Noramlly, the Fermi energy is read from DOSCAR. However, this does
not work in case of a self-consistent calculation in which DOSCAR
is not written between iterations. One of the options is
to modify slightly the output to LOCPROJ and add EFERMI to the
first line.
Since 'n_orbitals' can be a 2D array in case of spin-polarized
calculations, one should use 'numpy.max' instead of 'max' to
extract the maximum number of bands.
In the new version of VASP LOCPROJ contains the eigenvalues and
Fermi weights. Also, during a charge self-consistency calculation
the file EIGENVAL is not written at intermediate iterations. It is,
thus, preferential to use LOCPROJ to get the named data.
At the moment, EIGENVAL will still be used if it is complete but
in the future this dependence should be removed completely.
The band indices should be converted to Fortran convention,
i.e. starting from 1, in the output files because the are
used in the density matrix file which is read by a Fortran code.
The format of LOCPROJ has been modified again (in VASP 5.4.2
build from Dec 02, 2015).
Now, there is an additional line before each projector block
providing the spin, k-, and band indices, as well as
eigenvalues and Fermi weights.
Scripts 'run_plovasp.sh' have been replaced by a template in which
the path must be set by the user.
Also, .gitignore has been added to example 'lunio3'.
At one step of the orthogonaliztion procedure two matrix multiplications
have been replaced with one matrix multiplication and a element-wise
multiplication of a vector and a matrix.
Fermi weights are output next to eigenvalues. They will be needed
for the calculation of the KS density matrix in the charge
self-consistency implementation.
The part responsible for generating a mapping between the shell/ions
and block projector matrices has now been relocated to a separate
method 'get_block_matrix_map()'. This simplifies the source code
and makes testing easier.
The mapping for NORMION = True has been implemented.
Also, the orthogonalization loop has been fixed. First of all,
orthogonalization should be done separately for each block map 'bl_map'.
Second, one has to take into account that the orbital dimensions of the
block matrix can vary from block to block. To make that the overlap
matrix is non-singular one, thus, has to pass to
'orthogonalize_projector_matrix()' only a view of a submatrix of 'pmat'
corresponding to the current block.
Two tests to check the simplest cases have been added.
The implementation of the mapping of a set of projectors (belonging
to different shells and ions) onto a block matrix in the
orthogonalization routine has been generalized. Now, an implementation
of the choice between the full orthogoanlization and per-site one
is straightforward: it is just a matter of defining a proper mapping.
The mapping scheme itself is described in the doc-string of method
'ProjectorGroup.orthogonalize()'
There was a very nasty bug in the preparation of the block matrix
'p_mat'. The point is that this matrix is created once for all k-points
with the band dimension being the maximum possible. However, only
a part of the matrix is used at every k-point but the orthogonalization
is done for the whole matrix. The problem was that if the number of
bands for a given k-point was smaller than that for the next k-point
them for the next k-point some part of 'p_mat' still contained data from
the previous step, which messed up the orthonormalization. Now, 'p_mat'
is set to zero at each step of the loop.
Also, property 'nion' was added to ProjectorShell since it is used
very often.
First of all, suite '_plotools' is now split into three separate suites
'_plotools', '_proj_shell', '_proj_group', following the changes made
into the structure of the code.
Second, the two tests in 'test_projshells.py' have been fixed to conform
to the recent modifications in the code and input files.