JOURNAL OF COMPUTATIONAL PHYSICS 132, 123–129 (1997) ARTICLE NO. CP965627 A Compact Multigrid Solver for Convection-Diffusion Equations Murli M. Gupta,* Jules Kouatchou, and Jun Zhang Department of Mathematics, The George Washington University, Washington, DC 20052 E-mail: *mmg@math.gwu.edu Received April 29, 1996; revised October 30, 1996 In Section 2 we present the NPF difference scheme to solve the convection-diffusion equation (1), describe its We introduce a multigrid algorithm to solve the convection- diffusion equations using a nine-point compact difference scheme. implementation with multigrid, and carry out a Fourier We test the efficiency of the algorithm with various smoothers and smoothing analysis of the Gauss–Seidel operator. In Sec- intergrid transfer operators. The algorithm displays a grid-indepen- tion 3 we present numerical experiments that demonstrate dent convergence rate and produces solutions with high accuracy. the effectiveness and accuracy of the multigrid algorithm. Numerical results are presented to validate the conclusions. 1997 The paper ends with discussion and conclusions. Academic Press 2. THE MULTIGRID IMPLEMENTATION 1. INTRODUCTION 2.1. The Compact Nine-Point Stencil Consider the two-dimensional convection-diffusion Let h(=1/ N ) be the uniform mesh-size. The finite differ- equation ence approximation for convection-diffusion equation (1) at a grid point (x, y) which is denoted by x 0 (Fig. 1) involves Lu u xx + u yy + u x + u y = f (x, y), (x, y) , (1) eight neighboring mesh points at (x h, y), (x, y h), u(x, y) = g (x, y), (x, y) , (x h, y h). These points are denoted by x i , i = 1, 2, ..., 8, and the values of a function u at the point i are denoted which often appears in the description of transport phe- by u i . A compact fourth order approximation (NPF ) of nomena. The magnitudes of and determine ratios of the convection-diffusion equation (1) is given by [6] convection to diffusion. In many problems of practical interest the convective terms dominate the diffusion and the values of and are large. Many numerical simulations 8 i=0 i u i = h 2 2 [( f 4 + f 3 + f 2 + f 1 + 8f 0 ) (2) of (1) become increasingly difficult (converge slowly or even diverge) as the ratio of convection to diffusion in- + ( f 1 - f 3 ) + ( f 2 - f 4 )], creases. For convection-diffusion equations with constant coef- where u i and i are depicted by the stencil in Fig. 1 and ficients (1), Gupta et al. [6] proposed a compact fourth- = h/2 and = h/2 are the cell Reynolds numbers. order finite difference scheme which was shown to be both When the cell Reynolds numbers and are greater accurate and cost-effective; it is also stable for all values than 1.0, the coefficient matrix is no longer an M-matrix of and . In [7], this compact nine-point formula (NPF ) (see [13] for the definition of M-matrix). The numerical was extended to solve convection-diffusion equations with experiments conducted in [6] showed that this scheme con- variable coefficients. The new scheme also has a truncation verges for any values of and when classical iterative error of order h 4 and the resulting systems of linear equa- methods are used. tions could be solved by classical iterative methods for large values of and . 2.2. The Multigrid Method In this paper, we present a method that combines multigrid techniques with NPF to solve the convection- The multigrid method is among the fastest and most efficient iterative algorithms for solving linear systems aris- diffusion equation (1). For a wide range of and , we compare the effectiveness of a number of smoothers for ing from discretizing elliptic differential equations (see [3, 12]). This method offers convergence rates independent solving the linear systems resulting from the use of NPF . We also compare the choices of the intergrid transfer oper- of the grid size and is very effective for solving large scale computation-intensive problems. Structurally, the algo- ators. 123 0021-9991/97 $25.00 Copyright 1997 by Academic Press All rights of reproduction in any form reserved.