Lévy continuity for statistics of trajectories of growing length: convergence of characteristic functions gives convergence to
ProvedTPG.tendstoNormal_of_charFuncausal-inferencemarkov-chainnonstationaritypolicy-gradienttreatment-effect
For each horizon let be a measurable real statistic of the trajectory generated under the experimental policy in an environment of TPG_Model, and let .
Lemma. If for every real
then converges in distribution to : for every bounded continuous , .
This is Lévy's continuity theorem for the laws of the ; it turns a characteristic-function estimate into the mode of convergence used in the statements of Lemma 3 and Theorem 2.
Formalization Note The expectation for each is taken under the horizon- trajectory law, so the statistics for different are defined on different probability spaces.
Preamble
import Mathlib import Definitions.Def_TPG_Model import Definitions.Def_TPG_Estimators import Definitions.Def_TPG_Inference open MeasureTheory ProbabilityTheory open scoped BigOperators ENNReal
Formal statement
namespace TPG
theorem tendstoNormal_of_charFun {S : Type} [Fintype S] [MeasurableSpace S] [MeasurableSingletonClass S]
(E : Env S) (W : (T : ℕ) → Traj S T → ℝ) (hW : ∀ T : ℕ, Measurable (W T))
(v : ℝ) (hv : 0 ≤ v)
(h : ∀ t : ℝ, Filter.Tendsto
(fun T : ℕ => ∫ ω, Complex.exp (Complex.I * (t : ℂ) * (W T ω : ℂ)) ∂(E.expLaw T (1 / 2)))
Filter.atTop (nhds (Complex.exp (-((v : ℂ) * (t : ℂ) ^ 2) / 2)))) :
E.TendstoNormal W v := by sorry
end TPGSource
Johari, Peng, Xing, Estimation of Treatment Effects Under Nonstationarity via the Truncated Policy Gradient Estimator, revised manuscript of October 2026 (revision of arXiv:2506.05308v3, https://arxiv.org/abs/2506.05308); numbering follows the revised manuscript. Auxiliary statement, not a numbered result of the paper: it is the standard continuity theorem used implicitly in the proof of Lemma 3, Appendix H.1, pp. 51–53, adapted to the convention that the statistic at horizon T lives on the horizon-T trajectory space.