parallel overlap in fempy
We tried to write a simple overlapping Schwarz method in python using fempy.
I'm not 100% sure, but I couldn't find any way to code to correctly assemble the local matrices. We used a 1st order Lagrange space. What seems to be missing is:
- boundary conditions on the processor boundary (the "front" nodes are constraint 0)
- matrix assembly in the overlap (the default partition type in dune-fem C++ is
interiorborder
and I couldn't find any way to tell dune-fem to assemble also in the overlap)
What we did now to work around the problem is:
- patch the C++ code (threaditerator.hh) to use the
all
as default partition type - post-process the
scipy
matrix to explicitly eliminate rows corresponding to a processor boundary
I believe that this is really a bug in the python bindings, or did I miss something?