Stubborn epsilonm residual
 

Hello,

I am running RANS simulations using twoPhaseEulerFoam. While all other residuals decrease with time, that is not the case for epsilonm, which tends to stay constant. This is typical:

http://i.imgur.com/n3xFYCK.png

I noticed that the fluidisedBed tutorial exhibits the same behavior.
Is this expected then?
Any clues as to why this is happening would be appreciated.

Hello Albert,

did you find a solution for your problem? I see exactly the same behavior. It can also be reproduced by changing the water velocity to something > 0 in the twoPhaseEulerFoam bubbleColumn tutorial.

Thank you
Joachim

Hello - not really. We decided that the simulations had converged based on plots of the quantities we were interested in at locations we were interested in.

Thank you for your answer. Perhaps somebody else has any idea?

I filed a bug at the "official" bug tracker of OpenFAOM: http://www.openfoam.org/mantisbt/view.php?id=1916
It looks like it's also not clear to them what causes this problem.

Greetings to all!

I haven't checked the values and calculations, but here's what I know on this topic:

1. The residual values you're plotting are the normalized residual values. This means that they are divided by a value that changes it to a scale "0 to 1". The problem with this in some cases is that the normalization value isn't accurate or is based on an outdated result.
   • Therefore, in order to diagnose what are the non-normalized residual values you're getting, you will need to hack the code and rebuild it with a line that prints out the non-normalized values.
   • Let me see if I can find a thread that explains this... I don't know if this is still valid in the latest OpenFOAM versions: http://www.cfd-online.com/Forums/ope...tml#post209160 - post #2 explains how to output the normalization values as well, without the need to hack into the code.
2. The residual values for the turbulence models aren't always critical enough to have to be reduced. It depends on the modelling and equation resolution strategy, but sometimes what comes out of the residual calculations is both a reflection of the normalization issue and the actual nature of the turbulence modelling: it's just a model.
   • What I mean is that the pressure (continuity) and velocity (momentum) equations are pretty much proven to be accurate to the reality, at least in laminar flow and DNS. But things aren't as direct when turbulence modelling comes into play, because its exactly just that: it's a model that tries to give a good notion of how the fluid should behave in a particular flow environment, i.e. it's X turbulent in this region and Y turbulent in that other region, as if it's a probability function.

I'm not familiar with the "epsilonm" variable, but its name gives off a vibe of being something similar but very different from the usual epsilon variables, therefore it might be something that has a very smooth value transition, resulting in the odd residual plots.... imagine if the field would only range in [0.999 to 1.001], regardless of how turbulent the flow was.

Best regards,
Bruno

FYI: I was curious and took a look into this. See comment #5684: http://www.openfoam.org/mantisbt/view.php?id=1916#c5684