We have two probability spaces $(\Omega_1,\mathcal{F_1},P_1)$ and $(\Omega_2,\mathcal{F_2},P_2)$. Is it possible to construct probability space $(\Omega=\Omega_1\times\Omega_2,\mathcal{F},P)$ such that:

  1. $\forall A_1\in\mathcal{F_1}, A_2\in\mathcal{F_2}$ set $A_1\times A_2\in\mathcal{F}$ and its measure equals $P_1(A_1)P_2(A_2)$

  2. $\mathcal{F}$ contains all sets $A\subseteq\Omega$ such that $\forall \omega_1\ \{\omega_2:(\omega_1,\omega_2)\in A\}\in\mathcal{F_2}$ and $P_2(\{\omega_2:(\omega_1,\omega_2)\in A\})$ is constant for all $\omega_1$

  3. $P$-measure of such sets from (2) equals $P_2(\{\omega_2:(\omega_1,\omega_2)\in A\})$ (does not depend on $\omega_1)$

In other words, how to add to ($\Omega_1\times\Omega_2,\mathcal{F_1}\times\mathcal{F_2},P_1P_2)$ all sets from $\Omega$ with constant cross-$\Omega_2$ measure?

One can assume $\Omega_1=[0,1]$ with Lebesgue measure and $\Omega_2=[0,1]^\Gamma$ (infinite-dimensional cube).

This is needed for my research in Economics. Any advice on this problem will be greatly appreciated. Links to papers would be the best help. I am struggling with this question for several months already. Please, share any thoughts.


Consider the case $\Omega_1 = \Omega_2 = [0,1]$, equipping both factors with Lebesgue measure. Let $S$ be any subset of $[0,1]$. Then the set $$(S \times [0,1/2]) \cup (S^c \times (1/2,1])$$ is "cross-constant" in your sense --- each vertical slice is measurable and has measure $1/2$. So all such sets would belong to $\mathcal{F}$. But also $[0,1]\times [0,1/2]$ belongs to $\mathcal{F}$, and taking intersections with sets of the previous form yields $S \times [0,1/2] \in \mathcal{F}$. For every $S \subseteq [0,1]$.

Defining $\mu(S) = 2\cdot P(S \times [0,1/2])$ for all $S \subseteq [0,1]$ would give us a countably additive extension of Lebesgue measure to all subsets of $[0,1]$. Whether such a thing exists is set-theoretic. I believe the continuum hypothesis implies that such $\mu$ cannot exist. In any case, there is certainly no "natural" solution to the problem you pose.

  • $\begingroup$ (Minor typo in condition 2: surely you mean $A \subseteq \Omega$.) $\endgroup$ – Nik Weaver Dec 1 '14 at 4:59
  • $\begingroup$ Isn't it just the axiom of choice which excludes a "Lebesgue measure " on the power set of $[0,1]$? $\endgroup$ – Jochen Wengenroth Dec 1 '14 at 7:49
  • $\begingroup$ @Jochen: no, the axiom of choice excludes a translation-invariant measure on [0,1]. $\endgroup$ – Nik Weaver Dec 1 '14 at 13:06
  • $\begingroup$ I refer to a theorem of Banach and Kuratowski. $\endgroup$ – Nik Weaver Dec 1 '14 at 13:07
  • $\begingroup$ @NikWeaver, thank you for your answer - it is very helpful. While your idea seems obvious when written, I spent quite some time trying to get to an argument like this. Thank you again for your valuable help. $\endgroup$ – Andrei K Dec 1 '14 at 15:27

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.