(13) —
ProvedModelRiskOT.WorstProb.eq_13dualityoptimal-transportp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1worst-case-probability
Let be a Polish space, a cost satisfying (A1), a probability measure on , , and a nonempty closed set. Then the worst-case probability (12) is the value of a univariate convex problem:
The only measure on the right-hand side is the baseline , so the infinite-dimensional worst case is computed by a one-dimensional minimization. It is the specialization of the paper's strong duality theorem (Theorem 1, with Remark 1's (9)) to , which is upper semicontinuous because is closed.
Formalization Note Both sides are in ; the expectation is a lower Lebesgue integral, which equals the integral under the completion of since is universally measurable.
Preamble
import Mathlib import Definitions.Def_ModelRiskOT_WorstProb_WorstCaseProb open MeasureTheory open scoped ENNReal
Formal statement
namespace ModelRiskOT.WorstProb
/-- (13), p. 9: for a nonempty closed set `A`, the worst-case probability (12) is the value of a
univariate problem,
`sup {P(A) : d_c(μ, P) ≤ δ} = inf_{λ ≥ 0} {λδ + E_μ[(1 − λ c(X, A))⁺]}`. -/
theorem eq_13 {S : Type*} [TopologicalSpace S] [PolishSpace S] [MeasurableSpace S]
[BorelSpace S] (c : S → S → ℝ) (hc : ModelRiskOT.Duality.AssumptionA1 c) (μ : Measure S) [IsProbabilityMeasure μ]
(δ : ℝ) (hδ : 0 < δ) (A : Set S) (hA : IsClosed A) (hAne : A.Nonempty) :
worstProb c μ δ A = ⨅ lam ∈ Set.Ici (0 : ℝ), obj13 c μ δ A lam := by sorry
end ModelRiskOT.WorstProb
Source
Blanchet & Murthy, Quantifying Distributional Model Risk via Optimal Transport, arXiv:1604.01446v2, p. 9, Eq. (13)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.