Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

For a Markov chain (can the following discussion be for either discrete time or continuous time, or just discrete time?),

  1. if for an initial distribution i.e. the distribution of $X_0$, there exists a limiting distribution for the distribution of $X_t$ as $t \to \infty$, I wonder if there exists a limiting distribution for the distribution of $X_t$ as $t \to \infty$, regardless of the distribution of $X_0$?
  2. When talking about limiting distribution of a Markov chain, is it in the sense that some distributions converge to a distribution? How is the convergence defined?


share|improve this question

1 Answer 1

up vote 3 down vote accepted
  1. No, let $X$ be a Markov process having each state being absorbing, i.e. if you start from $x$ then you always stay there. For any initial distribution $\delta_x$, there is a limiting distribution which is also $\delta_x$ - but this distribution is different for all initial conditions.

  2. The convergence of distributions of Markov Chains is usually discussed in terms of $$ \lim_{t\to\infty}\|\nu P_t - \pi\| = 0 $$ where $\nu$ is the initial distribution and $\pi$ is the limiting one, here $\|\cdot\|$ is the total variation norm. AFAIK there is at least a strong theory for the discrete-time case, see e.g. the book by S. Meyn and R. Tweedie "Markov Chains and Stochastic Stability" - the first edition you can easily find online. In fact, there are also extension of this theory by the same authors to the continuous time case - just check out their work to start with.

share|improve this answer
Thanks! I was wondering if the limiting distribution which is independent of initial distributions is unique when it exists? –  Tim Dec 1 '12 at 15:10
@Tim: could you please define the uniqueness? Is far as my guess is true, you mean exactly its independence from the initial distribution. –  S.D. Dec 1 '12 at 17:33
By "uniqueness" of the limiting distribution, I mean if there are two different probability measures on the state space s.t. they can both be the limiting distribution for a Markov chain, and the limiting distribution is defined to be the same for all initial distributions. –  Tim Dec 1 '12 at 17:51
@Tim Given any initial distribution it admits (if admits) a unique limiting distribution. Thus, if the latter is independent of the former, the latter is unique. In other words, suppose there are two limiting distributions $\pi_1$ and $\pi_2$, then $\|\nu_1 P^n - \pi_1\| \to 0$ and $\|\nu_2 P^n - \pi_2\| \to 0$ for some $\nu_1, \nu_2$ which contradicts with the fact that limit of $\nu_1 P^n$ is the same as of $\nu_2 P^n$ –  S.D. Dec 2 '12 at 9:43
Thanks! (1) Do you mean that for any given initial distribution, its limiting distribution (not necessarily the same one for other initial distributions) is unique? Why is that? (2) also I feel the first comment on my another post is confusing math.stackexchange.com/questions/248609/…. If you can let me know what you think, that will be appreciated! –  Tim Dec 2 '12 at 10:06

Your Answer


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.