In at least some newer versions of OpenMPI, when using MPI pools, it's possible to see seg faults from the interaction_elph class coming from bcast calls invoking the intraPool MPI communicator.
It appears this issue is related to the use of the non-blocking collective call MPI_Ireduce -- therefore, we swap this with MPI_Reduce to fix the issue.
Major thanks to @anjohan for helping to figure this out.
Additionally, this PR contains:
two minor typo fixes in the docs
one improved error message, for the case the user requests a doping concentration outside the range of Wannierized bands.
A note in the docs about issues that can arise if one does not use exclude_bands in Wannier90 if there are bands below the disentanglement window.
Summary:
In at least some newer versions of OpenMPI, when using MPI pools, it's possible to see seg faults from the interaction_elph class coming from bcast calls invoking the intraPool MPI communicator.
It appears this issue is related to the use of the non-blocking collective call MPI_Ireduce -- therefore, we swap this with MPI_Reduce to fix the issue.
Major thanks to @anjohan for helping to figure this out.
Additionally, this PR contains:
exclude_bandsin Wannier90 if there are bands below the disentanglement window.