6
$\begingroup$

This may be a very basic question. Let $X$ be a complete metric space and let $T$ be a compact subset of $X$. Say that a function $\pi: X \to T$ is a projection if $$ d(x, \pi(x)) = d(x, T) \quad \forall x \in X\,. $$ If $T$ is a closed convex subset of $\mathbb{R}^d$ then the (unique) projection $\pi$ onto $T$ is always continuous; of course this will not happen in general. I am interested in the following weaker question:

Does there always exist a projection that is Borel measurable?

$\endgroup$
  • 2
    $\begingroup$ Did you try to use the measurable selection theorem of Kuratowski and Ryll-Nardzewski? $\endgroup$ – Jochen Wengenroth Apr 21 '17 at 8:55
5
$\begingroup$

Consider the closed multimap $\Pi:X\to 2^T$, whose graph is the closed set $\mathrm{graph}(\Pi):=\big\{(x,t)\in X\times T\, : d(x,t)=\min_{s\in T}d(x,s)\big\}\subset X\times T$. You want a measurable selection $\pi:X\to T$ of $\Pi$, that is $\mathrm{graph}(\pi)\subset\mathrm{graph}(\Pi)$. Since $K$ is a compact metric space, thus a Polish space, the classical Kuratowski–Ryll-Nardzewski measurable selection theorem does the job.

rmk. The Kuratowski–Ryll-Nardzewski measurable selection theorem admits an easy proof in the particular case of a closed, non-empty set valued multimap $\Pi:X\to2^T$ with $T$ a compact metric space. In this case (by the Alexandroff-Hausdorff theorem) $T$ is a continuous image $T=\kappa( C)$ of the Cantor set $C$ via some continuous surjective map $\kappa:C\to T$. The pre-image of the closed set $\mathrm{graph}(\Pi)$ via $\mathrm{id}_X\times\kappa:X\times C\to X\times T$ is therefore the graph of a closed multimap $X\to 2^C$, that is the multimap $x\mapsto \kappa^{-1}\Pi(x)\neq\emptyset;$ a lower semi-continuous, hence measurable selection of the latter, is $x\mapsto \min\kappa^{-1}\Pi(x) $, and a measurable selection of $\Pi$ is therefore

$$\pi(x):=\kappa\big(\min\kappa^{-1}\Pi(x)\big)\in\Pi(x).$$

$\endgroup$
  • 1
    $\begingroup$ Ah! I was unaware of (or had forgotten) the term "selection," which made it much harder to search for useful results. In my application $X$ is separable, so this does the job. Thanks! $\endgroup$ – Jon Apr 21 '17 at 9:02
  • $\begingroup$ Actually we don't need any further hypotheses on the metric space $(X,d)$ because $T$ is already a Polish space, being a compact metric space. (edited) $\endgroup$ – Pietro Majer Apr 22 '17 at 15:23

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.