INDEPENDENT PUBLICATION · AMERICAN AMBITION BECOME A FOUNDING SPONSOR NEWS & ANALYSIS / DATED AT PUBLICATION

MATHEMATICS / LATEST COVERAGE

A Claude-assisted proof of the Riemann Hypothesis passes independent checks.

Mathematician Levent Alpöge released the proof in Lean. Independent checkers confirmed that it proves the statement mathematicians have been asking about since 1859.

A straight line in the story of primes.

Prime numbers can seem to turn up wherever they please. But a deep clue to how often they occur lies in an equation that doesn’t look like a list of primes at all. In 1859, Bernhard Riemann proposed that the nontrivial zeros of the zeta function, the points where its value drops to zero other than a known set of exceptions, all line up on one vertical line. As the Clay Mathematics Institute explains, proving that would sharpen what mathematicians can say about how primes are distributed.

On November 20, mathematician Levent Alpöge released a proof developed with Anthropic’s Claude. The proof arrived as a library written in Lean, a language for stating and checking mathematical arguments. The Lean and Mathlib communities, with the endorsement of the organizations that support them, say they are confident that both the statement and the proof are correct.

More than typing up the math.

In his release note, Alpöge says he steered the work toward combining two ideas that had worked in a simpler setting. He credits Claude with technical suggestions near the finish, including how to scale up a key construction. In his account, the model helped figure out the argument itself, not just translate a finished proof into code.

The release runs to roughly 17 million lines of Lean. Alpöge says he wants to write up and explain the argument and expects it can be simplified. For now, the formal proof gives other mathematicians a result to build on while they work out which ideas carry the most weight.

Check the answer. Then check the question.

A proof checker can only tell you whether you proved the statement you gave it. The verification team also checked that this was the right statement: independent tools compared the theorem in Alpöge’s library with the standard definition of the Riemann Hypothesis in Mathlib, Lean’s community mathematics library. Multiple checkers then accepted the exported proof. Andrew Sutherland’s verification notes at MIT document those checks and an independent rebuild that reproduced the proof export exactly.

That’s the most useful lesson for other researchers working with Super Intelligence (SI): a model can help find an argument, and a separate, reproducible check can establish exactly what that argument proves.