riemann

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

The window bound VWin

The Lean code proves VWin from the row hypothesis Row0b. VWin bounds a weighted sum of lengths of Tate-Shafarevich groups over fields cut out by a primitive character and two auxiliary primes. This page says what VWin states and how the 55,805 declarations below vwin_of_row are put together.

draft

Result

This chapter is one branch of the trunk. The theorem everyCharacter_of_row (MainConclusion.lean#L7) needs two things from the row hypothesis: a structure of literature statements, and the statement VWin. This branch gives VWin.

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

In words: if the row hypothesis Row0b holds (for local parameters that satisfy IsRecT), then VWin holds. The Galois row chapter explains Row0b. The rest of this section says what VWin states.

Fix notation. A PChar ψ\psi is a primitive Dirichlet character of some conductor NψN_\psi. For an odd prime ll, a number field EE and an integer ee, let H1H^1 be continuous Galois cohomology of the Galois group that the code calls GS (attached to the finite set SS of places of EE above ll), with coefficients ℚl/ℤl\mathbb{Q}_l/\mathbb{Z}_l twisted by χ−e\chi^{-e}. Here χ\chi is a character to ℤl×\mathbb{Z}_l^\times whose pullback to the absolute Galois group is the ll-adic cyclotomic character. The code takes the Tate-Shafarevich submodule Sha D e of this H1H^1 (classes that are locally trivial at the chosen places in SS), and shaLen D e is its length as a ℤl\mathbb{Z}_l-module, a value in ℕ∪{∞}\mathbb{N} \cup \{\infty\}. Write ℓl(E,e)\ell_l(E, e) for it.

The statement VVle E t B (ExistsPageN.lean#L12) says: for each choice of these data, ℓp(E,e)<∞\ell_p(E, e) < \infty for each prime pp and 0<e<t0 < e < t, and for each finite set SS of primes

∑l∈Slog⁡l∑e=1t−1ℓl(E,e)≤B.\sum_{l \in S} \log l \sum_{e=1}^{t-1} \ell_l(E, e) \le B .

The statement IsChiField ψ e q d E says that EE embeds in a cyclotomic field ℚ(ζN)\mathbb{Q}(\zeta_N) with 2e+2,Nψ,q,d∣N2^{e+2}, N_\psi, q, d \mid N, as the fixed field of the classes a∈(ℤ/N)×a \in (\mathbb{Z}/N)^\times with a≡±1(mod2e+2)a \equiv \pm 1 \pmod{2^{e+2}}, ψ(a)=1\psi(a) = 1 and Jacobi symbols (aq)=(ad)=1\left(\tfrac{a}{q}\right) = \left(\tfrac{a}{d}\right) = 1.

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

def VWinAt (ψ : PChar) (Ct : ℝ) : Prop :=
  ∀ eps : ℝ, 0 < eps → eps < 1 / 52 → ∀ e : ℝ → ℕ, (∀ x ≥ (1 : ℝ), PageN eps x (e x)) →
    ∃ J : ℝ, 1 ≤ J ∧ ∃ X0 : ℝ, ∀ X ≥ X0, ∀ q d : ℕ, WinPair eps (X ^ (1 / J)) q d →
      ∃ Y : ℝ, X ^ (1 / J) ≤ Y ∧ Y ^ (2 : ℝ) ≤ X ∧ ∀ x : ℝ, Y ≤ x → x ≤ Y ^ (2 : ℝ) → VAt ψ eps Ct q d (e x) x

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

So VWin states this. Let ψ\psi be a primitive character, let 0<ε<1520 < \varepsilon < \tfrac{1}{52}, and let e(x)e(x) be any integer function with x2+8ε≤2e(x)≤2x2+8εx^{2+8\varepsilon} \le 2^{e(x)} \le 2x^{2+8\varepsilon} for x≥1x \ge 1 (this is PageN). Then there are J≥1J \ge 1 and X0X_0 such that for each X≥X0X \ge X_0 and each pair (q,d)(q, d) with WinPair ε (X^{1/J}) q d, there is YY with X1/J≤Y≤X1/2X^{1/J} \le Y \le X^{1/2} such that for every x∈[Y,Y2]x \in [Y, Y^2] and every field EE with IsChiField ψ (e x) q d E:

∑l∈Slog⁡l∑e=1⌊x401⌋−1ℓl(E,e)≤exp⁡ ⁣(12xε)for all finite sets S of primes,\sum_{l \in S} \log l \sum_{e=1}^{\lfloor x^{401} \rfloor - 1} \ell_l(E, e) \le \exp\!\left(\tfrac12 x^{\varepsilon}\right) \quad \text{for all finite sets } S \text{ of primes},

and each length in the sum is finite.

Idea

The proof of vwin_of_row has two layers.

The top layer is bookkeeping. The row hypothesis is placed in a structure OpenImageCondition. A long list of named statements is then produced, each one by a theorem of the Lean code. The names refer to results in the literature (Bernstein-Zelevinsky, Jacquet-Shalika, Poitou-Tate, Serre-Tate and others). The statements are packed into the structure LiteratureRankNHypothesis and then into a larger structure HypothesesH'. The literature chapter is the place for those statements.

The lower layer is a counting argument. The code splits the length ℓl(E,e)\ell_l(E,e) into three parts: the prime l=2l = 2, and for odd ll the (−1)e(-1)^e and (−1)e+1(-1)^{e+1} eigenspaces of complex conjugation (shaLen_eq_add). Each part gets its own bound (ThmBab, ThmBcFree, and ThmHV). The bound for the "plus" part comes from a case split. For a host datum (a character χ\chi, an exponent and two primes q1,dq_1, d) and a Dirichlet character ψ\psi of the host modulus, a prime ll falls in one of the cases I, II, III, IV, V or Z. Each case has a statement CountI, ..., CountZ that bounds a Selmer-type length polynomially in ll, the modulus and ee. The largest sub-branches of the tree prove these case bounds. They go through Galois representations of dimensions 2, 3 and 4: Kato-type vanishing for two-dimensional representations (case I), symplectic four-dimensional automorphy (cases II and III), unitary three-dimensional automorphy (case V), and a Hida-theory threshold statement.

Before you read

Plan

  1. vwin_of_row wraps the row as OpenImageCondition and calls automorphic_window_of_inputs with a base change theorem.
  2. automorphic_window_of_inputs builds the CM and abelian variety inputs and calls v_window_of_depth_zero_residual_bc_cm.
  3. v_window_of_depth_zero_residual_bc_cm packs four groups of inputs into LiteratureRankNHypothesis and calls v_window_of_literal_automorphy_n.
  4. v_window_of_automorphy_csd_ss_mono_odd extends this to HypothesesH' and calls vwin_of_hypotheses_H_E_alt.
  5. vwin_of_hypotheses_H_E_alt proves LemmaK and the four counts CountI, CountII, CountIII, CountV.
  6. of_four_alt adds ThmBab, ThmBcFree and CountZ, and calls vwin_of_counts.
  7. vwin_of_counts combines all bounds into JointWin 401 and then into VWin.

The steps

Step 1. The entry point

vwin_of_row uses the hypothesis hh twice. It builds OpenImageCondition.mk h, a structure with one field galoisRepOfPolarizedPairLit : RowT (TemperedProductOfSquareIntegrable.lean#L207). Here RowT is by definition Row0b fun h rc => IsRecT h rc (WeakOfAtt.lean#L47), the same Prop as the type of hh. It also passes hh as the argument gal : RowT.

The third input is baseChangeN : BaseChangeIsobaricN (DefiniteNHolds.lean#L5). It states: for a CM field EE such that EE over its maximal real subfield is unramified at all finite places, and for an automorphic representation π\pi of the unitary group in the sense of IsAutRepU, there is a base change datum IsBaseChangeN with isobaric pieces of sizes N1+⋯+Nc=nN_1 + \dots + N_c = n. This theorem is not in the dominator subtree of vwin_of_row. Other parts of the proof use it too.

IsRecT h rc says that a map rc from irreducible representations of GLn\mathrm{GL}_n of a local field to Weil-Deligne representations has a local Artin map aa and satisfies the list of laws LawsT (IsRecTOfIsRec.lean#L55): dimension, characters, the Langlands classification by segments, twists, central characters, duality, injectivity on "good" representations, depth and a type law. This page does not unfold these laws further.

Step 2. CM and abelian variety inputs

theorem UpperTheory.One.Join.automorphic_window_of_inputs (O : OpenImageCondition)
 (bc : BaseChangeIsobaricN) (gal : RowT) : UpperTheory.Top.VWin :=
  v_window_of_depth_zero_residual_bc_cm O bc gal (cm_abelian_inputs_from_literature O)

The theorem cm_abelian_inputs_from_literature (TopMainZj.lean#L59) fills the 11 fields of LiteratureCMHypothesis. Their names are GrothendieckDeformation, AbelianDeRham, AbelianDualCrystal, PeriodComparison, GaussManinPeriods, SerreTateGoodReductionSt, EllipticMainTheoremCM, HilbertClassFieldJ, SiegelKlingenDeligneRibet, PartialZetaFunctionalEquation and HeckeTateFunctionalEquation. The names refer to results in the literature. Each field is given by a theorem of the Lean code. For example elliptic_main_theorem_cm_of_open O is defined as PositivityTheory.Prw.ellipticMainTheoremCM_holds, and it does not use O. The largest part of this sub-branch (489 declarations) is UpperTheory.Lit.serreTateGoodReduction. The whole sub-branch has 594 declarations.

Step 3. The rank n package

theorem UpperTheory.One.Join.v_window_of_depth_zero_residual_bc_cm (O : OpenImageCondition)
 (bc : BaseChangeIsobaricN) (gal : RowT) (cm : LiteratureCMHypothesis) : UpperTheory.Top.VWin :=
  UpperTheory.V.v_window_of_literal_automorphy_n
    ⟨local_inputs_from_literature O, automorphic_inputs_from_literature O bc gal,
     galois_cohomology_from_literature O, cm⟩

LiteratureRankNHypothesis (LitAut.lean#L59) has four fields. loc : LitLocAlt has 18 local statements on GLn\mathrm{GL}_n of a local field (Bernstein-Zelevinsky finiteness, Langlands classification, local converse theorem, depth zero induction and others). aut : LiteratureAutomorphyHypothesis has 15 global statements. Two of them are the inputs bc and gal. So the row enters the counting argument here, as the field aut.gal. coh : LitCoh has 12 statements on Galois cohomology (Poitou-Tate, local finiteness, local Tate duality, Euler characteristics, Greenberg-Wiles, Bloch-Kato). cm is Step 2. The three builder theorems are small (23, 20 and 13 declarations), because each field comes from a theorem that is proved elsewhere.

Step 4. The extended hypotheses

theorem v_window_of_automorphy_csd_ss_mono_odd (hN : LiteratureRankNHypothesis) : VWin :=
  vwin_of_hypotheses_H_E_alt (published_hypotheses_from_patching_alt hN UpperTheory.V.Z.Tf.SR.descWeak)

v_window_of_literal_automorphy_n is equal to this theorem. published_hypotheses_from_patching_alt (PublishedHypothesesFromPatching.lean#L98) builds HypothesesH'. This structure keeps litN := hN and adds fields for Clozel's conjectures on cohomology (ClozelCohC, ClozelConjC), Galois representations for totally real fields in rank 2, strong multiplicity one, quadratic base change, newforms, a descent statement DescWeak, cyclotomic statements and the existence of a definite quaternion algebra. Its dominator subtree has 3,449 declarations. The largest parts are RankTwoDescent.descS (1,506) and clozelConjC (787).

Step 5. Lemma K and the four counts

theorem vwin_of_hypotheses_H_E_alt (hH : HypothesesH') : VWin :=
  have hK := lemmaK_litN_alt hH.litN
  have hE := ends0_alt hH
  have h4 := countII_and_countIII_of_printsH_of_sp4In hH hK (sp4In0_alt hH)
  of_four_alt hH.litN hK
    (count_i_of_printed_bound_k_kato_automorphy hH hK (katoTwoW_of_printsH hH)
      (automorphyUnramified_of_litN_of_gdEnd hH.litN hE.1))
    h4.1 h4.2 (count_v_of_printed_bound_k_reduced_unitary_lift hH hK hE.2)

This theorem is the main fork. Its dominator subtree has 48,127 declarations.

LemmaK CK (LemmaK.lean#L13) states: for a prime ll, a totally complex number field LL that contains a primitive root of unity of order ll (order 4 if l=2l = 2), and e≥1e \ge 1, the group H2H^2 with coefficients ℤl\mathbb{Z}_l twisted by χe+1\chi^{e+1} is finite, and log⁡#H2≤exp⁡ ⁣((2en)CKlog⁡(2D))\log \# H^2 \le \exp\!\big((2en)^{C_K} \log(2D)\big), where n=[L:ℚ]n = [L:\mathbb{Q}] and D=∣disc L∣D = |\mathrm{disc}\, L|.

The counts live on a Host (DirichletPrimeToLPartMulLPowerPart.lean#L148): a primitive character χ\chi of conductor qq, an exponent ex≥2\mathrm{ex} \ge 2 and primes q1≡3(mod8)q_1 \equiv 3 \pmod 8, d≡7(mod16)d \equiv 7 \pmod{16} coprime to qq. Its modulus is N=lcm(2ex+2,q) q1dN = \mathrm{lcm}(2^{\mathrm{ex}+2}, q)\, q_1 d. The set h.X c e is the set of Dirichlet characters ψ\psi modulo NN that are trivial on a subgroup HH and satisfy ψ(−1)=(−1)e\psi(-1) = (-1)^e. For such ψ\psi the code defines a length Vlen (a Tate-Shafarevich-type length for the twist ψ−1χ−e\psi^{-1}\chi^{-e} over ℚ\mathbb{Q}), and VLe Ω c e ψ B says that this length is finite and its normalized value times log⁡l\log l is at most BB. Then (Counts.lean#L60):

def CountShape (P : ∀ (h : Host) (l : ℕ) [Fact l.Prime] (c : Coef l) (_ : ℕ), DirichletCharacter c.O h.N → Prop) : Prop :=
  ∀ χ : PChar, ∃ C : ℕ, ∀ h : Host, h.chi = χ → ∀ t : ℕ, h.Big t →
    ∀ (Ω : Type) [Field Ω] [CharZero Ω] [IsAlgClosure ℚ Ω] (l : ℕ) [Fact l.Prime] (c : Coef l) (e : ℕ),
      l ≠ 2 → 0 < e → e < t → ∀ ψ ∈ h.X c e, P h l c e ψ → VLe Ω c e ψ (((l : ℝ) * h.N * e) ^ C)

In words: for each χ\chi there is an exponent CC such that, for each host with large enough primes, each odd prime ll, each 0<e<t0 < e < t and each ψ\psi in case PP, the length is at most (lNe)C(l N e)^C. CountII, CountIII and CountV are this shape for the cases CaseII, CaseIII and CaseV (with R1≤2c2(ψ)R_1 \le 2^{c_2(\psi)}). CountI has a different shape: it bounds the sum ∑llog⁡l∑ψVn\sum_{l} \log l \sum_{\psi} V_n over primes ll in case I by (Nt)C(N t)^C. The cases are conditions on ll, ee and ψ\psi, such as l−1∣2el - 1 \mid 2e, ll in a finite bad set Bchi χ, or the parity of the order of ψ\psi.

The inputs to the counts are statements on Galois representations.

  • KatoTwoW (KatoTwoW.lean#L10) states: if a two-dimensional representation WW of GℚG_\mathbb{Q}, twisted by χˉ−e\bar\chi^{-e}, is attached to a newform of weight 2e+12e+1, and WW satisfies the side conditions Used and WitAQ, then the Bloch-Kato Selmer groups of two twists of WW vanish. It feeds CountI.
  • SymplecticFourImageDense (IsAut4.lean#L14) states: for good raw data D in the case of (h,e,ψ)(h, e, \psi), the data is finite, has no "end", and all its irreducible pieces are automorphic four-dimensional representations in the sense of IsAut4Alt. It feeds CountII and CountIII.
  • ends0_alt gives GdEndAlt and ThmS6Core_s. The second one feeds CountV. Under ends0_alt sits UnitaryThreeAutomorphyDense (SymUse.lean#L27): for each bound BB there is m≥Bm \ge B with m≡j(modl−1)m \equiv j \pmod{l-1} and an irreducible three-dimensional representation of GLG_L with prescribed seed data. This page does not unfold IsSeedAlt, Raw, Used or WitAQ.

Step 6. Adding the other bounds

theorem of_four_alt (hN : LiteratureRankNHypothesis) (hK : ∃ CK : ℝ, LemmaK CK)
    (h1 : CountI) (h2 : CountII) (h3 : CountIII) (h5 : CountV) : VWin := by
  obtain ⟨B, hB⟩ := Win.thmBab_of_lemmaK (pt_s hN) hN.coh.lfK hN.coh.ltK FW.ferreroWashington hK
  obtain ⟨CK, hL⟩ := hK
  exact vwin_of_counts (pt_s hN) hN.coh.lfK hN.coh.ltK hL (hB 401 (by norm_num))
    (Odd.thmBcFree_of (ML.iv (pt_s hN) hN.coh.uo)) h1 h2 h3 h5 (countZ X.propZ)

The proof gets three more facts. ThmBab K 401 comes from Lemma K, Poitou-Tate, local finiteness, local Tate duality and a statement named ferreroWashington. It gives a window YY in which the l=2l = 2 part (BaAt) and the parts over cyclotomic layers (BbAt) are bounded by exp⁡(x4ε/5)\exp(x^{4\varepsilon/5}). ThmBcFree (PrimeSetBoundDataOfPrimitiveChar.lean#L64) bounds the "minus" part VVminusLe for every CM field of the family. CountZ comes from PropZ, which says that the length is 00 for the trivial character when l≥5l \ge 5, e≥2e \ge 2 is even and l−1∣2el - 1 \mid 2e.

Step 7. From counts to VWin

theorem vwin_of_counts (hP : PoitouTateDuality) (hF : LocalFinitenessK) (hT : LocalTateDualityK) {CK : ℝ}
    (hK : LemmaK CK) {K : BConst} (hB : ThmBab K 401) (hc : ThmBcFree) (h1 : CountI) (h2 : CountII) (h3 : CountIII)
    (h5 : CountV) (hZ : CountZ) : VWin :=
  vwin_of_joint (jointWin_of_counts hP hF hT hK hB h1 h2 h3 h5 hZ) ((thmBc_iff (by norm_num)).2 hc)

The six counts give PointHV and then ThmHV K 401. This is a bound ((cond E) t)C+Cexp⁡(x4ε/5)((\mathrm{cond}\,E)\, t)^C + C \exp(x^{4\varepsilon/5}) for the "plus" part VVplusLe over each CM field EE with IsChiField. JointWin 401 puts the windows of ThmBab and ThmHV together. Then vwin_of_joint (AdmissibleOfPageN.lean#L131) fixes ψ\psi, shows that a field with IsChiField is a CM field, adds the three parts with vvle_of_parts, and checks that the sum is at most exp⁡(12xε)\exp(\tfrac12 x^\varepsilon) for xx large.

Where it is in the code

Sizes are dominator subtree sizes: the number of declarations that only this declaration reaches from the root.

Declaration Module Size
UpperTheory.One.vwin_of_row VwinOfRow.lean 55,805
UpperTheory.One.Join.automorphic_window_of_inputs TopMainZj.lean 55,802
UpperTheory.One.Join.v_window_of_depth_zero_residual_bc_cm TopMainZj.lean 55,154
UpperTheory.V.v_window_of_literal_automorphy_n WindowBoundOfLiteratureLevel.lean 55,095
…DescentVanishingReductions.vwin_of_hypotheses_H_E_alt WindowBoundOfLiteratureLevel.lean 48,127
UpperTheory.V.Z.Kato.WeilSphericalRecords.katoTwoW_of_printsH KatoTwoWOfPublishedH.lean 9,555
…DescentVanishingReductions.sp4In0_alt Sp4In0.lean 7,295
UpperTheory.V.Z.Sym.W.unitary_three_automorphy_dense_of_prints UnitaryThreeInputsOfLiterature.lean 4,238
…DescentVanishingReductions.published_hypotheses_from_patching_alt PublishedHypothesesFromPatching.lean 3,449
UpperTheory.V.ZK.lemmaK_litN_alt KFinal.lean 2,836
UpperTheory.V.Z.Int.slopeAll SlopeAll.lean 2,414
UpperTheory.V.Th.core_alt_bis ThFinal.lean 1,145
UpperTheory.V.HV.Core.of_four_alt HVCore.lean 957
UpperTheory.One.Join.cm_abelian_inputs_from_literature TopMainZj.lean 594

Some notes on the largest sub-branches:

  • katoTwoW_of_printsH (9,555) proves KatoTwoW. It reads the row from hH.litN.aut.gal and uses quadBC2. Most of its size is twoS (9,174) and below it Kato.Final.zetaUr (8,219), with parts named after Kato's Euler system and Hecke projectors on newforms.
  • sp4In0_alt (7,295) proves SymplecticFourImageDense. Its largest parts are seedIIs_alt (3,287) and indCs (2,024).
  • unitary_three_automorphy_dense_of_prints (4,238) proves UnitaryThreeAutomorphyDense. It is reached through ends0_alt and linkedUnlE_alt.
  • lemmaK_litN_alt (2,836) proves ∃ CK, LemmaK CK from LiteratureRankNHypothesis. Its largest part is bOdd (1,788).
  • slopeAll (2,414) proves SlopeAll: for a cuspidal polarized π\pi on a CM field with regular infinitesimal character given by a strictly decreasing mm, the Satake parameters at each place have slopes in the range set by mm. It is used by intK and intP. One path to it starts at count_i_of_printed_bound_k_kato_automorphy.
  • core_alt_bis (1,145) proves DenseThresholdMainCase, a statement on ordinary points of a universal ordinary determinant ring in a Hida setting for 2≤n≤42 \le n \le 4. It is reached through linkedUnlE_alt.

Notes

  • The field of OpenImageCondition is never read in this branch. A search for galoisRepOfPolarizedPairLit finds only its definition here, and one use for a different structure in EveryCharacterOfLiteratureOfVWin.lean. The …_of_open O theorems that this page looked at do not use O. The row reaches the counts only through the argument gal, which becomes LiteratureAutomorphyHypothesis.gal. This page has not checked every …_of_open theorem one by one.
  • The names of the literature statements (BernsteinZelevinskyFiniteness, PoitouTateDuality, ferreroWashington and so on) refer to published results. This page does not check that the Lean statements match the published theorems. The literature chapter treats them.
  • This page does not unfold GS, Sha1, locFam, cycZ or ShaU. The reading of shaLen as the length of a Tate-Shafarevich group of ℚl/ℤl(−e)\mathbb{Q}_l/\mathbb{Z}_l(-e) follows the names and the visible definitions, not a full unfolding.
  • This page does not unfold the cases CaseI to CaseV past their definitions in Host, nor Bchi, pK, hStar, pOne and ordC, which enter Host.Big.
  • This page does not explain why the constant is 401, why ε<1/52\varepsilon < 1/52, or why PageN uses the exponent 2+8ε2 + 8\varepsilon. The code fixes these values. The reasons belong to the part of the proof that uses VWin, under every_character_of_literature_of_vWin.
  • The dominator sizes count declarations, not difficulty. Many small sub-branches (for example the three builders in Step 3) are small because their fields are proved in other parts of the tree.

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