Proposition 3.5 — a_µ(x) = ‖ϕ(x) − β_ϕ(µ)‖² + D(µ)/2, D(µ) = 2Var(ϕ(X)), d_µ = −2⟨ϕ − β, ϕ′ − β⟩
ProvedDistCov.Indep.proposition_3_5barycenterhilbert-spacenegative-typep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1variance
Let be a separable metric space with its Borel -field, with negative type witnessed by the embedding into a real Hilbert space (), and let be a probability measure on with finite first moment. Write . Then:
- for all ,
- if , where , that is
- for all ,
The third identity expresses the centred distance as a (negative) Gram kernel of the centred embedding; it is what turns into a squared norm in Proposition 3.7.
Formalization Note The variance is written out as the integral of , since . Hilbert spaces are real (Errata (viii), p. 27); separability is the standing assumption of Errata (i), p. 24.
Preamble
import Mathlib import Definitions.Def_DistCov_Indep_Setting
Formal statement
namespace DistCov.Indep
open MeasureTheory
open scoped InnerProductSpace
theorem proposition_3_5 {X : Type*} [MetricSpace X] [MeasurableSpace X] [BorelSpace X] [SecondCountableTopology X] {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
(ϕ : X → H) (hϕ : IsNegTypeEmbedding ϕ)
(μ : Measure X) [IsProbabilityMeasure μ] (h : FiniteFirstMoment μ) :
(∀ x, meanDist μ x = ‖ϕ x - bary ϕ μ‖ ^ 2 + D μ / 2) ∧
D μ = 2 * ∫ x, ‖ϕ x - bary ϕ μ‖ ^ 2 ∂μ ∧
(∀ x x', dcent μ x x' = -2 * ⟪ϕ x - bary ϕ μ, ϕ x' - bary ϕ μ⟫_ℝ) := by sorry
end DistCov.Indep
Source
Lyons, Distance covariance in metric spaces, arXiv:1106.5758 (version of 19 Dec. 2020), p. 13, Proposition 3.5 and the definition of Var(Z) before it; Errata (i), p. 24 and (viii), p. 27
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.