Let $X$ be a random variable such that $E[X] = \alpha$, $\alpha \in \mathbb{R}$ and $E[X^2] = \beta$. The problem is to find a lower bound on the following probability $$ P \left[|X| > \frac{|\alpha|}{2} \right]. $$

Any help would be appreciated.


You are going to minimise the probability in question if you put as much of the probability as possible at $\frac \alpha 2$ and the rest at a single other point, though you may need two other points when $\alpha=0$.

This happens when $X=\frac{\alpha}{2}$ with probability $\frac{4\beta-4\alpha^2}{4\beta-3\alpha^2}$ and $X=\frac{2\beta-\alpha^2}{\alpha}$ with probability $\frac{\alpha^2}{4\beta-3\alpha^2}$, and these give the desired mean and second moment, noting $\beta\ge \alpha^2$ since $E[X^2] \ge (E[X])^2$ so all the probabilities are between $0$ and $1$ and the "other point" is larger in magnitude than $\alpha$. Thus

$$P \left[|X| > \frac{|\alpha|}{2} \right] \ge \frac{\alpha^2}{4\beta-3\alpha^2}$$

When $\alpha=0$, for some arbitrarily small $\delta$, let $X=+\sqrt{\frac \beta\delta}$ with probability $\frac{\delta}2$, $X=-\sqrt{\frac \beta\delta}$ with probability $\frac{\delta}2$ and $X=0$ with probability $1 -{\delta}$. So the lower bound in this situation is $0$, as suggested by the earlier expression.

  • 1
    $\begingroup$ The key to this solution is the initial characterization of the minimum. How do you demonstrate that? $\endgroup$
    – whuber
    Sep 22 '21 at 15:15
  • 1
    $\begingroup$ @Henry, Can you please provide any reference to the proof, how did you get $P \left[|X| > \frac{|\alpha|}{2} \right] \ge \frac{\alpha^2}{4\beta-3\alpha^2} $? $\endgroup$
    – Bhisham
    Sep 28 '21 at 10:47
  • $\begingroup$ @Bhisham It is an immediate consequence of my first point. The explanation for that point (not proof) is that if you have any probability in $\left[-\frac{|\alpha|}{2},\frac{|\alpha|}{2}\right]$ not at $\frac{\alpha}{2}$ then you can shift some of it to $\frac{\alpha}{2}$ and the rest outside the interval while maintaining the first two moments; then if you have any probability outside the interval but not at $\frac{2\beta-\alpha^2}{\alpha}$ you can shift it and some of the probability at $\frac{\alpha}{2}$ to $\frac{2\beta-\alpha^2}{\alpha}$ while maintaining the first two moments $\endgroup$
    – Henry
    Sep 28 '21 at 11:04

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.