Parallel smoothing pressure correction solver for biomedical flow problems: convergence criteria, preconditioning, scalability

摘要

A fast, parallel flow solver based on a pressure-correction approach is analysed. It is especially suitable for biomedical flow problems due to the use of unstructured meshes and the ability of tackling large problems. The original smoothing pressure correction method was enhanced by parallelisation, efficient linear solvers, and the introduction of a generic linear stopping criterion. 'Lid-driven cavity' at Re = 400 served as steady-state test case and was solved in less than 2 min on a mesh with 84 k cells using a standard PC. Solving 'flow past a cylinder' demonstrated accuracy in transient mode, and the simulation of blood flow in the human aortic bifurcation applicability to biomedical problems. Scalability was investigated on a Cray XT5. High parallel efficiency could be achieved when solving the momentum equations with scalable low-cost preconditioning. Computationally more expensive multigrid preconditioning proved to be advantageous when solving the pressure correction equation, but restricted scalability to a range of ca. 1-30 computing cores.