3
$\begingroup$

Suppose that $(S,\Sigma)$ is a measurable space with $S$ Polish and $\Sigma$ its Borel sigma algebra. Let $\mathcal{C}$ be the collection of discrete probability measures on $S$ having countably infinite many distinct point masses. For any $P \in \mathcal{C}$ there exists a set $F_P \in \Sigma^{\infty}$ of full $P^{\infty}$ measure with the property that for any $(s_1,s_2,...) \in F_P$ we have that $s_i$ is a point mass of $P$ for all $i$, and that the empirical probability measures for $(s_1,s_2,...)$ converge setwise (sometimes called strongly) to $P$. Let

$$G = \bigcup_{P \in \mathcal{C}} F_P $$

My question is this: is $G \in \Sigma^{\infty}$?

Additionally, is $G \in \Sigma^{\infty}$ if we change $\mathcal{C}$ to be the collection of all discrete probability measures on $S$?

Many thanks for any help on this.

$\endgroup$
  • $\begingroup$ Can you explain what you mean by setwise convergence of probability measures? $\endgroup$ – Anthony Quas Jun 22 '17 at 21:24
  • $\begingroup$ @Anthony Quas: $P_n \rightarrow P$ setwise means that $P_n(A) \rightarrow P(A)$ for all $A \in \Sigma$. $\endgroup$ – shanex Jun 22 '17 at 21:29
  • $\begingroup$ Can you explain what you mean by all discrete probability measures on $\mathcal S$ (how does this differ from the discrete probability measures on $\mathcal S$ having countably many distinct point masses?) $\endgroup$ – Anthony Quas Jun 22 '17 at 22:29
  • $\begingroup$ @Anthony Quas: by all I mean those discrete $P$ having finite or countably many distinct point masses. $\endgroup$ – shanex Jun 22 '17 at 22:32
  • $\begingroup$ OK. So when you say "countable", you mean countably infinite (as finite sets are generally also defined to be countable). My answer is fine in either case. $\endgroup$ – Anthony Quas Jun 22 '17 at 22:41
1
$\begingroup$

Define $\bar f_n(x)=\limsup_{N\to\infty} \frac 1N\#\{j\le N\colon x_j=x_n\}$, $\underline f_n(x)=\liminf_{N\to\infty} \frac 1N\#\{j\le N\colon x_j=x_n\}$, $g_n(x)=\prod_{j<n}\mathbf 1_{\{x_j\ne x_n\}}$.

Then define $F_n(x)=g_n(x)\cdot \mathbf 1_{\bar f_n(x)=\underline f_n(x)}\cdot \bar f_n(x)$. This function is the limiting density of terms that are equal to $x_n$ provided that (1) $x_n$ is the first time this point has appeared; (2) the limiting density of $x_n$'s exists.

Then $$ G=\left\{x\colon \sum_{n=1}^\infty F_n(x)=1\right\}\cap \bigcap_n\{x\colon \bar f_n(x)>0\}. $$

$\endgroup$
  • $\begingroup$ Thanks very much. But wouldn't $\{x:\Sigma_{i=1}^{\infty} F_n(x) = 1\}$ contain tuples $x$ having entries that could repeat only finitely often? No tuple in any $F_P$ has this property. $\endgroup$ – shanex Jun 23 '17 at 22:34
  • $\begingroup$ OK. I hadn't seen you wanted each $x_i$ to have positive mass. But then just intersect with $\bigcap_n\{x\colon \bar f_n(x)>0\}$. I will edit my answer to reflect this. $\endgroup$ – Anthony Quas Jun 23 '17 at 23:36

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.