6

Let $(X,\Sigma,\mu)$ be a measurable space with $\mu(X)<\infty$.

Prove that $\mu(\Sigma)$ is closed.

I've been stumped with this for quite a while. I've tried every usual way of showing a set is closed, to no avail. The elements in the preimage of $\mu$ are sets, meaning I can't extract subsequences like I would do with real numbers.

Gabriel Romon
  • 35,428
  • 5
  • 65
  • 157

1 Answers1

0

Here's one approach you can take. Since $\mu(X)<\infty$, there can only be countably many atoms in $X$ (modulo null sets), so let $A_1,A_2,\dots$ be all the atoms, let $A=\bigcup A_n$, and let $B=X\setminus A$. Then the restriction of $\mu$ to $B$ is atomless, and thus takes every value in the interval $[0,\mu(B)]$. Note that $\mu(\Sigma)$ is just the image of $\mu(\Sigma\cap\mathcal{P}(A))\times\mu(\Sigma\cap\mathcal{P}(B))$ under the addition map $\mathbb{R}\times\mathbb{R}\to\mathbb{R}$. To show that $\mu(\Sigma)$ is closed, it thus sufffices to show $\mu(\Sigma\cap\mathcal{P}(A))\times\mu(\Sigma\cap\mathcal{P}(B))$ is compact. Since $\mu(\Sigma\cap\mathcal{P}(B))=[0,\mu(B)]$ is compact, it suffices to show $\mu(\Sigma\cap\mathcal{P}(A))$ is compact as well.

But the restriction of $\mu$ to $A$ is atomic: every measurable subset of $A$ has the form $\bigcup_{n\in S} A_n$ for some $S\subseteq\mathbb{N}$, up to null sets. So $\mu(\Sigma\cap\mathcal{P}(A))$ is just the image of the map $m:\{0,1\}^\mathbb{N}\to\mathbb{R}$ defined by $$m(f)=\mu\left(\bigcup_{f(n)=1} A_n\right)=\sum_{f(n)=1}\mu(A_n).$$ But $m$ is continuous with respect to the product topology on $\{0,1\}^\mathbb{N}$ (for instance, by the dominated convergence theorem--this is where we use the hypothesis $\mu(X)<\infty$). Since $\{0,1\}^\mathbb{N}$ is compact, the image of $m$ is thus compact, and so $\mu(\Sigma\cap\mathcal{P}(A))$ is compact.

Eric Wofsey
  • 330,363
  • Since the restriction of $\mu$ to $B$ is atomless, I can prove easily that $\mu$ takes arbitrary small values. How do you prove that it can take any value in $[0,\mu(B)]$ ? – Gabriel Romon Sep 02 '16 at 06:05
  • See http://math.stackexchange.com/questions/254728/simpler-proof-non-atomic-measures – Eric Wofsey Sep 02 '16 at 06:12
  • thanks, my Googling skills suck – Gabriel Romon Sep 02 '16 at 06:17
  • How do you prove that the set of atoms is at most countable ? It doesn't seem trivial to me. I can prove the set of singletons with positive measure is countable, but I'm deadlocked when it comes to atoms. – Gabriel Romon Sep 02 '16 at 15:39
  • Any two atoms are either equal or disjoint, modulo null sets. If there were uncountably many different atoms modulo null sets, there would be uncountably many of measure $>1/n$ for some $n$ and then a union of countably infinitely many of those would have to have infinite measure since they are disjoint up to null sets. – Eric Wofsey Sep 02 '16 at 15:57