§4, (4.7) — the equality constraints of 𝓜𝒮𝒫^l become those of (4.6)
ProvedPrimalDualLDR.Multistage.sec_4_eq_4_7linear-decision-rulesmultistagep2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-programming
Consider the multistage setting of §4 under its standing assumptions, with and . Fix a stage .
- Let () and be decision rules and , matrices related by (4.3): for every , and . Then the -th equality constraint of , , holds if and only if
- For all matrices , , (4.7) holds if and only if
the -th equality constraint of (4.6).
This is the step in the derivation of the linear program (4.6) where the multistage case differs from the one-stage case of §2.
Formalization Note The paper derives the forward direction of part 2 by multiplying (4.7) from the right by ; the stated equivalence also contains the converse (multiply by ), which the reformulation of as (4.6) needs. Invertibility of is not a hypothesis: it follows from the standing assumptions (" inherits invertibility from "). 's equality constraint is in the corrected reading in which covers only .
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
/-- §4, (4.7) (p. 21): the equality constraints of `𝓜𝒮𝒫^l` and of (4.6). Under the standing
assumptions of §4, for every stage `t`:
1. if the rules `x_s ∈ 𝓛²_{k^s,n_s}`, `s_t ∈ 𝓛²_{k^t,m_t}` and the matrices `X_s`, `S_t` are related by
(4.3), i.e. `X_s P_s M = 𝔼(x_s(ξ^s) ξᵀ)` for all `s` and `S_t P_t M = 𝔼(s_t(ξ^t) ξᵀ)`, then the `t`-th
equality constraint of `𝓜𝒮𝒫^l` holds iff (4.7):
`Σ_s A_ts X_s P_s M P_tᵀ + S_t P_t M P_tᵀ = B_t P_t M P_tᵀ`;
2. for all matrices `X_s`, `S_t`, (4.7) holds iff the equality constraint of (4.6):
`Σ_s A_ts X_s P_s N_t P_t + S_t P_t = B_t P_t`, where `N_t = M P_tᵀ (P_t M P_tᵀ)⁻¹`. -/
theorem sec_4_eq_4_7 (σ : Setting) (hσ : σ.Standing) (t : Fin σ.T) :
(∀ (x : σ.XRules) (s : (Fin (σ.kb t) → ℝ) → (Fin (σ.m t) → ℝ)) (X : σ.XFam)
(S : Matrix (Fin (σ.m t)) (Fin (σ.kb t)) ℝ),
(∀ r, σ.IsL2RuleAt r (x r)) → σ.IsL2RuleAt t s →
(∀ r, X r * σ.Pt r * σ.M = σ.moment r (x r)) → S * σ.Pt t * σ.M = σ.moment t s →
(σ.eqMSPl t x s ↔
∑ r, σ.A t r * X r * σ.Pt r * σ.M * (σ.Pt t)ᵀ + S * σ.Pt t * σ.M * (σ.Pt t)ᵀ =
σ.B t * σ.Pt t * σ.M * (σ.Pt t)ᵀ)) ∧
(∀ (X : σ.XFam) (S : Matrix (Fin (σ.m t)) (Fin (σ.kb t)) ℝ),
(∑ r, σ.A t r * X r * σ.Pt r * σ.M * (σ.Pt t)ᵀ + S * σ.Pt t * σ.M * (σ.Pt t)ᵀ =
σ.B t * σ.Pt t * σ.M * (σ.Pt t)ᵀ) ↔
∑ r, σ.A t r * X r * σ.Pt r * σ.N t * σ.Pt t + S * σ.Pt t = σ.B t * σ.Pt t) := 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, §4, (4.3), (4.6), (4.7)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.