Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 2 — under box uncertainty, worst-case CVaR minimization (16)–(20) is equivalent to the LP (24)–(30)

Proved
WorstCaseCVaR.Box.proposition_2

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

cvarlinear-programmingp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1portfolio-optimizationrobust-optimization

Let f:Rn×Rm→Rf : \mathbb R^n \times \mathbb R^m \to \mathbb Rf:Rn×Rm→R be a loss function, y[1],…,y[S]∈Rmy_{[1]}, \dots, y_{[S]} \in \mathbb R^my[1]​,…,y[S]​∈Rm the scenarios of the random vector, X⊆Rn\mathcal X \subseteq \mathbb R^nX⊆Rn the decision set, 0<β<10 < \beta < 10<β<1 the confidence level, and let the distribution π\piπ range over the box

PπB={π:π=π0+η, e⊤η=0, η‾≤η≤η‾}.\mathcal P_\pi^B = \{\pi : \pi = \pi^0 + \eta,\ e^\top\eta = 0,\ \underline\eta \le \eta \le \overline\eta\}.PπB​={π:π=π0+η, e⊤η=0, η​≤η≤η​}.

Then:

  1. If (x∗,u∗,z∗,ξ∗,ω∗,α∗,θ∗)(x^*, u^*, z^*, \xi^*, \omega^*, \alpha^*, \theta^*)(x∗,u∗,z∗,ξ∗,ω∗,α∗,θ∗) solves (24)–(30), then (x∗,u∗,α∗,θ∗)(x^*, u^*, \alpha^*, \theta^*)(x∗,u∗,α∗,θ∗) solves (16)–(20) with Pπ=PπB\mathcal P_\pi = \mathcal P_\pi^BPπ​=PπB​.
  2. Conversely, if (x~∗,u~∗,α~∗,θ~∗)(\tilde x^*, \tilde u^*, \tilde\alpha^*, \tilde\theta^*)(x~∗,u~∗,α~∗,θ~∗) solves (16)–(20) with Pπ=PπB\mathcal P_\pi = \mathcal P_\pi^BPπ​=PπB​, then (x~∗,u~∗,z~∗,ξ~∗,ω~∗,α~∗,θ~∗)(\tilde x^*, \tilde u^*, \tilde z^*, \tilde\xi^*, \tilde\omega^*, \tilde\alpha^*, \tilde\theta^*)(x~∗,u~∗,z~∗,ξ~​∗,ω~∗,α~∗,θ~∗) solves (24)–(30) for every optimal solution (z~∗,ξ~∗,ω~∗)(\tilde z^*, \tilde\xi^*, \tilde\omega^*)(z~∗,ξ~​∗,ω~∗) of the linear program (23) with u=u~∗u = \tilde u^*u=u~∗,
min⁡(z,ξ,ω){η‾⊤ξ+η‾⊤ω:ez+ξ+ω=u~∗, ξ≥0, ω≤0},\min_{(z,\xi,\omega)}\{\overline\eta^\top\xi + \underline\eta^\top\omega : e z + \xi + \omega = \tilde u^*,\ \xi \ge 0,\ \omega \le 0\},(z,ξ,ω)min​{η​⊤ξ+η​⊤ω:ez+ξ+ω=u~∗, ξ≥0, ω≤0},
  1. and such an optimal solution of (23) exists.

Here "solves" means: feasible, with objective value θ\thetaθ minimal among all feasible points. The proposition turns the semi-infinite worst-case CVaR problem (16)–(20) into a single finite problem (24)–(30), a linear program when fff is linear in xxx and X\mathcal XX is a polyhedron.

Formalization Note Constraint (18) is in its equivalent "for all π∈PπB\pi \in \mathcal P_\pi^Bπ∈PπB​" form. Clause 3 makes explicit the existence that the paper's "where (z~∗,ξ~∗,ω~∗)(\tilde z^*, \tilde\xi^*, \tilde\omega^*)(z~∗,ξ~​∗,ω~∗) is an optimal solution to (23)" presupposes. No hypothesis on fff, X\mathcal XX, π0\pi^0π0, η‾\underline\etaη​, η‾\overline\etaη​ is added: when PπB\mathcal P_\pi^BPπB​ is empty neither problem has an optimal solution, so the statement holds for all data.

Preamble
import Mathlib
import Definitions.Def_WorstCaseCVaR_Box_Problems

open Matrix
Formal statement
namespace WorstCaseCVaR.Box

/-- Proposition 2, p. 1159. (1) If `(x*, u*, z*, ξ*, ω*, α*, θ*)` solves (24)–(30), then
`(x*, u*, α*, θ*)` solves (16)–(20) with `𝒫_π = 𝒫_π^B`. (2) Conversely, if `(x̃*, ũ*, α̃*, θ̃*)`
solves (16)–(20) with `𝒫_π = 𝒫_π^B`, then `(x̃*, ũ*, z̃*, ξ̃*, ω̃*, α̃*, θ̃*)` solves (24)–(30)
for every optimal solution `(z̃*, ξ̃*, ω̃*)` of (23) with `u = ũ*`; (3) such an optimal
solution exists. -/
theorem proposition_2 {n m S : ℕ} (f : (Fin n → ℝ) → (Fin m → ℝ) → ℝ)
    (ys : Fin S → Fin m → ℝ) (𝒳 : Set (Fin n → ℝ)) (β : ℝ) (hβ0 : 0 < β) (hβ1 : β < 1)
    (π0 ηlo ηhi : Fin S → ℝ) :
    (∀ q : Var24 n S, Solves24 f ys 𝒳 β π0 ηlo ηhi q →
      Solves16 f ys 𝒳 β π0 ηlo ηhi q.toVar16) ∧
    (∀ p : Var16 n S, Solves16 f ys 𝒳 β π0 ηlo ηhi p →
      ∀ d : Dual23 S, IsOptimal23 ηlo ηhi p.u d →
        Solves24 f ys 𝒳 β π0 ηlo ηhi (p.withDual d)) ∧
    (∀ p : Var16 n S, Solves16 f ys 𝒳 β π0 ηlo ηhi p →
      ∃ d : Dual23 S, IsOptimal23 ηlo ηhi p.u d) := by sorry

end WorstCaseCVaR.Box
Source
Zhu & Fukushima, Worst-Case Conditional Value-at-Risk with Application to Robust Portfolio Management, Oper. Res. 57(5), 2009, p. 1159, Proposition 2 (proof pp. 1167–1168)
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 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