Binary SCF: GAMESS improvements for energy evaluation based on SCF methods

Abstract

This paper describes a new algorithm that improves computational efficiency for energy evaluation in GAMESS, named Binary SCF (BitSCF). This algorithm is based on the Conventional Self Consistent Field (CSCF) method and it presents two main innovations. The first one is the use of a bit map for the control of the 2e − integrals evaluated and previously stored in disk. The bit map reduces disk demand significantly when storing 2e − integrals. The second innovation is the use of i-loop balancing, a new proposal for static workload distribution through the nodes. The i-loop balancing optimizes parallelism in GAMESS, reducing an important sequential portion of its source code. The results obtained confirm a reduction of up to 49% in the disk demand for the 2e − integrals storage, considering different simulated molecules on a 13-node Beowulf cluster and 6-31G and STO-6G basis set. The total and CPU times stay below the times obtained with the GAMESS CSCF method for larger molecules. The total execution and CPU times were equivalent at the CSCF times for smaller molecules. The results show an improvement in the GAMESS scalability as well as a better cost vs. benefit relationship when executing the program.