Kelly’s Lemma provides a way of using the time reversal of a Markov chain to calculate its stationary distribution. In theory, it is just a rewrite of the stationary distribution and time-reversal results for continuous-time chains. However, in application, it is different; you can take a guess at the time reversal of a chain $Q’$, and if you are right, i.e., if it checks out in Kelly’s Lemma, you get the stationary distribution as a consequence.
Theorem (Kelly’s Lemma). Let be a continuous-time Markov chain with Q-matrix
. Suppose we can find positive rates
and a probability distribution
such that
for all
for all
Then is the Q-matrix of the time-reversed process, and
is the stationary distribution.
The idea of Kelly’s Lemma is that we can sometimes guess the time reversal of a Markov chain and use this to compute the stationary distribution.
Proof.
Thus is a stationary distribution for
. By a similar argument,
is also stationary for the chain with rates
.
The process is Markov, so conditioning on the present separates past and future. This implies that the reversed process
is also Markov.
We now compute the transition rates of the time-reversed process: