VibeMathedMath problems solved with AI

Castro-Silva's conjecture: large triangles in planar sets of positive upper density

Conjecture 1 of Castro-Silva: let A⊆R2A\subseteq\mathbb{R}^2 be a measurable set of positive upper density, lim sup⁡T→∞∣A∩[−T,T]2∣/(2T)2>0\limsup_{T\to\infty}|A\cap[-T,T]^2|/(2T)^2>0, and let u,v,w∈R2u,v,w\in\mathbb{R}^2 be non-collinear points. Is there t0>0t_0>0 such that for every t≥t0t\ge t_0 the set AA contains a configuration congruent to {tu,tv,tw}\{tu,tv,tw\}?

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI-discovered
Method
Argument
Field
Euclidean Ramsey theory, harmonic analysis
Posed by
Davi Castro-Silva
Year posed
2021
Years open
5y
Solved
2026-11-11
Model
GPT-6 Astra Pro, Claude Opus 5.5
Vendor
OpenAI; Anthropic
Collaborators
Polona Durcik
Verification
Lean-checked, statement unaudited
Publication
Preprint
Significance
35 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem 1.1: if Δ⊂R2\Delta\subset\mathbb{R}^2 is three noncollinear points and E⊂R2E\subset\mathbb{R}^2 is measurable with upper Banach density d∗(E)>0d^*(E)>0, there is λ0\lambda_0 such that for every λ≥λ0\lambda\ge\lambda_0 the set EE contains a translate of a rotation of λΔ\lambda\Delta. Since upper Banach density dominates the centred upper density and rotated translates are congruent copies, this gives Castro-Silva's Conjecture 1 in full, in a stronger orientation-preserving form. Castro-Silva's broader conjecture for d+1d+1 points in Rd\mathbb{R}^d in other dimensions is not addressed. Earlier: Falconer, Kovac and Yavicoli (2022) for d∗(E)>1/2d^*(E)>1/2; Furstenberg, Katznelson and Weiss for neighbourhoods of EE.

What the AI did

Per the Hexagon provenance statement and the paper's README: the manuscript was generated over three conversations with GPT-6 Astra Pro via Codex on 4-5 November 2026, directed by Polona Durcik toward trilinear oscillatory integral methods (as in papers of Christ and of Christ, Durcik and Roos), with otherwise generic prompts. The Lean formalization was generated by Claude Opus 5.5 (medium effort). Hexagon labels the text 'Primarily AI-generated' with 'Human understanding: some parts'.

Verification

No independent mathematician has checked this yet. Checked here on 15 November 2026: the Hexagon PDF (2611.00302v1, contributed by Polona Durcik) and its provenance page; Castro-Silva's Conjecture 1 read in arXiv 2102.10018 v1 and v3. Palomar entry PALOMAR-2026-11-10-000004 v1 read: Comparator configuration with theorem Auto.triangles_in_the_plane, permitted axioms propext, Classical.choice and Quot.sound only, checked by Palomar with nanoda and con-ron at commit 3fb29cd. The headline statement in Challenge.lean was read here and matches Theorem 1.1, including the upper Banach density definition. The repository at that commit (19 Lean files, about 29,000 lines) was grepped: no sorry outside the challenge stub, no axiom, native_decide, unsafe or implemented_by. Not rebuilt here. The challenge statement was written by the formalizing model and checked by the author, so it is not independently anchored.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion