0
$\begingroup$

Consider the following concentration inequality.

concentration-inequalities

The problem is :

enter image description here

The inequalities are easily proven with the help of Markov's inequality. But I cannot find a strategy to construct an example to show that it cannot be improved. My rough strategy is : For every $\epsilon>0$, to construct a random variable $X_{\epsilon}$ with $E(X_{\epsilon})=0$ and $var(X_{\epsilon})=\sigma^2$ such that $P(X_{\epsilon} \geq x)>\frac{\sigma^2}{\sigma^2+x^2}$ if $x>0$ and $P(X_{\epsilon} \geq x)<\frac{x^2}{\sigma^2+x^2}$ if $x<0$. How can I construct such $X_{\epsilon}$? Thanks in advance.

Source : Rohatgi-Saleh, p.$98$, problem $3$.

$\endgroup$
1
$\begingroup$

Why not let $X\equiv 0$? That achieves equality in both cases.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Can I imply that the CI's cannot be improved upon (for any random variable satisfying the prescribed conditions) just by producing an example where equality is achieved? $\endgroup$ – Samayita Mar 4 '18 at 3:13
  • $\begingroup$ @Samayita That's correct. If you produce an example where equality is achieved, you can't possibly produce a lower or higher bound otherwise it won't work for that example. $\endgroup$ – user223391 Mar 4 '18 at 3:23
  • $\begingroup$ For example we still may have bounds which are sharper or equal to the bounds given in the lemma, and both the bounds coincide (and achieves equality) in the particular example you've given $\endgroup$ – Samayita Mar 4 '18 at 3:24
  • $\begingroup$ @Samayita Yes, other random variables may indeed be lower or higher than the bounds we have. $\endgroup$ – user223391 Mar 4 '18 at 3:25
  • $\begingroup$ Okay I understand now, the point is whether or not the bounds can be improved for ALL random variables (satisfying the prescribed conditions). Thanks! $\endgroup$ – Samayita Mar 4 '18 at 3:31

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.