Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 1 — with fixed recourse and strict feasibility, SPu\mathcal{SP}^uSPu and SPl\mathcal{SP}^lSPl equal the LPs (2.3) and (2.8)

Proved
PrimalDualLDR.FixedRecourse.theorem_1

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

linear-decision-rulesp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1robust-optimizationstochastic-programming

Consider the one-stage stochastic program with fixed recourse

SP:min⁡x∈Lk,n2E(c(ξ)⊤x(ξ))s.t.Ax(ξ)≤b(ξ)  P-a.s.,\mathcal{SP}:\qquad \min_{x\in\mathcal L^2_{k,n}} \mathbb E\big(c(\xi)^\top x(\xi)\big)\quad\text{s.t.}\quad Ax(\xi)\le b(\xi)\ \ \mathbb P\text{-a.s.},SP:x∈Lk,n2​min​E(c(ξ)⊤x(ξ))s.t.Ax(ξ)≤b(ξ)  P-a.s.,

with c(ξ)=Cξc(\xi) = C\xic(ξ)=Cξ, b(ξ)=Bξb(\xi) = B\xib(ξ)=Bξ, and assume that P\mathbb PP has a polyhedral support Ξ={ξ:Wξ≥h}\Xi = \{\xi : W\xi \ge h\}Ξ={ξ:Wξ≥h} of the type (2.1): nonempty, bounded, with W,hW, hW,h of the form (2.1b), spanning Rk\mathbb R^kRk. Assume further that SP\mathcal{SP}SP is strictly feasible (2.9). Let M=E(ξξ⊤)M = \mathbb E(\xi\xi^\top)M=E(ξξ⊤). Then

  1. the primal linear-decision-rule problem SPu\mathcal{SP}^uSPu (min⁡Tr⁡(MC⊤X)\min \operatorname{Tr}(MC^\top X)minTr(MC⊤X) s.t. AXξ+Sξ=BξAX\xi+S\xi=B\xiAXξ+Sξ=Bξ, Sξ≥0S\xi\ge0Sξ≥0 P\mathbb PP-a.s.) has the same optimal value as the linear program (2.3),
min⁡X,Λ Tr⁡(MC⊤X)  s.t.  AX+ΛW=B, Λh≥0, Λ≥0;\min_{X,\Lambda}\ \operatorname{Tr}(MC^\top X)\ \ \text{s.t.}\ \ AX+\Lambda W=B,\ \Lambda h\ge0,\ \Lambda\ge0;X,Λmin​ Tr(MC⊤X)  s.t.  AX+ΛW=B, Λh≥0, Λ≥0;
  1. the dual linear-decision-rule problem SPl\mathcal{SP}^lSPl (min⁡E(c(ξ)⊤x(ξ))\min \mathbb E(c(\xi)^\top x(\xi))minE(c(ξ)⊤x(ξ)) over x∈Lk,n2x\in\mathcal L^2_{k,n}x∈Lk,n2​, s∈Lk,m2s\in\mathcal L^2_{k,m}s∈Lk,m2​ s.t. E([Ax(ξ)+s(ξ)−b(ξ)]ξ⊤)=0\mathbb E([Ax(\xi)+s(\xi)-b(\xi)]\xi^\top)=0E([Ax(ξ)+s(ξ)−b(ξ)]ξ⊤)=0, s(ξ)≥0s(\xi)\ge0s(ξ)≥0 P\mathbb PP-a.s.) has the same optimal value as the linear program (2.8),
min⁡X,S Tr⁡(MC⊤X)  s.t.  AX+S=B,  (W−he1⊤)MS⊤≥0.\min_{X,S}\ \operatorname{Tr}(MC^\top X)\ \ \text{s.t.}\ \ AX+S=B,\ \ (W-he_1^\top)MS^\top\ge0.X,Smin​ Tr(MC⊤X)  s.t.  AX+S=B,  (W−he1⊤​)MS⊤≥0.

The formal statement adds one hypothesis the paper does not state: W^≠0\widehat W \ne 0W=0, i.e. some row of WWW below the first two is nonzero (see the Formalization Note).

Since SPu\mathcal{SP}^uSPu is an upper and SPl\mathcal{SP}^lSPl a lower bound on SP\mathcal{SP}SP, the theorem brackets the optimal value of the (generally #P-hard) stochastic program between two explicit linear programs whose size is polynomial in k,l,m,nk, l, m, nk,l,m,n.

Formalization Note "Equivalent" is formalized as equality of optimal values, each the infimum of the objective over the feasible set in [−∞,+∞][-\infty,+\infty][−∞,+∞] (+∞+\infty+∞ if infeasible, −∞-\infty−∞ if unbounded). Strict feasibility is a hypothesis of both equalities, as printed, although the first holds without it. The paper's last sentence (the LP sizes are polynomial, so they are efficiently solvable) is informal and is not part of the statement. Repaired statement: the hypothesis W^≠0\widehat W \ne 0W=0 is added because the printed theorem is false without it: for k=1k = 1k=1, l=2l = 2l=2, W=(1,−1)⊤W = (1,-1)^\topW=(1,−1)⊤, h=(1,−1)⊤h = (1,-1)^\toph=(1,−1)⊤, P=δ1\mathbb P = \delta_1P=δ1​, n=m=1n = m = 1n=m=1, A=B=1A = B = 1A=B=1, C=−1C = -1C=−1, every assumption of the theorem holds, SPl\mathcal{SP}^lSPl has value −1-1−1, but W−he1⊤=0W - he_1^\top = 0W−he1⊤​=0, so (2.8) has no sign constraint on SSS and value −∞-\infty−∞. The paper's proof of Proposition 3 drops the constraint e1⊤z≥0e_1^\top z \ge 0e1⊤​z≥0 as redundant, which fails exactly when W^=0\widehat W = 0W=0. Under the standing assumptions W^≠0\widehat W \ne 0W=0 holds automatically when k≥2k \ge 2k≥2 (a bounded Ξ\XiΞ forces it), so the repair only excludes this degenerate one-point case.

Preamble
import Mathlib
import Definitions.Def_PrimalDualLDR_FixedRecourse_Basic
import Definitions.Def_PrimalDualLDR_FixedRecourse_Setting
import Definitions.Def_PrimalDualLDR_FixedRecourse_Problems

open MeasureTheory Matrix
Formal statement
namespace PrimalDualLDR.FixedRecourse

/-- Theorem 1 (p. 10): if `P` has a polyhedral support of the type (2.1) while `SP` has fixed recourse
and is strictly feasible, then `SP^u` and `SP^l` are equivalent to the linear programs (2.3) and
(2.8), respectively: their optimal values coincide.

Repair (trap 30): the hypothesis `hW` says `Ŵ ≠ 0` (some row `i ≥ 2` of `W` is nonzero; rows
`0`, `1` are `±e_1ᵀ`). The page omits it, and the printed statement is false without it: for
`k = 1`, `l = 2`, `W = (1, −1)ᵀ`, `h = (1, −1)`, `P = δ_1`, the matrix `W − h e_1ᵀ` is `0`, so
`𝒦 = ℝ` while `𝒦_P = [0, ∞)`; with `n = m = 1`, `A = B = 1`, `C = −1` (strictly feasible with
`x̄ = 0`, `s̄(ξ) = ξ`) the program (2.8) loses `S ≥ 0` and has value `−∞`, while (2.6) and `SP^l`
have value `−1`. Under the standing assumptions `Ŵ ≠ 0` is automatic when `k ≥ 2`. -/
theorem theorem_1 (σ : Setting) (hσ : σ.Standing)
    (hW : ∃ i : Fin σ.l, 2 ≤ i.val ∧ σ.W i ≠ 0) (hsf : σ.StrictlyFeasible) :
    σ.valSPu = σ.valLP23 ∧ σ.valSPl = σ.valLP28 := by sorry

end PrimalDualLDR.FixedRecourse
Source
Kuhn, Wiesemann, Georghiou, Primal and dual linear decision rules in stochastic and robust optimization, Optimization Online 2009/02/2218, p. 10, Theorem 1
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me