Dear All,

For a pipe flow with moderate Reynolds number,say 8,000, if we run a 3-D unsteady simulation with laminar solver, i.e., no turbulence model is involved, on a mesh which is fine enough or even finer for a large-eddy simulation (LES), what will happen? Will the results be as useful as that of running a LES on the same mesh? Obviously, the LES calculates a bit more than the laminar solver.

It seems that the DNS still use the laminar N-S equations without turblence model,but running on a very fine mesh and with very small time step.

Another question is: Can anybody give an estimation of the ratio of the subgrid scale velocity over the total instaneous velocity in LES?

Thanks for any comments.

You are touching a delicate problem for LES. A (the) major difference between DNS and LES is the subgrid modelling. Obviously it has an effect on the result, if it is turned on or not, however it is also a matter of which mesh you are using. On a very fine mesh, LES whithout subgrid modelling should be close to a DNS computation, hence it would be erronous to include the model. Any serious paper dealing with development of subgrid modelling should include the effect of the model, try e.g. Krajnovic.

And equally problematic area is the treatment of walls using LES. Properly used LES (not going into DNS) would result in fairly large cells close to walls, and hence the use of wallfunctions to bridge the nearwall region. The idea of using a model, which is based on equalibrium flate plate boundary layer to model, e.g. the wake behind a car is to say the least questionable.

Only some thoughts, regards Jonas

Thanks, Jonas.

Yes, it is a matter of mesh. DNS on a coarse mesh on which LES works can not resolve the field enough.

Jonas, do you have idea of my second question? What will be that ratio roughly? u'/u (u' belongs to SGS field.)

Kind regards,

Ray

Hi there,

A quick comment regarding the mesh density in case of an LES and a DNS. For LES, usually the first poit of the wall is set at y+=1. For a DNS, y+=0.1

There is an order of magnitude difference between the two. In addition, one needs to keep in mind that for DNS, all 3 directions have to respect this criterion of y+=0.1.

Now as far as the ratio u'/u, if you are looking at an estimate of the turbulent intensity, well it depends on the flow properties, but it is usually between 10-20%

I hope this helps. Sincerely,

Frederic Felten.

Thanks for the response.

Is this ratio so big? I am not estimating the turbulence intensity but the ratio of u'/u, note u' is only for the SGS part not the whole fluctuating velocity. If the velocity is decomposed in space into large scale velocity and subgrid scale (SGS) velocity, can the latter one be ignored?

Does DNS need an initial random perturbation ?