Time change of an inhomogeneous Poisson process (source code)

= Time change of an inhomogeneous Poisson process

Let an <Inhomogeneous Poisson process> $N$ have cumulative intensity $\Lambda(t)=\int_0^t\lambda(u)\,du$. Suppose $\Lambda$ is continuous, strictly increasing and unbounded, and put $g=\Lambda^{-1}$. Then $M_t=N_{g(t)}$ is a rate-one <Poisson process>: its disjoint increments are <independent random variables> and
$$
M_t-M_s\sim\operatorname{Poisson}(\Lambda(g(t))-\Lambda(g(s)))
=\operatorname{Poisson}(t-s).
$$
Conversely, a rate-one <Poisson process> $M$ gives $N_t=M_{\Lambda(t)}$, with <Poisson distribution> of mean $\Lambda(t)$. For example, $\lambda(t)=2t$ gives $\Lambda(t)=t^2$ and $M_u=N_{\sqrt u}$.

Unboundedness matters if the new process is to exist at every nonnegative time. The strictly positive rate $\lambda(t)=e^{-t}$ instead gives $\Lambda(t)=1-e^{-t}<1$: its inverse is defined only for $0\leq u<1$. Strictly positive intensity alone does not justify a rate-one time change on the entire half-line.