Answers are grounded in this paper and cite their sources.
Consider the Edwards–Anderson spin glass in a box V ⊂ ℤ^d, d ≥ 2, with i.i.d. couplings whose law is symmetric and atomless, and let ℰ_V be the event that the ground-state product σ_iσ_j at an interior nearest-neighbour bond {i, j} depends on the boundary condition. For a bond at the centre of a cube of side L and Gaussian couplings, Chatterjee proved ℙ(ℰ_V) ≥ 1/(32L). We prove ℙ(ℰ_V) ≥ 1/(8d(ρ_∥ + 1)) for every interior bond of every box, where ρ_∥ is the distance from the bond to the nearest face of V parallel to it, and the same bound, with ρ_∥ replaced by the distance to the hyperplane, for finite domains lying on one side of a lattice hyperplane parallel to the bond. Hence ℙ(ℰ_V) ≥ 1/(16dKr) for a bond at distance r from the boundary whenever a parallel face lies within distance Kr (K ≥ 1), and every interior bond of a cube of side L has ℙ(ℰ_V) ≥ 1/(4d(L + 2)), which exceeds 1/(32L) for d = 2 and every L, and for d ≤ 7 and L large. The proof abstracts Chatterjee's argument, which reverses the couplings on pieces of the boundary of a cube and follows the effect along paths; our witnesses are rotated and reflected copies of the box. We show that this flip-and-path method cannot give more than 1/ρ_∥: every instance of it, on a box or a torus, uses at least ρ_∥ witness edges. The case of a bond whose nearest face is perpendicular to it, with all parallel faces far away, is left open. On strips ℤ × Λ, whose long sides carry no boundary condition, the general-position bound fails: for laws with a bounded density bounded below near 0, the sensitivity of an edge at distances r ≤ R from the two ends is at most Ce^(−cr) + C/R.
No takes yet. Share an insight, caveat, or question.
Yan Ru Pei (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: