2
$\begingroup$

Suppose that $X$ and $Y$ are compact metric spaces. A Borel probability measure $\mu$ on $X\times Y$ satisfies $$ \mu(A\times B)=\int_A\mu(B|x)\mu_X(dx), $$ for $A$ and $B$ Borel sets in $X$ and $Y$ respectively, where $\mu(\cdot|x)$ is a transition probability measure, i.e., a measurable map $x\mapsto\mu(\cdot|x)$ from $X$ to the set of Borel probability measures on $Y$ (with the topology of weak convergence of measures) and where $\mu_X$ is the marginal of $\mu$ on $X$. The transition measure $\mu(\cdot|x)$ is continuous if the map $x\mapsto\mu(\cdot|x)$ is continuous. Note that here the map $x\mapsto\mu(\cdot|x)$ is a map between compact metric spaces.

Now suppose that $X=X_1\times X_2$ and $Y=Y_1\times Y_2$ ($X_1$, $X_2$, $Y_1$, $Y_2$ compact and metric). A transition probability measure $\mu(\cdot|x)$ is a product transition probability measure if $$\mu(B_1\times B_2|(x_1,x_2))=\mu_1(B_1|x_1)\mu_2(B_2|x_2)$$ for $B_1$ and $B_2$ Borel subsets of $Y_1$ and $Y_2$ respectively, where $\mu_1(\cdot|x_1)$ is a transition probability measure with source $X_1$ (with its Borel sets) and target $Y_1$ (with its Borel sets) and $\mu_2(\cdot|x_2)$ is a transition probability measure with source $X_2$ (with its Borel sets) and target $Y_2$ (with its Borel sets).

Endow the set $\mathfrak C$ of continuous transition probability measures with the topology of uniform convergence (which renders the space metric, complete and separable), and let $\mathfrak C^*$ be the subspace of continuous product transition probability measures, which is also complete and separable.

Let $\{\mu_n(\cdot|x)\}$ be an arbitrary sequence of continuous product transition probability measures and define for each $n$ a Borel probability measure $\nu_n$ on $\mathfrak C^*$ by $$ \nu_n:=\frac{1}{n}\delta_{\mu_1}+\cdots+\frac{1}{n}\delta_{\mu_n}, $$ where $\delta_{\mu_k}$ denotes the Dirac measure with pointmass on the transition measure $\mu_k$. The measure $\nu_n$ induces a (continuous) transition probability measure $\nu_n(\cdot|x)$ defined by $$\nu_n(B|x):=\frac{1}{n}\mu_1(B|x)+\cdots+\frac{1}{n}\mu_n(B|x).$$ Suppose that this transition measure converges uniformly to a (continuous) transition measure $\nu(\cdot|x)$. Is there a Borel probability measure $\lambda$ on $\mathfrak C^*$ such that for each $x\in X$ and $B$ a Borel set in $Y$, $$ \nu(B|x)=\int_{\mathfrak C^*}\mu(B|x)\lambda(d\mu)? $$

$\endgroup$

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.