2
$\begingroup$

Let $(X, \mathcal{A}, \mu)$ and $(Y, \mathcal{B}, \nu)$ be two probability spaces and $\pi: X \rightarrow Y$ be a measure preserving function (not necessarily invertible). Can we conclude there exists a measurable set $Y'\subset Y$, $\nu(Y')=1$, such that $Y'\subset Img(\pi)$?

Can we conclude it using more assumptions? In my case, $Y = \{0,1\}^\mathbb{N}$, equipped with the product $\sigma$-algebra (of the discrete $\sigma$-algebra), and $\nu$ is the product Bernoulli measure. I think this probability space is not complete. The other probability space is anything.

In general, we don't know if $Img(\pi)$ is measurable or not (if it was, the question was over). But we know any set $S$ of positive measure won't be disjoint of $Img(\pi)$, because $\mu(\pi^{-1}(S)) = \nu(S) >0$, so $\pi^{-1}(S) \neq \emptyset$ and there exists $x \in X$ so that $\pi(x) \in S$.

Any help is appreciated. Thanks.

$\endgroup$
  • $\begingroup$ For the second question, it sounds like you have an application - maybe you could explain why you think you need this result? $\endgroup$ – Dap Jan 19 '18 at 20:28
  • $\begingroup$ Yes, I have. I'm asking this question because I am trying to check what are the properties of a (measure theoretical) semi-conjugacy between and "IID dynamical system" and a Bernoulli shift. Here, an "IID dynamical system" is to be understood as a measure preserving system $(f,\mu)$ having a non $\mu$-constant observable $\psi$ such that $(\psi \circ f^n)_n$ is IID (now, in purely probabilistic terms). $\endgroup$ – Lucas Amorim Jan 20 '18 at 16:51
2
$\begingroup$

Just answering the first question.

The important thing is whether the image of $\pi$ is Lebesgue measurable, i.e. has inner measure $1.$ If it has inner measure less than $1$ then such a $Y'$ cannot exist.

The answer to the first question is no. From Fremlin's Measure Theory section 235H, "The image measure catastrophe", with minor changes to notation to match yours:

Suppose, for instance, that $\nu$ is Lebesgue measure on $Y = [0, 1],$ that $X \subseteq [0, 1]$ is a non-Lebesgue-measurable set of outer measure $1$ (134D), that $\mu$ is the subspace measure $\nu_X$ on $X,$ and that $\phi(x) = x$ for $x \in X.$ Then $$\mu(\phi^{−1}(F)) = \nu^*(X \cap F) = \nu(F)$$ for every Lebesgue measurable set $F \subseteq Y.$

Since $\phi(X)=X$ is non-measurable, its inner measure is less than $1.$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Dap, I was just looking around for what an inner measure is because I was taught with Folland's book and this concept is not in his expository text. I took a look and found out that a simple case of what you said can be shown: Let $(X,\mathcal{A},\mu) = ([0,1] \setminus \{0.5\}, \mathcal{B}_{[0,1] \setminus \{0.5\}}, Lesbegue)$ completed and $(Y, \mathcal{B}, \nu) = ([0,1], \{ \emptyset, [0,1] \}, \nu(\emptyset)=0$ $\nu([0,1])=1)$. Let $\pi(x) =x$. It is measure preserving. The image of $\pi$ is not measurable and does not contain another measurable of $\nu$-measure 1. Thanks! :) $\endgroup$ – Lucas Amorim Jan 20 '18 at 16:44

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.