Section 9.8 · PDF pages 716–738
The adapted frame and nonvanishing construction from result 9.49
The notation transcription is provisional. Check it against the printed mathematics when citing or reusing a formula.
Lemma 9.49 (The theta-line degree from its power formula). Write d_{K} = [K : ℚ] and g = dim B. Under the source’s key power formula, with its local unit clauses and archimedean theta normalization, the theta and determinant lines satisfy
deĝ(T^{−1}) =^{1}_{2}deĝ(D^{−1}) +^{g}_{4}d_{K} log(2π). (C.46)
Here for a metrized fractional line L = (I, n_{σ}) the convention is deĝL = − log N(I) −∑_{σ} log n_{σ}.
Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/ ThetaLineDualDegreeOfKeyFormulaPowerForm.lean:31--117.
Proof. For the fixed positive power N, local formula (C.44), together with the chosen local generators, says that the valuation of κ equals that of c^{8}_{a}^{N}/c^{4}_{d}^{N} at every prime. Apply the statement also to inverses to retain equality, not just one inequality. Unique factorization of fractional ideals gives
Selectable PDF notation • transcription unchecked
(κ) = I_{T}^{8N}I_{D}^{−4N}.Consequently ∑_{σ} log |σκ| = 8N log N(I_{T} )−4N log N(I_{D}) by the norm formula for a principal ideal. The archimedean clause of the same key power formula is
log |σκ| + 8N log n_{T,σ} −4N log n_{D,σ} = 2gN log(2π).
It uses the determinant of the chosen Lie tangent matrix and the volume of its period parallelepiped; their quotient is exactly the inverse square of the determinant-line norm. Sum this identity over all d_{K} embeddings and substitute the preceding ideal-norm identity. Group terms using the displayed degree convention and divide by 8N > 0. This gives (C.46). The power and its unit ambiguities cancel; the factor 2π and the factor 1/4 do not.
Lemma 9.50 (Hadamard bounds the degree of the coefficient line). Suppose every Lie-letter line has normal-ized degree Ω/2, ℒ ≥1, and retain the determinant and norm bounds proved above. Put P(Ω) = (1+|Ω|)/2.
Selectable PDF notation • transcription unchecked
−µ(v)
For a coefficient of total degree s, its line N_{µ} = (T^{−1})^{n} ∏_{v}ℓ_{e(v)} has
Selectable PDF notation • transcription unchecked
( )
deĝN_{µ}
Selectable PDF notation • transcription unchecked
1
≤ P(Ω) +_{2} log ℒ (s + gn). (C.47)
Selectable PDF notation • transcription unchecked
d_{K}Source: Formalism/LocalToGlobal/CMDeterminantIdeals/ CMIdealValuations/DetValuationLeOne.lean:170--245; StatementC.lean:20--42.
Proof. The local determinant inequalities give an inclusion of fractional ideals I_{ℓ1} · · · I_{ℓg}(det U) ⊂I_{D}. Their absolute norms therefore satisfy the reverse size inequality. At each complex embedding Hadamard’s deter-minant inequality, applied to the image-letter Gram matrix, gives
|σ det U| n_{D,σ} ≤ℒ^{g} ∏n_{ℓj,σ}.
Selectable PDF notation • transcription unchecked
j
One direct proof is to project each successive vector orthogonally off the earlier ones; its residual length is no larger than its original length, and the product of residual lengths is the absolute Gram volume. Taking logs, summing embeddings, and using ∑_{σ} log |σ det U| = log N((det U)) with the ideal inclusion yields
Selectable PDF notation • transcription unchecked
̂ ∑
deĝℓ_{j} degD ≥ −g log ℒ = gΩ/2 −g log ℒ.
Selectable PDF notation • transcription unchecked
d_{Kj} d_{K}Combine with (C.46) to bound the normalized degree of T^{−1} by g(−Ω/4 +^{1}_{2} log ℒ +^{ 1}_{4} log(2π)). Since log(2π) < 2 and −Ω/4 ≤|Ω|/2, this is at most g(P(Ω) +^{1}_{2} log ℒ). Degrees add on tensor products and change sign on duals. Therefore
deĝN_{µ}/d_{K} = ndeĝ(T^{−1})/d_{K} −sΩ/2.
Use −Ω/2 ≤P(Ω) and log ℒ ≥0 to obtain (C.47). This retains both the s and the gn terms.
Proposition 9.51 (The coefficient product formula gives the CM lower bound). Under the stated good-CM, key-power and line-degree hypotheses, any nonzero frame polynomial of a section integral at all the source’s integral models satisfies
Selectable PDF notation • transcription unchecked
( ) ∑
− P(Ω) +_{2}^{1} log ℒ (s + gn) −s log(rℒ) ≤^{1} log S_{σ}, (C.48)
Selectable PDF notation • transcription unchecked
d_{Kσ}provided each S_{σ} > 0. This is the source’s quantitative CM-frame lower bound, with s =∑_{b,i}a_{bi}.
Source: Formalism/LocalToGlobal/CMDeterminantIdeals/CMIdealValuations/ AbsNormSpanGalois.lean:240--370; Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/ CmPointLowerBoundOfCoefficientBound.lean:14--79; MeanLogSupNormLowerBoundCmFrameNonvanishing.lean:49--89.
Proof. Choose a nonzero coefficient c_{µ} of G; its degree is exactly s by Lemma 9.48. For each embedding apply (C.45) on the polydisc of radii ρ_{v} = 1/n_{ℓe(v),σ}. The coefficient bound used by the source can be proved by a finite Fourier average, avoiding any boundary-integral assertion. Choose an integer A larger than every exponent of every monomial of∏ G and of µ, and a primitive Ath root ζ of unity. Multiply G((ρ_{v}ζ^{kv})_{v}) by _{v}ζ^{−kvµ(v)} and average over all k_{v} ∈{0, . . . , A −1}. Expansion into its finite monomial sum kills every coefficient except c_{µ}, because ∑^{A}_{k=0}^{−1}ζ^{k(b−a)} is zero unless a = b in the chosen exponent range, when it equals A. Taking absolute values of the average proves
Selectable PDF notation • transcription unchecked
|σc_{µ}| n^{n}_{T−1,σ}∏n^{−}_{ℓe}^{µ}_{(v}^{(}_{)}^{v}_{,σ}^{)} ≤S_{σ}(rℒ)^{s}.
vThe left side is the length of c_{µ} in the metrized coefficient line N_{µ}.
Corollary 9.44 says c_{µ} ∈I_{Nµ}. Hence (c_{µ}) ⊂I_{Nµ} and N((c_{µ})) ≥N(I_{Nµ}). The principal-ideal norm identity then gives ∑
log ‖c_{µ}‖_{Nµ,σ} ≥−deĝN_{µ}.
Selectable PDF notation • transcription unchecked
σ
All lengths are positive, since the coefficient is nonzero. Take logs in the coefficient upper bounds, sum,∑ and combine with this inequality. One obtains −deĝN_{µ}/d_{K} −s log(rℒ) ≤d^{−}_{K}^{1}_{σ} log S_{σ}. Insert (C.47) to obtain (C.48).
When the point was defined over a finite extension of the original coefficient field, S_{σ} is pulled back from the restriction of σ to that field. Every embedding of the smaller field has the same number of extensions. Thus its normalized mean is unchanged, and (C.48) descends to that coefficient field. The nonzero coefficient is available because the polynomial evaluates nontrivially at the original adapted frame, whose moduli point and section were preserved by the CM construction. This verifies both the field average and the nonvanishing required in the source use.
Lemma 9.52 (The rank-one trace form and its discriminant norm). For a definite pure-type period datum of 𝒪_{E}-rank one, there are a nonzero fractional ideal 𝔞 ⊂E, an integral module identification e : M ≃𝔞, and a conjugation-invariant totally positive β ∈E such that
Selectable PDF notation • transcription unchecked
τ_{0}h(a, b) = ∑ τ(βe(a)e(b)) (C.49)
τ(^{τ}rt^{:})=^{E→}τ_{0}^{K}(rt)over every splitting field K. They satisfy
2 log N𝔞 + log |N_{E/ℚ}β|
Selectable PDF notation • transcription unchecked
1
=_{2} log d − log rd(E). (C.50)
Selectable PDF notation • transcription unchecked
[E : ℚ]
The order generator is w = (1 + rt)/2, with rt^{2}= −d.
Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/ ExistsFractionalIdealTraceFormRankOne.lean:20--260.
Proof. Choose an E-coordinate on the one-dimensional generic module. The integral module maps injec-tively onto a nonzero fractional ideal 𝔞, since it is finite over 𝒪_{E} and spans that coordinate line. The adjoint alternating form has expression
ϵB(a, b) = Tr_{E/ℚ}(βe(a)e(b)/rt)
for a nonzero β = β^{̄}; this follows by representing the trace-linear functional on the coordinate line, using nondegeneracy of the field trace. Its alternating and adjoint identities force the indicated conjugation sym-metry. The formula (C.30), together with ̄w = 1 −w, now gives (C.49). Indeed, expanding both rational traces over all embeddings weights the embedding τ by (τ(w) −1 + τ_{0}(w))/τ(rt). This weight is one if τ(rt) = τ_{0}(rt) and zero otherwise, as substitution of w = (1 + rt)/2 shows.
Definiteness implies Tr(βxx̄) > 0 for all nonzero x ∈E. To check this implication from the integral positivity hypothesis, multiply x by a nonzero integer so that it belongs to 𝔞 and use the formula for B(wa, a); the alternating term B(a, a) vanishes. Every complex value of β is real. If its value at one infinite place were negative, choose a ∈E with absolute value greater than one at that place and less than one at all the others. Such an a exists by approximation in the Minkowski coordinates: a ℚ-basis of E gives an isomorphism E ⊗_{ℚ} ℝ ≃ ℂ^{[E:ℚ]/2}, and rational coefficients are dense. The traces of βa^{n}ā^{n} would then eventually be negative, a contradiction. Nonvanishing of β excludes zero at every embedding. Hence β is totally positive.
Perfectness of the integral alternating form means that 𝔞 is self-dual for the trace pairing with scalar ϵβ/rt. The inverse-different calculation of Lemma 9.24 therefore gives
Selectable PDF notation • transcription unchecked
𝔞(ϵβ/rt)̄𝔞 = 𝔇^{−}_{E}^{1}.Take ideal norms. Conjugation preserves the norm, the unit sign has absolute norm one, and |N_{E/ℚ}rt| = d^{[E:ℚ]/2} by its square equation. Thus 2 log N𝔞+log |Nβ| = [E : ℚ] log d/2−log |D_{E}|, which is (C.50).
Proposition 9.53 (The exact integral eigenline of a pure tensor). Let the pure tensor abelian scheme be formed from the preceding rank-one datum and an integral elliptic curve with positive squared Lie norm n_{σ} at each embedding of the coefficient field. For any embedding τ : E →K of the matching pure type, its integral Lie eigenline has normalized degree
2 deg Lie_{τ} = − mean_{σ} log n_{σ} +_{2}^{1} log d − log rd(E). (C.51)
The line is the intersection with the actual integral Lie lattice, including at primes ramified in E.
Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/ TwoMulDegreeEigenLatticeTensorRankOne.lean:22--315; TwoMulDegDualLineCmIdeal.lean:15--181.
Proof. Work over a field splitting E. Index the matching embeddings by i, and put
Selectable PDF notation • transcription unchecked
ev(a)_{i} = τ_{i}(e(a)), dv(b)_{i} = τ_{i}(βe(b)).Their scalar pairing is the image of h(a, b) by (C.49). The vectors ev(a) span the full coordinate space. For otherwise a nontrivial linear relation between their coordinate functions would contradict independence of the distinct embeddings on 𝒪_{E}: multiply a fixed nonzero e(a_{0}) by every algebraic integer and apply the automorphism-independence argument of Lemma 16.2.
Let c(a) be the Lie tangent of the pure-tensor map at the integral elliptic Lie generator. The tensor Lie presentation shows that the integral Lie lattice is exactly ∑_{a}𝒪_{K}c(a). For the finite Hermitian dual system (x_{m}, y_{m}) of Lemma 9.33, define
Selectable PDF notation • transcription unchecked
∑ ∑ )
Φ(z) = z_{i} dv(y_{m})_{i} c(x_{m}).
Selectable PDF notation • transcription unchecked
m i
The dual-system formula gives Φ(ev(a)) = c(a). The polarization pairing implies that its Hermitian metric is ∑
Selectable PDF notation • transcription unchecked
H_{σ}(Φz, Φz^{′}) = σ(τ_{i}β)n_{σ}σz_{i} σz_{i}^{′}. (C.52)
iHere is a check that diagonal metric identities suffice. Their difference is a Hermitian biadditive form vanish-ing on every diagonal c(a). Substituting a + b kills its real part. Substituting a + wb then kills its imaginary part, because w acts on that pure type by a nonreal scalar. Thus the difference vanishes on all pairs of span-ning vectors, proving (C.52). Positivity of β and n_{σ} makes Φ injective. The integral E-action is diagonal in these coordinates, by its identity on the spanning c(a). Distinct embeddings are separated by an algebraic integer, so the τ_{i} eigenspace is precisely KΦ(e_{i}). ∑ We identify its integral lattice without inverting a discriminant. The coordinate lattice is L_{0} =
_{a}𝒪_{K} ev(a). A vector z belongs to L_{0} if and only if ∑_{i}z_{i} dv(b)_{i} ∈𝒪_{K} for every b. Neces-sity follows from the integral∑ ∑ h-pairing. For sufficiency, the same finite dual system expresses z as
_{m}(_{i}z_{i} dv(y_{m})_{i}) ev(x_{m}); the identity holds first on the spanning vectors and hence everywhere. Taking z = te_{i} gives
Selectable PDF notation • transcription unchecked
L_{0} ∩Ke_{i} = 𝔟^{−1}e_{i}, 𝔟 = 𝒪_{K}τ_{i}(β𝔞̄).This proves the required integral, ramification-sensitive intersection.
Put v = Φ(e_{i}). Its squared norm is σ(τ_{i}β)n_{σ} by (C.52). The normalized degree of the fractional line 𝔟^{−1}v is therefore
log N𝔟 1 −_{2} mean_{σ} log(σ(τ_{i}β)n_{σ}).
Selectable PDF notation • transcription unchecked
[K : ℚ]
Norms of extended fractional ideals are raised to [K : E]. Also every embedding of E has the same number of extensions, so the mean log of σ(τ_{i}β) is log |N_{E/ℚ}β|/[E : ℚ]. Substitute these two facts, double the degree, and obtain
2 log N𝔞 + log |Nβ| − mean_{σ} log n_{σ}.
Selectable PDF notation • transcription unchecked
[E : ℚ]
Formula (C.50) gives (C.51). The same computation of ideal norms and repeated embeddings shows that this normalized degree is unchanged under the finite coefficient extensions used in the construction; the source uses these base changes before descending the equality to the original pure CM object.
Lemma 9.54 (CM period averaging from the two multiplier families). Let k be imaginary quadratic, fix ι : k ↪→ℂ, and choose an integral ideal 𝔟_{C} in each ideal class C ∈ Cl(𝒪_{k}). Put h = # Cl(𝒪_{k}). Let M contain k, and let Λ_{0,σ} and Λ_{C,σ} be complex period lattices, for every σ : M ↪→ℂ. Assume Λ_{0,σ} is stable under σ𝒪_{k}. Suppose there are β_{C} ∈M^{×} and δ_{C} ∈𝒪_{M}^{×} with
Selectable PDF notation • transcription unchecked
∆(Λ_{C,σ}) = σδ_{C}, ∆(Λ) = g_{2}(Λ)^{3}−27g_{3}(Λ)^{2},and the following two conditions, including their integral assertions: for every x ∈𝔟^{−}_{C}^{1}, β_{C}x is integral and σ(β_{C}x)Λ_{0,σ} ⊂Λ_{C,σ}; for every y ∈𝔟_{C}, β_{C}^{−1}y is integral and σ(β_{C}^{−1}y)Λ_{C,σ} ⊂Λ_{0,σ}. Here x, y are mapped through the fixed inclusion k ⊂M. Writing A(Λ) for Euclidean area and D(Λ) = A(Λ)^{6}|∆(Λ)|, one has
Selectable PDF notation • transcription unchecked
1 ∑ ∑
log A(Λ_{0,σ}) = ^{1} log D(ι𝔟_{C}). (C.53)
Selectable PDF notation • transcription unchecked
[M : ℚ]_{σ} 6h_{C}Proof. Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/MeanLogCovolumeEqMeanClassGroup.lean:211--312.
First, the two integrality assertions imply 𝔟_{C}𝒪_{M} = (β_{C}): the first says (β_{C})𝔟^{−}_{C}^{1}𝒪_{M} ⊂𝒪_{M}, and the second says 𝔟_{C}𝒪_{M} ⊂(β_{C}). Invertibility of the ideal gives the reverse inclusion from the first. Taking ideal norms therefore gives ∑
log |σβ_{C}| = [M : k] log N𝔟_{C}. (C.54)
Selectable PDF notation • transcription unchecked
σ
The period inclusions likewise give the exact equality
Selectable PDF notation • transcription unchecked
Λ_{C,σ} = σβ_{C} Λ_{0,σ}𝔟^{−}_{C}^{1}.For clarity, the second inclusion characterizes membership on the right by the condition that multiplication by every y ∈𝔟_{C} lands in σβ_{C}Λ_{0,σ}. Conversely choose finitely many x_{j} ∈𝔟^{−}_{C}^{1}, y_{j} ∈𝔟_{C} with ∑_{j}x_{j}y_{j} = 1, and use the first inclusion on each term. A rank-two period lattice stable under 𝒪_{k} is homothetic to a fractional 𝒪_{k}-ideal. Indeed its rational span has dimension one over k; after choosing a nonzero generator, its inverse image is a finitely generated full 𝒪_{k}-submodule of k. Thus multiplication by 𝔟^{−}_{C}^{1} divides area by N𝔟_{C}, and
log A(Λ_{C,σ}) = 2 log |σβ_{C}| + log A(Λ_{0,σ}) − log N𝔟_{C}.
The area assertion can also be checked on an integral ideal: its index in 𝒪_{k} is its norm, and indices multiply; clearing a common denominator gives the fractional case. The sum of log |σδ_{C}| is zero since δ_{C} is a unit. Use (C.54) and [M : ℚ] = 2[M : k] to obtain, for each C,
Selectable PDF notation • transcription unchecked
∑ ∑
log D(Λ_{C,σ}) = 6 log A(Λ_{0,σ}). (C.55)
Selectable PDF notation • transcription unchecked
σ σ
There is a second way to sum the left side. For fixed σ, write Λ_{0,σ} = c_{σ}σ𝔞_{σ}. Then Λ_{C,σ} has ideal class [𝔞_{σ}]/C. Multiplication and inversion permute the finite class group. Also D is unchanged by homothety: area scales by |c|^{2}, whereas g_{2}, g_{3} scale by c^{−4}, c^{−6}. It is unchanged by conjugation, because the defining absolutely convergent lattice sums for g_{2}, g_{3} conjugate term by term. The restriction of σ to k is ι or its conjugate. Consequently, in either case,
Selectable PDF notation • transcription unchecked
∑ ∑
log D(Λ_{C,σ}) = log D(ι𝔟_{C}).
Selectable PDF notation • transcription unchecked
C C
Sum (C.55) over C and interchange the two finite sums. Division by 6h[M : ℚ] gives (C.53). This argument uses the actual two multiplier families; it does not require an additional assertion about the distribution of Galois conjugates among ideal classes.
Lemma 9.55 (The Lie metric and the eta normalization). In Lemma 9.54, suppose the lattices come from principally polarized elliptic schemes with integral Lie generators as in the source. Let n_{σ} be the squared Lie norm in the manuscript’s convention, which is π times the Hermitian form whose imaginary part is the integral Riemann form. If τ_{C} ∈ℍ represents class C, then
Selectable PDF notation • transcription unchecked
∑ ( )
mean_{σ} log n_{σ} = −^{1} log 4πℑτ_{C} |η(τ_{C})|^{4} . (C.56)
Selectable PDF notation • transcription unchecked
h_{C}Proof. Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/HermFormPrincipalDimOne.lean:13--174; IsLieNormSqIffEqPiDivCovolume.lean:13--28; MeanLogLieNormSqEqMeanLogEta.lean:60--99.
Write w for the nonzero tangent generator and H(w, w) = a > 0. If ω_{1}, ω_{2} form a period basis, the Riemann form on that basis has off-diagonal entry aℑ(ω_{1}ω̄_{2}). It is integral and unimodular, so that entry is 1 or −1. Explicitly every entry is an integer multiple of it, and surjectivity onto the integral dual forces that integer to divide 1. Hence a = 1/A(Λ), and n_{σ} = π/A(Λ_{0,σ}). ∑
The discriminant identity in [1, Theorem4.21, p.57] uses g_{2} = 60∑_{ω̸=0}ω^{−4}, g_{3} = 140_{ω̸=0}ω^{−6},
exactly the source’s invariants. With η(τ) = q^{1/24} ∏_{m≥1}(1 −q^{m}), q = e^{2πiτ}, that identity is ∆(ℤ + ℤτ) = (2π)^{12}η(τ)^{24}. Since A(ℤ + ℤτ) = ℑτ, homothety invariance gives
Selectable PDF notation • transcription unchecked
D(ι𝔟_{C}) =(4π^{2}ℑτ_{C} |η(τ_{C})|^{4})^{6}.These are positive quantities, so their real logarithms are valid. If the embedding defining τ_{C} is the other embedding of k, use conjugation invariance of D before the same calculation. Substitute in (C.53) and in log n_{σ} = log π − log A(Λ_{0,σ}). The extra log π cancels one of the two factors of π, giving exactly (C.56).
We also check how the scheme hypotheses supply the preceding lemma’s integral multiplier assertions. A generic elliptic homomorphism extends over each discrete valuation ring of 𝒪_{M} by the semiabelian Hom-extension result (Proposition 10.31). Its derivative in integral Lie bases is integral at every finite prime and thus lies in 𝒪_{M}. Base change preserves this derivative. Under complex uniformization a homomorphism has derivative c precisely when multiplication by c sends the source period lattice into the target lattice: it descends to the quotient in one direction, and the lifted linear map of the universal covers proves the other. Apply this first to the endomorphisms by 𝒪_{k}, then to every β_{C}x and β_{C}^{−1}y in the specified families. This proves all conditions of Lemma 9.54 for the same schemes, generators and coefficient field. The source’s construction of those schemes and their unit discriminants remains a separate input, not a consequence of this averaging.
Lemma 9.56 (The real Eisenstein constant term used by the source). For τ = x + iy ∈ℍ and real s > 1, put
Selectable PDF notation • transcription unchecked
∑_{y}s
E(τ, s) =_{2s} .
(m,n)∈ℤ^{2}\{0}^{|mτ + n|}This is the sum over all nonzero pairs, with no primitive-pair restriction. It converges, and as s ↓1,
Selectable PDF notation • transcription unchecked
π ( )
E(τ, s) −_{s −1} −→2π(γ − log 2) −π log y|η(τ)|^{4} . (C.57)
Proof. Source: Development/AbelianVarietyTheory/Proofs/HomogeneousIdeals/SummableAndAbsTsumSubIntegralLe.lean; TendstoRealEisensteinSubPole.lean:9--374.
The source subtracts the integral from each horizontal row. For Y > 0 set
Selectable PDF notation • transcription unchecked
∑ √
R(a, Y ; s) = ((a + n)^{2}+ Y^{2})^{−s}, B(s) =^{π Γ(s −1/2)}.
n∈ℤ^{Γ(s)}For s ≥1 the row is summable and
Selectable PDF notation • transcription unchecked
||R(a, Y ; s) −B(s)Y^{ 1−2s}|| ≤Y^{ −2s}. (C.58)Here is the sum–integral estimate behind this assertion. For a nonnegative integrable continuous function increasing up to 0 and decreasing thereafter, compare its value at the left endpoint of each unit interval with the interval integral. On intervals entirely to the left the errors are nonpositive and telescope between the endpoint values; on the right the errors are nonnegative and telescope the other way. On the single interval crossing 0, use the upper bound f(0) and the lower bound f(a)+f(a+1)−f(0) for the integral. The running error is between −f(0) and f(0). Letting the endpoints tend to infinity proves convergence of the nonnega-tive lattice sum and an error of at most f(0). Apply this to f(t) = (t^{2}+ Y^{2})^{−s}. Its integral is B(s)Y^{1−2s}: insert u^{s−1}e^{−u(t2+Y 2)}, integrate the Gaussian in t, and then the gamma integral in u. Nonnegativity permits the interchange; the resulting integrals are finite for s > 1/2.
Put e_{m}(s) = R(mx, my; s) −B(s)(my)^{1−2s} for m ≥ 1. For 1 ≤ s ≤ 2, (C.58) bounds |e_{m}(s)| by (my)^{−2}+ (my)^{−4}. Dominated convergence applies both to each row and to ∑_{m≥1}e_{m}(s). Summing rows gives ∑
Selectable PDF notation • transcription unchecked
E(τ, s) = 2y^{s}ζ(2s) + 2y^{s} e_{m}(s) + 2B(s)y^{1−s}ζ(2s −1). (C.59)
m≥1The same estimates prove absolute convergence for s > 1.
At s = 1, the cotangent partial fraction expansion [2, Equation4.22.3] and its exponential expression give
Selectable PDF notation • transcription unchecked
_{ππe}2πi(a+iY )
R(a, Y ; 1) =_{Y} + 2_{Y} ℜ_{1 −e2πi(a+iY )} .Indeed take imaginary parts of the symmetrically paired series for π cot π(a + iY ); those imaginary parts are absolutely summable, since Y > 0. If q = e^{2πiτ}, then |q| < 1 and
Selectable PDF notation • transcription unchecked
∑ 1 q^{m} ||||||∏ ||||||ℜ_{1 −qm} = − log (1 −q^{j}) .
Selectable PDF notation • transcription unchecked
m≥1^{m} j≥1To justify it, expand − log(1 −q^{j}) = ∑_{m≥1}q^{jm}/m, and interchange the absolutely convergent double series. For example |q|^{jm}/m ≤|q|^{j+m−1}, whose double sum is finite. Taking real parts and exponentiating the convergent logarithm gives the stated modulus of the product. Consequently
Selectable PDF notation • transcription unchecked
∑ 2π ||||||∏ ||||||
e_{m}(1) = −_{y} log (1 −q^{j}) .
Selectable PDF notation • transcription unchecked
m≥1 j≥1
For the last term of (C.59), the classical special values [2, Equations5.4.1,5.4.6,5.4.11–5.4.13] give B(1) = π and B^{′}(1) = −2π log 2. The Laurent expansion [2, Equation25.2.4] is ζ(1+t) = t^{−1}+γ +O(t). Thus
Selectable PDF notation • transcription unchecked
π
2B(s)y^{1−s}ζ(2s −1) −_{s −1} −→2π(γ − log 2) −π log y.
The first term of (C.59) tends to π^{2}y/3. Finally log |η(τ)|^{2}= −πy/6 + 2 log |∏_{j≥1}(1 −q^{j})|. Substitution cancels π^{2}y/3 and gives (C.57). All these limits are for the same fixed τ; only a finite sum of them over ideal classes will be taken below.
Proposition 9.57 (The elliptic period and the quadratic logarithmic derivative). Suppose k has discriminant −d with d > 4, and let χ be the quadratic Dirichlet character attached to that discriminant, in the source’s Kronecker convention. Under the scheme and multiplier hypotheses of Lemma 9.55,
Selectable PDF notation • transcription unchecked
^{√} L^{′}(1, χ)− mean_{σ} log n_{σ} = γ + log(2π/ d) −ℜ . (C.60)
Selectable PDF notation • transcription unchecked
L(1, χ) √
In particular the denominator here is positive and nonzero: L(1, χ) = πh/ d.
Proof. Source: Development/AbelianVarietyTheory/Proofs/HomogeneousIdeals/ExistsBasisEmbedding.lean:115--246; ExistsIsKroneckerCharOfDiscr.lean:507--524; DedekindZetaEqSumRealEisenstein.lean:14--59,83--154,222--243.
For a fractional ideal 𝔞 represented by ω(ℤτ + ℤ), the discriminant determinant formula gives
Selectable PDF notation • transcription unchecked
_{N𝔞 = }2|ω_{√}|^{2}ℑτ_{.}
dThis is first the index formula for integral ideals in the integral basis of discriminant −d, and then follows for fractional ideals by multiplying by a nonzero integer. Map 0 ̸= a ∈𝔞 to the integral ideal (a)𝔞^{−1}. Its norm is |ιa|^{2}/N𝔞. Each ideal in the inverse class occurs twice, because the only units are 1, −1 when d > 4. One can verify the latter directly: an imaginary quadratic unit has norm 1, integral trace t with |t| ≤2, and characteristic polynomial X^{2}−tX + 1. The exceptional traces 0, ±1 would give discriminant −4 or −3; the remaining possibilities give ±1. Taking sums, and letting C run over all classes, therefore gives for real
Selectable PDF notation • transcription unchecked
s > 1_{√s}∑
ζ_{k}(s) = ^{(2/d)} E(τ_{C}, s). (C.61)
^{2} CAbsolute convergence, already proved for each Eisenstein sum, justifies regrouping by the ideals.
The quadratic factorization is ζ_{k} = ζL(·, χ). Here the normalization can be checked at every rational prime: in the split, inert and ramified cases, putting t = p^{−s}, the local factors are respectively (1 −t)^{−2}, (1−t^{2})^{−1} and (1−t)^{−1}. They equal (1−t)^{−1}(1−χ(p)t)^{−1} for χ(p) = 1, −1, 0. The Kronecker convention gives precisely these values. At an odd unramified prime this follows by counting roots of the quadratic integral-basis polynomial; at 2 it is the same check with the discriminant modulo 8. The ramified case gives 0 on both sides. Multiplying convergent Euler products proves the factorization in the region used in (C.61)._{√}
Put c = 2/ d and A =∑_{C} log(ℑτ_{C}|η(τ_{C})|^{4}). Lemma 9.56 gives
Selectable PDF notation • transcription unchecked
∑ hπ
E(τ_{C}, s) =_{s −1 }+ 2hπ(γ − log 2) −πA + o(1).
Selectable PDF notation • transcription unchecked
C
The character is odd and hence nonprincipal; its Dirichlet L-function is holomorphic at 1. This also follows directly by partial summation of the bounded periodic partial sums of χ, uniformly on compact subsets of ℜs > 0. Expand c^{s}= c(1+(s−1) log c+o(s−1)) in (C.61), and compare first residues and then constant terms with
Selectable PDF notation • transcription unchecked
ζ(s)L(s, χ) =^{L}_{s}^{(1}_{ −}^{, χ}_{1 }^{)}+ γL(1, χ) + L^{′}(1, χ) + o(1).This yields L(1, χ) = chπ/2 and
Selectable PDF notation • transcription unchecked
A^{√} L^{′}(1, χ)_{h} = γ − log(2 d) − .
Selectable PDF notation • transcription unchecked
L(1, χ)
In particular the right side is real. Add log(4π) and use (C.56). The result is (C.60), with the sign and constant required by the source, rather than a period formula with an unspecified normalization.
Lemma 9.58 (Algebraizing the linear maps in the pure CM family). Let A = V /Λ, B = W/Γ be complex abelian varieties with symmetric ample line bundles and the source’s theta uniformizations. Suppose F : V → W is complex linear, FΛ ⊂ Γ, and F^{∗}H_{B} = nH_{A} for an integer n ≥ 0. Then the induced map A(ℂ) → B(ℂ) is an algebraic group homomorphism. Here the theta uniformization hypothesis includes the correspondence between algebraic sections of every line-bundle power and its theta functions.
Proof. Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/ExistsHomOfLinearScalingHermForm.lean:263--384.
Choose one power N of the target bundle whose sections s_{i} give affine nonvanishing charts with coordi-nate rings generated by ratios s_{j}/s_{i}. Let θ_{i} be their theta functions. A symmetric bundle has semicharacter with square 1: inversion changes the semicharacter to its inverse, whereas symmetry preserves it. The prod-uct θ_{i}(Fz)θ_{j}(Fz) consequently has the transformation law for the source bundle to the power 2nN; the two target semicharacter factors cancel. By the specified theta-section correspondence it is an algebraic section t_{ij}. In local trivializations the same nonzero frame factor occurs for every j, so on t_{ii} ̸= 0,
Selectable PDF notation • transcription unchecked
t_{ij(z) =} θ_{j}(Fz)_{.}
t_{ii} θ_{i}(Fz)These ratios pull back every generator of the target affine coordinate ring to a regular function. They satisfy all its relations: evaluate on every complex point, use the displayed identity, and use that the source is reduced. They therefore define morphisms on those open sets. They agree on overlaps by the same equality of values and separatedness of the target. The open sets cover, because at every Fz some s_{i} is nonzero; complex closed points detect an omitted closed subset on this variety. This gives an algebraic morphism inducing z →→Fz on the universal covers. The same reasoning on A×A, which is smooth and reduced, proves preservation of addition; F0 = 0 proves preservation of the identity. Thus this is precisely the source’s construction through products of theta sections and their ratios.
Lemma 9.59 (Descent and integral extension of the full CM family). For polarized abelian varieties defined over a number field k ⊂ℂ, every algebraic complex homomorphism is defined over a finite extension of k. A family additive in a finitely generated abelian group can be descended over one finite extension. If source and target have abelian scheme models over that extension’s integers, all these maps extend uniquely over the integers, preserving their action and pairing identities.
Proof. Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/PolCountableAndTangentExt.lean:20--64,118--149,354--407; CmFamilyDescendsToNumberField.lean:84--114; CmFamilyExtendsOverDedekindDomain.lean:13--22.
First complex homomorphisms form a countable set, and their differentials determine them. A homo-morphism lifts to a complex linear map of universal covers preserving period lattices. Its images on a finite integral basis of the source period lattice lie in the countable target lattice. These images determine the linear map because the lattice spans the real tangent space; surjectivity of the exponential determines the homomorphism. The same lift shows that equal differentials give equal homomorphisms.
Spread the given morphism to a finitely generated k-subalgebra R ⊂ℂ, along with the equations saying it preserves the identity and addition. Such a spread uses only finitely many affine charts, generators, relations and gluing equations; enlarge R to contain their coefficients. The group identities hold after inclusion into ℂ, hence already over R by flatness of the source and separatedness of the target. For every k-algebra embedding R → ℂ the specialization is therefore one of the countably many complex homomorphisms. Fix a tangent vector and an affine target chart containing the identity. Its image as a dual-number point lies entirely in that chart, since its reduced point is the identity. Write the coordinate functions in a k-basis of the dual numbers, after tensoring with R. Each coefficient has only countably many specialized values.
Lemma 9.26 forces every such coefficient to be algebraic over k. Choose a number-field specialization of R fixing its algebraic elements, as in that lemma. The specialized map has the original differential on every tangent vector. Differential uniqueness just proved makes its complex base change the original map. This yields a finite field of definition, with the specified embedding into ℂ.
For an additive family choose finitely many additive generators and their negatives, and take the com-positum of the finitely many fields of definition. Sums then descend every member. Every identity of homomorphisms and of rigidified pairing bundles can be tested after the faithfully flat extension to ℂ, so the same identities hold over that field. Over each discrete valuation ring of its integers extend the maps by
Proposition 10.31; these local extensions agree on the generic fibre and are unique, hence glue. For identities between the extended rigidified pairing bundles, use the source’s uniqueness of primitive normalization: the generic identity is normalized at the zero sections, and the extension law removes the possible bundle from the base. This is exactly the family-identity extension input used here; the primitive-normalization law was used in Lemma 9.38 as well. It is not an assertion that arbitrary line bundles are determined by a generic fibre without such normalization.
Proposition 9.60 (The same-type pure CM tensor decomposition). Let E be the source CM field, with quadratic generator w and r_{t} = 2w −1, r_{t}^{2} = −d. Let C be its principally polarized CM elliptic curve of character w →→ ω. Let T be a principally polarized abelian scheme with 𝒪_{E}-action over the integers of k ⊂ ℂ. Assume its complex Lie type is pure of multiplicity r: only the embeddings with r_{t} →→ 2ω −1 occur, each with the specified multiplicity. Then, after one finite extension k_{1}/k, there are a finite projective 𝒪_{E}-lattice M, an integral perfect alternating form B with the source adjoint action, ε ∈{1, −1}, and its quadratic Hermitian form h as in Lemma 9.33, such that
Selectable PDF notation • transcription unchecked
εB(wa, a) > 0 (a ̸= 0), T_{𝒪k1} ≃M ⊗_{𝒪k0} C_{𝒪k1}as the polarized objects with action and pairing used by the source. Here k_{0} = ℚ(r_{t}), 𝒪_{k0} = ℤ[w] is the quadratic order denoted OK H; the tensor uses this quadratic action of C, while M retains its full 𝒪_{E}-action. The same finite extension works for the entire family M → Hom(C, T), not separately for individual vectors of M.
Proof. Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/ExistsHomFamilyOfPureTypeComplexTriple.lean:25--259; ExistsTensorDecompositionPureTypeCmTriple.lean:51--120; NonemptyTensorOfOntoFullFamily.lean:28--47.
We spell out first the complex construction. Uniformize T by V /Λ. The action lifts to complex linear operators preserving Λ. Since E ⊗_{ℚ} ℂ is a product of copies of ℂ, the lifted action is the direct sum of its embedding eigenspaces. Choose a coordinate change F putting it in that form, with precisely the multiplic-ities in the given Lie type. Carry the period lattice and its unimodular Riemann form through F^{−1}. This gives M = Λ with period map P in the diagonal coordinates and alternating form B. It is finitely generated over ℤ, hence over 𝒪_{E}, and is torsion-free: every nonzero a ∈𝒪_{E} acts by nonzero scalars on every embed-ding coordinate. Over the Dedekind domain 𝒪_{E}, finite torsion-free modules are projective. Real ranks give rank_{𝒪E} M = r. The positive complex structure on ℝ ⊗M is F^{−1}iF. It squares to −1, preserves B, and B(Jx, x) > 0; its action traces are the sums of the specified embedding characters. These are exactly the period-datum conditions, including the source’s type convention.
Put ρ = 2ω −1. Purity says all nonzero coordinate characters restrict to this same root. Thus P(r_{t}a) = ρP(a) for every a ∈M. If the elliptic period lattice is c(ℤ + ℤω), the map of universal covers
Selectable PDF notation • transcription unchecked
G_{a}(cz) = F(zP(a))sends its periods into Λ: the two generators map to FP(a) and FP(wa). Comparison of the integral Rie-mann forms on this rank-two sublattice, followed by the one-dimensional calculation in Lemma 9.55, gives
G_{a}^{∗}H_{T} = |B(wa, a)|H_{C}. Choose the same sign ε with εℑω > 0. Positivity of H_{T} then gives εB(wa, a) > 0 for every nonzero a, so the absolute value is εB(wa, a), an integer. For a = 0 both sides vanish. Lemma 9.58 gives algebraic maps ψ_{a} : C →T with these lifts. Uniqueness on the universal covers gives ψ_{a+b} = ψ_{a}+ψ_{b} and ψ_{sa} = sψ_{a} for every s ∈𝒪_{E}.
The diagonal norm identity supplies the full pairing, not only its diagonal. Subtract the two diagonal identities from the one for a + b to recover the symmetric cross term; do the same with a + r_{t}b to recover the other cross term. Using ̄r_{t} = −r_{t} and r_{t} ̸= 0 gives
Selectable PDF notation • transcription unchecked
H_{T} (G_{a}(c), G_{b}(c)) =(B(wa, b) −B(a, b) + B(a, b)ω)εH_{C}(c, c).This is exactly evaluation of h(a, b) = ε(B(wa, b) −B(a, b)) + εB(a, b)w, with the same conjugate-slot convention as Lemma 9.33. The source’s Weil/Riemann-form dictionary and primitive normalization iden-tify the corresponding pairing bundles. This proves the full family pairing law, including its scalar, rather than choosing a separate polarization for each map. ∑ The vectors FP(a) span V over ℂ, since P(M) already spans its real tangent space. Given v =
_{j}z_{j}FP(a_{j}), surjectivity of the exponential writes exp_{T}(v) = ∑_{j}ψ_{aj}(exp_{C}(cz_{j})). Thus the family is jointly onto on the chosen complex fibre. Descend the full family over one k_{1} by Lemma 9.59, and extend it over 𝒪_{k1}. The same complex joint-surjectivity is preserved, since the pullback fibre identifications preserve the maps and their finite sums.
Construct the principally polarized tensor scheme A = M ⊗C by Lemmas 9.37–9.40. The tensor universal property and full family pairing give Φ : A →T with Φ(a ⊗x) = ψ_{a}(x) and Φ^{∗}λ_{T} = λ_{A}, by
Lemma 9.35 with scalar 1. Joint-surjectivity gives surjectivity on complex points. The proper-image and smooth-density argument of Lemma 9.28 makes Φ surjective over the base. Its kernel is trivial on every test scheme: translating the pulled-back pairing by x ∈ ker Φ makes its character trivial, and the tensor’s principal-pairing condition forces x = 0. Thus Φ is a monomorphism. It is proper, hence a closed immersion. Its target is reduced, being smooth over the reduced ring 𝒪_{k1}; a surjective closed immersion into a reduced scheme is an isomorphism, since its defining ideal has zero radical and is therefore zero. Transporting the tensor structure through this isomorphism proves the assertion with all original actions and pairings.
Corollary 9.61 (The pure CM eigenline degree). Assume the realized CM elliptic curve and theta uniformiza-tion inputs used above. For the rank-one pure CM object in the source degree calculation, the normalized degree of its integral Lie eigenline is ℜΩ/2, where
Selectable PDF notation • transcription unchecked
2π L^{′}(1, χ)Ω= γ + log − .
Selectable PDF notation • transcription unchecked
rd(E) L(1, χ)
Proof. Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/CmEigenLineDegreeFormula.lean:44--126.
Choose a complex embedding of the original coefficient field. Take the compositum with the field of the elliptic curve whose root 2ω −1 equals the image of the original root ρ. These are the same root, not merely roots of the same quadratic polynomial. Pull back both objects to that compositum, then apply
Proposition 9.60 with r = 1. The prescribed eigencharacter sends w to ω because 2w = 1 + r_{t} and 2 ̸= 0; this verifies the character used in the integral eigenline of Lemma 9.53. Normalized degrees and the mean log of the elliptic Lie norm are unchanged by the finite field extensions, as proved there. Substitute (C.60) into (C.51). The discriminant terms cancel:
Selectable PDF notation • transcription unchecked
√
log(2π/ d) +_{2}^{1} log d − log rd(E) = log(2π/ rd(E)).
This gives 2 deg Lie_{τ} = ℜΩ over the extension, hence over the original field. It supplies precisely the Ω/2 used in (C.47).
Lemma 9.62 (Integral discriminant-one Weierstrass models). An elliptic Weierstrass curve W over a number field with integral j acquires, over a finite extension, a global Weierstrass equation with integral coefficients and discriminant exactly 1. The source obtains this by adjoining the coordinates of all 12-torsion points and using the two normal forms below.
Proof. Source: Development/AbelianVarietyTheory/Proofs/HomogeneousIdeals/HasRationalTorsionIffMemRange.lean:185--369; ExistsIntegralModelOfForms.lean:77--169; IsDiscreteValuationRingValuationSubringAtPrime.lean:94--129.
The 12-torsion is finite: the 2-torsion coordinates satisfy the nonzero division polynomial Ψ^{2}_{2}, the 3-torsion coordinates satisfy Ψ_{3}, and each x-coordinate has at most two points above it. For torsion of a product of integers, multiply by one integer; its fibres are translates of that integer’s finite torsion subgroup. Adjoining the finitely many coordinates therefore gives a finite extension K on which all 12-torsion is ratio-nal.
Rational 2-torsion permits a change with scale u = 1 to
Selectable PDF notation • transcription unchecked
y^{2}= x(x −P)(x −Q).Complete the square and translate one root of the cubic to zero; all three roots are rational because they are the x-coordinates of the 2-torsion. Rational 4-torsion implies that −P, −Q, P, P −Q are squares. To see the first assertion, choose square roots α^{2}= −P, β^{2}= −Q over an algebraic closure. The point (αβ, αβ(α + β)) doubles to (0, 0), as substitution in the tangent-line addition formula verifies. It is 4-torsion, hence has rational coordinates p, q. Since P, Q, P −Q are nonzero, so are p, q, and (p −P)p/q = α. Interchange P, Q, and then translate x by P, to obtain the other three squares. In particular c^{2}= PQ(P −Q) for some c ∈K, and ∆= 16P^{2}Q^{2}(P −Q)^{2}= (2c)^{4}.
Rational 3-torsion gives a flex point, whose tangent can be made y = 0 by another scale-one change. The resulting equation is
Selectable PDF notation • transcription unchecked
y^{2}+ axy + by = x^{3}.Its remaining 3-division factor is 3z^{3}+a^{2}z^{2}+3abz +3b^{2}. Choose a root z; it is the coordinate of a 3-torsion point and hence lies in K. It is nonzero since b ̸= 0. Direct substitution shows (a + 3b/z)^{3} = a^{3}−27b. Thus ∆= b^{3}(a^{3}−27b) = (bd)^{3} with d = a+3b/z. Combining the fourth and third roots gives δ = bd/(2c) with δ^{12}= ∆. Scale the original curve by u = δ to make its discriminant 1. Torsion rationality is preserved.
It remains to make all coefficients integral simultaneously; this step is essential at primes above 2 and 3. Let S = 𝒪_{K}. For this normalized curve, c^{3}_{4}= j and c^{2}_{6}= c_{4}^{3}−1728, so c_{4}, c_{6} ∈S by integral closedness. Write its split and Deuring forms as above, both with scale 1. Put v = 16PQ and m = 4(P + Q). Their invariant equations give
Selectable PDF notation • transcription unchecked
v^{3}−c_{4}v^{2}+ 256 = 0, m^{2}= c_{4} + 3v.Thus v, m ∈S. The Deuring invariants similarly give
Selectable PDF notation • transcription unchecked
(a^{2})^{4}−6c_{4}(a^{2})^{2}−8c_{6}a^{2}−3c^{2}_{4}= 0, (27b)^{4}−a^{3}(27b)^{3}+ 19683 = 0.Consequently a^{2}, a, 27b ∈S.
Let [1, r, s, t] be the change from the split to the Deuring equation. Comparing its first three coefficients gives
Selectable PDF notation • transcription unchecked
2s = a, 12r = a^{2}+ m, 54(t −4rs) = 27b −108ra.Apply [1, −4r, 0, −28(t −4rs)] to the Deuring form, obtaining W^{′}. All its parameters and coefficients lie in S[1/3]. The same equation is obtained by applying [1, −3r, s, −27(t −4rs)] to the split form; all these parameters and coefficients lie in S[1/2], by the same three identities and m, v ∈S. But S[1/2] ∩S[1/3] =
S inside K: if 2^{u}x, 3^{v}x ∈ S, Bezout for the coprime integers 2^{u}, 3^{v} expresses x as their integral linear combination. Hence every coefficient of W^{′} is integral. Its discriminant is still 1, since every change in this last step has scale 1.
Proposition 9.63 (The unit-discriminant CM elliptic family). For the imaginary quadratic field k and its integral ideal representatives 𝔟_{C}, there is one number field M ⊂ℂ containing the chosen image of k, and principally polarized elliptic schemes P, Q_{C} over 𝒪_{M} with global Lie generators v, t_{C}, such that their period lattices at the chosen complex embedding are
Selectable PDF notation • transcription unchecked
Λ_{0} = c_{0}ι𝒪_{k}, Λ_{C} = c_{C}ι𝔟^{−}_{C}^{1}, c_{0}, c_{C} ̸= 0.Their discriminants in those Lie generators are units. After one further finite extension all endomorphisms and both multiplier families in Lemma 9.54 are defined over the same field, with β_{C} = c_{C}/c_{0} belonging to it. The conclusion uses the theta uniformization and normalized pairing laws already specified in this section.
Proof. Source: Development/AbelianVarietyTheory/Proofs/CubicalLineBundles/ExistsFieldCmCurvesUnitDiscriminant.lean:100--238; ExistsCmCurvesOfWeierstrassCurves.lean:77--169; ExistsNumberFieldDefiningCmCurveHoms.lean:52--122,187--279.
The invariant of a lattice is j(Λ) = 1728g_{2}(Λ)^{3}/∆(Λ). For an 𝒪_{k}-stable rank-two lattice it is an alge-braic integer. The precise established input is [1, Corollary12.16, p.125]. To check its hypothesis, express a noninteger element of 𝒪_{k} on the two period generators. Eliminating its two integer linear relations gives a quadratic equation for their ratio τ ∈ℍ; its leading coefficient is nonzero, since otherwise the endomorphism was an integer. Thus ℚ(τ) is imaginary quadratic. The source proves this same integrality via its modular-polynomial argument; the cited corollary supplies exactly that established input, with the conventional 1728 factor.
The possible j(Λ) for this fixed k form a finite set. The ideal classification in the proof of Lemma 9.54 associates an ideal class to each lattice, and lattices in the same class are homothetic, hence have the same j. Choose for each of these finitely many j a Weierstrass equation over ℚ(j) with that invariant. Such an equation exists also for j = 0, 1728, treated separately by y^{2}= x^{3}+1 and y^{2}= x^{3}−x; otherwise the usual coefficients proportional to j/(1728 −j) give one by substitution in c^{3}_{4}/∆. Apply Lemma 9.62 to each, and embed their finite fields in ℂ. Taking the compositum with ιk gives one number field containing all their integral Weierstrass coefficients. Thus all the models have discriminant 1 over this one ring of integers.
The Weierstrass-to-period construction in the source gives their principal polarizations and Lie genera-tors. In the projective cubic the identity is [0 : 1 : 0] and the formal parameter is −X/Y ; the dual-number point [ϵ : −1 : 0] is its tangent generator and remains one after every base change. In a complex uniformiza-tion with that tangent generator, the affine coordinates have principal parts z^{−2} and −z^{−3}. Subtracting the corresponding Weierstrass functions leaves entire periodic functions, hence constants. Substitution in the cubic gives c_{4} = 12g_{2}, c_{6} = 216g_{3}, and ∆= g_{2}^{3}−27g_{3}^{2}, precisely as in ExistsPolarisedAbelianSchemeLieBasisInvariants.lean:43--57. These identities determine the period lattice; equality of g_{2}, g_{3} gives equality of the lattices by the Weierstrass uniformization with its fixed differential. Equality of j gives homothety, since the scale of the differential may then change. Apply this to the classes 𝒪_{k} and 𝔟^{−}_{C}^{1} to obtain the two displayed lattices.
For each pair of these elliptic curves, its group of complex period multipliers is finitely generated over ℤ. Indeed multiplication by a fixed nonzero period embeds it in the target rank-two lattice, and subgroups of that lattice are finitely generated. Each multiplier is an algebraic homomorphism by the theta algebraization argument, and Lemma 9.59 descends a finite set of generators. Take the compositum for P →P, P →Q_{C} and Q_{C} → P, for every one of the finitely many C. It defines all these multiplier groups at once. Now β_{C} = c_{C}/c_{0} is itself a multiplier P →Q_{C}, because 𝒪_{k} ⊂𝔟^{−}_{C}^{1} for an integral ideal. Its derivative in the two chosen generators therefore belongs to the common number field and is nonzero. For every x ∈𝔟^{−}_{C}^{1}
and y ∈𝔟_{C}, the multipliers β_{C}x and β_{C}^{−1}y send the corresponding lattices into each other. They lie in the groups just descended, so all the required homomorphisms exist over that same field. The Lie generators, unit discriminants and lattice identities pull back unchanged. This proves the family and both universal multiplier conditions used in (C.53)–(C.60).
Lemma 9.64 (Patching key invariants across the denominator primes). Fix dimension g ≥ 1. Suppose that for each rational prime p the algebraic key-invariant construction supplies integers n_{p}, M_{p} ≥1, with p ∤M_{p}, and a field-valued unit κ_{p}(x), natural under extension of characteristic-zero fields, with the following properties. Its scaling weight is (det U)^{4np}/a^{8np} under change of Lie basis U and theta generator a. It is integral and a unit over any discrete valuation ring where M_{p} is a unit, with the stated base-change/reduction compatibility. Finally, over ℂ, in every half-characteristic theta normal form x = (τ, U, c),
κ_{p}(x) = (det U^{4}/c^{8})^{np}.
Assume key data descend to countable characteristic-zero fields. Then one can choose a single exponent N_{g} ≥1 and a single invariant κ_{g} of that weight, integral and a unit at every discrete valuation ring, with the same normal-form value. The exponent depends only on g.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ThetaKeyRuleExistsOfAlgebraicInvariants.lean:29--44,55--78,80--133.
Any countable characteristic-zero field embeds in ℂ: choose a transcendence basis over ℚ, embed it into a transcendence basis of ℂ, and extend to its algebraic closure. Descend x to such a field and then embed in ℂ. At a complex theta normal form, the two stated formulas give κ_{p}(x)^{nq} = κ_{q}(x)^{np}. Naturality and injectivity of field embeddings descend the equality first to that countable field, then to the original field. Hence this equality holds for all original x.
Choose N_{g} = n_{2} and κ_{g} = κ_{2}. Scaling, field naturality and the normal-form value already hold. Over a discrete valuation ring R with fraction field K, if M_{2} is a unit there is nothing to prove. Otherwise its nonunit image lies in the maximal ideal, so the residue characteristic is some rational prime p | M_{2}. For this p, M_{p} is a unit in R. Thus κ_{p}(x) ∈R^{×}, and
Selectable PDF notation • transcription unchecked
κ_{g}(x)^{np}= κ_{p}(x)^{Ng} ∈R^{×}.Taking the discrete valuation gives n_{p}v(κ_{g}) = 0; since n_{p} ≥ 1, v(κ_{g}) = 0. This proves that κ_{g} is a unit. Equivalently integral closedness puts it in R, and applying the same argument to its inverse makes it a unit. At primes already good for M_{2} use its original reduction law. At the bad primes the source asks compatibility only with characteristic-zero field specializations of that local ring; none exist as residue field maps in positive residue characteristic. Indeed a surjective map from a local ring to a characteristic-zero field makes every positive integer a unit. This verifies that clause too, without silently asserting a new positive-characteristic invariant. Since the initial choice n_{2} was made before R, K, x, its uniformity is exactly the required dimension-only uniformity.
Lemma 9.65 (The archimedean key-formula constant). In the source’s principally polarized theta normal form, write τ = X + iY with Y > 0, and θ = cθ[a, b] for real half-characteristics a, b. Let n_{θ} be the average squared theta norm over a period fundamental domain, and n_{η} the squared determinant Lie norm in its dual convention. If the Lie basis is represented in these coordinates by the rows of (2πi)^{−1}U, then
Selectable PDF notation • transcription unchecked
|||| ||||
|c|^{2}_{nη = }(4π)^{g} det Y det U^{4}
Selectable PDF notation • transcription unchecked
n_{θ} = √ , , n^{4}_{θ}n^{2}_{η}= (2π)^{2g}. (C.62)| det U|^{2} c^{8} det(2Y )
Selectable PDF notation • transcription unchecked
2N
Consequently a patched invariant of exponent N_{g} satisfies |κ_{g}|n^{4}_{θ}^{Ng}n_{η}^{g} = (2π)^{2gNg}, the archimedean key power formula used in (C.46).
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/KeyInNormalFormAndArchimedeanConstant.lean:16--31; ValueAtNormalForm.lean:47--91; KeyFormulaPowOfKeyRuleCharZero.lean:15--30.
The normal-form period lattice is ℤ^{g}+ τℤ^{g}. Its fundamental parallelepiped has volume det Y , by the real block determinant. Write its points as z = u + τv, with u, v ∈ [0, 1]^{g}. The theta metric contributes e^{−2π(ℑz)tY −1(ℑz)}. Expand the two theta series in the squared modulus and integrate first over u. Distinct indices have a nonzero integral-frequency difference, so their u-integrals vanish; diagonal terms remain. Completion of the square gives
Selectable PDF notation • transcription unchecked
∫ ∫ |θ[a, b](u + τv)|^{2}e^{−2πvtY v} du dv = ∑ ∫ e^{−2π(m+a+v)tY (m+a+v)} dv.
^{[0,1]g[0,1]g} m∈ℤ^{g[0,1]g}
∫ t √The shifted unit cubes tile ℝ^{g} up to null boundaries, so the right side is _{ℝg} e^{−2πtY t} dt = 1/ det(2Y ). The exchanges are justified by normal convergence of the theta series on compact z-sets, or by their Gaussian bounds and then dominated convergence. Multiplication by |c|^{2} gives n_{θ}; averaging divides out the common parallelepiped volume det Y .
By the source’s definition of the determinant Lie norm, if W = (2πi)^{−1}U it is det Y /(π^{g}| det W|^{2}). Since | det W|^{2} = (2π)^{−2g}| det U|^{2}, this is the displayed n_{η}. Now det(2Y ) = 2^{g} det Y ; substituting and cancelling |c|^{8}, | det U|^{4} and (det Y )^{2} proves (C.62).
To check that the field invariant really has this normal value for arbitrary U, c, replace the theta generator by c^{−1}θ and the Lie basis by U^{−1} times the old basis. Both are valid bases because c ̸= 0 and det U ̸= 0. The invariant of the resulting normal form is 1 by its normalization. Its scaling law therefore gives κ_{g} = (det U^{4}/c^{8})^{Ng} for the original data. Raise (C.62) to N_{g}. This proves the exact archimedean constant; it does not supply the algebraic invariants assumed in Lemma 9.64.
Lemma 9.66 (The key construction’s numerical data avoid any chosen prime). For every g ≥1 and rational prime p there are integers n, ℓ, ℓ^{′} and integer matrices A_{m} ∈M_{4}(ℤ) such that
Selectable PDF notation • transcription unchecked
∏g
n ≥4, gn < ℓ< ℓ^{′}, A^{t}_{m}A_{m}= mI_{4}, p ∤P := (ℓ−jn)(ℓ^{′}−jn).
j=0Moreover P ≥1 and c = 2P^{3g+2}g!n^{g}(ℓ^{′}−ℓ) ≥1.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ExistsAdmissibleKeyNumericalDataPrimeTo.lean:18--55.
Use n = 4p, ℓ = gn + 1 and ℓ^{′} = gn + 1 + p. Each factor of P is positive and congruent to 1 modulo p, so P is positive and prime to p. Lagrange’s four-square theorem [2, Section27.13(iii)] writes m = a^{2}+ b^{2}+ c^{2}_{0}+ d_{0}^{2}. Its quaternion multiplication matrix is
Selectable PDF notation • transcription unchecked
(a −b −c_{0} −d_{0})
((b a −d_{0} c_{0} ))
^{Am =}(c_{0} d_{0} a −b)^{.}
d_{0} −c_{0} b aMultiplying columns gives A^{t}_{m}A_{m}= mI_{4}. All factors in the displayed formula for c are positive, including ℓ^{′}−ℓ = p. Thus the same P, c are chosen before the field, abelian variety or sections. The four-square theorem is used only for these explicit integral matrices; it introduces no restriction on the later residue characteristic.
Lemma 9.67 (The split transverse Koszul determinant). Let A/R be one of the source’s principally polarized abelian schemes of dimension g, where R is a local Noetherian ring. Let s_{1}, . . . , s_{g} ∈ H^{0}(A, L^{n}) have common zero scheme consisting of distinct sections z_{a} and unit Jacobian determinants in the specified Lie basis. Suppose positive powers of L have no higher cohomology on the geometric fibres, their section modules commute with base change, and H^{0}(A, L^{m}) is free of rank m^{g}. For ℓ> gn choose bases of all H^{0}(A, L^{ℓ−jn}), 0 ≤j ≤g. Then the augmented global-section Koszul complex is split acyclic, the number of zero sections is d = g!n^{g}, and the square of its torsion in these bases is a unit of R.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ExistsKoszulTorsionTransversalSplitZeros.lean:28--70,116--167,347--396; ExistsIsUnitField.lean:145--151.
At a common zero the unit Jacobian says the local equations s_{i} form a basis of the cotangent space. In the regular local ring of a geometric abelian fibre, Nakayama makes them generators of its maximal ideal; they form a regular system of parameters and their Koszul complex resolves the residue field. Off the common zeros some s_{i} is a unit, and contraction by that coordinate makes the Koszul complex exact. Thus on every geometric fibre its sheaf complex, augmented by the split reduced zero scheme, is exact. Twist by L^{ℓ}. All its nonzero terms are positive powers because ℓ> gn. The assumed higher-cohomology vanishing makes the global-section complex exact, including surjectivity onto values at the zero points and the statement that its kernel is generated by the s_{i}.
To see the matrix directly, its term in Koszul degree j is indexed by j-element subsets I ⊂{1, . . . , g} and a basis vector of H^{0}(L^{ℓ−jn}). For q = I ∪{i}, multiplication by s_{i} gives the matrix block with sign (−1)^{#{j∈I:j<i}}. The augmentation is the evaluation matrix at z_{a} in the chosen local bundle generators. Mul-tiplications commute, so each pair of insertions cancels with the opposite order; evaluation kills each s_{i}. Hence the full differential D satisfies D^{2}= 0 over R.
Exactness of a matrix complex descends from an algebraic closure to the residue field: choose a linear retraction of that field extension onto the residue field and apply it coordinatewise to a preimage of every cycle. A bounded finite free complex exact modulo the maximal ideal is split acyclic over R. One direct elimination proof chooses a unit entry of a nonzero differential, uses invertible row and column operations to isolate a two-term identity complex, and repeats on its complement. If a nonzero complement had every differential zero modulo the maximal ideal, residual exactness would force all its terms to be zero. Thus the process exhausts it. It gives a degree-one contraction h with Dh + hD = 1 and h^{2} = 0. Consequently D + h is an isomorphism between the even and odd direct sums of the complex; their free ranks are equal and its determinant is a unit. Its square removes the sign from ordering these summands and is the source’s squared Koszul torsion. Changing the chosen contraction leaves that square unchanged: in the split identity complexes the change is block unipotent; its diagonal determinants are 1.
The equality of even and odd ranks gives
Selectable PDF notation • transcription unchecked
∑g (_{g})
d = (−1)^{j} (ℓ−jn)^{g}= g!n^{g}.
j=0^{j}For the last equality expand the polynomial in j. The gth finite difference annihilates all powers of degree less than g and sends j^{g} to g!; the two factors (−1)^{g} cancel. This proves both the exact number of zero points and the unit determinant over the original local ring. No complex-point argument was used to infer invertibility at a residue prime.
The two Koszul determinants and the square similitude comparisons are combined by the following exact source expression. We record it because it fixes the exponents whose eventual weight is 8c and 4c. Set P as
in Lemma 9.66. For squared torsions t, t^{′} and comparison units λ_{j}, λ^{′}_{j}, define
Selectable PDF notation • transcription unchecked
_{Tℓ= t4P} 3g+2∏^{g}_{λ}(−1)^{j}(^{g}_{j})(P/(ℓ−jn))^{3g+2}_{,}
j=0^{j}and define T_{ℓ}′ in the same way. If J is the product of the Jacobian determinants at the g!n^{g} zero sections, the invariant candidate is
Selectable PDF notation • transcription unchecked
κ = T_{ℓ}^{gn−2ℓ′}T_{ℓ}^{2}_{′}^{ℓ−gn}J^{8P 3g+2(ℓ′−ℓ)}. (C.63)All quotients P/(ℓ−jn) and P/(ℓ^{′}−jn) are integers. All bases and determinants in this formula are the specified ones; (C.63) is not defined by a choice-free assertion. It is a unit wherever the squared comparison units and the transverse Koszul data have been constructed.
Source: Development/AbelianVarietyTheory/Definitions/KeyFormula/DegProd.lean:30--49; IsSquareDeterminantComparison.lean:103--108.
Lemma 9.68 (The section bases in the key construction). For a principally polarized abelian scheme f : A → Spec R over a local Noetherian ring and every m ≥1, the module H^{0}(A, L^{m}) is free of rank m^{g}, has no higher cohomology, and commutes with arbitrary base change. Thus it supplies the bases and vanishing hypotheses of Lemma 9.67.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/BasisPowerSectionsLocalBaseChange.lean.
On each geometric fibre, the principal-polarization and positive-power Riemann–Roch/index results used in Proposition 10.27 give h^{0}(L^{m}) = m^{g} and H^{i}(L^{m}) = 0 for i > 0. The precise formulas are [20, Theorems9.11,9.14 and Proposition9.18(iv), pp.131–135]. They hold in every characteristic, so apply to every geometric fibre here. The scheme morphism f is proper, smooth and of finite presentation; the in-vertible sheaf L^{m} is finitely presented and flat over R. By [14, Tag0B91, Lemma36.30.4], Rf_{∗}L^{m} is perfect and commutes with arbitrary derived base change. Represent it over the local ring by a bounded complex of finite free modules. Cancel every unit differential entry as a two-term identity summand. The resulting complex has all differentials in the maximal ideal. Its tensor product with the residue field therefore has zero differentials, but its cohomology is concentrated in degree zero with dimension m^{g}, by the geometric-fibre calculation and faithful descent from an algebraic closure. The minimal complex consequently has only one nonzero term, R^{mg}, in degree zero. This proves both the vanishing and freeness over R. The arbitrary derived base-change assertion now becomes ordinary base change of this one free module. In the source’s local-function model, restriction to the trivializing charts identifies this module with its section coordinates, as in the section dictionary in Proposition 10.27.
Lemma 9.69 (The finite kernel used by the square comparison). Let A have dimension g, let m be invertible on the base, and let U ∈M_{4}(ℤ) satisfy U^{t}U = mI_{4}. Over an algebraically closed field, ker U^{t}: A^{4}→A^{4} has m^{4g} points and is killed by m. If the P-torsion is split into disjoint sections over a local domain, with m | P and P invertible, these kernel points lift uniquely to disjoint kernel sections.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ExistsInvariantOfSplitZerosTorsion.lean:25--135.
Taking determinants gives (det U)^{2} = m^{4}, so det U ̸= 0. Multiplying U^{t}U = mI by the adjugate also gives UU^{t} = mI. This kills every member of ker U^{t} by m. Smith reduction by unimodular integer matrices identifies the kernel with the product of the kernels of the positive diagonal integers d_{1}, . . . , d_{4}. Each d_{i} divides m, because the cokernel of U^{t} is killed by m, and ∏_{i}d_{i} = | det U| = m^{2}. Therefore
Proposition 10.27 gives ∏_{i}d^{2}_{i}^{g}= m^{4g} points.
For the lifting assertion, evaluate the split P-torsion sections on a geometric closed fibre. Their values are distinct and their number is P^{2g}, so they give all P-torsion points there. This reduction is also injective on all global P-torsion sections: two such sections of the finite etale group scheme that agree on that fibre agree on an open and closed subset of the connected local base, hence everywhere. Lift each coordinate of every residue-kernel point through this enumeration. The U^{t}-relations of their lifts are P-torsion sections reducing to zero, so injectivity makes them zero over the base. Distinct kernel points give sections separated in some coordinate on every fibre, by the same finite etale equalizer argument.
Lemma 9.70 (Square similitude determinant units). Under the abelian-scheme and normalized cube laws of the source, over a field or a discrete valuation ring where m is a unit, the split kernel of Lemma 9.69 supplies the square determinant comparison in (C.63). In particular its comparison scalar λ is a unit, including at finite residue primes.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ExistsL2ValFieldOrDvrSplitKernel.lean:39--115,117--188; EqOfSimilitudeIsoQuotientKernelInvariant.lean:61--117; ExistsNormalisingUnitSimilitudeOrigin.lean:120--155.
Put B = A^{4} and ℒ = L ⊠L ⊠L ⊠L. The matrix U defines an isogeny α : B → B. For every ξ ∈ ker U^{t} the cube and adjoint identities give a trivialization w_{ξ} of
Selectable PDF notation • transcription unchecked
⊗4
(α_{r} + ξ_{r})^{∗}L ⊗ℒ^{−m}.
r=1Indeed the quadratic terms are U^{t}U = mI, and the linear translation terms vanish by U^{t}ξ = 0. Dividing the product of the four pulled-back theta sections by w_{ξ} gives a section f_{ξ} of ℒ^{m}. Choose a basis b_{1}, . . . , b_{m}g of H^{0}(A, L^{m}). The external products of these basis elements form a basis of H^{0}(B, ℒ^{m}): apply the free base-change result of Lemma 9.68 successively to the four factors and their projections. Thus
Selectable PDF notation • transcription unchecked
f_{ξ} =∑Q_{ξI}b_{I1} ⊠· · · ⊠b_{I4}
Idefines a square matrix Q of size m^{4g}.
We verify its independence in an algebraically closed residue field. Translation by v ∈ ker α leaves the numerator of f_{ξ} unchanged. Ratios w_{ξ}/w_{ξ0} transform by characters of that finite group, since a ratio with its translate is a nowhere vanishing global function on the connected proper B, hence is constant. These characters are distinct. Here is the source’s useful direct check of this fact. If the characters for ξ, ξ^{′} were equal, their trivialization ratio would be invariant under ker α. Fix a row r and pull back along i_{r} : A →A^{4}, x →→(U_{r1}x, . . . , U_{r4}x). The identity UU^{t}= mI gives α_{s}i_{r} = [m] when s = r and 0 otherwise. Cancel the constant line factors. The pulled-back ratio is a trivialization of [m]^{∗}(t^{∗}_{ξr}L ⊗t_{ξ}^{∗}_{r′}L^{−1}) invariant under A[m], because i_{r}(A[m]) ⊂ ker α. It descends along the finite etale torsor [m] to a trivialization of t^{∗}_{ξr}L ⊗t^{∗}_{ξr′}L^{−1}. Explicitly the invariance is the equality of its two pullbacks to A ×_{[m],A,[m]} A, identified with A × A[m]; faithfully flat gluing then descends the generator. The source’s principal-pairing condition makes ξ_{r} = ξ_{r}^{′}. This holds for every r, a contradiction.
Each f_{ξ} is nonzero: the isogeny α is surjective, so a point in the product of the four nonvanishing theta loci lifts after subtracting ξ. Eigenfunctions with distinct group characters are linearly independent. For a short proof, translate a minimal nontrivial linear relation by v and subtract one character value times the relation; choose v distinguishing two of the characters, obtaining a shorter nontrivial relation. This contradiction proves det Q ̸= 0 in the field. Over the original local ring, base change preserves the section bases, the disjoint kernel and these equations. The determinant is nonzero modulo its maximal ideal and therefore is a unit.
It remains to verify the origin normalization in the definition of the square comparison. Let µ be the normalized generator of [m]^{∗}L ⊗L^{−m2}. Every ξ_{r} is killed by m. The fibre of w_{ξ} at the origin is ⊗_{r}L_{ξr}⊗ L^{−}_{0}^{4m}; raising it to m^{2} and multiplying by ∏_{r}µ(ξ_{r}) cancels all L_{ξr} factors. The remaining fibre is L^{4}_{0}^{−4m3}. Divide by that power of the fixed rigidification. This is a unit ρ_{ξ} of the base ring, since both expressions are generators of the same rank-one module. Finally
λ = (det Q)^{2m2} ∏ρ^{2}_{ξ}
Selectable PDF notation • transcription unchecked
ξ
is a unit and has precisely the product-section expansion and origin relation required in the source definition. All quantities are on the same base and in the specified bases; no arbitrary complex rescaling was used to make a finite-prime determinant a unit.
Lemma 9.71 (From independent generic values to one value). Let k be algebraically closed, I finite, and let V_{R}(T, a) be a condition on T ∈R^{I}, a ∈R, for k-algebras R. Assume it is preserved by base change, has at most one value over each field, and has a value for every algebraically independent T in every algebraically closed extension. Assume values for two jointly independent tuples agree. Finally assume any value over the residue field of a discrete valuation k-algebra lifts after an injective extension of discrete valuation rings, with the same residue field map. Then there is κ ∈k such that every field-valued V_{F} (T, a) has a = κ in F.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ExistsAlgHomOfIsTranscendenceBasis.lean:13--58,74--148.
Two individually independent tuples have equal values even when they are not jointly independent: en-large their common algebraically closed field by a fresh independent tuple X, compare each tuple with X, and use injectivity of the field extension. Let K be an algebraic closure of k(X_{i} : i ∈I). Choose the value a_{X} at X. Every k-embedding K →K maps it to a value at another independent tuple. The equality just proved shows that all these embeddings fix a_{X}. It follows that a_{X} ∈k. Indeed if a_{X} were transcendental, extend it to a transcendence basis. The assignment sending each basis element t to t + 1 preserves algebraic independence, and extends from the rational function field to an embedding of its algebraic closure into K. It would move a_{X}. Since k is algebraically closed, algebraic elements of K over k already belong to k. Denote this fixed value by κ.
Now let T ∈F^{I} have value a. In an algebraic closure Ω of F(X_{i}) form
Selectable PDF notation • transcription unchecked
S_{i} = T_{i} + (X_{i} −T_{i})t ∈Ω[[t]].The tuple S is algebraically independent over k: a polynomial relation in the polynomial variable t would, on evaluating at 1, be a relation among X. The inclusion of these polynomials in Ω[[t]] is injective, so this proves independence there too. Its reduction at t = 0 is T in Ω. Apply the assumed valuation-ring lift to the value a. We get an injective extension Ω[[t]] →R^{′} and a value u ∈R^{′} reducing to a. Embed its fraction field in an algebraic closure Ω^{′}. The image of S is still independent. A map K →Ω^{′} sending X to that tuple exists by algebraic closedness and the transcendence-basis property. The base change of its value κ and the base change of u are values at the same tuple, hence equal. Injectivity gives u = κ in R^{′}. Reducing gives a = κ in Ω, and injectivity F →Ω gives it in F. This proves the assertion for arbitrary tuples where a value exists, not only on a generic open set.
Lemma 9.72 (The finite family in holomorphic key normalization). Assume the source’s weight-zero Koecher conclusion: every holomorphic symplectic-invariant scalar function on the Siegel domain ℌ_{g} is constant for g ≥ 2. Let a finite family of holomorphic functions F_{i} on that domain be permuted under symplectic changes of period basis. Then each F_{i} is constant.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ProdCompEqOfFiniteFamily.lean:19--129.
Discard duplicate functions, obtaining a finite set ℱ. For any group element the asserted substitution maps this finite set injectively into itself: equality after composition with an invertible action implies equality of the original functions. It is thus a permutation. For fixed z ∈ ℂ the function ∏_{F∈ℱ}(z −F(τ)) is holomorphic and symplectic invariant, so is constant in τ. Fix τ_{0} and put z = F_{i}(τ). The product vanishes at τ, so also at τ_{0}. Therefore F_{i}(τ) belongs to the fixed finite set {F(τ_{0}) : F ∈ℱ}. The Siegel domain is convex in the real vector space of symmetric matrices: convex combinations preserve positive definiteness of the imaginary part. It is connected, and each F_{i} is continuous. Its image in that finite set is a single point, proving constancy.
Lemma 9.73 (Removing the key normalization constant). Suppose a key invariant in dimension 2g, g ≥1, of exponent c and denominator M has been constructed, with the source’s scaling, field and local unit laws, and its holomorphic values at half-characteristics. Assume the theta symplectic transformation law, the weight-zero Koecher conclusion, and the source’s elliptic discriminant test. Then there is a key invariant in dimension g of exponent 2c and the same denominator M whose normal-form constant is 1.
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/ExistsRescale.lean:68--164; ExistsNormalisedKeyInvariantRuleCharZero.lean:48--99,101--158,161--216.
Apply Lemma 9.72 to the finite family indexed by pairs of half-characteristics. The theta transformation law permutes this family, and the scaling formula transports its values to arbitrary Lie bases and theta gen-erators, as in Lemma 9.65. In particular for the zero characteristic in dimension 2g there is a fixed scalar a_{0} with value a_{0}(det U^{4}/c^{8}_{0})^{c} in every normal form.
Pull this invariant back along the square x →→x^{2} of an object of dimension g. The theta characteristic is repeated twice, hence is even, and the integral symplectic theta law moves it to zero characteristic. Its deter-minant/generator quotient becomes (det U^{4}/c^{8}_{0})^{2}; the theta-law factors cancel in that fourth/eighth quotient. The invariant so obtained has exponent 2c, retains the same local unit law, and has value a_{0}(det U^{4}/c^{8}_{0})^{2c} for every half-characteristic. All copies are actual products of the original object, so their field and local base-change data agree.
We must check that dividing by a_{0} preserves integrality. The elliptic test in the source says the normal determinant/generator quotient for an elliptic Weierstrass object equals its discriminant. Take its gth product and evaluate the invariant over ℚ for an integral Weierstrass equation W. Its value is a_{0}∆(W)^{2gc}. Field naturality therefore gives a_{0} = r ∈ ℚ^{×} after division by that nonzero rational discriminant power. Over any discrete valuation ring where M and ∆(W) are units, the same equality shows r is a unit. Use the two source equations y^{2}+ y = x^{3} and y^{2} = x^{3}+ x, whose discriminants are −27 and −64. Their rational constants both map to the same complex a_{0}, hence are the same r. In a local ring either 3 or −2 = 1 −3 is a unit. In the first case use ∆= −27, in the second use ∆= −64. Thus r is a unit at every discrete valuation ring where M is a unit.
Divide the invariant by r. This preserves scaling and field naturality. It preserves the local unit and residue-field laws too: if w ∈R^{×} represents r, the relation r_{den}w = r_{num} determines its image under every characteristic-zero field specialization as the same rational number. The new normal-form value is exactly (det U^{4}/c^{8}_{0})^{2c}, as required.
Lemma 9.74 (The power-series socle determinant). Let S = k[[X_{1}, . . . , X_{g}]] over a field. If F_{i} =∑_{j}a_{ij}X_{j} and each X_{j}^{N} belongs to (F_{1}, . . . , F_{g}), then det a /∈(F_{1}, . . . , F_{g}).
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/CoeffEqZeroOfMemSpanPow.lean:11--111,121--206.
We use the following determinant comparison, including its elementary justification. If CX = C^{′}X and all entries of that vector lie in an ideal I, then det C^{′}− det C ∈I, provided the variables X_{j} form a regular sequence in every order. Replace the rows one at a time. If∑ v is the difference of the row being replaced,
_{j}v_{j}X_{j} = 0; the determinant difference is ∑_{j}v_{j}C_{j}, with C_{j} the cofactors of that row. The adjugate identity gives C_{j}X_{l} −C_{l}X_{j} ∈I. For any syzygy ∑_{j∈T}v_{j}X_{j} = 0 these relations imply ∑_{j∈T}v_{j}C_{j} ∈I, by induction on T: regularity expresses the last coefficient in the ideal generated by the other X_{j}, and subtracting those relations reduces to the shorter syzygy. This proves the determinant comparison. Regularity here is explicit: modulo any subset of the variables the power-series ring is the power-series ring in the others, where multiplication by a remaining variable is injective by comparison of coefficients.
Choose B with BF = (X_{j}^{N+1})_{j}, and put I = (X_{1}^{N+1}, . . . , X_{g}^{N+1}). If det a ∈(F), the adjugate of B
gives det B F_{j} ∈I, hence det(Ba) ∈I. But (Ba)X = (X_{j}^{N+1})_{j} equals the action of diag(X_{j}^{N}) on X. The determinant comparison yields ∏_{j}X_{j}^{N} ∈ I. This is impossible: every series in I has zero coefficient on the monomial ∏_{j}X_{j}^{N}, whereas that monomial has coefficient 1 in itself. This proves the lemma, without a characteristic restriction.
Lemma 9.75 (Discriminant equals the norm of the Jacobian up to a unit). Let R be a local ring, S = R[[X_{1}, . . . , X_{g}]], and B = S/(F_{1}, . . . , F_{g}) finite free over R of rank d, with each F_{i}(0) in the maximal ideal 𝔪_{R}. For any R-basis b of B, its trace discriminant is associated in R to the norm N_{B/R}(J) of the class J = det(∂F_{i}/∂X_{j}) in B.
Suppose further that R is a discrete valuation ring and e_{1}, . . . , e_{d} : S → R are R-algebra homomor-phisms satisfying: (i) e_{a}(F_{i}) = 0 for all a, i; (ii) e_{a}(X_{i}) ∈𝔪_{R} for all a, i; (iii) ∩^{d}_{a=1} ker e_{a} = (F_{1}, . . . , F_{g}); and (iv) there exists 0 ̸= c ∈ R with cR^{d} ⊂ im(e_{1}, . . . , e_{d}). Then B = S/(F_{1}, . . . , F_{g}) is automatically finite free over R of rank d, there exist lifts β_{1}, . . . , β_{d} ∈S whose classes form an R-basis of B, the matrix P = (e_{a}(β_{a}′))_{a,a}′ has det P ̸= 0, and
Selectable PDF notation • transcription unchecked
∏d
(det P)^{2} ∼ e_{a}(det(∂F_{i}/∂X_{j})) . (C.64)
Selectable PDF notation • transcription unchecked
a=1
Proof. Source: Development/AbelianVarietyTheory/Proofs/ThetaLawProofs/IsUnitDetOfProductMapNeZero.lean:30--587; DiscrMulDetEqNormOfAnnihilator.lean:7--55; PointMatrixMulLeftMulMatrix.lean:7--55; DiscrEqJacobianOfIntegralPoints.lean:9--110.
For the first assertion, let µ : S ⊗_{R} B → B be the S-algebra surjection s ⊗w →→ sw̄, with kernel K = ker µ. Put J_{∆} = (X_{j} ⊗1 −1 ⊗X^{̄}_{j})_{1≤j≤g} ⊂S ⊗_{R} B and I = (X_{1} ⊗1, . . . , X_{g} ⊗1) ⊂S ⊗_{R} B. Because B is finite over R, the algebra S⊗_{R}B is finite (hence integral) over S; because B = S/(F_{1}, . . . , F_{g}) is finitely presented over S, the kernel K of the S-linear surjection µ from a finite S-module is finitely generated over S, hence finitely generated as an ideal of S ⊗_{R} B. Write δ(s) = s ⊗1 −1 ⊗s̄ ∈ K for s ∈S. Every power series decomposes as s = s(0) +∑^{g}_{j=1}X_{j}u_{j} (u_{j} ∈S). Since δ(s(0)) = 0 and
Selectable PDF notation • transcription unchecked
δ(X_{j}u_{j}) = δ(X_{j})(u_{j} ⊗1) + (X_{j} ⊗1)δ(u_{j}) −δ(X_{j})δ(u_{j}) ∈J_{∆}+ IK,every element t =∑_{α}s_{α} ⊗w_{α} ∈S ⊗_{R} B satisfies t −1 ⊗µ(t) = ∑_{α}δ(s_{α})(1 ⊗w_{α}) ∈J_{∆}+ IK. In particular K ⊂J_{∆}+ IK. Each X_{j} lies in the Jacobson radical of S (since 1 + X_{j}y has constant term 1, hence is a unit in S). By integrality of S⊗_{R}B over S, every maximal ideal of S⊗_{R}B contracts to a maximal ideal of S, so I lies in the Jacobson radical of S ⊗_{R} B. Nakayama’s lemma for the finitely generated ideal K therefore gives K = J_{∆}.
Since µ(F_{i} ⊗1) = F^{̄}_{i} = 0, we can write
Selectable PDF notation • transcription unchecked
∑g
F_{i} ⊗1 = a_{ij}(X_{j} ⊗1 −1 ⊗X^{̄}_{j}) (a_{ij} ∈S ⊗_{R} B).
j=1Let π : S ⊗_{R} B →B ⊗_{R} B be the quotient on the first factor and put ∆= π(det a) ∈B ⊗_{R} B. Multiplying the vector (X_{j}⊗1−1⊗X^{̄}_{j})_{j} by the adjugate of (a_{ij}) gives (det a)(X_{l}⊗1−1⊗X^{̄}_{l}) ∈∑_{i}(S⊗_{R}B)(F_{i}⊗1) ⊂ ker π. Since K = J_{∆} contains s ⊗1 −1 ⊗s̄ for every s ∈S, applying π gives
Selectable PDF notation • transcription unchecked
∆(x ⊗1 −1 ⊗x) = 0 (x ∈B).
For each k, the R-linear map der_{k} : S⊗_{R}B →B, s⊗w →→∂s/∂X_{k} w, satisfies the Leibniz rule der_{k}(yz) = µ(y) der_{k}(z)+µ(z) der_{k}(y). Applying der_{k} to the defining relations for F_{i}⊗1 annihilates all terms containing µ(X_{j} ⊗1 −1 ⊗X^{̄}_{j}) = 0 and leaves ∂F_{i}/∂X_{k} = µ(a_{ik}). Hence µ_{B⊗B}(∆) = µ(det a) = det(µ(a_{ij})) = J in B.
Write ∆=∑_{i,j}M_{ij}b_{i} ⊗b_{j} with M ∈M_{d}(R). We prove directly over the arbitrary local ring R that det M ∈ R^{×}. Let 𝔫 = ( X^{̄}_{1}, . . . ,X^{̄}_{g}) ⊂ B. Any prime P ⊂ B containing 𝔪_{R}B contracts to 𝔪_{R}, hence is maximal by integrality of B/R; its preimage in S is maximal and therefore contains every X_{j} ∈ Jac(S). Thus 𝔫 ⊂^{√}𝔪_{R}B, and since 𝔫 is finitely generated there exists N ≥1 with 𝔫^{N} ⊂𝔪_{R}B. Let k = R/𝔪_{R}, let F^{̄}_{i} ∈k[[X_{1}, . . . , X_{g}]] be the coefficientwise reduction of F_{i}, and put C = k[[X_{1}, . . . , X_{g}]]/( F^{̄}_{1}, . . . ,F^{̄}_{g}). Coefficient reduction induces an R-algebra map ρ : B → C vanishing on 𝔪_{R}B; hence ρ(𝔫^{N}) = 0, so X_{j}^{N} ∈( F^{̄}_{1}, . . . ,F^{̄}_{g}) for every j. Because each F_{i}(0) ∈𝔪_{R}, constant-term reduction induces an R-algebra map ev : B →k killing 𝔫, and every w ∈B satisfies w −c ∈𝔫 for some c ∈R. Lift ρ⊗ev : B ⊗_{R} B →C along π∑ to the ring map Φ = red ⊗ ev : S ⊗_{R} B → k[[X]]. Applying Φ to the diagonal relations yields F^{̄}_{i} =_{j}Φ(a_{ij})X_{j} (as ev( ̄X_{j}) = 0). Hence (ρ ⊗ ev)(∆) is the class of det(Φ(a_{ij})) in C, which is nonzero by Lemma 9.74. On the other hand,
Selectable PDF notation • transcription unchecked
∑d (∑d )
(ρ ⊗ ev)(∆) = ρ(b_{i}) · ev M_{ij}b_{j} ̸= 0,
Selectable PDF notation • transcription unchecked
i=1 j=1
so some index i_{0} satisfies ev(w) ̸= 0 for w = ∑_{j}M_{i0j}b_{j} ∈ B. This w is a unit in B: write w −c ∈ 𝔫 with c ∈R; then ̄c = ev(w) ̸= 0 forces c ∈R^{×}, whereas any maximal ideal P ⊂B containing w contracts to 𝔪_{R} by integrality, hence contains 𝔪_{R}B ⊃ 𝔫^{N}, hence contains w −c by primality, and hence contains c ∈ 𝔪_{R}, a contradiction. Let w^{′} = w^{−1} ∈ B. For any R-linear functional f : B → R and any u ∈ B, applying f ⊗ id to ∆(u ⊗1) = ∆(1 ⊗u) gives the swap identity
Selectable PDF notation • transcription unchecked
∑M_{ij}f(b_{i}u) b_{j} =∑M_{ij}f(b_{i}) (b_{j}u).
i,j i,jGiven any t ∈R^{d}, set u = w^{′} ∑_{j}t_{j}b_{j} and take f = b^{∗}_{i0} (the i_{0}-th coordinate functional of the basis b). The right side collapses to wu = ∑_{j}t_{j}b_{j}, so comparing basis coefficients yields ∑_{i}b_{i}^{∗}_{0}(b_{i}u)M_{ij} = t_{j}. Thus v →→vM is surjective on R^{d}, so M ∈ GL_{d}(R) and det M ∈R^{×}.
To identify N_{B/R}(J), apply the swap identity with f = b^{∗}_{k} and u = b_{k} and sum over k = 1, . . . , d: since Tr_{B/R}(b_{i}) = ∑_{k}b_{k}^{∗}(b_{i}b_{k}) and ∑_{k}b_{k}^{∗}(b_{i})b_{k} = b_{i}, this gives ∑_{i,j}M_{ij} Tr_{B/R}(b_{i}) b_{j} = ∑_{i,j}M_{ij}b_{i}b_{j} = J. Applying the swap identity again with f = Tr_{B/R} and u = b_{l} yields
Selectable PDF notation • transcription unchecked
∑ ∑d (∑d )
Jb_{l} = M_{ij} Tr_{B/R}(b_{i}) (b_{j}b_{l}) = M_{ij} Tr_{B/R}(b_{i}b_{l}) b_{j}.
Selectable PDF notation • transcription unchecked
i,j j=1 i=1
Hence the matrix of left multiplication by J in the basis b is M^{t}T, where T_{il} = Tr_{B/R}(b_{i}b_{l}). Taking determinants gives N_{B/R}(J) = (det M)(det T) = (det M) discr_{R}(b), which proves the first assertion since det M ∈R^{×}.
Now assume the discrete valuation ring hypotheses (i)–(iv). By (i) and (iii), the evaluation map (e_{1}, . . . , e_{d}) : S → R^{d} descends to an injective R-linear map φ : B ↪→ R^{d}. Over the principal
ideal domain R, every submodule of R^{d} is finite free of rank at most d; since cR^{d} ⊂ im φ with c ̸= 0, the rank is at least d, so B is finite free of rank d. Choose an R-basis b of B and lift it to β_{1}, . . . , β_{d} ∈ S. If P = (e_{a}(β_{a}′)) had det P = 0, there would be a nonzero v ∈ R^{d} with Pv = 0, giving φ(∑_{a}′ v_{a}′b_{a}′) = 0, contradicting injectivity of φ and linear independence of b; hence det P ̸= 0. If d = 0, both sides of (C.64) equal 1. If d ≥ 1, pick any index a; decomposing F_{i} = F_{i}(0) +∑_{j}X_{j}u_{ij} and applying e_{a} gives F_{i}(0) = −∑_{j}e_{a}(X_{j})e_{a}(u_{ij}) ∈ 𝔪_{R} by (i) and (ii), so the first part applies to B and b. For every y ∈ B, expanding yb_{a}′ in the basis b and applying e_{a} yields the matrix iden-tity PL_{y} = diag(e_{a}(y))P over R, where L_{y} is the matrix of left multiplication by y in the basis b. Multiplying on the left by adj(P) and taking traces gives (det P) Tr_{B/R}(y) = (det P)∑_{a}e_{a}(y); taking determinants directly gives (det P)N_{B/R}(y) = (det P)∏_{a}e_{a}(y). Since R is a domain and det P ̸= 0, cancelling det P gives Tr_{B/R}(y) = ∑_{a}e_{a}(y) and N_{B/R}(y) = ∏_{a}e_{a}(y). In particular T_{ij} = Tr_{B/R}(b_{i}b_{j}) = ∑_{a}e_{a}(b_{i})e_{a}(b_{j}) = (P^{ t}P)_{ij}, so discr_{R}(b) = det T = (det P)^{2}, and the first part yields (C.64). No assumption that the special fibre is reduced is needed.