Proposition 15, p. 617 — E|L̂ₙ|ʳ I{|L̂ₙ| ≥ M} ≤ C/(p ∧ (1 − p)) · E|Y|ʳ I{|Y| ≥ M}, and the same for Ûₙ
ProvedStoneRegression.Quantile.proposition_15conditional-quantilesnonparametric-regressionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1uniform-integrability
Fix . Let be a (jointly Borel) probability weight function satisfying condition (1) with a constant :
Let , and , and let be any real response with i.i.d. Then
and the same inequality holds with replaced by .
Since the right side does not depend on and tends to as when , the proposition gives uniform integrability of and , which turns the convergence in probability of Theorem 3 into convergence in .
Formalization Note. The page leaves unquantified; the Lean statement takes , the range in which Theorem 3 uses it. Condition (1) is required at the given only. Expectations are -valued lower integrals. The weight function is assumed jointly Borel.
Preamble
import Mathlib import Definitions.Def_StoneRegression_Criterion_Setting import Definitions.Def_StoneRegression_Quantile_Quantiles
Formal statement
namespace StoneRegression.Quantile
open MeasureTheory ProbabilityTheory Filter Topology
open scoped ENNReal NNReal
/-- Proposition 15, p. 617, for one `n ≥ 1`: if `Wₙ` is a (Borel) probability weight function satisfying
(1) at this `n` with constant `C ≥ 1`, then for `0 < p < 1`, `M > 0` and `r ≥ 1`,
`E|L̂ₙ(p|X)|ʳ I{|L̂ₙ(p|X)| ≥ M} ≤ C/(p ∧ (1 − p)) · E|Y|ʳ I{|Y| ≥ M}`, and the same with `Ûₙ(p|X)`.
The page leaves `r` unquantified; `r ≥ 1` is the range Theorem 3 uses. -/
theorem proposition_15 {d : ℕ} (μ : Measure (EuclideanSpace ℝ (Fin d))) [IsProbabilityMeasure μ]
(W : StoneRegression.Criterion.WeightSeq d) (n : ℕ) (hn : 1 ≤ n)
(hWn : Measurable
(fun q : EuclideanSpace ℝ (Fin d) × (Fin n → EuclideanSpace ℝ (Fin d)) => W n q.1 q.2))
(hPn : ∀ x xs, (∀ i, 0 ≤ W n x xs i) ∧ ∑ i, W n x xs i = 1)
(C : ℝ≥0) (hC : 1 ≤ C)
(h1 : ∀ f : EuclideanSpace ℝ (Fin d) → ℝ≥0, Measurable f →
∫⁻ ω, ∑ i : Fin n, ‖StoneRegression.Criterion.wAt W n ω i‖ₑ * (f (ω (i.val + 1)) : ℝ≥0∞) ∂(StoneRegression.Criterion.xLaw μ) ≤ C * ∫⁻ x, f x ∂μ)
(p : ℝ) (hp0 : 0 < p) (hp1 : p < 1) (M : ℝ) (hM : 0 < M) (r : ℝ) (hr : 1 ≤ r)
(κ : Kernel (EuclideanSpace ℝ (Fin d)) ℝ) [IsMarkovKernel κ] :
∫⁻ ω, ‖estLowerQ W n p ω‖ₑ ^ r * (if M ≤ |estLowerQ W n p ω| then 1 else 0) ∂(StoneRegression.Criterion.pairLaw μ κ) ≤
((C : ℝ≥0∞) / ENNReal.ofReal (min p (1 - p))) *
∫⁻ q, ‖q.2‖ₑ ^ r * (if M ≤ |q.2| then 1 else 0) ∂(μ ⊗ₘ κ) ∧
∫⁻ ω, ‖estUpperQ W n p ω‖ₑ ^ r * (if M ≤ |estUpperQ W n p ω| then 1 else 0) ∂(StoneRegression.Criterion.pairLaw μ κ) ≤
((C : ℝ≥0∞) / ENNReal.ofReal (min p (1 - p))) *
∫⁻ q, ‖q.2‖ₑ ^ r * (if M ≤ |q.2| then 1 else 0) ∂(μ ⊗ₘ κ) := by sorry
end StoneRegression.Quantile
Source
Stone (1977), Ann. Statist. 5, Proposition 15, p. 617 (§12)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.