Abstract

The parallel scaling of classical molecular dynamics simulations is limited by the communication of the 3D fast Fourier transform of the particle-mesh electrostatics methods, which are used by most molecular simulation packages. The Fast Multipole Method (FMM) has much lower communication requirements and would, therefore, be a promising alternative to mesh based approaches. However, the abrupt switch from direct particle-particle interactions to approximate multipole interactions causes a violation of energy conservation, which is required in molecular dynamics. To counteract this effect, higher accuracy must be requested from the FMM, leading to a substantially increased computational cost. Here, we present a regularization of the FMM that provides analytical energy conservation. This allows the use of a precision comparable to that used with particle-mesh methods, which significantly increases the efficiency. With an application to a 2D system of dipolar molecules representative of water, we show that the regularization not only provides energy conservation but also significantly improves the accuracy. The latter is possible due to the local charge neutrality in molecular systems. Additionally, we show that the regularization reduces the multipole coefficients for a 3D water model even more than in our 2D example.

Highlights

  • Molecular dynamics (MD) simulations are powerful for studying the static and dynamic behavior of molecules

  • It is well known that the Fast Multipole Method (FMM) introduces a large energy drift in the MD simulation unless full machine precision is used

  • We used the idea of Chartier et al.7 to develop a modified version of the FMM, which eliminates the long term energy drift

Read more

Summary

INTRODUCTION

Molecular dynamics (MD) simulations are powerful for studying the static and dynamic behavior of molecules. The problem with these particle-mesh based methods is that they use a 3D fast Fourier transform to solve the Poisson equation in Fourier space This requires a total of four transposes of the charge grid per integration step. Both hierarchically approximate that long-range interactions contribute less to the error in the final potential and forces. The FMM assigns particles spatially to cells and approximates interactions between more distant cells This splitting leads to discontinuities in the potential, which violate energy conservation in MD simulations. 3. Use of a multipole acceptance criterion (MAC) to optimize the interaction list size for the regularized FMM. Use of a multipole acceptance criterion (MAC) to optimize the interaction list size for the regularized FMM Without these techniques, the practical performance of the FMM will suffer from the overhead of the regularization.

Plain FMM
1: Tree construction 2: Upward pass
MAC: Multipole Acceptance Criterion
REGULARIZED FMM
Computational complexity and cost
NUMERICAL RESULTS
MULTIPOLES FOR WATER IN 3D
CONCLUSIONS
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call