8 · Intrinsic smooth forms, integration and Stokes

18 NAG Tags

8.1 Original smooth forms on an analytic space

Let KK be a complete non-Archimedean field. For the current constructions below, let XX be a good Hausdorff KK-analytic space of pure dimension nn. When discussing the algebraic presentation, we impose Gubler’s hypotheses: KK is algebraically closed and nontrivially valued, and XX is the analytification of an algebraic variety over KK. Statements on that presentation will not be used to enlarge its original hypotheses silently.

Definition 8.1.1. (Tropical presentation of an original smooth form). An analytic moment map on an analytic domain WW is an analytic morphism f=(f1,…,fr):W⟶(Gmr)an.f=(f_1,\ldots,f_r):W\longrightarrow(\mathbb G_m^r)^{\mathrm{an}}. The functions fif_i are invertible. We use the tropicalization convention ftrop(x)=(−log⁡∣f1(x)∣,…,−log⁡∣fr(x)∣).f_{\mathrm{trop}}(x)=(-\log|f_1(x)|,\ldots,-\log|f_r(x)|). An original smooth (p,q)(p,q)-form is locally represented by a pullback ftrop∗αf_{\mathrm{trop}}^*\alpha, where α\alpha is a smooth superform on the tropical image. Presentations are compared after common refinement, and compatible local presentations are glued as a sheaf. Denote the resulting sheaf by Asmp,qA^{p,q}_{\mathrm{sm}}.

Here a superform on a polyhedral subset is the restriction of a superform from a real open neighborhood, modulo forms whose restrictions to tangent spaces of all its cells vanish. The tangent-space quotient is essential: an ambient form can be nonzero while its restriction to the polyhedral set is zero. Subdivision does not change this quotient. Likewise, the analytic definition requires common refinements and sheafification; it is not the choice of one permanent tropical coordinate system on XX.

Remark 8.1.2 (Functions and sheafification). The sheaf Asm0,0A^{0,0}_{\mathrm{sm}} consists of functions locally of the form φ(−log⁡∣f1∣,…,−log⁡∣fr∣)\varphi(-\log|f_1|,\ldots,-\log|f_r|) with φ\varphi smooth. On a fixed tropical chart, an arbitrary smooth form need not be represented by a single superform on that same chart; see [Gubler] (Remark 5.7). This is why a definition that merely says “pull back superforms along a tropicalization” is incomplete.

Example 8.1.3. (Continuous and tropically continuous functions). On AK1,an\mathbb A_K^{1,\mathrm{an}}, the function x↦∣T(x)∣x\mapsto|T(x)| is continuous. Near the rigid point T=0T=0, every analytic invertible function has constant absolute value on a sufficiently small neighborhood. Every function given near 00 by a continuous function of finitely many logarithms of invertible functions is therefore locally constant there. The function ∣T∣|T| is not locally constant near 00 and has no such tropical presentation.

Consequently, continuity of a function or metric on XX does not mean continuity of its coefficients in one local tropical presentation. This is the distinction made in [CLD12] (§3.1.6).

Theorem 8.1.4. (Gubler’s algebraic presentation). Let KK be complete, algebraically closed, and nontrivially valued, and let ZZ be an algebraic variety over KK. The smooth-form sheaf constructed on ZanZ^{\mathrm{an}} using algebraic moment maps on very affine Zariski open subsets agrees with the original smooth-form sheaf constructed using analytic moment maps. The identification respects d′d', d′′d'', JJ, and wedge product.

Proof. Use the common variety class of NAG-003062 and NAG-003078: integral separated schemes of finite type over the stated field. Use the same coordinates tf=−log⁡∣f∣t_f=-\log|f| and the same slot interchange J0J_0 in both presentations. Restricting algebraic units to an ordinary affinoid neighbourhood gives an analytic moment germ (NAG-002423–002424).

Conversely choose an ordinary affinoid witness for an analytic germ and work inside its ordinary interior. Include in one tuple all units of any finite germ comparison under consideration. By NAG-003062 there are a compact Laurent affinoid neighbourhood CC, a very affine algebraic open UU and algebraic units gig_i on UU such that C⊂UanC\subset U^{\mathrm{an}} lies in the witness and ∣gi∣=∣fi∣|g_i|=|f_i| throughout CC. The construction truncates in the actual Laurent quotient with error smaller than min⁡i∥fi−1∥−1\min_i\|f_i^{-1}\|^{-1} for a nonempty tuple (NAG-003061). It gives an ordinary open chart about the point inside CC. On this chart the gig_i are constants times characters for the algebraic unit lattice, and their tropicalizations are integral affine functions of its coordinates (NAG-003077).

On the actual joint image of the two tuples the projections agree both on points and on every cell tangent. Thus they induce the same pullback in both alternating factors, including after adjoining further coordinates (NAG-003065). Two choices can be compared on an ordinary common intrinsic refinement inside both witnesses (NAG-003077). Consequently replacement descends through the germ relation and gives an inverse on stalks to the restriction of algebraic presentations; this is the algebraic identification in NAG-003078. Wedge, the two differentials and J0J_0 commute with these affine pullbacks on the ambient superforms and descend by NAG-002422 and NAG-002424. ◻

The theorem is an equality of two presentations of the original sheaf. Enlarging the allowed tropicalizations to harmonic ones, or adjoining polyhedral currents, makes a different construction and requires a separate comparison theorem.

8.2 Local structure near classical points

Smooth real functions can be locally constant at rigid points

Let x∈X(K)x\in X(K) and let ff be an analytic unit near xx. Continuity of ff gives a smaller neighbourhood on which ∣f(y)−f(x)∣<∣f(x)∣.|f(y)-f(x)|<|f(x)|. The strong triangle inequality then implies ∣f(y)∣=∣f(x)∣|f(y)|=|f(x)|. For finitely many units f1,…,frf_1,\ldots,f_r, choose one neighbourhood where all these equalities hold. Every original smooth function presented as F(−log⁡∣f1∣,…,−log⁡∣fr∣)F(-\log|f_1|,\ldots,-\log|f_r|) is constant there.

The corresponding positive-degree forms presented by these coordinates vanish there because the local tropical image is a point. This is a distinctive feature of the original real differential calculus. The word smooth should not evoke a real manifold chart of dimension 2n2n around a rigid point. The useful differential directions occur at nonclassical analytic points and their tropical neighbourhoods.

This does not force a global smooth function on a connected analytic space to be constant. The neighbourhoods on which it is constant can vary with the rigid point, and nonclassical points lie between them. On the projective line, a smooth regularization of the standard metric weight varies near the Gauss point while being constant in appropriate rigid neighbourhoods after local trivialization.

A direct consequence for singular potentials

On a curve, the point mass δx\delta_x at a rigid point is a perfectly meaningful top-degree current: it evaluates smooth scalar tests at xx. But it cannot be the curvature of an original smooth function near xx, because such a function has zero Hessian on some neighbourhood. Its logarithmic potential is singular. Poincaré–Lelong supplies precisely this kind of potential.

The distinction between a smooth form and a current is therefore already necessary in dimension one. A smooth positive metric can approximate a model metric whose mass concentrates at a nonclassical point; a logarithmic Green function can create a divisor mass at a rigid point. These are different limiting and singularity phenomena.

Restriction of a form and vanishing of its germ

In ordinary differential geometry, a positive-degree form can restrict to zero on a point while remaining nonzero arbitrarily near it. CLD’s tropical-dimension support lemmas give stronger vanishing statements for certain degrees at analytic points of small tropical dimension. Their proofs use the local structure of moment maps, not just the formal rule that a form restricts to zero on a lower-dimensional space.

This distinction is used in the higher-codimension extension arguments of CLD25. If a test has degree larger than the relevant tropical dimension along a cycle, the source’s support theorem can force its support away from that cycle. This is why a singular expression on the complement may still pair with the differentiated test in a canonical way. One must cite the support theorem; merely observing that the restriction to the cycle is zero would not justify the extension.

Smooth density does not imply positive density

Original smooth scalar functions are dense in continuous functions in the appropriate local or compact-support topology. This permits the measure-theoretic extension used in defining uTuT. The approximation can be made with prescribed scalar bounds under the relevant cutoff construction.

However, density of a vector space of smooth functions does not imply density of its psh cone in the continuous current-psh cone. An approximation may develop negative Hessian terms. This is analogous to the difference between approximating an arbitrary continuous function by polynomials and approximating a convex function by convex polynomials on a specified domain: the positivity constraint is an additional problem.

In real convex geometry, convolution on a smaller fixed open set preserves convexity. On an analytic space, a given continuous potential may not be represented by a convex function in one fixed moment chart, and refining charts introduces relations that the approximation must respect. This is the precise reason a local smooth-density theorem is insufficient to prove general psh regularization.

Example. (Local constancy of a unit logarithm). For f(T)=a+Tf(T)=a+T with a≠0a\ne0, the strict ultrametric inequality gives ∣f∣=∣a∣|f|=|a| on ∣T∣<∣a∣|T|<|a|. Hence log⁡∣f∣\log|f| is constant on that disc. At the Gauss point of radius r>∣a∣r>|a|, however, ∣f∣=max⁡(∣a∣,r)=r|f|=\max(|a|,r)=r. These two descriptions are the two regions of v(f)=min⁡(v(a),−log⁡r)v(f)=\min(v(a),-\log r) along the Gauss points. Local constancy near a classical point is compatible with variation along an annular skeleton.

References. [CLD12] (§§3.1–3.3, 5.4–5.6); [CLD25] (§§7.1, 15.6, 16.8 and 18.7).

8.3 Integration on weighted polyhedral spaces

The data needed for a top-degree integral

Suppose a pure nn-dimensional polyhedral complex has an integral tangent lattice NσN_\sigma on every maximal cell σ\sigma and a weight mσm_\sigma. Choose a lattice basis e1,…,ene_1,\ldots,e_n of NσN_\sigma. The basis fixes a normalized volume; replacing it by another lattice basis changes the determinant by ±1\pm1 and does not change the resulting density. A superform of degree (n,n)(n,n) can be contracted against this volume and integrated over the cell. The total integral is the sum over maximal cells with the weights mσm_\sigma.

The three ingredients have distinct roles. The polyhedral support says where integration occurs. The lattice fixes a unit of volume. The weight records the multiplicity of the analytic or algebraic object over that support. The Euclidean volume inherited from an arbitrary ambient embedding generally has the wrong normalization.

Example. (A parametrization of index two). Consider the map Gm→Gm\mathbb G_m\to\mathbb G_m, z↦z2z\mapsto z^2. On tropical coordinates it is t↦2tt\mapsto2t. Its generic algebraic degree is two, including the inseparable case when appropriate scheme-theoretic multiplicities are retained. If one integrates on the target using its standard lattice, the degree formula has a factor two. If instead one treats the source coordinate as a primitive target coordinate, the factor is lost. This simple calculation is the prototype for the lattice indices in tropical pushforward and the calibrated degree formulas of CLD25.

The boundary calculation behind balancing

For a compactly supported (n−1,n)(n-1,n)-superform, apply ordinary Stokes on every maximal cell. Each codimension-one face receives a sum of contractions with its outgoing primitive normals. The sum vanishes precisely when the weighted normal vectors balance in the quotient by the tangent space of that face. Thus balancing turns a cellwise boundary formula into a global closedness statement.

This also explains why subdividing a cell changes no integral or current. Newly introduced interior faces occur in opposite pairs with compatible lattice volumes. Their boundary contributions cancel. A genuine corner of a function behaves differently: the normal derivatives on the two sides need not agree, and their difference survives as the divisor weight. The same integration-by-parts calculation therefore proves both closedness of a tropical cycle and the corner formula for a PL function.

Example. (Distributional curvature and outgoing slopes). For f,η∈C2([a,b])f,\eta\in C^2([a,b]), integration by parts twice gives ∫abf′′η dt=∫abfη′′ dt+[f′η−fη′]ab.\int_a^b f''\eta\,dt=\int_a^b f\eta''\,dt+[f'\eta-f\eta']_a^b. If a<0<ba<0<b and η\eta is compactly supported in (a,b)(a,b), then ∫abmax⁡(0,t)η′′(t) dt=η(0).\int_a^b\max(0,t)\eta''(t)\,dt=\eta(0). The distributional second derivative is therefore δ0\delta_0, although the classical second derivative vanishes off the origin. On the balanced tropical line, let the test be induced by an ambient smooth function and compactly supported away from the outer ends. The terms involving the first derivative of the test cancel by balancing. The calculation on the outgoing parameters then gives (a1+a2+a3)η(0)(a_1+a_2+a_3)\eta(0) for a continuous affine-on-rays function with outgoing slopes a1,a2,a3a_1,a_2,a_3. The vertex coefficient is the sum of those slopes.

The tangent quotient

A curve can be embedded in a torus of arbitrarily large dimension. Its tropical image remains one-dimensional. A superform on the ambient real vector space is used only through its restrictions to tangent directions of the tropical image. Ambient components that vanish on all relevant tangent spaces represent the zero form on the image. This quotient is essential in the definition of CLD forms; the pullback notation alone does not specify equality of forms.

The example also explains why one cannot prove positivity by examining an arbitrary matrix in arbitrary ambient coordinates without passing to the quotient. Positivity is a condition on the actual tangent directions tested by the space. Refinement by additional moment coordinates must preserve that condition. In the intrinsic construction below, equality and positivity are formulated on common refinements so that the resulting forms do not depend on the original embedding.

8.4 Integration, canonical weights, and Stokes’ theorem

Compatibility of integral presentations

Let XX have pure dimension nn and let α\alpha be a compactly supported smooth (n,n)(n,n)-form. Locally write α=ftrop⁡∗β\alpha=f_{\operatorname{trop}}^*\beta for an analytic moment map ff. The image ftrop⁡(X)f_{\operatorname{trop}}(X) can have singular faces, several analytic branches can map to the same face, and an integral-affine map between two presentations can have a nontrivial lattice index. Therefore the unweighted integral of β\beta on the set-theoretic image cannot be the definition.

CLD assigns a canonical calibration to suitable tropical presentations. In the algebraic torus case, it agrees with the usual tropical multiplicity. In the analytic construction, finite maps and the degrees of the relevant analytic fibres supply the local weights. The compatibility theorem says that passing to a refined moment map and then projecting gives the same weighted integral. This is the essential geometric theorem behind the definition, not a consequence of the notation ftrop⁡∗f_{\operatorname{trop}}^*.

The construction in three steps

First, on a calibrated polyhedral cell, contract a top-degree superform with its calibrated tangent volume and integrate the resulting density. Second, sum over cells. Lower-dimensional overlaps do not contribute to a top-degree integral, and subdivision creates cancelling interior faces for boundary integrals. Third, use a smooth partition of unity on the analytic space to reduce a general compactly supported form to finitely many tropical presentations.

Independence must be proved at both transitions: common refinements compare local tropical presentations, while the partition-of-unity identity compares decompositions of the global form. For a paracompact good analytic space, the smooth partitions of unity constructed in CLD12 make this procedure available. Compact support is important because the actual integral is a finite local calculation even when the space itself is not compact.

Example. (The annular integral). On an annulus with tropical coordinate t=−log⁡∣T∣t=-\log|T|, take a compactly supported form α=f(t) d′t∧d′′t.\alpha=f(t)\,d't\wedge d''t. The canonical weight is one, and ∫Xα=∫f(t) dt\int_X\alpha=\int f(t)\,dt in the chosen lattice normalization. Replacing TT by TmT^m changes the real coordinate to s=mts=mt. The pullback of d′s∧d′′sd's\wedge d''s has two factors of mm, while the calibrated volume and the image parametrization account for the corresponding degree and Jacobian. One cannot change the coordinate and leave the integration normalization unchanged.

Stokes on the analytic space

On every cell, Stokes is ordinary integration by parts. At an interior codimension-one face, the sum of boundary contractions vanishes by the calibrated balancing condition. The remaining contributions lie on the analytic boundary. Thus, when the support of η\eta avoids that boundary, ∫Xd′η=0for η∈Acn−1,n(X),∫Xd′′ξ=0for ξ∈Acn,n−1(X).\int_X d'\eta=0\quad\text{for }\eta\in A_c^{n-1,n}(X), \qquad \int_X d''\xi=0\quad\text{for }\xi\in A_c^{n,n-1}(X). For a compact analytic domain with boundary, there are corresponding boundary integrals. Deleting those terms is justified by the support condition, not by compactness of the domain.

This is the bridge from the balancing equation of Chapter 5 to the closedness of an integration current. It also explains why arbitrary piecewise smooth functions cannot be differentiated cellwise and then treated as smooth global functions. Their internal interfaces contribute additional terms.

The signed boundary measure for ordinary native forms

Theorem. Let KK be complete for a specified real-valued non-Archimedean absolute value, possibly trivial. Let XX be a good Hausdorff KK-analytic space in the norm-admissible generalized affinoid atlas, analytically pure of dimension d≥1d\ge1, and let F\mathcal F be coherent on XGX_G. Take an ordinary real smooth moment form β\beta with compact ordinary germ support SS. Put D=d′D=d' for type (d−1,d)(d-1,d), or D=d′′D=d'' for type (d,d−1)(d,d-1). There is a unique finite signed Radon measure μX,F,βD\mu_{X,\mathcal F,\beta}^{D} supported in S∩∂KXS\cap\partial_K X, where ∂KX=X∖Int⁡(X/K)\partial_KX=X\setminus\operatorname{Int}(X/K), such that, for every ordinary smooth real scalar uu on XX, ∫Xu dμX,F,βD=IX,FN(D(uβ)).(S1)\int_X u\,d\mu_{X,\mathcal F,\beta}^{D} = I^N_{X,\mathcal F}\bigl(D(u\beta)\bigr). \tag{S1} Here INI^N is the native integral of NAG-002425. In particular the mass is IX,FN(Dβ)I^N_{X,\mathcal F}(D\beta). Here support of a signed measure means support of its total variation. For support contained in the analytic interior the measure vanishes.

We give the proof first for D=d′D=d'. All charts use −log⁡∣f∣-\log|f| and the interleaved positive lattice top generator. The strict integral structure-coefficient estimate (B) below is used only for a single global representative of the primitive. The multiplier in (B) is any ordinary smooth scalar.

The native bulk comparison under free extension.

On a pure compact affinoid Y=M(B)Y=\mathcal M(B), let MM be a finite BB-module. For a finite free polyradius let L=KrL=K_r, BL=B⊗^KLB_L=B\widehat\otimes_K L, ML=BL⊗BMM_L=B_L\otimes_B M, and p:YL→Yp:Y_L\to Y. For any ordinary top section ω\omega we have IYL,MLN(p♮ω)=IY,MN(ω).(S2)I^N_{Y_L,M_L}(p^\natural\omega)=I^N_{Y,M}(\omega). \tag{S2} To see this, extend the units and use the same ambient superform. The full inverse image of an affinoid domain with algebra CC has algebra CL=C⊗^KLC_L=C\widehat\otimes_K L: the relative completed tensor with BLB_L identifies with this algebra by multiplying simple tensors, with inverse c⊗l↦c⊗(1⊗l)c\otimes l\mapsto c\otimes(1\otimes l) (NAG-001616, 001688, 001846). The free Gauss section is a right inverse on each domain (001569); unit and joint images agree. The inverse image of the ordinary interior of a germ witness is open. Thus the same joint-image equality gives a germ witness upstairs, including at a point of the absolute boundary. Restrictions, products, both differentials and the Lagerberg slot swap commute with this rule, and p♮u=u∘pp^\natural u=u\circ p (002422–24). Finite module restrictions are the tensors with their canonical norms (001565, 001686, 002414). Purity is preserved by free extension (002291).

For a single unit chart on a presenting piece with algebra CC and restricted module PP, subdivide the common moment image with the boundary images over both fields (002358, 002405). In a top cell, at a relative-interior weight ww, both whole levels are interior. Choose an integral projection QQ injective on the saturated tangent lattice NN. Take the algebra of the whole ww-level cut out by all original equalities of moduli, before applying QQ. Those Laurent conditions base change, and 002306 gives a finite algebra over the exact-radius polytorus at QwQw. If H,H′H,H' are the Gauss fields for that polytorus over K,LK,L, and E,E′E,E' the respective Gauss fibre algebras, 002301 gives E′=E⊗HH′E'=E\otimes_H H' with H↪H′H\hookrightarrow H' isometric. In particular (CL⊗CP)⊗CLE′≃(P⊗CE)⊗HH′.(C_L\otimes_C P)\otimes_{C_L}E' \simeq (P\otimes_C E)\otimes_H H'. The dimension over H′H' equals the original dimension over HH; the weight divides that dimension by the same index [Zd:Q(N)][\mathbb Z^d:Q(N)] over each field (002362). This retains the whole Artinian fibre including its lengths and residue degrees. Lower-dimensional images carry zero top integral. Pairing with the same tangential coefficient and volume proves the single-chart identity. On a finite presentation all actual intersections base change; summing their equalities with the same signs proves (S2) by 002416 and 002425.

For BB integral and β=tf∗α\beta=t_f^*\alpha on all of YY, choose LL nontrivially valued with BLB_L strict by 001840 (identity domain). It is integral of the same analytic dimension by 002291. Apply (B) upstairs and (S2) to the full derivative downstairs: ∣IY,BN(d′(uβ))∣=∣IYL,BLN(d′((u∘p)p♮β))∣≤CYL,p♮β∥u∥C0(Y).(S3)|I^N_{Y,B}(d'(u\beta))| = |I^N_{Y_L,B_L}(d'((u\circ p)p^\natural\beta))| \le C_{Y_L,p^\natural\beta}\|u\|_{C^0(Y)}. \tag{S3} Choosing this one field gives a bound on the fixed original chart. There is no strictness condition on the local presentations of uu. This argument changes the field for the bulk integral, without a comparison of the inverse images of the two absolute boundaries.

Coherent coefficients and finite cuts.

On a compact pure YY, put Yp=M(B/p)Y_{\mathfrak p}=\mathcal M(B/\mathfrak p) and ℓp(M)=length⁡Bp(Mp)\ell_{\mathfrak p}(M)= \operatorname{length}_{B_{\mathfrak p}}(M_{\mathfrak p}) for p∈Min⁡B\mathfrak p\in\operatorname{Min}B. The complete quotient is an integral pure affinoid of dimension dd by 001562 and analytic purity. For every ordinary top section, IY,MN(ω)=∑p∈Min⁡Bℓp(M) IYp, ON(ω∣Yp).(S4)I^N_{Y,M}(\omega)=\sum_{\mathfrak p\in\operatorname{Min}B} \ell_{\mathfrak p}(M)\, I^N_{Y_{\mathfrak p},\,\mathcal O}(\omega|_{Y_{\mathfrak p}}). \tag{S4} Here restriction to a closed component is defined by restricting units and ambient coefficients: inverse images of the presenting neighbourhoods are neighbourhoods, and tangential equality restricts to their joint subimages (001846, 002406, 002422–24).

For the coefficient check, use a finite presentation and a nonempty chart intersection V=M(C)V=\mathcal M(C). On its inverse image in YpY_{\mathfrak p} the algebra is C/pCC/\mathfrak p C, and the tuple is the restricted tuple. Subdivide with all these actual images and all their boundary images. At ww generic in a top cell, every contributing image locally has that same tangent plane, and its whole level is interior. Empty inverse levels give zero. The level with algebra DwD_w in VV is also an affinoid domain of YY by domain transitivity (001609). It is finite over the chosen polytorus algebra RR by 002306. Put E=Dw⊗RHE=D_w\otimes_R H as above. The whole level on the inverse image in YpY_{\mathfrak p} is cut out by the same equalities, hence has algebra Dw/pDwD_w/\mathfrak pD_w and Gauss fibre E/pEE/\mathfrak p E (001565, 001702, 001846). NAG-002360 applied to B→DwB\to D_w gives the exact numerator dim⁡H(M⊗BE)=∑pℓp(M)dim⁡H(E/pE).\dim_H(M\otimes_B E) =\sum_{\mathfrak p}\ell_{\mathfrak p}(M)\dim_H(E/\mathfrak p E). In particular the component lengths refer to the original BB even when the multiplier varies through finer charts. Divide by the same index of Q(N)Q(N) on each contributing image. Integrate the common tangential coefficient and sum on the presenting intersections using 002401, 002405 and 002416. This proves (S4).

For an arbitrary compact affinoid primitive, choose a finite cover by affinoid neighbourhoods whose ordinary interiors cover YY, each carrying one representative of β\beta on the whole domain (002424). On a nonempty intersection YJ=M(BJ)Y_J=\mathcal M(B_J) choose one restricted representative, and put MJ=BJ⊗BMM_J=B_J\otimes_B M. These domains are affinoid and pure (002411, 002413). Apply (S4) to d′(uβ)∣YJd'(u\beta)|_{Y_J} and then (S3) to the restricted single-chart primitive on each integral component. Every restricted multiplier has norm at most ∥u∥C0(Y)\|u\|_{C^0(Y)}. The signed cover formula for this variable top test follows by intersecting the cover with a presenting cover for the top test, applying 002401 on each single chart and interchanging finite sums (002416, 002425). Every nonempty intersection enters, including ones with empty ambient ordinary interior. We get a bound independent of uu by summing the finitely many constants with their nonnegative component lengths. This proves boundedness of T(u)=IY,MN(d′(uβ))T(u)=I^N_{Y,M}(d'(u\beta)) on the entire compact affinoid.

Compact Radon representation within the ordinary calculus.

For F∈C(Y,R)F\in C(Y,\mathbb R) a finite ordinary open cover of oscillation <ϵ<\epsilon and a smooth partition from 002428 approximate FF uniformly by the partition-weighted sum of one value of FF per nonempty open. Thus TT extends uniquely and boundedly to C(Y)C(Y) by taking limits. For continuous f≥0f\ge0 set P(f)=sup⁡0≤g≤fT(g)P(f)=\sup_{0\le g\le f}T(g) over continuous gg. It is finite, nonnegative and positively homogeneous. Split each 0≤g≤f1+f20\le g\le f_1+f_2 as min⁡(g,f1)+(g−min⁡(g,f1))\min(g,f_1)+(g-\min(g,f_1)); in the reverse direction add two approximate maximizers. This gives additivity. Extending by differences is well defined by additivity; both PP and P−TP-T are positive linear functionals.

For each of these two functionals H0H_0 put m(U)=sup⁡{H0(a):a∈C(Y), 0≤a≤1, supp⁡a⊂U},m∗(E)=inf⁡U⊃E, U openm(U).m(U)=\sup\{H_0(a):a\in C(Y),\ 0\le a\le1,\ \operatorname{supp}a\subset U\},\qquad m^*(E)=\inf_{U\supset E,\ U\ {\rm open}}m(U). Then m(Y)=H0(1)m(Y)=H_0(1). For a countable union of opens, a test supported in the union has its compact support covered by finitely many of them. Splitting it by a partition near that support proves subadditivity on opens. Open supersets within errors ϵ2−i\epsilon 2^{-i} give countable subadditivity of m∗m^*; it is monotone, vanishes at the empty set and agrees with mm on opens. For an open UU and an open V⊃EV\supset E, choose a test aa supported on a compact K0⊂U∩VK_0\subset U\cap V with H0(a)>m(U∩V)−ϵH_0(a)>m(U\cap V)-\epsilon. Adding any test on V∖K0V\setminus K_0 keeps the sum at most one. Hence m(V)≥m∗(E∩U)+m∗(E∖U)−ϵm(V)\ge m^*(E\cap U)+m^*(E\setminus U)-\epsilon. Infimizing and using subadditivity proves that UU splits m∗m^* additively on every set. Such splitting sets are closed under complement and finite union by successive splitting. For disjoint splitting sets BiB_i, successive splittings bound m∗(E)m^*(E) below by each partial sum of m∗(E∩Bi)m^*(E\cap B_i) plus m∗(E∖⋃iBi)m^*(E\setminus\bigcup_i B_i). Subadditivity gives splitting for their union and countable additivity. Disjointification gives closure under countable unions. Restriction to Borel sets therefore gives a finite measure ν\nu, outer regular by construction; outer approximation of complements in compact YY gives compact inner approximation for Borel sets.

For 0≤f≤Nδ0\le f\le N\delta, approximate each m({f>jδ})m(\{f>j\delta\}) by tests aja_j, j=1,…,Nj=1,\ldots,N. Pointwise δ∑jaj≤f\delta\sum_j a_j\le f, so H0(f)≥δ∑jν({f>jδ})H_0(f)\ge\delta\sum_j\nu(\{f>j\delta\}). The corresponding simple function differs from ff by at most δ\delta. Let δ\delta tend to zero with Nδ≥max⁡fN\delta\ge\max f. Apply the resulting lower bound also to max⁡f−f\max f-f, using ν(Y)=H0(1)\nu(Y)=H_0(1), to obtain H0(f)=∫f dνH_0(f)=\int f\,d\nu. The difference of the measures for PP and P−TP-T represents TT.

We use the following uniqueness and localization test here and below. If a finite signed Radon measure annihilates smooth scalars compactly supported in an ordinary open UU, it vanishes there. For compact C⊂UC\subset U take an open C⊂V⊂UC\subset V\subset U with variation on V∖CV\setminus C as small as desired. A cutoff between zero and one supported compactly in VV and equal to one near CC exists by 002428. Its zero integral bounds the absolute measure of CC by that variation. Compact inner approximation then gives zero on Borel subsets of UU. A finite sum of the positive Radon measures in the construction dominates the variation, so the same regular approximations apply to their signed differences. This proves uniqueness. Tests supported outside supp⁡β\operatorname{supp}\beta give zero. For a scalar compactly supported in Int⁡(Y/K)\operatorname{Int}(Y/K), the product primitive is compactly supported there, and NAG-002577 annihilates its derivative. We have proved (S1) and the asserted support on YY. In particular (S2) and uniqueness identify the compact scalar-extension comparison on measures with p∗μYL,ML,p♮βd′=μY,M,βd′p_*\mu_{Y_L,M_L,p^\natural\beta}^{d'}=\mu_{Y,M,\beta}^{d'}: the pushforward is Radon as the map is continuous between compact Hausdorff spaces. This comparison needs no boundary-fibre equality.

Ordinary localization and the closed-cut formula.

On XX take finitely many compact affinoid neighbourhoods YiY_i whose ordinary interiors cover SS, and smooth cutoffs χi\chi_i compactly supported in those interiors, summing to one near SS (002428). Goodness gives these neighbourhoods, purity holds by 002413, and 002414 gives finite coefficient modules on all their intersections. With ιi:Yi↪X\iota_i:Y_i\hookrightarrow X, set μ=∑i(ιi)∗((χi∣Yi) μYi,F∣Yi, β∣Yid′).\mu=\sum_i(\iota_i)_*\bigl((\chi_i|_{Y_i})\, \mu_{Y_i,\mathcal F|_{Y_i},\,\beta|_{Y_i}}^{d'}\bigr). Compact-domain pushforwards are finite Radon measures: compact inner approximants have compact images, and outer regularity follows by approximating the complement from inside. Formula (S1) on YiY_i, tested on uχiu\chi_i, gives the integral of the derivative of the entire product. Its compact collar in the ordinary interior makes both derivative and integral compatible with zero extension by 002580. The sum equals IX,FN(d′(uβ))I^N_{X,\mathcal F}(d'(u\beta)) since ∑iχi=1\sum_i\chi_i=1 near SS. The uniqueness test on ordinary opens now proves independence and gives support in S∩∂KXS\cap\partial_KX: outside SS the product vanishes and on the analytic interior use 002577 on compactly supported scalar multiples. The ambient space itself need not be compact or paracompact.

In particular, on an ordinary open UU the restriction of μ\mu is characterized on compactly supported smooth scalars aa there by IU,F∣UN(d′(aβ∣U))I^N_{U,\mathcal F|_U}(d'(a\beta|_U)). Indeed aa extends by zero and both integrals use the same carrier inside UU by 002425. This also works on the ambient ordinary interior of a closed affinoid domain. More generally, for a finite ordinary cover of SS, local primitives with germ supports in S∩UiS\cap U_i have a common measure supported in SS precisely when their evaluations on compact smooth overlap scalars agree after the two zero extensions. Necessity is restriction. For sufficiency take the sum of the zero extensions of the measures of χiβi\chi_i\beta_i on UiU_i. Each product is compactly supported there. Testing against a scalar compactly supported in UkU_k, the products of that scalar with χi\chi_i are compactly supported in the overlaps; replace their local evaluations by the ones on UkU_k and sum. This recovers the evaluation on UkU_k. Uniqueness on each open proves the assertion. True boundary support follows as above locally: an ordinary open has the same structure-map interior on that open (001631). Equality of ordinary primitive germs on the overlaps suffices for the evaluation condition by the common-carrier argument.

There is also an expression without cutoffs. For any finite such closed affinoid cover of the support by ordinary interiors write DJ=⋂j∈JYj, BJ=O(DJ)D_J=\bigcap_{j\in J}Y_j,\ B_J=\mathcal O(D_J), MJ=Γ(DJ,F)M_J=\Gamma(D_J,\mathcal F), and write iJpi_{Jp} for the inclusion of M(BJ/p)\mathcal M(B_J/\mathfrak p) into XX. Then μX,F,βd′=∑∅≠J, DJ≠∅(−1)∣J∣+1∑p∈Min⁡BJℓp(MJ)(iJp)∗μM(BJ/p), O, iJp∗βd′.(S5)\mu_{X,\mathcal F,\beta}^{d'}= \sum_{\emptyset\ne J,\ D_J\ne\varnothing}(-1)^{|J|+1} \sum_{\mathfrak p\in\operatorname{Min}B_J} \ell_{\mathfrak p}(M_J)(i_{Jp})_* \mu_{\mathcal M(B_J/\mathfrak p),\,\mathcal O,\,i_{Jp}^*\beta}^{d'} . \tag{S5} Testing against uu, apply (S4) on each intersection and the finite cover formula on their compact union (002416, 002425). This yields (S1), hence equality by uniqueness. Each cut has its own absolute boundary, potentially including an artificial frontier. The support bound on XX is for the signed sum. Neither raw restriction to a closed overlap nor addition of the total variations is used.

Finally let JJ be the native Lagerberg involution. It fixes scalars, interchanges d′d' and d′′d'', and acts on interleaved top degree by (−1)d(-1)^d, since each of the dd pairs is reversed (002388, 002424, 002577). For a compact primitive γ\gamma of type (d,d−1)(d,d-1) this gives μX,F,γd′′=(−1)dμX,F,Jγd′.\mu_{X,\mathcal F,\gamma}^{d''} =(-1)^d\mu_{X,\mathcal F,J\gamma}^{d'} . Thus (S1) uses the derivative of uγu\gamma with its own plus convention, and the signed sum and localization statements follow in this bidegree as well.

A uniform scalar bound from the fixed moment image

Let KK be complete and nontrivially valued for its specified real non-Archimedean absolute value. Let Z=M(A)Z=\mathcal M(A) be a nonempty integral strict affinoid of analytic dimension d≥1d\ge1, with structure coefficient. Fix analytic units f=(f1,…,fn)f=(f_1,\ldots,f_n), t=tft=t_f with coordinates −log⁡∣fi∣-\log|f_i|, and a smooth real ambient form α\alpha of type (d−1,d)(d-1,d) on a neighbourhood of P=t(Z)P=t(Z). Write θ=t∗α\theta=t^*\alpha, and II for the native ordinary top integral. Then there exists C<∞C<\infty, independent of every chart and cut used for the scalar, such that ∣I(d′(uθ))∣≤C∥u∥C0(Z)(u∈AZ0,0(Z)).(B)|I(d'(u\theta))|\le C\|u\|_{C^0(Z)} \qquad (u\in\mathcal A^{0,0}_Z(Z)). \tag{B} The proof estimates a positive bulk integral in the fixed chart. No boundary measure or lifting of boundary sheets is a premise.

A cutoff in the original target.

Put H=t(∂KZ)H=t(\partial_K Z). By NAG-002358 and 002405, PP is compact finite polyhedral of dimension at most dd, HH is finite polyhedral of dimension at most d−1d-1, and the structure chain has fixed finite nonnegative weights mσm_\sigma on its dd-cells. If HH is empty, (B) follows with C=0C=0 from NAG-002577. If dim⁡P<d\dim P<d, the entire second slot of θ\theta vanishes tangentially, giving the same conclusion. Take a finite subdivision of PP adapted to HH.

For each cell of HH, choose a real affine hyperplane ℓj=0\ell_j=0 containing it and containing none of the affine spans of the dd-cells of PP. Such choices exist: the space of affine functions vanishing on the span of that cell is not contained in the subspace of functions vanishing on any specified dd-plane, by the dimension inequality. A finite union of proper linear subspaces cannot cover a real vector space (choose a nonzero linear equation for each and take their product, a nonzero polynomial). Use a smooth 0≤b≤10\le b\le1 equal to one on [−1,1][-1,1] and zero outside (−2,2)(-2,2), from the elementary smooth profile in the proof of NAG-002427. For 0<ϵ<10<\epsilon<1, put χϵ=1−∏j(1−b(ℓj/ϵ)),ψϵ=∑jb(ℓj/(2ϵ)).\chi_\epsilon=1-\prod_j(1-b(\ell_j/\epsilon)),\qquad \psi_\epsilon=\sum_j b(\ell_j/(2\epsilon)). Then 0≤χϵ≤10\le\chi_\epsilon\le1, it is one on an open neighbourhood of HH, and it vanishes off the union of the slabs ∣ℓj∣≤2ϵ|\ell_j|\le2\epsilon. On the support of its derivative ψϵ≥1\psi_\epsilon\ge1 and ∥∇χϵ∥≤c/ϵ\|\nabla\chi_\epsilon\|\le c/\epsilon, with cc fixed. Smoothness here is in real moment coordinates; the hyperplanes do not define new analytic domains.

For every coarse dd-cell σ\sigma and every jj, the lattice volume in σ\sigma of {∣ℓj∣≤4ϵ}\{|\ell_j|\le4\epsilon\} is at most cσjϵc_{\sigma j}\epsilon for sufficiently small ϵ\epsilon. If the restriction is constant nonzero the intersection is empty eventually. Otherwise in integral affine coordinates on the cell span some coordinate has nonzero coefficient aa in ℓj\ell_j. Enclose the compact cell in a fixed box and slice parallel to that axis. Each slice has length at most 8ϵ/∣a∣8\epsilon/|a|; the remaining box has fixed volume (NAG-002389). This proof applies equally on a compact fine dd-cell whenever its projection has rank dd and lands in σ\sigma, with a constant then allowed to depend on the fine cell.

A positive majorant independent of the multiplier.

Order the ambient coordinates once and put Ω=∑i1<⋯<id⋀k=1d(d′xik∧d′′xik).\Omega=\sum_{i_1<\cdots<i_d} \bigwedge_{k=1}^d(d'x_{i_k}\wedge d''x_{i_k}). In lattice coordinates on a dd-plane its interleaved top coefficient is the sum of the squares of its coordinate dd-minors, which is strictly positive. Use aσ>0a_\sigma>0 for this constant on aff⁡(σ)\operatorname{aff}(\sigma), and ωσ\omega_\sigma for the interleaved top generator there. On σ\sigma write d′χϵ∧α=qϵ,σωσd'\chi_\epsilon\wedge\alpha=q_{\epsilon,\sigma}\omega_\sigma. Compactness, the gradient bound and the finitely many coarse tangent planes give one c1c_1 such that ∣qϵ,σ∣≤(c1/ϵ)ψϵaσ.|q_{\epsilon,\sigma}|\le(c_1/\epsilon)\psi_\epsilon a_\sigma. Indeed qϵ,σq_{\epsilon,\sigma} is linear in the first derivatives of χϵ\chi_\epsilon with bounded coefficients, and vanishes off their support. Define Λϵ=(c1/ϵ)ψϵΩ\Lambda_\epsilon=(c_1/\epsilon)\psi_\epsilon\Omega. By the coarse chart integral (NAG-002405) and the slab estimate, I(t∗Λϵ)=c1ϵ∑σmσaσ∫σψϵ dvolσ≤c1∑σ,jmσaσcσj=:C.(1)I(t^*\Lambda_\epsilon) ={c_1\over\epsilon}\sum_\sigma m_\sigma a_\sigma \int_\sigma\psi_\epsilon\,d\mathrm{vol}_\sigma \le c_1\sum_{\sigma,j}m_\sigma a_\sigma c_{\sigma j}=:C. \tag{1}

We claim both top sections ∥u∥∞ t∗Λϵ ± u t∗(d′χϵ∧α)\|u\|_\infty\,t^*\Lambda_\epsilon \ \pm\ u\,t^*(d'\chi_\epsilon\wedge\alpha) are positive in the ordinary native sense. At each point use an ordinary affinoid scalar representative u=v(tg)u=v(t_g) and adjoin ff. Subdivide the joint image so projection π\pi to the ff-coordinates maps each cell into a coarse cell. If π\pi on a dd-cell has rank <d<d, both pulled-back tops vanish. Otherwise the two coarse top coefficients are pulled back with the same squared determinant in integral coordinates. This is positive whatever the orientation. Use ∣v(tg)∣≤∥u∥∞|v(t_g)|\le\|u\|_\infty on the actual joint image and the preceding coefficient inequality. This proves positivity on all local representatives so chosen. NAG-002425, whose positivity includes varying finite domain presentations, now gives ∣I(u t∗(d′χϵ∧α))∣≤C∥u∥∞.(2)|I(u\,t^*(d'\chi_\epsilon\wedge\alpha))| \le C\|u\|_\infty. \tag{2} It is essential to take this inequality for the ordinary integral. Applying an absolute value separately to all signed terms of an arbitrary closed cover would not give (1).

The term containing the derivative of the multiplier.

For each fixed uu, I((χϵ∘t)d′(uθ))→0.(3)I((\chi_\epsilon\circ t)d'(u\theta))\to0. \tag{3} Choose once for this uu a finite cover by ordinary affinoid presentation neighbourhoods, with their interiors covering ZZ. On every actual nonempty closed intersection use one restricted scalar presentation and its joint tuple with ff. The restriction and finite-cover formulae NAG-002416, 002424–25 allow the same finite family for all ϵ\epsilon. On its joint image the top representative before multiplication by χϵ∘π\chi_\epsilon\circ\pi is d′(v π∗α)=d′v∧π∗α+v π∗(d′α),d'(v\,\pi^*\alpha)=d'v\wedge\pi^*\alpha+v\,\pi^*(d'\alpha), where vv is composed with the scalar-coordinate projection. On each dd-cell on which π\pi has rank <d<d, this is zero: all dd double-primed covectors in both terms come from the coarse target. On each remaining cell the coefficient is bounded on the compact polytope for this fixed presentation. The portion on which χϵ∘π≠0\chi_\epsilon\circ\pi\ne0 has lattice volume O(ϵ)O(\epsilon) by the fine-cell slab estimate above. All weights are finite and fixed for this uu, and the finite signed sum tends to zero. This argument requires no uniform control on the derivatives, weights or number of charts of uu.

Finally (1−χϵ∘t)uθ(1-\chi_\epsilon\circ t)u\theta is identically zero on the inverse image of an ordinary neighbourhood of HH. Its support is compact in Int⁡(Z/K)\operatorname{Int}(Z/K); hence NAG-002577 gives I(d′((1−χϵ∘t)uθ))=0I(d'((1-\chi_\epsilon\circ t)u\theta))=0. The product rule with the scalar in first position yields I(d′(uθ))=I(u t∗(d′χϵ∧α))+I((χϵ∘t)d′(uθ)).I(d'(u\theta)) = I(u\,t^*(d'\chi_\epsilon\wedge\alpha)) + I((\chi_\epsilon\circ t)d'(u\theta)). Take ϵ↓0\epsilon\downarrow0 in (2)–(3). This proves (B) with the constant in (1), for multipliers with arbitrary ordinary local charts.

Normal and multiplicity check.

For comparison, on a single global joint tuple, take subdivisions adapted to the projection and boundary images. For an incident fine cell–facet flag mapping with full ranks to a coarse flag, let ee be the tangent-lattice index on the facet. Modulo the coarse facet lattice the primitive inward normal maps to kk times the coarse primitive inward normal, with k>0k>0. In bases completed by these normals the full index is ekek (block triangular determinant). In the normal-contraction formula of NAG-002389 the two lists of facet tangent vectors give e2e^2; the remaining normal gives kk, with its tangent component killed by alternation. Scalar facet volume changes by 1/e1/e upon passage from fine to coarse, leaving ekek times the original fine weight, with the common inward convention (−1)d(d+1)/2(-1)^{d(d+1)/2}. For a facet mapping to smaller facet dimension the tangent lists lose rank, and for a contracted incident dd-cell the entire second slot loses rank. Each such flag has zero contraction of the pulled-back primitive. For d=1d=1 the empty facet tangent lists have index one. The generic pushforward of top cells in NAG-002405 counts their full weights times the full index: NAG-002306 and 002362 take the whole Gauss fibre degree with all local lengths and residue degrees and divide by the chosen projection index. Formula (1) uses those resulting coarse weights once. A coarse signed normal can vanish even when the individual incidences have nonzero absolute contractions. The bulk majorant avoids any identification of individual boundary sheets or any assumption that absolute normal masses commute with projection.

A Green identity and its use

Let u,vu,v be smooth real functions, and let TT be a closed current of suitable degree. Assume all products under consideration have compact support away from the analytic boundary. Applying the two Leibniz rules and Stokes yields the integration-by-parts symmetry ∫u d′d′′v∧T=∫v d′d′′u∧T\int u\,\mathrm d'\mathrm d''v\wedge T=\int v\,\mathrm d'\mathrm d''u\wedge T when the remaining degrees make these top-degree integrals. A convenient sufficient support assumption is that one factor is compactly supported and the other is defined on a neighbourhood of that support. More general formulations use cutoffs.

The identity is the engine behind several later results: it moves derivatives away from an approximating potential, proves symmetry of mixed expressions, and gives estimates in terms of the sup norm of the potential. However, its use with a nonsmooth potential requires a limiting argument. The formula itself does not authorize a product that has not yet been defined.

Example. (Boundary terms in Green’s identity). For u,v∈C2([a,b])u,v\in C^2([a,b]), ∫ab(uv′′−vu′′) dt=[uv′−vu′]ab.\int_a^b(uv''-vu'')\,dt=[uv'-vu']_a^b. For u=1u=1 and v=t2v=t^2, this is 2(b−a)2(b-a), which is nonzero when a<ba<b. Compactness does not remove boundary terms. If one factor is compactly supported in (a,b)(a,b), it and its first derivative vanish near the endpoints, and the displayed boundary term is zero. This is the support mechanism in the current identity.

References. [CLD12] (§§3.3–3.12); the algebraic presentation and integration comparison in [Gubler].

The signed measure of an ordinary top form

Proposition NAG-002970. NAG-002970 (Native top-form measures; F8CR-M). Let YY be good, Hausdorff, and pure of dimension dd. For every real ordinary compact (d,d)(d,d)-form ω\omega, there is a unique finite signed Radon measure μY,ω\mu_{Y,\omega}, supported in S=Supp⁡ωS=\operatorname{Supp}\omega, with ∫Yϕ dμY,ω=∫Yϕωfor every ordinary smooth scalar ϕ.\begin{equation} \int_Y\phi\,d\mu_{Y,\omega}=\int_Y\phi\omega \quad\text{for every ordinary smooth scalar }\phi. \end{equation} The construction is linear in ω\omega and positive for positive ω\omega. If its compact support is contained in an open subspace, the measure on YY is the zero extension of the measure there. It defines a separate Borel-measure extension; it does not enlarge the class of coefficients admitted by the frozen grid integral. Empty support gives the zero measure.

Proof. We spell out the boundedness and representation steps.

Density and cutoffs on a compact set. For h∈C(S,R)h\in C(S,\mathbb R) and ϵ>0\epsilon>0, take finitely many ordinary open sets covering SS on each of which the oscillation of hh on SS is less than ϵ\epsilon. Choose a point xjx_j of each nonempty piece. By NAG-002428 there are nonnegative compact ordinary smooth χj\chi_j subordinate to these sets, summing to one near SS. Then ∑jh(xj)χj\sum_jh(x_j)\chi_j approximates hh uniformly on SS to error ϵ\epsilon, and is nonnegative on all of YY when h≥0h\geq0. Thus ordinary smooth scalar restrictions are dense in C(S)C(S). Moreover for a compact F⊂U⊂SF\subset U\subset S with UU relatively open, there is a continuous function on SS equal to one on FF, between zero and one, and with support contained in UU. Lift UU to an open set of YY, use finitely many point cutoffs from NAG-002427, and take 1−∏j(1−χj)1-\prod_j(1-\chi_j). Finite partitions on compact supports follow from the same frozen partition theorem restricted to SS.

A positive domination. Choose finitely many moment representatives αj\alpha_j near the compact support and a nonnegative ordinary partition ρj\rho_j with compact supports inside those chart opens, summing to one near SS. In the ambient real coordinates t1,…,trt_1,\ldots,t_r of a chart, put Θj=1d!(∑k=1rd′tk∧d′′tk)d.\Theta_j=\frac1{d!}\left(\sum_{k=1}^r d't_k\wedge d''t_k\right)^d. For d=0d=0 take Θj=1\Theta_j=1. On a dd-plane the coefficient of Θj\Theta_j is the Euclidean Gram square and is positive. Replace αj\alpha_j by (αj+(−1)dJαj)/2(\alpha_j+(-1)^dJ\alpha_j)/2; this does not change its restriction to any dd-plane, since the top generator has JJ-sign (−1)d(-1)^d. On the compact moment image of Supp⁡ρj\operatorname{Supp}\rho_j, its coefficient matrix on ⋀dRr\bigwedge^d\mathbb R^r is bounded. The elementary finite-dimensional inequality ∣vTCv∣≤cj∣v∣2|v^TCv|\leq c_j|v|^2, for a sufficiently large constant cjc_j, therefore shows that cjΘj±αjc_j\Theta_j\pm\alpha_j are nonnegative on all dd-planes there. This remains true on the tangent planes of every joint moment refinement, including rank-deficient ones. If r<dr<d, both restrictions are zero.

The zero extensions of ρjcjtg∗Θj\rho_jc_jt_g^*\Theta_j define a positive compact form θ\theta satisfying −θ≤ω≤θ-\theta\leq\omega\leq\theta. For M=sup⁡S∣ϕ∣M=\sup_S|\phi|, the forms Mθ±ϕωM\theta\pm\phi\omega are positive: on SS use the displayed bound and outside SS use ω=0\omega=0. Positivity of NAG-002425 gives ∣∫Yϕω∣≤Cωsup⁡S∣ϕ∣,Cω=∫Yθ<∞.\begin{equation} \left|\int_Y\phi\omega\right|\leq C_\omega\sup_S|\phi|, \qquad C_\omega=\int_Y\theta<\infty. \end{equation} Hence the native functional depends only on ϕ∣S\phi|_S and extends uniquely by uniform limits to a bounded linear functional Λω:C(S)→R\Lambda_\omega:C(S)\to\mathbb R. For positive ω\omega the positive approximants above prove that this extension is positive.

Representation of a positive functional, with the construction. For any positive bounded Λ:C(S)→R\Lambda:C(S)\to\mathbb R, define, for open U⊂SU\subset S and arbitrary E⊂SE\subset S, a(U)=sup⁡{Λ(g):g∈C(S), 0≤g≤1, Supp⁡g⊂U},a∗(E)=inf⁡{a(U):E⊂U, U open in S}.\begin{align*} a(U)&=\sup\{\Lambda(g):g\in C(S),\ 0\leq g\leq1, \ \operatorname{Supp}g\subset U\},\\ a^*(E)&=\inf\{a(U):E\subset U,\ U\text{ open in }S\}. \end{align*} Here a(S)=Λ(1)a(S)=\Lambda(1), and every quantity is finite. If an open set is covered by countably many open sets, the compact support of any test gg is covered by finitely many of them. A finite partition on that support splits gg into admissible summands, proving countable subadditivity for aa on open sets. Choosing open supersets with errors ϵ2−j\epsilon2^{-j} proves that a∗a^* is an outer measure; it agrees with aa on open sets.

Every open UU is measurable for this outer measure. Indeed, fix EE and choose open V⊃EV\supset E with a(V)≤a∗(E)+ϵa(V)\leq a^*(E)+\epsilon. Choose a test gg supported on a compact F⊂V∩UF\subset V\cap U with Λ(g)≥a(V∩U)−ϵ\Lambda(g)\geq a(V\cap U)-\epsilon. For a test hh supported in V∖FV\setminus F, the function g+hg+h is still between zero and one and supported in VV. Thus a(V∖F)≤a(V)−Λ(g)a(V\setminus F)\leq a(V)-\Lambda(g), and a∗(E∩U)+a∗(E∖U)≤a(V∩U)+a(V∖F)≤a∗(E)+2ϵ.a^*(E\cap U)+a^*(E\setminus U) \leq a(V\cap U)+a(V\setminus F) \leq a^*(E)+2\epsilon. Outer subadditivity supplies the reverse inequality. Let ϵ↓0\epsilon\downarrow0. The sets satisfying this splitting equality form a sigma-algebra. Complements are immediate. For disjoint measurable EjE_j, finite iteration gives, for every HH, a∗(H)≥∑j=1Na∗(H∩Ej)+a∗(H∖⋃j≥1Ej).a^*(H)\geq\sum_{j=1}^N a^*(H\cap E_j) +a^*\left(H\setminus\bigcup_{j\geq1}E_j\right). Pass to the supremum over NN and use outer subadditivity on H∩⋃EjH\cap\bigcup E_j to obtain measurability of the union. General countable unions reduce to disjoint ones by differences. The same finite-iteration inequality and outer subadditivity prove countable additivity of the restricted outer measure. Hence μ=a∗∣Borel(S)\mu=a^*|_{\text{Borel}(S)} is a finite Borel measure.

For a test gg above, Λ(g)≤μ(Supp⁡g)\Lambda(g)\leq\mu(\operatorname{Supp}g), since every open set containing that support admits gg in its defining supremum. It follows that the measure of an open set is the supremum of the measures of its compact subsets. Outer regularity is built into a∗a^*. Inner regularity for every Borel set follows by taking complements and using finiteness of μ(S)\mu(S). Thus μ\mu is Radon. Integrals of nonnegative Borel functions are defined by the supremum over dominated nonnegative simple functions; bounded functions may equivalently be approximated uniformly by their finite dyadic step functions.

To verify representation without importing it, let u≥0u\geq0 be continuous, 0≤u≤mδ0\leq u\leq m\delta. Tests supported in {u>kδ}\{u>k\delta\}, k=1,…,mk=1,\ldots,m, give δ∑k=1mμ{u>kδ}≤Λ(u).\delta\sum_{k=1}^m\mu\{u>k\delta\}\leq\Lambda(u). For the upper bound put vk=min⁡(1,max⁡(0,(u−kδ)/δ))v_k=\min(1,\max(0,(u-k\delta)/\delta)), k=0,…,m−1k=0,\ldots,m-1. Then u=δ∑vku=\delta\sum v_k, and for k≥1k\geq1 the support of vkv_k is contained in {u>(k−1)δ}\{u>(k-1)\delta\}. Consequently Λ(u)≤δμ(S)+δ∑k=1m−1μ{u>(k−1)δ}.\Lambda(u)\leq\delta\mu(S) +\delta\sum_{k=1}^{m-1}\mu\{u>(k-1)\delta\}. The lower simple sum differs from uu by at most δ\delta, and the upper one by at most 2δ2\delta. Let δ→0\delta\to0, choosing mδ≥∥u∥∞m\delta\geq\|u\|_\infty. Both bounds tend to ∫u dμ\int u\,d\mu. Linearity gives Λ(u)=∫u dμ\Lambda(u)=\int u\,d\mu for all continuous uu. Uniqueness follows from the cutoff formula for open sets and Radon inner regularity.

Signed functionals and locality. For a bounded real Λ\Lambda, define on u≥0u\geq0 Λ+(u)=sup⁡0≤v≤uΛ(v).\Lambda^+(u)=\sup_{0\leq v\leq u}\Lambda(v). This is finite, nonnegative, positively homogeneous, and additive: for one inequality split any v≤u1+u2v\leq u_1+u_2 into min⁡(v,u1)\min(v,u_1) and v−min⁡(v,u1)v-\min(v,u_1); for the other add two nearly maximizing choices. Extend linearly to all of C(S)C(S). It is bounded, since Λ+(u)≤∥Λ∥∥u∥∞\Lambda^+(u)\leq\|\Lambda\|\|u\|_\infty for u≥0u\geq0. The functional Λ−:=Λ+−Λ\Lambda^-:=\Lambda^+-\Lambda is positive and bounded. The positive construction gives μ=μ+−μ−\mu=\mu^+-\mu^-. Any two finite signed Radon representations are equal by comparing the positive measures in their two decompositions and using positive uniqueness. Apply this construction to Λω\Lambda_\omega on SS, then extend by zero along the closed inclusion S↪YS\hookrightarrow Y. This yields (8.4.1). The same uniqueness, density on a compact union of supports, and locality of the frozen integral prove linearity and compatibility with open restrictions wherever the form has compact support.

Cutoffs and the order-zero bound (lines 296–348).

For compact F⊂O⊂SF\subset O\subset S with OO relatively open, lift OO to an ordinary open of YY. Finitely many point cutoffs there give 1−∏j(1−χj)1-\prod_j(1-\chi_j) restricted to SS, equal to one on FF and supported in OO. More generally a finite relatively open cover of FF lifts to ambient opens; apply 002428 on FF and restrict. For u∈C(S)u\in C(S), a finite such cover of SS by sets of oscillation less than ϵ\epsilon gives an approximant ∑ju(xj)χj\sum_j u(x_j)\chi_j of uniform error at most ϵ\epsilon. If u≥0u\ge0 the approximant is nonnegative. No continuous extension of uu to YY is required.

In each moment presentation on a neighbourhood of SS, use the interleaved form Θ=(∑kd′tk∧d′′tk)d/d!\Theta=(\sum_k d't_k\wedge d''t_k)^d/d!. In interleaved coordinates a tangent dd-frame with Plücker vector vv evaluates Θ\Theta to ∑IvI2\sum_I v_I^2 and the given coefficient matrix to vtAvv^{\mathsf t}A v (absorbing the fixed coordinate sign in AA). The latter has absolute value bounded by c∑IvI2c\sum_I v_I^2 on the compact chart region used by its cutoff. Projection of a frame under refinement uses the same calculation, including v=0v=0. In dimension zero take Θ=1\Theta=1. With the finite nonnegative partition, sum the cutoff multiples of cΘc\Theta, extended by zero, to get a compact positive form θ\theta bounding ±ω\pm\omega.

Here is a way to use precisely M=sup⁡S∣ϕ∣M=\sup_S|\phi| even if ϕ\phi is much larger just outside SS. For ϵ>0\epsilon>0 choose a smooth real scalar truncation β\beta equal to the identity on [−M−ϵ/2,M+ϵ/2][-M-\epsilon/2,M+\epsilon/2] and bounded in modulus by M+ϵM+\epsilon. For example multiply the identity by the cutoff profile of 002427 applied to 2(M+ϵ−∣t∣)/ϵ2(M+\epsilon-|t|)/\epsilon; this factor is constant near zero. Then β(ϕ)ω=ϕω\beta(\phi)\omega=\phi\omega as sections. Nonnegativity in common charts and 002425 give ∣∫Yϕω∣≤(M+ϵ)∫Yθ.\left|\int_Y\phi\omega\right| \le (M+\epsilon)\int_Y\theta. Let ϵ\epsilon decrease to zero. For empty SS take the zero measure. Otherwise the estimate and the approximants above extend the functional uniquely to a bounded linear Λ\Lambda on C(S)C(S): use limits of the real integrals of uniform approximants. Positivity is preserved for positive ω\omega.

Positive representation (lines 350–425).

For Λ≥0\Lambda\ge0 let a(O)a(O) be the supremum of Λ(g)\Lambda(g) over 0≤g≤10\le g\le1 continuous with support in an open O⊂SO\subset S, and put m(H)=inf⁡O⊃Ha(O)m(H)=\inf_{O\supset H}a(O) over open supersets. An open cover of supp⁡g\operatorname{supp}g admits a finite subordinate partition there as above. Splitting gg by the partition gives countable subadditivity of aa on open covers. Taking open supersets of HjH_j within ϵ2−j\epsilon 2^{-j} of m(Hj)m(H_j) gives countable subadditivity of mm. It is monotone, m(∅)=0m(\varnothing)=0, m(O)=a(O)m(O)=a(O) for open OO, and m(S)=Λ(1)<∞m(S)=\Lambda(1)<\infty. For arbitrary HH and open OO, choose open V⊃HV\supset H within ϵ\epsilon of m(H)m(H) and an admissible gg in V∩OV\cap O within ϵ\epsilon of a(V∩O)a(V\cap O). Put F=supp⁡gF=\operatorname{supp}g. Any admissible test in V∖FV\setminus F can be added to gg without exceeding one. Thus m(H∩O)+m(H∖O)≤m(H)+2ϵm(H\cap O)+m(H\setminus O)\le m(H)+2\epsilon. Subadditivity gives the reverse inequality after passage to the limit.

The finite-intersection step implicit at lines 379–392 is as follows. Call AA splitting if m(H)=m(H∩A)+m(H∖A)m(H)=m(H\cap A)+m(H\setminus A) for every HH. Complements split. For splitting A,BA,B, split HH by AA, then H∩AH\cap A by BB. The last two pieces (H∩A)∖B(H\cap A)\setminus B and H∖AH\setminus A have union H∖(A∩B)H\setminus(A\cap B). Subadditivity yields the splitting inequality for A∩BA\cap B, and the opposite inequality always holds. Hence splitting sets form an algebra. For disjoint splitting AjA_j, A=⋃jAjA=\bigcup_j A_j, finite iteration and monotonicity give m(H) ≥ ∑j≤Nm(H∩Aj)+m(H∖A).m(H)\ \ge\ \sum_{j\le N}m(H\cap A_j)+m(H\setminus A). Taking the supremum over NN and using countable subadditivity on H∩AH\cap A proves AA splits. Finite differences in the algebra now disjointify any countable union. Taking H=AH=A gives countable additivity on disjoint splitting sets by the same inequalities. Since all opens split, μ=m∣Borel\mu=m|_{\rm Borel} is a finite measure.

For an admissible gg, Λ(g)≤m(supp⁡g)\Lambda(g)\le m(\operatorname{supp}g) by taking the infimum over open supersets of its support. This gives inner regularity on opens; outer regularity is by construction. For any Borel AA, an open outer approximation to S∖AS\setminus A has compact complement inside AA, giving inner regularity there too. To check representation, fix u≥0u\ge0 continuous, 0≤u≤m0δ0\le u\le m_0\delta. Put Ok={u>kδ}O_k=\{u>k\delta\}. For 1≤k≤m01\le k\le m_0 choose near-maximizing tests gkg_k in OkO_k. The pointwise inequality δ∑kgk≤u\delta\sum_k g_k\le u gives the lower sum δ∑k=1m0μ(Ok)≤Λ(u)\delta\sum_{k=1}^{m_0}\mu(O_k)\le\Lambda(u) by finite approximation. For the upper sum, u=δ∑k=0m0−1min⁡(1,max⁡(0,u/δ−k))u=\delta\sum_{k=0}^{m_0-1} \min(1,\max(0,u/\delta-k)). For k≥1k\ge1 each summand has support in Ok−1O_{k-1}. Thus the upper sum is δμ(S)+δ∑k=1m0−1μ(Ok−1)\delta\mu(S)+\delta\sum_{k=1}^{m_0-1}\mu(O_{k-1}). The two simple functions approximate uu uniformly to within 2δ2\delta. Finiteness gives Λ(u)=∫u dμ\Lambda(u)=\int u\,d\mu; differences give the identity on C(S)C(S). Any positive Radon representation λ\lambda satisfies λ(O)=a(O)\lambda(O)=a(O): use compact inner approximation and a cutoff equal to one on each chosen compact. Outer regularity then gives uniqueness on Borel sets.

Signed measures and locality (lines 427–446).

For arbitrary bounded real Λ\Lambda use the source construction Λ+(u)=sup⁡0≤v≤uΛ(v)\Lambda^+(u)=\sup_{0\le v\le u}\Lambda(v) on continuous u≥0u\ge0. Splitting v≤u1+u2v\le u_1+u_2 into min⁡(v,u1)\min(v,u_1) and the remainder, and adding nearly maximizing tests in the other direction, prove additivity. Positive homogeneity gives a linear extension by differences, since u−v=u′−v′u-v=u'-v' implies u+v′=u′+vu+v'=u'+v. Both Λ+\Lambda^+ and Λ−=Λ+−Λ\Lambda^-=\Lambda^+-\Lambda are positive and bounded; for the first use ∣Λ+(w)∣≤Λ+(∣w∣)≤∥Λ∥∥w∥∞|\Lambda^+(w)|\le\Lambda^+(|w|)\le\|\Lambda\|\|w\|_\infty. Let μ±\mu_\pm be their positive Radon representatives. Then σ=μ+−μ−\sigma=\mu_+-\mu_- represents Λ\Lambda.

For conventions requiring regular total variation, the following argument is internal to this proof. Put ν=μ++μ−\nu=\mu_++\mu_-. It is regular: for any Borel AA, take the union of two compact inner approximants and the intersection of two open outer approximants. Define v(A)v(A) by the supremum of ∑i∣σ(Ai)∣\sum_i|\sigma(A_i)| over finite Borel partitions of AA. Then v≤νv\le\nu. On disjoint Borel AjA_j, intersecting any finite partition of their union with each AjA_j gives v(⋃jAj)≤∑jv(Aj)v(\bigcup_j A_j)\le\sum_jv(A_j) by countable additivity of σ\sigma and the triangle inequality. Conversely combine near-optimal finite partitions on the first NN pieces, adding the remainder as one block; take the supremum and then let NN increase. Thus vv is a finite measure. Compact and open approximants F⊂A⊂OF\subset A\subset O with ν(O∖F)<ϵ\nu(O\setminus F)<\epsilon give v(O∖F)<ϵv(O\setminus F)<\epsilon, so v=∣σ∣v=|\sigma| is Radon. Conversely, any candidate signed measure with regular finite variation decomposes as (∣σ∣+σ)/2−(∣σ∣−σ)/2(|\sigma|+\sigma)/2-(|\sigma|-\sigma)/2 into positive measures dominated by its variation and hence regular by the same estimate. Signed uniqueness therefore reduces, by adding the opposite positive parts, to positive uniqueness. This proves what is needed here without borrowing the downstream NAG-002973.

The compact carrier SS is closed in Hausdorff YY. Push each positive measure forward along S↪YS\hookrightarrow Y; compact inner approximants stay compact, and an outer open of SS lifts to an open of YY to which Y∖SY\setminus S may be added. This gives the claimed signed Radon measure on YY. For linearity compare on the compact union of the carriers by density. For support inside an open, the local integral equals the ambient integral on every multiplier by 002425 and 002428. Uniqueness on the same compact carrier proves the zero-extension assertion. ◻

8.5 The tropical infrastructure: charts, calibration, and smoothness

Paralinéaire means piecewise affine geometry

The French term paralinéaire, abbreviated pl, refers to piecewise affine geometry. Local polyhedral pieces carry affine structures, and subdivisions allow maps and functions to be described affinely on smaller cells. This is not yet the integral structure supplied by the character lattice of an algebraic torus. The abstract real framework separates what is needed for differential calculus from the additional rational or integral structure needed for arithmetic applications.

The reason for this separation can be seen in an ordinary PL function. Its graph has an honest piecewise affine structure, even at a corner where the graph is not the graph of one affine function. A cellwise derivative exists. A global distributional derivative has an additional interface term. CLD25 arranges the definitions so that these two operations remain distinguishable.

G-local descriptions and ordinary smooth descriptions

A G-tropical chart may be defined on an analytic or polyhedral domain, and a G-cover may assemble several such domains along interfaces. Pulling back ambient smooth superforms through these charts gives G-forms. Ordinary smooth forms require the refined affine structure to provide suitable descriptions on ordinary neighbourhoods. A function can therefore be G-smooth without being ordinary smooth.

Take u(t)=∣t∣u(t)=|t|. On the two half-lines it is affine; as a G-local object it has zero cellwise Hessian away from the interface. As a function on the ordinary neighbourhood of 00, its current Hessian is 2δ02\delta_0. The distinction persists in higher degree and on analytic spaces. It is not removed by choosing a finer subdivision: refinement moves the interface into the combinatorics but does not eliminate its contribution to the current derivative.

This example also explains why a current originally defined on ordinary smooth tests cannot automatically be evaluated on every G-form. The larger testing operation requires either an extension theorem, measure coefficients with a compatible G-local construction, or a structured current for which the action is defined geometrically.

Calibrations are the replacement for an orientation and a density

On a smooth real manifold, integration uses an orientation and a top-degree density. On a tropical space there are several tangent pieces at a point, each possibly carrying a multiplicity. A calibration packages the appropriate tangent-volume data so that forms can be contracted and integrated. Positivity of a calibration corresponds to nonnegative weights in familiar integral polyhedral examples.

The calibration must be compatible under changes of chart. In a tropicalization of an analytic space, these compatibilities are degree formulas. In an abstract tropical space they are part of the structure and its axioms. The definitions involving skeletal subsets and their volume forms are designed to let integration depend on the intrinsic space instead of on a chosen embedding into Rr\mathbb R^r.

Richness and harmonic compatibility

The vocabulary of CLD25 Chapters 5–6 can initially obscure a simple requirement: there must be enough calibrated polyhedral pieces to detect forms and enough balancing to prove Stokes. Richness provides sufficient skeletal geometry to represent the local differential data. The stronger interior hypotheses ensure that the relevant top-dimensional data are visible. Differential separation ensures that the chosen functions and forms distinguish the directions needed by the calculus. Harmony expresses the cancellation of calibrated boundary contributions in tropical charts.

These descriptions explain the role of the axioms; they do not replace their exact definitions. In Definition 6.6.2, a tropical space of level nn is a calibrated and refined G-tropical space satisfying richness, together with a specified closed boundary and further interior conditions. The traces of boundaries on skeletal subsets must also be negligible in the required dimension. Thus the level and boundary belong to the structure.

An ordinary balanced weighted polyhedral complex is a useful model, but the framework is designed to include analytic spaces whose polyhedral descriptions vary with the chart. A single global complex is not required. CLD25 Chapters 12–14 prove the necessary analytic facts rather than assuming that a Berkovich space is literally a polyhedron.

Stokes in the abstract tropical framework

Once the calibration and harmony conditions hold, Stokes follows by applying the cellwise formula and cancelling interior faces. Partitions of unity pass from one chart to arbitrary compactly supported forms. The later analytic proof consists in identifying the canonical analytic calibration with this abstract one and verifying the hypotheses. This division makes the theory modular: a new class of spaces or charts can be tested against the same integration requirements.

The useful question when reading a new result is therefore: which part is formal tropical calculus, and which part verifies an analytic hypothesis? A proof of a formula for calibrated polyhedra does not by itself prove independence from analytic presentations. Conversely, once the analytic compatibility theorem is established, repeating every polyhedral calculation on Berkovich spaces is unnecessary.

References. [CLD25] (Chapters 2–6, especially Definition 6.6.2).

8.6 Analytic realization of the tropical framework

Graded algebra and residual intersection theory

CLD25 Chapter 10 supplies the algebra needed to carry residual geometry through changes of value group and field. A normed object has reductions at many radii, so the graded setting is the natural one. The chapter also treats Zariski–Riemann spaces and intersection numbers of residual bundles. Such a space organizes compatible data over modifications rather than choosing one preferred model.

The point of this machinery becomes visible later: a PL metric can have a residual line bundle at an analytic germ, and a curvature product can be compared with a residual intersection number. These numbers must be independent of auxiliary models and compatible with scalar extension. The descent results in the graded chapter are what makes that numerical comparison meaningful in the non-strict setting.

This is a different use of models from defining the entire analytic calculus as an inverse system of algebraic cycles. The ordinary test forms and currents are already analytic objects. Residual models are used to calculate and characterize particular currents arising from geometric metrics.

Analytic dimension and the source of polyhedra

CLD25 Chapter 11 collects the analytic dimension, finite-morphism, and reduction facts used later. CLD25 Chapter 12 studies the skeletal loci of moment maps and their characteristic polyhedra. The tropical dimension of a point is tied to how many independent real directions its analytic functions can expose. The Abhyankar invariants from Section 3.7 control this dimension.

The richness theorems in CLD25 Chapter 12 say, in the source’s precise sense, that there are enough analytic skeletal subsets to detect the differential data; the boundaryless case supplies the stronger interior properties. These are existence and compatibility theorems. They do not assert that every analytic neighbourhood is homeomorphic to one finite polyhedron or that one fixed skeleton detects every current.

Degrees, multiplicities, and canonical calibration

CLD25 Chapter 13 defines the relevant degrees of quasi-finite morphisms, incorporating generic multiplicities and residue-field extensions. CLD25 Chapter 14 uses these degrees to calibrate suitable tropical cells. A local analytic map can be many-to-one even when the real tropical map is injective on a cell; the calibration records the missing degree.

A proof of a canonical integral must show compatibility under three operations: subdivision of cells, refinement of the moment map, and passage through finite analytic maps. The calibrated degree formula handles the last operation; the variation results ensure that these weights behave correctly in analytic families. This is the analytic counterpart of the lattice-index pushforward formula used in ordinary tropical intersection theory.

A family calculation and why it matters for Green currents

Suppose a family of fibres is cut out by analytic functions and a fixed compactly supported test is tropicalized by additional units. Under the source’s equidimensional hypotheses, the calibrated tropical images of nearby fibres stabilize in the relevant sense. The integral of that fixed test over the fibres can therefore be compared with its integral over the central fibre.

This mechanism is used in the proof of the Bochner–Martinelli formula. A truncated logarithmic maximum has curvature concentrated over a small Gauss point in the parameter space. As the truncation level tends to zero, the calibrated fibre calculation identifies its action on the fixed test with integration over the complete intersection. Thus the higher-codimension formula depends on the analytic degree and variation machinery, not only on a formal differentiation of a maximum.

Analytic forms and finite pushforward

CLD25 Chapter 15 realizes the abstract differential calculus on analytic spaces, including scalar extension, finite maps, support, and integration. CLD25 Chapter 16 develops currents in this realization. Proper pushforward is defined by duality with smooth pullback, with the appropriate degree shift. Projection formulas are then proved on their stated domains.

One should keep finite-map trace and arbitrary proper pushforward separate. A finite map can preserve dimension and carries a degree normalization. A general proper map may change dimension and collapse directions. Neither operation supplies a general descent theorem for actual delta-form representatives merely because a pushforward current exists.

8.7 Tropical geometry in the analytic calculus

The same tropical constructions enter several analytic theories, but at different stages. In the CLD construction, the integral tangent lattice and canonical multiplicities normalize integration. Balancing then supplies Stokes’ theorem, so that differentiation of currents is independent of a cell decomposition. The passage to the analytic space still requires compatibility under refinements of moment maps; a formula on one polyhedron is only a presentation of that construction. These steps are developed in Sections 8.1 and 8.4.

In the GK calculus, a tropical cycle also occurs as a factor in a form. The corner formula becomes a differential identity with a cycle-valued contribution. Stable intersection supplies the multiplication of these contributions under the hypotheses of the delta-calculus. Thus GK uses more than the set-theoretic tropicalization: tangent lattices, multiplicities, refinements, and intersection products all enter. Sections 16.1 and 15.7 explain how the resulting objects descend through algebraic tropical charts.

Skeleta provide a different kind of comparison. Their integral-affine structures and slope formulas relate model divisors to tropical functions. Faithful tropicalization controls a chosen skeleton, while the multiplicity-one section selects analytic points over a suitable tropical locus. Neither assertion identifies every analytic form with data on one preassigned skeleton. The precise geometric statements are used in Section 6.3; the distinction between topological reconstruction and differential comparison is taken up in Section 24.1.

Extended tropicalizations retain the boundary strata required by Green functions. Their role in the BGK theory is therefore complementary to the affine corner calculus: they allow compatible data on toric compactifications and local-height calculations on a common tropicalization. The valuation convention remains the same throughout. A lattice index is an integer multiplicity, whereas a logarithmic distance records the scale of the absolute value.

References. For tropicalization and multiplicities, see [Gub13] [MS15]. For the analytic integration construction, see [CLD12] (Chapters 2–4) and [CLD25]. The intersection and skeletal developments are [GK17] [GRW16] [GRW17]; the extended tropical intersection theory is [BGK25].

8.8 Scope and geometric inputs

Throughout the detailed construction below, KK is complete, algebraically closed and nontrivially valued, and XX is a separated integral algebraic variety of dimension nn over KK. We work on a paracompact open subset W⊂XanW\subset X^{\mathrm{an}}. The statements below concern the original smooth forms, not harmonic or weakly smooth forms. These assumptions are retained in every analytic consequence.

A very affine variety is a closed subvariety of a split algebraic torus. For a very affine open U⊂XU\subset X, the group of units modulo constants gives a canonical torus TUT_U and a closed immersion φU:U→TU\varphi_U:U\to T_U, determined up to translation. Its tropicalization is τU:Uan⟶NU,R,x⟼(−log⁡∣f1(x)∣,…,−log⁡∣fr(x)∣).\tau_U:U^{\mathrm{an}}\longrightarrow N_{U,\mathbb R},\qquad x\longmapsto(-\log|f_1(x)|,\ldots,-\log|f_r(x)|). Here the chosen units give coordinates; their tropical image is denoted Trop⁡(U)\operatorname{Trop}(U). A tropical chart is V=τU−1(Ω)V=\tau_U^{-1}(\Omega), where Ω\Omega is relatively open in Trop⁡(U)\operatorname{Trop}(U).

Theorem 8.8.1. (Geometric inputs from Gubler). The following results are used with the hypotheses above.

  1. Tropical charts form a basis. Intersections have common refinements coming from intersections of very affine opens; transition maps of tropicalizations are integral affine.

  2. A superform on the image of a tropical chart defines zero on the analytic chart if and only if it is zero on that tropical image.

  3. The algebraic-chart sheaf agrees with the original CLD sheaf using analytic moment maps.

  4. Smooth partitions of unity subordinate to open covers exist, with locally finite compact supports.

Open proof obligation (RC2). Supply native proofs of the geometric moment-image, boundary and degree inputs. The historical source expressly imports the listed Gubler16 results.

NAG-900055 · Proof status

8.9 The quotient on a polyhedral image

Let PP be a locally finite polyhedral subset of a real affine space and let Ω⊂P\Omega\subset P be relatively open. Choose an ambient open neighbourhood. Restrict ambient superforms to each face using its tangent space, and declare two representatives equal when all these restrictions agree. Subdivision does not change the resulting space.

Proposition 8.9.1. The kernel of restriction is stable under d′d', d′′d'', JJ and wedge multiplication. Thus these operations descend to the polyhedral quotient. Pullbacks along affine maps mapping faces into faces also descend.

Proof. On the relative interior of a face, restriction is ordinary affine pullback. It commutes with the two differentials by their coordinate formulas, and with JJ and wedge by their algebraic definitions. A form restricting to zero has zero tangential coefficients; their tangential derivatives vanish on that relative interior. Continuity gives the same conclusion on its closure. Applying this argument face by face proves the kernel assertions. For an affine map, a representative vanishing on a target face pulls back to zero on every source face mapping to it. ◻

One must quotient by tangent restrictions, not merely by the values of the scalar coefficients on PP. For example, on the line x2=0x_2=0 in R2\mathbb R^2, the form d′x2d'x_2 vanishes after restriction although its ambient coefficient is the constant one. The two cotangent copies are both restricted.

8.10 Gluing the analytic sheaf

Definition. A smooth (p,q)(p,q)-form on WW is a family of superforms on the images of a tropical-chart cover, whose pullbacks agree on common refinements of overlaps. Two families are identified if they agree after refining their covers. Write this sheaf as AXp,qA^{p,q}_X.

Proposition 8.10.2. This defines a sheaf. The operations d′d', d′′d'', JJ and wedge descend to it and satisfy the identities of Chapter 7.

Proof. A common refinement of three covers is obtained by successive intersections, so agreement after refinement is transitive. Restriction is restriction of each family. If forms on an open cover agree on overlaps, take the union of their chart families; the agreement condition makes this a global compatible family. Uniqueness is tested on the same covering charts. Proposition 8.9.1 proves that applying an operation to equal representatives preserves equality. Its algebraic identities therefore pass to the quotient and then to the glued sheaf. ◻

The support is the complement of the locus where the germ vanishes. The chart injectivity in Theorem 8.8.1 identifies vanishing of a form presented on one chart with vanishing of its superform on that chart’s image. It does not say that every form on a fixed chart has a single presentation on that chart; a further cover may be needed [Gubler] (Remark 5.7).

Example. (Behaviour at a classical point). If x∈X(K)x\in X(K) and ff is an analytic unit near xx, then on a sufficiently small neighbourhood ∣f(y)−f(x)∣<∣f(x)∣|f(y)-f(x)|<|f(x)|. The strong triangle inequality gives ∣f(y)∣=∣f(x)∣|f(y)|=|f(x)|. For finitely many units, intersect these neighbourhoods. A smooth function expressed through their logarithms is constant there, and positive-degree forms in these tropical coordinates vanish there. Thus these coordinates are not real-manifold coordinates near a classical point.

8.11 Compactly supported tests

Write Acp,q(W)A_c^{p,q}(W) for forms with compact support. A fixed presentation consists of finitely many tropical charts and compact polyhedral subsets of their images, together with compatible coefficient representatives. Tropicalization is proper in the very affine setting, so their inverse images are compact. On each such presentation use the coefficient C∞C^\infty seminorms on its fixed compact sets, with tangential restrictions on the faces. The actual test space has the final locally convex topology of all these finite presentations, as in [Gubler] (§6.1–6.2). Equivalently, a current is a linear functional continuous on each of the presentation spaces. In this book a functional on (q,q)(q,q)-tests has bidimension (q,q)(q,q) and bidegree (n−q,n−q)(n-q,n-q). Gubler’s notation in §6.2 uses the test degree as the current index; this indexing difference changes no pairing. This topology is essential: a sequence that is smooth in a different chart at every stage is not automatically convergent in a fixed presentation. Arguments below that use uniform convergence of scalar functions will be justified by measures, rather than by pretending that the scalar functions all have a common chart.

8.12 Cutoffs and scalar density

Lemma 8.12.1. For compact C⊂WC\subset W and an open neighbourhood OO of CC, there is 0≤χ≤10\leq\chi\leq 1 smooth, compactly supported in OO, and equal one near CC.

Proof. Choose C⊂O′⋐OC\subset O'\Subset O using local compactness and compactness of CC. Apply the partition theorem to the cover {O,W∖O′‾}\{O,W\setminus\overline{O'}\}. Only finitely many members of the locally finite partition meet CC. Sum all members meeting a smaller neighbourhood of CC, chosen to avoid the other supports and to lie in O′O'. Every selected support is compact and lies in OO. Their sum is one on that smaller neighbourhood and lies between zero and one everywhere. ◻

Proposition 8.12.2. Every f∈Cc(W)f\in C_c(W) admits smooth approximants converging uniformly on WW, with all supports in one compact subset. If f≥0f\geq 0, the approximants may be chosen nonnegative.

Proof. Take a compactly supported cutoff χ\chi equal one near supp⁡f\operatorname{supp}f, and put L=supp⁡χL=\operatorname{supp}\chi. Restrictions of global smooth functions form a real algebra containing constants on LL. They separate points: a cutoff supported in a neighbourhood of one point and avoiding another separates those points. The real Stone–Weierstrass theorem gives smooth aja_j converging uniformly to ff on LL. Then χaj→f\chi a_j\to f uniformly on WW because χf=f\chi f=f, with supports in LL. For nonnegative ff, approximate f\sqrt f on LL by smooth bjb_j and use χbj2\chi b_j^2 instead. ◻

This theorem is scalar density only. It does not preserve a positive Hessian and gives no general regularization theorem for current-psh functions. The approximation class in Chapter 11 remains a separate hypothesis.

8.13 Ordinary-positive domination on actual tests

An actual symmetric form is positive if locally it has a chart presentation whose restrictions to the faces are ordinary positive. This property is preserved by refinement because affine pullback preserves positivity. Multiplication by a nonnegative smooth scalar also preserves it, even if that scalar requires a further chart.

Proposition 8.13.1. For every fixed finite test presentation and compact supports, there are a compactly supported actual positive form θ\theta and a constant cc such that every symmetric test η\eta in that presentation satisfies c∥η∥D,0θ+η≥0,c∥η∥D,0θ−η≥0.c\|\eta\|_{D,0}\theta+\eta\geq0,\qquad c\|\eta\|_{D,0}\theta-\eta\geq0. Here ∥η∥D,0\|\eta\|_{D,0} is a coefficient seminorm in that fixed presentation.

Proof. First localize by a fixed finite nonnegative smooth partition over the prescribed supports. For one chart, choose an ambient Euclidean norm and the constant form βq=ωq/q!\beta_q=\omega^q/q!. On every tangent space its restriction has identity coefficient matrix in an orthonormal exterior basis. Restricting a coefficient tensor to that tangent space has uniformly bounded norm: the chosen Euclidean metric makes orthogonal restriction contractive. Thus the matrix estimate of Proposition 7.3.2 gives positive domination on every face by a common multiple of βq\beta_q. Multiply this control form by a cutoff equal one on the localized support and compactly supported inside the chart, then extend by zero. It is an actual positive form. Summing the finitely many control forms and constants gives the assertion. Tangent spaces of dimension less than qq contribute zero to both the test and control form. ◻

8.14 Intrinsic continuous multiplication

The next construction avoids taking a quotient of arbitrary continuous chart coefficients. Each measure is attached directly to an actual smooth test form.

Theorem 8.14.1. Let TT be a symmetric linear functional on Acq,q(W)A_c^{q,q}(W) nonnegative on ordinary positive tests. Then TT is a current, of order zero in every fixed presentation. For every actual smooth compactly supported test η\eta, there is a signed Radon measure μT,η\mu_{T,\eta} determined by ∫ϕ dμT,η=T(ϕη),ϕ∈Cc∞(W).\int\phi\,d\mu_{T,\eta}=T(\phi\eta),\qquad \phi\in C_c^\infty(W). Its support lies in supp⁡η\operatorname{supp}\eta. For f∈C0(W)f\in C^0(W) the formula (fT)(η)=∫f dμT,η(fT)(\eta)=\int f\,d\mu_{T,\eta} defines an intrinsic current. It is positive when f≥0f\geq 0, and agrees with smooth multiplication when ff is smooth.

Proof. Actual domination gives current continuity exactly as in Proposition 9.9.1; that proposition is purely functional analytic and does not use this theorem. If η\eta is positive, then ϕ↦T(ϕη)\phi\mapsto T(\phi\eta) is nonnegative for nonnegative smooth ϕ\phi, and ∣T(ϕη)∣≤∥ϕ∥supp⁡η,∞ T(η).|T(\phi\eta)|\leq\|\phi\|_{\operatorname{supp}\eta,\infty}\,T(\eta). Scalar density and the Riesz theorem give a positive Radon measure. It is supported in supp⁡η\operatorname{supp}\eta because all tests outside that set pair to zero. For symmetric η\eta, Proposition 8.13.1 writes η\eta as a difference of two actual positive tests. Take the difference of their measures. Its values on smooth scalar tests determine it uniquely by scalar density; hence it is independent of the decomposition. Symmetrization handles all tests. This also proves linearity in the actual test η\eta, including every relation among its chart presentations.

It remains to check continuity of fTfT. In a fixed presentation put M=c∥η∥D,0M=c\|\eta\|_{D,0} and use its fixed control form θ\theta. The positivity of Mθ±ηM\theta\pm\eta implies −MμT,θ≤μT,η≤MμT,θ.-M\mu_{T,\theta}\leq\mu_{T,\eta}\leq M\mu_{T,\theta}. Consequently ∣μT,η∣≤MμT,θ|\mu_{T,\eta}|\leq M\mu_{T,\theta} and ∣(fT)(η)∣≤c∥η∥D,0∫∣f∣ dμT,θ.|(fT)(\eta)|\leq c\|\eta\|_{D,0}\int|f|\,d\mu_{T,\theta}. The final integral is finite because θ\theta has compact support. This proves continuity on every presentation. Positivity and agreement with smooth multiplication follow from the defining measure formula. ◻

Proposition 8.14.2. The construction commutes with restriction and satisfies f(gT)=(fg)Tf(gT)=(fg)T for continuous f,g≥0f,g\geq 0. The same equality holds for signed continuous multipliers by linearity.

Proof. For smooth compactly supported ϕ\phi, uniqueness of the representing measures gives μT,ϕη=ϕμT,η\mu_{T,\phi\eta}=\phi\mu_{T,\eta}. It follows that (gT)(ϕη)=∫gϕ dμT,η,μgT,η=gμT,η.(gT)(\phi\eta)=\int g\phi\,d\mu_{T,\eta}, \qquad \mu_{gT,\eta}=g\mu_{T,\eta}. Integrating ff proves associativity. The same uniqueness argument on open subsets proves restriction compatibility. Decompose a signed continuous multiplier into its positive and negative parts to obtain the signed version. ◻

Thus the relation problem for multiplication by a continuous scalar is resolved in the stated original algebraic CLD setting. This does not resolve the different kernel problem d′d′′[h]=0⇒d′d′′(hT)=0\mathrm d'\mathrm d''[h]=0\Rightarrow\mathrm d'\mathrm d''(hT)=0: the latter concerns the action of curvature-equivalent potentials, not relations between presentations of one actual smooth test.

8.15 The degree in change of tropical coordinates

The foundations identify the origin of the three factors in an integral: the analytic fibre degree, the tangent lattice, and the Euclidean density. We now prove the elementary change-of-variables calculation and isolate the geometric comparison needed to apply it to analytic charts. This section uses the algebraic CLD setting of Section 8.8.

Lemma 8.15.1. Let F:σ→τF:\sigma\to\tau be an affine isomorphism between nn-dimensional polyhedra whose linear part is integral and injective on their tangent lattices. Put e=[Nτ:F(Nσ)]e=[N_\tau:F(N_\sigma)]. For a compactly supported top-degree superform β\beta on τ\tau, in lattice-normalized integration, ∫σF∗β=e∫τβ.\int_\sigma F^*\beta=e\int_\tau\beta.

Proof. Choose integral bases of the two lattices, with real coordinates x,yx,y. Write y=Mx+by=Mx+b. The positive top-degree generator has two cotangent factors. Its pullback acquires (det⁡M)2(\det M)^2, whereas ordinary real change of variables contributes ∣det⁡M∣−1|\det M|^{-1}. The resulting factor is ∣det⁡M∣=e|\det M|=e. This calculation is independent of the chosen bases, whose transition determinants have absolute value one. ◻

After subdividing a polyhedral map into cells on which it is injective, apply the lemma cell by cell. If source weights are mσm_\sigma, the target weight is mτ=∑F(σ)=τmσ[Nτ:F(Nσ)].m_\tau=\sum_{F(\sigma)=\tau}m_\sigma [N_\tau:F(N_\sigma)]. Cells whose images have dimension below nn contribute zero because top-degree pullback vanishes there. For a finite collection of cells and a form whose relevant pullback has compact support, summing gives the projection formula for the weighted integrals.

Example 8.15.2. On an annulus write t=−log⁡∣T∣t=-\log|T| and replace the moment coordinate TT by TmT^m, with integer m≥1m\geq 1. Then s=mts=mt and ∫m2b(mt) dt=m∫b(s) ds.\int m^2 b(mt)\,dt=m\int b(s)\,ds. The tropical image for this presentation has calibration mm. The algebraic map T↦TmT\mapsto T^m is finite free of rank mm, with basis 1,T,…,Tm−11,T,\ldots,T^{m-1}, also when the residue characteristic divides mm. The factor records algebraic degree, not the number of visibly distinct points of a fibre. A top form a(t)d′t∧d′′ta(t)d't\wedge d''t is represented in the ss coordinate by m−2a(s/m)d′s∧d′′sm^{-2}a(s/m)d's\wedge d''s; integrating it with weight mm gives the original integral.

Independence of the analytic integral

For two analytic presentations, adjoining all their unit coordinates gives a common moment map. The geometric input says that its calibrated weights push forward to the weights of either presentation. In the algebraic chart construction this is the tropical degree comparison used in [Gubler] (Proposition 4.11 and Lemma 5.15); the analytic construction is in [CLD12] (§§3.3–3.8). The lemma proves equality of the two local integrals once this geometric weight identity is known. It does not prove the weight identity itself.

The remaining partition argument is formal but useful to record. Let (ρi)(\rho_i) and (σj)(\sigma_j) be two subordinate smooth partitions near the compact support of a form α\alpha, and denote the corresponding local integral functionals by IiI_i and JjJ_j. Only finitely many terms meet the support. Refining both presentations where ρiσjα\rho_i\sigma_j\alpha is supported, the local comparison gives ∑iIi(ρiα)=∑i,jIi(ρiσjα)=∑i,jJj(ρiσjα)=∑jJj(σjα).\begin{aligned} \sum_i I_i(\rho_i\alpha) &=\sum_{i,j}I_i(\rho_i\sigma_j\alpha)\\ &=\sum_{i,j}J_j(\rho_i\sigma_j\alpha) =\sum_jJ_j(\sigma_j\alpha). \end{aligned} Thus the global integral is independent of charts and partitions. The analytic-domain and sheaf constructions of Chapter 2 make the refinements meaningful; the tropical degree theorem supplies the nonformal part. For Stokes one further uses balancing and support away from the analytic boundary, as defined in Section 2.7. None of these steps is supplied by the homeomorphism type of a skeleton alone.

Ordinary presentation topologies on compact tests

Fix a complete non-Archimedean field KK with its original real-valued absolute value. Let XX be good and Hausdorff. We use the ordinary smooth moment sheaf AXp,q\mathcal A_X^{p,q} of Definition 5.52.2 and Theorem 5.52.3, and write Ep,q(X)=AX,cp,q(X)\mathcal E^{p,q}(X)=\mathcal A_{X,c}^{p,q}(X). All supports in this note are ordinary compact supports. A version with boundary-avoiding tests is obtained by applying the construction to a specified ordinary open subspace of the analytic interior. There is no boundary vanishing assertion in the topology construction. All one-form generators anticommute, and the positive top generator is d′x1∧d′′x1∧⋯∧d′xn∧d′′xnd'x_1\wedge d''x_1\wedge\cdots\wedge d'x_n\wedge d''x_n, as in Definition 5.46.1.

For an open Ω⊂RN\Omega\subset\mathbb R^N, let Ep,q(Ω)E^{p,q}(\Omega) be the space of smooth ambient superforms with the coefficient seminorms pL,m(η)=max⁡I,J, ∣a∣≤msup⁡u∈L∣∂aηI,J(u)∣,L⋐Ω.p_{L,m}(\eta)=\max_{I,J,\,|a|\leq m}\sup_{u\in L} |\partial^a\eta_{I,J}(u)|, \qquad L\Subset\Omega. Empty coefficient slots give the zero space.

Definition 8.15.3. NAG-002472 (The two generating families). A cutoff presentation i=(V,f,Ω,χ)i=(V,f,\Omega,\chi) consists of an affinoid domain V⊂XV\subset X, a unit tuple ff on VV, an open set Ω⊃tf(V)\Omega\supset t_f(V), and χ∈Ac0,0(X)\chi\in\mathcal A_c^{0,0}(X) with supp⁡χ⊂Int⁡top(V,X)\operatorname{supp}\chi\subset\operatorname{Int}_{\mathrm{top}}(V,X). Here tf=−log⁡∣f∣t_f=-\log|f|. Its map is Qi:Ep,q(Ω)⟶Ep,q(X),Qiη=ext⁡0(χtf∗η).Q_i:E^{p,q}(\Omega)\longrightarrow\mathcal E^{p,q}(X),\qquad Q_i\eta=\operatorname{ext}_0(\chi t_f^*\eta). Let τcut\tau_{\mathrm{cut}} be the locally convex topology generated by all seminorms PP on Ep,q(X)\mathcal E^{p,q}(X) for which PQiP Q_i is continuous for every ii. This is the definition F8-T1 in the supplied candidate.

An ordinary presentation ss consists of finitely many such charts (Vj,fj,Ωj)(V_j,f_j,\Omega_j), without cutoffs, and a compact set Ks⊂⋃jUjK_s\subset\bigcup_j U_j, where Uj=Int⁡top(Vj,X)U_j=\operatorname{Int}_{\mathrm{top}}(V_j,X). Let Hs⊂∏jEp,q(Ωj)H_s\subset\prod_j E^{p,q}(\Omega_j) be the linear subspace of tuples (ηj)(\eta_j) for which the induced ordinary sections agree on Uj∩UkU_j\cap U_k and vanish on Uj∖KsU_j\setminus K_s. Give HsH_s its coefficient subspace topology. The sections glue with zero on X∖KsX\setminus K_s to a section Rsη∈Ep,q(X)R_s\eta\in\mathcal E^{p,q}(X). Give Bs=Hs/ker⁡RsB_s=H_s/\ker R_s its quotient topology and let τord\tau_{\mathrm{ord}} be the final locally convex topology for the maps Bs→Ep,q(X)B_s\to\mathcal E^{p,q}(X), equivalently for all RsR_s.

Both definitions refer to ordinary sections. In particular, the relations in HsH_s are not compatibility solely on a finite closed affinoid cover. Every compactly supported section belongs to some im⁡Rs\operatorname{im}R_s: choose finitely many ordinary representing neighbourhoods over its support and take KsK_s to be that support. The cutoff maps also span the test space, by the finite partition of Theorem C.1.60 on these same neighbourhoods.

Theorem 8.15.4. NAG-002473 (Comparison of the two final topologies). In every bidegree, τcut=τord.\tau_{\mathrm{cut}}=\tau_{\mathrm{ord}}. More precisely, each RsR_s is a finite sum of cutoff presentation maps composed with continuous linear maps, and each QiQ_i factors continuously through one ordinary presentation space HsH_s. The charts, cutoffs and factorizing maps in each assertion are fixed before the variable ambient forms are chosen.

Proof. Fix ss. Apply Theorem C.1.60 to KsK_s and (Uj)(U_j), obtaining fixed χj\chi_j supported compactly in UjU_j, with sum one near KsK_s. If η=(ηj)∈Hs\eta=(\eta_j)\in H_s, then Rsη=∑jQ(Vj,fj,Ωj,χj)ηj.R_s\eta=\sum_j Q_{(V_j,f_j,\Omega_j,\chi_j)}\eta_j. On KsK_s this follows from the partition and compatibility; off KsK_s all relevant sections are zero. The coordinate projections from HsH_s are continuous. Thus every RsR_s is continuous for τcut\tau_{\mathrm{cut}}.

Conversely, fix i=(V,f,Ω,χ)i=(V,f,\Omega,\chi) and put Ks=supp⁡χK_s=\operatorname{supp}\chi. The empty-support case is zero. Otherwise choose finitely many affinoid neighbourhoods Wj⊂Int⁡top(V,X)W_j\subset\operatorname{Int}_{\mathrm{top}}(V,X) whose interiors cover KsK_s, such that χ\chi is represented on WjW_j by a fixed smooth scalar coefficient cjc_j in a fixed unit tuple gjg_j. To do this, shrink inside an ordinary representing neighbourhood of χ\chi before taking WjW_j; the neighbourhood basis and restriction assertion are part of the proof of Theorem 5.52.3. On WjW_j use the joint tuple hj=(f∣Wj,gj)h_j=(f|_{W_j},g_j) and the ambient open set Ω×Ωgj\Omega\times\Omega_{g_j}, where Ωgj⊃tgj(Wj)\Omega_{g_j}\supset t_{g_j}(W_j). Set Ljη=πg∗cj πf∗η.L_j\eta=\pi_g^*c_j\,\pi_f^*\eta. For every compact LL in this product and every mm, the product and chain rules give pL,m(Ljη)≤Cj,L,m pπf(L),m(η).p_{L,m}(L_j\eta)\leq C_{j,L,m}\,p_{\pi_f(L),m}(\eta). The derivatives of the fixed coefficient cjc_j are absorbed in Cj,L,mC_{j,L,m}. The tuple (Ljη)j(L_j\eta)_j belongs to the fixed space HsH_s: its components represent the same ordinary section χtf∗η\chi t_f^*\eta and vanish off KsK_s. It depends continuously on η\eta, and its glued section is QiηQ_i\eta. Therefore QiQ_i is continuous for τord\tau_{\mathrm{ord}}.

A seminorm is continuous for either final topology exactly when its composites with that topology’s generating maps are continuous. The two factorizations prove equality of these families of seminorms. ◻

Proposition 8.15.5. NAG-002476 (The available direction of the tangential comparison). Let P⊂Ω⊂RNP\subset\Omega\subset\mathbb R^N be a compact finite polyhedral set, and let NPN_P be the ambient smooth forms with zero tangential restriction on every face. Give Ep,q(Ω)/NPE^{p,q}(\Omega)/N_P its quotient topology. For each face σ\sigma, fix an affine tangent basis. The suprema, on compact subsets of σ\sigma, of tangential coefficient derivatives of any fixed finite order are well-defined continuous seminorms on this quotient. Consequently the topology they generate is no finer than the ambient quotient topology.

Proof. In a tangent basis, restriction of each covector factor is a fixed linear map. Each tangential derivative is a fixed linear combination of ambient partial derivatives. Its coefficient supremum is bounded by a constant times the ambient seminorm of the same order on the same compact set. If a form belongs to NPN_P, its restricted coefficients vanish on the relative interior of every face. Their derivatives along that face vanish there and, by continuity, on its boundary. Thus the seminorms annihilate NPN_P and descend. The defining property of the quotient gives continuity of each descended seminorm. ◻

The intrinsic tangential comparison here gives continuity from the ambient quotient for each fixed compact polyhedral image. A converse would require a bound for ambient representatives. A comparison with general chart and support families would also require matching the underlying sheaves and the systems of presentations.

Multiplication in chart and support topologies

Fix bidegrees and an order r∈N∪{∞}r\in\mathbf N\cup\{\infty\}. The coefficients are smooth real scalars. The order rr specifies which coefficient derivatives define the topology. For r=∞r=\infty, use every finite derivative order.

Let XX be a Hausdorff GG-tropical space with the following chart and sheaf data. Smooth forms on a compact chart image are tangential restrictions of local ambient smooth forms on a finite polyhedral set. For every compact domain WW in a chart f:V→Ef:V\to E, the image f(W)f(W) is compact and finite polyhedral, and a form on it is uniquely determined by its pullback to WW. Compact domains cover chart domains in the GG-topology. The smooth GG-form sheaf has restrictions and exterior products compatible with the corresponding affine pullbacks and is determined on GG-covers. Domains restrict to chart intersections, with joint coordinates available for common refinements.

Definition 8.15.6. (Intrinsic seminorm on a compact image). Let CC be a compact cell, use affine coordinates on its affine span, and write a tangential form as α=∑∣I∣=p,∣J∣=qαIJ d′xI∧d′′xJ\alpha=\sum_{|I|=p,|J|=q}\alpha_{IJ}\,d'x_I\wedge d''x_J. Put ∥α∥C,m=max⁡I,J, ∣a∣≤msup⁡C∣∂aαIJ∣,m≥0.\|\alpha\|_{C,m} =\max_{I,J,\,|a|\le m}\sup_C|\partial^a\alpha_{IJ}|, \qquad m\ge0. The derivatives at the boundary are the continuous restrictions of those on the relative interior. Coefficient slots with degree exceeding the dimension are zero. For a compact finite polyhedral set PP, choose a finite cell decomposition C\mathcal C and affine coordinates on each cell; a standard seminorm is ∥α∥P,m=max⁡C∈C∥α∣C∥C,m.\|\alpha\|_{P,m}=\max_{C\in\mathcal C}\|\alpha|_C\|_{C,m}. Use the seminorms for m≤rm\le r, all finite mm if r=∞r=\infty. This is the intrinsic cell topology on the space of smooth forms on PP.

Different affine coordinates give equivalent seminorms: the tangential covectors change by fixed minors, and their derivatives by fixed linear combinations. Passing to a common finite subdivision gives the same cell topology. On a subdivided cell of full relative dimension the smooth coefficient derivatives are restrictions of the old ones; conversely these subcells cover a dense subset of the original cell, so the original suprema follow by continuity. Lower-dimensional cells are handled by their tangent restrictions. This uses only finitely many coordinate changes. It asserts no bounded ambient extension across faces.

For an affine map a:P→Qa:P\to Q of compact polyhedral sets, subdivision of the source and target gives, for each mm, a finite constant ca,mc_{a,m} with ∥a∗α∥P,m≤ca,m∥α∥Q,m.\begin{equation} \|a^*\alpha\|_{P,m}\le c_{a,m}\|\alpha\|_{Q,m}. \end{equation} To see the estimate, restrict to a source cell mapped into a target cell. The coefficients of pullback are sums with coefficients given by the pp- and qq-minors of the fixed tangent matrix. A derivative of order ℓ≤m\ell\le m adds products of ℓ\ell matrix entries, by the chain rule. Take the maximum of these finitely many constants. Translation contributes no derivatives. This proves the forward continuity needed below; no strictness assertion for a surjective affine map is used.

Definition 8.15.7. (A single-chart space). Let f:V→Ef:V\to E be a chart on a domain of XX. Define Af,Gp,q(X)\mathcal A_{f,G}^{p,q}(X) to be the subspace of smooth GG-forms ω\omega on XX for which, for every compact domain W⊂VW\subset V, there is αf,W∈Ap,q(f(W))\alpha_{f,W}\in\mathcal A^{p,q}(f(W)) satisfying ω∣W=(f∣W)∗αf,W.\omega|_W=(f|_W)^*\alpha_{f,W}. The compact-nullity prerequisite makes αf,W\alpha_{f,W} unique. Transport the intrinsic topology from Ap,q(f(W))\mathcal A^{p,q}(f(W)) to its pullback space on WW, and give Af,Gp,q(X)\mathcal A_{f,G}^{p,q}(X) the initial topology for all these restriction maps. Explicitly it is generated by ρf,W,m(ω)=∥αf,W∥f(W),m,W⊂V compact,m≤r.\rho_{f,W,m}(\omega)=\|\alpha_{f,W}\|_{f(W),m}, \qquad W\subset V\text{ compact},\quad m\le r.

Definition 8.15.8. (A fixed chart family). For a family D={fi:Vi→Ei}\mathcal D=\{f_i:V_i\to E_i\} put AD,Gp,q(X)=⋂iAfi,Gp,q(X),\mathcal A_{\mathcal D,G}^{p,q}(X)=\bigcap_i\mathcal A_{f_i,G}^{p,q}(X), and give it the initial locally convex topology TD,rT_{\mathcal D,r} for all the inclusions into the single-chart spaces. Thus its seminorms are the restrictions of all ρfi,W,m\rho_{f_i,W,m}; neighbourhoods require only finitely many of these seminorms at a time. Families and compact domains need not be finite or countable. If the domains form a GG-cover, compact nullity and the sheaf property imply that this topology is Hausdorff: vanishing of every seminorm gives zero on every chart domain and hence on XX.

We use the word refinement with its geometric meaning. A chart g:V′→Fg:V'\to F refines f:V→Ef:V\to E if V′⊂VV'\subset V and f∣V′=a∘gf|_{V'}=a\circ g for a fixed affine map a:F→Ea:F\to E. A family refines another if every chart in it has such a factorization through a chart of the latter. From a coarse family D\mathcal D to a refining family R\mathcal R one obtains an inclusion AD,Gp,q(X)⟶AR,Gp,q(X).\mathcal A_{\mathcal D,G}^{p,q}(X) \longrightarrow\mathcal A_{\mathcal R,G}^{p,q}(X). It is continuous. For a compact WW in the domain of g∈Rg\in\mathcal R, the representative on g(W)g(W) is a∗αf,Wa^*\alpha_{f,W}. Estimate (8.15.1) bounds each target seminorm by one source seminorm, proving continuity of the inclusion.

Definition 8.15.9. (Support and chart families). The support of a GG-form is the complement of the largest ordinary open set on which it is zero. A support family Φ\Phi is a family of ordinary closed subsets of XX containing ∅\varnothing, closed under finite unions and under taking ordinary closed subsets. Let Γ\Gamma be a nonempty collection of GG-covering chart families, admitting a common refining family in Γ\Gamma for any two members.

For D∈Γ\mathcal D\in\Gamma and F∈ΦF\in\Phi, let AD,G,Fp,q(X)={ω∈AD,Gp,q(X):supp⁡ω⊂F},\mathcal A_{\mathcal D,G,F}^{p,q}(X)= \{\omega\in\mathcal A_{\mathcal D,G}^{p,q}(X):\operatorname{supp}\omega\subset F\}, with its subspace topology from TD,rT_{\mathcal D,r}. Define AΓ,Φp,q(X)=⋃D∈Γ, F∈ΦAD,G,Fp,q(X).\mathcal A_{\Gamma,\Phi}^{p,q}(X) =\bigcup_{\mathcal D\in\Gamma,\,F\in\Phi} \mathcal A_{\mathcal D,G,F}^{p,q}(X). Common refinements and unions of supports make this a vector space. Its CrC^r topology is the locally convex final topology of all these stage inclusions. Equivalently its continuous seminorms are exactly those seminorms whose restrictions to every stage are continuous.

Consequently a linear map from this space to a locally convex space is continuous exactly when its restriction to each stage is continuous. This follows from the defining seminorms by composing with every continuous seminorm of the target.

Definition 8.15.10. (Ordinary smooth specialization). In the ordinary affine specialization, the specified affine coordinates locally present the ordinary smooth forms on domain interiors. Use the nonempty family ΓX\Gamma_X of all affine chart families whose domain interiors cover XX. A common refinement is obtained on the intersections using joint coordinates. For a support family Φ\Phi the ordinary test space here is AΓX,Φp,q(X)\mathcal A^{p,q}_{\Gamma_X,\Phi}(X) with the topology of Definition 8.15.9.

Proposition 8.15.11. (Multiplication by a fixed smooth form). Fix r∈N∪{∞}r\in\mathbf N\cup\{\infty\}. Let Γ\Gamma and Φ,Ψ\Phi,\Psi be as in Definition 8.15.9. Suppose the fixed smooth GG-form α\alpha of bidegree (p′,q′)(p',q') belongs to AΓp′,q′(X)\mathcal A^{p',q'}_{\Gamma}(X), using all ordinary closed supports. Assume that every ordinary closed subset of supp⁡α\operatorname{supp}\alpha belonging to Φ\Phi also belongs to Ψ\Psi. Then Mα:AΓ,Φp,q(X)⟶AΓ,Ψp+p′,q+q′(X),ω⟼α∧ω,M_\alpha:\mathcal A^{p,q}_{\Gamma,\Phi}(X) \longrightarrow\mathcal A^{p+p',q+q'}_{\Gamma,\Psi}(X), \qquad \omega\longmapsto\alpha\wedge\omega, is well-defined, linear, and continuous for the CrC^r topologies. In particular this holds for ordinary smooth forms with Γ=ΓX\Gamma=\Gamma_X.

Proof. Choose once a family E∈Γ\mathcal E\in\Gamma that tropicalizes α\alpha. Fix a source stage (D,F)(\mathcal D,F). A common refining family R∈Γ\mathcal R\in\Gamma tropicalizes both α\alpha and every form in the source stage. Refinement gives a continuous linear inclusion from AD,G\mathcal A_{\mathcal D,G} to AR,G\mathcal A_{\mathcal R,G} by the preceding affine estimate.

On a compact domain WW of a chart h∈Rh\in\mathcal R write α∣W=h∗a\alpha|_W=h^*a and ω∣W=h∗b\omega|_W=h^*b. In tangential cell coordinates, each coefficient of a∧ba\wedge b is a finite signed sum of products of coefficients. For each finite m≤rm\le r, the product rule gives ∥a∧b∥h(W),m≤Np,q,p′,q′,h(W) 2m∥a∥h(W),m ∥b∥h(W),m.\begin{equation} \|a\wedge b\|_{h(W),m} \le N_{p,q,p',q',h(W)}\,2^m \|a\|_{h(W),m}\,\|b\|_{h(W),m}. \end{equation} Here NN is a finite upper bound for the number of coefficient products in a fixed output slot on these finitely many cells. No ambient normal derivatives occur. The coefficient ∥a∥h(W),m\|a\|_{h(W),m} is fixed and finite. Thus for every target chart seminorm, multiplication is bounded by a constant times a continuous source seminorm after refinement. For r=∞r=\infty apply this argument separately at each finite mm.

For every source form its product has support in GF=F∩supp⁡αG_F=F\cap\operatorname{supp}\alpha. This is an ordinary closed subset of both FF and supp⁡α\operatorname{supp}\alpha; hence GF∈ΦG_F\in\Phi and the support hypothesis gives GF∈ΨG_F\in\Psi. The stage map therefore has its image in AR,G,GFp+p′,q+q′(X)\mathcal A^{p+p',q+q'}_{\mathcal R,G,G_F}(X), equipped with the subspace topology. The continuous map into AR,G\mathcal A_{\mathcal R,G} just constructed is continuous into that subspace. Composing with its canonical continuous inclusion into the target final space proves continuity on the source stage. The linear universal property of the source final topology proves continuity of MαM_\alpha. All choices of R\mathcal R and GFG_F depend only on the source stage and the fixed α\alpha, not on ω\omega. ◻

In particular a fixed ordinary smooth cutoff acts continuously on the compact test space. If the compact support family is taken inside a specified ordinary open of the analytic interior, the same argument gives an endomorphism for that family because the support can only decrease.

Estimates on bounded sets in final test spaces

For an arbitrary locally convex space EE, write Eb′E'_b for its continuous dual with the topology of uniform convergence on bounded subsets of EE. Bounded means bounded for every continuous seminorm. No separation or completeness assumption will be needed. In this section EE is spanned by the images of linear maps Qi:Ei→EQ_i:E_i\to E, and has the locally convex final topology for these maps.

Theorem 8.15.12. NAG-003086 (Uniform eventual equality on bounded sets). Let DM:E→RD_M:E\to\mathbb R be linear, for real M≥0M\geq0. Assume that every DMQiD_M Q_i is continuous, and that for every ii there is Hi<∞H_i<\infty such that DMQi=0for all M≥Hi.\begin{equation} D_M Q_i=0\qquad\hbox{for all }M\geq H_i. \end{equation} Then each DMD_M is continuous. For every bounded set B⊂EB\subset E there exists HB<∞H_B<\infty such that DM(b)=0(M≥HB, b∈B).\begin{equation} D_M(b)=0\qquad(M\geq H_B,\ b\in B). \end{equation} In particular DM→0D_M\to0 in Eb′E'_b. No bounded-set lifting, countable cofinality, regular inductive limit, or uniform bound on the HiH_i is assumed.

Proof. Continuity of DMD_M is exactly the final seminorm criterion. Every x∈Ex\in E is a finite sum of elements of presentation images, so (8.15.3) makes DM(x)D_M(x) eventually zero. Suppose (8.15.4) fails. Choose Mj≥jM_j\geq j and bj∈Bb_j\in B with DMj(bj)≠0D_{M_j}(b_j)\ne0, and put cj=j∣DMj(bj)∣,P(x)=∑j≥1cj∣DMj(x)∣.c_j=\frac{j}{|D_{M_j}(b_j)|},\qquad P(x)=\sum_{j\geq1}c_j|D_{M_j}(x)|. The sum is finite for every xx, so PP is a finite-valued seminorm. For each fixed ii all terms with j≥Hij\geq H_i vanish on Qi(Ei)Q_i(E_i). Thus PQiP Q_i is a finite sum of continuous seminorms and is continuous. The final topology makes PP continuous on EE. But P(bj)≥jP(b_j)\geq j, contradicting boundedness of BB. This contradiction proves (8.15.4), for the full real parameter family. ◻

For the following estimate, let DM:E→RD_M:E\to\mathbb R be linear for each real M≥0M\geq0.

Proposition 8.15.13. NAG-003090 (A quantitative estimate across presentations). Let pip_i be a continuous seminorm on EiE_i, let wi>0w_i>0, and let ε(M)>0\varepsilon(M)>0. Suppose ∣DMQi(η)∣≤ε(M)wipi(η)for all i,M,η.\begin{equation} |D_M Q_i(\eta)|\leq\varepsilon(M)w_i p_i(\eta) \quad\hbox{for all }i,M,\eta. \end{equation} Then the finite-decomposition gauge Pw(x)=inf⁡x=∑i∈FQiηi∑i∈Fwipi(ηi),F finite,\begin{equation} P_w(x)=\inf_{x=\sum_{i\in F}Q_i\eta_i} \sum_{i\in F}w_i p_i(\eta_i),\qquad F\text{ finite}, \end{equation} is a continuous seminorm on EE, and sup⁡x∈B∣DM(x)∣≤ε(M)sup⁡x∈BPw(x)\begin{equation} \sup_{x\in B}|D_M(x)| \leq\varepsilon(M)\sup_{x\in B}P_w(x) \end{equation} for every bounded set BB. Hence ε(M)→0\varepsilon(M)\to0 gives strong-dual convergence, even when no presentation is fixed for the members of BB. The premise must hold with the actual generators of the topology whose bounded sets are used.

Proof. The spanning assumption makes the infimum finite. Concatenating finite decompositions proves subadditivity (with arbitrary small errors before taking infima); scaling proves absolute homogeneity. Repeated indices can be merged by the seminorm triangle inequality. Thus PwP_w is a seminorm, possibly with a kernel. The one-term decomposition gives PwQi≤wipiP_w Q_i\leq w_i p_i, so every composite is continuous. The final criterion gives continuity of PwP_w. Apply (8.15.5) to any finite decomposition, sum, and take its infimum to get ∣DM(x)∣≤ε(M)Pw(x)|D_M(x)|\leq\varepsilon(M)P_w(x). Boundedness now proves (8.15.7). ◻