Proposition 3.7 — dcov(θ) = 4‖β_{ϕ⊗ψ}(θ − µ × ν)‖²
ProvedDistCov.Indep.proposition_3_7Let and be separable metric spaces with their Borel -fields, of negative type as witnessed by embeddings and into real Hilbert spaces. Let be a probability measure on whose marginals and have finite first moments. Then has finite first moment, i.e. the tensor embedding is -integrable, so that is defined, and
The proposition identifies distance covariance with the squared distance, in the Hilbert tensor product, between the barycenters of and of the product of its marginals. In particular on spaces of negative type, and exactly when these two barycenters coincide.
Formalization Note is the completion of the algebraic tensor product with the inner product . The paper uses one space for both factors; allowing two spaces is a harmless generalization of a universally quantified statement. is written by linearity. "Finite first moment of " is stated as Bochner integrability of with respect to . Separability is the standing assumption of Errata (i), p. 24.
import Mathlib import Definitions.Def_DistCov_Indep_Setting
namespace DistCov.Indep
open MeasureTheory
theorem proposition_3_7 {X : Type*} [MetricSpace X] [MeasurableSpace X] [BorelSpace X] [SecondCountableTopology X] {Y : Type*} [MetricSpace Y] [MeasurableSpace Y] [BorelSpace Y] [SecondCountableTopology Y] {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] {K : Type*} [NormedAddCommGroup K] [InnerProductSpace ℝ K] [CompleteSpace K]
(ϕ : X → H) (ψ : Y → K) (hϕ : IsNegTypeEmbedding ϕ) (hψ : IsNegTypeEmbedding ψ)
(θ : Measure (X × Y)) [IsProbabilityMeasure θ]
(hμ : FiniteFirstMoment (θ.map Prod.fst)) (hν : FiniteFirstMoment (θ.map Prod.snd)) :
Integrable (tensorEmb ϕ ψ) θ ∧
dcov θ = 4 * ‖bary (tensorEmb ϕ ψ) θ
- bary (tensorEmb ϕ ψ) ((θ.map Prod.fst).prod (θ.map Prod.snd))‖ ^ 2 := by sorry
end DistCov.Indep
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.