periodicity of a Markov chain
Let be a stationary (http://planetmath.org/StationaryProcess) Markov chain with state space
. Let be the -step transition probability that the process goes from state at time to state at time :
Given any state , define the set
It is not hard to see that if , then . The period of , denoted by , is defined as
where is the greatest common divisor of all positive integers in .
A state is said to be aperiodic if . A Markov chain is called aperiodic if every state is aperiodic.
Property. If states communicate (http://planetmath.org/MarkovChainsClassStructure), then .
Proof.
We will employ a common inequality involving the -step transition probabilities:
for any and non-negative integers .
Suppose first that . Since , and for some . This implies that , which forces or , and hence .
Next, assume , this means that . Since , there are such that and , and so , showing . If we pick any , we also have , or . But this means divides both and , and so divides their difference, which is . Since is arbitrarily picked, . Similarly, . Hence .∎