Let $X_1, X_2, ...$ be random variables with $\mbox E X^2_j = 1$ for all $j$. Show that $\inf P[|X_n| > \epsilon] > 0$ for some $\epsilon > 0$ if and only if $\inf \mbox E |X_n| > 0$.

I have shown necessity, which is quite easy and doesn't use the hypothesis that $\mbox E X_j ^ 2 = 1$. I'm stuck on sufficiency, and can't seem to figure out how to work $\mbox E X_j ^ 2 = 1$ into anything useful.




1) $\Rightarrow$ : $E(|x|) = \int_{|x|\le e}|x| \; p(x) \, dx + \int_{|x|> e}|x| \; p(x) \, dx \ge \epsilon \; P(|x| >\epsilon )$

2) $\Leftarrow$ : Chebyshev's inequality

Added: Say we have $X$ with $E(X^2)=1$ and $E(|X|)\ge a > 0$ Let's apply Chebyshev to $Y=|X|$. $Var(Y) = E(Y^2)-E(Y)^2 \le 1- a^2$ Let's take (say) $\epsilon = a/2$. Then $P(Y<\epsilon) = P(a-Y > a/2)$ corresponds to a "tail", that we can bound strictly below 1, for example with the one-sided Chebyshev inequality.

  • $\begingroup$ I am not seeing how to get Chebyshev to help. It seems I need to bound $E[X_n]$ from above using $P(|X_n| > \epsilon)$. What should I apply Chebyshev to? $\endgroup$ – guy Jul 1 '11 at 1:28
  • $\begingroup$ Perhaps we are speaking of opposites senses of the implication. To show that $E(|X|) >0$ given that $P(|X|>\epsilon)>d$ you use the first one. $\endgroup$ – leonbloy Jul 1 '11 at 1:32
  • $\begingroup$ I have the first result. I can't prove that if $\inf_n E[|X_n|] > 0$ then $\inf P(|X_n| > \epsilon) > 0$ for some $\epsilon > 0$, the Chebyshev hint not withstanding. If I can bound $E[|X_n|]$ from above using $P(|X_n| > \epsilon)$ for some $\epsilon$ then I can prove the contrapositive, but that is as far as I can go. Perhaps I am completely on the wrong track? I can't see how to use Chebyshev to do this. $\endgroup$ – guy Jul 1 '11 at 1:43
  • $\begingroup$ I expanded my hint - still not a complete proof. $\endgroup$ – leonbloy Jul 1 '11 at 2:07
  • $\begingroup$ Okay, I think I see now. So bound $P(Y < \epsilon)$ strictly below $1$ using Cantelli's Inequality which gives a lower bound on $P(Y \ge \epsilon)$, correct? $\endgroup$ – guy Jul 1 '11 at 2:46

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.