All News
mathematicsriemann-hypothesisleanformal-verificationmathlib

Lean and Mathlib communities certify Alpöge's Riemann proof checks

The Lean and Mathlib communities certified that Alpöge's Riemann proof type-checks. That settles the machine-checked artifact, not the mathematics itself.

Vlad MakarovVlad Makarovreviewed and published
2 min read
Lean and Mathlib communities certify Alpöge's Riemann proof checks

One day after Levent Alpöge released a Lean library he says proves the Riemann Hypothesis, the people who maintain the mathematics it depends on put their names to it. A joint statement on 21 November, endorsed by the Lean FRO and the Mathlib Initiative, certifies that both the statement and the proof are correct, and that all four of the community's own verification criteria are satisfied.

What they actually checked

Alpöge's library, released on 20 November, is 17 million lines of Lean, and depends directly only on Mathlib. The statement runs through four tests. The theorem is stated as Mathlib's own RiemannHypothesis, vetted by Lean and number-theory experts. Dozens of people compiled the library and reproduced the same raw proof export, without errors or build failures. A new file isolating that statement imports only Mathlib, and two tools, comparator and con-leche --compare, confirmed it matches the theorem Alpöge proved. Finally, the export was checked by Lean's official kernel, by con-leche and con-ron, and by third-party checkers written in other languages.

The fourth test carries the weight

That last one answers the objection which dominated the first day: that nobody could know whether the file was even well-formed, or whether Lean's own logic contained a flaw the proof could exploit. con-leche and con-ron come with a formal proof, relative to set theory, that they cannot accept a false proof. The checks ran on an inert export, out of reach of hypothetical malicious code in the Lean sources, and MIT hosts an overview of the runs.

A certification by the people who own the statement

This certifies a formal object. It says the Lean statement is Mathlib's statement and that the proof term checks. It does not say a human can follow the argument, or that the ideas are new; the statement ends by looking forward to "all the activity in the mathematical community that will help explain the new ideas in this proof." It is also an insider certification, since the body doing the certifying maintains Mathlib and the four criteria are the community's own, published on the Xena project blog on 12 November and written for AI-generated Lean specifically. There is still no paper, no referee report and no readable exposition.

The same evening, TechCrunch reported a separate, all-human claim: a draft proof of the Generalized Riemann Hypothesis by Stanford postdoc Alessandro Masullo, circulated by Ravi Vakil as "all human, with no AI." We take no view on it.

What changed is narrower than the headlines and more useful than most of them. Yesterday's announcement was a file nobody could vouch for. Today it is a file two communities have vouched for, and a mathematics still waiting for its explanation.

Related Articles

Scroll down

to load the next article