Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

This exercise appears in K.L.Chung's A Course in Probability Theory, Chapter 7.

Ex.7.1-4

Let ${X_j}$ be independent r.v.'s such that $\max_{1\leqslant j\leqslant n} \frac{|X_j|}{b_n} \to 0$ in pr. and $(S_n - a_n)/b_n$ converges to a nondegenerate d.f. Then $b_n \to \infty$, $\frac{b_{n+1}}{b_n} \to 1$, and $\frac{a_{n+1} - a_n}{b_n} \to 0$.

I found it difficult, and I do not have any idea why this is put in the exercise of CLT. Anyone helps me solve this? Thanks.

share|improve this question
    
I'm guessing you at least mean that $b_n\rightarrow\infty$? Also, could you edit and put your math within dollar signs so that we can read it more easily? –  Noah Stein Jun 11 '10 at 14:44
    
Probably it is in a chapter on the CLT because it is sort of like a converse for a CLT. That is to say, if your centering and scaling sequences $a_n$ and $b_n$ are such that a CLT-like statement holds about your sequence of independent random variables $X_j$, then the sequences $a_n$ and $b_n$ behave asymptotically as the corresponding sequences in the CLT do. –  Noah Stein Jun 11 '10 at 14:54
add comment

1 Answer

The fact that $b_n\to \infty$ is quite easy to check: if not, there is a $M$ and a subsequence $(b_{n'})$ which remains below $M$, hence $\frac 1{b_{n'}}\max_{1\leqslant j\leqslant n'}|X_j|\geqslant \frac 1{M}\max_{1\leqslant j\leqslant n'}|X_j|$, hence $\max_{1\leqslant j\leqslant n'}|X_j|$ would converge to $0$ in probability, which is not possible.

Let us denote the convergence in distribution by $\Rightarrow$.

Theorem 7.6 in Durrett's book Probability: theory and examples, (second edition), provides an useful theorem here:

The convergence of types theorem. Assume that $Y_n\Rightarrow Y$ and there are constants $\alpha_n>0$, $\beta_n$ such that $Y'_n=\alpha_nY_n+\beta_n\Rightarrow Y'$, where $Y$ and $Y'$ are not degenerate. Then there are $\alpha>0$ and $\beta\in\Bbb R$ such that $\alpha_n\to \alpha$ and $\beta_n\to \beta$.

The proof uses the fact that in case of convergence in distribution, the sequence of corresponding characteristic functions actually converges uniformly on compact sets. Then we proves that for $\alpha$, there is an unique positive real number which can be a candidate, and the same for $\beta_n$.

Here, we use this theorem with $W_n:=\frac{S_n-a_n}{b_n}$, $\alpha_n:=\frac{b_n}{b_{n+1}}$ and $\beta_n:=\frac{a_n-a_{n+1}}{b_{n+1}}$. It works since $\frac{X_{n+1}}{b_{n+1}}$ goes to $0$ in probability.

Finally, we have to use independence in order to identify the obtained limits. We compute the characteristic function of $W_n$, and of $W_{n+1}$, and we take the limits.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.