16 · Analytic delta-forms and their positivity

13 NAG Tags

An analytic domain, a closed analytic subspace and a scalar extension are different constructions; Chapter 2 supplies each of them. The quotient defining a differential object must respect precisely the maps admitted by its theory. In the original GK construction below, the universal quantifier ranges over algebraic morphisms and compatible charts. The later Mihatsch comparison uses analytic morphisms under its own hypotheses. The existence of the underlying analytic map alone does not prove that substitution respects a presentation quotient.

16.1 Delta-forms and delta-currents

Throughout this section, KK is algebraically closed, complete, and nontrivially valued. Let XX be an integral separated KK-scheme of finite type, let n=dim⁡Xn=\dim X, and let W⊂XanW\subset X^{\mathrm{an}} be open. These are the hypotheses of the algebraic tropical-chart construction in [GK17] (§4) and of the positivity results in [GK19] (§0.1). No hypothesis on the residue characteristic is imposed.

From tropical preforms to forms on the analytic space

Let NN be a lattice of rank rr, put NR=N⊗ZRN_{\mathbf R}=N\otimes_{\mathbf Z} \mathbf R, and let Ω~⊂NR\widetilde\Omega\subset N_{\mathbf R} be open. A tropical cycle with smooth weights is represented by a finite integral real-affine polyhedral complex together with smooth weight functions mσm_\sigma on its maximal faces. For each codimension-one face τ\tau and x∈τx\in\tau, balancing means ∑σ⊃τmσ(x)νσ,τ∈Nτ,R,\sum_{\sigma\supset\tau}m_\sigma(x)\nu_{\sigma,\tau} \in N_{\tau,\mathbf R}, where νσ,τ\nu_{\sigma,\tau} is a lift of the primitive outgoing normal and Nτ,RN_{\tau,\mathbf R} is the tangent space of τ\tau. If CC has codimension ℓ\ell, its integration current δC\delta_C has bidegree (ℓ,ℓ)(\ell,\ell). The weights and coefficients in the following definition are real.

The tropical preforms were defined in Chapter 15.

A tropical chart of XanX^{\mathrm{an}} consists of a very affine open subvariety U⊂XU\subset X, its canonical torus embedding φU:U↪TU\varphi_U:U\hookrightarrow T_U, and V=trop⁡U−1(Ω),Ω=Ω~∩Trop⁡(U).V=\operatorname{trop}_U^{-1}(\Omega),\qquad \Omega=\widetilde\Omega\cap\operatorname{Trop}(U). Write NUN_U for the cocharacter lattice of TUT_U. If f:X′→Xf:X'\to X is an algebraic morphism and (V′,φU′)(V',\varphi_{U'}) is a chart with f(U′)⊂Uf(U')\subset U and fan(V′)⊂Vf^{\mathrm{an}}(V')\subset V, the torus maps induce an integral affine map F:NU′,R⟶NU,R.F:N_{U',\mathbf R}\longrightarrow N_{U,\mathbf R}. An ambient preform α~\widetilde\alpha can be pulled back by FF and then restricted to the tropical cycle Trop⁡(U′)\operatorname{Trop}(U'). Here restriction means the product with δTrop⁡(U′)\delta_{\operatorname{Trop}(U')}.

Definition 16.1.1. (The functorial chart quotient). Let Np,q(V,φU)\mathcal N^{p,q}(V,\varphi_U) be the subspace of ambient preforms α~∈Pp,q(Ω~)\widetilde\alpha\in P^{p,q}(\widetilde\Omega) such that (F∗α~∧δTrop⁡(U′))∣trop⁡U′(V′)=0\left. \bigl(F^*\widetilde\alpha\wedge \delta_{\operatorname{Trop}(U')}\bigr) \right|_{\operatorname{trop}_{U'}(V')}=0 for every algebraic morphism f:X′→Xf:X'\to X and every compatible chart as above. Set Pp,q(V,φU)=Pp,q(Ω~)/Np,q(V,φU).P^{p,q}(V,\varphi_U) = P^{p,q}(\widetilde\Omega)/\mathcal N^{p,q}(V,\varphi_U).

This quotient is independent of the choice of Ω~\widetilde\Omega. It has pullbacks under compatible algebraic morphisms [GK17] (Definition 4.4 and Remark 4.5). The quantifier over all morphisms is part of the definition. It cannot be replaced by the requirement that the restriction to the one cycle Trop⁡(U)\operatorname{Trop}(U) vanish. Indeed, [GK17] (Example 4.22) constructs a nonzero chart class whose restriction to that cycle is zero but whose pullback to a compatible chart is nonzero.

Definition 16.1.2. (Closed generators and delta-forms). An element ω∈Pp,q(V,φU)\omega\in P^{p,q}(V,\varphi_U) belongs to Zp,q(V,φU)Z^{p,q}(V,\varphi_U) if, after every compatible pullback and restriction as in Definition 16.1.1, the resulting supercurrent is both d′d'-closed and d′′d''-closed. Let AZ(V,φU)AZ(V,\varphi_U) be the algebra generated by A∙,∙(Ω)A^{\bullet,\bullet}(\Omega) and Z∙,∙(V,φU)Z^{\bullet,\bullet}(V,\varphi_U). Thus locally an element has the form β=∑j=1sαj∧ωj,αj∈A∙,∙(Ω),ωj∈Z∙,∙(V,φU).\beta=\sum_{j=1}^s\alpha_j\wedge\omega_j, \qquad \alpha_j\in A^{\bullet,\bullet}(\Omega), \quad \omega_j\in Z^{\bullet,\bullet}(V,\varphi_U). Sheafifying the chart algebras PP gives the sheaf of generalized delta-forms; sheafifying AZAZ gives the sheaf B∙,∙B^{\bullet,\bullet} of delta-forms.

The universally closed condition makes d′β=∑jd′αj∧ωj,d′′β=∑jd′′αj∧ωjd'\beta=\sum_j d'\alpha_j\wedge\omega_j,\qquad d''\beta=\sum_j d''\alpha_j\wedge\omega_j independent of the chosen expression. There are injective morphisms of sheaves A∙,∙ ↪ B∙,∙ ↪ P∙,∙,A^{\bullet,\bullet} \ \hookrightarrow\ B^{\bullet,\bullet} \ \hookrightarrow\ P^{\bullet,\bullet}, and BB is a bigraded differential algebra [GK17] (§4.6, Definitions 4.9 and 4.13, Proposition 4.15). The differentials need not preserve the larger algebra PP. This is the precise difference between a generalized delta-form and a delta-form in this construction.

The test topology and its dual

It is useful to spell out the continuity condition in the definition of a delta-current. On a fixed chart, freeze a finite tuple ωJ=(ωj)j∈J\omega_J=(\omega_j)_{j\in J} of closed generators. Let AZ(V,φU,ωJ)AZ(V,\varphi_U,\omega_J) consist of the sums ∑jαj∧ωj\sum_j\alpha_j\wedge\omega_j. If C⊂ΩC\subset\Omega is compact and s≥0s\geq0, write pC,s(α)p_{C,s}(\alpha) for the usual smooth coefficient seminorm, controlling derivatives up to order ss. The quotient seminorm is pC,s,ωJ(β)=inf⁡β=∑jαj∧ωjmax⁡j∈JpC,s(αj).p_{C,s,\omega_J}(\beta) = \inf_{\beta=\sum_j\alpha_j\wedge\omega_j} \max_{j\in J}p_{C,s}(\alpha_j). On a fixed chart covering of WW, choose such tuples chart by chart and use the induced locally convex topology. Finally fix a compact subset of WW containing the supports of the test forms.

Definition 16.1.3. (Delta-current). A delta-current on WW is a real linear functional on Bc∙,∙(W)B_c^{\bullet,\bullet}(W) whose restriction to every fixed-chart, fixed-generator, fixed-compact-support space just described is continuous. A delta-current of bidegree (p,q)(p,q) acts on Bcn−p,n−q(W)B_c^{n-p,n-q}(W); its space is denoted Ep,q(W)E^{p,q}(W).

This is [GK17] (Definition 6.3 and §6.4), with bidegree indexing. The two constructions of currents fit into natural maps Pp,q(W)⟶Ep,q(W)⟶Dp,q(W).P^{p,q}(W)\longrightarrow E^{p,q}(W) \longrightarrow D^{p,q}(W). The first associates a functional by integration, ⟨[α]E,β⟩=∫Wα∧β,β∈Bcn−p,n−q(W);\langle[\alpha]_E,\beta\rangle=\int_W\alpha\wedge\beta, \qquad \beta\in B_c^{n-p,n-q}(W); the second restricts a functional to the smaller space Acn−p,n−q(W)A_c^{n-p,n-q}(W). In particular, enlarging the space of test forms reverses the direction of the comparison map on duals.

Proposition 16.1.4. (Functorial detection of a generalized delta-form). Let α,β∈Pp,q(W)\alpha,\beta\in P^{p,q}(W). Then α=β\alpha=\beta if and only if [f∗α]D=[f∗β]D[f^*\alpha]_D=[f^*\beta]_D on every open W′⊂(X′)anW'\subset(X')^{\mathrm{an}} with fan(W′)⊂Wf^{\mathrm{an}}(W')\subset W, for every algebraic morphism f:X′→Xf:X'\to X.

Open proof obligation (RC2). Prove detection and its converse on actual tropical pullback tests; GK17 Proposition 6.8 supplies the criterion.

NAG-900291 · Proof status

Remark 16.1.5 (Which operations have been constructed). There is a product Ba,b×Ep,q→Ea+p,b+qB^{a,b}\times E^{p,q}\to E^{a+p,b+q}, and d′d', d′′d'' act on EE by transposition. Proper algebraic morphisms have pushforward on delta-currents [GK17] (§6.4, §6.9 and §6.10). These statements do not give a product of two arbitrary delta-currents or pullback of an arbitrary delta-current. Nor do the displayed comparison maps by themselves establish injectivity, surjectivity, or extension of every ordinary current to the larger test space.

Positivity and positive representatives

The adjectives strong, positive and weak refer to the three Lagerberg cones. They should be distinguished from the following two additional issues: which chart representatives are positive, and which test space is used by a current.

Definition 16.1.6. (Positive generalized delta-form). A chart class β∈Pp,p(V,φU)\beta\in P^{p,p}(V,\varphi_U) is positive if, for every compatible algebraic morphism and chart, its pulled-back restriction is a tropical delta-preform whose underlying supercurrent is positive. It is positively representable if it has a positive ambient representative β~∈Pp,p(Ω~)\widetilde\beta\in P^{p,p}(\widetilde\Omega). A generalized delta-form on WW has either property if the property holds locally on tropical charts.

There are corresponding strong and weak versions [GK19] (Definitions 2.2 and 2.6). Positive representability implies positivity, since ambient affine pullback and restriction to an effective tropical cycle preserve the relevant cones. The converse is not part of the definition or of this implication.

Definition 16.1.7. (Positive delta-current). A symmetric T∈Ep,p(W)T\in E^{p,p}(W) is positive if ⟨T,β⟩≥0\langle T,\beta\rangle\geq0 for every positive β∈Bcn−p,n−p(W)\beta\in B_c^{n-p,n-p}(W). It is weakly positive when tested against strongly positive forms, and strongly positive when tested against weakly positive forms.

Restriction of a positive delta-current is a positive ordinary current, because every positive smooth test form is a positive delta-form [GK19] (Definition 2.10 and Remark 2.12). For a generalized delta-form β\beta, the statements proved in that paper are more specific: β positively representable ⟹ [β]E positive,β positive ⟺ [f∗β]D positive for every algebraic f.\begin{split} \beta\text{ positively representable} &\ \Longrightarrow\ [\beta]_E\text{ positive},\\ \beta\text{ positive} &\ \Longleftrightarrow\ [f^*\beta]_D\text{ positive for every algebraic }f. \end{split} These are [GK19] (Corollary 2.11 and Proposition 2.13). In the second line the inverse image of WW is the domain of the current. In the first line, positive representability is an actual hypothesis.

Remark 16.1.8 (The smooth case and the middle-degree caution). For a smooth form, positivity of the form and of either associated current agree [GK19] (Remark 2.14). In middle degree, one must not supplement this with the false assertion that strongly positive real Lagerberg forms span all symmetric forms. The counterexample in [BGJK] (Example 2.3.6) also shows that the weak cone can contain a line. Thus [GK19] (Remark 1.12), which repeats the older spanning assertion, is not a valid general-purpose nondegeneracy argument. In degree (1,1)(1,1) the three cones agree, so this particular obstruction does not affect the curvature comparisons below.

16.2 Piecewise smooth weights and four plurisubharmonicity conditions

Definition 16.2.1. (The four conditions for a continuous weight). Let u:W→Ru:W\to\mathbf R be continuous.

  1. uu is current-psh for the original smooth test theory if d′d′′[u]Dd'd''[u]_D is positive.

  2. uu is delta-psh if d′d′′[u]Ed'd''[u]_E is positive.

  3. uu is functorial current-psh if u∘fanu\circ f^{\mathrm{an}} is current-psh on (fan)−1(W)(f^{\mathrm{an}})^{-1}(W) for every algebraic morphism f:X′→Xf:X'\to X.

  4. uu is functorial delta-psh if the analogous condition holds with delta-psh in place of current-psh.

These are the conditions in [GK19] (Definition 3.1 and §3.4); the explicit label current-psh records which of the several uses of psh is intended. For a continuous metric on a line bundle LL, the condition is imposed on each weight −log⁡∥s∥-\log\|s\|, where ss is a nowhere vanishing local algebraic frame. The ordinary or delta Poincare–Lelong formula for invertible functions makes the condition independent of the frame [GK19] (§3.5).

Proposition 16.2.2. (Implications requiring no comparison theorem). For continuous weights and for continuous metrics, functorial delta-psh ⟹ delta-psh ⟹ current-psh,functorial delta-psh ⟹ functorial current-psh ⟹ current-psh.\begin{gathered} \text{functorial delta-psh}\ \Longrightarrow\ \text{delta-psh}\ \Longrightarrow\ \text{current-psh},\\ \text{functorial delta-psh}\ \Longrightarrow\ \text{functorial current-psh}\ \Longrightarrow\ \text{current-psh}. \end{gathered}

Proof. Use the identity morphism to remove the functorial quantifier. For the remaining implication, restrict the positive delta-current to smooth test forms. Apply the same argument after every morphism for the functorial version. ◻

The proposition states implications, not their strictness. In particular, it is not a counterexample to the converse of any arrow. For smooth metrics the arrows collapse to equivalences: c1(L,∥⋅∥)≥0 as a smooth form ⟺ current-psh ⟺ functorial delta-psh.\begin{gathered} c_1(L,\|\cdot\|)\geq0\text{ as a smooth form}\\ \ \Longleftrightarrow\ \text{current-psh} \ \Longleftrightarrow\ \text{functorial delta-psh}. \end{gathered} This is [GK19] (Corollary 8.9); it also includes functorial current-psh, and delta-psh follows by Proposition 16.2.2.

Definition 16.2.3. (Delta-metric). A piecewise smooth metric is a delta-metric if every point admits a tropical frame with weight φ\varphi for which the facewise (1,1)(1,1)-form dP′dP′′φd'_Pd''_P\varphi is the restriction of an ambient superform with piecewise smooth coefficients. For such a metric, there is a unique γ∈B1,1(W)\gamma\in B^{1,1}(W) whose associated delta-current, after every algebraic pullback, is the curvature delta-current of the pulled-back metric. This γ\gamma is denoted c1(L,∥⋅∥)c_1(L,\|\cdot\|).

The local criterion and its equivalence with the functorial curvature representation are [GK17] (Definition 9.9, Proposition 9.11 and Definition 9.12). The existence of one associated curvature current does not by itself supply this stronger representation. For a delta-metric, the statement in [GK19] (Proposition 8.8) is the narrower equivalence c1(L,∥⋅∥)≥0 as a delta-form ⟺ functorial current-psh.c_1(L,\|\cdot\|)\geq0\text{ as a delta-form} \ \Longleftrightarrow\ \text{functorial current-psh}.

The Hessian term and the corner term

Let CC be a tropical cycle of pure dimension dd with positive constant weights. Suppose φ\varphi is a continuous piecewise smooth function on an open subset of ∣C∣|C|. Choose a subdivision on whose faces it is smooth. The operators dP′,dP′′d'_P,d''_P differentiate on those faces; they do not include jumps of the first derivatives. The tropical Poincare–Lelong formula is d′d′′[φ]=[dP′dP′′φ]+δφ⋅C.\begin{equation} d'd''[\varphi]=[d'_Pd''_P\varphi]+\delta_{\varphi\cdot C}. \end{equation} Here φ⋅C\varphi\cdot C is the corner cycle [GK17] (Corollary 3.19).

For precision, at a codimension-one face τ\tau choose primitive outgoing normal lifts νσ,τ\nu_{\sigma,\tau} for the maximal faces σ\sigma containing τ\tau. Balancing says that ντ=∑σ⊃τmσνσ,τ\nu_\tau=\sum_{\sigma\supset\tau}m_\sigma\nu_{\sigma,\tau} is tangent to τ\tau. The corner weight at x∈τx\in\tau is mφ⋅C,τ(x)=∑σ⊃τmσ dφσ(x)(νσ,τ)−dφτ(x)(ντ).m_{\varphi\cdot C,\tau}(x) = \sum_{\sigma\supset\tau} m_\sigma\,d\varphi_\sigma(x)(\nu_{\sigma,\tau}) - d\varphi_\tau(x)(\nu_\tau). Changing a normal lift by a vector tangent to τ\tau changes the two terms by equal amounts. This proves independence of the lifts. The full construction and its invariance under subdivision are given in [GK17] (§1.10 and Proposition 1.12).

Theorem 16.2.4. (Piecewise smooth criterion). Let ∥⋅∥\|\cdot\| be a piecewise smooth metric on L∣WL|W. On a tropical frame write −log⁡∥s∥=φ∘trop⁡U,-\log\|s\|=\varphi\circ\operatorname{trop}_U, where the tropical domain meets each face convexly. Then the metric is current-psh if and only if, on every such frame, the following hold:

  1. the restriction of φ\varphi to each maximal tropical face is convex;

  2. the corner cycle φ⋅Trop⁡(U)\varphi\cdot\operatorname{Trop}(U) is effective.

It suffices to verify the conditions on a basis of tropical frames. Equivalently, after a subdivision, the first condition can be replaced by smooth convexity on the maximal pieces, while retaining the corner condition.

Open proof obligation (RC2). Prove the piecewise-smooth corner-current criterion, including its precise conditions; GK19 Theorem 8.4 and Remark 8.5(a) are the source inputs.

NAG-900293 · Proof status

Example 16.2.5. (A corner invisible to the facewise Hessian). For φ(t)=∣t∣\varphi(t)=|t| and η∈Cc∞(R)\eta\in C_c^\infty(\mathbf R), integration by parts separately on the two half-lines gives ⟨d′d′′[φ],η⟩=∫R∣t∣η′′(t) dt=2η(0).\langle d'd''[\varphi],\eta\rangle =\int_{\mathbf R}|t|\eta''(t)\,dt=2\eta(0). Thus d′d′′[∣t∣]=2δ0d'd''[|t|]=2\delta_0, although the facewise Hessian is zero. The sign is reversed for −∣t∣-|t|. This atomic curvature is represented by a delta-preform, but not by a smooth (1,1)(1,1)-form.

Example 16.2.6. (The tropical tripod). For X={x1+x2=1}⊂Gm2X=\{x_1+x_2=1\}\subset\mathbf G_m^2, the tropicalization has three rays of weight one with primitive outgoing vectors (1,0),(0,1),(−1,−1)(1,0),(0,1),(-1,-1). Let φ\varphi be conic, with respective outgoing slopes a,b,ca,b,c, and put u=φ∘trop⁡u=\varphi\circ\operatorname{trop}. Every edge restriction is linear. Nevertheless a nonnegative test function η\eta, equal to one at the vertex, gives ⟨d′d′′[u],trop⁡∗η⟩=a+b+c.\langle d'd''[u],\operatorname{trop}^*\eta\rangle=a+b+c. In particular, (a,b,c)=(−1,0,0)(a,b,c)=(-1,0,0) disproves the claim that facewise convexity alone implies current-psh. This is [GK19] (Example 3.3). The choice (1,1,−2)(1,1,-2) has zero corner curvature: nonnegativity of each individual outgoing slope would be an unnecessarily strong test.

Proposition 16.2.7. (The precise graph criterion). Let φ\varphi be continuous and piecewise C2C^2 on a one-dimensional balanced tropical cycle, viewed as a metric graph with positive constant edge weights mem_e. Parametrize each edge by a coordinate tt in a primitive integral direction; all outgoing derivatives below use this lattice-normalized parametrization. Against restrictions of smooth ambient test functions supported away from the boundary, its curvature is Δφ=∑emeφe′′(t) dt+∑v(∑e∋vme∂νeφ(v))δv,\Delta\varphi = \sum_e m_e\varphi''_e(t)\,dt + \sum_v \left(\sum_{e\ni v}m_e\partial_{\nu_e}\varphi(v)\right)\delta_v, where the vertices include the breakpoints of the chosen subdivision. It is positive precisely when the densities and the vertex coefficients in this formula are all nonnegative.

Proof. Integrate by parts twice on every edge. Continuity and the balancing condition cancel the terms involving derivatives of the test function at vertices. The remaining boundary terms are the weighted outgoing derivatives of φ\varphi. The density measures and the point masses are mutually singular; their positivity can therefore be tested separately. ◻

Remark 16.2.8 (A scope correction to a graph formulation). The slope-only wording of [GK19] (Example 8.6) is correct for piecewise linear functions. For piecewise smooth functions it must retain convexity on the open edges. Indeed, φ(t)=−t2\varphi(t)=-t^2 on an open interval has sum of the two outgoing first derivatives equal to zero at every point, while φ′′=−2\varphi''=-2. Proposition 16.2.7 supplies the precise replacement.

16.3 Why the universal quotient is necessary

NAG-900295 · Conditional on stated inputs

The definitions at the start of this chapter build functoriality into the null space. We give the arguments which this choice makes possible. They also identify the remaining geometric input: pullback and intersection of ambient tropical preforms must already be constructed, with their projection and closedness properties. The following algebra does not replace those results.

Fix a chart (V,φU)(V,\varphi_U) in the algebraic category of Section 16.1. For a compatible map ff and chart upstairs, write Rf(α~)R_f(\widetilde\alpha) for F∗α~∧δTrop⁡(U′)F^*\widetilde\alpha\wedge\delta_{\operatorname{Trop}(U')}. Thus the null space consists of preforms annihilated by every RfR_f. The identity map is among these maps, but it is not the only one.

Proposition 16.3.1. The universal null spaces are stable under compatible pullback. Consequently pullback of a chart class is independent of its ambient representative and is compatible with composition.

Proof. Suppose α~\widetilde\alpha is null on VV, and pull it back along f:X′→Xf:X'\to X. To show that F∗α~F^*\widetilde\alpha is null upstairs, test it after an arbitrary g:X′′→X′g:X''\to X' and compatible chart. The resulting expression is the restriction of (F∘G)∗α~(F\circ G)^*\widetilde\alpha to the tropical cycle of that chart. It is zero because the composite f∘gf\circ g is one of the maps occurring in the original null condition. Thus the pullback descends. Composition and the identity rule hold before quotienting, and hence hold afterward. ◻

A condition using only RidR_{\mathrm{id}} would not support this proof: its hypothesis contains no information about the composite map. This is the precise reason for the quantifier over morphisms. The product on the quotient additionally uses the ideal property of tropical restriction, proved by the ambient intersection theory. Vanishing as a distribution on one support is not a license to multiply arbitrary distributions.

Proposition 16.3.2. Let β=∑iαi∧ωi\beta=\sum_i\alpha_i\wedge\omega_i with ωi\omega_i universally closed. Then the classes ∑id′αi∧ωi\sum_i d'\alpha_i\wedge\omega_i and ∑id′′αi∧ωi\sum_i d''\alpha_i\wedge\omega_i depend only on β\beta. They give differentials satisfying d′2=d′′2=0d'^2=d''^2=0 and d′d′′+d′′d′=0d'd''+d''d'=0.

Proof. It suffices to consider an expression for zero. After every compatible pullback and tropical restriction, it is a zero current. Differentiate that equality. The smooth-current Leibniz rule and closedness of every restricted generator identify its derivative with the restriction of ∑id′αi∧ωi\sum_i d'\alpha_i\wedge\omega_i. This vanishes after all maps and hence represents zero in the universal quotient. The same argument applies to d′′d''. The three differential identities now follow by applying the corresponding identities to the smooth coefficients in one expression. ◻

The proof is a relation argument, not merely a rule for differentiating a preferred representative. It is also why requiring closedness only on the original chart would be insufficient. Universal closedness is preserved by pullback through the same composition argument as Proposition 16.3.1. Closure under products uses GK’s tropical product theorem for closed preforms. With that input, the graded Leibniz rule descends and gives the full differential algebra.

The nodal example and what it detects

GK17 Example 4.22 uses an integral nodal cubic over Cp\mathbb C_p, with p≠2,3p\ne 2,3, and its normalization. The construction supplies a chart whose tropical image near a vertex has four outgoing branches. A PL potential has cancelling total slope at the vertex, so its corner current on that single tropical image is zero. After pullback to a chart of the normalization, the potential is zero on one interval and has slope one on the next; its corner current is a unit atom.

The underlying cubic is xy+px3+py3=0xy+px^3+py^3=0. Its normalization is parametrized on its affine part by x=−up(1+u3),y=−u2p(1+u3).x=-\frac{u}{p(1+u^3)},\qquad y=-\frac{u^2}{p(1+u^3)}. Substitution verifies the equation: the three numerators after multiplication by p2(1+u3)3p^2(1+u^3)^3 are u3(1+u3)u^3(1+u^3), −u3-u^3, and −u6-u^6. The points u=0,∞u=0,\infty map to the node. A regular unit on the nodal curve pulls back to a unit with equal values at these two preimages, which explains why the canonical tropical coordinates downstairs can identify data separated by normalization. The complete skeleton and unit-group computation is the geometric part of the cited example.

The conclusion is precise: restriction to one tropical image does not detect equality of chart classes. It is not, by itself, a counterexample to injectivity of the map from delta-forms to ordinary analytic currents. An ordinary analytic test may use a refined moment map; testing on one tropical image is less information than testing on every actual smooth analytic form. Proposition 16.1.4 avoids this ambiguity by using every algebraic pullback and all ordinary tests there.

Mihatsch’s skeleton definition and its comparison maps

Let XX be a pure-dimensional Hausdorff good analytic space without analytic boundary, over a complete valued field. Mihatsch permits trivial valuation; statements involving weakly smooth forms additionally use their nontrivial-valuation hypotheses. The data of an actual delta-form consist of compatible polyhedral currents on all admitted skeleta, locally obtained from one ambient delta-form through toric coordinates. Compatibility under adding coordinates is part of the definition.

There are two levels of degree. A term α∧[σ]\alpha\wedge[\sigma] with coefficient bidegree (p,q)(p,q) and support codimension cc has tridegree (p,q,c)(p,q,c) and total bidegree (p+c,q+c)(p+c,q+c). Thus a PL function has tridegree (0,0,0)(0,0,0), while its pure corner curvature has tridegree (0,0,1)(0,0,1) and total bidegree (1,1)(1,1). Confusing coefficient degree and total degree gives incorrect test degrees in Green identities.

Proposition 16.4.1. (The coefficient-codimension-zero comparison). In Mihatsch’s analytic setting, piecewise smooth forms identify with the codimension-zero part: PSXp,q≃BXp,q,0PS_X^{p,q}\simeq B_X^{p,q,0}. Original smooth forms embed in it. Under the hypotheses for the weakly smooth theory, weakly smooth forms also embed in this piecewise smooth sheaf.

NAG-900298 · Conditional on stated inputs

Reduction used in the comparison. The local presentation in codimension zero gives surjectivity: its coefficients are piecewise smooth forms in the toric coordinates. For injectivity, the top-degree case is decisive. If a piecewise smooth (n,n)(n,n)-form represents zero on the calibrated tropical part, its remaining possible support lies in lower-dimensional pieces and in the image of the analytic boundary of a compact chart. Those pieces have dimension at most n−1n-1 in the vanishing criterion of Mihatsch Proposition 4.7. An (n,n)(n,n)-form restricts to zero on every such tangent space. It is therefore zero as a piecewise smooth form. For general (p,q)(p,q), wedge with every complementary piecewise smooth form and apply the top-degree statement. The nondegeneracy of local wedge detection, proved for the analytic presentation spaces in the cited comparison, forces the original form to vanish. Finally harmonic coordinate functions are locally PL, so pullback of smooth superforms by them is piecewise smooth. The relation and injectivity statements for that inclusion are those of the weakly smooth construction. ◻

This proof records the reduction and its geometric inputs; it does not replace the analytic wedge-detection theorem by dimension counting alone. On an abstract tropical space with a prescribed sheaf of linear functions, the analogous inclusion into all piecewise smooth forms can be strict. The analytic local-presentation theorem is therefore substantive.

Which differential is preserved

On piecewise smooth forms, the facewise operators are dP′,dP′′d'_P,d''_P. The full current operators d′,d′′d',d'' also contain boundary terms. Mihatsch uses the convention ∂′=dP′−d′,∂′′=dP′′−d′′.\partial'=d'_P-d',\qquad\partial''=d''_P-d''. Thus it is not correct to say that inclusion of all piecewise smooth forms preserves their facewise differential as the full differential. On original smooth and weakly smooth forms the boundary operators vanish, which is precisely what makes those smaller inclusions compatible with the full calculus. For ∣t∣|t| on R\mathbb R, the facewise second derivative is zero, while the full second derivative is 2δ02\delta_0; the codimension-one term must occur in the delta-algebra.

Pullback is a theorem about relations

Mihatsch Theorem 4.5 proves pullback for every analytic morphism between pure-dimensional boundaryless spaces. Locally it is the expected substitution tf⋆γ↦tf∘π⋆γt_f^\star\gamma\mapsto t_{f\circ\pi}^\star\gamma. The substantive assertion is that a zero presentation remains zero after substitution. The proof reduces near Abhyankar points to a map of constant fibre dimension followed by a closed immersion. For the former it uses compatible flat maps of skeleta; for the latter it uses the tropical product description near a Cartier divisor. The regular-immersion geometry in Chapter 22 explains this second mechanism.

NAG-900299 · Comparison requirements remain

This gives actual pullback, including restriction to a closed analytic subspace in the stated category. It does not supply arbitrary actual pushforward. A proper map always suggests a transpose operation on currents, but the resulting current need not have local ambient presentations of the required kind. Mihatsch Example 2.45(3) explicitly notes that even a balanced polyhedral pushforward need not be a delta-form on the target tropical space. Additional representation control is necessary.

NAG-900300 · Open proof obligation

In the common algebraic category, Mihatsch also states an embedding of original GK delta-forms into his sheaf, specifying its action on ambient generators; the source states this comparison without a formal proof. We retain that status. The proved piecewise-smooth comparison above does give an inclusion of weakly smooth forms into Mihatsch’s algebra. It does not give an inclusion of weakly smooth forms into the original GK algebra. This distinction refines, rather than reverses, the comparison table in Chapter 18.

References. [GK17] (Definitions 4.4 and 4.9, Propositions 4.7–4.15, Example 4.22, Proposition 6.8); [Mih24] (Proposition 2.38, Example 2.45, Theorem 4.5, Proposition 4.7 and Section 4.2).

16.5 The six differential operators and their signs

The full derivative of a delta-form has a cellwise part and a boundary part. The distinction is already visible for a PL function, but becomes more important when products have several support codimensions. We use Mihatsch’s convention, in which the boundary operator is the cellwise derivative minus the current derivative. ∂′=dP′−d′,∂′′=dP′′−d′′,d′=dP′−∂′,d′′=dP′′−∂′′.\partial'=d'_P-d',\qquad\partial''=d''_P-d'', \qquad d'=d'_P-\partial',\quad d''=d''_P-\partial''. For coefficient bidegree (a,b)(a,b) and support codimension cc, the total bidegree is (a+c,b+c)(a+c,b+c). The homogeneous degree changes are operatorcoefficient/codimension changetotal changedP′(1,0,0)(1,0)dP′′(0,1,0)(0,1)∂′(0,−1,1)(1,0)∂′′(−1,0,1)(0,1).\begin{array}{c|c|c} \text{operator}&\text{coefficient/codimension change}&\text{total change}\\\hline d'_P&(1,0,0)&(1,0)\\ d''_P&(0,1,0)&(0,1)\\ \partial'&(0,-1,1)&(1,0)\\ \partial''&(-1,0,1)&(0,1). \end{array} An operator landing in a negative coefficient degree is zero. Thus the first derivatives of a continuous piecewise smooth scalar function are cellwise derivatives. Its second derivative can nevertheless acquire a boundary term.

Proposition 16.5.1. On the delta-algebra the following identities hold: (dP′)2=(dP′′)2=(∂′)2=(∂′′)2=0,dP′∂′+∂′dP′=0,dP′′∂′′+∂′′dP′′=0,dP′dP′′+dP′′dP′=0,∂′∂′′+∂′′∂′=0,dP′∂′′+∂′′dP′+dP′′∂′+∂′dP′′=0.\begin{gather*} (d'_P)^2=(d''_P)^2=(\partial')^2=(\partial'')^2=0,\\ d'_P\partial'+\partial'd'_P=0,\qquad d''_P\partial''+\partial''d''_P=0,\\ d'_Pd''_P+d''_Pd'_P=0,\qquad \partial'\partial''+\partial''\partial'=0,\\ d'_P\partial''+\partial''d'_P+d''_P\partial'+\partial'd''_P=0. \end{gather*} The last identity does not assert that each of its two mixed anticommutators vanishes separately.

Proof. The cellwise identities hold on each smooth coefficient form. Expand (d′)2=(dP′−∂′)2=0(d')^2=(d'_P-\partial')^2=0 on a trihomogeneous delta-form. Its three terms have support codimension changes zero, one and two. The direct trigrading makes each codimension part vanish separately, giving (∂′)2=0(\partial')^2=0 and the primed mixed identity. Repeat for d′′d''. Expanding d′d′′+d′′d′=0d'd''+d''d'=0 and separating the same three codimension changes gives the remaining identities. The codimension-one part contains both mixed anticommutators, with the same sign; there is no further grading that separates them. ◻

Example 16.5.2. For f(t)=∣t∣f(t)=|t| on R\mathbb R, one has DDf=2δ0DDf=2\delta_0 and dP′dP′′f=0d'_Pd''_Pf=0. Since ∂′f=∂′′f=0\partial'f=\partial''f=0, (dP′′∂′+∂′dP′′)f=−2δ0,(dP′∂′′+∂′′dP′)f=2δ0.(d''_P\partial'+\partial'd''_P)f=-2\delta_0, \qquad(d'_P\partial''+\partial''d'_P)f=2\delta_0. Thus the two anticommutators cancel but are individually nonzero. Treating all six operators as pairwise anticommuting would erase exactly the corner curvature that the delta-algebra was designed to retain.

The curvature of a piecewise smooth function

For any continuous piecewise smooth function ff, the degree table gives d′′f=dP′′fd''f=d''_Pf and hence DDf=dP′dP′′f−∂′dP′′f.DDf=d'_Pd''_Pf-\partial'd''_Pf. The first term has tridegree (1,1,0)(1,1,0); it records the Hessian on the cells. The second has tridegree (0,0,1)(0,0,1); it records the slope mismatch along codimension-one faces. For a PL function the first term vanishes, so the entire curvature is the negative of the displayed boundary derivative. This is compatible with the positive corner mass of a convex maximum.

If ff is harmonic in an enlarged coordinate system, this conclusion concerns its full curvature. It cannot be read as saying that every cellwise Hessian or every individual normal slope is zero. The compatibility theorem for that coordinate system identifies the cancellations in the full operator. This is another reason to retain the operator notation when comparing original smooth, weakly smooth and piecewise smooth forms.

16.6 Codimension as a filtration of the differential complex

Fix a total first degree pp. In the complex (Bp,∙,d′′)(B^{p,\bullet},d''), put FcBp,qF^cB^{p,q} equal to the sum of components of support codimension at least cc. Both terms of d′′=dP′′−∂′′d''=d''_P-\partial'' preserve this decreasing filtration: the first keeps codimension fixed and the second raises it by one. The filtration is finite in each degree.

Proposition 16.6.1. The differential on the associated graded complex is dP′′d''_P. For a dP′′d''_P-closed homogeneous term α\alpha of codimension cc, the first obstruction to correcting it to a full d′′d''-closed term is the class of ∂′′α\partial''\alpha in the dP′′d''_P-cohomology of codimension c+1c+1.

Proof. Modulo Fc+1F^{c+1} the boundary derivative disappears, proving the first assertion. To make α+β\alpha+\beta closed modulo codimension c+2c+2, where β\beta has codimension c+1c+1, one must solve dP′′β=∂′′α.d''_P\beta=\partial''\alpha. The right side is dP′′d''_P-closed by Proposition 16.5.1. Its class is therefore the obstruction to this equation. Replacing α\alpha by α+dP′′γ\alpha+d''_P\gamma changes the right side by −dP′′∂′′γ-d''_P\partial''\gamma, so the obstruction depends only on the associated-graded cohomology class. ◻

Further corrections can have further obstructions even though the differential has only these two codimension pieces. The standard spectral sequence of a finite filtered complex organizes those successive equations; its existence is not a statement that they all vanish. For a reference calculation, the explicit first obstruction is often more useful than an unsupported claim that cellwise closedness is sufficient. It explains algebraically how a facewise primitive can fail to be a primitive in the actual delta-calculus.

References. The operator convention and trigrading are [Mih24] (Definition 2.41 and Section 4.2). The identities and first-obstruction calculation are consequences of that trigrading and the current differential identities.

Polyhedral Monge–Ampère theory beyond the toric case

The toric theory of [BGJK25] uses a global convex model. On a balanced polyhedral space there need not be a single ambient convex function representing every positive potential. Botero–Mazzon–Pille-Schneider therefore start with polyhedral convexity. A function is polyhedrally convex, abbreviated PC, when it satisfies the convexity inequality for the polyhedral convex combinations allowed by the space. For a piecewise affine function this is equivalent to nonnegative intersection with every local positive polyhedral cycle. Positivity against the fundamental cycle alone is a weaker condition. Fix a piecewise affine reference function γ\gamma and its growth class. The regularizable class PCreg(X,γ)\mathrm{PC}_{\mathrm{reg}}(X,\gamma) consists of uniform decreasing limits of piecewise affine PC functions with growth γ+O(1)\gamma+O(1). Their Theorem 4.13 extends the mixed polyhedral Monge–Ampère operator to this class, with symmetry, positivity, fixed total mass and integration by parts. The measure is initially defined on their compactification; positivity of γ\gamma at infinity ensures that it is supported on XX. This is a construction in a specified polyhedral function class. It does not identify that class with UX\mathcal U_X on an arbitrary Berkovich space.

A continuous obstacle for which orthogonality fails

The following computation makes the obstruction to a general envelope argument explicit. It also corrects the negative-axis branch printed in Example 6.23 of [BMP26]. Equip the union XX of the two coordinate axes in R2\mathbb R^2 with weight one on its four rays, and put γ=∣x∣+∣y∣\gamma=|x|+|y| on XX.

Proposition 16.7.1. For ϵ>0\epsilon>0, define the continuous obstacle φ(x,0)=∣x∣+ϵ,φ(0,y)=∣∣y∣−ϵ∣.\varphi(x,0)=|x|+\epsilon,\qquad \varphi(0,y)=\bigl||y|-\epsilon\bigr|. Its PC envelope with growth bounded above by γ+O(1)\gamma+O(1) is h(x,0)=∣x∣,h(0,y)=max⁡(0,∣y∣−ϵ).h(x,0)=|x|,\qquad h(0,y)=\max(0,|y|-\epsilon). Moreover, with the polyhedral Monge–Ampère normalization of the slope formula, MA⁡poly(h)=2δ(0,0)+δ(0,ϵ)+δ(0,−ϵ),∫X(φ−h)MA⁡poly(h)=2ϵ>0.\operatorname{MA}_{\mathrm{poly}}(h) =2\delta_{(0,0)}+\delta_{(0,\epsilon)}+\delta_{(0,-\epsilon)}, \qquad \int_X(\varphi-h)\operatorname{MA}_{\mathrm{poly}}(h)=2\epsilon>0.

Proof. The function hh is the restriction of the ambient convex function max⁡{∣x∣,∣y∣−ϵ}\max\{|x|,|y|-\epsilon\}. It is piecewise affine PC, lies below φ\varphi, and differs boundedly from γ\gamma. Thus it is an admissible competitor. Conversely, let vv be any PC competitor. Its restriction to each entire axis is convex: the usual convex combinations on a line are polyhedral convex combinations after subdivision at the point in question. This first gives the assertion at rational points, and then at all points by continuity of PC functions. Since v(0,ϵ)v(0,\epsilon) and v(0,−ϵ)v(0,-\epsilon) are at most zero, convexity gives v(0,0)≤0v(0,0)\leq 0 and v(0,y)≤0v(0,y)\leq 0 for ∣y∣≤ϵ|y|\leq\epsilon. Choose CC with v≤γ+Cv\leq\gamma+C. For 0<t<R0<t<R on either horizontal ray, convexity between 00 and RR gives v(t,0)≤t+tC/Rv(t,0)\leq t+tC/R; letting R→∞R\to\infty gives v(t,0)≤tv(t,0)\leq t. On a vertical ray, interpolate instead between ϵ\epsilon and RR. For t>ϵt>\epsilon this gives v(0,±t)≤t−ϵR−ϵ(R+C)⟶t−ϵ.v(0,\pm t)\leq\frac{t-\epsilon}{R-\epsilon}(R+C) \longrightarrow t-\epsilon. Hence v≤hv\leq h, proving the envelope identity. The outgoing slopes of hh at the origin are 1,1,0,01,1,0,0; the slope jumps at (0,±ϵ)(0,\pm\epsilon) are one. There are no other corners. This proves the measure formula. The obstacle gap is ϵ\epsilon at the origin and zero at the other two atoms, giving the integral. ◻

In the source notation the envelope is Pγ(φ−γ)=hP_\gamma(\varphi-\gamma)=h, an unshifted function with growth γ+O(1)\gamma+O(1). The outside branch ∣y−ϵ∣|y-\epsilon| printed in the cited example has left limit 2ϵ2\epsilon at y=−ϵy=-\epsilon, whereas the middle branch has value zero. Replacing that outside branch by ∣y∣−ϵ|y|-\epsilon gives exactly the continuous obstacle above. The failure of orthogonality survives this repair. The total mass is four, equal to the total outgoing slope of γ\gamma. Thus positivity, correct mass, regularizability and an explicit envelope do not imply orthogonality. In particular the variational implication in [BMP26] (Proposition 6.12) must retain both regularity and orthogonality hypotheses, as well as the prescribed mass of the measure. This example concerns the PC class on the balanced cross; it is not a counterexample to the non-Archimedean envelope theorem in Section 19.19.

16.8 Local geometry in tropical Calabi–Yau problems

Amini–Piquerez–Wu work with compact prekähler tropical varieties and locally convex potentials. Their framework contains a background semipositive form ω\omega, and a Kähler form θ\theta used for approximation. Theorem 5.13 of [APW26] has two different conclusions: continuous θ\theta-psh functions admit uniform PL approximations in the same class; an ω\omega-psh function admits decreasing PL approximations in the enlarged classes ω+αjθ\omega+\alpha_j\theta, with αj↓0\alpha_j\downarrow 0. When the limit is continuous, the latter convergence is uniform by Dini’s argument. It is essential to retain the varying background form in the second statement. Their existence and nonexistence results distinguish local models. A point is regular in codimension one in their terminology if a neighbourhood has a tropical chart in L×Rd−1L\times\mathbb R^{d-1}, with LL a generalized tropical line. If every point of the open stratum has this property, Theorem 10.4 gives their tropical Calabi–Yau theorem. This phrase is a local chart condition at every point; it does not mean only that a singular set has codimension at least two. The obstruction in Theorem 11.4 is also precise. At a codimension-two face, let ZZ be the two-dimensional star fanfold and let EE be its linear span, of dimension r≥3r\geq 3. Condition (G) requires that the image of ZZ in E/FE/F have dimension two for every (r−2)(r-2)-dimensional subspace FF. If one such star satisfies (G), their universal tropical Calabi–Yau assertion fails for every strongly semipositive background form. This is a failure of surjectivity of the Monge–Ampère operator onto all admissible measures, not a claim that no equation has any solution.

What the analytic comparison additionally requires

Theorem 12.11 of the same preprint works over K=C((t))K=\mathbf C((t)) with a smooth projective variety and starts with an existing bounded tropical solution in its toric embedding setup. It assumes a full-dimensional SNC dual complex, strong faithfulness relative to that complex, and a probability measure on it absolutely continuous with respect to its top-dimensional Lebesgue measure. Under these hypotheses the tropical potential extends to a continuous non-Archimedean psh potential with the prescribed Monge–Ampère measure, and has the paper’s comparison property. Ordinary injectivity of one tropical chart is not the strong faithfulness assumption, and the theorem does not solve the tropical existence problem that precedes it. The mixed calculus cited there from the companion work is retained as an external dependency. The results just described are source theorems, not independently reconstructed proofs of the companion theory. They supplement the local CLD theory by global positivity and envelope geometry; they do not replace the test-space comparisons required for actual Green forms.

References. [BMP26] (Definitions 3.1–3.3, Theorem 4.13, Proposition 6.12 and Example 6.23); [APW26] (Theorems 5.13, 10.4, 11.4 and 12.11).