# Riemann Hypothesis proof certified — Lean checks Alpöge and Claude's proof

> Lean experts say the AI-assisted Riemann Hypothesis proof is correct, after several independent checkers and statement tools agreed.

- Org: Lean and Mathlib communities
- Date: 2026-11-21
- Category: showcase, paper
- Source: https://leanprover-community.github.io/blog/posts/riemann-hypothesis/
- Canonical page: https://ai-tldr.dev/releases/lean-mathlib-riemann-hypothesis-proof-certified/

## In short

The Lean and Mathlib communities certify that Levent Alpöge's 17-million-line Lean proof of the Riemann Hypothesis, built with Claude, is correct. Several independent proof checkers accept it, and its statement matches Mathlib's definition.

## What it is

On November 21, 2026 the Lean and Mathlib communities published a joint statement, endorsed by the Lean FRO and the Mathlib Initiative, saying they have full confidence that Levent Alpöge's proof of the Riemann Hypothesis is correct. Alpöge, a mathematician at Anthropic, released the proof one day earlier as a Lean library of 17 million lines, written with Claude. The statement certifies both the theorem statement and the proof.

## How it works

The certification follows four tests that Kevin Buzzard's Xena blog set out for AI-written Lean code. The proved theorem was compared with Mathlib's own RiemannHypothesis definition by two independent tools, comparator and con-leche. Dozens of people compiled the library and got the same proof export. That export was then accepted by Lean's kernel and by separate checkers such as nanoda, con-leche and con-ron, which ran on an inert copy so malicious code could not interfere.

## Why it matters

Machine-checked certification removes the main doubt about AI-written proofs: that the code proves a slightly different statement or exploits a bug in one checker. Andrew Sutherland's MIT verification page shows the five checked theorems cover the Riemann Hypothesis and its generalization to all Dirichlet L-functions. That makes this the first famous open problem settled with an AI model and confirmed by the formal-proof community, although a readable human write-up is still to come.

## Who it's for

mathematicians, formal-verification and AI-for-math researchers

## Try it

Read the verification log at https://math.mit.edu/~drew/grh/

## Quick facts

- Certified by: Lean and Mathlib communities, endorsed by Lean FRO and Mathlib Initiative
- Proof author: Levent Alpöge (Anthropic), working with Claude
- Proof size: 17 million lines of Lean, depends only on Mathlib
- Theorems checked: Riemann Hypothesis + GRH for all Dirichlet L-functions
- Axioms used: propext, Classical.choice, Quot.sound only
- Independent checkers: Lean kernel, nanoda, con-leche, con-ron, lean4lean
- Not covered: Dedekind, Hecke and general automorphic L-functions

## FAQ

### Who certified the Riemann Hypothesis proof?

The Lean and Mathlib communities certified Levent Alpöge's Riemann Hypothesis proof in a joint statement on November 21, 2026, endorsed by the Lean FRO and the Mathlib Initiative. MIT mathematician Andrew Sutherland published a separate verification page with logs. Scientific American quotes him saying the chance that anything is fishy is approximately zero.

### Could the Lean proof of the Riemann Hypothesis exploit a bug in Lean?

The certifiers tested for that. The exported Riemann Hypothesis proof was accepted by Lean's official kernel and by independent checkers written in other languages, including nanoda, con-leche and con-ron. According to the Lean statement, con-leche and con-ron come with a formal proof that they cannot accept a false proof, which addresses worries about flaws in Lean's own logic.

### Does Alpöge's proof also settle the Generalized Riemann Hypothesis?

Partly. Andrew Sutherland's MIT verification page lists five checked theorems, including the Riemann Hypothesis and the claim that zeros of every Dirichlet L-function in the critical strip lie on the critical line. Those cover all moduli and characters. Dedekind zeta functions, Hecke L-functions and general automorphic L-functions were not among the checked statements.

### Is there a readable paper for the certified Riemann Hypothesis proof?

Not yet. The certification covers the 17-million-line Lean library, not a human-written paper. Levent Alpöge's original statement says he intends to understand the argument, simplify it and write it up properly. The Lean community says it looks forward to work that will explain the new ideas in the proof.

## Links

- [Lean Joint Statement](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/)
- [MIT Verification Files](https://math.mit.edu/~drew/grh/)
- [Alpöge Statement (PDF)](https://alpo.ge/remark.pdf)
- [Scientific American](https://www.scientificamerican.com/article/the-ai-riemann-hypothesis-proof-is-almost-certainly-true-independent-mathematicians-attest/)
- [Xena: Lean Proof Tests](https://xenaproject.wordpress.com/2026/11/12/and-the-proof-is-checked-in-lean/)
- [HN Discussion](https://news.ycombinator.com/item?id=50589618)
