Let $X:\mathbb{R}\rightarrow\mathbb{R}$ be a non-constant function, measurable with respect to the Borel-Algebra $\mathcal{B}$ and $\sigma(X)$ the sigma-algebra generated by $X$. Let $\mathcal{A}\subsetneqq\sigma(X)$, $\mathcal{A}\neq\{\emptyset, \mathbb{R}\}$ be a sub-sigma algebra of $\sigma(X)$.


Does there always exist a $\mathcal{B}$-measurable function $Y:\mathbb{R}\rightarrow\mathbb{R}$ such that

  • $\sigma(X)\cap\sigma(Y)=\mathcal{A}$
  • But neither is $\sigma(X)\subset\sigma(Y)$ nor $\sigma(Y)\subset\sigma(X)$.

It was easy for me to find a few specific examples where $\sigma(X)$ has only a finite number of elements, but I am interested in the general case.

EDIT: The example given by Eric answers the question as stated. But due to his comment and answer I noticed that I actually would like to understand something slightly(?) different. I am really interested in the case where $\mathcal{A}$ is also generated by a function, i.e. $\mathcal{A}=\sigma(f\circ X)$. The countable-uncountable sigma-algebra is not generated by a function since it is not countably generated (see here) but a sigma-algebra generated by a function is.

  • 1
    $\begingroup$ Your "equivalent formulation" seems quite different from the original question. How does $f(X)=g(Y)$ say that $\sigma(X)\cap\sigma(Y)=\mathcal{A}$? $\endgroup$ Jul 11, 2016 at 20:50
  • $\begingroup$ Yes, Thanks for pointing this out! I naively assumed that every $\mathcal{A}$ being a sub-algebra were also generated by a function. Now that I start to understand your answer, I see this is not correct. I need to think about this. Probably I need to edit/revise the original question as well. $\endgroup$
    – g g
    Jul 12, 2016 at 8:40

1 Answer 1


This is certainly not always possible. For instance, if $X$ is the identity function, then $\sigma(X)=\mathcal{B}$, so for any Borel-measurable $Y$, $\sigma(Y)\subseteq\sigma(X)$.

For a slightly less trivial example, let $X$ be given by $X(t)=t$ unless $t=0$ and $X(0)=1$. Then $\sigma(X)$ is the algebra of all Borel sets that can't distinguish $0$ and $1$. Let $\mathcal{A}$ consist of all elements of $\sigma(X)$ that are either countable or cocountable. If $Y$ is Borel-measurable and $\mathcal{A}\subset\sigma (Y)$, then $Y$ must be injective on $\mathbb{R}\setminus\{0\}$; in particular, all the fibers of $Y$ must be countable.

Now let $$r=\inf\{r\in\mathbb{R}:Y^{-1}((r,\infty))\text{ is countable}\}$$ and $$s=\inf\{s\in\mathbb{R}:Y^{-1}((-\infty,s))\text{ is countable}\}.$$ Then clearly $s\leq r$, and in fact $s<r$ since if $r=s$ then the fact that $Y^{-1}(\{r\})$ is countable would imply $Y^{-1}(\mathbb{R})$ is countable. Choose $t\in(s,r)$ and let $S=(-\infty,t)\setminus\{Y(0),Y(1)\}$. Then $Y^{-1}(S)$ is Borel, is neither countable nor cocountable, and does not distinguish $0$ and $1$. Thus $Y^{-1}(S)\in\sigma(X)\setminus\mathcal{A}$. In particular, this shows that $\sigma(Y)\cap\sigma(X)$ cannot be equal to $\mathcal{A}$ for any Borel-measurable $Y$.


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.