Prove $a_nX_n+b_n \Rightarrow aX+b$ by means of characteristic functions

219 Views Asked by At

I want to solve the following exercise in Probability and Measure, Billingsly [1994]

According to Example 25.8, if $X_n \Rightarrow X$, $a_n \rightarrow a$ and $b_n \rightarrow b$, then $a_nX_n+b_n \Rightarrow aX+b$. Prove this by means of characteristic functions.

I believe we should use the continuity theorem for charateristic functions. Using rules for these, we can WLOG assume $b_n=0$ and just worry about $a_nX_n$. Then we need to show that $$ \varphi_{a_nX_n}(t) \rightarrow \varphi_{aX}(t) $$ for all $t \in \mathbb{R}$. Taking such a $t$, we slide in a "mixed term" have $$ \vert \varphi_{a_nX_n}(t) - \varphi_{aX}(t) \vert \leq \vert \varphi_{a_nX_n}(t) - \varphi_{aX_n}(t) \vert + \vert \varphi_{aX_n}(t) - \varphi_{aX}(t) \vert = \vert \varphi_{X_n}(a_nt) - \varphi_{X_n}(at) \vert + \vert \varphi_{X_n}(at) - \varphi_{X}(at) \vert, $$ where the the second term vanishes by the continuity theorem. However the first term is a bit more tricky as we have a "double" dependence on $n$.

Can anyone help me see how and why this term vanishes?

1

There are 1 best solutions below

0
On BEST ANSWER

You are on the right track by using the continuity theorem for characteristic functions. To show that the first term vanishes, you can use the fact that $a_n \rightarrow a$ and the uniform continuity of $\varphi_{X_n}$.

Since $X_n \Rightarrow X$, we have that $\varphi_{X_n}(t) \rightarrow \varphi_X(t)$ for all $t \in \mathbb{R}$. By Lévy’s continuity theorem, this implies that $\varphi_X$ is continuous. Since $\mathbb{R}$ is a metric space, this implies that $\varphi_X$ is uniformly continuous.

Now, let $\epsilon > 0$. Since $\varphi_X$ is uniformly continuous, there exists a $\delta > 0$ such that for all $s,t \in \mathbb{R}$ with $|s-t| < \delta$, we have $|\varphi_X(s) - \varphi_X(t)| < \epsilon/2$. Since $a_n \rightarrow a$, there exists an $N$ such that for all $n > N$, we have $|a_n - a| < \delta/|t|$ (assuming $t\neq 0$, otherwise both terms are equal to 1). Then for all $n > N$, we have $$|\varphi_{X_n}(a_nt) - \varphi_{X_n}(at)| = |\varphi_{X}(a_nt) - \varphi_{X}(at)| <\epsilon/2.$$ Thus, the first term vanishes as desired.