In my probability notes, I have the following definition and theorem:

Let $(\Omega_1,\mathcal{A}_1,P_1)$ be a probability space and $(\Omega_2,\mathcal{A}_2)$ a measure space.

Definition: We say that $p:\Omega_1\times\mathcal{A}_2\rightarrow [0,1]$ is a transition probability if: $$ \forall\,\omega_1\in\Omega_1\;\;p(\omega_1,\cdot):\mathcal{A}_2\rightarrow [0,1]\text{ is a probability,} $$ $$ \forall\,A\in\mathcal{A}_2\;\;p(\cdot,A):\Omega_1\rightarrow [0,1]\text{ is measurable.} $$

$ $

Theorem: Given the above notations, there exists a unique probability $Q$ in $(\Omega_1\times\Omega_2,\mathcal{A}_1\otimes \mathcal{A}_2)$ such that $$Q(A_1\times A_2)=\int_{A_{1}}p(\omega_1,A_2)\,P_1(d\omega_1),\;\forall\,A_1\in\mathcal{A}_1,\,A_2\in\mathcal{A}_2.$$ Moreover, for any measurable function $X:\Omega_1\times\Omega_2\rightarrow\mathbb{R}$ nonnegative or $Q$-integrable one has $$\int_{\Omega_1\times\Omega_2}X\,dQ=\int_{\Omega_1}\int_{\Omega_2}X(\omega_1,\omega_2)\,p(\omega_1,d\omega_2)\,P_1(d\omega_1).$$

I've always heard that Fubini "says" that the double integral is equal to the iterated integrals. Is the theorem that I wrote a generalization of Fubini's theorem (the particular case of $p(\omega_1,A)=p(A)=:P_2(A)$ and $Q(A_1\times A_2)=P_1(A_1)P_2(A_2)$)?


Yes. Replacing $p(\cdot, A):= P_2(A)$ with some measure $P_2$ and forcing $X \geq 0$ reduces your theorem to Tonellis Theorem. Fubinis Theorem extends this result for $Q$-integrable functions.

However I think you forgot to require that $(\Omega_2,\mathcal{A}_2)$ is a $\sigma$-finite measure space (otherwise you can construct counterexamples).

A proof of Tonellis Theorem and Fubinis Theorem can be found in the book Measure and integration theory from Heinz Bauer (Proposition 23.6 and corollary 23.7 ).


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.