riemann

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

The Weil bound and arithmetic positivity

The folder WeilBound does not contain Weil's explicit formula. Its largest branch on the spine proves a positivity statement of Arakelov geometry, the statement DeepStR. This page states DeepStR, shows the inputs it comes from, and says what the code has in place of the positivity steps of the timeline.

draft

Result

This page covers four steps of the timeline: step 4 (W(1S)≥0W(1_S) \geq 0 through a determinant), step 5 (the Guinand-Weil sign), step 6 (a positivity W(⋅)≥0W(\cdot) \geq 0 for a test function with a factor XηX^{\eta}) and step 13 (the Bost slope inequality). The Lean code has no direct counterpart for steps 4 and 6. It has the Guinand-Weil function of step 5 and a proved Bost slope inequality for step 13, but neither is in this branch. The Notes section gives the details.

What the code does have, under the folder Development/WeilBound, is a positivity theorem of a different kind. The anchor of this page is this theorem (AhSlimU.lean#L9).

theorem deep_slim_use (h₁ : ResolutionCharZero) (h₂ : PrincipalisationCharZero)
    (hPL : PoincareLelongInv) (hU : PositivityTheory.MbHsUse.HSUse)
    (hBig : MoriwakiBigStatement) (hF : HT.FarguesExact) (hZ : GR.ZarhinStatement)
    (hN : Pub.NefRR) : Plug.DeepStR

The conclusion DeepStR (ModelSmall.lean#L63) says this. There is a constant C≥0C \geq 0 with the following property. Take data DD (a structure D7), a core T : CoreJ D that satisfies the laws LawsJ, with m2≤km^2 \leq k for the two integers m,km, k of DD. Take 1≤j<m1 \leq j < m, a tower object Y : Deep T j that is good for some ee and oo, and whose level has at most 15 prime factors. Then

MeanAll (Y.U) (ξR(D)) (ℓ(D)) C\texttt{MeanAll}\ (Y.U)\ (\xi_R(D))\ (\ell(D))\ C

holds, where

ℓ(D)=log⁡rd(E)+∣Re⁡L′L(1,χd)∣.\ell(D) = \log \mathrm{rd}(E) + \left|\operatorname{Re} \frac{L'}{L}(1, \chi_d)\right|.

Here EE is the CM field of the host of DD (a number field), rd(E)=∣disc⁡E∣1/[E:ℚ]\mathrm{rd}(E) = |\operatorname{disc} E|^{1/[E:\mathbb{Q}]} is its root discriminant, and χd\chi_d is the quadratic Dirichlet character modulo the prime dd of DD (SemiposAtOfPositive.lean#L105). The number ξR(D)\xi_R(D) is the norm of a fixed vector φ0\varphi_0.

In words, the Lean code states: if eight published theorems hold, then on the unitary Shimura varieties of the tower, the mean volume slope of small cusp sections is at least −C-C times an error term. The error term grows like log⁡rd(E)+∣L′/L(1,χd)∣\log \mathrm{rd}(E) + |L'/L(1,\chi_d)| times the size of the weight.

Idea

A lower bound on a slope is a positivity statement. It says that a lattice of sections is not too thin. The branch gets it from two theorems of arithmetic intersection theory.

  • Moriwaki and Yuan: for a nef hermitian line bundle A‾\overline{A} on an arithmetic variety of dimension D+1D+1, the self-intersection A‾D+1\overline{A}^{D+1} is at most the arithmetic volume. So a lower bound for A‾D+1\overline{A}^{D+1} gives many small sections.
  • The arithmetic Hilbert-Samuel bound (Gillet-Soulé, Abbes-Bouche, Zhang): the number of small sections grows at most like A‾D+1\overline{A}^{D+1}. Together with Zhang's successive minima, this controls the intersection number from both sides.

The branch then builds a tower of integral models of unitary Shimura varieties (the RootTower of GoodTowerExistsX). On that tower it computes the arithmetic intersection numbers of the Hodge line bundles twisted by a weight uu (the statement VolumeTheoremX). A Bost-style bound on the slopes of the Lie algebra, and the Chowla-Selberg formula, give the error term ℓ(D)\ell(D). The volume theorem then gives the lower bound for the mean slope.

DeepStR is not the end. It feeds topCovolG_alt_alt, which gives a height lower bound (HeightCondition) and then the covolume bound CovolumeBoundOfX of the literature step.

Before you read

You need these ideas.

  • A number field KK, its ring of integers 𝒪K\mathcal{O}_K, CM fields and the root discriminant.
  • A Hermitian form and its unitary group. A unitary Shimura variety of signature (p,q)(p, q).
  • Cusp forms, as sections of an automorphic line bundle with a sup norm.
  • A hermitian lattice over 𝒪K\mathcal{O}_K: a projective 𝒪K\mathcal{O}_K-module with a hermitian norm at each complex place. Its degree, its slope μ\mu and its maximal slope μmax⁡\mu_{\max}.
  • Arithmetic intersection numbers and the arithmetic volume (Arakelov geometry).
  • Dirichlet characters, Dirichlet L-functions and the von Mangoldt function Λ(n)\Lambda(n) (only for the Notes on step 5).
  • For one input: p-adic Hodge theory (the Hodge-Tate sequence of Fargues).

Plan

  1. Unfold the error term and MeanAll.
  2. State the two positivity inputs, MoriwakiBigStatement and HSUse, and the volume they use.
  3. Get an absolute Siegel lemma (LineOfLargeDegree) and a slope bound for the Lie algebra (SlopeBoundSt).
  4. Get a good tower and the volume theorem (GoodTowerExistsX, VolumeTheoremX).
  5. Get DeepStR from the volume theorem (deepStR_of_volume).
  6. Say which inputs are literature and which are proved.

The steps

Step 1. The mean slope and the error term

The definitions are in the namespace GlobalTheory.CovHt (SemiposAtOfPositive.lean#L101). A weight uu is a family of integers uτ,iu_{\tau,i}, with one row for each index τ\tau and p+qp + q entries in each row. Write ∥u∥1=∑τ,i∣uτ,i∣\|u\|_1 = \sum_{\tau,i} |u_{\tau,i}|. The condition WtGap u g says that g>0g > 0, that each row is strictly decreasing, and that ∣uτ,a−uτ,b∣≥g|u_{\tau,a} - u_{\tau,b}| \geq g when a≠ba \neq b and a≥pa \geq p. For a number ss the error term is

err(s,u,g)=(log⁡B(u,g)+s) ∥u∥1+q∑τ(max⁡iuτ,i−min⁡iuτ,i),\mathrm{err}(s,u,g) = (\log B(u,g) + s)\,\|u\|_1 + q \sum_{\tau} \Big(\max_i u_{\tau,i} - \min_i u_{\tau,i}\Big),

where B(u,g)B(u,g) is an explicit bound (bndB). Then MeanAll s C says: for each uu and gg with WtGap u g, there are real numbers sl(Z)\mathrm{sl}(Z), one for each connected component ZZ, with

sl(Z)≤μvol(Z,u)and−C err(s,u,g)≤1#Comp∑Zsl(Z).\mathrm{sl}(Z) \leq \mu_{\mathrm{vol}}(Z, u) \quad\text{and}\quad -C\,\mathrm{err}(s,u,g) \leq \frac{1}{\#\mathrm{Comp}} \sum_{Z} \mathrm{sl}(Z).

The volume slope μvol(Z,u)\mu_{\mathrm{vol}}(Z,u) is compVolSlope (IsHomogeneousCompIdeal.lean#L280). For t∈ℝt \in \mathbb{R}, let V(t)V(t) count the lattice sections of weight τu\tau u on ZZ whose sup norm at the good embeddings is at most eτte^{\tau t}, normalized in τ\tau. Then s≤μvol(Z,u)s \leq \mu_{\mathrm{vol}}(Z,u) means δ(s+t)≤V(t)\delta (s + t) \leq V(t) for all tt, where δ>0\delta > 0 is the top degree of uu on ZZ. So a lower bound on the mean slope says: on average over the components, there are many small cusp sections.

In DeepStR the value of ss is ℓ(D)=log⁡rd(E)+∣Re⁡L′/L(1,χd)∣\ell(D) = \log \mathrm{rd}(E) + |\operatorname{Re} L'/L(1,\chi_d)|.

Step 2. Arithmetic positivity: Moriwaki and Hilbert-Samuel

The arithmetic volume is defined in PenVol.lean#L35. For a family of lattices Λn\Lambda_n and a smallness condition PP,

vol(P,Λ)=lim sup⁡n→∞log⁡#{s∈Λn:P(n,s)}nD+1/(D+1)!.\mathrm{vol}(P,\Lambda) = \limsup_{n \to \infty} \frac{\log \#\{s \in \Lambda_n : P(n, s)\}}{n^{D+1}/(D+1)!}.

MoriwakiBigStatement (MoriwakiSt.lean#L29) works with an arithmetic variety XX of dimension D+1D+1 inside a product of projective spaces, and a hermitian line bundle A‾\overline{A} (HermLine) that is semipositive, invariant under complex conjugation, nef on the fibres and nef on the generic fibre. It says

A‾D+1≤vol(A‾),\overline{A}^{D+1} \leq \mathrm{vol}(\overline{A}),

where A‾D+1\overline{A}^{D+1} is computed by an IntersectionSystem IIII that satisfies the axioms IsIntersectionSystem, and the volume counts sections with sup norm at most 1.

HSUse (SmallGeneratedCountUpperOfUse.lean#L12) is three statements. ArithKleiman says that a nef bundle with nonnegative heights on arithmetic curves has A‾D+1≥0\overline{A}^{D+1} \geq 0. G7 and ArithmeticVolumeNef say that the normalized count of small sections has lim sup⁡≤A‾D+1\limsup \leq \overline{A}^{D+1}. The proof hsLinks_of_use turns HSUse and MoriwakiBigStatement into six statements. Two of them are Zhang's successive minima bounds: if every curve CC has height at least edeg⁡Ce \deg C, then (D+1) edeg⁡X≤A‾D+1(D+1)\,e \deg X \leq \overline{A}^{D+1}, and if a Zariski dense set of curves has height at most e′deg⁡Ce' \deg C, then A‾D+1≤(D+1) e′deg⁡X\overline{A}^{D+1} \leq (D+1)\,e' \deg X.

Both inputs are proved in Lean without hypotheses: moriwakiBig_outright (Outright.lean#L9) and hsUse_outright (line 13 of the same file). Their proofs go through hsChain, which dominates 3,745 declarations. The site also lists them in the literature inputs, because the intersection numbers are abstract.

Step 3. The Siegel lemma and the slope bound

LineOfLargeDegree.Statement (KernelsLattice.lean#L85) is an absolute Siegel lemma. For each number field KK, each hermitian lattice FF of rank i≥1i \geq 1 over 𝒪K\mathcal{O}_K and each ε>0\varepsilon > 0, there is a finite extension K′K' and a vector v≠0v \neq 0 with

μ(F)−12log⁡i−ε≤deg⁡K′(v).\mu(F) - \tfrac12 \log i - \varepsilon \leq \deg_{K'}(v).

The proof lineOfLargeDegree_of_use gets it from the successive minima upper bound of Step 2.

SlopeBoundSt (SlopeBoundSt.lean#L72) is a bound on the Lie algebra of a semi-abelian scheme tt over 𝒪L\mathcal{O}_L with principal generic fibre. Let BB be the Lie algebra with the Hodge metric. Then there is a constant C≥0C \geq 0, independent of all data, with

μmax⁡(B)≤Clog⁡(2+rank⁡B+max⁡(−deg⁡^nB,0)).\mu_{\max}(B) \leq C \log\big(2 + \operatorname{rank} B + \max(-\widehat{\deg}_n B, 0)\big).

Here μmax⁡(B)\mu_{\max}(B) is the supremum of the slopes deg⁡^n(F)/rank⁡F\widehat{\deg}_n(F)/\operatorname{rank} F over the nonzero submodules FF, and deg⁡^n\widehat{\deg}_n is the degree divided by [L:ℚ][L:\mathbb{Q}] (Slopes.lean#L35). This is the place where the branch is closest to the Bost slope method. The proof TopSlope.slopeBound (SlopeBoundOfLargeDegreeLine.lean#L94) does not use a slope inequality. For each F≠0F \neq 0 it takes a short vector of FF from the Siegel lemma, over an extension L′L'. Then a key degree bound (−meanlog⁡∣x∣≤Cℓ-\text{mean}\log|x| \leq C\ell) bounds the degree of that vector. Together they give μ(F)≤(C+2)ℓ\mu(F) \leq (C+2)\ell.

Step 4. The good tower and the volume theorem

good_tower_volume_theorem_bis gives two statements (GoodTowerExistsX.lean#L22).

GoodTowerExistsX says: for a CM field EE, a host, a datum with p,q≥1p, q \geq 1 and a global integral model with split components, there is a root tower TT with a good origin that satisfies the condition Extra.

VolumeTheoremX (GoodTowerExistsX.lean#L28) says: there is C≥0C \geq 0 such that for EE abelian over ℚ\mathbb{Q}, q≥1q \geq 1, 3q+1≤p+q3q + 1 \leq p + q, and at most 4 #CMType (p+q)24\,\#\mathrm{CMType}\,(p+q)^2 bad primes in TT, six things hold for every full intersection system IIII. The last of them is MeanSlope. For a weight uu and B≥128B \geq 128 above the bad primes,

−C((log⁡(B rd(E))+∣Re⁡L′/L(1,χd)∣) ∥u∥1+min⁡(j,N−j) spread(u))≤1#Comp∑cslope(c,u),-C \Big( \big(\log(B\,\mathrm{rd}(E)) + |\operatorname{Re} L'/L(1,\chi_d)|\big)\,\|u\|_1 + \min(j, N - j)\,\mathrm{spread}(u) \Big) \leq \frac{1}{\#\mathrm{Comp}} \sum_{c} \mathrm{slope}(c, u),

where slope(c,u)\mathrm{slope}(c, u) is the self-intersection of the line bundle of weight uu on the component cc, divided by (dim⁡+1)(\dim + 1), the volume and the flag degree (Proportional.lean#L62). This proof is the largest part of the branch: volume_theorem_of_inputs dominates 8,967 declarations. The Chowla-Selberg formula GlobalTheory.LFunction_one_and_mean_log_eta gives the term with L′/L(1,χd)L'/L(1,\chi_d). It says L(1,χ)=πh/dL(1,\chi) = \pi h/\sqrt{d}, and the mean of log⁡(Im⁡τ ∣η(τ)∣4)\log(\operatorname{Im}\tau\,|\eta(\tau)|^4) over the ideal classes is γ−log⁡(2d)−L′/L(1,χ)\gamma - \log(2\sqrt{d}) - L'/L(1,\chi).

Step 5. From the volume theorem to DeepStR

deepStR_of_volume (DeepStROfVolume.lean#L21) takes NefRR, the bundle Top.Given, an intersection system, the birational pullback formula, GoodTowerExistsX and VolumeTheoremX. It checks the bad prime count and the conditions on (p,q)(p,q) for the data of DeepStR. Then it compares the intersection slope of Step 4 with the volume slope of Step 1. The lemma SlopeVolM.slope_sub_cost_le_compVolSlope (UniformAllNormHeightBound.lean#L300) gives

slope(c,u)−2 cost(l,B) ∥u∥1≤μvol(c,u).\mathrm{slope}(c, u) - 2\,\mathrm{cost}(l, B)\,\|u\|_1 \leq \mu_{\mathrm{vol}}(c, u).

This is where the Moriwaki inequality of Step 2 is used: an intersection number gives a count of small sections.

The proof of the anchor, deepStR_of_use (DeepStROfUse.lean#L12), joins Steps 2 to 5 in five lines. It builds the Siegel lemma, Top.Given, and the tower and volume pair. It takes Prov.HS and the birational pullback formula out of Top.Given, and calls deepStR_of_volume.

Step 6. What is literature and what is proved

The parent topCovolSlim2_use_alt_alt receives fifteen hypotheses and passes eight of them to deep_slim_use. In the proof of the Riemann Hypothesis, every_character_of_literature_of_vWin supplies them (EveryCharacterOfLiteratureOfVWin.lean#L812).

Hypothesis Source Published result (from links/literature.yaml)
ResolutionCharZero resolution_char_zero_of_literature Hironaka's resolution of singularities
PrincipalisationCharZero principalisation_char_zero_of_literature Principalisation of ideals (Hironaka)
PoincareLelongInv poincare_lelong_inv_of_literature The Poincaré-Lelong formula
HSUse hsUse_outright (proved in Lean) Arithmetic Hilbert-Samuel, upper bound
MoriwakiBigStatement moriwakiBig_outright (proved in Lean) Moriwaki and Yuan on volumes
FarguesExact fargues_exact_of_literature Exactness of the Hodge-Tate sequence (Fargues)
ZarhinStatement zarhin_statement Zarhin's trick
NefRR nef_rr_of_literature Asymptotic Riemann-Roch for nef divisors

Everything else in the branch is proved in Lean from these eight inputs. That includes the Siegel lemma, SlopeBoundSt, the tower, the volume theorem and the comparison of Step 5. The proofs have not been reviewed by a person.

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 File Size
SelfIntersectionVolumeEstimates.topCovolSlim2_use_alt_alt (parent) .../ThirdPlaceSeries/TopCovolSlim2Use.lean 52,322
PositivityTheory.HtPt.Ah.deep_slim_use .../HilbertSamuelApplications/AhSlimU.lean 30,358
GlobalTheory.CompanionTheory.Ar.Plug.deepStR_of_use .../HilbertSamuelApplications/DeepStROfUse.lean 29,968
GlobalTheory.CompanionTheory.Ar.Link.good_tower_volume_theorem_bis .../HilbertSamuelApplications/GoodTowerVolumeTheoremBis.lean 21,078
volume_theorem_of_inputs (under the previous line) 8,967
GlobalTheory.CompanionTheory.Ar.Plug.deepStR_of_volume .../GeometryProofs/DeepStROfVolume.lean 1,218
GlobalTheory.CompanionTheory.Ar.Giv.lineOfLargeDegree_of_use .../HilbertSamuelApplications/LineOfLargeDegreeOfUse.lean 1,206
GlobalTheory.CompanionTheory.Ar.Sl.TopSlope.slopeBound .../GeometryProofs/SlopeBoundOfLargeDegreeLine.lean 850
PositivityTheory.Pf1.HilbertSamuelMoriwaki.moriwakiBig_outright .../MoriwakiBigTheorem/Outright.lean 3
PositivityTheory.Pf1.HilbertSamuelMoriwaki.hsUse_outright .../MoriwakiBigTheorem/Outright.lean 3

The HilbertSamuelApplications and MoriwakiBigTheorem files are in Development/WeilBound/CoreStatements/AutomorphicKernels/MoriwakiBigness/. The GeometryProofs files are in Development/AutomorphicSpectrum/Proofs/. The parent file is in Development/AutomorphicArithmetic/ThetaCorrespondence/ThetaPairLifts/. The other child of the parent, topCovolG_alt_alt, dominates 12,516. The two outright theorems have size 3 because their proof hsChain (3,745) is shared: its dominator is every_character_of_literature_of_vWin. The full closure of deep_slim_use (all declarations it uses, shared or not) has 87,553 declarations. In the dominated part, about 14,800 declarations are in Development/AutomorphicSpectrum/Proofs, about 6,700 in Development/WeilBound/CoreStatements and about 6,400 in GroundTheory/HeightsAndTheta/Heights.

Notes

  • The name WeilBound of the folder does not refer to Weil's explicit formula. The names with "Weil" in this branch are about Weil pairings on abelian schemes. The words "Guinand", "Stepanov" and "Bost" do not occur in the Lean source.
  • Timeline step 4 (W(1S)≥0W(1_S) \geq 0 through a determinant, read as Stepanov with multiplicity one) has no counterpart in the code. There is no quantity W(1S)W(1_S) and no positivity proved by a determinant. See also the determinants page.
  • Timeline step 6 (W(1(0,1]+Xη1(0,1/X])≥0W(1_{(0,1]} + X^{\eta} 1_{(0,1/X]}) \geq 0 gives a bound larger than 1−η1 - \eta) has no counterpart in the code. There is no test function of this form and no exponent η\eta of this kind. The positivity in this branch is the lower bound on mean slopes of Step 1, with an error term in log⁡rd(E)\log \mathrm{rd}(E) and ∣L′/L(1,χd)∣|L'/L(1,\chi_d)|.
  • Timeline step 5 (the Guinand-Weil sign): the code has a Guinand-Weil type function, Gfun in Formalism/LocalRingCardinality/DataParams₂.lean, with G(y)=∑n≤yΛ(n)(y/n−1)−ylog⁡y+(1+γ) yG(y) = \sum_{n \leq y} \Lambda(n)(y/n - 1) - y \log y + (1+\gamma)\,y. It is used on the side of the theta matching. It is not in the closure of deep_slim_use. The only L-function term in this branch is L′/L(1,χd)L'/L(1,\chi_d), which comes from Chowla-Selberg, not from an explicit formula.
  • Timeline step 13 (the Bost slope inequality): the code proves a slope inequality for Euclidean ℤ\mathbb{Z}-lattices, Formalism.F.SlopeInequality (Statement.lean#L75). If φ:L→M\varphi : L \to M is injective and ∥φx∥≤C∥x∥\|\varphi x\| \leq C\|x\|, then μ(L)≤μmax⁡(M)+log⁡C\mu(L) \leq \mu_{\max}(M) + \log C. Its proof is slopeInequality_of. It is not used in this branch. Here the role of a slope bound is played by SlopeBoundSt, which is proved from the Siegel lemma and a degree bound, not from a slope inequality.
  • In TopSlope.slopeBound the second conclusion Prov.LemmaS is the same statement as SlopeBoundSt (an abbrev). The proof gives the same term twice.
  • MoriwakiBigStatement is weaker than the published theorem: it has ≤\leq where the published result for nef bundles has equality. HSUse has only the lim sup⁡\limsup upper bound. In both, the intersection numbers come from an abstract IntersectionSystem with the axioms IsIntersectionSystem, not from the Gillet-Soulé construction. So the results depend on those axioms.
  • This page does not unfold Top.Given (a large conjunction of about twenty statements), the structures D7, CoreJ, LawsJ and Deep, the condition Extra, or the first five parts of VolumeTheoremX.
  • The site does not claim that the proof is correct. The Lean code states the results above, and nobody has reviewed it. The trust report says what has been checked.

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