• I. Yamazaki, X.S. Li, E.G. Ng
  LBNL, Berkeley, California

Numerical simulations to design high-energy particle accelerators give rise to large-scale ill-conditioned highly indefinite linear systems of equations that are becoming increasingly difficult to solve using either a direct solver or a preconditioned iterative solver alone. In this paper, we describe our current effort to develop a parallel hybrid linear solver that balances the robustness of a direct solver with the efficiency of a preconditioned iterative solver. We demonstrate that our hybrid solver is more robust and efficient than the existing state-of-the-art software to solve these linear systems on a large number of processors.