Consider the following random experiment:

We play a game where in each round we either win or lose.Let $\{A_n\}$,$n\in\mathbb N$, be a sequence of events where $A_n$ is the event that we win in the $n$-round.Let $P\left(A_{n}\right)=\frac{1}{n^2}$, for each $n$.Finally, the events are independent (since the game is played successively in independent rounds).

This is a typical example of a random experiment where the Borel-Cantelli lemma can be applied.

We have $$\sum_{n=1}^{\infty} P\left(A_{n}\right)=\sum_{n=1}^{\infty} \frac{1}{n^2}=\frac{\pi^2}{6} < \infty,$$so $$P\left(\limsup\limits_{n\to\infty} A_n\right) = 0$$

Therefore, the probability that infinitely many of $A_n$ occur is $0$. Only a finite number of $A_n$ occur. Let $k$ be this finite number.

My question is the following:

Since the events are independent,we can apply the rule of product.We will apply it for the first $k+1$ events.So, we have $$P\left(\bigcap_{n=1}^{k+1} A_n\right)=\prod _{n=1}^{k+1} P(A_n)=\prod _{n=1}^{k+1} \frac{1}{n^2} =\frac{1}{((k+1)!)^2}\neq 0$$ This means that there is a non-zero probability that both $A_1,A_2,...,A_{k+1}$ occur.Thus,it is possible that $k+1$ occur.But,we assumed that only $k$ of $A_n$ occur.In fact,we can always consider more and more events to the rule of product,finding smaller and smaller probabilities (but still non-zero) for each new intersection of events.

Obviously,I am doing something wrong here.How is the Borel-Cantelli lemma connected to the rule of product in this case? Thank you very much in advance!

  • $\begingroup$ Don't forget that $k$ itself is a random variable. $\endgroup$ – John Dawkins Mar 31 '16 at 1:06
  • $\begingroup$ Thank you for your comment,John Dawkins.Yes,k itself is indeed a random variable.But,the fact that "we can always consider more and more events to the rule of product" is valid independently of k. $\endgroup$ – AS42713 Mar 31 '16 at 20:03
  • $\begingroup$ Because $k$ is random and strongly dependent on the $A_n$s, the equality $P(\cap_{n=1}^{k+1}A_n)=\prod_{n=1}^{k+1} P(A_n)$ is most likely not true, and in any case would require proof. $\endgroup$ – John Dawkins Mar 31 '16 at 21:31

The Borel-Cantelli Lemma applies to tail events "at infinity", where by the Kolmogorov 0-1 law such events have possible probability 1 or 0 only. So any finite projection up to $k$ samples (however large $k$ is) does not capture what happens at infinity.

  • $\begingroup$ Thanks for your answer,kodlu.This idea that infinity is "untouchable" by any finite projection was close to my original thinking.But,I couldn't get past the fact that the sequence (1,1/4,1/36,1/576,......) ,which is the sequence of P(A1),P(A1 and A2),P(A1 and A2 and A3),P(A1and A2 and A3 and A4)..., has infinitely many terms. $\endgroup$ – AS42713 Mar 31 '16 at 19:52

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.