4
$\begingroup$

Let $X_1, X_2, \ldots$ be independent random variables with $0 \leq X_n \leq 1$ and $\sum_n E(X_n) = \infty$. I'd like to show that $\sum_n X_n = \infty$ with probability 1. This seems like a Borel Cantelli Problem to be but I am having a hard time defining the sets to work with. Is there another easier approach here? Thanks!

$\endgroup$
  • $\begingroup$ Applying the 2nd Borel-Cantelli lemma together with your previous question immediately yields the desired result. $\endgroup$ – Sangchul Lee Dec 15 '15 at 18:15
4
$\begingroup$

We can use Kolmogorov's three-series theorem.

Let $X_1,X_2,\ldots$ be independent random variables. Let $A>0$ and let $Y_i=X_i1_{\{|X_i|\le A\}}$. In order that $\sum_{n=1}^\infty X_n$ converges almost surely, it is necessary and sufficient that $$ \sum_{n=1}^\infty\Pr\{|X_n|>A\}<\infty,\ \sum_{n=1}^\infty\operatorname EY_n\ \text{converges, and}\ \sum_{n=1}^\infty\operatorname{Var}Y_n<\infty. $$

Let us choose any $A\ge1$. Then we have that $$ \sum_n\operatorname E[X_n1_{\{|X_n|\le A\}}]=\sum_n\operatorname EX_n=\infty. $$ Hence, the series $\sum_nX_n$ diverges almost surely. Since $X_n\ge0$, we have that $\sum_nX_n=\infty$ almost surely.

$\endgroup$
  • $\begingroup$ Nice. See my answer for an alternative elementary solution. $\endgroup$ – Mizar Dec 14 '15 at 16:26
2
$\begingroup$

Independence and boundedness of the variables can be exploited efficiently with the following computation (which is an easy version of the Chernoff bound): for any fixed $\alpha>0$ we have $$\mathbf{P}(X_1+\dots+X_n\le \alpha)=\mathbf{P}(e^{-(X_1+\dots+X_n)}\ge e^{-\alpha})\le\frac{\mathbf{E}[e^{-(X_1+\dots+X_n)}]}{e^{-\alpha}} =e^\alpha\prod_{k=1}^n\mathbf{E}[e^{-X_k}].$$ Now for any $0\le s\le 1$ we have the simple inequality $$1-e^{-s}=se^\xi\ge se^{-1}$$ (we applied Lagrange's theorem to the function $x\mapsto e^x$ on the interval $[-s,0]$, which gives us some $\xi\in [-s,0]$ such that the first equality holds; then we used the fact that $\xi\ge -1$), thus $$\mathbf{E}[e^{-X_k}]\le \mathbf{E}[1-e^{-1}X_k]=1-e^{-1}\mathbf{E}[X_k]$$ and finally $$\prod_{k=1}^n\mathbf{E}[e^{-X_k}]\le\prod_{k=1}^n\left(1-e^{-1}\mathbf{E}[X_k]\right)\le\prod_{k=1}^n\exp\left(-e^{-1}\mathbf{E}[X_k]\right) =\exp\left(-e^{-1}\sum_{k=1}^n\mathbf{E}[X_k]\right)\to 0$$ as $n\to\infty$. So $$\mathbf{P}(\sum X_k\le \alpha)=\lim_{n\to\infty}\mathbf{P}(\sum_{k=1}^n X_k\le\alpha)=0$$ and since $\alpha$ was arbitrary we get the thesis.

$\endgroup$

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.