In a Markov chain $(X_t)$, is it possible to find an initial distribution for $X_0$, s.t. the distribution of $X_t$ doesn't converge in some sense (such as wrt total variation) as $t \to \infty$?

The MC can be continuous or discrete-time, finite or infinite-state, homogeneous or inhomogeneous, .... But it will be the best that the MC is as special as possible, e.g., discrete-time, finite-state, homogeneous, ....

Thanks and regards!


Suppose the finitely many states of a Markov chain are partitioned into "even" states and "odd" states, and whenever you're in an "even" state you always go from there to an "odd" state and vice-versa. Then there's no limiting distribution if the initial distribution makes the sum of the probabilities of all the even states equal to $1$.

  • $\begingroup$ Thanks! I see. This example is not an aperiodic MC. Otherwise and if the MC is irreducible, there will be a limiting distribution for all initial distributions. $\endgroup$ – Tim Dec 10 '12 at 4:20
  • $\begingroup$ I was wondering what condition can make every initial distribution has a limit distribution? Note that the limit distributions for different initial distributions may be different. $\endgroup$ – Tim Dec 10 '12 at 14:07

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.