(3.4), p. 11, and p. 18 — D(µ₁ − µ₂) = −2‖β_ϕ(µ₁ − µ₂)‖², with β_ϕ(µ)(w, s) = µ{u ; w(u) ≤ cs}
ProvedDistCov.Hilbert.Ddiff_eq_baryDiffbarycenterdistance-covariancegaussiannegative-typep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be Borel probability measures on with finite first moments, let , , and be as in the Gaussian Crofton embedding, and let
which is the barycenter of the signed measure under the embedding . Then is integrable with respect to and
This is identity (3.4), for , applied to the Gaussian embedding of . It reduces strong negative type of to showing that almost everywhere forces .
Formalization Note. is the energy expansion of the Setting file, and is written as the integral of . Finite first moment means for some . Indices start at .
Preamble
import Mathlib import Definitions.Def_DistCov_Hilbert_Setting
Formal statement
namespace DistCov.Hilbert
open MeasureTheory ProbabilityTheory
theorem Ddiff_eq_baryDiff (μ₁ μ₂ : Measure DistCov.Indep.L2N) [IsProbabilityMeasure μ₁]
[IsProbabilityMeasure μ₂] (h₁ : DistCov.Indep.FiniteFirstMoment μ₁) (h₂ : DistCov.Indep.FiniteFirstMoment μ₂) :
Integrable (fun p => baryDiff μ₁ μ₂ p ^ 2) (rho.prod volume) ∧
DistCov.Indep.Ddiff μ₁ μ₂ = -2 * ∫ p, baryDiff μ₁ μ₂ p ^ 2 ∂(rho.prod volume) := by sorry
end DistCov.Hilbert
Source
Lyons, Distance covariance in metric spaces, arXiv:1106.5758 (version of 19 Dec. 2020), p. 11, (3.4); p. 18, proof of Theorem 3.16, display β_ϕ(µ) : (w, s) ↦ µ{u ; w(u) ≤ cs}
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.