MOLECULAR SIMULATIONS ON SUPERCOMPUTERS

Work from our laboratory using molecular dynamics applied to the study of peptide/protein folding dynamics and thermodynamics in aqueous solution, and the exploration and development of molecular solvent interaction models including electronic polarizability is described. The basic computational techniques we have used in developing algorithms for vector/parallel, shared memory, multiple instruction, multiple data (MIMD) and massively parallel single instruction, multiple data (SIMD) architectures are discussed in the context of these application areas. The technique we have used to parallelize the molecular dynamics/molecular mechanics code CHARMM is detailed, with special focus given to the performance issues faced in developing algorithms for the CRAY Y-MP. New developments of a molecular dynamics kernel for simulation of electronically polarizable solvent systems using a parallel-extended Fortran language (Fortran-9X) on the massively parallel Connection Machine (CM-2) are outlined. We provide examples of applications using both of these platforms and discuss the development of software for such simulations in a heterogeneous computational environment.