Research

AI-generated text

Researchers release separate claims concerning the Riemann hypotheses

Levent Alpöge says he has found a Lean-formalized proof of the Riemann Hypothesis with help from Anthropic’s Claude.

Researchers release separate claims concerning the Riemann hypotheses

Mathematician Levent Alpöge, an employee of Anthropic, says he has found a proof of the Riemann Hypothesis that compiles in Lean. The conjecture dates to 1859. Around the same time, Alessandro Masullo, a postdoctoral researcher at Stanford University, released a draft proof of the General Riemann Hypothesis. Both claims still need to be examined by the mathematics community.

The Riemann Hypothesis concerns the structure underlying the distribution of prime numbers and is one of the Millennium Prize Problems. An accepted proof would be a major result in number theory.

Searching for a proof with Claude

Alpöge described the result and the circumstances surrounding it in an essay. He wrote that on the previous Sunday evening he had messaged Dario Amodei, Anthropic’s chief executive, to ask how to announce the discovery. Alpöge had been occupied with the problem since childhood and felt unable to wait longer to release his finding, even though a fully human-intelligible explanation of the proof was not yet available.

According to his account, he worked through considerable uncertainty with Anthropic’s Claude model to find the components of the argument. Alpöge has committed to pursuing an explanation that people can understand and communicate to one another.

Lean is an open-source language that helps verify mathematical proofs. Even so, understanding and assessing the formalized result may take time. In a professional discussion on Zulip, Chris Birbeck, a number theorist at the University of East Anglia, identified several advanced mathematical tools in the argument. Confidence in the formalized proof does not replace detailed scrutiny by the wider community.

Alpöge and Tristan Buckmaster, a mathematician at New York University, have also released another purportedly important proof related to the Navier–Stokes equations. Problems associated with those equations helped prompt leading AI labs and mathematicians to turn their attention to outstanding mathematical questions over the summer.

A separate draft without AI

Meanwhile, Ravi Vakil, a Stanford University professor and president of the American Mathematics Association, posted Masullo’s draft proof of the General Riemann Hypothesis. Vakil said the work was entirely human and did not use artificial intelligence.

The original plan was for Masullo to spend a year at Stanford refining the text and have it reviewed by enough specialists before an announcement. Vakil said the circumstances had instead made it preferable to release the current draft now.

In the draft’s introduction, Masullo wrote that the work was the product of twelve years. He also addressed mathematicians discouraged by the pace and tone of the current professional climate, emphasizing that human researchers remain present and at work.