Proposition 2.4 — 𝔼[1 ∧ e^A] = Φ(μ/σ) + exp(μ + σ²/2)Φ(−σ − μ/σ) for A ~ N(μ, σ²)
ProvedRoberts1997.RWM.expect_min_one_exp_gaussiangaussianmetropolisp2o-batch-p200ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let , and . With the standard normal distribution function,
This computes the expected acceptance probability when the log acceptance ratio is Gaussian; it yields the speed and the acceptance rate .
Formalization Note The paper writes and divides by ; the formula is read with (for the law is the same but the right side changes). The variance is (σ ^ 2).toNNReal. The integrand lies in , so the integral is a genuine expectation.
Preamble
import Definitions.Def_Roberts1997_RWM_Speed open MeasureTheory ProbabilityTheory
Formal statement
namespace Roberts1997.RWM
/-- Proposition 2.4 (p. 115). If `A ~ N(μ, σ²)` with `σ > 0`, then
`𝔼[1 ∧ e^A] = Φ(μ/σ) + exp(μ + σ²/2) Φ(-σ - μ/σ)`. -/
theorem expect_min_one_exp_gaussian (μ σ : ℝ) (hσ : 0 < σ) :
∫ a, min 1 (Real.exp a) ∂(gaussianReal μ (σ ^ 2).toNNReal) =
Phi (μ / σ) + Real.exp (μ + σ ^ 2 / 2) * Phi (-σ - μ / σ) := by sorry
end Roberts1997.RWM
Source
Roberts, Gelman, Gilks, Weak convergence and optimal scaling of random walk Metropolis algorithms, Ann. Appl. Probab. 7(1), 1997, p. 115, Proposition 2.4
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.