Stafford’s conjecture (AI agents, Lean)
Awaiting review
Status
- Claim
Awaiting review
- Our reporting
- Medium confidence
- Verification
- Lean-verified
- Importance
- 3 of 5
- Last verified
- 22 November 2026
Your AI and this story
- GPT-6.2 Astra171 days after its cutoff
- Claude Opus 5.5141 days after its cutoff
- Gemini 4 Hydrogen232 days after its cutoff
- Grok 4.7171 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 141 days before it.
Key facts
- Claim (arXiv 2611.24577, 18 Nov 2026): for every nonzero d ∈ A_n(k), char k = 0, there are F, R, S ∈ A_n(k) with 1 = dR + FdS (Stafford 1978); consequence: every finitely generated torsion right A_n(k)-module is cyclic
- Proof route: a symplectic change of variables to a normal form, then showing the cyclic module Q = A_n/(dA_n + x_1^N dA_n) vanishes, using lengths over the associated graded ring and Gabber’s involutivity theorem
- AI appendix: ‘AI agents assisted with proof discovery, counterexample searches, review, formalization and writing … The first complete argument was written by AI agents in the style of a paper proof. Agents then formalized it in Lean, and the formalization changed parts of the argument where the original steps could not be checked as written’
- Setup: one controlling agent kept ‘the authoritative state of the proof’, parallel agents tried separate approaches, independent reviewer agents checked each result; models used: OpenAI GPT-5.5 to GPT-6.1, Anthropic Claude Haiku 4.5 to Opus 5, local Qwen and DeepSeek; Claude Opus 5.5 helped draft the manuscript
- Lean: every statement of the paper corresponds to a Lean declaration; side-by-side supplement at itpplasma.github.io/stafford38-supplementary
- The working rules became ‘a reusable agent skill for open mathematical problems’
What happened
The paper appeared in the 19 November arXiv listing (math.RA). Its Appendix A describes the agent workflow in detail, including how the Lean formalization forced changes to the first argument and how the printed proof was then rebuilt from the Lean proof. The authors come from a plasma-physics group (the supplement is hosted under itpplasma). The sweep first listed it on 22 November; earlier runs had treated its line as an already-read re-listing.
Why it matters
Stafford’s question about Weyl algebras had stood since 1978. The result arrives with a full Lean development, which makes it checkable without trusting the agents, though Lean statements still have to be compared with the intended theorem. Not refereed; no public reaction found by 22 November.
What is disputed or not yet verified
| Verification | Formal proof in Lean (per the authors); preprint not refereed |
|---|
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 2 primary
Primary
- arXiv 2611.24577: Stafford’s conjecture on cyclicity of torsion modules (Albert, Moser, Philipp)arxiv.org, paper
- Lean supplement: statements next to their Lean versionsitpplasma.github.io, code
Changes
- Filed (science agent; arXiv PDF and Appendix A read)