1
$\begingroup$

Let $\mu_n$ and $\mu$ be Borel probability measures on a separable Hilbert space $X$. Assume that $\mu_n\to\mu$ vaguely. Is it necessarily true that $\mu_n\to\mu$ weakly?

I know that this is true if $X$ is a locally compact Polish space (because enlargements of compacta stay compact and we can get tightness, see below) . I have only seen the vague topology used on locally compact spaces so I suspect that things behave badly outside this setting.

A related question in case the answer is "no": denote the set of Borel probability measures on $X$ by $P(X)$. Is it at least true that the Borel sigma algebras generated from the vague and weak topologies on $P(X)$ are equal? I wanted to use something like the Lusin-Suslin theorem (11.4 here) but the vague topology is not complete when restricted to probability measures.

Proof when $X=\mathbb R^d$: let $K_1$ compact such that $\mu(K_1)>1-\epsilon$. Let $K_2=\{x:\|x - K_1\|\le \epsilon\}$, which is also compact. There exists a continuous function $0\le f\le 1$ with $f\equiv1$ on $K_1$ and $f\equiv0$ outside $K_2$. Then $$ \mu_n(K_2) \ge \int fd\mu_n \to \int fd\mu \ge \mu(K_1) $$ is larger then $1-2\epsilon$ for $n$ large. Since $K_2$ is compact $(\mu_n)$ is tight, has a weak limit, and that limit has to be $\mu$. This is easily extended to any locally compact space instead of $\mathbb R^d$ with a more painful construction of the set $K_2$.

$\endgroup$
1
$\begingroup$

The answer to both questions is no. In fact, for any $\mu_n,\mu\in P(X)$, $\mu_n\to\mu$ vaguely. This comes from the fact that $C_0(X)=0$.

To see this, assume $f\in C_0(X)$. For each $n$, there exists a compact set $K_n$ such that $|f|\le\frac1n$ outside $K_n$. Recall that compact sets in infinite dimensional Hilbert spaces have empty interior, and hence are nowhere dense. Defining $K:=\bigcup_nK_n$, we have that $K$ is nowhere dense by the Baire category theorem. But $f=0$ outside of $K$, and hence $f\equiv0$.

This is why you never see the vague topology discussed outside of the context of a locally compact space - without the assumption of local compactness, it may be entirely meaningless.

$\endgroup$
  • 2
    $\begingroup$ The conclusion is correct, but the argument is slightly wrong. Baire category says that $K$ has empty interior, but it need not be nowhere dense ($K$ need not be closed, and its closure could have nontrivial interior). But anyway you don't need Baire here: for each $n$ the set $\{|f| > 1/n\}$ is both open and contained in $K_n$, so it must be empty because, as you said, $K_n$ has empty interior. Hence you have $|f| \le 1/n$ everywhere, and $n$ was arbitrary. $\endgroup$ – Nate Eldredge Sep 20 '17 at 21:58
  • $\begingroup$ Thank you both for this discussion. I did not write this earlier because I did not have enough reputation for writing comments. $\endgroup$ – YZS Dec 12 '17 at 11:42

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.