GitLab

Note that there are much more bugs of this type in diagonalization/eigen_diag.F90.

Removed preprocessor switches CPP_SOC, CPP_INVERSION, CPP_APW, CPP_CORE.
Only a single fleur executable is built.

I don't know whether there was a problem or not.
But I think that this commit makes the routine a
little more stable.

For very large unit cells this default could become 0 and
this lead to an error. Now the lowest value is 1.

...it would not be compatible with the inp file format.

spg_gen became a limiting code part for large unit cells as
parts of it scaled with the number_of_atoms^4. I rewrote
certain parts of it in a more efficient way. Better scaling
now.

I hope I didn't damage anything.

After making CPP_CORE the default in Fleur runs some
tests didn't work any more. I adapted the reference values
from theses tests. Everything works again.

The parameter maxCubeAtoms is used in chkmt to determine the
size of some arrays. It was chosen too small for some limiting
cases.

Also in this commit: Added an error message for too small distances
between atoms.

This is a bug fix in response to the GitLab issue #49.

XML-Codepath has to be adjusted

Also in this commit: Correct translations in symmetries in XML code
path to multiples of 1/3 iff they are near to it.

The wrong data types lead to some arrays being overwritten and as
a consequence the error map2 was caused in vacfun if the xml code
path and parallelization was used.

It is possible to fix this at different places in the code. I
hope the here chosen code part is adequate for this fix.

The whole Bravais matrix was read in, even for film calculations. But
then it overwrites dtilda, stored in the (3,3) element. This lead
to errors in the z position of atoms in films.

...also renamed everything new in out.xml.
@Daniel: Sorry, could not resist. ;)

A new file was missing in the associated cmake dependency list.

The fix so far seems to work for bulk systems.

git version to the executable. This is done by cmake generating a file
init/compileinfo.h with the required data. This file is included in init/compile_descr.F90 and
ignored by git.

...in collaboration with Uliana.

For some numbers more than 10 digits were written out to inp.xml.
But it turned out that starting with the 11th digit numerical
noise becomes relevant so that the numbers became strange. It is
surprising that we already have numerical noise in the 11th digit.
I expected this to happen at digit 15 or so.

Time reversal symmetry is exploited to increase the number of symmetry
operations used for the generation of the Brillouin zone iff no inversion
symmetry is present. So far the inverse of each symmetry operation was
added, ignoring translation vectors. However, this could lead to further
identity symmetry operations which caused problems as this matrix has
to be excluded for several code parts.

The fix was developed in collaboration with D. Wortmann.