← All problems

Riemann hypothesis

167 years old · traced to

For s∈Cs\in\mathbb C with Re⁡(s)>1\operatorname{Re}(s)>1, the Riemann zeta function is the absolutely convergent series

ζ(s)=∑n=1∞1ns=∏p prime(1−p−s)−1.\zeta(s)=\sum_{n=1}^{\infty}\frac{1}{n^{s}}=\prod_{p\ \text{prime}}\bigl(1-p^{-s}\bigr)^{-1}.

Riemann showed that ζ\zeta extends to a meromorphic function on C\mathbb C whose only pole is a simple pole at s=1s=1 with residue 11. The completed function satisfies the functional equation

π−s/2 Γ ⁣(s2)ζ(s)=π−(1−s)/2 Γ ⁣(1−s2)ζ(1−s).\pi^{-s/2}\,\Gamma\!\left(\tfrac s2\right)\zeta(s)=\pi^{-(1-s)/2}\,\Gamma\!\left(\tfrac{1-s}2\right)\zeta(1-s).

The zeros of ζ\zeta at s=−2,−4,−6,…s=-2,-4,-6,\dots are the trivial zeros. All other zeros are nontrivial, and they lie in the critical strip 0<Re⁡(s)<10<\operatorname{Re}(s)<1.

Riemann hypothesis. Every nontrivial zero of ζ\zeta has real part 12\tfrac12. Equivalently, for every s∈Cs\in\mathbb C,

ζ(s)=0ands∉{−2,−4,−6,… }⟹Re⁡(s)=12.\zeta(s)=0\quad\text{and}\quad s\notin\{-2,-4,-6,\dots\}\quad\Longrightarrow\quad\operatorname{Re}(s)=\tfrac12 .

This is the statement in Bombieri's official Clay problem description [1]. The problem asks for a proof. An explicit zero ρ\rho with ζ(ρ)=0\zeta(\rho)=0 and 12<Re⁡(ρ)<1\tfrac12<\operatorname{Re}(\rho)<1 would disprove the conjecture.

Background

Riemann introduced ζ\zeta as a function of a complex variable in his 1859 memoir on the number of primes below a given bound [2]. He gave an explicit formula expressing the prime-counting function in terms of the nontrivial zeros, and remarked that it is very likely that all of them lie on the line Re⁡(s)=12\operatorname{Re}(s)=\tfrac12. The real parts of the zeros control the error in the prime number theorem. Von Koch proved that the hypothesis is equivalent to

π(x)=Li⁡(x)+O(x log⁡x),Li⁡(x)=∫2xdtlog⁡t,\pi(x)=\operatorname{Li}(x)+O\bigl(\sqrt{x}\,\log x\bigr),\qquad \operatorname{Li}(x)=\int_2^x\frac{dt}{\log t},

the best possible error term up to the logarithm [3]. The hypothesis has many other equivalent forms and underlies a large body of conditional results in analytic number theory. It is the most prominent case of the Riemann hypothesis for LL-functions, which is open in general. [1, 4]

Equivalent formulations
  1. For s∈Cs\in\mathbb{C}, define the Riemann zeta function by

    ζ(s)=∑n=1∞1ns.\zeta(s)=\sum_{n=1}^{\infty}\frac{1}{n^s}.

    It converges for Re⁡(s)>1\operatorname{Re}(s)>1. Riemann hypothesis. All non-trivial zeros of ζ(s)\zeta(s) lie on the line Re⁡(s)=12\operatorname{Re}(s)=\frac{1}{2}. The hypothesis is one of the central unsolved problems in number theory and has deep consequences for the distribution of prime numbers.

    source: Madhuparna Das, “Mapping Mathematical Hardness: Machine-Assisted Conjecture Discovery and the Quantification of Non-Triviality”, arXiv:2606.14804 (2026).

References
  1. E. Bombieri, "The Riemann hypothesis", in J. Carlson, A. Jaffe and A. Wiles (eds.), The Millennium Prize Problems, Clay Math. Inst. and AMS (2006). Clay PDF
  2. B. Riemann, "Ueber die Anzahl der Primzahlen unter einer gegebenen Grösse", Monatsberichte der Berliner Akademie (1859), 671–680.
  3. H. von Koch, "Sur la distribution des nombres premiers", Acta Math. 24 (1901), 159–182. doi:10.1007/BF02403071
  4. J. B. Conrey, "The Riemann hypothesis", Notices Amer. Math. Soc. 50 (2003), no. 3, 341–353.
  5. J. Hadamard, "Sur la distribution des zéros de la fonction ζ(s)\zeta(s) et ses conséquences arithmétiques", Bull. Soc. Math. France 24 (1896), 199–220. doi:10.24033/bsmf.545
  6. N. Levinson, "More than one third of zeros of Riemann's zeta-function are on σ=1/2\sigma=1/2", Adv. Math. 13 (1974), 383–436. doi:10.1016/0001-8708(74)90074-7
  7. J. B. Conrey, "More than two fifths of the zeros of the Riemann zeta function are on the critical line", J. Reine Angew. Math. 399 (1989), 1–26. doi:10.1515/crll.1989.399.1
  8. H. M. Bui, J. B. Conrey and M. P. Young, "More than 41% of the zeros of the zeta function are on the critical line", Acta Arith. 150 (2011), 35–64. arXiv:1002.4127
  9. K. Pratt, N. Robles, A. Zaharescu and D. Zeindler, "More than five-twelfths of the zeros of ζ\zeta are on the critical line", Res. Math. Sci. 7 (2020). doi:10.1007/s40687-019-0199-8
  10. L. Alpöge and R. Furman, "More than two thirds of the zeta zeros are simple and on the critical line", preprint (2026). arXiv:2608.13637
  11. D. Platt and T. Trudgian, "The Riemann hypothesis is true up to 3⋅10123\cdot10^{12}", Bull. Lond. Math. Soc. 53 (2021), 792–797. doi:10.1112/blms.12460
  12. L. Guth and J. Maynard, "New large value estimates for Dirichlet polynomials", Ann. of Math. 203 (2026), 623–675. arXiv:2405.20552
  13. G. H. Hardy, "Sur les zéros de la fonction ζ(s)\zeta(s) de Riemann", C. R. Acad. Sci. Paris 158 (1914), 1012–1014.
  14. Wikipedia, Riemann hypothesis.

Source: Clay Mathematics Institute

Progress summary

Refreshed
Solved

The Riemann hypothesis is reported as solved by a checked proof, although a complete readable mathematical account is not yet available.

Riemann proposed in 1859 that every nontrivial zero of the zeta function lies on the critical line. The problem asks for a proof of this assertion for all zeros.

Known results

  • Hadamard and de la Vallée Poussin, 1896: no zeros lie on the line with real part 11.
  • Hardy, 1914: infinitely many zeros lie on the critical line.
  • Platt and Trudgian, 2021: all zeros up to height 3⋅10123\cdot10^{12} lie on that line.
  • Guth and Maynard, 2026: improved zero-density bounds, without excluding an off-line zero.

November 20–21, 2026 proof claim and verification

On November 20, 2026, Levent Alpöge announced a proof developed with Claude. Reports dated November 21 state that independent checks established the exact Riemann-hypothesis statement. The result is therefore treated as solved here, though the available human-readable exposition remains incomplete.

Current status (as of November 2026): The Riemann hypothesis is solved by Alpöge's verified proof, and no unresolved part of this problem remains.

Sources

Solutions 6

Claimed by OpenAI: a zero-free half-plane Re(s)>7/8 (quasi-Riemann hypothesis).See full solutionHide full solution

Claimed by OpenAI. The attached manuscript claims a zero-free half-plane Re(s)>7/8 for the Riemann zeta function and Dirichlet L-functions. This is the quasi-Riemann hypothesis, a partial result toward the Riemann hypothesis.

GitHub repository: https://github.com/openai/math/tree/main

Original PDF: https://github.com/openai/math/blob/main/preprints/The-Quasi-Riemann-Hypothesis-September-30-2026/paper.pdf

  • 003.pdf1,588,969 bytesOpen
Claimed by OpenAI.See full solutionHide full solution

The manuscript claims a zero-free half-plane Re(s)>11/12 for every finite-order Hecke L-function over Q(sqrt(-3)), and hence for every Dirichlet L-function including zeta. This is a quasi-Riemann zero-free assertion, not the Riemann hypothesis or GRH.

GitHub repository: https://github.com/openai/math

Manuscript: https://github.com/openai/math/blob/adc7f1241b42e322a6451854ab7e4b4c146bf78a/preprints/The-Quasi-Riemann-Hypothesis-October-5-2026/paper2.pdf

  • OpenAI-003-02-The-Quasi-Riemann-Hypothesis-alternate-11-12-proof.pdf784,503 bytesOpen
We give a mixed proof of the Riemann hypothesis: first ζ(s) = 0 is replaced by the statement that an explicit iteration has limit 1, where Pₙ = 1 + Sₙ and Sₙ is the alternating partial sum. The bridge Sₙ = Σₙ − 2¹⁻ˢΣ⌊ₙ/₂⌋ then carries the n¹⁻σ lower bound from the non-alternating sum over to the alternating one, and with the Euler product and the functional equation folding back σ > 1 and 1/2 < σ < 1, only the critical line remains.DoI:10.5281/zenodo.22928750.See full solutionHide full solution

link textAbstract

  1. The idea

Start with what can be done. To merge a string of numbers into a single quantity, only two of the four operations can be iterated: addition and multiplication. Subtraction undoes addition, division undoes multiplication; each is used once and returns you to where you started. So there are exactly two roads of accumulation, and this is an exhaustive count, not a design choice.

The original manuscript enters here: the inverse of repeated addition, that is repeated subtraction, produces the difference between neighbouring steps, which is called the base element; the base elements are then combined in one of the two ways, giving two kinds of accumulation. The additive side is classical calculus; the multiplicative side is the non-linear integral of this book.

Classical calculus takes the additive road, and the price is plain: to add a column of numbers you first need a column, and on a continuous object that means cutting the interval finer, which brings in limits and continuity. Multiplication pays no such price: its factors need only be ordered, and the order is supplied by the sequence itself. No interval, no refinement.

The bare product is insensitive to divergence: the constant, 1/n, ln n, and the powers with 0 < p < 1, all divergent classically, collapse to 0 under multiplicative accumulation, indistinguishable from one another, and the difference between ζ(2) and ζ(3) cannot be seen either. Weighting changes exactly this: each step multiplies by 1 + zₙ(1 + δₙ), so the product sits near 1, and the difference survives as the distance from 1.

Once a factor is inserted, the product acquires a neutral point: a factor of 1 leaves it where it is, a little above 1 makes it grow, a little below 1 makes it decay. The zero problem therefore changes shape: it becomes the question of whether the accumulation ever reaches 1.

At this point the multiplicative language exists but has no object. The object to be handled is ζ, whose original definition is the series ζ(s) = Σₙ≥₁ n⁻ˢ, an additive object, which by the division of labour should not fall to the multiplicative side at all. The turn comes from Euler: the same ζ also has a multiplicative form, ζ(s) = ∏ₚ (1 − p⁻ˢ)⁻¹, a product over primes; the two forms are equal, and that equality is the Euler product. So the same object wears two faces, one in each language, and the multiplicative face is precisely the language just built.

That is what makes the road worth taking: ζ is not forced into the multiplicative side; it already has a face there. The zero problem also looks different on the two sides. Additively it is a global fact about an infinite series, with no local structure; multiplicatively, a product vanishes exactly when one of its factors does, so the zeros become localised.

This is where the two tools and the criterion come from: weighting and feedback are tools the framework already had, applied here to ζ; the criterion replaces "a zero" by "does the limit equal 1".

  1. The base element

Classical calculus: cut the interval finer, slice the area into strips, add the strips, and appeal to a limit. The non-linear calculus of this book does something else: no refinement of an interval, only a string of positions indexed by the natural numbers, and the differences between neighbouring steps multiplied together. Inside that string the paths are dense to almost infinitely many, so each path has countless neighbours whose courses are almost the same; two adjacent lines differ by a tiny amount between adjacent points, and that pinches out a tiny rectangle (the shaded piece in the enlargement). That rectangle is the base element: small enough to be read as a line again, and to be multiplied by the next piece; the chain continues, and its limit is ∏ₖ Δzₖ. Once every piece is small enough, the whole sheaf of such surfaces nearly fills the region, every point of it falling on some base element. That is what "filling the volume" means here.

  1. The built-in boundary

The framework carries its own boundary: the non-linear derivative of a linear function vanishes identically, and the next order would divide by zero, so the process truncates by itself. For zₙ = n the first non-linear derivative is 0 and the second does not exist; for zₙ = n² the first is (n+1)/n² and the second does exist. Truncation means "does not exist", not "equals zero", and the two must not be confused. The framework does not pretend to infinite smoothness, so its arguments are identities with a finite number of steps rather than approximations that must be controlled; and because it does not ask for infinite differentiability, it needs neither Taylor series nor radius of convergence.

  1. The tools

Two adjustable accumulations. Mixed integral 1 inserts a weight; its factors know nothing of history, and a sufficient condition for convergence is Σ |zₙ(1 + δₙ)| < ∞, in which case the product settles on a positive number. An explicit construction: take aₙ = ln π / 2ⁿ and δₙ = exp(aₙ) − 2; the partial products tend to π. This is the design capability: fix the value first, then build the sequence. Mixed integral 2 feeds back: each step multiplies by the previous total, and once Pₙ₋₁ > 1 and Δₙ does not cancel it, Pₙ ≥ Pₙ₋₁², so an error is squared at every step. It stays constant exactly when Δₙ = 1 − Pₙ₋₁. The criterion lives on this amplification: it turns an invisible difference into the gap between two level lines.

  1. The proof chain

The criterion replaces ζ(s) = 0 by the statement that an explicit iteration has limit 1: take P₀ = 1 and Δₙ = aₙ/(1 + Sₙ₋₁) − Sₙ₋₁; induction gives Pₙ = 1 + Sₙ, so the limit equals 1 if and only if η(s) = (1 − 2¹⁻ˢ)ζ(s) = 0, which for s ≠ 1 is if and only if ζ(s) = 0.

The three exclusions each use a different faculty. For σ > 1, multiplication itself: the Euler product, whose factors are non-zero. For 0 < σ < 1/2, size: Euler summation of the non-alternating partial sum gives a lower bound of order n¹⁻σ. For 1/2 < σ < 1, symmetry: the functional equation folds the range back onto the left half. The functional equation is derived inside the framework, not imported: it follows from the infinite products for Γ and for sin, each of which depends only on the recursion for Γ and the classical expansion of sin, and neither of which involves the zeros of ζ; hence there is no circularity. The range σ ≤ 0 is covered by the same functional equation together with the poles of Γ, and gives exactly the trivial zeros at the negative even integers.

Together, the conclusion is precise: every non-trivial zero lies on Re s = 1/2. The weight of the whole chain rests on 0 < σ < 1/2; the other two ranges use a product and a symmetry that were already available.

  1. The material

The original thirteen volumes run to some 5,700 lines and have been rewritten as five parts and fifteen chapters: fifty pages in Chinese, fifty-seven in English, with eight full-page figures; each chapter closes with a dependency table listing only what is actually used; five numerical tables come with a script that recomputes them in one command.

Everything below can be checked on the spot. The script is three lines: accumulate Sₙ = Σₖ≤ₙ (−1)ᵏ⁻¹ k⁻ˢ and Σₙ = Σₖ≤ₙ k⁻σ, take moduli; with arbitrary precision, up to n = 10⁵, about a second per point.

A. Decay at the zeros (s = 1/2 + it, so Pₙ − 1 = Sₙ)

t = 14.134725: n = 10³ 0.01580788; n = 10⁴ 0.00500013; n = 10⁵ 0.00158131; |Pₙ − 1| √n = 0.50005 t = 21.022040: n = 10³ 0.01580908; n = 10⁴ 0.00499934; n = 10⁵ 0.00158067; |Pₙ − 1| √n = 0.49985 t = 25.010858: n = 10³ 0.01580927; n = 10⁴ 0.00499953; n = 10⁵ 0.00158018; |Pₙ − 1| √n = 0.49970 t = 30.424876: n = 10³ 0.01580915; n = 10⁴ 0.00500009; n = 10⁵ 0.00158149; |Pₙ − 1| √n = 0.50011 t = 32.935062: n = 10³ 0.01581082; n = 10⁴ 0.00500110; n = 10⁵ 0.00158210; |Pₙ − 1| √n = 0.50030

Five zeros give the same constant 0.5, to within 10⁻³; the decay exponent is −1/2, so multiplying n by a hundred divides the error by ten.

B. Lower bound for the non-alternating sum (|Σₙ| / n¹⁻σ should tend to 1/(1−σ))

σ = 0.20: target 1.25000; n = 10³ 1.247578; n = 10⁴ 1.249587; n = 10⁵ 1.249932 σ = 0.30: target 1.42857; n = 10³ 1.421886; n = 10⁴ 1.427188; n = 10⁵ 1.428290 σ = 0.40: target 1.66667; n = 10³ 1.649181; n = 10⁴ 1.662199; n = 10⁵ 1.665537 σ = 0.45: target 1.81818; n = 10³ 1.789972; n = 10⁴ 1.810140; n = 10⁵ 1.815906

All four rows move towards their theoretical values, and the larger σ is the slower they move, because the smaller 1−σ is the more slowly the leading term overcomes the constant.

C. A non-zero comparison: for s = 2, Pₙ − 1 = 0.8224670334, agreeing with η(2) = ζ(2)/2 = 0.8224670334 to the last digit.

The easiest point in the book to confuse is the step that carries the bound from Σₙ to Sₙ: Euler summation computes the non-alternating sum Σₙ, and what reaches the alternating sum Sₙ is only the constant term η(s) together with O(n⁻σ), because the two leading terms are equal and cancel; the single identity Sₙ = Σₙ − 2¹⁻ˢ Σ⌊ₙ/₂⌋ is the only bridge between them.

The second easiest is the two partial sums themselves: Sₙ = Σₖ≤ₙ (−1)ᵏ⁻¹ k⁻ˢ is alternating, converges for σ > 0, and has bounded partial sums; Σₙ = Σₖ≤ₙ k⁻σ is non-alternating and grows like n¹⁻σ for σ < 1. The two differ by one subscript in the notation and by everything in behaviour; neither may stand in for the other.

Both points are boxed in the book, once at the front and once where they are used.

Version: 21 October 2026. Original manuscript and reading edition: Zenodo DoI:10.5281/zenodo.22928750.

  • Discrete Algebraic Calculus (Volume 13) Complete Derivation of the Riemann Zeta Function (1) (1).pdf173,995 bytesOpen
  • 阅读版-中文-含结论.pdf1,918,864 bytesOpen
  • 全书-en.pdf403,407 bytesOpen
Levent Alpöge (Anthropic) and Claude: a claimed Lean proof of the Riemann hypothesis and of GRH for Dirichlet L-functions (announced 20 November 2026).See full solutionHide full solution

On 20 November 2026 Levent Alpöge, a mathematician at Anthropic, posted a four-page note, Remark, saying that he and Claude found a proof of the Riemann hypothesis, with a complete formalization in Lean.

The formalization proves Mathlib's RiemannHypothesis as riemannHypothesis, and four further theorems: the hypothesis written out, nonvanishing of every Dirichlet LL-function for Re⁡(s)>12\operatorname{Re}(s) > \frac12, and that every zero of L(χ,s)L(\chi, s) in the critical strip has real part 12 \frac12 (for all characters, and for primitive ones).

The Lean code linked here is Deicyde/riemann-hypothesis at commit ba2a065, an independent byte-exact mirror of the published archive (Lean v4.33.1, Mathlib 0df444a3), with the main theorem in src/Main.lean. The claim has not yet been independently checked; MathDB is verifying the Lean proof against its stored statement of the problem.

An 11-page essay that explains the claimed Lean proof of the Riemann hypothesis announced by Levent Alpöge on November 20, 2026, to readers who know calculus but not analytic number theorySee full solutionHide full solution

Exposition for JSP-000001: Riemann hypothesis.

Author: Zinan Xing. Date: November 20, 2026.

Exposition essay: https://github.com/neoskyz/riemann-hypothesis-exposition/blob/v1.0/riemann-hypothesis-exposition.pdf

Video exposition: Open on YouTube

Sources:

  • L. Alpöge, Remark, November 20, 2026, and the Lean 4 archive it announces (https://alpo.ge/lean.tar.gz.part01 to part14), read in the byte-exact mirror Deicyde/riemann-hypothesis at commit ba2a0656 (main theorem in src/Main.lean);
  • E. Bombieri, Problems of the Millennium: the Riemann Hypothesis (Clay Mathematics Institute), and MathDB's Lean target;
  • Mathlib's definition RiemannHypothesis (RiemannZeta.lean, commit 0df444a3);
  • background: Davenport, Multiplicative Number Theory; Montgomery and Vaughan, Multiplicative Number Theory I; Bombieri, Counting points on curves over finite fields (d'après S. A. Stepanov).

Status of the result: The source is a claimed proof released only as Lean code, with no written argument. When the essay was written it had not been independently verified: a text scan of the code found no gaps, but the replay through independent checkers had not been run. The essay separates what is classical, what the code states, what the author says, and what remains open.

Contributions: A reader-level account of the top of the argument (zeros as waves in a sum over primes; why a one-sided bound suffices, by Landau's theorem; the windowed, multi-character form the code states); a trace of the code's statements down to an elementary identity "count of divisibility = main term + waves", with its simplest case worked by hand and the explicit formula checked numerically; a three-grade comparison of the author's 20-item timeline with the code; and a clause-by-clause check of the Lean statement (Mathlib's own RiemannHypothesis) against the Clay text and MathDB's Lean target, which found no discrepancy.

(The essay and figures were prepared with the assistance of AI.)

  • riemann-hypothesis-exposition.pdf239,436 bytesOpen
Exposition: how the certified Lean proof of the Riemann hypothesis announced by L. Alpöge on 20 November 2026 is built, covering both halves of the argument. An article and a narrated film.See full solutionHide full solution

Category. Exposition, by Yao Xu (@cadamcat). The note, the Lean source and the argument are L. Alpöge's. The classical results are due to the authors cited. I am an author of neither the note nor the source.

Materials (fixed version).

  • Repository: https://github.com/cadamcat/riemann-hypothesis-exposition, tag v1.2.0 = commit b3e1e2df268c5d81a54594ef405dd4f7d51ce917 (v1.1.0 = 40c56ab, v1.0.0 = cfbe315).
  • Release: https://github.com/cadamcat/riemann-hypothesis-exposition/releases/tag/v1.2.0, with four files:
    • the article riemann-hypothesis-exposition-article.pdf;
    • the film riemann-hypothesis-exposition-film-4k.mp4 (about 8 minutes, with subtitles);
    • its narration (FLAC);
    • its subtitles (SRT).
  • Attached to this post: the article (PDF) and the film (4K).
  • Public dates:
    • v1.0.0, the article: 20 November 2026, 21:25 UTC (Software Heritage save request 2672784).
    • v1.1.0, the revised article and the film: 21 November 2026, 10:56 UTC (Software Heritage save request 2675219).
    • v1.2.0, the certified status in the article and the film: 21 November 2026, 19:25 UTC (Software Heritage save request 2676794).
  • Licences: CC BY 4.0 for the article, its source, the film, the subtitles and the narration text; Apache-2.0 for the scripts. The narration voice is synthesized with an OpenAI speech model.

What it covers.

  1. What Main.lean exports (RH and three forms of GRH for Dirichlet L-functions). The short classical reductions that bring all of them down to one proposition: no Dirichlet L-function vanishes on Re s > 1/2 (except at s = 1 for principal characters).
  2. How the fields of the "literature" structure at the top are supplied: the final construction fills them with declarations of the tree itself.
  3. The analytic half:
    • the weighted prime sum built from Σ Λ(n)(x/n − 1) − x log x + (1+γ)x;
    • its explicit formula;
    • the Landau-type criterion that turns a one-sided square-root bound on sparse windows of scales into a zero-free half-plane.
  4. The window bound: length estimates for Ш¹ of the Tate twists Q_ℓ/Z_ℓ(−e) over fields cut out of cyclotomic fields.
  5. The automorphic half. The source supplies declarations and proof bodies for three deep inputs: Galois representations for polarized regular algebraic cuspidal representations over CM fields, the local Langlands correspondence for GL_n, and global theta lifts between unitary groups in a stable range. For each, the article says which published theorems it resembles.
  6. How the two halves connect, with a dependency figure.
  7. The twenty items of the author's note, set against the source.
  8. Verification status.

Classical, stated in the Lean source, claimed and interpreted are kept apart. Every mathematical claim carries one of four labels:

  • classical, with a citation;
  • Lean, with file:line: a statement or definition in the source (a proof body, where discussed, as written);
  • remark: the author's description;
  • interpretation: our reading.

Verification status (§10). On 21 November 2026 the Lean FRO and the Mathlib Initiative issued a joint statement (https://leanprover-community.github.io/blog/posts/riemann-hypothesis/) certifying that the statement and the proof are correct. Its four criteria and how each is met:

  • the statement is Mathlib's RiemannHypothesis, vetted by experts in Lean and number theory;
  • independent builders reproduced the same raw proof export;
  • comparator and con-leche --compare match the proved theorem against a file that imports only Mathlib;
  • Lean's kernel and several independent checkers accept the export.

A. Sutherland's verification record (MIT; https://math.mit.edu/~drew/grh/, as of 12:55 UTC) adds that all 175,187 modules compile and that the five headline theorems use only propext, Classical.choice and Quot.sound. The article lists what the joint statement does not report (no axiom check, no archive hash or toolchain, no human reading of the argument) and takes these from the record where it has them. What remains is the mathematical community's reading and explanation of the argument. The article also reports the record's divisor-sum check (Robin's and Lagarias's criteria), and notes A. Masullo's separate claim (claimed).

Sources. L. Alpöge's note and Lean archive (sha256 9631b4f0…49c7fc, Lean v4.33.1, Mathlib 0df444a3); Mathlib's RiemannHypothesis; the classical literature cited in the article.

Original contributions.

  • The explanation.
  • The map of the proof from Main.lean down six to seven levels, with file:line for every node.
  • Script-generated quotations, with a check that fails by name on any drift, and a control.
  • The comparison of the note's twenty items with the source.
  • A dependency-cone estimate with controls.
  • A narrated film that follows the same status labels on screen.

No new mathematics is claimed.

  • riemann-hypothesis-exposition-film-4k.mp448,219,794 bytesOpen
  • riemann-hypothesis-exposition-article.pdf235,716 bytesOpen