2
$\begingroup$

Let $(X_n)_{n \geq 1}$ be a sequence of i.i.d. random variables. Show that: $$\frac{X_n}{n} \rightarrow 0 \text{ P-a.s} \Leftrightarrow E[|X_1|] < \infty$$

I thought of using the Borel-Cantelli lemma's but i don't know hot to proceed..

$\endgroup$
  • $\begingroup$ Isn't that a strong law? Don't you also need that E[X]=0 because the sample mean converges to E[X]? $\endgroup$ – Jimmy R. Mar 25 '14 at 13:17
  • $\begingroup$ @Stefanos No, it is not. He doesn't want to show that $\frac1n\sum_{k=1}^n X_k\to0$. He only wants to show that $\frac{X_n}n\to 0$. $\endgroup$ – Cm7F7Bb Mar 25 '14 at 13:19
  • $\begingroup$ @V.C. Ok, yes... true $\endgroup$ – Jimmy R. Mar 25 '14 at 13:26
1
$\begingroup$

Indeed Borel-Cantelli, twice.

First note that for every positive $c$ and every real number $x$, $$ c\sum_{n=1}^\infty\mathbf 1_{|x|\geqslant cn}\leqslant|x|\leqslant c\sum_{n=0}^\infty\mathbf 1_{|x|\geqslant cn}, $$ hence, as soon as $(X_n)$ is identically distributed, $$ cU(c)\leqslant E(|X_1|)\leqslant c+cU(c),\qquad U(c)=\sum_{n=1}^\infty P(|X_n|\geqslant cn). $$ In particular, here is a result of independent interest:

Assume that the sequence $(X_n)$ is identically distributed. Then the expectation $E(|X_1|)$ is finite if and only if, for every positive $c$, the series $\sum\limits_nP(|X_n|\geqslant cn)$ converges.

Now to the proof of the exercise.

If $E(|X_1|)$ is finite, then the series $U(c)$ converges hence the so-called easy part of Borel-Cantelli lemma shows that the event $\limsup\limits_{n\to\infty}\,[|X_n|\geqslant cn]$ has probability $0$, thus, almost surely $\limsup\limits_{n\to\infty}\,|X_n|/n\leqslant c$. This holds for every $c\gt0$ hence $\lim\limits_{n\to\infty}\,|X_n|/n=0$ almost surely, which implies that $\lim\limits_{n\to\infty}\,X_n/n=0$ almost surely. This part does not use the independence.

On the other hand, if $E(|X_1|)$ is infinite, then the series $U(c)$ diverges and the sequence $(X_n)$ is independent hence the so-called difficult part of Borel-Cantelli lemma shows that the event $\limsup\limits_{n\to\infty}\,[|X_n|\geqslant cn]$ has probability $1$, thus, almost surely $\limsup\limits_{n\to\infty}\,|X_n|/n\geqslant c$. This holds for every $c\gt0$ hence $\limsup\limits_{n\to\infty}\,|X_n|/n=+\infty$ almost surely, which implies that the sequence of general term $X_n/n$ diverges almost surely.

$\endgroup$
0
$\begingroup$

Replace $X$ by $|X|$: we can then assume that $X>0$.

  • if $EX_1<\infty$: $$ \frac 1n X_n = \frac 1n X_0 + \frac 1n \sum_{k=1}^n (X_k - X_{k-1}) \to E(X_1 - X_{0}) = 0 $$a.s., because of the $LLN$.

  • if $\frac {X_n}n\to 0$ a.s.: $$\frac 1n \sum_{k=1}^n (X_k1_{X_k<N} - X_{k-1}1_{X_{k-1}<N})\to EX1_{X<N} $$ Now RHS $\to EX$ via monotonic convergence. $$ \frac 1n \sum_{k=1}^n (X_k1_{X_k\ge N} - X_{k-1}1_{X_{k-1}\ge N}) = o(1) + \frac 1n X_n1_{X_n\ge N}\to 0 $$ because $0\le\frac 1n X_n1_{X_n\ge N} \le \frac1n {X_n}$. Now make the sum and take the limit $n\to\infty$ and then $N\to\infty$, and you are done.

$\endgroup$

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.