MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I want to ask the following probability inequality:

Is it true that for any random variable $X\ge 0$, we have

$$ \sup_{t>0}(t\mathbb E(X\mathbf 1_{X\ge t})) \le 2\sup_{t>0}(t^2 \mathbb P(X \ge t))? $$

share|cite|improve this question
You should tell us why you want to know; why you think it might be true etc. – Anthony Quas Dec 9 '12 at 16:27
It's from a big proof of some theorem. The right hand side of this inequality is a weaker version of the second moment. This inequality gets a lower bound of this weak second moment. – honglangwang Dec 9 '12 at 19:37
up vote 6 down vote accepted

This is an interesting inequality.

Let $f$ be the density of $|X|$. Let $$g(t)=t\int_t^\infty xf(x)dx.$$ Then our inequality is $$\sup_t g(t)\leq \sup_t t^2\int_t^\infty f(x)dx.$$ Let us assume that $\sup$ of the LHS is attained at some point $t=a$, so that $g(a)$ be the maximal value of $g$.

(It will be easy to get rid of this assumption, as well as of the assumption about existence of the density in the end). Our $g$ is bounded. Then the RHS evaluated at the same point $a$ is $$-2a^2\int_a^\infty\left(\frac{g(t)}{t}\right)^\prime\frac{dt}{t}.$$ I integrate this by parts and obtain $$2a^2\left( g(a)a^{-2}-\int_a^\infty\frac{g(t)}{t^3} dt\right).$$ Now, using that $a$ is the maximum of $g$, I estimate this from below as $$2g(a)-2g(a)a^2\int_a^\infty\frac{dt}{t^3}\geq g(a),$$ which completes the proof.

Getting rid of the assumptions made before is routine: just approximate your distribution by a distribution satisfying those assumptions.

share|cite|improve this answer

Notice that for a non negative random variable $Y$, we have $$ \mathbb E(Y)=\int_0^{\infty}\mu(Y\geqslant s)\mathrm ds.$$ Fix $t\geqslant 0$. We have for $s\leqslant t$ that $\{X\mathbf 1_{X\geqslant t}\geqslant s\}=\{X\geqslant t\}$ and if $s\gt t$ then $\{X\mathbf 1_{X\geqslant t}\geqslant s\}=\{X\geqslant s\}$. Consequently, we have $$\mathbb E(X\mathbf 1_{X\geqslant t})=t\mu(X\geqslant t)+\int_t^\infty\mu(X\geqslant s)\mathrm ds\leqslant t\mu(X\geqslant t)+\sup_{s\in\mathbb R}s^2\mu(X\geqslant s)\cdot \int_t^\infty s^{-2}\mathrm ds.$$ Multiplying by $t$ on both sides yields the wanted result.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.