Theorem 4 — exact lifted linear decision rules with one recourse variable
DisprovedAdaptDRO.LiftedLDR.theorem_4_beta_eq_betaESuppose the second-stage problem has complete recourse and exactly one free second-stage decision variable. For the SOC ambiguity set of Definition 3 and its lifted ambiguity set, the worst-case expected optimal recourse cost equals the best worst-case expected cost among rules affine in the primary and auxiliary random variables. For every first-stage decision ,
Thus the extended affine decision-rule approximation is exact in this recourse regime, even though regular affine rules need not be feasible or exact.
Formalization Note The statement makes Assumption 1 and dual feasibility explicit. Theorem 4 does not repeat Assumption 1, but its route through Theorem 1 uses it; primal feasibility alone does not imply dual feasibility. The four model values use extended-real suprema and infima, while finite first moments in the ambiguity sets ensure that affine expectations are real.
import Mathlib import Definitions.Def_AdaptDRO_LiftedLDR_Model
namespace AdaptDRO.LiftedLDR
/-- Theorem 4 (p. 610): with one free recourse variable and complete
recourse, the lifted affine rule has the exact worst-case value. -/
theorem theorem_4_beta_eq_betaE
{I1 I2 I3 L1 L2 M N1 : ℕ} (m : Model I1 I2 L1 M N1 1)
(C : Matrix (Fin L2) (Fin I1) ℝ)
(D : Matrix (Fin L2) (Fin I2) ℝ)
(E : Matrix (Fin L2) (Fin I3) ℝ)
(h : Fin L2 → ℝ) (K : Set (Fin L2 → ℝ))
(hSOC : m.SOCData) (hAssumption1 : m.Assumption1 C D E h K)
(hrec : m.recourse.Complete) (hdual : m.recourse.DualFeasible) :
∀ x : Fin N1 → ℝ, m.beta x = m.betaE x := by sorry
end AdaptDRO.LiftedLDR
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.