Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I am reading the book Kolmogorov A.N., Gnedenko B.V. Limit distributions for sums of independent random variables.

From the general theory there it is known that if $X_i$ are symmetric i.i.d r.v such that $P(|X_1|>x)=x^{-\alpha},\, x \geq 1$, then $(X_1+\ldots+X_n)n^{-1/\alpha}\to Y$, where c.f. of $Y$ equals $\varphi_Y(t)=e^{-c|t|^{\alpha}}, \alpha \in (0,2]$, so $Y$ has stable law of distribution.

I want to check it without using that general theorems. So I start as the following, $X_1$ has density of distribution $f_X(x)=|x|^{-\alpha-1}\alpha/2, |x|>1$. Using Levy theorem one must prove that $\varphi^n_{X_1}(t/n^{1/\alpha})\to \varphi_Y(t),\, n \to \infty$ for all $t\in \mathbb R$. $$\varphi_{X_1}(t/n^{1/\alpha})=\int_{1}^{\infty}\cos(tx/n^{1/\alpha})\alpha x^{-\alpha-1}\,dx,$$ for all it is evident that $t$ $\varphi_{X_1}(t/n^{1/\alpha})\to 1, n \to \infty$ so we have indeterminate form $1^\infty$.

So we are to find $n(\varphi_{X_1}(t/n^{1/\alpha})-1)$, but $\varphi_{X_1}(t/n^{1/\alpha})\sim 1+1/2(2txn^{-1/\alpha})^2$, and I can only say something about $\alpha=2$ and I got stuck here. Perhaps, I made a mistake somewhere.

Could you please help me? Thanks.

share|improve this question
1  
(+1) Thanks for showing your thoughts and work. –  cardinal Apr 22 '12 at 17:39
    
Maybe it would help others to specify which "general theorems" you are trying not to use. For example, you are appearing to admit usage of Levy's continuity theorem, which I would consider a pretty general one. :) –  cardinal Apr 22 '12 at 18:00
    
@cardinal By 'general theorems' I mean neccessary and sufficient conditions for the convergence of a sum of the form $\dfrac{1}{B_n}(X_1+\ldots+X_n-A_n)$ to the specific stable law. –  Tarasenya Apr 22 '12 at 18:02
add comment

2 Answers

up vote 1 down vote accepted

The problem is, in fact, not so hard as I thought. Note, that it is enough to prove that $\varphi(t)=1-Ct^{\alpha}+o(t^{\alpha}),t \to 0+$. First, I make substitution $y=tx$ in $\int_{1}^{\infty}\cos(tx)\alpha x^{-\alpha-1}\,dx=t^{\alpha}\int_{t}^{\infty}\cos (y) \alpha y^{-\alpha-1}\,dy$. Integrating by parts gives me that the last equals $\cos t-t^{\alpha}\int_{t}^{\infty}\sin y y^{-\alpha}\,dy=1-C t^{\alpha}+o(t^{\alpha})$, and it is a crucial thing that $\alpha \in (0,2]$ to write the last equality, here $C=\int_{0}^{\infty}\sin y y^{-\alpha}\, dy$. Hence, we are done.

share|improve this answer
add comment

I yr integral make the change of variables $z = \frac y {n^{\frac 1 {\alpha}}}$. This brings a factor $\frac 1n$ out front. The write $cos(tz) = 1 + (cos(tz) -1)$. Integrate the 1 explicitly, and the integral invlving $cos(tz)-1$ converges because it is nice at zero.

share|improve this answer
1  
Are you sure that you will get the desired result? –  Tarasenya Apr 22 '12 at 18:20
    
Never ! I think it's right but I didn't do the details. –  mike Apr 24 '12 at 18:53
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.