18.1 Forms, positivity, and intersection theory
| Line of work | Main geometric problem | Relation to CLD |
|---|---|---|
| Tropicalization and canonical measures | Describe analytic images, multiplicities, degeneration and equidistribution | Supplies geometry used by the calculus; important parts predate CLD12. |
| Gubler’s forms and currents | Present original analytic forms using very affine algebraic charts | An equivalent presentation under the stated hypotheses, not an enlarged sheaf. |
| Gubler–Künnemann | Multiply polyhedral curvature and construct local intersections | Enlarges the differential algebra to delta-forms while retaining compatibility with original tests. |
| Gubler–Jell–Rabinoff | Use numerical harmonic functions as coordinates | Enlarges the smooth coordinate class and hence the test space; produces weakly smooth forms and strong currents. |
| Gubler–Rabinoff | Detect PL positivity and define classical psh analytically | Compares model geometry and curve tests; supplies independent routes to specified positivity equivalences. |
| Model, toric, and variational theories | Build semipositive metrics, envelopes, measures and solutions | Global or symmetry-restricted theories requiring explicit comparison with local currents. |
The purpose of the table is to prevent two opposite mistakes. These papers are not all reformulations of CLD, but neither are they unrelated theories that merely use the same words. Their common local calculations involve slopes, Hessians, weighted integration and metric variation. Their differences concern the allowable representatives, the tests, the category of spaces, and the mode of passage to limits.
The following comparison keeps the objects and their test spaces explicit. Chapters 19–21 then use these distinctions in the study of model metrics, canonical measures and local heights.
18.2 Objects, morphisms, and comparison hypotheses
Every comparison has four entries: the underlying space and field, the object being compared, the operation or property, and the tests used to detect equality. A statement about one entry cannot silently change the others. For example, equality of a Chern measure is a top-degree comparison; equality of first Chern currents tests lower-degree forms; equality of metrized line bundles also retains constants and transition data.
There are three especially useful kinds of theorem in this subject. An equivalence of presentations, such as Gubler’s algebraic presentation of original CLD forms, changes how the same object is described. An enlargement, such as the weakly smooth sheaf, changes the allowable coordinates and hence the tests. A comparison criterion, such as a PL model-positivity theorem, identifies properties on a restricted class of objects. None of these three kinds should be substituted for another.
The tables and proofs below can be used as a checklist while reading a new paper. They also explain why the same notation , , or psh can have different domains in different sources. A correct formula includes its domain even when that domain is suppressed typographically.
18.3 Comparison of forms, potentials, and metrics
The preceding definitions give two different kinds of comparison: maps between spaces of forms and currents, and implications between positivity conditions on a fixed function or metric. Keeping these two comparisons separate prevents several misleading identifications.
Forms and their dual test spaces
| Construction | Defining data | Comparison justified here |
|---|---|---|
| Original CLD forms | Analytic-unit tropicalizations, superforms, common refinements | Gubler’s algebraic presentation gives the same sheaf under Theorem 8.1.4. |
| GJR weakly smooth forms | Numerically harmonic coordinate functions | Original forms inject into weakly smooth forms; strictness can occur. |
| Piecewise smooth forms | Smooth formulas on polyhedral pieces | The derivatives are ; current derivatives also see corners. |
| GK delta-forms | Smooth coefficients and universally closed polyhedral generators | , with the generalized class. |
| Original currents | Continuous dual of | Other test theories restrict to these tests. |
| Strong currents | Continuous dual of | Restriction ; see Proposition 17.2.11 for its topological qualifications. |
| Delta-currents | Continuous dual of | Restriction ; functorial detection is a separate theorem. |
The table gives no general inclusion between weakly smooth forms and delta-forms. A larger test space reverses the natural arrow on continuous duals. Equality on original tests is consequently a weaker assertion than equality on all tests of an enlarged theory unless an additional detection theorem applies.
Regularity determines the available theorem
| Input | What is justified | Further conclusion needing another input |
|---|---|---|
| Original smooth | Positive Hessian form and positive current agree; pullback preserves psh. | No automatic global model approximation on a prescribed line. |
| Current positivity, first-stage decreasing approximation, and the continuous mixed product. | No automatic comparison with every continuous current-psh function. | |
| Original current positivity and generated-class functoriality. | No automatic one-step smooth approximation or equality of all continuous generated stages. | |
| Real PL under Proposition 19.2.11 | Model, generated, and classical curvewise positivity are equivalent under the stated good Hausdorff boundaryless strict hypotheses. | The original-current converse is the additional source comparison in Remark 19.3.1. |
| Arbitrary continuous metric | A well-defined first Chern current. | Local or global positive approximation must be established. |
| Unbounded potential | A current when locally integrable. | Multiplication by another current, a non-pluripolar product, and energy finiteness are separate requirements. |
Relation with adelic metrics and Deligne pairings
The same local distinction remains relevant in adelic geometry. A model line bundle, an actual collection of local norms, and the associated collection of curvature currents carry different data. An adelic completion also specifies compatibility of models and a mode of convergence. In particular, the equality of two local currents does not supply the lost constants of the metrics, nor does a local positivity statement assert a global numerical nef condition.
For example, let be the trivial line on a smooth projective analytic curve and compare the constant norms and . Both have zero curvature. Any construction retaining the norm of the distinguished section still distinguishes them when . This elementary observation already explains why a current comparison is not an isometry theorem.
The present book establishes the vocabulary and local comparison results needed before such applications. It does not use any companion research manuscript as a source for the definitions or for the corrections in this volume. The earlier internal research summaries are not used as additional theorems of this series.
A terminology concordance
The following translations retain the mathematical distinctions needed when reading the French sources. A shared English word is not an additional comparison theorem.
| Source term | Meaning used here | Precise reference in this book |
|---|---|---|
| forme lisse | An original smooth form obtained from analytic-unit tropicalizations | Definition 8.1.1 |
| courant de bidimension | A functional on tests of bidegree | Definition 9.1.2 |
| support propre in CLD12 | Compact support disjoint from the analytic boundary | Definition 9.1.1 |
| positif, faiblement positif, fortement positif | Three different cones, with complementary dual cones for currents | Definition 7.2.2, Definition 9.1.5 |
| localement psh-approchable in CLD12 | Local uniform smooth psh approximation on one fixed neighbourhood | Definition 11.2.1, Remark 11.2.6 |
| localement approchable in CLD12 | A difference of two functions in that positive approximation class | Remark 11.2.6 |
| localement approchable in CLD25 | The first decreasing smooth psh stage | Definition 11.2.2, Proposition 11.2.3 |
| psh continue in the two CLD versions | Current positivity in CLD12; a continuous generated sheaf in CLD25 | Definition 11.1.1, Definition 11.2.2 |
| Weakly smooth, in GJR | Smooth superforms pulled back along harmonic tropicalizations | Definition 17.2.5 |
| Strong current, in GJR | A continuous functional on weakly smooth test forms | Definition 17.2.10 |
| Strongly positive current | A symmetric current nonnegative on weakly positive tests | Definition 9.1.5 |
| Functorially psh, in GK | Current positivity after all morphisms in the stated algebraic category | Definition 16.2.1 |
| Classically psh, in GR | Subharmonicity after analytic extension and every smooth analytic curve test | Definition 19.2.1, Definition 19.2.2 |
For the GJR source, the earlier arXiv version and the version cited here have different numbering. In v2, model semipositivity implies current psh in Theorem 7.15, with harmonic consequence Corollary 7.16 and metric Remark 7.17. The corresponding v3 references are Theorem 7.14, Corollary 7.15, and Remark 7.16. The curve regularization result in the body of v3 is Proposition 15.11. The bibliography pins v3, and Theorem 17.2.16 preserves its decreasing-net formulation.
The two meanings of the word “strong” can be read directly from their definitions: Neither choice determines the other. Likewise, weak smoothness has no relation by definition to weak positivity.
18.4 Extensions of the original calculus
The original construction gives a local differential language, integration and currents, logarithmic divisor identities, a positive approximation calculus, and the comparison with model measures. Its scope is already substantially broader than a measure theory on proper algebraic varieties.
Three obstacles explain much of the subsequent literature. First, singular but geometric weights such as PL model functions are not automatically original smooth functions; an algebra that can multiply their curvature terms is desirable. This leads toward GK delta-forms. Second, model-harmonic functions can be larger than the original smooth pluriharmonic class; using them as coordinates leads toward GJR weakly smooth forms. Third, positivity of a distribution does not itself exhibit positive approximants or a local potential for an arbitrary closed current. This leads toward regularization, curve criteria, and further pluripotential theory.
There are also literal statements that need repair. The strong-positive cone does not span all symmetric forms in middle degree; a continuous facewise-convex function can have negative corner mass; and certain printed signs disagree with the total-degree algebra. These are treated in Appendix A, together with the usable replacements. The existence of such corrections does not erase the geometric construction of the theory, but it does require care when a later proof cites the affected lemma.
The standard model metric on gives a common instance of these constructions: its local weight has a slope jump, its Chern measure is the unit mass at the Gauss point, and its Green function satisfies Poincaré–Lelong. The metric and height calculations are given together in Examples 20.3.1 and 22.2.1. A constant rescaling leaves the curvature fixed but changes the Green function; the local-height recursion records that change.
18.5 What a comparison theorem preserves
The constructions in the preceding chapters can be compared at several levels: functions, metrics, forms, currents, measures and cohomology classes. Moving from one level to another usually forgets information. The following elementary calculations explain the resulting limitations. They are useful when a theorem is applied outside the setting in which it was first stated.
A top-degree measure does not determine every curvature factor
On , consider the convex quadratic functions Their Hessian matrices are and . Thus , while This equality is simply equality of the determinants. It neither identifies the matrices nor identifies their mixed products with an independently chosen form. For instance, wedging with detects the coefficient in the second direction and gives different results.
There is no conflict with a uniqueness theorem for a Monge–Ampère equation in a fixed class. Such a theorem supplies additional information: a fixed reference curvature, a boundary or asymptotic condition, and usually a normalization of the potential. The two quadratic functions above do not satisfy the same prescribed asymptotic condition. On a real torus their constant Hessians also define different cohomology classes.
Consequently, the identification of a Chambert–Loir measure with a top-degree analytic product is an important comparison theorem, but it is not by itself an identification of the underlying differential sheaves. The latter requires comparison in every bidegree and compatibility with restriction, multiplication and differentiation. The separate constructions in Chapters 8, 16 and 17 provide precisely the additional objects that must be compared.
A tropical pushforward can lose an analytic measure
Example 18.5.1. Let be complete and nontrivially valued. In , let be the rational point and let be the Gauss point with . Then , but both have tropical coordinate . Hence The bounded continuous function has value zero at and value at , so the two analytic measures are distinct.
For measures supported on a prescribed section of tropicalization, the situation changes. If , then , and therefore The section makes recovery possible. In the toric Monge–Ampère comparison, invariance and the canonical section are substantive hypotheses: they identify a class on which tropical pushforward retains all the required information. Removing those hypotheses would produce the ambiguity of Example 18.5.1.
The same reasoning applies to a finite family of moment coordinates. A larger family can distinguish more points, and Payne’s inverse-limit theorem uses a sufficiently rich system of them all. A result on one selected tropicalization must not be read as a result on that inverse limit without checking compatibility under refinement. Chapter 6 proves the affine reconstruction and identifies exactly where the complete coordinate family enters.
18.6 Changing the test space of a current
Let be real locally convex Hausdorff vector spaces, where the inclusion is continuous but the topology on need not be the topology induced by . Continuous duals have a restriction map The following facts separate detection of a current from extension of a current. They are elementary functional analysis, but both issues occur when original smooth tests are compared with weakly smooth or delta-form tests.
NAG-900326 Proof review pending
Proposition 18.6.1. The map is injective if and only if is dense in . A functional belongs to its image if and only if there is a continuous seminorm on such that
Proof. A continuous functional vanishing on a dense subspace is zero. Conversely, if the closure of is a proper closed subspace, the Hausdorff locally convex quotient by that closure has a nonzero continuous functional, by Hahn–Banach separation. Its pullback is a nonzero element of .
If extends continuously, continuity of the extension bounds it by a constant times the maximum of finitely many continuous seminorms; absorb that constant into . Conversely, the stated domination lets the real Hahn–Banach theorem extend to a linear functional on dominated by . That extension is continuous. ◻
Example 18.6.2. (Density does not imply existence of extension). Take with the uniform norm and with its usual smooth topology. The inclusion is continuous and dense, so restriction is injective. The functional is continuous on , but has no continuous extension to . Indeed, The required uniform-norm bound is impossible.
NAG-900327 Comparison requirements remain
NAG-900327 · Comparison requirements remain
For actual differential tests, this means that a statement about the direction of an arrow between sheaves does not by itself determine injectivity or surjectivity of the dual arrow. One needs density for uniqueness and a continuity estimate for existence. Moreover, the Hahn–Banach extension in Proposition 18.6.1 does not automatically preserve positivity, closedness, wedge products or geometric representatives. Each of those requirements adds another condition.
An equality first checked on decomposable exterior-product tests has the same issue. It extends to the full test space if their span is dense in the relevant test topology and both sides are continuous there. An algebraic spanning statement on one chart is not a substitute for this topological density on the actual quotient. In the Green-form constructions of Chapter 22, the diagonal and cutoff arguments are used to reach all actual compactly supported tests, not just one chosen family.
18.7 A common chain of comparison and its hypotheses
Consider a line bundle with a continuous metric. To pass from its geometric construction to an intersection measure, one often uses the following chain of results. Each arrow has a distinct mathematical content.
First, choose local frames and form their weights . On overlaps, . Analytic transition functions are units, so their logarithms are original smooth and have zero Hessian. This proves that the local currents glue. It says nothing yet about their positivity.
Second, a model or curvewise criterion may establish positivity in its specified class. If the available theorem produces a locally uniform positive approximation on one fixed neighbourhood, then the weights lie in the admitted class of Chapter 11. This is stronger information than a bare inequality against current tests. For real PL metrics, Chapter 19 identifies the hypotheses under which the geometric criterion supplies that approximation.
NAG-900328 Conditional on stated inputs
NAG-900328 · Conditional on stated inputs
Third, the continuous product theorem defines local mixed currents. Gluing these products uses the special transition functions of the line bundle. Their smooth zero-Hessian property gives the required kernel action on closed backgrounds; there is no need to assert that every continuous function with zero ambient curvature has the same property. The distinction is essential in view of the cross example.
Finally, comparison with the model measure is proved on the geometric approximants and passed to the limit in the theorem’s topology. At this stage one obtains a top-degree measure. Total mass is controlled by the algebraic intersection number on a proper space. Neither properness nor that mass identity proves local regularization of arbitrary current-psh weights. They enter after the potential class and its products have been established.
The construction therefore combines different contributions without identifying their domains by notation: CLD supplies the intrinsic differential language; Gubler’s algebraic presentation makes the original charts accessible in algebraic geometry; GK retains singular polyhedral curvature in an intersection algebra; GJR and GR supply harmonic and curvewise geometric comparisons; model and toric theories identify the resulting measures and heights in their respective classes. The proofs used at one stage should be cited at that stage, with their field, regularity and test-space assumptions retained.
References. The examples and functional-analytic criteria above are elementary. The geometric comparisons are the separately stated theorems of [CLD12] [Gubler] [GK17] [GK19] [GJR] [GRPL] [GRPSH] [BGJK25]; the proof locations in this book are Chapters 8, 13, 16, 19 and 20.