# Weaker condition for law of large numbers

$X_k$'s are i.i.d. Suppose $X_k$ is symmetric and $E[|X_k|^{3/4}]<\infty$. Do we have $S_n/n \rightarrow 0$ either in probability or almost surely, where $S_n$ is the partial sum.

-
That's a curious moment condition. Where did you get that from? – cardinal Nov 13 '11 at 18:51
basically a random one < 1. – epsilon Nov 13 '11 at 18:56

No.

Take the standard Cauchy distribution: $$\newcommand{\rd}{\,\mathrm{d}} \mathbb E |X|^{3/4} = 2 \int_{0}^{\infty} \frac{x^{3/4}}{\pi (1+x^2)} \rd x= \csc(\pi/8) < \infty \>,$$ but $S_n/n$ is equal in distribution to $X$. That is, it is also standard Cauchy no matter how large $n$ becomes.

More generally, note that this same counterexample works for any $0 < p < 1$ since $$\frac{\pi}{2} \mathbb E|X|^p = \int_{0}^\infty \frac{x^p}{1+x^2} \rd x \leq 1 + \int_{1}^\infty x^{p-2} \rd x = \frac{2-p}{1-p} < \infty \> .$$

-
Thank you. It's a nice example – epsilon Nov 13 '11 at 19:19
Thanks. The Cauchy serves as a counterexample for many statements regarding probability and mathematical statistics. Usually, at least in a setting that admits continuous distributions, it's a good one to try out on any new conjecture or "result". – cardinal Nov 13 '11 at 19:27