I am trying to read my probability book on the "Strong Law of Large Numbers", and came across this example that is really confusing me.

Let $X_i$ be a sequence of independent uniformly distributed random variables in $[0, 1]$ and $Y_n = \min(X_1, ..., X_n)$. Show that $Y_n$ converges to zero with probability 1.

The book says let $Y$ be the limit of the $Y_n$s which exists because nonincreasing and bounded below. Then for $1 > \epsilon > 0$, $$P(Y \ge \epsilon) = P(X_1\ge \epsilon \text{ & }\cdots\text{ & }X_n \ge \epsilon) = (1 - \epsilon)^n.$$ Here it means $Y_n$ I think, right?

Then it says $$P(Y \ge \epsilon) \le \lim_{n \to \infty} (1 - \epsilon)^n = 0.$$ So we can conclude $P(Y \ge \epsilon) = 0$, so $P(Y = 0) = 1$.

Why does it have a less than or equal sign for the last step? Shouldn't it be equality?

Thanks for any help.

  • 1
    $\begingroup$ The $Y_n$ is a typo. And yes, it should be equality. $\endgroup$ – André Nicolas Feb 21 '13 at 5:21

Why does it have a less than or equal sign for the last step?

The reason is that one starts from the almost sure inequality $Y\leqslant Y_n$ for each fixed $n$, which yields $[Y\geqslant\varepsilon]\subseteq[Y_n\geqslant\varepsilon]$, hence $\mathbb P(Y\geqslant\varepsilon)\leqslant\mathbb P(Y_n\geqslant\varepsilon)$. Finally, $\mathbb P(Y\geqslant\varepsilon)\leqslant\lim\limits_{n\to\infty}\mathbb P(Y_n\geqslant\varepsilon)$.


Shouldn't it be equality?

In the end, after the proof is finished, one knows the equality holds. But for the proof to be correct, one should definitely use an inequality.


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.