The Design of a New Out-of-Core Multifrontal Solver

Direct methods for solving large sparse linear systems of equations are popular because of their generality and robustness. Their main weakness is that the memory they require increases rapidly with problem size. We discuss the design and development of a new multifrontal solver that aims to circumvent this problem by allowing both the system matrix and its factors to be stored externally. We highlight some of the key features of our new out-of-core package, in particular its use of efficient implementations of dense linear algebra kernels to perform partial factorizations and its memory management system. We present numerical results for some large-scale problems arising from practical applications.