Stationary Markov processes with continuous paths

Daniel Ray

Transactions of the American Mathematical Society · 1956 · 72 citations · 16 references

DOIFull text

Open access

Concepts

Abstract

I. Introduction.Let P(x, I, E) be a stationary transition probability function on the real line.It is well known [IV; V] that for each probability measure p., the function P determines a Markov process {$2, SF, Pr} with initial distribution p.That is, a space fl of paths = {x(r, co)|r^0J can be chosen, together with a field SF of subsets of S2 containing all sets of the form {w|a<x(/, w) <b}, and a measure Pr{ } on SF, such thattor t>0, a, b real.(Here and throughout the paper we suppress the indication of the range of an integration, if that range is the real line.)For each t>0,(1.2) m -* nTt:pTt(E) = j P(x, t, E)p(dx)defines a positivity and norm preserving operator on the space of bounded additive Borel set functions.The hypothesis that the transition probabilities are stationary is reflected in the fact that the operators { Tt \ t > 0} form a semi-group: Tt+.= TtTs.Among the semi-groups defined in this way, those satisfying the following two conditions have been called diffusion processes by Feller [VIII]: Ao: For every bounded continuous real function f and for every t>0, (1.3) Ttf(x) = E{f(x(l)) | x} = ff(y)P(x, t, dy) defines a continuous function of the real variable x.B0: The transition probabilities are of local character: for every x and for every open interval I containing x, 1 (1.4) lim -P(x, t, T) = 0, <-o t

References

16