Example 5 — a pointwise compact class with integrable envelope is Glivenko–Cantelli
ProvedGenEmpLik.Consistency.example_5empirical-processesp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1uniform-law-of-large-numbers
Let be i.i.d. random elements of with law . Let be compact, let be measurable for each , and let be continuous on for -almost all . If there is a measurable envelope with
then is Glivenko–Cantelli.
This is the standard sufficient condition (van der Vaart, Asymptotic Statistics, Example 19.8) by which the Glivenko–Cantelli hypothesis of Theorem 7 is verified for continuous losses on a compact decision set; Corollary 1 uses it in exactly this way.
Formalization Note Samples are 0-based and is the law of ; "Glivenko–Cantelli" is the definition of this mission (integrable class, supremum in , convergence ∀ᵐ). The continuity is assumed for -almost every , as printed.
Preamble
import Mathlib import Definitions.Def_GenEmpLik_Consistency_IsGlivenkoCantelli open MeasureTheory ProbabilityTheory Filter Topology
Formal statement
namespace GenEmpLik.Consistency
/-- Example 5 (arXiv:1610.03425v3, p. 16; van der Vaart 1998, Example 19.8). Let `ξ₀, ξ₁, …`
be i.i.d. with law `P₀`, let `X` be compact, let `ℓ(·; ξ)` be continuous on `X` for
`P₀`-almost every `ξ`, and let `ℓ(x; ·)` be measurable for each `x ∈ X`. If there is a measurable envelope
`Z : Ξ → ℝ₊` with `|ℓ(x; ξ)| ≤ Z(ξ)` for all `x ∈ X`, `ξ ∈ Ξ` and `E_{P₀}[Z] < ∞`, then
`{ℓ(x; ·) : x ∈ X}` is Glivenko–Cantelli. -/
theorem example_5 {Ω Ξ : Type*} [MeasurableSpace Ω] [MeasurableSpace Ξ]
{μ : Measure Ω} [IsProbabilityMeasure μ] {d : ℕ}
(ξ : ℕ → Ω → Ξ) (hξm : ∀ i, Measurable (ξ i)) (hξind : iIndepFun ξ μ)
(hξid : ∀ i, IdentDistrib (ξ i) (ξ 0) μ μ)
(X : Set (EuclideanSpace ℝ (Fin d))) (hX : IsCompact X)
(ℓ : EuclideanSpace ℝ (Fin d) → Ξ → ℝ) (hℓm : ∀ x ∈ X, Measurable (ℓ x))
(hcont : ∀ᵐ s ∂(μ.map (ξ 0)), ContinuousOn (fun x => ℓ x s) X)
(Z : Ξ → ℝ) (hZm : Measurable Z) (hZ0 : ∀ s, 0 ≤ Z s)
(hZ : ∀ x ∈ X, ∀ s, |ℓ x s| ≤ Z s) (hZint : Integrable Z (μ.map (ξ 0))) :
IsGlivenkoCantelli μ ξ ((fun x => ℓ x) '' X) := by sorry
end GenEmpLik.Consistency
Source
Duchi, Glynn & Namkoong, Statistics of Robust Optimization: A Generalized Empirical Likelihood Approach, arXiv:1610.03425v3, p. 16, Example 5 (quoting van der Vaart 1998, Example 19.8)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.