1
$\begingroup$

Let $X$ be a compact metric space, and $\Theta$ be a finite space, endowed with their own $\sigma$-algebra.

Let $f \colon X \times \Theta \to \mathbb{R}$ be a Caratheodory function such that (1) for each $x \in X$, the function $f(x, \cdot) \colon \Theta \to \mathbb{R}$ is measurable; and (2) for each $\theta \in \Theta$, the function $f( \cdot, \theta) \colon X \to \mathbb{R}$ is continuous.

Given each $x \in X$, we have a probability distribution $\pi( \cdot \,| \, x) \colon 2^{\Theta} \to [0,1]$. In particular, given any fixed $x \in X$, it will generate a corresponding probability distribution $\pi$ on $2^\Theta$.

I am curious that

Under what kind of conditions (assumptions) imposed on this probability distribution $\pi$ , the map $$X \ni x \mapsto \int_\Theta f(x,\theta) \, \pi( \mathrm{d} \theta \,| \,x) \in \mathbb{R}$$ will be continuous on $X$?

Any idea or suggestions are most welcome!

Thank you so much!

$\endgroup$
  • $\begingroup$ Thanks @Michael, you’re right. Actually, the space $\Theta$ I considered was a compact metric space, but now I just want to simplify the question and restrict it to be finite. $\endgroup$ – Paradiesvogel Nov 18 '18 at 10:50
  • $\begingroup$ Thanks @Michael . I totally agree with you. In fact, I really need the probability distribution $\pi(\cdot | x)$ depending on $x$. This means for each $x \in X$, I have a different probability distribution defined on $2^\Theta$. Also, the space $\Theta$ is at least not trivial. In such a setting, what can I do to ensure the map $h(x)$ is continuous on $X$? Is it possible to do that? Thanks a million again :-) $\endgroup$ – Paradiesvogel Nov 18 '18 at 11:02
1
$\begingroup$

I think your integral is $$ h(x) = \sum_{\theta\in \Theta} f(x,\theta) \pi(\{\theta\}|x) \quad \forall x \in X $$ if $\pi(\{\theta\}|x) = \pi(\{\theta\})$ for all $x \in X$ then this is a sum of a finite number of functions that are continuous in $x$, and hence is continuous in $x$. More generally, if $\pi(\{\theta\}|x)$ is continuous in $x$ for each $\theta \in \Theta$, then this is a sum of a finite numer of functions that are continuous in $x$ (and hence is continuous in $x$).

Else, it is easy to get a discontinuous example (despite my incorrect comment from before that tried to do it with $\Theta$ being only a 1-element set) by defining $\pi(\{\theta\}|x)$ discontinuously. Define $X=[0,1]$, define $\Theta=\{0,1\}$, $f(x,0)=0$, $f(x,1) = 1$ for all $x \in [0,1]$, and define: $$ (\pi(\{0\}|x), \pi(\{1\}|x)) = \left\{ \begin{array}{ll} (1,0) &\mbox{ if $x \in [0,1/2)$} \\ (1/2,1/2) & \mbox{ if $x \in [1/2,1]$} \end{array} \right.$$ Then $$h(x)= \pi(\{1\}|x) = \left\{ \begin{array}{ll} 0 &\mbox{ if $x \in [0,1/2)$} \\ 1/2 & \mbox{ if $x \in [1/2,1]$} \end{array} \right.$$ and this is discontinuous in $x$.

$\endgroup$
  • $\begingroup$ Thanks so much for your answer @Michael. I should clarify initially that given any fixed $x \in X$, we have a different probability distribution; that can be reviewed as for each state variable $x$ in a State space $X$, I will have a different probability distribution on $\Theta$ and such a probability distribution depends on different observations of $x$. In this setting, could we still get some result for continuity of the map $h$? Would you mind rethinking about the question based on this setting please? I sincerely appreciate your kind help! $\endgroup$ – Paradiesvogel Nov 18 '18 at 11:16
  • $\begingroup$ I think this is what my example already does. Note that in this example I refined my answer to have $\Theta$ now a 2-element set rather than a 1-element set (for a 1-element set then the mass function must be 1 (in order to be a mass function) so I could not really define $\pi(\{\theta\}|x)$ discontinuously in $x$ when $\Theta$ is a 1-element set). $\endgroup$ – Michael Nov 18 '18 at 11:20
  • $\begingroup$ Thanks @Michael . Did you mean that a sufficient condition for the continuity of the function $h$ is $\pi(\{\theta\} | x)$ is continuous in $x$ for each $\theta \in \Theta$? Besides, may I ask what do you think if we extend the finite space of $\Theta$ to a compact metric space? Is it possible to do that? $\endgroup$ – Paradiesvogel Nov 18 '18 at 11:31
  • $\begingroup$ Yes, my first paragraph says it is sufficient to have $\pi(\{\theta\}|x)$ continuous in $x$ for each $\theta \in \Theta$, for the case when $\Theta$ is a finite set. $\endgroup$ – Michael Nov 18 '18 at 11:32
  • $\begingroup$ Thank you very much @Michael . I understand now. But still curious about what if the space $\Theta$ is allowed to be compact. In this setting, since the space $\Theta$ could be infinite or countably infinite, such a sufficient condition may fail. Do you think is there any reasonable assumption imposed on $\pi$ that would guarantee the continuity of $h$ with integral? Many thanks :-) $\endgroup$ – Paradiesvogel Nov 18 '18 at 11:41

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.