CM curves, the invariant derivation and divisibility
Four steps of the timeline speak about CM curves, the invariant derivation, divisibility by powers of a prime, norms from Chowla-Selberg and a gcd. The Lean code has a declaration for each of these words. But the declarations are smaller and more scattered than the steps suggest, and some of them do other jobs. This page follows each one and says what it really proves.
draft
Result
This page is about four steps of the timeline.
- Step 2: on a CM curve for inert in , where is the invariant derivation. This gives a divisibility of by a power of .
- Step 9: the norms of the lifts and the level, with divisibility from Chowla-Selberg, Weyl and the Siegel mass.
- Step 10: a lower bound on the divisibility of the -fold lift that is sufficient.
- Step 15: glue an integral structure across the Hecke characters to get a gcd.
The Lean code has these results that are close to the steps. All of them are proved in the code. None of them is a cut point in the literature inputs.
cmCurveGoodOutsidePrime. For each prime with , there is an elliptic curve over the number field , with CM by the integers of , with good reduction outside and with minimal discriminant .constantCoeff_pIterate_invariantDerivation_coord. For a lift of a formal group law, the constant term of is modulo . Here is the invariant derivation in direction , and comes from the coefficient of in a quasi-logarithm. This is a Hasse-Witt type formula.JetInt. Iterated derivatives at of an Eisenstein lattice sum, divided by powers of periods with of known -adic size, have a bounded -adic size.chowla_selberg_discriminant. For an imaginary quadratic field of discriminant with , the class group mean of is .betaArch_of_transport. The norm of an algebraic number , made from a Hecke L-value at a CM point, is bounded. The proof uses the Chowla-Selberg mean.flatProductDepthPeriodic_gcd. If a product of local test functions is periodic with periods and , then it is periodic with period .
Idea
The steps of the timeline tell one story: a derivative on a CM curve kills -th powers, so derivatives of theta values are divisible by ; the divisibility is then measured by norms through Chowla-Selberg, bounded from below, and combined into a gcd. The Lean code does not tell this story in one chain. The pieces live in different branches of the proof tree.
- The CM curve is a literature input. It is used once, to build the field data
FieldDatof a host field. It is not used to compute derivatives. - The formula for is about any lift of a formal group law. It is used to show that a dual Frobenius matrix is nilpotent, inside the proof of a prismatic Dieudonné statement. It does not mention a CM curve or theta.
- The divisibility of derivatives over periods is the statement
JetInt. Its proof uses bounds on values of Hilbert Eisenstein series at CM points, not the formula for . - The norm bound and the gcd are two small lemmas under the unitary hypotheses.
So the page has one section per piece, and it says at each point where the timeline and the code differ.
Before you read
You need these ideas.
- An algebraic number field, its places, its ring of integers and its class group.
- p-adic numbers and the -adic absolute value on . A local field.
- An elliptic curve with complex multiplication (CM), its -invariant, its lattice and the discriminant .
- A formal group law and its invariant derivation. For the third section, p-adic Hodge theory and -divisible groups.
- Eisenstein series and Hecke characters.
- A Dirichlet character and its L-function, for the Chowla-Selberg formula.
- Schwartz functions at the finite places, for the gcd.
Plan
- State the CM curve and see where the proof uses it.
- Define the invariant derivation and state the formula for .
- See where the formula is used: the nilpotence of a dual Frobenius.
- State
JetInt: derivatives over periods have bounded -adic size. - Turn this into lower bounds on valuations:
LatIntRam,LatIntUnrandToricDenominator. - State the Chowla-Selberg formula of the code.
- Bound the norm of with it:
BetaArch. - Get a gcd of depths, and use it for a minimal depth.
The steps
Step 1. The CM curve
The statement is in CmCurveGoodOutsidePrimePremises.lean#L35.
def CMCurveGoodOutsidePrimeSt : Prop :=
∀ p : ℕ, p.Prime → p % 4 = 3 → 3 < p →
∃ (F : Type) (_ : Field F) (_ : NumberField F) (A : WeierstrassCurve F),
IsGoodCMCurve p F A ∧ HasMinimalDiscriminant A (-(p : F) ^ 3)
IsGoodCMCurve p F A asks three things. is , where is the lattice of algebraic integers of . The -invariant of is . And has good reduction at each prime of that does not divide . HasMinimalDiscriminant A D asks that at each prime there is a minimal model of whose discriminant is a unit times . This is a curve of the kind that B. Gross studied.
The proof is CmCurveGoodOutsidePrime.lean#L47. It takes the lattice and the fact that is an algebraic integer. It scales the lattice, and it takes the twist of the curve by if and by if . Then at each prime it finds a local minimal model with discriminant a unit times . The proof dominates 169 declarations. The largest part (92) is the computation of the discriminant of the twisted model.
In the proof of the Riemann Hypothesis the statement enters through cm_curve_good_outside_prime_st_of_literature (EveryCharacterOfLiteratureOfVWin.lean#L277). This is one line: it ignores its argument O and returns cmCurveGoodOutsidePrime. It is a field of the bundle published_results_eigen (3,017).
The statement is used by baseCond_KPx (ExistsPrePin.lean#L30), which dominates 111 declarations.
theorem baseCond_KPx : RayField → PrincipalIdealSt → CMCurveGoodOutsidePrimeSt →
StandardLattice.SiegelMassRatioCM → BaseCond KPx
Inside it, nonempty_fieldDat applies the statement to , where is the discriminant of the imaginary quadratic host field and . The curve is stored in the structure FieldDat, and the code defines from it a torsion field and a discriminant field. So the CM curve has CM by the host field itself, and is ramified in that field, not inert. The same theorem takes SiegelMassRatioCM: for an unramified CM field and , the ratio of the masses of the standard Hermitian lattices of ranks and is an explicit value massRatioRightSide K M. The code proves it with massRatio_all_unramified_cmFields. This is the only place on this page where a Siegel mass appears.
Step 2. The invariant derivation and
Let be power series in over a ring . For , the code defines (InvDer.lean#L16)
When is a formal group law, is the invariant derivation in the direction .
The main statement is ConstantCoeffPIterateInvariantDerivationCoordStatement.lean#L8. It works in a "frame" IsFrame p Q θ: a surjection , an injection where is a unit, is -adically complete, has no -torsion, and for each we have with . The series is a lift (IsLift θ μ): its image under is a commutative formal group law. The series over is a quasi-logarithm for in coordinate : its linear part is the -th coordinate, and has coefficients in . Let with in . Then the statement is
In words: up to the ideal , the -th power of the invariant derivation acts on the coordinate through a -linear map, and the coefficients of this map are read off from the terms of the logarithm. When all are modulo this ideal, the constant term of is modulo it. This is the general form of "". The proof (constantCoeff_iterate_invDer_X_sub_mem) dominates 29 declarations, and the statement dominates 30.
The statement is about a formal group law in general. It does not mention an elliptic curve, CM, a prime inert in , a theta function or a period .
Step 3. Where the formula is used
The only user in the proof tree is isNil_dual_frobenius (IsNilDualFrobenius.lean#L24), which dominates 77 declarations. Its setting is a -divisible group over the ring of integers of an algebraically closed field , a frame built from a primitive system in the tilt, and a chart of by a formal group law. A basis of quasi-logarithms has a Frobenius matrix of the form with invertible. If the dual group is connected, the theorem proves that the matrix is nilpotent in the sense IsNil. The formula of Step 2 is what makes the dual Frobenius small modulo .
This lemma is used in window_matrices (1,206), and that is used in the proof of UpperTheory.Lit.prismaticDieudonneHodge (1,708). The statement PrismaticDieudonneHodge asks for a functor from -divisible groups over to free modules over Fontaine's ring , of rank equal to the height, with a Frobenius and a Hodge filtration . It is a literature input that the code proves. So in the code the invariant derivation serves -adic Hodge theory, not the divisibility of theta values.
Step 4. Derivatives over periods
The statement closest to "divisibility of " is JetInt (JetAnalytic.lean#L239).
It is stated for all data AllLat: a CM field that is abelian over , a CM type , an integer , values of partial zeta functions, a conductor , an integral lattice of the imaginary quadratic field of discriminant , a prime and a lattice of . For each embedding of into and each with , it asks for nonzero with two conditions.
- is algebraic with , and has a similar bound.
- For each Eisenstein lattice sum of weight attached to (
IsDefSum), and each list of directions with copies of , the number
is algebraic, and its -adic size is at most , with from the zeta values and from the conductor.
Here Sz l lam r x means that is algebraic and (TorsTransport.lean#L17). So JetInt says: derivatives of an Eisenstein sum at , divided by period powers, have bounded denominators at . This is a divisibility statement for derivatives over periods. The derivatives are partial derivatives on , not the invariant derivation of Step 2.
The proof is cmLatticeDividedPowerJetIntegrality_of_ellipticCM (CmLatticeDividedPowerJetIntegralityOfPublishedStatement.lean#L9), with 156 declarations. It takes four named inputs: SiegelKlingenDeligneRibet, PartialZetaFunctionalEquation, HilbertClassFieldJ and EllipticMainTheoremCM. The inner proof JetIntElliptic.jetInt starts with fun _ hPZ _ _ =>, so it uses only the functional equation of partial zeta functions. The other three inputs are not used. Its main parts are a bound on Hilbert Eisenstein values at CM points when is integral (norm_hilbertEisensteinValue_le_of_integral_j, 58) and a core lemma (66).
Step 5. Lower bounds on valuations
From JetInt the code gets two statements on the -adic valuation of (TorsTransport.lean#L33). The number is defined in CMVal.beta:
where is the value at of a lattice sum (IsLatSum), is an explicit constant and is the weight. LatIntRam (for that divides the conductor) and LatIntUnr (for that does not) both say
This is a lower bound on the divisibility of by . The theorem toricDenominator (ToricDenominatorOfZetaDenominators.lean#L39, 32 declarations) combines the two bounds with ZetaDenominators, a bound on the denominators of partial zeta values, into the statement ToricDenominator.
The chain is latIntUnr (20) and latIntRam (9), then toricDenominator (32), then joinOut_of_lat (180), then joinOut_of_jetInt (233), then joinOut_of_five (1,598), then join_out_of_literature (1,604) and published_results_eigen (3,017). The output JoinOut holds, among other things, the statement Transport of Step 7.
The timeline speaks of "the -fold lift". The code has no -fold lift here. The closest counterpart of step 10 is this lower bound on .
Step 6. The Chowla-Selberg formula
Let be an imaginary quadratic field with discriminant , , class number and character . For each ideal class pick an ideal in and let be its lattice in . The code proves (DedekindZetaEqSumRealEisenstein.lean#L259)
Here is Euler's constant. The left side is a mean over the class group. It does not depend on the choice of , because does not change when the lattice is scaled. The proof uses the Kronecker limit formula for the real analytic Eisenstein series (tendsto_realEisenstein_sub_pole, in TendstoRealEisensteinSubPole.lean of the same folder) and the class number formula (LFunction_one_eq_classNumber, in the same file). Both are proved in the code.
The version used by the proof tree is classMean_log_norm_disc_chowlaSelberg (ClassMeanLogNormDiscChowlaSelbergStatement.lean#L9). It is the same formula, written with nrm, latDisc and dlogL D 1 for the host field . It dominates 8 declarations.
Step 7. The norm bound
The statement BetaArch (NormSmall.lean#L96) bounds the norm of . Let be the value of a Hecke L-function at (IsTor), and let be an integral lattice. Then
where is a bound bTH fixed by the data. Up to sign, the constant coefficient of the minimal polynomial is the norm of from to .
The proof is betaArch_of_transport (BetaConjugateMeanLogAbsBoundOfTransport.lean#L43), with 70 declarations. It takes four inputs: Transport, FieldDat, LBound and TorEx. The argument has three parts.
- Find the imaginary quadratic subfield of the field data.
Transportgives each Galois conjugate as a value of the same shape, for an ideal and a lattice . This is Shimura reciprocity. WithLBound(a bound on for Hecke L-functions) it gives , where contains the term .- Average over the conjugates. They run over the ideal classes of , so the mean of the terms is given by the Chowla-Selberg formula of Step 6.
The chain above it is betaConjugate_mean_logAbs_bound_of_transport (71), then square_top_term_joined_estimates (2,312), then fittedFunction_at_zero_close_alt (4,154), and then the unitary hypotheses. So here "Chowla-Selberg gives divisibility" becomes "Chowla-Selberg controls the archimedean size of the conjugates of ".
Step 8. A gcd of depths
Fix data of a grid: places, local Schwartz test functions and a family . The function flatProd is the product over places of local factors. DepthOK S t Φ β e K (IsExpansionIff.lean#L20) says two things.
- If
flatProdis not zero at half-integral points , then and lie in the given lattices. flatProdis periodic: for lattice vectors .
The statement flatProductDepthPeriodic_gcd (FlatProductDepthPeriodicGcdStatement.lean#L9, 5 declarations) is
The proof is Bezout: with integers , and lattices and periods are closed under integer combinations.
The user is depthK_of_hypotheses_alt (DepthOfSumBound.lean#L12), with 49 declarations. Let expDepth be the least valid depth. For any valid , the gcd of expDepth and is valid, and it is at most expDepth, so it equals expDepth. Hence expDepth divides , and for each prime . With a bound on this gives DepthK. The chain above it is depthK_of_sumOKK (50), then sum_size_finite_away_archimedean_combined (1,150), then fittedFunction_at_zero_close_alt (4,154).
So the gcd in the code is a gcd of periods of one product of local test functions. It is not a gcd across Hecke characters.
Where it is in the code
Sizes are dominator sizes: the number of declarations that only this node reaches on the way down from the top. All paths are relative to lean/src. DivisorProofs is Development/AbelianVarietyTheory/Proofs/DivisorProofs/.
| Declaration | Module (file) | Size |
|---|---|---|
GlobalTheory.Published.cmCurveGoodOutsidePrime |
DivisorProofs/CmCurveGoodOutsidePrime.lean |
169 |
UpperTheory.LiteratureTheorems.cm_curve_good_outside_prime_st_of_literature |
.../ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin.lean |
170 |
GlobalTheory.Mem.Top.Bl.KPxCond.baseCond_KPx |
.../IwasawaDecomposition/ExistsPrePin.lean |
111 |
GlobalTheory.T.constantCoeff_pIterate_invariantDerivation_coord |
DivisorProofs/ConstantCoeffPIterateInvariantDerivationCoordStatement.lean |
30 |
GlobalTheory.DualFrobNil.isNil_dual_frobenius |
DivisorProofs/IsNilDualFrobenius.lean |
77 |
UpperTheory.Lit.prismaticDieudonneHodge |
.../PrismaticDieudonne/PrismaticDieudonneHodge.lean |
1,708 |
GlobalTheory.T.cmLatticeDividedPowerJetIntegrality_of_ellipticCM |
DivisorProofs/CmLatticeDividedPowerJetIntegralityOfPublishedStatement.lean |
156 |
GlobalTheory.ToricDenomRow.toricDenominator |
DivisorProofs/ToricDenominatorOfZetaDenominators.lean |
32 |
GlobalTheory.JoinFun.joinOut_of_five |
DivisorProofs/JoinOutOfFive.lean |
1,598 |
GlobalTheory.T.classMean_log_norm_disc_chowlaSelberg |
DivisorProofs/ClassMeanLogNormDiscChowlaSelbergStatement.lean |
8 |
GlobalTheory.SqTop.betaArch_of_transport |
DivisorProofs/BetaConjugateMeanLogAbsBoundOfTransport.lean |
70 |
GlobalTheory.T.flatProductDepthPeriodic_gcd |
DivisorProofs/FlatProductDepthPeriodicGcdStatement.lean |
5 |
GlobalTheory.DepthUnf.depthK_of_hypotheses_alt |
.../LocalTypeDictionary/DepthOfSumBound.lean |
49 |
JetInt, LatIntRam, LatIntUnr, BetaArch, Transport and DepthOK are definitions, with size 1. The small sizes of classMean_log_norm_disc_chowlaSelberg (8) and chowla_selberg_discriminant (6) mean that most of the Chowla-Selberg proof is shared with other parts of the tree, so it is counted higher up. The whole proof of riemannHypothesis has 757,483 declarations in the spine. The largest node on this page, prismaticDieudonneHodge, has 1,708.
The folder GroundTheory/HeightsAndTheta/Heights/ChowlaSelberg/ (namespace GroundTheory.Heights₂.Heights.ChowlaSelberg) has 58 files and 296 declarations in the index, for example a structure CMFamily and files on the class mean and the Kronecker limit. None of these declarations is in the spine, and no declaration outside the folder uses them. They are not part of the proof of the Riemann Hypothesis.
Notes
- The code is not reviewed. The site does not claim that the proof is correct. The trust report says what has been checked.
- Step 2 of the timeline, " on a CM curve for inert": the code has the formula for on any formal group lift, and it has a CM curve, but no declaration joins them. The CM curve of the code has CM by itself, so is ramified there, and in the proof is the discriminant of the host field. No declaration on this page speaks of a prime inert in .
- Step 2, "divisibility of ": the code has no theta function on a CM curve here. The closest statement is
JetInt, which is about partial derivatives of Eisenstein lattice sums for a CM field , divided by period powers. Its proof does not use the invariant derivation. JetIntElliptic.jetInttakes four published-named inputs and uses onlyPartialZetaFunctionalEquation. The namesSiegelKlingenDeligneRibet,HilbertClassFieldJandEllipticMainTheoremCMin its type do not do any work there.- Step 9, "the norms of the lifts and the level": the norm in the code is the norm of one algebraic number built from a Hecke L-value, not of a lift. The page does not show how the level enters. "Terms of order cancel" has no counterpart on this page. "Weyl" has no counterpart on this page. The Siegel mass appears only as the input
SiegelMassRatioCMofbaseCond_KPx, next to the CM curve, and not in a divisibility argument. - Step 10, "a lower bound on the divisibility of the -fold lift": there is no -fold lift in these declarations. The closest counterpart is the lower bound of
LatIntRamandLatIntUnr. This page does not check that this bound is "sufficient" in the sense of the timeline. - Step 15, "glue an integral structure across the Hecke characters to get a gcd": the code has a gcd, but it is the Bezout gcd of two periods of one product of local test functions. There is no integral structure glued across Hecke characters on this page. That idea has no counterpart here.
- None of the statements on this page has an entry in
links/literature.yaml. The CM curve, the prismatic Dieudonné statement and the Chowla-Selberg formula are proved in the code, so they are not cut points. - This page does not unfold
Transport,FieldDat,LBound,TorEx,ZetaDenominatorsor the constantbetaBound_vp.
The source of this page. Markdown. Something is wrong? Report it.