There is a newer version of the record available.

Published November 1, 2026 | Version 1.5

Dobrushin uniqueness for the four-dimensional SU(2) Wilson lattice gauge measure for beta < 2/9, with interval/Lean certificates (note v1.5)

Authors/Creators

  • 1. Independent Researcher, Oloron-Sainte-Marie, France

Description

Dobrushin uniqueness for the four-dimensional SU(2) Wilson lattice gauge measure for β < 2/9, with interval/Lean certificates (note v1.5, 1 November 2026; the previous version is v1.4 of 28 October 2026, doi:10.5281/zenodo.23712954).

Result. For the SU(2) Wilson action on Z4 with weight exp(−β Σp(1 − ½ Re Tr Up)), the Gibbs (DLR) measure is unique for 0 ≤ β < 2/9 ≈ 0.222 (equivalently g2 > 18; βtH < 1/18 in the 't Hooft normalisation). Method: Dobrushin's criterion in its Wasserstein (W1) form with the Euclidean (chordal) metric of R4 on SU(2) ≅ S3, the metric used by Levin (1980). Secondary result: with the geodesic metric the exact coefficient is 8β/(9π), which gives the weaker range β < π/16. Each threshold is sharp for its criterion.

New in this version (v1.5). (1) The one-site coefficient in a field of strength κ equals Eκ[x⊥2] = I2(κ)/(κ I1(κ)) for every κ ≥ 0 (v1.4: κ ≤ 4/3); for κ > 0 the maximum over directions is attained exactly for the directions orthogonal to the field, oblique directions are strictly smaller with explicit two-sided bounds, and the coefficient is strictly decreasing in κ, hence maximal at zero field, where it equals 1/4. Each of the 18 neighbouring links has influence exactly β/4 and the Dobrushin sum is 9β/2. The proof is analytic (divergence-free flux field, slice estimate, exact sum-of-squares identity, Chebyshev association argument); it is not formalised in Lean. The interval-arithmetic proof of v1.4 on κ ≤ 4/3, with its Lean-checked final steps, is kept as a second proof (Appendix A). (2) A corollary records the Poincaré inequality for β < 2/9, uniform in the volume and the boundary condition, that Wu's theorem (2006) gives from these coefficients (a corollary, not a new method). (3) Application (Section 7 and Appendix B): the same chordal coefficient, applied to the slab σ-models of Durhuus and Fröhlich through Dobrushin's comparison, gives Wilson's area law for β < 2/3, uniformly on tori, with free boundary conditions and for every Gibbs measure, with string tension at least arcosh(2/β − 2). This range is not new: Borgs and Seiler (1983, Appendix B of the preprint version) proved confinement for β < 12/17 by a transfer-matrix argument, and no priority is claimed for any threshold; the contribution is an independent proof with explicit constants.

Prior public statement. The uniqueness range β < 2/9 for the four-dimensional SU(2) model (with exponential decay and a lattice mass gap), the all-field bound Dch(κ) ≤ 1/4, i.e. W1(νb, νb') ≤ |b − b'|/4 on S3 (Levin's Remark 3.7 for N = 4), and the exactness of the influence coefficient β/4 were stated publicly, with proofs, in the preprint of D. Bhattacharjee and U. Bhattacharya, “Mass Gap of Lattice Yang-Mills Theory up to the Dobrushin Ceiling”, hexagon:2610.00341v1, 18 October 2026 (Theorems E, F and 7 and Proposition 13; the same statement is Theorem 125 of v4, with J. M. Lockwood; Zenodo 10.5281/zenodo.23786523). This note was prepared independently and makes no priority claim on them. Its own contributions: the exact value I2/(κ I1) of the supremum over directions for every κ ≥ 0, its maximisers, the two-sided estimate for oblique directions, the strict monotonicity in κ and an analytic proof; the exact geodesic coefficient 8β/(9π); two interval-arithmetic certificates with independent checks; Lean-checked final steps; explicit transfers of the published constants of Levin, Royer and Shen-Zhu-Zhu to the Wilson link; and the area-law application (an independent proof; the range itself is not new).

Status. The chordal identity for all κ is proved on paper by an analytic argument, not formalised in Lean, read by one independent reviewer (a language model) and not refereed; the area-law application of Section 7 and Appendix B is also proved on paper only, not formalised, and was read by the same reviewer; numerical and interval cross-checks (exact power series to κ39, Arb at 90 digits, interval certificates on [4/3, 20], optimal transport) are not used in the proof. The v1.4 proof on κ ≤ 4/3 (Arb interval arithmetic, independent mpmath recertification, final rational steps checked in Lean 4, Mathlib v4.34.0, axioms propext, Classical.choice, Quot.sound only) is kept as a second proof. The archive contains the note (PDF and LaTeX), the certificates with reproduction scripts, the independent checks, the cross-check scripts for Sections 4 and 7 and a standalone Lean project that builds from scratch.

Limitations. Uniqueness, the Poincaré inequality of the corollary and the area law of Section 7 for rectangular Wilson loops (range not new); no other decay of correlations, no transfer-matrix gap and no log-Sobolev inequality is claimed. The Borgs–Seiler threshold was read in the preprint version only. The geodesic bound is certified only for κ ≤ 59/50. Prior art: Levin (1980), Royer (1986), Gross (1979), Simon (1979) and Wu (2006) were read; the bound 1/4 for all fields and the range 2/9 are in the preprint of Bhattacharjee and Bhattacharya. Not consulted: Seiler's monograph (LNP 159), the content of Georgii's book, Simon's book on lattice gases, the journal version of Borgs–Seiler, and Brydges–Giffen–Durhuus–Fröhlich (1986) beyond its abstract. A strong-coupling statement: nothing is claimed about the continuum limit or the Yang–Mills mass gap.

Prepared with the assistance of Claude (Anthropic); the analytic proof (Section 4) and the area-law section (Section 7, Appendix B) were read by an independent reviewer, Kimi K3.1 Pro (Moonshot AI), a language model; every claim in the note carries a status label.

Files

dobrushin_su2_wilson_v1_5.pdf

Files (3.5 MB)

Name Size
md5:dc57070850c7cbdcef54778d2f84fa9a
915.7 kB Preview Download
md5:22e9e6d8dd1f6c1ff1dfe1a2966ebc6c
2.6 MB Preview Download

Additional details