Minimum residual methods for cloth simulation

This paper presents a new method for cloth simulation which improves the position based dynamics method, first popularized for games by Jakobsen. We are replacing the widely used iterative relaxation techniques (Jacobi and Gauss-Seidel) with conjugate gradient like methods (Minimum Residual and Conjugate Residuals) and show that they have better convergence and stability properties. They are also as easy to parallelize as Jacobi and can be optimized to faster versions, well suited for real-time applications. We tested our method on multi-core processors and obtained considerable speedups while maintaining iteration count and accuracy.