Kim’s dimension inequality claimed (Zehavi, GPT-6 Astra)
Awaiting review
Importance: major (4 of 5)The takeaway
Minhyong Kim’s non-abelian Chabauty method proves finiteness of rational points once a dimension inequality holds, which was known only in special cases such as CM Jacobians. A single-author preprint claims it for every curve of genus ≥ 2 over ℚ with a rational point, and says the main bound came from a ‘lengthy back-and-forth interaction with GPT 6 Astra’. Unrefereed.
Status
- Claim
Awaiting review
- Our reporting
- Medium confidence
- Verification
- Preprint only
- Importance
- Major (4 of 5)
- Last verified
- 18 November 2026
Your AI and this story
- GPT-6.2 Astra166 days after its cutoff
- Claude Opus 5.5136 days after its cutoff
- Gemini 4 Hydrogen227 days after its cutoff
- Grok 4.7166 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 136 days before it.
Key facts
- Claim (abstract): ‘We establish Kim’s conjectural dimension inequality for smooth projective geometrically connected curves C/Q of genus at least two with a rational basepoint… Consequently, the Chabauty–Kim loci are finite at all sufficiently large levels, recovering the Mordell theorem over Q through the Chabauty–Kim method’
- Method: noncommutative Iwasawa theory, bounding second Galois cohomology through kernels of Iwasawa operators on representations of the generalized symplectic group, ‘inspired by Coates and Kim’s work on curves with CM Jacobians’
- AI declaration (verbatim): ‘The development of the nontrivial bound of Theorem 3.1, as well as its application to the power-saving bound of Theorem 1.4 was carried out through a lengthy back-and-forth interaction with GPT 6 Astra (Open AI) on Ultra mode.’ The general strategy and the reduction to Iwasawa operators are the author’s, done about a year earlier; some ideas that came up with Astra ‘already appear in a paper of Weibo Fu’
- Status: unrefereed single-author preprint; no Lean; no expert reaction found by 16 Nov
- A second AI route to Mordell over ℚ via Kim’s method (Hexagon 2611.00576, published 18 Nov 2026, 26 pages, dated 20 Sept): Jit Wu Yap, ‘An averaged Iwasawa approach to Mordell over Q’, subtitled ‘Proof candidate for independent review’. It proposes an aggregate bound for multiplicities of a torsion Iwasawa module, averaged over symplectic highest weights, giving a cumulative Selmer dimension inequality along a subsequence of depths; it uses neither Parshin’s construction nor Faltings–Vojta. Provenance (verbatim): ‘The paper here was obtained by prompting GPT Astra 6 back in September. It was found as part of a directed search for new proofs of the Mordell conjecture. This particular proof took GPT 6 Astra only 30 minutes to produce, which was very surprising to the prompter.’ The manuscript says it ‘is not an announcement of an established new proof, and makes no priority claim’. Labels: primarily AI-generated; some parts checked by a human
What happened
Sa’ar Zehavi posted a proof of the dimension inequality that Kim’s method needs to show finiteness of rational points on curves. The inequality was previously known in special settings (for instance curves with CM Jacobians, after Coates and Kim). The author credits GPT-6 Astra with developing the central bound in an extended exchange, after his own reduction of the problem.
Why it matters
A general proof would make the Chabauty–Kim method an independent route to Faltings’s theorem over ℚ, a long-stated goal of the programme. It is a single-author, unrefereed claim; expert checking is needed.
On 18 Nov Jit Wu Yap deposited on Hexagon a second, independent proof candidate along the same Chabauty–Kim line, produced by GPT-6 Astra in about 30 minutes in September and presented explicitly as a candidate for review, not an established proof.
What is disputed or not yet verified
| Verification | Unrefereed preprint |
|---|
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 2 primary
Primary
Changes
- Added Jit Wu Yap’s Hexagon proof candidate (2611.00576, GPT-6 Astra, Mordell over ℚ via an averaged Iwasawa bound)
- Filed (science agent; abstract and AI declaration read in the PDF)