Let $Y$ be a locally compact Hausdorff topological space (further assumptions like metrizability, separability, etc., may be added if necessary) and let $\mathscr Y$ denote the Borel $\sigma$-algebra on it. Let $\Delta (Y)$ be the set of probability measures on $(Y,\mathscr Y)$ and endow it with the weak-$\star$ topology. This is, by definition, the weakest topology on $\Delta(Y)$ such that a net $(\mathbb P_{\alpha})$ in $\Delta(Y)$ converges to $\mathbb P\in\Delta(Y)$ if and only if $$\int_{y\in Y}f(y)\,\mathrm d\mathbb P_{\alpha}(y)\to\int_{y\in Y}f(y)\,\mathrm d\mathbb P(y)$$ for any bounded continuous function $f:Y\to\mathbb R$.

Now consider another measurable space $(X,\mathscr X)$ and a map $$\mathbb P:X\to\Delta (Y)$$ that is measurable, where the relevant $\sigma$-algebra on $\Delta(Y)$ is the Borel $\sigma$-algebra derived from the weak-$\star$ topology. Take any bounded measurable function $f:Y\to\mathbb R$, so that $f$ is integrable with respect to the probability measure $\mathbb P_x$ for each $x\in X$.

My question is as follows: given the measurability of the map $\mathbb P:X\to\Delta (Y)$, is the following map from $X$ to $\mathbb R$: $$x\mapsto\int_{y\in Y}f(y)\,\mathrm d\mathbb P_x(y)$$ measurable?

So far, I managed to reduce the problem to the following claim:

For any $c\in\mathbb R$ and any $B\in\mathscr Y$, the set $$\{\mathbb P\in\Delta(Y)\,|\,\mathbb P(B)\geq c\}$$ is in the Borel $\sigma$-algebra on $\Delta(Y)$ generated by the weak-$\star$ topology.

Is this claim true? I was thinking about using Lusin’s theorem to show it is, but to no avail so far.

Disclaimer: I also asked this question on Mathematics StackExchange. I repeat the question here in the hope that it can attract more attention and feedback.


If $Y$ is a metric space, then the arguments in the answers to The borel $\sigma-$algebra of the set of probability measures show that the map $I_f$ defined by $I_f(\mu) = \int f\,d\mu$ is measurable for any bounded measurable $f$. Then the answer to your question is affirmative, since your map is just $I_f \circ \mathbb{P}$. (The question there is about compact metric spaces, but compactness is not needed.)

If you drop metrizability, things get weird. For instance, if you let $Y = \omega_1+1$ where $\omega_1$ is the first uncountable ordinal, then the weak-* topology on $\Delta(Y)$ is not Hausdorff; if $\mu_1$ is the Dieudonne measure and $\mu_2$ is the point mass at the endpoint $\omega_1$, then $\int f\,d\mu_1 = \int f\,d\mu_2 = f(\omega_1)$ for every continuous $f$, so $\mu_1, \mu_2$ are topologically indistinguishable. Thus every Borel set in $\Delta(Y)$ contains both measures or neither. So if we take $f = 1_{\{\omega_1\}}$, which is a bounded Borel function on $Y$, then the map $\mu \mapsto \int f\,d\mu$ is not measurable, since $\int f\,d\mu_1 = 0 \ne 1 = \int f\,d\mu_2$. If you like, you may then take $X=Y$ and $\mathbb{P}$ the identity map.


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.