[mahdialloush]

Hi Everyone,

In Fluent, the full approximation scheme is only used for density-based solvers, not for pressure-based ones. Does anyone have any idea what could be the reason?

Many thanks!

[sbaffini]

I am no expert on fas, nor I know if the Fluent motivation is a general one or related to some specific decision (for example, that they only implement proven methods from tested published works), but fas is multigrid for non linear problems, and the Fluent version involves discretizing on multiple grid levels. This has a cost that is probably justified only for an explicit coupled solver, which means explicit density based in Fluent.

There might be other (more meaningful) reasons

[LuckyTran]

The A in AMG is algebraic. FAS is a geometric multigrid technique. In general, geometric multigrids handle non-linearities (i.e. coupling) better than AMG. AMG relies on iterations/sweeps to transfer non-linearities.


At the same time, the density-based solver is a coupled solver compared to the segregated pressure-based solver. You're basically asking why a multigrid method that handles coupling better is preferred for a coupled solver. Oh em gee. What makes a coupled solver preferred (not necessarily better or not) over a segregated solver in the first place?

[sbaffini]

As a side note, there is one istance where Fluent practically (but not formally) uses the fas approach also for the pressure based solvers. It is for the fmg initialization. Yet, it does so internally by temporarily switching to the explicit density based one

[mahdialloush]

Ah cool, I did not know that they switch internally to density-based upon using FMG initialisation. In fact, my question is exactly here, why do they, and possibly many other solvers, avoid the FAS with pressure-based solver? Is it numerically non-feasible or unstable for instance in a pressure-based framework?

[arjun]

I had long back discussion with a developer (Fluent) who has worked on this part of the code and he mentioned that in unstructured grid major issue is that at coarser levels they pretty much always end up with negative volumes.

So my guess is that it has something to do with quality of control volumes and the solution there.

[sbaffini]

Originally Posted by arjun

I had long back discussion with a developer (Fluent) who has worked on this part of the code and he mentioned that in unstructured grid major issue is that at coarser levels they pretty much always end up with negative volumes.

So my guess is that it has something to do with quality of control volumes and the solution there.

I had the suspect that it is, indeed, not all rainbows and unicorns, but does that concern the specific solver in use?

[arjun]

Originally Posted by sbaffini

I had the suspect that it is, indeed, not all rainbows and unicorns, but does that concern the specific solver in use?

I suspect it has to do with Ap of momentum equation that is needed for Rhie and Chow term for pressure based method. Once it goes to negative or 0 the method won't work.

In case of density based solver this is not the problem.