Abstract
A strategy is described for a fast all-atom molecular dynamics simulation of
multimillion-atom biological systems on massively parallel supercomputers. The strategy is
developed using benchmark systems of particular interest to bioenergy research, comprising
models of cellulose and lignocellulosic biomass in an aqueous solution. The approach involves
using the reaction field (RF) method for the computation of long-range electrostatic interactions,
which permits efficient scaling on many thousands of cores. Although the range of applicability
of the RF method for biomolecular systems remains to be demonstrated, for the benchmark
systems the use of the RF produces molecular dipole moments, Kirkwood G factors, other
structural properties, and mean-square fluctuations in excellent agreement with those obtained
with the commonly used Particle Mesh Ewald method. With RF, three million- and five million atom
biological systems scale well up to ∼30k cores, producing ∼30 ns/day. Atomistic simulations
of very large systems for time scales approaching the microsecond would, therefore, appear
now to be within reach.
