The Lean and Mathlib communities certified on November 21, 2026 that Levent Alpöge’s 17-million-line Lean library accurately states and proves the Riemann Speculation, following a day of compilation, statement-matching, and proof-kernel checks. Their joint assertion was endorsed by the Lean FRO and the Mathlib Initiative.
The certification follows Alpöge’s announcement on November 20, 2026, by which he stated he found and formalized the proof whereas working with Claude. The discharge contained solely the Lean formalization; Alpöge seen a standard write-up at this stage as an enormous, unreadable doc and deliberate to know, simplify, and clarify the arithmetic correctly.
How the Proof Was Checked
The codebase relies upon instantly solely on Mathlib, the group’s library of formalized arithmetic. The assertion being proved was already a part of Mathlib and had been vetted by specialists in Lean and quantity idea, addressing the requirement that the formal assertion specific the supposed speculation. Dozens of individuals reported compiling the library with out errors or construct failures and reproducing the identical uncooked proof export.
To test the concept’s identification, a community-added problem file remoted the Riemann Hypothesis assertion and imported solely Mathlib. Comparator and con-leche’s comparability mode independently confirmed that Alpöge’s theorem matched that assertion precisely. The uncooked proof export was checked with Lean’s official kernel, con-leche, con-ron, and extra third-party checkers written in several programming languages.
These checks ran on an inert export, past the attain of hypothetical malicious code within the Lean sources. Con-leche and con-ron even have formal proofs, relative to set idea, that they can’t settle for a false proof, addressing considerations {that a} flaw in Lean’s logic might be exploited.
What the Verification Establishes
Andrew Sutherland’s MIT verification record, final up to date at 12:55 UTC on November 21, 2026, stories that every one 175,187 modules compiled with zero errors utilizing Lean and Mathlib model 4.33.1; his checks had been run with Claude Code. An axiom audit discovered that the 5 headline theorems depended solely on propext, Classical.alternative, and Quot.sound, with no axioms declared outdoors Lean core. Nanoda, an impartial Rust kernel, and con-leche every accepted the complete exported proof, lean4lean accepted each compiled module, and con-leche’s comparability mode matched each problem assertion.
At that web page’s cutoff, the Lean default and paranoid kernel checks on the export and the complete lake comparator run remained in progress. Older bundled con-leche and con-ron binaries exited and not using a verdict as a result of they didn’t assist one reflexive inductive kind, not as a result of they rejected the proof. Sutherland additionally disclosed that the construct ran on unsandboxed cloud machines and that its export hash had not been in contrast with one from a sandboxed lake construct; a passing consequence nonetheless depends on the kernels used and on Mathlib’s definitions representing their supposed arithmetic.
The central theorem makes use of Mathlib’s formal definition: each zero of the Riemann zeta perform apart from the trivial zeros at damaging even integers should have actual half one-half. The separate situation s ≠ 1 seems as a result of Mathlib assigns the zeta perform a standard worth at its pole. The submission additionally verifies Generalized Riemann Speculation statements for Dirichlet L-functions throughout all moduli and characters; Sutherland expressly excludes Dedekind zeta capabilities, Hecke L-functions, and normal automorphic L-functions from the checked scope.
Claude’s Function within the Discovery
Alpöge’s account assigns Claude a task in discovery in addition to formalization: he guided the work from a toy setting into a similar goal state of affairs earlier than Claude outpaced him close to the end. In a single instance, Claude recommended {that a} von Mangoldt sum of size X ought to use a series of theta lifts of size X, a technical alternative Alpöge highlights in his rationalization of the argument.
Alpöge describes the present argument as extraordinarily lengthy, expects substantial simplification, and remains to be working to know and clarify it. His account describes human steering and Claude’s technical contributions, with each formalizing the arguments with out further assumptions. The certification considerations the formal artifact; that account provides the premise for describing Claude’s position, slightly than demonstrating an autonomous answer.
The certification resolves the formal-checking query with out resolving the explanatory one. Whether or not the proof’s new concepts could be simplified and absorbed by mathematicians stays the central open query across the verified Lean library.

