Is it known what the next-to-leading order term is in the variance of the central limit distribution for the average of $N$ variables each of which is distributed according to $P(x) \sim 1/x^{1+\alpha}$ for $1 < \alpha$?

So we have $\bar{x} = N^{-1} \sum_{i=1}^N x_i$ where $x_i$ are independent and distributed according to $P(x)$, then the distribution of $\bar{x}$, $P_N(\bar{x})$, will have a variance $\sigma_N$ which goes to zero as $N\to \infty$ and the first terms are

$\sigma_N \sim 1 / N^{1-1/\alpha}$ for $1 < \alpha < 2$

$\sigma_N \sim ( \log(N) / N )^{1/2}$ for $\alpha = 2$

$\sigma_N \sim 1 / N^{1/2}$ for $2 < \alpha$

What I couldn't find so far is the form of the next-to-leading order terms in particular for $1 < \alpha \leq 2$ (for $2 < \alpha$ I actually found it). I suspect that for $\alpha = 2$ we will have

$$\sigma_N \sim ( \log(N) / N )^{1/2} ( 1 + c / \log(N) + \ldots)$$

but I haven't found this in any of the probability text books so far.


1 Answer 1


You are looking for a generalization of the Edgeworth expansion to distributions that have slowly decaying tails.

For $\alpha>2$ this can be found here:

Corrections to the Central Limit Theorem for Heavy-Tailed Probability Densities, by Henry Lam, Jose Blanchet, Damian Burch, Martin Z. Bazant (2011).

It seems that Vinogradov also considers the case $\alpha=2$ in his 1994 book on Refined Large Deviation Limit Theorems, but I have only seen the table of contents.

No pointers to $\alpha<2$, regrettably.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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