Consider a finite sequence of $\sigma$-finite measure spaces $(\Omega_i, \mathcal{F}_i, \mu_i)$. Constructing the product measurable space $$ (\Omega_1 \times \cdots \times \Omega_n, \mathcal{F}_1 \otimes \cdots \otimes \mathcal{F}_n), $$ we can define the product measure $\mu = \mu_1 \otimes \cdots \otimes \mu_n$ such that $$ \mu(A_1 \times \cdots \times A_n) = \mu_1(A_1) \cdots \mu_n(A_n) $$ whenever $A_i \in \mathcal{F}_i$.

Is it that the product measure only behaves like this when measuring products? E.g., let $n=2$ and consider the "triangle" set $A = \{(x,y): x \in (0,a], y \in (0,x]\} \subset \mathbb{R}^2$ and the Borel $\sigma$-algebra $\mathcal{B}$. Perhaps naively I would put $A_1 = (0,a]$ and $A_2 = (0,x]$, but then I don't think that $A = A_1 \times A_2$. I would then try to measure $\mu(A) = \mu_1(A_1)\mu_2(A_2)$, which I would want to be $\frac{1}{2}a^2$. What would be the proper way to measure non-cartesian product sets?

  • $\begingroup$ The measure of such a set can be calculated using Fubini's theorem (also known as Cavallieris principle in this context). $\endgroup$ – PhoemueX Jun 30 '15 at 17:36

You're correct that the you can only apply the rule $\mu(A) = \mu_1(A_1) \mu_2(A_2)$ when $A = A_1\times A_2.$ A triangle does not meet this criteria. However, you can create sets of rectangles that tell us what the .

So, for the sake of convenience, let $a=1$ and cut the square $[1,0] \times [1,0]$ up into $n^2$ equal squares in the usual way. [Perhaps first convince yourself that whether a square contains its boundary or not doesn't change its measure.] I claim that each such square has measure $1/n^2$, by the definition of the product measure. Furthermore, there is a collection of $n(n+1)/2$ squares that cover the triangle, and a collection of $n(n-1)/2$ squares that are covered by the triangle.

Therefore, $$\frac{1}{n^2} \frac{n(n-1)}{2} = \frac{1}{2} -\frac{1}{2n} \leq \mu(A) \leq \frac{1}{n^2} \frac{n(n+1)}{2} = \frac{1}{2} +\frac{1}{2n}$$ so we may conclude that $\mu(A) = 1/2$, as this must be true for all $n$.

More generally, as mentioned by PhoemueX, you can calculate measures using Fubini's theorem by working with indicator functions, but you can also do these sorts of decompositions to get a sense of what's going on.


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.