All news articles

Lean community backs Anthropic mathematician’s Riemann hypothesis proof

The Lean and Mathlib communities say they have full confidence in Levent Alpöge’s AI-assisted proof after computer checks. Explaining the argument and gaining wider acceptance remain separate challenges.

AI-generated editorial illustration based on the published article topic; not a documentary photograph.
Loading statistics…

The communities behind the Lean proof-checking language and its Mathlib library say they have “full confidence” that Anthropic mathematician Levent Alpöge’s proof of the Riemann hypothesis is correct. Their assessment follows checks of both the statement and the formal proof, building on earlier computer tests.

Alpöge said on 20 November that he and Anthropic’s AI assistant Claude had found a proof. The hypothesis, formulated in 1859, concerns the distribution of prime numbers. Clay has offered a $1 million prize for a solution.

Rather than a conventional paper, Alpöge published the argument in Lean, which allows computers to check mathematical reasoning. He said he had guided the approach and that Claude had contributed to the argument. He also said he wanted to understand and explain the mathematics better.

Verification notes put the Lean source at more than 17 million lines. MIT mathematician Andrew Sutherland reported that all 175,187 modules compiled without errors and that two independently written checking programmes accepted the full exported proof. His checks used Claude Code.

A separate project on the LUMI supercomputer reported compiling the same modules in a network-isolated environment. It said the hashes, or digital fingerprints, of two compiled files matched Sutherland’s results.

The Lean and Mathlib communities said two tools confirmed that the theorem proved in Alpöge’s code matched Mathlib’s definition of the hypothesis. They also said several proof checkers, including Lean’s official kernel, had checked the exported proof.

In Sutherland’s report, updated at 12.55 UTC on 21 November, his own full Comparator run and a separate Lean-kernel replay of the export were still in progress.

The checks test whether the formal argument follows the rules of the system; they depend on the checking programmes and mathematical definitions used. They do not themselves provide a conventional, human-readable explanation of the reasoning.

Sutherland says the checked generalised result concerns a class of functions called Dirichlet L-functions, rather than every possible related function. Rutgers mathematician Alex Kontorovich told the Associated Press that he did not yet understand the proof and would need to do more work.

The Clay Mathematics Institute still lists the hypothesis among its unsolved problems. Before considering a proposed solution for a prize, its rules require publication in a qualifying outlet, Independently verifiedRules for the Millennium Prize Problems - Clay Mathematics Institute to have passed and general acceptance by the global mathematics community.

Clay’s president, Martin Bridson, called Alpöge’s announcement historically significant but said, in a statement reported by the Associated Press, that “the journey from today’s announcement to a place of real understanding will be long and difficult”.

Sign in to give this article a thumbs up or down.

From the discussion

Private test discussion. Comments are readers’ views and are not yet automatically fact-checked. Editing is available for 60 seconds after posting.

Sign in with a confirmed reader account and choose a username to read comments and participate.

Sorting applies to top-level comments; replies remain oldest first. New comments and likes can change the order. Refresh for the current ranking.

Loading comments…