Claimed AI-led

Computer-assisted proof of a uniform spectral gap for the spin-1 Heisenberg chain on even periodic rings

A manuscript in OpenAI's October 2026 mathematics release proves that the spin-1 antiferromagnetic Heisenberg chain on even periodic rings has a unique ground state and a spectral gap above (log⁡20)/784≈0.0038(\log 20)/784 \approx 0.0038 for every even length L≥2304L \ge 2304, the periodic formulation of Haldane's 1983 conjecture, by combining an analytic thermal-purity argument with certified finite computations.

Model
Unreleased internal OpenAI model, not named in the release
Field
Physics
Date
2026-10-06
Problem posed
1983 · open 43 yrs

What was found

Haldane predicted in 1983 that integer-spin antiferromagnetic chains have a gap above the ground state while half-integer chains do not; numerics put the spin-1 gap near 0.41, but no proof existed for the Heisenberg chain itself, which is not frustration free, so the finite-size criteria used for models such as AKLT do not apply. The manuscript, dated 24 September 2026 and published on 6 October 2026 as result 268 of the openai/math release that this registry records at 2026-10-06-openai-math-collection, bounds the gap through the purity of the Gibbs state, propagating finite-ring partition-function bounds to all longer even rings; it also gives ΔL≳4.79×10−4\Delta_L \gtrsim 4.79 \times 10^{-4} for L≥60L \ge 60. A companion manuscript treats open chains with boundary fields. The release ships a Python finite verifier for the numerical inputs and no Lean formalization of this result. Three outside papers have since built on the proof: Sirui Lu's tensor-network reformulation sharpens the bound to ΔL>0.0729\Delta_L > 0.0729 for every even L≥36L \ge 36 from the same inputs and certifies gaps for seven further chains and ladders, Hal Tasaki uses its machinery to prove uniqueness of the translation-invariant ground state of the infinite chain, and Tarun Grover extends the purity bootstrap to higher dimensions and fermions.

Novelty check

arXiv API, 2026-11-12: abs:"Haldane gap" AND abs:proof returned three records, two unrelated papers from 2000 and 2003 and Tasaki's October 2026 extension of this proof; abs:"Haldane conjecture", the 30 most recent records, held no rigorous proof for the spin-1 Heisenberg chain before October 2026. The manuscript's own introduction cites only partial rigorous results (Lieb-Schultz-Mattis for half-integer chains, and a 2025 theorem of Tasaki that assumes a gap for odd open chains), and Tasaki, who has worked on the problem, describes this paper as proving the Haldane conjecture in the standard periodic-chain formulation. No earlier entry in this registry covers it; it appeared only inside the collection entry.

Caveats and known objections

Entered at claimed as a lab release, and kept there because the outside checks so far are partial. Tasaki studied the analytic argument and wrote expository notes on it, but states that he has not independently checked the computer-assisted estimates. Lu verified the finite inputs in exact arithmetic and formalized the general gap theorem in Lean 4, but for the spin-1 chain his Lean theorem still takes the short-ring integer trace totals for 5≤n≤125 \le n \le 12 as hypotheses, and his paper declares that GPT-6-Astra and OpenAI Codex were used for calculations, formalization and writing, with every argument verified by the author. So the computer-assisted half has one outside checker. A second independent check of the finite inputs, or a Lean discharge of the remaining hypotheses, would support independent or formal. The theorem covers even periodic rings only, and its constant is about a hundred times smaller than the numerical gap. Autonomy is ai-led as for the rest of the release: OpenAI says its results were produced by an internal model under one fixed procedure, the manuscript names no human author, and the release does not say how much human review preceded publication.

Independent checks

Hal Tasaki: Studied the analytic argument carefully and published expository notes in continuous-time and discrete-time versions; states that the complete proof also uses computer-assisted estimates, which he has not independently checked. · link ↗

Sirui Lu (Max Planck Institute of Quantum Optics): Verified the finite inputs in exact arithmetic, recast the proof in tensor-network diagrams, sharpened the bound to ΔL>0.0729\Delta_L > 0.0729 for every even L≥36L \ge 36, and formalized the general theorem in Lean 4; for the spin-1 chain the Lean statement assumes the short-ring trace totals for 5≤n≤125 \le n \le 12. Declares AI assistance (GPT-6-Astra, OpenAI Codex), with all arguments verified by the author. · link ↗

Disagree with these grades?

Bring a citation: a grade moves on evidence, not on argument.

Or on GitHub: submit a check challenge the grade send a correction or send a pull request

Entry history (1 event)
  1. AddedEntered the registry graded Claimed and AI-led.

Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.

Graded claimed for verification and ai-led for autonomy. What these mean.

Cite this entry

Plain text
whataifound.org. (2026). Computer-assisted proof of a uniform spectral gap for the spin-1 Heisenberg chain on even periodic rings. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-06-periodic-spin-one-haldane-gap
BibTeX
@misc{whataifound-openai-2026-gap,
  title        = {Computer-assisted proof of a uniform spectral gap for the spin-1 Heisenberg chain on even periodic rings},
  author       = {{whataifound.org}},
  year         = {2026},
  howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
  note         = {Result by OpenAI. Verification: Claimed. Autonomy: AI-led.},
  url          = {https://whataifound.org/finding/2026-10-06-periodic-spin-one-haldane-gap}
}

Related findings

← All physics findings in the registry