Parallelization of a finite volumes discretization for anisotropic diffusion problems using an improved Schur complement technique (Q2056487)

In this paper a new algorithm for combining a finite volume method with an improved Schur complement technique is presented. The method is directed towards solving 2D anisotropic diffusion problems on general meshes. The convergence of the finite volume scheme is analysed thoroughly. To reduce storage and calculation time, a nonoverlapping domain decomposition strategy is discussed and summarized in an algorithm. Associated to this decomposition, the numerical scheme based on the Schur decomposition (DDFV-SC) is introduced which is an improvement over previous approaches as it removes the vector of unknowns on the interface. The theoretical results are complemented by several tests in 2D comparing the previous DDFV and new DDVF-SC algorithms. The addressed cases concern mild anisotropy, theterogeneous rotating anisotropy and discontinuous anisotropy diffusion. The experimental order of convergence is assessed as well as the CPU times underlying the advantages of the proposed algorithm.