3
$\begingroup$

Let $X\neq\emptyset$ be a set and let $\mu:{\cal P}(X)\to [0,1]$ be a probability measure. Is there a probability measure $$\bar{\mu}:{\cal P}({\cal P}(X))\to [0,1]$$ with the following property?

For all $S\subseteq X$ we have $\bar{\mu}({\cal P}(S)) =\mu(S)$.

$\endgroup$
2
  • $\begingroup$ Still wondering whether it is doable for $X=\mathbb{N}$? $\endgroup$ Jun 25, 2019 at 16:26
  • 1
    $\begingroup$ If the measure is atomic (as it necessarily is in the case of $\mathbb N$), just define $\bar\mu({{x}})=\mu({x})$ and set all other subsets to have measure 0. $\endgroup$ Jun 25, 2019 at 17:12

2 Answers 2

5
$\begingroup$

For notational and conceptual simplicity, assume that $X$ and $P(X)$ are disjoint.

As Bugs Bunny implicitly suggested: Let $\mathcal E = \{\{n\}: n \in X \}$ be the set of singletons. This is a subset of $P(X)$. The measure $\bar\mu$ will concentrate on this set.

For any $\mathcal A \subseteq P(X)$, we will have $\bar \mu(\mathcal A)=\bar \mu(\mathcal A\cap\mathcal E)$; note that $\mathcal A\cap\mathcal E$ is morally the same as a subset of $X$, so $\mu$ will measure it.

Formally, let $S_{\mathcal A} = \{n\in X: \{n\}\in \mathcal A\}$ be the ``singleton support'' of $\mathcal A$. Define $\bar \mu (\mathcal A) := \mu(S_{\mathcal A})$.

$\endgroup$
5
$\begingroup$

If $|X|>2$, then there is a single solution: $\overline{\mu}(\{x\})=\mu(x)$ and $\overline{\mu}(S)=0$ for any other set $S$.

Indeed, let $a=\overline{\mu}(\emptyset)$. By the condition for $S=\{x\}$, $\overline{\mu}(\{x\}) =\mu (x)-a$. By the condition for $S=\{x,y\}$ with $x\neq y$, $\overline{\mu}(\{x,y\}) =a$.

Now if $|X|>2$, there are at least $1+|X|$ two-element subsets. Thus, $\sum_{|S|\leq 2} \overline{\mu}(S) \geq a+ \sum_{x\in X}\mu (x) =1+a$. This forces $a=0$.

Now it is easy to show that for all larger subsets $\overline{\mu}(S)=0$.

$\endgroup$
3
  • 1
    $\begingroup$ I am afraid I cannot follow the argument. What is $x$? Is it an element of $X$? Then I do not see how to apply the map $\mu$ to it. Or is it a set with $1$ element? Then $\mathcal{P}(x)=\{\varnothing, x\}\neq \{x\}$. Moreover, it would be helpful if you could explain what you mean by the phrase "no measure [...] left". (Some probabilities can be zero I guess.) $\endgroup$ Jun 25, 2019 at 16:24
  • $\begingroup$ Grateful for the nice answer in the finite case, what about $X$ infinite? (I am especially curious for $X=\omega$.) $\endgroup$ Jun 25, 2019 at 16:24
  • $\begingroup$ I will edit my answer: I got it slightly off... $\endgroup$
    – Bugs Bunny
    Jun 25, 2019 at 19:07

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.