Lean community: 17-million-line Riemann proof made with Claude is correct
The Lean and Mathlib communities say the proof by Levent Alpöge and Claude proves exactly the Riemann hypothesis and passes several independent checks. Mathematicians do not yet understand the ideas behind it, however.
Anthropic mathematician Levent Alpöge released his proof on 20 November as Lean code rather than as a conventional paper. Lean is a proof assistant: the program mechanically checks that every step of reasoning follows from the basic rules of mathematics. The Riemann hypothesis concerns the distribution of prime numbers and has been regarded as the most important unsolved problem in mathematics.
Written out as ordinary text, the proof would run to more than 250,000 pages, Alpöge says. He also says the method solves the generalised Riemann hypothesis. The Lean community’s statement concerns the Riemann hypothesis.
The statement, which the Lean and Mathlib communities published on 21 November, rests on four checks. The formulation of the hypothesis is part of the Mathlib library and has been checked by experts. Dozens of people have compiled the code without errors. Two separate tools confirmed that the theorem Alpöge proved is exactly Mathlib’s formulation. Several independent checkers also accepted the proof. Two of them have been formally proven unable to accept a false proof.
The statement calls the third check decisive: a machine verifies only the theorem as written in the code. Critics faulted OpenAI’s September Navier–Stokes claim on exactly this point, Latent Space reported. They argued that its Lean proof did not match the natural-language result.
Machine acceptance does not mean understanding, though. ‘Does that mean I understand it? Not yet,’ Alex Kontorovich, a mathematician at Rutgers University, told AP. The Clay Mathematics Institute, which has offered a $1 million prize for a solution, still listed the hypothesis as unsolved hours after the announcement, AP reported. Alpöge acknowledges that he does not yet fully grasp the mathematics surrounding the proof. He plans to write a readable version in the coming months.
It has not been disclosed which Claude model was used or how much Alpöge directed the work, Latent Space reported. The proof was announced a day after Anthropic’s stock market listing. Alpöge denies that it is a PR stunt.
Hours after Alpöge’s announcement, Stanford professor Ravi Vakil published Alessandro Masullo’s draft proof, written entirely by a human. Latent Space put the draft at about 940 pages. Number theorist Brian Conrad had been reviewing it for months. About eight hours later, Vakil said several people had reported problems in the draft, The Decoder reported. The problems had been found with the help of different AI models.