Proof of Theorem 2.1, third to fifth equalities:
ProvedSoftRobust.Duality.scaling_identityconvex-risk-measuresp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1robust-optimization
Let be an admissible family of probability measures on (nondecreasing, convex, ), with penalty and generated risk measure . For every bounded measurable and every ,
Applied to , it rewrites the dual function of the previous step through , so that the maximum over becomes , the left side of (6).
Formalization Note The infimum on the left is taken in EReal (measures with contribute ); the statement asserts that it equals the real number on the right, so in particular it is finite.
Preamble
import Mathlib import Definitions.Def_SoftRobust_Duality_Mixture import Definitions.Def_SoftRobust_Duality_ConvexRiskMeasure import Definitions.Def_SoftRobust_Duality_Model open MeasureTheory open scoped NNReal ENNReal
Formal statement
namespace SoftRobust.Duality
/-- §2.3, proof of Theorem 2.1, p. 1224, third to fifth equalities ("simple algebra and the
definition of ρ^Q"): for every bounded measurable `Y` and `λ ≥ 0`,
`inf_{ℚ ∈ 𝒫} {E_ℚ[Y] + (λ + 1) α^Q(ℚ)} = −(λ + 1) ρ^Q(Y / (λ + 1))`, the infimum taken in `EReal`. -/
theorem scaling_identity {Ω : Type*} [MeasurableSpace Ω]
(𝒬 : ℝ≥0 → Set (ProbabilityMeasure Ω)) (h𝒬 : IsAdmissibleFamily 𝒬)
(Y : Ω → ℝ) (hY : IsBddPayoff Y) (l : ℝ) (hl : 0 ≤ l) :
(⨅ Q : ProbabilityMeasure Ω,
((∫ ω, Y ω ∂(Q : Measure Ω) : ℝ) : EReal) +
((l + 1 : ℝ) : EReal) * (penaltyOfFamily 𝒬 Q : EReal)) =
((-((l + 1) * rhoQ 𝒬 (fun ω => Y ω / (l + 1))) : ℝ) : EReal) := by sorry
end SoftRobust.Duality
Source
Ben-Tal, Bertsimas, Brown, A Soft Robust Model for Optimization Under Ambiguity, Oper. Res. 58(4) (2010), p. 1224, §2.3, proof of Theorem 2.1, third to fifth equalities ("simple algebra and the definition of ρ^Q")
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.