SIAgentNews.com
← All news
DevelopingClaudeResearch

Lean and Mathlib communities certify Claude-assisted Riemann Hypothesis proof

The formal verification of a 17-million-line proof marks a milestone for AI-assisted mathematics while leaving the human-readable explanation for now.

· First reported by Unite.AI

Ink and watercolour: picture for the story 'Lean and Mathlib communities certify Claude-assisted Riemann Hypothesi'
AI-made ink and watercolour for this story. No real people or products are shown.
17,518,7number of modules compiled with zero errors

The short version

  • The certified Lean library contains 17,518,7 modules and 17 million lines of code.
  • The proof was developed by Levent Alpöge with significant technical contributions from Claude.
  • Multiple independent kernels, including Nanoda and con-leche, accepted the exported proof.
  • The certification confirms the formal artifact but does not yet provide a simplified human explanation.

Why it matters

This event demonstrates that AI can assist in producing massive, verifiable mathematical proofs that exceed human capacity for manual checking. It shifts the focus of mathematical discovery toward formal verification as a primary standard of truth.

The formal certification

The Lean and Mathlib communities officially certified a formal proof of the Riemann Hypothesis on November 21, 2026. This certification followed an announcement by Levent Alpöge, who stated he discovered and formalized the proof while working with the AI model Claude. The community spent a full day performing statement-matching, compilation, and proof-kernel checks to validate the work.

The certification confirms that the formal artifact correctly states and proves the hypothesis within the Lean framework. However, the community noted that the proof's mathematical explanation remains unresolved at this stage. Alpöge has described the current write-up as a massive, unreadable document that requires significant simplification before it can be properly understood by the broader mathematical community.

The joint statement endorsing the work was supported by the Lean FRO and the Mathlib Initiative. This marks a significant milestone because the Riemann Hypothesis is one of the most famous unsolved problems in mathematics. While the formal logic is now verified, the underlying ideas still need to be distilled into a format that humans can easily absorb and study.

How the proof was verified

To ensure the proof was accurate, the community relied on Mathlib, which is a library of formalized mathematics vetted by experts. The specific statement being proved was already part of Mathlib and had been reviewed by number theory experts to ensure it expressed the intended hypothesis. Dozens of people reported compiling the library without any build failures or errors.

The verification process involved checking the theorem's identity using a community-added challenge file. This file isolated the Riemann Hypothesis statement and imported only Mathlib to ensure no outside code interfered. Comparison tools independently confirmed that Alpöge’s theorem matched that specific statement exactly, providing a layer of security against hidden errors.

The raw proof export was then tested against several different kernels, including the official Lean kernel, con-leche, and con-ron. These checkers were written in different programming languages to provide diverse validation. Because these checks ran on an inert export, they were protected from any hypothetical malicious code that might exist within the Lean sources themselves.

The role of Claude AI

Alpöge credits Claude with playing a role in both the discovery and the formalization of the proof. He guided the process from a simple toy setting into a complex target situation where Claude eventually outpaced him near the finish. For example, Claude suggested using a specific chain of theta lifts for a von Mangoldt sum, a technical choice Alpöge highlighted in his explanation.

The interaction between the human and the AI was collaborative rather than autonomous. Alpöge provided the human guidance necessary to navigate the problem, while Claude contributed technical choices and helped formalize the arguments without requiring extra assumptions. The certification focuses on the final formal artifact rather than the specific mechanics of how the AI arrived at the solution.

This partnership highlights a new workflow where AI can handle the heavy lifting of formalization. While the AI provided the technical path, the human researcher remained in control of the goals and the high-level direction. This distinction is important for understanding how AI is currently being integrated into high-level scientific research and mathematical verification.

Scope and limitations

The verification includes the Generalized Riemann Hypothesis statements for Dirichlet L-functions across all moduli and characters. However, the scope of the checked work specifically excludes certain types of functions. Sutherland expressly excluded Dedekind zeta functions, Hecke L-functions, and general automorphic L-functions from the verified scope.

The central theorem uses Mathlib’s formal definition regarding the zeros of the Riemann zeta function. This definition requires that every zero other than the trivial zeros at negative even integers must have a real part of one-half. A separate condition regarding the value at its pole was also included because Mathlib assigns the zeta function a conventional value there.

While the formal-checking question is resolved, the explanatory question remains the next major hurdle. The central challenge now is whether the new ideas within the 17-million-line proof can be simplified. Mathematicians will need to work with Alpöge to understand the logic so it can be absorbed into the existing body of mathematical knowledge.

What to watch

  • How Alpöge plans to simplify the 17-million-line proof into a human-readable format.
  • Whether other major mathematical conjectures will be targeted for AI-assisted formal verification next.

Try it

Look up the MIT verification record for the Riemann Hypothesis to see the list of kernels that accepted the proof.

Sources and links

Read the original at Unite.AI

First reported by Unite.AI. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news