Lemma 4 — variance bound from the light-tail coefficients
OpenQiFixedDesign.Concentration.lemma_4optimal-transportp2o-gran-per-chapterp2o-plan-paperp2o-v1probabilitystatistics
Let be a real Borel probability law of mean zero. Suppose , and . Then its variance satisfies
The numerical constant and exponent are exactly those of (33). The variance is a nonnegative extended integral, whose finiteness follows from the exponential-moment assumption.
Differences from the source:
- encoding: A is the finite exponential moment and innovations have mean zero → a nonnegative extended integral equals the finite real A, and the real first integral is zero; reason: light tails guarantee first and second integrability, so no default-zero integral changes an instance
- encoding: the variance of the mean-zero innovation → the nonnegative extended integral of its square; reason: mean zero identifies the second moment with variance and an infinite moment cannot turn into zero
Preamble
import Mathlib import Definitions.Def_QiFixedDesign_Concentration_LightTail open MeasureTheory Matrix Filter Topology open scoped ENNReal
Formal statement
namespace QiFixedDesign.Concentration
theorem lemma_4 (μ : Measure ℝ) [IsProbabilityMeasure μ]
(a A : ℝ) (hTail : LightTail μ a A) (hMean : ∫ u, u ∂μ = 0) :
(∫⁻ u, ENNReal.ofReal (u ^ 2) ∂μ) ≤ ENNReal.ofReal (kappa a A) := by sorry
end QiFixedDesign.Concentration
Source
Qi, Cao and Shen (2021), Distributionally Robust Conditional Quantile Prediction with Fixed Design, DOI 10.1287/mnsc.2020.3903, supplied source.pdf, Lemma 4, (33), PDF/printed p. 16
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.