TheoremBase

The Squared Wasserstein Distance to a Fixed Measure: a One-Sided Bound Along Couplings, Differentiability at a Uniquely Mapped Source, and the Translation and Centring Identities

The squared Wasserstein distance to a fixed measure satisfies a one-sided quadratic bound along every coupling out of a mapped source, is differentiable along couplings with gradient twice the optimal displacement when the source is uniquely mapped, and transforms explicitly under translations, whence the squared distance splits into its centred part and the squared distance of the means.

Statement

In the setting of The Intrinsic Calculus on the Wasserstein Space: Standing Notation, let μ,ν∈P2(Rd)\mu,\nu\in\mathcal{P}_{2}(\mathbb{R}^{d}). The mean m(ρ)∈Rdm(\rho)\in\mathbb{R}^{d} and the centred measure ρˉ∈P2(Rd)\bar{\rho}\in\mathcal{P}_{2}(\mathbb{R}^{d}) of ρ∈P2(Rd)\rho\in\mathcal{P}_{2}(\mathbb{R}^{d}) are those of those clauses, and (τa)#ρ∈P2(Rd)(\tau_{a})_{\#}\rho\in\mathcal{P}_{2}(\mathbb{R}^{d}) for a∈Rda\in\mathbb{R}^{d} by The Optimal Map as a Square-Integrable Vector Field: Integrability, Transport Cost and Uniqueness of the Class (i), the translations τa\tau_{a} being those fixed there. Then the following hold.

1. (A one-sided bound along couplings) Let T:Rd→RdT:\mathbb{R}^{d}\to\mathbb{R}^{d} be an optimal map from μ\mu to ν\nu, with id−T∈L2(μ;Rd)\mathrm{id}-T\in L^{2}(\mu;\mathbb{R}^{d}). Then for every ρ∈P2(Rd)\rho\in\mathcal{P}_{2}(\mathbb{R}^{d}) and every π∈Π(μ,ρ)\pi\in\Pi(\mu,\rho),

W2(ρ,ν)2 ≤ W2(μ,ν)2+2 J(id−T,π)+I(π).W_{2}(\rho,\nu)^{2}\ \le\ W_{2}(\mu,\nu)^{2}+2\,\mathcal{J}(\mathrm{id}-T,\pi)+I(\pi).

2. (Differentiability at a uniquely mapped source) Suppose that the ordered pair (μ,ν)(\mu,\nu) is uniquely mapped, and let TT be an optimal map from μ\mu to ν\nu. Then the function

P2(Rd)→R,ρ↦W2(ρ,ν)2,\mathcal{P}_{2}(\mathbb{R}^{d})\to\mathbb{R},\qquad\rho\mapsto W_{2}(\rho,\nu)^{2},

is differentiable along couplings at μ\mu, with gradient along couplings 2 (id−T)2\,(\mathrm{id}-T).

3. (Translations) For all a,b∈Rda,b\in\mathbb{R}^{d},

W2((τa)#μ,(τb)#ν)2=W2(μ,ν)2+2 (a−b)⋅(m(μ)−m(ν))+∥a−b∥2.W_{2}\bigl((\tau_{a})_{\#}\mu,(\tau_{b})_{\#}\nu\bigr)^{2}=W_{2}(\mu,\nu)^{2}+2\,(a-b)\cdot\bigl(m(\mu)-m(\nu)\bigr)+\lVert a-b\rVert^{2}.

4. (Centring)

W2(μ,ν)2=W2(μˉ,νˉ)2+∥m(μ)−m(ν)∥2.W_{2}(\mu,\nu)^{2}=W_{2}(\bar{\mu},\bar{\nu})^{2}+\lVert m(\mu)-m(\nu)\rVert^{2}.

Proofs

Log in to submit a proof.

Loading...

Citations

Loading…

Dependencies

Loading…

Related

0 relations

Curated associations between results. These are editable and subjective — they do not replace the dependency graph, which is derived from the references in the text.

No relations recorded yet.

Comments

Log in to comment.

Loading…