4
$\begingroup$

Let $(X,\Sigma)$ be a standard measurable space, and let $\,\preceq\,$ be a total order on $X$ with the property that $\,\{(x,y) \in X \times X: x \preceq y\} \in \Sigma \otimes \Sigma$.

Let $A \subset X$ be a set with the property that for all $x,z \in A$ and $y \in X$ with $x \preceq y \preceq z$, we have $y \in A$.

Is it necessarily the case that $A \in \Sigma$? If not, is $A$ necessarily universally measurable with respect to $\Sigma$?

$\endgroup$
1
  • 1
    $\begingroup$ A thought: is it possible for a Borel order to have a set isomorphic to $\omega_1$? It seems unlikely. If not, then take a cofinal well-ordered subset $C$ of $A$, which must then be countable. Fix some $x_0 \in A$ and then you have $A \cap [x,\infty) = \bigcup_{y \in C} [x,y]$ which is Borel. Applying the same argument to the reverse order, $A \cap (-\infty, x]$ is also Borel. $\endgroup$ Sep 20, 2015 at 17:24

1 Answer 1

2
$\begingroup$

I believe the answer is yes. (Please check this answer carefully, as this is rather outside my field. There may well be a much easier solution.)

In the paper "Borel Orderings" by Harrington, Marker and Shelah (Trans. AMS 310 (1988), 293-302, MR 0965754), the authors consider Borel (partial) orderings. A Borel ordering is said to be thin if there does not exist a perfect set of pairwise incomparable elements. Of course, in a total order there are no incomparable elements, so every Borel total order is thin. Corollary 3.2 states that a thin Borel order does not contain an $\omega_1$-chain. (The original reference given for this result is a 1982 paper of Harrington and Shelah, "Counting equivalence classes for co-$\kappa$-Souslin relations", MR 673790, text at ScienceDirect, for which I lack the appropriate subscription.)

Given this, suppose $\preceq$ is a Borel total order on the Polish space $X$ and $A \subset X$ is convex. Using Zorn's lemma, choose a subset $C \subset A$ which is well-ordered by $\preceq$ and cofinal in $A$ (see Well-ordered cofinal subsets). If $C$ is uncountable then it contains a subset with the order type of $\omega_1$, which is forbidden by the Harrington–Shelah result. So $C$ is countable. Fix some $x_0 \in A$; then we have $A \cap [x_0,\infty) := \{x \in A: x \succeq x_0\} = \bigcup_{y \in C} [x_0,y]$. (The $\subseteq$ direction is because $C$ is cofinal, and the $\supseteq$ direction is because $A$ is convex.) This is a countable union, and each closed interval $[x_0,y]$ is Borel. So $A \cap [x_0, \infty)$ is Borel. By applying the same argument to the reverse ordering, we also get that $A \cap (-\infty, x_0]$ is Borel. The union of these two sets is $A$, so $A$ is Borel.

$\endgroup$
4
  • $\begingroup$ Bravo! ${{{{}}}}$ $\endgroup$ Sep 21, 2015 at 19:42
  • $\begingroup$ @NateEldredge: Yes, thank you very much for this. I was actually about to post a comment myself that I have read somewhere that every Borel-totally ordered space $X$ admits no copies of $\omega_1$, since there exists a countable ordinal $\xi$ such that $X$ can be embedded into $2^\xi$. [I don't know whether this embedding is just an order embedding or has additional measurable-structure-respecting properties.] So then, by precisely the argument in your comment and in your answer, we have the desired result. $\endgroup$ Sep 21, 2015 at 20:06
  • $\begingroup$ @JulianNewman: In fact, maybe we can give a more direct proof by embedding $X$ into $2^\xi$, which I think would let us construct the desired countable cofinal set more explicitly. $\endgroup$ Sep 21, 2015 at 20:14
  • $\begingroup$ @NateEldredge: Yes, I have had precisely this thought - in particular, the "well-ordered cofinal subsets" argument probably requires at least $\omega_1$-dependent choice, whereas perhaps the embedding of $X$ into $2^\xi$ doesn't need anything like this. $\endgroup$ Sep 22, 2015 at 0:00

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.