A parallel grid modification and domain decomposition algorithm for local phenomena capturing and load balancing (Q1387816)

For solving one-dimensional convection-diffusion problems \textit{P. L. Lion}'s nonoverlapping Schwarz domain decomposition method [in: T. F. Chan, R. Glowinski, J. Periaux, O. B. Widlund (eds.). Domain decomposition methods for partial differential equations, Proc. 3rd Int. Symp. Houston/TX (USA) 1989, 202-223 (1990; Zbl 0704.65090)] is applied. These problems are discretized by constructing finite difference schemes within each subdomain. Here, two variants of discretizing the convection term are discussed. In order to capture local phenomena, as e.g. fronts and layers, fine grids in subdomains that contain fronts and layers are applied. In subdomains where the solution changes slowly coarse grids are used. The choice of the relaxation parameters in Lion's method is studied. Furthermore, the implementation of the algorithm on an nCUBE2 parallel computer is discussed. Finally, some numerical experiments demonstrating the convergence properties of the iterative method and the scalability of the algorithm are presented.