[Bram]

Hallo,

I'm looking for information on domain decomposition methods.
Does anyone know which domain decomposition method OpenFoam use?

Thanks a lot!
Bram

[fra76]

By heart, Metis and ParMetis, Scotch, Hierarchical, plus simple and manual decomposition.
To get a list, simply type a wrong method in system/decomposeParDict and try to decompose the case.

Hope this helps,
Francesco

[Bram]

Hey Francesco,

Thanks for the information. But I am more looking for the method OpenFoam use to exchange information between the various domains, not the way how the domains are split. Examples of such methods are:
-The Dirichlet-Neumann algorithm
-The Neumann-Neumann algorithme
-A FETI-methode
-The ADN an ARN methods
-The Robin-Robin algorithm

Thanks!
Bram

[fra76]

Ok, I did not understand your question.
And I cannot give you an answer, sorry...

Bye,
Francesco

[chegdan]

I know this is old, but...

@Bram
Were you able to figure out what sort of method openfoam uses to transfer information at the boundary (I know it communicates via openMP or other message passing code)? I do know the domain decomposition method used is non-overlapping and that the information across processor boundaries needs to be updated at each inner iteration of the linear solver...however, how is the information at the interfaces related to eachother (Neumann-Neumann,Dirichlet-Dirichlet, etc.)?

Dan

[elmo555]

Originally Posted by chegdan

I know this is old, but...

(I know it communicates via openMP or other message passing code)

To clarify, communication between processors is done with MPI, not OpenMP, and the MPI implementation (OpenMPI, Intel MPI, MPICH, ...) can be chosen by the user or operating system.