Let $\mu$ be a Borel probability measure on $\mathbb{R}^d$ and and let $(\mu_n : n \in \mathbb{N})$ be a sequence of such measures. Suppose that $\mu_n(f) \to \mu(f)$ for all smooth $f$ of compact support. How can I show that $\mu_n$ converges weakly to $\mu$ on $\mathbb{R}^d$?

I have tried to prove this result by contradiction and directly, however I am unable to make any serious progress on it. I know that if for any continuous bounded $g$ I could find smooth $f_k$ of compact support converging uniformly to $g$ then for any $\varepsilon > 0$ I have $$|\mu(g)-\mu_i(g)| \leq |\mu(g)-\mu(f_k)|+|\mu(f_k)-\mu_i(f_k)|+|\mu_i(f_k)-\mu_i(g)| $$ $$ \leq \varepsilon +|\mu(f_k)-\mu_i(f_k)|+\varepsilon$$ by uniform convergence and the fact that $\mu_i$ are probability measures so we can find some $k$ such that $\|g-f_k\|_\infty \leq \varepsilon$. Then we use the fact that $\mu_i(f_k) \to \mu(f_k)$ to get $|\mu(g)-\mu_i(g)| \leq 3\varepsilon$ for all large enough $i$, giving us weak convergence.

However I don't believe that there exist smooth $f_k$ of compact support converging uniformly to $g$ for this to work (or at least I can't prove it). I know the result is true if we had a compact subset of $\mathbb{R}^d$ by the Stone-Weierstrass theorem but that does not generalise to this case which leaves me stumped as I can't find any other way to attack the problem.

How should I proceed on this question?


Fix $\epsilon>0$. Since $\mu$ is a probability measure we can find $R>0$ such that $$\mu(\{y \in \mathbb{R}^d; |y| \leq R\}) \geq 1-\epsilon.$$ Let $\chi: \mathbb{R}^d \to [0,1]$ be a smooth function such that $\text{supp} \, \chi \subseteq B(0,2R)$ and $\chi(y)=1$ for $|y| \leq R$. As $$\lim_{n \to \infty} \mu_n(\chi) = \mu(\chi) \geq 1-\epsilon$$ we get $$\limsup_{n \to \infty} \int_{|y| \geq 2R} d\mu_n(y) \leq \limsup_{n \to \infty} \int (1-\chi(y)) \, \mu_n(dy) \leq \epsilon.$$This implies that we can find $M>0$ such that $$\sup_{n \in \mathbb{N}} \int_{|y| \geq M} \, \mu_n(dy) + \int_{|y| \geq M} \, d\mu(y) \leq 3 \epsilon. \tag{1}$$ This shows that the sequence $(\mu_n)_{n \in \mathbb{N}}$ is tight.

Now if $g$ is a bounded continuous function, choose a smooth function $f$ with compact support such that $\sup_{|y| \leq 2M} |f(y)-g(y)| \leq \epsilon$. Moreover, pick a smooth function $\varphi$ with compact support such that $0 \leq \varphi \leq 1$, $\varphi|_{B(0,M)}=1$ and $\varphi|_{B(0,2M)^c}=0$. Then

$$|\mu_n(g)-\mu(g)| \leq |\mu_n(g \cdot \varphi)-\mu(g \varphi)| + |\mu_n(g(1-\varphi))- \mu(g(1-\varphi))|.$$

For the first term you can proceed as you suggested in your question (because $g \cdot \varphi$ has compact support). By $(1)$, the second term on the right-hand side is bounded by

$$3 \|g\|_{\infty} \epsilon.$$

Combining both considerations we get

$$\lim_{n \to \infty} |\mu_n(g)-\mu(g)|=0.$$

| cite | improve this answer | |

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.