19 · Model metrics and semipositivity

56 NAG Tags

The construction of a formal metric, including frame independence and invariance under an admissible blow-up, is Proposition 3.11.1. We now study the additional numerical positivity conditions. A residual line bundle is represented on models of a germ reduction as explained in Section 3.10; its degree conditions retain information beyond the real dimension of a chosen skeleton.

For the distinction between formal and algebraic domination of models, and for cofinal models carrying ample extensions, see Section 11.15.1. The separate reduced-fibre hypothesis and its effect on section norms are treated in Sections 11.15.2–11.15.3.

19.1 Model semipositivity and the role of curve tests

A definition using formal models

Let KK be algebraically closed, complete and nontrivially valued, let XX be a strictly KK-analytic space, and let LL be an analytic line bundle. A metric is piecewise linear if there is a covering in the analytic Grothendieck topology on which LL has frames of norm identically one. On a compact reduced strictly analytic space this is equivalent to being induced by a formal model [GK19] (§5.9 and §6.1). The topology in this statement is material: a frame on an analytic domain need not extend to an ordinary open neighborhood.

Definition 19.1.1. (Local model semipositivity). Suppose first that XX is reduced. A piecewise linear metric on LL is semipositive at x∈Xx\in X if there are a compact strictly analytic neighborhood VV of xx and a formal model (V,L)(\mathfrak V,\mathfrak L) inducing the metric such that deg⁡(L∣C)≥0\deg(\mathfrak L|_C)\geq0 for every integral closed curve C⊂VsC\subset\mathfrak V_s that is proper over the residue field. For nonreduced XX, impose the condition on XredX_{\mathrm{red}}. The metric is semipositive on an open set if it is semipositive at each of its points.

This is [GK19] (Definition 6.2). Independence from the inducing model uses common dominating models and the projection formula [GK19] (Lemma 6.3). For a proper algebraic model, this recovers numerical nonnegativity of its line bundle on every special-fibre curve. It is a local non-Archimedean condition; it does not by itself specify an adelic metric at all places.

Theorem 19.1.2. (The algebraic piecewise linear comparison). Retain the field hypotheses above. Let XX be an algebraic variety over KK, let LL be a line bundle, and let W⊂XanW\subset X^{\mathrm{an}} be open. For a piecewise linear metric on L∣WL|W, the following conditions are equivalent:

  1. the metric is semipositive on WW in Definition 19.1.1;

  2. after every algebraic morphism f:X′→Xf:X'\to X, the pulled-back metric is delta-psh on (fan)−1(W)(f^{\mathrm{an}})^{-1}(W);

  3. after every such morphism, the pulled-back metric is current-psh for the original smooth test theory;

  4. its first Chern delta-form is positive on WW;

  5. for every closed integral algebraic curve C⊂XC\subset X, the restriction to W∩CanW\cap C^{\mathrm{an}} is current-psh.

Open proof obligation (RC2). Prove the model equivalence in the stated scope, including curve closure and the detection of each model degree; the text uses GK19 Theorems 6.10, 0.1 and 4.1 and Lemma 6.8.

NAG-900329 · Proof status

Proof retained in the RC2 evidence archive. Prove the model equivalence in the stated scope, including curve closure and the detection of each model degree; the text uses GK19 Theorems 6.10, 0.1 and 4.1 and Lemma 6.8.

Remark 19.1.3 (What the theorem does not identify). The condition d′d′′[−log⁡∥s∥]D≥0d'd''[-\log\|s\|]_D\geq0 on WW alone is not one of the five equivalent conditions as stated in [GK19] (Theorem 6.10). Nor does the theorem concern arbitrary continuous metrics in place of piecewise linear ones. These are changes of hypotheses that require separate results.

Analytic curve tests in the 2025 comparison

The analytic extension in [GRPL] is best stated for functions. Here kk is any complete nontrivially valued field, and XX is a good strictly kk-analytic space.

Definition 19.1.4. (Real piecewise linear function). A function h:X→Rh:X\to\mathbf R is piecewise linear if, locally at each point, finitely many strictly analytic domains VjV_j cover a neighborhood and there are gj∈OX(Vj)×g_j\in\mathcal O_X(V_j)^\times such that h∣Vj=log⁡∣gj∣h|_{V_j}=\log|g_j|. It is real piecewise linear if it is locally a finite real linear combination of such functions.

For a real piecewise linear hh, the metric ∥1∥=e−h\|1\|=e^{-h} is locally induced by a formal model of the trivial line bundle with real coefficients. Semipositivity at a point means that this model can be chosen with nonnegative degree on each proper curve in the special fibre. Equivalently, the residue class Lh(x)∈Pic⁡(red⁡(X,x))⊗ZR\mathcal L_h(x)\in \operatorname{Pic}(\operatorname{red}(X,x))\otimes_{\mathbf Z}\mathbf R is nef. Here red⁡(X,x)\operatorname{red}(X,x) is Temkin’s reduction of the germ; the equivalence with the formal-model definition is part of the comparison recalled in [GRPL] (§5).

Theorem 19.1.5. (Curve tests after analytic extension). Let h:X→Rh:X\to\mathbf R be real piecewise linear. The following conditions are equivalent:

  1. hh is semipositive at every point of XX;

  2. there exists an algebraically closed analytic extension F/kF/k such that, for every smooth analytic curve C/FC/F and every morphism φ:C→XF\varphi:C\to X_F, the function hF∘φh_F\circ\varphi is semipositive on CC.

When these conditions hold, the corresponding curvewise assertion holds over every analytic field extension. On smooth analytic curves, semipositivity of a real piecewise linear function is equivalent to subharmonicity in Thuillier’s sense.

Open proof obligation (RC2). Supply the exact curve and scalar tests and their relationship with model positivity; the text invokes GRPL Proposition 5.1 and Corollary 5.4.

NAG-900330 · Proof status

Theorem 19.1.6. (Closure and local approximation in the real PL class). Let XX be good and strictly kk-analytic.

  1. If hjh_j are semipositive real piecewise linear functions converging pointwise to a real piecewise linear function hh, then hh is semipositive.

  2. If XX is boundaryless and hh is real piecewise linear, then hh is semipositive at xx if and only if on some strictly analytic neighborhood of xx it is a uniform limit of locally Fubini–Study functions.

Open proof obligation (RC2). Prove the stated uniform closure and approximation claims with their precise hypotheses; GRPL Theorems 1.4 and 1.5 are supplied.

NAG-900331 · Proof status

Remark 19.1.7 (Current positivity and the dated status statement). The curvewise condition in Theorem 19.1.5 is not defined by testing a single current on XX. The March 2025 version of [GRPL] (Remark 1.8) explicitly records as unresolved whether a real piecewise linear function that is current-psh for the original CLD test forms must be semipositive, and relates this to pullback stability. This is a status statement about that version. Later claims must be compared with it using their definitions and hypotheses; the analytic curve theorem alone does not settle this additional comparison.

Uniform limits of model metrics

For a line bundle on a proper KK-variety, a semipositive approximable metric in Zhang’s sense is a uniform limit of semipositive rational-formal metrics. Uniform convergence means uniform convergence of the logarithm of the ratio to a fixed continuous reference metric.

Proposition 19.1.8. (Recognizing a formal metric in the closure). Assume the ground field hypotheses of Theorem 19.1.2, and let LL be a line bundle on a proper KK-variety. A formal metric on LL is semipositive if and only if it is a uniform limit of semipositive rational-formal metrics.

Open proof obligation (RC2). Prove the formal metric closure and current comparison in the stated scope; GK19 Proposition 7.2 supplies the comparison.

NAG-900332 · Proof status

The proposition recognizes semipositivity when the limiting metric is already formal. An arbitrary continuous limit need not be induced by one formal model. Similarly, a topology controlling adelic boundary behavior requires additional data beyond the uniform convergence of metrics on a proper analytic space.

19.2 Curvewise positivity and the 2025 comparison results

The curvewise definition of Gubler–Rabinoff in [GRPSH] is relevant to the comparison with both versions of CLD: it tests pullbacks of functions to curves and does not begin by pulling back a current. In this section, “classically psh” always has the meaning of [GRPSH] (Definition 4.1).

The quantifiers in the curvewise definition

Let KK be any complete non-Archimedean field, possibly trivially valued, and let XX be a KK-analytic space. For an analytic extension K′/KK'/K, write πK′/K:XK′=X⊗^KK′⟶X.\pi_{K'/K}:X_{K'}=X\widehat\otimes_K K'\longrightarrow X.

Definition 19.2.1. (Classically plurisubharmonic functions). An upper semicontinuous function u:X→R∪{−∞}u:X\to\mathbb R\cup\{-\infty\} is classically psh if, for every analytic extension K′/KK'/K, every K′K'-analytic curve CC, and every K′K'-analytic morphism f:C→XK′f:C\to X_{K'}, the function u∘πK′/K∘fu\circ\pi_{K'/K}\circ f is subharmonic on CC in the sense of [GRPSH] (Definition 3.4).

Here an analytic curve is separated and of pure dimension one. We spell out the one-dimensional condition, including the convention at −∞-\infty, to make the definition usable without an implicit change of category.

Definition 19.2.2. (Harmonic comparison on curves). First let FF be nontrivially valued and let CC be a smooth strict FF-analytic curve. On a strictly affinoid domain A⊂CA\subset C, a harmonic function is, locally after finite separable base extension, the pullback from a semistable skeleton of a real piecewise affine function whose weighted outgoing slopes sum to zero at every interior vertex. No balancing condition is imposed at boundary vertices.

An upper semicontinuous function v:C→R∪{−∞}v:C\to\mathbb R\cup\{-\infty\} is subharmonic in Thuillier’s sense if it is not identically −∞-\infty on any connected component and if, for every strictly affinoid domain A⊂CA\subset C and every harmonic a:A→Ra:A\to\mathbb R, v≤a on ∂A⟹v≤a on A.v\le a\text{ on }\partial A \quad\Longrightarrow\quad v\le a\text{ on }A. For a possibly singular or non-strict curve CC over an arbitrary complete field FF, Gubler–Rabinoff require that, after every nontrivially valued analytic extension F′/FF'/F and every morphism from a connected smooth strict F′F'-analytic curve into CF′C_{F'}, the pullback is either Thuillier-subharmonic or identically −∞-\infty.

The harmonic comparison and the general-curve extension are [GRPSH] (§§3.1–3.3 and Definition 3.4). Thus on a connected smooth strict curve over a nontrivially valued field, the general definition is Thuillier’s class with the constant function −∞-\infty adjoined. This explicit enlargement is necessary for decreasing limits. The source’s §4.1.3 shows that in Definition 19.2.1 it is enough to test connected smooth strict curves after analytic base extension; combining this with the curve definition permits nontrivially valued extensions throughout. The quantifier over extension fields must not be dropped.

Proposition 19.2.3. (Functoriality and the maximality property). Classically psh functions form a sheaf, and they are stable under analytic pullback and base change. Suppose F\mathcal F is any assignment of upper semicontinuous functions to analytic spaces which is stable under these operations and whose functions on analytic curves are subharmonic. Then every member of F(X)\mathcal F(X) is classically psh.

Open proof obligation (RC2). Prove the common-extension step in the converse direction; GRPSH Proposition 4.2(5).

NAG-900333 · Proof status

Thus the curve-test definition gives a maximal compatible class, not a definition by taking the smallest class generated from smooth functions. This logical distinction matters when comparing it with the generation process in CLD25. The word “smallest” in the introductory description in [GRPSH] (§1.2) should not be used as a replacement for its operative Definition 4.1; the implication in Proposition 19.2.3 follows directly from that definition.

What the PL comparison actually proves

Theorem 19.2.4. (Curvewise positivity and nef models). Assume that the valuation of KK is nontrivial. Let XX be good and strictly KK-analytic, and let h:X→Rh:X\to\mathbb R be an R\mathbb R-PL function. Then the following conditions are equivalent:

  1. hh is semipositive in the local model sense;

  2. the residue class Lh(x)L_h(x) is nef at every x∈Xx\in X;

  3. hh is classically psh.

Open proof obligation (RC2). Prove all equivalences between the model, residual and curvewise classes; the text supplies GJR Proposition 7.1 and GRPSH Theorem 4.3.

NAG-900334 · Proof status

Corollary 19.2.5. (Two harmonicity definitions on the PL class). Under the hypotheses of Theorem 19.2.4, an R\mathbb R-PL function hh is harmonic in the GJR model sense if and only if both hh and −h-h are classically psh.

Proof. Apply Theorem 19.2.4 to hh and to −h-h. The two residue classes are nef exactly when Lh(x)L_h(x) has degree zero on all the curves used to test nefness, which is numerical triviality. ◻

Remark (A PL hypothesis cannot be discarded). Gubler–Rabinoff call a real function pluriharmonic when both it and its negative are classically psh [GRPSH] (Definition 7.1). These functions are continuous. Corollary 19.2.5 identifies them with GJR harmonic functions only after imposing the R\mathbb R-PL condition. The cited paper proves this PL condition for a good quasicompact strict analytic space admitting a strictly semistable formal model [GRPSH] (Proposition 7.8); it does not assert that condition for every analytic space.

The relation with CLD25 psh functions

We now restrict to a good Hausdorff boundaryless strictly analytic space of pure dimension over a nontrivially valued field. We use P25\mathcal P_{25} from Definition 11.2.2 and write Psh⁡cl\operatorname{Psh}_{\mathrm{cl}} for the class of Definition 19.2.1.

Proposition 19.2.7. (A verified one-way comparison). Every CLD25 psh function is classically psh: P25(X)⊆Psh⁡cl(X).\mathcal P_{25}(X) \subseteq \operatorname{Psh}_{\mathrm{cl}}(X). Consequently, a CLD25 psh function which is also R\mathbb R-PL is semipositive in the model sense.

Open proof obligation (RC2). Prove the curvewise comparison in the precise source category; the argument appeals to GRPSH section 1.2.

NAG-900336 · Proof status

This argument uses pullbacks of functions at each stage of the construction. It does not supply a pullback of an arbitrary current. In particular, it cannot start with only the assumption d′d′′[u]≥0d'd''[u]\ge0 against the original CLD test forms unless an additional comparison has already placed uu in the CLD25 class.

Remark (A distinction between forward implications). For an R\mathbb R-PL function on the common boundaryless setting, the source-supported implications include P25 ⟹ Psh⁡cl ⟺ model semipositivity ⟹ d′d′′[u]≥0.\mathcal P_{25} \ \Longrightarrow\ \operatorname{Psh}_{\mathrm{cl}} \ \Longleftrightarrow\ \text{model semipositivity} \ \Longrightarrow\ d'd''[u]\ge0. The middle equivalence is restricted to PL functions; the final implication is [GJR] (Theorem 7.14). The inclusion at the left is not an assertion that the two full sheaves of possibly singular functions are equal. A separate claim in CLD25 about a converse current criterion must be assessed at its own stated hypotheses and proof, rather than inferred from the curvewise theorem.

Proposition 19.2.9. (Pointwise stability in the curvewise class). Let (uj)(u_j) be a net of classically psh functions on a KK-analytic space. If its pointwise limit u:X→R∪{−∞}u:X\to\mathbb R\cup\{-\infty\} is upper semicontinuous, then uu is classically psh.

Open proof obligation (RC2). Prove the curvewise psh limit property and its common-extension step in the exact scope, without using GRPSH Theorem 6.5 as proof.

NAG-900337 · Proof status

An independent identification on the PL class

The following comparison combines the local model approximation in GJR with the curvewise criterion of Gubler–Rabinoff. Its proof does not use the equality of the two continuous generated sheaves claimed in CLD25 Proposition 8.1.11, or the converse current criterion claimed in its Theorem 18.8.4.

Lemma 19.2.10. (Locally Fubini–Study functions belong to the continuous generated class). Let XX be good and strictly analytic over a nontrivially valued complete field KK. If an R\mathbb R-PL function hh is locally Fubini–Study, then h∈C25(X)h\in\mathcal C_{25}(X).

Open proof obligation (RC2). Establish the finite nonnegative combination of maxima of analytic-unit logarithms on an ordinary neighbourhood in the stated scope; GJR section 7.9 supplies this presentation.

NAG-900338 · Proof status

Proposition 19.2.11. (Four equivalent positivity conditions for PL functions). Let KK be complete and nontrivially valued, and let XX be a good Hausdorff boundaryless strictly KK-analytic space of pure dimension nn. For an R\mathbb R-PL function h:X→Rh:X\to\mathbb R, the following conditions are equivalent:

  1. hh is semipositive in the local model sense;

  2. h∈C25(X)h\in\mathcal C_{25}(X);

  3. h∈P25(X)h\in\mathcal P_{25}(X);

  4. hh is classically psh.

Each of these conditions implies positivity of the original current d′d′′[h]d'd''[h].

Open proof obligation (RC2). Prove the fixed-neighbourhood Fubini–Study approximation and the reverse curvewise model implication; GJR Proposition 7.10 and GRPSH Theorem 4.3.

NAG-900339 · Proof status

Remark (What the PL identification leaves separate). This proposition does not identify C25\mathcal C_{25} with C0∩P25C^0\cap\mathcal P_{25} for arbitrary continuous functions. It also does not prove that every semipositive PL function lies in the one-step smooth approximation class U\mathcal U: the local charts used to regularize different LFS approximants may vary. Most importantly, the proof does not supply a converse starting only with d′d′′[h]≥0d'd''[h]\ge0. The four-condition equivalence is obtained through model approximation and curvewise tests, independently of that additional current criterion.

19.3 The PL curvature comparison stated in CLD25

There is a substantive additional claim in the July 2025 text. It must be stated accurately even though the elementary implications in the preceding sections do not prove it.

Remark 19.3.1 (The residual criterion stated in the source). Let XX be a Hausdorff boundaryless analytic space and let LL carry an integral PL metric in the terminology of CLD25. At a point xx, write L~x\widetilde L_x for the residual line bundle on the reduction of the analytic germ. The degree-one case of [CLD25] (Theorem 18.8.4) states c1(L,∥⋅∥)≥0 on some neighbourhood of x⟺L~x is numerically nef.\begin{equation} \begin{gathered} c_1(L,\|\cdot\|)\geq0\text{ on some neighbourhood of }x\\ \Longleftrightarrow\quad\widetilde L_x\text{ is numerically nef}. \end{gathered} \end{equation} The full theorem treats mixed products of integral PL curvature currents and residual intersection classes. It is not a theorem about arbitrary finite continuous metrics.

The residual numerical condition is an intersection-theoretic condition: on a suitable proper premodel of the reduction, the residual line has nonnegative degree on the relevant proper integral curves. It is neither an effective-section condition nor the definition of positivity of a test current. The forward direction in (19.3.1) is consequently a substantial comparison.

The source proof reduces near a germ to residual dimension of definition, computes the curvature along fibres when the relevant dimension equals the number of factors, and uses a negligibility argument for the remaining locus. That proof relies on the earlier construction of residual bundles, tropical calibration, and fibre integration. The elementary counterexamples in Section A.1 do not refute this theorem. Conversely, those earlier constructions are not replaced by the positivity definition or by the scalar estimate in Proposition 11.1.2.

Remark 19.3.2 (The precise boundary of verification). Theorem 19.18.12 reconstructs the current criterion in the pure-dimensional category, including mixed products. It supplies the Laurent realization and noncharging argument while citing the deeper reduction, lifting and fibre-integration theorems. Those inputs have not all been independently re-proved here. Theorem 19.18.15 now proves first-stage approximation for integral PL metrics with semiample residue, and Corollary 19.18.18 gives the generated-psh comparison for strictly integral metrics over a nontrivially valued field. The general integral comparison is now proved by Corollary 19.18.23, using prescribed graded realization. A converse for arbitrary continuous metrics remains separate. The comparison with the earlier status statement in [GRPL] (Remark 1.8) is chronological: that statement dates from March 2025, whereas CLD25 dates from July 2025.

There are local issues which can be identified precisely. The numerical zero-cycle test on p. 539 must use subvarieties of dimension equal to the number of Chern factors. The proof of Corollary 18.8.8 also invokes a strictly integral PL result under a more general integral PL hypothesis; the required reduction is not provided there. Theorem 19.18.22 supplies a graded proof in the integral category without that strictness reduction. These points are described in Remark A.1.7. They do not justify a blanket rejection of the residual criterion.

Remark 19.3.3 (What an extension to continuous metrics would require). Suppose a finite continuous metric satisfies c1(L,∥⋅∥)≥0c_1(L,\|\cdot\|)\geq0. To apply a theorem about PL metrics to it, one would first need an appropriate family of PL approximants whose semipositivity is controlled on a fixed neighbourhood or on a fixed global line. Density of arbitrary model functions among continuous functions gives no such positivity. Likewise, positivity of the weak limit of currents gives no sign information about every approximating current. Thus even accepting (19.3.1) does not produce a converse for all continuous metrics.

19.4 Model semipositivity and global Monge–Ampère theory

Gubler–Martin: a common model language

Gubler–Martin studies semipositive metrics over general complete non-Archimedean fields, removing a recurring division between discretely valued fields and algebraically closed fields. Algebraic, formal, and piecewise linear descriptions are compared in their stated proper or strictly analytic settings. Density of rational PL metrics gives a flexible supply of approximating metrics.

Density alone is not a positivity theorem. Approximating a continuous weight by arbitrary PL functions need not preserve convexity or model nefness. A separate result says that pointwise limits of semipositive model metrics, when the limiting metric is itself a model metric in the theorem’s class, remain semipositive. The endpoint hypothesis matters: it is not a claim that every arbitrary pointwise limit is a finite continuous model metric.

The proof uses model geometry and numerical tests. It is not a corollary of weak closure of the positive-current cone unless a converse comparison between current positivity and model nefness has already been proved. This illustrates a recurrent relation to CLD: model theorems can supply an independent route to positivity and then be compared with analytic currents.

Two objects both called a closed (1,1)(1,1)-form

In model-based variational pluripotential theory, a closed (1,1)(1,1)-form may mean an element of a direct limit Z1,1(X)=lim→⁡XN1(X/K∘)R,\mathcal Z^{1,1}(X)=\varinjlim_{\mathcal X}N^1(\mathcal X/K^\circ)_{\mathbb R}, with transition maps given by pullback to dominating models. A model function has a curvature class in this space. This is an algebraic numerical object, not the original CLD sheaf of smooth (1,1)(1,1)-forms.

Given θ\theta in this model space, a model function ϕ\phi is θ\theta-psh when θ+ddcϕ\theta+dd^c\phi is represented by a nef class. The notation resembles a differential positivity condition because the intended analogy is strong. The definition still requires a comparison theorem before it can be replaced by positivity against all CLD tests.

For metrics in classes where the measure comparison is established, the resulting top-degree Monge–Ampère measure agrees with the analytic or model measure. That common endpoint does not identify all lower-degree objects or all singular potential classes.

Envelopes and the variational method

For a continuous obstacle ff, the θ\theta-psh envelope is, schematically, the upper regularization of the supremum of admitted θ\theta-psh functions below ff. The difficult question is whether the envelope remains continuous and has the required approximation properties. This is where algebraic positivity, multiplier or test ideals, and global generation enter.

The energy functional is built from mixed intersections of a reference metric and the variable metric. In an admissible continuous class, its first variation is integration against the Monge–Ampère measure. Maximizing energy minus the integral against a prescribed measure leads to the equation MA⁡(ϕ)=μ.\operatorname{MA}(\phi)=\mu. Turning this variational principle into a theorem requires compactness or coercivity, envelope regularity, and a differentiability theorem. The formal Euler–Lagrange equation alone is not a solution method.

This global programme concerns a fixed ample line bundle on a proper variety and a normalized measure of the correct total mass. It is different from proving a local potential theorem for every positive closed current on every analytic domain. Both problems are important, and neither formulation automatically implies the other.

Residue characteristic and the scope of the results

The foundational results of Boucksom–Favre–Jonsson assume residue characteristic zero in their initial forms. Gubler–Jell–Künnemann–Martin develops envelope continuity using test ideals in positive characteristic, with the geometric and resolution assumptions stated in that paper. Its surface case uses available resolution results in the relevant dimension.

Fang–Gubler–Künnemann treats mixed characteristic using global test-ideal methods. In the final version used here, Theorem 1.1 gives envelope continuity under the specified regular-model hypothesis, and more generally under resolution of singularities for the projective models. Theorem 1.2 solves the Monge–Ampère equation for positive Radon measures supported on the skeleton of an SNC-model, assuming both the stated resolution and embedded-resolution hypotheses. The solution is a continuous semipositive metric, unique up to scaling.

These assumptions must remain in the theorem. It would be incorrect to cite the result as an unconditional general mixed-characteristic local BT theory. Conversely, the conditional nature of the resolution input does not make the analytic and algebraic methods trivial: the test ideals supply the regularity needed for the variational argument.

References. [GM] [BFJ15] [GJKM] [FGK].

19.5 Maximum principles from analytic curves

The curvewise definition has a useful consequence which does not depend on smooth regularization: maxima can be detected geometrically. Two inputs of Gubler–Rabinoff are responsible. First, a connected boundaryless analytic space is connected by overconvergent analytic curves after suitable analytic field extensions. Second, the boundary of a pure-dimensional affinoid has a dimension-decreasing stratification whose minimal strata are its Shilov points. We state where these geometric results enter and prove their potential-theoretic consequences.

Theorem 19.5.1. (Local maximum principle for the curvewise class). Let uu be classically psh on a KK-analytic space XX. If uu attains a local maximum at x∈Int⁡(X)x\in\operatorname{Int}(X), then uu is constant on a neighbourhood of xx.

Open proof obligation (RC2). Supply the overconvergent curve-connectivity bridge and the curve maximum principle used on the lifted fibres; GRPSH Theorem 5.4 and Proposition 3.5(6).

NAG-900340 · Proof status

The compact-fibre argument is needed: local constancy upstairs at individual points of a fibre does not alone give local constancy on its image. Compactness permits all those neighbourhoods to be used simultaneously. The analytic extension fields are likewise part of the curve definition; they cannot be omitted merely by restricting to curves defined over the original field.

Theorem 19.5.2. (The Shilov maximum principle). If uu is classically psh on an affinoid space XX with Shilov boundary SS, then max⁡Xu=max⁡Su.\max_Xu=\max_Su.

Open proof obligation (RC2). Prove the boundary stratification down to Shilov points; GRPSH Lemma 6.3 is used in addition to the open local maximum principle.

NAG-900341 · Proof status

References. These are [GRPSH] (Theorems 6.1 and 6.4), with the geometric inputs in Theorem 5.4 and Lemma 6.3 separated from their consequences. In the proof of Theorem 6.4 in the consulted v2, the phrase asserting continuity of the psh function should be replaced by upper semicontinuity; that weaker property already makes its maximum set closed. This is a repair of the wording of the proof, not a failure of the maximum theorem.

19.6 Why pointwise limits suffice in the curvewise class

Theorem 19.6.1. Let (uj)(u_j) be a net of classically psh functions converging pointwise to u:X→R∪{−∞}u:X\to\mathbb R\cup\{-\infty\}. If uu is upper semicontinuous, then it is classically psh. No monotonicity assumption is needed.

Open proof obligation (RC2). Supply the native curve comparison maximum principle used after pointwise convergence on the finite boundary. The present route names the now-open Shilov maximum principle as an alternative.

NAG-900342 · Proof status

The finiteness of the testing boundary is the decisive feature. It converts separate pointwise inequalities into one simultaneous eventual inequality. There is no need to dominate a moving family of current coefficients, and no product of currents occurs. This is why the result is a pointwise-closure theorem for a function class, rather than a Bedford–Taylor continuity theorem.

References. [GRPSH] (Theorem 6.5). The distinction from convergence of mixed products is explained in Chapters 9 and 13.

19.7 Finite-dimensional spaces of pluriharmonic functions

Write PH⁡(X)\operatorname{PH}(X) for the real functions hh for which both hh and −h-h are classically psh. Such a function is continuous, since it is both upper and lower semicontinuous. The class is a real vector space by the convex-cone property of classically psh functions.

Theorem 19.7.1. Let a quasicompact analytic space XX be covered by finitely many affinoid domains U1,…,UsU_1,\ldots,U_s, and let SS be the union of their Shilov boundaries. Then evaluation induces an injective linear map PH⁡(X)⟶RS,h⟼(h(s))s∈S,\operatorname{PH}(X)\longrightarrow\mathbb R^S, \qquad h\longmapsto(h(s))_{s\in S}, and in fact ∥h∥C0(X)=max⁡s∈S∣h(s)∣\|h\|_{C^0(X)}=\max_{s\in S}|h(s)|. In particular dim⁡PH⁡(X)≤∣S∣\dim\operatorname{PH}(X)\leq|S|.

Open proof obligation (RC2). The Shilov maximum principle for both signs is open in this projection; prove it or give an independent norm-detection route.

NAG-900343 · Proof status

Corollary 19.7.2. On a quasicompact XX as above, a net in PH⁡(X)\operatorname{PH}(X) which converges pointwise to a finite real function converges uniformly, and its limit is pluriharmonic.

Open proof obligation (RC2). The finite-boundary norm and evaluation theorem used for the finite-dimensional argument remains open.

NAG-900344 · Proof status

The corollary is the finite-dimensional norm argument behind the locally uniform convergence result of Gubler–Rabinoff. On this quasicompact domain, the argument works for nets and itself establishes continuity of the limit. It is a direct consequence of the finite-boundary evaluation theorem, not a new regularization theorem. On a connected compact boundaryless space, the local maximum principle further gives PH⁡(X)=R\operatorname{PH}(X)=\mathbb R.

The finite set SS is a determining set for this particular pluriharmonic space. It does not determine arbitrary continuous functions, arbitrary currents, or every cohomology class of a differential sheaf. Likewise, identifying this kernel with the kernel of an original CLD Hessian on a larger function class requires a comparison theorem. The word pluriharmonic does not erase that requirement.

References. [GRPSH] (Proposition 7.3 and Theorems 7.4, 7.6). The net corollary above is the elementary finite-dimensional consequence of their evaluation map, written with its quasicompact-domain hypothesis.

19.8 Numerical nefness under proper pullback

The model comparison uses two kinds of independence. Numerical nefness is unchanged by a proper surjective pullback, while a model metric is unchanged by replacing the model with a dominating one. We first prove the numerical statement over an ordinary field, then explain how it enters the model calculation.

Proposition 19.8.1. Let f:Y→Zf:Y\to Z be a proper surjective morphism of proper varieties over a field kk, and let MM be a line bundle on ZZ. Then MM is nef if and only if f∗Mf^*M is nef. The statement also holds for real linear combinations of line bundles.

Proof. Let C⊂YC\subset Y be an integral proper curve. If f(C)f(C) is a point, f∗M∣Cf^*M|_C is trivial and has degree zero. Otherwise its image is an integral curve DD, and the cycle projection formula gives deg⁡k(f∗M∣C)=[k(C):k(D)]deg⁡k(M∣D).\deg_k(f^*M|_C)=[k(C):k(D)]\deg_k(M|_D). Thus nefness of MM implies nefness of f∗Mf^*M. Conversely, fix an integral curve D⊂ZD\subset Z. The fibre over its generic point is a nonempty finite-type scheme over k(D)k(D). A closed point of that fibre has residue field finite over k(D)k(D). Its closure in YY, with reduced structure, is an integral proper curve CC mapping onto DD. Applying the displayed equality and dividing by the positive field degree shows deg⁡k(M∣D)≥0\deg_k(M|_D)\geq 0. This proves the converse. All degree expressions are linear in the line-bundle class, giving the real-coefficient version. ◻

The proof includes inseparable morphisms: the field degree is the full algebraic degree. Replacing it by the number of distinct geometric points of a fibre would invalidate the argument in positive characteristic. It also explains why nefness is tested on curves, even when the model has much larger dimension.

19.9 Equal model metrics on a common normal model

For the next two results we restrict to a complete discretely valued field KK, with valuation ring RR, uniformizer π\pi and λ=−log⁡∣π∣\lambda=-\log|\pi|. Let XX be an integral projective KK-variety. All models are normal integral projective flat RR-models with generic fibre identified with XX. The normal and discretely valued hypotheses permit a direct Cartier-divisor calculation; this proof is not a substitute for the general formal-model comparison over arbitrary valuation rings.

Lemma 19.9.1. Let Xs=∑YmYY\mathcal X_s=\sum_Ym_YY, and let D=∑YaYYD=\sum_Ya_YY be a vertical Cartier divisor. Equip the generically trivial bundle O(D)\mathcal O(D) with its model metric and use its canonical rational section sDs_D as the generic trivialization. At the divisorial point ξY\xi_Y, −log⁡∥sD(ξY)∥=aYmYλ.-\log\|s_D(\xi_Y)\|=\frac{a_Y}{m_Y}\lambda. In particular, a vertical model metric with identically zero weight has D=0D=0.

Proof. At the generic point of YY, the normal model has a discrete valuation ring. Its valuation ord⁡Y\operatorname{ord}_Y satisfies ord⁡Y(π)=mY\operatorname{ord}_Y(\pi)=m_Y. The analytic valuation extending the fixed one on KK is therefore vξY=(λ/mY)ord⁡Yv_{\xi_Y}=(\lambda/m_Y)\operatorname{ord}_Y. If a local equation of DD is gg, the canonical section is gg times the local model frame, which has norm one. Hence its weight is vξY(g)=aYλ/mYv_{\xi_Y}(g)=a_Y\lambda/m_Y. If all weights vanish, every Weil coefficient is zero. On a normal integral scheme the map from Cartier divisors to Weil divisors is injective, so D=0D=0. ◻

Proposition 19.9.2. Suppose L\mathcal L and L′\mathcal L' on two such models extend the same line bundle LL with specified generic-fibre identifications. Assume a common normal projective model Y\mathcal Y dominates both. If their induced metrics on LanL^{\mathrm{an}} agree, the pullbacks of L\mathcal L and L′\mathcal L' to Y\mathcal Y are isomorphic compatibly with the generic identifications. Consequently their special-fibre nefness conditions agree.

Proof. On Y\mathcal Y, the quotient of the two pulled-back line bundles is generically trivial. The specified generic trivialization determines a rational section whose Cartier divisor is vertical. The quotient metric has weight zero because the original metrics agree and model metrics pull back along the generic identity. Lemma 19.9.1 forces its divisor to vanish. The rational section is therefore a global nonvanishing regular section and gives the required isomorphism.

Each morphism on special fibres is proper and surjective. Apply Proposition 19.8.1 to their reduced components, or directly to all integral curves as in its proof. Nefness on the source is equivalent to nefness of the original line bundle. The two pulled-back bundles are isomorphic, so the two original nefness conditions agree. For rational model metrics, first raise both to a common positive tensor power. ◻

The existence of common dominating normal models is a geometric input. In this projective DVR setting it is obtained from the closure of the generic graph and normalization; finiteness of normalization uses the usual excellence hypotheses, satisfied by a complete DVR. In the general analytic setting the corresponding work is carried out by admissible formal modifications. The elementary divisor argument above begins after this input, rather than assuming that arbitrary formal models can be compared by a set-theoretic common refinement.

19.10 Why the atomic model measure survives a modification

We retain the normal projective DVR setting and use the ordinary projection formula for algebraic intersection cycles. Let f:Y→Xf:\mathcal Y\to\mathcal X be a proper birational model morphism inducing the identity on XX. Put Ys=∑EmEE\mathcal Y_s=\sum_E m_EE and let L1,…,Ln\mathcal L_1,\ldots,\mathcal L_n be model line bundles on X\mathcal X, where n=dim⁡Xn=\dim X.

Proposition 19.10.1. The atomic sums defining the model measure on X\mathcal X and on Y\mathcal Y with pulled-back line bundles are equal as measures on XanX^{\mathrm{an}}. Exceptional components whose images have dimension less than nn contribute zero.

Proof. If dim⁡f(E)<n\dim f(E)<n, the proper pushforward of the nn-cycle [E][E] is zero. The projection formula therefore gives deg⁡E(f∗L1⋯f∗Ln)=0.\deg_E(f^*\mathcal L_1\cdots f^*\mathcal L_n)=0. If EE dominates a component YY of Xs\mathcal X_s, the local ring at the generic point of EE dominates the discrete valuation ring at the generic point of YY. Both have the same fraction field, because ff is birational. An overring of a discrete valuation ring in its fraction field is either that ring or the field; domination excludes the latter. Thus the two discrete valuation rings are equal. Properness and separatedness give a unique centre of this valuation on Y\mathcal Y, so there is exactly one such component over YY. It has the same residue function field, the same multiplicity of π\pi, and the same normalized analytic valuation. Hence mE=mY,ξE=ξY,deg⁡E(f∗L1⋯f∗Ln)=deg⁡Y(L1⋯Ln).m_E=m_Y,\qquad \xi_E=\xi_Y, \qquad \deg_E(f^*\mathcal L_1\cdots f^*\mathcal L_n) =\deg_Y(\mathcal L_1\cdots\mathcal L_n). For completeness, the existence of that component is the codimension-one part of the proper birational model comparison: extending the generic inverse over the discrete valuation ring gives its centre, whose closure dominates YY. Equivalently, a proper birational morphism to a normal variety is an isomorphism over the generic points of prime divisors. Summing the preceding equalities, with the zero contributions from exceptional components, proves equality of the two atomic measures. ◻

This proof isolates the two numerical mechanisms. Pullback does not create mass on exceptional components contracted to smaller-dimensional sets, and the old divisorial valuations retain their original masses. It does not say that every modification of a metric leaves the measure unchanged: replacing the pulled-back bundle by a twist involving an exceptional divisor changes the metric and can redistribute the masses. The blow-up example in Section 20.6 is of this latter kind.

References. [GK19] (Lemma 6.3 and Theorem 6.10); [GK17] (Theorem 10.5); [CL06]. The proofs here treat the normal projective DVR case; the more general formal-model statements retain the hypotheses of those sources.

19.11 From curves in a special fibre to analytic curve tests

The converse direction of a semipositivity criterion is geometric. A current inequality on an analytic curve can detect a negative model degree only if that curve meets the relevant residual component. The lifting theorem provides this incidence. We explain the subsequent degree argument before treating the pointwise-limit theorem, whose proof uses a different finite-support mechanism.

Throughout this section the ground field is complete and nontrivially valued, and analytic spaces are good and strict. A real PL function hh is represented locally by a formal model L\mathfrak L of the trivial line bundle, with real coefficients. Its semipositivity means nonnegativity of the degrees of Ls\mathfrak L_s on proper integral curves in a suitable local formal model.

Lemma 19.11.1. (The degree detected by a lifted curve). Let YY be a proper integral curve in the special fibre of a formal model W\mathfrak W. Suppose an analytic curve CC in a neighbourhood UU has a formal model C\mathfrak C mapping to W\mathfrak W, and has a proper integral special-fibre component Y′Y' dominating YY. If the restriction of the model function to CC is semipositive, then deg⁡(L∣Y)≥0.\deg(\mathfrak L|_Y)\geq0.

Proof. The nonconstant morphism of proper integral curves Y′→YY'\to Y is finite and surjective. Write its degree as e=[κ(Y′):κ(Y)]>0e=[\kappa(Y'): \kappa(Y)]>0, including inseparable degree. Pullback of the model metric to CC is represented by the restricted pullback of L\mathfrak L. Semipositivity on CC and the model-degree criterion give 0≤deg⁡(π∗L∣Y′)=edeg⁡(L∣Y).0\leq\deg(\pi^*\mathfrak L|_{Y'}) =e\deg(\mathfrak L|_Y). The equality is the projection formula on proper curves. Division by the positive number ee proves the assertion. Real coefficients cause no problem because both sides depend linearly on the line-bundle class. ◻

The construction in [GRPL] (Theorem 4.1 and proof of Theorem 5.3) supplies the situation of this lemma. Starting from the generic point η\eta of YY, it produces an analytic point zz reducing to η\eta and a one-dimensional Zariski-closed subspace through zz in a sufficiently small analytic neighbourhood. After choosing a formal model of the inclusion, the closure of the reduction of zz is the required Y′Y'. There are two dimension and boundary checks in this argument.

First, the formal closure of the analytic curve has a one-dimensional special fibre. A point in that fibre mapping to η\eta cannot have zero-dimensional closure, because its image contains the generic point of a curve. Its closure is therefore a component. Second, the closure of the reduction of an interior analytic point is proper over the residue field. Conversely, properness of YY makes zz an interior point of the original formal generic fibre. Choosing a neighbourhood of zz keeps it interior in the smaller domain, so the corresponding Y′Y' is proper as well. This is the interior-point criterion recalled in [GRPL] (Section 5.2).

Thus the proof needs a local lift with controlled reduction, not a global algebraic curve through an arbitrary analytic point. The affine lifting theorem and the local analytic lifting theorem have different scopes. A Zariski-closed curve in an affinoid domain need not extend to a Zariski-closed subspace of the whole analytic space.

Why normalization and field extension occur

Once all Zariski-closed analytic curves detect the model degrees, one can replace them by smooth curve tests. Over an algebraically closed analytic extension, first pass to the reduction of a curve and then normalize it. The normalization is a rig-smooth curve because the ground field is perfect. The normalization morphism is finite and surjective. Semipositivity descends under this morphism by the same positive-degree argument used above. Boundary points impose no proper residual-curve condition in dimension one; equivalently, one can use smooth boundaryless subdomains to test interior points.

This explains the quantifiers of Theorem 19.1.5. The test maps are allowed after analytic field extension and are not restricted to curves already embedded globally in XX. Descent under field extension is a theorem about the residue line-bundle classes. It is not obtained by defining a pullback of every CLD current.

19.12 Why pointwise convergence suffices for a PL limit

For general distributions, pointwise convergence of functions does not imply convergence of their derivatives as currents. The PL closure theorem avoids this invalid inference. It transfers the derivative to a specially chosen test whose own curvature has finite support, turning the limit into a finite sum of point evaluations.

Lemma 19.12.1. (Finite-support testing). Let CC be a proper analytic curve on which PL functions have finite signed curvature measures DDhDDh satisfying integration by parts. Let U⊂CU\subset C be open and let hj,hh_j,h be PL on UU, with hj→hh_j\to h pointwise. Suppose they admit PL extensions to CC. If a nonnegative PL function ϕ\phi has compact support in UU, then ∫Cϕ DDh~j⟶∫Cϕ DDh~,\int_C\phi\,DD\widetilde h_j \longrightarrow\int_C\phi\,DD\widetilde h, where tildes denote arbitrary such extensions. If DDhjDDh_j is positive on UU, the limit is nonnegative.

Proof. By locality, DDϕDD\phi is supported in the compact support of ϕ\phi, hence in UU. Write DDϕ=∑a=1NcaδxaDD\phi=\sum_{a=1}^N c_a\delta_{x_a}. Integration by parts gives ∫Cϕ DDh~j=∫Ch~j DDϕ=∑a=1Ncahj(xa)⟶∑a=1Ncah(xa)=∫Cϕ DDh~.\int_C\phi\,DD\widetilde h_j =\int_C\widetilde h_j\,DD\phi =\sum_{a=1}^N c_a h_j(x_a) \longrightarrow\sum_{a=1}^N c_a h(x_a) =\int_C\phi\,DD\widetilde h. There is no interchange of an infinite integral with a pointwise limit: the middle expression is a fixed finite sum. Nonnegativity follows from ϕ≥0\phi\geq 0, its support in UU, and positivity of DDhjDDh_j there. The chosen extensions do not affect the integrals. ◻

Proposition 19.12.2. (The curvewise pointwise-limit argument). On a smooth strict analytic curve, a pointwise limit of semipositive real PL functions is semipositive if the limit is again real PL.

Open proof obligation (RC2). Prove the curvewise limiting comparison and nontrivial model/geometric step imported from GRPL’s Theorem 5.6 argument.

NAG-900351 · Proof status

Corollary 19.12.3. On a good strict analytic space, a pointwise limit hh of semipositive real PL functions hjh_j is semipositive whenever hh is real PL.

Proof. Pull the functions back along each smooth curve test after analytic field extension. Pointwise convergence survives this operation and each pulled-back function remains real PL. Proposition 19.12.2 gives semipositivity on that curve. Theorem 19.1.5 then gives semipositivity of hh. ◻

The real PL hypothesis on the limit is used twice: it gives a formal model with finitely many curvature atoms on a curve, and it makes the residue-model criterion applicable. The proof does not assert a general pointwise-closedness theorem for arbitrary continuous current-psh functions. Nor does it establish weak convergence of DDhjDDh_j against every continuous test from pointwise convergence alone. It proves exactly the numerical inequalities needed for a PL endpoint.

References. [GRPL] (Theorems 4.1, 5.3, 5.6 and Corollary 5.4); the curve integration-by-parts formula is obtained there from formal intersection theory and local heights.

19.13 Why the approximation neighbourhood can be fixed

A pointwise perturbation of a nef residue class is not yet a local approximation theorem. If the neighbourhood on which the perturbation is positive shrinks with its coefficient, there may be no common domain on which to take a limit. Gubler–Jell–Rabinoff’s construction resolves this by realizing all perturbations on one formal model and using one reduction tube.

Work over a complete nontrivially valued field on a good strictly analytic space. Let xx be an interior point, and let hh be a semipositive real PL function at xx. The residue-model construction used below is [GJR] (Proposition 7.10): after shrinking once and taking common dominating models, there are a compact analytic neighbourhood VV of xx, a formal model V\mathfrak V, and a projective residue model PP with the following data.

The closure QQ of red⁡(x)\operatorname{red}(x) in Vs\mathfrak V_s maps properly onto PP. A nef real line-bundle class LL on PP represents the residue of hh. Choose an ample line bundle AA on PP. Its residue class is realized by a PL function aa on VV. The model line bundles representing hh and aa restrict to the pullbacks of LL and AA on QQ. These simultaneous equalities, not merely equality of numerical classes at xx, are arranged by the common-model construction.

Proposition 19.13.1. With these data, the single open set W=Int⁡X(V)∩red⁡−1(Q)W=\operatorname{Int}_X(V)\cap\operatorname{red}^{-1}(Q) contains xx, and hϵ=h+ϵa(ϵ∈Q>0)h_\epsilon=h+\epsilon a\quad(\epsilon\in\mathbb Q_{>0}) is locally Fubini–Study on WW for every ϵ>0\epsilon>0. Moreover hϵ→hh_\epsilon\to h uniformly on WW.

Proof. Anticontinuity of reduction makes the tube open in VV; intersecting with Int⁡X(V)\operatorname{Int}_X(V) makes WW open in XX and still containing xx. The numerical theorem that nef plus a positive multiple of ample is ample shows that L+ϵAL+\epsilon A is ample on PP, hence semiample. Its pullback to QQ remains semiample. For every y∈Wy\in W, the closure of red⁡(y)\operatorname{red}(y) lies in QQ; the residue bundle of hϵh_\epsilon is represented there by the restriction of that pullback, hence is semiample. This is exactly the locally Fubini–Study condition at yy.

The function aa is continuous on the fixed compact set VV, so ∥a∥V<∞\|a\|_V<\infty. Therefore ∥hϵ−h∥W≤ϵ∥a∥V→0\|h_\epsilon-h\|_W\leq\epsilon\|a\|_V\to 0. Neither QQ nor WW depends on ϵ\epsilon. ◻

The difficult geometric input is the simultaneous realization on a formal model. Once it is available, the fixed-domain and uniform-error assertions follow from the displayed estimate. This identifies the role of the model construction more precisely than a pointwise appeal to openness of the ample cone.

Which CLD approximation class follows

Proposition 19.13.2. On a good Hausdorff boundaryless strictly analytic space over a complete nontrivially valued field, a semipositive real PL function belongs locally to the continuous generated sheaf C25\mathcal C_{25} of Definition 11.2.2.

Open proof obligation (RC2). Supply the finite maxima presentation used for local Fubini–Study functions; GJR section 7.9.

NAG-900354 · Proof status

This is a two-stage argument. It does not choose one common smooth-approximation neighbourhood for all hϵh_\epsilon. Consequently the conclusion stated here is membership in C25\mathcal C_{25}, not an unproved identification with the one-step class U\mathcal U. This distinction is exactly why the generated-sheaf definition is useful.

Corollary 19.13.3. For a real PL function hh in this setting, the following conditions are equivalent: model semipositivity, membership in C25\mathcal C_{25}, membership in P25\mathcal P_{25}, and classical psh in the sense of Gubler–Rabinoff.

Proof. The preceding proposition gives the first implication. Proposition 11.2.4 gives C25⊆P25\mathcal C_{25}\subseteq\mathcal P_{25}. Proposition 19.2.7 gives classical psh. The PL curvewise comparison, Theorem 19.2.4, returns model semipositivity. Every implication is used on the common stated category. ◻

Single-current positivity is not included among these equivalent conditions. Obtaining it from any of them is already established. Inferring the others from that single-current condition is the separate residual-current comparison discussed earlier. This corollary is a synthesis of the cited theorems, not a resolution of that additional implication.

The complementary pointwise-closure theorem is Corollary 19.12.3. Its proof was given in Section 19.12 by finite-support testing on curves. The two results use different topologies: the fixed-tube construction gives a uniform approximation to a semipositive PL function, whereas pointwise closure starts with semipositive PL functions and requires a PL limit. Together they explain the two directions of the local PL comparison without identifying pointwise convergence with weak convergence of currents.

References. [GJR] (Proposition 7.10); [GRPL] (Theorem 5.6). The generated-sheaf corollary combines these constructions with the explicitly stated comparisons of Chapters 11 and 19.

19.14 The analytic step in a residual positivity criterion

The residual criterion of CLD25 Theorem 18.8.4 compares two different kinds of positivity: degrees of residual intersections and positivity against analytic test forms. Its proof first treats points where the residual construction has dimension at most the number of Chern factors, then removes an exceptional set by a support theorem. We isolate the latter inference so that its hypotheses can be checked independently of the residual geometry.

Lemma 19.14.1. (Positivity across a negligible set). Let TT be a symmetric current on XX whose scalar pairings with smooth tests are signed Radon measures. Let E⊂XE\subset X be closed. Assume that T∣X∖ET|_{X\setminus E} is positive and that ∣μT,η∣(E)=0|\mu_{T,\eta}|(E)=0 for every compactly supported smooth test η\eta of complementary degree. Then TT is positive on XX.

Proof. Fix a positive test η\eta. The scalar measure μT,η\mu_{T,\eta} restricts to a positive measure on X∖EX\setminus E, because testing it against a nonnegative smooth scalar is testing TT against a positive form there. Smooth scalar density extends this positivity to the associated Radon measure. Since its total variation gives zero mass to EE, it is positive on all of XX. In particular T(η)=μT,η(X)≥0T(\eta)=\mu_{T,\eta}(X)\geq 0. This holds for every positive test, proving the claim. ◻

It is not enough to say that EE has empty interior, or that it is small for one chosen tropicalization. The noncharging hypothesis concerns every actual test measure of TT. If TT is not yet known positive, one also cannot apply a theorem about positive currents directly to TT. A finite signed decomposition supplies a legitimate route.

Proposition 19.14.2. Suppose locally T=∑j=1NcjSjT=\sum_{j=1}^N c_j S_j, where cj∈Rc_j\in\mathbb R and each SjS_j is positive and closed of the same degree. If every SjS_j gives zero mass to EE after pairing with any complementary smooth form, then the noncharging hypothesis of Lemma 19.14.1 holds for TT.

Proof. For a fixed test η\eta, linearity gives μT,η=∑jcjμSj,η\mu_{T,\eta}=\sum_jc_j\mu_{S_j,\eta}. Total variations satisfy ∣μT,η∣≤∑j∣cj∣ ∣μSj,η∣.|\mu_{T,\eta}|\leq\sum_j|c_j|\,|\mu_{S_j,\eta}|. The right side vanishes on EE, so the left side does too. A finite partition subordinate to local decompositions proves the assertion on the compact support of any global test. ◻

For a product of Chern currents with local weights hi=ai−bih_i=a_i-b_i in an admitted continuous psh class, such a decomposition is explicit: ∏i=1dd′d′′hi=∑I⊆{1,…,d}(−1)d−∣I∣⋀i∈Id′d′′ai∧⋀i∉Id′d′′bi.\prod_{i=1}^d\mathrm d'\mathrm d''h_i =\sum_{I\subseteq\{1,\ldots,d\}}(-1)^{d-|I|} \bigwedge_{i\in I}\mathrm d'\mathrm d''a_i\wedge \bigwedge_{i\notin I}\mathrm d'\mathrm d''b_i. Every summand without its sign is positive and closed by the continuous-product theorem. Thus a support theorem for positive closed currents can be applied to those summands before positivity of the original signed product has been established. This is the correct order of the argument.

Dimension and the exceptional set

Suppose XX is pure of dimension nn and TT has bidegree (d,d)(d,d), hence bidimension (n−d,n−d)(n-d,n-d). CLD25 Theorem 16.8.5 uses negligible sets in tropical Hausdorff dimension n−dn-d. Its hypotheses include a nontrivial valuation, absence of analytic boundary, closedness and positivity. To use it for the preceding decomposition, verify these conditions for each positive summand and show that the exceptional set has negligible image in every relevant tropical chart.

The geometric part of CLD25 Theorem 18.8.4 produces such a set from strata of a residual model, using Proposition 11.10.4 and Example 16.8.4. The numerical input is tested on integral residual subvarieties WW of dimension dd: deg⁡(c1(M1)⋯c1(Md)∩[W])≥0.\deg\bigl(c_1(M_1)\cdots c_1(M_d)\cap[W]\bigr)\geq0. The dimension is forced: each Chern class lowers cycle dimension by one. A zero-cycle degree therefore requires dim⁡W=d\dim W=d, not an unspecified codimension. The index slip in the introductory explanation on p. 539 of CLD25 is corrected in this form.

Near a point where the residual construction has dimension dd, the product is a scalar multiple λδF\lambda\delta_F of a calibrated fibre current. Positivity is then exactly λ≥0\lambda\geq 0, and the comparison theorem identifies the sign of λ\lambda with the displayed residual degree. When the residual dimension is smaller than dd, the product vanishes for degree reasons. The remaining strata form the exceptional locus. Once its noncharging property has been verified, Lemma 19.14.1 extends positivity to the entire neighbourhood.

This separates the proof into verifiable components: the residual-model realization, the calibrated-fibre formula, the dimension bound for the exceptional locus, and the measure argument above. The measure and signed-decomposition steps have just been proved. Sections 19.18.1–19.18.7 supply the remaining assembly, with named geometric inputs and a precise pure-dimensional conclusion.

Example 19.14.3. (Why the noncharging condition matters). On the real line, the signed measure T=−δ0T=-\delta_0 is zero, hence positive, on R∖{0}\mathbb R\setminus\{0\}. It is not positive on R\mathbb R. The closed exceptional set has empty interior, but ∣T∣({0})=1|T|(\{0\})=1. Thus positivity off a closed set cannot be extended without an actual noncharging assertion.

References. [CLD25] (Theorem 16.8.5, Example 16.8.4 and Theorem 18.8.4). The measure-theoretic transfer is proved above; the residual geometric hypotheses remain explicit.

19.15 Envelope calculus with explicit analytic inputs

The global variational method has an elementary core once its analytic inputs are available. It is useful to prove that core separately: one can then see exactly where envelope regularity, orthogonality, and compactness enter. In this section XX is compact and θ\theta denotes the background class in a fixed global pluripotential theory, not an unspecified original smooth CLD form.

Assume the admitted θ\theta-psh class is stable under addition of constants and convex combinations. For a continuous obstacle ff, assume its envelope P(f)P(f) exists as the largest admitted function below ff. These are assumptions on the chosen theory, including any upper-semicontinuous regularization necessary to make the supremum admitted.

Proposition 19.15.1. The envelope satisfies monotonicity, translation invariance, concavity, and the sup-norm estimate f≤g ⟹ P(f)≤P(g),P(f+c)=P(f)+c,P((1−t)f+tg)≥(1−t)P(f)+tP(g)(0≤t≤1),∥P(f)−P(g)∥∞≤∥f−g∥∞.\begin{gather*} f\leq g\ \Longrightarrow\ P(f)\leq P(g),\qquad P(f+c)=P(f)+c,\\ P((1-t)f+tg)\geq(1-t)P(f)+tP(g)\quad(0\leq t\leq1),\\ \|P(f)-P(g)\|_\infty\leq\|f-g\|_\infty. \end{gather*} If uu is an admitted obstacle, then P(u)=uP(u)=u.

Proof. Every competitor below ff is below gg when f≤gf\leq g, proving monotonicity. Adding cc gives a bijection between the two competitor sets for translation invariance. The convex combination (1−t)P(f)+tP(g)(1-t)P(f)+tP(g) is admitted and lies below (1−t)f+tg(1-t)f+tg, proving concavity. If ε=∥f−g∥∞\varepsilon=\|f-g\|_\infty, then g−ε≤f≤g+εg-\varepsilon\leq f\leq g+\varepsilon; monotonicity and translation invariance give the norm estimate. Finally uu is its own largest competitor. ◻

Assume next that admitted continuous functions have positive Monge–Ampère measures MA⁡(u)\operatorname{MA}(u) of fixed mass VV, an increasing concave energy EE, and the first-variation and endpoint inequalities ∫X(u−v)MA⁡(u)≤E(u)−E(v)≤∫X(u−v)MA⁡(v).\int_X(u-v)\operatorname{MA}(u)\leq E(u)-E(v) \leq\int_X(u-v)\operatorname{MA}(v). The inequalities follow from concavity and the first variation along the segment from vv to uu; they are also proved by the energy calculation in Chapter 20 on its approximation domain. In the model-based theory their validity is a theorem of that theory, with its own domain.

Definition 19.15.2. The envelope has the orthogonality property if ∫X(f−P(f))MA⁡(P(f))=0.\int_X(f-P(f))\operatorname{MA}(P(f))=0. If ff and P(f)P(f) are continuous, this is equivalent to the measure being supported on the contact locus {P(f)=f}\{P(f)=f\}.

Contact-locus interpretation. The integrand is nonnegative. Each compact subset where f−P(f)≥ε>0f-P(f)\geq\varepsilon>0 has measure zero if the integral vanishes. Taking the union over positive reciprocal integers gives zero measure on the noncontact locus. Conversely a measure concentrated on the contact locus makes the integrand zero almost everywhere. Continuity makes that locus closed, giving the support statement. ◻

Theorem 19.15.3. (Differentiation through a continuous envelope). Assume every continuous obstacle has a continuous admitted envelope, orthogonality holds, and MA⁡\operatorname{MA} is weakly continuous under uniform convergence in this class. Then ddt∣t=0E(P(f+th))=∫Xh MA⁡(P(f)),f,h∈C(X).\left.\frac d{dt}\right|_{t=0}E(P(f+th)) =\int_X h\,\operatorname{MA}(P(f)),\qquad f,h\in C(X).

Proof. Put u=P(f)u=P(f) and ut=P(f+th)u_t=P(f+th). The norm estimate gives ∥ut−u∥∞≤∣t∣∥h∥∞\|u_t-u\|_\infty\leq|t|\|h\|_\infty. On the contact locus for uu, one has u=fu=f and ut≤f+thu_t\leq f+th, so ut−u≤thu_t-u\leq th almost everywhere for MA⁡(u)\operatorname{MA}(u). On the contact locus for utu_t, one has ut=f+thu_t=f+th and u≤fu\leq f, so ut−u≥thu_t-u\geq th almost everywhere for MA⁡(ut)\operatorname{MA}(u_t). The energy inequalities yield t∫h MA⁡(ut)≤E(ut)−E(u)≤t∫h MA⁡(u).t\int h\,\operatorname{MA}(u_t)\leq E(u_t)-E(u)\leq t\int h\,\operatorname{MA}(u). For positive tt, divide directly; for negative tt, reverse the inequalities. Uniform convergence of utu_t and weak continuity of MA⁡\operatorname{MA} make the two endpoint integrals tend to the same value. This proves both one-sided derivatives and their equality. ◻

Remark 19.15.4. This short proof deliberately assumes continuity of every envelope and uniform-continuity of the Monge–Ampère operator on the admitted class. It is a synthesis of the energy inequalities and contact-locus identity, not a new proof of those analytic inputs. BFJ’s differentiability theorem is formulated in a broader finite-energy framework and includes work needed when these stronger continuous hypotheses are unavailable. The argument above must not be used to remove their hypotheses or their envelope construction.

Proposition 19.15.5. (Differentiability forces orthogonality). Conversely, assume the derivative formula of Theorem 19.15.3 holds for all continuous directions and that P(f)P(f) is continuous. Then orthogonality follows.

Proof. Set h=P(f)−f≤0h=P(f)-f\leq 0. For 0≤t≤10\leq t\leq 1, the obstacle f+thf+th lies between P(f)P(f) and ff. Hence monotonicity and P(P(f))=P(f)P(P(f))=P(f) give P(f+th)=P(f)P(f+th)=P(f) exactly. The derivative formula gives ∫(P(f)−f)MA⁡(P(f))=0\int(P(f)-f)\operatorname{MA}(P(f))=0. ◻

The Euler–Lagrange step and what it does not prove

Let μ\mu be a positive measure of mass VV, and define Fμ(u)=E(u)−∫u dμF_\mu(u)=E(u)-\int u\,d\mu on the admitted class. Since E(u+c)=E(u)+cVE(u+c)=E(u)+cV, the functional is unchanged by constants. A normalization is required for compactness and uniqueness.

Theorem 19.16.1. Under the continuous-envelope hypotheses above, if an admitted continuous function uu maximizes FμF_\mu, then MA⁡(u)=μ\operatorname{MA}(u)=\mu. Conversely, a solution of this equation maximizes FμF_\mu among the admitted continuous functions.

Proof. For any h∈C(X)h\in C(X) put ut=P(u+th)u_t=P(u+th). Since ut≤u+thu_t\leq u+th and μ≥0\mu\geq 0, maximality gives E(ut)−E(u)≤∫(ut−u) dμ≤t∫h dμ.E(u_t)-E(u)\leq\int(u_t-u)\,d\mu\leq t\int h\,d\mu. Use positive and negative tt and the envelope derivative to deduce ∫h MA⁡(u)=∫h dμ\int h\,\operatorname{MA}(u)=\int h\,d\mu. Equality against every continuous test identifies the measures. Conversely, concavity gives E(v)−E(u)≤∫(v−u)MA⁡(u)=∫(v−u) dμ,E(v)-E(u)\leq\int(v-u)\operatorname{MA}(u)=\int(v-u)\,d\mu, hence Fμ(v)≤Fμ(u)F_\mu(v)\leq F_\mu(u). ◻

The theorem does not produce a maximizer. In BFJ’s method one enlarges the domain to finite-energy potentials to gain compactness, proves the appropriate upper semicontinuity and integrability, obtains a maximizer there, and then proves continuity under the measure hypotheses. None of these steps follows from the displayed Euler–Lagrange equation. Likewise, uniqueness modulo constants uses a global Hodge-index or energy rigidity theorem; concavity by itself need not be strict.

The precise BFJ existence theorem

In the version cited here, BFJ Theorem A assumes a complete discretely valued field of residue characteristic zero, a smooth projective variety which descends to a one-variable function field having this completion, and an ample line bundle. The prescribed positive Radon measure has the degree as its mass and is supported on the dual complex of an SNC model. Under these hypotheses it is the Monge–Ampère measure of a continuous semipositive metric, unique up to a constant factor. Here “algebraizable” is the source’s extra descent condition, not merely the assertion that the analytic space is the analytification of a projective variety over KK. The latter is already assumed. The difference matters when applying the theorem to a variety over a local field. Later envelope theorems may weaken particular inputs in specified settings, but their conclusions must be cited with their own geometric and characteristic hypotheses.

References. [BFJ15] (Proposition 2.15, Theorem 7.2, Remark 7.4, Theorems A and 8.1); [GJKM] [FGK].

19.17 A corrected negligibility statement

For a product of dd curvature factors on a pure nn-dimensional space, the relevant tropical Hausdorff dimension is n−dn-d. It is the dimension of the tests, or equivalently the bidimension of the current. It is not the number dd of curvature factors.

Proposition 19.17.1. (Noncharging for admitted differences). Work in the algebraic scope of Theorem 13.6.3, over a nontrivially valued complete field. Let XX be pure of dimension nn, Hausdorff and boundaryless. Locally let hi=ai−bih_i=a_i-b_i with ai,bi∈UXa_i,b_i\in\mathcal U_X, for 1≤i≤d≤n1\leq i\leq d\leq n. If EE is closed and tropically Hn−d\mathcal H^{n-d}-negligible, then for every actual compactly supported smooth test η\eta of bidegree (n−d,n−d)(n-d,n-d), the following signed measure has total variation zero on EE: d′d′′h1∧⋯∧d′d′′hd∧η.\mathrm d'\mathrm d''h_1\wedge\cdots\wedge\mathrm d'\mathrm d''h_d\wedge\eta.

Proof. Expand the product by multilinearity as in Proposition 19.14.2. Each unsigned term is an ordinary positive closed current of bidimension (n−d,n−d)(n-d,n-d) by Theorem 13.6.3. Apply CLD25 Theorem 16.8.5 with its parameter p=n−dp=n-d to each term. The finite signed sum is noncharging by the total-variation inequality in Proposition 19.14.2. A finite partition on the support of η\eta handles local decompositions. ◻

Example 19.17.2. (The printed exponent in CLD25 Corollary 16.8.6). Take X=Gm,KanX=\mathbf G_{m,K}^{\rm an} over an algebraically closed nontrivially valued field, and put u=max⁡(0,log⁡∣T∣)u=\max(0,\log|T|). This is a finite continuous member of UX\mathcal U_X: regularize the maximum of the two smooth pluriharmonic branches. The standard formal metric on O(1)\mathcal O(1), restricted to this chart, or the slope formula, gives d′d′′[u]=δζ,\mathrm d'\mathrm d''[u]=\delta_\zeta, where ζ\zeta is the unit-radius Gauss point. In the logarithmic coordinate the slopes jump from 00 to 11; all directions off the skeleton contribute zero. The singleton E={ζ}E=\{\zeta\} is tropically H1\mathcal H^1-negligible, since its image under every moment map has at most one point. A nonnegative smooth compactly supported scalar test equal to one near ζ\zeta gives mass one on EE. Thus the printed Hp\mathcal H^p hypothesis in Corollary 16.8.6, with n=p=1n=p=1, is false as written. The corrected exponent is n−pn-p; EE is not H0\mathcal H^0-negligible.

Proposition 19.17.1 proves the corrected assertion in its explicit nontrivial-valuation and admitted-potential scope. For PL metrics, Proposition 19.18.8 below supplies the additional Shilov-section argument over a trivially valued field. Correcting the exponent alone does not enlarge the full potential class.

19.18 The logical structure of the residual criterion

The following formulation separates the analytic inference from its geometric inputs. Let TT be a symmetric current on a neighbourhood UU of xx, with signed Radon scalar pairings. Suppose the proposed residual criterion assigns a number aWa_W to each relevant proper integral residual subvariety WW of dimension dd.

Proposition 19.18.1. (Conditional residual test). Assume the following two properties.

  1. If all aW≥0a_W\geq 0, there is a closed set E⊂UE\subset U such that TT is positive on U∖EU\setminus E and every scalar total variation of TT gives zero mass to EE.

  2. For each WW and every sufficiently small neighbourhood VV of xx, there is a nonempty open subset VW⊂VV_W\subset V on which T=cWaWδFWT=c_Wa_W\delta_{F_W}, with cW>0c_W>0 and δFW\delta_{F_W} a nonzero positive integration current.

Then TT is positive near xx if and only if all aW≥0a_W\geq 0.

Proof. The forward direction follows by restricting a positive current to VWV_W. If aW<0a_W<0, a positive test on which δFW\delta_{F_W} is nonzero gives a negative value. Conversely, use the first assumption and Lemma 19.14.1. ◻

For CLD25 Theorem 18.8.4, aWa_W is deg⁡(c1(M1)⋯c1(Md)∩[W])\deg(c_1(M_1)\cdots c_1(M_d)\cap[W]). The conditional proposition is not a replacement for the geometric assertions in the following list. It identifies exactly what they must supply.

  1. Residual realization and change of premodel: Chapter 17, especially the residual/analytic dimension comparison in Proposition 17.6.4. These remain geometric inputs.

  2. Calibrated fibre identity: Proposition 18.6.1 and paragraph 18.8.3. Their flatness, equidimensionality and native completed-residue-field normalizations must be retained.

  3. Dimension bound: applying Proposition 11.10.4 with threshold d+1d+1 gives dimension at most n−d−1n-d-1 for strata with residual dimension greater than dd. Example 16.8.4 converts this into tropical Hn−d\mathcal H^{n-d}-negligibility.

  4. The complement of the union of positive neighbourhoods need only be contained in that bad locus. Equality with the whole bad locus is unnecessary. Noncharging of every scalar pairing, not smallness alone, completes the argument.

  5. Passage from numerical nefness to a generated psh metric: Proposition 17.8.4 is stated for strictly integral PL metrics. Corollary 18.8.8 is written for integral PL metrics. An extension of the former statement is a separate requirement.

Theorem 19.18.12 below reconstructs the residual current criterion in the Hausdorff boundaryless pure-dimensional category, with the geometric inputs named explicitly. It does not strengthen the potential-class assertion of Theorem 13.6.3. The independent PL equivalence in Corollary 19.13.3 retains its good, strict, nontrivially valued setting and reaches the generated class C25\mathcal C_{25}, not the one-stage class U\mathcal U.

References. CLD25, fixed version of 25 July 2025, pp. 464–465, 497–498, 532–533, 538–541. The following sections distinguish the supplied deductions from the imported geometric theorems.

The coefficient belongs to the base

The scalar in the fibre description can be determined before any positivity argument. This also fixes a notational error in CLD25, paragraph 18.8.3: the bundles MiM_i live on YY, so their residual intersection and the accompanying multiplicity and defect must be evaluated at yy, not at xx. We give the deduction on actual smooth tests.

Proposition 19.18.2. (Fibre coefficient, flat case). Let f:V→Wf:V\to W be a flat morphism of Hausdorff boundaryless analytic spaces, pure of dimensions nn and dd, respectively. Let M1,…,MdM_1,\ldots,M_d be integral PL metrized line bundles on WW. Suppose their top Chern measure is supported on a closed discrete set S⊂WS\subset W, with μ=⋀i=1dc1(Mi)=∑z∈Sazδz.\mu=\bigwedge_{i=1}^d c_1(M_i)=\sum_{z\in S} a_z\delta_z. Assume the relative integration formula of CLD25 Proposition 18.3.5 and that every nonempty fibre over SS is pure of dimension n−dn-d. If Li⊗q≃f∗MiL_i^{\otimes q}\simeq f^*M_i isometrically for a common positive integer qq, then ⋀i=1dc1(Li)=q−d∑z∈SazδVz.\begin{equation} \bigwedge_{i=1}^d c_1(L_i) =q^{-d}\sum_{z\in S}a_z\delta_{V_z}. \end{equation} Here VzV_z is an H(z)\mathscr H(z)-analytic fibre with its native integration current, including its cycle multiplicities; an empty fibre contributes zero.

Proof. Let η\eta be any actual compactly supported smooth form of bidegree (n−d,n−d)(n-d,n-d) on VV. Its support KK has compact image f(K)f(K). Choose a compactly supported smooth scalar cutoff α\alpha on WW, equal to one near f(K)f(K). The support of η\eta is proper over WW, so Proposition 18.3.5 applies to η∧f∗α\eta\wedge f^*\alpha. With uη(z)=∫Vzηu_\eta(z)=\int_{V_z}\eta on fibres of the prescribed dimension, with value zero on the others. Tensor homogeneity and relative integration give qd⟨⋀ic1(Li),η⟩=∫Wαuη dμ=∑z∈S∩f(K)az∫Vzη.q^d\left\langle\bigwedge_i c_1(L_i),\eta\right\rangle =\int_W\alpha u_\eta\,d\mu =\sum_{z\in S\cap f(K)}a_z\int_{V_z}\eta. This sum is finite because SS is closed and discrete and f(K)f(K) is compact. The formula holds for every such η\eta, proving the current identity. No degree factor is introduced by regarding VzV_z over H(z)\mathscr H(z): this is already the normalization in the relative integration theorem. ◻

CLD25 Proposition 18.4.3 supplies the discrete top measure, while Theorem 18.4.7, not Proposition 18.4.3 alone, supplies its residual coefficient. In the notation f:(X,x)→(Y,y)f:(X,x)\to(Y,y), dim⁡Y=d\dim Y=d, shrinking away the other atoms gives the correctly typed formula T=q−dayδf−1(y),ay=mY(y)def⁡k(y)∫PH(y)~/k~∏i=1dc1(M~i,y).\begin{equation} T=q^{-d}a_y\delta_{f^{-1}(y)},\qquad a_y=m_Y(y)\operatorname{def}_k(y) \int_{P_{\widetilde{\mathscr H(y)}/\widetilde k}} \prod_{i=1}^d c_1(\widetilde M_{i,y}). \end{equation} The graded intersection integral, mY(y)m_Y(y) and def⁡k(y)\operatorname{def}_k(y) retain the definitions of that theorem. The positive prefactor affects the mass, although not its sign. A reduction to an ungraded premodel may introduce an additional graded-field degree; it must not be suppressed. Theorem 18.4.7 and the relative integration theorem remain imported inputs, not conclusions of the short proof above.

How residual dimension produces a flat presentation

Write rxr_x for simultaneous residual definition dimension. CLD25 Proposition 17.6.4 is the lifting input: on an almost good germ with trivializable underlying bundles, a graded field of definition F⊂H(x)~F\subset\widetilde{\mathscr H(x)} of transcendence degree rxr_x and with dim⁡FΩF/k~=rx\dim_F\Omega_{F/\widetilde k}=r_x gives a flat boundaryless presentation through a germ of dimension rxr_x. If k~\widetilde k is perfect, this applies without tensor powers. For a finite family of line bundles in general, it applies after a common power qq of the residue characteristic exponent.

Its construction has three distinct inputs. First, choose homogeneous generators with independent differentials and lift them to analytic units. The associated torus moment sends xx to a Gauss point. A finite separable extension of the graded residue field is realized by a tame extension of the completed residue field, hence by an etale morphism of germs. Second, the given embedding into H(x)~\widetilde{\mathscr H(x)} selects a section after base change; the germ-reduction correspondence restricts the target to the required quasi-compact open. The resulting map is boundaryless. Flatness at the Abhyankar target is supplied by the analytic flatness theorem cited there, not by boundarylessness alone. Third, the residual-bundle equivalence of Theorem 17.4.9 lifts the bundles and identifies their pullbacks isometrically. In the inseparable case, Lemma 10.6.14 descends the powered line bundles to the separable graded subfield before this construction.

Thus the power qq repairs inseparability in a field of definition. It is not a strictification of the metric. Nor does a premodel of dimension at most dd imply that the minimal definition dimension is exactly dd: it may be smaller, in which case the dd-fold curvature product vanishes. If the minimum is dd, formula (19.18.2) applies with its full coefficient. These two cases must be kept separate in testing a residual subvariety of dimension dd.

Reduction strata and the exceptional set

Two elementary consequences of the reduction theorems explain the geometry of the final positivity argument. They also specify which reduction topology is being used.

Lemma 19.18.3. (Specialization reaches every neighbourhood). Let VV be an affinoid space with graded reduction map ρ:V→V~\rho:V\to\widetilde V. Let x∈Vx\in V and let ww be a specialization of ρ(x)\rho(x). Then every neighbourhood of xx in VV meets ρ−1(w)\rho^{-1}(w).

Proof. CLD25 Lemma 11.7.2 identifies the closure of ρ−1(w)\rho^{-1}(w) with the union of fibres over the generizations of ww. The point ρ(x)\rho(x) is one such generization. Hence xx lies in that closure, which is precisely the asserted neighbourhood property. For clarity, the nontrivial input in that lemma is the closed-image statement for reduction of an affinoid neighbourhood at an interior point, together with surjectivity of reduction. Anticontinuity alone proves only one inclusion in the closure formula. ◻

Lemma 19.18.4. (The dimension threshold). Let 0≤d≤n0\leq d\leq n and let VV be an affinoid of dimension at most nn. The locus where the closure of the reduction point has dimension greater than dd is weakly analytic of dimension at most n−d−1n-d-1. Its image under every local moment map has zero (n−d)(n-d)-dimensional Hausdorff measure.

Proof. Choose finitely many homogeneous generators of the graded reduction, and lift them to analytic functions gjg_j of spectral radii rjr_j. A reduction point has dimension at least d+1d+1 exactly when some d+1d+1 of these generators have algebraically independent residues there. For the corresponding subtuple II, this says that its moment has value the Gauss point ηrI\eta_{r_I}. The locus is therefore a finite union of fibres gI−1(ηrI)g_I^{-1}(\eta_{r_I}). The analytic dimension inequality bounds their dimension over H(ηrI)\mathscr H(\eta_{r_I}) by n−(d+1)n-(d+1). This is the proof mechanism of Proposition 11.10.4 with threshold d+1d+1, not dd. Example 16.8.4 supplies the stated tropical Hausdorff negligibility. If d≥nd\geq n, the locus is empty. ◻

To assemble the criterion on a reduction neighbourhood, one needs a finite analytic-domain cover whose graded reductions glue to a space SS, together with line bundles Li\mathcal L_i realizing the residual bundles at every point in the interior. Put P={ρ(x)}‾P=\overline{\{\rho(x)\}}. Anticontinuity makes ρ−1(P)\rho^{-1}(P) open. Intersect with the interior of the analytic neighbourhood before applying a boundaryless positivity statement. If all degree-dd residual intersections on PP are nonnegative, the cases of minimal definition dimension at most dd give positive neighbourhoods around the corresponding points. The closed complement of their union is contained in the higher-dimensional reduction locus. Lemma 19.18.4 and a noncharging theorem then remove that complement. Only containment is required; equality with the entire higher-dimensional locus is generally unwarranted.

Conversely, let W⊂PW\subset P be an integral residual subvariety of dimension dd, with generic point ww. The closure property of Lemma 19.18.3, when available on the glued reduction neighbourhood, places a point over ww in every positive neighbourhood of xx. The residual realization at that point identifies the restrictions Li∣W\mathcal L_i|_W with its premodel data. The lower-definition-dimensional case gives zero intersection; the dimension-dd case detects its sign by (19.18.2). This proves the numerical implication once the realization and the premodel intersection comparison are supplied. The Laurent gluing and simultaneous realization are supplied in Propositions 19.18.9 and 19.18.10; Lemma 19.18.11 explains the numerical comparison. Under the boundaryless hypothesis, goodness already follows from CLD25 paragraph 11.4.1. Thus an affinoid neighbourhood is available in this argument.

There is a further propagation of the exponent correction: CLD25 Proposition 18.3.7 repeats the Hp\mathcal H^p hypothesis of Corollary 16.8.6 for a product of pp curvature factors. Example 19.17.2, viewed as the metric ∥1∥=e−u\|1\|=e^{-u} on the trivial line bundle, refutes this printed metric version as well. In the admitted-potential scope, Proposition 19.17.1 supplies the required Hn−p\mathcal H^{n-p} version. Proposition 19.18.8 extends the verified replacement to PL metrics over arbitrary complete base fields, including the trivial valuation.

Why a tensor power does not remove strictness

CLD25 Definition 17.2.3 distinguishes integral PL metrics from strictly integral PL metrics. The latter require a strict analytic space and local orthonormal frames on strict analytic domains. The common power in Proposition 17.6.4 addresses inseparability, whereas the following elementary example rules out a general strictification by tensor powers, even after finite extension.

Example 19.18.5. (An irrational constant norm). Take K=QpK=\mathbf Q_p with ∣p∣=p−1|p|=p^{-1}, X=M(K)X=\mathcal M(K) and the trivial line L=KeL=Ke with ∥e∥=c=p−2\|e\|=c=p^{-\sqrt 2}. This is an integral PL metric because constant norms are allowed. For every positive integer qq, an orthonormal basis of L⊗qL^{\otimes q} would have the form ae⊗qae^{\otimes q} with ∣a∣cq=1,∣a∣=pq2.|a|c^q=1,\qquad |a|=p^{q\sqrt2}. But ∣K×∣=pZ|K^\times|=p^{\mathbf Z}. After a finite extension K′/KK'/K of ramification index ee, the value group is p1eZp^{\frac 1e\mathbf Z}, which still cannot contain pq2p^{q\sqrt 2}. On the one-point space the local orthonormal-frame condition is the same condition. Hence no such tensor power becomes strictly integral, even after finite extension. Nevertheless its weight is constant, hence smooth and psh with zero curvature. This refutes the proposed reduction method, not Corollary 18.8.8 itself.

A constant change of weight preserves both curvature and the psh property, so a metric already known to be a constant rescaling of a strictly integral metric can be treated in that restricted situation. The general integral PL case needs a compatible treatment of the constants on all chart overlaps, or an extension theorem in graded geometry. Enlarging the value group by a transcendental valued extension merely moves the problem: one must then descend psh membership to the original domain. Neither tensor homogeneity nor Proposition 17.6.4 provides that descent. Theorem 19.18.15 below handles arbitrary constants for semiample residues. Theorem 19.18.17 supplies the fixed-neighbourhood argument in the strict, nontrivially valued setting. For general nef integral PL data, Proposition 19.18.20 supplies the required compatibility, and Theorem 19.18.22 proves the generated-class implication.

References. CLD25, fixed arXiv v2: Definition 17.2.3, Remark 17.2.4, Lemma 17.5.1, Proposition 17.6.4; Propositions 18.3.5 and 18.4.3, Theorem 18.4.7, Proposition 18.6.1 and paragraphs 18.8.2–18.8.3; Lemma 11.7.2 and Proposition 11.10.4. The fibre deduction, threshold argument and strictification counterexample are written out above; deep lifting and intersection inputs are attributed explicitly.

PL weights and the noncharging theorem

The general approximation question in Section 11.6 is unnecessary for PL weights. Their local polyhedral description gives an explicit first-stage approximation of each convex summand. We first make this observation precise, then deal with the trivially valued case of the support argument.

Proposition 19.18.6. (PL weights are admitted differences). Let XX be Hausdorff and almost good, and let h1,…,hsh_1,\ldots,h_s be finitely many real-valued PL functions. Every point has one open neighbourhood UU on which hi=ai−bi,ai,bi∈UX(U).h_i=a_i-b_i,\qquad a_i,b_i\in\mathcal U_X(U). The functions ai,bia_i,b_i can be chosen as pullbacks of finite maxima of affine functions under one analytic moment. They have decreasing smooth psh approximants on this same UU, with an explicit uniform error bound.

Proof. We use the local chart and polyhedral decomposition results of CLD25 Proposition 12.4.7, Proposition 6.3.4 and Proposition 8.8.2(b). Almost goodness makes the associated affine tropical space decent. After taking a compact domain neighbourhood and concatenating the finitely many moments, these results write each hih_i as the difference of two convex PL functions on the convex hull of the tropical image. A convex PL function on a finite polytope is a finite maximum of affine functions there. Extend those affine functions to the ambient real vector space. Their maximum is defined on that whole vector space and has the required restriction. Shrinking to the interior of the chosen domain gives one open UU for all the functions.

For v=max⁡1≤j≤Nℓjv=\max_{1\leq j\leq N}\ell_j, put vϵ(t)=ϵlog⁡ ⁣(∑j=1Neℓj(t)/ϵ),ϵ>0.v_\epsilon(t)=\epsilon\log\!\left(\sum_{j=1}^N e^{\ell_j(t)/\epsilon}\right),\qquad \epsilon>0. Then v≤vϵ≤v+ϵlog⁡Nv\leq v_\epsilon\leq v+\epsilon\log N. If pj=eℓj/ϵ/∑keℓk/ϵp_j=e^{\ell_j/\epsilon}/\sum_k e^{\ell_k/\epsilon} and bjb_j is the gradient of ℓj\ell_j, then ∇2vϵ=ϵ−1(∑jpjbjbjt−(∑jpjbj)(∑jpjbj)t)≥0,\nabla^2v_\epsilon =\epsilon^{-1}\left(\sum_jp_j b_jb_j^t -\Bigl(\sum_jp_jb_j\Bigr)\Bigl(\sum_jp_jb_j\Bigr)^t\right) \geq0, ∂vϵ∂ϵ=−∑jpjlog⁡pj≥0.\frac{\partial v_\epsilon}{\partial\epsilon} =-\sum_jp_j\log p_j\geq0. Thus vϵ↓vv_\epsilon\downarrow v as ϵ↓0\epsilon\downarrow 0. Pullback by the fixed moment produces original smooth psh functions on UU, with the same uniform error bound. Apply this construction to both summands of every hih_i. No assertion about later ordinal stages is used. ◻

The chart theorem is an external geometric input; the displayed regularization proves the first-stage assertion directly. In particular, this is stronger information than merely recording that a PL weight belongs to the generated difference class. For arbitrary analytic spaces below, we use the CLD continuous-product construction and its locally uniform continuity in Proposition 8.7.2 on these explicitly approximated functions. This is a named input in that category; it does not enlarge the algebraic scope of Theorem 13.6.3.

Lemma 19.18.7. (Tropical negligibility along a Gauss section). Suppose kk is trivially valued, fix 0<r<10<r<1, and let kr=k((t))k_r=k((t)) with ∣t∣=r|t|=r. Let π:Xr→X\pi:X_r\to X be scalar extension and σ:X→Xr\sigma:X\to X_r its continuous Gauss section. If a closed subset E⊂XE\subset X is tropically Ha\mathcal H^a-negligible, then σ(E)\sigma(E) is closed and tropically Ha\mathcal H^a-negligible in XrX_r. Here XX is Hausdorff and good and a≥0a\geq 0.

Proof. The image of σ\sigma is the fixed-point set of the continuous retraction σπ\sigma\pi, hence is closed in the Hausdorff space XrX_r. The section is a homeomorphism onto its image, so σ(E)\sigma(E) is closed. We explain the finite local description underlying CLD25 Lemma 16.8.7. Work over an affinoid V=M(A)⊂XV=\mathcal M(A)\subset X. On a compact analytic domain W⊂VrW\subset V_r, let ff be an analytic unit. A finite rational refinement of WW, followed by approximation in the rational-domain algebra, expresses its absolute value as ∣ϕ∣/∣ψ∣|\phi|/|\psi|, where ϕ,ψ∈A⊗^kkr\phi,\psi\in A\widehat\otimes_k k_r are nonzero on the relevant compact domain. The approximation can preserve absolute values: choose its error smaller than the positive minimum of the absolute value being approximated.

Approximate ϕ,ψ\phi,\psi in turn by finite Laurent sums with coefficients in AA, still preserving their absolute values there. The Gauss-section formula is ∣∑νaνtν∣(σ(z))=max⁡ν∣aν(z)∣rν.\left|\sum_\nu a_\nu t^\nu\right|(\sigma(z)) =\max_\nu |a_\nu(z)|r^\nu. Cover σ−1(W)\sigma^{-1}(W) by the finitely many compact sets on which specified numerator and denominator terms attain these maxima. On each nonempty piece KK, the selected coefficients are nonzero and their ratio gg is an analytic unit on a neighbourhood of KK in VV. For an integer ee, log⁡∣f∣∘σ=log⁡∣g∣+elog⁡ron K.\log|f|\circ\sigma=\log|g|+e\log r \quad\hbox{on }K. For a finite tuple of units, refine these pieces simultaneously. Its tropicalization on each piece is thus the tropicalization of a moment on XX, followed by translation. Cover the pieces by finitely many compact analytic domains on which those units are defined. The hypothesis on EE and translation invariance of Hausdorff measure now give zero Ha\mathcal H^a measure for the image of σ(E)∩W\sigma(E)\cap W. The assertion is G-local, so these affinoid computations suffice. ◻

Proposition 19.18.8. (Noncharging for PL curvature products). Let kk be any complete non-Archimedean field, including a trivially valued field. Let XX be Hausdorff, boundaryless and pure of dimension nn, and let L1,…,LdL_1,\ldots,L_d have PL metrics, with 1≤d≤n1\leq d\leq n. Put T=⋀ic1(Li)T=\bigwedge_i c_1(L_i), using the CLD continuous-product construction. If E⊂XE\subset X is closed and tropically Hn−d\mathcal H^{n-d}-negligible, then ∣μT,η∣(E)=0|\mu_{T,\eta}|(E)=0 for every actual compactly supported smooth test η\eta of bidegree (n−d,n−d)(n-d,n-d). No positivity of TT is assumed.

Proof. Boundarylessness implies goodness (CLD25 paragraph 11.4.1), so local frames and Proposition 19.18.6 apply. Expand the product of the local weight differences into a finite signed sum of ordinary positive closed products SS. Their existence and locally uniform approximation are supplied by the continuous-product theorem just specified. If the valuation is nontrivial, apply CLD25 Theorem 16.8.5 to each SS, with its bidimension parameter equal to n−dn-d. Proposition 19.14.2 finishes this case.

Suppose the valuation is trivial and use the extension and section of Lemma 19.18.7. Write SrS_r for the product of the pulled-back convex summands. The source inputs CLD25 Proposition 16.8.2 and Corollary 18.3.3(a), restricted here to the explicit smooth approximation sequences, give supp⁡(Sr∧π∗η)⊆σ(X),π∗(Sr∧π∗η)=S∧η.\operatorname{supp}(S_r\wedge\pi^*\eta)\subseteq\sigma(X), \qquad \pi_*(S_r\wedge\pi^*\eta)=S\wedge\eta. These are the Shilov-support and scalar-extension integration inputs. They follow first for smooth products from the canonical tropical calibrations and then pass to the locally uniform limits; no equality of the higher generated classes is needed here. By the lemma, σ(E)\sigma(E) is tropically Hn−d\mathcal H^{n-d}-negligible. The nontrivially valued theorem applied on XrX_r says that Sr∧π∗ηS_r\wedge\pi^*\eta has total variation zero there. Its support lies in σ(X)\sigma(X), and π−1(E)∩σ(X)=σ(E)\pi^{-1}(E)\cap\sigma(X)=\sigma(E). Consequently, ∣μS,η∣(E)≤π∗∣μSr,π∗η∣(E)=0.|\mu_{S,\eta}|(E) \leq \pi_*|\mu_{S_r,\pi^*\eta}|(E)=0. Finite signed summation and a finite partition of unity on the test support finish the proof. Scalar extension uses the native calibration of Corollary 18.3.3(a), with factor one. This is distinct from a finite analytic covering of calibrated degree ee, for which the corresponding factor is ee. ◻

This supplies the corrected PL-metric scope needed in the residual argument. It does not prove noncharging for every function in the full transfinite dpsh class, nor does it construct pullback of an arbitrary current under scalar extension. The argument only pulls back the functions defining the specified products.

Laurent gluing with arbitrary radii

We now fill in the gluing step of the residual argument. The boundaryless hypothesis already guarantees an affinoid neighbourhood of the point. Thus the non-good-germ concern recorded in version 0.8 is not an additional obstruction under the hypotheses of CLD25 Theorem 18.8.4. What needs proof is that the particular Laurent cover is a formal cover in the graded sense.

Proposition 19.18.9. (The reduction of a Laurent cover). Let V=M(A)V=\mathcal M(A) be affinoid, x∈Vx\in V, and f1,…,fm∈A×f_1,\ldots,f_m\in A^\times. Set ri=∣fi(x)∣r_i=|f_i(x)| and, for ϵ∈{±1}m\epsilon\in\{\pm 1\}^m, put Vϵ={z∈V:∣fi(z)∣ϵi≤riϵi for all i},Aϵ=O(Vϵ).V_\epsilon=\{z\in V:|f_i(z)|^{\epsilon_i}\leq r_i^{\epsilon_i}\ \text{for all }i\}, \qquad A_\epsilon=\mathcal O(V_\epsilon). The graded schemes Sϵ=Spec⁡(A~ϵ)S_\epsilon=\operatorname{Spec}(\widetilde A_\epsilon) glue along the reductions of the intersections to a separated graded k~\widetilde k-scheme SS of finite type. Their reduction maps glue to a surjective anticontinuous map ρ:V→S\rho:V\to S, with ρ−1(Sϵ)=Vϵ\rho^{-1}(S_\epsilon)=V_\epsilon. If s∈Ss\in S, then ρ−1(s)‾=⋃t a generization of sρ−1(t).\overline{\rho^{-1}(s)} =\bigcup_{t\text{ a generization of }s}\rho^{-1}(t).

Open proof obligation (RC2). Establish the residual-to-analytic Laurent-domain identification used for the coordinate bounds; TemLocal Proposition 3.1(ii).

NAG-900370 · Proof status

Proposition 19.18.10. (Simultaneous residual realization). Let xx be a point of a Hausdorff boundaryless analytic space XX, and let L1,…,LdL_1,\ldots,L_d have integral PL metrics. One can choose an affinoid neighbourhood VV of xx, a Laurent cover as above and graded line bundles Li\mathcal L_i on its glued reduction SS such that, for every y∈Int⁡X(V)y\in\operatorname{Int}_X(V), Py={ρ(y)}‾red,(Py,L1∣Py,…,Ld∣Py)P_y=\overline{\{\rho(y)\}}_{\mathrm{red}},\qquad (P_y,\mathcal L_1|_{P_y},\ldots,\mathcal L_d|_{P_y}) is a simultaneous proper premodel of the residual bundles on the germ (X,y)(X,y).

Open proof obligation (RC2). Prove the finite homogeneous refinement and lift required for simultaneous realization; TemLocal Lemma 2.4.

NAG-900371 · Proof status

Numerical tests and the residual criterion

A change of premodel should preserve the numerical test, not an unnormalized top intersection number. The following projection-formula argument explains the distinction. We state its ungraded form; CLD25 paragraphs 10.4.6–10.4.11 and 10.7.14 give its graded version by a splitting extension.

Lemma 19.18.11. (Numerical positivity under proper domination). Let g:Z→Yg:Z\to Y be a proper surjective morphism of proper integral varieties over a field, and let M1,…,MdM_1,\ldots,M_d be line bundles on YY. Their Chern product has nonnegative degree on every integral dd-dimensional subvariety of YY if and only if the pulled-back product has that property on ZZ. The same equivalence holds with all degrees required to vanish.

Proof. For an integral dd-dimensional A⊂ZA\subset Z, the projection formula expresses its intersection as that of g∗[A]g_*[A] on YY. This pushforward is zero if the image has smaller dimension; otherwise it is [k(A):k(g(A))][g(A)][k(A):k(g(A))][g(A)], with positive coefficient. This proves one direction. Conversely, given an integral dd-dimensional B⊂YB\subset Y, choose a closed point of the nonempty generic fibre over BB. Its residue field is finite over k(B)k(B). Its reduced closure A⊂g−1(B)A\subset g^{-1}(B) is proper and generically finite over BB, of some degree e>0e>0, and dim⁡A=d\dim A=d. Hence deg⁡ ⁣(∏ic1(g∗Mi)∩[A])=e deg⁡ ⁣(∏ic1(Mi)∩[B]).\deg\!\left(\prod_i c_1(g^*M_i)\cap[A]\right) =e\,\deg\!\left(\prod_i c_1(M_i)\cap[B]\right). Division by the positive integer ee gives the converse, for either nonnegativity or vanishing. ◻

Finite families of residual bundles admit common proper dominating premodels by CLD25 paragraphs 10.7.5–10.7.7. After such domination their pullbacks agree. The lemma, with the graded intersection conventions, explains the independence stated in paragraph 10.7.14. In particular, a presentation through a premodel of dimension m<dm<d has numerically zero dd-fold product. If m=dm=d, numerical positivity is detected by its top degree. Positive field-degree and multiplicity factors may change that degree; they preserve its sign.

Theorem 19.18.12. (Residual criterion in pure dimension). Let XX be Hausdorff, boundaryless and pure of dimension nn over a complete non-Archimedean field. Let L1,…,LdL_1,\ldots,L_d, 1≤d≤n1\leq d\leq n, have integral PL metrics, and put T=⋀ic1(Li)T=\bigwedge_i c_1(L_i). For x∈Xx\in X, the following are equivalent:

  1. TT is positive on an open neighbourhood of xx.

  2. On a proper simultaneous residual premodel at xx, the Chern product has nonnegative degree on every integral subvariety of dimension dd.

Likewise, TT vanishes near xx if and only if all these degrees vanish. The premodel may be the PxP_x of Proposition 19.18.10.

Proof. We retain three geometric inputs with their original hypotheses: residual-to-analytic lifting after a common power (CLD25 Proposition 17.6.4), the residual top-mass formula (Theorem 18.4.7), and flat relative fibre integration (Proposition 18.6.1). Section 19.18.2 describes the first input and Proposition 19.18.2 checks the normalization of the latter two. They are cited theorems, not being re-proved here.

Step 1: points of small definition dimension. At a point whose simultaneous residual definition dimension is at most dd, these inputs give, after a common tensor power qq, a flat presentation through a base of dimension m≤dm\leq d. If m<dm<d, both the curvature product and its numerical class vanish. If m=dm=d, after shrinking the presentation the product equals q−daz δf−1(z).q^{-d}a_z\,\delta_{f^{-1}(z)}. The fibre current is positive and nonzero near the point. Formula (19.18.2) and premodel independence identify the sign of aza_z with the numerical sign. This proves both implications in the small-dimension case.

Step 2: extend positivity across the remaining locus. Use Proposition 19.18.10 and write P=PxP=P_x. The set U=Int⁡X(V)∩ρ−1(P)U=\operatorname{Int}_X(V)\cap\rho^{-1}(P) is an open neighbourhood of xx. If the numerical condition holds on PP and y∈Uy\in U has dim⁡Py≤d\dim P_y\leq d, it holds on Py⊂PP_y\subset P as well. This premodel bounds the residual definition dimension by dd, so Step 1 gives a positive neighbourhood of yy. Let O⊂UO\subset U be the union of all these neighbourhoods and set E=U∖OE=U\setminus O. Then EE is closed in UU and is contained in {y:dim⁡Py>d}\{y:\dim P_y>d\}. On each Laurent chart the latter locus has weakly analytic dimension at most n−d−1n-d-1, by Lemma 19.18.4. Thus EE is tropically Hn−d\mathcal H^{n-d}-negligible. Proposition 19.18.8 gives zero total variation on EE for every scalar test measure of TT. Lemma 19.14.1 proves positivity on UU.

Step 3: recover every numerical test. Suppose TT is positive on a neighbourhood NN of xx contained in Int⁡X(V)\operatorname{Int}_X(V). Let W⊂PW\subset P be integral of dimension dd and let ww be its generic point. Since ww is a specialization of ρ(x)\rho(x), the closure formula of Proposition 19.18.9 gives a point y∈N∩ρ−1(w)y\in N\cap\rho^{-1}(w). Now Py=WP_y=W, and the restrictions Li∣W\mathcal L_i|_W realize the residual bundles at yy. Step 1 and premodel independence show that their degree on WW is nonnegative. This proves the converse.

Finally, replace L1L_1 by its dual to apply the positivity statement to −T-T. Vanishing of all degrees is equivalent to both numerical signs being nonnegative, hence to both TT and −T-T being positive. Positive test forms span the symmetric tests and TT is symmetric, so this is equivalent to T=0T=0 near xx. ◻

This is a source-based reconstruction of the residual current criterion in the stated pure-dimensional category. The Laurent overlaps, simultaneous bundle realization, corrected noncharging exponent and trivial-valuation passage have now been supplied. The conclusion concerns current positivity. It does not identify an arbitrary positive integral PL metric with a member of the generated psh class, and it does not prove smooth regularization for every continuous current-psh metric. The following sections treat semiample integral residues and the strict fixed-neighbourhood argument; Theorem 19.18.22 then supplies the general integral PL bridge by prescribed graded realization. Proposition 19.18.6 expresses a PL weight as a difference of admitted functions; positivity of its difference current alone is not a proof that the difference itself is admitted.

References. [Tem04] (Proposition 3.1(ii), Lemma 2.4, and Section 4); [CLD25] (paragraph 11.4.1, Propositions 6.3.4, 8.8.2, 12.4.7, 16.8.2, Theorem 16.8.5, Lemma 16.8.7, Corollary 18.3.3 and Theorem 18.8.4). The proof uses the named analytic product, reduction, lifting and mass theorems; it is not an independent reproof of those inputs.

Normalized maxima and residual sections

The numerical criterion does not by itself give an approximation of a metric. For that purpose, the residual sections must first be lifted with their norms intact. In the integral PL category these norms can be arbitrary positive real numbers. Keeping them in the formulas both repairs a normalization error in CLD25 and gives a direct proof of the semiample case.

Lemma 19.18.13. (The normalized maximum formula). Let (X,x)(X,x) be a Hausdorff almost-good analytic germ, let LL have an integral PL metric, and let ss be a local nonvanishing section of its underlying line bundle. Put C=∥s(x)∥−1C=\|s(x)\|^{-1}. Let a1,…,aNa_1,\ldots,a_N be analytic units near xx, and set ri=C∣ai(x)∣.\begin{equation} r_i=\frac{C}{|a_i(x)|}. \end{equation} If the homogeneous residual sections ai(x)~s(x)~\widetilde{a_i(x)}\widetilde{s(x)} are global sections generating L~x\widetilde L_x, then on one ordinary open neighbourhood of xx ∥s∥−1=max⁡iri∣ai∣,h=h(x)+max⁡ilog⁡∣ai∣∣ai(x)∣,h=log⁡∥s∥−1.\begin{equation} \|s\|^{-1}=\max_i r_i|a_i|, \qquad h=h(x)+\max_i\log\frac{|a_i|}{|a_i(x)|}, \quad h=\log\|s\|^{-1}. \end{equation} Conversely, this identity on a neighbourhood implies that the stated residual sections generate.

Proof. We use the correspondence between domains containing xx and their reductions as germs, and the residual norm of CLD25 Section 17.4. Let b=∥s∥−1b=\|s\|^{-1}. Define the analytic domains W={b≥ri∣ai∣ for every i},Vi={b=ri∣ai∣},V=W∩⋃iVi.W=\{b\geq r_i|a_i|\text{ for every }i\},\qquad V_i=\{b=r_i|a_i|\},\qquad V=W\cap\bigcup_i V_i. These are analytic domains locally because the metric is integral PL. Every one contains xx, by the choice of rir_i. The set VV is exactly the locus where b=max⁡iri∣ai∣b=\max_i r_i|a_i|. In the reduction of the germ, its defining conditions say v(s(x)~)−1≥v(ai(x)~)for every i,with equality for at least one i.v(\widetilde{s(x)})^{-1}\geq v(\widetilde{a_i(x)}) \quad\hbox{for every }i, \qquad\hbox{with equality for at least one }i. The positive real numbers rir_i normalize the analytic inequalities at xx; they are not required to be absolute values of scalars. Equivalently, the last display is the maximum identity of CLD25 Lemma 10.6.15. That lemma identifies it with generation by the residual sections. Thus generation is equivalent to (V,x)(V,x) having the full reduction of (X,x)(X,x). The germ-reduction criterion then says that VV is a neighbourhood of xx. Restrict to its ordinary interior and take logarithms. This proves both directions. ◻

Remark 19.18.14 (Corrections to the printed formulas). In the fixed CLD25 v2 PDF, page 484, Lemma 17.4.10 prints ri=∣ai(x)∣∥s(x)∥−1r_i=|a_i(x)|\|s(x)\|^{-1} and defines WW using ≤\leq. In the maximum formula these must be replaced by (19.18.3) and ≥\geq, respectively. The last line of that proof must also retain the inverse on the residual norm of ss. The same normalization is printed again in Corollaries 17.4.11 and 17.4.12, page 485. Proposition 17.8.2, page 497, drops the constants altogether when applying Corollary 17.4.12. The valid conclusion is the weighted maximum (19.18.4); this is sufficient for tropical convexity.

These are literal formula corrections, not counterexamples to the intended semiample criterion. To check the first correction, take X=M(Qp)X=\mathcal M(\mathbf Q_p), ∥s∥=1\|s\|=1 and a1=pa_1=p. Its nonzero homogeneous reduction generates the residual line. The printed right side is ∣p∣2|p|^2, whereas the left side is 11. With r1=∣p∣−1r_1=|p|^{-1} the equality holds. More generally, replacing aia_i by ciaic_i a_i, ci∈k×c_i\in k^\times, leaves every normalized ratio ∣ai∣/∣ai(x)∣|a_i|/|a_i(x)| unchanged.

Theorem 19.18.15. (Semiample residues and first-stage approximation). In the setting of Lemma 19.18.13, suppose that L~x⊗q\widetilde L_x^{\otimes q} is globally generated for some positive integer qq. There is one ordinary open neighbourhood UU of xx on which h=log⁡∥s∥−1h=\log\|s\|^{-1} belongs to the first-stage class UX(U)\mathcal U_X(U). More precisely, there are finitely many analytic units aia_i on UU such that h=h(x)+1qmax⁡iℓi,ℓi=log⁡∣ai∣∣ai(x)∣.h=h(x)+\frac1q\max_i\ell_i, \qquad \ell_i=\log\frac{|a_i|}{|a_i(x)|}. The functions hϵ=h(x)+ϵqlog⁡(∑i=1Neℓi/ϵ),ϵ>0,\begin{equation} h_\epsilon=h(x)+\frac{\epsilon}{q} \log\left(\sum_{i=1}^N e^{\ell_i/\epsilon}\right),\qquad\epsilon>0, \end{equation} are original CLD smooth psh functions on that same UU, decrease to hh as ϵ↓0\epsilon\downarrow 0, and satisfy 0≤hϵ−h≤ϵlog⁡Nq.0\leq h_\epsilon-h\leq\frac{\epsilon\log N}{q}. No strictness or nontriviality of the valuation is required.

Proof. Quasicompactness of the germ reduction provides a finite family of homogeneous generating sections of L~x⊗q\widetilde L_x^{\otimes q}. Relative to s(x)⊗q~\widetilde{s(x)^{\otimes q}}, write their nonzero coefficients in H(x)~\widetilde{\mathscr H(x)}. The homogeneous-unit lifting used in CLD25 Corollary 17.4.12 lifts these coefficients to analytic units aia_i near xx. This is a germ statement: after one finite shrinking all units and their inverses are defined on the same neighbourhood. Apply Lemma 19.18.13 to L⊗qL^{\otimes q} and s⊗qs^{\otimes q}.

The finite tuple (ai)(a_i) is an analytic moment. On its real tropical space, each ℓi\ell_i is affine, including its constant term. The log-sum-exp calculation in Proposition 19.18.6 gives a positive semidefinite covariance Hessian and nonnegative ϵ\epsilon-derivative. Its pullback therefore has positive d′d′′d'd'', and is smooth in the original moment-chart sense. The elementary maximum bounds give the stated uniform estimate on all of UU. This constructs one sequence on a fixed domain, rather than merely proving membership in a generated closure. ◻

Changing the frame to s′=fss'=fs subtracts the smooth pluriharmonic function log⁡∣f∣\log|f| from every weight in the construction. Consequently the conclusion is intrinsic to the metric. These are local approximating weights; no global approximation of metrics on an arbitrary nontrivial line bundle is asserted. Arbitrary constants in a finite maximum have thus been handled directly. No valued extension, and hence no descent of psh functions, enters this argument.

A fixed reduction tube for nef approximation

For a nef residue, one adds a small ample class and then applies the semiample result. The delicate point is not the inequality on a single residual variety. It is the existence of one open neighbourhood where all the perturbed metrics have semiample residues. The following formulation separates this geometric input from the elementary limiting argument.

Proposition 19.18.16. (Approximation from compatible premodels). Let VV be a compact analytic domain neighbourhood of xx in a Hausdorff boundaryless space. Let ρ:V→S\rho:V\to S be an anticontinuous reduction map, with SS a graded scheme, and put P={ρ(x)}‾redP=\overline{\{\rho(x)\}}_{\mathrm{red}}. Assume the following data have been constructed on this one domain:

  1. integral PL weights h,ah,a and graded line bundles L,A\mathcal L,\mathcal A on SS, whose restrictions to Py={ρ(y)}‾redP_y=\overline{\{\rho(y)\}}_{\mathrm{red}} are premodels of their residues for every y∈Int⁡X(V)y\in\operatorname{Int}_X(V);

  2. a proper graded variety QQ, line bundles M,AM,A on QQ, and a morphism g:P→Qg:P\to Q with L∣P≃g∗M\mathcal L|_P\simeq g^*M, A∣P≃g∗A\mathcal A|_P\simeq g^*A;

  3. for all positive integers m,nm,n, the line bundle M⊗n⊗A⊗mM^{\otimes n}\otimes A^{\otimes m} is semiample.

Then on the single ordinary open neighbourhood U=Int⁡X(V)∩ρ−1(P)\begin{equation} U=\operatorname{Int}_X(V)\cap\rho^{-1}(P) \end{equation} every h+ϵah+\epsilon a, ϵ∈Q>0\epsilon\in\mathbf Q_{>0}, is locally in the first-stage class and belongs to C25(U)\mathcal C_{25}(U). Moreover h∈C25(U)h\in\mathcal C_{25}(U).

Proof. Anticontinuity makes ρ−1(P)\rho^{-1}(P) open; the interior in (19.18.6) ensures that UU is an ordinary open neighbourhood of xx in the original space. For y∈Uy\in U, closedness of PP implies Py⊂PP_y\subset P. If ϵ=m/n\epsilon=m/n, the residue of the integral PL weight nh+manh+ma is represented by (L⊗n⊗A⊗m)∣Py≃(g∣Py)∗(M⊗n⊗A⊗m)(\mathcal L^{\otimes n}\otimes\mathcal A^{\otimes m})|_{P_y} \simeq (g|_{P_y})^*(M^{\otimes n}\otimes A^{\otimes m}) on PyP_y. Pullback and restriction preserve global generation of a power, hence semiampleness. Theorem 19.18.15, followed by division by nn, applies near every yy. The local neighbourhood may depend on both yy and ϵ\epsilon; the sheaf property nevertheless gives h+ϵa∈C25(U)h+\epsilon a\in\mathcal C_{25}(U) on this same UU.

Since aa is continuous on compact VV, choose a real constant BB with a+B≥0a+B\geq 0 on VV. For j≥1j\geq 1 set hj=h+(a+B)/jh_j=h+(a+B)/j. Constants preserve C25\mathcal C_{25}, so each hjh_j belongs to C25(U)\mathcal C_{25}(U), and hj↓h,sup⁡U∣hj−h∣≤sup⁡V∣a+B∣j.h_j\downarrow h,\qquad \sup_U|h_j-h|\leq\frac{\sup_V|a+B|}{j}. Closure under locally uniform limits, or under this decreasing continuous limit, gives the last assertion. No intersection of a countable family of neighbourhoods is used. Nor have we produced a single smooth sequence on UU for the limiting hh. ◻

For an ordinary projective variety QQ, condition (3) follows from MM nef and AA ample: M⊗n⊗A⊗mM^{\otimes n}\otimes A^{\otimes m} is ample, hence semiample. In the graded formulation we have stated the required semiampleness explicitly. This avoids hiding an algebraic descent or ampleness theorem in the analytic step. The proposition is a conditional criterion: it proves the consequence of compatible premodels, not their existence in every graded situation.

Theorem 19.18.17. (The strict nef case on a fixed neighbourhood). Let kk be complete and nontrivially valued, let XX be good, Hausdorff and strictly kk-analytic, and let x∈Int⁡(X)x\in\operatorname{Int}(X). If an R\mathbf R-PL function hh in the GJR sense is semipositive at xx, then there is an ordinary open neighbourhood UU of xx on which h∈C25(U)h\in\mathcal C_{25}(U). One can choose bounded a∈PLa\in\mathrm{PL} on a compact neighbourhood and B∈RB\in\mathbf R so that h+(a+B)/jh+(a+B)/j decreases uniformly to hh on UU, and each term is locally Fubini–Study up to an additive constant.

Open proof obligation (RC2). Supply the fixed-local approximation in the strict nef setting; GJR Proposition 7.10.

NAG-900377 · Proof status

This is a source-based expansion of GJR’s proof, not a new approximation theorem. It supplies the fixed-neighbourhood justification missing from a bare germwise reading of CLD25 Proposition 17.8.4 in the common strict, nontrivially valued setting. The smooth regularization neighbourhoods for the different perturbations need not agree: only the domain on which their generated-class membership is asserted must agree. The conclusion is deliberately C25\mathcal C_{25}; no diagonalization to UX\mathcal U_X has been established.

The strict comparison and the graded compatibility problem

Corollary 19.18.18. (Current positivity for strictly integral PL metrics). Let XX be Hausdorff, boundaryless, strictly analytic and pure of dimension n≥1n\geq 1 over a complete nontrivially valued field. Let LL have a strictly integral PL metric. At a point xx, the following conditions are equivalent:

  1. c1(L)c_1(L) is a positive current on a neighbourhood of xx, against the original CLD tests;

  2. the residual line bundle L~x\widetilde L_x is nef;

  3. a local weight of LL belongs to C25\mathcal C_{25} on a neighbourhood of xx.

The same conclusion holds after a constant rescaling of such a metric.

Proof. Boundarylessness implies goodness. With d=1d=1, Theorem 19.18.12 identifies (1) and (2). In this strict category the residual line of a strictly integral metric agrees with the residue of its GJR PL local weight, through the formal-model description; this is the comparison of CLD25 paragraph 17.7.10 and GJR Proposition 7.1. Thus Theorem 19.18.17 gives (2)⇒\Rightarrow(3). Finally, C25\mathcal C_{25} is generated locally from smooth psh functions by continuous locally uniform limits. Positivity of d′d′′d'd'' is local and passes through these limits: for each compactly supported smooth test, uniform convergence of the weights gives convergence of their distributional derivatives. This proves (3)⇒\Rightarrow(1). An additive constant in the weight preserves both curvature and generated-class membership, proving the last assertion. ◻

For an integral PL metric with semiample residue, Theorem 19.18.15 already works over every complete field. The strict nef proof uses the prescribed-model compatibility of GJR Lemma 6.14. For the general integral case, one must supply a graded analogue with the two bundle identifications required in Proposition 19.18.16; simultaneous realization of given metrics alone does not assert this additional domination. Proposition 19.18.20 below constructs precisely that analogue.

One could instead enlarge the valued field to accommodate the radii. That route must prove descent of generated psh membership to the original open set. Descent of numerical nefness, or the scalar-extension formula for Chern currents, only returns numerical information or current positivity. Using the desired converse to descend would be circular. Similarly, a continuous Gauss section is not an analytic morphism and does not automatically pull arbitrary smooth psh functions back to original smooth psh functions. The Gauss-section noncharging argument of Lemma 19.18.7 is a different statement.

Setting Conclusion established here
Integral PL; semiample residue First-stage smooth approximation on one neighbourhood; arbitrary complete base field.
GJR R\mathbf R-PL; nef residue; strict, nontrivial valuation Fixed-tube locally Fubini–Study approximation and C25\mathcal C_{25} membership.
Strictly integral PL; pure dimension; boundaryless Original-current positivity, residual nefness and local C25\mathcal C_{25} membership are equivalent over a nontrivially valued field.
General integral PL; nef residue The prescribed graded construction below proves local C25\mathcal C_{25} membership over every complete field; see Theorem 19.18.22.
General continuous current-positive weight No approximation or generated-class converse follows from the PL results alone.

References. CLD25 v2, Lemma 10.6.15, Theorem 17.4.9, Lemma 17.4.10, Corollaries 17.4.11–17.4.12, and Propositions 17.8.2–17.8.4; GJR arXiv v3, Corollary 5.9, Lemmas 6.11 and 6.14, Section 7.9 and Proposition 7.10. The normalization corrections were checked against the printed PDF. The fixed-tube proof is an expansion of GJR’s argument with its model-theoretic inputs retained.

Realizing a prescribed graded premodel

The conditional criterion in Proposition 19.18.16 requires more than separate realizations of two residual bundles. Their analytic realizations must be compared with the same prescribed projective premodel. We now construct that comparison. The construction takes place over the original valued field; arbitrary radii occur as homogeneous degrees in the reduction.

Write F=H(x)~F=\widetilde{\mathscr H(x)} and R=PF/k~R=P_{F/\widetilde k}. A proper premodel QQ comes with an embedding of its graded function field into FF and a centre morphism πQ:R→Q\pi_Q:R\to Q. It need not have function field equal to FF. This distinction matters when the residue is defined over a smaller graded field.

Lemma 19.18.19. (A finite cover adapted to a premodel). Let XX be Hausdorff and boundaryless over a complete non-Archimedean field kk, and let x∈Xx\in X. Let QQ be a proper graded premodel of F/k~F/\widetilde k, with a finite affine cover Qi=Spec⁡BiQ_i=\operatorname{Spec}B_i. Choose homogeneous nonzero generators aija_{ij} of the finitely generated graded k~\widetilde k-algebra Bi⊂FB_i\subset F. There exist an affinoid neighbourhood VV of xx, analytic units fij,g1,…,gtf_{ij},g_1,\ldots,g_t on VV, and an assignment ϵ↦i(ϵ)\epsilon\mapsto i(\epsilon), for ϵ∈{±1}t\epsilon\in\{\pm 1\}^t, with the following properties:

  1. fij(x)~=aij\widetilde{f_{ij}(x)}=a_{ij};

  2. the Laurent domains Vϵ={z∈V:∣gl(z)∣ϵl≤∣gl(x)∣ϵl for all l}V_\epsilon=\{z\in V:|g_l(z)|^{\epsilon_l}\leq |g_l(x)|^{\epsilon_l}\text{ for all }l\} cover VV, contain xx, and satisfy ∣fi(ϵ)j(z)∣≤∣fi(ϵ)j(x)∣(z∈Vϵ, all j);\begin{equation} |f_{i(\epsilon)j}(z)|\leq|f_{i(\epsilon)j}(x)| \quad(z\in V_\epsilon,\ \text{all }j); \end{equation}

  3. the reduction of the germ (Vϵ,x)(V_\epsilon,x) is contained in πQ−1(Qi(ϵ))\pi_Q^{-1}(Q_{i(\epsilon)}).

Open proof obligation (RC2). Construct the Laurent refinement and the simultaneous bounds in the residual setting; TemLocal Lemma 2.4 and Theorem 4.5 Step 1.

NAG-900379 · Proof status

Historical proof-local display withheld.\begin{equation} \text{Historical proof-local display withheld.} \end{equation}

The finite refinement is essential. Merely choosing lifts of coordinates on each QiQ_i does not show that their boundedness domains cover an analytic neighbourhood. The reflection of germ inclusions supplies exactly that missing implication.

Proposition 19.18.20. (Prescribed simultaneous realization).

Under the hypotheses of Lemma 19.18.19, let L1,…,LsL_1,\ldots,L_s be integral PL metrized line bundles near xx. Let M1,…,MsM_1,\ldots,M_s be graded line bundles on QQ, and fix residual isomorphisms πQ∗Ml≃L~l,x(1≤l≤s).\pi_Q^*M_l\simeq\widetilde L_{l,x}\qquad(1\leq l\leq s). One can choose VV and its Laurent cover so that the separated glued graded reduction ρ:V→S\rho:V\to S carries line bundles Ll\mathcal L_l realizing all the LlL_l. With P={ρ(x)}‾red,P=\overline{\{\rho(x)\}}_{\mathrm{red}}, there is a proper surjective morphism of premodels g:P→Qg:P\to Q and compatible isomorphisms Ll∣P≃g∗Ml.\begin{equation} \mathcal L_l|_P\simeq g^*M_l. \end{equation} For every y∈Int⁡X(V)y\in\operatorname{Int}_X(V), the closure Py={ρ(y)}‾redP_y=\overline{\{\rho(y)\}}_{\mathrm{red}} is a proper premodel of the reduction at yy, and Ll∣Py\mathcal L_l|_{P_y} represents L~l,y\widetilde L_{l,y}. All these data are chosen for the finite family before any tensor powers or positive rational perturbations are selected.

Proof. Choose the affine cover of QQ so that every MlM_l has a homogeneous frame tlit_{li} on every QiQ_i. Apply Lemma 19.18.19. We first construct the map of premodels and then arrange the line bundles on the same cover.

Put Aϵ=O(Vϵ)A_\epsilon=\mathcal O(V_\epsilon) and Sϵ=Spec⁡A~ϵS_\epsilon=\operatorname{Spec}\widetilde A_\epsilon. Proposition 19.18.9 glues these charts to SS. All contain the point q=ρ(x)q=\rho(x). The coordinate ring of P∩SϵP\cap S_\epsilon is Cϵ=im⁡ ⁣(A~ϵ⟶F).\begin{equation} C_\epsilon=\operatorname{im}\!\left(\widetilde A_\epsilon \longrightarrow F\right). \end{equation} For i=i(ϵ)i=i(\epsilon), each fijf_{ij} is bounded on VϵV_\epsilon by its absolute value at xx. Consequently its homogeneous reduction is defined in degree ∣fij(x)∣|f_{ij}(x)| and maps to aija_{ij} in FF. Thus Bi⊂CϵB_i\subset C_\epsilon. This inclusion gives a morphism P∩Sϵ→QiP\cap S_\epsilon\to Q_i. The morphisms agree on overlaps: they induce the same embedding of graded function fields into FF, their source is integral, and QQ is separated. They therefore glue to a dominant g:P→Qg:P\to Q. The premodel PP is proper by Proposition 19.18.10, or directly by the full-germ valuative criterion used in its proof. As QQ is separated, gg is proper. Dominance and properness make it surjective.

Notice that we have constructed a map from PP, not necessarily from all of SS. Polynomial relations among the aija_{ij} need only hold in the image ring (19.18.10). They need not hold among the chosen analytic lifts fijf_{ij}. This is why coordinate lifting suffices.

We now treat the metrics. Trivialize each underlying LlL_l near xx by a frame ele_l. Under the prescribed residual isomorphism, the generic value of tlit_{li} is of the form βliel(x)~\beta_{li}\widetilde{e_l(x)}, with βli∈F\beta_{li}\in F homogeneous and nonzero. Lift these finitely many βli\beta_{li} to analytic units blib_{li} and put eli=bliele_{li}=b_{li}e_l. For i=i(ϵ)i=i(\epsilon), the residual frame of elie_{li} generates L~l,x\widetilde L_{l,x} on the reduction of (Vϵ,x)(V_\epsilon,x). CLD25 Corollary 17.4.5, in rank one, implies that elie_{li} has constant norm cli=∥eli(x)∥c_{li}=\|e_{li}(x)\| on this domain after shrinking at xx. Apply it to all finitely many pairs (l,ϵ)(l,\epsilon) and shrink VV once more. The coordinate bounds and the Laurent refinement remain valid under this shrinking.

Reduce the normed free modules generated by these frames. On an overlap, the ratio of two frames is an analytic unit of constant absolute value, so its reduction is an invertible homogeneous transition function. These transitions glue to Ll\mathcal L_l on SS, exactly as in Proposition 19.18.10. Their degrees retain the constants clic_{li}; no normalization into ∣k×∣|k^\times| is made. On P∩SϵP\cap S_\epsilon, identify the reduced frame of elie_{li} with g∗tlig^*t_{li}. On overlaps, the transition ratios agree in FF by the prescribed residual isomorphism. Both are regular homogeneous units on an integral chart overlap, hence agree there. This proves (19.18.9), including compatibility with the original residual isomorphisms.

All the objects S,P,gS,P,g in the conclusion are understood to be reconstructed from the final shrunken VV by the displayed image-ring construction; the retained coordinate bounds justify the same maps. Finally, the assertion at every interior yy is the germ-reduction calculation in Proposition 19.18.10, applied to these same constant-norm frames. It identifies both the premodel and the residual bundles. Since only a finite cover, finitely many coordinates and finitely many frames were used, the construction does not depend on any later perturbation parameter. ◻

Nef perturbations in the graded category

The remaining algebraic input is semiampleness of positive rational perturbations. It can be proved after a graded splitting extension and then descended. This algebraic extension is distinct from extending the valued field of the analytic space.

Lemma 19.18.21. (Graded nef-plus-ample semiampleness). Let QQ be a proper graded variety over a graded field KK. Suppose MM is nef and AA is very ample. Then M⊗n⊗A⊗mM^{\otimes n}\otimes A^{\otimes m} is semiample for every pair of positive integers m,nm,n.

Proof. Choose a finite polyradius rr such that the graded field extension K′=K(T/r)K'=K(T/r) splits Q,M,AQ,M,A and the degrees of a projective embedding defined by AA; this is CLD25 paragraph 10.3.4. Taking degree-one parts gives an ordinary projective variety Q′1Q'^1 over K′1K'^1, an ordinary nef line bundle M′1M'^1, and a very ample line bundle A′1A'^1. Here nefness uses the graded base-change theorem of paragraph 10.4.11, while the very ample assertion follows from paragraphs 10.3.2–10.3.3. These are the named algebraic inputs to this reduction.

The classical nef-plus-ample theorem makes (M′1)⊗n⊗(A′1)⊗m(M'^1)^{\otimes n}\otimes(A'^1)^{\otimes m} ample. Choose a positive power qq which is globally generated. The equivalence between strict graded bundles and their degree-one parts shows that the corresponding graded bundle on QK′Q_{K'} is globally generated by homogeneous sections. Global generation descends to QQ by CLD25 paragraph 10.2.4. For completeness, the descent uses H0(Q,D)⊗KK′≃H0(QK′,DK′),D=(M⊗n⊗A⊗m)⊗q.H^0(Q,D)\otimes_K K'\simeq H^0(Q_{K'},D_{K'}), \qquad D=(M^{\otimes n}\otimes A^{\otimes m})^{\otimes q}. This flat base-change identity follows by computing sections on a finite affine cover and its intersections. Express a finite set of homogeneous generators on the right as sums of tensors. The finitely many homogeneous components of their coefficients in H0(Q,D)H^0(Q,D) generate after extension to K′K'. Faithful flatness makes the cokernel of their evaluation map zero already on QQ. Thus DD is globally generated, as required. The exponent qq may depend on m,nm,n; no uniform exponent is needed. ◻

The proof uses the classical nef-plus-ample theorem in the same way as the strict argument of GJR Proposition 7.10. The additional work is the graded splitting and descent of homogeneous generation. It supplies the precise semiampleness needed in Proposition 19.18.16, without claiming that a purely numerical descent proves analytic psh descent.

Theorem 19.18.22. (Integral PL nefness gives generated psh). Let kk be any complete non-Archimedean field, allowing trivial valuation. Let XX be Hausdorff and boundaryless, and let LL have an integral PL metric near x∈Xx\in X. If L~x\widetilde L_x is nef, then a local weight hh of LL belongs to C25(U)\mathcal C_{25}(U) on an ordinary open neighbourhood UU of xx. More precisely, one can choose an integral PL weight aa on a compact neighbourhood VV, a real constant BB, and a single U⊂VU\subset V such that hj=h+a+Bj↓h,sup⁡U∣hj−h∣≤sup⁡V∣a+B∣j,\begin{equation} h_j=h+\frac{a+B}{j}\downarrow h, \qquad \sup_U|h_j-h|\leq\frac{\sup_V|a+B|}{j}, \end{equation} and each hjh_j is locally in UX\mathcal U_X on UU. Neither strictness nor a nontrivial valuation is required.

Proof. Represent L~x\widetilde L_x by a nef graded line bundle MM on a proper premodel. Graded Chow’s lemma, CLD25 Lemma 10.7.3, gives a proper birational quasi-projective dominating premodel QQ. Since the original premodel is proper, QQ is proper and projective in the graded sense. Pullback preserves nefness. Replace MM by its pullback and choose a very ample graded line bundle AA on QQ. These operations concern only the two fixed residual bundles.

The essential surjectivity of CLD25 Theorem 17.4.9(b) lifts πQ∗A\pi_Q^*A to an integral PL metric on an analytically trivial line bundle near xx. Denote its weight by aa, and fix a residual isomorphism. Apply Proposition 19.18.20 simultaneously to LL and this auxiliary metrized line, with their prescribed M,AM,A on QQ. It gives V,S,ρ,P,g,L,AV,S,\rho,P,g,\mathcal L,\mathcal A satisfying the first two hypotheses of Proposition 19.18.16. Lemma 19.18.21 gives its third hypothesis.

Set U=Int⁡X(V)∩ρ−1(P)U=\operatorname{Int}_X(V)\cap\rho^{-1}(P). That proposition proves that h+ϵah+\epsilon a is locally first-stage and belongs to C25(U)\mathcal C_{25}(U) for every positive rational ϵ\epsilon. Choose BB with a+B≥0a+B\geq 0 on compact VV. Its final argument yields (19.18.11) and h∈C25(U)h\in\mathcal C_{25}(U). Every analytic unit and every open neighbourhood in this argument was constructed over kk. The splitting extension in Lemma 19.18.21 was used only to establish a statement about graded line bundles; its global generators were descended before the analytic regularization was applied. ◻

This fills the prescribed-premodel step isolated in version 0.10. It reconstructs the integral PL implication intended in CLD25 Proposition 17.8.4 by making its domain and bundle compatibility explicit. It is a synthesis of the cited reduction and algebraic results, not a new regularization theorem for arbitrary continuous weights. In particular, no assertion that h∈UX(U)h\in\mathcal U_X(U) has been made: the neighbourhoods supporting smooth approximations of the individual hjh_j may depend on jj.

The completed integral PL comparison

Corollary 19.18.23. (Integral PL current, residue and generated class). Let XX be Hausdorff, boundaryless and pure of dimension n≥1n\geq 1 over any complete non-Archimedean field. Let LL have an integral PL metric. At a point xx, the following conditions are equivalent:

  1. c1(L)c_1(L) is positive near xx against all original CLD compactly supported smooth tests;

  2. L~x\widetilde L_x is nef;

  3. a local weight of LL belongs to C25\mathcal C_{25} near xx.

Proof. For d=1d=1, Theorem 19.18.12 gives (1)⇔\Leftrightarrow(2), with its named product, noncharging, lifting and mass inputs. Theorem 19.18.22 gives (2)⇒\Rightarrow(3). To prove (3)⇒\Rightarrow(1), note that current positivity is local, holds for smooth psh weights, and is preserved by locally uniform limits of continuous weights. Indeed, for a fixed compactly supported smooth test η\eta, uniform convergence on its support gives ⟨d′d′′[hj],η⟩→⟨d′d′′[h],η⟩\langle d'd''[h_j],\eta\rangle\to\langle d'd''[h],\eta\rangle. The class of continuous current-positive weights is therefore closed under all the operations generating C25\mathcal C_{25}. ◻

Remark 19.18.24 (Scope of the completed implication). The semiample case still has the stronger conclusion of Theorem 19.18.15: one decreasing smooth sequence on one neighbourhood. The general integral nef case has the generated-class conclusion of Theorem 19.18.22. Neither result settles the continuous predecessor replacement in CLD25 paragraph 8.1.11 or the passage from an arbitrary continuous positive current potential to a locally smooth-approximable potential. The general analytic Bedford–Taylor, capacity and Green-form questions retain their separate hypotheses and unresolved steps.

For comparison with GJR, Theorem 19.18.17 allows real PL coefficients in the strict, nontrivially valued category and uses formal models. Theorem 19.18.22 treats integral PL metrics with arbitrary homogeneous degrees over any complete field, using graded reductions. Their common mechanism is a fixed reduction tube, an ample residual perturbation, local regularization of finite maxima, and a continuous limit. The categories and the strengths of the approximation conclusions should remain visible when either theorem is cited.

References. Temkin [Tem04] (Lemma 2.4, Proposition 3.1(ii), Lemma 4.4 and Theorem 4.5); CLD25 v2, paragraphs 10.2.4, 10.3.2–10.3.4, 10.4.11 and 10.7.1, Lemma 10.7.3, Lemma 17.3.2, Corollary 17.4.5, Theorem 17.4.9 and Proposition 17.8.4. The proofs above supply the prescribed-cover construction, map on the reduced closure, compatible bundle identifications, and fixed-domain deduction. The named germ correspondence, graded nef base change, Chow lemma and residual lifting remain imported theorems.

19.19 Relative volumes and the orthogonality mechanism

A local positive current does not by itself define a global envelope theory. The latter requires a global class of semipositive metrics and an approximation theorem in that class. Relative volumes provide a separate algebraic route to the derivative needed by the variational method. This route explains both the strength and the scope of the results of Burgos Gil–Gubler–Jell–Künnemann–Martin and Boucksom–Gubler–Martin. Use additive metric weights. Let XX be a projective geometrically integral KK-variety of dimension nn, let LL be ample, and let ϕ\phi be a continuous semipositive metric. The relative volume vol⁡(L,ϕ1,ϕ0)\operatorname{vol}(L,\phi_1,\phi_0) is the asymptotic relative determinant norm of the sup norms on H0(X,mL)H^0(X,mL), with the normalization of [BGM22]. Their Theorem A, valid over an arbitrary complete non-Archimedean field, asserts ddt∣t=0vol⁡(L,ϕ+tf,ϕ)=∫Xanf MA⁡(ϕ),f∈C0(Xan).\begin{equation} \left.\frac{d}{dt}\right|_{t=0} \operatorname{vol}(L,\phi+tf,\phi) =\int_{X^{\mathrm{an}}} f\,\operatorname{MA}(\phi),\qquad f\in C^0(X^{\mathrm{an}}). \end{equation} Here ϕ+tf\phi+tf need not remain semipositive. Differentiation transverse to the semipositive cone is part of the theorem. Over a nontrivially valued field semipositivity is the uniform closure of nef model metrics; the arbitrary-field statement uses the compatible Fubini–Study formulation. It must not be read as a differentiability theorem for every local CLD-psh function.

Proposition 19.19.1. (From the volume derivative to orthogonality). Assume continuity of semipositive envelopes for (X,L)(X,L). For a continuous metric ψ\psi, let PψP\psi be its semipositive envelope. Then ∫Xan(ψ−Pψ)MA⁡(Pψ)=0.\int_{X^{\mathrm{an}}}(\psi-P\psi)\operatorname{MA}(P\psi)=0.

Open proof obligation (RC2). Establish the required volume–energy identity for continuous envelopes; BGM22 Equation (0.2).

NAG-900384 · Proof status

Since the integrand is continuous and nonnegative, the measure is supported on the contact set {Pψ=ψ}\{P\psi=\psi\}. Indeed, on any compact subset where the gap is at least ϵ>0\epsilon>0, its mass is at most ϵ−1\epsilon^{-1} times the displayed integral. This explains the precise place where orthogonality enters the variational solution of a Monge–Ampère equation. The earlier discretely valued result [BGJKM20] (Theorem C), in residue characteristic zero on a smooth projective variety, removes the additional algebraicity condition in the corresponding BFJ orthogonality result. The arbitrary-field volume theorem is later and more general as a differentiability statement. Its orthogonality consequence still assumes continuity of envelopes. “Arbitrary field” in the former theorem does not erase the latter hypothesis. This also identifies why Proposition 16.7.1 is compatible with these theorems. Its polyhedral PC class has no automatic relative-volume identity with the stated derivative. The formal envelope inequalities alone do not force contact-locus support. The algebraic volume input is doing substantive work.

References. [BGJKM20] (Theorems B–D); [BGM22] (Theorems A–B and Equation (0.2)).

Finite-energy measures and their topology

Boucksom–Jonsson formulate a global theory from a space D\mathcal D of test functions, a positive class ω\omega, an energy EωE_\omega and an admissible cone Dω\mathcal D_\omega. With probability-measure normalization, the dual energy is Jω(μ)=sup⁡ϕ∈Dω(Eω(ϕ)−∫ϕ dμ),Mω1={μ:Jω(μ)<∞}.J_\omega(\mu)=\sup_{\phi\in\mathcal D_\omega} \left(E_\omega(\phi)-\int\phi\,d\mu\right), \qquad \mathcal M^1_\omega=\{\mu:J_\omega(\mu)<\infty\}. The strong topology is the coarsest refinement of weak convergence making JωJ_\omega continuous. It contains strictly more information than weak convergence alone. Positivity and total mass one do not imply finite energy. Their Theorem A constructs a complete quasi-metric on this space, compatible with the Dirichlet functional on Monge–Ampère images, under their synthetic orthogonality property. The density of those images is one conclusion; an actual local CLD potential for every measure is not. Their Definition 2.15 expresses orthogonality through a directed limit of admitted test potentials below an obstacle. It is not an assumption that every envelope has a continuous representative. The paper verifies its synthetic hypotheses for projective Berkovich spaces over arbitrary non-Archimedean fields. Thus it would be incorrect to carry the continuity-of-envelopes hypothesis from Proposition 19.19.1 into every finite-energy measure theorem, or conversely to infer that all continuous envelopes have been constructed from the synthetic result. The two formulations must be compared at the level of their definitions.

References. [BJ-S] (Definitions 2.1, 2.15 and Theorems A–B).