Fractional step method for stochastic evolution equations (Q1965900): Difference between revisions

A stochastic differential equation \[ dX = F(t,X) dt + B(t,X) dM + G(t,X) dt + C(t,X) dN \tag{1} \] in a Hilbert space \(H\) is considered, where \(M\), \(N\) are (possibly cylindrical) martingales and the operators \(F\), \(G\), \(B\) and \(C\) are in general unbounded, nonlinear and random. Very roughly speaking, the fractional step method consists in splitting the equation (1) into two equations \[ dY=F(t,Y) dt + B(t,Y)dM, \qquad dZ=G(t,Z) dt + C(t,Z) dN \] which are then recursively solved on (short) time intervals of some partition in such a way that the solution to the first equation provides an initial datum for the second one and vice versa. As the mesh of the partition goes to zero one obtains a sequence of processes that might approximate a solution of (1). Under some assumptions on the coefficients of the equation (1) (which are too technical to be listed here) the authors prove that the fractional step procedure converges and they use this result to establish existence of solutions to some quasilinear stochastic partial differential equations. Further, they apply the method to find invariant sets for the solution to (1), which implies comparison and positivity theorems. In this respect, particular attention is paid to bilinear stochastic reaction-diffusion equations. The proofs are rather involved, however, the fractional step method covers certain types of equations to which more traditional approaches do not seem to be applicable.