Theorem 1 — under mixture uncertainty,
ProvedWorstCaseCVaR.Mixture.theorem_1cvarmixture-distributionp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1robust-optimization
Let () be probability distributions on (the likelihood distributions), let , and let be a decision for which the loss is integrable under every . Then the worst-case CVaR of over the set of all mixtures , , is
where , and the minimum over is attained.
The theorem replaces a supremum over infinitely many distributions by a pointwise maximum of convex functions, so worst-case CVaR minimization over mixtures is a single convex program in .
Formalization Note The minimum is stated with an explicit minimizer of . Distributions are probability measures (Remark 1 of the paper); indices are zero-based (Fin l); is the standing reading of "confidence level".
Preamble
import Mathlib import Definitions.Def_WorstCaseCVaR_Mixture_Setting open MeasureTheory
Formal statement
namespace WorstCaseCVaR.Mixture
theorem theorem_1 {l m n : ℕ} [NeZero l] (P : Fin l → Measure (Fin m → ℝ))
[∀ i, IsProbabilityMeasure (P i)] (f : (Fin n → ℝ) → (Fin m → ℝ) → ℝ)
(β : ℝ) (hβ0 : 0 < β) (hβ1 : β < 1) (x : Fin n → ℝ)
(hf : ∀ i, Integrable (f x) (P i)) :
∃ α₀ : ℝ, IsMinOn (FL P f β x) Set.univ α₀ ∧ wcvar P f β x = FL P f β x α₀ := by sorry
end WorstCaseCVaR.Mixture
Source
Zhu & Fukushima, Worst-Case Conditional Value-at-Risk with Application to Robust Portfolio Management, Oper. Res. 57(5), 2009, p. 1157, Theorem 1, Eq. (6)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.