Does Markov Chain with infinite state space (i.e. S = {0,1,2,...} have equilibrium distribution?

364 Views Asked by At

This question might be super stupid and dumb, but I still have to ask. I thought equilibrium distribution only exists for finite states Markov chain. But my course book did give me a example with state space $S=\{0,1,2,... \}$.
enter image description here

It talked a lot, but somehow appears to claim there is no equilibrium distribution. Because it violates the condition that $\sum_{i\in S}\pi_i=1$

I am completely confused by it. Does equilibrium distribution for Markov Chain with infinite many state space exist, does it always exist or conditionally exist, or never?

1

There are 1 best solutions below

4
On

In that particular case there is no equilibrium distribution (rather obviously there is a tendency to drift right)

But suppose instead, for example you had $$X_{n+1}= \left\{\begin{array}{cl} X_{n}-1 & \text{with probability } \frac{X_n}{X_n+2} \\ X_n & \text{with probability } \frac{1}{X_n+2} \\ X_{n}+1 & \text{with probability } \frac{1}{X_n+2} \end{array}\right.$$

then there would be a stable distribution on the non-negative integers

In fact this seems to have a stable distribution of $\pi_j = \dfrac{j +2}{3e j!}$