In this paper, we present new types of exponential integrators for Stochastic Differential Equations (SDEs) that take advantage of the exact solution of (generalized) geometric Brownian motion. We examine both Euler and Milstein versions of the scheme and prove strong convergence, taking care to deal with the dependence on the noise in the solution operator. For the special case of linear noise we obtain an improved rate of convergence for the Euler version over standard integration methods. We investigate the efficiency of the methods compared with other exponential integrators for low dimensional SDEs and high dimensional SDEs arising from the discretization of stochastic partial differential equations. We show that, by introducing a suitable homotopy parameter, these schemes are competitive not only when the noise is linear, but also in the presence of nonlinear noise terms. Although our new schemes are derived and analysed under zero commutator conditions (1.2), our numerical investigations illustrate that the resulting methods rival traditional methods even when this does not hold.
4 March 2009 in International Journal for Numerical Methods in Engineering
Purpose: The purpose of this study is to propose a non-classical method to obtain efficient and accurate numerical solutions of the advection–diffusion–reaction equations. Design/methodology/approach: Unlike conventional numerical methods, this study proposes a numerical scheme using outer Newton iteration applied to a time-dependent PDE. The linearized time dependent PDE is discretized by trapezoidal rule, which is second order in time, and by spline-based finite difference method of fourth order in space. Findings: Using the proposed technique, even when relatively large time step sizes are used in computations, the efficiency of the proposed procedure is very clear for the numerical examples in comparison with the existing classical methods. Originality/value: This study, unlike these classical methods, proposes an alternative approach based on linearizing the nonlinear problem at first, and then discretizing it by an appropriate scheme. This technique helps to avoid considering the convergence issues of Newton iteration applied to nonlinear algebraic system containing many unknowns at each time step if an implicit method is used in time discretization. The linearized PDE can be solved by implicit time integrator, which enables the use of large time step size.
1 January 2022 in Journal of Computational and Applied Mathematics