Optimization of the Ewald method for calculating Coulomb interactions in molecular simulations
Kenton D. Hammonds, D. M. Heyes
Abstract
Practical implementations of the Ewald method used to compute Coulomb interactions in molecular dynamics simulations are hampered by the requirement to truncate its reciprocal space series. It is shown that this can be mitigated by representing the contributions from the neglected reciprocal lattice vector terms as a simple modification of the real space expression in which the real and reciprocal space series have slightly different charge spreading parameters. This procedure, called the α' method, enables significantly fewer reciprocal lattice vectors to be taken than is currently typical for Ewald, with negligible additional computational cost, which is validated on model systems representing different classes of charged system, a CsI crystal and melt, water, and a room temperature ionic liquid. A procedure for computing accurate energies and forces based on a periodic sampling of an additional number of reciprocal lattice vectors is also proposed and validated by the simulations. The convergence characteristics of expressions for the pressure based on the forces and the potential energy are compared, which is a useful assessment of the accuracy of the simulations in reproducing the Coulomb interaction. The techniques developed in this work can reduce significantly the total computer simulation times for medium sized charged systems, by factors of up to ∼5 for those in the classes studied here.