riemann

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

The unitary hypotheses node

The Lean code ends the trunk with one existence statement. A record of forty-six fields, called HypsAlt, has an element. This page says what that record is, what the words "definite side" and "closed bound" mean in the code, and what the three large branches prove.

draft

Result

The previous chapter, theta matching, ends at the theorem UpperTheory.Lit.every_character_of_theta_integrals_local_matching. That theorem calls one more node of the trunk. This node is the last node of the trunk. Its statement is:

theorem nonempty_hyps_definite_side_of_closed_bound :
    GlobalTheory.Fc.S0.UnitaryGroupThirdPlaceHypotheses →
    GlobalTheory.FunPr.ThetaLocalInputs →
    GlobalTheory.IndefiniteRowSideFullHecke.WeilTensorModels.ThetaIntegralsAgree →
    GlobalTheory.IndefiniteRowSideFullHecke.WeilTensorModels.LocalMatching →
    GlobalTheory.CovTop.CovolumeBoundOfX →
    ∀ (hL : UpperTheory.Zero.LLC.RecTExists),
    (∀ cDep : ℝ, GlobalTheory.Fc.RL.LinkedAt (parametersOfDependence hL cDep)
        (fun Q L => SlotNormCondition Q L fun 𝒞 =>
          Closed 𝒞 fun Fm _ => BoundAtTwoThreeFamilyFonAlt Fm)) →
    Nonempty (GlobalTheory.Fc.HypsAlt S0.NL GlobalTheory.Fc.DefiniteSideCondition)

(Namespaces are shortened here. The full text is at SiegelEisensteinSecondTermHypotheses.lean#L44.)

In words: assume seven inputs. The first is a list of about sixty statements named after published results on unitary groups and automorphic forms. The sixth, RecTExists, is an existence statement of local Langlands type. The seventh is a "closed bound" on certain families. Then there is at least one record of type HypsAlt NL DefiniteSideCondition. The caller every_character_of_theta_integrals_local_matching gives this record to UpperTheory.Top.every_character_of_green_v_win, together with VWin, and gets UpperTheory.EveryCharacter.

The node dominates 263,615 declarations. Its three largest dominator children are the three branches of this page: 88,131, 19,399 and 12,608 declarations.

Idea

The target HypsAlt is a record of forty-six fields. Most fields are statements about a class 𝒜\mathcal A of "families" of arithmetic data, often in the form "this holds once the constants of the family are large". The proof does not build the record field by field at the top. It works in four moves.

First, it proves twelve "open parameter statements" for one choice of parameters (branch 3). Second, it proves two estimates, SurjectiveOntoX and LambdaLowerBound₂ (branch 2). Third, the top node uses these to get a constant cDep and two more row statements, a functional condition and a covolume bound. Fourth, branch 1 turns all of this into a record AllRowsAlt (one "row" per field of HypsAlt) and then into an element of HypsAlt.

Each input that is named after the literature is a hypothesis. The Lean code does not prove those inputs in this subtree. It proves that they imply the conclusion.

Before you read

Plan

  1. State what the hypotheses record UnitaryGroupThirdPlaceHypotheses contains.
  2. Unfold Closed, the meaning of "closed bound".
  3. Unfold DefiniteSideCondition, the meaning of "definite side".
  4. Unfold HypsAlt, the record that is shown nonempty, and NL.
  5. Unfold RecTExists, LinkedAt and the parameters.
  6. Follow the proof of the top node.
  7. Describe branch 3 (open0_of_unitary_group_hypotheses).
  8. Describe branch 2 (ontoX_and_lamLow_of_unitaryGroupHypotheses').
  9. Describe branch 1 (hypotheses_nonempty_of_unramified_exponents_alt).

The steps

1. The input record of literature hypotheses

UnitaryGroupThirdPlaceHypotheses is a structure of type Prop with sixty fields (BernsteinUnitaryProductIrreducible.lean#L22). Each field is one Prop. Most are in the namespace GlobalTheory.Published. A few examples:

structure UnitaryGroupThirdPlaceHypotheses : Prop where
  clozelCohomological : GlobalTheory.Published.ClozelCohomological
  greenCuspidal : GlobalTheory.Published.GreenCuspidal
  cuspRapidDecay : GlobalTheory.Published.CuspRapidDecay
  siegelWeilUnitaryConvergent : GlobalTheory.Published.SiegelWeilUnitaryConvergent
  doublingMethod : GlobalTheory.Published.DoublingMethodLocalZetaIntegrals
  ...
  siegelEisensteinSimplePole : GlobalTheory.Published.UnusedSlot72
  litN : GlobalTheory.Mem.LiteralAutomorphy.NonEndoscopicHypothesesAlt
  ...

The field names refer to the literature (Clozel, Green, Siegel-Weil, the doubling method, Harish-Chandra, Casselman-Wallach, Bernstein-Zelevinsky and others). This page unfolds one of them as an example. CuspRapidDecay says: for each datum D : D7 and each admissible family of real embeddings, every function on the group GA D V that is automorphic, smooth, of uniform moderate growth and cuspidal has rapid decay. Here rapid decay means: for each N∈ℕN \in \mathbb N there is CC with ∣φ(g)∣⋅h(g)N≤C|\varphi(g)| \cdot h(g)^N \le C for all gg, where h(g)h(g) is the infimum of a height over the rational points γg\gamma g. This is the form of the classical statement that cusp forms decay rapidly, stated for the objects of this code base. The other fields are not unfolded on this page.

One field is the trivial statement. UnusedSlot72 is defined as True (DiscreteSpectrumResidualClassification.lean#L7), although the field that has this type is named siegelEisensteinSimplePole.

2. "Closed bound": the predicate Closed

A FamP ε is a structure ("family") with a real parameter ε\varepsilon. It has a threshold x1>1x_1 > 1, an index type IxI_x for each real xx, a datum D(x,i)D(x,i) of type D7 and size data L(x,i)L(x,i) for each xx and i∈Ixi \in I_x, and many named constants. A FamClass is a predicate on families, for all ε\varepsilon at once. The definition of Closed is:

def Closed (𝒞 : FamClass) (P : ∀ {eps : ℝ}, FamP eps → ℝ → Prop) : Prop :=
  ∃ ε1 : ℝ, 0 < ε1 ∧ ∀ eps : ℝ, 0 < eps → eps ≤ ε1 → ∃ K : TransportedAlt, ∀ (Fm : FamP eps) (CH : ℝ),
    𝒞 Fm → Fm.Large K CH → P Fm CH

In words: there is ε1>0\varepsilon_1 > 0 such that for each ε∈(0,ε1]\varepsilon \in (0, \varepsilon_1] there is a threshold record KK with this property. For each family FF with parameter ε\varepsilon in the class 𝒞\mathcal C and each real CHC_H, if FF is large with respect to KK and CHC_H, then P(F,CH)P(F, C_H). The record KK has a number and three functions from lists of reals to reals. Fm.Large K CH says that three lists of constants of FF are above thresholds, and that the thresholds may depend on a fourth list (cOut) of constants of FF. So Closed 𝒞 P means "PP holds for all small ε\varepsilon, for each family in 𝒞\mathcal C whose constants are large enough".

The "closed bound" of the theorem name is the seventh hypothesis. It applies Closed to BoundAtTwoThreeFamilyFonAlt. That predicate says: there are CC and x0x_0 such that for all x≥x0x \ge x_0, all i∈Ixi \in I_x and all auxiliary data that satisfy a setup condition SetupAlt, the statement BoundAtTwoThree (L x i) X.ρ C holds. This page does not unfold BoundAtTwoThree. The top caller gets this hypothesis from UpperTheory.N.boundAtTwoThree_linked_of_recT and the field nLink of the input record.

3. "Definite side": DefiniteSideCondition

The type DefiniteSideType is a predicate on a datum D : D7, an object O, a real cc, a "bundle" b : BundleCondition D O c and two reals C,CHC, C_H. DefiniteSideCondition is one such predicate (DefiniteSideCondition.lean#L25). Its shape is one long implication:

def DefiniteSideCondition : DefiniteSideType := fun D O c b C CH =>
  ∀ (hok1 : ...) (hsep : ...) (hP : CovolumeOntoIsometryConditions ...) (hI : PassIn3 ...)
    (lam0 C0 ce Sw C1 : ℝ) (H : ∀ u, ... .NumFacts ...) (NuC : ...) (I : DivisionCompatibilityDataThreeSided ...),
    DivTied3 ... → HEq I b.P.I → ∀ u, RowsU ...

In words: for each choice of side data for the bundle bb (two structural facts, a covolume and isometry condition, a pass condition, five reals with numerical facts, a family of submodules over the ring of integers of a field D.what₂.LL, and division compatibility data II), if the data are tied (DivTied3) and II is the division data of bb, then the row statement RowsU holds at every index uu. This page does not unfold RowsU, DivTied3 or the side conditions. The word "definite" is in the name only; the definition does not mention a definite form directly.

In HypsAlt, this predicate enters through the field rRows : SelectionDefiniteRows DR 𝒜 σ. That field says: Closed (D3 3 𝒜) of the statement "there is C≥0C \ge 0 such that, eventually along the selection, the bundles satisfy constant bounds and DR b C CH".

4. The record HypsAlt and the class NL

HypsAlt N DR is a structure in Type 1 with forty-six fields (Hyps.lean#L22). It takes a predicate N on family classes and a definite side predicate DR. The first fields are data:

  • a family class 𝒜\mathcal A,
  • a selection σ : RawSel BundleCondition 𝒜 (for each family in 𝒜\mathcal A, each large xx and each ii, a bundle),
  • a function cT : ℝ → ℝ → ℝ.

The other fields are properties. Many are of the form Closed (D3 3 𝒜) (...) or "for all xx after the selection threshold and all ii, the selected bundle has property ...". The properties have names such as twist condition, period admissibility, divisor condition, class number formula at the datum (StandardLattice.ClassNumberFormula₂), counting bounds, a surjectivity of a projection map (zL), and genus mass conditions. The last field is nRest : N (D3 3 𝒜). Here D3 3 𝒜 is a class derived from 𝒜\mathcal A (Des (WithChN 𝒜) 3). This page does not unfold Des or WithChN.

The theorem uses N = NL, where

def NL (𝒞 : FamClass) : Prop := Closed 𝒞 fun Fm _ => Formalism.MainNotions.LinkN Fm

LinkN F says: there are CC and x0x_0 such that for all x≥x0x \ge x_0, all ii for which ShaFinOdd (D x i) holds, and each lift index jj, if a series ∑lv(l)\sum_l v(l) converges then

η(L,j)≤d0(j)(C(f m b L2+(log⁡x)2Nc)+∑lv(l)).\eta(L, j) \le d_0(j) \Big( C \big( f\, m\, b\, L^2 + (\log x)^2 N_c \big) + \sum_l v(l) \Big).

Here η(L,j)=∑wℓ(w) ew ηj(w)\eta(L,j) = \sum_w \ell(w)\, e_w\, \eta_j(w), summed over places ww in the support of LL where the twist μw\mu_w differs from μ0\mu_0, and ewe_w is the ramification index. The letters f,m,b,L,Nc,d0(j)f, m, b, L, N_c, d_0(j) are fields of the datum and of the size data (GlobalTheoremInputs.lean#L45). So nRest asks that a weighted count of twisted places is bounded, in the "closed" sense.

5. RecTExists, parameters and LinkedAt

RecTExists (IsRecTOfIsRec.lean#L78) says: for each number field EE, algebraic closure Ω\Omega, valuation subring AA of Ω\Omega and finite place ww of EE under AA, there is a map rc from irreducible representations of GLn(Ew)\mathrm{GL}_n(E_w) (all nn) to Weil-Deligne representations with the laws LawsT. The laws include: dimension nn, agreement with local class field theory for n=1n = 1, compatibility with Langlands classification by segments, cuspidal goes to irreducible, compatibility with twists, central characters and duals. This is an existence statement in the style of the local Langlands correspondence for GLn\mathrm{GL}_n. The caller gets it from local_langlands_gl_recT_exists_of_literature.

A Parameters hL is a record of four parts: a position R, a norm choice extension N, an SExt and a GeoP. The parameters used here are

parametersOfDependence hL cDep =
  parametersOfExtension (extensionWithList hL cDep) SExt.none (geoCl anyGeometry)

This page does not unfold extensionWithList.

LinkedAt p r says: there is a further condition R on bases such that LargeBaseCondition holds for it, and the row statement r holds for the combined slot and the bundle linkage of p. A row statement Row Q Lk H says: there is a choice E of bundle properties with RowE Q Lk E, and H 𝒜 σ holds for every class 𝒜\mathcal A and selection σ\sigma that are selections for E. SlotNormCondition Q L N is Row Q L (fun 𝒜 _ => N (D3 3 𝒜)). So the "closed bound" hypothesis says: for every real cDepc_{\mathrm{Dep}}, along the parameters of dependence cDepc_{\mathrm{Dep}}, every admissible selection has Closed (D3 3 𝒜) BoundAtTwoThreeFamilyFonAlt.

6. The proof of the top node

The Lean proof (lines 45 to 68) does these steps.

  1. It gets FunctionalsSeparate from functionalsSeparate_of_howeUnram_of_siegelWeil (4,530 declarations).
  2. It gets S0.RowRat from test_vector_assembly_holds through rowRatW_of_rowTest.
  3. It gets ConstantFunctionFits from fourier_coefficient_fit_of_join_alt (4,155). This uses the second half of branch 2.
  4. evaluation_grid_of_join gives a constant CF1C_{F1} such that EvLgG (GridA CF) holds for CF≥CF1C_F \ge C_{F1}. rowFun_at_gridA_alt gives a constant CF0C_{F0}.
  5. It sets cDep=max⁡(CF0,CF1)c_{\mathrm{Dep}} = \max(C_{F0}, C_{F1}).
  6. Branch 3 gives OpenParameterStatements at the parameters of dependence cDepc_{\mathrm{Dep}}.
  7. open_and_linked_of_stronger_extension_alt moves these statements to the stronger parameters extensionWithAtProperty ... (GridA cDep).
  8. It gets the functional condition and the covolume bound CovolumeBoundThirdPlaceSeries at the stronger parameters. The second uses the input CovolumeBoundOfX.
  9. It applies branch 1 to all of the above.

7. Branch 3: the twelve open parameter statements

theorem GlobalTheory.T.open0_of_unitary_group_hypotheses :
    UnitaryGroupThirdPlaceHypotheses → ∀ (hL : RecTExists) (cDep : ℝ),
      OpenParameterStatements (parametersOfDependence hL cDep)

OpenParameterStatements p is a structure of twelve fields. Each field is LinkedAt p r for one row statement r: primitivity, a lambda prime condition, a divisor condition, a decomposition point condition, a selection jump condition, a tie condition, a label condition, a goodness condition, a weak geometric compatibility, a multiplicity condition, a kappa condition, and a defect condition. The defect condition RowDefectCondition is the row form of DefiniteSideCondition. So branch 3 already proves the definite side condition in row form.

The theorem at line 75 only calls Open0ProofU.open_orbit_of_unramified_exponents_alt (line 39). That proof builds each of the twelve fields from one lemma and the input record. The largest part is main_theorem_published_inputs_third_alt_alt (10,941 declarations). It gives the multiplicity condition, from fourteen named hypotheses (Harish-Chandra finiteness, rapid decay of cusp forms, Weil representation of the unitary pair, Siegel-Weil formula, doubling method, strong approximation and others). The next parts are weil_unitary_entries' (1,187, the defect condition) and good_ft_entries_alt_alt (262, the label condition).

8. Branch 2: SurjectiveOntoX and LambdaLowerBound₂

theorem ontoX_and_lamLow_of_unitaryGroupHypotheses' :
    UnitaryGroupThirdPlaceHypotheses →
      GlobalTheory.BigOn.SurjectiveOntoX ∧ GlobalTheory.BigTop.LambdaLowerBound₂

Both conclusions have the form Closed FamHyp (...) for one fixed class FamHyp.

  • SurjectiveOntoX: for large families FF in FamHyp, if DIVinN F holds, then there is C0≥0C_0 \ge 0 such that eventually, at every point pp with More p, and for every C≥C0C \ge C_0, the statement (p.S false).ON (p.N C CH) holds.
  • LambdaLowerBound₂: for large families, eventually at every point pp and for every CC, (p.S true).L6 (p.N C CH) holds.

Here "eventually at points" means: there is x0x_0 such that for every x>x1x > x_1 with x≥x0x \ge x_0, every ii and every point pp that satisfies PointConditions', the property holds. This page does not unfold FamHyp, ON or L6. The names suggest "onto" and "a lower bound for λ\lambda", but this page does not claim more than the definitions.

The proof is two calls. ontoX_of_hypotheses_alt_alt (19,069 declarations) proves the first part from twelve named hypotheses. One of them is RecTExists, which is made by llOf_alt from the field litN.loc.llc. Its main sub-branches are runHolds_of_parOK (5,710), geoSt_of_pageSt_alt (5,131) and expSt_of_base_change_and_crystalline_lattices (3,719). lamLow_of_hypotheses_alt_alt (312 declarations) proves the second part from seven named hypotheses: rapid decay, the Weil-Siegel formula, Siegel-Weil convergence, the doubling method, unramified doubling for GL\mathrm{GL}, Bernstein-Zelevinsky finiteness and formal degrees.

9. Branch 1: the record is nonempty

theorem hypotheses_nonempty_of_unramified_exponents_alt :
    UnitaryGroupThirdPlaceHypotheses → S0.RowRat →
    ∀ (hL : RecTExists) (cDep : ℝ) (A : GroupRetwisting.AtP), EvLgG A →
    LinkedAt (parametersOfDependence hL cDep) (closed bound) →
    LinkedAt (parametersOfExtension (extensionWithAtProperty ... A) ...) FunctionalCondition →
    LinkedAt (parametersOfExtension (extensionWithAtProperty ... A) ...) CovolumeBoundThirdPlaceSeries →
    Nonempty (HypsAlt S0.NL DefiniteSideCondition)

(The fifth line is the same "closed bound" hypothesis as in the top node.) The proof (HypothesesNonemptyOfUnramifiedExponents.lean#L42) has six steps.

  1. large_unramified_exponent_bundles_exist_alt (19,960 declarations) gives three facts: SiegelMassRatioCM, MemberE for the stronger parameters, and ExBL (there is a large base condition with bundles). Its main part is member_exists_baseLattice_alt (13,155).
  2. hyps_e_third_layer_ser_n_of_geometric_galois (1,263) gives two inputs prN and serN from the fields litN and toHypothesesX.
  3. It calls branch 2 and branch 3 again at cDepc_{\mathrm{Dep}}.
  4. UpperTheory.GalC.final_alt_alt (15,614) turns the closed bound into NL. Its type NLink.Final.mainTheorem says: from the named hypotheses and the closed bound for BoundAtTwoThreeFamilyFonAlt, the same LinkedAt statement holds with Closed 𝒞 LinkN. This is where the "closed bound" hypothesis is used. Its main part is final_of_alt (14,859).
  5. open_and_linked_of_stronger_extension_alt moves the open statements and the NL row to the stronger parameters.
  6. allRows_of_open0_alt (33,892) gives a record AllRowsAlt ... NL DefiniteSideCondition. Then nonempty_hyps_of_allRows_alt (419) builds the HypsAlt element from it.

AllRowsAlt has one "row" for each property field of HypsAlt, each in the form Lg1 KP Q Q' (fun S => ...). The last step joins these rows with RowL.and into one row, takes one selection from ExBL, and reads off the fields. The largest parts of allRows_of_open0_alt are lg1_sLp_sO_hTF (13,601; the LpL^p selection, order and torsion free rows, from one Siegel-Weil hypothesis) and hServ_of_linkR (10,829; the service row, from global Artin reciprocity).

What "unramified exponents" means here

The phrase "unramified exponents" is in the names of four files and theorems of this chapter. It is not in any hypothesis or conclusion of these theorems. No definition with that name enters the statements above. So this page does not give it a meaning. The same holds for "open orbit" in the name of branch 3, and for "surjective and low weight" in the file name of branch 2. The theorem names ontoX and lamLow do match the conclusions SurjectiveOntoX and LambdaLowerBound₂, and open0 matches the structure OpenParameterStatements (and its near copy S0.Open0).

Where it is in the code

All modules below are in Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/ThirdPlaceSeries/ unless the name says otherwise. Size is the number of declarations that the node dominates.

Declaration Module Size
UpperTheory.Lit.nonempty_hyps_definite_side_of_closed_bound SiegelEisensteinSecondTermHypotheses.lean#L44 263,615
GlobalTheory.Fc.S0.StageU.hypotheses_nonempty_of_unramified_exponents_alt HypothesesNonemptyOfUnramifiedExponents.lean#L41 88,131
GlobalTheory.Fc.AllAtMemberStage0.allRows_of_open0_alt LinkROfIsLinked.lean#L50 33,892
GlobalTheory.T.large_unramified_exponent_bundles_exist_alt MemberBundlesLargeUnramifiedExponents.lean#L69 19,960
UpperTheory.GalC.final_alt_alt CFinal.lean#L19 15,614
GlobalTheory.Fc.nonempty_hyps_of_allRows_alt NonemptyHypsOfAllRows.lean#L19 419
GlobalTheory.Fc.S0.ontoX_and_lamLow_of_unitaryGroupHypotheses' SurjectiveAndLowWeightOfUnramifiedExponents.lean#L16 19,399
GlobalTheory.BigOn.ontoX_of_hypotheses_alt_alt OntoXOfPrints.lean#L28 19,069
GlobalTheory.BigTop.lamLow_of_hypotheses_alt_alt LamLowOfPublished.lean#L17 312
GlobalTheory.T.open0_of_unitary_group_hypotheses OpenOrbitOfUnramifiedExponents.lean#L75 12,608
GlobalTheory.T.main_theorem_published_inputs_third_alt_alt MainTheoremPublishedInputsThird.lean#L34 10,941
GlobalTheory.Fc.HypsAlt Hyps.lean#L22 1
GlobalTheory.Fc.DefiniteSideCondition DefiniteSideCondition.lean#L25 1
GlobalTheory.Fc.S0.UnitaryGroupThirdPlaceHypotheses BernsteinUnitaryProductIrreducible.lean#L22 1
Formalism.MainNotions.Closed ParamsHeightRankConstant.lean#L159 1
UpperTheory.Zero.LLC.RecTExists IsRecTOfIsRec.lean#L78 1

Branches 2 and 3 are used twice: once by the top node and once inside branch 1. So in the dominator tree they hang under the top node, not under branch 1.

Notes

  • What is hypothesis and what is proof. The sixty fields of UnitaryGroupThirdPlaceHypotheses, the inputs ThetaLocalInputs, ThetaIntegralsAgree, LocalMatching, CovolumeBoundOfX, RecTExists and the closed bound are all assumptions of this node. The node proves that they imply the existence of a HypsAlt record. Whether each assumption is a correct form of the published theorem it is named after is a separate question. This page checked only CuspRapidDecay and RecTExists in any detail.
  • A trivial field. The field siegelEisensteinSimplePole has type UnusedSlot72, which is True. So this field asks for nothing.
  • A pass-through. llOf_alt takes RecTExists and eleven other hypotheses, and returns its first argument. The other eleven hypotheses are not used in its proof.
  • Not unfolded. This page does not unfold RowsU, DivTied3, BoundAtTwoThree, FamHyp, ON, L6, Des, WithChN, extensionWithList, extensionWithAtProperty, EvLgG, GridA beyond its first line, RowRat, MemberE beyond its four conjuncts, ExBL, SiegelMassRatioCM, or most fields of HypsAlt. The plain words for these objects stay at the level of their names and shapes.
  • Names. Many names in this subtree (_alt, _alt_alt, ', "third place", "unramified exponents", "open orbit") do not match a defined object. This page does not guess their meaning.
  • What the record is used for. This page stops at the element of HypsAlt. How every_character_of_green_v_win uses it, together with VWin and the Green character input, is not on this page.

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