Explicit and Parallelised Octupole and Hexadecapole Moment Interactions in Flexible Smooth Particle Mesh Ewald
Year: 2026
Type: preprint
Abstract: Long-range electrostatic interactions have long been known to be important in the accurate simulation of biomolecular systems and molecular crystals. Despite this fact, these interactions are absent in many machine learning force fields aiming to improve upon classical potentials. In the worst cases, this omission can lead to serious discrepancies with experimental results. FFLUX is a machine learning force field that predicts atomic energies and multipole moments up to the hexadecapole moment on-the-fly during molecular dynamics simulations. The predicted multipole moments are fed into an adapted smooth particle mesh Ewald algorithm allowing long-range electrostatic interactions to be calculated. Here we present refactoring of the Ewald routines and a new implementation of routines handling interactions involving octupole and hexadecapole moments. The new implementation leads to a speed-up of up to 10.5 times in serial compared to our previous implementation. Additionally, a hybrid parallelisation scheme using a combination of Message Passing Interface and OpenMP is made available to these high-electrostatic-rank calculations. This parallelisation offers speed-ups of up to two orders of magnitude, improving the feasibility of bulk simulations with high-rank electrostatic interactions using FFLUX force fields.
Source: ChemRxiv
Language: English
Subfield: Atomic and Molecular Physics, and Optics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7213452149