Extend an eventwise conditional probability law to bounded measurable observables
ProvedMeasureTheory.conditional_mean_of_eventwise_probability_lawbanditsp2o-batch-p100ap2o-gran-per-chapterp2o-plan-paperp2o-v1revenue-managementthompson-sampling
If the conditional probability of every measurable event of X agrees almost surely with a family of probability measures, then any measurable real observable uniformly bounded almost surely under every measure in that family is integrable when evaluated at X, and its conditional expectation equals its integral against that family. The bound may be zero, and the observable may take either sign.
Preamble
import Mathlib.MeasureTheory.Function.ConditionalExpectation.Basic open MeasureTheory open scoped MeasureTheory
Formal statement
theorem MeasureTheory.conditional_mean_of_eventwise_probability_law
{Ω : Type*} {E : Type*} [mΩ : MeasurableSpace Ω] [MeasurableSpace E]
(P : Measure Ω) [IsProbabilityMeasure P]
(m : MeasurableSpace Ω) (hm : m ≤ mΩ)
(X : Ω → E) (hX : Measurable X)
(ν : Ω → Measure E) (hν : ∀ ω, IsProbabilityMeasure (ν ω))
(hLaw : ∀ B : Set E, MeasurableSet B →
P[(X ⁻¹' B).indicator (fun _ => (1 : ℝ)) | m] =ᵐ[P]
fun ω => (ν ω B).toReal)
(f : E → ℝ) (hf : Measurable f)
(C : ℝ) (hC : 0 ≤ C)
(hbound : ∀ ω, ∀ᵐ x ∂(ν ω), ‖f x‖ ≤ C) :
Integrable (fun ω => f (X ω)) P ∧
P[(fun ω => f (X ω)) | m] =ᵐ[P] fun ω => ∫ x, f x ∂(ν ω) := by sorrySource
Ferreira, Simchi-Levi, Wang, Online Network Revenue Management Using Thompson Sampling, Oper. Res. 66(6), 2018, p. 1594, Theorem 1