p. 1159 —
ProvedWorstCaseCVaR.Box.max_decompositioncvarlinear-programmingp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1robust-optimization
Let , , , and let define the box uncertainty set of (21). Suppose the linear program (22),
is feasible. Then (22) attains its optimal value , and the maximum of over is attained and equals
This reduces the worst-case constraint (18) over the box to the optimal value of a single linear program, which is the first step toward the LP reformulation (24)–(30).
Formalization Note "Attained maximum with value " is IsGreatest of the image set. Feasibility of (22) is assumed; it is equivalent to nonemptiness of .
Preamble
import Mathlib import Definitions.Def_WorstCaseCVaR_Box_DualPair open Matrix
Formal statement
namespace WorstCaseCVaR.Box
/-- p. 1159, display before (22): if (22) is feasible, then (22) attains its optimal value
`γ*(u)`, and `max_{π ∈ 𝒫_π^B} α + (1 − β)⁻¹ πᵀu` exists and equals
`α + (1 − β)⁻¹ (π⁰)ᵀu + γ*(u)/(1 − β)`. -/
theorem max_decomposition {S : ℕ} (β α : ℝ) (hβ0 : 0 < β) (hβ1 : β < 1)
(π0 ηlo ηhi u : Fin S → ℝ) (hfeas : (lp22Feasible ηlo ηhi).Nonempty) :
IsGreatest ((fun η => u ⬝ᵥ η) '' lp22Feasible ηlo ηhi) (gammaStar ηlo ηhi u) ∧
IsGreatest ((fun π => α + (1 - β)⁻¹ * (π ⬝ᵥ u)) '' boxSet π0 ηlo ηhi)
(α + (1 - β)⁻¹ * (π0 ⬝ᵥ u) + gammaStar ηlo ηhi u / (1 - β)) := 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, display preceding Eq. (22)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.