Suppose that $X_n$ are independent random variables where $\sup_n E|X_n| < \infty$.

Then, $\frac{\sum_{i=1}^{n}X_i}{n^a}$ converges to 0 almost surely for all $a>1$.

I think that $\frac{1}{n^{a-1}}$ converges to 0 and $\frac{\sum_{i=1}^{n}X_i}{n}$ converges because the sup of expectation is bounded, but since $X_n$ are not identical, the SLLN cannot be applied directly. And it makes me hard to follow the process of proving SLLN. Could you please help me?


Let $\Delta\equiv \sup_n\mathsf{E}|X_n|$. Then $$ \sum_{k=1}^\infty\mathsf{P}\left(\left|\frac{X_k}{k^\alpha}\right|\ge 1\right)\le \sum_{k=1}^\infty \frac{\Delta}{k^\alpha}<\infty, $$ and for $Y_n=X_n/n^{\alpha}1\{|X_n|\le n^{\alpha}\}$, $$ \sum_{k=1}^\infty\mathsf{E}Y_k \quad\text{and}\quad \sum_{k=1}^\infty \operatorname{Var}(Y_k) $$ converge as well$^{(*)}$. Hence, by Kolmogorov's three-series theorem, $\sum_{k=1}^n X_k/k^{\alpha}$ converges a.s. in $\mathbb{R}$. Convergence of $\sum_{k=1}^n X_k/n^{\alpha}$ follows from Kroneker's lemma.

$^{(*)}$ \begin{align} \sum_{k=1}^\infty \operatorname{Var}(Y_k)&\le \sum_{k=1}^\infty \frac{1}{k^{2\alpha}}\mathsf{E}[X_k^21\{|X_k|\le k^{\alpha}\}] \\ &\le \sum_{k=1}^\infty \frac{1}{k^{\alpha}}\mathsf{E}[|X_k|1\{|X_k|\le k^{\alpha}\}]\le \sum_{k=1}^\infty\frac{\Delta}{k^{\alpha}}<\infty. \end{align}


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.