The problem goes: For some $k \ge 1$, let $\{X_i\}$ be a sequence of $k$-dependent random variables. That is, for each $n\ge 1$, $\sigma(X_1,\ldots,X_n)$ and $\sigma(X_{n+k+1},\ldots)$ are independent. In addition, $EX_i=0, \forall i$. If $\sum_n \text{var} X_n<\infty$, must $\sum_n X_n$ converge almost surely?

I know from the proof of the one series theorem that was given us, I only need to show $T_n \to_p 0$, where $T_n=\sup_{m\ge n}|S_m-S_n|$, and $S_n=\sum_{i=1}^n X_i$. However, in the proof of the one series theorem, this was done by Kolmogorov's maximal inequality, which hinges on $\{X_i\}$ being independent, which is not the case here. Thank you for any help.


We have for each $i_0\in\{0,\dots,k-1\}$ that the series $$\sum_{j=1}^\infty X_{kj+i_0}$$ converges almost surely. Hence there is a set $\Omega'\subset\Omega$ of probability one for which if $\omega\in \Omega'$, then for each $i_0\in\{0,\dots,k-1\}$, the series $\sum_{j=1}^\infty X_{kj+i_0}$ is convergent.

Assume we have a sequence $(c_n)_{n\geqslant 1}$ of real numbers such that $S':=\sum_n c_{2n}$ and $S'':=\sum_n c_{2n+1}$ are convergent. Define $S'_N:=\sum_{j=1}^Nc_{2j}$ and $S''_N:=\sum_{j=1}^Nc_{2j-1}$ and $S_N:=\sum_{j=1}^Nc_j$. Then $$S_{2N}=S'_N+S''_N\to S'+S''$$ and similarly, $S_{2N+1}\to S'+S''$. The same idea applies for $k$ not necessarily equal to $2$.

  • $\begingroup$ Thank you! I should have thought about breaking the series into $k$ subseries. $\endgroup$ – Fang Jing Dec 2 '13 at 18:40

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.