Strong approximations for stochastic differential equations with boundary conditions (Q1915841)

The authors present a Euler approximation scheme for the boundary value problems of stochastic differential equation (SDE). One is the one-dimensional case of \(dX_t= \sigma(X_t) dW_t+ b(X_t) dt\), \(t\in [0, 1]\), with the linear boundary condition of \(F_0 X_0+ F_1 X_1= h_0\) \((F_0, F_1, h_0\in \mathbb{R})\). The numerical approximations \(\overline X_0\) for \(X_0\) associated with various partitions \(\pi\) of \([0, 1]\) are first calculated from the stochastic flow of the SDE and the given boundary condition, and then the approximations \(\overline X_t\) are obtained via the initial value problem of the SDE. Another case is the multidimensional SDE with constant diffusion coefficient and general boundary condition, which is the same as studied by \textit{D. Nualart} and \textit{E. Pardoux} [Ann. Probab. 19, No. 3, 1118-1144 (1991; Zbl 0736.60052)], where they have proved that the solution \(X_t\) is related to a bijective nonlinear transformation \(T\) of the paths of the Wiener process. By reducing the problem to one of approximation of SDEs where the initial condition is the fixed point of some mapping related with \(T\), the approximations \(\overline X_t\) can be obtained. It is also proved that \(E\{\sup_{t\in [0, 1]} |\overline X_t- X_t|^p\}\leq C\delta^{p/2}\), \(p\geq 1\), for the first case and \(\leq C\delta^p\), \(p\geq 2\), for the second case, where \(\overline X_t\) is the obtained approximation and \(\delta\triangleq|\pi|\).