Molecular Dynamics Simulations Of Nano Oscillatory Flows Based On GPU Acceleration Algorithm

In the experiments of superlubricity, polyelectrolyte brushes andphase transitions in confined liquids etc, the study of viscoelasticbehavior of fluids through the oscillatory flows has attracted considerableattentions. However, some conclusions obtained by different groups werenot consistent. The main reason may be the difference of experimentconditions from different research groups. Therefore, in order to avoid theadverse effects of environment, scientific research workers attempt toprocess molecular dynamics numerical simulations using computers inthe experiments.However, in the processes of the computer simulation, if we want togive a very approximate simulation results with the real situation, we have to process numerical calculation for the large number of molecules.The high computational cost of the calculation process often tends toconsume for day after days, or week after weeks, or even a few months.Due to reduce the computational time, a lot of problems have to be settledthrough some degree predigestion, although these simplification willaffect the the accuracy of the results.According to the rapid development of GPU computing method, anew numerical computation method is presented in this paper. Theprimary viewpoint is that molecular dynamics simulations could becompleted mainly on the GPU clusters based on CUDA-MPI hybidprogramming. Compared with the CPU clusters, the speedup ratio of themethod has been increased by5times. This shows that this method has acertain application value for molecular dynamics numerical simulationsof the large-scale molecular systems. On this basis, we study the numerical simulation for nanoscale fluidoscillatory flow based on the GPU algorithm. In the actual simulations,we use the GPU computing method based on domain decomposition, andthe fast Fourier transformation on the GPU clusters. It is worth to notethat the computional time of the entire simulation processes is effectivelyreduced.In the paper we investigate oscillatory flows of nanoscale fluid viamolecular dynamics simulation. To induce the oscillatory flows, we usethe constrained dynamics of hybrid continuum-particle method ratherthan the direct non-equilibrium molecular dynamics simulation. Theresults show that when the amplitude of oscillation is constant, theresponse of the velocity profiles to the frequencies rapidly become weakwith growth of the oscillation frequency. The main reason is that theparticles far from Câ†'P can’t rapidly respond to the high oscillation frequencies. In particular, at the frequency ω=1, the fluid doesn’t showobvious oscillation in the near wall region, while at ω=0.1the oscillationalmost occurs in the whole fluid region. The temperature profiles ofoscillatory flows are not symmetrical. In addition, the fast fouriertransform of shear stress indicates that the fluid exhibits linearviscoelastic behavior under given conditions. In the Nanoscale fluid,these results have a certain meanings.