Parallelization of multireference perturbation calculations with GAMESS

The quasi-degenerate multireference second-order perturbation theory (MRMP2 ) routines in the GAMESS suite of program codes have been parallelized usin g a distributed data interface (DDI). Two typical kinds of molecules were c hosen for examination of parallelization speedup using one to eight PCs gat hered as a cluster and connected by Fast Ethernet. The first example, in wh ich total energies of several low-lying electronic states have been obtaine d for niobium monohydride, give parallelization speedup of 7.15 when eight PCs were used as a cluster. The second example is the ground-state total en ergy for a medium sized molecule, 4a,4b,8a,9a-tetrahydro-pyridino[1 ' ,2 ' -4,3]imidazolidino[1,5-a]pyridine. When distributed memory is employed, the parallelization speedup improves to 6.84 for the MRMP2 calculations when a n eight-PC cluster is used. These results demonstrate that our efforts to a chieve the parallelization of MRMP2 routines have been successful. (C) 2001 John Wiley & Sons, Inc.