{"schema":"postcutoff/event@1","as_of":"2026-10-21T14:07:00+02:00","url":"https://postcutoff.com/e/2026-10-19-neumann-zagier-conjecture-tree-lattices-ai-proof/","md":"https://postcutoff.com/e/2026-10-19-neumann-zagier-conjecture-tree-lattices-ai-proof/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"2026-10-19-neumann-zagier-conjecture-tree-lattices-ai-proof","date":"2026-10-19","date_precision":"day","short_title":"Neumann–Zagier 1985 conjecture on tree lattices settled after an AI proof with ChatGPT 5.5 Pro and Opus 4.8","deck":null,"takeaway":"Joshua Evan Greene, Ahmad Issa and Duncan McCoy prove that a positive definite tree lattice has an irreducible vertex element, which settles a 1985 conjecture of Neumann and Zagier.","category":"science","category_label":"Science & math","importance":3,"confidence":"high","status":{"key":"pending","labels":["Event confirmed","Awaiting review"]},"sources":[{"n":1,"title":"arXiv 2610.27095: Tree lattices and the Neumann–Zagier conjecture","url":"https://arxiv.org/abs/2610.27095","type":"paper","group":"primary","domain":"arxiv.org"}],"official":1,"filed":"2026-10-21","updated":"2026-10-21","orgs":["OpenAI","Anthropic"],"title":"Neumann–Zagier 1985 conjecture on tree lattices settled after an AI proof with ChatGPT 5.5 Pro and Opus 4.8","summary":"Joshua Evan Greene, Ahmad Issa and Duncan McCoy prove that a positive definite tree lattice has an irreducible vertex element, which settles a 1985 conjecture of Neumann and Zagier. Greene and McCoy had submitted their partial result to the First Proof AI benchmark. Issa then proved the theorem with ChatGPT 5.5 Pro and Opus 4.8, and the three authors simplified the AI argument.","key_facts":["Paper: arXiv 2610.27095 'Tree lattices and the Neumann–Zagier conjecture' (Greene, Issa, McCoy), 19 Oct 2026","Result: if a tree lattice L(T) is positive definite, some vertex element is irreducible; corollary: if L(T) ≅ Z^n then T has a vertex of weight one (Neumann–Zagier 1985)","AI methodology: 'In 2026, JEG and DM submitted this result to the First Proof AI benchmark test. Soon thereafter, JEG and DM were contacted by AI (the second author), who had used AI (artificial intelligence) to prove Theorem 1.1.'","'The AI proof contained two main innovations: the use of the Perron-Frobenius theorem to identify a vertex, and an inductive proof akin to Lemma 3.1 to establish its irreducibility. While the use of Perron-Frobenius appeared to be a stroke of genius, it turned out to be a red herring'; the inductive lemma 'persisted as an essential linchpin'","'The initial AI proof of Theorem 1.1 used ChatGPT 5.5 Pro and Opus 4.8 … We chucked AI efforts to write this paper in favor of our own, although we engaged it for proofreading.'","Prior human progress: Greene and McCoy proved the single-bad-vertex case in 2020 and strengthened it in 2023"],"key_numbers":[],"tags":["math","combinatorics","lattices","topology","first-proof","chatgpt","claude","claude-opus-4-8"],"science":{"field":"mathematics","subfield":"combinatorics / integral lattices (low-dimensional topology)","problem":"Neumann–Zagier conjecture (1985) on tree lattices isometric to Z^n","result":"Every positive definite tree lattice has an irreducible vertex; hence a tree presenting Z^n has a vertex of weight one","open_since":"1985","ai_system":["ChatGPT 5.5 Pro","Claude Opus 4.8"],"human_role":"AI-assisted: Issa obtained the first proof of Theorem 1.1 with the two models; the authors digested, simplified and strengthened it and wrote the paper themselves","verification":{"key":"preprint","label":"Preprint only","text":"Unrefereed preprint by named experts who checked and re-derived the AI argument"},"status":"pending","shock":""},"body_md":"## What happened\n\nA tree with integer weights on its vertices presents an integral \"tree lattice\". In 1985 Walter Neumann and Don Zagier conjectured that a tree lattice isometric to the standard lattice Z^n must\ncontain a vertex of weight one. Greene and McCoy proved special cases in 2020 and 2023. In 2026 they submitted the open case to\nthe First Proof benchmark of research problems for AI.\n\nAhmad Issa then contacted them with a proof he had obtained using ChatGPT 5.5 Pro and Claude Opus 4.8. The authors' AI\nmethodology section describes how they digested it. The model's Perron–Frobenius idea \"appeared to be a stroke of genius\" but\nwas a red herring. Its inductive irreducibility lemma stayed essential, and they strengthened the theorem. The paper itself was\nwritten by hand, with AI used only for proofreading.\n\n## Why it matters\n\nA problem posted as a benchmark task was solved with frontier models and then confirmed and improved by the experts who posed\nit. The paper also describes candidly what the AI contributed.","disputed":[{"label":"Verification","text":"Unrefereed preprint by named experts who checked and re-derived the AI argument","sources":[]}],"related":[{"id":"2026-10-21-ai-assisted-results-october-21-listing","url":"https://postcutoff.com/e/2026-10-21-ai-assisted-results-october-21-listing/","date":"2026-10-21","date_precision":"day","short_title":"AI-assisted results in the 21 Oct 2026 arXiv listing","deck":"A ChatGPT 6.1 Sol counterexample to Frankl–Ota–Tokushige, klt Fano proofs from ChatGPT, Vince's conjecture with GPT Astra, Pólya for spherical shells, a Claude-built 55-addition 3×3 scheme and a Claude Opus 5.5 lemma in reverse mathematics","takeaway":"Wednesday's arXiv announcement brought four results with their own entries: the MDS conjecture in even characteristic, the double-cap conjecture, the quasitrace/monotone completeness counterexamples and the Neumann–Zagier conjecture. An expert verification of Claude's Hadamard matrices was added to that entry. This page lists the next tier, about thirty papers.","category":"science","category_label":"Science & math","importance":3,"confidence":"medium","status":{"key":"pending","labels":["Awaiting review"]},"sources":18,"official":18,"filed":"2026-10-21","updated":"2026-10-21","orgs":["OpenAI","Anthropic","Google DeepMind","various mathematicians"]},{"id":"2026-10-14-first-proof-third-batch","url":"https://postcutoff.com/e/2026-10-14-first-proof-third-batch/","date":"2026-10-14","date_precision":"day","short_title":"First Proof batch 3: frontier systems solve 9 of 10 research problems, open-weight Kimi K3 harness none","deck":null,"takeaway":"Within a year the best result went from \"a few\" (February) to 6–7 (June) to 9 of 10 (October) on fresh research problems with strict peer-review grading.","category":"benchmark","category_label":"Benchmarks","importance":4,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":5,"official":3,"filed":"2026-10-14","updated":"2026-10-15","orgs":["First Proof Foundation","OpenAI","Anthropic","ETH Zurich","Moonshot AI"]},{"id":"2026-06-10-first-proof-second-batch","url":"https://postcutoff.com/e/2026-06-10-first-proof-second-batch/","date":"2026-06-10","date_precision":"day","short_title":"AI systems pass 7 of 10 unpublished research problems in First Proof's second batch","deck":null,"takeaway":"It is the most carefully refereed measurement so far of AI on genuine research problems.","category":"benchmark","category_label":"Benchmarks","importance":3,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":6,"official":4,"filed":"2026-10-05","updated":"2026-10-05","orgs":["First Proof Foundation","ETH Zurich","OpenAI","UCLA","Princeton University"]},{"id":"2026-02-14-first-proof-challenge","url":"https://postcutoff.com/e/2026-02-14-first-proof-challenge/","date":"2026-02-14","date_precision":"day","short_title":"AI solves about half of 10 unpublished research problems in the 'First Proof' challenge","deck":null,"takeaway":"It gave a cleaner measure than olympiads of whether AI can do research maths: at the time, about half the time.","category":"science","category_label":"Science & math","importance":3,"confidence":"high","status":{"key":"confirmed","labels":["Result confirmed"]},"sources":3,"official":2,"filed":"2026-09-29","updated":"2026-10-05","orgs":["Google DeepMind","OpenAI"]}],"people":[],"posts":[],"videos":[],"models":[],"changes":[{"date":"2026-10-21","type":"filed","text":"Created (PDF read)"}],"provenance":{"agents":[{"model":"Claude Opus 5.5","maker":"Anthropic","tool":"Claude Code"}],"filed":"2026-10-21","run":null,"sources_read":"PDF read","updated":"2026-10-21","human_review":null,"version":null},"gaps":[{"model_id":"gpt-6-astra","name":"GPT-6 Astra","cutoff":"2026-04","days_after":172,"in_training_data":false},{"model_id":"claude-opus-5-5","name":"Claude Opus 5.5","cutoff":"2026-06","days_after":111,"in_training_data":false},{"model_id":"gemini-4-argon","name":"Gemini 4 Argon","cutoff":"2026-03","days_after":202,"in_training_data":false},{"model_id":"grok-4-7","name":"Grok 4.7","cutoff":"2026-05","days_after":141,"in_training_data":false}],"short_url":null}