Leapfrog time-stepping for Hermite methods (Q2316219)

The authors consider initial-boundary value problems for a class of linear first-order hyperbolic systems. Finite difference time domain methods yield numerical solutions of these partial differential equations. \textit{J. Goodrich} et al., [Math. Comput. 75, No. 254, 595--630 (2006; Zbl 1103.35065)] used Hermite interpolation and a staggered grid to derive methods of high order. Now, the authors investigate a variant, which combines Hermite interpolation, leapfrog time-stepping, and staggered grids in both space and time. The Hermite interpolation operator applies derivative data up to some order \(m\). It follows that the method is consistent of order (at least) \(2m\), provided that the exact solution is sufficiently smooth. The authors show the stability of the method in one as well as in multiple space dimensions, where the Fourier transform of the solution is applied. The convergence is proven only in one space dimension. Results of numerical computations are presented for wave equations in one space dimension, which confirm the predicted convergence orders. In the case of even \(m\), the estimated convergence rates indicate an actually higher order of \(2m+2\). Moreover, wave equations are considered in two space dimensions, where the results indicate that the method is also convergent with rates as in the one dimensional case. Furthermore, the authors examine computations on graphics processing units in an example with three space dimensions. The performance demonstrates a potential for accelerated computations.