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

2 entries · 6 linked sources News found
A public Lean formalization of Helfgott’s ternary Goldbach proof Charlie Reams; Harald Andrés Helfgott is listed as having participated Published: 3 sources

A public repository describes a Lean 4 formalization of the weak (ternary) Goldbach theorem: every odd integer greater than 5 is a sum of three primes. The repository attributes the underlying analytic proof to Helfgott and its finite verification to Helfgott and David J. Platt. Its account says AI wrote most of the Lean code under human direction, and reports corrections or workarounds where some steps in source material did not match the proof. This formalizes an existing result; it is not a new proof of binary Goldbach.

Research relevance:

A large formalization of a major analytic number theory proof could make its dependencies, computational inputs, and source discrepancies easier to inspect. The project’s Lean infrastructure and explicit accounting of external inputs may also interest researchers building formalizations of analytic arguments.

Claim status:
The repository claims the Lean theorem is complete and machine-checkable. This digest has not built or independently checked the Lean project, audited the proof, or verified the correspondence between every formal statement and the informal theorem. The reported formalization is not independent confirmation of Helfgott’s original proof.
Limitations:
The result is the weak, ternary theorem, not the still-open binary Goldbach conjecture. The repository documents external-source and computational trust boundaries; its claims about completeness, AI authorship, and source corrections are project-reported. Treat the public Lean artifact as available for scrutiny, not as independently scrutinized here. The original theorem predates this release by years; the new material is the formalization and associated artifacts.

Sources & further reading

Correction clarifies what the Yang–Mills Lean construction establishes Scott Armstrong Published: 3 sources

In a supplied X post, Armstrong corrected his earlier description of the Yang–Mills construction: he says the Lean code proves the constructed theory is not free Yang–Mills and is interacting, including that its stress-tensor trace is nonzero. He distinguishes this from the stronger claim that the theory cannot be built from generalized free fields with continuous mass spectra, which he says is not proved. The public project repository provides a detailed map of the construction and manuscript, making this a substantive clarification of the earlier coverage.

Research relevance:

This narrows an important claim about a high-profile formal construction and illustrates why the precise theorem statement and its interpretation matter alongside a Lean artifact.

Claim status:
This is an author’s correction and interpretation, not an independent mathematical review. The repository makes the formal construction and explanatory materials available, but this digest has not checked the Lean code or verified the interpretation of its theorem.
Limitations:
The correction is based on Armstrong’s supplied post, whose author identity was not independently authenticated through X. The repository’s scope notes distinguish its reviewed statement and explanatory material from remaining summaries and drafts. The stronger generalized-free-field claim remains explicitly unresolved in Armstrong’s account.

Sources & further reading

Coverage and limitations
  • Window checked: 2026-11-13 10:41 to 2026-11-14 06:15 Europe/Paris. Dates below refer to the source’s stated publication or post date; search indexing and crawl dates were not treated as publication dates.
  • Direct X API results were supplied for __alpoge__, scottnarmstrong, wtgowers, sebastienbubeck, thomasfbloom, littmath, alexkontorovich, f_charton, leonard41111588, boazbaraktcs, leanprover, projectnumina, harmonicmath, axiommathai, and epochairesearch. Collection was bounded; replies and reposts were excluded, cached posts may be included, and account resolution does not independently verify identity. Empty results do not establish inactivity. The supplied feed contained relevant posts from Armstrong and Kontorovich, plus unrelated or promotional posts from other accounts.
  • Direct Bluesky collection returned no posts for xenaproject.bsky.social, with a 300-post cap and replies and reposts excluded. Collection for mathstodon.xyz failed with HTTPError, also with replies and reposts excluded and a 300-post cap. These are access gaps, not evidence of no posts.
  • I checked supplied leads against institutional pages and primary project sources, including Rutgers and NYU pages, GitHub project material, and a web search for recent Lean and AI mathematics research. I did not consult X timelines directly. The supplied sources contained no direct social feed for Tao’s blog or Mathstodon, the Xena blog, or Simon Frieder’s blog; coverage of these and other sources is therefore incomplete. The NYU page confirms Leonardo De Moura’s November 14 seminar, but an upcoming talk is not a research result.

Requested coverage window: 2026-11-13T10:41:06.584577+01:00 — 2026-11-14T06:15:01.742023+01:00.

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

Archives 79
Older editions (49)