4
$\begingroup$

How does one show that any finite-state time homogenous Markov Chain has at least one stationary distribution in the sense of $\pi = \pi Q$ where $Q$ is the transition matrix and $\pi$ is the stationary distribution? My instinct is it involved eigenvalues and the Perron Frobenius theorem but I'm having trouble completing the argument.

$\endgroup$

3 Answers 3

7
$\begingroup$

There are a number of ways to prove this. In a recent article "What is a stationary measure?", Alex Furman outlines a straightforward proof using only linear algebra. He notes that the existence of an invariant probability measure also follows from the Brouwer fixed point theorem.

$\endgroup$
1
  • 1
    $\begingroup$ The article cited does not seem to provide an elementary proof as there is no(?) justification for the statement: if $P^Tv = v$ then $P^Tv^+ = v^+$. $\endgroup$
    – Ben
    Dec 7, 2018 at 9:12
0
$\begingroup$

Any stochastic matrix $P$ has at least one stationary distribution $\pi$ in the sense that $\pi P=\pi$. This can be shown by the following statement from Wikipedia.

Brouwer Fixed Point Theorem (applied to the compact convex set of all probability distributions of the finite state space $\Omega$) implies that there is some left eigenvector which is also a stationary probability vector.

Of course, we could also prove this by Perron Frobenius theorem.

$\endgroup$
0
-1
$\begingroup$

Isn't it simple? Suppose you take the matrix to a power k where k-> infinity. Then all eigenvalues approach 0 while eigenvalue =1 is stationary. Note that the image is always the span of colloms so for eigenvalue 1 the picture must be one dimensional.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .