Let $h,g$ be continuous, $g > 0$, and $$\lim\limits_{|x|\to \infty} \frac{|h(x)|}{g(x)} = 0$$ Let $X_n \xrightarrow[]{D} X$ and let $\mu$ denote the probability measures induced by $X_n, X$, respectively. Suppose that $\int g d\mu_n \leq C < \infty$. Then show that $\int h d\mu_n \to \int h d\mu$.

What I have so far:

First note that by the given condition, for every $n$, $h\in L^1(\mu_n)$ since $|h| \leq kg \in L^1(\mu_n)$, for some $k$. For any $\epsilon > 0$, we choose $M$ large enough, $$0 \leq \left\lvert \int h d\mu_n - \int h d\mu\right\rvert \leq \left\lvert \int_{|x| > M} h d\mu_n \right\rvert + \left\lvert \int_{|x| \leq M} h d\mu_n - \int_{|x| \leq M} h d\mu \right\rvert + \left\lvert \int_{|x| > M} h d\mu \right\rvert \\ \leq \int_{|x| > M} \left\lvert h \right\rvert d\mu_n + \left\lvert \int_{|x| \leq M} h d\mu_n - \int_{|x| \leq M} h d\mu \right\rvert + \int_{|x| > M} \left\lvert h \right\rvert d\mu \\ \leq \int_{|x| > M} \frac{\left\lvert h \right\rvert}{g} g d\mu_n + \left\lvert \int_{|x| \leq M} h d\mu_n - \int_{|x| \leq M} h d\mu \right\rvert + \epsilon \\ \leq \epsilon \int_{|x| > M} g d\mu_n + \left\lvert \int_{|x| \leq M} h d\mu_n - \int_{|x| \leq M} h d\mu \right\rvert + \epsilon\ \\ \leq \epsilon \sup\limits_n \int g d\mu_n + \left\lvert \int_{|x| \leq M} h d\mu_n - \int_{|x| \leq M} h d\mu \right\rvert + \epsilon \\ \leq \epsilon C + \left\lvert \int_{|x| \leq M} h d\mu_n - \int_{|x| \leq M} h d\mu \right\rvert + \epsilon $$ where in the second line, I assumed integrability of $h$ w.r.t. $\mu$ (not sure if I can assume that) and in the third line by absolute continuity of integral since $\{|x| > M\} \to \emptyset$, which has measure 0. I can't seem to bound the middle term though. My motivation for this splitting was to somehow use continuity of $h$ on a compact interval to invoke boundedness.

  • $\begingroup$ For $M>0$ let $\chi_M: \mathbb{R} \to [0,1]$ be a continuous function such that $\chi_M(x)=1$ for all $|x| \leq M$ and $\chi_M(x)=0$ for all $|x|>M+1$. Write $$\int h = \int h \chi_M + \int h (1-\chi_M)$$ and use a very similar reasoning as you did in your computations to estimate each of the integrals; instead of $$\left| \int_{|x| \leq M} h \, d\mu_n- \int_{|x| \leq M} h \, d\mu \right|$$ you will get $$\left| \int \chi_M h \, d\mu_n - \int \chi_M h \, d\mu \right|$$ which converges to $0$ as $n \to \infty$ because of the weak convergence of $\mu_n$. $\endgroup$ – saz Dec 5 '17 at 7:22
  • $\begingroup$ Well, h is bounded on compacts and therefore $f \chi_M$ is bounded. $\endgroup$ – saz Dec 5 '17 at 11:55
  • $\begingroup$ Ah right and the boundedness on $[-M-1,M+1]^c$ comes from the fact that the product is 0 right? $\endgroup$ – James Yang Dec 5 '17 at 12:02
  • $\begingroup$ Yes, exactly... the function $f \chi_M$ has compact support. $\endgroup$ – saz Dec 5 '17 at 12:04

I will give s sketch of the proof. Your attempt above shows that you can complete this argument. To begin with the hypothesis does imply that h is integrable with respect to $\mu$. Instead of splitting integrals as above do the following: fix a positive integer N and choose a continuous function F which is 1 on $[-N,N]$, 0 outside $(-N-1,N+1)$ with its range contained in $[0,1]$. Write $ \int h d\mu_n$ as $ \int hF d\mu_n+\int h(1-F) d\mu_n$ and do a similar thing for $\int h d\mu$. $\int hF d\mu_n$ converges to $\int hFd\mu$ because hF is a bounded continuous function. To handle $\int h(1-F) d\mu_n$ bound it by $\int_{\mathbb R\setminus [-N,N]} h d\mu_n$ and use the argument you have used in your proof.


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.