Anthropic mathematician Levent Alpöge said on November 20 that he believes he and Claude have proved the Riemann hypothesis—and its broader generalized form. He has published a proposed proof in computer-checkable form, putting an extraordinary claim before mathematicians. Expert checks of both the encoded question and the argument remain pending, Scientific American reported.
An earlier attempt stopped short
First proposed in 1859, the Riemann hypothesis concerns the distribution of prime numbers. It predicts that the zeros of the Riemann zeta function that determine the primes lie along a particular vertical line. These zeros are places where the function takes the value zero; their positions describe increasingly fine details of the sequence of primes. Many mathematical results assume the hypothesis is true.
Claude had already attempted the problem earlier this year. In an August 10 account, Anthropic explicitly said that effort had not succeeded. Instead, the company reported that an unreleased research version improved a lower bound on the proportion of zeros on the line, from 41.6% to 67.2%. That was progress on a related question, not a proof of the hypothesis itself.
A subsequent paper by Alpöge and Ralph Furman states that at least two thirds of the relevant zeros lie on the line and are simple, meaning they are not repeated roots. Its provenance note credits Claude with discovering the proof autonomously and the authors with verifying and communicating it. The paper also states that the argument was formally verified in Lean 4.
Alpöge describes Claude pulling ahead
The new claim describes a different relationship between researcher and model. In his posted statement, Alpöge recounts months of work with Claude across a wide range of mathematics. He says he directed the effort toward combining two ideas that appeared in a simplified setting, then guided it toward the corresponding situation needed for the larger problem.
From there, he credits Claude with rapid technical advances near the finish. His account is not a claim that a model solved the problem without human direction: he describes his own choices about the approach while emphasizing that Claude eventually moved beyond his command of some of the mathematics. He specifically acknowledges limited mastery of one area used in the argument.
A rough timeline in the statement dates several key steps to November 15 and ends with a November 16 entry labeled “GRH,” short for Generalized Riemann Hypothesis. Alpöge says Claude assembled that timeline after searching through the work. He cautions that some technical descriptions may need editing because he is paraphrasing material he does not yet fully understand.
This puts us in the rather extraordinary situation this age has made possible, in which I am presenting the world with a complete argument before having mastered a massive neighbourhood around its mathematics.
Levent Alpöge, in his posted statement
Code now, an explanation later
For the public release, Alpöge chose Lean, a programming language used to express mathematical arguments so software can check their steps. He says the formalization is finished and links to an archive split into 14 parts. He also says the method resolves the generalized hypothesis, while describing an extension to the Extended Riemann Hypothesis only as an expectation for further work.
The distinction between an encoded proof and the intended mathematical claim is central to the checks ahead. Scientific American notes that experts must establish whether the hypothesis has been properly encoded, as well as whether the proof is correct. The presence of Lean files therefore does not, by itself, settle the announcement’s central claim.
Alpöge says the current argument is extremely long. He rejected publishing an enormous, unreadable prose version and instead released the formal files. He expects substantial simplification, but says he has not yet had time to search for shorter arguments himself. His stated next task is to understand the argument more fully, explain it in talks and produce a proper written account.
Reader comments
Newest comments first. Replies stay oldest first.