Let $X_1, X_2, \dots$ be independent, with

$$P(X_n=1) = P(X_n = -1) = \frac{1}{2} \left( 1 - \frac{1}{n^2} \right),$$


$$P(X_n = n) = P(X_n = -n) = \frac{1}{2n^2}.$$

Let $S_n = X_1 + \cdots + X_n$ and $S_n^* = S_n/\sqrt{n}$. The first part of the problem asked to show that $\text{Var}(S_n^*) \rightarrow 2$, which was a pretty straightforward calculation.

The second part asks us to show that $S_n^* \stackrel{d}{\underset{n\to\infty}{\longrightarrow}} Z$, where $Z$ is a normal random variable. I tried to do this by using characteristic functions. We have that

$$\varphi_{X_n}(t) = \cos t + \frac{1}{n^2}\left( \cos (nt) - \cos t \right).$$

However, writing the characteristic function of $S_n^*$ in terms of $\varphi_{X_n}(t)$ made it really messy and not workable much. Is there a more efficient way of approaching this problem? Would Lindeberg-Feller CLT work here?


The lindeberg CLT can be applied for this sequence of random variables.

Hint: figure out a convinient lower bound for the cumulative variance

$s_n^2=\sum_{i=1}^n \sigma_k^2$

  • $\begingroup$ I don't think Lindeberg-Feller CLT will give the result here. However, I did get the result by using characteristic functions. $\endgroup$ – symmetricuser Jan 27 '15 at 20:55

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.