Convergence Rates in Liquid Simulations – Københavns Universitet

Convergence Rates in Liquid Simulations

Master Thesis Defense by Sune Løje

Abstract:

In this thesis, the Stable Fluids method isimplemented on a GPU using the CUDA programming language. The Poisson-pressure equation is initially solved by using a traditional Jacobi iterative solver and its poor convergence is demonstrated. Two alternative methods, the Geometric Multigrid and Conjugate Gradient, are examined and implemented as replacement for the Jacobi solver. During this process, it is shown how the Geometric Multigrid and Conjugate Gradient can be efficiently implemented in terms of an implicit A matrix by reusing code from the Jacobi solver. The convergence capabilities of the different solvers are demonstrated and analyzed using three liquid configurations.

Censor: Niels Jørgen Christensen
Supervisor: Kenny Erleben