Command Palette

Search for a command to run...

Lean experts certify an AI-assisted Riemann proof—but the math still needs explaining

The groups behind the Lean proof assistant say the proposed proof passes their checks, while a human-readable explanation and Clay’s assessment remain outstanding.

Ken Richardson / Wikimedia Commons, CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/); cropped.

A landmark check

The organizations behind the Lean proof assistant and its mathematical library certified an AI-assisted proof of the Riemann hypothesis on Saturday, saying independent checks had confirmed both the statement and the argument. The result is an extraordinary endorsement for a proposed solution to a 167-year-old conjecture, although mathematicians still have to understand the enormous proof and the Clay Mathematics Institute continues to list the problem as unsolved.leanprover-community +1

Anthropic mathematician Levent Alpöge released the argument on Friday, saying he developed it with the company’s Claude AI. His submission contains about 17 million lines of Lean code, rather than a conventional paper. Alpöge said he intends to explain the argument and write it up properly.leanprover-community +1

Why primes are at stake

Proposed by Bernhard Riemann in 1859, the hypothesis concerns the distribution of prime numbers. It predicts that the relevant zeros of a mathematical function called the Riemann zeta function all lie on a particular vertical line. A proof would sharpen mathematicians’ understanding of how primes are distributed among the integers.claymath

The problem is one of Clay’s Millennium Prize Problems, each carrying a $1 million award. A formal verification is not itself a Clay prize decision; the institute’s page still calls the hypothesis unsolved.claymath +1

What the computers checked

The Lean FRO and Mathlib Initiative said reviewers confirmed that the formal statement matches the hypothesis, that Alpöge’s library compiles, and that independent tools checked the proof and matched its conclusion against the established statement. The groups said they had “full confidence” in the proof’s correctness.leanprover-community

MIT mathematician Andrew Sutherland separately published a detailed verification log. In his results as of 12:55 UTC Saturday, independent checkers accepted the exported proof and all 175,187 modules passed a separate replay. His own full comparator run was still listed as in progress; the later community statement reports a successful comparator check without identifying it as his run.leanprover-community +1

The remaining human challenge

Formal checking establishes that the encoded statement follows from specified assumptions according to the checking systems; it does not automatically make the underlying ideas easy to explain. Alpöge acknowledged that the argument was exceptionally long and said he had not yet mastered all the mathematics around it.npr +1

Number theorist James Maynard told Nature that, if the result fully checks out, it would be a historic milestone. The Lean groups now say they look forward to mathematicians explaining the proof’s new ideas.leanprover-community +1