Parallel implementation of isothermal and isoenergetic Dissipative Particle Dynamics using Shardlow-like splitting algorithms

详细信息    查看全文

• 作者：James P. Larentzos^a ; ^{James.Larentzos@engilitycorp.com" class="auth_mail} ; John K. Brennan^b ; Joshua D. Moore^b ; Martin Lí ; sal^c ; ^d ; William D. Mattson^b
• 关键词：Dissipative Particle Dynamics ; DPD ; Shardlow splitting algorithm ; Numerical integration ; Parallelization
• 刊名：Computer Physics Communications
• 出版年：July 2014
• 年：2014
• 卷：185
• 期：7
• 页码：1987-1998
• 全文大小：1241 K

文摘

A parallel implementation of the Shardlow splitting algorithm (SSA) for Dissipative Particle Dynamics (DPD) simulations is presented. The isothermal and isoenergetic SSA implementations are compared to the DPD version of the velocity-Verlet integrator in terms of numerical stability and performance. The integrator stability is assessed by monitoring temperature, pressure and total energy for both the standard and ideal DPD fluid models. The SSA requires special consideration due to its recursive nature resulting in more inter-processor communication as compared to traditional DPD integrators. Nevertheless, this work demonstrates that the SSA exhibits stability over longer time steps that justify its regular use in parallel, multi-core applications. For the computer architecture used in this study, a factor of 10–100 speedup is achieved in the overall time-to-solution for isoenergetic DPD simulations and a 15–34 speedup is achieved for the isothermal DPD simulations.