Green’s conjecture counterexamples (Astra in the search)
Awaiting review
Importance: major (4 of 5)The takeaway
Genus-10 curves that are W(E6)-covers of the projective line fail Green’s conjecture, a 9-dimensional family of counterexamples that also breaks the Green–Lazarsfeld secant conjecture. The authors used ChatGPT-6 Astra ‘for exploratory computations in the search for counterexamples’ and wrote the paper themselves. Unrefereed.
Status
- Claim
Awaiting review
- Our reporting
- Medium confidence
- Verification
- Preprint only
- Importance
- Major (4 of 5)
- Last verified
- 17 November 2026
Your AI and this story
- GPT-6.2 Astra169 days after its cutoff
- Claude Opus 5.5139 days after its cutoff
- Gemini 4 Hydrogen230 days after its cutoff
- Grok 4.7169 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 139 days before it.
Key facts
- Claim (abstract, verbatim): ‘We show that curves of genus 10 which are W(E6)-covers of P¹ fail Green’s Conjecture on syzygies of canonical curves. This provides a 9-dimensional family of counterexamples to Green’s Conjecture. This class of curves also provides counterexamples to the Green–Lazarsfeld Secant Conjecture.’
- AI statement (verbatim): ‘ChatGPT-6 Astra was used for exploratory computations in the search for counterexamples, reference search, and interactive feedback on the manuscript. The paper has been completely written by the authors who take full responsibility for all content.’
- Context (paper): Green’s conjecture (1984) states K_{p,2}(C, ω_C) = 0 if and only if p < Cliff(C); the direction ‘non-vanishing for p ≥ Cliff(C)’ is the Green–Lazarsfeld theorem. Aprodu and Farkas had earlier proved it for curves on arbitrary K3 surfaces; Farkas’s 2026 survey still listed it as open for arbitrary curves
- The E6-curves are not Petri general (dim W¹₆(C) = 1), which the paper notes places them outside earlier positive results
- Status: unrefereed preprint; AI role limited to exploration and feedback by the authors’ account; no reactions found by 17 Nov
What happened
Three algebraic geometers, among them Gavril Farkas, one of the main contributors to work on Green’s conjecture, posted a short preprint saying that a whole family of genus-10 curves with E6 symmetry violates it. They credit ChatGPT-6 Astra only with exploratory computations during the search and with feedback; they say they wrote the paper themselves.
Why it matters
Green’s conjecture has guided research on syzygies of curves for four decades and is proved for general curves. If the counterexamples hold up, the conjecture is false as stated for arbitrary curves. The AI role is a supporting one, which makes this a different kind of case from the many model-written proofs in recent weeks.
What is disputed or not yet verified
| Verification | Unrefereed preprint |
|---|
Sources
1 source from 1 site. Numbers match the chips in the text.
1 source: 1 primary
Primary
Changes
- Filed (science agent; abstract and AI statement read in the PDF)