Dynamic domain decomposition during run time
 

Hey!

I was wondering if there is an existing implementation for a case where I have a parallel run and I would like to keep as many cpus within a part of the domain e.g. where some scalar concentration is high enough. In other words I would like to decompose my domain again and again in runtime in order to keep as many sub-domains (cpus) in regions where intense computations are managed. Naturally this should be done only every N:th time step or so.

This kind of situations may appear in situations where one solves e.g. cell-wise independent high-load problems along your CFD solution. For an example chemistry and material property computations or some other multiphysics related problems may need this kind of approach.

If implementation does not exist, would a dynamic mesh libraries and decomposePar utility be the right way to begin with the implementation?