riemann

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

The literature layer

One file of the Lean code gathers about 190 results with names from the literature into a few large bundles. Only 13 of its 195 theorems use the argument that the code calls the open literature statements. This page shows how the file is built, and where its three large branches go.

draft

Result

Let 𝖤𝖢\mathsf{EC} be the statement UpperTheory.EveryCharacter of the overview. It says: for each modulus N≥1N \geq 1, each Dirichlet character χ\chi modulo NN and each complex ss with Re⁡(s)>12\operatorname{Re}(s) > \tfrac12 and (χ,s)≠(1,1)(\chi, s) \neq (1, 1), the Dirichlet L-function has L(s,χ)≠0L(s, \chi) \neq 0. This is the Generalized Riemann hypothesis for Dirichlet L-functions, in the form "no zeros to the right of the line".

Let 𝖮\mathsf{O} be the structure OpenLiteratureStatements and let 𝖵\mathsf{V} be the statement UpperTheory.Top.VWin. The Lean code proves

𝖮∧𝖵  ⟹  𝖤𝖢.\mathsf{O} \wedge \mathsf{V} \;\Longrightarrow\; \mathsf{EC}.
theorem UpperTheory.LiteratureTheorems.every_character_of_literature_of_vWin
    (O : OpenLiteratureStatements) (hV : UpperTheory.Top.VWin) : UpperTheory.EveryCharacter

The statement is at EveryCharacterOfLiteratureOfVWin.lean#L810. Its only caller is everyCharacter_of_row, one step below the top of the trunk.

Idea

The proof of this theorem is one application of a later theorem, every_character_of_theta_integrals_local_matching. That theorem needs seven inputs. Two of them are very large structures of hypotheses with 60 and 53 fields. Each field is a statement that the code names after a published result: the Hecke and Tate functional equation, the theorem of Deligne and Ribet, the Siegel and Weil formula, and so on.

The file fills these fields with 195 small theorems. Each one has the form NAME_of_literature (O : OpenLiteratureStatements) : P. The argument O is in the signature of every one of them. But 182 of the 195 theorems never read it. Most of them (152) are one line: they cite a theorem that the Lean code proves in another file, and they ignore O. Others (30) combine earlier theorems of the same file. Only 13 theorems read a field of O, directly or through other theorems.

The structure O has three fields. Two of them have the same type UpperTheory.One.M.Row, a statement on the existence of Galois representations. The third is a form of the local Langlands correspondence. At the call site, the Lean code fills the two Row fields with the same proof, and fills the third with a theorem that it proves. So the word "open" in the name is a label. In the final proof, nothing in O is left as a hypothesis.

Three branches carry most of the weight below this theorem: the main theorem of the next chapter (313,327 declarations in its dominator part), a volume bound (52,186) and a theta statement (8,520).

Before you read

Plan

  1. State the three fields of OpenLiteratureStatements.
  2. Show how the caller fills them.
  3. Sort the 195 theorems of the file into groups, and say which ones use O.
  4. Show how the nine bundle theorems collect the small theorems into structures.
  5. Show the proof of the main theorem as one call with seven inputs.
  6. Describe the two large branches that the file feeds: the theta branch and the volume branch.

The steps

Step 1. The three open statements

The structure is a Prop with three fields (L215):

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

The field names refer to the literature (a local-global compatibility result, the existence of the local Langlands correspondence, and Galois representations for polarized pairs). The types are what the Lean code uses. Two fields have the same type.

The type Row. It is Row0b with the predicate IsRecT put in (RecT.lean#L6). The definition Row0b (WeightsConjBis.lean#L69) says: for each CM field EE, each prime ℓ\ell, each field isomorphism ι ⁣:ℚ‾ℓ≅ℂ\iota \colon \overline{\mathbb{Q}}_\ell \cong \mathbb{C}, each a≥1a \geq 1, each cuspidal object π\pi of the type CuspSub E a and each Hecke character χ\chi of the maximal real subfield of EE, if π\pi is regular algebraic and (π,χ)(\pi, \chi) is a polarized pair, then there are a homomorphism r ⁣:Gal⁡(Ω/E)→GLa(ℚ‾ℓ)r \colon \operatorname{Gal}(\Omega/E) \to \mathrm{GL}_a(\overline{\mathbb{Q}}_\ell), an integer weight and a family of Hodge and Tate multisets with the properties PropsB. The fields of PropsB say: rr is continuous and semisimple, rr matches π\pi at places away from ℓ\ell (through each local map that satisfies IsRecT), the weights match the infinitesimal character of π\pi, the weights are symmetric under complex conjugation, and a condition AboveB holds at places above ℓ\ell. The chapter /proof/galois-row/ unfolds this statement.

The type RecTExists. It says (IsRecTOfIsRec.lean#L78): for each number field EE, each algebraic closure Ω\Omega, each valuation subring AA of Ω\Omega and each finite place ww of EE below AA, there is a map rc from irreducible representations of GLn(Ew)\mathrm{GL}_n(E_w) (all nn) to Weil and Deligne parameters, with IsRecT. The predicate IsRecT asks for a local Artin map aa and the 13 laws of LawsT: dimension, the case n=1n = 1 through aa, compatibility with the Langlands classification by segments, cuspidal goes to irreducible, twists by ∣det⁡∣s|\det|^s and by characters, central characters, duals, injectivity on "good" representations, tame parameters at depth zero, and a type law. This page does not unfold Good or TypeLaw.

Step 2. How the caller fills O

The caller is one line (MainConclusion.lean#L7):

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

In words: the two Row fields both get the same proof hh. The field localLanglandsExists gets recTExists, a theorem that the Lean code proves (its dominator part has 325 declarations). The statement 𝖵\mathsf{V} comes from hh too, by vwin_of_row. That step is the chapter /proof/vwin/. Then everyCharacter supplies hh from galoisLocalGlobalStrong_recParameter, which is the chapter /proof/galois-row/. So the Lean code does not leave 𝖮\mathsf{O} as an assumption of the final theorem.

Step 3. The 195 theorems and their use of O

Each of the 195 theorems in the file takes O. A script read each proof term and followed the references between the theorems of the file. The result is in this table.

Group Count What the proof does Uses O
Direct users 3 reads a field of O yes, directly
Indirect users 10 combines theorems of the file, and one of them reads O yes, through other theorems
Combiners and bundles 30 builds its result from earlier theorems of this file, by a theorem of another file or as a structure no
Leaf citations 151 is one line that cites a theorem proved in another file no
Trivial 1 siegel_eisenstein_simple_pole proves UnusedSlot72, which is True, by trivial no

The three direct users are these.

Theorem Line Conclusion Field of O
local_langlands_recT_exists_of_literature L283 RecTExists localLanglandsExists (it returns the field)
pair_u_of_literature L285 GalS.PairUAlt galoisRepOfPolarizedPairLit, also the first one
carU_local_global_of_literature L459 GalS.CarUAlt caraianiLocalGlobalDual

The ten indirect users are local_langlands_gl_recT_exists_of_literature, cusp_u_of_literature, hecke_l_adic_character_crystalline_of_literature, hecke_l_adic_character_above_of_literature, carU_above_l_of_literature, the bundles literature_local, literature_automorphy, literature_nonexistence and published_results_eigen, and the main theorem. The bundle literature_automorphy also reads O.caraianiLocalGlobalDual directly in its field car.

One indirect user has a form to note. The theorem local_langlands_gl_recT_exists_of_literature (L313) proves RecTExists by llOf_alt. The signature of llOf_alt is RecTExists → LocalConverseGL → … → BernsteinZelevinskyFiniteness → RecTExists. Its first input is RecTExists itself, from O. The output type is the same as the first input type. This page does not unfold the proof of llOf_alt.

The leaf citations cover many areas. The table sorts them by the names of their conclusions. The names come from the code. This page does not claim that each Lean statement is the published theorem of the same name. It unfolds three of them below, as samples.

Area (by name) Examples of conclusions Uses O
L-functions and class field theory HeckeTateFunctionalEquation, PartialZetaFunctionalEquation, DeligneRibet, GlobalArtinReciprocity, ChebotarevDegreeOne, Linnik, LogDerOne, HilbertClassFieldJ, RayClassFieldSt no
Representations of pp-adic GLn\mathrm{GL}_n BernsteinZelevinskyFiniteness, SegmentRepresentation, LanglandsExhaustionGL, LocalConverseGL, DepthZeroSupport, MoyPrasadJacquetIso, GreenRegularCharacters no
Real groups DiscreteSeriesExists, DiscreteSeriesClassification, HarishChandraFinite, CasselmanWallach, HolomorphicDiscreteSeries, KFiniteVectorsUnitary no
Automorphic forms CuspRapidDecay, CuspidalSpectrumGL, JacquetShalika, FlathTensorProduct, CuspidalMultiplicityOne, AutomorphicFormsFiniteGL no
Weil representation and theta WeilRepresentationExists, StoneVonNeumann, WeilSiegelFormulaUnitary, SiegelWeilStd, EisBL, GLHoweDuality, HoweDualityReal no
Galois cohomology and pp-adic Hodge theory PoitouTateNormalised, LocalTateDuality, GlobalEulerChar, BlochKatoLocal, FontaineBcris, BhattScholzeIntegral, DeRhamPotSemistable no
Arithmetic geometry ToroidalCompactification, KawamataViehweg, NadelVanishingAndNonKltDivisor, ArithHilbertSamuel, CycleHeights, FarguesExact, ZarhinStatement, ResolutionCharZero no

Three samples. The theorem heckeTateFunctionalEquation' (L223) is GlobalTheory.Top.heckeTateFE and ignores O. Its conclusion (HeckeTateFunctionalEquationCM.lean#L244) says: for a totally complex number field EE, a CM type φ\varphi, a unitary Hecke character χ\chi of EE with infinity type n≠0n \neq 0 relative to φ\varphi, and conductor exponents aa, there are functions Λ,Λ′\Lambda, \Lambda' that are completed L-functions of χ\chi and χ−1\chi^{-1}, with

Λ(s)=ε∞(n) εf(χ,a,s) Λ′(1−s)for all s∈ℂ.\Lambda(s) = \varepsilon_\infty(n) \, \varepsilon_f(\chi, a, s) \, \Lambda'(1 - s) \quad \text{for all } s \in \mathbb{C}.

This page does not unfold IsCompletedL, archEps (ε∞\varepsilon_\infty) or finEps (εf\varepsilon_f).

The theorem deligne_ribet_of_literature (L229) is PositivityTheory.PfDR.deligneRibet_holds and ignores O. Its conclusion DeligneRibet (SiegelKlingenDeligneRibet.lean#L45) is a congruence. For a totally real field FF, a prime ℓ\ell, n≥1n \geq 1, a function ε\varepsilon on ideals that is a ray function modulo ℓn\ell^n with values that satisfy InZl l, an ideal 𝔠\mathfrak{c} prime to ℓn\ell^n and k≥1k \geq 1: if rr is the value attached to ε\varepsilon and kk, and r′r' is the value attached to the translate of ε\varepsilon by 𝔠\mathfrak{c}, then r−N(𝔠)kr′r - N(\mathfrak{c})^k r' satisfies InZl l. This page does not unfold IsRayValue or InZl.

The theorem logDerivOne (L219) ignores O. Its conclusion LogDerOne (LogDerOne.lean#L12) says: there is a constant CC such that for each primitive Dirichlet character χ\chi modulo q>1q > 1,

∣L′(1,χ)L(1,χ)−Z(χ)∣≤Clog⁡(3q),\left| \frac{L'(1, \chi)}{L(1, \chi)} - Z(\chi) \right| \leq C \log(3q),

where Z(χ)Z(\chi) is zsum χ 0 1 (fun ρ => 1 / (1 - ρ)). This page does not unfold zsum.

Step 4. The nine bundles

Nine theorems of the file build structures. Each has the form theorem NAME (O) : S where field := ..., and each field is one of the small theorems applied to O. The one exception is the field car of literature_automorphy, which reads O directly.

Bundle Line Structure built Fields Uses O
literature_local L589 LocalRepresentationTheoryHypothesesAlt 19 yes, through llc
literature_automorphy L609 GlobalAutomorphicGaloisHypotheses' 16 yes
literature_cohomology L626 LitCoh 12 no
literature_nonexistence L639 NonEndoscopicHypothesesAlt 5 yes, through the first two bundles
published_results_analytic L645 GeometricGaloisRepCompatibility 12 no
published_results_local L658 NLink.Hypotheses₂ 7 no
published_results_eigen L666 UnitaryGroupThirdPlaceHypotheses 60 yes
published_results_first L727 ThetaLocalInputs 53 no
published_results_arithmetic L781 PublishedHodgeBundleTheorems 28 no

The names of the bundles refer to the literature, but the bundles hold no new mathematics. They are records of earlier results. The bundle published_results_eigen contains the next three bundles of the table as its fields litN, toHypothesesX and nLink.

Step 5. The proof of the main theorem

The proof is one term:

  every_character_of_theta_integrals_local_matching
    (published_results_eigen O) (published_results_first O)
    (theta_integrals_agree_and_local_matching_of_literature O).1
    (theta_integrals_agree_and_local_matching_of_literature O).2
    (topCovolSlim2_use_alt_alt (resolution_char_zero_of_literature O) ...
       (published_results_arithmetic O) (grothendieck_deformation_of_literature O))
    (local_langlands_gl_recT_exists_of_literature O) hV

The target theorem (SiegelEisensteinSecondTermHypotheses.lean#L69) has this signature:

theorem UpperTheory.Lit.every_character_of_theta_integrals_local_matching
    (P : UnitaryGroupThirdPlaceHypotheses) (F : ThetaLocalInputs)
    (hA : ThetaIntegralsAgree) (hM : LocalMatching) (hX : CovolumeBoundOfX)
    (hL : RecTExists) (hV : VWin) : UpperTheory.EveryCharacter

So the main theorem feeds seven inputs. PP is the 60-field bundle. FF is the 53-field bundle. hAhA and hMhM come from the theta branch. hXhX comes from the volume branch. hLhL is the local Langlands statement, through llOf_alt. hVhV is the hypothesis 𝖵\mathsf{V}, passed on with no change. Two inputs of the volume branch, hsUse_outright and moriwakiBig_outright, are theorems of other files. They are not theorems of this file.

The statement 𝖵\mathsf{V} is (ExistsPageN.lean#L49):

def UpperTheory.Top.VWin : Prop := ∀ ψ : PChar, VWinAt ψ 401

VWinAt ψ 401 is a statement about all ε\varepsilon with 0<ε<1520 < \varepsilon < \tfrac{1}{52} and windows [Y,Y2][Y, Y^2] in a range below XX, with the constant 401401 in an exponent. The chapter /proof/vwin/ unfolds it. This file does not use 𝖵\mathsf{V} for anything except to pass it on.

Step 6. The theta branch

The theorem theta_integrals_agree_and_local_matching_of_literature (L565) does not use O:

theorem theta_integrals_agree_and_local_matching_of_literature (O : OpenLiteratureStatements) :
    ThetaIntegralsAgree ∧ LocalMatching :=
  PositivityTheory.ThKR.theoremK_head2 (eis_bl_of_literature O) (siegel_weil_std_of_literature O)

The theorem theoremK_head2 (TheoremKOfMod.lean#L26) proves EisBL → SiegelWeilStd → ThetaIntegralsAgree ∧ LocalMatching. Both of its inputs are leaf citations: each one is one line that calls a theorem proved in another file. The chapter Siegel-Weil, the metaplectic representation and cusp forms follows the proof of SiegelWeilStd.

  • EisBL (HypothesesE.lean#L46) says: for a section ff of the right type on a doubled group, the Eisenstein series converges for Re⁡(s)>t\operatorname{Re}(s) > t for some tt, is holomorphic there, and near each point s0s_0 a power (s−s0)k(s - s_0)^k times it is a holomorphic family of functions of moderate growth.
  • SiegelWeilStd (HypothesesS.lean#L11) says: in a set of cases on two spaces (one Hermitian and one skew-Hermitian), the Eisenstein series at the point s=m−n2s = \tfrac{m - n}{2} equals a constant (11 or 22) times the average of a theta sum over a fundamental domain. This has the shape of a Siegel and Weil formula. This page does not unfold EisValueAt, thetaSum or IsDbl.
  • LocalMatching (WeilTensorModels.lean#L581) says: for each triple that satisfies IsTriple, there are data R, Rf with IsMatched.
  • ThetaIntegralsAgree (WeilTensorModels.lean#L585) is three statements: decay and a constant (OneConst, Coinv) for each matched triple, and two step relations StepBeta, StepAlpha that compare the constants for spaces of sizes b+2b + 2 and bb, with a nonzero factor CC and a value of a partial L-function.

The chapter /proof/theta-matching/ unfolds these. The dominator part of this branch has 8,520 declarations. Of these, theoremK_head2_of_mod holds 6,655 and siegel_weil_std_of_literature holds 1,652.

Step 7. The volume branch

The theorem topCovolSlim2_use_alt_alt (TopCovolSlim2Use.lean#L16) has 15 hypotheses and the conclusion CovolumeBoundOfX. Thirteen hypotheses come from theorems of this file, with names such as ResolutionCharZero, PoincareLelongInv, FarguesExact, ZarhinStatement, NefRR, SiegelToroidalIntegral, ToroidalForFanKS, ArDegree, GrothendieckDeformation, and the 28-field bundle PublishedHodgeBundleTheorems. None of them uses O. The other two are hsUse_outright and moriwakiBig_outright.

The proof builds two records HypothesesX and HypothesesH from the inputs, and then calls topCovolG_alt_alt with them and with deep_slim_use. In the dominator part (52,186 declarations), deep_slim_use holds 30,358 and topCovolG_alt_alt holds 12,516.

The conclusion is (LiftLambdaBound.lean#L60):

def CovolumeBoundOfX : Prop := Closed FamHyp fun Fm CH =>
  ∃ C0 : ℝ, 0 ≤ C0 ∧ EventuallyAtAllPoints Fm fun _ _ _ p => ∀ C, C0 ≤ C → (p.S false).L3 (p.N C CH)

In words: for each family Fm in the class that Closed FamHyp gives, there is C0≥0C_0 \geq 0 such that, for all large xx and all points pp of the family at xx that satisfy RowsTAlt, the property L3 holds for each constant C≥C0C \geq C_0. This page does not unfold Closed, FamHyp, L3 or the point data p.S, p.N.

Where it is in the code

Sizes are the sizes of the dominator parts: the declarations that only this node reaches from the top. The main theorem has 503,326.

Declaration Module Size
every_character_of_literature_of_vWin ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin 503,326
every_character_of_theta_integrals_local_matching ThirdPlaceSeries/SiegelEisensteinSecondTermHypotheses 313,327
topCovolSlim2_use_alt_alt ThirdPlaceSeries/TopCovolSlim2Use 52,186
theta_integrals_agree_and_local_matching_of_literature ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin 8,520
GlobalTheory.TorKSField.toroidalForFanKS (cited by toroidal_for_fan_ks_of_literature) 5,077
published_results_arithmetic ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin 5,089
siegel_toroidal_integral_of_literature ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin 3,656
published_results_eigen ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin 3,017
published_results_first ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin 2,346
OpenLiteratureStatements ThirdPlaceSeries/EveryCharacterOfLiteratureOfVWin 1

The module prefix is Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/. The file itself has 814 lines, of which 208 are imports.

Notes

  • The word "literature". The file names its theorems after published results. The Lean code does not take any of them as an axiom here. Each leaf theorem cites a proof in another file. Whether each Lean statement matches the published theorem is a separate question. This page checked only the three samples in step 3, and only for their shape. It did not check side conditions.
  • The argument O is mostly unused. 182 of the 195 theorems do not read it. It is in their signatures only. The count comes from a script that reads the proof terms of the file and follows references inside the file. A theorem that cites another file is counted as "no", because the cited theorems do not take O.
  • Two fields of one type. The fields caraianiLocalGlobalDual and galoisRepOfPolarizedPairLit both have the type Row, and the caller fills both with the same proof. The different names do not mean different statements in the Lean code.
  • A loop in a signature. llOf_alt takes RecTExists and returns RecTExists. With recTExists proved elsewhere, this is not a gap. But this page does not know why the code takes this path, and it does not unfold llOf_alt.
  • UnusedSlot72. One field of published_results_eigen, siegelEisensteinSimplePole, has the type UnusedSlot72, which is True. The name suggests a Siegel Eisenstein simple pole result, but the Lean statement is empty.
  • Not unfolded. This page does not unfold CuspSub, PropsB.above, Good, TypeLaw, IsCompletedL, archEps, finEps, IsRayValue, InZl, zsum, EisValueAt, thetaSum, IsTriple, IsMatched, StepBeta, StepAlpha, Closed, FamHyp, L3, PChar or VWinAt in full. The later chapters treat some of them.
  • Correctness. This page tells what the Lean code states. It does not say that the proof is correct. See the trust report.

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