Parallel computing applied to the molecular dynamics simulations

This paper discusses the use of high-performance algorithms for modeling the dynamics of dispersed systems using the method of molecular dynamics. Large-scale modeling, which makes it possible to determine various thermodynamic parameters and control the processes of phase transformations, requires consideration of a large number of particles, which imposes significant restrictions on the computational capabilities of the system. In work various problems of dynamics of disperse systems are considered. To solve the problems posed by the authors, a specialized data structure was developed, which reduces the computational complexity of the algorithm from quadratic to linear. The specificity of the molecular dynamics method makes it possible to achieve significant acceleration when using computations at heterogeneous stations equipped with a central processing unit (CPU) and graphic processing units (GPU). The authors showed that the presented technique can be effectively used for solving problems related to modeling processes occurring in areas with characteristic sizes of tens of nanometers on personal computers equipped with one or more GPUs. (paper)