Lévy continuity in CLT form: pointwise convergence of to gives
ProvedMartingale.tendstoInDistribution_gaussian_of_tendsto_charFunLet be measurable real random variables on a probability space , and let . If for every
then converges in distribution to .
What it is for. This is the packaging of Lévy's continuity theorem in the exact form that central limit arguments produce their output. A CLT proof — whether for independent summands, martingale differences, or mixing sequences — invariably ends with a pointwise statement about characteristic functions: for each fixed , the expectation converges to the Gaussian characteristic function. The conclusion that is actually wanted, weak convergence of the laws, then requires (i) recognising as the characteristic function of the pushforward law , (ii) computing the characteristic function of , and (iii) invoking Lévy continuity. This statement performs all three once, so that a CLT proof can stop at the characteristic-function estimate.
On the degenerate case. No positivity of is assumed beyond : for the conclusion is convergence in distribution to the Dirac mass at , which is the correct degenerate statement and is exactly what is needed when a CLT is applied to a functional of asymptotic variance zero. Formally this is handled by the coercion into , which is the identity precisely because .
Proof. The pushforward is a probability measure since is measurable, and by the change-of-variables formula its characteristic function at is
The characteristic function of is , which at is the assumed limit. Lévy's continuity theorem — pointwise convergence of characteristic functions to the characteristic function of a probability measure implies weak convergence — then gives convergence of the laws, which is the definition of convergence in distribution.
import Mathlib.MeasureTheory.Function.ConvergenceInDistribution import Mathlib.MeasureTheory.Measure.LevyConvergence import Mathlib.Probability.Distributions.Gaussian.Real open MeasureTheory ProbabilityTheory Filter open scoped ENNReal NNReal Topology ProbabilityTheory
theorem Martingale.tendstoInDistribution_gaussian_of_tendsto_charFun
{Ω : Type*} {m0 : MeasurableSpace Ω} (P : Measure Ω) [IsProbabilityMeasure P]
(X : ℕ → Ω → ℝ) (hX : ∀ n, Measurable (X n)) (v : ℝ) (hv : 0 ≤ v)
(h : ∀ t : ℝ, Tendsto (fun n : ℕ => ∫ ω, Complex.exp (Complex.I * t * (X n ω : ℂ)) ∂P) atTop
(𝓝 (Complex.exp (-(v * t ^ 2) / 2)))) :
TendstoInDistribution X atTop (id : ℝ → ℝ) (fun _ => P) (gaussianReal 0 v.toNNReal) := by sorry