[Starcatcher]

Hi Guys,


I work with OpenFoam only for some months. I need to run parallel DNS with chemical reactions with OpenFoam v6.
Therefore I use the rhoReactingFoam solver with no turbulence model ("simulationType laminar" in turbulenceProperties)


In fvSchemes I have to define interpolation/discretisation schemes (I guess).
I think for dns they should be of as high order as possible.
Unfortuantely I am not experiences with varying those.


The criteria for the choice of those schemes are:1. The scheme schould be avaiable in OpenFoam v6, or v7. (I currently don't have time to change the source code)
2. The scheme choice should be numerically stable. (from experience?)
3. The scheme should produce relyable results (from experience?)




Does anybody have recommendations for the choice of the right schemes or even can give me a fvScheme-file with appropriate settings for DNS?


Thanks in advance
Andrew

[HPE]

I definitely wouldn't use OF for DNS. I highly recommend you to find another software for your DNS studies.

[Starcatcher]

Thanks HPE for the quick reply!


may I ask why?


I have ansys fluent as alternative, but there i can not change much

[HPE]

Because OF's spatial and temporal error order in terms of Taylor's series is upper-bounded to mere second order. Also, OF does provide a single DNS solver: dnsFOAM which is only for simple isotropic turbulence cases. The reasons can be increased but I think these two should be enough.

[Starcatcher]

interesting.
about the temporal error order: Do you think the source cide can be ehanced properly?

About the solver: i use rhoReactingFoam with laminar setting for turbulence. That's basically dns asuming a suffisient resolution. right?

I am willing to put some work into the solver because it is still better then using other opensource which is even less maintained...i guess

[HPE]

>> About the solver: i use rhoReactingFoam with laminar setting for turbulence. That's basically dns asuming a suffisient resolution. right?

Absolutely not. 'laminar' entry turns off the Reynolds stress modelling contribution for a chosen mathematical approach, RANS, DES or LES bu assuming the modelled flow will be laminar. If the flow is expected to be turbulent, then it means that you remove the turbulence model from the closure problem. You will get unphysical mumbo-jumbo eventually. I strongly recommend you to study what DNS is for a considerable amount of time before jumping into its practice.

>> about the temporal error order: Do you think the source cide can be ehanced properly?

If you have months and if you are proficient with low level OF coding, yes. Otherwise, no. Honestly, I feel that you are not fully aware what you are about to deal with. DNS is in my opinion hard to construct, validate, and very expensive to run and analyse, and considerably limited to simple canonical flows.

There should be open-source or free at least DNS software out there in a form of research code suites.

And DNS with FLUENT, or riding a donkey with Mach-10 speed. I think the latter is more likely to happen.

[Starcatcher]

Thank you for the detailed answer. You seem to know that stuff.
As far as I thought to understand, there is no closure problem when you solve the NS-Equations directly which the rhoReactionFoam (with laminar setting) is doing. Am I wrong?

[HPE]

NS is not directly solved there, but the filtered/averaged NS. Yes, Im afraid that you were wrong. But thats OK, we all learn new things.