Proposition 8, p. 610 — in probability for normal forces
ProvedStoneRegression.Criterion.proposition_8Let be i.i.d. -valued with law , and let be independent standard normal real random variables, independent of . Let be a sequence of Borel weights such that
Then
Since , this is the step from consistency to condition (5) in the necessity half of Theorem 1.
Formalization Note. The variables and are realized as the coordinates of the product law of i.i.d. pairs with and independent of (the kernel ). This also produces a normal variable paired with itself, which the statement does not use. Since the joint law of is the same for every such sequence and convergence in probability depends only on the joint law, the paper's "there is a sequence " is captured exactly. Weights are assumed jointly Borel (standing assumption).
import Mathlib import Definitions.Def_StoneRegression_Criterion_Setting open MeasureTheory ProbabilityTheory Filter Topology open scoped ENNReal NNReal BoundedContinuousFunction
namespace StoneRegression.Criterion
/-- Proposition 8 (Stone 1977, p. 610). Let `Y₁, Y₂, …` be i.i.d. standard normal and independent of
`(X, X₁, X₂, …)`; these are the response coordinates of `pairLaw μ (Kernel.const _ (gaussianReal 0 1))`
(the response paired with `X` itself is also present there, and unused). If `∑ᵢ W_{ni}(X) Yᵢ → 0` in
probability, then `∑ᵢ W²_{ni}(X) → 0` in probability. -/
theorem proposition_8 {d : ℕ} (μ : Measure (EuclideanSpace ℝ (Fin d))) [IsProbabilityMeasure μ]
(W : WeightSeq d) (hW : MeasurableWeights W)
(hY : TendstoInMeasure (pairLaw μ (Kernel.const (EuclideanSpace ℝ (Fin d)) (gaussianReal 0 1)))
(fun n ω => estimate W n ω) atTop (fun _ => 0)) :
TendstoInMeasure (xLaw μ) (fun n ω => ∑ i : Fin n, wAt W n ω i ^ 2) atTop (fun _ => 0) := by sorry
end StoneRegression.Criterion
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.