1
$\begingroup$

I have a random variable $X$ that arose from a Graph Theory problem, but I believe the details of the problem are not essential to this part, since the instructions explicitly say to use Markov's inequality. What might be important is that $0\leq X \leq m$ for some $m>1$. We have shown earlier that $E[X]\geq m/2$ and now are asked to use Markov's inequality to demonstrate that $P(X\geq 49m/100)\geq 1/51$.

I can't see how Markov's Inequality could be used to give a lower bound on a probability. I saw this question and answer: A lower bound on the probability that a variable is 3/2 times the expected value

I didn't follow some of the steps, but even so, in the end it appeared to me that the probability that was lower-bounded was just for the complement of the event in question. So as far as I can tell that solution would not help here.

Just for completeness I will write the inequality and do my best to try to use it:

$$P(X\geq 49m/100) \leq E[X]/(49m/100) \Rightarrow$$ $$49mp/100 \leq E[X]$$

where $p=P(X\geq 49m/100)$. I have heard rumor of a reverse Markov inequality in which I could compute

$$p \leq E[m-X]/(m-49m/100)$$ $$51mp/100 \leq E[m-X]$$

and I would imagine this further implies

$$51mp/100\leq m-E[X]$$ $$49mp/100\leq - E[X]$$ $$E[X]\leq 49mp/100$$

But clearly this result is paradoxical so I must not be fully understanding what the reverse Markov inequality really says, if it even is helpful here.

$\endgroup$
1
$\begingroup$

Let $Y=m-X$ so $Y$ is non-negative and $E[Y] \le \frac m2$

so can say from Markov's inequality that $P\left(Y \gt\frac{51m}{100}\right) \lt \frac{E[Y]}{51m/100} \le \frac{m/2}{51m/100}= \frac{50}{51}$

and thus $P\left(Y \le\frac{51m}{100}\right) \ge \frac{1}{51}$ and $P\left(X \ge\frac{49m}{100}\right) \ge \frac{1}{51}$

$\endgroup$
1
$\begingroup$

Let $A=\{X\geqslant x\}$ with $x\geqslant0$, then the pointwise inequality between random variables $$X\leqslant x\mathbf 1_{A^c}+m\mathbf 1_A$$ implies the inequality between expectations $$E(X)\leqslant xP(A^c)+mP(A)=x+(m-x)P(A)$$ Assuming furthermore that $x<m$ and $E(X)\geqslant y$, one gets $$P(A)\geqslant\frac{y-x}{m-x}$$ In your case, $x=0.49m$ and $y=0.5m$ hence, indeed, $$\frac{y-x}{m-x}=\frac1{51}$$

$\endgroup$

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.