TheoremBase

The Noise Wasserstein Distance is a Metric on the Measures Noise-Connected to the Reference Measure: Existence of Noise-Optimal Couplings, Comparison with the Quadratic Wasserstein Distance and Lower Semicontinuity

The measures noise-connected to the reference measure have finite second moment and are pairwise noise-connected, noise-optimal couplings exist, and the noise Wasserstein distance is a metric on them that dominates the quadratic Wasserstein distance up to a constant and is lower semicontinuous under weak convergence.

Statement

In the setting of Probability Measures on a Hilbert Space Transported in the Noise Norm: Standing Notation, with the reference measure ρ∈P2(X)\rho\in\mathcal{P}_{2}(X) of Probability Measures on a Hilbert Space Transported in the Noise Norm: Standing Notation §reference, the sets Πa(μ,ν)\Pi^{a}(\mu,\nu) of couplings of finite noise cost and noise-connectedness are those of Couplings of Finite Noise Cost and Their Noise Cost; WaW_{a} is the noise Wasserstein distance, a nonnegative real-valued function on the noise-connected ordered pairs of members of P(X)\mathcal{P}(X); noise-optimal couplings are as defined there; and Pρa\mathcal{P}^{a}_{\rho} is the set of measures noise-connected to ρ\rho. Then claims 1 to 3 hold, claims 4 to 6 hold for all μ,ν,λ∈Pρa\mu,\nu,\lambda\in\mathcal{P}^{a}_{\rho}, and claims 7 to 9 hold.

1. (The reference measure and finite second moments) ρ∈Pρa\rho\in\mathcal{P}^{a}_{\rho}. Pρa⊆P2(X)\mathcal{P}^{a}_{\rho}\subseteq\mathcal{P}_{2}(X).

2. (Noise-connectedness) For any two members μ,ν\mu,\nu of Pρa\mathcal{P}^{a}_{\rho}, distinct or not, the ordered pair (μ,ν)(\mu,\nu) is noise-connected, so that Wa(μ,ν)W_{a}(\mu,\nu) is defined.

3. (Noise-optimal couplings) For all μ,ν∈P(X)\mu,\nu\in\mathcal{P}(X) such that (μ,ν)(\mu,\nu) is noise-connected there is a noise-optimal coupling π∈Πa(μ,ν)\pi\in\Pi^{a}(\mu,\nu).

4. (Symmetry) Wa(μ,ν)=Wa(ν,μ)W_{a}(\mu,\nu)=W_{a}(\nu,\mu).

5. (Separation) Wa(μ,ν)=0W_{a}(\mu,\nu)=0 if and only if μ=ν\mu=\nu.

6. (Triangle inequality) Wa(μ,λ)≤Wa(μ,ν)+Wa(ν,λ)W_{a}(\mu,\lambda)\le W_{a}(\mu,\nu)+W_{a}(\nu,\lambda).

7. (The noise Wasserstein space) The restriction of WaW_{a} to pairs of members of Pρa\mathcal{P}^{a}_{\rho} is a metric on Pρa\mathcal{P}^{a}_{\rho}; the metric space (Pρa,Wa)(\mathcal{P}^{a}_{\rho},W_{a}) is called the noise Wasserstein space based at ρ\rho.

8. (Comparison) For all μ,ν∈Pρa\mu,\nu\in\mathcal{P}^{a}_{\rho}, which belong to P2(X)\mathcal{P}_{2}(X) by claim 1, the quadratic Wasserstein distance satisfies W2(μ,ν)≤aˉ Wa(μ,ν)W_{2}(\mu,\nu)\le\sqrt{\bar{a}}\,W_{a}(\mu,\nu), with aˉ\bar{a} the positive bound on the noise weights of Probability Measures on a Hilbert Space Transported in the Noise Norm: Standing Notation §weights.

9. (Lower semicontinuity) Let μj,νj∈Pρa\mu_{j},\nu_{j}\in\mathcal{P}^{a}_{\rho} for j∈Nj\in\mathbb{N} and μ,ν∈P(X)\mu,\nu\in\mathcal{P}(X) satisfy μj⇒μ\mu_{j}\Rightarrow\mu and νj⇒ν\nu_{j}\Rightarrow\nu, and suppose that the sequence (Wa(μj,νj))j∈N(W_{a}(\mu_{j},\nu_{j}))_{j\in\mathbb{N}} of real numbers is bounded. Then (μ,ν)(\mu,\nu) is noise-connected and Wa(μ,ν)≤lim inf⁡jWa(μj,νj)W_{a}(\mu,\nu)\le\liminf_{j}W_{a}(\mu_{j},\nu_{j}), the limit inferior of that bounded sequence. If moreover μ∈Pρa\mu\in\mathcal{P}^{a}_{\rho}, then ν∈Pρa\nu\in\mathcal{P}^{a}_{\rho}.

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…