24 · Limits, cohomology and remaining comparisons

20 NAG Tags

24.1 Topological limits, differential limits, and Hodge questions

Payne’s inverse-limit theorem

Payne’s inverse-limit theorem reconstructs an analytification from a sufficiently rich system of tropicalizations as a topological space. A point is recovered from its compatible values in all those tropicalizations. The system of embeddings and transition maps is part of the theorem.

A sheaf of differential forms is additional data. To recover it as a direct limit of tropical sheaves, one must compare the local presentations, quotient relations, and restriction maps. A current is a continuous dual, so one must also compare topologies. Cohomology introduces another passage to kernels and images. None of these operations is determined solely by a homeomorphism of underlying spaces.

A concrete obstruction to an automatic Hodge transfer

Even if a filtered direct limit of complexes is exact in an algebraic category, the problem at hand may involve topological vector spaces, support conditions, and completed or continuous duals. Images need not be closed, and a primitive available on one tropical presentation may acquire a different singular support after refinement. A claimed Hodge or Aeppli comparison must therefore specify the category in which the limit is taken.

The curve calculation in Chapter 17 is an effective sanity check. Its weakly smooth Dolbeault dimensions involve the first Betti number of the analytic topology, while the algebraic genus also contains residual contributions. A proposed identification with algebraic Hodge theory must explain those contributions rather than assuming the two genera coincide.

Differential and cohomological comparison requirements

First compare finite-chart smooth complexes on a fixed relatively compact analytic domain. Next prove compatibility under refinement, including the zero relations and test topology. Then identify which compact support condition is preserved. Only after these steps should one compare cohomology or seek representatives with an exact prescribed support complement.

For Green problems, the last support step is essential. A primitive on an alteration, a current after proper pushforward, and an actual differential form on the original complement are different outputs. A limit theorem should state which output it produces and should prove descent of that output, not just convergence of its pairings on a smaller collection of tests.

References. [Pay09]; [GJRc]; the explicit representative constructions in [Mih24].

24.2 Further analytic problems

Regularization with a fixed test theory

The basic comparison asks when a continuous function with positive original CLD Hessian has local smooth positive approximants on one fixed neighbourhood. The forward implication from approximation to current positivity is elementary; the converse is a regularization theorem. A useful result should state the field, analytic category, topology of approximation, and whether the neighbourhood is uniform throughout the approximation.

Such a theorem would make many current inequalities usable in the established continuous BT calculus. It would also clarify comparisons between raw current positivity, generated CLD25 psh, numerical model positivity, and curvewise psh. The PL cases and curve results provide strong tests, but proving the general continuous statement requires controlling all positive tests, not only a selected finite chart or a family of curves whose detecting power has not been established.

A local potential theorem for closed currents

Given a positive closed (1,1)(1,1)-current TT, a local potential theorem seeks uu with d′d′′[u]=T\mathrm d'\mathrm d''[u]=T in a specified regularity class. This is logically distinct from regularizing a potential already known to exist. It also requires more than a degree-zero or primitive condition tested on a restricted subspace: the equality must hold on the full test space stated in the theorem.

A proof would need a local exactness statement for the differential complex, an analytic mechanism giving the required regularity, and a way to preserve positivity. The tropical and toric cases suggest techniques, but a primitive on one real chart need not descend through all analytic relations. This is one reason the intrinsic descent problem is central to the long-term programme.

Arbitrary-background action and kernel descent

The continuous recursion already defines an action of suitable potentials on an arbitrary ordinary positive closed background current. To promote this to an action of curvature objects, one needs the kernel condition (14.2.1). The cross example proves that abstract tropical positivity and closedness do not suffice in full generality.

A productive analytic theorem would identify a geometric class of kernel functions that annihilates every allowed background current, or a geometric class of background currents for which the full kernel acts trivially. Smooth pluriharmonic functions give the elementary case; skeletal BT currents give a structured case. The general smooth-Berkovich question must use the additional analytic geometry explicitly.

Bounded and unbounded potentials

Beyond continuous potentials, a complete theory should specify monotone convergence, locality on the appropriate topology, capacities, quasi-continuity, negligible sets, comparison and domination principles, and suitable Dirichlet problems. These are related but separate theorems. A non-pluripolar product also requires a definition for unbounded potentials and a proof that truncation and exceptional-set choices are compatible.

The toric theory and global energy theories give established frameworks in which many of these issues can be addressed. The goal of a general local analytic theory is to connect such frameworks with the original test calculus while retaining their domains and comparison maps. It is not achieved by relabelling every model-theoretic θ\theta-psh function as a locally integrable CLD potential without proof.

Actual Green forms and exact support

A strong Green statement for a codimension-rr cycle ZZ may ask for an actual form γ\gamma of degree (r−1,r−1)(r-1,r-1) on Xan∖∣Z∣anX^{\mathrm{an}}\setminus|Z|^{\mathrm{an}} with ∫Xan∖∣Z∣anγ∧d′d′′η=−∫Zanη\int_{X^{\mathrm{an}}\setminus|Z|^{\mathrm{an}}}\gamma\wedge\mathrm d'\mathrm d''\eta =-\int_{Z^{\mathrm{an}}}\eta for every actual compactly supported test in the specified calculus. The complete-intersection theorem gives an explicit model of such a statement in its own category. A general cycle requires a global construction and compatibility of local representatives.

The requirements are substantive: the complement is the original complement; the representative lies in the chosen sheaf; the testing is not restricted to external products or a chosen tropical chart; and field extension must descend with the correct degree and the original absolute value. A current identity after arbitrary proper pushforward does not by itself satisfy these requirements. It may be a useful intermediate step if its relation to the desired representative is proved.

24.3 What filtered limits do preserve

A topological inverse-limit description and a differential direct-limit description are different statements. Once a differential complex has actually been identified with a filtered direct limit, its cohomology is easy to compute algebraically. The difficult part is often establishing that identification with the required support and topology. We separate these steps.

Theorem 24.3.1. Let (Ci∙,di)(C_i^\bullet,d_i) be a filtered system of complexes of real vector spaces, with transition maps commuting with the differentials. Then the canonical map is an isomorphism: lim→⁡iHk(Ci∙) → ∼  Hk(lim→⁡iCi∙).\varinjlim_i H^k(C_i^\bullet) \ \xrightarrow{\ \sim\ }\ H^k\left(\varinjlim_i C_i^\bullet\right). Injectivity of the transition maps is not required.

Proof. In a filtered colimit of vector spaces, every element has a representative at one stage, and an element represented at a stage is zero precisely when it becomes zero at some later stage. Let zz represent a cocycle in the limit. Choose a representative zi∈Cikz_i\in C_i^k. Its differential is zero in the colimit, hence dizid_i z_i becomes zero at a later stage jj. The image zjz_j is then a cocycle at that stage, proving surjectivity.

For injectivity, suppose a stage cocycle ziz_i becomes a boundary in the limit, say z=dyz=dy. Represent yy at another stage and take a common later stage for both representatives. The equality z=dyz=dy holds at a further stage by the same zero criterion. Thus the image of the cohomology class of ziz_i is already zero at that stage, so it is zero in the colimit of cohomology groups. ◻

Corollary 24.3.2. For a filtered system of double complexes with commuting transition maps, filtered colimits commute with Bott–Chern and Aeppli cohomology, defined algebraically by HBCp,q=ker⁡d′∩ker⁡d′′im⁡d′d′′,HAp,q=ker⁡d′d′′im⁡d′+im⁡d′′.H^{p,q}_{BC}=\frac{\ker d'\cap\ker d''}{\operatorname{im}d'd''}, \qquad H^{p,q}_{A}=\frac{\ker d'd''}{\operatorname{im}d'+\operatorname{im}d''}.

Proof. A numerator condition involves finitely many equations, so each equation that holds in the colimit holds after passage to one common later stage. A denominator relation involves finitely many preimages, which can also be represented at one common stage. The surjectivity and injectivity argument of the theorem then applies verbatim. This proof concerns algebraic quotients, without any assertion that the images are topologically closed. ◻

For an analytic application one must first construct a system whose colimit is the desired sheaf of forms or global test complex. A finite collection of local presentations can be refined by concatenating moment maps, but that observation alone does not identify a global complex on a noncompact space. Covers, restrictions, support control and the allowed quotient relations still have to be carried through. The theorem does not supply that geometric step.

Example 24.3.3. (Compact support and pullback). The constant function one on a point has compact support. Its pullback along R→{∗}\mathbb R\to\{\ast\} has support all of R\mathbb R. Thus an arbitrary map does not induce pullback between compactly supported test spaces. Properness over the relevant support, or a specified alternative support construction, is needed before forming a direct system of compactly supported complexes.

24.4 Why inverse limits of spaces are not enough

Example 24.4.1. (A topological test of the inference). Let Xj={0,1}jX_j=\{0,1\}^j with the discrete topology and the maps that forget the last coordinate. Their inverse limit is the compact space X={0,1}NX=\{0,1\}^{\mathbb N}. The pullback colimit of Hsing0(Xj,R)H^0_{\rm sing}(X_j,\mathbb R) consists of real functions depending on finitely many coordinates. However Hsing0(X,R)H^0_{\rm sing}(X,\mathbb R) consists of all real-valued functions on XX.

Indeed, a continuous path into XX has constant projection to every discrete XjX_j, hence is constant. Every singular one-simplex is therefore constant, so every zero-cochain is a cocycle and there are no zero-coboundaries. The function x↦∑j≥1xj3−jx\mapsto\sum_{j\geq 1}x_j3^{-j} does not depend on finitely many coordinates and gives an explicit class outside the colimit.

This is not a counterexample to Payne’s theorem or to a specific Berkovich comparison. It disproves only the general inference that a homeomorphism to an inverse limit automatically identifies singular or differential cohomology with the colimit of the finite-stage groups. Different cohomology theories have different continuity properties. For the intended differential theory, one must prove the required complex-level comparison.

24.5 Duality changes direction under a limit

Proposition 24.5.1. For a filtered system of real vector spaces, there is a natural algebraic isomorphism Hom⁡R(lim→⁡iVi,R)≃lim←⁡iHom⁡R(Vi,R).\operatorname{Hom}_{\mathbb R}(\varinjlim_i V_i,\mathbb R) \simeq\varprojlim_i\operatorname{Hom}_{\mathbb R}(V_i,\mathbb R).

Proof. A functional on the colimit restricts to a compatible family of functionals on the stages. Conversely, a compatible family assigns the same scalar to any two representatives of a colimit element, so it defines a unique linear functional. These constructions are inverse. ◻

With locally convex topologies, continuous duals require the chosen inductive-limit topology and its continuity criterion. Even when that criterion is available, the direction remains inverse on duals. One cannot replace it by a second direct limit without an additional theorem.

Example 24.5.2. (Perfect finite-stage pairings need not remain perfect). Take Vj=Wj=RjV_j=W_j=\mathbb R^j, with the standard dot product and transition maps adding a zero last coordinate. Each pairing is perfect and the pairings are compatible. Both direct limits are the finite-support sequence space R(N)\mathbb R^{(\mathbb N)}. Their induced pairing is nondegenerate, but the resulting map R(N)⟶Hom⁡R(R(N),R)=RN\mathbb R^{(\mathbb N)}\longrightarrow \operatorname{Hom}_{\mathbb R}(\mathbb R^{(\mathbb N)},\mathbb R)=\mathbb R^{\mathbb N} is not surjective. The functional v↦∑iviv\mapsto\sum_i v_i is well defined because vv has finite support, but it would require the infinite coefficient sequence (1,1,…)(1,1,\ldots) in the second factor. Thus nondegeneracy survives in this example while algebraic perfectness does not.

Lemma 24.5.3. (One sufficient stabilization condition). If all transition maps of a filtered system of finite-dimensional vector spaces are injective and their dimensions have a uniform finite bound, then the system stabilizes after some stage and its colimit is finite dimensional.

Proof. The nonempty set of dimensions has a largest value. Choose a stage realizing it. Every later injective map from this stage has a target of the same dimension, hence is an isomorphism. Filteredness makes these later stages cofinal. ◻

Such stabilization is one possible route for transferring finite-dimensional duality, but it is a hypothesis to prove for the actual transition maps. A common tropical refinement need not induce an injective map on a cohomology group merely because the refinement map is surjective on the underlying support.

24.6 Cohomological annihilation and a current equation

There is another distinction when cohomology is used to construct a Green current. Vanishing on the kernel of a differential gives an algebraic factorization. A continuous current requires a topological estimate, and an actual differential form requires further regularity. The following elementary result makes the first two steps precise.

Theorem 24.6.1. (A continuous lifting criterion). Let D:E→FD:E\to F be a continuous linear map of real locally convex spaces and let L:E→RL:E\to\mathbb R be continuous linear. A continuous linear functional GG on FF satisfying G∘D=LG\circ D=L exists if and only if there is a continuous seminorm pp on FF such that ∣L(e)∣≤p(De)(e∈E).|L(e)|\leq p(De)\qquad(e\in E).

Proof. If GG exists, continuity bounds it by a continuous seminorm on FF, giving necessity. Conversely, the estimate implies that LL vanishes on ker⁡D\ker D. Hence g(De)=L(e)g(De)=L(e) is well defined on D(E)D(E) and satisfies ∣g(y)∣≤p(y)|g(y)|\leq p(y). The real Hahn–Banach theorem extends gg to a linear functional GG on FF with the same bound. Since pp is continuous, GG is continuous. ◻

Example 24.6.2. (The kernel condition is not enough). Take E=F=ℓ2(N)E=F=\ell^2(\mathbb N), D(x)n=xn/nD(x)_n=x_n/n, and L(x)=∑nxn/nL(x)=\sum_n x_n/n. The functional LL is continuous by Cauchy–Schwarz, and ker⁡D=0\ker D=0. If G∘D=LG\circ D=L, evaluation on the standard basis forces G(en)=1G(e_n)=1 for every nn. No bounded functional on ℓ2\ell^2 has these coefficients. Explicitly, take xn=nx_n=n for 1≤n≤N1\leq n\leq N and zero otherwise. Then L(x)=NL(x)=N and ∥Dx∥2=N\|Dx\|_2=\sqrt N, so no uniform bound ∣L(x)∣≤C∥Dx∥2|L(x)|\leq C\|Dx\|_2 holds.

For a current equation on a pure nn-dimensional space, take E=Acn−r,n−rE=A_c^{n-r,n-r}, F=Acn−r+1,n−r+1F=A_c^{n-r+1,n-r+1}, D=d′d′′D=d'd'', and L(η)=−∫ZηL(\eta)=-\int_Z\eta. A functional GG supplied by the theorem is a current of bidegree (r−1,r−1)(r-1,r-1) satisfying the required transposed equation, with the book’s fixed differential sign convention. Merely knowing that ∫Zη=0\int_Z\eta=0 whenever d′d′′η=0d'd''\eta=0 supplies only the kernel condition. One still needs the continuous estimate.

Even after this estimate, Hahn–Banach gives no actual representative on Xan∖∣Z∣anX^{\mathrm{an}}\setminus|Z|^{\mathrm{an}}, no local integrability of a chosen coefficient expression, and no control of additional singular support. Those belong to a regularity and representation theorem. If the target asks for equality on delta-tests, the spaces E,FE,F in the argument must already be the full delta-test spaces with their topology. Solving the equation only on original smooth tests does not perform that enlargement.

The local complete-intersection construction of Chapter 22 addresses these stronger requirements by an explicit geometric representative and a proof on actual tests. The functional argument here serves a different purpose: it identifies which estimates and comparisons a general cohomological existence strategy would have to supply.

24.7 The coefficients carried by a tropical face

A tropical cohomology class remembers tangent directions as well as the topology of the underlying space. The coefficient system is already nonconstant at a vertex where several cells meet. We describe it first for a finite polyhedral complex in a real vector space; strata at infinity are then handled by quotient maps.

Let C\mathscr C be a polyhedral complex in NRN_{\mathbb R}. For a face σ\sigma, let LσL_\sigma be its parallel linear space. Define Fp(σ)=∑τ⊇σ⋀pLτ ⊆⋀pNR,Fp(σ)=Hom⁡R(Fp(σ),R).F_p(\sigma)=\sum_{\tau\supseteq\sigma}\bigwedge^p L_\tau \ \subseteq\bigwedge^p N_{\mathbb R},\qquad F^p(\sigma)=\operatorname{Hom}_{\mathbb R}(F_p(\sigma),\mathbb R). The sum is a sum of subspaces, not a direct sum. If σ⊆τ\sigma\subseteq\tau, then Fp(τ)⊆Fp(σ)F_p(\tau)\subseteq F_p(\sigma); dualizing gives a restriction Fp(σ)→Fp(τ)F^p(\sigma)\to F^p(\tau). At a boundary stratum of an extended tropicalization, first project to the quotient tangent space of that stratum. These maps define the multi-cotangent sheaf Fp\mathcal F^p.

Example 24.7.1. Let Y⊂R2Y\subset\mathbb R^2 be the union of the two coordinate axes. At the origin F1(0)=R2F_1(0)=\mathbb R^2, whereas on the interior of either axis F1F_1 is its one-dimensional tangent line. Thus a degree-one coefficient at the vertex is a covector with two components, and restriction retains the component along the chosen axis. For the tropical line with rays e1,e2,−e1−e2e_1,e_2,-e_1-e_2, the three outgoing restrictions are a,b,−a−ba,b,-a-b. Their sum is zero. This relation comes from the common ambient covector; it is not three independent endpoint values.

Example 24.7.2. Let XX be the union in R4\mathbb R^4 of the planes spanned by e1,e2e_1,e_2 and by e3,e4e_3,e_4. At the origin, F1(0)=R4F_1(0)=\mathbb R^4, but F2(0)=R(e1∧e2)⊕R(e3∧e4),dim⁡F2(0)=2≠6=dim⁡⋀2F1(0).F_2(0)=\mathbb R(e_1\wedge e_2)\oplus\mathbb R(e_3\wedge e_4), \qquad \dim F_2(0)=2\ne6=\dim\bigwedge^2F_1(0). Thus one must take exterior powers before summing the tangent spaces. Replacing FpF_p by ⋀pF1\bigwedge^pF_1 changes the theory at singular points.

From local closed forms to a coefficient sheaf

For a polyhedral space XX in the sense of Jell–Shaw–Smacka, put Lp=ker⁡(d′′:Ap,0⟶Ap,1).\mathcal L^p=\ker(d'':\mathcal A^{p,0}\longrightarrow\mathcal A^{p,1}). On an ordinary real open set its sections are locally constant primed pp-forms. Across a face, only their restrictions to the tangent spaces of adjacent cells are visible. This is why the appropriate coefficient sheaf is Fp\mathcal F^p.

Theorem 24.7.3. (Polyhedral comparison). Let XX be a paracompact, second countable Hausdorff polyhedral space with the extended-affine charts and boundary compatibility of [JSS]. There is a canonical isomorphism Lp≃Fp\mathcal L^p\simeq\mathcal F^p, and Hd′′p,q(X)≃Hq(X,Fp),Hd′′,cp,q(X)≃Hcq(X,Fp).H^{p,q}_{d''}(X)\simeq H^q(X,\mathcal F^p),\qquad H^{p,q}_{d'',c}(X)\simeq H_c^q(X,\mathcal F^p). If XX has a face structure, these groups are also its tropical cohomology groups, respectively with and without compact support.

Open proof obligation (RC2). Prove the polyhedral Poincare lemma and the fine/acyclic chart-sheaf route; the proof uses the stated JSS theorem, proposition and lemma.

NAG-900476 · Proof status

For p=0p=0 the coefficient sheaf is R\mathbb R, so this row is ordinary real cohomology. For p>0p>0 it is generally not determined by the topology. The two plane example explains the extra data without referring to an analytic space. A claim that Payne’s topological inverse limit automatically recovers all rows would have to recover these coefficient systems and their transition maps as well.

24.8 Compact support changes the Poincaré lemma

A radial homotopy usually spreads the support to the star centre and does not solve the compactly supported problem. The following explicit construction provides the required replacement on Rn\mathbb R^n. It also identifies the obstruction that integration detects.

Lemma 24.8.1. Write coordinates on Rn\mathbb R^n as (x,t)∈Rn−1×R(x,t)\in\mathbb R^{n-1}\times\mathbb R, and an ordinary compactly supported qq-form as ω=α(x,t)+dt∧β(x,t)\omega=\alpha(x,t)+dt\wedge\beta(x,t), with no dtdt in α,β\alpha,\beta. Choose ρ∈Cc∞(R)\rho\in C_c^\infty(\mathbb R) with integral one and put χ(t)=∫−∞tρ(s) ds\chi(t)=\int_{-\infty}^t\rho(s)\,ds. Define Iω=∫Rβ(x,s) ds,Eη=ρ(t) dt∧η(x),I\omega=\int_{\mathbb R}\beta(x,s)\,ds,\qquad E\eta=\rho(t)\,dt\wedge\eta(x), Hω=∫−∞tβ(x,s) ds−χ(t)Iω.H\omega=\int_{-\infty}^t\beta(x,s)\,ds-\chi(t)I\omega. Then HωH\omega is compactly supported and dH+Hd=id⁡−EIdH+Hd=\operatorname{id}-EI. Moreover Id=−dIId=-dI and dE=−EddE=-Ed.

Proof. Below the supports of β\beta and ρ\rho both terms in HωH\omega vanish. Above them they are both IωI\omega and cancel. The xx-support is contained in the compact projection of supp⁡ω\operatorname{supp}\omega. Thus the support is compact. Direct differentiation gives dω=dxα+dt∧(∂tα−dxβ).d\omega=d_x\alpha+dt\wedge(\partial_t\alpha-d_x\beta). Integrating its second coefficient gives Idω=−dxIωId\omega=-d_xI\omega, because α\alpha vanishes at both ends. The same calculation gives dE=−EddE=-Ed. Finally the dtdt-part of dHωdH\omega is dt∧(β−ρIω)dt\wedge(\beta-\rho I\omega), while its xx-part cancels the integral terms of HdωHd\omega and leaves α\alpha. Their sum is ω−EIω\omega-EI\omega. This also proves the formula in degree zero, where Hω=0H\omega=0. ◻

Theorem 24.8.2. For superforms on Rn\mathbb R^n, Hd′′,cp,q(Rn)=0(q≠n),Hd′′,cp,n(Rn)≃⋀p(Rn)∗.H^{p,q}_{d'',c}(\mathbb R^n)=0\quad(q\ne n),\qquad H^{p,n}_{d'',c}(\mathbb R^n)\simeq\bigwedge^p(\mathbb R^n)^*. The second isomorphism integrates each double-primed top-degree coefficient. In particular a compactly supported (p,n)(p,n)-form is d′′d''-exact with compact support if and only if all these coefficient integrals vanish.

Proof. Keep the primed exterior factor fixed. Up to the fixed degree sign in d′′d'', the remaining complex is the ordinary compactly supported de Rham complex, with coefficients in ⋀p(Rn)∗\bigwedge^p(\mathbb R^n)^*. Lemma 24.8.1 gives a homotopy equivalence with the same complex on Rn−1\mathbb R^{n-1} shifted by one degree; IEIE is the identity. Iterate this reduction. On R0\mathbb R^0, only degree zero remains and its cohomology is R\mathbb R. The iterated fibre integral is the ordinary top-degree integral up to the orientation sign. This proves the assertions and constructs a primitive whenever the obstruction vanishes. ◻

Combining this calculation with ordinary Poincaré exactness gives Euclidean duality: Hd′′p,0(Rn)H^{p,0}_{d''}(\mathbb R^n) pairs perfectly with Hd′′,cn−p,n(Rn)H^{n-p,n}_{d'',c}(\mathbb R^n) by integration. The pairing is the exterior-product pairing of two finite-dimensional coefficient spaces. The compact support in the second factor is essential.

Where tropical Poincaré duality holds

Theorem 24.9.1. (Jell–Shaw–Smacka). If XX is an nn-dimensional tropical manifold, locally modelled on matroidal fans with the boundary charts specified in [JSS] (Definition 4.15), integration induces an isomorphism Hd′′p,q(X)⟶(Hd′′,cn−p,n−q(X))∗,[α]⟼([β]⟼∫Xα∧β).H^{p,q}_{d''}(X)\longrightarrow \bigl(H^{n-p,n-q}_{d'',c}(X)\bigr)^*,\qquad [\alpha]\longmapsto\left([\beta]\longmapsto\int_X\alpha\wedge\beta\right).

Open proof obligation (RC2). Prove the local matroidal and global Poincare duality in the stated scope; the text supplies JSS Theorem 4.33.

NAG-900479 · Proof status

Proposition 24.9.2. Give the union YY of the two coordinate axes in R2\mathbb R^2 weight one on each axis. Then Hd′′0,0(Y)=RH^{0,0}_{d''}(Y)=\mathbb R and Hd′′,c1,1(Y)=R2H^{1,1}_{d'',c}(Y)=\mathbb R^2. The integration pairing is the sum of the two coordinates and is not perfect.

Proof. A d′′d''-closed smooth function is locally constant and YY is connected, proving the first assertion. Restriction identifies the complexes Ac1,∙(Y)A_c^{1,\bullet}(Y) with the direct sum of the corresponding complexes on the two real lines. To check surjectivity, extend an arbitrary compactly supported coefficient a(x)a(x) on the horizontal line as a(x)ρ(y)d′xa(x)\rho(y)d'x or a(x)ρ(y)d′x∧d′′xa(x)\rho(y)d'x\wedge d''x, with ρ=1\rho=1 near zero. This restricts to zero on the vertical line. The analogous vertical extension realizes the other summand. Tangential equality on the two axes is precisely the quotient relation for these bidegrees, proving injectivity. These restrictions commute with d′′d''.

By Theorem 24.8.2 each line contributes one top compactly supported class, detected by its integral. Integration on YY adds the two integrals. The class with integrals (1,−1)(1,-1) is nonzero but pairs to zero with every constant. This proves the failure. ◻

The cross is balanced. Its defect is that the two one-dimensional branches have independent top classes while the whole space has only one constant class. This supplies a concrete obstruction to deriving duality from balancing, and a separate reason not to transfer a tropical Hodge theorem through a topological inverse limit alone. It is not a counterexample to the theorem for tropical manifolds or to any analytic comparison proved with stronger hypotheses.

References. [JSS] (Sections 2–4); the relation with the analytic limit question is developed in Section 6.5 and Chapter 24.

24.10 The same topological star can carry different cohomology

Let YY be a fan consisting of mm distinct rays with nonzero primitive directions v1,…,vmv_1,\ldots,v_m in a real vector space VV. Give the rays positive weights wiw_i satisfying ∑iwivi=0\sum_iw_iv_i=0. Each ray is parametrized by t↦tvit\mapsto tv_i, t≥0t\geq 0. Define R:V∗⟶Rm,λ⟼(λ(v1),…,λ(vm)).R:V^*\longrightarrow\mathbb R^m,\qquad \lambda\longmapsto(\lambda(v_1),\ldots,\lambda(v_m)).

Proposition 24.10.1. For the ambient superform sheaf on YY, edgewise integration induces an isomorphism Hd′′,c1,1(Y)≃Rm/im⁡R.H^{1,1}_{d'',c}(Y)\simeq\mathbb R^m/\operatorname{im}R. Consequently its dimension is m−dim⁡span⁡{vi}m-\dim\operatorname{span}\{v_i\}. Integration on the weighted fan is induced by (bi)↦∑iwibi(b_i)\mapsto\sum_iw_i b_i.

Proof. For α∈Ac1,0(Y)\alpha\in A_c^{1,0}(Y) write α∣i=ai(t)d′t\alpha|_i=a_i(t)d't. Since d′′α∣i=−ai′(t)d′t∧d′′td''\alpha|_i=-a'_i(t)d't\wedge d''t, its integral on the iith ray is ai(0)a_i(0). The values at the vertex come from a common ambient covector, so the vector of integrals lies in im⁡R\operatorname{im}R. Hence the map to the quotient is well-defined. It is surjective, since independent bump top forms supported away from the vertex realize arbitrary edge integrals.

For injectivity, first subtract d′′(χλ)d''(\chi\lambda), where λ\lambda is a constant ambient primed covector and χ\chi a cutoff equal to one near the vertex, to reduce to a form β\beta with zero edge integrals. The local polyhedral Poincaré lemma gives β=d′′α0\beta=d''\alpha_0 near the vertex. Choose another cutoff supported there. Subtracting d′′(χα0)d''(\chi\alpha_0) makes β\beta vanish near the vertex, but changes its integrals by the vector of vertex values of α0\alpha_0. Add back d′′(χλ0)d''(\chi\lambda_0) for an ambient covector λ0\lambda_0 with those same vertex values. Its derivative vanishes near the vertex. The resulting form β~\widetilde\beta is supported away from the vertex and has zero integral on every ray.

On each ray put ai(t)=∫t∞b~i(s) dsa_i(t)=\int_t^\infty\widetilde b_i(s)\,ds, where β~∣i=b~i(t)d′t∧d′′t\widetilde\beta|_i=\widetilde b_i(t)d't\wedge d''t. The zero integral makes aia_i vanish near zero, and compact support makes it vanish for large tt. These edgewise primitives extend to ambient forms in disjoint neighbourhoods away from the vertex and glue by zero. Their d′′d'' is β~\widetilde\beta. Restoring the subtracted exact forms proves injectivity. Finally balancing says precisely that the functional ∑iwibi\sum_iw_i b_i annihilates im⁡R\operatorname{im}R. ◻

For the rank-two uniform matroid on mm elements, the rays are the coordinate vectors in Rm/R1\mathbb R^m/\mathbb R\mathbf 1. Their span has dimension m−1m-1, and their only linear relation is their sum. The proposition gives Hc1,1=RH_c^{1,1}=\mathbb R. For the planar cross, m=4m=4 and the span has dimension two, so Hc1,1=R2H_c^{1,1}=\mathbb R^2. Both underlying spaces are a four-valent topological star when m=4m=4, but their superform coefficient sheaves differ.

This calculation is the local analogue of the harmonic-coordinate issue on analytic curves. Recording only an embedded graph may impose extra relations among its directions. Harmonic tropicalizations use the geometric harmonic functions needed for the intrinsic theory, rather than assuming one fixed low-dimensional embedding records every differential direction. The analytic comparison still requires the model and refinement theorems of GJR; the star calculation explains what those theorems must preserve.

References. For the flag description, [AK]; for the polyhedral Poincaré lemma and matroidal duality, [JSS] (Theorems 3.16 and 4.33). The rank-two balancing and star quotient calculations above are written out in the conventions of this book.

24.11 Hodge and Aeppli calculations on a real torus

A compact real torus provides a complete model in which the two differentials, their cohomology, and the d′d′′d'd'' equation can be computed directly. This model is useful for separating a valid differential statement from a proposed passage through tropical limits. The proof below uses Fourier series on one fixed torus. It does not require a projective embedding, and it does not assert a Hodge theorem for every tropical space.

Let M=V/ΛM=V/\Lambda be a real torus of dimension nn, with a fixed translation-invariant Euclidean inner product. Complexify the superform spaces during the computation and recover real forms by conjugation. The dual lattice is Λ∗\Lambda^*. Every smooth form has a Fourier expansion α(x)=∑k∈Λ∗e2πi⟨k,x⟩αk,αk∈⋀pVC∗⊗⋀qVC∗.\alpha(x)=\sum_{k\in\Lambda^*}e^{2\pi i\langle k,x\rangle}\alpha_k, \qquad \alpha_k\in\bigwedge^p V^*_{\mathbb C}\otimes\bigwedge^q V^*_{\mathbb C}. Its coefficients decay faster than every negative power of ∣k∣|k|. Conversely, such decay defines a smooth form, with termwise differentiation. These standard Fourier facts follow by repeated integration by parts on a fundamental domain and the absolute convergence of sufficiently high powers of (1+∣k∣)−1(1+|k|)^{-1}.

Contraction on each nonzero mode

For k≠0k\ne 0, write k′=∑kid′xik'=\sum k_i d'x_i and k′′=∑kid′′xik''=\sum k_i d''x_i in orthonormal coordinates. On that Fourier mode, the differential operators are exterior multiplication by 2πik′2\pi i k' and 2πik′′2\pi i k'', with the total-degree wedge convention. Let ιk′\iota'_k and ιk′′\iota''_k be contraction with the corresponding metric dual vectors in the first and second exterior factors. Define hk′=ιk′2πi∣k∣2,hk′′=ιk′′2πi∣k∣2.h'_k=\frac{\iota'_k}{2\pi i|k|^2},\qquad h''_k=\frac{\iota''_k}{2\pi i|k|^2}. The second contraction includes the total-exterior-algebra sign. For a primed pp-form a′a' and a double-primed form b′′b'', our convention is ιk′′(a′∧b′′)=(−1)pa′∧ιkb′′.\iota''_k(a'\wedge b'')=(-1)^p a'\wedge\iota_k b''. Without this sign, the two cross-factor anticommutation identities below would not hold.

Lemma 24.11.1. On every nonzero mode, d′hk′+hk′d′=1,d′′hk′′+hk′′d′′=1,d′hk′′+hk′′d′=0,d′′hk′+hk′d′′=0.\begin{gather*} d'h'_k+h'_kd'=1,\qquad d''h''_k+h''_kd''=1,\\ d'h''_k+h''_kd'=0,\qquad d''h'_k+h'_kd''=0. \end{gather*} The norms of hk′,hk′′h'_k,h''_k are bounded by a constant times ∣k∣−1|k|^{-1}. They therefore act on smooth Fourier series, and also on distributions.

Proof. For a covector λ\lambda and vector vv, exterior multiplication ελ\varepsilon_\lambda and contraction satisfy ελιv+ιvελ=λ(v)\varepsilon_\lambda\iota_v+\iota_v\varepsilon_\lambda=\lambda(v). Applying this in the matching factor gives ∣k∣2|k|^2; applying it across the two disjoint factors gives zero. Division by 2πi∣k∣22\pi i|k|^2 proves the identities. The contraction norm is ∣k∣|k|, proving the estimate. Smooth coefficients are rapidly decreasing, while distribution coefficients have at most polynomial growth, so these estimates preserve the corresponding Fourier classes. ◻

Let P0P_0 denote the projection to the zero Fourier mode, equivalently the coefficientwise average. Set h′=h′′=0h'=h''=0 on that mode and use the preceding formulas on the others. The four identities then hold globally with 11 replaced by 1−P01-P_0 in the first two. Thus the only obstruction visible in these complexes is the constant coefficient.

Theorem 24.11.2. (Dolbeault cohomology of a real torus). Every d′′d''-closed smooth (p,q)(p,q) form is cohomologous to its unique constant-coefficient average. In particular, Hd′′p,q(M)≅⋀pV∗⊗⋀qV∗,dim⁡Hd′′p,q(M)=(np)(nq).H^{p,q}_{d''}(M)\cong \bigwedge^p V^*\otimes\bigwedge^q V^*, \qquad \dim H^{p,q}_{d''}(M)=\binom np\binom nq. The same conclusion holds with distributional forms, and the inclusion of smooth forms induces the indicated cohomology isomorphism.

Proof. If d′′α=0d''\alpha=0, the homotopy identity gives α−P0α=d′′h′′α\alpha-P_0\alpha=d''h''\alpha. The primitive is smooth, or distributional, in the same category as α\alpha. Conversely, the average of a derivative is zero because its zero Fourier mode vanishes. Hence no nonzero constant form is exact. The construction commutes with conjugation of Fourier series, so it restricts to real forms. ◻

The d′d′′d'd'' lemma

Theorem 24.11.3. A smooth form α\alpha with d′α=d′′α=0d'\alpha=d''\alpha=0 has a smooth d′d′′d'd'' primitive if and only if P0α=0P_0\alpha=0. One primitive is β=h′′h′α,d′d′′β=α.\beta=h''h'\alpha,\qquad d'd''\beta=\alpha. The statement also holds for distributions. In particular, a form closed under both differentials which is exact under either one is d′d′′d'd''-exact.

Proof. Necessity follows because a derivative has average zero. For sufficiency, work on a nonzero mode. Since d′′h′α=−h′d′′α=0d''h'\alpha=-h'd''\alpha=0, d′d′′h′′h′α=d′h′α=α.d'd''h''h'\alpha=d'h'\alpha=\alpha. Summing is legitimate by the contraction estimates; the multiplier is of order at most ∣k∣−2|k|^{-2}. If the bidegree has a zero entry, the corresponding negative-degree primitive space is zero, and the same identities force every mean-zero form closed under both operators in that bidegree to vanish. Finally, exactness under either differential implies zero average. ◻

The two contractions and their anticommutation signs are essential. Solving separately for a d′d' primitive and a d′′d'' primitive does not by itself provide one common d′d′′d'd'' primitive on an arbitrary double complex. Here the explicit identities make the implication valid.

Definition 24.11.4. For the superform double complex, the Bott–Chern and Aeppli groups are HBCp,q=ker⁡d′∩ker⁡d′′⊂Ap,qd′d′′Ap−1,q−1,HAp,q=ker⁡(d′d′′:Ap,q→Ap+1,q+1)d′Ap−1,q+d′′Ap,q−1.\begin{align*} H^{p,q}_{BC}&=\frac{\ker d'\cap\ker d''\subset A^{p,q}} {d'd''A^{p-1,q-1}},\\ H^{p,q}_{A}&=\frac{\ker(d'd'':A^{p,q}\to A^{p+1,q+1})} {d'A^{p-1,q}+d''A^{p,q-1}}. \end{align*}

Proposition 24.11.5. On MM, both groups are represented uniquely by constant forms, so they have the same dimensions as Hd′′p,q(M)H^{p,q}_{d''}(M). The assertion holds for smooth forms and for distributions.

Proof. The Bott–Chern assertion is Theorem 24.11.3. For Aeppli cohomology, take a nonzero mode of α\alpha with d′d′′α=0d'd''\alpha=0. Then d′αd'\alpha is d′′d''-closed, and α=d′h′α+h′d′α=d′h′α+h′d′′h′′d′α=d′h′α−d′′h′h′′d′α.\begin{align*} \alpha&=d'h'\alpha+h'd'\alpha\\ &=d'h'\alpha+h'd''h''d'\alpha\\ &=d'h'\alpha-d''h'h''d'\alpha. \end{align*} This is a sum of an exact term for each differential. The multipliers have polynomially bounded size, so the formulas sum in the desired category. A sum of derivatives has zero average, proving uniqueness of the constant representative. ◻

Pairings and current equations

Integration gives a perfect pairing Hd′′p,q(M)×Hd′′n−p,n−q(M)⟶R,([α],[β])⟼∫Mα∧β.H^{p,q}_{d''}(M)\times H^{n-p,n-q}_{d''}(M)\longrightarrow\mathbb R, \qquad ([\alpha],[\beta])\longmapsto\int_M\alpha\wedge\beta. It is well defined by Stokes. To prove nondegeneracy, replace both classes by constant forms. A basis element d′xI∧d′′xJd'x_I\wedge d''x_J pairs, up to its fixed sign and the positive volume of the torus, with the complementary basis element indexed by Ic,JcI^c,J^c. The pairing matrix is invertible. The same argument identifies the Bott–Chern/Aeppli complementary pairing.

This calculation proves more than finite-dimensionality: it gives explicit continuous solution operators for all nonzero Fourier modes. Hence the obstruction to a closed current being d′d′′d'd''-exact is exactly its constant cohomology component. By contrast, on a proposed direct limit of tropical complexes, finite-stage nondegenerate pairings alone provide neither a limiting solution estimate nor an actual form representative. The Fourier operator has supplied both an algebraic homotopy and a topological bound in this particular model.

24.12 Positive currents and their global torus potentials

Theorem 24.12.1. Let TT be an ordinary positive closed (1,1)(1,1) supercurrent on MM. Its coefficientwise average is a constant positive semidefinite form θ\theta. There is a continuous real function uu, unique up to a constant, such that T=θ+d′d′′[u].T=\theta+d'd''[u]. On VV, the periodic lift of uu added to a quadratic potential for θ\theta is finite convex.

Proof. The average of the positive matrix-valued coefficient measure is positive semidefinite, giving θ\theta. The distributional d′d′′d'd'' lemma produces a real distribution uu with d′d′′u=T−θd'd''u=T-\theta. Locally write θ=d′d′′q\theta=d'd''q for a quadratic function. Then d′d′′(u+q)=T≥0d'd''(u+q)=T\geq 0. The real convex-distribution theorem, Corollary 11.5.2, makes u+qu+q a finite convex continuous function locally. Thus uu has a continuous representative, and these representatives agree on overlaps because they define the same distribution.

On VV, the lifted function u+qu+q has positive distributional Hessian globally, hence is convex. If two potentials give the same current, their difference has zero Hessian and is affine on VV. A Λ\Lambda-periodic affine function is constant. ◻

This conclusion is a feature of real convexity. Positive closed complex (1,1)(1,1) currents can have logarithmically singular potentials; a positive real Hessian distribution on an open Euclidean domain instead has a finite convex representative. The real torus theorem must not be transferred to a singular tropical quotient or an arbitrary analytic space merely because the notation for the two differential operators is similar.

Corollary 24.12.2. (Rigidity in a null direction). Let HH be the positive semidefinite matrix of θ\theta and uu the preceding potential. The lift of uu is constant along every vector in ker⁡H\ker H. If the image of ker⁡H\ker H is dense in MM, then uu is constant and T=θT=\theta.

Proof. For e∈ker⁡He\in\ker H, the quadratic function q(x)=12xtHxq(x)=\frac 12x^{\mathsf t}Hx is constant along x+Rex+\mathbb Re. The convex function q+uq+u restricts to a convex function of one real variable which is bounded, because uu is periodic and continuous. A convex function bounded on all of R\mathbb R is constant: any nonzero secant slope would force unbounded growth on one end. Thus u(x+te)=u(x)u(x+te)=u(x). Invariance under a dense subgroup and continuity give the final claim. ◻

Example 24.12.3. On R2/Z2\mathbb R^2/\mathbb Z^2, let H=(1aaa2)H=\left(\begin{smallmatrix}1&a\\a&a^2\end{smallmatrix}\right) with aa irrational. Its kernel is generated by (−a,1)(-a,1), whose image is dense in the torus. Every positive closed (1,1)(1,1) current with average HH is therefore the constant current HH. If a=0a=0, nonconstant potentials depending only on x1x_1 are possible, subject to 1+u′′≥01+u''\geq 0. The closure of the null foliation, rather than the rank of HH alone, accounts for the difference.

References. The arguments in this section combine the elementary Fourier contraction of a flat torus with the real convex-potential theorem of [Lag12] (Proposition 1.13). They are included as explicit model calculations. For the broader comparison of superform cohomology with tropical cohomology, see [JSS]; no passage from the flat-torus calculation to a general Berkovich Hodge theorem is used here.

24.13 The Hodge index calculation in bidegree (1,1)(1,1)

The flat-torus computation also gives an explicit index theorem. The statement concerns symmetric real (1,1)(1,1) classes. Their constant representatives are symmetric matrices, and the primitive condition is a trace condition. This makes the sign and the equality case completely visible.

Theorem 24.13.1. Assume n≥2n\geq 2, and let θ\theta be a constant positive definite (1,1)(1,1) form on MM. On the space of symmetric Bott–Chern (1,1)(1,1) classes, the pairing Q([α],[β])=∫Mα∧β∧θn−2Q([\alpha],[\beta])=\int_M\alpha\wedge\beta\wedge\theta^{n-2} has signature (1,n(n+1)2−1)(1,\frac{n(n+1)}2-1). Its restriction to the primitive subspace {[α]:∫Mα∧θn−1=0}\left\{[\alpha]:\int_M\alpha\wedge\theta^{n-1}=0\right\} is negative definite. In particular, a smooth symmetric closed form in that primitive subspace has Q([α],[α])=0Q([\alpha],[\alpha])=0 if and only if it is d′d′′d'd''-exact.

Proof. Stokes makes the pairing invariant under adding a d′d′′d'd''-exact form to either factor. By Proposition 24.11.5, replace all classes by constant representatives. Make a real linear coordinate change in which the matrix of θ\theta is the identity; the lattice changes but the torus volume stays positive. Let AA be the symmetric matrix of α\alpha. Expanding det⁡(I+tA)\det(I+tA), or diagonalizing AA, gives α∧θn−1=tr⁡An θn,α2∧θn−2=(tr⁡A)2−tr⁡(A2)n(n−1) θn.\begin{align*} \alpha\wedge\theta^{n-1}&=\frac{\operatorname{tr}A}{n}\,\theta^n,\\ \alpha^2\wedge\theta^{n-2} &=\frac{(\operatorname{tr}A)^2-\operatorname{tr}(A^2)}{n(n-1)}\,\theta^n. \end{align*} Primitivity is precisely tr⁡A=0\operatorname{tr}A=0. On this subspace the second expression is a negative multiple of tr⁡(A2)=∑iλi2\operatorname{tr}(A^2)=\sum_i\lambda_i^2, and vanishes only when A=0A=0. The line spanned by the identity is positive and orthogonal to the trace-zero subspace. Since the dimension of the symmetric matrices is n(n+1)/2n(n+1)/2, this gives the signature. A zero constant representative is equivalent to d′d′′d'd''-exactness by Theorem 24.11.3. ◻

For a general positive matrix HH of θ\theta, the same formulas use H−1/2AH−1/2H^{-1/2}AH^{-1/2}. Thus the invariant primitive condition is tr⁡(H−1A)=0\operatorname{tr}(H^{-1}A)=0. The coordinate change is only a device for the calculation; it does not replace the lattice normalization in integration.

Let u,vu,v be continuous periodic functions with αu=θ+DDu≥0\alpha_u=\theta+DDu\geq 0 and αv=θ+DDv≥0\alpha_v=\theta+DDv\geq 0. The global regularization of Section 21.23 defines their continuous mixed products. Set Iθ(u,v)=−∫M(u−v)(αun−αvn).I_\theta(u,v)=-\int_M(u-v)(\alpha_u^n-\alpha_v^n).

Proposition 24.13.2. One has Iθ(u,v)≥0I_\theta(u,v)\geq 0. For smooth u,vu,v, it is the sum of nonnegative Dirichlet terms Iθ(u,v)=∑j=0n−1∫Md′(u−v)∧d′′(u−v)∧αuj∧αvn−1−j.I_\theta(u,v)=\sum_{j=0}^{n-1}\int_M d'(u-v)\wedge d''(u-v)\wedge\alpha_u^j\wedge\alpha_v^{n-1-j}.

Proof. In the smooth case, factor the difference of powers and integrate by parts against u−vu-v. Every remaining factor is a closed positive (1,1)(1,1) form, so each integrand is nonnegative. For continuous potentials use common global smooth semipositive approximations. The measures converge weakly, the potentials converge uniformly, and their total masses remain fixed. The moving-test convergence of Chapter 9 therefore passes the displayed nonnegativity to the limit. The integral defining IθI_\theta is meaningful without assigning ordinary first derivatives to the continuous potentials. ◻

The index theorem is a nondegeneracy statement on a finite-dimensional cohomology space. The inequality for IθI_\theta is a positivity statement on potentials in one fixed class. They use related integration-by-parts signs, but their equality cases require different arguments. In the smooth strictly positive setting, Section 21.25 proves that Iθ(u,v)=0I_\theta(u,v)=0 forces u−vu-v constant. The full degenerate continuous uniqueness problem belongs to the corresponding convex Monge–Ampère theorem.

References. These are matrix and integration-by-parts consequences of the flat-torus calculations above. Their role in the torus metric setting is compared with [BGJK25] (Sections 7–8); no general tropical Hodge–Riemann theorem is assumed.

24.14 Subject and literature concordance

The following concordance groups the sources by the mathematical construction being used. It supplements the notation index in Section 0.2.3. A source listed in one row may also contribute to another; the grouping does not assert equivalence of its objects with every object in that row.

Construction Sources Comparison developed here
Tropical cycles and skeleta [Gub13] [GRW16] [GRW17] [Pay09] [DT26] Multiplicities, lattice degrees, faithful tropicalization, and the topology of analytic spaces; Sections 5.2, 6.3, 24.1.
Smooth forms and currents [CLD12] [Gubler] [CLD25] Analytic-unit charts, algebraic presentations, test topology, and the revised tropical framework; Chapters 7–9.
Positive cones and current estimates [CLD12] [CLD25] [BGJK] Dual positivity, order zero, and the distinction between ordinary and weak positivity; Section 9.1.1 and Appendix A.
Delta-forms and intersections [GK17] [GK19] [Mih24] [Mih23] [Cai23] [BGK25] Corner terms, products, analytic constructions, and compactification; Sections 16.1, 15.7, 22.2.
Harmonicity and PL positivity [GJR] [GJRc] [GRPL] [GRPSH] Enlarged coordinates and test spaces, model semipositivity, and curvewise criteria; Sections 17.2–19.2.4.
Measures, metrics, and heights [CL06] [GM] [BFJ15] [Gub10] [GS23] [Lin26] [CG26] Model measures, global Monge–Ampère theory, canonical measures, and local heights; Sections 19.4, 20.6, 22.2, 21.3.

24.15 Cycle classes: what cohomological vanishing gives

A construction of Green forms must first identify the obstruction represented by the cycle. Liu’s tropical cycle class map is directly relevant because its target is built from original CLD forms, rather than from a separately chosen finite polyhedral model. Let XX be smooth and separated of finite type over a complete nontrivially valued field. Liu constructs a rational sheaf Tp\mathscr T^p whose realification is the sheaf of d′′d''-closed (p,0)(p,0)-forms. The fine resolution by CLD forms relates Htropp,q(X)=Hq(Xan,Tp),Htropp,q(X)⊗QR≃Hd′′p,q(Xan).H^{p,q}_{\mathrm{trop}}(X)=H^q(X^{\mathrm{an}},\mathscr T^p), \qquad H^{p,q}_{\mathrm{trop}}(X)\otimes_{\mathbb Q}\mathbb R \simeq H^{p,q}_{d''}(X^{\mathrm{an}}). His cycle map cltrop:CHp(X)→Htropp,p(X)\mathrm{cl}_{\mathrm{trop}}:\mathrm{CH}^p(X)\to H^{p,p}_{\mathrm{trop}}(X) is functorial and product-compatible. Its integration formula, after realification, is ∫Xancltrop(Z)∧η=∫Zanη,d′′η=0,quadη∈Acn−p,n−p(Xan).\begin{equation} \int_{X^{\mathrm{an}}}\mathrm{cl}_{\mathrm{trop}}(Z)\wedge\eta =\int_{Z^{\mathrm{an}}}\eta, \qquad d''\eta=0,quad\eta\in A_c^{n-p,n-p}(X^{\mathrm{an}}). \end{equation} The left side means the pairing of a representative of the class with the closed test. Independence of that representative follows from Stokes. The theorem proves that this pairing agrees with cycle integration.

Theorem 24.15.1. (Liu’s de Rham vanishing consequence). Let kk be a finite extension of Qp\mathbb Q_p, let X/kX/k be proper and smooth, and let ZZ have codimension pp with cldR(Z)=0\mathrm{cl}_{\mathrm{dR}}(Z)=0. After base change to Cp\mathbf C_p, integration over ZanZ^{\mathrm{an}} vanishes on every d′′d''-closed original CLD test of complementary bidegree.

Open proof obligation (RC2). Establish the pairing-vanishing result in the stated scope; Liu20 Theorem 1.8 is supplied.

NAG-900490 · Proof status

24.16 Finite tropical models and Hodge theory

The fan theorems of Amini–Piquerez provide powerful finite-dimensional models, with hypotheses that are visible combinatorially. For a rational simplicial fan Σ\Sigma, let Σ‾\overline\Sigma be its canonical compactification. The tropical Feichtner–Yuzvinsky theorem identifies Hp,p(Σ‾,Q)≃Ap(Σ,Q).H^{p,p}(\overline\Sigma,\mathbb Q)\simeq A^p(\Sigma,\mathbb Q). For the full cohomology ring to be concentrated on the diagonal and equal to the Chow ring, their Theorem 1.3 additionally assumes that Σ\Sigma is a tropical homology manifold. Integral coefficients require further unimodularity, saturation and integral-duality hypotheses. Rational and integral statements should be recorded separately. Homological smoothness is itself a local condition. In [AP-D] (Theorem 1.1), a fan is homologically smooth precisely when the fan and all its star fans satisfy tropical Poincaré duality. Their local-to-global theorem gives duality between ordinary tropical cohomology and Borel–Moore homology. Compactly supported cohomology enters the associated pairing. Replacing Borel–Moore homology by ordinary homology on a noncompact space would change that theorem.

Example 24.16.1. (The complete one-dimensional fan). For the fan in R\mathbb R with rays generated by 11 and −1-1, the Stanley–Reisner relation is x+x−=0x_+x_-=0 and the linear relation is x+−x−=0x_+-x_-=0. Hence A∙(Σ,Q)=Q[h]/(h2),deg⁡h=1.A^\bullet(\Sigma,\mathbb Q)=\mathbb Q[h]/(h^2),\qquad\deg h=1. The compactification has H0,0=H1,1=QH^{0,0}=H^{1,1}=\mathbb Q and no other cohomology. Multiplication pairs 11 and hh nondegenerately. This is a complete finite calculation, using the stated fan comparison. It proves neither a general d′d′′d'd''-lemma on an analytic curve nor an identification of all forms on its Berkovich analytification with forms on one segment.

The remaining transition to Berkovich geometry has three parts. One must identify the relevant complexes and their transition maps, prove that cohomology and the needed support conditions pass through the chosen limit, and control the topology needed for duality or solution operators. Payne’s topological inverse-limit theorem addresses none of these three analytic assertions by itself. The limit and closed-range issues in Sections 24.5 and 24.6 remain in force even when every individual tropical model has perfect finite-dimensional duality.

References. [Liu20] (Theorems 1.1, 1.7–1.9); [AP-FY] (Theorems 1.1–1.3); [AP-D] (Theorems 1.1–1.3).

From local fans to global tropical Hodge classes

There are two further layers in the Amini–Piquerez programme. The local layer establishes the Kähler package for suitable Chow rings: for example, effective unimodular quasi-projective quasilinear fans are Chow-Kähler in [AP-F] (Theorem 1.10). The global layer constructs Kähler classes and uses a tropical Steenbrink complex to obtain Hard Lefschetz, Hodge–Riemann and weight-monodromy results for smooth projective tropical varieties, with rational coefficients and the definitions of [AP-H] (Theorem 1.4). Neither assertion is a statement about every balanced polyhedral complex. The tropical Clemens–Schmid sequence addresses the representation of cohomology by tropical cycles. Under the additional rational triangulability hypothesis, [AP-CS] (Theorem 1.1) identifies the classes generated by codimension-pp tropical cycles on a smooth projective tropical variety with ker⁡(N:Htropp,p(X,Q)⟶Htropp−1,p+1(X,Q)).\ker\bigl(N:H^{p,p}_{\mathrm{trop}}(X,\mathbb Q) \longrightarrow H^{p-1,p+1}_{\mathrm{trop}}(X,\mathbb Q)\bigr). The monodromy-kernel condition is part of the conclusion; one should not replace it by all of Hp,pH^{p,p}. These cycles are tropical cycles. An algebraic lifting statement would require an additional theorem. Likewise, a tropical cycle class and Liu’s cycle class of an algebraic cycle become comparable only after the relevant realization map has been constructed. The full proofs of the Steenbrink and Clemens–Schmid constructions are not reproduced here. The finite fan calculation, the cycle-pairing theorem and the limit requirements isolate the parts needed for the analytic programme. Extending this account to a self-contained global tropical Hodge theory would require a separate development of those complexes and their comparison maps.

24.17 The next mathematical comparisons

The preceding additions change the form of the remaining questions. Some unrestricted claims already have counterexamples; others have precise positive theorems in narrower categories. The following programme records the assertions still needed for the general analytic goals of this book. An external theorem with an adequate statement is not a missing proof merely because its proof is not repeated here.

  1. Approximation and residual geometry. For CLD25 Proposition 8.1.11, prove the predecessor-replacement step or restrict the conclusion to the proved first stage. The pure-dimensional residual current criterion is reconstructed in Theorem 19.18.12. The separate integral PL bridge is now supplied by Theorem 19.18.22 and Corollary 19.18.23. These tasks do not by themselves prove smooth regularization for every continuous current-psh metric.

  2. Larger potential classes and kernels. Theorem 13.6.3 already permits the stated arbitrary ordinary positive closed background and admitted continuous potentials. Extending the potential class requires new approximation or monotone-convergence results. Passing to curvature classes additionally requires a kernel theorem. The unrestricted abstract kernel statement is false; an analytic replacement must specify functions, backgrounds and actual relations.

  3. Capacity and global envelopes. Compare the toric, polyhedral, model-metric and synthetic finite-energy theories on their common domains. A general local analytic package of capacity, quasi-continuity, comparison and Dirichlet existence has not been established here. The counterexample in Section 16.7 excludes a universal orthogonality shortcut. The model and field hypotheses of each positive existence theorem remain essential.

  4. Representatives and local potentials. Polyhedral pullback and current pushforward do not supply a general analytic trace in the actual delta-sheaf. Prove that trace with its domain and support conditions before using it for descent. Likewise, existence of a distributional primitive is insufficient for a local psh potential theorem for arbitrary positive closed (1,1)(1,1)-currents.

  5. Cohomology and actual Green forms. Combine a precisely stated finite-stage Hodge comparison with control of limits, closed ranges and support. Liu’s vanishing theorem supplies a necessary pairing condition, not an actual primitive. The general strong Green problem still requires a representative on the exact complement and the identity on the full actual test space, with compatible finite-field descent.

  6. Energy and arithmetic compatibility. On an explicitly identified common domain, compare limiting measures and relative energies with CLD/GK currents and arithmetic intersection. Retain the boundary topology and the compactified reference object. In mixed characteristic, list resolution or model hypotheses theorem by theorem; do not replace a conditional global theorem by an unconditional local statement.

The next editorial expansion should develop the global tropical Hodge complexes and the mixed finite-energy comparison in more detail. The next proof repair should address continuous predecessor replacement after the integral PL bridge completed below. These are separate tasks: expanding a literature account is not evidence that an unresolved analytic implication has been proved.

Version 0.9 supplies the Laurent gluing, simultaneous residual realization and PL noncharging argument over arbitrary complete fields, then assembles the pure-dimensional residual criterion. It also corrects the previous manuscript’s non-good-germ concern: boundarylessness already implies goodness here. The integral psh comparison requires the additional argument below; it cannot be replaced by merely expressing a PL weight as a difference of admitted functions.

Version 0.11 supplies prescribed graded premodel realization, the simultaneous bundle identifications, and the nef integral PL implication on a fixed tube. P0-A4 is therefore complete in the stated integral PL category, with its named imported inputs. The next approximation task is P0-B: continuous predecessor replacement in CLD25 paragraph 8.1.11. A generated-class conclusion for a nef PL weight is not a first-stage smooth sequence for that weight.

Version 0.12 proves continuous endpoint collapse for a decreasing hierarchy on one fixed compact set. The remaining P0-B task is to justify a common compact representation, or to construct weaker continuous replacements in C25\mathcal C_{25} on one fixed neighbourhood. These are different targets: the first would give first-stage approximation, while the second is enough for the unindexed equality. The same-ordinal forward inclusion and the locally integrable hierarchy equality are already established in Section 11.6.3; they should not be listed as unresolved.

Version 0.13 adds a projective envelope route. Theorem 11.7.5 proves that bounded first-stage decreasing stability suffices for equality at every continuous ordinal stage; after that input is supplied, ordinal control is no longer a separate obstacle. The next source-verification task is the chain in [Wu26]: Theorem A.3, bounded localization, Theorem 4.17 and Corollary 4.18. Proposition 11.7.3 corrects the barrier normalization and background conversion, and Lemma 11.7.1 supplies sequential continuous obstacles from the local approximation property. Section 11.8 now supplies a fixed-domain diagonal lemma, bounded-family tropical localization and projective barrier construction. Section 11.9 supplies one such common-domain direction for explicitly specified rational PL generators over a nontrivially valued field. The approximation and single-generator tasks are now treated in Section 11.22 under its stated hypotheses. The common-domain first-stage comparison is now supplied by Corollary 11.24.4; remaining tasks are sheaf-level decreasing closure, complete source-class identification and trivial-valuation extension. Outside the projective envelope category, bounded decreasing stability remains the analytic target; no claim over arbitrary complete fields follows from the equicharacteristic-zero envelope theorem.

For an already strict R-PL germ, Section 11.13 now supplies coefficient-compatible decreasing rational approximation on one residue tube, and a projective original-smooth comparison. An arbitrary sequence still needs a common finite presentation for the ordered-family corollary. Identification of arbitrary pre-existing relative real PL presentations remains separate; continuous relative-support approximation is supplied by Section 11.22. Section 11.14 repairs the single-polarization cofinal input after exhibiting the fixed-model failure of GM Lemma 4.12. Finite simultaneous polarizations and the numerical and bounded-correction parts of GM Proposition 4.14 are reconstructed in the stated formal-domination scope. Section 11.15 supplies algebraic model-morphism domination using non-Noetherian formal GAGA, and reduced-special-fibre cofinality under the algebraically closed hypothesis. These particular interfaces are therefore closed; the broader analytic comparisons remain open. Section 11.17 narrows the remaining reverse-comparison problem. The explicit smooth family obstructing ambient convex extension does admit decreasing coefficient-compatible relative PL approximation in its original chart. More generally, finite coordinatewise-monotone convex compositions of bounded relative rational PL functions have such approximations, including the decreasing-tuple construction (11.17.9). Section 11.22 now supplies the required density directly for continuous relative-support functions on a fixed compact image. Proposition 11.22.14 applies it to one original smooth psh input using a boundary-adapted chart. Theorem 11.24.3 now supplies common-domain replacement for arbitrary original smooth sequences whose charts vary; the further sheaf-level closure is separate. The earlier fixed rational-fan result is extended to one-scale polytopal complexes by Theorem 11.18.1. To apply this result to real PL output from another source, one must verify the vertex condition or prove a replacement allowing the subdivision to vary. Examples 11.18.3 and 11.18.4 rule out freezing arbitrary subdivisions while merely rounding their coefficients.

Section 11.19 now gives the exact closure on a fixed complex: a relatively supported target must lie in the real span of admissible rational functions. Under the strict face-support certificate this reduces to the explicit continuity equations. Edge-connected rational simplices of dimension at least two provide a separate route to one common value-group scale. For this fixed-complex approach, the hypotheses must still be verified on the source output. Section 11.22 instead constructs new affine branches before gluing and so avoids fixing those subdivisions.

Lemma 11.19.12 and Proposition 11.19.13 complete the coefficient-compatible convex extension on individual cells. Their value estimates do not bound normal slopes at the original faces. Section 11.20 now proves the slope-controlled version and preserves relative supports for compatible compact complexes. Section 11.22 handles continuous relative-support inputs with arbitrary real affine supports on one compact domain, including extended images under its admissible boundary convention. The global twisted and simplicial source formulations still require separate identification.

For the current approximation status, Section 11.22 supersedes the earlier density tasks listed above. Continuous relative-support inputs on one compact ordinary or extended image admit coefficient-compatible decreasing approximation; the extended case retains its admissible boundary-support hypothesis. A single original smooth psh input, and a finite collection of such inputs, are covered by Section 11.22.6. Section 11.24 completes the arbitrary original-smooth-family replacement and the stated first-stage germ comparison. The open tasks are further sheaf-level decreasing closure, continuous predecessor replacement, and precise source-class identification. None of the abstract boundary or fixed-chart counterexamples negates those open analytic assertions.