Theorem 3.22 — R_σ(Y) is the infimum of E(h(Y)) over h with ∫ h*(σ(u)) du ≤ 0
OpenMultistageStochastic.distortion_inf_representationLet be a distortion function and on a probability space. Then
the infimum over measurable , where is the conjugate of .
This is Theorem 3.22, the "alternative description" of Section 3.4: it expresses a distortion functional as an infimum instead of the supremum of (3.15), so that a stochastic program with a risk-functional objective becomes a joint minimisation over the decision and over , as (3.3) does for the Average Value-at-Risk with . For that special case the infimum over collapses to an infimum over the scalar .
Formalization Note An admissible is measurable, has finite for almost every level, integrable on with integral at most , and integrable; these are the conditions under which the two expectations in (3.24) are the integrals the source intends, and without them Lean's integral of a non-integrable function is , which would let a meaningless into the constraint set. The conjugate is valued in the extended reals, so a supremum equal to is recorded as such rather than replaced by a junk real. The infimum is over the subtype of admissible ; the source's identity presupposes that this set is nonempty, and the proof exhibits admissible with close to .
import Definitions.Def_MultistageStochastic_RiskFunctional import Definitions.Def_MultistageStochastic_Distortion open MeasureTheory open scoped ENNReal
namespace MultistageStochastic
theorem distortion_inf_representation {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω)
[IsProbabilityMeasure P] (σ : ℝ → ℝ) (hσ : IsDistortionFunction σ)
(Y : Ω → ℝ) (hY : MemLinfty P Y) :
distortionFunctional P σ Y
= ⨅ h : {h : ℝ → ℝ // IsAdmissibleConjugate σ h ∧ Integrable (fun ω => h (Y ω)) P},
∫ ω, (h : ℝ → ℝ) (Y ω) ∂P := by sorry
end MultistageStochastic
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.