Log In Sign Up

Rate of convergence of the smoothed empirical Wasserstein distance

by   Adam Block, et al.

Consider an empirical measure ā„™_n induced by n iid samples from a d-dimensional K-subgaussian distribution ā„™ and let Ī³ = š’©(0,Ļƒ^2 I_d) be the isotropic Gaussian measure. We study the speed of convergence of the smoothed Wasserstein distance W_2(ā„™_n * Ī³, ā„™*Ī³) = n^-Ī± + o(1) with * being the convolution of measures. For K<Ļƒ and in any dimension dā‰„ 1 we show that Ī± = 12. For K>Ļƒ in dimension d=1 we show that the rate is slower and is given by Ī± = (Ļƒ^2 + K^2)^2 4 (Ļƒ^4 + K^4) < 1/2. This resolves several open problems in <cit.>, and in particular precisely identifies the amount of smoothing Ļƒ needed to obtain a parametric rate. In addition, we also establish that D_KL(ā„™_n * Ī³ā„™*Ī³) has rate O(1/n) for K<Ļƒ but only slows down to O((log n)^d+1 n) for K>Ļƒ. The surprising difference of the behavior of W_2^2 and KL implies the failure of T_2-transportation inequality when Ļƒ < K. Consequently, the requirement K<Ļƒ is necessary for validity of the log-Sobolev inequality (LSI) for the Gaussian mixture ā„™ * š’©(0, Ļƒ^2), closing an open problem in <cit.>, who established the LSI under precisely this condition.


page 1

page 2

page 3

page 4

āˆ™ 01/27/2020

Exact rate of convergence of the mean Wasserstein distance between the empirical and true Gaussian distribution

We study the Wasserstein distance W_2 for Gaussian samples. We establish...
āˆ™ 07/22/2018

On the rate of convergence of empirical measure in āˆž-Wasserstein distance for unbounded density function

We consider a sequence of identically independently distributed random s...
āˆ™ 01/20/2021

A short proof on the rate of convergence of the empirical measure for the Wasserstein distance

We provide a short proof that the Wasserstein distance between the empir...
āˆ™ 04/01/2019

Gaussian approximation for empirical barycenters

In this work we consider Wasserstein barycenters (average in Wasserstein...
āˆ™ 05/30/2019

Convergence of Smoothed Empirical Measures with Applications to Entropy Estimation

This paper studies convergence of empirical measures smoothed by a Gauss...
āˆ™ 08/31/2020

Note on approximating the Laplace transform of a Gaussian on a complex disk

In this short note we study how well a Gaussian distribution can be appr...
āˆ™ 11/12/2020

Implicit bias of any algorithm: bounding bias via margin

Consider n points x_1,ā€¦,x_n in finite-dimensional euclidean space, each ...