How to prove the one-sided Chebyshev's inequality which states that if X has mean 0 and variance $\sigma^2$, then for any $a > 0$

$P(X \geq a) \leq \frac{\sigma^2}{\sigma^2+a^2}$

Solution. I know the Chebyshev's inequality which States that

$P(|X-\mu| \geq a) \leq \frac{Var(X)}{a^2}$ If I first argue that for any $b > 0$

$P(X \geq a) \leq P{[(X+b)^2 \geq (a+b)^2]}$

$P(X\geq a) \leq \frac{E(X+b)^2}{(a+b)^2}$

$P(X \geq a) \leq \frac{[E(X^2)+2E(X)b+b^2]}{(a+b)^2}$

$P(X \geq a) \leq \frac{\sigma^2+ b^2}{(a+b)^2}$

I got the correct answer.

  • 1
    $\begingroup$ Hint: Expand $(X+b)^2$, take the expected value, and write in terms of $\sigma^2$. Substitute what you get into $\frac{E(X+b)^2}{(a+b)^2}$ and then minimize that. $\endgroup$ – grndl Oct 1 '16 at 15:03
  • 1
    $\begingroup$ There is a full proof on the Wikipedia page en.wikipedia.org/wiki/Cantelli%27s_inequality $\endgroup$ – Jack D'Aurizio Oct 1 '16 at 15:09
  • $\begingroup$ I got the correct answer. $\endgroup$ – Dhamnekar Winod Oct 1 '16 at 15:25

I have a couple of proofs at http://www.se16.info/hgb/cheb.htm#OTProof and http://www.se16.info/hgb/cheb2.htm

One of these, loosely based on Probability and Random Processes by Grimmett and Stirzaker, would give a proof like this:

With $a>0$, for any $b\ge 0$ $$P(X\ge a) = P(X+b \ge a+b) \le E\left[\dfrac{(X+b)^2}{(a+b)^2}\right] = \dfrac{\sigma^2+b^2}{(a+b)^2}$$

But treating $\dfrac{\sigma^2+b^2}{(a+b)^2}$ as a function of $b$, the minimum occurs at $b = \sigma^2 / a$, so $$P(X\ge a) \le \dfrac{\sigma^2+(\sigma^2/a)^2}{(a+\sigma^2/a)^2} =\dfrac{\sigma^2(a^2+\sigma^2)}{(a^2+\sigma^2)^2} = \dfrac{\sigma^2}{ \sigma^2+a^2}.$$


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.