in a proof of the markov's inequality I didn't understand the last step

the markov's inequality : $$P(X \geq a) \leq \frac{E[X]}{a}$$

let $X : \Omega \to \Bbb{R}$ be a random variable

let $$\mathbf{1}_{X \geq a} :\mathbb{R} \to \{0,1\}$$

$$\mathbf{1}_{X \geq a}(X) = \begin{cases} 0, & \text{if $X<a$ } \\ 1, & \text{if $X\geq a$ } \end{cases}$$

and let $$g = a \cdot \mathbf{1}_{X \geq a} :\mathbb{R} \to \{0,a\}$$

$$g(X) = \begin{cases} 0, & \text{if $X<a$ } \\ a, & \text{if $X\geq a$ } \end{cases}$$

then $$g(X) \leq X$$ $$E[g(X)] \leq E[X]$$ $$a\cdot E[\mathbf{1}_{X \geq a}(X)] \leq E[X]$$ $$E[\mathbf{1}_{X \geq a}(X)] \leq \frac{E[X]}{a} \implies P(X \geq a) \leq \frac{E[X]}{a}$$

I didn't understand this last step can someone please elucidate.

  • $\begingroup$ Also considering the expectation of the Bernoulli random variable may help you understand a bit I feel, e.g. $\bf{1}_{X\geq a}(X)$ is a Bernoulli random variable, its expectation is the probability that X is greater and equal to a in this sense. $\endgroup$ – lzstat Mar 26 '17 at 17:01

The expectation of an indicator variable related to an event $E$ is always the probability that $E$ happens :


$$I_E = \begin{cases} 1 & \text{if } E \text{ happens} \\ 0 & \text{if } E \text{ doesn't happen} \end{cases} $$ Then $$ E[I_E] \overset{(*)}{=} 1\cdot Pr[E \text{ happens}] + 0 \cdot Pr[E \text{ doesn't happens}] = Pr[E \text{ happens}] $$ $(*)$ By the definition of expecation.

In your case the event is "$X \geq a$". Therefore $$ E[I_{X \geq a}] = 1\cdot Pr[X \geq a \text{ happens}] + 0 \cdot Pr[X \geq a\text{ doesn't happen}] = Pr[X \geq a] $$


Taking expectation is to integrate a random variable over the outcome space w.r.t the probability measure. Recall that $\int_{\Omega} 1_{A} dP = \int_{A} 1 dP = P(A)$ by basic properties of Lebesgue integration.

Also note that the set $\{ X \geq a \}$ is the shorthand for $\{ \omega \in \Omega \mid X(\omega) \geq a \}$ which is always a subset of $\Omega$ and which is actually simply the preimage $X^{-1}([a, +\infty])$.

  • $\begingroup$ Don't want to offense you but I don't think we should go into "deep" measure theory for a question related to Markov inequality in a "simple probabilistic framework". I might be wrong though :) $\endgroup$ – Zubzub Mar 24 '17 at 13:50

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.