Molecular dynamics-Lattice Boltzmann hybrid method on graphics processors

molecular dynamics;lattice boltzmann;graphics processors;wavy;bellow;hybrid;Computational fluid dynamics (CFD)

Marsh, Daniel D. ; Vanka ; Surya Pratap

关键词: molecular dynamics;    lattice boltzmann;    graphics processors;    wavy;    bellow;    hybrid;    Computational fluid dynamics (CFD);   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/16140/Marsh_Daniel.pdf?sequence=2&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Molecular Dynamics is an atomistic computational tool that has become popular due to its ability to predict nano-scale fluid phenomena.For studying flows, a non-equilibrium method is implemented involving characterization of boundary and other conditions to properly simulate the flow. Using a leap-frog integration scheme in conjunction with immersed boundary treatment and a stochastic model, this code is suitable for flow calculations in any type of geometry.Lattice Boltzmann is a meso-scale continuum solver that has also become popular recently due to its relative efficiency in solving flow patterns compared to Navier-Stokes implementations. This work utilizes a Lattice Boltzmann code which is optimized for the novel architecture.Graphics processors are a novel paradigm for performing scientific computation. With recent advances in the programming structure on graphics processors, large speedups are able to be gained at an economical price compared to traditional CPU solvers.In this study, Molecular Dynamics and Lattice Boltzmann have been coupled to solve problems in the nano-scale regime. Molecular Dynamics is utilized to solve near wall regions while Lattice Boltzmann solves the rest of the domain. This is implemented using a novel method on the Graphics Processing Unit, showing speed increases of 5-10x for purely Molecular Dynamics, and 50-75x for purely Lattice Boltzmann compared to a modern CPU. We examine physics of straight and bellow channels using the individual solvers and the hybrid solver to show the viability of these methods implemented on Graphics Processing Units.