Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.3, p. 590 — with random right-hand side and known dual extreme points, a distribution in 𝒫 attains Ẑ(x) = Ẑ_DD(x)

Proved
MinimaxSLP.RhsExtremal.theorem_3_3

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

distributionally-robust-optimizationextremal-distributionmoment-problemp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-programming

Consider the two-stage minimax stochastic linear program with random right-hand side h~∈Rr\tilde h\in\mathbb R^rh~∈Rr and constant objective q∈Rdq\in\mathbb R^dq∈Rd, with second-stage cost Q(h,x)=max⁡{(h−Tx)′p:W′p≤q}\mathcal Q(h,x)=\max\{(h-Tx)'p : W'p\le q\}Q(h,x)=max{(h−Tx)′p:W′p≤q}, disutility U(t)=max⁡k(αkt+βk)\mathbb U(t)=\max_{k}(\alpha_k t+\beta_k)U(t)=maxk​(αk​t+βk​) with αk≥0\alpha_k\ge0αk​≥0, and moment class P={P:EP[h~]=μ, EP[h~h~′]=Q}\mathcal P=\{P : \mathbb E_P[\tilde h]=\mu,\ \mathbb E_P[\tilde h\tilde h']=Q\}P={P:EP​[h~]=μ, EP​[h~h~′]=Q}. Assume

  1. complete recourse: {Ww:w≥0}=Rr\{Ww : w\ge 0\}=\mathbb R^r{Ww:w≥0}=Rr (Assumption 2);
  2. {p∈Rr:W′p≤q}≠∅\{p\in\mathbb R^r : W'p\le q\}\neq\emptyset{p∈Rr:W′p≤q}=∅ (Assumption 3 at the constant qqq);
  3. QQQ is symmetric and Q≻μμ′Q\succ\mu\mu'Q≻μμ′ (Assumption 4);
  4. p1,…,pNp_1,\dots,p_Np1​,…,pN​ are the extreme points of {p:W′p≤q}\{p : W'p\le q\}{p:W′p≤q} (Assumption 5).

Theorem 3.3. For an arbitrary x∈Xx\in Xx∈X, there exists an extremal distribution in P\mathcal PP that achieves the optimal value Z^(x)\hat Z(x)Z^(x). Precisely, there is P∈PP\in\mathcal PP∈P such that U(Q(⋅,x))\mathbb U(\mathcal Q(\cdot,x))U(Q(⋅,x)) is integrable under every P′∈PP'\in\mathcal PP′∈P and

EP′[U(Q(h~,x))]≤EP[U(Q(h~,x))]∀P′∈P,\mathbb E_{P'}\big[\mathbb U(\mathcal Q(\tilde h,x))\big]\le\mathbb E_{P}\big[\mathbb U(\mathcal Q(\tilde h,x))\big]\quad\forall P'\in\mathcal P,EP′​[U(Q(h~,x))]≤EP​[U(Q(h~,x))]∀P′∈P,

the value EP[U(Q(h~,x))]\mathbb E_{P}[\mathbb U(\mathcal Q(\tilde h,x))]EP​[U(Q(h~,x))] is the attained maximum of (16), and

EP[U(Q(h~,x))]=Z^(x)=Z^DD(x).\mathbb E_{P}\big[\mathbb U(\mathcal Q(\tilde h,x))\big]=\hat Z(x)=\hat Z_{DD}(x).EP​[U(Q(h~,x))]=Z^(x)=Z^DD​(x).

The worst-case distribution is therefore a genuine member of the moment class, not only a limit, and it can be written down from an optimal solution of the semidefinite program (16) as a finite mixture of normal distributions.

Formalization Note As printed, Assumption 2 and Assumption 3 "for all qqq" contradict each other when r≥1r\ge1r≥1; the §3 reading uses complete recourse and dual feasibility at the one constant qqq. Assumption 1 (XXX bounded and nonempty) is not used and is omitted; x∈Xx\in Xx∈X is kept as printed. No strong-duality hypothesis is made: the proof does not need one. The maximality over P\mathcal PP is stated directly, so the claim does not depend on junk values of sSup.

Preamble
import Mathlib
import Definitions.Def_MomentDRO_Conf_Setting
import Definitions.Def_MinimaxSLP_RhsExtremal_Model
Formal statement
namespace MinimaxSLP.RhsExtremal

open MeasureTheory Matrix

/-- Theorem 3.3, p. 590. Under Assumptions 2 (complete recourse), 3 (at the constant `q`),
4 and 5, and `α_k ≥ 0`, for an arbitrary `x ∈ MinimaxSLP.ObjSDP.X` there is a distribution `P ∈ 𝒫` that attains
the optimal value `Ẑ(x)`: `𝕌(𝒬(·, x))` is integrable under every `P′ ∈ 𝒫` and
`E_{P′}[𝕌(𝒬(h̃, x))] ≤ E_P[𝕌(𝒬(h̃, x))]`; moreover `E_P[𝕌(𝒬(h̃, x))]` is the attained maximum
of (16), and `E_P[𝕌(𝒬(h̃, x))] = Ẑ(x) = Ẑ_DD(x)`. -/
theorem theorem_3_3 {m₁ n r d K N : ℕ} [NeZero K]
    (A : Matrix (Fin m₁) (Fin n) ℝ) (b : Fin m₁ → ℝ)
    (W : Matrix (Fin r) (Fin d) ℝ) (T : Matrix (Fin r) (Fin n) ℝ) (q : Fin d → ℝ)
    (α β : Fin K → ℝ) (μ : Fin r → ℝ) (Q : Matrix (Fin r) (Fin r) ℝ)
    (ps : Fin N → Fin r → ℝ)
    (hα : ∀ k, 0 ≤ α k)
    (hA2 : ∀ z : Fin r → ℝ, ∃ w : Fin d → ℝ, 0 ≤ w ∧ W *ᵥ w = z)
    (hA3 : ∃ π : Fin r → ℝ, Wᵀ *ᵥ π ≤ q)
    (hQ : Q.IsSymm) (hA4 : (Q - Matrix.vecMulVec μ μ).PosDef)
    (hA5 : Set.range ps = Set.extremePoints ℝ {π : Fin r → ℝ | Wᵀ *ᵥ π ≤ q})
    (x : Fin n → ℝ) (hx : x ∈ MinimaxSLP.ObjSDP.X A b) :
    ∃ P ∈ MinimaxSLP.ObjSDP.momentClass μ Q,
      (∀ P' ∈ MinimaxSLP.ObjSDP.momentClass μ Q,
        Integrable (fun h => MinimaxSLP.ObjSDP.disutility α β (MinimaxSLP.RhsSDP.Qval W T q h x)) P' ∧
        ∫ h, MinimaxSLP.ObjSDP.disutility α β (MinimaxSLP.RhsSDP.Qval W T q h x) ∂P' ≤ ∫ h, MinimaxSLP.ObjSDP.disutility α β (MinimaxSLP.RhsSDP.Qval W T q h x) ∂P) ∧
      IsGreatest (obj16 T α β ps x '' feasible16 K N μ Q)
        (∫ h, MinimaxSLP.ObjSDP.disutility α β (MinimaxSLP.RhsSDP.Qval W T q h x) ∂P) ∧
      MinimaxSLP.RhsSDP.Zhat W T q α β μ Q x = ∫ h, MinimaxSLP.ObjSDP.disutility α β (MinimaxSLP.RhsSDP.Qval W T q h x) ∂P ∧
      ZhatDD T α β ps μ Q x = ∫ h, MinimaxSLP.ObjSDP.disutility α β (MinimaxSLP.RhsSDP.Qval W T q h x) ∂P := by sorry

end MinimaxSLP.RhsExtremal
Source
Bertsimas, Doan, Natarajan & Teo, Models for Minimax Stochastic Linear Optimization Problems with Risk Aversion, Math. Oper. Res. 35(3), 2010, p. 590, Theorem 3.3 (proof pp. 590–591)
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