Let $X$ and $Y$ be compact metric spaces, $\pi: X \to Y$ a continuous onto map, and $\nu$ a Borel probability measure on $Y$ such that $\nu$-almost every $y \in Y$ has unique $\pi$-preimage. Does there exist a measurable map $s: Y \to X$ such that for $\nu$-almost every $y \in Y$, $(\pi \circ s)(y) = y$?

For the application I'm interested in, $X$ and $Y$ are shift spaces with finite alphabet over a countable group, if that helps at all.


Every continuous onto map $\pi: X \to Y$ between compact metric spaces has a Borel measurable section, i.e. there is a Borel measurable map $s: Y \to X$ such that $\pi \circ s (y) = y$ for all $y$. The assumptions about the measure $\nu$ and the preimages of $\pi$ are superfluous.

For a proof of this, you can look up Theorem 6.9.7 in Bogachev's Measure Theory, Volume 2. The idea of the proof is to start with the case where $X \subset [0,1]$. In this case, for a given $y \in Y$ you can select a preimage of $y$ by taking the smallest one, i.e. by setting $$ s(y) = \inf \; \pi^{-1}(y). $$ In the general case, the trick is to use the fact that for any compact metric space $X$ there is some compact set $K \subset [0,1]$ and a surjective continuous map $f: K \to X$ (in fact you can take $K$ to be the Cantor set). Then you can apply the previous case to the composition $\pi \circ f$.

  • $\begingroup$ Thanks! The assumptions about the measure $\nu$ and the preimages of $\pi$ are needed for my application, but it's good to know they aren't needed for the section to exist. $\endgroup$ – Sophie MacDonald Mar 27 at 16:57
  • $\begingroup$ My pleasure! Of course under these assumptions any two sections will agree $\nu$-almost everywhere, whereas in general one can't say much. $\endgroup$ – Dominique R.F. Mar 27 at 17:12
  • $\begingroup$ Yes, and in fact my situation is that $\nu = \pi_* \mu$ where $\mu$ is a fully supported probability measure on $X$; this means that $s$ is also a left inverse of $\pi$ $\mu$-almost everywhere. Moreover, $\mu$ and $\nu$ are both invariant under the continuous action of a countable group $G$, and $\pi$ is $G$-equivariant; the assumptions thus imply that $s$ is also $G$-equivariant almost everywhere. $\endgroup$ – Sophie MacDonald Mar 27 at 17:28

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.