--- id: "2026-11-16-reinhardt-conjecture-proof-claim-fei-xue-chatgpt" url: "https://postcutoff.com/e/2026-11-16-reinhardt-conjecture-proof-claim-fei-xue-chatgpt/" as_of: "2026-11-19T22:11:00+01:00" date: "2026-11-16" date_precision: day category: science importance: 4 confidence: medium status: [Awaiting review] verification: Preprint only sources: 3 editor: human editor human_review: null version: "2026-11-17" --- As of: 2026-11-19 22:11 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/e/2026-11-16-reinhardt-conjecture-proof-claim-fei-xue-chatgpt/ # Reinhardt conjecture: claimed proof (ChatGPT-assisted) Full title: Fei Xue claims a proof of Reinhardt's 1934 conjecture that the smoothed octagon packs worst, with ChatGPT 6.1 Sol credited for the proofs of the remaining cases arXiv 2611.21470 (math.MG, 16 Nov 2026), Fei Xue: "We prove the Reinhardt conjecture: among centrally symmetric convex disks in the plane, the smoothed regular octagon has the least optimal lattice packing density", with equality only for its affine images. The proof bounds the number of control arcs in Hales's optimal-control formulation and verifies the remaining switching types by rigorous finite computation. The acknowledgments credit ChatGPT with the exploration of local cost bounds and the proofs of the six remaining cases. Unrefereed; no expert reaction found by 17 Nov. ## Key facts - Claim (abstract): the smoothed regular octagon has the least optimal lattice packing density among centrally symmetric convex disks; equality holds precisely for its nonsingular affine images. Reinhardt proposed it in 1934 - Method (abstract): 'explicit bounds on the number of control arcs, local cost inequalities and rigorous finite verification of the remaining switching types', building on Hales's optimal-control formulation and the Hales–Vajjha finiteness theorem (arXiv 2405.04331) - AI statement (acknowledgments, verbatim): 'The author proposed the argument giving an explicit finite upper bound for the number of control arcs. ChatGPT dot carried out the sustained exploration of local cost lower bounds. ChatGPT 6.1 Sol carried out the proofs of the six remaining types R01–R06 and the organization of the manuscript.' - Length: running page numbers reach at least 472; a reader archive with algorithms and fixed data accompanies the finite checks - The author thanks Chuanming Zong and Thomas Hales for seminar discussions around 2015 that introduced him to the problem - Status: unrefereed preprint, no Lean formalization; the sweep's PDF scan missed the AI statement (it sits in the acknowledgments without the usual wording), found on manual re-scan 17 Nov ## What happened On 16 Nov 2026 Fei Xue posted a preprint claiming a proof of Reinhardt's conjecture, one of the classical open problems on packing convex shapes. Thomas Hales had reformulated it as an optimal-control problem, and with Koushik Vajjha proved that a minimizer uses only finitely many switches. The new paper bounds those switches explicitly and checks the remaining cases with certified computations. The author says the model did the sustained work on the local bounds and the proofs of the last six cases; he writes that he hopes later work "will lead to a shorter and more beautiful proof". ## Why it matters Reinhardt's problem is the planar analogue of the question of which convex body packs worst. It has been open since 1934 and Hales has worked on it for 15 years. A proof of this length that rests partly on model-written case analysis will need expert checking before it is accepted. ## What is disputed or not yet verified - Verification: Unrefereed preprint with computer-assisted finite verification; not checked independently ## Your AI and this story - GPT-6.2 Astra (training cutoff May 2026): 169 days after its cutoff - Claude Opus 5.5 (training cutoff June 2026): 139 days after its cutoff - Gemini 4 Hydrogen (training cutoff March 2026): 230 days after its cutoff - Grok 4.7 (training cutoff May 2026): 169 days after its cutoff ## Sources 1. [arXiv 2611.21470: A proof of the Reinhardt conjecture (Fei Xue)](https://arxiv.org/abs/2611.21470) (arxiv.org, paper) 2. [Hales and Vajjha, Packings of smoothed polygons (arXiv 2405.04331)](https://arxiv.org/abs/2405.04331) (arxiv.org, paper) 3. [Hales, The Reinhardt conjecture as an optimal control problem (arXiv 1703.01352)](https://arxiv.org/abs/1703.01352) (arxiv.org, paper) ## Changes - 2026-11-17 (filed): Created (science agent; the acknowledgments name ChatGPT) ## Related - 2026-11-17: [AI-assisted results, 17 Nov](https://postcutoff.com/e/2026-11-17-ai-assisted-results-november-17-listing/index.md) - People: [Thomas Hales](https://postcutoff.com/person/thomas-hales/)