Step 3 of the martingale CLT:
ProvedMartingale.expectation_prod_one_addThe step where the martingale property does its work. For a bounded martingale difference sequence adapted to and any real ,
By induction on : the partial product is -measurable and bounded, so it pulls out of the conditional expectation, and
Boundedness is what makes the complex product integrable and licenses pulling it out; in the main proof it is supplied by the truncation step.
Why this identity is the crux. For independent summands one proves a central limit theorem by factoring the characteristic function , which is unavailable for martingales. McLeish's device is to compare with the product , which is not a characteristic function but does have expectation exactly for the reason above — the martingale property replaces independence at precisely this point. Writing with , one then shows and that is uniformly integrable and tends to , so that , and Lévy's continuity theorem finishes.
import Mathlib.Probability.Martingale.Basic import Mathlib.MeasureTheory.Function.ConvergenceInDistribution import Mathlib.MeasureTheory.Function.ConvergenceInMeasure import Mathlib.Probability.Distributions.Gaussian.Real open MeasureTheory ProbabilityTheory Filter open scoped ENNReal NNReal Topology ProbabilityTheory
theorem Martingale.expectation_prod_one_add {Ω : Type*} {m0 : MeasurableSpace Ω}
(P : Measure Ω) [IsProbabilityMeasure P] (ℱ : Filtration ℕ m0)
(Z : ℕ → Ω → ℝ) (hmeas : ∀ k, Measurable (Z k))
(hadapt : ∀ k, Measurable[ℱ k] (Z k))
(hint : ∀ k, Integrable (Z k) P)
(hmds : ∀ k, P[Z (k + 1) | ℱ k] =ᵐ[P] 0)
(hcent : ∫ ω, Z 0 ω ∂P = 0)
(C : ℝ) (hbdd : ∀ k ω, |Z k ω| ≤ C) (θ : ℝ) (n : ℕ) :
∫ ω, (∏ k ∈ Finset.range n, (1 + Complex.I * θ * (Z k ω : ℂ))) ∂P = 1 := by
sorry