Answers are grounded in this paper and cite their sources.
The transfer-matrix gap of four-dimensional SU(2) lattice gauge theory tends to zero at weak coupling, with an explicit rate in infinite volume (note v1.3, 12 November 2026; first deposit). What the note is. A technical note with proofs written in full. For the SU(2) Wilson lattice gauge theory on Z4 it shows that, for every family μβ of translation-invariant Gibbs states that are reflection positive for the time reflection through a hyperplane of sites, the gap m(μβ) = −log‖T|Ω⊥‖ of the one-step transfer operator tends to zero as β → ∞, and gives an explicit power-law rate. Results. (1) For every fixed integer t ≥ 1, limsup m(μβ) ≤ B(t) = (2/t) log(c(0)/c(t)), B(t) = O((log t)/t), also in the gauge-invariant sector, on the positive part of the spectrum and after removing the eigenspaces at ±1. (2) New in v1.3: for β ≥ 10228 and every such state, m ≤ 24 β−1/120 log β. The proof quantifies the Gaussian (free Maxwell) limit by Stein's method for the conditional law in a box that grows with β. The bound is smaller than 1 only for β of order 10455 or more, and is far above the expected exponentially small gap. Status. Written by AI models and read by other AI models; no human has checked the proofs; not refereed. The non-abelian Stokes appendix and the constants of the rate were executed and recomputed by an agent other than their author (programs and outputs in the package); numerical values are used in no proof. The historical novelty is not established: the anteriority searches were bounded. Limitations. Not Problem 5.1 of Chatterjee (arXiv:1803.01950). The theorem at a fixed lattice size with rate O((log β)/β) of Bhattacharjee et al. (Zenodo 10.5281/zenodo.24147967) is a different statement; neither implies the other. No exponential rate, nothing at moderate β, no confinement statement (the bound also holds for U(1) in its Coulomb phase), no continuum mass gap. Prepared with the assistance of Claude (Anthropic: Opus 5.5, Fable 5.2, Sonnet 5.5), GPT-6 Astra and GPT-6.1 Sol (OpenAI), Qwen3.8 (Qwen team) and DeepSeek V4 Pro; the roles are listed in the note.
No takes yet. Share an insight, caveat, or question.
Kevin Remondière (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: