Corollary 1 — Gelbrich risk bound
DisprovedWassersteinDRO.Gelbrich.gelbrich_risk_boundoperations-researchprobabilityrobust-optimization
If the nominal distribution has mean vector and covariance matrix and , then for every loss function the worst-case risk over the Wasserstein ambiguity set is bounded above by the Gelbrich risk: .
Preamble
import Mathlib import Definitions.Def_WassersteinDRO_Gelbrich_worstCaseRisk import Definitions.Def_WassersteinDRO_Gelbrich_gelbrichRisk open MeasureTheory
Formal statement
namespace WassersteinDRO.Gelbrich
/-- Corollary 1 (Gelbrich risk), Kuhn et al. 2019, p. 18, eq. (18): if `P̂N` has mean vector
`μ̂` and covariance matrix `Ŝ` and `p ≥ 2`, then the worst-case risk of any fixed loss
function `ℓ` is bounded above by the Gelbrich risk, `R_{ε,p}(P̂N,ℓ) ≤ R_ε(μ̂,Ŝ,ℓ)`. (The
paper's companion statement for the worst-case *optimal* risk over a loss class `L`,
`R_{ε,p}(P̂N,L) ≤ R_ε(μ̂,Ŝ,L)`, follows immediately by taking the infimum over `ℓ ∈ L` on both
sides of this bound and is not separately formalized — see `STATUS.md`.) -/
theorem gelbrich_risk_bound {m : ℕ} (ε p : ℝ) (hp : 2 ≤ p) (Ξ : Set (EuclideanSpace ℝ (Fin m)))
(PN : Measure (EuclideanSpace ℝ (Fin m))) (μhat : EuclideanSpace ℝ (Fin m))
(SigmaHat : Matrix (Fin m) (Fin m) ℝ) (hμ : meanVector PN = μhat) (hS : covarianceMatrix PN = SigmaHat)
(ℓ : EuclideanSpace ℝ (Fin m) → ℝ) :
worstCaseRisk ε p Ξ PN ℓ ≤ gelbrichRisk ε Ξ μhat SigmaHat ℓ := by sorry
end WassersteinDRO.Gelbrich
Source
Kuhn, Mohajerin Esfahani, Nguyen & Shafieezadeh-Abadeh, Wasserstein Distributionally Robust Optimization, INFORMS TutORials 2019, Corollary 1, p. 18, eq. (18)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.