DNS and Fluent
 

On the topic of whether Fluent can use DNS,

someone gave some comments some years ago as follows,

---------------------------------

absolutely, just make sure your grid is fine enough to resolve the kolmogorov scales, and turn on the unsteady laminar equations, and there you go. Then, choose a time-step representative of the turbulent time-scales in your domain, something on the order of 10-10 seconds or smaller, and let it rip.

Now, you may need 10000 CPU's running in parallel, about 10,000,000 GB's of memory, and you'll need to run billions of time-steps just to simulate a few seconds of real time, but in principle, it can be done. (Some sarcasm of course embedded in that).

---------------------------------

So my question is : The reason that DNS is not practical in Fluent, is it also because Fluent's discretization schemes is not fine enough?

Regards,

Daniel

Re: DNS and Fluent
 

Daniel,

Basically, if you want to do DNS on a real problem of significance then you need a very fine grid. The resolution of this grid must match the smallest turbulent eddies which are indeed the Kolmogorov scales. You cannot resolve an eddy size if the eddy is smaller than the grid!

Now if you want to model high Reynolds number flow as in reality, you need this ultra fine grid to model the full range of turbulent scales which obviously includes the smallest ones.

To answer your question, there just wouldn't be enough memory on the best machine in the world to create all the nodes. Even if you could create meshes with the hundreds of billions of cells required for a real problem, you wouln't have a hope of solving the flow!

A compromise is to use LES which uses a grid which can resolve the larger eddies in a flow and a RANS model to 'model' but not properly solve the small scale eddies. This is currently state of the art in a tried and tested way.

I recommend you read an introductory CFD book such as the one by Versteeg or Lui.

Carlos.