Recent advances in sparse direct solvers

International audienceDirect methods for the solution of sparse systems of linear equations of the form A x = b are used in a wide range of numerical simulation applications. Such methods are based on the decomposition of the matrix into a product of triangular factors (e.g., A = L U ), followed by triangular solves. They are known for their numerical accuracy and robustness but are also characterized by a high memory consumption and a large amount of computations. Here we survey some research directions that are being investigated by the sparse direct solver community to alleviate these issues: memory-aware scheduling techniques, low-rank approximations, and distributed/shared memory hybrid programming