CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM (http://www.cfd-online.com/Forums/openfoam/)
-   -   Cartesian grid specific solverIterationpresure solver (http://www.cfd-online.com/Forums/openfoam/60980-cartesian-grid-specific-solveriterationpresure-solver.html)

anne May 26, 2006 06:04

Hello, I would like to know
 
Hello,

I would like to know if there is any flag option
in OpenFoam that could specify that the
solver has to be applied to a cartesian grid.

I asked this because (I think ?) it could
avoid some CPU time when running on a cartesian grid (it would avoid all the surfave normal vectors computations and so on).

Another question relates to the number
of iterations required to solve the
pressure. In a developped channel flow, when using
the IGCCg solver it requires about 30 iterations for each pressure resolution and when using AMG it requires about 10-15. Is there aby way to
optimize the iteration number ?

Finally, I would like to know if OpenFoam makes use of Rhie-Chow interpolation.


Thank You for your help,

Anne

hjasak May 26, 2006 06:21

Cartesian Meshes You are ri
 
Cartesian Meshes

You are right, on Cartesian meshes you could use special tricks to make the discretisation run faster by using the properties of the mesh. In order to do that, some methods (e.g. gradient) would have to be written again specifically for Cartesian meshes. This also leads to the issues with signalling (you would need 2 mesh classes) etc. and Cartesian meshes are not too useful. On balance, I would say this is not worth it - unless you wish to do things that are only possible on a Cartesian mesh (e.g. sixth order gradients). For this reason, OpenFOAM treats all meshes as unstructured polyhedral.

The code has a natural slot for this kind of development: you could copy the finiteVolume library and re-implement the discretisation for the new mesh type. All other stuff, from pre-processing to matrices and fields can be re-used as appropriate. An example of something simular would be the finiteArea library, solving FVM on a curved moving 2-D surface mesh.

Solver

Sounds like for your case ICCG works much better - just pick that. This is why we've got several solvers in the first place :-)

Rhie-Chow Interpolation

Yes, of course, but not "by the book". Have a look at my Thesis or try to catch Rasmus Hemph - he has recently written a note about this.

Enjoy,

Hrv

anne May 26, 2006 09:59

Dear Hrv, Thank you for you
 
Dear Hrv,

Thank you for your fast answer.

In our project we would like to perform LES for the study of contaminant dispersion in a urban context.
For this we are thinking in using OpenFoam because
it is an open source code.
I have just started with DNS and LES of fully channel flow for a first go (I don't know anything about c++). The second step I think about is to perform a surface mounted cube in a channel. As it is still
cartesian I was wondering about the possibility of
switch off the unstructuired-related part to gain
in CPU. But in viw of my poor knowledge
in c++ I am not sure I will re-write
the discrete operators.

However, could you let me know where I can access your thesis ? It would be very helpfull for me to have it.

I have two more quetions ....

The first one is on turbulent inlet conditions issued from a precursor channel flow computation
that I had posted. I still not have any answer,
I know that you must be very busy but
some indications on how to do it it
would be great.

The second is: is there any good tutorial to
start with tracking particules and scalar passive
transport?

Thanks again,

Anne


All times are GMT -4. The time now is 00:59.