The goal of the proposed Phase II research is to adapt a variety of high accurate, computationally expensive ab initio methods to run efficiently in parallel on a network of workstations. This will extend the capacity and availability of electronic structure theory methods to researchers, by multiplying the computing resources applied to the problem. Feasibility was established by Phase I research that parallelized Hartree-Fock SCF in Gaussian 92 through the use of the Linda parallel programming model and software, resulting in good speedups on networks of 4-6 workstations. Working with the Gaussian and Linda software, the principal objective is to parallelize (to the same level as in Phase I) the computation of MP2, MP3, CID, CISD, MP4, CCD, BD, QCISD, and CCSD energies, including non-interative triple excitation corrections, and gradients up to QCISD. In the process we plan to develop associated strategies for ERI transformations and iterative energy and gradient methods, to minimize data movement and redundant computations. A second Phase II objective is to enhance the parallel performance of SCF functions on larger numbers of workstations. Candidates for parallelization include ERI (shell quartet) selection, matrix diagonalization and other matrix operations used in the ERI transformation and back-transformation.
