Lemma 4.1, p.13 — Hubbard–Stratonovich representation of
ProvedCritWinSpin.lemma_4_1p2o-gran-per-chapterp2o-plan-paperp2o-v1spin-systems
Let be positive semidefinite with , and let , be as in (19). If has density and , then has law , that is, for every ,
This reduces discrete sampling to sampling a continuous vector.
Differences from the source:
- encoding: X has marginal law π → the integral ratio at every x, with integrability of both integrands stated as conclusions; reason: the definition of the marginal; integrability rules out the junk value of the Bochner integral
Preamble
import Mathlib import Definitions.Def_CritWinSpin_Wpot open MeasureTheory set_option autoImplicit false
Formal statement
namespace CritWinSpin
/-- **Lemma 4.1, Hubbard–Stratonovich representation** (Liu–Zhang–Ren, STOC 2027 submission, manuscript
sha256 f77584da…, p.13). For `J = LLᵀ` positive semidefinite, if `Z` has density `∝ e^{−V}` and
`X ∼ P(· | Z)`, then `X` has law `π^ρ_{J,h}`. -/
theorem lemma_4_1 (S : SpinLaw) (hS : IsGHS S) {N : ℕ} (J L : Matrix (Fin N) (Fin N) ℝ)
(h : Fin N → ℝ) (hJ : J.PosSemidef) (hL : J = L * L.transpose) (x : Config S (Fin N)) :
Integrable (fun z : Fin N → ℝ => Real.exp (-hsV S L h z)) ∧
Integrable (fun z : Fin N → ℝ => condP S L h z x * Real.exp (-hsV S L h z)) ∧
(∫ z : Fin N → ℝ, condP S L h z x * Real.exp (-hsV S L h z)) /
(∫ z : Fin N → ℝ, Real.exp (-hsV S L h z)) = gibbs S J h x := by sorry
end CritWinSpin
Source
Liu, Zhang, Ren, The Critical Window for Sampling and Learning-to-Sample in Spin Systems, STOC 2027 submission (manuscript sha256 f77584da…), Lemma 4.1, p.13, eq. (19)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.