[Gmsh] partitioning in gmsh -- repost

Christophe Geuzaine cgeuzaine at ulg.ac.be
Sun Oct 17 11:58:24 CEST 2010

On 15/10/10 23:30, Umut Tabak wrote:
> Dear all,
> I asked the question given in the link below but no replies have
> appeared so far. So it is time to fire again ;-)

Hello Umut!

> http://www.geuz.org/pipermail/gmsh/2010/005773.html

I've had a quick look at your files. To answer your two questions:

* currently the partitioner creates an actual topology for the 
partitioned model, hence the new line elements (with the negative 
numbers) in the output file. The option to change this is currently 
hardcoded (see createPartitionBoundaries in Mesh/meshPartitionOptions.h)

* currently Gmsh always renumbers the elements when writing a mesh file. 
We should maybe provide an option not to touch the numering...

As yiou can see the partitioning code in Gmsh is just not flexible 
enough at the moment. It might thus be easier/faster for you to just 
write your own little wrapper around Metis to partition your meshes 
according to your criteria?

> I was processing Ansys generated mesh structures in gmsh. There are two
> questions from my side, the questions are in the above link. However, my
> question on partitioning is somewhat more urgent now so here that comes,
> I am writing the input mesh file in a mixed order(there is also an
> example file attached to the above link) since Ansys optimizes the mesh
> structure for the the bandwidth of the resulting matrices. If I write
> the elements in different order than a consecutive order, say [1: nelm
> ], I guess gmsh wrappers for the partitioners are not that smart yet to
> recognize this difference or I have a conceptual misunderstanding.
> Because giving the input file to gmsh with this mixed order screws up
> everything after the partitioning operation, however sorting the
> elements to be in a consecutive order solves the problem. I wanted to
> confirm that this is the right way for me to follow, I found this by
> some reasoning and trials.
> Any comments are appreciated, best wishes for all.
> Umut
> _______________________________________________
> gmsh mailing list
> gmsh at geuz.org
> http://www.geuz.org/mailman/listinfo/gmsh

Prof. Christophe Geuzaine
University of Liege, Electrical Engineering and Computer Science