5
$\begingroup$

Let $X$ be a topological vector space, let $f:X\to\Bbb R$ be a continuous linear function and let $P(X)$ denote the set of all Borel probability measures on $X$. For any $M\subseteq X$ we define the strong convex hull as $$ \operatorname{sco}(M) = \left\{\int y\;\mathrm d\nu: \nu\in P(X) \text{ and } \nu^*(M) =1\right\}. $$ Is that true that for any non-empty $M$ it holds that $\sup_M f = \sup_{\operatorname{sco(M)}}f$? In case the latter fact is true, I would be happy if someone can provide a reference to this.

I am not sure whether the integral is always defined over topological vector spaces, so the motivation was the case $X=P(A)$ where $A$ is a Borel space, and $P(A)$ is endowed with the topology of weak convergence. In the latter situation the integration is well-defined, but I guess that a similar result shall hold for a more general case as well. Feel free to correct me.

$\endgroup$
5
  • $\begingroup$ By "bounded linear function", do you mean continuous, or "bounded" as in $\bigl(\forall B \subset X\bigr)(B\text{ bounded } \Rightarrow f(B)\text{ bounded})$? $\endgroup$ – Daniel Fischer Jul 8 '13 at 17:53
  • $\begingroup$ I know very little about the field, but it seems you are looking for stuff in Choquet theory. $\endgroup$ – Michael Greinecker Jul 8 '13 at 19:22
  • $\begingroup$ What is $\nu^*$ ? $\endgroup$ – Bunder Jul 9 '13 at 6:50
  • $\begingroup$ @Bunder: an outer measure induced by the probability measure $\nu$ $\endgroup$ – Ilya Jul 9 '13 at 7:03
  • $\begingroup$ @Daniel: sorry, I meant a continuous function $\endgroup$ – Ilya Jul 9 '13 at 7:50
2
$\begingroup$

For continuous $f$, I think - I'm not sure enough about my measure theory to be near certain - that you can argue as follows:

$M \subset \operatorname{sco}(M)$ follows by choosing $\delta_x$ for $x \in M$ as the probability measure. Hence $\sup_M f \leqslant \sup_{\operatorname{sco}(M)} f$. Thus if $\sup_M f = \infty$, you have $\sup_{\operatorname{sco}(M)} f = \sup_M f$. And if $c = \sup_M f < \infty$, for any $\varepsilon > 0$, the open half space $H_\varepsilon := f^{-1}((-\infty,\, c+\varepsilon))$ is a neighbourhood of $M$, thus

$$x_\nu := \int_X y\,d\nu = \int_{H_\varepsilon} y\,d\nu \in \overline{H_\varepsilon} \subset f^{-1}((-\infty,\, c+\varepsilon])$$

for each $\nu \in P(X)$ with $\nu^\ast(M) = 1$. Letting $\varepsilon \to 0$ then shows $f(x_\nu) \leqslant c$, hence $\sup_{\operatorname{sco}(M)} f = \sup_M f$.

$\endgroup$
2
  • $\begingroup$ I think the idea is clear to me, yet indeed some technical details I have to check (as I said, we even don't talk here which integral do we use). I guess, you don't have an idea about a book where a proof of a similar fact can be found, do you? $\endgroup$ – Ilya Jul 9 '13 at 9:55
  • $\begingroup$ I'm afraid I can't name a book where something like that is treated. I expect that one can make the argument hold for every reasonable integral, by filling in the details, but as I said, I'm not sure about it. $\endgroup$ – Daniel Fischer Jul 9 '13 at 10:11

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.