I have studied the paper, N.V. Nikitin, "An approach to wall modeling in large-eddy simulations," Physics of Fluids, Vol12, No. 7.

There are someting which I cannot understand. At 1631 pp. in the middle of the second paragraph, if DELTA/h is raised to 0.1 to 0.2. the resolved turbulence usually dies leading to a steady 1-d solution. This solution is inaccurate, because bar d < d , so that the model dose not reduce to the RANS model,... Full RANS is recovered only when DELTA >> 1.5

Q1) Is above sentence correct ???

Q2) Why bar d < d ??

I can't understand these sentences well. How about you ???

Hi KS,

Let's first ensure that the definition are correct: - d is the distance to the wall. - \bar{d} = bar d = min(d;C_{DES} Delta): the modified distance to the wall (eg distance to the wall upper-clipped by the cell size).

The DES background turbulence model is a Spalart-Allmaras. The RANS turbulence model only relies on d; while the DES turbulence model relies on \bar{d}.

The answer to Q1 is "yes". Delta/h will denotes the mesh resolution. Nikitin indicates that if the mesh resolution is too coarse (DELTA/h raised from 0.1 to 0.2), the resolved turbulence become small until it completely disappear. Since the turbulence is no longer resolved, one could expect that the model operate in a RANS mode. However, since \bar{d}<d (eg the cell size is smaller than the distance to the wall), the results from the DES turbulence model are not identical to those from the RANS turbulence model that relies on d only.

To simplify, I summarize the characteristic of RANS and DES models:

- RANS: (1) only relies on the distance to the wall d; (2) the turbulence is always modeled.

- DES: (1) relies on \bar{d} (mix between distance to the wall and grid size); (2) the turbulence is resolved (C_{DES} Delta < d + equilibrium conditions) or modeled (C_{DES} Delta > d).

The idea behind (and that extend) what Nikitin show is: DES will not enable you to model the turbulence in one cell and to resolve it in the cell next to it. There is a "grey area" between modeled and resolved turbulence when using DES.

If you are interested there is a "Young-Person's Guide to Detached-Eddy Simulation Grids, NASA/CR-2001-211032, Jul" written by Spalart. You should be able to download it from:

http://webs.uolsinectis.com.ar/acoirmejor/NASA-LTRS%20Search.htm

Hoppe this may help. Julien

Thank you very much for your answers. O.K. I can understand it.

I mean "1-d solution" "1-dimensional solution". Your meaning of it is the solution of shear stress in channel ?? (y-1)

I can't download the file you refered. I am not in USA,so I can't. ^^

In external flow case, the phenonmenon like the results of Nikitin does not come out ?? Maybe above answer can be acquired in the paper of Spalart, "Comments on the feasibility of LES for wings, and on a hybrid RANS/LES approach" But I can't find it. ><

Thank you. Best regards.

You can get the report from

http://techreports.larc.nasa.gov/ltrs/2001-cit.html

Thank you for your information. But I can't access to this site.

I think I am not in USA, so I can't access to NASA download site. Otherwise, do I have some problems in my PC ? From last year, I can't access to this.

Can others access to this site ??

Hi,

I am not in USA but can still access the file. Do you want a copy sent by email (it is about 8Meg)?

I don't catch completely what is puzzling you. Nikitin say that if one made a coarse grid, DES of a channel flow can get from the state "turbulence is resolved" (the average solution is 1-d but the instantaneous is not) to "turbulence is modeled" (the instantaneous solution is a 1-D steady solution that is wrong since \bar{d}<d).

In Nikitin's article, they focus on channel flow. In such configuration, turbulence is generated through shear stress at the wall. Therefore, turbulent fluctuations are higher at the wall and diffuse to the center of the channel.When one want to do LES of channel, the mesh should enable to resolve these wall fluctuation.

On the contrario, when doing DES of channel, the near wall fluctuations are not resolved since they are modeled as RANS. Therefore, the fluctuations in the center of the channel can not be correctly generated.

For external flows, DES (in RANS mode) will be accurate to predict detachment points or wall shear without requiring as much points as a LES. Once the flow is detached, turbulent fluctuations will be mainly generated throught mixing shear. DES (in LES mode) will be able to capture accurately these type of fluctuations.

Hoppe this may help. Julien

Let me see if I can throw my two cents in.

For "natural" applications of DES, the streamwise/spanwise grid spacing is larger than the boundary layer thickness, so the boundary layer is handled entirely by RANS (d<Delta). Nikitin et al. was an "unnatural" application of DES. You can think of the channel as a boundary layer with a thickness of h/2. So his grids had a streamwise/spanwise spacing of 10 to 20 points per boundary layer thickness. If you do this, the RANS/LES interface is close to the wall, and the center of the channel is in LES mode. I.e. turbulent stresses are resolved. Near the wall, in the RANS region, they are modeled.

If d > Delta, you are in the LES region. But to get true LES in this region, the grid has to be fine enough to support fluctuations. For the channel, he found that if he had 20 points per boundary layer thickness, this was enough resolution to support fluctuations in the center, and get a decent log layer there. If the resolution was worse than this, then there was not enough resolution to support fluctuations in the LES region, and the solution was degraded. To get back into full RANS mode, you would have to have the streamwise/spanwise spacing about equal to the boundary layer thickness (channel half height). So for intermediate grids (not fine enough for LES, but too coarse to have the channel in RANS mode), you can get a bad answer.

Hope this helps a little. Bottom line is that for external aero problems, the goal is to not do what Nikitin et al did. I.e. don't make the grid in the streamwise/spanwise direction much finer than the Bl thickness. Nikitin was an intellectual excercise that says, "what if we try to use DES in LES mode in the boundary layer". If we do, there are errors (because this is not what DES was designed to do). But for a model not designed for wall-layer modeled LES, it was reasonable (skin friction errors around 15%, a resolved log layer, but a mismatch between resolved and modeled layers).

I also have Spalart's "Strategies" paper - this is better than the original 1997 paper proposing DES. Let me know if you want it. Also, we have many papers we have done using DES with our code - let me know if any of these are of interest.

Jim Forsythe www.cobaltcfd.com

Thank you for your kindness. Your explanations are much helpful to me. I want to research more about this problem.

Does Spalart's "Strategies" paper mean "Strategies for turbulence modeling and simulation" ? If it's right, I have it. Thanks.

I serched the web site about Cobalt code which you mentioned. It is very exciting and impressive to me. But Is it commercial code? or necessary cost? ^^

After all, thank you very much.

Bye~

Best regrads.

Yes, that is "Strategies".

Yes, Cobalt is a commercial code. We've been working with DES since '98, shortly after the method was proposed. We work closely with Kyle Squires, who has done a lot of the pioneering work in the US with DES. If you want more info about his work on the channel, you may want to contact him directly (www.eas.asu.edu/~squires - has some of his papers on there as well).

Jim