hi,all I am fresh in LES programming,I have a unsteady laminar flow pragram using cartesian grid ,colocated variable arrangement,and pressure is solved using SIMPLE method,implicit Euler method for time marching . Now I want to develop a LES code based on this laminar code. (first smagrinsky then dynamic LES code)I have the following questions:

1.Is the following statement right? The main difference between laminar and lES is the latter adds a eddy viscosity,so I need only add a subroutine for calculating the eddy viscosity,and the unsteady terms ,convecting terms and the source terms can remain unchanged .The difference between smagrinsky and dynamic model lies in only the eddy viscosity solver.

2.Can i still use SIMPLE to calculate Pressure,if fractional step method is better?

3.What is the most simple case to test the LES code? Maybe cases without walls is better? Can someone help to tell me the definite domain size and resolution in domain and time? How many time steps is enough?

4. Using fine enough grid and time step ,Can the laminar code be used to do DNS calculation?

5.If cases with wall boudary is include what changes should be made to the laminar code to account for the LES wall boundary?

Any comment is greatly appreciated. Thanks

Sorry ,I forgot to say that the solver is for incompressibel flows

(1) is correct. You only need to calculate an additional subgrid stress (making use of spatial filtering). And yes, the difference between smagrinsky and dynamic model lies in only the eddy viscosity solver.

(2)Yes, you can still use the Simple method. I do not know if fractional step method is better/worse. You might think about which one is faster.

(3) A simple test case would be a 3D turbulence decay simulation with periodic boundary conditions.

(4) Yes the laminar code can be used to do DNS simulations with high enough grid resolution and small time steps.

(5) The wall boundary effect in LES is an ongoing research topic, but most LES calculations include a damping function such that the eddy viscosity goes to zero at the walls.

So many thanks, dear agg, you really helped me a lot. I am still a little confused about the following:

1)the smagrinsky model doesn't need spatial filtering to get the eddy viscosity(Mij=-2*C*Delta*Delta*Abs(S)*Sij) and Sij can be calculated without filtering, since the velocity is filtered by grid filter in the NSE. So only the dynamic model needs explicit filtering. Is that right?

2)I guess the fractional step method is faster,is it right?

3)The 3d decay case ,you mean 3 periodic bc is the most simple one?

4)I read that the dynamic model's coefficient vanish in laminar flow and at a solid boundary,so does it still need a damping function near the wall?

best regards ben

1)the smagrinsky model doesn't need spatial filtering to get the eddy viscosity(Mij=-2*C*Delta*Delta*Abs(S)*Sij) and Sij can be calculated without filtering, since the velocity is filtered by grid filter in the NSE. So only the dynamic model needs explicit filtering. Is that right?

Yes, no explicit filtering for Smagorinsky. One level of (sub-test) filtering for dynamic model. Easiest filter to implement is the top-hat filter: simply average over cells.

2)I guess the fractional step method is faster,is it right?

I don't know!

3)The 3d decay case ,you mean 3 periodic bc is the most simple one?

Yes, I think so.

4)I read that the dynamic model's coefficient vanish in laminar flow and at a solid boundary,so does it still need a damping function near the wall?

No, dynamic models do not use (van Driest type) damping functions near the walls.

Yes, you are correct for parts (1), (3) and (4). I think the fractional step method is faster, but not sure. In any case the bottle neck is the poisson solver and hence try to make it as fast as possible.

dear agg, I want to know 1.how to give the initial field 2.how to implentment the periodic bc at each time step of a 3d decay turbulent flow with 3 periodic directions. I guess it maybe not be that simpley implentmented as in laminar flow (Ui(1,j,k)=Ui(N,j,k),Ui(i,1,k)=Ui(i,N,k),Ui(i,j,1) =Ui(i,j,N))

3.I know that for channel flow the inlet is given a speed profile plus random fluctuations on each CV center in the inlet plane at each time step. Does the 3d decay turbulent flow need such random fluctuations to be given? If need ,how and when to add it ?

best regards

ben

A periodic BC is a numerical BC and does not change with flow (laminar or turbulent).

Here are some references for initial conditions for 3D decaying turbulence

1) Numerical experiments in homogeneous turbulence - Robert S. Rogallo, NASA Technical Memorandum 81315 (1981).

2) Kwak, D., Reynolds, W.C. and Ferziger, J.H. "Three-dimensional time dependent computation of turbulent flow", Report No. TF-5, Stanford University (1975).

3) Erlebacher, G., Hussaini, Kreiss, H.O. and Sarkar, S., "Tha analysis and simulation of compressible turbulence", Theoretical and Computational fluid dynamics, Vol. 2, pp. 73-95 (1990).

thanks dear agg ,it is very helpful

best regards

ben

hi,

I am also trying to solve the same problem, but my doubt is that, to change a code from laminar to LES, is it not enough to change the Viscosity alone, and add it to the lamiar one, like

vis(effective or LES) = vis(laminar)+ vis(turbulent)

and may be under-relax the viscosity.

vis(turbulent)= rho*Cs*Delta*Delta*Abs(S).

Correct me,

krishna.