riemann

The Lean code claims a proof of the Riemann Hypothesis. This site has not checked the claim yet. See what is checked.

Siegel-Weil, the metaplectic representation and cusp forms

The literature node of the proof names a Siegel-Weil formula, two metaplectic representations, a convergent Siegel-Weil formula and a cusp decay bound. In the Lean code each of them is a theorem with a proof, not a citation. The largest is the Siegel-Weil formula on doubled unitary groups. This page follows its proof, and says what the code has in place of the period integrals and Poincaré series of the note.

draft

Result

The node every_character_of_literature_of_vWin (see the literature inputs) takes a bundle O : OpenLiteratureStatements and gets many named statements from it. Five of them are the subject of this page.

Theorem What it gives Line
siegel_weil_std_of_literature SiegelWeilStd, the Siegel-Weil formula for doubled unitary groups L563
metaplectic_representation_of_literature MetaplecticRepresentation, the Weil representation of the metaplectic cover of Sp2n(ℝ)\mathrm{Sp}_{2n}(\mathbb{R}) L463
real_metaplectic_representation_of_literature RealMetaplecticRepresentation, the same with a Lie algebra action L491
siegelWeilU_unitary_convergent_of_literature SiegelWeilUnitaryConvergent, finite volume and Siegel densities L581
cusp_decay_sobolev_of_literature CuspDecaySobolev, rapid decay of cusp forms with a Sobolev bound L337

Each theorem takes O as an argument and does not read it. Each one is a call to a theorem that the code proves. So in the Lean code these statements are not assumptions. They are proved results with names from the literature.

The main result is SiegelWeilStd (HypothesesS.lean#L11). In words: let E/FE/F be a CM extension of number fields. Let AA be a definite Hermitian (or skew-Hermitian) space of size mm, and YY a skew-Hermitian (or Hermitian) space of size nn. Let ω\omega be a Weil representation of the doubled group G2=U(Y⊕−Y)G_2 = U(Y \oplus -Y) on Schwartz functions Φ\Phi, with a Hecke character χ\chi. Then the Siegel Eisenstein series E(x,s,Φ)E(x, s, \Phi) has a value at s0=m−n2s_0 = \tfrac{m-n}{2}, and

E(x,s0,Φ)=c⋅1μ(D)∫Dθ(ω(x)SgΦ) dμ(g),c={1n<m,2n≥m. E(x, s_0, \Phi) = c \cdot \frac{1}{\mu(D)} \int_{D} \theta\big(\omega(x) S_g \Phi\big) \, d\mu(g), \qquad c = \begin{cases} 1 & n < m, \\ 2 & n \ge m. \end{cases}

Here DD is a fundamental domain for UA(F)U_A(F) in UA(𝔸)U_A(\mathbb{A}), μ\mu is a Haar measure, SgS_g is the action of g∈UA(𝔸)g \in U_A(\mathbb{A}), and θ(Φ)=∑ξΦ(ξ)\theta(\Phi) = \sum_{\xi} \Phi(\xi) is the theta sum over rational points. The theorem PositivityTheory.SiegelWeilFormula.Fin.siegelWeilStd (SiegelWeilStd.lean#L8) proves it with no hypotheses.

Idea

Timeline step 3 of the note says that the quantity NN of the note is automorphic, because a function μ\mu comes from a theta function on a sixfold cover of GL5\mathrm{GL}_5, as a period integral of thetas. Step 16 uses a relative trace formula on the same cover, and bounds a projected Poincaré series with Cauchy-Schwarz. Step 18 is about covolumes and the index of theta series.

The Lean code does not contain a cover of GL5\mathrm{GL}_5, a relative trace formula, or a Cauchy-Schwarz bound on a projected Poincaré series. What it does contain is near to these ideas, on unitary groups.

  • A theta integral (a period of a theta function over UA(F)\UA(𝔸)U_A(F) \backslash U_A(\mathbb{A})) is an Eisenstein series. This is the Siegel-Weil formula above. It is the closest counterpart to "a period integral of thetas is automorphic". The proof is the classical one: compare Fourier coefficients, then use an induction on nn.
  • The metaplectic cover is the double cover of Sp2n(ℝ)\mathrm{Sp}_{2n}(\mathbb{R}), not a sixfold cover of GL5\mathrm{GL}_5.
  • Covolumes appear as the finite volume of U(F)\U(𝔸)U(F) \backslash U(\mathbb{A}) and as Siegel densities, in SiegelWeilUnitaryConvergent.
  • Poincaré series appear in one place: to build a cusp form whose theta period is not zero. The identity term of the series is larger than the rest at large level. This is a lower bound, not a Cauchy-Schwarz upper bound.

Before you read

You need these ideas.

Plan

  1. Read the statement SiegelWeilStd.
  2. The Schrödinger model: the formulas that the proof uses.
  3. The case n<mn < m: Fourier coefficients of both sides agree.
  4. The case n=mn = m: the constant 2.
  5. The case m<nm < n: an induction on nn with a constant term split.
  6. Holomorphy at s0s_0: the SiegelWeilProof modules.
  7. Where the formula is used: theoremK_head2.
  8. The two metaplectic statements.
  9. Finite volume and Siegel densities (timeline step 18).
  10. Cusp decay and Poincaré series (timeline steps 3 and 16).

The steps

Step 1. The statement

The statement is long. It quantifies over the fields F,EF, E, the conjugation σ\sigma, a base datum B (with an additive character ψ\psi), the sizes m,n≥1m, n \ge 1, and the spaces A : Spc F E m and Y : Spc F E n. It asks that AA is definite (min r s = 0), and that one of A,YA, Y is Hermitian and the other is skew-Hermitian. Then it takes:

  • a Hecke character χ\chi with ResEtaPow χ m (its restriction to 𝔸F×\mathbb{A}_F^\times is ηm\eta^m, where η\eta is 1 on norms from EE and on a given subgroup KK, and −1-1 elsewhere);
  • a doubling datum IsDbl B.ψ dd χ om S l: a Weil representation om of G2(𝔸)G_2(\mathbb{A}), the action S of UA(𝔸)U_A(\mathbb{A}), and a functional l with ∑yl(⋅)\sum_{y} l(\cdot) over rational yy equal to thetaSum;
  • a frame b for the doubled form, the standard maximal compact K, and a height a with a(pk)=∣det⁡t∣a(pk) = |\det t|;
  • a Haar measure μ\mu and a fundamental domain DD;
  • a KK-finite Schwartz function Φ\Phi.

The conclusion is EisValueAt. It says: for each xx, there is a meromorphic function that is equal to the Eisenstein series of the section x↦l(ω(x)Φ)x \mapsto l(\omega(x)\Phi) for large Re⁡(s)\operatorname{Re}(s), is analytic at s0s_0, and takes the value c⋅1μ(D)∫Dθ(ω(x)SgΦ) dμ(g)c \cdot \tfrac{1}{\mu(D)}\int_D \theta(\omega(x)S_g\Phi)\,d\mu(g) there. The section is flattened by flat: fs(x)=a(x)s−s0f(x)f_s(x) = a(x)^{s - s_0} f(x) (Hypotheses2.lean#L48).

Step 2. The Schrödinger model

The proof works with the data bundle SD in AuxiliaryPrimeFunction.lean#L34. It defines a transform TΦ(X)=l(ρ(Xb2T)Φ)T\Phi(X) = l(\rho(X b_2^T)\Phi) on m×nm \times n matrices XX. Five properties are bundled in TAll.

  • TBij: TT is a bijection onto the Schwartz space of matrices.
  • TPar (L178): the Siegel parabolic acts by explicit formulas. The unipotent element (1β01)\begin{pmatrix} 1 & \beta \\ 0 & 1 \end{pmatrix} multiplies TΦ(X)T\Phi(X) by ψ(−12tr⁡(β gram(X,X)))\psi\big(-\tfrac12 \operatorname{tr}(\beta \, \mathrm{gram}(X,X))\big). The Levi element (t00u)\begin{pmatrix} t & 0 \\ 0 & u \end{pmatrix} gives χ(det⁡t) ∣det⁡t∣m/2 TΦ(Xu−T)\chi(\det t)\,|\det t|^{m/2}\, T\Phi(X u^{-T}). The group UAU_A acts by X↦g−1XX \mapsto g^{-1}X.
  • TWeyl (L186): the Weyl element acts by a Fourier transform, T(ω(w)Φ)(X)=∫ψ(εtr⁡ gram(X,Z)) TΦ(Z) dν(Z)T(\omega(w)\Phi)(X) = \int \psi\big(\varepsilon \operatorname{tr}\,\mathrm{gram}(X,Z)\big)\, T\Phi(Z)\,d\nu(Z).
  • TTheta: ∑XTΦ(X)=θ(Φ)\sum_X T\Phi(X) = \theta(\Phi) over rational XX.
  • ThetaAuto: 0<μ(D)<∞0 < \mu(D) < \infty, the theta integral I(Φ)I(\Phi) is continuous, and it is left invariant under G2(F)G_2(F).

The proof gets TAll from a Poisson summation theorem (TW.Twp.poisson) in siegelWeilStd_pre (SiegelWeilStdPre.lean#L19).

Step 3. The case n < m

Here the constant is 1. The proof compares Fourier coefficients along the unipotent radical of the Siegel parabolic. For a nondegenerate Hermitian matrix TT of size nn, the orbit integral is

orbT(Φ)=1μ(D)∫D∑X : gram(X,X)=−2TTΦ(g−1X) dμ(g). \mathrm{orb}_T(\Phi) = \frac{1}{\mu(D)}\int_D \sum_{X \,:\, \mathrm{gram}(X,X) = -2T} T\Phi(g^{-1}X)\, d\mu(g).

The code uses the letter TT both for this matrix and for the transform TΦT\Phi of Step 2.

Two statements carry the comparison.

  • ThetaCoef (L197): the TT-th Fourier coefficient of the theta integral I(Φ)I(\Phi) at xx is orbT(ω(x)Φ)\mathrm{orb}_T(\omega(x)\Phi).
  • WhitK (L266): for n<mn < m, the TT-th Fourier coefficient of the Eisenstein series (a Whittaker-type integral over β\beta) is holomorphic at s0s_0 with value orbT(ω(x)Φ)\mathrm{orb}_T(\omega(x)\Phi).

From these, Main (L271) says that the Eisenstein series is holomorphic at s0s_0 with value I(Φ)(x)I(\Phi)(x). The proof also needs the degenerate coefficients and absolute convergence (AbsThm). Then HeadOf turns Main into SWStdLt, the statement for n<mn < m.

Step 4. The case n = m

Here s0=0s_0 = 0 and the constant is 2. The statement MainEq covers this case. It is proved in Atop.mainEq_of, from the same inputs as Main and a statement WhitEq on the Fourier coefficients in the case m=nm = n.

Step 5. The case m < n

Here the proof is an induction on nn. A lower rank datum LD (LowerRankDatum.lean#L94) embeds a problem of size n−1n-1 into the problem of size nn. Three statements do the work.

  • OrbRes (L175): for m<nm < n there is no XX of full rank, so each nondegenerate orbit integral is 0. The other orbit integrals are restrictions of orbit integrals of the smaller problem.
  • CTSplit (L155): the constant term along the unipotent of the embedding is a Fourier coefficient of a lower rank Eisenstein series, plus a second term that scales by eτ(n−2z)e^{\tau(n-2z)}.
  • Step (L178): if the formula with constant 2 holds at rank n−1n-1, it holds at rank nn, for m<nm < n.

MainLe puts MainEq and Step together and gets the formula with constant 2 for all m≤nm \le n. The statement FieldOf (L194) says that EisBL, SecK, SWStdLt (the case n<mn < m) and MainLe (the case m≤nm \le n) give SiegelWeilStd. The theorem Atop.fieldOf proves it. The theorem Atop.siegelWeilStd_of (SiegelWeilStdLeaves.lean#L40) does this assembly. It is a list of 16 have lines and one exact. The constant term split itself is proved in Act.ctSplit_of (CtSplitOf.lean#L10).

Step 6. Holomorphy at s0

The induction needs to know that the Eisenstein series has no pole at s0s_0 in the range m<nm < n. This is HolLead (LowerRankDatum.lean#L20). It is proved by holLeadB (HolLeadB.lean#L9), in the folder CayleyCosetSums. The folder Development/WeilBound/RepresentationTheoryGL/SiegelWeilProof has 635 modules and 5,141 declarations. 4,940 of them are in the spine. Most of its large nodes are shared with other parts of the proof, so their dominator is everyCharacter, not siegelWeilStd. Examples are holLeadB (660), Rg.Sw.Glob.top7 (658) and Rg.Sw.lc_of (619). The closure of Fin.siegelWeilStd has 21,500 declarations. 3,682 of them are in the SiegelWeilProof modules.

The meromorphic continuation of the Eisenstein series is a separate statement, EisBL (HypothesesE.lean#L46). It is proved by PositivityTheory.RgEis.Hd.eisBL.

Step 7. Where the formula is used

The theorem theta_integrals_agree_and_local_matching_of_literature (L565) is theoremK_head2 (eis_bl O) (siegel_weil_std O). It gives ThetaIntegralsAgree and LocalMatching, the two theta inputs of the theta matching chapter. So the Siegel-Weil formula is the source of the statement that two theta integrals into the same unitary group are proportional. Other users of SiegelWeilStd in the spine include theoremK_head2_of_mod, closeSph, swAnisoX_of and siegelWeilAniso_of.

Step 8. The metaplectic representation

MetaplecticRepresentation (MetaplecticRepresentation.lean#L94) says: for each finite index set, there are a topological group GG and a covering map pr:G→Sp(ℝ)\mathrm{pr} : G \to \mathrm{Sp}(\mathbb{R}) onto, with a kernel of 2 elements, a unitary representation Ω\Omega on L2L^2 and a representation ω\omega on Schwartz space. They agree on Schwartz functions, ω\omega is continuous, ω\omega intertwines the Heisenberg action, and the nontrivial kernel element acts by −1-1. For the Gaussian φ0(x)=e−πx⋅x\varphi_0(x) = e^{-\pi x \cdot x} it fixes the square of the vacuum matrix coefficient:

(∫ω(g)φ0 φ0‾)2det⁡(holPart(pr g))=(∫φ0φ0‾)2,holPart(ABCD)=12((A+D)+i(C−B)). \Big(\int \omega(g)\varphi_0 \, \overline{\varphi_0}\Big)^2 \det\big(\mathrm{holPart}(\mathrm{pr}\,g)\big) = \Big(\int \varphi_0\overline{\varphi_0}\Big)^2, \qquad \mathrm{holPart}\begin{pmatrix} A & B \\ C & D\end{pmatrix} = \tfrac12\big((A+D) + i(C-B)\big).

The proof is GlobalTheory.metaplecticRepresentation_holds (MetaplecticRepresentationHolds.lean#L17).

RealMetaplecticRepresentation (RealMetaplecticRepresentation.lean#L351) is a second version. It has the same kind of double cover. It also says that ω\omega is smooth along smooth curves, and that each XX in the Lie algebra 𝔰𝔭\mathfrak{sp} gives a one-parameter subgroup that lifts exp⁡(tX)\exp(tX), with derivative the quadratic operator of XX. Both statements are also fields of the bundle ThetaLocalInputs.

Step 9. Finite volume and Siegel densities

SiegelWeilUnitaryConvergent (SiegelWeilUnitaryConvergent.lean#L46) says: for a CM extension E/FE/F and a nondegenerate Hermitian matrix JJ of size MM, two facts hold. The fundamental domain ℱ\mathcal{F} of UJ(F)U_J(F) in UJ(𝔸)U_J(\mathbb{A}) has finite volume. And SiegelDensities (L18) holds: for 2n<M2n < M and nondegenerate Hermitian TT, the mean over ℱ\mathcal{F} of the theta sum over XX with gram(X)=T\mathrm{gram}(X) = T is a product of local densities, g∞(T)∏vgv(T)g_\infty(T)\prod_v g_v(T). This is the convergent range of the Siegel-Weil formula, in the form of Siegel's mass formula.

This is the code's counterpart of "covolumes" in timeline step 18. The proof CompanionTheory.SW.siegelWeilUnitaryConvergent has two parts: finVol_unitaryGroup and siegel_densities_of_hermitian_form. The spine has 43 users of this statement. The covolume bound CovolumeBoundOfX of the theta matching chapter is a different statement (LiftLambdaBound.lean#L60).

Step 10. Cusp decay and Poincaré series

CuspDecaySobolev (ExistsGrowthSobNeTopOfIsAutFunSm.lean#L18) says: for each open UU and each N,MN, M, there are dd and CC so that each smooth, automorphic, UU-invariant cusp function φ\varphi with finite Sobolev growth norm satisfies

decayM(φ)≤C⋅growthSobd,N(φ). \mathrm{decay}_M(\varphi) \le C \cdot \mathrm{growthSob}_{d,N}(\varphi).

So cusp forms decay faster than each power of the height, with a constant controlled by a Sobolev norm. The proof goes through cuspDecaySobolev_of_bound and CuspGeo.cuspDecayBound.

The Poincaré series of the code are in the proof of FunctionalsSeparate, in the theorem functionalsSeparate_of_howeUnram_of_siegelWeil (FunctionalsSeparateOfHoweUnramOfSiegelWeil.lean#L17). The chain is this.

  • exists_cusp_form_theta_period_ne_zero (Period.lean#L18): for each nonzero row tt there are a cusp form yy in CoreForms and vectors v,wv, w with per0(y(v),t(w))≠0\mathrm{per}_0(y(v), t(w)) \ne 0. This is a theta period that is not zero.
  • exists_multT_eq_poin (ExistsMultTEqPoin.lean#L75): a Poincaré series is a cusp form. It uses absolute convergence and moderate growth of the series.
  • poincareSeries_pairing_ne_zero_of_large_level (PoincareSeriesIdentityTermDominatesStatement.lean#L11): if 2eQ+(6r+9)∣τ∣≤s∣τ∣N𝔪2e^{Q + (6r + 9)|\tau|} \le s^{|\tau|}\sqrt{N\mathfrak{m}}, where rr is the rank and N𝔪N\mathfrak{m} the norm of the level, then ∫ℱpoin⋅φ≠0\int_{\mathcal{F}} \mathrm{poin}\cdot\varphi \ne 0. The identity term is larger than the sum of the other terms.

The theorem functionalsSeparate_of_howeUnram_of_siegelWeil takes GlobalTheory.Published.SiegelWeilUnitaryConvergent as an input. That is a second statement with the same name as the one in Step 9, in a different namespace.

Where it is in the code

Sizes are dominator sizes from the current spine: the number of declarations that only this node reaches on the way down from the top. A small size does not mean a small proof. It can mean that the proof is shared with other parts of the code. All paths are relative to lean/src.

Declaration Module (file) Size
UpperTheory.LiteratureTheorems.theta_integrals_agree_and_local_matching_of_literature ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin.lean 8,520
PositivityTheory.ThKR.theoremK_head2 ThetaLifts/Reductions/FiniteClosure/TheoremKOfMod.lean 6,823
UpperTheory.LiteratureTheorems.siegel_weil_std_of_literature ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin.lean 1,652
PositivityTheory.SiegelWeilFormula.Fin.siegelWeilStd SiegelWeilFormula/SiegelWeilStandard/SiegelWeilStd.lean 1,651
PositivityTheory.SiegelWeilFormula.Fin.siegelWeilStd_of_ct SiegelWeilFormula/SiegelWeilStandard/SiegelWeilStdOfCt.lean 1,172
PositivityTheory.SiegelWeilFormula.Fin.siegelWeilStd_pre SiegelWeilFormula/SiegelWeilStandard/SiegelWeilStdPre.lean 1,128
PositivityTheory.SiegelWeilFormula.Fin.siegelWeilStd_mid SiegelWeilFormula/SiegelWeilStandard/SiegelWeilStdMid.lean 798
PositivityTheory.SiegelWeilFormula.Fin.siegelWeilStd_of SiegelWeilFormula/SiegelWeilStandard/SiegelWeilStdOf.lean 621
PositivityTheory.SiegelWeilFormula.Act.ctSplit_of SiegelWeilFormula/SiegelWeilActions/CtSplitOf.lean 464
PositivityTheory.SiegelWeilFormula.Atop.siegelWeilStd_of SiegelWeilFormula/FockVacuumBounds/SiegelWeilStdLeaves.lean 54
PositivityTheory.Rg.Sw.HSt.holLeadB SiegelWeilProof/HolomorphicContinuation/CayleyCosetSums/HolLeadB.lean 660 (shared)
UpperTheory.LiteratureTheorems.real_metaplectic_representation_of_literature ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin.lean 133
UpperTheory.LiteratureTheorems.metaplectic_representation_of_literature same file 42
UpperTheory.LiteratureTheorems.siegelWeilU_unitary_convergent_of_literature same file 6
UpperTheory.LiteratureTheorems.cusp_decay_sobolev_of_literature same file 5
UpperTheory.Lit.functionalsSeparate_of_howeUnram_of_siegelWeil LocalTypeDictionary/FunctionalsSeparateOfHoweUnramOfSiegelWeil.lean 4,530
UpperTheory.Lit.Cusp.exists_cusp_form_theta_period_ne_zero RealBands/Period.lean 448
UpperTheory.Lit.Cusp.exists_multT_eq_poin RealBands/ExistsMultTEqPoin.lean 186
GlobalTheory.T.poincareSeries_pairing_ne_zero_of_large_level TwistedOperators/PoincareSeriesIdentityTermDominatesStatement.lean 65

The ThirdPlaceSeries and LocalTypeDictionary files are in Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/. The RealBands files are in Development/AutomorphicArithmetic/ThetaCorrespondence/. The ThetaLifts file is in Development/WeilBound/. The SiegelWeilFormula files are in Development/WeilBound/CoreStatements/AutomorphicKernels/. The SiegelWeilProof files are in Development/WeilBound/RepresentationTheoryGL/. The TwistedOperators file is in Development/AbelianVarietyTheory/Proofs/.

Closures (all declarations used, shared or not, without Mathlib) are larger: Fin.siegelWeilStd 21,500, theoremK_head2 23,711, theta_integrals_agree_and_local_matching_of_literature 38,926, the metaplectic proof 1,094, the real metaplectic proof 1,049, the convergent Siegel-Weil proof 5,387 and the cusp decay proof 2,722.

Notes

  • No counterpart in the code: the sixfold cover of GL5\mathrm{GL}_5 (timeline steps 3 and 16). A search for a cover of GL5\mathrm{GL}_5 finds nothing. The only name with GL5 is a one-line statement GL5LogLValueBoundByZeta. The only cover in this part of the code is the double cover of Sp2n(ℝ)\mathrm{Sp}_{2n}(\mathbb{R}).
  • No counterpart in the code: the statement "NN is automorphic" (timeline step 3). The code has no automorphic form built from the counting function NN of the note. The nearest result is the Siegel-Weil formula, which says that a theta integral on a unitary group is an Eisenstein series.
  • No counterpart in the code: the relative trace formula (timeline step 16). The name relTrace exists, but it is a trace of fields and is not related.
  • No counterpart in the code: a Cauchy-Schwarz bound on a projected Poincaré series (timeline step 16). Cauchy-Schwarz appears only as general lemmas. The Poincaré series of the code give a lower bound (a period is not zero), not an upper bound.
  • No counterpart in the code: the index of theta series (timeline step 18). The covolume part of step 18 corresponds to the finite volume in SiegelWeilUnitaryConvergent and to CovolumeBoundOfX. Congruences between lifts and forms are not on this page.
  • The folder Development/AbelianVarietyTheory/BlockTheory/PoincareCuspForms has 18 files and 390 declarations, but only 66 are in the spine and each has size at most 3. Its name does not describe its content. The Poincaré series of Step 10 are in other folders.
  • The literature inputs calls siegel_weil_std_of_literature a leaf citation. That word means a one-line theorem that calls a proof in another file. This page follows that proof (Steps 2 to 6).
  • The Galois row uses a separate Siegel-Weil statement, siegelWeilStd_k, with its own copies of the definitions (TPar_k, ThetaCoef_k, CTSplit_k). See the Galois row. This page does not compare the two.
  • This page does not unfold the proofs of WhitK, WhitEq, CTSplit, HolLead or AbsThm. It does not check that each statement matches the published theorem of the same name.
  • The Lean code is not reviewed by people. The site does not claim that the proof is correct. The trust report says what has been checked.

The source of this page. Markdown. Something is wrong? Report it.