Theorem 2 — with C₂ bounded, j = J, inf Q = inf P, the same first-stage minimizers, and the infimum over ℒ^∞ recourses is attained
ProvedStochConvexProg.FirstStage.theorem_2first-stage-problemp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1recoursestochastic-programming
Consider the two-stage stochastic convex program under its standing assumptions, and suppose the second-stage constraint set is bounded. Then the conclusions of Theorem 1 hold:
- for every (3.8);
- ;
- gives the minimum in the intrinsic first-stage problem if and only if it gives the minimum in the first-stage problem induced by ;
and moreover
- for each the infimum in (3.8) is attained: there is with
Thus when is bounded, restricting recourse to essentially bounded measurable functions loses no generality, and an optimal recourse function exists for every first-stage decision.
Formalization Note "Gives the minimum" is " the value at every other point". is the EReal infimum over all of (Lp (Fin n₂ → ℝ) ⊤ σ).
Preamble
import Mathlib import Definitions.Def_StochConvexProg_FirstStage_Problem import Definitions.Def_DupacovaWets_Consistency_expect import Definitions.Def_StochConvexProg_FirstStage_FirstStage open MeasureTheory
Formal statement
namespace StochConvexProg.FirstStage
/-- Theorem 2, p. 187: if `C₂` is bounded, the conclusions of Theorem 1 hold (`j = J`, `inf Q = inf P`,
the same first-stage minimizers), and the infimum in (3.8) is attained for each `x₁` by some
`x₂ ∈ ℒ^∞_{n₂}`. -/
theorem theorem_2 {S : Type*} [MeasurableSpace S] {σ : Measure S} [IsProbabilityMeasure σ]
{n₁ n₂ m₁ m₂ : ℕ} (pr : Problem S σ n₁ n₂ m₁ m₂) (hC₂ : Bornology.IsBounded pr.C₂) :
(∀ x₁, pr.j x₁ = pr.J x₁) ∧ pr.infQ = pr.infP ∧
(∀ x₁, (∀ x₁', pr.j x₁ ≤ pr.j x₁') ↔ (∀ x₁', pr.J x₁ ≤ pr.J x₁')) ∧
(∀ x₁, ∃ x₂ : Lp (Fin n₂ → ℝ) ⊤ σ, pr.f (x₁, x₂) = pr.J x₁) := by sorry
end StochConvexProg.FirstStage
Source
Rockafellar and Wets, Stochastic convex programming: basic duality, Pacific J. Math. 62(1) (1976), p. 187, Theorem 2 (with Theorem 1, p. 186)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.