Theorem 4 — Gelbrich bound
DisprovedWassersteinDRO.Gelbrich.gelbrich_boundoperations-researchprobabilityrobust-optimization
If is the Euclidean norm and the distributions have mean vectors and covariance matrices , then , with equality when and are elliptical distributions with the same density generator .
Preamble
import Mathlib import Definitions.Def_WassersteinDRO_Gelbrich_wassersteinDistance import Definitions.Def_WassersteinDRO_Gelbrich_meanVector import Definitions.Def_WassersteinDRO_Gelbrich_covarianceMatrix import Definitions.Def_WassersteinDRO_Gelbrich_psdSqrt import Definitions.Def_WassersteinDRO_Gelbrich_IsElliptical open MeasureTheory
Formal statement
namespace WassersteinDRO.Gelbrich
/-- Theorem 4 (Gelbrich bound), Kuhn et al. 2019, p. 8, eq. (8): if `Q, Q'` have mean vectors
`μ, μ'` and covariance matrices `S, S' ∈ S^m_+`, then
`W2(Q,Q') ≥ (‖μ-μ'‖² + Tr[S+S'-2(S^{1/2}S'S^{1/2})^{1/2}])^{1/2}`, with equality if `Q, Q'`
are elliptical distributions with the same density generator `g`. -/
theorem gelbrich_bound {m : ℕ} (Q Q' : Measure (EuclideanSpace ℝ (Fin m)))
(μ μ' : EuclideanSpace ℝ (Fin m)) (S S' : Matrix (Fin m) (Fin m) ℝ)
(hS : S.PosSemidef) (hS' : S'.PosSemidef)
(hμ : meanVector Q = μ) (hμ' : meanVector Q' = μ')
(hSeq : covarianceMatrix Q = S) (hS'eq : covarianceMatrix Q' = S') :
wassersteinDistance 2 Q Q' ≥
ENNReal.ofReal (Real.sqrt (‖μ - μ'‖ ^ 2 +
(S + S' - (2 : ℝ) • psdSqrt (psdSqrt S * S' * psdSqrt S)).trace)) ∧
(∀ g : ℝ → ℝ, IsElliptical Q g μ S → IsElliptical Q' g μ' S' →
wassersteinDistance 2 Q Q' =
ENNReal.ofReal (Real.sqrt (‖μ - μ'‖ ^ 2 +
(S + S' - (2 : ℝ) • psdSqrt (psdSqrt S * S' * psdSqrt S)).trace))) := by sorry
end WassersteinDRO.Gelbrich
Source
Kuhn, Mohajerin Esfahani, Nguyen & Shafieezadeh-Abadeh, Wasserstein Distributionally Robust Optimization, INFORMS TutORials 2019, Theorem 4, p. 8, eq. (8)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.