(591d) Advancing Molecular Dynamics Models for Future Computers: A Case Study of Truncated Electrostatics

The growing use of GPUs and other coprocessors along with the trend towards heterogeneous computing have made it increasingly desirable to reevaluate computational methods and models with an eye toward approaches that allow for increased concurrency and data locality to fully exploit the potential performance of new computer hardware. Enhanced truncation approaches have been proposed as an alternative method to traditional approaches for evaluating long-range Coulombic interactions, as to improve computational efficiency while allowing for a tunable degree of accuracy. While these methods are particularly well suited for many-core architectures and graphics processing units (GPUs) due to the inherent fine-grain parallelism, care must be taken, particularly for systems with inhomogeneous charge distribution. We compare efficient truncation-based approximations to evaluation of electrostatic forces with the more traditional particle-particle particle-mesh (P3M) method for the molecular dynamics simulation of polyelectrolyte brush layers. We show that with the use of GPU accelerators, large-scale parallel simulations on the hybrid Titan supercomputer using P3M can be 3 times faster due to a reduction in the mesh-size required. Alternatively, using a truncation-based scheme can be up to 4 times faster than GPU-accelerated P3M and results in accurate calculation of shear velocities and disjoining pressures for brush layers. For configurations with highly nonuniform charge distributions, however, we find that it is more efficient to use P3M; for these systems, computationally efficient parametrizations of the truncation-based approach do not produce accurate counterion density profiles or brush morphologies.