Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Given disjoint nonempty subsets $X_1, X_2$ of the state space of a finite irreducible Markov chain, there are unique measures $\mu_1$ on $X_1$ and $\mu_2$ on $X_2$ such that (a) starting from a $\mu_1$-random point $x$, if one walks until one first hits $X_2$ at some point $y$, the point $y$ is distributed according to $\mu_2$, and (b) vice versa (with the roles of $X_1$ and $X_2$ reversed). E.g., if the Markov chain is (unbiased) random walk on $Z/5Z$ and $X_1$ is {0,1} and $X_2$ is {3,4}, then $\mu_1(0)=\mu_2(4)=2/3$ and $\mu_1(1)=\mu_2(3)=1/3$. Is there a way to think about these measures in terms of standard constructions in discrete probability theory?

share|improve this question
    
I take it you are looking at finite state space? And irreducible? In which case you can look at the sequence of hitting times $T_1,T_2,\ldots$ of, alternately, $X_1$ and $X_2$. Then, if $M_t$ is your markov chain, $M_{T_k}$ is an irreducible chain on $X_1\cup X_2$ with unique stationary distribution $(\mu_1+\mu_2)/2$. –  George Lowther Feb 17 '11 at 22:59
    
I added the words "finite" and "irreducible". Thanks, George. George's comments made me realize that my problem has a simple answer: Just create a new Markov chain with one or two states corresponding to each state of the original chain, according to whether the walk's most recent visit to $X_1 \cup X_2$ visited $X_1$ or $X_2$. Then $\mu_1$ and $\mu_2$ are stationary measures for return-maps of this chain. Sorry not to have thought harder before posting... –  James Propp Feb 18 '11 at 1:40
add comment

1 Answer

Just create a new Markov chain with one or two states corresponding to each state of the original chain, according to whether the walk's most recent visit to $X_1 \cup X_2$ visited $X_1$ or $X_2$. $\mu_1$ and $\mu_2$ are stationary measures for return-maps of this chain.

share|improve this answer
add comment

Your Answer

 
discard

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.