§4.3, proof of Theorem 3, p. 25, display — L(ξ, ζ) ≤ (‖ξ‖₁ + 4‖ζ‖₁) max_{z∈𝒵} ‖z‖_∞
ProvedPDScenRed.Mean.lossL_leintegrabilityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1scenario-reduction
Let be a nonempty bounded polytope and let be a selection of minimizers, for every . Let be a bound on the coordinates of the feasible decisions, for every and every (for instance ). Then for all ,
where .
This is the bound the proof uses to see that is finite whenever is finite, and that differentiation can be moved inside the expectation.
Formalization Note The three inequalities of the display are stated as three conjuncts. The paper's is replaced by any upper bound of it, which is equivalent: the case is included, and every larger makes the right-hand sides larger.
Preamble
import Mathlib import Definitions.Def_PDScenRed_Mean_Setting open scoped InnerProductSpace open MeasureTheory
Formal statement
namespace PDScenRed.Mean
theorem lossL_le {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))
(hsel : ∀ ξ, zsel ξ ∈ SmartPTO.Fisher.Wstar (polytope P qv) ξ)
(B : ℝ) (hB : ∀ z ∈ polytope P qv, ∀ i, |z i| ≤ B) :
∀ ξ ζ : EuclideanSpace ℝ (Fin d),
lossL (polytope P qv) zsel ξ ζ ≤
|SmartPTO.Fisher.xi (polytope P qv) (ξ - (2 : ℝ) • ζ)| + 2 * |⟪ζ, zsel ξ⟫_ℝ| ∧
|SmartPTO.Fisher.xi (polytope P qv) (ξ - (2 : ℝ) • ζ)| + 2 * |⟪ζ, zsel ξ⟫_ℝ| ≤
(∑ i, |(ξ - (2 : ℝ) • ζ) i| + 2 * ∑ i, |ζ i|) * B ∧
(∑ i, |(ξ - (2 : ℝ) • ζ) i| + 2 * ∑ i, |ζ i|) * B ≤
(∑ i, |ξ i| + 4 * ∑ i, |ζ i|) * B := 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, second paragraph, displayed chain
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.