1

Everywhere I look I see that the period of a state $i$ in a Markov chain is given by $$ \gcd\{n>0 : P_{ii}^n>0 \} $$ but what do we mean if the set $\{n>0 : P_{ii}^n>0 \} = \emptyset$? For instance in a two state Markov chain with $P_{00}=1$ and $P_{10}=1$, what is the period of the state $i=1$?

nullUser
  • 27,877

2 Answers2

1

Not many authors pay attention to this case, it is true. One author who does is Pierre Bremaud. In his Markov Chains: Gibbs Fields, Monte Carlo Simulation, and Queues he explicitly defines the period to be "infinity" when return is impossible. I guess that definition is as reasonable as any.

0

Periods generally only make sense in irreducible components of a Markov chain. In your example, 0 and 1 are not in the same component. Communication is a two way street, you need to be able to get from $i$ to $j$ and from $j$ to $i$ with positive probability.

Alex R.
  • 32,771
  • What does that have to do with anything? Period is a state property not a transition property. We speak of the period of state $i$, not of the period between $i$ and $j$. – nullUser Feb 27 '13 at 02:12
  • @nullUser: Period is precisely a "transition property." $P^n_{ii}$ is precisely the $n$ state return probability for a state $i$ which is a sum over every possible intermediate state. How else do you plan to come back to a state $i$ if not through some state $j$ which it communicates with? – Alex R. Feb 27 '13 at 03:18