Anthropic's Levent Alpöge claims a Riemann hypothesis proof
The Anthropic mathematician says he and Claude proved the 1859 conjecture and posted a Lean-formalised proof online, but no one has yet independently verified it.
- What happened
- Anthropic mathematician Levent Alpöge said on 20 November that he had found a proof of the Riemann hypothesis while working with Claude.
- What’s new
- He posted the proof, written in the Lean language, on his own website and says the method also settles the Generalized Riemann Hypothesis.
- Why it matters
- If it holds up, the result would settle what Scientific American calls the most famous open problem in mathematics, and with an AI's help.
Levent Alpöge, a mathematician who works at Anthropic, said on 20 November that he has proved the Riemann hypothesis with help from the company's Claude model. "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," he wrote in a statement on his website.
The hypothesis was proposed by Bernhard Riemann in 1859 and concerns the distribution of prime numbers. It is one of the seven Millennium Prize problems, each carrying a $1 million award from the Clay Mathematics Institute. According to Tom's Hardware, the institute says the first 10,000,000,000,000 relevant solutions are known to fit the conjecture, short of a proof that it holds for all of them.
What Alpöge claims
Alpöge says the same method goes further and resolves the Generalized Riemann Hypothesis, and he expects that with more time it will also cover the Extended Riemann Hypothesis. He says he and Claude formalised the arguments without assumptions. Hector Pasten of the Pontifical Catholic University of Chile told Scientific American that a correct, understood proof would link branches of mathematics whose connections had been only partial.
The Verge reports that Alpöge's document says he texted Anthropic CEO Dario Amodei about the result and suggests he delayed publishing until after Anthropic's IPO on Thursday. Anthropic has not commented on the claim or taken part in the announcement, The Verge says. In August, Anthropic said it was working on the problem with Claude but did not expect the techniques used to lead to a proof.
How it will be checked
The proof has so far appeared only on Alpöge's website and has not been independently verified. Scientific American notes that it is written in Lean, a language that checks the correctness of each step; problems can still occur, but the outlet expects experts to determine fairly quickly whether the hypothesis was encoded properly and the proof is sound. Formal recognition is slower: Clay's rules, as The Verge describes them, require a published solution to survive at least two years of scrutiny and win general acceptance among mathematicians. Only the Poincaré conjecture has been officially solved since the prize was set up.
A year of AI mathematics claims
Alpöge previously said in July that he had used Claude to solve the Jacobian conjecture. He was also caught up in the September dispute over OpenAI's claimed Navier-Stokes solution, when NYU mathematician Tristan Buckmaster suggested OpenAI had used the route he and Alpöge were pursuing. OpenAI has denied wrongdoing, The Verge reports.