GitLab

Don't ask me why. This is magic :D

for inversion symmetry throw away the imaginary part of the interstitial output density in mix/kerker

For easier debugging of the preconditioner

(Due to unknown reasons we now only need 19 iterations.)

...there was a case where it was still too large.

This build is only done on the nightly builds. The reason for this is that the
pytest plugin for running the workflow regression tests has no support for MPI
executables yet and even if they are called without MPI they can lead to strange test
failures. We don't use the non-hdf5 build since the hdf5 build leads to relevant
behaviour changes in aiida-fleur and are assumed to be the main way of using aiida-fleur

Some additional information can be written out into a test.log file in the work
directory. For this use the test_logger fixture

The arrays il and kl used in the kerker preconditioner were shared but have to be private
This lead to race conditions for the cases with atomtypes with different lmax or muffin tin radius

Since this was undiscovered for quite some time the loop is switched to default none to avoid further
bugs

Make LDA+U double counting consistent for noco calculations

See merge request fleur/fleur!187

to make sure it stays consistent

This reverts commit 15666ce7.

Also adjusted tests that changed results because of this

.. to see if it is the cause for the failing tests

This is a bit suboptimal since the formula for AMF is easiest when combining
U energy and double counting term. For FLL however it is simpler to calculate it separately

The electron interaction part of the DFT+U potential is moved to a different file
to be able to reuse the calculation in the AMF double counting, which boils down to
a DFT+U potential with a modified density matrix

If the spin quantization axis is not aligned with the actual magnetization
The double counting expression gets more complicated

In this commit only the FLL double counting is tackled

(It looks like the output string was too long for the temporary internal storage.)

Correct type of posforce output in relax.xml to produce the correct the representation in masci-tools

This increases stability when adding LOs. Before there was too often an error
that there were not enough linearly independent basis functions.