Implementation of analytic derivative density functional theory codes on scalar and parallel architectures

The implementation of our Analytic Derivative Density Functional Theory code, CADPAC/DFT, is described. The code uses regular gaussian basis sets, evaluates the overlap, kinetic, nuclear attraction and coulomb integrals analytically, and the DFT integrals by quadrature. Special attention is given to the reduction in the number of quadrature points during the calculation. A detailed discussion on the analytic evaluation of first and second derivatives is given, concluding that for first derivatives one may proceed by (a) evaluating weight derivatives and working in any coordinate system or (b) not evaluating weight derivatives and working in internal coordinates. For second derivatives it is necessary to evaluate all weight derivative contributions. Supporting examples are given. An implementation of the CAD-PAC/DFT code on the iPSC/860 hypercube is described. The DFT integrals were evaluated by distributing the quadrature points in batches to each node achieving up to 90% efficiency with 16 nodes. copyright 1995 American Institute of Physics