Proof of Lemma 5.1, p. 446 — F_X⁻¹(U) has the law of X, and F_{X₂} ≤ F_{X₁} gives F_{X₁}⁻¹(U) ≤ F_{X₂}⁻¹(U)
ProvedConvexRiskFn.Order.quantile_couplingcouplingp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1quantilestochastic-order
Let be a probability space carrying a uniform random variable (measurable, for ). For a random variable with cdf , write and .
- For every measurable , is a random variable with the same distribution as :
- For measurable with for all ,
This is the quantile construction of the proof of Lemma 5.1: it realizes two random variables ordered by as pointwise ordered random variables with the same laws on the same space.
Formalization Note The paper says "for all ". The Lean states the order almost surely: only almost surely, and outside the paper's is while the Lean real infimum returns .
Preamble
import Mathlib import Definitions.Def_StochasticOrders_MonotoneConvex_IcxOrder import Definitions.Def_ConvexRiskFn_Order_Setting open MeasureTheory
Formal statement
namespace ConvexRiskFn.Order
/-- Proof of Lemma 5.1, p. 446: for a uniform random variable `U`, `X̂ := F_X⁻¹(U)` is a random
variable with the distribution of `X`, and `F_{X₂} ≤ F_{X₁}` gives `F_{X₁}⁻¹(U) ≤ F_{X₂}⁻¹(U)`
(almost surely: `U ∈ (0, 1)` only almost surely). -/
theorem quantile_coupling {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(U : Ω → ℝ) (hU : IsUniformRV P U) :
(∀ X : Ω → ℝ, Measurable X →
Measurable (fun ω => cdfInv (cdfOf P X) (U ω)) ∧
∀ t : ℝ, P {ω | cdfInv (cdfOf P X) (U ω) ≤ t} = P {ω | X ω ≤ t}) ∧
(∀ X₁ X₂ : Ω → ℝ, Measurable X₁ → Measurable X₂ →
(∀ t : ℝ, cdfOf P X₂ t ≤ cdfOf P X₁ t) →
∀ᵐ ω ∂P, cdfInv (cdfOf P X₁) (U ω) ≤ cdfInv (cdfOf P X₂) (U ω)) := by sorry
end ConvexRiskFn.Order
Source
Ruszczyński, Shapiro, Optimization of convex risk functions, Math. Oper. Res. 31 (2006), p. 446, proof of Lemma 5.1, second paragraph
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.