The Lean and Mathlib communities have certified Levent Alpöge’s proof of the Riemann Hypothesis. What began as a dramatic proof announcement now carries an explicit endorsement supported by published verification records.
Their November 21 joint statement, endorsed by the Lean FRO and the Mathlib Initiative, addresses both essential questions: whether the formal theorem says what mathematicians mean by the Riemann Hypothesis, and whether its proof is logically valid. Alpöge released the formalization on November 20, 2026, as a Lean library containing 17 million lines of code.
“We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct,” the joint statement says.
The AI connection needs more precise attribution. MIT mathematician Andrew Sutherland’s verification page explicitly credits Anthropic’s Claude Code with helping run checks. The opened sources do not establish how much Claude contributed to constructing Alpöge’s proof. What they document in detail is the verification.
What Lean and Mathlib Certified
The Lean and Mathlib communities’ certification covers four separate checks: the mathematical statement, successful compilation, the identity of the theorem proved, and the validity of the proof.
Those distinctions matter. A project can compile while proving a different proposition from the one advertised. A theorem can also depend on an additional assumption that effectively grants the desired conclusion. Checking the final result requires more than a successful build.
The community statement reports that:
- Mathlib’s existing definition of the Riemann Hypothesis has been vetted by Lean and number theory experts.
- Dozens of people compiled Alpöge’s library and reproduced the same raw proof export.
- Two independent tools confirmed that Alpöge’s theorem matches Mathlib’s statement.
- Multiple proof checkers accepted the exported proof.
My view is that the separation between these checks is the most consequential detail. The certification is stronger because it identifies several distinct ways a formalization could go wrong and provides evidence addressing each one.
The Exact Riemann Hypothesis Statement
Alpöge’s project, built with Lean v4.33.1 and Mathlib v4.33.1, proves Mathlib’s existing RiemannHypothesis proposition, according to the MIT verification record.
In ordinary mathematical language, the statement says that every nontrivial zero of the Riemann zeta function has real part one-half. The formal definition excludes the trivial zeros at −2, −4, −6 and so on. It also excludes the point (s = 1), because Mathlib assigns the zeta function a conventional value at its pole.
The MIT page lists the headline theorem declaration as:
theorem riemannHypothesis : RiemannHypothesis
That declaration identifies the proposition being proved. The proof itself is supplied by the formal library; the one-line declaration is not a standalone proof.
Crucially, the community created a separate file importing only Mathlib to isolate the target statement. Comparator and con-leche --compare then checked its identity against the theorem in Alpöge’s library. Both confirmed a match, the joint statement reports.
That prevents a subtle but serious failure mode: proving something with a familiar name whose underlying definitions differ from the accepted target.
Five Headline Theorems Were Audited
The MIT page records five headline theorems:
| Theorem | Mathematical scope |
|---|---|
| Riemann Hypothesis | Mathlib’s existing definition |
| Expanded zeta statement | The same assertion written explicitly in the submission |
| Dirichlet L-function nonvanishing | No zeros to the right of real part one-half, with the stated pole exception |
| Dirichlet L-function zeros | Zeros inside the critical strip have real part one-half |
| Primitive-character version | The corresponding assertion for primitive Dirichlet characters |
MIT explicitly limits the generalized Riemann Hypothesis scope here to Dirichlet L-functions. Readers should not expand that description into a claim about every broader conjecture carrying the GRH name.
How the Independent Proof Checkers Were Used
The Lean v4.33.1 project was checked through compilation, kernel replay, exported-proof validation and statement comparison. These operations inspect different artifacts, so their results should not be treated as interchangeable.
A proof assistant’s kernel checks whether the formal derivation follows its logical rules. Checking an exported proof with separately implemented kernels reduces reliance on any one implementation.
The MIT verification page, updated November 22 at 04:35 UTC, records the following:
| Check | Recorded result | Important condition |
|---|---|---|
| Lean compilation | All 175,187 modules compiled | Lean and Mathlib v4.33.1; zero errors |
leanchecker module replay | All 175,187 modules passed | Checks compiled modules through the Lean kernel |
nanoda | Full exported proof accepted | Independently written Rust kernel |
con-leche | 1,468,418 declarations accepted | Updated build running in verified mode |
con-ron --compare | All five target theorems matched | Updated build checked solution and challenge declarations |
lean4lean | All 175,187 compiled modules passed | Five required a higher computation-step limit |
The community statement additionally reports that Lean’s official kernel, con-leche, con-ron and other independently implemented checkers accepted the raw proof.
There is a reporting distinction worth preserving: MIT’s snapshot still lists its separate default and paranoid Lean-kernel runs over the complete export as running. Its module-by-module kernel replay had finished. Those are different checks; the snapshot should not be summarized as saying every listed run had completed.
Checker versions also matter. Older con-leche and con-ron binaries bundled with Lean v4.36.0 declined the proof because they lacked support for one inductive type. MIT identifies that as a compatibility limitation, not a proof rejection; updated builds accepted the relevant checks.
The community statement says con-leche and con-ron have formal soundness proofs relative to set theory. That strengthens confidence while retaining explicit foundations and implementation assumptions.
Reproduction, Axioms and the Audit Trail
The Lean v4.33.1 proof export was reproduced byte-for-byte in an independent sandboxed build, according to MIT’s verification record.
The sandbox had no network access and read-only source files. Its export matched the published artifact at , approximately 61.6 GB in decimal units. MIT notes that one compiled module differed nondeterministically, even though the exported proof matched exactly.
Frequently Asked Questions
3 questions
1What Did Lean and Mathlib Certify?
The communities certified both the formal statement and the proof of Alpöge’s Riemann Hypothesis result. Their joint statement cites successful compilation, a match to Mathlib’s vetted definition and acceptance by multiple proof checkers. The certification concerns formal correctness, while mathematical exposition remains separate work.