A Parallel Self-consistent Field Code

Abstract

We describe a version of an algorithm for evolving self-gravitating collections of particles that should be nearly ideal for parallel architectures. Our method is derived from the ``self-consistent field'' (SCF) approach suggested previously by Clutton-Brock and others. Owing to the use of a global description of the gravitational field, the particles in an SCF simulation do not interact with one another directly, minimizing communications overhead between nodes in a parallel implementation. Ideal load-balancing is achieved since precisely the same number of operations are needed to compute the acceleration for each particle. Consequently, the SCF technique is perfectly scalable and the size of feasible applications will grow in simple proportion to advances in computational hardware. We describe an SCF code developed for and tested on a Connection Machine 5. Empirical tests demonstrate the efficient and scalable nature of the algorithm. Depending on the application, simulations with particle numbers in the range $N\sim 10^7 - 10^{8.5}$ are now possible. Larger platforms should make simulations with billions of particles feasible in the near future. Specific astrophysical applications are discussed in the context of collisionless dynamics.