2
$\begingroup$

Can one define on the measurable space $(\Omega = \{0,1\}^\mathbb{N}, P(\Omega))$ a measure $\mathbb{P}$ uniform in the sense that sets of sequences with $k$ fixed values should have measure $\dfrac{1}{2^k}$. Note that I require that the measure be defined on all of $P(\Omega)$. I also assume the axiom of choice to be true.

My guess would be it's not possible as my intuition was that since $\Omega$ is in bijection with $\mathbb{R}$, the measurable space looks a bit like $(\mathbb{R}, P(\mathbb{R}))$ which I think cannot have a translation invariant measure supporting Vitali sets for example.

I read this question: Uniform probability measure on $\{0,1\}^\omega$ but the link seems dead. Does Kolmogorov theorem indeed provides a positive answer to my question? I saw the other simple construction from Lebesgue measure but it supports only Lebesgue measurable sets, not all of $P(\Omega)$ as far as I understand.

$\endgroup$
  • 1
    $\begingroup$ Kolmogorov's theorem provides a measure on the product $\sigma$-algebra and not on $P(\Omega)$. $\endgroup$ – p4sch Mar 10 '18 at 22:01
2
$\begingroup$

Actually I suspect that the answer to the question exactly as you asked it is yes: If I recall correctly Lebesgue measure can be extended to a measure on all of $P(\Bbb R)$; what's impossible is a translation-invariant extension.

Which leads to a question you didn't quite ask: If you define an operation making $\{0,1\}$ into a group with identity $0$ then $\Omega$ is a (compact abelian) group. This gives an elegant way to describe the "uniform" measure on a subalgebra of $P(\Omega)$: it's just Haar measure. In fact this Haar measure cannot be extended to a translation-invariant measure on all of $P(\Omega)$.

If $G$ is an infinite group then there is no translation-invariant probability measure defined on all of $P(G)$.

Exactly like the standard proof of the existence of a non-Lebesgue-measurable set: Say $H$ is a countably infinite subgroup (for example the subgroup generated by any countably infinite subset), and let $C$ consist of one element from each coset. Then $$G=\bigcup_{h\in H}hC,$$so $\mu(C)=0$ implies $\mu(G)=0$, while $\mu(C)>0$ implies $\mu(G)=\infty$.

(If $G$ is not abelian I'll leave it to you to insert the words "left" and "right" in a few places so as to make sense of the above.)

$\endgroup$

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.