Standard Gaussian reciprocal shifted-tail integral bound
ProvedBanditAlgorithm.standardGaussian_reciprocal_shift_integral_boundanalysisgaussianprobabilitythompson-sampling
Let . Conditional on , let and put
There is a universal constant such that
This is the scale-free analytic core of the optimal-arm estimate: after standardizing a sample mean based on unit-variance Gaussian rewards, one has .
Preamble
import Mathlib.Probability.Distributions.Gaussian.Real open MeasureTheory ProbabilityTheory ENNReal
Formal statement
namespace BanditAlgorithm
/-- The standardized Gaussian reciprocal-tail integral in Exercise 36.6(a). -/
theorem standardGaussian_reciprocal_shift_integral_bound :
∃ C : ℝ, 0 < C ∧
∀ a : ℝ, 0 < a →
(∫⁻ z, ENNReal.ofReal
(1 / (gaussianReal z 1).real {x | -a < x} - 1)
∂gaussianReal 0 1) ≤
ENNReal.ofReal (C / a ^ 2) := by
sorry
end BanditAlgorithmSource
Lattimore and Szepesvári, Bandit Algorithms (CUP 2020), Exercise 36.6(a) and its Gaussian-tail hint, printed p. 475 / PDF p. 484. This is exactly the standardized one-rank Gaussian integral in that exercise.