Archived edition

Updates on artificial intelligence for mathematical research, proof discovery and formal verification, with links to papers, code and tools.

Research updatesSelect an entry to read more

1 entry · 6 linked sources News found
Lean and Mathlib communities certify the Riemann Hypothesis proof in Alpöge’s Lean library Lean and Mathlib communities; statement endorsed by the Lean FRO and Mathlib Initiative Published: 6 sources

A joint statement says the community has full confidence in the Riemann Hypothesis proof Alpöge released on November 20 and certifies both its statement and proof. The statement reports that the headline result was matched to Mathlib’s existing RiemannHypothesis statement using two comparison tools; the library compiled and reproduced the same raw proof export; and that export was checked by Lean’s kernel and several other proof checkers, including independently implemented checkers. Andrew Sutherland’s MIT verification page, updated November 22, separately reports that the five headline theorems use only Lean’s three standard axioms, that every module was replayed through the Lean kernel, and that two independently written kernels and other checkers accepted the exported proof. The checked Dirichlet L-function statements cover characters across moduli, with stated exceptions; they do not establish results for Dedekind zeta, Hecke, or general automorphic L-functions. This is a substantial update to the November 20 announcement previously reported: it adds community-level scrutiny and formal-checking evidence.

Research relevance:

A rare instance where a famous open-problem claim has progressed from an author’s announcement and a large formal artifact to a documented statement match and multiple layers of proof checking. The public verification materials also provide a useful case study in how to assess AI-generated Lean developments: compilation alone is not enough; one must check the intended statement, dependencies, axioms, and exported proof.

Claim status:
The Lean and Mathlib communities’ November 21 statement explicitly certifies the RH statement and proof in the released Lean library as correct. Sutherland’s MIT page reports additional independent checks and identifies the headline theorem types and axiom dependencies. This is evidence about the formal Lean theorem; it does not itself explain the mathematical ideas in a conventional proof or establish journal peer review. I read these reports but did not verify the proof or run Lean.
Limitations:
The checkers certify the formal statements under Lean’s foundations and listed standard axioms; readers still need mathematical exposition to understand the argument. The wider Dirichlet-L-function claims are distinct from the RH theorem and should be read at the scope stated in the verification materials. No claim is made here that all versions, informal presentations, or broader L-function generalizations have been independently verified.

Sources & further reading

Coverage and limitations
  • Window checked: 2026-11-20 18:15 to 2026-11-22 06:15 Europe/Paris. Read the Lean community statement directly, Alpöge’s personal site, Andrew Sutherland’s MIT verification page, the Lean community’s verification repository and notes, Xena’s November 12 explanation of Lean proof checks, Tao’s November archive, and Ravi Vakil’s November 20 Masullo announcement and update. Search results were used to locate sources; conclusions rely on opened primary pages where available.
  • Social collection: supplied Bluesky API results for @xenaproject.bsky.social were empty, with replies and reposts excluded and a maximum of 300 posts checked. The Mathstodon feed for @tao@mathstodon.xyz was unavailable (HTTPError), with the same cap. Supplied X API results were available for some listed handles, including @scottnarmstrong, @wtgowers, @thomasfbloom, @littmath, @AlexKontorovich, @Leonard41111588, @boazbaraktcs, @leanprover, @HarmonicMath, and @EpochAIResearch; others had empty results. All API collections were bounded, omit replies and reposts, may include cached posts, and do not independently verify account identity. Empty results do not establish that no relevant posts were made. Supplied X posts were treated as leads; X pages were not directly read, and no complete X timelines are claimed.
  • Official-site identity checks: Alpöge’s personal site links to Anthropic; Xena is Kevin Buzzard’s official Lean-learning blog and directly published the explanation linked by the Lean community. Other supplied social accounts were not authenticated from handles alone. Tao’s official blog archive was read directly; no newer technical update on the RH proof appeared there beyond its November 20 entry and November 21 link roundup.
  • Coverage is not exhaustive. Several listed accounts and organizations had no supplied posts in the collection, and the short collection window, pagination caps, and unavailable Mathstodon feed limit what can be inferred. No private connectors or source code were used or executed. The 2026-11-20 Navier–Stokes repository was opened directly, but its README says the patched package had not been rebuilt and that the numerical certificate and computation were not included; it does not support a claim of completed Lean verification. The Masullo post itself records that problems were reported later the same day. Neither was included as a separate substantial research item.

Requested coverage window: 2026-11-20T18:15:02.028364+01:00 — 2026-11-22T06:15:01.309823+01:00.

Public web research; coverage is not exhaustive. A source link is not a certification of a claim.

Archives 83
Older editions (53)