6
$\begingroup$

I am trying to solve the following problem, which provides an example for a central limit theorem in spite of the fact that the variance is infinite.
Consider the two-sided Pareto distribution with density $$ f(x) = \begin{cases} \frac{1}{|x|^3}, \enspace &\textrm{for} |x| \geq 1 \\ 0, \enspace &\textrm{otherwise}. \end{cases} $$ (a) Prove that the characteristic function $$ \varphi(t) = 1 - t^2 \left( \log \frac{1}{|t|} + O(1) \right) \enspace \textrm{as} \enspace t \rightarrow 0$$ (b) Let $X_1, X_2, \dots$ be independent, identically distributed random variable with the density as above. Prove that $$ \frac{\sum_{k=1}^{n} X_k}{\sqrt{n \log n}} \overset{d}{\rightarrow} \mathcal{N}(0,1) \enspace \textrm{as} \enspace n \rightarrow \infty.$$ I have proven (b) using (a). But I have no idea how to prove (a). I cannot use the same method as in the proof of the central limit theorem, since the variance is not finite. And I tried to calculate the characteristic function but this was not beneficial.
Thank you very much for your help.

$\endgroup$
2
  • $\begingroup$ What have you tried? $\endgroup$ Commented Apr 26, 2024 at 18:43
  • 1
    $\begingroup$ I updated the question $\endgroup$ Commented Apr 26, 2024 at 20:11

1 Answer 1

1
$\begingroup$

$$ \begin{align*} \varphi(t) &= \mathbb{E}\left(e^{itX}\right)\\ &= \int_{\mathbb{R}} e^{itx}f(x)dx = \int_{-\infty}^{-1} \dfrac{e^{itx}}{-x^3}dx + \int_1^\infty \dfrac{e^{itx}}{x^3}dx\\ &= \int_{1}^\infty \dfrac{e^{-itx}}{x^3}dx + \int_1^\infty \dfrac{e^{itx}}{x^3}dx \\ &= \int_1^\infty \dfrac{e^{itx} + e^{-itx}}{x^3}dx \end{align*} $$ Since $$ e^{itx} = \cos(tx) + i \sin(tx), \text{ and } e^{-itx} = \cos(tx) - i\sin(tx) $$ we have $$ \varphi(t) = 2 \int_1^\infty \dfrac{\cos(tx)}{x^3}dx $$ Now, notice that $$ \int_1^\infty \dfrac{1}{x^3}dx = \dfrac{1}{2} $$ Thus, $$ \varphi(t) = 1 - 2 \int_1^\infty \dfrac{1 - \cos(tx)}{x^3}dx $$ Since we only care about the behavior of $\varphi$ as $t \rightarrow 0$, we can assume that $0 < \vert t \vert < 1$. With this assumption, we split the integral into two: $$ \varphi(t) = 1 - 2 \int_1^{\frac{1}{\vert t \vert}} \dfrac{1 - \cos(tx)}{x^3}dx - 2 \int_{\frac{1}{\vert t \vert}}^\infty \dfrac{1 - \cos(tx)}{x^3}dx $$

The second integral, notice that $$ 0 \le \int_{\frac{1}{\vert t \vert}}^\infty \dfrac{1 - \cos(tx)}{x^3}dx \le \int_{\frac{1}{\vert t \vert}}^\infty \dfrac{2}{x^3}dx = t^2 $$ Thus, $$ \int_{\frac{1}{\vert t \vert}}^\infty \dfrac{1 - \cos(tx)}{x^3}dx = t^2 O(1) $$

Now, for the first integral, consider the Taylor expansion of $\cos$: $$ \cos(tx) = 1 - \dfrac{t^2x^2}{2} + t^4 O\left(x^4\right) $$

So, $$ \int_1^{\frac{1}{\vert t \vert}} \dfrac{1 - \cos(tx)}{x^3}dx = \int_1^{\frac{1}{\vert t \vert}} \left[\dfrac{t^2}{2x} + t^4O(x) \right]dx = \dfrac{t^2}{2} \ln\left(\dfrac{1}{\vert t \vert}\right) + t^2 O(1) $$

Therefore, $$ \varphi(t) = 1 - t^2\left(\ln\left(\dfrac{1}{\vert t \vert}\right) + O(1)\right), \text{ as } t \rightarrow 0 $$

$\endgroup$
3
  • $\begingroup$ If I use the Taylor expansion I get $ t^2/2x - t^4O(x) $ not $t^2/2x + t^4O(x)$. How do you get the plus, since the minus is problematic? $\endgroup$ Commented Apr 27, 2024 at 18:44
  • $\begingroup$ @EnergySkiller The definition of Big-Oh is as follows $$ f(x) = O(g(x)), \text{ as } x \rightarrow 0, \text{ if } \exists \delta, M > 0, \vert f(x) \vert \le M g(x) \ \forall x, 0 < \vert x \vert < \delta $$ So if $f = O(g)$ then $-f = O(g)$ $\endgroup$ Commented Apr 28, 2024 at 0:43
  • 1
    $\begingroup$ Thank you. I had an error in my reasoning. $\endgroup$ Commented Apr 28, 2024 at 10:45

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.