HACPar: An efficient parallel multiscale framework for hybrid atomistic–continuum simulation at the micro- and nanoscale

Introduction

The nanoscale fluid phenomena play a quite important role in many applications, such as micro- and nanoscale channel flow and boundary layer flow.^1–3 Micro-electromechanical system (MEMS) devices and lab-on-a-chip devices are two typical cases that feature flow at micro- and nanometer scales. In order to investigate these phenomena, researchers usually use the multiscale modeling methods. The advantages of multiscale modeling lie in the ability of revealing properties of multiscale systems by capturing phenomena that appear in a wide range of time and length scales, which are beyond any single solver and method.^4,5 Especially, the hybrid atomistic–continuum (HAC) method based on geometric coupling has seen rapid development since it allows on-the-fly information exchange and interaction between multiple simulation regions.

In the previous research, there are few investigations on the coupling simulation framework and coupling-oriented parallel issues. Delgado-Buscalioni⁶ proposes a coupling simulation framework for molecular-continuum simulation that deployed on grid-like computer architectures. Tang et al.⁷ provide a universal interface for multiscale simulation but based on only particle framework. For the existing molecular dynamics (MD) software, FrantzDale et al.⁸ extend the multiscale simulation ability of LAMMPS⁹ software. For the mesh-particle configuration of the coupling framework, Neumann et al.¹⁰ propose the MaMico framework that provides interfaces for multiscale simulation and several investigations on parallel issues. Cosden and Lukes¹¹ design a preliminary framework based on OpenFOAM¹² and LAMMPS⁹ but less of deep parallelization design and abstraction.

Although there have been substantial theoretical research, there are still nontrivial challenges when turning the theory into practice. In the previous research, the researcher often uses the house design codes and performs several work on the open-source software frameworks. Nevertheless, the designers tend to regard the open-source software as the black box or the gray box and lack the deep investigation of inside software configuration. Therefore, the coupling computational platform serves less effectively and systematically. Although the open-source platforms such as OpenFOAM and LAMMPS have certain parallel optimization, there is still an urgent need of how to design the optimized parallel issue oriented to coupling methods which has not been in-depth study. Moreover, the field experts have less experience on hand-on modifying the existing frameworks when they want to add new coupling strategies,...