Documentation

LeanPool.Feige.KernelAveraging

Averaging conditional probability bounds #

This file records the elementary final step used after the latent two-point decomposition: an almost-everywhere event bound for the conditional Markov kernel survives averaging over a probability law.

theorem Feige.measureReal_bind_apply_le {A : Type u_1} {B : Type u_2} [MeasurableSpace A] [MeasurableSpace B] (κ : ProbabilityTheory.Kernel A B) [ProbabilityTheory.IsMarkovKernel κ] (ν : MeasureTheory.Measure A) [MeasureTheory.IsProbabilityMeasure ν] {s : Set B} (hs : MeasurableSet s) {c : } (hc0 : 0 c) (hc : ∀ᵐ (a : A) ν, (κ a).real s c) :
(ν.bind κ).real s c

An almost-everywhere real-valued probability bound for the fibres of a Markov kernel is inherited by the mixture measure.