22 · Green objects and higher codimension

15 NAG Tags

The geometric ingredients of the local Green construction are now available separately: completed residue fields and analytic fibres in Section 2.6, analytic flatness in Section 2.9, and Abhyankar dimension in Section 3.7. Proposition 2.6.4 gives the degree of a finite locally free fibre as a sum of lengths times residue-field degrees. In the argument below, proving that the required local map is finite and flat remains a geometric step; counting its underlying points would lose the multiplicity.

Bochner–Martinelli and Levine: what higher codimension changes

A complete intersection supplies a logarithmic maximum

Let XX be a pure nn-dimensional boundaryless analytic space in the source category. Let f1,…,frf_1,\ldots,f_r be a regular sequence defining a closed analytic subspace YY of pure codimension rr. On U=X∖YU=X\setminus Y, set u=log⁡max⁡(∣f1∣,…,∣fr∣).u=\log\max(|f_1|,\ldots,|f_r|). This is finite and continuous on UU. Its singularity along YY is exactly what is needed for a Green current. The source constructs the relevant products and extensions and states d′d′′(u(d′d′′u)r−1)=δY.\begin{equation} \mathrm d'\mathrm d''\bigl(u(\mathrm d'\mathrm d''u)^{r-1}\bigr)=\delta_Y. \end{equation} This is Theorem 18.7.2 of CLD25, explicitly attributed there to Mihatsch. For r=1r=1 it is Poincaré–Lelong.

The expression on the left has two levels. On UU, the continuous-potential product is available. Extending the resulting current to XX uses support and dimension results. The formula does not define an unrestricted product (d′d′′u)r(\mathrm d'\mathrm d''u)^r for every unbounded uu on all of XX. CLD itself points out this limitation immediately after the theorem.

The truncation argument and the nonformal step

For ε>0\varepsilon>0 define uε=log⁡max⁡(ε,∣f1∣,…,∣fr∣)=max⁡(log⁡ε,u).u_\varepsilon=\log\max(\varepsilon,|f_1|,\ldots,|f_r|) =\max(\log\varepsilon,u). The truncated function is finite and continuous. Its top relative curvature is described by a fibre over a small Gauss point in the parameter space. For a fixed test η\eta of degree (n−r,n−r)(n-r,n-r), the support and dimension lemmas ensure that the differentiated test can be treated away from YY in the extension argument. On the relevant compact support, sufficiently small truncations agree with uu.

The remaining step compares the calibrated fibre current with δY\delta_Y. This uses the variation theorem from CLD25 Chapter 14 and the regular-sequence geometry. It is the step that retains the scheme-theoretic multiplicities. The equality is therefore a theorem about the analytic complete intersection, not merely an identity of real functions on a tropical support.

There are several local typographical slips in the printed proof, including the second expression for uεu_\varepsilon and dimension indices. The corrected truncation above is the one used in this exposition. The theorem is also available through Mihatsch’s independent delta-form treatment, discussed in Chapter 16; the categories of representatives and tests must still be kept distinct.

The Green sign

With φ=min⁡iv(fi)=−u\varphi=\min_i v(f_i)=-u, the expression γ=(−1)r−1φ(d′d′′φ)r−1=−u(d′d′′u)r−1\gamma=(-1)^{r-1}\varphi(\mathrm d'\mathrm d''\varphi)^{r-1} =-u(\mathrm d'\mathrm d''u)^{r-1} has bidegree (r−1,r−1)(r-1,r-1) and satisfies d′d′′[γ]+δY=0\mathrm d'\mathrm d''[\gamma]+\delta_Y=0 in the appropriate current realization. This computation reconciles the logarithmic-maximum convention in CLD25 with the valuation-minimum convention used in the delta-form literature. It also shows why the sign cannot be guessed from the divisor case without counting the r−1r-1 additional factors.

Levine’s formula on projective space

Let X=PnX=\mathbb P^n and let Y≃Pn−rY\simeq\mathbb P^{n-r} be the linear subspace defined by x0=⋯=xr−1=0x_0=\cdots=x_{r-1}=0. On U=X∖YU=X\setminus Y, the rational projection f:U⟶Pr−1,[x0:⋯:xn]⟼[x0:⋯:xr−1]f:U\longrightarrow\mathbb P^{r-1},\qquad [x_0:\cdots:x_n]\longmapsto[x_0:\cdots:x_{r-1}] is a morphism. Give O(1)\mathcal O(1) the standard formal metric, and write α=c1(OPn(1)),β=c1(f∗OPr−1(1)),h=log⁡max⁡0≤i≤n∣xi∣max⁡0≤i<r∣xi∣.\alpha=c_1(\mathcal O_{\mathbb P^n}(1)),\qquad \beta=c_1(f^*\mathcal O_{\mathbb P^{r-1}}(1)),\qquad h=\log\frac{\max_{0\leq i\leq n}|x_i|}{\max_{0\leq i<r}|x_i|}. The homogeneous ratios make hh a well-defined function on UU. Set ω=∑j=0r−1αr−1−jβj.\omega=\sum_{j=0}^{r-1}\alpha^{r-1-j}\beta^j. After the source’s extension of hωh\omega to a current Λ\Lambda on XX, Theorem 18.7.5 gives d′d′′Λ+δY=αr.\begin{equation} \mathrm d'\mathrm d''\Lambda+\delta_Y=\alpha^r. \end{equation}

Away from YY, the algebra is transparent: d′d′′h=α−β\mathrm d'\mathrm d''h=\alpha-\beta, so (α−β)∑j=0r−1αr−1−jβj=αr−βr=αr,(\alpha-\beta)\sum_{j=0}^{r-1}\alpha^{r-1-j}\beta^j =\alpha^r-\beta^r=\alpha^r, because β\beta comes from a space of dimension r−1r-1. The nonformal part identifies the singular contribution along YY. The proof blows up the centre of projection, uses Poincaré–Lelong for the exceptional divisor, and pushes the resulting intersection current down. The exceptional fibres supply degree one, producing δY\delta_Y.

Why this does not solve every Green problem

The local formula uses a regular sequence of global functions on the chosen domain. The projective formula uses a specific linear cycle and a specific curvature term. A general homologically trivial cycle may not be globally a complete intersection, and local representatives may fail to glue in a prescribed differential-form sheaf. Even if a global Green current exists, it need not have an actual delta-form representative on the exact complement of the cycle.

Thus the formulas give substantial explicit cases and a proof mechanism. To reach a general strong Green statement, one must still specify the representative class, the full test class, the support condition, the gluing obstruction, and the descent under field extension. Chapter 24 turns these requirements into a precise comparison problem.

22.2 Green objects, star products, and local heights

The equation behind the construction

For the statements taken directly from original GK in this subsection and the next, work over a complete algebraically closed nontrivially valued field. The model examples can be compared after the base change just described. Mihatsch’s statements below have their own analytic hypotheses, which we state separately.

For a codimension-pp cycle ZZ, the integration current δZ\delta_Z has bidegree (p,p)(p,p). A Green object gZg_Z has bidegree (p−1,p−1)(p-1,p-1) and satisfies an equation d′d′′gZ+δZ=ωZ,\begin{equation} \mathrm d'\mathrm d''g_Z+\delta_Z=\omega_Z, \end{equation} where ωZ\omega_Z belongs to a specified class of forms. One must always state that class. In classical complex arithmetic intersection it is a class of smooth forms; in GK it is a class of δ\delta-forms; in the model theory discussed below, “forms” and “currents” are classes assembled from Chow groups of models. Sharing equation (22.2.1) does not make these categories identical.

For a metrized line bundle L‾\overline L with a nonzero meromorphic section ss, the basic Green potential is gD=−log⁡∥s∥g_D=-\log\|s\|, where D=div⁡(s)D=\operatorname{div}(s). The Poincaré–Lelong equation reads d′d′′[gD]+δD=c1(L‾).\mathrm d'\mathrm d''[g_D]+\delta_D=c_1(\overline L). It separates two sources of singularity: the zero or pole of the section, and the curvature of the metric. For a model metric, that curvature may itself be polyhedrally singular. This is precisely why the δ\delta-calculus is useful.

Suppose DD meets ZZ properly and the metric belongs to the GK class. The Green star product is gD∗gZ=gD∧δZ+c1(L‾)∧gZ.\begin{equation} g_D*g_Z=g_D\wedge\delta_Z+c_1(\overline L)\wedge g_Z. \end{equation} The first term means restrict the logarithmic potential to the cycle and push its current forward to XX. Proper intersection ensures that this restriction is the permitted one. The second term is the action of an allowed δ\delta-form on a current. These operations are part of the construction; the formula is not an invitation to multiply any two singular distributions. Applying d′d′′\mathrm d'\mathrm d'' and using the Green equations gives d′d′′(gD∗gZ)+δD⋅Z=c1(L‾)∧ωZ.\mathrm d'\mathrm d''(g_D*g_Z)+\delta_{D\cdot Z}=c_1(\overline L)\wedge\omega_Z. For two divisors, changing the order changes the representative by a d′\mathrm d'- and d′′\mathrm d''-boundary. The class in the appropriate quotient is symmetric; integration on a proper space kills these boundaries by Stokes. This is the mechanism behind symmetry of local heights.

References. [GK17] (Definition 11.1, §11.2, Propositions 11.4–11.5 and Corollary 11.6).

The local-height recursion and metric variation

Let XX be proper of dimension nn, and take n+1n+1 metrized divisors D^i\widehat D_i with properly meeting supports, or the more flexible pseudodivisors of GK with empty total intersection of supports. The integral of their star product defines a local height. Its useful computational form is the recursion λD^0,…,D^n(X)=λD^0,…,D^n−1(div⁡(sn))−∫Xanlog⁡∥sn∥ c1(L‾0)⋯c1(L‾n−1).\begin{align} \lambda_{\widehat D_0,\ldots,\widehat D_n}(X) ={}&\lambda_{\widehat D_0,\ldots,\widehat D_{n-1}}(\operatorname{div}(s_n))\\ &-\int_{X^{\mathrm{an}}}\log\|s_n\|\, c_1(\overline L_0)\cdots c_1(\overline L_{n-1}). \end{align} Each application replaces one intersection with a lower-dimensional cycle and one analytic integral. For model metrics, this agrees with the local height defined by model intersection theory.

If only the first metric is changed to ∥⋅∥0′=e−ρ∥⋅∥0\|\cdot\|_0'=e^{-\rho}\|\cdot\|_0, then λ′−λ=∫Xanρ c1(L‾1)⋯c1(L‾n).\begin{equation} \lambda'-\lambda=\int_{X^{\mathrm{an}}}\rho\, c_1(\overline L_1)\cdots c_1(\overline L_n). \end{equation} When the remaining metrics are semipositive, the mixed measure is positive and has mass deg⁡X(L1⋯Ln)\deg_X(L_1\cdots L_n). Hence ∣λ′−λ∣≤∥ρ∥∞deg⁡X(L1⋯Ln).\begin{equation} |\lambda'-\lambda| \leq\|\rho\|_\infty\deg_X(L_1\cdots L_n). \end{equation} This estimate explains why uniform approximation can extend intersection constructions in a controlled positive class. A limit argument still needs the measures, positivity, and independence of approximation to be established; continuity of functions alone supplies none of those automatically.

References. [GK17] (Propositions 12.4 and 12.6, Remark 12.7).

Example 22.2.1. (A height changes while curvature stays fixed). Use the standard model metric on OP1(1)\mathcal O_{\mathbb P^1}(1) from Example 20.3.1. Write z=Z1/Z0z=Z_1/Z_0 and t=−log⁡∣z∣t=-\log|z|. The Green functions of the points 00 and ∞\infty are g0=max⁡(0,t),g∞=max⁡(0,−t),g_0=\max(0,t),\qquad g_\infty=\max(0,-t), and both Chern measures are δξG\delta_{\xi_G}, where ξG\xi_G is the Gauss point. Formula (22.2.3) gives λ(0,∞)=g0(∞)+∫g∞ δξG=0.\lambda(0,\infty)=g_0(\infty)+\int g_\infty\,\delta_{\xi_G}=0. The closures of the two points in PR1\mathbb P^1_R are disjoint, giving the same model intersection. Now multiply the first norm by e−ce^{-c}. Its Green function becomes g0+cg_0+c, its curvature is unchanged, and its local height increases by cc.

Thus even the simplest local height cannot be recovered from the two curvature measures alone. Later we shall encounter a further loss of information coming from degree-zero line bundles, but scalar changes already establish the point. A theory that retains metrized line bundles contains more arithmetic information than a theory of their Chern currents.

Remark 22.2.2 (The variation bound for a signed measure). Write μ=c1(L‾1)⋯c1(L‾n)\mu=c_1(\overline L_1)\cdots c_1(\overline L_n) in (22.2.4). For a finite signed mixed measure, the general estimate is ∣λ′−λ∣≤∥ρ∥∞∥μ∥TV.|\lambda'-\lambda|\leq\|\rho\|_\infty\|\mu\|_{\mathrm{TV}}. If μ\mu is positive, its total variation is its mass, and the degree formula gives (22.2.5). For signed μ\mu, the signed mass need not control the total variation. Positivity is therefore essential to that particular degree bound.

Higher codimension: an explicit Green form and a length formula

Mihatsch supplies a particularly concrete higher-codimension construction. Let XX be a good Hausdorff pure nn-dimensional analytic space without boundary, and suppose Z=V(f1,…,fr)Z=V(f_1,\ldots,f_r) is a complete intersection of codimension rr. On X∖ZX\setminus Z set φ=min⁡iv(fi),γ=(−1)r−1φ(d′d′′φ)r−1.\varphi=\min_i v(f_i),\qquad \gamma=(-1)^{r-1}\varphi(\mathrm d'\mathrm d''\varphi)^{r-1}. Although φ\varphi is expressed using a minimum, the product is defined in the specified δ\delta-algebra. The theorem states d′d′′[γ]+δZ=0,\begin{equation} \mathrm d'\mathrm d''[\gamma]+\delta_Z=0, \end{equation} including the corresponding integration identity against compactly supported δ\delta-form tests. The codimension is visible in the exponent: γ\gamma has bidegree (r−1,r−1)(r-1,r-1), as a Green object should. For r=1r=1, it is the familiar logarithmic function v(f)v(f).

A separate theorem recovers formal intersection lengths. Assume RR is a DVR with v(π)=1v(\pi)=1. Let X\mathfrak X be separated and flat over Spf⁡R\operatorname{Spf}R, special, and pure of formal dimension n+1n+1. Let Z\mathfrak Z be a local complete intersection of pure codimension rr, and let Y⊂X\mathfrak Y\subset\mathfrak X be closed, Cohen–Macaulay, flat over RR, and pure of formal dimension rr. If Y∩Z\mathfrak Y\cap\mathfrak Z is Artinian, the distance function defined by the ideal of Z\mathfrak Z gives the same expression for γ\gamma, and ∫Y∖∂Yγ=length⁡ROY∩Z,Y=Yan.\begin{equation} \int_{Y\setminus\partial Y}\gamma =\operatorname{length}_R\mathcal O_{\mathfrak Y\cap\mathfrak Z}, \qquad Y=\mathfrak Y^{\mathrm{an}}. \end{equation} The theorem also proves the compact support needed for this integral. Flatness matters: a purely vertical formal subscheme can have empty generic fibre while retaining a nonzero intersection length. The general local-complete-intersection Green statement appears separately as a conjecture in that paper; the proved length formula must not be confused with a proof of that whole conjecture.

References. [Mih24] (Theorem 5.5, Conjecture 5.9, Corollary 5.10, and Theorem 5.11).

Example 22.2.3. (Intersection length and valuation scale). In Spf⁡R⟨x⟩\operatorname{Spf}R\langle x\rangle, the sections x=0x=0 and x=πmx=\pi^m have intersection ring R⟨x⟩/(x,x−πm)≃R/(πm).R\langle x\rangle/(x,x-\pi^m)\simeq R/(\pi^m). Its RR-length is mm. The Green function −log⁡∣x∣-\log|x| of the first section takes the value −log⁡∣πm∣=mλ-\log|\pi^m|=m\lambda at the generic point of the second, where λ=−log⁡∣π∣\lambda=-\log|\pi|. Thus the logarithmic value equals the algebraic length under the normalization v(π)=1v(\pi)=1; for the original absolute value it is the length multiplied by λ\lambda.

For global calculations, the 2025 BGK preprint adds another useful endpoint. Prescribed piecewise smooth Green data on an arbitrary projective variety can be expressed on a common extended tropicalization, and the corresponding local heights admit tropical formulas. The extended target includes strata at infinity; it is not merely one bounded graph. The result does not assert that every continuous singular metric is captured by one fixed finite tropicalization. It tells us that a finite collection of data in the stated class can be treated together, which is exactly what an intersection calculation requires.

References. [BGK25] (Theorem F, Proposition 8.2.2, and Theorem 8.2.3).

22.3 Truncation for a coordinate complete intersection

The complete-intersection theorem combines a differential calculation with a geometric multiplicity calculation. To make both visible, we give the full truncation argument for the coordinate origin in arbitrary dimension. The one external measure input will be the standard projective model-measure formula of Chapter 20. This case explains the signs, the extension of the current and the convergence of the truncated curvature without assuming a general extension theorem.

Let KK be complete, algebraically closed and nontrivially valued. Choose a∈Ka\in K with 0<∣a∣<10<|a|<1, and put X=(AKr)anX=(\mathbb A_K^r)^{\mathrm{an}}, r≥1r\geq 1. Write z1,…,zrz_1,\ldots,z_r for the coordinates and oo for the KK-rational origin. On U=X∖{o}U=X\setminus\{o\} define u=log⁡max⁡i∣zi∣,uj=log⁡max⁡(∣a∣j,∣z1∣,…,∣zr∣).u=\log\max_i|z_i|,\qquad u_j=\log\max\bigl(|a|^j,|z_1|,\ldots,|z_r|\bigr). Each uju_j is a finite Fubini–Study weight on the affine chart: it comes from the sections ajZ0,Z1,…,Zra^j Z_0,Z_1,\ldots,Z_r of OPr(1)\mathcal O_{\mathbb P^r}(1). Theorem 20.8.1 makes it admissible for the local uniform smooth-potential calculus. The function uu has the same property locally on UU, because one coordinate is nonzero near each of its points.

Lemma 22.3.1. (Smooth forms near a rational point). Every original smooth function is locally constant near a KK-rational analytic point. Every original smooth form of positive total degree vanishes on some neighbourhood of that point. For forms in one fixed finite local presentation, the vanishing neighbourhood may be chosen independently of the smooth coefficient functions.

Proof. An original smooth presentation uses finitely many analytic units fif_i. At a KK-rational point xx, the number fi(x)f_i(x) belongs to K×K^\times. Continuity gives a neighbourhood on which ∣fi−fi(x)∣<∣fi(x)∣|f_i-f_i(x)|<|f_i(x)|. The ultrametric equality then gives ∣fi∣=∣fi(x)∣|f_i|=|f_i(x)| throughout that neighbourhood. Thus every tropical coordinate −log⁡∣fi∣-\log|f_i| is constant there. Smooth scalar composites are constant and their positive-degree differential factors vanish. Taking the intersection for the finitely many coordinates proves the uniform statement for a fixed presentation. ◻

The lemma is particular to the original smooth sheaf. It is not an assertion about all continuous functions near a rational point, nor about every enlarged differential-form sheaf. It explains why some logarithmic extensions are easier on original tests than on stronger test classes.

Lemma 22.3.2. Let ξj\xi_j be the Gauss point of the closed polydisc ∣zi∣≤∣a∣j|z_i|\leq|a|^j. With the standard algebraic-degree normalization, (d′d′′uj)r=δξjon X.(d'd''u_j)^r=\delta_{\xi_j}\quad\text{on }X. Moreover ξj→o\xi_j\to o in XX, and therefore these measures converge weakly on compactly supported continuous tests to δo\delta_o.

Proof. The standard formal model of Pr\mathbb P^r has irreducible special fibre of multiplicity one, and deg⁡PK~rO(1)r=1\deg_{\mathbb P^r_{\widetilde K}}\mathcal O(1)^r=1. The model-measure comparison therefore identifies its top Chern measure with the single Dirac mass at the standard Gauss point. Apply the projective automorphism [Z0:Z1:⋯:Zr]⟼[ajZ0:Z1:⋯:Zr].[Z_0:Z_1:\cdots:Z_r]\longmapsto [a^jZ_0:Z_1:\cdots:Z_r]. The pullback weight in the frame Z0Z_0 is uju_j, and the inverse image of the standard Gauss point is ξj\xi_j. Functoriality for an analytic isomorphism proves the formula. This uses the model comparison, rather than inferring an analytic Dirac mass from a tropical vertex alone.

For a polynomial f=∑αaαzαf=\sum_\alpha a_\alpha z^\alpha, ∣f(ξj)∣=max⁡α∣aα∣ ∣a∣j∣α∣⟶∣a0∣=∣f(o)∣.|f(\xi_j)|=\max_\alpha |a_\alpha|\,|a|^{j|\alpha|} \longrightarrow |a_0|=|f(o)|. The analytic topology of affine space is generated by these evaluations, so ξj→o\xi_j\to o. Evaluation of a continuous compactly supported test then proves convergence of the Dirac masses. ◻

Theorem 22.3.3. The current u(d′d′′u)r−1u(d'd''u)^{r-1} on UU has a unique extension GG of bidegree (r−1,r−1)(r-1,r-1) to XX on original smooth tests. It satisfies d′d′′G=δo.d'd''G=\delta_o. Consequently γ=−G\gamma=-G is a zero-curvature Green current for the coordinate origin.

Proof. A compactly supported test for GG has bidegree (1,1)(1,1). By Lemma 22.3.1, it vanishes on a neighbourhood of oo, so its support is a compact subset of UU. Pair it with the current already defined on UU. This defines GG on all such tests. On a fixed finite presentation the neighbourhood of vanishing is common to all coefficients, so continuity reduces to continuity of the current on UU. The inductive test topology then gives continuity on the full test space. If two extensions existed, their difference would be supported at oo; every test vanishes near oo, so that difference would be zero.

Let η\eta be a smooth compactly supported scalar test. Its form d′d′′ηd'd''\eta vanishes near oo. On its compact support, max⁡i∣zi∣\max_i|z_i| has a strictly positive minimum. For all sufficiently large jj, the functions uju_j and uu therefore agree on a neighbourhood of this support. Locality of the admitted mixed product gives ⟨G,d′d′′η⟩=⟨uj(d′d′′uj)r−1,d′d′′η⟩=⟨(d′d′′uj)r,η⟩.\langle G,d'd''\eta\rangle =\left\langle u_j(d'd''u_j)^{r-1},d'd''\eta\right\rangle =\left\langle(d'd''u_j)^r,\eta\right\rangle. The second equality is the recursive mixed-product identity on the whole affine space, where uju_j is finite and admissible. Lemma 22.3.2 identifies the last term with η(ξj)\eta(\xi_j), which tends to η(o)\eta(o). The distributional d′d′′d'd'' convention gives ⟨d′d′′G,η⟩=⟨G,d′d′′η⟩\langle d'd''G,\eta\rangle=\langle G,d'd''\eta\rangle, proving the theorem. ◻

The algebraic multiplicity that the coordinate proof does not supply

Replace the coordinate equations by z1a1,…,zrarz_1^{a_1},\ldots,z_r^{a_r} with positive integers aia_i. Their common zero has the same underlying point, but the coordinate-ring quotient has basis z1b1⋯zrbr,0≤bi<ai,z_1^{b_1}\cdots z_r^{b_r},\qquad 0\leq b_i<a_i, and length a1⋯ara_1\cdots a_r. The complete-intersection current is therefore (a1⋯ar)δo(a_1\cdots a_r)\delta_o. This number is not recoverable from the support set {o}\{o\} alone. A proof for the logarithmic maximum of these powered functions must retain the degree of the finite monomial map or carry out an equivalent intersection calculation.

For a general regular sequence, the geometric fibre theorem performs this task, including nonreduced fibres and inseparable multiplicities. The coordinate proof above should therefore be read as a full proof of its stated case and an explanation of the general proof’s structure. It does not replace the variation-of-fibres theorem used in CLD25, nor the actual delta-form construction in Mihatsch.

There is a second distinction concerning tests. The extension in Theorem 22.3.3 was possible because original positive-degree tests vanish near a rational point. A theorem on all actual delta-form tests must verify its own extension and support statements; those tests need not satisfy the same vanishing lemma. Equality of the induced original currents does not by itself settle equality on the stronger test space.

References. [CLD25] (Theorem 18.7.2); [Mih24] (Theorem 5.5); [GK17] (Theorem 10.5). The proof above is restricted to the coordinate origin and explicitly uses the standard model-measure comparison.

22.4 Extension across a closed analytic subset

The coordinate proof extends because positive-degree original tests vanish near a rational point. In higher dimension the relevant inequality compares the test degree with the dimension of the removed analytic subset. It is the test degree, not the current degree, which controls this argument.

Let XX be a pure nn-dimensional boundaryless analytic space in the original CLD smooth-test category. Let Z⊂XZ\subset X be a closed analytic subset of dimension at most dd, and put U=X∖ZU=X\setminus Z. We use the following geometric input: for max⁡(a,b)>d\max(a,b)>d, restriction induces a topological isomorphism Aca,b(X)⟶Aca,b(U).A_c^{a,b}(X)\longrightarrow A_c^{a,b}(U). Its inverse extends tests by zero. In particular every such test on XX vanishes on a neighbourhood of ZZ. This is the smooth-test dimension argument underlying CLD25 Lemma 7.1.13, using its Lemmas 3.3.6 and 6.4.5. We retain the test bidegrees explicitly: the complementary indices in the last sentence of the printed proof of 7.1.13 must not replace the degrees of the tests actually being paired.

Proposition 22.4.1. If a current TT on UU has bidegree (p,q)(p,q) and max⁡(n−p,n−q)>d\max(n-p,n-q)>d, it has a unique extension T~\widetilde T to XX. It is given by ⟨T~,α⟩=⟨T,α∣U⟩.\langle\widetilde T,\alpha\rangle=\langle T,\alpha|_U\rangle. If the same degree inequality holds for a differential of TT, then taking that differential commutes with this unique extension.

Proof. Take the continuous dual of the test-space isomorphism above with (a,b)=(n−p,n−q)(a,b)=(n-p,n-q). This proves existence, continuity and uniqueness. For the second assertion, restriction of the differentiated extension agrees with the differential on UU. Uniqueness in its new degree therefore identifies it with the extension of that differential. The new-degree inequality is indispensable to this last step. ◻

For a codimension-rr subset, d=n−rd=n-r. A current of bidegree (r−1,r−1)(r-1,r-1) extends uniquely, since its tests have degree (n−r+1,n−r+1)(n-r+1,n-r+1). Its d′d′′d'd'' has bidegree (r,r)(r,r) and tests of degree (n−r,n−r)(n-r,n-r). The strict inequality has disappeared. A nonzero current supported on ZZ, such as δZ\delta_Z, is now possible.

Example 22.4.2. On the punctured analytic line, u=log⁡∣z∣u=\log|z| has d′d′′u=0d'd''u=0. Its original-current extension satisfies d′d′′[u]=δ0d'd''[u]=\delta_0. Extending the zero current gives zero, whereas differentiating the extension gives δ0\delta_0. Thus an unqualified assertion that extension by zero commutes with differentiation is false when the relevant support condition fails in the differentiated degree.

22.5 The complete-intersection calculation with its geometric inputs

We now separate the analytic limiting argument from the geometric fibre comparison. Let f1,…,frf_1,\ldots,f_r define a pure codimension-rr analytic complete intersection ZZ in XX, and write u=log⁡max⁡i∣fi∣on U,uε=max⁡(log⁡ε,u),ε>0.u=\log\max_i|f_i|\quad\text{on }U, \qquad u_\varepsilon=\max(\log\varepsilon,u),\quad\varepsilon>0. The latter expression means log⁡max⁡(ε,∣f1∣,…,∣fr∣)\log\max(\varepsilon,|f_1|,\ldots,|f_r|) on all of XX. Assume the continuous-potential products in this calculation are available with locality and the recursion d′d′′(uε(d′d′′uε)r−1)=(d′d′′uε)r.d'd''\bigl(u_\varepsilon(d'd''u_\varepsilon)^{r-1}\bigr) =(d'd''u_\varepsilon)^r. Assume also the following geometric comparison on the full original test space: (d′d′′uε)r⇀δZ(ε↓0).\begin{equation} (d'd''u_\varepsilon)^r\rightharpoonup\delta_Z \quad(\varepsilon\downarrow0). \end{equation} This is the substantive input. In CLD25 it is obtained by identifying truncated products with calibrated fibre currents in Corollary 18.6.2 and then applying the fibre comparison in Proposition 18.7.3. It includes the multiplicities of the complete intersection.

Theorem 22.5.1. Under these hypotheses, the unique original-current extension GG of u(d′d′′u)r−1u(d'd''u)^{r-1} satisfies d′d′′G=δZd'd''G=\delta_Z.

Proof. Proposition 22.4.1 gives GG. Fix a compactly supported smooth test η\eta of degree (n−r,n−r)(n-r,n-r). The form d′d′′ηd'd''\eta has degree (n−r+1,n−r+1)(n-r+1,n-r+1) and therefore vanishes near ZZ. Its support CC is a compact subset of UU. If CC is empty, all pairings below vanish. Otherwise the continuous function uu has a finite minimum MM on CC. Choose ε<eM\varepsilon<e^M. By continuity and compactness, after decreasing ε\varepsilon if necessary, uε=uu_\varepsilon=u on a neighbourhood of CC. Locality gives ⟨d′d′′G,η⟩=⟨G,d′d′′η⟩=⟨uε(d′d′′uε)r−1,d′d′′η⟩=⟨(d′d′′uε)r,η⟩.\begin{align*} \langle d'd''G,\eta\rangle &=\langle G,d'd''\eta\rangle\\ &=\left\langle u_\varepsilon(d'd''u_\varepsilon)^{r-1},d'd''\eta\right\rangle\\ &=\left\langle(d'd''u_\varepsilon)^r,\eta\right\rangle. \end{align*} Let ε\varepsilon tend to zero and use (22.5.1). The result is ⟨δZ,η⟩\langle\delta_Z,\eta\rangle. Every original test of the required degree was allowed, so this is the asserted current identity. ◻

Only a finite lower bound for uu was used; the printed requirement that this bound be strictly positive is unnecessary and generally false. The small parameter is compared with eMe^M, not with MM. This corrects both the truncation expression and the inequality used to make truncations agree on the test support.

What the fibre comparison must verify

CLD25 Theorem 14.7.2 is a theorem about compact analytic XX satisfying Serre’s condition (Sr)(S_r), a regular rr-dimensional target, and a morphism purely of relative dimension n−rn-r. After fixing a moment map, it gives local constancy of the calibrated top-dimensional tropical fibre near a rigid target point. Its hypotheses should be retained when the theorem is used. In particular, the relative dimension in this application is n−rn-r, not rr.

If a fixed test has a finite chart decomposition to which this theorem applies, equality of calibrated fibres gives equality of its fibre integrals on each chart. Summing the chart contributions gives the desired equality for that test. The neighbourhood in the target may depend on the test and its chart decomposition. No uniform neighbourhood valid for every test is needed for weak convergence.

The reduction from a general regular sequence to those geometric hypotheses is not reproduced by the truncation calculation. In particular, Serre conditions concern localizations, not only the depth at one point. The present proof keeps the calibrated-fibre statement as an explicit input. It proves the complete analytic passage once that input is supplied, without claiming to have independently repaired every geometric reduction in the source.

Why extension uniqueness ends at the critical degree

For s<r−1s<r-1, the same extension argument applies in both degrees (s,s)(s,s) and (s+1,s+1)(s+1,s+1). Thus locality and uniqueness give d′d′′(u(d′d′′u)s~)=(d′d′′u)s+1~.d'd''\bigl(\widetilde{u(d'd''u)^s}\bigr) =\widetilde{(d'd''u)^{s+1}}. At s=r−1s=r-1 the right-hand extension is no longer uniquely determined by its restriction to UU. The geometric fibre limit selects the missing contribution δZ\delta_Z. This is the precise place where the analytic complete intersection, rather than only its complement, enters the Green identity.

References. [CLD25] (Lemma 7.1.13, Theorem 14.7.2, Corollary 18.6.2, and Section 18.7). The geometric inputs and the conditional analytic proof are distinguished above.

22.6 The algebra behind finite-flat multiplicities

The finite-flat step in a regular-immersion formula ensures that the same scheme-theoretic multiplicity appears on the zero fibre and on nearby tropical fibres. In the finite case, the relevant local flatness criterion has a short module-theoretic proof. This proof also shows why nilpotents must be retained.

Lemma 22.6.1. Let (R,m)(R,\mathfrak m) be a Noetherian regular local ring of dimension rr, with regular parameters t1,…,trt_1,\ldots,t_r. If a finite RR-module MM has this sequence as an MM-regular sequence, then MM is free. Its rank is dim⁡R/mM/mM.\dim_{R/\mathfrak m}M/\mathfrak mM.

Proof. Induct on rr. If r=0r=0, the ring is a field. For r>0r>0, put t=t1t=t_1. The quotient ring R/tRR/tR is regular local of dimension r−1r-1, and the remaining parameters are regular on M/tMM/tM. By induction, M/tMM/tM is free over R/tRR/tR, say of rank ee. Lift a basis to MM. Nakayama’s lemma makes the resulting map Re→MR^e\to M surjective. Let its kernel be KK, which is finite because RR is Noetherian.

Since tt is a nonzerodivisor on MM, one has Tor⁡1R(M,R/tR)=0\operatorname{Tor}_1^R(M,R/tR)=0, as follows from the two-term resolution of R/tRR/tR. Tensoring the exact sequence with R/tRR/tR gives 0⟶K/tK⟶(R/tR)e⟶M/tM⟶0.0\longrightarrow K/tK\longrightarrow(R/tR)^e \longrightarrow M/tM\longrightarrow0. The right-hand map is the chosen basis isomorphism, so K/tK=0K/tK=0. Nakayama gives K=0K=0, proving freeness. Reduction modulo m\mathfrak m identifies the rank. ◻

Corollary 22.6.2. (Finite miracle flatness). Let R→AR\to A be a finite local homomorphism of Noetherian local rings, with RR regular, AA Cohen–Macaulay, and dim⁡A=dim⁡R\dim A=\dim R. Then AA is a finite free RR-module.

Proof. Parameters of RR generate an ideal whose quotient in AA is finite-dimensional over the residue field of RR, hence artinian. Their images therefore form a system of parameters of AA. In a Cohen–Macaulay local ring every system of parameters is regular. Apply Lemma 22.6.1 to the finite RR-module AA. ◻

The remaining commutative-algebra input is the system-of-parameters characterization of Cohen–Macaulay rings. In the analytic application, one must separately establish finiteness of the map and the dimension equality at the relevant points. The presence of a regular sequence somewhere in the source does not establish those geometric facts.

Example 22.6.3. (Equal dimensions without flatness). Let R=k[[t]]R=k[[t]] and A=R[ϵ]/(ϵ2,tϵ)A=R[\epsilon]/(\epsilon^2,t\epsilon). This is a finite local RR-algebra of dimension one. As an RR-module it is R⊕kϵR\oplus k\epsilon, so it is not flat: tt kills the nonzero element ϵ\epsilon. Its generic fibre has dimension one over k((t))k((t)), while its special fibre has length two. The missing Cohen–Macaulay condition is detected by depth zero at the closed point.

Example 22.6.4. (Flatness retains nonreduced degree). In contrast, R[ϵ]/(ϵ2)R[\epsilon]/(\epsilon^2) is free of rank two over RR, with basis 1,ϵ1,\epsilon. Every fibre has length two, although its reduced support has one point. Thus finite-flat degree counts scheme length, not the number of distinct points.

For a finite locally free morphism of rank ee, tensoring a local free basis with any residue field shows that every fibre algebra has dimension ee. This is the precise constancy used in a tropical weight comparison. It survives inseparability and nonreduced fibres. In a complete-intersection example fi=xiaif_i=x_i^{a_i} in a regular coordinate ring, the zero-fibre quotient has basis x1b1⋯xrbrx_1^{b_1}\cdots x_r^{b_r} with 0≤bi<ai0\leq b_i<a_i, hence length ∏iai\prod_i a_i. The corresponding cycle in a Green identity must retain this multiplicity.

References. This is the finite-module proof of the local flatness criterion underlying [Mih24] (Proposition 3.13). Compare Stacks Project, Tags 00R4 and 07DY.

22.7 Regular sequences and the finite-flat comparison

The geometric input in the preceding truncation proof has an independent construction in Mihatsch’s Proposition 3.13. The decisive points are Abhyankar points of the complete intersection. At these points the ambient local ring has dimension exactly equal to the length of the regular sequence. This produces the Cohen–Macaulay hypothesis required for a finite-flat comparison. It does not require a Serre condition on an entire neighbourhood.

Lemma 22.7.1. Let (A,m)(A,\mathfrak m) be a Noetherian local ring of dimension rr. If f1,…,fr∈mf_1,\ldots,f_r\in\mathfrak m is an AA-regular sequence, then AA is Cohen–Macaulay. In this situation, if a finite local homomorphism B→AB\to A has regular local source of dimension rr and dim⁡A=r\dim A=r, then AA is flat over BB.

Proof. The sequence gives depth⁡A≥r\operatorname{depth}A\geq r, whereas depth is at most dimension. Thus equality holds. For the second assertion, let t1,…,trt_1,\ldots,t_r be a regular system of parameters of BB. Finiteness makes A/(t1,…,tr)AA/(t_1,\ldots,t_r)A artinian, so their images are a system of parameters of AA. Every system of parameters in a Cohen–Macaulay local ring is regular. Corollary 22.6.2 now applies. Equivalently this is the finite case of miracle flatness. The first assertion supplies the Cohen–Macaulay assumption used in the second. ◻

The dimension hypothesis cannot be replaced by the mere existence of a regular sequence. In particular, a regular sequence of length rr at one local ring does not imply that this ring satisfies (Sr)(S_r) at all its localizations. The following example makes the issue explicit.

Example 22.7.2. For any field kk and r≥1r\geq 1, put A=(k[x,y,t1,…,tr]/(x2,xy))(x,y,t1,…,tr).A=\bigl(k[x,y,t_1,\ldots,t_r]/(x^2,xy)\bigr)_{(x,y,t_1,\ldots,t_r)}. The elements t1,…,trt_1,\ldots,t_r form a regular sequence, but AA does not satisfy (S1)(S_1), hence does not satisfy (Sr)(S_r). Its underlying spectrum is pure of dimension r+1r+1, and the quotient by the sequence is pure of dimension one.

Proof. A polynomial variable is a nonzerodivisor over any coefficient ring, and this remains true after successively quotienting by the previous variables and localizing. Thus the sequence is regular. The nilradical is (x)(x) and the reduced ring is the regular local ring k[y,t1,…,tr](y,t1,…,tr)k[y,t_1,\ldots,t_r]_{(y,t_1,\ldots,t_r)}, proving the dimension and purity assertions.

The nonzero class of xx has annihilator p=(x,y)\mathfrak p=(x,y), a nonminimal prime. In ApA_{\mathfrak p} this class remains nonzero and is killed by the maximal ideal. Hence depth⁡Ap=0\operatorname{depth}A_{\mathfrak p}=0, while dim⁡Ap=1\dim A_{\mathfrak p}=1. This violates (S1)(S_1). Every open neighbourhood of the closed point of Spec⁡A\operatorname{Spec}A contains its generization p\mathfrak p, so shrinking around that point does not remove the failure. ◻

This is a counterexample to the unrestricted commutative-algebra inference, not to the Green identity. It explains why the sentence in the proof of CLD25 Proposition 18.7.3 asserting a neighbourhood with (Sr)(S_r) needs justification beyond the regular sequence alone. Mihatsch instead makes the Cohen–Macaulay argument only at the relevant Abhyankar points, where the required dimension equality holds.

The openness theorem cited at this point, [Duc09] (Theorem 4.4), is a theorem about the locus where the local ring has the specified property; its definition is given in Section 3.4 of that source. Openness does not prove that the complete intersection lies in that locus. In Example 22.7.2, the depth at the closed maximal ideal is at least rr, but the local ring fails Serre’s condition because one of its localizations has depth zero and positive dimension. This distinction between one depth inequality and the ring condition is the missing commutative-algebra issue.

The weighted product near a regular immersion

Let XX be a pure nn-dimensional affinoid analytic space and let Z=V(f1,…,fr)Z=V(f_1,\ldots,f_r) be defined by a regular sequence. Put m=n−rm=n-r, U×=X∖V(f1⋯fr),U≥C×={x∈U×:v(fi(x))≥C for all i}.U^\times=X\setminus V(f_1\cdots f_r),\qquad U^\times_{\geq C}=\{x\in U^\times:v(f_i(x))\geq C\text{ for all }i\}. Choose a moment map g:X→Gmsg:X\to\mathbb G_m^s. Write T(X,g)T(X,g) for its weighted tropicalization in the top dimension of XX; lower-dimensional pieces of the set-theoretic image are not included in this notation.

Theorem 22.7.3. (Mihatsch’s product theorem). For all sufficiently large CC, one has an equality of weighted top-dimensional polyhedra T(U≥C×,(f,g))=[C,∞)r×T(Z,g),T(U^\times_{\geq C},(f,g))=[C,\infty)^r\times T(Z,g), where the first factor has its standard lattice weight.

Open proof obligation (RC2). Establish the analytic moment-image, boundary dimension and finite-fibre properness inputs in the exact regular immersion scope; the proof takes them from Mihatsch Proposition 3.13.

NAG-900454 · Proof status

The proof keeps the regular sequence until the finite-flat step, so it retains nonreduced multiplicities. Replacing ZZ by its reduced support before this step would generally change eσe_\sigma. The product structure is not a product decomposition of the analytic space, and it does not say that every local test is originally expressed in the coordinates (f,g)(f,g). The latter issue is the reason for the refinement argument in the next section.

References. [Mih24] (Proposition 3.13 and its proof); [CLD25] (Theorem 14.7.2 and Proposition 18.7.3). The commutative-algebra inputs are the regular-sequence criterion for Cohen–Macaulayness and the local flatness criterion, Stacks Project, Tags 07DY and 00R4; Serre conditions are defined in Tag 031O.

22.8 The complete-intersection identity on delta-form tests

We now use Mihatsch’s analytic delta-sheaf Bp,qB^{p,q}, not the original smooth sheaf Ap,qA^{p,q}. The support theorem, integral and Stokes theorem are those of [Mih24] (Corollary 4.9, Definition 4.10 and Proposition 4.11). They are part of the input calculus. In particular, equality will be tested on every actual compactly supported delta-form of the indicated degree.

Let XX be pure of dimension nn and boundaryless, and let Z=V(f1,…,fr)Z=V(f_1,\ldots,f_r) be a complete intersection of codimension rr. Distinguish the two open sets U=X∖Z,U×=X∖V(f1⋯fr)⊆U.U=X\setminus Z,\qquad U^\times=X\setminus V(f_1\cdots f_r)\subseteq U. On UU, the function φ=min⁡iv(fi)\varphi=\min_i v(f_i) is finite and piecewise smooth. Near a point where some fif_i vanish, discard those indices: their valuations are larger than the finite minimum on a sufficiently small neighbourhood. The remaining functions are analytic units there and give a tropical presentation of φ\varphi. Thus φ\varphi is defined on the exact complement UU, not only on U×U^\times.

Lemma 22.8.1. (The diagonal current). Let ϕ(x)=min⁡(x1,…,xr)\phi(x)=\min(x_1,\ldots,x_r) on Rr\mathbb R^r, and let Δ\Delta be the diagonal with primitive parameter t↦(t,…,t)t\mapsto(t,\ldots,t). In the polyhedral delta-calculus, (−1)r−1(d′d′′ϕ)r−1=δΔ,(−1)r−1ϕ(d′d′′ϕ)r−1=tδΔ.(-1)^{r-1}(\mathrm d'\mathrm d''\phi)^{r-1}=\delta_\Delta, \qquad (-1)^{r-1}\phi(\mathrm d'\mathrm d''\phi)^{r-1}=t\delta_\Delta.

Open proof obligation (RC2). Prove the fan computation with the stated diagonal support and coefficients; Mihatsch Lemma 5.7 is supplied.

NAG-900455 · Proof status

Define the actual delta-form γ=(−1)r−1φ(d′d′′φ)r−1∈Br−1,r−1(U).\gamma=(-1)^{r-1}\varphi(\mathrm d'\mathrm d''\varphi)^{r-1}\in B^{r-1,r-1}(U). Its support is contained in the locus where all finite v(fi)v(f_i) agree, hence in U×U^\times. On this locus it is the pullback of tδΔt\delta_\Delta. Away from ZZ, d′d′′γ=0\mathrm d'\mathrm d''\gamma=0. This last identity follows from the affine function tt on the closed diagonal and the differential rules in the delta-calculus.

Why one tropical presentation is insufficient

For an original smooth test, one can use a finite affinoid covering and inclusion–exclusion: the test is a pullback of a superform, and its high-degree restriction to lower-dimensional boundary images vanishes. For an arbitrary delta-form, its restriction to a skeleton can involve the other charts in its definition. The smooth argument cannot simply be repeated by replacing the letter AA with BB.

Fix η∈Bcn−r,n−r(X)\eta\in B_c^{n-r,n-r}(X). Near Supp⁡(η∣Z)\operatorname{Supp}(\eta|_Z) choose finitely many affinoid domains KiK_i, moment maps gig_i and presentations of η\eta in their interiors. The support theorem puts η∣Z\eta|_Z on the top-dimensional skeleton of ZZ. Each such skeleton has finite fibres under its tropicalization. Shrinking the affinoids separates the relevant fibre points from the boundary. Choose a smooth partition of unity λi\lambda_i with support in these interiors and let ε=∑iλi\varepsilon=\sum_i\lambda_i, equal to one near Supp⁡(η∣Z)\operatorname{Supp}(\eta|_Z).

Lemma 22.8.2. (Local product identity with a cutoff). With these choices, let νi\nu_i be the pushforward of λiη∣Z\lambda_i\eta|_Z from its skeleton under gig_i. For sufficiently large CC, and every polyhedral delta-form α\alpha on Rr\mathbb R^r, the corresponding pushforward upstairs, restricted to (C,∞)r(C,\infty)^r, is (f,gi)trop,∗(ftrop⋆α∧λiη)=α×νi.(f,g_i)_{{\rm trop},*} \bigl(f_{\rm trop}^{\star}\alpha\wedge\lambda_i\eta\bigr) =\alpha\times\nu_i. All restrictions to the skeletal domain and to {φ>C}\{\varphi>C\} are understood in this formula. Pushforward is taken only where the support is proper over the target.

Open proof obligation (RC2). Prove the supported cutoff product and its compact gluing step for actual delta tests; Mihatsch Lemma 5.8 is supplied.

NAG-900456 · Proof status

Theorem 22.8.3. (Complete-intersection Green identity). For every actual η∈Bcn−r,n−r(X)\eta\in B_c^{n-r,n-r}(X), ∫X∖Zγ∧d′d′′η=−∫Zη∣Z.\begin{equation} \int_{X\setminus Z}\gamma\wedge\mathrm d'\mathrm d''\eta=-\int_Z\eta|_Z. \end{equation} In particular d′d′′[γ]+δZ=0\mathrm d'\mathrm d''[\gamma]+\delta_Z=0 on the original smooth test space as well.

Open proof obligation (RC2). Prove degree vanishing for the actual supported delta tests after the cutoff; the proof applies Mihatsch equation (5.22).

NAG-900457 · Proof status

This is the proof of [Mih24] (Theorem 5.5), reorganized around the product theorem, the cutoff refinement and the one-dimensional integral. It supplies the complete-intersection identity on the full stated test class. It does not solve a Green problem for an arbitrary higher-codimension cycle, where global regular-sequence equations need not exist. Nor does the proof descend an actual form through an alteration. Those are different geometric problems, discussed in Chapter 23.

22.9 The blow-up proof of Levine’s formula

We now prove the projective-space formula from the divisor identity and the degree of a projective-space fibre. This isolates the geometric step suppressed by the telescoping calculation on the complement. Work in the common algebraic setting of original GK, with KK algebraically closed, complete and nontrivially valued. Use its delta-metrics, intersection products, and proper pushforward of currents.

Let Y⊂PnY\subset\mathbb P^n be the linear subspace X0=⋯=Xr−1=0X_0=\cdots=X_{r-1}=0, with 1≤r≤n1\leq r\leq n. Denote the blow-up by π:X~→Pn\pi:\widetilde X\to\mathbb P^n and the resolved projection by b:X~→Pr−1b:\widetilde X\to\mathbb P^{r-1}. For r=1r=1, the blow-up in the Cartier divisor is the identity; the following formulas still apply.

Lemma 22.9.1. The blow-up is the incidence variety in Pn×Pr−1\mathbb P^n\times\mathbb P^{r-1} defined by XiUj=XjUiX_iU_j=X_jU_i for 0≤i,j<r0\leq i,j<r. Its exceptional Cartier divisor EE satisfies E≃Y×Pr−1,O(E)≃π∗OPn(1)⊗b∗OPr−1(−1).E\simeq Y\times\mathbb P^{r-1},\qquad \mathcal O(E)\simeq\pi^*\mathcal O_{\mathbb P^n}(1)\otimes b^*\mathcal O_{\mathbb P^{r-1}}(-1).

Proof. On an affine chart of Pn\mathbb P^n, the ideal of YY is generated by the first rr affine coordinates. In the blow-up chart Ui≠0U_i\ne 0, set uj=Uj/Uiu_j=U_j/U_i. Its equations are Xj=XiujX_j=X_i u_j for j<rj<r. These are precisely the usual Rees-algebra charts of the blow-up of this coordinate ideal. They glue to the incidence variety. On YY, all the equations in the UU-coordinates disappear, giving the product description of EE. The pullback ideal becomes the invertible ideal O(−E)\mathcal O(-E), locally generated by XiX_i. The sections X0,…,Xr−1X_0,\ldots,X_{r-1} of π∗O(1)\pi^*\mathcal O(1) have common factor cutting out EE; after dividing by it, they generate π∗O(1)⊗O(−E)\pi^*\mathcal O(1)\otimes\mathcal O(-E) and define bb. Hence this line is b∗O(1)b^*\mathcal O(1), which is the asserted line-bundle identity. ◻

Give both hyperplane lines their standard model metrics and set a=π∗αa=\pi^*\alpha, β=b∗c1(O‾(1))\beta=b^*c_1(\overline{\mathcal O}(1)). The quotient metric on O(E)\mathcal O(E) makes the Green function of its canonical section, outside EE, equal to H=log⁡max⁡0≤i≤n∣Xi∣max⁡0≤i<r∣Xi∣.H=\log\frac{\max_{0\leq i\leq n}|X_i|} {\max_{0\leq i<r}|X_i|}. This is an equality of norms, not just of curvatures: it follows by evaluating the quotient of the two maximum norms in a chart Ui≠0U_i\ne 0. Poincaré–Lelong therefore gives d′d′′[H]=a−β−δE.\mathrm d'\mathrm d''[H]=a-\beta-\delta_E.

Theorem 22.9.2. Put Ω=∑j=0r−1ar−1−jβj\Omega=\sum_{j=0}^{r-1}a^{r-1-j}\beta^j and define the current Λ=π∗([H]∧Ω)\Lambda=\pi_*([H]\wedge\Omega). Then d′d′′Λ+δY=αr.\mathrm d'\mathrm d''\Lambda+\delta_Y=\alpha^r. On Pn∖Y\mathbb P^n\setminus Y, it is represented by the formula h∑j=0r−1αr−1−jβjh\sum_{j=0}^{r-1}\alpha^{r-1-j}\beta^j of Section 22.1.

Proof. All factors of Ω\Omega are closed delta-forms, so multiplication of the divisor equation and the current Leibniz rule give d′d′′([H]∧Ω)=ar−βr−δE∧Ω=ar−δE∧Ω.\mathrm d'\mathrm d''([H]\wedge\Omega)=a^r-\beta^r-\delta_E\wedge\Omega =a^r-\delta_E\wedge\Omega. Here βr=0\beta^r=0 because it is pulled back from the (r−1)(r-1)-dimensional projective space. Proper pushforward commutes with the differential, and π∗ar=αr\pi_*a^r=\alpha^r by the projection formula and the generic degree one of π\pi. It remains to compute the exceptional term. On E=Y×Pr−1E=Y\times\mathbb P^{r-1}, the form aa comes from YY and β\beta from the second factor. Pair π∗(δE∧ar−1−jβj)\pi_*(\delta_E\wedge a^{r-1-j}\beta^j) with any complementary admitted test. The test also comes from YY after restriction to EE. If j<r−1j<r-1, the integrand has insufficient degree in the projective-space fibre, so its fibre integral is zero. For j=r−1j=r-1, the fibre integral of βr−1\beta^{r-1} is deg⁡OPr−1(1)r−1=1\deg\mathcal O_{\mathbb P^{r-1}}(1)^{r-1}=1. The remaining integral is exactly the test integral over YY. Hence π∗(δE∧Ω)=δY\pi_*(\delta_E\wedge\Omega)=\delta_Y. This proves the current identity. Outside YY, the map π\pi is an isomorphism, so the representative is the displayed one. The argument constructs a current globally and a specified form on the exact complement; it does not assert that arbitrary proper pushforward preserves actual delta-forms. ◻

The product structure of EE was essential to this short proof. For a general regular immersion, the exceptional divisor is the projectivized normal bundle, and pushing powers of its tautological class produces Segre classes. The linear-subspace formula cannot simply be copied with a general cycle substituted for YY. Such a generalization requires both the normal-cone intersection formulas and admissible Green representatives.

22.10 Metric variation and completion of local heights

The one-metric variation formula gives a simultaneous continuity theorem without differentiating singular functions. Fix line bundles L0,…,LnL_0,\ldots,L_n on a proper nn-dimensional variety and fixed pseudodivisor data for which the initial local heights are defined. Suppose all metrics under consideration are semipositive, and the model or delta-metric height theory has symmetry and the one-metric variation formula.

Theorem 22.10.1. For two metric tuples m,m′\boldsymbol m,\boldsymbol m', put εi=d(mi,mi′)\varepsilon_i=d(m_i,m_i') and Vi=deg⁡X∏j≠iLjV_i=\deg_X\prod_{j\ne i}L_j. Then ∣λ(m)−λ(m′)∣≤∑i=0nεiVi.|\lambda(\boldsymbol m)-\lambda(\boldsymbol m')| \leq\sum_{i=0}^n\varepsilon_i V_i. Consequently local height extends uniquely to uniform limits of admitted semipositive metrics, with the fixed pseudodivisor data.

Proof. Replace the metrics one at a time, in the order 0,…,n0,\ldots,n. The iith difference is the integral of the iith weight difference against a mixed positive measure formed from the other metrics, some old and some new. Its mass is ViV_i, since the underlying line bundles did not change. Its absolute value is at most εiVi\varepsilon_iV_i. Sum the n+1n+1 inequalities. For uniformly Cauchy metric sequences, the estimate makes the height sequence Cauchy. If two sequences have the same limiting metrics, their mutual distances tend to zero, so the estimate makes their height limits equal. This proves existence and independence of approximation. Uniqueness follows from density of the original metric class in its uniform completion. ◻

Remark 22.10.2. This is a completion theorem after a height theory has been constructed on a positive class. It is not a construction of arbitrary mixed measures from uniform convergence alone. Symmetry, positivity, the degree formula and the one-metric variation identity are the inputs that make the proof work. For signed integrable metrics, expand into differences of semipositive metrics or use a uniform total-variation bound; the signed total degree alone is insufficient.

From local metric variation to global height variation

For a global field with fixed product-formula weights nvn_v, changing a metric only at finitely many places gives the sum of the local variations: h′−h=∑vnv∫Xvanρv c1(L‾1,v)⋯c1(L‾n,v).h'-h=\sum_v n_v\int_{X_v^{\mathrm{an}}}\rho_v\, c_1(\overline L_{1,v})\cdots c_1(\overline L_{n,v}). For point heights the mixed measure is replaced by the normalized finite orbit measure. A change by a principal scalar has local weights −log⁡∣a∣v-\log|a|_v; the product formula makes the corresponding global sum zero. An arbitrary independent rescaling at each place does not have this cancellation. This is why local constants matter even though principal global changes leave heights unchanged.

References. [CLD25] (Theorem 18.7.5); [GK17] (Sections 7, 11–12); [GH]. The Levine proof uses only the stated divisor, projection and degree theorems after the explicit blow-up calculation.

22.11 Three meanings of a non-Archimedean Green object

The word Green is used for three different constructions in the recent literature. Their equations share some notation but their unknowns belong to different spaces. Keeping the categories explicit prevents a solution in one theory from being mistaken for a solution in another.

Model-theoretic currents

Dubocs works in the non-Archimedean Arakelov theory of Bloch–Gillet–Soulé. The ground data are an excellent discrete valuation ring and a proper smooth variety over its fraction field. Section 1.1 assumes the existence of regular proper flat models and common dominating regular models. These model hypotheses hold, for example, under the indicated resolution hypothesis; excellence alone is not stated to imply them. Forms and currents are built from bivariant intersection classes and Chow groups of models and their special fibres. They are not defined as continuous linear functionals on CLD test forms. Let FF be a metrized vector bundle of rank rr with a regular section whose zero scheme is ZZ. In [Dub26] (Theorem 4.1), blow up ZZ, let ss be the canonical exceptional section, and let QFQ_F be the tautological quotient with its induced metric. The constructed model current is Λ=π∗(div⁡v(s)∧f∗cr−1(QF)),ddcΛ+δZ=[cr(F)].\Lambda=\pi_*\bigl(\operatorname{div}_v(s)\wedge f^*c_{r-1}(Q_F)\bigr), \qquad dd^c\Lambda+\delta_Z=[c_r(F)]. The formula is useful because it expresses the higher-codimension Green current through an exceptional divisor and a quotient Chern class. When F=L⊕rF=L^{\oplus r}, Theorem 6.1 gives the corresponding Levine formula. Neither statement asserts that Λ\Lambda is an actual Mihatsch delta-form on Xan∖∣Z∣anX^{\mathrm{an}}\setminus|Z|^{\mathrm{an}}. A comparison would have to define realization maps, match Chern classes and pushforwards, and verify the identity against the required analytic tests.

Monge–Ampère Green functions

A Green function in Boucksom–Jonsson’s pluripotential theory solves a nonlinear measure equation, for example MA⁡ω(ϕx)=δx\operatorname{MA}_\omega(\phi_x)=\delta_x with their normalized measure. It is a scalar psh potential. Their Theorem A shows that a divisorial Green function is PL in dimension one but need not be PL in higher dimension. More strongly, Theorem B constructs on PC((t))3\mathbf P^3_{\mathbf C((t))} a continuous potential whose Monge–Ampère measure has finite support in a dual complex, while the potential does not factor through the retraction of any SNC model. Thus a skeletal support theorem for a measure is not a skeletal factorization theorem for its potential. This result rules out a general shortcut from finitely many measure atoms to a potential on a finite skeleton. It does not contradict the comparison theorem in Section 16.8, where strong faithfulness, a pre-existing tropical solution and absolute continuity are additional hypotheses.

Actual forms for cycle equations

The strong Green problem considered here asks for an actual form of bidegree (r−1,r−1)(r-1,r-1) on the exact cycle complement, satisfying a linear d′d′′d'd'' identity against all admitted actual tests. Mihatsch’s complete-intersection result belongs to this third category. It has stronger representative and testing content than a current identity on original smooth tests, but retains its own geometric hypotheses. Dubocs’s construction and Monge–Ampère Green functions provide comparisons and possible methods; neither removes those hypotheses. For zero curvature one must also eliminate the curvature term [cr(F)][c_r(F)] or its analogue. Homological triviality of a cycle is not a construction of a primitive in the actual delta-sheaf. Section 24.15 supplies a relevant vanishing theorem on closed CLD tests; Sections 22.8 and 23.4 explain the additional representative and descent requirements.

References. [Dub26] (Section 1.1 and Theorems 4.1, 6.1); [BJ24] (Theorems A–B); [Mih24] (Theorem 5.5).