On the splitting-up method and stochastic partial differential equations (Q1394518)

The authors consider stochastic partial differential equations essentially (generalisations are also given) of the form \[ du(t,x) = Lu(t,x)dt + M_ku(t,x) \circ dw_t^k, \quad x \in \mathbb{R}^d, \;t\in [0,T], \tag{1} \] and some initial data \(u_0 = u_0(x)\). Here \(w_t^k\) are \(k\) independent one-dimensional Wiener processes and equation (1) is to be interpreted in the Stratonovich sense. The \(M_k\)'s are first-order operators and \(L\) is a second-order elliptic operator, all acting on functions defined on \(R^d\). The authors propose and analyse splitting-up methods for the numerical approximation of the solution of (1). By stretching out the time-scale in a certain way they can reformulate the splitting-up method as a stochastic equation and apply stochastic calculus for their proofs (previous proofs relied on semigroup theory). One of their theorems says that for each \(T>0\) and \(p>0\) there is a constant \(N\) such that \(E \max_{t \in T_n} \|u_n(t) -u(t)\|_0^p \leq N/h^p\) for all \(n>1\), where \(u_n(t)\) is the numerical approximation to (1), \(E\) the expectation, \(T_n\) a grid on \([0,T]\) and \(\|.\|_0\) the usual \(L_2\)-norm on \(R^d\). Auxiliary results, the proof of the above theorem, as well as generalisations of it are presented in Sections 2 to 5. The last section provides an application to nonlinear filtering.