[alberto]

Originally Posted by lakeat

Hey, what! I feel I left behind in life. Any Good papers on that? Could you share some links? And why you think it's not easy to implement them?

Well, "a good paper on that" is hard to find given that there are many. You can start looking here (warning, marketing of stuff done in the university where I work ) http://zjw.public.iastate.edu/papers.htm

It is not easy to implement them in OF because it would mean a redesign of the data storage (DG and IP methods require more storage than standard FV), operators and such.

Best,

[alberto]

Originally Posted by lakeat

Again, I heard a rumor that openfoam will adopt cmake instead of wmake, is it true? Any other rumors

Do you know who is key writer of so many shell scripts in openfoam? Very good jobs, you know, Brilliant!

This has nothing to do with this thread. Why did you ask this here?

[mmahdinia]

Dear Alberto,

I ran the simulations according to your instructions. And here are the results:

http://mech.sharif.edu/~mahdinia/Q.htm

The span-wise instabilities in the QUICK case are called lobe and Cleft instabilities and should exist. The linear case has no lobe and cleft and the cubic case has unreal disturbances.

There is a small point that I should mention. The maximum speed in our simulation is nearly the speed of the head of the current (foremost point of the heavy fluid), which is 0.1 m/s. Assuming the viscosity doesn't change for solution, it has a value of 10^-6 m^/s of water. The horizontal length of the domain is 2 m. So in order to get Pe=2, one has to have 100000 cells in the horizontal direction (without refinement near the walls!). If we assume that we have a small number of cells, namely 80 cells in the span-wise and 80 cells in the vertical direction, then this give 640 million meshes. So I think no one would be able have to such mesh. Also there are articles published with 30 million cells for the exact same case:

S. K. Ooi, G. Constantinescu, and L. Weber, “Numerical simulations of lock-exchange compositional gravity current,” J. Fluid Mech. 635, 361 (2009).

They use QUICK for convection in concentration equation but not in the momentum (they use central here).

Currently my mesh is 800*80*80, with the refinement near walls so that y+=1 near all the walls. And my result match well both with experiments and the above work (as you know my mesh is coarser, so I can't see some small vortexes, but the agreement nevertheless is quite good).

What do you think? For my case with Pe>2, should the QUICK method give more acceptable results since convection>>diffusion (Pe>>2) ?

Sincerely,
Maani.

[alberto]

Hi,

the grid requirement is indeed Dx < 2e-5 m in your case, which might be challenging.

In such a case you are forced to use limiters or some high-order upwind scheme. If you are interested, you might investigate what happens with the limitedLinearV scheme for the convection, which is a linear scheme with a limiter to avoid oscillations where Pe > 2. It is not usually recommended for LES, but it might be competitive with QUICK, and it seems to be less diffusive than the linearUpwind (second order upwind like scheme).
With this you should have a good view of the most frequently used schemes in OF, and find out which one gives the best results for your case.

Best,

[ata]

Hi Mani
Your phone is switched off or the number is incorrect.
Best regards

Ata

[mmahdinia]

Dear Alberto,

I know that this might not be correlated to the topic, but I have a little question about dynamic LES itself. Since this model may become unstable, people use averaging. I have two questions:

1) In one of the posts you mentioned that to stabilize the code in the case of total inhomogeneity (all directions), one may clip nuEff to be non-zero. Is averaging necessary in this case? or is the averaging done on a small neighborhood of the particles? If yes how much should be the size of this neighborhood?

I'd really appreciate it, if you could also mention an article about the clipping of nuEff (was it done before?)

2) Some people use temporal averaging. For example if C is the dyn. coeff., then:

C(n+1)=(1-eps)*c(n)+eps*c(n+1,calculated);

where n is the time step an eps is nearly 10^-3.

Is this applicable for unsteady flows like mine?

Sincerely,
Maani

[mmahdinia]

Hi,

Does anyone has a journal paper for "clipping" of effective viscosity such that (nuSGS+nu)>0 in dynamic LES mehod?

Sincerely,
Maani

[ehsan]

Dear all

I need to know the exact formula for fourth order discritization of P gradient in OpenFOAM. Could you be of any help?

Best Regards
Ehsan

[ehsan]

Hello,

If anyone could advise whether there is any similar to discretization scheme described in

Symmetry-preserving discretization of turbulent flow

http://www.sciencedirect.com/science...21999103001268

in OpenFOAM?

Regards