Parallel implementation of the treecode Ewald method

Summary form only given. Here, we discuss an efficient parallel implementation of the treecode Ewald method for fast evaluation of long-range Coulomb interactions in a periodic system for molecular dynamics simulations. The parallelization is based on an adaptive decomposition scheme using the Morton order of the particles. This decomposition scheme takes advantage of the data locality and involves minimum changes to the original sequential code. The message passing interface (MPI) is used for interprocessor communications, making the code portable to a variety of parallel computing platforms. We also discuss communication and performance models for our parallel algorithm. The predicted communication time and parallel performance from these models match the measured results well. Timing results obtained using a system of water molecules on the IA32 Cluster at the Ohio Supercomputer Center demonstrate high speedup and efficiency of the parallel treecode Ewald method.