9
$\begingroup$

Let $\Omega$ and $\Gamma$ be two nonempty sets and $\mathscr{A}$ and $\mathscr{B}$ be $\sigma$-algebras over $\Omega$ and $\Gamma$, respectively. The product $\sigma$-algebra of $\mathscr{A}$ and $\mathscr{B}$ is defined to be the smallest $\sigma$-algebra over $\Omega\times\Gamma$ that contains the sets of the forma $A\times B$, for $A\in\mathscr{A}$ and $B\in\mathscr{B}$, and it is denoted by $\mathscr{A}\times\mathscr{B}$. When referring to measurable subsets of $\Omega$, $\Gamma$ or $\Omega\times\Gamma$, I will mean elements of the respective $\sigma$-algebras$.

Given a set $L\subseteq\Omega\times\Gamma$ and elements $\omega\in\Omega$ and $\gamma\in\Gamma$, the sections of $L$ by $\omega$ and $\gamma$ are, respectively, $$L_\omega=\left\{y\in\Gamma:(\omega,y)\in L\right\}\qquad\text{and}\qquad L^\gamma=\left\{w\in\Omega:(w,\gamma)\in L\right\}.$$

The following fact is very easily proved:

Fact: If $L\subseteq\Omega\times\Gamma$ is measurable, then all its sections are measurable.

Now, the converse is not true: If $\Omega$ is a set with cardinality $>2^{\aleph_0}$ equipped with the sigma algebra $\mathscr{P}(\Omega)$, the power set of $\Omega$, then the diagonal $\Delta_\Omega=\left\{(\omega,\omega):\omega\in\Omega\right\}$ is not in the product $\sigma$-algebra $\mathscr{P}(\Omega)\times\mathscr{P}(\Omega)$, but all its sections are (obviously) measurable. For a proof, see Does $\mathcal P ( \mathbb R ) \otimes \mathcal P ( \mathbb R ) = \mathcal P ( \mathbb R \times \mathbb R )$? and the reference given in the question.

What I want is a simpler example of it. Summarizing:

Question: Give an (simple) example of a non-measurable set in the product space s.t. all its sections are measurable.

All I could find out is that the product $\sigma$-algebra cannot be finite, so neither $\mathscr{A}$ nor $\mathscr{B}$ can.

$\endgroup$
8
$\begingroup$

Pick $\Omega$ to be any uncountable set and let $\mathcal{A}$ be the $\sigma$-algebra of subsets which are countable or have countable complement. Then there exists a non-measurable set $E\subset \Omega$, i.e., an uncountable set with uncountable complement. Let $L:=\{(e,e) : e\in E\}$ (i.e., the "$E$-diagonal"). The set $L$ has measurable sections in $\mathcal{A} \otimes \mathcal{A}$, because they are all either empty or points. The diagonal embedding $\omega \mapsto (\omega,\omega)$ of $\Omega$ into $\Omega \times \Omega$ is measurable (since preimages of measurable rectangles under this embedding are easily seen to be measurable), and since the preimage of $L$ under this embedding is the non-measurable set $E$, we see that $L$ itself is not measurable.

$\endgroup$
  • $\begingroup$ Perfect. This kind of non-separable space seems to appear a lot in counter-examples. $\endgroup$ – Luiz Cordeiro Dec 3 '13 at 0:47
  • $\begingroup$ Actually, there is a little problem with the argument since the projection of a measurable set is not always measurable. Not sure how to save the proof in the general case, but it can be modified by using the Borel $\sigma$-algebra instead of the abstract one given. $\endgroup$ – Lukas Geyer Dec 3 '13 at 3:51
  • $\begingroup$ I think the general argument is perfectly fine. The mapping $\omega\mapsto (\omega,\omega)$ from $\Omega$ into the diagonal of $\Omega\times \Omega$ is certainly measurable, so the inverse image of any measurable subset of the diagonal must be measurable. But this is just the projection. $\endgroup$ – user940 Dec 3 '13 at 4:13
  • $\begingroup$ @ByronSchmuland: Thanks, I got confused by overdosing on counterexamples, your argument works nicely, so I should probably edit it back. $\endgroup$ – Lukas Geyer Dec 3 '13 at 4:21
1
$\begingroup$

Here is an example on real line: There is a bijection $f: \mathbb{R} \rightarrow \mathbb{R}$ whose graph has full outer measure in plane. Such a function can be constructed by a simple transfinite induction.

Contrast this with the following: If $F: \mathbb{R}^2 \rightarrow \mathbb{R}$ is such that for each fixed $a$, the functions $y \mapsto F(a, y)$ and $x \mapsto F(x, a)$ are continuous, then $F$ is Borel. By previous example, one cannot replace "continuous" by "Borel" here.

$\endgroup$

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.