§4.3, proof of Theorem 3, p. 25, first paragraph — ζ ↦ L(ξ, ζ) is convex
ProvedPDScenRed.Mean.lossL_convexconvexityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1scenario-reduction
Let be a nonempty bounded polytope, let be any map, and fix . Then the function
is convex on .
Convexity in the reduced scenario is what makes the upper-bound problem (16) a convex programme, and it is the reason a zero subgradient certifies optimality in the proof of Theorem 3.
Formalization Note No property of is needed: for fixed , is a constant vector. The paper's is nonempty and compact (§1); the polytope is closed, so boundedness is the stated hypothesis.
Preamble
import Mathlib import Definitions.Def_PDScenRed_Mean_Setting open scoped InnerProductSpace open MeasureTheory
Formal statement
namespace PDScenRed.Mean
theorem lossL_convex {d r : ℕ} (P : Matrix (Fin r) (Fin d) ℝ) (qv : Fin r → ℝ)
(hne : (polytope P qv).Nonempty) (hbdd : Bornology.IsBounded (polytope P qv))
(zsel : EuclideanSpace ℝ (Fin d) → EuclideanSpace ℝ (Fin d))
(ξ : EuclideanSpace ℝ (Fin d)) :
ConvexOn ℝ Set.univ (fun ζ => lossL (polytope P qv) zsel ξ ζ) := by sorry
end PDScenRed.Mean
Source
Bertsimas & Mundru, Optimization-based Scenario Reduction for Data-Driven Two-stage Stochastic Optimization, author manuscript (MIT DSpace), p. 25, §4.3, proof of Theorem 3, first paragraph, last sentence
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.