Methods of reflections: relations with Schwarz methods and classical stationary iterations, scalability and preconditioning. (Q2287384)

First, the authors of this article consider the method of reflections. This is a method successively approximating the interaction of particles within a fluid. Second, they consider the Schwarz domain decomposition method, where the subdomains are considered as the complements of the particle domains. The authors show that there is a direct correspondence between those two methods. With this, they are able to provide a new convergence analysis which is based on the maximum principle yielding now precise convergence estimates that one could not obtain otherwise. Further, they show that the two methods are different in the case of more than two particles/subdomains. Next, they show using the substructuring techniques from domain decomposition that the method of reflections are classical block Jacobi and block Gauss-Seidel methods for the interface traces. With this, new relaxed versions of the method of reflections are derived having better convergence properties. Finally, the authors give coarse corrections for the method of reflections in order to make them scalable when the number of particles become large. Numerical experiments complement the theory. The scalability and the possibility of using preconditioning in the Krylov methods are illustrated.