Laplace-mixture envelope for reciprocal Gaussian tails
ProvedBanditAlgorithm.gaussian_reciprocal_tail_laplace_envelopeanalysisgaussianprobabilitythompson-sampling
Let for a standard Gaussian random variable . There is a universal constant such that, for every real ,
This inequality packages the reciprocal Gaussian-tail estimate as a nonnegative Laplace mixture. It is designed for reuse with exponential supermartingale or moment-generating-function bounds.
Preamble
import Mathlib.Probability.Distributions.Gaussian.Real open MeasureTheory ProbabilityTheory
Formal statement
namespace BanditAlgorithm
/-- A Laplace-mixture envelope for the reciprocal standard-Gaussian upper tail. -/
theorem gaussian_reciprocal_tail_laplace_envelope :
∃ C : ℝ, 0 < C ∧ ∀ u : ℝ,
ENNReal.ofReal
(1 / (gaussianReal 0 1).real (Set.Ioi u) - 1) ≤
ENNReal.ofReal C *
(∫⁻ l : ℝ in Set.Ioi 0,
ENNReal.ofReal
(l * Real.exp (l * u - l ^ 2 / 2))) := by
sorry
end BanditAlgorithmSource
Lattimore and Szepesvári, Bandit Algorithms (CUP 2020), Gaussian-tail hint to Exercise 36.6(a), printed p. 475 / PDF p. 484. This is a purely analytic Laplace-mixture reformulation of the hinted Gaussian reciprocal-tail calculation.