Skip to main content
Live
Main content

Lean community certifies Claude-assisted Riemann hypothesis proof

Joint statement backed by the Lean FRO and Mathlib Initiative says Levent Alpöge's 17-million-line Lean proof checks out, a day after release.

Jaeden Schafer
Editor in Chief · · 5 min read
Anthropic logo

The Lean and Mathlib communities have certified Levent Alpöge's Claude-assisted proof of the Riemann hypothesis, issuing a joint statement on November 21 that reads: "We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct." The statement, endorsed by the Lean FRO and the Mathlib Initiative, landed one day after Alpöge, a mathematician at Anthropic, released the proof.

Alpöge's release was not a paper. It was a Lean library of about 17 million lines of code that depends directly only on Mathlib, the community-maintained library of formalized mathematics. The Lean and Mathlib communities said they have "full confidence" the proof is correct and congratulated Alpöge on the result.

The certification rests on four checks. The theorem being proved is Mathlib's own RiemannHypothesis definition, which number theory and Lean experts had vetted before Alpöge's release. The library compiles, and dozens of people reproduced the same raw proof export. Two tools, comparator and con-leche in comparison mode, confirmed Alpöge's theorem matches that statement exactly. And the exported proof passed several independent checkers, including Lean's kernel, con-leche, con-ron and third-party checkers written in other languages.

Key facts

  • 01The Lean and Mathlib communities certified Levent Alpöge's Riemann hypothesis proof on November 21, 2026.
  • 02The proof is a Lean library of about 17 million lines, released November 20 and depending directly only on Mathlib.
  • 03MIT's Andrew Sutherland reports all 175,187 modules compile with zero errors on Lean and Mathlib v4.33.1.
  • 04Checked theorems cover RH and GRH for Dirichlet L-functions, but not Dedekind, Hecke or automorphic L-functions.
  • 05Lean's default and paranoid kernel replays were still running as of 02:25 UTC on November 22.

The most detailed public record comes from MIT mathematician Andrew Sutherland, who ran his checks with Claude Code. As of 02:25 UTC on November 22, his page reports that all 175,187 modules compile with zero errors on Lean and Mathlib v4.33.1. The five headline theorems rely only on Lean's three standard axioms: propext, Classical.choice and Quot.sound.

Sutherland's independent kernel replays also passed. The Rust-based checker nanoda accepted the full export in 27 minutes and 42 seconds on 80 threads, and con-leche accepted 1,468,418 declarations in verified mode. A rebuild on a separate machine, with no network access and a read-only source tree, produced an export matching the published 61.6-billion-byte file.

“the chance that anything is fishy here is approximately zero.”
— Andrew Sutherland, MIT mathematician

Two checks were still running at Sutherland's last update: Lean's default and paranoid kernels replaying the export, and an end-to-end lake comparator run. Sutherland notes that a passing result shows the formal statements follow from the standard axioms according to the kernels used, and that it relies on Mathlib's definitions meaning what they are intended to mean.

The scope reaches past the classic conjecture. The checked theorems include the Generalized Riemann Hypothesis for Dirichlet L-functions across all moduli and characters. Dedekind zeta functions, Hecke L-functions and automorphic L-functions in general are not covered. Sutherland also combined the proof with an existing Lean formalization to derive Robin's inequality for every n greater than 5040.

Alpöge's account of how the proof was built describes a collaboration. He guided the work from a toy setting toward an analogous target before Claude outpaced him near the finish. In one example he cites, Claude proposed that a von Mangoldt sum of length X should use a chain of theta lifts of the same length. The certification concerns the formal artifact; the description of Claude's role comes from Alpöge.

Related · from this week
Nvidia launches $4,999 DGX Spark 64GB for local AI development
Jaeden Schafer · 4 min read →

Humans have not yet absorbed the argument. Alpöge said the AI-written informal explanation, which runs past 250,000 pages, is not worth publishing, and he promised a human-readable account. He expects substantial simplification. Brian Conrey, a leading Riemann hypothesis specialist who reviewed Anthropic's earlier work on zeta zeros, said he "really did not see this coming at all."

The conjecture dates to 1859 and carries a $1 million Clay Mathematics Institute prize. Clay's rules require publication in a qualifying outlet, a two-year wait and general acceptance by the mathematical community, so formal verification alone does not settle the award.

Anthropic has not issued a separate statement on the proof. Alpöge posted it a day after the company's Nasdaq debut, wrote that "the precise timing is horrible," and said leadership approved the release. The result follows Anthropic's September formalization of Fermat's Last Theorem, about 13 million lines of Lean written by Claude agents in roughly 11 days, and an August result in which an unreleased Claude model raised the proven share of zeta zeros on the critical line from 41.6% to 67.2%.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Models

Nvidia logo
Models

Nvidia launches $4,999 DGX Spark 64GB for local AI development

The new SKU runs 100B-parameter models on device; two units cluster to 128GB and 200B via Nvidia Sync Cluster Assistant.

Jaeden Schafer4 min read
OpenAI logo
Models

OpenAI's Astra and Anthropic's Claude Opus 5 crack unsolved Enigma messages

Two cryptanalysts used frontier LLMs to break wartime codes that had resisted human researchers for two decades.

Jaeden Schafer4 min read
Anthropic logo
Models

Anthropic says Claude autonomously discovered a Crispr-like enzyme system

Nearly 950 Claude agents worked 21 hours across 210 million tokens to flag a previously uncharacterized enzyme found in bacteriophages.

Jaeden Schafer4 min read