d′log⁡∣bi∣∧d′′log⁡∣bi∣+d′log⁡∣ci∣∧d′′log⁡∣ci∣)≥∑id′ti∧d′′ti.Dg=2\sum_i\bigl(d'\log|b_i|\wedge d''\log|b_i| +d'\log|c_i|\wedge d''\log|c_i|\bigr) \geq\sum_i d't_i\wedge d''t_i. The inequality is the elementary square inequality (ξ−η)2≤2ξ2+2η2(\xi-\eta)^2\leq2\xi^2+2\eta^2. On any common presentation, scalar Cauchy–Schwarz and (F′)2≤2FF′′(F')^2\leq2FF'' give d′g∧d′′g≤2g Dg.d'g\wedge d''g\leq2g\,Dg. This also holds at a zero branch by smoothness; equivalently, apply Cauchy–Schwarz to the nonzero summands and then take limits. Since Ω‾\overline\Omega is compact, gg has a finite bound there. Set N=max⁡(1,sup⁡Ω‾g)N=\max(1,\sup_{\overline\Omega}g) and a=g/Na=g/N. The two claimed inequalities follow. ◻

Theorem (A faithful curvature measure dominating all coefficients). There are original smooth classical functions 0≤al≤10\leq a_l\leq1 on Ω\Omega and a continuous classical function h=∑l≥12−lal,0≤h≤1,h=\sum_{l\geq1}2^{-l}a_l,\qquad 0\leq h\leq1, such that the positive Radon measure

λ=D(hS)=∑l≥12−lD(alS)\lambda=D(hS)=\sum_{l\geq1}2^{-l}D(a_lS)

has the following properties. For every original compact smooth (1,1)(1,1)-test η\eta, there is a finite CηC_\eta with

∣μS,η∣≤Cηλ.|\mu_{S,\eta}|\leq C_\eta\lambda.

For every Borel set E⊂ΩE\subset\Omega,

λ(E)=0⟺1ES=0.\lambda(E)=0\quad\Longleftrightarrow\quad\mathbf1_E S=0.

Thus F=supp⁡λ=supp⁡SF=\operatorname{supp}\lambda=\operatorname{supp}S. The measure is locally finite; its total mass on Ω\Omega is not required to be finite. For every original smooth gg of compact support, D(gS)=bgλ,bg∈L∞(λ),ΓS(g)≤Cgλ,D(gS)=b_g\lambda,\quad b_g\in L^\infty(\lambda), \qquad \Gamma_S(g)\leq C_g\lambda, and bgb_g has compact support in FF up to a null set.

Status: NEW_PROVED. The countable cofinal moment controls are a named direct consequence of R and B, as recorded in Lemma “A countable family of coefficient controls”. The new construction makes their dominating measure the curvature of one bounded continuous potential and retains its exact coefficient-null ideal.

Proof. Use the countable positive controls from that lemma, in degree (1,1)(1,1). They have the form Θl=χl∑id′tli∧d′′tli,0≤χl≤1,\Theta_l=\chi_l\sum_i d't_{li}\wedge d''t_{li}, \qquad 0\leq\chi_l\leq1, with compact support on a rational unit presentation. The cutoffs can be chosen between zero and one from the start. The exact rational replacement uses B, Lemma D.1, and a countable dense subfield of the original KK; it changes no pointwise moment norm or absolute value. Apply Lemma “A global smooth compensator for a finite moment list” to each compact support, with constant NlN_l. Positivity and the local smooth product rule give the global measure inequality D(alS)≥Nl−1μS,Θl.D(a_lS)\geq N_l^{-1}\mu_{S,\Theta_l}. Outside the support of Θl\Theta_l its right side is zero, so no extension of the current is involved.

The original smooth classical functions hm=∑l=1m2−lal+2−mh_m=\sum_{l=1}^m2^{-l}a_l+2^{-m} decrease uniformly to hh, with 0≤hm≤10\leq h_m\leq1. The local relative CLN estimate bounds the masses of D(hmS)D(h_mS) on each compact stage, independently of mm. Their measures are increasing, since D(hmS)=∑l≤m2−lD(alS)D(h_mS)=\sum_{l\leq m}2^{-l}D(a_lS). Their locally finite sum equals D(hS)D(hS) by uniform convergence against the fixed coefficient measures of SS. This proves (eq:r4-reference-lambda); it also proves classicality and continuity of hh by the bounded decreasing limit property.

For each ll, μS,Θl≤2lNlλ\mu_{S,\Theta_l}\leq2^lN_l\lambda. Finite cofinal domination therefore gives (eq:r4-all-coefficients-lambda) for every original test, including tests with additional moment coordinates. This is not a claim based on one preferred tropical chart. If λ(E)=0\lambda(E)=0, that inequality makes all coefficient restrictions of SS to EE zero. Conversely, every D(alS)=Dal∧SD(a_lS)=Da_l\wedge S is a smooth coefficient measure of SS locally. Hence 1ES=0\mathbf1_E S=0 makes every term of the positive series vanish on EE. This proves (eq:r4-faithful-lambda). Applying it to ordinary open sets identifies the supports. Finally D(gS)=Dg∧SD(gS)=Dg\wedge S and ΓS(g)=d′g∧d′′g∧S\Gamma_S(g)=d'g\wedge d''g\wedge S are coefficient measures of compact original tests, so the remaining assertions follow from (eq:r4-all-coefficients-lambda). ◻

Closability, locality, and coercivity

Let C\mathscr C be the restrictions to FF of real original smooth compactly supported functions on Ω\Omega. Set E(f,g)=∫ΩΓS(f,g)=−∫Ffbg dλ(f,g∈C).\mathcal E(f,g)=\int_\Omega\Gamma_S(f,g) =-\int_F f b_g\,d\lambda \qquad(f,g\in\mathscr C). The second equality is compact Stokes. These formulas are independent of the chosen ambient representatives: if a core function is zero λ\lambda-almost everywhere, its self-energy is zero by the second formula, and all mixed energies vanish by Cauchy–Schwarz.

Theorem (A regular strongly local coercive form). The form (E,C)(\mathcal E,\mathscr C) is densely defined and closable on L2(F,λ)L^2(F,\lambda). Its closure (E,D)(\mathcal E,\mathscr D) is a regular strongly local symmetric Dirichlet form. It satisfies

E(f,f)≥∫Ff22−h dλ≥12∥f∥L2(λ)2(f∈D).\mathcal E(f,f)\geq \int_F\frac{f^2}{2-h}\,d\lambda \geq\tfrac12\|f\|_{L^2(\lambda)}^2 \qquad(f\in\mathscr D).

In particular E1/2\mathcal E^{1/2} itself is a Hilbert norm on D\mathscr D, equivalent to the usual form norm. The construction is valid also when S=0S=0, in which case the support space and the Hilbert space are empty and zero.

Status: NEW_PROVED. Here regularity and strong locality concern the support space FF with its ordinary subspace topology. No intrinsic metric, doubling property, or strict psh exhaustion is presumed.

Proof. The space FF is locally compact, metrizable, and second countable. Original smooth compact functions form a real algebra and separate its points. Indeed distinct Berkovich points in the affine analytification give different seminorms to some regular function. A smooth truncation of its logarithm distinguishes them, and an original compact cutoff preserves these two values. The cutoffs also give a nonzero value at each point. The locally compact Stone–Weierstrass theorem, followed by multiplication by one cutoff on any prescribed compact stage, gives uniform density in Cc(F)C_c(F) in the usual regular-core sense, and hence density in L2(F,λ)L^2(F,\lambda).

For closability, let fj→0f_j\to0 in L2(λ)L^2(\lambda) and assume that fjf_j is Cauchy for the energy seminorm. It has a limit in the abstract Hilbert completion of that seminorm. For every g∈Cg\in\mathscr C, E(fj,g)=−∫fjbg dλ⟶0,\mathcal E(f_j,g)=-\int f_jb_g\,d\lambda\longrightarrow0, because bgb_g is bounded and compactly supported, hence belongs to L2(λ)L^2(\lambda). The limit is orthogonal to the dense core of that completion and is therefore zero. This is closability.

For every smooth normal contraction PP with P(0)=0P(0)=0 and ∣P′∣≤1|P'|\leq1, the original chain rule gives E(P(f),P(f))≤E(f,f)\mathcal E(P(f),P(f))\leq\mathcal E(f,f) on the core. Approximate 0∨t∧10\vee t\wedge1 by such smooth contractions. Their images converge in L2L^2 and have bounded form norm; weak compactness in the closed form space and lower semicontinuity give the contraction inequality. Approximating an arbitrary form-domain function by core functions proves the same inequality on D\mathscr D. This is the Markov property. The identical argument supplies any scalar normal contraction, in particular absolute value and positive part. Core density in both the form norm and the uniform topology proves regularity.

We verify strong locality on FF, rather than assume it from ambient representatives. For any fixed original compact smooth χ\chi, the coefficient domination gives E(χf,χf)≤2∥χ∥∞2E(f,f)+2Cχ∥f∥L2(λ)2.\mathcal E(\chi f,\chi f) \leq 2\|\chi\|_\infty^2\mathcal E(f,f) +2C_\chi\|f\|_{L^2(\lambda)}^2. Thus multiplication by χ\chi extends boundedly to D\mathscr D. Suppose f∈Df\in\mathscr D has compact support in FF and g∈Dg\in\mathscr D equals a constant cc almost everywhere on a relative ordinary open neighborhood OO of that support. Choose χ=1\chi=1 near this support and with supp⁡χ∩F⊂O\operatorname{supp}\chi\cap F\subset O. For core fj→ff_j\to f in form norm, put kj=χfjk_j=\chi f_j. Then kj→fk_j\to f in form norm. The measure D(kjS)D(k_jS) is carried by OO, has a bounded compactly supported density against λ\lambda, and has total mass zero by compact Stokes. The Green identity extends to gg by its L2L^2 convergence from the core. Hence E(kj,g)=−∫g D(kjS)=−c∫D(kjS)=0.\mathcal E(k_j,g)=-\int g\,D(k_jS) =-c\int D(k_jS)=0. Passing to the limit proves strong locality.

It remains to prove the stated coercivity, which is stronger than closability. Write vm=2−hmv_m=2-h_m, so 1≤vm≤21\leq v_m\leq2, and λm=D(hmS)\lambda_m=D(h_mS). For a core ff, the positive square identity for the two covectors d′fd'f and (f/vm)d′vm(f/v_m)d'v_m gives ΓS(f)−ΓS(vm,f2/vm)≥0.\Gamma_S(f)-\Gamma_S(v_m,f^2/v_m)\geq0. All functions in this identity are original smooth, and f2/vmf^2/v_m has compact support. Compact Stokes therefore yields E(f,f)≥−∫(f2/vm)D(vmS)=∫(f2/vm) dλm.\mathcal E(f,f)\geq -\int(f^2/v_m)D(v_mS) =\int(f^2/v_m)\,d\lambda_m. The positive measures λm\lambda_m increase locally to λ\lambda, and vm→2−hv_m\to 2-h uniformly. On the fixed compact support of ff this proves the first inequality in (eq:r4-picone-coercivity). The second follows from 2−h≤22-h\leq2. Both pass to the closed form domain by form approximation and L2L^2 convergence, since 1/(2−h)1/(2-h) is bounded. Completeness for the energy norm follows from ∥f∥22≤2E(f,f)\|f\|_2^2\leq2\mathcal E(f,f). ◻

Lemma (The reference potential belongs locally to the form domain). For every original compact smooth cutoff χ\chi, χhm→χh\chi h_m\to\chi h in the closed form norm. In particular hh belongs locally to D\mathscr D.

Proof. The compensators satisfy ΓS(al)≤2D(alS)\Gamma_S(a_l)\leq2D(a_lS) on Ω\Omega. For k>mk>m, weighted Cauchy–Schwarz consequently gives ΓS(hk−hm)≤2(∑l=m+1k2−l)∑l=m+1k2−lD(alS)≤21−mλ.\Gamma_S(h_k-h_m) \leq 2\Bigl(\sum_{l=m+1}^k2^{-l}\Bigr) \sum_{l=m+1}^k2^{-l}D(a_lS) \leq 2^{1-m}\lambda. Also ∣hk−hm∣≤2−m|h_k-h_m|\leq2^{-m}. The product square estimate, integrated over the support of χ\chi, gives E(χ(hk−hm),χ(hk−hm))≤21−2mE(χ,χ)+22−m∥χ∥∞2λ(supp⁡χ).\mathcal E\bigl(\chi(h_k-h_m),\chi(h_k-h_m)\bigr) \leq2^{1-2m}\mathcal E(\chi,\chi) +2^{2-m}\|\chi\|_\infty^2\lambda(\operatorname{supp}\chi). The L2L^2 difference tends to zero as well. Thus the sequence is form Cauchy, and its L2L^2 limit is the stated function. ◻

A linear relative Dirichlet problem on all original tests

For an ordinary open W⋐ΩW\Subset\Omega, let D0(W)\mathscr D_0(W) be the form-norm closure of the original smooth compact functions supported in WW, regarded as zero off WW in L2(F,λ)L^2(F,\lambda). This definition is the boundary condition in the next theorem. It is not an assertion of a pointwise ordinary boundary limit.

Theorem (Coercive energy solutions of the relative Poisson equation). For every real f∈L2(W∩F,λ)f\in L^2(W\cap F,\lambda), there is a unique u∈D0(W)u\in\mathscr D_0(W) such that

E(u,ϕ)=−∫Wfϕ dλ(ϕ∈D0(W)).\mathcal E(u,\phi)=-\int_W f\phi\,d\lambda \quad(\phi\in\mathscr D_0(W)).

Its coefficient-defined current satisfies D(uS)=fλon WD(uS)=f\lambda\quad\hbox{on }W against every original compact smooth scalar test. Moreover E(u,u)1/2≤2∥f∥L2(λ),∥u∥L2(λ)≤2∥f∥L2(λ).\mathcal E(u,u)^{1/2}\leq\sqrt2\|f\|_{L^2(\lambda)}, \qquad \|u\|_{L^2(\lambda)}\leq2\|f\|_{L^2(\lambda)}. The solution depends linearly and continuously on ff. If f≥0f\geq0, then u≤0u\leq0 almost everywhere. If f∈L∞(W∩F,λ)f\in L^\infty(W\cap F,\lambda), then

∣u∣≤(2−h)∥f∥∞≤2∥f∥∞λ-almost everywhere on W.|u|\leq(2-h)\|f\|_\infty\leq2\|f\|_\infty \quad\lambda\hbox{-almost everywhere on }W.

Thus positive bounded densities have bounded energy solutions whose relative curvature is positive. The theorem does not assert that these solutions are ambient classical psh or belong to the bounded psh capacity completion.

Status: NEW_PROVED, with a fully specified energy boundary condition. The functional-analytic Riesz representation step is standard; the faithful original coefficient control and coercive barrier needed to apply it are proved above.

Proof. The preceding coercivity makes D0(W)\mathscr D_0(W) a Hilbert space for E\mathcal E. The functional ϕ↦∫Wfϕ dλ\phi\mapsto\int_W f\phi\,d\lambda has norm at most 2∥f∥2\sqrt2\|f\|_2. Riesz representation gives a unique w∈D0(W)w\in\mathscr D_0(W) with E(w,ϕ)=∫fϕ dλ\mathcal E(w,\phi)=\int f\phi\,d\lambda. Set u=−wu=-w. Taking ϕ=w\phi=w and using coercivity proves both quantitative bounds and continuity in the data.

The multiplication uSuS is well defined on all original compact complementary tests: their coefficient variations are bounded by CηλC_\eta\lambda, and uu is locally L1(λ)L^1(\lambda) by Cauchy–Schwarz and local finiteness. An L2L^2 approximation by core functions extends Green’s identity to E(u,ϕ)=−∫u D(ϕS)=−⟨D(uS),ϕ⟩(ϕ∈Ac0,0(W;R)).\mathcal E(u,\phi)=-\int u\,D(\phi S) =-\langle D(uS),\phi\rangle \quad(\phi\in A_c^{0,0}(W;\mathbf R)). This proves the full original-current equation, rather than only its value on a chosen tropical subfamily of tests.

If f≥0f\geq0, the vector ww uniquely minimizes E(z,z)/2−∫fz dλ\mathcal E(z,z)/2-\int fz\,d\lambda on the Hilbert space. The normal contraction z↦∣z∣z\mapsto|z| preserves D0(W)\mathscr D_0(W) and does not increase energy. Therefore replacing ww by ∣w∣|w| does not increase the objective. Uniqueness gives w=∣w∣w=|w|. The same argument shows positivity of this linear solution operator for every nonnegative datum.

We supply the boundary justification for comparison with v=2−h≥1v=2-h\geq1. Choose an original smooth 0≤χ≤10\leq\chi\leq1 equal to one near W‾\overline W and compactly supported in Ω\Omega. By Lemma “The reference potential belongs locally to the form domain”, v~=χv∈D\widetilde v=\chi v\in\mathscr D and χ(2−hm)→v~\chi(2-h_m)\to\widetilde v in form norm. For every core test in WW, Stokes and D(vS)=−λD(vS)=-\lambda give E(v~,ϕ)=∫Wϕ dλ.\mathcal E(\widetilde v,\phi)=\int_W\phi\,d\lambda. This extends to every ϕ∈D0(W)\phi\in\mathscr D_0(W), since λ(W)<∞\lambda(W)<\infty and both sides are form continuous.

For completeness, if w∈D0(W)w\in\mathscr D_0(W) and M≥0M\geq0, then z=(w−Mv~)+z=(w-M\widetilde v)^+ belongs to D0(W)\mathscr D_0(W). Approximate ww by original core wjw_j supported in WW, and v~\widetilde v by the nonnegative smooth functions χ(2−hm)\chi(2-h_m). The positive part of their difference is supported inside supp⁡wj\operatorname{supp}w_j. Smooth normal contractions that vanish on (−∞,0](-\infty,0] approximate this positive part, with energy bounded by that of the difference; their supports remain in this same compact subset of WW. Closedness and weak compactness in form norm therefore put the positive part in D0(W)\mathscr D_0(W). First let m→∞m\to\infty and then j→∞j\to\infty; the L2L^2 limits and the same bounded-energy argument prove the claim for zz. No pointwise boundary regularity is used here.

Assume first 0≤f≤M0\leq f\leq M. Pair the two equations with this zz to obtain E(w−Mv~,z)=∫W(f−M)z dλ≤0.\mathcal E(w-M\widetilde v,z)=\int_W(f-M)z\,d\lambda\leq0. For every real form-domain rr, the normal contraction inequality E(∣r∣)≤E(r)\mathcal E(|r|)\leq\mathcal E(r) implies E(r+,r−)≤0\mathcal E(r^+,r^-)\leq0. Hence E(r,r+)≥E(r+,r+)\mathcal E(r,r^+)\geq\mathcal E(r^+,r^+). Apply this with r=w−Mv~r=w-M\widetilde v. Coercivity makes z=0z=0, so 0≤w≤Mv0\leq w\leq Mv on WW. For signed ff with ∣f∣≤M|f|\leq M, positivity and linearity of the solution operator applied to M±fM\pm f give ∣wf∣≤wM≤Mv|w_f|\leq w_M\leq Mv. This proves (eq:r4-energy-linfty). ◻

Relation to higher mixed degree and to the remaining gates.

For any fixed allowed (p,p)(p,p)-background with q=n−p≥1q=n-p\geq1, one may freeze q−1q-1 bounded differentiated slots and apply this section to their positive closed mixed current of degree (n−1,n−1)(n-1,n-1). This supplies a linear energy theory for that fixed mixed background. It does not identify these separate reference measures uniformly as the slots vary, solve a nonlinear Monge–Ampere equation, or prove raw relative capacity outer regularity. The ordinary topology on FF is retained, and no equality on additional δ\delta-tests is asserted.

A one-way comparison with the regular energy capacity

Retain the degree-one setting, support space F=supp⁡SF=\operatorname{supp}S, faithful curvature measure λ=D(hS)\lambda=D(hS), and regular closed form (E,D)(\mathcal E,\mathscr D) of Section “A faithful curvature measure and a coercive relative energy form”. In particular 0≤h≤10\leq h\leq1, every fixed compact original coefficient measure is bounded by a constant times λ\lambda, and E(f,f)≥12∥f∥L2(λ)2\mathcal E(f,f)\geq\frac12\|f\|_{L^2(\lambda)}^2. Put E1(f,f)=E(f,f)+∥f∥L2(λ)2.\mathcal E_1(f,f)=\mathcal E(f,f)+\|f\|_{L^2(\lambda)}^2. The relative BT capacity below always uses the original bounded classical testers on the indicated analytic domain. The energy capacity instead lives on the ordinary support space FF.

The external capacity input and its hypotheses

Let cap⁡E,1\operatorname{cap}_{\mathcal E,1} be the 11-capacity of this regular Dirichlet form: for a relative open G⊂FG\subset F, cap⁡E,1(G)=inf⁡{E1(f,f):f∈D, f≥1 λ-a.e. on G},\operatorname{cap}_{\mathcal E,1}(G) =\inf\{\mathcal E_1(f,f):f\in\mathscr D,\ f\geq1\ \lambda\text{-a.e. on }G\}, and for arbitrary A⊂FA\subset F take the infimum over relative open neighborhoods. We use the standard Choquet property and the following compact-core formula:

cap⁡E,1(K)=inf⁡{E1(f,f):f∈C, f>1 on K},K⋐F.\operatorname{cap}_{\mathcal E,1}(K) =\inf\{\mathcal E_1(f,f):f\in\mathscr C,\ f>1\text{ on }K\}, \qquad K\Subset F.

Here C\mathscr C is the restriction of the original smooth compact functions on Ω\Omega.

Status of this input: EXTERNAL_KNOWN. The precise primary reference is Fukushima–Sato–Taniguchi [FST91, §3, pp. 519–520, conditions (C.1), (C.2) and formula (3.1)]. Their space is locally compact separable metric, their reference measure is positive Radon with full support, and their regular core is a uniformly dense algebra with compact cutoffs and smooth contraction stability. These hypotheses hold here on FF: regularity and full support were proved above; original smooth cutoffs restrict to the required compact cutoffs on FF; and smooth scalar contractions vanishing at zero preserve C\mathscr C. In particular for a compact K⊂G⋐FK\subset G\Subset F the ambient cutoff may be chosen with support meeting FF only inside GG. If the transience hypothesis stated later in that paper is retained, it also holds here. Indeed, σ\sigma-finiteness of λ\lambda gives a bounded strictly positive g∈L1(λ)∩L2(λ)g\in L^1(\lambda)\cap L^2(\lambda) with ∥g∥2≤1/2\|g\|_2\leq1/\sqrt2, and coercivity gives ∫F∣f∣g dλ≤∥g∥2∥f∥2≤E(f,f)1/2.\int_F |f|g\,d\lambda \leq\|g\|_2\|f\|_2\leq\mathcal E(f,f)^{1/2}. No finiteness of λ(F)\lambda(F) is needed. For S=0S=0 the statements below have their evident zero-capacity interpretation.

A uniform estimate against all normalized testers

Lemma (The smooth core controls the BT measures). Fix an ordinary relatively compact W⋐ΩW\Subset\Omega. There is a finite CWC_W such that every real original smooth compact function ff and every w∈Cb(Ω)w\in\mathcal C_b(\Omega) with 0≤w≤10\leq w\leq1 satisfy

∫Wf2 D(wS)≤CW∥f∥L∞(Ω)E1(f∣F,f∣F)1/2.\int_W f^2\,D(wS) \leq C_W\|f\|_{L^\infty(\Omega)} \mathcal E_1(f|_F,f|_F)^{1/2}.

The same assertion holds with any permitted smaller capacity domain in place of Ω\Omega, with WW compactly contained in that domain. The energy on the right can still be taken on the original full Ω\Omega.

Status: DIRECT_COROLLARY of the R3 fixed-cutoff square-energy estimate and the faithful coefficient bound. We give the argument to specify its original-test scope.

Proof. Choose original smooth 0≤χ,κ≤10\leq\chi,\kappa\leq1 with χ=1\chi=1 near W‾\overline W, κ=1\kappa=1 near supp⁡χ\operatorname{supp}\chi, and compact support in Ω\Omega. One can first work on a smaller norm-box stage on which the prescribed-domain smooth regularization is available. A finite sum of such stages treats an arbitrary W‾\overline W. The coefficient bound gives ΓS(χ)≤Cχλ\Gamma_S(\chi)\leq C_\chi\lambda.

First suppose that ww is smooth and 0≤w≤20\leq w\leq2 near these supports. Compact Stokes gives ∫χf2D(wS)=−2∫χf ΓS(f,w)−∫f2 ΓS(χ,w).\int\chi f^2D(wS) =-2\int\chi f\,\Gamma_S(f,w) -\int f^2\,\Gamma_S(\chi,w). The original positive cone supplies the measure-valued Cauchy–Schwarz inequality for these energy measures. Moreover ∫κ ΓS(w)=12∫w2 Dκ∧S−∫κw D(wS)≤2 ∣μS,Dκ∣(supp⁡κ).\int\kappa\,\Gamma_S(w) =\frac12\int w^2\,D\kappa\wedge S -\int\kappa w\,D(wS) \leq 2\,|\mu_{S,D\kappa}|(\operatorname{supp}\kappa). This last bound is independent of ww. Thus the two terms in the Stokes formula are bounded in absolute value by a constant times, respectively, ∥f∥∞E(f∣F,f∣F)1/2,∥f∥∞(∫f2ΓS(χ))1/2≤Cχ1/2∥f∥∞∥f∥L2(λ).\|f\|_\infty\mathcal E(f|_F,f|_F)^{1/2}, \qquad \|f\|_\infty \left(\int f^2\Gamma_S(\chi)\right)^{1/2} \leq C_\chi^{1/2}\|f\|_\infty\|f\|_{L^2(\lambda)}. This proves the claim for such smooth ww.

For a general normalized tester use its smooth decreasing classical majorants on a neighborhood of the fixed compact stage. After discarding a prefix they lie in [0,2][0,2] there: their sublevel sets below two are an increasing open cover of that compact stage. The fixed original test χf2\chi f^2 and bounded convergence against the coefficient measures of SS give D(wjS)→D(wS)D(w_jS)\to D(wS) against this test. The preceding bound therefore passes to ww. This passage keeps SS fixed. The same cutoffs and proof inside a smaller capacity domain prove the final assertion; its local energy is at most the global nonnegative energy on Ω\Omega. ◻

Theorem (Energy capacity dominates relative BT outer capacity). For every compact H⋐ΩH\Subset\Omega there is a finite CHC_H such that every subset A⊂HA\subset H, whether Borel or not, satisfies

Cap⁡S∗(A;Ω)≤CH cap⁡E,1(A∩F)1/2.\operatorname{Cap}_S^*(A;\Omega) \leq C_H\, \operatorname{cap}_{\mathcal E,1}(A\cap F)^{1/2}.

In particular every energy-polar subset is locally of zero relative BT outer capacity. Every normalized measure D(wS)D(wS) vanishes on such a set, uniformly in the sense of (eq:r4-energy-BT-bridge).

Status: NEW_PROVED. This is a one-way quantitative comparison. Neither equality of the two capacities nor the reverse inequality is asserted.

Proof. Choose H⊂W⋐ΩH\subset W\Subset\Omega and a fixed cutoff stage as in Lemma “The smooth core controls the BT measures”. First let K⊂W‾∩FK\subset\overline W\cap F be compact. Take f∈Cf\in\mathscr C with f>1f>1 on KK. For ϵ>0\epsilon>0 use a smooth nondecreasing 11-Lipschitz contraction PϵP_\epsilon which fixes [0,1][0,1], vanishes at zero, and has image in [−ϵ,1+ϵ][-\epsilon,1+\epsilon]. It may be chosen constant outside a bounded interval. Applying it to an ambient smooth representative of ff gives an original compact smooth function, still denoted Pϵ(f)P_\epsilon(f), which is at least one on KK. The chain rule gives E1(Pϵ(f),Pϵ(f))≤E1(f,f).\mathcal E_1(P_\epsilon(f),P_\epsilon(f)) \leq\mathcal E_1(f,f). Use the tester estimate on a slightly larger fixed core containing W‾\overline W. For every normalized ww it gives D(wS)(K)≤CH(1+ϵ)E1(f,f)1/2.D(wS)(K) \leq C_H(1+\epsilon)\mathcal E_1(f,f)^{1/2}. Taking the infimum in (eq:r4-energy-compact-core) and then ϵ↓0\epsilon\downarrow0 proves

D(wS)(K)≤CHcap⁡E,1(K)1/2,D(wS)(K)\leq C_H\operatorname{cap}_{\mathcal E,1}(K)^{1/2},

with the same constant for every such KK and every ww.

Let G⊂FG\subset F be any relative open set containing A∩FA\cap F. Choose an ordinary open O⊂ΩO\subset\Omega with O∩F=GO\cap F=G. The measure D(wS)D(wS) is supported on FF, since SS is zero on the ordinary open complement of FF. Its Radon inner regularity, (eq:r4-energy-compact-measure), and energy capacity monotonicity give D(wS)(O∩W)=sup⁡K⋐O∩W∩FD(wS)(K)≤CHcap⁡E,1(G)1/2.D(wS)(O\cap W) =\sup_{K\Subset O\cap W\cap F}D(wS)(K) \leq C_H\operatorname{cap}_{\mathcal E,1}(G)^{1/2}. This holds for every normalized ww, with a common constant. The ordinary open set (O∩W)∪(Ω∖F)(O\cap W)\cup(\Omega\setminus F) contains AA and has the same mass as O∩WO\cap W for every one of these measures. Taking their supremum therefore bounds its raw BT capacity by the same right side. The definition of BT outer capacity, followed by the infimum over GG, proves (eq:r4-energy-BT-bridge). This proof uses inner regularity only for each fixed Radon measure after a uniform compact estimate has been established. It does not interchange an infimum over neighborhoods with a supremum over testers. ◻

Corollary (A canonical relative capacitary representative of energy functions). Every u∈Du\in\mathscr D determines a canonical Borel function class on Ω\Omega modulo local relative outer-capacity zero: it is characterized as the local relative capacity limit of the original smooth core approximations of uu. A selected representative u~\widetilde u agrees with the L2(λ)L^2(\lambda) representative on FF λ\lambda-almost everywhere and can be set to zero off FF. It is locally relatively quasicontinuous. Arbitrary changes on a merely λ\lambda-null set are not asserted to preserve this canonical capacitary class. For each compact H⋐ΩH\Subset\Omega there is a finite CHC_H such that

Cap⁡S∗(H∩{∣u~−v~∣>δ};Ω)≤CHδ−1E1(u−v,u−v)1/2(u,v∈D, δ>0).\operatorname{Cap}_S^* (H\cap\{|\widetilde u-\widetilde v|>\delta\};\Omega) \leq C_H\delta^{-1}\mathcal E_1(u-v,u-v)^{1/2} \quad(u,v\in\mathscr D,\ \delta>0).

The construction is linear in these relative classes and injective from D\mathscr D. No psh or completed-psh membership of u~\widetilde u is asserted.

Status: DIRECT_COROLLARY of the core estimate and elementary capacitary completion.

Proof. For a real original compact smooth ff, choose a smooth P:R→[0,1]P:\mathbf R\to[0,1] which is zero near zero, is one on {∣t∣≥1}\{|t|\geq1\}, and has finite Lipschitz constant CPC_P. Apply the core estimate to P(f/δ)P(f/\delta). Its form norm is at most CPδ−1E1(f,f)1/2C_P\delta^{-1}\mathcal E_1(f,f)^{1/2}. The ordinary open gap W∩{∣f∣>δ}W\cap\{|f|>\delta\} has raw capacity equal to its outer capacity, because it is itself an admissible open neighborhood. Consequently Cap⁡S∗(H∩{∣f∣>δ};Ω)≤CHδ−1E1(f,f)1/2.\operatorname{Cap}_S^* (H\cap\{|f|>\delta\};\Omega) \leq C_H\delta^{-1}\mathcal E_1(f,f)^{1/2}. The same proof works on every permitted compact/domain pair inside Ω\Omega, by the last assertion of the core lemma.

Now take original smooth fj→uf_j\to u in the form norm. The displayed inequality makes this sequence capacity Cauchy on a countably cofinal collection of such pairs. Choose a subsequence with summable capacities of the ordinary open successive gaps, using thresholds tending summably to zero. Outside the union of their tails the sequence converges uniformly on each compact core. The intersection of these tails is relatively outer-null. Assign value zero there. This gives a Borel finite limit and local quasicontinuity; the open complement of FF, itself of zero relative capacity, may be included in every exceptional open set. The same gap inequality proves convergence of the entire sequence and independence of the core approximation.

The reference measure itself is one normalized tester: λ=D(hS)\lambda=D(hS) with 0≤h≤10\leq h\leq1. Hence relative capacity convergence implies local convergence in λ\lambda-measure. It follows that the constructed function agrees λ\lambda-almost everywhere with the prescribed L2L^2 limit. Passing the smooth gap estimate by the level-set triangle inequality proves (eq:r4-energy-convergence-BT). The same argument applied to sums and scalar multiples proves linearity of the resulting classes. If that class is zero, then its L2(λ)L^2(\lambda) representative is zero, so injectivity follows from the definition of D\mathscr D as a subspace of L2(λ)L^2(\lambda). ◻

What the comparison leaves open.

The energy capacity is Choquet by the verified external regular-form theorem. This fact has not been transferred backwards to raw relative BT capacity. In particular the general compact outer-regularity gate for arbitrary SS is not closed by this section. Smooth core approximations are not psh approximations, so the final corollary also does not turn the energy Poisson solution into a member of AS,b\mathcal A_{S,b} or a classical psh solution.

Bounded classical potentials in the energy domain and polar splitting

Use the arbitrary degree-one background SS, the norm-box domain Ω\Omega, and the faithful reference measure λ=D(hS)\lambda=D(hS) of Section “A faithful curvature measure and a coercive relative energy form”. The geometric functions ala_l and hh in that construction can be fixed independently of SS: their moment controls, normalizations, and smooth compensators depend only on the domain. We make this common choice when comparing several backgrounds of the same degree.

Proposition (The closed energy agrees with bounded BT energy). Every u∈Cb(Ω)u\in\mathcal C_b(\Omega) belongs locally to the closed energy domain: χu∈D\chi u\in\mathscr D for every original smooth compact function χ\chi. The same holds for finite differences of such potentials. If uj↓uu_j\downarrow u are the bounded smooth classical regularizations on a neighborhood of the support of χ\chi, then χuj⟶χuin the closed form norm.\chi u_j\longrightarrow\chi u \quad\hbox{in the closed form norm}. The energy measures furnished by these strong limits coincide with the previously defined bounded BT measures ΓS(u,v)\Gamma_S(u,v). Locally these measures are absolutely continuous with respect to λ\lambda. For every original compact scalar ϕ\phi, with χ=1\chi=1 near its support, E(χu,ϕ)=−∫u D(ϕS).\mathcal E(\chi u,\phi)=-\int u\,D(\phi S).

Status: DIRECT_COROLLARY of the proved weighted bounded calculus, its energy identity, and the faithful closable form. Absolute continuity is proved by energy convergence, not inferred from weak convergence of measures.

Proof. Work first on a compact regularization stage containing supp⁡χ\operatorname{supp}\chi. The approximants have one finite bound there and converge at every actual point. For two indices j,kj,k, put rjk=uj−ukr_{jk}=u_j-u_k. The bounded energy identity reads ΓS(rjk)=12D(rjk2S)−rjkD(rjkS).\Gamma_S(r_{jk}) =\tfrac12D(r_{jk}^2S)-r_{jk}D(r_{jk}S). For any two increasing index sequences jl,kl→∞j_l,k_l\to\infty, the first term tends to zero on every fixed compact test by bounded convergence against the coefficients of SS. Expand the second into its four weighted one-potential terms. The weighted joint decreasing theorem makes each tend to uD(uS)uD(uS); their signs cancel. Hence the positive measures ΓS(rjlkl)\Gamma_S(r_{j_lk_l}) tend to zero on compact smooth tests. If their masses against χ2\chi^2 did not tend to zero as min⁡(j,k)→∞\min(j,k)\to\infty, a violating sequence would admit a subsequence with both indices increasing. Thus ∫χ2 ΓS(rjk)⟶0.\int\chi^2\,\Gamma_S(r_{jk})\longrightarrow0. The smooth product square inequality gives E(χrjk,χrjk)≤2∫χ2ΓS(rjk)+2∫rjk2ΓS(χ)⟶0.\mathcal E(\chi r_{jk},\chi r_{jk}) \leq2\int\chi^2\Gamma_S(r_{jk}) +2\int r_{jk}^2\Gamma_S(\chi) \longrightarrow0. The last passage uses the fixed finite coefficient measure ΓS(χ)≤Cχλ\Gamma_S(\chi)\leq C_\chi\lambda and bounded convergence. The L2(λ)L^2(\lambda) limit is χu\chi u. This proves strong form convergence and local membership. Finite compact localization and a partition give the statement for arbitrary χ\chi when more than one regularization stage is needed.

On a fixed smaller compact core the energies of uju_j are bounded by the local CLN energy estimate. The measure Cauchy–Schwarz inequality then gives ∥ΓS(uj)−ΓS(uk)∥TV,H≤(∫HΓS(uj−uk))1/2((∫HΓS(uj))1/2+(∫HΓS(uk))1/2)⟶0,\|\Gamma_S(u_j)-\Gamma_S(u_k)\|_{\rm TV,H} \leq \left(\int_H\Gamma_S(u_j-u_k)\right)^{1/2} \left(\left(\int_H\Gamma_S(u_j)\right)^{1/2} +\left(\int_H\Gamma_S(u_k)\right)^{1/2}\right) \longrightarrow0, where a larger compact smooth cutoff may replace the indicators on the right. Each smooth energy measure is absolutely continuous with respect to λ\lambda by the all-original coefficient bound. Its total-variation limit is therefore also absolutely continuous. The distributional energy identity and weighted decreasing calculus identify this limit with the existing ΓS(u)\Gamma_S(u). Polarization gives the same conclusion for mixed energies and finite differences. The smooth Green and product identities pass through the same strong limits. In particular they give the last displayed formula of the proposition. ◻

The proposition supplies the local product and chain identities used next. For example, for a smooth compact ff and bounded classical aa, the closed energy of fafa is the integral of a2ΓS(f)+2faΓS(f,a)+f2ΓS(a).a^2\Gamma_S(f)+2fa\Gamma_S(f,a)+f^2\Gamma_S(a). This follows by the preceding smooth approximations and measure-valued Cauchy–Schwarz.

Theorem (Orthogonal energy splitting across a complete pole set). Let ϕ≤0\phi\leq0 be an extended classical psh function on Ω\Omega, and put P={ϕ=−∞}P=\{\phi=-\infty\}. Suppose the coefficient restrictions SP=1PS,SPc=1Ω∖PSS_P=\mathbf1_PS,\qquad S_{P^c}=\mathbf1_{\Omega\setminus P}S are separately closed, as supplied by the proved pole restriction theorem. Use the same geometric hh for all three backgrounds. Then λP=D(hSP)=1Pλ,λPc=D(hSPc)=1Pcλ.\lambda_P=D(hS_P)=\mathbf1_P\lambda, \qquad \lambda_{P^c}=D(hS_{P^c})=\mathbf1_{P^c}\lambda. Multiplication by 1P\mathbf1_P and 1Pc\mathbf1_{P^c} are orthogonal projections for both the L2L^2 and the energy inner products on D\mathscr D. More precisely the closed forms have the isometric direct-sum decomposition

(ES,DS)=(ESP,DSP)⊕(ESPc,DSPc)(\mathcal E_S,\mathscr D_S) =(\mathcal E_{S_P},\mathscr D_{S_P}) \oplus(\mathcal E_{S_{P^c}},\mathscr D_{S_{P^c}})

under the Borel carrier identification of the two L2L^2 spaces with L2(F,λ)L^2(F,\lambda). The support spaces of the two components may intersect topologically; disjoint topological supports are not assumed. The decomposition also holds for each part space D0(W)\mathscr D_0(W), W⋐ΩW\Subset\Omega.

Status: NEW_PROVED for the closed-form splitting. The closedness of the current restrictions is a previously proved input, not a new extension theorem in this section.

Proof. For every smooth geometric compensator ala_l, D(alSP)=Dal∧SP=1P(Dal∧S).D(a_lS_P)=Da_l\wedge S_P =\mathbf1_P(Da_l\wedge S). The positive summable curvature construction therefore gives the asserted identities for the reference measures, and similarly for PcP^c.

Set bN=max⁡{0,1+ϕ/N},N≥1.b_N=\max\{0,1+\phi/N\},\qquad N\geq1. These are actual bounded classical functions, with 0≤bN≤10\leq b_N\leq1, zero on PP, and increasing to one on PcP^c. All their bounded products over SPS_P vanish: bNSP=bN2SP=0b_NS_P=b_N^2S_P=0 and hence D(bNSP)=ΓSP(bN)=0D(b_NS_P)=\Gamma_{S_P}(b_N)=0. For any nonnegative original compact smooth χ\chi, the bounded energy identity over SPcS_{P^c} gives 0≤∫χΓSPc(bN)=12∫bN2Dχ∧SPc−∫χbND(bNSPc)≤12∫bN2Dχ∧SPc.0\leq\int\chi\Gamma_{S_{P^c}}(b_N) =\tfrac12\int b_N^2D\chi\wedge S_{P^c} -\int\chi b_N D(b_NS_{P^c}) \leq\tfrac12\int b_N^2D\chi\wedge S_{P^c}. The last measures before multiplication by bN2b_N^2 are fixed coefficient measures and give zero mass to PP. Bounded convergence and separate closedness make the last expression tend to 12⟨SPc,Dχ⟩=0\tfrac12\langle S_{P^c},D\chi\rangle=0. Consequently

∫χΓS(bN)⟶0.\int\chi\Gamma_S(b_N)\longrightarrow0.

No continuity of an increasing mixed Monge–Ampere sequence has been used in this deduction.

For an original smooth compact ff, the preceding proposition makes fbNfb_N a vector of D\mathscr D. The product square inequality and ΓS(bN−bM)≤2ΓS(bN)+2ΓS(bM)\Gamma_S(b_N-b_M)\leq2\Gamma_S(b_N)+2\Gamma_S(b_M) show that these vectors are form Cauchy. Indeed, their derivative terms tend to zero by (eq:r4-pole-selector-zero-energy), and their terms against the fixed ΓS(f)\Gamma_S(f) tend to zero by bounded convergence. Their L2L^2 limit is f1Pcf\mathbf1_{P^c}. Thus both f1Pcf\mathbf1_{P^c} and f1Pf\mathbf1_P belong to D\mathscr D.

The product identity and Cauchy–Schwarz, again using (eq:r4-pole-selector-zero-energy), give ES(f1Pc,f1Pc)=∫PcΓS(f)=ESPc(f,f),\mathcal E_S(f\mathbf1_{P^c},f\mathbf1_{P^c}) =\int_{P^c}\Gamma_S(f)=\mathcal E_{S_{P^c}}(f,f), and the analogous equality on PP. For two smooth compact functions f,gf,g, approximate the two carrier products by f(1−bN)f(1-b_N) and gbNgb_N. In their mixed energy, every term involving ΓS(bN)\Gamma_S(b_N) or a mixed derivative with bNb_N tends to zero by the same estimate and Cauchy–Schwarz. The remaining term is ∫bN(1−bN)ΓS(f,g)\int b_N(1-b_N)\Gamma_S(f,g), which tends to zero by bounded convergence against a fixed coefficient measure. Hence the two carrier products are energy orthogonal.

For smooth core functions, multiplication by either indicator is therefore a contraction in the form norm. It extends continuously to all of D\mathscr D, and its L2L^2 action is still multiplication by that indicator. Their sum is the identity and they are orthogonal projections. The range for PP is the form closure of the vectors f1Pf\mathbf1_P for original core ff. The displayed energy and reference-measure identities identify that closure isometrically with the core closure defining DSP\mathscr D_{S_P}; the other range is identical with the PcP^c form domain. This proves (eq:r4-energy-orthogonal-sum).

If ff has compact support in WW, the approximating products fbNfb_N have that same compact support. Their smooth regularizations multiplied by ff are original core functions supported in WW. Thus the whole argument applies to the part-form closures, proving the final assertion. ◻

Corollary (Canonical polar and nonpolar energy components). On a norm-box exhaustion domain where the proved canonical decomposition S=Spp+SnpS=S_{\rm pp}+S_{\rm np} is represented by one complete pole set, the energy form and its zero-boundary linear Poisson problem split orthogonally into those two components. The splitting is independent of the chosen pole witness. In particular the energy solution for f∈L2(λ)f\in L^2(\lambda) is the direct sum of the solutions for the restricted data fλppf\lambda_{\rm pp} and fλnpf\lambda_{\rm np}.

Proof. Apply the theorem to the complete-pole representative of the canonical current decomposition. Current uniqueness identifies the two components independently of its witness. The part-space orthogonal sum identifies the two Riesz equations componentwise, so uniqueness of the energy solutions proves the last assertion. ◻

A boundary of this conclusion.

The energy splitting does not prove additivity of raw relative BT capacities for Spp+SnpS_{\rm pp}+S_{\rm np}. The vectors 1Pf\mathbf1_Pf belong to a linear energy domain; they have not been shown to admit global bounded classical psh defining rows. In particular, replacing the ambient psh tester class by all positive relative-energy-curvature functions would change the capacity in the research problem. The distinction is visible in the smooth nonpolar Dirichlet counterexample and its explicit energy solution.

An explicit solution beyond the ambient psh constraint

The smooth nonpolar counterexample has a positive replacement: its equation admits a bounded original smooth solution with the requested ordinary boundary values when the ambient psh constraint is removed. This also tests the energy construction on all original currents, not only on toric scalar tests.

Retain V=t−1(B)V=t^{-1}(B), B={∣x∣<1}⊂RnB=\{|x|<1\}\subset\mathbf R^n, n≥2n\geq2, β=∑id′ti∧d′′ti\beta=\sum_i d't_i\wedge d''t_i, and S=[βn−1]S=[\beta^{n-1}] from Section “A smooth nonpolar background can still obstruct completed solvability”. Choose 0<r0<10<r_0<1 and a smooth nonnegative radial function f(x)=φ(∣x∣2)f(x)=\varphi(|x|^2) supported in ∣x∣<r0|x|<r_0, with 0≤f≤10\leq f\leq1 and f≢0f\not\equiv0. Write f(r)f(r) for its radial value. Set

A(r)=n∫0rsn−1f(s) ds,U(r)=−∫r1τ1−nA(τ) dτ,u=U(∣t∣).A(r)=n\int_0^r s^{n-1}f(s)\,ds,\qquad U(r)=-\int_r^1\tau^{1-n}A(\tau)\,d\tau, \qquad u=U(|t|).

Proposition (An ordinary-boundary solution in the energy space). The function uu is bounded and original smooth on a neighborhood of V‾\overline V, is nonpositive on VV, and has ordinary boundary value zero. It satisfies D(uS)=f(t)βnD(uS)=f(t)\beta^n against all original compact smooth scalar tests on VV. It is not classical psh. In fact, the equation has no solution with zero ordinary boundary limit in the global bounded classical capacity closure GS,b(V)\mathscr G_{S,b}(V).

Choose a larger ordinary algebraic norm box Ω\Omega containing V‾\overline V, and construct λ=D(hS)\lambda=D(hS) there by Theorem “A faithful curvature measure dominating all coefficients”. Then u∈D0(V)u\in\mathscr D_0(V) and is the unique energy solution of this equation. Its datum has a bounded density against λ\lambda.

Status: NEW_PROVED for the explicit non-psh energy solution and the comparison of the two solution classes. The nonsolvability in the bounded classical closure is the previously proved counterexample.

Proof. The integrand in (eq:r4-explicit-radial-potential) is O(τ)O(\tau) as τ↓0\tau\downarrow0, so UU is finite. More precisely, U′(r)=r1−nA(r),U′′(r)+n−1rU′(r)=nf(r)(r>0).U'(r)=r^{1-n}A(r),\qquad U''(r)+\frac{n-1}{r}U'(r)=nf(r)\quad(r>0). Since f(r)=φ(r2)f(r)=\varphi(r^2) is smooth and even near zero, r1−nA(r)r^{1-n}A(r) is rr times a smooth function of r2r^2. Thus x↦U(∣x∣)x\mapsto U(|x|) is smooth also at zero. The same formula extends smoothly beyond r=1r=1, because ff vanishes near that sphere. It gives U(1)=0U(1)=0 and U≤0U\leq0 in BB. Properness of the torus moment map over compact real sets and continuity prove the ordinary boundary assertion for uu.

As an identity of original forms, D(U(∣t∣))∧βn−1=ΔxU(∣t∣)nβn=f(t)βn.D\bigl(U(|t|)\bigr)\wedge\beta^{n-1} =\frac{\Delta_xU(|t|)}n\beta^n =f(t)\beta^n. The off-diagonal Hessian terms vanish upon wedging with βn−1\beta^{n-1}, and each diagonal term contributes the same factor 1/n1/n. This identity of forms proves the current equation on the full original test sheaf.

Let c=A(r0)>0c=A(r_0)>0. For r0<r<1r_0<r<1, U′(r)=cr1−n,U′′(r)=−(n−1)cr−n<0.U'(r)=c r^{1-n},\qquad U''(r)=-(n-1)c r^{-n}<0. At (r,0,…,0)(r,0,\ldots,0) this is the second derivative in the first coordinate. The full-annulus trace argument in Lemma “Original torus integration and separate convexity” would force this restriction to be convex for a classical psh function. Thus uu is not classical psh. The same datum is nonzero, compactly supported and bounded by the smooth matching subsolution measure D(12(∣t∣2−1)S)D(\tfrac12(|t|^2-1)S), so Theorem “Nonpolar bounded-subsolution failure” excludes every solution in GS,b(V)\mathscr G_{S,b}(V) with these boundary values.

We finally verify the asserted energy boundary condition, which does not follow merely from a pointwise boundary limit in an arbitrary domain. In this paragraph the Hilbert-space vector denoted by uu means its zero extension from VV to Ω\Omega, rather than its smooth continuation beyond the boundary. Choose smooth radial cutoffs χϵ(t)\chi_\epsilon(t) which equal one for ∣t∣≤1−2ϵ|t|\leq1-2\epsilon, vanish for ∣t∣≥1−ϵ|t|\geq1-\epsilon, take values in [0,1][0,1], and have derivative bounded by C/ϵC/\epsilon. Then uϵ=χϵuu_\epsilon=\chi_\epsilon u is an original smooth compact function on VV. The original torus calibration gives, for every smooth function G(t)G(t), ΓS(G(t))=∣∇G(t)∣2nβn=(n−1)! σ∗(∣∇G∣2dx).\Gamma_S(G(t))=\frac{|\nabla G(t)|^2}{n}\beta^n =(n-1)!\,\sigma_*(|\nabla G|^2dx). Near ∣t∣=1|t|=1, the already displayed formulas give ∣u∣≤C(1−∣t∣)|u|\leq C(1-|t|) and ∣∇u∣≤C|\nabla u|\leq C. The gradients of u−uϵu-u_\epsilon are consequently uniformly bounded on their collar of real volume O(ϵ)O(\epsilon). Applying this estimate first to two compact core functions gives E(uϵ−uδ,uϵ−uδ)≤Cmax⁡(ϵ,δ).\mathcal E(u_\epsilon-u_\delta,u_\epsilon-u_\delta) \leq C\max(\epsilon,\delta). Thus the compact core sequence is energy Cauchy. Its L2(λ)L^2(\lambda) limit is uu: λ\lambda is finite on the compact V‾\overline V, and bounded pointwise convergence applies to the zero extensions. Hence u∈D0(V)u\in\mathscr D_0(V); passing δ↓0\delta\downarrow0 in the displayed inequality also gives E(u−uϵ,u−uϵ)=O(ϵ)\mathcal E(u-u_\epsilon,u-u_\epsilon)=O(\epsilon).

Take an original compact smooth scalar equal to one near V‾\overline V. Applying all-coefficient domination to its product with β\beta gives βn∣V≤Cλ\beta^n|_V\leq C\lambda. The datum f(t)βnf(t)\beta^n therefore has a bounded, nonnegative λ\lambda-density, and is in particular an L2L^2 datum. The current equation just proved, the core Green identity, and form density identify uu with the unique solution in Theorem “Coercive energy solutions of the relative Poisson equation”. ◻

This example isolates the analytic issue. Positivity of D(uS)D(uS) for a fixed (n−1,n−1)(n-1,n-1)-background is a linear subharmonicity condition relative to SS; it need not impose ambient plurisubharmonicity. The energy space can solve that linear problem, while the ambient psh constraint may rule out the same boundary datum. No conclusion about the entire locally defined sheaf AS,b\mathcal A_{S,b} follows from the global defining-row obstruction.

Fourth-round correspondence and current proof boundaries

The original theorem-by-theorem literature comparison remains Section “Theorem-by-theorem comparison with the classical theory”. Section “Current theorem correspondence and the advance beyond R2” and the frozen R3 proof map are the baseline for this continuation. This section records changes in mathematical status; it does not count the accepted results of R, B, A2, or R1–R3 as new proofs. A2’s geometric adelic nefness theorem remains a completed input under its stated global hypotheses.

Throughout, K/QpK/\mathbf Q_p is the prescribed finite extension with its original absolute value. Unless a theorem explicitly specializes further, SS is a fixed symmetric original-positive (p,p)(p,p)-current, separately closed under d′d' and d′′d'', and all identities concern the full original compact smooth CLD test sheaf. No wedge of two arbitrary positive closed currents, unrestricted varying-background weighted continuity, or equality on every GK δ\delta-test is asserted.

The two degree ranges must be distinguished

The pole-carrier capacity theorem, simultaneous tester pasting, completed strong-image results and constrained psh variational theorems apply to every remaining degree q=n−p≥1q=n-p\geq1, with the individual domain, bound and collar hypotheses stated in their sources. The new faithful Hilbert energy form and linear Poisson theory instead concern an arbitrary (n−1,n−1)(n-1,n-1)-background, that is, remaining degree one. They can be applied after freezing q−1q-1 bounded differentiated slots of a higher-degree problem. This gives a linear theory for each such fixed mixed background; it is not a uniform nonlinear theory while those slots vary.

All eleven Bedford–Taylor directions

Direction

Frozen result through R3

This-round advance

Remaining status

1. Locally bounded psh

B’s actual classical class, the ambient representative theorem, and the arbitrary-relative representative counterexample were already established.

Bounded classical potentials now embed locally in the faithful closed energy form, with strong form convergence of their smooth decreasing rows. General form vectors have a canonical relatively quasicontinuous class, specified by their core approximations. See Proposition “The closed energy agrees with bounded BT energy” and Corollary “A canonical relative capacitary representative of energy functions”.

Neither energy membership nor relative completion supplies a general ambient classical representative. The unrestricted representative assertion remains FALSE.

2. Positive closed currents and mixed products

The arbitrary-fixed-SS bounded and completed BT calculi, and finite-coefficient-mass zero extension across closed locally complete pole sets, were already proved.

Compatible original piecewise tests admit uniform approximation. Finite etale trace, disc/polydisc projection and arbitrary same-KK analytic morphisms now preserve the bounded potential DD-kernel. Exact test spreading further proves kernel transfer to finite extensions of finite-polyradius Gauss fields; see Theorem “Gauss residue-field preservation of the potential kernel” and Corollary “Finite extensions after a Gauss extension”.

The quotient for every original positive closed background remains OPEN. These are assertions about specified potentials, not general current pullbacks.

3. CLN inequalities

Finite-test relative CLN was already a DIRECT_COROLLARY of R/B. R3 added the all-degree energy-to-capacity estimate.

New wall-normal mass and normal-factor estimates have orders O(ϵ)O(\epsilon) and O(ϵ1/2)O(\epsilon^{1/2}). In degree one, a bounded continuous psh curvature measure dominates every original compact coefficient of SS; see Theorem “A faithful curvature measure dominating all coefficients”.

Constants retain finite local controls and the stated geometry. No global strictly psh exhaustion is presumed.

4. Symmetry and weighted monotone convergence

The requested joint decreasing theorem, completed capacity continuity and localized strong psh-dual convergence were already complete.

Theorem “Uniform approximation of compatible piecewise tests” adds weak continuity on tangentially compatible original piecewise tests for positive closed currents. Completed collar energy inequalities and strong image completeness are direct consequences of the frozen calculus.

Arbitrary piecewise jumps fail. Unrestricted varying-background weighted stability and preservation of nonlinear weak limits by convex blocks remain FALSE.

5. Locality and maxima

Ordinary locality, Borel maximum locality and the relative maximum/contact measure inequality were already proved.

Small global psh compensators make one maximum branch strictly active on each separated compact piece. The same construction works outside controlled capacity-error sets; see Theorem “Pasting uniformly close normalized testers” and Proposition “Pasting outside a small capacity error”.

An arbitrary strict gap is not ordinarily open. The true maximum measure inequality does not supply the false unrestricted relative obstacle-contact formula.

6. Bedford–Taylor capacities

Ambient core Choquet theory, degree-one nonpolar compact regularity, the general degree-one completed compact formula, separated analytic-carrier additivity and positive tail approximation were already proved.

Theorem “Finite pole-carrier additivity” replaces analytic carriers by complete pole carriers with mutual relative nullness, in every degree. Close-tester pasting repairs an actual toric subfamily defect. The energy capacity controls relative BT outer capacity in one direction.

General sharp compact control C1 and general Borel outer regularity C2 remain OPEN. The energy capacity is not identified with raw BT capacity.

7. Quasicontinuity and capacity convergence

Relative bounded quasicontinuity, monotone capacity convergence, and equivalence of common-collar energy and strong data topologies were established.

Form-norm convergence now gives local relative outer-capacity convergence for canonical energy representatives. The completed image theorems retain one global defining bound and an actual collar or exterior under their limits.

Energy representatives need not be psh. The whole normalized psh family is still not capacity precompact.

8. Pluripolar and negligible sets

Qualified ambient polar/negligible/capacity-zero equivalences, pole restriction, and canonical current polar decomposition were available.

Theorem “Orthogonal energy splitting across a complete pole set” gives an orthogonal closed-form decomposition across a complete pole set, even when the two topological supports overlap. Energy-polar sets are locally relative BT outer-null.

Relative BT measures may still charge ambient polar sets. Energy splitting does not imply raw capacity additivity for an arbitrary polar/nonpolar decomposition.

9. Comparison and domination

Fixed-SS compact-gap comparison and relative q.e. domination, plus the qualified ambient Dini stability theory, were proved.

The constrained variational solutions obey the exact monotonicity bound IS(aν,aν′)≤4B dpsh(ν,ν′)I_S(a_\nu,a_{\nu'})\leq4B\,d_{\rm psh}(\nu,\nu'). The completed interior image has relative-q.e. uniqueness with the prescribed exterior and barriers.

No general relative pointwise uniqueness is deduced. Ordinary weak data remain insufficient even under uniform power-capacity domination.

10. Relative extremal functions

Ambient obstacle orthogonality, extremal mass formulas and relative capacitary envelopes were retained; unrestricted relative contact formulas had explicit counterexamples.

Theorems “The constrained relative variational problem” and “Strong stability and a retraction onto the image” give a unique constrained energy maximizer and a continuous retraction onto its actual Monge–Ampere image, for capacity-continuous datum pairings.

The Euler output is an inequality. A pointwise obstacle gap does not justify arbitrary two-sided psh variations or unrestricted orthogonality.

11. Monge–Ampere and Dirichlet problems

The qualified ambient bounded-subsolution and Dini-capacity existence theorems, their boundary extensions, and finite-carrier completed gluing were established. Arbitrary-relative classical existence and its TV-closed image were already false.

Specified completed collar/interior images are strong-metric complete and closed. A new smooth nonpolar degree-one example rules out bounded-subsolution existence in the global bounded-row psh completion. The coercive linear energy space does solve its relative Poisson problem, and an explicit radial solution verifies the distinction.

General completed-psh solvability remains OPEN; its unrestricted global-row subsolution version is FALSE. Broader ordinary boundary data still need proved hypotheses.

What is new, and what is a direct consequence
The original projected-test obstruction has been reduced further.

The new wall estimates, compatible piecewise approximation, split-moment projection formula, and finite etale original trace are NEW_PROVED. Finite etale preservation of the bounded kernel is their DIRECT_COROLLARY. Rational localization, finite etale splitting away from a discriminant, and removal of that fixed analytic exceptional locus prove the full ordinary-disc kernel theorem in its smooth affine algebraic-neighborhood scope. The polydisc version is a direct consequence. Thus the earlier conditional projected-test criterion is now discharged in that scope, not for every imaginable geometric pushforward. The separately proved same-KK analytic-morphism theorem is also NEW_PROVED: an actual analytic translation and a compact zero-tube argument reduce its graph to the established zero-section trace. Product preservation and the analytic-immersion background corollary are direct consequences. This enlarges the proved functorial scope without upgrading the arbitrary-background quotient or, by itself, settling arbitrary scalar extension.

Which scalar extensions are now covered.

Finite K′/KK'/K and analytic morphisms over such finite extensions are covered by Corollary “Finite extensions with their inherited calibration”, a DIRECT_COROLLARY with the inherited absolute value and the verified finite-constant calibration. Lemma “Exact spreading from an algebraic parameter fiber” is NEW_PROVED: for any point aa of a smooth affine algebraic parameter variety, every original compact H(a)\mathscr H(a)-test on a scalar extension of an original-KK space is the exact restriction of a compact original-KK test on an ordinary parameter neighborhood. It uses neither regularization nor the bounded BT calculus over H(a)\mathscr H(a).

The full Gauss-fiber current formula is EXTERNAL_KNOWN, specifically CLD25 Proposition 18.6.1 and Corollary 18.6.2, printed pp. 532–534 [CLD25, 18.6.1–18.6.2]. Its fixed Borel-weighted form is a direct Radon-measure consequence. Together with exact spreading and the proved original-KK ambient-product quotient, it gives DIRECT_COROLLARY kernel transfer to Kr=H(ζr),L/Kr finite,K_{\mathbf r}=\mathscr H(\zeta_{\mathbf r}), \qquad L/K_{\mathbf r}\text{ finite}, for every finite positive real polyradius r\mathbf r. The finite-over-Gauss proof returns to a parameter space over the original KK and retains the one factor [L:Kr][L:K_{\mathbf r}] when comparing the two calibrations of the same fiber. All uses of R/B and bounded BT products remain over the original finite field. These statements do not cover arbitrary complete fields or every analytic morphism defined over KrK_{\mathbf r} or LL.

Lemma “Exact finite-coefficient descent of original tests” is a further NEW_PROVED finite-test statement: finitely many original tests over any complete extension L/KL/K descend exactly to E=K(c1,…,cm)‾≃H(a)E=\overline{K(c_1,\ldots,c_m)}\simeq\mathscr H(a) for one finite-dimensional algebraic parameter point. The scalar integration calibration has factor one [CLD25, Proposition 15.3.1, p. 425]; its fixed Borel-weighted version is a direct Radon-measure consequence. These facts prove the precise CONDITIONAL reduction in Theorem “Conditional reduction to algebraic parameter residue fields”: scalar kernel transfer for every such H(a)\mathscr H(a) is equivalent to scalar kernel transfer for every complete extension. The premise for general non-Gauss parameter points remains unproved. The reduction does not add extension-field analytic-map functoriality or settle the arbitrary-background quotient.

A faithful analytic energy object, with a different solution class.

Theorem “A faithful curvature measure dominating all coefficients” constructs λ=D(hS)\lambda=D(hS), 0≤h≤10\leq h\leq1, with the same Borel coefficient-null ideal as SS. The new closed form is regular, strongly local and coercive: E(f,f)≥∫f22−h dλ≥12∥f∥L2(λ)2.\mathcal E(f,f)\geq \int\frac{f^2}{2-h}\,d\lambda \geq\tfrac12\|f\|_{L^2(\lambda)}^2. This proves a full-original-test linear equation D(uS)=fλD(uS)=f\lambda on the specified part domain, with a form boundary condition. An ordinary pointwise boundary limit is verified separately for the explicit radial example; it is not assumed for every energy solution. The bounded-classical embedding is a DIRECT_COROLLARY; the closed-form pole splitting is NEW_PROVED.

The new quantitative comparison Cap⁡S∗(A;Ω)≤CH cap⁡E,1(A∩supp⁡S)1/2,A⊂H⋐Ω,\operatorname{Cap}_S^*(A;\Omega) \leq C_H\, \operatorname{cap}_{\mathcal E,1} (A\cap\operatorname{supp}S)^{1/2}, \qquad A\subset H\Subset\Omega, uses the external regular-form compact-core formula with its verified hypotheses. That formula and the energy capacity’s Choquet property are EXTERNAL_KNOWN, not new BT results. The resulting canonical energy representative is fixed by smooth core limits. Merely changing values on a λ\lambda-null set need not preserve that capacitary class. The coefficient-null ideal, the energy-capacity-null ideal and the BT outer-capacity-null ideal are not identified. In particular the proved local absolute continuity of ΓS(u,v)\Gamma_S(u,v) does not assert D(uS)≪λD(uS)\ll\lambda. The pole projections act on the L2L^2 form domain; their existence for a complete pole set is not a theorem about arbitrary Borel indicators.

All-degree capacity and constrained variation.

General complete-pole-carrier additivity and close normalized tester pasting are NEW_PROVED; outer additivity, positive-tail extensions and the capacity-error pasting corollaries are direct consequences of their stated mechanisms. The new constrained variational theorem assumes a nonempty capacity-closed convex subset of one globally bounded completed collar class and datum pairings continuous in outer capacity. It proves the exact Euler inequality and a retraction onto the actual solution image. It does not turn every admissible datum into a Monge–Ampere datum.

The collar and interior strong-image closure theorems are DIRECT_COROLLARY of R3 energy and R2 completeness. For the interior theorem, selected representatives share an actual exterior including ∂V\partial V. Thus a−b=0a-b=0 there, and IS(a,b)=∫K(a−b) d(νb−νa)I_S(a,b)=\int_K(a-b)\,d(\nu_b-\nu_a) even when the full measures have different boundary masses. Continuous upper and lower barriers retain the ordinary boundary value; a cutoff supported strictly inside VV gives the final strong convergence of interior data.

False assertions and excluded proof routes
  1. Unrestricted bounded-subsolution existence in the global bounded-row completion is false. Theorem “Nonpolar bounded-subsolution failure” uses a smooth nonpolar (n−1,n−1)(n-1,n-1)-background, n≥2n\geq2, nonatomic compact data and a continuous matching subsolution. Its scope is the completion by rows defined on the whole full torus-preimage domain with one bound. It does not establish nonexistence in the entire locally defined sheaf AS,b\mathcal A_{S,b}.

  2. Arbitrary bounded piecewise tests are insufficient. Example “Tangential matching cannot be omitted” has a tangential jump whose pairings fail to converge along uniformly convergent continuous potentials. Compatibility on moment faces is an essential hypothesis of the new test theorem.

  3. A full Gauss fiber is not its independent Gauss section. In Example “A single Gauss section misses correlated fiber points”, an original test depending on x−zx-z has positive full-fiber pairing while vanishing on every independent radius-one Gauss lift. Correlated points with ∣x−z∣<1|x-z|<1 must be retained. The shortcut is FALSE; the external full-fiber identity itself is not in question.

  4. Separation cost cannot be ignored. The annulus calculation in Section “Two simultaneous-pasting mechanisms for relative capacity” forces the bounded compensator size to grow at least as ϵ−1−1\epsilon^{-1}-1 for the displayed nearby compact pieces. Pasting near-maximizers with moving geometry therefore requires control of the product AKδA_K\delta.

  5. Regular energy capacity does not settle C2. Even a weakly compact, uniformly finite-energy restricted family can have a raw/outer defect. The actual vertical-fiber subfamily in Example “An actual toric subfamily and its missing testers” has values 1/21/2 and 11 on the displayed Borel set. The full BT tester class repairs this particular defect by the new pasting theorem. It is not a counterexample to full BT outer regularity.

  6. Energy selectors are not psh selectors. The orthogonal projections 1P\mathbf1_P and 1Pc\mathbf1_{P^c} act on a linear form domain. This does not provide both selectors in a cone of global bounded classical psh limits, or prove raw capacity additivity for a nested polar/nonpolar pair. Likewise positivity of D(uS)D(uS) need not make uu ambient psh; Proposition “An ordinary-boundary solution in the energy space” gives a smooth explicit example.

  7. The frozen weak-limit and quotient warnings remain. Convex blocks need not preserve nonlinear weak Monge–Ampere limits; ordinary weak data need not give uniform or capacity convergence of solutions; actual-potential faithfulness does not identify the distributional DD-kernel. Same-field kernel equalities are not silently extended to new tests after arbitrary complete scalar extensions.

Minimal remaining gates and the next proof tasks
Q: arbitrary-background potential quotients — OPEN.

Finite etale and disc/polydisc kernel preservation, followed by arbitrary same-KK analytic-morphism functoriality, are now proved. So are finite-KK extension functoriality and scalar kernel transfer to finite extensions of finite-polyradius Gauss fields. The latter result is not a theorem for arbitrary maps over those Gauss fields. These statements do not establish the all-complete-extension curve condition or a continuous representative of every bounded classical difference in the original DD-kernel. The precise goal remains D[u]=D[v]⇒D((u−v)S)=0D[u]=D[v]\Rightarrow D((u-v)S)=0 for every allowed fixed SS, beyond the already established immersion and weak-closure background classes. A valid regularity mechanism or a direct arbitrary-background annihilation theorem is still needed. Exact spreading applies at general algebraic parameter points, but does not provide the needed bounded ambient-BT representation of every such fiber. Density of Gauss points does not settle that fixed-fiber, Borel-weighted issue. The exact finite-coefficient descent now reduces the scalar-extension part to those finite-dimensional parameter residue fields, with the full quantifier over parameter points still an OPEN premise.

C1: sharp compact weak-closure domination — OPEN.

For the normalized measure family on one fixed capacity domain, prove ν(K)≤CS(K;Ω)\nu(K)\leq C_S(K;\Omega) for every weak-closure measure ν\nu and every compact core KK. The degree-one nonpolar case and the proved carrier/positive-tail classes remain available; no density theorem says they exhaust arbitrary positive closed backgrounds. Completion, energy compactness and strong-image closedness do not supply this sharp raw classical-test bound.

The final focused compact-image check sharpens the quantifiers without adding a theorem beyond R3. In remaining degree one, on the same full normalized domain, MA=Mcl‾ local vague,CA(K)=Ccl∗(K)(K⋐Ω).\mathcal M_{\mathcal A} =\overline{\mathcal M_{\rm cl}}^{\,\mathrm{local\ vague}}, \qquad C_{\mathcal A}(K)=C_{\rm cl}^*(K)\quad(K\Subset\Omega). The completed image is locally vaguely compact and its compact mass maximum is attained. A fixed compact cutoff gives a weakly compact family of finite measures; global total mass need not be finite. These facts recover Corollary “The outer capacity on compact sets is the completed capacity”. The missing claim is still: for each compact KK and each ϵ>0\epsilon>0, there exists a normalized classical tester on the same domain with mass on KK at least CA(K)−ϵC_{\mathcal A}(K)-\epsilon. A classical representative of the completed maximizer is not logically necessary, but weak density alone does not give this lower approximation of compact mass. The route is recorded as a FAILED_ATTEMPT to close C1/C2, not a new compact-regularity proof.

C2: ordinary-open Borel outer regularity — OPEN.

One must still prove the additional increasing-set/outer-capacity step beyond compact control, or the equivalent Borel weak-closure identification of Proposition “The exact missing outer-regularity implication”. A concrete next target is simultaneous extraction from open near-maximizers into finitely many separated compact pieces whose testers have a common capacity limit, with the pasting loss and AKδA_K\delta tending to zero. The new pasting lemmas prove the final combination step, not that extraction statement.

D: a verifiable relative solvability criterion — OPEN.

The specified completed images are now closed and carry exact relative-q.e. uniqueness. The constrained variational map is available. The missing step is a useful independent criterion for a datum to be fixed by that map, or another appropriate solution class with justified boundary behavior. A universal bounded-subsolution criterion in the global bounded-row psh completion is excluded by the new example. In degree one the faithful energy space is a proved alternative; its nonlinear extension in changing mixed degree is not supplied here.

Comparison with CLD, GK and arithmetic applications.

All new identities remain ordinary CLD-current identities. A comparison on the entire GK δ\delta-test sheaf needs its own theorem, and the nonpositive, nonclosed, off-diagonal GK calculus is not subsumed by these results. No change to the completed A2 nefness input is required. Strong Green forms and heights remain subsequent applications, not replacements for Q, C1, C2 or the remaining Dirichlet criteria.

The continuation should address these gates directly, using the moving-atom, crossing-support, nonpolar full-torus, tangential-jump and vertical-fiber examples as scope tests. Reproving the frozen bounded BT calculus would not resolve them.

References

  1. [BT76]

    E. Bedford and B. A. Taylor, The Dirichlet problem for a complex Monge–Ampère equation, Invent. Math. 37 (1976), 1–44. https://www.math.stonybrook.edu/~ebedford/PapersForM655/BT1.pdf.

  2. [BT82]

    E. Bedford and B. A. Taylor, A new capacity for plurisubharmonic functions, Acta Math. 149 (1982), 1–40. https://www.math.stonybrook.edu/~ebedford/PapersForM655/BTNewCapacity.pdf.

  3. [Dem12]

    J.-P. Demailly, Complex Analytic and Differential Geometry, online book, version June 21, 2012. https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf.

  4. [CLD12]

    A. Chambert-Loir and A. Ducros, Formes différentielles réelles et courants sur les espaces de Berkovich, arXiv:1204.6277v1 (2012). https://arxiv.org/pdf/1204.6277v1.

  5. [CLD25]

    A. Chambert-Loir and A. Ducros, Formes différentielles réelles et courants sur les espaces de Berkovich, arXiv:1204.6277v2 (2025), 608 pp. https://arxiv.org/pdf/1204.6277v2.

  6. [GK17]

    W. Gubler and K. Künnemann, A tropical approach to non-Archimedean Arakelov geometry, Algebra Number Theory 11 (2017), 77–180. The theorem numbering used here is checked in arXiv:1406.7637v3 (2016). https://arxiv.org/pdf/1406.7637v3.

  7. [GR25]

    W. Gubler and J. Rabinoff, Classically psh and pluriharmonic functions on Berkovich spaces, arXiv:2506.13548v2 (September 17, 2025). https://gubler.app.uni-regensburg.de/Publikationen/2506.13548.pdf.

  8. [Wu26]

    L. Wu, On the local theory of non-Archimedean psh functions, arXiv:2609.27610v1 (September 23, 2026). https://arxiv.org/pdf/2609.27610v1.

  9. [BGJK21]

    J. I. Burgos Gil, W. Gubler, P. Jell, and K. Künnemann, Pluripotential theory for tropical toric varieties and non-Archimedean Monge–Ampère equations, arXiv:2102.07392v1 (February 15, 2021). The cited theorem numbers refer to this author-hosted version. https://gubler.app.uni-regensburg.de/Publikationen/2102.07392.pdf.

  10. [GM19]

    W. Gubler and F. Martin, On Zhang’s semipositive metrics, Doc. Math. 24 (2019), 331–372. https://ems.press/content/serial-article-files/26466.

  11. [Brouwer11]

    L. E. J. Brouwer, Über Abbildung von Mannigfaltigkeiten, Math. Ann. 71 (1911), 97–115. https://doi.org/10.1007/BF01456931.

  12. [DE06]

    K. Dabbek and F. Elkhadhra, Capacité associée à un courant positif fermé, Doc. Math. 11 (2006), 469–486. https://ems.press/content/serial-article-files/26001?nt=1.

  13. [ElMir84]

    H. El Mir, Sur le prolongement des courants positifs fermés, Acta Math. 153 (1984), 1–45. https://doi.org/10.1007/BF02392374.

  14. [Kol96]

    S. Kołodziej, Some sufficient conditions for solvability of the Dirichlet problem for the complex Monge–Ampère operator, Ann. Polon. Math. 65 (1996), 11–21. https://matwbn.icm.edu.pl/ksiazki/apm/apm65/apm6512.pdf.

  15. [Kol98]

    S. Kołodziej, The complex Monge–Ampère equation, Acta Math. 180 (1998), 69–117. https://archive.ymsc.tsinghua.edu.cn/pacm_download/117/6558-11511_2006_Article_BF02392879.pdf.

  16. [Mih24]

    A. Mihatsch, δ\delta-Forms on Lubin–Tate spaces, arXiv:2112.10018v4 (7 September 2024). https://arxiv.org/pdf/2112.10018v4.

  17. [FST91]

    M. Fukushima, K.-I. Sato, and S. Taniguchi, On the closable parts of pre-Dirichlet forms and the fine supports of underlying measures, Osaka J. Math. 28 (1991), 517–535. https://ir.library.osaka-u.ac.jp/repo/ouka/all/6785/ojm28_03_03.pdf.

Source text and edition identity

This reading copy omits personal attribution and places existing mathematical question marks over their relation symbols. The mathematical content and source status are retained.

Reading-copy SHA-256 854252899cbbb065fdd7d93983fecac1f9bb8e232ac0861ababfd444927417c2

Frozen-original SHA-256 854252899cbbb065fdd7d93983fecac1f9bb8e232ac0861ababfd444927417c2

\documentclass[11pt,a4paper]{article}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{lmodern}
\ifdefined\pdfmapfile
  \pdfmapfile{=lm.map}
  \pdfmapfile{+cm.map}
  \pdfmapfile{+symbols.map}
  \pdfmapfile{+euler.map}
  \pdfmapfile{+rsfs.map}
\fi
\usepackage[a4paper,margin=27mm]{geometry}
\usepackage{amsmath,amssymb,amsthm,mathtools,mathrsfs}
\usepackage{booktabs,tabularx,longtable,array,enumitem}
\usepackage{microtype,xcolor,etoolbox}
\usepackage{hyperref}
\hypersetup{colorlinks=true,linkcolor=blue!45!black,citecolor=blue!45!black,
 urlcolor=blue!45!black,pdfauthor={Research record prepared with ChatGPT},pdftitle={Relative Bedford--Taylor theory on Berkovich spaces: fourth consolidated research record}}
\usepackage[capitalise,noabbrev]{cleveref}
\numberwithin{equation}{section}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{corollary}[theorem]{Corollary}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\theoremstyle{remark}
\newtheorem{remark}[theorem]{Remark}
\newcommand{\Cb}{\mathcal C_b}
\newcommand{\R}{\mathbf R}
\newcommand{\Q}{\mathbf Q}
\newcommand{\Z}{\mathbf Z}
\newcommand{\PSH}{\operatorname{PSH}}
\newcommand{\BTcap}{\operatorname{Cap}}
\newcommand{\supp}{\operatorname{supp}}
\newcommand{\dd}{d'd''}
\newcommand{\an}{\mathrm{an}}
\setlist{topsep=4pt,itemsep=2pt,parsep=0pt}
\setlength{\parindent}{1.3em}
\setlength{\parskip}{2pt}
\setlength{\emergencystretch}{2em}
\makeatletter
\patchcmd{\l@section}{1.5em}{2.1em}{}{}
\renewcommand*\l@subsection{\@dottedtocline{2}{1.5em}{3.6em}}
\makeatother

\title{Relative Bedford--Taylor theory on Berkovich spaces\\[5pt]
\large Fourth research record: original-test functoriality,\\
faithful energy forms, capacities, and limits of relative solvability}
\author{Research record prepared with ChatGPT\\
\small Based on the authorized manuscripts R, B, and A2}
\date{11 November 2026 (UTC+08)}
\begin{document}
\maketitle
\begin{abstract}
This continuation retains the arbitrary fixed-background bounded
Bedford--Taylor calculus proved in the preceding records. It develops
uniform approximation of tangentially compatible original piecewise
CLD tests, reconstructs finite etale traces from finitely many Newton
polygons, and proves preservation of the bounded potential curvature
kernel under disc projections and arbitrary analytic morphisms over
the prescribed original field. Exact spreading of original tests from
algebraic parameter fibers, combined with the known full Gauss-fiber
formula, further gives scalar kernel transfer to finite extensions
of finite polyradius Gauss residue fields. A separate exact finite-coefficient
descent reduces the remaining scalar question to residue fields of
finite-dimensional algebraic parameter points, without solving those
general fibers. In remaining degree one, a bounded
continuous psh function provides a faithful curvature measure with
exactly the coefficient-null ideal of the background. The associated
energy form is regular, strongly local and coercive. Its Poisson
problem has unique energy solutions; its capacity controls relative
BT outer capacity in one direction, and complete pole sets give an
orthogonal form decomposition. In every remaining degree, complete
pole carriers yield qualified capacity additivity, while finite tester
pasting and a constrained variational retraction extend the relative
theory. A smooth nonpolar example shows that a bounded classical
subsolution does not suffice for solvability in the global bounded-row
psh completion; an explicit non-psh energy solution isolates the
obstruction. General potential quotients, arbitrary-background raw
capacity regularity, and unrestricted completed-psh Dirichlet
existence remain open or false in the precise scopes recorded here.
\end{abstract}

\section*{Scope and reading guide}
\addcontentsline{toc}{section}{Scope and reading guide}
The current comparison across all eleven Bedford--Taylor directions
is Section~\ref{sec:r4-status}. The exact original-literature theorem
comparison remains Section~\ref{sec:comparison}. Part I retains the
R1--R3 mathematical sources verbatim, with their original section
numbering. Historical open statements are superseded only by the
qualified later theorem explicitly identified in Part II.
The early projection and analytic-graph gates recorded within the R4
proof-development sections are resolved, in their stated scopes, by
Sections~\ref{sec:r4-disc-projection-kernel} and
\ref{sec:r4-same-field-analytic-kernel}. Section~\ref{sec:r4-gauss-extension-kernel}
then treats the stated Gauss extensions; earlier narrower scope
paragraphs are retained as part of the dependency record.

For the new proofs, start with
Section~\ref{sec:r4-same-field-analytic-kernel} for the same-field
analytic kernel theorem, Section~\ref{sec:r4-faithful-energy} for the
faithful reference measure and linear energy theory,
Section~\ref{sec:r4-energy-capacity-bridge} for the one-way capacity
comparison, and Section~\ref{sec:r4-status} for their limitations.
The smooth nonpolar obstruction and its radial energy solution have
separate full proofs in Part II. Original compact smooth CLD tests
and the original absolute value are retained throughout.

R and B are authorized internal inputs in their actual hypotheses;
they have not been subjected to a new comprehensive proof audit.
A2's same-model geometric adelic nefness theorem is already complete
under its stated global assumptions. An arbitrary positive separately
closed background is allowed in the fixed-background results. The
faithful Hilbert energy construction has remaining degree one; it
applies in higher degree only after fixing the other mixed slots.
Every boundary or global defining-row hypothesis is part of its
individual theorem.

The provenance labels \texttt{INTERNAL\_INPUT},
\texttt{EXTERNAL\_KNOWN}, \texttt{DIRECT\_COROLLARY}, and
\texttt{NEW\_PROVED} are distinct from the logical statuses
\texttt{PROVED}, \texttt{CONDITIONAL}, \texttt{OPEN}, and
\texttt{FALSE}. A failed argument is labeled
\texttt{FAILED\_ATTEMPT}. Here \texttt{NEW\_PROVED} means a complete
argument supplied within this project relative to its declared inputs;
it does not assert worldwide priority, external refereeing, or formal
verification. Independent internal readings and exact source versions
are recorded in the accompanying archive.

No general pullback of a current, wedge of two arbitrary positive
currents, automatic extension of kernel identities beyond the
specified finite and Gauss field extensions, or equality on all GK $\delta$-tests is asserted.
Raw BT capacity, its open outer regularization, and energy capacity
are kept distinct. Energy representatives and global bounded psh
defining rows are different solution classes.

The archive records elapsed working intervals and the complete
continuation state. Parallel branches are not added to elapsed time;
pure proof time was not separately instrumented. This record makes no
claim that its present continuation lasted eight hours.
\tableofcontents
\clearpage
\part*{Part I. Frozen R1--R3 mathematical dependencies}
\addcontentsline{toc}{part}{Part I. Frozen R1--R3 mathematical dependencies}

% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/inputs.tex
\section{Scope, inputs, and conventions}
\label{sec:inputs}

Fix a finite extension $K/\mathbf Q_p$ with its given absolute value.
Let $X$ be a good, Hausdorff, boundaryless, quasi-smooth $K$-analytic
space of pure dimension $n\geq1$, and let $U\subset X$ be an ordinary open
set. We use only the original Chambert--Loir--Ducros sheaf
$A^{a,b}$, its compactly supported smooth tests, and its original
calibration. We assume $0\leq p\leq n$. A current of bidegree $(p,p)$ acts on
$A_c^{n-p,n-p}$. We write $D=d'd''$.

The class $\Cb(U)$ consists of the actual real-valued, locally bounded,
upper-semicontinuous classical psh functions of \textup{B}, Definition
2.2: pullback to every analytic curve after every complete isometric
extension is subharmonic. No function is changed on a null or polar
set. A norm in this report always means the actual pointwise bound
\[
 \|u\|_{V,\infty}=\sup_{x\in V}|u(x)|,
\]
even where the traditional notation $L^\infty(V)$ is used. There is
no unspecified reference volume measure behind this notation.

\subsection{The accepted internal results}

The user has authorized the following theorems as established inputs.
They have been checked for scope and dependencies, but their proofs
have not been re-audited in this project.

\begin{description}
\item[\textup{R}.] \emph{Continuous CLD-positive potentials and adelic
mixed measures}, 10 November 2026, 19 pages.
Theorem 2.9 gives local uniform smooth positive approximation of finite
continuous original-current-positive potentials on smooth
quasi-projective algebraic analytifications over a $p$-adic completion.
Lemma 3.1 gives finite coefficient Radon control for the original
positive cone and stability under a fixed continuous multiplier.
Lemma 3.2 smooths pure Fubini--Study metrics on their original line.
Proposition A.1 gives regularized maxima with exact inactive branches.
Theorem 2.10 identifies positivity of the prescribed Yuan--Zhang
integrable adelic mixed measure under its stated hypotheses.

\item[\textup{B}.] \emph{Locally bounded classical plurisubharmonic
functions and mixed currents}, 10 November 2026, 32 pages.
Theorems 2.5--2.7 identify $\Cb$ with the local decreasing PL class,
and construct strongly positive mixed currents and weighted joint
decreasing continuity with background $[1]$.
Definition 2.4 and Lemma 7.2 specify actual-space coefficient Radon
measures and Borel multiplication.
Theorem 4.4, Lemma D.2, and Proposition 7.1 provide the common
projective line, frame, ordinary neighborhood, and decreasing pure
Fubini--Study approximants for a finite family of decreasing sequences.
Lemma D.4 supplies ordinary algebraic neighborhoods without changing
$K$. These common-domain inputs are needed in
Lemma~\ref{lem:common-majorants}; separate local regularizations with
shrinking domains would not suffice.

\item[\textup{A2}.] \emph{A non-Archimedean criterion for geometric
adelic nefness}, revised 12 October 2026, 31 pages.
Theorem 1.1 is the completed same-model geometric nefness theorem under
its number-field, integrable adelic line, and full Cartier-boundary
hypotheses. It preserves the open line and its prescribed
identifications. This is an accepted global application, not an open
problem in the present project. It is not used to prove the local
relative BT theorems below.
\end{description}

Any use of the broader analytic setting passes through the local
algebraic neighborhoods already supplied by \textup{B}. A finite
extension of $\mathbf Q_p$ can be realized as a number-field completion;
the use of \textup{R} does not change the prescribed absolute value.
No argument below makes a new base change or renormalizes integration.

\subsection{Three positivity cones}

On a finite original moment presentation, strongly positive forms are
generated by decomposable squares. Positive forms are represented by
positive semidefinite symmetric coefficient tensors. Weakly positive
forms are nonnegative on decomposable vectors. These cones need not
coincide in middle degree. A \emph{positive current} in this report is
nonnegative on positive complementary original forms, as in
\textup{R}, Convention 2.2. A strongly positive current is nonnegative
on weakly positive complementary forms, as in \textup{B}, Convention
2.3. Thus the arbitrary background considered here is not assumed
strongly positive.

We use the finite-matrix domination from \textup{R}, not a decomposition
of every symmetric test into decomposable squares. The latter would
lose a middle-degree annihilator unless one had strong positivity and
used the separate argument in \textup{B}, Lemma 5.1. Wedge with a
positive $(1,1)$-form preserves the original positive cone. More
generally, wedge with a strongly positive form preserves it; these are
the only cone-preserving wedge operations used in the proofs.

\subsection{Meaning of the result labels}

\begin{center}
\begin{tabularx}{\textwidth}{@{}lX@{}}
\toprule
Label & Meaning in this record\\
\midrule
\texttt{EXTERNAL\_KNOWN} & A verified statement of a cited external source.\\
\texttt{INTERNAL\_INPUT} & A theorem accepted from the supplied manuscripts.\\
\texttt{DIRECT\_COROLLARY} & A consequence obtained directly from named accepted inputs or preceding proved results.\\
\texttt{NEW\_PROVED} & A statement with a complete proof in this record, relative to the adopted inputs; this does not claim bibliographic priority or external peer review.\\
\texttt{CONDITIONAL} & A stated implication whose extra hypothesis is not established here.\\
\texttt{OPEN} & An unresolved assertion in the declared scope.\\
\texttt{FALSE} & An assertion ruled out by a specified counterexample.\\
\texttt{FAILED\_ATTEMPT} & A proof route with an identified invalid or missing step.\\
\bottomrule
\end{tabularx}
\end{center}

The separate analogue-status table uses \texttt{PROVED},
\texttt{CONDITIONAL}, \texttt{OPEN}, and \texttt{FALSE}.
Here \texttt{PROVED} includes the user's accepted internal inputs and
proofs derived from them. It does not mean that the input manuscripts
have been independently validated by this research run.

\subsection{What is not being constructed}

The product has an arbitrary \emph{fixed} positive closed background
and bounded classical psh potentials. It is not a wedge product of two
entirely arbitrary positive currents. It is not the full GK
$\delta$-form algebra, and its current identities do not quantify over
all $\delta$-test forms. Later applications to Green forms or
Beilinson--Bloch heights do not enter the proof of the present calculus.

% END source: base_r3/base_r2/base_r1/sections/inputs.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/relative_cln_agent.tex
% Independent contribution: /root/relative_cln, 10 November 2026.
% Fragment; the master file supplies theorem environments and bibliography.

\section{Ordered products and relative finite-test estimates}
\label{sec:relative-cln-agent}

Throughout this section, $K/\mathbf Q_p$ is finite with its prescribed
absolute value. The space $X$ is good, Hausdorff, boundaryless and
quasi-smooth, of pure dimension $n$. All open sets are ordinary open sets,
and all forms are original CLD forms. We write $D=d'd''$.
The word \emph{positive} means dual positivity for the original positive
form cone used in R, Convention 2.2. It does not mean strong positivity.
The background $S$ is a symmetric positive $(p,p)$-current satisfying
$d'S=d''S=0$.
All $L^\infty$ bounds below mean actual pointwise suprema on the stated
sets; no ambient measure is used to discard exceptional values.

\subsection{The precise order-zero input}

\begin{lemma}[Positive controls on a finite test stage]
\label{lem:finite-positive-controls}
Let $W$ be an ordinary open set, let $L\subset W$ be compact, and fix a
finite original moment presentation of a smooth $(q,q)$-form $\eta$ near
$L$. After replacing $\eta$ by a compactly supported extension agreeing
with it near $L$, there are positive forms
$\Theta_1,\ldots,\Theta_N\in A_c^{q,q}(W)$ and a constant $C_\eta$ such that
every positive $(n-q,n-q)$-current $T$ satisfies
\begin{equation}\label{eq:finite-positive-controls}
 |\mu_{T,\eta}|(L)
 \leq C_\eta\sum_{a=1}^N\langle T,\Theta_a\rangle.
\end{equation}
The same controls work for all $T$. For forms on one fixed finite
presentation, with supports in one fixed compact set, $C_\eta$ is bounded
by a constant times the supremum norm of the coefficient matrices.
\end{lemma}

\begin{proof}
This is the local domination argument of R, Lemma 3.1. We recall the
details that matter for subsequent estimates. A finite smooth partition
of unity reduces the assertion to one moment presentation with real
coordinates $t_1,\ldots,t_r$. Put
\[
 \beta=\sum_{a=1}^r d't_a\wedge d''t_a,
 \qquad \Omega_q=\frac{\beta^q}{q!},
\]
with $\Omega_0=1$. Choose a nonnegative compact smooth cutoff $\rho$,
equal to one near the relevant support. The symmetric coefficient
matrix of $\Omega_q$ is positive definite. Hence, with a constant
$C_\eta$ controlled by the coefficient supremum norm,
\[
 C_\eta\|h\|_\infty\rho\Omega_q\pm h\eta_{\rm sym}\geq0
\]
for every compact smooth scalar $h$. Antisymmetric forms and their
negatives belong to the null part of this positivity convention; a
positive current annihilates them. Therefore
\[
 |\langle T,h\eta\rangle|
 \leq C_\eta\|h\|_\infty\langle T,\rho\Omega_q\rangle.
\]
If $h$ uses further moment coordinates, pull the displayed cone
inequalities to a common refinement. Positivity is preserved, so the
constant does not depend on those additional scalar coordinates.

Original smooth scalars are uniformly dense on compact sets. Riesz
representation thus gives the coefficient Radon measure and the stated
total-variation bound. Summing over the finite partition gives
\eqref{eq:finite-positive-controls}. The local algebraic neighborhoods in
B, Proposition 7.1 and Lemma D.4, let us apply R's argument on the
analytic spaces considered here.
\end{proof}

\paragraph{Status and dependence.}
Lemma~\ref{lem:finite-positive-controls} is
\texttt{DIRECT\_COROLLARY}, from R, Lemma 3.1 and B's local algebraic
transport. In particular it must not be presented as a new positivity
theorem. Strong positivity, and B's decomposable-square annihilator in
middle degree, are not used in this argument.

The scalar coefficient relations hold as identities of Radon measures:
test each relation after multiplication by every original smooth scalar
and use uniqueness of the representing measure. Consequently, for every
locally bounded Borel function $a$,
\[
 \langle aT,\eta\rangle=\int a\,d\mu_{T,\eta}
\]
defines an intrinsic coefficient-Radon current. It commutes with
restriction and with common refinements. The proof is the measure
uniqueness argument of B, Lemma 7.2, applied to the original-positive
coefficient measures supplied by R. It does not replace $a$ by an
almost-everywhere representative chosen using another measure.

\subsection{One multiplication and ordered iteration}

\begin{proposition}[One bounded potential over an arbitrary background]
\label{prop:one-potential-arbitrary-background}
For every $u\in\mathcal C_b(W)$ the current $D(uS)$ is symmetric,
positive, coefficient Radon, and separately $d'$- and $d''$-closed.
If $a_j$ are locally uniformly bounded Borel functions converging
pointwise to $a$, then $a_jS\to aS$ on every original compact test.
In particular,
\[
 D(u_jS)\longrightarrow D(uS)
\]
whenever $u_j\to u$ pointwise and the $u_j$ have a common local bound.
In the last assertion the $u_j$ need not be monotone.
\end{proposition}

\begin{proof}
The convergence assertions are dominated convergence for the fixed
finite coefficient measures of $S$, followed by continuity of
distributional differentiation. For a continuous classical psh
function $v$, the accepted local approximation results of R and B give
smooth psh functions $v_m\to v$ locally uniformly on a fixed local
neighborhood. The identity
\[
 D(v_mS)=Dv_m\wedge S
\]
and preservation of the original positive cone under wedge with a
positive $(1,1)$-form show positivity. Passing to the limit proves
positivity of $D(vS)$. For general $u$, use B, Theorem 2.5 to obtain
continuous semipositive PL functions $u_j\downarrow u$ on a fixed local
neighborhood. They are uniformly bounded on each compact subdomain by
their first term and the locally bounded limit. Dominated convergence
for the fixed coefficient measures of $S$ gives
$D(u_jS)\to D(uS)$. The cone of positive currents is weakly closed.

Closedness follows from $d'D=d''D=0$, and symmetry follows from the
CLD involution. Lemma~\ref{lem:finite-positive-controls} supplies the
coefficient-Radon property. A finite localization proves the assertions
on $W$.
\end{proof}

\begin{corollary}[Canonical ordered products]
\label{cor:ordered-arbitrary-background}
For $0\leq k\leq n-p$, the recursion
\[
 T_{S,0}=S,\qquad
 T_{S,k}(u_1,\ldots,u_k)
 =D\bigl(u_1T_{S,k-1}(u_2,\ldots,u_k)\bigr)
\]
defines a positive, symmetric-as-a-current, coefficient-Radon, separately
closed current for every ordered tuple $u_i\in\mathcal C_b(W)$.
The recursion is intrinsic, multilinear on the cone, positively
homogeneous, unchanged by adding a constant in a differentiated slot,
and compatible with ordinary restriction and gluing.
\end{corollary}

\begin{proof}
Apply Proposition~\ref{prop:one-potential-arbitrary-background}
inductively. Borel multiplication and distributional differentiation
are intrinsic and local. Their linearity proves additivity and positive
homogeneity by induction. A constant in the outer slot contributes
$D(cT)=0$ because the inner current is separately closed; the assertion
for other slots follows inductively. On overlaps, equal coefficient
currents have equal Borel multiples, so the recursion glues.
\end{proof}

Both preceding statements are \texttt{DIRECT\_COROLLARY}. At this
point ``symmetric-as-a-current'' refers only to the CLD involution.
It does \emph{not} assert invariance under permutations of two
discontinuous differentiated potentials. Similarly, intrinsic recursion
does not yet assert convergence along every simultaneous regularization
of all slots. These are separate analytical assertions.

\begin{lemma}[Continuous slots]
\label{lem:continuous-slot-exchange}
Let $u\in\mathcal C_b(W)$ and let $v$ be continuous and classical psh.
Then
\[
 D\bigl(uD(vS)\bigr)=D\bigl(vD(uS)\bigr).
\]
\end{lemma}

\begin{proof}
First take both potentials continuous. Smooth local regularization,
R, Lemma 3.1, and the coefficient mass bounds show that joint locally
uniform convergence of continuous potentials passes through the
recursion. Explicitly, if $Q_j\to Q$ are positive currents and
$a_j\to a$ locally uniformly, with $a$ continuous, then
\[
 a_jQ_j-aQ=(a_j-a)Q_j+a(Q_j-Q)\longrightarrow0.
\]
The first term is controlled by the locally uniform coefficient mass
bound, and the second by R, Lemma 3.1. Smooth commutativity therefore
passes to continuous potentials over the fixed arbitrary background.

Choose continuous psh $u_j\downarrow u$ on one fixed local
neighborhood. The left side of
\[
 D\bigl(u_jD(vS)\bigr)=D\bigl(vD(u_jS)\bigr)
\]
converges by dominated convergence against the \emph{fixed} coefficient
measures of $D(vS)$. On the right, $D(u_jS)\to D(uS)$ by
Proposition~\ref{prop:one-potential-arbitrary-background}; multiplication
by the fixed continuous function $v$ passes to the limit by R,
Lemma 3.1. Differentiate and localize.
\end{proof}

This is again \texttt{DIRECT\_COROLLARY}. No simultaneously varying
discontinuous multiplier is used in its proof.

\subsection{Relative Chern--Levine--Nirenberg bounds}

\begin{theorem}[Finite-test relative CLN estimate]
\label{thm:finite-test-cln}
Fix $K_0\Subset V\Subset W$, an integer $k\leq n-p$, and a finite
original moment presentation of a $(q,q)$-test $\eta$ near $K_0$, where
$q=n-p-k$. There are positive compact tests
\[
 \theta_1,\ldots,\theta_N\in A_c^{n-p,n-p}(V)
\]
and a constant $C$, determined by these fixed data, such that for every
positive separately closed background $S$ and every tuple
$u_1,\ldots,u_k\in\mathcal C_b(W)$,
\begin{equation}\label{eq:relative-cln-product}
 \bigl|\mu_{T_{S,k}(u_1,\ldots,u_k),\eta}\bigr|(K_0)
 \leq C\prod_{i=1}^k\|u_i\|_{L^\infty(V)}
       \sum_{\ell=1}^N\langle S,\theta_\ell\rangle.
\end{equation}
One may replace each norm by $\operatorname{osc}_V(u_i)$.
The empty product is one when $k=0$.

For a locally bounded Borel weight $a$, the same right side multiplied
by $\|a\|_{L^\infty(V)}$ bounds
$|\mu_{aT_{S,k},\eta}|(K_0)$.
The controls and constants are independent of the potentials and of
every regularization parameter. On a fixed finite test stage, the
constant is bounded by a fixed constant times the coefficient
supremum norm of $\eta$.
\end{theorem}

\begin{proof}
Choose nested compact neighborhoods between $K_0$ and $V$, with enough
room for $k+1$ cutoff choices. Lemma~\ref{lem:finite-positive-controls}
first bounds the left side by finitely many evaluations
\[
 C_0\sum_a\langle T_{S,k},\Theta_a^{(0)}\rangle,
 \qquad \Theta_a^{(0)}\in A_c^{q,q}(V),\quad
 \Theta_a^{(0)}\geq0.
\]
All these controls are chosen before the potentials. For every one of
them, the defining recursion gives
\[
 \langle T_{S,k},\Theta_a^{(0)}\rangle
 =\int u_1\,d\mu_{T_{S,k-1}(u_2,\ldots,u_k),D\Theta_a^{(0)}}.
\]
Apply Lemma~\ref{lem:finite-positive-controls} to the fixed smooth test
$D\Theta_a^{(0)}$, using the next compact neighborhood. We obtain
positive tests $\Theta_{ab}^{(1)}$ of degree $(q+1,q+1)$ and constants
independent of the potentials such that
\[
 \langle T_{S,k},\Theta_a^{(0)}\rangle
 \leq C_a\|u_1\|_{L^\infty(V)}
       \sum_b\langle T_{S,k-1},\Theta_{ab}^{(1)}\rangle.
\]
Repeat this operation $k$ times. At each stage only a finite collection
of fixed smooth compact tests is differentiated and controlled.
The final forms have degree $(q+k,q+k)=(n-p,n-p)$, are positive,
and are supported in $V$. Absorb the finitely many constants into $C$
and enumerate the final controls as $\theta_\ell$. This proves
\eqref{eq:relative-cln-product}.

Adding a constant to any differentiated potential leaves the ordered
product unchanged. Subtracting $\inf_Vu_i$ therefore replaces the norm
by its oscillation. Local boundedness and $\overline V\subset W$
make these extrema finite. The weighted estimate follows from
$|a\mu|\leq\|a\|_\infty|\mu|$.

For the last assertion, choose the initial cutoff controls using the
fixed stage and use the matrix norm only in its first domination
constant. All later differentiated controls are then fixed. No
derivative of $\eta$ enters a varying constant. This proves the stated
uniformity.
\end{proof}

\paragraph{Status.}
The estimate is a relative extension needed in this project, with a
complete proof above. Under the user's stipulated inputs its logical
status is \texttt{DIRECT\_COROLLARY}: it follows by finite iteration of
R's fixed-test domination and the ordered recursion. It should not be
advertised as a new independent analytical mechanism.

\paragraph{Charts and refinements.}
The auxiliary forms need not come from a single ambient Euclidean ball,
and no strictly psh exhaustion is assumed. Each domination is an
inequality of intrinsic original CLD forms. Pulling it to a common
moment refinement preserves positivity; R's estimate already permits
scalar coefficients using additional coordinates. Different choices
of finite charts can give different numerical constants and controls,
but they bound the same intrinsic coefficient measures. The proof
therefore establishes compatibility of the estimate without asserting
a chart-independent numerical norm.

\paragraph{What the estimate does not prove by itself.}
A uniform total-variation bound does not allow a discontinuous
multiplier to pass through weakly convergent coefficient measures.
Consequently \eqref{eq:relative-cln-product}, by itself, proves neither
exchange of two discontinuous slots nor weighted joint decreasing
continuity. A further comparison, energy, or capacity argument is
required. In particular, one cannot replace the fixed continuous
multiplier in R, Lemma 3.1 by an arbitrary bounded psh function merely
by invoking this bound.


% END source: base_r3/base_r2/base_r1/sections/relative_cln_agent.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/relative_monotone.tex
\section{A common collar and weighted monotone convergence}
\label{sec:monotone}

The results of this section use the internal inputs listed in
Section~\ref{sec:inputs}. In particular, they do not assert an independent
proof of the regularization theorems in \textup{R} and \textup{B}.
All currents and test forms belong to the original CLD calculus.
Positivity means positivity in the cone used in \textup{R}, not strong
positivity. Set $D=d'd''$.

We first record the exact local consequence of the internal approximation
results that is needed below. A smooth psh function in this section has a
positive original $(1,1)$ curvature \emph{form}.

\begin{lemma}[A common domain for smooth majorants]
\label{lem:common-majorants}
Let $x\in U$, and let $u_{i,j}\downarrow u_i$ be finitely many decreasing
families in $\Cb(U)$ with locally bounded limits. After shrinking around
$x$, there is one ordinary neighborhood $W$, independent of $i,j$, with
the following property. Each of the functions $a=u_{i,j}$ and $a=u_i$
has smooth psh majorants $a_\nu$ on $W$ such that
\[
 a_\nu\geq a,\qquad a_\nu(x')\longrightarrow a(x')\quad(x'\in W).
\]
The majorants are bounded uniformly in $\nu$ on each fixed compact subset
after a finite prefix is discarded. If $a\leq M$ on a compact neighborhood
$C\Subset W$, its majorants may be chosen with $a_\nu\leq M+1$ on $C$.
The index discarded may depend on the function $a$.
\end{lemma}

\begin{proof}
Use \textup{B}, Proposition 7.1, to work on one ordinary algebraic
neighborhood, with the original field and calibration. Apply \textup{B},
Lemma D.2, to the entire finite family of sequences, before choosing
approximants. It gives one projective variety, one ample line, one frame
on $W$, and bounded classical metrics whose weights on $W$ are the given
functions plus one fixed constant. By \textup{B}, Theorem 4.4, every such
metric is a decreasing limit of pure Fubini--Study metrics on this same
line. Their weights $p_\nu$ are continuous, satisfy $p_\nu\geq a$, and
decrease pointwise to $a$ after the fixed constant is removed.

Smooth each pure Fubini--Study metric uniformly, using \textup{R},
Lemma 3.2 and Proposition A.1. If the uniform error is at most
$\epsilon_\nu\downarrow0$, add $\epsilon_\nu$ to the smoothed weight.
The resulting smooth psh function $a_\nu$ satisfies
\[
 p_\nu\leq a_\nu\leq p_\nu+2\epsilon_\nu.
\]
This proves pointwise convergence and the majorant property.
On $C$, the increasing open sets $\{p_\nu<M+1/2\}$ cover $C$ whenever
$a\leq M$. Compactness gives a common index on $C$. Discard this prefix
and require $2\epsilon_\nu<1/2$. The same argument with any bound for
$a$ gives local uniform boundedness of the row. No monotonicity of the
\emph{smoothed} row is asserted or needed.
\end{proof}

\begin{lemma}[A smooth psh collar]
\label{lem:smooth-collar}
Under the assumptions of Lemma~\ref{lem:common-majorants}, there are
compact polynomial norm boxes
\[
 x\in\operatorname{Int}C_1,\qquad
 C_1\subset\operatorname{Int}C_{\mathrm{mid}},\qquad
 C_{\mathrm{mid}}\subset\operatorname{Int}C_2\Subset W,
\]
a smooth psh function $b$ on a neighborhood of $C_2$, and a nonnegative
original smooth cutoff $\chi$ compactly supported in
$\Omega=\operatorname{Int}C_2$, with the following properties.
After adding one constant $c$ to all the original functions, put
\[
 v_{i,j}=\max\{u_{i,j}+c,b\},\qquad
 v_i=\max\{u_i+c,b\}.
\]
Then:
\begin{enumerate}
\item $v_{i,j},v_i\in\Cb(\Omega)$ and $v_{i,j}\downarrow v_i$;
\item $v_{i,j}=u_{i,j}+c$ and $v_i=u_i+c$ on a neighborhood of $C_1$;
\item $\chi=1$ on a neighborhood of $C_{\mathrm{mid}}$;
\item all $v_{i,j},v_i$ equal $b$ on one fixed neighborhood of
      $\operatorname{supp}d'\chi\cup\operatorname{supp}d''\chi$;
\item each patched function $v$ has smooth psh majorants $v_\nu\geq v$
      converging pointwise to $v$, uniformly bounded on $\operatorname{supp}\chi$,
      and equal to $b$ on that same collar.
\end{enumerate}
The majorant rows in (5) need not be decreasing.
\end{lemma}

\begin{proof}
Choose the boxes by the construction in \textup{B}, Lemma D.2. In
particular, lower-bound polynomials have no zero near $C_2$. Choose
thresholds strictly between the defining thresholds for $C_1$ and
$C_{\mathrm{mid}}$. The finite maximum
\[
 b_0=\max\{0,\log|P_a^{N_a}/c_a|,
                     \log|d_l/Q_l^{M_l}|\}
\]
is zero on a neighborhood of $C_1$ and at least some $\delta>0$ on a
neighborhood in $C_2$ of $C_2\setminus\operatorname{Int}C_{\mathrm{mid}}$.
All its branches other than the zero branch are logarithms of analytic
units where they are active. A branch with a zero is strictly inactive
near that zero.

Apply the regularized maximum of \textup{R}, Proposition A.1, with
smoothing radius smaller than all the fixed gaps. The result $\widetilde b$
is an original smooth psh function: zero branches at polynomial zeros
are deleted by the exact inactivity property. On a neighborhood of
$C_1$, only the constant zero branch is active, so $\widetilde b=0$.
On the outer collar, $\widetilde b\geq\delta$.

The endpoint bounds give a constant $c$ and a finite $M$ such that
\[
 1\leq u_i+c\leq u_{i,j}+c\leq M
 \quad\hbox{on a neighborhood of }C_2.
\]
Multiply $\widetilde b$ by a positive constant so that the resulting $b$
is greater than $M+3$ on the outer collar. Choose $\chi$, with
$0\leq\chi\leq1$, equal to one near $C_{\mathrm{mid}}$ and compactly
supported in $\Omega$. Smooth cutoff existence and a finite localization
are part of the original CLD test calculus. Its derivative support lies
where the fixed outer-collar inequality holds.

Classical psh functions are stable under maxima and addition of constants.
This proves (1)--(4). To prove (5), apply Lemma~\ref{lem:common-majorants}
to the \emph{unpatched} function $a=u_{i,j}+c$ or $u_i+c$. Discard a
prefix so that its smooth psh majorants $a_\nu$ are at most $M+1$ on
$C_2$. Define
\[
 v_\nu=\operatorname{rmax}_{\epsilon_\nu}(a_\nu,b),
 \qquad \epsilon_\nu\downarrow0,
\]
with all radii less than the fixed collar gap. The regularized maximum
majorizes the ordinary maximum and has error at most $\epsilon_\nu$.
It is exactly $b$ on the collar because the other branch is inactive
there. These facts give all the assertions in (5).
\end{proof}

\begin{remark}
Neither lemma supplies a strictly psh exhaustion of an arbitrary analytic
domain. The collar is constructed on a chosen small norm box. The closed
positive form used to detect a test below is chosen separately and may
be degenerate in directions not occurring in that test.
\end{remark}

\begin{lemma}[Green exchange on a common collar]
\label{lem:collar-green}
Let $Q$ be a symmetric positive $d'$- and $d''$-closed current of
bidegree $(n-1,n-1)$ on $\Omega$. Let $\chi\in A_c^{0,0}(\Omega)$.
Suppose that $a$ is smooth, $g$ is locally bounded Borel, and both equal
one smooth function $b$ on a neighborhood of
$\operatorname{supp}d'\chi\cup\operatorname{supp}d''\chi$.
Whenever $D(gQ)$ has coefficient Radon type, one has
\[
 \langle aD(gQ),\chi\rangle
 =\langle gD(aQ),\chi\rangle.
\]
In this statement positivity of $a$ and $g$ is not required.
\end{lemma}

\begin{proof}
Since $Q$ is closed and $a$ is smooth, $D(aQ)=Da\wedge Q$. The
definition of the distributional derivative gives
\begin{align*}
 \langle aD(gQ)-gD(aQ),\chi\rangle
  &=\langle gQ,D(\chi a)-\chi Da\rangle\\
  &=\langle Q,g\{aD\chi+d'\chi\wedge d''a
                         +d'a\wedge d''\chi\}\rangle.
\end{align*}
Every term in braces is supported in the collar. Substituting $a=g=b$
there makes the last expression
\[
 \langle bQ,D(\chi b)-\chi Db\rangle=0.
\]
The final equality follows by applying the same distributional identity
to the smooth pair $(b,b)$. The substitutions are valid identities of
coefficient measures. No derivative of the Borel function $g$ is used.
\end{proof}

\begin{lemma}[Monotonicity of the ordered trace integral]
\label{lem:trace-monotonicity}
Let $S$ have bidegree $(p,p)$, let $0\leq k\leq n-p$, and let
$\alpha$ be a smooth, positive, $d'$- and $d''$-closed form of bidegree
$(n-p-k,n-p-k)$ such that wedging it with a positive current preserves
positivity. Let $\chi\geq0$ be compactly supported in $\Omega$.

For $0\leq i\leq k$, suppose $f_i,g_i\in\Cb(\Omega)$,
$f_i\leq g_i$, and all these functions equal the same smooth psh
function $b$ near the derivative support of $\chi$. Assume that each
$g_i$ has the smooth majorants with this exact collar supplied by
Lemma~\ref{lem:smooth-collar}. Define, using the ordered products,
\[
 J(f_0;f_1,\ldots,f_k)
   =\langle f_0T_{S,k}(f_1,\ldots,f_k),\chi\alpha\rangle.
\]
Then
\[
 J(f_0;f_1,\ldots,f_k)
 \leq J(g_0;g_1,\ldots,g_k).
\]
Only existence and positivity of the ordered products are used here.
\end{lemma}

\begin{proof}
The case $k=0$ is measure monotonicity. For $k>0$, fix smooth psh
majorants $G_i\geq g_i$ with the common collar. First replace the weight
$f_0$ by $G_0$, using positivity of
$T_{S,k}(f_1,\ldots,f_k)\wedge\alpha$. Apply
Lemma~\ref{lem:collar-green} with
\[
 Q=T_{S,k-1}(f_2,\ldots,f_k)\wedge\alpha
\]
to move $G_0$ into a differentiated position and $f_1$ into the weight.
The resulting current $DG_0\wedge Q$ is positive and closed. Replace
the weight $f_1$ by $G_1$, and repeat. At the end one obtains
\[
 J(f_0;f_1,\ldots,f_k)
 \leq\langle G_k\,DG_0\wedge\cdots\wedge DG_{k-1}\wedge S,
                     \chi\alpha\rangle.
\]
A final application of the same Green identity, together with
commutation of smooth even-degree factors, makes the right-hand side
$J(G_0;G_1,\ldots,G_k)$.

Now let the index of $G_k$ tend to infinity, keeping every other index
fixed. Bounded convergence against the fixed coefficient measures of
$S$ implies $D(G_kS)\to D(g_kS)$. The outer factors and the weight are
smooth and fixed, so this convergence can be tested directly. Next let
the index of $G_{k-1}$ tend to infinity. The background
$D(g_kS)$ is now fixed, so the same argument applies. Continue in the
order $k,k-1,\ldots,1$. Finally, with the ordered product fixed, let the
weight $G_0$ tend to $g_0$ by bounded convergence. The result is exactly
$J(g_0;g_1,\ldots,g_k)$ in its original order. At no step are two
discontinuous limits interchanged.
\end{proof}

\begin{lemma}[An upper bound for decreasing weights]
\label{lem:upper-weight}
Let $P_j,P$ be positive coefficient Radon currents of one bidegree,
with $P_j\to P$ on all original compact smooth tests. Let $a_j\downarrow a$
be locally uniformly bounded functions. Suppose that, near the support
of a fixed positive compact test $\Theta$, every fixed $a_l$ has bounded
continuous majorants converging pointwise to $a_l$.
Then
\[
 \limsup_{j\to\infty}\langle a_jP_j,\Theta\rangle
       \leq\langle aP,\Theta\rangle.
\]
\end{lemma}

\begin{proof}
For $j\geq l$ and a continuous majorant $a_{l,\nu}\geq a_l$,
positivity gives
\[
 \langle a_jP_j,\Theta\rangle
       \leq\langle a_{l,\nu}P_j,\Theta\rangle.
\]
Keep $l,\nu$ fixed and pass to the limit in $j$, using \textup{R},
Lemma 3.1. Bounded convergence against the fixed positive Radon measure
$\mu_{P,\Theta}$ first allows $\nu\to\infty$ and then $l\to\infty$.
\end{proof}

\begin{theorem}[Weighted decreasing continuity for a fixed arbitrary background]
\label{thm:relative-monotone}
Assume the internal inputs \textup{R} and \textup{B}. Let $S$ be any
symmetric positive original CLD current of bidegree $(p,p)$ on $U$,
separately $d'$- and $d''$-closed. Let $0\leq k\leq n-p$.
If $u_{i,j}\downarrow u_i$ in $\Cb(U)$ for $0\leq i\leq k$, with
locally bounded limits, then
\[
 u_{0,j}T_{S,k}(u_{1,j},\ldots,u_{k,j})
 \longrightarrow
 u_0T_{S,k}(u_1,\ldots,u_k)
\]
on every original compact smooth complementary test form. The ordered
unweighted products have the corresponding joint decreasing continuity.
The background $S$ is fixed throughout.
\end{theorem}

\begin{proof}
Write $W_k$ for the weighted assertion with $k$ differentiated slots.
Bounded convergence for the fixed coefficient measures of $S$ proves
$W_0$. Distributional differentiation shows that $W_{k-1}$ implies
joint unweighted convergence in degree $k$. We prove $W_k$ assuming
this unweighted convergence.

The statement is local on compact test supports. Use a finite original
smooth partition to reduce to one test $\eta$ supported in a small
norm box $C_1$ as in Lemma~\ref{lem:smooth-collar}. Include in the chosen
ordinary neighborhood all the unit coordinates needed to present
$\eta$. Let these coordinates be $t_1,\ldots,t_N$, and set
\[
 \beta=\sum_{a=1}^N d't_a\wedge d''t_a,
 \qquad r=n-p-k,\qquad \alpha=\frac{\beta^r}{r!}.
\]
For $r=0$ set $\alpha=1$. The form $\alpha$ is closed and strongly
positive. In the fixed finite coefficient space it is a positive
definite trace form. After replacing $\eta$ by its symmetric part,
there is $C>0$ such that
\[
 \Theta_\pm=C\chi\alpha\pm\eta
\]
are positive original compact tests. This is precisely the matrix
domination used in \textup{R}, Lemma 3.1; additional scalar coordinates
are handled by pullback to a common refinement. No definiteness in
all possible moment directions is required.

Patch all the potentials to $v_{i,j},v_i$ by
Lemma~\ref{lem:smooth-collar}, and put
\[
 P_j=T_{S,k}(v_{1,j},\ldots,v_{k,j}),\qquad
 P=T_{S,k}(v_1,\ldots,v_k).
\]
The induction hypothesis gives $P_j\to P$. Applying
Lemma~\ref{lem:trace-monotonicity} to the ordered tuples
$(v_0,\ldots,v_k)\leq(v_{0,j},\ldots,v_{k,j})$ yields
\[
 \langle v_0P,\chi\alpha\rangle
       \leq\langle v_{0,j}P_j,\chi\alpha\rangle.
\]
Lemma~\ref{lem:upper-weight} gives the reverse limsup inequality.
Consequently
\begin{equation}
 \label{eq:trace-limit}
 \langle v_{0,j}P_j,\chi\alpha\rangle
       \longrightarrow\langle v_0P,\chi\alpha\rangle.
\end{equation}
Apply Lemma~\ref{lem:upper-weight} to $\Theta_+$ and $\Theta_-$.
Together with \eqref{eq:trace-limit}, the two resulting inequalities
give respectively the upper and lower bounds for
$\langle v_{0,j}P_j,\eta\rangle$. Thus this pairing converges to
$\langle v_0P,\eta\rangle$.

On a neighborhood of $\operatorname{supp}\eta$, the patched potentials
are $u_{i,j}+c$ and $u_i+c$. Ordinary locality of the recursion identifies
the differentiated products there with the original products, since
constants in differentiated slots vanish. Remove the constant $c$ from
the weight by the unweighted degree-$k$ convergence already established.
Finite summation over the test partition proves $W_k$.

This induction proves all $W_k$ and all allowed unweighted assertions.
It uses neither capacity nor energy convergence, nor any interchange
of two discontinuous approximation parameters.
\end{proof}

\begin{corollary}[Symmetry and the first-stage calculus]
\label{cor:full-calculus}
Under the hypotheses of Theorem~\ref{thm:relative-monotone}, the currents
$T_{S,k}$ are symmetric in the differentiated arguments,
positive, symmetric as currents, and separately $d'$- and $d''$-closed.
They are additive and positively homogeneous in each potential, commute
with ordinary restriction and gluing, and are independent of the
regularization used to compute them. In particular,
\[
 D\bigl(uD(vS)\bigr)=D\bigl(vD(uS)\bigr)
 \qquad(u,v\in\Cb(U),\ p\leq n-2).
\]
\end{corollary}

\begin{proof}
All assertions except permutation symmetry were proved for the ordered
recursion. Choose, locally on one fixed domain, continuous semipositive
PL approximants decreasing to each potential, as in \textup{B}.
Continuous arbitrary-background products are symmetric, by uniform
smooth approximation and fixed-continuous-multiplier convergence.
Theorem~\ref{thm:relative-monotone}, with a constant weight, passes this
identity to the given potentials. The recursion identifies the result
independently of the approximants and on overlaps.
\end{proof}

\begin{corollary}[Top-degree convergence as measures]
\label{cor:top-measures}
If $p+k=n$, the weighted convergence in
Theorem~\ref{thm:relative-monotone} holds against every compactly
supported continuous scalar function, locally as signed Radon measures.
\end{corollary}

\begin{proof}
Unweighted convergence against one nonnegative smooth cutoff equal to
one near a prescribed compact set bounds the masses of the positive
measures there. The weights have a common actual pointwise bound on a
slightly larger compact neighborhood. Thus the weighted variations are
uniformly bounded. Uniform density of original smooth scalar functions,
with a fixed cutoff, extends convergence to continuous scalar tests.
\end{proof}

\begin{remark}[What independence means here]
The construction is intrinsic to the \emph{given potentials} and to $S$.
Independence of regularization does not assert that $D(uS)$ depends only
on the ambient current $D[u]$ for every possible pair of potentials.
That stronger quotient-by-potentials statement is distinct from the
claims proved here and from the special Bedford--Taylor-current
condition in CLD 2025, Definition 8.6.2. Equality on all original tests
also does not assert equality on all $\delta$-tests.
\end{remark}

% END source: base_r3/base_r2/base_r1/sections/relative_monotone.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/functional_weights.tex
\section{Continuous functions of several monotone weights}
\label{sec:functional-weights}

This section records a \texttt{DIRECT\_COROLLARY} of
Theorem~\ref{thm:relative-monotone}. It is useful when a localization
factor is a continuous function of several psh potentials, although that
factor need not itself be psh.

\begin{lemma}[Increasing convex compositions]
\label{lem:convex-compositions}
Let $h\in\Cb(V)$, and suppose that its values lie in a compact interval
$I$. If $\varphi:I\to\mathbf R$ is continuous, convex, and
nondecreasing, then $\varphi\circ h\in\Cb(V)$. In particular, if
$h\geq0$, every positive integral power of $h$ is classical psh.
If $h_j\downarrow h$ have values in $I$, then
$\varphi(h_j)\downarrow\varphi(h)$.
\end{lemma}

\begin{proof}
Interpolate $\varphi$ linearly on successively refined dyadic partitions
of $I$. The resulting functions $\varphi_m$ are convex and
nondecreasing, are finite maxima of affine functions with nonnegative
slopes, and decrease uniformly to $\varphi$. Each
$\varphi_m(h)$ is therefore a finite maximum of nonnegative multiples
of $h$ plus constants. The defining classical class is stable under
these operations. Its decreasing-limit property, with the local
boundedness of the limit, gives $\varphi(h)\in\Cb(V)$.
The last assertion follows from continuity and monotonicity of
$\varphi$.
\end{proof}

\begin{proposition}[Functional weighted convergence]
\label{prop:functional-weights}
Let $S$ be a fixed background as in
Theorem~\ref{thm:relative-monotone}, and let $0\leq k\leq n-p$.
Suppose that
\[
 u_{i,j}\downarrow u_i\quad(1\leq i\leq k),
 \qquad h_{a,j}\downarrow h_a\quad(1\leq a\leq s)
\]
are finite families in $\Cb(U)$ with locally bounded limits. Set
\[
 P_j=T_{S,k}(u_{1,j},\ldots,u_{k,j}),\qquad
 P=T_{S,k}(u_1,\ldots,u_k).
\]
For every continuous function $F:\mathbf R^s\to\mathbf R$,
\[
 F(h_{1,j},\ldots,h_{s,j})P_j
 \longrightarrow F(h_1,\ldots,h_s)P
\]
on every original compact smooth complementary test. The factors
$F(h_{1,j},\ldots,h_{s,j})$ are the actual Borel functions, and no
psh assertion about those factors is required.
\end{proposition}

\begin{proof}
Fix a compact test $\eta$ and a relatively compact neighborhood $V$ of
its support. The endpoint bounds provide one compact interval for all
values of each $h_{a,j}$ and $h_a$ on $V$. Add fixed constants so that
the shifted functions $x_{a,j}=h_{a,j}+c_a$ and $x_a=h_a+c_a$ are
nonnegative there. Absorb these shifts into $F$.

We first prove convergence for a polynomial in the $x_a$. Every
monomial of total degree $d\geq1$ can be regarded as a product
$z_1\cdots z_d$ in which variables may repeat. The polarization
identity
\[
 z_1\cdots z_d
 =\frac1{d!}\sum_{A\subset\{1,\ldots,d\}}
       (-1)^{d-|A|}\left(\sum_{a\in A}z_a\right)^d
\]
expresses it as a linear combination of powers of sums with
nonnegative coefficients. Each nonzero sum of shifted psh functions
is nonnegative and classical psh. By
Lemma~\ref{lem:convex-compositions}, its $d$th power is also classical
psh and decreases along the given sequences. The weighted theorem
therefore gives convergence for every term in the displayed identity.
Constants are covered by its constant-weight case. Finite linearity
proves the assertion for every polynomial.

The unweighted convergence $P_j\to P$, together with \textup{R},
Lemma 3.1, supplies a finite bound
\[
 \sup_j|\mu_{P_j,\eta}|(\operatorname{supp}\eta)
       +|\mu_{P,\eta}|(\operatorname{supp}\eta)<\infty.
\]
Approximate $F$ uniformly by polynomials on the common compact
rectangle of shifted values. An approximation error of at most
$\epsilon$ changes each pairing by at most $\epsilon$ times this
fixed variation bound. First pass to the limit in $j$ for the
polynomial, and then let $\epsilon\downarrow0$. This proves the
assertion on $\eta$.
\end{proof}

\begin{remark}
The proposition permits, for example, bounded continuous cutoffs of
$u_j-v_j$ and products of finitely many bounded psh weights.
It does not permit replacing $F$ by a discontinuous function without
an additional argument. In applications to strict maximum sets, one
first uses continuous scalar cutoffs and only afterwards takes a
bounded pointwise limit against the already fixed coefficient measures.
\end{remark}


% END source: base_r3/base_r2/base_r1/sections/functional_weights.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/scope_stability.tex
\section{A sufficient condition for varying backgrounds}
\label{sec:dominated-backgrounds}

The main theorem fixes the background. The following separate result
gives a sufficient additional hypothesis under which a background may
vary. It is an elementary \texttt{DIRECT\_COROLLARY} of the
coefficient-Radon domination argument and the ordered construction;
it does not assert unrestricted varying-background stability.

\begin{lemma}[Order domination and Borel multipliers]
\label{lem:order-dominated-borel}
Let $P_j,P,R$ be original positive currents of one bidegree on $U$.
Suppose that $P_j\to P$ on all original compact smooth tests and that
\[
 0\leq P_j\leq C R
\]
for one constant $C<\infty$. Let $a_j$ be Borel functions converging
pointwise to $a$, with one common bound on each compact subset.
Then
\[
 a_jP_j\longrightarrow aP
\]
on every original compact smooth test. Closedness of these currents
is not required in this lemma.
\end{lemma}

\begin{proof}
Weak closure of the positive cone gives $0\leq P\leq CR$.
Fix a compact test $\eta$. The finite positive control construction
in R, Lemma 3.1 supplies a finite positive Radon measure $\nu_\eta$,
depending only on $R$, $C$ and the test, such that
\begin{equation}\label{eq:fixed-measure-domination}
 |\mu_{P_j,\eta}|\leq\nu_\eta,
 \qquad |\mu_{P,\eta}|\leq\nu_\eta.
\end{equation}
Here these are inequalities of measures, not only bounds on their
total masses. Indeed, on each moment piece choose a positive compact
control $\Theta$ with $A\Theta\pm\eta\geq0$. For every nonnegative
smooth scalar $h$,
\[
 |\langle P_j,h\eta\rangle|
 \leq A\langle P_j,h\Theta\rangle
 \leq AC\langle R,h\Theta\rangle.
\]
Radon uniqueness and scalar density give domination by
$AC\mu_{R,\Theta}$. Sum over the finite pieces to obtain
\eqref{eq:fixed-measure-domination}.

For a fixed bounded Borel function $b$, approximate $b$ in
$L^1(\nu_\eta)$ by continuous functions on a compact neighborhood
of the test support. Such approximation is available for a finite
Radon measure; cutoffs keep the functions on that neighborhood.
R, Lemma 3.1 gives convergence with each fixed continuous multiplier,
and \eqref{eq:fixed-measure-domination} bounds the approximation error
uniformly in $j$. It follows that
\[
 \int b\,d\mu_{P_j,\eta}\longrightarrow
 \int b\,d\mu_{P,\eta}.
\]
Finally,
\[
 \left|\int(a_j-a)\,d\mu_{P_j,\eta}\right|
 \leq\int|a_j-a|\,d\nu_\eta\longrightarrow0
\]
by bounded convergence for the fixed measure $\nu_\eta$. Apply the
fixed-multiplier assertion to $b=a$ and combine the two limits.
\end{proof}

\begin{proposition}[Order-dominated varying backgrounds]
\label{prop:dominated-background-stability}
Let $R_0,S_j,S$ be symmetric positive separately closed original
$(p,p)$-currents on $U$. Assume
\[
 S_j\longrightarrow S,
 \qquad 0\leq S_j\leq C R_0
\]
on $U$, with one finite constant $C$. Fix
$u_1,\ldots,u_k\in\Cb(U)$, where $0\leq k\leq n-p$.
Then
\[
 T_{S_j,k}(u_1,\ldots,u_k)
 \longrightarrow T_{S,k}(u_1,\ldots,u_k).
\]
Moreover, for every locally uniformly bounded sequence of Borel
functions $a_j\to a$ pointwise,
\[
 a_jT_{S_j,k}(u_1,\ldots,u_k)
 \longrightarrow aT_{S,k}(u_1,\ldots,u_k).
\]
All convergence is on the full original compact smooth test sheaf.
The differentiated potential tuple is fixed.
\end{proposition}

\begin{proof}
For a fixed potential tuple the ordered construction is linear in
the background. Positivity therefore gives
\begin{equation}\label{eq:product-order-domination}
 0\leq T_{S_j,r}(u_{k-r+1},\ldots,u_k)
 \leq C T_{R_0,r}(u_{k-r+1},\ldots,u_k)
\end{equation}
for every $0\leq r\leq k$: the difference is the ordered product
with the positive closed background $CR_0-S_j$.

Induct on $r$. The case $r=0$ is the assumed weak convergence.
Assume convergence at degree $r-1$. Its currents are dominated by
the one fixed positive current on the right of
\eqref{eq:product-order-domination}. Apply
Lemma~\ref{lem:order-dominated-borel} with the fixed multiplier
$u_{k-r+1}$, and then apply $D$. This gives convergence at degree
$r$ by the recursion. At $r=k$, apply the same lemma once more,
now with the varying Borel weights $a_j$, to obtain the final
assertion.
\end{proof}

\begin{remark}
The hypothesis $S_j\leq CR_0$ supplies domination by fixed coefficient
measures. It is stronger than a uniform bound on local masses.
The conclusion permits discontinuous Borel weights precisely because
this additional domination is available. No varying differentiated
tuple, and no assertion for arbitrary weakly convergent backgrounds,
is included.
\end{remark}

\subsection{Continuous changes of potential}

\begin{lemma}[Continuous harmonic multipliers]
\label{lem:continuous-harmonic-multipliers}
Assume the local regularization theorem of \textup{R}. Let $S$ be an
allowed positive separately closed original current of bidegree
$(p,p)$, with $p\leq n-1$. If $h$ is finite continuous and
$D[h]=0$ as an ambient original current, then
\[
 D(hS)=0.
\]
Consequently, if $u,v$ are finite continuous and $D[u]=D[v]$, then
$D(uS)=D(vS)$. This conclusion concerns continuous potentials.
\end{lemma}

\begin{proof}
Work on the ordinary local algebraic neighborhoods supplied by
\textup{B}, Proposition 7.1. Both $h$ and $-h$ have positive ambient
curvature currents, since those currents vanish. By \textup{R},
Theorem 2.9, each is locally uniformly approximable by smooth
functions with positive curvature forms. For a smooth approximant
$h_m$, the current
\[
 D(h_mS)=Dh_m\wedge S
\]
is positive. Locally uniform convergence passes through multiplication
by the fixed coefficient-Radon current $S$ and then through $D$.
Hence $D(hS)$ is positive. Apply the same argument to $-h$: the
current $-D(hS)$ is positive as well. The original positive cone of
currents is pointed, by the finite-test domination in R, Lemma 3.1,
so $D(hS)=0$. Take $h=u-v$ for the final assertion and localize.
No classical-psh assertion about $h$ is needed.
\end{proof}

This is a \texttt{DIRECT\_COROLLARY} of \textup{R} and the fixed-
background one-potential construction. Applying it to the positive
closed current formed by the remaining slots also proves potential
independence for a continuous slot in a mixed product. It does not
establish the same statement for an arbitrary discontinuous
difference $u-v$ with $D[u-v]=0$.


% END source: base_r3/base_r2/base_r1/sections/scope_stability.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/potential_independence_scope.tex
\section{A limited quotient-by-potentials statement}
\label{sec:potential-independence-scope}

The main theorem constructs products of specified, actual pointwise
potentials. The assertion that $D(uS)$ depends only on the ambient
current $D[u]$ is stronger. The following cases do follow from the
available inputs. They are \texttt{DIRECT\_COROLLARY} statements, not
an assertion of this stronger property for every background.

\begin{lemma}[Ambient mixed backgrounds]
\label{lem:potential-independence-ambient-mixed}
Let $u,v\in\Cb(U)$ satisfy $D[u]=D[v]$ on all original compact smooth
tests. Let $0\leq p\leq n-1$, and suppose locally that
\[
 S=\sum_{\ell=1}^{N}c_\ell
      T_{[1],p}(a_{\ell,1},\ldots,a_{\ell,p}),
 \qquad a_{\ell,i}\in\Cb(U),\quad c_\ell\in\mathbf R.
\]
Then
\[
 D(uS)=D(vS).
\]
The conclusion also holds for a locally finite sum of this kind.
In particular, it applies when such a current is an allowed positive
closed background.
\end{lemma}

\begin{proof}
It suffices to consider one summand. By the symmetry already present
in \textup{B}, Theorem 2.6,
\begin{align*}
 D\bigl(uT_{[1],p}(a_1,\ldots,a_p)\bigr)
  &=T_{[1],p+1}(a_1,\ldots,a_p,u)\\
  &=D\bigl(a_1D(a_2\cdots D(a_pD[u])\cdots)\bigr).
\end{align*}
Replace the innermost current $D[u]$ by $D[v]$. At every subsequent
step, equality of coefficient-Radon currents implies equality after
multiplication by the same actual Borel function, by \textup{B},
Lemma 7.2. Distributional differentiation preserves that equality.
Reversing the preceding symmetry gives $D(vS)$. Finite linearity and
localization prove the stated extensions. For $p=0$, the assertion is
the original equality $D[u]=D[v]$.
\end{proof}

\begin{lemma}[Smooth closed backgrounds]
\label{lem:potential-independence-smooth}
Under the same hypothesis on $u,v$, let $S=[\alpha]$, where $\alpha$
is an original smooth $(p,p)$-form satisfying
$d'\alpha=d''\alpha=0$. Then $D(uS)=D(vS)$.
\end{lemma}

\begin{proof}
Set $h=u-v$. The given local boundedness makes all products below
well defined. Borel multiplication of the integration current agrees
with smooth-form multiplication:
\[
 h[\alpha]=\alpha\wedge[h].
\]
The distributional Leibniz rule and the two closedness assumptions give
\[
 D(hS)=D(\alpha\wedge[h])=\alpha\wedge D[h]=0.
\]
This is an equality on the entire original compact smooth test sheaf.
\end{proof}

\begin{proposition}[Passage through a dominated weak closure]
\label{prop:potential-independence-dominated-closure}
Fix $u,v\in\Cb(U)$ with $D[u]=D[v]$. Let $S_j,S,R_0$ be allowed
positive separately closed backgrounds of degree $(p,p)$, with
\[
 S_j\longrightarrow S,\qquad 0\leq S_j\leq C R_0
\]
for one finite constant $C$. Suppose that $D(uS_j)=D(vS_j)$ for every
$j$, for example by either preceding lemma. Then
\[
 D(uS)=D(vS).
\]
\end{proposition}

\begin{proof}
The function $h=u-v$ is an actual locally bounded Borel function.
Lemma~\ref{lem:order-dominated-borel}, with the fixed multiplier $h$,
gives $hS_j\to hS$ on all original compact smooth tests.
Apply $D$ to this convergence. Every $D(hS_j)$ is zero by hypothesis,
so $D(hS)=0$.
\end{proof}

\begin{remark}[The remaining scope question]
These arguments give neither a representation of an arbitrary
original-positive closed current by ambient bounded-potential products,
nor a dominated approximation of every such current by the backgrounds
above. They also do not show that a locally bounded difference
$h=u-v$ satisfying $D[h]=0$ is a continuous function at its actual
point values. This original-current kernel regularity question remains
open in this investigation. The previously proved continuous-harmonic
multiplier lemma applies if that additional continuity is known.
\end{remark}


% END source: base_r3/base_r2/base_r1/sections/potential_independence_scope.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/relative_energy_agent.tex
\section{Relative energy and convergence in capacity}
\label{sec:relative-energy-capacity}

Throughout this section, the fixed-background calculus of
Theorem~\ref{thm:relative-monotone} is available. Its proof uses no energy
or capacity assertion from this section. All measures below are measures
on the actual Berkovich space, and all multipliers use their given point
values. We continue to use the original positive cone of \textup{R}.

\subsection{A relative energy form}

\begin{lemma}[Regularization on a prescribed smaller norm box]
\label{lem:compact-box-regularization}
Work in one ordinary algebraic neighborhood supplied by \textup{B}.
Let $C_{\mathrm{in}}\Subset\operatorname{Int}C_{\mathrm{out}}$ be
compact polynomial norm boxes, and let finitely many bounded classical
psh functions be defined on a neighborhood of $C_{\mathrm{out}}$.
They admit smooth decreasing psh approximants on one neighborhood of
the entire $C_{\mathrm{in}}$. Thus the smaller domain need not be
chosen separately at each point of a prescribed compact test support.
\end{lemma}

\begin{proof}
This is the compact-box form of the construction in \textup{B},
Lemma D.2, followed by its Theorem 4.4. We spell out the distinction
from applying local regularization separately at every point.
Insert a middle norm box and construct the finite polynomial maximum
$b$ used in that lemma: it is zero on $C_{\mathrm{in}}$ and is at
least a fixed $\delta>0$ on the outer collar. Homogenizing its branches
and adding inactive global generating sections realizes $b$ as the
weight of one pure-Fubini--Study metric $q$, in a frame nonzero near
$C_{\mathrm{out}}$. After adding one common constant, all given
functions lie in $[1,M]$ there. Choose $m\delta>M+1$ and give each
extended metric the local weight $\max\{a+c,mb\}$, setting it equal
to $mq$ outside the middle box. These prescriptions agree on their
overlap. They define bounded classical metrics on one projective
variety and one ample line, with weights exactly $a+c$ on a
neighborhood of $C_{\mathrm{in}}$.

Theorem 4.4 of \textup{B} gives decreasing global pure-Fubini--Study
approximants for each metric. Their smooth global approximations from
\textup{R}, Lemma 3.2, restrict to the same frame near the whole
$C_{\mathrm{in}}$. If the uniform smoothing error is at most
$\epsilon_m=4^{-m}$, adding $2\epsilon_m$ gives a decreasing smooth
row: its values lie between $p_m+\epsilon_m$ and
$p_m+3\epsilon_m$, while $p_{m+1}\leq p_m$.
These particular smooth approximants are classical psh. Indeed, after
any complete field extension and pullback to an analytic curve, their
active branches are logarithms of norms of analytic units, hence
harmonic functions. The regularized maximum is convex and
nondecreasing in each variable, so its composition with these branches
is subharmonic, by the curve chain rule in \textup{B}, Lemma B.1.
A section vanishing identically or at a point gives
an inactive branch nearby, which is deleted. Positive scaling and addition of
constants preserve this property. Thus the rows used below belong to
$\Cb$; no general identification of original-form positivity with
classical plurisubharmonicity is being assumed. The same convex,
coordinatewise nondecreasing composition argument applies to the
additional regularized maxima with the smooth collar below.
Remove the common constant $c$ to finish. No extension of a background
current is used in this argument.

\end{proof}

Let $Q$ be a symmetric positive, separately $d'$- and $d''$-closed
current of bidegree $(n-1,n-1)$ on an ordinary open $\Omega$. Let
$\mathcal D_b(\Omega)$ be the real vector space generated locally by
$\Cb(\Omega)$. If $f,g\in\mathcal D_b(\Omega)$, define
\begin{equation}
\label{eq:relative-gamma}
 \Gamma_Q(f,g)
   =\frac12\bigl\{D(fgQ)-fD(gQ)-gD(fQ)\bigr\},
 \qquad \Gamma_Q(f)=\Gamma_Q(f,f).
\end{equation}
Here $D(fQ)$ is a signed coefficient Radon current, by linearity of
the ordered one-potential construction. The first term in
\eqref{eq:relative-gamma} is initially a distribution. The following
lemma proves that the full expression is a measure.

\begin{lemma}[Relative energy measures]
\label{lem:relative-energy-measures}
For $f,g\in\mathcal D_b(\Omega)$, $\Gamma_Q(f,g)$ is a signed Radon
measure and $\Gamma_Q(f)$ is nonnegative. For every nonnegative compact
smooth $\chi$, one has
\begin{equation}
\label{eq:relative-energy-CS}
 \left|\int\chi\,d\Gamma_Q(f,g)\right|^2
 \leq\left(\int\chi\,d\Gamma_Q(f)\right)
       \left(\int\chi\,d\Gamma_Q(g)\right).
\end{equation}
Suppose, for the following integration formula, that $f=a-b$, where
$a,b\in\Cb$ are functions on one common norm-box domain and agree
outside a compact $L$ contained in a strictly smaller norm box.
Suppose also that the psh summands of $g$ are defined on that larger
domain. This includes the common-collar differences used below.
Then
\begin{equation}
\label{eq:relative-green-energy}
 \int f\,D(gQ)=\int g\,D(fQ)
       =-\int\Gamma_Q(f,g),
 \qquad
 \int\Gamma_Q(f)=-\int f\,D(fQ).
\end{equation}
The integrals in \eqref{eq:relative-green-energy} are supported on a
fixed compact neighborhood of $\operatorname{supp}f$.
\end{lemma}

\begin{proof}
First suppose that all functions are smooth. Then
\[
 \Gamma_Q(f,g)
  =\frac12(d'f\wedge d''g+d'g\wedge d''f)\wedge Q.
\]
Symmetry of $Q$ makes the pairing symmetric. Since
$d'f\wedge d''f$ is a strongly positive $(1,1)$-form, the measure
$\Gamma_Q(f)$ is positive. Positivity of
$\Gamma_Q(f+t g)$ for every real $t$ proves
\eqref{eq:relative-energy-CS} by the discriminant inequality.
Stokes proves \eqref{eq:relative-green-energy}.

We explain carefully the approximation used to pass these identities
to bounded potentials. On a common smaller domain, let $p_m\downarrow a$
be the continuous pure-Fubini--Study weights supplied by \textup{B} for
a given $a\in\Cb$. Smooth $p_m$ uniformly with error at most
$\epsilon_m=4^{-m}$, and call the result $s_m$. Then
\[
 a_m=s_m+2\epsilon_m,
 \qquad
 p_m+\epsilon_m\leq a_m\leq p_m+3\epsilon_m,
 \qquad a_{m+1}<a_m.
\]
Thus $a_m$ are smooth psh functions decreasing pointwise to $a$.
The same construction is made for finitely many psh summands.

If an exact common collar is required, apply a regularized maximum
with the fixed smooth collar function from
Lemma~\ref{lem:smooth-collar}. Its smoothing radii may decrease to zero.
For completeness, the regularized maximum is nondecreasing in the
smoothing radius: for fixed input, its expression as the integral of
a maximum of affine functions of the radius is convex, and symmetry
of the mollifier makes it even. Hence it is nondecreasing for
nonnegative radius. The patched smooth rows remain decreasing and
equal the prescribed collar function exactly.

Write $f=\sum c_i a_i$ and $g=\sum d_l b_l$, with classical psh
$a_i,b_l$, and use such simultaneous smooth rows. Products of bounded
functions converge pointwise and are uniformly bounded on each compact
set. Therefore $D(f_mg_mQ)\to D(fgQ)$ by bounded convergence against
the fixed coefficient measures of $Q$. Each term in $f_mD(g_mQ)$
is a constant multiple of $a_{i,m}D(b_{l,m}Q)$, so it converges by
Theorem~\ref{thm:relative-monotone} with one differentiated slot.
The same holds with $f$ and $g$ exchanged. Consequently
\[
 \Gamma_Q(f_m,g_m)\longrightarrow\Gamma_Q(f,g)
\]
as distributions. A positive distribution in top degree is a positive
Radon measure. Apply this to $f+t g$, and polarize, to obtain the
measure assertion and \eqref{eq:relative-energy-CS}.

It remains to justify compact-support integration by parts without
assuming differentiability of $f$. Write $f=a-b$ as in the statement.
Choose several nested norm boxes between $L$ and the boundary of the
given domain. First apply Lemma~\ref{lem:compact-box-regularization}
with a smaller box still containing all the boxes to be used below.
This gives smooth decreasing rows for $a,b$ and all psh summands of
$g$ on a neighborhood of that entire box. This step is essential:
applying a pointwise regularization theorem separately would not
produce a domain containing the full support of $f$.

Inside this common regularization domain, choose a second smooth psh
collar which is zero near $L$ and dominates the bounded shifted values
of $a$ and $b$ on an outer collar. Patch their smooth rows by a
regularized maximum with this collar, with radii decreasing to zero.
The limiting patched difference is still exactly $f$: on $L$ the
collar is inactive, and outside $L$ the two limiting functions agree.
The smooth differences $f_m$ have support in one fixed compact set
because their two rows have the same exact outer collar. The smooth
rows of the psh summands of $g$ are already defined near this compact
set. No approximation or global extension of the fixed $Q$ is involved.

For these smooth rows, insert one compact smooth cutoff equal to one
on the common support of $f_m$. Stokes gives the two Green equalities.
Every term is a finite sum of weighted one-potential products, so
Theorem~\ref{thm:relative-monotone} passes it to the limit. The
energy terms converge by the already proved distributional formula
and positivity, with the same cutoff. All resulting multipliers are the
Borel multipliers in \eqref{eq:relative-gamma}; no exceptional-set
representative is chosen. This local integration formula is the only
compact-support version used below.
\end{proof}

\begin{remark}[A noncircular estimate before joint continuity]
\label{rem:energy-before-continuity}
There is a useful estimate that precedes the preceding lemma. Suppose
$a_j\downarrow a$ are \emph{smooth} psh functions, uniformly bounded
near a compact set, and all equal one smooth function outside a fixed
compact $L\Subset\Omega$. Choose $\chi\geq0$, compactly supported,
with $\chi=1$ near $L$, and put
\[
 E_\chi(a)=\int\chi\,d'a\wedge d''a\wedge Q.
\]
For $j<l$, $g=a_j-a_l\geq0$ has compact support in $L$, and smooth
integration by parts gives
\begin{equation}
\label{eq:smooth-energy-cauchy}
 E_\chi(a_l)-E_\chi(a_j)
   =E_\chi(g)+2\int g\,D(a_jQ)\geq E_\chi(g).
\end{equation}
The identity
\[
 E_\chi(a_j)=\frac12\langle Q,a_j^2D\chi\rangle
                  -\langle D(a_jQ),\chi a_j\rangle
\]
and the relative CLN estimate bound these energies uniformly. They
therefore increase to a finite limit, and the gradients are Cauchy
in the local Hilbert seminorm induced by $Q$. In fact the smooth
energy measures are Cauchy in total variation:
\[
 \|\chi\Gamma_Q(a_j)-\chi\Gamma_Q(a_l)\|_{\mathrm{TV}}
 \leq E_\chi(a_j-a_l)^{1/2}
       \bigl(E_\chi(a_j)^{1/2}+E_\chi(a_l)^{1/2}\bigr).
\]
To see this, polarize the difference and use the measure-valued
Cauchy--Schwarz inequality. That inequality follows by writing the
positive $2\times2$ matrix of energy measures relative to the sum of
its two diagonal measures, applying the pointwise discriminant
inequality to its densities, and then ordinary Cauchy--Schwarz.
This statement uses only the one-potential construction and relative
CLN. We do not use
it to prove Theorem~\ref{thm:relative-monotone}; that theorem has the
independent trace-squeeze proof.
\end{remark}

\subsection{A localized relative capacity}

Fix a positive closed background $S$ of bidegree $(p,p)$, and put
$q=n-p$. Work on a relatively compact ordinary moment domain $\Omega$
on which a smooth closed strongly positive $(1,1)$-form $\beta$ is
defined. A typical choice is the trace form in the proof of
Theorem~\ref{thm:relative-monotone}. No strict positivity of $\beta$
in all analytic directions is assumed. For $1\leq k\leq q$ and
a Borel subset $E\subset\Omega$, define
\begin{equation}
\label{eq:relative-capacity}
 \operatorname{Cap}_{S,\beta,k}(E;\Omega)
  =\sup_{\substack{v\in\Cb(\Omega)\\0\leq v\leq1}}
    \int_ET_{S,k}(v,\ldots,v)\wedge\beta^{q-k}.
\end{equation}
For $k=q$, $\beta$ drops out, and we write
$\operatorname{Cap}_S(E;\Omega)$. In particular, $S=[1]$ gives the
usual candidate non-Archimedean Bedford--Taylor capacity in top degree.
We make no nondegeneracy assertion for an arbitrary $S$ or $\Omega$.

\begin{proposition}[Elementary properties and mixed domination]
\label{prop:capacity-elementary}
The capacity in \eqref{eq:relative-capacity} is monotone and countably
subadditive on Borel sets, finite on compact subsets of $\Omega$, and
continuous from below on increasing Borel sets. Restriction to a
smaller domain can only increase it. If
$a_1,\ldots,a_k\in\Cb(\Omega)$ satisfy $0\leq a_i\leq M_i$, then
\begin{equation}
\label{eq:mixed-capacity-domination}
 \int_ET_{S,k}(a_1,\ldots,a_k)\wedge\beta^{q-k}
 \leq\frac{k^k}{k!}\left(\prod_{i=1}^kM_i\right)
       \operatorname{Cap}_{S,\beta,k}(E;\Omega).
\end{equation}
If some $M_i=0$, the left side vanishes. Constants may first be
subtracted from the $a_i$, so bounds on their oscillations suffice.
\end{proposition}

\begin{proof}
The first assertions follow from the same properties of positive
measures and the supremum in \eqref{eq:relative-capacity}; for an
increasing sequence, the two suprema over the sequence index and the
admissible functions commute. Compact finiteness is the uniform
relative CLN bound. Domain monotonicity follows by restricting every
admissible function on the larger domain.

If all $M_i>0$, set $v_i=a_i/M_i$ and
$v=k^{-1}\sum_i v_i$. Then $0\leq v\leq1$. Symmetry, multilinearity,
and positivity show that the expansion of
$T_{S,k}(v,\ldots,v)$ contains
$k!k^{-k}T_{S,k}(v_1,\ldots,v_k)$, with all other terms
positive. This gives \eqref{eq:mixed-capacity-domination}.
\end{proof}

\begin{theorem}[A quantitative relative capacity estimate]
\label{thm:relative-capacity-estimate}
For this theorem, take $\Omega$ to be the ordinary interior of one
polynomial norm box in an algebraic neighborhood, and assume that
the compact set $L$ below is contained in a strictly smaller nested
norm box. Let $u_j,u\in\Cb(\Omega)$ satisfy $u_j\downarrow u$.
Suppose that there are such a compact $L\Subset\Omega$ and one smooth psh function $h$
such that $u_j=u=h$ on $\Omega\setminus L$, and that the functions
have a common bound on a neighborhood of $L$. Set $g_j=u_j-u\geq0$.
For $1\leq k\leq q$, there is a finite constant $C$, independent of
$j$, $v$, and $\delta>0$, such that
\begin{equation}
\label{eq:capacity-rate}
 \operatorname{Cap}_{S,\beta,k}(\{g_j>\delta\};\Omega)
 \leq\frac{C}{\delta}
   \left(\int g_j\,
      T_{S,k}(u,\ldots,u)\wedge\beta^{q-k}\right)^{2^{-k}}.
\end{equation}
The constant depends on the compact collar, $\beta$, a bound for $u$
on a slightly larger compact set, and finitely many positive cutoff
evaluations of $S$. Consequently $u_j\to u$ in this relative capacity.
\end{theorem}

\begin{proof}
Choose a nonnegative compact smooth $\chi$ equal to one near $L$.
Fix an admissible $v$, and abbreviate products using the symmetric
calculus already proved. For $0\leq r\leq k$, set
\[
 I_r=\int g_j\,(Dv)^r(Du)^{k-r}\wedge S\wedge\beta^{q-k}.
\]
These numbers are finite and nonnegative. For $r\geq1$, put
\[
 Q_r=(Dv)^{r-1}(Du)^{k-r}\wedge S\wedge\beta^{q-k}.
\]
This is an arbitrary allowed positive closed current of bidegree
$(n-1,n-1)$; the energy lemma therefore applies to it. Since $g_j$
has compact support in $L$, Green's identity and Cauchy--Schwarz give
\begin{align}
 I_r
  &= -\int\Gamma_{Q_r}(g_j,v),\notag\\
 I_r^2
  &\leq\left(\int\Gamma_{Q_r}(g_j)\right)
          \left(\int\chi\,\Gamma_{Q_r}(v)\right). 
 \label{eq:capacity-energy-step}
\end{align}
The first factor satisfies the one-sided estimate
\begin{align}
 0\leq\int\Gamma_{Q_r}(g_j)
  &= -\int g_jD(g_jQ_r)\notag\\
  &=\int g_jD(uQ_r)-\int g_jD(u_jQ_r)
    \leq I_{r-1}.
 \label{eq:capacity-one-sided}
\end{align}
The final inequality uses $g_j\geq0$ and positivity of
$D(u_jQ_r)$. Thus there is no need to estimate a signed curvature
by exchanging two limits.

For the second factor, the energy identity gives
\[
 \int\chi\,\Gamma_{Q_r}(v)
   =\frac12\langle Q_r,v^2D\chi\rangle
                         -\langle D(vQ_r),\chi v\rangle.
\]
It is nonnegative. Since $0\leq v\leq1$, it is bounded above by
the absolute coefficient estimate for the first term. Apply the
relative CLN bound to $Q_r$, whose differentiated arguments consist
of $r-1$ copies of $v$ and $k-r$ copies of the fixed bounded $u$.
The smooth $\beta$ factors are fixed. Hence
\[
 0\leq\int\chi\,\Gamma_{Q_r}(v)\leq A_r,
\]
where $A_r$ is finite and independent of $v$ and $j$. Combining this
with \eqref{eq:capacity-energy-step}--\eqref{eq:capacity-one-sided}
and iterating yields
\[
 I_k\leq
 A_k^{1/2}A_{k-1}^{1/4}\cdots A_1^{1/2^k}I_0^{1/2^k}.
\]
On $\{g_j>\delta\}$, the inequality
$1\leq g_j/\delta$ gives \eqref{eq:capacity-rate} after taking the
supremum over $v$.

Finally, $I_0$ is the integral of $g_j\downarrow0$ against the
\emph{fixed} locally finite positive measure
$T_{S,k}(u,\ldots,u)\wedge\beta^{q-k}$. The functions
$g_j$ have a common bound and fixed compact support. Bounded
convergence gives $I_0\to0$. This last step does not use a claim about
varying backgrounds.
\end{proof}

\begin{corollary}[Local convergence in capacity and quasicontinuity]
\label{cor:relative-quasicontinuity}
On a sufficiently small ordinary norm-box neighborhood, let
$V_0\Subset\Omega$ be a smaller core and let $u_j\downarrow u$ in
$\Cb$ on a neighborhood of $\overline\Omega$. Then, for every
$\delta>0$,
\[
 \operatorname{Cap}_{S,\beta,k}
   (V_0\cap\{u_j-u>\delta\};\Omega)\longrightarrow0.
\]
Every $u\in\Cb$ is locally quasicontinuous for this capacity: for
every $\epsilon>0$, there is an ordinary open $G\subset\Omega$
with $\operatorname{Cap}_{S,\beta,k}(G;\Omega)<\epsilon$ such that
$u|_{V_0\setminus G}$ is continuous in the subspace topology.
For $k=q$, this statement uses the intrinsic relative capacity
$\operatorname{Cap}_S$ and needs no trace form.
\end{corollary}

\begin{proof}
Patch the entire decreasing family by the common collar of
Lemma~\ref{lem:smooth-collar}. The patched functions agree with the
original ones plus a fixed constant on a neighborhood of $V_0$,
and their differences have support in one fixed compact set. Apply
Theorem~\ref{thm:relative-capacity-estimate} to them.

For quasicontinuity, take continuous psh approximants decreasing to
the patched $u$, with the same collar. Choose indices $j_l$ such that
\[
 \operatorname{Cap}_{S,\beta,k}
       (\{u_{j_l}-u>2^{-l}\};\Omega)
       <\epsilon 2^{-l-1}.
\]
Each exceptional set is ordinary open: $u_{j_l}$ is continuous and
$u$ is upper semicontinuous, so $u_{j_l}-u$ is lower semicontinuous.
Their union $G$ is open and has capacity less than $\epsilon$ by
countable subadditivity. On $\Omega\setminus G$, the chosen
subsequence converges uniformly to the patched $u$. Its restriction
to $V_0$ is the original $u$ up to the fixed constant.
\end{proof}

\begin{remark}[Scope of the capacity conclusions]
The preceding results concern the Borel capacity defined by the
supremum of the constructed measures. They do not establish outer
regularity, continuity from above on compact sets, Choquet
capacitability, relative extremal functions, or an identification of
capacity-zero sets with ambient pluripolar sets. Degeneracy is allowed.
In particular, an arbitrary $S$ can be supported on an ambient
pluripolar set, and the relative measures can charge that set.

The relative energy lemma is a direct consequence of the first-stage
calculus. The quantitative estimate \eqref{eq:capacity-rate} and its
local quasicontinuity consequence provide an additional general
relative theorem. Their proof depends on the already proved
fixed-background symmetry, so they are not used retrospectively in
the proof of that symmetry.
\end{remark}

% END source: base_r3/base_r2/base_r1/sections/relative_energy_agent.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/capacity_stability.tex
\section{Stability under local convergence in capacity}
\label{sec:capacity-stability}

This section is a \texttt{DIRECT\_COROLLARY} of the relative calculus,
mixed domination, and local quasicontinuity proved above. In particular,
it is not used in establishing any of those assertions. The background
$S$ remains fixed. Put $q=n-p$ and assume $q\geq1$.

\begin{lemma}[Top capacity controls each finite coefficient stage]
\label{lem:top-capacity-coefficients}
Let $\eta$ be an original compact smooth test of bidegree
$(q-r,q-r)$, $0\leq r\leq q$, presented by unit moment coordinates
$t_1,\ldots,t_N$ on an ordinary domain $\Omega$. Suppose these
coordinates are bounded on $\Omega$. Let
$a_1,\ldots,a_r\in\Cb(\Omega)$ have actual bounds $|a_i|\leq M$.
There is a constant $C_\eta$, independent of these potentials, such
that, for every Borel set $E\subset\operatorname{supp}\eta$,
\[
 |\mu_{T_{S,r}(a_1,\ldots,a_r),\eta}|(E)
       \leq C_\eta\operatorname{Cap}_S(E;\Omega).
\]
The constant may depend on $M$, the test, and the bounded moment
coordinates. For $r=0$ the tuple is empty.
\end{lemma}

\begin{proof}
Set $\rho=\frac12\sum_{a=1}^Nt_a^2$ and
$\beta=D\rho=\sum_a d't_a\wedge d''t_a$. The function $\rho$ is
bounded, smooth and classical psh: after every complete field extension
and curve pullback, each unit logarithm is harmonic, and this convex
function of their values is subharmonic. Finite coefficient-matrix
domination, as in Lemma~\ref{lem:finite-positive-controls}, gives an
inequality of measures on the test support,
\[
 |\mu_{T_{S,r}(a_1,\ldots,a_r),\eta}|
 \leq C\,T_{S,r}(a_1,\ldots,a_r)\wedge\beta^{q-r}.
\]
This is a measure inequality, not merely a total-mass bound: multiply
the positive test inequalities by every nonnegative smooth scalar and
use Radon uniqueness. Smooth factors commute with the recursion, so
the measure on the right is
\[
 T_{S,q}(a_1,\ldots,a_r,
             \underbrace{\rho,\ldots,\rho}_{q-r}).
\]
Shift the $a_i$ by $M$, and apply the mixed domination inequality
\eqref{eq:mixed-capacity-domination} in degree $q$. The bounds $2M$
and $\sup_\Omega\rho$ give the required constant. A zero bound
contributes a constant differentiated slot, whose product vanishes.
No positivity of $D\rho$ outside the test's moment directions is
needed.
\end{proof}

\begin{lemma}[A capacity-controlled multiplier limit]
\label{lem:capacity-multiplier}
Let $P_j\to P$ be positive currents of one fixed bidegree on $\Omega$,
with convergence on all original compact smooth complementary tests.
Fix such a compact test $\eta$ supported in $L\Subset\Omega$.
Assume that the coefficient measures for $P_j$ and $P$ satisfy
\[
 |\mu_{P_j,\eta}|(E),\ |\mu_{P,\eta}|(E)
       \leq C_\eta\operatorname{Cap}_S(E;\Omega)
       \qquad(E\subset L\text{ Borel}).
\]
Let $a_j,a$ be real-valued Borel functions, uniformly bounded near $L$, with
\[
 \operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};\Omega)
       \longrightarrow0\qquad(\delta>0).
\]
Assume that $a$ is continuous outside ordinary open sets of arbitrarily
small $\operatorname{Cap}_S(\,\cdot\,;\Omega)$ on a compact
neighborhood of $L$. Then
\[
 \langle a_jP_j,\eta\rangle\longrightarrow\langle aP,\eta\rangle.
\]
\end{lemma}

\begin{proof}
Weak convergence of the positive currents bounds their coefficient
variations on $L$ by one number $A$. If $|a_j|,|a|\leq M$, then
\[
 \left|\int(a_j-a)\,d\mu_{P_j,\eta}\right|
 \leq \delta A+2MC_\eta
       \operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};\Omega).
\]
First let $j\to\infty$, and then let $\delta\downarrow0$.

To handle the fixed multiplier $a$, choose an ordinary open exceptional
set $G$ of capacity less than $\epsilon$. On a compact neighborhood
$H$ of $L$, the restriction of $a$ to $H\setminus G$ is continuous.
The space $H$ is compact Hausdorff, hence normal. Tietze extension and
a cutoff give a continuous function $a_\epsilon$ near $L$, bounded
by $M$, which equals $a$ on $L\setminus G$. Thus
\[
 \left|\int(a-a_\epsilon)\,d\mu_{P_j,\eta}\right|,
 \left|\int(a-a_\epsilon)\,d\mu_{P,\eta}\right|
 \leq 2MC_\eta\epsilon.
\]
For fixed $a_\epsilon$, R, Lemma 3.1 gives convergence. Let
$\epsilon\downarrow0$ and combine the two estimates. Neither outer
regularity of the capacity nor a choice of a different representative
for $a$ is used.
\end{proof}

\begin{definition}[Local convergence in the relative capacity]
\label{def:local-capacity-convergence}
We say that real-valued Borel functions $a_j$ on $U$ converge to $a$ locally in
$S$-capacity if, for every relatively compact polynomial norm-box
domain $\Omega$ whose closure is contained in an ordinary algebraic
neighborhood $W\subset U$, every compact $L\Subset\Omega$, and
every $\delta>0$,
\[
 \operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};\Omega)
       \longrightarrow0.
\]
Within any prescribed finite unit-moment presentation one may first
choose such a neighborhood $W$, and then choose boxes with closure
contained in $W$. The quantifier covers all these boxes, including
their further shrinks. We do not identify this condition with
convergence for a single larger-domain capacity: such an identification
would require an additional comparison between capacities of domains.
\end{definition}

\begin{theorem}[Weighted stability in local relative capacity]
\label{thm:capacity-stability}
Let $0\leq k\leq q$. For $0\leq i\leq k$, let
$u_{i,j},u_i\in\Cb(U)$ have one actual uniform bound on every
compact subset of $U$, independent of $j$. Suppose
$u_{i,j}\to u_i$ locally in $S$-capacity in the sense of
Definition~\ref{def:local-capacity-convergence}. Then
\[
 u_{0,j}T_{S,k}(u_{1,j},\ldots,u_{k,j})
       \longrightarrow u_0T_{S,k}(u_1,\ldots,u_k)
\]
on every original compact smooth complementary test. No monotonicity
is required. Unweighted products have the same stability.
\end{theorem}

\begin{proof}
Localize a test to finitely many relatively compact norm boxes on
which its unit moment coordinates are defined and bounded. The assumed
local uniform bounds apply on their closures. On each piece,
Lemma~\ref{lem:top-capacity-coefficients} controls the coefficient
measures of every degree-$r$ product, $0\leq r\leq k$, uniformly
in $j$, and also controls its proposed limit.

Each $u_i$ is locally quasicontinuous by
Corollary~\ref{cor:relative-quasicontinuity}. If that corollary is
applied on still smaller boxes, domain monotonicity gives
$\operatorname{Cap}_S(G;\Omega)\leq
\operatorname{Cap}_S(G;\Omega')$ for $G\subset\Omega'\subset\Omega$.
A finite cover and countable subadditivity therefore give the ordinary
open exceptional set required near the test support, with arbitrarily
small capacity measured in $\Omega$. Choose a compact neighborhood of
the test support inside the union of the smaller cores; continuity
glues on its complement of that exceptional set. This supplies the
compact-neighborhood hypothesis of Lemma~\ref{lem:capacity-multiplier}.

Now induct on the number of differentiated slots. In degree zero the
background is fixed, and Lemma~\ref{lem:capacity-multiplier} applies
with $P_j=P=S$. At each later degree, weighted convergence at the
preceding degree passes through $D$ and gives unweighted convergence
of the new products. Apply Lemma~\ref{lem:capacity-multiplier} with
weight $u_{0,j}$ to obtain the weighted assertion at this degree.
The capacity controls and quasicontinuity just verified supply all its
hypotheses. Sum the finitely many localized tests.
\end{proof}

\begin{remark}
Theorem~\ref{thm:capacity-stability} strengthens the mode of convergence
available once the monotone theory has been established. It neither
gives compactness of arbitrary bounded psh families in this capacity,
nor proves that every weak limit of their Monge--Amp\`ere measures
comes from a bounded psh function. Those are different assertions.
The result does not change the fixed-background requirement.
\end{remark}

% END source: base_r3/base_r2/base_r1/sections/capacity_stability.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/outer_regularity_attempt.tex
\section{The remaining compact-regularity obstruction}
\label{sec:outer-regularity-attempt}

This section records a precise reduction, not a proof of outer
regularity for the full Bedford--Taylor capacity. The initial capacity estimate concerns decreasing potential sequences.
The later nonmonotone stability theorem assumes convergence in capacity;
it does not extract a convergent subsequence from an arbitrary normalized
family. Compact continuity of the capacity instead permits a different
maximizing potential at every index. This distinction is essential.

\subsection{A measure-family reduction}

\begin{proposition}[Weak closure and compact outer regularity]
\label{prop:capacity-weak-closure}
Let $L$ be a compact metrizable space and let $\mathcal M$ be a
nonempty family of positive Radon measures on $L$ satisfying
\[
 \sup_{\mu\in\mathcal M}\mu(L)<\infty.
\]
For a Borel set $E\subset L$, put
\[
 c(E)=\sup_{\mu\in\mathcal M}\mu(E),
 \qquad
 c^*(K)=\inf_{\substack{G\supset K\\G\text{ open in }L}}c(G)
 \quad(K\subset L\text{ compact}).
\]
Let $\overline{\mathcal M}^{\,w}$ be the weak closure in the space
of finite measures. Then:
\begin{enumerate}
\item $c$ is inner regular on Borel sets:
\[
 c(E)=\sup_{\substack{K\subset E\\K\text{ compact}}}c(K).
\]
\item For every compact $K\subset L$,
\begin{equation}
\label{eq:compact-capacity-closure}
 c^*(K)=\sup_{\mu\in\overline{\mathcal M}^{\,w}}\mu(K).
\end{equation}
\item The following conditions are equivalent:
\begin{enumerate}
\item $c(K_j)\downarrow c(K)$ whenever $K_j\downarrow K$ are compact;
\item $c^*(K)=c(K)$ for every compact $K$;
\item every $\mu\in\overline{\mathcal M}^{\,w}$ satisfies
      $\mu(K)\leq c(K)$ for every compact $K$.
\end{enumerate}
\end{enumerate}
\end{proposition}

\begin{proof}
For the first assertion, Radon inner regularity gives
\[
 \sup_{\mu\in\mathcal M}\mu(E)
  =\sup_{\mu\in\mathcal M}
       \sup_{K\Subset E}\mu(K)
  =\sup_{K\Subset E}\sup_{\mu\in\mathcal M}\mu(K).
\]
This interchanges two suprema, which is legitimate. It does not
interchange a supremum with an infimum over open neighborhoods.

The mass bound makes $\overline{\mathcal M}^{\,w}$ compact. Suppose
$\mu_j\in\mathcal M$ converges weakly to $\mu$. For an open
$G\supset K$, the open-set inequality for weak convergence yields
\[
 \mu(K)\leq\mu(G)\leq\liminf_j\mu_j(G)\leq c(G).
\]
Taking the infimum over $G$ proves that the right side of
\eqref{eq:compact-capacity-closure} is at most the left side.

Choose decreasing open neighborhoods $G_j$ of $K$, with compact
closures $F_j$, such that $F_j\downarrow K$ and the $G_j$ form a
neighborhood basis of $K$. Choose $\mu_j\in\mathcal M$ with
\[
 \mu_j(G_j)\geq c(G_j)-1/j.
\]
After passing to a subsequence, $\mu_j$ converges weakly to some
$\mu\in\overline{\mathcal M}^{\,w}$. For each fixed $l$, all
sufficiently large $j$ satisfy $G_j\subset F_l$, so the closed-set
inequality gives
\[
 \lim_j c(G_j)
 \leq\limsup_j\mu_j(F_l)\leq\mu(F_l).
\]
Let $l\to\infty$. The measure is finite, hence
$\mu(F_l)\downarrow\mu(K)$, while
$\lim_jc(G_j)=c^*(K)$. This proves
\eqref{eq:compact-capacity-closure}.

Condition (a) implies (b) by applying it to compact neighborhoods
$F_j\downarrow K$ and using $c(K)\leq c^*(K)\leq c(F_j)$.
Conversely, assume (b) and let $K_j\downarrow K$. Given $\epsilon>0$,
choose an open $G\supset K$ with $c(G)\leq c(K)+\epsilon$.
The nested compact sets $K_j$ eventually lie in $G$; otherwise their
intersections with the closed complement of $G$ would have a common
point. Thus $\lim_jc(K_j)\leq c(K)+\epsilon$. The reverse inequality
is monotonicity. Finally, equivalence with (c) follows from
\eqref{eq:compact-capacity-closure}.
\end{proof}

\noindent\textbf{Status.} All assertions of the proposition are
\texttt{DIRECT\_COROLLARY} elementary reductions from Radon regularity
and weak compactness of bounded positive measure families. They locate
the missing analytic assertion and are not counted as new BT theorems.

\paragraph{Application to the constructed relative capacity.}
Fix its domain $\Omega$, its background $S$, and its trace form
$\beta$, and write
\[
 \mu_v=T_{S,k}(v,\ldots,v)\wedge\beta^{n-p-k},
 \qquad v\in\Cb(\Omega),\quad0\leq v\leq1.
\]
Choose an ordinary open core $W\Subset\Omega$ lying in one smaller
compact norm box, and an original smooth cutoff $\chi$, with
$0\leq\chi\leq1$, supported inside that box, and
$\chi=1$ on a neighborhood of $\overline W$. The measures
\[
 \mathcal M_\chi=\{\chi\mu_v:0\leq v\leq1\}
\]
have uniformly bounded total mass by relative CLN and can be viewed
on one compact metrizable norm-box neighborhood of
$\operatorname{supp}\chi$. Their capacity on subsets of $W$ is
exactly $\operatorname{Cap}_{S,\beta,k}(\,\cdot\,;\Omega)$.
Proposition~\ref{prop:capacity-weak-closure} therefore identifies the
precise remaining local assertion.

\begin{quote}
\textbf{Weak-closure domination (open).}
If $0\leq v_j\leq1$ are classical psh functions on the
\emph{fixed} domain $\Omega$, and
$\chi\mu_{v_j}\to\nu$ weakly, then, for every compact
$K\subset W$,
\[
 \nu(K)\leq
 \operatorname{Cap}_{S,\beta,k}(K;\Omega).
\]
\end{quote}

This condition is weaker than representing $\nu|_W$ as the
Monge--Amp\`ere measure of one admissible potential. It also retains
the original capacity domain: representing the limit by a bounded
psh function only on a smaller domain would bound it by that
\emph{larger} domain-restricted capacity, which is insufficient.
If weak-closure domination is proved, compact continuity from above
and outer regularity on compact sets follow immediately. Outer
regularity on all Borel sets and a capacitability theorem still need
their own argument; neither is assumed here.

\subsection{Cases where the reduction closes}

\begin{proposition}[Finite families and uniformly compact potential families]
\label{prop:restricted-family-regularity}
\begin{enumerate}
\item If $\mathcal M$ is finite, its capacity $c$ is outer regular
      on every Borel set.
\item If $\mathcal M$ is weakly compact, its capacity is continuous
      from above on compact sets and outer regular on compact sets.
\item In the relative Bedford--Taylor setting, let $\mathcal F$ be
      a compact family, for local uniform convergence, of continuous
      classical psh functions on $\Omega$ with values in $[0,1]$.
      The capacity obtained by restricting the supremum to
      $v\in\mathcal F$ has the two compact-set regularity properties
      on every relatively compact core.
\end{enumerate}
\end{proposition}

\begin{proof}
For (1), write $\mathcal M=\{\mu_1,\ldots,\mu_N\}$. Given a
Borel $E$ and $\epsilon>0$, choose open $G_i\supset E$ with
$\mu_i(G_i)\leq\mu_i(E)+\epsilon$. The finite intersection
$G=\bigcap_iG_i$ is open and satisfies
$c(G)\leq c(E)+\epsilon$. Assertion (2) follows directly from
Proposition~\ref{prop:capacity-weak-closure}.

For (3), uniform stability of the continuous fixed-background
products makes $v\mapsto\chi\mu_v$ continuous from local uniform
convergence to weak convergence of finite measures on the compact
cutoff support. Its image is compact, so (2) applies. These are
restricted-family results. No local-uniform compactness of the
entire normalized classical psh class is asserted.
\end{proof}

\subsection{Why the direct extensions fail}

\paragraph{Radon regularity is not uniform family regularity.}
On $L=[0,1]$, take
\[
 A=\mathbf Q\cap(0,1],\qquad
 \mathcal M=\{\delta_x:x\in A\}.
\]
The set $A$ is dense in $L$, but $0\notin A$. Every measure is Radon
and has mass one. Nevertheless,
\[
 c(\{0\})=0,\qquad c([0,1/m])=1,\qquad c^*(\{0\})=1.
\]
The weak closure contains $\delta_0$, for example as the limit of
$\delta_{1/j}$, exactly accounting for the
defect in \eqref{eq:compact-capacity-closure}. Thus positivity,
Radon regularity, uniform mass bounds, countable subadditivity, and
continuity from below do not by themselves imply compact continuity
from above. This is an abstract obstruction to a proposed proof,
not a counterexample to the full non-Archimedean BT capacity.

\paragraph{Finite maxima do not supply an admissible upper envelope.}
On the ordinary unit disc in $\mathbf A_K^{1,\mathrm{an}}$, the
bounded continuous classical psh functions
\[
 a_j=\max\{-1,j^{-1}\log|T|\}
\]
increase pointwise. Their limit is zero at every point with $T\ne0$
and is $-1$ at the rigid zero. It is not upper semicontinuous.
Thus closure under finite maxima and bounded decreasing limits
does not justify taking the pointwise supremum of a bounded family.
An upper-semicontinuous envelope theorem, with its all-extension
curve condition, remains an additional assertion.

\paragraph{Potential order does not order curvature measures.}
Even if a tail envelope is available, $v_j\leq w_j$ does not imply
$\mu_{v_j}\leq\mu_{w_j}$. On an annulus crossing the unit circle,
a suitable positive rescaling of $\max\{0,\log|T|\}$ lies in
$[0,1]$ and has a nonzero positive Laplacian, whereas the larger
constant function $1$ has zero Laplacian. An envelope route therefore
also needs a contact or extremal-measure comparison; pointwise
domination alone is insufficient.

\paragraph{Status.}
The measure-family reduction and the restricted-family propositions
have complete proofs above. They isolate the weak-closure domination
lemma as the next analytic target. For the full normalized
$\Cb(\Omega)$ family, that lemma, compact continuity from above,
and outer regularity remain \texttt{OPEN}. The failed arguments
recorded here are not used elsewhere in the report.

% END source: base_r3/base_r2/base_r1/sections/outer_regularity_attempt.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/maximum_locality.tex
\section{Maximum locality and relative comparison}
\label{sec:maximum-locality}

This section uses the fixed-background mixed-current theorem proved above. Thus
\(S\) is a fixed symmetric positive original CLD current of bidegree
\((p,p)\), with \(d'S=d''S=0\), and all potentials belong to
\(\mathcal C_b\). Positivity means the cone used in R, not strong positivity.
No varying-background convergence is used.

For a coefficient-Radon current \(T\) and a Borel set \(E\), write
\(\mathbf 1_E T\) for its Borel multiple. In particular,
\[
 \mu_{\mathbf 1_E T,\eta}=\mathbf 1_E\mu_{T,\eta}
\]
for every original compact smooth complementary test \(\eta\). This notation
does not assert that \(E\) is ordinarily open, nor that \(\mathbf 1_E T\) is
closed.

\begin{theorem}[Maximum locality for a fixed background]
\label{thm:maximum-locality-relative}
Let \(1\leq k\leq n-p\), let \(u,v,w_2,\ldots,w_k\in\mathcal C_b(V)\), and
put \(M=\max(u,v)\) and \(E=\{u>v\}\). Then
\[
 \mathbf 1_E T_{S,k}(M,w_2,\ldots,w_k)
 =\mathbf 1_E T_{S,k}(u,w_2,\ldots,w_k).
\tag{ML}
\]
The equality holds as coefficient-Radon currents on the original CLD test
sheaf. The corresponding identity on \(\{v>u\}\) holds with \(v\) in the
first differentiated slot.
\end{theorem}

\noindent\textbf{Status.} \texttt{NEW\_PROVED} consequence of the
fixed-background theorem. Its \([1]\)-background case was already
a \texttt{DIRECT\_COROLLARY} of B; see below.

\begin{proof}
The assertion is local on ordinary open sets. By B, Theorem 2.5, choose on one
ordinary neighborhood simultaneous decreasing continuous semipositive PL
approximants
\[
 u_j\downarrow u,\qquad v_j\downarrow v,\qquad w_{i,j}\downarrow w_i.
\]
Finite intersections of the fixed neighborhoods suffice. Set
\[
 M_j=\max(u_j,v_j),\qquad h_j=M_j-v_j,
 \qquad h=M-v=(u-v)_+.
\]
Both \(M_j\) and \(v_j\) decrease to locally bounded classical psh functions.
The functions \(h_j\) themselves need not be monotone; no monotonicity of
\(h_j\) is asserted or used.

Write
\[
 A_j=T_{S,k}(M_j,w_{2,j},\ldots,w_{k,j}),\qquad
 B_j=T_{S,k}(u_j,w_{2,j},\ldots,w_{k,j}).
\]
The set \(E_j=\{u_j>v_j\}\) is ordinarily open. Ordinary locality gives
\(A_j=B_j\) there, while \(h_j=0\) on its complement. Hence, as
coefficient-Radon currents,
\[
 h_j A_j=h_j B_j.
\tag{1}
\]
Apply weighted joint decreasing continuity to each of
\(M_j A_j,v_j A_j,M_j B_j,v_j B_j\). Subtracting the resulting limits in
(1) gives
\[
 h(A-B)=0,
 \quad A=T_{S,k}(M,w_2,\ldots,w_k),\quad
 B=T_{S,k}(u,w_2,\ldots,w_k).
\tag{2}
\]
This passage uses two admissible psh weights, not a convergence theorem for
an arbitrary changing Borel weight.

For an original compact test \(\eta\), multiply (2) by arbitrary original
smooth scalar functions. Uniqueness of its Radon coefficient measure gives
\(h\mu_{A-B,\eta}=0\). On \(E_m=\{h\geq 1/m\}\), the Borel function
\(1/h\) is bounded. Therefore
\(\mathbf 1_{E_m}\mu_{A-B,\eta}=0\). Since
\(E=\bigcup_{m\geq1}E_m\), it follows that
\(\mathbf 1_E\mu_{A-B,\eta}=0\). This proves (ML) on every original
test, and the local identities glue. Interchanging \(u\) and \(v\) proves
the final assertion.
\end{proof}

\begin{theorem}[Relative comparison with a compact boundary gap]
\label{thm:relative-comparison}
Let \(m=n-p\geq1\). For \(a\in\mathcal C_b(V)\), put
\[
 \mu_a=T_{S,m}(a,\ldots,a).
\]
Let \(u,v\in\mathcal C_b(V)\). Assume that for every \(\varepsilon>0\),
\[
 \overline{\{u<v-\varepsilon\}}\ \text{is compact and contained in }V.
\tag{BG}
\]
The closure in (BG) is the ordinary topological closure of a Borel set. Then
\[
 \int_{\{u<v\}}d\mu_v\ \leq\ \int_{\{u<v\}}d\mu_u.
\tag{CP}
\]
The integrals are allowed to be infinite. If \(V\Subset U\) and
\(\liminf_{x\to\partial V}(u(x)-v(x))\geq0\), condition (BG) holds.
\end{theorem}

\noindent\textbf{Status.} \texttt{NEW\_PROVED} consequence of the
fixed-background theorem. It requires no strictly psh exhaustion and no
capacity theorem.

\begin{proof}
Fix \(\varepsilon>0\) and let \(w=\max(u,v-\varepsilon)\). Form the
positive closed current
\[
 Q=\sum_{j=0}^{m-1}
 T_{S,m-1}(\underbrace{w,\ldots,w}_{j},
                  \underbrace{u,\ldots,u}_{m-1-j}).
\]
For \(m=1\), this means \(Q=S\). Multilinearity, symmetry, and the
recursive definition give the telescoping identity
\[
 \mu_w-\mu_u=D\bigl((w-u)Q\bigr).
\tag{3}
\]
Here \(w-u\) is a bounded Borel function on its compact support; the identity
is obtained by subtracting two well-defined Borel multiples. It does not
require \(w-u\) to be psh.

By (BG), the support of \((w-u)Q\) is contained in a compact subset of
\(V\). Choose an original smooth compact cutoff \(\chi\) equal to one
near that support. Equation (3) and distributional differentiation give
\[
 \langle\mu_w-\mu_u,\chi\rangle
 =\langle(w-u)Q,D\chi\rangle=0.
\tag{4}
\]
The signed measure \(\mu_w-\mu_u\) has compact support, since the two
potentials agree on the ordinary complement of
\(\overline{\{u<v-\varepsilon\}}\). Thus (4) is its total signed mass.
No subtraction of two possibly infinite total masses on \(V\) is made.

Set
\[
 A_\varepsilon=\{u<v-\varepsilon\},\qquad
 B_\varepsilon=\{u=v-\varepsilon\},\qquad
 C_\varepsilon=\{u>v-\varepsilon\}.
\]
The set \(A_\varepsilon\cup B_\varepsilon\) is contained in the compact
set \(\overline{\{u<v-\varepsilon/2\}}\), so every measure below has
finite mass on it. Maximum locality and symmetry give
\[
 \mathbf 1_{A_\varepsilon}\mu_w
   =\mathbf 1_{A_\varepsilon}\mu_v,
 \qquad
 \mathbf 1_{C_\varepsilon}\mu_w
   =\mathbf 1_{C_\varepsilon}\mu_u.
\]
For the first equality, constants in differentiated slots have zero
curvature. Applying (4) and partitioning into these three Borel sets yields
\[
 \mu_v(A_\varepsilon)+\mu_w(B_\varepsilon)
   =\mu_u(A_\varepsilon\cup B_\varepsilon).
\]
Positivity therefore implies
\[
 \mu_v(\{u<v-\varepsilon\})
 \leq\mu_u(\{u\leq v-\varepsilon\}).
\tag{5}
\]
Along any positive sequence \(\varepsilon\downarrow0\), both sets in
(5) increase to \(\{u<v\}\). Continuity from below of positive measures
proves (CP).

For the final assertion, the boundary hypothesis means that every boundary
point has a neighborhood in which \(u-v>-\varepsilon\). Compactness of
\(\partial V\) gives a finite collar of this kind. Its complement in
\(\overline V\) is compact and contained in \(V\), proving (BG).
\end{proof}

\paragraph{What comparison does not prove.}
The theorem does not by itself imply pointwise domination or uniqueness
for an arbitrary background. Section~\ref{sec:capacity-domination}
proves precise conclusions modulo capacity zero. For instance, \(S=0\) makes both Monge--Amp\`ere measures zero
for all potentials. More generally, relative products can only detect
the support and the curvature directions of \(S\). Domination requires
an additional mechanism that detects a strict inequality of potentials;
the complex proof uses strictly psh perturbations for this purpose.

\subsection{The part already implied by B}
\label{subsec:baseline-maximum-locality}

\begin{proposition}[Maximum locality for the previously constructed products]
On every analytic space permitted by B, identity (ML) for B's products
\(T_k=T_{[1],k}\) already follows from B, Theorems 2.5--2.7, without the
fixed-background extension proved in this report. The same is true when
the background is itself one of B's mixed currents.
\end{proposition}

\noindent\textbf{Status.} \texttt{DIRECT\_COROLLARY}.

\begin{proof}
Repeat the proof of Theorem~\ref{thm:maximum-locality-relative} with
\(S=[1]\). Every input of that proof is already supplied by B:
Theorem 2.5 supplies decreasing continuous regularizations on a fixed
ordinary neighborhood; Theorem 2.6 supplies ordinary locality and
coefficient-Radon products; Theorem 2.7 supplies weighted joint decreasing
continuity on all original complementary tests. The last Borel division
argument uses the Borel multiplication from B, Definition 2.4 and
Lemma 5.1. Thus the proof does not use any theorem first established in
this report. If the background is a B mixed current, absorb its fixed
potentials into the differentiated slots and use B's symmetry and weighted
continuity in those additional slots.
\end{proof}

% END source: base_r3/base_r2/base_r1/sections/maximum_locality.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/capacity_domination.tex
\section{Domination and uniqueness modulo capacity zero}
\label{sec:capacity-domination}

There is a precise consequence of comparison that does not require a
nondegeneracy theorem. It separates the proved measure-theoretic
conclusion from the additional condition needed for pointwise order.
The results in this section are \texttt{DIRECT\_COROLLARY} of
Theorem~\ref{thm:relative-comparison} and the positive mixed calculus.

Let $S$ have bidegree $(p,p)$ and put $q=n-p\geq1$. On an ordinary
open $V$, use the top-degree capacity
\[
 \operatorname{Cap}_S(E;V)
   =\sup_{\substack{h\in\Cb(V)\\0\leq h\leq1}}
       \int_E T_{S,q}(h,\ldots,h),
       \qquad E\subset V\text{ Borel}.
\]
This definition makes sense without a moment frame. Relative CLN gives
compact finiteness, and countable subadditivity is unchanged. Write
$\mu_a=T_{S,q}(a,\ldots,a)$. For two potentials $a,b$, the compact-gap
condition means
\[
 \overline{\{a<b-\delta\}}\Subset V
             \qquad\text{for every }\delta>0,
\]
where the closure is the ordinary topological closure.

\begin{proposition}[Domination modulo capacity zero]
\label{prop:capacity-domination}
Let $u,v\in\Cb(V)$ satisfy the compact-gap condition in the order
$(u,v)$. If $\mu_u(\{u<v\})=0$, then
\[
 \operatorname{Cap}_S(\{u<v\};V)=0.
\]
This does not assert that the strict inequality set is empty.
\end{proposition}

\begin{proof}
Fix $\delta>0$, $0<\epsilon<\delta$, and any admissible
$h\in\Cb(V)$ with $0\leq h\leq1$. Put
\[
 v_{\delta,\epsilon}=v+\epsilon(h-1)-\delta,
 \qquad E=\{u<v_{\delta,\epsilon}\}.
\]
Then
\[
 \{u<v-2\delta\}\subset E\subset\{u<v-\delta\}.
\]
The pair $(u,v_{\delta,\epsilon})$ satisfies the compact-gap condition.
Comparison and positivity of the mixed expansion give
\[
 0\leq\epsilon^q\int_{\{u<v-2\delta\}}\mu_h
 \leq\int_E\mu_{v_{\delta,\epsilon}}
 \leq\int_E\mu_u=0.
\]
Constants disappear in differentiated slots, which explains the
$\epsilon^q\mu_h$ term. Since $h$ was arbitrary, the capacity of
$\{u<v-2\delta\}$ is zero. Take $\delta$ through a sequence
decreasing to zero and use countable subadditivity. All sets are Borel;
none is assumed ordinary open.
\end{proof}

\begin{proposition}[Uniqueness modulo capacity zero]
\label{prop:capacity-uniqueness}
Let $u,v\in\Cb(V)$ satisfy the compact-gap condition in both orders.
If $\mu_u=\mu_v$ as Radon measures, then
\[
 \operatorname{Cap}_S(\{u\ne v\};V)=0.
\]
\end{proposition}

\begin{proof}
Fix $\delta>0$, $0<\epsilon<\delta$, and an admissible $h$ as above.
Set $u_{\delta,\epsilon}=u+\epsilon(h-1)-\delta$ and
$E=\{v<u_{\delta,\epsilon}\}$. Then
\[
 \{v<u-2\delta\}\subset E\subset\{v<u-\delta\}\Subset V,
\]
where the last compact containment refers to its closure. Comparison
for $(v,u_{\delta,\epsilon})$ and positivity yield
\[
 \int_E\mu_u+\epsilon^q\int_E\mu_h
 \leq\int_E\mu_{u_{\delta,\epsilon}}
 \leq\int_E\mu_v=\int_E\mu_u.
\]
The term being cancelled is finite, because $E$ has compact closure
in $V$. Hence $\mu_h(\{v<u-2\delta\})=0$. Taking the supremum
over $h$ and then a countable union in $\delta$ gives capacity zero
for $\{v<u\}$. Interchange $u$ and $v$ and take a union.
\end{proof}

\begin{corollary}[A conditional pointwise conclusion]
\label{cor:conditional-pointwise-domination}
Assume, in addition, the following property for $(S,V)$: every
nonempty strict gap $\{a<b-\delta\}$, for $a,b\in\Cb(V)$ satisfying
the compact-gap condition and $\delta>0$, has positive
$\operatorname{Cap}_S(\,\cdot\,;V)$. Then
Proposition~\ref{prop:capacity-domination} gives $u\geq v$
pointwise, and Proposition~\ref{prop:capacity-uniqueness} gives
$u=v$ pointwise.
\end{corollary}

\begin{proof}
If a prohibited pointwise inequality held at a point, it would hold
there with some positive margin $\delta$. Its strict gap is a subset
of the capacity-zero set already proved, contradicting the additional
property.
\end{proof}

The last corollary has status \texttt{CONDITIONAL}. The condition on
strict gaps is a new research obligation for any proposed class of
domains and backgrounds; positivity on nonempty ordinary opens alone
does not identify it. It fails for $S=0$, and may fail away from the
support of a nonzero background. The propositions retain their literal
meaning even when the capacity degenerates.

% END source: base_r3/base_r2/base_r1/sections/capacity_domination.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/counterexamples.tex
\section{Counterexamples to unrestricted extensions}
\label{sec:counterexamples}

All three examples use the original field \(K\), with its prescribed
absolute value. They concern statements stronger than the fixed-background
theorem.

\subsection{A bounded discontinuous classical psh function over \texorpdfstring{$K$}{K}}

\begin{proposition}[A sequence of branch wells]
\label{prop:branch-wells}
There exist an ordinary open disc \(V\subset\mathbf A_K^{1,\mathrm{an}}\), a
function \(w\in\mathcal C_b(V)\), a Gauss point \(\xi\in V\), and rigid
points \(x_i\in V\) such that
\[
 x_i\longrightarrow\xi,\qquad w(\xi)=0,\qquad w(x_i)=-1.
\]
The function \(w\) is the pointwise decreasing limit of continuous classical
psh functions on one fixed disc, uniformly bounded there.
\end{proposition}

\begin{proof}
Let \(\widetilde K\) be the finite residue field. Choose pairwise distinct
monic irreducible polynomials
\(\widetilde P_i\in\widetilde K[T]\), and lift them to monic polynomials
\(P_i\in K^\circ[T]\). Write \(d_i=\deg P_i\) and \(a_i=2^{-i}\), with
indices starting at \(i=1\). Thus \(\sum_i a_i=1\). Choose \(R>1\), put
\(C=\log R\), and work on
\[
 V=\{x:|T(x)|<R\}.
\]
For example one may take \(R=|\pi|^{-1}\); no normalization of the given
absolute value is made. Set
\[
 f_i(x)=\frac{a_i}{d_i}
       \max\!\left\{\log|P_i(x)|,-\frac{d_i}{a_i}\right\}.
\tag{7}
\]
At a zero of \(P_i\), the first entry is interpreted as \(-\infty\).
Each \(f_i\) is continuous and classical psh: logarithms of analytic
functions are classical psh, and this class is closed under maxima with
constants and positive scalar multiples (Gubler--Rabinoff,
Proposition 4.2).

On \(\{|T|\leq1\}\), one has \(-1\leq f_i\leq0\). Moreover, at any
point of this closed disc at most one \(f_i\) is nonzero. Indeed, for
\(i\ne j\), coprimality of the reductions gives integral polynomials
\(A,B,H\) with
\[
 1=A P_i+B P_j+\pi H.
\]
If both \(|P_i(x)|<1\) and \(|P_j(x)|<1\), evaluation at a point with
\(|T(x)|\leq1\) contradicts the non-Archimedean triangle inequality.
Consequently every finite sum of the \(f_i\) is bounded below by \(-1\)
on the closed unit disc.

If \(1<|T(x)|<R\), monicity and integrality give
\[
 |P_i(x)|=|T(x)|^{d_i},\qquad f_i(x)=a_i\log|T(x)|.
\]
In particular, \(f_i\leq a_i C\) everywhere on \(V\). Define
\[
 w_N=C+\sum_{i=1}^N(f_i-a_i C)
     =\sum_{i=1}^N f_i+C\sum_{i>N}a_i.
\tag{8}
\]
The functions \(w_N\) are continuous classical psh on the same open disc,
and
\[
 w_{N+1}-w_N=f_{N+1}-a_{N+1}C\leq0,
 \qquad -1\leq w_N\leq C.
\]
Thus \(w=\lim_N w_N\) is locally bounded classical psh, by the decreasing
limit property in Gubler--Rabinoff, Proposition 4.2(4).

Let \(\xi\) be the Gauss point of the closed unit disc. Since
\(|P_i(\xi)|=1\), equations (7)--(8) give \(w(\xi)=0\). Let \(x_i\)
be the rigid Berkovich point defined by \(P_i\). The irreducibility of
\(\widetilde P_i\) implies irreducibility of \(P_i\) over \(K\).
The point \(x_i\) is a point of the original \(K\)-analytic line; using
its finite residue extension to describe its seminorm is not a base
change of the analytic space. At this point, \(f_i(x_i)=-1\), while
\(f_j(x_i)=0\) for \(j\ne i\). Therefore \(w(x_i)=-1\).

It remains to verify the convergence of the points. For a nonzero
\(F\in K[T]\), multiply by a scalar so that its Gauss norm equals one.
Its reduction is then a nonzero polynomial over \(\widetilde K\).
Only finitely many distinct irreducibles \(\widetilde P_i\) divide that
reduction. For every other \(i\), evaluation at \(x_i\) is a unit,
so \(|F(x_i)|=\|F\|_{\mathrm{Gauss}}\). Hence
\(|F(x_i)|\to|F(\xi)|\) for every polynomial \(F\). This is precisely
convergence to \(\xi\) in the topology of the Berkovich affine line.
It follows in particular that \(w\) is discontinuous at \(\xi\).
\end{proof}

\begin{corollary}[Failure for varying backgrounds]
\label{cor:varying-background-false}
Weak convergence of positive closed backgrounds does not, by itself,
imply weighted convergence for a fixed locally bounded classical psh
weight. This fails already with no differentiated potential.
\end{corollary}

\noindent\textbf{Status of the unrestricted assertion.} \texttt{FALSE}.

\begin{proof}
On the curve in Proposition~\ref{prop:branch-wells}, take
\(S_i=\delta_{x_i}\) and \(S=\delta_\xi\). These are symmetric positive
top-degree original currents, and are separately \(d'\)- and
\(d''\)-closed for degree reasons. The convergence \(x_i\to\xi\)
gives \(S_i\to S\) on every compact original smooth scalar test.
Nevertheless,
\[
 wS_i=-\delta_{x_i}\longrightarrow-\delta_\xi,
 \qquad wS=0.
\]
Thus the failure occurs for \(p=n=1\) and \(k=0\), with a constant
sequence of admissible weights. It does not involve changing a function
on a null set, changing the base field, or changing the test sheaf.
\end{proof}

\begin{remark}[A strict inequality set need not be ordinary open]
The function $w$ above satisfies $w(\xi)=0$, $w(x_i)=-1$, and
$x_i\to\xi$. Thus $\{w>-1/2\}$ contains $\xi$ but contains no
$x_i$, and is not an ordinary open neighborhood of $\xi$. Both $w$
and the constant $-1/2$ are bounded classical psh functions. This
checks explicitly the distinction used in maximum locality.
\end{remark}

\subsection{A nonzero relative product carried by an ambient pluripolar set}

\begin{proposition}[Concentration on a smooth divisor]
\label{prop:polar-relative-mass}
For every \(m\geq1\), there are a smooth \((m+1)\)-dimensional ordinary
analytic open \(V\), a nonzero positive closed \((1,1)\)-current \(S\),
and bounded continuous classical psh functions \(u_1,\ldots,u_m\) such
that \(T_{S,m}(u_1,\ldots,u_m)\) is a nonzero positive measure supported
on an ambient pluripolar set.
\end{proposition}

\noindent\textbf{Status of universal ambient non-concentration.}
\texttt{FALSE}, even for continuous potentials and an integration-current
background.

\begin{proof}
Choose \(0<r<1<R\) and \(R_0>0\). With coordinates
\(z_0,z_1,\ldots,z_m\), put
\[
 V=\{|z_0|<R_0\}\times
       \prod_{i=1}^m\{r<|z_i|<R\},
 \qquad Y=\{z_0=0\}\cap V.
\]
These are ordinary opens in the relevant smooth algebraic
analytifications. Let \(\iota:Y\hookrightarrow V\) be the closed
embedding and take \(S=\iota_*[1]_Y\). This is a symmetric positive
closed original CLD current. The set \(Y\) is ambient pluripolar,
since the nontrivial classical psh function \(\log|z_0|\) equals
\(-\infty\) precisely there.

Set
\[
 u_i=\max\{0,\log|z_i|\},\qquad 1\leq i\leq m.
\]
Each \(u_i\) is bounded and continuous on \(V\). Pushforward commutes
with distributional \(D\), and continuous multiplication satisfies the
projection formula. Recursion therefore gives
\[
 T_{S,m}(u_1,\ldots,u_m)
   =\iota_* T^Y_{[1],m}(u_1|_Y,\ldots,u_m|_Y).
\tag{9}
\]
The right side can be calculated on
\((\mathbf P^1_K)^m\). For the standard model
\((\mathbf P^1_{K^\circ})^m\), take the pullback of
\(\mathcal O(1)\) from each of its \(m\) factors, with its standard
model metric. In the affine frames, their weights are the displayed
\(u_i\). The special fiber is irreducible and reduced, and the mixed
intersection degree of these \(m\) line bundles is one. The model
measure formula, together with R, Lemma 3.3, identifies their original
CLD mixed measure with the Dirac mass at the product Gauss point
\(\xi_Y\). This point lies in the displayed polyannulus. Restricting
to \(Y\) and using (9) gives
\[
 T_{S,m}(u_1,\ldots,u_m)=\delta_{\iota(\xi_Y)}.
\]
In particular it has positive mass on \(Y\).
\end{proof}

\paragraph{Implication for capacity.}
An ambient pluripolar set need not be negligible for a relative
capacity. Indeed, in this example a positive rescaling of
\(u_1+\cdots+u_m\) lies between zero and one on \(V\), and its
\(m\)-fold curvature relative to \(S\) has a positive contribution
on \(Y\). Any non-concentration theorem must impose a condition on
the background, or formulate polarity intrinsically on its support.

\subsection{The unqualified capacity can vanish identically}

\begin{proposition}[Compact-domain degeneration]
\label{prop:compact-capacity-zero}
Let \(X\) be a connected compact boundaryless \(K\)-analytic space on
which the classical maximum principle holds, for example
\(X=\mathbf P_K^{n,\mathrm{an}}\) with \(n\geq1\). Every finite
locally bounded classical psh function on \(X\) is constant. Hence
\[
 \sup_{\substack{u\in\mathcal C_b(X)\\0\leq u\leq1}}
       \int_E (Du)^n=0
\]
for every Borel set \(E\subset X\).
\end{proposition}

\noindent\textbf{Status of nondegeneracy on every ordinary domain.}
\texttt{FALSE}. The calculation itself is a
\texttt{DIRECT\_COROLLARY} of the classical maximum principle.

\begin{proof}
An upper-semicontinuous real function on a compact space attains its
maximum \(M\). Its maximum set \(F=\{u=M\}\) is nonempty and closed.
Every point of \(F\) is interior because \(X\) is boundaryless.
Gubler--Rabinoff, Theorem 6.1, makes \(u\) locally constant at such a
point. Thus \(F\) is also open, and connectedness gives \(F=X\).
The curvature of a constant vanishes, proving the capacity assertion.
\end{proof}

The entire space \(X\) is not pluripolar when the definition excludes
identically \(-\infty\) local witnesses. Thus an unqualified equivalence
between zero capacity and pluripolarity also fails for this capacity.
Two different constants have the same zero Monge--Amp\`ere measure;
with empty boundary, no Dirichlet uniqueness follows without a
normalization or additional geometric data.

% END source: base_r3/base_r2/base_r1/sections/counterexamples.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/comparison.tex
\section{Theorem-by-theorem comparison with the classical theory}
\label{sec:comparison}

\subsection{Scope and source conventions}

The comparison below separates the statement of a theorem from the category
in which it is proved. All complex statements use their authors' normalization
of \(dd^c\); replacing it by a positive constant multiple changes numerical
constants, not the conclusions listed here. All statements proved in this
report use \(D=d'd''\), the prescribed absolute value on the finite extension
\(K/\mathbf Q_p\), and the original compactly supported CLD smooth test
forms. No renormalization of the valuation is implicit.

The following source versions have been read and pinned: BT76 and BT82 are
the original journal articles; Dem12 is the June 21, 2012 version; CLD12 is
arXiv:1204.6277v1; CLD25 is arXiv:1204.6277v2; GR25 is
arXiv:2506.13548v2; Wu26 is arXiv:2609.27610v1; GK17 is the
arXiv:1406.7637v3 text of the subsequently published article; BGJK21 is
arXiv:2102.07392v1. The theorem numbers below refer to these versions.
The source register records URLs, hashes, and verification notes. Page
numbers are printed page numbers, which agree with PDF page numbers in the
quoted classical and CLD files.

\paragraph{The 2025 test-category issue.}
CLD25 develops Chapter 8 on its tropical spaces and subsequently applies
that formalism to analytic spaces; it also develops harmonic tropicalizations
and additional differential structures. Its theorems are recorded as
\texttt{EXTERNAL\_KNOWN} in their own stated categories. An arbitrary
functional on the original CLD12 test sheaf is not automatically a current
on an enlarged test sheaf. Accordingly, no CLD25 assertion is used here to
extend an arbitrary original current \(S\) without a proved comparison of
the relevant current categories. The proofs in this report instead use the
original-category inputs R and B. Products arising from common smooth or
model data can be compared by their actual definitions; this does not prove
an extension theorem for arbitrary currents.

\paragraph{Internal provenance.}
R, Theorem 2.9, provides continuous-potential regularization in its stated
algebraic setting, with R, Lemma 3.1, supplying the original positive-cone
coefficient-Radon and continuous-multiplier control. B, Theorems 2.5--2.7,
provide classical-psh regularization and the entire bounded-potential
calculus with background \([1]\), including weighted joint decreasing
continuity. These are \texttt{INTERNAL\_INPUT}, not new results of this
round. A2, Theorem 1.1, is already an established input under its number-field,
integrable-adelic, fixed-boundary hypotheses. It is not an open direction in
the present comparison.

\subsection{Status matrix}

Each status below refers to the precise assertion in its cell. In particular,
an \texttt{OPEN} entry is not a claim that every special case is unknown in
the literature, and a \texttt{FALSE} entry concerns the explicitly unrestricted
version. Local capacity and energy refinements proved elsewhere in this
report have their own hypotheses; they are not silently identified with a
nondegenerate global Bedford--Taylor capacity.

\begin{longtable}{@{}>{\raggedright\arraybackslash}p{.05\textwidth}>{\raggedright\arraybackslash}p{.24\textwidth}>{\raggedright\arraybackslash}p{.31\textwidth}>{\raggedright\arraybackslash}p{.31\textwidth}@{}}
\hline
 & Classical target & Established non-Archimedean baseline & Assessment in this project\\
\hline
\endfirsthead
\hline
 & Classical target & Established non-Archimedean baseline & Assessment in this project\\
\hline
\endhead
1 & Locally bounded psh functions and decreasing regularization &
GR25 gives the classical curve-test class; B 2.5 identifies it with its
decreasing semipositive PL regularization class. &
\texttt{PROVED / INTERNAL\_INPUT}. Continuous original-current positivity
is covered by R in R's scope; unrestricted distributional positivity is not
silently equated with classical psh.\\[3pt]
2 & Positive closed currents and mixed products &
CLD12 treats approachable continuous potentials; CLD25 treats continuous
potentials on arbitrary backgrounds in its category; B 2.6 treats \([1]\). &
Arbitrary fixed original positive closed \(S\), bounded classical psh
potentials: \texttt{PROVED / NEW\_PROVED} in the fixed-background theorem.
Ordered existence alone is a \texttt{DIRECT\_COROLLARY} of R/B.\\[3pt]
3 & Chern--Levine--Nirenberg estimates &
CLD12 5.6.2 gives the smooth estimate; CLD25 8.4.1 gives a relative smooth
estimate in its category. &
Original coefficient-Radon relative mass control for arbitrary fixed
\(S\): \texttt{PROVED / DIRECT\_COROLLARY}. No global strictly psh form is assumed.\\[3pt]
4 & Symmetry and weighted monotone convergence &
B 2.6--2.7 gives the full \([1]\)-background result; CLD25 8.7.2 concerns
local uniform convergence of continuous potentials. &
Fixed arbitrary original \(S\): \texttt{PROVED / NEW\_PROVED}.
Unrestricted varying-background weighted stability:
\texttt{FALSE}, Corollary~\ref{cor:varying-background-false}.\\[3pt]
5 & Locality and maximum principles &
Ordinary locality is in B; strict-set maximum locality is already a
\texttt{DIRECT\_COROLLARY} of B 2.5--2.7. GR25 6.1 gives the classical
psh maximum principle. &
The extension of Borel maximum locality to arbitrary fixed \(S\):
\texttt{PROVED / NEW\_PROVED},
Theorem~\ref{thm:maximum-locality-relative}. The set \(\{u>v\}\) is Borel;
ordinary openness is not asserted.\\[3pt]
6 & Bedford--Taylor capacities &
The mixed measures supply candidate set functions; neither a formal
supremum nor CLN alone proves nondegeneracy or full Choquet continuity. &
The stated Borel axioms and mixed domination are \texttt{PROVED}; the
norm-box estimate in Theorem~\ref{thm:relative-capacity-estimate} is
\texttt{PROVED / NEW\_PROVED}. Outer regularity, continuity from above
on compact sets, and Choquet capacitability remain \texttt{OPEN}.
Nondegeneracy on every analytic domain: \texttt{FALSE},
Proposition~\ref{prop:compact-capacity-zero}. A complete intrinsic domain
classification remains \texttt{OPEN}.\\[3pt]
7 & Quasicontinuity and convergence in capacity &
No unrestricted such theorem follows just from weak monotone convergence.
The complex energy proof supplies an exact template, with a dependency
warning explained below. &
Local bounded convergence in capacity and quasicontinuity on smaller
cores of sufficiently small norm boxes:
\texttt{PROVED / NEW\_PROVED}, Corollary~\ref{cor:relative-quasicontinuity}.
They use the stated Borel capacity and ordinary open exceptional sets;
the decreasing family is defined near the closure of the box.
With actual local uniform bounds, convergence in top relative capacity
on all prescribed local norm boxes gives nonmonotone weighted stability:
\texttt{PROVED / DIRECT\_COROLLARY},
Theorem~\ref{thm:capacity-stability}.
Full global/unbounded analogues and outer regularity remain \texttt{OPEN}.\\[3pt]
8 & Pluripolar and negligible sets &
CLD25 16.9.2--4 gives integrability and non-concentration for continuous
approachable potentials against \([1]\)/smooth tests, with hypotheses on
the polar witness. &
Universal ambient non-concentration for arbitrary \(S\): \texttt{FALSE},
Proposition~\ref{prop:polar-relative-mass}. The full zero-capacity/polar/
negligible equivalence is \texttt{OPEN} in the intended domain class.\\[3pt]
9 & Comparison and domination &
The compact-gap comparison for \([1]\) and B mixed backgrounds is a
\texttt{DIRECT\_COROLLARY} of B. Wu26 4.15--4.16 gives separate global
projective results with its stated hypotheses. &
Extension of compact-gap comparison to arbitrary fixed \(S\):
\texttt{PROVED / NEW\_PROVED}, Theorem~\ref{thm:relative-comparison}.
Domination and uniqueness modulo capacity zero are
\texttt{PROVED / DIRECT\_COROLLARY},
Propositions~\ref{prop:capacity-domination}--\ref{prop:capacity-uniqueness}.
Pointwise conclusions require strict-gap capacity detection and are
\texttt{CONDITIONAL}; unrestricted arbitrary-background pointwise
domination remains \texttt{FALSE}.\\[3pt]
10 & Relative extremal functions &
Global envelope theorems in Wu26 have residue-characteristic and global
geometric hypotheses; they do not give the required local theorem over
finite extensions of \(\mathbf Q_p\). &
Local envelope regularity, equilibrium support, and capacity/mass
identities in a suitable domain class: \texttt{OPEN}. Proposed statements
under additional barrier/envelope assumptions are \texttt{CONDITIONAL}.\\[3pt]
11 & Monge--Amp\`ere and Dirichlet problems &
CLD supplies the continuous operator and model comparisons; BGJK21 solves
specified invariant toric/abelian equations. &
The bounded \([1]\) operator is \texttt{PROVED / INTERNAL\_INPUT} (B);
its fixed arbitrary-background extension is also proved here. Appropriate general local
Dirichlet existence and uniqueness are \texttt{OPEN}. Unqualified
existence/uniqueness on every domain is \texttt{FALSE}.\\
\hline
\end{longtable}

\subsection{1. Locally bounded plurisubharmonic functions}

\paragraph{Classical statement and location.}
Dem12, Chapter I, Definition 5.1 and Theorems 5.4--5.8, pp.~39--40,
define psh functions by upper semicontinuity and subharmonicity on complex
lines, prove decreasing-limit stability, decreasing smooth convolution
regularization on smaller opens, convex increasing composition, and the
distributional curvature characterization. In Theorem 5.5 the function is
not identically \(-\infty\) on any component, and the regularization
\(u*\rho_\varepsilon\) is defined on \(\Omega_\varepsilon\), not on an
arbitrary unchanged domain. The upper-envelope theorem is I.5.7; its
proof uses Choquet's lemma I.4.23 and the submean inequality.
\cite[Ch.~I, pp.~39--40]{Dem12}

\paragraph{Non-Archimedean comparison.}
GR25, Definition 4.1 and Proposition 4.2, pp.~21--22, use pullbacks to
analytic curves after every complete valued extension. Their basic
properties include maxima, positive linear combinations, logarithms of
analytic functions, and decreasing nets. Theorem 4.3, pp.~22--23, relates
classical psh PL functions to semipositivity under its good/strict/nontrivial
valuation hypotheses. B, Theorem 2.5, is the stronger accepted input
needed here: actual pointwise locally bounded classical psh functions admit
decreasing continuous semipositive PL approximations on a fixed ordinary
neighborhood over the prescribed field.\cite[\S4]{GR25}

CLD12, Definition 5.5.1, p.~84, uses a distributional definition for
continuous functions. Definition 5.6.3 and Remark 5.6.4, p.~87, introduce
local uniform psh approachability and explicitly identify the then-missing
implication from continuous current positivity to approachability. R,
Theorem 2.9, resolves the stated continuous problem in R's setting.
CLD25, Definition 8.1.1, p.~195, instead defines a sheaf by closure under
decreasing sequences; Definition 8.1.7 and Proposition 8.1.11,
pp.~196--198, specify its continuous class. Theorem names or similar
notation do not identify these classes with classical psh without a bridge.
\cite[\S\S5.5--5.6]{CLD12}\cite[\S8.1]{CLD25}

\paragraph{Required change.}
The present theory fixes B's actual pointwise class \(\mathcal C_b\).
There is no replacement of a function by an almost-everywhere representative:
an arbitrary background can charge sets invisible to ambient volume.

\subsection{2. Positive currents and the recursive product}

\paragraph{Classical statement and proof ingredients.}
Dem12, III.1.14, p.~133, gives measure coefficients and trace control for
positive currents; III.1.23, p.~136, gives local weak compactness under mass
bounds. III.3.2, p.~145, states that for locally bounded psh \(u\) and
arbitrary positive closed \(T\), \(dd^c(uT)\) is positive closed. Its proof
uses decreasing smooth regularization, bounded convergence for the fixed
coefficient measures of \(T\), and continuity of distributional
differentiation. Iteration gives an ordered product before symmetry is
known. BT82, equation (2.1), p.~4, records the recursive definition for the
\([1]\)-background products.\cite[III.1.14, III.1.23, III.3.2]{Dem12}
\cite[\S2, p.~4]{BT82}

\paragraph{Non-Archimedean comparison.}
CLD12, Definition 5.4.1, pp.~81--82, distinguishes the three positivity
cones. Proposition 5.4.6 and Corollary 5.4.7, pp.~83--84, treat top positive
currents as Radon measures and continuous multiplication. Corollaries
5.6.5--5.6.6, p.~88, construct approachable continuous mixed products.
Paragraph 5.6.8, p.~89, explicitly discusses \(D(uT)\) for bounded Borel
\(u\) and explains that positivity needs appropriate decreasing psh
regularization; it is not a completed locally bounded theory.
\cite[\S\S5.4, 5.6]{CLD12}

CLD25, Lemma 8.5.5, p.~207, and \S8.7.1, p.~209, give continuous
recursive products on arbitrary positive closed backgrounds in that
formalism. Its Definition 8.6.2, p.~208, calls a current ``Bedford--Taylor''
if the resulting operation depends only on the curvature of its potential.
That extra property is different from the fixed-potential recursion asked
for here and is not assumed for our arbitrary \(S\).
\cite[\S\S8.5--8.7]{CLD25}

R's coefficient-Radon theorem concerns the original positive cone, whereas
B's \([1]\)-products are strongly positive. These assertions are not
interchangeable in middle degree. The arbitrary-background ordered
construction uses R's exact cone, B's regularization, bounded convergence
against fixed coefficient measures, and continuous-stage positivity. Thus
ordered existence is correctly labeled \texttt{DIRECT\_COROLLARY}; its
later joint continuity and symmetry require further analysis.

\subsection{3. Relative Chern--Levine--Nirenberg estimates}

\paragraph{Classical statement.}
For compact \(L\subset K^\circ\) on a complex manifold and fixed positive
closed \(T\), Dem12, III.3.3, p.~146, proves
\[
 \|dd^cu_1\wedge\cdots\wedge dd^cu_q\wedge T\|_L
 \leq C_{K,L}\prod_{i=1}^q\|u_i\|_{L^\infty(K)}\,\|T\|_K.
\]
The mass seminorm is the sum of absolute variations of local coefficient
measures; its equivalence to a positive trace is explained on p.~145.
The proof iterates a cutoff integration by parts and domination of the
cutoff's Hessian by a local positive trace. BT82, Theorem 2.10, p.~10,
also records mass, bounded-weight, and energy estimates.
\cite[III.3.3]{Dem12}\cite[Theorem 2.10]{BT82}

\paragraph{Non-Archimedean comparison and direct consequence.}
CLD12, Proposition 5.6.2, pp.~86--87, proves its smooth estimate.
CLD25, Proposition 8.4.1, pp.~202--203, gives a smooth relative version;
Proposition 8.4.3, pp.~203--205, gives an additional weighted estimate
under a smooth psh defining-function hypothesis. These are not statements
that every analytic open has a strictly psh exhaustion.
\cite[Proposition 5.6.2]{CLD12}\cite[\S8.4]{CLD25}

For the original arbitrary \(S\), Theorem~\ref{thm:finite-test-cln}
provides a finite family of positive original tests controlling each
requested coefficient test, uniform constants after differentiation and
cutoff, and compatibility with tropical refinements. Its status is
\texttt{DIRECT\_COROLLARY}: finite iteration of R's positive coefficient
control and the ordered recursion supplies the estimate. It is not counted
as a new analytic breakthrough. The result controls arbitrary original
complementary tests, not just positive scalar tests against top-degree
measures.

\subsection{4. Symmetry and weighted joint decreasing continuity}

\paragraph{Classical statement.}
BT82, Theorem 2.1 and Corollary 2.2, p.~4, give joint decreasing
continuity and symmetry for the \([1]\)-products. Theorem 2.4,
pp.~7--9, includes a simultaneously decreasing locally bounded psh weight.
For arbitrary fixed positive closed \(T\), Dem12, III.3.7,
pp.~147--149, proves
\[
 u_{0,j}\,dd^cu_{1,j}\wedge\cdots\wedge dd^cu_{q,j}\wedge T
 \longrightarrow
 u_0\,dd^cu_1\wedge\cdots\wedge dd^cu_q\wedge T,
\]
when all \(u_{i,j}\downarrow u_i\) and the limits are locally bounded.
III.3.10, p.~149, then deduces symmetry.\cite[Theorems 2.1, 2.4]{BT82}
\cite[III.3.7--3.10]{Dem12}

\paragraph{The decisive classical mechanism.}
Dem12 first max-patches every potential to one smooth psh function in a
common boundary collar. CLN yields compactness. Lemma III.3.9, p.~148,
gives a one-sided inequality for a weak limit of decreasing
upper-semicontinuous multipliers times varying positive measures.
Integration by parts moves one smooth regularization at a time; ordered
limits then give the reverse inequality for the total positive trace.
Trace positivity eliminates the defect. This is more than formal
integration by parts, and it is not valid to exchange two discontinuous
limits without the trace argument. The present proof supplies a
test-adapted original-CLD replacement for the required collar and trace.

\paragraph{Baseline and forbidden strengthening.}
B, Theorem 2.7, already proves the whole weighted assertion for background
\([1]\). CLD25, Proposition 8.7.2, pp.~209--210, instead proves local
\emph{uniform} convergence of continuous potentials with weakly varying
backgrounds; Proposition 8.7.4 gives continuous symmetry. The change from
uniform to decreasing convergence is substantive. The branch-well
counterexample in Corollary~\ref{cor:varying-background-false} shows why
one must keep the background fixed in the bounded discontinuous theorem.
\cite[Propositions 8.7.2, 8.7.4]{CLD25}

\subsection{5. Ordinary locality, maximum locality, and maximum principles}

\paragraph{Exact locality assertion.}
For \(u,v,w_2,\ldots,w_k\) locally bounded psh, the desired maximum
identity is
\[
 \mathbf1_{\{u>v\}}\,
 dd^c\max(u,v)\wedge dd^cw_2\wedge\cdots\wedge T
 =\mathbf1_{\{u>v\}}\,
 dd^cu\wedge dd^cw_2\wedge\cdots\wedge T.
\]
It follows directly from the verified weighted theorem BT82, Theorem 2.4,
by the continuous-approximation and weighted-difference argument proved
in Theorem~\ref{thm:maximum-locality-relative}. We do not assign it an
invented numbered location in BT82. Ordinary-open locality follows
immediately from recursion; Borel strict-set locality is a separate
statement.

Wu26, Theorem 4.11, p.~27, supplies local maximum locality for its
top-degree products. Part (1) in the pinned v1 has \(\mathrm{MA}(v)\)
where the right-hand side on \(\{u>v\}\) must be
\(\mathrm{MA}(u)\); part (2) has the correct orientation. This textual
issue is recorded, not used as a premise. B 2.5--2.7 alone already imply
the \([1]\) maximum identity by the proof above, including all
complementary tests.\cite[Theorem 4.11]{Wu26}

The classical local maximum principle for psh functions has a distinct
non-Archimedean counterpart: GR25, Theorem 6.1, pp.~30--31, states local
constancy at an interior local maximum on the connected situation of that
theorem; Theorem 6.4, pp.~33--34, gives a Shilov-boundary maximum
principle on affinoids. Neither assertion identifies ordinary opens with
sets defined by inequalities of discontinuous psh functions.
\cite[\S6]{GR25}

\subsection{6. Capacities and their elementary axioms}

\paragraph{Classical definition and exact axioms.}
BT82, Definition 3.1, p.~11, defines for compact \(F\subset\Omega\)
\[
 \begin{aligned}
 C(F,\Omega)&=\sup_{0<u<1}\int_F(dd^cu)^n,\\
 C(E,\Omega)&=\sup_{\substack{F\subset E\\F\text{ compact}}}C(F,\Omega).
 \end{aligned}
\]
For Borel \(E\), the latter agrees with the supremum of the integrals
over \(E\). Proposition 3.2 gives monotonicity, contravariance in the
domain, countable subadditivity, and continuity from below for Borel
sets. Section 6, p.~22, defines the outer capacity
\(C^*(E)=\inf_{\substack{G\supset E\\G\text{ open}}}C(G)\);
Proposition 6.1 gives its elementary properties.
\cite[Definition 3.1, Proposition 3.2, \S6]{BT82}

Theorem 8.2, p.~29, proves the full Choquet axioms on strongly
pseudoconvex domains, including continuity along decreasing compact sets
and along arbitrary increasing sets. The Choquet theorem stated there
then gives capacitability of analytic sets. These last assertions are
not formal consequences of defining a supremum over measures.
\cite[Theorem 8.2 and the following Choquet theorem]{BT82}

\paragraph{Established non-Archimedean properties and remaining requirements.}
Proposition~\ref{prop:capacity-elementary} proves monotonicity, countable
subadditivity on Borel sets, compact finiteness, continuity from below on
increasing Borel sets, domain contravariance, and mixed-measure domination
for \(\operatorname{Cap}_{S,\beta,k}\). It uses the constructed positive
measures, symmetry, and relative CLN. For this Borel capacity, equality
with its outer envelope remains \texttt{OPEN}. Defining a separate outer
capacity does not establish that equality or transfer the proved estimates
to it. Continuity from above on compact sets and Choquet capacitability
also remain \texttt{OPEN}. Nondegeneracy requires enough admissible psh curvature on
the chosen domain; it fails identically on a connected proper space for
the untwisted capacity, by Proposition~\ref{prop:compact-capacity-zero}.
No theorem in the checked CLD12 \S\S5.4--5.7 or CLD25 Chapter 8 supplies
the full bounded-discontinuous relative capacity package on all ordinary
domains. A useful local norm-box capacity is a more limited statement and
must retain its localization data.

\subsection{7. Quasicontinuity and convergence in capacity}

\paragraph{Classical theorem and dependency.}
BT82, Proposition 3.3, p.~12, gives the energy Cauchy--Schwarz inequality.
Theorem 3.4, pp.~12--13, proves capacity convergence for decreasing
potentials in the common-collar class \(P^*(\Omega,F,\rho,A,B)\),
defined in (2.2), pp.~4--5, by agreement with \(A\rho+B\) off a
compact set. The proof repeatedly applies Cauchy--Schwarz and ends by
using the \emph{already proved} weighted Theorem 2.4; see p.~13.
Consequently it cannot be imported to prove that weighted theorem in a
circle. Theorem 3.5, pp.~13--14, gives quasicontinuity. Its unbounded
case uses the sublevel-capacity estimate Theorem 6.10, p.~26.
\cite[Proposition 3.3; Theorems 3.4, 3.5, 6.10]{BT82}

Theorem 7.2, pp.~26--27, gives quasi-uniform convergence for monotone
psh sequences converging almost everywhere to a psh limit. Theorem 7.4,
p.~27, gives a stronger mixed continuity theorem with all but one
sequence monotone. These require the intervening negligible-set and
capacity theory; they are not corollaries of fixed-background weak
decreasing continuity alone.\cite[Theorems 7.2, 7.4]{BT82}

\paragraph{Transfer achieved in this round and its scope.}
After weighted continuity and symmetry have been established independently,
Theorem~\ref{thm:relative-capacity-estimate} proves the uniform estimate
over the relative capacity tests on a polynomial norm box. Its hypotheses
include a common smooth psh collar and a compact support for the differences
inside a strictly smaller nested box. Corollary~\ref{cor:relative-quasicontinuity}
then proves convergence in capacity on a smaller core for a decreasing
\(\mathcal C_b\) family defined near the closure of the box, and local
bounded quasicontinuity with ordinary open exceptional sets. These are
\texttt{PROVED / NEW\_PROVED} relative results. They concern the defined
Borel capacity; no outer-regularity or nondegeneracy assertion is implicit.
For \(k=n-p\), the trace form \(\beta\) drops out. A general sequence of
Borel multipliers cannot replace the admissible psh differences. Unbounded
truncation estimates, global capacity nondegeneracy criteria, and the full
monotone-a.e. theorem remain \texttt{OPEN}.

\paragraph{Nonmonotone stability now available.}
Theorem~\ref{thm:capacity-stability} is a
\texttt{DIRECT\_COROLLARY} of this established theory. Its hypotheses
are actual bounds uniform in the sequence index on every compact set,
and convergence of every potential and the psh weight in
\(\operatorname{Cap}_S(\,\cdot\,;\Omega)\) on every relatively compact
polynomial norm-box domain \(\Omega\) whose closure lies in an ordinary
algebraic neighborhood in \(U\). The quantified family includes boxes
inside any prescribed finite moment presentation. It proves weighted
joint convergence on all original compact smooth complementary tests
without monotonicity. It does not assert equivalence with convergence
for one larger-domain capacity. On each finite presentation, bounded
\(\rho=\tfrac12\sum t_a^2\) fills the missing differentiated slots
and controls lower-degree coefficient variations by the top relative
capacity. Quasicontinuity then supplies the fixed-multiplier limit.
These arguments use neither outer regularity nor a capacity compactness
theorem, and are not used in proving the earlier monotone or capacity
estimates.

\subsection{8. Pluripolar sets and negligible sets}

\paragraph{Classical statements.}
BT82, Proposition 5.1, p.~18, proves that a negligible set
\(\{h<h^*\}\), where \(h\) is a locally upper bounded psh envelope,
has capacity zero. Theorem 6.9, pp.~25--26, identifies outer-capacity
zero sets with pluripolar sets on strongly pseudoconvex domains;
Corollary 6.6 and Proposition 6.3 supply the equivalent extremal-function
condition. Theorem 7.1, p.~26, concludes that negligible sets are
pluripolar. Theorem 8.3, p.~29, characterizes pluripolarity of analytic
sets by their compact subsets.\cite[\S\S5--8]{BT82}

\paragraph{What CLD already proves.}
CLD25, Theorem 8.12.1 and Corollary 8.12.3, pp.~223--224, give
integrability and polar non-concentration for continuous locally
approachable potentials, assuming suitable smooth psh compact
sublevels and an approachable locally integrable polar witness. In the
analytic setting, Proposition 16.9.1 and Theorem 16.9.2,
p.~466, and Corollaries 16.9.3--16.9.4, p.~467, give the corresponding
conclusions for \(Du_1\cdots Du_m\wedge\alpha\), where \(\alpha\)
is a smooth complementary form. In particular they prove the indicated
measures do not charge lower-dimensional Zariski-closed sets.
\cite[\S8.12, \S16.9]{CLD25}

These theorems are not an assertion for an arbitrary positive closed
background. Proposition~\ref{prop:polar-relative-mass} gives a positive
integration-current background whose nonzero differentiated relative
product is entirely carried by an ambient polar divisor. Thus the
correct relative theory must impose a condition on \(S\), use a
background-relative notion of negligible set, or make a statement on
the support of \(S\). Even for \([1]\), extending the full
zero-capacity/polar/negligible equivalence to the intended bounded theory
requires envelope and unbounded-witness analysis not supplied by the
first-stage calculus.

\subsection{9. Comparison, domination, and uniqueness}

\paragraph{Classical theorem.}
BT82, Theorem 4.1, pp.~14--16, proves for bounded psh \(u,v\) on a
bounded open \(\Omega\subset\mathbf C^n\), with
\(\liminf_{z\to\partial\Omega}(u-v)\geq0\), that
\[
 \int_{\{u<v\}}(dd^cv)^n\leq
 \int_{\{u<v\}}(dd^cu)^n.
\]
Corollaries 4.4--4.5, p.~17, give uniqueness with common boundary data
and the domination principle under the stated zero-mass hypothesis.
The printed proof of uniqueness uses a strictly psh perturbation and
the fact that a nonempty strict inequality set of psh functions has
positive ambient volume. This detection step is additional to comparison.
\cite[Theorem 4.1, Corollaries 4.4--4.5]{BT82}

\paragraph{Non-Archimedean statement now available.}
Theorem~\ref{thm:relative-comparison} proves the relative inequality for
arbitrary fixed original \(S\) when every
\(\overline{\{u<v-\varepsilon\}}\) is compactly contained in the
domain. Its proof uses Borel maximum locality and the compact-support
identity
\[
 T_{S,m}(w^m)-T_{S,m}(u^m)
  =D\!\left((w-u)\sum_{j=0}^{m-1}
       T_{S,m-1}(w^j,u^{m-1-j})\right),
 \qquad w=\max(u,v-\varepsilon).
\]
There is no subtraction of two infinite total masses. However,
\(S=0\), or a background missing directions or regions, prevents
comparison from implying domination. Wu26, Theorems 4.15--4.16,
pp.~29--30, gives separate global projective comparison/domination
results with its polarization and envelope-property hypotheses; those
are not the arbitrary-background local assertion.\cite[Theorems 4.15--4.16]{Wu26}

\paragraph{Comparison already implied by B.}
For \(S=[1]\), the same compact-gap comparison proof uses only B's
symmetry, recursion, coefficient-Radon multiplication, and the maximum
locality already derived from B 2.5--2.7. Thus this case is
\texttt{DIRECT\_COROLLARY}, as is the case where \(S\) is already a B
mixed current, by absorbing its fixed potentials into the differentiated
slots. The \texttt{NEW\_PROVED} assertion is the extension to arbitrary
fixed original positive closed backgrounds.

\paragraph{The measure-theoretic domination conclusion.}
Proposition~\ref{prop:capacity-domination} now proves, under the same
compact-gap condition in the order \((u,v)\), that
\[
 \mu_u(\{u<v\})=0
 \quad\Longrightarrow\quad
 \operatorname{Cap}_S(\{u<v\};V)=0.
\]
If \(\mu_u=\mu_v\) and the compact-gap condition holds in both
orders, Proposition~\ref{prop:capacity-uniqueness} proves
\(\operatorname{Cap}_S(\{u\ne v\};V)=0\). These are
\texttt{DIRECT\_COROLLARY} statements for the defined Borel capacity;
they do not identify a capacity-zero set with the empty set or an ambient
pluripolar set. The additional property that every nonempty strict gap
of such potentials has positive relative capacity gives the pointwise
conclusions in Corollary~\ref{cor:conditional-pointwise-domination},
which is explicitly \texttt{CONDITIONAL}. Establishing that detection
property for a useful domain/background class remains \texttt{OPEN};
positivity of capacity on ordinary open sets alone is insufficient.

\subsection{10. Relative extremal functions and equilibrium measures}

\paragraph{Classical statement.}
For \(E\subset\Omega\), set
\[
 h_E=\sup\{v\in\operatorname{PSH}(\Omega):v\leq0,\ v|_E\leq-1\}.
\]
BT82, Proposition 5.3, p.~19, proves for compact \(E\) in a
strictly pseudoconvex domain that \((dd^ch_E^*)^n\) is supported on
\(E\) and that its total mass equals \(C(E,\Omega)\). The envelope
is psh after upper regularization; the proof also needs increasing
convergence (Proposition 5.2), negligible sets, comparison, and local
homogeneous Dirichlet replacements. The simultaneous induction is
explicit on pp.~19--22. Merely writing the envelope does not prove its
equilibrium identity.\cite[\S5, especially Proposition 5.3]{BT82}

\paragraph{Actual non-Archimedean frontier.}
One must first specify ordinary domains, admissible boundary barriers,
and a class for which upper regularized envelopes stay classical psh.
Then one must prove maximality off the obstacle and identify its
Monge--Amp\`ere measure. Wu26's principal envelope theorem,
Theorem 7.1, p.~51, concerns geometrically unibranch projective
varieties (in particular normal varieties) over discretely or trivially
valued fields of equicharacteristic zero; the residue
characteristic excludes the finite extensions of \(\mathbf Q_p\)
fixed here. B's accepted local regularization remains available but
does not by itself prove the required obstacle-envelope theorem.
\cite[Theorem 7.1]{Wu26}

\subsection{11. Monge--Amp\`ere operators and Dirichlet problems}

\paragraph{Original classical theorem.}
BT76, Theorem D, p.~4, states: on a bounded strictly pseudoconvex
\(\Omega\subset\mathbf C^n\), for continuous boundary data
\(\varphi\in C(\partial\Omega)\) and nonnegative
\(f\in C(\overline\Omega)\), there is a unique
\(u\in\operatorname{PSH}(\Omega)\cap C(\overline\Omega)\)
with the prescribed boundary data and \((dd^cu)^n=f\), using the
paper's volume-form convention. Its existence proof is in \S8,
pp.~39--43, and uses the regularity result Theorem C on balls plus
spherical modifications. The theorem does not assert solvability for
every positive Radon measure.\cite[Theorems C--D and \S8]{BT76}

\paragraph{Non-Archimedean consequence and remaining problem.}
B already supplies the bounded \([1]\)-background operator. The present
first-stage theorem extends its calculus to arbitrary fixed original
backgrounds, and the comparison theorem proves the relative compact-gap
inequality; that is necessary but not
sufficient for existence or uniqueness. One still needs a domain
geometry, barriers, envelope regularity, and a measure class compatible
with the actual capacity. BGJK21 proves invariant toric and abelian
Monge--Amp\`ere existence results, with specified global geometry and
data; it is a model for selected mechanisms, not an arbitrary local
Dirichlet theorem. A proper connected untwisted space already shows
why empty-boundary uniqueness fails for this formulation.
\cite[\S\S6--8]{BGJK21}

\subsection{Compatibility with tropical, \texorpdfstring{$\delta$}{delta}-form, and adelic theories}

BGJK21, Theorem 3.1.4, pp.~13--15, shows that locally bounded tropical psh
functions on its toric partial compactifications are finite and
continuous. Theorem 4.1.2, pp.~20--21, gives arbitrary positive closed
Lagerberg-background products, with local uniform convergence of the
potentials and weak convergence of the background. Theorem 6.1.1,
p.~33, compares the corresponding tropical and non-Archimedean
continuous products. Its bounded class is therefore not a source of
the missing discontinuous-potential argument on general Berkovich spaces.
\cite[Theorems 3.1.4, 4.1.2, 6.1.1]{BGJK21}

GK17, Proposition 4.15, constructs the differential algebra of
\(\delta\)-forms; its Poincar\'e--Lelong theorem is Theorem 7.2, and
Theorem 10.5 identifies formal-metric mixed measures with
Chambert--Loir measures. These statements include nonpositive,
nonclosed, and nondiagonal objects beyond the present BT calculus.
Agreement of two ordinary currents on the original smooth tests does
not imply agreement on every \(\delta\)-test. On common semipositive
PL/model data, ordinary-current agreement is supplied by the
continuous-stage identification, R's model-measure comparison, and
recursion; a broader \(\delta\)-test comparison remains a separate
theorem.\cite[Proposition 4.15, Theorems 7.2, 10.5]{GK17}

R, Theorems 2.9--2.10 and Lemma 3.3, already supply the stated continuous
regularization, positivity, and model/adelic mixed-measure comparisons.
A2, Theorem 1.1, already obtains same-model geometric adelic nefness
from one-place original-current positivity for the specified
integrable arithmetic adelic rational line. Its fixed compactification,
full Cartier boundary, prescribed open-line isomorphisms, and number-field
hypotheses remain part of that result. None is removed by the local BT
extension, and none needs to be proved again in this project.

\paragraph{Dependency discipline.}
The useful classical order is
\[
 \begin{aligned}
 &\text{regularization and coefficient measures}\\
 &\quad\Longrightarrow\ \text{ordered products and CLN}\\
 &\quad\Longrightarrow\ \text{weighted monotone continuity and symmetry}\\
 &\quad\Longrightarrow\ \text{energy/capacity estimates and maximum locality}.
 \end{aligned}
\]
Extremal functions, pluripolar equivalences, and Dirichlet existence
require further inputs. The comparison proof given here follows directly
from maximum locality and compact-support mass conservation and does
not assume those later inputs. This separation prevents capacity from
being used circularly to justify the weighted theorem on which its
classical proof already depends.

% END source: base_r3/base_r2/base_r1/sections/comparison.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/base_r1/sections/frontiers.tex
\section{Minimal remaining problems and the next research step}
\label{sec:frontiers}

The fixed-background bounded-potential calculus is proved relative to
the accepted inputs. The following questions are not hidden hypotheses
of that proof. They concern genuinely later parts of the proposed
theory, or a stronger notion of independence than the one requested
for the recursive product.

\subsection{Compact regularity, then envelopes}

The first local analytic target is the weak-closure domination statement
in Section~\ref{sec:outer-regularity-attempt}. Its quantifiers are
important: the normalized potentials live on one fixed domain,
their measures need only converge weakly, and the resulting bound must
use the capacity of that same domain. It suffices for compact outer
regularity and compact continuity from above. It is weaker than a
compactness-and-representation theorem for the potentials, and is
therefore the more economical initial target.

The next envelope problem is to specify a class of ordinary domains
and prove that, for each compact obstacle $E$, the function
\[
 h_E^*=\left(\sup\{v\in\Cb(V):v\leq0,\ v|_E\leq-1\}\right)^*
\]
belongs to the intended classical psh class, with all complete
extensions in its curve test. The upper regularization can change
values on $E$. One must therefore prove, rather than assume, that
the obstacle condition is retained in the required pointwise or
capacity sense. Maximality away from $E$, support of its equilibrium
measure, and a formula relating that measure to capacity are separate
steps. The finite-max counterexample in the preceding section explains
why a formal supremum is insufficient.

Neither compact regularity alone nor local bounded quasicontinuity
has been used to assert Borel outer regularity, negligible-set
theorems, or Choquet capacitability. Each of these remains
\texttt{OPEN} in the full stated relative setting.

\subsection{An exact sufficient condition for polar non-concentration}

\begin{proposition}[A conditional truncation criterion]
\label{prop:polar-truncation-criterion}
Let $q=n-p\geq1$, let $K_0\Subset V$, and let
$\varphi:V\to[-\infty,0]$ be a Borel function such that
$\varphi_N=\max\{\varphi,-N\}\in\Cb(V)$ for every integer $N\geq1$.
Suppose that one number $A<\infty$ satisfies
\begin{equation}
\label{eq:uniform-polar-weight}
 \sup_{N\geq1}\ sup_{\substack{v\in\Cb(V)\\0\leq v\leq1}}
       \int_{K_0}(-\varphi_N)T_{S,q}(v,\ldots,v)\leq A.
\end{equation}
Then, for every $t>0$,
\[
 \operatorname{Cap}_S(K_0\cap\{\varphi<-t\};V)\leq A/t,
 \qquad
 \operatorname{Cap}_S(K_0\cap\{\varphi=-\infty\};V)=0.
\]
Every mixed relative measure of bounded potentials whose oscillations
are bounded on $V$ gives zero mass to the latter set.
\end{proposition}

\begin{proof}
Choose an integer $N\geq t$. On $\{\varphi<-t\}$ one has
$-\varphi_N\geq t$. For each admissible $v$, Markov's inequality
and \eqref{eq:uniform-polar-weight} give
\[
 t\int_{K_0\cap\{\varphi<-t\}}T_{S,q}(v,\ldots,v)\leq A.
\]
Take the supremum over $v$. The pole set is contained in every
sublevel set, so its capacity is at most $A/t$ for every $t>0$.
Let $t\to\infty$. The mixed-measure assertion follows from
\eqref{eq:mixed-capacity-domination}, after shifting and normalizing
the bounded potentials.
\end{proof}

\noindent\textbf{Status.} \texttt{CONDITIONAL}. The implication is
proved, but the uniform weighted bound is not. The bounded CLN estimate
alone gives a bound growing with $N$ and cannot supply
\eqref{eq:uniform-polar-weight}. A plausible next analytic problem is
to derive such a bound from integrability of $\varphi$ against a
specified finite collection of coefficient controls of $S$, together
with an appropriate local barrier. The exact sufficient assumptions
must be proved. In the integration-current counterexample supported
on a polar divisor, $-\varphi_N=N$ on the relative mass; the left
side grows linearly. Thus an unrestricted assertion for all $S$ is
false, not merely unproved.

\subsection{From capacity-zero conclusions to pointwise order}

Section~\ref{sec:capacity-domination} proves domination and uniqueness
modulo zero capacity. The remaining pointwise problem is to determine
for which pairs $(S,V)$ the capacity detects every nonempty strict
psh gap in Corollary~\ref{cor:conditional-pointwise-domination}.
Positivity on nonempty ordinary open sets is not automatically the
same assertion: differences of discontinuous psh functions need not
have ordinary open strict-sign sets. An appropriate relative notion
may be needed if $S$ is carried by a proper analytic subset.

The local stability theorem in Section~\ref{sec:capacity-stability}
now handles nonmonotone sequences that converge in the specified local
capacity. It does not prove precompactness of arbitrary bounded psh
families, nor replace the domain quantifier by a single larger-domain
capacity. Those would require additional theorems.

\subsection{Dirichlet problems and compatibility}

A local Dirichlet theorem must begin by choosing the domain class and
the boundary on which data are prescribed. Ordinary topological,
Shilov, and projective model boundaries serve different purposes.
For a proposed domain class, the next concrete tasks are to construct
barriers, prove the required envelope property, identify measures
controlled by the relative capacity, and establish pointwise
uniqueness under a verified detection condition. General existence
for arbitrary positive Radon data is not assumed.

There is also an elementary obstruction to a domain-free exact
Monge--Amp\`ere existence statement. On a proper boundaryless space,
the positive top product has
\[
 \int_X T_{S,q}(u,\ldots,u)
   =\left\langle D\bigl(uT_{S,q-1}(u,\ldots,u)\bigr),1\right\rangle
   =0
\]
for $q\geq1$, since $1$ is a compactly supported smooth test and
$D1=0$. Thus no nonzero positive measure is such an exact product
there. A curvature class or a suitable local boundary geometry is
substantive additional data.

Ordinary-current compatibility on the already identified continuous
PL/model overlap is retained. Extending identities to all
$\delta$-tests requires its own test-space and multiplication theorem.
The full GK algebra includes objects outside the positive closed
diagonal setting. A2's same-model geometric adelic nefness theorem is
already established and is not included among the open problems.

\subsection{Failed routes retained for continuation}

\begin{enumerate}
\item \texttt{FAILED\_ATTEMPT}: replace fixed-continuous-multiplier
convergence by bounded-measurable-multiplier convergence using only
uniform masses. The branch-well example disproves the resulting
varying-background assertion.
\item \texttt{FAILED\_ATTEMPT}: exchange two discontinuous limits in
the smooth Green formula. The proved replacement is the ordered
one-parameter trace argument; the order of limits is explicit.
\item \texttt{FAILED\_ATTEMPT}: prove weighted monotone continuity
using a capacity convergence theorem whose proof already invokes
weighted monotone continuity. The proof in Section~\ref{sec:monotone}
does not use capacity or the later energy measures.
\item \texttt{FAILED\_ATTEMPT}: use a globally strictly positive
Euclidean trace on every analytic domain. The replacement is a
polynomial norm-box collar and a trace chosen for each finite test
presentation.
\item \texttt{FAILED\_ATTEMPT}: represent every symmetric middle-degree
test by decomposable positive squares. The proof instead uses R's
positive coefficient-matrix cone and respects its distinction from
strong positivity.
\item \texttt{FAILED\_ATTEMPT}: pass compact-support energy identities
through pointwise local approximations or multiply a difference of
psh functions by an arbitrary smooth partition. The latter need not
remain a difference of psh functions. The corrected proof first
regularizes on a prescribed compact box and then uses a second
collar that leaves the original difference unchanged.
\item \texttt{FAILED\_ATTEMPT}: infer outer regularity of a supremum
of measures from the individual Radon properties. The Dirac-family
example shows the exact failure; weak-closure domination is the
remaining analytic statement.
\item \texttt{FAILED\_ATTEMPT}: treat $\{u>v\}$ as an ordinary open
set or infer ambient pluripolar non-concentration for every $S$.
The explicit examples in Section~\ref{sec:counterexamples} rule out
these assertions.
\end{enumerate}

\subsection{Recommended order for the next research session}

First attack fixed-domain weak-closure domination, keeping track of
where an envelope/contact statement is actually needed. In parallel,
test the exact uniform truncation estimate
\eqref{eq:uniform-polar-weight} under an explicit $S$-integrability
hypothesis. Then establish a domain and background criterion that
detects strict psh gaps, so that the proved comparison and capacity-zero
uniqueness become pointwise statements. Only after those steps should
an obstacle-envelope or Dirichlet existence theorem be attempted.
The archived main proof and inputs should be reused; no general
re-audit of R, B, or A2 is part of this plan.

% END source: base_r3/base_r2/base_r1/sections/frontiers.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/upper_envelopes.tex
\section{Upper envelopes supplied by the internal projective theory}
\label{sec:r2-upper-envelopes}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
The following assertions use \textup{B}, Theorems C.6 and 4.4, as
accepted inputs. They do not require another proof of envelope
continuity or of the internal contact theorem. The first record did
not use the full force of these inputs in discussing local envelopes.
This section corrects that omission; it is not counted as an
independent new regularization theorem.

\begin{lemma}[A bounded classical competitor lies below the continuous envelope]
\label{lem:r2-envelope-competitor}
Let $F$ be one of the fields allowed in \textup{B}, let $Y/F$ be smooth,
projective and geometrically integral, and let $L$ be ample. If $r$ is
a bounded classical metric on $L$, $f$ is a continuous metric on $L$,
and $r\leq f$, then $r\leq P_L f$ at every point.
\end{lemma}

\begin{proof}
By \textup{B}, Theorem 4.4(2), there are continuous model-semipositive
metrics $r_j\downarrow r$ on the same identified line. Fix
$\epsilon>0$. The open sets
\[
 \{r_j<f+\epsilon\}
\]
increase and cover the compact space $Y^{\an}$. Consequently one of
them is all of $Y^{\an}$. The metric $r_j-\epsilon$ is a continuous
model-semipositive minorant of $f$ and is therefore at most $P_Lf$.
Thus $r\leq P_Lf+\epsilon$. Letting $\epsilon\downarrow0$ proves the
assertion with the actual pointwise representatives. Real translations
preserve continuous model-semipositivity; equivalently they can be
obtained as uniform limits of the permitted rational valuation
translations. No comparison of Monge--Amp\`ere measures is used.
\end{proof}

\begin{proposition}[Envelopes of bounded upper-semicontinuous obstacles]
\label{prop:r2-usc-obstacle}
In the preceding setting, assume $Y^{\an}$ is metrizable, as it is
for the finite $p$-adic fields and the finite unit-Gauss towers used
here. Let $f$ be an upper-semicontinuous metric whose difference from
one continuous metric is bounded. There is a greatest bounded
classical metric $P_L^{\mathrm{b}}f\leq f$. If $f_j\downarrow f$ are
continuous metrics, with a common bounded range relative to a fixed
continuous reference metric, then
\[
 P_L^{\mathrm{b}}f=\lim_j P_L f_j.
\]
In particular, this limit is independent of the continuous majorants.
\end{proposition}

\begin{proof}
Choose a continuous model-semipositive reference metric $a$. A bounded
upper-semicontinuous function on a compact metric space is the
decreasing limit of uniformly bounded continuous functions: for a
compatible metric $d$, the functions
\[
 x\longmapsto\sup_y\bigl((f-a)(y)-j d(x,y)\bigr)
\]
give one such choice. Add $a$ to obtain $f_j$.
By \textup{B}, Theorem C.6, $p_j=P_L f_j$ is continuous
model-semipositive. The sequence decreases. For a sufficiently large
constant $M$, the semipositive metric $a-M$ lies below every $f_j$;
thus $a-M\leq p_j\leq f_j$, and $p=\lim_jp_j$ is bounded and
classical by decreasing closure. Moreover $p\leq f$.

If $r\leq f$ is any bounded classical metric, then $r\leq f_j$ for
every $j$. Lemma~\ref{lem:r2-envelope-competitor} gives $r\leq p_j$;
hence $r\leq p$. Thus $p$ is the greatest permitted minorant, which
also proves independence. The metrizability used here follows, for
these fields, from a countable dense coefficient subfield and
polynomial seminorm inequalities with rational radii on a finite
affine cover. It does not posit metrizability for arbitrary complete
non-Archimedean fields.
\end{proof}

\begin{corollary}[Upper envelopes of bounded projective families]
\label{cor:r2-projective-upper-envelope}
Let $(r_\alpha)_{\alpha\in A}$ be a nonempty family of bounded
classical metrics on $L$, uniformly bounded relative to one continuous
reference metric. Write $h=\sup_\alpha r_\alpha$, with its upper
semicontinuous regularization $h^*$. Then $h^*$ is a bounded classical
metric on $L$.
\end{corollary}

\begin{proof}
The regularization is taken in local frames; changes of frame add a
continuous function and therefore commute with this operation. Put
$p=P_L^{\mathrm b}(h^*)$. Every $r_\alpha$ is a bounded classical
minorant of $h^*$, so $h\leq p$. Since $p$ is upper semicontinuous,
$h^*\leq p$. The defining obstacle inequality gives $p\leq h^*$.
Thus $p=h^*$.
\end{proof}

\begin{theorem}[Local bounded upper-envelope theorem]
\label{thm:r2-local-upper-envelope}
Let $U$ have the fixed analytic hypotheses of the first record. Let
$(u_\alpha)_{\alpha\in A}$ be a nonempty family in $\Cb(U)$ that is
locally uniformly bounded above. Then
\[
 \left(\sup_{\alpha\in A}u_\alpha\right)^*\in\Cb(U).
\]
The regularization is the ordinary upper-semicontinuous regularization
on the actual Berkovich space. The conclusion includes the complete
field-extension and curve tests in the definition of classical psh.
\end{theorem}

\begin{proof}
Work near an arbitrary point. By \textup{B}, Lemmas D.4--D.5 and the
compactification used in Proposition 4.5, work on an ordinary open
of a smooth projective geometrically integral variety over an allowed
finite $p$-adic constant field. Choose the nested compact polynomial
norm boxes $C_1\subset\operatorname{Int}C_{\rm mid}
\subset C_{\rm mid}\subset\operatorname{Int}C_2\subset U$ used
in \textup{B}, Lemma D.2.

Fix one member $u_{\alpha_0}$. It has a finite lower bound on $C_2$.
Choose $M$ with $u_{\alpha_0}\geq-M$ there and replace each
$u_\alpha$ on a neighborhood of $C_2$ by
$v_\alpha=\max\{u_\alpha,-M\}$. These functions have one common
two-sided bound on $C_2$, and
\[
 \sup_\alpha v_\alpha=\sup_\alpha u_\alpha\quad\hbox{on }C_2.
\]
The collar construction in the proof of \textup{B}, Lemma D.2
depends only on these bounds and the chosen boxes. Explicitly, its
pure-FS metric $q$ has a weight $b$ that is zero on $C_1$ and bounded
below by $\delta>0$ on the outer collar. Choose a single translation
$c$ and one integer $m$ with
$0<v_\alpha+c\leq M'$ and $m\delta>M'+1$ for every $\alpha$.
On the same ample line and in the same frame, put
\[
 R_\alpha=\max\{v_\alpha+c,mb\}
       \quad\hbox{on }\operatorname{Int}C_2,
 \qquad R_\alpha=mq\quad\hbox{off }C_{\rm mid}.
\]
The prescriptions agree on the ordinary overlap. Maximum closure and
locality give bounded classical metrics; their differences from $mq$
all lie in $[0,M']$. This argument applies to an arbitrary index set:
the finiteness assumption in the statement of the cited decreasing
family lemma is replaced here by the displayed common bound, and no
simultaneous regularization of this family is being presumed.

Corollary~\ref{cor:r2-projective-upper-envelope} applies to these
metrics. On $\operatorname{Int}C_1$ their weights are
$v_\alpha+c$. Upper regularization is local on ordinary open sets;
subtracting $c$ therefore gives the asserted function near the point.
All metrics in this argument are classical in the original sense,
and restrictions preserve all its field-extension and curve tests.
Classicality is local, so the conclusion holds on $U$.
\end{proof}

\begin{remark}[What this result does not assert]
The theorem does not say that the unregularized supremum is upper
semicontinuous, or that its exceptional set has zero capacity for an
arbitrary background. The latter assertion is false, even when the
family is an increasing sequence of continuous bounded potentials;
see Section~\ref{sec:r2-support-envelopes}.
\end{remark}

% END source: base_r3/base_r2/sections/upper_envelopes.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/weak_closure.tex
\section{Quasi-open sets and the weak closure of bounded BT families}
\label{sec:r2-weak-closure}

The fixed field, original CLD test sheaf, positivity cone, and actual
point-value convention are those of the frozen first-round report.
We use its fixed-background calculus, relative coefficient CLN bounds,
local capacity convergence, and local quasicontinuity as inputs. We do
not assume outer regularity on arbitrary Borel sets or a theorem about
upper envelopes. The capacity domain is fixed throughout each argument.

Fix an ordinary norm-box domain $\Omega$, a symmetric positive closed
background $S$ of bidegree $(p,p)$, and put $q=n-p\geq1$. Write
\[
 C(E)=\operatorname{Cap}_S(E;\Omega)
   =\sup_{\substack{v\in\mathcal C_b(\Omega)\\0\leq v\leq1}}
      \int_E T_{S,q}(v,\ldots,v)
\]
for Borel $E\subset\Omega$. The same arguments apply to the localized
$\operatorname{Cap}_{S,\beta,k}$ capacities of the first round.
Define its outer regularization by
\[
 C^*(E)=\inf_{\substack{G\supset E\\G\text{ ordinary open in }\Omega}}
          C(G).
\]
This is a definition; equality with $C(E)$ is not presumed.

\subsection{Outer regularity on the proved class of quasi-open sets}

\begin{definition}
A Borel set $E\subset\Omega$ is $C$-quasi-open if for every
$\epsilon>0$ there is an ordinary open $G\subset\Omega$ such that
$C(G)<\epsilon$ and $E\cup G$ is ordinary open.
\end{definition}

\begin{proposition}[Quasi-open sets are outer regular]
\label{prop:r2-quasiopen-outer}
Every $C$-quasi-open Borel set satisfies $C^*(E)=C(E)$.
Countable unions and finite intersections of $C$-quasi-open sets
are $C$-quasi-open.

Let $V\Subset\Omega$ be an ordinary open core on which real-valued
functions $a,b$ are $C$-quasicontinuous in the sense proved in the
first round. Then
\[
 V\cap\{a>b\},\qquad V\cap\{|a-b|>\delta\}\quad(\delta>0)
\]
are $C$-quasi-open, and hence outer regular for $C$.
In particular this applies to locally bounded classical psh
functions on the permitted local cores.
\end{proposition}

\begin{proof}
Monotonicity gives $C(E)\leq C^*(E)$. If $E\cup G$ is open and
$C(G)<\epsilon$, countable subadditivity gives
\[
 C^*(E)\leq C(E\cup G)\leq C(E)+\epsilon.
\]
Letting $\epsilon\downarrow0$ proves the first assertion. For a
countable family $E_i$, choose exceptional opens $G_i$ with
$C(G_i)<\epsilon2^{-i-1}$ and take $G=\bigcup_iG_i$. Then
$\bigcup_iE_i\cup G$ is open. For a finite intersection, choose
such a common $G$ and use
\[
 \left(\bigcap_{i=1}^N E_i\right)\cup G
     =\bigcap_{i=1}^N(E_i\cup G).
\]

For the final assertion, choose an ordinary open $G$ of capacity
less than $\epsilon$ such that both $a$ and $b$ are continuous on
$V\setminus G$. Their strict inequality set there is relatively
open in $V\setminus G$. Thus there is an ordinary open $O\subset V$
such that
\[
 (V\cap\{a>b\})\setminus G=O\setminus G.
\]
Consequently $(V\cap\{a>b\})\cup G=O\cup G$ is ordinary open.
The proof for $|a-b|>\delta$ is identical. Notice that the argument
does not declare the original strict inequality sets ordinarily open.
\end{proof}

\begin{corollary}[Outer-small exceptional sets]
\label{cor:r2-outer-zero}
If a local strict inequality set between two bounded classical psh
functions has zero capacity, it has ordinary open neighborhoods
with arbitrarily small capacity. Thus the exceptional sets in the
first-round domination and uniqueness statements modulo capacity
zero may also be taken to have outer capacity zero.
\end{corollary}

\begin{proof}
Apply Proposition~\ref{prop:r2-quasiopen-outer} to each strict
inequality set. Equality of two functions fails on the union of
the two strict inequality sets; use countable subadditivity.
\end{proof}

\subsection{What weak limits already preserve}

\begin{lemma}[Open-set and quasi-open-set bounds survive weak limits]
\label{lem:r2-open-limit}
Let $\nu_j$ be positive Radon measures on a relatively compact
ordinary region $W\subset\Omega$, converging locally weakly to
$\nu$. Suppose that for every ordinary open $G\subset W$,
\[
 \nu_j(G)\leq A C(G)
\]
with one finite constant $A$. Then
\[
 \nu(G)\leq A C(G)
\]
for such opens, and
\[
 \nu(E)\leq A C(E)
\]
for every $C$-quasi-open $E\subset W$ whose outer neighborhoods
are taken inside $W$.

The analogous statement holds for total variations of signed
Radon measures converging locally weakly and satisfying the same
uniform open-set bounds.
\end{lemma}

\begin{proof}
For positive measures, the weak-convergence inequality on opens
gives $\nu(G)\leq\liminf_j\nu_j(G)$. For signed measures, use
\[
 |\nu|(G)
  =\sup\left\{\left|\int f\,d\nu\right|:
       f\in C_c(G),\ |f|\leq1\right\}
  \leq\liminf_j|\nu_j|(G).
\]
The local form follows by exhaustion with compactly supported
continuous tests. For a Borel $E\subset W$, Radon outer regularity
of the individual limiting measure gives
\[
 |\nu|(E)\leq A\inf_{G\supset E} C(G)=A C^*(E),
\]
with the neighborhood infimum restricted to $W$.
Proposition~\ref{prop:r2-quasiopen-outer} identifies this with
$A C(E)$ for the stated quasi-open sets.
\end{proof}

\begin{corollary}[Partial weak-closure domination]
\label{cor:r2-quasiopen-domination}
Let $0\leq v_j\leq1$ be classical psh functions on the fixed
domain $\Omega$, with no monotonicity assumption. If
\[
 T_{S,q}(v_j,\ldots,v_j)\longrightarrow\nu
\]
locally weakly, then $\nu(E)\leq C(E)$ on every local
$C$-quasi-open set. In particular this holds on the strict
inequality sets of Proposition~\ref{prop:r2-quasiopen-outer}.
The assertion applies to all members of the local weak closure
of the normalized BT measure family, including limits for which
no potential representation is known.
\end{corollary}

\begin{proof}
Every measure in the original family satisfies the capacity bound
on all Borel sets by the definition of $C$. Apply
Lemma~\ref{lem:r2-open-limit} locally. Positive smooth cutoffs equal
to one on a prescribed core give finite uniformly bounded measures,
so the same argument applies to the weak closure there.
\end{proof}

\subsection{A multiplier theorem for a capacity-controlled weak closure}

\begin{theorem}[Quasicontinuous multipliers through weak closure]
\label{thm:r2-weak-closure-multipliers}
Let $P_j\to P$ be positive coefficient-Radon original CLD currents
of one bidegree, with convergence on all original compact smooth
tests. Fix a compact test $\eta$. Suppose that on a compact
neighborhood $H$ of its support there is one finite constant $A_\eta$
such that
\begin{equation}
\label{eq:r2-open-coefficient-bound}
 |\mu_{P_j,\eta}|(G)\leq A_\eta C(G)
\end{equation}
for every ordinary open $G$ in a fixed slightly larger region.
Let $a$ be bounded and $C$-quasicontinuous on that region. Then
\[
 \langle aP_j,\eta\rangle\longrightarrow\langle aP,\eta\rangle.
\]
No Borel capacity bound for $P$ is required as an extra hypothesis.

More generally, suppose $a_j,a$ are uniformly bounded,
$C$-quasicontinuous there, and $a_j\to a$ locally in $C$-capacity
on every compact subset of the region. Then
\[
 \langle a_jP_j,\eta\rangle\longrightarrow\langle aP,\eta\rangle.
\]
All multipliers are their actual Borel functions.
\end{theorem}

\begin{proof}
The positive-current coefficient theorem implies weak convergence
of the signed coefficient measures against fixed continuous compact
scalar tests. Their total variations are locally uniformly bounded.
Lemma~\ref{lem:r2-open-limit} passes
\eqref{eq:r2-open-coefficient-bound} to $\mu_{P,\eta}$ on opens.

Let $M$ bound $|a|$ on $H$. For $\epsilon>0$, choose an ordinary
open $G$ with $C(G)<\epsilon$ such that $a$ is continuous on
$H\setminus G$. By Tietze extension on the compact Hausdorff
neighborhood $H$, followed by a cutoff equal to one near
$\operatorname{supp}\eta$, there is a continuous compact scalar
$a_\epsilon$, with $|a_\epsilon|\leq M$, equal to $a$ on
$\operatorname{supp}\eta\setminus G$. Therefore
\[
 \left|\int(a-a_\epsilon)\,d\mu_{P_j,\eta}\right|
 +\left|\int(a-a_\epsilon)\,d\mu_{P,\eta}\right|
 \leq4M A_\eta\epsilon.
\]
The integrals with $a_\epsilon$ converge by the fixed-continuous
multiplier theorem. Let $\epsilon\downarrow0$.

For the varying-weight assertion, choose an ordinary open core
$V$ with $\operatorname{supp}\eta\subset V\Subset H$. The Borel
set
\[
 E_{j,\delta}=V\cap\{|a_j-a|>\delta\}
\]
is $C$-quasi-open by Proposition~\ref{prop:r2-quasiopen-outer}.
Individual Radon outer regularity and the open-set bound imply
\[
 |\mu_{P_j,\eta}|(E_{j,\delta})
       \leq A_\eta C(E_{j,\delta}).
\]
The latter capacity tends to zero, because it is bounded by the
capacity of the corresponding set on the larger compact $H$.
If $N$ bounds the coefficient variations on
$\operatorname{supp}\eta$, then
\[
 \left|\int(a_j-a)\,d\mu_{P_j,\eta}\right|
 \leq N\delta+2M A_\eta C(E_{j,\delta}).
\]
First let $j\to\infty$ and then $\delta\downarrow0$. Combine
this estimate with the fixed-weight assertion just proved.
\end{proof}

\begin{corollary}[Fixed additional BT slots on a weak closure]
\label{cor:r2-fixed-slots-closure}
On a fixed local norm-box domain, let
\[
 P_j=T_{S,r}(v_{1,j},\ldots,v_{r,j}),\qquad0\leq v_{i,j}\leq1,
\]
and suppose $P_j\to P$ on original compact smooth tests. The
generating potentials need not converge and need not be monotone.
For every fixed bounded classical psh function $u$,
\[
 uP_j\longrightarrow uP,
 \qquad D(uP_j)\longrightarrow D(uP)
\]
on all original compact tests, whenever the indicated degrees
are allowed. More generally, any fixed tuple of additional bounded
classical psh differentiated slots, and a fixed bounded classical
psh weight, passes through this weak convergence.
\end{corollary}

\begin{proof}
The coefficient-capacity domination proved in the first round
gives \eqref{eq:r2-open-coefficient-bound}, uniformly in $j$, on
each compact original test stage. More precisely, first divide a
given compact test into finitely many moment presentations and
choose the local norm boxes required by that domination statement.
The capacities used for these coefficient estimates have those
fixed local domains. A domain may depend on the test presentation,
but never on $j$; this localization makes no comparison assertion
between capacities on different domains. The restrictions of all
$v_{i,j}$ remain in $[0,1]$ there. The given $u$ is locally
quasicontinuous by the first-round theorem. Apply
Theorem~\ref{thm:r2-weak-closure-multipliers}, and then use
continuity of distributional $D$.

For iteration, after one additional slot the approximating currents
remain bounded mixed BT currents with the same original background
$S$. The additional fixed potential has a finite bound on the
slightly larger local domain. Positive rescaling and addition of
constants put it into a normalized bounded range; the first-round
mixed coefficient-capacity estimate therefore remains uniform.
The limit is the corresponding positive closed ordered product
with background $P$. Repeat one slot at a time and finally apply
the multiplier theorem to the fixed weight.

This is a capacity-controlled varying-background theorem, not an
assertion for arbitrary weakly convergent positive backgrounds.
The branch-well/Dirac counterexample in the first round lacks the
uniform capacity control required here.
\end{proof}

\subsection{The unresolved compact-set step}

The preceding arguments do not establish
\[
 \nu(K)\leq C(K)
 \qquad(K\text{ compact})
\]
for an arbitrary weak limit of normalized BT measures. A compact
set is ordinarily closed; it need not be quasi-open. The theorem
above does not turn its characteristic function into a
quasicontinuous multiplier. The remaining lemma is precisely
weak-closure domination on compact sets from the first-round
reduction, with the original capacity domain unchanged.

Two proposed stronger shortcuts fail. First, individual
quasicontinuity gives a small exceptional open for each potential;
even a common exceptional open for a countable family supplies
continuity of each member, not equicontinuity or compactness of the
normalized family. Second, replacing a sequence by finite or tail
maxima does not order its curvature measures. Moreover the tail
supremum need not be upper semicontinuous. No upper-envelope or
contact theorem is assumed in this section.

The new proved assertions concern quasi-open sets and weak-closure
calculus under capacity bounds. Full compact continuity, full
Borel outer regularity, and general weak-closure domination for an
arbitrary background remain \texttt{OPEN}. For the ambient background
on the later exhaustion domains, the extremal arguments in
Sections~\ref{sec:r2-ambient-extremals} and
\ref{sec:r2-polar-converse-choquet} prove the corresponding compact
and relatively compact Borel assertions. They are additional
arguments, not consequences of weak convergence alone.

% END source: base_r3/base_r2/sections/weak_closure.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/capacitary_closure.tex
\section{A bounded capacitary closure and its BT calculus}
\label{sec:r2-capacitary-closure}

This section introduces an auxiliary class for a specific purpose:
an increasing bounded sequence of classical psh functions has an
actual Borel limit which need not be upper semicontinuous, and its
ambient upper regularization can change values on a set detected by
$S$. We construct the calculus on limits in outer relative capacity
without making either of those replacements. The original class
$\Cb$ and all of its pointwise theorems remain as in the first record.
The new class is relative to the fixed current $S$.

Put $q=n-p\geq1$. On an ordinary norm-box domain $\Omega$, write
$C_\Omega(E)=\operatorname{Cap}_{S}(E;\Omega)$ and let
$C_\Omega^*$ denote its outer regularization, defined in
Section~\ref{sec:r2-weak-closure}. All sets occurring as function
level sets below are Borel. The outer capacity is also defined on
arbitrary subsets.

\begin{lemma}[Elementary outer-capacity facts]
\label{lem:r2-outer-facts}
The function $C_\Omega^*$ is monotone and countably subadditive,
and $C_\Omega^*(G)=C_\Omega(G)$ on ordinary open sets. If
$\Omega'\subset\Omega$ is ordinary open and $E\subset\Omega'$, then
\[
 C_\Omega^*(E)\leq C_{\Omega'}^*(E).
\]
For decreasing compact sets $K_j\downarrow K$ contained in
$\Omega$, one has
\[
 C_\Omega^*(K_j)\downarrow C_\Omega^*(K).
\]
The last assertion is about the outer regularization; it does not
establish $C_\Omega^*(K)=C_\Omega(K)$.
\end{lemma}

\begin{proof}
For countable subadditivity choose open covers whose capacities
approximate the respective infima, with errors summing to
$\epsilon$, and take their union. Monotonicity and the assertion
on opens follow directly from the definition. For domain
monotonicity, an open $G\subset\Omega'$ is also open in $\Omega$,
and every normalized potential on $\Omega$ restricts to an
admissible potential on $\Omega'$. Thus
$C_\Omega(G)\leq C_{\Omega'}(G)$, and taking infima proves the
claim. Finally, every open neighborhood $G$ of $K$ contains all
sufficiently large $K_j$, by compactness of $K_1$. Hence
$\lim_j C_\Omega^*(K_j)\leq C_\Omega(G)$. Take the infimum
over $G$ and use monotonicity for the reverse inequality.
\end{proof}

\begin{definition}[Local outer-capacity convergence]
\label{def:r2-outer-convergence}
Real Borel functions $a_j$ converge to $a$ locally in outer
$S$-capacity if, for every norm-box domain $\Omega$ whose compact
closure lies in an allowed ordinary algebraic neighborhood, every
compact $H\Subset\Omega$, and every $\delta>0$,
\[
 C_\Omega^*(H\cap\{|a_j-a|>\delta\})\longrightarrow0.
\]
The Cauchy condition is the corresponding limit with $a_j-a_l$,
as both indices tend to infinity. We quantify over all these
domains, not just one larger domain. Two functions are locally
equal quasi-everywhere if the outer capacity of their disagreement
set on each such compact $H$ is zero.
\end{definition}

\begin{lemma}[Quasicontinuity of capacity limits]
\label{lem:r2-qc-limit}
Let $a_j$ be real, locally bounded, locally $S$-quasicontinuous
functions, and let $a$ be a locally bounded Borel function. If
$a_j\to a$ locally in outer $S$-capacity, then $a$ is locally
$S$-quasicontinuous. The conclusion requires neither compact
outer regularity of $C_\Omega$ nor nondegeneracy of the capacity.
\end{lemma}

\begin{proof}
Fix $H\Subset\Omega$ and a slightly larger compact neighborhood
$H'$ in $\Omega$. On $H'$ choose a subsequence $a_{j_l}$ such that
\[
 C_\Omega^*(H'\cap\{|a_{j_l}-a|>2^{-l}\})
       <\epsilon2^{-l-3}.
\]
Choose ordinary open covers $G_l$ of these sets with
$C_\Omega(G_l)<\epsilon2^{-l-2}$. Local quasicontinuity and a
finite cover of $H'$ give ordinary open sets $F_l$, with the same
capacity bound, such that $a_{j_l}$ is continuous on
$H'\setminus F_l$. Domain monotonicity allows all these estimates
to be measured in the fixed domain $\Omega$.
The union $G=\bigcup_l(G_l\cup F_l)$ has capacity less than
$\epsilon$. On $H'\setminus G$ the functions $a_{j_l}$ are
continuous and converge uniformly to $a$. Their uniform limit is
continuous there. Restriction to $H$ proves the assertion.
\end{proof}

\begin{definition}[Bounded capacitary closure]
\label{def:r2-capacitary-class}
For an ordinary algebraic norm-box domain $W$ and $M<\infty$,
let $\mathcal A^M_S(W)$ consist of the Borel functions $a$ with
$|a|\leq M$ which are limits, locally in outer $S$-capacity on
$W$, of sequences
\[
 a_j\in\Cb(W),\qquad |a_j|\leq M\quad\hbox{on }W.
\]
We retain actual Borel representatives but may also take the
quotient by local quasi-everywhere equality. Let
$\mathcal A_{S,b}(U)$ be the local class: near each point a member
restricts to some $\mathcal A^M_S(W)$ as above. The particular
bound and approximating domain may depend on the neighborhood.
\end{definition}

\begin{proposition}[Existence of capacitary limits]
\label{prop:r2-completion}
On a norm-box domain $W$ over the fixed finite $p$-adic field,
every sequence $a_j\in\Cb(W)$ with $|a_j|\leq M$ which is
locally Cauchy in $S$-capacity has a Borel limit
$a\in\mathcal A^M_S(W)$ locally in outer $S$-capacity. It is
unique up to local quasi-everywhere equality.

The quotient of $\mathcal A^M_S(W)$ is complete for this local
capacity convergence. Addition, nonnegative scalar multiplication,
and finite maxima preserve the local class, with the corresponding
bounds. Every member is locally quasicontinuous.
\end{proposition}

\begin{proof}
The space $W$ has a countable basis of norm-box neighborhoods:
use a countable dense coefficient subfield on an affine algebraic
chart and polynomial seminorm inequalities with rational
thresholds. Choose countably many pairs
\[
 V_l\Subset\Omega_l\Subset W
\]
of ordinary opens, with $\Omega_l$ permitted norm boxes, cofinal
in the following sense: whenever $H\Subset\Omega\subset W$ is
a permitted compact/domain pair, finitely many $V_l$ cover $H$
and their corresponding $\Omega_l$ lie in $\Omega$.
The $V_l$ can be chosen to have compact closure in $\Omega_l$.

Each gap $V_l\cap\{|a_j-a_k|>\delta\}$ is quasi-open for
$C_{\Omega_l}$ by the proved quasicontinuity of $\Cb$ and
Proposition~\ref{prop:r2-quasiopen-outer}. Its outer capacity is
therefore its capacity. The assumed Cauchy condition on
$\overline V_l$ lets us choose a subsequence $a_{j_m}$ such that,
for all $l\leq m$,
\[
 C_{\Omega_l}^*(V_l\cap
       \{|a_{j_{m+1}}-a_{j_m}|>2^{-m}\})<2^{-m}.
\]
For each pair $(l,m)$ with $l\leq m$, choose an ordinary open
cover $G_{l,m}\subset\Omega_l$ of that set with capacity less
than $2^{1-m}$. For fixed $l$ the tails
$G_l^{(N)}=\bigcup_{m\geq N}G_{l,m}$, $N\geq l$, satisfy
\[
 C_{\Omega_l}(G_l^{(N)})\leq 2^{2-N}.
\]
Outside their intersection the subsequence is eventually pointwise
Cauchy on $V_l$, since $\sum_m2^{-m}<\infty$.
Define the Borel function $a=\limsup_m a_{j_m}$ at every point;
it lies in $[-M,M]$. Outside $G_l^{(N)}$,
\[
 |a-a_{j_N}|\leq 2^{1-N}\quad\hbox{on }V_l.
\]
This proves convergence of the subsequence in outer capacity on
each $V_l$. The Cauchy condition and the triangle inequality for
level sets extend convergence to the entire sequence. Cofinality,
finite subadditivity and domain monotonicity then give the stated
convergence on every permitted compact/domain pair. Indeed, a
cover inside each $\Omega_l\subset\Omega$ is an ordinary open
cover in $\Omega$ with no greater capacity.

Two limits have disagreement sets covered by the countable union
of their positive level gaps; the level-set triangle inequality
gives zero outer capacity for each such gap. This proves
uniqueness. Lemma~\ref{lem:r2-qc-limit} gives quasicontinuity.

For completeness, start with a Cauchy sequence in
$\mathcal A^M_S(W)$. For its $j$th member choose a defining
classical approximant within $2^{-j}$ in capacity, at threshold
$2^{-j}$, on the first $j$ compact/domain pairs of a cofinal
enumeration. Such a choice is possible by its definition, and
every chosen approximant has the same bound $M$. The resulting
classical sequence is Cauchy on every pair. The preceding
construction gives its limit, which is also the limit of the
original sequence by the level-set triangle inequality.
Finally, sums, nonnegative multiples and maxima of defining
classical sequences converge to the corresponding pointwise
operations: use the inequalities
\[
 |(a+b)-(c+d)|\leq|a-c|+|b-d|,
 \qquad |\max(a,b)-\max(c,d)|\leq\max(|a-c|,|b-d|).
\]
The classical class has these closure properties. Localizing
proves the assertions for $\mathcal A_{S,b}$.
\end{proof}

\begin{theorem}[BT calculus on the bounded capacitary closure]
\label{thm:r2-completed-calculus}
Let $S$ be the arbitrary fixed original-positive separately closed
background. For $0\leq k\leq q$ and
$a_1,\ldots,a_k\in\mathcal A_{S,b}(U)$, the recursion
\[
 \mathcal T_{S,0}=S,\qquad
 \mathcal T_{S,k}(a_1,\ldots,a_k)
  =D\bigl(a_1\mathcal T_{S,k-1}(a_2,\ldots,a_k)\bigr)
\]
defines positive, symmetric, separately closed coefficient-Radon
currents. It agrees with the original BT product on $\Cb$,
is independent of defining approximants and of local
quasi-everywhere changes of representatives, and is symmetric in
all differentiated slots. It is additive and positively homogeneous
in those slots and respects restriction and gluing.

If $a_{i,j}\in\Cb(U)$, $0\leq i\leq k$, have actual uniform
bounds on compact subsets and converge locally in outer
$S$-capacity to locally bounded Borel functions $a_i$, then
\begin{equation}
\label{eq:r2-completed-weighted-limit}
 a_{0,j}T_{S,k}(a_{1,j},\ldots,a_{k,j})
 \longrightarrow
 a_0\mathcal T_{S,k}(a_1,\ldots,a_k)
\end{equation}
on all original compact smooth complementary tests. The same
stability holds for sequences in the capacitary closure if, locally,
their defining approximants admit one common bound.
\end{theorem}

\begin{proof}
Work on a compact test stage, localizing into finitely many smaller
norm boxes $\Omega$ where the test's unit moment coordinates are
defined and bounded. R1 finite coefficient domination gives, for
every bounded classical approximating product $P_j$ and each fixed
test $\eta$ on this stage,
\begin{equation}
\label{eq:r2-approx-coefficient-cap}
 |\mu_{P_j,\eta}|(E)\leq C_\eta C_\Omega(E)
       \quad(E\text{ Borel}),
\end{equation}
with a constant depending only on the common bounds and the test
stage. In degree zero this holds for $P_j=S$ itself, by filling
the complementary degree with the bounded quadratic moment
potential. If positive currents $P_j$ converge to $P$, their
coefficient measures converge weakly against compact continuous
scalars by \textup{R}. Lower semicontinuity of total variation on
opens and individual Radon outer regularity imply
\begin{equation}
\label{eq:r2-limit-coefficient-outer}
 |\mu_{P,\eta}|(E)\leq C_\eta C_\Omega^*(E)
       \quad(E\text{ Borel}).
\end{equation}
This is precisely the weak-closure argument of
Lemma~\ref{lem:r2-open-limit}; it does not replace $C_\Omega^*$
by $C_\Omega$ on compact sets.

We induct on $k$. The case $k=0$ follows from Borel multiplication
of $S$, the coefficient bound, and the multiplier theorem
\ref{thm:r2-weak-closure-multipliers}. Limits $a_i$ are
quasicontinuous by Lemma~\ref{lem:r2-qc-limit}. Suppose the
assertions through degree $k-1$ are known. Weighted convergence
there, followed by distributional $D$, gives convergence of the
degree-$k$ products to the displayed recursive expression. A weak
limit of positive symmetric closed currents retains these
properties. The coefficient-Radon theorem of \textup{R} applies
to it, and \eqref{eq:r2-limit-coefficient-outer} supplies its
outer-capacity control. Apply the multiplier theorem once more
to get \eqref{eq:r2-completed-weighted-limit} in degree $k$.
Its varying-weight proof also follows directly from
\[
 \left|\int(a_{0,j}-a_0)\,d\mu_{P_j,\eta}\right|
 \leq\delta A+2M C_\eta
      C_\Omega^*(H\cap\{|a_{0,j}-a_0|>\delta\}),
\]
and continuous extension of $a_0$ off an ordinary exceptional
open of arbitrarily small capacity. Here $A$ bounds coefficient
variation on the compact test support and $M$ bounds the weights.

The recursion uses the actual Borel values of $a_i$ and therefore
does not contain any approximation choice. On overlaps the same
recursion gives the same current, proving gluing. Formula
\eqref{eq:r2-limit-coefficient-outer} shows that changing a slot
on a locally outer-capacity-zero set leaves every relevant Borel
product unchanged; induction proves representative independence.
Symmetry, addition and positive homogeneity follow by taking
limits of the corresponding identities for classical defining
sequences, with simultaneous finite-slot approximation.

For sequences already in the closure, their coefficient measures
obey \eqref{eq:r2-limit-coefficient-outer}, with uniform constants
when the defining bounds are uniform. The same multiplier
argument and induction apply. All localizations are over finitely
many actual original test stages, so the conclusion holds for
every original compact complementary test, including tests not
coming from one initially chosen moment map.
\end{proof}

\paragraph{Status and role.}
The analytic extension theorem is recorded as \texttt{NEW\_PROVED},
relative to the R1 fixed-background theory and the R2 weak-closure
multiplier lemma. The outer-capacity and completion lemmas are
\texttt{DIRECT\_COROLLARY} measure-capacity arguments with proofs
included. The construction is useful even when $C_\Omega$ is
degenerate. It does not assert that every member of
$\mathcal A_{S,b}$ has a classical psh representative on the
ambient space, or that outer capacity equals the original Borel
capacity. In particular it does not solve the unresolved compact
weak-closure domination problem by definition. For the ambient
background alone, Corollary~\ref{cor:r2-ambient-classical-representative}
later identifies each equivalence class with a unique classical
representative; the arbitrary-background distinction remains.

% END source: base_r3/base_r2/sections/capacitary_closure.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/envelope_capacity_attempt.tex
\section{Increasing sequences and relative Aubin--Mabuchi energy}
\label{sec:r2-increasing-energy}

This section uses the frozen first-round calculus and its common-collar
Green identity. Let $S$ be any allowed positive closed original current
of bidegree $(p,p)$, put $q=n-p\geq1$, and fix one ordinary polynomial
norm-box domain $\Omega$. The capacity is always
$C(E)=\operatorname{Cap}_S(E;\Omega)$, with this same domain.

Choose a bounded smooth classical psh function $b$ on $\Omega$ and a compact
$L\Subset\Omega$, contained in a strictly smaller norm box. Let
$\mathcal P_b$ denote the locally bounded classical psh functions $u$
which equal $b$ on $\Omega\setminus L$. All families below have a common
bound on a fixed compact neighborhood of $L$, hence on all of $\Omega$
because of the common collar. Such families arise from
the common-collar construction; its maximum operation also preserves
increasing sequences.

\subsection{A compactly supported relative energy}

\begin{definition}
For $u\in\mathcal P_b$, define
\[
 \mathcal E_b(u)=\frac1{q+1}\sum_{i=0}^q
       \int_\Omega (u-b)(Du)^i\wedge(Db)^{q-i}\wedge S.
\]
Every integral is a Borel integral supported in $L$. Products use the
proved fixed-background calculus; in particular this definition does
not require a smooth background or a strict psh exhaustion.
\end{definition}

\begin{theorem}[Relative energy difference formula]
\label{thm:r2-energy-difference}
For $u,v\in\mathcal P_b$,
\begin{equation}
\label{eq:r2-energy-difference}
 \mathcal E_b(v)-\mathcal E_b(u)
   =\frac1{q+1}\sum_{i=0}^q
      \int_\Omega(v-u)(Dv)^i\wedge(Du)^{q-i}\wedge S.
\end{equation}
Consequently $u\leq v$ implies $\mathcal E_b(u)\leq\mathcal E_b(v)$,
and in that case
\begin{equation}
\label{eq:r2-lower-energy-control}
 \int_\Omega(v-u)(Du)^q\wedge S
     \leq(q+1)\bigl(\mathcal E_b(v)-\mathcal E_b(u)\bigr).
\end{equation}
On a uniformly bounded subset of $\mathcal P_b$, the energies are
uniformly bounded in absolute value.
\end{theorem}

\begin{proof}
Put $g=v-u$ and $u_t=(1-t)u+tv$, for $0\leq t\leq1$. All these
functions have the same collar. Multilinearity makes
$F(t)=\mathcal E_b(u_t)$ a finite polynomial in $t$; differentiating it
is therefore an algebraic operation on finitely many finite integrals.
Write $\alpha_t=Du_t$ and $\beta=Db$. The derivative is
\begin{align*}
 (q+1)F'(t)
 &=\sum_{i=0}^q\int g\,\alpha_t^i\beta^{q-i}S\\
 &\quad+\sum_{i=1}^q i\int(u_t-b)Dg\,
                                  \alpha_t^{i-1}\beta^{q-i}S.
\end{align*}
In each term of the second sum, the background
$\alpha_t^{i-1}\beta^{q-i}S$ is positive closed of bidegree
$(n-1,n-1)$, and both differences have compact support in the common
norm box. The first-round Green identity gives
\[
 \int(u_t-b)Dg\,\alpha_t^{i-1}\beta^{q-i}S
  =\int g(\alpha_t-\beta)\alpha_t^{i-1}\beta^{q-i}S.
\]
The resulting sums telescope, leaving
\[
 F'(t)=\int g\,\alpha_t^qS.
\]
Expand $\alpha_t^q=((1-t)Du+tDv)^q$ and integrate from zero to one.
The identity
\[
 \binom qi\int_0^1t^i(1-t)^{q-i}\,dt=\frac1{q+1}
\]
proves \eqref{eq:r2-energy-difference}.

If $g\geq0$, every summand is nonnegative. Retaining the term $i=0$
gives \eqref{eq:r2-lower-energy-control}. Finally, the relative CLN
estimate bounds the masses of all mixed products on $L$ uniformly
when the potentials have one common bound. Multiplying by the bounded
function $u-b$ gives the asserted energy bound.
\end{proof}

\begin{lemma}[The pairwise capacity bound]
\label{lem:r2-pair-capacity}
On a uniformly bounded subset of $\mathcal P_b$, there is a constant
$A$ such that $u\leq v$ implies, for every $\delta>0$,
\begin{equation}
\label{eq:r2-pair-capacity}
 C(\{v-u>\delta\})
 \leq\frac A\delta
   \left((q+1)\bigl(\mathcal E_b(v)-\mathcal E_b(u)\bigr)
   \right)^{2^{-q}}.
\end{equation}
The same estimate holds with $C^*$ on the left.
\end{lemma}

\begin{proof}
The proof of the first-round quantitative capacity estimate applies
to any single pair $u\leq v$ with compactly supported difference.
Indeed its iteration uses only $g=v-u\geq0$, Green's identity,
Cauchy--Schwarz, positivity of $D(vQ)$, and uniform CLN bounds.
Before its final limit step it gives
\[
 C(\{v-u>\delta\})
 \leq\frac A\delta\left(\int(v-u)(Du)^qS\right)^{2^{-q}}.
\]
Apply \eqref{eq:r2-lower-energy-control}. The strict inequality set is
quasi-open by the preceding section, so its outer capacity equals its
Borel capacity. This proves the last assertion without assuming compact
outer regularity.
\end{proof}

\subsection{Increasing limits converge to their actual pointwise limit}

\begin{theorem}[Increasing sequences are Cauchy in outer capacity]
\label{thm:r2-increasing-cauchy}
Let $u_j\in\mathcal P_b$ be increasing and uniformly bounded on the
fixed compact neighborhood. Put
\[
 u_\infty(x)=\sup_j u_j(x),\qquad
 e_\infty=\sup_j\mathcal E_b(u_j).
\]
Then $e_\infty$ is finite. For every $\delta>0$,
\begin{equation}
\label{eq:r2-increasing-actual-rate}
 C^*(\{u_\infty-u_j>\delta\})
 \leq\frac A\delta
     \left((q+1)(e_\infty-\mathcal E_b(u_j))\right)^{2^{-q}}.
\end{equation}
The actual pointwise limit $u_\infty$ is locally
$C$-quasicontinuous on each ordinary core. In particular
$u_j\to u_\infty$ in outer capacity on the common difference support,
with the original domain $\Omega$ unchanged.

The theorem does not assert that $u_\infty$ is upper semicontinuous,
classically psh, or equal outside a capacity-zero set to its upper
semicontinuous regularization.
\end{theorem}

\begin{proof}
Theorem~\ref{thm:r2-energy-difference} makes the energies increasing
and uniformly bounded. For $l\geq j$, apply
Lemma~\ref{lem:r2-pair-capacity} to $(u_j,u_l)$. Its right side is
bounded by the right side of \eqref{eq:r2-increasing-actual-rate},
independently of $l$. The strict gap sets increase with $l$ and their
union is $\{u_\infty-u_j>\delta\}$. Continuity from below of the
Borel capacity, already proved in the first round, gives the stated
bound with $C$ on the left. Each strict gap set is quasi-open, and
countable unions of quasi-open sets are quasi-open. Therefore the
union has outer capacity equal to its Borel capacity. This proves
\eqref{eq:r2-increasing-actual-rate} directly, before any regularity
of $u_\infty$ is known.

Here is a direct quasicontinuity argument, with no assumption about
upper envelopes. Fix an ordinary core $V$ and $\epsilon>0$. Choose
a subsequence $u_{j_l}$ so late that the pairwise estimate guarantees
\[
 C^*(\{u_{j_{l+1}}-u_{j_l}>2^{-l}\})
       <\epsilon2^{-l-3}.
\]
This choice can be made successively from the bound involving
$e_\infty-\mathcal E_b(u_{j_l})$. Choose ordinary open covers $G_l$
of these gap sets with $C(G_l)<\epsilon2^{-l-2}$. Also use
first-round quasicontinuity to choose ordinary opens $H_l$ with
$C(H_l)<\epsilon2^{-l-2}$ such that $u_{j_l}$ is continuous on
$V\setminus H_l$. The union
\[
 G=\bigcup_l(G_l\cup H_l)
\]
has capacity less than $\epsilon$. On $V\setminus G$, the continuous
functions $u_{j_l}$ have consecutive increments bounded by $2^{-l}$.
They therefore converge uniformly there. Their pointwise limit is
the specified function $u_\infty$, so this function is continuous
on $V\setminus G$.

The convergence assertion now follows from
\eqref{eq:r2-increasing-actual-rate}.
\end{proof}

\begin{corollary}[Local increasing convergence without a prescribed collar]
\label{cor:r2-local-increasing}
Let $u_j\in\Cb(U)$ increase pointwise and be uniformly bounded on
every compact subset of $U$. Set $u_\infty=\sup_j u_j$, with its
actual Borel values. For every permitted norm-box domain
$\Omega\Subset U$, every compact $H\Subset\Omega$, and every
$\delta>0$,
\[
 \operatorname{Cap}_S^*(H\cap\{u_\infty-u_j>\delta\};\Omega)
       \longrightarrow0.
\]
Thus the convergence is local outer-capacity convergence in the
all-domain sense of Definition~\ref{def:r2-outer-convergence}.
The limit is locally quasicontinuous and belongs to
$\mathcal A_{S,b}(U)$.
\end{corollary}

\begin{proof}
Fix $(H,\Omega)$. Cover $H$ by finitely many ordinary smaller cores
$V_a$ contained in norm-box domains $\Omega_a\Subset\Omega$.
On each such domain use the common-collar construction on the whole
sequence. Its construction requires a common two-sided bound on a
slightly larger box; it does not use the direction of monotonicity.
After one fixed shift $c_a$, it gives a bounded smooth classical psh
barrier $b_a$ and
\[
 v_{j,a}=\max\{u_j+c_a,b_a\},
\]
equal to $u_j+c_a$ on $V_a$, and equal to $b_a$ outside a fixed
compact subset of $\Omega_a$. The maximum preserves increasing
order. Its pointwise supremum equals $u_\infty+c_a$ on $V_a$.
Theorem~\ref{thm:r2-increasing-cauchy} therefore gives
\[
 \operatorname{Cap}_S^*\bigl(
 V_a\cap\{u_\infty-u_j>\delta\};\Omega_a\bigr)
       \longrightarrow0.
\]
Domain monotonicity of outer capacity, with
$\Omega_a\subset\Omega$, and finite subadditivity imply the
asserted estimate on $H$. This repeats the collar construction
inside each requested domain; convergence for one larger capacity
domain has not been substituted for convergence on its smaller
subdomains.

The actual limit is a locally bounded Borel function. Applying the
quasicontinuity lemma for outer-capacity limits gives local
quasicontinuity. On each relatively compact neighborhood the
original sequence itself is uniformly bounded and gives the
defining approximation for $\mathcal A_{S,b}$.
\end{proof}

\begin{corollary}[Weighted increasing limits with actual representatives]
\label{cor:r2-increasing-weighted}
Let $0\leq k\leq q$, and suppose that
$u_{i,j}\in\Cb(U)$, $0\leq i\leq k$, increase with $j$ and
have a common bound on every compact subset for each fixed slot.
Write $u_i=\sup_j u_{i,j}$ with actual Borel values. Then
\[
 u_{0,j}T_{S,k}(u_{1,j},\ldots,u_{k,j})
   \longrightarrow
 u_0\mathcal T_{S,k}(u_1,\ldots,u_k)
\]
on every original compact smooth complementary test form, where
$\mathcal T$ is the calculus on the bounded capacitary closure.
No ambient upper regularization of any $u_i$ is made.
\end{corollary}

\begin{proof}
Apply Corollary~\ref{cor:r2-local-increasing} in each slot, followed
by Theorem~\ref{thm:r2-completed-calculus}. These are successive
uses: the energy estimate and the local increasing convergence were
proved using only the frozen classical-potential calculus, before
applying its extension to the capacitary closure.
\end{proof}

\subsection{The precise upper-envelope question}

Suppose an upper-envelope theorem supplies a classical psh function
\[
 U=(\sup_j u_j)^*
\]
on the same local domain. Because all $u_j$ equal the same smooth
$b$ outside $L$, so does $U$ there. The following formulation separates
the available energy argument from the remaining analytic step.

\begin{proposition}[The energy no-jump criterion]
\label{prop:r2-envelope-energy-criterion}
Under these hypotheses, the following are equivalent:
\begin{enumerate}
\item $\mathcal E_b(U)=\sup_j\mathcal E_b(u_j)$;
\item $u_j\to U$ locally in the relative capacity, on a core
      containing the common compact difference support;
\item $U=u_\infty$ outside a set of relative capacity zero on that core.
\end{enumerate}
The zero set in (3) has outer capacity zero, since both functions
are quasicontinuous on the core.
\end{proposition}

\begin{proof}
For (1)$\Rightarrow$(2), apply
Lemma~\ref{lem:r2-pair-capacity} to $(u_j,U)$; the energy difference
tends to zero. For (2)$\Rightarrow$(1), use
\eqref{eq:r2-energy-difference} with $(u_j,U)$. Every measure on its
right is a mixed top-degree measure of uniformly bounded psh
potentials on the \emph{same} $\Omega$. After translating and
rescaling the potentials, the first-round mixed-capacity bound gives
one uniform bound by a constant times $C(E)$ on every Borel set $E$.
Their masses on $L$ are uniformly bounded by CLN. Split each integral
of $U-u_j$ into the sets where this gap is at most $\delta$ and
greater than $\delta$. The first part is bounded by a uniform mass
times $\delta$, and the second by a uniform bound times
$C(\{U-u_j>\delta\}\cap L)$. First let $j\to\infty$, then
$\delta\downarrow0$. This proves energy convergence without changing
the capacity domain or invoking a stronger convergence on smaller
domains.

Theorem~\ref{thm:r2-increasing-cauchy} gives capacity convergence to
the actual limit $u_\infty$. If (2) holds, subadditivity applied to
the two error-level sets gives $C(\{|U-u_\infty|>\delta\})=0$
for every $\delta>0$ on a slightly smaller core. Taking their
countable union proves (3). Conversely, (3), subadditivity, and the
already proved convergence to $u_\infty$ give (2). Quasicontinuity
of the two functions makes the strict error-level sets quasi-open,
so their outer capacities also vanish.
\end{proof}

For arbitrary backgrounds $S$, the upper-envelope identification
cannot be assumed: an ambient envelope can raise the limit on a
subvariety which carries positive relative capacity. For $S=[1]$,
the criterion isolates a step which is proved later by the ordinary
divisorial barrier in Theorem~\ref{thm:r2-increasing-no-jump}.
Continuity and orthogonality for a continuous obstacle in
\textup{B}, Theorem C.6, and the upper contact statement in
\textup{B}, Theorem 4.4, do not by their statements identify that
energy. The later barrier proof supplies the additional argument;
it was not inserted into the criterion's proof.

% END source: base_r3/base_r2/sections/envelope_capacity_attempt.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/relative_capacitary_envelopes.tex
\section{Relative capacitary upper envelopes}
\label{sec:r2-capacitary-envelopes}

The upper envelope constructed here belongs to the bounded
capacitary closure. Its order is the order modulo relative
outer-capacity-zero sets. The construction does not replace it by
an ambient upper semicontinuous regularization. In particular it
continues to apply when the background is carried by an ambient
pluripolar subvariety.

\subsection{One common collar and one capacity domain}

Use the notation of Section~\ref{sec:r2-increasing-energy}:
$q=n-p\geq1$, a fixed norm box $\Omega$, a bounded smooth
classical psh collar $b$, and the class $\mathcal P_b$ with one
compact difference support $L\Subset\Omega$. Write
$C_\Omega^*=\operatorname{Cap}_S^*(\,\cdot\,;\Omega)$.
For Borel functions define $a\preceq_\Omega d$ to mean
$C_\Omega^*(\{a>d\})=0$.

\begin{theorem}[A capacitary envelope from a maximizing sequence]
\label{thm:r2-one-domain-envelope}
Let $\mathcal F\subset\mathcal P_b$ be nonempty and uniformly
bounded. Let $\mathcal F^\vee$ be its closure under nonempty finite
maxima, and set
\[
 e_* = \sup_{v\in\mathcal F^\vee}\mathcal E_b(v).
\]
There is an increasing sequence $u_j\in\mathcal F^\vee$ such that
$\mathcal E_b(u_j)\uparrow e_*$. Its actual pointwise limit
\[
 H=\sup_j u_j
\]
belongs to $\mathcal A_{S,b}(\Omega)$ and has the following properties:
\begin{enumerate}
\item For each $v\in\mathcal F$, one has $v\preceq_\Omega H$.
\item If a Borel function $h$ satisfies $v\preceq_\Omega h$ for
      every $v\in\mathcal F$, then $H\preceq_\Omega h$.
\item The equivalence class of $H$ modulo $C_\Omega^*$-zero sets
      is independent of the maximizing sequence.
\end{enumerate}
Moreover $H=b$ outside $L$, and the extended BT calculus gives
$\mathcal E_b(H)=e_*$, with the same finite-sum definition of the
energy.
\end{theorem}

\begin{proof}
The energy is uniformly bounded on $\mathcal F^\vee$ by the
common bounds and CLN. Choose $v_j\in\mathcal F^\vee$ with
$\mathcal E_b(v_j)>e_*-2^{-j}$, and put
$u_j=\max(v_1,\ldots,v_j)$. Monotonicity of the energy gives
\[
 e_*-2^{-j}<\mathcal E_b(v_j)\leq\mathcal E_b(u_j)\leq e_*.
\]
The preceding increasing convergence theorem, including its
all-domain local version, applies to $H$ and gives membership in
the capacitary closure.

Fix $v\in\mathcal F$ and put $w_j=\max(u_j,v)$. These functions
are in $\mathcal F^\vee$, and
\[
 0\leq\mathcal E_b(w_j)-\mathcal E_b(u_j)
       \leq e_*-\mathcal E_b(u_j)\longrightarrow0.
\]
For every $\delta>0$, the pairwise energy estimate implies
\[
 C_\Omega^*(\{v>H+\delta\})
 \leq C_\Omega^*(\{w_j-u_j>\delta\})\longrightarrow0.
\]
Taking the countable union over $\delta=1/m$ proves (1).

Every $u_j$ is a finite maximum of members of $\mathcal F$.
If $h$ is as in (2), then $u_j\leq h$ off a finite union of
outer-capacity-zero sets. Taking the countable union of these
exceptional sets over $j$ gives $H\preceq_\Omega h$.
Applying (2) to two constructed upper bounds proves (3).

Finally, $u_j-b$ and $H-b$ vanish outside $L$. Choose a smooth
compact cutoff equal to one near $L$. Apply weighted increasing
BT convergence separately to the two weights $u_j$ and $b$ in
each summand of $\mathcal E_b(u_j)$. The cutoff makes this a
finite collection of original compact smooth tests. Their limits
are precisely the summands of $\mathcal E_b(H)$, proving the
last identity.
\end{proof}

The exceptional set in (1) is allowed to depend on $v$.
For an uncountable family, (1) does not assert that the pointwise
supremum of the entire family is bounded by $H$ outside one
common exceptional set.

\subsection{A countable energy construction for the local order}

The preceding theorem uses one fixed capacity domain. The
following construction enforces the estimates in every smaller
domain, without assuming equivalence of capacities on different
domains.

Let $W$ be an allowed ordinary norm-box domain over the fixed
finite $p$-adic field. Define $a\preceq_{S,\mathrm{loc}}d$ when,
for every permitted $H_0\Subset\Omega\subset W$,
\[
 \operatorname{Cap}_S^*(H_0\cap\{a>d\};\Omega)=0.
\]

\begin{theorem}[Local relative capacitary envelope]
\label{thm:r2-local-capacitary-envelope}
Let $\mathcal F\subset\Cb(W)$ be nonempty and uniformly bounded
above on each compact subset of $W$. There is an increasing sequence
$u_j\in\mathcal F^\vee$ whose actual Borel supremum $H$ belongs
to $\mathcal A_{S,b}(W)$ and is the least upper bound of
$\mathcal F$ for $\preceq_{S,\mathrm{loc}}$:
\[
 v\preceq_{S,\mathrm{loc}}H\quad(v\in\mathcal F),
 \qquad
 \bigl[\ v\preceq_{S,\mathrm{loc}}h\ \text{for all }v\in\mathcal F\ \bigr]
 \Longrightarrow H\preceq_{S,\mathrm{loc}}h.
\]
Its local quasi-everywhere class is independent of all choices.
\end{theorem}

\begin{proof}
Fix $v_0\in\mathcal F$ and first replace the family by
$\{\max(v,v_0):v\in\mathcal F\}$. This family is uniformly
bounded on every compact subset, since $v_0$ supplies a common
lower bound. It has exactly the same local quasi-everywhere upper
bounds as the original family, and every finite maximum of its
members is a finite maximum of original members. We may therefore
assume the two-sided bounds in the rest of the construction.

Choose countably many pairs $V_a\Subset\Omega_a\Subset W$
cofinal for compact/domain pairs, as in
Proposition~\ref{prop:r2-completion}. Thus every
$H_0\Subset\Omega\subset W$ is covered by finitely many $V_a$
whose corresponding $\Omega_a$ lie in $\Omega$.
The common bound of $\mathcal F^\vee$ on a slightly larger box
lets us choose, for each $a$, one fixed collar operation
\[
 P_a(v)=\max\{v+c_a,b_a\}\quad\text{on }\Omega_a.
\]
It equals $v+c_a$ on $V_a$ and the bounded smooth psh function
$b_a$ outside one fixed compact subset of $\Omega_a$.
Its construction is uniform over $\mathcal F^\vee$, and
\[
 P_a(\max(v,w))=\max(P_a(v),P_a(w)).
\]
Write $E_a(v)=\mathcal E_{b_a}(P_a(v))$. The relative energy
theorem gives a finite $B_a\geq1$ with $|E_a(v)|\leq B_a$ on
$\mathcal F^\vee$. Define
\[
 \Phi(v)=\sum_{a=1}^{\infty}2^{-a}
             \frac{E_a(v)+B_a}{2B_a+1}.
\]
The series converges uniformly, takes values in $[0,1]$, and is
monotone for pointwise order.

Put $e=\sup_{v\in\mathcal F^\vee}\Phi(v)$. Choose
$v_j\in\mathcal F^\vee$ with $\Phi(v_j)>e-2^{-j}$ and set
$u_j=\max(v_1,\ldots,v_j)$. Then $\Phi(u_j)\uparrow e$.
The functions $u_j$ are locally uniformly bounded, so their
actual limit $H$ belongs to $\mathcal A_{S,b}(W)$ by
Corollary~\ref{cor:r2-local-increasing}.

Fix $v\in\mathcal F$ and set $w_j=\max(u_j,v)$. For each $a$
the difference $E_a(w_j)-E_a(u_j)$ is nonnegative. Consequently
\[
 0\leq E_a(w_j)-E_a(u_j)
 \leq 2^a(2B_a+1)\bigl(\Phi(w_j)-\Phi(u_j)\bigr)
 \leq 2^a(2B_a+1)(e-\Phi(u_j))\longrightarrow0.
\]
For $\delta>0$, the pairwise energy estimate in $\Omega_a$ gives
\[
 \operatorname{Cap}_S^*\bigl(V_a\cap\{v>H+\delta\};\Omega_a\bigr)
 \leq\operatorname{Cap}_S^*\bigl(
       \{P_a(w_j)-P_a(u_j)>\delta\};\Omega_a\bigr)
       \longrightarrow0.
\]
For a requested pair $H_0\Subset\Omega$, take the finite cofinal
cover of $H_0$. Domain monotonicity and finite subadditivity
transfer the zero bounds to $\Omega$. Taking the countable
union over $\delta=1/m$ proves
$v\preceq_{S,\mathrm{loc}}H$.

The least-upper-bound assertion follows from countability,
exactly as in Theorem~\ref{thm:r2-one-domain-envelope}: each
$u_j$ uses finitely many members of $\mathcal F$, and there
are only countably many $j$. The union of the corresponding
local outer-capacity-zero sets remains locally outer-capacity
zero. Uniqueness in the local ordered quotient follows.
\end{proof}

\begin{corollary}[Countable cofinal families]
If $\mathcal F$ has a pointwise cofinal countable subfamily, its
pointwise supremum is Borel and agrees locally quasi-everywhere
with $H$. This includes every countable family.
\end{corollary}

\begin{proof}
Let $v_m\in\mathcal F$ be pointwise cofinal. Then
$\sup\mathcal F=\sup_m v_m$, and $v_m\preceq_{S,\mathrm{loc}}H$
for all $m$. Countable subadditivity gives
$\sup\mathcal F\preceq_{S,\mathrm{loc}}H$. Conversely every
$u_j$ is a finite maximum of members of $\mathcal F$, so
$H\leq\sup\mathcal F$ pointwise.
\end{proof}

\subsection{Obstacle families and the scope of the result}

Let $\mathcal F$ be as in the local theorem, and suppose every
member satisfies the actual pointwise condition $v\leq\psi$
on a specified set $E$. Nonempty finite maxima preserve this
condition. The constructed representative $H=\sup_j u_j$
therefore satisfies $H\leq\psi$ at every point of $E$.
This gives an extremal object in the capacitary closure which
retains a prescribed pointwise obstacle while respecting the
background $S$.

These envelope constructions are recorded as \texttt{NEW\_PROVED}
consequences of the relative energy estimates and the completed
calculus. They do not assert that $H$ is classically psh, that
$H=(\sup\mathcal F)^*$, or that its Monge--Amp\`ere measure is
supported on a contact set. They do not establish compact outer
regularity or weak-closure domination for the original Borel
capacity for an arbitrary background. For the ambient background,
the later negligible-set theorem identifies $H$ with the ordinary
usc envelope quasi-everywhere, and the exhaustion-domain extremal
theorems supply the stated equilibrium and capacity formulas.
These later conclusions use additional arguments.

% END source: base_r3/base_r2/sections/relative_capacitary_envelopes.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/domain_comparison.tex
\section{Local comparison with the capacity of a fixed affine domain}
\label{sec:r2-domain-comparison}

Fix a smooth integral affine $K$-variety $Y$ of dimension $n\geq1$
and an ordinary open set
\[
 V\subset Y^{\mathrm{an}},\qquad
                    \overline V\Subset Y^{\mathrm{an}}.
 \tag{D}\label{eq:r2-domain-comparison-scope}
\]
Let $S$ be any symmetric original-positive, separately $d'$- and
$d''$-closed current of bidegree $(p,p)$ on $V$, and put
$q=n-p\geq1$. Restrictions of $S$ to smaller ordinary domains are
understood in the notation $\operatorname{Cap}_S(\,\cdot\,;\Omega)$.
All functions retain their actual point values and the all-extensions
classical definition.

The purpose of the construction is to compare admissible functions
on different domains without assuming that they extend unchanged.
An explicit smooth pluriharmonic correction on the inner core is
sufficient, because that correction disappears from every
differentiated BT slot over an arbitrary closed background.

\subsection{The effect of a change of frame}

\begin{lemma}[Smooth pluriharmonic invariance]
\label{lem:r2-harmonic-frame-invariance}
Let $O\subset V$ be ordinary open. Let $a_i,b_i\in\Cb(O)$ satisfy
$a_i=b_i+\psi_i$, where $\psi_i$ is original smooth and
$D\psi_i=0$. Then, for $0\leq k\leq q$,
\[
 T_{S,k}(a_1,\ldots,a_k)=T_{S,k}(b_1,\ldots,b_k)
 \quad\text{on }O.
\]
In particular this applies when each $\psi_i$ is a real constant
plus a real multiple of the logarithm of the absolute value of an
analytic unit.
\end{lemma}

\begin{proof}
For every separately closed current $Q$ the original smooth product
rule gives
\[
 D(\psi_i Q)=D\psi_i\wedge Q=0.
\]
All cross terms vanish because $d'Q=d''Q=0$. This is an identity
against all original compact smooth complementary tests. The result
now follows inductively from the ordered recursion: first replace
the inner $k-1$ slots by induction and then use the displayed identity
for the outer slot. Only multiplication by an original smooth
function is used in this last step. A unit logarithm is original
smooth and has zero $D$, as it is linear in its unit moment
coordinate. Its positive and negative are classical psh after every
complete extension.
\end{proof}

This \texttt{DIRECT\_COROLLARY} of the ordered calculus is much
narrower than a claim that arbitrary-background products depend only
on the ambient currents $D[a_i]$ for arbitrary potentials. Only the
specified smooth zero-curvature differences are used below.

\subsection{Uniform extension from one inner norm box}

Choose a closed affine embedding $Y\hookrightarrow\mathbf A^N_K$.
Let $Z$ be the normalization of its projective closure, let
\[
 A=\nu^*\mathcal O_{\mathbf P^N}(1),
\]
and write $e_\infty=\nu^*X_0$ for its fixed frame on $Y$. The line
$A$ is ample, and $Y$ identifies with the inverse image of the
standard affine chart because it is already normal. We do not assume
that $Z$ is smooth. No current or BT product on its singular locus
will be used.

The polynomial norm boxes below are those in the constructive part of
\textup{B}, Lemmas D.1--D.2. They use one finite list of polynomial
upper and lower bounds, with strict slack between the successive
thresholds and with upper bounds on the affine coordinates. In
particular the closed boxes are compact. They may always be chosen
locally inside a prescribed ordinary neighborhood.

\begin{lemma}[A uniform extension up to one fixed frame term]
\label{lem:r2-uniform-framed-extension}
Let $\Omega\subset V$ be ordinary open, and fix nested polynomial
norm boxes
\[
 C_1\subset\operatorname{Int}C_{\mathrm{mid}},\qquad
 C_{\mathrm{mid}}\subset\operatorname{Int}C_2,\qquad
 C_2\Subset\Omega,
\]
with the strict-threshold properties just described. Put
$W=\operatorname{Int}C_1$. There are a number $A_0>0$ and one
original smooth pluriharmonic function $\psi$ on a neighborhood of
$C_1$, depending on the boxes and $V$ but not on $v$ or $S$, such
that every $v\in\Cb(\Omega)$ with $0\leq v\leq1$ admits
$\widetilde v\in\Cb(V)$ satisfying
\[
 0\leq\widetilde v\leq1\text{ on }V,
 \qquad
 \widetilde v=A_0^{-1}v+\psi\text{ on a neighborhood of }C_1.
 \tag{E}\label{eq:r2-uniform-framed-extension}
\]
Consequently, for every $k\leq q$,
\[
 T_{S,k}(\widetilde v,\ldots,\widetilde v)
       =A_0^{-k}T_{S,k}(v,\ldots,v)\quad\text{on }W.
 \tag{P}\label{eq:r2-extension-product}
\]
\end{lemma}

\begin{proof}
We give the geometric construction in full to avoid invoking either a
smooth compactification or \textup{B}, Theorem~4.4.

Choose intermediate thresholds between those defining $C_1$ and
$C_{\mathrm{mid}}$. By clearing rational powers, the polynomial
collar can be written
\[
 b=\max\{0,\log|P_a^{N_a}/c_a|,
                       \log|d_l/Q_l^{M_l}|\},
 \qquad c_a,d_l\in K^\times.
\]
Every denominator $Q_l$ is nonzero on a neighborhood of $C_2$.
The chosen strict gaps give
\[
 b=0\text{ near }C_1,
 \qquad b\geq\delta>0\text{ on }C_2\setminus\operatorname{Int}
                                    C_{\mathrm{mid}}.
\]
The function is finite and continuous on a neighborhood of $C_2$.
It is classical psh there: each nonconstant branch is a logarithm of
an analytic function or of an analytic unit; the constant branch
makes branches with a zero inactive in a neighborhood of that zero.
These assertions persist after every complete extension.

Clear the denominators and homogenize in the fixed projective
embedding. After taking a common sufficiently high degree $m_0$,
all displayed branches are ratios $s_a/s_0$ of global sections of
$A^{m_0}$. Here $s_0$ is a product of homogenized denominator
polynomials and a power of $X_0$, pulled back to $Z$, and is nowhere
zero on a neighborhood of $C_2$. Include the branch $s_0/s_0=1$.
Append small nonzero $K$-scalar multiples of a finite global
generating tuple of $A^{m_0}$. The scalars can be chosen so that
all their ratios to $s_0$ have absolute value strictly less than one
on $C_2$, by compactness. They do not change the maximum $b$ there,
and the full tuple generates the line everywhere on $Z$.

Its pure Fubini--Study metric $q_0$ on $A^{m_0}$ has weight $b$
in the frame $s_0$ near $C_2$. This metric is classical on $Z$ in
the literal sense that every local frame weight is classical psh
after every complete extension. Indeed, pullback along any analytic
curve gives the finite maximum of logarithms of analytic functions
with no common zero. This argument uses generation and the sheaf
and maximum properties, and needs no smoothness of $Z$.

Fix $c>0$, put $M=c+1$, and choose an integer $m>0$ with
$m\delta>M+1$. Set
\[
 d=mm_0,\qquad L=A^d,\qquad e=s_0^m,\qquad Q=mq_0.
\]
All these objects are independent of $v$. Define a metric $R_v$ on
$L$ using the two ordinary opens
$\operatorname{Int}C_2$ and $Z^{\mathrm{an}}\setminus C_{\mathrm{mid}}$:
its $e$-weight on the first is
\[
 \max\{v+c,mb\},
\]
and on the second it equals $Q$. The opens cover $Z^{\mathrm{an}}$
because $C_{\mathrm{mid}}\subset\operatorname{Int}C_2$. On their
overlap the inequality $mb\geq m\delta>M\geq v+c$ makes the two
prescriptions identical. They therefore glue as metrics on the same
identified line. Their weights are all-extensions classical by
restriction, maxima, and locality on curves. This is an explicit
gluing construction on the normal variety, not an application of a
global approximation or envelope theorem there.

The metric difference is a globally defined bounded Borel function
and satisfies
\[
 0\leq R_v-Q\leq M.
\]
On $\operatorname{Int}C_2$ this is the identity
$R_v-Q=\max\{v+c-mb,0\}$ and the bound follows from $b\geq0$.
On the other open the difference is zero.

Now use the fixed frame $e_\infty^d$ on all of $Y$. Let
\[
 w_v=\operatorname{wt}_{e_\infty^d}(R_v),\qquad
 Q_\infty=\operatorname{wt}_{e_\infty^d}(Q).
\]
The function $w_v$ is classical psh on $Y^{\mathrm{an}}$, and
$Q_\infty$ is continuous there. In particular they are bounded on
$V$, uniformly in $v$, because
\[
 Q_\infty\leq w_v\leq Q_\infty+M
 \quad\text{and}\quad\overline V\Subset Y^{\mathrm{an}}.
\]
Write
\[
 a=\min_{\overline V}Q_\infty,
 \qquad
 A_0=1+\operatorname{osc}_{\overline V}(Q_\infty)+M,
 \qquad
 \widetilde v=\frac{w_v-a}{A_0}\Big|_V.
\]
This gives $0\leq\widetilde v\leq1$ on the original domain $V$.
The actual bounds and $A_0$ are independent of the competitor.

Near $C_1$ we have $b=0$ and $v+c>0$, so the $e$-weight of $R_v$
is exactly $v+c$. The ratio
\[
 f=\frac{s_0^m}{e_\infty^d}
\]
is an algebraic unit near $C_2$. Since $e=f e_\infty^d$, the metric
transition law gives
\[
 w_v=v+c+\log|f|\quad\text{near }C_1.
\]
Thus \eqref{eq:r2-uniform-framed-extension} holds with the fixed
function
$\psi=A_0^{-1}(c+\log|f|-a)$. It is original smooth
pluriharmonic. Positive homogeneity and
Lemma~\ref{lem:r2-harmonic-frame-invariance} prove
\eqref{eq:r2-extension-product}, with the original background and on
all original tests. The normal projective variety was used only to
construct the metric and its frame; every current in this final step
is on the smooth open $Y$.
\end{proof}

The lemma is recorded as \texttt{NEW\_PROVED} in the stated uniform
framed-affine scope, by the constructive collar mechanism of
\textup{B}, Lemma D.2. Its proof does not use that lemma's smooth
projective ambient hypothesis, a resolution of singularities, or
\textup{B}, Theorem~4.4. The following capacity consequences are
\texttt{DIRECT\_COROLLARY}.

\subsection{Borel and outer capacities on different domains}

For arbitrary subsets of a domain $O$, write
\[
 \operatorname{Cap}_S^*(E;O)
   =\inf_{\substack{G\supset E\\G\text{ ordinary open in }O}}
                 \operatorname{Cap}_S(G;O).
\]
This is the outer capacity by definition; no assertion that it agrees
with the original Borel capacity is used here.

\begin{theorem}[Uniform local comparison with a fixed domain]
\label{thm:r2-domain-capacity-comparison}
Assume \eqref{eq:r2-domain-comparison-scope}. For every ordinary
$\Omega\subset V$ and every ordinary $W$ with
$\overline W\Subset\Omega$, there is a finite constant
$C=C(Y,V,\Omega,W,q)>0$, independent of $S$, such that
\[
 \operatorname{Cap}_S(E;V)
 \leq\operatorname{Cap}_S(E;\Omega)
 \leq C\operatorname{Cap}_S(E;V)
 \quad(E\subset W\text{ Borel}).
 \tag{C}\label{eq:r2-domain-capacity-comparison}
\]
For arbitrary $E\subset W$ one also has
\[
 \operatorname{Cap}_S^*(E;V)
 \leq\operatorname{Cap}_S^*(E;\Omega)
 \leq C\operatorname{Cap}_S^*(E;V).
 \tag{C*}\label{eq:r2-domain-outer-comparison}
\]
For a single inner core in
Lemma~\ref{lem:r2-uniform-framed-extension}, one may take
$C=A_0^q$.
\end{theorem}

\begin{proof}
The first inequality of \eqref{eq:r2-domain-capacity-comparison}
is restriction of competitors: a normalized function on $V$ remains
normalized and classical on $\Omega$, and its product restricts
there by ordinary locality.

First suppose that $W$ is the core of one nested-box construction.
For an arbitrary competitor $0\leq v\leq1$ on $\Omega$, the extension
lemma and equality of actual Radon measures on $W$ give
\[
 \int_E T_{S,q}(v,\ldots,v)
 =A_0^q\int_E T_{S,q}(\widetilde v,\ldots,\widetilde v)
 \leq A_0^q\operatorname{Cap}_S(E;V).
\]
Taking the supremum over $v$ proves the second inequality for that
core. The constant is uniform over all competitors and all allowed
backgrounds.

For general $W$, cover the compact $\overline W$ by finitely many
inner cores $W_1,\ldots,W_N$ of nested polynomial norm boxes whose
outer boxes lie in $\Omega$. Let $A_i$ be the constants from their
extension lemmas. The same competitor $v$ on $\Omega$ can be
extended separately on each core. Positivity and a finite cover give
\[
 \begin{split}
 \int_E T_{S,q}(v,\ldots,v)
 &\leq\sum_{i=1}^N\int_{E\cap W_i}T_{S,q}(v,\ldots,v)\\
 &\leq\sum_{i=1}^NA_i^q\operatorname{Cap}_S(E\cap W_i;V)\\
 &\leq\left(\sum_{i=1}^NA_i^q\right)
                                 \operatorname{Cap}_S(E;V).
 \end{split}
\]
Take the supremum. Thus $C=\sum_iA_i^q$ works. This finite-cover
step avoids assuming that an arbitrary compact core lies in one
strictly smaller box of a previously chosen presentation. In
particular, no strict-inequality description of an ordinary interior
is silently imposed.

For outer capacities, every ordinary open of $\Omega$ is ordinary
open in $V$, so restriction and the first Borel inequality give the
first inequality of \eqref{eq:r2-domain-outer-comparison}. Conversely,
if $G\subset V$ is ordinary open and contains $E$, then $G\cap W$
is an ordinary open of $\Omega$ containing $E$. The already proved
Borel comparison and monotonicity give
\[
 \operatorname{Cap}_S^*(E;\Omega)
 \leq\operatorname{Cap}_S(G\cap W;\Omega)
 \leq C\operatorname{Cap}_S(G\cap W;V)
 \leq C\operatorname{Cap}_S(G;V).
\]
Take the infimum over such $G$. This argument invokes the definition
of outer capacity and no outer-regularity theorem.
\end{proof}

The endpoint $p=n$ can be included by defining the degree-zero
capacity to be $S(E)$; domain restriction then makes both Borel
capacities identical, with constant one.

\subsection{Equivalent local convergence and null sets}

\begin{corollary}[One fixed capacity domain suffices locally]
\label{cor:r2-fixed-domain-local-convergence}
Assume \eqref{eq:r2-domain-comparison-scope}, and let $a_j,a$ be
real Borel functions on $V$. The following are equivalent:
\begin{enumerate}
\item For every compact $L\Subset V$ and every $\delta>0$,
\[
 \operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};V)
                          \longrightarrow0.
\]
\item The functions converge in the all-domain local $S$-capacity
      sense of Definition~\ref{def:local-capacity-convergence},
      restricted to $V$.
\end{enumerate}
The same equivalence holds with the outer capacities throughout.
No uniform boundedness or classicality of the functions $a_j,a$ is
needed for this equivalence of convergence conditions.
\end{corollary}

\begin{proof}
Suppose (1) holds. Given a permitted relatively compact domain
$\Omega\subset V$ and a compact $L\Subset\Omega$, choose an
ordinary $W$ with $L\subset W$ and $\overline W\Subset\Omega$.
The second inequality of
\eqref{eq:r2-domain-capacity-comparison}, with this fixed geometry,
implies the required convergence using $\Omega$ as admissibility
domain.

Conversely, cover a prescribed compact $L\Subset V$ by finitely many
ordinary $W_i$ with compact closure in permitted small norm-box
domains $\Omega_i\Subset V$. Domain monotonicity and finite
subadditivity give
\[
 \begin{split}
 \operatorname{Cap}_S(L\cap\{|a_j-a|>\delta\};V)
 &\leq\sum_i\operatorname{Cap}_S
       (\overline W_i\cap\{|a_j-a|>\delta\};\Omega_i).
 \end{split}
\]
Each summand tends to zero by (2). The same proof uses
\eqref{eq:r2-domain-outer-comparison} and subadditivity of outer
capacity for the outer version.
\end{proof}

\begin{corollary}[Equivalence of local null classes]
\label{cor:r2-fixed-domain-null-classes}
For a Borel set $E\subset V$, its capacity with domain $V$ is zero
if and only if
\[
 \operatorname{Cap}_S(E\cap L;\Omega)=0
\]
for every permitted small domain $\Omega\subset V$ and compact
$L\Subset\Omega$. For arbitrary subsets $E$, the same assertion
holds with outer capacities.
\end{corollary}

\begin{proof}
One direction is the comparison theorem. For the converse, $V$ is
second countable in the present finite $p$-adic affine setting.
Choose countably many ordinary cores $W_i$ covering $V$, with
$\overline W_i\Subset\Omega_i$ for permitted small norm-box domains
$\Omega_i\Subset V$. Monotonicity under restriction and countable
subadditivity give
\[
 \operatorname{Cap}_S(E;V)
 \leq\sum_i\operatorname{Cap}_S(E\cap\overline W_i;\Omega_i)=0.
\]
The outer-capacity proof is identical. No ambient upper-semicontinuous
representative is substituted in this argument.
\end{proof}

These statements identify the local convergence and local null-set
notions on the qualified fixed domain $V$. They do not assert uniform
constants as the core approaches the boundary of $V$, convergence in
capacity on the entire noncompact domain from local convergence, or
capacity nondegeneracy. They remain valid for $S=0$ and backgrounds
supported on a proper analytic subvariety. They do not apply to an
arbitrary proper ambient space with only constant untwisted global
psh competitors. Every comparison is between ordinary domains; the
sets being measured may be arbitrary Borel sets, including non-open
strict gaps.

% END source: base_r3/base_r2/sections/domain_comparison.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/support_envelopes.tex
\section{Analytic supports and a relative-envelope obstruction}
\label{sec:r2-support-envelopes}

\begin{proposition}[BT products and capacity on an analytic retract]
\label{prop:r2-retract}
Let $i:Y\hookrightarrow U$ be a closed analytic embedding of allowed
pure-dimensional quasi-smooth spaces, of dimensions $m$ and $n$.
Let $R$ be a symmetric original-positive separately closed current of
bidegree $(a,a)$ on $Y$, with $0\leq a\leq m$, and put
\[
 S=i_*R,\qquad p=n-m+a,\qquad q=m-a=n-p.
\]
For $0\leq k\leq q$ and $u_1,\ldots,u_k\in\Cb(U)$,
\begin{equation}
\label{eq:r2-retract-product}
 T^U_{S,k}(u_1,\ldots,u_k)
   =i_*T^Y_{R,k}(u_1\circ i,\ldots,u_k\circ i)
\end{equation}
on all original compact smooth complementary tests.
If there is an analytic retraction $r:U\to Y$ with $r\circ i=\mathrm{id}$,
then, for every Borel $E\subset U$,
\begin{equation}
\label{eq:r2-retract-capacity}
 \operatorname{Cap}_{S}(E;U)
     =\operatorname{Cap}_{R}(i^{-1}E;Y).
\end{equation}
The corresponding equality holds for outer capacities of arbitrary
subsets. In degree zero the capacity is simply the measure of the set.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
This is a functorial consequence of the fixed-background recursion,
not a restriction of that recursion to integration-current backgrounds.

\begin{proof}
The map $i$ is proper. Its pullback preserves the original positive
test cone and commutes with the two differentials; hence $S$ has the
stated positivity and closedness. For any coefficient-Radon current
$P$ on $Y$ and locally bounded Borel function $f$ on $U$,
\[
 f\,i_*P=i_*((f\circ i)P).
\]
Indeed, the coefficient measure of $i_*P$ against a test $\eta$ is
the pushforward of the coefficient measure of $P$ against $i^*\eta$.
This follows first on smooth scalar tests and then on Borel functions
by Radon uniqueness. Apply this identity and commutation of $D$ with
proper pushforward inductively through the recursion. Analytic
pullback preserves classical psh, so every restricted slot is allowed.
This proves \eqref{eq:r2-retract-product}.

Every normalized potential on $U$ restricts to a normalized potential
on $Y$, giving the inequality $\leq$ in
\eqref{eq:r2-retract-capacity}. Conversely every normalized $v$ on
$Y$ extends to $v\circ r$ on $U$, still classical and between zero
and one. Equation~\eqref{eq:r2-retract-product} gives the opposite
inequality. For outer capacities, restricting any ordinary open
neighborhood of $E$ gives one of $i^{-1}E$. Conversely, if $H\subset Y$
is open and contains $i^{-1}E$, choose an open $G_0\subset U$ with
$i^{-1}G_0=H$. The open set $G_0\cup(U\setminus i(Y))$ contains
$E$ and has capacity $\operatorname{Cap}_{R}(H;Y)$. Taking infima
proves the assertion.
\end{proof}

\begin{proposition}[Ambient negligible sets can have positive relative capacity]
\label{prop:r2-negligible-positive}
For every $m\geq1$, there is an allowed $(m+1)$-dimensional domain
$U$, a nonzero positive closed background $S$ of bidegree $(1,1)$,
and an increasing sequence $a_j\in\Cb(U)\cap C^0(U)$, with
$-1\leq a_j\leq0$, such that, writing $h=\sup_j a_j$,
\[
 h^*=0,\qquad
 \operatorname{Cap}_{S}(\{h<h^*\};U)>0.
\]
There is even a compact singleton $K\subset U$ of positive relative
capacity for which the ambient relative extremal function
\[
 h_K=\sup\{v\in\Cb(U):v\leq0,\ v|_K\leq-1\}
\]
satisfies $h_K^*=0$.
\end{proposition}

\noindent\textbf{Status of the assertions refuted: \texttt{FALSE}.}
The construction refutes both universal relative negligibility of
ambient envelope exceptions and the unqualified formula
$\operatorname{Cap}_S(K;U)=\int_U T_{S,m}((h_K^*)^m)$.

\begin{proof}
Choose $0<r<1<R$ and set
\[
 U=\{|z_0|<1\}\times\prod_{i=1}^m\{r<|z_i|<R\},
 \qquad Y=\{z_0=0\},\qquad S=i_*[1]_Y.
\]
The projection onto the last $m$ coordinates is an analytic
retraction. Put
\[
 a_j=\max\{-1,j^{-1}\log|z_0|\}.
\]
The logarithm is interpreted as $-\infty$ on $Y$. The maximum is
continuous and classical, it equals $-1$ on $Y$, and it increases
to zero at every point outside $Y$. Hence $h=-1$ on $Y$ and $h=0$
outside $Y$.

The complement of $Y$ is dense in $U$. For an explicit verification
at $y\in Y$, use the Gauss seminorm in the $z_0$ variable over the
completed residue field $\mathcal H(y)$, with radius $s>0$. It defines
a point $y_s$ of the original $K$-analytic product with
$|z_0(y_s)|=s$. As $s\downarrow0$, evaluations of polynomials tend
to their values at $y$, and so do local analytic functions by uniform
approximation on a compact product neighborhood. Thus $y_s\to y$.
Consequently $h^*=0$.

Let $\xi_Y$ be the product Gauss point of $Y$ and $K=\{i(\xi_Y)\}$.
For $u_i=\max\{0,\log|z_i|\}$, the standard model metric on
$(\mathbf P^1_{K^\circ})^m$ gives
\[
 T_{S,m}(u_1,\ldots,u_m)=\delta_{i(\xi_Y)}.
\]
This is exactly the mixed model calculation of the first record,
Proposition~\ref{prop:polar-relative-mass}, with the original
absolute value and calibration. Put $M=m\log R$ and
$v=M^{-1}\sum_i u_i$. Then $0\leq v\leq1$ on $U$ and positivity
and multilinearity give
\[
 T_{S,m}(v^m)(K)\geq \frac{m!}{(m\log R)^m}>0.
\]
In particular both $K$ and $Y=\{h<h^*\}$ have positive relative
capacity.

Every $a_j$ is a competitor for $h_K$. Thus $h_K=0$ off $Y$, while
$h_K\leq0$ everywhere. Density again gives $h_K^*=0$. The constant
$-1$ is a competitor and every competitor is at most $-1$ on $K$,
so $h_K=-1$ on $K$ itself. All differentiated slots of $h_K^*=0$
vanish, whereas the displayed capacity of $K$ is positive. This
proves the claimed failure without an appeal to an unspecified
pluripolar exceptional set.
\end{proof}

\paragraph{A correct replacement in this setting.}
For a background supported on an analytic retract, the intrinsic
capacity and extremal problem on $Y$ are the appropriate comparison
objects, by Proposition~\ref{prop:r2-retract}. Ambient upper
regularization can erase values on $Y$ that the background detects.
For a general positive closed current its support need not be an
analytic space; this observation does not by itself define an
intrinsic psh envelope on that support. Such a construction remains a
separate analytic problem.

% END source: base_r3/base_r2/sections/support_envelopes.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/gap_detection.tex
\section{Detection of strict gaps and actual point values}
\label{sec:r2-gap-detection}

All fields, currents, and test forms retain the conventions of the first
round. In particular, $K/\mathbf Q_p$ is finite with its prescribed
absolute value, $D=d'd''$, and $\Cb$ denotes the actual pointwise,
locally bounded, all-extensions classical psh class. No representative
is altered on a null set. This section supplies a sufficient geometric
condition for the strict-gap hypothesis in
Corollary~\ref{cor:conditional-pointwise-domination}. The condition is
proved for $S=[1]$ on the domains specified below, and for an explicit
family of positive closed backgrounds. It is not asserted for every
positive closed background or every ordinary domain.

\subsection{Shilov boundaries determine pointwise order}

\begin{definition}[A family of affinoid neighborhoods]
Let $V$ be an ordinary open in a good strictly analytic space. A family
$\mathscr A$ of strict affinoid domains contained in $V$ is called an
affinoid neighborhood basis if, whenever $x\in O\subset V$ and $O$ is
ordinary open, some $A\in\mathscr A$ satisfies
\[
 x\in\operatorname{Int}_V(A)\subset A\subset O.
\]
Write $\Gamma(A)$ for its finite Shilov boundary and put
$\mathscr D=\bigcup_{A\in\mathscr A}\Gamma(A)$.
\end{definition}

The external input used here is \cite[Theorem~6.4, pp.~33--34]{GR25}:
a classical psh function on an affinoid space attains its maximum at a
Shilov point. Its definition of classical psh uses curve pullback after
every complete extension. The theorem applies to the restriction of a
function in $\Cb(V)$ to each $A$; it does not require that the function
be continuous. The following lemma is a \texttt{DIRECT\_COROLLARY} of
that theorem and upper semicontinuity.

\begin{lemma}[A strict gap contains a Shilov point]
\label{lem:r2-shilov-gap}
Let $a$ be a finite classical psh function on $V$, let $b$ be a finite
upper-semicontinuous function on $V$, and let $\delta\geq0$. If
$a(x)>b(x)+\delta$, then every sufficiently small ordinary neighborhood
of $x$ contains a point $z\in\mathscr D$ with
$a(z)>b(z)+\delta$. Consequently,
\[
 a\leq b\text{ on }\mathscr D\quad\Longrightarrow\quad a\leq b
 \text{ on }V.
\]
In particular, two members of $\Cb(V)$ that agree on $\mathscr D$
agree at every point of $V$.
\end{lemma}

\begin{proof}
Choose $\epsilon>0$ with
$b(x)+\delta+2\epsilon<a(x)$. Upper semicontinuity gives an ordinary
neighborhood $O$ of $x$ on which $b<b(x)+\epsilon$. Choose
$A\in\mathscr A$ with $x\in\operatorname{Int}_V(A)$ and $A\subset O$.
The affinoid maximum principle gives $z\in\Gamma(A)$ such that
$a(z)=\max_A a\geq a(x)$. Therefore
\[
 a(z)\geq a(x)>b(x)+\delta+2\epsilon>b(z)+\delta.
\]
The same construction can be made inside any prescribed ordinary
neighborhood of $x$, by intersecting it with $O$. This proves the first
assertion. If pointwise order failed, apply it with $\delta=0$ to
contradict order on $\mathscr D$. Apply that implication in both orders
to obtain equality. The strict-gap set itself has nowhere been treated
as ordinary open.
\end{proof}

For example, for a fixed $x\in V$ the same maximum principle and upper
semicontinuity give the recovery formula
\[
 a(x)=\inf_{\substack{A\in\mathscr A\\
                         x\in\operatorname{Int}_V(A)}}
                  \max_{z\in\Gamma(A)}a(z).
\]
The infimum is over the members of the full neighborhood basis. This
is a pointwise statement, rather than an assertion about equality
classes modulo any measure.

\subsection{An atom at a prescribed divisorial point}

For the fixed-domain result, assume that $Y/K$ is a smooth integral
affine variety of dimension $n\geq1$ and that
\[
 V\subset Y^{\mathrm{an}}
 \quad\text{is ordinary open with}\quad
 \overline V\Subset Y^{\mathrm{an}}.
 \tag{G}\label{eq:r2-framed-domain}
\]
The closure here is the ordinary topological closure. More generally,
the argument applies when $V$ has compact closure inside an algebraic
open of a normal projective variety on which a fixed ample line bundle
has a specified nowhere-vanishing frame, and that open is smooth.
No convexity, exhaustion, or regularity of the boundary of $V$ is
assumed. Small norm boxes in the algebraic neighborhoods supplied by
\textup{B} satisfy \eqref{eq:r2-framed-domain}, after choosing one
affine chart.

We use the following external facts with their original field:
\cite[Proposition~A.3, p.~365; Corollary~A.6, p.~367;
Proposition~A.9, p.~368]{GM19} identify Shilov points of strict
affinoids with divisorial points, show locality of divisoriality, and
realize divisorial points on algebraic $K^\circ$-models of a proper
variety. The ample extension theorem is
\cite[Proposition~4.11, p.~353]{GM19}. The ordinary CLD model-measure
formula is \cite[Theorem~6.9.3 and Proposition~6.9.7]{CLD12}; its
generically smooth and stable-field hypotheses apply here. For smooth
projective varieties the same normalization is explicitly recorded in
\textup{R}, Lemma~3.3, and \textup{B}, equation~(4.1). We only use the
ordinary-current measure, not equality on all $\delta$-test forms.

\begin{lemma}[A bounded model tester on the original domain]
\label{lem:r2-atomic-tester}
Assume \eqref{eq:r2-framed-domain}. For every divisorial point
$z\in V$ there exists a continuous semipositive real PL function
$h_z\in\Cb(V)$ such that
\[
 0\leq h_z\leq1\quad\text{on }V,
 \qquad (Dh_z)^n(\{z\})>0.
\]
The function is defined and bounded on a neighborhood of $\overline V$.
If a larger relatively compact ordinary $V^+$ with
$\overline V\subset V^+\Subset Y^{\mathrm{an}}$ is fixed in advance,
one can arrange $0\leq h_z\leq1$ on this same $V^+$ for every $z$.
\end{lemma}

\begin{proof}
Choose a closed embedding $Y\hookrightarrow\mathbf A^N_K$ and take
the normalization $Z$ of its closure in $\mathbf P^N_K$. Because $Y$
is normal, the inverse image of the standard affine chart in $Z$ is
identified with $Y$. If $\nu:Z\to\overline Y$ is the normalization,
then $L=\nu^*\mathcal O_{\mathbf P^N}(1)$ is ample, and the pullback
of the homogeneous coordinate $X_0$ gives a fixed frame $e$ of $L$
on $Y$. Thus no modification of the open line or its frame has been
made.

The point $z$ is divisorial in $Z^{\mathrm{an}}$. By the cited model
realization theorem there is an algebraic $K^\circ$-model
$\mathscr Z_0$ of $Z$ on which $z$ is associated with an irreducible
component of the special fiber. Normalize this model. Normalization is
finite in the present excellent discretely valued situation, and the
center selected by $z$ is still a special-fiber component. Apply the
ample extension theorem to obtain a dominating model $\mathscr Z$
and an ample line bundle $\mathscr L$ with a specified identification
\[
 \mathscr L|_Z\simeq L^{\otimes m},\qquad m>0.
\]
Normalize once more if necessary and pull back $\mathscr L$; finite
pullback preserves ampleness. A dominating model of a normal model
preserves each codimension-one divisorial valuation represented on the
latter. Indeed, after localizing the target at the generic point of
its component one has a discrete valuation ring, and the proper
birational morphism is an isomorphism at that generic point. Hence $z$
is still associated with a special-fiber component $E$ of
$\mathscr Z$.

For clarity, the ample extension used here does not require resolution
of mixed-characteristic models. One may first choose a projective
model carrying an ample extension of a power of $L$, take a vertical
blowup dominating $\mathscr Z_0$, and tensor a sufficiently high
pullback power with its relatively ample tautological line. The center
of the blowup lies in the special fiber, so the tautological line has
trivial generic fiber. This produces an ample extension of the stated
power of $L$, before normalization. This is an existence of a
dominating model, not a claim about an arbitrary previously fixed
model.

The model metric of $\mathscr L$ gives the weight
\[
 \phi=-\log\|e^{\otimes m}\|_{\mathscr L}
 \quad\text{on }Y^{\mathrm{an}}.
\]
It is continuous and semipositive PL. In particular it is classical
psh after all complete extensions by
\cite[Theorem~4.3, pp.~22--23]{GR25}. Equivalently, replace
$\mathscr L$ by a positive power that is generated by model sections.
Its weight is then a pure Fubini--Study maximum in the same frame;
the inactive-branch rule handles zeros of the sections. This also
shows directly that all functions used are defined over the original
$K$.

On a normal algebraic model the original CLD model formula gives the
atom
\[
 (D\phi)^n(\{z\})
   =b_E\deg_{\widetilde K,E}\bigl((\mathscr L|_E)^n\bigr)>0,
\]
where $b_E$ is the special-fiber multiplicity and the structural degree
retains residue-field degrees. Ampleness makes the displayed degree
strictly positive. This formula is interpreted in the original
calibration; neither the absolute value nor the integration current
is rescaled. The continuous BT product on the smooth open $Y$ is this
ordinary CLD measure by the accepted continuous compatibility of
\textup{R} and \textup{B}. If a power was used to obtain generating
sections, its tensor degree is retained in this formula and in the
frame, or divided out in every slot.

Fix $V^+$ as in the statement, taking one if none was prescribed.
The continuous function $\phi$ is bounded on $\overline {V^+}$. Choose
real numbers $A<B$ with
$A\leq\phi\leq B$ there, and set $h_z=(\phi-A)/(B-A)$. A positive
constant multiple and a constant translate preserve classicality and
semipositive real PL structure. They multiply the displayed atom by
$(B-A)^{-n}>0$. This proves all assertions on the fixed domain $V$.
\end{proof}

The lemma is a \texttt{DIRECT\_COROLLARY} of the cited model geometry,
model-measure formula, and the accepted continuous compatibility. It
does not solve a prescribed Dirac Monge--Amp\`ere equation: other
positive atoms of $(Dh_z)^n$ are allowed.

\begin{theorem}[Strict-gap detection for the ambient capacity]
\label{thm:r2-ambient-gap-detection}
Assume \eqref{eq:r2-framed-domain}. If $a,b\in\Cb(V)$ and
$\{a>b+\delta\}$ is nonempty for some $\delta\geq0$, then
\[
 \operatorname{Cap}_{[1]}(\{a>b+\delta\};V)>0.
\]
Here the supremum defining capacity uses functions on the original
domain $V$. In particular the strict-gap hypothesis of
Corollary~\ref{cor:conditional-pointwise-domination} holds for
$S=[1]$ on $V$.
\end{theorem}

\begin{proof}
Choose a strict affinoid neighborhood basis in $V$. By
Lemma~\ref{lem:r2-shilov-gap}, the gap contains a Shilov point $z$
of one of these affinoids. Such a point is divisorial in
$Y^{\mathrm{an}}$ by the cited locality of divisorial points. Choose
the tester $h_z$ from Lemma~\ref{lem:r2-atomic-tester}. The gap is
Borel, since both potentials are upper semicontinuous. Thus
\[
 \operatorname{Cap}_{[1]}(\{a>b+\delta\};V)
 \geq\int_{\{a>b+\delta\}}(Dh_z)^n
 \geq(Dh_z)^n(\{z\})>0.
\]
Every tester in this inequality is defined on $V$. There is no use of
an implication from positive capacity on a smaller domain to positive
capacity on $V$.
\end{proof}

This is a proved consequence of the two lemmas, independently of any
outer regularity or identification of polar sets. In the project
ledger it is a \texttt{DIRECT\_COROLLARY} of those inputs and the
definition of capacity.

\begin{corollary}[Pointwise ambient domination and uniqueness]
\label{cor:r2-pointwise-ambient}
Assume \eqref{eq:r2-framed-domain} and let $u,v\in\Cb(V)$. If
\[
 \overline{\{u<v-\epsilon\}}\Subset V\quad(\epsilon>0),
 \qquad (Du)^n(\{u<v\})=0,
\]
then $u\geq v$ at every point of $V$. If the compact-gap condition
holds in both orders and $(Du)^n=(Dv)^n$, then $u=v$ at every point.
\end{corollary}

\begin{proof}
Apply Proposition~\ref{prop:capacity-domination} or
Proposition~\ref{prop:capacity-uniqueness}, respectively, and then
Theorem~\ref{thm:r2-ambient-gap-detection} to any prohibited strict
inequality with a positive margin. This uses only Borel restrictions
of the measures.
\end{proof}

\subsection{Faithfulness of original trace currents}

\begin{proposition}[Actual classical representatives are determined by
their original currents]
\label{prop:r2-current-faithfulness}
Let $U$ have the local algebraic neighborhoods supplied by \textup{B},
and let $u,v\in\Cb(U)$. Then
\[
 [u]\leq[v]\text{ as original CLD }(0,0)\text{-currents}
 \quad\Longleftrightarrow\quad u\leq v\text{ pointwise on }U.
\]
On the left, $[v]-[u]$ is nonnegative on every positive original
compactly supported $(n,n)$-form. In particular $[u]=[v]$ implies
actual pointwise equality. This proposition uses the fixed-background
weighted monotone theorem of the first round; it makes no assertion
of stability for arbitrary varying backgrounds.
\end{proposition}

\begin{proof}
Pointwise order plainly implies current order. For the converse it is
enough to work in a domain satisfying
\eqref{eq:r2-framed-domain}, inside an algebraic neighborhood of
each point. Fix a divisorial point $z$ there, and choose its model
tester $h$ by Lemma~\ref{lem:r2-atomic-tester}. On a fixed smaller
ordinary neighborhood $W$ of $z$, the smooth approximation theorem of
\textup{R} gives smooth psh functions converging uniformly to $h$.
The pure-FS realization in the proof of the lemma allows this also
to be obtained directly from \textup{R}, Proposition~A.1. After
taking a subsequence $s_j$ with
$\|s_j-h\|_W\leq4^{-j}$, set
\[
 h_j=s_j+2\cdot4^{-j}.
\]
Then $h_j\downarrow h$ on the same $W$, since
$h+4^{-j}\leq h_j\leq h+3\cdot4^{-j}$.
Each $h_j$ is smooth and belongs to the all-extensions classical
class: locally it is a convex, coordinatewise nondecreasing smooth
function of logarithms of analytic units, with zero-section branches
inactive. Its curvature is an original positive $(1,1)$-form.

For every nonnegative original smooth scalar
$\chi\in A_c^{0,0}(W)$, the form $\chi(Dh_j)^n$ is positive and
compactly supported. The assumed current order therefore gives
\[
 \big\langle (v-u)(Dh_j)^n,\chi\big\rangle\geq0.
\]
Apply Theorem~\ref{thm:relative-monotone} with fixed background
$[1]$, all differentiated slots equal to $h_j$, and the weight equal
to the constant sequence $u$ or $v$. It yields
\[
 (v-u)(Dh_j)^n\longrightarrow(v-u)(Dh)^n
\]
on all original scalar tests. The limit is therefore a positive Radon
measure. Its atom at $z$ is
$(v(z)-u(z))(Dh)^n(\{z\})$, and the second factor is strictly
positive. Hence $u(z)\leq v(z)$.

This holds at every point in the Shilov determining set of the local
domain. Lemma~\ref{lem:r2-shilov-gap} gives $u\leq v$ throughout
that domain, and these domains cover $U$ by \textup{B}. This proves
the converse. Equality follows by interchanging $u$ and $v$.
\end{proof}

Thus equality almost everywhere for all local original trace measures
does determine the actual classical representatives: it implies
equality of their $(0,0)$-currents, to which the proposition applies.
The essential passage to an atomic tester uses weighted convergence
of a decreasing potential with fixed background $[1]$. Weak
convergence of arbitrary background measures would not justify it.
The proposition is a \texttt{DIRECT\_COROLLARY} of that theorem,
the atomic testers, and the Shilov maximum principle.

\subsection{A precise condition on an arbitrary background}

Let $S$ be an original symmetric positive separately $d'$- and
$d''$-closed current of bidegree $(p,p)$ on $V$, with
$q=n-p\geq1$. An explicit sufficient condition is the following:
there is an affinoid neighborhood basis $\mathscr A$ such that, for
each $z\in\mathscr D=\bigcup_{A\in\mathscr A}\Gamma(A)$, there is
$h_z\in\Cb(V)$ satisfying
\[
 0\leq h_z\leq1,
 \qquad T_{S,q}(h_z,\ldots,h_z)(\{z\})>0.
 \tag{A}\label{eq:r2-relative-atomic-condition}
\]
The condition is stated in actual Borel measures on $V$, not only in
one tropical image.

\begin{proposition}[Pointwise conclusions under atomic detection]
\label{prop:r2-relative-gap-detection}
If \eqref{eq:r2-relative-atomic-condition} holds, every nonempty
strict gap of two members of $\Cb(V)$ has positive
$\operatorname{Cap}_S(\,\cdot\,;V)$. Consequently the pointwise
domination and uniqueness conclusions of
Corollary~\ref{cor:r2-pointwise-ambient} hold with
$(Da)^n$ replaced by $T_{S,q}(a,\ldots,a)$.
\end{proposition}

\begin{proof}
Lemma~\ref{lem:r2-shilov-gap} supplies a point $z$ of the gap in
$\mathscr D$. Its tester in \eqref{eq:r2-relative-atomic-condition}
assigns positive measure to the gap. The definition of capacity and
the first-round domination and uniqueness propositions finish the
proof.
\end{proof}

For a general $S$ this is \texttt{CONDITIONAL}: the condition is not
deduced from positivity, nonzero mass, or full support alone. The
next construction shows that it is nonvacuous in every allowed degree.

\begin{proposition}[Backgrounds with detection in every degree]
\label{prop:r2-faithful-backgrounds}
Assume \eqref{eq:r2-framed-domain}. There is a sequence of normalized
continuous semipositive PL functions $h_j$ on a fixed neighborhood
of $\overline V$ such that, for every $0\leq p<n$, the series
\[
 S_p=\sum_{j\geq1}2^{-j}T_{[1],p}(h_j,\ldots,h_j)
\]
defines a symmetric positive separately closed original current on
$V$ satisfying \eqref{eq:r2-relative-atomic-condition}. Here
$T_{[1],0}=[1]$. More generally, any positive closed background $S$
of the same degree with $S-\epsilon S_p$ positive for some
$\epsilon>0$ satisfies that condition.

The measure
\[
 \sigma=\sum_{j\geq1}2^{-j}(Dh_j)^n
\]
is a finite positive Radon measure on $V$ with full support. If two
members of $\Cb(V)$ are ordered $\sigma$-almost everywhere, they
are ordered at every point of $V$. In particular this one fixed
measure determines their actual representatives.
\end{proposition}

\begin{proof}
The space $Y^{\mathrm{an}}$ is second countable in the present setting.
Here is the elementary reason needed for this assertion. A finite
extension of $\mathbf Q_p$ has a countable dense subfield. On each
bounded coordinate region, polynomials with coefficients in that
subfield approximate arbitrary $K$-polynomials uniformly. Their
absolute-value functions therefore determine the analytic topology,
and inverse images of intervals with rational endpoints give a
countable basis. Passing to a closed affine subvariety and an ordinary
open preserves second countability.

For each member $O$ of a countable ordinary basis, cover $O$ by
interiors of strict affinoid domains contained in $O$. Good strict
analyticity supplies the cover, and second countability makes $O$
Lindel\"of, so a countable subcover suffices. The resulting countable
family is an affinoid neighborhood basis. Its Shilov determining set
$\mathscr D=\{z_1,z_2,\ldots\}$ is countable. Fix
$\overline V\subset V^+\Subset Y^{\mathrm{an}}$ and choose the
testers $h_j=h_{z_j}$ of Lemma~\ref{lem:r2-atomic-tester}, normalized
on this one $V^+$.

Relative CLN with background $[1]$ bounds the coefficient variations of
$T_{[1],p}(h_j,\ldots,h_j)$ uniformly in $j$ on every fixed compact
finite original presentation in $V^+$. The bounds use only
$0\leq h_j\leq1$ there. Thus the positive series defining $S_p$
converges absolutely in each such coefficient measure. The resulting
measures satisfy the original relations and refinement compatibilities,
and so define an original current. Positivity and symmetry pass term
by term. Closedness follows by applying $d'$ or $d''$ to a fixed
compact smooth test and summing the absolutely convergent pairings.
The same estimate on the compact set $\overline V\subset V^+$
shows that $\sigma(V)<\infty$.

Fix $j$ and let $R_j=S_p-2^{-j}T_{[1],p}(h_j,\ldots,h_j)$.
The current $R_j$ is positive and separately closed. Finite additivity
in the background follows directly from the defining recursion and
linearity of coefficient Borel multiplication. Therefore, for
$q=n-p$,
\[
 \begin{split}
 T_{S_p,q}(h_j,\ldots,h_j)
 &=2^{-j}T_{T_{[1],p}(h_j,\ldots,h_j),q}(h_j,\ldots,h_j)
     +T_{R_j,q}(h_j,\ldots,h_j)\\
 &\geq 2^{-j}(Dh_j)^n.
 \end{split}
\]
The first term is the indicated $n$-fold product by its literal ordered
recursion. In particular it has a positive atom at $z_j$. This proves
\eqref{eq:r2-relative-atomic-condition} for $S_p$. If
$S-\epsilon S_p$ is positive, the same finite decomposition proves it
for $S$. No assertion of continuity under arbitrary varying
backgrounds is used to pass from the series to these products.

Finally, $\sigma(\{z_j\})>0$ for every $j$. The set $\mathscr D$ is
dense because every nonempty ordinary open contains one of the
affinoids in the basis and hence a Shilov point. Thus $\sigma$ has
full support. Almost-everywhere order forces order at each of its
positive atoms, and Lemma~\ref{lem:r2-shilov-gap} then gives order
everywhere. The conclusion uses these particular determining atoms,
not merely the assertion that $\sigma$ has full support.
\end{proof}

This construction and its proof are recorded as
\texttt{NEW\_PROVED} relative to the adopted inputs, without a claim
of bibliographic priority. They exhibit backgrounds to which the
conditional criterion applies. They do not identify all backgrounds
with the desired property.

\subsection{Limits of the conclusion and excluded arguments}

\begin{enumerate}
\item The domain in \eqref{eq:r2-framed-domain} is essential to the
fixed-domain proof. On a connected proper positive-dimensional
analytic space, the global maximum principle makes every finite
global classical psh function constant, so the untwisted ambient
capacity defined using such global functions is zero. Distinct
constants then give nonempty capacity-zero strict gaps, while their
Monge--Amp\`ere measures both vanish. The first-round statements modulo
capacity zero remain true; a pointwise uniqueness assertion on a proper
space needs a normalization. No domain-independent detection theorem
is claimed.

\item A positive closed background may vanish off a proper analytic
subset. In an ordinary ball disjoint from that subset, every product
with this background vanishes by restriction. Nonempty strict gaps
there have zero relative capacity. In particular ambient positivity
does not furnish condition \eqref{eq:r2-relative-atomic-condition}.
The ambient-polar concentration example of the first round is fully
compatible with the present results.

\item Positive capacity on all nonempty ordinary opens, or full
support of one trace measure, is not substituted for strict-gap
detection. A gap of two upper-semicontinuous functions need not be
ordinary open. Lemma~\ref{lem:r2-shilov-gap} is the step that actually
relates its pointwise content to detectable points.

\item Local positivity of capacity on a smaller domain does not by
itself prove positivity for the capacity of a larger prescribed
domain: admissible competitors must extend to the latter. The ample
frame and bounded closure in Lemma~\ref{lem:r2-atomic-tester} were used
precisely to construct the tester on that larger fixed domain.

\item An unconditional pointwise domination or uniqueness theorem for
arbitrary $S$ is \texttt{FALSE}. For example, take $S=0$ on the open
unit disc, $u=\max\{\log|T|,-1\}$, and $v=0$. Both functions are
continuous bounded classical psh; their relative Monge--Amp\`ere
measures vanish. For $0<\epsilon<1$, the ordinary closure of
$\{u<v-\epsilon\}$ is contained in the compact closed disc of radius
$e^{-\epsilon}<1$;
for $\epsilon\geq1$ the set is empty. The reverse gaps are empty.
Thus the compact-gap hypotheses hold in both orders, but $u<v$
throughout the domain. What remains \texttt{OPEN} here is finding
necessary conditions, or weaker useful sufficient conditions, for
backgrounds not covered by
\eqref{eq:r2-relative-atomic-condition}. The atomic condition may be
stronger than necessary. No implication from an unspecified smooth
strict-positivity condition to detection is asserted.
\end{enumerate}

% END source: base_r3/base_r2/sections/gap_detection.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/gap_detection_addendum.tex
\subsection{A connected example with a nonzero background}
\label{subsec:r2-nonzero-background-counterexample}

The failure of unconditional pointwise domination is not limited to
the zero current or disconnected spaces.

\begin{example}[A nonzero connected-domain counterexample]
\label{ex:r2-nonzero-domination-failure}
Let $n\geq2$, let $R>1$, and let
\[
 U=\{\,|z_i|<R:0\leq i<n\,\}\subset(\mathbf A_K^n)^{\mathrm{an}},
 \qquad Y=\{z_0=0\}\cap U,
 \qquad S=[Y].
\]
The open polydisc $U$ is connected and $S$ is a nonzero symmetric
positive separately closed original current of bidegree $(1,1)$.
Choose $0<r<1$ and $0<c<\log(1/r)$, and set
\[
 b=\max\bigl\{0,\log(|z_0-1|/r),
                    \log(|z_1|/r),\ldots,
                    \log(|z_{n-1}|/r)\bigr\},
 \qquad u=b,\quad v=\max\{c,b\}.
\]
Then $u,v$ are continuous bounded classical psh on $U$. They satisfy
the compact-gap condition in both orders and
\[
 T_{S,n-1}(u,\ldots,u)=T_{S,n-1}(v,\ldots,v),
 \qquad
 T_{S,n-1}(u,\ldots,u)(\{u<v\})=0,
\]
but $u<v$ on a nonempty relatively compact ordinary polydisc.
Consequently unconditional pointwise relative domination and
uniqueness are \texttt{FALSE} even for a nonzero background on a
connected domain satisfying \eqref{eq:r2-framed-domain}.
\end{example}

\begin{proof}
Truncation by the constant zero makes the displayed maxima finite and
continuous across the zero sets of the coordinates. Logarithms of
analytic functions are classical psh after all complete extensions,
and finite maxima and positive constant translations preserve that
class. The coordinate bounds on $U$ give actual uniform upper bounds
for $b$; it is bounded below by zero.

Put $\rho=re^c<1$. Then
\[
 \{u<v\}=\{b<c\}
  =\{\,|z_0-1|<\rho,
              |z_i|<\rho\ (1\leq i<n)\,\}.
\]
This polydisc is nonempty. Its closure is contained in the closed
polydisc with the same center and radius. That closed polydisc is
compact and lies in $U$, since $|z_0|=1<R$ on it and
$|z_i|<1<R$ for $i\geq1$.
It is disjoint from $Y$. More precisely, for $0<\epsilon<c$ the gap
$\{u<v-\epsilon\}$ is the same polydisc with radius
$re^{c-\epsilon}<1$, while for $\epsilon\geq c$ it is empty.
The reverse gaps are empty because $v\geq u$. These facts check every
compact-gap quantifier.

On the ordinary neighborhood $\{|z_0|<1\}\cap U$ of $Y$, the
non-Archimedean triangle inequality gives $|z_0-1|=1$. Hence
$b\geq\log(1/r)>c$ there and $u=v$ there. Every recursively defined
product with background $[Y]$ is supported on $Y$: Borel
multiplication does not enlarge the support, and distributional
differentiation does not enlarge it. The products for $u$ and $v$
are equal on this neighborhood by ordinary locality, and vanish
outside $Y$; hence they are equal on $U$. Their common measure assigns
zero mass to the displayed gap because it is disjoint from $Y$.
No base change and no modified calibration enter the example.
\end{proof}

\subsection{Detection by backgrounds constructed from smooth factors}
\label{subsec:r2-smooth-factor-detection}

Atomic detection by $S$-products is a sufficient condition, but the
proof can also use a controlled family of smooth factors approaching
the ambient atomic testers. The following construction makes no claim
about whether its resulting measures have atoms.

Continue to assume \eqref{eq:r2-framed-domain}. Use the determining
set $\mathscr D=\{z_j\}$ and the continuous model testers $h_j$ in
Proposition~\ref{prop:r2-faithful-backgrounds}, normalized on the fixed
$V^+$. For each $j$ choose smooth classical psh functions $g_{j,m}$ on
$Y^{\mathrm{an}}$ such that
\[
 g_{j,m}\downarrow h_j,\qquad
 h_j+4^{-m}\leq g_{j,m}\leq h_j+3\cdot4^{-m}.
 \tag{S}\label{eq:r2-smooth-detectors}
\]
Here $h_j$ denotes its original model weight with the same constant
normalization, also defined on $Y^{\mathrm{an}}$. These approximants
exist on the whole required domain: after taking a generated power,
the model metric is a finite pure-FS maximum of algebraic section
ratios in its specified frame. The smooth regularized maxima of
\textup{R}, Proposition~A.1 have uniformly bounded error everywhere;
zeros of sections are inactive locally. Apply the constant-translation
construction of \textup{R}, Lemma~A.2 to that global uniform error.
The functions remain all-extensions classical by the convex
unit-logarithm description. Set
\[
 f_{j,m}=\tfrac12 g_{j,m}.
\]
Thus $0\leq f_{j,m}\leq1$ on $V^+$, and
$f_{j,m}\downarrow h_j/2$ on the same domain.

\begin{proposition}[A sufficient condition using smooth factors]
\label{prop:r2-smooth-factor-condition}
Let $0\leq p<n$ and $q=n-p$. Suppose that a symmetric positive
separately closed original $(p,p)$-current $S$ on $V$ has the following
property: for every $j,m$ there is a constant $a_{j,m}>0$ such that
\[
 S-a_{j,m}(Df_{j,m})^p\text{ is a positive current on }V.
 \tag{F}\label{eq:r2-smooth-factor-condition}
\]
As usual, the right-hand factor is interpreted as
$(Df_{j,m})^p\wedge[1]$, and equals $[1]$ when $p=0$.
Then every nonempty strict gap of two members of $\Cb(V)$ has
positive $\operatorname{Cap}_S(\,\cdot\,;V)$. The pointwise relative
domination and uniqueness conclusions hold under the first-round
compact-gap hypotheses.
\end{proposition}

\begin{proof}
Finite additivity in the background and positivity give
\[
 T_{S,q}(f_{j,m},\ldots,f_{j,m})
       \geq a_{j,m}(Df_{j,m})^n.
\]
Let $u,v\in\Cb(V)$ and $\delta\geq0$, and suppose that
$E=\{u>v+\delta\}$ has $S$-capacity zero. Since each $f_{j,m}$ is an
admissible capacity tester, $(Df_{j,m})^n(E)=0$ for every $j,m$.
Put $w=\max\{u,v+\delta\}\in\Cb(V)$. The nonnegative Borel
function $w-v-\delta$ is zero off $E$, so
\[
 (w-v-\delta)(Df_{j,m})^n=0.
\]
Fix $j$. Theorem~\ref{thm:relative-monotone}, applied with fixed
background $[1]$ and differentiated slots $f_{j,m}\downarrow h_j/2$,
passes this identity to the limit. The two nonconstant weights $w$
and $v$ are constant sequences in that theorem; the scalar constant
$\delta$ uses its unweighted case. Thus
\[
 (w-v-\delta)(D(h_j/2))^n=0.
\]
All statements are local on compact original tests, so local
boundedness of $u,v,w$ suffices. The limiting measure has a positive
atom at $z_j$, which forces
$w(z_j)-v(z_j)-\delta=0$, or $u(z_j)\leq v(z_j)+\delta$.
Lemma~\ref{lem:r2-shilov-gap} then gives $u\leq v+\delta$ everywhere
on $V$. Hence $E$ was empty. The pointwise comparison consequences
follow from the first-round results modulo capacity zero.
\end{proof}

\begin{corollary}[Explicit backgrounds and a faithful smooth trace measure]
\label{cor:r2-smooth-factor-backgrounds}
For every $0\leq p<n$, the series
\[
 \widehat S_p=\sum_{j,m\geq1}2^{-j-m}(Df_{j,m})^p
\]
defines a positive separately closed original current on $V$ that
satisfies \eqref{eq:r2-smooth-factor-condition}. So does any
background $S\geq\epsilon\widehat S_p$ with $\epsilon>0$.
Moreover,
\[
 \widehat\sigma=\sum_{j,m\geq1}2^{-j-m}(Df_{j,m})^n
\]
is a finite positive Radon measure of full support on $V$, formed from
original smooth top-form trace measures. For $u,v\in\Cb(V)$,
\[
 u\leq v\quad\widehat\sigma\text{-almost everywhere}
 \quad\Longleftrightarrow\quad u\leq v\quad\text{everywhere on }V.
\]
\end{corollary}

\begin{proof}
Uniform coefficient bounds and convergence of the positive series
follow from relative CLN, exactly as in
Proposition~\ref{prop:r2-faithful-backgrounds}, since all $f_{j,m}$
lie in $[0,1]$ on the fixed $V^+$. Removing one summand leaves a
positive separately closed current, which proves
\eqref{eq:r2-smooth-factor-condition} with $a_{j,m}=2^{-j-m}$.
The same argument applies after multiplication by $\epsilon$ and
addition of a positive background.

If $u\leq v$ almost everywhere for $\widehat\sigma$, the positive
coefficients in its defining sum imply
$(Df_{j,m})^n(\{u>v\})=0$ for every $j,m$. Apply the weighted limit
argument in the preceding proof with $\delta=0$. It gives pointwise
order on $\mathscr D$, hence everywhere. The reverse implication is
immediate.

Finally take a nonempty ordinary open $O\subset V$, choose
$z_j\in O\cap\mathscr D$, and choose a nonnegative original smooth
cutoff supported in $O$ and equal to one near $z_j$. Unweighted
monotone convergence gives
\[
 \int\chi(Df_{j,m})^n
     \longrightarrow 2^{-n}\int\chi(Dh_j)^n>0.
\]
Some summand assigns positive mass to $O$, and consequently
$\widehat\sigma(O)>0$. This proves full support without asserting
any atomic or non-atomic description of these smooth-factor products.
\end{proof}

The counterexample above and this construction are
\texttt{NEW\_PROVED} relative to the adopted inputs. The implication
under condition \eqref{eq:r2-smooth-factor-condition} is a
\texttt{DIRECT\_COROLLARY} of fixed-background weighted convergence
and the established Shilov detectors. These are sufficient conditions;
a necessary characterization of pointwise-faithful backgrounds is
not proved here.

% END source: base_r3/base_r2/sections/gap_detection_addendum.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/weighted_l1.tex
\section{A relative weighted $L^1$ Chern--Levine--Nirenberg estimate}
\label{sec:r2-weighted-l1}

The background is fixed throughout. Let $X$ be an allowed analytic
space over the prescribed finite extension $K/\mathbf Q_p$, of pure
dimension $n$, and let $S$ be a symmetric, original-positive,
separately $d'$- and $d''$-closed current of bidegree $(p,p)$.
Put $q=n-p$ and $D=d'd''$. All products, coefficient measures and
tests below belong to the original CLD calculus. No extension of $S$
to the additional test forms in CLD 2025 is used.

We prove a relative version of the weighted-form argument in
CLD 2025, Proposition 8.4.3. The passage to bounded classical
potentials uses the fixed-background calculus proved in R1. The
result is recorded as \texttt{NEW\_PROVED} in R2, with this method
and these dependencies stated explicitly; no claim of literature
priority is made.

\subsection{The positive cutoff forms}

Let $O$ be an ordinary open set and let $\rho$ be an original smooth
psh function on $O$, meaning that $A=D\rho$ is a positive
$(1,1)$-form. Suppose that
\[
 V=\{\rho<0\}\Subset O.
\]
Write $t=(-\rho)_+$ and $B=d'\rho\wedge d''\rho$. Fix $a=4$.
For $0\leq k\leq q$, put
\begin{equation}
\label{eq:r2-cutoff-forms}
 \beta_k=t^{k+a}A^{q-k}
       +(k+a)t^{k+a-1}B\wedge A^{q-k-1},
\end{equation}
where the second term is omitted for $k=q$. The forms are extended
by zero off $V$. They have compact support in $\overline V$, are
strongly positive, and have continuous coefficients. In fact their
coefficients are of class $C^2$ on each original moment presentation:
the smallest power occurring is $t^3$.

The coefficient-Radon property of $S$ defines $\beta_k\wedge S$ and
its products with smooth forms. Integration by parts with these
$C^2$ cutoff forms is legitimate. To check this within the original
test category, approximate their one-variable coefficient functions
in $C^2$ by smooth functions of $\rho$, and multiply by one fixed
original smooth cutoff equal to one near $\overline V$.
The approximants have one compact support. The coefficient measures
permit passage in every term involving at most two derivatives.
The approximating forms need not be positive; positivity of the
limiting forms follows directly from \eqref{eq:r2-cutoff-forms}.

On $V$, set $b=k+a$, $r=q-k$, and
\[
 c_k=\begin{cases}2,&k<q,\\1,&k=q.\end{cases}
\]
For $1\leq k\leq q$, direct differentiation gives
\begin{align}
 d''\beta_k
   &=-c_k b\,t^{b-1}d''\rho\wedge A^r,
       \label{eq:r2-dbeta}\\
 D\beta_k
   &=c_k b\{-t^{b-1}A^{r+1}
                 +(b-1)t^{b-2}B\wedge A^r\}.
       \label{eq:r2-ddbeta}
\end{align}
The endpoint $k=q$ is distinguished because $\beta_q=t^{q+a}$
has no second summand. The following proof uses both cases explicitly.

\begin{lemma}[Smooth relative weighted estimate]
\label{lem:r2-smooth-weighted-cln}
Assume $q\geq1$. Let $F\leq0$ and $u_1,\ldots,u_k$ be original
smooth psh functions on a neighborhood of $\overline V$, where
$0\leq k\leq q$. Put $M_i=\sup_V|u_i|$. Then
\begin{equation}
\label{eq:r2-smooth-l1}
 \int(-F)\,\beta_k\wedge Du_1\wedge\cdots\wedge Du_k\wedge S
 \leq C(k,a)\prod_{i=1}^kM_i
                 \int(-F)\,\beta_0\wedge S,
\end{equation}
where $C(0,a)=1$ and
\begin{equation}
\label{eq:r2-constant}
 C(k,a)=\prod_{i=1}^{k}
           \frac{4e(a+i)(a+i+1)}{a+i-1}.
\end{equation}
The constants do not depend on $S$, $F$, or the chosen functions.
\end{lemma}

\begin{proof}
All integrals have compact support. Write
$\Psi_{k-1}=Du_1\wedge\cdots\wedge Du_{k-1}$ and denote the
left-hand side of \eqref{eq:r2-smooth-l1} by $J_k$.
We first estimate the last slot when its function $u$ is strictly
positive on a neighborhood of $\overline V$. Twice integrating by
parts against the closed current $\Psi_{k-1}\wedge S$ gives
\begin{equation}
\label{eq:r2-integration-identity}
 J_k
 =\int\bigl\{u(FD\beta_k-DF\wedge\beta_k)
              +2F\,d'u\wedge d''\beta_k\bigr\}
                       \wedge\Psi_{k-1}\wedge S.
\end{equation}
Here and below cross terms can be replaced by their symmetric parts
because the remaining current is symmetric. For example, the two
cross terms in $D(F\beta_k)$ have equal pairings. Stokes applied to
$Fu\,d''\beta_k\wedge\Psi_{k-1}\wedge S$ yields
\eqref{eq:r2-integration-identity}, with the displayed signs.

Put $w=-F\geq0$ and
\[
 \Gamma(f,g)=\tfrac12(d'f\wedge d''g+d'g\wedge d''f),
 \qquad \Gamma(f)=d'f\wedge d''f.
\]
For every positive scalar $h$, the positive-square identity gives
\[
 2\Gamma(u,\rho)\leq h^{-1}\Gamma(u)+hB.
\]
Apply this pointwise on $V$ with $h=c_kbu/t$. Equations
\eqref{eq:r2-dbeta}--\eqref{eq:r2-integration-identity}, and
positivity of $uDF\wedge\beta_k$, give
\begin{align*}
 J_k
 &\leq\int wu\{-D\beta_k+c_k^2b^2t^{b-2}B\wedge A^r\}
                          \wedge\Psi_{k-1}\wedge S\\
 &\hspace{1cm}
       +\int w\,t^b u^{-1}\Gamma(u)\wedge A^r
                          \wedge\Psi_{k-1}\wedge S.
\end{align*}
The form in braces is
\[
 c_kb\,t^{b-1}A^{r+1}
  +c_kb\bigl((c_k-1)b+1\bigr)t^{b-2}B\wedge A^r.
\]
Since $c_k$ is either $1$ or $2$, it is bounded above by
\[
 C_k\beta_{k-1},\qquad
 C_k=\frac{2b(b+1)}{b-1}.
\]
Also $t^bA^r\leq\beta_k$. Wedging each displayed inequality with
the other strongly positive factors and then with the
original-positive current $S$ preserves it. We obtain
\begin{equation}
\label{eq:r2-positive-slot}
 J_k(F;u_1,\ldots,u_{k-1},u)
 \leq C_k\|u\|_{V}J_{k-1}
       +\int w\,\beta_k\wedge\Psi_{k-1}\wedge
                          u^{-1}\Gamma(u)\wedge S.
\end{equation}

Now let $u_k$ have arbitrary sign and set $M=\sup_V|u_k|$.
If $M=0$, its curvature vanishes on $V$ and there is nothing to
prove. Otherwise put $z=\exp(B_0u_k)$, where $B_0>0$. Then
\[
 Dz=z^{-1}\Gamma(z)+B_0z\,Du_k
       \geq z^{-1}\Gamma(z)+B_0e^{-B_0M}Du_k.
\]
Apply \eqref{eq:r2-positive-slot} to $z$. The energy term cancels
with the identical term in this lower bound for $Dz$, leaving
\[
 B_0e^{-B_0M}J_k\leq C_ke^{B_0M}J_{k-1}.
\]
Choose $B_0=(2M)^{-1}$. This gives
\[
 J_k\leq\frac{4e(a+k)(a+k+1)}{a+k-1}\,M_kJ_{k-1}.
\]
Induction from $J_0$ proves the assertion.
\end{proof}

\subsection{Bounded potentials and negative psh witnesses}

We now work in a common algebraic norm-box neighborhood of the kind
used in R1. The compact set $\overline V$ lies in a prescribed
smaller norm box, and all potentials are defined on a larger one.
Thus the compact-box regularization lemma of R1 gives smooth
decreasing classical psh approximants on a neighborhood of the
entire $\overline V$. This domain condition is only local geometry;
it places no restriction on the positive closed current $S$.

For a negative extended-real classical psh function
$\phi:O\to[-\infty,0]$, put
\[
 \phi_N=\max\{\phi,-N\}\in\Cb(O),
 \qquad
 A_{S,\rho}(\phi)=\int(-\phi)\,\beta_0\wedge S\in[0,\infty].
\]
This is an integral of the actual Borel function against one
specified positive Radon measure. It is not an ambient-volume
integrability condition.

\begin{theorem}[Uniform truncation estimate over an arbitrary background]
\label{thm:r2-weighted-l1}
Assume $q\geq1$ and $A_{S,\rho}(\phi)<\infty$.
Let $K\Subset V$ and let $u_1,\ldots,u_q\in\Cb(O)$ have
pointwise bounds $|u_i|\leq M_i$ on a fixed neighborhood of
$\overline V$. Then, with
$\delta_K=\min_K(-\rho)>0$,
\begin{equation}
\label{eq:r2-uniform-truncation}
 \sup_{N\geq1}\int_K(-\phi_N)
       T_{S,q}(u_1,\ldots,u_q)
 \leq \delta_K^{-(q+a)} C(q,a)
             \prod_{i=1}^qM_i\,A_{S,\rho}(\phi).
\end{equation}
In particular, $\phi$ is integrable on $K$ for every such mixed
measure, and the same bound holds with $-\phi$ in place of
$-\phi_N$.
\end{theorem}

\begin{proof}
First fix $N$ and $\epsilon>0$. The bounded classical psh function
$F=\phi_N-\epsilon$ is at most $-\epsilon$.
Choose a smooth decreasing row $F_j\downarrow F$ on one
neighborhood of $\overline V$. After a prefix is discarded,
$F_j\leq-\epsilon/2$ there. This follows from compactness applied
to the increasing open sets where the continuous majorants are
less than $-\epsilon/2$.

Simultaneously choose smooth decreasing classical psh rows
$u_{i,j}\downarrow u_i$ on the same neighborhood. Given
$\varepsilon>0$, a prefix makes $|u_{i,j}|\leq M_i+\varepsilon$
on $\overline V$. The lower bound follows from $u_{i,j}\geq u_i$;
the upper bound is the same compactness argument. All choices are
made for this fixed $N$. No regularization theorem for the
unbounded function $\phi$ is assumed.

Apply Lemma~\ref{lem:r2-smooth-weighted-cln} with $k=q$.
The R1 weighted decreasing theorem, in top degree against
continuous compact scalar tests, passes the left side to
\[
 \int(-\phi_N+\epsilon)\,\beta_q\,
                          T_{S,q}(u_1,\ldots,u_q).
\]
On the right, bounded convergence is against the \emph{fixed}
measure $\beta_0\wedge S$. Let $\varepsilon\downarrow0$ and
then $\epsilon\downarrow0$. We obtain
\[
 \int(-\phi_N)\,\beta_q\,
                          T_{S,q}(u_1,\ldots,u_q)
 \leq C(q,a)\prod_iM_i\int(-\phi_N)\,\beta_0\wedge S.
\]
Because $\beta_q=t^{q+a}\geq\delta_K^{q+a}$ on $K$ and
$-\phi_N\leq-\phi$, this is
\eqref{eq:r2-uniform-truncation}. Finally $-\phi_N\uparrow-\phi$.
Monotone convergence for each fixed mixed measure proves the
integrability assertion and its bound.
\end{proof}

\begin{corollary}[Uniformity over normalized potentials]
\label{cor:r2-normalized-witness}
Under the same hypotheses,
\[
 \sup_{N\geq1}\ \sup_{\substack{v\in\Cb(O)\\0\leq v\leq1}}
 \int_K(-\phi_N)\,T_{S,q}(v,\ldots,v)
 \leq \delta_K^{-(q+a)}C(q,a)A_{S,\rho}(\phi).
\]
The same assertion holds with different normalized potentials in
the differentiated slots.
\end{corollary}

\begin{corollary}[A precise relative polar conclusion]
\label{cor:r2-relative-polar}
For the capacity with the \emph{fixed admissibility domain} $O$,
\[
 \operatorname{Cap}_{S}(E;O)
   =\sup_{\substack{v\in\Cb(O)\\0\leq v\leq1}}
          \int_E T_{S,q}(v,\ldots,v),
\]
one has, for every $N>0$,
\[
 \operatorname{Cap}_{S}\bigl(K\cap\{\phi\leq-N\};O\bigr)
 \leq \frac{\delta_K^{-(q+a)}C(q,a)}{N}
                   A_{S,\rho}(\phi).
\]
Hence $K\cap\{\phi=-\infty\}$ has zero relative capacity and
zero mass for every bounded mixed measure in
Theorem~\ref{thm:r2-weighted-l1}.
\end{corollary}

\begin{proof}
On the indicated set, $-\phi\geq N$. Apply the integrability
bound in Theorem~\ref{thm:r2-weighted-l1} and take the supremum.
The pole set is contained in every such level set. Its zero mass
also follows directly from the proved integrability of $\phi$.
\end{proof}

\begin{corollary}[Outer capacity and quasicontinuity of an integrable witness]
\label{cor:r2-witness-outer-capacity}
Keep the hypotheses of Theorem~\ref{thm:r2-weighted-l1}, and let
$W$ be ordinary open with $\overline W\subset K\Subset V$.
Define outer capacity by
\[
 \operatorname{Cap}_S^*(E;O)
  =\inf_{\substack{G\subset O\text{ ordinary open}\\E\subset G}}
                  \operatorname{Cap}_S(G;O).
\]
Write $C_K=\delta_K^{-(q+a)}C(q,a)A_{S,\rho}(\phi)$.
Then, for $N>0$,
\begin{equation}
\label{eq:r2-open-sublevel-bound}
 \operatorname{Cap}_S(W\cap\{\phi<-N\};O)\leq C_K/N.
\end{equation}
Consequently $W\cap\{\phi=-\infty\}$ has outer capacity zero.
The truncations $\phi_N$ converge to $\phi$ locally in outer
capacity: for every $\delta>0$,
\[
 \operatorname{Cap}_S^*
       (W\cap\{\phi_N-\phi>\delta\};O)\leq C_K/N
                  \longrightarrow0,
\]
where the difference is $+\infty$ at a pole of $\phi$.
Finally, for every $\epsilon>0$ there is an ordinary open
$G\subset O$ with $\operatorname{Cap}_S(G;O)<\epsilon$ such
that $\phi|_{W\setminus G}$ is finite and continuous.
These are \texttt{DIRECT\_COROLLARY} assertions; no outer-regularity
theorem for arbitrary Borel sets is assumed.
\end{corollary}

\begin{proof}
The function $\phi$ is upper semicontinuous, so
$H_N=W\cap\{\phi<-N\}$ is ordinary open. It is contained in
$K\cap\{\phi\leq-N\}$, so
Corollary~\ref{cor:r2-relative-polar} proves
\eqref{eq:r2-open-sublevel-bound}. Each $H_N$ contains the pole
set in $W$, giving its zero outer capacity. Since $\phi_N=\phi$
off $\{\phi<-N\}$, the same ordinary open set covers every
displayed truncation exceptional set.

For the last assertion choose $N$ with $C_K/N<\epsilon/2$.
The bounded classical function $\phi_N$ is quasicontinuous by R1.
Precisely, cover $\overline W$ by finitely many smaller norm-box
cores in $O$ and apply the bounded local quasicontinuity theorem
on each larger box with a total error less than $\epsilon/2$.
Restriction of admissible functions gives
$\operatorname{Cap}_S(E;O)\leq\operatorname{Cap}_S(E;\Omega)$
for sets in any of these boxes $\Omega\subset O$. Thus the union
$G_N\subset O$ of their ordinary open exceptional sets has
$\operatorname{Cap}_S(G_N;O)<\epsilon/2$, and $\phi_N$ is
continuous on $W\setminus G_N$. Put $G=G_N\cup H_N$.
Countable subadditivity bounds its capacity by $\epsilon$.
On $W\setminus G$ one has $\phi=\phi_N\geq-N$, proving
the assertion in the subspace topology.
\end{proof}

\subsection{Intrinsic local hypotheses and scope}

One sufficient hypothesis, independent of a particular barrier, is
local coefficient integrability of $\phi$ with respect to $S$:
\begin{equation}
\label{eq:r2-coefficient-integrability}
 \int|\phi|\,d\mu_{S,\Theta}<\infty
 \quad\hbox{for every positive original compact smooth
 $(q,q)$-test $\Theta$.}
\end{equation}
The finite positive controls of R imply that this is equivalent
to integrability against all local coefficient variations.
They also dominate the continuous-coefficient compact form
$\beta_0$ by finitely many positive smooth controls. Therefore
\eqref{eq:r2-coefficient-integrability} implies
$A_{S,\rho}(\phi)<\infty$ for every barrier used above.

The R1 collar construction supplies the required barriers locally:
take a smooth psh function $b$ which is zero on the inner norm box
and large on an outer collar, and put $\rho=b-c$ with $c>0$
between the two levels. Then $\{\rho<0\}\Subset O$ and contains
the chosen core. No strict positivity of $D\rho$ is required.
A finite covering gives the stated integrability and mass bounds
on arbitrary compact subsets for which
\eqref{eq:r2-coefficient-integrability} holds.

Lower-degree coefficient estimates follow by inserting bounded
smooth quadratic moment potentials into the remaining slots and
using R's finite-test domination. Thus the same hypothesis makes
$\phi T_{S,k}(u_1,\ldots,u_k)$ a coefficient-Radon current for
every bounded tuple and every $k\leq q$.

For $q=0$, the asserted uniform truncation bound is simply
integrability of $\phi$ against the fixed top-degree measure $S$;
no differentiated slot is present. Finally, the capacity conclusion
keeps its admissibility domain $O$. To obtain the corresponding
statement on a smaller domain, choose its collar and verify its
integrability condition there; monotonicity under domain restriction
does not transfer the displayed upper bound automatically.

The hypothesis on $\phi$ and $S$ is essential to the scope of this
result. If $S=[Y]$ and $\phi$ is identically $-\infty$ on $Y$,
it is generally not satisfied. Nothing here asserts that arbitrary
relative products avoid every ambient pluripolar set. Nor does the
uniform bound alone establish weighted convergence when both an
unbounded witness and its background measure vary.


% END source: base_r3/base_r2/sections/weighted_l1.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/polar_restriction.tex
\section{Restriction of a positive closed current to a complete pole set}
\label{sec:r2-polar-restriction}

Let $S$ be a symmetric original-positive, separately $d'$- and
$d''$-closed current of bidegree $(p,p)$ on an allowed ordinary
open $U$, and put $q=n-p$. Let $\phi:U\to[-\infty,\infty)$ be
an extended-real classical psh function, and set
\[
 P=\{\phi=-\infty\}.
\]
All restrictions in this section mean Borel multiplication of the
actual coefficient-Radon measures. In particular, they are not
restrictions to a tropical image. The proof uses the bounded
capacitary closure developed in this record and the original CLD
positive cone of R throughout.

\subsection{A first-order estimate on a fixed test presentation}

For a bounded classical psh function $a$, define the distribution
\begin{equation}
\label{eq:r2-relative-gamma-current}
 \Gamma_S(a)=\tfrac12D(a^2S)-aD(aS).
\end{equation}
It is a positive coefficient-Radon current of bidegree $(p+1,p+1)$.
Indeed, locally add a constant so that $a\geq0$ and use a decreasing
smooth classical psh row. The square of each term is classical psh
and decreases to $a^2$. The first-round weighted theorem passes
\eqref{eq:r2-relative-gamma-current} to the limit of the positive
currents $d'a_j\wedge d''a_j\wedge S$. Adding the constant does
not change \eqref{eq:r2-relative-gamma-current}, since $S$ is
separately closed. Positivity and R's coefficient-Radon theorem
give the assertion. Wedging with a fixed closed smooth trace form
recovers the first-round relative energy measures.

\begin{lemma}[First-order relative Cauchy--Schwarz estimate]
\label{lem:r2-first-order-CS}
Suppose $q\geq1$. Localize an original compact smooth test
$\eta$ of bidegree $(q-1,q)$ to one finite moment presentation
with coordinates $t_1,\ldots,t_r$. Let
\[
 \beta=\sum_{a=1}^r d't_a\wedge d''t_a,
\]
and choose a nonnegative compact original smooth cutoff $\chi$
equal to one near $\operatorname{supp}\eta$. There is a constant
$C_\eta<\infty$, independent of $S$ and of the bounded classical
psh function $a$, such that
\begin{equation}
\label{eq:r2-first-order-CS}
 \bigl|\langle d'(aS),\eta\rangle\bigr|^2
 \leq C_\eta
       \langle\Gamma_S(a),\chi\beta^{q-1}\rangle
       \langle S,\chi\beta^q\rangle.
\end{equation}
The same assertion holds for $d''$ and tests of bidegree $(q,q-1)$.
The constant remains valid when the moment presentation is refined
to include additional coordinates used by $a$ or its approximants.
\end{lemma}

\begin{proof}
First let $a$ be smooth and write $\xi=d'a$.
The finite expansion of $\eta$ has terms
$f_{I,J}d't_I\wedge d''t_J$, where $|I|=q-1$ and $|J|=q$.
Under the standard symmetric-matrix convention for original
$(q,q)$-forms, the symmetric part of $\xi\wedge\eta$ is a
finite sum, with fixed signs, of the symmetric tensors between
\[
 A_I=\xi\wedge d't_I,\qquad B_J=d't_J.
\]
For two arbitrary $q$-covectors $A,B$, the square
$(\sqrt\epsilon A\pm\epsilon^{-1/2}B)^2$ is in the original
positive cone. The associated matrix Young inequality therefore
bounds either sign of their symmetric cross tensor by
$\epsilon A^2+\epsilon^{-1}B^2$, up to a fixed harmless factor.
Apply it to the finitely many terms and use the fixed bounds of
the coefficients $f_{I,J}$ on the test support. Summing the
$A_I^2$ terms gives a fixed multiple of
$d'a\wedge d''a\wedge\beta^{q-1}$; summing the $B_J^2$ terms
is bounded by a fixed multiple of $\beta^q$.
After rescaling $\epsilon$, we obtain, for every $\epsilon>0$,
\[
 \pm\operatorname{Sym}(d'a\wedge\eta)
 \leq \epsilon\chi\,d'a\wedge d''a\wedge\beta^{q-1}
             +\frac{C'\!_\eta}{\epsilon}\chi\beta^q.
\]
Outside the test support the left side is zero. The cutoff makes
the right side a compact positive test. No strict positivity in
additional analytic directions is used: $A_I$ contains the full
covector $d'a$, and only its remaining $q-1$ factors come from
the fixed frame. The same matrix-square argument works on every
common refinement, with the same finite index sets and constants.

The current $S$ annihilates the antisymmetric component, by the
original-positive coefficient argument in R. Evaluation and
separate closedness give
\[
 |\langle d'(aS),\eta\rangle|
 \leq\epsilon E+\epsilon^{-1}C'\!_\eta M,
 \quad
 E=\langle\Gamma_S(a),\chi\beta^{q-1}\rangle,
 \quad M=\langle S,\chi\beta^q\rangle.
\]
Optimizing proves \eqref{eq:r2-first-order-CS}; if $E$ or $M$
is zero, take a limit in $\epsilon$ to obtain the same conclusion.

For bounded classical $a$, choose a smooth decreasing row on
a fixed neighborhood of the compact test stage, as in R1.
Multiplication against the fixed coefficient measures of $S$
gives $a_jS\to aS$. The weighted first-round theorem gives
$\Gamma_S(a_j)\to\Gamma_S(a)$, as explained above.
Pass to the limit in the smooth inequality. The involution $J$,
or the identical calculation with primed factors interchanged,
proves the $d''$ assertion. A finite partition handles every
original compact test, not only those in one initially fixed chart.
\end{proof}

\subsection{The idempotent identity}

\begin{lemma}[Cubic chain identity in the bounded capacitary closure]
\label{lem:r2-cubic-chain}
Let $0\leq a\leq M$ be a member of $\mathcal A_{S,b}$, locally
with a common bound for its defining approximants. Then
\begin{equation}
\label{eq:r2-cubic-chain}
 D(a^3S)=3aD(a^2S)-3a^2D(aS).
\end{equation}
The square and cube belong to the same local capacitary closure.
Moreover \eqref{eq:r2-relative-gamma-current} is a positive
coefficient-Radon current, and it is continuous under bounded
outer-capacity convergence of classical approximants.
\end{lemma}

\begin{proof}
For a smooth function, the usual product rule gives
\[
 D(a^2S)=2aD(aS)+2d'a\wedge d''a\wedge S,
\]
\[
 D(a^3S)=3a^2D(aS)+6a\,d'a\wedge d''a\wedge S,
\]
which imply \eqref{eq:r2-cubic-chain}.
For bounded nonnegative classical $a$, choose decreasing smooth
nonnegative classical approximants. Their squares and cubes are
again classical psh and decrease to the corresponding powers.
Pass the two weighted one-slot terms in
\eqref{eq:r2-cubic-chain} by the first-round weighted theorem;
pass the left side distributionally against the fixed $S$.

For a nonnegative member of the capacitary closure, replace its
classical defining approximants by their maxima with zero. The
approximants remain uniformly bounded and converge in outer
capacity. Squaring and cubing are Lipschitz on that bounded
range, so their powers also converge in outer capacity. They
are classical psh by the preceding paragraph and decreasing
closure, hence define $a^2,a^3$ in the capacitary class.
The completed weighted theorem now passes every term of the
classical identity. The same argument passes
$\tfrac12D(a_j^2S)-a_jD(a_jS)$ to
\eqref{eq:r2-relative-gamma-current}. Positivity passes from the
classical energy currents, and R gives coefficient-Radon type.
\end{proof}

\begin{theorem}[Positive closed restriction to a complete pole set]
\label{thm:r2-polar-restriction}
For the arbitrary background $S$ and the complete pole set $P$
specified above, both currents
\[
 S_P=\mathbf1_P S,\qquad S_{P^c}=\mathbf1_{U\setminus P}S
\]
are symmetric, original-positive, and separately $d'$- and
$d''$-closed. For every $0\leq k\leq q$ and
$u_1,\ldots,u_k\in\Cb(U)$, one has
\begin{equation}
\label{eq:r2-polar-restriction-BT}
 \mathbf1_P T_{S,k}(u_1,\ldots,u_k)
       =T_{S_P,k}(u_1,\ldots,u_k),
\end{equation}
and the analogous identity for $P^c$, on every original compact
smooth complementary test form.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}, relative to R1 and the
proved R2 capacitary calculus.}
This is a restriction theorem for a current already defined on
$U$. It is not an extension theorem for a current initially
defined only on $U\setminus P$.

\begin{proof}
Positivity, symmetry and coefficient-Radon type of both Borel
restrictions are immediate. In top degree $q=0$, separate
closedness is automatic and there are no differentiated slots.
Suppose therefore $q\geq1$.

Work locally and subtract a constant from $\phi$ so that
$\phi\leq0$ on the neighborhood. Set
\[
 w_N=1+\frac{\max\{\phi,-N\}}{N},\qquad N\geq1.
\]
Then $0\leq w_N\leq1$, the functions are classical psh, and
they increase pointwise to the actual Borel function
$w=\mathbf1_{P^c}$. The local increasing-convergence theorem
puts $w$ in $\mathcal A_{S,b}$ and gives outer-capacity
convergence of $w_N$ to $w$. Its square and cube equal itself.
The cubic identity therefore gives
\[
 D(wS)=3wD(wS)-3wD(wS)=0,
 \qquad
 \Gamma_S(w)=\tfrac12D(wS)-wD(wS)=0.
\]
Completed weighted convergence gives
\[
 w_NS\longrightarrow wS,
 \qquad \Gamma_S(w_N)\longrightarrow0.
\]
Fix an arbitrary first-order original compact test and localize
it to the presentations of Lemma~\ref{lem:r2-first-order-CS}.
The fixed factor $\langle S,\chi\beta^q\rangle$ is finite,
whereas the energy factor tends to zero. The lemma implies
$d'(w_NS)\to0$ on that test. Distributional continuity also
gives $d'(w_NS)\to d'(wS)$, so $d'(wS)=0$.
The same proof gives $d''(wS)=0$. Subtracting from the separately
closed $S$ proves closedness of $S_P$. These local conclusions
glue because the Borel restrictions have already been defined
globally.

For \eqref{eq:r2-polar-restriction-BT}, first take $k=1$ and
a smooth classical psh function $u$. Separate closedness just
proved and smooth Borel multiplication give
\[
 D(uS_P)=Du\wedge S_P
              =\mathbf1_P(Du\wedge S)=\mathbf1_PD(uS).
\]
For general $u\in\Cb$, use a decreasing smooth classical row
$u_j\downarrow u$ on a fixed compact norm-box stage. It has
one common bound on the slightly larger domain. The left side
converges by the fixed-background first-round theorem for $S_P$.
The currents $D(u_jS)$ converge to $D(uS)$ and satisfy uniform
coefficient bounds by the top relative capacity of that same
domain. The multiplier $\mathbf1_P=1-w$ is quasicontinuous for
this capacity because $w\in\mathcal A_{S,b}$. The R2
quasicontinuous multiplier theorem therefore passes the right
side to $\mathbf1_PD(uS)$. This proves the one-slot identity
on all original tests without asserting stability for arbitrary
varying backgrounds.

For several slots, apply the one-slot identity to the arbitrary
positive closed background
$Q=T_{S,k-1}(u_2,\ldots,u_k)$. The restriction theorem already
applies to this $Q$ and the same classical pole set. Induction
gives
\[
 \mathbf1_PQ=T_{S_P,k-1}(u_2,\ldots,u_k),
\]
and hence
\[
 \mathbf1_PD(u_1Q)=D(u_1\mathbf1_PQ)
                         =T_{S_P,k}(u_1,\ldots,u_k).
\]
The complement follows in the same way, or by finite linearity
in the decomposition $S=S_P+S_{P^c}$.
\end{proof}

\begin{corollary}[Exact preservation of pole-set mass]
\label{cor:r2-pole-mass-preservation}
If $\mathbf1_PS=0$, then
$\mathbf1_PT_{S,k}(u_1,\ldots,u_k)=0$ for every bounded
classical tuple. This conclusion needs no integrability estimate
for the original witness. Conversely, the theorem allows nonzero
$S_P$; it does not assert that arbitrary relative products avoid
ambient pole sets.
\end{corollary}

\begin{proof}
Apply \eqref{eq:r2-polar-restriction-BT} with $S_P=0$.
\end{proof}

\paragraph{The ambient vanishing needs a separate argument.}
For $S=[1]$, the theorem shows that $\mathbf1_P[1]$ is a
positive separately closed original current of degree zero.
This calculation alone does not imply its vanishing. Under the
connectedness and nontriviality assumptions on the witness,
Theorem~\ref{thm:r2-ambient-polar-traces} proves that vanishing
by an independent argument using divisorial points and bounded
model approximations. No distributional elliptic regularity is invoked.
CLD 2025, Remark 7.8.3, explicitly distinguishes the equation
$D[u]=0$ from a continuity theorem on its general tropical
spaces, and Proposition 8.1.12 retains local integrability as
an extra condition in its psh class. The proof in the cited
ambient trace theorem supplies the additional analytic input,
rather than silently interpreting a closed degree-zero density
as a continuous pluriharmonic function.

% END source: base_r3/base_r2/sections/polar_restriction.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/divisorial_poles.tex
\section{Divisorial points are not poles of a nontrivial classical psh function}
\label{sec:r2-divisorial-poles}

The field, ordinary topology, and all-extensions classical definition
remain fixed. The geometric lemma below uses strict affinoid domains
inside the algebraic neighborhoods supplied by \textup{B}. Its
normalization step is essential: a formal model with only one
irreducible special-fiber component need not have a unique Shilov
point unless the corresponding generic-point fiber is controlled.

\begin{lemma}[A domain with one prescribed Shilov point]
\label{lem:r2-one-shilov-domain}
Let $U$ be an ordinary open in an allowed good, Hausdorff,
boundaryless, quasi-smooth, pure $n$-dimensional $K$-analytic space,
with $n\geq1$. If $z\in U$ is divisorial, there exists a strict
affinoid domain $A\subset U$ such that
\[
 \Gamma(A)=\{z\},\qquad \operatorname{Int}_U(A)\ne\varnothing.
\]
The lemma does not assert that $z$ itself belongs to this ordinary
interior. In fact it belongs to the boundary of $A$.
\end{lemma}

\begin{proof}
Work in a strict algebraic neighborhood of $z$, as permitted by
\textup{B}, and choose a strict affinoid neighborhood $C\subset U$
of $z$. Divisoriality is unchanged on passing to $C$, by
\cite[Corollary~A.6, p.~367]{GM19}. By the definition of a divisorial
point, equivalently \cite[Proposition~A.5, pp.~366--367]{GM19}, there
is an admissible formal $K^\circ$-model $\mathfrak C$ of $C$ such
that the reduction of $z$ is the generic point of a special-fiber
component.

We may and do make this formal model normal. Indeed, $C$ is normal
because it is quasi-smooth. The formal affine rings are topologically
of finite type over the excellent complete discrete valuation ring
$K^\circ$. Their normalizations in their reduced generic fibers are
finite, and glue to a normal admissible formal model. The generic
fiber remains $C$, since it is already normal. The reduction of $z$
on this new model is still generic in the special fiber: it lies
over the former generic point under a finite integral map, and a
prime above a minimal prime over the uniformizer is again minimal
over that uniformizer, by incomparability. Denote the new generic
point by $\eta$ and its component by $E$.

Remove the other irreducible components of the special fiber and
choose a formal affine open neighborhood $\mathfrak A$ of $\eta$
inside the remaining formal open. Such a choice is possible because
the formal model is locally noetherian and $\eta$ lies on no other
component; one first works in an affine neighborhood and then avoids
its finitely many other components. The special fiber of
$\mathfrak A$ has exactly one irreducible component, an open part
of $E$. Set $A=\mathfrak A_\eta$, its generic fiber. Formal open
immersions induce analytic-domain inclusions on generic fibers, so
$A$ is a strict affinoid domain of $C$, hence of $U$, and it contains
$z$.

The model $\mathfrak A$ is normal and its generic fiber is reduced.
The uniqueness of the point over its special-fiber generic point is
the normal-model assertion in
\cite[Proposition~6.9.7, pp.~119--120]{CLD12}. It can also be read
directly at the height-one local ring of the normal model: the
resulting discrete valuation, normalized by the prescribed absolute
value of the uniformizer, determines that point. By
\cite[Proposition~A.3, p.~365]{GM19}, the points reducing to generic
points of this formal special fiber are exactly the Shilov boundary
of $A$. Hence $\Gamma(A)=\{z\}$.

We check the ordinary-interior assertion separately. The special
fiber of $\mathfrak A$ has a closed point $\xi$, and the reduction
map is surjective, so some $y\in A$ reduces to $\xi$. The reduced
closure of $\xi$ is a finite scheme over $\widetilde K$, hence is
proper. The formal reduction criterion
\cite[Lemma~6.5.1, p.~105]{CLD12} therefore gives
$y\in\operatorname{Int}(A/K)$. Equivalently, one can use the
analytic Nullstellensatz and the fact that rigid points of a strict
analytic space over a nontrivially valued field do not belong to its
analytic boundary; see \cite[Remark~4.3.9, p.~67]{CLD12}.

For the analytic-domain inclusion $A\hookrightarrow U$, relative
interior is ordinary topological interior. Since $U$ is boundaryless
and locally separated, the composition formula gives
\[
 \operatorname{Int}_U(A)
   =\operatorname{Int}(A/U)=\operatorname{Int}(A/K);
\]
these are \cite[\S2.11(3)--(4), pp.~7--8]{GR25}. Thus the ordinary
interior is nonempty. Finally $A$ is pure of dimension $n$, and its
Shilov points belong to its analytic boundary by
\cite[Lemma~6.2, p.~31]{GR25}. The same identification of interiors
shows that $z\notin\operatorname{Int}_U(A)$, as asserted.
\end{proof}

The geometric lemma is a \texttt{DIRECT\_COROLLARY} of the cited
formal-model and boundary facts, with the needed normalization
spelled out. It makes no field extension. Rigid or reduction points
with a finite residue-field extension, when used in the proof, are
points of the original $K$-analytic space.

\begin{remark}[Why one component alone is insufficient]
Over the same discrete valuation ring with uniformizer $\pi$, the
formal scheme
\[
 \operatorname{Spf}\bigl(K^\circ\langle T\rangle/(T^2-\pi^2)\bigr)
\]
has only one irreducible special-fiber component, but its generic
fiber consists of two rigid points, $T=\pi$ and $T=-\pi$, and its
Shilov boundary consists of both. The characteristic of $K$ is zero,
so these points are distinct even when the residue characteristic is
two. This formal scheme is not normal. Thus replacing the
normalization paragraph by a claim that ``one component implies one
Shilov point'' would be an invalid argument.
\end{remark}

\begin{proposition}[A nontrivial pole set has empty ordinary interior]
\label{prop:r2-no-interior-poles}
Let $U$ satisfy the allowed analytic hypotheses and be connected.
Let $\phi:U\to\mathbf R\cup\{-\infty\}$ be classical psh and not
identically $-\infty$. Then
\[
 P=\{\phi=-\infty\}
\]
has empty ordinary interior in $U$.
\end{proposition}

\begin{proof}
For integers $N\geq1$ define
\[
 h_N=\frac{\max\{\phi,-N\}}{N}.
\]
These are members of $\Cb(U)$ and are bounded below by $-1$.
Upper semicontinuity of $\phi$, whose values do not include $+\infty$,
gives a finite upper bound $M$ on a neighborhood of each point. On
such a neighborhood,
\[
 h_N\leq \frac{\max\{M,0\}}{N}.
\]
In particular the family is locally uniformly bounded above, although
it need not be increasing. Apply the bounded upper-envelope theorem,
Theorem~\ref{thm:r2-local-upper-envelope}, to each tail and put
\[
 g_j=\left(\sup_{N\geq j}h_N\right)^*\in\Cb(U).
\]
The star is ordinary upper-semicontinuous regularization. The tail
envelopes decrease, and on the same neighborhood the estimates give
\[
 -1\leq g_j\leq\frac{\max\{M,0\}}j.
\]
Consequently $g=\lim_j g_j$ is in $\Cb(U)$ by decreasing closure
of the classical class, and $-1\leq g\leq0$ everywhere. At every
point where $\phi$ is finite, $h_N$ tends to zero, so every tail
supremum, and hence every $g_j$, is at least zero there. Thus $g=0$
at that point. Since $\phi$ is not identically $-\infty$, $g$
attains its maximum zero somewhere on $U$.

The set $\{g=0\}$ is closed by upper semicontinuity and open by
the local maximum principle \cite[Theorem~6.1, pp.~30--31]{GR25},
because all points of $U$ are interior points. It is nonempty, so
connectedness implies $g\equiv0$. If $P$ contained a nonempty
ordinary open $O$, every $h_N$ and every tail supremum would equal
$-1$ on $O$. Locality of ordinary upper regularization would give
$g_j=-1$ on $O$ for all $j$, hence $g=-1$ there, a contradiction.
\end{proof}

\begin{theorem}[No divisorial poles]
\label{thm:r2-no-divisorial-poles}
Under the assumptions of Proposition~\ref{prop:r2-no-interior-poles},
the value $\phi(z)$ is finite at every divisorial point $z\in U$.
\end{theorem}

\begin{proof}
Suppose instead that $\phi(z)=-\infty$ at such a point. Choose
$A\subset U$ from Lemma~\ref{lem:r2-one-shilov-domain}. For every
$N\geq1$, the finite classical truncation
$\phi_N=\max\{\phi,-N\}$ is bounded on the compact domain $A$.
It takes its maximum on $A$ at a Shilov
point, by \cite[Theorem~6.4, pp.~33--34]{GR25}. Since the unique
Shilov point is $z$ and $\phi_N(z)=-N$, it follows that
$\phi_N\equiv-N$ on $A$. Letting $N\to\infty$ gives
$\phi\equiv-\infty$ on $A$. Its nonempty ordinary interior is
therefore contained in $P$, contrary to
Proposition~\ref{prop:r2-no-interior-poles}.
\end{proof}

Both pole statements are \texttt{DIRECT\_COROLLARY} of the bounded
upper-envelope theorem already obtained from the accepted internal
inputs, the external maximum principles, and the geometric lemma.
They do not use any capacity-zero theorem for pole sets, a trace
nonconcentration assertion, or a relative increasing-continuity
statement. Thus they may be used as inputs to those subsequent
questions without a capacity argument being fed back into itself.

The conclusion concerns divisorial points in the ambient dimension.
For example, a proper analytic subvariety can still be a pole set and
can carry an integration-current background. Divisorial points of
that smaller subvariety are not ambient divisorial points. This
theorem therefore does not assert nonconcentration on ambient polar
sets for arbitrary positive closed backgrounds.

% END source: base_r3/base_r2/sections/divisorial_poles.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/ambient_polar_traces.tex
\section{Ambient pole sets and all original trace measures}
\label{sec:r2-ambient-polar-traces}

This section combines the absence of divisorial poles from
Theorem~\ref{thm:r2-no-divisorial-poles} with the proved local
increasing-capacity theorem. Density of divisorial points alone
would not justify the conclusion. The passage from model atoms
to a smooth trace uses the quasicontinuity of the actual pole
indicator and a uniform bound on one fixed capacity domain.

\begin{theorem}[Vanishing of ambient smooth traces on a classical pole set]
\label{thm:r2-ambient-polar-traces}
Let $U$ be a connected allowed ordinary open and let
$\phi:U\to[-\infty,\infty)$ be classical psh and not identically
$-\infty$. Put $P=\{\phi=-\infty\}$. For every original compact
smooth $(n,n)$-test $\eta$, one has
\[
 |\mu_{[1],\eta}|(P)=0.
\]
Equivalently, $\mathbf1_P[1]=0$ as an original coefficient-Radon
current. No local $L^1$ assertion about the original function
$\phi$ is assumed in this theorem.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}, relative to the
stated internal inputs and the preceding R2 theorems.}

\begin{proof}
Use a finite partition of the test and ordinary algebraic
localization from B. Work on one moment presentation, with unit
coordinates $t_1,\ldots,t_r$ defined on a fixed larger compact
norm-box neighborhood. Choose a smaller norm-box domain
$\Omega\Subset U$ containing the localized test support, on
which the coordinates are bounded. Set
\[
 \rho=\tfrac12\sum_{a=1}^r t_a^2,
 \qquad \beta=D\rho=\sum_{a=1}^r d't_a\wedge d''t_a.
\]
The function $\rho$ is smooth and classical psh. Indeed its
pullback to every curve after every complete field extension is
a convex function of the harmonic logarithms of analytic units;
the curve chain rule gives nonnegative Laplacian.
The finite matrix domination of R supplies a cutoff $\chi$ and
a finite constant $C_\eta$ such that
\begin{equation}
\label{eq:r2-ambient-trace-domination}
 |\mu_{[1],\eta}|\leq C_\eta\,\chi\beta^n\wedge[1].
\end{equation}
Any antisymmetric component is annihilated, as in R. It therefore
suffices to show that the positive smooth trace
$\mu=\beta^n\wedge[1]$ vanishes on $P$ near the test support.

Apply the common local projective globalization from B, Lemma
D.2, to $\rho$, and then its projective pure-Fubini--Study
regularization theorem. More explicitly, this is the constructive
globalization part of the first-round
Lemma~\ref{lem:compact-box-regularization}, before its final
smoothing step; it applies on the entire prescribed smaller
norm box. In the fixed line frame, after removing
the common additive constant, this gives
\[
 \rho_j\downarrow\rho
\]
on a neighborhood of the entire prescribed compact box. The
functions $\rho_j$ are restrictions of rational pure-Fubini--Study
metric weights on one projective variety and one identified ample
line. They are bounded uniformly on the fixed $\Omega$: the
first row and the continuous limit bound the whole decreasing
row on its compact closure. Choose $M>0$ with
$|\rho_j|\leq M$ there for every $j$.

The positive measures
\[
 \mu_j=T_{[1],n}(\rho_j,\ldots,\rho_j)
\]
are restrictions of the corresponding calibrated model measures.
Each is a finite sum of atoms at divisorial points. Ordinary
algebraic localization, including the possible finite constant
field in B, Lemma D.5, preserves this assertion; the calibration
may acquire its stated fixed degree factor, which does not affect
the support or zero-mass conclusion. By
Theorem~\ref{thm:r2-no-divisorial-poles}, none of these points
belongs to $P$. Consequently
\[
 \mathbf1_P\mu_j=0.
\]
The fixed-background decreasing theorem gives
$\mu_j\to\mu$ on all original scalar tests, and hence weakly
on compact continuous scalar tests.

These weak limits have a uniform bound on the \emph{same}
admissibility domain. In fact
$v_j=(\rho_j+M)/(2M)$ belongs to $\Cb(\Omega)$ and is between
zero and one, so for every Borel $E\subset\Omega$,
\[
 \mu_j(E)\leq(2M)^n\operatorname{Cap}_{[1]}(E;\Omega).
\]
Subtract a constant from $\phi$ so that it is nonpositive on
a neighborhood of $\overline\Omega$, and use
\[
 w_N=1+N^{-1}\max\{\phi,-N\}\uparrow\mathbf1_{P^c}.
\]
The local increasing theorem gives quasicontinuity of this actual
limit for the same capacity. Thus $\mathbf1_P$ is a bounded
quasicontinuous multiplier. The R2 weak-closure multiplier theorem
applies to $\mu_j\to\mu$ and yields
\[
 \mathbf1_P\mu
       =\lim_j\mathbf1_P\mu_j=0
\]
on the compact core. This passage, rather than density of the
model points, is the essential step. Equation
\eqref{eq:r2-ambient-trace-domination} now proves the assertion
for the localized test. The finite partition proves it for every
original compact smooth $(n,n)$-form.
\end{proof}

\begin{corollary}[Ambient bounded BT products avoid classical pole sets]
\label{cor:r2-ambient-bt-poles}
Under the same hypotheses, for every bounded classical tuple
and every $0\leq k\leq n$,
\[
 \mathbf1_P T_{[1],k}(u_1,\ldots,u_k)=0.
\]
Every coefficient measure of these currents has total variation
zero on $P$. In particular every bounded mixed Monge--Amp\`ere
measure gives $P$ zero mass.
\end{corollary}

\begin{proof}
The theorem gives $\mathbf1_P[1]=0$. Apply the exact restriction
identity of Theorem~\ref{thm:r2-polar-restriction}. Equality of
currents and uniqueness of their coefficient-Radon measures give
the coefficient assertion.
\end{proof}

\begin{corollary}[Local outer-capacity zero of ambient pluripolar sets]
\label{cor:r2-ambient-polar-outer-zero}
For every permitted norm-box domain $\Omega\Subset U$ and
compact $H\Subset\Omega$,
\[
 \operatorname{Cap}_{[1]}^*(H\cap P;\Omega)=0.
\]
More generally, every locally pluripolar subset has local outer
capacity zero, where locally pluripolar means locally contained
in the pole set of a classical psh function not identically
$-\infty$ on its connected witness domain. This statement does
not itself assert the converse. The converse and qualified pole
patching on exhaustion domains are proved later in
Section~\ref{sec:r2-polar-converse-choquet}.
\end{corollary}

\begin{proof}
Every normalized bounded potential on $\Omega$ gives a measure
that vanishes on $P$, by the preceding corollary applied locally.
Thus the Borel capacity of $W\cap P$ is zero for an ordinary
open core $H\subset W\Subset\Omega$.
The pole indicator is quasicontinuous on $W$ by the increasing
argument used above. Hence
\[
 W\cap P=W\cap\{\mathbf1_{P^c}<1/2\}
\]
is quasi-open. The proved outer regularity of quasi-open sets
gives outer capacity zero, and monotonicity gives the assertion
on $H$. For a locally pluripolar set, cover its intersection
with a prescribed compact core by smaller open witness cores
whose larger witness domains lie in $\Omega$. Second
countability in the fixed finite $p$-adic setting gives a
countable subcover. Apply the local conclusion in each and
use countable subadditivity together with the correct direction
of capacity domain monotonicity.
\end{proof}

\begin{corollary}[Quasicontinuity of extended classical psh functions]
\label{cor:r2-extended-psh-qc}
Every extended-real classical psh function that is not identically
$-\infty$ on any connected component is locally finite and
continuous outside ordinary open sets of arbitrarily small
ambient outer BT capacity. Its lower truncations converge
locally in outer capacity. No $L^1$ bound on the unattenuated
witness is asserted by this conclusion.
\end{corollary}

\begin{proof}
Work on a compact core, subtract a local upper bound, and thus
write the function as a negative witness $\phi$. Choose the
collar trace $\nu=\beta_0\wedge[1]$ of the weighted theorem.
The ambient trace theorem and finite smooth domination of this
continuous-coefficient trace give $\nu(P)=0$.
Lemma~\ref{lem:r2-attenuation} supplies
\[
 \psi=-\vartheta(-\phi),\qquad
 \vartheta(t)=\sum_{j\geq1}\min\{t/L_j,1\},
 \qquad \int(-\psi)\,d\nu<\infty.
\]
The function $\vartheta$ is increasing and tends to infinity:
for any integer $m$, all of its first $m$ summands equal one
once $t\geq\max_{j\leq m}L_j$.
For every $N>0$,
\[
 \{\phi<-N\}\subset\{\psi\leq-\vartheta(N)\}.
\]
The weighted theorem applied to $\psi$ therefore bounds the
capacity of the ordinary open sublevel set on the core by
$C/\vartheta(N)$, which tends to zero. The original truncation
$\max\{\phi,-N\}$ agrees with $\phi$ outside that open set.
This proves local outer-capacity convergence. Finally combine
one such truncation's bounded quasicontinuity with the small
ordinary open sublevel set, exactly as in
Corollary~\ref{cor:r2-witness-outer-capacity}. The original
function is finite and continuous on the complement.
\end{proof}

% END source: base_r3/base_r2/sections/ambient_polar_traces.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/witness_criteria.tex
\section{Attenuated witnesses and an obstruction to exponential tails}
\label{sec:r2-witness-criteria}

The notation and original test category are those of
Section~\ref{sec:r2-weighted-l1}. The first observation weakens the
hypothesis needed for a pole-set conclusion; it does not assert
integrability of the original witness under that weaker hypothesis.

\begin{lemma}[An integrable witness with the same pole set]
\label{lem:r2-attenuation}
Let $\phi:O\to[-\infty,0]$ be classical psh, and let $\nu$ be a
finite positive Radon measure on $O$. If
\[
 \nu(\{\phi=-\infty\})=0,
\]
there is a negative classical psh function $\psi$ on the same
ordinary open set such that
\[
 \{\psi=-\infty\}=\{\phi=-\infty\},\qquad
                         \int(-\psi)\,d\nu<\infty.
\]
More precisely, one can choose $L_j\geq2^j$, tending increasingly
to infinity, and set
\begin{equation}
\label{eq:r2-attenuated-witness}
 \psi=\sum_{j\geq1}\frac{\max\{\phi,-L_j\}}{L_j},
 \qquad \int(-\psi)\,d\nu\leq1.
\end{equation}
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
This is an elementary measure-theoretic attenuation combined with
decreasing closure of extended-real classical psh functions
(Gubler--Rabinoff, Proposition 4.2(3)--(4)).

\begin{proof}
Put $t=-\phi$. The function $t$ is finite $\nu$-almost everywhere.
For $L\to\infty$, the functions $\min(t,L)/L$ are bounded by one
and converge to zero $\nu$-almost everywhere. Since $\nu$ is finite,
bounded convergence permits the choice of increasing $L_j\geq2^j$
such that
\[
 \int\frac{\min(t,L_j)}{L_j}\,d\nu\leq2^{-j}.
\]
Every finite partial sum in \eqref{eq:r2-attenuated-witness} is a
negative bounded classical psh function. The partial sums decrease,
so their extended-real limit is classical psh. At a point where
$t<\infty$, the absolute sum is bounded by
\[
 t\sum_j L_j^{-1}<\infty.
\]
At a point where $t=\infty$, each summand is $-1$, so the limit
is $-\infty$. This proves equality of the actual pole sets, including
the complete-extension classicality supplied by decreasing closure.
Finally monotone convergence gives
\[
 \int(-\psi)\,d\nu
   =\sum_j\int\frac{\min(t,L_j)}{L_j}\,d\nu\leq1.
\]
\end{proof}



\begin{corollary}[A pole-set criterion using one background trace]
\label{cor:r2-no-pole-trace}
In the geometry of Theorem~\ref{thm:r2-weighted-l1}, let
$\nu=\beta_0\wedge S$ and let $W$ be ordinary open with
$\overline W\Subset V$. If
\[
 (\beta_0\wedge S)(\{\phi=-\infty\})=0,
\]
then $W\cap\{\phi=-\infty\}$ has zero outer capacity for the
fixed admissibility domain $O$. Every bounded mixed top-degree
measure over $S$ gives this set zero mass. No $L^1$ assumption on
the original $\phi$ is necessary for these conclusions.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
The compactly supported positive trace $\nu$ is finite. Apply
Lemma~\ref{lem:r2-attenuation}, then apply
Theorem~\ref{thm:r2-weighted-l1} and
Corollary~\ref{cor:r2-witness-outer-capacity} to $\psi$.
The pole sets coincide. All domains remain the originally chosen
ones. For a mixed tuple that is only locally bounded, take its
finite bounds on a neighborhood of the compact core, as in the
weighted theorem.
\end{proof}

\begin{corollary}[Preservation of non-concentration on a specified pole set]
\label{cor:r2-background-pole-preservation}
Let $P=\{\phi=-\infty\}$ for a negative classical psh function
$\phi$ on $U$. Suppose that every coefficient measure of $S$
has total variation zero on $P$. Then every coefficient measure
of $T_{S,k}(u_1,\ldots,u_k)$ has total variation zero on $P$,
for $u_i\in\Cb(U)$ and $0\leq k\leq q$.
In particular, a background whose coefficient measures avoid all
such pole sets retains this property after every bounded BT product.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
Work locally on nested norm boxes with a collar as in the weighted
theorem. The hypothesis and finite positive controls imply
$(\beta_0\wedge S)(P)=0$, including for the continuous-coefficient
cutoff trace. The previous corollary makes every bounded top-degree
mixed measure over $S$ vanish on $P$ in a smaller core.

For a coefficient of a lower-degree product, localize its original
test to a finite moment presentation, and let
$\beta=D\rho_0$ be the quadratic trace form of that presentation,
where $\rho_0$ is smooth psh and bounded on the larger box.
The finite-test domination from R bounds the variation of that
coefficient on the core by a constant multiple of
\[
 T_{S,k}(u_1,\ldots,u_k)\wedge\beta^{q-k}
   =T_{S,q}(u_1,\ldots,u_k,
                 \underbrace{\rho_0,\ldots,\rho_0}_{q-k}).
\]
The equality is compatibility with smooth slots, already proved in
R1. This mixed measure vanishes on $P$ by the preceding paragraph.
Finite local presentations and ordinary restriction give the
assertion for every original coefficient. The case $k=0$ is the
hypothesis itself.
\end{proof}

\paragraph{The ambient specialization.}
For $S=[1]$, the criterion reduces the general polar-set question to
zero mass of pole sets for the specified original smooth volume
traces. That input is proved separately in
Theorem~\ref{thm:r2-ambient-polar-traces}, using divisorial
non-poles, bounded model approximations and the quasicontinuous
multiplier theorem. Neither the attenuation argument nor that
trace theorem asserts local $L^1$ integrability of the original
unattenuated witness. Complex distributional ellipticity is not
used. The criterion also makes clear why an arbitrary background
concentrated on the pole set is excluded.

\begin{proposition}[No uniform exponential improvement over arbitrary backgrounds]
\label{prop:r2-no-exponential-tails}
Over the original finite extension $K/\mathbf Q_p$, there are a
two-dimensional allowed ordinary open set $U$, a bounded continuous
classical psh function $f$, a background $S=D(f[1])$, a negative
classical psh witness $\phi$, and a compact $K_0\Subset U$ with
the following properties. The background trace integrability
condition of Theorem~\ref{thm:r2-weighted-l1} holds for every
fixed collar used below, but there is a normalized continuous
classical psh $v$ such that
\[
 \int_{K_0}\exp(-a\phi)\,T_{S,1}(v)=\infty
                   \quad\text{for every }a>0.
\]
Moreover the relative capacities of the deep sublevels have a
polynomial lower bound. Consequently the weighted $L^1$ theorem
cannot be strengthened, under its stated hypotheses, to an
exponential tail bound uniform over normalized potentials.
\end{proposition}

\noindent\textbf{Status of the exponential assertion refuted:
\texttt{FALSE}.}

\begin{proof}
Choose a uniformizer $\pi$ and put $\lambda=-\log|\pi|>0$.
Choose $0<r<1<R$ and set
\[
 U=\{|z_0|<1\}\times\{r<|z_1|<R\}.
\]
For $j\geq1$, put
\[
 f_j=\max\{\log|z_0|,-j\lambda\},\qquad
 f=\sum_{j\geq1}j^{-4}f_j,\qquad
 \phi=\log|z_0|,\qquad
 v=\frac{\max\{0,\log|z_1|\}}{\log R}.
\]
Each $f_j$ is continuous classical psh and takes values in
$[-j\lambda,0]$. The series for $f$ converges uniformly on $U$
because $\sum j^{-3}<\infty$. Its partial sums decrease, so $f$
is bounded continuous classical psh. The current
$S=D(f[1])$ is symmetric, original-positive and separately closed.
Writing $S_j=D(f_j[1])$, linearity and convergence give
\[
 S=\sum_{j\geq1}j^{-4}S_j.
\]
The positive series converges locally as a coefficient-Radon
current. Indeed, relative CLN bounds its mass against each fixed
positive cutoff by a constant times
$\sum j^{-4}(1+j\lambda)<\infty$.

The support of $S_j$ lies in $\{|z_0|=|\pi|^j\}$: on one
side $f_j$ is constant and on the other it is the logarithm of
the norm of an analytic unit. Thus $-\phi=j\lambda$ on this
support. For a fixed positive collar trace $\beta_0$, finite-test
domination and CLN give a constant $C_\rho$, independent of $j$,
such that
\[
 \int\beta_0\wedge S_j\leq C_\rho(1+j\lambda).
\]
The same is valid for its continuous coefficients by positive
smooth controls. Equality of the positive coefficient measures of
the series and monotone convergence imply
\[
 A_{S,\rho}(\phi)
  =\sum_{j\geq1}j^{-4}\,j\lambda
                       \int\beta_0\wedge S_j
  \leq C_\rho\sum_{j\geq1}j^{-4}j\lambda(1+j\lambda)<\infty.
\]

Let $x_j$ be the product Gauss point with radii $(|\pi|^j,1)$.
The usual model metric of $\mathcal O(1)$ on the first projective
line, after the coordinate rescaling $z_0\mapsto z_0/\pi^j$,
has affine weight $f_j+j\lambda$. Together with the standard
model metric in $z_1$, the mixed model intersection has degree
one. The original calibrated model-measure identity therefore gives
\[
 T_{S_j,1}(v)=(\log R)^{-1}\delta_{x_j}.
\]
Fixed continuous multiplication by $v$ and distributional $D$
pass the positive partial sums to the limit, yielding
\begin{equation}
\label{eq:r2-polynomial-tail-measure}
 T_{S,1}(v)=(\log R)^{-1}
                            \sum_{j\geq1}j^{-4}\delta_{x_j}.
\end{equation}
Choose $r<r_0<1<R_0<R$ and take
\[
 K_0=\{|z_0|\leq|\pi|\}
                      \times\{r_0\leq|z_1|\leq R_0\}.
\]
This is compact in $U$, contains every $x_j$, and lies in local
norm boxes admitting the required collar. Since
$\phi(x_j)=-j\lambda$, equation
\eqref{eq:r2-polynomial-tail-measure} gives
\[
 \int_{K_0}(-\phi)\,T_{S,1}(v)
      =\frac{\lambda}{\log R}\sum_{j\geq1}j^{-3}<\infty,
 \qquad
 \int_{K_0}e^{-a\phi}\,T_{S,1}(v)
      =\frac1{\log R}\sum_{j\geq1}j^{-4}e^{aj\lambda}=\infty.
\]
The potential $v$ is between zero and one on all of $U$. Hence,
for every $N>0$,
\[
 \operatorname{Cap}_S(K_0\cap\{\phi<-N\};U)
 \geq\frac1{\log R}\sum_{j>N/\lambda}j^{-4}
 \geq\frac1{3\log R}(1+N/\lambda)^{-3}.
\]
This cannot be bounded above by $A e^{-aN}$ with fixed finite
$A$ and $a>0$. All powers of $\pi$, all Gauss points, and all
model metrics used here are defined over the original field with
its prescribed absolute value.
\end{proof}

\begin{proposition}[A uniform first moment does not give uniform integrability]
\label{prop:r2-unbounded-weight-failure}
Even for $S=[1]$ on a smooth analytic curve, a fixed negative psh
witness satisfying the collar integrability condition need not
commute with weak limits of normalized bounded BT measures.
There can be positive measures $\mu_j\to0$ with
\[
 \int_{K_0}(-\phi)\,d\mu_j=1\quad\text{for all }j,
 \qquad (-\phi)\mu_j\longrightarrow\delta_0\ne0
\]
on one fixed compact $K_0$ containing their supports.
\end{proposition}

\noindent\textbf{Status of the asserted unbounded-weight stability:
\texttt{FALSE}.}

\begin{proof}
On $U=\{|z|<1\}\subset\mathbf A^{1,\an}_K$, set
$\phi=\log|z|$, with value $-\infty$ at $0$, and put
\[
 v_j=1+\frac{\max\{\log|z|,-j\lambda\}}{j\lambda},
 \qquad \lambda=-\log|\pi|.
\]
These are continuous classical psh functions with $0\leq v_j\leq1$.
Let $\xi_j$ be the Gauss point of radius $|\pi|^j$. The calibrated
one-variable model calculation gives
\[
 \mu_j=D(v_j[1])=(j\lambda)^{-1}\delta_{\xi_j},
 \qquad -\phi(\xi_j)=j\lambda.
\]
All points $\xi_j$ lie in the fixed compact disc
$K_0=\{|z|\leq|\pi|\}\Subset U$ and tend to $0$.
Thus $\mu_j\to0$ even in total variation, whereas
$(-\phi)\mu_j=\delta_{\xi_j}\to\delta_0$ as Radon measures,
and hence on every original compact smooth scalar test.
The witness is finite $\mu_j$-almost everywhere. Its first moment
is exactly one, but for every $M>0$,
\[
 \sup_j\int_{K_0\cap\{-\phi>M\}}(-\phi)\,d\mu_j=1.
\]
This is the precise failure of uniform integrability.

The background trace hypothesis can be checked directly. Choose
a radial smooth psh collar $\rho$ that is constant on a disc
containing $K_0$, with $\{\rho<0\}\Subset U$ still containing
that disc. Such a collar is a convex smooth regularized maximum
of zero and an affine function of $\log|z|$, minus a constant.
For $q=1$, its trace
$\beta_0=(-\rho)_+^4D\rho+
4(-\rho)_+^3d'\rho\wedge d''\rho$ is supported in a compact
annulus away from $0$.
The function $\phi$ is bounded there, so
$\int(-\phi)\beta_0\wedge[1]<\infty$.

The sequence $v_j$ increases. Its actual pointwise supremum is
one off $0$ and zero at $0$, so it is not upper semicontinuous.
It belongs to the bounded capacitary closure by
Corollary~\ref{cor:r2-local-increasing}. No claim is made that
this supremum is a classical psh limit. The bounded-weight
calculus is consistent with $\mu_j\to0$; the failure concerns
only the attempted extension to the fixed unbounded multiplier.
\end{proof}

% END source: base_r3/base_r2/sections/witness_criteria.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/obstacle_maximality.tex
\section{Quasicontinuous obstacles and local maximality}
\label{sec:r2-obstacle-maximality}

This section concerns the ambient background $[1]$. It uses the
continuous-obstacle envelope and orthogonality theorem of
\textup{B}, Theorem C.6, without changing its hypotheses. A separate
limit argument extends orthogonality to a class of bounded
discontinuous obstacles. It then supplies the local equivalence
between maximality and vanishing Monge--Amp\`ere measure. No
relative obstacle orthogonality for an arbitrary background is
asserted.

\begin{lemma}[Continuous majorants of a quasicontinuous function]
\label{lem:r2-continuous-majorants-capacity}
Let $f_j\downarrow f$ be continuous real functions on an ordinary
neighborhood of a compact $H$, with $f$ bounded and
$C_\Omega$-quasicontinuous on a compact neighborhood of $H$.
Then
\[
 C_\Omega^*(H\cap\{f_j-f>\delta\})\longrightarrow0
       \qquad(\delta>0).
\]
Here $\Omega$ is the fixed local capacity domain containing that
neighborhood.
\end{lemma}

\begin{proof}
For $\epsilon>0$, choose an ordinary open $G$ of capacity less
than $\epsilon$ such that $f$ is continuous on the chosen compact
neighborhood minus $G$. On this compact set, the continuous
functions $f_j$ decrease to a continuous function. Dini's theorem
gives uniform convergence there. Thus the indicated level set is
contained in $G$ for every sufficiently large $j$, and its outer
capacity is at most $C_\Omega(G)<\epsilon$.
\end{proof}

\begin{theorem}[Orthogonality for bounded quasicontinuous usc obstacles]
\label{thm:r2-qc-orthogonality}
Let $Y/F$ be smooth projective geometrically integral over a finite
extension $F/\mathbf Q_p$, with its prescribed absolute value,
and let $L$ be ample. Let $f$ be a bounded
upper-semicontinuous metric on $L$, relative to a continuous
reference metric. Assume that its local weights are locally
quasicontinuous for the ambient capacities $\operatorname{Cap}_{[1]}$
on the permitted local norm boxes. Put
\[
 p=P_L^{\mathrm b} f,
\]
the greatest bounded classical minorant from
Proposition~\ref{prop:r2-usc-obstacle}. Then
\begin{equation}
\label{eq:r2-qc-orthogonality}
 \int_{Y^{\an}}(f-p)c_1(L,p)^n=0.
\end{equation}
In particular this applies to $f=r+g$, where $r$ is a bounded
classical metric and $g$ is any continuous real function on
$Y^{\an}$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
The new step is the extension to these discontinuous obstacles;
the continuous-obstacle theorem is an \texttt{INTERNAL\_INPUT}.

\begin{proof}
Choose uniformly bounded continuous metrics $f_j\downarrow f$.
Let $p_j=P_L f_j$. The accepted envelope theorem makes each
$p_j$ continuous model-semipositive and gives
\[
 \int(f_j-p_j)c_1(L,p_j)^n=0.
\]
Proposition~\ref{prop:r2-usc-obstacle} gives $p_j\downarrow p$,
with a common bound relative to a continuous semipositive
reference metric. Write $\mu_j=c_1(L,p_j)^n$ and
$\mu=c_1(L,p)^n$. The fixed-background monotone theorem, in
local frames, gives $\mu_j\to\mu$ on all original scalar tests.
The measures have the same finite total mass $\deg_F(L^n)$,
by continuous compatibility and passage to the decreasing limit.

Fix a local compact test stage. In its frame the weights of all
$p_j$ have uniform bounds on a slightly larger box. The R1
coefficient-capacity estimate therefore controls $\mu_j$
uniformly by that fixed box's ambient capacity. By
Lemma~\ref{lem:r2-continuous-majorants-capacity}, the local weights
of $f_j$ converge to those of $f$ in outer capacity. These weights
are quasicontinuous, as is their limit. The multiplier theorem
\ref{thm:r2-weak-closure-multipliers} yields convergence of the
integrals with multiplier $f_j$. The weighted decreasing theorem
yields convergence with multiplier $p_j$. Subtract the two
convergences. A finite original smooth partition of unity on the
compact projective space makes the differences of local weights
equal to the global scalar functions $f_j-p_j$ and $f-p$.
Summing gives \eqref{eq:r2-qc-orthogonality}.

Finally, bounded classical local weights are quasicontinuous by
R1, and addition of a continuous function preserves
quasicontinuity and upper semicontinuity. Thus $r+g$ satisfies all
the stated assumptions. No monotone-convergence theorem for
arbitrary discontinuous weights was used: their convergence was
proved using Dini's theorem off ordinary small-capacity opens.
\end{proof}

\begin{definition}[Local maximality]
\label{def:r2-maximality}
A function $u\in\Cb(U)$ is locally maximal if every point has
an ordinary neighborhood $V$ with compact closure in a smooth
affine algebraic chart such that the following holds. Whenever
$v\in\Cb(V)$ and $v=u$ on $V\setminus K$ for some compact
$K\Subset V$, one has $v\leq u$ on $V$.

Equivalently one may use competitors whose strict positive gap
over $u$ has compact closure in $V$: replace such a competitor
by its maximum with $u$. This definition is local and does not
use the whole of a proper analytic space as a boundaryless
comparison domain.
\end{definition}

\begin{theorem}[Maximality and the homogeneous equation]
\label{thm:r2-maximal-ma-zero}
Let $u\in\Cb(U)$ on any allowed good, Hausdorff, boundaryless,
quasi-smooth pure-dimensional space over the fixed $K$. Then
\[
 u\text{ is locally maximal}
       \quad\Longleftrightarrow\quad (Du)^n=0\text{ on }U.
\]
The equality is that of the ordinary original-CLD Radon measure.
No continuity assumption on $u$ is imposed.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}, relative to the adopted
inputs and the preceding orthogonality extension.}

\begin{proof}
Assume first that $(Du)^n=0$. On a relatively compact framed
affine neighborhood $V$, a competitor as in
Definition~\ref{def:r2-maximality} satisfies every compact-gap
condition of Corollary~\ref{cor:r2-pointwise-ambient}. Its strict
gap has zero $(Du)^n$-mass. That corollary gives $u\geq v$
pointwise. Such neighborhoods exist by \textup{B}, so $u$ is
locally maximal.

Conversely, fix a point where maximality holds on $V$. The
common-collar construction of \textup{B}, Lemma D.2, realizes
$u+c$, on a smaller ordinary neighborhood $W\Subset V$, as the
weight of a bounded classical metric $r$ on an ample line $L$
over a smooth projective variety. Its frame is fixed on $W$.
Choose any nonnegative original smooth scalar $\chi$ compactly
supported in $W$; it extends by zero to the projective space.
For $\epsilon>0$, let
\[
 f=r+\epsilon\chi,\qquad p=P_L^{\mathrm b}f.
\]
The metric $r$ is a permitted minorant, so
$r\leq p\leq r+\epsilon\chi$. In particular $p=r$ off the
compact support of $\chi$. Its weight on $W$, minus $c$, glues
with $u$ outside that support to give a member of $\Cb(V)$
which equals $u$ off a compact subset of $V$. Maximality forces
this function to be at most $u$. Therefore $p=r$ everywhere,
also outside $W$ by the obstacle inequalities.

The obstacle $f=r+\epsilon\chi$ satisfies
Theorem~\ref{thm:r2-qc-orthogonality}. Its conclusion is
\[
 0=\int(f-p)c_1(L,p)^n
   =\epsilon\int\chi\,c_1(L,r)^n
   =\epsilon\int_W\chi\,(Du)^n.
\]
This holds for every such $\chi$, so positivity and Radon
uniqueness give $(Du)^n=0$ on $W$. Cover $U$ by these
neighborhoods. If the local algebraic chart uses the finite
constant-field convention in \textup{B}, Lemma D.5, its original
calibration changes by a positive constant; the vanishing
assertion is unchanged. Thus it holds for the prescribed $K$
and its original absolute value.
\end{proof}

\begin{remark}[Scope for Dirichlet theory]
The theorem identifies bounded solutions of the homogeneous
equation with a local maximality property. It does not prove
that a Perron envelope attains arbitrary prescribed boundary
values, that every ordinary domain admits boundary barriers,
or that a nonzero prescribed measure admits a solution. Those
are separate Dirichlet existence and boundary-regularity
questions. For an arbitrary background $S$, even pointwise
domination can fail, as the connected nonzero-background
counterexample in this record shows; the ambient theorem above
cannot be transferred to that setting without further hypotheses.
\end{remark}

% END source: base_r3/base_r2/sections/obstacle_maximality.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/perron_peak_barriers.tex
\section{A homogeneous Dirichlet theorem with explicit peak barriers}
\label{sec:r2-perron}

The results in this section concern the ambient background $[1]$.
The geometric hypothesis is stated separately: no assertion is
made that an arbitrary Berkovich analytic domain has these
barriers. The solution will be bounded and classical psh, with
the stated ordinary boundary limits. Interior continuity is not
part of the conclusion.

Let $V$ be a connected ordinary open set whose closure $\overline V$
is compact and is contained in a smooth affine algebraic chart.
Write
\[
 \partial V=\overline V\setminus V
\]
for its ordinary topological boundary, and assume that this set
is nonempty. All closures and limits in this section refer to
the ordinary Berkovich topology.

\begin{definition}[Continuous psh peak barriers]
\label{def:r2-peak-barriers}
The domain $V$ has continuous psh peak barriers if for each
$\xi\in\partial V$ there is a function
\[
 \beta_\xi\in \Cb(V)\cap C(\overline V)
\]
such that $\beta_\xi(\xi)=0$ and
$\beta_\xi(x)<0$ for every $x\in\overline V\setminus\{\xi\}$.
\end{definition}

\begin{lemma}[Boundary maximum principle]
\label{lem:r2-boundary-maximum}
Suppose that $h\in\Cb(V)$ is bounded above and
\[
 \limsup_{V\ni x\to\xi}h(x)\leq0
       \qquad(\xi\in\partial V).
\]
Then $h\leq0$ on $V$.
\end{lemma}

\begin{proof}
If $h(x_0)>0$, choose $0<a<h(x_0)$. For each boundary point the
limsup condition gives a neighborhood on which $h<a$ in $V$.
The union of these neighborhoods contains the compact boundary.
Thus $\{h\geq a\}$ is contained in a compact subset of $V$; it
is closed there by upper semicontinuity. The bounded usc
function $h$ attains its global positive maximum at a point
of $V$. The strong maximum principle for classical psh
functions, \cite[Theorem 6.1]{GR25}, makes it constant on the
connected domain $V$. A positive constant contradicts the
boundary condition at the nonempty boundary.
\end{proof}

\begin{theorem}[Homogeneous Dirichlet problem under peak barriers]
\label{thm:r2-peak-dirichlet}
Assume that $V$ has the barriers in
Definition~\ref{def:r2-peak-barriers}. For every
$g\in C(\partial V)$ there is a unique bounded classical psh
function $H_g$ on $V$ such that
\[
 (DH_g)^n=0\quad\hbox{on }V,\qquad
 \lim_{V\ni x\to\xi}H_g(x)=g(\xi)
       \quad(\xi\in\partial V).
\]
The measure equation is tested against all original compactly
supported smooth CLD scalar tests. Moreover
\[
 \min_{\partial V}g\leq H_g\leq\max_{\partial V}g,
 \qquad
 \|H_{g_1}-H_{g_2}\|_{L^\infty(V)}
       \leq\|g_1-g_2\|_{L^\infty(\partial V)} .
\]
In particular the solution is monotone with respect to its
boundary values.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED} under the displayed
barrier hypothesis.} The corresponding assertion for general
ordinary analytic domains remains \texttt{OPEN}; no existence
of barriers in general dimension is supplied here.

\begin{proof}
Put $m=\min_{\partial V}g$ and $M=\max_{\partial V}g$, and let
$\mathcal F_g$ consist of all $v\in\Cb(V)$ satisfying
\[
 m\leq v\leq M,\qquad
 \limsup_{V\ni x\to\eta}v(x)\leq g(\eta)
       \quad(\eta\in\partial V).
\]
This family contains the constant $m$. Let
$H=\sup_{v\in\mathcal F_g}v$, pointwise, and let $H^*$ denote
its ordinary upper-semicontinuous regularization on $V$.
Theorem~\ref{thm:r2-local-upper-envelope} gives
$H^*\in\Cb(V)$ and $m\leq H^*\leq M$.

Fix $\xi\in\partial V$ and $\epsilon>0$. Continuity of $g$
gives a boundary neighborhood of $\xi$ on which
$|g-g(\xi)|<\epsilon$. On the compact complement,
$\beta_\xi$ has a strictly negative maximum. Consequently
one can choose a finite $A>0$ such that, on all of $\partial V$,
\[
 b_-=g(\xi)-\epsilon+A\beta_\xi\leq g,\qquad
 b_+=g(\xi)+\epsilon-A\beta_\xi\geq g .
\]
The function $b_-$ is classical psh on $V$, continuous on the
closure, and at most $M$. Thus $\max(m,b_-)\in\mathcal F_g$,
and $H^*\geq b_-$. For every $v\in\mathcal F_g$, the function
\[
 v-b_+=v+A\beta_\xi-g(\xi)-\epsilon
\]
is bounded classical psh and has boundary limsup at most zero.
Lemma~\ref{lem:r2-boundary-maximum} implies $v\leq b_+$.
Since $b_+$ is continuous, also $H^*\leq b_+$. Taking limits
at $\xi$ in
\[
 g(\xi)-\epsilon+A\beta_\xi
       \leq H^*
       \leq g(\xi)+\epsilon-A\beta_\xi
\]
and then letting $\epsilon$ decrease to zero proves
$H^*(x)\to g(\xi)$ as $x\to\xi$ from $V$. Hence
$H^*\in\mathcal F_g$, so $H=H^*$ pointwise.

It remains to prove the equation. Let $W\Subset V$ be an
ordinary neighborhood, and let $v\in\Cb(W)$ agree with $H$
outside a compact set $K\Subset W$. The function equal to
$\max(v,H)$ on $W$ and to $H$ off $K$ is well defined and
classical psh by ordinary gluing and maximum stability.
Call it $w$. It is globally bounded, since its modification
is confined to $K$, and it has boundary limits $g$.
Lemma~\ref{lem:r2-boundary-maximum}, applied to $w-M$,
gives $w\leq M$; also $w\geq H\geq m$. Therefore
$w\in\mathcal F_g$ and $w\leq H$ by the definition of $H$.
It follows that $v\leq H$. This proves local maximality
of $H$, and Theorem~\ref{thm:r2-maximal-ma-zero} yields
$(DH)^n=0$.

If $H_1,H_2$ are two bounded classical solutions with the
same boundary values, then for each $\epsilon>0$ the set
$\{H_1<H_2-\epsilon\}$ has compact closure in $V$.
Indeed, the ordinary boundary limits give a neighborhood
of each boundary point on which this set is empty, and
compactness of the boundary gives a common compact core.
The analogous assertion holds after interchanging the
functions. Since their Monge--Amp\`ere measures vanish,
pointwise ambient domination
(Corollary~\ref{cor:r2-pointwise-ambient}) gives both
$H_1\geq H_2$ and $H_2\geq H_1$.

More generally set
$\delta=\|g_1-g_2\|_{L^\infty(\partial V)}$.
The pair $H_{g_2}+\delta,H_{g_1}$ satisfies the required
one-sided compact-gap condition and has zero measures, so domination
gives $H_{g_1}\leq H_{g_2}+\delta$. Reversing the indices
proves the norm estimate. If $g_1\leq g_2$, apply this
comparison directly with no shift to obtain
$H_{g_1}\leq H_{g_2}$.
\end{proof}

\begin{remark}[What the theorem does not settle]
A classification or construction of suitable domains in
arbitrary dimension is a separate geometric problem.
The proof uses continuous peak barriers to control the
ordinary boundary and the discontinuous-obstacle theorem
to identify maximality with the homogeneous equation.
It supplies neither interior continuity for $H_g$ nor
existence for an arbitrary prescribed nonzero measure.
For an arbitrary positive closed background $S$, the
pointwise domination step would require an additional
gap-detection hypothesis.
\end{remark}


% END source: base_r3/base_r2/sections/perron_peak_barriers.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/capacity_noncompactness.tex
\section{A bounded family which is not precompact in capacity}
\label{sec:r2-capacity-noncompactness}

This example uses the fixed ambient background $S=[1]$. All
potentials, including their pointwise limit, are continuous
classical psh. It rules out replacing the capacity-convergence
hypothesis by bounded pointwise convergence, and rules out a
capacity-precompactness argument for the whole normalized test
family. It is not a counterexample to weak-closure domination
of that family's measures.

\subsection{A moving residue branch over the original field}

Let $\pi$ be a uniformizer of the fixed finite extension
$K/\mathbf Q_p$, and retain its original absolute value. Put
\[
 \lambda=-\log|\pi|>0,\qquad
 V=\{x\in\mathbf A_K^{1,\an}:|T(x)|<|\pi|^{-1}\}.
\]
Choose distinct monic irreducible polynomials
$\overline P_i\in\widetilde K[T]$, of degrees $d_i\geq2$,
and monic lifts $P_i\in K^\circ[T]$. Define
\[
 h_i=\frac1{d_i}\max\{\log|P_i(T)|,-d_i\lambda\},
 \qquad
 \phi_i=\lambda^{-1}h_i,\qquad
 \psi_i=\frac{1+\phi_i}{2}.
\]
These are continuous classical psh functions. In fact $h_i$ is
the local weight of the pure Fubini--Study metric on
$\mathcal O_{\mathbf P^1}(1)$ specified, after tensoring by
$d_i$, by the two sections
\[
 P_i^{\mathrm{hom}}(X,Z),\qquad \pi^{d_i}Z^{d_i}.
\]
They have no common zero. The factors $d_i^{-1}$ and
$\lambda^{-1}$ above are factors in the potential, not a
renormalization of the field or of the integration current.

Let $\zeta$ be the unit Gauss point. Let $\xi_i$ be the unique
Shilov point of the closed residue disc
\[
 B_i=\{|P_i(T)|\leq|\pi|^{d_i}\}.
\]
More explicitly, $P_i$ has $d_i$ simple Hensel roots in the
unramified extension of degree $d_i$. Over that extension,
$B_i$ is the disjoint union of the closed discs of radius
$|\pi|^{d_i}$ about those roots. Their conjugate Gauss points
restrict to the same original $K$-point $\xi_i$.
The residue field of its branch has degree $d_i$.

\begin{lemma}[The branch weights and their measures]
\label{lem:r2-moving-branch-measures}
On $V$ one has $0\leq\psi_i<1$ and
\[
 D h_i=\delta_{\xi_i},\qquad
 D\psi_i=\frac1{2\lambda}\delta_{\xi_i}.
\]
Moreover $\xi_i\to\zeta$, and pointwise on $V$
\[
 \psi_i\longrightarrow
 \psi=\frac12+\frac1{2\lambda}\max\{0,\log|T|\}.
\]
The limit is continuous classical psh, and
\[
 D\psi=\frac1{2\lambda}\delta_\zeta.
\]
At the indicated points,
\[
 \psi_i(\xi_i)=0,\qquad
 \psi_j(\xi_i)=\frac12\ (j\ne i),\qquad
 \psi(\xi_i)=\psi(\zeta)=\frac12.
\]
\end{lemma}

\begin{proof}
Irreducibility of the separable reduction gives the unramified
root extension and the described discs by Hensel's lemma.
The closed $K$-domain $B_i$ is the closed disc over this
unramified residue extension, viewed as a $K$-analytic domain;
it has one Shilov point. On its interior $h_i=-\lambda$.
Off $B_i$, the function is $d_i^{-1}\log|P_i|$, which is
harmonic there because $P_i$ has no zero there. Thus the
positive curvature measure of the displayed model metric is
supported at $\xi_i$. Its total mass is
$\deg_K\mathcal O_{\mathbf P^1}(1)=1$ in the original model
formula, so it is $\delta_{\xi_i}$. Equivalently, the radial
slope $d_i^{-1}$ at that branch is multiplied by its residue
degree $d_i$; these two factors cancel. This is the same
normalization, not a degree-one assertion about the residue
field. Scaling gives the formula for $D\psi_i$.

For $|T|\leq1$, all $|P_i|\leq1$, so $-\lambda\leq h_i\leq0$.
For $|T|>1$, monicity and integral coefficients give
$|P_i(T)|=|T|^{d_i}$, hence $h_i=\log|T|$. The asserted
normalization on $V$ follows.

At a point of the closed unit disc, at most one of the numbers
$|P_i|$ can be strictly less than one. Indeed, two distinct
irreducible reductions are coprime; a lifted Bezout identity
would otherwise have left side of norm less than one and right
side of norm one. Therefore $h_i(x)\to0$ at every such point.
Outside that disc all $h_i$ equal $\log|T|$. This proves the
pointwise convergence. The formula for $D\psi$ is the
one-variable canonical model formula at the unit Gauss point.
The same coprimality argument gives all the stated point values.

Finally, take a nonzero $F\in K[T]$ and divide by a coefficient
of maximum norm so that its reduction is a nonzero polynomial.
Only finitely many of the $\overline P_i$ divide that
reduction. For all other $i$,
\[
 |F(\xi_i)|=\max_\nu|\operatorname{coeff}_\nu(F)|
           =|F(\zeta)|.
\]
Polynomial seminorms define the Berkovich topology, so
$\xi_i\to\zeta$.
\end{proof}

\subsection{The counterexample in every dimension}

Let $\Omega=V^n$, an ordinary open in $\mathbf A_K^{n,\an}$.
For $2\leq l\leq n$ put
\[
 b_l(T_l)=\lambda^{-1}\max\{0,\log|T_l|\},
 \qquad
 u_i=\frac1n\left(\psi_i(T_1)+\sum_{l=2}^n b_l(T_l)\right).
\]
For $n=1$ the sum is empty. Then $0\leq u_i<1$ on
$\Omega$, and $u_i$ converges pointwise to the continuous
classical psh function
\[
 u=\frac1n\left(\psi(T_1)+\sum_{l=2}^n b_l(T_l)\right).
\]
Let $z_i$ be the product Gauss point with first coordinate
$\xi_i$ and the remaining coordinates unit Gauss points, and
let $z$ be the all-unit Gauss point. Explicitly, its seminorm
on polynomials in the remaining coordinates is the maximum
of the coefficient seminorms at $\xi_i$. This description
shows that $z_i\to z$.

\begin{proposition}[Failure of capacity precompactness and pointwise
weighted stability]
\label{prop:r2-capacity-noncompactness}
Set
\[
 c_n=\frac{n!}{2n^n\lambda^n}>0.
\]
For the fixed ambient background $[1]$,
\[
 (Du_i)^n=c_n\delta_{z_i},\qquad
 (Du)^n=c_n\delta_z.
\]
For $i\ne j$,
\[
 \operatorname{Cap}_{[1]}
    \left(\left\{|u_i-u_j|>\frac1{4n}\right\};\Omega\right)
       \geq c_n.
\]
Consequently the normalized continuous classical family
$(u_i)$ has no capacity-Cauchy subsequence. The same lower
bound holds for the error sets with $u_j$ replaced by the
pointwise limit $u$.
This failure also holds for local capacity convergence in the
all-domain sense: one compact subset of one fixed smaller
domain already detects every pairwise error.

Although $(Du_i)^n\to(Du)^n$ weakly, the weighted products fail
to converge to the product of the pointwise limits:
\[
 u_i(Du_i)^n=0\quad\text{for every }i,\qquad
 u(Du)^n=\frac{c_n}{2n}\delta_z\ne0.
\]
\end{proposition}

\begin{proof}
The mixed product with one factor from each coordinate is
\[
 D\psi_i(T_1)\wedge Db_2(T_2)\wedge\cdots\wedge Db_n(T_n)
       =\frac1{2\lambda^n}\delta_{z_i}.
\]
This is the product model formula: the mixed intersection
number of the $n$ distinct pullbacks of
$\mathcal O_{\mathbf P^1}(1)$ is one. The first coordinate
retains the residue-degree cancellation proved in the lemma.
There is no factorial in this mixed product. A term with a
repeated coordinate vanishes; it is the product of two
curvatures pulled back from the same curve. Equivalently,
the relevant positive product on $(\mathbf P^1)^n$ has zero
mixed intersection number and hence zero mass.
Expanding the $n$th power of the sum defining $u_i$ therefore
contributes exactly $n!$ equal mixed terms and the factor
$n^{-n}$. This gives $c_n\delta_{z_i}$; the same calculation
gives the formula for $u$.

At $z_i$ one has
\[
 u_i(z_i)=0,\qquad
 u_j(z_i)=u(z_i)=\frac1{2n}\quad(j\ne i).
\]
The error set thus contains $z_i$. The function $u_i$ itself
is admissible in the capacity supremum, so its Monge--Amp\`ere
mass $c_n$ at $z_i$ proves both lower bounds. They hold for
every pair of distinct indices, excluding every
capacity-Cauchy subsequence.

For the local assertion, put
\[
 K_*=\{z\}\cup\{z_i:i\geq1\},\qquad
 \Omega_0=\operatorname{Int}
   \{\,|T_l|\leq|\pi|^{-1/2}\text{ for }1\leq l\leq n\,\}.
\]
The convergence $z_i\to z$ makes $K_*$ compact, and
$K_*\Subset\Omega_0\Subset\Omega$. All $u_i$ remain normalized
admissible tests after restriction to $\Omega_0$. Therefore
\[
 \operatorname{Cap}_{[1]}
  \left(K_*\cap\left\{|u_i-u_j|>\frac1{4n}\right\};\Omega_0\right)
       \geq c_n\qquad(i\ne j),
\]
and the same estimate holds with $u_j$ replaced by $u$.
Thus passing to local capacities does not restore a
capacity-Cauchy subsequence.

Weak convergence of the unweighted measures follows from
$z_i\to z$. The displayed point values give the weighted
identities, with $u(z)=1/(2n)$. These are different ordinary
Radon measures and are distinguished by an original smooth
compact scalar cutoff which equals one near $z$.
\end{proof}

\paragraph{Status and excluded route.}
The claims of capacity precompactness of the whole normalized
family, and of weighted continuity under bounded pointwise
convergence, are \texttt{FALSE}; this is an explicit
counterexample in every dimension. Each individual ingredient
uses the established continuous/model calculus. The example
does not contradict monotone continuity, convergence in
capacity, the completion theorem, or weak-closure domination:
its unweighted weak limit is itself the Monge--Amp\`ere measure
of an admissible continuous potential.

% END source: base_r3/base_r2/sections/capacity_noncompactness.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/divisorial_determination.tex
\section{Divisorial determination of the ambient capacitary order}
\label{sec:r2-divisorial-determination}

The statement here is an additional consequence of the model
approximation and controlled-multiplier arguments already proved.
It gives the pointwise reduction used to solve the
increasing-envelope problem in the following section. Density of divisorial points by itself
would not suffice for any assertion about discontinuous functions.

\begin{proposition}[Divisorial test for local quasi-everywhere order]
\label{prop:r2-divisorial-capacitary-order}
Let $V$ be an ordinary open with compact closure in a smooth
affine algebraic chart. Let $a,b$ be locally bounded real Borel
functions on $V$, locally quasicontinuous for the ambient
capacities on all permitted local norm boxes. Then the following
conditions are equivalent:
\begin{enumerate}[label=(\roman*)]
\item $a\leq b$ locally outside sets of outer
$\operatorname{Cap}_{[1]}$-capacity zero;
\item $a(z)\leq b(z)$ for every ambient divisorial point $z\in V$.
\end{enumerate}
In particular the local ambient capacitary equivalence class of
a bounded quasicontinuous function is determined by its actual
values at ambient divisorial points.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
The inputs are the atomic testers, the whole-box model
regularization from B, and the controlled quasicontinuous
multiplier theorem; a new regularization theorem is not asserted.

\begin{proof}
For (i)$\Rightarrow$(ii), a divisorial point has positive
singleton capacity on every sufficiently small framed affine
capacity domain containing it, by Lemma~\ref{lem:r2-atomic-tester}.
It therefore cannot lie in a local outer-capacity-zero
exceptional set.

For the converse fix a permitted capacity domain $\Omega\subset V$
and an ordinary open $W\Subset\Omega$ on a compact neighborhood
of whose closure $a$ and $b$ are bounded and
$\operatorname{Cap}_{[1]}(\,\cdot\,;\Omega)$-quasicontinuous.
Let $H\subset W$ be compact. Set $g=(a-b)_+$. This is a bounded
nonnegative quasicontinuous Borel function near $\overline W$,
and it vanishes at every ambient divisorial point there.

Let $v\in\Cb(\Omega)$ with $0\leq v\leq1$. Cover $H$ by finitely
many smaller cores $H_\alpha$ and choose nonnegative original
smooth scalar cutoffs $\chi_\alpha$, supported in slightly
larger norm boxes and equal to one on $H_\alpha$.
The whole-box regularization from B gives, on each such larger
box, a uniformly bounded decreasing sequence of continuous
semipositive rational pure Fubini--Study weights
$v_{\alpha,j}\downarrow v$. Here rational pure-FS is the
model class specified in B, Theorem 4.4(2); arbitrary real
weights in a Fubini--Study presentation are not needed.
Their Monge--Amp\`ere measures
$\mu_{\alpha,j}=(Dv_{\alpha,j})^n$ are restrictions of model
measures and hence are supported at ambient divisorial points.
If the construction shifts $v$ by a fixed constant, subtract
that constant; curvature and the argument are unchanged.
Consequently
\[
 \int \chi_\alpha g\,d\mu_{\alpha,j}=0 .
\]
The measures converge weakly to $(Dv)^n$ and, on the support
of $\chi_\alpha$, have a common bound by the capacity of one
fixed intermediate norm box. The function $g$ is quasicontinuous
for this same box. Theorem~\ref{thm:r2-weak-closure-multipliers}
therefore gives
\[
 \int \chi_\alpha g\,(Dv)^n=0 .
\]
Only a fixed bounded quasicontinuous multiplier has been passed
through this controlled family of measures. No arbitrary
Borel-multiplier continuity is used.

It follows that $(Dv)^n(H\cap\{g>\delta\})=0$ for every
$\delta>0$. Since this holds for every normalized $v$ on
$\Omega$, the raw capacity of this set is zero. More usefully,
the argument applies to every compact core in $W$. Exhausting
$W$ by compact sets and using continuity from below of the raw
capacity shows that $W\cap\{g>\delta\}$ has raw capacity zero
relative to $\Omega$. This set is quasiopen relative to $\Omega$:
off an arbitrarily small ordinary exceptional open, $g$ is
continuous on a neighborhood of $\overline W$, and $W$ is open.
Proposition~\ref{prop:r2-quasiopen-outer} thus gives zero outer
capacity as well. Taking the countable union over positive
rational $\delta$ proves (i). The all-domain local
formulation follows by repeating the argument in every
requested norm box, or by the framed-domain comparison
Theorem~\ref{thm:r2-domain-capacity-comparison}.
\end{proof}

\begin{corollary}[The exact divisorial form of the no-jump problem]
\label{cor:r2-divisorial-nojump-reduction}
Let $u_j\in\Cb(U)$ increase and be locally uniformly bounded.
Put
\[
 h=\sup_j u_j,\qquad U_h=h^*.
\]
The preceding results give local ambient quasicontinuity of
$h$, and $U_h\in\Cb(U)$. The assertion
\[
 h=U_h\quad\hbox{locally }
       \operatorname{Cap}_{[1]}^*\hbox{-quasi-everywhere}
\]
is equivalent to the assertion
\[
 \sup_j u_j(z)
   =\limsup_{x\to z}\sup_j u_j(x)
       \quad\hbox{for every ambient divisorial }z.
\]
When the sequence has a common compact collar, these conditions
are also equivalent to the energy no-jump condition proved in
Proposition~\ref{prop:r2-envelope-energy-criterion}.
\end{corollary}

\begin{proof}
The actual increasing-limit theorem gives the required
quasicontinuity of $h$; the bounded upper-envelope theorem
gives that of $U_h$. Apply the proposition to the pair, using
$h\leq U_h$. For the energy equivalence, the earlier
common-collar criterion first gives quasi-everywhere equality
for its fixed capacity domain. The atomic testers on that
same domain force equality at each divisorial point. The
present proposition then upgrades this to the all-domain
local quasi-everywhere assertion. Conversely that local
assertion implies the fixed-domain one on the compact support
of the common-collar difference by a finite cover and domain
monotonicity. Thus it implies the earlier energy criterion as well.
\end{proof}

\begin{remark}[Why Shilov control alone did not close the problem]
For a fixed psh function, the Shilov maximum principle controls
its values on an affinoid by its Shilov boundary. The one-Shilov
domain constructed at a divisorial point in this record need
not contain that point in its ordinary interior. Thus that
domain alone does not control the limsup over all ordinary
directions appearing in the displayed equality. A uniform
upper-control statement for bounded psh families at divisorial
points is needed in addition to this reduction. It is supplied
by Theorem~\ref{thm:r2-divisorial-upper-control}; the resulting
no-jump conclusion is Theorem~\ref{thm:r2-increasing-no-jump}.
\end{remark}


% END source: base_r3/base_r2/sections/divisorial_determination.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/divisorial_no_jump.tex
\section{Divisorial barriers and the absence of an upper-envelope jump}
\label{sec:r2-divisorial-no-jump}

The argument below supplies the missing ordinary-neighborhood control
at a divisorial point. It uses a local model potential whose
Monge--Amp\`ere measure has only one atom on that neighborhood. The
potential need not vanish only at that point: its zero set is an
affinoid domain. The proof uses the already established pointwise
ambient domination principle, not an assertion that negligible sets
are pluripolar.

All model constructions are over the fixed field $K$, with its original
absolute value and integration calibration. Write $R=K^\circ$.
Throughout this section $n\geq1$.

\begin{lemma}[An algebraic affine model inside a prescribed neighborhood]
\label{lem:r2-algebraic-one-shilov}
Let $Y/K$ be smooth integral affine of dimension $n$, let
$O\subset Y^{\an}$ be ordinary open, and let $z\in O$ be divisorial.
There is a normal, integral, flat, finite-type $R$-algebra $B$ with
the following properties. Its generic fiber
$Y_B=\operatorname{Spec}(B\otimes_RK)$ is a smooth affine algebraic
open in $Y$. The strict affinoid generic fiber
\[
 A=\bigl(\operatorname{Spf}\widehat B\bigr)_\eta
       \subset Y_B^{\an}
\]
of its uniformizer-adic completion is contained in $O$ and contains
$z$. The special fiber of $\operatorname{Spec}B$ has exactly one
irreducible component, and its generic point is the reduction of
$z$. In particular $\Gamma(A)=\{z\}$.
\end{lemma}

\begin{proof}
Choose a compact polynomial norm box $C$ with
\[
 z\in\operatorname{Int}(C),\qquad C\subset O.
\]
After clearing rational powers in its thresholds, write its finitely
many defining inequalities as
\[
 |a_i|\leq1\quad(1\leq i\leq s)
\]
on an affine algebraic open $Y_0\subset Y$. Here the $a_i$ are
regular on $Y_0$: an upper constraint uses a ratio $F^N/c$, and a
lower constraint uses $d/G^M$ after removing $G=0$. Bounds on affine
coordinates are included, so the displayed locus is the compact
box $C$. Such thresholds can be chosen strictly around the values
at $z$, since rational powers of $|K^\times|$ are dense in
$\mathbf R_{>0}$. In particular $|a_i(z)|\leq1$.

Take a normal projective compactification $\overline Y$ of $Y$.
Proposition~A.9 of \cite{GM19} realizes $z$ on a proper algebraic
$R$-model of $\overline Y$. Normalize that model. Normalization is
finite because the model is of finite type over the excellent DVR
$R$; it is an isomorphism on the normal generic fiber. The center
of $z$ is still the generic point $\eta_E$ of a special-fiber
component $E$. Equivalently, one can use the strict transform of
the component on a normal dominating model. Its height-one local
ring is the valuation ring of the same divisorial valuation.

For each $i$, the inequality $|a_i(z)|\leq1$ says exactly that
the rational function $a_i$ belongs to the DVR
$\mathcal O_{\mathscr Y,\eta_E}$. It is therefore regular on a
Zariski neighborhood of $\eta_E$. Intersect these finitely many
neighborhoods. Remove the other special-fiber components. Also
remove the closure of the generic boundary
$\overline Y\setminus Y_0$. This last removal does not remove
$\eta_E$. Indeed, the ideal $I$ of the schematic closure is
uniformizer-saturated. At the DVR
$\mathcal O_{\mathscr Y,\eta_E}$ its localization is nonzero,
since the boundary does not contain the generic point of
$\overline Y$. Any nonzero ideal of that DVR contains a power
of the uniformizer of $R$. Saturation then forces
$I_{\eta_E}=\mathcal O_{\mathscr Y,\eta_E}$, so the closure
does not contain $\eta_E$.

Choose an affine open $\operatorname{Spec}B$ containing $\eta_E$
inside the resulting open. Then $B$ is normal, flat and finite type
over $R$, its special fiber has only the component coming from
$E$, and $B\otimes_RK$ defines a smooth affine open $Y_B$ of
$Y_0$. All $a_i$ belong to $B$. Thus $|a_i|\leq1$ on the
generic fiber of $\operatorname{Spf}\widehat B$, so $A\subset C$.
It contains $z$ because its reduction is $\eta_E\in\operatorname{Spec}B$.

Normality supplies the unique point above this special-fiber
generic point, by \cite[Proposition~6.9.7]{CLD12}; the completion
has the corresponding relatively normal formal model, as $R$ is
excellent. The affine formal-model description of the Shilov
boundary in \cite[Proposition~A.3]{GM19} now gives
$\Gamma(A)=\{z\}$. No generic modification has changed $Y_0$:
the only normalization above is an isomorphism on the already
normal generic compactification. In particular $A\subset O$ is
an actual domain of the originally specified analytic space.
\end{proof}

\begin{lemma}[A local model barrier with a single divisorial atom]
\label{lem:r2-local-dirac-barrier}
Under the hypotheses of Lemma~\ref{lem:r2-algebraic-one-shilov},
there are a smooth affine algebraic open $Y_B\subset Y$, a
continuous semipositive PL function $\rho\in\Cb(Y_B^{\an})$,
and a real number $\epsilon>0$ such that
\[
 \rho\geq0,\qquad \rho(z)=0,\qquad
 \Omega=\{\rho<\epsilon\},\qquad
 \{\rho\leq2\epsilon\}\Subset O\cap Y_B^{\an},
\]
and, on the ordinary open neighborhood $\Omega$ of $z$,
\[
 (D\rho)^n=c_z\delta_z\qquad(c_z>0).
\]
Every sublevel $\{\rho\leq t\}$ is compact in $Y_B^{\an}$.
All statements about the measure are ordinary CLD-current
statements, with the accepted continuous BT compatibility.
\end{lemma}

\begin{proof}
Use $B$ from the preceding lemma and choose finitely many
$R$-algebra generators $g_1,\ldots,g_N$. Their closed embedding
\[
 \operatorname{Spec}B\hookrightarrow\mathbf A_R^N
\]
identifies its formal generic fiber with
\[
 A=\{|g_i|\leq1\text{ for every }i\}\subset Y_B^{\an}.
\]
Define
\[
 \rho=\log\max\{1,|g_1|,\ldots,|g_N|\}.
\]
The function is finite, continuous and classical after all complete
extensions: pullback to a curve is a finite maximum of logarithms
of analytic functions and the constant zero. It is semipositive PL.
Its zero set is exactly $A$. Closed sublevels are intersections of
the closed affine variety $Y_B^{\an}$ with compact closed
polydiscs, hence are compact.

Take the projective closure of $\operatorname{Spec}B$ in
$\mathbf P_R^N$, and normalize it. Denote this normal projective
flat model by $\mathscr Z$. Its normalization is finite, and its
standard affine chart $X_0\ne0$ is still $\operatorname{Spec}B$
because $B$ is normal. Let
\[
 \mathscr L=\nu^*\mathcal O_{\mathbf P_R^N}(1).
\]
It is ample and generated, and its model weight in the frame
$X_0$ on $Y_B$ is exactly $\rho$, with $g_i=X_i/X_0$.
The generic fiber is normal and generically smooth over the
characteristic-zero field $K$. The original CLD model formula
therefore applies, since the discretely valued field $K$ is stable;
see \cite[Theorem~6.9.3, \S6.9.6 and Proposition~6.9.7]{CLD12}.
It gives the finite measure
\[
 c_1(\mathscr L)^n
   =\sum_F b_F\deg_{\widetilde K,F}
                    ((\mathscr L|_F)^n)\,\delta_{z_F}.
\]
Here $F$ runs over special-fiber components, $b_F>0$ is its
multiplicity, and the degree retains its structural residue-field
degree. Ampleness makes every displayed coefficient positive.
The points $z_F$ lie in the analytification of every nonempty
generic algebraic open, so $\rho(z_F)$ is finite.

We determine exactly which of these atoms have height zero.
At the generic point of $F$, choose a local frame $e$ of
$\mathscr L$. The sections $X_i/e$ are regular in the
corresponding DVR, and at least one is a unit because the sections
generate $\mathscr L$. Consequently
\[
 \rho(z_F)
  =\log\max_i|X_i/X_0|(z_F)
  =-\log|X_0/e|(z_F).
\]
This equals zero precisely when $X_0/e$ is a unit, that is,
when the generic point of $F$ belongs to the affine chart
$X_0\ne0$. That chart has exactly one special-fiber component.
Its associated point is $z$. Thus $z$ is the only height-zero
atom, and every other atom has strictly positive height.

Choose $2\epsilon$ smaller than the finitely many positive
heights, if there are any. Shrink $\epsilon$ further so that
$\{\rho\leq2\epsilon\}\subset O$: the compact sublevels
decrease to $A\subset O$, so compactness gives this inclusion.
They are compact subsets of the open $O\cap Y_B^{\an}$, as
required. The restriction of the displayed model measure to
$\Omega=\{\rho<\epsilon\}$ is $c_z\delta_z$, where
\[
 c_z=b_E\deg_{\widetilde K,E}((\mathscr L|_E)^n)>0.
\]
On the smooth open $Y_B$ its model curvature measure is the
ordinary continuous BT product by the accepted compatibility
from R and B. This proves the claimed identity against all
original compact smooth scalar tests. No equation with a
prescribed mass has been solved, and no change of field or
normalization has been made.
\end{proof}

\begin{theorem}[Uniform upper control at a divisorial point]
\label{thm:r2-divisorial-upper-control}
Use $\rho,\epsilon,\Omega$ from
Lemma~\ref{lem:r2-local-dirac-barrier}. Given real numbers
$m\leq M$, every $u\in\Cb(\Omega)$ satisfying
\[
 u\leq M\text{ on }\Omega,\qquad u(z)\geq m
\]
obeys
\[
 u(x)\leq u(z)+A\rho(x)\quad(x\in\Omega),
 \qquad A=\frac{2(M-m+1)}\epsilon.
\]
In particular, a family with these same bounds is uniformly
upper semicontinuous at $z$ in the ordinary topology: for each
$\delta>0$ the ordinary neighborhood
$\Omega\cap\{A\rho<\delta\}$ works simultaneously for
every member of the family.
\end{theorem}

\begin{proof}
Put $v=u(z)+A\rho$. This is continuous classical psh and
\[
 (Dv)^n=A^n c_z\delta_z.
\]
On $\{\rho\geq\epsilon/2\}\cap\Omega$ one has
\[
 v\geq m+A\epsilon/2=M+1>u.
\]
Thus for every $\delta>0$,
\[
 \overline{\{v<u-\delta\}}
     \subset\{\rho\leq\epsilon/2\}\Subset\Omega.
\]
This verifies the compact-gap condition at every part of the
ordinary boundary, without treating the discontinuous strict
gap as an ordinary open set. At $z$ one has $v(z)=u(z)$, so
\[
 (Dv)^n(\{v<u\})=0.
\]
The domain $\Omega$ has compact closure in the smooth affine
chart $Y_B$, and therefore satisfies the framed affine
hypothesis of Corollary~\ref{cor:r2-pointwise-ambient}.
Apply that proved pointwise domination principle to the pair
$(v,u)$. It gives $v\geq u$ everywhere. Since $\rho$ is
continuous and vanishes at $z$, the simultaneous-neighborhood
assertion follows.
\end{proof}

The two geometric lemmas are direct consequences of model
realization, normality and the model-measure formula, with
the neighborhood construction made explicit. The uniform upper
control is recorded as \texttt{NEW\_PROVED} relative to the
previous ambient domination theorem. Its proof uses neither
the upper-envelope identification nor a polar-set theorem.

\begin{corollary}[Upper regularization preserves divisorial values]
\label{cor:r2-divisorial-no-jump}
Let $U$ have the allowed analytic hypotheses. Let
$\mathcal F\subset\Cb(U)$ be nonempty and locally uniformly
bounded above, and put
\[
 h=\sup_{u\in\mathcal F}u,\qquad U_h=h^*.
\]
Then
\[
 U_h(z)=h(z)\qquad\text{at every ambient divisorial }z\in U.
\]
The family can be uncountable; no measurability of $h$ is used
in this assertion.
\end{corollary}

\begin{proof}
Fix $z$ and pass to an ordinary smooth affine algebraic
neighborhood, using the accepted B, Lemma D.4. Choose
$u_0\in\mathcal F$ and replace every $u$ by $\max(u,u_0)$.
The supremum is unchanged. The new family is uniformly bounded
above and below on a compact neighborhood of $z$, since
$u_0$ is locally bounded. Apply the preceding theorem on a
smaller $\Omega$, with common constants $m,M,A$ and the
fixed barrier $\rho$. Taking the supremum gives
\[
 h(x)\leq h(z)+A\rho(x)\quad(x\in\Omega).
\]
Ordinary upper regularization at $z$, where $\rho(z)=0$,
yields $U_h(z)\leq h(z)$. The reverse inequality holds by
definition. All algebraic charts and barriers here are over
the original $K$; no geometric-integrality field change is
needed for this step.
\end{proof}

\begin{theorem}[Increasing envelopes have no ambient capacity jump]
\label{thm:r2-increasing-no-jump}
Let $u_j\in\Cb(U)$ increase and be locally uniformly bounded
above. Set $h=\sup_j u_j$ and $U_h=h^*$. Then
\[
 U_h\in\Cb(U),\qquad
 h=U_h\quad\text{locally outside sets of outer }
           \operatorname{Cap}_{[1]}\text{-capacity zero}.
\]
Moreover $u_j\to U_h$ locally in outer ambient capacity on
every permitted norm-box domain. If the sequence has one
common compact collar, its ambient energy satisfies
\[
 \mathcal E_b(U_h)=\sup_j\mathcal E_b(u_j).
\]
\end{theorem}

\begin{proof}
The established bounded upper-envelope theorem gives
$U_h\in\Cb(U)$. Actual increasing convergence gives local
ambient quasicontinuity of the Borel function $h$, and
quasicontinuity of $U_h$ is already known. The preceding
corollary identifies their values at every ambient divisorial
point. Proposition~\ref{prop:r2-divisorial-capacitary-order}
therefore gives the asserted local outer-capacity-zero equality.
Combine it with actual increasing convergence to $h$ and
subadditivity of outer capacity to obtain convergence to $U_h$.
The common-collar energy assertion is then
Proposition~\ref{prop:r2-envelope-energy-criterion}, for the
ambient background $[1]$.
\end{proof}

\begin{theorem}[Negligible upper-envelope sets have ambient outer capacity zero]
\label{thm:r2-negligible-outer-zero}
Let $\mathcal F\subset\Cb(U)$ be nonempty and locally uniformly
bounded above. Write
\[
 h=\sup_{u\in\mathcal F}u,\qquad U_h=h^*.
\]
Then the set $\{h<U_h\}$ has locally zero outer
$\operatorname{Cap}_{[1]}$-capacity, whether or not the raw
supremum $h$ is Borel. Locally, a countable family of finite
maxima of members of $\mathcal F$ has an increasing actual
supremum $H$ with
\[
 H\leq h\leq U_h,
 \qquad H=U_h\quad\text{locally ambient-quasi-everywhere}.
\]
\end{theorem}

\begin{proof}
Work on one permitted local norm-box domain $W$. The local
capacitary-envelope theorem, with background $[1]$, supplies
an increasing sequence of finite maxima of members of
$\mathcal F|_W$ whose actual Borel supremum $H$ belongs to
$\mathcal A_{[1],b}(W)$. It satisfies $H\leq h$ pointwise,
and every $u\in\mathcal F$ satisfies $u\leq H$ locally
ambient-quasi-everywhere.

Fix an ambient divisorial $z\in W$. Its singleton has
positive capacity on each sufficiently small framed affine
domain containing it. Thus the exceptional set for each
individual inequality $u\leq H$ cannot contain $z$.
Consequently $u(z)\leq H(z)$ for every $u\in\mathcal F$,
even though their exceptional sets may depend on $u$.
Taking the supremum and using $H\leq h$ gives
\[
 H(z)=h(z)=U_h(z),
\]
where the last equality is
Corollary~\ref{cor:r2-divisorial-no-jump}.

Both $H$ and $U_h$ are locally bounded, Borel and locally
ambient-quasicontinuous. Proposition~\ref{prop:r2-divisorial-capacitary-order}
gives $H=U_h$ locally outside a set of outer capacity zero.
The pointwise sandwich implies
\[
 \{h<U_h\}\subset\{H<U_h\}.
\]
Outer capacity is defined for arbitrary sets and is monotone,
so no Borel measurability of the left side is needed. This
proves the theorem locally on every requested domain.
\end{proof}

\paragraph{Scope and dependencies.}
The upper-envelope conclusions in this section are
\texttt{NEW\_PROVED} consequences of the model barrier and
the previously proved ambient calculus. They concern the
ambient capacity $\operatorname{Cap}_{[1]}$; the assertion
for unrestricted positive closed backgrounds remains false,
as the subvariety-supported example already shows. Zero
ambient outer capacity has been proved here. A converse
identification of all zero-capacity sets with pluripolar sets
is not used or asserted.

The proof order is model realization and the original model
formula, then the established pointwise ambient domination,
then uniform divisorial upper control. The quasicontinuous
divisorial criterion and actual increasing convergence are
used only after that control is proved. Thus no assertion
about negligible sets is inserted into its own proof.

% END source: base_r3/base_r2/sections/divisorial_no_jump.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/ambient_completion_representative.tex
\section{Classical representatives of the ambient capacitary closure}
\label{sec:r2-ambient-completion-representative}

The completion retains actual Borel functions. The result here
identifies their classes modulo ambient outer-capacity-zero sets;
it does not identify all of their actual point values. It is
specific to the ambient background $[1]$.

\begin{corollary}[The unique classical representative]
\label{cor:r2-ambient-classical-representative}
Let $U$ have the allowed analytic hypotheses. Every
$a\in\mathcal A_{[1],b}(U)$ has a unique function
$r_a\in\Cb(U)$ such that
\[
 a=r_a\quad\text{locally outside sets of outer }
          \operatorname{Cap}_{[1]}\text{-capacity zero}.
\]
At each ambient divisorial point $z$ one has $r_a(z)=a(z)$.
If $a\in\mathcal A^M_{[1]}(W)$ on a norm-box domain $W$,
then $|r_a|\leq M$ there. For any defining approximants
$a_j\in\Cb(W)$ with $|a_j|\leq M$, one formula is
\[
 r_a=\lim_{m\to\infty}
           \left(\sup_{j\geq m}a_j\right)^*.
\]
The right side is a decreasing limit of bounded classical psh
functions, and the final actual classical function is independent
of the approximants.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
The inputs are divisorial no-jump, the divisorial-capacitary order
criterion, and the already proved existence of capacitary limits.

\begin{proof}
First work on a defining domain $W$ with
$a\in\mathcal A^M_{[1]}(W)$ and approximants $a_j$ as stated.
Set
\[
 v_m=\left(\sup_{j\geq m}a_j\right)^*.
\]
The bounded upper-envelope theorem gives $v_m\in\Cb(W)$,
and $-M\leq v_m\leq M$. The tails decrease, so $v_m$ decreases
to a bounded classical psh function $r$ with the same bound.

Fix an ambient divisorial point $z\in W$ and choose a small
framed affine norm-box capacity domain containing $z$. Its
singleton has positive capacity by the atomic tester lemma.
Local outer-capacity convergence $a_j\to a$, applied to the
compact singleton $\{z\}$, therefore implies
$a_j(z)\to a(z)$. Otherwise one fixed positive error threshold
would put that singleton into infinitely many error sets,
contradicting their capacities tending to zero.
Corollary~\ref{cor:r2-divisorial-no-jump}, applied to each tail,
gives
\[
 v_m(z)=\sup_{j\geq m}a_j(z).
\]
It follows that $r(z)=a(z)$ at every ambient divisorial point.
The function $a$ is locally quasicontinuous by the completion
theorem; $r$ is locally quasicontinuous as a bounded classical
potential. Proposition~\ref{prop:r2-divisorial-capacitary-order}
therefore gives $r=a$ locally ambient-quasi-everywhere.

For uniqueness, suppose $r,s\in\Cb(W)$ have this property.
Positive singleton capacity forces $r(z)=s(z)$ at every
ambient divisorial point. Shilov boundaries of strict affinoid
neighborhoods consist of such points. The actual pointwise
order lemma, Lemma~\ref{lem:r2-shilov-gap}, applied in both
orders, gives $r=s$ at every point of $W$. This also proves
independence of the defining approximants.

For a general local member $a\in\mathcal A_{[1],b}(U)$,
perform the construction on each defining neighborhood.
On an ordinary overlap the two resulting classical functions
are both locally quasi-everywhere equal to $a$, hence are
actually equal by the uniqueness just proved. They glue to
$r_a\in\Cb(U)$ by the classical sheaf property, which retains
the all-extensions curve tests. Its divisorial values and
uniqueness are local and hence hold on $U$.
\end{proof}

\begin{remark}[The representative need not be the usc regularization of the Borel function]
Take a connected ordinary polydisc $W\subset\mathbf A_K^{n,\an}$,
$n\geq1$, containing the $K$-rational origin $o$, and put
$a=\mathbf1_{\{o\}}$. The singleton is locally pluripolar:
the classical function $\log|T_1|$ is not identically $-\infty$
and has a pole there. The already proved ambient polar theorem
therefore makes $\{o\}$ locally outer-capacity zero. Consequently
the constant zero sequence converges to $a$ in local outer
capacity, so $a\in\mathcal A_{[1],b}(W)$ and $r_a=0$.

The point $o$ is closed, so its indicator is upper
semicontinuous and $a^*=a$. This function is not classical psh:
it attains the strict value one at an interior point and is zero
elsewhere, contradicting the classical strong maximum principle
on the connected domain. Thus $r_a$ is not in general $a^*$.
For the actual supremum of an increasing classical sequence,
in contrast, the proved no-jump theorem does identify its
classical representative with that supremum's usc regularization.
This example uses only the ambient polar-zero theorem; it does
not depend on a converse from capacity zero to pluripolarity.
\end{remark}

\begin{corollary}[The completed ambient calculus uses the classical representative]
\label{cor:r2-ambient-completed-calculus}
For $0\leq k\leq n$ and
$a_0,\ldots,a_k\in\mathcal A_{[1],b}(U)$,
\[
 \mathcal T_{[1],k}(a_1,\ldots,a_k)
       =T_{[1],k}(r_{a_1},\ldots,r_{a_k}),
\]
and
\[
 a_0\mathcal T_{[1],k}(a_1,\ldots,a_k)
       =r_{a_0}T_{[1],k}(r_{a_1},\ldots,r_{a_k}).
\]
These are equalities of coefficient-Radon currents on every
original complementary smooth test form. A uniformly bounded
classical sequence that is locally Cauchy in ambient capacity
thus converges locally in outer capacity to a unique actual
classical function, with the established joint weighted BT
continuity whenever all the slots have the stated bounds.
\end{corollary}

\begin{proof}
The completed calculus is independent of changes on local
outer-capacity-zero sets and agrees with the classical calculus
on $\Cb$, by Theorem~\ref{thm:r2-completed-calculus}.
This gives the unweighted identity. The same theorem's
coefficient bounds by outer capacity make every coefficient
measure of this product vanish on the local disagreement set
of $a_0$ and $r_{a_0}$, proving the weighted identity as a
Borel multiplication identity. A local capacity-Cauchy sequence
has a limit in the completed class; the first corollary
identifies its unique classical representative, and zero-capacity
changes of the limit preserve convergence. The weighted
convergence assertion is then the already established completed
calculus theorem.
\end{proof}

No corresponding conclusion for an arbitrary background $S$
is claimed. For example, a background supported on an ambient
pluripolar subvariety can distinguish the actual increasing
limit from its ambient upper regularization. The relative
capacitary closure remains necessary in that setting.

% END source: base_r3/base_r2/sections/ambient_completion_representative.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/dirac_dirichlet.tex
\section{A Dirac Dirichlet problem on the constructed divisorial domains}
\label{sec:r2-dirac-dirichlet}

\begin{corollary}[Constant boundary data and one divisorial atom]
\label{cor:r2-dirac-dirichlet}
Let $z,\rho,\epsilon,\Omega$ and $c_z$ be supplied by
Lemma~\ref{lem:r2-local-dirac-barrier}. For every real
$m>0$ and $b\in\mathbf R$, the function
\[
 u_{m,b}=b+\left(\frac m{c_z}\right)^{1/n}
                         (\rho-\epsilon)
\]
is continuous on a neighborhood of $\overline\Omega$,
belongs to $\Cb(\Omega)$, and satisfies
\[
 (Du_{m,b})^n=m\delta_z\quad\text{on }\Omega,
 \qquad u_{m,b}=b\quad\text{on }\partial\Omega.
\]
Here the closure and boundary are ordinary ones in the smooth
affine chart $Y_B^{\an}$. It is the unique bounded classical
solution with this measure and the boundary condition
\[
 \lim_{\substack{x\to y\\x\in\Omega}}w(x)=b
                   \quad(y\in\partial\Omega).
\]
In particular uniqueness holds among classical solutions
continuous on $\overline\Omega$ with boundary value $b$.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
The domain and its barrier have been constructed together.
This corollary concerns that family of domains and a single
divisorial atom. It makes no prescribed-domain assertion for
general atomic or diffuse measures.

\begin{proof}
Positive scaling and translation preserve classicality,
and the proved homogeneity of the BT product gives
\[
 (Du_{m,b})^n=\frac m{c_z}(D\rho)^n=m\delta_z.
\]
Continuity is inherited from $\rho$. Since
$\Omega=\{\rho<\epsilon\}$ and $\rho$ is continuous,
each point of its ordinary boundary satisfies
$\rho=\epsilon$. This proves the boundary value.

Suppose $w$ is another solution with the stated boundary
limit. Both $w$ and $u_{m,b}$ have the same finite boundary
limit at every point. For each $\delta>0$, every boundary
point therefore has an ordinary neighborhood on which their
difference has absolute value less than $\delta$, at points
of $\Omega$. The boundary is compact, being a closed subset
of the compact $\overline\Omega$. A finite union of these
neighborhoods covers it. Consequently both strict-gap sets
with margin $\delta$ have closure compactly contained in
$\Omega$. The equality of the Monge--Amp\`ere measures
and Corollary~\ref{cor:r2-pointwise-ambient} now give
$w=u_{m,b}$ at every point.
\end{proof}

% END source: base_r3/base_r2/sections/dirac_dirichlet.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/ambient_extremals.tex
\section{Ambient extremals on domains with a bounded negative exhaustion}
\label{sec:r2-ambient-extremals}

This section concerns $S=[1]$. The domain hypothesis is explicit
and weaker than a strictly psh exhaustion. Its conclusions use
the proved divisorial no-jump theorem; none is claimed here for
an arbitrary positive closed background.

\begin{definition}[The exhaustion domains used in this section]
\label{def:r2-exhaustion-domain}
Let $Y/K$ be smooth affine of dimension $n\geq1$. An exhaustion
domain is a connected ordinary open $V\Subset Y^{\an}$ with
nonempty ordinary boundary and a function
\[
 \rho\in\Cb(V)\cap C(\overline V),\qquad
 \rho<0\text{ on }V,\qquad \rho=0\text{ on }\partial V.
\]
In particular $\rho$ is bounded and
$\{\rho\leq-\epsilon\}\Subset V$ for every $\epsilon>0$.
No positivity stronger than classical psh is required of $\rho$,
and no equation for $(D\rho)^n$ is assumed.
\end{definition}

\begin{lemma}[Local supply of exhaustion domains]
\label{lem:r2-local-exhaustion-domains}
Every point in an allowed ordinary algebraic neighborhood has
arbitrarily small exhaustion domains of the preceding type.
This does not assert that every ordinary analytic domain is an
exhaustion domain.
\end{lemma}

\begin{proof}
Inside a prescribed ordinary neighborhood $O$, choose a compact
rational polynomial norm box $C$ containing the point in its
ordinary interior. After passing to an affine algebraic open
$Y_0$ and clearing powers in the rational thresholds, write
\[
 C=\{|a_i|\leq1\text{ for }1\leq i\leq N\}.
\]
The regular functions $a_i$ include the required bounds on
affine coordinates and on inverses used for lower constraints.
Consequently each locus $\{|a_i|\leq e^t\text{ for all }i\}$,
with $t\geq0$, is compact in $Y_0^{\an}$. Equivalently, one
can include scaled powers of a set of affine algebra generators.
Put
\[
 \rho_0=\log\max(1,|a_1|,\ldots,|a_N|).
\]
This is continuous classical psh by the finite maximum rule
after complete extensions and curve pullbacks. Its compact
sublevels decrease to $C$ as $t\downarrow0$. Choose a small
positive rational valuation threshold $\epsilon$ such that
$\{\rho_0\leq2\epsilon\}\subset O$.

Analytic spaces are locally connected, as recalled in
\cite{GR25} from Berkovich's local connectedness theorem.
Let $V$ be the connected component containing the specified
point of $\{\rho_0<\epsilon\}$. It is ordinary open, and its
compact closure is contained in $\{\rho_0\leq\epsilon\}$.
Since a connected component is relatively closed in that
ordinary open, its boundary is contained in
$\{\rho_0=\epsilon\}$. Thus $\rho=\rho_0-\epsilon$ is the
required bounded continuous negative exhaustion.

For completeness the boundary is nonempty. Otherwise $V$
would be a compact open-and-closed subdomain of the strict
affinoid $\{\rho_0\leq2\epsilon\}$, hence an affinoid
component. It is pure of positive dimension and therefore has
a Shilov point on its analytic boundary, by
\cite[Lemma~6.2]{GR25}. This contradicts its being an ordinary
open in the boundaryless space $Y_0^{\an}$ and the
interior-composition rule \cite[\S2.11(3)--(4)]{GR25}.
\end{proof}

The lemma is a \texttt{DIRECT\_COROLLARY} of the ordinary
norm-box geometry and classical maximum calculus already used
in the record. It does not supply strict positivity or peak
barriers at every boundary point.

Fix an exhaustion domain $V$ for the rest of this section and
keep the capacity domain fixed:
\[
 C(E)=\operatorname{Cap}_{[1]}(E;V),\qquad C^*(E)=C_V^*(E).
\]
For any subset $E\subset V$, define
\[
 \mathcal F_E=\{v\in\Cb(V):-1\leq v\leq0,\ v=-1\text{ on }E\},
 \qquad h_E=\sup_{v\in\mathcal F_E}v,\qquad U_E=h_E^*.
\]
The constraints in $\mathcal F_E$ are actual pointwise
constraints, including when $E$ is not Borel. The family
contains the constant $-1$. When $E=\varnothing$, $U_E=0$.

\begin{lemma}[Basic extremal properties and a countable selection]
\label{lem:r2-extremal-basics}
One has $U_E\in\Cb(V)$ and $-1\leq U_E\leq0$.
If $E\subset F$, then $U_F\leq U_E$ pointwise.
The exceptional subset
\[
 E\cap\{U_E>-1\}
\]
has locally zero outer ambient capacity. If $E\Subset V$,
then $U_E$ has ordinary boundary limit zero, and the
exceptional subset has $C^*$-capacity zero in the fixed domain.
There is an increasing sequence $u_j\in\mathcal F_E$ with
\[
 \left(\sup_j u_j\right)^*=U_E,\qquad
 u_j\longrightarrow U_E
   \quad\text{locally in outer ambient capacity}.
\]
\end{lemma}

\begin{proof}
The bounded upper-envelope theorem gives classicality.
Monotonicity follows from inclusion of the admissible families.
Since $h_E=-1$ on $E$, the exceptional subset is contained
in $\{h_E<U_E\}$. The arbitrary-family negligible-set theorem
gives its local outer-capacity zero, without requiring that
$h_E$ or $E$ be Borel. On a relatively compact $E$, a finite
local cover and domain monotonicity give zero for the fixed
domain $V$.

If $E\Subset V$, choose $c>0$ such that $c\rho\leq-1$ on
$\overline E$. The function $\max(-1,c\rho)$ belongs to
$\mathcal F_E$. It tends to zero at every ordinary boundary
point, and
\[
 \max(-1,c\rho)\leq U_E\leq0.
\]
This proves the boundary assertion. The same lower barrier
works simultaneously for all $E$ contained in one fixed
compact subset of $V$.

Here is an elementary countable selection, independent of
Choquet's capacitability theorem. The finite $p$-adic field
makes $V$ second countable. For a countable ordinary basis
$(B_l)$, let
\[
 M_l=\sup\{v(x):v\in\mathcal F_E,\ x\in B_l\}.
\]
For each $l,m$ choose $v_{l,m}\in\mathcal F_E$ and
$x_{l,m}\in B_l$ with $v_{l,m}(x_{l,m})>M_l-1/m$.
Enumerate these functions and take their successive finite
maxima $u_j$. They remain in $\mathcal F_E$.
For $a=\sup_j u_j$, one has
$\sup_{B_l}a=M_l=\sup_{B_l}h_E$ for each $l$.
The neighborhood formula for upper regularization therefore
gives $a^*=h_E^*=U_E$. The proved increasing no-jump theorem
then gives the claimed local outer-capacity convergence.
\end{proof}

\begin{lemma}[Order of total Monge--Amp\`ere masses]
\label{lem:r2-total-mass-order}
Let $u,v\in\Cb(V)$ be bounded, with
$u\leq v\leq0$ and ordinary boundary limits zero. Then
\[
 \int_V(Dv)^n\leq\int_V(Du)^n.
\]
The assertion allows infinite masses and contains no
subtraction of infinities.
\end{lemma}

\begin{proof}
If $v\equiv0$ the assertion is immediate. Otherwise the
strong maximum principle on the connected domain gives
$v<0$ everywhere. For $0<t<1$, put $w=(1-t)v$.
Then $w>v\geq u$ on all of $V$. The boundary limits imply
\[
 \overline{\{u<w-\epsilon\}}\Subset V\qquad(\epsilon>0).
\]
Indeed each such gap is absent on a neighborhood of the
compact ordinary boundary. The comparison principle gives
\[
 (1-t)^n\int_V(Dv)^n
  =\int_{\{u<w\}}(Dw)^n
  \leq\int_{\{u<w\}}(Du)^n
  =\int_V(Du)^n.
\]
Let $t\downarrow0$.
\end{proof}

\begin{theorem}[The compact extremal and its equilibrium mass]
\label{thm:r2-compact-extremal}
For every compact $K\Subset V$, the function $U_K$ has
ordinary boundary limit zero and is locally maximal on
$V\setminus K$. Its measure
\[
 \mu_K=(DU_K)^n
\]
is supported in $K$, is finite, and satisfies
\begin{equation}
\label{eq:r2-compact-capacity-mass}
 C(K)=\mu_K(K)=\mu_K(V).
\end{equation}
The equality $U_K=-1$ on $K$ is asserted outside a
$C^*$-zero subset; it is not asserted at every point of $K$.
\end{theorem}

\begin{proof}
The boundary statement and the zero-capacity exceptional set
are in Lemma~\ref{lem:r2-extremal-basics}. Choose there an
increasing sequence $u_j\in\mathcal F_K$ whose upper-regularized
supremum is $U_K$.

We first prove that the measure vanishes off $K$.
Work in a small ordinary $W\Subset V\setminus K$ on which
the constructive common projective globalization applies,
and fix a nonnegative original smooth scalar $\chi$ compactly
supported in $W$. Globalize the whole bounded family $u_j$
on one identified ample line, obtaining bounded classical
metrics $R_j$ whose weights on a neighborhood of
$\operatorname{supp}\chi$ are $u_j+c$ with one fixed $c$.
The construction is a maximum with one common collar, so
the metrics increase and have one common bound.
Extend $\chi$ by zero to the projective space. For a fixed
$\epsilon>0$, put
\[
 p_j=P_L^{\mathrm b}(R_j+\epsilon\chi).
\]
The exact obstacle inequalities give
$R_j\leq p_j\leq R_j+\epsilon\chi$, hence $p_j=R_j$
off $\operatorname{supp}\chi$. Envelope monotonicity makes
$p_j$ increasing.

Its local weight minus $c$ glues with $u_j$ off that support
to a function $w_j\in\Cb(V)$. This modification is compact
and uniformly bounded. It satisfies $w_j\geq u_j\geq-1$,
and has boundary limsup at most zero. The boundary maximum
principle gives $w_j\leq0$. Since the modification is away
from $K$, the constraint $w_j=-1$ on $K$ holds at every
point. Therefore $w_j\in\mathcal F_K$, and
\[
 u_j\leq w_j\leq h_K,\qquad
 \left(\sup_jw_j\right)^*=U_K.
\]
Both increasing rows converge locally in outer capacity
to $U_K$ by the no-jump theorem.

The proved quasicontinuous-obstacle orthogonality gives,
in local weights,
\[
 0=\int_W(u_j+\epsilon\chi-w_j)(Dw_j)^n .
\]
The integrand vanishes off the fixed compact support of
$\chi$. Insert a fixed smooth cutoff equal to one near
that support. Weighted increasing BT convergence sends
the terms with weights $u_j$ and $w_j$ to the same limit,
while weak convergence handles the fixed continuous
weight $\chi$. Consequently
\[
 \epsilon\int_W\chi(DU_K)^n=0.
\]
If the globalization uses B's finite constant-field
convention, its local calibration multiplies this zero
identity by a positive constant, exactly as in the
maximality proof. The conclusion is for the original
$K$-measure. Such $W$ cover $V\setminus K$, and all original
nonnegative compact scalar tests are allowed there.
Thus $\mu_K$ is zero on $V\setminus K$.
It is finite by local CLN on $K$, and local maximality
follows from the proved homogeneous-equation criterion.

This argument deliberately lifts the ACTUAL admissible
functions $u_j$. Directly modifying $U_K$ and treating a
quasi-everywhere obstacle inequality as a pointwise
admissibility condition would not be justified.

Since $U_K+1$ is a normalized test on the original $V$,
$\mu_K(K)\leq C(K)$. For the reverse inequality take
$v\in\Cb(V)$ with $0\leq v\leq1$ and $0<t<1$, and set
\[
 w=(1-t)v-1+t/2.
\]
It lies between $-1+t/2$ and $-t/2$. On $K$ outside the
exceptional zero-capacity set, $U_K=-1<w$.
That exceptional Borel set has zero $(Dv)^n$-mass, because
$(Dv)^n(E)\leq C(E)\leq C^*(E)$.
The boundary limit $U_K=0$ gives the compact-gap condition
for $(U_K,w)$. Comparison and the established support give
\[
 (1-t)^n\int_K(Dv)^n
 \leq\int_{\{U_K<w\}}(Dw)^n
 \leq\int_{\{U_K<w\}}\mu_K
 \leq\mu_K(K).
\]
Take the supremum over normalized $v$ and let $t\downarrow0$.
This proves \eqref{eq:r2-compact-capacity-mass}.
\end{proof}

\begin{theorem}[Compact continuity and compact outer regularity]
\label{thm:r2-compact-capacity-continuity}
If $K_j\Subset V$ are decreasing compact sets with
$K=\bigcap_jK_j$, then
\[
 C(K_j)\downarrow C(K),\qquad
 (\,\sup_j U_{K_j}\,)^*=U_K.
\]
Moreover $U_{K_j}\to U_K$ locally in outer ambient capacity
and their Monge--Amp\`ere measures converge weakly on all
original compact scalar tests. Every compact $K\Subset V$
satisfies
\[
 C^*(K)=C(K).
\]
All these capacities use the same original domain $V$.
\end{theorem}

\begin{proof}
The case of an empty intersection is eventually empty and
is immediate by compactness. Otherwise put
$U=(\sup_jU_{K_j})^*$. Monotonicity gives $U\leq U_K$.
For $v\in\mathcal F_K$ and $\epsilon>0$, upper
semicontinuity makes
$O_\epsilon=\{v<-1+\epsilon\}$ an ordinary open
neighborhood of $K$. Nested compactness gives
$K_j\subset O_\epsilon$ for all sufficiently large $j$.
The function
\[
 v_\epsilon=\max(-1,v-\epsilon)
\]
then belongs to $\mathcal F_{K_j}$. Hence
$v_\epsilon\leq U_{K_j}\leq U$. Letting
$\epsilon\downarrow0$ gives $v\leq U$.
Taking the supremum and upper regularization gives
$U_K\leq U$, proving equality.

The increasing no-jump and BT theorems now give the stated
capacity and measure convergence. Every measure involved
is supported in the one compact set $K_1$. Testing with
a compact smooth cutoff equal to one near $K_1$ therefore
gives convergence of their total masses. Formula
\eqref{eq:r2-compact-capacity-mass} proves compact
continuity of the raw capacity.

Choose decreasing compact neighborhoods $L_j\Subset V$
of $K$ with intersection $K$, using local compactness
and metrizability. Since $K\subset\operatorname{Int}L_j$,
\[
 C(K)\leq C^*(K)
 \leq C(\operatorname{Int}L_j)
 \leq C(L_j)\longrightarrow C(K).
\]
This proves compact outer regularity without interchanging
an infimum of neighborhoods with a supremum of individual
Radon measures.
\end{proof}

\begin{theorem}[Open-set extremals and their total mass]
\label{thm:r2-open-extremal}
Let $G\Subset V$ be ordinary open. Then $U_G=-1$ at every
point of $G$, $U_G$ has ordinary boundary limit zero on
$\partial V$, and
\[
 \operatorname{supp}(DU_G)^n\subset\partial_VG,
 \qquad
 \int_V(DU_G)^n=C(G).
\]
Here $\partial_VG=\overline G\setminus G$ is the ordinary
relative boundary. In particular the mass formula is a
TOTAL-mass formula; $(DU_G)^n(G)=0$.
If $K_m\Subset G$ increase with
$G=\bigcup_m\operatorname{Int}K_m$, then
$U_{K_m}\downarrow U_G$ pointwise.
\end{theorem}

\begin{proof}
The family constraint makes $h_G=-1$ on the ordinary open
$G$, so its upper regularization is also exactly $-1$
there. The exhaustion barrier from
Lemma~\ref{lem:r2-extremal-basics}, using $\overline G$,
gives the boundary limit zero.

Choose the indicated compact exhaustion and put
$U=\lim_m U_{K_m}$. This is a bounded decreasing limit
of classical psh functions and is therefore classical.
For each $x\in G$, one has $x\in\operatorname{Int}K_m$
for some $m$. On that interior $U_{K_m}=-1$ exactly,
by the same open-set regularization observation.
Thus $U=-1$ on $G$, and $U\in\mathcal F_G$.
Conversely $\mathcal F_G\subset\mathcal F_{K_m}$ for
every $m$, so $U_G\leq U_{K_m}$ and hence $U_G\leq U$.
The admissibility of $U$ gives the reverse inequality.

The decreasing BT theorem gives convergence of the measures.
They are all supported in the fixed compact $\overline G$.
A cutoff equal to one near it gives convergence of total
masses. Since the raw capacity is continuous from below,
\[
 \int_V(DU_G)^n
 =\lim_m\int_V(DU_{K_m})^n
 =\lim_m C(K_m)=C(G).
\]
The limit measure is supported in $\overline G$. Ordinary
locality and the actual equality $U_G=-1$ on $G$ make
the measure zero on $G$, proving the boundary support.
\end{proof}

\paragraph{Status and scope.}
The compact and open extremal mass formulas and compact
continuity are \texttt{NEW\_PROVED} on the displayed
exhaustion domains, relative to the reviewed ambient no-jump
and obstacle-orthogonality theorems. The results do not
identify every arbitrary ordinary domain as such a domain,
do not assert a strictly psh exhaustion, and do not transfer
these mass formulas to an arbitrary background $S$.
Compact outer regularity is proved here for the raw ambient
capacity. Full Borel outer regularity and Choquet
capacitability require the additional steps stated in the
subsequent section.

% END source: base_r3/base_r2/sections/ambient_extremals.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/polar_converse_choquet.tex
\section{Polar witnesses and capacitability from ambient extremals}
\label{sec:r2-polar-converse-choquet}

\noindent\textbf{Status: \texttt{NEW\_PROVED}, relative to the adopted
inputs and the preceding reviewed R2 theorems.}
The witness construction, arbitrary-set extremal mass formula, and
ordinary-open proof of increasing continuity are supplied below.
The Choquet and patching consequences are
\texttt{DIRECT\_COROLLARY}. The two theorem groups
\textup{(N)} and \textup{(E)} are separated to display dependencies;
both have been proved in the preceding sections. Neither a general
Josefson theorem nor Borel outer regularity is an input.

\subsection{Domain and the two theorem inputs}

Let $V$ be a connected ordinary open with compact closure in a smooth
affine algebraic chart over the fixed finite extension $K/\mathbf Q_p$.
Assume that its ordinary boundary is nonempty and that there is
\[
 \rho\in\Cb(V)\cap C(\overline V),\qquad
 \rho<0\text{ on }V,\qquad \rho=0\text{ on }\partial V.
\]
Thus each negative sublevel of $\rho$ has compact closure in $V$.
All capacities in this section use the ambient background $[1]$ and
the \emph{fixed admissibility domain $V$}. Write
\[
 C(E)=\operatorname{Cap}_{[1]}(E;V),\qquad
 C^*(E)=\inf_{\substack{G\supset E\\G\text{ ordinary open in }V}}C(G).
\]
The first notation is initially for Borel sets; the second is defined
for arbitrary subsets. All Monge--Amp\`ere measures and their weak
limits use the original CLD scalar test functions.

For any subset $A\subset V$, define
\[
 \mathcal F_A=\{v\in\Cb(V):-1\leq v\leq0,\ v=-1\text{ on }A\},
 \qquad h_A=\sup_{v\in\mathcal F_A}v,\qquad U_A=h_A^*.
\]
The ordinary upper regularization $U_A$ is bounded classical by
Theorem~\ref{thm:r2-local-upper-envelope}. The conditions $v=-1$
and $v\leq-1$ are identical for this normalized family. The raw
supremum is never substituted for its regularization without a
stated argument.

The two theorem groups are recorded separately to make the dependency
order explicit:
\begin{enumerate}[label=\textup{(\Alph*)}]
\item[\textup{(N)}] For every locally uniformly upper-bounded family
of bounded classical psh functions, the difference between its actual
supremum and its ordinary usc regularization is contained locally in
a Borel set of ambient outer capacity zero. In particular a locally
uniformly bounded increasing sequence has its actual supremum equal
to its regularization outside such a set, and their values agree at
every ambient divisorial point. Local and fixed-domain null classes
are identified only through
Corollary~\ref{cor:r2-fixed-domain-null-classes}.
These are Theorem~\ref{thm:r2-negligible-outer-zero},
Theorem~\ref{thm:r2-increasing-no-jump}, and
Corollary~\ref{cor:r2-divisorial-no-jump}.

\item[\textup{(E)}] If $G\Subset V$ is ordinary open, then $U_G=-1$
at every point of $G$, $U_G$ has ordinary boundary limit zero,
$\operatorname{supp}(DU_G)^n\subset\overline G$, and
\[
 \int_V(DU_G)^n=C(G).
\]
For a compact $K\Subset V$, the corresponding compact extremal
identity is
\[
 \int_V(DU_K)^n=C(K).
\]
The open-set identity concerns \emph{total mass}. It does not assert
$\int_G(DU_G)^n=C(G)$: the function is constant on $G$.
These are Theorems~\ref{thm:r2-open-extremal} and
\ref{thm:r2-compact-extremal}, with the domain and boundary
properties of Lemma~\ref{lem:r2-extremal-basics}.
\end{enumerate}

For every fixed $A\subset W\Subset V$, choose $a>0$ sufficiently
large that $a\rho\leq-1$ on $\overline W$. Then
\[
 b=\max(-1,a\rho)\in\mathcal F_A,
 \qquad b\leq U_A\leq0.
\]
This gives the ordinary boundary limit zero of $U_A$, uniformly for
all such $A$. In particular $\{U_A<-1+\eta\}\Subset V$ for every
$0<\eta<1$, with a common compact bound for fixed $W$ and $\eta$.
The capacity of $W$ is finite by relative CLN on a slightly larger
compact stage.

We also use the total-mass order of
Lemma~\ref{lem:r2-total-mass-order}: if $u\leq v\leq0$ are bounded
classical functions with ordinary boundary limits zero, then
$\int_V(Dv)^n\leq\int_V(Du)^n$, including infinite masses.

\subsection{A nontrivial pole witness for an outer-capacity-zero set}

\begin{theorem}[A witness on the specified exhaustion domain]
\label{thm:r2-capacity-zero-polar-witness}
Let $E\subset W\Subset V$ be an arbitrary subset with $C^*(E)=0$.
For every fixed ambient divisorial point $z\in V$ there is a nonpositive
extended classical psh function $\phi$ on $V$ such that
\[
 E\subset\{\phi=-\infty\},\qquad -1\leq\phi(z)\leq0.
\]
In particular the witness is not identically $-\infty$.
\end{theorem}

\begin{proof}
Choose an ordinary $W'$ with
$\overline W\subset W'\Subset V$. From $C^*(E)=0$ choose ordinary
opens $O_j\supset E$ with $C(O_j)<2^{-j}$ and set
\[
 G_j=W'\cap\bigcap_{i=1}^jO_i,\qquad u_j=U_{G_j}.
\]
The $G_j$ decrease, while $-1\leq u_j\uparrow h\leq0$.
They share a lower barrier $b=\max(-1,a\rho)$ as above.
Their measures are supported in the one compact set
$\overline W'$, and their total masses tend to zero by \textup{(E)}.

Put $U=h^*\in\Cb(V)$. By \textup{(N)}, $h=U$ locally outside
sets of outer capacity zero. The actual increasing-limit theorem
\ref{cor:r2-local-increasing}, the completed calculus
\ref{thm:r2-completed-calculus}, and independence under such changes
of representatives give
\[
 (Du_j)^n\longrightarrow(DU)^n
\]
on all original compact scalar tests. Their vanishing total masses
imply $(DU)^n=0$. Also $b\leq U\leq0$, so $U$ has ordinary
boundary limit zero. Pointwise ambient domination
\ref{cor:r2-pointwise-ambient}, applied to $U$ and $0$, now gives
$U=0$ on $V$. Indeed all sets $\{U<-\epsilon\}$ are relatively
compact by the lower barrier, and $(DU)^n$ vanishes on the whole
strict gap.

By the divisorial part of \textup{(N)}, $u_j(z)\uparrow0$.
Choose a strictly increasing sequence $j_l$ such that
$u_{j_l}(z)>-2^{-l}$. The finite sums
\[
 \phi_m=\sum_{l=1}^m u_{j_l}
\]
are nonpositive classical psh and decrease with $m$. Their pointwise
limit $\phi$ is extended classical psh by the classical decreasing
closure, and $\phi(z)\geq-\sum_l2^{-l}=-1$.
For $x\in E$, every $G_j$ contains $x$ and the open-set extremal
satisfies $u_j(x)=-1$ exactly. Thus $\phi_m(x)=-m$ and
$\phi(x)=-\infty$. No integrability of the original set or a
global pole-patching theorem has been used.
\end{proof}

\begin{corollary}[The local zero-set equivalence on this domain]
\label{cor:r2-local-polar-equivalence}
On the same exhaustion domain, for every arbitrary $E\subset W\Subset V$,
\[
 C^*(E)=0
 \quad\Longleftrightarrow\quad
 E\subset\{\phi=-\infty\}
 \text{ for some nonpositive classical }\phi\text{ on }V,
 \ \phi\not\equiv-\infty.
\]
The implication from right to left is the already proved ambient
polar theorem, not a new nonconcentration hypothesis.
\end{corollary}

\begin{proof}
The forward implication is the theorem. For the reverse implication,
Corollary~\ref{cor:r2-ambient-polar-outer-zero} gives local outer
capacity zero on every compact/domain pair. The fixed-domain
comparison \ref{cor:r2-fixed-domain-null-classes} gives $C^*(E)=0$.
\end{proof}

\begin{definition}[Local negligibility in the ordinary topology]
\label{def:r2-local-negligibility}
Let $E$ be an arbitrary subset of an allowed ordinary open $U$.
It is locally negligible if every $x\in U$ has an ordinary
neighborhood $O\subset U$ and a nonempty locally uniformly
upper-bounded family $\mathcal F\subset\Cb(O)$ such that
\[
 E\cap O\subset
 \left\{\sup_{v\in\mathcal F}v<
                    \left(\sup_{v\in\mathcal F}v\right)^*\right\}.
\]
It is locally increasing-negligible if the family can be chosen as
one increasing locally uniformly bounded sequence in $\Cb(O)$.
Local pluripolarity means that every $x\in U$ has a connected ordinary
neighborhood $O\subset U$ and an extended classical psh function
$\phi$ on $O$, not identically $-\infty$ there, with
$E\cap O\subset\{\phi=-\infty\}$.
Local outer-capacity zero means that
\[
 \operatorname{Cap}_{[1]}^*(E\cap H;\Omega)=0
\]
for every permitted ordinary norm-box domain $\Omega\subset U$
and every compact $H\Subset\Omega$.
\end{definition}

\begin{corollary}[Local polar, negligible, and outer-zero sets]
\label{cor:r2-polar-negligible-equivalence}
On every allowed ordinary open $U$, the following properties of an arbitrary subset
$E\subset U$ are equivalent:
\begin{enumerate}[label=\textup{(\roman*)}]
\item $E$ is locally pluripolar;
\item $E$ has local ambient outer-capacity zero;
\item $E$ is locally increasing-negligible;
\item $E$ is locally negligible.
\end{enumerate}
These are local witness assertions. A single global psh witness on
$U$ is not asserted.
\end{corollary}

\begin{proof}
The implication (i)$\Rightarrow$(ii) is
Corollary~\ref{cor:r2-ambient-polar-outer-zero}.
For (ii)$\Rightarrow$(i), fix $x\in U$ and use
Lemma~\ref{lem:r2-local-exhaustion-domains} in an ordinary algebraic
neighborhood to choose an exhaustion domain $V$ containing $x$.
Choose a connected ordinary $W$ with $x\in W\Subset V$ and an
ambient divisorial point $z\in W$. The local null-class comparison
gives $\operatorname{Cap}_{[1]}^*(E\cap W;V)=0$.
Theorem~\ref{thm:r2-capacity-zero-polar-witness} supplies a
nonpositive classical witness $\phi$ on $V$, finite at $z$, whose
pole set contains $E\cap W$. Its restriction to $W$ is a
nontrivial classical witness, as required.

For (i)$\Rightarrow$(iii), shrink a connected witness neighborhood
if necessary and subtract a finite upper bound so that $\phi\leq0$.
The pole set $P=\{\phi=-\infty\}$ has empty ordinary interior by
Proposition~\ref{prop:r2-no-interior-poles}; hence the witness remains
nontrivial on every smaller connected ordinary neighborhood.
The actual bounded classical functions
\[
 w_N=1+\frac{\max(\phi,-N)}{N}
\]
increase to $\mathbf1_{P^c}$. Because $P$ has empty ordinary
interior, the ordinary usc regularization of this actual limit is
the constant one. Its discrepancy set is exactly $P$, which
contains the specified part of $E$. This proves increasing
negligibility. The implication (iii)$\Rightarrow$(iv) is immediate.

Finally (iv)$\Rightarrow$(ii) follows from
Theorem~\ref{thm:r2-negligible-outer-zero}. To express it on a
prescribed compact/domain pair $(H,\Omega)$, restrict the local
witness families to smaller permitted domains contained in $\Omega$.
Choose countably many ordinary cores covering $E\cap H$; such a
choice exists in the locally second-countable finite $p$-adic
setting. Each core intersection has outer capacity zero on its
own domain and therefore on $\Omega$ by domain monotonicity.
Countable subadditivity gives the required zero capacity on $H$.
No uncountable union of exceptional sets is taken.
\end{proof}

\begin{corollary}[Pole patching on one exhaustion domain]
\label{cor:r2-exhaustion-domain-pole-patching}
On the same connected exhaustion domain $V$, let $E\subset V$
have local ambient outer-capacity zero. In particular this applies
to a locally pluripolar $E$. Then there is one nonpositive extended
classical psh function $\phi$ on $V$, not identically $-\infty$,
such that $E\subset\{\phi=-\infty\}$. No relative compactness
of $E$ itself is required for this conclusion.
\end{corollary}

\noindent\textbf{Provenance: \texttt{DIRECT\_COROLLARY}.}
The domain $V$ and its exhaustion hypothesis are retained.

\begin{proof}
Choose ordinary $W_m\Subset V$ increasing to $V$ and fix one
ambient divisorial point $z\in V$. Put $E_m=E\cap W_m$.
The fixed-domain null-class comparison gives $C^*(E_m)=0$.
The witness theorem gives nonpositive classical $\phi_m$ on $V$
with $E_m\subset\{\phi_m=-\infty\}$ and $\phi_m(z)\geq-1$.
The series
\[
 \phi=\sum_{m\geq1}2^{-m}\phi_m
\]
is a decreasing limit of nonpositive classical partial sums and
is therefore extended classical. It satisfies $\phi(z)\geq-1$,
so it is nontrivial. Every point of $E$ belongs to some $E_m$,
and the corresponding positive summand forces $\phi=-\infty$
there.
\end{proof}

\begin{remark}[Why the domain qualification is essential]
Untwisted pole globalization on every analytic space is
\texttt{FALSE}. For example, the rational point $0$ of
$\mathbf P^{1,\an}_K$ is locally pluripolar, with local witness
$\log|t|$. A nontrivial global extended classical psh function
on this proper connected space attains a finite maximum and is
constant by the strong maximum principle
\cite[Theorem 6.1]{GR25}. It therefore has no
pole. The preceding corollary is globalization on the one
specified exhaustion domain; it is not that false unrestricted
assertion, nor a theorem about globally twisted psh functions.
\end{remark}

\subsection{An outer-capacity formula for arbitrary relatively compact sets}

\begin{theorem}[The extremal total-mass formula]
\label{thm:r2-arbitrary-extremal-mass}
For every arbitrary $A\subset W\Subset V$,
\[
 C^*(A)=\int_V(DU_A)^n<\infty.
\]
This assertion neither assumes Borel measurability of $A$ nor
identifies the raw capacity with its outer regularization beforehand.
\end{theorem}

\begin{proof}
If $G\Subset V$ is ordinary open and contains $A$, then
$\mathcal F_G\subset\mathcal F_A$, hence $U_G\leq U_A\leq0$.
Both functions have boundary limit zero. The mass-order lemma and
\textup{(E)} imply
\[
 \int_V(DU_A)^n\leq\int_V(DU_G)^n=C(G).
\]
Taking the infimum over such $G$ gives
$\int_V(DU_A)^n\leq C^*(A)$. Relatively compact opens suffice here:
intersect any open cover with a fixed relatively compact open
neighborhood of $\overline W$. This also proves finiteness.

Since $h_A=-1$ on $A$, input \textup{(N)} shows that
\[
 A\cap\{U_A>-1\}\subset N
\]
for some set $N\subset W$ with $C^*(N)=0$. We may use a Borel
outer-zero cover of this defect if $A$ itself is not Borel.
Fix $0<\eta<1$ and put
\[
 G_\eta=\{U_A<-1+\eta\},\qquad
 v_\eta=\max\left(-1,\frac{U_A}{1-\eta}\right).
\]
The set $G_\eta$ is ordinary open and relatively compact, because
$U_A$ is usc and bounded below by the common exhaustion barrier.
The function $v_\eta$ is classical, has boundary value zero,
belongs to $[-1,0]$, and equals $-1$ on $G_\eta$. Consequently
\[
 \frac{U_A}{1-\eta}\leq v_\eta\leq U_{G_\eta}.
\]
Twice applying the mass-order lemma gives
\[
 C(G_\eta)=\int_V(DU_{G_\eta})^n
 \leq\int_V(Dv_\eta)^n
 \leq(1-\eta)^{-n}\int_V(DU_A)^n.
\]
Finally $A\subset G_\eta\cup N$. Monotonicity and subadditivity
of outer capacity, and its equality with raw capacity on ordinary
open sets, imply
\[
 C^*(A)\leq C(G_\eta)
 \leq(1-\eta)^{-n}\int_V(DU_A)^n.
\]
Let $\eta$ decrease to zero. The two inequalities prove the formula.
\end{proof}

\subsection{The missing continuity axiom and the Choquet conclusion}

\begin{theorem}[Continuity from below on arbitrary subsets of a core]
\label{thm:r2-outer-capacity-increasing-sets}
If $A_j\subset A_{j+1}\subset W\Subset V$ are arbitrary subsets
and $A=\bigcup_jA_j$, then
\[
 C^*(A)=\lim_j C^*(A_j).
\]
\end{theorem}

\begin{proof}
Write $U_j=U_{A_j}$; these bounded classical functions decrease
with $j$. By \textup{(N)}, each defect
$A_j\cap\{U_j>-1\}$ is contained in an outer-zero set $N_j$.
The countable union of the $N_j$ has outer capacity zero.
For $\epsilon>0$, choose one ordinary open set $G$ containing
that union with $C(G)<\epsilon$.

Fix $0<\eta<1$ and put
\[
 O_j=G\cup\{U_j<-1+\eta\}.
\]
These are \emph{increasing ordinary open sets}, since the $U_j$
decrease. Each $O_j$ contains $A_j$. The truncation estimate in
the preceding proof and its mass formula give
\[
 C(O_j)\leq\epsilon+
      (1-\eta)^{-n}C^*(A_j).
\]
The raw capacity is continuous from below on Borel sets, directly
from its definition as a supremum of Radon measures. Applying
this only to the increasing \emph{open} sets $O_j$ yields
\[
 C^*(A)\leq C\Bigl(\bigcup_jO_j\Bigr)
       =\lim_j C(O_j)
       \leq\epsilon+(1-\eta)^{-n}\lim_j C^*(A_j).
\]
Let $\epsilon$ and then $\eta$ decrease to zero. Monotonicity
supplies the reverse inequality.

This is the ordinary-open argument of the classical capacitability
proof, \cite[\S8, Theorem 8.2]{BT82}, with its analytic hypotheses
proved separately here. An infinite intersection of open covers is
not declared open, and no infimum/supremum interchange is made.
\end{proof}

\begin{corollary}[Local Choquet capacitability and Borel outer regularity]
\label{cor:r2-local-choquet}
Fix an ordinary core $W\Subset V$. The set function
\[
 A\longmapsto C^*(A),\qquad A\subset W,
\]
is a finite monotone countably subadditive Choquet capacity: it is
continuous from below on arbitrary increasing sets and from above
on decreasing compact sets. Every Suslin subset $A\subset W$
(analytic in the descriptive-set-theoretic sense) is capacitable:
\[
 C^*(A)=\sup_{\substack{K\subset A\\K\text{ compact}}}C(K).
\]
In particular every Borel $A\subset W$ satisfies
\[
 C(A)=C^*(A).
\]
\end{corollary}

\begin{proof}
Monotonicity and countable subadditivity are already proved for
outer capacity. Compact continuity from above is
Lemma~\ref{lem:r2-outer-facts}; increasing continuity is the preceding
theorem. Finiteness follows by covering $\overline W$ with one
relatively compact ordinary open and applying CLN to its closure.
The core $W$ is a locally compact second-countable metrizable space
in the fixed finite $p$-adic algebraic setting. The Choquet
capacitability theorem, in the form recalled in
\cite[\S8, following Theorem 8.2]{BT82}, therefore gives
\[
 C^*(A)=\sup_{K\subset A\text{ compact}}C^*(K)
\]
for such Suslin $A$. This terminology concerns analytic sets of the
Polish space $W$, not analytic subspaces over the base field $K$.
For compact $K$, the mass formula of the
preceding theorem and the compact part of \textup{(E)} give
$C^*(K)=C(K)$.

For Borel $A$, raw inner regularity is elementary:
\[
 C(A)=\sup_{v}\sup_{K\subset A\text{ compact}}
       \int_K(Dv)^n
     =\sup_{K\subset A\text{ compact}}C(K).
\]
Only two suprema are interchanged in this equality. Combining it
with capacitability gives $C(A)=C^*(A)$.
\end{proof}

\paragraph{Scope and dependency order.}
The capacity-zero witness uses only the open extremal identity,
ambient no-jump, the proved increasing calculus, and pointwise
ambient domination. The arbitrary-set mass formula and Choquet
continuity do not use that witness theorem: their only null-set
input is \textup{(N)}. Thus the argument does not prove negligible
sets polar by first assuming that conclusion. The capacitability
statement is on each fixed relatively compact core of this explicitly
qualified domain. It does not assert a global Josefson theorem on an
arbitrary analytic space, or discard the negative-exhaustion-domain
hypothesis.

\paragraph{A distinction in abstract capacity theory.}
For a Borel capacity $c$, its Borel-envelope extension
$\inf\{c(B): A\subset B,\ B\text{ Borel}\}$ is different from
the ordinary-open extension used here. The familiar increasing-set
argument using $\bigcap_{k\geq j}G_k$ yields Borel sets, not open
ones. Raw Borel inner regularity and raw Borel continuity from below
therefore do not, by that argument, supply the missing continuity
axiom for the ordinary-open extension. Theorem
\ref{thm:r2-outer-capacity-increasing-sets} supplies the required
axiom by actual ordinary sublevel sets of extremal functions.

% END source: base_r3/base_r2/sections/polar_converse_choquet.tex


% BEGIN frozen or reviewed source: base_r3/base_r2/sections/round2_status.tex
\section{Second-round theorem and status increment}
\label{sec:r2-status}

This section updates, rather than replaces, the source-by-source comparison
in Section~\ref{sec:comparison}. The first-round statements remain frozen
with their original hypotheses. An earlier \texttt{OPEN} entry is closed
only by the particular second-round theorem cited below. No claim of
literature priority is encoded by \texttt{NEW\_PROVED}: that label records
a proof supplied in this project, relative to its accepted inputs.

\subsection{Two independent status axes and the fixed category}

The labels \texttt{EXTERNAL\_KNOWN}, \texttt{INTERNAL\_INPUT},
\texttt{DIRECT\_COROLLARY}, and \texttt{NEW\_PROVED} record provenance.
The labels \texttt{PROVED}, \texttt{CONDITIONAL}, \texttt{OPEN}, and
\texttt{FALSE} assess a precisely qualified proposed analogy. A theorem
proved under an explicit additional geometric hypothesis is
\texttt{PROVED} with that hypothesis; the corresponding general-domain
analogy is \texttt{CONDITIONAL} or \texttt{OPEN}. A counterexample to an
unrestricted assertion does not change the status of its qualified
replacement.

All local assertions retain the finite extension $K/\mathbf Q_p$, its
prescribed absolute value, and the allowed good, Hausdorff, boundaryless,
quasi-smooth pure-dimensional analytic spaces. Unless a narrower theorem
states otherwise, $S$ is an arbitrary \emph{fixed} symmetric current of
bidegree $(p,p)$, positive in the original CLD cone and separately
$d'$- and $d''$-closed. Products and current equalities are tested on
\emph{all original compactly supported smooth complementary CLD forms}.
Neither a change of positivity cone nor an enlargement of the test sheaf
is implicit. Put $q=n-p$; the relative top-capacity statements below use
$q\geq1$.

R and B remain \texttt{INTERNAL\_INPUT}. In particular, the bounded
classical class and its $[1]$-background calculus are not second-round
discoveries. The completed geometric adelic nefness theorem A2 also
remains an \texttt{INTERNAL\_INPUT} under its own number-field,
integrability, and boundary hypotheses. The original classical, CLD,
GR, and GK theorems recorded in Section~\ref{sec:comparison} remain
\texttt{EXTERNAL\_KNOWN} in their stated categories.

\subsection{Increment for the eleven Bedford--Taylor directions}

\begin{longtable}{@{}>{\raggedright\arraybackslash}p{.045\textwidth}>{\raggedright\arraybackslash}p{.19\textwidth}>{\raggedright\arraybackslash}p{.40\textwidth}>{\raggedright\arraybackslash}p{.28\textwidth}@{}}
\hline
 & Target & Qualified second-round increment & Remaining qualification or obstruction\\
\hline
\endfirsthead
\hline
 & Target & Qualified second-round increment & Remaining qualification or obstruction\\
\hline
\endhead
1 & Locally bounded psh functions &
\texttt{PROVED / DIRECT\_COROLLARY}: upper regularizations of locally
uniformly upper-bounded classical families are classical, by
Theorem~\ref{thm:r2-local-upper-envelope}; the projective bounded usc
obstacle envelope is Proposition~\ref{prop:r2-usc-obstacle}.
\texttt{PROVED / DIRECT\_COROLLARY}: original trace-current order determines
the actual bounded classical representative,
Proposition~\ref{prop:r2-current-faithfulness}. &
The new capacitary class is relative to $S$ and retains Borel
representatives. For $S=[1]$, every such class now has a unique
actual classical representative modulo local outer-capacity zero,
Corollary~\ref{cor:r2-ambient-classical-representative}
(\texttt{DIRECT\_COROLLARY}). Actual Borel values on the exceptional
set are not identified. No corresponding assertion for arbitrary
$S$ is made.\\[5pt]

2 & Positive closed currents and mixed products &
The arbitrary-fixed-$S$ bounded classical calculus is already
\texttt{PROVED} in R1. Theorem~\ref{thm:r2-completed-calculus} extends
it to $\mathcal A_{S,b}$, with well-defined Borel recursion,
quasi-everywhere independence, symmetry, and restriction/gluing:
\texttt{PROVED / NEW\_PROVED}. &
This does not define the wedge of two arbitrary positive closed
currents. Nor does it settle dependence only on the currents $D[u_i]$:
the general discontinuous potential-quotient assertion remains
\texttt{OPEN}.\\[5pt]

3 & Chern--Levine--Nirenberg estimates &
\texttt{PROVED / NEW\_PROVED}: the weighted $L^1$ estimate and uniform
truncation estimate of Theorem~\ref{thm:r2-weighted-l1}, under the
specified collar-trace integrability condition. The deep-sublevel
outer-capacity estimate and integrable-witness consequences are
\texttt{DIRECT\_COROLLARY},
Corollaries~\ref{cor:r2-normalized-witness} and
\ref{cor:r2-witness-outer-capacity}. &
The integrability hypothesis concerns an explicit trace of the fixed
$S$; it is not automatic for arbitrary $S$. Uniform exponential tails
and unbounded-weight stability from a uniform first moment alone are
\texttt{FALSE}, Propositions~\ref{prop:r2-no-exponential-tails} and
\ref{prop:r2-unbounded-weight-failure}.\\[5pt]

4 & Symmetry and weighted monotone convergence &
R1 already proves symmetry and weighted joint decreasing continuity
for actual bounded classical limits and fixed $S$.
\texttt{PROVED / NEW\_PROVED}: bounded increasing sequences converge
in outer relative capacity to their \emph{actual} Borel supremum,
Theorem~\ref{thm:r2-increasing-cauchy} and
Corollary~\ref{cor:r2-local-increasing}; weighted increasing continuity
in the completed class is Corollary~\ref{cor:r2-increasing-weighted}.
The ambient no-jump theorem
\ref{thm:r2-increasing-no-jump} identifies this supremum with its usc
regularization locally quasi-everywhere:
\texttt{PROVED / NEW\_PROVED}.
Theorem~\ref{thm:r2-weak-closure-multipliers} proves the qualified
quasicontinuous multiplier passage through weak closure. &
Ambient no-jump and its common-collar energy equality are closed
questions in this record. Replacing the actual supremum by its
ambient usc regularization remains \texttt{FALSE} without
qualification for arbitrary $S$. Unrestricted varying-$S$ weighted
stability remains \texttt{FALSE}.\\[5pt]

5 & Locality and maximum principles &
R1 Borel maximum locality and ordinary restriction/locality remain
\texttt{PROVED}. Theorem~\ref{thm:r2-polar-restriction} proves that
$\mathbf1_P S$ and $\mathbf1_{P^c}S$ are separately closed for a
classical pole set $P$, and that BT products commute with this
restriction: \texttt{PROVED / NEW\_PROVED}.
Theorem~\ref{thm:r2-maximal-ma-zero} proves local classical maximality
equivalent to $(Du)^n=0$ for the ambient background. &
Strict inequality sets of discontinuous potentials are Borel and
locally quasi-open, not automatically ordinary open. The polar
restriction theorem starts with a current defined on the entire
domain; it is not a general extension theorem for a current defined
only off the pole set.\\[5pt]

6 & Bedford--Taylor capacities &
\texttt{PROVED / DIRECT\_COROLLARY}: outer regularity on quasi-open
sets, Proposition~\ref{prop:r2-quasiopen-outer}, and the elementary
outer-capacity properties in Lemma~\ref{lem:r2-outer-facts}.
\texttt{PROVED / NEW\_PROVED}: two-sided local comparison with one
fixed relatively compact affine capacity domain,
Theorem~\ref{thm:r2-domain-capacity-comparison}. For the ambient
capacity on an exhaustion domain, compact raw outer regularity and
compact continuity are Theorem~\ref{thm:r2-compact-capacity-continuity};
outer-capacity continuity for arbitrary increasing subsets of a
fixed relatively compact core is
Theorem~\ref{thm:r2-outer-capacity-increasing-sets}.
Choquet capacitability and Borel outer regularity on that core are
Corollary~\ref{cor:r2-local-choquet}. &
Raw compact outer regularity and the corresponding Choquet theory
for \emph{arbitrary relative backgrounds $S$} remain \texttt{OPEN}.
The ambient result retains the exhaustion-domain and fixed-core
scope. General-domain nondegeneracy remains \texttt{FALSE}; the R1
proper-space example still applies.\\[5pt]

7 & Quasicontinuity and capacity convergence &
R1 bounded quasicontinuity is retained. Theorems
\ref{thm:r2-increasing-cauchy} and \ref{thm:r2-completed-calculus}
give the increasing and capacitary-completion extensions.
Corollary~\ref{cor:r2-extended-psh-qc} gives local finite
quasicontinuity and lower-truncation outer-capacity convergence for
every nontrivial extended classical psh function:
\texttt{PROVED / DIRECT\_COROLLARY} of the new polar-trace and
weighted estimates. &
Capacity precompactness of the full normalized continuous psh family,
and weighted BT stability under bounded pointwise convergence alone,
are \texttt{FALSE},
Proposition~\ref{prop:r2-capacity-noncompactness}. This counterexample
does not refute compact-set domination by the weak closure of the
admissible measure family.\\[5pt]

8 & Pluripolar and negligible sets &
\texttt{PROVED / NEW\_PROVED}: every original smooth ambient trace
vanishes on a nontrivial classical pole set,
Theorem~\ref{thm:r2-ambient-polar-traces}.
\texttt{DIRECT\_COROLLARY}: all ambient bounded BT coefficients avoid
such sets, and every locally pluripolar subset has local outer
capacity zero, Corollaries~\ref{cor:r2-ambient-bt-poles} and
\ref{cor:r2-ambient-polar-outer-zero}.
Theorem~\ref{thm:r2-capacity-zero-polar-witness} constructs a
nontrivial witness for an outer-zero set. The resulting local
equivalence of pluripolar, outer-zero, and negligible sets is
Corollary~\ref{cor:r2-polar-negligible-equivalence}.
Pole patching on one exhaustion domain is
Corollary~\ref{cor:r2-exhaustion-domain-pole-patching}. &
These local ambient equivalences are \texttt{PROVED}.
Unqualified untwisted pole globalization on every analytic space is
\texttt{FALSE}, already on $\mathbf P^{1,\an}_K$.
Arbitrary relative backgrounds may charge ambient pluripolar sets;
their mass is governed by
Theorem~\ref{thm:r2-polar-restriction}, not forced to vanish.\\[5pt]

9 & Comparison and domination &
The R1 compact-gap comparison theorem for arbitrary fixed $S$
remains \texttt{PROVED}.
\texttt{PROVED / DIRECT\_COROLLARY}: ambient strict-gap detection and
pointwise domination/uniqueness on the stated relatively compact
framed affine domains, Theorem~\ref{thm:r2-ambient-gap-detection}
and Corollary~\ref{cor:r2-pointwise-ambient}.
Relative atomic and smooth-factor detection criteria and explicit
faithful backgrounds are proved in
Propositions~\ref{prop:r2-relative-gap-detection} and
\ref{prop:r2-smooth-factor-condition}. &
Pointwise domination for arbitrary $S$ remains
\texttt{CONDITIONAL} on detection; the unrestricted version is
\texttt{FALSE} even for nonzero $S$ on a connected domain,
Example~\ref{ex:r2-nonzero-domination-failure}.
Capacity-modulo uniqueness and actual pointwise uniqueness are
distinct assertions.\\[5pt]

10 & Relative extremal functions &
\texttt{PROVED / NEW\_PROVED}: least upper bounds in the local
$S$-quasi-everywhere order, represented in $\mathcal A_{S,b}$,
Theorem~\ref{thm:r2-local-capacitary-envelope}.
Ambient bounded usc obstacle envelopes follow directly from B;
orthogonality for bounded quasicontinuous usc projective obstacles
is Theorem~\ref{thm:r2-qc-orthogonality},
\texttt{PROVED / NEW\_PROVED}.
For the ambient background on an exhaustion domain, compact/open
extremal equilibrium formulas are Theorems
\ref{thm:r2-compact-extremal} and \ref{thm:r2-open-extremal};
arbitrary relatively compact sets satisfy the outer-capacity
total-mass formula \ref{thm:r2-arbitrary-extremal-mass}. &
The ambient and relative categories remain distinct. General
\emph{relative} equilibrium-measure/contact-set formulas and their
capacity identities remain \texttt{OPEN}; unrestricted replacement
of an $S$-capacitary envelope by the ambient envelope is \texttt{FALSE},
Proposition~\ref{prop:r2-negligible-positive}.\\[5pt]

11 & Monge--Amp\`ere and Dirichlet theory &
\texttt{PROVED / NEW\_PROVED}: the ambient local equivalence between
maximality and the homogeneous equation,
Theorem~\ref{thm:r2-maximal-ma-zero}.
The explicit peak-barrier theorem
\ref{thm:r2-peak-dirichlet} gives a bounded classical solution of
$(DH_g)^n=0$, ordinary boundary values, uniqueness, and the boundary
sup-norm stability estimate under
Definition~\ref{def:r2-peak-barriers}.
Corollary~\ref{cor:r2-dirac-dirichlet} gives an explicit continuous
solution and uniqueness for a single positive divisorial atom and
constant boundary data on the domains constructed with its barrier:
\texttt{PROVED / DIRECT\_COROLLARY}. &
As a general-domain analogy the Dirichlet conclusion is
\texttt{CONDITIONAL} on those explicit barriers. Their general
geometric construction, interior continuity for the Perron solution,
and existence for a general prescribed nonzero measure remain
\texttt{OPEN}. Bounded negative exhaustion domains are explicitly
defined and locally supplied; they are not silently assumed to have
peak barriers at every boundary point.\\[3pt]
\hline
\end{longtable}

\subsection{Scope distinctions that survive the second round}

\paragraph{Fixed backgrounds and controlled varying backgrounds.}
The primary calculus keeps $S$ fixed. The order-dominated replacement
in Proposition~\ref{prop:dominated-background-stability} allows
$0\leq S_j\leq C R_0$, $S_j\to S$, with one fixed positive closed
$R_0$ and a fixed differentiated tuple; its multiplier hypotheses
must also be retained. The second-round weak-closure multiplier
theorem instead requires uniform coefficient bounds by a fixed local
capacity and quasicontinuous weights. Neither statement removes the
counterexample in Corollary~\ref{cor:varying-background-false}.

\paragraph{Actual representatives and ambient upper regularization.}
An actual increasing supremum is now an element of the bounded
capacitary closure, and its BT products are established there. For the
ambient background, Theorem~\ref{thm:r2-increasing-no-jump} identifies
it with its ordinary usc regularization locally outside outer-zero
sets and proves the common-collar energy no-jump equality. More
generally, Corollary~\ref{cor:r2-ambient-classical-representative}
assigns each $a\in\mathcal A_{[1],b}$ a unique classical representative
$r_a$ modulo such sets. For an arbitrary Borel representative $a$,
$r_a$ need not equal $a^*$; arbitrary modifications on polar sets are
allowed in that capacitary class. This distinction does not affect
the actual increasing-supremum theorem.
Theorem~\ref{thm:r2-local-capacitary-envelope} obtains a least upper
bound without taking an uncountable union of exceptional sets: each
competitor has its own quasi-everywhere comparison, and a countable
energy construction determines the resulting class. For an arbitrary
relative background $S$, the capacitary representative cannot in
general be replaced by the ambient usc envelope.

\paragraph{Raw capacity and outer capacity.}
For arbitrary $S$, the identity $C^*(E)=C(E)$ is proved for quasi-open
$E$; full raw compact outer regularity remains open. For $S=[1]$ on
an exhaustion domain, it is now proved for compact sets and for
every Borel subset of each fixed relatively compact core. The
ordinary-open proof of Theorem~\ref{thm:r2-outer-capacity-increasing-sets}
supplies the additional continuity axiom needed for the local
Choquet theorem; compact continuity of $C^*$ alone would not supply
that proof. The raw-capacity closure criterion in
Proposition~\ref{prop:capacity-weak-closure} is therefore resolved
for the stated ambient compact stages but remains an exact target
for arbitrary relative $S$. The explicit potential-noncompactness
example has an admissible unweighted weak limit and does not refute
compact-set domination by the admissible measure family's weak closure.

\paragraph{Polar sets and arbitrary support.}
The general ambient pole-trace problem is now \texttt{PROVED}; it is
no longer a missing hypothesis in the ambient nonconcentration result.
The proof uses the absence of divisorial poles, whole-box model
approximation, and the capacity-controlled multiplier passage, as
recorded in Theorem~\ref{thm:r2-ambient-polar-traces}. For arbitrary
$S$, the exact statement is
$\mathbf1_P T_{S,k}=T_{\mathbf1_PS,k}$ from
Theorem~\ref{thm:r2-polar-restriction}. In particular, a background
already carried by $P$ is not made nonpolar by applying bounded BT
operators. Ambient outer-zero sets now admit local nontrivial pole
witnesses, and the local pluripolar/negligible/outer-zero classes agree
by Corollary~\ref{cor:r2-polar-negligible-equivalence}. There is also
pole patching on a single connected exhaustion domain. A global
untwisted version for all analytic spaces is false on proper connected
spaces; these local and qualified global statements are not conflated.

\paragraph{Potential quotients.}
Proposition~\ref{prop:r2-current-faithfulness} concerns $[u]$ and
$[v]$, not merely $D[u]$ and $D[v]$. General independence under
replacing a differentiated potential by another actual bounded
classical potential with the same $D$-current remains \texttt{OPEN}.
The continuous harmonic-difference result
\ref{lem:continuous-harmonic-multipliers}, the ambient-mixed and smooth
background cases \ref{lem:potential-independence-ambient-mixed} and
\ref{lem:potential-independence-smooth}, and the dominated-closure
case \ref{prop:potential-independence-dominated-closure} remain proved
partial answers. None is silently promoted to the general assertion.

\paragraph{Comparison with other differential calculi.}
Agreement on ordinary CLD tests with compatible model and
Chambert--Loir measures is retained where the accepted R/B comparison
applies. Equality on every $\delta$-test form is a different statement.
The report does not extend arbitrary original currents to an enlarged
CLD25 test sheaf, or claim the nonpositive, nonclosed, or nondiagonal
parts of the GK calculus. The completed A2 adelic theorem is an input
to subsequent geometric applications, not an unresolved item in this
local analytic program.

\subsection{Three minimal targets for the next round}

The former ambient no-jump and local polar-witness targets are now
proved. The following remaining targets are not assumed anywhere in
the second-round arguments.

\begin{enumerate}
\item \textbf{Compact domination for the raw capacity's measure
closure for arbitrary relative backgrounds.} For an arbitrary allowed
fixed $S$, on one fixed permitted capacity domain and compact test
stage, prove or refute
\[
 \nu(K)\leq\operatorname{Cap}_S(K;\Omega)
 \quad\text{for every compact }K
\]
for every weak limit $\nu$ of normalized top BT measures (with the
localization conventions of
Proposition~\ref{prop:capacity-weak-closure}). It is enough to resolve
this precise closure statement; a capacity-convergent subsequence of
all normalized potentials is not available. A counterexample must
violate the displayed inequality, rather than only potential
precompactness or weighted pointwise continuity. The ambient result
on exhaustion domains is already closed and is not assigned again
as new work.

\item \textbf{The general potential-quotient lemma.} Given actual
$u,v\in\Cb(U)$ with $D[u]=D[v]$ as original currents, establish
or refute
\[
 D\bigl((u-v)S\bigr)=0
\]
for every allowed fixed positive closed background $S$.
The continuous-difference and specified-background cases are already
proved. Equality of the original currents $[u]$ and $[v]$, and the
unique ambient capacitary representative, do not by themselves
settle this statement about equality of their $D$-currents.

\item \textbf{Nonzero data on one prescribed domain.} Starting with
an explicitly qualified domain and boundary barriers, construct and
control solutions for a sum of at least two prescribed positive
divisorial atoms on that \emph{same} domain. The single-atom result
\ref{cor:r2-dirac-dirichlet} constructs its domain together with the
atom and barrier. It does not provide this next fixed-domain step;
adding its potentials produces mixed terms. A passage from finite
atomic data to more general measures would then require uniform
potential bounds, boundary control, and a proved capacity stability
estimate for that family of solutions.
\end{enumerate}

The homogeneous peak-barrier theorem yields bounded classical
solutions with ordinary boundary limits; interior continuity has not
been proved there. General prescribed nonzero measures and general
ordinary domains still require additional geometric and analytic
hypotheses. The relative equilibrium/contact problem and comparison
on enlarged $\delta$-test categories remain separate subsequent tasks.

\subsection{Excluded routes and the conclusions they actually exclude}

\begin{enumerate}
\item \texttt{FALSE}: arbitrary varying-background weighted stability,
even at order zero; see
Corollary~\ref{cor:varying-background-false}. Fixed-background
monotonicity and the qualified order-dominated replacement survive.

\item \texttt{FALSE}: relative nonconcentration on every ambient
pluripolar set, or universal relative negligibility of the discrepancy
between an actual supremum and its ambient usc envelope; see
Proposition~\ref{prop:polar-relative-mass} and
Proposition~\ref{prop:r2-negligible-positive}. The valid replacement
is support restriction and the $S$-capacitary envelope.

\item \texttt{FALSE}: pointwise domination solely from relative
measure inequalities for an arbitrary nonzero $S$ on a connected
domain; see Example~\ref{ex:r2-nonzero-domination-failure}.
Relative gap detection is a real additional condition.

\item \texttt{FALSE}: nondegeneracy of the untwisted capacity on every
analytic domain, including proper ones; see
Proposition~\ref{prop:compact-capacity-zero}. Local framed domains
and projective metric capacities must be stated separately.

\item \texttt{FALSE}: compactness in capacity of the whole normalized
continuous psh family, or weighted BT continuity under bounded
pointwise convergence; see
Proposition~\ref{prop:r2-capacity-noncompactness}.
\texttt{OPEN}: the compact raw-capacity weak-closure inequality for
arbitrary relative $S$. The latter is not contradicted by this example;
the qualified ambient compact inequality is now proved.

\item \texttt{FALSE}: a uniform first-moment estimate implies uniform
integrability of a fixed unbounded multiplier, or universal exponential
tail control for arbitrary $S$; see
Propositions~\ref{prop:r2-unbounded-weight-failure} and
\ref{prop:r2-no-exponential-tails}. The proved weighted estimate is
used with its actual trace-integrability and truncation conclusions.

\item \texttt{FAILED\_ATTEMPT}: deriving raw outer regularity by
interchanging an infimum over ordinary neighborhoods with a supremum
over a measure family. Individual Radon outer regularity does not
justify that interchange. For arbitrary $S$, the quasi-open argument
and weak-closure characterization remain the available replacements.
For the ambient exhaustion-domain capacity, the extremal mass formula
and increasing ordinary sublevels supply the additional proof.

\item \texttt{FALSE}: untwisted pole globalization on every analytic
space. The rational point of $\mathbf P^{1,\an}_K$ in
Section~\ref{sec:r2-polar-converse-choquet} is locally pluripolar,
while nontrivial global classical psh functions there are constant.
Corollary~\ref{cor:r2-exhaustion-domain-pole-patching} proves the
qualified replacement on one connected exhaustion domain.

\item \texttt{FAILED\_ATTEMPT}: inferring smooth-trace nullity merely
from density of divisorial points, or passing an indicator through
weak convergence without coefficient control. The successful polar
proof additionally uses quasicontinuity of the actual indicator and
uniform same-domain capacity estimates.

\item \texttt{FAILED\_ATTEMPT}: swapping two discontinuous moving
limits by formal integration by parts, using the desired joint
continuity to prove its own energy estimate, or treating an
uncountable union of quasi-everywhere exceptional sets as negligible.
The established R1 trace argument and R2 countable energy-envelope
construction avoid these inferences.
\end{enumerate}

The second-round record therefore closes several analytic gaps while
retaining explicit frontier statements. It does not label the full
non-Archimedean Bedford--Taylor program complete.

% END source: base_r3/base_r2/sections/round2_status.tex


% BEGIN frozen or reviewed source: base_r3/sections/energy_convex_blocks.tex
\section{Mixed energy controls capacity in every remaining degree}
\label{sec:r3-all-degree-energy}

The fixed background in this section is arbitrary: $S$ is symmetric,
original-positive, of bidegree $(p,p)$, and separately closed, and
$q=n-p\geq1$. The original field, ordinary topology, and all original
CLD tests are retained. The estimates resolve the varying-tester
energy obstruction mentioned in
Section~\ref{sec:r3-relative-compact-capacity}. They do not identify
nonlinear weak limits of the original unaveraged measures.

Expressions such as $(Da)^i(Db)^jS$ abbreviate the already constructed
mixed products over the fixed $S$. An energy expression always means
$\Gamma_Q(f,g)$ for the specified mixed background $Q$ of bidegree
$(n-1,n-1)$. It never means a wedge of an arbitrary energy current
with another current.

\subsection{A common-collar energy distance}

Let $\Omega$ be a sufficiently small ordinary norm-box domain and
$L\Subset\Omega$ a compact set in a strictly smaller norm box. Fix
a smooth classical psh function $h$ with semipositive original
curvature form. Consider $u,v\in\Cb(\Omega)$ with
\[
 u=v=h\quad\hbox{on }\Omega\setminus L.
\]
There is one fixed bound for $u,v,h$ near $L$. Put $f=u-v$ and
\begin{equation}
 \label{eq:r3-I-energy}
 I_S(u,v)=\int f\bigl((Dv)^qS-(Du)^qS\bigr).
\end{equation}
This integral is finite and has compact support.

\begin{lemma}[The positive energy identity]
\label{lem:r3-I-positive}
For the actual point values of the potentials one has
\[
 I_S(u,v)=\sum_{a=0}^{q-1}
    \int\Gamma_{(Du)^a(Dv)^{q-1-a}S}(f)\geq0.
\]
\end{lemma}

\begin{proof}
Symmetry and multilinearity of the proved mixed calculus give
\[
 (Du)^qS-(Dv)^qS
   =D\left(f\sum_{a=0}^{q-1}(Du)^a(Dv)^{q-1-a}S\right).
\]
Apply $-\int fD(fQ)=\int\Gamma_Q(f)$ to each positive separately
closed summand. No weakly varying arbitrary background is used.
\end{proof}

\begin{theorem}[An all-degree mixed-energy capacity estimate]
\label{thm:r3-I-capacity}
Under these common-collar assumptions, for every $\delta>0$,
\begin{equation}
 \label{eq:r3-all-degree-I-capacity}
 \operatorname{Cap}_S^*(\{|u-v|>\delta\};\Omega)
       \leq C\delta^{-2} I_S(u,v)^{\,2^{-q}}.
\end{equation}
The finite constant depends on the fixed compact collar, the common
bound for the potentials, $q$, and finitely many positive coefficient
evaluations of $S$. It is independent of the normalized capacity
tester and all regularization indices. Restriction to a smaller
compact core gives the corresponding cutoff version.
\end{theorem}

\begin{proof}
Fix $w\in\Cb(\Omega)$ with $0\leq w\leq1$. For $0\leq r\leq q-1$ put
\[
 E_r(w)=\int\Gamma_{(Dw)^r(D(u+v))^{q-1-r}S}(f).
\]
These are nonnegative integrals on the same compact support, since
$f=0$ off $L$. Expansion and the positive energy identity give
\begin{equation}
 \label{eq:r3-E-zero}
 E_0(w)\leq2^{q-1}I_S(u,v).
\end{equation}
For $r\geq1$ write
$Q=(Dw)^{r-1}(D(u+v))^{q-1-r}S$, of bidegree $(n-2,n-2)$.
First suppose that the potentials are smooth. Stokes for the
compactly supported first-order expression gives
\begin{equation}
 \label{eq:r3-E-transfer}
 E_r(w)=\int d'f\wedge d''w\wedge(Du-Dv)\wedge Q.
\end{equation}
For the sign, apply $d'$ to
$d'f\wedge d''f\wedge d''w\wedge Q$; the result is
$-d'f\wedge Df\wedge d''w\wedge Q+
d'f\wedge d''f\wedge Dw\wedge Q$.

Choose a fixed nonnegative compact smooth cutoff $\kappa$ equal
to one near $L$. Symmetry and Cauchy--Schwarz, separately for
the positive backgrounds $Du\,Q$ and $Dv\,Q$, imply
\begin{align*}
 E_r(w)
 &\leq\left(\int\Gamma_{Du\,Q}(f)\right)^{1/2}
       \left(\int\kappa\,\Gamma_{Du\,Q}(w)\right)^{1/2}\\
 &\quad+\left(\int\Gamma_{Dv\,Q}(f)\right)^{1/2}
       \left(\int\kappa\,\Gamma_{Dv\,Q}(w)\right)^{1/2}.
\end{align*}
The first factors squared sum to $E_{r-1}(w)$. The second factors
squared have a uniform finite sum: for either $R=Du\,Q$ or
$R=Dv\,Q$ the cutoff identity is
\[
 \int\kappa\,\Gamma_R(w)
  =\tfrac12\langle R,w^2D\kappa\rangle-\int\kappa wD(wR).
\]
The subtracted term is nonnegative. The relative finite-test CLN
estimate bounds the first term for the fixed form $D\kappa$.
All slots of $R$ are bounded copies of $u,v,w$ or their sum.
Thus, for a fixed $A_r$ independent of $w$,
\begin{equation}
 \label{eq:r3-E-recurrence}
 E_r(w)\leq A_r E_{r-1}(w)^{1/2}.
\end{equation}
For $q=1$ this recurrence is empty.

The compact Green identity for $f^2$ now gives
\[
 \int f^2(Dw)^qS
       =-2\int f\,\Gamma_{(Dw)^{q-1}S}(f,w).
\]
Cauchy--Schwarz with the same compact cutoff and another uniform
CLN bound yield
\begin{equation}
 \label{eq:r3-E-square-final}
 \int f^2(Dw)^qS\leq A E_{q-1}(w)^{1/2}.
\end{equation}
The common bound for $f$ is absorbed in $A$. Iteration proves
\begin{equation}
 \label{eq:r3-all-degree-square-control}
 \int f^2(Dw)^qS\leq C I_S(u,v)^{\,2^{-q}}.
\end{equation}

For bounded potentials, use the protected prescribed-box
construction for simultaneous smooth decreasing rows. If necessary,
first modify $u,v$ outside a neighborhood of $L$ to a common
auxiliary psh collar, preserving them up to the same constant
near $L$ and preserving their difference $f$ everywhere. Their
rows then have an exact common smooth collar outside a fixed
larger compact set $L'\Subset\Omega$. This does not assert that
arbitrary smoothing preserves the originally prescribed $h$.
All limiting weighted and energy expressions in the argument
are supported in $L$, so the auxiliary modification leaves
them unchanged. Choose all cutoffs for the larger fixed stage.
Take a simultaneous smooth decreasing row for $w$ there.
After discarding a prefix the latter lies in $[0,2]$,
so all constants remain uniform. Every background in
the preceding formulas is an explicit mixed product over the
same $S$. Expand each energy as
\[
 \Gamma_R(a,b)=\tfrac12\{D(abR)-aD(bR)-bD(aR)\}.
\]
Its scalar factors are polynomials in finitely many decreasing
psh rows. The functional-weighted joint monotone theorem passes
every term to the asserted bounded-potential expression on a
fixed cutoff. The differences retain fixed compact support.
This proves the identities and inequalities without using
capacity continuity to justify the energy calculation.

Chebyshev and the supremum over $w$ give the asserted raw
capacity bound. The strict gap set is relatively quasi-open
by quasicontinuity of $u,v$. Proposition~\ref{prop:r2-quasiopen-outer}
therefore identifies its raw and outer capacities. No such
set was assumed ordinarily open.
\end{proof}

\subsection{Concavity selects simultaneous convex blocks}

For the fixed collar define
\[
 \mathcal E_{S,h}(u)
   =\frac1{q+1}\sum_{j=0}^q
       \int(u-h)(Du)^j(Dh)^{q-j}S.
\]
All nonzero weighted integrands lie on the fixed compact core.
No nondegeneracy assumption on $Dh$ is imposed.

\begin{lemma}[Quantitative midpoint concavity]
\label{lem:r3-energy-midpoint}
The energy is concave and
\begin{equation}
 \label{eq:r3-midpoint-I-bound}
 \mathcal E_{S,h}\!\left(\frac{u+v}{2}\right)
  -\frac{\mathcal E_{S,h}(u)+\mathcal E_{S,h}(v)}2
      \geq c_q I_S(u,v),\qquad c_q=\frac q{4^{q+1}}.
\end{equation}
For $u_t=(1-t)u+tv$,
\[
 \frac{d}{dt}\mathcal E_{S,h}(u_t)=\int(v-u)(Du_t)^qS,
 \qquad
 \frac{d^2}{dt^2}\mathcal E_{S,h}(u_t)
       =-q\int\Gamma_{(Du_t)^{q-1}S}(v-u).
\]
\end{lemma}

\begin{proof}
The energy is a polynomial in $t$. In its derivative, compact
Green exchange replaces
\[
 \int(u_t-h)D(v-u)(Du_t)^{j-1}(Dh)^{q-j}S
\]
by
$\int(v-u)(Du_t-Dh)(Du_t)^{j-1}(Dh)^{q-j}S$.
The finite sum telescopes to the first derivative. Differentiating
again and using compact Green gives the second derivative and
concavity. These are identities in the established bounded
mixed calculus; smooth common-collar rows give the same proof.

The one-variable midpoint Green formula expresses the gap as
\[
 \frac q2\int_0^1\min(t,1-t)
       \left(\int\Gamma_{(Du_t)^{q-1}S}(u-v)\right)dt.
\]
On $[1/4,3/4]$, expansion into positive mixed energy measures gives
\[
 \int\Gamma_{(Du_t)^{q-1}S}(u-v)
 \geq4^{-(q-1)}\int\Gamma_{(D(u+v))^{q-1}S}(u-v)
 \geq4^{-(q-1)}I_S(u,v).
\]
The weight is at least $1/4$ on this interval of length $1/2$.
This yields the displayed constant.
\end{proof}

\begin{corollary}[An energy criterion for capacity convergence]
\label{cor:r3-energy-capacity-topology}
For common-collar $u,v$ put
\[
 J_{S,h}(u,v)=\mathcal E_{S,h}(u)-\mathcal E_{S,h}(v)
                 +\int(v-u)(Du)^qS.
\]
Then
\[
 \frac{I_S(u,v)}{q+1}
       \leq J_{S,h}(u,v)
       \leq\frac{qI_S(u,v)}{q+1}.
\]
For a uniformly bounded common-collar sequence $u_j$ and $u$,
the following conditions are equivalent:
\begin{enumerate}
\item $I_S(u_j,u)\to0$;
\item $u_j\to u$ locally in outer relative capacity on all
      permitted compact/domain pairs;
\item $\mathcal E_{S,h}(u_j)\to\mathcal E_{S,h}(u)$ and
      $\int(u_j-u)(Du)^qS\to0$.
\end{enumerate}
In particular, for an actual pointwise convergent sequence with
these bounds and collar, convergence of its energies suffices
for capacity convergence.
\end{corollary}

\begin{proof}
Taylor's formula and the preceding second derivative give
\[
 J_{S,h}(u,v)=
 \sum_{a=0}^{q-1}\frac{q-a}{q+1}
       \int\Gamma_{(Dv)^a(Du)^{q-1-a}S}(v-u).
\]
Indeed the coefficient is
$q\binom{q-1}{a}\int_0^1t^a(1-t)^{q-a}dt=(q-a)/(q+1)$.
Each measure is positive, proving the two bounds.
Condition (3) therefore implies (1). The all-degree
estimate gives convergence on the fixed capacity domain
$\Omega$ from (1), and
Corollary~\ref{cor:r2-fixed-domain-local-convergence} gives
the all-domain local assertion (2).

Conversely, the completed weighted capacity-continuity
theorem applies to every finite term in the energy, with
a compact cutoff equal to one near $L$. It gives both
limits in (3), and also the defining limit for $I$.
The common support and local bounds justify all these
compact pairings. Under actual pointwise convergence,
the integral against the fixed measure $(Du)^qS$ tends
to zero by bounded convergence on $L$. This proves the
last assertion. This corollary is
\texttt{DIRECT\_COROLLARY} of the preceding estimates
and the established completed calculus.
\end{proof}

\begin{theorem}[Convex-block compactness in every positive remaining degree]
\label{thm:r3-all-degree-convex-blocks}
Let $\Omega$ be an ordinary open in one allowed smooth affine
algebraic neighborhood, and let $S$ be the fixed background above.
Every sequence $u_j\in\Cb(\Omega)$ uniformly bounded on each
compact neighborhood admits finite tail convex combinations
\[
 v_N\in\operatorname{conv}\{u_j:j\geq N\}
\]
and an actual Borel $a\in\mathcal A_{S,b}(\Omega)$ such that
$v_N\to a$ locally in outer relative capacity on every permitted
compact/domain pair.

For an ordinary $\Omega$ as in this theorem, use the same definition
of $\mathcal A_S^{[0,1]}(\Omega)$ as in
Definition~\ref{def:r3-normalized-completion}: its normalized
approximating functions must be defined on the entire $\Omega$.
If $0\leq u_j\leq1$ on that entire original $\Omega$, then
$0\leq v_N\leq1$ there and one may take
$a\in\mathcal A_S^{[0,1]}(\Omega)$. Its normalized approximants
are defined on that same domain. The completed weighted mixed
calculus applies to these convex blocks and their limit.
\end{theorem}

\begin{proof}
Choose a countable cofinal family of norm-box stages
$V_l\Subset\Omega_l\Subset\Omega$ as in the R2 completion theorem.
Every permitted compact/domain pair is covered by finitely many
$V_l$ whose $\Omega_l$ lie inside its prescribed capacity domain.

For each $l$, the common bounds on the original sequence and
the smooth collar construction give $c_l$, a smooth psh collar
$h_l$, and a fixed compact $L_l\Subset\Omega_l$, such that
\[
 u_{j,l}=\max(u_j+c_l,h_l)
\]
equals $u_j+c_l$ near $\overline V_l$ and equals $h_l$ off $L_l$.
These functions have a common bound near $L_l$. The construction
uses the uniform bound only; no monotonicity of $u_j$ is required.

Put $U_j=(u_{j,l})_{l\geq1}$. A finite convex vector
$W=\sum_jb_jU_j$ has components $W_l=\sum_jb_ju_{j,l}$ with
the same collar $h_l$. On $V_l$ it agrees with the global
convex combination $\sum_jb_ju_j+c_l$. These componentwise
convex combinations are retained; the nonlinear patching
operation is not applied a second time to the global sum.

Write $\mathcal E_l=\mathcal E_{S,h_l}$. Relative CLN gives
$B_l<\infty$ such that $|\mathcal E_l(W_l)|\leq B_l$ for
all these convex vectors. The functional
\[
 F(W)=\sum_{l\geq1}\frac{2^{-l}}{1+B_l}\mathcal E_l(W_l)
\]
is bounded and concave. The series converges uniformly and
absolutely on the convex hull. Let
\[
 m_N=\sup\{F(W):W\in\operatorname{conv}(U_j:j\geq N)\}.
\]
Then $m_N\downarrow m_\infty$ with finite limit. By the
definition of supremum, choose a finite vector $W_N$ in
the indicated tail with $F(W_N)\geq m_N-\epsilon_N$,
where $\epsilon_N=2^{-N}$. No compactness of the convex
hull or existence of a maximizing vector is assumed.

For $M\geq N$ the midpoint of $W_N,W_M$ belongs to the
$N$-tail. Its concavity gap is nonnegative and at most
\[
 \frac{m_N-m_M+\epsilon_N+\epsilon_M}{2}\longrightarrow0
       \quad(N,M\to\infty).
\]
Every summand of the gap is nonnegative. The midpoint
estimate consequently gives, for each fixed $l$,
\[
 I_S(W_{N,l},W_{M,l})\longrightarrow0.
\]
The all-degree estimate makes the component sequences
Cauchy in capacity on their fixed stages.

Use the same finite weights defining $W_N$ to put
$v_N=\sum_jb_{N,j}u_j$ on all of $\Omega$. It is in
the required tail convex hull and equals $W_{N,l}-c_l$
on $V_l$. Thus, for every $l$ and every $\delta>0$,
\[
 \operatorname{Cap}_S^*
   (V_l\cap\{|v_N-v_M|>\delta\};\Omega_l)\longrightarrow0.
\]
Finite subadditivity and domain monotonicity give this
Cauchy property on every prescribed compact/domain pair.
Run the summable-subsequence construction of the R2 completion
theorem on this one global countable cofinal list: at its
$m$th step make the first $m$ capacity errors summable.
The pointwise limsup of that single subsequence is one
actual Borel function on all of $\Omega$. It is the local
capacity limit on every stage, and preserves the common
local bounds and the interval $[0,1]$ when present.
The approximants remain the global $v_N$. This proves
all assertions. Weighted mixed convergence is then an
application of the already proved completed calculus.
\end{proof}

\begin{corollary}[Actual bounded pointwise limits have a completed calculus]
\label{cor:r3-pointwise-completed}
In the preceding setting, suppose $u_j\in\Cb(\Omega)$ are locally
uniformly bounded and converge at every point to $u$. Then the
actual Borel function $u$ belongs to $\mathcal A_{S,b}(\Omega)$.
It has local relative quasicontinuity and the completed weighted
BT calculus. There are finite tail convex combinations of the
given sequence converging locally in outer $S$-capacity to these
same point values modulo relative outer-null sets.

This does not assert weighted or higher-degree unweighted
BT convergence for the original unaveraged $u_j$.
\end{corollary}

\begin{proof}
Choose $v_N$ and its capacity limit $a$ from the theorem.
At every fixed point, every finite convex combination of
the $N$th tail tends to $u$ as $N\to\infty$, so $v_N\to u$
pointwise. The summable-subsequence construction for the
known capacity convergence gives a subsequence converging
pointwise to $a$ off locally relative-outer-null sets on
the same global cofinal list. Thus $u=a$ outside these
sets. Subadditivity shows that $v_N\to u$ in capacity
on each exact compact/domain pair. This is membership
in the completed class with the given Borel values.
The remaining assertions are the completed-calculus
and quasicontinuity theorems, and are
\texttt{DIRECT\_COROLLARY}.
\end{proof}

\paragraph{The nonlinear distinction.}
For $q=1$, linearity identifies a weak limit of $D(u_jS)$
with $D(aS)$, as in Theorem~\ref{thm:r3-linear-convex-blocks}.
For $q>1$ the measure of a convex combination contains mixed
terms. The new theorem does not identify a weak limit of
$(Du_j)^qS$ with $(Da)^qS$. It provides capacity compactness
after convex averaging, not capacity precompactness of the
original sequence; the latter remains false even for the
ambient background and continuous potentials.

\paragraph{Status and dependence.}
The all-degree energy-capacity estimate and the simultaneous
concave-tail selection argument are \texttt{NEW\_PROVED}.
The positive energy identity and the Aubin--Mabuchi variation
formulas are \texttt{DIRECT\_COROLLARY} of the bounded mixed
and common-collar Green calculus. No compact outer regularity,
strictly psh exhaustion, arbitrary-current wedge, or arbitrary
varying-background stability is used.


% END source: base_r3/sections/energy_convex_blocks.tex


% BEGIN frozen or reviewed source: base_r3/sections/strong_relative_topology.tex
% Reconstructed after the scratch reset from the complete proof record.
% Previous independently reviewed source SHA256:
% b5b8bb8d7156c7574dc8297ee05c5759aab617fdacda65f5b87d5df7db2e4dfe.
\section{Strong psh-dual topology for arbitrary fixed backgrounds}
\label{sec:r3-strong-relative-topology}

Keep the fixed finite extension $K/\mathbf Q_p$, its original
absolute value, and the allowed analytic spaces. Let $S$ be a
fixed symmetric original-positive, separately $d'$- and
$d''$-closed current of bidegree $(p,p)$, and put $q=n-p\geq1$.
All identities below use the full original compact smooth test
sheaf. Write $\mu_u=(Du)^qS$ for $u\in\mathcal C_b(U)$.
For a compactly supported finite signed measure $\sigma$ set
\[
 \|\sigma\|_{\mathrm{psh},U}
 =\sup_{\substack{a\in\mathcal C_b(U)\\0\leq a\leq1}}
          \left|\int a\,d\sigma\right|,
 \qquad
 d_{\mathrm{psh},U}(\nu,\nu')
       =\|\nu-\nu'\|_{\mathrm{psh},U}.
\]
This is a seminorm; no faithfulness assertion is implicit.

\subsection{The common-collar estimate}

Let $\Omega$ be a prescribed ordinary norm-box domain. Let
$L\Subset\Omega$ lie in a strictly smaller norm box, and let $h$
be smooth classical psh with semipositive original curvature form.
Suppose
\[
 u,v\in\mathcal C_b(\Omega),\qquad |u|,|v|\leq B,\qquad
 u=v=h\quad\hbox{on }\Omega\setminus L.
\]
Choose original smooth $0\leq\chi\leq1$ of compact support,
equal to one on a neighborhood of $L$. Locality shows that
$\mu_u-\mu_v$ is supported in $L$. Thus
\[
 \mathfrak d_S(u,v)
   :=\|\mu_u-\mu_v\|_{\mathrm{psh},\Omega}
    =d_{\mathrm{psh},\Omega}(\chi\mu_u,\chi\mu_v)
\]
does not depend on the cutoff. Put
$I_S(u,v)=\int(u-v)(\mu_v-\mu_u)$.

\begin{theorem}[Energy and strong psh-dual distance]
\label{thm:r3-relative-energy-dual}
Under these hypotheses there is a finite constant $C$, depending
only on the fixed compact stages, the common bound, $q$, and
finitely many positive coefficient evaluations of $S$, such that
\begin{equation}
 \mathfrak d_S(u,v)\leq C I_S(u,v)^{1/2},
 \qquad I_S(u,v)\leq4B\,\mathfrak d_S(u,v).
 \label{eq:r3-relative-energy-dual}
\end{equation}
The constant is independent of the normalized tester and of
regularization indices.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the uniform
forward estimate; the reverse estimate is a
\texttt{DIRECT\_COROLLARY} of normalization and positivity of $I_S$.

\begin{proof}
Set $f=u-v$ and
\[
 Q=\sum_{b=0}^{q-1}(Du)^b(Dv)^{q-1-b}S .
\]
Each summand is an already defined positive separately closed
current of bidegree $(n-1,n-1)$. Symmetry, telescoping, and the
compact Green identity give
\[
 D(fQ)=\mu_u-\mu_v,\qquad
 \int\Gamma_Q(f)=I_S(u,v)\geq0 .
\]
For a normalized actual classical tester $a$, the protected
common-collar approximation and the bounded Green identity yield
\[
 \int a(\mu_u-\mu_v)=-\int\chi\,\Gamma_Q(a,f).
\]
There is no cutoff error: $\Gamma_Q(\chi,f)=0$, since $\chi$ is
constant near $L$ and $f=0$ off $L$. This identity is legitimate
for bounded, possibly discontinuous $a,u,v$ by the fixed-background
energy calculus; one first uses smooth rows on the prescribed box
and then its weighted convergence. No varying arbitrary background
is passed to a limit.

For $0\leq a\leq1$, the chain identity and positivity give
\[
 0\leq\int\chi\Gamma_Q(a)
 =\tfrac12\langle a^2Q,D\chi\rangle
                  -\int\chi a\,D(aQ)
 \leq\tfrac12|\mu_{Q,D\chi}|(\operatorname{supp}\chi).
\]
Relative CLN bounds the last expression uniformly in $u,v,a$.
Cauchy--Schwarz therefore bounds the preceding pairing by
$C I_S(u,v)^{1/2}$. Taking the supremum proves the first inequality.
If $B>0$, the functions $(u+B)/(2B)$ and $(v+B)/(2B)$ are
normalized testers. Their difference is $f/(2B)$, so
$I_S(u,v)\leq4B\mathfrak d_S(u,v)$. The case $B=0$ is trivial.
\end{proof}

\begin{corollary}[Equivalent relative topologies]
\label{cor:r3-strong-relative-topology}
For uniformly bounded common-collar potentials, convergence of
$I_S(u_j,u)$ to zero, convergence of $\mathfrak d_S(u_j,u)$ to
zero, and local outer $S$-capacity convergence on all permitted
compact/domain pairs are equivalent. The corresponding Cauchy
conditions are equivalent as well. These statements concern
relative capacity classes, not equality of actual values at
capacity-zero points.
\end{corollary}

\begin{proof}
Combine Theorem~\ref{thm:r3-relative-energy-dual} with
Theorem~\ref{thm:r3-I-capacity} and
Corollary~\ref{cor:r3-energy-capacity-topology}. For completeness,
the capacity-Cauchy implication for pairs follows directly from
uniform coefficient-capacity control:
\[
 I_S(u_j,u_k)
 \leq\int_L|u_j-u_k|(\mu_{u_j}+\mu_{u_k})
 \leq C\delta+
 C'\operatorname{Cap}_S^*(L\cap\{|u_j-u_k|>\delta\};\Omega).
\]
First let $j,k$ tend to infinity, then $\delta\downarrow0$.
Reverse framed domain comparison supplies the all-domain local
form of fixed-domain capacity convergence.
\end{proof}

\subsection{Localized strong convergence without a collar}

\begin{theorem}[Capacity convergence implies strong measure convergence]
\label{thm:r3-relative-capacity-strong-dual}
Let $u_j,u\in\mathcal C_b(U)$ have one common actual bound on every
compact subset. Suppose $u_j\to u$ locally in outer $S$-capacity,
on all permitted compact/domain pairs. For every original smooth
compact scalar $0\leq\chi\leq1$,
\[
 d_{\mathrm{psh},U}(\chi\mu_{u_j},\chi\mu_u)\longrightarrow0.
\]
Neither continuity nor monotonicity of the potentials is required.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
A finite nonnegative smooth partition reduces the cutoff to
one prescribed regularization box. On this box choose original
compact cutoffs $0\leq\eta\leq\kappa\leq1$, with $\eta=1$ near
$\operatorname{supp}\chi$ and $\kappa=1$ near
$\operatorname{supp}\eta$. Put
\[
 f_j=u_j-u,\qquad
 Q_j=\sum_{b=0}^{q-1}(Du_j)^b(Du)^{q-1-b}S,\qquad
 e_j=\int\eta\,\Gamma_{Q_j}(f_j).
\]
All these expressions use the established bounded energy calculus.

First let $a$ be a smooth classical psh function with $0\leq a\leq2$
on a neighborhood of the cutoff stages. Stokes gives
\begin{equation}
 \int a\chi\,D(f_jQ_j)
 =-\int\chi\,\Gamma_{Q_j}(a,f_j)
          -\int a\,\Gamma_{Q_j}(\chi,f_j).
 \label{eq:r3-local-strong-green}
\end{equation}
Here is a precise limiting justification. Fix $j$ and the background
$Q_j$ first, and smooth only $u_j,u$ downward on the prescribed
compact box. The smooth identity passes by distributional and
energy convergence against the fixed smooth tests $\chi$ and
$\kappa a$. A smooth cutoff is locally a difference of two smooth
classical functions after adding a sufficiently large convex
quadratic in the unit moment coordinates. The established
bilinear energy identity therefore applies. This passage does not
pair a changing nonsmooth multiplier with a changing coefficient
measure.

The chain identity, $a\geq0$, and relative CLN show
\[
 \int\eta\,\Gamma_{Q_j}(a)
 \leq\int\kappa\,\Gamma_{Q_j}(a)
 \leq2|\mu_{Q_j,D\kappa}|(\operatorname{supp}\kappa)\leq C.
\]
The energy of the fixed smooth $\chi$ is uniformly bounded by the
same coefficient estimates. Apply the weighted Cauchy--Schwarz
inequality to each term in
\eqref{eq:r3-local-strong-green}; $a\leq2$ controls the second.
One obtains
\begin{equation}
 \left|\int a\chi\,d(\mu_{u_j}-\mu_u)\right|
                  \leq C_\chi e_j^{1/2}.
 \label{eq:r3-local-strong-bound}
\end{equation}
The constant is independent of this tester.

Now let $a\in\mathcal C_b(U)$ be any actual normalized tester.
Its restriction has a smooth decreasing classical row on the
fixed compact stages; late terms lie between zero and two there.
For fixed $j$, bounded convergence against the fixed finite signed
measure $\chi(\mu_{u_j}-\mu_u)$ passes
\eqref{eq:r3-local-strong-bound} to $a$. No energy limit involving
this tester is needed. The same bound holds uniformly for every
normalized actual tester on $U$.

It remains to prove $e_j\to0$. The energy expansion is
\[
 e_j=\tfrac12\langle f_j^2Q_j,D\eta\rangle
                    -\int\eta f_j\,d(\mu_{u_j}-\mu_u).
\]
All coefficient measures in this formula belong to explicit mixed
products over the one fixed $S$. Formula
\eqref{eq:r2-approx-coefficient-cap}, with the common local bounds,
gives uniform mass and capacity control on the fixed stages.
Splitting into $|f_j|\leq\delta$ and $|f_j|>\delta$ bounds the first
term in absolute value by
$C\delta^2+C'\operatorname{Cap}_S^*(\{|f_j|>\delta\}\cap H;\Omega)$
and the second by
$C\delta+C'\operatorname{Cap}_S^*(\{|f_j|>\delta\}\cap H;\Omega)$,
for a fixed compact stage $H$. Let $j\to\infty$ and then
$\delta\downarrow0$. Thus $e_j\to0$.
Taking the supremum in \eqref{eq:r3-local-strong-bound} and summing
the finitely many cutoff pieces proves the theorem.
\end{proof}

\begin{corollary}[Bounded monotone and mixed strong convergence]
\label{cor:r3-bounded-mixed-strong-dual}
The localized psh-dual monotone convergence theorem
\ref{thm:r3-psh-dual-monotone} holds for locally uniformly bounded
actual classical decreasing rows, without requiring that their
terms be continuous. More generally, mixed top-degree products
whose finitely many slots converge locally in outer $S$-capacity
with common bounds converge in the localized psh-dual seminorm.
\end{corollary}

\begin{proof}
The proved decreasing convergence in relative capacity and
Theorem~\ref{thm:r3-relative-capacity-strong-dual} give the first
assertion. For the second use finite polarization,
\[
 q!T_{S,q}(a_1,\ldots,a_q)
 =\sum_{J\subset\{1,\ldots,q\}}(-1)^{q-|J|}
       \left(D\sum_{i\in J}a_i\right)^qS.
\]
Every sum has common local bounds and the required capacity
convergence. The triangle inequality finishes the proof.
\end{proof}

\paragraph{Scope.}
The background may charge ambient pluripolar sets. The theorem
does not replace relative quasi-everywhere equality by pointwise
equality, solve the general potential-quotient problem, or allow
an arbitrary changing background $S_j$. For $q=0$ the differentiated
measure is the fixed background itself, so the corresponding
convergence assertion is immediate.

% END source: base_r3/sections/strong_relative_topology.tex


% BEGIN frozen or reviewed source: base_r3/sections/completed_strong_topology.tex
\section{Strong convergence to relative completed potentials}
\label{sec:r3-completed-strong-topology}

Keep the arbitrary fixed background $S$ and the original psh-dual
seminorm of Section~\ref{sec:r3-strong-relative-topology}. For
$a\in\mathcal A_{S,b}(U)$, write
\[
 \mathcal\mu_a=\mathcal T_{S,q}(a,\ldots,a),\qquad q=n-p\geq1.
\]
This section does not assume that $a$ has an ambient classical
representative.

\begin{corollary}[Strong convergence to the completed measure]
\label{cor:r3-completed-strong-dual}
Let $u_j\in\mathcal C_b(U)$ have common actual bounds on compact
subsets and converge locally in outer $S$-capacity, on all
permitted compact/domain pairs, to a locally bounded Borel function
$a$. Then $a\in\mathcal A_{S,b}(U)$ and, for every original
smooth compact scalar $0\leq\chi\leq1$,
\begin{equation}
 d_{\mathrm{psh},U}
       \bigl(\chi(Du_j)^qS,\chi\mathcal\mu_a\bigr)
                     \longrightarrow0.
 \label{eq:r3-completed-strong-dual}
\end{equation}
The limit uses the actual completed relative potential, not its
ambient usc regularization. In particular, whenever the proved
bounded relative-capacity completion supplies such a limit to a
capacity-Cauchy sequence, its localized measures are Cauchy and
converge in this strong seminorm as well.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
pair estimate in
Theorem~\ref{thm:r3-relative-capacity-strong-dual} and the
completed weighted BT calculus.

\begin{proof}
Capacity convergence implies the corresponding capacity Cauchy
condition. Apply the proof of
Theorem~\ref{thm:r3-relative-capacity-strong-dual} to a pair
$u_j,u_k$. On each of its finitely many cutoff stages put
\[
 f_{j,k}=u_j-u_k,\qquad
 Q_{j,k}=\sum_{b=0}^{q-1}(Du_j)^b(Du_k)^{q-1-b}S.
\]
The constants are uniform for both indices, and the proof gives
\[
 d_{\mathrm{psh},U}\bigl(\chi(Du_j)^qS,\chi(Du_k)^qS\bigr)
       \leq C\left(\int\eta\,\Gamma_{Q_{j,k}}(f_{j,k})
                                                   \right)^{1/2}
\]
on a single stage; for a general cutoff one sums finitely many
such bounds. The energy expansion and uniform coefficient-capacity
estimates in that proof show that every displayed energy tends
to zero as $j,k\to\infty$. This step involves only classical
potentials and explicit mixed backgrounds over the one fixed $S$.
No energy measure of an unproved Borel representative is introduced.

The completed weighted calculus already supplies
$\mathcal\mu_a$ as a positive Radon measure. For each fixed
normalized tester $b\in\mathcal C_b(U)$ it gives
\[
 \int b\chi(Du_k)^qS\longrightarrow\int b\chi\,d\mathcal\mu_a.
\]
Indeed $b$ is simply a constant sequence in the weighted slot.
Given $\epsilon>0$, choose $N$ so that the preceding pair
seminorm is at most $\epsilon$ for all $j,k\geq N$. For
fixed $j\geq N$ and fixed $b$, let $k\to\infty$ to obtain
\[
 \left|\int b\chi\,d\bigl((Du_j)^qS-\mathcal\mu_a\bigr)\right|
                         \leq\epsilon.
\]
The bound is independent of $b$. Taking its supremum only now
proves \eqref{eq:r3-completed-strong-dual}. Thus the argument
does not interchange a weak limit with an uncontrolled supremum.
\end{proof}

\begin{corollary}[Stability within the bounded completed class]
\label{cor:r3-strong-within-completion}
Suppose $a_j,a\in\mathcal A_{S,b}(U)$ converge locally in outer
$S$-capacity and, locally, their defining classical approximants
have one common actual bound. Then
\[
 d_{\mathrm{psh},U}(\chi\mathcal\mu_{a_j},
                         \chi\mathcal\mu_a)\longrightarrow0
\]
for every cutoff as above. The corresponding statement holds for
mixed top-degree products with finitely many convergent slots and
common defining bounds.
\end{corollary}

\begin{proof}
Work first on one of the fixed local cutoff stages. For every
$j$, choose a sufficiently late classical defining approximant
$v_j$ of $a_j$ so that its capacity error tends to zero on the
finitely many larger compact stages used by the local estimate,
and its localized measure differs from that of $a_j$ in
psh-dual seminorm by at most $1/j$. Such a choice is possible
by Corollary~\ref{cor:r3-completed-strong-dual}, applied to the
defining sequence for this fixed $j$. The common defining bound
keeps all constants uniform. A countable diagonal, if necessary,
also makes $v_j\to a$ in the local all-domain capacity sense.
The same corollary applied to $v_j$ gives its strong convergence
to $\mathcal\mu_a$. The triangle inequality proves the assertion.
Local tester estimates apply to restrictions of normalized testers
on $U$, and a finite partition of the compact cutoff gives the
stated global seminorm bound. Finally use the finite polarization
identity of Corollary~\ref{cor:r3-bounded-mixed-strong-dual}
for the mixed statement. No classical representative of the
completed limits is required.
\end{proof}

\paragraph{Scope.}
This strengthens stability of the already defined relative
completed operator. It does not make the ambient classical class
closed under these limits, and it does not add an arbitrary
varying-background convergence theorem. The explicit completed
solution in Proposition~\ref{prop:r3-relative-completed-dirichlet-example}
is one application where an ambient classical zero-boundary
solution is impossible.

% END source: base_r3/sections/completed_strong_topology.tex


% BEGIN frozen or reviewed source: base_r3/sections/pluripolar_decomposition.tex
\section{A canonical pluripolar decomposition of positive closed currents}
\label{sec:r3-pluripolar-decomposition}

The background in this section is arbitrary. The purpose of the
decomposition is to locate the part on which replacing an actual
increasing limit by its ambient upper regularization can lose
information. The construction uses actual coefficient measures,
not an unspecified ambient volume measure.

We work over the prescribed finite extension $K/\mathbf Q_p$.
A current is positive in the original CLD cone, is symmetric,
and is separately $d'$- and $d''$-closed. All tests are original
compactly supported smooth CLD forms. A current $T$ is said to be
\emph{carried by} a Borel set $E$ if $\mathbf1_E T=T$.
This is a coefficient-measure assertion; it does not assert that
the topological support of $T$ is contained in a nonclosed $E$.

\subsection{A finite measure detecting all coefficient-null sets}

\begin{lemma}[A countable family of coefficient controls]
\label{lem:r3-countable-trace-controls}
Let $U$ be an ordinary open in the analytification of a smooth
integral affine $K$-variety of dimension $n$. Fix $0\leq p\leq n$.
There is a countable family of positive original forms
$\Theta_l\in A_c^{n-p,n-p}(U)$ with this property: for every
original compact smooth complementary test $\eta$, there are
finitely many indices $l_1,\ldots,l_N$ and a finite constant
$C_\eta$ such that every original-positive $(p,p)$-current $T$
satisfies the inequality of Radon measures
\[
 |\mu_{T,\eta}|
       \leq C_\eta\sum_{j=1}^N\mu_{T,\Theta_{l_j}}.
 \tag{1}\label{eq:r3-countable-trace-control}
\]
The controls depend on $U,p$, but not on $T$.
\end{lemma}

\begin{proof}
Choose a countable dense subfield $K_0\subset K$ containing the
finitely many coefficients needed to define the affine variety
and its fixed affine embedding. Such a field exists since $K$ is
finite over $\mathbf Q_p$ and is separable as a topological field.
There are countably many rational functions with coefficients
in $K_0$ and countably many finite lists of these functions.

Every finite analytic unit presentation can locally be replaced
by one of these lists without changing any logarithmic absolute
value. Indeed B, Lemma D.1, first replaces the analytic units
on a compact affinoid neighborhood by rational functions over
$K$. Write their numerators and denominators in the fixed
affine coordinates. The numerators and denominators are units
on that compact neighborhood. Approximate their finitely many
coefficients by elements of $K_0$ closely enough that the
uniform errors are smaller than the positive minima of the
original absolute values. The ultrametric inequality preserves
each numerator and denominator norm, and hence each rational
function norm, at every point of the neighborhood. Their
original smooth logarithmic differentials also agree on its
ordinary interior. This is an exact replacement of the moment
coordinates, not a limiting identification of currents.

For each finite rational list, let $O$ be its algebraic unit
locus intersected with $U$, and write its logarithmic
coordinates as $t_1,\ldots,t_s$. The space $O$ is locally
compact and second countable, so choose a countable compact
exhaustion with interiors covering $O$, and nonnegative
original smooth compact cutoffs $\chi_j$ equal to one near
these compact sets. Put $r=n-p$ and include the forms
\[
 \chi_j\frac1{r!}
       \left(\sum_{i=1}^s d't_i\wedge d''t_i\right)^r
\]
in the countable family, extending them by zero off $O$.
For $r=0$ the parenthesized factor is one. Original smooth
cutoffs and compact exhaustions are available by the same
ordinary topology and partition-of-unity inputs as in R1.

Now divide a given compact test $\eta$ into finitely many
pieces on neighborhoods where the exact rational replacements
apply. Each piece has one of the listed moment presentations.
On its fixed compact support, its symmetric coefficient
matrix is bounded by a multiple of the positive definite
matrix of the displayed power. Choose $\chi_j=1$ near that
support. R's finite-test cone domination, recalled in
Lemma~\ref{lem:finite-positive-controls}, then gives
\eqref{eq:r3-countable-trace-control}. More explicitly, its
cone inequalities remain valid after multiplication by any
original smooth scalar on a common refinement; Riesz
uniqueness therefore makes the domination an inequality of
coefficient measures. Positive currents annihilate the
antisymmetric null part. Summing the finitely many pieces
proves the assertion.
\end{proof}

\begin{corollary}[A faithful finite control measure]
\label{cor:r3-faithful-control-measure}
For every such positive current $S$, there is a finite positive
Radon measure $\lambda_S$ on $U$ such that, for every Borel
$E\subset U$,
\[
 \lambda_S(E)=0
 \quad\Longleftrightarrow\quad
 \mathbf1_E S=0.
 \tag{2}\label{eq:r3-faithful-null-ideal}
\]
The measure need not be canonical; its null ideal is canonical.
\end{corollary}

\begin{proof}
Set $\mu_l=\mu_{S,\Theta_l}$. Each is a finite positive
Radon measure, since $\Theta_l$ has compact support. Define
\[
 \lambda_S=\sum_{l\geq1}
       \frac{2^{-l}}{1+\mu_l(U)}\,\mu_l.
\]
Its total mass is at most one. It is Radon: finite initial
sums are Radon, and the tails have uniformly small total
mass, so compact inner approximation passes to the sum.
If $\lambda_S(E)=0$, then $\mu_l(E)=0$ for every $l$;
the preceding lemma makes every coefficient restriction
$\mathbf1_E\mu_{S,\eta}$ vanish. Conversely,
$\mathbf1_E S=0$ makes all $\mu_l(E)$ vanish. This proves
\eqref{eq:r3-faithful-null-ideal}.
\end{proof}

These two control statements are \texttt{DIRECT\_COROLLARY}
of the original coefficient positivity in R, exact moment
replacement in B, and separability of the prescribed field.
No globally strictly positive form is used. The countable
family may contain arbitrarily many different moment
presentations.

\subsection{The maximal pluripolar part}

Call $S$ \emph{non-pluripolar} if
$\mathbf1_E S=0$ for every Borel locally pluripolar set $E$.
This term refers to all its coefficient measures, not only
to one selected top-degree trace.

\begin{theorem}[Pluripolar decomposition on an exhaustion domain]
\label{thm:r3-pluripolar-decomposition}
Let $V$ be a connected exhaustion domain of
Definition~\ref{def:r2-exhaustion-domain}, of positive dimension.
Every allowed positive closed current $S$ on $V$ has a unique
decomposition
\[
 S=S_{\rm pp}+S_{\rm np}
 \tag{3}\label{eq:r3-polar-decomposition}
\]
into positive symmetric separately closed currents such that
$S_{\rm pp}$ is carried by a complete pole set in $V$ and
$S_{\rm np}$ is non-pluripolar.
There is a nonpositive, nontrivial extended classical psh
function $\phi$ on $V$, with $Q=\{\phi=-\infty\}$, for which
\[
 S_{\rm pp}=\mathbf1_QS,
 \qquad S_{\rm np}=\mathbf1_{V\setminus Q}S.
 \tag{4}\label{eq:r3-polar-split-by-witness}
\]
The two currents are independent of the control measure and
the chosen witness. They commute with ordinary restriction,
addition of positive closed currents, and nonnegative scaling.
\end{theorem}

\begin{proof}
Use the finite control measure $\lambda_S$ above. Let
$\mathscr P$ consist of the complete pole sets of nonpositive,
nontrivial extended classical psh functions on $V$, and put
\[
 m=\sup_{P\in\mathscr P}\lambda_S(P)\leq\lambda_S(V).
\]
The empty set belongs to $\mathscr P$. Choose $P_j\in\mathscr P$
with $\lambda_S(P_j)\to m$, and choose nonpositive witnesses
$\phi_j$. Fix one ambient divisorial point $z\in V$.
By the R2 no-divisorial-poles theorem, every $\phi_j(z)$ is
finite. Choose positive $b_j$ such that
$b_j|\phi_j(z)|\leq2^{-j}$. The decreasing sum
\[
 \phi=\sum_{j\geq1}b_j\phi_j
\]
is extended classical psh and is nontrivial, since
$\phi(z)\geq-1$. This is the negative-series construction
already used in the R2 pole-witness theorem. Its pole set
$Q$ contains every $P_j$, so
$m\leq\lambda_S(Q)\leq m$. In particular the supremum is
attained by this one complete pole set.

Let $E\subset V$ be Borel and locally pluripolar.
The R2 same-domain pole-patching theorem provides a member
of $\mathscr P$ containing $E$. Applying the same negative
sum construction to that witness and $\phi$ gives
$R\in\mathscr P$ containing $E\cup Q$. Maximality gives
\[
 \lambda_S(E\setminus Q)
       \leq\lambda_S(R)-\lambda_S(Q)=0.
\]
By \eqref{eq:r3-faithful-null-ideal},
$\mathbf1_{E\setminus Q}S=0$. Thus the second current in
\eqref{eq:r3-polar-split-by-witness} is non-pluripolar.
The R2 pole-restriction theorem makes both displayed
restrictions positive and separately closed. This proves
existence, including the actual coefficient interpretation.

For uniqueness, suppose $S=A+B$ is another such
decomposition, with $A$ carried by a locally pluripolar
Borel set $P$ and $B$ non-pluripolar. Positivity gives
$0\leq A\leq S$. Since
$\mathbf1_{P\setminus Q}S=0$, positivity of coefficient
controls gives $\mathbf1_{P\setminus Q}A=0$. Hence
$\mathbf1_Q A=A$. Also $\mathbf1_QB=0$. Restricting
$S=A+B$ to $Q$ now gives $A=S_{\rm pp}$ and
$B=S_{\rm np}$.

These properties are local. In particular restriction of
a non-pluripolar current to an ordinary open $W\subset V$
is still non-pluripolar. To check this precisely, exhaust
$W$ by compact subsets: a locally pluripolar set intersected
with one of these compacts is locally pluripolar in $V$,
and coefficient measures pass to the countable union.
Uniqueness therefore identifies the restricted decomposition
with the one constructed on a smaller exhaustion domain.
For sums, the sum of the non-pluripolar parts is
non-pluripolar, and the sum of the pluripolar parts is
carried by the union of their two pole sets, contained in
one complete pole set by the negative-sum construction.
Uniqueness proves additivity. Scaling is immediate.
\end{proof}

\begin{corollary}[Local decomposition on the allowed analytic spaces]
\label{cor:r3-local-polar-decomposition}
On any allowed analytic space, the preceding local
decompositions glue to a canonical decomposition
$S=S_{\rm pp}+S_{\rm np}$. Locally, $S_{\rm pp}$ is carried
by a complete pole set; $S_{\rm np}$ gives zero mass to
every locally pluripolar Borel set in all coefficient measures.
No global countability assumption on the analytic space is
required for this local assertion.
\end{corollary}

\begin{proof}
Use B's ordinary algebraic neighborhoods and the R2 local
basis of exhaustion domains. The uniqueness just proved
identifies the decompositions on every smaller such domain
in an overlap. Currents and their positivity and separate
closedness are local for ordinary opens, so the pieces glue.
In dimension zero the only locally pluripolar set is empty;
the assertion there is $S_{\rm pp}=0$, $S_{\rm np}=S$.
\end{proof}

\subsection{Compatibility with the bounded Bedford--Taylor calculus}

\begin{theorem}[Decomposition commutes with bounded mixed products]
\label{thm:r3-polar-decomposition-bt}
For every $0\leq k\leq n-p$ and $u_1,\ldots,u_k\in\Cb(U)$,
\[
 \bigl(T_{S,k}(u_1,\ldots,u_k)\bigr)_{\rm pp}
       =T_{S_{\rm pp},k}(u_1,\ldots,u_k),
\]
\[
 \bigl(T_{S,k}(u_1,\ldots,u_k)\bigr)_{\rm np}
       =T_{S_{\rm np},k}(u_1,\ldots,u_k).
\]
These are current equalities against all original tests.
In particular all mixed products over a non-pluripolar
background have non-pluripolar coefficient measures.
\end{theorem}

\begin{proof}
Work on an exhaustion neighborhood where
$S_{\rm pp}=\mathbf1_QS$ as above. Linearity of the
background in the recursive construction gives
\[
 T_{S,k}=T_{S_{\rm pp},k}+T_{S_{\rm np},k}.
\]
The R2 pole-restriction identity gives
$\mathbf1_QT_{S_{\rm pp},k}=T_{S_{\rm pp},k}$.
For any locally pluripolar Borel set $E$, choose a complete
pole set $P$ containing it on this neighborhood. Since
$\mathbf1_PS_{\rm np}=0$, the same identity gives
\[
 \mathbf1_PT_{S_{\rm np},k}
      =T_{\mathbf1_PS_{\rm np},k}=0.
\]
Thus the two terms have exactly the required polar and
non-pluripolar properties. Uniqueness of the decomposition,
applied to the resulting positive closed current, identifies
them with its canonical parts. Locality proves the assertion
on $U$.
\end{proof}

\paragraph{Status and limits.}
The decomposition and its BT compatibility are recorded as
\texttt{NEW\_PROVED} relative to the proved R2 pole
restriction and pole-patching theorems. Original smooth
positive backgrounds and bounded ambient BT backgrounds are
non-pluripolar by R2's ambient trace theorem. Subvariety
integration backgrounds can instead be entirely pluripolar.
This is not an analytic-stratum or Siu decomposition: no
analyticity of a pole set is asserted. Nor is the capacity
of a sum asserted to be the sum of the capacities, since
the same test potential enters both summands.


% END source: base_r3/sections/pluripolar_decomposition.tex


% BEGIN frozen or reviewed source: base_r3/sections/pluripolar_zero_extension.tex
\section{Zero extension across a closed complete pluripolar set}
\label{sec:r3-zero-extension}

The restriction theorem of R2 starts with a current on the whole
ordinary open set. Here the current is initially defined only on the
complement of the exceptional set. Its closedness across that set is
part of the conclusion. The proof adapts the classical El Mir
argument, but the finite-mass hypothesis and the first-order estimate
are formulated for all original CLD coefficient tests.

The base field, analytic category, and original positivity cone are
unchanged. A closed set $P\subset U$ is \emph{locally complete
pluripolar} here if, near each of its points, it is exactly the pole
set of a nontrivial extended classical psh function. On a smaller
neighborhood such a witness can be taken nonpositive. No assertion
below replaces this equality by containment in an unspecified
larger pole set.

\subsection{The mass condition and a first-order estimate}

Let $S$ be a symmetric original-positive $(p,p)$-current on
$U\setminus P$, and put $q=n-p$. For an original compact smooth
$(q,q)$-form $\eta$ on $U$, denote by $\mu_{S,\eta}$ the coefficient
measure on $U\setminus P$ obtained by restricting the form there.
This notation does not require that its restricted support be
compact in $U\setminus P$.

We say that $S$ has \emph{locally finite coefficient mass across
$P$} if, locally on $U$,
\begin{equation}
 \label{eq:r3-finite-mass-across-pole}
 |\mu_{S,\eta}|(U\setminus P)<\infty
 \quad\hbox{for every }\eta\in A_c^{q,q}(U).
\end{equation}
Equivalently, it suffices to require this for the finite positive
controls of each original test stage. In particular the condition
includes controls for the coordinates of compact cutoffs. It is
not a finite-mass assumption for just one initially selected
tropical trace.

\begin{lemma}[Coefficientwise extension and weighted first-order control]
\label{lem:r3-zero-extension-controls}
Under \eqref{eq:r3-finite-mass-across-pole}, the coefficientwise
zero extension
\[
 \langle\widetilde S,\eta\rangle
       =\int_{U\setminus P}d\mu_{S,\eta}
\]
is an original-positive symmetric coefficient-Radon current on $U$,
with $\mathbf1_P\widetilde S=0$.

Suppose $q\geq1$. Fix a moment presentation of an original
$(q-1,q)$-test $\alpha$ near its compact support, let $t_1,\ldots,t_s$
be its moment coordinates, and put
\[
 \beta=\sum_i d't_i\wedge d''t_i.
\]
Let $\chi\geq0$ be an original compact cutoff equal to one near
$\operatorname{supp}\alpha$, with support in that moment
presentation's ordinary domain. For every original smooth real $f$
and real Borel $b$ for which the displayed integrals are finite,
\begin{equation}
 \label{eq:r3-weighted-first-order-control}
 \left|\int_{U\setminus P}b\,d'f\wedge\alpha\wedge S\right|^2
 \leq C_\alpha
 \left(\int_{U\setminus P}\chi\,d'f\wedge d''f
                           \wedge\beta^{q-1}\wedge S\right)
 \left(\int_{U\setminus P}\chi b^2\beta^q\wedge S\right).
\end{equation}
There is the analogous inequality for $(q,q-1)$-tests and $d''f$.
The constant depends on the fixed test presentation and its bounded
coefficients, and not on additional moment coordinates used by $f$
or by bounded approximations to $b$. For a test with several
presentations one uses a finite partition and the corresponding
finite sum of these controls.
\end{lemma}

\begin{proof}
Extend each coefficient measure by zero on the closed set $P$.
The local finite-mass assumption makes these extensions Radon.
Their linear relations and common-refinement identities hold
off $P$ and both sides have zero restriction to $P$, so the same
identities hold on $U$. On each fixed test stage, the positive
controls in Lemma~\ref{lem:finite-positive-controls} give
an order-zero bound with finite right hand side. Hence the
extended functionals are continuous in the original test
topology and define a current. Positivity and symmetry pass
through zero extension. This proves the first assertion without
using either closedness or a Bedford--Taylor product.

For the second assertion, expand $\alpha$ as a finite sum of
bounded smooth coefficients times
$d't_I\wedge d''t_J$, where $|I|=q-1$ and $|J|=q$.
Use the standard Lagerberg signs for squares. On a common
refinement, set
\[
 A_I=d'f\wedge d't_I,\qquad B_J=b\,d't_J.
\]
The coefficient matrix of $S$ defines a positive semidefinite
bilinear measure on these $q$-vectors: the positivity needed
here is duality with the original positive form cone. The
two-by-two square inequality, followed by the scalar
Cauchy--Schwarz inequality for its densities, bounds each
mixed term by the square root of
\[
 \int\chi\, [A_I]^2\wedge S
       \quad\hbox{times}\quad
 \int\chi\, [B_J]^2\wedge S.
\]
Here $[A_I]^2$ denotes the positive square with that standard
sign. Bounded coefficients of $\alpha$ contribute a fixed
factor. Summing the first squares is bounded by a fixed
multiple of $d'f\wedge d''f\wedge\beta^{q-1}$, and summing
the second squares is bounded by a fixed multiple of
$b^2\beta^q$. The finite-sum Cauchy--Schwarz inequality
gives \eqref{eq:r3-weighted-first-order-control}.

One may first take $b$ smooth and then use coefficient-Radon
measure identities and bounded approximation, or apply the
same two-by-two inequality to Radon densities directly.
Pullback to additional moment coordinates preserves the
original positive cone. Thus the bounds use the fixed
coordinates of $\alpha$ but impose no common fixed
presentation on the varying $f$. The $d''$ assertion follows
by the CLD involution. A finite original partition proves
the last assertion.
\end{proof}

\subsection{Smooth cutoffs adapted to the pole set}

\begin{lemma}[Bounded psh cutoffs with a zero neighborhood]
\label{lem:r3-psh-zero-cutoffs}
Work on a sufficiently small ordinary algebraic neighborhood,
and let $H$ be a compact norm-box core in a larger such
neighborhood. Suppose $\phi\leq0$ is an extended classical
psh function there and $P=\{\phi=-\infty\}$ is closed.
On one fixed ordinary neighborhood of $H$ there are original
smooth classical psh functions $w_N$, with semipositive
original $(1,1)$-forms $Dw_N$, such that
\[
 0\leq w_N\leq1+N^{-2}\quad\hbox{on }H,\qquad
 w_N=0\quad\hbox{near }P\cap H,
\]
and $w_N(x)\to1$ for every $x\in H\setminus P$.
No monotonicity in $N$ is asserted or needed.
\end{lemma}

\begin{proof}
The truncation $\phi_N=\max(\phi,-2N)$ is bounded classical
psh on the larger neighborhood. By the prescribed-box
regularization in Lemma~\ref{lem:compact-box-regularization},
it has smooth classical psh approximants with semipositive
original curvature forms decreasing to
$\phi_N$ on a fixed smaller neighborhood containing $H$.
The smaller geometric box can be chosen independently of
$N$; the projective line and regularization index need
not be independent of $N$.

Choose one approximant $\psi_N\geq\phi_N$ such that
\[
 \psi_N<N^{-1}\quad\hbox{on }H,\qquad
 \psi_N<-3N/2\quad\hbox{on }P\cap H.
\]
Both choices follow from compactness and the decreasing
continuous rows: their limits are at most zero on $H$
and equal $-2N$ on $P\cap H$. The second inequality
holds on an ordinary neighborhood of $P\cap H$.

Let $F$ be the convolution of $t\mapsto\max(t,0)$ with
an even nonnegative smooth probability kernel supported
in $(-1/4,1/4)$. Then $F$ is nonnegative, smooth, convex
and nondecreasing, equals zero for $t\leq-1/4$, and
equals $t$ for $t\geq1/4$. Put
\[
 w_N=F(1+\psi_N/N).
\]
Convex nondecreasing composition preserves the classical
psh property, by the all-extensions curve definition;
the result is also an original smooth function. Moreover,
$Dw_N=F'(1+\psi_N/N)D\psi_N/N+
F''(1+\psi_N/N)d'\psi_N\wedge d''\psi_N/N^2$
is a semipositive original form.
The stated upper bound follows by monotonicity of $F$,
and the negative inequality on $\psi_N$ gives the zero
neighborhood. If $\phi(x)$ is finite, then, for all large
$N$, $\psi_N(x)\geq\phi(x)$ and
$1+\phi(x)/N>1/4$. Hence
\[
 1+\phi(x)/N\leq w_N(x)\leq1+N^{-2}.
\]
This proves the pointwise limit. No diagonal approximation
at all points is involved.
\end{proof}

\begin{theorem}[A Skoda--El Mir zero-extension theorem]
\label{thm:r3-zero-extension}
Let $P$ be a closed locally complete pluripolar subset of
an allowed ordinary open $U$. Let $S$ be a symmetric
original-positive $(p,p)$-current on $U\setminus P$,
separately $d'$- and $d''$-closed there. Assume
\eqref{eq:r3-finite-mass-across-pole}. Then its
coefficientwise zero extension $\widetilde S$ to $U$
is separately $d'$- and $d''$-closed.
All assertions are on all original compact smooth tests.
\end{theorem}

\begin{proof}
Lemma~\ref{lem:r3-zero-extension-controls} constructs $\widetilde S$ and proves
its positivity, symmetry and coefficient-Radon property.
If $p=n$, separate closedness follows from degree. Suppose
$q=n-p\geq1$. Closedness is local for ordinary opens.
Near a point of $P$, choose a nonpositive complete pole
witness and nested norm boxes as in
Lemma~\ref{lem:r3-psh-zero-cutoffs}. A finite partition
reduces each compact test to this situation and, within
it, to one fixed moment presentation.

Fix an original $(q-1,q)$-test $\alpha$. Choose a
nonnegative compact original smooth $\chi$, equal to
one near its support, with support in a compact core
$H$ on which the cutoffs $w_N$ are defined and within
the ordinary domain of that fixed moment presentation. Put
$\beta=\sum_i d't_i\wedge d''t_i$ for the coordinates
of $\alpha$. The positive current
$S\wedge\beta^{q-1}$ is separately closed on the
complement of $P$. The function $w_N$ is zero on
an ordinary neighborhood of $P\cap H$, so every term
below can be integrated by parts on that complement.
One can insert a further compact cutoff away from $P$
equal to one on the relevant supports; its differentiated
terms multiply zero functions.

The smooth identity
$D(w_N^2)=2w_NDw_N+2d'w_N\wedge d''w_N$ gives
\begin{align*}
 E_N
 &:=\int_{U\setminus P}
        \chi\,d'w_N\wedge d''w_N\wedge\beta^{q-1}\wedge S\\
 &=\frac12\int_{U\setminus P}
        w_N^2D\chi\wedge\beta^{q-1}\wedge S
   -\int_{U\setminus P}
        \chi w_NDw_N\wedge\beta^{q-1}\wedge S\\
 &\leq 2\,|\mu_{S,D\chi\wedge\beta^{q-1}}|
                         (\operatorname{supp}\chi)
 =:C<\infty .
\end{align*}
Here $0\leq w_N\leq2$ on the fixed compact set, and
the last term being subtracted is nonnegative.
The finite constant concerns the one fixed smooth
form $D\chi\wedge\beta^{q-1}$, including the moment
coordinates of $\chi$. This step has used closedness
only where $S$ was originally given.

Set
\[
 a_N=2w_N-w_N^2=1-(1-w_N)^2.
\]
On $H$ one has $0\leq a_N\leq1$; it is zero near
$P\cap H$, and tends to one at every point of
$H\setminus P$. Consequently $a_N\alpha$ is a
valid compact test on $U\setminus P$. Separate
closedness there gives, with the harmless sign
determined by total degree,
\[
 \left|\int_{U\setminus P}a_Nd'\alpha\wedge S\right|
 =2\left|\int_{U\setminus P}
              (1-w_N)d'w_N\wedge\alpha\wedge S\right|.
\]
By \eqref{eq:r3-weighted-first-order-control}, its
square is at most
\[
 4C_\alpha E_N
       \int_{U\setminus P}\chi(1-w_N)^2\beta^q\wedge S .
\]
The latter integral tends to zero by dominated
convergence for the fixed finite coefficient measure
$\chi\beta^q\wedge\widetilde S$. Its restriction to
$P$ is zero by construction. Uniform boundedness of
$E_N$ therefore makes the preceding error tend to zero.
Dominated convergence for the fixed signed coefficient
measure of $d'\alpha$ identifies the left hand side
in the limit with $|\langle\widetilde S,d'\alpha\rangle|$.
This proves $d'\widetilde S=0$. The $d''$ estimate and
the same argument prove $d''\widetilde S=0$.

Each localization and inequality uses original
forms and preserves common-refinement identities.
No convergence statement for a varying background
current or an unbounded multiplier was invoked.
\end{proof}

\subsection{Minimality and extension of bounded products}

\begin{corollary}[The least positive extension]
\label{cor:r3-minimal-zero-extension}
If $R$ is any symmetric original-positive separately closed
extension of $S$ to $U$, then
\[
 R=\widetilde S+A,
\]
where $A$ is positive, symmetric, separately closed and
carried by $P$. Conversely, adding any such $A$ gives
an extension. Thus $\widetilde S$ is the least positive
extension. It is the unique extension whose coefficient
measures give zero mass to $P$.
\end{corollary}

\begin{proof}
Coefficientwise, the difference equals $\mathbf1_PR$,
which is positive and symmetric. The theorem proves
that $\widetilde S$ is closed, so the difference is
separately closed by subtraction. It is carried by
$P$. The converse and uniqueness follow immediately
from the coefficient restrictions. No assertion is
made that a positive current carried by $P$ must
have an analytic-support description.
\end{proof}

\begin{corollary}[Bounded mixed products extend without extra mass]
\label{cor:r3-zero-extension-bt}
Let $u_1,\ldots,u_k\in\Cb(U)$, where $0\leq k\leq n-p$.
The current $T_{S,k}(u_1|_{U\setminus P},\ldots,
u_k|_{U\setminus P})$ has locally finite coefficient
mass across $P$, and its coefficientwise zero extension
is
\[
 T_{\widetilde S,k}(u_1,\ldots,u_k).
\]
This current gives zero coefficient mass to $P$.
The same coefficient equality holds after multiplication
by any locally bounded Borel weight on $U$.
\end{corollary}

\begin{proof}
The theorem makes $\widetilde S$ an admissible fixed
background on $U$. Its bounded mixed products exist
and have locally finite coefficients by the established
relative calculus and relative CLN estimate. Ordinary
restriction identifies them with the asserted products
on $U\setminus P$. The R2 pluripolar restriction identity
gives
\[
 \mathbf1_P T_{\widetilde S,k}(u_1,\ldots,u_k)
 =T_{\mathbf1_P\widetilde S,k}(u_1,\ldots,u_k)=0.
\]
These two coefficient assertions identify the product
with the zero extension and prove the mass claim.
Actual Borel multiplication preserves the equality.
The potentials in this corollary are required on $U$;
it does not assert a removable-singularity theorem
for arbitrary potentials given only off $P$.
\end{proof}

\paragraph{Status and dependencies.}
The zero-extension theorem is \texttt{NEW\_PROVED} in
the prescribed non-Archimedean setting. The proof uses
R's original coefficient cone and B's prescribed-box
regularization. Its cutoff mechanism is the classical
Skoda--El Mir mechanism; no novelty is claimed for
that classical argument. The mass and first-order
lemma makes the required transfer to all original
tests explicit. The last two corollaries are
\texttt{DIRECT\_COROLLARY}, the last also using the
R2 pole-restriction identity. Mere support avoidance
in CLD's zero-extension convention does not prove
the finite-mass extension theorem stated here.

% END source: base_r3/sections/pluripolar_zero_extension.tex


% BEGIN frozen or reviewed source: base_r3/sections/zero_extension_counterexamples.tex
\section{Two sharp scope tests for the zero-extension theorem}
\label{sec:r3-zero-extension-counterexamples}

These examples test two hypotheses of
Theorem~\ref{thm:r3-zero-extension}. They do not concern the
previously proved restriction theorem for a current already
defined on the whole space. The field and its absolute value
remain the prescribed finite $p$-adic ones; every form below
is an original CLD form.

\begin{proposition}[Closed pluripolar containment is insufficient]
\label{prop:r3-zero-extension-noncomplete-false}
There are a closed pluripolar set $E$ in a smooth two-dimensional
ordinary analytic domain $U$ and an original-positive symmetric
$(1,1)$-current $S$ on $U\setminus E$ such that:
\begin{enumerate}
\item $S$ is separately $d'$- and $d''$-closed;
\item all its original coefficient masses are locally finite
      across $E$;
\item its coefficientwise zero extension to $U$ is not closed.
\end{enumerate}
The exceptional set is not locally complete pluripolar.
Thus replacing \emph{closed locally complete pluripolar} by
\emph{closed pluripolar} in Theorem~\ref{thm:r3-zero-extension}
is \texttt{FALSE}.
\end{proposition}

\begin{proof}
Let $\pi$ be a uniformizer and $\lambda=-\log|\pi|>0$. Use
coordinates $(t,z)$ on $\mathbf G_{m,K}\times\mathbf A^1_K$ and put
\[
 s=\log|t|,\qquad
 U=\{-\lambda<s<\lambda,\ |z|<1\},\qquad
 Y=U\cap\{z=0\},\qquad E=Y\cap\{s=0\}.
\]
The set $U$ is an ordinary open in the smooth affine surface,
and $Y$ is its closed smooth analytic curve. No normalization
of the prescribed absolute value has been changed.
The set $E$ is closed in $U$. It is pluripolar because
\[
 E\subset Y=\{\log|z|=-\infty\},
\]
and $\log|z|$ is a nontrivial extended classical psh function
on $U$.

On $U\setminus E$, define the current of integration over
the component
\[
 Z=Y\cap\{s>0\}.
\]
Equivalently, this current is $[Y]$ locally where $s>0$
and is zero locally where $s<0$ or $z\ne0$. These neighborhoods
cover $U\setminus E$, so it is a well-defined symmetric
original-positive separately closed current $S$.
There is no boundary point of $Z$ inside $U\setminus E$.
Its zero extension is the actual Borel multiplication
\[
 T=\mathbf1_{\{s>0\}}[Y]\quad\hbox{on }U.
\]
For every original compact smooth $(1,1)$-test $\eta$,
\[
 |\mu_{S,\eta}|(U\setminus E)
       \leq|\mu_{[Y],\eta}|(U)<\infty.
\]
Thus the full coefficient-mass condition holds, including all
test presentations, not just the radial one used below.

Choose $f\in C_c^\infty((-\lambda,\lambda))$ with $f(0)=1$.
Choose an original smooth scalar cutoff $\kappa(z)$ on
$\{|z|<1\}$, compactly supported there and equal to one on a
neighborhood of $z=0$. One explicit construction takes a
smooth function of $\log|z|$ on $z\ne0$, equal to one for
$\log|z|\leq-\lambda$ and zero for
$\log|z|\geq-\lambda/2$, and gives it value one at zero.
It is original smooth because it is constant on an ordinary
neighborhood of zero. Now put
\[
 \alpha=\kappa(z)f(s)d''s\in A_c^{0,1}(U).
\]
Its support is compact: it is contained in the product of a
closed smaller $z$-disc and a closed smaller $t$-annulus.
Pullback to $Y$ gives
\[
 d'\alpha|_Y=f'(s)d's\wedge d''s.
\]
The original calibration of the identity tropicalization of
$\mathbf G_m$ has multiplicity one. Hence the defining
CLD integral, equivalently its one-dimensional Lebesgue
formula, gives
\begin{equation}
\label{eq:r3-annular-boundary-current}
 \langle T,d'\alpha\rangle
   =\int_{\{s>0\}\subset Y}f'(s)d's\wedge d''s
   =\int_0^\lambda f'(s)\,ds=-1.
\end{equation}
Here $f$ vanishes near $\lambda$.
Thus $d'T\ne0$. Symmetry, or the same computation with a
$(1,0)$-test, also gives $d''T\ne0$. The nonzero boundary
functional is an original test functional; no additional
$\delta$-test is being introduced.

For completeness, the local-completeness failure can be
verified independently of the extension theorem. Let
$x\in Y$ be the unit Gauss point of the $t$-annulus.
Suppose on an ordinary neighborhood $W$ of $x$ that
$E\cap W=\{\phi=-\infty\}$ for an extended classical psh
function $\phi$. Choose a connected ordinary neighborhood
$C$ of $x$ in $Y\cap W$. Such neighborhoods exist by local
connectedness of analytic curves; the same fact is used in
\cite[proof of Proposition~3.5(3)]{GR25}.
The restriction $\phi|_C$ is extended classical psh.

The map sending $r$ to the radius-$e^r$ Gauss point is
continuous. Since $C$ is a neighborhood of $x$, it contains
points of this map with sufficiently small $r>0$, which are
outside $E$. Thus $\phi|_C$ is not identically $-\infty$.
But $x$ is a divisorial point of this smooth curve and
$\phi(x)=-\infty$, contradicting the proved
Theorem~\ref{thm:r2-no-divisorial-poles} applied on $C$.
This proves that $E$ is not locally complete pluripolar.
\end{proof}

\paragraph{Interpretation.}
The proposition is an explicit \texttt{NEW\_PROVED}
counterexample within the allowed original analytic category.
Its purpose is the necessity of the complete-pole hypothesis
for this extension statement. It is not a counterexample to
the extension theorem with all its hypotheses. The current
has already finite mass; strengthening only the mass bound
cannot repair the false version.

\begin{example}[The local mass hypothesis cannot be omitted]
\label{ex:r3-zero-extension-infinite-mass}
Let $U=\{|t|<1\}\subset\mathbf A_K^{1,\mathrm{an}}$ and
$P=\{0\}=\{\log|t|=-\infty\}$. For $j\geq1$ let
$x_j=\zeta_{0,|\pi|^j}$ be the corresponding Gauss point and set
\[
 \mu=\sum_{j=1}^{\infty}\delta_{x_j}
       \quad\hbox{on }U\setminus P.
\]
The points $x_j$ tend to zero. Every compact subset of
$U\setminus P$ has a positive minimum for $|t|$, hence meets
only finitely many of the $x_j$. Thus $\mu$ is a positive
Radon measure there, and defines an original top-degree
$(1,1)$-current; separate closedness is automatic in that
degree. The closed set $P$ is a complete pole set.

Any positive Radon extension to $U$ would give infinite mass
to the compact closed disc $\{|t|\leq|\pi|\}$, since that disc
contains every $x_j$. This is impossible. Therefore no
positive coefficient-Radon extension exists. This elementary
example has remaining degree $q=0$; it tests deletion of the
mass hypothesis from the all-bidegrees theorem and is not
presented as a new remaining-degree-one obstruction.
\end{example}


% END source: base_r3/sections/zero_extension_counterexamples.tex


% BEGIN frozen or reviewed source: base_r3/sections/potential_quotient.tex
\section{Potential quotients: normal traces and the remaining regularity gate}
\label{sec:r3-potential-quotient}

Fix the original absolute value on a finite extension $K/\mathbf Q_p$.
All currents, positive cones, and compactly supported tests in this
section are original CLD objects. The functions in $\Cb$ retain their
actual values and the all-complete-extensions classical definition.
For the normal-coordinate statements, the spaces are ordinary opens
in smooth algebraic analytifications over this same $K$.

The question is whether
\begin{equation}
 D[u]=D[v],\quad u,v\in\Cb(U)
 \quad\Longrightarrow\quad
 D\bigl((u-v)S\bigr)=0
 \label{eq:r3-general-potential-quotient}
\end{equation}
for every symmetric original-positive, separately $d'$- and
$d''$-closed current $S$ of bidegree $(p,p)$, $p<n$.
The full implication remains \texttt{OPEN}. The results below give
additional backgrounds and a precise sufficient regularity statement.
They do not replace the quantifier over arbitrary $S$ by a restricted
one in the main research question.

\subsection{Bounded classical weights through a normal Gauss kernel}

Let $B$ be an ordinary open in a smooth $d$-dimensional
$K$-variety, let $\mathbb D_s=\{|t|<s\}$, and write
\[
 W=B\times_K\mathbb D_s,\qquad
 \pi:W\longrightarrow B,\qquad i:B\longrightarrow W
\]
for the projection and zero section. For $a\in B$ and $0<r<s$,
let $\sigma_r(a)$ be the image in $W$ of the radius-$r$ Gauss
point of the $\mathscr H(a)$-disc over $a$.

Choose a smooth convex function $\psi:\mathbf R\to\mathbf R$
which equals $0$ on $(-\infty,-\epsilon]$ and equals the
identity on $[\epsilon,\infty)$, where $\epsilon>0$ is fixed.
Put
\[
 b_N(t)=-N+\psi(\log|t|+N),\qquad \beta_N=Db_N.
\]
The function is extended by the constant $-N$ at $t=0$.
For sufficiently large $N$, $\beta_N$ is a smooth strongly positive
closed form supported in
$e^{-N-\epsilon}\leq|t|\leq e^{-N+\epsilon}<s$.

\begin{lemma}[The normal trace for a bounded classical weight]
\label{lem:r3-bounded-normal-trace}
For $u\in\Cb(W)$ and every original
$\eta\in A_c^{d,d}(B)$,
\begin{equation}
 \lim_{N\to\infty}\int_W u\,\beta_N\wedge\pi^*\eta
       =\int_B (u\circ i)\eta.
 \label{eq:r3-normal-trace}
\end{equation}
The same conclusion holds for any finite real linear combination
of such $u$. All integrals use the prescribed Borel representatives.
\end{lemma}

\begin{proof}
Let $\mu_\eta$ be the signed Radon measure of the smooth top
form $\eta$ on $B$. The product calibration and Borel coarea give
\begin{equation}
 \int_W u\,\beta_N\wedge\pi^*\eta
 =\int_B\left(\int_{\mathbf R}
      u\bigl(\sigma_{e^x}(a)\bigr)\psi''(x+N)\,dx\right)
      d\mu_\eta(a).
 \label{eq:r3-normal-coarea}
\end{equation}
Here the inner integral is restricted to the support of
$\psi''(x+N)$, which lies below $\log s$.
For clarity, this is coarea for smooth forms, with a Borel scalar
coefficient, rather than an external product of arbitrary currents.
On a finite moment presentation of the base top form, keep its
calibrated skeleton cells separate. Adjoining the coordinate $t$
gives their products with the radial interval; integration on a
product cell is the base calibrated measure times $dx$. The normal
coefficient is $\psi''(x+N)$. Borel Fubini proves
\eqref{eq:r3-normal-coarea} for bounded Borel coefficients, exactly
as in the cellwise argument of \textup{B}, Lemma A.3. A finite
partition and signed linearity recover every $\eta$. Any structural
degree of the base is already in $\mu_\eta$; the normal factor has
mass one, since $\int\psi''(x+N)\,dx=1$.
The Gauss lift is continuous in $(a,r)$ on an affinoid base patch:
for a polynomial in $t$ its value is
$|\sum_i f_it^i|(\sigma_r(a))=\max_i|f_i(a)|r^i$,
and uniform approximation gives the assertion for analytic
functions on a compact radial range. Thus the Borel integrand
in \eqref{eq:r3-normal-coarea} is evaluated on actual analytic
points and is measurable.

Fix $a\in B$. Pull $u$ back to the $\mathscr H(a)$-disc over
$a$. This is a locally bounded subharmonic function by the
all-extensions classical hypothesis. Its maximum on a closed
radius-$r$ disc occurs at the Gauss point, so
\[
 u(i(a))\leq u(\sigma_r(a)).
\]
As $r\downarrow0$, $\sigma_r(a)\to i(a)$ in the ordinary
topology. Upper semicontinuity gives the opposite limiting upper
bound. Therefore
\begin{equation}
 u(\sigma_r(a))\longrightarrow u(i(a)).
 \label{eq:r3-radial-actual-limit}
\end{equation}
This argument concerns only the pullback of the given function to
a curve. It does not assume base extension of a current equality.

Choose $s_0<s$ and a compact base neighborhood of
$\operatorname{supp}\eta$. Its product with the closed
radius-$s_0$ disc is contained in a finite union of compact
analytic domains in $W$. Local boundedness of $u$ gives one
bound there. For large $N$, every normal radius in
\eqref{eq:r3-normal-coarea} is at most $s_0$.
The inner measures are probability measures supported at radii
tending uniformly to zero. Equation
\eqref{eq:r3-radial-actual-limit} and dominated convergence against
the fixed finite measure $|\mu_\eta|$ prove
\eqref{eq:r3-normal-trace}. Finite real linearity proves the
last assertion.
\end{proof}

This is a short \texttt{NEW\_PROVED} extension of the continuous
normal-kernel argument in \textup{R}, Lemma 5.1. The additional
input is the actual radial limit supplied by classical
subharmonicity; weak convergence of $\beta_N$ alone would not
justify the weighted limit.

\begin{proposition}[Restriction of a bounded difference with positive curvature]
\label{prop:r3-normal-restriction}
Let $f=u-v$, with $u,v\in\Cb(W)$. If $D[f]$ is an
original-positive current on $W$, then $D[f\circ i]$ is
an original-positive current on $B$. If $D[u]=D[v]$, then
\[
 D[u\circ i]=D[v\circ i]
\]
on every original compact smooth complementary test on $B$.

Consequently, if $j:Z\hookrightarrow X$ is a smooth closed
immersion of smooth $K$-varieties and $u,v\in\Cb(U)$ on an
ordinary $U\subset X^{\an}$ satisfy $D[u]=D[v]$, then
\begin{equation}
 D[u|_{Z^{\an}\cap U}]=D[v|_{Z^{\an}\cap U}].
 \label{eq:r3-subvariety-restriction}
\end{equation}
\end{proposition}

\begin{proof}
The assertion has no curvature content when $d=0$.
For $d\geq1$, take a positive original
$\alpha\in A_c^{d-1,d-1}(B)$. The form
$\beta_N\wedge\pi^*\alpha$ is positive, smooth, and compactly
supported in $W$. Separate closedness of $\beta_N$ and
distributional integration by parts give
\[
 0\leq\langle D[f],\beta_N\wedge\pi^*\alpha\rangle
   =\int_W f\,\beta_N\wedge\pi^*(D\alpha).
\]
The normal trace lemma, applied to $u$ and $v$ separately, gives
the limit $\langle D[f\circ i],\alpha\rangle$.
This proves positivity. In the equality case the same computation
starts with zero and applies to every original test $\alpha$,
without a positivity restriction; it proves the full current equality.

By \textup{R}, Lemma 5.2, each point of the smooth immersion
has one ordinary analytic product neighborhood with a normal
polydisc over an ordinary neighborhood in $Z^{\an}$.
Apply the zero-section result successively to its normal
coordinates. At every step the restricted functions remain actual
locally bounded classical psh functions by their definition.
Locality gives \eqref{eq:r3-subvariety-restriction}.
No equality on tests after an arbitrary extension of $K$ has been
used or asserted.
\end{proof}

\begin{corollary}[Backgrounds carried by a smooth subvariety]
\label{cor:r3-subvariety-potential-quotient}
In the last setting, put $h=u-v$. Then
\[
 D(h[Z^{\an}\cap U])=0.
\]
More generally the same conclusion holds for $S=j_*Q$ when,
locally on $Z^{\an}\cap U$, $Q$ is a finite linear combination
of bounded-potential ambient BT products on $Z$, or is a smooth
separately closed form current. The pushforward is taken only
in the degrees where it is defined. In particular this includes
every positive closed background of either displayed kind.
\end{corollary}

\begin{proof}
Borel multiplication and the proper local pushforward satisfy
$h\,j_*Q=j_*((h|_Z)Q)$ on all original tests, with the
original calibration. The restricted curvature equality is
\eqref{eq:r3-subvariety-restriction}. For ambient bounded-potential
products on $Z$, the symmetry argument of
Lemma~\ref{lem:potential-independence-ambient-mixed} puts that
curvature in the innermost differentiated slot; for a smooth
closed form, use the distributional Leibniz rule. Thus
$D((h|_Z)Q)=0$. Pushforward commutes with $D$, proving the claim.
The first assertion is the case $Q=[1]$.
\end{proof}

The corollary is a \texttt{DIRECT\_COROLLARY} of the normal trace
and the previously proved restricted quotient statements. The
relative background $[Z]$ is allowed to charge the ambient
pluripolar set $Z$; ambient nonconcentration has not been imposed.

\subsection{A further closure class supplied by R2}

\begin{corollary}[Bounded ambient BT weak closures]
\label{cor:r3-bounded-bt-quotient-closure}
Let $u,v\in\Cb(U)$ satisfy $D[u]=D[v]$. Suppose that locally
on each fixed norm-box domain one has
\[
 S_j=T_{[1],p}(a_{1,j},\ldots,a_{p,j})\longrightarrow S,
 \qquad 0\leq a_{i,j}\leq1,
 \qquad 0\leq p<n,
\]
on all original compact smooth tests. Then
$D((u-v)S)=0$. No monotonicity or convergence of the
generating potentials is required.
\end{corollary}

\begin{proof}
For every $j$, ambient BT symmetry gives $D(uS_j)=D(vS_j)$.
Corollary~\ref{cor:r2-fixed-slots-closure}, applied separately
to the fixed weights $u$ and $v$, gives
$uS_j\to uS$ and $vS_j\to vS$. Distributional differentiation
passes the equality to the limit. The local argument glues.
\end{proof}

This is a \texttt{DIRECT\_COROLLARY} of R2, not a regularization
theorem for arbitrary currents. The uniform bounded-potential
family provides the capacity control in that result. Neither an
arbitrary positive closed current nor a current supported on an
ambient polar set has been shown to admit such an approximation.

\subsection{Continuous representatives: a precise sufficient gate}

\begin{lemma}[A continuous current representative is the actual difference]
\label{lem:r3-continuous-representative-actual}
Let $u,v\in\Cb(U)$ and let $H$ be a finite continuous function
on $U$. If
\[
 [u]-[v]=[H]
\]
on every original compact smooth top-degree test, then
$u-v=H$ at every point of $U$.
No classical-psh hypothesis on $H$ is required.
\end{lemma}

\begin{proof}
Work in a framed affine ordinary neighborhood as in
Section~\ref{sec:r2-gap-detection}. At a prescribed divisorial
point $z$, choose its continuous model tester $a$ with
$(Da)^n(\{z\})>0$. On one smaller neighborhood, its pure-FS
construction gives smooth classical psh $a_j\downarrow a$.
For every original compact smooth scalar $\chi$,
\[
 \int \chi(u-v-H)(Da_j)^n=0.
\]
The fixed-background weighted decreasing theorem passes the
$u$ and $v$ terms to their actual Borel limits. The $H$ term
passes by continuity of the fixed multiplier and weak convergence
of $(Da_j)^n$. Hence
$(u-v-H)(Da)^n=0$ as a signed Radon measure. Its atom at $z$
gives $u(z)-v(z)=H(z)$.

This holds at every divisorial point in the Shilov determining
family of the neighborhood. The function $v+H$ is finite and
upper semicontinuous, although it has not been declared psh.
Lemma~\ref{lem:r2-shilov-gap}, applied to the classical function
$u$ and the upper-semicontinuous function $v+H$, gives
$u\leq v+H$ everywhere. Apply the same lemma to $v$ and
$u-H$ for the reverse inequality. Locality finishes the proof.
\end{proof}

The lemma is a \texttt{DIRECT\_COROLLARY} of weighted convergence,
the divisorial testers, and Shilov determination. It is a recovery
statement \emph{after} a continuous representative has been supplied.

\begin{proposition}[A sufficient local regularity statement]
\label{prop:r3-quotient-regularity-gate}
Assume the following statement on the permitted analytic spaces:
\begin{quote}
For $u,v\in\Cb(U)$ with $D[u]=D[v]$, every point has an
ordinary neighborhood $W$ and a finite continuous $H_W$ such that
$[u]-[v]=[H_W]$ on $W$.
\end{quote}
Then \eqref{eq:r3-general-potential-quotient} holds for every
allowed positive separately closed background $S$.
\end{proposition}

\begin{proof}
Lemma~\ref{lem:r3-continuous-representative-actual} gives
$u-v=H_W$ at the actual points of $W$. Distributional
differentiation gives $D[H_W]=0$. The continuous-harmonic
multiplier lemma from R1, based on \textup{R}, Theorem 2.9,
then gives $D(H_WS)=0$ on $W$. Indeed $H_W$ and $-H_W$
are locally uniform limits of smooth functions with positive
curvature forms; multiplying those forms by $S$, passing the
uniform limit through its coefficient-Radon measures, and using
the pointed positive cone gives both signs of the required zero
current. These local equalities glue.
\end{proof}

The proposition is \texttt{CONDITIONAL}; the displayed existence
of a continuous representative remains \texttt{OPEN}. This is
strictly a sufficient reduction, not a claimed equivalence.
The known faithfulness of $u\mapsto[u]$ does not construct
$H_W$. Likewise $u-v$ is not a psh slot merely because both
$u$ and $v$ are psh.

\begin{remark}[The external harmonicity statements have different scopes]
In CLD's 2025 revision, Section 7.8 begins with a boundaryless
\emph{tropical space} of level $n$. Definition 7.8.2 defines its
harmonic sheaf using locally real linear combinations of
paralinear harmonic functions. Remark 7.8.3 discusses both the
possible gap between that sheaf and continuous functions with
$D[u]=0$, and the unresolved continuity, up to current-equivalence,
of locally integrable functions with $D[u]=0$ in that tropical
setting. It is not an established theorem about the present
bounded differences of classical psh functions over $K$, and it
is not used as a proof that our specific gate is equivalent to
that broader problem.

Similarly, GR, Definition 7.1 and Proposition 7.2, make a function
pluriharmonic when it and its negative are classically psh; its
continuity then follows from the two upper-semicontinuity
conditions. The current equality $D[u]=D[v]$ has not been shown
to put either sign of $u-v$ in that class. These are source
comparisons, not substitute regularity proofs.
\end{remark}

\subsection{A false extension to arbitrary ambient-capacitary representatives}

\begin{example}[Ambient current zero does not imply a relative multiplier zero]
\label{ex:r3-polar-representative-quotient}
Let $R>1$ and
\[
 U=\{|z|<R,\ |t|<R\}\subset(\mathbf A^2_K)^{\an},
 \qquad Y=\{z=0\}\cap U,
 \qquad q(t)=\max(0,\log|t|).
\]
Define the actual Borel function
\[
 f(x)=\begin{cases}q(t(x)),&x\in Y,\\0,&x\notin Y.\end{cases}
\]
It is bounded and upper semicontinuous: $Y$ is closed, $q$ is
continuous and nonnegative, and the outside value is zero.
It is locally ambient-capacity-quasicontinuous, since it vanishes
off the analytic polar set $Y$, which has local outer ambient
capacity zero. It represents the zero class in the ambient
capacitary completion.

Every original smooth top-degree trace measure on $U$ gives
zero mass to $Y$. Hence $[f]=0$ and $D[f]=0$. Nevertheless,
with the original positive separately closed background $S=[Y]$,
\[
 D(fS)=i_*D[q]=\delta_{(0,\zeta_{0,1})}\ne0,
\]
where $\zeta_{0,1}$ is the unit Gauss point of the $t$-disc.
The atom has mass one in the original $K$ calibration, by the
one-variable model formula for $\max(0,\log|t|)$.

This does not refute \eqref{eq:r3-general-potential-quotient}:
$f$ cannot be a difference $a-b$ of two members of $\Cb(U)$.
Otherwise $[a]=[b]$ and R2 current-faithfulness would force
$a=b$ at every point, whereas $f$ is nonzero on part of $Y$.
The example instead proves \texttt{FALSE} the shortcut which
replaces the actual difference hypothesis by boundedness,
upper semicontinuity, and ambient quasicontinuity. Its ingredients
are direct consequences of the already proved polar and model
measure facts; it is not counted as a new BT construction.
\end{example}

\subsection{What the normal trace does not supply}

The normal-kernel proof gives restriction along a smooth immersion
over the original field. It does not establish that an equality
of original currents pulls back to every complete field extension,
or to an arbitrary analytic curve map after such an extension.
Nor does it approximate an arbitrary positive closed current by
smooth-subvariety integration currents with multiplier control.
The bounded-difference restriction statement and the continuous
representative gate are therefore kept separate. The unresolved
step in \eqref{eq:r3-general-potential-quotient} is not hidden in
an unproved change of test category or in an exchange of two
discontinuous limits.

% END source: base_r3/sections/potential_quotient.tex


% BEGIN frozen or reviewed source: base_r3/sections/potential_quotient_projection.tex
\section{The projection route for potential quotients: a precise test gate}
\label{sec:r3-potential-quotient-projection}

This section supplements Section~\ref{sec:r3-potential-quotient}.
It does not resolve its unrestricted potential-independence question.
All spaces, analytic products, and absolute values are over the original
finite extension $K/\mathbf Q_p$. All tests are original CLD smooth tests.
Let $U$ be an allowed ordinary smooth $n$-dimensional algebraic
analytic neighborhood, $n\geq1$, let $\mathbb D_s$ be an ordinary
disc, and put $W=U\times_K\mathbb D_s$, with projection $\pi$.
Suppose $u,v\in\Cb(U)$ and $D[u]=D[v]$, and set $f=u-v$.
The functions $u\circ\pi,v\circ\pi$ are classical by the
all-extensions definition. Pulling back the equality of currents is
a separate question.

\begin{proposition}[A sufficient projected-test criterion]
\label{prop:r3-projection-continuous-coefficients}
For $\alpha\in A_c^{n,n}(W)$ define the compactly supported
ordinary current $P_\alpha=\pi_*[\alpha]$ of bidegree $(n-1,n-1)$.
The pushforward is defined because the projection is proper on the
support of $\alpha$. The current
\[
 \nu_\alpha=DP_\alpha=\pi_*[D\alpha]
\]
is a finite signed Radon measure, and
\begin{equation}\label{eq:r3-projection-exact-reduction}
 \langle D[f\circ\pi],\alpha\rangle
       =\int_U f\,d\nu_\alpha.
\end{equation}
Suppose, as a compact current on $U$, that
\[
 P_\alpha=\sum_{\ell=1}^N c_\ell[\eta_\ell],\qquad
 \eta_\ell\in A_c^{n-1,n-1}(U),
\]
where the $c_\ell$ are continuous and bounded on the relevant
compact supports. Then $\langle D[f\circ\pi],\alpha\rangle=0$.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
The conclusion for every $\alpha$ is \texttt{CONDITIONAL} on the
displayed representation, or an equally strong pairing-continuity
statement, for every projected test.

\begin{proof}
The first identity is the definition of $D$ on the locally bounded
Borel function $f\circ\pi$, followed by pushforward of the fixed
signed measure $D\alpha$. It uses no pullback of nonsmooth currents.
Use a finite original smooth partition on the compact support to
reduce to a neighborhood on which the accepted regularization supplies
smooth psh $u_j\downarrow u$ and $v_j\downarrow v$, with uniform
local bounds. Write $f_j=u_j-v_j$.
Since $Du_j$ and $Dv_j$ converge weakly to the same positive current,
the fixed-continuous-multiplier theorem of R gives
\[
 \sum_\ell\int_U c_\ell Df_j\wedge\eta_\ell\longrightarrow0.
\]
For the smooth functions $f_j$, ordinary smooth pullback and Stokes
identify this sum with
\[
 \langle P_\alpha,Df_j\rangle
      =\int_U f_j\,d\nu_\alpha.
\]
Bounded convergence against the one fixed finite signed measure
$\nu_\alpha$ gives the asserted zero limit. If a representation is
initially stated only near the support, multiply it by a compact
cutoff equal to one there before using it as an identity of compact
currents. The finite partition reduction preserves continuity of the
coefficients and the degrees of the test forms.
\end{proof}

\begin{lemma}[Separated radial tests]
\label{lem:r3-radial-separated-kernel}
Let $I\Subset(-\infty,\log s)$, $x=\log|t|$, and let
$\gamma$ be a compactly supported Lagerberg form of bidegree
$(r,s')$ on $I$, where $r,s'\in\{0,1\}$. If
$\eta\in A_c^{n-r,n-s'}(U)$, then the form
\[
 \alpha=\pi^*\eta\wedge\gamma(x),
\]
extended by zero off the annulus, satisfies
$\langle D[f\circ\pi],\alpha\rangle=0$.
The same holds for finite sums, limits in the original compact test
topology, and analytic changes of centre $t'=t-a(x)$ for which
$a$ is analytic and $|a|<s$ on a neighborhood of the base support.
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
Use separate moment coordinates on the base and the radial annulus.
On each calibrated product cell the integration formula is iterated
integration in the base coordinates and the radial coordinate.
Consequently it is valid also for the bounded Borel multiplier pulled
back from $f$: first approximate a bounded Borel function by simple
functions in the resulting fixed coefficient measures.

If $r=s'=1$, the only contribution is the base pairing
$\langle D[f],\eta\rangle$ times the radial integral of $\gamma$,
and is zero. If $r=s'=0$, the only possible contribution contains
$D\gamma$ and its integral is zero by one-dimensional Stokes. In
the remaining two bidegrees the possible cross term has radial
factor $d'\gamma$ or $d''\gamma$, again with zero integral.
Degree considerations eliminate the other terms. This proves the
claim without differentiating $f$. Finite sums and convergence in
the test topology preserve the equality by the definition of a
current. Under the stated bound on $a$, translation is an analytic
automorphism of the disc over the relevant base neighborhood and
preserves $\pi$; applying the same computation in $t'$ proves the
last assertion. No assertion that these tests generate all original
tests on the analytic product is made.
\end{proof}

\begin{example}[The analytic product is not the topological product]
\label{ex:r3-analytic-product-not-topological-product}
The assertion that products of functions of the two factors are
dense in all original smooth functions on the Berkovich analytic
product is \texttt{FALSE}.
Choose a uniformizer $\varpi$, and work in an ordinary bidisc of
radius $R>1$ with coordinates $(x,t)$. Let $z_0$ be the two-variable
unit Gauss point. Let $z_1$ be the image of the two-variable unit
Gauss point in $(X,S)$ under
\[
 x=X,\qquad t=X+\varpi S.
\]
Both points project to the unit Gauss point in each factor. Indeed,
for a polynomial $P$, the Gauss norm of $P(X+\varpi S)$ is at most
the maximum coefficient norm of $P$, and its coefficient of $S^0$
is $P(X)$, which gives the opposite inequality. However,
\[
 |t-x|(z_0)=1,\qquad |t-x|(z_1)=|\varpi|.
\]
Thus the original smooth function $\log|t-x|$ on the ordinary
nonvanishing locus of $t-x$ separates the two points. Multiplying by an
original compact cutoff equal to one near both gives a compact
smooth separating function. Every finite sum of products of
functions from the two factors, and every uniform limit of such
sums on the compact set $\{z_0,z_1\}$, takes equal values at the
two points. Equivalently, the nonzero top-degree current
$\delta_{z_0}-\delta_{z_1}$ annihilates every such tensor test.
\end{example}

The example disproves this density shortcut, not the desired
projection-kernel implication. It also does not contradict the
translated radial tests, which can involve $t-x$.

\paragraph{What the published projection results supply.}
CLD's 2025 manuscript, Section 9.1, constructs the product of its
topological tropical spaces. Its factor-generated refinements in
Section 9.2 (in particular Corollary 9.2.4), and the tensor-current
construction and differential identity in 9.4.4 and 9.4.8, have
that product as their domain. The preceding example prevents their
factor-density conclusion from being substituted for the missing
assertion on the analytic fiber product.

Mihatsch, \emph{$\delta$-Forms on Lubin--Tate spaces},
arXiv:2112.10018v4, Theorem 3.23, printed page 58, provides flat
refined skeletons for locally dominant analytic morphisms.
Definition 2.18, printed page 22, is formulated with piecewise
smooth fiber integrals. Example 2.45(3), printed pages 41--42,
also illustrates that a balanced pushforward need not be a
$\delta$-form. It is an abstract sheaf-level warning, not a
counterexample for the present disc projection. These results do
not state the continuous-coefficient original-test representation
in Proposition~\ref{prop:r3-projection-continuous-coefficients}.
Gubler--K\"unnemann, Section 6.10, concerns proper pushforward of
$\delta$-currents and likewise does not prove that representation.

\noindent\textbf{Status: \texttt{FAILED\_ATTEMPT}.}
The inference ``flat after refinement, hence an admissible projected
original smooth test'' is not justified. The missing statement is
pairing continuity for the projected test through cell changes and
refinements, in exactly the original sheaf.

\paragraph{The conditional graph consequence.}
Suppose the projection-kernel implication has been established for
iterated disc products. Let $C$ be an ordinary open in a smooth
algebraic $K$-curve, with a closed immersion
$\iota:C\hookrightarrow\mathbb D^m$, and let $g:C\to U$ be an
analytic map. Assume explicitly that the graph $(g,\iota)$ is
locally in the smooth algebraic closed-immersion scope of the
already proved normal-trace restriction theorem; for example,
both maps may be induced by algebraic maps with an algebraic
closed embedding for $\iota$. This hypothesis is stronger than
merely requiring an arbitrary analytic closed immersion.
Its graph is then a permitted smooth closed immersion into
$U\times_K\mathbb D^m$.
Applying that normal-trace theorem to the pulled-back kernel
identity would give $D[f\circ g]=0$. This is a
\texttt{CONDITIONAL} same-$K$ consequence. It neither supplies
the kernel identity nor changes the ground field; it therefore
does not by itself verify the all-complete-extensions curve test
in Gubler--Rabinoff, Definition 4.1.

\begin{proposition}[The positive square of a kernel difference]
\label{prop:r3-kernel-positive-square}
Under $D[u]=D[v]$, the established bounded energy calculus gives
\begin{equation}\label{eq:r3-kernel-positive-square}
 \Gamma_{[1]}(f)
       =\tfrac12D[f^2]-fD[f]
       =\tfrac12D[f^2]\geq0.
\end{equation}
For every fixed smooth positive closed $(1,1)$-form $\beta$ and
every nonnegative compact smooth $\chi$ on its domain,
\[
 0\leq\langle\Gamma_{[1]}(f),\chi\beta^{n-1}\rangle
  =\tfrac12\int f^2D\chi\wedge\beta^{n-1}
  \leq\tfrac12\|f\|_{\infty,\operatorname{supp}D\chi}^2
                 \int|D\chi\wedge\beta^{n-1}|.
\]
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
On a regularization neighborhood use smooth decreasing psh rows
$u_j,v_j$ and $f_j=u_j-v_j$. The smooth currents
$d'f_j\wedge d''f_j$ are positive. Expand $f_jDf_j$ into its
four weighted psh terms; the established fixed-background weighted
monotone theorem gives the limit $fD[f]$. Bounded convergence gives
$D[f_j^2]\to D[f^2]$. The smooth square identity thus gives the
first formula and its positivity, which glue by locality. Pairing
with $\chi\beta^{n-1}$ and using that $\beta$ is separately closed
gives the second formula and the displayed elementary bound.
\end{proof}

This estimate is not zero energy. On a relatively compact annulus,
$f=\log|t|$, $u=f+C$ and $v=C$ are bounded classical potentials
and $D[f]=0$, whereas
$\Gamma_{[1]}(f)=d'\log|t|\wedge d''\log|t|$ has positive radial
integral. Nor is the Borel function $f^2$ shown continuous by the
positive-square identity. Consequently R's continuous-potential
input cannot be applied to it without a separate regularity proof.
CLD 2025, Remark 7.8.3, is in its enlarged tropical test category;
Gubler--Rabinoff 2025, Definition 7.1 and Proposition 7.2, concern
functions whose two signs satisfy the classical psh conditions.
Neither statement turns the original current equation $D[f]=0$
alone into that hypothesis.

\paragraph{Remaining gate.}
The unrestricted implication
$D[u]=D[v]\Rightarrow D((u-v)S)=0$ for an arbitrary allowed
positive closed background remains \texttt{OPEN}. The proved
projected-test criterion, the radial cases, and the positive-square
identity are retained; neither the tensor-density shortcut nor a
formal passage to another test sheaf closes the gap.

% END source: base_r3/sections/potential_quotient_projection.tex


% BEGIN frozen or reviewed source: base_r3/sections/relative_compact_capacity.tex
\section{Relative compact regularity: a linear-degree reduction and a contact obstruction}
\label{sec:r3-relative-compact-capacity}

The field, actual classical class $\Cb$, original CLD tests, and
original-positive cone remain those of the first two records. The
background is fixed throughout. No equality on additional
$\delta$-tests is asserted. The target of this section is the sharp
inequality
\[
 \nu(K)\leq\operatorname{Cap}_S(K;\Omega)
 \tag{RC}\label{eq:r3-sharp-relative-closure}
\]
for weak limits of normalized top relative BT measures on one fixed
domain. We do not prove \eqref{eq:r3-sharp-relative-closure} for an
arbitrary $S$. We prove a representation and an exact reduction when
the remaining degree is one, and give counterexamples to a proposed
contact-measure proof in every intermediate degree.

\paragraph{External comparison.}
For complex positive closed currents, Dabbek--Elkhadhra define the
analogous capacity in \cite[Definition~2.1, p.~471]{DE06}. Their list
of open problems on p.~472 explicitly asks for continuity on
decreasing compact sets, equality with a contact-constrained
capacity, and attainment of the capacity of a compact set by one
potential. This is an \texttt{EXTERNAL\_KNOWN} statement about that
paper; it is not a claim that all these questions remain open in the
present literature. In particular, that paper cannot be cited as
already proving the compact outer regularity sought here.

\subsection{An energy estimate for one remaining degree}

For this subsection let $S$ have bidegree $(n-1,n-1)$. It is
arbitrary within the fixed original-positive, symmetric, separately
closed class. Thus
\[
 C_{\Omega'}(E)=\operatorname{Cap}_S(E;\Omega')
   =\sup_{\substack{w\in\Cb(\Omega')\\0\leq w\leq1}}
        D(wS)(E).
\]
The relative energy measure $\Gamma_S(f,g)$ for differences and
real linear combinations of bounded classical potentials is the one
in Lemma~\ref{lem:relative-energy-measures}.
For an original smooth real scalar $\chi$ we also use
\[
 \Gamma_S(\chi)=d'\chi\wedge d''\chi\wedge S.
\]
This is a positive measure: the $(1,1)$-factor is strongly positive,
so pairing it with the specified original-positive $S$ is legitimate.
The cutoff is not being assumed psh.

Choose permitted local stages
\[
 V\Subset\Omega'\Subset\Omega,
\]
with $V$ ordinary open, and original smooth cutoffs
$0\leq\chi,\kappa\leq1$ compactly supported in $\Omega'$ such
that $\chi=1$ near $\overline V$ and $\kappa=1$ near
$\operatorname{supp}\chi$. The cutoffs may be chosen with support
inside a smaller compact norm box, so prescribed-domain
regularization applies near their entire supports. Put
\begin{equation}
\label{eq:r3-cutoff-energy-norm}
 N_\chi(f)^2
    =\int\chi\,d\Gamma_S(f)
       +\int f^2\,d\Gamma_S(\chi).
\end{equation}
This is the square of a positive semidefinite quadratic seminorm
on the real span of bounded classical potentials. The associated
bilinear form is obtained by polarization.

\begin{lemma}[A fixed-energy bound for relative capacity]
\label{lem:r3-linear-energy-capacity}
For the preceding stage there is a finite constant $A_{\chi,\kappa,S}$
such that, whenever $a,b\in\Cb(\Omega)$ satisfy $0\leq a,b\leq1$,
$f=a-b$, and $\delta>0$, one has
\begin{align}
 \int\chi f^2\,D(wS)
   &\leq A_{\chi,\kappa,S} N_\chi(f)
       &&(w\in\Cb(\Omega'),\ 0\leq w\leq1),
       \label{eq:r3-energy-square-bound}\\
 C_{\Omega'}^*(V\cap\{|f|>\delta\})
   &\leq A_{\chi,\kappa,S}\delta^{-2}N_\chi(f).
       \label{eq:r3-energy-capacity-bound}
\end{align}
The same bound holds with any compact subset of $V$ in place of
$V$. The constant is independent of $a,b,w$ and of all
regularization indices.
\end{lemma}

\begin{proof}
First take smooth $a,b,w$. For compactly supported expressions,
Stokes and symmetry of $S$ give
\begin{equation}
\label{eq:r3-square-green}
 \int\chi f^2D(wS)
    =-2\int\chi f\,d\Gamma_S(f,w)
       -\int f^2\,d\Gamma_S(\chi,w).
\end{equation}
There is no boundary term. The positive $2\times2$ matrices of
energy measures give their measure-valued Cauchy--Schwarz
inequality, also after bounded Borel multiplication. Since
$\chi\leq\kappa$ and $\kappa=1$ near the support of
$d\chi$, it follows that
\begin{align*}
 \left|\int\chi f\,d\Gamma_S(f,w)\right|
   &\leq \|f\|_\infty
       \left(\int\chi\,d\Gamma_S(f)\right)^{1/2}
       \left(\int\kappa\,d\Gamma_S(w)\right)^{1/2},\\
 \left|\int f^2\,d\Gamma_S(\chi,w)\right|
   &\leq \|f\|_\infty
       \left(\int f^2\,d\Gamma_S(\chi)\right)^{1/2}
       \left(\int\kappa\,d\Gamma_S(w)\right)^{1/2}.
\end{align*}
All norms in this proof can be taken on the fixed compact
neighborhood of the cutoff supports.

The energy of a normalized nonnegative tester is uniformly bounded.
Indeed, the defining identity for $\Gamma$ gives
\[
 \int\kappa\,d\Gamma_S(w)
  =\frac12\langle S,w^2D\kappa\rangle
       -\int\kappa w\,D(wS).
\]
The last integral is nonnegative. The first is bounded in absolute
value by the variation of the fixed coefficient measure
$\mu_{S,D\kappa}$ times $\|w\|_\infty^2/2$.
This measure is finite on the compact stage by the accepted
coefficient-Radon property. In particular one finite bound
$M_{\kappa,S}$ works for all $0\leq w\leq2$ near the stage.
Combining the preceding inequalities yields, in the smooth case,
\[
 \int\chi f^2D(wS)
 \leq \sqrt{5M_{\kappa,S}}\,\|f\|_\infty N_\chi(f).
\]

For general bounded classical $a,b,w$, use the simultaneous smooth
decreasing rows of Lemma~\ref{lem:compact-box-regularization} on
one neighborhood of the whole compact stage. After discarding a
prefix, each row lies in $[0,2]$ there. To see the upper bound,
the continuous approximants decrease to a function at most one;
their open sublevel sets below two form an increasing cover of
the compact stage. A finite subcover gives the uniform index.
The smooth differences therefore have absolute value at most two.

Proposition~\ref{prop:functional-weights}, applied to
$F(x,y)=(x-y)^2$, passes the left side to
$\int\chi f^2D(wS)$. The first summand of
$N_\chi(f)^2$ converges by the distributional formula for
$\Gamma$ and the weighted fixed-background theorem. The second
converges by bounded convergence against the fixed measure
$\Gamma_S(\chi)$. Passing the smooth inequality proves
\eqref{eq:r3-energy-square-bound}, for example with
$A_{\chi,\kappa,S}=2\sqrt{5M_{\kappa,S}}$.
This argument uses no continuity theorem with a varying background.

Since $\chi=1$ on $V$, Chebyshev's inequality and the supremum
over normalized $w$ give the estimate for the raw capacity of
$V\cap\{|f|>\delta\}$. Both $a$ and $b$ are relatively
quasicontinuous. That strict level set on the ordinary core $V$
is therefore quasi-open, and
Proposition~\ref{prop:r2-quasiopen-outer} identifies its raw and
outer capacities. This proves
\eqref{eq:r3-energy-capacity-bound}. Monotonicity gives the compact
version. In particular no strict level set has been silently
treated as ordinarily open.
\end{proof}

\paragraph{Status.}
Lemma~\ref{lem:r3-linear-energy-capacity} is
\texttt{NEW\_PROVED}. Its inputs are the established original-cone
energy measures, weighted monotone calculus, coefficient-Radon
control, and quasicontinuity. It needs no global strictly psh
exhaustion and no compact outer regularity.

\subsection{Convex blocks give actual capacitary limits}

The following class preserves the normalization on the original
domain; it is narrower than merely asking that an arbitrary
member of $\mathcal A_S^1(\Omega)$ take values in $[0,1]$.

\begin{definition}
\label{def:r3-normalized-completion}
Let $\Omega$ be a permitted ordinary algebraic norm-box domain.
Write $\mathcal A_S^{[0,1]}(\Omega)$ for the Borel functions
$a:\Omega\to[0,1]$ admitting a sequence
\[
 a_j\in\Cb(\Omega),\qquad 0\leq a_j\leq1\text{ on }\Omega,
\]
converging to $a$ locally in outer $S$-capacity in the all-domain
sense of Definition~\ref{def:r2-outer-convergence}. Actual Borel
representatives are retained. Its BT calculus is the restriction
of Theorem~\ref{thm:r2-completed-calculus}.
\end{definition}

\begin{theorem}[Linear-degree convex-block compactness]
\label{thm:r3-linear-convex-blocks}
Let $S$ have bidegree $(n-1,n-1)$ on the fixed norm-box domain
$\Omega$. Given any sequence $u_j\in\Cb(\Omega)$ with
$0\leq u_j\leq1$, there exist a subsequence $(u_{j_l})$, finite
convex combinations
\[
 v_m\in\operatorname{conv}\{u_{j_l}:l\geq m\},
\]
and $a\in\mathcal A_S^{[0,1]}(\Omega)$ such that
$v_m\to a$ locally in outer $S$-capacity on every permitted
compact/domain pair in $\Omega$.

If in addition $D(u_jS)\to\nu$ locally weakly on all original
compact smooth scalar tests, then
\begin{equation}
\label{eq:r3-linear-weak-representation}
 \nu=D(aS),\qquad
 D(v_mS)\longrightarrow D(aS),\qquad
 v_mD(v_mS)\longrightarrow aD(aS).
\end{equation}
The last convergence is for the convex blocks; it is not a
weighted convergence assertion about the original sequence.
\end{theorem}

\begin{proof}
Choose the countable cofinal pairs
$V_l\Subset\Omega_l\Subset\Omega$ used in
Proposition~\ref{prop:r2-completion}, and cutoffs
$\chi_l,\kappa_l$ as in the preceding lemma. Let $\mathscr V$
be the real span of all normalized classical potentials on
$\Omega$. For each $l$, write $N_l=N_{\chi_l}$. Positivity of
$\Gamma$ and the elementary $L^2$ term in
\eqref{eq:r3-cutoff-energy-norm} show that $N_l$ is a Hilbert
seminorm on $\mathscr V$.

There is a finite $B_l$ such that
\[
 N_l(u)^2\leq B_l
 \quad(u\in\Cb(\Omega),\ 0\leq u\leq1).
\]
For the first summand use the cutoff energy identity from the
preceding proof with $\chi_l$ in place of $\kappa$; for the
second use $u^2\leq1$ and the finite mass of
$\Gamma_S(\chi_l)$. Define
\begin{equation}
\label{eq:r3-total-hilbert-norm}
 \|f\|_{\mathcal H}^2
   =\sum_{l=1}^{\infty}
          \frac{2^{-l}}{1+B_l}N_l(f)^2.
\end{equation}
This is finite on $\mathscr V$: for
$f=\sum_{i=1}^r c_i u_i$, the seminorm triangle inequality
gives $N_l(f)^2\leq(\sum_i|c_i|)^2 B_l$.
Quotient by its null space and complete to obtain a real Hilbert
space $\mathcal H$. Each normalized $u_j$ has norm at most one.
This construction makes no claim that a Hilbert vector is
already a pointwise function or that a separately chosen
Dirichlet form is closable in a trace $L^2$ space.

The closed span of the sequence in $\mathcal H$ is separable.
Weak sequential compactness of its unit ball gives a subsequence
$u_{j_l}$ converging weakly to $h\in\mathcal H$. For every
$m$, $h$ belongs to the norm closure of the convex hull of the
tail $\{u_{j_l}:l\geq m\}$. Here is the needed elementary
Hilbert argument. If $h$ were outside that closed convex set and
$p$ its nearest point, then
\[
 \langle h-p,v-p\rangle\leq0
 \quad\text{for every }v\text{ in the set}.
\]
Passing to the weak limit along its tail would give
$\|h-p\|^2\leq0$, a contradiction.
Consequently choose finite convex combinations $v_m$ as stated
with $\|v_m-h\|_{\mathcal H}<2^{-m}$.

The $v_m$ are actual classical potentials on the whole original
$\Omega$, still in $[0,1]$. For each fixed $l$,
\[
 N_l(v_m-v_r)\longrightarrow0\quad(m,r\to\infty).
\]
Lemma~\ref{lem:r3-linear-energy-capacity} makes the sequence
Cauchy in outer capacity on $V_l$ with capacity domain
$\Omega_l$. For any permitted $H\Subset\Omega'\subset\Omega$,
finitely many such $V_l$ cover $H$, with $\Omega_l\subset\Omega'$.
Finite subadditivity and domain monotonicity therefore give the
Cauchy property on this exact pair $(H,\Omega')$. Thus it holds
on all permitted pairs, not only for one larger capacity domain.

Proposition~\ref{prop:r2-completion} now supplies an actual Borel
limit locally in outer capacity. Its construction takes a
pointwise limsup of a subsequence of the $v_m$, so the resulting
representative has values in $[0,1]$ at every point. It is in
$\mathcal A_S^{[0,1]}(\Omega)$ by the very same normalized
approximants. This step supplies the pointwise representative
which the abstract Hilbert argument alone would not supply.

Suppose finally that $D(u_jS)\to\nu$. Linearity in one
differentiated slot makes $D(v_mS)$ the corresponding finite
convex combination of tail measures. For each fixed original
test its scalar pairings converge to the pairing with $\nu$.
The completed calculus instead identifies their limit as
$D(aS)$, hence $\nu=D(aS)$. Its weighted capacity-continuity
theorem also gives the last limit in
\eqref{eq:r3-linear-weak-representation}. All conclusions use
the fixed $S$ and all original tests.
\end{proof}

\paragraph{Status.}
Theorem~\ref{thm:r3-linear-convex-blocks} is
\texttt{NEW\_PROVED}. It holds in every ambient dimension, but
its remaining degree is specifically one. Convex combinations
are essential: the theorem does not contradict the normalized
family without a capacity-Cauchy subsequence in
Proposition~\ref{prop:r2-capacity-noncompactness}.

\subsection{What the representation does and does not settle}

Define, for Borel $E\subset\Omega$,
\[
 C_{\mathcal A}(E;\Omega)
   =\sup_{a\in\mathcal A_S^{[0,1]}(\Omega)}D(aS)(E).
\]
There is no enlargement of the capacity domain in this definition.

\begin{corollary}[The outer capacity on compact sets is the completed capacity]
\label{cor:r3-linear-completed-capacity}
In the degree-one setting of
Theorem~\ref{thm:r3-linear-convex-blocks},
\begin{align}
 C_\Omega(E)&\leq C_{\mathcal A}(E;\Omega)
                   \leq C_\Omega^*(E)
                  &&(E\text{ Borel}),\label{eq:r3-completed-cap-sandwich}\\
 C_{\mathcal A}(K;\Omega)&=C_\Omega^*(K)
                  &&(K\Subset\Omega\text{ compact}).
                  \label{eq:r3-completed-compact-cap}
\end{align}
Consequently raw compact outer regularity for the original
classical capacity is equivalent here to the no-gain assertion
\begin{equation}
\label{eq:r3-completion-no-gain}
 \sup_{a\in\mathcal A_S^{[0,1]}(\Omega)}D(aS)(K)
       \leq
 \sup_{\substack{u\in\Cb(\Omega)\\0\leq u\leq1}}D(uS)(K)
 \quad(K\Subset\Omega).
\end{equation}
This last assertion is not proved for an arbitrary $S$.
\end{corollary}

\begin{proof}
The first inequality of \eqref{eq:r3-completed-cap-sandwich}
follows from constant approximating sequences. If $a$ is in
the normalized completion, choose its normalized approximants
$a_j$ on $\Omega$. The completed calculus and local mass bounds
give local weak convergence of $D(a_jS)$ to $D(aS)$, including
against compact continuous scalar tests. For every ordinary
open $G\subset\Omega$, the open-set inequality implies
\[
 D(aS)(G)\leq\liminf_jD(a_jS)(G)\leq C_\Omega(G).
\]
Taking open neighborhoods of a Borel $E$ and then taking the
supremum over $a$ proves the second inequality. This argument
does not use the unavailable compact-set inequality for weak
limits.

Fix compact $K\Subset\Omega$. Choose an ordinary neighborhood
$W$ of $K$ and a compactly supported original smooth cutoff
$0\leq\zeta\leq1$ equal to one near $\overline W$. Put
\[
 \mathcal M_\zeta
    =\{\zeta D(uS):u\in\Cb(\Omega),\ 0\leq u\leq1\}.
\]
On a compact metrizable neighborhood of the cutoff support this
is a uniformly mass-bounded family. Its capacity on subsets of
$W$ is exactly $C_\Omega$; open neighborhoods of $K$ may be
restricted to $W$. Proposition~\ref{prop:capacity-weak-closure}
therefore gives
\[
 C_\Omega^*(K)
       =\sup_{\widetilde\nu\in
                    \overline{\mathcal M_\zeta}^{\,w}}
                    \widetilde\nu(K).
\]

Take a sequence $\zeta D(u_jS)$ converging to one such
$\widetilde\nu$. Relative CLN bounds $D(u_jS)$ on every compact
subset of $\Omega$. By a countable compact exhaustion and
diagonal weak compactness, pass to a subsequence for which
$D(u_jS)$ converges locally weakly on all of $\Omega$ to a
positive Radon measure $\nu$. More explicitly, use compact
cutoffs equal to one on successive compact neighborhoods;
the finite weak limits agree on their overlapping interiors
and hence define $\nu$. The already convergent cutoff sequence
then satisfies $\widetilde\nu=\zeta\nu$.

Theorem~\ref{thm:r3-linear-convex-blocks} represents
$\nu=D(aS)$ for some
$a\in\mathcal A_S^{[0,1]}(\Omega)$. Since $\zeta=1$ on $K$,
\[
 \widetilde\nu(K)=D(aS)(K)\leq C_{\mathcal A}(K;\Omega).
\]
Take the supremum over the weak closure and combine with
\eqref{eq:r3-completed-cap-sandwich} to obtain
\eqref{eq:r3-completed-compact-cap}. Finally, the equivalence
with \eqref{eq:r3-completion-no-gain} is immediate from that
equality and the inclusion of classical potentials.
\end{proof}

\paragraph{Scope and minimal remaining gap.}
The corollary is a \texttt{DIRECT\_COROLLARY} of the new convex-block
theorem and the proved measure-family reduction. It replaces a
weak-limit question by the exact completed-class inequality
\eqref{eq:r3-completion-no-gain}. It does not replace that
inequality by a constant-factor estimate or a smaller-domain
capacity. For an arbitrary $S$, an actual completed representative
has not been shown to have a classical representative on the same
domain. Ambient upper regularization can change values on a set
charged by $S$, as the support examples of
Section~\ref{sec:r2-support-envelopes} show. The ambient
$S=[1]$ representative theorem is therefore not an input here.

\subsection{The non-pluripolar-background case closes in degree one}

The next two assertions allow every remaining degree $q=n-p\geq1$.
We say that a fixed positive closed background $S$ is
\emph{locally non-pluripolar} if the total variations of all its
coefficient-Radon measures give zero mass to every locally
pluripolar Borel set. Equivalently,
on each ordinary connected witness domain one has
$\mathbf1_P S=0$ whenever
$P=\{\phi=-\infty\}$ and $\phi$ is extended classical psh and
not identically $-\infty$. The equivalence follows by the
definition of local pluripolarity, countable local covers in
the present finite $p$-adic setting, and coefficient-Radon
multiplication. This is an explicit extra hypothesis, not a
property of all positive closed currents.

\begin{lemma}[Ambient polar sets are relatively outer-null under nonconcentration]
\label{lem:r3-nonpolar-relative-null}
Let $S$ be locally non-pluripolar. Every locally pluripolar set
has locally zero outer $\operatorname{Cap}_S$-capacity, on every
permitted compact/domain pair.
\end{lemma}

\begin{proof}
First work where $\phi\leq0$ is a nontrivial classical witness
and put $P=\{\phi=-\infty\}$. By the hypothesis and
Theorem~\ref{thm:r2-polar-restriction},
\[
 \mathbf1_P T_{S,q}(v,\ldots,v)
     =T_{\mathbf1_PS,q}(v,\ldots,v)=0
\]
for every normalized classical tester on this same domain.
Thus the raw capacity of $P$ is zero there.

It remains essential to obtain an outer-zero assertion.
The bounded increasing classical sequence
\[
 w_N=1+\frac{\max(\phi,-N)}{N}
\]
has actual limit $w=\mathbf1_{P^c}$.
Corollary~\ref{cor:r2-local-increasing} and the completion
theorem put $w$ in $\mathcal A_{S,b}$ and make it locally
relatively quasicontinuous. On an ordinary relatively compact
core $V$,
\[
 P\cap V=V\cap\{w<1/2\}
\]
is quasi-open. Proposition~\ref{prop:r2-quasiopen-outer}
therefore turns its raw zero capacity into outer zero
capacity. This argument does not assume compact outer
regularity.

Finally, fix an arbitrary permitted pair $H\Subset\Omega$.
Cover the part of the locally pluripolar set in $H$ by
countably many ordinary cores with witness domains contained
in $\Omega$. On each core the preceding argument applies.
Domain monotonicity and countable subadditivity of the outer
capacity give zero capacity on the original pair.
Subsets need not be Borel, since outer capacity is monotone.
\end{proof}

\begin{proposition}[Classical representatives for a non-pluripolar background]
\label{prop:r3-nonpolar-classical-representative}
Let $S$ be locally non-pluripolar on the fixed norm-box domain
$\Omega$, in any remaining degree $q\geq1$. For every
$a\in\mathcal A_S^{[0,1]}(\Omega)$ there is
$r\in\Cb(\Omega)$, $0\leq r\leq1$ on $\Omega$, such that
$a=r$ locally outside sets of outer $S$-capacity zero.
The representative is not asserted to be unique.

For any finite tuple of such completed potentials, any choices
of these representatives give the same weighted and unweighted
BT currents on all original compact smooth complementary tests.
In particular the normalized completed class does not enlarge
the capacity of any Borel set in this case.
\end{proposition}

\begin{proof}
Choose normalized defining approximants $a_j$ on the entire
$\Omega$. The summable-subsequence construction in
Proposition~\ref{prop:r2-completion}, now applied to their known
capacity limit $a$, gives a subsequence $a_{j_l}$ converging
pointwise to $a$ outside a locally relative-outer-null set $N$.
For clarity, on the first $l$ members of a countable cofinal
list of compact/domain pairs require the capacity of
$\{|a_{j_l}-a|>2^{-l}\}$ to be at most $2^{-l}$.
Open covers and the countable subadditivity bound on their
limsup sets give the asserted $N$ on every pair.

Set
\[
 h_m=\sup_{l\geq m}a_{j_l},\qquad r_m=h_m^*.
\]
The bounded upper-envelope theorem,
Theorem~\ref{thm:r2-local-upper-envelope}, gives
$r_m\in\Cb(\Omega)$ with $0\leq r_m\leq1$.
The sequence decreases, so
$r=\lim_m r_m$ is classical on the same domain with the same
normalization. For each $m$, the discrepancy set
\[
 E_m=\{h_m<r_m\}
\]
is locally pluripolar by
Theorem~\ref{thm:r2-negligible-outer-zero} and
Corollary~\ref{cor:r2-polar-negligible-equivalence}.
Lemma~\ref{lem:r3-nonpolar-relative-null} makes $E_m$ locally
relative-outer-null. Outside $N\cup\bigcup_mE_m$ one has
\[
 r=\lim_m h_m=\limsup_l a_{j_l}=a.
\]
This proves the asserted equality modulo local relative outer
capacity zero, without replacing any of the given Borel point
values in the original completed class.

The completed calculus is independent of precisely these
changes of representatives, including the weight slot and
the degree-zero background multiplication, by
Theorem~\ref{thm:r2-completed-calculus}. It therefore identifies
each completed product with the product of the chosen
classical representatives. In particular, if $a$ is normalized,
its measure $T_{S,q}(a,\ldots,a)$ is the measure of a normalized
classical $r$ on the original $\Omega$. Taking suprema over
Borel sets gives equality of the two capacity families.
Uniqueness is intentionally absent: for $S=0$ all functions
are relatively quasi-everywhere equal.
\end{proof}

\noindent\textbf{Status.}
The preceding lemma and representative proposition are
\texttt{DIRECT\_COROLLARY} consequences of the proved R2
polar-restriction, envelope, negligible-set, and completion
theorems. They retain the stated nonconcentration hypothesis.

\begin{theorem}[Sharp compact regularity for a non-pluripolar degree-one background]
\label{thm:r3-nonpolar-linear-compact-regularity}
Let $S$ be an arbitrary fixed locally non-pluripolar,
symmetric original-positive separately closed current of
bidegree $(n-1,n-1)$ on the permitted fixed norm-box domain
$\Omega$. Every locally weak limit of
\[
 D(u_jS),\qquad u_j\in\Cb(\Omega),\quad0\leq u_j\leq1,
\]
is $D(rS)$ for some $r\in\Cb(\Omega)$ with $0\leq r\leq1$.
Consequently, for every compact $K\Subset\Omega$,
\[
 \operatorname{Cap}_S^*(K;\Omega)
       =\operatorname{Cap}_S(K;\Omega).
\]
The raw capacity is continuous on decreasing compact subsets
of $\Omega$. Equivalently, every localized weak-closure
measure satisfies the sharp bound
\eqref{eq:r3-sharp-relative-closure} on the core where its
cutoff is one.
\end{theorem}

\begin{proof}
Theorem~\ref{thm:r3-linear-convex-blocks} represents any such
weak limit by a normalized completed potential on $\Omega$.
Proposition~\ref{prop:r3-nonpolar-classical-representative}
replaces its class by a normalized classical potential on
this same domain, preserving the current.
The compact assertion follows either from
Corollary~\ref{cor:r3-linear-completed-capacity} or directly
from Proposition~\ref{prop:capacity-weak-closure}.
That proposition also gives continuity from above on compact
sets. Its localized version is exactly the final assertion.
\end{proof}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
The conclusion holds in all ambient dimensions under the
explicit non-pluripolarity hypothesis and with remaining
degree one. The representative proposition alone does not
give weak-closure representation for $q>1$.

\subsection{Why an ambient contact measure is not a relative one}

We first give a compact-obstacle example in remaining degree one.
It applies to all classical competitors, not merely toric ones.

\begin{proposition}[An ambient compact extremal with relative mass off the obstacle]
\label{prop:r3-compact-relative-contact-false}
Let $n\geq2$, choose a uniformizer $\pi$ of the original $K$,
put $\lambda=-\log|\pi|>0$, and work on the analytic torus
$X=(\mathbf G_{m,K}^n)^{\mathrm{an}}$. Set
\[
 t_i=\log|T_i|,\qquad
 \theta_i=d't_i\wedge d''t_i,\qquad
 \beta=\sum_{i=1}^n\theta_i,
 \qquad S=[\beta^{n-1}].
\]
For $0<a<\lambda$ define the ordinary domain and its compact core
\[
 \Omega=\{\max_i|t_i|<\lambda\},\qquad
 K_a=\{\max_i|t_i|\leq a\},
\]
and the continuous classical psh function
\[
 H=\max\left\{-1,
          \frac{\max_i|t_i|-\lambda}{\lambda-a}\right\}.
\]
Then $H$ is the actual pointwise supremum of all
$v\in\Cb(\Omega)$ satisfying $v\leq0$ on $\Omega$ and
$v\leq-1$ on $K_a$. Nevertheless
\[
 D(HS)(\Omega\setminus K_a)>0.
\]
Thus the ambient compact extremal does not in general have a
relative Monge--Amp\`ere measure supported on its obstacle.
\end{proposition}

\begin{proof}
The domain is an ordinary open with compact closure in the
smooth affine torus, and $K_a$ is compact in it. If a strict
norm-box core is desired, take, for example, $a=\lambda/2$
and express its bounds using $T_i^2$ and the original value
$|\pi|$; no valuation is rescaled. Logarithms of analytic units
are harmonic after every complete extension and curve pullback.
Finite maxima of their affine combinations and constants are
classical psh. This proves that $H\in\Cb(\Omega)$, that
$-1\leq H<0$, that $H=-1$ on $K_a$, and that $H$ has ordinary
boundary value zero.

We claim $(DH)^n=0$ off $K_a$. Near any point there the constant
branch $-1$ is inactive. Among the branches $t_i$ and $-t_i$
of the maximum, at most one sign for each $i$ is active, since
the maximum is strictly positive. Delete inactive branches
on a neighborhood. If the remaining affine functions are
$\ell_1,\ldots,\ell_m$, then $m\leq n$, and
\[
 \max(\ell_1,\ldots,\ell_m)
    =\ell_m+\max(\ell_1-\ell_m,\ldots,
                         \ell_{m-1}-\ell_m,0).
\]
The differences are logarithms of analytic units. Smooth the
finite maximum by translation-invariant convex regularized
maxima. Their curvature factors through at most $m-1$ such
moment coordinates, so its $n$th exterior power vanishes
identically as an original form. The uniform continuous
fixed-background calculus passes this equality to $H$.
This proves the claim without a regularity assertion for
arbitrary psh functions on the skeleton.

Let $v$ be an admissible competitor. For every $\epsilon>0$,
the set $\{H<v-\epsilon\}$ is contained in
$\{H<-\epsilon\}$, whose closure is compact in $\Omega$.
Also $\{H<v\}$ is disjoint from $K_a$, because $H=-1$ and
$v\leq-1$ there. Thus $(DH)^n(\{H<v\})=0$.
Corollary~\ref{cor:r2-pointwise-ambient}, on this relatively
compact affine domain, gives $v\leq H$ at every point.
Since $H$ itself is admissible, it is the stated actual
supremum. This comparison includes all non-toric competitors.

To compute relative mass outside $K_a$, choose a tropical open
rectangle meeting
\[
 t_1=t_2=r,\qquad a<r<\lambda,
 \qquad |t_i|<r\ (i\geq3),
\]
small enough that the only potentially maximal branches are
$t_1,t_2$, and that their maximum is greater than $a$.
There
\[
 H=\frac{t_2-\lambda+\max(0,t_1-t_2)}{\lambda-a}.
\]
The one-variable distributional second derivative of
$\max(0,s)$ is $\delta_0$. One may equivalently obtain the
following formula by smooth convex regularization and the
original CLD tropical integration formula. For a nonnegative
original smooth tester $\psi\circ\operatorname{trop}$ supported
in that rectangle,
\[
 \int(\psi\circ\operatorname{trop})\,D(HS)
   =\frac{2(n-1)!}{\lambda-a}
     \int\psi(t_2,t_2,t_3,\ldots,t_n)
                    \,dt_2\,dt_3\cdots dt_n.
\]
Indeed, the square of the gradient of $s=t_1-t_2$ in the
diagonal form $\beta$ is two, and
$d's\wedge d''s\wedge\beta^{n-1}
 =2(n-1)!\theta_1\wedge\cdots\wedge\theta_n$.
Choose $\psi$ positive at a point of the indicated diagonal.
The displayed integral is strictly positive and is supported
outside $K_a$, proving the assertion.
\end{proof}

There is also a smooth common-collar obstruction in every
intermediate remaining degree. It directly tests the relative
energy envelope rather than only the ambient compact extremal.

\begin{proposition}[A relative energy maximum need not satisfy contact orthogonality]
\label{prop:r3-relative-energy-contact-false}
On the same torus and domain, let $1\leq q<n$ and put
\[
 S=[\beta^{n-q}],\qquad b=\sum_{i=1}^q t_i^2.
\]
Choose nonzero $\varphi\in C_c^\infty((-\lambda,\lambda)^n)$,
$\varphi\geq0$, and let $\chi=\varphi\circ\operatorname{trop}$,
$L=\operatorname{supp}\chi$, and $f=b+\chi$. Consider
\[
 \mathcal F=\{v\in\Cb(\Omega):
       b-1\leq v\leq f,\quad v=b\text{ on }\Omega\setminus L\}.
\]
The greatest actual member and the relative capacitary energy
envelope of this uniformly bounded common-collar family are $b$
(the latter modulo local relative capacity zero). However
\begin{equation}
\label{eq:r3-failed-relative-orthogonality}
 \int_\Omega(f-b)T_{S,q}(b,\ldots,b)
    =2^q q!(n-q)!\int_{\mathbf R^n}\varphi(t)\,dt>0.
\end{equation}
In particular, relative energy maximization over ambient
classical psh competitors does not imply the expected contact
orthogonality, even for a smooth strongly positive background.
\end{proposition}

\begin{proof}
The smooth function $b$ is classical psh: it is a convex smooth
function of harmonic logarithms of units, after every complete
extension and curve pullback. It is bounded on $\Omega$ and
\[
 Db=2\sum_{i=1}^q\theta_i,
 \qquad (Db)^n=0,
 \qquad T_{S,q}(b,\ldots,b)
       =2^q q!(n-q)!\theta_1\wedge\cdots\wedge\theta_n.
\]
For any $v\in\mathcal F$, the closures of all positive-margin
sets $\{b<v-\epsilon\}$ lie in the fixed compact $L$.
Pointwise ambient domination again gives $v\leq b$ on all of
$\Omega$. Since $b\in\mathcal F$, it is the greatest member.
Relative Aubin--Mabuchi energy is monotone on this common-collar
class, so $\mathcal E_b(v)\leq\mathcal E_b(b)=0$. The constant
sequence $b$ is a maximizing sequence in
Theorem~\ref{thm:r2-one-domain-envelope}, identifying its
capacitary envelope with $b$.

The original CLD integration formula for the identity tropical
chart of the torus has multiplicity one and integrates
$\theta_1\wedge\cdots\wedge\theta_n$ as Lebesgue measure
with its positive orientation. Substitution gives
\eqref{eq:r3-failed-relative-orthogonality}. Its strict positivity
uses only the chosen nonzero nonnegative smooth $\varphi$.
All products in this example are ordinary original currents.
\end{proof}

\paragraph{Status of the counterexamples.}
The two counterexamples are \texttt{NEW\_PROVED}; the contact or
orthogonality implications they disprove have status
\texttt{FALSE}. Neither is a counterexample to
\eqref{eq:r3-sharp-relative-closure}. The second concerns a
continuous obstacle and a common-collar family; it does not
identify every possible relative compact-capacity extremal.
The smooth backgrounds in these examples are locally
non-pluripolar, by the ambient smooth-trace polar-zero theorem.
Thus the degree-one contact failure coexists with the compact
outer regularity proved above; it is a failure of the proposed
proof mechanism, rather than a failure of that regularity.

\subsection{The remaining nonlinear weak-limit obstruction}

For $q>1$, applying the same square integration-by-parts argument
to
\[
 \int\chi f^2(Dw)^q\wedge S
\]
uses the energy of $f$ relative to
\[
 Q_w=(Dw)^{q-1}\wedge S.
\]
This is a positive closed background of bidegree $(n-1,n-1)$,
but it changes with the capacity tester $w$. The direct Hilbert
argument above therefore does not by itself give a fixed
seminorm controlling every tester: convergence in countably
many selected seminorms would not justify that supremum.
The independent mixed-energy recurrence in
Theorem~\ref{thm:r3-I-capacity} resolves this uniform-control
problem. Together with quantified concavity, it gives the
all-degree convex-block theorem
\ref{thm:r3-all-degree-convex-blocks}. Thus tester-dependent
energy is no longer an open obstruction in this record.

The separate nonlinear weak-limit problem remains. For $q>1$,
the MA measure of a convex combination contains mixed terms;
it is not the corresponding convex combination of the original
MA measures. The new capacity-convergent convex blocks do not
therefore identify an arbitrary weak limit of those original
measures with the MA measure of their completed limit.

The current status of arbitrary-background compact outer
regularity remains \texttt{OPEN}. The precise reusable advances
are the fixed-energy capacity estimate, normalized convex-block
representation in degree one, its exact completed-capacity
reduction, and sharp compact regularity under the additional
non-pluripolarity hypothesis in that degree. The stronger
convex-block compactness result for every remaining degree is
proved separately in Section~\ref{sec:r3-all-degree-energy}.
The contact
counterexamples rule out the ambient contact and relative-energy
orthogonality routes above.



% END source: base_r3/sections/relative_compact_capacity.tex


% BEGIN frozen or reviewed source: base_r3/sections/relative_contact_inequality.tex
% Reconstructed after the scratch reset. Original reviewed SHA256:
% e004c695696c6709a129f0d7f3bd93cffe7a96ca6f9b70813743372f21eeec43.
\section{A contact inequality for every fixed positive background}
\label{sec:r3-relative-contact-inequality}

Let $S$ be any allowed symmetric original-positive separately
closed $(p,p)$-current on an ordinary open set $U$, and put
$q=n-p\geq1$. Write $\mu_u=T_{S,q}(u,\ldots,u)$ for actual
$u\in\mathcal C_b(U)$. Restrictions below are Borel restrictions
of the coefficient-Radon top measures.

\begin{proposition}[Relative maximum and contact inequalities]
\label{prop:r3-relative-contact-inequality}
For $u,v\in\mathcal C_b(U)$,
\begin{equation}
 \mu_{\max(u,v)}
       \geq{\bf1}_{\{u\geq v\}}\mu_u
                  +{\bf1}_{\{u<v\}}\mu_v.
 \label{eq:r3-relative-max-inequality}
\end{equation}
In particular, if $u\leq h$,
\begin{equation}
 {\bf1}_{\{u=h\}}\mu_u
             \leq{\bf1}_{\{u=h\}}\mu_h.
 \label{eq:r3-relative-contact-inequality}
\end{equation}
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
proved Borel maximum locality and decreasing BT continuity.

\begin{proof}
Put $w_j=\max(u+1/j,v)$. This sequence decreases to
$w=\max(u,v)$. On the Borel set $E=\{u\geq v\}$ the strict
inequality $u+1/j>v$ holds, so Borel maximum locality in each
differentiated slot and invariance under constants give
\begin{equation}
 {\bf1}_E\mu_{w_j}={\bf1}_E\mu_u.
 \label{eq:r3-contact-before-limit}
\end{equation}
The top measures converge weakly against compact continuous
functions, by the original-current decreasing theorem and the
coefficient-Radon convergence result. For a compact $K\subset E$,
local Portmanteau therefore gives
\[
 \mu_w(K)\geq\limsup_j\mu_{w_j}(K)=\mu_u(K).
\]
For an arbitrary Borel subset of $E$, inner regularity of the
fixed Radon measure $\mu_u$ extends this inequality from its
compact subsets. Work first in a relatively compact stage and
then exhaust if necessary. This step does not require $E$ to
be closed or ordinary open. On the complementary Borel set
$\{u<v\}$, strict Borel maximum locality gives
$\mu_w=\mu_v$. The two disjoint restrictions prove
\eqref{eq:r3-relative-max-inequality}.
If $u\leq h$, take $v=h$, so that $w=h$ and
$\{u\geq h\}=\{u=h\}$, giving
\eqref{eq:r3-relative-contact-inequality}.
\end{proof}

\begin{corollary}[Zero mass on contact with a maximal potential]
\label{cor:r3-relative-zero-contact-mass}
If $u\leq h$ are actual bounded classical potentials and
$T_{S,q}(h,\ldots,h)=0$, then
$ {\bf1}_{\{u=h\}}T_{S,q}(u,\ldots,u)=0$.
\end{corollary}

The sets in this statement need only be Borel. No arbitrary
varying-background theorem, capacity regularity assertion, or
exchange of discontinuous limits is used.

% END source: base_r3/sections/relative_contact_inequality.tex


% BEGIN frozen or reviewed source: base_r3/sections/nonzero_dirichlet.tex
\section{Nonzero atomic Dirichlet problems on one fixed domain}
\label{sec:r3-nonzero-dirichlet}

Fix a finite extension $K/\mathbf Q_p$ with its original absolute
value. Let $Y/K$ be smooth integral affine of pure dimension $n\geq1$.
Throughout this section the domain is fixed in advance:
\[
 V\Subset Y^{\an}
\]
is connected and ordinary open, has nonempty ordinary boundary, and
admits
\[
 \rho\in\Cb(V)\cap C(\overline V),\qquad
 \rho<0\text{ on }V,\qquad \rho=0\text{ on }\partial V.
 \tag{H}\label{eq:r3-dirichlet-domain}
\]
Closures and boundary limits are ordinary ones in $Y^{\an}$.
The points and masses below are chosen after $V$ has been fixed.
We use $D=d'd''$ and the ambient background $[1]$, with the original
CLD calibration. All weak equalities and limits are tested against
the full original compactly supported smooth scalar class.

The inputs from the preceding records are the bounded mixed calculus,
ordinary upper-envelope closure, divisorial no-jump, local
maximality equivalence $(Du)^n=0$, the boundary maximum principle,
and pointwise ambient comparison, domination and uniqueness.
These inputs retain the all-extensions classical definition. The
projective input B, Theorems C.6 and 4.4, has already been used in
proving those local results; no new projective or residue-characteristic
hypothesis is inserted here.

\subsection{Point obstacles and their actual envelopes}

Fix pairwise distinct ambient divisorial points
$Z=\{z_1,\ldots,z_r\}\subset V$, where $r\geq1$. For
$a=(a_1,\ldots,a_r)\in[0,\infty)^r$ put $A(a)=\max_i a_i$ and
\[
 \mathcal F_a=\{v\in\Cb(V):-A(a)\leq v\leq0,
                         \ v(z_i)\leq-a_i\text{ for every }i\},
 \qquad U_a=\left(\sup_{v\in\mathcal F_a}v\right)^*.
\]
The constant $-A(a)$ makes this family nonempty.

\begin{proposition}[Actual weighted point envelopes]
\label{prop:r3-weighted-point-envelope}
For every $a$ the function $U_a$ belongs to $\mathcal F_a$ and
\[
 U_a=\sup_{v\in\mathcal F_a}v
\]
at every point. It is also the greatest member of $\Cb(V)$ with
$v\leq0$ and $v(z_i)\leq-a_i$, even without imposing the lower
bound in the definition of $\mathcal F_a$. It has ordinary
boundary limit zero.

For all $a,b\in[0,\infty)^r$ and $t\geq0$,
\[
 \|U_a-U_b\|_{L^\infty(V)}\leq\|a-b\|_\infty,
 \qquad U_{ta}=tU_a.
 \tag{1}\label{eq:r3-envelope-lipschitz}
\]
For every fixed $M<\infty$, all parameters $a\in[0,M]^r$ have
one common continuous lower barrier with zero boundary limit.
\end{proposition}

\begin{proof}
Upper-envelope closure gives $U_a\in\Cb(V)$ and
$-A(a)\leq U_a\leq0$. Divisorial no-jump gives the crucial
actual constraint
\[
 U_a(z_i)=\sup_{v\in\mathcal F_a}v(z_i)\leq-a_i.
\]
Thus $U_a$ itself is in $\mathcal F_a$. The pointwise supremum
is consequently at least $U_a$, and is at most $U_a$ by the
definition of upper regularization. This proves the equality
and the greatest-member assertion for the displayed family.
If $v\in\Cb(V)$ is any other nonpositive admissible function,
then $\max(v,-A(a))\in\mathcal F_a$ and dominates $v$.
This proves the assertion without the artificial lower bound.

Set
\[
 d_Z=\min_i(-\rho(z_i))>0.
\]
For $M>0$ the continuous classical function
\[
 b_M=\max\left(-M,\frac{M}{d_Z}\rho\right)
\]
equals $-M$ at every $z_i$ and has boundary limit zero.
For $a\in[0,M]^r$, the greatest-member characterization gives
$b_M\leq U_a\leq0$. This proves the common barrier and the
boundary limit. When $a=0$, the family consists of zero and
$U_0=0$.

Let $\delta=\|a-b\|_\infty$. The function
\[
 w=\max(-A(b),U_a-\delta)
\]
lies between $-A(b)$ and zero. At each $z_i$, both branches
are at most $-b_i$, because $b_i\leq a_i+\delta$.
Hence $w\in\mathcal F_b$ and $U_b\geq U_a-\delta$.
Interchanging the parameters proves the first inequality in
\eqref{eq:r3-envelope-lipschitz}. For $t>0$, multiplication by
$t$ is a bijection between the two admissible families, and
commutes with ordinary upper regularization. The case $t=0$
has already been noted.
\end{proof}

\begin{proposition}[A continuous homogeneous map of obstacle depths to masses]
\label{prop:r3-atomic-mass-map}
There are nonnegative real functions $c_i(a)$ such that
\[
 (DU_a)^n=\sum_{i=1}^r c_i(a)\delta_{z_i}.
 \tag{2}\label{eq:r3-atomic-mass-map}
\]
Each $c_i$ is locally Lipschitz on $[0,\infty)^r$, satisfies
$c_i(ta)=t^n c_i(a)$, and obeys
\[
 a_i=0\ \Longrightarrow\ c_i(a)=0.
 \tag{3}\label{eq:r3-mass-face}
\]
For every $a\ne0$ the total mass
$m(a)=\sum_i c_i(a)$ is strictly positive.
\end{proposition}

\begin{proof}
Write $Z_+(a)=\{z_i:a_i>0\}$. We first prove that $U_a$
is locally maximal off $Z_+(a)$. Take an ordinary
$W\Subset V\setminus Z_+(a)$ and a competitor
$v\in\Cb(W)$ which equals $U_a$ outside a compact
$L\Subset W$. Glue $\max(v,U_a)$ on $W$ to $U_a$ off
$L$. The resulting function $w\in\Cb(V)$ is bounded above:
only the compact set $L$ has changed. It has boundary limit
zero, so the boundary maximum principle gives $w\leq0$.
Moreover $w\geq U_a\geq-A(a)$. At each positive-depth
constraint it equals $U_a$; each zero-depth constraint is
automatic from $w\leq0$. Thus $w\in\mathcal F_a$.
The actual greatest-member property forces $w\leq U_a$,
and hence $v\leq U_a$. This is local maximality.

The proved homogeneous-equation criterion now gives
$(DU_a)^n=0$ on $V\setminus Z_+(a)$. This positive Radon
measure is supported on a finite compact set, so it is finite
and has the form \eqref{eq:r3-atomic-mass-map}. Its support
also proves \eqref{eq:r3-mass-face}. Homogeneity follows from
\eqref{eq:r3-envelope-lipschitz} and the proved BT homogeneity.

Here is a direct continuity estimate, with no varying-background
assertion. Choose original smooth compact scalar cutoffs
$\chi_i$ which equal one near $z_i$ and zero near every other
point of $Z$. Such cutoffs exist in the original CLD smooth
sheaf. For $a,b\in[0,M]^r$, the bounded mixed calculus gives
\[
 (DU_a)^n-(DU_b)^n
   =D\left((U_a-U_b)
          \sum_{j=0}^{n-1}(DU_a)^j\wedge(DU_b)^{n-1-j}\right).
\]
Pair with $\chi_i$. The coefficient-Radon estimates for the
fixed smooth test $D\chi_i$ and relative CLN, on a compact
neighborhood of its support, give a constant $C_{i,M}$ with
\[
 \begin{split}
 |c_i(a)-c_i(b)|
 &\leq C_{i,M}\|U_a-U_b\|_{L^\infty(V)}\\
 &\leq C_{i,M}\|a-b\|_\infty.
 \end{split}
\]
The constant depends on the fixed test stage, $M,n$ and the
domain geometry, not on $a,b$. Borel multiplication by the
bounded difference is legitimate for these coefficient-Radon
currents. When $n=1$, the sum contains the single background
$[1]$ and the same argument applies. Alternatively, uniform
convergence in \eqref{eq:r3-envelope-lipschitz}, the proved
BT stability, and the fixed finite support give continuity
of every coefficient directly. No mass can escape to the
boundary because the support is always contained in $Z$.

Finally suppose $a\ne0$ and $m(a)=0$. Then $(DU_a)^n=0$.
The common lower barrier makes every set
$\{U_a<-\delta\}$ relatively compact in $V$. Pointwise
ambient domination applied to $(U_a,0)$ gives $U_a\geq0$.
This contradicts $U_a(z_i)\leq-a_i<0$ for an index with
$a_i>0$. Therefore $m(a)>0$.
\end{proof}

\subsection{A finite-dimensional surjectivity argument}

\begin{lemma}[Surjectivity of a face-preserving simplex map]
\label{lem:r3-simplex-surjectivity}
Let
\[
 \Delta^{r-1}=\{a\in[0,\infty)^r:\textstyle\sum_i a_i=1\}.
\]
If a continuous $F:\Delta^{r-1}\to\Delta^{r-1}$ satisfies
$a_i=0\Rightarrow F_i(a)=0$, then $F$ is surjective.
\end{lemma}

\begin{proof}
First let $y$ be in the relative interior of the simplex.
Put
\[
 d_i(a)=(y_i-F_i(a))_+,
 \qquad \sigma(a)=\sum_i d_i(a),
 \qquad
 T_i(a)=\frac{a_i+d_i(a)}{1+\sigma(a)}.
\]
This is a continuous self-map of the simplex. Brouwer's
fixed-point theorem, used here as an external topological
input \cite{Brouwer11}, gives a fixed point $a$. Its equations are
\[
 d_i(a)=\sigma(a)a_i.
\]
If $a_i=0$, the face condition gives $F_i(a)=0$, whereas
$y_i>0$ gives $d_i(a)>0$, a contradiction. Thus all
$a_i>0$. If $\sigma(a)>0$, every $d_i(a)>0$, so
$y_i>F_i(a)$ for every $i$, contradicting that both
vectors have sum one. Hence $\sigma(a)=0$. Then
$y_i\leq F_i(a)$ for every $i$, and equality of their sums
gives $F(a)=y$.

For a boundary target, restrict to its support face. The
face condition makes that face invariant, and the preceding
argument applies to its relative interior. A one-point
simplex is immediate. This proves surjectivity.
\end{proof}

For $r=2$ only the intermediate value theorem is needed:
the normalized first mass along $a=(t,1-t)$ is continuous,
equals zero at $t=0$, and equals one at $t=1$.

\begin{theorem}[Finite atomic Dirichlet problem on the fixed domain]
\label{thm:r3-finite-atomic-dirichlet}
Assume \eqref{eq:r3-dirichlet-domain}. Let $z_1,\ldots,z_r$
be distinct ambient divisorial points of $V$, let $m_i\geq0$,
and let $b\in\mathbf R$. There exists a unique bounded
classical psh function $u\in\Cb(V)$ such that
\[
 (Du)^n=\sum_{i=1}^r m_i\delta_{z_i},\qquad
 \lim_{\substack{x\to\xi\\x\in V}}u(x)=b
                         \quad(\xi\in\partial V).
 \tag{4}\label{eq:r3-atomic-dirichlet}
\]
The equality of measures has the original calibration.
For a fixed finite set $Z$, there is a finite constant
$C_Z$, depending on $V$ and $Z$ but not on the masses or
$b$, such that
\[
 \|u-b\|_{L^\infty(V)}
       \leq C_Z\left(\sum_i m_i\right)^{1/n}.
 \tag{5}\label{eq:r3-finite-atomic-bound}
\]
With $V,Z$ fixed, the solution depends continuously in the
uniform norm on the nonnegative mass vector and on $b$.
No continuity of $u$ at arbitrary interior points is asserted
by this theorem alone.
\end{theorem}

\begin{proof}
It is enough first to take $b=0$. If all masses are zero,
take $u=0$. Otherwise let $M_\mu=\sum_i m_i>0$ and
$y_i=m_i/M_\mu$. On $\Delta^{r-1}$ define
\[
 F_i(a)=\frac{c_i(a)}{m(a)}.
\]
The denominator is positive; the map is continuous and
preserves faces by Proposition~\ref{prop:r3-atomic-mass-map}.
Choose $a\in\Delta^{r-1}$ with $F(a)=y$, using the
simplex lemma. Put
\[
 t=\left(\frac{M_\mu}{m(a)}\right)^{1/n},\qquad u=tU_a.
\]
Homogeneity gives every prescribed coefficient $m_i$, and
the boundary limit is zero. Adding $b$ proves existence in
\eqref{eq:r3-atomic-dirichlet}. The same previously fixed
domain has been used throughout; no domains adapted to
individual atoms have been glued together.

For uniqueness, two solutions have equal ordinary boundary
limits. For every positive margin, both strict-gap closures
are therefore compact in $V$: cover the compact ordinary
boundary by finitely many neighborhoods on which the
difference has smaller absolute value. Pointwise ambient
uniqueness gives equality at every point.

Since $m(a)$ is continuous and positive on the compact
simplex, $m_Z=\min_{\Delta^{r-1}}m(a)>0$. Also
$\|U_a\|_\infty\leq A(a)\leq1$ there. Thus
$C_Z=m_Z^{-1/n}$ proves \eqref{eq:r3-finite-atomic-bound}.

To prove stability, let the nonnegative mass vectors tend
to a vector $\mu$. If their total masses tend to zero,
\eqref{eq:r3-finite-atomic-bound} gives uniform convergence
to zero. Otherwise represent their zero-boundary solutions
as $t_jU_{a_j}$ with $a_j\in\Delta^{r-1}$ as above.
The parameters $t_j$ are bounded. From any subsequence one
can extract a further subsequence with $a_j\to a$ and
$t_j\to t$. The envelope Lipschitz estimate gives uniform
convergence to $tU_a$. The atomic mass map and the common
boundary barrier show that this limit solves the limiting
problem. Uniqueness identifies it with the specified
solution. All subsequential limits therefore agree, which
proves uniform convergence of the entire sequence.
Translations handle the varying boundary constant.
\end{proof}

The analytic construction and the simplex argument are recorded
as \texttt{NEW\_PROVED}, relative to the preceding R1/R2
theorems. Brouwer's theorem is \texttt{EXTERNAL\_KNOWN}.
The construction does not sum single-atom solutions. Its
coefficients $c_i(a)$ come from the full nonlinear BT measure,
so every mixed term is included before prescribing the masses.

\subsection{Order and a summable countable atomic class}

\begin{lemma}[Monge--Amp\`ere order reverses zero-boundary solution order]
\label{lem:r3-measure-order}
Let $u_1,u_2\in\Cb(V)$ be bounded with boundary limits zero.
If $(Du_1)^n\geq(Du_2)^n$ as Radon measures, then
$u_1\leq u_2$ at every point. The measures need not have
finite total mass on $V$.
\end{lemma}

\begin{proof}
Write $\mu_i=(Du_i)^n$. For $t,\delta>0$ put
\[
 w=(1+t)u_1-\delta,\qquad E=\{u_2<w\}.
\]
The boundary limit of $u_2-w$ is $\delta$, so $E$ itself
has compact closure in $V$. The compact-gap comparison
principle gives
\[
 (1+t)^n\mu_1(E)\leq\mu_2(E)\leq\mu_1(E).
\]
The masses being compared are finite on this compact set.
Hence $\mu_1(E)=\mu_2(E)=0$. Pointwise ambient domination
for $(u_2,w)$ gives $u_2\geq w$ everywhere. Let
$\delta\downarrow0$ and then $t\downarrow0$. The extra
constant shift is what permits the cancellation without
assuming a finite total mass on the whole domain.
\end{proof}

Write $G(\mu)$ for a zero-boundary solution when one has
been constructed. Positive homogeneity and uniqueness give
$G(s\mu)=s^{1/n}G(\mu)$ for $s\geq0$. If the relevant
solutions exist, positivity of all mixed BT terms gives
\[
 G(\mu)+G(\nu)
       \leq G(\mu+\nu)
       \leq\min\{G(\mu),G(\nu)\}.
 \tag{6}\label{eq:r3-sum-sandwich}
\]
Indeed the full mixed expansion makes the Monge--Amp\`ere
measure of the sum at least $\mu+\nu$, so the lemma gives
the lower inequality. The two measure-order comparisons
give the upper inequalities. This explains precisely why
adding single-atom solutions is a useful subsolution but
does not compute the solution in dimension greater than one.

\begin{corollary}[A countable atomic Dirichlet class]
\label{cor:r3-summable-atomic-dirichlet}
Let $z_i\in V$ be pairwise distinct ambient divisorial points
and $m_i>0$, for $i\geq1$. Let
$g_i=G(\delta_{z_i})$ be their single-atom zero-boundary
solutions on this same fixed domain. Suppose
\[
 \sum_{i=1}^{\infty}m_i^{1/n}\|g_i\|_{L^\infty(V)}<\infty.
 \tag{7}\label{eq:r3-atomic-summability}
\]
Then $\mu=\sum_i m_i\delta_{z_i}$ is a locally finite
Radon measure on $V$. It has a unique bounded classical
zero-boundary solution. The solutions for its finite
partial sums decrease uniformly to that solution, with the
explicit error bound
\[
 0\leq u_N-u\leq
 \sum_{i>N}m_i^{1/n}\|g_i\|_{L^\infty(V)}.
 \tag{8}\label{eq:r3-atomic-tail-bound}
\]
Their weighted BT measures converge on every original compact
smooth test.
Adding a constant gives the same conclusion for any constant
boundary value.
\end{corollary}

\begin{proof}
Set $\mu_N=\sum_{i=1}^N m_i\delta_{z_i}$ and
$u_N=G(\mu_N)$, using the finite atomic theorem. The measure
order lemma makes $u_N$ decrease. The mixed sandwich gives
\[
 \sum_{i=1}^N m_i^{1/n}g_i\leq u_N\leq0.
\]
Condition \eqref{eq:r3-atomic-summability} makes the series
$s=\sum_i m_i^{1/n}g_i$ uniformly convergent and bounded
on $V$. Its finite sums decrease and are classical, so $s$
is classical. Its boundary limit is zero: first make the
uniform tail small, then use the boundary limits of the
finite initial sum. Thus
\[
 s\leq u_N\downarrow u\leq0
\]
gives a bounded classical limit with zero boundary limit.
In fact, repeated use of the mixed sandwich gives, for every
$N,M\geq1$,
\[
 u_N+\sum_{i=N+1}^{N+M}m_i^{1/n}g_i
       \leq u_{N+M}\leq u_N.
\]
Let $M\to\infty$. This proves
\eqref{eq:r3-atomic-tail-bound}, and hence uniform convergence.

The fixed-background decreasing BT theorem gives
$(Du_N)^n\to(Du)^n$. For a nonnegative original smooth
compact scalar test $\chi$ it follows that
\[
 \sum_i m_i\chi(z_i)
    =\lim_N\int\chi\,d\mu_N
    =\int\chi(Du)^n<\infty.
\]
Choose such a cutoff at least one near any prescribed
compact subset of $V$. This proves local finiteness of the
atomic sum; on this locally compact metrizable space it is
the corresponding Radon measure. The same identity gives
$(Du)^n=\mu$ on all original tests. The weighted decreasing
theorem also gives $u_N(Du_N)^n\to u(Du)^n$ locally.
Uniqueness follows from the two compact-gap conditions and
pointwise ambient uniqueness, without a global total-mass
assumption.
\end{proof}

For the singleton, the R2 compact-extremal formula identifies
\[
 g_z=\operatorname{Cap}_{[1]}(\{z\};V)^{-1/n}U_{\{z\}},
 \qquad
 \|g_z\|_\infty
       =\operatorname{Cap}_{[1]}(\{z\};V)^{-1/n}.
\]
The point constraint is exact at the divisorial point, so
$U_{\{z\}}(z)=-1$. Thus \eqref{eq:r3-atomic-summability}
is equivalently the explicit condition
\[
 \sum_i
 \left(\frac{m_i}
 {\operatorname{Cap}_{[1]}(\{z_i\};V)}\right)^{1/n}<\infty.
\]
The order, sandwich and countable-atomic conclusions are
\texttt{DIRECT\_COROLLARY} of the finite atomic theorem and
the proved positive mixed calculus, with their proofs given
above. No arbitrary diffuse-measure existence or stability
under moving supports is asserted here.

\begin{corollary}[A quantitative bound for atomic changes of mass]
\label{cor:r3-atomic-quantitative-stability}
Let $\mu=\sum_i m_i\delta_{z_i}$ and
$\nu=\sum_i n_i\delta_{z_i}$ be locally finite nonnegative atomic
measures on one finite or countable set of distinct divisorial
points in $V$. Suppose bounded classical zero-boundary solutions
$u_\mu,u_\nu$ exist. For
\[
 \kappa_i=\operatorname{Cap}_{[1]}(\{z_i\};V)>0,
 \qquad
 d_+=\sum_{m_i>n_i}\left(\frac{m_i-n_i}{\kappa_i}\right)^{1/n},
 \quad
 d_-=\sum_{n_i>m_i}\left(\frac{n_i-m_i}{\kappa_i}\right)^{1/n},
\]
assume $d_++d_-<\infty$. Then
\[
 -d_+\leq u_\mu-u_\nu\leq d_-,
 \qquad
 \|u_\mu-u_\nu\|_{L^\infty(V)}
       \leq\max(d_+,d_-)
       \leq\sum_i\left(\frac{|m_i-n_i|}{\kappa_i}\right)^{1/n}.
 \tag{9}\label{eq:r3-atomic-stability-bound}
\]
In particular this applies to the finite atomic solutions and to
any two countable solutions constructed under
\eqref{eq:r3-atomic-summability} on their combined support.
\end{corollary}

\begin{proof}
Put $\alpha=(\mu-\nu)_+$ and $\beta=(\nu-\mu)_+$, with
the positive and negative parts taken coefficient by coefficient.
The summability hypothesis and the preceding corollary produce
zero-boundary solutions $G(\alpha),G(\beta)$ with
\[
 -d_+\leq G(\alpha)\leq0,
 \qquad -d_-\leq G(\beta)\leq0.
\]
Positivity of every mixed BT term gives
\[
 (D(u_\nu+G(\alpha)))^n\geq\nu+\alpha\geq\mu.
\]
The measure-order lemma therefore yields
$u_\nu+G(\alpha)\leq u_\mu$. Interchanging $\mu,\nu$
gives $u_\mu+G(\beta)\leq u_\nu$. The displayed bounds
on $G(\alpha),G(\beta)$ prove the conclusion. This proof does
not need the existence of a separate solution for $\mu+\nu$.
\end{proof}

On the extendible-exhaustion domains of
Theorem~\ref{thm:r3-local-atomic-continuity}, every finite partial
solution extends continuously to $\overline V$. The uniform
tail bound consequently makes every countable solution satisfying
\eqref{eq:r3-atomic-summability} continuous on $\overline V$
as well. This last statement uses that additional domain hypothesis;
the uniform estimate itself holds on every domain fixed at the
start of this section.

% END source: base_r3/sections/nonzero_dirichlet.tex


% BEGIN frozen or reviewed source: base_r3/sections/atomic_contact.tex
\section{Contact and unique depth parameters for positive atomic data}
\label{sec:r3-atomic-contact}

Keep the fixed domain $V$, the distinct divisorial points
$z_1,\ldots,z_r$, and the actual envelopes $U_a$ from
Section~\ref{sec:r3-nonzero-dirichlet}. The statements below are
\texttt{DIRECT\_COROLLARY} of that construction and local maximality.
They do not require interior continuity of $U_a$.

\begin{proposition}[Complementarity at a finite point obstacle]
\label{prop:r3-atomic-complementarity}
For every $a\in[0,\infty)^r$ and every index $i$,
\[
 c_i(a)>0\quad\Longrightarrow\quad U_a(z_i)=-a_i.
\]
Equivalently, the atomic measure of $U_a$ is supported on those
point constraints at which equality holds.
\end{proposition}

\begin{proof}
Suppose that $U_a(z_i)<-a_i$, and write
$\delta=-a_i-U_a(z_i)>0$. Choose an ordinary
$W\Subset V$ containing $z_i$ and no other $z_j$. Let
$v\in\Cb(W)$ agree with $U_a$ off a compact subset
$L\Subset W$. The function $\max(v,U_a)-U_a$ vanishes off
$L$ and is bounded there, so it has a finite upper bound $M\geq0$.
Choose $0<t\leq1$ with $tM<\delta$. On $W$ put
\[
 w=(1-t)U_a+t\max(v,U_a),
\]
and set $w=U_a$ off $L$. The formulas agree on the ordinary
overlap. Positive linear combinations, the maximum rule, and
ordinary gluing show that $w$ is classical psh. It is bounded,
has ordinary boundary limit zero, and satisfies $w\geq U_a$.
The boundary maximum principle gives $w\leq0$.
At $z_i$ one has
\[
 w(z_i)\leq U_a(z_i)+tM<-a_i;
\]
at every other constraint point $w=U_a$. Thus
$w\in\mathcal F_a$, using $w\geq U_a\geq-A(a)$ for the
lower bound. The actual greatest-member property implies
$w\leq U_a$. Since $t>0$, it follows that $v\leq U_a$.

This proves local maximality of $U_a$ on $W$. The proved
maximality criterion gives $(DU_a)^n|_W=0$, so $c_i(a)=0$.
The contrapositive is the assertion. Only actual constraints
were used; no quasi-everywhere constraint was substituted.
\end{proof}

\begin{corollary}[Unique depth coordinates for strictly positive masses]
\label{cor:r3-active-mass-homeomorphism}
Let
\[
 \mathcal A=\{a\in[0,\infty)^r:c_i(a)>0\text{ for all }i\}.
\]
Then $\mathcal A$ is an open subset of $(0,\infty)^r$, and
the full mass map
\[
 c:\mathcal A\longrightarrow(0,\infty)^r,
 \qquad a\longmapsto(c_1(a),\ldots,c_r(a)),
\]
is a homeomorphism. Its inverse is the actual point-depth map
\[
 (m_1,\ldots,m_r)\longmapsto
 \bigl(-G(\textstyle\sum_i m_i\delta_{z_i})(z_j)\bigr)_{j=1}^r.
\]
Consequently the normalized mass map $F$ restricts to a
homeomorphism from $\mathcal A\cap\Delta^{r-1}$ onto the
relative interior of $\Delta^{r-1}$. For a target
$y\in\operatorname{relint}\Delta^{r-1}$ its unique normalized
depth vector is
\[
 a_i(y)=
 \frac{-G(\sum_jy_j\delta_{z_j})(z_i)}
 {\sum_{\ell=1}^r-G(\sum_jy_j\delta_{z_j})(z_\ell)}.
\]
\end{corollary}

\begin{proof}
Continuity of the coefficients makes $\mathcal A$ open, and
face preservation puts it inside $(0,\infty)^r$. Given a
strictly positive mass vector, the finite atomic theorem
constructs its unique solution as $tU_a=U_{ta}$ for some
$t>0$ and $a\in\Delta^{r-1}$. Its mass vector is exactly the
target, so $ta\in\mathcal A$. Complementarity gives
$-U_{ta}(z_i)=ta_i$ for every $i$. This proves surjectivity
and the displayed inverse formula.

If $c(a)=c(b)$ for $a,b\in\mathcal A$, actual Dirichlet
uniqueness gives $U_a=U_b$. Complementarity then gives
$a_i=-U_a(z_i)=-U_b(z_i)=b_i$ for every $i$. Thus the
map is injective. The inverse is continuous because the
fixed-support solution map is continuous in the uniform
norm, and evaluation at a fixed point is continuous for
that norm. The mass map itself is continuous by its proved
local Lipschitz estimate.

For normalized masses, scaling the unique solution scales
all its depths by the same positive factor. Dividing those
depths by their sum gives the displayed point of the simplex.
The sum is positive because the full depth vector is in
$\mathcal A$. The formula is continuous, and homogeneity
shows it is the inverse of the restricted normalized map.
\end{proof}

This does not assert that the mass map is injective on the
whole nonnegative cone. A constraint with strict slack has
zero mass, and its prescribed depth need not be recoverable
from the solution. The inverse theorem is restricted to the
region on which every prescribed atom has positive mass.

% END source: base_r3/sections/atomic_contact.tex


% BEGIN frozen or reviewed source: base_r3/sections/divisorial_obstacle_continuity.tex
\section{Continuous envelopes with divisorial point constraints}
\label{sec:r3-divisorial-obstacle-continuity}

This section uses the continuous ample-envelope theorem of B and the
divisorial no-jump theorem proved in R2. The first result is a
\texttt{DIRECT\_COROLLARY} of those inputs. The local application also
uses the explicit order-preserving common-collar construction. No
continuity theorem for general homogeneous Dirichlet solutions is
assumed or obtained.

\subsection{A continuous insertion argument}

\begin{lemma}[Continuous insertion with a uniform error]
\label{lem:r3-continuous-insertion}
Let $Z$ be a compact Hausdorff space. Let $a$ be a bounded upper
semicontinuous real function, and let $f$ be a bounded lower
semicontinuous real function with $a\leq f$. For every $\epsilon>0$
there is a continuous function $g$ such that
\[
 a-\epsilon\leq g\leq f.
\]
\end{lemma}

\begin{proof}
For each $x\in Z$, choose a neighborhood $O_x$ on which
$a(y)<a(x)+\epsilon/4$ and $f(y)>f(x)-\epsilon/4$.
The constant $c_x=a(x)-\epsilon/2$ then satisfies
$a(y)-\epsilon<c_x<f(y)$ on $O_x$. Take a finite subcover
and a continuous partition of unity subordinate to it. The same
partition-weighted sum of the constants $c_x$ is the required $g$.
\end{proof}

Let $Y/K$ be smooth, projective and geometrically integral, and let
$L$ be ample. All metric inequalities below are inequalities of local
weights; their differences are globally defined real functions.
The fixed field and its absolute value are unchanged.

\begin{theorem}[A lower semicontinuous obstacle with an attained envelope]
\label{thm:r3-lsc-obstacle-continuity}
Let $f$ be a lower semicontinuous metric on $L$ whose difference from
a continuous reference metric is bounded. Put
\[
 h_f=\sup\{r:r\text{ is a bounded classical metric on }L, r\leq f\},
 \qquad P_f=h_f^*.
\]
If $P_f\leq f$ at every point, then $h_f=P_f$ and $P_f$ is continuous.
More precisely, for every $\epsilon>0$ there is a continuous obstacle
$g_\epsilon\leq f$ such that
\[
 P_f-\epsilon\leq P_Lg_\epsilon\leq P_f.
\]
\end{theorem}

\begin{proof}
Choose a continuous semipositive reference metric $r_0$.
A sufficiently negative translate of $r_0$ is a competitor, and
every competitor is bounded above by $r_0+C$ for one constant $C$.
Replacing competitors by their maxima with one such negative
translate gives a uniformly bounded family without changing the
supremum. Corollary~\ref{cor:r2-projective-upper-envelope} therefore
makes $P_f$ a bounded classical metric. Under the stated hypothesis
it is itself a competitor. Thus $h_f=P_f$ pointwise.

Apply Lemma~\ref{lem:r3-continuous-insertion} to $P_f-r_0$ and
$f-r_0$, and add $r_0$ back. This gives a continuous metric
$g_\epsilon$ with $P_f-\epsilon\leq g_\epsilon\leq f$.
By B, Theorem C.6, $P_Lg_\epsilon$ is continuous and semipositive.
The actual-competitor comparison
Lemma~\ref{lem:r2-envelope-competitor} gives
$P_f-\epsilon\leq P_Lg_\epsilon$.
On the other hand, $P_Lg_\epsilon\leq g_\epsilon\leq f$, so it is
a competitor for $h_f$ and is at most $P_f$. These uniform
inequalities prove continuity of $P_f$.
\end{proof}

\begin{corollary}[Discontinuities confined to divisorial points]
\label{cor:r3-divisorial-lsc-obstacle}
In the preceding theorem, suppose that every discontinuity point
of $f$ is ambient divisorial. Then $P_f\leq f$ pointwise, and the
conclusion of the theorem holds.
\end{corollary}

\begin{proof}
At a continuity point of $f$, upper regularization of $h_f\leq f$
gives $P_f\leq f$. At a divisorial point $z$, the local
divisorial no-jump result, Corollary~\ref{cor:r2-divisorial-no-jump},
gives $P_f(z)=h_f(z)\leq f(z)$. These two cases cover the space.
\end{proof}

\begin{corollary}[Finite divisorial obstacles]
\label{cor:r3-finite-divisorial-obstacle}
Let $F$ be a continuous semipositive metric on $L$, let
$z_1,\ldots,z_N$ be distinct divisorial points, and let $a_i\geq0$.
The greatest bounded classical metric subject to
\[
 r\leq F,\qquad r(z_i)\leq F(z_i)-a_i\quad(1\leq i\leq N)
\]
exists and is continuous. Denote it by $P_a$. Then
\[
 F-\max_i a_i\leq P_a\leq F,
 \qquad
 \|P_a-P_b\|_\infty\leq\max_i|a_i-b_i|.
\]
\end{corollary}

\begin{proof}
The obstacle $f_a=F-\sum_i a_i\mathbf1_{\{z_i\}}$ is lower
semicontinuous and is continuous off the finite divisorial set.
The preceding corollary applies. The displayed lower bound is an
admissible continuous metric. If $\delta=\max_i|a_i-b_i|$, then
$P_a-\delta$ is admissible for the $b$-constraints, whence
$P_a-\delta\leq P_b$. Interchange $a$ and $b$.
\end{proof}

The lower semicontinuity hypothesis is deliberate. The point
obstacle has downward jumps and is generally not upper
semicontinuous. It must not be inserted into the upper
semicontinuous-obstacle theorem without the preceding argument.

\subsection{An application on domains with an extendible exhaustion}

\begin{lemma}[A framed smooth compactification of an affine variety]
\label{lem:r3-framed-compactification}
A smooth geometrically integral affine variety $Y/K$ has a smooth
projective compactification $\overline Y$, an ample line $A$ on
$\overline Y$, and a section $s$ with
$\overline Y\setminus\operatorname{div}(s)=Y$.
After taking a power, $A$ is very ample and $s$ is one coordinate
section of its projective embedding.
\end{lemma}

\begin{proof}
Embed $Y$ in affine space, normalize its projective closure, and
pull back the hyperplane line and its section at infinity.
Normalization is finite and is an isomorphism on the smooth open
$Y$. The pulled-back line is ample, and the nonvanishing locus of
its section is exactly $Y$.

Use the characteristic-zero projective resolution preserving $Y$
already invoked in B, Proposition 4.5. The ample-boundary property
can be retained at each blowup in this resolution. Indeed, suppose
$H$ is ample and $s_H$ has nonvanishing locus $Y$, and let
$\pi:Z'\to Z$ be a blowup of an ideal $I$ whose cosupport is in
the boundary. Write $I\mathcal O_{Z'}=\mathcal O_{Z'}(-E)$.
For sufficiently large $N$, the line
\[
 H'=\pi^*H^N\otimes\mathcal O_{Z'}(-E)
\]
is ample. Increasing $N$ if necessary, the section $s_H^N$
belongs locally to $I H^N$: this follows from the containment
of the cosupport in $\operatorname{div}(s_H)$ and the
Noetherian radical criterion, using finitely many affine charts.
The induced section of $H'$ has divisor
$N\pi^*\operatorname{div}(s_H)-E$. For still larger $N$ this
divisor is effective and has exactly the full boundary as its
support. Thus the property persists. The same finite-pullback
argument handles a normalization if one is inserted. The
resulting smooth projective compactification has the required
ample line and section. A sufficiently high very ample power,
with the powered section included in a basis, gives the last
assertion.
\end{proof}

Let $Y/K$ now be smooth, geometrically integral and affine.
Suppose that $\rho$ is a continuous classical psh function on
$Y^{\an}$ and that $V$ is a connected component of
$\{\rho<0\}$ with
\[
 V\Subset Y^{\an},\qquad
 \varnothing\ne\partial V\subset\{\rho=0\}.
\]
This includes the norm-box exhaustion domains constructed from
$\log\max(1,|g_1|,\ldots,|g_N|)-\epsilon$ in R2. It is stronger
than merely having an exhaustion defined on $V$.

For distinct divisorial points $z_i\in V$ and $a_i\geq0$, let
\[
 U_a=\sup\{u\in\Cb(V):u\text{ is bounded},\ u\leq0,
                         \ u(z_i)\leq-a_i\text{ for every }i\}.
\]
The divisorial no-jump theorem makes this actual supremum
classical and admissible, as in the finite-atomic construction.

\begin{theorem}[Continuity of the finite-point Dirichlet envelopes]
\label{thm:r3-local-atomic-continuity}
Under the displayed extendible-exhaustion hypotheses, $U_a$ is
continuous on $\overline V$, with value zero on $\partial V$.
Consequently the finite divisorial-atomic Dirichlet solutions with
constant boundary values constructed in this round are continuous
on $\overline V$ for these domains.
\end{theorem}

\begin{proof}
Choose $C>0$ with $C\rho(z_i)\leq-a_i$ for every $i$.
The bounded function $C\rho|_V$ is a competitor. Hence
\[
 C\rho\leq U_a\leq0\quad\hbox{on }V.
\]
These inequalities already give the claimed boundary values.

For $\delta>0$, define a function on $Y^{\an}$ by
\[
 w_\delta=
 \begin{cases}
 \max(U_a,C\rho+\delta),&\text{on }V,\\
 C\rho+\delta,&\text{off }V.
 \end{cases}
\]
Near every point of $\partial V$, one has $C\rho+\delta>0\geq U_a$
on the part lying in $V$. Thus the two prescriptions agree on an
ordinary neighborhood of the boundary. Ordinary gluing and the
maximum rule show that $w_\delta$ is classical psh.
As $\delta\downarrow0$, these functions decrease locally boundedly
to the classical psh function
\[
 w=
 \begin{cases}U_a,&\text{on }V,\\ C\rho,&\text{off }V.
 \end{cases}
\]
Moreover $w\leq g:=\max(0,C\rho)$, and $g$ is continuous
classical psh on $Y^{\an}$.

Use the framed compactification of
Lemma~\ref{lem:r3-framed-compactification}. Its one affine
coordinate chart contains the entire $Y$. Large upper bounds
on its finitely many affine coordinates give nested compact
norm boxes whose innermost ordinary interior contains
$\overline V$; all outer boxes remain in $Y^{\an}$.
The functions $w$ and $g$ are defined on a neighborhood of the
outer box and have common bounds there. Apply the explicit
common-collar construction from B, Lemma D.2, in the
order-preserving form used in
Theorem~\ref{thm:r2-local-upper-envelope}.
For clarity, in one fixed frame that construction is
\[
 \mathcal G(f)=\max(f+c,mb)
\]
on the outer box and the same fixed metric off its middle box.
Here $b=0$ on the inner box, the common translation $c$ makes
$f+c>0$ there, and $m$ forces equality with the fixed metric on
the outer collar. It gives bounded classical metrics
\[
 R=\mathcal G(w)\leq F=\mathcal G(g)
\]
on one ample line. The metric $F$ is continuous and classical.
B, Theorem 4.4, gives decreasing continuous model-semipositive
metrics converging to $F$; compactness and Dini's theorem make
this convergence uniform. Thus $F$ is a continuous semipositive
metric, as required by the projective finite-point corollary.
Their weights
on $V$ are respectively $U_a+c$ and $c$, since $g=0$ on $V$.

Let $P_a$ be the continuous projective finite-point envelope of
Corollary~\ref{cor:r3-finite-divisorial-obstacle}, with upper
obstacle $F$ and depths $a_i$. The metric $R$ is a competitor,
so $R\leq P_a$. Conversely, the local weight $P_a-c$ on $V$
is bounded classical, is at most zero, and satisfies the actual
point constraints. It is therefore at most $U_a$. On $V$ this
gives
\[
 U_a+c=R\leq P_a\leq U_a+c.
\]
Thus $U_a=P_a-c$ on $V$, proving interior continuity. Combined
with the previously established ordinary boundary limits, this
proves continuity on $\overline V$.
\end{proof}

\paragraph{Scope.}
The projective envelope statement is a direct consequence of B
and R2. The local theorem is a \texttt{DIRECT\_COROLLARY} after
the extension and common-collar construction above. Its extra
exhaustion hypothesis is explicit. This argument neither proves
interior continuity for the general peak-barrier Perron solution
of R2 nor replaces arbitrary boundary values by constant ones.

% END source: base_r3/sections/divisorial_obstacle_continuity.tex


% BEGIN frozen or reviewed source: base_r3/sections/skeletal_dirichlet.tex
\section{Non-atomic Dirichlet data on fixed Gauss skeleton boxes}
\label{sec:r3-skeletal-dirichlet}

Fix the field, the smooth integral affine $Y/K$, and the connected
ordinary exhaustion domain $V\Subset Y^{\an}$ of
\eqref{eq:r3-dirichlet-domain}. In particular $n=\dim Y\geq1$,
$\rho<0$ on $V$, and $\rho$ is continuous on $\overline V$ with
ordinary boundary value zero. The domain remains fixed throughout.
All Monge--Amp\`ere measures use $[1]$, the original absolute value,
and the original CLD tests and calibration.

The new step is a compactness mechanism for a specified class of
supports. It proves weak-measure stability in that class; it does not
assume such stability. The main existence theorem is recorded as
\texttt{NEW\_PROVED}, relative to the finite atomic construction and
the preceding R1/R2 results.

\subsection{The support geometry and its function estimates}

For a bounded open rectangular box $I\subset\mathbf R^n$, put
\[
 \mathbb A(I)=
 \{x\in(\mathbf G_{m,K}^n)^{\an}:
              (\log|T_1(x)|,\ldots,\log|T_n(x)|)\in I\}.
\]
Its Gauss section $\sigma:I\to\mathbb A(I)$ is characterized by
\[
 \left|\sum_{m\in\mathbf Z^n}c_mT^m\right|(\sigma(x))
          =\max_m |c_m|e^{m\cdot x}
\]
for Laurent polynomials. Suppose that, for $1\leq\ell\leq L$,
there are $K$-analytic isomorphisms onto ordinary opens
\[
 \Phi_\ell:\mathbb A(I_\ell)\xrightarrow{\sim}\Omega_\ell
          \subset V
\]
and compact rectangular boxes $J_\ell\Subset I_\ell$, each with
nonempty Euclidean interior. Set
\[
 K_\ell=\Phi_\ell(\sigma(J_\ell)),\qquad
 K=\bigcup_{\ell=1}^L K_\ell\Subset V.
 \tag{S}\label{eq:r3-gauss-box-support}
\]
The larger full polyannulus, and the positive margin between its box
and $J_\ell$, are part of the hypotheses. A compact subset merely
called a skeleton is not being substituted for these data.
The measure may be supported on any closed subset of this fixed $K$;
the approximating divisorial points below are allowed to lie anywhere
in $K$. Thus an irrational-radius face need not itself contain a
dense set of divisorial points.

\begin{lemma}[Uniform continuity on an inner Gauss box]
\label{lem:r3-gauss-box-equicontinuity}
If $u\in\Cb(\mathbb A(I))$ and $-B\leq u\leq0$, then
$F_u(x)=u(\sigma(x))$ is convex in each coordinate separately.
For $J\Subset I$ as above there is a constant $C_{I,J}$,
independent of $u$ and $B$, such that
\[
 |F_u(x)-F_u(y)|\leq C_{I,J}B\|x-y\|_1
                         \quad(x,y\in J).
\]
In particular a uniformly bounded family of classical functions
restricts to an equicontinuous family on $K$ in
\eqref{eq:r3-gauss-box-support}.
\end{lemma}

\begin{proof}
Fix a coordinate $i$ and fix all the other real coordinates $x_j$.
Let $F/K$ be the completed residue field of the weighted Gauss
point with these $n-1$ coordinates. For $n=1$, take $F=K$.
The corresponding coordinate elements $c_j\in F^\times$ have
$|c_j|=e^{x_j}$ and the Gauss independence property. The map from
the $F$-annulus with $\log|t|\in I_i$ given by
\[
 T_i=t,\qquad T_j=c_j\quad(j\ne i)
\]
lands in $\mathbb A(I)_F$. At its radius-$e^s$ Gauss point,
the image in the original $K$-space is
$\sigma(x_1,\ldots,s,\ldots,x_n)$. This follows directly by
grouping a Laurent polynomial according to its exponent of $T_i$
and applying the Gauss norm first to the remaining coefficients.

The all-complete-extensions classical hypothesis makes the
pullback of $u$ classical subharmonic on that annulus. Let
$a<b$ be any two logarithmic radii strictly inside $I_i$.
Subtract the affine function of $\log|t|$ interpolating its
values at the two Gauss endpoints. This affine function is
pluriharmonic, so the difference is still classical psh.
Its values at the two Shilov points of the closed subannulus
are zero. The affinoid maximum principle
\cite[Theorem~6.4]{GR25} makes the difference nonpositive.
This is exactly the convexity inequality for the $i$th
coordinate of $F_u$.

A real convex function with range in $[-B,0]$ on an interval
has uniformly bounded secant slopes on every inner interval:
comparison with secants to the outer endpoints bounds their
absolute values by the oscillation divided by the margin.
If $\delta>0$ is the smallest coordinate margin from $J$ to
the endpoints of $I$, the bound $2B/\delta$ is sufficient.
Change one coordinate at a time inside $J$ and add these
estimates. This proves the displayed inequality.

Each Gauss section is continuous and is a homeomorphism onto
its image, since its coordinates recover $x$. Pullback by
$\Phi_\ell$ preserves the classical definition. Apply the
estimate on each of the finitely many boxes. All auxiliary
complete extensions above were used only for the curve tests
of functions, not for any asserted base change of currents.
\end{proof}

\begin{lemma}[A uniform singleton capacity on the original domain]
\label{lem:r3-gauss-box-point-capacity}
There is $c_K>0$, depending on the data in
\eqref{eq:r3-gauss-box-support} and on $V$, such that
\[
 \operatorname{Cap}_{[1]}(\{z\};V)\geq c_K
                                      \quad(z\in K).
\]
Moreover $K$ has a dense set of ambient divisorial points over
the original field $K$.
\end{lemma}

\begin{proof}
First work on $\mathbb A(I_\ell)$. Choose $A_\ell>0$ with
\[
 A_\ell>\sum_{i=1}^n
       (\sup (I_\ell)_i-\inf(J_\ell)_i).
\]
For $a\in J_\ell$ the continuous classical function
\[
 h_a=\frac1{A_\ell}\sum_{i=1}^n
                       \max(0,\log|T_i|-a_i)
\]
lies in $[0,1]$ on the same $\mathbb A(I_\ell)$, independently
of $a$. Its ordinary Monge--Amp\`ere measure is
\[
 (Dh_a)^n=\frac{n!}{A_\ell^n}\delta_{\sigma(a)}.
 \tag{10}\label{eq:r3-gauss-box-test-atom}
\]
Here is a check of that normalization on all original tests.
Replace each maximum by a smooth convex regularization
$\gamma_\epsilon(\log|T_i|-a_i)$, with second derivative
a nonnegative approximate identity of mass one. The ordinary
product calibration on the split torus gives, for every
original compact smooth scalar test $\chi$,
\[
 \int_{\mathbb A(I_\ell)}\chi
       \bigwedge_{i=1}^nD\gamma_\epsilon(\log|T_i|-a_i)
  =\int_{I_\ell}\chi(\sigma(x))
       \prod_{i=1}^n\gamma_\epsilon''(x_i-a_i)\,dx.
\]
To check the formula when $\chi$ requires additional moments,
adjoin those moments to the coordinate map and subdivide its
tropicalization. On cells whose projection to the coordinate
box has rank less than $n$, the displayed top form vanishes.
On a full-rank cell the canonical calibration is the degree
over the coordinate Gauss skeleton. The coordinate map here
is the identity inclusion into the split torus, so its degree
is one and its point over $x$ is precisely $\sigma(x)$.
The construction of the canonical calibration and its
compatibility with projections
\cite[\S3.5.1, Lemma~3.5.2, and \S\S3.7--3.8]{CLD12}
therefore give the formula with the actual value
$\chi(\sigma(x))$, also after every required refinement.
For small
$\epsilon$ its support lies in $I_\ell$, and the right side
tends to $\chi(\sigma(a))$. Locally uniform BT convergence
gives the corresponding mixed product of the maxima.
Repeated coordinate factors vanish already for the smooth
rank-one forms. Expansion of the $n$th power leaves precisely
$n!$ distinct-coordinate terms, proving
\eqref{eq:r3-gauss-box-test-atom}. No finite extension or
change in the integration current was made.

Choose an intermediate open box $J_\ell^+$ with
$J_\ell\Subset J_\ell^+\Subset I_\ell$ and put
$W_\ell=\Phi_\ell(\mathbb A(J_\ell^+))$.
Its closure is compact in $\Omega_\ell$. The reverse
direction of Theorem~\ref{thm:r2-domain-capacity-comparison}
provides a fixed $C_\ell<\infty$ such that
\[
 \operatorname{Cap}_{[1]}(E;\Omega_\ell)
       \leq C_\ell\operatorname{Cap}_{[1]}(E;V)
                                  \quad(E\subset W_\ell).
\]
The isomorphism $\Phi_\ell$ has degree one and preserves
the original products and tests. Apply the last inequality
to a singleton and use \eqref{eq:r3-gauss-box-test-atom}.
One may take
\[
 c_K=\min_\ell\frac{n!}{C_\ell A_\ell^n}>0.
\]
This lower bound uses the proved reverse collar comparison;
ordinary domain monotonicity alone has the opposite direction.

Write $\lambda=-\log|\pi|>0$ for a uniformizer of the fixed
field. The points $a\in J_\ell\cap(\lambda\mathbf Q)^n$
are dense in $J_\ell$. Their Gauss points are the unique
Shilov points of strict fixed-radius polyannuli: rational
valuation thresholds are expressed by finitely many integral
power inequalities over $K$. They are therefore divisorial,
and remain so on ordinary neighborhoods and under the analytic
isomorphism, by \cite[Proposition~A.3 and Corollary~A.6]{GM19}.
This gives a dense divisorial subset of $K$, without extending
the ground field used for the data or currents.
\end{proof}

\subsection{From skeletal compactness to uniform compactness of solutions}

\begin{lemma}[A common support determines the uniform distance]
\label{lem:r3-support-uniform-isometry}
Let $u,v\in\Cb(V)$ be bounded, with ordinary boundary limit
zero, and suppose both $(Du)^n$ and $(Dv)^n$ are supported
on a compact subset $H\Subset V$. Then
\[
 \|u-v\|_{L^\infty(V)}=\sup_H|u-v|.
\]
\end{lemma}

\begin{proof}
Put $b=\sup_H|u-v|$. For every $\epsilon>b$, the Borel gap
$E=\{u<v-\epsilon\}$ is disjoint from $H$, so
$(Du)^n(E)=0$. The boundary values make its closure compact
in $V$, and give every compact-gap condition required for
pointwise ambient domination. That principle gives
$u\geq v-\epsilon$ on $V$. Interchange $u,v$ and let
$\epsilon\downarrow b$. The reverse norm inequality follows
from $H\subset V$. No strict-gap set was treated as an
ordinary open set.
\end{proof}

\begin{lemma}[Uniform bounds for finite atomic data in $K$]
\label{lem:r3-skeletal-finite-bounds}
Let $\nu$ be a finite nonnegative sum of divisorial atoms in
$K$, and let $u_\nu$ be its zero-boundary solution from
Theorem~\ref{thm:r3-finite-atomic-dirichlet}. Put
$M=\nu(K)$ and
\[
 d_K=\min_K(-\rho)>0,
 \qquad b_K=\max(-1,\rho/d_K).
\]
Then
\[
 \|u_\nu\|_\infty\leq(M/c_K)^{1/n},
 \qquad (M/c_K)^{1/n}b_K\leq u_\nu\leq0.
 \tag{11}\label{eq:r3-skeletal-uniform-bounds}
\]
In particular all such solutions with bounded total mass have
one common ordinary boundary barrier.
\end{lemma}

\begin{proof}
The assertion is immediate for $M=0$. Otherwise discard
zero-mass points. The finite atomic construction writes
$u_\nu=tU_a$, with $t>0$ and $a\in\Delta^{r-1}$.
Put $A=\max_i a_i>0$ and choose $i$ with $a_i=A$.
The defining lower bound and the actual constraint give
$U_a(z_i)=-A$. Consequently $B=\|u_\nu\|_\infty=tA$.
The function $u_\nu/B$ lies in $[-1,0]$ and equals $-1$
at $z_i$. It is a competitor for the singleton extremal,
so $u_\nu/B\leq U_{\{z_i\}}\leq0$. Both functions have
ordinary boundary limit zero. The exact total-mass order
Lemma~\ref{lem:r2-total-mass-order} and the compact-extremal
formula give
\[
 c_K\leq\operatorname{Cap}_{[1]}(\{z_i\};V)
      =\int_V(DU_{\{z_i\}})^n
      \leq\int_V(D(u_\nu/B))^n=\frac{M}{B^n}.
\]
All masses here are finite; the cited order theorem in fact
also permits infinite masses without cancellation.
This proves the norm bound.

At every constraint point $z_j\in K$, the function $b_K$
equals $-1$, which is at most $-a_j/A$. The greatest-member
description of $U_{a/A}$ therefore gives
$b_K\leq U_{a/A}=u_\nu/B$. Since $b_K\leq0$ and
$B\leq(M/c_K)^{1/n}$, the second bound follows. The same
$b_K$ works for every choice of the finite support in $K$.
\end{proof}

\subsection{Arbitrary finite measures on the prescribed skeleton}

\begin{theorem}[A non-atomic fixed-domain Dirichlet class]
\label{thm:r3-skeletal-dirichlet}
Assume \eqref{eq:r3-gauss-box-support}. For every finite
nonnegative Radon measure $\mu$ supported on $K$ and every
$b\in\mathbf R$, there is a unique bounded actual classical
psh function $u\in\Cb(V)$ with
\[
 (Du)^n=\mu,
 \qquad\lim_{\substack{x\to\xi\\x\in V}}u(x)=b
                                      \quad(\xi\in\partial V).
\]
The bounds \eqref{eq:r3-skeletal-uniform-bounds} hold for
$u-b$ with $M=\mu(K)$.

If $\mu_j\to\mu$ weakly as finite measures on this same
compact $K$, and $b_j\to b$, then their solutions converge
uniformly on $V$. Their weighted measures also satisfy
\[
 u_j(Du_j)^n\longrightarrow u(Du)^n
\]
against every compact continuous scalar test, hence against
every original compact smooth scalar test.

If $V$ additionally satisfies the extendible-exhaustion
hypotheses of Theorem~\ref{thm:r3-local-atomic-continuity},
all these solutions are continuous on $\overline V$.
\end{theorem}

\begin{proof}
First take $b=0$. The compact space $K$ is metrizable; it is
a finite union of compact Gauss boxes. Its dense divisorial
subset permits finite atomic measures $\mu_j$ supported in
$K$, with $\mu_j(K)=\mu(K)$, that converge weakly to $\mu$.
For example take finite Borel partitions of successively
smaller diameter and replace each part's mass by a nearby
divisorial point in $K$. Uniform continuity of functions in
$C(K)$ verifies the asserted weak convergence.

Let $u_j$ be the finite atomic solutions. Their uniform
bound and common boundary barrier are
\eqref{eq:r3-skeletal-uniform-bounds}. Lemma
\ref{lem:r3-gauss-box-equicontinuity} makes their restrictions
equicontinuous on each $K_\ell$. Arzel\`a--Ascoli and a
finite diagonal extraction give a subsequence uniformly
Cauchy on all of $K$. Lemma
\ref{lem:r3-support-uniform-isometry} then makes this
subsequence uniformly Cauchy on the entire original domain
$V$. Denote its uniform limit by $u$.

Uniform limits here are classical. Explicitly, take a further
subsequence with uniform errors at most $4^{-j}$; adding
$2\cdot4^{-j}$ makes it decrease to $u$. The decreasing
closure of the classical class applies. The common finite
bound makes $u$ locally bounded, and the common $b_K$
barrier gives its ordinary zero boundary limit.

Uniform convergence implies local capacity convergence on
every prescribed small norm box. The proved fixed-background
BT stability, Theorem~\ref{thm:capacity-stability}, therefore
gives $(Du_j)^n\to(Du)^n$ on all original compact tests.
But $(Du_j)^n=\mu_j\to\mu$ on those tests. The associated
Radon measures are determined by the original smooth scalar
tests, so $(Du)^n=\mu$. This passage uses uniform convergence
of the solutions obtained above, not an assumed stability
under weak convergence of their right-hand sides.
The bounds pass to the uniform limit. Pointwise ambient
uniqueness, with the two boundary compact-gap conditions,
gives uniqueness among all bounded classical solutions.

Now let arbitrary $\mu_j\to\mu$ on $K$. Their total masses
are bounded. The solutions just constructed satisfy the same
uniform bounds, boundary barrier, and skeletal equicontinuity.
From any subsequence, the preceding compactness and support
argument extracts a further subsequence converging uniformly
on $V$. Uniform BT stability identifies its measure with
$\mu$, and uniqueness identifies its limit with $u$.
Thus the entire sequence converges uniformly.

Every bounded classical function is continuous on $K$, by
Lemma~\ref{lem:r3-gauss-box-equicontinuity}. For a fixed
continuous compact scalar $\chi$, write
\[
 \int\chi u_j\,d\mu_j-\int\chi u\,d\mu
  =\int_K\chi(u_j-u)\,d\mu_j
       +\int_K\chi u\,d(\mu_j-\mu).
\]
The first term tends to zero by the uniform convergence and
mass bound; the second tends to zero because $\chi u$ is
continuous on $K$. This proves the stated weighted limit
with the actual representatives. Adding the constants $b_j$
proves all assertions for varying constant boundary values.

Under the additional extendible-exhaustion hypotheses, every
finite atomic approximant extends continuously to $\overline V$,
with value zero on the boundary. Uniform convergence on $V$
is also uniform on $\overline V$ after those extensions.
Their limit is therefore continuous on $\overline V$.
\end{proof}

For fixed $M$, the solutions with $\mu\geq0$, support in $K$,
and $\mu(K)\leq M$ form a compact set for the uniform norm:
the finite positive measures of bounded mass on compact $K$
are weakly compact, and the theorem gives a continuous
solution map. This is compactness of this prescribed-measure
class, not compactness of every uniformly bounded psh family.

\begin{example}[Non-atomic data in every dimension]
\label{ex:r3-skeletal-nonatomic}
Let $R>1$ and fix
\[
 V=\{\max_i|T_i|<R\}\subset(\mathbf A_K^n)^{\an},
 \qquad
 \rho=\log\max(1,|T_1|,\ldots,|T_n|)-\log R.
\]
This is an extendible-exhaustion domain. Fix finite
$a<b<\log R$ and let $K=\sigma([a,b]^n)$.
Every finite positive Radon measure on this $K$ has a
unique bounded classical Dirichlet solution with any
constant boundary value, continuous on $\overline V$.
For instance
\[
 \mu=\sigma_*(f(x)\,dx),\qquad
       0\leq f\in L^1([a,b]^n),
\]
is non-atomic and nonzero when $f\not\equiv0$ in $L^1$.
No equation on a new domain adapted to its mass distribution
has been used. The local testers in
\eqref{eq:r3-gauss-box-test-atom} are already bounded on
this whole $V$, so with $A=n(\log R-a)$ one can use
$c_K=n!/A^n$ in the uniform norm bound.
\end{example}

\paragraph{The remaining diffuse-data scope.}
The argument has not proved existence for every compactly
supported measure, or for every measure dominated by
$(Dw)^n$ for a continuous bounded classical $w$.
Outside the stated Gauss-box geometry, the uniform skeletal
continuity needed for Arzel\`a--Ascoli has not been supplied.
The earlier moving-branch counterexample rules out replacing
that step by unrestricted compactness of bounded psh families.
Nor may an atomization of a non-atomic measure be assumed to
remain dominated by that measure: a positive point mass
already contradicts such domination on its singleton.
The present proof obtains its testing continuity and uniform
compactness from the specified support geometry instead of
assuming a general variational compactness theorem.

% END source: base_r3/sections/skeletal_dirichlet.tex


% BEGIN frozen or reviewed source: base_r3/sections/moving_atomic_dirichlet.tex
% Reconstructed after the scratch reset. Original reviewed SHA256:
% cc54ba0d930f0684881340e6964306db5eea9e0d9e1b9846b3d47a3bd9c4e55e.
\section{Moving atoms and failure of weak-data Dirichlet stability}
\label{sec:r3-moving-atomic-dirichlet}

Let $\pi$ be a uniformizer, $r=|\pi|$, $\lambda=-\log r$,
$R=r^{-1}$, and
$\Omega=\{|T_l|<R:1\leq l\leq n\}$.
Choose pairwise distinct monic irreducibles $\overline P_i$ of
degrees $d_i\geq2$ over the residue field, and monic integral
lifts $P_i$. Let $\xi_i$ be the Shilov point of
$\{|P_i(T_1)|\leq r^{d_i}\}$ in the original analytic affine line.
It is a single original $K$-analytic point of residual degree
$d_i$. After the unramified splitting extension, it consists of
the conjugate Gauss points of discs centered at the roots of $P_i$
and of radius $r^{d_i}$.

Let $z_i$ be the product Gauss point with first factor $\xi_i$
and remaining coordinate radii $r$, and let $z$ be the product
Gauss point with first radius one and remaining radii $r$.
Let $y$ have all coordinate radii one. The distinct residue branches
give $\xi_i\to\zeta_{0,1}$, hence $z_i\to z$.
For completeness, a fixed polynomial has only finitely many
irreducible residue factors; its norm at $\xi_i$ eventually equals
its unit Gauss norm. Uniform polynomial approximation on compact
polydiscs gives the asserted convergence for the product points.

Define
\begin{equation}
 g_i=\max\left\{d_i^{-1}\log|P_i(T_1)|-\lambda,\,
       \log|T_2|-\lambda,\ldots,\log|T_n|-\lambda,\,-2\lambda\right\}.
 \label{eq:r3-moving-atomic-potentials}
\end{equation}
The list of remaining coordinates is empty when $n=1$.

\begin{theorem}[Explicit zero-boundary weak instability]
\label{thm:r3-moving-atomic-dirichlet}
The functions $g_i$ are continuous classical psh on $\overline\Omega$,
take values in $[-2\lambda,0]$, have ordinary boundary value zero,
and satisfy
\[
 (Dg_i)^n=\delta_{z_i}.
\]
They converge pointwise, indeed eventually at each point, to
\[
 g=\max\{0,\log|T_1|,\ldots,\log|T_n|\}-\lambda,\qquad
 (Dg)^n=\delta_y.
\]
Put $B=2^{(n-1)/n}\lambda$ and
\[
 H=\max\left\{
     \frac B\lambda(\log|T_1|-\lambda),\,
     \frac B{2\lambda}(\log|T_l|-\lambda)\ (2\leq l\leq n),\,
     -B\right\}.
\]
Then $H$ is the continuous zero-boundary solution of
$(DH)^n=\delta_z$, but
\[
 \|g_i-H\|_\infty\geq2\lambda-B>0.
\]
For $n\geq2$, $z\ne y$ and $g(z)=-\lambda\ne-B=H(z)$.
Thus weak convergence of the data on a fixed compact support
does not imply uniform or capacity convergence of the solutions;
in dimension at least two it does not even imply pointwise
convergence to the solution of the limit datum.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} counterexample;
the corresponding unrestricted weak-data stability assertion is
\texttt{FALSE}.

\begin{proof}
Finite maxima of logarithms with a finite floor are continuous
classical after every complete extension. At an ordinary boundary
point, some $|T_l|=R$. If $l=1$, monicity and integrality give
$|P_i(T_1)|=|T_1|^{d_i}$; otherwise the displayed $l$th term
is zero. All other terms are nonpositive on the closed polydisc.
This proves the boundary and boundedness assertions.

We compute the atom over the original field. The separable
irreducible reduction splits in the unramified extension of
degree $d_i$. Its roots are units, and distinct roots have
distance one. Near a conjugate root $\alpha$,
$\log|P_i(T_1)|=\log|T_1-\alpha|$ on the corresponding open
residue disc, since every other factor is a unit of norm one.
The tie of all terms in \eqref{eq:r3-moving-atomic-potentials}
occurs at radii $r^{d_i},r,\ldots,r$. The slope simplex has
volume $1/(d_i n!)$. Original torus integration contributes
$n!$, and passage to the original degree-$d_i$ point contributes
the factor $d_i$. Its original mass is exactly one.

There is a useful independent check that no other mass was
omitted. On $\mathbf P_K^n$, homogenize the sections
\[
 \pi^{d_i}P_i(T_1),\quad
 \pi^{d_i}T_l^{d_i}\ (2\leq l\leq n),\quad \pi^{2d_i}.
\]
They are base-point free sections of $\mathcal O(d_i)$:
on the affine chart the last is nonzero; at infinity their
leading terms are the coordinate $d_i$th powers. Their pure-FS
weight divided by $d_i$ is the displayed $g_i$ in this frame.
Its positive ordinary top measure has total mass
$\deg(\mathcal O(1)^n)=1$, by the established ordinary-current
model compatibility. The original atom already has mass one.
This proves $(Dg_i)^n=\delta_{z_i}$ on all original tests,
without assuming a pullback formula for an arbitrary delta form.

If $|T_1|>1$, monicity gives $g_i=g$. If $|T_1|\leq1$, at most
one of the distinct residue polynomials can have norm below one.
For every other index the first logarithmic term is $-\lambda$,
again giving $g_i=g$. Thus the equality is eventual at every
point, and always
\[
 0\leq g-g_i\leq\lambda.
\]
The slope simplex for $g$ at $y$ is the standard unit simplex;
elsewhere the active slope differences have rank below $n$.
Original tropical calibration gives $(Dg)^n=\delta_y$.
For $H$, the floor ties precisely at $z$. The slopes are
$B/\lambda$ in the first coordinate and $B/(2\lambda)$ in
the others, so its top mass is
\[
 \frac{B^n}{2^{n-1}\lambda^n}=1.
\]
The same rank calculation excludes further mass. Real slopes
are justified by smooth convex regularization and the proved
bounded continuity, so no rationality of $B/\lambda$ is needed.
The ordinary boundary values are zero. Actual zero-boundary
uniqueness identifies $g_i$ and $H$ as the respective solutions.
At $z_i$, one has $g_i=-2\lambda$ and $H=-B$, proving the
uniform lower bound.

Finally set
$K_*=\{z,z_1,z_2,\ldots\}$ and
$\Omega_0=\{|T_l|<r^{-1/2}\text{ for all }l\}\Subset\Omega$.
This is a fixed compact set in $\Omega_0$. The tester
$1+g_i/(2\lambda)$, restricted to $\Omega_0$, is normalized and
has mass $(2\lambda)^{-n}$ at $z_i$. Therefore any strict error
set on $K_*$ containing $z_i$ has outer capacity at least this
constant on $\Omega_0$. Taking an error threshold less than
$2\lambda-B$ rules out local capacity convergence to $H$.
\end{proof}

\begin{corollary}[Convex blocks do not preserve the original weak limit]
\label{cor:r3-moving-convex-block-mismatch}
For $n\geq2$ the example has tail convex blocks converging
uniformly to $g$, whose top measures converge to $\delta_y$,
whereas the unaveraged top measures converge weakly to
$\delta_z\ne\delta_y$.
\end{corollary}

\begin{proof}
The open residue branches
$O_i=\{|P_i(T_1)|<1\}\cap\Omega$ are pairwise disjoint.
The difference $g-g_i$ vanishes off $O_i$ and is at most
$\lambda$. Thus for any $N$ distinct indices,
\[
 0\leq g-\frac1N\sum g_i\leq\frac{\lambda}{N}
\]
uniformly. Uniform BT continuity gives the measure limit
$\delta_y$. The original measures have weak limit $\delta_z$.
This explicitly excludes deducing a nonlinear unaveraged
weak-limit representation solely from capacity compactness
of convex blocks.
\end{proof}

\paragraph{Additional scope checks.}
The weighted measures $g_i(Dg_i)^n$ tend to $-2\lambda\delta_z$.
For distinct $i,j$, $g_j(z_i)=-\lambda$, so the same fixed
compact capacity testers rule out a capacity-Cauchy subsequence.
The compact $K_*$ is not contained in finitely many fixed full
Gauss boxes on which all these bounded potentials have a common
interior Lipschitz estimate: such equicontinuity would contradict
$z_i\to z$ and $g_i(z_i)-g_i(z)=-\lambda$.
There is no contradiction with sharp weak-closure capacity
domination, since $\delta_z$ is itself the measure of $H$.
Nor is there a common locally bounded Monge--Ampere subsolution
dominating all $\delta_{z_i}$: its locally finite measure would
have mass at least one at infinitely many points of $K_*$.

% END source: base_r3/sections/moving_atomic_dirichlet.tex


% BEGIN frozen or reviewed source: base_r3/sections/bounded_subsolution.tex
\section{A bounded-subsolution theorem on extendible domains}
\label{sec:r3-bounded-subsolution}

The final Dirichlet theorem in this section has ambient background
$[1]$.  Its preliminary uniform weighted estimate holds over every
fixed allowed positive closed background.  The topology of the data
is stronger than ordinary weak convergence, which is insufficient
even for the explicit moving-atom examples.

We keep the original finite field extension $K/\mathbf Q_p$, its
absolute value, and all original CLD compact smooth tests.  Positive
backgrounds are symmetric, original-positive, and separately
$d'$- and $d''$-closed.  Actual bounded classical psh functions are
used throughout.  Products and energies mean the already established
fixed-background BT products, not wedges of arbitrary currents.

\subsection{Uniformity over bounded psh weights}

For finite signed Radon measures carried by a fixed compact subset
of an ordinary domain $O$, define
\[
 \|\sigma\|_{\mathrm{psh},O}
   =\sup_{\substack{a\in\mathcal C_b(O)\\0\leq a\leq1}}
                 \left|\int a\,d\sigma\right|,
 \qquad
 d_{\mathrm{psh},O}(\mu,\nu)=\|\mu-\nu\|_{\mathrm{psh},O}.
\]
The bound $0\leq a\leq1$ is on the same full domain $O$.

\begin{theorem}[Uniform psh-weight convergence for decreasing rows]
\label{thm:r3-psh-dual-monotone}
Let $S$ be a fixed positive background of bidegree $(p,p)$ on an
allowed ordinary domain $O$, and put $q=n-p\geq1$.  Suppose
$u_j\in\mathcal C_b(O)\cap C^0(O)$ decrease to
$u\in\mathcal C_b(O)$.  If $f$ is original smooth with compact
support in $O$ and $0\leq f\leq1$, then
\[
 \sup_{\substack{a\in\mathcal C_b(O)\\0\leq a\leq1}}
 \left|\int af\bigl(T_{S,q}(u_j^q)-T_{S,q}(u^q)\bigr)\right|
       \longrightarrow0.
\]
Here $u_j^q$ denotes $q$ repeated slots, not a scalar power.
All constants used in the proof depend only on fixed compact
stages, local bounds, and finitely many positive evaluations of $S$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
Split $f$ by a finite original smooth nonnegative partition subordinate
to smaller norm boxes covering its support.  It suffices to prove
the assertion for one summand, still denoted $f$, supported strictly
inside one such box.  Choose an original nonnegative compact cutoff
$\chi$ equal to one near its support.  All potentials are uniformly
bounded on a fixed larger compact stage.  For $j\leq k$ put
\[
 g_{jk}=u_j-u_k,\quad
 Q_{jk}=\sum_{r=0}^{q-1}(Du_j)^r(Du_k)^{q-1-r}S,\quad
 \mu_j=(Du_j)^qS.
\]
Every summand of $Q_{jk}$ is a positive separately closed current
of bidegree $(n-1,n-1)$, and the proved mixed calculus gives
\[
 D(g_{jk}Q_{jk})=\mu_j-\mu_k.
\]
Let
\[
 e_{jk}=\int\chi\,\Gamma_{Q_{jk}}(g_{jk})\geq0.
\]
For an original smooth psh tester $a$ with $0\leq a\leq2$ on
the fixed stage, compact Green and the product rule give
\[
 \int af(\mu_j-\mu_k)
   =-\int f\,\Gamma_{Q_{jk}}(a,g_{jk})
     -\int a\,\Gamma_{Q_{jk}}(f,g_{jk}).
\]
The pairing involving the original smooth $f$ has its usual smooth
first factor; no differentiated arbitrary Borel function is used.
Cauchy--Schwarz gives
\begin{equation}
 \left|\int af(\mu_j-\mu_k)\right|\leq A e_{jk}^{1/2},
 \label{eq:r3-uniform-weight-energy}
\end{equation}
with $A$ independent of $a,j,k$.  Indeed the required factors other
than $e_{jk}$ are bounded by
\[
 \int\chi\,\Gamma_{Q_{jk}}(a),\qquad
 \int\chi\,\Gamma_{Q_{jk}}(f).
\]
For the first, the cutoff identity is
\[
 \int\chi\,\Gamma_{Q_{jk}}(a)
 =\tfrac12\langle Q_{jk},a^2D\chi\rangle
       -\int\chi aD(aQ_{jk})
 \leq\tfrac12\left|\langle Q_{jk},a^2D\chi\rangle\right|.
\]
The subtracted measure is nonnegative.  The finite-test CLN estimate
bounds the last expression uniformly, using $a^2\leq4$ and the
fixed local bounds for the slots of $Q_{jk}$.  The second factor
is bounded by the same CLN theorem against the fixed original form
$\chi\,d'f\wedge d''f$.  The product rule and Cauchy--Schwarz can
first be applied to smooth approximants of $u_j,u_k$ and then passed
to their limits by functional weighted monotone convergence.  Every
background occurring in this passage is an explicit mixed product
over the same $S$.

For an actual $a\in\mathcal C_b(O)$ with $0\leq a\leq1$, take a
smooth decreasing majorizing row on the prescribed box.  After a
prefix it is between zero and two on the compact stage.  For fixed
$j,k$, dominated convergence against the fixed finite measures
$f\mu_j,f\mu_k$ passes \eqref{eq:r3-uniform-weight-energy} to $a$.
The bound $A$ has not changed.  This avoids introducing a first-order
product with a simultaneously varying discontinuous tester.

The energy has the exact scalar expression
\begin{equation}
 e_{jk}
   =\tfrac12\langle Q_{jk},g_{jk}^2D\chi\rangle
       -\int\chi g_{jk}(\mu_j-\mu_k).
 \label{eq:r3-dual-energy-expansion}
\end{equation}
Fix $j$ and let $k\to\infty$.  Functional weighted joint monotone
convergence applies to every term after expanding the scalar
polynomials in $u_j,u_k$.  It gives a finite nonnegative limit
$e_j$.  Letting $j\to\infty$ in that resulting expression gives
$e_j\to0$, because its scalar difference is now $u_j-u\to0$ and
the same functional weighted theorem applies to the mixed tuples
$(u_j,u)$.  The constant row $u$ is an allowed decreasing row.

For each fixed bounded psh tester $a$, weighted monotone continuity
gives $a\mu_k\to a\mu$, where $\mu=(Du)^qS$, against the test $f$.
Pass $k\to\infty$ in \eqref{eq:r3-uniform-weight-energy}; then
\[
 \left|\int af(\mu_j-\mu)\right|\leq A e_j^{1/2}
 \quad\text{for every normalized }a.
\]
Take the supremum only after this fixed-tester limit.  The right
side tends to zero.  Summing the finitely many initial localized
estimates proves the assertion on $O$.  No capacity convergence
was used to prove the uniform weighted estimate.
\end{proof}

\begin{lemma}[The psh dual distance detects weak convergence]
\label{lem:r3-dual-determines-weak}
Let $V$ be relatively compact in a smooth affine algebraic analytification
and fix a compact $L\subset V$.  On finite positive Radon measures
carried by $L$, $d_{\mathrm{psh},V}$ is a metric.  Convergence in this
metric implies ordinary weak convergence of these measures.
\end{lemma}

\begin{proof}
Constants are testers, so convergence controls total masses.  The
differences of bounded continuous classical psh functions on $V$
form a real vector lattice: for $a,b,c,d$ in that cone,
\[
 \max(a-b,c-d)=\max(a+d,c+b)-(b+d).
\]
This lattice contains constants and separates the points of $L$.
For the separation assertion, an affine polynomial $P$ whose
absolute values differ at two Berkovich points yields, for a suitable
finite $c$, the bounded continuous classical function
$\max(\log|P|,c)$, still separating them.  It is bounded above on
$\overline V$.  Affine polynomial seminorms separate the points of
the affine analytification.  The real lattice Stone--Weierstrass
theorem therefore gives uniform density of this lattice in $C(L)$.
Every bounded psh function in its definition can be shifted and
positively rescaled into $[0,1]$ on $V$.  The metric consequently
controls each fixed lattice test.  Uniform approximation and the
bounded masses give convergence against every continuous test on
$L$.  The same argument with distance zero proves equality of the
measures.  Symmetry and the triangle inequality are immediate.
\end{proof}

\subsection{The domain and the common exterior}

For the remainder of the section, $Y$ is a smooth geometrically
integral affine $K$-variety and $\rho$ is continuous classical psh
on all of $Y^{\an}$.  Let $V$ be a connected ordinary component
of $\{\rho<0\}$ with compact closure in $Y^{\an}$, nonempty ordinary
boundary, and $\rho=0$ on that boundary.  This is the extendible
exhaustion-domain hypothesis.  No strict psh exhaustion is assumed.
Write $\operatorname{MA}(v)=(Dv)^n$ on $V$.

\begin{lemma}[Weak boundary order for a subsolution]
\label{lem:r3-weak-boundary-subsolution-order}
Let $a,b\in\mathcal C_b(V)$ be bounded.  Suppose $b$ has ordinary
boundary limit zero, $\limsup_{x\to\partial V}a(x)\leq0$, and
$\operatorname{MA}(a)\geq\operatorname{MA}(b)$.  Then $a\leq b$
at every actual point of $V$.
\end{lemma}

\begin{proof}
The boundary maximum principle gives $a,b\leq0$.  Fix $t,\delta>0$
and put $w=(1+t)a-\delta$.  The strict gap set $E=\{b<w\}$ is
relatively compact: the boundary limsup of $w-b$ is at most
$-\delta$, and the boundary is compact.  It need not be ordinary
open.  The proved comparison theorem gives
\[
 (1+t)^n\operatorname{MA}(b)(E)
 \leq\operatorname{MA}(w)(E)
 \leq\operatorname{MA}(b)(E).
\]
These masses are finite, so $\operatorname{MA}(b)(E)=0$.
Pointwise ambient domination with the strict boundary gap yields
$w\leq b$.  Let $\delta\downarrow0$ and then $t\downarrow0$.
\end{proof}

\begin{lemma}[A common lower barrier and exterior]
\label{lem:r3-common-exterior-extension}
Fix a compact $L\Subset V$ and $B>0$.  Every bounded classical
zero-boundary solution $v$ with
\[
 -B\leq v\leq0,\qquad\operatorname{MA}(v)\text{ carried by }L
\]
satisfies $v\geq C\rho$ on $V$, where $C$ depends only on
$B,L,\rho$.  It has a classical extension $\widetilde v$ to
$Y^{\an}$, equal to $v$ on $V$ and to $C\rho$ on $Y^{\an}\setminus V$.
For the whole family one can make a common prescribed-box collar
extension $W_v$ such that, on a fixed neighborhood of $\overline V$,
$W_v=\widetilde v+c$ for one constant $c$, and all $W_v$ equal one
fixed smooth classical collar outside one larger compact set.
In particular $W_v-W_{v'}$ vanishes at every point outside $V$,
including every point of $\partial V$.
\end{lemma}

\begin{proof}
If $L$ is empty, domination gives $v=0$ and the assertion is
immediate.  Otherwise choose $C$ with
$C\min_L(-\rho)\geq B$.  The function
$b=\max(-B,C\rho)$ has zero boundary value and satisfies $b\leq v$
on $L$.  The measure $\operatorname{MA}(v)$ therefore vanishes on
$\{v<b\}$.  Pointwise domination gives $b\leq v$, and in particular
$C\rho\leq v$.

For $\varepsilon>0$, use $\max(v,C\rho+\varepsilon)$ inside $V$
and $C\rho+\varepsilon$ outside it.  Near every boundary point the
second branch is active, by the ordinary boundary limit of $v$.
The ordinary gluing theorem makes this a global classical function.
Let $\varepsilon\downarrow0$.  Decreasing closure gives precisely
the claimed actual extension, including its value zero on the
boundary.  The family has uniform bounds on every fixed compact
neighborhood of $\overline V$: inside $V$ use $B$, and outside it
all functions equal $C\rho$.

Apply the constructive common-collar part of the prescribed-box
regularization lemma (the B extension step) to this one uniformly
bounded family.  Choose the inner compact stage to contain
$\overline V$ in its interior and the outer barrier above the
common bound on the outer collar.  The same shift and maximum
construction preserve every $\widetilde v+c$ near $\overline V$
and give the same smooth outer collar.  Outside $V$ all inputs
were already identical, so the same pointwise construction keeps
them identical there.  This proves the final assertion.  The
algebraic compactification used for this collar does not change
the field or the original current test category.
\end{proof}

\begin{theorem}[Compact-support stability in the psh dual metric]
\label{thm:r3-psh-dual-dirichlet-stability}
On the specified extendible $V$, let $v_j$ be bounded actual
classical zero-boundary solutions, with one common bound
$-B\leq v_j\leq0$.  Suppose all $\mu_j=\operatorname{MA}(v_j)$
are carried by one compact $L\Subset V$ and form a Cauchy sequence
for $d_{\mathrm{psh},V}$.  Then there is a unique bounded actual
classical zero-boundary $v$ such that
\[
 v_j\longrightarrow v\quad\text{in all-domain local outer ambient
 capacity},\qquad
 d_{\mathrm{psh},V}(\mu_j,\operatorname{MA}(v))\longrightarrow0.
\]
Its measure is carried by $L$.  In particular, if
$d_{\mathrm{psh},V}(\mu_j,\mu)\to0$ for a finite positive measure
$\mu$ carried by $L$, then $\operatorname{MA}(v)=\mu$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
The case $B=0$ is immediate.  Use the common exterior and collar
from Lemma~\ref{lem:r3-common-exterior-extension}.  All $W_j=W_{v_j}$
are uniformly bounded on one fixed collar domain $\Omega$.  Their
differences vanish outside $V$, including its boundary.  Hence the
positive mixed energy has the exact expression
\[
 \begin{split}
 I_{[1]}(W_j,W_k)
 &=\int_V(v_j-v_k)(\mu_k-\mu_j)\\
 &\leq2B\,d_{\mathrm{psh},V}(\mu_j,\mu_k).
 \end{split}
\]
For the bound, write $v_j-v_k=B(a_j-a_k)$ with
$a_j=1+v_j/B$, so both $a_j,a_k$ are normalized psh testers on $V$.
No equality of the extension measures on $\partial V$ is asserted:
their actual multiplier $W_j-W_k$ is zero there, which is the
reason no boundary mass occurs in the identity.

The all-degree estimate of Theorem~\ref{thm:r3-I-capacity} makes
$W_j$ Cauchy in outer ambient capacity on $\Omega$.  Domain
comparison gives all-domain local convergence and the completed
calculus supplies a limit in $\mathcal A_{[1],b}$.  The proved
ambient representative theorem supplies its unique actual classical
representative.  More explicitly, take a summably capacity-convergent
subsequence and use its decreasing usc-regularized tail suprema.
Near $\overline V$ the continuous barriers
\[
 C\rho+c\leq W_j\leq\max(0,C\rho)+c
\]
pass through this construction at actual points.  Subtracting $c$
on $V$ gives a bounded classical representative $v$ satisfying
$C\rho\leq v\leq0$.  Thus it has ordinary boundary limit zero.
The Cauchy property identifies the same limit for the full sequence.

The completed weighted calculus gives weak convergence of
$\operatorname{MA}(v_j)$ to $\operatorname{MA}(v)$ on every original
compact test in $V$.  These measures are supported in $L$, so the
limit is carried by $L$ and is finite.  For every fixed normalized
actual psh tester $a$ on $V$, insert an original smooth cutoff equal
to one near $L$.  Weighted capacity continuity then gives
\[
 \int a\,d\mu_k\longrightarrow
              \int a\,d\operatorname{MA}(v).
\]
For fixed $j$, pass $k\to\infty$ in the defining dual inequality,
and only afterwards take the supremum in $a$.  This yields
\[
 d_{\mathrm{psh},V}(\mu_j,\operatorname{MA}(v))
       \leq\limsup_{k\to\infty}d_{\mathrm{psh},V}(\mu_j,\mu_k)
       \longrightarrow0.
\]
This proves strong convergence without assuming in advance that
the measure metric is complete.  It identifies any prescribed
metric limit by Lemma~\ref{lem:r3-dual-determines-weak}.
Uniqueness of the actual zero-boundary solution is the proved
ambient comparison and domination theorem.
\end{proof}

\subsection{Strong atomic approximation from a subsolution}

\begin{theorem}[The bounded-subsolution criterion]
\label{thm:r3-bounded-subsolution-dirichlet}
On the specified extendible domain $V$, let $\mu$ be a finite
positive Radon measure with compact support in $V$.  The following
are equivalent:
\begin{enumerate}
\item There is a bounded actual classical $u$ on $V$ with ordinary
boundary limit zero and $(Du)^n=\mu$.
\item There is a bounded actual classical $w$ on $V$ with
$(Dw)^n\geq\mu$.
\end{enumerate}
The solution in the first assertion is unique.  No boundary
condition is required of the subsolution in the second assertion.
This theorem asserts a bounded classical solution; it does not
assert interior continuity from the bounded-subsolution hypothesis
alone.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
Necessity is immediate.  For sufficiency the zero measure has
solution zero, so suppose $\mu\ne0$, and let $K=\operatorname{supp}\mu$.
Subtract a constant from $w$ so that $-M\leq w\leq0$.  Choose
$C$ so large that $C\rho<-M-1$ on an ordinary neighborhood of $K$.
Then
\[
 b=\max(w,C\rho)
\]
is bounded classical, equals $w$ on that neighborhood, and has
ordinary boundary value zero.  Ordinary locality and positivity
give $(Db)^n\geq\mu$ on all of $V$: on the neighborhood of $K$
this follows from the subsolution, and outside $K$ the datum is
zero.  This uses no presumed ordinary openness of a comparison
set involving $w$; the displayed strict constant bound supplies
the required open neighborhood.

The same epsilon-maximum construction as in the preceding lemma
extends $b$ to a global classical function $\widetilde b$, equal
to $C\rho$ off $V$.  We now use the accepted constructive input
from B, with the entire compact $\overline V$ protected.  Extend
the local potential to one semipositive metric on a projective
algebraic model in a fixed frame, and take its decreasing pure
Fubini--Study rows.  In that frame, after the common constant or
unit-log correction, their local potentials $b_j$ are continuous
classical on a fixed neighborhood of $\overline V$ and satisfy
\[
 b_j\downarrow\widetilde b.
\]
Their ordinary top-degree measures are finite positive sums of
Dirac masses at divisorial points of the original $K$-analytic
space.  This last fact is the accepted semipositive PL/model
measure compatibility applied to these particular algebraic
pure-Fubini--Study rows.  It is not a claim that every locally
defined PL function has globally finite atomic curvature.

On the compact boundary $\partial V$, the continuous functions
$b_j$ decrease to the continuous zero function.  Dini's theorem
on that compact set gives
\[
 \varepsilon_j=\max_{\partial V}b_j\downarrow0.
\]
The maximum principle gives $b_j\leq\varepsilon_j$ on $V$.
Also $b_j\geq b$ there.  Only the boundary restriction is used in
this Dini argument; continuity of $b$ in the interior is not assumed.

Put $\nu=(Db)^n$.  Since $\mu\leq\nu$, Radon--Nikodym gives
$\mu=f\nu$ for a Borel $0\leq f\leq1$, chosen zero outside $K$.
Choose a compact $L\Subset V$ with $K\subset L^{\circ}$.
There exist original smooth $f_l$ with common compact support in
$L$, $0\leq f_l\leq1$, and
\[
 \int|f_l-f|\,d\nu\longrightarrow0.
\]
Here only the finite measure on a compact collar is involved.
Radon regularity first approximates a bounded Borel function by
continuous compactly supported functions; original smooth
partitions and smooth approximation on finitely many moment
presentations, followed by a smooth truncation into $[0,1]$ and
a fixed support cutoff, give the asserted original smooth functions.
Thus neither a changed test sheaf nor an infinite-volume density
assertion is needed.

For each fixed $l$, Theorem~\ref{thm:r3-psh-dual-monotone}, with
$S=[1]$, gives
\[
 d_{\mathrm{psh},V}\bigl(f_l(Db_j)^n,f_l\nu\bigr)
                  \longrightarrow0\quad(j\to\infty).
\]
Choose strictly increasing $j(l)$ so that this distance is at most
$2^{-l}$, and set
\[
 \mu_l=f_l(Db_{j(l)})^n.
\]
These are finite positive divisorial atomic measures carried by
the same compact $L$, and
\[
 d_{\mathrm{psh},V}(\mu_l,\mu)
    \leq2^{-l}+\int|f_l-f|\,d\nu\longrightarrow0.
\]
This is the needed strong approximation, not merely weak
approximation by atomic measures.  Also
$\mu_l\leq(Db_{j(l)})^n$ on all of $V$.

The finite-divisorial-atom Dirichlet theorem on this same
exhaustion domain gives bounded actual zero-boundary classical
solutions $v_l$ with $\operatorname{MA}(v_l)=\mu_l$.
Apply Lemma~\ref{lem:r3-weak-boundary-subsolution-order} with
$a=b_{j(l)}-\varepsilon_{j(l)}$ and $b=v_l$.  Its hypotheses hold
by the preceding measure domination and boundary maximum.  Hence
\[
 0\geq v_l\geq b_{j(l)}-\varepsilon_{j(l)}
              \geq b-\varepsilon_1.
\]
This is one common global bound, independent of the atoms and
their masses.  Theorem~\ref{thm:r3-psh-dual-dirichlet-stability}
now gives a bounded actual classical zero-boundary $u$ with
$\operatorname{MA}(u)=\mu$.  It also supplies capacity convergence
of the constructed solutions.  Actual pointwise uniqueness follows
from the ambient domination theorem.
\end{proof}

\paragraph{Scope of stability and of the algebraic input.}
The compact support, common uniform bound, and extendible exhaustion
are used explicitly in the common-exterior energy argument.  The
construction retains the original field and invokes B only for its
accepted algebraic collar and pure-Fubini--Study approximation, with
the already known original-current/model-measure comparison.  It
does not derive arbitrary-background Dirichlet existence, ordinary
weak-data stability, or continuity under merely bounded subsolutions.
The preliminary uniform weighted theorem, in contrast, is valid for
arbitrary fixed $S$ and all its permitted remaining degrees.

For the moving-atom measures $\delta_{z_j}\rightharpoonup\delta_z$
of the explicit counterexample, the normalized tester
$1+g_j/(2\lambda)$ has values zero at $z_j$ and $1/2$ at $z$.
Consequently their psh dual distances are at least $1/2$.
The strong topology used here therefore excludes that actual weak
convergence defect.

% END source: base_r3/sections/bounded_subsolution.tex


% BEGIN frozen or reviewed source: base_r3/sections/capacity_dirichlet.tex
\section{The fixed-domain Dirichlet problem under power capacity growth}
\label{sec:r3-capacity-dirichlet}

Throughout this section $K/\mathbf Q_p$ is the prescribed finite
extension with its original absolute value. Let $Y/K$ be smooth,
geometrically integral and affine, of dimension $n\geq1$. Let
$\rho$ be a continuous classical psh function on all of $Y^{\an}$.
Fix a connected component $V$ of $\{\rho<0\}$ such that
\[
 V\Subset Y^{\an},\qquad
 \varnothing\ne\partial V\subset\{\rho=0\}.
\]
All boundaries are ordinary boundaries. Write
$C(E)=\operatorname{Cap}_{[1]}(E;V)$. The background in this
Dirichlet section is $[1]$, not an arbitrary relative current.
The currents and their convergence use all original CLD tests;
the classical functions retain actual values and all extension tests.

\begin{lemma}[An affine projective collar for a fixed lower obstacle]
\label{lem:r3-capdir-affine-collar}
Fix $B>0$ and put
\[
 \mathcal F_B=\{u\in\Cb(V):B\rho\leq u\leq0\}.
\]
There is a smooth projective compactification $Z$ of $Y$, an ample
line $L$ on $Z$, a fixed frame on an ordinary neighborhood of
$\overline V$, and an affine order-preserving map
$u\mapsto U$ from $\mathcal F_B$ to bounded classical metrics
on $L$ with the following properties. In that frame the weight
of $U$ is $u+c$ on $V$, with one constant $c$ independent of $u$.
All the metrics are equal on $Z^{\an}\setminus V$. They have
common bounds relative to a continuous reference metric, and the
metrics $F$ and $L_B$ associated with $0$ and $B\rho$ are
continuous. One has $L_B\leq U\leq F$.

Conversely, any bounded classical metric $P$ with
$L_B\leq P\leq F$ and $P=F$ outside $V$ corresponds in this
frame to a member of $\mathcal F_B$. The affine family has energy
\[
 \mathcal E(U)=\frac1{n+1}\sum_{j=0}^n
  \int_{Z^{\an}}(U-F)c_1(L,U)^j c_1(L,F)^{n-j}.
\]
It satisfies the mixed energy identities, midpoint inequality and
local energy-to-capacity estimates already proved.
\end{lemma}

\begin{proof}
For $u\in\mathcal F_B$ and $\epsilon>0$, glue
$\max(u,B\rho+\epsilon)$ on $V$ to $B\rho+\epsilon$ outside
$V$. Near every ordinary boundary point the latter function is
positive while $u\leq0$, so these prescriptions agree on an
ordinary neighborhood. The glued functions are classical and
decrease locally boundedly as $\epsilon\downarrow0$ to
\[
 \widetilde u=u\text{ on }V,\qquad
 \widetilde u=B\rho\text{ off }V.
\]
The lower bound guarantees the actual first equality. This gives
a classical function on $Y^{\an}$. The map $u\mapsto\widetilde u$
is affine and order preserving. Its upper member is the continuous
function equal to zero on $V$ and to $B\rho$ off $V$; its lower
member is the continuous function $B\rho$.

Use the framed smooth compactification
Lemma~\ref{lem:r3-framed-compactification}, with one frame whose
nonvanishing locus contains all of $Y$. Choose nested compact
norm boxes in that frame, the innermost containing $\overline V$
in its ordinary interior. On the outer box the family
$\widetilde u$ has one common bound. The constructive collar of
B, Lemma D.2, in the form used in
Theorem~\ref{thm:r2-local-upper-envelope}, gives on one ample
power the weights
\[
 U=\max(\widetilde u+c,mb)
\]
on the outer box, and a fixed metric outside the middle box.
Here $b=0$ on the inner box, $c$ makes every
$\widetilde u+c$ positive there, and the integer $m$ makes the
fixed branch strictly dominate on the outer collar. Since the
inputs all agree outside $V$, the outputs all agree there.
On $V$ the maximum is inactive and its weight is $u+c$.
Thus this particular family remains affine, even though the
collar operation on arbitrary inputs is a maximum. The upper
and lower members are continuous, and ordinary restriction
proves the converse assertion.

Global mixed Green exchange gives the stated energy identities;
differences of the metrics vanish outside the fixed compact
$\overline V$. Equivalently one can compute in this frame on a
larger norm box. When applying the smooth-collar formulation of
Theorem~\ref{thm:r3-I-capacity}, use its prescribed-box auxiliary
collar reduction: preserve the weights near $\overline V$, up
to a common constant, and replace only their already common
exterior by a protected smooth collar. Their difference and all
energy integrands supported by that difference are unchanged.
The constants are uniform for the present bounded family. This
does not assert that the original $B\rho$ is smooth.
\end{proof}

\begin{lemma}[A constrained maximizer]
\label{lem:r3-capdir-maximizer}
Let $H\Subset V$ be compact, and let $\mu$ be a finite positive
Radon measure carried by $H$, with
\[
 \mu(E)\leq A C(E)^{1+\alpha}\quad
 (E\subset H\text{ Borel}),\qquad A,\alpha>0.
\]
For every $B>0$, the functional
\[
 \mathcal F(U)=\mathcal E(U)-\int_H u\,d\mu
\]
on the preceding affine family attains its supremum.
\end{lemma}

\begin{proof}
The common bounds, the fixed projective intersection mass and
finiteness of $\mu$ bound the functional above and below. Choose
a maximizing sequence $U_j$, with local weights $u_j+c$ on $V$.
The family is convex. The integral term is affine, so its midpoint
gap cancels, and Lemma~\ref{lem:r3-energy-midpoint} gives
\[
 0\leq c_n I(U_j,U_k)
 \leq\sup\mathcal F-
              \frac{\mathcal F(U_j)+\mathcal F(U_k)}2
 \longrightarrow0\quad(j,k\to\infty).
\]
The energy-to-capacity estimate on the fixed surrounding norm box
therefore makes the weights capacity Cauchy on every smaller
compact stage. The reverse framed domain comparison,
Corollary~\ref{cor:r2-fixed-domain-local-convergence}, upgrades
this to all-domain local capacity Cauchy before taking any
completion. The ambient completion and unique classical
representative, Corollary~\ref{cor:r2-ambient-classical-representative},
produce a bounded classical limit metric $U$.

Its actual bounds $L_B\leq U\leq F$ and its equality to $F$
outside $V$ follow, for example, from the decreasing tail-usc
formula for that representative: the two bounding metrics are
continuous, so both bounds pass through the tail regularization
and its decreasing limit. Thus the limit is an admissible member
of the same affine family. The completed weighted calculus on
a finite cover of $\overline V$ gives
$\mathcal E(U_j)\to\mathcal E(U)$; outside $V$ all weighted
differences in this formula vanish.

Let $M=\mu(V)$ and let $L$ be a bound for $|u_j|,|u|$. For
$\delta>0$, set $E_j=H\cap\{|u_j-u|>\delta\}$. Local capacity
convergence and a finite cover of $H$ give $C^*(E_j)\to0$ on
the fixed domain $V$. Hence
\[
 \int_H|u_j-u|\,d\mu
 \leq\delta M+2LA\,C^*(E_j)^{1+\alpha}.
\]
First let $j\to\infty$, then $\delta\downarrow0$. This proves
continuity of the measure term, so the admissible limit attains
the supremum. No weak convergence theorem for arbitrary varying
Monge--Amp\`ere measures was used.
\end{proof}

\begin{lemma}[The upper variation]
\label{lem:r3-capdir-upper-variation}
If $\mu\ne0$ and $U$ is the constrained maximizer, with local
function $u$, then $u<0$ everywhere on $V$ and
\[
 (Du)^n\leq\mu.
\]
In particular its Monge--Amp\`ere measure is carried by $H$.
\end{lemma}

\begin{proof}
The zero function is not a maximizer. Indeed the competitor
$tB\rho$, $0<t\leq1$, is admissible. The reference metric $F$
has constant weight on $V$, and all differences are zero off $V$.
Consequently its energy is
\[
 \mathcal E(U_{tB\rho})=
 \frac{t^{n+1}}{n+1}\int_V B\rho\,(BD\rho)^n.
\]
This is a finite constant times $t^{n+1}$. The change in the
measure term is $-t\int_H B\rho\,d\mu>0$, of first order in
$t$. Thus $u\ne0$. The classical strong maximum principle on
the connected domain makes the nonpositive $u$ strictly negative.

Let $\phi\geq0$ be an original compact smooth scalar test in
$V$, extended by zero to $Z$. Since $u$ is usc and negative,
its maximum on $\operatorname{supp}\phi$ is strictly negative.
For small $t>0$, $u+t\phi\leq0$ there. Let
\[
 P_t=P_L^{\mathrm b}(U+t\phi).
\]
The bounded-usc obstacle theorem supplies this actual greatest
classical minorant, and $U\leq P_t\leq U+t\phi$. It is equal
to $U$ off the support of $\phi$, and is an admissible member
of the same constrained family. Put $\Delta_t=P_t-U$.
Quasicontinuous-obstacle orthogonality,
Theorem~\ref{thm:r2-qc-orthogonality}, gives
\[
 \int\Delta_t c_1(L,P_t)^n
       =t\int\phi c_1(L,P_t)^n.
\]
Energy concavity and $0\leq\Delta_t\leq t\phi$ yield
\[
 \int\phi c_1(L,P_t)^n
 \leq\frac{\mathcal E(P_t)-\mathcal E(U)}t
 \leq\int\phi c_1(L,U)^n.
\]
Uniform convergence $P_t\to U$ and the proved bounded capacity
continuity identify both limits with $\int_V\phi(Du)^n$.
Maximality of $\mathcal F$ gives
\[
 \mathcal E(P_t)-\mathcal E(U)
       \leq\int_H\Delta_t\,d\mu
       \leq t\int_H\phi\,d\mu.
\]
Thus $(Du)^n\leq\mu$ on all nonnegative original scalar tests.
The coefficient-Radon identification turns this into the claimed
measure inequality. All perturbations have support inside $V$;
no boundary mass is discarded in these calculations.
\end{proof}

\begin{lemma}[An estimate independent of the lower obstacle]
\label{lem:r3-capdir-apriori}
Let $v\in\Cb(V)$ be bounded, nonpositive, with zero ordinary
boundary limit, and suppose $(Dv)^n\leq\mu$. For $M=\mu(V)>0$
put
\[
 S=\left(1+\frac2{1-2^{-\alpha}}\right)
                  (AM^\alpha)^{1/[n(1+\alpha)]}.
\]
Then $v\geq-S$. If its measure is carried by $H$, then with
$d_H=\min_H(-\rho)>0$ one has
\[
 v\geq\max\{-S,(S/d_H)\rho\}.
\]
\end{lemma}

\begin{proof}
For $s>0$ put $f(s)=C(\{v<-s\})^{1/n}$. These sets have compact
closure because of the ordinary boundary limit. Given a tester
$a\in\Cb(V)$, $0\leq a\leq1$, compare $v$ with
$w=-s-t+ta$. One has
\[
 \{v<-s-t\}\subset\{v<w\}\subset\{v<-s\}.
\]
The middle gap has compact closure. Comparison gives
\[
 t^n\int_{\{v<-s-t\}}(Da)^n
 \leq\int_{\{v<w\}}(Dw)^n
 \leq(Dv)^n(\{v<w\})\leq\mu(\{v<-s\}).
\]
Taking the supremum and using the measure bound on
$H\cap\{v<-s\}$ yields
\[
 t f(s+t)\leq A^{1/n} f(s)^{1+\alpha}.
\]
The same estimate without the capacity bound, followed by
$s\downarrow0$ and increasing-set continuity of raw capacity,
gives $t f(t)\leq M^{1/n}$.

Set $T=(AM^\alpha)^{1/[n(1+\alpha)]}$ and start with $s_0=T$.
As long as $f(s_j)>0$, take
$t_j=2A^{1/n}f(s_j)^\alpha$ and $s_{j+1}=s_j+t_j$.
Then $f(s_{j+1})\leq f(s_j)/2$, and the initial estimate gives
$t_j\leq2T2^{-j\alpha}$. Their sum is at most
$2T/(1-2^{-\alpha})$. If the iteration stops, all deeper
sublevels already have zero capacity; otherwise $f(S)=0$ by
monotonicity and the geometric decay. The strict sublevel
$\{v<-S\}$ is ordinary open because $v$ is usc. Ambient
capacity nondegeneracy on the fixed framed domain makes it empty.

The bounded classical barrier
$b=\max(-S,(S/d_H)\rho)$ equals $-S$ on $H$ and has zero
boundary limit. Thus $(Dv)^n(\{v<b\})=0$ when the measure is
carried by $H$. Pointwise domination, applied first with the
positive margin $b-\delta$ and then $\delta\downarrow0$,
gives $v\geq b$. No strict gap between two discontinuous
potentials has been called ordinary open.
\end{proof}

\begin{theorem}[Power-capacity Dirichlet existence]
\label{thm:r3-capacity-dirichlet}
Let $\mu$ be a finite positive Radon measure carried by a compact
$H\Subset V$, and assume the power bound in
Lemma~\ref{lem:r3-capdir-maximizer}. There is a unique bounded
$u\in\Cb(V)$ with zero ordinary boundary limit and
\[
 (Du)^n=\mu.
\]
For $M>0$ it satisfies
$\max(-S,(S/d_H)\rho)\leq u\leq0$, with $S$ as above.
For $M=0$ the solution is zero. All equalities hold on the full
original test sheaf. Continuity on $\overline V$ is proved later
by the one-sided Dini estimate.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
Assume $M>0$ and choose $B>S/d_H$. The constrained maximizer
exists, and its upper variation gives $(Du)^n\leq\mu$.
The preceding lemma gives $u\geq(S/d_H)\rho$. Therefore
\[
 u-B\rho\geq(B-S/d_H)(-\rho)>0
 \quad\text{at every point of }V.
\]
This is a uniform positive margin on every compact perturbation
support, and releases the artificial lower obstacle there.

Let $\phi\geq0$ be original compact smooth in $V$. For sufficiently
small $t>0$, the global lower metric satisfies
$L_B\leq U-t\phi$. Put $P_t=P_L^{\mathrm b}(U-t\phi)$.
Since $L_B$ is a competitor and $U-t\|\phi\|_\infty$ is also
a global classical minorant,
\[
 L_B\leq P_t\leq U-t\phi\leq U,
 \qquad U-t\|\phi\|_\infty\leq P_t.
\]
Outside $V$ the lower metric equals $U$, so $P_t=U$ there.
Thus $P_t$ is admissible. Write $\Delta_t=U-P_t$.
The obstacle orthogonality identity is now
\[
 \int\Delta_t c_1(L,P_t)^n
       =t\int\phi c_1(L,P_t)^n.
\]
Since $\Delta_t\geq t\phi$, concavity gives
\[
 \int\phi c_1(L,U)^n
 \leq\frac{\mathcal E(U)-\mathcal E(P_t)}t
 \leq\int\phi c_1(L,P_t)^n.
\]
Uniform convergence again shows that the middle expression tends
to $\int_V\phi(Du)^n$. Maximality gives the opposite measure
inequality:
\[
 \mathcal E(U)-\mathcal E(P_t)
       \geq\int_H\Delta_t\,d\mu
       \geq t\int_H\phi\,d\mu.
\]
Hence $(Du)^n\geq\mu$, proving equality. The original sandwich
$B\rho\leq u\leq0$ gives the ordinary zero boundary limit,
and the a priori lemma gives the stated sharper bounds.

If $M=0$, take $u=0$. For uniqueness, if $u,v$ have the same
measure and zero boundary values, compare $v$ with
$(1+\epsilon)u-\delta$ on their compact strict gap. The
comparison inequality and the common measure force that gap to
have zero measure; ambient pointwise domination makes it empty.
Let $\delta\downarrow0$ and then $\epsilon\downarrow0$ and
repeat in the other order. All cancellations occur on compact
gaps of finite mass. This proves uniqueness without a weak-data
stability assumption.
\end{proof}

% END source: base_r3/sections/capacity_dirichlet.tex


% BEGIN frozen or reviewed source: base_r3/sections/dini_capacity_dirichlet.tex
\section{Dini capacity moduli for the fixed-domain Dirichlet problem}
\label{sec:r3-dini-capacity-dirichlet}

Retain the extendible domain $(Y,\rho,V)$ and the original ambient
background $[1]$ of Section~\ref{sec:r3-capacity-dirichlet}. Write
$C(E)=\operatorname{Cap}_{[1]}(E;V)$. All functions retain their
actual classical values over the prescribed field.

\begin{definition}[Dini capacity modulus]
\label{def:r3-dini-capacity-modulus}
A finite nondecreasing function $\omega:[0,\infty)\to[0,\infty)$
with $\omega(0)=0$ is a Dini capacity modulus if
\begin{equation}\label{eq:r3-dini-integrability}
 \int_0^a\omega(t)\,\frac{dt}{t}<\infty
\end{equation}
for some $a>0$. No continuity or strict positivity is imposed.
\end{definition}

Monotonicity implies measurability. The condition forces
$\omega(t)\to0$ at zero, and then holds for every finite positive
upper endpoint, since the integrand is bounded on compact intervals
away from zero.

\begin{theorem}[Dirichlet solvability under a Dini modulus]
\label{thm:r3-dini-capacity-dirichlet}
Let $H\Subset V$ be compact, and let $\mu$ be a finite positive
Radon measure carried by $H$. Suppose
\begin{equation}\label{eq:r3-dini-capacity-hypothesis}
 \mu(E)\leq C(E)\omega(C(E)^{1/n})^n
       \qquad(E\subset H\text{ Borel})
\end{equation}
for a Dini capacity modulus. There is a unique bounded
$u\in\Cb(V)$ with ordinary boundary limit zero and $(Du)^n=\mu$.
For $M=\mu(V)>0$, put
\begin{equation}\label{eq:r3-dini-S}
 S_\omega(s)=s+2\sum_{j\geq0}\omega(2^{-j}M^{1/n}/s),
 \qquad S_* =\inf_{s>0}S_\omega(s).
\end{equation}
Each series is finite. With $d_H=\min_H(-\rho)$, one has
\begin{equation}\label{eq:r3-dini-solution-bound}
 \max\{-S_*,(S_*/d_H)\rho\}\leq u\leq0.
\end{equation}
For $M=0$, $u=0$. The infimum defining $S_*$ need not be attained.
\end{theorem}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
preceding variational construction, with the scalar iteration below.
Continuity is established separately in
Theorem~\ref{thm:r3-dini-solution-continuity}.

\begin{lemma}[Dini sublevel iteration]
\label{lem:r3-dini-sublevel-iteration}
If $v\in\Cb(V)$ is bounded, nonpositive, has zero boundary limit,
and $(Dv)^n\leq\mu$, then $\|v\|_\infty\leq S_\omega(s)$ for
every $s>0$. If its measure is carried by $H$, it also satisfies
$v\geq\max(-S_\omega(s),(S_\omega(s)/d_H)\rho)$.
\end{lemma}

\begin{proof}
Set $f(s)=C(\{v<-s\})^{1/n}$. The compact-gap comparison used in
Lemma~\ref{lem:r3-capdir-apriori}, with normalized testers on this
same domain, gives
\begin{equation}\label{eq:r3-dini-recurrence}
 t f(s+t)\leq f(s)\omega(f(s)),\qquad
 t f(t)\leq M^{1/n}\quad(s,t>0).
\end{equation}
For the first inequality apply the measure bound on
$H\cap\{v<-s\}$ and use monotonicity of
$c\mapsto c\omega(c^{1/n})^n$. The second follows by increasing
continuity of raw capacity as the shift $s$ decreases to zero.

Starting at $s_0>0$, while $f(s_j)>0$, take
$t_j=2\omega(f(s_j))$ and $s_{j+1}=s_j+t_j$. If
$\omega(f(s_j))=0$, the first inequality makes $f(s_j+t)=0$
for every $t>0$. Increasing-set continuity then gives $f(s_j)=0$,
so the iteration has already stopped. Otherwise
\[
 f(s_{j+1})\leq\tfrac12 f(s_j),\qquad
 t_j\leq2\omega(2^{-j}M^{1/n}/s_0).
\]
The sum is finite because
\begin{equation}\label{eq:r3-dini-dyadic-bound}
 \sum_{j\geq0}\omega(2^{-j}a)
 \leq\omega(a)+\frac1{\log2}\int_0^a\omega(t)\,\frac{dt}{t}.
\end{equation}
Indeed the integral over $[2^{-j}a,2^{-j+1}a]$ is at least
$(\log2)\omega(2^{-j}a)$ for $j\geq1$. Monotonicity of $f$
now gives $f(S_\omega(s_0))=0$, whether the iteration terminates
or continues indefinitely. The strict sublevel
$\{v<-S_\omega(s_0)\}$ is an ordinary open set, since $v$ is usc.
Ambient capacity nondegeneracy makes it empty. The barrier
$b=\max(-S_\omega(s_0),(S_\omega(s_0)/d_H)\rho)$ equals
$-S_\omega(s_0)$ on $H$. Thus $v\geq b$ on the carrying set
of its measure, and compact-gap pointwise domination gives
$v\geq b$ throughout $V$.
\end{proof}

\begin{proof}[Proof of the theorem]
For $M>0$, fix $s_0>0$ and choose $B>S_\omega(s_0)/d_H$.
Use the class $\mathcal F_B$, its common affine collar and energy
from Lemma~\ref{lem:r3-capdir-affine-collar}. The power bound in
the constrained-maximizer argument was used for continuity of
integration against $\mu$ under bounded capacity convergence on
$H$. Here, if $|v_j|,|v|\leq L$ and
$c_j=C(H\cap\{|v_j-v|>\delta\})$, then
\[
 \int_H|v_j-v|\,d\mu
 \leq\delta M+2L c_j\omega(c_j^{1/n})^n.
\]
The last term tends to zero when $c_j\to0$, with value zero at
$c_j=0$. The same midpoint argument and ambient representative
therefore give a maximizer of the energy minus $\int v\,d\mu$.
The test $tB\rho$ again proves that it is not the zero function,
and the classical strong maximum principle makes it strictly negative.

The positive obstacle variation of
Lemma~\ref{lem:r3-capdir-upper-variation} yields $(Du_B)^n\leq\mu$.
The preceding lemma makes $u_B\geq(S_\omega(s_0)/d_H)\rho$.
The strict choice of $B$ leaves a positive lower-obstacle margin
on each compact perturbation support. The negative obstacle
variation in Theorem~\ref{thm:r3-capacity-dirichlet} therefore
applies without any change of domain or envelope input and gives
$(Du_B)^n\geq\mu$. These variations use the already proved ambient
quasicontinuous-obstacle orthogonality, not an arbitrary-relative
orthogonality assertion. Ambient pointwise comparison gives uniqueness.

Apply the a priori lemma to this one solution for every $s>0$.
Infimizing gives $u\geq-S_*$. Along any sequence of parameters
with $S_\omega(s)\to S_*$, the barrier inequalities give the
other bound in \eqref{eq:r3-dini-solution-bound}. No minimum is needed.
\end{proof}

\begin{remark}
For $\omega(t)=\mathbf1_{[1,\infty)}(t)$ and $M=1$,
$S_\omega(s)=s$ when $s>1$, whereas $S_\omega(s)\geq s+2$
when $0<s\leq1$. Thus its infimum is $1$ and is not attained.
\end{remark}

\begin{corollary}[Power specialization]
\label{cor:r3-dini-power-specialization}
For $\omega(t)=A^{1/n}t^\alpha$, $A,\alpha>0$, the measure
condition is $\mu(E)\leq A C(E)^{1+\alpha}$. For $M>0$,
\[
 \|u\|_\infty\leq
 (1+\alpha)\alpha^{-\alpha/(1+\alpha)}
 \left(\frac2{1-2^{-\alpha}}\right)^{1/(1+\alpha)}
 (AM^\alpha)^{1/[n(1+\alpha)]}.
\]
\end{corollary}
\begin{proof}
Sum the geometric series to obtain $S_\omega(s)=s+Qs^{-\alpha}$,
where $Q=2A^{1/n}M^{\alpha/n}/(1-2^{-\alpha})$, and minimize at
$s=(\alpha Q)^{1/(1+\alpha)}$.
\end{proof}

\begin{corollary}[Logarithmic modulus]
\label{cor:r3-dini-logarithmic}
For $L>0$ and $\beta>1$, the sufficient condition is
\begin{equation}\label{eq:r3-dini-logarithmic-hypothesis}
 \mu(E)\leq
 \frac{L^n C(E)}{[\log(e+C(E)^{-1/n})]^{n\beta}}.
\end{equation}
For $M>0$, $s\geq eM^{1/n}$ and $\ell=\log(s/M^{1/n})$,
\begin{equation}\label{eq:r3-dini-log-bound}
 \|u\|_\infty\leq
 s+2L\left(\ell^{-\beta}
       +\frac{\ell^{1-\beta}}{(\beta-1)\log2}\right).
\end{equation}
\end{corollary}
\begin{proof}
Use $\omega(t)=L[\log(e+1/t)]^{-\beta}$ for $t>0$ and
$\omega(0)=0$. Its Dini integral converges precisely for $\beta>1$.
For $a\leq e^{-1}$, bound $\omega(a)$ by
$L(\log(1/a))^{-\beta}$ and the Dini integral by
$L(\log(1/a))^{1-\beta}/(\beta-1)$. Insert these bounds into
\eqref{eq:r3-dini-dyadic-bound} with $a=M^{1/n}/s$.
\end{proof}

It is enough to prescribe the modulus and the measure condition
at capacities with $C(E)^{1/n}<a$. Extend it for $t\geq a$ by
$\max\{\omega(a),M^{1/n}/a\}$. The remaining sets have mass at
most $M$, so the extended condition holds. This does not assert
that every individual datum under a non-Dini bound is unsolvable;
the separate radial construction concerns a uniform guarantee.

% END source: base_r3/sections/dini_capacity_dirichlet.tex


% BEGIN frozen or reviewed source: base_r3/sections/dini_stability_continuity.tex
\section{One-sided capacity stability and continuity of Dini solutions}
\label{sec:r3-dini-stability-continuity}

Here the background is $[1]$. All functions are actual bounded
classical psh functions over the fixed original field, and all
currents use the original CLD tests. On a relatively compact ordinary
open $V$ in a permitted smooth affine neighborhood, write
$C(E)=\operatorname{Cap}_{[1]}(E;V)$. For a Dini modulus put
\[
 \Phi_\omega(b)=2\sum_{j\geq0}\omega(2^{-j}b),\qquad
 \Phi_\omega(0)=0.
\]
The dyadic integral comparison proves
\begin{equation}\label{eq:r3-stability-dyadic-tail}
 \Phi_\omega(b)\leq2\omega(b)+\frac2{\log2}
          \int_0^b\omega(r)\,\frac{dr}{r},\qquad
 \Phi_\omega(b)\longrightarrow0\quad(b\downarrow0).
\end{equation}

\begin{theorem}[One-sided support-capacity estimate]
\label{thm:r3-one-sided-dini-stability}
Let $u,v\in\Cb(V)$ be bounded and suppose
\[
 \overline{\{u<v-\delta\}}\Subset V\qquad(\delta>0).
\]
Suppose $\mu=(Du)^n$ is carried by a compact $H\Subset V$ and
\begin{equation}\label{eq:r3-stability-dini-measure}
 \mu(E)\leq C(E)\omega(C(E)^{1/n})^n
             \quad(E\subset H\text{ Borel}).
\end{equation}
Then for every $s>0$,
\begin{equation}\label{eq:r3-one-sided-dini-bound}
 \sup_V(v-u)\leq s+\Phi_\omega\bigl(
                  C(H\cap\{u<v-s\})^{1/n}\bigr).
\end{equation}
In particular, if $v_j$ satisfies the same gap condition and
converges to $u$ in outer capacity on $H$, with capacity domain
$V$, then $\limsup_j\sup_V(v_j-u)\leq0$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
Put $E_s=\{u<v-s\}$ and $f(s)=C(H\cap E_s)^{1/n}$.
It is finite by CLN on $H$. For any normalized classical tester
$0\leq b\leq1$, set $w=v-s-t+tb$. Then
\[
 E_{s+t}\subset\{u<w\}\subset E_s.
\]
The middle set has compact closure. Positivity of the mixed
expansion and compact-gap comparison give
\[
 t^n\int_{E_{s+t}}(Db)^n
 \leq\int_{\{u<w\}}(Dw)^n
 \leq\mu(\{u<w\})\leq\mu(H\cap E_s).
\]
Taking the supremum over $b$ gives the stronger full-gap bound
\begin{equation}\label{eq:r3-stability-recurrence}
 t^n C(E_{s+t})\leq f(s)^n\omega(f(s))^n,
 \qquad t f(s+t)\leq f(s)\omega(f(s)).
\end{equation}
Starting at $s$, repeat the halving iteration of
Lemma~\ref{lem:r3-dini-sublevel-iteration}, with increments
$2\omega(f(s_j))$. The total increment is at most
$\Phi_\omega(f(s))$. A zero increment with $f(s_j)>0$ is
impossible: the recurrence and increasing-set continuity give
$f(s_j)=0$ in that case. Hence, for
$L=s+\Phi_\omega(f(s))$, one has $f(L)=0$.

The measure condition gives $\mu(E_L)=\mu(H\cap E_L)=0$.
The strict gap $E_L$ is only Borel; it is not called ordinary
open. Apply the proved ambient pointwise domination theorem to
$u$ and $v-L$. Its gap has compact closure, and the measure of
$u$ gives it no mass, so $u\geq v-L$ at every point. This proves
\eqref{eq:r3-one-sided-dini-bound}. The convergence statement
follows first for any fixed $s>0$ from
\eqref{eq:r3-stability-dyadic-tail}, and then by $s\downarrow0$.
\end{proof}

\begin{theorem}[Continuity on the ordinary closure]
\label{thm:r3-dini-solution-continuity}
On the extendible exhaustion domains of
Theorem~\ref{thm:r3-dini-capacity-dirichlet}, its solution extends
to a continuous function on $\overline V$ with value zero on
$\partial V$. In particular this holds for power-capacity data.
\end{theorem}

\begin{proof}
The proved compact-support barrier gives $u\geq C\rho$ for a
finite $C$. The common-exterior extension lemma produces a
classical function $U$ on a neighborhood of $\overline V$,
equal to $u$ on $V$ and to $C\rho$ on the exterior. The function
is zero on the ordinary boundary. This is a bounded classical
extension, not yet an assertion of interior continuity.

B's regularization provides continuous classical $w_j\downarrow U$
on one neighborhood of $\overline V$, with a common bound there.
Since $U=0$ on the compact boundary, Dini's lemma applied only to
the continuous restrictions on $\partial V$ gives
\[
 \epsilon_j=\max_{\partial V}w_j\downarrow0.
\]
Put $v_j=w_j-\epsilon_j$ on $V$. The boundary upper limit of
$v_j-u$ is at most zero. Thus each positive-margin gap has compact
closure. Decreasing relative capacity convergence of the $w_j$,
and the fixed-domain comparison on the compact $H$, give
$v_j\to u$ in outer ambient capacity there. The preceding theorem
implies $\limsup_j\sup_V(v_j-u)\leq0$.
Since $w_j\geq u$, it follows that $w_j\to u$ uniformly on $V$.
On the boundary the errors are at most $\epsilon_j$. Hence the
continuous $w_j|_{\overline V}$ converge uniformly to the extension
of $u$ by zero, proving continuity on the whole ordinary closure.
\end{proof}

\begin{theorem}[Strong-data uniform stability for power growth]
\label{thm:r3-power-dual-uniform-stability}
Fix an extendible $(Y,\rho,V)$, a compact $H\Subset V$, $B>0$,
and $A,\alpha>0$. Suppose $u,v\in\Cb(V)$ have ordinary boundary
limit zero, $-B\leq u,v\leq0$, and their measures $\mu,\nu$
are carried by $H$ and obey
\[
 \mu(E),\nu(E)\leq A C(E)^{1+\alpha}
             \qquad(E\subset H\text{ Borel}).
\]
With
\[
 d=d_{\mathrm{psh},V}(\mu,\nu)
   =\sup_{\substack{b\in\Cb(V)\\0\leq b\leq1}}
           \left|\int b\,d(\mu-\nu)\right|,
\]
there is a finite constant depending only on the fixed data and
the specified collar and finite CLN controls such that
\begin{equation}\label{eq:r3-power-dual-uniform-bound}
 \|u-v\|_\infty\leq C_0
             d^{\alpha/[2^n(n+2\alpha)]}.
\end{equation}
No optimality of this exponent is asserted.
\end{theorem}

\begin{proof}
The common compact support and bound give, by
Lemma~\ref{lem:r3-common-exterior-extension}, one larger domain
$\Omega$ and classical extensions $U,V'$ agreeing with a fixed
collar function outside $V$. Their mixed energy is
\[
 I=\int_H(u-v)\,d(\nu-\mu)\leq2B d.
\]
Indeed $(u+B)/B$ and $(v+B)/B$ are normalized testers on $V$;
the two constant terms cancel in their difference. The extensions
have one bound on the chosen larger compact stage. Apply
Theorem~\ref{thm:r3-I-capacity}, then reverse domain comparison
only on $H\Subset V$, to obtain
\begin{equation}\label{eq:r3-dual-gap-capacity}
 C(H\cap\{|u-v|>s\})\leq C_1s^{-2}d^{2^{-n}}.
\end{equation}
The reverse comparison has been used on a compact core, not at
the boundary of $V$.

For the power modulus, $\Phi_\omega(b)=
2A^{1/n}b^\alpha/(1-2^{-\alpha})$. Apply the one-sided estimate
in both orders and use \eqref{eq:r3-dual-gap-capacity}:
\[
 \|u-v\|_\infty\leq
 s+C_2s^{-2\alpha/n}d^{\alpha/(n2^n)}.
\]
For $0<d\leq1$ take $s=d^{\alpha/[2^n(n+2\alpha)]}$.
For $d=0$ let $s\downarrow0$ in the gap estimate and domination,
or use that the dual distance determines compact measures and
ambient uniqueness. For $d>1$, the uniform bound $2B$ is absorbed
in the constant. This proves the claimed exponent.
\end{proof}

\begin{corollary}[Strong-data stability under a common Dini modulus]
\label{cor:r3-dini-dual-uniform-stability}
Under the same domain, compact support and uniform potential bound,
replace the two power bounds by one common Dini bound. If
$d_{\mathrm{psh},V}(\mu_j,\mu)\to0$, the corresponding bounded
zero-boundary solutions converge uniformly on $\overline V$.
\end{corollary}
\begin{proof}
The energy argument still gives
\eqref{eq:r3-dual-gap-capacity}. For a fixed $s>0$, the capacity
on the right tends to zero. Apply the two one-sided Dini estimates
and then \eqref{eq:r3-stability-dyadic-tail}; finally let $s$ decrease
to zero. The boundary values agree. This is a
\texttt{DIRECT\_COROLLARY}, not a weak-measure stability theorem.
\end{proof}

% END source: base_r3/sections/dini_stability_continuity.tex


% BEGIN frozen or reviewed source: base_r3/sections/maximal_boundary_dirichlet.tex
\section{Nonconstant boundary data with a continuous maximal lift}
\label{sec:r3-maximal-boundary-dirichlet}

Keep the original field and the extendible domain $(Y,\rho,V)$
of Section~\ref{sec:r3-capacity-dirichlet}. Let $h$ be a continuous
classical psh function on all of $Y^{\an}$ and suppose
\[
 (Dh)^n=0\quad\text{on }V.
\]
Write $g=h|_{\partial V}$. This is a hypothesis on a specified
classical lift of the boundary function; it is not an assertion
that every continuous boundary function admits such a lift.

\begin{theorem}[Dini data with a continuous maximal boundary lift]
\label{thm:r3-maximal-boundary-dirichlet}
Let $K_0\Subset V$ be compact and let $\mu$ be a finite positive
Radon measure carried by $K_0$. Suppose, for a Dini modulus
$\omega$ as in Definition~\ref{def:r3-dini-capacity-modulus}, that
\[
 \mu(E)\leq C(E)\omega(C(E)^{1/n})^n,
 \qquad E\subset K_0\text{ Borel},\qquad
 C(E)=\operatorname{Cap}_{[1]}(E;V).
\]
There is a unique bounded classical function $u$ on $V$,
continuous on $\overline V$, with boundary value $g$ and
$(Du)^n=\mu$. If $M=\mu(V)>0$, set
\[
 d=\min_{K_0}(-\rho),\qquad
 S_\omega(s)=s+2\sum_{j\geq0}\omega(2^{-j}M^{1/n}/s),
 \qquad S_* =\inf_{s>0}S_\omega(s).
\]
Then the actual pointwise bounds are
\[
 h+\max\{-S_*,(S_*/d)\rho\}\leq u\leq h.
\]
If $M=0$, the solution is $u=h$. The uniqueness assertion holds
among all bounded classical solutions with the stated ordinary
boundary limits, without an assumed interior continuity.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for this explicit
nonconstant-boundary extension. The mixed calculus, contact
inequality, projective obstacle theorems and scalar Dini iteration
are previously established inputs.

\begin{proof}
Assume $M>0$. Fix $s_0>0$, put $S=S_\omega(s_0)$, and choose
$B>S/d$. The constrained family is
\[
 \mathcal F_B^h=\{v\in\Cb(V):h+B\rho\leq v\leq h\}.
\]
Its members already have the ordinary boundary limit $g$.

\emph{The common affine extension.}
For $\epsilon>0$ glue $\max(v,h+B\rho+\epsilon)$ on $V$
to $h+B\rho+\epsilon$ off $V$. Near each boundary point the
second branch is larger than $h$, hence larger than $v$.
The prescriptions agree on an ordinary neighborhood, and their
decreasing limit is classical, equal to $v$ on $V$ and to
$h+B\rho$ outside. All members agree outside $V$ and have one
bound on every chosen surrounding compact box. The proof of
Lemma~\ref{lem:r3-capdir-affine-collar} therefore gives a single
ample line on a smooth projective compactification, and metrics
$U_v$ with weight $v+c$ on a neighborhood of $\overline V$.
The map remains affine, since its maximum collar is inactive
on $V$ and all inputs are identical outside $V$. Its continuous
lower and upper metrics $L_B,F$ have weights $h+B\rho+c$ and
$h+c$ on $V$, and agree off $V$. Use the global mixed energy
$\mathcal E_F$ relative to $F$.

\emph{Existence of a constrained maximizer.}
Maximize
\[
 \mathcal F(U_v)=\mathcal E_F(U_v)-\int_{K_0}v\,d\mu.
\]
The common bounds and projective intersection mass bound this
functional. The midpoint argument of
Lemma~\ref{lem:r3-capdir-maximizer} applies without a change:
the integral is affine, the midpoint energy gap controls $I$,
and the prescribed-box estimate followed by reverse framed domain
comparison makes a maximizing sequence locally capacity Cauchy
on every permitted local domain. The ambient completion has a
unique actual classical representative. The continuous upper and
lower metrics pass through its tail-usc formula, so this limit
is admissible and equals the common exterior metric off $V$.
Weighted capacity continuity gives continuity of the energy.
For the measure term, if $|v_j|,|v|\leq L$ on $K_0$ and
$c_j=C^*(K_0\cap\{|v_j-v|>\delta\})$, then
\[
 \int_{K_0}|v_j-v|\,d\mu
 \leq\delta M+2L c_j\omega(c_j^{1/n})^n\longrightarrow\delta M.
\]
Letting $\delta\downarrow0$ proves the required continuity.
Thus a constrained maximizer $U$ exists, with local function $u$.

\emph{The positive variation and the contact set.}
The set $O=\{u<h\}$ is ordinary open because $u-h$ is usc and
$h$ is continuous. For a nonnegative original smooth $\phi$
with compact support in $O$, choose $t>0$ small enough that
$u+t\phi\leq h$ there. Extend $\phi$ by zero to the projective
space and set $P_t=P_L^{\mathrm b}(U+t\phi)$. The actual
usc-obstacle envelope satisfies $U\leq P_t\leq U+t\phi$, is
unchanged off the perturbation support, and is admissible.
Orthogonality and energy concavity give, exactly with these
global metrics,
\[
 \int\phi c_1(L,P_t)^n
 \leq\frac{\mathcal E_F(P_t)-\mathcal E_F(U)}t
 \leq\int\phi c_1(L,U)^n.
\]
Uniform convergence of $P_t$ to $U$ gives the common limit
$\int\phi(Du)^n$. Maximality bounds the middle expression by
$\int\phi\,d\mu$. Thus $(Du)^n\leq\mu$ on $O$.

On the Borel contact set $A=\{u=h\}$ the proved contact
inequality, Corollary~\ref{cor:r3-relative-zero-contact-mass},
gives
\[
 \mathbf1_A(Du)^n\leq\mathbf1_A(Dh)^n=0.
\]
Combining the two disjoint pieces yields $(Du)^n\leq\mu$
on all of $V$. In particular this measure is carried by $K_0$.
There is no use of a strong maximum principle for $u-h$, which
is not asserted to be psh.

\emph{The estimate for the difference without differentiating it.}
Put $f(s)=C(\{u<h-s\})^{1/n}$. For a normalized classical
tester $0\leq a\leq1$ on $V$, compare $u$ with the classical
function
\[
 w=h-s-t+ta.
\]
The mixed expansion is positive and gives
$(Dw)^n\geq t^n(Da)^n$. Moreover
\[
 \{u<h-s-t\}\subset\{u<w\}\subset\{u<h-s\}.
\]
All positive-margin gaps have compact closure by the common
boundary values. Comparison and $(Du)^n\leq\mu$ therefore
give
\[
 t f(s+t)\leq f(s)\omega(f(s)),\qquad
 t f(t)\leq M^{1/n}.
\]
The second estimate follows by decreasing the auxiliary positive
shift to zero and using increasing-set continuity of raw capacity.
Apply the scalar halving argument from
Lemma~\ref{lem:r3-dini-sublevel-iteration}: starting at $s_0$,
the increments $2\omega(f(s_j))$ make $f$ decrease by factors
of two, and their sum is bounded by
$2\sum_{j\geq0}\omega(2^{-j}M^{1/n}/s_0)$. If the modulus
vanishes at a positive value encountered, the recurrence makes
all deeper capacities zero and the iteration stops. In either
case $f(S)=0$. The set $\{u<h-S\}$ is ordinary open, so
ambient capacity nondegeneracy makes it empty. Thus $u\geq h-S$.
This argument has used the psh functions $u,h,w,a$ themselves;
it has not treated $u-h$ as a psh potential.

The classical barrier
\[
 b=h+\max\{-S,(S/d)\rho\}
\]
equals $h-S$ on $K_0$, so $b\leq u$ there. Since $(Du)^n$
is carried by $K_0$, the strict gap $\{u<b\}$ has zero
$(Du)^n$-mass. Pointwise domination applied first to $b-\delta$
and then with $\delta\downarrow0$ gives $b\leq u$ everywhere.
In particular
\[
 u-(h+B\rho)\geq(B-S/d)(-\rho)>0\quad\text{on }V.
\]

\emph{Release of the lower obstacle.}
For an arbitrary nonnegative original compact smooth $\phi$ in
$V$, this strict margin permits $L_B\leq U-t\phi$ for all
sufficiently small $t>0$. Let $P_t=P_L^{\mathrm b}(U-t\phi)$.
The lower metric and the global constant shift give
\[
 L_B\leq P_t\leq U-t\phi\leq U,
 \qquad U-t\|\phi\|_\infty\leq P_t.
\]
Outside $V$ the lower metric equals $U$, so $P_t=U$ there and
$P_t$ is admissible. Orthogonality and concavity, now in the
negative direction, give
\[
 \int\phi c_1(L,U)^n
 \leq\frac{\mathcal E_F(U)-\mathcal E_F(P_t)}t
 \leq\int\phi c_1(L,P_t)^n.
\]
For clarity, orthogonality reads
$\int(U-P_t)c_1(L,P_t)^n=t\int\phi c_1(L,P_t)^n$;
the lower inequality uses $U-P_t\geq t\phi$.
The common limit is $\int\phi(Du)^n$. Maximality gives
\[
 \mathcal E_F(U)-\mathcal E_F(P_t)
       \geq\int_{K_0}(U-P_t)\,d\mu
       \geq t\int_{K_0}\phi\,d\mu.
\]
Thus $(Du)^n\geq\mu$, and equality follows. All these energy
differences are global expressions on one line; no boundary mass
is inferred to vanish merely by restriction.

The construction gives one solution. Repeat only the a priori
estimate for that solution for each $s>0$ and take a sequence
with $S_\omega(s)\to S_*$. This gives the two displayed bounds
with $S_*$; no minimizing parameter is required.

\emph{Continuity and uniqueness.}
The constructed global metric admits bounded continuous classical
decreasing regularizations on one neighborhood of $\overline V$
by B. In the fixed frame write their local functions as
$w_j\downarrow u$ on $V$; on the boundary their limit is $h$.
Compactness of the boundary and Dini's lemma give
\[
 \epsilon_j=\max_{\partial V}(w_j-h)\downarrow0.
\]
The functions $v_j=w_j-\epsilon_j$ have boundary upper limit
at most $g$. Their positive-margin gaps above $u$ have compact
closure. Decreasing capacity convergence and domain comparison
give $v_j\to u$ in outer capacity on $K_0$, with domain $V$.
Theorem~\ref{thm:r3-one-sided-dini-stability} applies; its gap
hypothesis does not require constant boundary values. It gives
$\limsup_j\sup_V(v_j-u)\leq0$. Since $w_j\geq u$ and
$\epsilon_j\to0$, the $w_j$ converge uniformly on $V$, and
also on its boundary, to the extension of $u$ by $g$. This is
the required continuity on $\overline V$.

If two bounded classical solutions have the same ordinary
boundary limits $g$, the gaps of one above the other with any
positive margin have compact closure. Compact-gap comparison
and ambient pointwise domination give equality. One can verify
this with the scaling argument after subtracting one sufficiently
large constant from both functions: their boundary values then
are strictly negative, so the comparison of $b$ with
$(1+\epsilon)a-\delta$ has compact gap. Equality of their
measures and their finite mass on that compact gap force the
gap to have zero measure, and domination removes it. Let
$\delta,\epsilon\downarrow0$ and interchange $a,b$.
For $M=0$, $h$ itself solves the problem and the same uniqueness
argument applies. This completes the proof.
\end{proof}

\begin{corollary}[An upper lift with curvature dominated by the datum]
\label{cor:r3-dominated-curvature-boundary-lift}
The preceding theorem remains valid if the assumption $(Dh)^n=0$
is replaced by
\[
 (Dh)^n\leq\mu\quad\text{on }V.
\]
All other hypotheses, the ordinary boundary values, continuity,
uniqueness and estimates are unchanged.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
The contact inequality now gives
$\mathbf1_A(Du)^n\leq\mathbf1_A(Dh)^n\leq\mathbf1_A\mu$.
Combined with the positive variation on $\{u<h\}$ it still
gives $(Du)^n\leq\mu$ everywhere. Every subsequent step uses
that inequality, the positivity of mixed terms, and the fixed
boundary lift, and is unchanged. If $\mu=0$, the new hypothesis
also forces $(Dh)^n=0$. The direction of the curvature inequality
is substantive: this is an upper lift of smaller curvature than
the prescribed datum, not an arbitrary continuous extension.
\end{proof}

\begin{example}[A nonconstant maximal lift in every dimension at least two]
On $Y=\mathbf A_K^n$, $n\geq2$, take
$h=\max(0,\log|T_1|)$ and
\[
 \rho=\max\{-1,\log|T_1|-\log R,\ldots,
                       \log|T_n|-\log R\},\qquad R>1.
\]
These are continuous classical psh functions, and
$V=\{|T_i|<R\text{ for every }i\}$ is an extendible domain
of the required type. The curvature $Dh$ is nonzero, but its
top power is zero because $h$ depends on one coordinate only
and $n\geq2$. Its boundary restriction is nonconstant: it is
zero at a boundary point with $T_1=0$ and $|T_2|=R$, and is
$\log R$ where $|T_1|=R$. Thus the theorem includes explicit
nonconstant boundary data with a nonconstant classical lift.
\end{example}

% END source: base_r3/sections/maximal_boundary_dirichlet.tex


% BEGIN frozen or reviewed source: base_r3/sections/subsolution_exhaustion.tex
\section{Exhaustion of subsolution-dominated data}
\label{sec:r3-subsolution-exhaustion}

Keep the extendible domain $(Y,\rho,V)$ of
Section~\ref{sec:r3-capacity-dirichlet}, the original field and
absolute value, and the ambient background $[1]$. Boundary limits
in this section are ordinary boundary limits at every point of
$\partial V$. The measures need only be locally finite.

\begin{theorem}[The bounded subsolution criterion without compact support]
\label{thm:r3-exhausted-subsolution-dirichlet}
Let $\mu$ be a locally finite positive Radon measure on $V$.
There exists a bounded $u\in\Cb(V)$ with zero ordinary boundary
limit and $(Du)^n=\mu$ if and only if there exists a bounded
$w\in\Cb(V)$ with zero ordinary boundary limit and
\[
 (Dw)^n\geq\mu.
\]
The solution is unique and satisfies $w\leq u\leq0$ for each
such subsolution $w$. The measure $\mu$ is not required to have
finite total mass or compact support. A constant boundary value
is handled by adding that constant to the solution and subsolution.
\end{theorem}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}
The inputs are the proved compact-data bounded-subsolution theorem,
the fixed-background decreasing BT theorem, and ambient pointwise
comparison. The matching boundary condition on $w$ is essential
to this exhaustion argument.

\begin{proof}
The reverse implication follows by taking $w=u$.
For the forward construction choose nested compact sets
$H_j\Subset V$ with $H_j\subset H_{j+1}^{\circ}$ and
$\bigcup_jH_j^{\circ}=V$, and put $\mu_j=\mathbf1_{H_j}\mu$.
These are finite compactly carried measures and $\mu_j\uparrow\mu$.
The same bounded $w$ is a subsolution for each $\mu_j$.
Theorem~\ref{thm:r3-bounded-subsolution-dirichlet} supplies bounded
classical $u_j$ with zero boundary limit and $(Du_j)^n=\mu_j$.
The weak-boundary order lemma
\ref{lem:r3-weak-boundary-subsolution-order} gives
\[
 w\leq u_{j+1}\leq u_j\leq0.
\]
Consequently $u_j$ decreases to an actual bounded classical
function $u$, with $w\leq u\leq0$. The boundary sandwich implies
that $u$ has zero limit at every ordinary boundary point. The
decreasing BT theorem gives $(Du_j)^n\to(Du)^n$ on every original
compact smooth scalar test. If such a test has support $H$, then
$H\subset H_j$ for every sufficiently large $j$, so its pairing
with $\mu_j$ equals its pairing with $\mu$. Thus $(Du)^n=\mu$.
No passage of total mass through the boundary is used.

For completeness, the comparison argument for uniqueness uses only
finite mass on compact gaps, even if $\mu(V)=\infty$.
Let $a,b$ be two bounded zero-boundary classical functions with
$(Da)^n\geq(Db)^n$. For $\epsilon,\delta>0$ the strict gap
\[
 E=\{b<(1+\epsilon)a-\delta\}
\]
has compact closure in $V$ by their boundary limits. The compact-gap
comparison theorem gives
\[
 (1+\epsilon)^n(Da)^n(E)\leq(Db)^n(E)\leq(Da)^n(E).
\]
All three quantities are finite on this compact gap, so both
measures vanish on $E$. Ambient pointwise domination, with lower
function $b$, implies $b\geq(1+\epsilon)a-\delta$ on $V$.
Let $\delta\downarrow0$ and then $\epsilon\downarrow0$ to obtain
$a\leq b$. Equality of the measures gives equality of the
functions by applying the argument in both orders. The same
argument verifies the order used in the exhaustion, and uses
strict gaps as Borel sets, without asserting that they are ordinary
open. Translation by a constant proves the final assertion.
\end{proof}

\begin{corollary}[Mixed data dominated by bounded potentials]
\label{cor:r3-mixed-datum-dirichlet}
Let $u_1,\ldots,u_n\in\Cb(V)$ be globally bounded and let
$f$ be a Borel function with $0\leq f\leq C<\infty$. Set
\[
 \mu=f\,Du_1\wedge\cdots\wedge Du_n.
\]
If $f$ is compactly supported, this measure has a unique bounded
zero-boundary classical solution. No boundary condition on the
$u_i$ is needed in this compactly supported case. If instead every
$u_i$ has zero ordinary boundary limit, the same conclusion holds
without compact support of $f$ and without a finite-total-mass
assumption.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
If $C=0$, take the zero solution. Otherwise put
\[
 w=\left(\frac{C}{n!}\right)^{1/n}\sum_{i=1}^n u_i.
\]
This is a bounded classical function. Positivity of all mixed
terms in the expansion and the coefficient $n!$ of the product
with all $n$ different factors give
\[
 (Dw)^n
  =\frac{C}{n!}\left(\sum_i Du_i\right)^n
  \geq C\,Du_1\wedge\cdots\wedge Du_n\geq\mu.
\]
The mixed measure is locally finite by the local CLN theorem, so
bounded Borel multiplication defines a locally finite Radon measure.
For compactly supported $f$, apply
Theorem~\ref{thm:r3-bounded-subsolution-dirichlet}, which requires
only a bounded subsolution in this case. If all the $u_i$ have
zero boundary limit, then so does $w$, and the preceding theorem
applies to the possibly noncompact data. This is an ambient
diagonal Dirichlet conclusion; it makes no arbitrary-background
relative Dirichlet existence assertion.
\end{proof}

% END source: base_r3/sections/subsolution_exhaustion.tex


% BEGIN frozen or reviewed source: base_r3/sections/capacity_growth_instability.tex
% Reconstructed after the scratch reset. Original reviewed SHA256:
% 988aa4a454cc77b8fefb752f2fcdc210c46e80b5eb1cb542d69f07af2a470f6a.
\section{Uniform power capacity growth does not imply weak-data stability}
\label{sec:r3-capacity-growth-instability}

Keep $\Omega,\lambda,g_i,H,z_i,z,K_*,\Omega_0$ from
Section~\ref{sec:r3-moving-atomic-dirichlet}. Write
$\mu_i=\delta_{z_i}$ and $\mu=\delta_z$.

\begin{corollary}[Weak instability under every fixed power bound]
\label{cor:r3-power-capacity-weak-instability}
For each fixed $\alpha>0$ there is one constant
$A_\alpha=(2\lambda)^{n(1+\alpha)}$ such that all Borel
$E\subset\Omega$ satisfy
\begin{equation}
 \mu_i(E),\ \mu(E)
       \leq A_\alpha\operatorname{Cap}_{[1]}(E;\Omega)^{1+\alpha}
       \quad\hbox{for every }i.
 \label{eq:r3-common-power-capacity-bound}
\end{equation}
Nevertheless the corresponding continuous bounded zero-boundary
solutions $g_i$ do not converge uniformly or locally in outer
capacity to $H$, and have no local capacity-Cauchy subsequence.
For $n\geq2$ they do not converge pointwise to $H$.
Moreover
\[
 d_{\mathrm{psh},\Omega}(\mu_i,\mu)\geq\tfrac12
 \quad\hbox{for every }i .
\]
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
explicit moving-atom construction. The assertion that weak data
convergence, common compact support, common zero boundary values,
and a uniform power capacity bound force capacity or uniform
solution convergence is \texttt{FALSE}.

\begin{proof}
The functions $b_i=1+g_i/(2\lambda)$ and
$b=1+H/(2\lambda)$ are normalized classical psh on $\Omega$,
because $B<2\lambda$. Their top measures are
$(2\lambda)^{-n}\delta_{z_i}$ and
$(2\lambda)^{-n}\delta_z$. Hence every set containing the
respective atom has capacity at least $(2\lambda)^{-n}$.
This proves \eqref{eq:r3-common-power-capacity-bound}; sets
without that atom have zero measure. The convention at infinite
capacity is harmless.

Let $0<\epsilon<2\lambda-B$. The set
$K_*\cap\{|g_i-H|>\epsilon\}$ contains $z_i$.
The same normalized tester restricted to the smaller ordinary
domain $\Omega_0$ shows that its raw, and thus its outer,
capacity on that fixed domain is at least $(2\lambda)^{-n}$.
This is failure of local all-domain capacity convergence, not
merely an assertion on the original large domain.
For $i\ne j$, $|g_i-g_j|(z_i)=\lambda$; a fixed threshold less
than $\lambda$ gives the same lower bound for every pair and
rules out a capacity-Cauchy subsequence. Uniform and, for
$n\geq2$, pointwise failures were explicitly computed in the
preceding theorem. Finally $b_i(z_i)=0$ and $b_i(z)=1/2$,
so this one normalized tester proves the psh-dual lower bound.
\end{proof}

This is consistent with the proved capacity-growth existence
theorems and their a priori bounds. It excludes ordinary weak
data convergence as a substitute for the strong psh-dual
topology in the corresponding stability assertion.

% END source: base_r3/sections/capacity_growth_instability.tex


% BEGIN frozen or reviewed source: base_r3/sections/linear_capacity_obstruction.tex
% Reconstructed after the scratch reset. Original reviewed SHA256:
% 21e71bd66b0528578d68192e12f48767d20d020f720e5d8bcd8ea5c555cf2248.
\section{Linear capacity domination does not ensure a bounded solution}
\label{sec:r3-linear-capacity-obstruction}

Fix $r=|\pi|$, $\lambda=-\log r>0$, $R=r^{-1}$, and the
ordinary polydisc $\Omega=\{|T_l|<R\}_{l=1}^n$. Set
\[
 t_j=2^j\lambda,\qquad r_j=R e^{-t_j}=r^{\,2^j-1},\qquad
 s=\max_l\log|T_l|-\log R .
\]
Let $z_j$ be the diagonal radius-$r_j$ Gauss point, let
$B_j=\{|T_l|\leq r_j\}_{l=1}^n$, and let $o$ be the rigid origin.
Thus $z_j\to o$. Write $K_j=\{o,z_j,z_{j+1},\ldots\}$.

\begin{lemma}[Exact radial compact capacities]
\label{lem:r3-radial-compact-capacity}
For every $j$,
\[
 U_j=\max(-1,s/t_j),\qquad
 (DU_j)^n=t_j^{-n}\delta_{z_j},
\]
and, on the same fixed domain $\Omega$,
\[
 \operatorname{Cap}_{[1]}(\{z_j\};\Omega)
 =\operatorname{Cap}_{[1]}(K_j;\Omega)
 =\operatorname{Cap}_{[1]}(B_j;\Omega)=t_j^{-n}.
\]
\end{lemma}

\begin{proof}
The ordinary torus slope simplex has volume $t_j^{-n}/n!$;
the original factor $n!$ gives the measure identity, with no
mass on lower-rank strata or coordinate zero loci. The function
$1+U_j$ is a normalized tester, giving the singleton lower bound.
For an arbitrary normalized $a\in\mathcal C_b(\Omega)$ and
$\epsilon>0$, compare $v=(1+\epsilon)U_j$ with $a-1$.
The compact set $B_j$ lies in $\{v<a-1\}$. For every
$\delta>0$ the closure of $\{v<a-1-\delta\}$ is compactly
contained in $\Omega$, since it lies in $\{v<-\delta\}$.
Thus the proved comparison principle gives
\[
 \int_{B_j}(Da)^n
 \leq\int_{\{v<a-1\}}(Da)^n
 \leq\int_{\{v<a-1\}}(Dv)^n
 \leq(1+\epsilon)^n t_j^{-n}.
\]
Let $\epsilon\downarrow0$, then take the supremum over $a$.
The inclusions $\{z_j\}\subset K_j\subset B_j$ give all three
equalities. No outer regularity interchange is used.
\end{proof}

\begin{theorem}[A finite nonpolar linearly dominated datum]
\label{thm:r3-linear-capacity-no-solution}
The measure
\begin{equation}
 \mu=\sum_{j=1}^\infty m_j\delta_{z_j},\qquad
 m_j=t_j^{-n}-t_{j+1}^{-n}>0
 \label{eq:r3-linear-capacity-datum}
\end{equation}
is finite, supported in $K_1\Subset\Omega$, and gives no mass
to locally pluripolar sets. It satisfies, for every Borel $E$,
\begin{equation}
 \mu(E)\leq\operatorname{Cap}_{[1]}(E;\Omega).
 \label{eq:r3-linear-capacity-bound}
\end{equation}
There is no locally bounded actual classical psh solution with
ordinary boundary limit zero and $(Du)^n=\mu$.
For every $\alpha>0$ there is no finite constant $A$ for which
$\mu(E)\leq A\operatorname{Cap}_{[1]}(E;\Omega)^{1+\alpha}$
holds for all Borel $E$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} counterexample.
The assertion that finite nonpolar data and linear capacity
domination suffice for bounded zero-boundary solvability is
\texttt{FALSE} in every dimension.

\begin{proof}
The total mass is $t_1^{-n}$. Every $z_j$ is divisorial and
has positive singleton capacity, so an ambient locally polar
set contains none of these points. The origin has zero mass.
If $E$ contains any atoms, let $j$ be its least atom index.
Then
\[
 \mu(E)\leq\sum_{l\geq j}m_l=t_j^{-n}
       =\operatorname{Cap}_{[1]}(\{z_j\};\Omega)
       \leq\operatorname{Cap}_{[1]}(E;\Omega).
\]
Otherwise $\mu(E)=0$. Moreover
$\mu(K_j)=\operatorname{Cap}_{[1]}(K_j;\Omega)=t_j^{-n}$,
which disproves every uniform higher power bound.

We compute the finite partial solutions, rather than assume
weak stability for their data. Put
\[
 A_j^{(N)}=\left(\sum_{k=j}^N m_k\right)^{1/n}
 \quad(1\leq j\leq N),\qquad A_{N+1}^{(N)}=0
\]
and
\begin{equation}
 u_N=\sum_{j=1}^N
       (A_j^{(N)}-A_{j+1}^{(N)})\max(s,-t_j).
 \label{eq:r3-radial-partial-solution}
\end{equation}
It is continuous classical, has ordinary zero boundary values,
and is bounded for each fixed $N$. Its scalar radial slope
between successive breaks is $A_j^{(N)}$.
At $z_j$ the centered local tropical function, up to an additive
constant, is
\[
 \max\{A x_1,\ldots,A x_n,B x_1,\ldots,B x_n\},
 \qquad A=A_j^{(N)},\quad B=A_{j+1}^{(N)}.
\]
The convex hull of the slopes is
$\{\xi_l\geq0,\ B\leq\sum_l\xi_l\leq A\}$, with volume
$(A^n-B^n)/n!$. Its original top mass is therefore
$A^n-B^n=m_j$.

Here the calibration can be verified directly by smoothing the
finite maximum with log-sum-exp. Its Hessian is positive definite
when the slope hull has interior. For any interior slope target
$\xi$, the smooth function minus $\xi\cdot x$ is coercive and
strictly convex, so its gradient is a diffeomorphism onto that
interior. Change of variables gives the displayed volume.
The original coordinate tropical integration has multiplicity
one and factor $n!$. Scaling the smoothing parameter concentrates
this mass at the original Gauss point. At other strata the active
slope differences have rank below $n$; coordinate zero branches
are locally inactive. Smooth original integration under
refinements and bounded BT convergence therefore prove, on all
original compact tests,
\begin{equation}
 (Du_N)^n=\sum_{j=1}^N m_j\delta_{z_j}.
 \label{eq:r3-radial-partial-measure}
\end{equation}
This computation applies to any positive finite weights at these
radial points and will be used below with other weights.

With $t_0=0$, summation by parts gives the actual origin value
\[
 -u_N(o)=\sum_{j=1}^N A_j^{(N)}(t_j-t_{j-1}).
\]
Since $A_j^{(N)}\geq m_j^{1/n}
 =(1-2^{-n})^{1/n}/t_j$, it follows that
\[
 -u_N(o)\geq(1-2^{-n})^{1/n}
                  \left(1+\frac{N-1}{2}\right)\longrightarrow\infty.
\]
If $u$ were a locally bounded zero-boundary solution for $\mu$,
measure order and the ordinary boundary limits would imply
$u\leq u_N$ for every $N$, because
$(Du)^n\geq(Du_N)^n$. This contradicts the finite actual value
$u(o)$. The comparison is valid on compact gaps, so it does
not require subtracting infinite total masses near the boundary.
\end{proof}

% END source: base_r3/sections/linear_capacity_obstruction.tex


% BEGIN frozen or reviewed source: base_r3/sections/general_modulus_obstruction.tex
% Reconstructed after the scratch reset. Original reviewed SHA256:
% 82c69b3cb16af56d2894d025da513a65b7cbe1cb5948b142aa0a235754ae8f05.
\section{A non-Dini modulus obstruction in every dimension}
\label{sec:r3-nondini-capacity-obstruction}

Use the same fixed ordinary polydisc, radial points, and scales
$t_j=2^j\lambda$ as in Section~\ref{sec:r3-linear-capacity-obstruction}.
Write $q_j=t_j^{-1}$ and $C_j=t_j^{-n}$. For a finite nondecreasing
function $\omega:(0,\infty)\to(0,\infty)$ set
\[
 F(x)=x\,\omega(x^{1/n})^n\quad(0<x<\infty),\qquad
 F(0)=0,\qquad F(\infty)=\infty.
\]

\begin{theorem}[Failure of the universal guarantee outside the Dini range]
\label{thm:r3-nondini-modulus-obstruction}
Suppose
\begin{equation}
 \int_0^a\omega(t)\,\frac{dt}{t}=\infty
 \quad\hbox{for some }a>0.
 \label{eq:r3-nondini-hypothesis}
\end{equation}
There is a finite nonpolar positive measure $\mu$ supported in
the fixed compact $K_1=\{o,z_1,z_2,\ldots\}\Subset\Omega$ such that
every Borel $E\subset\Omega$ satisfies
\begin{equation}
 \mu(E)\leq F\bigl(\operatorname{Cap}_{[1]}(E;\Omega)\bigr),
 \label{eq:r3-nondini-growth}
\end{equation}
but $\mu$ has no locally bounded actual classical zero-boundary
Monge--Ampere solution.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} counterexample,
using the already calibrated radial finite solutions. The uniform
existence guarantee for this prescribed non-Dini modulus is
\texttt{FALSE}; this is not an if-and-only-if criterion for an
individual measure.

\begin{proof}
Set
\[
 W_j=C_j\omega(q_j)^n,\qquad m_j=W_j-W_{j+1},\qquad
 \mu=\sum_{j=1}^{\infty}m_j\delta_{z_j}.
\]
Monotonicity of $\omega$ and $q_{j+1}=q_j/2$ imply
\begin{equation}
 W_{j+1}\leq2^{-n}W_j,\qquad
 m_j\geq(1-2^{-n})W_j>0.
 \label{eq:r3-modulus-tail-ratio}
\end{equation}
Since $\omega(q_j)\leq\omega(q_1)<\infty$, $W_j\to0$.
The total mass is $W_1$, and its tail from index $j$ is $W_j$.
As before it is nonpolar: all its atoms are divisorial, and the
origin has no mass.

If a Borel set contains atoms, take its least atom index $j$.
Lemma~\ref{lem:r3-radial-compact-capacity} and monotonicity of
$F$ give
\[
 \mu(E)\leq W_j=F(C_j)
 =F(\operatorname{Cap}_{[1]}(\{z_j\};\Omega))
 \leq F(\operatorname{Cap}_{[1]}(E;\Omega)).
\]
Sets containing no atom have zero mass. This proves the growth
bound, including the cases of zero or infinite capacity.

For the first $N$ weights use the explicit finite solution
\eqref{eq:r3-radial-partial-solution}, now with
$A_j^{(N)}=(\sum_{k=j}^Nm_k)^{1/n}$.
The calibrated identity \eqref{eq:r3-radial-partial-measure}
holds for these weights as well. Its origin depth obeys
\[
 \begin{split}
 -u_N(o)
 &=\sum_{j=1}^N A_j^{(N)}(t_j-t_{j-1})\\
 &\geq(1-2^{-n})^{1/n}
       \left(\omega(q_1)+\frac12\sum_{j=2}^N\omega(q_j)\right).
 \end{split}
\]
For every $j$, monotonicity gives
\[
 (\log2)\omega(q_{j+1})
 \leq\int_{q_{j+1}}^{q_j}\omega(t)\,\frac{dt}{t}
 \leq(\log2)\omega(q_j).
\]
The integral over any compact interval bounded away from zero
is finite. Hypothesis \eqref{eq:r3-nondini-hypothesis} therefore
forces $\sum_j\omega(q_j)=\infty$. The depths tend to infinity.
Any bounded zero-boundary solution for $\mu$ would lie below
every finite partial solution by measure order, contradicting
its finite actual value at the origin.
\end{proof}

\begin{corollary}[The logarithmic threshold]
\label{cor:r3-logarithmic-modulus-obstruction}
For $0<\beta\leq1$, there is such a measure with
\[
 \mu(E)\leq
 \frac{\operatorname{Cap}_{[1]}(E;\Omega)}
 {\bigl[\log(e+\operatorname{Cap}_{[1]}(E;\Omega)^{-1/n})\bigr]^{n\beta}}
\]
at finite positive capacity, yet with no bounded zero-boundary
solution. At zero and infinite capacity use the preceding
conventions.
\end{corollary}

\begin{proof}
Take $\omega(t)=[\log(e+1/t)]^{-\beta}$. It is finite,
positive, and nondecreasing, and its Dini integral diverges
exactly for $0<\beta\leq1$ in this range.
\end{proof}

Allowing a finite nonnegative nondecreasing modulus does not
enlarge the non-Dini case: if it vanished at any positive
argument, it would vanish on the whole smaller interval and
its Dini integral there would be finite. Combined with the
proved Dini sufficiency theorem, the construction gives the
sharp universal-modulus threshold in the stated fixed-domain,
compact-support class. It makes no assertion that every
individual datum outside the Dini range fails.

% END source: base_r3/sections/general_modulus_obstruction.tex


% BEGIN frozen or reviewed source: base_r3/sections/universal_modulus_criterion.tex
\section{The sharp universal capacity-modulus criterion on a polydisc}
\label{sec:r3-universal-modulus-criterion}

Fix $n\geq1$, the original finite extension $K/\mathbf Q_p$, a
uniformizer $\pi$, and the same polydisc as in the radial obstruction:
\[
 R=|\pi|^{-1},\qquad
 \Omega=\{|T_l|<R:1\leq l\leq n\},\qquad
 C(E)=\operatorname{Cap}_{[1]}(E;\Omega).
\]
Let $\omega:[0,\infty)\to[0,\infty)$ be finite, nondecreasing,
and zero at zero. Put $F(c)=c\omega(c^{1/n})^n$ for finite
$c>0$, with $F(0)=0$ and $F(+\infty)=+\infty$.

\begin{corollary}[A universal rather than individual equivalence]
\label{cor:r3-universal-modulus-criterion}
The following assertions are equivalent.
\begin{enumerate}
\item $\int_0^a\omega(t)\,dt/t<\infty$ for some $a>0$.
\item Every finite positive Radon measure $\mu$ carried by a compact
subset of $\Omega$ and satisfying
$\mu(E)\leq F(C(E))$ for every Borel $E\subset\Omega$ has
a bounded actual classical zero-boundary solution of $(Du)^n=\mu$.
\item Every measure in (2) has such a solution continuous on
$\overline\Omega$.
\end{enumerate}
Such bounded zero-boundary solutions are unique. If (1) fails,
a counterexample can be chosen to be a sum of positive atoms at
original divisorial points tending to the rigid origin, with no
mass at the origin or on any ambient locally pluripolar set.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
proved sufficient Dini theorem, its continuity improvement, and
the non-Dini radial construction. The quantifier is over all
measures with the specified bound, not over one individual measure.

\begin{proof}
The function
\[
 \rho=\max\{-1,\log|T_1|-\log R,\ldots,
                          \log|T_n|-\log R\}
\]
is continuous classical on the whole affine space and makes
$\Omega=\{\rho<0\}$ an extendible exhaustion domain. Therefore
(1) implies (3) by Theorems~\ref{thm:r3-dini-capacity-dirichlet}
and \ref{thm:r3-dini-solution-continuity}. Compactness of the
ordinary closure makes (3) imply (2).

If (1) fails, $\omega(t)>0$ for every $t>0$. Otherwise
$\omega(t_0)=0$ would make it vanish on $(0,t_0]$, and finite
monotonicity would make its Dini integral finite on every compact
interval away from zero. Thus the restriction of $\omega$ to
$(0,\infty)$ satisfies the hypotheses of
Theorem~\ref{thm:r3-nondini-modulus-obstruction}. Its explicit
measure on this same polydisc contradicts (2), and has the stated
nonpolar atomic form. Uniqueness is ambient pointwise comparison.
\end{proof}

The zero measure, for example, satisfies every such growth bound
and has the zero solution. Thus failure of the Dini condition
does not forbid solutions for all individual data. The assertion
also does not extend to arbitrary relative backgrounds with ambient
classical solutions, as
Theorem~\ref{thm:r3-relative-dirichlet-no-classical} shows.

% END source: base_r3/sections/universal_modulus_criterion.tex


% BEGIN frozen or reviewed source: base_r3/sections/relative_representative_obstruction.tex
\section{A nonzero polar background with no classical representative}
\label{sec:r3-relative-representative-obstruction}

Let
\[
 V=\{|x|<1,\ |y|<1\}\subset(\mathbf A_K^2)^{\an},
 \qquad Y_1=\{x=0\}\cap V,\quad Y_2=\{y=0\}\cap V,
\]
and $S=[Y_1]+[Y_2]$. This is a nonzero symmetric original-positive
separately closed $(1,1)$-current on a connected smooth space.
The remaining degree is one. All capacities retain normalized
ambient actual classical testers.

\begin{proposition}[Incompatible branch values in a completed class]
\label{prop:r3-no-relative-classical-representative}
There is $a\in\mathcal A_S^{[0,1]}(V)$ which is an actual
increasing limit of continuous normalized classical functions,
but no actual classical psh function is equal to $a$ locally
outside an outer $S$-capacity-zero set. One can take
\[
 a_j=\max\{0,1+j^{-1}\log|x|\}\uparrow a,
 \qquad a=\mathbf1_{\{x\ne0\}}.
\]
Its completed relative curvature is $D(aS)=0$.
\end{proposition}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} counterexample;
the unrestricted classical-representative assertion is \texttt{FALSE}.

\begin{proof}
Each $a_j$ is continuous classical and lies in $[0,1]$. The finite
floor removes the logarithmic singularity. Because $\log|x|<0$
on $V$, the rows increase to the displayed actual Borel limit.
The proved increasing-capacity theorem places it in the normalized
relative completion on every permitted local compact/domain pair.

Let $\xi_{i,r}$ be the centered Gauss point of radius $0<r<1$
on $Y_i$. Each singleton has strictly positive relative capacity.
For example, with $\ell_r=-\log r$, the normalized tester
\[
 b_r(y)=\max\{0,1+\ell_r^{-1}\log|y|\}
\]
has $D(b_rS)=\ell_r^{-1}\delta_{\xi_{1,r}}$; its restriction
to $Y_2$ is zero. Exchanging $x,y$ gives the other singleton.
These are ordinary original-field curve calibrations. The same
positive atomic mass can be tested on a smaller local domain
containing the given point, so a local outer-null exceptional set
cannot contain any $\xi_{i,r}$.

If a classical $w$ represented the relative class, it would satisfy
$w(\xi_{1,r})=0$ and $w(\xi_{2,r})=1$ for every $r$. Restrict
$w$ to the first analytic axis. The classical maximum theorem on
its closed radius-$r$ disc gives $w(o)\leq w(\xi_{1,r})=0$.
On the other axis, $\xi_{2,r}\to o$ as $r\downarrow0$.
Upper semicontinuity gives $w(o)\geq1$. This contradiction proves
nonrepresentability without assuming that $w$ is continuous.

On $Y_1$ every row is zero; on $Y_2$ its Laplacian is the atom
$j^{-1}\delta_{\xi_{2,e^{-j}}}$. Hence $D(a_jS)\to0$ on all
original compact tests. The completed increasing calculus identifies
the limit with $D(aS)$, proving the last assertion.
\end{proof}

This is not a counterexample to compact capacity regularity: its
completed curvature is zero. Nor is $a$ a difference of two locally
bounded classical functions. For any such function $h$, the maximum
theorem and upper semicontinuity imply
$h(\xi_{i,r})\to h(o)$ on each axis. A difference must therefore
have the same limiting value on both axes, whereas $a$ has limits
zero and one. Thus this example does not refute the still open
potential quotient for differences of classical potentials.

% END source: base_r3/sections/relative_representative_obstruction.tex


% BEGIN frozen or reviewed source: base_r3/sections/arbitrary_relative_dirichlet_obstruction.tex
% Reconstructed after the scratch reset. Original reviewed SHA256:
% d67c630596611b08c8baaf9ec86d267d55b867bd1472763e1fcf1f2c47ed4ab6.
\section{A relative Dirichlet obstruction in every positive codimension}
\label{sec:r3-arbitrary-relative-dirichlet-obstruction}

Fix integers $p,q\geq1$, put $n=p+q$, and let
\[
 V=\{|T_1|<1,\ldots,|T_n|<1\}\subset(\mathbf A_K^n)^{\rm an}.
\]
Choose a uniformizer $\pi$, and write $r=|\pi|$ and
$\lambda=-\log r>0$. The actual function
\[
 G=\max\{\log|T_1|,\ldots,\log|T_n|,\log r\}
\]
is bounded continuous classical psh on $V$, lies between
$-\lambda$ and zero, and has ordinary boundary limit zero.
Let
\[
 \begin{split}
 Z_1&=\{T_1=T_{q+2}=\cdots=T_n=0\}\cap V,\\
 Z_2&=\{T_2=T_{q+2}=\cdots=T_n=0\}\cap V ,
 \end{split}
\]
with the additional equations absent when $p=1$. These are smooth
$q$-dimensional coordinate polydiscs. Write $i_i:Z_i\hookrightarrow V$
and $S=[Z_1]+[Z_2]$. This is a nonzero symmetric original-positive
separately closed $(p,p)$-current with connected support.

Let $z_i$ be the Gauss point of the radius-$r$ closed polydisc in
the $q$ free coordinates of $Z_i$. Both are points of the original
$K$-analytic space. They lie outside $Z_1\cap Z_2$ and are
divisorial intrinsically on $Z_i$. Put
\[
 \mu=\delta_{z_1}+2^q\delta_{z_2}.
\]
All statements below concern all original compact smooth tests.

\begin{theorem}[A bounded matching subsolution does not suffice]
\label{thm:r3-relative-dirichlet-no-classical}
There is no $u\in\mathcal C_b(V)$ with ordinary boundary limit zero
and
\[
 T_{S,q}(u,\ldots,u)=\mu.
\]
Nevertheless the bounded continuous classical function $w=2G$ has
the same boundary values and satisfies
\begin{equation}
 T_{S,q}(w,\ldots,w)
       =2^q\delta_{z_1}+2^q\delta_{z_2}\geq\mu.
 \label{eq:r3-relative-matching-subsolution}
\end{equation}
For every $\alpha>0$, all Borel $E\subset V$ satisfy
\begin{equation}
 \mu(E)\leq A_\alpha\operatorname{Cap}_S(E;V)^{1+\alpha},
 \qquad A_\alpha=(1+2^q)\lambda^{q(1+\alpha)}.
 \label{eq:r3-relative-power-datum}
\end{equation}
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} counterexample.
Thus the arbitrary-background classical bounded-subsolution
Dirichlet assertion is \texttt{FALSE}, even with this power
capacity bound in every remaining degree $q\geq1$.

\begin{proof}
On $Z_i$, let $g_i=G|_{Z_i}$. The original tropical calibration of
the radial maximum gives
\begin{equation}
 (Dg_i)^q=\delta_{z_i}.
 \label{eq:r3-relative-branch-measure}
\end{equation}
Indeed near the radius-$r$ Gauss point its slope polytope is
$\operatorname{conv}(0,e_1,\ldots,e_q)$ of volume $1/q!$;
the original CLD top-product normalization is $q!$ and the
coordinate tropical multiplicity is one. Elsewhere the slope
differences have rank below $q$. One can smooth the maximum and
pass by the proved bounded BT continuity, so the equality is an
ordinary-current equality on all original tests.

For a hypothetical $u$, analytic restriction preserves the actual
classical class. Integration-current compatibility and background
linearity give
\begin{equation}
 T_{S,q}(u,\ldots,u)
 =i_{1*}(D(u|_{Z_1}))^q+i_{2*}(D(u|_{Z_2}))^q .
 \label{eq:r3-relative-axis-splitting}
\end{equation}
Both summands are positive and carried by their respective
coordinate subspaces. Since $z_1\notin Z_2$ and $z_2\notin Z_1$,
equality with $\mu$ forces the intrinsic measures to be
$\delta_{z_1}$ and $2^q\delta_{z_2}$, respectively. In particular
there is no mass on the intersection: a positive summand cannot
cancel mass there.

The ordinary boundary of each coordinate polydisc lies in the
ordinary boundary of $V$, so both restrictions have boundary
limit zero. The proved intrinsic ambient uniqueness theorem,
which determines actual bounded classical potentials, and
\eqref{eq:r3-relative-branch-measure} imply
\[
 u|_{Z_1}=g_1,\qquad u|_{Z_2}=2g_2 .
\]
At the common rigid origin these values are $-\lambda$ and
$-2\lambda$, a contradiction.

Formula \eqref{eq:r3-relative-matching-subsolution} follows from
the same integration identity and positive homogeneity. Finally
$b=1+G/\lambda$ is an actual normalized tester on $V$ and
\[
 T_{S,q}(b,\ldots,b)
           =\lambda^{-q}(\delta_{z_1}+\delta_{z_2}).
\]
If $E$ contains an atom of $\mu$, its capacity is at least
$\lambda^{-q}$, whereas $\mu(E)\leq1+2^q$. This gives
\eqref{eq:r3-relative-power-datum}; if it contains no atom the
inequality is immediate.
\end{proof}

\begin{proposition}[An actual completed solution]
\label{prop:r3-relative-completed-dirichlet-example}
Define, with the actual value $\log0=-\infty$,
\[
 a_j=\max\{2G,\ G+j^{-1}\log|T_2|\}.
\]
These are bounded continuous classical psh functions, increase
pointwise, and have boundary limit zero. Their actual limit is
\[
 a=
 \begin{cases}
 G,&T_2\ne0,\\
 2G,&T_2=0.
 \end{cases}
\]
It belongs to the bounded relative capacitary completion
$\mathcal A_{S,b}(V)$, has boundary limit zero, and satisfies
\begin{equation}
 \mathcal T_{S,q}(a,\ldots,a)=\mu.
 \label{eq:r3-relative-completed-solution}
\end{equation}
The sequence converges locally in outer $S$-capacity on all
permitted compact/domain pairs. In contrast, $a^*=G$ and
$T_{S,q}(a^*,\ldots,a^*)=\delta_{z_1}+\delta_{z_2}\ne\mu$.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the actual
increasing-limit calculus, integration compatibility, and intrinsic
ambient polar zero, applied to this explicit construction.

\begin{proof}
The maximum with the bounded floor handles the logarithmic pole.
It is classical by truncating the logarithm and decreasing to the
bounded maximum. It is continuous at the zero locus because
the logarithmic branch is inactive on a neighborhood of each such
point. Since $\log|T_2|\leq0$ on $V$, the sequence increases and
\[
 2G\leq a_j\leq G\leq0.
\]
These bounds give its boundary limit and the stated actual
pointwise limit. The relative increasing-limit theorem gives
membership in $\mathcal A_{S,b}$ and local all-domain capacity
convergence.

To identify the measure, restrict only the classical approximants.
On $Z_2$, every restriction is $2g_2$. On $Z_1$ their actual
increasing limit equals $g_1$ off the proper intrinsic analytic
divisor $Z_1\cap Z_2$, and equals $2g_1$ on that divisor.
The divisor has intrinsic ambient outer capacity zero. Hence the
intrinsic completed class and its measure are those of $g_1$.
Intrinsic increasing continuity gives
\[
 (D(a_j|_{Z_1}))^q\longrightarrow\delta_{z_1},\qquad
 (D(a_j|_{Z_2}))^q=2^q\delta_{z_2}.
\]
Apply \eqref{eq:r3-relative-axis-splitting} to each classical $a_j$.
Pushforward against all original tests, followed by the relative
completed limit, proves
\eqref{eq:r3-relative-completed-solution}. No pullback theorem for
an arbitrary completed Borel representative is assumed.

The complement of $\{T_2=0\}$ is dense: normal Gauss points of
positive radius approach every point of that coordinate
hyperplane. Continuity of $G$ and $a\leq G$ therefore give
$a^*=G$. The last measure formula follows from
\eqref{eq:r3-relative-branch-measure}.
\end{proof}

\paragraph{Scope.}
This example works for every positive codimension $p$ and every
remaining degree $q$. It does not refute the ambient $S=[1]$
Dirichlet theorem. It supplies a particular completed solution,
not a general arbitrary-background completed Dirichlet theorem,
and it does not define a wedge of two arbitrary currents.

% END source: base_r3/sections/arbitrary_relative_dirichlet_obstruction.tex


% BEGIN frozen or reviewed source: base_r3/sections/relative_dirichlet_tv_limit.tex
% Reconstructed after the scratch reset. Original reviewed SHA256:
% 19209703b0dd390cd7d0bd8249c5e0d6827340143d494ebd5182ca58e517a86a.
\section{Total-variation limits may require a completed solution}
\label{sec:r3-relative-dirichlet-tv-limit}

Use $p,q,V,Z_i,S,G,a_j,\mu$ from
Section~\ref{sec:r3-arbitrary-relative-dirichlet-obstruction}.
Let $\zeta_j\in Z_1$ be the Gauss point whose free coordinate
$T_2$ has radius $r^j$ and whose other $q-1$ free coordinates
have radius $r$. Thus $\zeta_1=z_1$.

\begin{proposition}[The exact measures of the approximants]
\label{prop:r3-relative-dirichlet-tv-formula}
For every integer $j\geq1$ one has the ordinary measure identity
\begin{equation}
 T_{S,q}(a_j,\ldots,a_j)
      =\mu+\frac{2^q-1}{j}\delta_{\zeta_j}.
 \label{eq:r3-relative-dirichlet-tv-formula}
\end{equation}
Consequently
\begin{equation}
 \|T_{S,q}(a_j^q)-\mu\|_{\rm TV}
 =d_{\mathrm{psh},V}(T_{S,q}(a_j^q),\mu)
 =\frac{2^q-1}{j}.
 \label{eq:r3-relative-tv-and-dual}
\end{equation}
All these measures have one common compact support in $V$.
\end{proposition}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the exact
formula. The convergence and closure obstruction below are
\texttt{DIRECT\_COROLLARY}.

\begin{proof}
On $Z_2$, $a_j=2g_2$ and its intrinsic top measure is
$2^q\delta_{z_2}$. On $Z_1$ write $t_1=T_2$ and
$t_l=T_{l+1}$ for $2\leq l\leq q$. On its torus let
$s_l=\log|t_l|<0$ and set
\[
 h=\max(-\lambda,s_2,\ldots,s_q),\quad
 g=\max(s_1,h),\quad k_j=\max(s_1/j,h).
\]
For $q=1$, $h=-\lambda$. Since $s_1\leq s_1/j$, a direct
three-case comparison with $h$ gives
\begin{equation}
 a_j|_{Z_1}=\max(2g,g+s_1/j)=g+k_j.
 \label{eq:r3-relative-radial-two-maxima}
\end{equation}
The equality extends over the coordinate zero loci by continuity.
For $j=1$ this is $2g$, giving $2^q\delta_{z_1}$, exactly the
asserted formula when $\zeta_1=z_1$.

Suppose $j>1$. At $z_1$, all $s_l=-\lambda$, and $k_j=s_1/j$
is affine near this tropical point. Hence the mass there is the
mass of $g$, namely one. At $\zeta_j$ the tropical coordinates
are $s_1=-j\lambda$ and $s_l=-\lambda$ for $l\geq2$. In centered
coordinates $x_1=s_1+j\lambda$, $x_l=s_l+\lambda$ one has
\[
 g+k_j=-2\lambda+
       \max(0,x_2,\ldots,x_q)
       +\max(0,x_1/j,x_2,\ldots,x_q).
\]
The slope polytope at this vertex is the Minkowski sum
\begin{equation}
 P_j=\operatorname{conv}(0,e_2,\ldots,e_q)
           +\operatorname{conv}(0,e_1/j,e_2,\ldots,e_q).
 \label{eq:r3-relative-shell-polytope}
\end{equation}
For $q=1$ it is the interval $[0,1/j]$. In general its slice at
$\xi_1=t$, $0\leq t\leq1/j$, is
\[
 \xi_2,\ldots,\xi_q\geq0,\qquad
             \xi_2+\cdots+\xi_q\leq2-jt.
\]
Indeed a point of the second simplex with first coordinate $t$
has remaining coordinate sum at most $1-jt$, and the first
simplex contributes at most one; conversely any nonnegative
vector with the indicated total can be split between these
two bounds. Thus
\[
 \operatorname{vol}_q(P_j)
 =\int_0^{1/j}\frac{(2-jt)^{q-1}}{(q-1)!}\,dt
 =\frac{2^q-1}{j q!}.
\]

Here the use of the original torus normalization can be checked
without an unproved delta pullback. Smooth each maximum by its
log-sum-exp regularization. The sum of the two regularizations is
the log-sum-exp for all sums of their slope vectors. Its Hessian
is positive definite, since the slopes affinely span $\mathbf R^q$.
Its gradient maps $\mathbf R^q$ diffeomorphically onto
$\operatorname{int}P_j$: for an interior target $\xi$, the
regularization minus $\xi\cdot x$ is strictly convex and coercive,
and hence has a unique critical point. The change of variables
formula integrates its Hessian determinant to
$\operatorname{vol}_q(P_j)$. The original CLD top wedge on the
coordinate tropical chart has the factor $q!$ and multiplicity
one. Rescaling the regularization parameter concentrates the
mass at the displayed Gauss point. The smooth local computation,
the original tropical integration formula under refinements,
and bounded BT continuity identify the limit on every original
compact smooth test. Its mass is therefore $(2^q-1)/j$.

For completeness there is no other top-degree mass. Away from
the ties among $s_1$, $s_1/j$, and $h$, both terms are affine,
or one is affine and the other depends on fewer than $q$
directions, or both depend only on $s_2,\ldots,s_q$. At
$s_1=h<s_1/j$, the second term is affine, so the only possible
top vertex is the radial vertex $z_1$ already counted.
At $s_1<h=s_1/j$ with $h>-\lambda$, the constant floor is
inactive. Every active total gradient satisfies
$j\xi_1+\xi_2+\cdots+\xi_q=2$; their differences have rank
at most $q-1$. If $h=-\lambda$ but fewer than $q-1$ of the
remaining coordinates equal the floor, fewer than $q$ directions
are active. The only remaining case is $\zeta_j$.
These rank conclusions are also read from the smooth
regularizations and the same bounded limit. Near a zero of
some $t_l$, its logarithmic branch is inactive; near $t_1=0$
both maxima equal $h$. Thus there is no additional mass on
coordinate zero loci.

The intrinsic result on $Z_1$ is
$\delta_{z_1}+(2^q-1)j^{-1}\delta_{\zeta_j}$.
Integration-current compatibility and linearity in $S$ prove
\eqref{eq:r3-relative-dirichlet-tv-formula} on $V$.
The extra measure is positive; its total variation equals its
total mass, and the normalized constant tester one attains the
same psh-dual value. Finally $\zeta_j$ tends to the point of
$Z_1\cap Z_2$ with $t_1=0$ and the remaining coordinates of
radius $r$. This proves common compact support, including when
$q=1$, where the limit is the rigid origin.
\end{proof}

\begin{corollary}[Failure of classical closure in total variation]
\label{cor:r3-relative-dirichlet-not-tv-closed}
For this fixed background, top measures of uniformly bounded
continuous classical potentials with common ordinary zero
boundary values need not form a closed family, even under
total-variation convergence on a fixed compact support.
The same failure holds for strong psh-dual convergence. The
limit measure has a continuous matching subsolution and satisfies
every fixed power capacity growth condition in
\eqref{eq:r3-relative-power-datum}, but has no ambient classical
zero-boundary solution. Its actual relative completed solution
is $a$ from Proposition~\ref{prop:r3-relative-completed-dirichlet-example}.
\end{corollary}

\begin{proof}
Use \eqref{eq:r3-relative-tv-and-dual} and
Theorem~\ref{thm:r3-relative-dirichlet-no-classical}.
The common bounds and boundary values of the $a_j$, their
capacity convergence to $a$, and the completed equation were
proved in the cited proposition.
\end{proof}

% END source: base_r3/sections/relative_dirichlet_tv_limit.tex


% BEGIN frozen or reviewed source: base_r3/sections/finite_carrier_completed_gluing.tex
\section{Constructive gluing on finitely many analytic carriers}
\label{sec:r3-finite-carrier-completed-gluing}

This construction gives completed relative potentials with
independently prescribed branch operators. The hypotheses explicitly
use local \emph{outer} capacity zero; raw capacity zero is not
silently substituted for it.

\begin{proposition}[Logarithmic separation of finitely many carriers]
\label{prop:r3-finite-carrier-completed-gluing}
Let $U$ be an allowed ordinary open set and let
$S_1,\ldots,S_N$ be symmetric original-positive separately closed
currents of one bidegree $(p,p)$. Put
$S=\sum_iS_i$ and $q=n-p\geq1$. Suppose that $S_i$ is
carried by a closed set $A_i$, and there are analytic functions
$f_i$ on $U$, with $|f_i|\leq1$, such that
\[
 A_i\subset F_i:=\{f_i=0\}.
\]
Assume each $\log|f_i|$ is an allowed extended classical psh
function, and, for $i\ne l$, assume that $F_l$ has locally
zero outer $S_i$-capacity on all permitted compact/domain pairs.

Let $h_1,\ldots,h_N,w\in\mathcal C_b(U)$ satisfy
$w\leq h_i$ on $A_i$. Define
\[
 \psi_i=\sum_{l\ne i}\log|f_l|,
 \qquad
 a_j=\max\{w,h_1+j^{-1}\psi_1,\ldots,
                         h_N+j^{-1}\psi_N\}.
\]
For $N=1$ the empty sum is zero. Then $a_j$ is classical,
locally uniformly bounded, and increases to an actual Borel
function $a\in\mathcal A_{S,b}(U)$. For every $0\leq k\leq q$,
\begin{equation}
 \mathcal T_{S,k}(a,\ldots,a)
       =\sum_{i=1}^N T_{S_i,k}(h_i,\ldots,h_i)
 \label{eq:r3-finite-carrier-glued-operator}
\end{equation}
on all original compact smooth complementary tests. If the
$h_i$ and $w$ are continuous, each $a_j$ is continuous as well.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
proved increasing-limit and completed-calculus theorems, with
an explicit gluing construction. It is not a wedge product of
two arbitrary currents.

\begin{proof}
The functions $\psi_i$ are extended classical psh and
nonpositive. Taking their sums with $h_i$, and then the finite
maximum with the locally bounded floor $w$, gives an actual
locally bounded classical function. Moreover
\[
 w\leq a_j\leq\max(w,h_1,\ldots,h_N),
\]
so the bounds are uniform on each compact set, and $a_j$
increases with $j$. The relative increasing-limit theorem gives
all-domain local outer-capacity convergence to the actual limit
$a$, both for $S$ and separately for each fixed $S_i$.
Thus $a$ lies in every completed class required below.

At a point of $A_i\setminus\bigcup_{l\ne i}F_l$, the
penalty $\psi_i$ is finite. Every other branch penalty contains
$\log|f_i|=-\infty$ and is inactive. Hence
\[
 a=\max(w,h_i)=h_i
        \quad\hbox{on }A_i\setminus\bigcup_{l\ne i}F_l.
\]
The ordinary open set $U\setminus A_i$ has zero local outer
$S_i$-capacity: the background and every one of its BT products
are zero there by ordinary restriction, and the set is itself
an open zero-capacity neighborhood. The assumed outer-null
sets $F_l$, $l\ne i$, therefore account for all possible
discrepancies between $a$ and $h_i$ in the $S_i$-class.
Finite subadditivity gives
$a=h_i$ locally outside an outer $S_i$-capacity-zero set.
The completed calculus is invariant under exactly these changes,
so
\[
 \mathcal T_{S_i,k}(a,\ldots,a)
                 =T_{S_i,k}(h_i,\ldots,h_i).
\]

For every classical approximant $a_j$, the original recursion
is linear in its background:
\[
 T_{S,k}(a_j,\ldots,a_j)
          =\sum_{i=1}^N T_{S_i,k}(a_j,\ldots,a_j).
\]
Pass to the completed limits along this common increasing
sequence and use the preceding identities. This proves
\eqref{eq:r3-finite-carrier-glued-operator}; for $k=0$ it is
the identity $S=\sum_iS_i$.

If $w,h_i$ are continuous, the branches are continuous off
their logarithmic zero sets. Near a point where a penalty is
$-\infty$, that branch lies below the continuous finite floor
on a sufficiently small neighborhood, because the other
summands are locally bounded above. It can therefore be deleted
locally. The remaining finite maximum is continuous, proving
the last assertion.
\end{proof}

\begin{corollary}[Completed Dirichlet gluing on coordinate branches]
\label{cor:r3-coordinate-completed-dirichlet}
Let $V$ be the open unit polydisc in
$(\mathbf A_K^n)^{\an}$, and let $Z_1,\ldots,Z_N$ be
distinct coordinate subspaces of one dimension $q\geq1$ and
positive codimension $p=n-q$. Put
$S=\sum_i[Z_i]$, with all $Z_i$ restricted to $V$.
For each $i$, suppose a finite positive Radon measure $\nu_i$
on $Z_i$ has a bounded continuous classical solution $h_i$
with intrinsic ordinary boundary value zero, and suppose
\[
 (Dh_i)^q=\nu_i,\qquad
 C_iG_i\leq h_i\leq0
\]
for some finite $C_i\geq0$, where, for one fixed $0<r<1$,
\[
 G=\max(\log|T_1|,\ldots,\log|T_n|,\log r),
 \qquad G_i=G|_{Z_i}.
\]
Then the relative completed class contains an actual bounded
solution with ambient ordinary boundary limit zero of
\[
 \mathcal T_{S,q}(a,\ldots,a)
                  =\sum_i(i_i)_*\nu_i.
\]
It is an increasing limit of uniformly bounded continuous
ambient classical zero-boundary potentials. No matching of
the intrinsic values of $h_i$ at intersections is required.

In particular, this conclusion applies when each $\nu_i$ has
compact support in $Z_i$ and satisfies an intrinsic Dini
capacity bound on that coordinate polydisc.
\end{corollary}

\begin{proof}
Let $\pi_i:V\to Z_i$ be the coordinate retraction and put
$H_i=h_i\circ\pi_i$. These are bounded continuous classical
psh functions on $V$, and $H_i\leq0$. Choose
$C\geq\max(1,C_1,\ldots,C_N)$ and $w=CG$. Then
$w|_{Z_i}\leq h_i=H_i|_{Z_i}$.

For every $i$ one can choose a linear function $f_i$ which
vanishes on $Z_i$ and is not identically zero on any $Z_l$
with $l\ne i$. Indeed, the linear forms vanishing on $Z_i$
form a finite-dimensional $K$-space. For each distinct
same-dimensional coordinate space $Z_l$, the forms also
vanishing there form a proper subspace. An infinite field
is not a finite union of such proper subspaces. Scale the
chosen form so that its coefficients are integral; then
$|f_i|\leq1$ on $V$.

For $i\ne l$, the locus $\{f_l=0\}\cap Z_i$ is a proper
analytic subset of the smooth $Z_i$. It has zero local outer
ambient capacity there. For each compact subset of $V$, its
intersection with this locus consequently has outer capacity zero
with capacity domain $Z_i$, by finite localization and domain
monotonicity. Retraction compatibility
Proposition~\ref{prop:r2-retract} gives outer
$[Z_i]$-capacity zero on that compact set with domain $V$.
The fixed-domain null-class comparison
Corollary~\ref{cor:r2-fixed-domain-null-classes} gives the
same local all-domain assertion. Thus the preceding proposition
applies with carrier $A_i=Z_i$, branch function $H_i$, and
floor $CG$. It constructs continuous ambient classical $a_j$
and an actual completed limit with the required operator, by
integration-current compatibility.

Since all penalties are nonpositive,
\[
 CG\leq a_j\leq0,\qquad CG\leq a\leq0.
\]
The continuous floor $CG$ tends to zero at every ordinary
boundary point of $V$, proving the ambient boundary limits.
It also gives a uniform global bound. This argument does not
require the projected branch functions $H_i$ to vanish on
every ambient boundary face.

Finally the intrinsic unit polydiscs have the continuous
global classical exhaustions $G_i$. The proved Dini Dirichlet
theorem and its continuity improvement supply $h_i$, and their
compact-support barrier estimate supplies the finite constants
$C_i$. This verifies the last assertion using only the
already established ambient theory on each intrinsic branch.
\end{proof}

\paragraph{Scope.}
The construction is finite and its carrier intersections satisfy
an explicit relative outer-null condition. It does not identify
ambient pluripolar sets with relative null sets for an arbitrary
background. It also does not supply an ambient classical
representative when branch values are incompatible. The examples
of Sections~\ref{sec:r3-arbitrary-relative-dirichlet-obstruction}
and \ref{sec:r3-relative-dirichlet-tv-limit} show why the
completed target class is necessary even under favorable
intrinsic data hypotheses.

% END source: base_r3/sections/finite_carrier_completed_gluing.tex


% BEGIN frozen or reviewed source: base_r3/sections/analytic_carrier_capacity_additivity.tex
\section{Capacity additivity on finitely many analytic carriers}
\label{sec:r3-analytic-carrier-capacity}

This section gives an all-degree class of backgrounds for which raw
relative capacity is outer regular on compact sets.  The argument
uses separate ambient testers on the individual carriers and glues
them into one normalized ambient classical potential.  It does not
replace an actual relative representative by an ambient usc envelope.

Work over the fixed finite extension $K/\mathbf Q_p$, with its original
absolute value.  Let $\Omega$ be an ordinary open subset of an allowed
good, Hausdorff, boundaryless, quasi-smooth, pure $n$-dimensional
$K$-analytic space.  Fix $0\leq p<n$ and $q=n-p$.  All backgrounds
below are symmetric, original CLD-positive, and separately
$d'$- and $d''$-closed currents of bidegree $(p,p)$.  All current
identities are against the full original compact smooth test sheaf.
For such a background $R$, write
\[
 \mu_{R,h}=T_{R,q}(h,\ldots,h),\qquad
 C_R(E)=\sup_{\substack{h\in\mathcal C_b(\Omega)\\0\leq h\leq1}}
             \mu_{R,h}(E)
\]
for Borel $E\subset\Omega$, and
\[
 C_R^*(A)=\inf_{\substack{G\supset A\\G\text{ ordinary open in }\Omega}}
                  C_R(G)
\]
for arbitrary $A\subset\Omega$.  These are capacities on the one
fixed domain $\Omega$.  Infinite values are permitted unless a
relatively compact core is specified.

\begin{theorem}[Finite analytic-carrier additivity]
\label{thm:r3-analytic-carrier-additivity}
Let $S_1,\ldots,S_m$ be backgrounds as above.  Suppose there are
analytic functions $f_1,\ldots,f_m$ on $\Omega$ such that
\[
 |f_i|\leq1,\qquad A_i=\{f_i=0\},
\]
and every coefficient measure of $S_i$ is carried by $A_i$.
Assume, for all distinct $i,l$, that
\begin{equation}
 C_{S_i}(A_l)=0.
 \label{eq:r3-carrier-cross-null}
\end{equation}
Put $S=\sum_{i=1}^m S_i$.  Then every Borel $E\subset\Omega$
satisfies
\begin{equation}
 C_S(E)=\sum_{i=1}^m C_{S_i}(E).
 \label{eq:r3-carrier-raw-additivity}
\end{equation}
For every arbitrary subset $A\subset\Omega$ one also has
\begin{equation}
 C_S^*(A)=\sum_{i=1}^m C_{S_i}^*(A).
 \label{eq:r3-carrier-outer-additivity}
\end{equation}
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
This label applies to the simultaneous testing assertion
\eqref{eq:r3-carrier-raw-additivity}.

The outer identity \eqref{eq:r3-carrier-outer-additivity} is a
\texttt{DIRECT\_COROLLARY}.  The carrier condition is a condition
on actual coefficient measures, not an identification of an abstract
current support with an intrinsic analytic space.

\begin{proof}
Linearity of the fixed-background recursion gives, for every
normalized $h$,
\[
 \mu_{S,h}=\sum_{i=1}^m\mu_{S_i,h}.
\]
Hence $C_S(E)\leq\sum_iC_{S_i}(E)$.  We prove the reverse inequality
by constructing one tester for any chosen finite tuple of testers.

Fix $h_i\in\mathcal C_b(\Omega)$ with $0\leq h_i\leq1$, and fix
$0<\delta<1/2$.  Give analytic logarithms their actual value
$-\infty$ on their zero sets.  Put
\[
 \psi_i=\sum_{l\ne i}\log|f_l|\leq0,
 \qquad
 b_{\varepsilon,\delta}
   =\max\bigl\{0,\,
         \delta+(1-2\delta)h_i+\varepsilon\psi_i:
         1\leq i\leq m\bigr\},\qquad\varepsilon>0.
\]
An empty sum is zero.  This function belongs to
$\mathcal C_b(\Omega)$ and satisfies
\begin{equation}
 0\leq b_{\varepsilon,\delta}\leq1-\delta<1.
 \label{eq:r3-carrier-pasted-normalized}
\end{equation}
Here membership can be checked entirely with bounded classical
functions: replace each $\log|f_l|$ by
$\max\{\log|f_l|,-N\}$, take the same finite maximum, and let
$N\to\infty$.  The resulting bounded classical functions decrease
to $b_{\varepsilon,\delta}$ and have the common lower bound zero.
The accepted decreasing closure of the actual classical class gives
the assertion.  Truncated analytic logarithms, positive linear
combinations, and finite maxima are classical after every complete
extension, so this argument uses the required classical class.

For $R>0$ set
\[
 A_{i,R}=A_i\cap\bigcap_{l\ne i}\{f_l\ne0\}
                    \cap\{\psi_i>-R\}.
\]
Choose $\varepsilon R<\delta/2$ and consider the ordinary open set
\begin{equation}
 N_i=
 \bigcap_{l\ne i}\{f_l\ne0\}
       \cap\{\psi_i>-R\}
       \cap\{|f_i|<\exp(-1/\varepsilon)\}.
 \label{eq:r3-carrier-active-open}
\end{equation}
The finite sum $\psi_i$ is continuous on the first intersection,
so this is indeed ordinary open, and it contains $A_{i,R}$.
On $N_i$ the $i$th branch of the maximum is greater than
$\delta-\varepsilon R>\delta/2$.  If $l\ne i$, all logarithms
are nonpositive and $\psi_l\leq\log|f_i|$.  The $l$th branch is
therefore less than
\[
 \delta+(1-2\delta)\cdot1-1=-\delta.
\]
Consequently, as an equality of actual functions on this ordinary
open neighborhood,
\begin{equation}
 b_{\varepsilon,\delta}
      =\delta+(1-2\delta)h_i+\varepsilon\psi_i
      \quad\text{on }N_i.
 \label{eq:r3-carrier-active-branch}
\end{equation}
The bounds $0\leq h_i\leq1$ are what make this an ordinary-open
argument even when the $h_i$ are discontinuous.  No set defined by
comparing two discontinuous psh functions is declared open.

Every $f_l$, $l\ne i$, is an analytic unit on $N_i$.  Thus $\psi_i$
is original smooth and $D\psi_i=0$ there.  Ordinary locality,
positive homogeneity, and the smooth pluriharmonic invariance of
Lemma~\ref{lem:r2-harmonic-frame-invariance} imply
\begin{equation}
 \mu_{S_i,b_{\varepsilon,\delta}}|_{N_i}
       =(1-2\delta)^q\mu_{S_i,h_i}|_{N_i}.
 \label{eq:r3-carrier-local-measure}
\end{equation}
This uses only unit logarithms and the original smooth product rule
over separately closed tail currents.  It does not assume potential
independence for an arbitrary difference whose ambient current has
zero $D$.

The measure $\mu_{S_i,h_i}$ is carried by $A_i$: the background
vanishes on the ordinary open complement of $A_i$, and restriction
and recursion give vanishing of its products there.  Moreover
\eqref{eq:r3-carrier-cross-null} gives
\[
 \mu_{S_i,h_i}(A_l)=0\quad(l\ne i).
\]
The increasing union of $A_{i,R}$, as $R\to\infty$, is
$A_i\setminus\bigcup_{l\ne i}A_l$, which therefore carries the
whole measure.  Positivity, background linearity,
\eqref{eq:r3-carrier-pasted-normalized}, and
\eqref{eq:r3-carrier-local-measure} yield, for every Borel $E$,
\[
 \begin{split}
 C_S(E)
 &\geq\mu_{S,b_{\varepsilon,\delta}}(E)\\
 &\geq(1-2\delta)^q
            \sum_{i=1}^m\mu_{S_i,h_i}(E\cap A_{i,R}).
 \end{split}
\]
The current identities on the ordinary neighborhoods imply the
displayed identities on their Borel subsets by uniqueness of the
coefficient-Radon measures.  For each $R$ choose a permitted
$\varepsilon$; no convergence theorem for these changing testers
is needed.  Monotone convergence on the right as $R\to\infty$,
followed by $\delta\downarrow0$, gives
\[
 C_S(E)\geq\sum_{i=1}^m\mu_{S_i,h_i}(E).
\]
Taking independent suprema over the finite tuple $(h_1,\ldots,h_m)$
proves \eqref{eq:r3-carrier-raw-additivity}, including infinite
values.

Apply the raw identity to ordinary open sets.  Taking the infimum
over one common open neighborhood of $A$ first gives
$C_S^*(A)\geq\sum_iC_{S_i}^*(A)$.  If the right side is finite,
choose, for any $\eta>0$, ordinary opens $G_i\supset A$ with
\[
 C_{S_i}(G_i)\leq C_{S_i}^*(A)+\eta/m.
\]
The \emph{finite} intersection $G=\bigcap_iG_i$ is ordinary open,
contains $A$, and satisfies
\[
 C_S(G)=\sum_iC_{S_i}(G)
       \leq\sum_iC_{S_i}^*(A)+\eta.
\]
Let $\eta\downarrow0$.  If the right side is infinite the earlier
lower bound already proves equality.  This proves
\eqref{eq:r3-carrier-outer-additivity}.
\end{proof}

\begin{corollary}[Transfer of regularity through finitely many carriers]
\label{cor:r3-carrier-regularity-transfer}
Under Theorem~\ref{thm:r3-analytic-carrier-additivity}, fix an ordinary
core $W\Subset\Omega$.  If each $C_{S_i}$ is Borel outer regular
on $W$, the same is true of $C_S$.  If each $C_{S_i}^*$ on subsets
of $W$ is a finite Choquet capacity, the same is true of $C_S^*$.
If, in addition, each such capacity has compact inner capacitability
on Suslin subsets of $W$, so does $C_S^*$.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
Borel outer regularity follows from the two additive formulas.
Finiteness, monotonicity, countable subadditivity, increasing-set
continuity, and decreasing-compact continuity are preserved by a
finite sum.  For compact inner capacitability, approximate each of
the finitely many summands from below by a compact $K_i\subset A$.
The finite union of the $K_i$ is a compact subset of $A$ and
simultaneously gives the required lower approximation for the sum.
\end{proof}

\begin{corollary}[The crossing coordinate supports have regular capacity]
\label{cor:r3-crossing-support-capacity}
In the notation of
Theorem~\ref{thm:r3-relative-dirichlet-no-classical}, let
$S=[Z_1]+[Z_2]$ on the ordinary unit polydisc $V$.  For every Borel
$E\subset V$ one has the exact same-domain identity
\begin{equation}
 \operatorname{Cap}_S(E;V)
   =\operatorname{Cap}_{[1]_{Z_1}}(i_1^{-1}E;Z_1)
     +\operatorname{Cap}_{[1]_{Z_2}}(i_2^{-1}E;Z_2).
 \label{eq:r3-crossing-capacity-sum}
\end{equation}
The corresponding equality holds for outer capacities of arbitrary
subsets.  On every fixed ordinary core $W\Subset V$, the relative
capacity is Borel outer regular, and its outer capacity is a finite
Choquet capacity with compact inner capacitability on Suslin subsets
of $W$.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
Take $f_1=T_1$ and $f_2=T_2$.  The additional equations defining
$Z_i$ when $p>1$ cause no problem: the theorem only asks that
$[Z_i]$ be carried by the corresponding zero set, not that this
zero set equal $Z_i$.  The restrictions of $f_2$ to $Z_1$ and of
$f_1$ to $Z_2$ are nonzero coordinate functions.  Their zero sets
are proper intrinsic analytic divisors, whose intrinsic ambient
capacity is zero.  This follows from the proved ambient polar
theorem, Corollary~\ref{cor:r2-ambient-polar-outer-zero}, on a
countable relatively compact exhaustion of these polydiscs; every
normalized tester on the full polydisc restricts to each local
domain.  Thus the full raw intrinsic capacities of the divisors
are zero.

Each coordinate projection $V\to Z_i$ is an analytic retraction.
Proposition~\ref{prop:r2-retract} identifies raw and outer relative
capacities for $[Z_i]$ with the respective intrinsic capacities.
It verifies \eqref{eq:r3-carrier-cross-null} and, together with
Theorem~\ref{thm:r3-analytic-carrier-additivity}, proves
\eqref{eq:r3-crossing-capacity-sum} and its outer version.

For fixed $W\Subset V$, its inverse image $i_i^{-1}W$ is an ordinary
relatively compact core in the intrinsic smooth polydisc $Z_i$.
Apply the ambient exhaustion-domain result
Corollary~\ref{cor:r2-local-choquet} on each such intrinsic core,
and then Corollary~\ref{cor:r3-carrier-regularity-transfer}.
Pullback of a decreasing compact family to the closed subspace
$Z_i$ is again a decreasing compact family, and inverse image
preserves increasing unions.  For Suslin sets, intersection with
the closed coordinate support remains Suslin; compact inner
approximants on the two supports have compact images in $W$.
These observations verify every stated core property for the
ambient relative capacity.
\end{proof}

In particular, for this background and any compact $H\subset V$,
\[
 \operatorname{Cap}_S(H;V)=\operatorname{Cap}_S^*(H;V).
\]
If $0\leq\chi\leq1$ is original smooth with compact support and
$\chi=1$ near $H$, every ordinary weak limit $\nu$ of the finite
measures $\chi\,\mu_{S,h}$, $0\leq h\leq1$, satisfies
\[
 \nu(H)\leq\operatorname{Cap}_S(H;V).
\]
Indeed, for an ordinary open neighborhood $G$ of $H$ contained in
$\{\chi=1\}^{\circ}$, the weak open-set inequality gives
$\nu(H)\leq\nu(G)\leq\operatorname{Cap}_S(G;V)$; take the
infimum over such $G$ and use compact outer regularity.  The
local CLN estimate gives the required uniform finite mass for this
compactly localized family.

\paragraph{Scope.}
This is an all-$q$ positive result for an explicit class of
backgrounds carried by ambient polar analytic sets.  It coexists
with the failure of an ambient classical relative Dirichlet solution
in Theorem~\ref{thm:r3-relative-dirichlet-no-classical} and with the
change of measure under ambient usc regularization in
Proposition~\ref{prop:r3-relative-completed-dirichlet-example}.
Thus good raw capacity regularity does not by itself supply an
ambient classical representative for every relative completed
potential.  The proof supplies neither compact outer regularity
for an arbitrary fixed $S$ nor a countable-carrier additivity
theorem.  Its finite pasting and pairwise relative-capacity-null
hypothesis are substantive.

% END source: base_r3/sections/analytic_carrier_capacity_additivity.tex


% BEGIN frozen or reviewed source: base_r3/sections/capacity_order_approximation.tex
\section{Order approximation of backgrounds and countable carrier sums}
\label{sec:r3-capacity-order-approximation}

All capacities below use the same normalized actual classical testers
on the fixed ordinary domain $\Omega$ and the original CLD test sheaf.
Fix $q=n-p\geq1$.  A positive background means a symmetric,
original-positive, separately $d'$- and $d''$-closed current of
bidegree $(p,p)$.

\begin{proposition}[Uniform capacity control under order approximation]
\label{prop:r3-capacity-order-approximation}
Let $S$ and $S_N$ be positive backgrounds on $\Omega$ such that
\[
 0\leq S_N\leq S,\qquad S_N\longrightarrow S
\]
as currents against every original compact smooth complementary
test.  The sequence need not be increasing.  Fix a compact set $K_0$
and ordinary opens satisfying
\[
 K_0\subset W,\qquad \overline W\subset W_1\Subset\Omega.
\]
There are a constant $C$ and finitely many original positive compact
smooth complementary test forms $\theta_\ell$, supported in $W_1$,
which depend only on these compact stages, the degree, and the
chosen finite CLN presentations, such that
\[
 e_N=C\sum_\ell\langle S-S_N,\theta_\ell\rangle\longrightarrow0
\]
and
\begin{align}
 0\leq C_S(E)-C_{S_N}(E)&\leq e_N
       &&\text{for every Borel }E\subset K_0,\label{eq:r3-order-raw}\\
 0\leq C_S^*(A)-C_{S_N}^*(A)&\leq e_N
       &&\text{for every subset }A\subset K_0.\label{eq:r3-order-outer}
\end{align}
All capacities in these formulas are finite.  Consequently Borel
outer regularity on $K_0$ passes from all the $C_{S_N}$ to $C_S$.
The properties of being a finite Choquet capacity and of compact
inner capacitability on Suslin subsets of $K_0$ likewise pass from
the corresponding outer capacities to $C_S^*$.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
Put $R_N=S-S_N$.  This is a positive background.  The finite-test
relative CLN estimate, with the normalized bound $0\leq h\leq1$,
gives
\begin{equation}
 C_{R_N}(W)
    \leq C\sum_\ell\langle R_N,\theta_\ell\rangle=e_N.
 \label{eq:r3-order-tail-cln}
\end{equation}
The forms and the constant are fixed before choosing $N$ or the
background: the background occurs in this estimate only through
its displayed positive pairings.  One obtains the estimate on $W$
by a finite covering of $\overline W$ and the CLN cutoffs supported
in $W_1$.  Weak convergence on the original test sheaf proves
$e_N\to0$.  The same estimate for $S$ proves finiteness on $W$.

For every normalized tester $h$, background linearity gives the
identity of positive Radon measures
\[
 \mu_{S,h}=\mu_{S_N,h}+\mu_{R_N,h}.
\]
Consequently every Borel $F\subset W$ satisfies
\[
 C_{S_N}(F)\leq C_S(F)
       \leq C_{S_N}(F)+C_{R_N}(W)
       \leq C_{S_N}(F)+e_N.
\]
This proves \eqref{eq:r3-order-raw}.

For $A\subset K_0$, the ordinary opens containing $A$ and contained
in $W$ are cofinal for each outer-capacity infimum: replace an
arbitrary open neighborhood $G$ by $G\cap W$.  Applying the last
inequalities to these ordinary opens and taking infima proves
\eqref{eq:r3-order-outer}.  There is no exchange of an infimum with
an infinite sum and no infinite intersection of open sets.

Suppose every $C_{S_N}$ is Borel outer regular on $K_0$.  For Borel
$E\subset K_0$ the two estimates give
\[
 C_S(E)\leq C_S^*(E)
       \leq C_{S_N}^*(E)+e_N
       =C_{S_N}(E)+e_N\leq C_S(E)+e_N.
\]
Let $N\to\infty$.  This proves Borel outer regularity.

Outer capacities are monotone and countably subadditive.  Uniform
convergence in \eqref{eq:r3-order-outer} preserves increasing-set
continuity and decreasing-compact continuity.  Explicitly, if
$A_j\uparrow A\subset K_0$, then for every $N$,
\[
 C_S^*(A)\leq C_{S_N}^*(A)+e_N
       =\lim_j C_{S_N}^*(A_j)+e_N
       \leq\lim_j C_S^*(A_j)+e_N;
\]
letting $N\to\infty$ proves continuity from below.  For decreasing
compacts $L_j\downarrow L\subset K_0$, the analogous inequality
\[
 \lim_j C_S^*(L_j)
       \leq\lim_j C_{S_N}^*(L_j)+e_N
       =C_{S_N}^*(L)+e_N\leq C_S^*(L)+e_N
\]
proves continuity from above.  Finiteness was already proved.

Finally, if $A\subset K_0$ is Suslin and each approximating outer
capacity has compact inner capacitability, then
\[
 C_S^*(A)\leq C_{S_N}^*(A)+e_N
       =\sup_{L\subset A\text{ compact}}C_{S_N}^*(L)+e_N
       \leq\sup_{L\subset A\text{ compact}}C_S^*(L)+e_N.
\]
Let $N\to\infty$ and use monotonicity for the reverse inequality.
If compact outer regularity also holds, the final compact terms can
be written as raw capacities.  This proves all the assertions.
\end{proof}

\begin{corollary}[Countably many separated analytic carriers]
\label{cor:r3-countable-carrier-capacity}
Let $(S_i)_{i\geq1}$ be positive backgrounds and suppose the partial
sums $P_N=\sum_{i=1}^N S_i$ converge weakly, on every original
compact test, to a locally coefficient-Radon positive background
$S$.  Assume that there are analytic functions $f_i$ on $\Omega$
with $|f_i|\leq1$ such that $S_i$ is carried by $\{f_i=0\}$ and
\[
 C_{S_i}(\{f_l=0\};\Omega)=0\qquad(i\ne l).
\]
On every compact $K_0\subset\Omega$ one then has
\begin{align*}
 C_S(E;\Omega)&=\sum_{i\geq1}C_{S_i}(E;\Omega)
       &&\text{for Borel }E\subset K_0,\\
 C_S^*(A;\Omega)&=\sum_{i\geq1}C_{S_i}^*(A;\Omega)
       &&\text{for arbitrary }A\subset K_0.
\end{align*}
The series are finite and their tails tend to zero uniformly over
the stated subsets of $K_0$.  If each component capacity has Borel
outer regularity and the stated Choquet properties on such compact
cores, the same is true of $S$.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
Positivity of the series implies $0\leq P_N\leq S$: for fixed $N$,
$S-P_N$ is the weak limit of its positive tails.  It is separately
closed.  Apply Proposition~\ref{prop:r3-capacity-order-approximation}
to $P_N$.  For every finite $N$, Theorem
\ref{thm:r3-analytic-carrier-additivity} identifies the raw and outer
capacities of $P_N$ with the corresponding finite sums.  The uniform
errors of Proposition~\ref{prop:r3-capacity-order-approximation}
then give exactly the two asserted series identities, together with
uniform convergence of their tails.  Finite-component regularity
passes first to each $P_N$ and then to $S$ by the proposition.
\end{proof}

\paragraph{Scope.}
The additional step for a countable family is the uniform CLN tail
estimate, not an infinite logarithmic maximum or an infinite
intersection of ordinary open neighborhoods.  The order-approximation
result does not assert that an arbitrary positive closed background
admits approximants with already regular capacity.  In particular
it does not close the remaining unrestricted arbitrary-background
compact outer-regularity problem.

% END source: base_r3/sections/capacity_order_approximation.tex


% BEGIN frozen or reviewed source: base_r3/sections/weak_closure_capacity_attempt.tex
\section{What weak compactness does and does not give for outer capacity}
\label{sec:r3-weak-closure-capacity-attempt}

The following abstract reduction isolates a remaining relative-capacity
gate.  Its counterexample concerns a family of ordinary Radon measures;
it is not a realization by non-Archimedean BT measures over one fixed
background.  In particular it does not refute arbitrary-background
compact outer regularity.

Let $X$ be a compact metrizable space and let $\mathcal M$ be a
nonempty family of finite positive Radon measures with
$\sup_{\mu\in\mathcal M}\mu(X)<\infty$.  Put
$\mathcal N=\overline{\mathcal M}^{\,w}$, a weakly compact family.
For Borel $E$ and arbitrary $A\subset X$, respectively, define
\[
 c(E)=\sup_{\mu\in\mathcal M}\mu(E),\qquad
 c^{\mathrm o}(A)=\inf_{\substack{G\supset A\\G\text{ open}}}c(G),
 \qquad
 \widehat c(A)=\sup_{\nu\in\mathcal N}\nu^*(A).
\]
Here $\nu^*$ is the Radon outer measure, equivalently its open-cover
outer regularization.  The open outer capacity $c^{\mathrm o}$ must
not be confused with a Borel-envelope extension.

\begin{proposition}[The weakly closed measure-family capacity]
\label{prop:r3-weak-closure-capacity-facts}
The function $\widehat c$ is finite, monotone, countably subadditive,
continuous from below on arbitrary increasing subsets, and continuous
from above on decreasing compact sets.  Moreover
\begin{align}
 \widehat c(G)&=c(G)&&\text{for open }G,\label{eq:r3-closure-open}\\
 \widehat c(K)&=c^{\mathrm o}(K)&&\text{for compact }K.\label{eq:r3-closure-compact}
\end{align}
For Borel $E$ one has
\[
 c(E)\leq\widehat c(E)\leq c^{\mathrm o}(E),\qquad
 \widehat c(E)=\sup_{K\subset E\text{ compact}}c^{\mathrm o}(K).
\]
The latter compact inner-capacitability identity also holds for Suslin
subsets, with $\widehat c$ defined above on arbitrary subsets.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of elementary
Radon regularity, weak compactness, and the Choquet capacitability
theorem.  This is not a new analytic estimate for BT products.

\begin{proof}
Finiteness, monotonicity, and countable subadditivity follow directly
from those of the individual finite Radon outer measures.  We spell
out continuity from below on arbitrary, possibly nonmeasurable sets.
For a fixed finite Radon measure $\nu$ and $A_j\uparrow A$, choose
Borel hulls $B_j\supset A_j$ with $\nu(B_j)=\nu^*(A_j)$, and put
$H_j=\bigcap_{k\geq j}B_k$.  Then $A_j\subset H_j$, the Borel sets
$H_j$ increase, and $\nu(H_j)=\nu^*(A_j)$.  Thus
\[
 \nu^*(A)\leq\nu\!\left(\bigcup_jH_j\right)
             =\lim_j\nu^*(A_j)\leq\nu^*(A).
\]
Taking suprema commutes with this increasing supremum, so
$\widehat c(A_j)\uparrow\widehat c(A)$.

Let $K_j\downarrow K$ be compact.  Choose
$\nu_j\in\mathcal N$ with
$\nu_j(K_j)\geq\widehat c(K_j)-1/j$ and pass to a weakly convergent
subsequence with limit $\nu\in\mathcal N$.  For every fixed $l$,
eventually $K_j\subset K_l$, and the closed-set weak inequality gives
\[
 \lim_j\widehat c(K_j)\leq\limsup_j\nu_j(K_l)\leq\nu(K_l).
\]
Let $l\to\infty$.  The result is at most $\nu(K)\leq\widehat c(K)$;
monotonicity supplies the reverse inequality.

If $\nu\in\mathcal N$, take $\mu_j\in\mathcal M$ converging weakly
to it.  For open $G$, the open-set weak inequality gives
$\nu(G)\leq\liminf_j\mu_j(G)\leq c(G)$.  The inclusion
$\mathcal M\subset\mathcal N$ gives \eqref{eq:r3-closure-open}.
It follows at once that $\widehat c(K)\leq c^{\mathrm o}(K)$.
For the reverse inequality let
$G_j=\{x:\operatorname{dist}(x,K)<1/j\}$ and choose
$\mu_j\in\mathcal M$ with $\mu_j(G_j)\geq c(G_j)-1/j$.
After passing to a weak limit $\nu\in\mathcal N$, compare $G_j$
with each fixed compact neighborhood
$F_l=\{x:\operatorname{dist}(x,K)\leq1/l\}$.  The same closed-set
inequality yields
\[
 c^{\mathrm o}(K)=\lim_jc(G_j)\leq\nu(F_l).
\]
Let $l\to\infty$ to get $c^{\mathrm o}(K)\leq\nu(K)$, proving
\eqref{eq:r3-closure-compact}.  Empty sets are immediate.

The two inequalities for Borel $E$ follow from the inclusion of
families and from open covers.  For Borel inner regularity, only
two suprema are interchanged:
\[
 \widehat c(E)
 =\sup_{\nu\in\mathcal N}\sup_{K\subset E\text{ compact}}\nu(K)
 =\sup_{K\subset E\text{ compact}}\widehat c(K).
\]
For Suslin sets the established finite Choquet axioms imply the
same inner-capacitability formula by the Choquet theorem, in the
classical form recalled in \cite[\S8]{BT82}.  Suslin means analytic
in descriptive set theory, not an analytic subspace over $K$.
\end{proof}

\begin{proposition}[The exact missing outer-regularity implication]
\label{prop:r3-outer-closure-equivalence}
The ordinary-open extension $c^{\mathrm o}$ is finite, monotone,
countably subadditive, outer regular by definition, and continuous
from above on decreasing compact sets.  The following are equivalent:
\begin{enumerate}
\item $c^{\mathrm o}$ is continuous from below on all increasing subsets;
\item $c^{\mathrm o}(E)=\widehat c(E)$ for every Borel $E$.
\end{enumerate}
Under either condition the equality also holds for every Suslin set.
In contrast, the sharp compact weak-closure bound
\[
 \nu(K)\leq c(K)\quad(\nu\in\mathcal N,\ K\text{ compact})
\]
implies $c(E)=\widehat c(E)$ for Borel $E$, but by itself does not
imply $c(E)=c^{\mathrm o}(E)$ for Borel $E$.
\end{proposition}

\begin{proof}
The elementary outer-capacity properties follow by taking open covers;
for countable subadditivity choose errors summing to a prescribed
$\varepsilon$.  If $K_j\downarrow K$ are compact, every open
neighborhood of $K$ contains some $K_j$, which proves compact
continuity from above.

Under the first condition, $c^{\mathrm o}$ is itself a finite Choquet
capacity.  Its compact inner-capacitability on Borel sets, together
with \eqref{eq:r3-closure-compact}, gives
\[
 c^{\mathrm o}(E)=\sup_{K\subset E\text{ compact}}c^{\mathrm o}(K)
                 =\widehat c(E).
\]
Conversely suppose the second condition.  For arbitrary
$A_j\uparrow A$, choose Borel sets $B_j\supset A_j$ with
$c^{\mathrm o}(B_j)=c^{\mathrm o}(A_j)$.  Such hulls are obtained
by taking countable intersections of nearly minimizing open covers;
they are Borel and are not asserted open.  Put
$H_j=\bigcap_{k\geq j}B_k$ and $H=\bigcup_jH_j$.
Then $A_j\subset H_j$, $c^{\mathrm o}(H_j)=c^{\mathrm o}(A_j)$,
and $A\subset H$.  All $H_j,H$ are Borel.  Therefore
\[
 c^{\mathrm o}(A)\leq c^{\mathrm o}(H)=\widehat c(H)
 =\lim_j\widehat c(H_j)=\lim_jc^{\mathrm o}(A_j).
\]
The reverse inequality is monotonicity.  Suslin equality follows
from the same Choquet inner-capacitability formulas.

Finally the sharp compact bound and
\eqref{eq:r3-closure-compact} give $c(K)=\widehat c(K)$ on compacts.
Both raw Borel capacities have compact inner regularity, so they
agree on Borel sets.  The last assertion is proved false by the
following explicit example.
\end{proof}

\begin{example}[A weakly compact family with a Borel outer-capacity jump]
\label{ex:r3-vertical-fiber-capacity}
Take $X=[0,1]^2$ and
\[
 \mathcal M=\{\mu_x:x\in[0,1]\},\qquad
 \mu_x=\delta_x\otimes\mathcal L^1|_{[0,1]}.
\]
This family is weakly compact, and its raw capacity is
$c(E)=\sup_x\mathcal L^1(E_x)$ for Borel $E$.  Define
\[
 E=\bigl(\{0\}\times[0,\tfrac12]\bigr)
    \ \cup\ \bigcup_{j\geq1}
               \bigl(\{1/j\}\times(\tfrac12,1]\bigr).
\]
Then
\[
 c(E)=\widehat c(E)=\tfrac12,
 \qquad c^{\mathrm o}(E)=1.
\]
The set $E$ is both $F_\sigma$ and $G_\delta$.  There are compact
$K_l\uparrow E$ with $c^{\mathrm o}(K_l)=1/2$ for every $l$.
The same example remains valid after replacing $\mathcal M$ by its
weakly compact convex hull.
\end{example}

\noindent\textbf{Status: \texttt{FALSE}} for the proposed abstract
implication ``weakly compact measure family implies ordinary-open
Borel outer regularity.''  Its use to close the BT capacity problem
is a \texttt{FAILED\_ATTEMPT}.

\begin{proof}
For every continuous function on the compact square, its fiber
integral depends continuously on $x$.  Thus $x\mapsto\mu_x$ is
weakly continuous and its image is compact.  All nonempty fibers
of $E$ have length $1/2$, proving the raw and weakly closed values.

Let $G$ be any relatively open subset of the square containing $E$.
It contains the compact set $\{0\}\times[0,1/2]$, so for some
$\eta>0$ it contains
$[0,\eta)\times[0,1/2+\eta)$.  For sufficiently large $j$,
the remainder $\{1/j\}\times(1/2,1]$ is already contained in $E$.
Hence the entire fiber at $1/j$ is contained in $G$, and $c(G)=1$.
This proves $c^{\mathrm o}(E)=1$.

Put
\[
 K_l=\bigl(\{0\}\times[0,\tfrac12]\bigr)
       \cup\bigcup_{1\leq j\leq l}
          \bigl(\{1/j\}\times[\tfrac12+1/(l+2),1]\bigr).
\]
These are increasing compacts with union $E$.  Their raw capacities
are $1/2$, and \eqref{eq:r3-closure-compact} gives
$c^{\mathrm o}(K_l)=1/2$.  This is the claimed failure of continuity
from below, even along an increasing sequence of compact sets.

The displayed union makes $E$ an $F_\sigma$.  Let
$D=\{0\}\cup\{1/j:j\geq1\}$, a closed subset of $[0,1]$.
The complement of $E$ is the union of
\[
 ([0,1]\setminus D)\times[0,1],\qquad
 \{0\}\times(\tfrac12,1],\qquad
 \bigcup_{j\geq1}\{1/j\}\times[0,\tfrac12].
\]
Each is $F_\sigma$, so $E$ is also $G_\delta$.

The weakly compact convex hull is
$\{\tau\otimes\mathcal L^1:\tau\text{ a probability on }[0,1]\}$.
For Borel $E$, integration of the fiber lengths against $\tau$
has supremum exactly their pointwise supremum, since Dirac
probabilities are allowed.  Thus the raw capacity on all Borel
sets, and hence its ordinary-open extension, is unchanged.
\end{proof}

The family need not even give a strongly subadditive capacity.
For example, let $F_1=\{0,1\}\times[0,1/2]$ and
$F_2=(\{0\}\times[1/2,1])\cup(\{1\}\times[0,1/2])$.
Then $c(F_1)=c(F_2)=1/2$, whereas
$c(F_1\cup F_2)=1$ and $c(F_1\cap F_2)=1/2$.

\paragraph{Application and limitation for the present project.}
For a compactly localized family of normalized BT measures, the
relative CLN estimate supplies the finite mass hypothesis.  Weak
closure then gives precisely the identities on opens and compacts
proved above.  It does not identify the resulting inner Choquet
capacity with the original ordinary-open outer capacity on Borel
sets.  The abstract example is not asserted to satisfy any BT
maximum or locality structure, and is not a counterexample to an
arbitrary-$S$ theorem.  A BT-specific approximation, contact, or
separation argument is still required for the unrestricted problem.
The analytic-carrier results in this round provide such arguments
for a concrete class; they do not rely on the disproved shortcut.

% END source: base_r3/sections/weak_closure_capacity_attempt.tex


% BEGIN frozen or reviewed source: base_r3/sections/round3_status.tex
\section{Current theorem correspondence and the advance beyond R2}
\label{sec:r3-status}

Section~\ref{sec:comparison} gives the original theorem-by-theorem
comparison with BT76/82, Demailly, CLD12/25, GK, GR, and Wu.
Section~\ref{sec:r2-status} records the frozen second-round state.
This section states the present assessment. An old open entry is
closed only with the hypotheses of the specific theorem cited below.

The original finite extension $K/\mathbf Q_p$, its absolute value,
and all original compact smooth complementary CLD tests are retained.
An arbitrary background means a fixed symmetric current in the
original positive cone, separately $d'$- and $d''$-closed.
Put $q=n-p$. The all-degree relative energy and top-capacity
statements have $q\geq1$. They do not construct the wedge of two
arbitrary positive closed currents.

\subsection{What was already complete before this continuation}

R and B are \texttt{INTERNAL\_INPUT}; their classical regularization,
original coefficient-Radon control and background-$[1]$ mixed calculus
are not discoveries of this round. A2's geometric adelic nefness
theorem is also complete under its stated global hypotheses.

R1 already established the seven requested fixed-background bounded
BT properties, including symmetry and weighted joint decreasing
continuity. Its finite-test relative CLN estimate is a
\texttt{DIRECT\_COROLLARY} of R/B. R2 supplied the completed class
$\mathcal A_{S,b}$, actual increasing limits, polar restriction,
fixed-domain capacity comparison, ambient classical representatives,
the qualified ambient polar/negligible/capacity-zero equivalence,
ambient extremal mass formulas and local Choquet theory. Those
results remain frozen dependencies.

\subsection{The eleven directions after the third round}

\begin{longtable}{@{}>{\raggedright\arraybackslash}p{.04\textwidth}>{\raggedright\arraybackslash}p{.19\textwidth}>{\raggedright\arraybackslash}p{.40\textwidth}>{\raggedright\arraybackslash}p{.28\textwidth}@{}}
\hline
 & Direction & Current qualified result & Remaining limitation\\
\hline\endfirsthead
\hline
 & Direction & Current qualified result & Remaining limitation\\
\hline\endhead
1 & Locally bounded psh &
\texttt{PROVED}: B's actual classical class and R2's ambient
representative theorem. Proposition
\ref{prop:r3-no-relative-classical-representative} gives a nonzero
polar background whose completion has no classical representative. &
The unrestricted arbitrary-$S$ representative assertion is
\texttt{FALSE}, even on a connected smooth bidisc.\\[4pt]
2 & Positive closed currents and products &
\texttt{PROVED}: the arbitrary-fixed-$S$ bounded calculus and its
completion. New locally finite coefficient-mass zero extension across a closed locally
complete pole set is Theorem~\ref{thm:r3-zero-extension}. &
The arbitrary-background potential quotient remains \texttt{OPEN}.
Normal traces settle smooth algebraic-immersion and specified weak-closure
backgrounds, not every positive closed current.\\[4pt]
3 & CLN estimates &
The finite-test mass estimate remains a \texttt{DIRECT\_COROLLARY}.
Theorem~\ref{thm:r3-I-capacity} adds a genuinely relative
all-degree energy-to-capacity bound. &
Constants retain the finite local controls, common bound and collar
hypotheses. No global strictly psh form is assumed.\\[4pt]
4 & Symmetry and weighted continuity &
The requested decreasing theorem was already proved. New results
give simultaneous tail-convex-block compactness and strong local
measure convergence in the normalized-psh dual distance, including
completed limits. &
Nonlinear weak limits need not survive convexification. Unrestricted
varying-background weighted stability remains \texttt{FALSE}.\\[4pt]
5 & Locality and maxima &
Ordinary locality, Borel maximum locality and ambient maximality
are retained. Proposition~\ref{prop:r3-relative-contact-inequality}
adds the contact measure inequality as a direct corollary. &
Strict gap sets of discontinuous potentials are not automatically
ordinary open. Closed polar sets need not satisfy the locally
complete hypothesis needed for zero extension.\\[4pt]
6 & BT capacities &
Ambient core Choquet theory is retained. New results treat nonpolar
remaining degree one, separated analytic carriers in every degree,
and order limits with a uniform CLN tail. See
Theorems~\ref{thm:r3-nonpolar-linear-compact-regularity} and
\ref{thm:r3-analytic-carrier-additivity}, and
Proposition~\ref{prop:r3-capacity-order-approximation}. &
For entirely arbitrary $S$, sharp compact regularity and the further
Borel outer-regularity gate remain \texttt{OPEN}. Abstract weak
compactness alone does not supply them.\\[4pt]
7 & Quasicontinuity and capacity convergence &
Bounded relative quasicontinuity and monotone capacity convergence
are retained. The common-collar energy, psh-dual distance and
relative capacity determine the same bounded topology. &
The full normalized psh family is not capacity precompact.
The limit class may be $\mathcal A_{S,b}$ without an ambient
classical representative.\\[4pt]
8 & Pluripolar and negligible sets &
The qualified ambient equivalences are retained. Section
\ref{sec:r3-pluripolar-decomposition} gives a canonical polar/nonpolar
decomposition with its countable coefficient-control hypothesis. &
Relative measures may charge ambient polar analytic sets. A current
carried by a pole set need not have topological support contained
in that nonclosed set.\\[4pt]
9 & Comparison and domination &
Fixed-$S$ compact-gap comparison and q.e. domination are retained.
Theorem~\ref{thm:r3-one-sided-dini-stability} gives one-sided
ambient stability; power bounds yield a quantitative uniform
estimate for strong convergence of data. &
Ordinary weak-data convergence is insufficient, even under one
uniform power-capacity bound. General relative pointwise
domination still requires detection of its capacity ambiguity.\\[4pt]
10 & Extremal functions &
Ambient obstacle orthogonality and capacity/mass formulas are
retained. On the prescribed exhaustion domains, weighted divisorial obstacles
give a continuous mass map, a finite-data existence proof, and
active-set complementarity. &
The ambient extremal-contact formula is false for general relative
capacity, including specified smooth backgrounds. No arbitrary-$S$
orthogonality theorem is used in the existence proofs.\\[4pt]
11 & Dirichlet problems &
\texttt{PROVED}: ambient compactly supported data with a bounded
subsolution, and compactly supported Dini-capacity data on extendible
domains; the latter solutions are continuous. Specified maximal
boundary lifts give nonconstant boundary versions. A matching bounded
zero-boundary subsolution permits noncompact data. &
An arbitrary-relative classical existence theorem from a continuous
matching subsolution is \texttt{FALSE} in every $p,q\geq1$.
Completed branch gluing supplies a qualified replacement;
general domains, boundary data and completed equations remain open.\\
\hline
\end{longtable}

\subsection{Principal new proofs and their dependencies}

\paragraph{Relative energy and strong topology.}
For bounded potentials with one common smooth classical collar value,
or with the proved protected reduction to such a collar,
\[
 I_S(u,v)=\int (u-v)
          \bigl(\operatorname{MA}_S(v)-\operatorname{MA}_S(u)\bigr)
       \geq0.
\]
The mixed carré-du-champ identity and repeated Cauchy--Schwarz
transfer give
\[
 \operatorname{Cap}_S^*(\{|u-v|>\delta\};\Omega)
       \leq C\delta^{-2} I_S(u,v)^{2^{-q}}.
\]
The midpoint energy gap yields Cauchy tail convex blocks. The
normalized-psh dual distance satisfies
$d_S(u,v)\leq C I_S(u,v)^{1/2}$ and $I_S(u,v)\leq4B d_S(u,v)$.
The localized strong-convergence proof retains the cutoff derivative
term. For each pair of potentials it first fixes the already
constructed mixed background, and only then regularizes. This
prevents a circular use of an unproved varying Borel energy multiplier.

\paragraph{Relative capacity on analytic carriers.}
For $S_i$ carried by $\{f_i=0\}$, $|f_i|\leq1$, and pairwise
relative-null cross carriers, normalized testers $h_i$ are combined by
\[
 b_{\varepsilon,\delta}
  =\max\left(0,
       \delta+(1-2\delta)h_i+\varepsilon
                        \sum_{l\ne i}\log|f_l|:1\leq i\leq m\right).
\]
Explicit uniform margins make one branch active on an ordinary open
neighborhood of the relevant part of its carrier, even when $h_i$
is discontinuous. There the logarithms are smooth units with zero
curvature. This proves exact raw capacity additivity on all Borel
sets and outer additivity on arbitrary subsets. For countable sums,
the extra ingredient is the uniform CLN estimate for the positive
tail background, not an infinite maximum or an infinite open
intersection. These are actual normalized testers on the same domain.

\paragraph{Ambient nonzero Dirichlet existence.}
An extendible exhaustion gives a common exterior for the energy.
The constrained energy maximum is attained using the midpoint
capacity estimate and the ambient representative theorem. Positive
obstacle variation gives one measure inequality. The sublevel
capacity iteration gives a bound independent of the initially imposed
floor. The floor is then inactive on every compact perturbation,
and negative variation yields the other inequality. The obstacle
orthogonality is the already proved ambient theorem based on B;
no arbitrary-relative orthogonality is assumed.

For $\mu(E)\leq C(E)\omega(C(E)^{1/n})^n$, the sufficient scalar
condition is $\int_0^a\omega(t)\,dt/t<\infty$. Its dyadic estimate
gives an explicit bound and one-sided stability. Uniform convergence
of B's regularizing rows is then proved; it is not assumed from the
start. Dini's elementary uniform-convergence lemma is used only on
the already continuous ordinary boundary. The radial measure in
Theorem~\ref{thm:r3-nondini-modulus-obstruction} proves the converse
for a universal guarantee on a fixed polydisc. It does not classify
each individual measure satisfying a weaker bound.

\paragraph{A necessary change in the relative solution class.}
For two intersecting coordinate $q$-planes and $S=[Z_1]+[Z_2]$,
the datum $\delta_{z_1}+2^q\delta_{z_2}$ forces the two intrinsic
zero-boundary solutions to take incompatible values at their common
origin. Yet a continuous matching subsolution and every fixed
positive power-capacity bound are available. This proves failure of
the proposed ambient classical solution class. The explicit increasing
completed solution has continuous approximants with
\[
 \operatorname{MA}_S(a_j)
   =\mu+\frac{2^q-1}{j}\delta_{\zeta_j}.
\]
Thus the classical relative Dirichlet image is not even closed under
total variation with one common compact support and potential bound.
The completed gluing theorem handles independent intrinsic branch
solutions, without requiring incompatible intersection values to agree.
The same crossing background has regular capacity by analytic-carrier
additivity. Capacity regularity and classical representability are
therefore separate issues.

\subsection{Additional original-source pins}

El Mir's extension theorem is Theorem II.1, printed p.~15,
with proof on pp.~15--20; the restriction corollary II.5 is on
pp.~21--22. Demailly III.2.3, pp.~139--140, is the corresponding
complex template. The CLD12 support/dimension lemma 3.2.5 and
CLD25 7.1.12--13 have their particular hypotheses; they are not
silently substituted for the arbitrary-original-current extension
proved here.\cite[II.1, II.5]{ElMir84}\cite[III.2.3]{Dem12}

Ko\l odziej 1996, Theorem 1 and Proposition 2, pp.~13--15, give
the capacity-sublevel template. Its Theorem 2, p.~16, has additional
local-capacity and modulus hypotheses for existence. In the 1998
paper, Lemma 2.3.1 is on pp.~81--83; Theorem B, p.~96, supplies
bounds and continuity for an existing solution. Theorem C, p.~97,
with proof in \S3.4, pp.~103--111, is the bounded-subsolution
existence theorem, with the boundary qualification stated on p.~111.
The present non-Archimedean existence argument is supplied in full,
rather than inferred from the name of either classical theorem.
\cite[Theorems 1--2]{Kol96}\cite[Theorems B--C]{Kol98}

Dabbek--Elkhadhra 2006, Definition 2.1 and Proposition 2.2,
p.~471, with questions on p.~472, provide a precise relative-capacity
template. The follow-up source register is a scoped search, not a
claim that those classical questions are globally open today.
\cite[\S2]{DE06}

The projection discussion checks the actual topological-product
category in CLD25 \S9.1 and its tensor-density results. Mihatsch
Theorem 3.23, printed p.~58, gives refined flat skeleton maps;
Definition 2.18 gives piecewise smooth fibre integrals, and
Example 2.45(3), pp.~41--42, distinguishes balanced pushforwards
from $\delta$-forms. These do not establish the original projected-test
continuity required by the potential-quotient argument.
\cite[\S9.1--9.4]{CLD25}\cite[2.18, 2.45(3), 3.23]{Mih24}

\subsection{The smallest unresolved gates}

\paragraph{Q: potential quotients for arbitrary original backgrounds.}
For $u,v\in\Cb(U)$ with $D[u]=D[v]$ on all original tests,
prove $D((u-v)S)=0$ for every allowed $S$. A continuous representative
of the bounded classical difference would suffice. Normal traces
settle smooth algebraic immersions over the original field. The projection
route has an explicit continuous-coefficient sufficient criterion,
but no theorem covering every original projected test. An analytic
fibre product is not the topological product of its factor spaces,
and a same-field graph argument is not the all-extensions curve test.

\paragraph{C1: sharp compact control for a completely arbitrary $S$.}
For the fixed localized family of normalized relative BT measures,
show that every weak-closure measure $\nu$ satisfies
$\nu(K)\leq\operatorname{Cap}_S(K;\Omega)$ on compact cores.
The all-degree convex-block result does not preserve the original
nonlinear weak limit. The analytic-carrier and order-approximation
theorems do not assert that every positive closed current admits
approximants of those types.

\paragraph{C2: Borel outer regularity beyond compact control.}
Establish increasing-set continuity of open outer relative capacity,
or the equivalent Borel identification with the weak-closure capacity
in Proposition~\ref{prop:r3-outer-closure-equivalence}. The explicit
vertical-fibre example has a convex weakly compact measure family
and sharp compact control, but fails this outer-regularity step.
It is an abstract counterexample to that shortcut, not an actual
non-Archimedean BT counterexample.

\paragraph{D: the appropriate relative completed Dirichlet problem.}
The unrestricted ambient classical assertion is false. The replacement
must specify a completed solution class, a boundary condition which
survives its limits, and a justified uniqueness statement. The finite
analytic-branch construction gives a concrete model, not the general
existence theorem. For the ambient problem, broader domains and
arbitrary continuous boundary data still require proved barriers or
an appropriate homogeneous solution.

\paragraph{Comparison with the other calculi.}
An equality on all GK $\delta$-tests needs its own comparison theorem.
The full nonpositive, nonclosed and off-diagonal GK calculus is not a
consequence of the positive BT construction. A2 remains a completed
geometric input. Strong Green forms and general heights remain later
applications and do not replace any of Q, C1, C2 or D.

The next continuation should work on these specific gates and test
each proposed mechanism against the moving-atom, radial-modulus,
intersecting-support, vertical-fibre and analytic-product examples.
Reproving the accepted bounded calculus or auditing R/B/A2 afresh
would not address the remaining obstacles.

% END source: base_r3/sections/round3_status.tex

\clearpage
\part*{Part II. Fourth-round proofs and counterexamples}
\addcontentsline{toc}{part}{Part II. Fourth-round proofs and counterexamples}

% BEGIN frozen or reviewed source: sections/potential_quotient_next.tex
\section{Compatible piecewise smooth tests for the potential-quotient problem}
\label{sec:r4-potential-quotient-next}

The full question remains whether
\[
 D[u]=D[v],\quad u,v\in\mathcal C_b(U)
       \quad\Longrightarrow\quad D((u-v)S)=0
\]
for every allowed fixed positive separately closed current $S$.
The results in this section address the projected-test obstruction
identified in R3. They enlarge the admissible test class in a
quantitative way; they do not assume that all projected tests
belong to that enlarged class.

The field is the prescribed finite extension $K/\mathbf Q_p$,
with its original absolute value. All smooth forms, current
equalities and coefficient measures are original CLD objects.
Positivity is the original positive semidefinite coefficient cone
of R, not an assumption of strong positivity. Let $U$ be an
allowed ordinary open, with a fixed finite moment presentation
$t=(t_1,\ldots,t_r)$ on a neighborhood of a compact set $H$.
Each $t_i$ is the logarithm of the norm of an analytic unit.
Enlarge this presentation once to contain every coordinate used
by the finitely many forms and cutoffs below, and put
\[
 \beta=\sum_{i=1}^r d't_i\wedge d''t_i.
\]
All currents $T,T_j$ below are symmetric. Let $T$ have bidegree
$(p,p)$ and put $q=n-p$.

\subsection{A uniform estimate near an affine moment wall}

\begin{lemma}[Normal mass in a shrinking wall tube]
\label{lem:r4-wall-normal-mass}
Assume $q\geq1$. Let
$\ell=a_0+\sum_i a_i t_i$ be a nonconstant affine expression
in the moment variables. There are finitely many original
positive compact smooth test forms $\Theta_b$ and a constant
$C$, fixed independently of $T$ and $0<\epsilon\leq1$, such that
every original-positive separately closed $T$ satisfies
\begin{equation}\label{eq:r4-wall-normal-mass}
 \bigl(d'\ell\wedge d''\ell\wedge\beta^{q-1}\wedge T\bigr)
       (H\cap\{|\ell|\leq\epsilon\})
       \leq C\epsilon\sum_b\langle T,\Theta_b\rangle.
\end{equation}
In particular the displayed positive measure gives zero mass
to $H\cap\{\ell=0\}$.
\end{lemma}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
The inputs are the original coefficient-Radon domination and
the smooth product rule over a separately closed current.

\begin{proof}
Choose a nonnegative compact original smooth cutoff $\chi$
equal to one on a neighborhood of $H$. Put
\[
 g_\epsilon=\sqrt{\ell^2+\epsilon^2}.
\]
Since $D\ell=0$, one has
\[
 Dg_\epsilon
  =\frac{\epsilon^2}{(\ell^2+\epsilon^2)^{3/2}}
                          d'\ell\wedge d''\ell
  \geq\frac1{2^{3/2}\epsilon}
       \mathbf1_{\{|\ell|\leq\epsilon\}}
                          d'\ell\wedge d''\ell.
\]
The inequality is an inequality of positive coefficient measures
after wedging with $T\wedge\beta^{q-1}$. Separate closedness
and compact Stokes give
\[
 \langle Dg_\epsilon\wedge T,\chi\beta^{q-1}\rangle
       =\langle T,g_\epsilon D\chi\wedge\beta^{q-1}\rangle.
\]
The left side is nonnegative. The functions $g_\epsilon$ have
one finite bound on $\operatorname{supp}\chi$ for
$0<\epsilon\leq1$. Apply the fixed finite positive controls
to the one smooth form $D\chi\wedge\beta^{q-1}$; Borel
coefficient multiplication bounds the right side by
$C\sum_b\langle T,\Theta_b\rangle$. Multiplication by
$2^{3/2}\epsilon$ proves the estimate. Decreasing the tube to
the wall proves the final assertion by continuity of the fixed
locally finite measure. Constant affine expressions require no
estimate: their normal covector is zero.
\end{proof}

\begin{lemma}[Variation of a normal-factor coefficient]
\label{lem:r4-wall-normal-factor}
Under the preceding assumptions, let $\xi$ and $\zeta$ be
forms of bidegrees $(q-1,q)$ and $(q,q-1)$, respectively, whose
Borel coefficients on the fixed presentation have a fixed bound
on $H$. For the finite coefficient measures defined by that
presentation,
\begin{equation}\label{eq:r4-normal-factor-variation}
 \begin{split}
 &|\mu_{T,d'\ell\wedge\xi}|(H\cap\{|\ell|\leq\epsilon\})
  +|\mu_{T,d''\ell\wedge\zeta}|(H\cap\{|\ell|\leq\epsilon\})\\
 &\hspace{25mm}\leq C\epsilon^{1/2}
                           \sum_b\langle T,\Theta_b\rangle.
 \end{split}
\end{equation}
The controls can be chosen for any prescribed finite list of
such bounded-coefficient forms. Their coefficient measures vanish
on the wall. All constants remain valid on common refinements
of this fixed presentation.
\end{lemma}

\begin{proof}
This is the finite-matrix Cauchy--Schwarz argument, now with a
Borel indicator of the tube. A coefficient of $d'\ell\wedge\xi$
is a symmetric cross tensor between a $q$-covector containing
$d'\ell$ and one of the fixed coordinate $q$-covectors.
The original positive cone contains every positive semidefinite
square, including nondecomposable ones. Matrix Young inequality,
summed over the fixed finite index sets, therefore gives
\[
 |\mu_{T,d'\ell\wedge\xi}|(E)
 \leq C
 \bigl((d'\ell\wedge d''\ell\wedge\beta^{q-1}\wedge T)(E)\bigr)^{1/2}
 \bigl((\beta^q\wedge T)(E)\bigr)^{1/2}
\]
for Borel $E\subset H$. To obtain total variation, multiply the
coefficients also by an arbitrary bounded Borel scalar of norm
at most one before the same matrix inequality and take the
supremum. Antisymmetric components are annihilated by the
original-positive convention. The identical argument with the
primed factors interchanged treats the other term. The first
factor is $O(\epsilon^{1/2})$ by the preceding lemma, and the
second is uniformly controlled by the fixed positive trace.
Absorb both controls into their finite sum. The matrix inequalities
pull back to every common refinement, so no new constant is
needed for additional scalar test coordinates. At $\epsilon=0$
use decreasing continuity of the variation measures.
\end{proof}

\subsection{Piecewise smooth tests with tangential matching}

\begin{definition}[The finite original piecewise test class]
\label{def:r4-piecewise-tests}
On an open subset of $\mathbf R^r$, fix a finite affine hyperplane
arrangement. A compatible piecewise smooth $(q,q)$-form is a
collection of smooth forms on the closed chambers, smooth on
neighborhoods of those chambers, whose restrictions to the
tangent space of each common face agree. Only finitely many
chambers meet the compact stages considered. Pull it back by
the original unit moment map $t$ and multiply by a fixed
original compact smooth scalar cutoff. Finite sums of such
expressions are called finite original piecewise tests here.

On a face, choose the ambient tensor supplied by any incident
chamber when representing such a test by Borel coefficients.
The next theorem shows that its pairing with a positive separately
closed current is independent of that choice. This definition
does not include an arbitrary piecewise form on an abstract
skeleton that has not been represented by an original moment map.
\end{definition}

\begin{theorem}[Uniform approximation of compatible piecewise tests]
\label{thm:r4-piecewise-test-continuity}
Let $\eta$ be a finite original piecewise test of bidegree $(q,q)$.
There are original compact smooth forms $\eta_\epsilon$, on one
fixed compact support stage, and finitely many fixed positive
controls $\Theta_b$ such that for every original-positive
separately closed $(p,p)$-current $T$,
\begin{equation}\label{eq:r4-piecewise-uniform-pairing}
 |\langle T,\eta_\epsilon\rangle-\langle T,\eta\rangle|
       \leq C\epsilon^{1/2}\sum_b\langle T,\Theta_b\rangle
       \qquad(0<\epsilon\leq\epsilon_0).
\end{equation}
The Borel-coefficient pairing on the right is independent of the
chamber representatives on every face. Consequently, if
$T_j\to T$ weakly on all original compact smooth tests and all
these currents are original-positive and separately closed, then
\[
 \langle T_j,\eta\rangle\longrightarrow\langle T,\eta\rangle.
\]
For $q=0$, the conclusion instead follows from uniform
approximation of continuous piecewise smooth scalar functions;
no wall-normal estimate for top-degree measures is asserted.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
It suffices to work in one fixed presentation with one compact
scalar cutoff. Enlarge the compact image stage slightly so that
every chamber coefficient and its first derivatives are bounded
there. Choose a standard smooth compactly supported convolution
kernel of radius $\epsilon$ in $\mathbf R^r$, convolve the
Borel coefficient tensors, and then pull back and multiply by
the fixed cutoff. This gives the forms $\eta_\epsilon$ with
fixed compact support.

We give the error estimate explicitly. If there is just one wall
$\ell=0$, choose a linear coordinate system with first coordinate
$\ell$. Tangential matching and the ordinary Hadamard lemma give
for the difference of the two smooth branches
\[
 \eta_+-\eta_-
       =\ell\theta+d'\ell\wedge\xi+d''\ell\wedge\zeta,
\]
with bounded smooth coefficients on the compact stage.
The tangential part therefore has size $O(\epsilon)$ inside
the tube, while the other two parts are controlled by
Lemma~\ref{lem:r4-wall-normal-factor}.

For finitely many walls the same decomposition of the error
follows without imposing a normal-crossing condition. Consider
two points $x,y$ with $|x-y|\leq\epsilon$ and the line segment
joining them. For a generic segment, it crosses each affine
wall at most once and avoids intersections of distinct walls.
Along each chamber segment the change of the coefficients is
bounded by a constant times that segment's length. Their total
is $O(\epsilon)$. At a crossing of the wall $\ell_b=0$, the
jump has zero restriction to its tangent hyperplane. Linear
algebra therefore writes that jump as
$d'\ell_b\wedge\xi_b+d''\ell_b\wedge\zeta_b$, with uniformly
bounded coefficients. A fixed linear right inverse on this
finite-dimensional normal ideal supplies these bounds. A crossing
can occur only when $|\ell_b(x)|\leq C_b\epsilon$.

If $x$ lies on walls, first move it an arbitrarily small distance
into its chosen incident chamber and take the limit. The same
finite bound holds; any initial changes of incident chamber are
normal-ideal terms for walls through $x$. Degenerate crossing
segments can likewise be obtained by limits of generic ones.
All choices may be made by the finite chamber ordering, so the
resulting bounded coefficients are Borel. Averaging over $y$
in the convolution gives the pointwise coefficient decomposition
\[
 \eta_\epsilon-\eta
   =\epsilon\theta_\epsilon+
       \sum_b\bigl(d'\ell_b\wedge\xi_{b,\epsilon}
                    +d''\ell_b\wedge\zeta_{b,\epsilon}\bigr),
\]
where the coefficients have one bound and each normal-factor
term is supported in $|\ell_b|\leq C_b\epsilon$. The fixed
compact cutoff is included in these coefficients. Finite
coefficient domination controls the first term by $O(\epsilon)$,
and Lemma~\ref{lem:r4-wall-normal-factor} controls the remaining
finite sum by $O(\epsilon^{1/2})$. This proves
\eqref{eq:r4-piecewise-uniform-pairing}.

The difference between any two ambient representatives on a
face lies in the ideal generated by the normals of the walls
containing that face. The last assertion of
Lemma~\ref{lem:r4-wall-normal-factor} makes its pairing zero,
proving independence there. In degree zero, face matching is
ordinary continuity and convolution converges uniformly, so R's
coefficient estimate suffices.

Finally weak convergence makes every finite positive control
$\langle T_j,\Theta_b\rangle$ bounded. Fix $\epsilon$, pass to
the weak limit with the original smooth $\eta_\epsilon$, and
then let $\epsilon\downarrow0$ in the two uniform error bounds.
This proves the stated convergence. A finite localization proves
the theorem for every test in the defined class. Neither a
strong-positivity assumption nor a change of original test sheaf
was used.
\end{proof}

\subsection{The resulting improvement of the projection gate}

\begin{proposition}[Piecewise projected tests annihilate the kernel]
\label{prop:r4-piecewise-projection-gate}
Let $\pi:U\times_K\mathbb D_s\to U$, $\dim U=n\geq1$, and let
$u,v\in\mathcal C_b(U)$ satisfy $D[u]=D[v]$. For
$\alpha\in A_c^{n,n}(U\times_K\mathbb D_s)$, put
$P_\alpha=\pi_*[\alpha]$. Suppose, locally on a finite compact
cover, that $P_\alpha$ is the ordinary current of a finite
original piecewise $(n-1,n-1)$-test as in
Definition~\ref{def:r4-piecewise-tests}. Then
\[
 \langle D[(u-v)\circ\pi],\alpha\rangle=0.
\]
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
new test-continuity theorem and the fixed-measure reduction in R3.
The universal assertion for all projected tests is still
\texttt{CONDITIONAL} on the stated geometric representation.

\begin{proof}
The fixed signed measure
$\nu_\alpha=DP_\alpha=\pi_*[D\alpha]$ is finite.
Choose simultaneous locally bounded smooth decreasing psh rows
$u_j\downarrow u$, $v_j\downarrow v$ on each compact stage.
Their positive separately closed currents $Du_j,Dv_j$ converge
to the same current. Theorem~\ref{thm:r4-piecewise-test-continuity}
therefore gives
\[
 \langle P_\alpha,D(u_j-v_j)\rangle\longrightarrow0.
\]
Smooth Stokes identifies this with
$\int(u_j-v_j)\,d\nu_\alpha$. Bounded convergence against this
one fixed signed measure gives
$\int(u-v)\,d\nu_\alpha=0$. The original projection identity
then gives the assertion. Finite original smooth partitions
handle the stated finite local cover. No pullback of an arbitrary
original current has been presumed.
\end{proof}

\begin{example}[Tangential matching cannot be omitted]
\label{ex:r4-incompatible-wall-test}
On the original two-dimensional torus take an ordinary product
annulus with coordinates $x=\log|T_1|$, $y=\log|T_2|$ in
$(-2,2)$. Put $c_j=(-1)^j/j$ and
\[
 a_j=\max(x,c_j),\qquad T_j=D[a_j],\qquad T=D[\max(x,0)].
\]
The functions are continuous classical psh and converge uniformly,
so $T_j\to T$ as positive separately closed original currents.
Choose nonnegative smooth compact scalar functions $\rho,\sigma$
in $(-1,1)$ with $\rho(0)>0$ and $\int\sigma(y)\,dy>0$.
The Borel test
\[
 \eta=\mathbf1_{\{x>0\}}\rho(x)\sigma(y)\,d'y\wedge d''y
\]
has a tangential jump at $x=0$. The original torus calibration gives
\[
 \langle T_j,\eta\rangle
  =\mathbf1_{\{c_j>0\}}\rho(c_j)\int\sigma(y)\,dy.
\]
This pairing has two distinct subsequential limits. Thus arbitrary
bounded piecewise coefficient tests do not satisfy the theorem.
Every point and current is over the original $K$; real radii define
Berkovich Gauss points and do not require a field extension.
This is a counterexample to the enlarged-test shortcut, not to
the potential-quotient conjecture.
\end{example}

\paragraph{What remains unproved.}
Mihatsch's \emph{$\delta$-Forms on Lubin--Tate spaces},
arXiv:2112.10018v4, Section 2.2 defines compatible piecewise smooth
forms, and Definition 2.18 and Theorem 3.23 provide piecewise
smooth fiber integration on suitably refined flat skeletons.
Example 2.45(3) explicitly warns that a flat pushforward need
not descend to the specified smaller sheaf of forms. The theorem
above handles original moment-piecewise tests, but does not prove
that the skeletonwise pushforward $P_\alpha$ has such a common
original representation. Establishing that descent, with the
original calibration and compatibility under further refinements,
is the remaining geometric gate for this route.

Even after that projection gate, the already proved normal-trace
argument applies only to graphs lying in its smooth algebraic
closed-immersion scope over the original field. It does not
automatically supply arbitrary analytic graph restrictions after
every complete field extension. Nor has a continuous representative
of every bounded classical difference in the original $D$-kernel
been constructed here. The unrestricted arbitrary-background
potential quotient is therefore still \texttt{OPEN}.

% END source: sections/potential_quotient_next.tex


% BEGIN frozen or reviewed source: sections/split_projection_tests.tex
\section{A projection class with an explicit original piecewise formula}
\label{sec:r4-split-projection-tests}

This section verifies a concrete subclass of the geometric hypothesis
in Proposition~\ref{prop:r4-piecewise-projection-gate}. The relative
coordinate functions are split over the original base. No finite
cover, scalar extension, or altered normalization is used.

Let $U$ be an allowed smooth $n$-dimensional ordinary algebraic
analytic neighborhood over the fixed $K$, with $n\geq1$.
Let $a_1,\ldots,a_m$, $m\geq1$,
be analytic functions on $U$, such that $a_i-a_j$ is an analytic
unit whenever $i\ne j$. Let $t=(t_1,\ldots,t_r)$ be an original
unit moment map on $U$, enlarged to include
\[
 b_{ij}=\log|a_i-a_j|\quad(i\ne j).
\]
Thus each $b_{ij}$ is a coordinate, or a fixed affine expression
in these coordinates. Put $b_{ii}=-\infty$ as a notation, not
as a smooth coordinate. On the complement of the graphs of the
$a_i$ in $U\times_K\mathbf A^{1,\an}$, with vertical coordinate
$T$, consider the original moment map
\[
 \Phi(x,T)=\bigl(t(x),\log|T-a_1(x)|,\ldots,
                             \log|T-a_m(x)|\bigr).
\]

A split-moment test is a finite sum of forms
\begin{equation}\label{eq:r4-split-test}
 \alpha=(\chi\circ\pi)\Phi^*\omega,
 \qquad \chi\in A_c^{0,0}(U),
 \qquad\omega\text{ smooth of bidegree }(n,n),
\end{equation}
where the coefficient support of $\omega$ is contained in one
fixed bounded box in the last $m$ coordinates. One may also
use a compact cutoff in the first $r$ real coordinates.
All coefficients are defined on a real open neighborhood of
the relevant compact stages and extended with such cutoffs.
The form is extended by zero across the graphs. This extension
is original smooth: near each graph its own logarithmic
coordinate leaves the lower end of the support box. The base
support is compact, the functions $a_i$ are bounded there, and
the vertical norm is bounded on the support. Thus $\alpha$ has
compact support in the actual analytic product.

\begin{theorem}[The split-moment projection formula]
\label{thm:r4-split-projection-formula}
For every split-moment test $\alpha$, the ordinary current
$\pi_*[\alpha]$ is represented by a finite original piecewise
smooth $(n-1,n-1)$-test in the sense of
Definition~\ref{def:r4-piecewise-tests}. The representation has
no additional current supported on base walls. It is compatible
with arbitrary further original moment refinements on the base.
Consequently, if $u,v\in\mathcal C_b(U)$ and $D[u]=D[v]$, then
\[
 \langle D[(u-v)\circ\pi],\alpha\rangle=0
\]
for every split-moment test. The same assertion applies to a
split-moment test supported in $U\times_K\mathbb D_s$ and then
viewed on that ordinary disc product.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the explicit
finite-original presentation. The resulting kernel equality is
a \texttt{DIRECT\_COROLLARY}. The weighted description of a
split line-fiber skeleton used below is an
\texttt{EXTERNAL\_KNOWN} input, also verified directly by the
Gauss norm and product calibration in the proof.

\begin{proof}
It suffices to treat one summand of
\eqref{eq:r4-split-test}. At a base point $x$, write
$\sigma_{i,s}(x)$ for the image of the radius-$e^s$ Gauss
point centred at $a_i(x)$ in the $\mathscr H(x)$-line. The
original point lies in the analytic fiber over $x$. The Gauss
norm gives the exact identities
\begin{equation}\label{eq:r4-split-Gauss-coordinates}
 \log|T-a_k|(\sigma_{i,s}(x))
       =\max\{s,b_{ik}(x)\},\qquad
 \sigma_{i,s}(x)=\sigma_{j,s}(x)
       \Longleftrightarrow s\geq b_{ij}(x).
\end{equation}
For $k=i$ the maximum is simply $s$.

The part of the line-fiber skeleton seen by the functions
$T-a_i$ is the union of these rays. A point off their union
lies in a complementary disc on which all their logarithmic
norms are constant, so it contributes no vertical differential
direction. Count each common ray only once by assigning it to
its first centre. With
\[
 L_1=+\infty,\qquad L_i=\min_{j<i}b_{ij}\quad(i\geq2),
\]
the $i$th assigned ray has parameter range $s<L_i(x)$.
Indeed it has not yet met any earlier ray exactly when
$s<b_{ij}(x)$ for every $j<i$. Endpoints have one-dimensional
measure zero. Each noncollapsed ray has the standard normal
weight $ds$, with no residue-degree factor: its centre is
$\mathscr H(x)$-rational. Any structural degree of a base
skeleton cell is already in the base calibration.

Here is a formula in one finite real moment space. Define the
continuous piecewise affine map
\[
 F_i(t,s)=
   \bigl(t,\max(s,b_{i1}(t)),\ldots,
                       \max(s,b_{im}(t))\bigr).
\]
Let $\mathcal I_s$ denote contraction with the normal top
coefficient in the product calibration: for a form of bidegree
$(n,n)$ it extracts its $d's\wedge d''s$ coefficient as a
form of bidegree $(n-1,n-1)$ in the base variables, with the
usual product-integration sign convention. Put
\begin{equation}\label{eq:r4-split-fiber-formula}
 \eta(t)=\sum_{i=1}^m
          \int_{-\infty}^{L_i(t)}\mathcal I_s(F_i^*\omega)(t,s)\,ds.
\end{equation}
For $i=1$ the upper endpoint is $+\infty$.
This is a finite ordinary integral: the $i$th vertical coordinate
of $F_i$ is $s$, so the bounded support of $\omega$ in that
coordinate supplies one fixed compact $s$-interval for every
term, including real $t$ in a neighborhood of the compact
base moment image. In particular the formula is defined in
that full real neighborhood, not just on its tropical image.

We verify that $\eta$ is a compatible piecewise smooth form
there. Subdivide by the finitely many hyperplanes ordering the
affine functions $b_{ij}$ and the fixed vertical support endpoints.
On each chamber, $L_i$ is one fixed affine function, and the
breakpoints of the maxima in $F_i$ have a fixed order. Splitting
the integral at these breakpoints expresses every coefficient
as a finite sum of integrals of smooth functions with affine
endpoints. The coefficients are therefore smooth up to that
closed chamber, with smooth extensions obtained from the same
signed-integral formulas. They and their first derivatives are
bounded on the chosen compact stage.

For matching on a common base face, fix tangent vectors to
that face. Except for the finitely many limiting values of $s$
at a breakpoint, the maps $F_i$ and their derivatives in those
tangent vectors have the same limiting values from the adjacent
chambers. This follows directly from continuity of the maximum:
where two affine expressions agree on the face, their tangential
derivatives agree there. The same is true of the endpoint $L_i$.
The integrands are uniformly bounded on the fixed compact
$s$-interval. Dominated convergence of these tangential
coefficients therefore proves that the two restrictions of
$\eta$ to the face agree. When an interval collapses its
integral tends to zero; no term is retained on that lower
dimensional base face. Repeating the argument on intersections
of faces proves the full compatibility required in
Definition~\ref{def:r4-piecewise-tests}.

It remains to prove the ordinary current identity, including its
test scope. Let $\gamma$ be any original compact smooth
$(1,1)$-test on $U$, and refine the base moment presentation
to include all its coordinates and a finite partition. On each
calibrated full-dimensional base cell, adjoining $T-a_i$ gives
the above Gauss rays with the product calibration. The original
integration formula and Borel Fubini on these finite cell
presentations give
\[
 \int_{U\times\mathbf A^1}\alpha\wedge\pi^*\gamma
       =\int_U\chi\,t^*\eta\wedge\gamma.
\]
One can equivalently read this product description in Mihatsch,
arXiv:2112.10018v4, proof of Proposition 3.21, equations
(3.27)--(3.31) and its final weight verification. In the present
split case the centres and all pairwise distances are already
defined on the original base, so none of the preliminary field
or finite-cover reductions in that general proposition is used.
Under a refinement, the formulas
\eqref{eq:r4-split-Gauss-coordinates} are unchanged and only
the base calibrated cells subdivide. Their weights sum to the
original weight. This verifies the displayed identity for all
original tests, not only for tests in the first chosen chart.

The identity is exactly
$\pi_*[\alpha]=[\chi t^*\eta]$. It also excludes any extra
ordinary current on a collapsed base cell: the identity has
already been checked on the entire original complementary test
sheaf. Geometrically, a full-dimensional source cell here has
one relative direction and full base dimension; a cell over a
base face has total dimension at most that face's dimension
plus one and contributes no additional top integral.

The projected test is thus in the finite original piecewise
class. Proposition~\ref{prop:r4-piecewise-projection-gate}
proves the kernel equality. If the original support is in the
disc product, extension by zero into the affine-line product
is original smooth and the same argument applies.
\end{proof}

\begin{corollary}[Split polynomial and rational vertical moments]
\label{cor:r4-split-polynomial-projection}
The theorem also applies when the finitely many vertical units
in a test presentation have the form
\[
 g_\ell(x,T)=b_\ell(x)
                  \prod_{i=1}^m(T-a_i(x))^{e_{\ell i}},
 \qquad b_\ell\in\mathcal O_U^\times,\quad e_{\ell i}\in\mathbf Z,
\]
and the compact test admits the indicated split refinement.
The centres and their pairwise differences have the same
assumptions as in the theorem. Strictly smaller analytic
remainders may be discarded only on an ordinary neighborhood
where they preserve the absolute values of every chosen unit.
\end{corollary}

\begin{proof}
Adjoin the base units $b_\ell$ and all the linear factors to
the moment presentation. Then each $\log|g_\ell|$ is an
affine linear combination of the enlarged original logarithms,
so the test becomes a split-moment test. For a permitted
remainder, ultrametricity gives equality of the two logarithmic
functions on the stated ordinary neighborhood, hence equality
of their original smooth differentials and of the represented
test there. This is a \texttt{DIRECT\_COROLLARY}; it does not
assert a splitting of arbitrary polynomials on the original base.
\end{proof}

\paragraph{The remaining descent problem.}
General original units in the vertical variable need not have
this split presentation over the original base. A finite flat
cover or an extension used to split them cannot simply pull back
$D[u]=D[v]$, since precisely that original-current functoriality
is unproved. Nor may pairwise collision loci be removed without
showing that the projected test extends with the required
compatibility there. The next geometric question is therefore
descent of the displayed piecewise formula through finite flat
base presentations, or an independent original-base formula for
nonsplit vertical units. The theorem settles a concrete
multiple-centre projection class and does not claim that it
generates all original tests on the analytic product.

% END source: sections/split_projection_tests.tex


% BEGIN frozen or reviewed source: sections/finite_etale_piecewise_trace.tex
\section{Finite etale trace in the original piecewise test class}
\label{sec:r4-finite-etale-trace}

Throughout this section the ground field is the original finite extension
$K/\mathbf Q_p$, with its original absolute value. A finite etale map
is used only to compute the pushforward of forms and their currents.
No pullback of a general current is presupposed.

\begin{lemma}[Finite tomography of unordered vectors]
\label{lem:r4-finite-vector-tomography}
Fix integers $d,m\geq1$ and let $A=\{0,\ldots,d\}^m$.
For an unordered list $Z=(z_1,\ldots,z_d)$ of vectors in
$\mathbf R^m$, including repetitions, record the ordered list
of $a\cdot z_l$, $1\leq l\leq d$, for every $a\in A$.
These records determine $Z$ uniquely. The set of consistent
records is a finite union of polyhedral sets. On a finite
polyhedral subdivision of that set, the vectors can be listed
as affine linear functions of the records. If two such listings
meet on a face, they agree there as unordered lists.
\end{lemma}

\begin{proof}
The records give the polynomial in $X$
\[
 P_Z(X,a)=\prod_{l=1}^d(X-a\cdot z_l)
\]
for every $a$ in the grid. Each coefficient has degree at most
$d$ in each coordinate of $a$. Successive one-variable polynomial
interpolation on $\{0,\ldots,d\}$ determines $P_Z(X,a)$ as a
polynomial in $X,a_1,\ldots,a_m$. Unique factorization into its
monic linear factors $X-a\cdot z_l$ determines the unordered
list, with multiplicities.

Here is the finite polyhedral assertion explicitly. The records
for the coordinate vectors $e_i\in A$ list the $i$th coordinates
of all $z_l$. Fix their first coordinate in its given order.
Each possible joint listing is obtained by permuting the other
$m-1$ coordinate lists; there are only finitely many choices.
For each choice and each $a$, also choose a permutation matching
the projected coordinates with the recorded ordered list for
$a$. Its validity is a finite system of linear equalities,
together with the ordering inequalities on the records. The
consistent set is the finite union of these polyhedra. A common
subdivision makes one valid listing affine on each cell. The
uniqueness just proved gives the last assertion.
\end{proof}

\begin{lemma}[Extension from a finite polyhedral locus]
\label{lem:r4-polyhedral-form-extension}
Let $C\subset\mathbf R^N$ be a closed finite polyhedral set.
Suppose a form $\tau$ on $C$ is smooth up to each cell of a
finite subdivision and has matching tangential restrictions
on common faces. For every compact $L\subset C$, there is a
real open neighborhood of $L$ on which $\tau$ extends to a
finite compatible piecewise smooth form. On compact stages
the extended coefficients and their first derivatives are bounded.
\end{lemma}

\begin{proof}
Choose a large closed polyhedral box containing $L$ in its
interior and triangulate it together with $C$ and the given
subdivision. After barycentric subdivision, the subcomplex
contained in $C$ is full: the vertices of any simplex which
belong to that subcomplex span a face of that simplex lying
in $C$. Let $\lambda_v$ be the continuous piecewise affine
hat functions of this triangulation and put
\[
 q(x)=\sum_{v\in C}\lambda_v(x),\qquad
 R(x)=\frac{\sum_{v\in C}\lambda_v(x)v}{q(x)}.
\]
On $q>0$ the point $R(x)$ belongs to the face spanned by the
$C$-vertices of the simplex containing $x$. Thus $R$ maps
to $C$ and is the identity on $C$. It is continuous and
piecewise smooth. Work in the interior of the box with
$q>1/2$, which is an open neighborhood of $L$.
On each closed cell the denominator is bounded away from zero.
Pull back the smooth form on the corresponding $C$-face by
$R$. These pullbacks have smooth extensions near their closed
cells and match tangentially, because $R$ and its tangential
derivatives agree on common faces. Their pullbacks agree with
$\tau$ on $C$. A further finite hyperplane subdivision puts
the extension in the form of Definition~\ref{def:r4-piecewise-tests}.
All asserted bounds follow on compact stages from the finite
construction and the lower bound on $q$.
\end{proof}

\begin{theorem}[Original finite etale trace]
\label{thm:r4-finite-etale-piecewise-trace}
Let $p:Z\to U$ be a finite etale map of smooth good Hausdorff
boundaryless $K$-analytic spaces of pure dimension $n$.
The ordinary pushforward of a compact original smooth form,
or of a finite original compatible piecewise smooth form,
is locally represented by finite original compatible piecewise
smooth forms on $U$. Their ordinary currents agree with the
pushforward on every original smooth complementary test.
The representation is compatible with arbitrary further
original moment refinements. In particular it introduces no
additional current supported on a base wall.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
The finite-flat tropical degree formula used to verify the
ordinary-current identity is \texttt{EXTERNAL\_KNOWN}.
The joint-norm reconstruction and the finite-original descent
are proved here.

\begin{proof}
All assertions are local on the base. Fix $x\in U$.
The fiber $p^{-1}(x)$ is finite. Choose pairwise disjoint
ordinary neighborhoods of its points on each of which the
given form has one finite moment presentation. Properness of
$p$ permits shrinking the base so that its entire preimage
is contained in their union. The resulting finitely many
open-and-closed pieces are finite etale over the base. Shrink
again so that their ranks are constant. Treat each piece
separately and sum. We may therefore assume that $p$ has
constant degree $d$ and that the given form is
\[
 \omega(\log|h_1|,\ldots,\log|h_m|)
\]
for analytic units $h_1,\ldots,h_m$ on the whole preimage,
with $\omega$ a smooth or finite compatible piecewise smooth
superform. Finitely many initial moment presentations may be
combined by adjoining their units. Work on a relatively compact
base neighborhood whose preimage closure lies in this presentation.
All $h_i$ and their inverses are uniformly bounded there.

For each $a\in A=\{0,\ldots,d\}^m$, form the unit
$h^a=\prod_i h_i^{a_i}$ and its characteristic polynomial
for the finite locally free algebra $p_*\mathcal O_Z$:
\[
 Q_a(T)=T^d+c_{a,1}T^{d-1}+\cdots+c_{a,d}.
\]
Its coefficients are analytic on the original base, and its
constant coefficient is a unit. At a base point $y$, the roots
over an algebraic closure of $\mathscr H(y)$ are the values
of $h^a$ at its $d$ geometric sheets, counted with multiplicity.
Every embedding of a finite extension of the complete field
$\mathscr H(y)$ preserves its absolute value. Thus repeated
geometric sheets at one Berkovich fiber point contribute exactly
its finite local degree.

We only use these algebraic roots to compute their norms.
Gauss norm multiplicativity gives, for every real $s$,
\begin{equation}\label{eq:r4-characteristic-gauss-norm}
 \max_{0\leq k\leq d}
       \bigl(\log|c_{a,k}(y)|+(d-k)s\bigr)
       =\sum_{l=1}^d\max(s,r_{a,l}(y)),
 \qquad c_{a,0}=1,
\end{equation}
where $r_{a,l}$ are the log norms of these roots. The changes
of slope on the right determine their ordered list. Equivalently,
this is the elementary Newton polygon formula. Its ordered roots
are continuous piecewise affine functions of the finite coefficient
log norms: a choice of polygon vertices gives affine slopes,
and the finitely many possible choices are specified by affine
inequalities.

There is no need to use the logarithm of a coefficient vanishing
at $x$. Choose $B>0$ so that every root log norm of every $Q_a$
on the chosen neighborhood belongs to $[-B,B]$.
The valuations of all lower-hull vertices are then at most $dB$.
If $c_{a,k}(x)=0$, shrink the base so that
$|c_{a,k}|<\exp(-dB-1)$; such a coefficient lies strictly
above the lower Newton hull and may be omitted throughout
this neighborhood. All remaining nonzero coefficients at $x$
are analytic units after a further shrinking. The leading and
constant coefficients are retained. Consequently finitely many
\emph{original analytic unit} moments
\[
 t=(\log|c_{a,k}|)_{a,k\text{ retained}}
\]
determine the ordered root records $r(t)$ by finite continuous
piecewise affine formulas on a full real moment space.

Let $C$ be the subset of that real space where these records
are jointly consistent in the sense of
Lemma~\ref{lem:r4-finite-vector-tomography}. It is a closed
finite polyhedral set after a common subdivision, and it
contains the actual base moment image. On each of its cells
choose an affine listing $z_1(t),\ldots,z_d(t)$ and put
\begin{equation}\label{eq:r4-unordered-form-trace}
 \tau_C=\sum_{l=1}^d z_l^*\omega.
\end{equation}
This does not require the individual sheets to be analytic
sections. It is an expression in their reconstructed norm vectors.

The forms in \eqref{eq:r4-unordered-form-trace} match
tangentially on every common face. Indeed on that face the
two vector lists agree as unordered lists. Subdivide further
by their finitely many coordinate-equality hyperplanes.
On the relative interior of each resulting face a fixed
permutation identifies the affine vector maps. They therefore
have identical tangential derivatives on that face. If some
vectors coincide identically there, their tangential maps and
derivatives coincide as well. Summing the pullbacks proves
the assertion. If $\omega$ was piecewise smooth, also subdivide
by the inverse images of its finitely many walls; its own
tangential compatibility gives the same conclusion.

Lemma~\ref{lem:r4-polyhedral-form-extension} extends $\tau_C$
near each compact base moment image to a finite compatible
piecewise smooth form $\tau$ on a full real neighborhood.
Its pullback $t^*\tau$ is therefore a finite \emph{original}
piecewise form. This extension step is what distinguishes the
claim from an assertion only about forms on an abstract skeleton.

We verify the current identity, which also makes the choice
of extension immaterial. Take any original smooth compact
complementary test $\gamma$ on the base. Refine its moment
presentation to include all the retained coefficients and
use a finite partition on the support. The finite-flat tropical
degree formula identifies source calibrated full-dimensional
cells over every base cell with their finite fiber weights.
Those weights are exactly the multiplicities of the geometric
sheets used above. The source unit log norms on those cells
are the affine vector listings $z_l$; grouping identical
listings adds their multiplicities. Thus the original
integration formula gives
\begin{equation}\label{eq:r4-finite-trace-current-identity}
 \int_Z \omega(\log|h|)\wedge p^*\gamma
           =\int_U t^*\tau\wedge\gamma.
\end{equation}
For precision, the finite-flat degree input is Mihatsch,
arXiv:2112.10018v4, Lemma 3.15; its use after adjoining source
units is Proposition 3.19 and its proof, especially equation
(3.20). The characteristic-polynomial coefficients ensure that
adjoining those units does not add graded transcendence degree.
Indeed at any actual root $\theta$ with $|\log|\theta||\leq B$,
an omitted coefficient satisfies
\[
 \log|c_{a,k}\theta^{-k}|
       <-dB-1+kB\leq-1.
\]
Its monomial is strictly smaller than the leading monomial
$\theta^d$, hence cannot be a maximal term of the initial
characteristic relation. The nontrivial initial relation uses
only retained unit coefficients and still proves graded
algebraicity. This explicitly covers coefficients vanishing
at the original base point.
The same argument works after every additional base refinement.
The equality concerns calibrated weighted piecewise linear
spaces; we do not use an equality of their sheaves of linear
functions, which Remark 3.20 explicitly distinguishes.

On every refined base cell the moment map takes values in
$C$ and its tangent space maps to the tangent space of the
appropriate face of $C$. Consequently the pullback of the
extension agrees with $\tau_C$ there. This justifies using
\eqref{eq:r4-unordered-form-trace} in the full original
integration formula, regardless of how it was extended off $C$.
For piecewise forms the same cell integrals apply; face choices
are immaterial by the original normal-factor wall estimate.
This proves \eqref{eq:r4-finite-trace-current-identity} on
every original test, and hence the stated ordinary-current
identity. It rules out any additional wall-supported current.

Finally cover the compact projected support by finitely many
base neighborhoods just constructed and multiply by an original
partition of unity. Finite sums preserve the asserted class
and its tangential compatibility. This proves the theorem.
\end{proof}

\begin{corollary}[Finite etale preservation of the bounded potential kernel]
\label{cor:r4-finite-etale-kernel}
In the preceding setting assume $n\geq1$, let
$u,v\in\mathcal C_b(U)$, and suppose $D[u]=D[v]$ on all
original tests. Then
\[
 D[u\circ p]=D[v\circ p]
\]
on all original tests of $Z$. This is a statement about the
specified bounded potentials, not a definition of pullback
for arbitrary currents.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
For an original compact test $\alpha$ of bidegree $(n-1,n-1)$
on $Z$, the theorem represents $P_\alpha=p_*[\alpha]$ by
finite original compatible piecewise tests. The signed Radon
measure $\nu=D P_\alpha=p_*D[\alpha]$ is fixed.
Take local bounded smooth decreasing psh rows $u_j\downarrow u$
and $v_j\downarrow v$ on neighborhoods of the projected support,
using a finite original partition if needed. Their positive
closed curvatures converge to the same current. Theorem
\ref{thm:r4-piecewise-test-continuity} gives
\[
 \langle P_\alpha,D(u_j-v_j)\rangle\longrightarrow0.
\]
Compact Stokes identifies the left side with
$\int(u_j-v_j)\,d\nu$. Fixed-measure dominated convergence
identifies its limit with $\int(u-v)\,d\nu$, which is
$\langle D[(u-v)\circ p],\alpha\rangle$ by the definition
of pushforward. This proves the assertion. The compositions
are actual classical functions by their curve definition.
\end{proof}

\paragraph{What this does and does not settle.}
This theorem supplies finite etale descent in the original
piecewise category and makes finite etale kernel pullback a
proved consequence of that descent. Its proof does not appeal
to a pullback of an arbitrary original current. It does not
by itself prove the corresponding statement for arbitrary
complete extensions, for an analytic map of positive relative
dimension, or for arbitrary analytic graph immersions. These
remaining scopes must still be proved separately before the
general background quotient problem can be declared solved.

% END source: sections/finite_etale_piecewise_trace.tex


% BEGIN frozen or reviewed source: sections/disc_projection_kernel.tex
\section{Preservation of the bounded potential kernel by disc projection}
\label{sec:r4-disc-projection-kernel}

We retain the original field $K$ and absolute value. Let $Y$ be
a smooth integral affine $K$-variety of dimension $n\geq1$,
let $U\subset Y^{\rm an}$
be an ordinary open set in the scope of B, and let
$\mathbb D_s\subset\mathbf A_K^{1,\rm an}$ be an ordinary open
disc. All tests and currents in this section are original CLD
tests and currents. Write $\pi:U\times_K\mathbb D_s\to U$.

\begin{lemma}[Rational moment localization of an original test]
\label{lem:r4-rational-test-localization}
An original compact smooth form on an ordinary open subset of
$(Y\times_K\mathbf A^1)^{\rm an}$ is a finite sum of forms
which have the following description. On a neighborhood of
their support their unit moments are rational functions in
$K(Y)(T)$, and their compact scalar localization is a smooth
function of finitely many truncated logarithms of global
regular functions in $K[Y][T]$. Off the zeros of finitely many
nonzero polynomials in that ring, each summand is one global
rational moment expression, extended by zero outside its
compact localization. All equalities are actual pointwise
equalities of norms and of original smooth forms.
\end{lemma}

\begin{proof}
Let $L$ be the compact support. Around each point choose an
original unit presentation and a smaller compact neighborhood
inside it. B, Lemma D.1, replaces its finitely many analytic
units on a slightly larger compact neighborhood by rational
functions with exactly the same pointwise absolute values.
The replacement is on the original space and field. Equality
of logarithmic functions on the ordinary interior also gives
equality of their original smooth differentials.

The ordinary topology of an affine analytification has a basis
of finite norm conditions on its global regular functions.
Adding upper bounds on fixed affine coordinates makes the
closed norm boxes compact. Within the preceding presentation
choose nested such boxes and a scalar bump equal to one on
the inner box and supported on the outer box. Explicitly it
is a finite product of smooth functions of $\log|G|$ for
regular $G$, constant for sufficiently negative arguments
whenever the box includes $G=0$. The constant value extends
smoothly across that zero locus. These are original smooth
compact functions on the whole affine analytification.

Choose finitely many such bumps $b_i$ whose regions of value
one cover $L$. Put $b=\sum_i b_i$, and choose a smooth real
function $R$ which equals $1/t$ for $t\geq1/2$ and is zero
for $t\leq1/4$. Then $\chi_i=b_iR(b)$ have compact support
inside their respective presentations and $\sum_i\chi_i=1$
near $L$. Their scalar expressions involve only the finitely
many regular functions used in the norm boxes. Multiplying
the form by these $\chi_i$ gives the required finite sum.

Write every rational unit as a quotient of polynomials in
$K[Y][T]$. Its numerator and denominator are nonvanishing
near the compact support of the corresponding $\chi_i$.
Extend the smooth real coefficient forms from a neighborhood
of that compact moment image by a real cutoff. The expression
$\chi_i\omega(\log|r_1|,\ldots,\log|r_m|)$ then defines
the given summand globally on the unit locus of the rational
functions, and is zero near its complement because the scalar
support is compactly contained in that unit locus.
After deleting the zero loci of the regular functions used in
the bumps, their truncated logarithms are also ordinary unit
moments. This proves the last assertion. Identically zero
regular functions contribute only constant scalars and may
be omitted from the finite list.
\end{proof}

\begin{lemma}[A common finite etale splitting away from a base divisor]
\label{lem:r4-polynomial-etale-splitting}
Let $P_1,\ldots,P_N\in K[Y][T]$ be nonzero. There is a
nonzero $\Delta\in K[Y]$ and a finite etale surjective map
$p:Z\to Y_0=Y\setminus\{\Delta=0\}$ of constant positive
degree such that on $Z$
\[
 P_l(x,T)=b_l(x)\prod_{i=1}^m(T-a_i(x))^{e_{li}},
 \qquad b_l\in\mathcal O_Z^\times,\quad e_{li}\geq0.
\]
The $a_i$ are regular and pairwise distinct, with each
$a_i-a_j$ a unit. One may adjoin $P_0=T$ so that $m\geq1$.
\end{lemma}

\begin{proof}
Work over the fraction field $K(Y)$. Remove repeated factors
from the product of the $P_l$ and make the resulting polynomial
monic. Characteristic zero makes it separable. A finite normal
splitting field is a finite separable extension of $K(Y)$.
The normalization of the affine variety in that extension
is finite. After deleting the nonzero leading coefficients,
the finitely many denominators just introduced, and the
discriminant of the monic squarefree polynomial, its integral
splitting algebra is finite etale. Equivalently, restrict the
normalization to this etale open. The roots are integral there
and hence regular on the normalization. Their pairwise
differences are units because their product, up to sign and
square, is the invertible discriminant. The original factors
retain their multiplicities, and their leading coefficients
are units. The finite separable extension gives a constant
positive rank on this integral open. Clearing all deleted
denominators gives one nonzero $\Delta\in K[Y]$.
No alteration of the ground field is used.
\end{proof}

\begin{theorem}[The original disc-projection kernel theorem]
\label{thm:r4-disc-projection-kernel}
Let $u,v\in\mathcal C_b(U)$ and suppose
$D[u]=D[v]$ on every original compact complementary test on
$U$. Then
\[
 D[u\circ\pi]=D[v\circ\pi]
\]
on every original compact complementary test on
$U\times_K\mathbb D_s$. The equality holds for the actual
locally bounded classical pullbacks over the original $K$.
It does not define the pullback of an arbitrary positive closed
current under $\pi$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
Its inputs beyond R and B are the already proved R2 ambient
coefficient nonpolarity, the finite etale original trace theorem,
and the split projection formula of this round.

\begin{proof}
Put $f=u-v$ and
\[
 Q=D[u\circ\pi]-D[v\circ\pi].
\]
Composition preserves actual classicality by its all-extension
curve definition. Thus the two terms are positive closed
ambient bounded Bedford--Taylor currents. Every one of their
coefficient variations vanishes on analytic zero loci, by the
proved ambient coefficient-polar theorem. The same is true
of the signed coefficient variations of $Q$.

Fix an arbitrary original compact test $\alpha$ of bidegree
$(n,n)$ on the product, where $n=\dim Y$. Use
Lemma~\ref{lem:r4-rational-test-localization} and linearity
to treat one compactly localized rational moment summand.
Let $P_1,\ldots,P_N$ contain all polynomial numerators and
denominators of its rational moments and all regular functions
used by its scalar norm-box localization. Adjoin $T$.
Apply Lemma~\ref{lem:r4-polynomial-etale-splitting} to this
fixed finite list, obtaining $\Delta$ and $p:Z\to Y_0$.

Let
\[
 H(x,T)=\Delta(x)\prod_{l=0}^N P_l(x,T),
 \qquad P_0=T.
\]
This is a nonzero regular function on the integral affine
variety $Y\times\mathbf A^1$. Choose smooth functions
$0\leq\theta_j\leq1$ on $\mathbf R$ with
$\theta_j(s)=0$ for $s\leq-j-1$,
$\theta_j(s)=1$ for $s\geq-j$, and
$\theta_j(s)\to1$ for every finite $s$. The function
\[
 \zeta_j=\theta_j(\log|H|)
\]
extends by zero at $H=0$ and is original smooth. The form
$\alpha_j=\zeta_j\alpha$ has compact support where every
chosen $P_l$ and $\Delta$ is nonzero. It is a global finite
rational unit moment expression there. It need not have a
bound independent of $j$ in any smooth norm; no such bound
will be used.

Let $U_0=U\cap Y_0^{\rm an}$ and restrict the finite etale
map to $Z_U=p^{-1}(U_0)$. Theorem
\ref{thm:r4-finite-etale-piecewise-trace} and Corollary
\ref{cor:r4-finite-etale-kernel} give
\[
 D[u\circ p]=D[v\circ p]\quad\hbox{on }Z_U.
\]
Let $q=p\times\operatorname{id}$ on the corresponding
disc products. Its pullback $q^*\alpha_j$ is original smooth
and compactly supported. The polynomial factorization expresses
all its vertical unit moments as affine linear combinations
of the base unit logs and the logs $\log|T-a_i|$.
The support avoids every graph $T=a_i$. Because that support
is compact, all these linear-factor logs have finite upper
and lower bounds there. Multiplying their real coefficient
expression by smooth cutoffs equal to one near those ranges
puts its coefficient support inside a bounded box in all
the linear-factor coordinates. An original base cutoff equal
to one near the projected support may likewise be inserted.
It is therefore exactly a split-moment test in the sense of
\eqref{eq:r4-split-test}, on the original finite etale base
$Z_U$. Its extension into the affine-line product remains
smooth and compactly supported. The split projection theorem
gives
\begin{equation}\label{eq:r4-split-upstairs-zero}
 \bigl\langle D[f\circ p\circ\pi_Z],q^*\alpha_j\bigr\rangle=0.
\end{equation}

We now descend this numerical equality without invoking a
pullback of $Q$. If $d$ is the finite etale degree, the ordinary
finite-flat integration formula for smooth top forms gives
the equality of signed Radon measures
\[
 q_*[q^*(D\alpha_j)]=d[D\alpha_j].
\]
The equality holds on every original compact smooth scalar
test and hence on bounded Borel multipliers by Radon uniqueness.
The function $f\circ\pi$ is Borel and bounded on a
neighborhood of the fixed projected support. The definition
of a distributional derivative and the smooth pullback rule
$D(q^*\alpha_j)=q^*(D\alpha_j)$ therefore imply
\[
 \bigl\langle D[f\circ p\circ\pi_Z],q^*\alpha_j\bigr\rangle
      =d\langle D[f\circ\pi],\alpha_j\rangle
      =d\langle Q,\alpha_j\rangle.
\]
Equation~\eqref{eq:r4-split-upstairs-zero} proves
$\langle Q,\alpha_j\rangle=0$ for each $j$.

Finally $0\leq\zeta_j\leq1$ converges to one away from
the fixed analytic set $\{H=0\}$. Every coefficient variation
of $Q$ vanishes on that set. Fixed-measure dominated convergence
on the compact support of $\alpha$ gives
\[
 \langle Q,\alpha\rangle
       =\lim_j\langle Q,\zeta_j\alpha\rangle=0.
\]
There is no simultaneously changing coefficient measure in
this last step. Summing the finitely many original localized
summands proves the theorem for the original arbitrary test.
\end{proof}

\begin{corollary}[Finite-dimensional ordinary polydisc factors]
\label{cor:r4-polydisc-projection-kernel}
Under the same hypotheses, the kernel equality is preserved
by projection from $U$ times any finite-dimensional ordinary
polydisc over $K$.
\end{corollary}

\begin{proof}
Apply Theorem~\ref{thm:r4-disc-projection-kernel} successively.
Each intermediate space is an ordinary open in the smooth
integral affine analytification $Y\times\mathbf A^r$, and
the pulled-back potentials remain actual locally bounded
classical. This is a \texttt{DIRECT\_COROLLARY}.
\end{proof}

\paragraph{Remaining scope of the background quotient question.}
The new theorem resolves the original-test disc and polydisc
projection gate in the stated algebraic-neighborhood scope.
Combining it with the already proved same-field smooth
algebraic normal-trace theorem applies only when the graph
immersion lies in that theorem's established algebraic scope.
It does not silently enlarge that scope to arbitrary analytic
graphs or arbitrary complete field extensions. Nor does it
yet prove that every original distributional $D$-kernel
difference has a continuous representative. The full statement
$D[u]=D[v]\Rightarrow D((u-v)S)=0$ for arbitrary $S$
therefore remains open beyond the separately established
background classes until those additional steps are supplied.

% END source: sections/disc_projection_kernel.tex


% BEGIN frozen or reviewed source: sections/same_field_analytic_kernel.tex
\section{Functoriality of the bounded potential kernel over the original field}
\label{sec:r4-same-field-analytic-kernel}

The ground field throughout the main theorem is the prescribed
finite extension $K/\mathbf Q_p$ with its original absolute value.
An allowed space is good, Hausdorff, boundaryless and quasi-smooth,
as in B. Currents and tests are the original CLD objects.
Only pullbacks of specified functions, and pullbacks of original
smooth test forms, occur below. There is no definition of pullback
for an arbitrary current.

\begin{lemma}[Kernel preservation by a product with an allowed space]
\label{lem:r4-arbitrary-smooth-product-kernel}
Let $X,Z$ be allowed spaces over $K$, and let $u,v\in\mathcal C_b(X)$
satisfy $D[u]=D[v]$. For the projection $p_X:X\times_K Z\to X$,
\[
 D[u\circ p_X]=D[v\circ p_X]
\]
on every original compact complementary test.
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
proved polydisc kernel and the established same-field algebraic
normal-trace restriction.

\begin{proof}
Work locally around an arbitrary point of the analytic product.
Its two images have ordinary algebraic neighborhoods by B,
Lemma D.4. Their product is an ordinary open neighborhood of
the chosen product point. We may therefore replace $X$ and
$Z$ by ordinary opens in smooth affine varieties $X_0,Z_0$
over the same $K$. Restrict to the relevant integral components.
Choose a closed algebraic immersion $Z_0\hookrightarrow\mathbf A^M$.
Corollary~\ref{cor:r4-polydisc-projection-kernel}, on increasing
ordinary polydiscs in $\mathbf A^M$, proves the equality on
$X\times\mathbf A^{M,\rm an}$. The algebraic immersion
\[
 X_0\times Z_0\hookrightarrow X_0\times\mathbf A^M
\]
is a smooth closed immersion of smooth varieties. Apply
Proposition~\ref{prop:r3-normal-restriction} to the two actual
classical pullbacks and then restrict to $X\times Z$.
This proves the assertion locally at every product point.
The degree-zero case for $X$ is immediate, since a finite
function on a zero-dimensional smooth space is locally constant.
\end{proof}

\begin{lemma}[An ordinary rectangular neighborhood of a zero section]
\label{lem:r4-actual-zero-tube}
Let $C$ be a good Hausdorff analytic space, let $c\in C$, and
let $O$ be an ordinary open neighborhood of $(c,0)$ in
$C\times_K\mathbf A^{N,\rm an}$. After shrinking an ordinary
base neighborhood $C_1$ of $c$, there is $s>0$ with
\[
 C_1\times_K\mathbb D_s^N\subset O.
\]
If the zero section over a compact affinoid neighborhood $B$
is already contained in $O$, one may choose $C_1=\operatorname{Int}_C B$
after making $s$ sufficiently small.
\end{lemma}

\begin{proof}
The inverse image of $O$ under the zero section is ordinary
open and contains $c$. Choose an affinoid neighborhood $B$
of $c$ contained in this inverse image. Fix any $R>0$.
The actual analytic domains
\[
 B\times_K\overline{\mathbb D}_r^N\qquad(0<r\leq R)
\]
are compact, are nested as $r$ decreases, and their intersection
is the zero section over $B$. The last statement follows
because a multiplicative seminorm with all normal coordinates
of norm zero factors through their common zero ideal.
If none of these compact tubes were contained in $O$, their
intersections with the closed complement of $O$ inside the
fixed radius-$R$ compact tube would have the finite intersection
property. Compactness would give a point in the zero section
outside $O$, a contradiction. Thus a sufficiently small closed
tube lies in $O$; its base interior times the open disc is
the required ordinary product neighborhood. This argument is
about analytic products, not a replacement by Cartesian
products of their sets of points.
\end{proof}

\begin{theorem}[Same-field analytic functoriality]
\label{thm:r4-same-field-analytic-kernel}
Let $F:Z\to X$ be an arbitrary analytic morphism of allowed
$K$-analytic spaces. If $u,v\in\mathcal C_b(X)$ satisfy
$D[u]=D[v]$ on all original tests, then
\[
 D[u\circ F]=D[v\circ F]
\]
on all original tests of $Z$. The functions are their actual
locally bounded classical pullbacks.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} relative to
the disc-projection result of this round and the accepted
algebraic tubular-neighborhood input R, Lemma 5.2.
The graph in this theorem is allowed to be analytic. Its
restriction is proved by an explicit translation to a zero
section, not by extending the algebraic-immersion hypothesis
of the old restriction theorem without proof.

\begin{proof}
The assertion is local at a point $z\in Z$. If the target is
zero-dimensional near $F(z)$, the two functions there are
locally constant and the conclusion is immediate. Assume
from now on that its local dimension is positive.
By B, Lemma D.4,
choose ordinary algebraic neighborhoods of $z$ and of $F(z)$;
shrink the former so that its image lies in the latter. Thus
the source is an ordinary open in a smooth affine $K$-variety
and the target is an ordinary open in a smooth affine
$K$-variety $Y$. Choose a closed algebraic immersion
$Y\hookrightarrow\mathbf A^N$.

R, Lemma 5.2, gives an ordinary neighborhood $V$ of $F(z)$
in $Y^{\rm an}$, an ordinary ambient neighborhood
$W\subset\mathbf A^{N,\rm an}$, and an analytic isomorphism
\[
 \Phi:V\times_K\mathbb D_s^r\xrightarrow{\sim}W
\]
whose zero section is the given algebraic immersion. Here
$r$ is its codimension. We may shrink $V$ inside the target
open and correspondingly restrict $W$. Let
$R_W=\operatorname{pr}_V\circ\Phi^{-1}:W\to V$.
The functions
\[
 \widetilde u=u\circ R_W,\qquad
 \widetilde v=v\circ R_W
\]
are actual locally bounded classical functions on $W$, since
all-extension classicality is preserved by analytic composition.
They restrict to $u,v$ on the embedded $V$.
The polydisc kernel theorem, followed by the analytic
isomorphism $\Phi$, gives
$D[\widetilde u]=D[\widetilde v]$ on $W$.
An analytic isomorphism identifies the original test sheaves
and their calibrated integrals, so this transfer does not
invoke a general current pullback.

Shrink the source to an ordinary neighborhood $C$ of $z$
such that $F(C)\subset V$. Regard $F$ as an analytic map
$C\to\mathbf A^N$. By
Lemma~\ref{lem:r4-arbitrary-smooth-product-kernel}, the two
functions $\widetilde u(x)$ and $\widetilde v(x)$ on
$C\times W$ have equal original curvature currents.
The analytic automorphism
\[
 A:C\times\mathbf A^{N,\rm an}\longrightarrow
       C\times\mathbf A^{N,\rm an},\qquad
 A(c,t)=(c,F(c)+t)
\]
has inverse $(c,x)\mapsto(c,x-F(c))$. Let
$O=A^{-1}(C\times W)$, an ordinary open containing the
zero section over $C$. On $O$ put
\[
 U(c,t)=\widetilde u(F(c)+t),\qquad
 V'(c,t)=\widetilde v(F(c)+t).
\]
These are actual classical functions and $D[U]=D[V']$
on every original test, by the same isomorphism rule.

Lemma~\ref{lem:r4-actual-zero-tube} supplies an ordinary
neighborhood $C_1$ of $z$ and a radius $a>0$ such that
$C_1\times\mathbb D_a^N\subset O$. Restrict the equality
there. Apply the established zero-section normal-trace lemma
successively in its $N$ normal coordinates. Its base at each
stage is an ordinary open in a smooth algebraic
analytification; its potentials remain actual locally bounded
classical functions. The conclusion is
\[
 D[U|_{t=0}]=D[V'|_{t=0}]
             \quad\hbox{on }C_1.
\]
At the zero section these functions are exactly $u\circ F$
and $v\circ F$. This proves the claimed local equality and
hence the theorem. A zero-dimensional source has no curvature
tests and requires no additional argument.
\end{proof}

\begin{corollary}[Smooth analytic immersion backgrounds]
\label{cor:r4-analytic-immersion-quotient}
Let $j:Z\hookrightarrow X$ be a smooth closed analytic
immersion of allowed spaces over this same $K$. If
$D[u]=D[v]$, then
\[
 D((u-v)[Z])=0
\]
where the local pushforward is defined. More generally the
same is true for $S=j_*Q$ when $Q$ is locally a finite linear
combination of bounded ambient BT products on $Z$, a smooth
separately closed form current, or belongs to the bounded
ambient BT weak-closure class already covered in R3.
\end{corollary}

\begin{proof}
The theorem gives $D[u\circ j]=D[v\circ j]$ on $Z$.
Apply the corresponding R3 quotient statement there and
push forward using Borel multiplication and the ordinary
current pushforward, which commute with $D$ in this local
closed-immersion setting. This is a \texttt{DIRECT\_COROLLARY}.
The ambient current may be carried by the analytic polar
subspace $Z$; no nonpolarity assumption is imposed on it.
\end{proof}

\paragraph{The complete-extension gate remains distinct.}
The theorem applies to arbitrary analytic morphisms over the
original $K$, not merely algebraic maps. It does not view a
curve over an arbitrary complete extension as a finite-type
$K$-analytic curve. Such a change would be invalid in general.
The original equality has not yet been transferred to all
those extension spaces. Consequently this section does not
deduce that both signs of $u-v$ are classically psh, does not
assert a continuous representative for every kernel difference,
and does not declare the arbitrary-background quotient solved.

% END source: sections/same_field_analytic_kernel.tex


% BEGIN frozen or reviewed source: sections/finite_extension_kernel.tex
\section{The finite-extension part of the functoriality question}
\label{sec:r4-finite-extension-kernel}

\begin{corollary}[Finite extensions with their inherited calibration]
\label{cor:r4-finite-extension-kernel}
Let $K'/K$ be a finite extension, endowed with the unique
extension of the prescribed absolute value, and let $X$ be
an allowed $K$-analytic space. If $u,v\in\mathcal C_b(X)$
and $D[u]=D[v]$ on all original $K$-tests, then
\[
 D[u\circ\pi_{K'/K}]=D[v\circ\pi_{K'/K}]
\]
on all original $K'$-tests of $X_{K'}$.
More generally, for any analytic morphism $F:Z\to X_{K'}$
from an allowed $K'$-analytic space,
\[
 D[u\circ\pi_{K'/K}\circ F]
       =D[v\circ\pi_{K'/K}\circ F]
\]
on all its original $K'$-tests.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}}.
Only a finite scalar extension is involved, and the absolute
value is not renormalized. This does not settle the corresponding
assertion for arbitrary complete extensions.

\begin{proof}
Viewed over $K$, the projection $X_{K'}\to X$ is finite
etale. Its source remains good, Hausdorff, boundaryless and
quasi-smooth over $K$. Thus
Corollary~\ref{cor:r4-finite-etale-kernel} gives the first
equality on the original $K$-test sheaf of this same analytic
space.

There is no hidden change of test functions. Viewing a
$K'$-analytic space over $K$ changes neither its analytic
functions nor their absolute values. Its original unit moment
presentations, and hence its original smooth form sheaf,
are identical. Its calibrated top integration currents obey
\[
 [1]_K=[K':K]\,[1]_{K'}.
\]
One checks this on any calibrated full-dimensional moment
cell: the finite degree of its structural map to the original
$K$-torus is the degree of the same map to the $K'$-torus
multiplied by $[K':K]$. Refinement and a finite partition
give the displayed identity on every original smooth top
test. This is the finite-constant calibration rule of B,
Lemma D.5; it does not replace the given absolute value.
The same factor applies to integration with a locally bounded
Borel coefficient, and distributional differentiation
preserves it. Dividing the $K$-current equality by this positive
constant proves the equality on all original $K'$-tests.

For the second assertion, view $Z$ over $K$. Since $K'/K$
is finite separable, this again is an allowed $K$-analytic
space. Apply
Theorem~\ref{thm:r4-same-field-analytic-kernel} to the
$K$-analytic composite $Z\to X_{K'}\to X$. The identical
calibration argument on $Z$ gives the asserted $K'$ equality.
\end{proof}

In particular all analytic curve maps over finite extensions
are covered at the level of these original current equalities.
An analytic curve over a general complete extension need not
be an allowed finite-type $K$-analytic space; the proof does
not treat it as one. The unrestricted complete-extension gate
in the classical definition therefore remains separate.

% END source: sections/finite_extension_kernel.tex


% BEGIN frozen or reviewed source: sections/gauss_extension_kernel.tex
\section{Gauss residue fields: full fibers and exact test spreading}
\label{sec:r4-gauss-extension-kernel}

The field $K$ remains the prescribed finite extension of $\mathbf Q_p$,
with its original absolute value.  Every use of the bounded BT calculus
in this section takes place on an allowed space over this $K$.
No version of R or B over a general complete extension is assumed.
For a positive real polyradius $\mathbf r=(r_1,\ldots,r_s)$, write
\[
 \zeta_{\mathbf r}\in(\mathbf G_m^s)^{\rm an},\qquad
 K_{\mathbf r}=\mathscr H(\zeta_{\mathbf r})
   =\widehat{K(z_1,\ldots,z_s)}_{\mathbf r}
\]
for the Gauss point and its completed residue field.  The number of
parameters $s$ is finite.  Radii are arbitrary positive real numbers.

\subsection{The fixed full-fiber formula}

Let $X$ be an allowed space of pure dimension $n$, and let $A$ be an
ordinary open neighborhood of $\zeta_{\mathbf r}$ in the parameter
torus.  Denote the projections of $X\times_K A$ by $q$ and $p$, and put
\[
 b=\log\max(1,|z_1|/r_1,\ldots,|z_s|/r_s),\qquad
 B_{\mathbf r}=(D[b\circ p])^s.
\]
The function $b\circ p$ is continuous classical psh.  The actual fiber
of $p$ over $\zeta_{\mathbf r}$ is the whole $K_{\mathbf r}$-analytic
space $X_{K_{\mathbf r}}$.  Write $j$ for its identification with the
corresponding closed subset of $X\times_K A$.

\begin{proposition}[The calibrated full Gauss fiber]
\label{prop:r4-full-gauss-fiber}
For every original compact smooth form $\eta$ of bidegree $(n,n)$ on
$X\times_K A$,
\begin{equation}\label{eq:r4-full-gauss-fiber}
 \langle B_{\mathbf r},\eta\rangle
       =\int_{X_{K_{\mathbf r}}}j^*\eta.
\end{equation}
The integral on the right uses the original $K_{\mathbf r}$-calibration.
Its coefficient is exactly one, with no additional factorial.
Moreover, for every locally bounded Borel function $w$ on the product,
\begin{equation}\label{eq:r4-borel-gauss-fiber}
 \langle wB_{\mathbf r},\eta\rangle
       =\int_{X_{K_{\mathbf r}}}(w\circ j)j^*\eta.
\end{equation}
Here pullback of a smooth test means restriction of its analytic unit
moments to the fiber.  No pullback of an arbitrary current is used.
\end{proposition}

\noindent\textbf{Status: \texttt{EXTERNAL\_KNOWN}} for
\eqref{eq:r4-full-gauss-fiber}, and \texttt{DIRECT\_COROLLARY} for its
fixed Borel-weighted version.

\begin{proof}
The unweighted identity is precisely
\cite[Proposition 18.6.1 and Corollary 18.6.2, pp.~532--534]{CLD25},
applied to the analytic functions $z_1,\ldots,z_s$ on the product,
with the parameter denoted $t$ in that corollary equal to one.
Proposition 18.6.1 explicitly defines the right-hand current by
integration on the full completed-residue-field analytic fiber.
Corollary 18.6.2 computes the coefficient to be one.

This use of the 2025 statement concerns common original data only.
For example, locally smooth convex approximations to
$\max(0,t_1-\log r_1,\ldots,t_s-\log r_s)$ in the original unit moment
coordinates identify its curvature products, on every original test,
with the continuous-potential products used here.  Both use the same
calibrated smooth integration.  This comparison does not extend an
arbitrary original current to the larger 2025 test sheaf.

Fix $\eta$.  Apply the identity also to $\chi\eta$, where $\chi$ is an
arbitrary original compact smooth scalar function.  Uniqueness of
Radon measures gives
\[
 \mu_{B_{\mathbf r},\eta}=j_*\mu_{j^*\eta}.
\]
Indeed original smooth scalar functions determine Radon measures on
the actual Hausdorff analytic space.  The right-hand measure is Radon:
the fiber is closed and its intersection with the compact support of
$\eta$ is compact.  Integration of a fixed locally bounded Borel
function in this signed-measure identity proves
\eqref{eq:r4-borel-gauss-fiber}.  There is no varying-measure limit in
this step.
\end{proof}

\subsection{Exact spreading of every original fiber test}

\begin{lemma}[Exact spreading from an algebraic parameter fiber]
\label{lem:r4-exact-fiber-test-spreading}
Let $T$ be a smooth affine $K$-variety, let $a\in T^{\rm an}$, and let
$F=\mathscr H(a)$.  Let $X$ be an ordinary open in the analytification
of a smooth affine $K$-variety $Y$.  For every original compact smooth
form $\beta$ on $X_F$, there are an ordinary open neighborhood
$A\subset T^{\rm an}$ of $a$ and an original compact smooth form
$\alpha$ of the same bidegree on $X\times_K A$ such that
\[
 j_a^*\alpha=\beta.
\]
This is equality of original forms, not only approximation of their
integrals.  Consequently $j_a^*D\alpha=D\beta$.
The same assertion holds locally for an arbitrary allowed $X$ over
$K$, after a finite partition subordinate to its ordinary algebraic
neighborhoods.
\end{lemma}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} as a test-spreading
lemma.  Its proof uses affinoid density and the original definition of
smooth forms, not a regularization theorem over $F$.

\begin{proof}
We give the support and zero-locus details, since mere density of
$K(T)$ in a residue field would not by itself prove this assertion.
The image in $X$ of $\operatorname{supp}\beta$ is compact.  Choose an
original smooth compact cutoff $\chi_X$ equal to one near this image.
Its support is covered by finitely many compact polynomial norm
domains in $Y^{\rm an}$, all contained in $X$.  Their scalar extensions
are compact; in particular all the finitely many affine coordinates
used below have uniform bounds on their union, denoted $D_F$.
Finite localization permits working on one such stage at a time.

Around the compact support of $\beta$, choose finitely many compact
polynomial or rational norm boxes in $Y_F^{\rm an}$ whose slightly
larger boxes remain in original unit-moment presentations for
$\beta$.  Such boxes form a neighborhood basis: affine-coordinate
bounds first give an affinoid domain, and finitely many strict norm
inequalities give smaller rational domains.  Their algebras are the
completed algebras obtained from $F[Y]$ by the finitely many rational
localizations specifying the boxes.  Thus fractions of elements of
$F[Y]$, with the prescribed denominators inverted, are uniformly
dense in their analytic functions.

On a slightly larger box, approximate each analytic unit $h$ occurring
in the presentation by a rational function $P/Q$ with error strictly
less than the positive minimum of $|h|$.  The non-Archimedean triangle
inequality then gives $|P/Q|=|h|$ throughout that box.  In particular
the logarithms, and all their original smooth differentials, agree
on its ordinary interior.  This replaces the finitely many analytic
unit moments by rational $F$-moments exactly where the local term is
to be supported.

Choose localizing bumps by smooth profiles of the finitely many
polynomial norms defining the boxes.  When a polynomial may vanish,
the profile is constant below a fixed positive norm threshold; this
defines an original smooth scalar function across its zero locus.
Include extra lower cutoffs for every numerator and denominator of
a rational unit moment.  They are zero below a positive threshold,
so that all those polynomials are units on a neighborhood of the
support of the local term.  Upper norm cutoffs give compact supports.
If $b_1,\ldots,b_N$ are such bumps with $\sum b_i\geq1$ near
$\operatorname{supp}\beta$, use
\[
 \chi_i=b_i R\Bigl(\sum b_i\Bigr),
\]
where $R$ is smooth, equals $1/t$ for $t\geq1/2$, and is zero for
$t\leq1/4$.  This is a partition equal to one near the required
support.  Extend the real coefficient forms of the moment
presentations with a real smooth cutoff around the compact moment
images.  We have now written $\beta$ as a finite sum of localized
rational-moment forms, globally extended by zero, with the lower
cutoffs just described.  Only finitely many coefficients in $F$ occur
in the polynomials used for these moments and bumps.

The evaluations at $a$ of rational functions on $T$ whose denominators
do not vanish at $a$ are dense in $F$.  In fact their images contain
the fraction field of the coordinate ring modulo the prime kernel
of $a$, and $F$ is its completion.  Approximate all the preceding
finite coefficients by such evaluations.  Bounds on the affine
coordinates on $D_F$ make the resulting polynomial approximations
uniform on the whole of $D_F$.

There are only finitely many positive cutoff thresholds.  Choose the
coefficient approximations with errors smaller than all of them,
and smaller than the unit lower bounds on the active supports.
For clarity, if $P$ is one of the polynomials and $\widetilde P$ its
approximation, then
\[
 \|P-\widetilde P\|_{D_F}<\delta
 \quad\Longrightarrow\quad
 \begin{cases}
  |P|=|\widetilde P|,& |P|\geq\delta,\\
  |P|,|\widetilde P|<\delta,& |P|<\delta.
 \end{cases}
\]
Every profile used at a possible zero of $P$ is constant on the
second region.  Its value is therefore unchanged everywhere on
$D_F$, including the zero locus.  Every unit moment is protected by
a lower cutoff above this threshold, so its logarithm is unchanged
on an ordinary neighborhood of the active support.  Hence the
localized differential forms, not merely their scalar values, are
unchanged.  In particular the localized expression is still zero
on precisely the same part of the fiber outside its support.

The approximating rational functions on $T$ are analytic on an
ordinary neighborhood $A$ of $a$ after removing their finitely many
denominator zero loci.  Their coefficients define analytic
polynomials on $Y\times_K A$.  Use the same moment expressions and
norm-profile cutoffs there.  Lower cutoffs enforce the required unit
loci, and profiles constant near zero give original smoothness
across the remaining polynomial zero loci.  Multiply by
$\chi_X\circ q$ and by an original compact smooth cutoff on $A$
equal to one near $a$.  The resulting form $\alpha$ has compact
support: it is contained in a finite union of actual products of
compact affinoid domains.  Its fiber restriction is exactly $\beta$
by the preceding uniform norm and profile equalities.  Restriction
of unit moments commutes with $d'$ and $d''$, giving the assertion
about $D$.

For a general allowed $X$, apply B, Lemma D.4, only over the original
field $K$ to the compact image of the support.  A finite original
partition reduces to the preceding construction on ordinary affine
algebraic neighborhoods.  Intersect the finitely many parameter
neighborhoods and add the resulting forms.
\end{proof}

The neighborhood $A$ in this lemma is an ordinary neighborhood.  Even
when $T$ is a torus and $a$ is its Gauss point, $A$ need not be an
entire annulus or a polyradius box: a rational coefficient may have
zeros of its denominator at the same radius, away from the Gauss
point.  Removing those zeros is legitimate and is needed.

\subsection{The kernel over a finite polyradius Gauss field}

\begin{theorem}[Gauss residue-field preservation of the potential kernel]
\label{thm:r4-gauss-extension-kernel}
Let $u,v\in\mathcal C_b(X)$ on an allowed $K$-analytic space and
suppose $D[u]=D[v]$ on every original compact complementary test.
For every finite positive real polyradius $\mathbf r$ and the scalar
projection $\pi:X_{K_{\mathbf r}}\to X$,
\[
 D[u\circ\pi]=D[v\circ\pi]
\]
on every original $K_{\mathbf r}$-test.  The pullbacks are the actual
locally bounded classical functions.
\end{theorem}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the preceding
exact spreading lemma, the external full-fiber identity, and the
same-field product kernel already proved in this round.

\begin{proof}
The assertion in dimension zero is immediate.  Assume $n\geq1$ and
write $f=u-v$.  On $X\times_K A$, the original-field product result
Lemma~\ref{lem:r4-arbitrary-smooth-product-kernel} gives
\[
 D[u\circ q]=D[v\circ q].
\]
The background $B_{\mathbf r}$ is an ambient BT product.  Therefore
the already established ambient mixed-background independence,
Lemma~\ref{lem:potential-independence-ambient-mixed}, gives
\begin{equation}\label{eq:r4-gauss-weighted-kernel}
 D\bigl((f\circ q) B_{\mathbf r}\bigr)=0.
\end{equation}
This step uses only the proved ambient-product special case of the
potential quotient; it does not invoke the open arbitrary-background
assertion.

Let $\beta$ be any original compact $K_{\mathbf r}$-smooth form of
bidegree $(n-1,n-1)$ on $X_{K_{\mathbf r}}$.  Apply
Lemma~\ref{lem:r4-exact-fiber-test-spreading}, with parameter
$T=\mathbf G_m^s$ and $a=\zeta_{\mathbf r}$, to obtain $A$ and
$\alpha$ with $j^*\alpha=\beta$.  Test
\eqref{eq:r4-gauss-weighted-kernel} on $\alpha$ and use the fixed
Borel-weighted fiber identity on $D\alpha$.  It gives
\[
 0=\langle(f\circ q)B_{\mathbf r},D\alpha\rangle
   =\int_{X_{K_{\mathbf r}}}(f\circ\pi)D\beta
   =\langle D[f\circ\pi],\beta\rangle.
\]
All the BT operations in this argument occurred on the original
$K$-space $X\times_K A$.  The assertion for $s$ parameters was proved
in one step; no induction applies R or B over an intermediate
infinite extension.
\end{proof}

\subsection{Finite extensions of these Gauss fields}

\begin{lemma}[An algebraic etale realization of a finite fiber field]
\label{lem:r4-etale-realization-finite-gauss}
Let $E=K_{\mathbf r}$ and let $L/E$ be finite.  There are a Zariski
open $T_0\subset\mathbf G_m^s$ containing $\zeta_{\mathbf r}$ in its
analytification and a finite etale map $T'\to T_0$ of degree
$e=[L:E]$ such that the analytic fiber over $\zeta_{\mathbf r}$ is
a single point $a'$ with completed residue field isometric to $L$.
The spaces and the map are defined over the original $K$.
\end{lemma}

\noindent\textbf{Status: \texttt{EXTERNAL\_KNOWN}} elementary
Hensel--Krasner realization; the following proof fixes its precise
field and calibration scope.

\begin{proof}
Characteristic zero makes $L/E$ separable.  Choose $\theta$ with
$L=E(\theta)$, and let $P$ be its monic minimal polynomial, of
degree $e$.  Approximate its coefficients by elements of the dense
field $K(z_1,\ldots,z_s)$.  For a sufficiently close monic polynomial
$Q$ of the same degree, Hensel's argument in the complete field $L$
gives a root $\theta'\in L$ arbitrarily close to $\theta$.
Explicitly, on a sufficiently small closed ball about $\theta$ the
map
\[
 t\longmapsto t-Q(t)/Q'(\theta)
\]
is a strict contraction into that ball: the coefficient errors make
$Q(\theta)$ sufficiently small, while $Q'(\theta)$ stays close to
the nonzero number $P'(\theta)$ and the higher Taylor terms are
dominated by the linear term.  Completeness gives its fixed point.

Choose the closeness smaller than the minimum distance between
$\theta$ and its distinct $E$-conjugates.  Then $E(\theta')=L$.
For completeness, if an automorphism over $E(\theta')$ of an
algebraic closure moved $\theta$, invariance of the unique extended
absolute value would put both $\theta$ and that conjugate at the
same, strictly smaller distance from $\theta'$ than their mutual
distance.  The ultrametric inequality contradicts this.  Thus
$\theta\in E(\theta')$, and the reverse inclusion follows from
$\theta'\in L$.  Consequently $Q$ is irreducible over $E$, has
nonzero discriminant, and is also irreducible over $K(z_1,\ldots,z_s)$.

Clear the finitely many coefficient denominators and invert the
discriminant on an affine Zariski open $T_0$ of the parameter torus.
Every nonzero rational function has nonzero value at its Gauss point,
so $\zeta_{\mathbf r}\in T_0^{\rm an}$.  The monic algebra
\[
 \mathcal O(T_0)[t]/(Q(t))
\]
defines the required finite etale $K$-variety $T'$.  Its completed
fiber algebra at the Gauss point is $E[t]/(Q)\simeq L$, a field.
The extension of the complete absolute value is unique, so this
fiber is precisely one analytic point with residue field $L$.
\end{proof}

\begin{corollary}[Finite extensions after a Gauss extension]
\label{cor:r4-finite-over-gauss-kernel}
Under the hypotheses of Theorem~\ref{thm:r4-gauss-extension-kernel},
let $L$ be any finite extension of $K_{\mathbf r}$, with the inherited
absolute value.  Then
\[
 D[u\circ\pi_{L/K}]=D[v\circ\pi_{L/K}]
\]
on all original $L$-tests of $X_L$.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}}.  The following
argument returns to an original-$K$ parameter space; it does not
apply the finite-$K$ theorem over $K_{\mathbf r}$.

\begin{proof}
Use the realization $T'\to T_0$ just proved, and an ordinary
neighborhood $A'$ of its point $a'$.  On the original-$K$ space
$X\times_K A'$ pull back the parameter function
$b=\log\max(1,|z_i|/r_i)$.  Denote its $s$-fold curvature product
by $B'$.  The external full-fiber formula gives the integration
current on the fiber of the parameter map to $T_0$ at
$\zeta_{\mathbf r}$.  That fiber is $X_L$ viewed as an
$E=K_{\mathbf r}$-analytic space.  Finite-constant calibration gives
\begin{equation}\label{eq:r4-finite-gauss-fiber-calibration}
 B'=e\,j'_*[1]_{X_L,L},\qquad e=[L:E].
\end{equation}
Indeed on an $L$-analytic space viewed over $E$ its integration
current is $e$ times the $L$-integration current; see
\cite[Remark 14.4.3, p.~410]{CLD25}.  This is not the different
pushforward normalization for scalar base change.  Equality
\eqref{eq:r4-finite-gauss-fiber-calibration} also holds after a fixed
locally bounded Borel multiplication, by the coefficient-Radon
argument used in Proposition~\ref{prop:r4-full-gauss-fiber}.

The same-field product kernel on $X\times_K A'$ and ambient BT
independence give $D((f\circ q)B')=0$, with $f=u-v$.
Given any original $L$-test $\beta$ on $X_L$, use exact spreading
with parameter variety $T'$ and point $a'$ to obtain a compact
original $K$-test $\alpha$ restricting to $\beta$.  Pairing the
last equality with $\alpha$ gives
\[
 0=e\int_{X_L}(f\circ\pi_{L/K})D\beta.
\]
Division by the positive integer $e$ proves the result.  All
regularization, product, and potential-independence inputs were
again used only over the original finite field $K$.
\end{proof}

\subsection{A failed shortcut and the remaining gate}

\begin{example}[A single Gauss section misses correlated fiber points]
\label{ex:r4-correlated-gauss-fiber}
Take $X=\mathbf A^{1,\rm an}_x$ and a parameter $z$ of radius one.
Put $w=x-z$.  Choose $\chi\in C_c^\infty(\mathbf R)$ with
$\chi(0)=1$, and choose a nonzero nonnegative
$\psi\in C_c^\infty(( -\infty,0))$.  On the unit locus of $zw$ put
\[
 \alpha=\chi(\log|z|)\psi(\log|w|)
                d'\log|w|\wedge d''\log|w|,
\]
and extend it by zero.  This is an original smooth compact form:
its support has both $z$ and $w$ bounded above and away from zero,
and $x=z+w$ is bounded there.  The algebraic coordinate change
$(x,z)\mapsto(w,z)$ and unit torus calibration give
\[
 \bigl\langle D\max(0,\log|z|),\alpha\bigr\rangle
                  =\int_{\mathbf R}\psi(t)\,dt>0.
\]
In contrast, the independent radius-one Gauss lift over any
$a\in X$ satisfies
\[
 |x-z|=\max(|x(a)|,1)\geq1.
\]
Thus $\alpha$ vanishes on that entire independent Gauss section.
The full fiber over the parameter Gauss point contains points with
$|x-z|<1$ and cannot be replaced by this section.  The shortcut is
\texttt{FALSE}; the full-fiber identity used above retains precisely
these additional points.
\end{example}

The statements proved here concern finite extensions of finite
polyradius Gauss residue fields, including a single such field.
They do not assert kernel preservation for an arbitrary complete
extension of $K$, or for every analytic morphism defined over such
an extension.  In particular an arbitrary extension-analytic curve
is not silently viewed as an allowed finite-type $K$-analytic space.
Passing from this explicitly treated family of fields to all the
complete extensions in the definition of classical psh remains an
\texttt{OPEN} gate.  Although the exact spreading lemma also applies
at non-Gauss algebraic parameter points, no representation of their
general residue-field fibers by the bounded ambient BT backgrounds
used above has been proved.  Density of Gauss points is not a
substitute for that missing fixed-fiber statement.  The
arbitrary-background potential quotient therefore remains
\texttt{OPEN} at this stage.

% END source: sections/gauss_extension_kernel.tex


% BEGIN frozen or reviewed source: sections/finite_parameter_kernel_reduction.tex
\section{Reduction of scalar kernel functoriality to finite parameter fields}
\label{sec:r4-finite-parameter-kernel-reduction}

This section isolates the remaining scalar-extension question.  It
does not identify a general parameter fiber with a bounded ambient
BT product.  The base field $K$ and its absolute value remain fixed.

\begin{lemma}[Exact finite-coefficient descent of original tests]
\label{lem:r4-finite-coefficient-test-descent}
Let $L/K$ be any complete extension, with its prescribed extending
absolute value, and let $X$ be an allowed $K$-analytic space.  Given
finitely many original compact smooth forms $\beta_1,\ldots,\beta_N$
on $X_L$, there exist a finite tuple $c=(c_1,\ldots,c_m)$ of elements
of $L$, a closed complete subfield
\[
 E=\overline{K(c_1,\ldots,c_m)}\subset L,
\]
and original compact smooth forms $\eta_i$ on $X_E$, of the same
bidegrees as the $\beta_i$, such that
\[
 \beta_i=p^*\eta_i,\qquad p:X_L\longrightarrow X_E.
\]
The equality is exact.  The field $E$ is isometric to
$\mathscr H(a)$ for the point $a\in\mathbf A_K^{m,\rm an}$ defined
by evaluating its coordinates at $c_1,\ldots,c_m$.  In particular
$p^*D\eta_i=D\beta_i$.
\end{lemma}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} finite-coefficient
descent.  The analytic mechanism is the rational localization already
proved in Lemma~\ref{lem:r4-exact-fiber-test-spreading}; the present
argument extracts its finite coefficient field inside a given $L$.

\begin{proof}
The union of the images in $X$ of the test supports is compact.
Over the original field $K$, choose a finite original smooth
partition on this compact set, subordinate to ordinary affine
algebraic neighborhoods supplied by B, Lemma D.4.  Pulling it back
to $X_L$ reduces to one original test in an ordinary open
$X\subset Y^{\rm an}$, with $Y$ smooth affine over $K$.
Choose an original $K$-smooth compact cutoff $\chi_X$ equal to one
near the image of its support.  Cover $\operatorname{supp}\chi_X$
by finitely many compact polynomial or rational norm domains
$D_\ell\subset X$.  Their extensions $(D_\ell)_L$ are compact
affinoid domains and provide bounds on every affine coordinate.

Apply the rational-localization part of
Lemma~\ref{lem:r4-exact-fiber-test-spreading}, now with the complete
field denoted $F$ there equal to $L$.  Here this means its explicit
affinoid-density argument, not an invocation of R or B over $L$.
On finitely many slightly enlarged norm boxes, approximate each
analytic unit in the test presentations by a rational expression
with coefficients in $L$, with error smaller than its positive
minimum norm.  This preserves its logarithm exactly on the box.
Use smooth norm-profile bumps to localize there, with the following
two protections: profiles involving possible polynomial zeros are
constant below a positive threshold, and every actual unit numerator
or denominator has an additional lower cutoff.  The latter makes
each localized term identically zero near its non-unit locus.
Extend the real coefficient forms with real smooth cutoffs and add
the finitely many localized terms.  Multiplied by $\chi_X$, the
result is exactly the original test everywhere on $X_L$, including
outside its support.  These are precisely the support and zero-locus
steps justified in the cited lemma.

Only finitely many elements of $L$ are now used as coefficients of
the rational moments and of the polynomials specifying their
cutoffs.  Include all of them, for all charts and all $\beta_i$, in
the tuple $c$.  A $K$-analytic function used by $\chi_X$ needs no
new coefficient: it has its scalar pullback over every intermediate
field.  The finite real cutoff thresholds and the real smooth
coefficient functions are also unchanged by scalar extension.

Define $\eta_i$ over $E$ by the very same finite localized formulas.
There is no new coefficient approximation in this step.  The lower
cutoffs ensure that every rational unit moment is defined wherever
its localized term can be nonzero, over $E$ as well as over $L$;
profiles constant near zero ensure original smoothness across all
other zero loci.  Thus these formulas give genuine original forms,
extended by zero on the stated ordinary opens.  Their supports are
contained in the inverse image of $\operatorname{supp}\chi_X$,
which is a closed subset of the finite union of compact affinoid
domains $(D_\ell)_E$.  They consequently have compact support on
$X_E$.  This proves compactness using scalar extensions of
affinoid domains, without replacing analytic products by Cartesian
products of point sets.

Scalar pullback of each formula recovers exactly the rationalized
$L$-test.  This proves $p^*\eta_i=\beta_i$ on each piece; extend by
zero and add the pieces to obtain the global assertion.  All
finitely many coefficient lists can be united before taking $E$.

Finally the seminorm
$P\mapsto|P(c_1,\ldots,c_m)|$ on $K[T_1,\ldots,T_m]$ defines the
point $a$.  Its prime-kernel quotient has fraction field
$K(c_1,\ldots,c_m)$, and completing inside the complete field $L$
gives exactly $E$.  This identifies $E$ isometrically with
$\mathscr H(a)$.  Smooth pullback commutes with $d'$ and $d''$ by
the original unit-moment definition, giving the last assertion.
\end{proof}

\begin{proposition}[The fixed Borel integration identity under scalar extension]
\label{prop:r4-scalar-borel-integration}
Let $E\subset L$ be a complete extension, let $Z$ be a good Hausdorff
pure $n$-dimensional $E$-analytic space, and let $p:Z_L\to Z$ be scalar
base change.  If $\eta$ is an original compact smooth $(n,n)$-form
and $f$ is a locally bounded Borel function on $Z$, then
\[
 \int_{Z_L}(f\circ p)p^*\eta=\int_Z f\eta.
\]
There is no degree factor in this formula.
\end{proposition}

\noindent\textbf{Status: \texttt{EXTERNAL\_KNOWN}} for the measure
identity, and \texttt{DIRECT\_COROLLARY} for its fixed Borel weight.

\begin{proof}
The canonical scalar-base-change calibration gives
\[
 p_*\mu_{p^*\eta}=\mu_\eta;
\]
this is \cite[Proposition 15.3.1, p.~425]{CLD25}, restricted to the
common original test category.  Apply the definition of pushforward
to the fixed Borel function $f$.  Compact supports and local
boundedness make the integrals finite.  Equivalently one can first
localize by a compact cutoff equal to one on the test support.
This is an identity of fixed signed Radon measures; it makes no
continuity assertion for changing measures or changing Borel
weights.  Its factor one is distinct from the finite-constant
calibration on the same analytic space viewed over a smaller field.
\end{proof}

\begin{theorem}[Conditional reduction to algebraic parameter residue fields]
\label{thm:r4-finite-parameter-kernel-reduction}
Fix $u,v\in\mathcal C_b(X)$ on an allowed $K$-analytic space, with
$D[u]=D[v]$ on all original $K$-tests.  Assume the following precise
additional property:
\begin{quote}
For every finite $m$ and every $a\in\mathbf A_K^{m,\rm an}$, the
actual scalar pullbacks to $X_{\mathscr H(a)}$ satisfy
$D[u_{\mathscr H(a)}]=D[v_{\mathscr H(a)}]$ on all original
$\mathscr H(a)$-tests.
\end{quote}
Then for every complete extension $L/K$ the scalar pullbacks satisfy
\[
 D[u_L]=D[v_L]
\]
on all original $L$-tests.  Thus scalar kernel transfer for all
algebraic parameter residue fields is equivalent to scalar kernel
transfer for all complete extensions, with these fixed $X,u,v$.
\end{theorem}

\noindent\textbf{Status: \texttt{CONDITIONAL}} for the scalar kernel
conclusion; the reduction itself follows from the two proved
statements above.  The additional property has not been proved for
general non-Gauss parameter points.

\begin{proof}
Dimension zero is immediate.  Otherwise, take any original compact
$L$-test $\beta$ of bidegree $(n-1,n-1)$.  By
Lemma~\ref{lem:r4-finite-coefficient-test-descent}, it equals
$p^*\eta$ for an original compact test on $X_E$, where
$E\simeq\mathscr H(a)$ for a finite-dimensional algebraic parameter
point.  Put $f=u-v$ and denote its actual scalar pullbacks by $f_E$
and $f_L$.  They are locally bounded Borel functions and
$f_L=f_E\circ p$.  Using the fixed Borel integration identity gives
\[
 \begin{aligned}
 \langle D[f_L],\beta\rangle
  &=\int_{X_L}f_L D\beta
    =\int_{X_L}(f_E\circ p)p^*(D\eta)\\
  &=\int_{X_E}f_E D\eta
    =\langle D[f_E],\eta\rangle=0.
 \end{aligned}
\]
The last equality is exactly the additional hypothesis.  Since
$\beta$ was arbitrary, the asserted original-current equality
holds.  Conversely the all-complete-extension assertion applies
in particular to every $\mathscr H(a)$, so gives that hypothesis.
No BT operation or regularization theorem over $E$ or $L$ enters
this reduction.
\end{proof}

The unresolved geometric gate is therefore already visible at
residue fields of finite-dimensional algebraic parameter points.
The Gauss and finite-over-Gauss cases proved earlier supply only
part of this family.  Exact descent of tests does not represent a
general parameter fiber by a bounded ambient BT current, and
density of Gauss points does not supply such a representation.
This reduction asserts scalar base-change equalities only.  It
does not assert functoriality under every analytic morphism over an
arbitrary extension, or turn the still open arbitrary-background
potential quotient into a proved statement.

% END source: sections/finite_parameter_kernel_reduction.tex


% BEGIN frozen or reviewed source: sections/relative_capacity_next.tex
\section{Two simultaneous-pasting mechanisms for relative capacity}
\label{sec:r4-relative-capacity-next}

The unrestricted ordinary-open Borel outer regularity problem remains
open.  We prove two precise statements that use the BT structure beyond
weak compactness of a measure family.  The first extends separation by
analytic zero sets to complete pole sets of extended classical psh
functions.  The second pastes uniformly close normalized testers on
finitely many separated compact pieces, using a bounded global
classical compensator.  The cost of the latter construction depends
on the separating cutoffs; this dependence cannot be discarded.

The field is the fixed finite extension $K/\mathbf Q_p$ with its
original absolute value.  All backgrounds are symmetric,
original CLD-positive, and separately $d'$- and $d''$-closed.
Every current identity uses all original compact smooth complementary
tests.  Write $q=n-p\geq1$ and
\[
 \mu_{S,h}=T_{S,q}(h,\ldots,h),\qquad
 C_S(E;\Omega)=\sup_{\substack{h\in\mathcal C_b(\Omega)\\0\leq h\leq1}}
                         \mu_{S,h}(E).
\]
The raw capacity is defined on Borel sets; $C_S^*$ always denotes
its outer regularization by ordinary open neighborhoods in the same
specified domain.  Bounds on potentials mean actual pointwise bounds.

\subsection{Separation by complete pole sets}

\begin{theorem}[Finite pole-carrier additivity]
\label{thm:r4-pole-carrier-additivity}
Let $\Omega$ be an allowed ordinary open set, and let
$S_1,\ldots,S_m$ have the same bidegree $(p,p)$.
Suppose $\phi_i\leq0$ are extended classical psh functions on
$\Omega$, with $P_i=\{\phi_i=-\infty\}$, such that
\[
 \mathbf1_{P_i}S_i=S_i,
 \qquad C_{S_i}(P_l;\Omega)=0\quad(i\ne l).
\]
The first equality concerns every coefficient measure.  The pole
sets need not be closed, analytic, or principal.  Put $S=\sum_iS_i$.
For every Borel $E\subset\Omega$ and every arbitrary subset
$A\subset\Omega$, respectively, one has
\begin{align}
 C_S(E;\Omega)&=\sum_i C_{S_i}(E;\Omega),
                  \label{eq:r4-pole-raw-sum}\\
 C_S^*(A;\Omega)&=\sum_i C_{S_i}^*(A;\Omega).
                  \label{eq:r4-pole-outer-sum}
\end{align}
Infinite values are allowed.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the extension
to general complete pole carriers.  The outer identity is a
\texttt{DIRECT\_COROLLARY} of the raw identity.  Analytic unit-log
invariance is not used in this extension.

\begin{proof}
Fix normalized actual $h_i\in\mathcal C_b(\Omega)$ and
$0<\delta<1/2$.  For $\epsilon>0$ put
\[
 \psi_i=\sum_{l\ne i}\phi_l,\qquad
 b_i=\max\{0,\delta+(1-2\delta)h_i+\epsilon\psi_i\},
 \qquad b=\max_i b_i.
\]
The empty sum is zero.  Every $b_i$ and $b$ is an actual member
of $\mathcal C_b(\Omega)$, between zero and $1-\delta$.  Indeed,
replace each $\phi_l$ by $\phi_{l,N}=\max(\phi_l,-N)$.
The resulting bounded classical maxima decrease to the displayed
functions and have the common lower bound zero.  This uses only
bounded decreasing closure, not an unbounded BT operator.

For $R>0$ let
\[
 A_{i,R}=P_i\cap\{\psi_i>-R\},
\]
and choose $\epsilon R<\delta/2$.  On $A_{i,R}$ the $i$th branch
is greater than $\delta/2$.  All other $b_l$ vanish there, since
their penalties contain $\phi_i=-\infty$.  Strict Borel maximum
locality for the bounded functions $b_1,\ldots,b_m$ therefore gives
\begin{equation}
 \mathbf1_{A_{i,R}}\mu_{S_i,b}
       =\mathbf1_{A_{i,R}}\mu_{S_i,b_i}.
 \label{eq:r4-pole-active-branch}
\end{equation}
Neither $A_{i,R}$ nor $\{\psi_i>-R\}$ is asserted ordinarily open.

We next prove the required measure lower bound for this branch.
Set
\[
 c_{i,N}=\delta+(1-2\delta)h_i
                +\epsilon\sum_{l\ne i}\phi_{l,N},
 \qquad b_{i,N}=\max(0,c_{i,N}).
\]
All these functions are bounded classical.  Since truncation raises
the penalties, $c_{i,N}>\delta/2$ on $A_{i,R}$ for every $N$.
Borel maximum locality and positivity of the mixed expansion give
\[
 \mathbf1_{A_{i,R}}\mu_{S_i,b_{i,N}}
  =\mathbf1_{A_{i,R}}\mu_{S_i,c_{i,N}}
  \geq(1-2\delta)^q\mathbf1_{A_{i,R}}\mu_{S_i,h_i}.
\]
This is the point where positivity replaces the zero-curvature
identity used for analytic logarithms in R3.

The sequence $b_{i,N}\downarrow b_i$ has a bounded limit.
For a compact $L\subset A_{i,R}$, decreasing BT continuity and
the local compact-set weak inequality imply
\[
 \mu_{S_i,b_i}(L)
   \geq\limsup_N\mu_{S_i,b_{i,N}}(L)
   \geq(1-2\delta)^q\mu_{S_i,h_i}(L).
\]
One may insert a fixed original smooth cutoff equal to one near
$L$ to apply the finite-measure weak inequality.  Radon inner
regularity now extends the inequality from compact subsets to
every Borel subset of $A_{i,R}$.  Thus no unrestricted Borel
restriction is passed through a varying-measure limit.

The pole restriction identity
\eqref{eq:r2-polar-restriction-BT} shows that every
$\mu_{S_i,h_i}$ is carried by $P_i$, even when $P_i$ is not closed.
The cross-capacity hypothesis makes it vanish on $P_l$, $l\ne i$.
Consequently the increasing sets $A_{i,R}$ capture its entire
mass as $R\to\infty$.  By positivity and background linearity,
\[
 C_S(E;\Omega)
   \geq(1-2\delta)^q\sum_i\mu_{S_i,h_i}(E\cap A_{i,R}).
\]
For each $R$ use a new admissible $\epsilon$; no convergence of
these changing testers is invoked.  Let $R\to\infty$, then
$\delta\downarrow0$, and take the independent finite suprema in
the $h_i$.  This gives the lower bound in
\eqref{eq:r4-pole-raw-sum}.  The upper bound follows immediately
from $\mu_{S,h}=\sum_i\mu_{S_i,h}$.

For the outer formula, one common open neighborhood first gives
$C_S^*(A)\geq\sum_iC_{S_i}^*(A)$.  If that sum is finite, choose
separate open neighborhoods nearly minimizing each summand and
intersect the finitely many neighborhoods.  The raw formula and
monotonicity give the reverse inequality.  If a summand is infinite,
the earlier lower bound already proves equality.  All domains
remain $\Omega$ throughout.
\end{proof}

\begin{corollary}[A coefficient condition and countable sums]
\label{cor:r4-pole-carrier-closure}
The cross-capacity condition of the theorem holds if instead
\[
 \mathbf1_{P_l}S_i=0\qquad(i\ne l).
\]
Suppose a countable family satisfies these hypotheses pairwise,
and its positive partial sums converge on all original compact
tests to a locally coefficient-Radon positive closed background $S$.
Then the two capacity sum formulas hold on every fixed compact
core, with a countable sum.  If the component capacities are
Borel outer regular, respectively finite Choquet capacities,
on the required cores, these properties pass to $S$.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
For the first assertion, pole restriction commutation gives
$\mathbf1_{P_l}\mu_{S_i,h}=\mu_{\mathbf1_{P_l}S_i,h}=0$ for
every normalized $h$.  For the countable assertion apply the finite
theorem to each partial sum and then use
Proposition~\ref{prop:r3-capacity-order-approximation} on a compact
core $K_0\subset W\Subset\Omega$.  Its fixed-geometry CLN estimate
makes the capacity of the positive current tail on $W$ tend to
zero uniformly over subsets of $K_0$, both for raw and outer
capacities.  This proves the countable identities and transfers
the stated regularity properties.  No infinite maximum or infinite
intersection of ordinary open sets is used.
\end{proof}

\subsection{A global compensator for finitely many original cutoffs}

In the rest of this section let $Y$ be a smooth affine $K$-variety
and let $\Omega\Subset Y^{\an}$ be ordinary open.  No globally
strict psh exhaustion is assumed.  This affine scope is used for
global regular numerators and denominators in the next construction.

\begin{lemma}[A bounded classical compensator]
\label{lem:r4-cutoff-compensator}
For any finite family of original real compact smooth functions
$\chi_1,\ldots,\chi_m$ on $\Omega$, there is a nonnegative
bounded original smooth classical psh function $\rho$ on $\Omega$
such that $\rho+2\chi_i$ and $\rho-2\chi_i$ are classical psh
for every $i$.  The function $\rho$ can be chosen as a finite
positive sum of functions
\[
 F(\log|a|+A),\qquad a\in K[Y],
\]
with finite real $A$, where $F$ is one fixed smooth convex
nondecreasing function that vanishes on $(-\infty,-1]$ and has
$F''=2$ on $[0,\infty)$.  Each displayed function is extended
by zero at the zeros of $a$.  The compensator depends on the cutoffs and compact
presentations, not on any positive current.
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of exact
rational moment replacement in B and the classical curve chain
rule, with the following explicit global construction.

\begin{proof}
Cover the union of the cutoff supports by finitely many smaller
ordinary neighborhoods with compact closure, on each of which
all needed functions have smooth unit-moment presentations.
Use B's exact rational-unit replacement, as recorded in
Lemma~\ref{lem:r3-countable-trace-controls}, to write these moments
as $\log|a/b|$, with $a,b\in K[Y]$ nonvanishing on the respective
compact stages.  Replacement preserves their norms exactly on
those neighborhoods, so it preserves the original smooth functions
and their logarithmic differentials there.

Choose a nonnegative smooth bump of integral one supported in
$(-1,0)$, and integrate it twice from $-\infty$ to obtain a smooth
convex nondecreasing $\zeta$ with $\zeta=0$ on $(-\infty,-1]$
and $\zeta'=1$ on $[0,\infty)$.  Put $F=\zeta^2$.  Then $F$
has all the stated properties.  For every selected numerator or
denominator choose $A$ such that $|a|>e^{-A}$ on its compact stage,
and set
\[
 F_a=F(\log|a|+A),
\]
extended by zero near the zeros of $a$.  This is globally original
smooth and classical psh on $Y^{\an}$: off the zero set it is a
smooth convex nondecreasing function of a unit logarithm, and it
is identically zero on an ordinary neighborhood of the zero set.
The classical assertion holds after every complete extension by
the curve chain rule.  On the chosen unit stage,
\[
 DF_a=2\,d'\log|a|\wedge d''\log|a|.
\]

On each presentation use the logarithms of its numerators and
denominators as independent real coordinates.  The old rational
moments are their linear differences.  A sufficiently large
positive multiple of the sum of their squares has Hessian
dominating the positive and negative of twice each cutoff Hessian
on a neighborhood of the compact moment image.  Equivalently one
can use $(dt_a-dt_b)^2\leq2(dt_a^2+dt_b^2)$.  Choose a finite
positive sum of the $F_a$ providing this domination for each of
the finitely many presentations, and call the sum $\rho$.

On any one smaller neighborhood, its selected summands together
with $\pm2\chi_i$ form a locally convex smooth function of harmonic
unit logarithms.  Their pullbacks to every complete-extension
curve are subharmonic by the chain rule.  The other summands of
$\rho$ are already classical.  This proves classicality of
$\rho\pm2\chi_i$ on the neighborhood.  Off the cutoff support
the assertion is just classicality of $\rho$, so the local statements
glue.  This argument does not invoke an unproved general implication
from original-current positivity to classicality.

Every $F_a$ is nonnegative and is bounded above on the compact
$\overline\Omega$, since $a$ is globally regular.  Thus $\rho$
is nonnegative, original smooth and bounded on $\Omega$, as required.
\end{proof}

\begin{theorem}[Pasting uniformly close normalized testers]
\label{thm:r4-close-tester-pasting}
Fix pairwise disjoint compact sets $K_1,\ldots,K_m\subset\Omega$.
There is a finite $A_K$, depending on chosen separating original
smooth cutoffs and the fixed affine domain, with the following
property.  Let $0<\delta\leq1$ and let
$u_i\in\mathcal C_b(\Omega)$ satisfy $0\leq u_i\leq1$ and
\[
 |u_i(x)-u_j(x)|\leq\delta
                  \quad(x\in\Omega,\ 1\leq i,j\leq m).
\]
For every allowed fixed background $S$ there is one normalized
$v\in\mathcal C_b(\Omega)$ such that for every choice of Borel
$E_i\subset K_i$,
\begin{equation}
 \mu_{S,v}\!\left(\bigcup_iE_i\right)
       \geq(1+A_K\delta)^{-q}\sum_i\mu_{S,u_i}(E_i).
 \label{eq:r4-close-tester-pasting}
\end{equation}
The same $v$ works for all the subsets $E_i$.  Its construction and
$A_K$ are independent of $S$.  As $\delta\to0$, if all $u_i$
converge uniformly to one function, the constructed $v$ has the
same uniform limit.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}

\begin{proof}
Choose $0\leq\chi_i\leq1$ original smooth with pairwise disjoint
compact supports, each equal to one on an ordinary neighborhood
$N_i$ of $K_i$.  The disjoint compact sets admit such choices by
ordinary smooth partitions of unity.  Let $\rho$ be the preceding
compensator, put $M=\sup_\Omega\rho$, and set
\[
 \psi_i=\rho+2\chi_i,\quad A_K=M+2,\quad
 w=\max_i(u_i+\delta\psi_i),\quad
 v=\frac{w}{1+A_K\delta}.
\]
Every $\psi_i$ is classical and lies in $[0,A_K]$, so $w$ is
classical and $0\leq w\leq1+A_K\delta$.  Thus $v$ is normalized.
On $N_i$, one has $\chi_i=1$ and $\chi_j=0$ for $j\ne i$.
Consequently
\[
 (u_i+\delta\psi_i)-(u_j+\delta\psi_j)
                  \geq-\delta+2\delta=\delta>0.
\]
The $i$th branch is therefore uniquely active on this prescribed
ordinary open neighborhood, despite possible discontinuity of the
$u_i$.  Ordinary maximum locality and positivity of the mixed
expansion give there
\[
 \mu_{S,v}
   =(1+A_K\delta)^{-q}(D(u_i+\delta\psi_i))^qS
   \geq(1+A_K\delta)^{-q}\mu_{S,u_i}.
\]
The disjointness of the $E_i$ proves
\eqref{eq:r4-close-tester-pasting}.  All current assertions are
identities or inequalities on the full original test sheaf before
their Radon restrictions are taken.  Finally the correction and
normalizing factor differ uniformly from zero and one by
$O(A_K\delta)$, which proves the uniform-limit assertion.
\end{proof}

\begin{proposition}[Pasting outside a small capacity error]
\label{prop:r4-capacity-close-pasting}
Keep the domain, disjoint compact sets, cutoffs and constant $A_K$
of Theorem~\ref{thm:r4-close-tester-pasting}.  Let
$u_1,\ldots,u_m$ be any normalized classical testers; no uniform
closeness is assumed.  For $0<\delta\leq1$ put
\[
 B_i=\left\{\max_l(u_l-u_i)\geq2\delta\right\},\qquad
 v=\frac{\max_i(u_i+\delta(\rho+2\chi_i))}{1+A_K\delta}.
\]
Then $v$ is normalized classical and, for every Borel $E_i\subset K_i$,
\begin{equation}
 \mu_{S,v}\!\left(\bigcup_iE_i\right)
 \geq(1+A_K\delta)^{-q}
       \sum_i\left(\mu_{S,u_i}(E_i)
                       -C_S^*(K_i\cap B_i;\Omega)\right).
 \label{eq:r4-capacity-close-pasting}
\end{equation}
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
same explicit maximum construction and strict Borel locality.

\begin{proof}
The bounds and classicality of $v$ did not use closeness of the
$u_i$.  On $K_i\setminus B_i$ its $i$th branch is strictly
larger than every other branch.  Strict Borel maximum locality
and mixed positivity give the previous measure lower bound on
that Borel set.  Since $u_i$ is normalized on the entire domain,
\[
 \mu_{S,u_i}(E_i\cap B_i)
       \leq C_S(K_i\cap B_i;\Omega)
       \leq C_S^*(K_i\cap B_i;\Omega).
\]
Subtract this possible loss and sum over the disjoint $E_i$.
All capacities and test functions have the same domain $\Omega$.
\end{proof}

\begin{corollary}[Simultaneous extraction near a capacity limit]
\label{cor:r4-capacity-cluster-pasting}
Let $K_i$ be fixed disjoint compacts and $E_i\subset K_i$ Borel.
For each $i$, let $u_{i,j}$ be normalized classical testers.
Suppose all these rows converge to one actual Borel function $a$
in local outer $S$-capacity with domain $\Omega$ on
$\bigcup_iK_i$.  Then
\[
 C_S\!\left(\bigcup_iE_i;\Omega\right)
       \geq\sum_i\limsup_j\mu_{S,u_{i,j}}(E_i).
\]
\end{corollary}

\begin{proof}
Choose a subsequence separately for each of the finitely many rows
so that its mass on its own $E_i$ tends to its limsup.  Relabel
these selected rows with the same index.  Each still converges
in capacity to $a$ on every $K_l$.  For fixed $\delta>0$,
\[
 K_i\cap B_{i,j}\subset
   \bigcup_l\bigl(K_i\cap\{|u_{l,j}-a|\geq\delta\}\bigr)
      \ \cup\ \bigl(K_i\cap\{|u_{i,j}-a|\geq\delta\}\bigr).
\]
Finite subadditivity makes all error capacities tend to zero.
Apply Proposition~\ref{prop:r4-capacity-close-pasting}, first
let $j\to\infty$ with $\delta$ fixed, and then let
$\delta\downarrow0$.  CLN bounds all displayed masses on these
fixed compact sets.  No convergence on the arbitrary Borel $E_i$
is assumed or deduced from weak convergence of measures.
\end{proof}

\begin{corollary}[Simultaneous mass extraction near one uniform limit]
\label{cor:r4-uniform-cluster-pasting}
Let $K_i$ be fixed disjoint compacts and $E_i\subset K_i$ Borel.
For each $i$, let $u_{i,j}$ be normalized classical testers converging
uniformly on $\Omega$ to the same function.  Then
\[
 C_S\!\left(\bigcup_iE_i;\Omega\right)
       \geq\sum_i\limsup_j\mu_{S,u_{i,j}}(E_i).
\]
\end{corollary}

\begin{proof}
For each of the finitely many indices choose a subsequence attaining
its limsup; all resulting rows still have the same uniform limit.
Their maximal pairwise distance tends to zero.  Apply the theorem
to each finite row and let its index tend to infinity.  CLN makes
all masses on these fixed compacts finite and uniformly bounded.
\end{proof}

\subsection{The full BT tester class repairs the vertical-fiber subfamily}

\begin{example}[An actual toric subfamily and its missing testers]
\label{ex:r4-vertical-fiber-repair}
On the original two-dimensional split torus take
\[
 \Omega=\{-2<x=\log|T_1|<2,\ -2<y=\log|T_2|<2\},\qquad
 S=[d'x\wedge d''x+d'y\wedge d''y].
\]
For $0\leq t\leq1$ define the continuous normalized classical
tester
\[
 u_t=\frac{|x-t|}{2+t}.
\]
Let $\sigma$ denote the coordinate Gauss section.  The ordinary
tropical calibration, valid on all original tests, gives
\[
 D(u_tS)=\frac{2}{2+t}\,
        \sigma_*(\delta_t\otimes dy).
\]
Consider the Borel set
\[
 E=\sigma\!\left(
      \{0\}\times[0,\tfrac12]
      \ \cup\ \bigcup_{j\geq1}\{1/j\}\times(\tfrac12,1]
                      \right).
\]
For the measure subfamily generated only by the $u_t$, the raw
capacity of $E$ is $1/2$ and its ordinary-open outer capacity is
$1$.  For the full relative BT capacity one instead has
\[
 C_S(E;\Omega)\geq1.
\]
Thus this embedded vertical-fiber subfamily is not a counterexample
to Borel outer regularity of the full BT capacity.
\end{example}

\begin{proof}
The current formula follows by smoothing the convex absolute value:
its one-variable second derivative has mass $2/(2+t)$, and its
wedge with $S$ keeps only the $d'y\wedge d''y$ term.  Original
product calibration on the split torus has degree one; the
same common-refinement argument used in
\eqref{eq:r3-gauss-box-test-atom} identifies the actual scalar test
value on the Gauss section.  Bounded BT convergence gives the
displayed ordinary-current equality.

Every nonempty fiber of $E$ has length $1/2$, and the coefficient
$2/(2+t)$ is at most one, with equality at zero.  This proves
the subfamily raw value.  Any ordinary open set containing $E$
restricts to an open set on the Gauss box containing its displayed
subset.  Compactness of the lower half of the zero fiber forces
the entire $1/j$ fiber over $y\in[0,1]$ into that open set for
large $j$.  Their coefficients tend to one, so the outer value
is at least one.  Conversely use open covers lying over
$-\eta<y<1+\eta$; every subfamily measure of such a cover is at
most $1+2\eta$.  Letting $\eta\downarrow0$ gives the upper bound.

Fix $0<\eta<1/2$.  The compacts
\[
 K_1=\sigma(\{0\}\times[0,\tfrac12]),\qquad
 K_2=\sigma\!\left((\{0\}\cup\{1/j:j\geq1\})
                              \times[\tfrac12+\eta,1]\right)
\]
are disjoint and lie strictly inside $\Omega$.  Take
$E_i=E\cap K_i$.  One has $\|u_t-u_0\|_\infty\leq t$.
Apply Corollary~\ref{cor:r4-uniform-cluster-pasting} to the constant
row $u_0$ for $E_1$ and the row $u_{1/j}$ for $E_2$.  It gives
\[
 C_S(E;\Omega)\geq\tfrac12+(\tfrac12-\eta)=1-\eta.
\]
Let $\eta\downarrow0$.  The added testers are actual normalized
classical maxima with small positive psh corrections; the proof
does not replace the capacity by the weak closure of the displayed
subfamily.  No exact upper bound for the full capacity is claimed.
\end{proof}

\subsection{A necessary dependence and the unresolved extraction gate}

The separating-cutoff cost in Theorem~\ref{thm:r4-close-tester-pasting}
cannot be bounded solely by the ambient domain.  On the annulus
$-2<\log|T|<2$, choose an original smooth $0\leq\chi_\epsilon\leq1$
whose values at the Gauss radii $1$ and $e^\epsilon$ are zero and
one, respectively.  If $0\leq\rho\leq M$ and
$\rho+2\chi_\epsilon$ are classical, their restrictions to the
inner Gauss interval are convex and have Lipschitz bounds at most
$M$ and $M+2$, respectively.  These are the elementary inner
secant bounds from the proof of
Lemma~\ref{lem:r3-gauss-box-equicontinuity}.  For $0<\epsilon<1$,
\[
 2=2|\chi_\epsilon(e^\epsilon)-\chi_\epsilon(1)|
                 \leq(2M+2)\epsilon,
 \qquad M\geq\epsilon^{-1}-1.
\]
Thus a uniform separation-independent compensator estimate is
\texttt{FALSE}.  Here values at radii mean values at their original
Gauss points; no field extension is used.

For an arbitrary Borel $E$ in a fixed compact core, a sufficient
next step would be the following quantitative extraction statement:
from near-maximizers on ordinary neighborhoods of $E$, obtain
finitely many pairwise disjoint compact pieces $K_i$, normalized
testers $u_i$ with pairwise uniform distance at most $\delta$, and
\[
 \sum_i\mu_{S,u_i}(E\cap K_i)
                  \geq C_S^*(E;\Omega)-\eta,
 \qquad A_K\delta\longrightarrow0\quad(\eta\downarrow0).
\]
The proved pasting inequality would then give $C_S(E)=C_S^*(E)$.
This extraction statement is \texttt{OPEN}.  Neither abstract weak
compactness nor the R3 convex-block theorem supplies it: the latter
does not preserve the unaveraged nonlinear measures, and the
cutoff cost can diverge when the pieces move together.

The pole-carrier theorem supplies a separate sufficient mechanism,
including nonanalytic pole supports, but it requires the stated
mutual null conditions.  It does not separate an arbitrary
non-pluripolar summand from a nested pluripolar summand at no cost.
No previously frozen R3 theorem is changed or contradicted here.

% END source: sections/relative_capacity_next.tex


% BEGIN frozen or reviewed source: sections/completed_dirichlet_next.tex
\section{A complete strong image for a prescribed relative collar class}
\label{sec:r4-completed-collar-image}

The background is arbitrary and fixed. We retain the prescribed
finite extension $K/\mathbf Q_p$, its absolute value, the original
positive cone, and all original compact smooth complementary CLD
tests. Let $\Omega$ be an ordinary algebraic norm-box domain
relatively compact in a smooth affine analytification of pure
dimension $n$. Let $S$ be symmetric, original-positive, separately
$d'$- and $d''$-closed, of bidegree $(p,p)$, and put $q=n-p\geq1$.

Fix a compact $L\Subset\Omega$ contained in a strictly smaller
norm box. Fix a smooth classical psh function $h$ on a neighborhood
of $\overline\Omega$, with semipositive original curvature form,
and a number $B>0$ such that $|h|\leq B$ on $\Omega$. Define
\[
 \mathscr P_{h,L}^{B}
 =\{u\in\mathcal C_b(\Omega): |u|\leq B,\
                   u=h\text{ on }\Omega\setminus L\}.
\]
The exact equality here is an equality of actual point values.
In particular the family is nonempty and convex.

\begin{definition}[The completed admissible collar class]
\label{def:r4-completed-collar-class}
Let $\mathscr D_{S,h,L}^{B}$ be the set of local relative
quasi-everywhere classes having a bounded Borel representative $a$
which is a limit of a sequence $u_m\in\mathscr P_{h,L}^{B}$
locally in outer $S$-capacity on every permitted compact/domain
pair in $\Omega$. The bound and the collar are the same for
all members of each defining sequence and for the entire class.
We use representatives with $|a|\leq B$ and
$a=h$ on $\Omega\setminus L$.
\end{definition}

Such representatives always exist: the summably Cauchy subsequence
construction in the bounded capacity-completion theorem uses the
actual limsup of defining rows, which has both stated properties.
This is a closure of a specified \emph{global} family on $\Omega$.
It is not identified with every locally defined member of the
sheaf $\mathcal A_{S,b}$ having an informal boundary value.

For $a\in\mathscr D_{S,h,L}^{B}$ write
$\mu_a=\mathcal T_{S,q}(a,\ldots,a)$. The completed calculus
already defines this positive Radon measure. On the ordinary open
$\Omega\setminus L$, it equals $\mu_h$ by locality. Thus
$\mu_a-\mu_b$ is a finite signed measure carried by $L$, even
when the individual measures have infinite total mass on $\Omega$.
Put
\[
 d_S(a,b)=
 \sup_{\substack{v\in\mathcal C_b(\Omega)\\0\leq v\leq1}}
       \left|\int v\,d(\mu_a-\mu_b)\right|,
 \qquad
 I_S(a,b)=\int(a-b)\,d(\mu_b-\mu_a).
\]
The latter integral is finite and independent of the representatives:
the completed weighted calculus and coefficient outer-capacity
control annihilate their local relative-null changes.

\begin{lemma}[The energy inequalities pass to the completed class]
\label{lem:r4-completed-collar-energy}
There is a finite constant depending only on $S$, the fixed finite
CLN controls, the compact collar stages, $B$ and $q$ such that
every $a,b\in\mathscr D_{S,h,L}^{B}$ satisfy
\begin{align}
 0\leq I_S(a,b),\qquad
 d_S(a,b)&\leq C I_S(a,b)^{1/2},
       &I_S(a,b)&\leq4B\,d_S(a,b),\label{eq:r4-completed-energy-dual}\\
 \operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega)
   &\leq C\delta^{-2}I_S(a,b)^{2^{-q}}
       &&(\delta>0).\label{eq:r4-completed-energy-capacity}
\end{align}
The gap sets in the second line lie in $L$ for the chosen
representatives. In particular
\begin{equation}
 \operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega)
 \leq C\delta^{-2}
       \bigl(4B\,d_S(a,b)\bigr)^{2^{-q}} .
 \label{eq:r4-data-capacity-inverse}
\end{equation}
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
proved R3 energy inequalities, the completed weighted calculus,
and completed strong convergence. No new energy of an arbitrary
Borel difference is assumed.

\begin{proof}
Choose defining rows $u_m,v_m$ in the same
$\mathscr P_{h,L}^{B}$ for $a,b$. Let $\chi$ be original smooth,
compactly supported in $\Omega$, and equal to one near $L$.
The completed weighted theorem gives
\[
 \int\chi(u_m-v_m)(\mu_{v_m}-\mu_{u_m})
       \longrightarrow
       \int\chi(a-b)(\mu_b-\mu_a)=I_S(a,b).
\]
Each integral on the left is the ordinary compact-collar energy.
R3 strong convergence to the completed measures gives
\[
 \|\mu_{u_m}-\mu_a\|_{\mathrm{psh},\Omega}\to0,\qquad
 \|\mu_{v_m}-\mu_b\|_{\mathrm{psh},\Omega}\to0.
\]
Indeed the measure differences are carried by $L$, so insertion
of $\chi$ does not change them. Consequently
$d_S(u_m,v_m)\to d_S(a,b)$. Passing the original two energy
inequalities to the limit proves
\eqref{eq:r4-completed-energy-dual}. In particular the reverse
inequality has not treated $(a+B)/(2B)$ as a classical tester;
it was proved before passing to the completed limit.

Fix $0<2\epsilon<\delta$. The gap $\{|a-b|>\delta\}$ is
contained in the union of
\[
 L\cap\{|a-u_m|>\epsilon\},\quad
 L\cap\{|b-v_m|>\epsilon\},\quad
 \{|u_m-v_m|>\delta-2\epsilon\}.
\]
The first two sets have outer capacity tending to zero.
Apply the original energy-capacity bound to the third, then
let $m\to\infty$ and $\epsilon\downarrow0$. Outer subadditivity
is sufficient; no continuity from below or compact identification
of raw and outer capacity is used. This proves
\eqref{eq:r4-completed-energy-capacity}, hence the last estimate.
\end{proof}

\begin{proposition}[Capacity completeness with an actual collar]
\label{prop:r4-collar-completion}
The class $\mathscr D_{S,h,L}^{B}$ is complete for local outer
$S$-capacity convergence modulo local relative quasi-everywhere
equality. Every limit has an actual bounded representative
equal to $h$ on $\Omega\setminus L$. Therefore its ordinary
boundary limit on $\partial\Omega$ is $h|_{\partial\Omega}$.
If a fixed continuous lower and upper bound are imposed on all
defining rows, they can be retained by an actual limit
representative as well.
\end{proposition}

\begin{proof}
Use the countably cofinal compact/domain pairs of the R2 completion
theorem. For a capacity-Cauchy sequence $a_j$ in this class,
choose a defining $u_j\in\mathscr P_{h,L}^{B}$ whose capacity
error from $a_j$, at threshold $2^{-j}$, is at most $2^{-j}$
on the first $j$ pairs. The level-set triangle inequality makes
$(u_j)$ capacity Cauchy on every pair. The existing bounded
completion theorem gives a Borel limit. Since all $u_j$ belong
to this one admissible family, the limit belongs to
$\mathscr D_{S,h,L}^{B}$. Its actual limsup representative
from the summable subsequence is bounded by $B$ and equals
$h$ outside $L$. The same triangle inequality identifies it
as the limit of $a_j$ and proves uniqueness modulo the stated
relative-null sets.

A compact subset of $\Omega$ cannot meet a sufficiently small
neighborhood of a prescribed point of $\partial\Omega$.
Thus this representative equals the continuous $h$ near every
ordinary boundary point, proving the boundary assertion.
Any fixed pointwise bounds on the defining rows pass to the
same actual limsup. This last assertion concerns the selected
representative; arbitrary changes on null sets need not preserve
ordinary boundary limits.
\end{proof}

\begin{theorem}[A closed complete strong measure image]
\label{thm:r4-completed-collar-closed-image}
Let $\mathfrak M_{h,L}$ be the positive Radon measures on
$\Omega$ which equal $\mu_h$ on $\Omega\setminus L$, equipped
with
\[
 d_{\mathrm{psh},\Omega}(\nu,\nu')
 =\sup_{\substack{v\in\mathcal C_b(\Omega)\\0\leq v\leq1}}
          \left|\int v\,d(\nu-\nu')\right|.
\]
This is a metric on $\mathfrak M_{h,L}$. The image
\[
 \mathfrak I_{S,h,L}^{B}
   =\{\mu_a:a\in\mathscr D_{S,h,L}^{B}\}
\]
is complete for this metric and hence closed in
$\mathfrak M_{h,L}$. Every measure in the image has a unique
solution class in $\mathscr D_{S,h,L}^{B}$. More precisely,
if $\mu_{a_j}$ is Cauchy in this metric, there is a unique
relative quasi-everywhere class $a$ in the same admissible
completion such that
\[
 a_j\to a\text{ locally in outer }S\text{-capacity},
 \qquad
 d_{\mathrm{psh},\Omega}(\mu_{a_j},\mu_a)\to0.
\]
If the data have a prescribed metric limit $\nu\in\mathfrak M_{h,L}$,
then $\mu_a=\nu$. No ambient classical representative is asserted.
\end{theorem}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
preceding completed estimates and R2 completeness. This is a
solution-image theorem for the specified admissible class,
not an existence criterion for arbitrary dominated measures.

\begin{proof}
First verify that the displayed seminorm separates the measures.
Their difference is a finite signed Radon measure supported on
$L$. Bounded continuous classical psh functions on $\Omega$
include constants and suitable finite truncations of logarithms
of affine regular functions. Such functions separate the points
of the compact set $L$. Their differences form a real vector
lattice, since
\[
 \max(a-b,c-d)=\max(a+d,c+b)-(b+d).
\]
The lattice Stone--Weierstrass theorem gives density in $C(L)$.
Every bounded psh function in this lattice construction can be
shifted and positively scaled into $[0,1]$ on $\Omega$.
Consequently distance zero annihilates a dense set of continuous
tests and hence annihilates the signed measure. This proves
the metric assertion.

Suppose the image measures are metric Cauchy. Estimate
\eqref{eq:r4-data-capacity-inverse} makes the $a_j$ Cauchy
in capacity on the fixed domain $\Omega$. Reverse framed domain
comparison on compact cores gives the required all-domain local
Cauchy condition. Proposition~\ref{prop:r4-collar-completion}
produces $a$ in the same class. The completed strong convergence
theorem gives convergence of $\chi\mu_{a_j}$ to $\chi\mu_a$
in the same psh-dual seminorm, for a compact $\chi=1$ near $L$.
Outside $L$ the measures are already equal to $\mu_h$.
Thus the asserted global difference seminorm tends to zero.
This proves completeness without presuming that
$\mathfrak M_{h,L}$ itself is complete.

If $\mu_a=\mu_b$, then $d_S(a,b)=0$ and the energy-capacity
estimate gives zero outer capacity to every positive gap.
Countable subadditivity and fixed-domain null-class comparison
show that $a=b$ locally relative quasi-everywhere. The converse
is the invariance of the completed calculus. Thus the solution
class is unique. A prescribed metric limit agrees with $\mu_a$
by the metric separation just proved. Completeness implies
closedness, or the same argument applies directly to any
convergent sequence of image measures.
\end{proof}

\paragraph{Scope of the boundary and of existence.}
All defining rows are defined on the same full $\Omega$, obey
one common bound, and agree with one actual collar. This is
stronger than merely sharing a boundary limit. The theorem
neither supplies a classical representative nor proves a
bounded-subsolution existence theorem for arbitrary $S$.
For $q=1$ the admissible family and its completion are convex
and the measure map is linear, so this particular closed image
is convex as well. For higher degree no convexity of its
nonlinear image is asserted. The arguments use no arbitrary
varying-background continuity and no relative obstacle
orthogonality.

% END source: sections/completed_dirichlet_next.tex


% BEGIN frozen or reviewed source: sections/completed_interior_dirichlet.tex
\section{A closed interior Dirichlet image with fixed actual exterior values}
\label{sec:r4-interior-completed-image}

Retain the arbitrary fixed $S$, degree $q=n-p\geq1$, norm-box
domain $\Omega$, smooth outer collar $h$, compact $L$ and bound
$B>0$ of Section~\ref{sec:r4-completed-collar-image}.
Choose an ordinary open $V$ with
$\overline V\subset L\Subset\Omega$, and a compact
$K\Subset V$. Let $e$ be a fixed bounded actual function on
$\Omega\setminus V$, equal to $h$ off $L$. In particular
$e$ is specified at every point of $\partial V$.

Fix continuous functions $\ell,r$ on $\overline V$ with
$\ell\leq r$ and
\[
 \ell|_{\partial V}=r|_{\partial V}
                 =e|_{\partial V}=g.
\]
Assume that the following class is nonempty:
\[
 \mathscr P_{\rm bd}
 =\{u\in\mathscr P_{h,L}^{B}:
       u=e\text{ on }\Omega\setminus V,\ 
       \ell\leq u\leq r\text{ on }V\}.
\]
All equalities and inequalities in this definition are actual.
No assertion of nonemptiness for arbitrary data or barriers
is implicit in the notation.

Let $\mathscr D_{\rm bd}$ be the closure of this global family
in the all-domain local outer $S$-capacity sense, modulo
local relative quasi-everywhere equality. Its representatives
are selected by the actual limsup construction from defining
rows. Thus they satisfy $|a|\leq B$,
$a=e$ at every point of $\Omega\setminus V$, and
$\ell\leq a\leq r$ on $V$. In particular every such selected
representative has ordinary boundary value $g$.
Define
\[
 \mathscr D_{{\rm bd},K}
 =\{a\in\mathscr D_{\rm bd}:
                \nu_a:=\mathbf1_V\mu_a
                   \text{ is carried by }K\},
 \qquad
 \mu_a=\mathcal T_{S,q}(a,\ldots,a).
\]
The interior measure $\nu_a$ is finite. We do not require
$\mu_a(\partial V)=0$, or require the boundary restrictions
of $\mu_a$ to agree for different members.

\begin{lemma}[The boundary part disappears from the energy multiplier]
\label{lem:r4-interior-energy}
For $a,b\in\mathscr D_{{\rm bd},K}$, the completed collar
energy satisfies
\begin{equation}
 I_S(a,b)=\int_K(a-b)\,d(\nu_b-\nu_a),
 \qquad
 0\leq I_S(a,b)
 \leq4B\,d_{\mathrm{psh},V}(\nu_a,\nu_b),
 \label{eq:r4-interior-energy}
\end{equation}
where
\[
 d_{\mathrm{psh},V}(\nu,\nu')
 =\sup_{\substack{v\in\mathcal C_b(V)\\0\leq v\leq1}}
                 \left|\int_K v\,d(\nu-\nu')\right|.
\]
Consequently
\begin{equation}
 \operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega)
 \leq C\delta^{-2}
       \bigl(4B\,d_{\mathrm{psh},V}(\nu_a,\nu_b)\bigr)^{2^{-q}}.
 \label{eq:r4-interior-inverse-capacity}
\end{equation}
The constants have the same fixed dependencies as the
outer-collar energy-capacity estimate.
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}.}

\begin{proof}
For the selected representatives $a-b$ is exactly zero on
$\Omega\setminus V$, including every point of $\partial V$.
The difference $\mu_b-\mu_a$ is a finite signed measure on
the fixed compact $L$. Borel multiplication therefore gives
\[
 (a-b)(\mu_b-\mu_a)
       =\mathbf1_V(a-b)(\mu_b-\mu_a)
       =(a-b)(\nu_b-\nu_a),
\]
as equality of finite signed measures. This is not a
statement that the boundary measures themselves vanish.
The completed weighted calculus makes each product invariant
under relative-null modifications, but the displayed
cancellation uses the chosen actual equality on the boundary,
not an unsupported representative convention.
Integration proves the first formula. Nonnegativity follows
from Lemma~\ref{lem:r4-completed-collar-energy}.

Choose global defining rows $u_m,v_m\in\mathscr P_{\rm bd}$
for $a,b$. On the compact $K$, each of the fixed measures
$\nu_a,\nu_b$ is uniformly absolutely continuous with respect
to the relative outer capacity with domain $V$. Indeed the
rows restricted to $V$ have the common bound $B$;
normalizing them into $[0,1]$ and using the completed
coefficient outer-capacity bound yields
\[
 \nu_c(E)\leq (2B)^q
                     \operatorname{Cap}_S^*(E;V),
 \qquad E\subset K\text{ Borel},\quad c=a,b.
\]
This conclusion can also be obtained from the open-set
weak-limit inequality followed by the outer regularity of
each fixed Radon measure. It uses no compact equality of
raw and outer BT capacity.
Capacity convergence and the uniform bounds therefore imply
\[
 \int_K(u_m-v_m)\,d(\nu_b-\nu_a)
       \longrightarrow\int_K(a-b)\,d(\nu_b-\nu_a).
\]
Both $(u_m+B)/(2B)$ and $(v_m+B)/(2B)$ are legitimate
normalized classical testers on $V$. Their constant terms
cancel in the difference, so the absolute value of each
integral on the left is at most
$4B\,d_{\mathrm{psh},V}(\nu_a,\nu_b)$.
Passing to the limit proves \eqref{eq:r4-interior-energy}.
Apply the already proved completed energy-capacity estimate
to obtain \eqref{eq:r4-interior-inverse-capacity}.
\end{proof}

\begin{theorem}[Completeness and closedness of the interior image]
\label{thm:r4-interior-closed-image}
Let $\mathfrak M_K^+(V)$ be the finite positive Radon
measures carried by $K$, equipped with
$d_{\mathrm{psh},V}$. This is a metric space. The image
\[
 \mathfrak I_{{\rm bd},K}
      =\{\nu_a:a\in\mathscr D_{{\rm bd},K}\}
\]
is complete, and hence closed, in that metric space.
Every datum in this image has a unique local relative
quasi-everywhere solution class in
$\mathscr D_{{\rm bd},K}$.

More precisely, if $\nu_{a_j}$ is Cauchy in this metric,
then for a unique class $a\in\mathscr D_{{\rm bd},K}$,
\[
 a_j\longrightarrow a
       \quad\text{locally in outer }S\text{-capacity
                    on every permitted pair in }\Omega,
 \qquad
 d_{\mathrm{psh},V}(\nu_{a_j},\nu_a)\longrightarrow0.
\]
The selected limit representative retains the actual exterior
$e$, the two continuous barriers, and the ordinary boundary
value $g$. If the given data already converge in the metric
to $\nu\in\mathfrak M_K^+(V)$, then $\nu_a=\nu$.
\end{theorem}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
completed energy estimate, bounded capacity completeness,
and completed localized strong convergence.

\begin{proof}
Bounded continuous classical functions on $V$ obtained by
truncating logarithms of affine regular functions separate
points of $K$. Their differences form a vector lattice
containing constants. The same lattice Stone--Weierstrass
argument as in Theorem~\ref{thm:r4-completed-collar-closed-image}
shows that $d_{\mathrm{psh},V}$ separates finite signed
measures carried by $K$. It is finite because every tester
lies between zero and one. Thus it is a metric.

The class $\mathscr D_{\rm bd}$ is capacity complete.
For a Cauchy sequence in this class, choose a diagonal
sequence of its defining rows in the one family
$\mathscr P_{\rm bd}$, controlling the first $j$ compact/domain
pairs at step $j$. The existing capacity completion then
produces a limit defined by those rows. Their actual limsup
preserves $e$ at all points outside $V$, the bound $B$, and
the barriers. Since the barriers are continuous and agree
at each boundary point, they give the asserted ordinary
boundary limit. This is the argument of
Proposition~\ref{prop:r4-collar-completion} applied to the
more restricted defining family.

For the sequence in the theorem,
\eqref{eq:r4-interior-inverse-capacity} first gives capacity
Cauchyness on $\Omega$. Reverse framed domain comparison
on compact cores gives the all-domain local condition.
Hence there is $a\in\mathscr D_{\rm bd}$ with the stated
capacity convergence. Completed weak convergence on $V$
shows that $\mu_a$ vanishes on the ordinary open $V\setminus K$:
all the positive measures $\mu_{a_j}$ vanish there.
Thus $a\in\mathscr D_{{\rm bd},K}$.

Choose an original compact smooth $\chi$ on $V$ equal to one
near $K$. Then
$\chi\mu_{a_j}=\nu_{a_j}$ and $\chi\mu_a=\nu_a$.
Completed localized strong convergence with the fixed
background $S|_V$ gives
\[
 \sup_{\substack{v\in\mathcal C_b(V)\\0\leq v\leq1}}
   \left|\int v\chi\,d(\mu_{a_j}-\mu_a)\right|\longrightarrow0.
\]
This is exactly the required metric convergence of the
interior data. In particular variable boundary measures
of the global extensions never enter this final passage:
the cutoff has support strictly inside $V$.
This proves completeness. It also identifies any prescribed
metric limit by separation.

If $\nu_a=\nu_b$, then \eqref{eq:r4-interior-energy} gives
$I_S(a,b)=0$. The completed energy-capacity estimate makes
every positive gap outer-null, and the countable union over
rational thresholds gives local relative quasi-everywhere
equality. Conversely that equality preserves the completed
operator. This proves the asserted uniqueness of classes.
\end{proof}

\paragraph{Scope.}
The theorem retains a single global defining bound and a
specified exterior extension. It does not assert that a
general datum dominated by a relative Monge--Ampere measure
belongs to this closed image. The smooth nonpolar
degree-one counterexample of
Theorem~\ref{thm:r4-nonpolar-completed-subsolution-failure}
rules out such an unrestricted inference even for a natural
global bounded completion. Nor is pointwise uniqueness or
an ambient classical representative asserted here.

% END source: sections/completed_interior_dirichlet.tex


% BEGIN frozen or reviewed source: sections/completed_variational_retraction.tex
\section{A relative variational retraction and its exact Euler inequality}
\label{sec:r4-variational-retraction}

Retain the arbitrary fixed $S$, $q\geq1$, norm-box domain
$\Omega$, compact $L$, smooth classical collar $h$, common bound
$B>0$, and completed class $\mathscr D=\mathscr D_{S,h,L}^{B}$
of Section~\ref{sec:r4-completed-collar-image}. Every class has
global defining rows with that bound and that collar. Equality
and inequalities between completed potentials are understood
locally relative quasi-everywhere unless actual representatives
are specified.

The purpose of this section is to replace an unjustified relative
obstacle orthogonality assertion by an exact constrained variational
statement. Its output need not solve a prescribed Monge--Ampere
equation. The failure of an unrestricted existence assertion is
illustrated by Theorem~\ref{thm:r4-nonpolar-completed-subsolution-failure}.

\begin{definition}[Admissible data and admissible convex constraints]
\label{def:r4-variational-data}
Let $\mathfrak N_{S,h,L}$ consist of the measures
$\nu\in\mathfrak M_{h,L}$ whose restriction to $L$ is uniformly
absolutely continuous at zero with respect to the fixed outer
capacity: for every $\epsilon>0$ there is $\delta>0$ such that
\[
 E\subset L\text{ Borel},\quad
 \operatorname{Cap}_S^*(E;\Omega)<\delta
       \quad\Longrightarrow\quad \nu(E)<\epsilon .
\]
In particular $\nu$ annihilates the relevant relative-null sets.
Fix a nonempty convex subset $\mathscr K\subset\mathscr D$ which
is closed for local outer-capacity convergence.
\end{definition}

All image measures $\mu_a$, $a\in\mathscr D$, belong to
$\mathfrak N_{S,h,L}$. Indeed the completed coefficient estimate
\eqref{eq:r2-limit-coefficient-outer}, with a fixed compact
scalar equal to one near $L$, gives
$\mu_a(E)\leq C\operatorname{Cap}_S^*(E;\Omega)$ for
$E\subset L$. The common defining bound makes this control
uniform on the whole class.

An example of $\mathscr K$ is a nonempty family defined by
bounded Borel obstacles $\ell\leq a\leq r$ quasi-everywhere.
It is convex. It is capacity closed because, for $\delta>0$,
the set $\{a>r+\delta\}$ is contained in
$\{|a-a_j|>\delta\}\cup\{a_j>r\}$ whenever $a_j\to a$;
take outer capacities on each compact stage, then take a
countable union of positive gaps. The lower inequality is
identical. This does not assert that the pointwise obstacle
envelope is classical or has any contact measure property.

Define the completed compact-collar energy by
\[
 \mathcal E(a)=\frac1{q+1}\sum_{r=0}^q
   \int (a-h)\,
     \mathcal T_{S,q}(\underbrace{a,\ldots,a}_{r},
                       \underbrace{h,\ldots,h}_{q-r}).
\]
Each weighted integral is supported in $L$. For
$\nu\in\mathfrak N_{S,h,L}$ put
\[
 \mathcal F_\nu(a)=\mathcal E(a)
                       -\int_L(a-h)\,d\nu.
\]

\begin{lemma}[Completed energy and datum pairings]
\label{lem:r4-variational-energy-pairing}
The energy and these pairings are finite and well defined on
relative classes. They are continuous under local outer-capacity
convergence in $\mathscr D$, for each fixed admissible $\nu$.
The energy is concave, and for $a_t=(1-t)a+tb$,
\begin{align}
 \frac{d}{dt}\mathcal E(a_t)
       &=\int_L(b-a)\,d\mu_{a_t},\label{eq:r4-energy-variation}\\
 \mathcal E((a+b)/2)
       -\tfrac12\mathcal E(a)-\tfrac12\mathcal E(b)
       &\geq c_q I_S(a,b),\qquad
             c_q=\frac{q}{4^{q+1}}.\label{eq:r4-energy-midpoint}
\end{align}
For $\nu,\nu'\in\mathfrak N_{S,h,L}$ and $a,b\in\mathscr D$,
\begin{equation}
 \left|\int_L(a-b)\,d(\nu-\nu')\right|
       \leq4B\,d_{\mathrm{psh},\Omega}(\nu,\nu').
 \label{eq:r4-completed-data-pairing}
\end{equation}
\end{lemma}

\begin{proof}
Relative CLN and $|a-h|\leq2B$ give a uniform finite bound
for every term of the energy, by passage from the defining
classical rows. Their weighted completed convergence proves
continuity of the energy. It also proves independence of
relative-null changes, including in the weight slot.
The data term is finite because $\nu(L)<\infty$.
Its independence follows from the defining absolute continuity.

If $a_j\to a$ in capacity, split $L$ into
$|a_j-a|\leq\epsilon$ and its complementary gap. Then
\[
 \int_L|a_j-a|\,d\nu
 \leq\epsilon\nu(L)+2B\,\nu(L\cap\{|a_j-a|>\epsilon\})
                 \longrightarrow\epsilon\nu(L).
\]
The fixed-domain capacity of the gap tends to zero, by the
all-domain hypothesis and finite localization. Let
$\epsilon\downarrow0$. This proves continuity of the pairing.

For defining rows $u_m,v_m$ of $a,b$, the energy along
$(1-t)u_m+tv_m$ is a polynomial of degree at most $q+1$.
The finite coefficients converge by the completed weighted
calculus. Passing the original finite-polynomial derivative
identity therefore gives \eqref{eq:r4-energy-variation}.
Passing concavity and the R3 midpoint inequality, together
with the convergence of $I_S(u_m,v_m)$, gives
\eqref{eq:r4-energy-midpoint}. No differentiation of an
uncontrolled weak limit is involved.

For the last estimate, the globally defined functions
$(u_m+B)/(2B)$ and $(v_m+B)/(2B)$ are normalized classical
testers on $\Omega$. Hence
\[
 \left|\int_L(u_m-v_m)\,d(\nu-\nu')\right|
        \leq4B\,d_{\mathrm{psh},\Omega}(\nu,\nu').
\]
For these fixed two data measures, the capacity continuity
of their integrals just proved permits $m\to\infty$.
This yields \eqref{eq:r4-completed-data-pairing}.
The completed functions themselves were not declared
admissible classical testers in the definition of the metric.
\end{proof}

\begin{theorem}[The constrained relative variational problem]
\label{thm:r4-variational-maximizer}
For every $\nu\in\mathfrak N_{S,h,L}$, the functional
$\mathcal F_\nu$ has a unique maximizing relative class
$a_\nu\in\mathscr K$. It is characterized by
\begin{equation}
 \int_L(b-a_\nu)\,d(\mu_{a_\nu}-\nu)\leq0
                          \qquad(b\in\mathscr K).
 \label{eq:r4-variational-inequality}
\end{equation}
Every maximizing sequence converges to this class locally
in outer $S$-capacity.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for this
constrained completed variational construction over arbitrary
fixed $S$. Its inputs are the already proved energy and
completion theorems; it is not a new CLN estimate or a relative
contact-set theorem.

\begin{proof}
There is a uniform finite bound on $|\mathcal E(a)|$ for
$a\in\mathscr D$, and
$|\int_L(a-h)d\nu|\leq2B\nu(L)$. Thus the supremum
$m_\nu=\sup_{\mathscr K}\mathcal F_\nu$ is finite, and
$\mathscr K\ne\varnothing$ gives a maximizing sequence $a_j$.
Convexity makes $(a_j+a_k)/2$ admissible. The measure term
is affine, so \eqref{eq:r4-energy-midpoint} gives
\[
 0\leq c_q I_S(a_j,a_k)
 \leq m_\nu-\tfrac12\mathcal F_\nu(a_j)
                   -\tfrac12\mathcal F_\nu(a_k)
                   \longrightarrow0.
\]
The sign is essential: the concave midpoint value lies above
the arithmetic mean and below the supremum.
The completed energy-capacity estimate makes this sequence
capacity Cauchy, first on the fixed domain and then on every
smaller domain by reverse framed comparison.
The collar completion is complete, and the capacity-closed
constraint gives a limit $a_\nu\in\mathscr K$.
Lemma~\ref{lem:r4-variational-energy-pairing} passes the
objective to its limit, proving attainment.

If $a,b$ are two maximizers, the same midpoint bound gives
$I_S(a,b)=0$, so the completed energy-capacity estimate gives
$a=b$ relative quasi-everywhere. The preceding Cauchy argument
and uniqueness apply to every maximizing sequence.

For $b\in\mathscr K$, the segment
$a_\nu+t(b-a_\nu)$ stays admissible for $0\leq t\leq1$.
Its one-sided derivative at zero is nonpositive at a maximum.
Use \eqref{eq:r4-energy-variation} to obtain
\eqref{eq:r4-variational-inequality}.
Conversely, concavity gives
\[
 \mathcal F_\nu(b)-\mathcal F_\nu(a)
       \leq\int_L(b-a)\,d(\mu_a-\nu).
\]
If all the displayed inequalities hold, this is nonpositive
for every $b\in\mathscr K$, proving maximality and the stated
characterization.
\end{proof}

\begin{theorem}[Strong stability and a retraction onto the image]
\label{thm:r4-variational-retraction}
For $\nu,\nu'\in\mathfrak N_{S,h,L}$, put
$a=a_\nu$, $b=a_{\nu'}$. Then
\begin{align}
 I_S(a,b)
 &\leq\int_L(a-b)\,d(\nu'-\nu)
       \leq4B\,d_{\mathrm{psh},\Omega}(\nu,\nu'),
       \label{eq:r4-variational-monotonicity}\\
 d_{\mathrm{psh},\Omega}(\mu_a,\mu_b)
 &\leq C\bigl(4B\,d_{\mathrm{psh},\Omega}(\nu,\nu')\bigr)^{1/2},
       \label{eq:r4-retraction-strong-stability}\\
 \operatorname{Cap}_S^*(\{|a-b|>\delta\};\Omega)
 &\leq C\delta^{-2}
       \bigl(4B\,d_{\mathrm{psh},\Omega}(\nu,\nu')\bigr)^{2^{-q}} .
       \label{eq:r4-retraction-capacity-stability}
\end{align}
The map
\[
 R_{\mathscr K}:\mathfrak N_{S,h,L}\longrightarrow
       \mathfrak I_{\mathscr K}:=\{\mu_c:c\in\mathscr K\},
 \qquad R_{\mathscr K}(\nu)=\mu_{a_\nu},
\]
is a retraction: it is the identity on
$\mathfrak I_{\mathscr K}$. This image is complete and closed
in $\mathfrak M_{h,L}$. A datum has a solution in
$\mathscr K$ exactly when it is fixed by $R_{\mathscr K}$,
and that solution class is unique.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the
variational retraction and its stability; the closed-image
assertion is the preceding collar theorem applied to a
capacity-closed subset.

\begin{proof}
Use \eqref{eq:r4-variational-inequality} for the competitor
$b$ at the maximizer $a$, and for the competitor $a$ at $b$:
\[
 \int(b-a)\,d(\mu_a-\nu)\leq0,\qquad
 \int(a-b)\,d(\mu_b-\nu')\leq0.
\]
Adding and rearranging yields the first inequality in
\eqref{eq:r4-variational-monotonicity}, since
$I_S(a,b)=\int(a-b)d(\mu_b-\mu_a)$.
The datum-pairing lemma proves the second inequality.
The completed energy-dual and energy-capacity estimates
give the next two bounds. Their constants depend on the
fixed admissible family, not on moduli of absolute continuity
chosen separately for the two data measures.

For $c\in\mathscr K$ and $\nu=\mu_c$, concavity gives
$\mathcal F_\nu(b)\leq\mathcal F_\nu(c)$ for every
$b\in\mathscr K$. Uniqueness of the maximizer implies
$a_\nu=c$, so $R_{\mathscr K}(\nu)=\nu$.
Every image measure is admissible data by the coefficient
outer-capacity estimate stated above, making this a retraction
with its asserted domain and codomain.

For completeness of the image, a metric-Cauchy sequence
$\mu_{c_j}$ makes $c_j$ capacity Cauchy by the collar estimate.
Its limit lies in $\mathscr D$ by completeness and in
$\mathscr K$ by capacity closedness. Completed strong
convergence gives the image limit. The metric separates
the ambient measures, so this complete image is closed.
Finally, the retraction identity proves the fixed-point
criterion, and the injectivity of the collar measure map
gives uniqueness modulo relative-null sets.
\end{proof}

\begin{corollary}[What the Euler inequality actually permits]
\label{cor:r4-feasible-direction-equality}
Suppose a bounded Borel direction $\xi$, zero outside $L$,
satisfies $a_\nu+t\xi\in\mathscr K$ and
$a_\nu-t\xi\in\mathscr K$ for some $t>0$. Then
\[
 \int_L\xi\,d(\mu_{a_\nu}-\nu)=0.
\]
\end{corollary}

\begin{proof}
Use the two stated competitors in
\eqref{eq:r4-variational-inequality} and divide by $t$.
\end{proof}

This is equality only in directions known to remain inside
the completed psh constraint. A pointwise gap below an obstacle
does not by itself make arbitrary smooth positive and negative
perturbations admissible. Consequently the variational inequality
does not imply a contact-support formula or
$\mu_{a_\nu}=\nu$ for arbitrary data. Those would require an
additional theorem supplying the missing feasible directions.
This precise distinction is compatible with the smooth nonpolar
counterexample and avoids the false general ambient-relative
contact mechanism.

% END source: sections/completed_variational_retraction.tex


% BEGIN frozen or reviewed source: sections/nonpolar_trace_dirichlet_obstruction.tex
\section{A smooth nonpolar background can still obstruct completed solvability}
\label{sec:r4-nonpolar-trace-obstruction}

This counterexample has remaining degree one, in every ambient
dimension $n\geq2$. It rules out a general bounded-subsolution
criterion even for the global bounded capacitary closure of
classical potentials. The global defining-row hypothesis is
stated explicitly; no assertion identifying it with the entire
locally defined sheaf $\mathcal A_{S,b}$ is needed.

Let $Y=(\mathbf G_{m,K}^{n})^{\rm an}$ over the original finite
extension $K/\mathbf Q_p$, and put
\[
 t_i=\log|T_i|,\qquad
 t=(t_1,\ldots,t_n),\qquad
 B=\{x\in\mathbf R^n:|x|<1\},\qquad V=t^{-1}(B).
\]
The closure of $V$ is compact in the torus: it is contained in
the compact polyannulus $-1\leq t_i\leq1$. Its ordinary boundary
is contained in $\{|t|=1\}$. The function
\[
 \rho=|t|^2-1
\]
is smooth continuous classical psh on the torus, negative on
$V$, and zero on its ordinary boundary. It is an exhaustion
there, since each negative sublevel is compact in $V$.
Write
\[
 \beta=\sum_{i=1}^n d't_i\wedge d''t_i,\qquad
 S=[\beta^{n-1}].
\]
The brackets indicate the ordinary current of this smooth form.
This is a nonzero symmetric original-positive separately closed
current of bidegree $(n-1,n-1)$.

Let $\sigma:\mathbf R^n\to Y$ be the original Gauss section,
characterized on Laurent polynomials by
\[
 \left|\sum_m c_mT^m\right|_{\sigma(x)}
                  =\max_m |c_m|e^{m\cdot x}.
\]
Choose a nonzero real function $f\in C_c^\infty(B)$ with
$0\leq f\leq1$, and define the positive ordinary top measure
\[
 \mu=f(t)\,\beta^n .
\]
All equalities below concern original CLD currents and tests.

\begin{lemma}[Original torus integration and separate convexity]
\label{lem:r4-trace-calibration}
On the standard torus one has the equality of Radon measures
\begin{equation}
 \beta^n=n!\,\sigma_*(dx),
 \label{eq:r4-torus-volume}
\end{equation}
where $dx$ is Lebesgue measure. For any actual classical psh
$u$ on $V$, its Gauss restriction $U(x)=u(\sigma(x))$ is
convex in each coordinate separately on every real box contained
in $B$. If $|u_j|\leq M$ on $V$, the restrictions $U_j$ are
uniformly bounded and uniformly Lipschitz on every smaller real
box compactly contained in $B$.
\end{lemma}

\begin{proof}
The original tropical integration formula on the coordinate torus
has tropical multiplicity one. The expansion of the top power
contains $n!$ copies of the oriented coordinate volume form,
which proves \eqref{eq:r4-torus-volume}. This is an
ordinary-current statement: upon refining an original test
presentation, the full-rank cells projecting to the coordinate
tropicalization carry the Gauss-section integral with the same
calibration, whereas cells of smaller projection rank contribute
zero to the displayed top form. Equivalently, the ordinary
coordinate-torus integration formula may be tested first against
arbitrary original smooth compact scalars and then extended by
Radon uniqueness. This is the same original torus calibration
used in the proved Gauss-box results.

For clarity, the separate convexity uses the actual classical
condition after complete extensions. Fix all coordinates but
the $i$th, and take the completed residue field of their
$(n-1)$-dimensional Gauss point. Their coordinate images are
constants in that field with the prescribed absolute values
and Gauss independence. The analytic curve with $T_i=z$ and
the remaining $T_l$ equal to these constants maps the entire
annulus corresponding to an admissible interval into $V$ after
that complete extension. Its Gauss point of radius $e^s$ maps
back to $\sigma(x_1,\ldots,s,\ldots,x_n)$.

The pullback of $u$ is subharmonic by the classical definition.
On any closed subannulus subtract the affine function of
$\log|z|$ interpolating its two endpoint Gauss values. This is
a harmonic unit logarithm, and the subharmonic maximum theorem
with the two Shilov endpoints makes the difference nonpositive.
This is exactly the convexity inequality. It uses a full
annulus, not merely an ordinary neighborhood of its Gauss point.
The argument is also Lemma~\ref{lem:r3-gauss-box-equicontinuity}.

A one-dimensional convex function bounded by $M$ in absolute
value has slopes of absolute value at most $2M/\delta$ on an
interval at distance $\delta$ from the endpoints of a larger
interval. Apply this to each coordinate on nested real boxes.
Changing one coordinate at a time gives the uniform local
Lipschitz bound for the $\ell^1$ distance.
\end{proof}

\begin{definition}[Global bounded defining rows on $V$]
\label{def:r4-global-completed-row}
A bounded Borel function $a$ belongs to
$\mathscr G_{S,b}(V)$ if, for some finite $M$, there are
$u_j\in\mathcal C_b(V)$ with $|u_j|\leq M$ on the whole $V$
such that $u_j\to a$ locally in outer $S$-capacity on every
permitted compact/domain pair in $V$. The representative $a$
is required to be bounded as an actual function.
\end{definition}

This is the bounded global closure with precisely the same
convergence used in R2. Its members belong locally to
$\mathcal A_{S,b}$, and for remaining degree one their
completed operator is $D(aS)$. Restrictions of globally
bounded defining rows from a larger collar domain are included.

\begin{lemma}[Traces of global completed potentials]
\label{lem:r4-completed-trace-convexity}
If $a\in\mathscr G_{S,b}(V)$, the Borel function
$A(x)=a(\sigma(x))$ has, up to Lebesgue-almost-everywhere
equality, a continuous separately convex representative $U$
on $B$. If $a$ has ordinary boundary limit zero on $V$, then
$U$ has limit zero on the real sphere $\partial B$.
\end{lemma}

\begin{proof}
The global function $v=|t|^2$ is a normalized classical tester
on $V$ and satisfies
\[
 D(vS)=2\beta^n.
\]
Consequently, for every Borel $E\subset V$,
\begin{equation}
 2\beta^n(E)\leq\operatorname{Cap}_S(E;V)
                         \leq\operatorname{Cap}_S^*(E;V).
 \label{eq:r4-capacity-controls-volume}
\end{equation}
For a real compact $F\Subset B$, the subset $t^{-1}(F)$
is compact in $V$. Local all-domain capacity convergence and
a finite covering by permitted compact norm-box stages imply
convergence in outer capacity on this compact set with capacity
domain $V$, by domain monotonicity and subadditivity.

Take a common bound for $a$ and its defining row. Split
$t^{-1}(F)$ into the gap $|u_j-a|>\epsilon$ and its complement.
Formula \eqref{eq:r4-capacity-controls-volume} and
\eqref{eq:r4-torus-volume} give
\[
 \int_F|u_j(\sigma(x))-A(x)|\,dx\longrightarrow0.
\]
More explicitly, the integral is at most
$\epsilon\,\operatorname{vol}(F)$ plus a fixed constant times
the outer capacity of that gap; let $j\to\infty$ and then
$\epsilon\downarrow0$.

The preceding lemma gives local equicontinuity and boundedness
of the restrictions $U_j=u_j\circ\sigma$. Arzela--Ascoli on
an exhaustion of $B$ by compact sets, followed by a diagonal
subsequence, gives local uniform convergence to a continuous
separately convex $U$. The $L^1_{\rm loc}$ limit just identified
makes $U=A$ almost everywhere.

Every $\sigma(y)$ for $y\in\partial B$ is an ordinary boundary
point of $V$, since it is approached through the continuous
Gauss section from $B$ and has $|t|=1$. If $a$ has zero
ordinary boundary limit, compactness of this Gauss sphere
implies that for every $\epsilon>0$ there is $r_\epsilon<1$
such that
\[
 |A(x)|<\epsilon\quad(r_\epsilon<|x|<1).
\]
Otherwise a sequence tending to the sphere, and a convergent
subsequence of its Gauss images, contradict the boundary limit.
Since $U=A$ almost everywhere and $U$ is continuous, the same
collar satisfies $|U|\leq\epsilon$. This proves the last claim.
It does not assume that an arbitrary relative classical
representative automatically inherits boundary values.
\end{proof}

\begin{theorem}[Nonpolar bounded-subsolution failure]
\label{thm:r4-nonpolar-completed-subsolution-failure}
The current $S$ above has no coefficient mass on any ambient
locally pluripolar set. The measure $\mu$ is nonzero, finite,
nonatomic and compactly supported in $V$. The smooth bounded
classical function
\[
 w=\tfrac12(|t|^2-1)
\]
has ordinary boundary value zero and satisfies
\[
 D(wS)=\beta^n\geq\mu .
\]
Nevertheless there is no $a\in\mathscr G_{S,b}(V)$ with
ordinary boundary limit zero and
\[
 D(aS)=\mu .
\]
In particular there is no bounded actual classical solution.
Thus a general bounded-subsolution criterion in this globally
defined bounded relative completion is \texttt{FALSE}, even
for a smooth nonpolar background and compact nonatomic data.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} counterexample.

\begin{proof}
The smooth potential $\psi=|t|^2/2$ has $D\psi=\beta$ and is
bounded on $V$. Hence $S$ is the ambient bounded BT product
$(D\psi)^{n-1}$. Ambient coefficient-capacity control and the
proved ambient polar outer-zero theorem imply that every
coefficient variation of $S$ gives zero mass to locally
pluripolar sets. This proves the precise nonpolar assertion;
it is stronger than an assertion about only one top trace.
Formula \eqref{eq:r4-torus-volume} gives
\[
 \mu=n!\,\sigma_*(f(x)\,dx).
\]
It is therefore finite, nonatomic, nonzero, and carried by the
compact $\sigma(\operatorname{supp}f)\Subset V$.
The displayed subsolution calculation follows from $D\rho=2\beta$.

Suppose the asserted solution $a$ exists, and let $U$ be the
continuous separately convex representative of its Gauss trace
from Lemma~\ref{lem:r4-completed-trace-convexity}. For
$\varphi\in C_c^\infty(B)$, $\varphi(t)$ is an original smooth
compact test on $V$, because inverse images of compact
tropical sets are compact in the torus. The original smooth
form identity is
\[
 D\varphi(t)\wedge\beta^{n-1}
                  =\frac{\Delta\varphi(t)}{n}\,\beta^n .
\]
The equation $D(aS)=\mu$, Borel multiplication of the
coefficient-Radon current, and
\eqref{eq:r4-torus-volume} consequently give
\[
 (n-1)!\int_B U\,\Delta\varphi\,dx
             =n!\int_B f\varphi\,dx.
\]
Thus
\begin{equation}
 \Delta U=n f
 \quad\text{as real distributions on }B .
 \label{eq:r4-trace-poisson}
\end{equation}
Only legitimate original tests were used; equality on their
toric subclass is a necessary consequence of the proposed
full-current equation.

Choose $0<r_0<1$ with
$\operatorname{supp}f\subset\{|x|<r_0\}$. On the connected
annulus $A=\{r_0<|x|<1\}$, separate convexity makes each
distribution $\partial_i^2U$ positive. This follows directly
by one-dimensional convexity and Fubini on small real boxes.
Their sum is zero by \eqref{eq:r4-trace-poisson}; hence each
$\partial_i^2U=0$ there.

On every real box in $A$, the continuous function $U$ is
therefore affine in each coordinate, and successive linear
interpolation gives one multiaffine polynomial on that box.
Polynomials from overlapping boxes agree on their open
intersection, and thus agree identically. Since $A$ is
connected when $n\geq2$, a chain of overlapping boxes gives
a single multiaffine polynomial $P$ on the whole annulus.
The boundary assertion of the preceding lemma gives $P=0$
on the real unit sphere. Write
$P(x',x_n)=P_0(x')+x_nP_1(x')$.
For every $|x'|<1$, the two sphere points
$x_n=\pm\sqrt{1-|x'|^2}$ imply $P_0(x')=P_1(x')=0$.
Both polynomials vanish identically, so $P=0$.

It follows that $U=0$ on the outer annulus and is compactly
supported inside $B$. Choose $\eta\in C_c^\infty(B)$ equal
to one on $\{|x|\leq r_0\}$. Its derivatives are supported
where $U=0$, and \eqref{eq:r4-trace-poisson} yields
\[
 0=\int_B U\,\Delta\eta\,dx=n\int_B f\,dx>0.
\]
This contradiction proves the theorem.
\end{proof}

\paragraph{Precise scope of the negative result.}
The global rows on the whole torus-preimage domain are essential
in the trace-convexity argument. A classical potential defined only
on an ordinary neighborhood of a Gauss point need not have a
convex restriction to a preassigned skeleton. For example,
$-\log|T-1|$ is harmonic on a neighborhood avoiding $T=1$,
whereas on the usual Gauss line its restriction is
$-\max(t,0)$, which is concave. The proof therefore makes no
unjustified passage from local representatives of the sheaf
$\mathcal A_{S,b}$ to global defining rows.

The example is also compatible with the closed-image theorem
\ref{thm:r4-completed-collar-closed-image}. That theorem closes
the image of one bounded admissible family in a strong data
topology; it does not assert that every measure dominated by
a member belongs to that image. Here $0\leq\mu\leq D(wS)$
does not supply even a globally completed psh solution with
the specified zero boundary limit. A larger, non-psh energy
solution space may therefore be needed for a linear relative
Poisson problem. No claim that such a solution is psh is made.

% END source: sections/nonpolar_trace_dirichlet_obstruction.tex


% BEGIN frozen or reviewed source: sections/faithful_coercive_energy.tex
\section{A faithful curvature measure and a coercive relative energy form}
\label{sec:r4-faithful-energy}

Fix the original finite extension $K/\mathbf Q_p$ and its absolute
value. Let $\Omega$ be an ordinary algebraic norm box relatively
compact in the analytification of a smooth integral affine
$K$-variety $Y$ of dimension $n\geq1$. Let $S$ be an arbitrary
symmetric original CLD-positive current of bidegree $(n-1,n-1)$
on $\Omega$, separately closed under $d'$ and $d''$.
Every current and coefficient measure below uses the original
smooth CLD test sheaf. The background need not be strongly
positive, smooth, or supported on a subvariety.

The purpose is to construct an actual reference measure and
close the energy form. This fills the trace and core hypotheses
left conditional in the separate R3 degree-one research note.
It does not identify a Dirichlet-form capacity with raw relative
Bedford--Taylor capacity.

For real original smooth functions set
\[
 \Gamma_S(f,g)=\tfrac12(d'f\wedge d''g+d'g\wedge d''f)\wedge S,
 \qquad \Gamma_S(f)=\Gamma_S(f,f).
\]
These are signed, respectively positive, Radon measures, by the
original positive cone and coefficient-Radon theorem. Mixed
energies are interpreted with this symmetric convention.

\subsection{A countable smooth curvature construction}

\begin{lemma}[A global smooth compensator for a finite moment list]
\label{lem:r4-smooth-reference-compensator}
Let $H\Subset\Omega$ and let $t_i=\log|r_i|$, $1\leq i\leq m$,
be a finite rational unit presentation on a neighborhood of $H$.
There is an original smooth continuous classical psh function
$a$ on $Y^{\rm an}$, normalized so that $0\leq a\leq1$ on
$\Omega$, and a finite $N>0$, such that
\begin{equation}\label{eq:r4-compensator-curvature}
 Da\ \geq\ N^{-1}\sum_i d't_i\wedge d''t_i
       \quad\hbox{near }H,
 \qquad
 d'a\wedge d''a\ \leq\ 2a\,Da\quad\hbox{on }\Omega.
\end{equation}
Both inequalities are inequalities of positive $(1,1)$-forms
on common original moment presentations.
\end{lemma}

\begin{proof}
Write $r_i=b_i/c_i$, where $b_i,c_i$ are regular functions on
the affine variety $Y$ and neither vanishes on a neighborhood
of $H$. Choose $A>0$ so that $\log|b_i|+A$ and
$\log|c_i|+A$ are positive on a slightly smaller neighborhood
of $H$. Fix a nonnegative smooth function $\kappa$ supported
in $(-1,0)$ with integral one, and put
\[
 \zeta(s)=\int_{-\infty}^{s}(s-r)\kappa(r)\,dr,
 \qquad F(s)=\zeta(s)^2.
\]
Then $\zeta$ is smooth, convex and nondecreasing, is zero for
$s\leq-1$, and satisfies $\zeta'=1$ for $s\geq0$.
Consequently $F\geq0$, $F'\geq0$, $F''\geq0$,
\[
 F''=2\quad(s\geq0),\qquad (F')^2\leq2FF''\quad\hbox{everywhere}.
\]
For a regular function $b$, the function $F(\log|b|+A)$ is
extended by zero where $b=0$. It is original smooth: it is a
smooth function of the unit logarithm on the nonvanishing
locus and is identically zero on an ordinary neighborhood
of the zero locus. It is classical psh after every complete
isometric extension, by convex nondecreasing composition on
curves, or directly by its nonnegative one-variable second
derivative on each unit locus. The constant branch handles
zeros and identically zero pullbacks.

Let $g$ be the sum of these expressions for all $b_i,c_i$.
Near $H$, the identity $D\log|b_i|=D\log|c_i|=0$ gives
\[
 Dg=2\sum_i\bigl(d'\log|b_i|\wedge d''\log|b_i|
                      +d'\log|c_i|\wedge d''\log|c_i|\bigr)
       \geq\sum_i d't_i\wedge d''t_i.
\]
The inequality is the elementary square inequality
$(\xi-\eta)^2\leq2\xi^2+2\eta^2$.
On any common presentation, scalar Cauchy--Schwarz and
$(F')^2\leq2FF''$ give
\[
 d'g\wedge d''g\leq2g\,Dg.
\]
This also holds at a zero branch by smoothness; equivalently,
apply Cauchy--Schwarz to the nonzero summands and then take
limits. Since $\overline\Omega$ is compact, $g$ has a finite
bound there. Set $N=\max(1,\sup_{\overline\Omega}g)$ and
$a=g/N$. The two claimed inequalities follow.
\end{proof}

\begin{theorem}[A faithful curvature measure dominating all coefficients]
\label{thm:r4-faithful-curvature-measure}
There are original smooth classical functions $0\leq a_l\leq1$
on $\Omega$ and a continuous classical function
\[
 h=\sum_{l\geq1}2^{-l}a_l,\qquad 0\leq h\leq1,
\]
such that the positive Radon measure
\begin{equation}\label{eq:r4-reference-lambda}
 \lambda=D(hS)=\sum_{l\geq1}2^{-l}D(a_lS)
\end{equation}
has the following properties.
For every original compact smooth $(1,1)$-test $\eta$, there
is a finite $C_\eta$ with
\begin{equation}\label{eq:r4-all-coefficients-lambda}
 |\mu_{S,\eta}|\leq C_\eta\lambda.
\end{equation}
For every Borel set $E\subset\Omega$,
\begin{equation}\label{eq:r4-faithful-lambda}
 \lambda(E)=0\quad\Longleftrightarrow\quad\mathbf1_E S=0.
\end{equation}
Thus $F=\operatorname{supp}\lambda=\operatorname{supp}S$.
The measure is locally finite; its total mass on $\Omega$
is not required to be finite. For every original smooth $g$
of compact support,
\[
 D(gS)=b_g\lambda,\quad b_g\in L^\infty(\lambda),
 \qquad \Gamma_S(g)\leq C_g\lambda,
\]
and $b_g$ has compact support in $F$ up to a null set.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
The countable cofinal moment controls are a named direct
consequence of R and B, as recorded in
Lemma~\ref{lem:r3-countable-trace-controls}. The new construction
makes their dominating measure the curvature of one bounded
continuous potential and retains its exact coefficient-null ideal.

\begin{proof}
Use the countable positive controls from that lemma, in degree
$(1,1)$. They have the form
\[
 \Theta_l=\chi_l\sum_i d't_{li}\wedge d''t_{li},
 \qquad 0\leq\chi_l\leq1,
\]
with compact support on a rational unit presentation.
The cutoffs can be chosen between zero and one from the start.
The exact rational replacement uses B, Lemma D.1, and a
countable dense subfield of the original $K$; it changes no
pointwise moment norm or absolute value.
Apply Lemma~\ref{lem:r4-smooth-reference-compensator} to each
compact support, with constant $N_l$. Positivity and the local
smooth product rule give the global measure inequality
\[
 D(a_lS)\geq N_l^{-1}\mu_{S,\Theta_l}.
\]
Outside the support of $\Theta_l$ its right side is zero, so
no extension of the current is involved.

The original smooth classical functions
\[
 h_m=\sum_{l=1}^m2^{-l}a_l+2^{-m}
\]
decrease uniformly to $h$, with $0\leq h_m\leq1$.
The local relative CLN estimate bounds the masses of
$D(h_mS)$ on each compact stage, independently of $m$.
Their measures are increasing, since
$D(h_mS)=\sum_{l\leq m}2^{-l}D(a_lS)$.
Their locally finite sum equals $D(hS)$ by uniform
convergence against the fixed coefficient measures of $S$.
This proves \eqref{eq:r4-reference-lambda}; it also proves
classicality and continuity of $h$ by the bounded decreasing
limit property.

For each $l$,
$\mu_{S,\Theta_l}\leq2^lN_l\lambda$.
Finite cofinal domination therefore gives
\eqref{eq:r4-all-coefficients-lambda} for every original test,
including tests with additional moment coordinates. This is
not a claim based on one preferred tropical chart.
If $\lambda(E)=0$, that inequality makes all coefficient
restrictions of $S$ to $E$ zero. Conversely, every
$D(a_lS)=Da_l\wedge S$ is a smooth coefficient measure of
$S$ locally. Hence $\mathbf1_E S=0$ makes every term of
the positive series vanish on $E$. This proves
\eqref{eq:r4-faithful-lambda}.
Applying it to ordinary open sets identifies the supports.
Finally $D(gS)=Dg\wedge S$ and
$\Gamma_S(g)=d'g\wedge d''g\wedge S$ are coefficient
measures of compact original tests, so the remaining assertions
follow from \eqref{eq:r4-all-coefficients-lambda}.
\end{proof}

\subsection{Closability, locality, and coercivity}

Let $\mathscr C$ be the restrictions to $F$ of real original
smooth compactly supported functions on $\Omega$. Set
\[
 \mathcal E(f,g)=\int_\Omega\Gamma_S(f,g)
                    =-\int_F f b_g\,d\lambda
 \qquad(f,g\in\mathscr C).
\]
The second equality is compact Stokes. These formulas are
independent of the chosen ambient representatives: if a core
function is zero $\lambda$-almost everywhere, its self-energy
is zero by the second formula, and all mixed energies vanish
by Cauchy--Schwarz.

\begin{theorem}[A regular strongly local coercive form]
\label{thm:r4-coercive-dirichlet-form}
The form $(\mathcal E,\mathscr C)$ is densely defined and
closable on $L^2(F,\lambda)$. Its closure
$(\mathcal E,\mathscr D)$ is a regular strongly local
symmetric Dirichlet form. It satisfies
\begin{equation}\label{eq:r4-picone-coercivity}
 \mathcal E(f,f)\geq
      \int_F\frac{f^2}{2-h}\,d\lambda
      \geq\tfrac12\|f\|_{L^2(\lambda)}^2
           \qquad(f\in\mathscr D).
\end{equation}
In particular $\mathcal E^{1/2}$ itself is a Hilbert norm
on $\mathscr D$, equivalent to the usual form norm.
The construction is valid also when $S=0$, in which case
the support space and the Hilbert space are empty and zero.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
Here regularity and strong locality concern the support space
$F$ with its ordinary subspace topology. No intrinsic metric,
doubling property, or strict psh exhaustion is presumed.

\begin{proof}
The space $F$ is locally compact, metrizable, and second
countable. Original smooth compact functions form a real
algebra and separate its points. Indeed distinct Berkovich
points in the affine analytification give different seminorms
to some regular function. A smooth truncation of its logarithm
distinguishes them, and an original compact cutoff preserves
these two values. The cutoffs also give a nonzero value at
each point. The locally compact Stone--Weierstrass theorem,
followed by multiplication by one cutoff on any prescribed
compact stage, gives uniform density in $C_c(F)$ in the
usual regular-core sense, and hence density in $L^2(F,\lambda)$.

For closability, let $f_j\to0$ in $L^2(\lambda)$ and assume
that $f_j$ is Cauchy for the energy seminorm. It has a limit
in the abstract Hilbert completion of that seminorm. For
every $g\in\mathscr C$,
\[
 \mathcal E(f_j,g)=-\int f_jb_g\,d\lambda\longrightarrow0,
\]
because $b_g$ is bounded and compactly supported, hence belongs
to $L^2(\lambda)$. The limit is orthogonal to the dense core
of that completion and is therefore zero. This is closability.

For every smooth normal contraction $P$ with $P(0)=0$ and
$|P'|\leq1$, the original chain rule gives
$\mathcal E(P(f),P(f))\leq\mathcal E(f,f)$ on the core.
Approximate $0\vee t\wedge1$ by such smooth contractions.
Their images converge in $L^2$ and have bounded form norm;
weak compactness in the closed form space and lower
semicontinuity give the contraction inequality. Approximating
an arbitrary form-domain function by core functions proves
the same inequality on $\mathscr D$. This is the Markov
property. The identical argument supplies any scalar normal
contraction, in particular absolute value and positive part.
Core density in both the form norm and the uniform topology
proves regularity.

We verify strong locality on $F$, rather than assume it from
ambient representatives. For any fixed original compact smooth
$\chi$, the coefficient domination gives
\[
 \mathcal E(\chi f,\chi f)
 \leq 2\|\chi\|_\infty^2\mathcal E(f,f)
                   +2C_\chi\|f\|_{L^2(\lambda)}^2.
\]
Thus multiplication by $\chi$ extends boundedly to
$\mathscr D$. Suppose $f\in\mathscr D$ has compact support
in $F$ and $g\in\mathscr D$ equals a constant $c$ almost
everywhere on a relative ordinary open neighborhood $O$ of
that support. Choose $\chi=1$ near this support and with
$\operatorname{supp}\chi\cap F\subset O$. For core
$f_j\to f$ in form norm, put $k_j=\chi f_j$.
Then $k_j\to f$ in form norm. The measure $D(k_jS)$ is
carried by $O$, has a bounded compactly supported density
against $\lambda$, and has total mass zero by compact Stokes.
The Green identity extends to $g$ by its $L^2$ convergence
from the core. Hence
\[
 \mathcal E(k_j,g)=-\int g\,D(k_jS)
                         =-c\int D(k_jS)=0.
\]
Passing to the limit proves strong locality.

It remains to prove the stated coercivity, which is stronger
than closability. Write $v_m=2-h_m$, so $1\leq v_m\leq2$,
and $\lambda_m=D(h_mS)$. For a core $f$, the positive
square identity for the two covectors $d'f$ and
$(f/v_m)d'v_m$ gives
\[
 \Gamma_S(f)-\Gamma_S(v_m,f^2/v_m)\geq0.
\]
All functions in this identity are original smooth, and
$f^2/v_m$ has compact support. Compact Stokes therefore yields
\[
 \mathcal E(f,f)\geq
       -\int(f^2/v_m)D(v_mS)
       =\int(f^2/v_m)\,d\lambda_m.
\]
The positive measures $\lambda_m$ increase locally to
$\lambda$, and $v_m\to 2-h$ uniformly. On the fixed compact
support of $f$ this proves the first inequality in
\eqref{eq:r4-picone-coercivity}. The second follows from
$2-h\leq2$. Both pass to the closed form domain by form
approximation and $L^2$ convergence, since $1/(2-h)$ is
bounded. Completeness for the energy norm follows from
$\|f\|_2^2\leq2\mathcal E(f,f)$.
\end{proof}

\begin{lemma}[The reference potential belongs locally to the form domain]
\label{lem:r4-reference-local-energy}
For every original compact smooth cutoff $\chi$,
$\chi h_m\to\chi h$ in the closed form norm. In particular
$h$ belongs locally to $\mathscr D$.
\end{lemma}

\begin{proof}
The compensators satisfy
$\Gamma_S(a_l)\leq2D(a_lS)$ on $\Omega$.
For $k>m$, weighted Cauchy--Schwarz consequently gives
\[
 \Gamma_S(h_k-h_m)
 \leq 2\Bigl(\sum_{l=m+1}^k2^{-l}\Bigr)
                  \sum_{l=m+1}^k2^{-l}D(a_lS)
 \leq 2^{1-m}\lambda.
\]
Also $|h_k-h_m|\leq2^{-m}$. The product square estimate,
integrated over the support of $\chi$, gives
\[
 \mathcal E\bigl(\chi(h_k-h_m),\chi(h_k-h_m)\bigr)
 \leq2^{1-2m}\mathcal E(\chi,\chi)
       +2^{2-m}\|\chi\|_\infty^2\lambda(\operatorname{supp}\chi).
\]
The $L^2$ difference tends to zero as well. Thus the sequence
is form Cauchy, and its $L^2$ limit is the stated function.
\end{proof}

\subsection{A linear relative Dirichlet problem on all original tests}

For an ordinary open $W\Subset\Omega$, let
$\mathscr D_0(W)$ be the form-norm closure of the original
smooth compact functions supported in $W$, regarded as zero
off $W$ in $L^2(F,\lambda)$. This definition is the boundary
condition in the next theorem. It is not an assertion of a
pointwise ordinary boundary limit.

\begin{theorem}[Coercive energy solutions of the relative Poisson equation]
\label{thm:r4-linear-energy-dirichlet}
For every real $f\in L^2(W\cap F,\lambda)$, there is a unique
$u\in\mathscr D_0(W)$ such that
\begin{equation}\label{eq:r4-energy-poisson}
 \mathcal E(u,\phi)=-\int_W f\phi\,d\lambda
                         \quad(\phi\in\mathscr D_0(W)).
\end{equation}
Its coefficient-defined current satisfies
\[
 D(uS)=f\lambda\quad\hbox{on }W
\]
against every original compact smooth scalar test. Moreover
\[
 \mathcal E(u,u)^{1/2}\leq\sqrt2\|f\|_{L^2(\lambda)},
 \qquad \|u\|_{L^2(\lambda)}\leq2\|f\|_{L^2(\lambda)}.
\]
The solution depends linearly and continuously on $f$.
If $f\geq0$, then $u\leq0$ almost everywhere.
If $f\in L^\infty(W\cap F,\lambda)$, then
\begin{equation}\label{eq:r4-energy-linfty}
 |u|\leq(2-h)\|f\|_\infty\leq2\|f\|_\infty
                          \quad\lambda\hbox{-almost everywhere on }W.
\end{equation}
Thus positive bounded densities have bounded energy solutions
whose relative curvature is positive. The theorem does not
assert that these solutions are ambient classical psh or
belong to the bounded psh capacity completion.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}}, with a fully
specified energy boundary condition. The functional-analytic
Riesz representation step is standard; the faithful original
coefficient control and coercive barrier needed to apply it
are proved above.

\begin{proof}
The preceding coercivity makes $\mathscr D_0(W)$ a Hilbert
space for $\mathcal E$. The functional
$\phi\mapsto\int_W f\phi\,d\lambda$ has norm at most
$\sqrt2\|f\|_2$. Riesz representation gives a unique
$w\in\mathscr D_0(W)$ with
$\mathcal E(w,\phi)=\int f\phi\,d\lambda$.
Set $u=-w$. Taking $\phi=w$ and using coercivity proves
both quantitative bounds and continuity in the data.

The multiplication $uS$ is well defined on all original
compact complementary tests: their coefficient variations
are bounded by $C_\eta\lambda$, and $u$ is locally
$L^1(\lambda)$ by Cauchy--Schwarz and local finiteness.
An $L^2$ approximation by core functions extends Green's
identity to
\[
 \mathcal E(u,\phi)=-\int u\,D(\phi S)
                  =-\langle D(uS),\phi\rangle
       \quad(\phi\in A_c^{0,0}(W;\mathbf R)).
\]
This proves the full original-current equation, rather than
only its value on a chosen tropical subfamily of tests.

If $f\geq0$, the vector $w$ uniquely minimizes
$\mathcal E(z,z)/2-\int fz\,d\lambda$ on the Hilbert space.
The normal contraction $z\mapsto|z|$ preserves
$\mathscr D_0(W)$ and does not increase energy. Therefore
replacing $w$ by $|w|$ does not increase the objective.
Uniqueness gives $w=|w|$. The same argument shows positivity
of this linear solution operator for every nonnegative datum.

We supply the boundary justification for comparison with
$v=2-h\geq1$. Choose an original smooth $0\leq\chi\leq1$
equal to one near $\overline W$ and compactly supported in
$\Omega$. By Lemma~\ref{lem:r4-reference-local-energy},
$\widetilde v=\chi v\in\mathscr D$ and
$\chi(2-h_m)\to\widetilde v$ in form norm.
For every core test in $W$, Stokes and
$D(vS)=-\lambda$ give
\[
 \mathcal E(\widetilde v,\phi)=\int_W\phi\,d\lambda.
\]
This extends to every $\phi\in\mathscr D_0(W)$, since
$\lambda(W)<\infty$ and both sides are form continuous.

For completeness, if $w\in\mathscr D_0(W)$ and $M\geq0$,
then $z=(w-M\widetilde v)^+$ belongs to
$\mathscr D_0(W)$. Approximate $w$ by original core $w_j$
supported in $W$, and $\widetilde v$ by the nonnegative
smooth functions $\chi(2-h_m)$. The positive part of their
difference is supported inside $\operatorname{supp}w_j$.
Smooth normal contractions that vanish on $(-\infty,0]$
approximate this positive part, with energy bounded by that
of the difference; their supports remain in this same compact
subset of $W$. Closedness and weak compactness in form norm
therefore put the positive part in $\mathscr D_0(W)$.
First let $m\to\infty$ and then $j\to\infty$; the
$L^2$ limits and the same bounded-energy argument prove the
claim for $z$. No pointwise boundary regularity is used here.

Assume first $0\leq f\leq M$. Pair the two equations with
this $z$ to obtain
\[
 \mathcal E(w-M\widetilde v,z)=\int_W(f-M)z\,d\lambda\leq0.
\]
For every real form-domain $r$, the normal contraction
inequality $\mathcal E(|r|)\leq\mathcal E(r)$ implies
$\mathcal E(r^+,r^-)\leq0$. Hence
$\mathcal E(r,r^+)\geq\mathcal E(r^+,r^+)$.
Apply this with $r=w-M\widetilde v$. Coercivity makes
$z=0$, so $0\leq w\leq Mv$ on $W$.
For signed $f$ with $|f|\leq M$, positivity and linearity
of the solution operator applied to $M\pm f$ give
$|w_f|\leq w_M\leq Mv$. This proves
\eqref{eq:r4-energy-linfty}.
\end{proof}

\paragraph{Relation to higher mixed degree and to the remaining gates.}
For any fixed allowed $(p,p)$-background with $q=n-p\geq1$,
one may freeze $q-1$ bounded differentiated slots and apply
this section to their positive closed mixed current of degree
$(n-1,n-1)$. This supplies a linear energy theory for that
fixed mixed background. It does not identify these separate
reference measures uniformly as the slots vary, solve a
nonlinear Monge--Ampere equation, or prove raw relative
capacity outer regularity. The ordinary topology on $F$ is
retained, and no equality on additional $\delta$-tests is
asserted.

% END source: sections/faithful_coercive_energy.tex


% BEGIN frozen or reviewed source: sections/energy_capacity_bridge.tex
\section{A one-way comparison with the regular energy capacity}
\label{sec:r4-energy-capacity-bridge}

Retain the degree-one setting, support space
$F=\operatorname{supp}S$, faithful curvature measure
$\lambda=D(hS)$, and regular closed form
$(\mathcal E,\mathscr D)$ of
Section~\ref{sec:r4-faithful-energy}. In particular
$0\leq h\leq1$, every fixed compact original coefficient measure
is bounded by a constant times $\lambda$, and
$\mathcal E(f,f)\geq\frac12\|f\|_{L^2(\lambda)}^2$.
Put
\[
 \mathcal E_1(f,f)=\mathcal E(f,f)+\|f\|_{L^2(\lambda)}^2.
\]
The relative BT capacity below always uses the original bounded
classical testers on the indicated analytic domain. The energy
capacity instead lives on the ordinary support space $F$.

\subsection{The external capacity input and its hypotheses}

Let $\operatorname{cap}_{\mathcal E,1}$ be the $1$-capacity of
this regular Dirichlet form: for a relative open $G\subset F$,
\[
 \operatorname{cap}_{\mathcal E,1}(G)
 =\inf\{\mathcal E_1(f,f):f\in\mathscr D,\ 
                            f\geq1\ \lambda\text{-a.e. on }G\},
\]
and for arbitrary $A\subset F$ take the infimum over relative
open neighborhoods. We use the standard Choquet property and
the following compact-core formula:
\begin{equation}
 \operatorname{cap}_{\mathcal E,1}(K)
 =\inf\{\mathcal E_1(f,f):f\in\mathscr C,\ f>1\text{ on }K\},
 \qquad K\Subset F.
 \label{eq:r4-energy-compact-core}
\end{equation}
Here $\mathscr C$ is the restriction of the original smooth
compact functions on $\Omega$.

\noindent\textbf{Status of this input: \texttt{EXTERNAL\_KNOWN}.}
The precise primary reference is Fukushima--Sato--Taniguchi
\cite[\S3, pp.~519--520, conditions (C.1), (C.2) and
formula (3.1)]{FST91}. Their space is locally compact separable
metric, their reference measure is positive Radon with full
support, and their regular core is a uniformly dense algebra
with compact cutoffs and smooth contraction stability.
These hypotheses hold here on $F$: regularity and full support
were proved above; original smooth cutoffs restrict to the
required compact cutoffs on $F$; and smooth scalar contractions
vanishing at zero preserve $\mathscr C$. In particular for a
compact $K\subset G\Subset F$ the ambient cutoff may be chosen
with support meeting $F$ only inside $G$.
If the transience hypothesis stated later in that paper is
retained, it also holds here. Indeed, $\sigma$-finiteness of
$\lambda$ gives a bounded strictly positive
$g\in L^1(\lambda)\cap L^2(\lambda)$ with
$\|g\|_2\leq1/\sqrt2$, and coercivity gives
\[
 \int_F |f|g\,d\lambda
 \leq\|g\|_2\|f\|_2\leq\mathcal E(f,f)^{1/2}.
\]
No finiteness of $\lambda(F)$ is needed. For $S=0$ the
statements below have their evident zero-capacity interpretation.

\subsection{A uniform estimate against all normalized testers}

\begin{lemma}[The smooth core controls the BT measures]
\label{lem:r4-energy-core-tester}
Fix an ordinary relatively compact $W\Subset\Omega$.
There is a finite $C_W$ such that every real original smooth
compact function $f$ and every
$w\in\mathcal C_b(\Omega)$ with $0\leq w\leq1$ satisfy
\begin{equation}
 \int_W f^2\,D(wS)
 \leq C_W\|f\|_{L^\infty(\Omega)}
                   \mathcal E_1(f|_F,f|_F)^{1/2}.
 \label{eq:r4-energy-core-tester}
\end{equation}
The same assertion holds with any permitted smaller capacity
domain in place of $\Omega$, with $W$ compactly contained in
that domain. The energy on the right can still be taken on
the original full $\Omega$.
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the R3
fixed-cutoff square-energy estimate and the faithful coefficient
bound. We give the argument to specify its original-test scope.

\begin{proof}
Choose original smooth $0\leq\chi,\kappa\leq1$ with
$\chi=1$ near $\overline W$, $\kappa=1$ near
$\operatorname{supp}\chi$, and compact support in $\Omega$.
One can first work on a smaller norm-box stage on which the
prescribed-domain smooth regularization is available. A finite
sum of such stages treats an arbitrary $\overline W$.
The coefficient bound gives
$\Gamma_S(\chi)\leq C_\chi\lambda$.

First suppose that $w$ is smooth and $0\leq w\leq2$ near
these supports. Compact Stokes gives
\[
 \int\chi f^2D(wS)
 =-2\int\chi f\,\Gamma_S(f,w)
       -\int f^2\,\Gamma_S(\chi,w).
\]
The original positive cone supplies the measure-valued
Cauchy--Schwarz inequality for these energy measures.
Moreover
\[
 \int\kappa\,\Gamma_S(w)
 =\frac12\int w^2\,D\kappa\wedge S
                  -\int\kappa w\,D(wS)
 \leq 2\,|\mu_{S,D\kappa}|(\operatorname{supp}\kappa).
\]
This last bound is independent of $w$. Thus the two terms in
the Stokes formula are bounded in absolute value by a constant
times, respectively,
\[
 \|f\|_\infty\mathcal E(f|_F,f|_F)^{1/2},
 \qquad
 \|f\|_\infty
       \left(\int f^2\Gamma_S(\chi)\right)^{1/2}
 \leq C_\chi^{1/2}\|f\|_\infty\|f\|_{L^2(\lambda)}.
\]
This proves the claim for such smooth $w$.

For a general normalized tester use its smooth decreasing
classical majorants on a neighborhood of the fixed compact
stage. After discarding a prefix they lie in $[0,2]$ there:
their sublevel sets below two are an increasing open cover of
that compact stage. The fixed original test $\chi f^2$ and
bounded convergence against the coefficient measures of $S$
give $D(w_jS)\to D(wS)$ against this test. The preceding
bound therefore passes to $w$. This passage keeps $S$ fixed.
The same cutoffs and proof inside a smaller capacity domain
prove the final assertion; its local energy is at most the
global nonnegative energy on $\Omega$.
\end{proof}

\begin{theorem}[Energy capacity dominates relative BT outer capacity]
\label{thm:r4-energy-BT-capacity}
For every compact $H\Subset\Omega$ there is a finite $C_H$
such that every subset $A\subset H$, whether Borel or not,
satisfies
\begin{equation}
 \operatorname{Cap}_S^*(A;\Omega)
 \leq C_H\,
       \operatorname{cap}_{\mathcal E,1}(A\cap F)^{1/2}.
 \label{eq:r4-energy-BT-bridge}
\end{equation}
In particular every energy-polar subset is locally of zero
relative BT outer capacity. Every normalized measure $D(wS)$
vanishes on such a set, uniformly in the sense of
\eqref{eq:r4-energy-BT-bridge}.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}.}
This is a one-way quantitative comparison. Neither equality
of the two capacities nor the reverse inequality is asserted.

\begin{proof}
Choose $H\subset W\Subset\Omega$ and a fixed cutoff stage as
in Lemma~\ref{lem:r4-energy-core-tester}. First let
$K\subset\overline W\cap F$ be compact. Take
$f\in\mathscr C$ with $f>1$ on $K$. For $\epsilon>0$ use a
smooth nondecreasing $1$-Lipschitz contraction $P_\epsilon$
which fixes $[0,1]$, vanishes at zero, and has image in
$[-\epsilon,1+\epsilon]$. It may be chosen constant outside
a bounded interval. Applying it to an ambient smooth
representative of $f$ gives an original compact smooth
function, still denoted $P_\epsilon(f)$, which is at least
one on $K$. The chain rule gives
\[
 \mathcal E_1(P_\epsilon(f),P_\epsilon(f))
                    \leq\mathcal E_1(f,f).
\]
Use the tester estimate on a slightly larger fixed core
containing $\overline W$. For every normalized $w$ it gives
\[
 D(wS)(K)
 \leq C_H(1+\epsilon)\mathcal E_1(f,f)^{1/2}.
\]
Taking the infimum in
\eqref{eq:r4-energy-compact-core} and then
$\epsilon\downarrow0$ proves
\begin{equation}
 D(wS)(K)\leq
       C_H\operatorname{cap}_{\mathcal E,1}(K)^{1/2},
 \label{eq:r4-energy-compact-measure}
\end{equation}
with the same constant for every such $K$ and every $w$.

Let $G\subset F$ be any relative open set containing $A\cap F$.
Choose an ordinary open $O\subset\Omega$ with $O\cap F=G$.
The measure $D(wS)$ is supported on $F$, since $S$ is zero
on the ordinary open complement of $F$. Its Radon inner
regularity, \eqref{eq:r4-energy-compact-measure}, and energy
capacity monotonicity give
\[
 D(wS)(O\cap W)
  =\sup_{K\Subset O\cap W\cap F}D(wS)(K)
  \leq C_H\operatorname{cap}_{\mathcal E,1}(G)^{1/2}.
\]
This holds for every normalized $w$, with a common constant.
The ordinary open set
$(O\cap W)\cup(\Omega\setminus F)$ contains $A$ and has the
same mass as $O\cap W$ for every one of these measures.
Taking their supremum therefore bounds its raw BT capacity
by the same right side. The definition of BT outer capacity,
followed by the infimum over $G$, proves
\eqref{eq:r4-energy-BT-bridge}. This proof uses inner
regularity only for each fixed Radon measure after a uniform
compact estimate has been established. It does not interchange
an infimum over neighborhoods with a supremum over testers.
\end{proof}

\begin{corollary}[A canonical relative capacitary representative of energy functions]
\label{cor:r4-energy-relative-representative}
Every $u\in\mathscr D$ determines a canonical Borel function
class on $\Omega$ modulo local relative outer-capacity zero:
it is characterized as the local relative capacity limit of
the original smooth core approximations of $u$.
A selected representative $\widetilde u$ agrees with the
$L^2(\lambda)$ representative on $F$
$\lambda$-almost everywhere and can be set to zero off $F$.
It is locally relatively quasicontinuous. Arbitrary changes
on a merely $\lambda$-null set are not asserted to preserve
this canonical capacitary class. For each compact
$H\Subset\Omega$ there is a finite $C_H$ such that
\begin{equation}
 \operatorname{Cap}_S^*
    (H\cap\{|\widetilde u-\widetilde v|>\delta\};\Omega)
 \leq C_H\delta^{-1}\mathcal E_1(u-v,u-v)^{1/2}
 \quad(u,v\in\mathscr D,\ \delta>0).
 \label{eq:r4-energy-convergence-BT}
\end{equation}
The construction is linear in these relative classes and
injective from $\mathscr D$. No psh or completed-psh
membership of $\widetilde u$ is asserted.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the core
estimate and elementary capacitary completion.

\begin{proof}
For a real original compact smooth $f$, choose a smooth
$P:\mathbf R\to[0,1]$ which is zero near zero, is one on
$\{|t|\geq1\}$, and has finite Lipschitz constant $C_P$.
Apply the core estimate to $P(f/\delta)$. Its form norm is
at most $C_P\delta^{-1}\mathcal E_1(f,f)^{1/2}$.
The ordinary open gap
$W\cap\{|f|>\delta\}$ has raw capacity equal to its outer
capacity, because it is itself an admissible open neighborhood.
Consequently
\[
 \operatorname{Cap}_S^*
   (H\cap\{|f|>\delta\};\Omega)
 \leq C_H\delta^{-1}\mathcal E_1(f,f)^{1/2}.
\]
The same proof works on every permitted compact/domain pair
inside $\Omega$, by the last assertion of the core lemma.

Now take original smooth $f_j\to u$ in the form norm.
The displayed inequality makes this sequence capacity Cauchy
on a countably cofinal collection of such pairs. Choose a
subsequence with summable capacities of the ordinary open
successive gaps, using thresholds tending summably to zero.
Outside the union of their tails the sequence converges
uniformly on each compact core. The intersection of these
tails is relatively outer-null. Assign value zero there.
This gives a Borel finite limit and local quasicontinuity;
the open complement of $F$, itself of zero relative capacity,
may be included in every exceptional open set. The same
gap inequality proves convergence of the entire sequence
and independence of the core approximation.

The reference measure itself is one normalized tester:
$\lambda=D(hS)$ with $0\leq h\leq1$. Hence relative capacity
convergence implies local convergence in $\lambda$-measure.
It follows that the constructed function agrees
$\lambda$-almost everywhere with the prescribed $L^2$ limit.
Passing the smooth gap estimate by the level-set triangle
inequality proves \eqref{eq:r4-energy-convergence-BT}.
The same argument applied to sums and scalar multiples proves
linearity of the resulting classes. If that class is zero,
then its $L^2(\lambda)$ representative is zero, so injectivity
follows from the definition of $\mathscr D$ as a subspace of
$L^2(\lambda)$.
\end{proof}

\paragraph{What the comparison leaves open.}
The energy capacity is Choquet by the verified external
regular-form theorem. This fact has not been transferred
backwards to raw relative BT capacity. In particular the
general compact outer-regularity gate for arbitrary $S$
is not closed by this section. Smooth core approximations
are not psh approximations, so the final corollary also
does not turn the energy Poisson solution into a member
of $\mathcal A_{S,b}$ or a classical psh solution.

% END source: sections/energy_capacity_bridge.tex


% BEGIN frozen or reviewed source: sections/energy_polar_splitting.tex
\section{Bounded classical potentials in the energy domain and polar splitting}
\label{sec:r4-energy-polar-splitting}

Use the arbitrary degree-one background $S$, the norm-box domain
$\Omega$, and the faithful reference measure $\lambda=D(hS)$
of Section~\ref{sec:r4-faithful-energy}. The geometric functions
$a_l$ and $h$ in that construction can be fixed independently
of $S$: their moment controls, normalizations, and smooth
compensators depend only on the domain. We make this common
choice when comparing several backgrounds of the same degree.

\begin{proposition}[The closed energy agrees with bounded BT energy]
\label{prop:r4-classical-in-energy}
Every $u\in\mathcal C_b(\Omega)$ belongs locally to the
closed energy domain: $\chi u\in\mathscr D$ for every
original smooth compact function $\chi$. The same holds for
finite differences of such potentials. If $u_j\downarrow u$
are the bounded smooth classical regularizations on a
neighborhood of the support of $\chi$, then
\[
 \chi u_j\longrightarrow\chi u
                  \quad\hbox{in the closed form norm}.
\]
The energy measures furnished by these strong limits coincide
with the previously defined bounded BT measures
$\Gamma_S(u,v)$. Locally these measures are absolutely
continuous with respect to $\lambda$. For every original
compact scalar $\phi$, with $\chi=1$ near its support,
\[
 \mathcal E(\chi u,\phi)=-\int u\,D(\phi S).
\]
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the
proved weighted bounded calculus, its energy identity, and
the faithful closable form. Absolute continuity is proved
by energy convergence, not inferred from weak convergence
of measures.

\begin{proof}
Work first on a compact regularization stage containing
$\operatorname{supp}\chi$. The approximants have one finite
bound there and converge at every actual point. For two
indices $j,k$, put $r_{jk}=u_j-u_k$. The bounded energy identity
reads
\[
 \Gamma_S(r_{jk})
   =\tfrac12D(r_{jk}^2S)-r_{jk}D(r_{jk}S).
\]
For any two increasing index sequences $j_l,k_l\to\infty$,
the first term tends to zero on every fixed compact test by
bounded convergence against the coefficients of $S$.
Expand the second into its four weighted one-potential terms.
The weighted joint decreasing theorem makes each tend to
$uD(uS)$; their signs cancel. Hence the positive measures
$\Gamma_S(r_{j_lk_l})$ tend to zero on compact smooth tests.
If their masses against $\chi^2$ did not tend to zero as
$\min(j,k)\to\infty$, a violating sequence would admit a
subsequence with both indices increasing. Thus
\[
 \int\chi^2\,\Gamma_S(r_{jk})\longrightarrow0.
\]
The smooth product square inequality gives
\[
 \mathcal E(\chi r_{jk},\chi r_{jk})
 \leq2\int\chi^2\Gamma_S(r_{jk})
                         +2\int r_{jk}^2\Gamma_S(\chi)
                         \longrightarrow0.
\]
The last passage uses the fixed finite coefficient measure
$\Gamma_S(\chi)\leq C_\chi\lambda$ and bounded convergence.
The $L^2(\lambda)$ limit is $\chi u$. This proves strong
form convergence and local membership. Finite compact
localization and a partition give the statement for arbitrary
$\chi$ when more than one regularization stage is needed.

On a fixed smaller compact core the energies of $u_j$ are
bounded by the local CLN energy estimate. The measure
Cauchy--Schwarz inequality then gives
\[
 \|\Gamma_S(u_j)-\Gamma_S(u_k)\|_{\rm TV,H}
 \leq
 \left(\int_H\Gamma_S(u_j-u_k)\right)^{1/2}
 \left(\left(\int_H\Gamma_S(u_j)\right)^{1/2}
       +\left(\int_H\Gamma_S(u_k)\right)^{1/2}\right)
 \longrightarrow0,
\]
where a larger compact smooth cutoff may replace the
indicators on the right. Each smooth energy measure is
absolutely continuous with respect to $\lambda$ by the
all-original coefficient bound. Its total-variation limit
is therefore also absolutely continuous. The distributional
energy identity and weighted decreasing calculus identify
this limit with the existing $\Gamma_S(u)$. Polarization
gives the same conclusion for mixed energies and finite
differences. The smooth Green and product identities pass
through the same strong limits. In particular they give
the last displayed formula of the proposition.
\end{proof}

The proposition supplies the local product and chain identities
used next. For example, for a smooth compact $f$ and bounded
classical $a$, the closed energy of $fa$ is the integral of
\[
 a^2\Gamma_S(f)+2fa\Gamma_S(f,a)+f^2\Gamma_S(a).
\]
This follows by the preceding smooth approximations and
measure-valued Cauchy--Schwarz.

\begin{theorem}[Orthogonal energy splitting across a complete pole set]
\label{thm:r4-energy-pole-splitting}
Let $\phi\leq0$ be an extended classical psh function on
$\Omega$, and put $P=\{\phi=-\infty\}$. Suppose the
coefficient restrictions
\[
 S_P=\mathbf1_PS,\qquad S_{P^c}=\mathbf1_{\Omega\setminus P}S
\]
are separately closed, as supplied by the proved pole
restriction theorem. Use the same geometric $h$ for all
three backgrounds. Then
\[
 \lambda_P=D(hS_P)=\mathbf1_P\lambda,
 \qquad
 \lambda_{P^c}=D(hS_{P^c})=\mathbf1_{P^c}\lambda.
\]
Multiplication by $\mathbf1_P$ and $\mathbf1_{P^c}$ are
orthogonal projections for both the $L^2$ and the energy
inner products on $\mathscr D$. More precisely the closed
forms have the isometric direct-sum decomposition
\begin{equation}\label{eq:r4-energy-orthogonal-sum}
 (\mathcal E_S,\mathscr D_S)
   =(\mathcal E_{S_P},\mathscr D_{S_P})
                   \oplus(\mathcal E_{S_{P^c}},\mathscr D_{S_{P^c}})
\end{equation}
under the Borel carrier identification of the two $L^2$
spaces with $L^2(F,\lambda)$. The support spaces of the two
components may intersect topologically; disjoint topological
supports are not assumed. The decomposition also holds for
each part space $\mathscr D_0(W)$, $W\Subset\Omega$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the
closed-form splitting. The closedness of the current
restrictions is a previously proved input, not a new
extension theorem in this section.

\begin{proof}
For every smooth geometric compensator $a_l$,
\[
 D(a_lS_P)=Da_l\wedge S_P
                  =\mathbf1_P(Da_l\wedge S).
\]
The positive summable curvature construction therefore gives
the asserted identities for the reference measures, and
similarly for $P^c$.

Set
\[
 b_N=\max\{0,1+\phi/N\},\qquad N\geq1.
\]
These are actual bounded classical functions, with
$0\leq b_N\leq1$, zero on $P$, and increasing to one on
$P^c$. All their bounded products over $S_P$ vanish:
$b_NS_P=b_N^2S_P=0$ and hence
$D(b_NS_P)=\Gamma_{S_P}(b_N)=0$.
For any nonnegative original compact smooth $\chi$, the
bounded energy identity over $S_{P^c}$ gives
\[
 0\leq\int\chi\Gamma_{S_{P^c}}(b_N)
   =\tfrac12\int b_N^2D\chi\wedge S_{P^c}
            -\int\chi b_N D(b_NS_{P^c})
   \leq\tfrac12\int b_N^2D\chi\wedge S_{P^c}.
\]
The last measures before multiplication by $b_N^2$ are
fixed coefficient measures and give zero mass to $P$.
Bounded convergence and separate closedness make the last
expression tend to
$\tfrac12\langle S_{P^c},D\chi\rangle=0$.
Consequently
\begin{equation}\label{eq:r4-pole-selector-zero-energy}
 \int\chi\Gamma_S(b_N)\longrightarrow0.
\end{equation}
No continuity of an increasing mixed Monge--Ampere sequence
has been used in this deduction.

For an original smooth compact $f$, the preceding proposition
makes $fb_N$ a vector of $\mathscr D$.
The product square inequality and
$\Gamma_S(b_N-b_M)\leq2\Gamma_S(b_N)+2\Gamma_S(b_M)$
show that these vectors are form Cauchy. Indeed, their
derivative terms tend to zero by
\eqref{eq:r4-pole-selector-zero-energy}, and their terms
against the fixed $\Gamma_S(f)$ tend to zero by bounded
convergence. Their $L^2$ limit is $f\mathbf1_{P^c}$.
Thus both $f\mathbf1_{P^c}$ and $f\mathbf1_P$ belong to
$\mathscr D$.

The product identity and Cauchy--Schwarz, again using
\eqref{eq:r4-pole-selector-zero-energy}, give
\[
 \mathcal E_S(f\mathbf1_{P^c},f\mathbf1_{P^c})
       =\int_{P^c}\Gamma_S(f)=\mathcal E_{S_{P^c}}(f,f),
\]
and the analogous equality on $P$. For two smooth compact
functions $f,g$, approximate the two carrier products by
$f(1-b_N)$ and $gb_N$. In their mixed energy, every term
involving $\Gamma_S(b_N)$ or a mixed derivative with $b_N$
tends to zero by the same estimate and Cauchy--Schwarz.
The remaining term is
$\int b_N(1-b_N)\Gamma_S(f,g)$, which tends to zero
by bounded convergence against a fixed coefficient measure.
Hence the two carrier products are energy orthogonal.

For smooth core functions, multiplication by either indicator
is therefore a contraction in the form norm. It extends
continuously to all of $\mathscr D$, and its $L^2$ action
is still multiplication by that indicator. Their sum is
the identity and they are orthogonal projections.
The range for $P$ is the form closure of the vectors
$f\mathbf1_P$ for original core $f$. The displayed energy
and reference-measure identities identify that closure
isometrically with the core closure defining
$\mathscr D_{S_P}$; the other range is identical with the
$P^c$ form domain. This proves
\eqref{eq:r4-energy-orthogonal-sum}.

If $f$ has compact support in $W$, the approximating products
$fb_N$ have that same compact support. Their smooth
regularizations multiplied by $f$ are original core functions
supported in $W$. Thus the whole argument applies to the
part-form closures, proving the final assertion.
\end{proof}

\begin{corollary}[Canonical polar and nonpolar energy components]
\label{cor:r4-canonical-energy-polar-components}
On a norm-box exhaustion domain where the proved canonical
decomposition $S=S_{\rm pp}+S_{\rm np}$ is represented by
one complete pole set, the energy form and its zero-boundary
linear Poisson problem split orthogonally into those two
components. The splitting is independent of the chosen pole
witness. In particular the energy solution for
$f\in L^2(\lambda)$ is the direct sum of the solutions
for the restricted data $f\lambda_{\rm pp}$ and
$f\lambda_{\rm np}$.
\end{corollary}

\begin{proof}
Apply the theorem to the complete-pole representative of
the canonical current decomposition. Current uniqueness
identifies the two components independently of its witness.
The part-space orthogonal sum identifies the two Riesz
equations componentwise, so uniqueness of the energy
solutions proves the last assertion.
\end{proof}

\paragraph{A boundary of this conclusion.}
The energy splitting does not prove additivity of raw
relative BT capacities for $S_{\rm pp}+S_{\rm np}$.
The vectors $\mathbf1_Pf$ belong to a linear energy domain;
they have not been shown to admit global bounded classical
psh defining rows. In particular, replacing the ambient
psh tester class by all positive relative-energy-curvature
functions would change the capacity in the research problem.
The distinction is visible in the smooth nonpolar Dirichlet
counterexample and its explicit energy solution.

% END source: sections/energy_polar_splitting.tex


% BEGIN frozen or reviewed source: sections/radial_energy_solution.tex
\section{An explicit solution beyond the ambient psh constraint}
\label{sec:r4-radial-energy-solution}

The smooth nonpolar counterexample has a positive replacement:
its equation admits a bounded original smooth solution with
the requested ordinary boundary values when the ambient psh
constraint is removed. This also tests the energy construction
on all original currents, not only on toric scalar tests.

Retain $V=t^{-1}(B)$, $B=\{|x|<1\}\subset\mathbf R^n$,
$n\geq2$, $\beta=\sum_i d't_i\wedge d''t_i$, and
$S=[\beta^{n-1}]$ from
Section~\ref{sec:r4-nonpolar-trace-obstruction}.
Choose $0<r_0<1$ and a smooth nonnegative radial function
$f(x)=\varphi(|x|^2)$ supported in $|x|<r_0$, with
$0\leq f\leq1$ and $f\not\equiv0$. Write $f(r)$ for
its radial value. Set
\begin{equation}\label{eq:r4-explicit-radial-potential}
 A(r)=n\int_0^r s^{n-1}f(s)\,ds,\qquad
 U(r)=-\int_r^1\tau^{1-n}A(\tau)\,d\tau,
 \qquad u=U(|t|).
\end{equation}

\begin{proposition}[An ordinary-boundary solution in the energy space]
\label{prop:r4-explicit-nonpsh-energy-solution}
The function $u$ is bounded and original smooth on a
neighborhood of $\overline V$, is nonpositive on $V$, and
has ordinary boundary value zero. It satisfies
\[
 D(uS)=f(t)\beta^n
\]
against all original compact smooth scalar tests on $V$.
It is not classical psh. In fact, the equation has no solution
with zero ordinary boundary limit in the global bounded
classical capacity closure $\mathscr G_{S,b}(V)$.

Choose a larger ordinary algebraic norm box $\Omega$ containing
$\overline V$, and construct $\lambda=D(hS)$ there by
Theorem~\ref{thm:r4-faithful-curvature-measure}. Then
$u\in\mathscr D_0(V)$ and is the unique energy solution of
this equation. Its datum has a bounded density against
$\lambda$.
\end{proposition}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the
explicit non-psh energy solution and the comparison of the
two solution classes. The nonsolvability in the bounded
classical closure is the previously proved counterexample.

\begin{proof}
The integrand in \eqref{eq:r4-explicit-radial-potential}
is $O(\tau)$ as $\tau\downarrow0$, so $U$ is finite.
More precisely,
\[
 U'(r)=r^{1-n}A(r),\qquad
 U''(r)+\frac{n-1}{r}U'(r)=nf(r)\quad(r>0).
\]
Since $f(r)=\varphi(r^2)$ is smooth and even near zero,
$r^{1-n}A(r)$ is $r$ times a smooth function of $r^2$.
Thus $x\mapsto U(|x|)$ is smooth also at zero. The same
formula extends smoothly beyond $r=1$, because $f$ vanishes
near that sphere. It gives $U(1)=0$ and $U\leq0$ in $B$.
Properness of the torus moment map over compact real sets
and continuity prove the ordinary boundary assertion for $u$.

As an identity of original forms,
\[
 D\bigl(U(|t|)\bigr)\wedge\beta^{n-1}
           =\frac{\Delta_xU(|t|)}n\beta^n
           =f(t)\beta^n.
\]
The off-diagonal Hessian terms vanish upon wedging with
$\beta^{n-1}$, and each diagonal term contributes the same
factor $1/n$. This identity of forms proves the current
equation on the full original test sheaf.

Let $c=A(r_0)>0$. For $r_0<r<1$,
\[
 U'(r)=c r^{1-n},\qquad U''(r)=-(n-1)c r^{-n}<0.
\]
At $(r,0,\ldots,0)$ this is the second derivative in the
first coordinate. The full-annulus trace argument in
Lemma~\ref{lem:r4-trace-calibration} would force this
restriction to be convex for a classical psh function.
Thus $u$ is not classical psh. The same datum is nonzero,
compactly supported and bounded by the smooth matching
subsolution measure $D(\tfrac12(|t|^2-1)S)$, so
Theorem~\ref{thm:r4-nonpolar-completed-subsolution-failure}
excludes every solution in $\mathscr G_{S,b}(V)$ with
these boundary values.

We finally verify the asserted energy boundary condition,
which does not follow merely from a pointwise boundary
limit in an arbitrary domain. In this paragraph the Hilbert-space
vector denoted by $u$ means its zero extension from $V$ to
$\Omega$, rather than its smooth continuation beyond the boundary.
Choose smooth radial cutoffs
$\chi_\epsilon(t)$ which equal one for
$|t|\leq1-2\epsilon$, vanish for $|t|\geq1-\epsilon$,
take values in $[0,1]$, and have derivative bounded by
$C/\epsilon$. Then $u_\epsilon=\chi_\epsilon u$ is an
original smooth compact function on $V$. The original
torus calibration gives, for every smooth function $G(t)$,
\[
 \Gamma_S(G(t))=\frac{|\nabla G(t)|^2}{n}\beta^n
                  =(n-1)!\,\sigma_*(|\nabla G|^2dx).
\]
Near $|t|=1$, the already displayed formulas give
$|u|\leq C(1-|t|)$ and $|\nabla u|\leq C$.
The gradients of $u-u_\epsilon$ are consequently uniformly
bounded on their collar of real volume $O(\epsilon)$.
Applying this estimate first to two compact core functions gives
\[
 \mathcal E(u_\epsilon-u_\delta,u_\epsilon-u_\delta)
                         \leq C\max(\epsilon,\delta).
\]
Thus the compact core sequence is energy Cauchy.
Its $L^2(\lambda)$ limit is $u$: $\lambda$ is finite on
the compact $\overline V$, and bounded pointwise convergence
applies to the zero extensions. Hence $u\in\mathscr D_0(V)$;
passing $\delta\downarrow0$ in the displayed inequality also gives
$\mathcal E(u-u_\epsilon,u-u_\epsilon)=O(\epsilon)$.

Take an original compact smooth scalar equal to one near
$\overline V$. Applying all-coefficient domination to its
product with $\beta$ gives $\beta^n|_V\leq C\lambda$.
The datum $f(t)\beta^n$ therefore has a bounded, nonnegative
$\lambda$-density, and is in particular an $L^2$ datum.
The current equation just proved, the core Green identity,
and form density identify $u$ with the unique solution in
Theorem~\ref{thm:r4-linear-energy-dirichlet}.
\end{proof}

This example isolates the analytic issue. Positivity of
$D(uS)$ for a fixed $(n-1,n-1)$-background is a linear
subharmonicity condition relative to $S$; it need not impose
ambient plurisubharmonicity. The energy space can solve that
linear problem, while the ambient psh constraint may rule
out the same boundary datum. No conclusion about the entire
locally defined sheaf $\mathcal A_{S,b}$ follows from the
global defining-row obstruction.

% END source: sections/radial_energy_solution.tex


% BEGIN frozen or reviewed source: sections/round4_status.tex
\section{Fourth-round correspondence and current proof boundaries}
\label{sec:r4-status}

The original theorem-by-theorem literature comparison remains
Section~\ref{sec:comparison}. Section~\ref{sec:r3-status} and the
frozen R3 proof map are the baseline for this continuation.
This section records changes in mathematical status; it does
not count the accepted results of R, B, A2, or R1--R3 as new
proofs. A2's geometric adelic nefness theorem remains a completed
input under its stated global hypotheses.

Throughout, $K/\mathbf Q_p$ is the prescribed finite extension
with its original absolute value. Unless a theorem explicitly
specializes further, $S$ is a fixed symmetric original-positive
$(p,p)$-current, separately closed under $d'$ and $d''$, and
all identities concern the full original compact smooth CLD
test sheaf. No wedge of two arbitrary positive closed currents,
unrestricted varying-background weighted continuity, or equality
on every GK $\delta$-test is asserted.

\subsection{The two degree ranges must be distinguished}

The pole-carrier capacity theorem, simultaneous tester pasting,
completed strong-image results and constrained psh variational
theorems apply to every remaining degree $q=n-p\geq1$, with the
individual domain, bound and collar hypotheses stated in their
sources. The new faithful Hilbert energy form and linear Poisson
theory instead concern an arbitrary $(n-1,n-1)$-background,
that is, remaining degree one. They can be applied after freezing
$q-1$ bounded differentiated slots of a higher-degree problem.
This gives a linear theory for each such fixed mixed background;
it is not a uniform nonlinear theory while those slots vary.

\subsection{All eleven Bedford--Taylor directions}

\begingroup
\setlength{\tabcolsep}{2pt}
\begin{longtable}{@{}>{\raggedright\arraybackslash}p{.18\textwidth}>{\raggedright\arraybackslash}p{.28\textwidth}>{\raggedright\arraybackslash}p{.34\textwidth}>{\raggedright\arraybackslash}p{.16\textwidth}@{}}
\hline
Direction & Frozen result through R3 & This-round advance & Remaining status\\
\hline\endfirsthead
\hline
Direction & Frozen result through R3 & This-round advance & Remaining status\\
\hline\endhead
1. Locally bounded psh &
B's actual classical class, the ambient representative theorem,
and the arbitrary-relative representative counterexample were
already established. &
Bounded classical potentials now embed locally in the faithful
closed energy form, with strong form convergence of their smooth
decreasing rows. General form vectors have a canonical relatively
quasicontinuous class, specified by their core approximations.
See Proposition~\ref{prop:r4-classical-in-energy} and
Corollary~\ref{cor:r4-energy-relative-representative}. &
Neither energy membership nor relative completion supplies a
general ambient classical representative. The unrestricted
representative assertion remains \texttt{FALSE}.\\[4pt]

2. Positive closed currents and mixed products &
The arbitrary-fixed-$S$ bounded and completed BT calculi, and
finite-coefficient-mass zero extension across closed locally
complete pole sets, were already proved. &
Compatible original piecewise tests admit uniform approximation.
Finite etale trace, disc/polydisc projection and arbitrary
same-$K$ analytic morphisms now preserve the bounded potential
$D$-kernel. Exact test spreading further proves kernel transfer
to finite extensions of finite-polyradius Gauss fields;
see Theorem~\ref{thm:r4-gauss-extension-kernel} and
Corollary~\ref{cor:r4-finite-over-gauss-kernel}. &
The quotient for every original positive closed background
remains \texttt{OPEN}. These are assertions about specified
potentials, not general current pullbacks.\\[4pt]

3. CLN inequalities &
Finite-test relative CLN was already a
\texttt{DIRECT\_COROLLARY} of R/B. R3 added the all-degree
energy-to-capacity estimate. &
New wall-normal mass and normal-factor estimates have orders
$O(\epsilon)$ and $O(\epsilon^{1/2})$. In degree one, a bounded
continuous psh curvature measure dominates every original
compact coefficient of $S$; see
Theorem~\ref{thm:r4-faithful-curvature-measure}. &
Constants retain finite local controls and the stated geometry.
No global strictly psh exhaustion is presumed.\\[4pt]

4. Symmetry and weighted monotone convergence &
The requested joint decreasing theorem, completed capacity
continuity and localized strong psh-dual convergence were already
complete. &
Theorem~\ref{thm:r4-piecewise-test-continuity} adds weak continuity
on tangentially compatible original piecewise tests for positive
closed currents. Completed collar energy inequalities and strong
image completeness are direct consequences of the frozen calculus. &
Arbitrary piecewise jumps fail. Unrestricted varying-background
weighted stability and preservation of nonlinear weak limits by
convex blocks remain \texttt{FALSE}.\\[4pt]

5. Locality and maxima &
Ordinary locality, Borel maximum locality and the relative
maximum/contact measure inequality were already proved. &
Small global psh compensators make one maximum branch strictly
active on each separated compact piece. The same construction
works outside controlled capacity-error sets; see
Theorem~\ref{thm:r4-close-tester-pasting} and
Proposition~\ref{prop:r4-capacity-close-pasting}. &
An arbitrary strict gap is not ordinarily open. The true maximum
measure inequality does not supply the false unrestricted
relative obstacle-contact formula.\\[4pt]

6. Bedford--Taylor capacities &
Ambient core Choquet theory, degree-one nonpolar compact
regularity, the general degree-one completed compact formula,
separated analytic-carrier additivity and positive tail
approximation were already proved. &
Theorem~\ref{thm:r4-pole-carrier-additivity} replaces analytic
carriers by complete pole carriers with mutual relative nullness,
in every degree. Close-tester pasting repairs an actual toric
subfamily defect. The energy capacity controls relative BT outer
capacity in one direction. &
General sharp compact control C1 and general Borel outer
regularity C2 remain \texttt{OPEN}. The energy capacity is not
identified with raw BT capacity.\\[4pt]

7. Quasicontinuity and capacity convergence &
Relative bounded quasicontinuity, monotone capacity convergence,
and equivalence of common-collar energy and strong data topologies
were established. &
Form-norm convergence now gives local relative outer-capacity
convergence for canonical energy representatives. The completed
image theorems retain one global defining bound and an actual
collar or exterior under their limits. &
Energy representatives need not be psh. The whole normalized psh
family is still not capacity precompact.\\[4pt]

8. Pluripolar and negligible sets &
Qualified ambient polar/negligible/capacity-zero equivalences,
pole restriction, and canonical current polar decomposition were
available. &
Theorem~\ref{thm:r4-energy-pole-splitting} gives an orthogonal
closed-form decomposition across a complete pole set, even when
the two topological supports overlap. Energy-polar sets are locally
relative BT outer-null. &
Relative BT measures may still charge ambient polar sets.
Energy splitting does not imply raw capacity additivity for an
arbitrary polar/nonpolar decomposition.\\[4pt]

9. Comparison and domination &
Fixed-$S$ compact-gap comparison and relative q.e. domination,
plus the qualified ambient Dini stability theory, were proved. &
The constrained variational solutions obey the exact monotonicity
bound
$I_S(a_\nu,a_{\nu'})\leq4B\,d_{\rm psh}(\nu,\nu')$.
The completed interior image has relative-q.e. uniqueness with
the prescribed exterior and barriers. &
No general relative pointwise uniqueness is deduced. Ordinary
weak data remain insufficient even under uniform power-capacity
domination.\\[4pt]

10. Relative extremal functions &
Ambient obstacle orthogonality, extremal mass formulas and
relative capacitary envelopes were retained; unrestricted relative
contact formulas had explicit counterexamples. &
Theorems~\ref{thm:r4-variational-maximizer}
and~\ref{thm:r4-variational-retraction} give a unique constrained
energy maximizer and a continuous retraction onto its actual
Monge--Ampere image, for capacity-continuous datum pairings. &
The Euler output is an inequality. A pointwise obstacle gap
does not justify arbitrary two-sided psh variations or
unrestricted orthogonality.\\[4pt]

11. Monge--Ampere and Dirichlet problems &
The qualified ambient bounded-subsolution and Dini-capacity
existence theorems, their boundary extensions, and finite-carrier
completed gluing were established. Arbitrary-relative classical
existence and its TV-closed image were already false. &
Specified completed collar/interior images are strong-metric
complete and closed. A new smooth nonpolar degree-one example
rules out bounded-subsolution existence in the global bounded-row
psh completion. The coercive linear energy space does solve its
relative Poisson problem, and an explicit radial solution verifies
the distinction. &
General completed-psh solvability remains \texttt{OPEN};
its unrestricted global-row subsolution version is
\texttt{FALSE}. Broader ordinary boundary data still need
proved hypotheses.\\
\hline
\end{longtable}
\endgroup

\subsection{What is new, and what is a direct consequence}

\paragraph{The original projected-test obstruction has been reduced further.}
The new wall estimates, compatible piecewise approximation,
split-moment projection formula, and finite etale original trace
are \texttt{NEW\_PROVED}. Finite etale preservation of the bounded
kernel is their \texttt{DIRECT\_COROLLARY}. Rational localization,
finite etale splitting away from a discriminant, and removal of
that fixed analytic exceptional locus prove the full ordinary-disc
kernel theorem in its smooth affine algebraic-neighborhood scope.
The polydisc version is a direct consequence. Thus the earlier
conditional projected-test criterion is now discharged in that
scope, not for every imaginable geometric pushforward.
The separately proved same-$K$ analytic-morphism theorem is
also \texttt{NEW\_PROVED}: an actual analytic translation and
a compact zero-tube argument reduce its graph to the established
zero-section trace. Product preservation and the analytic-immersion
background corollary are direct consequences. This enlarges the
proved functorial scope without upgrading the arbitrary-background
quotient or, by itself, settling arbitrary scalar extension.

\paragraph{Which scalar extensions are now covered.}
Finite $K'/K$ and analytic morphisms over such finite extensions
are covered by Corollary~\ref{cor:r4-finite-extension-kernel},
a \texttt{DIRECT\_COROLLARY} with the inherited absolute value
and the verified finite-constant calibration.
Lemma~\ref{lem:r4-exact-fiber-test-spreading} is
\texttt{NEW\_PROVED}: for any point $a$ of a smooth affine
algebraic parameter variety, every original compact
$\mathscr H(a)$-test on a scalar extension of an original-$K$
space is the exact restriction of a compact original-$K$ test
on an ordinary parameter neighborhood. It uses neither
regularization nor the bounded BT calculus over $\mathscr H(a)$.

The full Gauss-fiber current formula is
\texttt{EXTERNAL\_KNOWN}, specifically CLD25
Proposition~18.6.1 and Corollary~18.6.2, printed pp.~532--534
\cite[18.6.1--18.6.2]{CLD25}.
Its fixed Borel-weighted form is a direct Radon-measure consequence.
Together with exact spreading and the proved original-$K$
ambient-product quotient, it gives \texttt{DIRECT\_COROLLARY}
kernel transfer to
\[
 K_{\mathbf r}=\mathscr H(\zeta_{\mathbf r}),
 \qquad
 L/K_{\mathbf r}\text{ finite},
\]
for every finite positive real polyradius $\mathbf r$.
The finite-over-Gauss proof returns to a parameter space over
the original $K$ and retains the one factor $[L:K_{\mathbf r}]$
when comparing the two calibrations of the same fiber.
All uses of R/B and bounded BT products remain over the
original finite field. These statements do not cover arbitrary
complete fields or every analytic morphism defined over
$K_{\mathbf r}$ or $L$.

Lemma~\ref{lem:r4-finite-coefficient-test-descent} is a further
\texttt{NEW\_PROVED} finite-test statement: finitely many original
tests over any complete extension $L/K$ descend exactly to
$E=\overline{K(c_1,\ldots,c_m)}\simeq\mathscr H(a)$ for one
finite-dimensional algebraic parameter point. The scalar
integration calibration has factor one
\cite[Proposition 15.3.1, p.~425]{CLD25}; its fixed Borel-weighted
version is a direct Radon-measure consequence. These facts prove
the precise \texttt{CONDITIONAL} reduction in
Theorem~\ref{thm:r4-finite-parameter-kernel-reduction}: scalar
kernel transfer for every such $\mathscr H(a)$ is equivalent to
scalar kernel transfer for every complete extension. The premise
for general non-Gauss parameter points remains unproved. The
reduction does not add extension-field analytic-map functoriality
or settle the arbitrary-background quotient.

\paragraph{A faithful analytic energy object, with a different solution class.}
Theorem~\ref{thm:r4-faithful-curvature-measure} constructs
$\lambda=D(hS)$, $0\leq h\leq1$, with the same Borel coefficient-null
ideal as $S$. The new closed form is regular, strongly local and
coercive:
\[
 \mathcal E(f,f)\geq
       \int\frac{f^2}{2-h}\,d\lambda
       \geq\tfrac12\|f\|_{L^2(\lambda)}^2.
\]
This proves a full-original-test linear equation
$D(uS)=f\lambda$ on the specified part domain, with a form
boundary condition. An ordinary pointwise boundary limit is
verified separately for the explicit radial example; it is not
assumed for every energy solution. The bounded-classical embedding
is a \texttt{DIRECT\_COROLLARY}; the closed-form pole splitting is
\texttt{NEW\_PROVED}.

The new quantitative comparison
\[
 \operatorname{Cap}_S^*(A;\Omega)
 \leq C_H\,
       \operatorname{cap}_{\mathcal E,1}
          (A\cap\operatorname{supp}S)^{1/2},
 \qquad A\subset H\Subset\Omega,
\]
uses the external regular-form compact-core formula with its
verified hypotheses. That formula and the energy capacity's
Choquet property are \texttt{EXTERNAL\_KNOWN}, not new BT results.
The resulting canonical energy representative is fixed by smooth
core limits. Merely changing values on a $\lambda$-null set need
not preserve that capacitary class. The coefficient-null ideal,
the energy-capacity-null ideal and the BT outer-capacity-null
ideal are not identified. In particular the proved local
absolute continuity of $\Gamma_S(u,v)$ does not assert
$D(uS)\ll\lambda$. The pole projections act on the $L^2$ form
domain; their existence for a complete pole set is not a theorem
about arbitrary Borel indicators.

\paragraph{All-degree capacity and constrained variation.}
General complete-pole-carrier additivity and close normalized
tester pasting are \texttt{NEW\_PROVED}; outer additivity,
positive-tail extensions and the capacity-error pasting
corollaries are direct consequences of their stated mechanisms.
The new constrained variational theorem assumes a nonempty
capacity-closed convex subset of one globally bounded completed
collar class and datum pairings continuous in outer capacity.
It proves the exact Euler inequality and a retraction onto
the actual solution image. It does not turn every admissible
datum into a Monge--Ampere datum.

The collar and interior strong-image closure theorems are
\texttt{DIRECT\_COROLLARY} of R3 energy and R2 completeness.
For the interior theorem, selected representatives share an
actual exterior including $\partial V$. Thus $a-b=0$ there,
and
\[
 I_S(a,b)=\int_K(a-b)\,d(\nu_b-\nu_a)
\]
even when the full measures have different boundary masses.
Continuous upper and lower barriers retain the ordinary boundary
value; a cutoff supported strictly inside $V$ gives the final
strong convergence of interior data.

\subsection{False assertions and excluded proof routes}

\begin{enumerate}
\item \textbf{Unrestricted bounded-subsolution existence in the
global bounded-row completion is false.}
Theorem~\ref{thm:r4-nonpolar-completed-subsolution-failure}
uses a smooth nonpolar $(n-1,n-1)$-background, $n\geq2$,
nonatomic compact data and a continuous matching subsolution.
Its scope is the completion by rows defined on the whole full
torus-preimage domain with one bound. It does not establish
nonexistence in the entire locally defined sheaf
$\mathcal A_{S,b}$.

\item \textbf{Arbitrary bounded piecewise tests are insufficient.}
Example~\ref{ex:r4-incompatible-wall-test} has a tangential jump
whose pairings fail to converge along uniformly convergent
continuous potentials. Compatibility on moment faces is an
essential hypothesis of the new test theorem.

\item \textbf{A full Gauss fiber is not its independent Gauss section.}
In Example~\ref{ex:r4-correlated-gauss-fiber}, an original test
depending on $x-z$ has positive full-fiber pairing while vanishing
on every independent radius-one Gauss lift. Correlated points
with $|x-z|<1$ must be retained. The shortcut is
\texttt{FALSE}; the external full-fiber identity itself is not
in question.

\item \textbf{Separation cost cannot be ignored.}
The annulus calculation in
Section~\ref{sec:r4-relative-capacity-next} forces the bounded
compensator size to grow at least as
$\epsilon^{-1}-1$ for the displayed nearby compact pieces.
Pasting near-maximizers with moving geometry therefore requires
control of the product $A_K\delta$.

\item \textbf{Regular energy capacity does not settle C2.}
Even a weakly compact, uniformly finite-energy restricted
family can have a raw/outer defect. The actual vertical-fiber
subfamily in Example~\ref{ex:r4-vertical-fiber-repair} has
values $1/2$ and $1$ on the displayed Borel set. The full BT
tester class repairs this particular defect by the new pasting
theorem. It is not a counterexample to full BT outer regularity.

\item \textbf{Energy selectors are not psh selectors.}
The orthogonal projections $\mathbf1_P$ and $\mathbf1_{P^c}$
act on a linear form domain. This does not provide both selectors
in a cone of global bounded classical psh limits, or prove raw
capacity additivity for a nested polar/nonpolar pair.
Likewise positivity of $D(uS)$ need not make $u$ ambient psh;
Proposition~\ref{prop:r4-explicit-nonpsh-energy-solution}
gives a smooth explicit example.

\item \textbf{The frozen weak-limit and quotient warnings remain.}
Convex blocks need not preserve nonlinear weak Monge--Ampere
limits; ordinary weak data need not give uniform or capacity
convergence of solutions; actual-potential faithfulness does not
identify the distributional $D$-kernel. Same-field kernel
equalities are not silently extended to new tests after arbitrary
complete scalar extensions.
\end{enumerate}

\subsection{Minimal remaining gates and the next proof tasks}

\paragraph{Q: arbitrary-background potential quotients --- \texttt{OPEN}.}
Finite etale and disc/polydisc kernel preservation, followed by
arbitrary same-$K$ analytic-morphism functoriality, are now proved.
So are finite-$K$ extension functoriality and scalar kernel
transfer to finite extensions of finite-polyradius Gauss fields.
The latter result is not a theorem for arbitrary maps over those
Gauss fields. These statements do not establish the
all-complete-extension curve condition or a continuous
representative of every bounded classical difference in the
original $D$-kernel. The precise goal remains
$D[u]=D[v]\Rightarrow D((u-v)S)=0$ for every allowed fixed $S$,
beyond the already established immersion and weak-closure
background classes. A valid regularity mechanism or a direct
arbitrary-background annihilation theorem is still needed.
Exact spreading applies at general algebraic parameter points,
but does not provide the needed bounded ambient-BT representation
of every such fiber. Density of Gauss points does not settle
that fixed-fiber, Borel-weighted issue. The exact finite-coefficient
descent now reduces the scalar-extension part to those
finite-dimensional parameter residue fields, with the full
quantifier over parameter points still an \texttt{OPEN} premise.

\paragraph{C1: sharp compact weak-closure domination --- \texttt{OPEN}.}
For the normalized measure family on one fixed capacity domain,
prove $\nu(K)\leq C_S(K;\Omega)$ for every weak-closure measure
$\nu$ and every compact core $K$. The degree-one nonpolar
case and the proved carrier/positive-tail classes remain
available; no density theorem says they exhaust arbitrary
positive closed backgrounds. Completion, energy compactness
and strong-image closedness do not supply this sharp raw
classical-test bound.

The final focused compact-image check sharpens the quantifiers
without adding a theorem beyond R3. In remaining degree one,
on the same full normalized domain,
\[
 \mathcal M_{\mathcal A}
   =\overline{\mathcal M_{\rm cl}}^{\,\mathrm{local\ vague}},
 \qquad
 C_{\mathcal A}(K)=C_{\rm cl}^*(K)\quad(K\Subset\Omega).
\]
The completed image is locally vaguely compact and its compact
mass maximum is attained. A fixed compact cutoff gives a weakly
compact family of finite measures; global total mass need not be
finite. These facts recover
Corollary~\ref{cor:r3-linear-completed-capacity}. The missing
claim is still: for each compact $K$ and each $\epsilon>0$,
there exists a normalized classical tester on the same domain
with mass on $K$ at least $C_{\mathcal A}(K)-\epsilon$.
A classical representative of the completed maximizer is
not logically necessary, but weak density alone does not give
this lower approximation of compact mass. The route is recorded
as a \texttt{FAILED\_ATTEMPT} to close C1/C2, not a new
compact-regularity proof.

\paragraph{C2: ordinary-open Borel outer regularity --- \texttt{OPEN}.}
One must still prove the additional increasing-set/outer-capacity
step beyond compact control, or the equivalent Borel
weak-closure identification of
Proposition~\ref{prop:r3-outer-closure-equivalence}.
A concrete next target is simultaneous extraction from open
near-maximizers into finitely many separated compact pieces
whose testers have a common capacity limit, with the pasting
loss and $A_K\delta$ tending to zero. The new pasting lemmas
prove the final combination step, not that extraction statement.

\paragraph{D: a verifiable relative solvability criterion --- \texttt{OPEN}.}
The specified completed images are now closed and carry exact
relative-q.e. uniqueness. The constrained variational map is
available. The missing step is a useful independent criterion
for a datum to be fixed by that map, or another appropriate
solution class with justified boundary behavior. A universal
bounded-subsolution criterion in the global bounded-row psh
completion is excluded by the new example. In degree one the
faithful energy space is a proved alternative; its nonlinear
extension in changing mixed degree is not supplied here.

\paragraph{Comparison with CLD, GK and arithmetic applications.}
All new identities remain ordinary CLD-current identities.
A comparison on the entire GK $\delta$-test sheaf needs its
own theorem, and the nonpositive, nonclosed, off-diagonal GK
calculus is not subsumed by these results. No change to the
completed A2 nefness input is required. Strong Green forms
and heights remain subsequent applications, not replacements
for Q, C1, C2 or the remaining Dirichlet criteria.

The continuation should address these gates directly, using the
moving-atom, crossing-support, nonpolar full-torus, tangential-jump
and vertical-fiber examples as scope tests. Reproving the frozen
bounded BT calculus would not resolve them.

% END source: sections/round4_status.tex

\clearpage

% BEGIN frozen or reviewed source: base_r3/references_all.tex
\begin{thebibliography}{BGJK21}
\bibitem[BT76]{BT76}
E.~Bedford and B.~A.~Taylor,
\emph{The Dirichlet problem for a complex Monge--Amp\`ere equation},
Invent. Math. \textbf{37} (1976), 1--44.
\url{https://www.math.stonybrook.edu/~ebedford/PapersForM655/BT1.pdf}.

\bibitem[BT82]{BT82}
E.~Bedford and B.~A.~Taylor,
\emph{A new capacity for plurisubharmonic functions},
Acta Math. \textbf{149} (1982), 1--40.
\url{https://www.math.stonybrook.edu/~ebedford/PapersForM655/BTNewCapacity.pdf}.

\bibitem[Dem12]{Dem12}
J.-P.~Demailly,
\emph{Complex Analytic and Differential Geometry},
online book, version June 21, 2012.
\url{https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf}.

\bibitem[CLD12]{CLD12}
A.~Chambert-Loir and A.~Ducros,
\emph{Formes diff\'erentielles r\'eelles et courants sur les espaces de Berkovich},
arXiv:1204.6277v1 (2012).
\url{https://arxiv.org/pdf/1204.6277v1}.

\bibitem[CLD25]{CLD25}
A.~Chambert-Loir and A.~Ducros,
\emph{Formes diff\'erentielles r\'eelles et courants sur les espaces de Berkovich},
arXiv:1204.6277v2 (2025), 608 pp.
\url{https://arxiv.org/pdf/1204.6277v2}.

\bibitem[GK17]{GK17}
W.~Gubler and K.~K\"unnemann,
\emph{A tropical approach to non-Archimedean Arakelov geometry},
Algebra Number Theory \textbf{11} (2017), 77--180.
The theorem numbering used here is checked in arXiv:1406.7637v3 (2016).
\url{https://arxiv.org/pdf/1406.7637v3}.

\bibitem[GR25]{GR25}
W.~Gubler and J.~Rabinoff,
\emph{Classically psh and pluriharmonic functions on Berkovich spaces},
arXiv:2506.13548v2 (September 17, 2025).
\url{https://gubler.app.uni-regensburg.de/Publikationen/2506.13548.pdf}.

\bibitem[Wu26]{Wu26}
L.~Wu,
\emph{On the local theory of non-Archimedean psh functions},
arXiv:2609.27610v1 (September 23, 2026).
\url{https://arxiv.org/pdf/2609.27610v1}.

\bibitem[BGJK21]{BGJK21}
J.~I.~Burgos Gil, W.~Gubler, P.~Jell, and K.~K\"unnemann,
\emph{Pluripotential theory for tropical toric varieties and non-Archimedean
Monge--Amp\`ere equations}, arXiv:2102.07392v1 (February 15, 2021).
The cited theorem numbers refer to this author-hosted version.
\url{https://gubler.app.uni-regensburg.de/Publikationen/2102.07392.pdf}.
\bibitem[GM19]{GM19}
W.~Gubler and F.~Martin,
\emph{On Zhang's semipositive metrics},
Doc. Math. \textbf{24} (2019), 331--372.
\url{https://ems.press/content/serial-article-files/26466}.


\bibitem[Brouwer11]{Brouwer11}
L.~E.~J.~Brouwer, \emph{\"Uber Abbildung von Mannigfaltigkeiten},
Math. Ann. \textbf{71} (1911), 97--115.
\url{https://doi.org/10.1007/BF01456931}.
\bibitem[DE06]{DE06}
K.~Dabbek and F.~Elkhadhra,
\emph{Capacit\'e associ\'ee \`a un courant positif ferm\'e},
Doc. Math. \textbf{11} (2006), 469--486.
\url{https://ems.press/content/serial-article-files/26001?nt=1}.
\bibitem[ElMir84]{ElMir84}
H.~El Mir, \emph{Sur le prolongement des courants positifs ferm\'es},
Acta Math. \textbf{153} (1984), 1--45.
\url{https://doi.org/10.1007/BF02392374}.
\bibitem[Kol96]{Kol96}
S.~Ko\l odziej, \emph{Some sufficient conditions for solvability
of the Dirichlet problem for the complex Monge--Amp\`ere operator},
Ann. Polon. Math. \textbf{65} (1996), 11--21.
\url{https://matwbn.icm.edu.pl/ksiazki/apm/apm65/apm6512.pdf}.
\bibitem[Kol98]{Kol98}
S.~Ko\l odziej, \emph{The complex Monge--Amp\`ere equation},
Acta Math. \textbf{180} (1998), 69--117.
\url{https://archive.ymsc.tsinghua.edu.cn/pacm_download/117/6558-11511_2006_Article_BF02392879.pdf}.
\bibitem[Mih24]{Mih24}
A.~Mihatsch, \emph{$\delta$-Forms on Lubin--Tate spaces},
arXiv:2112.10018v4 (7 September 2024).
\url{https://arxiv.org/pdf/2112.10018v4}.
\bibitem[FST91]{FST91}
M.~Fukushima, K.-I.~Sato, and S.~Taniguchi,
\emph{On the closable parts of pre-Dirichlet forms and the fine
supports of underlying measures}, Osaka J. Math. \textbf{28}
(1991), 517--535.
\url{https://ir.library.osaka-u.ac.jp/repo/ouka/all/6785/ojm28_03_03.pdf}.

\end{thebibliography}

% END source: base_r3/references_all.tex

\end{document}