High order recombination and an application to cubature on Wiener space (Q453236)

The subject of this paper is a new method of approximation for a parabolic partial differential equation or a filtering problem. A simple algorithm that associates to a discrete measure another measure with smaller support and same center of mass is first presented. This is called the reduction operation. Then the cubature method developed by \textit{T. Lyons} and \textit{N. Victoir} [Proc. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 460, No. 2041, 169--198 (2004; Zbl 1055.60049)] following \textit{S. Kusuoka} [Adv. Stud. Pure Math. 31, 147--165 (2001; Zbl 1028.60052)] is recalled. It provides higher order approximations if the test function is Lipschitz and the vector fields satisfy Kusuoka's UFG condition. This KLV method is iterated over a partition of the time interval and after each application of the KLV operation intermediate measures are replaced by reduced measures. This dynamic recombination allows to retain the high order accuracy, whereas the computational effort involved grows polynomially with the number of time steps. A numerical one-dimensional toy example is displayed in the Appendix.