TheoremBase

The Torus Wasserstein Distance: Comparison with the Euclidean Distance, Wrapping, and Integrals of Periodic Functions

Measures on the flat torus have finite second moment; the torus cost of a coupling is at most its Euclidean quadratic cost; wrapping two measures on Euclidean space does not increase their Wasserstein distance; and integrals of continuous periodic functions depend continuously, and Lipschitz-continuously for Lipschitz ones, on the measure in the torus Wasserstein distance.

Statement

In the setting of The Flat Torus: Standing Notation, used with n=dn=d for a natural number dd with 1≤d1\le d, and of Probability Measures on Euclidean Space and Random Vectors: Standing Notation, let π\pi be the wrapping map and dTd_{\mathbb{T}} the flat torus distance; let P(Td)\mathcal{P}(\mathbb{T}^{d}), the torus cost ITI_{\mathbb{T}} and the torus Wasserstein distance WTW_{\mathbb{T}} be as in Probability Measures on the Flat Torus, the Torus Cost of a Coupling, the Torus Wasserstein Distance and Optimal Couplings §measures. For μ,ν∈P(Rd)\mu,\nu\in\mathcal{P}(\mathbb{R}^{d}) let Π(μ,ν)\Pi(\mu,\nu) and the quadratic cost II be as in Couplings of Two Probability Measures on Euclidean Space and Their Quadratic Cost §coupling, let M2M_{2} be the second moment and P2(Rd)\mathcal{P}_{2}(\mathbb{R}^{d}) as in The Second Moment of a Probability Measure on Euclidean Space and the Probability Measures with Finite Second Moment §space, and let W2W_{2} be the quadratic Wasserstein distance. Then the following hold.

1. (Inclusion) Every μ∈P(Td)\mu\in\mathcal{P}(\mathbb{T}^{d}) belongs to P2(Rd)\mathcal{P}_{2}(\mathbb{R}^{d}), with M2(μ)≤dM_{2}(\mu)\le d.

2. (Comparison with the Euclidean cost) For μ,ν∈P(Td)\mu,\nu\in\mathcal{P}(\mathbb{T}^{d}) and γ∈Π(μ,ν)\gamma\in\Pi(\mu,\nu) one has IT(γ)≤I(γ)I_{\mathbb{T}}(\gamma)\le I(\gamma); hence WT(μ,ν)≤W2(μ,ν)W_{\mathbb{T}}(\mu,\nu)\le W_{2}(\mu,\nu).

3. (Wrapping) For μ′∈P(Rd)\mu'\in\mathcal{P}(\mathbb{R}^{d}) one has π#μ′∈P(Td)\pi_{\#}\mu'\in\mathcal{P}(\mathbb{T}^{d}). For μ′,ν′∈P(Rd)\mu',\nu'\in\mathcal{P}(\mathbb{R}^{d}) and γ′∈Π(μ′,ν′)\gamma'\in\Pi(\mu',\nu'), the map π×π=(π∘pr1,π∘pr2)\pi\times\pi=(\pi\circ\mathrm{pr}_{1},\pi\circ\mathrm{pr}_{2}) of Rd+d\mathbb{R}^{d+d} is Borel, (π×π)#γ′∈Π(π#μ′,π#ν′)(\pi\times\pi)_{\#}\gamma'\in\Pi(\pi_{\#}\mu',\pi_{\#}\nu'), and

IT((π×π)#γ′)=∫Rd+ddT(pr1(w),pr2(w))2 γ′(dw)≤I(γ′).I_{\mathbb{T}}\bigl((\pi\times\pi)_{\#}\gamma'\bigr)=\int_{\mathbb{R}^{d+d}}d_{\mathbb{T}}\bigl(\mathrm{pr}_{1}(w),\mathrm{pr}_{2}(w)\bigr)^{2}\,\gamma'(dw)\le I(\gamma').

Hence WT(π#μ′,π#ν′)≤W2(μ′,ν′)W_{\mathbb{T}}(\pi_{\#}\mu',\pi_{\#}\nu')\le W_{2}(\mu',\nu') whenever μ′,ν′∈P2(Rd)\mu',\nu'\in\mathcal{P}_{2}(\mathbb{R}^{d}).

4. (Integrals of periodic functions) Let f∈Cperf\in C_{\mathrm{per}}, the continuous Zd\mathbb{Z}^{d}-periodic functions. Then ff is bounded, by Elementary Properties of Lattice-Periodic Functions §bounded, and Borel, being continuous (Probability Measures on Euclidean Space and Random Vectors: Standing Notation §borel-maps), hence integrable with respect to every μ∈P(Rd)\mu\in\mathcal{P}(\mathbb{R}^{d}) by Probability Measures on Euclidean Space and Random Vectors: Standing Notation §measures. If ff is Lipschitz with constant LL, for a real L≥0L\ge0, then ∣∫f dμ−∫f dν∣≤L WT(μ,ν)\bigl|\int f\,d\mu-\int f\,d\nu\bigr|\le L\,W_{\mathbb{T}}(\mu,\nu) for all μ,ν∈P(Td)\mu,\nu\in\mathcal{P}(\mathbb{T}^{d}). If μj,μ∈P(Td)\mu_{j},\mu\in\mathcal{P}(\mathbb{T}^{d}) for j∈Nj\in\mathbb{N} and the real sequence (WT(μj,μ))j∈N(W_{\mathbb{T}}(\mu_{j},\mu))_{j\in\mathbb{N}} converges to 00, then (∫f dμj)j∈N(\int f\,d\mu_{j})_{j\in\mathbb{N}} converges to ∫f dμ\int f\,d\mu.

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…