I came across this problem:


If I have $X_1, X_2, ..., X_n$ $n$ iid random variables which pdf is $$ f_X(x) = \begin{cases} \dfrac{x^{\mu-1} e^{-x}}{\Gamma{(\mu)}} &0<x<\infty, \\0&\text{elsewhere}\end{cases} $$ namely a $\text{Gamma}(\alpha=\mu, \beta=1)$ distribution. Let $n\bar{X}_n= X_1+X_2+\cdots+X_n$. I need to prove that $$T_n\stackrel{\text{dist}}{\stackrel{n\to\infty}{\longrightarrow}}N(0,{1}/{\mu})$$ where $$T_n= \frac{\sqrt{n}(\bar{X}_n-\mu)}{\bar{X}_n}$$

My Solution:

I thought of a possible solution in two steps:

  • First, we need to find the pdf of $\bar{X}_n$ and then of $T_n$.
  • Then we take the limit of it and if we get a Normal distribution then, we solved the question.

\begin{align*} F_T(t) &= P(T \le t) \\& = P(\frac{\sqrt{n}(\bar{X}_n-\mu)}{\bar{X}_n} \le t) \\& = P(\frac{(\bar{X}_n-\mu)}{\bar{X}} \le \frac{t}{\sqrt{n}}) \\& = P(1- \frac{\mu}{\bar{X}_n} \le \frac{t}{\sqrt{n}}) \\& = P(\frac{\mu}{\bar{X}_n} \ge 1- \frac{t}{\sqrt{n}}) \\& = P(\frac{\bar{X}_n}{\mu} \le \frac{\sqrt{n}}{\sqrt{n}-t}) \\& = P(\bar{X}_n \le \frac{\mu\sqrt{n}}{\sqrt{n}-t}) \end{align*}

Now, I should do the integration $\int_0^\frac{\mu\sqrt{n}}{\sqrt{n}-t}$ of the pdf of $\bar{X}_n$. But it is not the same distribution as $X_i$. It is something else. This is where I stuck in my solution.

To clarify: My goal is to prove $T_n\stackrel{\text{d}}{\longrightarrow}Normal$, not finding the distribution of $\bar{X}_n$.

Any help will be appreciated!

  • $\begingroup$ @Xi'an Is it $Ga(\mu,n)$? $\endgroup$ – iTurki Dec 27 '14 at 20:24
  • $\begingroup$ yes indeed. But given your final question you should rather check whether or not the Central Limit Theorem applies to your setting. $\endgroup$ – Xi'an Dec 27 '14 at 20:26
  • $\begingroup$ @Xi'an Can you clarify, please? How to check if it applies or not? $\endgroup$ – iTurki Dec 27 '14 at 20:27
  • $\begingroup$ this is a self-study question: start with your favourite textbook or if nothing else with the Wikipedia entry on the CLT. $\endgroup$ – Xi'an Dec 27 '14 at 20:30
  • $\begingroup$ once you have gone through those basics, check for Slutsky's theorem. $\endgroup$ – Xi'an Dec 27 '14 at 20:35

This problem is a direct application of Slutsky's theorem:

If $X_n$ converges in distribution to a random element $X$ and if $Y_n$ converges in probability to a constant $c$, then $X_n/Y_n$ converges in distribution to $X/c$.

In your setting, the theorem numerator "$X_n$" is replaced with $\sqrt{n}(\bar{X}_n-\mu)$, since, by virtue of the Central Limit Theorem, $$\sqrt{n}(\bar{X}_n-\mu)\stackrel{\text{dist}}{\stackrel{n\to\infty}{\longrightarrow}}\mathcal{N}(0,\text{var}(X_i))$$ and here $\text{var}(X_i)=\mu$, hence $$\sqrt{n}(\bar{X}_n-\mu)\stackrel{\text{dist}}{\stackrel{n\to\infty}{\longrightarrow}}\mathcal{N}(0,\mu)\,.$$ And the theorem denominator "$Y_n$" is replaced with $\bar{X}_n$ which converges in probability to its expectation $\mu$ by the weak law of large numbers. Hence, $$\dfrac{\sqrt{n}(\bar{X}_n-\mu)}{\bar{X}_n}\stackrel{\text{dist}}{\stackrel{n\to\infty}{\longrightarrow}}\mathcal{N}(0/\mu,\mu/\mu^2) $$i.e. $$\dfrac{\sqrt{n}(\bar{X}_n-\mu)}{\bar{X}_n}\stackrel{\text{dist}}{\stackrel{n\to\infty}{\longrightarrow}}\mathcal{N}(0,1/\mu)$$ Q.E.D.

| cite | improve this answer | |
  • $\begingroup$ Thanks for the answer. I though $\bar{X}$ will converge to a variable. I was Wrong. Thanks again! $\endgroup$ – iTurki Dec 27 '14 at 21:28
  • $\begingroup$ I am quite sorry you could not find the solution by yourself... $\endgroup$ – Xi'an Dec 27 '14 at 21:30

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.