Dev.to AI 🤖 Ai 👁 0 📖 2 min read

How we track verification of a Lean proof claim: the Riemann Hypothesis Claude log

On November 20, 2026, Levent Alpöge posted a four-page note saying that he, working with Claude, believes he has found a proof of the Riemann Hypothesis (and the Generalized Riemann Hypothesis for Dirichlet L-functions).

On November 20, 2026, Levent Alpöge posted a four-page note saying that he, working with Claude, believes he has found a proof of the Riemann Hypothesis (and the Generalized Riemann Hypothesis for Dirichlet L-functions). Unusually, the evidence is not a paper but a Lean formalization: 14 archive files, no human-readable write-up yet.

That raises a practical problem for anyone who wants to cite the claim responsibly: what exactly has been checked, by whom, and when? Within hours there were build reports, axiom audits, statement comparisons and expert comments scattered across Zulip, social media and news articles. I started keeping a single dated log so there is one place to point to.

What "verified" means for a Lean proof claim

A formal proof is only as good as three things, and the log tracks each of them separately:

  1. Does it build? Someone independent has to compile the archive with a stated Lean and Mathlib version, and ideally replay it through the kernel and through independent checkers.
  2. Does the main theorem say what it claims? A proof of the wrong statement is still a valid proof — of the wrong statement. The theorem should use standard definitions (here, Mathlib's own RiemannHypothesis) and be compared against a trusted challenge file.
  3. Which axioms does it rely on? #print axioms should show only the standard ones (propext, Classical.choice, Quot.sound), with no extra axioms declared in the project.

After that come the slower, human steps: a readable write-up, expert review, broad acceptance and — eventually — the Clay Millennium Prize rules (journal publication plus two years).

How each log entry is structured

Every entry records four things:

  • Event time in UTC (date only when the source gives no exact time).
  • Original source — preferring primary sources: the author's note, the Lean archive, build logs, posts by the experts themselves.
  • Evidence — what the source actually shows, not what it implies. Claims, readings and verifications are labelled as such.
  • Status change — which verification step moved forward or back.

Each entry has a stable anchor, so articles and forum posts can link to a specific event instead of a moving page. Corrections are logged, never silently overwritten.

Where things stand (at the time of writing)

  • Lean archive published — done.
  • Independent build — done: an independent build reported all modules compiling with zero errors, kernel replay, and acceptance by independent checkers.
  • Statement faithfulness — preliminary: automated comparison against a Mathlib-only challenge file matched; a published human audit of the statement is still pending.
  • Axiom audit — done for the headline theorems (standard axioms only).
  • Human-readable write-up, expert review, Clay prize — not yet.

Statuses change as evidence comes in, so check the live log rather than this snapshot.

The log

👉 Riemann Hypothesis × Claude — Verification Log

It is available in English, 简体中文, 繁體中文, 日本語, 한국어 and Français. If you have a source that adds to or contradicts an entry, the site has a corrections page.

Disclaimer: the proof is a claim under verification. The site is independent and not affiliated with Anthropic or the Clay Mathematics Institute.

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.