Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2.1, p. 583 — with random objective, the risk-averse minimax problem (2) equals the semidefinite program (7)

Proved
MinimaxSLP.ObjSDP.theorem_2_1

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

distributionally-robust-optimizationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1semidefinite-programmingstochastic-programming

Consider the two-stage minimax problem (2) with random objective q~∈Rd\tilde q\in\mathbb R^dq~​∈Rd and constant right-hand side hhh:

Z=min⁡x∈X(c′x+sup⁡P∈PEP[U(Q(q~,x))]),Z=\min_{x\in X}\Big(c'x+\sup_{P\in\mathcal P}\mathbb E_P\big[\mathbb U(\mathcal Q(\tilde q,x))\big]\Big),Z=x∈Xmin​(c′x+P∈Psup​EP​[U(Q(q~​,x))]),

where X={x:Ax=b, x≥0}X=\{x : Ax=b,\ x\ge 0\}X={x:Ax=b, x≥0}, Q(q,x)=min⁡{q′w:Ww=h−Tx, w≥0}\mathcal Q(q,x)=\min\{q'w : Ww=h-Tx,\ w\ge 0\}Q(q,x)=min{q′w:Ww=h−Tx, w≥0}, U(t)=max⁡k=1,…,K(αkt+βk)\mathbb U(t)=\max_{k=1,\dots,K}(\alpha_k t+\beta_k)U(t)=maxk=1,…,K​(αk​t+βk​) with αk≥0\alpha_k\ge 0αk​≥0, and P\mathcal PP is the class of distributions with mean μ\muμ and second-moment matrix EP[q~q~′]=Q\mathbb E_P[\tilde q\tilde q']=QEP​[q~​q~​′]=Q. Assume:

  1. XXX is nonempty and bounded (Assumption 1);
  2. X(x)={w:Ww=h−Tx, w≥0}≠∅X(x)=\{w : Ww=h-Tx,\ w\ge 0\}\neq\emptysetX(x)={w:Ww=h−Tx, w≥0}=∅ for every x∈Xx\in Xx∈X;
  3. {π∈Rr:W′π≤q}≠∅\{\pi\in\mathbb R^r : W'\pi\le q\}\neq\emptyset{π∈Rr:W′π≤q}=∅ for every q∈Rdq\in\mathbb R^dq∈Rd (Assumption 3);
  4. QQQ is symmetric and Q−μμ′≻0Q-\mu\mu'\succ 0Q−μμ′≻0 (Assumption 4).

Then ZZZ equals the optimal value of the semidefinite program

ZSDP=min⁡x,Y,y,y0,wk c′x+Q⋅Y+μ′y+y0s.t.(Y12(y−αkwk)12(y−αkwk)′y0−βk)⪰0,  Wwk+Tx=h,  wk≥0  (k=1,…,K),Ax=b, x≥0,(7)Z_{\mathrm{SDP}}=\min_{x,Y,y,y_0,w_k}\ c'x+Q\cdot Y+\mu'y+y_0\quad\text{s.t.}\quad \begin{pmatrix}Y & \tfrac12(y-\alpha_k w_k)\\ \tfrac12(y-\alpha_k w_k)' & y_0-\beta_k\end{pmatrix}\succeq 0,\ \ Ww_k+Tx=h,\ \ w_k\ge 0\ \ (k=1,\dots,K),\quad Ax=b,\ x\ge 0, \tag{7}ZSDP​=x,Y,y,y0​,wk​min​ c′x+Q⋅Y+μ′y+y0​s.t.(Y21​(y−αk​wk​)′​21​(y−αk​wk​)y0​−βk​​)⪰0,  Wwk​+Tx=h,  wk​≥0  (k=1,…,K),Ax=b, x≥0,(7)

both values being finite (the objective values of (2) are bounded below with greatest lower bound ZZZ, those of (7) with greatest lower bound ZSDPZ_{\mathrm{SDP}}ZSDP​), and Z=ZSDPZ=Z_{\mathrm{SDP}}Z=ZSDP​. Moreover, for every x∈Xx\in Xx∈X the worst-case expected disutility Z(x)Z(x)Z(x) equals the optimal value of the semidefinite program (8) obtained from (7) by fixing xxx.

The theorem reduces a minimax stochastic linear program over an infinite-dimensional set of distributions to a single semidefinite program whose size is polynomial in nnn, ddd, rrr and KKK.

Formalization Note Assumption 2 (complete recourse) and Assumption 3 for all qqq, as printed on p. 582, cannot both hold when r≥1r\ge 1r≥1; the proof of Theorem 2.1 uses Assumption 3 for all qqq and only the nonemptiness of X(x)X(x)X(x) for x∈Xx\in Xx∈X, which replaces Assumption 2 here. The printed minima are read as infima; attainment is not claimed. Distributions are measures with finite second moments rather than densities. The per-xxx identity Z(x)=Z(8)(x)Z(x)=Z_{(8)}(x)Z(x)=Z(8)​(x) is the content of the last line of the paper's proof ("ZD(x)=Z(x)Z_D(x)=Z(x)ZD​(x)=Z(x) for all x∈Xx\in Xx∈X").

Preamble
import Mathlib
import Definitions.Def_MomentDRO_Conf_Setting
import Definitions.Def_MinimaxSLP_ObjSDP_Model

open MeasureTheory Matrix
Formal statement
namespace MinimaxSLP.ObjSDP

/-- Theorem 2.1 (p. 583): under Assumption 1, relatively complete recourse on `X` (in place of
Assumption 2, see the mission notes), Assumption 3 for all `q` and Assumption 4, the risk-averse
minimax problem (2) with random objective `q̃` and constant right-hand side `h` is equivalent to
the semidefinite program (7): both optimal values are genuine infima and `Z = Z_SDP`; moreover,
for every `x ∈ X`, `Z(x)` equals the optimal value of (8). -/
theorem theorem_2_1 {m₁ n r d K : ℕ} [NeZero K]
    (A : Matrix (Fin m₁) (Fin n) ℝ) (b : Fin m₁ → ℝ) (c : Fin n → ℝ)
    (W : Matrix (Fin r) (Fin d) ℝ) (T : Matrix (Fin r) (Fin n) ℝ) (h : Fin r → ℝ)
    (α β : Fin K → ℝ) (hα : ∀ k, 0 ≤ α k)
    (μ : Fin d → ℝ) (Q : Matrix (Fin d) (Fin d) ℝ)
    (hA1 : (X A b).Nonempty ∧ Bornology.IsBounded (X A b))
    (hrec : ∀ x ∈ X A b, (recourseSet W T h x).Nonempty)
    (hA3 : ∀ q : Fin d → ℝ, ∃ π : Fin r → ℝ, Wᵀ *ᵥ π ≤ q)
    (hQ : Q.IsSymm) (hA4 : (Q - Matrix.vecMulVec μ μ).PosDef) :
    IsGLB (firstStageValues A b c W T h α β μ Q) (Z A b c W T h α β μ Q) ∧
    IsGLB (values7 A b c W T h α β μ Q) (ZSDP A b c W T h α β μ Q) ∧
    Z A b c W T h α β μ Q = ZSDP A b c W T h α β μ Q ∧
    ∀ x ∈ X A b, Zx W T h α β μ Q x = ZD8 W T h α β μ Q x := by sorry

end MinimaxSLP.ObjSDP
Source
Bertsimas, Doan, Natarajan & Teo, Models for Minimax Stochastic Linear Optimization Problems with Risk Aversion, Math. Oper. Res. 35(3), 2010, p. 583, Theorem 2.1 (proof pp. 583–584; Assumptions 1–4, p. 582)
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