Nucleation of nanoparticles in a coarse grained fluid using OpenCL

Abstract

In this thesis, the nucleation rate of almost hard spheres in a course-grained fluid is measured to study the effects of an explicit solvent on the nucleation rate. Previous measurements show a discrepancy between physical measurements and simulations, where the latter all used implicit solvents. In this thesis, the fluid is approximated using Stochastic Rotation Dynamics (SRD), which natively contains Brownian dynamics as well as long range hydrodynamic forces between particles, and obeys the Navier-Stokes equations. The Poisseuille flow of the SRD fluid is examined, and found to match the Navier-Stokes equations. We also measured diffusion and velocity autocorrelation of nanoparticles in the fluid. The nucleation rate matches experimental data, but is in significant disagreement with other simulations. The nucleation rate progresses as a less steep curve than implicit-solvent methods and soft-particle simulations suggest. Since nucleation simulations are time-consuming, we used a graphics card in combination with OpenCL to speed up calculations. This sped up computation by roughly 40 times compared to standard processors. However, we found that specific attention must be paid to parallelisation issues and memory optimisations in this case.