Matthias Krack

`make ARCH=local VERSION="psmp" TESTOPTS+="--mpiranks 2 --ompthreads 2 --maxtasks 64" test` using 2 MPI ranks per test case looks like the proper choice.

I suggest to create a separate DFT+U section for SIRIUS and leave the CP2K section untouched. The implementations are too different.

> cp2k implementation of dft+U is incomplete and takes for granted that the hubbard correction should be applied to the last (partially) quantum atomic level of each relevant atomic species....

> I realized early on that cp2k implementation of dft+u is fairly incomplete (it is the simplified version) that's why I had to add extra parameters to the dft+U section....

> the cp2k implementation of DFT+U is a simplified version of the entire theory. > I could not find where n and l information are given in the input file...

If we want to keep one DFT+U section, we have two options: - we generalise to U and J and drop U_MINUS_J as keyword (which breaks compatibility with old inputs)...

> l should be added as well. By default if n and l are not present and SIRIUS is not used then the parameters are ignored completely. It means in...

> Yes it is. l is mandatory which is also needed by sirius. So we could simply ignore N when SIRIUS is not used explicitly. @mtaillefumier Add N for SIRIUS...

> SIRIUS supports multi-orbitals on a single atomic type (and multi-orbital between atom pairs as well). We put this in place when we developed LDA+U+V. Supporting this in cp2k is...

@mtaillefumier Are you planning to commit this PR soon? Conflicts start to accumulate and I would like to continue revising the DFT+U implementation.