I love to do DNS for my backward-facing step.

Even though CFX does not include DNS, it seems to me a way to conduct DNS is to impose Cs = 0.0 for Smagorinsky model.

I wonder wheter any one tried this. Also if you can see any possible problems in doing so, please let me know.

Thank you in advance.

Well, I might be wrong here, but to do DNS wouldn't you just need to use a very fine grid and timestep to capture the smallest scales? That would mean no turbulence model (i.e., laminar).

It depends on the Reynolds number too.

If you do laminar flow calculation, you can't process turbulent properties such as Reynolds stress in CFX.

Hi,

If you want to try DNS in CFX you will want to use the laminar turbulence model. The RANS and LES turbulence include simplifications to the stress vector which are normally not important but for DNS you may well need them to get the dissipation right. I hope the flow is a very low Reynolds number. Good luck, you will need it.

Glenn Horrocks

Thank you.

And an element high number!

Have you worked out how fine your grid would have to be? You will have to resolve all the scales in your flow. Therefore you need to calculate the length of kolmogorov scales in your flow. How many processors do you have to run your problem?

Regards

I used LES in two dimension flow, with sub-grid scale structure function, using the FORTRAN language. I had the 7.2 point of recirculation.

Hi,

Did you say you ran a 2D LES simulation? You can't run LES 2D, it is 3D only.

Regards, Glenn Horrocks

What I wanted to try was not real DNS but coarse DNS.

I think I have a proper mesh for three-dimensional LES with yplus =1 approximately. I just wanted to turn off subgrid modelling without refining grid further. I hope this may result in between a proper DNS and and a LES with standard smagorinsky model.

Regards, Sungho