Molecular Dynamics Simulations on Shared-Memory Multiple Processor Computers

A parallel algorithm for molecular dynamics simulations of polyatomic molecular systems is presented. The algorithm is optimized for shared-memory multiple processor computers by assigning each molecule to be a parallelization unit which is usually taken for each atom or interaction pair. A method to distribute all interaction pairs to each molecule in almost equal numbers is described. The algorithm is tested by simulation of a liquid crystal molecule system on four-and eight-processor workstations. Overall speedups of 3.6 and 5.7 are obtained on four-and eight- processors, respectively. A non-iterative algorithm to constrain bond-length degrees of freedom is also tested with regard to its efficiency for parallel computation. A performance superior to that of the iterative SHAKE algorithm was obtained