Development of buffer method in parallelized molecular orbital calculation

The development of a modified direct SCF-MO method or buffer method to improve the parallelization efficiency in Fock matrix generation was discussed. The parallelization efficiency was improved by using a PC cluster that does not have secondary storage on each processor. Part of the electron repulsion integrals was stored in a buffer and the stored integrals were reused in subsequent SCF cycles. The results show that this simple method achieved super-linear scalability.