Let $X$ and $Y$ be two random variables with first order moments, i.e. $E[|X|]$, $E[|Y|]<+\infty$. Assume further that


Set $Law(X)=\mu$ and $Law(Y)=\nu$, it is clear that $\mu$ and $\nu$ are close in the Prokhorov metric, see


for definition. Denote by $\rho(\cdot,\cdot)$ the Prokhorov metric. My question is how to estimate $\rho(\mu,\nu)$. For example, could we show that $\rho(\mu,\nu)<\varepsilon$ or $\rho(\mu,\nu)<\sqrt{\varepsilon}$? Thanks for the reply!


1 Answer 1


$\sqrt{\varepsilon}$ works. Assume that for some set $A$ we have $\mu(A)=a$ and $\nu(A^{\sqrt{\varepsilon}})<a-\sqrt{\varepsilon}$. Then with probability more then $\sqrt{\varepsilon}$ we have $X\in A$, $Y\notin A^{\sqrt{\varepsilon}}$. In this case $|X-Y|\geqslant \sqrt{\varepsilon}$ for sure. Hence expectation of $|X-Y|$ is more than $\varepsilon$.

  • $\begingroup$ Thanks a lot for your reply. Indeed, that will work even we replace the Prokhorov metric by the Wasserstein metric. $\endgroup$
    – CodeGolf
    Jan 12, 2016 at 14:12
  • $\begingroup$ Kantorovich optimal transportation distance (is it the same thing that you call Wasserstein metric?) may be defined as supremum of integrals $E(f(X)-f(Y))$ for 1-Lipschitz functions $f$, clearly this does not exceed $E(|X-Y|)<\varepsilon$. $\endgroup$ Jan 12, 2016 at 14:22
  • $\begingroup$ Exactly. It follows from the dual formulation. $\endgroup$
    – CodeGolf
    Jan 12, 2016 at 19:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.