MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Hi all, question: Let $Z_t$ be an iid sequence with $$\mathbb{E}\log(Z_t^2)<0 $$ Show that $$\sum_{j=0}^\infty Z_t^2 Z_{t-1}^2 ... Z_{t-j}^2 < \infty$$ almost surely

I am supposed to use LLN to solve this... but i can't make ends meet (this is exam preparation sheet question)

share|cite|improve this question
This is just a straightforward exercise. Don't think it is really the kind of question this site is intended for. To answer, take the log of the product, divide by n and use LLN to deduce that the product is less than one (almost surely) for all large n. – George Lowther Jun 6 '10 at 17:13
Actually I misread it. The formula doesn't seem to make sense. Shouldn't the product be $Z_1\cdots Z_j$, in which case the LLN shows that the terms are almost surely bounded by a geometric series with ratio less than 1, so absoluty convergent. – George Lowther Jun 6 '10 at 17:22

If the term $Z_i^2$ of the independent and identically distributed random sequence is less than 1, then you can find a $q$ with $Z_i^2 \leq q < 1$ for almost all $i$ (acording to the law of large numbers). Then the $k$-fold product is less than $q^k$ such that the geometric series limits your sum by $\frac{1}{1-q}$.

I presume that you meant $Z_{t+j}^2$ instead of $Z_{t-j}^2$. Otherwise you'd get negative indices.

share|cite|improve this answer

For every $t$, let $Y_t=\log(Z_t^2)$. Fix some $t$. The sequence $(Y_{t-k})_{k\geqslant0}$ is i.i.d. with $E[Y_t]\lt0$ hence the usual law of large numbers yields $\frac1j\sum\limits_{k=0}^{j-1}Y_{t-k}\to E[Y_1]$. Fix some negative $m\gt E[Y_1]$.

Then $\frac1j\sum\limits_{k=0}^{j-1}Y_{t-k}\leqslant m$ for every $j$ large enough, that is, for every $j\geqslant J$ where $J$ is random and almost surely finite. In particular, for every $j\geqslant0$, $\sum\limits_{k=0}^{j}Y_{t-k}\leqslant mj+X$, for some almost surely finite random $X$. This implies the pointwise convergence of the series since $$ \sum_{j\geqslant0}\exp\left(\sum\limits_{k=0}^{j}Y_{t-k}\right)\leqslant\sum_{j\geqslant0}\mathrm e^X\mathrm e^{mj}=\mathrm e^X(1-\mathrm e^m)^{-1}. $$ Note that the RHS above is almost surely finite since $m\lt0$ but is not (a priori) uniformly bounded since $X$ may be unbounded.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.