Denis Davydov

Here's a slightly cleaned up version of the code: [pbc.cc.zip](https://github.com/dealii/dealii/files/2607827/pbc.cc.zip) To get better understanding of what's wrong, I added other output. As local constraints/dofs we have: ``` === Process 0...

> It will probably matter in the case of AffineConstraints::distribute_local_to_global Good point. That would indeed give wrong result.

My guess is that when make_periodicity_constraints is called, it gives different master/slave DoFs on two processes. That’s because on proc 0 it has periodic face with hanging nodes, whereas for...

Or we can augment algorithms so that all equality constraints are such that master DoF < slave DoF

> the dofs in question here are not locally active. That is certainly true, but this also means that we should not have issue during assembly of RHS vector or...

> However, after I increase the number of refinement levels in pbc.cc to 4, it does trip the AffineConstraints::is_consistent_in_parallel if run with more than 12 mpi tasks. ok, so the...

I will try digging into `make_periodicity_constraints` to see if there is a quick fix for this problem, at least for equality constraints...

p.s. [documentation](https://www.dealii.org/developer/doxygen/deal.II/namespaceDoFTools.html#a7ff4a580b384bbfdb9edf05d3a5cb36d) says that @tamiko is an author of `make_periodicity_constraints`, so you might know more about this issue 😸

what is `msglvl` ?

@caliarim sorry, I am lost here as I don't really work with Fortran much. I am using C-wrappers to pArpack functions from a C++ application. There is indeed [debug.h](https://github.com/opencollab/arpack-ng/blob/master/PARPACK/SRC/MPI/debug.h) in...