GitLab

The preconditioner already improves the convergence of Cu and Fe files, but the starting distance is higher than for the older fleur version and tests run up to test 7 only (does not fail, but hangs).

in potden: introduce subroutines SpinsToChargeAndMagnetisation and ChargeAndMagnetisationToSpins
Writing preconditioned quantities to files is now substituted by passing types. This changes a lot.

With this bugfix my preconditioner input files should work again.

Bugfixes for LO part of Hamiltonian setup. Was broken for multiple atoms with LOs and inversion symmetry...


(cherry picked from commit 87f4143a)

At the moment this variable is without function. Later it will be used as a switch between the Yukawa and Coulomb cases in the potential generation.
Since the preconditioning parameter is now stored in the input file and since both the Coulomb and Yukawa potentials are generated in a calculation with preconditioner, the switch cannot simply be based on the value of the preconditioning parameter. This is quite ugly and we might want to change this somehow later.

I added preconditioning_param in the SCFLoopType in inp.xml as an optional parameter with default value 0.0.
It does not appear in the old input file.
Tests: a) generation of input file with default value; b) using this input file; c) using an inp.xml file with erased parameter; d) ctest. All tests (except ctests that currently also failed without this change) were OK.

...together with Miriam Hinzen.

...also: Some code beautification in that routine

pw_w is the warped potential (with step function)
pw is the potential without the step function

Also in this commit: Slight beautification of the configure script output for ChASE.

The user is responsible for providing the required paths and libraries.

ChASE needs several libraries to be available to work.
This includes MPI, LAPACK, LIBSTDC++.

My way to let the ChASE test succeed at the moment is to invoke
the configure script with

FLEUR_LIBRARIES="-L**PATH_TO:libchase_fleur.a**;-lchase_fleur;-lstdc++" ./fleur/configure.sh IFF

...test built together with Miriam.

This is to remove false Valgrind errors due to unused parts of those arrays.

The problem was that the output variable namex wasn't set any more. Of course,
this variable is not used for most things but it is at least used in w_inpXML
which is called at several points in the code. At some point this variable
should be eliminated in the whole code.

...this also fixes the related "gained too much charge" error.