1
$\begingroup$

Suppose $P(x,dy)$ and $Q(x,dy)$ are two Markov transition kernels on a topological space $E$ equipped with Borel $\sigma$-algebra $\mathcal B(E)$. Suppose for every $x \in E$, $P(x,\cdot)$ and $Q(x, \cdot)$ are equivalent, i.e. for any $A \in \mathcal B(E)$ and $x \in E$, $P(x,A) > 0 \Leftrightarrow Q(x,A) > 0$.

Suppose there exists a unique measure which is invariant for $Q$ (up to multiplicative constant). I explicitly do not wish to exclude the case that this invariant measure has infinite mass.

Is it true that, up to a multiplicative constant, there exists at most one invariant measure for $P$?

$\endgroup$
2
  • $\begingroup$ Can you be define "invariant measure for Q"? Is it what's usually called a stationary measure? $\endgroup$ Jun 16, 2016 at 18:29
  • $\begingroup$ Yes, invariant measure = stationary measure. $\endgroup$ Jun 16, 2016 at 19:18

1 Answer 1

2
$\begingroup$

Since you don't want to exclude the case of infinite stationary measures, it is very easy to produce a counterexample. The simple random walk on $\mathbb Z$ has a unique stationary measure (which coincides with the counting one), whereas for $p\neq q$ the random walk with the transition probabilities $p(n,n+1)=p$ and $p(n,n-1)=q$ has two minimal stationary measures.

$\endgroup$
1
  • $\begingroup$ Of course, nice example, the stationary measures being the constant distribution as well as the one for which $\mu(n+1) = \alpha \mu(n)$ for some $\alpha \neq 1$. Thanks! $\endgroup$ Jun 16, 2016 at 19:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.