$X_n$ is a sequence of random variables, and $\{\varphi_n(t)\}$ is the corresponding sequence of characteristic functions. Show that $X_n \stackrel{d}{\longrightarrow} 0$ iff $\{\varphi_n(t)\}$ converges to 1 in some neighbourhood of $t=0$.

The necessity follows immediately from Levy's continuity theorem. I am struggling with showing the sufficiency: it remains to show that $\varphi_n(t) \rightarrow 1$ for all $t \in \mathbb{R}$, but how to do that? Appreciate for any help!

  • 1
    $\begingroup$ Hint: what are all the moments? $\endgroup$ – Adam Hughes Jul 31 '14 at 5:05
  • $\begingroup$ Can you clarify which directions you're referring to as "necessity" and "sufficiency"? I'm a little confused. In particular, you don't need Levy's continuity theorem to prove "if $X_n \overset{d}{\to} 0$ then $\varphi_n \to 1$", only the definition of convergence in distribution. $\endgroup$ – Nate Eldredge Jul 31 '14 at 5:50

Suppose $X_n \overset{d}{\to} 0$. It should either be the definition of weak convergence, or a theorem you know, that $E f(X_n) \to f(0)$ for all bounded continuous $f$. Take $f(x) = e^{itx}$ to conclude $\varphi_n(t) \to 1$ for all $t$. (You don't need Lévy here.)

Suppose $\varphi_n(t) \to 1$ on some neighborhood of 0. Following the proof of the Lévy continuity theorem, conclude that $\{X_n\}$ is tight. Now use the double subsequence trick. If it is not the case that $X_n \overset{d}{\to} 0$, then there is a subsequence $X_{n_k}$ which does not have 0 as a limit point. (This uses the fact that convergence in distribution is metrizable.) By tightness, there is a further subsequence $X_{n_{k_j}}$ which converges in distribution to some random variable $X$ which is not 0. But now show that the chf of $X$ is 1 on a neighborhood of 0. Differentiate under the integral sign to show $EX^2 = 0$, a contradiction.

  • $\begingroup$ By sufficiency I mean showing that $\varphi(t) \overset{d}{\to} 1$ on some neighbourhood of $t=0$ implies $X_n \overset{d}{\to} 0$. I am confused how do you use continuity theorem to conclude that $\lim_{n \to \infty} \varphi_n(t)$ exists for all $t \in \mathbb{R}$? $\endgroup$ – math major Jul 31 '14 at 6:51
  • $\begingroup$ I guess the usual statement of the continuity theorem has as its hypothesis that $\varphi(t) := \lim_{n \to \infty} \varphi_n(t)$ exists for all $t$ and that $\varphi$ is continuous at 0. However, if you inspect the proof, you should find you only need the fact that the limit exists on a neighborhood of 0 and is continuous at 0. $\endgroup$ – Nate Eldredge Jul 31 '14 at 6:58
  • $\begingroup$ Yes, the statement of continuity theorem in the textbook that I am using is exactly the same as what you mentioned. I have gone through the whole proof again, but I can't see still why existence of $\varphi(t) = \lim_{n \to \infty}\varphi_n(t)$ on a neighbourhood of $t = 0$ implies the existence of $\lim_{n}\varphi_n(t)$ on the real line. Under the special condition that you gave, I can deduce that the distribution functions of $X_n$ are uniformly tight, which by Prokhorov's theorem implies existence of a weakly convergent subsequence, and I am stopped here .. $\endgroup$ – math major Jul 31 '14 at 8:00
  • $\begingroup$ Sorry, I claimed a little more than I actually know how to prove. But see my edit for something that does work. $\endgroup$ – Nate Eldredge Jul 31 '14 at 13:08
  • $\begingroup$ Ah, yes, I got what you mean now! Thanks so much! $\endgroup$ – math major Aug 2 '14 at 4:38

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.