A well known fact in probability is that a uniform random variable on $[0,1]$ can be used to simulate any other probability distribution on $\mathbb{R}$.

A standard way of doing this is to define, given $\mu$ a probability on $\mathbb{R}$, the random variable $F(u,\mu) = \min\lbrace x \in \mathbb{R}: \mu((-\infty,x]) \ge u\rbrace$. The random variable $F(u,\mu)$ has distribution $\mu$.

Suppose now that $(X,d)$ is a compact metric space and let $\mathcal{P}(X)$ be the space of Borel probability measures on $X$ endowed with the topology of weak convergence.

I'm looking for a reference for the following statement:

There exists a Borel function $F:[0,1]\times \mathcal{P}(X) \to X$ such that if $u$ is a random variable whose distribution is uniform on $[0,1]$ then for each $\mu$ the random variable $F(u,\mu)$ has distribution $\mu$.

  • 3
    $\begingroup$ I haven't seen this written down; we might consider it "folklore". But you have basically already written down the proof on $\mathbb{R}$ (just take the inf over rationals to make it clear that $F$ is Borel), and for any other compact metric space $X$, there is a Borel isomorphism from $X$ to $\mathbb{R}$, so the general case follows at once. $\endgroup$ – Nate Eldredge Aug 27 '16 at 16:10

This is a result of Blackwell and Dubins, "An extension of Skorohod's almost sure representation theorem". In fact, your function F can be constructed to be almost surely continuous in the measure argument, and X can be any Polish space.

| cite | improve this answer | |
  • $\begingroup$ According to the comment guidelines I'm supposed to avoid saying thanks. But thanks! $\endgroup$ – Pablo Lessa Aug 27 '16 at 17:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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