Abstract de la publi numéro 14072

Matrices coming from elliptic PDEs have been shown to have a low-rank property : their off-diagonal blocks can be approximated by low-rank blocks. This representation offers a substantial reduction of the memory requirement and gives efficient means to perform many of the basic algebra operations. In this talk, we present how these results can be used to significantly improve multifrontal solver. Low-rank blocks matrices and hierarchical matrices are analyzed and compared. The first one approximates the fronts blockwise. The second one defines a hierarchy of blocks to recursivly approximate the fronts. Both decrease memory consumption, complexity and parallel communication costs.