Lemma 3 — the non-anticipative slack constraint (4.5b) is equivalent to its full-information form (4.5c)
ProvedPrimalDualLDR.Multistage.lemma_3linear-decision-rulesmultistagep2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-programming
Consider the multistage setting of §4 under its standing assumptions and the linear-conditional-mean assumption -a.s., with . Fix a stage and a matrix . Then the constraint
is equivalent to
In (4.5b) the slack is non-anticipative (a function of the history ), while in (4.5c) it may depend on the whole of . The equivalence makes the one-stage cone description of Proposition 3 applicable to the slack constraints of .
Formalization Note The page prints the sign condition of (4.5c) as "" for a function of the full ; it is read as . Moment matrices are entrywise Bochner integrals.
Preamble
import Mathlib import Definitions.Def_PrimalDualLDR_Multistage_Basic import Definitions.Def_PrimalDualLDR_Multistage_Setting import Definitions.Def_PrimalDualLDR_Multistage_Problems open MeasureTheory Matrix
Formal statement
namespace PrimalDualLDR.Multistage
/-- Lemma 3 (p. 21): under the standing assumptions of §4 and the linear-conditional-mean assumption,
for any given `S_t ∈ ℝ^{m_t × k^t}`, constraint (4.5b) — there is a non-anticipative
`s_t ∈ 𝓛²_{k^t,m_t}` with `𝔼(s_t(ξ^t) ξᵀ) = S_t P_t M` and `s_t(ξ^t) ≥ 0` `P`-a.s. — is equivalent to
(4.5c) — there is `s̃_t ∈ 𝓛²_{k,m_t}`, a function of the whole `ξ`, with `𝔼(s̃_t(ξ) ξᵀ) = S_t P_t M` and
`s̃_t(ξ) ≥ 0` `P`-a.s. -/
theorem lemma_3 (σ : Setting) (hσ : σ.Standing) (hM : σ.LinearCondMean) (t : Fin σ.T)
(S : Matrix (Fin (σ.m t)) (Fin (σ.kb t)) ℝ) :
(∃ s : (Fin (σ.kb t) → ℝ) → (Fin (σ.m t) → ℝ), σ.IsL2RuleAt t s ∧
σ.moment t s = S * σ.Pt t * σ.M ∧ ∀ᵐ ξ ∂σ.P, ∀ i, 0 ≤ s (σ.tr t ξ) i) ↔
(∃ st : (Fin σ.k → ℝ) → (Fin (σ.m t) → ℝ), PrimalDualLDR.FixedRecourse.IsL2Rule σ.P st ∧
(fun i j => ∫ ξ, st ξ i * ξ j ∂σ.P) = S * σ.Pt t * σ.M ∧ ∀ᵐ ξ ∂σ.P, ∀ i, 0 ≤ st ξ i) := by sorry
end PrimalDualLDR.Multistage
Source
Kuhn, Wiesemann, Georghiou, Primal and dual linear decision rules in stochastic and robust optimization, Optimization Online 2009/02/2218, p. 21, Lemma 3, (4.5b), (4.5c)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.