Theorem 2 — integration at high temperature
OpenSpatialSorting.Supplement.integration_high_temperatureLet be a number of colours, a cost matrix (symmetric, zero diagonal, positive off-diagonal entries, largest entry ) and a density vector with positive entries. Consider the fixed-magnetization Generalized Potts Model on the rhombus of the triangular lattice, with collar wired to colour and prescribed colour counts, and its Gibbs distribution ; let be the number of free sites.
Given and , set
There is , depending only on , and , such that for every , every , every and every ,
At high temperature colours are spread nearly uniformly: every low-perimeter region carries its fair share of each colour up to , with overwhelming probability. Together with Theorem 1 this separates the two temperature regimes (Corollary 1.7).
Formalization Note The dependence is expressed by the quantifier order: is chosen after and before .
Differences from the source: none.
import Mathlib import Definitions.Def_SpatialSorting_IsCostMatrix import Definitions.Def_SpatialSorting_amax import Definitions.Def_SpatialSorting_IsDensity import Definitions.Def_SpatialSorting_N import Definitions.Def_SpatialSorting_Supplement_integratedProb open Classical Filter Topology
namespace SpatialSorting.Supplement
/-- Theorem 2 — integration at high temperature. With `d = min{δ, 1}`, `β_high = d² / (192 A_max)` and
`ζ = e^{−d²/16}`, there is `L_0 = L_0(c, δ, q)` such that for `0 ≤ β ≤ β_high` and `L ≥ L_0`,
`π_{Λ^L,β,ρ}(Integrated(c, δ)) ≥ 1 − ζ^{N_L}`.
Calvert, Oh, Randall, *Spatial Sorting and the Geometry of Lattice Spin Systems at Fixed Magnetization*, STOC 2027 submission manuscript (sha256 4d4d4b7d…7fae), Theorem 2, p. 9 (PDF p. 10).
Formalization Notes:
- encoding: `L_0 = L_0(c, δ, q)` → `L_0` is chosen after `q, c, δ` and before `A, ρ, β`; quantifier
order expresses the dependence.
- encoding: `A_max` is the base `amax A`, positive for `q ≥ 2` and `A ∈ 𝒜_q`; `N_L` is the base `N L`.
- encoding: `ζ^{N_L}` is written `(e^{−d²/16})^{N_L}` with a natural-number power.
- `β` ranges over the closed interval `[0, β_high]` (β = 0 is the uniform law on `Ω_{Λ^L,ρ}`);
`δ > 0` has no upper bound and `c > 0` is arbitrary, as on the page. -/
theorem integration_high_temperature :
∀ (q : ℕ) [NeZero q], 2 ≤ q → ∀ c δ : ℝ, 0 < c → 0 < δ →
∃ L0 : ℕ, ∀ A : Matrix (Fin q) (Fin q) ℝ, IsCostMatrix A →
∀ ρ : Fin q → ℝ, IsDensity ρ →
∀ β : ℝ, 0 ≤ β → β ≤ (min δ 1) ^ 2 / (192 * amax A) →
∀ L : ℕ, L0 ≤ L →
1 - Real.exp (-((min δ 1) ^ 2 / 16)) ^ (N L) ≤ integratedProb A β ρ L c δ := by sorry
end SpatialSorting.Supplement