next up previous
Next: About this document ...

Yury Gryazin
Simplified Krylov subspace algorithm for a high-order compact approximation of 3-D Helmholtz equation.

Mathematics Department
Idaho State University
Stop 8085
Pocatello ID 83209
gryazin@isu.edu

In recent years, the problem of increasing the resolution of existing numerical solvers has become an urgent task in many areas of science and engineering. Most of the existing efficient solvers for structured matrices were developed for lower-order approximation discretized partial differential equations. The need for improved accuracy of underlying algorithms leads to modified discretized systems and as a result to the modification of the numerical solvers. As an example, one can consider a modification of Fast Fourier Transform (FFT) type algorithms for the solution of the Helmholtz equation when the accuracy of the discretization of the original equation changes from a second to a fourth order approximation. In many situations, this modification has been successfully implemented but it usually has taken a major revision of the underlying numerical algorithm. In this paper, we consider the use of the lower-order approximation solver as a preconditioner for an efficient implementation of a high-order approximation scheme for the Helmholtz equation in the Krylov subspace method framework. As a first step, we develop a compact sixth-order approximation finite-difference scheme for the 3D Helmholtz equation with Dirichlet, Neumann and Sommerfeld-like boundary conditions. Then we consider the solution of the proposed discretized system by using the Generalized Minimal Residual (GMRES) method with a second order approximation system as a preconditioner. The numerical implementation of this approach requires the solution of the low-order approximation system on each step of the iterative algorithm, which can be efficiently done by a FFT-type method. The convergence analysis of the GMRES method in the case of a 1-D preconditioned system suggests a simplified Krylov subspace method which theoretically converges slower than the original GMRES algorithm but that does not require the Arnoldi orthogonalization procedure. We will use the abbreviation SKSM to describe this algorithm. In our numerical experiments, we demonstrate the accuracy of the proposed compact six-order approximation scheme and the efficiency of the proposed iterative procedures based on the combination of GMRES-FFT and SKSM-FFT. Both methods exhibit excellent convergence in most considered test problems and can be viewed as general blueprints for the improvement of the accuracy of existing lower-order approximation solvers. We also present the results of a numerical experiment which demonstrates that in some situations the GMRES-FFT combination experiences stagnation and needs to be restarted, while the SKSM-FFT method continues to exhibit excellent convergence properties.




next up previous
Next: About this document ...
root 2012-02-20