3
$\begingroup$

The problem:

Let $(X_n)_{n\geq 1}$ be a real-valued sequence of i.i.d. random variables and let $c > 0$. Use Borel-Cantelli's lemma to show that

$$\sum_{n=1}^\infty P(X_n^2 > n) < \infty \Rightarrow P(|X_n|\geq c\sqrt{n} \hspace{7pt} \text{i.o.} \hspace{7pt} )=0.$$

My attempt: So from Borel-Cantelli we have $$P(X_n^2 > n \hspace{7pt}\text{i.o.}\hspace{7pt})=0$$

and using the definition of "infinitely often":

$$\bigcap_{m=1}^\infty \bigcup_{n=1}^\infty (X_n^2 > n)=\bigcap_{m=1}^\infty \bigcup_{n=1}^\infty (|X_n| > \sqrt{n})$$

But I don't see how I get the inclusion into the event containing $c$.

$\endgroup$
  • $\begingroup$ This sounds like a homework question so I'll only give a hint. You want to show that $\sum_{n=1}^\infty P( X_n^2 > cn) = \infty$ but you only know that $\sum_{n=1}^\infty P(X_n^2 > n) = \infty$. Try choosing $k > c$ and dividing the latter sum into $k$ pieces, corresponding to the $k$ residue classes mod $k$. The sums are going to be 'approximately' equal therefore they are all infinite. $\endgroup$ – Jair Taylor Oct 21 '18 at 23:16
  • 1
    $\begingroup$ I don't have a complete solution, but I suspect one way forward might be to actually not use the given condition directly. In particular I think $\sum^{\infty} P(X_n^2 > n) < \infty$ implies that $E(X_n^2)<\infty$, which you might be able to bound with Chebyshev's inequality and then in turn feed to the BC lemma to. That might resolve the difficulty with $c$. Just an idea. $\endgroup$ – gogurt Oct 21 '18 at 23:47
3
$\begingroup$

The reasoning in the opening post gives indeed the result when $0\lt c\leqslant 1$. In order to extend it to all the values of $c$, we use the following facts:

  1. Since the sequence $(X_n)_{n\geqslant 1}$ is identically distributed, $\Pr\{X_n^2\gt n\}=\Pr\{X_1^2\gt n\}$;
  2. If $Y$ is a non.negative integrable random variable, them $\sum_{n\geqslant 1}\Pr\{Y\gt n\}\leqslant \mathbb E[Y]$.

Consequently, for all positive $c$, the series $\sum_{n\geqslant 1}\Pr\{X_n^2\gt cn\}$ converges.

Notice that we only use the fact that the sequence $(X_n)_{n\geqslant 1}$ is identically distributed. Its independence is not needed.

$\endgroup$

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.