Chapman-Kolmogorov equations, continuous time Markov chains, why do they have this solution.

379 Views Asked by At

$P^{'}(t)=QP(t)=P(t)Q$

These are the Chapman-Kolmogorov backward and forward equations for a countable sate space continuous time Markov Chain. They have the solution,

$ P(t)=e^{Qt} $

I have two questions

1) in the finite state space case, why is this a solution?

2) In the case where we have an infinite state space and the above are the infinite matrices how do we get this as a solution? I know this is a much harder question so maybe just a brief outline of an answer will help. Does the problem come from bringing the derivative inside the infinite sum?

1

There are 1 best solutions below

0
On

1) This just the solution of this ODE. You can compute it using for instance the definition of the matrix exponential: $$e^A = \sum_{j=0}^\infty \frac{1}{j!}A^j.$$ We have $$P(t) = e^{tQ} = \sum_{j=0}^\infty \frac{1}{j!}t^jQ^j.$$ When differentiating $P(t)$ w.r.t. to $t$, you obtain $$P'(t) = \sum_{j=1}^\infty \frac{1}{j!}(jt^{j-1})Q^j = \left(\sum_{j=1}^\infty \frac{1}{(j-1)!}t^{j-1}Q^{j-1}\right)Q = P(t)Q.$$ Equivalently you can bracket the single $Q$ out to the left and obtain $P'(t) = QP(t)$.

2) I think, when using the definition of the matrix exponential above, you can prove the same statement also for the infinite state space case.