Theta integrals and local matching
The Lean code takes a list of named inputs, two statements about theta integrals on unitary groups, a covolume bound, a local Langlands map and the statement VWin, and gets EveryCharacter from them. This page unfolds the two theta statements and follows the short branch that ends in L-function nonvanishing.
draft
Result
The trunk of the proof (see the map) reaches the theorem every_character_of_theta_integrals_local_matching from every_character_of_literature_of_vWin (see the literature inputs). The theorem has this form.
theorem every_character_of_theta_integrals_local_matching
(P : GlobalTheory.Fc.S0.UnitaryGroupThirdPlaceHypotheses)
(F : GlobalTheory.FunPr.ThetaLocalInputs)
(hA : GlobalTheory.IndefiniteRowSideFullHecke.WeilTensorModels.ThetaIntegralsAgree)
(hM : GlobalTheory.IndefiniteRowSideFullHecke.WeilTensorModels.LocalMatching)
(hX : GlobalTheory.CovTop.CovolumeBoundOfX)
(hL : UpperTheory.Zero.LLC.RecTExists) (hV : VWin) :
UpperTheory.EveryCharacter
The conclusion EveryCharacter is this statement. Let , let be a Dirichlet character modulo and let with . If or , then . Here is the Dirichlet L-function of Mathlib (DirichletCharacter.LFunction).
In words, the Lean code states: if seven inputs hold, then each Dirichlet L-function has no zero in the half-plane . The seven inputs are two large bundles of named statements (P and F), two statements on theta integrals (hA and hM), a covolume bound (hX), the existence of a local Langlands type map (hL) and the statement VWin (see the VWin page). In the proof of the Riemann Hypothesis, all seven come from the previous step.
Idea
The proof of the theorem has two parts, and it is three lines of Lean.
The first part is the trunk. The theorem nonempty_hyps_definite_side_of_closed_bound uses all inputs except VWin, and builds one large structure HypsAlt. This structure has 46 fields. They are bounds and choices for families of objects that the code calls FamP. This part dominates 263,615 declarations. It is the subject of the next chapter.
The second part is the branch every_character_of_green_v_win. It takes three inputs: the statement Green (about cuspidal characters of ), the statement VWin, and the structure HypsAlt. From them it proves ZeroFreeAll, which is the same nonvanishing for primitive characters only. Then a short Mathlib argument removes the condition "primitive". This branch dominates 12,070 declarations.
The theta inputs hA and hM are not used in the branch. They go only into the trunk part. So the name of the theorem refers to the two inputs, and the work that uses them is in the next chapter. This page still unfolds them, because their statements are concrete. ThetaIntegralsAgree says that two theta integrals into the same unitary group are proportional, with one nonzero constant, when their test functions match. LocalMatching says that test functions can be matched: at each place there are quotient maps of the two local Weil representations onto one common irreducible space.
Before you read
You need these ideas.
- A number field , a CM extension with conjugation , and the adele ring .
- A Hermitian form or skew-Hermitian form on and its unitary group.
- The Weil representation of a pair of unitary groups, and the theta correspondence.
- Schwartz functions on adelic spaces, Haar measure and a fundamental domain.
- Hecke characters of .
- Dirichlet characters, Dirichlet L-functions and the von Mangoldt function .
Plan
- Define the theta integral for one Weil model , as the Lean code does.
- Unfold
ThetaIntegralsAgree: convergence, one nonzero proportionality constant, and two recursion laws for that constant. - Unfold
LocalMatching: matched local quotient maps exist. - Build
HypsAltfrom the inputs (next chapter). - Get
GreenfromGreenRegularCharacters, a field ofP. - In the branch, prove
ZeroFreeAllfromGreen,VWinandHypsAlt. - Remove the condition "primitive" with
imprimitive.
The steps
Step 1. Theta integrals
All definitions are in the namespace GlobalTheory.IndefiniteRowSideFullHecke.WeilTensorModels, in the file WeilTensorModels.lean.
A space Spc F E n has a matrix , an -lattice , a matrix at each infinite place , and a basis at each finite place. The group UA σ J is the adelic unitary group with entries in . A Model σ X W for spaces of size and of size has Darboux coordinates for the symplectic form on , a vacuum vector at each finite place, a measure on and a set dom. A representation om of acts on functions on (the Lean index set is Fin n × Fin d).
def theta (om : Rep σ X W) (Φ : (Fin n × Fin d → AF F) → ℂ) (g : UA σ X.J) (h : UA σ W.J) : ℂ :=
thetaSum F (om (g, h) Φ)
def Model.I (M : Model σ X W) (om : Rep σ X W) (Φ : (Fin n × Fin d → AF F) → ℂ) (h : UA σ W.J) : ℂ :=
∫ g in M.dom, theta om Φ g h ∂M.μ
Here thetaSum is over the rational points (HilbertNinetyQuotientHomEqIff.lean#L107). So the theta function is , and the theta integral is
The predicate Model.IsGood asks that is a Haar measure and that dom is a fundamental domain for in . So is a function on . This is the usual global theta lift from to , written out by hand.
Step 2. The data of a triple
The two theta statements quantify over a "triple". The structure IsTriple B G X om om' (WeilTensorModels.lean#L388) has five parts.
B.IsGood: is a CM extension with conjugation , is Galois, is unramified, the additive character of is standard, and at each inert place the residue field has more than 3 elements and .G.IsGood: there are three spaces. is Hermitian of size and signature . is Hermitian of size and signature . is skew-Hermitian of size and signature . The sizes satisfy , is even, and . Also at each infinite place , where are the weights of two characters.X.IsGood: and are Hecke characters of with infinity types , with given restrictions to , unramified outside a finite set of places, and with ramified at each place of .weil,weil':omis a Weil representation of with splitting characters , andom'is one of with splitting characters .
So a triple gives two theta lifts into the same group : one from with , one from with .
Step 3. ThetaIntegralsAgree
def ThetaIntegralsAgree : Prop :=
(∀ F E ... σ B a a' G X om om', IsTriple B G X om om' → Decay B G om om' ∧
∀ R Rf, IsMatched B G X om om' R Rf → OneConst B G X om om' R Rf ∧ Coinv B G X om om' Rf) ∧
(∃ e₁ e₂ s : ℤˣ, ∀ F E ... σ, StepBeta σ e₁ e₂ s) ∧
∃ e₁ e₂ s : ℤˣ, ∀ F E ... σ, StepAlpha σ e₁ e₂ s
The statement is at WeilTensorModels.lean#L585. It has three parts.
The first part is about one triple. Decay says that is integrable on dom for each adelic Schwartz function . It says the same for and each in a subspace So, with a bound and moderate growth of . OneConst says three things: some is not zero, there is exactly one with IsConst c, and that is not zero. Here IsConst c means
Coinv says that at good places the kernels of the local maps are exactly the coinvariant spaces augAt, and that So is stable under .
The second and third parts, StepBeta and StepAlpha (WeilTensorModels.lean#L559), are recursion laws for the constant . Take a geometry and a smaller geometry where one space is smaller by 2 (one hyperbolic plane split off). Then there is a so that the constants and of the two triples satisfy a relation. In StepAlpha it is
with . In StepBeta the partial L-value multiplies instead, and the step factor is inverted. Here is the partial Euler product of over the places of not above (charL). The factor stepFac (WeilTensorModels.lean#L552) is a product of values of on two fractional ideals and, for each place , a value of times an epsilon factor built from a Gauss sum (eps, gauss).
Step 4. LocalMatching
def LocalMatching : Prop :=
∀ (F E : Type) ... (σ : E →ₐ[F] E) (B : Base F E) (a a' : ℕ)
(G : Geo σ a a') (X : Chars F E) om om', IsTriple B G X om om' → ∃ R Rf, IsMatched B G X om om' R Rf
The statement is at WeilTensorModels.lean#L581. For each triple there are real maps R and finite maps Rf with IsMatched.
IsFinMaps asks, for each finite place , two linear maps and from the local Schwartz-Bruhat spaces of and onto one space . The maps are onto. They kill the and coinvariants. They respect the action of : if , then the same holds after acts on both. The space is nontrivial and has no proper nonzero -stable subspace. Outside the two vacuum vectors have the same nonzero image. At the condition IsZetaAt relates the images to two explicit local integrals (bigP and smallP). IsRealMaps asks the same at the infinite places, with Lie algebra and maximal compact actions on the polynomial Fock space Pol.
So LocalMatching says: the local theta lifts of and to have a common irreducible quotient, at every place, with matched spherical vectors. A pair is matched (Match, PolMatch) when their pure tensor pieces have the same image in for some finite set , up to limits.
In the proof of the Riemann Hypothesis, both hA and hM come from theta_integrals_agree_and_local_matching_of_literature, which applies PositivityTheory.ThKR.theoremK_head2 to two statements named EisBL and SiegelWeilStd. That proof dominates 8,520 declarations. The names refer to Eisenstein series and the Siegel-Weil formula in the literature. This page does not unfold EisBL or SiegelWeilStd.
Step 5. The trunk part
The Lean statement is long. Here it is shortened: the part (...) stands for a condition SlotNormCondition on the family bound BoundAtTwoThreeFamilyFonAlt.
theorem nonempty_hyps_definite_side_of_closed_bound :
UnitaryGroupThirdPlaceHypotheses → ThetaLocalInputs → ThetaIntegralsAgree → LocalMatching →
CovolumeBoundOfX → ∀ (hL : RecTExists),
(∀ cDep : ℝ, LinkedAt (parametersOfDependence hL cDep) (...)) →
Nonempty (HypsAlt S0.NL DefiniteSideCondition)
The main proof passes P.nLink through boundAtTwoThree_linked_of_recT to supply the last hypothesis. The result is a term of HypsAlt N DefiniteSideCondition with N = S0.NL. By definition, S0.NL 𝒞 is Closed 𝒞 fun Fm _ => LinkN Fm (GeometricWeakTiesOfStrong.lean#L36). This is exactly the form that the branch asks for. In the proof, hA and hM go into test_vector_assembly_holds, fourier_coefficient_fit_of_join_alt, evaluation_grid_of_join, twisted_local_factors_match and twisted_square_theta_integrals_eq. The next chapter follows this proof.
Step 6. Green from regular characters
theorem green_of_regular (hG : GlobalTheory.Published.GreenRegularCharacters) : GlobalTheory.GLq.Green
GreenRegularCharacters is a field of P (GreenRegularCharacters.lean#L13). For a prime , an integer and a field of degree over , it gives a map from characters of to class functions on . For a regular (the powers , , are distinct), is cuspidal, has degree , and on an element with irreducible characteristic polynomial and a root it is . Two regular characters give the same exactly when they are Frobenius conjugate, and each cuspidal character is some . The name refers to Green's theorem on the characters of finite general linear groups. Green (LabelDatum.lean#L108) is the same kind of parametrisation, by "prime labels" (polynomials over ) in place of characters. The proof green_of_regular is in the file SumRangeEqRoots.lean#L225.
Step 7. The branch: ZeroFreeAll
theorem UpperTheory.Top.zero_free_all_of_green_v_win (hG : GlobalTheory.GLq.Green) (hV : UpperTheory.Top.VWin)
(h : Nonempty (HypsAlt (fun 𝒞 => Closed 𝒞 fun Fm _ => LinkN Fm) DefiniteSideCondition)) :
Formalism.HC.ZeroFreeAll :=
h.elim fun h => UpperTheory.Top.zeroFreeAll_of_topG (la := 3 / 2) (by norm_num)
(forall_goal_of_shape_of_hyps _ _ _ _ (shape_closed_link_n_of_green hG) trivial trivial h)
fun ψ => UpperTheory.Top.supplyChi ψ (hV ψ)
A PChar is a primitive Dirichlet character of some conductor . ZeroFreeAll says: for each PChar and each with , (ExplicitFormulaErrorOfHostConditions.lean#L18). The statement has no exception at . The trivial character modulo 1 is primitive, and its L-function is , so at the statement says that riemannZeta 1 is not zero. Mathlib gives a value at its pole by a convention, and the lemma riemannZeta_one says that this value is , which is not zero. So the case is true, but only because of the convention. The proof has three pieces.
First, shape_closed_link_n_of_green (9,550 declarations) proves TopShapeAlt True True N TopG from Green. TopShapeAlt V NR N G is a long implication: if about 50 named conditions on a family class hold (and , , and for a derived class), then holds. The theorem forall_goal_of_shape_of_hyps feeds the 46 fields of HypsAlt into these slots. It is a direct pass of fields, with two small conversions. The result is TopG.
Second, TopG (ZeroFreeAllOfTopG.lean#L12) says: for each and each requirement , if suitable hosts exist on windows of scale (HostExW la G.req), and each family in the class G.cls la satisfies HypV, then the class is FamilyClassConnected. That means: for small there is a family Fm in the class whose members satisfy the inequality ChainTied for all large in its window set.
Third, supplyChi (1,462 declarations) turns VWinAt ψ 401 into HostExW (3/2) (GChi ψ).req. So VWin supplies the hosts for each , at .
The theorem zeroFreeAll_of_topG (417 declarations) then ends the proof. For , it takes the level Level.cyc ψ, built from the subgroup generated by . The code defines
where the sum is over the nontrivial characters of the level, is a weight, is Euler's constant, and with the constant logDerivativeAtOne. Two proved facts combine. level_gm_le_on_family_scales (300 declarations) says that on the windows of a family where ChainTied holds, . formB (95 declarations) gives FormB: if for all , for arbitrarily large , then for and all in the level. The lemma level_of_tiedWin takes with , chooses so small that , and applies the two facts.
Step 8. Imprimitive characters
theorem UpperTheory.Top.imprimitive :
ZeroFreeAll → ∀ (N : ℕ) [NeZero N] (χ : DirichletCharacter ℂ N) (s : ℂ), 1 / 2 < s.re → (χ ≠ 1 ∨ s ≠ 1) →
DirichletCharacter.LFunction χ s ≠ 0
This is the last step of the branch (Imprimitive.lean#L5). Let be the primitive character that induces . Mathlib gives
The first factor is not zero by ZeroFreeAll. For each , , so the factor is not zero. The condition or is needed only for the Mathlib lemma LFunction_changeLevel. The theorem every_character_of_green_v_win is this step applied to zero_free_all_of_green_v_win.
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.
| Declaration | Module (file) | Size |
|---|---|---|
UpperTheory.Lit.every_character_of_theta_integrals_local_matching |
ThirdPlaceSeries/SiegelEisensteinSecondTermHypotheses.lean |
313,327 |
UpperTheory.Lit.nonempty_hyps_definite_side_of_closed_bound |
same file | 263,615 |
UpperTheory.Top.every_character_of_green_v_win |
ThirdPlaceSeries/TopPlug.lean |
12,070 |
UpperTheory.Top.zero_free_all_of_green_v_win |
ThirdPlaceSeries/TopPlug.lean |
12,065 |
UpperTheory.Top.shape_closed_link_n_of_green |
ThirdPlaceSeries/TopShape.lean |
9,550 |
UpperTheory.Top.supplyChi |
CharacterBoxes/SupplyChi.lean |
1,462 |
UpperTheory.Top.zeroFreeAll_of_topG |
CharacterBoxes/ZeroFreeAllOfTopG.lean |
417 |
UpperTheory.Top.imprimitive |
CharacterBoxes/Imprimitive.lean |
1 |
UpperTheory.LiteratureTheorems.theta_integrals_agree_and_local_matching_of_literature |
ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin.lean |
8,520 |
GlobalTheory.IndefiniteRowSideFullHecke.WeilTensorModels.ThetaIntegralsAgree |
BlockTheory/DivisorBase/WeilTensorModels.lean |
definition |
GlobalTheory.IndefiniteRowSideFullHecke.WeilTensorModels.LocalMatching |
BlockTheory/DivisorBase/WeilTensorModels.lean |
definition |
The ThirdPlaceSeries files are in Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/. The CharacterBoxes files are in Development/AutomorphicArithmetic/. The full closure of the main theorem (all declarations it uses, shared or not) has 452,823 declarations. The closure of every_character_of_green_v_win has 50,160. Besides the two big children, the main theorem dominates some smaller nodes that both children reach, for example theorem_A_closed_form_holds (3,015) and globalLatticeIndexTheorem_of_lengthBoundAtPlaces (2,223).
Notes
- The theorem is named after the theta inputs, but the branch
every_character_of_green_v_windoes not use them. They act only throughHypsAlt, which the next chapter builds. This page does not show howOneConst,StepAlphaorStepBetaare used there. UnitaryGroupThirdPlaceHypotheseshas about 60 fields andThetaLocalInputshas 53 fields. Most are named after published theorems (for examplejacquetShalika,casselmanWallach,howeDualityReal). This page uses onlygreenRegularCharacters,nLinkandlitNby name. It does not check that each field matches the published theorem of the same name. See the literature inputs for where these fields come from.- This page does not unfold
CovolumeBoundOfXpast its first level. It says that for families in the classFamHyp, for small and large parameters, a condition namedL3holds. RecTExistssays that for each number field , each finite place and each choice of a place above in an algebraic closure, there is a maprcfrom irreducible representations of to Weil-Deligne representations that satisfies the lawsLawsT(compatibility with twists, duals, central characters, cuspidal support, and more). This page does not list all the laws.TopShapeAlt,HypsAlt,DefiniteSideCondition,ChainTiedandLinkNare very long. This page names them and their role, and does not unfold their content.ZeroFreeAllhas no exception at for the trivial character modulo 1.EveryCharacterhas this exception. This page does not check how the chain treats the point for the trivial character.- The site does not claim that the proof is correct. The Lean code states the results above. The trust report says what has been checked.
The source of this page. Markdown. Something is wrong? Report it.