What was certified
The Lean and Mathlib communities certified on Saturday that Levent Alpöge’s proof of the Riemann hypothesis is correct, closing the central question around the most famous claim in mathematics this year. Alpöge, a mathematician at Anthropic, built the proof over months working with the company’s Claude model.
The joint statement, endorsed by the Lean Focused Research Organization and the Mathlib Initiative, says the communities have full confidence in the proof and congratulate Alpöge. It certifies that both the statement and the proof of the Riemann hypothesis are correct, after a day of checks by dozens of mathematicians.
Alpöge released the proof on Friday 20 November as a Lean library of about 17 million lines of code, with a note on his website. The hypothesis, first posed by Bernhard Riemann in 1859, holds that the nontrivial zeros of the zeta function all lie on a single line.
The statement checked is Mathlib’s own vetted definition of the hypothesis, reviewed by experts in Lean and number theory. That detail matters, because a proof in code is only as good as the statement it proves, and the communities confirmed the code proves exactly that definition.
How the checks ran
Dozens of people independently compiled the 175,187-module library without errors, reproducing the same raw proof export. Andrew Sutherland of MIT rechecked it with only the three standard axioms and said the chance anything is fishy is approximately zero, in a running log of checks.
Two independent tools, comparator and con-leche –compare, confirmed the theorem proved is exactly Mathlib’s statement of the hypothesis. The raw proof was also checked by Lean’s official kernel and by independent checkers written in different programming languages, on an inert export.
What is still missing
No human-readable paper exists yet: the proof lives only as Lean code, and Alpöge has said he does not yet fully understand every argument Claude used to produce it. He plans an account, and the communities look forward to work explaining the new ideas.
The Clay Mathematics Institute still lists the hypothesis as unsolved, and its one-million-dollar prize is years away under its own rules. Its president, Martin Bridson, called the announcement historic but said the road to a place of real understanding will be long and difficult.
The result is the most striking yet from Claude’s work in mathematics. The same model recently helped solve two black-hole physics problems and a centuries-old Fermat problem in about thirty hours, in research Anthropic published earlier this month, the latest in a run of AI-assisted mathematics results.
Our Read
What changed this weekend is not the claim but the verdict on it. On Friday the proof was an announcement with checks pending; by Saturday the people who maintain the checking tools had certified it. That is the difference between a headline and a result mathematicians can build on.
The certification also draws a line the earlier coverage could not. Compilation alone never settled whether the code proved the hypothesis as stated, and the comparator run answered that question. Two independent tools agreeing on the statement turns a machine check into a mathematical fact.
Why It Matters
A certified proof of the Riemann hypothesis would be the first Millennium Prize problem settled with AI assistance, and it changes what frontier models are credibly for. Mathematics has been the proving ground for machine reasoning; a result at this scale moves the debate from demos to discoveries.
Background
Alpöge says the same method also settles the broader Generalized Riemann Hypothesis, a claim the certification does not separately address. A separate GRH proof submitted by Stanford postdoc Alessandro Masullo, posted by mathematician Ravi Vakil, remains unverified and is not part of this certification.
Source: Lean community blog, Scientific American


0 Comments