Implementation and Parallel Optimization of the Lees-Edwards Boundary Condition in ESPResSo++

TL;DR

This paper presents a new implementation and parallel optimization of the Lees-Edwards boundary condition in ESPResSo++, enabling efficient shear flow simulations of complex fluids with validated physical results and excellent scalability on supercomputers.

Contribution

The paper introduces a novel implementation of LEbc in ESPResSo++ with a parallelization scheme that scales efficiently up to 1024 processors, validated through physical simulations.

Findings

Successful reproduction of shear velocity profiles and shear thinning.

Good agreement with previous literature results.

Linear scaling of performance up to 1024 processors.

Abstract

The Lees-Edwards boundary condition (LEbc) was first developed by Lees and Edwards in 1970s. The current implementation of LEbc in the ESPResSo++ MD software package provides a new possibility in simulating molecular or coarse grained systems under such non-equilibrium conditions, namely by introducing a shear flow which has potential applications in high-speed fluids, thermoplastic and other non-equilibrium processes etc. Using the LEbc code, shear flow simulations were carried out in model systems such as the Lennard-Jones fluids and the Kremer-Grest polymer melts. Some important physical properties and phenomena, including the linear profiles of shear velocities, non-layered density distribution and shear thinning, are successfully reproduced or captured. Results also show in good agreement with those from the previous literature even with unphysical simulation conditions, which on the other side gives a solid validation to our implementation of LEbc. Regarding the high performance computing (HPC) environments, we considered in depth and paid efforts on the parallelization of the LEbc code in ESPResSo++. By this, a modified scheme for data communication has been introduced within the framework of the domain decomposition. The benchmarks show a linear and good scaling for simulations which are parallelized up to 1024 processors in supercomputer systems.