(34), p. 25 — moment-region value at risk upper bound
ProvedDataDrivenRO.Moment.worst_case_var_lemoment-uncertaintyp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1robust-optimizationvalue-at-risk
Fix , nonnegative moment thresholds , estimates , and a risk level . For every probability law in the supported moment region and every vector ,
The bound transfers uncertainty in the mean and covariance to a bound on every linear projection's upper quantile. It is the probability input to Theorem 10.
Formalization Note Equation (34) prints equality after a supremum over . The region here also restricts support to the radius- ball, and the source does not establish attainment of that equality under the support restriction. This statement takes only the upper-bound direction needed by Theorem 10. The law's ball support ensures the moments are genuine integrals.
Preamble
import Mathlib import Definitions.Def_DataDrivenRO_Moment_Setting
Formal statement
namespace DataDrivenRO.Moment
/-- The upper-bound direction of (34), p. 25. The printed equality is not
asserted because the region also restricts distributions to a support ball. -/
theorem worst_case_var_le {d : ℕ} (R Γ₁ Γ₂ : ℝ)
(μhat : Fin d → ℝ) (Shat : Matrix (Fin d) (Fin d) ℝ)
(hR : 0 ≤ R) (hΓ₁ : 0 ≤ Γ₁) (hΓ₂ : 0 ≤ Γ₂)
(ε : ℝ) (hε0 : 0 < ε) (hε1 : ε < 1) :
∀ P ∈ PCS R Γ₁ Γ₂ μhat Shat, ∀ v : Fin d → ℝ,
VaR P ε v ≤ csValue μhat Shat Γ₁ Γ₂ ε v := by sorry
end DataDrivenRO.Moment
Source
Bertsimas, Gupta & Kallus, Data-Driven Robust Optimization, arXiv:1401.0212v2, (34), p. 25
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.