Conjecture Atlas

Proof source

Continuous CLD-positive potentials and mixed currents

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. Introduction
  2. Continuous coefficients and approachable potentials
  3. Metrics and Green identities on curves
  4. Restriction and an ambient extension
  5. Curve approximation and contact rigidity
  6. Envelopes on a fixed regular model
  7. Proof of local regularization
  8. Consequences for continuous potentials
  9. Comparison with adelic mixed measures
  10. A quantitative comparison of mixed measures
  11. Notation
  12. Declaration on the use of generative AI

Abstract

Let KK be a non-Archimedean completion of a number field and let X/KX/K be smooth and quasi-projective. A finite continuous function on an ordinary open of XanX^\mathrm{an} has positive Chambert–Loir–Ducros curvature current if and only if it admits smooth positive approximations on a fixed neighbourhood of each point. The proof combines restriction to curves, deterministic complete-intersection approximation, contact rigidity, and continuity of semipositive envelopes. Contact rigidity identifies a continuous curve-positive metric from its contact with a uniform Fubini–Study limit. On a fixed regular model, sums of divisor test ideals and global generation give one-sided O(m−1)O(m^{-1}) Fubini–Study approximations for model-obstacle envelopes. A single projective extension then completes local regularization. As consequences, continuous current-positive potentials admit symmetric mixed products with any positive closed CLD current, and these products depend only on their curvature currents. We also obtain a projective curve criterion under a regular-model hypothesis and positivity of the prescribed Yuan–Zhang local mixed measures.

Introduction

Positivity of a Chambert–Loir–Ducros current is a condition on smooth test forms. To construct positive mixed products, one must also control d′d′′(uT)d'd''(uT) for each intermediate positive closed current TT. Multiplication by a continuous potential is available, but positivity after differentiation requires an additional argument. Local approximation by smooth functions with positive curvature provides that argument. The first question addressed here is whether the current inequality itself supplies such an approximation.

Local regularization

Fix a number field kk, a non-Archimedean place vv, and K=kvK=k_v, with its given absolute value. Its valuation ring is RR. Choose a uniformizer π\pi and set λ=−log⁡∣π∣>0\lambda=-\log|\pi|>0. All smooth forms, positive cones, and currents in this paper are those of the first version of Chambert–Loir–Ducros [3]. In particular, positivity of currents is dual to the positive cone of [3, Definition 5.1.1(c)]. An ordinary open is open in the Berkovich topology.

Let X/KX/K be smooth of pure dimension nn, let O⊂XanO\subset X^\mathrm{an} be an ordinary open, and let u:O→Ru:O\to\mathbf R be finite and continuous. For n≥1n\geq1, the condition d′d′′[u]≥0d'd''[u]\geq0 means

⟨d′d′′[u],η⟩=∫Ou d′d′′η≥0for every positive η∈Acn−1,n−1(O).\langle d'd''[u],\eta\rangle =\int_O u\,d'd''\eta\geq0 \quad\text{for every positive }\eta\in\mathcal A_c^{n-1,n-1}(O).

Every compactly supported smooth test is allowed. For components of different dimensions the condition is imposed componentwise, and there is no curvature condition in dimension zero.

We write u∈Ploc(O)u\in\mathcal P_{\mathrm{loc}}(O) if every x∈Ox\in O has one ordinary neighbourhood V⊂OV\subset O on which there are smooth functions uju_j with d′d′′uj≥0d'd''u_j\geq0 as forms and uj→uu_j\to u uniformly on compact subsets of VV. The neighbourhood is fixed for the entire sequence.

Theorem (Local regularization). Let K=kvK=k_v be a non-Archimedean completion of a number field, with its given absolute value. Let X/KX/K be smooth and quasi-projective, O⊂XanO\subset X^\mathrm{an} an ordinary open, and u:O→Ru:O\to\mathbf R finite continuous. Then the following are equivalent:

  1. The original CLD curvature current d′d′′[u]d'd''[u] is positive on OO in the sense of (eq:current-positive).

  2. The function uu belongs to Ploc(O)\mathcal P_{\mathrm{loc}}(O).

After shrinking the fixed neighbourhood of each point once, the approximating sequence can be chosen decreasing and uniformly convergent on that neighbourhood.

The implication from smooth positive approximation to current positivity follows by testing locally uniform convergence against d′d′′ηd'd''\eta. The reverse implication does not assume that uu is a difference of approachable functions. The field and continuity hypotheses remain in force throughout the theorem; in particular, the assertion does not concern arbitrary upper-semicontinuous potentials. The proof is completed in Section “Completion of the proof”.

Contact rigidity

The projective argument uses a criterion that identifies a metric from its contact with a semipositive lower metric. Recall that a pure Fubini–Study metric on an ample line bundle LL is given by weights m−1max⁡ilog⁡∣si∣m^{-1}\max_i\log|s_i|, where the sections sis_i generate mLmL. A uniform FS limit is a uniform limit of such metrics. A continuous metric is curve-positive if its pullback by every algebraic morphism from a smooth KK-curve has positive Chern current. These conventions are developed in Section “Metrics and Green identities on curves”.

Contact rigidity 1 (Theorem “Contact rigidity”). Let Y/KY/K be smooth, projective, and integral of dimension n≥1n\ge1, and let LL be ample. Let pp be a uniform FS limit on LL, and let rr be a finite continuous curve-positive metric on LL. Then

r≥p,∫Yan(r−p)c1(L,p)n=0⟹r=p.r\ge p,\qquad \int_{Y^\mathrm{an}}(r-p)c_1(L,p)^n=0 \quad\Longrightarrow\quad r=p.

The metric pp need not be constructed as an envelope of rr. The criterion uses the Monge–Ampère measure of the lower metric and positivity of the upper metric only on curves. It therefore applies before a higher-dimensional positive product for rr is available. The proof in Section “Contact rigidity” passes curve Green identities through complete-intersection approximation and then transfers the contact identity to smooth divisors.

Continuous mixed currents

A current is called closed below if both d′T=0d'T=0 and d′′T=0d''T=0. Positive currents admit multiplication by continuous functions (Lemma “Continuous coefficients”). Thus, for a positive closed current SS, the recursive rule

BS()=S,BS(u1,…,um)=d′d′′(u1BS(u2,…,um))\mathcal B_S()=S,\qquad \mathcal B_S(u_1,\ldots,u_m) =d'd''\bigl(u_1\mathcal B_S(u_2,\ldots,u_m)\bigr)

is meaningful whenever its intermediate currents are positive. This is the usual Bedford–Taylor construction.

Theorem (Continuous potentials on a positive closed background). Let K=kvK=k_v, let X/KX/K be smooth quasi-projective of pure dimension nn, and let O⊂XanO\subset X^\mathrm{an} be an ordinary open. Let SS be an original CLD positive current of bidegree (p,p)(p,p), where 0≤p≤n0\leq p\leq n, with d′S=d′′S=0d'S=d''S=0. Suppose 0≤m≤n−p0\leq m\leq n-p and u1,…,um:O→Ru_1,\ldots,u_m:O\to\mathbf R are finite continuous functions with d′d′′[ui]≥0d'd''[u_i]\geq0. Then (eq:recursive-product) defines a positive closed current of bidegree (p+m,p+m)(p+m,p+m). It has the following properties.

  1. It is local, symmetric in the potentials, additive and positively homogeneous in each potential and in the background.

  2. If Sa→SS_a\to S on every compact smooth test, where all SaS_a are positive and closed of bidegree (p,p)(p,p), and the finite continuous functions ui,a:O→Ru_{i,a}:O\to\mathbf R converge to uiu_i locally uniformly with d′d′′[ui,a]≥0d'd''[u_{i,a}]\geq0, then BSa(u1,a,…,um,a)⟶BS(u1,…,um)\mathcal B_{S_a}(u_{1,a},\ldots,u_{m,a}) \longrightarrow\mathcal B_S(u_1,\ldots,u_m) on every compact smooth test.

  3. Each potential enters only through its curvature current: if ww is finite continuous and d′d′′[w]=d′d′′[ui]d'd''[w]=d'd''[u_i], replacing uiu_i by ww leaves the product unchanged.

In top degree the product is a positive Radon measure.

The recursion is described in [3, Section 5.6.8]. For continuous psh potentials, its positivity, continuity, and symmetry are established in [10, Lemma 8.5.5 and Propositions 8.7.2, 8.7.4]. Theorem “Local regularization” allows us to start from current positivity alone. Potential independence then has a short two-sided proof. If d′d′′[h]=0d'd''[h]=0, then both hh and −h-h satisfy Theorem “Local regularization”. For every positive closed TT this gives

d′d′′(hT)≥0,−d′d′′(hT)=d′d′′((−h)T)≥0,hence d′d′′(hT)=0.d'd''(hT)\geq0,\qquad -d'd''(hT)=d'd''((-h)T)\geq0, \qquad\text{hence }d'd''(hT)=0.

No representation of hh as a piecewise-linear or logarithmic function is required. Theorem “Continuous potentials on a positive closed background” concerns continuous potentials; it does not assert a monotone convergence theorem for bounded discontinuous ones. The theorem is proved in Section “Proof of the mixed-current theorem”.

Adelic mixed measures

For the arithmetic application, use the rational boundary completion of [9, Sections 2.4–2.5]. Its objects retain the specified line-bundle identifications. A strongly nef object has a Cauchy presentation by nef model objects, and an integrable object is a difference of two strongly nef objects, with rational tensor powers allowed. The order defining the completion compares the full Cartier-and-metric boundary datum, as recalled in Section “Comparison with adelic mixed measures”.

Theorem (Positivity of the prescribed adelic mixed measure). Let kk be a number field, vv a non-Archimedean place, and K=kvK=k_v. Let U/kU/k be smooth quasi-projective of pure dimension n≥1n\geq1. Let L‾1,…,L‾n\overline L_1,\ldots,\overline L_n be rational integrable adelic line bundles with their specified identifications in the boundary completion of [9, Sections 2.4–2.5]. Suppose their local weights at vv are finite continuous and have positive original CLD curvature currents. Then the signed measure prescribed in [9, Sections 3.6.7–3.6.8] is positive: c1(L‾1)v⋯c1(L‾n)v≥0on UKan.c_1(\overline L_1)_v\cdots c_1(\overline L_n)_v\geq0 \quad\text{on }U_K^\mathrm{an}.

The signed measure is defined by expanding integrable presentations. Its positivity follows by identifying that expansion with the local positive product of the target weights. This conclusion concerns the given measure and does not assert arithmetic nefness of the factors. The comparison is given in Section “Identification and positivity of the prescribed measure”.

Relation to earlier work

Chambert–Loir–Ducros distinguish distributional positivity from local psh approachability in [3, Definition 5.6.3 and Remark 5.6.4]; their stable products are treated in [3, Corollary 5.6.5 and Section 5.6.8]. The later continuous psh class is defined by uniform closure [10, Definition 8.1.7]. We do not identify a continuous current-positive function with a member of that class before proving Theorem “Local regularization”.

On curves, Wanner compares CLD and Thuillier subharmonicity [11, Corollaries 4.8 and 5.4]. Gubler–Jell–Rabinoff provide descent, harmonic tropicalizations, and smooth approximation [4, 5]. The finite residue field is important here: weakly smooth functions on the relevant curves are CLD-smooth. Section “Metrics and Green identities on curves” retains the passage from the original current inequality over KK to those curve results.

Torus-invariant pluripotential theory is developed in [12]. Piecewise-linear semipositivity and positivity tested on curves are studied in [14, 15], and Wu compares local approximation classes in [16, Appendix A]. Our starting hypothesis is instead the original current inequality (eq:current-positive) for an arbitrary finite continuous potential on a smooth variety.

The envelope argument follows the test-ideal method of Fang–Gubler–Künnemann [13], using the mixed-characteristic test ideals of [6]. We use test ideals of section divisors directly. The comparison on a higher normal model allows an obstacle on a higher normal model while subadditivity is applied on a fixed regular model. It is not a direct application of [13, Theorems 5.1 and 7.1]. Orthogonality, once continuity is established, is [8, Theorem 6.7]. The final adelic application identifies a measure already constructed by Yuan–Zhang; it does not introduce another mixed measure.

Proof strategy

The proof separates restriction, contact, and envelopes so that none uses regularization in a form still to be established.

Step 1: restrict the current inequality and extend locally. In a normal disc, let βN=d′d′′R2,ε(log⁡∣t∣,−N)\beta_N=d'd''R_{2,\varepsilon}(\log|t|,-N), where R2,εR_{2,\varepsilon} is a smooth convex regularized maximum. For the zero section i:B↪B×Dsi:B\hookrightarrow B\times D_s, Poincaré–Lelong and continuous multiplication give

βN⟶i∗[B],uβN⟶i∗[u∣B].\beta_N\longrightarrow i_*[B], \qquad u\beta_N\longrightarrow i_*[u|_B].

Testing against pullbacks from BB proves positivity on the zero section. The étale normal form of a smooth pair and Berkovich’s theorem on étale germs then give restriction to smooth subvarieties. Curve regularization and removal of finitely many closed points prove positivity after every algebraic curve map. An algebraic étale presentation also supplies an ambient analytic retraction, extending uu to a continuous curve-positive function vv near the original point (Section “Restriction and an ambient extension”).

Step 2: approximate mixed currents by curves and prove contact. For a generating tuple s0,…,sbs_0,\ldots,s_b, choose smooth cuts of the form σN=∑i=0bai,NsiN,ai,N∈πi(1+πR).\sigma_N=\sum_{i=0}^b a_{i,N}s_i^N, \qquad a_{i,N}\in\pi^i(1+\pi R). On a tropical presentation of any fixed smooth test, equality of the norms of two terms lies on an affine hyperplane, apart from finitely many exceptional NN. Distinct coefficient valuations exclude ties that persist for all NN. Absolute tropical integration therefore gives an eventual O(N−1)O(N^{-1}) error on that test. Successive cuts and weak closure yield normalized individual smooth complete-intersection curves, with fixed intersection degrees (Proposition “Complete-intersection approximation”).

For the contact criterion (eq:contact-intro), put g=r−pg=r-p and α=c1(L,p)\alpha=c_1(L,p). Passing curve Green identities through the approximation of T=αn−1T=\alpha^{n-1} gives ν=d′d′′(gT)=σ−αn,σ≥0,0≤−∫g dν=−∫g dσ≤0.\nu=d'd''(gT)=\sigma-\alpha^n,\qquad \sigma\geq0,\qquad 0\leq-\int g\,d\nu=-\int g\,d\sigma\leq0. The nonnegative energy of g+thg+th, for smooth hh and t∈Rt\in\mathbf R, then forces ν=0\nu=0. For a smooth divisor D=div(s)D=\mathop{\mathrm{div}}(s) in mLmL, set ρN=max⁡{mp,log⁡∣s∣+N}\rho_N=\max\{mp,\log|s|+N\}. Fixed-potential continuity gives

c1(mL,ρN)∧αn−1⟶i∗(c1(L∣D,p∣D)n−1).c_1(mL,\rho_N)\wedge\alpha^{n-1} \longrightarrow i_*\bigl(c_1(L|_D,p|_D)^{n-1}\bigr).

Green symmetry transfers contact to DD, and induction on dimension proves (eq:contact-intro) (Section “Curve approximation and contact rigidity”). No mixed product with the upper metric rr is assumed in this argument.

Step 3: construct continuous envelopes on a fixed regular model. Represent a model obstacle on a higher normal model W\mathscr W as r=θ+log⁡∣E∣r=\theta+\log|\mathscr E|, with E\mathscr E an invertible vertical ideal. For each admissible section s∈H0(Y,lL)s\in H^0(Y,lL), let DsD_s be its divisor on a fixed regular model X\mathscr X. We take the sum jm=∑l≥1∑0≠s∈VlτX ⁣(mlDs),Vl={s:∥s∥lr,sup⁡≤1}.\mathfrak j_m=\sum_{l\ge1}\sum_{0\ne s\in V_l} \tau_{\mathscr X}\!\left(\frac mlD_s\right), \qquad V_l=\{s:\|s\|_{lr,\sup}\le1\}. Subadditivity gives the lower bound below. Birational comparison, followed by clearing a fixed discrepancy denominator, gives the upper bound:

πBjmOW⊆Em,PL(r)≤θ+1mlog⁡∣jm∣≤r+Bλm.\pi^B\mathfrak j_m\mathcal O_{\mathscr W}\subseteq\mathscr E^m,\qquad P_L(r)\leq\theta+\frac1m\log|\mathfrak j_m| \leq r+\frac{B\lambda}{m}.

Global generation and one fixed auxiliary line bundle turn these bounds into pure-FS approximations of PL(r)P_L(r) on LL. On X=PRa\mathscr X=\mathbf P_R^a, the identity KX+(a+1)O(1)=0K_{\mathscr X}+(a+1)\mathcal O(1)=0 cancels the auxiliary twist directly and gives

0≤qm−PL(r)≤Bλm.0\leq q_m-P_L(r)\leq\frac{B\lambda}{m}.

Model-metric density and the one-Lipschitz property of the envelope treat continuous obstacles. Orthogonality [8, Theorem 6.7] then applies (Section “Uniform approximation and orthogonality”).

Step 4: extend once to projective space and apply contact rigidity. After adding a constant and shrinking the ambient neighbourhood, arrange 0<v+c<λ/20<v+c<\lambda/2. Choose one compact polynomial norm box CC and an FS metric qq, with a section s0s_0, such that b=−log⁡∥s0∥q,b=0 near x,{b≤λ}=C.b=-\log\|s_0\|_q,\qquad b=0\text{ near }x, \qquad\{b\le\lambda\}=C. On {b<λ}\{b<\lambda\}, use the weight max⁡{v+c,b}\max\{v+c,b\}; on {b>λ/2}\{b>\lambda/2\}, use qq. The two definitions agree on the overlap and give a global curve-positive metric rr (Lemma “Projective extension”). Its envelope p=PL(r)p=P_L(r) satisfies the contact identity, so Theorem “Contact rigidity” gives r=pr=p. Smooth FS approximation and restriction to one fixed neighbourhood recover uu. Explicit constant translations make the approximants decreasing (Section “Completion of the proof”). The same envelope–contact argument gives the projective curve criterion under the retained regular-model hypothesis (Corollary “A projective curve criterion”).

Organization

Section “Continuous coefficients and approachable potentials” proves the continuous-coefficient estimate and the continuity of a fixed approachable potential acting on positive closed currents. Section “Metrics and Green identities on curves” supplies smooth Fubini–Study approximation, the measure normalization, and the Green identities on curves. Section “Restriction and an ambient extension” establishes restriction and the ambient extension. Section “Curve approximation and contact rigidity” combines complete-intersection approximation with the direct divisor limit to prove contact rigidity. Section “Envelopes on a fixed regular model” proves envelope continuity on a fixed regular model, with projective space as a corollary. Section “Proof of local regularization” constructs the projective extension and completes local regularization. The mixed-current and curve-criterion consequences are proved in Section “Consequences for continuous potentials”; Section “Comparison with adelic mixed measures” identifies the prescribed adelic measure. Appendix “A quantitative comparison of mixed measures” records a quantitative comparison of mixed measures, independently of the contact argument.

Continuous coefficients and approachable potentials

We use the smooth cutoff and density results [3, Corollary 3.3.4 and Propositions 3.3.5–3.3.6], the Poincaré–Lelong formula [3, Theorem 4.6.5], and uniform stability of products [3, Corollary 5.6.5]. Weak convergence of currents always means convergence on every compactly supported smooth test. Convergence of measures on continuous tests will be stated explicitly.

Definition. On an ordinary open O⊂XanO\subset X^\mathrm{an}, a smooth function is locally of the form F(log⁡∣f1∣,…,log⁡∣fr∣),F(\log|f_1|,\ldots,\log|f_r|), where the fif_i are analytic units and FF is smooth near their logarithmic image. The class Ploc(O)\mathcal P_{\mathrm{loc}}(O) is defined in Section “Introduction”. We write u∈Dloc(O)u\in\mathcal D_{\mathrm{loc}}(O) if uu is locally a difference of two members of Ploc\mathcal P_{\mathrm{loc}}. For metrics, these conditions apply to their weights in local frames. The weight in a frame ee is −log⁡∥e∥-\log\|e\|.

A change of frame changes a weight by the logarithm of the norm of an analytic unit. That function is smooth and has zero d′d′′d'd''. Continuous multiplication of positive currents is known from [3, Corollary 5.4.7] and [10, Lemma 8.5.1]. We give the coefficient construction because it also controls weakly convergent nets.

Lemma (Continuous coefficients). Let OO be an ordinary open in a smooth pure nn-dimensional KK-variety. Let TaT_a be positive currents of bidegree (n−q,n−q)(n-q,n-q), 0≤q≤n0\leq q\leq n, converging weakly to TT. Then TT is positive. For S=TS=T and for each S=TaS=T_a, every finite continuous f:O→Rf:O\to\mathbf R determines a unique current fSfS by ⟨fS,η⟩=lim⁡j→∞⟨S,fjη⟩,\langle fS,\eta\rangle =\lim_{j\to\infty}\langle S,f_j\eta\rangle, where η\eta is any compactly supported smooth test and fjf_j are smooth functions converging uniformly to ff near suppη\mathop{\mathrm{supp}}\eta. The limit exists and is independent of the choice of fjf_j.

On a fixed compact test chart, the scalar coefficient measures of TaT_a have eventually uniformly bounded total variations. The same holds after wedging TaT_a with any fixed smooth form, in the bidegrees where the wedge is defined. The cone of positive currents is pointed. If fa→ff_a\to f locally uniformly, with fa,ff_a,f finite continuous, then

faTa⟶fTf_aT_a\longrightarrow fT

on every compact smooth test. Nets are allowed.

Proof. Positivity of TT follows by taking the limit on each positive test. We establish the coefficient estimate locally and then use a finite smooth partition. Fix a moment chart with rr moment coordinates, a compactly supported (q,q)(q,q)-form η\eta, and a nonnegative smooth cutoff χ\chi equal to one near its support. Let JJ denote the involution interchanging the two differential factors, and write ηs=12(η+(−1)qJη),ηa=η−ηs.\eta_{\mathrm s}=\tfrac12\bigl(\eta+(-1)^qJ\eta\bigr), \qquad \eta_{\mathrm a}=\eta-\eta_{\mathrm s}. Wedging ηa\eta_{\mathrm a} with γ∧Jγ\gamma\wedge J\gamma, for an (r−q,0)(r-q,0)-form γ\gamma, gives zero. In the convention [3, Definition 5.1.1(c)], both signs of ηa\eta_{\mathrm a} are therefore positive. A positive current annihilates ηa\eta_{\mathrm a}, also after multiplication by a smooth real scalar.

For symmetric forms the positivity test is a finite-dimensional positive-semidefinite matrix condition. The form Ω=1q!(∑j=1rd′xj∧d′′xj)q(Ω=1 if q=0)\Omega=\frac1{q!}\left(\sum_{j=1}^r d'x_j\wedge d''x_j\right)^q \quad(\Omega=1\text{ if }q=0) has a positive-definite coefficient matrix. Boundedness of the coefficient matrix of ηs\eta_{\mathrm s} on the fixed compact set gives CηC_\eta such that, for every smooth real hh, Cη∥h∥suppηχΩ±hηs≥0.C_\eta\|h\|_{\mathop{\mathrm{supp}}\eta}\chi\Omega\pm h\eta_{\mathrm s}\geq0. If hh uses additional moment coordinates, pull this inequality to a common refinement; positivity is preserved. The forms extend by zero outside the chart. For S=TaS=T_a or TT, we obtain

∣⟨S,hη⟩∣≤Cη∥h∥suppη⟨S,χΩ⟩.|\langle S,h\eta\rangle| \leq C_\eta\|h\|_{\mathop{\mathrm{supp}}\eta}\langle S,\chi\Omega\rangle.

At this stage hh is smooth, so the pairing is already defined.

The right-hand evaluations for TaT_a converge and are eventually bounded. Apply (eq:coefficient-estimate) to the finitely many coordinate test forms. Their scalar functionals extend uniquely to continuous compactly supported functions, and the Riesz theorem represents them by signed Radon measures. The same estimate bounds their total variations. These extensions agree on overlaps by uniform smooth density. On a fixed test stage CηC_\eta is bounded by a constant times the supremum norm of the coefficient matrix, so the extended functionals define currents.

We can now define fSfS, for S=TS=T or S=TaS=T_a, by multiplying these coefficient measures by ff. Smooth density supplies the functions fjf_j in the statement; estimate (eq:coefficient-estimate) proves that their pairings converge to this product independently of the approximation. The same local estimate proves continuity on each smooth test stage.

For fixed continuous ff, choose smooth fj→ff_j\to f uniformly on an enlarged compact set. For an eventually uniform MηM_\eta, ∣⟨fTa−fT,η⟩∣≤∣⟨Ta−T,fjη⟩∣+Mη∥f−fj∥∞.|\langle fT_a-fT,\eta\rangle| \leq|\langle T_a-T,f_j\eta\rangle| +M_\eta\|f-f_j\|_\infty. First fix jj, let aa tend to its limit, and then let jj increase. This proves fTa→fTfT_a\to fT. For varying faf_a, the additional term satisfies ∣⟨(fa−f)Ta,η⟩∣≤Mη∥fa−f∥suppη⟶0.|\langle(f_a-f)T_a,\eta\rangle| \leq M_\eta\|f_a-f\|_{\mathop{\mathrm{supp}}\eta}\longrightarrow0. A finite partition treats any compact test. The coefficient bounds also control wedges with a fixed smooth form, whose coefficients are bounded on compact charts. Finally, if both SS and −S-S are positive, their evaluations on χΩ\chi\Omega vanish; applying (eq:coefficient-estimate) to SS gives S=0S=0. ◻

Proposition (A fixed approachable potential). Let TT be a positive current on OO with d′T=d′′T=0d'T=d''T=0, and let u∈Ploc(O)u\in\mathcal P_{\mathrm{loc}}(O). Whenever the bidegrees permit it, Mu(T):=d′d′′(uT)M_u(T):=d'd''(uT) is positive and closed. For fixed uu, the operation MuM_u is continuous on the cone of positive closed currents for weak convergence. The same is true jointly when ua→uu_a\to u locally uniformly and all ua,uu_a,u belong to Ploc\mathcal P_{\mathrm{loc}}. Successive such operations are symmetric in their potentials.

Proof. On a fixed smaller neighbourhood choose smooth uj→uu_j\to u with d′d′′uj≥0d'd''u_j\geq0. Closedness gives d′d′′(ujT)=d′d′′uj∧T≥0.d'd''(u_jT)=d'd''u_j\wedge T\geq0. The curvature d′d′′ujd'd''u_j is a symmetric positive (1,1)(1,1)-form, hence strongly positive, so its wedge preserves positivity. Lemma “Continuous coefficients” and continuity of differentiation give the positive limit d′d′′(uT)d'd''(uT). Both differentials annihilate this limit. Equation (eq:moving-coefficients) proves the two continuity assertions. Induction gives joint continuity for successive operations. Simultaneous smooth positive approximations of finitely many potentials then give symmetry from the smooth wedge product. ◻

For a continuously metrized line bundle with approachable weights, write c1(L,p)∧Tc_1(L,p)\wedge T for this operation in local frames. The local definitions agree because a transition logarithm ℓ\ell is smooth, d′d′′ℓ=0d'd''\ell=0, and d′d′′(ℓT)=0d'd''(\ell T)=0.

Metrics and Green identities on curves

Fubini–Study metrics and normalization

Definition. Let LL be an ample rational line bundle on a projective Y/KY/K. A pure Fubini–Study metric on LL is defined by an integer d>0d>0, with dLdL an actual line bundle, and a generating tuple s0,…,sb∈H0(Y,dL)s_0,\ldots,s_b\in H^0(Y,dL). Its weight in a local frame is qe=1dmax⁡0≤j≤blog⁡∣sj/e⊗d∣.q_e=\frac1d\max_{0\leq j\leq b}\log|s_j/e^{\otimes d}|. Coefficients multiplying the sections belong to KK. A uniform FS limit is a uniform limit of these metrics. A metric is globally smoothly approximable if it is a uniform limit of global smooth metrics with positive curvature forms.

A model metric is nef if, after clearing tensor denominators, its model line bundle has nonnegative degree on every integral curve of the special fibre. A continuous metric is semipositive if it is a uniform limit of nef model metrics. For an ample rational line, this is equivalent to being a uniform pure-FS limit [7, Corollary 7.6 and Theorem 7.8]. All uniform differences of metrics are scalar differences on the same specified line bundle.

We use the standard regularized maximum from [3, proof of Proposition 6.3.2].

Proposition (Regularized maximum). Fix an even nonnegative C∞C^\infty function ρ\rho, supported in [−1,1][-1,1], with integral one. For h≥1h\geq1 and ε>0\varepsilon>0, put Rh,ε(t)=∫Rhmax⁡i(ti+εzi)∏i=1hρ(zi) dz.R_{h,\varepsilon}(t) =\int_{\mathbf R^h}\max_i(t_i+\varepsilon z_i) \prod_{i=1}^h\rho(z_i)\,dz. This function is smooth, convex, and nondecreasing in each variable. It satisfies max⁡iti≤Rh,ε(t)≤max⁡iti+ε,Rh,ε(t+c1)=Rh,ε(t)+c.\max_i t_i\leq R_{h,\varepsilon}(t)\leq\max_i t_i+\varepsilon, \qquad R_{h,\varepsilon}(t+c\mathbf1)=R_{h,\varepsilon}(t)+c. An entry more than 2ε2\varepsilon below the maximum is inactive on a neighbourhood of that point. If smooth a1,…,aha_1,\ldots,a_h have positive curvature, so does Rh,ε(a1,…,ah)R_{h,\varepsilon}(a_1,\ldots,a_h).

Translation covariance gives the metric transition law, and the inactive-coordinate property handles zeros of the generating sections. These are precisely the features needed in the next lemma.

Lemma (Smooth approximation of FS metrics). A pure-FS metric, and hence a uniform FS limit, is globally smoothly approximable. If pp is globally smoothly approximable on LL, if m>0m>0 is an integer for which mLmL is a line bundle, and if s∈H0(Y,mL)s\in H^0(Y,mL), then for each N∈RN\in\mathbf R the metric ρN=max⁡{mp,log⁡∣s∣+N}\rho_N=\max\{mp,\log|s|+N\} on mLmL is globally smoothly approximable.

Proof. For a pure-FS metric from sections of dLdL, pull back the smooth positive approximations of the canonical metric on OPb(1)\mathcal O_{\mathbf P^b}(1) from [3, Proposition 6.3.2], and divide their weights by dd. A diagonal choice treats uniform FS limits.

The second assertion is immediate for s=0s=0. Otherwise, take smooth positive pj→pp_j\to p and regularize max⁡{mpj,log⁡∣s∣+N}\max\{mp_j,\log|s|+N\}. The logarithmic entry is inactive near the divisor of ss; elsewhere it is smooth with zero curvature. The resulting global metrics are positive and have error at most m∥pj−p∥∞+εjm\|p_j-p\|_\infty+\varepsilon_j. This is the same regularized-maximum construction, applied to a smooth weight and a section weight. ◻

Lemma (Degree and comparison of measures). On a smooth projective Y/KY/K of dimension nn, globally smoothly approximable metrics on L1,…,LnL_1,\ldots,L_n have total mixed mass deg⁡K(L1⋯Ln)\deg_K(L_1\cdots L_n). For uniform FS limits, the model-theoretic mixed measure agrees with the CLD product.

The mass assertion is [3, Corollary 6.4.4]. The model comparison is [3, Theorem 6.9.3 and Proposition 6.9.7], with the discretely valued normalization of [3, Section 6.9.6]. For uniform FS limits, use uniform continuity of the model measures [8, Proposition 2.23(d) and Remark 2.24] and of the approachable CLD products. Rational tensor denominators are cleared and divided on both sides.

Degrees in this paper are over the stated ground field. A variety over a finite extension F/KF/K, regarded as a KK-variety, contributes the structural factor [F:K][F:K] to intersection and integration. Indeed, a moment map followed by (Gmr)F→(Gmr)K(\mathbf G_m^r)_F\to(\mathbf G_m^r)_K has that degree. This operation is distinct from scalar extension of a KK-variety: the normalized scalar-extension integration formula used next has no extra degree factor.

Definition. A continuous function on an ordinary algebraic analytic open is curve-positive if its pullback by every algebraic morphism from a smooth KK-curve has positive CLD curvature current on the inverse-image open. For a continuous metric the same condition is imposed on the pulled-back line bundle. Curves may be disconnected. The definition uses algebraic curve maps over the given KK; it does not build positivity after arbitrary field extension into the hypothesis.

Regularization and energy in dimension one

Proposition (Curve regularization). Let C/KC/K be a smooth algebraic curve and W⊂CanW\subset C^\mathrm{an} an ordinary open. A finite continuous CLD-positive function on WW belongs to Ploc(W)\mathcal P_{\mathrm{loc}}(W). A finite continuous function with zero Laplacian on a connected proper smooth curve is constant.

Proof. For a finite Galois extension F/KF/K, average a nonnegative compact smooth scalar test η\eta on WFW_F over the Galois group. The invariant average descends by [4, Proposition 14.2]. It is smooth because weakly smooth and CLD-smooth forms coincide on these curves [4, Remark 15.10]: their residue fields are algebraic over a finite field. If η‾\overline\eta is the descended test, the scalar-extension integration formula [4, Proposition 14.5] gives ∫WFuF d′d′′η=∫Wu d′d′′η‾.\int_{W_F}u_F\,d'd''\eta=\int_Wu\,d'd''\overline\eta. Uniform smooth density permits the continuous coefficient uu. Consequently current positivity passes to finite Galois extensions.

Let YY be a compact strictly affinoid curve domain inside WW. After such an extension choose a semistable skeleton Γ\Gamma with retraction τ\tau, edge lengths, and positive weights wew_e. For an edgewise smooth φ\varphi, constant near vertices and zero near boundary vertices, φ∘τ\varphi\circ\tau, extended by zero, is a compact smooth test. Its Laplacian has no vertex atoms, and integration on the annular edges gives

∫Γφ dτ∗(d′d′′[u]∣Y)=∑ewe∫eu∣e φe′′ ds.\int_\Gamma\varphi\,d\tau_*(d'd''[u]|_Y) =\sum_e w_e\int_e u|_e\,\varphi_e''\,ds.

This is also the retraction comparison [5, Proposition 6.3.2]; it does not assume that uu factors through τ\tau.

Tests inside an edge show that u∣eu|_e is convex. The positive Radon measure d′d′′[u]d'd''[u] has finite mass on the compact domain. Thus its edge second derivatives have finite mass and finite one-sided derivatives at the endpoints. Integration by parts in (eq:curve-skeleton), allowing arbitrary constants near the interior vertices, gives a nonnegative weighted sum of outgoing slopes at each such vertex. The finite-graph maximum principle therefore applies to uu.

To pass from graph inequalities to subharmonicity, we follow the harmonic-comparison argument of [11, Theorem 4.7]. A harmonic comparison function becomes skeletal after semistable refinement [4, Section 15.6 and Proposition 15.7]. Its boundary comparison with uu holds on that skeleton. Any prescribed finite set of type-II points can be included in a refinement after a further finite Galois extension [5, Remark 5.4.2]. The boundary and harmonic compatibilities are [5, Lemma 5.4.1 and Proposition 5.6.8]. Density of these points and continuity prove harmonic comparison on all of YY. Hence uu is Thuillier-subharmonic.

By [4, Proposition 15.11], uu has a decreasing net of weakly smooth psh approximants on a fixed relatively compact open. They are CLD-smooth by the finite-residue-field comparison. On a smaller compact neighbourhood, Dini’s finite-cover argument selects a uniformly convergent sequence. On a curve, positivity of the curvature current of a smooth function implies positivity of its curvature form: a negative coefficient on an open one-dimensional tropical cell is detected by a nonnegative scalar test there; continuity handles faces. This proves approachability.

For a zero-Laplacian function on a proper curve, (eq:curve-skeleton) makes its restriction to every refined skeleton harmonic. A harmonic function on a connected compact graph without boundary is constant. Refinement through arbitrary type-II points and descent give the final assertion. ◻

The smooth Green identity is [3, Theorem 3.12.2]. We extend it and the smooth chain rule to the continuous gaps below using curve regularization and bounded Laplacian variation.

Lemma (Green identities on curves). Let r,pr,p be finite continuous metrics with positive Chern currents on the same rational line LL on a smooth proper curve C/KC/K. Their gap g=r−pg=r-p has smooth approximants aj→ga_j\to g uniformly with sup⁡j∥d′d′′aj∥TV<∞\sup_j\|d'd''a_j\|_{\mathrm{TV}}<\infty. The two Chern measures have mass deg⁡KL\deg_KL, so ∥d′d′′g∥TV≤2deg⁡KL\|d'd''g\|_{\mathrm{TV}}\leq2\deg_KL.

On every finite real span EE of such gaps and smooth functions,

∫Ca d′d′′b=∫Cb d′d′′a,QC(a):=−∫Ca d′d′′a≥0(a,b∈E).\int_C a\,d'd''b=\int_C b\,d'd''a,\qquad Q_C(a):=-\int_C a\,d'd''a\geq0 \quad(a,b\in E).

Moreover, ΓC(a)=d′d′′(a2/2)−a d′d′′a\Gamma_C(a)=d'd''(a^2/2)-a\,d'd''a is a positive measure of mass QC(a)Q_C(a), and d′d′′Φ(a)=Φ′(a)d′d′′a+Φ′′(a)ΓC(a)d'd''\Phi(a)=\Phi'(a)d'd''a+\Phi''(a)\Gamma_C(a) for every smooth real Φ\Phi defined near the range of aa. If QC(a)=0Q_C(a)=0, then d′d′′a=0d'd''a=0 and aa is constant on each connected component.

Proof. In finitely many fixed frame neighbourhoods, use Proposition “Curve regularization” to approximate the weights by smooth subharmonic ri,j,pi,jr_{i,j},p_{i,j}. For a compact FF in one such neighbourhood, choose a nonnegative smooth cutoff ξ=1\xi=1 near FF. Then (d′d′′ri,j)(F)≤∫ξ d′d′′ri,j=∫ri,j d′d′′ξ≤∥ri,j∥suppd′d′′ξ∥d′d′′ξ∥TV.(d'd''r_{i,j})(F) \leq\int\xi\,d'd''r_{i,j} =\int r_{i,j}\,d'd''\xi \leq\|r_{i,j}\|_{\mathop{\mathrm{supp}}d'd''\xi}\|d'd''\xi\|_{\mathrm{TV}}. This bounds the local Laplacians of gi,j=ri,j−pi,jg_{i,j}=r_{i,j}-p_{i,j}. For a compactly supported ρ≥0\rho\geq0, the square identity gives ∫ρ d′gi,j∧d′′gi,j=∫(gi,j2/2)d′d′′ρ−∫ρgi,jd′d′′gi,j,\int\rho\,d'g_{i,j}\wedge d''g_{i,j} =\int(g_{i,j}^2/2)d'd''\rho -\int\rho g_{i,j}d'd''g_{i,j}, so the local energies are bounded as well.

Take a finite smooth partition χi\chi_i and set aj=∑iχigi,ja_j=\sum_i\chi_i g_{i,j}. Cauchy–Schwarz for integration on the curve bounds the cross terms by ∥d′χi∧d′′gi,j+d′gi,j∧d′′χi∥TV≤2QC(χi)1/2(∫ρi d′gi,j∧d′′gi,j)1/2,\|d'\chi_i\wedge d''g_{i,j} +d'g_{i,j}\wedge d''\chi_i\|_{\mathrm{TV}} \leq2Q_C(\chi_i)^{1/2} \left(\int\rho_i\,d'g_{i,j}\wedge d''g_{i,j}\right)^{1/2}, where ρi=1\rho_i=1 near suppχi\mathop{\mathrm{supp}}\chi_i. The product rule therefore gives the uniform global variation bound.

The mass formula is [10, Corollary 18.1.7]. Uniform convergence and the variation bound imply weak convergence of d′d′′ajd'd''a_j on continuous tests and −∫ajd′d′′aj⟶−∫ad′d′′a.-\int a_jd'd''a_j\longrightarrow-\int ad'd''a. The cited smooth Green formula and d′d′′(aj2/2)−ajd′d′′aj=d′aj∧d′′ajd'd''(a_j^2/2)-a_jd'd''a_j=d'a_j\wedge d''a_j pass to the limit. The latter positive measures have convergent masses. The smooth chain rule passes as well. Choosing approximations simultaneously proves these assertions on a fixed finite span EE. If QC(a)=0Q_C(a)=0, the positive measure ΓC(a)\Gamma_C(a) is zero. The nonnegative quadratic QC(a+th)Q_C(a+th), for smooth hh and arbitrary t∈Rt\in\mathbf R, has zero linear coefficient. By Green symmetry, ∫h d′d′′a=0\int h\,d'd''a=0 for every such hh. Thus d′d′′a=0d'd''a=0, and Proposition “Curve regularization” gives constancy. ◻

Restriction and an ambient extension

Restriction of a current to a subvariety is not a formal operation. For continuous potentials it follows from a positive kernel in the normal direction.

Lemma (A normal kernel). Let BB be an ordinary open in the analytification of a smooth KK-variety of pure dimension d≥1d\geq1, let Ds={∣t∣<s}D_s=\{|t|<s\}, and let i:B↪B×KDsi:B\hookrightarrow B\times_KD_s be the zero section. For a fixed ε>0\varepsilon>0, put vN=R2,ε(log⁡∣t∣,−N),βN=d′d′′vN.v_N=R_{2,\varepsilon}(\log|t|,-N),\qquad \beta_N=d'd''v_N. Then, for every finite continuous uu on the product, βN⟶i∗[B],uβN⟶i∗[u∣B]\beta_N\longrightarrow i_*[B],\qquad u\beta_N\longrightarrow i_*[u|_B] as currents. In particular, d′d′′[u]≥0d'd''[u]\geq0 implies d′d′′[u∣B]≥0d'd''[u|_B]\geq0.

Proof. The form βN\beta_N is smooth and strongly positive, with support in e−N−2ε≤∣t∣≤e−N+2ε.e^{-N-2\varepsilon}\leq|t|\leq e^{-N+2\varepsilon}. The logarithm is locally integrable. Dominated convergence and Poincaré–Lelong give d′d′′[vN]→d′d′′[log⁡∣t∣]=i∗[B]d'd''[v_N]\to d'd''[\log|t|]=i_*[B], with multiplicity one. Lemma “Continuous coefficients” permits the fixed multiplier uu. For a positive η∈Acd−1,d−1(B)\eta\in\mathcal A_c^{d-1,d-1}(B), pr⁡∗η∧βN\operatorname{pr}^*\eta\wedge\beta_N is a positive compact test for large NN. Indeed, its support lies above a fixed compact subset of BB and inside a fixed closed normal disc. Thus 0≤⟨d′d′′[u],pr⁡∗η∧βN⟩=⟨uβN,pr⁡∗(d′d′′η)⟩⟶⟨d′d′′[u∣B],η⟩.\begin{aligned} 0&\leq\langle d'd''[u],\operatorname{pr}^*\eta\wedge\beta_N\rangle\\ &=\langle u\beta_N,\operatorname{pr}^*(d'd''\eta)\rangle \longrightarrow\langle d'd''[u|_B],\eta\rangle. \end{aligned} Fixed base and normal cutoffs make the middle expression a compact test pairing without altering it. ◻

Lemma (A local product neighbourhood). Let i:Z↪Yi:Z\hookrightarrow Y be a closed immersion of smooth KK-varieties, let x∈Zanx\in Z^\mathrm{an}, and let rr be the codimension near xx. There are ordinary neighbourhoods V⊂ZanV\subset Z^\mathrm{an} and W⊂YanW\subset Y^\mathrm{an}, a radius s>0s>0, and an analytic isomorphism Φ:V×KDsr→∼W\Phi:V\times_KD_s^r\xrightarrow{\sim}W whose zero section is i∣Vi|_V. Consequently restriction of a finite continuous positive potential along ii preserves current positivity.

Proof. Use the étale local form of a smooth pair [2, Lemma 37.17.1, Tag 0FUE]. After Zariski shrinking there is an étale map Ψ=(z,F):Y⟶AKd+r,Z=Ψ−1(AKd×{0}),\Psi=(z,F):Y\longrightarrow\mathbf A_K^{d+r},\qquad Z=\Psi^{-1}(\mathbf A_K^d\times\{0\}), where d=dim⁡Zd=\dim Z and z∣Zz|_Z is étale. Form E=Y×AKdZE=Y\times_{\mathbf A_K^d}Z. The two maps E→ p Y,E→ q=(pr⁡Z,F∘p) Z×AKrE\xrightarrow{\,p\,}Y,\qquad E\xrightarrow{\,q=(\operatorname{pr}_Z,F\circ p)\,} Z\times\mathbf A_K^r are étale by base change. At the diagonal point ee above xx, both induce the identity on the completed residue field H(x)\mathscr H(x). Berkovich’s equivalence for étale germs [1, Theorem 3.4.1] therefore makes both analytic germs isomorphisms. By the definition of these germs [1, Section 3.4], they are represented by isomorphisms of ordinary open neighbourhoods. Their composition identifies a neighbourhood of xx in YanY^\mathrm{an} with an ordinary open G⊂Zan×AKr,anG\subset Z^\mathrm{an}\times\mathbf A_K^{r,\mathrm{an}}, matching ZZ with the zero section.

Choose a compact affinoid neighbourhood V0V_0 of xx in ZanZ^\mathrm{an} whose zero section lies in GG. In the product of V0V_0 with a fixed closed normal polydisc, the complement of GG is compact and disjoint from the zero section. If nonempty, it has a positive minimum of max⁡i∣ti∣\max_i|t_i|. Choose ss smaller than the initial polydisc radius and, when the complement is nonempty, smaller than that minimum. Then V×Dsr⊂GV\times D_s^r\subset G, where V=IntV0V=\mathop{\mathrm{Int}}V_0. Its image is the required ordinary open WW. Finally apply Lemma “A normal kernel” in each normal coordinate; a zero-dimensional target has no curvature condition. ◻

Proposition (Curve maps and extension). Let uu be a finite continuous CLD-positive function on an ordinary open O⊂XanO\subset X^\mathrm{an}, where X/KX/K is smooth. Then:

  1. For every algebraic morphism f:C→Xf:C\to X from a smooth KK-curve, u∘fanu\circ f^\mathrm{an} is CLD-positive on (fan)−1(O)(f^\mathrm{an})^{-1}(O).

  2. Near each point of OO, an affine embedding of XX has a finite continuous curve-positive ambient extension vv of uu.

Proof. Argue componentwise. For a nonconstant map the algebraic image is a curve. Away from its finitely many singular points, Lemma “A local product neighbourhood” restricts uu to the smooth image. Proposition “Curve regularization” supplies smooth positive approximations, whose pullbacks prove positivity on the source outside finitely many closed points.

Continuity removes each such point PP. A sufficiently small neighbourhood is a disc over K(P)K(P), viewed over KK. Choose 0≤wN≤10\leq w_N\leq1, equal to one near PP, with supports shrinking to PP, by translating one smooth profile in log⁡∣t∣\log|t|. Then ∫d′d′′wN=0,sup⁡N∥d′d′′wN∥TV<∞.\int d'd''w_N=0,\qquad \sup_N\|d'd''w_N\|_{\mathrm{TV}}<\infty. The bound contains only the fixed structural degree [K(P):K][K(P):K]. A smooth scalar φ\varphi is constant near PP, since each of its finitely many analytic unit norms is constant there. For large NN, ∣⟨d′d′′[u],φwN⟩∣≤∣φ(P)∣ sup⁡suppd′d′′wN∣u−u(P)∣ ∥d′d′′wN∥TV⟶0.|\langle d'd''[u],\varphi w_N\rangle| \leq|\varphi(P)|\, \sup_{\mathop{\mathrm{supp}}d'd''w_N}|u-u(P)|\,\|d'd''w_N\|_{\mathrm{TV}} \longrightarrow0. Apply positivity off PP to φ(1−wN)\varphi(1-w_N). Constant curve maps pull back uu to a constant.

For the extension, choose affine coordinates (z,y)(z,y) of an embedding X⊂AKd+rX\subset\mathbf A_K^{d+r}, with z:X→AKdz:X\to\mathbf A_K^d étale. The algebraic morphism Θ:X×AKr⟶AKd+r,(a,h)⟼(z(a),y(a)+h)\Theta:X\times\mathbf A_K^r\longrightarrow\mathbf A_K^{d+r},\qquad (a,h)\longmapsto(z(a),y(a)+h) is étale: after translating the second coordinate it is a base change of zz. At (x,0)(x,0), the map induces an isomorphism of completed residue fields. The same germ theorem [1, Theorem 3.4.1] gives an analytic inverse branch ι(w)=(ρ(w),y(w)−y(ρ(w)))\iota(w)=(\rho(w),y(w)-y(\rho(w))) on one ordinary ambient open. This branch agrees with the diagonal on XX. Shrinking so that ρ\rho takes values in OO, set v=u∘ρv=u\circ\rho. For any smooth algebraic curve mapping to the ambient affine space, the fibre product with Θ\Theta is a smooth algebraic curve. Its map to XX is covered by the first part, and the inverse branch identifies the relevant ordinary curve open with an open of that fibre product. Thus vv is curve-positive over KK. ◻

Curve approximation and contact rigidity

We first approximate the mixed currents of approachable metrics by algebraic curves. The coefficients are chosen over the original field KK, and the normalized intersection degrees remain fixed. These degrees supply the variation bounds needed to pass the curve Green identities to a higher-dimensional mixed current.

Complete-intersection approximation

Proposition (Complete-intersection approximation). Let Y/KY/K be smooth projective of dimension n≥2n\geq2. For 1≤i<n1\leq i<n, let pip_i be a uniform FS limit on an ample rational line bundle LiL_i. Then T=⋀i=1n−1c1(Li,pi)T=\bigwedge_{i=1}^{n-1}c_1(L_i,p_i) is a weak limit, on all compact smooth tests, of individual normalized smooth complete-intersection curve cycles TaT_a defined over KK. For every line bundle HH, each TaT_a has normalized HH-degree deg⁡K(HL1⋯Ln−1)\deg_K(HL_1\cdots L_{n-1}).

Proof. Step 1: a smooth cut with prescribed coefficient valuations. Let Z⊂YZ\subset Y be smooth projective of pure dimension d≥1d\geq1. First consider an ample line bundle MM generated by s0,…,sbs_0,\ldots,s_b, with FS weight q=max⁡ilog⁡∣si∣q=\max_i\log|s_i|. For every N≥1N\geq1, choose

ai,N∈πi(1+πR),σN=∑i=0bai,NsiNa_{i,N}\in\pi^i(1+\pi R),\qquad \sigma_N=\sum_{i=0}^b a_{i,N}s_i^N

so that DN=divZ(σN)D_N=\mathop{\mathrm{div}}_Z(\sigma_N) is smooth. Such a choice exists: the pure powers siNs_i^N have no common zero, and characteristic-zero Bertini gives a nonempty Zariski-open smooth locus in their coefficient space. The product ∏iπi(1+πR)\prod_i\pi^i(1+\pi R) is Zariski dense, since a nonzero polynomial cannot vanish on a product of infinite subsets of KK. If ZZ has several components, impose the finitely many smoothness conditions simultaneously. These choices are made independently of all test forms.

Step 2: ties have zero mass for each smooth test. Fix a smooth (d−1,d−1)(d-1,d-1)-test η\eta on ZanZ^\mathrm{an}, and let νη\nu_\eta be the signed measure defined by d′d′′ηd'd''\eta. On each component, discard the generators that vanish identically, retaining the original indices ii. By [3, Lemma 3.2.5], the compact support of d′d′′ηd'd''\eta avoids the zero divisors of the remaining generators. Cover it by finitely many compact moment charts, refined by the unit ratios si/sjs_i/s_j. On each cell of the resulting finite polyhedral complexes, the functions ℓij=log⁡∣si/sj∣\ell_{ij}=\log|s_i/s_j| are affine.

The absolute-value tropical integration formula [3, Section 3.7 and Proposition 3.7.2] implies that ∣νη∣|\nu_\eta| gives zero mass to inverse images of proper affine hyperplanes in the dd-dimensional cells; lower-dimensional cells carry no top-degree mass. Equivalently, on each compact chart one can dominate both signs of its Lagerberg representative by a positive top form and apply [4, Proposition 11.1(3)]. This is a statement about total variation, not just about the signed pushforward.

Since ∣ai,N∣=∣π∣i|a_{i,N}|=|\pi|^i, two summands in (eq:deterministic-section) have the same norm exactly when Nℓij=(i−j)λ.N\ell_{ij}=(i-j)\lambda. If ℓij\ell_{ij} is nonconstant on a top-dimensional cell, this is a null hyperplane. If it is the constant cc, the equality is impossible for c=0c=0 and holds for at most one positive integer NN when c≠0c\ne0. There are finitely many cells and pairs. Thus, for all sufficiently large NN, the largest summand is unique at ∣νη∣|\nu_\eta|-almost every point.

Step 3: the one-cut estimate. Put wN=Nq−log⁡∣σN∣≥0w_N=Nq-\log|\sigma_N|\geq0. Unique dominance in the ultrametric inequality gives log⁡∣σN∣=max⁡i{Nlog⁡∣si∣−iλ},0≤wN≤bλ∣νη∣-almost everywhere\log|\sigma_N| =\max_i\{N\log|s_i|-i\lambda\},\qquad 0\leq w_N\leq b\lambda \quad |\nu_\eta|\text{-almost everywhere} for all sufficiently large NN. Poincaré–Lelong therefore gives

∣⟨[DN]N−c1(M,q)∧[Z],η⟩∣=1N∣∫ZanwN d′d′′η∣≤bλN∥d′d′′η∥TV.\begin{aligned} \left|\left\langle \frac{[D_N]}N-c_1(M,q)\wedge[Z],\eta \right\rangle\right| &=\frac1N\left|\int_{Z^\mathrm{an}}w_N\,d'd''\eta\right|\\ &\leq\frac{b\lambda}{N}\|d'd''\eta\|_{\mathrm{TV}}. \end{aligned}

Here the threshold in NN depends on η\eta, but the chosen sequence (σN)(\sigma_N) does not. The formula is interpreted intrinsically on ZZ, or after pushforward to YY. It proves one-cut convergence on every original smooth test. The bound is a statement about integration against d′d′′ηd'd''\eta: wNw_N is unbounded along DND_N. The argument does not give the same estimate against an arbitrary positive background current, whose coefficient measures may charge the tie loci.

Step 4: successive cuts and uniform limits. For pure-FS metrics pip_i, let the generating tuple lie in diLid_iL_i, and put αi=c1(Li,pi)\alpha_i=c_1(L_i,p_i). At the ii-th cut, restrict that tuple to the preceding smooth intersection and apply Steps 1–3. Its divisor has class NidiLiN_id_iL_i, and its current is divided by NidiN_id_i. The Bertini conditions can also be imposed on the ambient divisors.

For clarity, let Ck\mathcal C_k be the currents of these smooth codimension-kk complete intersections, normalized by ∏i=1kNidi\prod_{i=1}^kN_id_i, with C0={[Y]}\mathcal C_0=\{[Y]\}. The one-cut limit places αk∧S\alpha_k\wedge S in the weak closure of Ck\mathcal C_k for every S∈Ck−1S\in\mathcal C_{k-1}. Continuity in the positive background (Proposition “A fixed approachable potential”) then gives inductively α1∧⋯∧αk∈Ck‾ weak.\alpha_1\wedge\cdots\wedge\alpha_k \in\overline{\mathcal C_k}^{\,\mathrm{weak}}. Indexing a weak neighbourhood by finitely many smooth tests and an error tolerance gives the asserted net of individual curves. Each member has normalized HH-degree deg⁡K ⁣(H∏i=1n−1(NidiLi))∏i=1n−1Nidi=deg⁡K(HL1⋯Ln−1).\frac{\deg_K\!\left(H\prod_{i=1}^{n-1}(N_id_iL_i)\right)} {\prod_{i=1}^{n-1}N_id_i} =\deg_K(HL_1\cdots L_{n-1}). Finally, approximate the pip_i uniformly by pure-FS metrics and take the union of the corresponding families Cn−1\mathcal C_{n-1} over all approximating tuples and tensor degrees. Every member is still an individual normalized smooth complete intersection with the displayed degree. Joint continuity in Proposition “A fixed approachable potential” puts the limiting mixed current in the weak closure of this union. ◻

Concentration on a divisor

The next limit uses only continuity in the positive background. In particular, the other metrics remain fixed while the truncated section metric tends to its divisor current.

Lemma (Concentration on a smooth divisor). Let Y/KY/K be smooth projective of dimension n≥1n\geq1, and let ss define a smooth effective Cartier divisor i:D↪Yi:D\hookrightarrow Y in a line bundle MM. Let q,p1,…,pn−1q,p_1,\ldots,p_{n-1} be globally smoothly approximable metrics on M,L1,…,Ln−1M,L_1,\ldots,L_{n-1}, respectively. Then

c1 ⁣(M,max⁡{q,log⁡∣s∣+N})∧⋀j=1n−1c1(Lj,pj)⟶i∗ ⁣(⋀j=1n−1c1(Lj∣D,pj∣D))\begin{aligned} &c_1\!\left(M,\max\{q,\log|s|+N\}\right) \wedge\bigwedge_{j=1}^{n-1}c_1(L_j,p_j)\\ &\hspace{2em}\longrightarrow i_*\!\left(\bigwedge_{j=1}^{n-1} c_1(L_j|_D,p_j|_D)\right) \end{aligned}

as N→∞N\to\infty, on every continuous scalar test on YanY^\mathrm{an}.

Proof. Put ρN=max⁡{q,log⁡∣s∣+N}\rho_N=\max\{q,\log|s|+N\}, which is smoothly approachable by Lemma “Smooth approximation of FS metrics”. In a local frame, ρN−N=max⁡{q−N,log⁡∣s∣}↘log⁡∣s∣\rho_N-N=\max\{q-N,\log|s|\}\searrow\log|s|. Local integrability and Poincaré–Lelong give c1(M,ρN)→[D]c_1(M,\rho_N)\to[D]. Apply the fixed-potential continuity of Proposition “A fixed approachable potential” in the remaining slots. The projection formula for approachable products [10, Proposition 8.7.5] identifies the limit with the right-hand side of (eq:divisor-limit). All measures are positive and have the common mass deg⁡K(ML1⋯Ln−1)\deg_K(ML_1\cdots L_{n-1}), by Lemma “Degree and comparison of measures”. Uniform smooth density extends convergence to continuous tests. ◻

Contact rigidity

The following criterion compares two metrics on the same ample line. Only the lower metric is assumed to be a uniform FS limit; the upper metric is tested on algebraic curves. The proof uses the preceding curve approximation and divisor limit, independently of envelope continuity.

Theorem (Contact rigidity). Let Y/KY/K be smooth, projective, and integral of dimension n≥1n\geq1, and let LL be ample. Let pp be a uniform FS limit on LL, and let rr be a finite continuous metric on LL whose pullback by every morphism from a smooth algebraic KK-curve has positive Chern current. If g:=r−p≥0,∫Yang c1(L,p)n=0,g:=r-p\geq0, \qquad \int_{Y^\mathrm{an}}g\,c_1(L,p)^n=0, then r=pr=p.

Proof. Step 1: pass from curves to mixed measures. Put α=c1(L,p),T=αn−1,μ=αn,D=deg⁡K(Ln)>0.\alpha=c_1(L,p),\qquad T=\alpha^{n-1},\qquad \mu=\alpha^n,\qquad D=\deg_K(L^n)>0. For n>1n>1, approximate TT by the normalized smooth curve cycles TaT_a of Proposition “Complete-intersection approximation”; for n=1n=1, take Ta=[Y]T_a=[Y]. The pullbacks of rr and pp to each curve have positive Chern measures. Their normalized pushforwards σa,μa\sigma_a,\mu_a have mass DD, and νa=d′d′′(gTa)=σa−μa,∥νa∥TV≤2D.\nu_a=d'd''(gT_a)=\sigma_a-\mu_a, \qquad\|\nu_a\|_{\mathrm{TV}}\leq2D. By Proposition “A fixed approachable potential”, μa→μ\mu_a\to\mu. Lemma “Continuous coefficients” gives νa→ν:=d′d′′(gT)\nu_a\to\nu:=d'd''(gT) on smooth tests. The variation bounds extend these limits to continuous scalar tests. Thus σa→σ:=ν+μ\sigma_a\to\sigma:=\nu+\mu, where σ\sigma is a positive measure of mass DD.

Step 2: use zero energy to annihilate the mixed Laplacian. We pass the curve Green identities of Lemma “Green identities on curves” to the limit. For a fixed smooth function hh, Lemma “Continuous coefficients” bounds the total variations of d′d′′h∧Tad'd''h\wedge T_a: a finite collection of local positive test evaluations bounds the fixed coefficient form d′d′′hd'd''h. Consequently,

∫g d′d′′(hT)=∫h dν,−∫(g+th)d′d′′((g+th)T)≥0(t∈R).\int g\,d'd''(hT)=\int h\,d\nu, \qquad -\int(g+th)d'd''((g+th)T)\geq0\quad(t\in\mathbf R).

All integrals are over YanY^\mathrm{an}. The contact assumption implies 0≤−∫g dν=∫g dμ−∫g dσ≤0.0\leq-\int g\,d\nu =\int g\,d\mu-\int g\,d\sigma\leq0. The quadratic polynomial in tt from (eq:contact-energy) therefore has zero constant term. Green symmetry writes its linear term as −2t∫h dν-2t\int h\,d\nu. Nonnegativity for every t∈Rt\in\mathbf R forces that coefficient to vanish. Since this holds for every smooth hh, we have ν=0\nu=0.

If ρ\rho is any finite continuous curve-positive metric on mLmL, set b=ρ−mpb=\rho-mp. Its normalized curve Laplacians have variations at most 2mD2mD. The same passage of Green symmetry, now for the two continuous gaps gg and bb, gives

∫g d′d′′(bT)=∫b dν=0.\int g\,d'd''(bT)=\int b\,d\nu=0.

Only products with the approachable lower metric pp have been used; the upper metric rr enters through its curve restrictions.

Step 3: restrict the contact identity and argue by dimension. When n=1n=1, Lemma “Green identities on curves” and the zero energy already proved make gg constant. Its integral against the positive measure μ\mu of mass DD makes that constant zero. Suppose n>1n>1 and proceed by induction. Fix a closed point x∈Yx\in Y. For sufficiently large mm, choose a smooth H=div(s)H=\mathop{\mathrm{div}}(s) in mLmL through xx. To impose this condition, use H0(Y,mL⊗Ix)H^0(Y,mL\otimes\mathcal I_x): for m≫0m\gg0 it is generated away from xx and realizes arbitrary first jets at xx. A nonzero first jet ensures smoothness there, and characteristic-zero Bertini ensures smoothness elsewhere. The nonempty open conditions admit a KK-rational choice since KK is infinite.

Set ρN=max⁡{mp,log⁡∣s∣+N}\rho_N=\max\{mp,\log|s|+N\}. These metrics are smoothly approximable by Lemma “Smooth approximation of FS metrics”, hence curve-positive. Equation (eq:contact-green), together with ∫g dμ=0\int g\,d\mu=0, yields ∫Yang c1(mL,ρN)∧αn−1=0.\int_{Y^\mathrm{an}}g\,c_1(mL,\rho_N)\wedge\alpha^{n-1}=0. Lemma “Concentration on a smooth divisor” now gives ∫Hang∣H c1(L∣H,p∣H)n−1=0.\int_{H^\mathrm{an}}g|_H\,c_1(L|_H,p|_H)^{n-1}=0. The integrand is nonnegative. It therefore has zero integral on each smooth integral component of HH. The curve condition passes to that component by composition, L∣HL|_H is ample, and p∣Hp|_H is a uniform FS limit by restriction of the generating tuples. Induction gives g(x)=0g(x)=0. Closed points give a dense set of rigid points of YanY^\mathrm{an}; continuity implies g=0g=0 everywhere. ◻

Envelopes on a fixed regular model

For an ample line bundle LL on a projective KK-variety and a continuous metric rr, set

PL(r)=sup⁡{q≤r:q is a continuous semipositive metric on L}.P_L(r)=\sup\{q\le r:q\text{ is a continuous semipositive metric on }L\}.

We use the standard finiteness, translation, scaling, and monotonicity properties of this envelope [8, Remark 6.2 and Proposition 6.3]. In particular,

∥PL(r)−PL(r′)∥∞≤∥r−r′∥∞.\left\lVert P_L(r)-P_L(r')\right\rVert_\infty\le \left\lVert r-r'\right\rVert_\infty.

The envelope is not assumed continuous at this stage.

Test ideals of admissible section divisors

Let X/R\mathscr X/R be regular, integral, projective and flat, with smooth generic fibre YY of dimension n≥1n\ge1. Extend LL to a line bundle L\mathscr L on X\mathscr X, and write θ\theta for its model metric. Such an extension exists because the closure of a generic Cartier divisor is Cartier on the regular scheme X\mathscr X. After clearing a fixed tensor denominator and adjusting the reference identification by a power of π\pi, a model metric rr can be written

r=θ+log⁡∣E∣,E=OW(−F)⊆OW,r=\theta+\log|\mathscr E|, \qquad \mathscr E=\mathcal O_{\mathscr W}(-F)\subseteq\mathcal O_{\mathscr W},

where f:W→Xf:\mathscr W\to\mathscr X is a normal projective model inducing the identity on YY, and FF is an effective vertical Cartier divisor. Put p=PL(r)p=P_L(r).

Proposition (The test-ideal estimate). There are coherent vertical ideals jm⊆OX\mathfrak j_m\subseteq\mathcal O_{\mathscr X}, for m≥1m\ge1, and an integer B≥0B\ge0, independent of mm, such that

πBjmOW⊆Em,p≤θ+ψm≤r+Bλm,ψm=1mlog⁡∣jm∣.\pi^B\mathfrak j_m\mathcal O_{\mathscr W}\subseteq\mathscr E^m,\qquad p\le\theta+\psi_m\le r+\frac{B\lambda}{m}, \qquad \psi_m=\frac1m\log|\mathfrak j_m|.

For every globally generated ample Cartier divisor HH on X\mathscr X,

jmL⊗m⊗OX(KX+(n+1)H)is globally generated.\mathfrak j_m\mathscr L^{\otimes m}\otimes \mathcal O_{\mathscr X}\bigl(K_{\mathscr X}+(n+1)H\bigr) \quad\text{is globally generated.}

Proof. For l≥1l\ge1, let Vl=H0(W,f∗L⊗l⊗El).V_l=H^0\bigl(\mathscr W,f^*\mathscr L^{\otimes l}\otimes\mathscr E^l\bigr). Since f∗OW=OXf_*\mathcal O_{\mathscr W}=\mathcal O_{\mathscr X}, every member of VlV_l is a section of L⊗l\mathscr L^{\otimes l} on X\mathscr X. Normality also gives

Vl={s∈H0(Y,lL):∥s∥lr,sup⁡≤1}.V_l=\{s\in H^0(Y,lL):\left\lVert s\right\rVert_{lr,\sup}\le1\}.

Indeed, a generic section has no horizontal poles, while the norm inequalities at vertical divisorial points are exactly the remaining codimension-one extension conditions on W\mathscr W. For 0≠s∈Vl0\ne s\in V_l, denote its effective divisor on X\mathscr X by DsD_s. Then Ds∼lL,f∗Ds≥lF.D_s\sim l\mathscr L,\qquad f^*D_s\ge lF. Define

jm=∑l≥1 ∑0≠s∈VlτX ⁣(mlDs).\mathfrak j_m =\sum_{l\ge1}\ \sum_{0\ne s\in V_l} \tau_{\mathscr X}\!\left(\frac mlD_s\right).

This sum equals a finite subsum by Noetherianity. For a sufficiently large ll, a generating tuple of H0(Y,lL)H^0(Y,lL) can be scaled into VlV_l. At every point of YY one of these sections is nonvanishing. Generic-fibre comparison with multiplier ideals [6, Corollary 7.21(e)] therefore shows that jm∣Y=OY\mathfrak j_m|_Y=\mathcal O_Y. Thus jm\mathfrak j_m is vertical.

Lower bound. Write τ0=τX(0)\tau_0=\tau_{\mathscr X}(0). This fixed ideal contains a power of π\pi, since its generic restriction is OY\mathcal O_Y. Cartier translation and subadditivity on the regular scheme X\mathscr X give, for every k≥1k\ge1,

OX(−kmDs)τ0=τX(kmDs)⊆τX ⁣(mlDs)kl⊆jmkl.\mathcal O_{\mathscr X}(-kmD_s)\tau_0 =\tau_{\mathscr X}(kmD_s) \subseteq \tau_{\mathscr X}\!\left(\frac mlD_s\right)^{kl} \subseteq\mathfrak j_m^{kl}.

Here we use [6, Proposition 7.16(b) and Theorem 7.22(e)]. Taking logarithmic norms, dividing by kmlkml, and letting k→∞k\to\infty yields 1llog⁡∥s∥lθ≤ψm.\frac1l\log\left\lVert s\right\rVert_{l\theta}\le\psi_m. If a pure-FS metric qq lies below rr, its defining sections belong to the corresponding VlV_l by (eq:weighted-unit-ball); taking their maximum gives q≤θ+ψmq\le\theta+\psi_m. Every continuous semipositive competitor below rr is a uniform limit of pure-FS metrics shifted down by constants in λQ\lambda\mathbf Q tending to zero [7, Corollary 7.6 and Theorem 7.8]. The lower bound in (eq:test-ideal-bounds) follows.

Upper bound. Choose compatible canonical divisors and set ΔW=f∗KX−KW\Delta_{\mathscr W}=f^*K_{\mathscr X}-K_{\mathscr W}. The fractional test ideal τW(ΔW)\tau_{\mathscr W}(\Delta_{\mathscr W}) is defined because KW+ΔWK_{\mathscr W}+\Delta_{\mathscr W} is Cartier [6, Definition 7.20]. Birational inclusion, monotonicity, and Cartier translation give

τX ⁣(mlDs)OW⊆τW ⁣(ΔW+mlf∗Ds)⊆τW(ΔW+mF)=EmτW(ΔW).\begin{aligned} \tau_{\mathscr X}\!\left(\frac mlD_s\right)\mathcal O_{\mathscr W} &\subseteq \tau_{\mathscr W}\!\left(\Delta_{\mathscr W}+\frac ml f^*D_s\right) \notag\\ &\subseteq \tau_{\mathscr W}(\Delta_{\mathscr W}+mF) =\mathscr E^m\tau_{\mathscr W}(\Delta_{\mathscr W}). \end{aligned}

The cited properties are [6, Proposition 7.16(a),(b) and Theorem 7.22(d)]. The fixed fractional ideal on the right is coherent and restricts to OY\mathcal O_Y. A single power πB\pi^B clears its denominators on a finite affine cover of W\mathscr W. Summing (eq:birational-bound) gives πBjmOW⊆Em\pi^B\mathfrak j_m\mathcal O_{\mathscr W}\subseteq\mathscr E^m, and hence the upper bound.

Global generation. Each effective boundary Δ=(m/l)Ds\Delta=(m/l)D_s is Q\mathbf Q-linearly equivalent to mLm\mathscr L. For M=mL+KX+HM=m\mathscr L+K_{\mathscr X}+H, one has M−KX−Δ∼QHM-K_{\mathscr X}-\Delta\sim_{\mathbf Q}H. Since the special fibre has dimension nn, [6, Corollary 7.24] makes τX(Δ)L⊗m(KX+(n+1)H)\tau_{\mathscr X}(\Delta)\mathscr L^{\otimes m}(K_{\mathscr X}+(n+1)H) globally generated. A finite subsum in (eq:divisor-sum) proves (eq:test-ideal-generation). ◻

Uniform approximation and orthogonality

Theorem (Envelopes on a regular model). Let X/R\mathscr X/R be regular, integral, projective and flat, with smooth generic fibre YY of dimension nn, and let LL be an ample rational line bundle on YY. For every finite continuous metric rr on LL, the envelope p=PL(r)p=P_L(r) is continuous, is a uniform limit of pure-FS metrics, and satisfies

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

Here the measure is the CLD product. For a model obstacle, after clearing fixed tensor denominators, there are pure-FS metrics qℓq_\ell, for all sufficiently large integers ℓ\ell, such that

0≤qℓ−p≤Cℓ,0\le q_\ell-p\le \frac C\ell,

where CC depends on the obstacle and the fixed auxiliary data.

Proof. If n=0n=0, then Y=Spec⁡FY=\operatorname{Spec}F for a finite extension F/KF/K, and YanY^\mathrm{an} is a point. In a fixed frame, weights of pure-FS metrics include λQ\lambda\mathbf Q, so every continuous metric is semipositive and p=rp=r. A model metric is pure-FS after clearing its fixed tensor denominator, which also gives the stated approximation. Assume henceforth that n≥1n\ge1.

First treat a model obstacle in the form (eq:model-obstacle). Choose a globally generated ample divisor HH sufficiently positive that A=OX(KX+(n+1)H)\mathscr A=\mathcal O_{\mathscr X}\bigl(K_{\mathscr X}+(n+1)H\bigr) is ample and globally generated. Put A=A∣YA=\mathscr A|_Y, and let qAq_A be its model FS metric. Choose k≥1k\ge1 such that kL−AkL-A is very ample and a pure-FS metric ν\nu defined by a generating tuple on that line bundle. Multiplying this tuple by a sufficiently large negative power of π\pi, arrange that h:=qA+νk≥r.h:=\frac{q_A+\nu}{k}\ge r. By Proposition “The test-ideal estimate”, the generators of jmL⊗mA\mathfrak j_m\mathscr L^{\otimes m}\mathscr A define a pure-FS metric tm=mθ+qA+log⁡∣jm∣on mL+A.t_m=m\theta+q_A+\log|\mathfrak j_m| \quad\text{on }mL+A. Tensoring with the tuple defining ν\nu gives a pure-FS metric on LL, hm=tm+νm+k=mm+k(θ+ψm)+km+kh.h_m=\frac{t_m+\nu}{m+k} =\frac m{m+k}(\theta+\psi_m)+\frac k{m+k}h. Consequently, p≤hm≤r+Cm+k,C=Bλ+k∥h−r∥∞.p\le h_m\le r+\frac C{m+k}, \qquad C=B\lambda+k\left\lVert h-r\right\rVert_\infty. The semipositive metric hm−C/(m+k)h_m-C/(m+k) lies below rr, hence below pp. Thus 0≤hm−p≤Cm+k.0\le h_m-p\le\frac C{m+k}. These metrics are represented by generating sections in degree m+km+k. Relabeling ℓ=m+k\ell=m+k proves (eq:regular-model-approximation). A fixed tensor denominator is divided out using PdL(dr)=dPL(r)P_{dL}(dr)=dP_L(r).

For a continuous obstacle, use uniform density of model metrics [7, Proposition 5.16(vi) and Theorem 5.18]. Equation (eq:envelope-lipschitz) shows that their envelopes converge uniformly to PL(r)P_L(r), and a diagonal choice gives pure-FS approximation. Finally, YY is geometrically reduced, so [8, Theorem 6.7] gives (eq:regular-model-orthogonality). Lemma “Degree and comparison of measures” identifies its measure with the CLD product. ◻

Corollary (Projective space). Let a,b≥1a,b\ge1 and L=OPKa(b)L=\mathcal O_{\mathbf P^a_K}(b). The conclusions of Theorem “Envelopes on a regular model” hold for every continuous metric on LL. For a model obstacle represented as in (eq:model-obstacle) on a model above PRa\mathbf P^a_R, the generators of jmO(bm)\mathfrak j_m\mathcal O(bm) give pure-FS metrics qmq_m satisfying 0≤qm−PL(r)≤Bλ/m.0\le q_m-P_L(r)\le B\lambda/m.

Proof. Take X=PRa\mathscr X=\mathbf P^a_R. For the more precise assertion, choose the hyperplane divisor HH; then KX+(a+1)H=0K_{\mathscr X}+(a+1)H=0. Equation (eq:test-ideal-generation) makes qm=θ+m−1log⁡∣jm∣q_m=\theta+m^{-1}\log|\mathfrak j_m| pure-FS. By (eq:test-ideal-bounds), p≤qm≤r+Bλ/mp\le q_m\le r+B\lambda/m; the envelope property applied to qm−Bλ/mq_m-B\lambda/m finishes the proof. ◻

Proof of local regularization

The next lemma places a local curve-positive function inside one global metric on projective space. It uses only algebraic sections and the maximum property on curves.

A projective extension

Lemma (Projective extension). Let W⊂AKa,anW\subset\mathbf A_K^{a,\mathrm{an}} be an ordinary open, a≥1a\ge1, and let v:W→Rv:W\to\mathbf R be finite, continuous and curve-positive. For every x∈Wx\in W, there are d≥1d\ge1, a finite continuous curve-positive metric rr on OPKa(d)\mathcal O_{\mathbf P^a_K}(d), a section s0s_0 nonvanishing near xx, and a constant c∈Rc\in\mathbf R, such that −log⁡∥s0∥r=v+con an ordinary neighbourhood of x.-\log\left\lVert s_0\right\rVert_r=v+c \quad\text{on an ordinary neighbourhood of }x.

Proof. Set c=λ/4−v(x)c=\lambda/4-v(x). After shrinking WW, continuity gives

0<v+c<λ/2on W.0<v+c<\lambda/2\quad\text{on }W.

Choose a compact polynomial norm box C={∣Fi∣≤∣ci∣, ∣Gj∣≥∣dj∣}⊂WC=\{|F_i|\le|c_i|,\ |G_j|\ge|d_j|\}\subset W with all defining inequalities strict at xx, including upper bounds on the affine coordinates. Here Fi,Gj∈K[z1,…,za]F_i,G_j\in K[z_1,\ldots,z_a] and ci,dj∈K×c_i,d_j\in K^\times. Such boxes form a neighbourhood basis: choose thresholds in ∣K×∣Q|K^\times|^{\mathbf Q} inside an ordinary basic neighbourhood and encode them by powers of the polynomials.

For N≥1N\ge1 sufficiently large, every logarithmic term in the following maximum is negative on a neighbourhood of xx:

b0=max⁡{0, log⁡∣FiNπciN∣,log⁡∣djNπGjN∣}b_0=\max\left\{ 0,\ \log\left|\frac{F_i^N}{\pi c_i^N}\right|, \log\left|\frac{d_j^N}{\pi G_j^N}\right| \right\}

On the open where every GjG_j is nonzero, this function satisfies b0=0b_0=0 near xx and {b0≤λ}=C\{b_0\le\lambda\}=C.

Put t0=∏jGjNt_0=\prod_jG_j^N, with the empty product equal to 11. Clear the denominators in (eq:collar-weight) and homogenize t0,t0FiNπciN,t0djNπGjNt_0,\qquad \frac{t_0F_i^N}{\pi c_i^N},\qquad \frac{t_0d_j^N}{\pi G_j^N} to a common degree dd, chosen strictly larger than deg⁡t0\deg t_0. Write s0s_0 for the homogenization of t0t_0. Add sufficiently small KK-multiples of Z0d,…,ZadZ_0^d,\ldots,Z_a^d, so that their ratios to s0s_0 have norm less than 11 on CC. The resulting tuple generates O(d)\mathcal O(d); let qq be its FS metric. Then b:=−log⁡∥s0∥q∈[0,+∞]b:=-\log\left\lVert s_0\right\rVert_q\in[0,+\infty] equals b0b_0 on CC, vanishes near xx, and satisfies

{b≤λ}=C.\{b\le\lambda\}=C.

Indeed, s0s_0 vanishes at infinity because d>deg⁡t0d>\deg t_0, and the remaining original ratios impose precisely the inequalities defining CC. The added sections preserve those inequalities on CC. We interpret b=+∞b=+\infty on the zero divisor of s0s_0.

The ordinary opens U={b<λ},V={b>λ/2}U=\{b<\lambda\},\qquad V=\{b>\lambda/2\} cover PKa,an\mathbf P_K^{a,\mathrm{an}}, and U⊂WU\subset W. On UU, define a metric by its s0s_0-weight max⁡{v+c,b}\max\{v+c,b\}; on VV, use qq. The definitions agree on U∩VU\cap V by (eq:normalization). They therefore give a finite continuous metric rr. Near xx, one has b=0<v+cb=0<v+c, so its weight is v+cv+c.

On every smooth algebraic KK-curve, the weights just used are continuous subharmonic functions or maxima of two such functions. Their maxima are subharmonic by the curve regularization of Proposition “Curve regularization” and the regularized maximum construction. Thus rr is curve-positive. ◻

Completion of the proof

Proof of Theorem “Local regularization”. Local uniform convergence of smooth positive potentials implies current positivity by testing against d′d′′ηd'd''\eta. Conversely, fix x∈Ox\in O on a positive-dimensional component of XX. Proposition “Curve maps and extension” supplies an ordinary ambient open W⊂AKa,anW\subset\mathbf A_K^{a,\mathrm{an}} and a finite continuous curve-positive function vv agreeing with uu along Xan∩WX^{\mathrm{an}}\cap W. Apply Lemma “Projective extension”, and retain its r,s0,cr,s_0,c.

By Corollary “Projective space”, the envelope p=PO(d)(r)p=P_{\mathcal O(d)}(r) is a uniform FS limit and p≤r,∫PKa,an(r−p)c1(O(d),p)a=0.p\le r,\qquad \int_{\mathbf P_K^{a,\mathrm{an}}}(r-p)c_1(\mathcal O(d),p)^a=0. Theorem “Contact rigidity” gives p=rp=r. By Lemma “Smooth approximation of FS metrics”, choose global smooth positive metrics rjr_j with ∥rj−r∥∞≤4−j\left\lVert r_j-r\right\rVert_\infty\le4^{-j}. On one fixed ordinary neighbourhood VxV_x of xx in XanX^{\mathrm{an}} where the weight identity holds, put uj=−log⁡∥s0∥rj∣Vx−c.u_j=-\log\left\lVert s_0\right\rVert_{r_j}\big|_{V_x}-c. These functions are smooth, have positive curvature, and satisfy ∥uj−u∥Vx≤4−j\left\lVert u_j-u\right\rVert_{V_x}\le4^{-j}. For the decreasing assertion, define wj=uj+2⋅4−j.w_j=u_j+2\cdot4^{-j}. Then u+4−j≤wj≤u+3⋅4−j,wj+1≤u+34 4−j<wj.u+4^{-j}\le w_j\le u+3\cdot4^{-j}, \qquad w_{j+1}\le u+\tfrac34\,4^{-j}<w_j. Thus wj↓uw_j\downarrow u uniformly on the same VxV_x. Constant translations preserve curvature. On a zero-dimensional component, every finite continuous function is locally constant and already smooth. ◻

Consequences for continuous potentials

We first apply local regularization to the operation constructed in Proposition “A fixed approachable potential”. The background current remains arbitrary; in particular, it need not itself be a product of curvature currents.

Proof of the mixed-current theorem

Proof of Theorem “Continuous potentials on a positive closed background”. Theorem “Local regularization” places every uiu_i in Ploc(O)\mathcal P_{\mathrm{loc}}(O). Iterating Proposition “A fixed approachable potential” in (eq:recursive-product) proves positivity, closedness, symmetry, and joint continuity, including nets. Locality, additivity, and positive homogeneity follow from the same recursion. A positive current of top degree is a Radon measure.

It remains to prove independence of the potentials. Let hh be finite continuous with d′d′′[h]=0d'd''[h]=0, and let TT be any positive closed current for which the bidegrees permit the operation. Both hh and −h-h have positive curvature currents, so Theorem “Local regularization” and Proposition “A fixed approachable potential” give d′d′′(hT)≥0,−d′d′′(hT)=d′d′′((−h)T)≥0.d'd''(hT)\geq0,\qquad -d'd''(hT)=d'd''((-h)T)\geq0. The positive cone is pointed by Lemma “Continuous coefficients”; hence

d′d′′[h]=0⟹d′d′′(hT)=0.d'd''[h]=0\quad\Longrightarrow\quad d'd''(hT)=0.

If d′d′′[w]=d′d′′[ui]d'd''[w]=d'd''[u_i], apply this implication to h=w−uih=w-u_i and to the positive closed product of the other slots. Symmetry permits us to place the changed slot first, proving the assertion. ◻

Chern products and closure operations

Corollary (Continuous Chern products). Let X/KX/K be smooth quasi-projective of pure dimension nn, let O⊂XanO\subset X^\mathrm{an} be an ordinary open, and let SS be a positive closed CLD current of bidegree (p,p)(p,p) on OO. Let L‾1,…,L‾m\overline L_1,\ldots,\overline L_m be continuously metrized analytic line bundles on OO, with finite local weights and positive CLD curvature currents, where m≤n−pm\leq n-p. Their local products glue to a positive closed current c1(L‾1)∧⋯∧c1(L‾m)∧S.c_1(\overline L_1)\wedge\cdots\wedge c_1(\overline L_m)\wedge S. It depends only on the curvature currents and SS, agrees with the smooth wedge product when all weights are smooth, and is a positive Radon measure if m+p=nm+p=n. The same assertions hold for rational line bundles after clearing and dividing their tensor denominators.

Proof. Apply Theorem “Continuous potentials on a positive closed background” in local frames. On overlaps, weights differ by logarithms of analytic units, whose curvature is zero by Poincaré–Lelong. Equation (eq:continuous-kernel) identifies the products. The remaining assertions, including rational normalization, follow from the same theorem. ◻

Corollary (Products with a smooth subvariety). Let i:Z↪Xi:Z\hookrightarrow X be a closed immersion of smooth quasi-projective KK-varieties, with XX of pure dimension nn and ZZ of pure dimension dd. Let O⊂XanO\subset X^\mathrm{an} be ordinary open, put OZ=(ian)−1(O)O_Z=(i^\mathrm{an})^{-1}(O), and let u1,…,umu_1,\ldots,u_m be finite continuous functions on OO with d′d′′[uj]≥0d'd''[u_j]\geq0, where m≤dm\leq d. Write [Z]O=i∗[OZ][Z]_O=i_*[O_Z] for the integration current on OO, and write u~j=uj∣OZ\widetilde u_j=u_j|_{O_Z}. Then

B[Z]O(u1,…,um)=i∗B[OZ](u~1,…,u~m).\mathcal B_{[Z]_O}(u_1,\ldots,u_m) =i_*\mathcal B_{[O_Z]}(\widetilde u_1,\ldots,\widetilde u_m).

Proof. Positive smooth approximants from Theorem “Local regularization” restrict to positive smooth approximants on OZO_Z. The assertion is then the projection formula for approachable mixed products [10, Proposition 8.7.5], applied to the closed immersion ii. ◻

Formula (eq:subvariety-projection) identifies a product with the integration current of ZZ with the intrinsic product on ZZ. It does not require a pullback operation for arbitrary currents.

Corollary (Closure operations). Let u1,…,uru_1,\ldots,u_r be finite continuous current-positive functions on an ordinary open O⊂XanO\subset X^\mathrm{an}, where X/KX/K is smooth quasi-projective. If F:Rr→RF:\mathbf R^r\to\mathbf R is convex and nondecreasing in each coordinate, then F(u1,…,ur)F(u_1,\ldots,u_r) is current-positive. In particular, finite maxima and sums with nonnegative coefficients preserve this class.

If V⊂YanV\subset Y^\mathrm{an} is an ordinary open in a smooth quasi-projective KK-variety and f:V→Of:V\to O is an analytic morphism, then ui∘fu_i\circ f is current-positive on VV.

Proof. By Theorem “Local regularization”, the uiu_i are locally uniformly approachable by smooth positive potentials. The convex composition assertion is therefore [10, Corollary 8.2.4]. For the pullback assertion, pull back those smooth approximants; smooth pullback preserves positivity, and locally uniform convergence is preserved because continuous maps send compact sets to compact sets. ◻

The projective curve criterion

Corollary (A projective curve criterion). Let Y/R\mathscr Y/R be a regular integral projective flat model with smooth generic fibre YY of positive dimension. Let LL be an ample line bundle on YY, and let rr be a finite continuous metric on LL. The following conditions are equivalent:

  1. For every algebraic morphism f:C→Yf:C\to Y from a smooth KK-curve, the pulled-back metric f∗rf^*r has positive Chern current.

  2. The original CLD current c1(L,r)c_1(L,r) is positive.

  3. The metric rr is semipositive, equivalently a uniform limit of pure Fubini–Study metrics on LL.

  4. The metric rr is a uniform limit of global smooth metrics on LL with positive curvature forms.

The criterion also holds for an ample rational line bundle by clearing its tensor denominator.

Proof. Suppose the first condition holds. Theorem “Envelopes on a regular model” gives a finite continuous semipositive envelope p=PL(r)p=P_L(r), which is a uniform Fubini–Study limit and satisfies p≤r,∫Yan(r−p)c1(L,p)dim⁡Y=0.p\leq r,\qquad \int_{Y^\mathrm{an}}(r-p)c_1(L,p)^{\dim Y}=0. Theorem “Contact rigidity” applies to the curve-positive upper metric rr and gives r=pr=p. This proves the third condition. Lemma “Smooth approximation of FS metrics” gives the fourth condition from the third, and positivity passes to the current limit, giving the second. Finally, Proposition “Curve maps and extension”, applied in local frames, gives the first condition from the second. Positive tensor powers scale each curvature by the same positive integer, which proves the rational-line assertion. ◻

Comparison with adelic mixed measures

Definition (Boundary completion). Let U/kU/k be quasi-projective. We use the rational boundary completion of [9, Sections 2.4–2.5], retaining its specified line-bundle identifications. A boundary datum consists of a projective compactification, an effective Cartier boundary with its multiplicities, and strictly positive Green data. On common dominating models a Cauchy presentation satisfies, for every ε>0\varepsilon>0 and all sufficiently large j,lj,l,

−εΔ‾∂≤D‾j−D‾l≤εΔ‾∂.-\varepsilon\overline\Delta_\partial \leq\overline D_j-\overline D_l \leq\varepsilon\overline\Delta_\partial .

In a fixed local identification at vv, this implies ∣wj−wl∣≤εbvon UKan,|w_j-w_l|\leq\varepsilon b_v\quad\text{on }U_K^\mathrm{an}, where the boundary Green function bvb_v is finite continuous on that open. The completion uses the full order (eq:boundary-order), not merely this scalar inequality. A strongly nef adelic line bundle has a Cauchy presentation by nef model objects. An integrable one is a difference of two strongly nef objects, allowing rational tensor powers and the specified isometries [9, Definition 2.5.2].

Identification and positivity of the prescribed measure

Proof of Theorem “Positivity of the prescribed adelic mixed measure”. Choose integrable presentations with their prescribed isometries, L‾i⊗ri≃P‾i⊗Q‾i−1,ri∈Z>0,\overline L_i^{\otimes r_i} \simeq\overline P_i\otimes\overline Q_i^{-1}, \qquad r_i\in\mathbf Z_{>0}, where P‾i,Q‾i\overline P_i,\overline Q_i are strongly nef. Their metrics at vv are locally psh-approachable, and their mixed measures are the stable CLD products [9, Lemma 3.6.7 and Sections 3.6.7–3.6.8]. Let ui,pi,qiu_i,p_i,q_i be weights in compatible local frames. The specified isometries give riui=pi−qi+hi,hi=−log⁡∣ai∣,d′d′′hi=0,r_i u_i=p_i-q_i+h_i,\qquad h_i=-\log|a_i|,\qquad d'd''h_i=0, for analytic units aia_i. By the definition in [9, Section 3.6.8], the prescribed signed measure is

μvYZ=1∏iri∑ϵ∈{0,1}n(−1)∣ϵ∣∏i=1nc1(R‾i,ϵi)v,R‾i,0=P‾i,R‾i,1=Q‾i.\mu_v^{\mathrm{YZ}} =\frac1{\prod_i r_i} \sum_{\epsilon\in\{0,1\}^n}(-1)^{|\epsilon|} \prod_{i=1}^n c_1(\overline R_{i,\epsilon_i})_v, \qquad \overline R_{i,0}=\overline P_i,\quad \overline R_{i,1}=\overline Q_i.

Its independence of presentation is part of that construction.

Theorem “Local regularization” makes each target weight uiu_i approachable. Apply additivity of approachable products to riui+qi=pi+hir_i u_i+q_i=p_i+h_i in every slot; the zero-curvature terms hih_i disappear by (eq:continuous-kernel). The alternating expansion is exactly

μvYZ=⋀i=1nd′d′′ui.\mu_v^{\mathrm{YZ}}=\bigwedge_{i=1}^nd'd''u_i.

The right-hand side is positive by Theorem “Continuous potentials on a positive closed background”. Positivity of Radon measures is local, so this proves the assertion on all of UKanU_K^\mathrm{an}. ◻

A quantitative comparison of mixed measures

For completeness, we record a comparison estimate in which the constant is independent of the metrics being compared. It is separate from the proof of contact rigidity: the divisor limit there uses only the continuity of a fixed potential acting on positive closed currents.

Lemma (Global domination of finitely many forms). Let Y/KY/K be smooth projective, and let η1,…,ηs\eta_1,\ldots,\eta_s be smooth symmetric real (1,1)(1,1)-forms on YanY^\mathrm{an}. There are an ample line bundle HH, a global smooth metric ϕ\phi on HH, and its positive curvature ω=c1(H,ϕ)\omega=c_1(H,\phi), such that −ω≤ηi≤ω(1≤i≤s).-\omega\leq\eta_i\leq\omega\qquad(1\leq i\leq s). The inequalities are in the original CLD form cone, and all the constructions take place over KK.

Proof. Choose a very ample line bundle AA on YY. We first construct one global positive curvature that controls the square of a local moment coordinate. On a compact polynomial norm box CC contained in an affine chart of the embedding defined by AA, let ff be an analytic unit on a neighbourhood of CC. The box is defined by finitely many upper bounds and positive lower bounds on polynomial functions, with bounds in ∣K×∣Q|K^\times|^\mathbf Q. After clearing powers, its affinoid algebra is the completion of the affine coordinate algebra with the lower-bounded polynomials inverted. Thus ff can be uniformly approximated on CC by a rational function hh from this algebra. Choose the error smaller than min⁡C∣f∣\min_C|f|. The ultrametric inequality gives ∣h∣=∣f∣throughout C.|h|=|f|\quad\text{throughout }C. In particular hh is a rational unit there. This equality of moment maps gives the same differential forms on IntC\mathop{\mathrm{Int}}C, by [3, Lemma 3.1.10].

Homogenizing a numerator and denominator for hh expresses it as h=s1/s0h=s_1/s_0, where s0,s1s_0,s_1 are sections of some positive tensor power AdA^d, both nonzero on CC. Set M=∥log⁡∣h∣∥CM=\|\log|h|\|_C, and choose ε>2M\varepsilon>2M. In the construction of Proposition “Regularized maximum”, choose ρ\rho strictly positive on (−1,1)(-1,1), and write Hε(t)=R2,ε(t,0).H_\varepsilon(t)=R_{2,\varepsilon}(t,0). Then Hε′′(t)=1ε(ρ∗ρˇ)(t/ε)>0(∣t∣<2ε),ρˇ(t)=ρ(−t).H_\varepsilon''(t) =\frac1\varepsilon (\rho*\check\rho)(t/\varepsilon)>0 \qquad(|t|<2\varepsilon), \qquad\check\rho(t)=\rho(-t). Hence Hε′′≥c>0H_\varepsilon''\geq c>0 on [−M,M][-M,M].

Take a global generating tuple of AdA^d, multiply its members by sufficiently small elements of K×K^\times, and append them to s0,s1s_0,s_1. If aa is any appended section, choose the coefficient so that log⁡∣a/s0∣<−2ε−1on C\log|a/s_0|<-2\varepsilon-1 \quad\text{on }C where the value is −∞-\infty at a zero. Thus its logarithmic entry lies more than 2ε2\varepsilon below the maximum of 00 and log⁡∣h∣\log|h| throughout CC. Every appended entry is exactly inactive in the regularized maximum on a neighbourhood of CC. The full tuple generates AdA^d everywhere, and regularizing its maximum gives a global smooth positive metric by Lemma “Smooth approximation of FS metrics”. In the frame s0s_0, its weight on IntC\mathop{\mathrm{Int}}C is precisely Hε(log⁡∣h∣)H_\varepsilon(\log|h|). Its curvature there is Hε′′(log⁡∣h∣) d′log⁡∣h∣∧d′′log⁡∣h∣≥c d′log⁡∣h∣∧d′′log⁡∣h∣.H_\varepsilon''(\log|h|) \,d'\log|h|\wedge d''\log|h| \geq c\,d'\log|h|\wedge d''\log|h|.

Now cover YanY^\mathrm{an} by the interiors of finitely many such boxes, using common moment presentations for the finitely many forms ηi\eta_i on each box. Replace the analytic moment units by rational units as above. The symmetric coefficient matrices of all the ηi\eta_i are bounded on these compact boxes. The construction for each moment unit supplies a global positive curvature controlling its coordinate square. Suitable positive integer multiples of these curvatures therefore dominate both signs of every coefficient matrix on the corresponding box. Their sum is a global smooth positive curvature, dominates every ηi\eta_i on the finite cover, and is the curvature of a positive tensor power of AA. If the list of required coordinate squares is empty, add any smooth positive approximation of a pure-FS metric on AA. This gives the asserted ample line HH and metric ϕ\phi. ◻

Proposition (Comparison on a fixed smooth test). Let Y/KY/K be smooth projective of dimension n≥1n\geq1. For 1≤i≤n1\leq i\leq n, let pi,qip_i,q_i be globally smoothly approximable metrics on the same specified line bundle LiL_i. Fix a real smooth function φ\varphi on YanY^\mathrm{an}. Choose an ample HφH_\varphi and a global smooth positive curvature ωφ\omega_\varphi such that −ωφ≤d′d′′φ≤ωφ,-\omega_\varphi\leq d'd''\varphi\leq\omega_\varphi, as in Lemma “Global domination of finitely many forms”. Then

∣∫Yanφ(⋀i=1nc1(Li,pi)−⋀i=1nc1(Li,qi))∣≤∑i=1n∥pi−qi∥∞ deg⁡K ⁣(Hφ∏h≠iLh).\begin{aligned} &\left|\int_{Y^\mathrm{an}}\varphi\left( \bigwedge_{i=1}^n c_1(L_i,p_i) -\bigwedge_{i=1}^n c_1(L_i,q_i)\right)\right|\\ &\hspace{1em}\leq \sum_{i=1}^n\|p_i-q_i\|_\infty\, \deg_K\!\left(H_\varphi\prod_{h\ne i}L_h\right). \end{aligned}

The norm is the uniform norm of the scalar difference of two weights on the same line. The line HφH_\varphi is fixed once φ\varphi and its calibration have been chosen; the degree constants are independent of all the metrics pi,qip_i,q_i. The assertion also holds for rational line bundles after clearing and dividing tensor denominators.

Proof. Put αi=c1(Li,pi)\alpha_i=c_1(L_i,p_i), βi=c1(Li,qi)\beta_i=c_1(L_i,q_i), and Ti=⋀h<iαh∧⋀h>iβhT_i=\bigwedge_{h<i}\alpha_h\wedge\bigwedge_{h>i}\beta_h. Telescoping and current duality give ∫φ(⋀iαi−⋀iβi)=∑i∫(pi−qi) d′d′′φ∧Ti.\int\varphi\left(\bigwedge_i\alpha_i-\bigwedge_i\beta_i\right) =\sum_i\int(p_i-q_i)\,d'd''\varphi\wedge T_i. Each TiT_i is positive and closed, so ∣d′d′′φ∧Ti∣≤ωφ∧Ti\lvert d'd''\varphi\wedge T_i\rvert\leq\omega_\varphi\wedge T_i. By Lemma “Degree and comparison of measures”, the latter measure has mass deg⁡K(Hφ∏h≠iLh)\deg_K(H_\varphi\prod_{h\ne i}L_h). Estimating each summand by its uniform coefficient norm proves (eq:metric-comparison). Multilinearity gives the rational-line version. ◻

Notation

Symbol

Meaning

Reference

K,R,π,λK,R,\pi,\lambda

Fixed completion, valuation ring, uniformizer, and −log⁡∣π∣-\log|\pi|

Section “Introduction”

d′d′′,Acp,qd'd'',\mathcal A_c^{p,q}

Original CLD differentials and compact smooth test forms

Section “Introduction”

Ploc,Dloc\mathcal P_{\mathrm{loc}},\mathcal D_{\mathrm{loc}}

Local smooth positive approximation and its difference class

Definition def:approachable

Mu(T)M_u(T)

The operation d′d′′(uT)d'd''(uT) for approachable uu

Proposition “A fixed approachable potential”

BS\mathcal B_S

Mixed product with a fixed positive closed background

(eq:recursive-product)

QC,ΓCQ_C,\Gamma_C

Curve Dirichlet energy and its energy measure

Lemma “Green identities on curves”

TaT_a

Normalized smooth complete-intersection curve currents

Proposition “Complete-intersection approximation”

p,r,gp,r,g

Lower approachable metric, upper curve-positive metric, and g=r−pg=r-p

Theorem “Contact rigidity”

PL(r)P_L(r)

Semipositive envelope of an obstacle metric

Section “Envelopes on a fixed regular model”

Vl,jmV_l,\mathfrak j_m

Admissible sections and sums of divisor test ideals

Proposition “The test-ideal estimate”

E,ΔW\mathscr E,\Delta_{\mathscr W}

Vertical obstacle ideal and canonical discrepancy boundary

Proposition “The test-ideal estimate”

ωφ,Hφ\omega_\varphi,H_\varphi

Fixed positive calibration and its ample line

Proposition “Comparison on a fixed smooth test”

bv,Δ‾∂b_v,\overline\Delta_\partial

Boundary Green function and full boundary datum

Definition “Boundary completion”

μvYZ\mu_v^{\mathrm{YZ}}

Prescribed signed adelic mixed measure

(eq:adelic-expansion)

Rh,εR_{h,\varepsilon}

Convex regularized maximum

Proposition “Regularized maximum”

Declaration on the use of generative AI

Generative AI was used in developing and drafting mathematical arguments and in revising the exposition of this manuscript.

References

  1. [1]

    V. G. Berkovich, Étale cohomology for non-Archimedean analytic spaces, Publications Mathématiques de l’IHÉS 78 (1993), 5–161.

  2. [2]

    The Stacks Project Authors, The Stacks Project, Tag 0FUE, Lemma 37.17.1.

  3. [3]

    A. Chambert-Loir and A. Ducros, Formes différentielles réelles et courants sur les espaces de Berkovich, arXiv:1204.6277v1, 2012.

  4. [4]

    W. Gubler, P. Jell, and J. Rabinoff, Forms on Berkovich spaces based on harmonic tropicalizations, arXiv:2111.05741v3, 2025.

  5. [5]

    W. Gubler, P. Jell, and J. Rabinoff, Dolbeault cohomology of graphs and Berkovich curves, arXiv:2111.05747v1, 2021.

  6. [6]

    B. Bhatt, L. Ma, Z. Patakfalvi, K. Schwede, K. Tucker, J. Waldron, and J. Witaszek, with an appendix by R. Datta, Test ideals in mixed characteristic: a unified theory up to perturbation, arXiv:2401.00615v2.

  7. [7]

    S. Boucksom and D. Eriksson, Spaces of norms, determinant of cohomology and Fekete points in non-Archimedean geometry, arXiv:1805.01016v3, 2021.

  8. [8]

    S. Boucksom, W. Gubler, and F. Martin, Non-Archimedean volumes of metrized nef line bundles, Épijournal de Géométrie Algébrique 5 (2021), Article 13; arXiv:2011.06986v3.

  9. [9]

    X. Yuan and S.-W. Zhang, Adelic line bundles on quasi-projective varieties, arXiv:2105.13587v9.

  10. [10]

    A. Chambert-Loir and A. Ducros, Formes différentielles réelles et courants sur les espaces de Berkovich, arXiv:1204.6277v2, 2025.

  11. [11]

    V. Wanner, Comparison of two notions of subharmonicity on non-archimedean curves, arXiv:1801.04713v3, 2018.

  12. [12]

    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, 2021.

  13. [13]

    Y. Fang, W. Gubler, and K. Künnemann, On the non-archimedean Monge–Ampère equation in mixed characteristic, arXiv:2203.12282v3, 2025.

  14. [14]

    W. Gubler and J. Rabinoff, On semipositive piecewise linear functions in non-archimedean analytic geometry, arXiv:2503.05513v1, 2025.

  15. [15]

    W. Gubler and J. Rabinoff, Classically psh and pluriharmonic functions on Berkovich spaces, arXiv:2506.13548v2, 2025.

  16. [16]

    L. Wu, On the local theory of non-archimedean psh functions, arXiv:2609.27610v1, 2026.

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 2c4ee1dbe9c55f0a76607b3bf4bf1b4154b416ab59cf359e12c0c69a08c556b5

Frozen-original SHA-256 83153cdc67adbe4983c1c7f265662d277032968b5364d37afcbdc75cc80737fc

\documentclass[11pt]{amsart}
% Prefer Latin Modern; keep a Computer Modern fallback for minimal TeX setups.
\IfFileExists{lmodern.sty}{%
 \usepackage[T1]{fontenc}\usepackage{lmodern}%
}{%
 \usepackage[OT1]{fontenc}%
 \ifdefined\pdfmapfile
  \IfFileExists{pdftex.map}{}{%
   \pdfmapfile{+cm.map}%
   \pdfmapfile{+cmextra.map}%
   \pdfmapfile{+symbols.map}%
   \pdfmapfile{+euler.map}%
   \pdfmapfile{+latxfont.map}%
   \pdfmapfile{+rsfs.map}%
  }%
 \fi
}
\usepackage{amsmath,amssymb,mathtools,mathrsfs}
\usepackage{microtype}
\usepackage{enumitem}
\usepackage{needspace}
\usepackage{longtable,booktabs,array}
\usepackage{etoolbox}
\usepackage{xcolor}
\definecolor{linkblue}{RGB}{28,64,120}
\definecolor{sectionteal}{RGB}{17,103,111}
\definecolor{equationviolet}{RGB}{105,70,142}
\definecolor{citegreen}{RGB}{24,97,77}
\definecolor{urlbrown}{HTML}{A33A00}
\usepackage[colorlinks=true,linkcolor=linkblue,citecolor=citegreen,
 urlcolor=urlbrown,linktoc=all,bookmarksnumbered=true]{hyperref}
\hypersetup{pdftitle={Continuous CLD-positive potentials and mixed currents},
 pdfauthor={},pdfsubject={Local regularization and continuous mixed currents; revised version 2.6}}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\theoremstyle{remark}
\newtheorem*{contactintro}{Contact rigidity}
\numberwithin{equation}{section}
\setlist[enumerate,1]{label=\textup{(\roman*)},leftmargin=2.4em}
\newcommand{\ddc}{d'd''}
\newcommand{\an}{\mathrm{an}}
\newcommand{\Ploc}{\mathcal P_{\mathrm{loc}}}
\newcommand{\Dloc}{\mathcal D_{\mathrm{loc}}}
\newcommand{\Ac}{\mathcal A_c}
\newcommand{\R}{\mathbf R}
\newcommand{\Q}{\mathbf Q}
\newcommand{\Z}{\mathbf Z}
\newcommand{\PP}{\mathbf P}
\newcommand{\Aff}{\mathbf A}
\newcommand{\OO}{\mathcal O}
\newcommand{\cX}{\mathscr X}
\newcommand{\cW}{\mathscr W}
\newcommand{\cL}{\mathscr L}
\newcommand{\cA}{\mathscr A}
\newcommand{\cE}{\mathscr E}
\newcommand{\fc}{\mathfrak c}
\newcommand{\fb}{\mathfrak b}
\newcommand{\norm}[1]{\left\lVert#1\right\rVert}
% Named results, sections, displayed equations, citations, and URLs have
% distinct, restrained link colors; the whole named reference is clickable.
\newcommand{\namedref}[2]{%
 \begingroup
 \ifstrequal{#1}{Section}{\hypersetup{linkcolor=sectionteal}}{%
  \ifstrequal{#1}{Appendix}{\hypersetup{linkcolor=sectionteal}}{}}%
 \hyperref[#2]{#1~\ref*{#2}}%
 \endgroup}
\renewcommand{\eqref}[1]{%
 \begingroup\hypersetup{linkcolor=equationviolet}%
 \hyperref[#1]{\textup{(\ref*{#1})}}%
 \endgroup}
\newcommand{\TV}{\mathrm{TV}}
\DeclareMathOperator{\supp}{supp}
\DeclareMathOperator{\Int}{Int}
\DeclareMathOperator{\divisor}{div}
\DeclareMathOperator{\im}{im}
\hyphenation{smooth}
\title[Continuous CLD-positive potentials]{Continuous CLD-positive potentials\\and mixed currents}
\author{}
\date{15 November 2026}

\begin{document}
\begin{abstract}
Let \(K\) be a non-Archimedean completion of a number field and let
\(X/K\) be smooth and quasi-projective. A finite continuous function on an
ordinary open of \(X^\an\) has positive Chambert--Loir--Ducros curvature
current if and only if it admits smooth positive approximations on a
fixed neighbourhood of each point. The proof combines restriction to
curves, deterministic complete-intersection approximation, contact
rigidity, and continuity of semipositive envelopes. Contact rigidity
identifies a continuous curve-positive metric from its contact with a
uniform Fubini--Study limit.
On a fixed regular model, sums of divisor test ideals and global generation
give one-sided \(O(m^{-1})\) Fubini--Study approximations for model-obstacle
envelopes. A single projective extension then completes local regularization. As consequences, continuous current-positive potentials admit
symmetric mixed products with any positive closed CLD current, and these
products depend only on their curvature currents. We also obtain a
projective curve criterion under a regular-model hypothesis and positivity
of the prescribed Yuan--Zhang local mixed measures.
\end{abstract}
\maketitle
\setcounter{tocdepth}{1}
\tableofcontents
\clearpage

\section{Introduction}\label{sec:introduction}

Positivity of a Chambert--Loir--Ducros current is a condition on smooth
test forms. To construct positive mixed products, one must also control
\(\ddc(uT)\) for each intermediate positive closed current \(T\).
Multiplication by a continuous potential is available, but positivity
after differentiation requires an additional argument. Local
approximation by smooth functions with positive curvature provides
that argument. The first question addressed here is whether the
current inequality itself supplies such an approximation.

\subsection{Local regularization}
Fix a number field \(k\), a non-Archimedean place \(v\), and
\(K=k_v\), with its given absolute value. Its valuation ring is \(R\).
Choose a uniformizer \(\pi\) and set \(\lambda=-\log|\pi|>0\).
All smooth forms, positive cones, and currents in this paper are those
of the first version of Chambert--Loir--Ducros \cite{CLD12}.
In particular, positivity of currents is dual to the positive cone
of \cite[Definition 5.1.1(c)]{CLD12}. An \emph{ordinary open} is open
in the Berkovich topology.

Let \(X/K\) be smooth of pure dimension \(n\), let \(O\subset X^\an\)
be an ordinary open, and let \(u:O\to\R\) be finite and continuous.
For \(n\geq1\), the condition \(\ddc[u]\geq0\) means
\begin{equation}\label{eq:current-positive}
 \langle\ddc[u],\eta\rangle
 =\int_O u\,\ddc\eta\geq0
 \quad\text{for every positive }\eta\in\Ac^{n-1,n-1}(O).
\end{equation}
Every compactly supported smooth test is allowed. For components of
different dimensions the condition is imposed componentwise, and there
is no curvature condition in dimension zero.

We write \(u\in\Ploc(O)\) if every \(x\in O\) has one ordinary
neighbourhood \(V\subset O\) on which there are smooth functions \(u_j\)
with \(\ddc u_j\geq0\) as forms and \(u_j\to u\) uniformly on compact
subsets of \(V\). The neighbourhood is fixed for the entire sequence.

\begin{theorem}[Local regularization]\label{thm:main}
Let \(K=k_v\) be a non-Archimedean completion of a number field, with
its given absolute value. Let \(X/K\) be smooth and quasi-projective,
\(O\subset X^\an\) an ordinary open, and \(u:O\to\R\) finite continuous.
Then the following are equivalent:
\begin{enumerate}
\item The original CLD curvature current \(\ddc[u]\) is positive on \(O\)
in the sense of \eqref{eq:current-positive}.
\item The function \(u\) belongs to \(\Ploc(O)\).
\end{enumerate}
After shrinking the fixed neighbourhood of each point once, the
approximating sequence can be chosen decreasing and uniformly convergent
on that neighbourhood.
\end{theorem}

The implication from smooth positive approximation to current positivity
follows by testing locally uniform convergence against \(\ddc\eta\).
The reverse implication does not assume that \(u\) is a difference of
approachable functions. The field and continuity hypotheses remain in
force throughout the theorem; in particular, the assertion does not
concern arbitrary upper-semicontinuous potentials. The proof is completed
in \namedref{Section}{subsec:regularization-proof}.

\subsection{Contact rigidity}
The projective argument uses a criterion that identifies a metric from
its contact with a semipositive lower metric. Recall that a \emph{pure
Fubini--Study metric} on an ample line bundle \(L\) is given by weights
\(m^{-1}\max_i\log|s_i|\), where the sections \(s_i\) generate \(mL\).
A \emph{uniform FS limit} is a uniform limit of such metrics.
A continuous metric is \emph{curve-positive} if its pullback by every
algebraic morphism from a smooth \(K\)-curve has positive Chern current.
These conventions are developed in \namedref{Section}{sec:metrics-curves}.

\begin{contactintro}[\namedref{Theorem}{thm:contact}]
Let \(Y/K\) be smooth, projective, and integral of dimension \(n\ge1\),
and let \(L\) be ample. Let \(p\) be a uniform FS limit on \(L\), and let
\(r\) be a finite continuous curve-positive metric on \(L\). Then
\begin{equation}\label{eq:contact-intro}
 r\ge p,\qquad \int_{Y^\an}(r-p)c_1(L,p)^n=0
 \quad\Longrightarrow\quad r=p.
\end{equation}
\end{contactintro}

The metric \(p\) need not be constructed as an envelope of \(r\).
The criterion uses the Monge--Amp\`ere measure of the lower metric and
positivity of the upper metric only on curves. It therefore applies
before a higher-dimensional positive product for \(r\) is available.
The proof in \namedref{Section}{subsec:contact-proof} passes curve Green
identities through complete-intersection approximation and then
transfers the contact identity to smooth divisors.

\subsection{Continuous mixed currents}
A current is called \emph{closed} below if both \(d'T=0\) and \(d''T=0\).
Positive currents admit multiplication by continuous functions
(\namedref{Lemma}{lem:continuous-coefficients}). Thus, for a positive closed
current \(S\), the recursive rule
\begin{equation}\label{eq:recursive-product}
 \mathcal B_S()=S,\qquad
 \mathcal B_S(u_1,\ldots,u_m)
 =\ddc\bigl(u_1\mathcal B_S(u_2,\ldots,u_m)\bigr)
\end{equation}
is meaningful whenever its intermediate currents are positive.
This is the usual Bedford--Taylor construction.

\begin{theorem}[Continuous potentials on a positive closed background]
\label{thm:background-products}
Let \(K=k_v\), let \(X/K\) be smooth quasi-projective of pure dimension
\(n\), and let \(O\subset X^\an\) be an ordinary open. Let \(S\) be an
original CLD positive current of bidegree \((p,p)\), where \(0\leq p\leq n\), with
\(d'S=d''S=0\). Suppose \(0\leq m\leq n-p\) and
\(u_1,\ldots,u_m:O\to\R\) are finite continuous functions with
\(\ddc[u_i]\geq0\). Then \eqref{eq:recursive-product} defines a positive
closed current of bidegree \((p+m,p+m)\). It has the following properties.
\begin{enumerate}
\item It is local, symmetric in the potentials, additive and positively
homogeneous in each potential and in the background.
\item If \(S_a\to S\) on every compact smooth test, where all \(S_a\)
are positive and closed of bidegree \((p,p)\), and the finite
continuous functions \(u_{i,a}:O\to\R\) converge to \(u_i\) locally
uniformly with \(\ddc[u_{i,a}]\geq0\), then
\[
 \mathcal B_{S_a}(u_{1,a},\ldots,u_{m,a})
 \longrightarrow\mathcal B_S(u_1,\ldots,u_m)
\]
on every compact smooth test.
\item Each potential enters only through its curvature current:
if \(w\) is finite continuous and \(\ddc[w]=\ddc[u_i]\), replacing
\(u_i\) by \(w\) leaves the product unchanged.
\end{enumerate}
In top degree the product is a positive Radon measure.
\end{theorem}

The recursion is described in \cite[Section 5.6.8]{CLD12}.
For continuous psh potentials, its positivity, continuity, and symmetry
are established in
\cite[Lemma 8.5.5 and Propositions 8.7.2, 8.7.4]{CLD25}.
\namedref{Theorem}{thm:main} allows us to start from current positivity alone.
Potential independence then has a short two-sided proof. If \(\ddc[h]=0\), then both \(h\) and \(-h\) satisfy
\namedref{Theorem}{thm:main}. For every positive closed \(T\) this gives
\begin{equation}\label{eq:two-sided-intro}
 \ddc(hT)\geq0,\qquad -\ddc(hT)=\ddc((-h)T)\geq0,
 \qquad\text{hence }\ddc(hT)=0.
\end{equation}
No representation of \(h\) as a piecewise-linear or logarithmic function
is required. \namedref{Theorem}{thm:background-products} concerns continuous
potentials; it does not assert a monotone convergence theorem for
bounded discontinuous ones. The theorem is proved in
\namedref{Section}{subsec:mixed-current-proof}.

\subsection{Adelic mixed measures}
For the arithmetic application, use the rational boundary completion of
\cite[Sections 2.4--2.5]{YZ}. Its objects retain the specified
line-bundle identifications. A strongly nef object has a Cauchy
presentation by nef model objects, and an integrable object is a
difference of two strongly nef objects, with rational tensor powers
allowed. The order defining the completion compares the full
Cartier-and-metric boundary datum, as recalled in
\namedref{Section}{sec:adelic}.

\begin{theorem}[Positivity of the prescribed adelic mixed measure]
\label{thm:adelic}
Let \(k\) be a number field, \(v\) a non-Archimedean place, and \(K=k_v\).
Let \(U/k\) be smooth quasi-projective of pure dimension \(n\geq1\).
Let \(\overline L_1,\ldots,\overline L_n\) be rational integrable adelic
line bundles with their specified identifications in the boundary
completion of \cite[Sections 2.4--2.5]{YZ}. Suppose their local weights
at \(v\) are finite continuous and have positive original CLD curvature
currents. Then the signed measure prescribed in
\cite[Sections 3.6.7--3.6.8]{YZ} is positive:
\[
 c_1(\overline L_1)_v\cdots c_1(\overline L_n)_v\geq0
 \quad\text{on }U_K^\an.
\]
\end{theorem}

The signed measure is defined by expanding integrable presentations.
Its positivity follows by identifying that expansion with the local
positive product of the target weights. This conclusion concerns the
given measure and does not assert arithmetic nefness of the factors.
The comparison is given in \namedref{Section}{subsec:adelic-proof}.

\subsection{Relation to earlier work}
Chambert--Loir--Ducros distinguish distributional positivity from local
psh approachability in \cite[Definition 5.6.3 and Remark 5.6.4]{CLD12};
their stable products are treated in
\cite[Corollary 5.6.5 and Section 5.6.8]{CLD12}.
The later continuous psh class is defined by uniform closure
\cite[Definition 8.1.7]{CLD25}. We do not identify a continuous
current-positive function with a member of that class before proving
\namedref{Theorem}{thm:main}.

On curves, Wanner compares CLD and Thuillier subharmonicity
\cite[Corollaries 4.8 and 5.4]{Wanner}.
Gubler--Jell--Rabinoff provide descent, harmonic tropicalizations,
and smooth approximation \cite{GJRforms,GJRcurves}.
The finite residue field is important here: weakly smooth functions
on the relevant curves are CLD-smooth. \namedref{Section}{sec:metrics-curves}
retains the passage from the original current inequality over \(K\)
to those curve results.

Torus-invariant pluripotential theory is developed in \cite{BGJK}.
Piecewise-linear semipositivity and positivity tested on curves are
studied in \cite{GRpl,GRpsh}, and Wu compares local approximation
classes in \cite[Appendix A]{Wu}. Our starting hypothesis is instead
the original current inequality \eqref{eq:current-positive} for an
arbitrary finite continuous potential on a smooth variety.

The envelope argument follows the test-ideal method of
Fang--Gubler--K\"unnemann \cite{FGK}, using the mixed-characteristic
test ideals of \cite{BMP}. We use test ideals of section divisors directly.
The comparison on a higher normal model allows an
obstacle on a higher normal model while subadditivity is applied on
a fixed regular model. It is not a direct application of
\cite[Theorems 5.1 and 7.1]{FGK}. Orthogonality, once continuity is
established, is \cite[Theorem 6.7]{BGM}.
The final adelic application identifies a measure already constructed
by Yuan--Zhang; it does not introduce another mixed measure.

\subsection{Proof strategy}
The proof separates restriction, contact, and envelopes so that none
uses regularization in a form still to be established.

\emph{Step 1: restrict the current inequality and extend locally.}
In a normal disc, let
\(\beta_N=\ddc R_{2,\varepsilon}(\log|t|,-N)\), where
\(R_{2,\varepsilon}\) is a smooth convex regularized maximum.
For the zero section \(i:B\hookrightarrow B\times D_s\),
Poincar\'e--Lelong and continuous multiplication give
\begin{equation}\label{eq:normal-limit-intro}
 \beta_N\longrightarrow i_*[B],
 \qquad u\beta_N\longrightarrow i_*[u|_B].
\end{equation}
Testing against pullbacks from \(B\) proves positivity on the zero
section. The \'etale normal form of a smooth pair and Berkovich's
theorem on \'etale germs then give restriction to smooth
subvarieties. Curve regularization and removal of finitely many
closed points prove positivity after every algebraic curve map.
An algebraic \'etale presentation also supplies an ambient analytic
retraction, extending \(u\) to a continuous curve-positive function
\(v\) near the original point
(\namedref{Section}{sec:restriction}).

\emph{Step 2: approximate mixed currents by curves and prove contact.}
For a generating tuple \(s_0,\ldots,s_b\), choose smooth cuts of
the form
\[
 \sigma_N=\sum_{i=0}^b a_{i,N}s_i^N,
 \qquad a_{i,N}\in\pi^i(1+\pi R).
\]
On a tropical presentation of any fixed smooth test, equality of
the norms of two terms lies on an affine hyperplane, apart from
finitely many exceptional \(N\). Distinct coefficient valuations
exclude ties that persist for all \(N\).
Absolute tropical integration therefore gives an eventual
\(O(N^{-1})\) error on that test. Successive cuts and weak closure
yield normalized individual smooth complete-intersection curves,
with fixed intersection degrees
(\namedref{Proposition}{prop:curve-approximation}).

For the contact criterion \eqref{eq:contact-intro}, put
\(g=r-p\) and \(\alpha=c_1(L,p)\).
Passing curve Green identities through the approximation of
\(T=\alpha^{n-1}\) gives
\[
 \nu=\ddc(gT)=\sigma-\alpha^n,\qquad \sigma\geq0,\qquad
 0\leq-\int g\,d\nu=-\int g\,d\sigma\leq0.
\]
The nonnegative energy of \(g+th\), for smooth \(h\) and \(t\in\R\),
then forces \(\nu=0\).
For a smooth divisor \(D=\divisor(s)\) in \(mL\), set
\(\rho_N=\max\{mp,\log|s|+N\}\). Fixed-potential continuity gives
\begin{equation}\label{eq:divisor-intro}
 c_1(mL,\rho_N)\wedge\alpha^{n-1}
 \longrightarrow i_*\bigl(c_1(L|_D,p|_D)^{n-1}\bigr).
\end{equation}
Green symmetry transfers contact to \(D\), and induction on
dimension proves \eqref{eq:contact-intro}
(\namedref{Section}{sec:contact}). No mixed product with the upper metric
\(r\) is assumed in this argument.

\emph{Step 3: construct continuous envelopes on a fixed regular model.}
Represent a model obstacle on a higher normal model \(\cW\) as
\(r=\theta+\log|\cE|\), with \(\cE\) an invertible vertical ideal.
For each admissible section \(s\in H^0(Y,lL)\), let \(D_s\) be its
divisor on a fixed regular model \(\cX\). We take the sum
\[
 \mathfrak j_m=\sum_{l\ge1}\sum_{0\ne s\in V_l}
       \tau_{\cX}\!\left(\frac mlD_s\right),
 \qquad V_l=\{s:\|s\|_{lr,\sup}\le1\}.
\]
Subadditivity gives the lower bound below. Birational comparison,
followed by clearing a fixed discrepancy denominator, gives the upper bound:
\begin{equation}\label{eq:ideal-intro}
 \pi^B\mathfrak j_m\OO_{\cW}\subseteq\cE^m,\qquad
 P_L(r)\leq\theta+\frac1m\log|\mathfrak j_m|
 \leq r+\frac{B\lambda}{m}.
\end{equation}
Global generation and one fixed auxiliary line bundle turn these bounds
into pure-FS approximations of \(P_L(r)\) on \(L\).
On \(\cX=\PP_R^a\), the identity
\(K_{\cX}+(a+1)\OO(1)=0\) cancels the auxiliary twist directly and gives
\begin{equation}\label{eq:rate-intro}
 0\leq q_m-P_L(r)\leq\frac{B\lambda}{m}.
\end{equation}
Model-metric density and the one-Lipschitz property of the envelope
treat continuous obstacles. Orthogonality \cite[Theorem 6.7]{BGM}
then applies (\namedref{Section}{subsec:envelope-proof}).

\emph{Step 4: extend once to projective space and apply contact rigidity.}
After adding a constant and shrinking the ambient neighbourhood, arrange
\(0<v+c<\lambda/2\). Choose one compact polynomial norm box \(C\) and
an FS metric \(q\), with a section \(s_0\), such that
\[
 b=-\log\|s_0\|_q,\qquad b=0\text{ near }x,
 \qquad\{b\le\lambda\}=C.
\]
On \(\{b<\lambda\}\), use the weight \(\max\{v+c,b\}\); on
\(\{b>\lambda/2\}\), use \(q\). The two definitions agree on the
overlap and give a global curve-positive metric \(r\)
(\namedref{Lemma}{lem:projective-extension}). Its envelope \(p=P_L(r)\)
satisfies the contact identity, so \namedref{Theorem}{thm:contact} gives
\(r=p\). Smooth FS approximation and restriction to one fixed
neighbourhood recover \(u\). Explicit constant translations make the
approximants decreasing (\namedref{Section}{subsec:regularization-proof}).
The same envelope--contact argument gives the projective curve criterion
under the retained regular-model hypothesis
(\namedref{Corollary}{cor:projective-criterion}).

\subsection{Organization}
\namedref{Section}{sec:coefficients} proves the continuous-coefficient
estimate and the continuity of a fixed approachable potential acting
on positive closed currents. \namedref{Section}{sec:metrics-curves} supplies
smooth Fubini--Study approximation, the measure normalization, and
the Green identities on curves. \namedref{Section}{sec:restriction}
establishes restriction and the ambient extension.
\namedref{Section}{sec:contact} combines complete-intersection approximation
with the direct divisor limit to prove contact rigidity.
\namedref{Section}{sec:envelopes} proves envelope continuity on a fixed regular
model, with projective space as a corollary. \namedref{Section}{sec:local-proof}
constructs the projective extension and completes local regularization. The mixed-current and curve-criterion consequences are
proved in \namedref{Section}{sec:consequences}; \namedref{Section}{sec:adelic}
identifies the prescribed adelic measure.
\namedref{Appendix}{app:comparison} records a quantitative comparison of
mixed measures, independently of the contact argument.

\section{Continuous coefficients and approachable potentials}
\label{sec:coefficients}

We use the smooth cutoff and density results
\cite[Corollary 3.3.4 and Propositions 3.3.5--3.3.6]{CLD12},
the Poincar\'e--Lelong formula \cite[Theorem 4.6.5]{CLD12}, and
uniform stability of products
\cite[Corollary 5.6.5]{CLD12}. Weak convergence of currents always
means convergence on every compactly supported smooth test.
Convergence of measures on continuous tests will be stated explicitly.

\begin{definition}\label{def:approachable}
On an ordinary open \(O\subset X^\an\), a smooth function is locally
of the form
\[
 F(\log|f_1|,\ldots,\log|f_r|),
\]
where the \(f_i\) are analytic units and \(F\) is smooth near their
logarithmic image. The class \(\Ploc(O)\) is defined in
\namedref{Section}{sec:introduction}. We write \(u\in\Dloc(O)\) if \(u\)
is locally a difference of two members of \(\Ploc\).
For metrics, these conditions apply to their weights in local frames.
The weight in a frame \(e\) is \(-\log\|e\|\).
\end{definition}

A change of frame changes a weight by the logarithm of the norm of an
analytic unit. That function is smooth and has zero \(d'd''\).
Continuous multiplication of positive currents is known from
\cite[Corollary 5.4.7]{CLD12} and \cite[Lemma 8.5.1]{CLD25}.
We give the coefficient construction because it also controls
weakly convergent nets.

\Needspace{14\baselineskip}
\begin{lemma}[Continuous coefficients]\label{lem:continuous-coefficients}
Let \(O\) be an ordinary open in a smooth pure \(n\)-dimensional
\(K\)-variety. Let \(T_a\) be positive currents of bidegree
\((n-q,n-q)\), \(0\leq q\leq n\), converging weakly to \(T\).
Then \(T\) is positive. For \(S=T\) and for each \(S=T_a\), every finite
continuous \(f:O\to\R\) determines a unique current \(fS\) by
\[
 \langle fS,\eta\rangle
   =\lim_{j\to\infty}\langle S,f_j\eta\rangle,
\]
where \(\eta\) is any compactly supported smooth test and \(f_j\)
are smooth functions converging uniformly to \(f\) near
\(\supp\eta\). The limit exists and is independent of the choice
of \(f_j\).

On a fixed compact test chart, the scalar coefficient measures of
\(T_a\) have eventually uniformly bounded total variations. The same
holds after wedging \(T_a\) with any fixed smooth form, in the
bidegrees where the wedge is defined. The cone of positive currents
is pointed. If \(f_a\to f\) locally uniformly, with \(f_a,f\)
finite continuous, then
\begin{equation}\label{eq:moving-coefficients}
 f_aT_a\longrightarrow fT
\end{equation}
on every compact smooth test. Nets are allowed.
\end{lemma}

\begin{proof}
Positivity of \(T\) follows by taking the limit on each positive test.
We establish the coefficient estimate locally and then use a finite
smooth partition. Fix a moment chart with \(r\) moment coordinates,
a compactly supported \((q,q)\)-form \(\eta\), and a nonnegative
smooth cutoff \(\chi\) equal to one near its support. Let \(J\) denote
the involution interchanging the two differential factors, and write
\[
 \eta_{\mathrm s}=\tfrac12\bigl(\eta+(-1)^qJ\eta\bigr),
 \qquad \eta_{\mathrm a}=\eta-\eta_{\mathrm s}.
\]
Wedging \(\eta_{\mathrm a}\) with \(\gamma\wedge J\gamma\), for an
\((r-q,0)\)-form \(\gamma\), gives zero. In the convention
\cite[Definition 5.1.1(c)]{CLD12}, both signs of
\(\eta_{\mathrm a}\) are therefore positive. A positive current
annihilates \(\eta_{\mathrm a}\), also after multiplication by a
smooth real scalar.

For symmetric forms the positivity test is a finite-dimensional
positive-semidefinite matrix condition. The form
\[
 \Omega=\frac1{q!}\left(\sum_{j=1}^r d'x_j\wedge d''x_j\right)^q
 \quad(\Omega=1\text{ if }q=0)
\]
has a positive-definite coefficient matrix. Boundedness of the
coefficient matrix of \(\eta_{\mathrm s}\) on the fixed compact set
gives \(C_\eta\) such that, for every smooth real \(h\),
\[
 C_\eta\|h\|_{\supp\eta}\chi\Omega\pm h\eta_{\mathrm s}\geq0.
\]
If \(h\) uses additional moment coordinates, pull this inequality
to a common refinement; positivity is preserved. The forms extend
by zero outside the chart. For \(S=T_a\) or \(T\), we obtain
\begin{equation}\label{eq:coefficient-estimate}
 |\langle S,h\eta\rangle|
 \leq C_\eta\|h\|_{\supp\eta}\langle S,\chi\Omega\rangle.
\end{equation}
At this stage \(h\) is smooth, so the pairing is already defined.

The right-hand evaluations for \(T_a\) converge and are eventually
bounded. Apply \eqref{eq:coefficient-estimate} to the finitely many
coordinate test forms. Their scalar functionals extend uniquely to
continuous compactly supported functions, and the Riesz theorem
represents them by signed Radon measures. The same estimate bounds
their total variations. These extensions agree on overlaps by
uniform smooth density. On a fixed test stage \(C_\eta\) is bounded
by a constant times the supremum norm of the coefficient matrix,
so the extended functionals define currents.

We can now define \(fS\), for \(S=T\) or \(S=T_a\), by multiplying
these coefficient measures by \(f\). Smooth density supplies
the functions \(f_j\) in the statement; estimate
\eqref{eq:coefficient-estimate} proves that their pairings converge
to this product independently of the approximation. The same local
estimate proves continuity on each smooth test stage.

For fixed continuous \(f\), choose smooth \(f_j\to f\) uniformly
on an enlarged compact set. For an eventually uniform \(M_\eta\),
\[
 |\langle fT_a-fT,\eta\rangle|
 \leq|\langle T_a-T,f_j\eta\rangle|
       +M_\eta\|f-f_j\|_\infty.
\]
First fix \(j\), let \(a\) tend to its limit, and then let \(j\)
increase. This proves \(fT_a\to fT\).
For varying \(f_a\), the additional term satisfies
\[
 |\langle(f_a-f)T_a,\eta\rangle|
 \leq M_\eta\|f_a-f\|_{\supp\eta}\longrightarrow0.
\]
A finite partition treats any compact test. The coefficient bounds
also control wedges with a fixed smooth form, whose coefficients are
bounded on compact charts. Finally, if both \(S\) and \(-S\) are
positive, their evaluations on \(\chi\Omega\) vanish; applying
\eqref{eq:coefficient-estimate} to \(S\) gives \(S=0\).
\end{proof}

\begin{proposition}[A fixed approachable potential]\label{prop:fixed-potential}
Let \(T\) be a positive current on \(O\) with \(d'T=d''T=0\), and let
\(u\in\Ploc(O)\). Whenever the bidegrees permit it,
\[
 M_u(T):=\ddc(uT)
\]
is positive and closed. For fixed \(u\), the operation \(M_u\) is
continuous on the cone of positive closed currents for weak
convergence. The same is true jointly when \(u_a\to u\) locally
uniformly and all \(u_a,u\) belong to \(\Ploc\).
Successive such operations are symmetric in their potentials.
\end{proposition}

\begin{proof}
On a fixed smaller neighbourhood choose smooth \(u_j\to u\) with
\(\ddc u_j\geq0\). Closedness gives
\[
 \ddc(u_jT)=\ddc u_j\wedge T\geq0.
\]
The curvature \(\ddc u_j\) is a symmetric positive \((1,1)\)-form,
hence strongly positive, so its wedge preserves positivity. \namedref{Lemma}{lem:continuous-coefficients}
and continuity of differentiation give the positive limit
\(\ddc(uT)\). Both differentials annihilate this limit.
Equation~\eqref{eq:moving-coefficients} proves the two continuity
assertions. Induction gives joint continuity for successive
operations. Simultaneous smooth positive approximations of finitely
many potentials then give symmetry from the smooth wedge product.
\end{proof}

For a continuously metrized line bundle with approachable weights, write
\(c_1(L,p)\wedge T\) for this operation in local frames. The local
definitions agree because a transition logarithm \(\ell\) is smooth,
\(\ddc\ell=0\), and \(\ddc(\ell T)=0\).

\section{Metrics and Green identities on curves}\label{sec:metrics-curves}

\subsection{Fubini--Study metrics and normalization}
\label{app:maxima}% Historical anchor for the relocated standard construction.
\begin{definition}\label{def:fs}
Let \(L\) be an ample rational line bundle on a projective \(Y/K\).
A \emph{pure Fubini--Study metric} on \(L\) is defined by an integer
\(d>0\), with \(dL\) an actual line bundle, and a generating tuple
\(s_0,\ldots,s_b\in H^0(Y,dL)\). Its weight in a local frame is
\[
 q_e=\frac1d\max_{0\leq j\leq b}\log|s_j/e^{\otimes d}|.
\]
Coefficients multiplying the sections belong to \(K\). A
\emph{uniform FS limit} is a uniform limit of these metrics.
A metric is \emph{globally smoothly approximable} if it is a uniform
limit of global smooth metrics with positive curvature forms.
\end{definition}

A model metric is \emph{nef} if, after clearing tensor denominators,
its model line bundle has nonnegative degree on every integral curve
of the special fibre. A continuous metric is \emph{semipositive}
if it is a uniform limit of nef model metrics. For an ample rational
line, this is equivalent to being a uniform pure-FS limit
\cite[Corollary 7.6 and Theorem 7.8]{BE}. All uniform differences of
metrics are scalar differences on the same specified line bundle.

We use the standard regularized maximum from
\cite[proof of Proposition 6.3.2]{CLD12}.

\begin{proposition}[Regularized maximum]\label{prop:regularized-max}
Fix an even nonnegative \(C^\infty\) function \(\rho\), supported
in \([-1,1]\), with integral one. For \(h\geq1\) and
\(\varepsilon>0\), put
\[
 R_{h,\varepsilon}(t)
 =\int_{\R^h}\max_i(t_i+\varepsilon z_i)
                    \prod_{i=1}^h\rho(z_i)\,dz.
\]
This function is smooth, convex, and nondecreasing in each variable.
It satisfies
\[
 \max_i t_i\leq R_{h,\varepsilon}(t)\leq\max_i t_i+\varepsilon,
 \qquad
 R_{h,\varepsilon}(t+c\mathbf1)=R_{h,\varepsilon}(t)+c.
\]
An entry more than \(2\varepsilon\) below the maximum is inactive
on a neighbourhood of that point. If smooth \(a_1,\ldots,a_h\)
have positive curvature, so does \(R_{h,\varepsilon}(a_1,\ldots,a_h)\).
\end{proposition}

Translation covariance gives the metric transition law, and the
inactive-coordinate property handles zeros of the generating sections.
These are precisely the features needed in the next lemma.

\begin{lemma}[Smooth approximation of FS metrics]\label{lem:smooth-fs}
A pure-FS metric, and hence a uniform FS limit, is globally smoothly
approximable. If \(p\) is globally smoothly approximable on \(L\),
if \(m>0\) is an integer for which \(mL\) is a line bundle, and if
\(s\in H^0(Y,mL)\), then for each \(N\in\R\) the metric
\[
 \rho_N=\max\{mp,\log|s|+N\}
\]
on \(mL\) is globally smoothly approximable.
\end{lemma}

\begin{proof}
For a pure-FS metric from sections of \(dL\), pull back the smooth
positive approximations of the canonical metric on
\(\OO_{\PP^b}(1)\) from \cite[Proposition 6.3.2]{CLD12}, and
divide their weights by \(d\). A diagonal choice treats uniform
FS limits.

The second assertion is immediate for \(s=0\). Otherwise, take
smooth positive \(p_j\to p\) and regularize \(\max\{mp_j,\log|s|+N\}\).
The logarithmic entry is inactive near the divisor of \(s\);
elsewhere it is smooth with zero curvature. The resulting global
metrics are positive and have error at most
\(m\|p_j-p\|_\infty+\varepsilon_j\).
This is the same regularized-maximum construction, applied to a
smooth weight and a section weight.
\end{proof}

\begin{lemma}[Degree and comparison of measures]\label{lem:degree}
On a smooth projective \(Y/K\) of dimension \(n\), globally smoothly
approximable metrics on \(L_1,\ldots,L_n\) have total mixed mass
\(\deg_K(L_1\cdots L_n)\). For uniform FS limits, the model-theoretic
mixed measure agrees with the CLD product.
\end{lemma}

The mass assertion is \cite[Corollary 6.4.4]{CLD12}.
The model comparison is
\cite[Theorem 6.9.3 and Proposition 6.9.7]{CLD12}, with the
discretely valued normalization of \cite[Section 6.9.6]{CLD12}.
For uniform FS limits, use uniform continuity of the model
measures \cite[Proposition 2.23(d) and Remark 2.24]{BGM} and of
the approachable CLD products. Rational tensor denominators are
cleared and divided on both sides.

Degrees in this paper are over the stated ground field. A variety
over a finite extension \(F/K\), regarded as a \(K\)-variety,
contributes the structural factor \([F:K]\) to intersection and
integration. Indeed, a moment map followed by
\((\mathbf G_m^r)_F\to(\mathbf G_m^r)_K\) has that degree.
This operation is distinct from scalar extension of a \(K\)-variety:
the normalized scalar-extension integration formula used next has
no extra degree factor.

\begin{definition}\label{def:curve-positive}
A continuous function on an ordinary algebraic analytic open is
\emph{curve-positive} if its pullback by every algebraic morphism
from a smooth \(K\)-curve has positive CLD curvature current on the
inverse-image open. For a continuous metric the same condition is
imposed on the pulled-back line bundle. Curves may be disconnected.
The definition uses algebraic curve maps over the given \(K\);
it does not build positivity after arbitrary field extension into
the hypothesis.
\end{definition}

\subsection{Regularization and energy in dimension one}
\begin{proposition}[Curve regularization]\label{prop:curves}
Let \(C/K\) be a smooth algebraic curve and \(W\subset C^\an\)
an ordinary open. A finite continuous CLD-positive function on \(W\)
belongs to \(\Ploc(W)\). A finite continuous function with zero
Laplacian on a connected proper smooth curve is constant.
\end{proposition}

\begin{proof}
For a finite Galois extension \(F/K\), average a nonnegative compact
smooth scalar test \(\eta\) on \(W_F\) over the Galois group.
The invariant average descends by
\cite[Proposition 14.2]{GJRforms}. It is smooth because weakly
smooth and CLD-smooth forms coincide on these curves
\cite[Remark 15.10]{GJRforms}: their residue fields are algebraic
over a finite field. If \(\overline\eta\) is the descended test,
the scalar-extension integration formula
\cite[Proposition 14.5]{GJRforms} gives
\[
 \int_{W_F}u_F\,\ddc\eta=\int_Wu\,\ddc\overline\eta.
\]
Uniform smooth density permits the continuous coefficient \(u\).
Consequently current positivity passes to finite Galois extensions.

Let \(Y\) be a compact strictly affinoid curve domain inside \(W\).
After such an extension choose a semistable skeleton \(\Gamma\)
with retraction \(\tau\), edge lengths, and positive weights \(w_e\).
For an edgewise smooth \(\varphi\), constant near vertices and zero
near boundary vertices, \(\varphi\circ\tau\), extended by zero,
is a compact smooth test. Its Laplacian has no vertex atoms, and
integration on the annular edges gives
\begin{equation}\label{eq:curve-skeleton}
 \int_\Gamma\varphi\,d\tau_*(\ddc[u]|_Y)
 =\sum_e w_e\int_e u|_e\,\varphi_e''\,ds.
\end{equation}
This is also the retraction comparison
\cite[Proposition 6.3.2]{GJRcurves}; it does not assume that \(u\)
factors through \(\tau\).

Tests inside an edge show that \(u|_e\) is convex. The positive
Radon measure \(\ddc[u]\) has finite mass on the compact domain.
Thus its edge second derivatives have finite mass and finite
one-sided derivatives at the endpoints. Integration by parts in
\eqref{eq:curve-skeleton}, allowing arbitrary constants near the
interior vertices, gives a nonnegative weighted sum of outgoing
slopes at each such vertex. The finite-graph maximum principle
therefore applies to \(u\).

To pass from graph inequalities to subharmonicity, we follow the
harmonic-comparison argument of \cite[Theorem 4.7]{Wanner}.
A harmonic comparison function becomes skeletal after semistable
refinement \cite[Section 15.6 and Proposition 15.7]{GJRforms}.
Its boundary comparison with \(u\) holds on that skeleton.
Any prescribed finite set of type-II points can be included in a
refinement after a further finite Galois extension
\cite[Remark 5.4.2]{GJRcurves}. The boundary and harmonic
compatibilities are \cite[Lemma 5.4.1 and Proposition 5.6.8]{GJRcurves}.
Density of these points and continuity prove harmonic comparison
on all of \(Y\). Hence \(u\) is Thuillier-subharmonic.

By \cite[Proposition 15.11]{GJRforms}, \(u\) has a decreasing net
of weakly smooth psh approximants on a fixed relatively compact
open. They are CLD-smooth by the finite-residue-field comparison.
On a smaller compact neighbourhood, Dini's finite-cover argument
selects a uniformly convergent sequence. On a curve, positivity of
the curvature current of a smooth function implies positivity of
its curvature form: a negative coefficient on an open
one-dimensional tropical cell is detected by a nonnegative scalar
test there; continuity handles faces. This proves approachability.

For a zero-Laplacian function on a proper curve,
\eqref{eq:curve-skeleton} makes its restriction to every refined
skeleton harmonic. A harmonic function on a connected compact
graph without boundary is constant. Refinement through arbitrary
type-II points and descent give the final assertion.
\end{proof}

The smooth Green identity is \cite[Theorem 3.12.2]{CLD12}.
We extend it and the smooth chain rule to the continuous gaps
below using curve regularization and bounded Laplacian variation.

\begin{lemma}[Green identities on curves]\label{lem:curve-green}
Let \(r,p\) be finite continuous metrics with positive Chern currents
on the same rational line \(L\) on a smooth proper curve \(C/K\).
Their gap \(g=r-p\) has smooth approximants \(a_j\to g\) uniformly
with \(\sup_j\|\ddc a_j\|_{\TV}<\infty\). The two Chern measures
have mass \(\deg_KL\), so \(\|\ddc g\|_{\TV}\leq2\deg_KL\).

On every finite real span \(E\) of such gaps and smooth functions,
\begin{equation}\label{eq:curve-green}
 \int_C a\,\ddc b=\int_C b\,\ddc a,\qquad
 Q_C(a):=-\int_C a\,\ddc a\geq0 \quad(a,b\in E).
\end{equation}
Moreover,
\[
 \Gamma_C(a)=\ddc(a^2/2)-a\,\ddc a
\]
is a positive measure of mass \(Q_C(a)\), and
\[
 \ddc\Phi(a)=\Phi'(a)\ddc a+\Phi''(a)\Gamma_C(a)
\]
for every smooth real \(\Phi\) defined near the range of \(a\).
If \(Q_C(a)=0\), then \(\ddc a=0\) and \(a\) is constant on each
connected component.
\end{lemma}

\begin{proof}
In finitely many fixed frame neighbourhoods, use
\namedref{Proposition}{prop:curves} to approximate the weights by smooth
subharmonic \(r_{i,j},p_{i,j}\). For a compact \(F\) in one such
neighbourhood, choose a nonnegative smooth cutoff \(\xi=1\) near \(F\).
Then
\[
 (\ddc r_{i,j})(F)
 \leq\int\xi\,\ddc r_{i,j}
 =\int r_{i,j}\,\ddc\xi
 \leq\|r_{i,j}\|_{\supp\ddc\xi}\|\ddc\xi\|_{\TV}.
\]
This bounds the local Laplacians of
\(g_{i,j}=r_{i,j}-p_{i,j}\). For a compactly supported
\(\rho\geq0\), the square identity gives
\[
 \int\rho\,d'g_{i,j}\wedge d''g_{i,j}
 =\int(g_{i,j}^2/2)\ddc\rho
   -\int\rho g_{i,j}\ddc g_{i,j},
\]
so the local energies are bounded as well.

Take a finite smooth partition \(\chi_i\) and set
\(a_j=\sum_i\chi_i g_{i,j}\). Cauchy--Schwarz for integration on
the curve bounds the cross terms by
\[
 \|d'\chi_i\wedge d''g_{i,j}
       +d'g_{i,j}\wedge d''\chi_i\|_{\TV}
 \leq2Q_C(\chi_i)^{1/2}
 \left(\int\rho_i\,d'g_{i,j}\wedge d''g_{i,j}\right)^{1/2},
\]
where \(\rho_i=1\) near \(\supp\chi_i\).
The product rule therefore gives the uniform global variation bound.

The mass formula is \cite[Corollary 18.1.7]{CLD25}.
Uniform convergence and the variation bound imply weak convergence
of \(\ddc a_j\) on continuous tests and
\[
 -\int a_j\ddc a_j\longrightarrow-\int a\ddc a.
\]
The cited smooth Green formula and
\(\ddc(a_j^2/2)-a_j\ddc a_j=d'a_j\wedge d''a_j\) pass to the limit.
The latter positive measures have convergent masses. The smooth
chain rule passes as well. Choosing approximations simultaneously
proves these assertions on a fixed finite span \(E\).
If \(Q_C(a)=0\), the positive measure \(\Gamma_C(a)\) is zero.
The nonnegative quadratic \(Q_C(a+th)\), for smooth \(h\) and
arbitrary \(t\in\R\), has zero linear coefficient. By Green symmetry,
\(\int h\,\ddc a=0\) for every such \(h\). Thus \(\ddc a=0\), and
\namedref{Proposition}{prop:curves} gives constancy.
\end{proof}

\section{Restriction and an ambient extension}\label{sec:restriction}

Restriction of a current to a subvariety is not a formal operation.
For continuous potentials it follows from a positive kernel in the
normal direction.

\begin{lemma}[A normal kernel]\label{lem:normal-kernel}
Let \(B\) be an ordinary open in the analytification of a smooth
\(K\)-variety of pure dimension \(d\geq1\), let
\(D_s=\{|t|<s\}\), and let \(i:B\hookrightarrow B\times_KD_s\)
be the zero section. For a fixed \(\varepsilon>0\), put
\[
 v_N=R_{2,\varepsilon}(\log|t|,-N),\qquad \beta_N=\ddc v_N.
\]
Then, for every finite continuous \(u\) on the product,
\[
 \beta_N\longrightarrow i_*[B],\qquad
 u\beta_N\longrightarrow i_*[u|_B]
\]
as currents. In particular, \(\ddc[u]\geq0\) implies
\(\ddc[u|_B]\geq0\).
\end{lemma}

\begin{proof}
The form \(\beta_N\) is smooth and strongly positive, with support in
\[
 e^{-N-2\varepsilon}\leq|t|\leq e^{-N+2\varepsilon}.
\]
The logarithm is locally integrable. Dominated convergence and
Poincar\'e--Lelong give
\(\ddc[v_N]\to\ddc[\log|t|]=i_*[B]\), with multiplicity one.
\namedref{Lemma}{lem:continuous-coefficients} permits the fixed multiplier \(u\).
For a positive \(\eta\in\Ac^{d-1,d-1}(B)\),
\(\operatorname{pr}^*\eta\wedge\beta_N\) is a positive compact test
for large \(N\). Indeed, its support lies above a fixed compact
subset of \(B\) and inside a fixed closed normal disc. Thus
\[
 \begin{aligned}
 0&\leq\langle\ddc[u],\operatorname{pr}^*\eta\wedge\beta_N\rangle\\
  &=\langle u\beta_N,\operatorname{pr}^*(\ddc\eta)\rangle
    \longrightarrow\langle\ddc[u|_B],\eta\rangle.
 \end{aligned}
\]
Fixed base and normal cutoffs make the middle expression a compact
test pairing without altering it.
\end{proof}

\begin{lemma}[A local product neighbourhood]\label{lem:tubular}
Let \(i:Z\hookrightarrow Y\) be a closed immersion of smooth
\(K\)-varieties, let \(x\in Z^\an\), and let \(r\) be the codimension
near \(x\). There are ordinary neighbourhoods \(V\subset Z^\an\)
and \(W\subset Y^\an\), a radius \(s>0\), and an analytic isomorphism
\[
 \Phi:V\times_KD_s^r\xrightarrow{\sim}W
\]
whose zero section is \(i|_V\). Consequently restriction of a finite
continuous positive potential along \(i\) preserves current positivity.
\end{lemma}

\begin{proof}
Use the \'etale local form of a smooth pair
\cite[Lemma 37.17.1, Tag 0FUE]{Stacks}. After Zariski shrinking there
is an \'etale map
\[
 \Psi=(z,F):Y\longrightarrow\Aff_K^{d+r},\qquad
 Z=\Psi^{-1}(\Aff_K^d\times\{0\}),
\]
where \(d=\dim Z\) and \(z|_Z\) is \'etale. Form
\(E=Y\times_{\Aff_K^d}Z\). The two maps
\[
 E\xrightarrow{\,p\,}Y,\qquad
 E\xrightarrow{\,q=(\operatorname{pr}_Z,F\circ p)\,}
                       Z\times\Aff_K^r
\]
are \'etale by base change. At the diagonal point \(e\) above \(x\),
both induce the identity on the completed residue field
\(\mathscr H(x)\). Berkovich's equivalence for \'etale germs
\cite[Theorem 3.4.1]{Ber93} therefore makes both analytic germs
isomorphisms. By the definition of these germs
\cite[Section 3.4]{Ber93}, they are represented by isomorphisms of
ordinary open neighbourhoods. Their composition identifies a
neighbourhood of \(x\) in \(Y^\an\) with an ordinary open
\(G\subset Z^\an\times\Aff_K^{r,\an}\), matching \(Z\) with the
zero section.

Choose a compact affinoid neighbourhood \(V_0\) of \(x\) in \(Z^\an\)
whose zero section lies in \(G\). In the product of \(V_0\) with a
fixed closed normal polydisc, the complement of \(G\) is compact
and disjoint from the zero section. If nonempty, it has a positive
minimum of \(\max_i|t_i|\). Choose \(s\) smaller than the initial
polydisc radius and, when the complement is nonempty, smaller than
that minimum. Then \(V\times D_s^r\subset G\), where \(V=\Int V_0\).
Its image is the required ordinary open \(W\).
Finally apply \namedref{Lemma}{lem:normal-kernel} in each normal coordinate;
a zero-dimensional target has no curvature condition.
\end{proof}

\begin{proposition}[Curve maps and extension]\label{prop:curve-restriction}
Let \(u\) be a finite continuous CLD-positive function on an ordinary
open \(O\subset X^\an\), where \(X/K\) is smooth. Then:
\begin{enumerate}
\item For every algebraic morphism \(f:C\to X\) from a smooth
\(K\)-curve, \(u\circ f^\an\) is CLD-positive on
\((f^\an)^{-1}(O)\).
\item Near each point of \(O\), an affine embedding of \(X\) has
a finite continuous curve-positive ambient extension \(v\) of \(u\).
\end{enumerate}
\end{proposition}

\begin{proof}
Argue componentwise. For a nonconstant map the algebraic image is a curve. Away from its
finitely many singular points,
\namedref{Lemma}{lem:tubular} restricts \(u\) to the smooth image.
\namedref{Proposition}{prop:curves} supplies smooth positive approximations,
whose pullbacks prove positivity on the source outside finitely many
closed points.

Continuity removes each such point \(P\). A sufficiently small
neighbourhood is a disc over \(K(P)\), viewed over \(K\). Choose
\(0\leq w_N\leq1\), equal to one near \(P\), with supports shrinking
to \(P\), by translating one smooth profile in \(\log|t|\).
Then
\[
 \int\ddc w_N=0,\qquad \sup_N\|\ddc w_N\|_{\TV}<\infty.
\]
The bound contains only the fixed structural degree \([K(P):K]\).
A smooth scalar \(\varphi\) is constant near \(P\), since each of
its finitely many analytic unit norms is constant there. For large \(N\),
\[
 |\langle\ddc[u],\varphi w_N\rangle|
 \leq|\varphi(P)|\,
 \sup_{\supp\ddc w_N}|u-u(P)|\,\|\ddc w_N\|_{\TV}
 \longrightarrow0.
\]
Apply positivity off \(P\) to \(\varphi(1-w_N)\).
Constant curve maps pull back \(u\) to a constant.

For the extension, choose affine coordinates \((z,y)\) of an
embedding \(X\subset\Aff_K^{d+r}\), with \(z:X\to\Aff_K^d\) \'etale.
The algebraic morphism
\[
 \Theta:X\times\Aff_K^r\longrightarrow\Aff_K^{d+r},\qquad
 (a,h)\longmapsto(z(a),y(a)+h)
\]
is \'etale: after translating the second coordinate it is a base
change of \(z\). At \((x,0)\), the map induces an isomorphism of completed residue
fields. The same germ theorem \cite[Theorem 3.4.1]{Ber93} gives an
analytic inverse branch
\[
 \iota(w)=(\rho(w),y(w)-y(\rho(w)))
\]
on one ordinary ambient open. This branch agrees with the diagonal
on \(X\). Shrinking so that \(\rho\) takes values in \(O\), set
\(v=u\circ\rho\).
For any smooth algebraic curve mapping to the ambient affine space,
the fibre product with \(\Theta\) is a smooth algebraic curve.
Its map to \(X\) is covered by the first part, and the inverse
branch identifies the relevant ordinary curve open with an open
of that fibre product. Thus \(v\) is curve-positive over \(K\).
\end{proof}

\section{Curve approximation and contact rigidity}\label{sec:contact}

We first approximate the mixed currents of approachable metrics by
algebraic curves. The coefficients are chosen over the original field
\(K\), and the normalized intersection degrees remain fixed. These
degrees supply the variation bounds needed to pass the curve Green
identities to a higher-dimensional mixed current.

\subsection{Complete-intersection approximation}
\label{subsec:curve-approximation}
\begin{proposition}[Complete-intersection approximation]
\label{prop:curve-approximation}
Let \(Y/K\) be smooth projective of dimension \(n\geq2\). For
\(1\leq i<n\), let \(p_i\) be a uniform FS limit on an ample rational
line bundle \(L_i\). Then
\[
 T=\bigwedge_{i=1}^{n-1}c_1(L_i,p_i)
\]
is a weak limit, on all compact smooth tests, of individual normalized
smooth complete-intersection curve cycles \(T_a\) defined over \(K\).
For every line bundle \(H\), each \(T_a\) has normalized \(H\)-degree
\(\deg_K(HL_1\cdots L_{n-1})\).
\end{proposition}

\begin{proof}
\emph{Step 1: a smooth cut with prescribed coefficient valuations.}
Let \(Z\subset Y\) be smooth projective of pure dimension \(d\geq1\).
First consider an ample line bundle \(M\) generated by
\(s_0,\ldots,s_b\), with FS weight \(q=\max_i\log|s_i|\).
For every \(N\geq1\), choose
\begin{equation}\label{eq:deterministic-section}
 a_{i,N}\in\pi^i(1+\pi R),\qquad
 \sigma_N=\sum_{i=0}^b a_{i,N}s_i^N
\end{equation}
so that \(D_N=\divisor_Z(\sigma_N)\) is smooth.
Such a choice exists: the pure powers \(s_i^N\) have no common
zero, and characteristic-zero Bertini gives a nonempty Zariski-open
smooth locus in their coefficient space. The product
\(\prod_i\pi^i(1+\pi R)\) is Zariski dense, since a nonzero polynomial
cannot vanish on a product of infinite subsets of \(K\).
If \(Z\) has several components, impose the finitely many smoothness
conditions simultaneously. These choices are made independently
of all test forms.

\emph{Step 2: ties have zero mass for each smooth test.}
Fix a smooth \((d-1,d-1)\)-test \(\eta\) on \(Z^\an\), and let
\(\nu_\eta\) be the signed measure defined by \(\ddc\eta\).
On each component, discard the generators that vanish identically,
retaining the original indices \(i\).
By \cite[Lemma 3.2.5]{CLD12}, the compact support of \(\ddc\eta\)
avoids the zero divisors of the remaining generators.
Cover it by finitely many compact moment charts, refined by the
unit ratios \(s_i/s_j\). On each cell of the resulting finite
polyhedral complexes, the functions
\(\ell_{ij}=\log|s_i/s_j|\) are affine.

The absolute-value tropical integration formula
\cite[Section 3.7 and Proposition 3.7.2]{CLD12}
implies that \(|\nu_\eta|\) gives zero mass to inverse images of
proper affine hyperplanes in the \(d\)-dimensional cells;
lower-dimensional cells carry no top-degree mass.
Equivalently, on each compact chart one can dominate both signs
of its Lagerberg representative by a positive top form and apply
\cite[Proposition 11.1(3)]{GJRforms}. This is a statement about
total variation, not just about the signed pushforward.

Since \(|a_{i,N}|=|\pi|^i\), two summands in
\eqref{eq:deterministic-section} have the same norm exactly when
\[
 N\ell_{ij}=(i-j)\lambda.
\]
If \(\ell_{ij}\) is nonconstant on a top-dimensional cell, this
is a null hyperplane. If it is the constant \(c\), the equality
is impossible for \(c=0\) and holds for at most one positive
integer \(N\) when \(c\ne0\).
There are finitely many cells and pairs. Thus, for all sufficiently
large \(N\), the largest summand is unique at
\(|\nu_\eta|\)-almost every point.

\emph{Step 3: the one-cut estimate.}
Put \(w_N=Nq-\log|\sigma_N|\geq0\).
Unique dominance in the ultrametric inequality gives
\[
 \log|\sigma_N|
   =\max_i\{N\log|s_i|-i\lambda\},\qquad
 0\leq w_N\leq b\lambda
 \quad |\nu_\eta|\text{-almost everywhere}
\]
for all sufficiently large \(N\).
Poincar\'e--Lelong therefore gives
% Preserve the old source anchor although the argument is now deterministic.
\begin{equation}\label{eq:haar-error}
 \begin{split}
 \left|\left\langle
 \frac{[D_N]}N-c_1(M,q)\wedge[Z],\eta
 \right\rangle\right|
 &=\frac1N\left|\int_{Z^\an}w_N\,\ddc\eta\right|\\
 &\leq\frac{b\lambda}{N}\|\ddc\eta\|_{\TV}.
 \end{split}
\end{equation}
Here the threshold in \(N\) depends on \(\eta\), but the chosen
sequence \((\sigma_N)\) does not. The formula is interpreted
intrinsically on \(Z\), or after pushforward to \(Y\).
It proves one-cut convergence on every original smooth test.
The bound is a statement about integration against \(\ddc\eta\):
\(w_N\) is unbounded along \(D_N\). The argument does not give
the same estimate against an arbitrary positive background current,
whose coefficient measures may charge the tie loci.

\emph{Step 4: successive cuts and uniform limits.}
For pure-FS metrics \(p_i\), let the generating tuple lie in
\(d_iL_i\), and put \(\alpha_i=c_1(L_i,p_i)\).
At the \(i\)-th cut, restrict that tuple to the preceding smooth
intersection and apply Steps 1--3. Its divisor has class
\(N_id_iL_i\), and its current is divided by \(N_id_i\).
The Bertini conditions can also be imposed on the ambient divisors.

For clarity, let \(\mathcal C_k\) be the currents of these smooth
codimension-\(k\) complete intersections, normalized by
\(\prod_{i=1}^kN_id_i\), with \(\mathcal C_0=\{[Y]\}\).
The one-cut limit places
\(\alpha_k\wedge S\) in the weak closure of \(\mathcal C_k\)
for every \(S\in\mathcal C_{k-1}\).
Continuity in the positive background
(\namedref{Proposition}{prop:fixed-potential}) then gives inductively
\[
 \alpha_1\wedge\cdots\wedge\alpha_k
 \in\overline{\mathcal C_k}^{\,\mathrm{weak}}.
\]
Indexing a weak neighbourhood by finitely many smooth tests and
an error tolerance gives the asserted net of individual curves.
Each member has normalized \(H\)-degree
\[
 \frac{\deg_K\!\left(H\prod_{i=1}^{n-1}(N_id_iL_i)\right)}
      {\prod_{i=1}^{n-1}N_id_i}
 =\deg_K(HL_1\cdots L_{n-1}).
\]
Finally, approximate the \(p_i\) uniformly by pure-FS metrics and
take the union of the corresponding families \(\mathcal C_{n-1}\)
over all approximating tuples and tensor degrees. Every member
is still an individual normalized smooth complete intersection
with the displayed degree. Joint continuity in
\namedref{Proposition}{prop:fixed-potential} puts the limiting mixed
current in the weak closure of this union.
\end{proof}

\subsection{Concentration on a divisor}\label{subsec:divisor-limit}
The next limit uses only continuity in the positive background. In
particular, the other metrics remain fixed while the truncated
section metric tends to its divisor current.

\begin{lemma}[Concentration on a smooth divisor]\label{lem:divisor-limit}
Let \(Y/K\) be smooth projective of dimension \(n\geq1\), and let
\(s\) define a smooth effective Cartier divisor
\(i:D\hookrightarrow Y\) in a line bundle \(M\).
Let \(q,p_1,\ldots,p_{n-1}\) be globally smoothly approximable
metrics on \(M,L_1,\ldots,L_{n-1}\), respectively. Then
\begin{equation}\label{eq:divisor-limit}
 \begin{split}
 &c_1\!\left(M,\max\{q,\log|s|+N\}\right)
   \wedge\bigwedge_{j=1}^{n-1}c_1(L_j,p_j)\\
 &\hspace{2em}\longrightarrow
 i_*\!\left(\bigwedge_{j=1}^{n-1}
              c_1(L_j|_D,p_j|_D)\right)
 \end{split}
\end{equation}
as \(N\to\infty\), on every continuous scalar test on \(Y^\an\).
\end{lemma}

\begin{proof}
Put \(\rho_N=\max\{q,\log|s|+N\}\), which is smoothly approachable by
\namedref{Lemma}{lem:smooth-fs}. In a local frame,
\(\rho_N-N=\max\{q-N,\log|s|\}\searrow\log|s|\).
Local integrability and Poincar\'e--Lelong give
\(c_1(M,\rho_N)\to[D]\). Apply the fixed-potential continuity of
\namedref{Proposition}{prop:fixed-potential} in the remaining slots.
The projection formula for approachable products
\cite[Proposition 8.7.5]{CLD25} identifies the limit with the
right-hand side of \eqref{eq:divisor-limit}.
All measures are positive and have the common mass
\(\deg_K(ML_1\cdots L_{n-1})\), by \namedref{Lemma}{lem:degree}.
Uniform smooth density extends convergence to continuous tests.
\end{proof}

\subsection{Contact rigidity}\label{subsec:contact-proof}
The following criterion compares two metrics on the same ample line.
Only the lower metric is assumed to be a uniform FS limit; the upper
metric is tested on algebraic curves. The proof uses the preceding
curve approximation and divisor limit, independently of envelope continuity.

\begin{theorem}[Contact rigidity]\label{thm:contact}
Let \(Y/K\) be smooth, projective, and integral of dimension
\(n\geq1\), and let \(L\) be ample. Let \(p\) be a uniform FS
limit on \(L\), and let \(r\) be a finite continuous metric on
\(L\) whose pullback by every morphism from a smooth algebraic
\(K\)-curve has positive Chern current. If
\[
 g:=r-p\geq0,
 \qquad
 \int_{Y^\an}g\,c_1(L,p)^n=0,
\]
then \(r=p\).
\end{theorem}

\begin{proof}
\emph{Step 1: pass from curves to mixed measures.}
Put
\[
 \alpha=c_1(L,p),\qquad T=\alpha^{n-1},\qquad
 \mu=\alpha^n,\qquad D=\deg_K(L^n)>0.
\]
For \(n>1\), approximate \(T\) by the normalized smooth curve
cycles \(T_a\) of \namedref{Proposition}{prop:curve-approximation}; for
\(n=1\), take \(T_a=[Y]\). The pullbacks of \(r\) and \(p\)
to each curve have positive Chern measures. Their normalized
pushforwards \(\sigma_a,\mu_a\) have mass \(D\), and
\[
 \nu_a=\ddc(gT_a)=\sigma_a-\mu_a,
 \qquad\|\nu_a\|_{\TV}\leq2D.
\]
By \namedref{Proposition}{prop:fixed-potential}, \(\mu_a\to\mu\).
\namedref{Lemma}{lem:continuous-coefficients} gives
\(\nu_a\to\nu:=\ddc(gT)\) on smooth tests. The variation bounds
extend these limits to continuous scalar tests. Thus
\(\sigma_a\to\sigma:=\nu+\mu\), where \(\sigma\) is a positive
measure of mass \(D\).

\emph{Step 2: use zero energy to annihilate the mixed Laplacian.}
We pass the curve Green identities of \namedref{Lemma}{lem:curve-green}
to the limit. For a fixed smooth function \(h\),
\namedref{Lemma}{lem:continuous-coefficients} bounds the total variations of
\(\ddc h\wedge T_a\): a finite collection of local positive test
evaluations bounds the fixed coefficient form \(\ddc h\).
Consequently,
\begin{equation}\label{eq:contact-energy}
 \int g\,\ddc(hT)=\int h\,d\nu,
 \qquad
 -\int(g+th)\ddc((g+th)T)\geq0\quad(t\in\R).
\end{equation}
All integrals are over \(Y^\an\). The contact assumption implies
\[
 0\leq-\int g\,d\nu
   =\int g\,d\mu-\int g\,d\sigma\leq0.
\]
The quadratic polynomial in \(t\) from
\eqref{eq:contact-energy} therefore has zero constant term.
Green symmetry writes its linear term as \(-2t\int h\,d\nu\).
Nonnegativity for every \(t\in\R\) forces that coefficient to
vanish. Since this holds for every smooth \(h\), we have
\(\nu=0\).

If \(\rho\) is any finite continuous curve-positive metric on
\(mL\), set \(b=\rho-mp\). Its normalized curve Laplacians have
variations at most \(2mD\). The same passage of Green symmetry,
now for the two continuous gaps \(g\) and \(b\), gives
\begin{equation}\label{eq:contact-green}
 \int g\,\ddc(bT)=\int b\,d\nu=0.
\end{equation}
Only products with the approachable lower metric \(p\) have been
used; the upper metric \(r\) enters through its curve restrictions.

\emph{Step 3: restrict the contact identity and argue by dimension.}
When \(n=1\), \namedref{Lemma}{lem:curve-green} and the zero energy
already proved make \(g\) constant. Its integral against the
positive measure \(\mu\) of mass \(D\) makes that constant zero.
Suppose \(n>1\) and proceed by induction. Fix a closed point
\(x\in Y\). For sufficiently large \(m\), choose a smooth
\(H=\divisor(s)\) in \(mL\) through \(x\). To impose this
condition, use \(H^0(Y,mL\otimes\mathcal I_x)\): for \(m\gg0\)
it is generated away from \(x\) and realizes arbitrary first jets
at \(x\). A nonzero first jet ensures smoothness there, and
characteristic-zero Bertini ensures smoothness elsewhere. The
nonempty open conditions admit a \(K\)-rational choice since
\(K\) is infinite.

Set \(\rho_N=\max\{mp,\log|s|+N\}\). These metrics are smoothly
approximable by \namedref{Lemma}{lem:smooth-fs}, hence curve-positive.
Equation~\eqref{eq:contact-green}, together with
\(\int g\,d\mu=0\), yields
\[
 \int_{Y^\an}g\,c_1(mL,\rho_N)\wedge\alpha^{n-1}=0.
\]
\namedref{Lemma}{lem:divisor-limit} now gives
\[
 \int_{H^\an}g|_H\,c_1(L|_H,p|_H)^{n-1}=0.
\]
The integrand is nonnegative. It therefore has zero integral on
each smooth integral component of \(H\). The curve condition
passes to that component by composition, \(L|_H\) is ample, and
\(p|_H\) is a uniform FS limit by restriction of the generating
tuples. Induction gives \(g(x)=0\). Closed points give a dense
set of rigid points of \(Y^\an\); continuity implies \(g=0\)
everywhere.
\end{proof}

\section{Envelopes on a fixed regular model}\label{sec:envelopes}

For an ample line bundle \(L\) on a projective \(K\)-variety and a
continuous metric \(r\), set
\begin{equation}\label{eq:envelope-definition}
 P_L(r)=\sup\{q\le r:q\text{ is a continuous semipositive metric on }L\}.
\end{equation}
We use the standard finiteness, translation, scaling, and monotonicity
properties of this envelope \cite[Remark~6.2 and Proposition~6.3]{BGM}.
In particular,
\begin{equation}\label{eq:envelope-lipschitz}
 \norm{P_L(r)-P_L(r')}_\infty\le \norm{r-r'}_\infty.
\end{equation}
The envelope is not assumed continuous at this stage.

\subsection{Test ideals of admissible section divisors}\label{subsec:test-ideals}

Let \(\cX/R\) be regular, integral, projective and flat, with smooth
generic fibre \(Y\) of dimension \(n\ge1\). Extend \(L\) to a line bundle
\(\cL\) on \(\cX\), and write \(\theta\) for its model metric.
Such an extension exists because the closure of a generic Cartier divisor
is Cartier on the regular scheme \(\cX\).
After clearing a fixed tensor denominator and adjusting the reference
identification by a power of \(\pi\), a model metric \(r\) can be written
\begin{equation}\label{eq:model-obstacle}
 r=\theta+\log|\cE|,
 \qquad \cE=\OO_{\cW}(-F)\subseteq\OO_{\cW},
\end{equation}
where \(f:\cW\to\cX\) is a normal projective model inducing the identity
on \(Y\), and \(F\) is an effective vertical Cartier divisor.
Put \(p=P_L(r)\).

\begin{proposition}[The test-ideal estimate]\label{prop:test-ideal-estimate}
There are coherent vertical ideals \(\mathfrak j_m\subseteq\OO_{\cX}\),
for \(m\ge1\), and an integer \(B\ge0\), independent of \(m\), such that
\begin{equation}\label{eq:test-ideal-bounds}
 \pi^B\mathfrak j_m\OO_{\cW}\subseteq\cE^m,\qquad
 p\le\theta+\psi_m\le r+\frac{B\lambda}{m},
 \qquad \psi_m=\frac1m\log|\mathfrak j_m|.
\end{equation}
For every globally generated ample Cartier divisor \(H\) on \(\cX\),
\begin{equation}\label{eq:test-ideal-generation}
 \mathfrak j_m\cL^{\otimes m}\otimes
 \OO_{\cX}\bigl(K_{\cX}+(n+1)H\bigr)
 \quad\text{is globally generated.}
\end{equation}
\end{proposition}

\begin{proof}
For \(l\ge1\), let
\[
 V_l=H^0\bigl(\cW,f^*\cL^{\otimes l}\otimes\cE^l\bigr).
\]
Since \(f_*\OO_{\cW}=\OO_{\cX}\), every member of \(V_l\) is a section
of \(\cL^{\otimes l}\) on \(\cX\). Normality also gives
\begin{equation}\label{eq:weighted-unit-ball}
 V_l=\{s\in H^0(Y,lL):\norm{s}_{lr,\sup}\le1\}.
\end{equation}
Indeed, a generic section has no horizontal poles, while the norm
inequalities at vertical divisorial points are exactly the remaining
codimension-one extension conditions on \(\cW\).
For \(0\ne s\in V_l\), denote its effective divisor on \(\cX\) by \(D_s\).
Then
\[
 D_s\sim l\cL,\qquad f^*D_s\ge lF.
\]
Define
\begin{equation}\label{eq:divisor-sum}
 \mathfrak j_m
 =\sum_{l\ge1}\ \sum_{0\ne s\in V_l}
       \tau_{\cX}\!\left(\frac mlD_s\right).
\end{equation}
This sum equals a finite subsum by Noetherianity.
For a sufficiently large \(l\), a generating tuple of \(H^0(Y,lL)\)
can be scaled into \(V_l\). At every point of \(Y\) one of these
sections is nonvanishing. Generic-fibre comparison with multiplier
ideals \cite[Corollary~7.21(e)]{BMP} therefore shows that
\(\mathfrak j_m|_Y=\OO_Y\). Thus \(\mathfrak j_m\) is vertical.

\smallskip\noindent
\emph{Lower bound.}
Write \(\tau_0=\tau_{\cX}(0)\). This fixed ideal contains a power of \(\pi\),
since its generic restriction is \(\OO_Y\).
Cartier translation and subadditivity on the regular scheme \(\cX\)
give, for every \(k\ge1\),
\begin{equation}\label{eq:lower-ideal}
 \OO_{\cX}(-kmD_s)\tau_0
 =\tau_{\cX}(kmD_s)
 \subseteq
 \tau_{\cX}\!\left(\frac mlD_s\right)^{kl}
 \subseteq\mathfrak j_m^{kl}.
\end{equation}
Here we use \cite[Proposition~7.16(b) and Theorem~7.22(e)]{BMP}.
Taking logarithmic norms, dividing by \(kml\), and letting \(k\to\infty\)
yields
\[
 \frac1l\log\norm{s}_{l\theta}\le\psi_m.
\]
If a pure-FS metric \(q\) lies below \(r\), its defining sections belong
to the corresponding \(V_l\) by \eqref{eq:weighted-unit-ball}; taking their
maximum gives \(q\le\theta+\psi_m\).
Every continuous semipositive competitor below \(r\) is a uniform
limit of pure-FS metrics shifted down by constants in
\(\lambda\mathbf Q\) tending to zero
\cite[Corollary~7.6 and Theorem~7.8]{BE}.
The lower bound in \eqref{eq:test-ideal-bounds} follows.

\smallskip\noindent
\emph{Upper bound.}
Choose compatible canonical divisors and set
\(\Delta_{\cW}=f^*K_{\cX}-K_{\cW}\).
The fractional test ideal \(\tau_{\cW}(\Delta_{\cW})\) is defined because
\(K_{\cW}+\Delta_{\cW}\) is Cartier
\cite[Definition~7.20]{BMP}.
Birational inclusion, monotonicity, and Cartier translation give
\begin{align}
 \tau_{\cX}\!\left(\frac mlD_s\right)\OO_{\cW}
 &\subseteq
 \tau_{\cW}\!\left(\Delta_{\cW}+\frac ml f^*D_s\right) \notag\\
 &\subseteq
 \tau_{\cW}(\Delta_{\cW}+mF)
 =\cE^m\tau_{\cW}(\Delta_{\cW}).
 \label{eq:birational-bound}
\end{align}
The cited properties are
\cite[Proposition~7.16(a),(b) and Theorem~7.22(d)]{BMP}.
The fixed fractional ideal on the right is coherent and restricts to
\(\OO_Y\). A single power \(\pi^B\) clears its denominators on a finite
affine cover of \(\cW\). Summing \eqref{eq:birational-bound} gives
\(\pi^B\mathfrak j_m\OO_{\cW}\subseteq\cE^m\), and hence the upper bound.

\smallskip\noindent
\emph{Global generation.}
Each effective boundary \(\Delta=(m/l)D_s\) is
\(\mathbf Q\)-linearly equivalent to \(m\cL\).
For \(M=m\cL+K_{\cX}+H\), one has
\(M-K_{\cX}-\Delta\sim_{\mathbf Q}H\).
Since the special fibre has dimension \(n\),
\cite[Corollary~7.24]{BMP} makes
\(\tau_{\cX}(\Delta)\cL^{\otimes m}(K_{\cX}+(n+1)H)\)
globally generated. A finite subsum in \eqref{eq:divisor-sum}
proves \eqref{eq:test-ideal-generation}.
\end{proof}

\subsection{Uniform approximation and orthogonality}\label{subsec:envelope-proof}

\begin{theorem}[Envelopes on a regular model]\label{thm:envelope}
Let \(\cX/R\) be regular, integral, projective and flat, with smooth generic
fibre \(Y\) of dimension \(n\), and let \(L\) be an ample rational line
bundle on \(Y\).
For every finite continuous metric \(r\) on \(L\), the envelope
\(p=P_L(r)\) is continuous, is a uniform limit of pure-FS metrics, and
satisfies
\begin{equation}\label{eq:regular-model-orthogonality}
 \int_{Y^{\an}}(r-p)c_1(L,p)^n=0.
\end{equation}
Here the measure is the CLD product.
For a model obstacle, after clearing fixed tensor denominators,
there are pure-FS metrics \(q_\ell\), for all sufficiently large
integers \(\ell\), such that
\begin{equation}\label{eq:regular-model-approximation}
 0\le q_\ell-p\le \frac C\ell,
\end{equation}
where \(C\) depends on the obstacle and the fixed auxiliary data.
\end{theorem}

\begin{proof}
If \(n=0\), then \(Y=\operatorname{Spec}F\) for a finite extension
\(F/K\), and \(Y^\an\) is a point. In a fixed frame, weights of pure-FS
metrics include \(\lambda\mathbf Q\), so every continuous metric is
semipositive and \(p=r\). A model metric is pure-FS after clearing its
fixed tensor denominator, which also gives the stated approximation.
Assume henceforth that \(n\ge1\).

First treat a model obstacle in the form \eqref{eq:model-obstacle}.
Choose a globally generated ample divisor \(H\) sufficiently positive that
\[
 \cA=\OO_{\cX}\bigl(K_{\cX}+(n+1)H\bigr)
\]
is ample and globally generated. Put \(A=\cA|_Y\), and let \(q_A\)
be its model FS metric.
Choose \(k\ge1\) such that \(kL-A\) is very ample and a pure-FS metric
\(\nu\) defined by a generating tuple on that line bundle.
Multiplying this tuple by a sufficiently large negative power of \(\pi\),
arrange that
\[
 h:=\frac{q_A+\nu}{k}\ge r.
\]
By \namedref{Proposition}{prop:test-ideal-estimate}, the generators of
\(\mathfrak j_m\cL^{\otimes m}\cA\) define a pure-FS metric
\[
 t_m=m\theta+q_A+\log|\mathfrak j_m|
 \quad\text{on }mL+A.
\]
Tensoring with the tuple defining \(\nu\) gives a pure-FS metric on \(L\),
\[
 h_m=\frac{t_m+\nu}{m+k}
     =\frac m{m+k}(\theta+\psi_m)+\frac k{m+k}h.
\]
Consequently,
\[
 p\le h_m\le r+\frac C{m+k},
 \qquad C=B\lambda+k\norm{h-r}_\infty.
\]
The semipositive metric \(h_m-C/(m+k)\) lies below \(r\), hence below
\(p\). Thus
\[
 0\le h_m-p\le\frac C{m+k}.
\]
These metrics are represented by generating sections in degree \(m+k\).
Relabeling \(\ell=m+k\) proves \eqref{eq:regular-model-approximation}.
A fixed tensor denominator is divided out using
\(P_{dL}(dr)=dP_L(r)\).

For a continuous obstacle, use uniform density of model metrics
\cite[Proposition~5.16(vi) and Theorem~5.18]{BE}.
Equation~\eqref{eq:envelope-lipschitz} shows that their envelopes converge
uniformly to \(P_L(r)\), and a diagonal choice gives pure-FS approximation.
Finally, \(Y\) is geometrically reduced, so
\cite[Theorem~6.7]{BGM} gives \eqref{eq:regular-model-orthogonality}.
\namedref{Lemma}{lem:degree} identifies its measure with the CLD product.
\end{proof}

\begin{corollary}[Projective space]\label{cor:projective-envelope}
Let \(a,b\ge1\) and \(L=\OO_{\PP^a_K}(b)\).
The conclusions of \namedref{Theorem}{thm:envelope} hold for every continuous metric
on \(L\). For a model obstacle represented as in
\eqref{eq:model-obstacle} on a model above \(\PP^a_R\), the generators
of \(\mathfrak j_m\OO(bm)\) give pure-FS metrics \(q_m\) satisfying
\[
 0\le q_m-P_L(r)\le B\lambda/m.
\]
\end{corollary}

\begin{proof}
Take \(\cX=\PP^a_R\).
For the more precise assertion, choose the hyperplane divisor \(H\);
then \(K_{\cX}+(a+1)H=0\).
Equation~\eqref{eq:test-ideal-generation} makes
\(q_m=\theta+m^{-1}\log|\mathfrak j_m|\) pure-FS.
By \eqref{eq:test-ideal-bounds}, \(p\le q_m\le r+B\lambda/m\);
the envelope property applied to \(q_m-B\lambda/m\) finishes the proof.
\end{proof}

\section{Proof of local regularization}\label{sec:local-proof}
The next lemma places a local curve-positive function inside one
global metric on projective space. It uses only algebraic sections
and the maximum property on curves.

\subsection{A projective extension}\label{subsec:projective-extension}

\begin{lemma}[Projective extension]\label{lem:projective-extension}
Let \(W\subset\Aff_K^{a,\an}\) be an ordinary open, \(a\ge1\),
and let \(v:W\to\mathbf R\) be finite, continuous and curve-positive.
For every \(x\in W\), there are \(d\ge1\), a finite continuous
curve-positive metric \(r\) on \(\OO_{\PP^a_K}(d)\), a section \(s_0\)
nonvanishing near \(x\), and a constant \(c\in\mathbf R\), such that
\[
 -\log\norm{s_0}_r=v+c
 \quad\text{on an ordinary neighbourhood of }x.
\]
\end{lemma}

\begin{proof}
Set \(c=\lambda/4-v(x)\).
After shrinking \(W\), continuity gives
\begin{equation}\label{eq:normalization}
 0<v+c<\lambda/2\quad\text{on }W.
\end{equation}
Choose a compact polynomial norm box
\[
 C=\{|F_i|\le|c_i|,\ |G_j|\ge|d_j|\}\subset W
\]
with all defining inequalities strict at \(x\), including upper bounds
on the affine coordinates.
Here \(F_i,G_j\in K[z_1,\ldots,z_a]\) and \(c_i,d_j\in K^\times\).
Such boxes form a neighbourhood basis: choose thresholds in
\(|K^\times|^{\mathbf Q}\) inside an ordinary basic neighbourhood and
encode them by powers of the polynomials.

For \(N\ge1\) sufficiently large, every logarithmic term in the following
maximum is negative on a neighbourhood of \(x\):
\begin{equation}\label{eq:collar-weight}
 b_0=\max\left\{
 0,\ \log\left|\frac{F_i^N}{\pi c_i^N}\right|,
       \log\left|\frac{d_j^N}{\pi G_j^N}\right|
 \right\}
\end{equation}
On the open where every \(G_j\) is nonzero, this function
satisfies \(b_0=0\) near \(x\) and \(\{b_0\le\lambda\}=C\).

Put \(t_0=\prod_jG_j^N\), with the empty product equal to \(1\).
Clear the denominators in \eqref{eq:collar-weight} and homogenize
\[
 t_0,\qquad
 \frac{t_0F_i^N}{\pi c_i^N},\qquad
 \frac{t_0d_j^N}{\pi G_j^N}
\]
to a common degree \(d\), chosen strictly larger than \(\deg t_0\).
Write \(s_0\) for the homogenization of \(t_0\).
Add sufficiently small \(K\)-multiples of
\(Z_0^d,\ldots,Z_a^d\), so that their ratios to \(s_0\) have norm
less than \(1\) on \(C\).
The resulting tuple generates \(\OO(d)\); let \(q\) be its FS metric.
Then
\[
 b:=-\log\norm{s_0}_q\in[0,+\infty]
\]
equals \(b_0\) on \(C\), vanishes near \(x\), and satisfies
\begin{equation}\label{eq:collar-sublevel}
 \{b\le\lambda\}=C.
\end{equation}
Indeed, \(s_0\) vanishes at infinity because \(d>\deg t_0\), and the
remaining original ratios impose precisely the inequalities defining \(C\).
The added sections preserve those inequalities on \(C\).
We interpret \(b=+\infty\) on the zero divisor of \(s_0\).

The ordinary opens
\[
 U=\{b<\lambda\},\qquad V=\{b>\lambda/2\}
\]
cover \(\PP_K^{a,\an}\), and \(U\subset W\).
On \(U\), define a metric by its \(s_0\)-weight
\(\max\{v+c,b\}\); on \(V\), use \(q\).
The definitions agree on \(U\cap V\) by \eqref{eq:normalization}.
They therefore give a finite continuous metric \(r\).
Near \(x\), one has \(b=0<v+c\), so its weight is \(v+c\).

On every smooth algebraic \(K\)-curve, the weights just used are
continuous subharmonic functions or maxima of two such functions.
Their maxima are subharmonic by the curve regularization of
\namedref{Proposition}{prop:curves} and the regularized maximum construction.
Thus \(r\) is curve-positive.
\end{proof}

\subsection{Completion of the proof}\label{subsec:regularization-proof}

\begin{proof}[Proof of \namedref{Theorem}{thm:main}]
Local uniform convergence of smooth positive potentials implies current
positivity by testing against \(\ddc\eta\).
Conversely, fix \(x\in O\) on a positive-dimensional component of \(X\).
\namedref{Proposition}{prop:curve-restriction} supplies an ordinary ambient open
\(W\subset\Aff_K^{a,\an}\) and a finite continuous curve-positive
function \(v\) agreeing with \(u\) along \(X^{\an}\cap W\).
Apply \namedref{Lemma}{lem:projective-extension}, and retain its \(r,s_0,c\).

By \namedref{Corollary}{cor:projective-envelope}, the envelope \(p=P_{\OO(d)}(r)\) is a uniform
FS limit and
\[
 p\le r,\qquad
 \int_{\PP_K^{a,\an}}(r-p)c_1(\OO(d),p)^a=0.
\]
\namedref{Theorem}{thm:contact} gives \(p=r\).
By \namedref{Lemma}{lem:smooth-fs}, choose global smooth positive metrics \(r_j\) with
\(\norm{r_j-r}_\infty\le4^{-j}\).
On one fixed ordinary neighbourhood \(V_x\) of \(x\) in \(X^{\an}\)
where the weight identity holds, put
\[
 u_j=-\log\norm{s_0}_{r_j}\big|_{V_x}-c.
\]
These functions are smooth, have positive curvature, and satisfy
\(\norm{u_j-u}_{V_x}\le4^{-j}\).
For the decreasing assertion, define
\[
 w_j=u_j+2\cdot4^{-j}.
\]
Then
\[
 u+4^{-j}\le w_j\le u+3\cdot4^{-j},
 \qquad w_{j+1}\le u+\tfrac34\,4^{-j}<w_j.
\]
Thus \(w_j\downarrow u\) uniformly on the same \(V_x\).
Constant translations preserve curvature.
On a zero-dimensional component, every finite continuous function is
locally constant and already smooth.
\end{proof}

\section{Consequences for continuous potentials}\label{sec:consequences}

We first apply local regularization to the operation constructed in
\namedref{Proposition}{prop:fixed-potential}. The background current remains
arbitrary; in particular, it need not itself be a product of curvature
currents.

\subsection{Proof of the mixed-current theorem}
\label{subsec:mixed-current-proof}
\begin{proof}[Proof of \namedref{Theorem}{thm:background-products}]
\namedref{Theorem}{thm:main} places every \(u_i\) in \(\Ploc(O)\).
Iterating \namedref{Proposition}{prop:fixed-potential} in
\eqref{eq:recursive-product} proves positivity, closedness, symmetry,
and joint continuity, including nets. Locality, additivity, and
positive homogeneity follow from the same recursion. A positive
current of top degree is a Radon measure.

It remains to prove independence of the potentials. Let \(h\) be
finite continuous with \(\ddc[h]=0\), and let \(T\) be any positive
closed current for which the bidegrees permit the operation. Both
\(h\) and \(-h\) have positive curvature currents, so
\namedref{Theorem}{thm:main} and \namedref{Proposition}{prop:fixed-potential} give
\[
 \ddc(hT)\geq0,\qquad
 -\ddc(hT)=\ddc((-h)T)\geq0.
\]
The positive cone is pointed by \namedref{Lemma}{lem:continuous-coefficients}; hence
\begin{equation}\label{eq:continuous-kernel}
 \ddc[h]=0\quad\Longrightarrow\quad\ddc(hT)=0.
\end{equation}
If \(\ddc[w]=\ddc[u_i]\), apply this implication to \(h=w-u_i\)
and to the positive closed product of the other slots. Symmetry permits
us to place the changed slot first, proving the assertion.
\end{proof}

\subsection{Chern products and closure operations}
\begin{corollary}[Continuous Chern products]
\label{cor:continuous-chern-products}
Let \(X/K\) be smooth quasi-projective of pure dimension \(n\), let
\(O\subset X^\an\) be an ordinary open, and let \(S\) be a positive
closed CLD current of bidegree \((p,p)\) on \(O\). Let
\(\overline L_1,\ldots,\overline L_m\) be continuously metrized
analytic line bundles on \(O\), with finite local weights and positive
CLD curvature currents, where \(m\leq n-p\). Their local products
glue to a positive closed current
\[
 c_1(\overline L_1)\wedge\cdots\wedge c_1(\overline L_m)\wedge S.
\]
It depends only on the curvature currents and \(S\), agrees with the
smooth wedge product when all weights are smooth, and is a positive
Radon measure if \(m+p=n\). The same assertions hold for rational line
bundles after clearing and dividing their tensor denominators.
\end{corollary}

\begin{proof}
Apply \namedref{Theorem}{thm:background-products} in local frames. On overlaps,
weights differ by logarithms of analytic units, whose curvature is zero
by Poincar\'e--Lelong. Equation~\eqref{eq:continuous-kernel} identifies
the products. The remaining assertions, including rational
normalization, follow from the same theorem.
\end{proof}

\begin{corollary}[Products with a smooth subvariety]
\label{cor:smooth-subvariety-products}
Let \(i:Z\hookrightarrow X\) be a closed immersion of smooth
quasi-projective \(K\)-varieties, with \(X\) of pure dimension \(n\)
and \(Z\) of pure dimension \(d\).
Let \(O\subset X^\an\) be ordinary open, put
\(O_Z=(i^\an)^{-1}(O)\), and let \(u_1,\ldots,u_m\) be finite
continuous functions on \(O\) with \(\ddc[u_j]\geq0\), where
\(m\leq d\). Write \([Z]_O=i_*[O_Z]\) for the integration current
on \(O\), and write \(\widetilde u_j=u_j|_{O_Z}\). Then
\begin{equation}\label{eq:subvariety-projection}
 \mathcal B_{[Z]_O}(u_1,\ldots,u_m)
 =i_*\mathcal B_{[O_Z]}(\widetilde u_1,\ldots,\widetilde u_m).
\end{equation}
\end{corollary}

\begin{proof}
Positive smooth approximants from \namedref{Theorem}{thm:main} restrict to
positive smooth approximants on \(O_Z\). The assertion is then the
projection formula for approachable mixed products
\cite[Proposition 8.7.5]{CLD25}, applied to the closed immersion \(i\).
\end{proof}

Formula~\eqref{eq:subvariety-projection} identifies a product with the
integration current of \(Z\) with the intrinsic product on \(Z\).
It does not require a pullback operation for arbitrary currents.

\begin{corollary}[Closure operations]\label{cor:closure-operations}
Let \(u_1,\ldots,u_r\) be finite continuous current-positive functions
on an ordinary open \(O\subset X^\an\), where \(X/K\) is smooth
quasi-projective. If \(F:\R^r\to\R\) is convex and nondecreasing in
each coordinate, then \(F(u_1,\ldots,u_r)\) is current-positive.
In particular, finite maxima and sums with nonnegative coefficients
preserve this class.

If \(V\subset Y^\an\) is an ordinary open in a smooth
quasi-projective \(K\)-variety and \(f:V\to O\) is an analytic
morphism, then \(u_i\circ f\) is current-positive on \(V\).
\end{corollary}

\begin{proof}
By \namedref{Theorem}{thm:main}, the \(u_i\) are locally uniformly
approachable by smooth positive potentials. The convex composition
assertion is therefore \cite[Corollary 8.2.4]{CLD25}.
For the pullback assertion, pull back those smooth approximants;
smooth pullback preserves positivity, and locally uniform convergence
is preserved because continuous maps send compact sets to compact sets.
\end{proof}

\subsection{The projective curve criterion}
\begin{corollary}[A projective curve criterion]
\label{cor:projective-criterion}
Let \(\mathscr Y/R\) be a regular integral projective flat model with
smooth generic fibre \(Y\) of positive dimension. Let \(L\) be an
ample line bundle on \(Y\), and let \(r\) be a finite continuous
metric on \(L\). The following conditions are equivalent:
\begin{enumerate}
\item For every algebraic morphism \(f:C\to Y\) from a smooth
\(K\)-curve, the pulled-back metric \(f^*r\) has positive Chern current.
\item The original CLD current \(c_1(L,r)\) is positive.
\item The metric \(r\) is semipositive, equivalently a uniform limit
of pure Fubini--Study metrics on \(L\).
\item The metric \(r\) is a uniform limit of global smooth metrics on
\(L\) with positive curvature forms.
\end{enumerate}
The criterion also holds for an ample rational line bundle by clearing
its tensor denominator.
\end{corollary}

\begin{proof}
Suppose the first condition holds. \namedref{Theorem}{thm:envelope} gives a
finite continuous semipositive envelope \(p=P_L(r)\), which is a
uniform Fubini--Study limit and satisfies
\[
 p\leq r,\qquad
 \int_{Y^\an}(r-p)c_1(L,p)^{\dim Y}=0.
\]
\namedref{Theorem}{thm:contact} applies to the curve-positive upper metric
\(r\) and gives \(r=p\). This proves the third condition.
\namedref{Lemma}{lem:smooth-fs} gives the fourth condition from the third,
and positivity passes to the current limit, giving the second.
Finally, \namedref{Proposition}{prop:curve-restriction}, applied in local
frames, gives the first condition from the second. Positive tensor
powers scale each curvature by the same positive integer, which proves
the rational-line assertion.
\end{proof}

\section{Comparison with adelic mixed measures}\label{sec:adelic}

\begin{definition}[Boundary completion]\label{def:boundary}
Let \(U/k\) be quasi-projective. We use the rational boundary
completion of \cite[Sections 2.4--2.5]{YZ}, retaining its specified
line-bundle identifications. A boundary datum consists of a
projective compactification, an effective Cartier boundary with
its multiplicities, and strictly positive Green data.
On common dominating models a Cauchy presentation satisfies, for
every \(\varepsilon>0\) and all sufficiently large \(j,l\),
\begin{equation}\label{eq:boundary-order}
 -\varepsilon\overline\Delta_\partial
 \leq\overline D_j-\overline D_l
 \leq\varepsilon\overline\Delta_\partial .
\end{equation}
In a fixed local identification at \(v\), this implies
\[
 |w_j-w_l|\leq\varepsilon b_v\quad\text{on }U_K^\an,
\]
where the boundary Green function \(b_v\) is finite continuous
on that open. The completion uses the full order
\eqref{eq:boundary-order}, not merely this scalar inequality.
A strongly nef adelic line bundle has a Cauchy presentation by
nef model objects. An integrable one is a difference of two
strongly nef objects, allowing rational tensor powers and the
specified isometries \cite[Definition 2.5.2]{YZ}.
\end{definition}

\subsection{Identification and positivity of the prescribed measure}
\label{subsec:adelic-proof}
\begin{proof}[Proof of \namedref{Theorem}{thm:adelic}]
Choose integrable presentations with their prescribed isometries,
\[
 \overline L_i^{\otimes r_i}
 \simeq\overline P_i\otimes\overline Q_i^{-1},
 \qquad r_i\in\Z_{>0},
\]
where \(\overline P_i,\overline Q_i\) are strongly nef. Their metrics
at \(v\) are locally psh-approachable, and their mixed measures are
the stable CLD products
\cite[Lemma 3.6.7 and Sections 3.6.7--3.6.8]{YZ}.
Let \(u_i,p_i,q_i\) be weights in compatible local frames. The
specified isometries give
\[
 r_i u_i=p_i-q_i+h_i,\qquad
 h_i=-\log|a_i|,\qquad \ddc h_i=0,
\]
for analytic units \(a_i\). By the definition in
\cite[Section 3.6.8]{YZ}, the prescribed signed measure is
\begin{equation}\label{eq:adelic-expansion}
 \mu_v^{\mathrm{YZ}}
 =\frac1{\prod_i r_i}
   \sum_{\epsilon\in\{0,1\}^n}(-1)^{|\epsilon|}
       \prod_{i=1}^n c_1(\overline R_{i,\epsilon_i})_v,
 \qquad
 \overline R_{i,0}=\overline P_i,\quad
 \overline R_{i,1}=\overline Q_i.
\end{equation}
Its independence of presentation is part of that construction.

\namedref{Theorem}{thm:main} makes each target weight \(u_i\) approachable.
Apply additivity of approachable products to
\(r_i u_i+q_i=p_i+h_i\) in every slot; the zero-curvature terms
\(h_i\) disappear by \eqref{eq:continuous-kernel}.
The alternating expansion is exactly
\begin{equation}\label{eq:adelic-local}
 \mu_v^{\mathrm{YZ}}=\bigwedge_{i=1}^n\ddc u_i.
\end{equation}
The right-hand side is positive by
\namedref{Theorem}{thm:background-products}. Positivity of Radon measures
is local, so this proves the assertion on all of \(U_K^\an\).
\end{proof}

\appendix
\section{A quantitative comparison of mixed measures}
\label{app:comparison}

For completeness, we record a comparison estimate in which the constant
is independent of the metrics being compared. It is separate from the
proof of contact rigidity: the divisor limit there uses only the
continuity of a fixed potential acting on positive closed currents.

\begin{lemma}[Global domination of finitely many forms]
\label{lem:global-calibration}
Let \(Y/K\) be smooth projective, and let
\(\eta_1,\ldots,\eta_s\) be smooth symmetric real \((1,1)\)-forms
on \(Y^\an\). There are an ample line bundle \(H\), a global smooth
metric \(\phi\) on \(H\), and its positive curvature
\(\omega=c_1(H,\phi)\), such that
\[
 -\omega\leq\eta_i\leq\omega\qquad(1\leq i\leq s).
\]
The inequalities are in the original CLD form cone, and all the
constructions take place over \(K\).
\end{lemma}

\begin{proof}
Choose a very ample line bundle \(A\) on \(Y\). We first construct
one global positive curvature that controls the square of a local
moment coordinate. On a compact polynomial norm box \(C\) contained
in an affine chart of the embedding defined by \(A\), let \(f\) be
an analytic unit on a neighbourhood of \(C\). The box is defined by
finitely many upper bounds and positive lower bounds on polynomial
functions, with bounds in \(|K^\times|^\Q\). After clearing powers,
its affinoid algebra is the completion of the affine coordinate
algebra with the lower-bounded polynomials inverted.
Thus \(f\) can be uniformly approximated on \(C\) by a rational
function \(h\) from this algebra. Choose the error smaller than
\(\min_C|f|\). The ultrametric inequality gives
\[
 |h|=|f|\quad\text{throughout }C.
\]
In particular \(h\) is a rational unit there. This equality of
moment maps gives the same differential forms on \(\Int C\),
by \cite[Lemma 3.1.10]{CLD12}.

Homogenizing a numerator and denominator for \(h\) expresses it
as \(h=s_1/s_0\), where \(s_0,s_1\) are sections of some
positive tensor power \(A^d\), both nonzero on \(C\).
Set \(M=\|\log|h|\|_C\), and choose
\(\varepsilon>2M\). In the construction of
\namedref{Proposition}{prop:regularized-max}, choose \(\rho\) strictly
positive on \((-1,1)\), and write
\[
 H_\varepsilon(t)=R_{2,\varepsilon}(t,0).
\]
Then
\[
 H_\varepsilon''(t)
 =\frac1\varepsilon
   (\rho*\check\rho)(t/\varepsilon)>0
       \qquad(|t|<2\varepsilon),
 \qquad\check\rho(t)=\rho(-t).
\]
Hence \(H_\varepsilon''\geq c>0\) on \([-M,M]\).

Take a global generating tuple of \(A^d\), multiply its members
by sufficiently small elements of \(K^\times\), and append them
to \(s_0,s_1\). If \(a\) is any appended section, choose the
coefficient so that
\[
 \log|a/s_0|<-2\varepsilon-1
 \quad\text{on }C
\]
where the value is \(-\infty\) at a zero. Thus its logarithmic
entry lies more than \(2\varepsilon\) below the maximum of
\(0\) and \(\log|h|\) throughout \(C\).
Every appended entry is exactly inactive in the regularized
maximum on a neighbourhood of \(C\). The full tuple generates
\(A^d\) everywhere, and regularizing its maximum gives a global
smooth positive metric by \namedref{Lemma}{lem:smooth-fs}.
In the frame \(s_0\), its weight on \(\Int C\) is precisely
\(H_\varepsilon(\log|h|)\). Its curvature there is
\[
 H_\varepsilon''(\log|h|)
       \,d'\log|h|\wedge d''\log|h|
 \geq c\,d'\log|h|\wedge d''\log|h|.
\]

Now cover \(Y^\an\) by the interiors of finitely many such boxes,
using common moment presentations for the finitely many forms
\(\eta_i\) on each box. Replace the analytic moment units by
rational units as above. The symmetric coefficient matrices of
all the \(\eta_i\) are bounded on these compact boxes.
The construction for each moment unit supplies a global positive
curvature controlling its coordinate square. Suitable positive
integer multiples of these curvatures therefore dominate both
signs of every coefficient matrix on the corresponding box.
Their sum is a global smooth positive curvature, dominates every
\(\eta_i\) on the finite cover, and is the curvature of a positive
tensor power of \(A\). If the list of required coordinate squares
is empty, add any smooth positive approximation of a pure-FS
metric on \(A\). This gives the asserted ample line \(H\) and
metric \(\phi\).
\end{proof}

\begin{proposition}[Comparison on a fixed smooth test]
\label{prop:metric-comparison}
Let \(Y/K\) be smooth projective of dimension \(n\geq1\). For
\(1\leq i\leq n\), let \(p_i,q_i\) be globally smoothly
approximable metrics on the same specified line bundle \(L_i\).
Fix a real smooth function \(\varphi\) on \(Y^\an\).
Choose an ample \(H_\varphi\) and a global smooth positive
curvature \(\omega_\varphi\) such that
\[
 -\omega_\varphi\leq\ddc\varphi\leq\omega_\varphi,
\]
as in \namedref{Lemma}{lem:global-calibration}. Then
\begin{equation}\label{eq:metric-comparison}
 \begin{split}
 &\left|\int_{Y^\an}\varphi\left(
      \bigwedge_{i=1}^n c_1(L_i,p_i)
       -\bigwedge_{i=1}^n c_1(L_i,q_i)\right)\right|\\
 &\hspace{1em}\leq
  \sum_{i=1}^n\|p_i-q_i\|_\infty\,
      \deg_K\!\left(H_\varphi\prod_{h\ne i}L_h\right).
 \end{split}
\end{equation}
The norm is the uniform norm of the scalar difference of two
weights on the same line. The line \(H_\varphi\) is fixed once
\(\varphi\) and its calibration have been chosen; the degree
constants are independent of all the metrics \(p_i,q_i\).
The assertion also holds for rational line bundles after clearing
and dividing tensor denominators.
\end{proposition}

\begin{proof}
Put \(\alpha_i=c_1(L_i,p_i)\), \(\beta_i=c_1(L_i,q_i)\), and
\(T_i=\bigwedge_{h<i}\alpha_h\wedge\bigwedge_{h>i}\beta_h\).
Telescoping and current duality give
\[
 \int\varphi\left(\bigwedge_i\alpha_i-\bigwedge_i\beta_i\right)
 =\sum_i\int(p_i-q_i)\,\ddc\varphi\wedge T_i.
\]
Each \(T_i\) is positive and closed, so
\(\lvert\ddc\varphi\wedge T_i\rvert\leq\omega_\varphi\wedge T_i\).
By \namedref{Lemma}{lem:degree}, the latter measure has mass
\(\deg_K(H_\varphi\prod_{h\ne i}L_h)\). Estimating each summand by
its uniform coefficient norm proves \eqref{eq:metric-comparison}.
Multilinearity gives the rational-line version.
\end{proof}

\section*{Notation}
\renewcommand{\arraystretch}{1.06}
\begin{longtable}{@{}>{\raggedright\arraybackslash}p{0.18\textwidth}>{\raggedright\arraybackslash}p{0.50\textwidth}>{\raggedright\arraybackslash}p{0.24\textwidth}@{}}
\toprule
Symbol & Meaning & Reference\\
\midrule
\endhead
\(K,R,\pi,\lambda\)& Fixed completion, valuation ring, uniformizer,
and \(-\log|\pi|\)&\namedref{Section}{sec:introduction}\\
\(\ddc,\Ac^{p,q}\)& Original CLD differentials and compact smooth
test forms&\namedref{Section}{sec:introduction}\\
\(\Ploc,\Dloc\)& Local smooth positive approximation and its
difference class&\namedref{Definition}{def:approachable}\\
\(M_u(T)\)& The operation \(\ddc(uT)\) for approachable \(u\)&
\namedref{Proposition}{prop:fixed-potential}\\
\(\mathcal B_S\)& Mixed product with a fixed positive closed
background&\eqref{eq:recursive-product}\\
\(Q_C,\Gamma_C\)& Curve Dirichlet energy and its energy measure&
\namedref{Lemma}{lem:curve-green}\\
\(T_a\)& Normalized smooth complete-intersection curve currents&
\namedref{Proposition}{prop:curve-approximation}\\
\(p,r,g\)& Lower approachable metric, upper curve-positive metric,
and \(g=r-p\)&\namedref{Theorem}{thm:contact}\\
\(P_L(r)\)& Semipositive envelope of an obstacle metric&
\namedref{Section}{sec:envelopes}\\
\(V_l,\mathfrak j_m\)& Admissible sections and sums of divisor test ideals&
\namedref{Proposition}{prop:test-ideal-estimate}\\
\(\cE,\Delta_{\cW}\)& Vertical obstacle ideal and canonical
discrepancy boundary&\namedref{Proposition}{prop:test-ideal-estimate}\\
\(\omega_\varphi,H_\varphi\)& Fixed positive calibration and its ample line&
\namedref{Proposition}{prop:metric-comparison}\\
\(b_v,\overline\Delta_\partial\)& Boundary Green function and full
boundary datum&\namedref{Definition}{def:boundary}\\
\(\mu_v^{\mathrm{YZ}}\)& Prescribed signed adelic mixed measure&
\eqref{eq:adelic-expansion}\\
\(R_{h,\varepsilon}\)& Convex regularized maximum&
\namedref{Proposition}{prop:regularized-max}\\
\bottomrule
\end{longtable}

\section*{Declaration on the use of generative AI}
Generative AI was used in developing and drafting mathematical
arguments and in revising the exposition of this manuscript.

\begin{thebibliography}{99}
\bibitem{Ber93}
V.~G.~Berkovich,
\emph{\'Etale cohomology for non-Archimedean analytic spaces},
Publications Math\'ematiques de l'IH\'ES \textbf{78} (1993), 5--161.
\bibitem{Stacks}
The Stacks Project Authors,
\emph{The Stacks Project},
\href{https://stacks.math.columbia.edu/tag/0FUE}{Tag 0FUE},
Lemma 37.17.1.
\bibitem{CLD12}
A.~Chambert-Loir and A.~Ducros,
\emph{Formes diff\'erentielles r\'eelles et courants sur les espaces
de Berkovich}, \href{https://arxiv.org/abs/1204.6277v1}{arXiv:1204.6277v1},
2012.
\bibitem{GJRforms}
W.~Gubler, P.~Jell, and J.~Rabinoff,
\emph{Forms on Berkovich spaces based on harmonic tropicalizations},
\href{https://arxiv.org/abs/2111.05741v3}{arXiv:2111.05741v3}, 2025.
\bibitem{GJRcurves}
W.~Gubler, P.~Jell, and J.~Rabinoff,
\emph{Dolbeault cohomology of graphs and Berkovich curves},
\href{https://arxiv.org/abs/2111.05747v1}{arXiv:2111.05747v1}, 2021.
\bibitem{BMP}
B.~Bhatt, L.~Ma, Z.~Patakfalvi, K.~Schwede, K.~Tucker, J.~Waldron,
and J.~Witaszek, with an appendix by R.~Datta,
\emph{Test ideals in mixed characteristic: a unified theory up to
perturbation},
\href{https://arxiv.org/abs/2401.00615v2}{arXiv:2401.00615v2}.
\bibitem{BE}
S.~Boucksom and D.~Eriksson,
\emph{Spaces of norms, determinant of cohomology and Fekete points
in non-Archimedean geometry},
\href{https://arxiv.org/abs/1805.01016v3}{arXiv:1805.01016v3}, 2021.
\bibitem{BGM}
S.~Boucksom, W.~Gubler, and F.~Martin,
\emph{Non-Archimedean volumes of metrized nef line bundles},
\'Epijournal de G\'eom\'etrie Alg\'ebrique \textbf{5} (2021),
Article 13;
\href{https://arxiv.org/abs/2011.06986v3}{arXiv:2011.06986v3}.
\bibitem{YZ}
X.~Yuan and S.-W.~Zhang,
\emph{Adelic line bundles on quasi-projective varieties},
\href{https://arxiv.org/abs/2105.13587v9}{arXiv:2105.13587v9}.
\bibitem{CLD25}
A.~Chambert-Loir and A.~Ducros,
\emph{Formes diff\'erentielles r\'eelles et courants sur les espaces
de Berkovich},
\href{https://arxiv.org/abs/1204.6277v2}{arXiv:1204.6277v2}, 2025.
\bibitem{Wanner}
V.~Wanner,
\emph{Comparison of two notions of subharmonicity on non-archimedean
curves},
\href{https://arxiv.org/abs/1801.04713v3}{arXiv:1801.04713v3}, 2018.
\bibitem{BGJK}
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},
\href{https://arxiv.org/abs/2102.07392v1}{arXiv:2102.07392v1}, 2021.
\bibitem{FGK}
Y.~Fang, W.~Gubler, and K.~K\"unnemann,
\emph{On the non-archimedean Monge--Amp\`ere equation in mixed
characteristic},
\href{https://arxiv.org/abs/2203.12282v3}{arXiv:2203.12282v3}, 2025.
\bibitem{GRpl}
W.~Gubler and J.~Rabinoff,
\emph{On semipositive piecewise linear functions in non-archimedean
analytic geometry},
\href{https://arxiv.org/abs/2503.05513v1}{arXiv:2503.05513v1}, 2025.
\bibitem{GRpsh}
W.~Gubler and J.~Rabinoff,
\emph{Classically psh and pluriharmonic functions on Berkovich spaces},
\href{https://arxiv.org/abs/2506.13548v2}{arXiv:2506.13548v2}, 2025.
\bibitem{Wu}
L.~Wu,
\emph{On the local theory of non-archimedean psh functions},
\href{https://arxiv.org/abs/2609.27610v1}{arXiv:2609.27610v1}, 2026.
\end{thebibliography}
\end{document}