Let $(X_n)$ be a sequence of independent uniform random variables in $[-1,1]$. Prove of disprove: $\lim_{n\to\infty} (1+X_1)(1+\frac{X_2}{2})...(1+\frac{X_n}{n})$ exists and belongs to $(0,\infty)$ almost surely.

Attempt: If we denote $Y_n=(1+X_1)(1+\frac{X_2}{2})...(1+\frac{X_n}{n})$ then it is well known that $(Y_n)$ is a martingale. Since it is non negative we conclude from the martingale convergence theorem that it must converge almost surely to a finite limit. So the only question is: is the limit almost surely in $(0,\infty)$ or it might be $0$ with a positive probability?

The limit is zero if and only if (we can forget about the case when $X_1=-1$ because it happens with probability $0$) for each $n$ we have $\prod_{k=n}^\infty (1+\frac{X_k}{k})=0$. So the event $\{Y_n\to 0\}$ belongs to the tail sigma algebra of $(X_n)$ and by Kolmogorov's $0-1$ law its probability is either $0$ or $1$. So now I just need to check if the probability of the limit of $Y_n$ being zero is positive or not. I had an idea to show that if at a point $\omega$ we have $(1+X_1(\omega))...(1+\frac{X_n(\omega)}{n})\to 0$ then at this same point we have $(1-X_1(\omega))...(1-\frac{X_n(\omega)}{n})\to\infty$. Since $Z_n:=(1-X_1(\omega))...(1-\frac{X_n(\omega)}{n})$ is a non negative martingale it must converge to a finite limit almost surely as well. So if my idea was correct then we would get that the set of such points $\omega$ must have probability zero, hence that would imply that almost surely the limit of $Y_n$ is in $(0,\infty)$. However, I failed to prove that my idea was correct. Any ideas?

  • $\begingroup$ Kolmogorov 3-series theorem tells that $$\sum_{n=1}^{\infty}\frac{X_n}{n}$$ converges almost surely. Now notice that the infinite product in question converges to a positive value if and only if this series converges. $\endgroup$ May 28, 2019 at 19:21
  • $\begingroup$ I can see that the series converges almost surely. I suppose you mean that $\prod_{n=1}^\infty (1+a_n)$ converges to a positive number if and only if $\sum_{n=1}^\infty a_n$ converges. This is a known theorem, but don't we need the terms of the sequence $a_n$ to be non negative to use it? $\endgroup$
    – Mark
    May 28, 2019 at 19:58
  • $\begingroup$ Good point. I was lazy enough to skip explaining that I am using the fact that $\log(1 + X_k/k) = (X_k/k) + \mathcal{O}(1/k^2)$. $\endgroup$ May 28, 2019 at 20:00
  • 1
    $\begingroup$ Oh, so here it's just Taylor expansion of logarithm. I guess this is the way to work with infinite products in general. Thanks a lot for your help. $\endgroup$
    – Mark
    May 28, 2019 at 20:05

1 Answer 1


Here is a problem-specific solution. Notice that

$$ \prod_{k=2}^{n} \left( 1 - \frac{X_k}{k} \right) = \prod_{k=2}^{n} \left( 1 + \frac{X_k}{k} \right)^{-1} e^{\mathcal{O}(1/k^2)}, $$

which follows from $1 - x = \frac{1-x^2}{1+x}$. From this, we find that $\left[ \prod_{k=1}^{n} \left( 1 + \frac{X_k}{k} \right) \right]^{-1}$ converges almost surely in $(0, \infty)$, and so, its reciprocal cannot be zero.


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.