The purported proof of the Riemann hypothesis has surfaced exclusively in the form of Lean code. Initial independent verification has found no issues, though a competing proof released on the same day by a human mathematician has already faced scrutiny for potential errors.

Levent Alpöge, a mathematician employed by Anthropic, announced on Friday that he and the company's Claude AI model have solved the Riemann hypothesis, widely regarded as mathematics' most celebrated unsolved problem. His announcement took the form of a four-page document posted to his website, with the actual proof rendered only in Lean, a formal verification system that validates each logical step.

Bernhard Riemann first posed the hypothesis in 1859, concerning the distribution of prime numbers along the number line. The Clay Mathematics Institute lists it among its Millennium Prize Problems, each carrying a $1 million reward. According to Alpöge, his approach simultaneously establishes the Generalized Riemann Hypothesis, an even more expansive mathematical claim.

I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis

Levent Alpöge

Code instead of paper

No traditional mathematical paper accompanies the announcement. Alpöge explained that a human-readable version would prove impractically large, noting in a footnote that his own draft manuscript exceeded 250,000 pages. He characterized releasing such a document as a "travesty of 'slop'", justifying his decision to publish only the Lean code.

Alpöge acknowledges that he has not yet fully grasped all the mathematical reasoning underlying the proof. He attributes the crucial final steps to Claude's capabilities. A timeline included in his announcement, which he states Claude assembled, spans from 6 October through 16 November.

Claude far outpaced me near the finish line. No, Claude flew, and danced, and sparkled near that finish line

Levent Alpöge

Alpöge has committed to eventually understanding the argument more thoroughly, refining it and preparing a proper writeup.

Initial verification

MIT mathematician Andrew Sutherland independently compiled the Lean proof on Friday, according to Scientific American's reporting.

Honestly the chance that anything is fishy here is approximately zero

Andrew Sutherland, MIT

An automated verification process, known as a comparator run, remained outstanding. This check would confirm that the code actually proves the Riemann hypothesis as stated, rather than demonstrating something subtly different. James Maynard of the University of Oxford expressed to The New York Times that he was "98 or 99 percent" confident in the result.

Terence Tao, the UCLA-based mathematician, posted on his blog that fully comprehending the proof will require considerable time. His preliminary assessment indicates that the proof leverages the complete scope of contemporary number theory.

Timing questions

The announcement arrived just one day following Anthropic's landmark initial public offering. Alpöge noted that the mathematical breakthrough occurred late on Sunday evening, after which he contacted chief executive Dario Amodei regarding the timing of disclosure. He referenced "a constraint to respect" in his decision-making.

There is an obvious cynical view, to do with the timing, but it is false

Levent Alpöge

Anthropic's stock price declined on Friday as broader market weakness in AI-related shares took hold.

Pushback and competing claims

Stanford mathematician Brian Conrad voiced objections to Alpöge's approach. Releasing a proof that its author does not fully comprehend struck Conrad as "frankly completely outrageous", he told The New York Times.

On the same day, AMS president Ravi Vakil introduced an alternative proof authored by mathematician Alessandro Masullo. Vakil characterized this as entirely human-generated, without AI involvement, and noted that Conrad had been reviewing it for several months. Within approximately eight hours, Vakil posted an update indicating that multiple reviewers had identified issues in Masullo's proof using AI tools.

Alpöge previously gained attention in July when he posted a counterexample to the Jacobian conjecture that Claude had identified. On Friday, he and Tristan Buckmaster of NYU also announced that they have demonstrated the Navier-Stokes equations can develop singularities without external forcing.

Source: The Next Web