Suppose as in [8] (3.41) that X with state space E admits a jointly measurable local time process relative to a reference measure dx on E, so that for
For example, X could be a Markov chain with countable state space, with , or a one-dimensional diffusion [56].
Fix a Markov killing time T for X. Then serves as a density for the Green's kernel of X killed at time T. From (22), for all there is the formula
So for each a such that the covariance
is a symmetric non-negative definite function of . Suppose now that a finite subset F of E is such that , for all . The results of Subsection 3.6 show that for any initial distribution on F the joint distribution of is determined by the values of the Green's function g(x,y) for . In particular, all product moments for non-negative integer n(y) have finite values which can be read from (27). And the joint moment-generating function of the converges in a neighborhood of the origin and is given there by the formula
where . Put another way, (57) states that for v in a neighborhood of , the function
is the unique solution f of the system of equations
For X a one-dimensional diffusion, Ray [55] (2.1) derived this system of equations for an h-process obtained conditioning on , and went on to show that these equations imply the Ray-Knight descriptions for the distribution of local times of one-dimensional diffusions stopped at a Markov killing time. See also Sheppard [61] who recovered most of Ray's results with the help of Dynkin's isomorphism theorem, and [53] for further discussion.
Note that the class of possible finite-dimensional distributions for as above is precisely the class of joint distributions of total occupation times of various states in a finite state Markov chain. This can be understood by considering the time-changed Markov chain where is the inverse of . Williams [65, 66] used a similar time change argument to derive variations of formula (55) for local time processes associated with both ordinary and fictitious states of a countable state Markov chain. See Theorem 6.1 of [66].
An altogether different application of the F-K formula to the local times of one-dimensional Lévy processes can be found in [6]. This work concerns the law of the Hilbert transform of with respect to x (for certain random T); it extends and simplifies [23], in which Kac's moment formula (4) is used. See also [31] regarding connections between the F-K formula and path decompositions for one-dimensional BM.
Acknowledgement. We are grateful to Jay Rosen for his illuminating comments on the manuscript.