Composition driven free convection
 

Hello.

I am performing CFD-calculations on a case with two connected vessels in which natural convection is important, driven by composition not by temperature differences. Using Rayleigh- and Grashof-numbers, the expected regime is fully turbulent. However, this is only a local phenomena. Far away, but still relevant, conditions are laminar.

The problem I am facing is that the transient calculations suddenly diverge because epsilon becomes small and the turbulent viscosity large (standard k-epsilon). I tried the fully deferred correction but this doesn't help.

Does anyone have experience with such a case? Then could you share some proper settings in terms of:

- convergence criterium (epsilon, velocities?)

- turbulence model (does a low reynolds turbulence model help?)

- grid quality (I know rectangular is preferable but then I will end up with millions of cells)

- differencing scheme (I don't expect any influence)

- anything else (like: number of iterations of a specific equation?)

Every help or suggestion is appreciated, P.Fonteijn.