Department of Mathematics

The George Washington University

Washington, DC 20052

and

Jun Zhang

Department of Computer Science and Engineering

University of Minnesota

4-192 EE/CS Building, 200 Union Street S.E.

Minneapolis, MN 55455, USA

We present an explicit fourth-order compact finite difference scheme for approximating the three dimensional convection-diffusion equation with variable coefficients. This 19-point formula is defined on a uniform cubic grid. Fourier smoothing analysis is performed to show that the smoothing factor of some relaxation techniques with our scheme is smaller than 1. We further design a parallelization-oriented multigrid method for fast solution of the resulting linear system. The multigrid method employs a four-color Gauss-Seidel relaxation technique for robustness and efficiency. We also propose a scaled residual injection operator to reduce the cost of multigrid inter-grid transfer operator. Numerical experiments on a $16$ processor vector computer are used to test the high accuracy of the discretization scheme as well as the fast convergence and the parallelization or vectorization efficiency of the solution method. Effects of using different residual projection operators are compared on both vector and serial computers.

Key words: 3D convection-diffusion equation, fourth-order compact discretization, multigrid method.