Conditional Hoeffding lemma: bounded and conditionally centred is sub-Gaussian
ProvedhasCondSubgaussianMGF_of_mem_Icc_of_condExp_eq_zeroconcentrationconditionalhoeffdingprobabilitysub-gaussian
Conditional Hoeffding lemma. Let be a measurable random variable on a standard Borel probability space that is almost surely in and has conditional expectation zero given a sub-sigma-algebra ( a.e.). Then is conditionally sub-Gaussian over the conditional-expectation kernel with parameter . This is the fiberwise (kernel) analog of the measure-level Hoeffding lemma hasSubgaussianMGF_of_mem_Icc_of_integral_eq_zero, and is the bridge that turns a bounded, conditionally-centered martingale difference into the HasCondSubgaussianMGF hypothesis consumed by the Azuma-Hoeffding inequality (the engine of McDiarmid bounded-differences).
Preamble
import Mathlib.Probability.Moments.SubGaussian import Mathlib.Probability.Kernel.Condexp import Mathlib.MeasureTheory.Measure.Real open MeasureTheory ProbabilityTheory Real open scoped NNReal ENNReal
Formal statement
theorem hasCondSubgaussianMGF_of_mem_Icc_of_condExp_eq_zero
{Ω : Type*} {m : MeasurableSpace Ω} [mΩ : MeasurableSpace Ω]
[StandardBorelSpace Ω] {μ : Measure Ω} [IsProbabilityMeasure μ]
(hm : m ≤ mΩ) {X : Ω → ℝ} {a b : ℝ}
(hX : Measurable X)
(hb : ∀ᵐ ω ∂μ, X ω ∈ Set.Icc a b)
(hc : μ[X | m] =ᵐ[μ] 0) :
HasCondSubgaussianMGF m hm X ((‖b - a‖₊ / 2) ^ 2) μ := by sorrySource
Boucheron, Lugosi, Massart, Concentration Inequalities (Oxford 2013), Sec 2.6 (Hoeffding lemma) and Sec 6.1 (martingale/bounded-differences method); Hoeffding, JASA 58 (1963) Sec 2. Conditional version = fiberwise Hoeffding over the regular conditional distribution (condExpKernel), a Markov kernel.