Supervised parallel-in-time algorithm for long-time Lagrangian simulations of stochastic dynamics: Application to hydrodynamics

JOURNAL OF COMPUTATIONAL PHYSICS

卷:393

Article

Blumers, Ansel L.¹  Li, Zhen²  Karniadakis, George Em² 

[1] Brown Univ, Dept Phys, Providence, RI 02912 USA

[2] Brown Univ, Div Appl Math, Providence, RI 02912 USA

关键词: Multiscale modeling;    Lagrangian method;    Parallel-in-time;    Particle simulations;    Dissipative particle dynamics;   

DOI  :  10.1016/j.jcp.2019.05.016

来源: Elsevier

【 摘 要 】

Lagrangian particle methods based on detailed atomic and molecular models are powerful computational tools for studying the dynamics of microscale and nanoscale systems. However, the maximum time step is limited by the smallest oscillation period of the fastest atomic motion, rendering long-time simulations very expensive. To resolve this bottleneck, we propose a supervised parallel-in-timealgorithm for stochastic dynamics (SPASD) to accelerate long-time Lagrangian particle simulations. Our method is inspired by bottom-up coarse-graining projections that yield mean-field hydrodynamic behavior in the continuum limit. Here as an example, we use the dissipative particle dynamics (DPD) as the Lagrangian particle simulator that is supervised by its macroscopic counterpart, i.e., the Navier-Stokes simulator. The low-dimensional macroscopic system (here, the Navier-Stokes solver) serves as a predictor to supervise the high-dimensional Lagrangian simulator, in a predictorcorrector type algorithm. The results of the Lagrangian simulation then correct the meanfield prediction and provide the proper microscopic details (e.g., consistent fluctuations, correlations, etc.). The unique feature that setsSPASDapart from other multiscale methods is the use of a low-fidelity macroscopic model as a predictor. The macro-model can be approximate and even inconsistent with the microscale description, butSPASDanticipates the deviation and corrects it internally to recover the true dynamics. We first present the algorithm and analyze its theoretical speedup, and subsequently we present the accuracy and convergence of the algorithm for the time-dependent plane Poiseuille flow, demonstrating thatSPASDconverges exponentially fast over iterations, irrespective of the accuracy of the predictor. Moreover, the fluctuating characteristics of the stochastic dynamics are identical to the unsupervised (serial in time) DPD simulation. We also compare the performance ofSPASDto the conventional spatial decomposition method, which is one of the most parallel-efficient methods for particle simulations. We find that the parallel efficiency ofSPASDand the conventional spatial decomposition method are similar for a small number of computing cores, but for a large number of cores the performance ofSPASDis superior. Furthermore, SPASDcan be used in conjunction with spatial decomposition for enhanced performance. Lastly, we simulate a two-dimensional cavity flow that requires more iterations to converge compared to the Poiseuille flow, and we observe thatSPASDconverges to the correct solution. Although a DPD solver is used to demonstrate the results, SPASD is a general framework and can be readily applied to other Lagrangian approaches including molecular dynamics and Langevin dynamics. (C) 2019 Elsevier Inc. All rights reserved.

【 授权许可】

Free