3
$\begingroup$

I would like to show that

If $E(X_n) = 1/n$ and $\mathrm{Var}(X_n) = 1/n^2$ then $X_n\to0$ almost surely.

I can not find some relation between the almost sure convergence and $E(X_n)$ and $Var(X_n)$.

$\endgroup$
  • 1
    $\begingroup$ To complete the proof above, you might want to check that $$\liminf\{|X_n-1/n|<\epsilon\}\subseteq\{\limsup|X_n|\leqslant\epsilon\}$$ $\endgroup$ – Did Jun 7 '17 at 7:52
5
$\begingroup$

HINT

Chebyshev gives $P(|X_n-1/n| \ge \epsilon )\le \frac{1}{\epsilon^2n^2}.$ Then apply Borel-Cantelli. The key relationship here is that often "in probability" + "Borel-Cantelli" = "almost sure"

$\endgroup$
  • $\begingroup$ I am trying to prove $\left[X_n \nrightarrow 0 \right] \subset \hbox{lim sup} \left[|X_n - \frac{1}{n}| > \epsilon \right] $ $\endgroup$ – orrillo Jun 7 '17 at 6:26
  • 1
    $\begingroup$ @orrillo Intuition here is that for any $\epsilon$ borel cantelli shows that $|X_n-1/n|< \epsilon$ for all sufficiently large $n$ with probability $1$. Claim this means that for any $\epsilon,$ $|X_n|< \epsilon$ for all sufficiently large $n$ With prob 1. Claim this means $X_n \to 0$ w p 1. Now prove and fill in gaps. This is really just definition chasing in real analysis, not a lot of probability after the application of borel cantelli. $\endgroup$ – spaceisdarkgreen Jun 7 '17 at 15:52
  • 1
    $\begingroup$ @orrillo to put it more simply $|X_n-1/n|\to 0$ implies $X_n\to 0.$ so if you can show borel cantelli implies the first is true a.s. then you're done. $\endgroup$ – spaceisdarkgreen Jun 7 '17 at 15:58
  • $\begingroup$ Let $Y_n = (X_n - \frac{1}{n})$. By Borel Cantelli, we have that $P(\hbox{lim sup}\left[|Y_n|> \epsilon \right])= 0$ for all $\epsilon > 0$. We have that $\left[Y_n \nrightarrow 0 \right] = \bigcup_{\epsilon>0} \hbox{lim sup}\left[|Y_n|> \epsilon \right] = \bigcup_{n \in \mathbb{N}} \hbox{lim sup}\left[|Y_n|> 1/n \right]$. So, $P(\left[Y_n \nrightarrow 0 \right] ) = P(\bigcup_{n \in \mathbb{N}} \hbox{lim sup}\left[|Y_n|> 1/n \right]) \leq \Sigma_{n \in \mathbb{N}} P(\hbox{lim sup}\left[|Y_n|> 1/n \right])= 0$. It is OK? $\endgroup$ – orrillo Jun 7 '17 at 17:59
  • 1
    $\begingroup$ @orrillo looks fine to me other than a bit of a notation confusion using $n$ in two different ways. $\endgroup$ – spaceisdarkgreen Jun 7 '17 at 18:45

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.