Proof of Theorem 5, p. 34 — K(ν, ν′) ≤ ∫ K(ν, δ_y) dν′(y)
ProvedOTSensitivity.MaxFun.dirac_average_boundconvexitydirac-measureoptimal-transportp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a Polish space and continuous with . For all probability measures on ,
The paper uses this with the law of the output and a conditional law, "for every ". It says that lies below the average of its values at Dirac masses; combined with (62) it shows that forces equality for almost every , which is (63).
Formalization Note The statement is given for every probability measure , which contains the page's instance for each . is the published RWPI.SqrtLasso.transportCost; values lie in .
Preamble
import Mathlib import Definitions.Def_RWPI_SqrtLasso_transportCost import Definitions.Def_OTSensitivity_MaxFun_Setting open MeasureTheory ProbabilityTheory open scoped ENNReal
Formal statement
namespace OTSensitivity.MaxFun
/-- Proof of Theorem 5, p. 34, display before (63): `K(ν, ·)` lies below the average of its values
at Dirac masses, `K(ν, ν') ≤ ∫ K(ν, δ_y) dν'(y)`, for every probability measure `ν'` (the page
applies it to `ν' = ν^𝓕_x`). -/
theorem dirac_average_bound {𝒴 : Type*} [TopologicalSpace 𝒴] [PolishSpace 𝒴] [MeasurableSpace 𝒴]
[BorelSpace 𝒴] (k : 𝒴 → 𝒴 → ℝ≥0∞) (hk : IsCost k)
(ν ν' : Measure 𝒴) [IsProbabilityMeasure ν] [IsProbabilityMeasure ν'] :
RWPI.SqrtLasso.transportCost k ν ν' ≤
∫⁻ y, RWPI.SqrtLasso.transportCost k ν (Measure.dirac y) ∂ν' := by sorry
end OTSensitivity.MaxFun
Source
Borgonovo, Figalli, Plischke & Savaré, Global Sensitivity Analysis via Optimal Transport, authors' manuscript of 7 March 2024 (Management Science, doi:10.1287/mnsc.2023.01796), Appendix A, proof of Theorem 5, display before (63), p. 34
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.