VibeMathedMath problems solved with AI

Equality of the Mirković-Vilonen and theta bases (Kamnitzer's conjecture)

Let GG be a connected complex semisimple group and NN a maximal unipotent subgroup. C[N]\mathbb{C}[N] carries the Mirković-Vilonen (MV) basis, from MV cycles in the affine Grassmannian via geometric Satake (Baumann-Kamnitzer-Knutson), and the theta basis of Gross-Hacking-Keel-Kontsevich for its cluster structure. Kamnitzer's ICM 2022 report closes its cluster-algebra section with the conjecture: the MV and theta bases for C[N]\mathbb{C}[N] coincide. Do they?

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI-discovered
Method
Argument
Field
Geometric representation theory; cluster algebras
Posed by
Joel Kamnitzer, Perfect bases in representation theory: three mountains and their springs, Proc. ICM 2022 (arXiv 2201.02289)
Year posed
2022
Years open
4y
Solved
2026-11-02
Model
ChatGPT-6-Pro (with Codex; reviewed by Rethlas-OSV)
Vendor
OpenAI
Collaborators
Qiyue Tang
Verification
Unreviewed
Publication
Preprint
Significance
25 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem 17.1: for every connected complex semisimple group GG, the normalized MV basis of C[N]\mathbb{C}[N] equals the regular theta basis. The bases are paired by weight and leading data; MV structure constants come from spatial fusion in the Beilinson-Drinfeld Grassmannian (Baumann-Kamnitzer-Knutson) and theta structure constants from punctured logarithmic curve counts on an affine Richardson variety, and the two are shown to agree, so the multiplicative pairing is the identity. Caveats: AI-drafted and AI-repaired proof, no independent expert review, author disclaims priority.

What the AI did

From the paper's 'AI use and scholarly credit' section: the initial draft was developed through seven ChatGPT-6-Pro conversations (about 23 hours), with four hours of research using Pharos, then polished with Codex and reviewed by Rethlas-OSV. That review found an error invalidating the main proof; ChatGPT-6-Pro was asked to spend six hours repairing it, a separate ChatGPT conversation reviewed the repair, and Codex expanded the manuscript from 21 to 52 pages. A second Rethlas-OSV review found further errors, one again blocking the proof as written, with corrections supplied; after these the manuscript passed. Hexagon labels the work 'a mix of human-written and AI-generated text', human understanding 'some parts'. The author does not claim priority or ownership of the results.

Verification

No independent mathematician has checked this yet. Checked here on 15 November 2026: the Hexagon record and PDF (hexagon 2611.00006v1, 52 pages), its AI-use section, the introduction and the statement of Theorem 17.1, read against the TeX source of Kamnitzer's ICM report (arXiv 2201.02289v2), whose conjecture reads 'The MV and theta bases for C[N]\mathbb{C}[N] coincide'. The statement matches; the paper cites it as Conjecture 6.9. The proof itself was not read or refereed. By the paper's own account the argument was drafted by ChatGPT-6-Pro and twice repaired after AI review found errors blocking the main theorem, and Hexagon records human understanding as 'some parts'. Hexagon is a moderated but not peer-reviewed repository. No earlier claim was found in this catalog or in an arXiv API search. Resolution is Candidate until someone independent checks the argument in public.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion