Fast numerical contour integral method for fractional diffusion equations (Q283284)

This paper is concerned with the numerical solution of one- and two-dimensional space-fractional diffusion equations. After applying spatial discretization by appropriate finite difference schemes, large and structured systems of ordinary differential equations are obtained. Instead of traditional time-stepping methods, the authors use the numerical contour integral method based on efficient schemes for the Laplace inversion. The key points for the success of the integral method is to choose a suitable type of contour and to parametrize it optimally. Here, the authors choose the hyperbolic contour due to the sectorial nature of the spatial discretized matrices. By exploiting the Topelitz-like structure of the coefficient matrices, concrete regions containing the spectra of the coefficient matrices are given. The resolvent norms of the resulting coefficient matrices are shown to be bounded outside of these regions. According to these regions, suitable parameters of the hyperbolic contour are determined. To reduce the computational cost of the numerical contour integral method, the pre-conditioned GMRES method with circulant preconditioners is used to solve the resulting shifted linear systems. Three numerical examples are given for illustration.