Science

Anthropic mathematician says Claude produced a Riemann hypothesis proof

23 views

What was claimed

On Friday, Levent Alpöge, a mathematician on the technical staff at Anthropic, published a short note and a formal proof file. "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," he wrote. He added that the method extends further, resolving the Generalized Riemann Hypothesis as well.

The Riemann hypothesis, proposed by Bernhard Riemann in 1859, concerns the zeros of the Riemann zeta function and, through them, the distribution of prime numbers. It is one of the seven Millennium Prize problems. A correct proof would settle a question that has shaped number theory for more than a century and would validate a large body of results that currently depend on it being true.

How the work was done

Alpöge said the collaboration lasted about six weeks and used a version of Claude that is not publicly available. The output was written in Lean, a programming language used to check mathematical proofs line by line, and runs past 250,000 lines. Anthropic confirmed that two of its mathematicians reviewed the paper and wrote a short explanatory note for experts.

Formal verification is not automatic proof of correctness. Errors can enter when a statement is translated into Lean, and a verified file only shows that the coded argument follows from the coded assumptions. Checking that those assumptions match the claim is the part that takes human attention.

Why mathematicians are cautious

Terence Tao said verification is still some way off, though the presence of a human author who can answer questions should help. Others noted that Anthropic published a related result in August, on a lower bound for the share of zeros satisfying the hypothesis, and said at the time that the techniques involved were not expected to reach the full conjecture.

The claim also lands in a crowded year. AI systems have produced a stream of results in higher mathematics, some verified and some still under review, and journals are still working out how to handle machine-assisted proofs. Alpöge's note is unusually direct: it asserts the full conjecture, not a step toward it.

Until outside experts finish reading the Lean file, the honest description is a claim awaiting verification.

Source: BBC / Scientific American / The Verge