Step 0 of the martingale CLT: truncation is asymptotically harmless
DisprovedMartingale.clt_truncation_matchesFirst step of the proof of the martingale CLT. Replace the array by its truncation
(the indicator is -measurable, so is again a martingale difference array — that is the point of stopping the sum before ). The claim is that this substitution does not change the limit: the difference of the two row sums tends to in probability.
The reason is that under the limiting-variance hypothesis , the event on which the running sum of squares ever exceeds the threshold has probability tending to (for ); on its complement the two row sums coincide exactly. The truncation is what makes the increments bounded, so that the characteristic-function expansion in the main argument converges.
Isolating this step means the analytic core of the proof may be carried out for a bounded array.
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.clt_truncation_matches {Ω : Type*} {m0 : MeasurableSpace Ω}
(P : Measure Ω) [IsProbabilityMeasure P] (D : ℕ → ℕ → Ω → ℝ)
(hmeas : ∀ n k, Measurable (D n k))
(σ : ℝ)
(hvar : TendstoInMeasure P
(fun (n : ℕ) ω => ∑ k ∈ Finset.range n, D n k ω ^ 2) atTop (fun _ => σ ^ 2)) :
TendstoInMeasure P
(fun (n : ℕ) ω => (∑ k ∈ Finset.range n, D n k ω)
- ∑ k ∈ Finset.range n, (Set.indicator
{ω' | ∑ j ∈ Finset.range k, D n j ω' ^ 2 ≤ 2} (D n k) ω))
atTop (fun _ => 0) := by
sorry