Conjecture Atlas
Frozen public edition v0.5.1. Read the current version of this page.
Analytic draft; conditional transfer

17 November 2026 · 38 PDF pages

Download standalone PDF

Results and proofs

Local Green Forms and Chow Correspondences

Conditional reductions and construction draft

17 November 2026

Abstract

We organize two geometric reductions for actual non-Archimedean Green forms. Algebraically trivial classes are treated through additive Poincaré families, rational-fibre rigidity and compactness of pp-adic abelian groups. Homologically trivial classes on abelian varieties are treated through vertical corrections on semistable models; Chow correspondences then transport the resulting numerical certificates. The principal conclusions are stated relative to their actual-form and numerical-transfer hypotheses. The complete available construction arguments are included: model Chern forms, scalar–curvature finite trace, exact-support Green pairs, compact-parameter primitives, abelian numerical descent and the regular-model Chow criterion. Products of curves and abelian varieties occur as consequences of the latter structure.

1. The two geometric reductions

Fix a finite extension K/QpK/\mathbf Q_p and its absolute value. Let X/KX/K be smooth, projective and geometrically integral of dimension nn. We seek a Green form for a prescribed rational cycle ZZ of codimension rr on the exact open set UZ=Xan∖∣Z∣an,1≤r≤n.U_Z=X^{\mathrm{an}}\setminus |Z|^{\mathrm{an}},\qquad 1\le r\le n. Here ∣Z∣|Z| is the union of the components with nonzero coefficient. Let Bp,q\mathcal B^{p,q} denote the ordinary actual δ\delta-form sheaf of Mihatsch [1], and write D=d′d′′D=d'd''. The desired identity is ∫UZγZ∧Dη=−∫Zanη,η∈Bcn−r,n−r(Xan). \int_{U_Z}\gamma_Z\wedge D\eta =-\int_{Z^{\mathrm{an}}}\eta, \qquad \eta\in\mathcal B_c^{n-r,n-r}(X^{\mathrm{an}}).(1.1) Its integral on the right retains the rational coefficients and cycle multiplicities. The form on the left belongs to Br−1,r−1(UZ)\mathcal B^{r-1,r-1}(U_Z), and the identity includes the existence of the displayed pairings.

There are two geometric ways to remove the curvature from a Green pair. Algebraic equivalence supplies a compact parameter group on which the obstruction becomes locally constant. A homologically trivial class on an abelian variety instead supplies a numerical special-fibre correction. The latter is transported through a Chow diagonal factorization. We state these reductions before developing the local constructions they require.

Definition 1.1 (Actual obstruction space). Set HBC,actr,r(X)={ω∈Br,r(Xan):d′ω=d′′ω=0}DBr−1,r−1(Xan).H^{r,r}_{\mathrm{BC,act}}(X)= \frac{\{\omega\in\mathcal B^{r,r}(X^{\mathrm{an}}):d'\omega=d''\omega=0\}} {D\mathcal B^{r-1,r-1}(X^{\mathrm{an}})}. A prescribed curved Green pair for ZZ consists of GZ∈Br−1,r−1(UZ)G_Z\in\mathcal B^{r-1,r-1}(U_Z) and a global separately closed ΩZ∈Br,r(Xan)\Omega_Z\in\mathcal B^{r,r}(X^{\mathrm{an}}) satisfying ∫UZGZ∧Dη+∫Zanη=∫XanΩZ∧η(η∈Bcn−r,n−r(Xan)). \int_{U_Z}G_Z\wedge D\eta+\int_{Z^{\mathrm{an}}}\eta =\int_{X^{\mathrm{an}}}\Omega_Z\wedge\eta \qquad\left(\eta\in\mathcal B_c^{n-r,n-r}(X^{\mathrm{an}})\right).(1.2)

Hypothesis 1.2 (Actual curvatures and geometric transfer). The following properties are named separately, so each conclusion can retain precisely the properties it uses.

(A1)

A rational-linear, Chow-invariant map oV,Kr:CH⁡r(V)Q→HBC,actr,r(V)\mathfrak o^r_{V,K}:\operatorname{CH}^r(V)_{\mathbf Q}\to H^{r,r}_{\mathrm{BC,act}}(V) is computed by actual degree-rr Chern characters of virtual bundles on projective flat models. It is functorial for the genuine analytic pullbacks used below. Two model characters with the same generic degree-rr Chow class differ by DD of a global actual form.

(A2)

Every prescribed rational cycle has a curved Green pair as in (1.2), with [ΩZ]=oX,Kr([Z])[\Omega_Z]=\mathfrak o^r_{X,K}([Z]).

(R)

Actual forms on (X×S)an(X\times S)^{\mathrm{an}} have the uniform rational-fibre constancy of Theorem 6.1: for smooth separated S/KS/K and a0∈S(K)a_0\in S(K), the pulled-back actual form is constant for all a∈S(K)a\in S(K) in one ordinary neighbourhood of a0a_0.

(N)

Let h:V→Xh:V\to X be finite dominant between smooth integral projective KK-varieties. If a projective flat model of VV carries ξ∈K0(V)Q\xi\in K_0(\mathcal V)_{\mathbf Q} such that ch⁡r(ξ∣V)=h∗z,deg⁡(ch⁡r(ξs)∩[T])=0\operatorname{ch}_r(\xi|_V)=h^*z,\qquad \deg\bigl(\operatorname{ch}_r(\xi_s)\cap[T]\bigr)=0 for every integral vertical rr-cycle TT, then oX,Kr(z)=0\mathfrak o^r_{X,K}(z)=0. A variety over a finite larger constant field may be regarded as a KK-variety; the special-fibre multiplicities, residue degrees and full native degree are retained.

(C)

The kernels of the obstruction maps are preserved by algebraic Chow correspondences of the specified degree, including proper algebraic pushforward. For a finite extension L/KL/K, this is interpreted in the native KK-theory and the algebraic projection π:XL→X\pi:X_L\to X satisfies π∗π∗=[L:K]\pi_*\pi^*=[L:K].

These properties are considered for every smooth projective integral variety occurring in the relevant argument, componentwise after geometric base change when its constant field is larger.

Theorem 1.3 (Algebraic equivalence). Assume (A1), (A2) and (R). If [Z][Z] is algebraically equivalent to zero over K‾\overline K, then oX,Kr([Z])=0\mathfrak o^r_{X,K}([Z])=0. Consequently there exists γZ∈Br−1,r−1(UZ)\gamma_Z\in\mathcal B^{r-1,r-1}(U_Z) satisfying (1.1) over the original valued field and against every compact actual complementary test.

The proof is completed in Subsection 9.1. The Poincaré family converts the cycle into an additive family on a Jacobian. Its algebraic norm gives a family over KK with one fibre equal to [L:K][Z][L:K][Z]. The obstruction-valued map is locally constant by (R), and compactness annihilates it.

For the second reduction, a degree-aa correspondence V⇝WV\rightsquigarrow W means a class in CH⁡dim⁡V+a(V×W)Q\operatorname{CH}^{\dim V+a}(V\times W)_{\mathbf Q}, acting from codimension rr to codimension r+ar+a.

Definition 1.4 (Potential abelian Chow retraction). The variety XX has a potential abelian Chow retraction if there are a finite extension L/KL/K, abelian varieties Aj/LA_j/L, integers aja_j, and correspondences uj∈CH⁡n+aj(XL×LAj)Q,vj∈CH⁡dim⁡Aj−aj(Aj×LXL)Qu_j\in\operatorname{CH}^{n+a_j}(X_L\times_LA_j)_{\mathbf Q},\qquad v_j\in\operatorname{CH}^{\dim A_j-a_j}(A_j\times_LX_L)_{\mathbf Q} such that ΔXL=∑jvj∘ujin CH⁡n(XL×LXL)Q. \Delta_{X_L}=\sum_jv_j\circ u_j \quad\text{in }\operatorname{CH}^n(X_L\times_LX_L)_{\mathbf Q}.(1.3) The zero-dimensional abelian variety Spec⁡L\operatorname{Spec}L is allowed.

Theorem 1.5 (Abelian correction and Chow transfer). Assume (A1), (A2), (N), (C), and the semistable abelian model theorem recalled in Theorem 7.4. If XX has a potential abelian Chow retraction, then oX,Kr(CH⁡r(X)Q,hom)=0.\mathfrak o^r_{X,K}\bigl(\operatorname{CH}^r(X)_{\mathbf Q,\mathrm{hom}}\bigr)=0. For 1≤r≤n1\le r\le n, every prescribed cycle in this group therefore has a Green form satisfying (1.1). In a fixed codimension, it is enough that the right-hand side of (1.3) act on the homologically trivial group as a fixed nonzero rational multiple of the identity.

Here CH⁡r(X)Q,hom\operatorname{CH}^r(X)_{\mathbf Q,\mathrm{hom}} is the kernel of the geometric rational ℓ\ell-adic cycle class for a fixed ℓ≠p\ell\ne p. The proof, completed in Subsection 9.2, first produces a vertically numerical extension on a semistable abelian model. Property (N) converts that extension into zero obstruction. Property (C) then applies the actual Chow identity term by term.

Corollary 1.6 (Products of curves and abelian varieties). Under the hypotheses of Theorem 1.5, let X=A×KC1×K⋯×KCs,X=A\times_KC_1\times_K\cdots\times_KC_s, where A/KA/K is abelian and the Ci/KC_i/K are smooth projective geometrically integral curves. Every prescribed geometrically homologically trivial cycle of codimension 1≤r≤dim⁡X1\le r\le\dim X satisfies (1.1) over KK.

Proof. After a finite extension choose rational points on all the curves. The actual Chow diagonal factorization of Theorem 8.6 factors through the single abelian variety A×∏iJac⁡(Ci)A\times\prod_i\operatorname{Jac}(C_i) with Tate shifts. Apply Theorem 1.5. ◻

The available construction of correspondence transfer also has a model-sensitive form. Theorem 8.5 requires a regular integral projective flat target model after finite extension, together with a regular semistable abelian model and the Chow identity. Theorem 8.7 supplies the models for products. This realizes the product argument through concrete numerical certificates, rather than assuming the abstract kernel property (C) for every correspondence.

1.1. How the local construction is assembled

The curvature construction precedes the support construction. Model Chern forms and the numerical detector are developed in Section 3. The restricted trace in Section 4 then handles original scalar–curvature generators; its one-parameter formula recovers the scalar potential without expanding its domain to arbitrary actual forms. In Section 5, supported perfect complexes and finite-flat multisections produce the prescribed residue and exact-open Green pair. The construction draft derives (A1) by producing global primitives for generic Chow-zero characters.

These operations are applied in two ways. Section 6 constructs primitives from compact additive families. Section 7 constructs them from semistable numerical certificates and traces the original generator expressions along composite finite maps. Finally, Section 8 transports certificates by algebraic intersection theory on regular models and proves the product consequences.

2. Conventions and cited foundations

For a projective flat model X/K∘\mathcal X/K^\circ, a model line bundle L\mathcal L determines a PL metric on its generic fibre. Its actual curvature c1(L‾)=D(−log⁡∥e∥)c_1(\overline L)=D(-\log\|e\|) has pure tridegree (0,0,1)(0,0,1); pure tridegree (0,0,r)(0,0,r) has bidegree (r,r)(r,r). Special-fibre intersections are rational and operational. Native tropical calibrations and multiplicities are retained in integration. Finite extensions carry the extension of the fixed absolute value.

The local arguments use Mihatsch’s ordinary sheaf and analytic pullback [1, Definition 4.2 and Theorem 4.5], the witnessing-skeleton, restriction and support statements [1, Lemma 4.4, Corollary 4.9 and Proposition 4.11], native finite-flat top-degree integration [1, Lemma 3.15 and Definition 4.10], and scalar Poincaré–Lelong [1, Theorem 5.5 and Corollary 5.6]. Polyhedral projection and pullback use [2, Section 2 and Proposition/Definition 4.2] and [1, Proposition 2.38 and Definition 2.41]. The analytifications in these applications are pure-dimensional and boundaryless.

The model detector uses scalar uniform density of rational model functions on proper analytifications [4, Proposition 2.15]. Its measure calculation uses the recursive model intersections and component-mass formula of Chambert-Loir–Ducros [3, Sections 8.7–8.8 and Theorem 18.5.5]. The localized Riemann–Roch calculation is the Cartier case of [5, Corollary 18.1.2].

The algebraic arguments use rational Chern characters on regular schemes, vector-bundle resolutions on smooth projective varieties, supported proper and refined Gysin identities, Bertini, miracle flatness and characteristic-zero resolution. The relevant references are [13; 7; 8; 9]. Curve-parameter generation of algebraic equivalence is used as in [15], and normalized Poincaré bundles as in [14]. The precise semistable component statement needed for abelian varieties is recalled in Theorem 7.4.

3. Actual Chern forms of model vector bundles

Let KK be a complete nontrivially discretely valued field, let R=K∘R=K^\circ, and let X\mathcal X be an integral projective flat RR-scheme with smooth projective generic fiber XX of pure dimension nn. All analytic forms below are Mihatsch actual delta-forms on the ordinary topology of XanX^{\rm an}; their sheaf is denoted B\mathcal B and D=d′d′′D=d'd''. In the intrinsic trigrading, a form in B0,0,r\mathcal B^{0,0,r} has bidegree (r,r)(r,r). The special fiber and its components need not be smooth, reduced, or normal. All Chow operations are rational and operational unless otherwise stated.

3.1. A scalar-density detector for pure delta forms

Lemma 3.1 (Detection with model Hessians). Let Ω∈B0,0,r(Xan)\Omega\in\mathcal B^{0,0,r}(X^{\rm an}) satisfy d′Ω=d′′Ω=0d'\Omega=d''\Omega=0, with 0≤r≤n0\leq r\leq n, and put q=n−rq=n-r. Suppose that for every tuple of global rational model functions ϕ1,…,ϕq\phi_1,\ldots,\phi_q, Ω∧Dϕ1∧⋯∧Dϕq=0\Omega\wedge D\phi_1\wedge\cdots\wedge D\phi_q=0 as a signed Radon measure. Then Ω=0\Omega=0 as an actual form. For q=0q=0, the hypothesis is the vanishing of the measure of Ω\Omega.

Proof. An actual top-degree form has a locally finite signed measure: in finitely many polyhedral charts over a compact set this follows by integration of its smooth coefficient densities. A partition of unity gives the same measure as actual integration. If that measure is zero, Mihatsch Proposition 4.11 implies that the top form is zero. This deals with q=0q=0.

Assume that vanishing of the displayed measure has already been proved with kk model-function slots replaced by arbitrary global original-smooth functions. Fix those kk functions and all but one of the remaining model functions, and denote their product with Ω\Omega by TT. Then TT has bidegree (n−1,n−1)(n-1,n-1) and is separately closed. For every model function ϕ\phi and every global original-smooth scalar ψ\psi, two applications of actual Stokes give ∫Xanϕ T∧Dψ=∫Xanψ T∧Dϕ=0.\int_{X^{\rm an}}\phi\,T\wedge D\psi =\int_{X^{\rm an}}\psi\,T\wedge D\phi=0. For fixed T,ψT,\psi, the left-hand side tests the single fixed finite signed measure T∧DψT\wedge D\psi against ϕ\phi. Scalar uniform density therefore implies T∧Dψ=0T\wedge D\psi=0. Repeating this argument proves Ω∧Dχ1∧⋯∧Dχq=0(D)\Omega\wedge D\chi_1\wedge\cdots\wedge D\chi_q=0 \tag{D} for arbitrary global original-smooth functions χi\chi_i. The density step concerns the continuous function ϕ\phi against one fixed signed measure, and consequently requires no convergence of Hessians.

If Ω\Omega is nonzero, choose an nn-dimensional witnessing skeleton in a moment chart, using Mihatsch Definition 3.8 and Lemma 4.4. On an open polyhedron σ\sigma of dimension qq its pure tridegree means that the nonzero coefficient is scalar. Refine the chart and subdivide so that its moment coordinates ua=−log⁡∣ha∣u_a=-\log|h_a| have injective tangent map on σ\sigma; this follows from the finite-fiber property of skeleton tropicalization. At a point in the relative interior of σ\sigma, choose a smooth cutoff equal to one nearby and supported in the moment-chart open set. It globalizes χ=12∑aua2\chi=\tfrac12\sum_a u_a^2 to an original-smooth scalar on XanX^{\rm an}. Near the chosen point, Dχ=∑ad′ua∧d′′uaD\chi=\sum_a d'u_a\wedge d''u_a is positive definite on the tangent space of σ\sigma. Thus (Dχ)q(D\chi)^q is a nonzero volume form there. Coefficientwise wedge with smooth forms now shows that Ω∧(Dχ)q\Omega\wedge(D\chi)^q is a nonzero actual top form. This contradicts ({D}) and Proposition 4.11. ◻

3.2. From operational numerical triviality to actual zero curvature

For clarity, the CLD top measure agrees with the measure of the actual wedge used above. Locally let uu be a PWS model metric weight and let TT be a product of previously treated actual model curvatures. Such a current is locally of order zero by its polyhedral description, so coefficientwise multiplication by the continuous function uu is defined and [uT]=u[T][uT]=u[T]. This equality is tested against original-smooth forms and follows directly from the same order-zero polyhedral integrals on both sides. Separate closedness and Mihatsch Proposition 4.11 give [Du∧T]=D[uT]=D(u[T]).[Du\wedge T]=D[uT]=D\bigl(u[T]\bigr). Induction first identifies the actual wedge with the CLD recursive product for semipositive model factors. For signed model metrics, express them as differences of psh model metrics and use multilinearity. This establishes the comparison over KK itself, before any scalar extension.

Theorem 3.2 (Numerical model curvature vanishing). Let PP be a homogeneous polynomial of degree rr in the first Chern classes of model line bundles L1,…,Lt\mathcal L_1,\ldots,\mathcal L_t on X\mathcal X. Suppose that ξ=P(c1(L1,s),…,c1(Lt,s))∈Ar(Xs)Q\xi=P\bigl(c_1(\mathcal L_{1,s}),\ldots, c_1(\mathcal L_{t,s})\bigr)\in A^r(\mathcal X_s)_{\mathbb Q} has degree zero against every integral rr-dimensional closed subvariety of Xs\mathcal X_s. Then P(c1(L‾1),…,c1(L‾t))=0in B0,0,r(Xan).P\bigl(c_1(\overline L_1),\ldots,c_1(\overline L_t)\bigr)=0 \quad\hbox{in }\mathcal B^{0,0,r}(X^{\rm an}).

Proof. Let Ω\Omega be the displayed actual model polynomial. Given q=n−rq=n-r model functions, choose a common proper model refinement f:Y→Xf:\mathcal Y\to\mathcal X on which they are represented, after clearing denominators, by vertical Cartier divisors D1,…,DqD_1,\ldots,D_q. One may work with formal completions; algebraic models and projective common refinements suffice here. The induced metrics of the original bundles are unchanged by pullback.

For a point xx reducing to the generic point of a special component V⊂YsV\subset\mathcal Y_s, the top wedge Ω∧∏jDϕj\Omega\wedge\prod_jD\phi_j has mass equal to the native CLD weight wY(x)w_{\mathcal Y}(x) times deg⁡(fs∗ξ∩∏j=1qc1(O(Dj)∣V)∩[V]).\deg\left(f_s^*\xi\cap \prod_{j=1}^{q}c_1(\mathcal O(D_j)|_V)\cap[V]\right). The proper operational projection formula identifies this degree with deg⁡(ξ∩fs∗(∏j=1qc1(O(Dj)∣V)∩[V]))=0.\deg\left(\xi\cap f_{s*} \left(\prod_{j=1}^{q}c_1(\mathcal O(D_j)|_V)\cap[V]\right)\right)=0. The pushed-forward class has dimension rr. Terms whose images have smaller dimension have zero proper pushforward; all remaining terms are annihilated by the numerical hypothesis. Operational pullback of a Cartier polynomial is available for arbitrary fsf_s: neither flatness of fsf_s nor smoothness of VV has been used. The CLD support and mass formula now gives a zero measure, with all native multiplicities and defect factors retained. Lemma 3.1 concludes the proof. ◻

3.3. Descent along locally split morphisms

Lemma 3.3 (Descent for a locally split morphism). Let a:Y→Xa:Y\to X be a morphism of pure-dimensional boundaryless analytic spaces. Suppose XX has an ordinary open cover (Ui)(U_i) and morphisms si:Ui→Ys_i:U_i\to Y with a∘si=id⁡Uia\circ s_i=\operatorname{id}_{U_i}. For Θ∈B(Y)\Theta\in\mathcal B(Y) the following are equivalent:

  1. There is an actual form θ\theta on XX with a∗θ=Θa^*\theta=\Theta.

  2. On Y×XYY\times_XY, p1∗Θ=p2∗Θp_1^*\Theta=p_2^*\Theta.

When the fiber product is pure-dimensional and the displayed pullbacks are defined, the descended form is unique and is given by θ∣Ui=si∗Θ\theta|_{U_i}=s_i^*\Theta.

Proof. The forward implication is functoriality. Conversely pull the equality back along (si,sj):Ui∩Uj→Y×XY(s_i,s_j):U_i\cap U_j\to Y\times_XY to obtain equality of the two local forms. They glue by the sheaf property. Pulling the same equality along (id⁡Y,si∘a):a−1(Ui)→Y×XY(\operatorname{id}_Y,s_i\circ a):a^{-1}(U_i)\to Y\times_XY proves Θ=a∗si∗Θ\Theta=a^*s_i^*\Theta on a−1(Ui)a^{-1}(U_i), hence Θ=a∗θ\Theta=a^*\theta. The local sections also prove uniqueness and injectivity of a∗a^* on actual forms. ◻

Theorem 3.4 (Actual model Chern forms by local flag sections). Let E1,…,Es\mathcal E_1,\ldots,\mathcal E_s be vector bundles on X\mathcal X and PP a homogeneous rational polynomial of degree rr in their Chern classes. There is a unique actual form cX(P)∈B0,0,r(Xan)c_{\mathcal X}(P)\in\mathcal B^{0,0,r}(X^{\rm an}) whose pullback to the simultaneous full flag bundle equals the corresponding symmetric polynomial in the model first Chern forms of the tautological line quotients. The form is separately closed, independent of the chosen splitting presentation, and functorial under morphisms of projective flat models with smooth projective generic fibers. Universal operational identities, in particular Whitney and the additivity of the Chern character in exact sequences of model bundles, hold as strict identities of actual forms.

If the special-fiber operational class P(c(Ej,s))P(c(\mathcal E_{j,s})) is numerically zero against every rr-cycle, then cX(P)=0c_{\mathcal X}(P)=0.

Proof. Put a:F=∏XFlag⁡(Ej)→Xa:\mathcal F=\prod_{\mathcal X}\operatorname{Flag}(\mathcal E_j) \to\mathcal X and denote its generic fiber by a:F→Xa:F\to X. Let ΩF\Omega_F be the symmetric model-root polynomial. On F×XF\mathcal F\times_{\mathcal X}\mathcal F, the two root polynomials represent the same operational Chow class by the two tautological filtrations and Whitney’s identity. Their difference is therefore operationally zero on the special fiber, before taking a degree or capping with the fundamental class. Theorem 3.2, applied to the smooth projective generic fiber F×XFF\times_XF, gives p1∗ΩF=p2∗ΩFp_1^*\Omega_F=p_2^*\Omega_F as actual forms.

Simultaneously trivialize the finitely many generic bundles EjE_j on a Zariski open cover of XX. Their standard full flags give algebraic sections si:Ui→Fs_i:U_i\to F, hence analytic sections on ordinary opens UianU_i^{\rm an}. These sections need not extend to all of the model and need not preserve a model trivialization. This is harmless: they pull back an already defined global actual form, and Mihatsch Theorem 4.5 applies to arbitrary analytic morphisms. Applying Lemma 3.3 defines cX(P)∣Uian=si∗ΩFc_{\mathcal X}(P)|_{U_i^{\rm an}}=s_i^*\Omega_F and proves its characterizing pullback equality and uniqueness. Pullback preserves the intrinsic trigrading and the differential operators, which proves purity and separate closedness.

For any operational identity among finitely many model bundles, use their simultaneous flag bundle. The difference of the root polynomials is operationally zero, so the numerical detector kills it upstairs. Faithfulness supplied by local sections gives the identity downstairs. The same argument on the product of two flag presentations proves independence of presentation. Functoriality follows by pulling the model flags back and then using their characteristic equality and the faithfulness of the new flag pullback.

Finally suppose ξ=P(c(Ej,s))\xi=P(c(\mathcal E_{j,s})) is numerically zero. For every integral rr-cycle WW on Fs\mathcal F_s, the operational projection formula gives deg⁡(as∗ξ∩[W])=deg⁡(ξ∩as∗[W])=0.\deg(a_s^*\xi\cap[W])= \deg(\xi\cap a_{s*}[W])=0. If the image drops dimension then as∗[W]=0a_{s*}[W]=0; otherwise the numerical hypothesis applies. Thus the root polynomial is numerically zero on the whole flag special fiber. Theorem 3.2 gives ΩF=0\Omega_F=0, and local sections give cX(P)=0c_{\mathcal X}(P)=0. Although asa_s happens to be smooth and flat here, the projection argument is the operational one and does not use a smooth special fiber. ◻

The construction gives a homomorphism K0(X)Q⟶⨁r=0nB0,0,r(Xan),[E]⟼∑rcX(ch⁡rE).K_0(\mathcal X)_{\mathbb Q}\longrightarrow \bigoplus_{r=0}^{n}\mathcal B^{0,0,r}(X^{\rm an}), \qquad [\mathcal E]\longmapsto \sum_r c_{\mathcal X}(\operatorname{ch}_r\mathcal E). Each degree factors through its special-fiber operational numerical class. This statement concerns classes of model bundles. The dependence on the model is controlled by the numerical class of its special fiber.

4. Finite traces of scalar–curvature forms

The trace will be constructed for the scalar–curvature module generated by products u0 Du1⋯Duk,D=d′d′′,(MT1)u_0\,Du_1\cdots Du_k,\qquad D=d'd'', \tag{MT1} where the uiu_i are piecewise linear analytic potentials.

Throughout this section KK is a complete nontrivially valued field of characteristic zero, and all analytic spaces are the analytifications of normal integral separated KK-varieties, with their ordinary topology. All forms are Mihatsch actual delta-forms. A PL scalar means a function locally presented by a finite piecewise affine function of logarithms of analytic units. Its DD is a pure form of tridegree (0,0,1)(0,0,1). Let Mk\mathscr M^k be the actual image sheaf of finite sums of ({MT1}). Relations in Mk\mathscr M^k are actual-form relations; no chosen presentation is part of a section.

4.1. Native degree under finite maps

For a finite dominant map q:Y→Xq:Y\to X of generic degree dd and any compact actual top form τ\tau on XX, these inputs give ∫Yq∗τ=d∫Xτ.(MT2)\int_Yq^*\tau=d\int_X\tau. \tag{MT2} Indeed remove a proper closed subset outside which qq is finite flat of constant degree dd. The support of τ\tau avoids this subset by Corollary 4.9; so does the support of its pullback above it. Lemma 3.15 and the native skeletal integration formula apply on the remaining open. It also proves that finite dominant pullback on actual forms is faithful: pair a putatively zero pullback with pulled-back original-smooth complementary tests, use ({MT2}), and apply Proposition 4.11.

4.2. Descent of scalar PL functions through finite quotients

Lemma 4.1 (Ordinary local scalar PL descent). Let q:Y→Xq:Y\to X be finite Galois between normal integral varieties, with group GG. A GG-invariant source PL scalar descends to a PL scalar on XX. If finitely many invariant potentials on Yan×(Gm)anY^{\rm an}\times(\mathbf G_m)^{\rm an} are jointly PL in fixed source unit logarithms and the parameter t=v(z)t=v(z), their descended potentials are locally jointly PL in a fixed finite tuple of base-unit logarithms and tt. The base tuple and base shrinking are independent of tt.

Proof. Work around a whole finite analytic fiber. Choose disjoint ordinary neighborhoods of its points on which the finitely many source potentials have moment presentations. Properness permits shrinking the base so that its full inverse image lies in their union. The corresponding source pieces are then clopen. Analytic units can be pasted on these pieces. Add constant units with distinct valuations as markers; after shrinking compact neighborhoods, their moment images lie in separated compact polyhedral boxes. Finite PL extension on these boxes combines the presentations. A smooth cutoff of a PL potential is neither required nor used.

Include the full GG-orbit of the resulting source-unit tuple. Its logarithms ℓ∈Rs\ell\in\mathbf R^s then carry a coordinate-permutation GG-action; add a constant unit if necessary so that s≥1s\geq1. If an ambient PL presentation φ(ℓ)\varphi(\ell) is invariant only after restriction to the actual source moment image, replace it by ∣G∣−1∑g∈Gφ(gℓ)|G|^{-1}\sum_{g\in G}\varphi(g\ell). This is an ambient invariant PL function with the same source restriction. For ordered orbit potentials the sorting itself already gives this ambient invariance. The same operation leaves an independent parameter fixed. Put d=∣G∣d=|G| and N=d(s−1)+1N=d(s-1)+1. For distinct integers aja_j take the Vandermonde directions mj=(1,aj,…,ajs−1)∈Zs.m_j=(1,a_j,\ldots,a_j^{s-1})\in\mathbf Z^s. Define J(ℓ)J(\ell) by recording, for each jj, the ordered multiset {⟨mj,gℓ⟩:g∈G}\{\langle m_j,g\ell\rangle:g\in G\}. It separates GG-orbits. In fact if J(x)=J(y)J(x)=J(y), for each jj one can choose gjg_j with ⟨mj,x⟩=⟨mj,gjy⟩\langle m_j,x\rangle=\langle m_j,g_jy\rangle. One gg occurs for at least ss indices. The resulting polynomial of degree at most s−1s-1 has ss roots, hence x=gyx=gy.

The map JJ is proper: any ss independent Vandermonde directions bound ℓ\ell when all the displayed coordinates are bounded. It is finite piecewise linear. On a common polyhedral subdivision its linearity cells map affinely and injectively to their images; a noninjective tangent direction would produce an infinite fiber. Every invariant PL scalar therefore factors as a PL scalar through JJ, after common image subdivision and PL ambient extension. The identical argument applies to J×id⁡RJ\times\operatorname{id}_{\mathbf R} and jointly PL potentials with the fixed parameter tt.

It remains to realize JJ downstairs without an illicit nonlinear pullback of forms. For each source monomial Fj=fmjF_j=f^{m_j} take its monic orbit polynomial ∏g∈G(T−gFj).\prod_{g\in G}(T-gF_j). Its coefficients are invariant analytic functions, hence analytic functions on the normal finite quotient. Algebraically the quotient is XX; analytically this follows on affinoids from finite coherent base change and the Reynolds idempotent in characteristic zero. With the convention that the coefficient of TiT^i gives the point (i,v(ai))(i,v(a_i)), the ordered root valuations are the negatives of the lower Newton polygon’s slopes, with multiplicity. They are therefore PL in the valuations of the coefficients that contribute to that polygon.

At a fixed base point retain the nonzero coefficients and shrink so that they are units. The constant coefficient a0a_0 is a unit because it is a product of source units; the leading coefficient is one. Choose a uniform upper bound BB for v(a0)v(a_0) on the shrunken base. Every coefficient vanishing at the fixed point can, by continuity and a further common shrinking, be made to have valuation greater than max⁡(B,0)+1\max(B,0)+1. Its point is then above the endpoint chord of the Newton polygon and cannot affect the lower hull. Such coefficients can be omitted uniformly. Thus JJ is PL in finitely many base-unit logarithms. These coefficients involve source units alone, so their choice and shrinking are independent of tt. Substitution is only a substitution into a scalar PL function. Taking its actual derivatives happens afterwards; no nonlinear PL pullback of a general delta-form has been postulated. ◻

4.3. Sorting gives a pure-curvature trace

Lemma 4.2 (Max/min identities). For source PL scalars u,vu,v, put M=max⁡(u,v)M=\max(u,v) and m=min⁡(u,v)m=\min(u,v). For every k≥0k\geq0 there are actual identities (DM)k+(Dm)k=(Du)k+(Dv)k,(MT3)(DM)^k+(Dm)^k=(Du)^k+(Dv)^k, \tag{MT3} M(DM)k+m(Dm)k=u(Du)k+v(Dv)k.(MT4)M(DM)^k+m(Dm)^k=u(Du)^k+v(Dv)^k. \tag{MT4}

Proof. Write h=u−vh=u-v, h+=max⁡(h,0)h_+=\max(h,0), h−=min⁡(h,0)h_-=\min(h,0), and A=Dh+A=Dh_+, B=Dh−B=Dh_-. Coefficientwise multiplication on support polyhedra gives h+B=h−A=0h_+B=h_-A=0: on one open side the appropriate function is zero, and on the common wall its continuous scalar coefficient is zero. Applying DD and using separate closedness gives AB=0AB=0. Since M=v+h+,m=v+h−,u=v+h++h−,M=v+h_+,\quad m=v+h_-,\quad u=v+h_++h_-, expansion proves ({MT3}). The same expansion, now using both AB=0AB=0 and the two scalar-weighted zero identities, proves ({MT4}). ◻

The second identity is useful but has a definite limitation: polarizing u(Du)ku(Du)^k recovers symmetrized energies, not an arbitrary ordered scalar-curvature monomial u0Du1⋯Duku_0Du_1\cdots Du_k. The parameter argument below supplies precisely that missing step.

Lemma 4.3 (Trace of pure PL curvature polynomials). For finite dominant q:Y→Xq:Y\to X between normal integral varieties, a finite sum CC of products Du1⋯DukDu_1\cdots Du_k has an intrinsic actual trace Tr⁡qcurv(C)\operatorname{Tr}^{\rm curv}_q(C), characterized by ∫XTr⁡qcurv(C)∧η=∫YC∧q∗η(MT5)\int_X\operatorname{Tr}^{\rm curv}_q(C)\wedge\eta =\int_YC\wedge q^*\eta \tag{MT5} for all compact complementary actual tests. With a parameter, jointly PL source potentials in fixed source logarithms and tt yield a trace with a product-moment presentation using fixed base logarithms and tt.

Proof. First let qq be Galois. Sort the orbit of a scalar uu by a finite sorting network. Repeated use of ({MT3}) gives ∑g∈G(Dg∗u)k=∑j=1d(Du(j))k.\sum_{g\in G}(Dg^*u)^k =\sum_{j=1}^d(Du_{(j)})^k. The ordered potentials u(j)u_{(j)} are invariant and descend by Lemma 4.1. The right side is consequently the pullback of a downstairs actual form. Define it to be the trace of (Du)k(Du)^k. Polarization, Du1⋯Duk=1k!∑I⊂{1,…,k}(−1)k−∣I∣(D∑i∈Iui)k,Du_1\cdots Du_k =\frac1{k!}\sum_{I\subset\{1,\ldots,k\}} (-1)^{k-|I|}\left(D\sum_{i\in I}u_i\right)^k, does the same for all mixed products. For k=0k=0 set Tr⁡qcurv(1)=d\operatorname{Tr}^{\rm curv}_q(1)=d.

This construction satisfies the actual orbit identity q∗Tr⁡qcurv(C)=∑g∈Gg∗C.q^*\operatorname{Tr}^{\rm curv}_q(C) =\sum_{g\in G}g^*C. Wedge with q∗ηq^*\eta, integrate, and use ({MT2}) and invariance of native integration under automorphisms. Division by dd proves ({MT5}). Finite pullback faithfulness proves independence of every actual relation. Current injectivity gives gluing on ordinary overlaps. The pairing permits arbitrary actual downstairs multipliers; they have not been treated as arbitrary source coefficients.

For a general separable finite map take a normal Galois closure Y~→X\widetilde Y\to X factoring through r:Y~→Yr:\widetilde Y\to Y, and divide the Galois trace of r∗Cr^*C by e=deg⁡re=\deg r. Applied to the top form Cq∗ηCq^*\eta, ({MT2}) proves the required pairing and independence of the closure. Characteristic zero covers every map used here. All spaces and parameter tori remain over the original KK; native generic degrees are not normalized away.

For the last assertion use the parameter version of Lemma 4.1 at the sorting step. The descended potentials have fixed base moments, so their Hessians and products have the required product-moment presentation. ◻

4.4. Integration along a parameter torus

4.4.0.1. The boundary issue.

Compact analytic support does not imply that an ambient moment coefficient avoids the tropicalized analytic boundary of an affinoid chart. We do not use the single-chart integral formula with that unverified hypothesis. Instead we first prove a pushforward identity for polyhedral currents on the product skeleton, retaining the boundary exclusion of Proposition 2.38 at every chart calculation. The intrinsic skeletal definition of integration then gives the analytic pairing. In this subsection f⋆f^\star denotes tropical-space pullback in that proposition, distinguished from the notation tc∗t_c^* for the resulting actual analytic form.

Lemma 4.4 (The coordinate product skeleton, including its structures). Let VV be a boundaryless ordinary analytic open of pure dimension nn, let bb be a tuple of analytic units on VV, and put S=Σ(V,b)S=\Sigma(V,b). The relative Gauss section identifies ιb:S×R → ∼  Σ(V×(Gm)an,(b,z)),(y,t)⟼νt(y),(PS1)\begin{split} \iota_b:S\times\mathbf R&\ \xrightarrow{\ \sim\ }\ \Sigma(V\times(\mathbf G_m)^{\rm an},(b,z)),\\ (y,t)&\longmapsto \nu_t(y), \end{split} \tag{PS1} where νt(y)\nu_t(y) is the Gauss point of radius e−te^{-t} in the torus over H(y)\mathcal H(y). This is an isomorphism of weighted pwl spaces and of tropical spaces. Its weight is the product of the native weight on SS and the standard weight on R\mathbf R, and its linear function sheaf is generated by the base linear functions and tt. The identification is compatible with every ordinary base shrinking and every additional base moment refinement.

Proof. Let xx lie above yy in the analytic product. The graded field generated by bb and zz has transcendence degree n+1n+1 precisely when the graded field generated by bb at yy has transcendence degree nn and z~\widetilde z is transcendental over H(y)~\widetilde{\mathcal H(y)}. Indeed adjoining zz increases that degree by at most one; in the degree-nn case H(y)~\widetilde{\mathcal H(y)} is algebraic over the graded field generated by bb. This is the graded-transcendence calculation in (3.27) of the cited proof, here with the single coordinate zz.

The condition on z~\widetilde z characterizes the Gauss valuation over H(y)\mathcal H(y): no cancellation among leading monomials is possible, and for every Laurent polynomial one has ∣∑iaizi∣νt(y)=max⁡i∣ai∣e−it.\left|\sum_i a_i z^i\right|_{\nu_t(y)} =\max_i |a_i|e^{-it}. Conversely these Gauss valuations have the required graded transcendence degree. Thus Definition 3.8 and the absence of analytic boundary give the claimed bijection. No valued extension of KK, or rationality of the radius in its value group, is needed.

The inverse is (p,v(z))(p,v(z)) on the displayed skeleton. It is pwl: a base skeleton chart is described by an affinoid domain and valuations of its analytic units; their pullbacks are valuations of the same units on the product, and v(z)v(z) is the extra coordinate. The corresponding closed chart domains pull back to closed pwl domains. This inverse is also proper. A compact subset of S×RS\times\mathbf R is contained in a finite union of compact base analytic domains times a bounded interval. Its inverse image is closed in the corresponding finite union of products with compact analytic annuli, and is therefore compact. It follows that the continuous bijective inverse is a homeomorphism. Subdivision into local polyhedral charts makes its injective affine pieces invertible onto their images, so both directions are pwl. This is the coordinate specialization of the topology and atlas argument following (3.31), now without its changes of field or center.

For the weights, take a top-dimensional base cell and a local toric projection used in Definition 3.8(2) to compute its native degree ee. Its product with the identity on Gm\mathbf G_m is finite flat of the same degree ee. Hence its induced weight is the base weight times the standard parameter weight. This calculation retains residue-field degrees and is exactly the last weight calculation in the proof of Proposition 3.21. It is local on cells and hence determines the entire product weight.

Under the bijection the moments are (tb,t)(t_b,t). Definition 3.8(3) therefore gives precisely the product linear function sheaf, not merely a set-theoretic identification. For additional base moments b′b', the same Gauss section applies. The inclusions of base and product skeletons agree under it, and their weights agree by Remark 3.9. The new linear functions are the new base ones together with the same tt. This proves the stated compatibility for all refinements and ordinary restrictions. ◻

Lemma 4.5 (Tropical projection with boundary control). Let SS be a pure-dimensional tropical space and f:S→RNf:S\to\mathbf R^N a linear map. Give S×RS\times\mathbf R the product weight and linear functions. Write pSp_S for its projection and π:RN×R→RN\pi:\mathbf R^N\times\mathbf R\to\mathbf R^N for the coordinate projection. If Γ\Gamma is a Euclidean delta-form with support in a fixed compact set of parameter values, then (pS)∗((f,id⁡)⋆Γ)=f⋆(π∗Γ)(PS2)(p_S)_*\bigl((f,\operatorname{id})^\star\Gamma\bigr) =f^\star(\pi_*\Gamma) \tag{PS2} as polyhedral currents on SS. The left side uses only the proper-on-support pwl-current pushforward of Proposition 2.17.

Proof. The product is a tropical space: on a product polyhedral subdivision, the vertical balancing conditions are those of SS, and horizontal internal facets cancel in pairs with the standard parameter weight. The support of B=(f,id⁡)⋆ΓB=(f,\operatorname{id})^\star\Gamma is contained in S×J0S\times J_0 for a compact interval J0J_0. Thus it is proper over SS, and Proposition 2.17 supplies the polyhedral current P=(pS)∗BP=(p_S)_*B before any assertion that it is a delta-form.

Work in a polyhedral neighborhood carrying a finite-fiber linear map hh to Rm\mathbf R^m, as supplied by the axioms of a sheaf of linear functions. Append ff and put g=(f,h)g=(f,h). Then gg has finite fibers and f=A∘gf=A\circ g, where AA is a literal coordinate projection. Let HH be any compact polyhedral subset of the interior of this neighborhood, and put EH=g(∂SH),WH=RN+m∖EH,Hg∘=H∖g−1(EH).E_H=g(\partial_S H),\qquad W_H=\mathbf R^{N+m}\setminus E_H,\qquad H_g^\circ=H\setminus g^{-1}(E_H). The restricted g:Hg∘→WHg:H_g^\circ\to W_H is proper. Write CH=g∗[Hdim⁡S,μS]∣WHC_H=g_*[H_{\dim S},\mu_S]|_{W_H}. It is a tropical cycle on WHW_H by Corollary 2.35.

Choose an interval J=[a,b]J=[a,b] containing J0J_0 in its interior. The boundary of H×JH\times J is contained in (∂SH×J) ∪ (H×{a,b}).(\partial_S H\times J)\ \cup\ (H\times\{a,b\}). Apply the characterizing identity (2.40) of Proposition 2.38 to this compact product and the refinement g×id⁡g\times\operatorname{id}. On the boundary-free region WH×(a,b)W_H\times(a,b) it gives (g×id⁡)∗(B∣Hg∘×(a,b))=(CH×R)∧(A×id⁡)∗Γ.(PS3)(g\times\operatorname{id})_* \bigl(B|_{H_g^\circ\times(a,b)}\bigr) =(C_H\times\mathbf R)\wedge (A\times\operatorname{id})^*\Gamma. \tag{PS3} Both sides vanish near the artificial parameter endpoints because Γ\Gamma does. They may therefore be viewed on WH×RW_H\times\mathbf R, but the exclusion of EHE_H in the base remains in force. There is no assertion of this chart identity across g(∂SH)g(\partial_S H).

Let πg:WH×R→WH\pi_g:W_H\times\mathbf R\to W_H be projection. Functoriality of pwl-current pushforward and the Euclidean projection formula give, entirely on WHW_H, g∗(P∣Hg∘)=πg,∗ ⁣((CH×R)∧(A×id⁡)∗Γ)=CH∧πg,∗(A×id⁡)∗Γ=CH∧A∗(π∗Γ).(PS4)\begin{split} g_*\bigl(P|_{H_g^\circ}\bigr) &=\pi_{g,*}\!\left((C_H\times\mathbf R)\wedge (A\times\operatorname{id})^*\Gamma\right)\\ &=C_H\wedge \pi_{g,*}(A\times\operatorname{id})^*\Gamma\\ &=C_H\wedge A^*(\pi_*\Gamma). \end{split} \tag{PS4} The last step needs only coordinate Fubini. For a coordinate projection AA, its Euclidean delta-pullback adds free standard coordinates. On each weighted carrier of Γ\Gamma, fiber integration in tt before or after adding those coordinates has the same fibers, coefficient integral and quotient weights. Thus πg,∗(A×id⁡)∗Γ=A∗(π∗Γ)\pi_{g,*}(A\times\operatorname{id})^*\Gamma=A^*(\pi_*\Gamma). This carrierwise argument applies to an arbitrary Euclidean delta-form, with compact parameter support; it does not invoke a general affine base-change lemma.

The last expression in ({PS4}) is exactly the characterizing expression (2.40) for f⋆(π∗Γ)f^\star(\pi_*\Gamma). Varying the compact polyhedral set HH inside the fixed finite-fiber gg-chart, Lemma 2.37 shows that P=f⋆(π∗Γ)P=f^\star(\pi_*\Gamma) on that chart. These charts cover SS, which proves ({PS2}). In particular the polyhedral pushforward has the asserted delta-form representative; this is a conclusion of the calculation. ◻

Lemma 4.6 (Integration on the product-moment subspace). Let p:U×(Gm)an→Up:U\times(\mathbf G_m)^{\rm an}\to U be projection. Suppose β=t(c,z)∗Γ\beta=t_{(c,z)}^*\Gamma for a fixed base analytic-unit tuple cc, t=v(z)t=v(z), and a Euclidean delta-form Γ(u,t)\Gamma(u,t) supported in a fixed compact set of tt. With π(u,t)=u\pi(u,t)=u, define Ip(β)=tc∗(π∗Γ).(MT6)I_p(\beta)=t_c^*(\pi_*\Gamma). \tag{MT6} This is intrinsic among product-moment presentations and satisfies ∫UIp(β)∧η=∫U×Gmβ∧p∗η(MT7)\int_UI_p(\beta)\wedge\eta =\int_{U\times\mathbf G_m}\beta\wedge p^*\eta \tag{MT7} for every compact actual complementary test. It commutes with d′,d′′d',d'' on these presentations.

Proof. 1. A skeletal identity before integration. Euclidean pushforward supplies the actual candidate θ=tc∗(π∗Γ)\theta=t_c^*(\pi_*\Gamma). On an arbitrary ordinary V⊆UV\subseteq U, refine the base moments by tuple concatenation, b=(c,h)b=(c,h), and put S=Σ(V,b)S=\Sigma(V,b). Under ({PS1}), the restriction of β\beta to the product skeleton is (tc∣S,id⁡)⋆Γ(t_c|_S,\operatorname{id})^\star\Gamma. Lemma 4.5 therefore gives θ∣S=(pS)∗(β∣S×R).(PS5)\theta|_S =(p_S)_*\bigl(\beta|_{S\times\mathbf R}\bigr). \tag{PS5} This is an equality of intrinsic polyhedral currents on SS for every base refinement. Its proof used the boundary-excluded chart identities ({PS3})–({PS4}), not an integral over a global tropicalization of VV.

If two product-moment presentations define the same actual β\beta, concatenate both base tuples and any further test tuple. Their restrictions to every resulting product skeleton agree by Definition 4.2. Equation ({PS5}) makes their candidate outputs agree on every corresponding base skeleton. Lemma 4.4, using all refinements, gives actual equality. This proves independence of presentations and their relations, and also their compatibility on ordinary overlaps, before any current pairing.

2. Products with arbitrary actual base forms. On an open VV where an actual form η\eta has a moment presentation, take bb to be the concatenation of its moments and cc, and write η=tb∗ηb\eta=t_b^*\eta_b. Let AA be the coordinate projection selecting cc. The product β∧p∗η\beta\wedge p^*\eta has Euclidean coefficient (A×id⁡)∗Γ∧πb∗ηb(A\times\operatorname{id})^*\Gamma\wedge\pi_b^*\eta_b in the fixed moments (b,z)(b,z), with the same compact parameter support. Euclidean projection and coordinate Fubini give πb,∗ ⁣((A×id⁡)∗Γ∧πb∗ηb)=A∗(π∗Γ)∧ηb.\pi_{b,*}\!\left( (A\times\operatorname{id})^*\Gamma\wedge\pi_b^*\eta_b\right) = A^*(\pi_*\Gamma)\wedge\eta_b. Apply ({PS5}) to this coefficient. For S=Σ(V,b)S=\Sigma(V,b) it gives (pS)∗((β∧p∗η)∣S×R)=(θ∧η)∣S.(PS6)(p_S)_* \left((\beta\wedge p^*\eta)|_{S\times\mathbf R}\right) =(\theta\wedge\eta)|_S. \tag{PS6} The coefficient ηb\eta_b here is an arbitrary Euclidean delta-form. No restriction to smooth tests or to products of model curvatures has occurred.

3. Intrinsic skeletal integration and ordinary partitions. Now let η\eta be compact and of complementary bidegree. Cover its support by finitely many ordinary opens ViV_i on which it has the presentations used in the preceding step, and choose an original-smooth partition (λi)(\lambda_i) subordinate to this cover on a neighborhood of the support. Put Si=Σ(Vi,bi)S_i=\Sigma(V_i,b_i). On ViV_i the unweighted form θ∧η\theta\wedge\eta is top-degree and is presented in bib_i; on Vi×GmV_i\times\mathbf G_m the unweighted form β∧p∗η\beta\wedge p^*\eta is top-degree and is presented in (bi,z)(b_i,z). Corollary 4.9(2) puts their supports in SiS_i and Si×RS_i\times\mathbf R, respectively.

Multiplication by λi\lambda_i or p∗λip^*\lambda_i cannot enlarge these supports. Their paired supports are compact: the relevant base support is compact inside ViV_i, and the parameter values lie in a fixed compact interval, hence in a compact analytic annulus. Finitely many affinoid neighborhoods of the base support give compact total paired support. This compactness justifies the integrals; it is not used to exclude any tropicalized base boundary.

The restriction λi∣Si\lambda_i|_{S_i} need only be PWS. Remark 4.3(1) makes its multiplication coefficientwise, and Proposition 2.17 gives the coefficientwise projection formula (pSi)∗ ⁣(pSi∗(λi∣Si) T)=(λi∣Si)(pSi)∗T(p_{S_i})_*\!\left(p_{S_i}^*(\lambda_i|_{S_i})\,T\right) =(\lambda_i|_{S_i})(p_{S_i})_*T for the relevant polyhedral current TT. Use Definition 4.10(1) to evaluate the analytic top integrals on these containing skeletons, followed by (2.17) and the integrable current pushforward formula in Definition 4.10(4). With ({PS6}), this gives ∫U×Gmp∗λi β∧p∗η=∫Si×RpSi∗(λi∣Si)(β∧p∗η)∣Si×R=∫Si(λi∣Si)(θ∧η)∣Si=∫Uλi θ∧η.\begin{split} \int_{U\times\mathbf G_m}p^*\lambda_i\,\beta\wedge p^*\eta &=\int_{S_i\times\mathbf R} p_{S_i}^*(\lambda_i|_{S_i}) (\beta\wedge p^*\eta)|_{S_i\times\mathbf R}\\ &=\int_{S_i}(\lambda_i|_{S_i})(\theta\wedge\eta)|_{S_i}\\ &=\int_U\lambda_i\,\theta\wedge\eta. \end{split} Definition 4.10(2) permits summing this identity over the ordinary partition. It proves ({MT7}) for every compact actual test. The partition is not required to have a single Euclidean presentation in the preassigned tuple bib_i.

Finally, Euclidean proper-on-support pushforward commutes with d′,d′′d',d'', and differentiation retains the compact parameter support. Applying this to ({MT6}), whose actual independence is already proved, gives the asserted differential compatibility. Equivalently it follows from the linear-map current identity of Definition 2.29(3) in ({PS5}). ◻

Lemma 4.7 (One-parameter evaluation of a nonlinear PL scalar). Let uu be a bounded PL scalar on an ordinary open VV. Choose χ∈Cc∞(R)\chi\in C_c^\infty(\mathbf R) equal to one near its range. On V×(Gm)anV\times(\mathbf G_m)^{\rm an} put H=max⁡(u−t,0).H=\max(u-t,0). For every actual form CC on VV with a common moment presentation, Ip(χ(t)t DH∧p∗C)=uC.(MT8)I_p\bigl(\chi(t)t\,DH\wedge p^*C\bigr)=uC. \tag{MT8} Only one parameter occurs, independently of the number of source moments needed to present uu.

Proof. Write u=φ(ℓ)u=\varphi(\ell) for a finite Euclidean PL scalar. Work first on its Euclidean moment domain with H(x,t)=max⁡(φ(x)−t,0)H(x,t)=\max(\varphi(x)-t,0). On each cell where φ\varphi is affine, the graph face t=φ(x)t=\varphi(x) in DHDH has projection weight one. The coefficient of tt in the defining affine difference is −1-1, so projection of its calibrated divisor measure is the standard base measure. Weighting by χ(t)t\chi(t)t gives χ(φ(x))φ(x)\chi(\varphi(x))\varphi(x).

There are also vertical faces above walls of φ\varphi, in the region t<φ(x)t<\varphi(x). It would be false to identify DHDH with the graph alone. On every such vertical face, however, projection drops the dimension of its pure carrier; its pure Euclidean pushforward is zero, exactly as in Mihatsch’s Example 2.5. Thus π∗(χ(t)t Dmax⁡(φ−t,0))=χ(φ)φ.\pi_*\bigl(\chi(t)t\,D\max(\varphi-t,0)\bigr) =\chi(\varphi)\varphi. The affine-cell equality extends across walls as an equality of actual scalar functions; the right side is continuous and the Euclidean pushforward is a delta-form of bidegree zero.

The affine projection formula now permits an arbitrary delta-form CC: π∗(χ(t)t DH∧π∗C)=π∗(χ(t)t DH)∧C.\pi_*\bigl(\chi(t)t\,DH\wedge\pi^*C\bigr) =\pi_*(\chi(t)t\,DH)\wedge C. Apply Lemma 4.6 after moment refinement. Only after actual moment pullback do we use the cutoff choice: χ(φ(ℓ))φ(ℓ)=χ(u)u=u\chi(\varphi(\ell))\varphi(\ell)=\chi(u)u=u. The ambient extension φ\varphi need not be bounded, so χ(φ)φ=φ\chi(\varphi)\varphi=\varphi is not asserted on its entire Euclidean domain. This argument has not pulled CC through the nonlinear PL graph; only the affine projection pullback and projection formula are used. ◻

4.5. The minimal finite trace theorem

Theorem 4.8 (Finite trace on the scalar-curvature module). Let q:Y→Xq:Y\to X be finite dominant between normal integral separated varieties over KK. Every ω∈Mk(Yan)\omega\in\mathscr M^k(Y^{\rm an}) has a unique actual trace on XanX^{\rm an} satisfying ∫XTr⁡qM(ω)∧η=∫Yω∧q∗η(MT9)\int_X\operatorname{Tr}^{\mathscr M}_q(\omega) \wedge\eta =\int_Y\omega\wedge q^*\eta \tag{MT9} for all compact complementary actual tests. It is local on the ordinary topology, respects actual relations, and retains the full native generic degree. For a locally presented generator ω=u0Du1⋯Duk\omega=u_0Du_1\cdots Du_k, a local formula is Tr⁡qM(ω)=IpX ⁣(χ(t)t Tr⁡q×id⁡curv(Dmax⁡(u0−t,0)∧∏i=1kDui)).(MT10)\boxed{\displaystyle \operatorname{Tr}^{\mathscr M}_q(\omega) = I_{p_X}\!\left( \chi(t)t\, \operatorname{Tr}^{\rm curv}_{q\times\operatorname{id}} \left(D\max(u_0-t,0)\wedge\prod_{i=1}^kDu_i\right) \right).} \tag{MT10}

Proof. Shrink around an entire finite source fiber so that finitely many generator presentations are combined as in Lemma 4.1, and u0u_0 is bounded there. Choose χ=1\chi=1 near that whole range. The curvature trace inside ({MT10}) exists by Lemma 4.3. Its parameter-uniform assertion supplies a product-moment presentation, so multiplication by χ(t)t\chi(t)t and application of IpXI_{p_X} are legitimate.

Pair with an arbitrary compact actual test η\eta. First Lemma 4.6, then ({MT5}) with the actual downstairs multiplier χ(t)t pX∗η\chi(t)t\,p_X^*\eta, and finally Lemma 4.7, give ∫XTr⁡qM(ω)η=∫X×Gmχ(t)t Tr⁡q×id⁡curv(DH∏iDui) pX∗η=∫Y×Gmχ(t)t DH∏iDui pY∗q∗η=∫Yu0∏iDui q∗η.\begin{split} \int_X\operatorname{Tr}^{\mathscr M}_q(\omega)\eta &=\int_{X\times\mathbf G_m}\chi(t)t\, \operatorname{Tr}^{\rm curv}_{q\times\operatorname{id}} (DH\prod_iDu_i)\,p_X^*\eta\\ &=\int_{Y\times\mathbf G_m}\chi(t)t\,DH\prod_iDu_i\, p_Y^*q^*\eta\\ &=\int_Yu_0\prod_iDu_i\,q^*\eta. \end{split} All these integrals have compact paired support. The identity is proved for all actual tests, not merely for model or original-smooth tests. Add generators. If an expression represents zero as an actual source form, its trace has zero pairing with every test, and is zero by Proposition 4.11. The same argument removes dependence on the cutoff, source moments, quotient moments, Galois closure, and all local choices. It also glues the local answers in the ordinary topology and proves uniqueness. ◻

5. Supported perfect complexes and exact-support Green forms

Let KK be complete discretely valued of characteristic zero, R=K∘R=K^\circ, and let X/R\mathcal X/R be integral, flat and projective, with smooth integral projective generic fiber XX of dimension nn. The special fiber may be singular or nonreduced. Let C=(0⟶CN⟶⋯⟶C0⟶0)C=(0\longrightarrow C_N\longrightarrow\cdots \longrightarrow C_0\longrightarrow0) be a bounded complex of model vector bundles, exact on U=X∖Z\mathcal U=\mathcal X\setminus\mathcal Z, with generic open U=X∖ZU=X\setminus Z. Assume ZZ is pure of codimension rr, 1≤r≤n1\leq r\leq n. The same construction is linear in finite rational combinations of such complexes. Define ΩC=∑i(−1)ich⁡r(C‾i)∈B0,0,r(Xan),zC=ch⁡rZ(C∣X)∩[X]∈CH⁡n−r(Z)Q.(PC1)\Omega_C=\sum_i(-1)^i\operatorname{ch}_r(\overline C_i) \in\mathcal B^{0,0,r}(X^{\rm an}), \qquad z_C=\operatorname{ch}^{Z}_r(C|_X)\cap[X] \in\operatorname{CH}_{n-r}(Z)_{\mathbb Q}. \tag{PC1} The first expression is an actual form by Theorem 3.4. The second is a cycle with coefficients, because CH⁡n−r(Z)\operatorname{CH}_{n-r}(Z) is the free group on the top-dimensional components of ZZ. It is not merely an ambient Chow class.

5.1. Finite multisections of flag bundles

Lemma 5.1 (A smooth finite flat multisection). Let F→YF\to Y be a smooth projective surjection of smooth projective integral varieties over the infinite field KK, of pure relative dimension dd. There is a smooth projective integral closed subscheme T0⊂FT_0\subset F whose map to YY is finite flat surjective. It may be taken as a connected component of a smooth complete intersection. If the source is geometrically integral and dim⁡Y≥1\dim Y\geq1, T0T_0 can be chosen geometrically integral.

Proof. Choose a very ample HH on FF and use M=HmM=H^m with m≥max⁡(dim⁡Y,1)m\geq\max(\dim Y,1). For any positive-dimensional integral component AA of any intermediate fiber complete intersection, two sections a0,a1a_0,a_1 of HH have nonconstant ratio on AA. Consequently a0m,a0m−1a1,…,a1ma_0^m,a_0^{m-1}a_1,\ldots,a_1^m are linearly independent on AA. In the parameter space V=H0(F,M)V=H^0(F,M), the sections vanishing identically on AA therefore have codimension at least m+1m+1.

Fix a geometric fiber, which is pure Cohen–Macaulay of dimension dd. Induct on j≤dj\leq d to see that the bad jj-tuples, whose common zero locus has dimension greater than d−jd-j, have codimension at least m+1m+1 in VjV^j. Above a good prefix, a bad next section vanishes on one of the finitely many components just considered; above a bad prefix the inductive bound applies. Upper semicontinuity of fiber dimension makes the universal bad locus closed. Adding dim⁡Y\dim Y still leaves its dimension strictly smaller than dim⁡Vd\dim V^d. Properness of YY makes its projection to VdV^d closed. Thus a nonempty open of dd-tuples has zero-dimensional fibers everywhere. The fibers are nonempty, because dd ample divisors meet every nonempty pure dd-dimensional projective fiber.

Smooth Bertini supplies a second nonempty open on which the total intersection is smooth of dimension dim⁡Y\dim Y. Geometric irreducibility Bertini supplies the stated refinement when relevant. Their intersection has a KK-rational point since KK is infinite. The resulting proper quasi-finite morphism is finite and surjective. Its source is Cohen–Macaulay and its target regular of the same dimension, so miracle flatness applies. If necessary choose a connected component; it is integral, finite flat, and still surjective. ◻

The bound m+1>dim⁡Ym+1>\dim Y, not just Bertini, is what rules out a positive-dimensional fiber anywhere on the base. The primary ingredients are Stacks Tags 0D4I, 0FD6, 0G4C and 00R4.

Lemma 5.2 (Relative diagonalization with the original open fixed). There is a projective morphism of integral flat models f:T→Xf:\mathcal T\to\mathcal X with smooth projective generic fiber TT, such that f:T→Xf:T\to X is generically finite of degree dd and finite flat over the entire prescribed open UU. There are model line maps sα:Lα→Mαs_\alpha:L_\alpha\to M_\alpha, isomorphisms above U\mathcal U, and signs ϵα\epsilon_\alpha, satisfying [f∗C]=∑αϵα[Lα→Mα](PC2)[f^*C] =\sum_\alpha\epsilon_\alpha[L_\alpha\to M_\alpha] \tag{PC2} in supported perfect K0K_0. Their generic divisors Dα=div⁡(sα∣T)D_\alpha=\operatorname{div}(s_\alpha|_T) are supported in f−1Zf^{-1}Z.

Proof. On U\mathcal U, exactness of a bounded vector-bundle complex is locally split. Hence Si∘=ker⁡di=im⁡di+1S_i^\circ=\ker d_i=\operatorname{im}d_{i+1} are subbundles, with S0∘=C0S_0^\circ=C_0 and SN∘=0S_N^\circ=0. Close the graph of this section in the product of the corresponding Grassmann bundles. This gives a projective birational integral model Y→X\mathcal Y\to\mathcal X, an isomorphism over U\mathcal U, with tautological subbundles Si⊂CiS_i\subset C_i and quotient bundles QiQ_i. Every asserted vanishing of a bundle map that holds on the dense open holds on this integral model. The differential therefore factors as ai:Qi⟶Si−1,a_i:Q_i\longrightarrow S_{i-1}, an equal-rank map that is an isomorphism over U\mathcal U.

There is an actual filtration by subcomplexes: its ppth term has CiC_i for i>pi>p, SpS_p in degree pp, and zero below pp. The successive two-term quotients are Qp+1→SpQ_{p+1}\to S_p in degrees p+1,pp+1,p. They are exact on the given open, and yield [C]=∑i(−1)i−1[Qi→Si−1][C]=\sum_i(-1)^{i-1}[Q_i\to S_{i-1}] in supported K0K_0, rather than only in ordinary K0K_0.

Resolve the generic YK\mathcal Y_K in characteristic zero, without changing its smooth open UU, and extend the projective resolution by graph closure over RR. This uses resolution over the field KK only, not mixed-characteristic resolution of a model. Call the resulting smooth projective generic fiber YY.

Take the product of the full source flag bundles of all QiQ_i over YY. Before extending the transported target flags, choose a smooth finite flat multisection T0→YT_0\to Y by Lemma 5.1. Take its flat model closure. Over U\mathcal U, the isomorphisms aia_i transport every source flag to a target flag. Close the graph of these target flags, and resolve the resulting generic fiber without changing its smooth inverse image of UU. Its final generic fiber TT is therefore still exactly T0T_0 over UU. In particular T→XT\to X is finite flat over all of UU; the successive modifications have introduced no new excluded image.

The maps aia_i preserve the two full flags. Their associated graded maps are model line maps Lα→MαL_\alpha\to M_\alpha, isomorphisms on the original model open. Termwise filtration of the two-term complexes gives ({PC2}). A nonzero map of line bundles on the integral model is injective and defines an effective Cartier divisor. Its generic divisor can only lie above ZZ. Empty generic divisors are allowed, for instance for a purely vertical line map. ◻

5.2. Localized characters and cycle coefficients

For degree-one variables l,ml,m put Br−1(l,m)=1r!∑a=0r−1mr−1−ala,(m−l)Br−1(l,m)=mr−lrr!.(PC3)B_{r-1}(l,m)=\frac1{r!}\sum_{a=0}^{r-1}m^{r-1-a}l^a, \qquad (m-l)B_{r-1}(l,m)=\frac{m^r-l^r}{r!}. \tag{PC3} For a nonzero line map s:L→Ms:L\to M on TT, with divisor DD, the all-degree Cartier localized character formula is ch⁡rD([L→M])∩[T]=Br−1(c1L,c1M)∩[D].(PC4)\operatorname{ch}^{D}_r([L\to M])\cap[T] =B_{r-1}(c_1L,c_1M)\cap[D]. \tag{PC4} Here LL has degree one and MM degree zero. To check the full polynomial, the complex resolves M∣DM|_D and the normal bundle of D↪TD\hookrightarrow T is (M⊗L−1)∣D(M\otimes L^{-1})|_D. Localized Riemann–Roch for this regular immersion gives em 1−e−(m−l)m−l∩[D]=em−elm−l∩[D].e^m\,\frac{1-e^{-(m-l)}}{m-l}\cap[D] =\frac{e^m-e^l}{m-l}\cap[D]. Its homogeneous degree r−1r-1 is exactly ({PC4}). This is Fulton, Intersection Theory, Corollary 18.1.2, specialized to a Cartier immersion; the proof uses deformation to the normal bundle. A leading-term statement alone, such as Stacks 0FEH, would not justify this all-degree polynomial.

Additivity on supported perfect K0K_0, compatibility with arbitrary base change, and the proper bivariant projection formula now give ∑αϵα(f∣Dα)∗(Br−1(c1Lα,c1Mα)∩[Dα])=d zCin CH⁡n−r(Z)Q.(PC5)\sum_\alpha\epsilon_\alpha (f|_{D_\alpha})_* \left(B_{r-1}(c_1L_\alpha,c_1M_\alpha)\cap[D_\alpha]\right) =d\,z_C \quad\hbox{in }\operatorname{CH}_{n-r}(Z)_{\mathbb Q}. \tag{PC5} The underlying equality is f∗(ch⁡f−1Z(f∗C)∩[T])=ch⁡Z(C)∩f∗[T]f_*(\operatorname{ch}^{f^{-1}Z}(f^*C)\cap[T]) =\operatorname{ch}^Z(C)\cap f_*[T], and f∗[T]=d[X]f_*[T]=d[X]. These are supported identities; one has not first forgotten the support and then attempted to recover coefficients. Nilpotent thickenings can be used to make every indicated factorization through ZZ scheme-theoretic. Chow groups are unchanged by those thickenings. Since the target Chow group is top-dimensional on ZZ, ({PC5}) is equality of actual cycle coefficients. See Stacks 0FB3 (base change), 0FBE (additivity), and 0FET (supported perfect K0K_0).

5.3. Chern degrees and actual top-degree tests

Lemma 5.3 (Chern degree against an arbitrary actual top test). Let h:B→Ah:B\to A be proper between integral projective KK-varieties, dim⁡A=m\dim A=m, dim⁡B=m+t\dim B=m+t. Let PP be a homogeneous degree-tt polynomial in first Chern forms of algebraic line bundles on BB with PWS metrics. For every compact actual top form η\eta on AA, ∫BP∧h∗η=∫h∗(Palg∩[B])η.(PC6)\int_BP\wedge h^*\eta =\int_{h_*(P_{\rm alg}\cap[B])}\eta. \tag{PC6} The statement is linear in cycles and includes their generic lengths.

Proof. Write each algebraic line bundle as a quotient of two very ample ones. Choose PWS metrics on those factors so that their quotient is the given metric. Expand PP by multilinearity. For each resulting monomial, choose a section of each very ample factor successively, meeting the finitely many current integral cycle components properly. The infinite field KK supplies such choices. Work componentwise when an intermediate divisor is reducible or nonreduced.

At every step apply scalar strong Poincare–Lelong, Mihatsch Corollary 5.6, with the remaining closed curvature factors and the pullback of η\eta as test. The latter is separately closed because η\eta is top-degree on AA. The Green correction is therefore zero. One is left with the pullback of η\eta integrated on a cycle of dimension mm.

A component mapping to a lower-dimensional subset of AA contributes zero, because the support of the actual top test avoids that subset. For a dominating component, remove a proper closed subset of AA so that the component maps finite flat of its native generic degree. The top test again avoids the removed subset. The native finite degree formula evaluates the integral with exactly that degree. This is precisely algebraic proper pushforward of the terminal intersection cycle, proving ({PC6}). ◻

This scalar equality applies to each Dα→ZjD_\alpha\to Z_j in ({PC5}), with m=n−rm=n-r and t=r−1t=r-1. Components whose image has smaller dimension contribute zero by the same support argument.

5.4. The exact-open Green construction

Theorem 5.4 (A strong Green pair for a supported model complex). Under the assumptions of ({PC1}), there is an actual form GC∈Br−1,r−1(Uan)G_C\in\mathcal B^{r-1,r-1}(U^{\rm an}) such that, for every compact actual η∈Bcn−r,n−r(Xan)\eta\in\mathcal B_c^{n-r,n-r}(X^{\rm an}), ∫UGC∧Dη+∫zCη=∫XΩC∧η.(PC7)\int_UG_C\wedge D\eta+\int_{z_C}\eta =\int_X\Omega_C\wedge\eta. \tag{PC7} The domain is the prescribed U=X∖ZU=X\setminus Z, with no auxiliary closed subset removed. If a tube is disjoint from the full model support Z\mathcal Z, the construction can be chosen zero on that tube.

Proof. Use Lemma 5.2 and give the line bundles their model metrics. Write lα=c1(L‾α),mα=c1(M‾α),gα=−log⁡∥sα∥.l_\alpha=c_1(\overline L_\alpha),\qquad m_\alpha=c_1(\overline M_\alpha),\qquad g_\alpha=-\log\|s_\alpha\|. On f−1Uf^{-1}U each gαg_\alpha is a PL scalar. Put GT=∑αϵαgαBr−1(lα,mα),GC=1dTr⁡f∣f−1UM(GT).(PC8)G_T=\sum_\alpha\epsilon_\alpha g_\alpha B_{r-1}(l_\alpha,m_\alpha), \qquad G_C=\frac1d\operatorname{Tr}^{\mathscr M}_{f|_{f^{-1}U}}(G_T). \tag{PC8} Every summand lies in the scalar-curvature module of Theorem 4.8. Source and target of the finite restriction are smooth, hence normal; all its assumptions hold. The resulting form is actual on the full ordinary open UU.

Strict model Chern additivity and functoriality from Theorem 3.4 convert ({PC2}), after forgetting support, into the actual identity on TT ∑αϵα(mα−lα)Br−1(lα,mα)=f∗ΩC.(PC9)\sum_\alpha\epsilon_\alpha (m_\alpha-l_\alpha)B_{r-1}(l_\alpha,m_\alpha) =f^*\Omega_C. \tag{PC9}

Apply scalar strong Poincare–Lelong to each sαs_\alpha with test Br−1(lα,mα)f∗ηB_{r-1}(l_\alpha,m_\alpha)f^*\eta. These tests are compact because ff is proper. Curvature factors are separately closed. Summation and ({PC9}) yield ∫T∖f−1ZGTf∗Dη+∑αϵα∫[Dα]Br−1(lα,mα)f∗η=∫Tf∗(ΩCη).(PC10)\int_{T\setminus f^{-1}Z}G_Tf^*D\eta +\sum_\alpha\epsilon_\alpha \int_{[D_\alpha]}B_{r-1}(l_\alpha,m_\alpha)f^*\eta =\int_Tf^*(\Omega_C\eta). \tag{PC10} For the first term, extending each off-divisor integral to the displayed common open changes nothing: DηD\eta has bidegree (n−r+1,n−r+1)(n-r+1,n-r+1), so its closed support avoids ZZ by Mihatsch Corollary 4.9. Properness puts its pulled-back support inside the finite locus f−1Uf^{-1}U. The minimal trace pairing identifies this first term with d∫UGCDηd\int_UG_CD\eta.

The form ΩCη\Omega_C\eta is top-degree on XX, so its support also avoids ZZ. On that support ff is finite flat of degree dd; the last term is d∫XΩCηd\int_X\Omega_C\eta by the native degree formula. The middle term equals d∫zCηd\int_{z_C}\eta by Lemma 5.3 and ({PC5}). Divide by dd.

All tests defining the off-ZZ integration functional of GCG_C have bidegree (n−r+1,n−r+1)(n-r+1,n-r+1) and therefore support away from ZZ. Thus its canonical extension and the all-actual meaning of ({PC7}) require no integrability assumption near ZZ beyond those already encoded by support avoidance.

Finally, above the model open U\mathcal U each sαs_\alpha is a model isomorphism of line bundles, so its model norm is one and gα=0g_\alpha=0. The local trace then vanishes over every tube avoiding Z\mathcal Z. This concerns the full model support, including possible vertical cohomology, not just the closure of its generic support. ◻

5.4.0.1. Support of the prescribed representative.

The source of a supported complex can have more components than its associated cycle after coefficient cancellation. Choosing one complex for each nonzero integral component of a prescribed cycle gives the form on its exact complement, as follows.

5.5. Arbitrary prescribed cycles and the good-reduction sector

Corollary 5.5 (Every prescribed cycle has a strong model-curvature pair). Let X/KX/K be smooth geometrically integral projective and let Z=∑jaj[Zj]Z=\sum_j a_j[Z_j] be a rational codimension-rr cycle, written with distinct integral components and nonzero coefficients. After choosing projective model data, there are an actual model Chern form ΩZ\Omega_Z and an actual GZ∈Br−1,r−1(Xan∖∣Z∣an)G_Z\in\mathcal B^{r-1,r-1} (X^{\rm an}\setminus|Z|^{\rm an}) satisfying D[GZ]δ+δZ=[ΩZ]δD[G_Z]_\delta+\delta_Z=[\Omega_Z]_\delta on all compact actual tests. No generic cycle representative is moved.

Proof. On the regular projective generic fiber, OZj\mathcal O_{Z_j} has a finite globally locally free resolution. Its degree-rr supported character is [Zj][Z_j], including the leading coefficient; this is Stacks 0FEW, or the leading-term part of localized Riemann–Roch. It is not necessary for ZjZ_j to be a regular immersion.

These finitely many generic bundle complexes extend to a common projective model after a vertical modification. For completeness, extend each bundle coherently, twist to obtain finitely many generators, and close the graph of its generic quotient in a Grassmann bundle. Its tautological quotient is a model vector bundle with the original generic fiber. Use the product of the finitely many Grassmann bundles. Every generic differential extends after multiplying by a common power of a uniformizer; the resulting successive products are zero because they vanish on the dense generic fiber. The scaled generic complex is chain-isomorphic to the original one by degreewise scalar rescaling. Its full model cohomology support has generic support exactly ZjZ_j.

Apply Theorem 5.4 to each complex and sum with coefficients aja_j. The union of the removed sets is exactly ∣Z∣|Z|, because the prescribed cycle was first written with distinct nonzero components. The zero cycle is handled by the zero form. ◻

Corollary 5.6 (Strong G for arbitrary cycles on a smooth good model). Suppose XX admits a smooth projective RR-model X\mathcal X. For every rational codimension-rr cycle ZZ on XX whose geometric ℓ\ell-adic class is zero, ℓ≠char⁡(k)\ell\ne\operatorname{char}(k), the form in Corollary 5.5 can be chosen with ΩZ=0\Omega_Z=0: ∫X∖∣Z∣GZDη=−∫Zη(η∈Bcn−r,n−r(Xan)).(PC11)\int_{X\setminus|Z|}G_ZD\eta=-\int_Z\eta \qquad \bigl(\eta\in\mathcal B_c^{n-r,n-r}(X^{\rm an})\bigr). \tag{PC11} This includes arbitrary singular or non-complete-intersection cycle components and retains the exact support and original field.

Proof. For each prescribed component take its scheme-theoretic flat closure Zj⊂X\mathcal Z_j\subset\mathcal X. The smooth model over a DVR is regular of finite dimension and has an ample line bundle. Hence OZj\mathcal O_{\mathcal Z_j} has a finite globally locally free resolution on this fixed model. One explicit justification is to surject finite sums of negative ample twists onto a coherent sheaf and its successive coherent kernels. After at most dim⁡X\dim\mathcal X steps the remaining kernel is locally free, since every regular local ring has global dimension its Krull dimension. Thus the global assertion is not inferred from perfectness without using the resolution property. No modification of the good model is needed. Use those model complexes in the preceding construction.

Their alternating degree-rr Chern character on the generic fiber is [Z][Z] in CH⁡r(X)Q\operatorname{CH}^r(X)_{\mathbb Q}. Smooth proper base change and compatibility of etale Chern classes identify its geometric special-fiber Chern class with the zero generic class. Its degree against every integral rr-cycle on the smooth special fiber is therefore zero by the etale trace pairing.

Theorem 3.2 now gives ΩZ=0\Omega_Z=0 as an actual form, not just as a current or a cohomology class. Substitute into the strong equation already proved against every actual test. ◻

Corollary 5.7 (Finite good covers and potential good reduction). Let X/KX/K be smooth integral projective. Suppose there are a finite valued extension K′/KK'/K, a smooth integral projective K′K'-variety YY with a smooth projective R′=K′∘R'=K'^\circ-model, and a finite dominant KK-morphism h:Y→Xh:Y\to X. Then every rational geometrically homologically trivial codimension-rr cycle ZZ on XX admits a zero-curvature strong Green form on exactly Xan∖∣Z∣anX^{\rm an}\setminus|Z|^{\rm an}, over the original field KK, against every compact KK-actual complementary test.

In particular this holds when XX has potential good reduction: take Y=XK′Y=X_{K'} and its finite projection to XX, working componentwise if the base change is disconnected. It also holds for a finite dominant smooth projective good-reduction cover over KK itself.

Proof. Regard YY as a KK-variety. It is smooth and integral, though it need not be geometrically integral over KK. This is enough for the preceding supported-complex theorem and the minimal trace: the multisection proof permits choosing a smooth connected component finite flat over an integral base. Since YY is Cohen–Macaulay and XX is regular of the same dimension, the finite dominant hh is flat by miracle flatness. Write a=deg⁡Kha=\deg_K h and set Z′=h∗ZZ'=h^*Z, with the scheme-theoretic generic lengths in flat pullback. Its support is exactly h−1∣Z∣h^{-1}|Z|, its geometric cohomology class is zero, and h∗Z′=aZ.(PC12)h_*Z'=aZ. \tag{PC12}

Resolve the closures of the distinct components of Z′Z' on the fixed smooth R′R'-model Y\mathcal Y, with their rational coefficients. View Y\mathcal Y also as a projective flat RR-model of the KK-variety YY. If ee is the ramification index, its special fiber over RR is the eeth nilpotent thickening of its good special fiber over R′R': locally a uniformizer of RR is a unit times the eeth power of one of R′R'.

Every integral rr-cycle on this thickened special fiber lies on its reduction. The restriction of the alternating operational degree-rr Chern class there is geometrically homologically zero by smooth proper base change over R′R'. Thus it has degree zero against every such cycle. Computing degrees over the smaller residue field only multiplies them by [k′:k][k':k], so they remain zero. Apply Theorem 3.2 directly to the RR-model Y\mathcal Y. It gives ΩZ′=0as a K-actual form on Yan.\Omega_{Z'}=0 \quad\hbox{as a $K$-actual form on $Y^{\rm an}$}. Native special-fiber multiplicities, including ee, are retained; none has been replaced by one. This step does not infer the KK-actual statement from equality on a smaller collection of K′K'-tests.

Perform the relative line construction over this RR-model simultaneously for the finitely many chosen resolutions. Close the joint kernel-Grassmann graph over their common exact open, then take the product of all source flag bundles before choosing one smooth finite multisection. Transport and close all target flags on that same multisection. This gives one common map carrying the rationally weighted sum of all the line complexes; it does not use a fiber product of separately chosen branched finite multisections. Keep its line sum GTG_T before applying any trace. Let b:T→Yb:T\to Y be the resulting map of generic degree dd, finite flat over Y∖h−1∣Z∣Y\setminus h^{-1}|Z|. The composite f=h∘b:T→Xf=h\circ b:T\to X is finite flat of degree adad above the whole U=X∖∣Z∣U=X\setminus|Z|. Define directly over KK GZ=1adTr⁡f∣f−1UM(GT∣f−1U).(PC13)G_Z=\frac1{ad} \operatorname{Tr}^{\mathscr M}_{f|_{f^{-1}U}} (G_T|_{f^{-1}U}). \tag{PC13} This is a permitted trace: GTG_T is the original finite sum of PL scalars times model curvature polynomials. We have not assumed that an already constructed GZ′G_{Z'} lies in the trace module.

Repeat the scalar Poincare–Lelong proof of Theorem 5.4 with an arbitrary compact actual test η\eta on XX. The source curvature is b∗ΩZ′=0b^*\Omega_{Z'}=0. Its residue pushed to YY is dZ′dZ', hence pushed to XX is daZdaZ by ({PC12}). The support of DηD\eta avoids ∣Z∣|Z|, so the first term is evaluated entirely on the finite open for the composite ff. Its minimal-trace pairing is ad∫UGZDηad\int_U G_ZD\eta. Division by adad proves ∫UGZDη=−∫Zη\int_U G_ZD\eta=-\int_Z\eta on all the stated tests and on the exact final complement.

For potential good reduction with geometrically integral XX, the native degree of the projection XK′→XX_{K'}\to X is [K′:K][K':K]. If that base change is disconnected, a smooth integral connected component and its good-model component may instead be used, with its actual native degree over XX. The possible ramification factor is already part of the degree and of the model calibrations; no additional normalization is inserted. ◻

5.6. Global primitives for generic Chow-zero model characters

Corollary 5.8 (Generically acyclic complexes). If a bounded model complex is exact on all of XX, then ∑i(−1)ich⁡r(C‾i)\sum_i(-1)^i\operatorname{ch}_r(\overline C_i) has a global actual DD-primitive on XanX^{\rm an}.

Proof. The same relative construction is made with generic open U=XU=X; its final generic multisection is finite flat over all of XX and all generic line divisors are empty. Formula ({PC8}) is global, and ({PC7}) has zero residue. Actual current injectivity identifies its DD with the displayed actual curvature. ◻

Proposition 5.9 (A generic K0K_0 relation has an actual primitive). Let ξ∈K0(X)Q\xi\in K_0(\mathcal X)_{\mathbb Q}. If ξ∣X=0\xi|_X=0 in K0(X)QK_0(X)_{\mathbb Q}, then every positive-degree actual ch⁡r(ξ)\operatorname{ch}_r(\xi) has a global actual DD-primitive.

Proof. A relation in K0(X)K_0(X) is a finite combination of short exact sequence relations and differences of isomorphic vector bundles. Extend the finitely many bundles to a common vertical model refinement, which does not change any original model metric on the generic fiber. For a generic exact sequence 0→A→B→C→00\to A\to B\to C\to0, close the graph of the subbundle A⊂BA\subset B in a Grassmann bundle of the already chosen model extension of BB. This produces a model exact sequence 0→S→B→Q→00\to S\to B\to Q\to0 with S∣X≃AS|_X\simeq A and Q∣X≃CQ|_X\simeq C. Strict model Whitney gives ch⁡r(B−A−C)=ch⁡r(S−A)+ch⁡r(Q−C).\operatorname{ch}_r(B-A-C) =\operatorname{ch}_r(S-A)+\operatorname{ch}_r(Q-C). After clearing uniformizer powers, the two generic isomorphisms are model maps giving two-term generically acyclic complexes. Their global primitives exist by Corollary 5.8. The same handles the isomorphic-bundle differences in the chosen finite presentation of the K0K_0 relation. Add the primitives and divide by the cleared integer denominator. ◻

Proposition 5.10 (Only the generic degree-rr Chow class matters). If ξ∈K0(X)Q\xi\in K_0(\mathcal X)_{\mathbb Q} satisfies ch⁡r(ξ∣X)=0\operatorname{ch}_r(\xi|_X)=0 in CH⁡r(X)Q\operatorname{CH}^r(X)_{\mathbb Q}, then its actual model ch⁡r(ξ)\operatorname{ch}_r(\xi) has a global actual DD-primitive.

Proof. Let ψ2\psi^2 be the Adams operation on model vector-bundle K0K_0 and set er(T)=∏0≤j≤nj≠rT−2j2r−2j,ξr=er(ψ2)ξ.e_r(T)=\prod_{\substack{0\leq j\leq n\\j\ne r}} \frac{T-2^j}{2^r-2^j}, \qquad \xi_r=e_r(\psi^2)\xi. On the smooth projective generic fiber the rational Chern character K0(X)Q≃⨁j=0nCH⁡j(X)QK_0(X)_{\mathbb Q}\simeq\bigoplus_{j=0}^n \operatorname{CH}^j(X)_{\mathbb Q} is an isomorphism, and ch⁡j(ψ2ζ)=2jch⁡j(ζ)\operatorname{ch}_j(\psi^2\zeta)=2^j\operatorname{ch}_j(\zeta). All generic Chern components of ξr\xi_r are zero, so ξr∣X=0\xi_r|_X=0.

The same Adams identity is strict for actual model Chern forms: it is a universal symmetric-root identity, hence follows from Theorem 3.4. Therefore ch⁡r(ξr)=ch⁡r(ξ)\operatorname{ch}_r(\xi_r)=\operatorname{ch}_r(\xi) as actual forms. Apply Proposition 5.9 to ξr\xi_r. ◻

Corollary 5.11 (Chow-zero cycles without a reduction hypothesis). For every prescribed rational codimension-rr cycle ZZ with [Z]=0[Z]=0 in CH⁡r(X)Q\operatorname{CH}^r(X)_{\mathbb Q}, strong G holds on exactly Xan∖∣Z∣anX^{\rm an}\setminus|Z|^{\rm an}, over KK, without a good-reduction assumption.

Proof. Construct the exact-support pair (GZ,ΩZ)(G_Z,\Omega_Z) above. Its model curvature is ch⁡r(ξ)\operatorname{ch}_r(\xi) for a model virtual class with generic degree-rr character [Z]=0[Z]=0. Choose a global actual Ψ\Psi with DΨ=ΩZD\Psi=\Omega_Z by the preceding proposition. Then GZ−Ψ∣X∖∣Z∣G_Z-\Psi|_{X\setminus|Z|} has zero curvature and the same source cycle, with no new poles. ◻

6. Algebraic equivalence and compact parameters

Throughout this section K/QpK/\mathbf Q_p is finite and X/KX/K is smooth, projective and geometrically integral. Write D=d′d′′D=d'd'' and let B\mathcal B be the actual delta-form sheaf.

Recall the algebraic quotient HBC,actr,r(X)={ω∈Br,r(Xan):d′ω=d′′ω=0}DBr−1,r−1(Xan).H_{\mathrm{BC,act}}^{r,r}(X)= \frac{\{\omega\in\mathcal B^{r,r}(X^{\rm an}): d'\omega=d''\omega=0\}} {D\mathcal B^{r-1,r-1}(X^{\rm an})}. The preceding theorems define the model-independent class oXr:CH⁡r(X)Q⟶HBC,actr,r(X),oXr(a)=[ch⁡r(ξˉ)],ch⁡r(ξ∣X)=a.(R12.P1)\mathfrak o_X^r:\operatorname{CH}^r(X)_{\mathbb Q} \longrightarrow H_{\mathrm{BC,act}}^{r,r}(X), \qquad \mathfrak o_X^r(a)=[\operatorname{ch}_r(\bar\xi)], \quad \operatorname{ch}_r(\xi|_X)=a. \tag{R12.P1}

6.1. Actual forms are locally constant in a rigid parameter

Lemma 6.1 (Uniform rigid-fiber constancy). Let S/KS/K be smooth and separated, let a0∈S(K)a_0\in S(K), and let ω∈B((X×S)an)\omega\in\mathcal B((X\times S)^{\rm an}). Denote by sa:Xan→(X×S)ans_a:X^{\rm an}\to(X\times S)^{\rm an} the section with parameter a∈S(K)a\in S(K). There is an ordinary analytic open neighborhood VV of a0a_0 in SanS^{\rm an} such that sa∗ω=sa0∗ωfor every a∈V∩S(K),(R12.P2)s_a^*\omega=s_{a_0}^*\omega \quad\text{for every }a\in V\cap S(K), \tag{R12.P2} as actual forms on the whole XanX^{\rm an}.

Proof. Put p:(X×S)an→Xanp:(X\times S)^{\rm an}\to X^{\rm an} and q:(X×S)an→Sanq:(X\times S)^{\rm an}\to S^{\rm an}. The analytic section s=sa0s=s_{a_0} is closed and ps=id⁡ps=\operatorname{id}. We first check the retracted germ argument directly.

Near a point of s(Xan)s(X^{\rm an}), write ω=tf⋆γ\omega=t_f^\star\gamma for a finite tuple fif_i of analytic units and a Euclidean delta-form γ\gamma. Shrink to where both fif_i and fi0=p∗s∗fif_i^0=p^*s^*f_i are defined. Each fi−fi0f_i-f_i^0 vanishes along the section. The inequalities ∣fi−fi0∣<∣fi0∣|f_i-f_i^0|<|f_i^0| therefore define an ordinary open neighborhood of that part of the section. On it ∣fi∣=∣fi0∣|f_i|=|f_i^0|. On every common refining skeleton the corresponding valuation tuples are equal as linear functions, so their intrinsic tropical pullbacks of γ\gamma agree. Mihatsch’s Definition 4.2 and Lemma 4.4 identify the actual forms, and Theorem 4.5 gives the compatibility with analytic pullback [1]. Taking the union of the resulting neighborhoods gives an ordinary open WW containing the entire section, on which ω=p∗s∗ω.\omega=p^*s^*\omega.

The projection qq is proper because XX is proper. In particular it is closed on the ordinary topology. Its closed set q((X×S)an∖W)q((X\times S)^{\rm an}\setminus W) does not contain a0a_0: the analytic fiber over a KK-rational point is exactly the image of the corresponding section. Its complement VV is an ordinary open with q−1(V)⊂Wq^{-1}(V)\subset W. Pull the displayed equality back by sas_a for a∈V∩S(K)a\in V\cap S(K) to obtain ({R12.P2}).

This argument uses properness of the analytic projection, not an identification of the underlying analytic product with a Cartesian product of point sets. All pullbacks are of actual forms by genuine analytic morphisms, including the closed section. ◻

Lemma 6.2 (A multiple enters a prescribed identity neighborhood). Let A/KA/K be an abelian variety. For every a∈A(K)a\in A(K) and every open neighborhood UU of 00 in A(K)A(K), there is an integer m>0m>0 such that ma∈Uma\in U.

Proof. The group A(K)A(K) is compact: a projective embedding realizes it as a closed subset of compact PN(K)\mathbf P^N(K). Choose a neighborhood VV of 00 with V−V⊂UV-V\subset U. Finitely many translates of VV cover A(K)A(K). Two of sufficiently many points 0,a,2a,…,Na0,a,2a,\ldots,Na belong to one translate; their difference is ma∈Uma\in U for some m>0m>0. This remains valid when aa is torsion. ◻

6.2. Additive parameter families and global primitives

Definition 32 (Additive cycle family). For an abelian variety A/KA/K, a class Θ∈CH⁡r(X×A)Q\Theta\in\operatorname{CH}^r(X\times A)_{\mathbb Q} is called additive on rational fibers if its refined Gysin fibers satisfy Θ0=0\Theta_0=0 and Θa+b=Θa+Θbin CH⁡r(X)Q(a,b∈A(K)).(R12.P3)\Theta_{a+b}=\Theta_a+\Theta_b \quad\text{in }\operatorname{CH}^r(X)_{\mathbb Q} \qquad(a,b\in A(K)). \tag{R12.P3} The requirement is about Chow classes. An arbitrary family on an abelian parameter variety is not presumed to satisfy it.

Theorem 6.4 (Compact additive parameters). Let Θ\Theta be as in ({R12.P3}). For every a∈A(K)a\in A(K), oXr(Θa)=0.(R12.P4)\mathfrak o_X^r(\Theta_a)=0. \tag{R12.P4} More precisely, any actual degree-rr model character representing Θa\Theta_a has a global actual DD-primitive that is a finite rational linear combination of the restricted finite-flat traces of global original PL-scalar/model-curvature line sums from Theorem 4.8.

Proof. Choose a virtual bundle on X×AX\times A whose degree-rr generic Chern character is Θ\Theta, and extend its finitely many bundles to a common projective flat model. This uses the rational Chern character isomorphism on the smooth projective generic fiber and the model-extension argument of Corollary 5.5. Let Ω\Omega be its actual degree-rr model character and write Ωb=sb∗Ω\Omega_b=s_b^*\Omega. Each Ωb\Omega_b is a model character on a graph model of XX. For any finite list of parameters, all of these model bundles can be compared on one projective flat refinement.

Lemma 6.1 gives a neighborhood UU of 00 in A(K)A(K) on which Ωb=Ω0\Omega_b=\Omega_0 literally. Fix aa, and choose m>0m>0 with ma∈Uma\in U by Lemma 6.2. On a common model the actual character Δ=Ωma−Ω0−m(Ωa−Ω0)=−m(Ωa−Ω0)(R12.P5)\Delta=\Omega_{ma}-\Omega_0-m(\Omega_a-\Omega_0) =-m(\Omega_a-\Omega_0) \tag{R12.P5} has generic degree-rr Chow class Θma−mΘa=0\Theta_{ma}-m\Theta_a=0. Proposition 5.10 supplies a global actual HH with DH=ΔDH=\Delta. Consequently βa=−1mH,Dβa=Ωa−Ω0.(R12.P6)\beta_a=-\frac{1}{m}H, \qquad D\beta_a=\Omega_a-\Omega_0. \tag{R12.P6}

If Ωa′\Omega_a' is any other actual degree-rr model character with generic class Θa\Theta_a, then Ωa′−(Ωa−Ω0)\Omega_a'-(\Omega_a-\Omega_0) has generic class zero. Give this difference a global primitive by the same proposition and add it to βa\beta_a. This proves ({R12.P4}) and independence of the chosen model metric at the level of actual primitives.

For the asserted form of the primitive, the generic-Chow-zero proof first applies an Adams projector, then expresses a generic K0K_0 relation by finitely many generically acyclic model complexes. Their prescribed generic open is the whole XX. Each primitive is therefore of the form ∑jqjdj−1Tr⁡bjM(Gj),qj∈Q,(R12.P7)\sum_j q_jd_j^{-1}\operatorname{Tr}^{\mathscr M}_{b_j}(G_j), \qquad q_j\in\mathbb Q, \tag{R12.P7} where bj:Tj→Xb_j:T_j\to X is finite flat on all of XX, djd_j is its native degree, and GjG_j is a global original line sum on TjT_j. Taking the finite rational linear combinations in ({R12.P6}) and in the final correction preserves this description. No trace of an arbitrary already constructed actual primitive is used. ◻

Remark 34 (A shorter quotient argument). By Lemma 6.1, the map a↦oXr(Θa)a\mapsto\mathfrak o_X^r(\Theta_a) is locally constant as a map from the compact space A(K)A(K) to an abstract set. It therefore has finite image. Additivity makes that image a finite subgroup of a real vector space, hence zero. This proves the vanishing without putting any topology on HBC,actr,r(X)H_{\mathrm{BC,act}}^{r,r}(X). The proof above additionally retains the composable representation ({R12.P7}).

6.3. The Poincare correspondence and descent over the original field

Lemma 6.6 (An additive family attached to a pointed curve). Let F/KF/K be a finite extension, let C/FC/F be a smooth projective geometrically integral curve with c0∈C(F)c_0\in C(F), and let Γ∈CH⁡r(XF×C)Q\Gamma\in\operatorname{CH}^r(X_F\times C)_{\mathbb Q}. Put J=Pic⁡C/F0J=\operatorname{Pic}^0_{C/F}. There is a class Θ∈CH⁡r(XF×J)Q\Theta\in\operatorname{CH}^r(X_F\times J)_{\mathbb Q} such that Θt=Γ∗c1(Pt),Θ0=0,Θt+t′=Θt+Θt′(t,t′∈J(F)),(R12.P8)\Theta_t=\Gamma_*c_1(\mathcal P_t),\qquad \Theta_0=0,\qquad \Theta_{t+t'}=\Theta_t+\Theta_{t'} \quad(t,t'\in J(F)), \tag{R12.P8} where P\mathcal P is the Poincare bundle normalized along c0×Jc_0\times J and C×{0}C\times\{0\}.

Proof. The pointed Picard functor has this normalized universal line bundle; see Milne’s Theorem 1.2 and Lemma 1.3 [14]. On XF×C×JX_F\times C\times J define Θ=(pXJ)∗(pXC∗Γ⋅pCJ∗c1(P)).(R12.P9)\Theta=(p_{XJ})_*\bigl(p_{XC}^*\Gamma\cdot p_{CJ}^*c_1(\mathcal P)\bigr). \tag{R12.P9} Here pXJp_{XJ} has relative dimension one, so the codimension is rr. Every pushforward in this formula is the usual algebraic proper pushforward on Chow groups. Refined Gysin compatibility with that pushforward and with intersection gives the fiber formula. The rigidified Picard law identifies Pt+t′\mathcal P_{t+t'} with Pt⊗Pt′\mathcal P_t\otimes\mathcal P_{t'}, and P0\mathcal P_0 is trivial. First Chern classes and correspondence action are additive, proving all of ({R12.P8}). ◻

Lemma 6.7 (Algebraic norm of an additive family). Let L/KL/K be finite Galois, let J/LJ/L be an abelian variety, and let Θ∈CH⁡r(XL×J)Q\Theta\in\operatorname{CH}^r(X_L\times J)_{\mathbb Q} be additive on J(L)J(L). Put A=Res⁡L/KJA=\operatorname{Res}_{L/K}J. There is a class ΘN∈CH⁡r(X×A)Q\Theta^N\in\operatorname{CH}^r(X\times A)_{\mathbb Q}, additive on A(K)A(K), whose fiber at the point aa corresponding to t∈J(L)=A(K)t\in J(L)=A(K) is (ΘN)a=h∗Θt,h:XL⟶X.(R12.P10)(\Theta^N)_a=h_*\Theta_t, \qquad h:X_L\longrightarrow X. \tag{R12.P10}

Proof. For completeness, AA can be constructed by descending the product ∏σ∈Gal⁡(L/K)σJ\prod_{\sigma\in\operatorname{Gal}(L/K)}{}^\sigma J with its canonical semilinear permutation datum. Finite Galois descent for projective varieties gives a smooth projective group, geometrically connected, hence an abelian variety over KK; see [14, Proposition 1.8]. Its KK-points are the conjugate tuples determined by a point of J(L)J(L). Let e:AL→Je:A_L\to J be the evaluation morphism, which is a group morphism. With hA:(X×A)L→X×Ah_A:(X\times A)_L\to X\times A, set ΘN=(hA)∗(id⁡XL×e)∗Θ.(R12.P11)\Theta^N=(h_A)_*(\operatorname{id}_{X_L}\times e)^*\Theta. \tag{R12.P11} The pullback here is an algebraic Chow pullback between smooth schemes, and hAh_A is finite flat. The proper-pushforward/refined- Gysin base-change formula gives ({R12.P10}). Since ee is additive and h∗h_* is a homomorphism, these fibers are additive; their zero fiber is zero. This proof only uses algebraic norms. It neither pulls actual forms back to LL nor descends them through an analytic finite trace. ◻

Theorem 6.8 (Geometrically algebraically trivial curvature). Let 1≤r≤n1\leq r\leq n and let z∈CH⁡r(X)Qz\in\operatorname{CH}^r(X)_{\mathbb Q} become algebraically trivial over K‾\overline K. Then oXr(z)=0.(R12.P12)\mathfrak o_X^r(z)=0. \tag{R12.P12} Every actual degree-rr model character representing zz has a global actual DD-primitive of the form ({R12.P7}). No reduction assumption is needed. For every prescribed rational cycle ZZ representing zz, strong G holds on exactly Xan∖∣Z∣anX^{\rm an}\setminus|Z|^{\rm an}, over the original KK, and on every compact actual test.

Proof. Over K‾\overline K, algebraically trivial Chow classes are generated by degree-zero divisor correspondences on smooth projective curves. This is also a consequence of the curve case of [15, Theorems 1–2, Proposition 3.14]. Write zK‾z_{\overline K} as a finite rational sum of such expressions. All curves, correspondences, marked points, divisor classes and rational equivalence witnesses involve finitely many algebraic coefficients. They therefore descend, after enlarging once, to a finite Galois extension L/KL/K. Enlarge LL also so that each curve has a chosen LL-point. Rational coefficients can either be retained throughout or cleared by one positive integer at this step.

Lemma 6.6, followed by taking a product of the finitely many Jacobians and adding the pullbacks of their families, gives an abelian variety J/LJ/L, a class Θ\Theta additive on J(L)J(L), and t∈J(L)t\in J(L) such that Θt=zL.\Theta_t=z_L. Form the native family ΘN\Theta^N from Lemma 6.7. Its fiber at the corresponding a∈A(K)a\in A(K) is (ΘN)a=h∗zL=[L:K]zin CH⁡r(X)Q,(R12.P13)(\Theta^N)_a=h_*z_L=[L:K]z \quad\text{in }\operatorname{CH}^r(X)_{\mathbb Q}, \tag{R12.P13} by the algebraic flat-pullback/proper-pushforward degree formula. Apply Theorem 6.4 entirely over KK and divide by d=[L:K]d=[L:K] to obtain ({R12.P12}).

Here is the primitive with every correction exposed. Choose a model character Ω\Omega for ΘN\Theta^N on X×AX\times A, and let βa\beta_a be ({R12.P6}). If Ωz\Omega_z is the prescribed model character for zz, then Ωa−Ω0−dΩz\Omega_a-\Omega_0-d\Omega_z has generic Chow character zero. Give it a global primitive HzH_z by Proposition 5.10. Then Ψz=βa−Hzd,DΨz=Ωz.(R12.P14)\Psi_z=\frac{\beta_a-H_z}{d}, \qquad D\Psi_z=\Omega_z. \tag{R12.P14} Every term retains ({R12.P7}).

Finally take the exact-support pair (GZ,ΩZ)(G_Z,\Omega_Z) supplied by the supported-complex construction, and apply the preceding result to its model curvature. Put γZ=GZ−ΨZ∣Xan∖∣Z∣an.\gamma_Z=G_Z-\Psi_Z\big|_{X^{\rm an}\setminus|Z|^{\rm an}}. The correction is global, so introduces no additional pole. Stokes and actual-current injectivity give, for every η∈Bcn−r,n−r(Xan)\eta\in\mathcal B_c^{n-r,n-r}(X^{\rm an}), ∫Xan∖∣Z∣anγZDη=−∫Zanη.\int_{X^{\rm an}\setminus|Z|^{\rm an}}\gamma_ZD\eta =-\int_{Z^{\rm an}}\eta. This is the required exact-support, all-actual-test statement. ◻

7. Finite numerical certificates and abelian varieties

In this section K/QpK/\mathbf Q_p is finite, with valuation ring RR and finite residue field kk. For a smooth integral projective KK-variety VV of dimension dd, write CH⁡r(V)Q,hom=ker⁡(CH⁡r(V)Q⟶H2r(VK‾,Qℓ(r))),ℓ≠p.\operatorname{CH}^r(V)_{\mathbf Q,\mathrm{hom}} =\ker\bigl(\operatorname{CH}^r(V)_{\mathbf Q}\longrightarrow H^{2r}(V_{\overline K},\mathbf Q_\ell(r))\bigr), \qquad \ell\ne p. When VV is not geometrically integral, the geometric class is understood on every component. The construction of the preceding sections applies to smooth integral VV with this convention: the finite multisection argument, model Chern forms, finite trace, and supported-complex Green construction use integrality over KK; geometric integrality enters the algebraic-equivalence theorem only.

7.1. A finite numerical certificate

Definition 7.1. Let V/R\mathcal V/R be integral, projective and flat, with smooth integral generic fiber VV. A virtual model bundle ξ∈K0(V)Q\xi\in K_0(\mathcal V)_{\mathbf Q} is an rr-numerical certificate for b∈CH⁡r(V)Qb\in\operatorname{CH}^r(V)_{\mathbf Q} if ch⁡r(ξ∣V)=banddeg⁡k(ch⁡r(ξs)∩[T])=0\operatorname{ch}_r(\xi|_V)=b \quad\text{and}\quad \deg_k\bigl(\operatorname{ch}_r(\xi_s)\cap[T]\bigr)=0 for every integral rr-dimensional closed subscheme T⊂VsT\subset\mathcal V_s. The degree is that of the resulting zero-cycle on the proper kk-scheme TT.

Lemma 7.2 (Model refinements preserve numerical certificates). If ρ:V′→V\rho:\mathcal V'\to\mathcal V is a projective morphism of integral projective flat RR-models that is an isomorphism on generic fibers, then ρ∗ξ\rho^*\xi is an rr-numerical certificate for the same generic class whenever ξ\xi is.

Proof. The equality on the generic fiber is immediate. For an integral vertical rr-cycle T′T' on V′\mathcal V', the operational projection formula gives deg⁡k(ch⁡r(ρ∗ξs)∩[T′])=deg⁡k(ch⁡r(ξs)∩ρs∗[T′]).\deg_k\bigl(\operatorname{ch}_r(\rho^*\xi_s)\cap[T']\bigr) =\deg_k\bigl(\operatorname{ch}_r(\xi_s)\cap\rho_{s*}[T']\bigr). The proper pushforward on the right is zero if the image has dimension smaller than rr; otherwise it is a positive integer multiple of an integral rr-cycle on Vs\mathcal V_s. The numerical certificate makes its degree zero. ◻

Theorem 7.3 (Finite numerical descent). Let h:V→Xh:V\to X be finite dominant between smooth integral projective KK-varieties, and let 1≤r≤dim⁡X1\le r\le\dim X. If h∗zh^*z has an rr-numerical certificate on a projective flat model of VV, then oX,Kr(z)=0(z∈CH⁡r(X)Q).\mathfrak o^r_{X,K}(z)=0 \qquad(z\in\operatorname{CH}^r(X)_{\mathbf Q}). Each prescribed rational cycle ZZ in class zz consequently has an ordinary actual zero-curvature Green form on Xan∖∣Z∣anX^{\mathrm{an}}\setminus|Z|^{\mathrm{an}} satisfying (1.1) for every compact actual complementary test.

The assertion also applies when VV is the underlying KK-variety of a smooth integral projective variety over a finite extension F/KF/K, and its model is projective flat over F∘F^\circ.

Proof. Put e=deg⁡K(h)e=\deg_K(h). A finite map between smooth integral varieties of the same dimension is flat, so hh is finite flat of rank ee. Choose a projective flat model X\mathcal X and a virtual bundle representing zz; let ΩX\Omega_X be its actual degree-rr character. The existence of this model character and its dependence only on the generic Chow class in actual Bott–Chern cohomology were proved in Proposition 5.10.

Let ξ\xi be the supplied certificate on V\mathcal V. Form the closure of the generic graph of hh in V×RX\mathcal V\times_R\mathcal X. It is a projective flat model V′\mathcal V' of VV, with morphisms to both original models. By Lemma 7.2, the pullback of ξ\xi remains numerically trivial on vertical rr-cycles. Replacing V\mathcal V by this model makes the two model characters comparable.

The numerical vanishing theorem Theorem 3.2 and Theorem 3.4 gives the strict equality of actual forms ch⁡r(ξˉ)=0on Van.\operatorname{ch}_r(\bar\xi)=0 \quad\text{on }V^{\mathrm{an}}. The virtual bundles h∗ξXh^*\xi_X and ξ\xi have the same generic degree-rr Chow character. Thus h∗ΩX−ch⁡r(ξˉ)=h∗ΩXh^*\Omega_X-\operatorname{ch}_r(\bar\xi)=h^*\Omega_X has a global actual DD-primitive by the generic-Chow-zero theorem Proposition 5.10. We use its constructive form: h∗ΩX=∑jcj Ωj,βV=∑jcjejTr⁡bjM(Gj),DGj=bj∗Ωj. h^*\Omega_X=\sum_jc_j\,\Omega_j, \qquad \beta_V=\sum_j\frac{c_j}{e_j} \operatorname{Tr}^{\mathscr M}_{b_j}(G_j), \qquad DG_j=b_j^*\Omega_j.(7.1) Here cj∈Qc_j\in\mathbf Q, each bj:Tj→Vb_j:T_j\to V is finite flat of native degree eje_j, and GjG_j is a global sum of the original PL scalar–curvature generators. In particular the composite hbjh b_j is finite flat on the whole generic fiber XX.

The finite trace of Theorem 4.8 therefore defines the following actual form on the original XanX^{\mathrm{an}}: βX=1e∑jcjejTr⁡hbjM(Gj). \beta_X=\frac1e\sum_j\frac{c_j}{e_j} \operatorname{Tr}^{\mathscr M}_{h b_j}(G_j).(7.2) For an arbitrary compact complementary actual test η\eta, the finite-trace pairing and actual Stokes give ∫XβX∧Dη=1e∑jcjej∫TjGj∧(hbj)∗Dη=1e∑jcj∫VΩj∧h∗η=1e∫Vh∗(ΩX∧η)=∫XΩX∧η.\begin{aligned} \int_X\beta_X\wedge D\eta &=\frac1e\sum_j\frac{c_j}{e_j} \int_{T_j}G_j\wedge(hb_j)^*D\eta\\ &=\frac1e\sum_jc_j\int_V\Omega_j\wedge h^*\eta\\ &=\frac1e\int_Vh^*(\Omega_X\wedge\eta) =\int_X\Omega_X\wedge\eta. \end{aligned} The last equality uses the full native degree ee, and the first three equalities hold for every compact actual test by the finite-trace theorem. Mihatsch’s injectivity of the map from actual forms to currents now yields DβX=ΩXD\beta_X=\Omega_X. The obstruction class vanishes; subtracting βX\beta_X from the exact-support curved pair of Corollary 5.5 yields the asserted Green form.

For a model over F∘F^\circ, regard it as a model over RR. The scheme-theoretic special fiber retains its ramification multiplicity. Each vertical numerical degree over kk is the corresponding degree over the residue field of FF multiplied by that residue-field degree. The same proof uses the complete native generic degree of hh, including the constant-field factor. ◻

7.2. The special fiber of a semistable abelian model

Let R′R' be a complete discrete valuation ring with finite residue field k′k' and fraction field FF. Let A/R′\mathcal A/R' be a regular projective strictly semistable model of an abelian variety A/FA/F of dimension dd, and let Y=⋃iDiY=\bigcup_iD_i be its reduced special fiber. With Y(1)=∐iDiY^{(1)}=\coprod_iD_i and Y(2)=∐i<j(Di∩Dj)Y^{(2)}=\coprod_{i<j}(D_i\cap D_j), define As(D)=im⁡(CH⁡s(D)Q→H2s(Dkˉ′,Qℓ(s)))A^s(D)=\operatorname{im}\bigl(\operatorname{CH}^s(D)_{\mathbf Q} \to H^{2s}(D_{\bar k'},\mathbf Q_\ell(s))\bigr) for each smooth projective stratum DD and set As(Y)=ker⁡(As(Y(1))→ρAs(Y(2))),As(Y)=coker⁡(Ad−s−1(Y(2))→γAd−s(Y(1))). A^s(Y)=\ker\bigl(A^s(Y^{(1)})\xrightarrow{\rho}A^s(Y^{(2)})\bigr), \qquad A_s(Y)=\operatorname{coker}\bigl( A^{d-s-1}(Y^{(2)})\xrightarrow{\gamma}A^{d-s}(Y^{(1)})\bigr).(7.3) The two arrows are the signed restrictions and Gysin maps. The natural map CH⁡s(Y)Q↠As(Y)\operatorname{CH}_s(Y)_{\mathbf Q}\twoheadrightarrow A_s(Y) is surjective.

Theorem 7.4 (Künnemann’s local semistable theorem). After a finite extension of any finite pp-adic field KK, every abelian variety over KK admits a regular projective strictly semistable model A/R′\mathcal A/R' for which, in all degrees, the cycle-class map As(Y)⟶H2seˊt(Ykˉ′,Qℓ(−s))A_s(Y)\longrightarrow H^{\mathrm{\acute et}}_{2s}(Y_{\bar k'},\mathbf Q_\ell(-s)) is injective, and the middle term of Ad+1−r(Y)→i∗i∗Ar(Y)→∩[Y]Ad−r(Y) A_{d+1-r}(Y)\xrightarrow{i^*i_*} A^r(Y) \xrightarrow{\cap[Y]} A_{d-r}(Y)(7.4) is exact, where i:Y↪Ai:Y\hookrightarrow\mathcal A.

Source of the published theorem. This is the complete-DVR construction and special-fiber theorem of [6, Theorems 5.6 and 6.6]. The base ring is complete and has finite residue field, as in Section 6.2 of the cited work. The groups in (7.3) are defined using the geometric cycle-class images on the individual strata. ◻

Lemma 7.5 (A pure, vertically numerical extension). Let A/FA/F have a model furnished by Theorem 7.4. For every geometrically homologically trivial class z∈CH⁡r(A)Qz\in\operatorname{CH}^r(A)_{\mathbf Q}, 1≤r≤d1\le r\le d, there is Θ0∈CH⁡r(A)Q\Theta_0\in\operatorname{CH}^r(\mathcal A)_{\mathbf Q} restricting to zz, such that its refined restriction to every DiD_i is homologically trivial. In particular zz has an rr-numerical certificate on A\mathcal A.

Proof. Choose a Chow extension Θ\Theta of zz by taking closures of cycle representatives. Its tuple aa of component restriction classes belongs to Ar(Y)A^r(Y). The class of a∩[Y]∈Ad−r(Y)a\cap[Y]\in A_{d-r}(Y) is the specialization of zz. Compatibility of proper specialization with geometric ℓ\ell-adic cycle classes makes its image in geometric homology zero. By the injectivity in Theorem 7.4, a∩[Y]=0a\cap[Y]=0.

Exactness of (7.4) gives b∈Ad+1−r(Y)b\in A_{d+1-r}(Y) with i∗i∗b=ai^*i_*b=a. Choose β∈CH⁡d+1−r(Y)Q\beta\in\operatorname{CH}_{d+1-r}(Y)_{\mathbf Q} mapping to bb and set Θ0=Θ−i∗β.\Theta_0=\Theta-i_*\beta. The generic fiber of i∗βi_*\beta is empty, so Θ0∣A=z\Theta_0|_A=z. The tuple of its component restrictions is a−i∗i∗b=0a-i^*i_*b=0 in Ar(Y)A^r(Y). Since Ar(Y)A^r(Y) is a subspace of ⨁iAr(Di)\bigoplus_i A^r(D_i), each component restriction has zero geometric ℓ\ell-adic class and is numerically trivial.

The regularity of A\mathcal A and the rational Chern-character isomorphism on regular schemes [7, Tag 0FEY] supply a unique ξ∈K0(A)Q\xi\in K_0(\mathcal A)_{\mathbf Q} satisfying ch⁡(ξ)∩[A]=Θ0. \operatorname{ch}(\xi)\cap[\mathcal A] =\Theta_0.(7.5) For completeness, applying the Adams operation 2−rψ22^{-r}\psi^2 to (7.5) yields the same codimension-rr class. Uniqueness therefore gives ψ2ξ=2rξ\psi^2\xi=2^r\xi. Because ch⁡j(ψ2ξ)=2jch⁡j(ξ)\operatorname{ch}_j(\psi^2\xi)=2^j\operatorname{ch}_j(\xi), all operational character components except j=rj=r vanish.

Every integral vertical rr-cycle T⊂AsT\subset\mathcal A_s is contained in a component DiD_i. The projection formula for the refined restriction gives deg⁡k′(ch⁡r(ξs)∩[T])=deg⁡k′((Θ0∣Di)∩[T])=0.\deg_{k'}\bigl(\operatorname{ch}_r(\xi_s)\cap[T]\bigr) =\deg_{k'}\bigl((\Theta_0|_{D_i})\cap[T]\bigr)=0. The last equality follows from numerical triviality of the component restriction. Thus ξ\xi is a numerical certificate. ◻

7.3. Zero-curvature Green forms on abelian targets

Theorem 7.6 (Full actual strong Green forms for abelian varieties). Let A/KA/K be an abelian variety of arbitrary dimension. For every 1≤r≤dim⁡A1\le r\le\dim A and every prescribed rational cycle ZZ whose Chow class is geometrically homologically trivial, there is an actual form γZ∈Br−1,r−1(Aan∖∣Z∣an)\gamma_Z\in\mathcal B^{r-1,r-1} (A^{\mathrm{an}}\setminus|Z|^{\mathrm{an}}) satisfying (1.1) for all compact actual tests over the original KK.

Proof. Choose a finite field extension F/KF/K for which AFA_F has the semistable model of Theorem 7.4. Its rr-numerical certificate for ZFZ_F is supplied by Lemma 7.5. The finite dominant map h:AF→Ah:A_F\to A has native degree [F:K][F:K] and is finite flat as a map of smooth integral KK-varieties. Apply Theorem 7.3 to hh. It supplies a global actual primitive for the obstruction of ZZ on the original AanA^{\mathrm{an}}, and hence the asserted Green form on the prescribed support complement. ◻

Corollary 7.7 (Finite abelian domination). Let X/KX/K be smooth projective geometrically integral, and assume there is a finite surjective morphism f:A→Xf:A\to X from an abelian variety A/KA/K. Every geometrically homologically trivial rational cycle on XX satisfies the strong Green identity in all codimensions, with prescribed exact support and all compact original-field actual tests.

Proof. For z∈CH⁡r(X)Q,homz\in\operatorname{CH}^r(X)_{\mathbf Q,\mathrm{hom}}, the class f∗zf^*z is geometrically homologically trivial by functoriality of the geometric cycle-class map. Extend KK to FF as in Theorem 7.6; Lemma 7.5 gives a numerical certificate for its pullback to AFA_F. The composite finite dominant map AF→A→XA_F\to A\to X has degree [F:K]deg⁡f[F:K]\deg f. Apply Theorem 7.3 directly to this composite. ◻

8. Numerical certificates under Chow correspondences

This section uses ordinary rational intersection theory on regular projective models. The only analytic input is the finite numerical descent theorem already proved in Theorem 7.3.

8.1. A supported numerical projection formula

Let V/R\mathcal V/R be a regular, integral, projective and flat model, and let Θ∈CH⁡p(V)Q\Theta\in\operatorname{CH}^p(\mathcal V)_{\mathbf Q} be homogeneous. By the rational Chern-character isomorphism for regular schemes [7, Tag 0FEY], there is a unique ξΘ∈K0(V)Q\xi_\Theta\in K_0(\mathcal V)_{\mathbf Q} such that ch⁡(ξΘ)∩[V]=Θ. \operatorname{ch}(\xi_\Theta)\cap[\mathcal V]=\Theta.(8.1) The Adams identity, applied to (8.1), gives ψ2ξΘ=2pξΘ\psi^2\xi_\Theta=2^p\xi_\Theta and hence ch⁡j(ξΘ)=0\operatorname{ch}_j(\xi_\Theta)=0 as an operational class for j≠pj\ne p. We write cΘ=ch⁡p(ξΘ)c_\Theta=\operatorname{ch}_p(\xi_\Theta). It is an operational class on every closed support and satisfies cΘ∩[V]=Θc_\Theta\cap[\mathcal V]=\Theta.

We call Θ\Theta vertically pp-numerically trivial if deg⁡k(cΘ∩[T])=0for every integral vertical p-cycleT⊂Vs. \deg_k(c_\Theta\cap[T])=0 \quad\text{for every integral vertical $p$-cycle} \quad T\subset\mathcal V_s.(8.2) Equivalently, ξΘ\xi_\Theta is a pp-numerical certificate for Θ\Theta on the generic fiber.

Lemma 8.1 (Numerical pullback). Let f:W→Vf:\mathcal W\to\mathcal V be a projective morphism between regular integral projective flat RR-models. If Θ\Theta is vertically pp-numerically trivial, then its refined pullback f!Θ=(f∗cΘ)∩[W]f^!\Theta=(f^*c_\Theta)\cap[\mathcal W] is vertically pp-numerically trivial.

Proof. The schemes are regular, so ff is a local complete intersection morphism, and the projective factorization supplies its refined Gysin operation. For any integral vertical pp-cycle T⊂WsT\subset\mathcal W_s, the proper projection formula gives deg⁡k((f∗cΘ)∩[T])=deg⁡k(cΘ∩f∗[T])=0.\deg_k((f^*c_\Theta)\cap[T]) =\deg_k(c_\Theta\cap f_*[T])=0. Indeed f∗[T]f_*[T] is a rational linear combination of vertical pp-cycles, or zero. The displayed refined pullback has pure operational representative f∗cΘf^*c_\Theta by (8.1) and Gysin compatibility. ◻

Lemma 8.2 (Supported numerical projection). Let f:W→Vf:\mathcal W\to\mathcal V be a projective morphism between regular integral projective flat RR-models of total dimensions MM and NN. Let Θ∈CH⁡p(W)Q\Theta\in\operatorname{CH}^p(\mathcal W)_{\mathbf Q} be vertically pp-numerically trivial. Put δ=M−N\delta=M-N, q=p−δq=p-\delta, and assume 0≤q≤N−10\le q\le N-1. Then f∗Θ∈CH⁡q(V)Qf_*\Theta\in\operatorname{CH}^q(\mathcal V)_{\mathbf Q} is vertically qq-numerically trivial. More precisely, for every integral vertical qq-cycle T⊂VsT\subset\mathcal V_s one has the supported projection identity (f∗Θ)⋅[T]=f∗(Θ⋅f![T])in CH⁡0(V)Q. (f_*\Theta)\cdot[T] = f_*\bigl(\Theta\cdot f^![T]\bigr) \qquad\text{in }\operatorname{CH}_0(\mathcal V)_{\mathbf Q}.(8.3)

Proof. Since V\mathcal V and W\mathcal W are regular, the projective morphism ff is a local complete intersection morphism: factor it through a projective bundle over V\mathcal V, then use the regular embedding of the regular source into that regular bundle. The fundamental class [T][T] has codimension N−qN-q in the regular scheme V\mathcal V. Its refined pullback f![T]∈CH⁡M−(N−q)(W)Q=CH⁡p(W)Qf^![T]\in\operatorname{CH}_{M-(N-q)}(\mathcal W)_{\mathbf Q} =\operatorname{CH}_p(\mathcal W)_{\mathbf Q} is supported on f−1(T)f^{-1}(T), hence is a rational combination of vertical pp-cycles. The bivariant projection formula for the lci pullback on regular schemes [5, Chapter 17] and [9, Tags 0FF6 and 0FFA] gives (8.3). The degree of Θ⋅f![T]\Theta\cdot f^![T] is zero by the numerical hypothesis on Θ\Theta. Since cf∗Θ∩[T]c_{f_*\Theta}\cap[T] is the intersection product (f∗Θ)⋅[T](f_*\Theta)\cdot[T], the degree required for vertical numerical triviality vanishes. ◻

Lemma 8.3 (Products of numerical and operational classes). Let W\mathcal W be a regular projective flat model. Suppose α∈CH⁡p(W)Q\alpha\in\operatorname{CH}^p(\mathcal W)_{\mathbf Q} is vertically pp-numerically trivial and β∈CH⁡t(W)Q\beta\in\operatorname{CH}^t(\mathcal W)_{\mathbf Q}. If p+tp+t lies between zero and the relative dimension of W\mathcal W, then α⋅β\alpha\cdot\beta is vertically (p+t)(p+t)-numerically trivial.

Proof. For an integral vertical (p+t)(p+t)-cycle TT, the operational intersection cβ∩[T]c_\beta\cap[T] is a rational combination of vertical pp-cycles. The degrees of cα∩(cβ∩[T])c_\alpha\cap(c_\beta\cap[T]) vanish termwise. Purity of the rational operational representatives identifies cαβ=cαcβc_{\alpha\beta}=c_\alpha c_\beta. ◻

8.2. Transfer along algebraic correspondences

Let VV and XX be smooth projective geometrically integral FF-varieties of dimensions mm and nn. A correspondence Γ∈CH⁡m+a(V×FX)Q\Gamma\in\operatorname{CH}^{m+a}(V\times_F X)_{\mathbf Q} acts by Γ∗b=(pr⁡X)∗(pr⁡V∗b⋅Γ)∈CH⁡r+a(X)Q,b∈CH⁡r(V)Q.\Gamma_*b=(\operatorname{pr}_X)_* (\operatorname{pr}_V^*b\cdot\Gamma) \in\operatorname{CH}^{r+a}(X)_{\mathbf Q}, \qquad b\in\operatorname{CH}^r(V)_{\mathbf Q}.

Proposition 8.4 (Algebraic correspondence transfer). Assume that VV and XX have regular integral projective flat RFR_F-models V\mathcal V and X\mathcal X. Suppose that b∈CH⁡r(V)Qb\in\operatorname{CH}^r(V)_{\mathbf Q} has an rr-numerical certificate on V\mathcal V. For every correspondence Γ∈CH⁡m+a(V×X)Q\Gamma\in\operatorname{CH}^{m+a}(V\times X)_{\mathbf Q} with 1≤r+a≤n1\le r+a\le n, its image Γ∗b\Gamma_*b has an (r+a)(r+a)-numerical certificate on X\mathcal X.

Proof. Write R′=RFR'=R_F. The projective flat product V×R′X\mathcal V\times_{R'}\mathcal X is integral: its generic fiber V×FXV\times_F X is geometrically integral, and flatness excludes any additional irreducible component supported on the special fiber. By the projective semistable alteration theorem of de Jong [10, Theorem 6.5], there is a projective generically finite morphism g:W⟶V×R′Xg:\mathcal W\longrightarrow \mathcal V\times_{R'}\mathcal X from a regular integral projective flat model (possibly over a finite extension of R′R'). The native degree of its generic map gF:W→V×FXg_F:W\to V\times_F X is a positive integer dd. Indeed, de Jong first constructs a semistable open source inside a projective compactification. Since the target here is proper, the alteration source is proper over the base, so this open immersion is closed as well. The integral compactification therefore equals the source. Let fVf_V and fXf_X be the composite model morphisms to the respective factors.

Choose a pure vertically numerical class Θ∈CH⁡r(V)Q\Theta\in\operatorname{CH}^r(\mathcal V)_{\mathbf Q} with Θ∣V=b\Theta|_V=b. The generic class gF∗Γg_F^*\Gamma has a Chow extension Γ~∈CH⁡m+a(W)Q\widetilde\Gamma\in\operatorname{CH}^{m+a}(\mathcal W)_{\mathbf Q}; the closure of cycle representatives gives such an extension. Set Ξ=(fV!Θ)⋅Γ~∈CH⁡r+m+a(W)Q.\Xi=(f_V^!\Theta)\cdot\widetilde\Gamma \in\operatorname{CH}^{r+m+a}(\mathcal W)_{\mathbf Q}. By Lemmas 8.1 and 8.3, Ξ\Xi is vertically (r+m+a)(r+m+a)-numerically trivial. Since dim⁡W−dim⁡X=m\dim\mathcal W-\dim\mathcal X=m, the supported numerical projection lemma applies to fXf_X, and gives the vertically (r+a)(r+a)-numerically trivial class Λ=1d(fX)∗Ξ∈CH⁡r+a(X)Q.\Lambda=\frac1d(f_X)_*\Xi \in\operatorname{CH}^{r+a}(\mathcal X)_{\mathbf Q}.

On smooth generic fibers, pullback is an ordinary lci Chow pullback, and the projection formula for the generically finite gFg_F gives Λ∣X=1d(pr⁡X)∗(gF)∗gF∗(pr⁡V∗b⋅Γ)=(pr⁡X)∗(pr⁡V∗b⋅Γ)=Γ∗b.\begin{aligned} \Lambda|_X &=\frac1d(\operatorname{pr}_X)_* (g_F)_*g_F^*(\operatorname{pr}_V^*b\cdot\Gamma)\\ &=(\operatorname{pr}_X)_* (\operatorname{pr}_V^*b\cdot\Gamma) =\Gamma_*b. \end{aligned} The pure virtual model bundle ξΛ\xi_\Lambda defined by (8.1) is the required numerical certificate. ◻

Theorem 8.5 (A Chow-correspondence criterion for actual Green forms). Let X/KX/K be smooth projective geometrically integral. Suppose there exists a finite extension F/KF/K for which the following properties hold:

  1. XFX_F has a regular integral projective flat RFR_F-model;

  2. there is an abelian variety B/FB/F with a regular projective semistable model satisfying Theorem 7.4;

  3. there are finitely many correspondences uj:XF⇝Bu_j:X_F\rightsquigarrow B and vj:B⇝XFv_j:B\rightsquigarrow X_F of opposite degree shifts with ∑jvj∘uj=ΔXFin CH⁡dim⁡X(XF×XF)Q. \sum_jv_j\circ u_j=\Delta_{X_F} \qquad\text{in }\operatorname{CH}^{\dim X}(X_F\times X_F)_{\mathbf Q}.(8.4)

Then every prescribed rational cycle ZZ on XX with geometrically homologically trivial class admits an actual zero-curvature Green form on the exact complement Xan∖∣Z∣anX^{\mathrm{an}}\setminus |Z|^{\mathrm{an}} satisfying (1.1) against every compact actual test over KK.

Proof. Write z=[Z]z=[Z] and zF=[ZF]z_F=[Z_F]. Equation (8.4) gives zF=∑j(vj)∗(uj)∗zF.z_F=\sum_j(v_j)_*(u_j)_*z_F. Each (uj)∗zF(u_j)_*z_F is geometrically homologically trivial, by compatibility of algebraic correspondences with geometric ℓ\ell-adic cycle classes. Terms of codimension outside 1,…,dim⁡B1,\ldots,\dim B vanish (in codimension zero this follows from homological triviality); for the remaining terms Lemma 7.5 supplies a numerical certificate on the regular semistable model of BB.

Apply Proposition 8.4 to vjv_j, taking the given regular model of XFX_F as the target. Its resulting numerical certificates add to a certificate for zFz_F. Finally, XF→XX_F\to X is finite flat of native degree [F:K][F:K]. The finite numerical descent theorem Theorem 7.3 produces the asserted actual Green form on the original XX. ◻

8.3. Products of curves and abelian varieties

Lemma 8.6 (Chow diagonal of a product of curves). Let FF have characteristic zero, let A/FA/F be an abelian variety, and let C1,…,Cs/FC_1,\ldots,C_s/F be smooth projective geometrically integral curves with FF-rational points. Put X=A×FC1×F⋯×FCsX=A\times_F C_1\times_F\cdots\times_F C_s. Then the diagonal of XX admits a finite decomposition (8.4), where every intermediate variety is the single abelian variety B=A×F∏iJac⁡(Ci)B=A\times_F\prod_i\operatorname{Jac}(C_i), and all correspondences are defined over FF.

Proof. Choose points Pi∈Ci(F)P_i\in C_i(F). The three mutually orthogonal Chow projectors π0,i=[Pi×Ci],π2,i=[Ci×Pi],π1,i=ΔCi−π0,i−π2,i\pi_{0,i}=[P_i\times C_i],\qquad \pi_{2,i}=[C_i\times P_i],\qquad \pi_{1,i}=\Delta_{C_i}-\pi_{0,i}-\pi_{2,i} split the motive of each curve into 1\mathbf 1, its degree-one motive, and the Lefschetz motive L\mathbb L. Idempotence follows from deg⁡(Pi)=1\deg(P_i)=1 and the mixed compositions vanish because [Pi]2=0[P_i]^2=0 on the curve. The Abel–Jacobi map and the Poincaré correspondence identify the degree-one motive with the degree-one direct summand of the Chow motive of Ji=Jac⁡(Ci)J_i=\operatorname{Jac}(C_i). The latter is split by the rational Chow–Künneth projector of the abelian variety, so the inverse correspondences exist in the category of rational Chow motives. These standard identities are established in [12, Sections 3 and 5].

Take the tensor product of the finitely many splittings, including the identity motive of AA. Every resulting summand factors through the motive of AA times a product of some JiJ_i, with an integral Tate twist recording the number of Lefschetz factors. The projection π\pi from B=A×∏iJiB=A\times\prod_i J_i onto any such subproduct admits a zero-section ii. The identity i∗π∗=id⁡i^*\pi^*=\operatorname{id}, implemented by transpose-graph correspondences, splits the motive of the subproduct from the motive of BB. Thus every summand can be factored through this one abelian variety, with the same Tate shift. Express the inclusion and retraction morphisms as algebraic correspondences uju_j and vjv_j; their composition is the tensor product of the chosen projectors. Summing over all choices of the three projectors on each curve gives ∑jvj∘uj=ΔX\sum_jv_j\circ u_j=\Delta_X in rational Chow groups. ◻

Lemma 8.7 (Regular models for products). Let A/KA/K be an abelian variety and let C1,…,Cs/KC_1,\ldots,C_s/K be smooth projective geometrically integral curves. After one finite extension F/KF/K there are simultaneously:

  1. rational points on all the Ci,FC_{i,F};

  2. regular projective strictly semistable models of the Ci,FC_{i,F}, and a regular projective strictly semistable model of AF×∏iCi,FA_F\times\prod_i C_{i,F};

  3. a regular projective semistable model of the abelian variety B=AF×∏iJac⁡(Ci,F)B=A_F\times\prod_i\operatorname{Jac}(C_{i,F}) satisfying Theorem 7.4.

Proof. Each curve acquires a rational point and a projective regular strictly semistable model after finite extension, by the semistable reduction theorem for curves. Enlarge this initial field so that AA has a regular projective strictly semistable model by Theorem 7.4. Now apply that theorem to the single abelian variety B=A×∏iJac⁡(Ci)B=A\times\prod_i\operatorname{Jac}(C_i), obtaining a possibly larger field on which BB has the required model. The earlier models of AA and the curves are carried to this final field by the ramified base-change regularization below. Semistable curve and abelian models remain regular and strictly semistable after an unramified extension. After ramified extension, Hartl’s explicit sequence of blowups recovers such a model without changing the generic fiber [11, Proposition 2.2]. Finally, his regularization of the product of two strict semistable models [11, Proposition 2.1] can be iterated over the finite family. All operations are projective blowups of projective schemes; thus the resulting model is projective, regular and flat, with the prescribed smooth generic fiber. ◻

Theorem 8.8 (Products of curves and abelian varieties). Let A/KA/K be an abelian variety and let C1,…,Cs/KC_1,\ldots,C_s/K be smooth projective geometrically integral curves, with K/QpK/\mathbf Q_p finite. On X=A×KC1×K⋯×KCs,X=A\times_K C_1\times_K\cdots\times_K C_s, every prescribed rational cycle of codimension 1≤r≤dim⁡X1\le r\le\dim X whose class is geometrically homologically trivial has an actual zero-curvature Green form satisfying (1.1) on its exact support complement over the original field, against all compact actual tests.

Proof. Choose F/KF/K and the regular semistable models of Lemma 8.7. The Chow-diagonal identity of Lemma 8.6 meets every hypothesis of Theorem 8.5. Apply that theorem. ◻

Corollary 8.9 (Finite domination by products). Let Y/KY/K be smooth projective geometrically integral, and suppose that there is a finite surjective morphism A×∏iCi→YA\times\prod_i C_i\to Y from a product as in Theorem 8.8. Then the conclusion of that theorem holds for every geometrically homologically trivial rational cycle on YY.

Proof. Pull a cycle back through the finite map and choose the common extension and models of Lemma 8.7. The proof of Theorem 8.5 constructs an actual numerical certificate for the pullback on a regular model of the product after that extension. The composite map from the base-changed product to YY is finite flat and has degree [F:K][F:K] times the degree of the original finite map. Apply Theorem 7.3 directly to this composite. ◻

9. The conditional geometric conclusions

9.1. Algebraic equivalence over the original field

Proof of Theorem 1.3. Write z=[Z]z=[Z]. Choose a finite Galois extension L/KL/K over which the curves, degree-zero divisors and correspondence identities expressing zLz_L by algebraic equivalence are defined. Theorem 6.6 provides an additive family Θ\Theta on XL×JX_L\times J, for a product of Jacobians J/LJ/L, with a fibre Θt=zL\Theta_t=z_L.

Let A=Res⁡L/KJA=\operatorname{Res}_{L/K}J. Evaluation AL→JA_L\to J and algebraic finite-flat pushforward give the norm family of Theorem 6.7. It is additive and has a KK-rational fibre (ΘN)a=[L:K]z.(\Theta^N)_a=[L:K]z. By (A1), the family defines the additive obstruction map φ:A(K)⟶HBC,actr,r(X),b⟼oX,Kr((ΘN)b).\varphi:A(K)\longrightarrow H^{r,r}_{\mathrm{BC,act}}(X), \qquad b\longmapsto\mathfrak o^r_{X,K}((\Theta^N)_b). Represent the family obstruction by a global model Chern form. Property (R) makes its pulled-back form constant near zero, so ker⁡φ\ker\varphi contains an identity neighbourhood. Compactness of A(K)A(K) implies that a positive multiple of any point lies in that neighbourhood, as in Theorem 6.2. The target is torsion-free, hence φ=0\varphi=0. Thus [L:K]oX,Kr(z)=0[L:K]\mathfrak o^r_{X,K}(z)=0, and division gives oX,Kr(z)=0\mathfrak o^r_{X,K}(z)=0.

Use (A2) for the original representative ZZ, not a moved cycle. Since [ΩZ]=0[\Omega_Z]=0, choose a global actual ΨZ\Psi_Z with DΨZ=ΩZD\Psi_Z=\Omega_Z. Set γZ=GZ−ΨZ∣UZ\gamma_Z=G_Z-\Psi_Z|_{U_Z}. Actual Stokes and (1.2) give (1.1) for every stated test, on the same open set UZU_Z. ◻

9.2. Abelian correction and diagonal transfer

Proof of Theorem 1.5. First let AA be an abelian variety over a finite extension of KK. Choose a further finite extension with the semistable model of Theorem 7.4. For a geometrically homologically trivial class, Theorem 7.5 gives a pure virtual model bundle whose special-fibre character is numerically trivial on all vertical cycles of the required dimension. Property (N), applied to the native finite map to the original KK-variety, gives zero obstruction. This uses the entire degree of the extension.

Now fix the retraction data (1.3) and z∈CH⁡r(X)Q,homz\in\operatorname{CH}^r(X)_{\mathbf Q,\mathrm{hom}}. Every (uj)∗zL(u_j)_*z_L is geometrically homologically trivial by compatibility of correspondences with cycle classes. Terms outside positive codimension on the abelian variety vanish; in codimension zero this follows from homological triviality. The first paragraph therefore annihilates the obstruction of each remaining term. Property (C) annihilates the obstruction of its image under vjv_j. Summing the Chow identity gives zero obstruction for zLz_L.

Apply the algebraic pushforward π∗:XL→X\pi_*:X_L\to X and (C). Since π∗zL=[L:K]z\pi_*z_L=[L:K]z, rational linearity gives oX,Kr(z)=0\mathfrak o^r_{X,K}(z)=0. If the correspondences act as λid⁡\lambda\operatorname{id} rather than the identity in the chosen codimension, the same argument gives λ[L:K]oX,Kr(z)=0\lambda[L:K]\mathfrak o^r_{X,K}(z)=0, and λ≠0\lambda\ne0 permits division. Finally apply (A2) and the global correction used in the preceding proof. ◻

References

[1] A. Mihatsch, δ\delta-Forms on Lubin–Tate spaces, arXiv:2112.10018, version 4 (2024), https://arxiv.org/abs/2112.10018v4.

[2] A. Mihatsch, On tropical intersection theory, Selecta Math. (N.S.) 29 (2023), art. 17, https://doi.org/10.1007/s00029-022-00818-7.

[3] A. Chambert-Loir and A. Ducros, Formes différentielles réelles et courants sur les espaces de Berkovich, arXiv:1204.6277, version 2 (2025), https://arxiv.org/abs/1204.6277.

[4] W. Gubler and F. Martin, On Zhang’s semipositive metrics, Doc. Math. 24 (2019), 331–372, https://doi.org/10.4171/DM/683.

[5] W. Fulton, Intersection Theory, second edition, Ergebnisse der Mathematik, Springer, 1998.

[6] K. Künnemann, Height pairings for algebraic cycles on abelian varieties, Ann. Sci. École Norm. Sup. (4) 34 (2001), 503–523, https://www.numdam.org/article/ASENS_2001_4_34_4_503_0.pdf.

[7] The Stacks Project Authors, Lemma 42.58.1, Tag 0FEY, https://stacks.math.columbia.edu/tag/0FEY.

[8] The Stacks Project Authors, Proposition 42.57.1, Tag 0FEW, https://stacks.math.columbia.edu/tag/0FEW.

[9] The Stacks Project Authors, Lemmas 42.59.7 and 42.59.13, Tags 0FF6 and 0FFA, https://stacks.math.columbia.edu/tag/0FF6, https://stacks.math.columbia.edu/tag/0FFA.

[10] A. J. de Jong, Smoothness, semi-stability and alterations, Publ. Math. Inst. Hautes Études Sci. 83 (1996), 51–93, https://www.numdam.org/item/PMIHES_1996__83__51_0.pdf.

[11] U. T. Hartl, Semi-stability and base change, Arch. Math. (Basel) 77 (2001), 215–221, https://doi.org/10.1007/PL00000484.

[12] A. J. Scholl, Classical motives, in Motives, Proc. Sympos. Pure Math. 55, Part 1 (1994), 163–187, https://www.dpmms.cam.ac.uk/~ajs1005/preprints/classical_motives.pdf.

[13] The Stacks Project Authors, Stacks Project, particularly Tags 0FB3, 0FBE, 0FET, 0FEW, 0D4I, 0FD6, 0G4C and 00R4, https://stacks.math.columbia.edu.

[14] J. S. Milne, Jacobian varieties, in Arithmetic Geometry, G. Cornell and J. H. Silverman (eds.), Springer, 1986, https://www.jmilne.org/math/xnotes/JVs.pdf.

[15] J. D. Achter, S. Casalaina-Martin and C. Vial, Parameter spaces for algebraic equivalence, Int. Math. Res. Not. IMRN (2019), no. 6, 1863–1893.

PDF and source

Download the paper

The PDF contains the full article, numbered statements, proofs, and references. The LaTeX download is self-contained, with the included proof sources expanded.

Edition and evidence

The frozen analytic arguments are draft proof sources. The principal transfer statements display their assumptions; no independent certification is added by this edition.

Source and PDF fingerprints

47 proof environments retained from the source. This checks document preservation, not mathematical validity.

Expanded LaTeX SHA-256

968f0560d434f7fb071eedff83c9d90c73a5387113975eb9e8a3a6697d56b263

PDF SHA-256

af692f792aebd3b0804afb05f9b9f720a3810a9dd9c4033ac975eb847bfff4dc

Permanent records and reader corrections

These links retain individual claim identities and revision-bound feedback. They are the reference index for this paper and its associated open problems.