Theorem 3.7 — chi-square convergence from log-Sobolev inequalities
ProvedLogSobolevMC.ChiSquare.theorem_3_7chi-square-distancehypercontractivitylog-sobolev-inequalitymarkov-chainsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be an irreducible finite-state Markov chain with positive stationary distribution . Write for its log-Sobolev constant, for its variational spectral gap, and . Fix a state with and a real number . Then
If is reversible, the same bound holds at the earlier time
The theorem bounds the chi-square distance of a heat-kernel row from stationarity in terms of the log-Sobolev constant and spectral gap. The two conclusions retain the distinct general and reversible constants.
Preamble
import Mathlib import Definitions.Def_LogSobolevMC_ChiSquare_Setting
Formal statement
namespace LogSobolevMC.ChiSquare
theorem theorem_3_7 {V : Type*} [Fintype V] [DecidableEq V]
(K : Matrix V V ℝ) (hK : MarkovMixing.IsStochastic K)
(π : V → ℝ) (hπ : MarkovMixing.IsStationary K π)
(hπpos : ∀ x, 0 < π x) (hirr : MarkovMixing.Irreducible K)
(x : V) (hx : π x ≤ Real.exp (-1)) (c : ℝ) (hc : 0 < c) :
let L := Real.log (Real.log (1 / π x))
(lpNorm π 2 (fun y =>
density K π ((2 * logSobolev K π)⁻¹ * L + c / gap K π) x y - 1) ≤
Real.exp (1 - c)) ∧
(MarkovMixing.DetailedBalance K π →
lpNorm π 2 (fun y =>
density K π ((4 * logSobolev K π)⁻¹ * L + c / gap K π) x y - 1) ≤
Real.exp (1 - c)) := by sorry
end LogSobolevMC.ChiSquare
Source
Diaconis and Saloff-Coste, Logarithmic Sobolev inequalities for finite Markov chains, Ann. Appl. Probab. 6 (1996), p. 723, Theorem 3.7, https://doi.org/10.1214/aoap/1034968224
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.