Incorporating Krylov Subspace Methods in the ETDRK4 Scheme

Abstract

A modification of the (2,2)-Pade algorithm developed by Wade et al. for implementing the exponential time differencing fourth order Runge-Kutta (ETDRK4) method is introduced. The main computational difficulty in implementing the ETDRK4 method is the required approximation to the matrix exponential. Wade et al. use the fourth order (2,2)-Pade approximant in their algorithm and in this thesis we incorporate Krylov subspace methods in an attempt to improve efficiency. A background of Krylov subspace methods is provided and we describe how they are used in approximating the matrix exponential and how to implement them into the ETDRK4 method. The (2,2)-Pade and Krylov subspace algorithms are compared in solving the one and two dimensional Allen-Cahn equation with the ETDRK4 scheme. We find that in two dimensions, the Krylov subspace algorithm is faster, provided we have a spatial discretization that produces a symmetric matrix.