I have been wondering how to define the product of conditional measures as defined by Renyi-Popper. I spell the details below.

If $(X,\Sigma)$ is a measurable space, then the function $\mu : \Sigma\times\Sigma \to [0,1]$ is called a Conditional Probability measure if it satisfies (For any $A,B,C \in \Sigma$):

  1. $\mu(A,B) \geq 0$; and $\mu(B,B)= 1$,

  2. For any $B\neq \emptyset$, $\mu(\cdot,B)$ is $\sigma$-additive

  3. $\mu(A\cap B,C) = \mu(A,B\cap C)\cdot \mu(B,C)$

It is a standard definition for the product for (non-conditional) probability function:

If $X$ and $Y$ are the support of measurable spaces and $\mu,\eta$ are probability measures over them, then the product measure over $X\times Y$ is given by:

$\lambda(U) = \int\mu(U_y)d\eta(y)= \int\eta(U_x)d\mu(x)$

where $U_x = \{y \in Y : (x,y)\in U\}$

On a similar way we could (try to) define the product of conditional probability measures as follows(Since I am more interested on the case where one of the spaces is finite and discrete I will make this assumption):

Let $X$ be a measurable space and $\mu$ a conditional probability measure over $X$. Let $Y$ be a finite discrete conditional probability space and $\eta$ a conditional probability measure over $Y$.

Define the following function: \begin{equation} \lambda(A,B) = \sum_{y\in Y}\mu(A_y,B_y)\cdot\eta(y,\pi_2(B))\quad (*)\end{equation}

where $\pi_2(B) = \{y \in Y:$for some $x\in X, (x,y)\in B\}$.

With this definition we have that if $U$ and $V$ are rectangles in $X\times Y$, i.e., $U = A\times B$ and $V = C\times D$, then $\lambda(U,V)= \mu(A,C)\cdot\eta(B,D)$.

Also the properties 1. and 2. from the definition of a conditional probability function hold.

However we have the following concerning property 3.:

On one hand:

$\lambda(A\cap B,C) = \sum_{y\in Y} \mu(A_y\cap B_y,C_y)\cdot \eta(y,\pi_2(C))$

On the other hand:

$\lambda(A,B\cap C)\cdot \lambda(B,C) = \sum_{y\in Y} \mu(A_y, B_y\cap C_y)\cdot \eta(y,\pi_2(B\cap C)\cdot \sum_{x\in Y}\mu(B_x,C_x)\cdot \eta(x,\pi_2(C))$.

Which I do not know if one can prove that an equality holds between both equations.


  1. If the definition $(*)$ makes sense, how one could prove that the product is a conditional probability measure?

  2. Does anyone know a reference for the product of conditional probability spaces?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.