Interface with VASP

The VASP interface relies on new options introduced since version 5.4.x In particular, a new INCAR-option LOCPROJ, the new LORBIT modes 13 and 14 have been added, and the new ICHARG mode 5 for charge self-consistent DFT+DMFT calculations have been added.

The VASP interface methodologically builds on the so called projection on localized orbitals (PLO) scheme, where the resulting KS states from DFT are projected on localized orbitals, which defines a basis for setting up a Hubbard-like model Hamiltonian. Resulting in lattice object stored in SumkDFT. The implementation is presented in M. Schüler et al. 2018 J. Phys.: Condens. Matter 30 475901.

The interface consists of two parts, PLOVASP, a collection of python classes and functions converting the raw VASP output to proper projector functions, and the python based VaspConverter, which creates a h5 archive from the PLOVASP output readable by SumkDFT. Therefore, the conversion consist always of two steps.

Here, we will present a guide how the interface can be used to create input for a DMFT calculation, using SrVO3 as an example. Full examples can be found in the tutorial section of DFTTools.

Limitations of the interface

The interface works correctly only if the k-point symmetries are turned off during the VASP run (ISYM=-1).
Generation of projectors for k-point lines (option Lines in KPOINTS) needed for Bloch spectral function calculations is not possible at the moment.
The interface currently supports only collinear-magnetism calculation (this implies no spin-orbit coupling) and spin-polarized projectors have not been tested.
The converter needs the correct Fermi energy from VASP, which is read from the LOCPROJ file. However, VASP by default does not output this information. Please see Remarks on the VASP version.

VASP: generating raw projectors

The VASP INCAR option LOCPROJ selects a set of localized projectors that will be written to the file LOCPROJ after a successful VASP run. A projector set is specified by site indices, labels of the target local states, and projector type:

LOCPROJ = <sites> : <shells> : <projector type>

where <sites> represents a list of site indices separated by spaces, with the indices corresponding to the site position in the POSCAR file; <shells> specifies local states (see below); <projector type> chooses a particular type of the local basis function. The recommended projector type is Pr 2. This will perform a projection of the Kohn-Sham states onto the VASP PAW projector functions. The number specified behind Pr is selecting a specific PAW channel, see the VASP wiki page for more information. The formalism for this type of projectors is presented in M. Schüler et al. 2018 J. Phys.: Condens. Matter 30 475901. For further details on the LOCPROJ flag also have a look in the VASP wiki.

The allowed labels of the local states defined in terms of cubic harmonics are (mind the order):

Entire shells: s, p, d, f

p-states: py, pz, px

d-states: dxy, dyz, dz2, dxz, dx2-y2

f-states: fy(3x2-y2), fxyz, fyz2, fz3, fxz2, fz(x2-y2), fx(x2-3y2).

For projector type Pr, one should ideally also set LORBIT = 14 in the INCAR file and provide parameters EMIN, EMAX, defining, in this case, an energy range (energy window) corresponding to the valence states. Note that, as in the case of a DOS calculation, the position of the valence states depends on the Fermi level, which can usually be found at the end of the OUTCAR file. Setting LORBIT=14 will perform an automatic optimization of the PAW projector channel as described in M. Schüler et al. 2018 J. Phys.: Condens. Matter 30 475901, by using a linear combination of the PAW channels, to maximize the overlap in the chosen energy window between the projector and the Kohn-Sham state. Therefore, setting LORBIT=14 will let VASP ignore the channel specified after Pr. This optimization is only performed for the projector type Pr, not for Ps and obviously not for Hy. We recommend to specify the PAW channel anyway, in case one forgets to set LORBIT=14.

In case of SrVO3 one may first want to perform a self-consistent calculation to know the Fermi level and the rough position of the target states. In the next step one sets ICHARG = 1 and adds the following additional lines into INCAR (provided that V is the second ion in POSCAR):

EMIN = 3.0

EMAX = 8.0

LORBIT = 14

LOCPROJ = 2 : d : Pr 2

The energy range does not have to be precise. Important is that it has a large overlap with valence bands and no overlap with semi-core or high unoccupied states. This INCAR will calculate and write-out projections for all five d-orbitals.

VASP input-output

The calculated projections \(\langle \chi_L | \Psi_\mu \rangle\) are written into files PROJCAR and LOCPROJ. The difference between these two files is that LOCPROJ contains raw matrices without any reference to sites/orbitals, while PROJCAR is more detailed. In particular, the information that can be obtained for each projector from PROJCAR is the following:

site (and species) index

for each k-point and band: a set of complex numbers for labeled orbitals

At the same time, LOCPROJ contains the total number of projectors (as well as the number of k-points, bands, and spin channels) in the first line, which can be used to allocate the arrays before parsing.

Conversion for the DMFT self-consistency cycle

The projectors generated by VASP require certain post-processing before they can be used for DMFT calculations. The most important step is to (ortho-)normalize them within an energy window that selects band states relevant for the impurity problem. This will create proper Wannier functions of the initial projections produced by VASP. Note that this energy window is different from the one described above and it must be chosen independently of the energy range given by EMIN, EMAX in the INCAR VASP input file. This part is done in PLOVASP.

PLOVASP: converting VASP output

PLOVASP is a collection of python functions and classes, post-processing the raw VASP LOCPROJ output creating proper projector functions.

The following VASP files are used by PLOVASP:

PROJCAR, LOCPROJ: raw projectors generated by VASP-PLO interface
EIGENVAL: Kohn-Sham eigenvalues as well as k-points with weights and Fermi weights
IBZKPT: k-point data (\(\Gamma\))
POSCAR: crystal structure data

To run PLOVASP, one first prepares an input file <name>.cfg (default name plo.cfg) that describes the definition of the correlated subspace. For SrVO3 this input file would look like this:

[General]
DOSMESH = -3.0 3.0 2001

[Shell 1]
LSHELL = 2
IONS = 2
EWINDOW = -1.4 2.0

TRANSFORM = 1.0  0.0  0.0  0.0  0.0
            0.0  1.0  0.0  0.0  0.0
            0.0  0.0  0.0  1.0  0.0

In the [section] general, the DOSMESH defines an energy window and number of data points, which lets the converter calculate the density of states of the created projector functions in a given energy window. Each projector shell is defined by a section [Shell 1] where the number can be arbitrary and used only for user convenience. Several parameters are required

IONS: list of site indices which must be a subset of indices given earlier in the VASP INCAR LOCPROJ flag. Note: If projections are performed for multiple sites one can specify symmetry equivalent sites with brackets: [2 3]. Here the projector are generated for ions 2 and 3, but they will be marked as symmetry equivalent later in ‘SumkDFT’.
LSHELL: \(l\)-quantum number of the projector shell; the corresponding orbitals must be present in LOCPROJ.
EWINDOW: energy window in which the projectors are normalized; note that the energies are defined with respect to the Fermi level.

The Option TRANSFORM is optional here, and it is specified to extract only the three \(t_{2g}\) orbitals out of the five d orbitals given by \(l = 2\). A detailed explanation of all input parameters can be found further below PLOVASP detailed guide.

Next, the converter is executed. This can be done by calling PLOVASP directly in the command line with the input file as an argument, e.g.:: plovasp plo.cfg

or embedded in a python script as:

import triqs_dft_tools.converters.plovasp.converter as plo_converter
# Generate and store PLOs
plo_converter.generate_and_output_as_text('plo.cfg', vasp_dir='./')

This will create the xml files vasp.ctrl and vasp.pg1 containing the orthonormalized projector functions readable by the VaspConverter. Moreover, PLOVASP will output important information of the orthonormalization process, such as the density matrix of the correlated shell and the local Hamiltonian.

Running the VASP converter

The actual conversion to a h5-file is performed with the orthonormalized projector functions readable by the VaspConverter in the same fashion as with the other DFTTools converters:

from triqs_dft_tools.converters.vasp import *
Converter = VaspConverter(filename = 'vasp')
Converter.convert_dft_input()

As usual, the resulting h5-file can then be used with the SumkDFT class::: sk = SumkDFTTools(hdf_file=’vasp.h5’)

Note that the automatic detection of the correct block structure might fail for VASP inputs. This can be circumvented by setting a bigger value of the threshold in SumkDFT, e.g.:

SK.analyse_block_structure(threshold = 1e-4)

However, this should only be done after a careful study of the density matrix and the projected DOS in the localized basis. For the complete process for SrVO3 see the tutorial for the VASP interface here.

PLOVASP detailed guide

The general purpose of the PLOVASP tool is to transform raw, non-normalized projectors generated by VASP into normalized projectors corresponding to user-defined projected localized orbitals (PLO). To enhance the performance parsing the raw VASP output files, the parser is implemented in plain C. The idea is that the python part of the parser first reads the first line of LOCPROJ and then calls the C-routine with necessary parameters to parse PROJCAR. The resulting PLOs can then be used for DFT+DMFT calculations with or without charge self-consistency. PLOVASP also provides some utilities for basic analysis of the generated projectors, such as outputting density matrices, local Hamiltonians, and projected density of states.

PLOs are determined by the energy window in which the raw projectors are normalized. This allows to define either atomic-like strongly localized Wannier functions (large energy window) or extended Wannier functions focusing on selected low-energy states (small energy window).

In PLOVASP, all projectors sharing the same energy window are combined into a projector group. This allows one in principal to define several groups with different energy windows for the same set of raw projectors. Note: multiple groups are not yet implemented.

A set of projectors defined on sites related to each other either by symmetry or by an atomic sort, along with a set of \(l\), \(m\) quantum numbers, forms a projector shell. There could be several projectors shells in a projector group, implying that they will be normalized within the same energy window.

Projector shells and groups are specified by a user-defined input file whose format is described below. Additionally, each shell can be marked correlated or non-correlated, to tell SumkDFT whether or not these should be treated in the DMFT impurity problem.

Input file format

The input file is written in the standard config-file format. Parameters (or ‘options’) are grouped into sections specified as [Section name]. All parameters must be defined inside some section.

A PLOVASP input file can contain three types of sections:

[General]: includes parameters that are independent of a particular projector set, such as the Fermi level, additional output (e.g. the density of states), etc.
[Group <Ng>]: describes projector groups, i.e. a set of projectors sharing the same energy window and normalization type. At the moment, DFTtools support only one projector group, therefore there should be no more than one projector group.
[Shell <Ns>]: contains parameters of a projector shell labelled with <Ns>. If there is only one group section and one shell section, the group section can be omitted but in this case, the group required parameters must be provided inside the shell section.

Section [General]

The entire section is optional and it contains four parameters:

BASENAME (string): provides a base name for output files. Default filenames are vasp.*.
DOSMESH ([float float] integer): if this parameter is given, the projected density of states for each projected orbital will be evaluated and stored to files pdos_<s>_<n>.dat, where s is the shell index and n the ion index. The energy mesh is defined by three numbers: EMIN EMAX NPOINTS. The first two can be omitted in which case they are taken to be equal to the projector energy window. Important note: at the moment this option works only if the tetrahedron integration method (ISMEAR = -4 or -5) is used in VASP to produce LOCPROJ.
EFERMI (float): provides the Fermi level. This value overrides the one extracted from VASP output files.
HK (True/False): If True, the projectors are applied the the Kohn-Sham eigenvalues which results in a Hamitlonian H(k) in orbital basis. The H(k) is written for each group to a file Basename.hk<Ng>. It is recommended to also set COMPLEMENT = True (see below). Default is False.

There are no required parameters in this section.

Section [Shell]

This section specifies a projector shell. Each [Shell] section must be labeled by an index, e.g. [Shell 1]. These indices can then be referenced in a [Group] section.

In each [Shell] section two parameters are required:

IONS (list of integer): indices of sites included in the shell. The sites can be given either by a list of integers IONS = 5 6 7 8 or by a range IONS = 5..8. The site indices must be compatible with the POSCAR file. Morever, sites can be marked to be identical by grouping them with brackets, i.e. IONS = [5 6] [7 8] will mark the sites 5 and 6 in the POSCAR (and of course also 7 and 8) to be idential. This will mark these correlated site as equivalent, and only one impurity problem per bracket group is generated.
LSHELL (integer): \(l\) quantum number of the desired local states.

It is important that a given combination of site indices and local states given by LSHELL must be present in the LOCPROJ file.

There are additional optional parameters that allow one to transform the local states:

CORR (True/False): Determines if shell is correlated or not. At least one shell has to be correlated. Default is True.
SORT (integer): Overrides the default detection of ion sorts by supplying an integer. Default is None, for which the default behavior is retained.
TRANSFORM (matrix): local transformation matrix applied to all states in the projector shell. The matrix is defined by a (multiline) block of floats, with each line corresponding to a row. The number of columns must be equal to \(2 l + 1\), with \(l\) given by LSHELL. Only real matrices are allowed. This parameter can be useful to select certain subset of orbitals or perform a simple global rotation.
TRANSFILE (string): name of the file containing transformation matrices for each site. This option allows for a full-fledged functionality when it comes to local state transformations. The format of this file is described below.

Section [Group]

Each defined projector shell must be part of a projector group. In the current implementation of DFTtools only a single group (labelled by any integer, e.g. [Group 1]) is supported. This implies that all projector shells must be included in this group.

Required parameters for any group are the following:

SHELLS (list of integers): indices of projector shells included in the group. All defined shells must be grouped.
EWINDOW (float float): the energy window specified by two floats: bottom and top. All projectors in the current group are going to be normalized within this window. Note: This option must be specified inside the [Shell] section if only one shell is defined and the [Group] section is omitted.

Optional group parameters:

NORMALIZE (True/False): specifies whether projectors in the group are to be normalized. The default value is True.
NORMION (True/False): specifies whether projectors are normalized on a per-site (per-ion) basis. That is, if NORMION = True, the orthogonality condition will be enforced on each site separately but the Wannier functions on different sites will not be orthogonal. If NORMION = False, the Wannier functions on different sites included in the group will be orthogonal to each other. The default value is False
BANDS (int int): the energy window specified by two ints: band index of lowest band and band index of highest band. Using this overrides the selection in EWINDOW.
COMPLEMENT (True/False). If True, the orthogonal complement is calculated resulting in unitary (quadratic) projectors, i.e., the same number of orbitals as bands. It is required to have an equal number of bands in the energy window at each k-point. Default is False.

File of transformation matrices

Warning

The description below applies only to collinear cases (i.e., without spin-orbit coupling). In this case, the matrices are spin-independent.

The file specified by option TRANSFILE contains transformation matrices for each ion. Each line must contain a series of floats whose number is either equal to the number of orbitals \(N_{orb}\) (in this case the transformation matrices are assumed to be real) or to \(2 N_{orb}\) (for the complex transformation matrices). The total number of lines \(N\) must be a multiple of the number of ions \(N_{ion}\) and the ratio \(N / N_{ion}\), then, gives the dimension of the transformed orbital space. The lines with floats can be separated by any number of empty or comment lines (starting from #), which are ignored.

A very simple example is a transformation matrix that selects the \(t_{2g}\) manifold. For two correlated sites, one can define the file as follows:

# Site 1
0   0.0   0.0   0.0   0.0
0   1.0   0.0   0.0   0.0
0   0.0   0.0   1.0   0.0

# Site 2
0   0.0   0.0   0.0   0.0
0   1.0   0.0   0.0   0.0
0   0.0   0.0   1.0   0.0

Remarks on the VASP version

In the current version of the interface the Fermi energy is extracted from the DOSCAR. However, if one pursues to do charge self-consistent calculations one needs to write the Fermi energy to the projectors (LOCPROJ file), as the DOSCAR is only updated after a full SCF/NSCF run. The file should contain the Fermi energy in the header. One can either copy the Fermi energy manually there after a successful VASP run, or modify the VASP source code slightly, by replacing the following line in locproj.F (around line 695):

<   WRITE(99,'(4I6,"  # of spin, # of k-points, # of bands, # of proj" )') NS,NK,NB,NF
---
>   WRITE(99,'(4I6,F12.7,"  # of spin, # of k-points, # of bands, # of proj, Efermi" )') W%WDES%NCDIJ,NK,NB,NF,EFERMI

Now one needs to pass additionally the variable EFERMI to the function, by changing (at arount line 560):

<   SUBROUTINE LPRJ_WRITE(IU6,IU0,W)
---
>   SUBROUTINE LPRJ_WRITE(IU6,IU0,W,EFERMI)
    REAL(q) :: EFERMI

Next, we need to pass this option when calling from electron.F and main.F (just search for LPRJ_WRITE in the files) and change all occurences as follows:

<   CALL LPRJ_WRITE(IO%IU6, IO%IU0, W)
---
>   CALL LPRJ_WRITE(IO%IU6, IO%IU0, W, EFERMI)

Now Vasp should print in the header of the LOCPROJ file additionally the Fermi energy.

Another critical point for CSC calculations is the function call of LPRJ_LDApU in VASP. This function is not needed, and was left there for debug purposes, but is called every iteration. Removing the call to this function in electron.F in line 644 speeds up the calculation significantly in the ICHARG=5 mode. Moreover, this prevents VASP from generating the GAMMA file, which should ideally only be done by the DMFT code after a successful DMFT step, and then be read by VASP.

Furthermore, there is a bug in fileio.F around line 1710 where VASP tries to print “reading the density matrix from Gamma”. This should be done only by the master node, and VASP gets stuck sometimes. Adding a

IF (IO%IU0>=0) THEN
...
ENDIF

statement resolves this issue. A similar problem occurs, when VASP writes the OSZICAR file and a buffer is stuck. Adding a flush to the buffer in electron.F around line 580 after

CALL STOP_TIMING("G",IO%IU6,"DOS")
flush(17)
print *, ' '

resolves this issue. Otherwise the OSZICAR file is not written properly.