I just altered my equations slightly to include rho in the flux and it made a very small difference.

What we really need is good test cases for multiphase turbulence! At the moment you could pretty much put a random number generator in multiphase codes and get just as good results.

I think Juho is right, because alpha is there if you start with the localized equation and the conditionally average them. However, the current k-epsilon model in bubbleFoam/twoPhaseEulerFoam is quite rough.

A good starting point could be the work done in Prof. Simonin's group. They have both a k-eps-kp and a full RSM model (they considered a the full set of second-order moments to account for anisotropy). Their applications were mainly related to gas-particle flows, but there are equivalent models for gas-liquid and tri-phase applications.

The rest is still advanced research. Prof Subramaniam group has been actively working on that too, especially on the derivation of closures from DNS.

Let me know if you need references.

I think the first step should be to match the single phase results at the dilute limit.

I noticed that in the twoPhaseEulerFoam the near wall G is calculated from turbulent viscosity and set to zero if the y* is less than 11.6:



While in the single phase epsilonWallFunction the G is calculated at all y* values and is calculated from the effective viscosity (nu+nut).

In my test case, the twoPhaseEulerFoam results improve with the effective viscosity based G-calculation and the Spalding turbulent viscosity wall function á la nutSpalartAllmarasWallFunction. The near wall behaviour matches quite well with single phase results and correlations. However, there is still a 1/3 of a difference in the centerline turbulent viscosity.

Interesting, there are some changes such as the one you mention between the single phase turbulence wall calculations in Of-1.6.x and Of-1.7.x.

So there's still a difference between turbulent viscosity calculated by single phase turbulence models and by the one in twoPhaseEulerFoam when the flow is dilute? Keep us updated if you find anything further.

I found the reason for the differing results: the lookup or add keyword for alphak was incorrect in the createRASTurbulence.H file.

Bug report:
www.openfoam.com/mantisbt/view.php?id=67
Fix commit:
github.com/OpenCFD®/OpenFOAM®-1.7.x/commit...d965e0473e059c70f70f

Now the results of my modified twoPhaseEulerFoam in a dilute case match those obtained with buoyantPimpleFoam.