Skip to content

Section 12.1 · PDF pages 821–823

The inverse Poincaré–Lelong identity

The notation transcription is provisional. Check it against the printed mathematics when citing or reusing a formula.

12.1 The inverse Poincaré–Lelong identity

The metric input is an integration identity, rather than an assertion about positivity. We first identify its coordinate integral, since an incorrect normalization here would change the arithmetic intersection formula used later.

Lemma 12.1 (The source’s sheet integral). Let P define an integral horizontal multiprojective arithmetic variety of arithmetic dimension e + 1. Let ϕ_{1}, . . . , ϕ_{e} be smooth real potentials on its complex multicone, transforming under blockwise nonzero scalars by real linear combinations of log |c_{k}|. Let g be an invariant function which is either smooth or the logarithmic norm of a nonzero algebraic section. Then the source’s ptsIntegral is ∫

Selectable PDF notation • transcription unchecked
i

g dd^{c}ϕ_{1} ∧· · · ∧dd^{c}ϕ_{e}, dd^{c}= ∂∂.^{̄} (MN.1)

Selectable PDF notation • transcription unchecked
X(ℂ)^{π}

The integral is over the reduced complex cycle, and may equivalently be computed on a resolution with pulled-back forms. This equality also applies to each horizontal component of a divisor.

Source: ReciprocityTheory/ReductiveGroups/ArithmeticGeometry/BirationalGeometry/IsIntersectionFamily.lean:83--148; ReciprocityTheory/ReductiveGroups/ArithmeticGeometry/BirationalGeometry/IsSemipositiveOn.lean:206--219.

Proof. Write z_{j} = x_{j} + iy_{j} and H for the real Hessian of a real function ϕ. Direct differentiation gives

Selectable PDF notation • transcription unchecked
∂^{2}ϕ_{=} H(e_{i}, e_{j}) + H(ie_{i}, ie_{j}) + i{H(e_{i}, ie_{j}) −H(ie_{i}, e_{j})}_{.}
∂z_{i}∂z̄_{j} 4

This is exactly cxHessian. Expanding the exterior product of the e forms gives, relative to ordinary Lebesgue measure on ℂ^{e},

Selectable PDF notation • transcription unchecked
( 2)^{e} (1 ∑ (∑ ))

e! Re ( (−1)^{e−|S|} det(cxHessian ϕ_{j})). (MN.2)

Selectable PDF notation • transcription unchecked
^{πe!}S⊆{1,...,e} j∈S

Indeed inclusion–exclusion extracts the term using each of the e matrices once; i dz_{j} ∧dz̄_{j} = 2 dx_{j} ∧dy_{j} accounts for the factor 2^{e}. Formula (MN.2) is precisely the definition of maDensity. No curvature sign or factor of two is implicit.

Embed X_{ℂ} by the Segre map. A general linear projection to P^{e}_{ℂ} has centre disjoint from X_{ℂ} and is finite: choose e + 1 linear forms giving a homogeneous Noether normalization of its coordinate ring. This supplies a GoodProj; hence the branch returning zero when no such projection exists is inapplicable. On the affine chart where the first projection coordinate is 1, outside the branch and singular loci, the finite map is a finite union of holomorphic sheets. Their complement is algebraic of smaller complex dimension and therefore has Lebesgue measure zero. A sheet lifts locally to nonzero multicone coordinates. Two such lifts differ by nonzero holomorphic scalars in the blocks. The differences of the pulled-back potentials are sums of log |c_{k}|, whose dd^{c} is zero; g is unchanged. The density is thus independent of the arbitrary sheet chosen in the source’s existential definition. The real change-of-variables formula on these sheets, with (MN.2), identifies its finite sum and Lebesgue integral with (MN.1). The missing projective hyperplane also has smaller complex dimension. For e = 0 the same argument is just the finite sum over the points of the zero-dimensional reduced cycle, with density 1.

To justify convergence at the omitted sets, take a projective resolution of the complex variety, using

Proposition 11.4. The pulled-back line metrics are smooth on this compact manifold. The logarithmic norm of a nonzero section is locally integrable: in a local frame it is log |f| minus a smooth function, and the local integrability assertion is part of the Poincaré–Lelong theorem [45, III,(2.15),p. 143]. It is therefore absolutely integrable against each smooth top-degree density on the compact resolution. Smooth g is simpler. The exceptional sets and omitted hyperplanes have measure zero there. Applying change of variables on increasing relatively compact portions of the sheets and then absolute integrability gives the stated equality of the full integrals. The argument is unchanged for a horizontal prime divisor. Vertical arithmetic components have empty complex fibre and contribute no such integral.

Proposition 12.2 (The precise metric identity). Let P be as in the source, of arithmetic dimension n + 2. Let L_{1}, . . . , L_{n}, A, B be smooth conjugation-invariant hermitian lines in its multicone convention. Assume that B has degree zero in each block, and that s is a nonzero multihomogeneous section of A on P. Then

ptsIntegral_{P} (log h_{B}, log h_{L1}, . . . , log h_{Ln})(log ‖s‖_{A})

Selectable PDF notation • transcription unchecked
∑

= m_{W} ptsIntegral_{W} (log h_{L1}, . . . , log h_{Ln})(log h_{B}) W∈divSupp(P,s) W horizontal − ptsIntegral_{P} (log h_{A}, log h_{L1}, . . . , log h_{Ln})(log h_{B}),

where m_{W} is the localized length specified by multAt. This is PoincareLelongInv, including its sign and normalization.

Source: Development/AutomorphicSpectrum/PublishedTheorems/Metrics2/PoincareLelongInv.lean:9--21.

Proof. Put u = log h_{B}, α = dd^{c} log h_{L1} ∧· · · ∧dd^{c} log h_{Ln} and v = log ‖s‖_{A}. The zero block degrees make u a globally defined smooth real function on X(ℂ). The homogeneity of the other metrics makes their curvature forms global. In a local frame, v = log |f| − log h_{A}, exactly as in the source’s definition log(|s(z)|/h_{A}(z)). The Poincaré–Lelong current equation is therefore

dd^{c}v = [div s] −dd^{c} log h_{A}. (MN.3)

This is Demailly’s formula V,(13.2), p. 271, divided by 2π and written for log ‖s‖ rather than log ‖s‖^{2} [45]. On a compact smooth resolution, integration by parts for the locally integrable current v and smooth uα gives ∫ ∫ ∫

Selectable PDF notation • transcription unchecked
∫

v dd^{c}u ∧α = u dd^{c}v ∧α = uα − u dd^{c} log h_{A} ∧α. (MN.4)

Selectable PDF notation • transcription unchecked
div s

There is no boundary term. Equivalently this is the distributional extension of the smooth integration-by-parts formula III,(3.1),p. 144 of the same reference; smoothness and closedness of α leave only dd^{c}u ∧α when applying dd^{c} to the test form.

We check the multiplicities when pushing (MN.4) down to the original, possibly singular variety. At a codimension-one local ring R of an integral component, f is a nonzerodivisor and R/(f) has finite length. Let R̃ be its finite semilocal normalization. The module R̃/R has finite length. Apply multiplication by f to 0 →R →R̃ →R̃/R →0. The kernel and cokernel of multiplication by f on the last finite-length module have equal length. The resulting exact sequence consequently shows

Selectable PDF notation • transcription unchecked
∑

length_{R}(R/fR) = length_{R}(R̃/fR̃) = [κ(𝔮) : κ(R)] ord_{𝔮}(f).

Selectable PDF notation • transcription unchecked
𝔮

Thus the pushed-forward zero cycle has exactly the source’s localized lengths, not just set-theoretic compo-nents. Subsequent exceptional divisors mapping into codimension at least two contribute zero: the pullback of n curvature forms to an n-dimensional divisor whose image has dimension at most n −1 has zero top exterior product on its dense smooth locus. Arithmetic vertical components have no complex points. Finally

Lemma 12.1 translates all three terms of (MN.4) into the source’s integrals. Its divSupp is finite, so the finite-support sum is the displayed ordinary finite sum. All conjugation hypotheses are preserved under this construction; no division by two is introduced because ptsIntegral integrates the full complex cycle.