Abstract: Physically based fluid simulation method can obtain realistic simulation results by solving the fluid governing equation directly, but numerical dissipations are liable to occur, thus causing the loss of fluid details. In this research, it is proposed to simulate the fluid with vortex particles. Firstly, the vorticity field is obtained by solving the curl form of the governing equation, upon which it is converted into an incompressible velocity field. This method greatly reduces numerical dissipations, automatically guaranteeing that the final velocity field is divergence-free, and thus can preserve much more fluid details. In addition, aiming at the performance bottleneck of the algorithm in solving Poisson equation for the vorticity-to-velocity conversion, an efficient preconditioned conjugate gradient method is designed and implemented based on GPU (graphics processing unit) to solve this equation, which can be more than ten times faster than the existing solvers. Experimental results show that the proposed algorithm can achieve more realistic fluid simulation results than the existing methods, and the simulation speed is significantly improved.

Key words: smoke simulation, vortex particle method, preconditioned conjugate gradient, Poisson equation, graphics processing unit