Adaptive Mesh Refinement

[mortbauer]

Hello

Is there some way to use adaptive mesh refinement? Or even just interface to some external software which does refinement lets say every tenth iteriation? I must admit that I never have done adaptive mesh refinement so far but it would be useful to have and I thought since you are optimizing code saturne for petascale and whatever it should be also a need for you, or am I wrong?

Greets Martin

[Yvan Fournier]

Hello,

Some years ago, there have been tests using Code_Saturne with the HOMARD tool (from the SALOME platform), but this was experimental only, and has been broken since version 1.3, when we switched to generating post-processing files with "true" polyhedra at non-conforming mesh joing junctions. As HOMARD does not currently support all cell types supported by Code_Saturne, we have not prioritized working on this so far.

On the HPC side, there has been interesting work done using regular mesh multiplication (http://www.prace-project.eu/IMG/pdf/Par ... urne-2.pdf, or presentation from the 2013 user meeting), and work is ongoing towards an adaptive mesh version, but plugging this in the code will require more work (the work required to update internal face cell-connectivity for turbomachinery helps us estimate what will be necessary).

Regards,

Yvan

[mortbauer]

Thank you Yvan for the detailed answer, I was yesterday also wrapping my head around using homard but found several problems, some easier and some more difficult, i think i will lskip that road.

About the mesh multiplication, i read about it, it sound really interesting, but actually don't know if it is designed to do local refinement as well, and as you said it isn't ready and i couldn't find any source code either. (your links aren't working for me)

So it seems that mesh adaption isn't used that much anyways in CFD since you in EDF aren't working with it either. I guess you create global refinend meshes to have a clue about convergence?

[Yvan Fournier]

Hello,

I edited the links in my previous post to fix them...

Local (adaptive) refinement is planned, but still in progress. Integration in the main version might take additional time, do don't expect anything in the near future, except maybe updated compatibility with HOMARD late 2014 or early 2015.

Regards,

Yvan

[James_T]

Thank you for the links - that's exactly what I was looking for. I am very hopeful for 2015 then

[zeph67]

Hi,

A few years after this discussion was initiated, is adaptive mesh refinement in progress in Code_Saturne ?

Thanks in advance.

[Yvan Fournier]

Hello,

Since version 6.0, mesh refinement is available as a preprocessing stage. There has been some work on coarsening so as to go towards adaptive, but it is not finished, and I need to find time to get back to it.

The mesh refinement works with any sort of cell. If you refine some zones multiple times, it is better for mesh quality to first refine all cells requiring one level of refinement, then all those requiring a second level, ... In case of several layers of refinemement, rebalancing the mesh by forcing intermediate partitionings may be needed (we have some examples available). It is available through user defined functions, not in the GUI yet.

Best regards,

Yvan

[zeph67]

Thanks Yvan for this detailed answered.

Do you think that one day, some dynamic AMR will be available in Code_Saturne ?

[Yvan Fournier]

Hello,

Yes, this is planned. I have a prototype (unfinished) coarsening (based on refinement history), and then will need to add the data mapping, but I still hope to be able to do this by v7.0 next year (with coarsening in v6.2 or 6.3).

Best regards,

Yvan

[zeph67]

OK great !

And will it include local refinement/coarsening ?