Twenty-Four Hours to a Verified Theorem: How the Lean Community Signed Off on a Riemann Hypothesis Proof
The people who maintain the Lean proof assistant and its Mathlib library have issued a joint statement with an unusual headline: "we have full confidence that Alpöge's proof of the Riemann Hypothesis is correct."
The statement, published on the Lean community blog on November 21 and endorsed by the Lean FRO and the Mathlib Initiative, closes the verification loop on a claim that would have taken human peer reviewers years. On November 20, Anthropic mathematician Levent Alpöge released a proof of the Riemann Hypothesis as a Lean library of 17 million lines of code. By the next afternoon, dozens of community members had rebuilt it, cross-examined what it actually proves, and run the exported proof through independent checkers.
We covered the claim itself on November 20, when the formalization was still unverified, and the competing 940-page human proof by Carlo Masullo on November 21. The certification is the new event, and it is worth reading closely, because the method matters as much as the theorem.
What the community checked
The joint statement evaluates the release against four criteria drawn from a post by Timothy Chow on the xena project blog. Each maps to a distinct failure mode that a machine-checked proof can still have.
First, the statement itself has to be the real Riemann Hypothesis. The community points to Mathlib's own RiemannHypothesis definition, which has been vetted over time by Lean users and number theorists. A proof of a subtly different proposition would be worthless, so this is checked separately from the proof.
Second, the library has to compile. Dozens of people reported building Alpöge's codebase without errors and reproducing the same raw proof export file. A theorem that only compiles on one machine in one configuration is not yet a result.
Third, the code has to prove that vetted statement. Here the community did something more careful than trusting an import chain. They added their own file that isolates the Riemann Hypothesis and imports only Mathlib, then compared it against the theorem Alpöge's library proves, using two independent tools, comparator and con-leche --compare. Both confirmed the two are the same statement.
Fourth, the proof object itself has to be valid. The exported raw proof was checked by Lean's official kernel, by the con-leche and con-ron checkers, and by several third-party checkers written in other programming languages. The two named checkers come with a formal proof, relative to set theory, that they cannot accept a false proof. That is a direct answer to the standing worry that a bug in Lean's logic itself could be exploited. All of this ran on an inert export of the proof, out of reach of any hypothetical malicious code living in the Lean sources.
An overview of the checking effort is hosted by MIT number theorist Andrew Sutherland at math.mit.edu/~drew/grh/.

Why this is fast, and what it does not cover
Traditional peer review of a serious proof runs on human attention. A 17-million-line library is beyond any review panel, which is exactly the argument for formalization: the machine does the verification and humans only have to audit the machine's specification. The Lean community's four-step process is designed so that no single trusted component decides everything. Compilation was reproduced by dozens of people. Statement equivalence was confirmed by two tools. Proof validity was confirmed by checkers in multiple languages, two of which are themselves formally proven sound.
What the statement does not claim is understanding. The community's closing line is a look forward to "all the activity in the mathematical community that will help explain the new ideas in this proof." Correctness is certified; the mathematical contents of the proof are still to be absorbed, and the broader question of how much of the work Claude did, which framed the original November 20 coverage, sits outside the Lean statement entirely.
The Riemann Hypothesis, the 167-year-old conjecture about the zeros of the zeta function, has now been proven under conditions where the verification does not depend on the reader trusting one author or one toolchain. That is a new bar for what "proof accepted" means, and it was cleared in about a day.
Sources
- Joint statement by the Lean and Mathlib communities: https://leanprover-community.github.io/blog/posts/riemann-hypothesis/
- Alpöge's proof release (remark.pdf): https://alpo.ge/remark.pdf
- Timothy Chow, "And the proof is checked in Lean?": https://xenaproject.wordpress.com/2026-11-12/and-the-proof-is-checked-in-lean/
- Andrew Sutherland's verification overview: https://math.mit.edu/~drew/grh/
- Mathlib RiemannHypothesis definition: https://leanprover-community.github.io/mathlib4_docs/Mathlib/NumberTheory/LSeries/RiemannZeta.html#RiemannHypothesis
- Techmeme pick: https://www.techmeme.com/261122/p3
