4
$\begingroup$

Let $X_k, k \geq 1$, be i.i.d. random variables such that $\limsup_{n\rightarrow\infty} \frac{X_n}{n}<\infty$ a.s, then show that $\limsup_{n\rightarrow\infty} \frac{\sum_{k=1}^{n} X_k}{n}<\infty$ a.s.

I'm thinking to use Borel-Cantelli lemma but don't know where to start. Any hints would be helpful. Thanks.

$\endgroup$
  • $\begingroup$ Probably too simple but subadditivity of limsup? en.wikipedia.org/wiki/… $\endgroup$ – BCLC Dec 12 '15 at 20:45
  • $\begingroup$ @BCLC But it's finite almost surely so I'm not sure whether we can use the property of limsup directly here. $\endgroup$ – Chris Gartland Dec 12 '15 at 22:56
  • $\begingroup$ What do you mean? $\endgroup$ – BCLC Dec 13 '15 at 2:24
  • $\begingroup$ @BCLC I mean this is almost surely, not pointwise. $\endgroup$ – Chris Gartland Dec 13 '15 at 2:58
  • $\begingroup$ I don't follow. Why can't we just split up the sum on the right? $\endgroup$ – BCLC Dec 13 '15 at 3:02
3
$\begingroup$

Suppose $\limsup_{n\rightarrow\infty} X_n/n < \infty$ with prob 1. Define the nonnegative i.i.d. random variables $Y_n = \max[X_n,0]$ for all $n \in \{1, 2, 3, …\}$. Then with prob 1 we have $\limsup_{n\rightarrow\infty} Y_n/n < \infty$ and: $$ \limsup_{n\rightarrow\infty} \frac{1}{n}\sum_{i=1}^n X_i \leq \lim_{n\rightarrow\infty} \frac{1}{n}\sum_{i=1}^n Y_i = E[Y] \quad \mbox{(by LLN)}$$ where $Y=Y_1$. We either have $E[Y]<\infty$ or $E[Y] = \infty$. If $E[Y]<\infty$ then we are done. Suppose $E[Y]=\infty$ (we reach a contradiction). Fix $m$ as a positive integer. Since $Y$ is nonnegative we have \begin{align} \infty &= E[Y] \\ &= \int_0^{\infty} P[Y>t]dt \\ &=\sum_{n=0}^{\infty} \int_{mn}^{m(n+1)} P[Y>t]dt \\ &\leq \sum_{n=0}^{\infty}mP[Y>mn] \\ &= mP[Y>0] + m\sum_{n=1}^{\infty} P[Y_n>mn] \\ &= mP[Y>0] + m \sum_{n=1}^{\infty} P[Y_n/n > m] \end{align} So $$\sum_{n=1}^{\infty} P[Y_n/n > m] = \infty $$ and by Borel-Cantelli it means that $\{Y_n/n>m\}$ for infinitely many indices $n$ (with prob 1), so that $\limsup_{n\rightarrow\infty} Y_n/n \geq m$ with probability 1. This holds for all $m>0$, so $\limsup_{n\rightarrow\infty} Y_n/n = \infty$ with probability 1, a contradiction. $\Box$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ The main idea is to show that if $Z_n$ are i.i.d. random variables that satisfy $E[Z_i]=\infty$, then $\limsup_{n\rightarrow\infty} Z_n/n = \infty$ with prob 1. $\endgroup$ – Michael Jun 30 '18 at 19:22
0
$\begingroup$

I think the Stolz-Cesaro's lemma will solve the problem. https://en.wikipedia.org/wiki/Stolz%E2%80%93Ces%C3%A0ro_theorem

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Welcome to the site! This is better suited as a comment; an answer should first of all be certain in what it is stating; that you think it is useful is not enough, because it actually doesn't answer the question - it is simply a suggestion (which is fine in the form of a comment). And second, an answer should show why the lemma could be used, not just link to it. Cheers! $\endgroup$ – Bobson Dugnutt Feb 28 '16 at 11:56
  • $\begingroup$ Third, the suggestion to use Stolz-Cesaro does not work. $\endgroup$ – Did Jun 30 '18 at 20:54

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.