riemann

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

From the row statement to every character

The Lean code proves EveryCharacter from one statement, Row0b. This page says what Row0b, VWin, RecTExists and OpenLiteratureStatements state, how the chain uses them, and which chapter proves each input.

draft

Result

Let N≥1N \geq 1, let χ\chi be a Dirichlet character modulo NN, and let L(s,χ)L(s, \chi) be its Dirichlet L-function. The overview calls the statement below EveryCharacter:

Re⁡(s)>12 and (χ≠1 or s≠1)  ⟹  L(s,χ)≠0.\operatorname{Re}(s) > \tfrac12 \ \text{and} \ (\chi \neq 1 \ \text{or} \ s \neq 1) \implies L(s, \chi) \neq 0 .

This page is about one theorem of the Lean code:

Row0b(IsRecT)  ⟹  EveryCharacter.\texttt{Row0b}(\texttt{IsRecT}) \implies \texttt{EveryCharacter} .

Here Row0b(IsRecT)\texttt{Row0b}(\texttt{IsRecT}) is a statement about ℓ\ell-adic Galois representations attached to cuspidal automorphic representations of GLa\mathrm{GL}_a over CM fields. The Lean code proves it in a separate theorem, galoisLocalGlobalStrong_recParameter. The two theorems together give everyCharacter.

Idea

The theorem everyCharacter is one line. It applies everyCharacter_of_row to a proof of Row0b. All the work is below these two names.

The theorem everyCharacter_of_row takes one hypothesis hh, a proof of Row0b(IsRecT)\texttt{Row0b}(\texttt{IsRecT}). It uses hh two times.

  1. It builds a value of the structure OpenLiteratureStatements. This structure has three fields. Two fields are hh itself. The third field is RecTExists, which says that a local Langlands map with a list of properties exists at each place of each number field. The theorem recTExists proves it.
  2. It gives hh to vwin_of_row, which proves a statement called VWin. VWin is a bound on lengths of Tate-Shafarevich type groups of certain abelian number fields. It is stated for every primitive Dirichlet character.

Then every_character_of_literature_of_vWin takes the structure and VWin and gives EveryCharacter. Its proof makes seven inputs from the structure and passes them to every_character_of_theta_integrals_local_matching. That theorem calls nonempty_hyps_definite_side_of_closed_bound, which shows that a certain type of hypotheses is not empty. Then every_character_of_green_v_win turns this, with VWin, into the statement that L(s,ψ)L(s, \psi) has no zero with Re⁡(s)>12\operatorname{Re}(s) > \tfrac12 for each primitive character ψ\psi. A last step passes from primitive characters to all characters.

So the trunk is a chain of five theorems. Each link is short. Each large input is a separate theorem with its own chapter.

Before you read

Plan

  1. everyCharacter applies everyCharacter_of_row to galoisLocalGlobalStrong_recParameter.
  2. everyCharacter_of_row builds OpenLiteratureStatements from hh and recTExists, and gets VWin from vwin_of_row h.
  3. every_character_of_literature_of_vWin turns the structure into seven inputs and calls the next theorem.
  4. every_character_of_theta_integrals_local_matching makes a nonempty type of hypotheses and calls every_character_of_green_v_win.
  5. every_character_of_green_v_win proves the zero-free statement for primitive characters and then for all characters.

The steps

Step 1. The top line

theorem UpperTheory.FinalTheorems.everyCharacter : UpperTheory.EveryCharacter :=
  UpperTheory.FinalTheorems.everyCharacter_of_row UpperTheory.Gal.Int.galoisLocalGlobalStrong_recParameter

The proof is a function application. The argument galoisLocalGlobalStrong_recParameter (RowSrc.lean#L6) has type Row0b recT. Here recT is the predicate fun h rc => IsRecT h rc (Development.lean#L18). This is the same predicate that everyCharacter_of_row asks for. The chapter The Galois row explains this proof.

Step 2. The hypothesis Row0b

Row0b takes a parameter Rc of type RecNotion. A RecNotion is a predicate on maps rc from isomorphism classes of irreducible smooth representations of GLn(Ew)\mathrm{GL}_n(E_w) (all nn together, the type IrrGL) to isomorphism classes of Weil-Deligne representations (the type WD h). Here EwE_w is the completion of a number field EE at a finite place ww, and h : IsPlaceOver A w says that a valuation subring AA of an algebraic closure Ω\Omega of EE lies over ww.

def Row0b : Prop :=
  ∀ (E Ω : Type) [Field E] [NumberField E] [IsCMField E] [Field Ω] [Algebra E Ω] [IsAlgClosure E Ω]
    (l : ℕ) [Fact l.Prime] (ι : Qbar l ≃+* ℂ) (a : ℕ), 1 ≤ a →
    ∀ (π : CuspSub E a) (χ : HChar (maximalRealSubfield E)), (CG.mk π).IsRegAlg → GlobalTheory.ContraPair.PolarizedPair π χ →
      ∃ (r : G E Ω →* GL (Fin a) (Qbar l)) (wt : ℤ) (HT : (E →+* Qbar l) → Multiset ℤ), PropsB Rc π ι r wt HT

In words. Let EE be a CM number field with algebraic closure Ω\Omega. Let ℓ\ell be a prime, let ι:ℚ‾ℓ≅ℂ\iota : \overline{\mathbb{Q}}_\ell \cong \mathbb{C}, and let a≥1a \geq 1. Let π\pi be an irreducible space of cusp forms on GLa\mathrm{GL}_a over EE (the structure CuspSub). Let χ\chi be a Hecke character of the maximal totally real subfield E+E^+. Suppose that the infinity type of π\pi is regular algebraic (IsRegAlg). Suppose that (π,χ)(\pi, \chi) is a polarized pair: χv(−1)=(−1)a\chi_v(-1) = (-1)^a at each infinite place vv of E+E^+, and the dual of π\pi is the complex conjugate of π\pi up to a twist by χ∘NE/E+\chi \circ \mathrm{N}_{E/E^+} (PolarizedPair, DualIsConjByOne.lean#L53). Then there are a homomorphism r:Gal(Ω/E)→GLa(ℚ‾ℓ)r : \mathrm{Gal}(\Omega/E) \to \mathrm{GL}_a(\overline{\mathbb{Q}}_\ell), an integer wtwt and a function HTHT from embeddings τ:E→ℚ‾ℓ\tau : E \to \overline{\mathbb{Q}}_\ell to multisets of integers, with the properties PropsB (WeightsConjBis.lean#L54).

The six fields of PropsB say this.

  • cont: rr is continuous.
  • ss: rr is semisimple (IsSSRep).
  • away: at each finite place ww not above ℓ\ell, for each map rc with Rc h rc and each local component σ\sigma of π\pi at ww, the Weil-Deligne representation of rr at ww and the value rc σ satisfy AwayB. AwayB says that the two have the same dimension. It also says that they are equal under the conditions Tor and mono0, and that under further conditions (IsQpBig, TmW, Sqf) their characteristic polynomials agree on elements of degree 0.
  • weights: for each τ\tau, the multiset HT(τ)HT(\tau), read in ℂ\mathbb{C}, is {a−12−z}\{ \tfrac{a-1}{2} - z \}, where zz runs over the infinitesimal character of π\pi at ι∘τ\iota \circ \tau.
  • weights_conj: HT(τ∘c)={wt−n:n∈HT(τ)}HT(\tau \circ c) = \{ wt - n : n \in HT(\tau) \}, where cc is complex conjugation of EE.
  • above: at places above ℓ\ell, the conditions of AboveB. They use the predicates IsCrysAt, IsdRAt, IsPstAt and IsJumpsAt on the restriction of rr to a decomposition group.

The page does not unfold Tor, mono0, TmW, Sqf, IsCrysAt, IsdRAt, IsPstAt or IsJumpsAt. The names point to crystalline, de Rham and potentially semistable conditions, but this page has not checked the definitions.

The trunk uses Row0b with Rc = IsRecT. The code also gives this instance a name:

def UpperTheory.One.M.Row : Prop := UpperTheory.GalStar.Row0b fun h rc => UpperTheory.Zero.LLC.IsRecT h rc

So One.M.Row and the type of the hypothesis of everyCharacter_of_row are the same proposition (RecT.lean#L6).

Step 3. The predicate IsRecT

def IsRecT (rc : IrrGL (w.adicCompletion E) → WD h) : Prop := ∃ a ha, LawsT (h := h) a ha rc

In words, IsRecT h rc says that there is a homomorphism aa from the Weil group at ww to Ew×E_w^\times, with the property IsLocalArtin, such that rc satisfies the laws LawsT (IsRecTOfIsRec.lean#L55). The laws are these.

  • dim: rc sends a representation of GLn\mathrm{GL}_n to a Weil-Deligne representation of dimension nn.
  • one: on a smooth character χ\chi of Ew×=GL1(Ew)E_w^\times = \mathrm{GL}_1(E_w), rc gives the character χ∘a\chi \circ a.
  • langlands: rc is compatible with the Langlands classification by segments (IsL, segSum).
  • cusp: a cuspidal class goes to an irreducible Weil-Deligne representation.
  • nu: the twist by ∣det⁡∣s|\det|^s goes to the twist lpTw s.
  • absCentral: the absolute value of the determinant of rc x at σ\sigma equals ∣ω(a(σ))∣|\omega(a(\sigma))|, where ω\omega is the central character of xx.
  • twist, central, dual, inj, depth: compatibility with twists, central characters and contragredients, injectivity, and depth zero goes to tame. These five hold only for classes xx with the property Good.
  • type: the law TypeLaw, which fixes the inertia polynomial of rc on some depth-zero cuspidal types over fields E′E' with Ew≅ℚgE_w \cong \mathbb{Q}_g.

The property Good x says that the cuspidal support of xx contains only characters, or twists of cuspidal representations with a nonzero vector fixed by the first congruence subgroup, over a field isomorphic to ℚg\mathbb{Q}_g (IsRecTOfIsRec.lean#L20). So IsRecT is a list of properties of a local Langlands map for GLn\mathrm{GL}_n. The list does not ask for a bijection, and it does not mention ε\varepsilon-factors. The same file has a stronger predicate IsRec with an ε\varepsilon family, and the theorem isRecT_of_isRec shows that IsRec gives IsRecT (with the extra input DepthZeroTypeInertia).

Step 4. The structure and RecTExists

structure UpperTheory.LiteratureTheorems.OpenLiteratureStatements : Prop where
  caraianiLocalGlobalDual : _root_.UpperTheory.One.M.Row
  localLanglandsExists : _root_.UpperTheory.Zero.LLC.RecTExists
  galoisRepOfPolarizedPairLit : _root_.UpperTheory.One.M.Row

The structure has three fields. The first and the third have the same type, One.M.Row, which is Row0b IsRecT. The second has type RecTExists:

def RecTExists : Prop :=
  ∀ (E Ω : Type) [Field E] [NumberField E] [Field Ω] [Algebra E Ω] [IsAlgClosure E Ω] (A : ValuationSubring Ω)
    (w : HeightOneSpectrum (𝓞 E)) (h : IsPlaceOver A w), ∃ rc, IsRecT h rc

In words: for each number field EE, each algebraic closure Ω\Omega, and each finite place ww with a valuation subring AA over it, there is a map rc with IsRecT h rc.

The field names refer to the literature (a local-global result with the name Caraiani, the existence of local Langlands, and Galois representations of polarized pairs). The Lean types are the ones above. This page does not claim that a published theorem has exactly these statements. In the Lean code, no field is an axiom. The theorem everyCharacter_of_row fills all three fields with proofs:

theorem UpperTheory.FinalTheorems.everyCharacter_of_row (h : UpperTheory.GalStar.Row0b fun h rc => UpperTheory.Zero.LLC.IsRecT h rc) : UpperTheory.EveryCharacter :=
  UpperTheory.LiteratureTheorems.every_character_of_literature_of_vWin ⟨h, UpperTheory.Zero.LLC.recTExists, h⟩ (UpperTheory.One.vwin_of_row h)

The anonymous constructor ⟨h, recTExists, h⟩ puts hh in the first and third fields.

The theorem recTExists (RecTExists.lean#L20) proves RecTExists. The Lean proof does this. It takes a local Artin map aa from artEx. It takes a map r on cuspidal classes from cuspEx. It extends r to all classes with exists_isLExt, through the Langlands classification. Then it checks each law of LawsT with the lemmas extDim, extTw, extDual, extInj, extDepth and typeIff. Its dependency closure has 11,316 declarations, which is small for this proof. This page does not go below it. No later chapter in the list above covers it in detail.

Step 5. The statement VWin

def VWin : Prop := ∀ ψ : PChar, VWinAt ψ 401

A PChar is a primitive Dirichlet character ψ\psi modulo some N≥1N \geq 1. VWinAt ψ Ct (ExistsPageN.lean#L44) unfolds as follows. Let 0<ε<1520 < \varepsilon < \tfrac{1}{52}. Let e:ℝ→ℕe : \mathbb{R} \to \mathbb{N} satisfy x2+8ε≤2e(x)≤2x2+8εx^{2+8\varepsilon} \leq 2^{e(x)} \leq 2x^{2+8\varepsilon} for x≥1x \geq 1. Then there are J≥1J \geq 1 and X0X_0 such that for each X≥X0X \geq X_0 and each pair of numbers (q,d)(q, d) with WinPair ε (X^(1/J)) q d, there is YY with X1/J≤YX^{1/J} \leq Y and Y2≤XY^2 \leq X, and for each x∈[Y,Y2]x \in [Y, Y^2] the statement VAt ψ ε Ct q d (e x) x holds.

WinPair ε Y q d says: qq is a prime with q≡3(mod8)q \equiv 3 \pmod 8 and 12Yε/5<log⁡q≤Yε/5\tfrac12 Y^{\varepsilon/5} < \log q \leq Y^{\varepsilon/5}; dd is a prime with d≡7(mod16)d \equiv 7 \pmod{16} and 12Yε/5(log⁡Y)2<log⁡d≤Yε/5(log⁡Y)2\tfrac12 Y^{\varepsilon/5}(\log Y)^2 < \log d \leq Y^{\varepsilon/5}(\log Y)^2; q∣d+1q \mid d+1; and each odd prime p≤Yε/5p \leq Y^{\varepsilon/5} divides d+1d + 1.

VAt ψ ε Ct q d e x says: for each number field EE with IsChiField ψ e q d E, the bound VVle E ⌊x^Ct⌋ (√(exp(x^ε))) holds. IsChiField says that EE embeds in a cyclotomic field ℚ(ζM)\mathbb{Q}(\zeta_M) with 2e+22^{e+2}, the modulus of ψ\psi, qq and dd all dividing MM, and that EE is the fixed field of the classes a mod Ma \bmod M with a≡±1(mod2e+2)a \equiv \pm 1 \pmod{2^{e+2}}, ψ(a)=1\psi(a) = 1, (aq)=1\left(\tfrac{a}{q}\right) = 1 and (ad)=1\left(\tfrac{a}{d}\right) = 1.

VVle E t B (ExistsPageN.lean#L12) says: for each algebraic closure Ω\Omega and each choice of data LData E Ω p for each prime pp, the lengths len⁡ Ш(Dp,e′)\operatorname{len}\,\text{Ш}(D_p, e') are finite for 0<e′<t0 < e' < t, and for each finite set SS of primes

∑ℓ∈Slog⁡ℓ∑e′=1t−1len⁡ℤℓШ(Dℓ,e′)≤B.\sum_{\ell \in S} \log \ell \sum_{e'=1}^{t-1} \operatorname{len}_{\mathbb{Z}_\ell} \text{Ш}(D_\ell, e') \leq B .

Here Ш(D,e′)\text{Ш}(D, e') is the submodule Sha D e' of H1H^1 of ℚℓ/ℤℓ\mathbb{Q}_\ell/\mathbb{Z}_\ell with the action through the (−e′)(-e')-th power of a character chi of a Galois group unramified outside the places above ℓ\ell. The field hchi of LData ties chi to the map cycZ. The page does not unfold Sha1, cycZ or locFam.

The theorem vwin_of_row proves VWin from hh:

theorem UpperTheory.One.vwin_of_row (h : UpperTheory.GalStar.Row0b fun h rc => UpperTheory.Zero.LLC.IsRecT h rc) : UpperTheory.Top.VWin :=
  UpperTheory.One.Join.automorphic_window_of_inputs (UpperTheory.LiteratureTheorems.OpenImageCondition.mk h) UpperTheory.BaseChangeIdentities.baseChangeN h

It wraps hh as an OpenImageCondition, adds the theorem baseChangeN, and calls automorphic_window_of_inputs. The chapter The window statement explains it.

Step 6. From the structure to seven inputs

theorem UpperTheory.LiteratureTheorems.every_character_of_literature_of_vWin (O : OpenLiteratureStatements) (hV : UpperTheory.Top.VWin) : UpperTheory.EveryCharacter

The proof (shortened here) is one application of every_character_of_theta_integrals_local_matching to seven arguments:

  • P := published_results_eigen O, of type UnitaryGroupThirdPlaceHypotheses;
  • F := published_results_first O, of type ThetaLocalInputs;
  • hA and hM, the two parts of theta_integrals_agree_and_local_matching_of_literature O, of types ThetaIntegralsAgree and LocalMatching;
  • hX, of type CovolumeBoundOfX, from topCovolSlim2_use_alt_alt applied to fifteen theorems;
  • hL := local_langlands_gl_recT_exists_of_literature O, of type RecTExists;
  • hV, which is VWin.

The file has many helper theorems of the form NAME_of_literature (O : OpenLiteratureStatements) : T. Most of them do not use O. They are theorems of the Lean code with an unused argument. Only four places in the file read a field of O: O.localLanglandsExists (line 284), O.galoisRepOfPolarizedPairLit (line 286), and O.caraianiLocalGlobalDual (lines 460 and 618). So all three fields reach the seven inputs. The chapter The literature structure follows these paths, and explains topCovolSlim2_use_alt_alt and theta_integrals_agree_and_local_matching_of_literature.

Step 7. From the seven inputs to EveryCharacter

theorem every_character_of_theta_integrals_local_matching (P : UnitaryGroupThirdPlaceHypotheses) (F : ThetaLocalInputs) (hA : ThetaIntegralsAgree)
    (hM : LocalMatching) (hX : CovolumeBoundOfX) (hL : RecTExists) (hV : VWin) :
    UpperTheory.EveryCharacter :=
  every_character_of_green_v_win (green_of_regular P.greenRegularCharacters) hV
    (nonempty_hyps_definite_side_of_closed_bound P F hA hM hX hL fun cDep => boundAtTwoThree_linked_of_recT P.nLink (parametersOfDependence hL cDep))

(Namespaces are cut here; the full text is at SiegelEisensteinSecondTermHypotheses.lean#L69.) The proof makes three things.

  1. A proof of Green from the field greenRegularCharacters of P.
  2. For each real number cDep, a linking condition, from the field nLink of P.
  3. With these, nonempty_hyps_definite_side_of_closed_bound gives a value of type Nonempty (HypsAlt S0.NL DefiniteSideCondition).

Then every_character_of_green_v_win (TopPlug.lean#L16) uses Green, VWin and the nonempty type. It first proves ZeroFreeAll: for each primitive character ψ\psi and each ss with Re⁡(s)>12\operatorname{Re}(s) > \tfrac12, L(s,ψ)≠0L(s, \psi) \neq 0. Here VWin enters character by character, as supplyChi ψ (hV ψ). Then UpperTheory.Top.imprimitive (Imprimitive.lean#L5) passes from ψ\psi to an arbitrary character χ\chi through the primitive character that induces χ\chi. The chapter Theta integrals and local matching explains these two theorems.

The theorem nonempty_hyps_definite_side_of_closed_bound (SiegelEisensteinSecondTermHypotheses.lean#L44) has the same six inputs and the linking condition. Its tactic proof builds a list of intermediate facts (separation of functionals, a rationality statement RowRat, a Fourier coefficient fit, two grid constants CF0 and CF1, and open and linked parameter statements). It ends with hypotheses_nonempty_of_unramified_exponents_alt. The chapter The unitary hypotheses explains it.

Where it is in the code

All paths are relative to lean/src. "Closure" is the number of declarations that the name uses, directly or not (rh graph -count closure). "Dominated" is the size from the spine data: the number of declarations that the proof reaches only through this name. The page gives it where the spine data lists it.

Declaration Module Closure Dominated
FinalTheorems.everyCharacter MainConclusion.lean#L10 757,477
FinalTheorems.everyCharacter_of_row MainConclusion.lean#L7
Gal.Int.galoisLocalGlobalStrong_recParameter RowSrc.lean#L6 181,653 51,591
GalStar.Row0b WeightsConjBis.lean#L69 707
Zero.LLC.IsRecT, RecTExists IsRecTOfIsRec.lean#L76 152 (RecTExists)
Zero.LLC.recTExists RecTExists.lean#L20 11,316
One.vwin_of_row VwinOfRow.lean#L4 164,523 55,805
Top.VWin ExistsPageN.lean#L49 38
LiteratureTheorems.OpenLiteratureStatements EveryCharacterOfLiteratureOfVWin.lean#L215
LiteratureTheorems.every_character_of_literature_of_vWin EveryCharacterOfLiteratureOfVWin.lean#L810 634,302
SelfIntersectionVolumeEstimates.topCovolSlim2_use_alt_alt TopCovolSlim2Use.lean#L16 147,778 52,186
LiteratureTheorems.theta_integrals_agree_and_local_matching_of_literature EveryCharacterOfLiteratureOfVWin.lean#L565 38,926 8,520
Lit.every_character_of_theta_integrals_local_matching SiegelEisensteinSecondTermHypotheses.lean#L69 452,823
Lit.nonempty_hyps_definite_side_of_closed_bound SiegelEisensteinSecondTermHypotheses.lean#L44 440,406
Top.every_character_of_green_v_win TopPlug.lean#L16 50,160 12,070

The whole proof of riemannHypothesis has 757,483 declarations in its closure. The spine data puts eight names on the trunk, from riemannHypothesis down to nonempty_hyps_definite_side_of_closed_bound. The largest single branch below the trunk is hypotheses_nonempty_of_unramified_exponents_alt, with 88,131 dominated declarations.

Notes

  • everyCharacter_of_row has one real hypothesis, Row0b(IsRecT)\texttt{Row0b}(\texttt{IsRecT}). The other ingredient, RecTExists, is a theorem. So the name OpenLiteratureStatements does not mean that the statements are open in the Lean code. All three fields get proofs.
  • The structure has two fields with the same type. The Lean code fills both with the same proof hh. The two names point to two different sources in the literature, but in Lean they are one proposition.
  • This page does not check that Row0b matches a published theorem. The names caraianiLocalGlobalDual and galoisRepOfPolarizedPairLit refer to the literature. The Lean statement is the one in Step 2, with the conditions AwayB and AboveB in the special form of this code.
  • IsRecT is weaker than a full local Langlands correspondence in some ways. Several laws hold only on the class Good, there is no surjectivity, and there are no ε\varepsilon-factors. The page does not say whether this matters for the proof.
  • The page does not unfold Tor, mono0, TmW, Sqf, IsCrysAt, IsdRAt, IsPstAt, IsJumpsAt, Sha1, cycZ, locFam, IsLocalArtin, TypeLaw in full, Green, HypsAlt, DefiniteSideCondition, S0.NL, UnitaryGroupThirdPlaceHypotheses, ThetaLocalInputs, ThetaIntegralsAgree, LocalMatching and CovolumeBoundOfX. Later chapters must unfold the ones they use.
  • The type that nonempty_hyps_definite_side_of_closed_bound gives is written HypsAlt S0.NL DefiniteSideCondition. The type that every_character_of_green_v_win asks for is written HypsAlt (fun 𝒞 => Closed 𝒞 fun Fm _ => LinkN Fm) DefiniteSideCondition. Lean accepts the proof, so the two types are equal by definition. The page has not unfolded S0.NL to see this.
  • The constant 401 in VWin and the bound ε<152\varepsilon < \tfrac{1}{52} come from the code. The page does not explain where they come from.
  • The closure and dominator counts come from the syntactic graph tool rh graph and the spine data. They count declarations, not lines.

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