Highly Scalable and Memory Efficient Ultra-Coarse-Grained Molecular Dynamics Simulations.

ABSTRACT

The use of coarse-grained (CG) models can significantly increase the time and length scales accessible to computational molecular dynamics (MD) simulations. To address very large-scale phenomena, however, requires a careful consideration of memory requirements and parallel MD load balancing in order to make efficient use of current supercomputers. In this work, a CG-MD code is introduced which is specifically designed for very large, highly parallel simulations of systems with markedly non-uniform particle distributions, such as those found in highly CG models having an implicit solvent. The CG-MD code uses an unorthodox combination of sparse data representations with a Hilbert space-filling curve (SFC) to provide dynamic topological descriptions, reduced memory overhead, and advanced load-balancing characteristics. The results of representative large-scale simulations indicate that our approach can offer significant advantages over conventional MD techniques, and should enable new classes of CG-MD systems to be investigated.