What values of under-relaxation factors to use?
 

Hi,

I wonder whether there is a general guideline on how to set the under-relaxation factors and how to decrease it step-by-step? I saw in some tutorials that in simpleFoam, it is advisable to set the under-relaxation factors for U and p to be 0.7 and 0.3 respectively. However, when that fails, how would you decrease them step-by-step? Is it advisable to reduce them together? Is there a way to check which one is causing the solver to diverge so that we can just reduce the under-relaxation factor for that particular quantity alone. Also, is there a lower bound where you should not decrease the under-relaxation factors beyond that?

Thanks!

The relaxation factors of 0.7 and 0.3 are typical values for simple algorithms. simplec algorithms (consistent true; inside fvOptions) can use 0.9 or 0.95. Temperature and pressure for compressible or heat transfer solvers are the only ones i'd suggest tweaking.

And yes you can slightly accelerate your solution by increasing those from 0.3 to 0.5 for example. And you can add stability by lowering them. BUT:



If your solver crashes changing the relaxation won't help you. Atleast 95% of the time. It is much more likely that your boundary conditions or mesh are the culprit. These standard values are a result of the algorithm. These 0.3 and 0.7 values are stable for nearly all problems. Compressible solvers are a bit more tricky though.


You can lower them as you'd like though. Going down to 0.2 and 0.5 is what I'd do, but again this is not usually what you do. The discritization of the convective term div(phi,U), mesh quality and boundary conditions are far far more important to get right.

A side note: at the time of writing, programmable under-relaxation factors with openfoam.com is also possible.

Under relaxation means that you apply only a part of the calculated changes. It is used for damping (non physical) oscillations of the calculated values.

You may use over relaxation too. If the precess is very stable you can reach steady state faster.