A high accuracy defect-correction multigrid method for the steady incompressible Navier-Stokes equations (Q1339546): Difference between revisions

The authors discuss a high accuracy defect-correction multigrid method for the solution of steady incompressible Navier-Stokes equations. The discretized form of the Navier-Stokes equations often requires the solution of a large system of equations. Increase in the number of equations to be solved is associated with a higher cost per iteration, thereby limiting the practical size of the problem that can be solved. Applying a higher-order method which decreases the number of equations while preserving high accuracy can partially alleviate this problem. Consequently a fourth-order approach to the solution of the Navier-Stokes equations is considered. Although the higher-order scheme will reduce the size of the ensuing linear system, for two- and three-dimensional problems the size will still be large. One way of overcoming this difficulty is through multigrid techniques which rapidly accelerate the convergence of the underlying iteration schemes. However, the authors consider a defect-correction multigrid approach. This technique uses only the fourth-order discretization for the defect evaluation on the finest multigrid level. For the rest of the algorithm it employs a second-order discretization, further improving the overall efficiency of the algorithm. This approach would also provide an efficient implementation in a parallel architecture environment since the defect can be evaluated separately from the multigrid process. Some numerical comparisons are made between the defect-correction approach and a standard multigrid implementation with a fourth-order discretization. In all the cases the defect-correction approach is shown to be more efficient for a given tolerance. Further a general family of second-order discretization is suggested and the effect that this parametrized family has on the performance of the defect-correction multigrid techniques is investigated numerically. An analysis is given of the convergence behaviour of this parameterized scheme when applied to the Poisson equation on a unit square with Dirichlet boundary conditions. The theoretical results are shown to agree closely with numerical results.