| Abstract: | The earlier numerical multiconfiguration Hartree–Fock atomic structure package was not designed with high-performance computers in mind. In this paper, some new algorithms based on spline–Galerkin methods are described that are appropriate for concurrent/vector architectures. The goal is to improve the level of numerical accuracy by several orders of magnitude using fewer basis functions than points in a numerical grid. Of critical importance is the robustness of the code: The most serious problems in the numerical implementation were associated with orthogonality constraints. In a spline basis approach, the orthogonality requirements can be integrated into quadratically convergent update procedures. These procedures are evaluated for a number of cases. |
