The backward canonical representations and interpolations for multiple Markov Gaussian processes (Q1208954)

It is introduced a concept of backward representation of Gaussian processes, i.e. for which time development is observed backward. The existing theory of representation may now be called the forward representation. The main result says that the backward and forward so- called Goursat representations of multiple Markov Gaussian processes are completely characterized by the maximal and minimal solutions for an equation derived from the covariance function. By using it, a concrete expression of the interpolation is obtained for a multiple Markov process.