Markov processes on time-like graphs (Q717881)

Classical stochastic processes are families of random variables where the parameter-set \(T\) is a subset of \(\mathbb R\), for example, the positive integers or \([0,+\infty)\). A notable exception is the family of Gaussian processes, for which the structure of the parameter space \(T\) can be virtually arbitrary. For the other two most popular families of stochastic processes, that is, Markov processes and martingales, the situation is much more complicated. The theories of Markov processes and martingales with the parameter set \(T\) equal to an orthant in \(\mathbb R^d\) are hard and little developed. In the present paper, tha authors introduce a class of stochastic processes where the ``time'' parameter has been replaced by paths in an oriented graph. They construct a time-structure that matches the Markov property better than other non-rectilinear time sets known in the literature. The parameter-set which is considered is a deterministic one, while the structure of the graph plays a basic rôle in the existence and uniqueness of the above processes. In Section 2, a collection of graphs which support such processes, called ``time-like'' graphs with no co-terminal cells, are defined. In Section 3, the ``natural Markov process on a time-like graph'' is constructed and its uniqueness in law is proved. In Section 4, examples of graphs that do not satisfy the imposed conditions and do not support a ``natural Brownian motion'', are provided. In Section 5, the authors define the ``harnesses'', which include all integrable Lévy processes and their corresponding bridges. Finally, Section 6 is devoted to the Brownian motion on the honeycomb graph.