Stokes flow
Hi: I have a conventional face-based cell centered unstructured segregated rhie-chow code. When applied to a very low Reynolds number flow (Re=10^-3) convergence is very slow. The problem is almost linear of course so I assume that a coupled solver would do the trick but a friend told me that even the coupled solver in a common commercial code he used for such a problem converged very poorly. It was suggested that I compute all explicit viscous terms (cross-diffusion and non-face-parallel) between momentum equations, but even turning these off completely so all viscous terms are implicit does not help. Any ideas? Experience? Thanks Rob
|
Re: Stokes flow
You are doing false time-stepping using a high as possible CFL number??? If so, at that reynolds number I would have thought you could push the CFL number to crazily high number.
The trick to achieving a high CFL is using a sensible restart file, or putting in your code that the CFL number should increase with iteration number. Iain ps I am concerned you say "turning these off completely", for me making your viscous terms implicit is switching them on!!! pps you are essentially using the wrong kind of solver, since viscosity and pressure do not have a strong link (in fluid mechanics steady-state there isn't one, though there is when you derive PISO or SIMPLE schemes). I suppose you really should be using stream-function / velocity-potential based solver. |
Re: Stokes flow
Thanks for the thoughtful response. Yes I know I can use another clas of solver that is more suitable, but it most convenient for me to use my own if I can get it to perform reasonably efficiently!! As far as viscous terms go I only retain "normal" diffusion terms and these are fully implicit, all other viscous terms are zeroed.
I will try you infinite CFL idea. Thanks!! Rob |
All times are GMT -4. The time now is 16:17. |