Analysis and Improvments of the Memory Usage of a Multifrontal Solver

We are concerned with the memory usage of sparse direct solvers. We particula- rly focus on the influence of state-of-the-art sparse matrix reordering techniques on the dynamic memory usage of a multifrontal solver, MUMPS, and present algorithms to modify the assembly tree traversal that aim at reducing the memory usage. A large number of experiments show the interest of the approach for sequential executions.