Let $X_n$ be a Markov chain. Use Levy's zero-one law to show that if $$P(\cup_{m=n+1}^{\infty}\{X_m\in B_m\}|X_n)\geq\delta>0 \text{ on }\{X_n\in A_n\}\text{, then}$$ $$P(\{X_n\in A_n\text{ i.o.}\} - \{X_n\in B_n\text{ i.o.}\})=0$$

My attempt: $$\text{Let }C_n = \cup_{m=n+1}^{\infty}\{X_m\in B_m\}\implies \cap C_n=\{X_n\in B_n \text{ i.o.}\}$$ $$P(\cup_{m=n+1}^{\infty}\{X_m\in B_m\}|X_n) = E(1_{C_n}|X_n)=E(1_{C_n}|\mathcal{F}_n)\rightarrow E(1_{\cap C_n}|\mathcal{F}_{\infty}) = 1_{\cap C_n} \text{ by 0-1 law}$$

where $\mathcal{F}_n = \sigma(X_1,..,X_n)$ and the second equality holds because of Markov property. We know that on $$\{X_n\in A_n\}, E(1_{C_n}|X_n)\geq \delta>0 \text{ and thus on } \{X_n\in A_n \text{ i.o.}\}, E(1_{C_n}|X_n)\geq \delta>0 \text{ i.o.}$$ and since the limit of $E(1_{C_n}|X_n) = \cap C_n = \{X_n\in B_n \text{ i.o.}\}$, I guess $$\{X_n\in A_n \text{ i.o.}\}\subset \{X_n\in B_n \text{ i.o.}\}$$

May I know how that leads to the answer, I am missing something small. Thanks.


Since the convergence is hold a.s. You actually have shown $$1_{\{X_n\in A_n \text{ i.o.}\}}\leq 1_{\{X_n\in B_n \text{ i.o.}\}} \text{ a.s.}$$ Note that two sets $U$ and $V$ with $U\subset V$ imply that $U-V=\emptyset$. Hence $$\{X_n\in A_n \text{ i.o.}\}- \{X_n\in B_n \text{ i.o.}\}$$ is a null set and then has probability $0$.


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.