Scalability and Performance of a Program that Uses Domain Decomposition for Monte Carlo Simulation of Molecular Liquids

The main factors hindering the development of supercomputer programs for molecular simulation by the Monte Carlo method within the framework of classical physics are considered, and possible ways to eliminate the problems that arise in this case are discussed. Thus, the use of molecular models with moderate stiffness of covalent bonds between fragments makes it possible not only to increase the efficiency of scanning the configuration space of the model, but also to abandon the complex apparatus of kinematics with rigid links, which significantly limits the possibilities of domain decomposition. Based on the domain decomposition strategy and a simplified treatment of the deformation energy of covalent bonds and angles, an original parallel algorithm for calculating the properties of large all-atomic models of aqueous solutions of biopolymers by the Monte Carlo method was developed. To speed up computations within the framework of this approach, each domain is assigned its own group of processors/cores using local data replication and splitting the loop over the interacting partners. The article discusses the logical scheme of the computational algorithm and the main components of the software package (fortran77, MPI 1.2). Test calculations performed for water and n-hexane demonstrated the high performance and scalability of the program in which the proposed algorithm was implemented © The Author 2022. This paper is published with open access at SuperFri.org