Proof of Theorem 2 — is convex in and affine in
ProvedWorstCaseCVaR.Discrete.convex_affine_Gconvexitycvarp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let , the scenarios , the decision and the function be as in the setting of §2.2, with confidence level . Then:
- for every probability vector (that is, and ), the function is convex on ;
- for every , the function is an affine function on :
These are the two structural properties under which the minimax lemma (Lemma 1) applies to on a product of an interval of thresholds and the ambiguity set of distributions ; affinity in gives in particular the concavity in that the lemma needs.
Formalization Note Convexity in is stated as ConvexOn ℝ Set.univ for each in stdSimplex ℝ (Fin S); only and matter for it. Affinity in is stated as the existence of an affine map (Fin S → ℝ) →ᵃ[ℝ] ℝ that agrees with everywhere.
Preamble
import Mathlib import Definitions.Def_WorstCaseCVaR_Discrete_Setting
Formal statement
namespace WorstCaseCVaR.Discrete
/-- Proof of Theorem 2, Zhu & Fukushima (2009), p. 1167: for fixed `x`, `G_β(x, α, π)` is convex
in `α` (for each probability vector `π`) and affine in `π` (for each `α`). -/
theorem convex_affine_G {n m S : ℕ} (f : (Fin n → ℝ) → (Fin m → ℝ) → ℝ)
(ys : Fin S → Fin m → ℝ) (x : Fin n → ℝ) (β : ℝ) (hβ0 : 0 < β) (hβ1 : β < 1) :
(∀ π ∈ stdSimplex ℝ (Fin S), ConvexOn ℝ Set.univ (fun α : ℝ => G f ys β x α π)) ∧
(∀ α : ℝ, ∃ g : (Fin S → ℝ) →ᵃ[ℝ] ℝ, ∀ π : Fin S → ℝ, G f ys β x α π = g π) := by sorry
end WorstCaseCVaR.Discrete
Source
Zhu & Fukushima, Worst-Case Conditional Value-at-Risk with Application to Robust Portfolio Management, Oper. Res. 57(5), 2009, p. 1167, proof of Theorem 2 (convexity in α cited from Rockafellar and Uryasev 2002)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.