Technology

A Claimed Riemann Hypothesis Proof From an Anthropic Mathematician, and a Separate Human Proof

Martin HollowayPublished 20m ago4 min readBased on 1 source
Reading level
A Claimed Riemann Hypothesis Proof From an Anthropic Mathematician, and a Separate Human Proof
Image by athree23 from Pixabay

Levent Alpöge, a mathematician employed by Anthropic, has announced in an essay a claimed solution to the Riemann Hypothesis, which he says is Lean-compilable. He says he found pieces of the argument with help from Claude. TechCrunch

The Riemann Hypothesis dates to 1859 and is among the Millennium Prize problems. It concerns a hidden structure behind the distribution of prime numbers, and it has resisted resolution for well over a century. Any claimed solution therefore faces scrutiny that few mathematical announcements ever receive.

According to the report, Alpöge messaged Dario Amodei, Anthropic's CEO, on a Sunday night to ask how he should announce his finding. The detail is unusual. A researcher asking his employer's chief executive for advice on a public announcement is not the usual route by which a result enters the mathematical literature.

Lean is an open-source language used to verify mathematical proofs. An argument described as compiling in Lean has been encoded in a formal language, so a proof assistant can check each inferential step mechanically. The check is only as meaningful as the formal statement it verifies. Whether that statement faithfully captures the original conjecture is a question for specialists who read the formalization. The compiler cannot answer it.

Alpöge also released a proof related to the Navier-Stokes equations, which govern fluid flow. He co-authored it with Tristan Buckmaster, a mathematician at NYU. The two should be read as separate threads.

On the same problem, Ravi Vakil, a Stanford professor and president of the American Mathematics Association, posted a proof of the General Riemann Hypothesis written by Alessandro Masullo, a Stanford postdoc. That proof was described as all human, with no AI involved. Masullo said it was the result of twelve years of work. TechCrunch

Anthropic has become the first publicly traded frontier AI lab, and Amodei leads it as chief executive. The listing means the company's research claims now reach public-market investors as well as the mathematics community, a new condition for a result of this kind.

In my view, the more useful question is not which proof came from which process, but how the community will check either one. Formal verification moves part of the burden from reputation to specification, which is a real gain. The specification step is still human work, as is the judgment about whether a formal statement says what the original conjecture says. Read together, the two announcements point to a shift in where the bottleneck sits. Generating candidate arguments is becoming easier with machine help. Judging those arguments is still slow, and it still depends on people who have spent years in the field.

The broader context here is encouraging, if cautiously so. If machine-assisted search can surface a line of attack that a researcher then develops and formalizes, the tools that speed up checking may matter as much as the tools that generate candidates. Whether the Alpöge claim or the Masullo proof holds will be settled by specialists reading the work line by line. That process has always been slow, and in this case it is also the part of the story that will determine what the announcements mean for the field.