Proof of Theorem A.2, pp. 745–746 — a = log[(1 − θ)/θ]/(1 − 2θ), s = (1 − 2θ)/(2θ(1 − θ)) solve (A.1)
ProvedLogSobolevMC.TwoPoint.eq_A_1_solutionlog-sobolev-inequalityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1real-analysis
Fix and let and be the functions of the proof of Theorem A.2 (p. 743–744), namely and of the function taking the values and on the two-point space with . Put
Then is a nonzero point of and solves the system
This is the explicit solution that, by the shape analysis of Figure 2, identifies in Theorem A.2.
Formalization Note and are the derivatives in (Lean's deriv); both functions are differentiable at this , where and are positive. The page prints the interval as ""; p. 743 says varies between and , which is the interval used here.
Preamble
import Mathlib import Definitions.Def_mm_basic import Definitions.Def_mm_lower import Definitions.Def_LogSobolevMC_TwoPoint_Setting
Formal statement
namespace LogSobolevMC.TwoPoint
/-- Proof of Theorem A.2, pp. 745–746: for `0 < θ < 1/2`, `a = log[(1 − θ)/θ]/(1 − 2θ)` and
`s = (1 − 2θ)/(2θ(1 − θ))`, `s` is a nonzero point of `[−(1 − θ)⁻¹, θ⁻¹]` solving the system (A.1)
`l(s) − a e(s) = 0`, `l′(s) − a e′(s) = 0`. -/
theorem eq_A_1_solution (θ : ℝ) (h0 : 0 < θ) (h1 : θ < 1 / 2) :
let a := Real.log ((1 - θ) / θ) / (1 - 2 * θ)
let s := (1 - 2 * θ) / (2 * θ * (1 - θ))
s ≠ 0 ∧ -(1 - θ)⁻¹ ≤ s ∧ s ≤ θ⁻¹ ∧
lFun θ s - a * eFun θ s = 0 ∧
deriv (lFun θ) s - a * deriv (eFun θ) s = 0 := by sorry
end LogSobolevMC.TwoPoint
Source
Diaconis and Saloff-Coste, Logarithmic Sobolev inequalities for finite Markov chains, Ann. Appl. Probab. 6 (1996), pp. 743–746, proof of Theorem A.2, (A.1) and the displays for a and s on p. 745
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.