Abstract
DL_FFLUX performs molecular dynamics for flexible molecules endowed with polarisable QTAIM atomic multipole moments, predicted by the machine learning method Gaussian Process Regression. Newly optimised and parallelised using domain-decomposition MPI, DL_FFLUX now operates in reasonable time frames. DL_FFLUX is delivered as an add-on to the widely distributed molecular dynamics code DL_POLY 4.08. For the systems studied here (10**3-10**5 atoms), DL_FFLUX adds minimal computational cost to the standard DL_POLY package. The parallel DL_FFLUX preserves the quality of the scaling of the MPI implementation in standard DL_POLY. For the first time it is feasible to use the full capability of DL_FFLUX to study systems that are large enough to be of real world interest. For example, a fully flexible, high-rank polarised (up to and including quadrupole moments) 1 ns simulation of a system of 10,125 atoms (3,375 water molecules) takes 30 hours (wall time) on 18 cores.
Supplementary materials
Title
FFLUX: a parallel, quantum chemical topology force field
Description
Section 1 on weak scaling contains Figures S1‐S4 showing runtime as a function of the number of atoms for varying L’ and Np values. Section 2 on the cut‐off radius gives timings in Table S1. Section 3 on L’ shows Figure S5 with timings as a function of L’ and the number of atoms. Section 4 on scaling contains Figures S6, S7 and S8, with relative speed‐ups as a function of L’ and an increasing number of atoms. Section 5 on profiling
shows Figures S9 and S10 with pie charts, and Figure S11 with the fraction of total runtime spent in MPI
communications as a function of Np. Section 6 provides technical details on optimisation.
Actions