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

4 entries · 13 linked sources Partial coverage
Axiom publishes thirty Lean applications of its CFSG corpus, including a formalization extending Galois’s claim Axiom Published: 3 sources

Axiom reports formalizing thirty published results outside its Classification of Finite Simple Groups development, using its large finite-group corpus. The new repository separates results proved conditionally from unconditional versions that supply the required corpus theorems. The applications include a version of a claim in Galois’s 1832 letter, extended from prime fields to every finite field, and results in coding theory, graph theory, and group theory. The repository gives theorem statements, source references, and counts of new and reused Lean lines; for example, it lists 1,287 new lines and 14,196 reused lines for the Galois result.

Research relevance:

A concrete demonstration that a large formal library can support downstream results across several areas, with explicit statements and code available for inspection. It may be useful to researchers considering how reusable formalizations can lower the cost of verifying adjacent results.

Claim status:
Axiom’s article announces the results, and its public repository contains Lean formalizations, including conditional and unconditional forms. The repository describes the unconditional forms as relying only on Lean’s standard axioms. This is evidence of published artifacts, not independent confirmation that the code was built or that the underlying mathematics has received external review.
Limitations:
The release is by the organization that produced the work. We read the repository page and article but did not execute Lean, build the repository, or verify any proof. The stated historical claim is extended to finite fields of prime-power order, beyond Galois’s prime-field wording. The feed post appeared on November 18, while Axiom’s article is dated November 16; the repository does not establish a separate date for each formalization.

Sources & further reading

AMS awards inaugural Milestone Prize to Johan Commelin and Adam Topaz for formal verification work American Mathematical Society Published: 4 sources

The AMS announced Johan Commelin and Adam Topaz as recipients of its 2027 Milestone Prize for Foundational Work in Formal Verification. The cited work centers on the Liquid Tensor Experiment and the paper “Abstraction Boundaries and Spec Driven Development in Pure Mathematics,” describing a method for formalizing advanced mathematics in Lean and using formalization as research develops. This is recognition of a research workflow and methodological contribution, rather than a new mathematical theorem in this reporting window.

Research relevance:

The prize highlights a practical model for collaborative formalization of frontier mathematics and for using precise specifications and abstraction boundaries to make formal developments tractable.

Claim status:
The award and its stated rationale are attributed to the AMS. The announcement is corroborated by an AMS social announcement and a summary of the AMS citation. This is an award announcement, not evidence of a new proof or an independent assessment of the formalization.
Limitations:
The AMS news page returned 403 when opened, so its content was not directly read. The date comes from the AMS announcement mirrored in search results and on the AMS account snapshot. The research and paper predate the award and are not presented as newly published work.

Sources & further reading

Lean 4.37.0-rc1 reports lower Mathlib build instruction count Lean FRO / Lean contributors Published: 4 sources

The Lean 4.37.0 release candidate includes performance changes, including improved type-class result caching. The official release notes report that building Mathlib takes about 2.7% less wall-clock time and about 6.7% less time in type-class resolution from that change. The Lean account’s supplied X snapshot additionally claims the full Mathlib build used 23.6% fewer instructions than with Lean 4.36.0, based on benchmark data. The public release notes establish that this is a release candidate, subject to change.

Research relevance:

Build and elaboration costs matter directly to large Lean research projects and formal libraries. The release notes provide a concrete, bounded performance figure, while the account’s instruction-count metric points to a potentially important scalability measure for growing libraries.

Claim status:
The release candidate and its documented performance changes are confirmed by Lean’s official release notes and release page. The 23.6% instruction-count comparison is a claim in the supplied Lean X snapshot and was not independently verified from the linked benchmark in this research pass.
Limitations:
The figures describe particular benchmarks, not universal improvements for every project. The release is a candidate, not a final release. We did not run any build or benchmark. The 23.6% figure comes from a bounded X snapshot and linked measurement lead, not a directly inspected benchmark report.

Sources & further reading

Harmonic reports formalized implementation of a plasma-control algorithm from a research paper Harmonic, with DIII-D National Fusion Facility engineers and physicists Published: 3 sources

Harmonic’s DIII-D case study says its Aristotle system implemented a published detachment-control algorithm in C from the 2022 paper by D. Eldon and coauthors, proved the implementation free of undefined behavior under its stated model, and generated synthesizable fixed-point Verilog. Harmonic says the Verilog was proved equivalent to a fixed-point C model, with each control cycle taking a constant 435 clock cycles. The example links mathematical equations in a research paper to verified software and hardware for a plasma-control setting.

Research relevance:

This is a practical research workflow connecting a scientific paper, algorithm implementation, formal verification, and hardware translation. It may be useful to mathematicians and computational scientists interested in whether formal agents can make research algorithms easier to deploy and audit.

Claim status:
These are vendor-reported case-study claims on Harmonic’s official page, not an independent evaluation. The page describes proof outcomes and deployment context, but the case study is not itself a published mathematical proof.
Limitations:
The case study page was directly read, but its verification artifacts and code were not independently inspected or run. Results depend on the specified requirements, model, and assumptions; proving properties of the implementation does not independently establish the scientific validity of the source paper or the physical controller’s behavior.

Sources & further reading

Coverage and limitations
  • The supplied direct X API snapshots were reviewed as leads. Feed status was ok for all 14 supplied accounts, but snapshots were bounded, omitted replies and reposts, included cached posts, and do not authenticate account identity. Only Scott Armstrong, Thomas Bloom, Daniel Litt, Alex Kontorovich, Boaz Barak, Lean, Harmonic, Axiom, and Epoch AI had supplied posts in-window; the others had no supplied posts. Empty results do not show that no posts were made. X posts were checked against official pages where relevant; no complete X timeline was read.
  • The supplied Bluesky snapshot for xenaproject.bsky.social returned no posts and was bounded at 300 posts, excluding replies and reposts. The Mathstodon feed for tao@mathstodon.xyz was unavailable due to HTTPError, with a stated 300-post cap. Kevin Buzzard’s Imperial College page confirms his Xena project, but no in-window Xena blog item was found. Terence Tao’s official blog archive was consulted and showed only older Lean Grimoire material relevant to the supplied prior edition.
  • Official sources directly read included Axiom’s CFSG application repository and research article, Lean’s v4.37.0 release notes and GitHub release listing, and the official Harmonic DIII-D case study. The AMS news page at https://www.ams.org/news?news_id=7711 could not be opened because it returned 403; its prize announcement was found through an AMS social post mirror and summarized by a search result. That award announcement is included with this limitation stated.
  • The supplied prior-items list was used to avoid repeating earlier releases. The Axiom result is a new downstream application of the already-covered CFSG corpus; Lean 4.37.0-rc1 is a new release candidate after the previously covered 4.36.0 release. The Harmonic DIII-D case study is an in-window research-adjacent workflow report; its broader C/Verilog product launch dates to October and is not treated as new.

Requested coverage window: 2026-11-18T18:15:01.230672+01:00 — 2026-11-20T06:15:01.215544+01:00.

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

Archives 83
Older editions (53)