Corollary 3.19 — R_σ(Y) is the maximum of E(Y·σ(U)) over uniform U, attained by a co-monotone coupling
OpenMultistageStochastic.distortion_comonotone_maxLet be a distortion function and on an atomless probability space. Then for every random variable that is uniformly distributed on under ,
there is a uniformly distributed for which equality holds, and equality holds for every uniform coupled with in a co-monotone way, i.e. with for all (the source's footnote 5):
This is Corollary 3.19, obtained from the dual representation by observing that is feasible for (3.15) — its quantile function is itself — and that the maximum is attained when and are coupled co-monotonically, that is, when almost surely. It is the form in which distortion functionals are evaluated on scenario models.
Formalization Note The maximum is stated as its two halves, an upper bound for every uniform and attainment by some uniform , which is what "max" asserts; the corollary's second sentence, "the maximum is attained whenever and are coupled in a co-monotone way", is the third conjunct, with co-monotonicity as defined in footnote 5. The atomless hypothesis (mission II's splitting form) is added because a uniformly distributed exists on only when the space carries no atoms; on a finite space the family is empty and the source's maximum would not exist. The source states the corollary in the setting of Section 3.3, which assumes such a space. Uniformity is on ; the integrand is integrable for every uniform since and is bounded.
import Definitions.Def_MultistageStochastic_RiskFunctional import Definitions.Def_MultistageStochastic_Distortion open MeasureTheory open scoped ENNReal
namespace MultistageStochastic
theorem distortion_comonotone_max {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω)
[IsProbabilityMeasure P] (hatom : Atomless P)
(σ : ℝ → ℝ) (hσ : IsDistortionFunction σ) (Y : Ω → ℝ) (hY : MemLinfty P Y) :
(∀ U : Ω → ℝ, IsUniform P U →
∫ ω, Y ω * σ (U ω) ∂P ≤ distortionFunctional P σ Y) ∧
(∃ U : Ω → ℝ, IsUniform P U ∧
∫ ω, Y ω * σ (U ω) ∂P = distortionFunctional P σ Y) ∧
∀ U : Ω → ℝ, IsUniform P U →
(∀ y z : ℝ, P {ω | Y ω ≤ y ∧ U ω ≤ z} = min (P {ω | Y ω ≤ y}) (P {ω | U ω ≤ z})) →
∫ ω, Y ω * σ (U ω) ∂P = distortionFunctional P σ Y := by sorry
end MultistageStochastic
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.