0% found this document useful (0 votes)
149 views31 pages

WIEN2k Paper

This document summarizes the WIEN2k program, which uses the augmented plane wave plus local orbitals (APW+lo) method to solve the Kohn-Sham equations of density functional theory for calculating the properties of solids. The APW+lo method considers all electrons self-consistently in a full-potential treatment. WIEN2k implements the method efficiently with parallelization and optimized numerical libraries. It can calculate many material properties, from basic ones like band structure and atomic structure to more specialized ones like NMR shielding tensor and electric polarization.

Uploaded by

daniel.fidelis
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
149 views31 pages

WIEN2k Paper

This document summarizes the WIEN2k program, which uses the augmented plane wave plus local orbitals (APW+lo) method to solve the Kohn-Sham equations of density functional theory for calculating the properties of solids. The APW+lo method considers all electrons self-consistently in a full-potential treatment. WIEN2k implements the method efficiently with parallelization and optimized numerical libraries. It can calculate many material properties, from basic ones like band structure and atomic structure to more specialized ones like NMR shielding tensor and electric polarization.

Uploaded by

daniel.fidelis
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 31

WIEN2k: An APW+lo program for

calculating the properties of solids


Cite as: J. Chem. Phys. 152, 074101 (2020); https://doi.org/10.1063/1.5143061
Submitted: 19 December 2019 . Accepted: 24 January 2020 . Published Online: 19 February 2020

Peter Blaha , Karlheinz Schwarz , Fabien Tran , Robert Laskowski , Georg K. H. Madsen , and
Laurence D. Marks

COLLECTIONS

Note: This paper is part of the JCP Special Topic on Electronic Structure Software.

This paper was selected as an Editor’s Pick

J. Chem. Phys. 152, 074101 (2020); https://doi.org/10.1063/1.5143061 152, 074101

© 2020 Author(s).
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

WIEN2k: An APW+lo program for calculating


the properties of solids
Cite as: J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061
Submitted: 19 December 2019 • Accepted: 24 January 2020 •
Published Online: 19 February 2020

Peter Blaha,1,a) Karlheinz Schwarz,1 Fabien Tran,1 Robert Laskowski,2 Georg K. H. Madsen,1
and Laurence D. Marks3

AFFILIATIONS
1
Institute of Materials Chemistry, Vienna University of Technology, Getreidemarkt 9/165-TC, A-1060 Vienna, Austria
2
Institute of High Performance Computing, A∗ STAR, 1 Fusionopolis Way, #16-16, Connexis 138632, Singapore
3
Department of Materials Science and Engineering, Northwestern University, Evanston, Illinois 60208, USA

Note: This paper is part of the JCP Special Topic on Electronic Structure Software.
a)
Author to whom correspondence should be addressed: [email protected]

ABSTRACT
The WIEN2k program is based on the augmented plane wave plus local orbitals (APW+lo) method to solve the Kohn–Sham equations
of density functional theory. The APW+lo method, which considers all electrons (core and valence) self-consistently in a full-potential
treatment, is implemented very efficiently in WIEN2k, since various types of parallelization are available and many optimized numer-
ical libraries can be used. Many properties can be calculated, ranging from the basic ones, such as the electronic band structure or
the optimized atomic structure, to more specialized ones such as the nuclear magnetic resonance shielding tensor or the electric polar-
ization. After a brief presentation of the APW+lo method, we review the usage, capabilities, and features of WIEN2k (version 19) in
detail. The various options, properties, and available approximations for the exchange-correlation functional, as well as the external
libraries or programs that can be used with WIEN2k, are mentioned. References to relevant applications and some examples are also
given.
© 2020 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license
(http://creativecommons.org/licenses/by/4.0/). https://doi.org/10.1063/1.5143061., s

I. INTRODUCTION approaches focused on dealing with the electrons via wave func-
tions,1 many current methods use density functional theory (DFT),2
Quantum mechanical calculations play a central role in under- which has significant speed advantages. Following the method out-
standing the properties of materials and, increasingly, predicting the lined by Kohn and Sham3 (KS), the interacting many-body system
properties of new materials. While in the early days, the emphasis of electrons is mapped onto a non-interacting system of quasi-
was mainly on understanding the energy, atom positions, and band particles, characterized by KS orbitals with a specific KS energy.
structure, modern codes now calculate a large number of differ- They have many of the properties of the true electron wave func-
ent properties ranging from piezoelectric response to nuclear mag- tions and of particular importance is that one can fill up these
netic resonance (NMR) shielding, examples of which will be given KS orbitals as a function of their KS energy yielding the true
later. With the advent of increasingly sophisticated methods and the electron density. The KS approach needs an exchange-correlation
ever increasing speed of computers over the last decades, in some (XC) functional and the corresponding XC potential. However, the
cases the accuracy of quantum mechanical calculations rivals or even exact functional is unknown and approximations are needed (see
surpasses the accuracy of experimental measurements. Sec. II B).
There are many different methods of theoretically modeling A second split in terms of methods is how the atomic positions
the behavior of electrons and atoms in materials. While earlier are considered, and there are two main methods: cluster calculations

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-1


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

for a finite number of atoms, which focus on the local properties of II. THEORY
some atomic arrangement, and those which are designed to exploit In the APW-based methods, the unit cell is decomposed into
the periodic nature of most solids; the WIEN2k code is an example spheres centered at the nuclear sites and an interstitial region,7 as
of the latter. We represent the solid by a unit cell, which is repeated shown in Fig. 1. These atomic spheres with radii RMT must not
in all three directions, corresponding to periodic boundary condi- overlap, but should be chosen for computational efficiency as large
tions. This assumes that the solid is perfect, ordered, and infinite; as possible with the additional constraint that RMT for d-elements
however, a real crystal differs from this ideal situation, since it is should be chosen to be about 10%–20% bigger than for sp-elements,
finite, may contain defects or impurities, and may deviate from its while f -elements should get even larger spheres because for iden-
ideal stoichiometry. For these important aspects and how to handle tical sphere sizes the number of plane-waves (PWs) to reach con-
them using supercells, see Chap. 8.2 of Ref. 4. vergence is largest for the localized 4f (5f ) electrons, medium for
There are many computational methods for solving the KS 3d (4d, 5d)-electrons, and much smaller for sp-states. An excep-
equations, for instance, linear combination of atomic orbitals tion is the H atom, whose sphere with short C–H or O–H bonds
(LCAO), numerical basis sets, pseudopotential schemes, or space should be chosen approximately half the size of RMT (C) or RMT (O).
partitioning methods. A recent comparison5 of these methods In WIEN2k, these sphere radii can be set automatically in an opti-
showed that especially all-electron codes predict essentially identical mal way using the setrmt utility. Note that non-optimal sphere sizes
results, demonstrating a high reproducibility, whereas some pseu- may lead to poor convergence (eventually only for one particular
dopotential codes lead to large deviations. One of the most accurate atom) and significantly longer computing time or suffer from trun-
codes is our WIEN2k code,6 which is the focus of this paper and cated Fourier or spherical-harmonic expansions. In the worst case,
is based on the augmented plane wave (APW) method. Detailed they can even produce “ghost-states” (unphysical eigenvalues in the
descriptions including many conceptual and mathematical details occupied spectrum) if the RMT of an sp-element is much larger than
are given in Ref. 7. The term all-electron (see Chap. 8.4 of Ref. 4) that of the other atoms.
means that all electrons from the core (starting from the 1s shell) to The electron density ρ and KS potential vKS (defined in Sec. II B)
the valence states are included. are expanded as a Fourier series in the interstitial (I) region (K
Turning to some historical specifics of our approach (see also denotes a reciprocal lattice vector in units of inverse bohr) and
Ref. 8), Slater9 proposed the original APW method. Unfortunately, as lattice harmonics (symmetry adapted combinations of spheri-
the original formulation leads to a nonlinear eigenvalue problem due cal harmonics ZLM )24 times radial functions ρLM (r) [vKS LM (r) for the
to the energy-dependent radial basis functions, which is computa- potential] inside the spheres (St , where t is the atom index),
tionally expensive. An important improvement came from Ander-
sen,10 who introduced a linearization of this energy dependency, ⎧

⎪ ∑ ρLM (r)ZLM (r̂), r ∈ St

and Koelling and Arbman11 made the linearized-APW (LAPW) ⎪
ρ(r) = ⎨L,M iK⋅r (1)
method a practical computational scheme using the muffin-tin (MT) ⎪


⎪∑ ρK e , r ∈ I,
approximation (see Sec. II). This was taken a step further by Free- ⎩K
man and collaborators who made the LAPW method a full-potential
⎧ KS
all-electron total energy method.12,13 ⎪ ∑ v (r)ZLM (r̂),
⎪L,M LM

⎪ r ∈ St
KS
This LAPW method formed the basis for the original WIEN v (r) = ⎨ KS iK⋅r (2)
code.14 However, the LAPW method had the drawback that only



⎪∑ vK e , r ∈ I.
⎩K
one principal quantum number per angular momentum ℓ could be
described and thus failed to give reliable results for all elements By default, the Fourier expansion runs up to |K| = 12 for large
on the left of the periodic table because these atoms require a RMT (16 for RMT < 1.2; 20 for RMT < 0.7 bohr), while the angular
proper description of shallow core states (semi-core) and valence momentum expansion truncates at L = 6. Note that the old “MT”
states at the same time (e.g., 1s and 2s in Li or 3sp and 4sp in Ti). approximation uses a constant value in the interstitial (i.e., only
This problem was solved by Singh,15 who introduced local orbitals K = 0) and a spherically symmetric density/potential inside the
(LOs) for the description of semi-core states. He also noted that spheres (i.e., only L = 0).
the LAPW method needed a larger plane-wave basis set than the
APW method. To overcome this problem, he suggested the aug-
mented plane wave plus local orbitals (APW+lo) method,16,17 where
the linearization of the energy-dependent radial wave function was
facilitated by an extra local orbital (lo, different from an LO, see
Sec. II), which has a superior plane-wave convergence compared to
LAPW. Last but not least, the linearization of the energy depen-
dency can introduce some inaccuracy in high precision calcula-
tions. This problem was finally solved by introducing additional
higher (second) derivative LOs (HDLOs).18,19 These latest develop-
ments form the basis of the present WIEN2k_19 code,6 while pre-
vious versions have been described in several reviews.7,8,20–23 The
method of our choice could be named (L)APW+lo+LO+HDLO,
FIG. 1. Schematic unit cell with large transition metal (TM), medium O and small
but we use a shorter acronym APW+lo. It is described in detail in
H spheres, and the interstitial region in between.
Sec. II A.

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-2


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

This space decomposition plays a crucial role in the definition where u̇tℓ is the first energy derivative of utℓ . The coefficients Alo
tℓm
of core and valence electrons, which are treated differently in APW- and Blo lo
tℓm are chosen such that ϕtℓm is zero at RMT and normalized.
type methods. Core states are defined as having wave functions (den- The APW+lo basis set has the advantage of a superior PW con-
sities) completely confined inside the atomic spheres. Thus, we do vergence as compared to the standard LAPW method,7,17 reducing
not use the standard definitions of core and valence, but, e.g., in 3d the number of PWs by almost 50%, but it needs additional lo basis
transition metals (TMs), the 3s and 3p states are also considered as functions. Thus, in WIEN2k, the default is to restrict the ℓmax to
valence, since a couple of percent of their charge leaks out of the the chemical ℓ-values sp(d, f ), which are hard to converge, but use
atomic sphere. To distinguish them from the conventional defini- a standard LAPW basis set inside the spheres for the higher ℓ values
tion, we call them semi-core states. Typically, these states are treated (by default up to ℓmax = 10),
using LOs (see below) and their energies are less than 6 Ry below the
⎧ k+K k+K
Fermi energy, but in special situations (small spheres due to short ⎪
⎪ ∑ [Atℓm utℓ (r, Etℓ ) + Btℓm u̇tℓ (r, Etℓ )]Yℓm (r̂), r ∈ St
⎪ℓ,m
nearest neighbor distances or high pressure), even lower lying states ϕLAPW
k+K (r) = ⎨
⎪ 1 i(k+K)⋅r
(such as Al-2p) have to be included. Relativistic effects are impor- ⎪ √Ω e


, r ∈ I,
tant for the core states, and thus, they are calculated by numerically (5)
solving the radial Dirac equation in the spherical symmetric part of
the potential vKS . Core states are constrained to be localized and not where the coefficients Ak+K k+K LAPW
tℓm and Btℓm are chosen such that ϕk+K and
hybridized with states at the neighboring atoms, but we use a thawed its first derivative are continuous at the sphere boundary.
core (no frozen core approximation), i.e., the core states are recalcu- As mentioned before, semi-core states (or also high-lying
lated in each self-consistent field cycle.7 The semi-core and valence empty states) cannot be described accurately by APW+lo. For these
electrons are commonly treated scalar relativistically, i.e., including states, the basis set has to be improved, and this can be done by
mass velocity and Darwin s-shift corrections, but neglecting spin– adding another type of local orbitals, the LOs, containing radial
orbit (SO) interactions.7,25 The SO effects can later on be included functions utℓ calculated at the appropriate (e.g., semi-core) energy
LO,i
in a second variational step using the scalar-relativistic orbitals as Etℓ ,
a basis.7,26 Since p1/2 radial wave functions differ considerably from ⎧ LO,i LO,i LO,i
scalar relativistic (or p3/2 ) orbitals, one can also enrich the basis set ⎪[Atℓm utℓ (r, Etℓ ) + Ctℓm utℓ (r, Etℓ )]Yℓm (r̂), r ∈ St


with additional p1/2 local orbitals, specifically, an LO (see below) with ϕLO,i
tℓm (r) =⎨

⎪0,
⎪ r ∈ I.
a p1/2 radial wave function, which is added in the second-variational ⎩
SO calculation.27 (6)
For instance, for TiO2 , one would use the Ti-3p energy to calculate
LO,i
A. The APW+lo method as implemented in WIEN2k uTi,1 (r, ETi,1 ) and then add some Ti-4p radial function uTi,1 (r, ETi,1 ),
The basis functions for the valence electrons consist of APWs, choosing the coefficients ALO,i LO,i
tℓm and Ctℓm such that the LO is zero at
which are plane waves in the interstitial region augmented with the RMT and is normalized. By adding such LOs (representing Ti-3p
radial wave functions utℓ (r, Etℓ ) defined at a fixed energy Etℓ , and states), a consistent and accurate description of both the Ti-3p
lo.7,16,17 An APW is given by semi-core and Ti-4p valence states is possible, retaining orthogo-
nality, which is not assured when the multiple-window approach
⎧ k+K is used (see Ref. 7). Cases where this improvement is essential is

⎪ ∑ Atℓm utℓ (r, Etℓ )Yℓm (r̂),
⎪ℓ,m r ∈ St the electric field gradient (EFG) calculation of rutile TiO2 28 or lat-
ϕAPW
k+K (r) = ⎨ (3)

⎪ √1 e
i(k+K)⋅r
, r ∈ I, tice parameter calculations of compounds with such elements. Note

⎩ Ω how lo [Eq. (4)] and LO [Eq. (6)] differ in their respective second
terms.
where k is a point in the first Brillouin zone (BZ), Yℓm (r̂) are spher- A clever choice of energy parameters Etℓ in Eqs. (3)–(6) is
ical harmonics, and utℓ are solutions of the scalar-relativistic radial essential for accurate results, and WIEN2k has several automatic
KS equation7 inside the sphere St . Note that these radial functions ways to make an optimal choice in most cases.6 Etℓ of semi-core
utℓ are recalculated in each self-consistent-field (SCF) cycle, allow- states (actually, of all states whose energy in the free atom is more
ing for an expansion/contraction corresponding to the given charge than 0.5 Ry below the highest occupied atomic orbital, e.g., also
state (ionicity) of the atom. These adaptive basis functions are part of C-2s or Ar-3s states) are determined by taking the average of the
the reason for the high accuracy of APW-based methods. The coef- two energies Ebottom and Etop , where the corresponding utℓ (RMT ) is
ficients Ak+K
tℓm are chosen such that the interstitial and sphere parts zero or has zero slope. For localized d or f valence electrons, the same
of the APW match at the sphere boundary. However, these APWs procedure is used, but Etop is searched only 0.5 Ry above EF to ensure
allow no variations of the radial functions for eigenvalues different that the energy parameters are set below EF . The energy parameters
than Etℓ and thus would be a poor basis. To overcome this con- of all other valence states are set to 0.2 Ry below EF (0.2 Ry above EF
straint, the energy dependency is handled by a lo, which is nonzero if there is a high lying semi-core LO). Thus, all our energy parame-
only inside a MT sphere, and given by ters are dynamically updated during the SCF cycle and not fixed by
input.
⎧ lo lo Implicit in this approximation is a linearization of the
⎪[Atℓm utℓ (r, Etℓ ) + Btℓm u̇tℓ (r, Etℓ )]Yℓm (r̂),

⎪ r ∈ St
ϕlo
tℓm (r) = ⎨ (4) energy dependency of the radial wave functions. Since the true
⎪ r ∈ I,
⎩0,

⎪ utℓ (r, Etℓ = εi ) varies most for more localized states (e.g., 3d or

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-3


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

4f ) at larger distances from the nucleus, large spheres and a large The specific setup of all these basis functions can, of course,
valence bandwidth could cause a poor description of the variations be selected manually by experts, but one of the great strengths of
of utℓ (r, εi ) with energy, leading to a significant dependency of the WIEN2k is that the default input usually works quite well and is
results on RMT (where smaller RMT yield more correct results but fairly robust.
with a larger computational effort). This can be solved by adding
an LO, which involves the second energy derivative of utℓ , called an
B. Available DFT approximations
HDLO,18,19
In DFT, the total energy of the system is given by3
[AHDLOutℓ (r, Etℓ) + Ctℓm
HDLO
ütℓ (r, Etℓ )]Yℓm (r̂), r ∈ St
ϕHDLO
tℓm (r) = { tℓm
0, r ∈ I. Etot = Ts + Een + EH + Exc + Enn . (9)
(7)
The terms on the right-hand side represent the noninteracting
Figure 2 illustrates the effect of adding HDLOs for the lattice kinetic, electron–nucleus, Hartree, XC, and nucleus–nucleus ener-
parameter of fcc-La as a function of RMT . The lattice parameter gies, respectively. The variational principle leads to the KS [or gen-
becomes independent of RMT when both d and f -HDLOs are added eralized KS29 (gKS)] equations (in this section, the orbital index i is
to the basis set, while the standard APW+lo+LO basis (LOs for 5s, a shorthand notation for valence and core orbitals),
5p states) produces an error of 0.04 bohr for the largest RMT .
1
The KS orbitals are expanded using the combined basis set (− ∇2 + vKS (r))ψi (r) = εi ψi (r), (10)
described above (n is the band index), 2
i
ψnk = ∑ cnk ϕi , (8) where vKS is the KS potential,
i
i
and the coefficients cnk are determined by the Rayleigh–Ritz vari- vKS (r) = ven (r) + vH (r) + vxc (r), (11)
ational principle. The number of APW (or LAPW) basis functions
Eq. (3) [or Eq. (5)] is determined by the cutoff value K max for the which is the sum of the electron–nucleus, Hartree, and XC poten-
reciprocal lattice vectors K such that ∣k + K∣ ≤ Kmax and depends on tials. Choosing an appropriate functional Exc in Eq. (9) [and poten-
the smallest of the atomic radii Rmin tial vxc in Eq. (11)] for the XC term is crucial in order to obtain
MT and the type of atom. Typically,
the necessary Rmin reliable results for the problem at hand.30–32 Several hundred33,34
MT Kmax values range from 3 (for small H-spheres)
to 7 for sp-elements, 8 for TM-d elements, and 9 for 4f lanthanides. different functionals are available in the literature; some of them
These values can be reduced by 0.5–1 for low quality screening cal- were proposed as general-purpose functionals, while others were
culations and increased by 0.5–2 for highest precision. It should be devised for a specific property (e.g., bandgap) or types of systems
mentioned that the efficiency of the APW+lo method depends cru- (e.g., van der Waals). Numerous functionals have been implemented
cially on the possible RMT values. For instance, the O-2p states con- in the WIEN2k code, and below, we provide a brief overview of the
verge well with RO O different families of functionals. Note that for XC functionals that
MT Kmax = 7. In MgO, one can use RMT = 2 bohrs,
leading to a very small PW cutoff energy of 170 eV. However, in depend explicitly on the electron density ρ, e.g., the local density
Mg(OH)2 , one has to use RO approximation (LDA) or the generalized gradient approximation
MT = 1.1 bohrs due to the short O–H dis-
tances, leading to a PW cutoff of 550 eV, i.e., an order of magnitude (GGA), the XC potential vxc is multiplicative, while for functionals
larger effort. that depend implicitly on ρ, e.g., meta-GGA (MGGA) or hybrids,
the XC potential is non-multiplicative when implemented in the gKS
scheme.

1. LDA, GGA, and MGGA


The LDA, GGA, and MGGA represent the first three rungs
of Jacob’s ladder of XC functionals.35 These approximations are
semilocal, since Exc is defined as

Exc = ∫ εxc (r)d3 r, (12)

and the XC energy density εxc depends only locally on some proper-
ties of the system. In the LDA, εxc depends on the electron density
ρ = ∑Ni=1 ∣ψi ∣2 , while in the GGA, εxc depends also on the first deriva-
tive ∇ρ. At the MGGA level, the functionals depend additionally on
the Laplacian of the electron density ∇2 ρ and/or the kinetic-energy
density t = (1/2) ∑Ni=1 ∇ψi∗ ⋅ ∇ψi . Semilocal functionals are the most
commonly used methods in the solid-state community for the cal-
FIG. 2. Lattice parameter (bohr) using PBE of fcc La as a function of RMT using the culation of properties depending on the total energy such as the
standard APW+lo+LO basis set, or with additional f -HDLO or d+f -HDLO.
geometry, cohesive energy, or the adsorption energy of a molecule

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-4


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

on a surface. The main reason is that they are faster than all other potential.12 The treatment of the Coulomb singularity is done by
types of approximations and therefore allow calculations of larger multiplying the Coulomb potential by a step function,68 which is
systems. very efficient compared to other methods.69 The screened hybrid
There is a huge literature on the performance of semilocal functionals use the Yukawa potential, and in Ref. 66, it was shown
functionals, concerning the geometry and cohesive energy of solids. that the results obtained with the PBE-based hybrid YS-PBE0 for
Extensive benchmark studies have been conducted by us36–39 and the bandgap are almost identical to those obtained with HSE06
others (see, e.g., Refs. 40 and 41). The results of these works showed (which uses the erfc screened potential), provided that the screen-
that among the GGA functionals, those with a small enhance- ing parameter is chosen appropriately [λ = (3/2)μ, see Ref. 70].
ment factor such as AM05,42 PBEsol,43 or a few others44–46 are Because of the double integral and summations over orbitals in the
the most accurate for the lattice constant and bulk modulus, while HF exchange [Eq. (14)], the calculations are much more expensive
the standard PBE47 is the best choice for the cohesive energy. (between 10 and 1000 times) than semilocal methods; however, there
At the MGGA level, the SCAN functional48 is becoming increas- are a couple of ways to speed-up such calculations significantly.
ingly popular and has been shown to be simultaneously as good For instance, one can first use a rather crude k-mesh and later on
as the best GGAs for the geometry (e.g., PBEsol) and the cohe- improve the k convergence in a few additional iterations continu-
sive energy (PBE).39,41 However, it should be mentioned that SCAN ing the previous calculations. Furthermore, a reduced k-mesh for
can be quite problematic for iterant magnetic systems49,50 or alkali the internal loop in the HF potential is possible,71 and finally, often
metals.51 a one-shot procedure72,73 is sufficient, where the hybrid orbitals
Many semilocal functionals have been implemented directly in and eigenvalues are calculated perturbatively on top of a calcula-
the WIEN2k code, but basically all existing semilocal functionals can tion with the semilocal functional on which the hybrid functional is
be used because WIEN2k is interfaced to the Libxc33,34 library of XC based.
functionals. One current limitation is that the MGGA functionals Calculations using hybrid functionals in WIEN2k can be found
are not yet implemented self-consistently (by default, the GGA PBE in Refs. 74–77 for applications and in Refs. 39, 64, and 78–80 for
potential is used for generating the orbitals although the user can various benchmark studies.
choose another potential).
3. On-site methods for strongly correlated electrons
2. Hybrid functionals
The high computational cost of hybrid methods discussed in
Beginning in the 21st century, hybrid functionals,52 which Sec. II B 2 limits the size of the systems that can be treated. Alter-
belong to the fourth rung of Jacob’s ladder, started to be extensively natively, one can use an on-site method, namely, DFT+U,81 exact
used for calculations of solids.53–55 The one that is currently the most exchange for correlated electrons (EECE),82 or on-site hybrids,83
popular is HSE06,56–58 which is a screened version of the other well- which can be viewed as approximate but cheap versions of the hybrid
known PBE0.59,60 In (screened) hybrid functionals, the exchange or HF methods. In these methods, a hybrid/HF treatment is applied
energy is a linear combination of a semilocal (SL) functional and the only to the electrons of a particular angular momentum belonging to
Hartree–Fock (HF) expression, a selected atom. However, using such an on-site scheme only makes
hybrid SL sense when the considered electrons are well localized around the
Exc = Exc + αx (Ex(scr)HF − Ex(scr)SL ), (13) atom, which is, in general, the case for strongly correlated electrons.
The on-site methods are mostly applied to open 3d-, 4f -, or 5f -shells
where
in strongly correlated materials in order to improve the description
1 N N of the electronic and magnetic properties. For such systems, the stan-
Ex(scr)HF = − ∑ ∑ δσi σj ∬ ψi∗ (r)ψj (r) dard GGA methods provide results that are often even qualitatively
2 i=1 j=1
inaccurate.84
× v(∣r − r′ ∣)ψj∗ (r′ )ψi (r′ )d3 rd3 r′ . (14) Different versions of DFT+U exist in the literature, and those
available in WIEN2k are the following: (1) the original version,81
In Eq. (14), v is either the bare Coulomb potential v = 1/∣r − r′ ∣ for called Hubbard in mean field (HMF) in WIEN2k, (2) the fully local-
unscreened hybrids or a potential that is screened at short or long ized limit version,85,86 called self-interaction correction (SIC), and
range for screened hybrids. For solids, it is computationally advanta- (3) the around mean-field (AMF) version.86 The details of the imple-
geous to use a potential that is short range, for instance, the Yukawa mentation of DFT+U in the LAPW method can be found in the work

potential v = e−λ∣r−r ∣ /∣r − r′ ∣ (Ref. 61) or v = erfc(μ∣r − r′ ∣)/∣r − r′ ∣ of Shick et al.,87 while a very good summary and discussion of the
(Ref. 56), where erfc is the complementary error function. Although DFT+U flavors is given in Ref. 88. Note that since the on-site term
hybrid functionals are also used for the total energy (geometry opti- is applied only inside the sphere surrounding the atom of interest,87
mization, and cohesive energy), they are particularly interesting the results may depend on the radius RMT of this sphere (see Refs.
for properties derived from the electronic band structure such as 89 and 90 for illustrations), which is a drawback of the on-site meth-
the bandgap, for which they significantly improve upon standard ods. Among our works reporting DFT+U calculations, we mention
GGA functionals such as PBE (see Refs. 62–65 for recent extensive Refs. 77 and 91–93.
benchmarking). For many technical aspects, the EECE and on-site hybrid meth-
In WIEN2k, unscreened and screened hybrid functionals are ods are quite similar to DFT+U; however, there are two con-
implemented66 according to the scheme of Massidda et al.,67 which ceptual differences. The first one concerns the double-counting
is based on the pseudo-charge method for calculating the Coulomb term. While in DFT+U, the double-counting term is derived using

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-5


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

concepts from the Hubbard model (see Ref. 88 for a summary of being the average of ∣∇ρ∣/ρ in the unit cell. The parameters in
the various expressions), in EECE and on-site hybrids, it is given by Eq. (16) are α = −0.012, β = 1.023 bohrs1/2 , and p = 1/2 and were
the semilocal expression of Exc (e.g., PBE) evaluated with the den- determined by minimizing the mean absolute error of the bandgap
sity of the strongly correlated electrons.82 The second difference is for a set of solids.105 As shown in benchmark studies,64,65,80 the
the calculation of the Slater integrals in the Coulomb and Hartree– TB-mBJ potential is currently the most accurate semilocal method
Fock terms. In DFT+U, they are parameterized with screened intra- for bandgap prediction. Other parameterizations of Eq. (16) were
atomic Coulomb (U) and exchange (J) interactions, which are usu- proposed in Refs. 110 and 111 and are also available in WIEN2k.
ally chosen empirically. However, in EECE and on-site hybrids, the The GLLB-SC potential112,113 is given by
Slater integrals are calculated explicitly using the orbitals of the
strongly correlated electrons.82 The results obtained with the on-site vGLLB-SC
xc (r) = 2ePBEsol
x (r) + KxLDA
hybrid methods can be found in Refs. 83 and 94–98. √ ∣ψnk (r)∣2
The results obtained with DFT+U and on-site hybrids should ×∑ εH − εnk + vPBEsol
c (r), (18)
be qualitatively similar in many (but not necessary all) cases.83,99–101 n,k ρ(r)
Actually, both methods contain empirical parameters: U and J (or
only U eff = U − J) in DFT+U and αx [Eq. (13)] in on-site hybrids. where ePBEsol
x is the PBEsol exchange-energy density per electron,
In both cases, the results will depend crucially on the value of the vPBEsol
c √ = δEc
PBEsol
/δρ is the PBEsol correlation potential, and KxLDA
parameters (U, J) or αx . For applications, the EECE method82 is 2
= 8 2/(3π ). Since the GLLB-SC potential depends on the orbital
less interesting since it consists of 100% of unscreened Hartree– energies (εH is the one at the valence band maximum), a non-zero
Fock exchange applied to correlated electrons, which usually is not derivative discontinuity114,115 can be calculated and added to the KS
accurate. bandgap for comparison with the experimental value.113,116 Similar
Technically, we mention that calculations with the on-site to TB-mBJ, the GLLB-SC potential is significantly more accurate
methods can only be done in a spin-polarized mode, i.e., with than traditional GGA functionals, as shown in Refs. 80, 104, 113,
runsp_lapw. However, it is possible to apply on-site methods to and 117. On the other hand, we note that these potentials are not
non-magnetic systems by using the script runsp_c_lapw, which obtained as the functional derivative of an energy functional.
constrains the system to have no spin polarization. Among other DFT methods, which have been shown to
provide bandgaps more accurately than PBE, are the GGAs
EV93PW91,118,119 AK13,120,121 and HLE16,122 as well as the LDA-
4. Methods for bandgaps
type functional Sloc.123 As mentioned in Sec. II B 1, the potential
It is well known that the GGA functionals that are commonly of MGGA energy functionals is not implemented in WIEN2k; how-
used for total-energy calculations, such as PBE, provide bandgaps ever, it is still possible to calculate bandgaps non-self-consistently
that are much smaller than experiment.57 Thus, one has to resort using the total energy (see Ref. 124 for details). Such MGGAs
to other methods to get reliable results for the bandgap. Hybrid that are particularly interesting for bandgaps are HLE17125 and
functionals and the GW method102,103 (see Sec. III I 4) provide TASK.126
much more accurate values; however, they are also significantly Figure 3 shows results for the bandgap of 76 solids, which
more expensive than semilocal methods and cannot be applied eas- we considered in our previous works.64,80 Compared to the stan-
ily to very large systems. Therefore, fast semilocal methods have dard PBE functional, the results are much improved when TB-mBJ,
been proposed specifically intended for bandgap calculations, and GLLB-SC, or HSE06 is used, since the mean absolute error (MAE)
those which are available in WIEN2k are discussed below. Note drops from 1.99 eV with PBE to 0.47 eV, 0.64 eV, or 0.82 eV, respec-
that a more detailed discussion of the DFT methods for bandgaps tively, for the other methods. Among all methods considered in
is provided in Ref. 104. Refs. 64 and 80, TB-mBJ leads not only to the smallest MAE but also
The Tran–Blaha modified Becke–Johnson (TB-mBJ) poten- to a slope (b = 0.97) of the linear fit that is closest to 1 (the AK13120
tial105 consists of a modified version of the BJ potential106 for functional also leads to a slope of 0.97).
exchange (which reproduces the exact KS potential of atoms very Finally, we also mention that the Slater127 and Krieger–
well106,107 ) and LDA108 for correlation. The exchange part, which is Li–Iafrate128 potentials have been implemented in the WIEN2k
a MGGA since it depends on the kinetic-energy density t, is code.129,130 However, these ab initio potentials, which are as expen-
√ ¿ sive as the HF/hybrid methods, are not really intended for bandgap
mBJ BR 1 5ÁÀ t(r) ,
Á calculations, but may be interesting for more fundamental studies or
vx (r) = cvx (r) + (3c − 2) (15) as a better starting point for approximating the exact KS exchange
π 6 ρ(r)
potential.
where vBR
x is the Becke–Roussel (BR) potential
109
and
5. Methods for van der Waals systems
c = α + βg p (16) The semilocal and hybrid functionals are, in general, quite inac-
curate for describing weak interactions.131,132 This is mainly due
with to the London dispersion forces that are not included properly
1 ∣∇ρ(r′ )∣ 3 ′ in these approximations. Nevertheless, much better results can be
g= d r (17) obtained by adding to the semilocal/hybrid functional a correla-
Vcell ∫ ρ(r′ )
cell tion term (Ec,disp ) accounting for the dispersion forces. There are

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-6


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

FIG. 3. Bandgaps of 76 solids calculated with the PBE, TB-mBJ, GLLB-SC, and HSE06 methods compared to experiment. The MAE compared to experiment, a linear fit
(y = a + bx, dashed lines) of the data, and the corresponding root mean square deviation (RMSD) are also shown. The calculated results are from Refs. 64 and 80.

essentially two types of dispersion corrections. The first one is of the 1


Ec,NLdisp = ρ(r)Φ(r, r′ )ρ(r′ )d3 rd3 r′ , (20)
atom-pairwise (at-pw) type, 2∬
where the kernel Φ is a function of ρ, ∇ρ, and the interelectronic
at-pw damp CnAB distance ∣r − r′ ∣. Compared to the single integral for semilocal func-
Ec, disp = − ∑ ∑ fn (RAB ) , (19)
A<B n=6,8,10,... RnAB tionals [Eq. (12)], evaluating such a double integral is clearly more
involved computationally. In order to make the calculations afford-
where CnAB are the dispersion coefficients for atom pair A–B sepa- able, Román-Pérez and Soler138 proposed a very efficient method
damp based on fast Fourier transformation (FFT) for evaluating Eq. (20).
rated by the distance RAB and fn is a damping function. Among
Their method is used for the implementation of NL-vdW functionals
the at-pw dispersion methods, Grimme’s DFT-D2133 and DFT-
in WIEN2k. However, to apply the FFT-based method efficiently in
D3134 can be used with WIEN2k through the separate software DFT-
an all-electron method, it is necessary to make ρ smoother by remov-
D3135 that supports periodic boundary conditions.136 A feature of
ing the high-density region close to the nucleus. This is done with the
the DFT-D3 method is that the dispersion coefficients are not pre-
following formula:
computed and fixed but depend on the coordination number of the
system. ⎧
⎪ρ(r),
⎪ ρ(r) ⩽ ρc
The second type of dispersion methods available in WIEN2k ρs (r) = ⎨ ρ(r)+Aρc (ρ(r)−ρc ) (21)
⎩ 1+A(ρ(r)−ρc ) , ρ(r) > ρc ,

are the nonlocal van der Waals (NL-vdW) functionals137 ⎪

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-7


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

t orb t orb
where A = 1 bohr3 and ρc is the density cutoff that determines how (D(ρn , vorb
n ), Fn (ρn , Rn , vn )) = G(ρn , Rn , vn ) = 0. (24)
smooth the new density ρs should be. More details can be found in
Ref. 139, where it is shown that converged benchmark results can This is equivalent to solving the SCF problem for an extended
easily be obtained at a relatively modest cost. KS-equation, finding the variational minimum of both the atomic
In our previous works,39,140 a plethora of DFT-D3 and NL-vdW positions and densities.
functionals were assessed on solids. The test set consists of strongly The Hellman–Feynman forces due to the input density ρn are
bound solids and van der Waals solids such as rare gases or layered calculated within lapw0, while the Pulay corrections are calculated
compounds. The results showed that among the at-pw methods, in lapw2 using the new density KS(ρn , Rtn , vorb
n ), so the forces above
PBE-D3/D3(BJ)134,141 seems to be a pretty good choice, while among are, in general, not true derivatives of the energy, rather pseudo-
the nonlocal methods, rev-vdW-DF2142 is the most balanced and forces that converge to them as the density converges. The general
actually more or less the best among all tested functionals. method is to expand to first order, i.e., write for the next value of the
variables,

C. SCF convergence, total energies, forces, (ρn+1 , Rtn+1 , vorb t orb t orb
n+1 ) = (ρn , Rn , vn ) + Hn G(ρn , Rn , vn ), (25)
and structure optimization
where Hn is an approximation to the inverse Jacobian, which is con-
The total energy Etot of a periodic solid (with frozen nuclear structed as a Simplex gradient, i.e., a multi-dimensional numerical
positions) is given by Eq. (9). The individual terms are of opposite derivative using some number of the prior steps. The approximate
sign and, in an all-electron method, very large. In order to cancel the inverse Jacobian has two components:
Coulomb singularity, we follow the algorithm of Weinert et al.,143
where the kinetic and potential-energy terms are combined and a 1. A predicted component where the density, positions, and other
numerically stable method is obtained. variables have changed in a way that maps onto the prior steps
The force exerted on an atom t residing at position Rt , in the SCF iterations, so some information is already available.
defined as 2. An unpredicted component where the changes in the variables
t are new, so no prior information is available.
F = −∇Rt Etot , (22)
is calculated from the Hellman–Feynman theorem and includes Equation (25) ignores higher-order terms in the expansion,
Pulay corrections,144 thereby taking into account that parts of the which can break down far from the fixed point. In addition, because
basis set used in WIEN2k depend upon the position of atoms (see it is generated by a type of numerical differentiation, it can have
Refs. 17 and 145–147 for the derivation specific to APW-based limited accuracy if the step sizes are inappropriate. In addition, the
methods). unpredicted component can lead to instabilities. The approach taken
WIEN2k exploits the self-consistency of the KS equations, run- is to control the algorithm greed for the unpredicted step and also
ning through a sequence of calculations where the target is a density use trust regions for both the predicted and unpredicted steps. The
(and other parameters), which when passed through these SCF cal- general idea of trust regions is that Eq. (25) is only reasonably accu-
culations yield the same density. This is equivalent to finding the rate for changes of the variables, which are smaller than some value,
solution to a set of simultaneous equations and is, in the most gen- which is called the trust radius, as illustrated in Fig. 4. Starting from
eral case, referred to mathematically as a “fixed-point problem,” some initial defaults, at each iteration, the algorithm will check to
although it goes under the different name of “mixing” in the DFT see if the step the mixer proposes to use is small enough; if it is
literature. The general method used for just the electron density ρ too large, then the step is reduced. In the next iteration, if the step
and other density-like variables (for instance, the density matrix) is used led to an adequate improvement when bounded by the trust
discussed in Ref. 148, while the extension to include atomic positions
is described in Ref. 149. At any given iteration n, the relevant vari-
ables can be written as a vector (ρn , vorb
n ), including the density ρ at
the start of an iteration as a function of the Cartesian coordinates r
and an orbital potential vorb (if used as in DFT+U or on-site hybrids,
see Sec. II B 3), as well as other relevant variables. After running
through the SCF sequence, a new density [symbolized by the SCF
mapping KS(ρn , vorbn )] is produced. The fixed-point for just the den-
sity is when the two are equal, i.e., for all of the variables, the set of
simultaneous equations

KS(ρn , vorb orb orb


n ) − (ρn , vn ) = D(ρn , vn ) = 0, (23)
FIG. 4. Illustration of a trust-region approach. With contours shown dashed, from
where D(ρn , vorb
n ) is the density residual. Since the total energy the initial point, the best route following the gradient of the contours is shown
with respect to the position of the atoms also has to be minimal, in red. However, only a linear step shown in blue is predicted by the multise-
when these are allowed to vary, this can be expanded to include the cant expansion. The trust region (brown) limits the step along this direction so it
makes adequate progress downhill and not too far, which is less efficient and can
forces Fn (ρn , Rtn , vorb t orb
n ) = −∇Rtn E(ρn , Rn , vn ), i.e., solve the larger diverge.
problem

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-8


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

radius, then this radius is increased; if the step was not good, then For the optimization of lattice parameters, WIEN2k offers a
the trust radius is decreased. In addition to changing the trust radii couple of workflows and utilities to generate structures with differ-
based upon improvement (or not), the most recent algorithm150 also ent lattice parameters, running the corresponding SCF calculations
looks at the last step to see how large it should have been for both the and analyzing the results. The optimized lattice parameters, how-
predicted and unpredicted parts. This approach is significantly more ever, are found only from the lowest total energy since there is no
stable and often leads to much smaller steps than earlier versions of stress tensor in WIEN2k yet. This makes the optimization tedious
the mixer used. for low symmetry cases and practically impossible for triclinic
By design, the algorithm requires minimal user input beyond lattices.
an estimate for the initial step to take for the unpredicted step; the
most recent version of the algorithm150 automatically controls all
the internal parameters. No algorithm is perfect, and the conver- D. User interface and utilities
gence of the mixing depends significantly upon the nature of the WIEN2k consists of a large set of individual programs (mostly
physical problem being considered. The better the description of the written in Fortran 90), which are linked together via tcsh-shell scripts
underlying quantum mechanical problem, primarily the XC poten- representing a particular workflow. With this modular structure,
tial, the more rapidly it will converge. A very badly posed problem, WIEN2k is, on the one hand, very flexible and one can run a ded-
in contrast, may converge only very slowly or not at all. icated program for a particular task. On the other hand, there is
Two main algorithms are used within WIEN2k. The first is not just one program and the specific task will be determined by
MSEC3, which is an updated version of a multisecant Broyden directives in the input file, but a user has to know which program
method148 with trust region controls. This is a conservative algo- performs this specific task.
rithm, which uses the least greedy approach at every iteration. It is WIEN2k can be driven either from the command line or using
recommended for problems that converge badly. The more pushy a web-based graphical user interface (GUI), called w2web, which can
MSR1149 uses a more aggressive algorithm, which is significantly bet- be accessed by any web browser. Most likely, an experienced user will
ter for problems with soft modes that may converge only very slowly use the command line and explore all advanced features of WIEN2k,
with MSEC3. but for the beginner, the web-based GUI provides a very good start-
One unique feature of WIEN2k is that it can simultaneously ing point and it also teaches the user the corresponding command
converge both the density and atomic positions by solving the fixed- line.
point problem of Eq. (24) using a multisecant approach.149 This
is often considerably faster than converging them independently
as done in many other DFT codes and different from molecular 1. Structure generation
dynamics approaches such as Car–Parrinello.151 The convergence The first task of every calculation is to define the structural
rate depends upon the number and width of the eigenvalue clus- data. As an example, the StructGen@w2web page is shown in Fig. 6
ters149 of the combined electron and atomic position Jacobian. This for the case of TiCoSb. The necessary basic input consists of the
approach does not follow the Born–Oppenheimer surface, which following:
is the energy surface when the density is converged, rather some
● the lattice type (P, B, F, and H for primitive, body centered,
other surface, which is a balance between having converged den-
face centered, and hexagonal, respectively) or, if already
sities and pseudo-forces as illustrated in Fig. 5. As such it can
known, one of the 230 space groups; for the Heusler com-
be somewhat confusing to the user, particularly as the pseudo-
pound TiCoSb, we can select F lattice or space group 216
forces can vary in a strange fashion. This mode can be used with
(F43m);
both the more conservative MSEC3 and the more aggressive MSR1
● the lattice parameters a, b, and c (in bohr or Å) and the
algorithm.
angles α, β, and γ (in degree);
● the atoms and their positions; if the space group is given,
only one of the equivalent atoms has to be specified.
When the new structure is saved, the setrmt utility determines the
nearest neighbor distances and automatically sets optimized atomic
sphere sizes RMT for this structure. The choice of RMT has nothing to
do with ionic radii but depends on the convergence properties of the
atoms as discussed before (see the end of Sec. II A). It is important
to note that if one wants to compare total energies for a series of
calculations (e.g., for volume optimization), the RMT should be kept
constant.
An alternative on the command line is the makestruct utility,
which works analogous to StructGen@w2web. More complex struc-
tures can be converted from cif or xyz files using the cif2struct or
FIG. 5. The combined density and position algorithm does not follow the Born– xyz2struct utilities. The generated structures can be conveniently
Oppenheimer surface (indigo) where the density is converged or the surface where visualized using XCrysDen152 or VESTA.153
the pseudo-forces are zero (orange), instead it finds a fixed point of a combination
Starting from a basic (simple) structure, WIEN2k has powerful
(green contours) following the red path.
tools to generate supercells and manipulate them. supercell generates

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-9


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

FIG. 6. Screenshot of the StructGen@w2web page of the w2web GUI of WIEN2k for TiCoSb.

quickly h × k × l supercells (with/without B- or F-centering so that 2. Input generation


the supercell size can be increased by factors of two) and can add
vacuum for surface slab generation. The structeditor, a collection of As mentioned above, WIEN2k consists of many individual pro-
GNU Octave scripts, is √even√more powerful, since it can create arbi- grams and most of them have their own input file. Although this
trary supercells (e.g., 3 × 3 × l), rotate or merge structures, and sounds very tedious at first, there are default inputs for all programs
delete or add atoms. and several tools for changing the most important parameters on

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-10


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

the fly. In w2web, the next step would be to check the symmetry of number of k-points (typical starting values for the SCF cycle would
the newly generated structure and generate the input files for the be a 10 × 10 × 10 mesh), while large cells (100 atoms) and insulators
SCF calculation (initialize@w2web). The user can provide a cou- can be started with only one k-point. In any case, after the first SCF
ple of parameters (only needed if one wants to change the defaults, cycle, the k-mesh should be increased and the results (e.g., the forces
see below) and run the following steps in batch mode or step by on the atoms) should be checked. Certain properties (DOS, optics,
step: and NMR) may need an even denser mesh, which in WIEN2k can be
easily done and is fairly cheap, since it is only used for the property
● nn: Determines the distances between all atoms up to twice
of interest.
the nearest neighbor distance. In addition, it checks for over-
lapping spheres and will issue an error message if the spheres
overlap. It also checks if identical elements have the same 3. SCF cycle
environment and eventually regroups them into equivalent The SCF cycle consists in WIEN2k of a complex workflow using
sets. several different programs. The main steps are as follows:
● sgroup: Checks the structure and determines the space-
● lapw0: Calculates the Coulomb and XC potential from the
group. It will group the atoms into sets of equivalent ones
density.
according to the Wyckoff positions of the corresponding
● lapw1: Calculates the valence and semi-core eigenvalues and
space-group. In addition, it will check and determine the
eigenvectors at all requested k-points
smallest possible (primitive) cell and create the correspond-
● lapw2: Calculates the valence electron density
ing structure file if necessary. For instance, if one enters the
● lcore: Calculates the core eigenvalues and the core density
NaCl structure as a primitive cubic structure with four Na
● mixer: Adds up the core and valence densities and mixes the
and four Cl atoms, it will automatically create a primitive
total density with densities from previous iterations. In addi-
FCC cell with only one Na and Cl atom.
tion, it may update the atomic positions according to the cal-
● symmetry: Finds the symmetry operations of the space group
culated forces (see Sec. II C) and also the density matrices or
as well as the point group symmetry of each atom and
orbital potentials when DFT+U or on-site hybrid methods
the corresponding LM expansion for the density/potential
are used.
[Eqs. (1) and (2)].
● lstart: Solves numerically the radial Dirac equation for free Additional programs may be called depending on the requested
atoms and creates atomic densities. Using the eigenvalues options to include SO coupling or one of the specialized func-
(or the localization within the atomic spheres) of all atomic tionals discussed in Sec. II B (DFT-D3, NL-vdW, DFT+U, on-site
states, it groups them into core and valence states. It selects hybrid/EECE, or hybrid-DFT/HF).
automatically LOs for semi-core states and writes the start- In w2web, the SCF cycle can be started by clicking on
ing energy parameters Eℓ to case.in1 (during the SCF cycle, SCF@w2web. In this interface, one can then specify several parame-
they are searched and adapted automatically to ensure best ters such as convergence criteria, parallelization, simultaneous opti-
possible settings in all cases). mization of internal atomic positions, or adding SO coupling.
● kgen: Generates a shifted or non-shifted equidistant k-mesh The most important parameters for the corresponding com-
with a user specified density in the irreducible part of the BZ. mand line script are as follows:
● dstart: Superposes the atomic densities and creates the start- run_lapw [-ec 0.0001 -cc 0.0001 -fc 1.0 -p -so -min]
ing density for the SCF cycle. The SCF cycle will stop when the (optional) convergence crite-
ria -ec (energy in Ry), -cc (charge in e− ), and -fc (forces in mRy/bohr)
On the command line, a corresponding script is called, which are fulfilled three times in a row. SO coupling (only possible after a
optionally allows us to specify various parameters (the most impor- previous init_so_lapw step) is switched on using -so, -min relaxes the
tant ones are given below with their default values for reference): atomic positions simultaneously with the electron density (Sec. II C),
init_lapw [-b -vxc PBE -ecut -6.0 -rkmax 7.0 -numk 1000] and -p switches on parallelization (Sec. II E).
The switches are described as follows: -b indicates batch mode The basic summary of the SCF cycle is written into the
(instead of step by step), -vxc selects the DFT functional, -ecut gives case.scf file and all relevant quantities are labeled :LABEL: and can
the core–valence separation energy (in Ry), -rkmax determines the be searched/monitored using analyse@Utils@w2web or the Linux
plane wave cut-off parameter Rmin MT Kmax , and -numk determines the grep command. If the desired convergence has been reached, it
total number of k-points in the full BZ. is advisable to save all relevant input/output files using either
The most critical parameter is Rmin
MT Kmax , which determines not save@Utils@w2web or the save_lapw utility. One can now either
only the quality but also the required computing time. The type of
check results using more k-points (kgen@single_prog@Execution@
atom with the smallest RMT determines this value because the RMT
w2web) or modify other inputs (input files@Files@w2web) such as
for other atoms are set such that when the smallest atom is converged
Rmin
MT Kmax or the XC functional and then continue with the SCF cycle.
with the number of PWs, all others are also converged with the num-
Later on, it is always possible to come back to a previously saved
ber of PWs. If the smallest sphere is a H atom (for instance, in short
calculation using restore@Utils@w2web (restore_lapw).
O–H bonds), Rmin MT Kmax = 3 is sufficient, most sp/d/f -elements con-
verge with Rmin
MT Kmax = 7/8/9. For lower (higher) precision, one can
decrease (increase) these values by 10%–20%. 4. Tasks
Of similar importance is the selection of a k-mesh. Gener- Once these steps have been finished, one could, for instance,
ally speaking, small unit cells and metallic character require a large optimize the lattice parameters (optimize@w2web) or perform

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-11


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

various other tasks (Tasks@w2web) such as Bandstructure@Tasks@ “real” instead of the “complex” version of the code, thus saving half
w2web, DOS@Tasks@w2web, ElectronDensities@Tasks@w2web, of the memory and running almost four times as fast. In addition,
XSPEC@Tasks@w2web, TELNES@Tasks@w2web, or OPTIC@ it is highly optimized and efficiently parallelized at three different
Tasks@w2web (see Fig. 6). Each of these tasks consists of a guided levels, which can be optimally chosen depending on the size of the
workflow and let the user prepare the necessary inputs, executes problem and the available hardware.
various small programs, and visualizes the results. Except for OpenMP parallelization (see below), parallelization
is activated by a -p switch in our scripts and needs a .machines file
E. Software requirements and parallelization as listed and described below:
1. Software # .machines file for parallelization
# OpenMP parallelization:
WIEN2k runs on any Linux platform and also on Mac. It is omp_global:4
written mainly in Fortran 90 (a few programs are written in C), omp_lapw0:16
and the workflows are managed by tcsh scripts. Most of the time- ...
critical parts use libraries such as BLAS and LAPACK, and efficient # k-point parallelization (speed:hostname):
libraries are therefore mandatory. There is direct installation support 1:host1
for the standard Linux tools GFortran+OpenBLAS (at least gcc 6.x) 1:host2
and Intel IFORT+MKL. The latter still gives the best performance. ....
w2web is a specialized web server written in perl and listens on a # MPI parallelization:
user-defined high port. Its access is, of course, password protected 1:host1:16 host2:16 ...
and can be limited to specific IP addresses. ....
For the optional installation of the MPI-parallel version (useful lapw0:host1:16 host2:16 ...
only on clusters with InfiniBand network or larger shared mem-
ory workstations with at least 16 cores), one needs obviously MPI ● OpenMP parallelization:
(e.g., Open MPI or Intel MPI) and also ScaLAPACK (included The main (time consuming) programs are all parallelized
in the MKL), FFTW,154 and, optional but highly recommended, using OpenMP and can use the corresponding threaded
ELPA.155,156 BLAS, LAPACK, and FFTW libraries. It is activated by
The following Linux tools are necessary (not all of them are either setting the OMP_NUM_THREADS variable globally
always installed by default): tcsh, Perl 5, Ghostscript, gnuplot, GNU or using omp_prog:N directives in .machines. While many
Octave, and Python 2.7.x+NumPy. parts of the code scale very well with the number of parallel
Optional, but highly recommended, programs for certain tasks threads on a multi-core shared memory machine, unfortu-
include the following: nately, the scaling of the matrix diagonalization is, at present,
limited to 2–4 cores due to performance bottlenecks in the
● XCrysDen152 and VESTA153 for structure and electron den-
corresponding OpenBLAS or MKL libraries.
sity visualization but also generation of band structure k-
● k-point parallelization:
meshes or plotting Fermi-surfaces.
This together with OpenMP is a very simple and highly
● Libxc33,34 for XC functionals not directly implemented in
efficient parallelization, which works even on a loosely
WIEN2k.
coupled cluster of simple PCs with a slow network for
● DFT-D3135 for DFT+D3 calculations of van der Waals sys-
small to medium sized cases (up to 100 atoms/cell) where
tems.
the eigenvalue problem needs to be solved for several
● Wannier90157,158 for constructing Wannier functions using
k-points. It requires a common (NFS) filesystem on all
the wien2wannier utility.
machines and password-less ssh (private/public keys). N
● phonopy,159,160 Phonon,161,162 or PHON163,164 for phonon
lines speed:hostname in .machines will split the list of k-
calculations.
points into N junks, and N jobs will be started in parallel
● BoltzTraP2165 for transport calculations (see Sec. III I 1).
on the corresponding hosts, followed by a summation step
● fold2bloch166 to fold supercell band structures back to the
of the partial densities. Since WIEN2k can use temporary
primitive BZ.
(local) storage for the eigenvectors, we are not limited in
● SKEAF167 to extract de Haas-van Alphen frequencies from
the number of k-points and our personal record is a NMR
WIEN2k.
chemical shift calculation for fcc Al with 106 k-points.
● Critic2168,169 is an alternative program to the WIEN2k pro-
● MPI-parallelization:
gram aim to analyze 3D scalar fields such as the electron
With a sufficiently powerful hardware (at least 16 cores
density by using the “atoms in molecules” (AIM) theory of
or a cluster with InfiniBand network) and for medium to
Bader.170,171
large sized problems (more than 50 atoms/cell), it is possi-
ble, and actually necessary, to parallelize further using MPI.
2. Parallelization Besides a tremendous speedup that can be achieved by par-
WIEN2k is a highly accurate all-electron code based on the allelization over atoms and, in particular, over basis func-
APW method and thus certainly not as fast as some other (pseu- tions, this version will distribute the necessary memory on
dopotential or minimal basis set) codes. However, it takes advantage all requested computers, thus allowing calculations for unit
of inversion symmetry and when present it will automatically use the cells with more than 1000 atoms.172–174 Such cells require

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-12


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

basis sets of about 105 APWs, and the resulting Hamil- (crystal field splitting), while for lower symmetry, a splitting into five
tonian, overlap, and eigenvector matrices may need about different d-orbitals is obtained. A review paper8 (Sec. 6.2) illustrates
500 GB of memory, which are distributed over the nodes these tools for TiC, a refractory metal (crystallizing in the sodium
in the standard ScaLAPACK block-cyclic distribution. For chloride structure) that is almost as hard as diamond but has metal-
such large systems, the solution of the general eigenvalue lic, covalent, and ionic bonding contributions. These data are very
problem can become the time-limiting (N 3 ) step (depend- useful to analyze the electronic structure, and we illustrate this for
ing on the number, type, and RMT of the atoms, the setup the band structure of the Heusler compound TiCoSb (see Ref. 177).
of the complicated matrix elements can take a compara- If one wants to know which atomic states (e.g., Co-d, Ti-d, or Sb-p
ble fraction of the total time), but the ELPA library pro- states) contribute most to a certain band, one can show the character
vides a highly efficient and scalable (1000 cores) diagonal- of the bands, sometimes called “fat bands.” For each eigenvalue εnk ,
ization. In cases with fewer atoms in large cells (isolated the size of the circle represents the weight of the chosen character
molecules in a large box or surface slabs with sufficient vac- (e.g., a particular qtℓm ). Figure 7 shows which band states originate
uum), we can use an iterative diagonalization175 using the mainly from Co-d, Ti-d, or Sb-p states, giving the band structure a
previous eigenvectors as start. Depending on the requested chemical interpretation.
number of eigenvalues, this method may be up to 100 times From the KS eigenvalues calculated on a sufficiently fine k-grid
faster than full diagonalization and still scales very well with in the irreducible BZ, one can obtain the density of states (DOS),
the number of cores. The MPI version of the code is used usually by means of the (modified) tetrahedron method.178 By using
when lines with speed:hostname:N (or lines with more than the partial charges [Eq. (26)], one can decompose the total DOS
one hostname) in .machines are specified. Of course, cou- into partial DOS (PDOS), which are useful for understanding chem-
pling of k- and MPI-parallelization (and/or OpenMP) is ical bonding and interpreting various spectroscopic data. Figure 8(a)
possible. shows how much each region (atomic spheres of Co, Ti, Sb, and
the interstitial) contributes to the total DOS. We show the domi-
nating valence contributions from the Co-3d and Ti-3d electrons in
III. PROPERTIES AND FEATURES Fig. 8(b) and those from Sb-5s/5p in Fig. 8(c). Unfortunately, the
A. Energy bands, density of states, electron densities interstitial PDOS cannot be decomposed into atomic and ℓ-like con-
tributions uniquely. However, by analyzing the atomic orbitals in the
Once a self-consistent solution for a chosen atomic structure is free atom, we see (Table I) that only a fraction of the related electron
done, one can focus on the electronic structure. The energy eigen- density resides inside the corresponding atomic sphere, e.g., 81% for
values as a function of the k-vector obtained at the end of the KS the Co-3d but only 15% for the Co-4s orbital. Therefore, a signifi-
calculation define the band structure. The k-path along high sym- cant part of the density lies outside the atomic sphere, leading to a
metry lines in the irreducible BZ (see the Bilbao Crystallographic
Server176 ) can be either obtained from WIEN2k default templates
or generated graphically using XCrysDen.152 In WIEN2k, one can
plot the energy bands by using the program spaghetti, indicating
that its interpretation is difficult. However, there are some tools to
help. A first tool is a symmetry analysis, which determines the irre-
ducible representation (of the corresponding point group) for each
KS eigenvalue. With this knowledge, one can connect the KS eigen-
states to bands by using compatibility relations and satisfying the
non-crossing rule. The chemical bonding information of state nk is
contained in the corresponding wave function ψ nk , which is complex
and three dimensional. However, when computing the square of its
modulus, one obtains an electron density, which is a real function
and easy to visualize. Integrating this (normalized) electron density,
one obtains a charge q, which can be decomposed into contribu-
tions from the interstitial region I and the atomic spheres St (labeled
by the atom number t and the quantum number ℓ according to the
atomic-like basis set),

1 = ∑ qtℓ + qI . (26)
t,ℓ

This allows us to compress the detailed information contained


in the wave function of a single eigenstate state ψ nk to a few numbers
that can be stored and analyzed. In addition, WIEN2k decomposes
the qtℓ according to the symmetry of the corresponding point group. FIG. 7. Band structure of TiCoSb with emphasis on Ti-d (blue), Co-d (red), and
Sb-p (black). The size of the circles in this fat band plot is proportional to the
For example, the five d-orbitals of a TM atom surrounded by lig-
corresponding partial charge.
ands in octahedral symmetry are split into the t 2g and eg manifold

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-13


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

DOS = ∑ PDOStℓ + PDOSI = ∑ PDOStℓ /qren


tℓ . (27)
t,ℓ t,ℓ

This sum runs only over the “chemical” ℓ, which are the main con-
tributions. In Table I, we see that the qren tℓ are close to the free
atom situation for more localized orbitals but differs significantly,
for example, for Co-4s, which is more localized in the solid than in
the free atom. For the importance of this effect, see Sec. III B.
The fundamental variable in DFT is the electron density ρ,
which can be compared to experimental data. The total ρ, which is
obtained by summing over all occupied states, can be decomposed
into its contributions coming from the core, semi-core, and valence
states. A variety of tools (such as XCrysDen152 or VESTA153 ) allows
one to visualize the density ρ along a line, in a plane (2D), or in the
unit cell (3D). One can easily compute the density corresponding
to a selected energy window of electronic states in order to visu-
alize their bonding character. By taking the difference between the
crystalline density and a superposition of atomic densities (placed at
the atomic position of the crystal), one obtains a difference density
Δρ = ρcrystal − ρatoms , which shows chemical bonding effects much
more clearly than the total or valence density. Figure 9 provides an
illustration of Δρ for TiCoSb within the (110) plane, where we can
observe the strong asphericities in the electron density around the Ti
and Co atoms originating from different occupations of the five 3d
orbitals, as well as the charge transfer (discussed below using Bader
charges).
When we want to compare the computed electron densities or
the related X-ray structure factors (computed using the lapw3 mod-
ule) with experimental data, we must take into account the motion
of the nuclei. In DFT calculations, we assume that the nuclei are at
rest, whereas in an experiment, this motion must be considered, for
FIG. 8. Total, partial, and renormalized partial DOS of TiCoSb. (a) Total DOS example, by means of the Debye–Waller factors, which can also be
decomposed into atoms and interstitial. (b) Ti-d and Co-d (renormalized) PDOS. calculated by phonon calculations.
(c) Sb-s, p (renormalized) PDOS. (d) Ti-s, p (renormalized) PDOS. (e) Co-s, p
(renormalized) PDOS.

non-negligible PDOS from the interstitial region [Fig. 8(a)]. A sim-


ple scheme to eliminate this interstitial part is to renormalize the
partial DOS with a factor qren
tℓ (determined by a least squares fit) such
that the sum of the renormalized PDOS contributions yields the total
DOS,

TABLE I. Fraction qfree


tℓ of the charge density of atomic orbitals (atom t and momen-
tum ℓ) that resides inside the corresponding atomic sphere of the free atom and the
renormalized charge qren
tℓ in the solid.

qfree
tℓ qren
tℓ

Co-4s 0.34 0.48


Co-4p ... 0.45
Co-3d 0.95 0.90
Ti-4s 0.15 0.30
Ti-4p ... 0.43
Ti-3d 0.81 0.86
Sb-5s 0.60 0.61
FIG. 9. Difference density Δρ = ρcrystal − ρatoms (e− /bohr3 ) of TiCoSb in the (110)
Sb-5p 0.35 0.40 plane.

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-14


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

It is a strength of theory to allow various decompositions (of B. Photoelectron spectroscopy


the DOS or electron density, for instance), which are often useful
1. Valence-band photoelectron spectroscopy
for interpreting properties, but these may depend on the basis set
used in a calculation, for example, when deriving atomic charges. In Experimental valence band photoelectron spectra (PES) are
a LCAO scheme, one takes the weights of all atomic orbitals cen- often just compared to the total DOS. Such a comparison, how-
tered at a given atom to determine how much charge corresponds ever, can at best reproduce certain peak positions, but usually not
to that atom (Mulliken’s population analysis). In an APW scheme, the experimental intensities. This is even more true with mod-
the charge inside the related atomic sphere would give an atomic ern synchrotron-based hard X-ray PES (HAXPES), where the spec-
charge, but this value clearly depends on the chosen atomic radius tra differ considerably depending on the excitation energy. This is
and lacks the interstitial contribution. However, the renormalized because the cross sections of different atomic orbitals change dra-
partial charges Qren
t obtained from an integral over the renormalized matically as a function of excitation energy and this effect should be
PDOS gives a meaningful measure of charge transfer, as shown in taken into account. The pes module179 of WIEN2k uses the partial
Table II. DOS (PDOStℓ ) and multiplies it with the corresponding energy-
A basis-set independent alternative is the AIM procedure pro- dependent atomic orbital cross sections σ tℓ , taken from various
posed by Bader,170,171 which is based on a topological analysis of tables,180,181
the density. It uniquely defines volumes (called “atomic basins”)
that contain exactly one nucleus by enforcing a zero-flux boundary: I = ∑ PDOStℓ σtℓ . (28)
t,ℓ
∇ρ ⋅ n̂ = 0. Inside such an atomic basin, this scheme uniquely defines
the Bader charge for a given density independently of the basis- In addition, pes can use the renormalized PDOS (see Sec. III A),
set method that was used to calculate the electron density.170,171 An so that the contributions from the less localized orbitals (whose wave
example of the application of the AIM method is given in Table II, functions are mainly in the interstitial region) are also properly taken
which shows the charge inside the atomic basins of TiCoSb deter- into account. This module allows us to specify the X-ray energy and
mined with the aim module of WIEN2k. QBader t,crystal is the nuclear
can handle unpolarized and linearly polarized light as well as lin-
charge Zt minus the number of electrons (a positive value indi- ear dichroism in angular distribution (LDAD). It was successfully
cates a depletion of electrons) using the SCF density, and QBader applied for various examples179 (SiO2 , PbO2 , CeVO4 , In2 O3 , and
t,super is
the same quantity but using a density from a superposition of the ZnO).
free neutral atoms. In this crystal structure, according to a Bader Here, we compare in Fig. 10 the experimental HAXPES spec-
analysis, even the superposition of neutral densities leads to a sig- trum177 of TiCoSb at 6 keV with the theoretical calculation. The
nificant charge transfer from Ti to Co and Sb, which is enhanced theoretical spectrum reproduces the experimental intensities very
for Co and Ti but reduced for Sb during the SCF cycle. These well, but the bandwidth is too small so that the Sb-s peak has about
Bader charges can be compared to the Qren 1 eV less binding energy. This is a well-known DFT problem and
t . As we can see, in both
methods, there is a transfer of electrons from the Ti to the Co concerns all states at lower energy. The decomposition of the total
atom, but the specific amount and, in particular, the charge state spectrum allows us to analyze the contributions to the different
of Sb differs significantly depending on the way it is calculated. peaks in the spectrum. The low energy feature is almost exclusively
An inspection of the difference density (Fig. 9) shows a negative from Sb-s states, the double peak at −5 eV is from Sb-p states, and
Δρ around Sb and thus indicates a positive charge of the Sb atom the double peak at −2 eV is from Co-d states. However, Co-s also
in contrast to the Bader charges, which seems to pick up a lot of contributes significantly to the feature at −6 eV, and Sb-p contribu-
charges in the interstitial region leading to a negative Sb charge. tions are even larger than Co-3d for the lowest binding energy peak
In essence, one should be careful with quantitative charge state
assignments.
An alternative to aim is the Critic2 package,168,169 which deter-
mines Bader charges using a pre-calculated 3D mesh of densities. It
is very fast; however, the integration of total charges on such a crude
mesh is inaccurate and one should restrict its usage for magnetic
moments (integrating the spin densities) or valence charges densities
(be careful with the 3D mesh).

TABLE II. Bader charges QBader Bader


t,crystal and Qt,super using the SCF and the free-atom
superposed density, respectively, and renormalized atomic charges Qren
t of TiCoSb.

Atom QBader
t,crystal QBader
t,super Qren
t

Co −0.89 −0.18 −1.25


FIG. 10. Experimental177 and theoretical PES of TiCoSb at 6 keV. The theoreti-
Ti +1.28 +0.82 +0.85
cal spectrum is further decomposed into its main contributions Sb-s, p and Co-s,
Sb −0.39 −0.64 +0.40 p, d.

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-15


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

at −1.5 eV. All Ti contributions are very small and thus not shown in The dipole selection rule is valid when the X-ray energy is not
Fig. 10. too large and limits transitions between a core state on atom X and
angular momentum ℓ into/from a conduction/valence band state
2. Core-level photoelectron spectroscopy with Δℓ ± 1 on the same atom. In essence, the spectrum is calcu-
X-ray photoelectron spectroscopy (XPS) determines the bind- lated from the corresponding partial DOS times the squared radial
ing energy (BE) of core states. These BEs are specific to certain matrix element. In the case of polarized light and oriented samples,
atoms, but the possible small changes of BEs (core level shifts) pro- the orientation-dependent spectra can be obtained by substituting
vide important additional information about the chemical environ- the ℓ-like partial DOS by an appropriate ℓm-like DOS, e.g., for a
ment and, in particular, the oxidation state of that element. Since K-spectrum of a tetragonal/hexagonal system by replacing the total
WIEN2k is an all-electron method, it has the self-consistent core p-DOS by pz and px + py -DOS.189
eigenvalues available and one could calculate their BE as energy Such a scheme leads to very good results for XES spectra, where
difference with respect to the Fermi level. However, according to the final state has a filled core-hole and the valence-hole is usu-
Janak’s theorem,182 DFT eigenvalues represent the partial deriva- ally well screened. For XANES and EELS spectra, however, the final
tive of the total energy with respect to the orbital occupancy and state190 determines the spectrum. The final state has a core hole
are therefore not necessarily good approximations of experimen- and an excited electron in the conduction band, and they will inter-
tal excitation energies. Such BEs are typically 10%–20% too small. act with each other leading to strong excitonic effects. In order to
In fact, even BE differences (core level shifts) from ground-state describe this effect in a DFT-based band-structure code, one has to
calculations might not be reliable because screening effects of the create a supercell (as large as possible, but depending on the spe-
final state are not included. Much better approximations to exper- cific system and the hardware resources) of about 32–256 atoms and
imental BEs can be obtained according to Slater’s transition state remove a core electron from one of the atoms. This electron should
theory,183 where half of a core electron is removed.173,184,185 The cor- be added either to the valence electrons (if there are proper states
responding SCF eigenvalue, which represents the slope of the total in the conduction band, e.g., in B–K edges of BN) or to the con-
energy vs occupation at half occupation, is a much better approx- stant background charge (if the lowest conduction band states are of
imation to the actual energy difference for n and n − 1 occupa- completely wrong character, e.g., for the O–K edge of a TM-oxide) to
tion, and typical BE errors are reduced to a few percent. For solids, keep the system neutral. In the following SCF cycle, the valence states
such calculations should employ large supercells where only one of on the atom with the core hole will get a lower energy and localize,
the atoms gets excited. This method also allows for some possible but the surrounding electrons are allowed to partially contribute to
screening due to the valence electrons but suffers from the fact that, the screening.
in solids, a neutral unit cell is required. The missing half electron As an example, the Cs-L3 spectrum of CsK2 Sb is shown in
can be compensated by adding a negative background charge, by Fig. 11(a) and compared with state-of-the-art calculations191 using
increasing the number of valence electrons by one half, or by play- the Bethe–Salpeter equation (BSE) (for further details on BSE, see
ing slightly with the nuclear charge186 according to a virtual crys- Sec. III I 3). Obviously, the spectrum calculated with the ground
tal approximation. However, it is not always clear which of these state electronic structure is very different from core hole super-
methods should be preferred. Successful applications include, for cell or BSE calculations, where the spectral weight is redistributed
instance, the N-1s shifts of h-BN covered Pt, Rh, and Ru(111) sur- into the first (excitonic) peak. In particular, the Cs-5d states come
faces (with a unit cell of the “nanomesh” containing more than 1000 down in energy and partially screen the core hole [Fig. 11(b)].
atoms)173 or to Pb-5d and Ta-4f shifts in the misfit layer compound
(PbS)1.14 TaS2 .185
The work function can also be obtained from surface slab calcu-
lations as the difference of the Fermi level and the Coulomb potential
in the middle of the vacuum region. An example can be found for
free and h-BN covered Ni and Rh(111) surfaces in Refs. 187 and
188. One has to carefully check the convergence of the work function
with respect to the size of the vacuum region.

C. X-ray absorption/emission spectroscopy


and electron energy loss spectroscopy
Experimental techniques such as X-ray emission (XES), near-
edge X-ray absorption (XAS, NEXAFS, and XANES), and electron
energy loss spectroscopy (EELS) represent an electronic transition
between a core state and a corresponding valence/conduction band
state, which leads to the measurement of emitted/absorbed X-rays or FIG. 11. (a) Cs-L3 spectrum of CsK2 Sb calculated using the ground state or a core
the energy loss of transmitted electrons. The intensity of such a spec- hole in 32 or 128 atom supercells. These calculations are compared with the BSE
trum is given by Fermi’s golden rule according to dipole transitions results from Ref. 191, which are aligned at the first main peak as it is not possible
between an initial (ΨI ) and a final (ΨF ) state, to calculate such spectra on an absolute energy scale. The spectra are broadened
with a Gaussian of 0.5 eV and a Lorentzian of 0.2 eV. (b) Partial Cs-d DOS for a
Cs atom with and without core hole.
I(E) ∝ ⟨ΨI ∣εr∣ΨF ⟩2 δ(εF − εI − E). (29)

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-16


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

The Cs-6s contributions (not shown) are 1–2 orders of magnitude certain momentum transfer q to yield a differential cross sec-
smaller because both their dipole matrix elements and the 6s-PDOS tion, which can be compared to the experiment. This formal-
are smaller than the corresponding Cs-5d quantities. On the other ism allows the calculation of relativistic EELS including transi-
hand, the BSE and core hole supercell calculations agree quite well. tions of arbitrary order (i.e., non-dipole transitions), and it can
Note that the BSE data have a larger broadening. In this example, the take into account the relative orientation between the sample and
size of the supercell is easy to converge, but it should be noted that beam.199 Practical aspects on how to perform EELS calculations
for details of the spectrum, a rather good k-mesh (8 × 8 × 8 for the have been given by Hébert,200 and some examples can be found in
128 atom supercell) is necessary. Refs. 201–204.
While the core hole approach works generally quite well, it also
has clear limitations or needs extensions:
D. Optics
(i) For metals or small bandgap semiconductors, a full core hole The optics module of WIEN2k uses the independent-particle
is sometimes too much because the static screening in the approximation (IPA) and calculates the direct transitions (conserv-
supercell might not be enough. Better results can be obtained ing k) between occupied nk and unoccupied n′ k states, where for
using a “partial” hole,192,193 although adjusting the size of the both states, KS eigenvalues are used.205 The joint density of states
hole until the resulting spectra match the experiment is not is modified by transition probabilities given by the square of the
fully ab initio anymore. momentum matrix elements M = ⟨n′ k|A⋅p|nk⟩ between these states,
(ii) The used DFT approximation may not be accurate. This can which determine the intensity of optical spectra using dipole selec-
concern the ground state, for instance, the O–K edge of NiO tion rules and clearly distinguish between optically allowed and
could be greatly improved using the TB-mBJ approxima- forbidden transitions. From the resulting imaginary part ε2 of the
tion194 instead of PBE or PBE+U. Even more problematic dielectric function, its real part ε1 can be obtained by the Kramers–
are excited-state effects due to the additional d electron in Kronig transformation and then additional optical functions such
strongly correlated materials (for instance, 3d in TM oxides), as conductivity, reflectivity, absorption, or the loss function can
where very poor L2,3 edges are obtained in single particle also be calculated. In a metallic solid, an additional Drude term
approaches and sophisticated methods such as dynamical accounts for the free-electron intraband contribution. For insula-
mean field theory (DMFT, see Sec. III I 5) or configuration tors and semiconductors, where the DFT gap is often too small
interactions195 are needed. when compared to the experiment, one can use a “scissor oper-
(iii) For early (“d0 ”) TM compounds, the L2,3 edges are influenced ator.” This sounds complicated but is nothing else than a rigid
strongly by interactions and interference effects between the shift of the unoccupied DFT bands to adjust the (too small) DFT
2p1/2 and 2p3/2 states, which are split only by a few eV. This bandgap, either using the experimental gap or, more ab initio, using
can be accounted for using fully relativistic BSE calculations the gap calculated with TB-mBJ. Note that TB-mBJ usually gives very
(see Sec. III I 3), where both the 2p1/2 and 2p3/2 states are good bandgaps, but the bandwidth of both the valence and con-
taken into account simultaneously.196 duction bands are too small, and thus, the optical properties with
(iv) The B-K edge of hexagonal BN has been investigated many TB-mBJ might not be very accurate, but still more accurate than
times in the literature.189,197 While the first strong excitonic standard GGA (see Sec. III B 2 in Ref. 104 for a brief summary
peak originating from antibonding B-pz (π∗ ) states is well of literature results with TB-mBJ). Alternatively, hybrid-DFT func-
described by supercell calculations and in full agreement tionals66 can be used, which give quite good bandgaps for semicon-
with BSE calculations,197 the experimental double peak at ductors, but one should be aware that the optical properties usually
around 7 eV above the π∗ peak originating from σ ∗ (B-px,y ) require a quite dense k-mesh, which makes hybrid calculations fairly
states shows up in the calculations only as a single peak. This expensive.
can only be fixed by taking electron–phonon interactions As an example we present in Fig. 12 the imaginary part of
into account. The approach is based on statistical averages the dielectric function ε2 for CsK2 Sb using various approximations.
over all vibrational eigenmodes of the system.197 Thus, one First, we note that very dense k-meshes are necessary for converged
calculates first the vibrational modes of h-BN and then, using results, which makes the application of more expensive many-body
a supercell of e.g., 128 atoms, all atoms are displaced accord- perturbation methods such as GW even more difficult. As expected,
ing to the vibrational eigenmodes with amplitudes deter- PBE calculations yield the smallest bandgap of 1.06 eV, while hybrid
mined by the Bose–Einstein occupations. Even at T = 0 K, the YS-PBE0 gives 1.68 eV and TB-mBJ 2.08 eV. This can be compared
zero-point motion is enough to split the degenerate px and to G0 W 0 results191 of 1.62 eV or early experimental estimates of
py states, since, on average, each B will have one N neighbor 1.0–1.2 eV, which, however, have been criticized. It should be noted
at a smaller (larger) distance than the two others. This leads that the GW result191 for ε2 probably suffers from an under con-
to the desired splitting of the single σ ∗ peak into a double verged k-mesh and even with their large smearing (note the large
peak.197 tail below 1.62 eV in the GW results shown in Fig. 12), a distinct
peak structure emerges, which is not present in the k-converged
EELS has fairly similar basic principles as XAS but differs results.
slightly because of the finite momentum of the electrons.198 The Experimental results for optical conductivity, reflectivity, or
telnes3 module of WIEN2k calculates the double differential scatter- absorption as well as the low energy valence electron energy
ing cross section on a grid of energy loss values and impulse transfer loss spectrum (VEELS) can often be successfully interpreted in
vectors. This double differential cross section is integrated over a the IPA.203,204 However, sometimes (in particular, for wide gap

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-17


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

shows a Verwey transition. At low temperature, this system has a


charge-ordered state (with Fe2+ and Fe3+ at different Fe-sites), but
above the Verwey transition temperature (at about 309 K), a valence
mixed state with the formal oxidation state Fe2.5+ appears. DFT cal-
culations made it possible to interpret this complicated situation (for
details see Ref. 207 and Sec. 7.4.1 of Ref. 208). Another complex sys-
tem is PrO2 , which has a Jahn–Teller-distorted CaF2 structure.209 It
contains Pr-4f electrons, which form a localized band (lower Hub-
bard band) for one 4f electron, but the others hybridize with the
valence electrons forming a situation between Pr3+ and Pr4+ . This
compound is an antiferromagnetic insulator that requires a relativis-
tic treatment. PBE+U calculations93 provide results that are consis-
tent with all experimental data for the bandgap, magnetic moment,
and structural distortion.
In the 1980s, a numerical problem occurred in connection with
FIG. 12. Imaginary part of the dielectric function ε2 for CsK2 Sb with various func- several studies of the Fe65 Ni35 INVAR alloy, which has a vanish-
tionals. The PBE calculations are presented with a 6 × 6 × 6 and 20 × 20 × 20 ing thermal expansion around room temperature. This is one of the
k-mesh, and all other calculations have used the larger mesh. The GW results are systems for which the magnetization shows a hysteresis when a mag-
from Ref. 191. netic field is applied. The hysteresis causes numerical difficulties,
since for a certain magnetic field, there are three solutions (mag-
netic moments) with very similar total energies causing difficulties
insulators) the frequency-dependent dielectric function ε in the IPA in the convergence of a conventional SCF procedure. In order to
may have little in common with the experimental situation. This solve this problem, the fixed spin moment (FSM) method was pro-
is because excitations are two-particle processes and the missing posed.210,211 It is a computational trick interchanging dependent and
electron–hole interaction (i.e., the excitonic effect already men- independent variables. Physically speaking, one applies a magnetic
tioned in Sec. III C) can significantly affect the calculated optical field and obtains a moment, but computationally one chooses the
response of a material when they are strong. In order to overcome moment (as input) and calculates the field afterwards. In a conven-
this problem, one needs to include the electron–hole correlation tional spin-polarized calculation, the Fermi energy must be the same
explicitly by solving the BSE (see Sec. III I 3). for the spin-up and spin-down electrons. The magnetic moment M
is an output. In the FSM scheme, one does several constrained cal-
culations, where the moment M is an input, but allowing different
E. Magnetism
Fermi energies for the two spin states. One can interpret the differ-
When magnetism occurs in a solid, it may come from local- ence in the Fermi energies as a magnetic field. Although one needs to
ized electrons (e.g., from f electrons of rare-earth atoms) or itin- perform several calculations (instead of a single conventional one),
erant (delocalized) electrons (e.g., in Fe, Co, or Ni). In any case, they converge rather rapidly. The FSM method allows expanding the
magnetism comes mainly from exchange splitting causing a par- usual total-energy vs volume curve to an energy surface Etot (V, M)
tial occupation of states, which differ between the spin-up (N ↑ ) and as a function of volume V and moment M, which also provides new
spin-down (N ↓ ) electrons. The corresponding magnetic moment insights.
M is defined as the difference between these occupation numbers
(M = N ↑ − N ↓ ). For such cases, one must perform spin-polarized
F. Hyperfine fields and electric field gradients
calculations (runsp_lapw) and needs the spin density in addition to
the total electron density. The default is collinear magnetic order as All aspects of nucleus–electron interactions, which go beyond
found in ferromagnets, for example, in Fe, Co, Ni, or antiferromag- the electric point-charge model for a nucleus, define the hyperfine
nets, for example, in Cr. In addition to collinear magnets mentioned interactions. Nuclei with a nuclear quantum number I ⩾ 1 have
here, one can also handle non-collinear magnetism (for example, an electric quadrupole moment Q. The nuclear quadrupole inter-
systems with canted magnetic moments or spin spirals), as described action (NQI) stems from the interaction of such a moment and
in Sec. III I 2. the EFG, the second derivative of the Coulomb potential at the
For spin-polarized calculations of a specific complex antiferro- corresponding nuclear site. One can measure the EFG with Möss-
magnetic structure, in most cases, it is essential to specify proper bauer spectroscopy, NMR, nuclear quadrupole resonance (NQR),
(antiferromagnetic) atomic spin-moments as an input for the SCF or perturbed angular correlation (PAC). The NQI determines the
cycle and the WIEN2k tool instgen allows us to set this easily. If one product of Q and the EFG, a traceless tensor. The latter has a prin-
is interested in the orientation of the magnetic moments with respect cipal component and an asymmetry parameter η. The EFG is a
to the crystal structure (easy or hard axis) or the magneto-crystalline ground-state property that can be determined experimentally (mea-
anisotropy,206 the SO interaction must be included. For heavy ele- suring NQI), provided the nuclear quadrupole moment is known.
ments or when orbital moments become important, one needs this In early studies, the EFG was interpreted as a simple point charge
full relativistic treatment by including the SO interaction. model with additional corrections (Sternheimer factor212 ). However,
Nowadays, one can study complicated systems, for exam- later it was shown that the EFG can be calculated with DFT, as
ple, BaFe2 O5 , an oxygen-deficient perovskite-like structure, which was illustrated213 for LiN3 . The EFG is sensitive to the asymmetric

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-18


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

charge distribution around a given nucleus and thus is a local probe,


which often helps in clarifying the local atomic arrangement. The
reader can find a short description of several EFG calculations for
selected examples in Chap. 6.4 of Ref. 8, and here, we describe some
important aspects and examples below.
Computationally, it is important to treat both valence and semi-
core states very accurately because, due to the 1/r3 factor in the EFG
expression, even small asphericities near the nucleus lead to impor-
tant contributions. This was demonstrated for the first time for TiO2
in the rutile structure, where the radial functions of the fully occu-
pied Ti-3p semi-core orbitals are slightly different for px , py , and pz
states and thus contribute significantly to the EFG.28
The mapping of the two Cu-EFGs in the high-temperature
superconductor YBa2 Cu3 O7 to the “plane” and “chain” Cu sites
provides insight into which Cu atom is responsible for supercon-
ductivity, and the analysis of the EFG on all other sites helped us to
interpret chemical bonding.214 It should be mentioned that the EFG
at the (superconducting) Cu-plane site comes out quite inaccurate
and GGA+U calculations are necessary, leading to a redistribution
of 0.07 e− from a 3dx2 −y2 orbital to a 3dz2 orbital and an EFG in
agreement with the experiment.92
Next, we briefly discuss the study on 16 fluoroaluminates from
Ref. 215, for which experimental NMR data were compared to DFT
results for the 27 Al EFG. In all of these compounds, the aluminum
atoms occur in AlF3− 6 octahedra, which have a wide diversity of con-
nectivity and distortions. One of these structures is shown in the
inset of Fig. 13 for Ba3 Al2 F12 . These fluoroaluminates illustrate how
sensitive the EFG is to the exact position of neighboring atoms. A
perfect octahedral symmetry would have a vanishing EFG, but small
distortions cause an EFG. The calculations were first done using FIG. 13. The calculated EFG (principal component) vs the experimental 27 Al
the less accurate powder diffraction data for the atomic positions quadrupole frequency for 16 fluoroaluminates is shown for two cases based on
[Fig. 13(a)], and the correlation between experiment and theory is (a) the experimental structure and (b) the DFT optimized structure. For further
not very good. Then, a DFT structure optimization was done leading details, see Ref. 215. The inset shows as one example of the fluoroaluminates,
the AlF3−
6 octahedra in Ba3 Al2 F12 (Ba: blue, Al: red, and F: green).
to an almost perfect correlation between experimental and theo-
retical EFGs [Fig. 13(b)]. This structure optimization has an even
more pronounced effect on the asymmetry parameter, as shown in
is extremely useful. We recalculate the densities of all electrons,
Ref. 215. We should mention that there can also be a sensitivity
including the core, in each cycle of the SCF scheme, in contrast
to DFT functionals for EFG calculations as described in Refs. 66
to the frozen-core approximation. The resulting core polarization
and 80.
can often be the main contribution to the HFF. The high quality
Last but not least, we demonstrate how one can determine the
of such calculations was demonstrated for the double perovskite
nuclear quadrupole moment Q from a combination of theoretical
BaFe2 O5 , for which the DFT calculations of EFG and HFF provided
EFG calculations and experimental measurements of the quadrupole
new insights (for details, see Ref. 207).
splitting, which is proportional to the product of EFG and Q. From
the slope of a linear regression for the EFG of several Fe compounds,
we could deduce the nuclear quadrupole moment Q of 57 Fe, the G. NMR chemical and Knight shifts
most important Mössbauer isotope. It was found to be about twice The NMR shielding ←

σ tensor is defined as a constant between
as large (Q = 0.16 b) as the previous literature value (Q = 0.082 b), an induced magnetic field Bind at the nucleus at site R and the
suggesting to revise this nuclear property using electronic structure external uniform field Bext ,
calculations.216
The magnetic hyperfine field (HFF) at a nucleus originates Bind (R) = −←

σ (R)Bext . (30)
from a Zeeman interaction between the magnetic moment I of this
nucleus and the magnetic field at this site produced by the spin- Its value is usually in the range of ppm (part per million). Since the
polarized electrons in a ferromagnet. The HFF has contributions magnetic field cannot be controlled with such a precision, the tensor
from the Fermi-contact term (the spin density at the nucleus), an is measured only with respect to some reference,
orbital, and a spin dipolar contribution. Here, we skip the details
but mention that an all-electron treatment is crucial, especially for δ(R) = σref − σ(R), (31)
the Fermi-contact term, since one needs accurate values of the spin-
density close to the nucleus, for which the basis set used in WIEN2k and often only its isotropic part σ(R) = tr[←

σ (R)] is known.

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-19


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

The external magnetic field is a relatively weak perturbation and involves a sum over empty states (first term) as well as a sum
compared to the typical energy scale of the electronic structure; over core states (second term) because the core states have been cal-
therefore, its effect on the spin and orbit of an electron can be sepa- culated before [Eq. (34)] and Eq. (36) is only correct if all states
rated in the theoretical calculations. Here, we only outline the main of a system are included. Note that core wave functions appear in
features that are specific to the APW+lo method and the WIEN2k Eq. (36) as if they were unoccupied states. H (1) is the perturbation
code and are vital for achieving high accuracy of the computed NMR due to the external magnetic field in symmetric gauge,
tensor. A more in-depth discussion can be found in the original
publications. The formalism for computing the orbital part of the 1
H (1) = r × p ⋅ Bext . (37)
shielding (chemical shift) has been described in Refs. 217 and 218, 2c
and the spin part (Knight shift) of the response has been described
in Ref. 219. The formalism has been applied for computing shielding As discussed in Sec. II A, the basis functions of the APW+lo
in various insulating78,220,221 and metallic systems.222–224 method are highly tuned to describe the occupied Bloch states every-
As will be explained below, this approach can reach the basis set where in the unit cell (in particular, also close to the nucleus), but
limit for NMR calculations, and benchmark calculations for small they are not a complete basis set. Therefore, depending on the per-
molecular systems have proven that standard quantum chemistry turbation of the Hamiltonian, they may not be well suited to expand
methods can only reach this precision with very large uncontracted the perturbations of wave functions. In fact, magnetic fields and
quintuple-zeta basis sets and only for light atoms.225 NMR are such cases where the perturbed wave functions, in partic-
ular, near the nucleus, are very different, which means that the sum
1. Orbital component in Eq. (36) cannot be converged with the available set of orbitals. In
order to remedy this issue, we had to enhance the original WIEN2k
The orbital part of the shielding, i.e., the orbital component of
basis set. First, the standard set of local orbitals is extended signifi-
the induced field Bind , is obtained directly from the Biot–Savart law
cantly, both in the number of LOs per ℓ (typically to 5–8 LOs) and
(in atomic units, with c as speed of light),
also in ℓ (typically to ℓ + 1), where ℓ refers to the maximal “chemical”
1 r−R 3 ℓ of this atom.217 These extra local orbitals are referred to as NMR-
Bind (R) = j(r) × d r, (32) LOs, and the energy at which the radial functions of those NMR-LOs
c∫ ∣r − R∣3
are computed is chosen such that each of the NMR-LO radial func-
where j(r) is the induced orbital current, evaluated as expectation tions has zero value at the sphere boundary and the number of nodes
value of the current operator inside the sphere of subsequent LOs increases by one corresponding
to the next principal quantum number.217 However, these NMR-
p∣r⟩⟨r∣ + ∣r⟩⟨r∣p Bext × r LOs alone cannot completely improve the variational flexibility close
J(r) = − − ∣r⟩⟨r∣. (33)
2 2c to the nuclei.
The perturbation of the Hamiltonian due to the external mag-
WIEN2k separates the calculation of valence and core states. netic field is proportional to a product of position and momentum
The core state contribution to the induced current is computed using operators. As a result, the perturbation of the radial wave func-
the spherically symmetric core density only, tion utℓ contains components proportional to utℓ±1 and their radial

1 derivative r ∂r utℓ±1 . A direct introduction of basis functions based
jcore (r) = − ρcore (r)Bext × r. (34) ∂
on r ∂r u is not convenient within the APW formalism because such
2c
functions are not eigenstates of the radial Schrödinger equation.
The method for computing the valence contribution to j(r) is Therefore, we have proposed to add the desired term directly to the
based on a linear response approach226–228 originally developed by Green’s function present in the formula for the first-order perturba-
Mauri, Pfrommer, and Louie.226 The expression for the induced cur- tion of the valence state wave function.218 It is referred to as the “ ∂r

u
rent involves only the first-order terms with respect to the external correction” (DUC).
field Bext , The convergence test with respect to the number of NMR-LO
(1) (0) (0) (1) and DUC corrections is presented in Fig. 14. The induced current
j(r) = ∑[⟨ψo ∣J(0) (r)∣ψo ⟩ + ⟨ψo ∣J(0) (r)∣ψo ⟩ and shielding calculated within our linear response formalism is
o
compared to the exact value for an isolated Ar atom. The current and
(0) (0)
+ ⟨ψo ∣J(1) (r)∣ψo ⟩], (35) shielding for a spherically symmetric atom can be computed exactly
using only its density and the same formula [Eq. (34)] as for the core
where ψo is an unperturbed KS occupied orbital, J0 (r) is the para- states (diamagnetic current).
(0)

magnetic part of the current operator [the first term in Eq. (33)], and Both DUC and several additional NMR-LOs are needed to
J 1 (r) is the diamagnetic component of the current operator [the sec- reproduce the shape of the exact diamagnetic current in a region
(1)
ond term in Eq. (33)]. ψo is the first-order perturbation of ψo ,
(0) within 0.5 bohr from the nucleus. It appears in Fig. 14(a) that any
given by the standard formula involving a Green’s function, error in the representation of the current in this region results in
substantial errors of the computed shielding values [Fig. 14(b)]. This
(0)
⟨ψe ∣H (1) ∣ψo ⟩
(0) (0)
(1) (0) method can therefore reach the basis set limit.
(1) (0) (0) ⟨ψcore ∣H ∣ψo ⟩
∣ψo ⟩ = ∑ ∣ψe ⟩ + ∑ ∣ψcore ⟩ , The all-electron nature and the modular concept of WIEN2k
e εo − εe core ε o − ε core
makes it very easy to perform NMR calculations with wave func-
(36) tions including SO interactions for heavy nuclei or using different

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-20


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

FIG. 15. 19 F NMR shielding σ (in ppm) in alkali fluorides (data taken from Ref. 221).
Black lines: Total σ (full line) and contribution from within the F-sphere (dashed
line). Further decomposition into core F-1s (green), F-2s band (blue), metal-p band
(red), and F-2p band (brown) is also shown.

bonding (in the metal-d band) and anti-bonding (in the F-p band)
metal-p–F-p interaction giving slightly different (non-canceling)
diamagnetic and paramagnetic contributions. (v) The trend of σ
within the series comes mainly from the F-2p valence band. The
most important ingredient, which determines the size of the (mostly
FIG. 14. (a) Comparison of the induced current computed for an Ar atom (valence
states 3s and 3p) with and without DUC corrections and for 2 and 7 NMR-LOs. j ρ paramagnetic) F-2p contribution, is the position of the unoccupied
is the (exact) diamagnetic current [see Eq. (34)]. (b) Convergence of NMR abso- metal-d band. The perturbation due to the magnetic field couples the
lute shielding σ with respect to the number of NMR-LOs in the APW basis and occupied F-p states with Δℓ ± 1 to unoccupied d states, and due to
with/without DUC corrections. The vertical blue line represents the exact value of the energy denominator in Eq. (36), the Cs-d states give the largest
the absolute shielding σ Ar = 1245.7 ppm computed with j ρ . contribution because they are the closest in energy to the valence
bands. We can even artificially apply a (large) U value to the empty
Cs-5d states shifting them further up. In this way, we do not alter
DFT approximations, including DFT+U (Sec. II B 3), but, in par- the occupied states but still can increase the F shielding in CsF to
ticular, also hybrid functionals (Sec. II B 2). Thus, we can compare reproduce the LiF or NaF shifts.
the theoretical shielding with the experimental chemical shifts for By a similar analysis, we could explain why the 33 S magnetic
several different compounds, and from the correlation and slope of shielding decreases with the metal nuclear charge Z in the ionic
the linear regression curve, the quality of a particular approximation alkali/alkali-earth sulfides but increases in TM sulfides.220
to the XC effects can be evaluated. Ideally, the slope of this linear
regression line should be −1, but typically for ionic compounds,78 2. Spin component
the slope with PBE is too large (−1.2), while with the hybrid func-
tional YS-PBE0,66 it is too small (−0.8). This is quite in contrast to In order to compute the induced spin density and spin part of
organic molecules, where hybrid functionals perform much better the NMR shielding tensor, we use a direct approach,219 instead of
than the GGAs. Surprisingly, the BJ potential106 performs quite well applying the linear response formalism proposed, for instance, in
for ionic oxides or halides and yields slopes close to −1. Ref. 229. This is possible because the interaction of the spin with
Most importantly, theory should not only reproduce measured the external magnetic field does not break the periodicity. Therefore,
experimental values but also provide insights. WIEN2k allows us to we perform self-consistent spin polarized calculations with a finite
analyze and identify the contributions to the NMR shielding, and external magnetic field Bext acting on the electron spin only. The
for instance, for the F-shielding in the alkali-fluoride series (Fig. 15), interaction with Bext is cast into a spin-dependent potential lead-
the following observations can be made:221 (i) Basically, all contri- ing to a spin splitting of eigenstates and a finite spin magnetization.
butions to the F-shielding come from a region inside the F atomic The induced magnetic field at a given nucleus is computed using an
sphere (the current in the rest of the unit cell contributes negligi- expression for the magnetic hyperfine field,230
bly). (ii) The large (diamagnetic) shielding comes from the constant 8π S(r)
F-1s core contribution. (iii) The contributions from the F-2s bands Bhf = μB mav + ⟨Φ1 ∣ 3 [3(μ ⋅ r̂)r̂ − μ]∣Φ1 ⟩. (38)
3 r
are still diamagnetic, but much smaller and again constant within
the series. (iv) The diamagnetic metal-p semi-core contributions The first term (Bc ) is the Fermi contact term, where mav is the aver-
(Na-2p to Cs-5p) increase within the series. This can be explained age of the spin density in a region near the nucleus with a diam-
by the fact that for heavier elements the metal p-states increase in eter equal to the Thomson radius. The second term (Bsd ) captures
energy and come closer to the F-2p band. This leads to an increased the spin-dipolar contribution to the hyperfine field, where Φ1 is

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-21


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

the large component of the wave function, S is the reciprocal rela- The maximally localized WFs are calculated by Wan-
tivistic mass enhancement, and μ is the magnetic moment operator nier90,231 and the transformation matrices U k are provided by the
of the electron. Bsd comes almost entirely from within the atomic wien2wannier232 module of WIEN2k. The resulting WF can be
sphere, which simplifies its calculation. The spin contribution (σ s ), used for various tasks. They can be visualized and are useful for
i.e., the Knight shift to the shielding, is therefore given by two the interpretation of chemical bonding to generate tight-binding
terms, models or interpolations to very fine k-meshes for properties that
require fine k-meshes. Such properties can be transport, anoma-
Bhf = −←
σ→s Bext = −(←
σ→c + ←
σ→
sd )Bext . (39) lous Hall conductivity, linear and non-linear optics, Berry cur-
In order to obtain a sizable response and evaluate the NMR shield- vatures and topology, or electron–phonon interactions. In partic-
ing with a numerical precision at the level of 1 ppm, we apply in ular, for the improved description of electronic correlations via
our calculations an external magnetic field of 100 T, which induces a the DMFT approximation,233,234 WFs provide a realistic starting
spin-splitting of approximately 1 mRy. These small changes require point.
an extremely fine k-point sampling (for fcc Al, 106 k-points are In connection with wien2wannier, there is also the BerryPi235
needed), and this must always be carefully checked. module in WIEN2k, which calculates the polarization of solids
Further details and results of our approach can be found in using the Berry phase approach.236 BerryPi can calculate the change
Refs. 219, 222, and 223, but, here, we summarize the main find- of polarization ΔP in response to an external perturbation to
ings: (i) The previously accepted point of view, namely, that Knight study ferroelectricity, the Born effective charges, pyroelectric coef-
shifts are proportional to the partial s-DOS at the Fermi energy ficients, or the piezoelectric tensor. In addition, one can define
and the orbital contribution σ orb is identical to that in the (ionic) Wilson loops and calculate Chern numbers to study topological
reference compound, is only true for simple sp metals. (ii) In TM properties.237
d-elements or metallic compounds, the orbital part σ orb can be
as important as the spin part σ s . (iii) The s-DOS at EF is always I. External programs
important, but an induced TM-d magnetic moment (proportional
to the partial d-DOS at EF ) polarizes the core states in the oppo- 1. Thermoelectric transport coefficients
site direction so that the valence and core polarizations can partly WIEN2k is interfaced with the BoltzTraP2 program165 for cal-
cancel. (iv) The dipolar contribution σ sd is usually small, but, in culating transport coefficients within the relaxation time approxi-
anisotropic materials, a large dominance of one particular orbital at mation. The calculation is based on evaluating the transport distri-
EF can eventually lead to a very large contribution. We have found bution function,
this in BaGa2 , where the pz -DOS has a large and sharp peak at EF
leading to an aspherical magnetization density and a large dipolar d3 k
σ(ε, T) = ∫ ∑ vnk ⊗ vnk τnk δ(ε − εnk ) , (43)
contribution.222 n 8π3

using a fine mesh in k-space. To obtain the group velocities, vnk ,


H. Wannier functions and Berry phases and also quasi-particle energies on a fine mesh or effective masses,
A single particle state of a periodic system is conventionally rep- BoltzTrap2 relies on interpolating the eigenvalues, εnk , and possibly
resented as a Bloch state ψ nk (r), which is labeled by a band index n also the relaxation times, τ nk , using Fourier sums.
and a vector k inside the first BZ. It satisfies Bloch’s theorem The interpolation is performed so that the calculated eigen-
value energies are reproduced exactly. Within KS theory, the mul-
ψnk (r) = unk (r)eik ⋅ r , (40) tiplicative potential [Eq. (10)] means that it is often computation-
ally very efficient to calculate a fine mesh of eigenvalues, which
where unk (r) = unk (r + R) is a lattice periodic function and R is a can then be interpolated further to evaluate Eq. (43). This argu-
Bravais lattice vector. Alternatively, one can define Wannier func- ment no longer holds when hybrid functionals (Sec. II B 2), or the
tions (WFs) wnR (r) in unit cell R for a set of J bands as231,232 GW method (Sec. III I 4), are used to obtain the band structure.
J Therefore, BoltzTraP2 can also include the k-space derivatives in
V
wnR (r) = 3 ∫
k
e−ik⋅R ∑ Umn ψmk (r) d3 k, (41) the interpolation scheme, as can be obtained from the momentum
(2π) BZ m=1 matrix elements introduced in Sec. III D. This allows a more effi-
cient interpolation and the use of a coarser k-mesh in the actual DFT
where V is the unit cell volume and U k are unitary transformation calculation.
matrices that mix Bloch states at a given k. Because of the arbitrary Once the transport distribution has been obtained, the temper-
phase of ψ nk (r), the resulting WFs are usually not localized. Maxi- ature and chemical potential dependent transport coefficients
mally localized WFs can be obtained by choosing the U k such that
the spread Ω is minimized, ∂f (0) (ε; μ, T)
L
(α)
(μ; T) = q2 ∫ σ(ε, T)(ε − μ)α (− )dε (44)
J ∂ε
Ω = ∑ [⟨wmR ∣r2 ∣wmR ⟩ − ⟨wmR ∣r∣wmR ⟩2 ]. (42)
m=1 can be obtained by a simple numerical integration. Figure 16 shows
the highest valence bands of TiCoSb calculated using the inter-
k,b
This involves overlap integrals Mmn = ⟨umk ∣unk+b ⟩ between the polation scheme of BoltzTraP2. Compared to Fig. 7, a very good
periodic part of the wave functions on a uniform grid in the BZ. agreement with the band structure obtained using DFT is found.

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-22


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

Our implementation is based on a mixed spinor basis set


approach.241,242 In the interstitial region, the basis functions are pure
spinors given in a global (g) spin coordinate frame,
g
φK+k,σ = ei(K+k)⋅r χσ , (45)

g 1 g 0
where χ↑ = ( ) and χ↓ = ( ). Inside the atomic spheres, the basis
0 1
functions are a combination of both up and down spinors, which are
set in a local spin coordinate frame with a quantization axis pointing
along the direction of the average magnetization of the given atomic
sphere. This direction does not have to be the same for each sphere,
and the basis functions are [for a LAPW basis set, where we drop the
(r, E)-dependency in the radial u functions]
FIG. 16. Highest valence bands of TiCoSb obtained with the BoltzTraP2 inter-
polation together with the calculated thermoelectric power factor at 300 K using t t t t

a constant relaxation time of τ = 10−14 s. The inset shows the constant energy φLAPW K+k,σσ σ
K+k,σ = ∑ (Atℓm
K+k,σσ σ
utℓ + Btℓm u̇tℓ )Yℓm χσ t , (46)
surface at the energy marked by the dotted line. ℓ,m,σ t

where χσ t is a spinor given in a local coordinate frame. This choice


of the spin coordinate frame allows us to use spin-polarized radial
Since the Fourier interpolation is done bandwise (marked by color functions with the quantization axis along the direction of the aver-
in Fig. 16), the band crossing along the Γ − X direction is not repro- age magnetization. The matching of the φLAPW K+k,σ basis to the plane
duced. However, the error is hardly visible by eye and, in accordance waves at r = RMT is done for up and down plane waves in a global
with the intention of the original algorithm,238,239 no Fourier rip- spin coordinate frame,
ples are seen, which means that errors in the derivatives are isolated
t t t t
g g
to the points where the crossing occurs. The calculated thermoelec- ei(K+k)⋅r χσ = ∑ (AK+k,σσ
tℓm uσtℓ + BK+k,σσ
tℓm u̇σtℓ )Yℓm χσ t . (47)
tric power factor using a constant relaxation time is also shown. The ℓ,m,σ t
power factor peaks at a high value close to the band edge, where a
steep transport distribution can be expected. The high power factor Thus, the Atℓm and Btℓm depend on global σ and local σ t spin indices.
g ∗
can be attributed to the complex constant energy surface shown in Multiplying both sides of Eq. (47) by (χσ t ) , integrating over the
the inset, which is typical for half-Heusler compounds with favorable spin variable, and comparing to the standard collinear expression,
p-type thermoelectric performance.240
t t
the AK+k,σσ
tℓm and BK+k,σσ
tℓm are given by
BoltzTraP2 is written mainly in PYTHON3 and can be used as
t t
PYTHON library. The interpolation is handled by a single PYTHON K+k,σσ g ∗ g
Atℓm = (χσ t ) χσ AK+k,σ
tℓm , (48)
call, fitde3D. Once the Fourier coefficients have been obtained, the
interpolation of the bands onto the direction needed for plotting the K+k,σσ t
g ∗ g t

band structure (getBands), or the fine mesh needed for obtaining Btℓm = (χσ t ) χσ BK+k,σ
tℓm , (49)
the transport distribution (getBTPbands), can be performed. The t t

use of BoltzTraP2 as a library gives a reproducible and flexible work where AK+k,σ
tℓm and BK+k,σ
tℓm are the collinear matching coefficients
flow. The analysis associated with Fig. 16 can thus be performed calculated for “local” spins. We have extended the original formal-
with a single PYTHON script, which is included in the most recent ism241,243 beyond the atomic moment approximation, and the code
distribution. processes the spin density matrices without any approximations also
inside the atomic spheres. Note that the adaptation of the above
equations for LOs or the APW+lo basis set is trivial.
2. Non-collinear magnetism A few more details of the implementation are the following: In
WIEN2k can only compute the electronic structure of mag- WIENNCM, the default scalar relativistic Hamiltonian is extended
netic systems with a collinear spin arrangement. For performance with SO interactions and this doubles the size of the Hamiltonian.
benefits, WIEN2k assumes that the spin density matrix along some It is also possible to use the DFT+U method for correlated systems.
direction is diagonal for each eigenstate. When SO interactions are The setup and execution of WIENNCM is as in WIEN2k; however,
taken into account, this condition is sometimes not satisfied. In such the atomic structure has to be augmented with definitions of the
cases, the off-diagonal terms of the spin density matrix are simply magnetic structure, which requires to define the orientation of the
ignored during the SCF procedure and only the z component of average magnetic moment for each atom. This is to some extent
the spin density is properly converged. Generally, this is not a big automatized in such a way that the user only needs to provide the
issue for cases with collinear spin arrangement and is, in fact, com- orientation for the “magnetic atoms” (e.g., only U atoms in UO2 ).
mon practice. However, such an approximation cannot be applied The orientation for “non-magnetic” atoms (O in UO2 ) is generated
for systems with a non-collinear spin arrangement for which the automatically. WIENNCM makes use of the spin symmetry, which
full spin density matrix has to be considered. For that purpose, we simplifies the calculations.
have written a non-collinear spin version of WIEN2k, referred to as If one wants to calculate spin spirals, one can either handle
WIENNCM. this by (big) supercells or, more efficiently, by using the generalized

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-23


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

Bloch theorem244 (neglecting SO interactions) so that these calcula- fully relativistic treatment of core states to study XANES at L2,3 edges
tions can be done in the small crystallographic cell. of 3d TM compounds.196 In particular, for formally 3d0 compounds
Examples of application of our implementation can be found in such as TiO2 , the correct L2 /L3 branching ratio can be obtained due
Refs. 91 and 245. to interference effects of the 2p1/2 and 2p3/2 core states. Even fine dif-
ferences in line shape between the rutile and anatase modifications
3. Electron–hole interactions are in agreement with the experiment demonstrating the power of
this approach.
The state-of-the-art method to include electron–hole interac-
tions is based on the solution of the equation of motion of the two- 4. GW approximation for quasi-particle calculations
particle Green’s function, known as the BSE.246–248 The WIENBSE
code allows the calculation of the optical response taking into The GW method102,103 is considered as the state-of-the-art
account excitonic effects. The BSE is solved in an approximate man- method for a first-principles description of the electronic quasi-
ner by representing them in the form of an effective eigenvalue particle band structure in solids. The name of this many-body per-
problem with the so-called BSE Hamiltonian,249,250 turbation theory based method comes from the interacting Green’s
function G(r, r′ , ω), whose poles in the complex frequency plane
e λ λ λ
∑ Hvck,v′ c′ k′ Av′ c′ k′ = E Avck , (50) determine the single-particle excitation energies and W, the dynami-
v′ ,c′ ,k′ cally screened Coulomb potential, which is obtained using the polar-
where the sum runs over occupied (they form the hole upon exci- izability in the random-phase approximation. The central quan-
tation) valence (v) and unoccupied (they become occupied upon tity, namely, the self-energy Σ(r, r′ , ω) from which the first-order
excitation) conduction (c) bands and k points (supplied by a DFT correction to the KS eigenvalues can be calculated, is obtained
calculation performed with WIEN2k) and the electron–hole Hamil- from
tonian consists of three terms, H e = H diag + H dir + H x , which are i ′

given by [x = (r, σ)], Σ(r, r′ , ω) = G(r, r′ , ω + ω′ )W(r′ , r, ω′ )eiω η dω′ , (55)


2π ∫
diag
Hvck,v′ c′ k′ = (εvk − εck + Δ)δvv′ δcc′ δkk′ , (51) where η is an infinitesimal positive number.
The GAP2 code257,258 is the second version of an all-electron
dir ∗ ′ ′ ∗ ′ 3 3 ′
GW implementation based on the WIEN2k code. As WIEN2k,
Hvck,v ′ c′ k′ = − ∫ ψvk (x)ψck (x )W(r, r )ψv′ k′ (x)ψc′ k′ (x )d xd x ,
this highly parallelized code can use an arbitrary number of
(52) HDLOs,259,260 which ensures that one can obtain fully converged
GW results even in difficult cases such as ZnO. Since it is based on
an all-electron method and can use orbitals from DFT+U, a major
x ∗ ′ ∗ ′ ′ 3 3 ′
Hvck,v ′ c′ k′ = ∫ ψvk (x)ψck (x)v̄(r, r )ψv′ k′ (x )ψc′ k′ (x )d xd x . advantage of this code is the possibility to explore d- and f -electron
systems in a meaningful way.260 We note that for materials that are
(53) traditionally categorized as strongly correlated (e.g., the oxides), the
The H diag
term depends only on the eigenvalues and accounts for standard semilocal functionals usually fail (see Sec. II B 3). For such
the response in the non-interacting limit. The exchange H x and systems, using the DFT+U (or hybrid) orbitals as input for one-shot,
G0 W 0 performs much better.261,262 In addition, the code allows for
the direct H dir Coulomb terms couple the electron–hole pairs.251
partially self-consistent GW 0 calculations by updating G with the
The direct term, in principle, depends on the dynamically screened
modified eigenvalues.
Coulomb electron–hole interaction and on the excitation energy
Another important feature is the possibility for a first-principles
(Eλ ), but here we apply the usual approximation and only account determination of the Hubbard U using the constrained random
for non-local but static screening. The coupling coefficients Aλvck phase approximation263 and maximally localized WFs with an inter-
define the electron–hole correlation function and enter the expres- face to Wannier90231 by wien2wannier.232
sion for the imaginary part of the dielectric function,
R RR2 5. Dynamical mean field theory
8π2 RRRR λ ⟨vk∣ − i∇x ∣ck⟩ RRR
ε2 (ω) = ∑RRR ∑ Avck R δ(Eλ − ω). (54) Many systems have valence electrons in orbitals, which are
Ω λ RRv,c,k εvk − εck RRRR
R R quite extended in space and overlap strongly with their neighbors.
The BSE approach is very successful in dealing with excitons and This leads usually to strong bonding-antibonding effects, large band-
the response in the optical regime. The excitonic effects are some- width W, and dominant contributions from the kinetic energy. If,
times small (e.g., in small gap semiconductors), but still important, in addition, the bare Coulomb interaction U between two electrons
with binding energies of some tens of meV, sometimes large (in par- on the same site is strongly screened (maybe due to metallic char-
ticular, in insulators) with binding energies of a couple of eV such acter), such systems are usually quite well described by standard
that the resulting optical functions have hardly any resemblance DFT approximations. However, as already mentioned in Sec. II B 3,
with independent-particle results. The WIENBSE implementation 3d or 4f electrons may be more localized so that they participate
has been used in several works for various semiconductors,252–255 but much less in bonding and have a more atomic-like character. In
together with supercell calculations and GW- or TB-mBJ-based sin- these cases, the bandwidth W (metal) and Coulomb interaction U
gle particle states, it can also successfully describe F centers in wide (Mott insulator) compete as does the crystal-field splitting (low-spin
bandgap alkali halides.76,256 It has also been extended to include a state) and Hund’s rule coupling J (high-spin state). We talk about

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-24


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

“correlated electrons” and standard semilocal DFT approximations and perpendicular to the electric field (the so-called LO-TO split-
may fail badly in certain cases. The DFT+U and hybrid methods ting) in the small wave-vector limit can be obtained when addi-
discussed in Secs. II B 3 and II B 2, respectively, can be much tionally the Born effective charges (see Sec. III H) are supplied to
more accurate depending on the investigated property. However, the phonon programs. An alternative approach for phonon calcu-
the state-of-the-art approach for these correlated electron systems lations, namely, density functional perturbation theory,273 is not
is DMFT,233,234 which is based on the Hubbard model on a lattice, implemented.
described by the following Hamiltonian: Phonon calculations can be used to investigate various prop-

erties of materials. Frequencies at Γ are analyzed according to their
H = ∑ tij ciσ cjσ + U ∑ ni↑ ni↓ . (56) symmetry and can be compared to IR and Raman spectra (see, e.g.,
<ij>,σ i
Ref. 274 for application on PbFBr1−x Ix ). The full phonon band struc-
The first term (kinetic energy) describes the hopping t ij of an elec- ture and the corresponding phonon-DOS can be calculated and inte-
tron with spin σ from lattice site j to lattice site i, while the second grated, yielding thermodynamic quantities such as the mean square
term (potential energy) accounts for the strong Coulomb repul- thermal displacements, the specific heat, entropy, or free energy,
sion U between two electrons at the same lattice site i, which which together with the quasi-harmonic approximation can be used
is responsible for the correlations in the system. Within DMFT, to determine thermal expansion. Imaginary frequencies at certain
the complicated lattice problem is replaced by a single-site impu- k-points indicate an instability of this phase (at 0 K) and occur,
rity model, which hybridizes with a self-consistently determined for instance, in all cubic perovskites.275 Freezing in one of the cor-
non-interacting bath. responding eigenmodes with a certain amplitude and subsequent
The basic idea of DFT+DMFT is to divide the electrons in structure relaxation yields a more stable phase in a particular space
the system into two groups: weakly correlated electrons (i.e., elec- group of lower symmetry and can be used to detect and analyze
trons in s- and p-orbitals) that are well described by an approxi- second-order phase transitions in various materials.276–278
mate DFT functional and strongly correlated electrons (i.e., d- and
f -electrons) well described using DMFT. The model Hamiltonian 7. Band structure unfolding
for DFT+DMFT is then constructed for the correlated subset with
a suitable basis usually defined by Wannier functions as discussed The standard way to model defects, vacancies, alloys (disorder),
in Sec. III H. The full-orbital KS Hamiltonian H KS is then projected or surfaces is by means of a supercell approach. While sufficiently
onto the correlated subspace of the partially filled orbitals and many- large supercells can handle the energetics of these problems quite
body terms H U as well as a double counting correction H DC are well, it is fairly difficult to describe the effect of the perturbation on
added. the bulk electronic structure. A band structure from a supercell cal-
Several such DFT+DMFT codes264–269 use WIEN2k as basis, culation usually looks like a bunch of spaghetti and is very difficult
and numerous applications have proven the power of the com- to interpret. It is therefore highly desirable to display the band struc-
bined DFT+DMFT approach. The DMFT approach is often applied ture in the original BZ of the bulk material and indicate the original
to explain optical, XAS, or ARPES spectra (see, e.g., Ref. 270 for Bloch character as much as possible. This unfolding can be done
V2 O3 ) and can also estimate the intensities of the spectral features conveniently by the fold2Bloch utility.166,279
due to lifetime broadening. Recently, free energies and forces have In an h × k × l supercell band structure, each k-point transforms
also been made available,269,271 which allows us to study structural into h × k × l k-points of the bulk BZ. fold2Bloch calculates the
(e.g., α-γ Ce), magnetic (e.g., bcc-fcc Fe), or metal–insulator (e.g., corresponding spectral weights wn (k), which amounts to the Bloch
NdNiO3 ) phase transitions with temperature.272 character k of the nth eigenvalue εn , subject to the normalization
that ∑k wn (k) = 1, and displays wn (k) in the unfolded band struc-
ture so that one can distinguish between regular bulk and defect
6. Phonons
states.
WIEN2k does not have its own program to calculate phonon
spectra, but it is interfaced with at least three different external
phonon programs: phonopy,159,160 Phonon,161,162 and PHON.163,164 8. de Haas–van Alphen effect
They all employ the finite-displacement method162 and a harmonic The knowledge of the Fermi surface (FS) of a metallic com-
approximation. First, the crystal structure must be very well relaxed pound is important to understand its electronic and transport prop-
so that residual forces on all atoms are very small. For this structure, erties. de Haas–van Alphen (dHvA) measurements of the quantum
the phonon codes suggest a systematic set of displacements (depend- oscillatory magnetization contain detailed information about the FS
ing on symmetry) in a chosen supercell and WIEN2k calculates the and report frequencies that are proportional to extremal FS cross
forces for these displacements. These forces are then used by the sections perpendicular to the magnetic field direction.280 However,
phonon codes to calculate harmonic force constants and setup and it is not so easy to reconstruct from the measured data the actual
diagonalize the dynamical matrices at the desired k-points, which multi-band FS.
yields the phonon frequencies and their eigenmodes. For Γ-phonons On the other hand, FS calculations in WIEN2k are rather triv-
(infrared or Raman spectroscopy), a supercell is not required; oth- ial and can be well presented using XCrysDen.152 For a quantitative
erwise, the supercell should be large enough (typically more than 50 comparison with experiment, it is highly desirable to calculate the
atoms/cell) such that the force constants between atoms separated corresponding dHvA frequencies as well as the corresponding effec-
by more than the supercell size become negligible. In ionic solids, tive masses. This can be done conveniently using the SKEAF (Super-
the frequency splitting of the optical vibrational modes parallel cell k-space Extremal Area Finder) tool.167,280 An application can

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-25


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

be found for the determination of transport properties in the HoBi281 mandatory. We also described programs that are not part of the
compound. WIEN2k distribution but are compatible with it. This includes,
for instance, WIENNCM (non-collinear magnetism), WIENBSE
(electron–hole interactions), BoltzTraP2 (thermoelectric transport
IV. DISCUSSION AND SUMMARY
coefficients), GAP2 (GW), or the various programs that calculate
In this paper, we have reviewed the widely used WIEN2k code, phonons.
which is based on the APW+lo method to solve the KS equations of The main advantage of working with an all-electron code such
DFT. Particular emphasis was placed on the various types of basis as WIEN2k is the possibility to implement methods for calculating
functions that are available. One of the strengths of the WIEN2k properties exactly. However, implementations within the APW+lo
code is the possibility to use an arbitrary number of local orbitals, method are not always straightforward, since the dual basis-set rep-
which allows an accurate calculation of all states, from the low-lying resentation, atomic-like functions inside the atomic spheres, and
occupied semi-core to the high-lying unoccupied states. For the lat- plane waves between the atoms may lead to complicated equations.
ter, the use of local orbitals is crucial in order to get converged results In addition, the discontinuity of the derivatives of the basis functions
for a property that is calculated using perturbation theory such as the at the sphere boundary may require a careful treatment. DFT codes
NMR chemical shift. using a basis set consisting only of plane waves or only of localized
Various types of approximations for the treatment of XC effects basis functions (e.g., Gaussian) lead, in principle, to easier imple-
are mentioned, and the large number of functionals that are available mentations. However, once a method has been implemented in the
constitutes another strength of the WIEN2k code. They range from APW+lo method, it can then be applied to any element of the peri-
the semilocal approximations (all the existing ones can be used via odic table, from hydrogen to the actinides, without any restriction,
the Libxc library) to the more sophisticated approximations such as which is the great power of the APW+lo method.
DFT+U, the hybrids, or functionals specifically developed for van
der Waals interactions. In particular, the popular Tran–Blaha mBJ
potential is implemented in WIEN2k, which is a cheap but accurate ACKNOWLEDGMENTS
method to calculate bandgaps in solids. Since the WIEN2k code is
a full-potential all-electron code, it is, in principle, able to provide We would like to thank the current and former colleagues
the exact result within a chosen XC approximation. Thus, WIEN2k of the WIEN2k group and external collaborators who have con-
is ideally suited for the testing of XC functionals. tributed to the development of the WIEN2k code (see our web-
The structure of the WIEN2k code, as well as the workflow of site282 or the WIEN2k user’s guide283 ). We also thank all the users
programs in a SCF calculation, has been described. WIEN2k has also keeping the WIEN2k mailing284 list alive. L.D.M. acknowledges
a user-friendly interface that is especially useful for beginners. In support from the National Science Foundation, USA, under Grant
principle, an APW-based method needs many specific input param- No. DMR-1507101. P.B. acknowledges support from the Austrian
eters (various PW and LM cutoffs and case specific LM expansions Science Foundation (FWF) for Project W1243 (Solids4Fun).
and specific E-parameters for each atom and angular momentum),
but one of the great strengths of our implementation is that for all
these parameters very good defaults are provided automatically to REFERENCES
1
the user so that WIEN2k can also be mastered by non-experts. Of W. J. Hehre, L. Radom, P. von R. Schleyer, and J. Pople, Ab Initio Molecular
course, an all-electron code cannot be as fast as PW pseudopotential Orbital Theory (Wiley-VCH, 1986).
2
codes, where the extensive use of FFTs speeds up the calculations. P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964).
3
Despite this, an APW-based method can be fairly efficient when W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
4
large atomic spheres can be used because of the relatively fast PW K. Schwarz and P. Blaha, “DFT calculations for real solids,” in Handbook of
Solid State Chemistry, Theoretical Description Vol. 5, edited by R. Dronkowski,
convergence in such cases. It also has a fast and robust method
S. Kikkawa, and A. Stein (Wiley-VCH Verlag, Weinheim, Germany, 2017),
to solve the SCF problem, including a simultaneous optimization Chap. 8, p. 227.
of the atomic positions. WIEN2k is a very efficient implementa- 5
K. Lejaeghere, G. Bihlmayer, T. Björkman, P. Blaha, S. Blügel, V. Blum, D. Caliste,
tion of the APW+lo method from the computational point of view. I. E. Castelli, S. J. Clark, A. Dal Corso, S. de Gironcoli, T. Deutsch, J. K. Dewhurst,
The code is highly optimized and uses whenever possible efficient I. Di Marco, C. Draxl, M. Dułak, O. Eriksson, J. A. Flores-Livas, K. F. Garrity,
numerical libraries (BLAS, LAPACK, and ELPA). It has three dif- L. Genovese, P. Giannozzi, M. Giantomassi, S. Goedecker, X. Gonze, O. Grånäs, E.
ferent parallelization schemes, which allows us to run the code effi- K. U. Gross, A. Gulans, F. Gygi, D. R. Hamann, P. J. Hasnip, N. A. W. Holzwarth,
D. Iuşan, D. B. Jochym, F. Jollet, D. Jones, G. Kresse, K. Koepernik, E. Küçük-
ciently on a laptop as well as on a huge high performance computing benli, Y. O. Kvashnin, I. L. M. Locht, S. Lubeck, M. Marsman, N. Marzari,
cluster. U. Nitzsche, L. Nordström, T. Ozaki, L. Paulatto, C. J. Pickard, W. Poelmans,
WIEN2k can calculate a large number of different properties. M. I. J. Probert, K. Refson, M. Richter, G.-M. Rignanese, S. Saha, M. Scheffler,
Besides the basic quantities such as the optimized atomic struc- M. Schlipf, K. Schwarz, S. Sharma, F. Tavazza, P. Thunström, A. Tkatchenko,
ture, cohesive energy, electronic band structure, or magnetism, M. Torrent, D. Vanderbilt, M. J. van Setten, V. Van Speybroeck, J. M. Wills, J.
numerous more specialized properties are available and can be R. Yates, G.-X. Zhang, and S. Cottenier, Science 351, aad3000 (2016).
6
P. Blaha, K. Schwarz, G. K. H. Madsen, D. Kvasnicka, J. Luitz, R. Laskowski,
readily calculated. Among them, those whose corresponding pro-
F. Tran, and L. D. Marks, WIEN2k: An Augmented Plane Wave Plus Local Orbitals
grams or modules are part of the WIEN2k code are, for instance, Program for Calculating Crystal Properties (Vienna University of Technology,
the optical properties, electric polarization, electric-field gradients, Austria, 2018).
NMR chemical and Knight shifts, or magnetic hyperfine fields. In 7
D. J. Singh and L. Nordström, Planewaves, Pseudopotentials, and the LAPW
particular, for the latter two quantities, an all-electron method is Method, 2nd ed. (Springer, New York, 2006).

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-26


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

8 51
K. Schwarz, P. Blaha, and S. B. Trickey, Mol. Phys. 108, 3147 (2010). P. Kovács, F. Tran, P. Blaha, and G. K. H. Madsen, J. Chem. Phys. 150, 164119
9 (2019).
J. C. Slater, Phys. Rev. 51, 846 (1937).
10 52
O. K. Andersen, Phys. Rev. B 12, 3060 (1975). A. D. Becke, J. Chem. Phys. 98, 5648 (1993).
11 53
D. D. Koelling and G. O. Arbman, J. Phys. F: Met. Phys. 5, 2041 (1975). T. Bredow and A. R. Gerson, Phys. Rev. B 61, 5194 (2000).
12 54
M. Weinert, J. Math. Phys. 22, 2433 (1981). J. K. Perry, J. Tahir-Kheli, and W. A. Goddard III, Phys. Rev. B 63, 144510
13
E. Wimmer, H. Krakauer, M. Weinert, and A. J. Freeman, Phys. Rev. B 24, 864 (2001).
55
(1981). J. Muscat, A. Wander, and N. M. Harrison, Chem. Phys. Lett. 342, 397 (2001).
14 56
P. Blaha, K. Schwarz, P. Sorantin, and S. B. Trickey, Comput. Phys. Commun. J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003); 124,
59, 399 (1990). 219906 (2006).
15 57
D. Singh, Phys. Rev. B 43, 6388 (1991). J. Heyd, J. E. Peralta, G. E. Scuseria, and R. L. Martin, J. Chem. Phys. 123, 174101
16
E. Sjöstedt, L. Nordström, and D. J. Singh, Solid State Commun. 114, 15 (2000). (2005).
58
17
G. K. H. Madsen, P. Blaha, K. Schwarz, E. Sjöstedt, and L. Nordström, Phys. A. V. Krukau, O. A. Vydrov, A. F. Izmaylov, and G. E. Scuseria, J. Chem. Phys.
Rev. B 64, 195134 (2001). 125, 224106 (2006).
59
18
G. Michalicek, M. Betzinger, C. Friedrich, and S. Blügel, Comput. Phys. Com- M. Ernzerhof and G. E. Scuseria, J. Chem. Phys. 110, 5029 (1999).
60
mun. 184, 2670 (2013). C. Adamo and V. Barone, J. Chem. Phys. 110, 6158 (1999).
19 61
F. Karsai, F. Tran, and P. Blaha, Comput. Phys. Commun. 220, 230 (2017). D. M. Bylander and L. Kleinman, Phys. Rev. B 41, 7868 (1990).
62
20
K. Schwarz, “Computation of materials properties at the atomic scale,” in J. M. Crowley, J. Tahir-Kheli, and W. A. Goddard III, J. Phys. Chem. Lett. 7,
Selected Topics in Application of Quantum Mechanics, edited by M. R. Pahlavani 1198 (2016).
63
(InTechOpen, 2015), Chap. 10, p. 275. A. J. Garza and G. E. Scuseria, J. Phys. Chem. Lett. 7, 4165 (2016).
21 64
K. Schwarz, P. Blaha, and G. K. H. Madsen, Comput. Phys. Commun. 147, 71 F. Tran and P. Blaha, J. Phys. Chem. A 121, 3318 (2017).
65
(2002). P. Borlido, T. Aull, A. W. Huran, F. Tran, M. A. L. Marques, and S. Botti,
22 J. Chem. Theory Comput. 15, 5069 (2019).
K. Schwarz, J. Solid State Chem. 176, 319 (2003).
66
23
K. Schwarz and P. Blaha, Comput. Mater. Sci. 28, 259 (2003). F. Tran and P. Blaha, Phys. Rev. B 83, 235118 (2011).
67
24
M. Kara and K. Kurki-Suonio, Acta Crystallogr., Sect. A 37, 201 (1981). S. Massidda, M. Posternak, and A. Baldereschi, Phys. Rev. B 48, 5058 (1993).
68
25
D. D. Koelling and B. N. Harmon, J. Phys. C: Solid State Phys. 10, 3107 (1977). J. Spencer and A. Alavi, Phys. Rev. B 77, 193110 (2008).
69
26
A. H. MacDonald, W. E. Pickett, and D. D. Koelling, J. Phys. C: Solid State Phys. R. Sundararaman and T. A. Arias, Phys. Rev. B 87, 165122 (2013).
70
13, 2675 (1980). T. Shimazaki and Y. Asai, Chem. Phys. Lett. 466, 91 (2008).
27 71
J. Kuneš, P. Novák, R. Schmid, P. Blaha, and K. Schwarz, Phys. Rev. B 64, 153102 J. Paier, M. Marsman, K. Hummer, G. Kresse, I. C. Gerber, and J. G. Ángyán,
(2001). J. Chem. Phys. 124, 154709 (2006); 125, 249901 (2006).
28 72
P. Blaha, D. J. Singh, P. I. Sorantin, and K. Schwarz, Phys. Rev. B 46, 1321 (1992). A. Alkauskas and A. Pasquarello, Physica B 401-402, 670 (2007).
29 73
A. Seidl, A. Görling, P. Vogl, J. A. Majewski, and M. Levy, Phys. Rev. B 53, 3764 F. Tran, Phys. Lett. A 376, 879 (2012).
74
(1996). A. S. Botana, F. Tran, V. Pardo, D. Baldomir, and P. Blaha, Phys. Rev. B 85,
30 235118 (2012).
A. J. Cohen, P. Mori-Sánchez, and W. Yang, Chem. Rev. 112, 289 (2012).
31 75
K. Burke, J. Chem. Phys. 136, 150901 (2012). F. Tran, D. Koller, and P. Blaha, Phys. Rev. B 86, 134406 (2012).
32 76
A. D. Becke, J. Chem. Phys. 140, 18A301 (2014). F. Karsai, P. Tiwald, R. Laskowski, F. Tran, D. Koller, S. Gräfe, J. Burgdörfer,
33
M. A. L. Marques, M. J. T. Oliveira, and T. Burnus, Comput. Phys. Commun. L. Wirtz, and P. Blaha, Phys. Rev. B 89, 125429 (2014).
77
183, 2272 (2012). F. Tran, F. Karsai, and P. Blaha, Phys. Rev. B 89, 155106 (2014).
34 78
S. Lehtola, C. Steigemann, M. J. T. Oliveira, and M. A. L. Marques, SoftwareX 7, R. Laskowski, P. Blaha, and F. Tran, Phys. Rev. B 87, 195130 (2013).
79
1 (2018). D. Koller, P. Blaha, and F. Tran, J. Phys.: Condens. Matter 25, 435503 (2013).
35 80
J. P. Perdew and K. Schmidt, AIP Conf. Proc. 577, 1 (2001). F. Tran, S. Ehsan, and P. Blaha, Phys. Rev. Mater. 2, 023802 (2018).
36 81
F. Tran, R. Laskowski, P. Blaha, and K. Schwarz, Phys. Rev. B 75, 115131 (2007). V. I. Anisimov, J. Zaanen, and O. K. Andersen, Phys. Rev. B 44, 943 (1991).
37 82
P. Haas, F. Tran, and P. Blaha, Phys. Rev. B 79, 085104 (2009); 79, 209902(E) P. Novák, J. Kuneš, L. Chaput, and W. E. Pickett, Phys. Status Solidi B 243, 563
(2009). (2006).
38 83
P. Haas, F. Tran, P. Blaha, L. S. Pedroza, A. J. R. da Silva, M. M. Odashima, and F. Tran, P. Blaha, K. Schwarz, and P. Novák, Phys. Rev. B 74, 155108 (2006).
K. Capelle, Phys. Rev. B 81, 125136 (2010). 84
K. Terakura, T. Oguchi, A. R. Williams, and J. Kübler, Phys. Rev. B 30, 4734
39
F. Tran, J. Stelzl, and P. Blaha, J. Chem. Phys. 144, 204120 (2016). (1984).
40 85
A. E. Mattsson, R. Armiento, J. Paier, G. Kresse, J. M. Wills, and T. R. Mattsson, V. I. Anisimov, I. V. Solovyev, M. A. Korotin, M. T. Czyżyk, and G. A. Sawatzky,
J. Chem. Phys. 128, 084714 (2008). Phys. Rev. B 48, 16929 (1993).
41 86
G.-X. Zhang, A. M. Reilly, A. Tkatchenko, and M. Scheffler, New J. Phys. 20, M. T. Czyżyk and G. A. Sawatzky, Phys. Rev. B 49, 14211 (1994).
063020 (2018). 87
A. B. Shick, A. I. Liechtenstein, and W. E. Pickett, Phys. Rev. B 60, 10763 (1999).
42 88
R. Armiento and A. E. Mattsson, Phys. Rev. B 72, 085108 (2005). E. R. Ylvisaker, W. E. Pickett, and K. Koepernik, Phys. Rev. B 79, 035103 (2009).
43 89
J. P. Perdew, A. Ruzsinszky, G. I. Csonka, O. A. Vydrov, G. E. Scuseria, L. K. Nawa, T. Akiyama, T. Ito, K. Nakamura, T. Oguchi, and M. Weinert, Phys.
A. Constantin, X. Zhou, and K. Burke, Phys. Rev. Lett. 100, 136406 (2008); 102, Rev. B 97, 035117 (2018).
039902(E) (2009). 90
Y.-C. Wang and H. Jiang, J. Chem. Phys. 150, 154116 (2019).
44
Z. Wu and R. E. Cohen, Phys. Rev. B 73, 235116 (2006). 91
R. Laskowski, G. K. H. Madsen, P. Blaha, and K. Schwarz, Phys. Rev. B 69,
45
G. K. H. Madsen, Phys. Rev. B 75, 195108 (2007). 140408(R) (2004).
46 92
Y. Zhao and D. G. Truhlar, J. Chem. Phys. 128, 184109 (2008). P. Blaha, K. Schwarz, and P. Novák, Int. J. Quantum Chem. 101, 550 (2005).
47 93
J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996); 78, F. Tran, J. Schweifer, P. Blaha, K. Schwarz, and P. Novák, Phys. Rev. B 77,
1396(E) (1997). 085123 (2008).
48 94
J. Sun, A. Ruzsinszky, and J. P. Perdew, Phys. Rev. Lett. 115, 036402 (2015). L. D. Marks, A. N. Chiaramonti, F. Tran, and P. Blaha, Surf. Sci. 603, 2179
49
E. B. Isaacs and C. Wolverton, Phys. Rev. Mater. 2, 063801 (2018). (2009).
50 95
Y. Fu and D. J. Singh, Phys. Rev. Lett. 121, 207201 (2018). R. Atta-Fynn and A. K. Ray, Europhys. Lett. 85, 27008 (2009).

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-27


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

96 141
X. Rocquefelte, M.-H. Whangbo, A. Villesuzanne, S. Jobic, F. Tran, K. Schwarz, S. Grimme, S. Ehrlich, and L. Goerigk, J. Comput. Chem. 32, 1456 (2011).
and P. Blaha, J. Phys.: Condens. Matter 22, 045502 (2010). 142
I. Hamada, Phys. Rev. B 89, 121103(R) (2014); 91, 119902(E) (2015).
97 143
Y. Lin, J. Wen, L. Hu, R. M. Kennedy, P. C. Stair, K. R. Poeppelmeier, and M. Weinert, E. Wimmer, and A. J. Freeman, Phys. Rev. B 26, 4571 (1982).
L. D. Marks, Phys. Rev. Lett. 111, 156101 (2013). 144
P. Pulay, Mol. Phys. 17, 197 (1969).
98 145
Q. C. Sherman, P. W. Voorhees, and L. D. Marks, Acta Mater. 181, 584 (2019). R. Yu, D. Singh, and H. Krakauer, Phys. Rev. B 43, 6411 (1991).
99 146
D. Torumba, P. Novák, and S. Cottenier, Phys. Rev. B 77, 155101 (2008). B. Kohler, S. Wilke, M. Scheffler, R. Kouba, and C. Ambrosch-Draxl, Comput.
100
F. Jollet, G. Jomard, B. Amadon, J. P. Crocombette, and D. Torumba, Phys. Phys. Commun. 94, 31 (1996).
Rev. B 80, 235109 (2009). 147
F. Tran, J. Kuneš, P. Novák, P. Blaha, L. D. Marks, and K. Schwarz, Comput.
101
C. A. Mizzi, P. Koirala, and L. D. Marks, Phys. Rev. Mater. 2, 025001 (2018). Phys. Commun. 179, 784 (2008).
102 148
L. Hedin, Phys. Rev. 139, A796 (1965). L. D. Marks and D. R. Luke, Phys. Rev. B 78, 075114 (2008).
103 149
M. S. Hybertsen and S. G. Louie, Phys. Rev. B 34, 5390 (1986). L. D. Marks, J. Chem. Theory Comput. 9, 2786 (2013).
104 150
F. Tran, J. Doumont, L. Kalantari, A. W. Huran, M. A. L. Marques, and L. D. Marks, “Hands-free DFT mixing” (unpublished).
151
P. Blaha, J. Appl. Phys. 126, 110902 (2019). R. Car and M. Parrinello, Phys. Rev. Lett. 55, 2471 (1985).
105 152
F. Tran and P. Blaha, Phys. Rev. Lett. 102, 226401 (2009). A. Kokalj, Comput. Mater. Sci. 28, 155 (2003).
106 153
A. D. Becke and E. R. Johnson, J. Chem. Phys. 124, 221101 (2006). K. Momma and F. Izumi, J. Appl. Crystallogr. 44, 1272 (2011).
107 154
F. Tran, P. Blaha, M. Betzinger, and S. Blügel, Phys. Rev. B 91, 165121 (2015). See http://www.fftw.org for information about FFTW.
108 155
J. P. Perdew and Y. Wang, Phys. Rev. B 45, 13244 (1992); 98, 079904(E) (2018). See https://elpa.mpcdf.mpg.de for information about ELPA.
109 156
A. D. Becke and M. R. Roussel, Phys. Rev. A 39, 3761 (1989). A. Marek, V. Blum, R. Johanni, V. Havu, B. Lang, T. Auckenthaler, A. Hei-
110
D. Koller, F. Tran, and P. Blaha, Phys. Rev. B 85, 155109 (2012). necke, H.-J. Bungartz, and H. Lederer, J. Phys.: Condens. Matter 26, 213201
111
R. A. Jishi, O. B. Ta, and A. A. Sharif, J. Phys. Chem. C 118, 28344 (2014). (2014).
157
112
O. Gritsenko, R. van Leeuwen, E. van Lenthe, and E. J. Baerends, Phys. Rev. A See http://www.wannier.org for information about Wannier90.
158
51, 1944 (1995). G. Pizzi, V. Vitale, R. Arita, S. Blügel, F. Freimuth, G. Géranton, M.
113
M. Kuisma, J. Ojanen, J. Enkovaara, and T. T. Rantala, Phys. Rev. B 82, 115106 Gibertini, D. Gresch, C. Johnson, T. Koretsune, J. Ibañez-Azpiroz, H. Lee,
(2010). J.-M. Lihm, D. Marchand, A. Marrazzo, Y. Mokrousov, J. I. Mustafa, Y. Nohara,
114 Y. Nomura, L. Paulatto, S. Poncé, T. Ponweiser, J. Qiao, F. Thöle, S. S. Tsirkin,
J. P. Perdew, R. G. Parr, M. Levy, and J. L. Balduz, Jr., Phys. Rev. Lett. 49, 1691
M. Wierzbowska, N. Marzari, D. Vanderbilt, I. Souza, A. A. Mostofi, and
(1982).
115 J. R. Yates, J. Phys.: Condens. Matter 32, 165902 (2020).
L. J. Sham and M. Schlüter, Phys. Rev. Lett. 51, 1888 (1983). 159
116
See https://atztogo.github.io/phonopy/ for information about phonopy.
E. J. Baerends, Phys. Chem. Chem. Phys. 19, 15639 (2017). 160
117
A. Togo and I. Tanaka, Scr. Mater. 108, 1 (2015).
I. E. Castelli, T. Olsen, S. Datta, D. D. Landis, S. Dahl, K. S. Thygesen, and 161
See http://www.computingformaterials.com for information about Phonon.
K. W. Jacobsen, Energy Environ. Sci. 5, 5814 (2012). 162
118 K. Parlinski, Z. Q. Li, and Y. Kawazoe, Phys. Rev. Lett. 78, 4063 (1997).
E. Engel and S. H. Vosko, Phys. Rev. B 47, 13164 (1993). 163
119 D. Alfè, Comput. Phys. Commun. 180, 2622 (2009).
J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, 164
D. J. Singh, and C. Fiolhais, Phys. Rev. B 46, 6671 (1992); 48, 4978(E) (1993). See http://www.homepages.ucl.ac.uk/∼ucfbdxa/phon/ for information about
120 PHON.
R. Armiento and S. Kümmel, Phys. Rev. Lett. 111, 036402 (2013). 165
121 G. K. H. Madsen, J. Carrete, and M. J. Verstraete, Comput. Phys. Commun.
V. Vlček, G. Steinle-Neumann, L. Leppert, R. Armiento, and S. Kümmel, Phys.
231, 140 (2018).
Rev. B 91, 035107 (2015). 166
122 See https://github.com/rubel75/fold2Bloch-Wien2k/ for information about
P. Verma and D. G. Truhlar, J. Phys. Chem. Lett. 8, 380 (2017).
123 fold2Bloch.
K. Finzel and A. I. Baranov, Int. J. Quantum Chem. 117, 40 (2017). 167
124 See http://www.wien2k.at/reg_user/unsupported/ for information about
F. Tran, J. Doumont, P. Blaha, M. A. L. Marques, S. Botti, and A. P. Bartók, SKEAF.
J. Chem. Phys. 151, 161102 (2019). 168
125 A. Otero-de-la-Roza, M. A. Blanco, A. Martín Pendás, and V. Luaña, Comput.
P. Verma and D. G. Truhlar, J. Phys. Chem. C 121, 7144 (2017). Phys. Commun. 180, 157 (2009).
126
T. Aschebrock and S. Kümmel, Phys. Rev. Res. 1, 033082 (2019). 169
127
A. Otero-de-la-Roza, E. R. Johnson, and V. Luaña, Comput. Phys. Commun.
J. C. Slater, Phys. Rev. 81, 385 (1951). 185, 1007 (2014).
128
J. B. Krieger, Y. Li, and G. J. Iafrate, Phys. Rev. A 46, 5453 (1992). 170
See https://www.chemistry.mcmaster.ca/bader/aim/ for the AIM theory of
129
F. Tran, P. Blaha, and K. Schwarz, J. Chem. Theory Comput. 11, 4717 (2015). Bader.
130
F. Tran, P. Blaha, M. Betzinger, and S. Blügel, Phys. Rev. B 94, 165149 (2016). 171
R. F. W. Bader, Atoms in Molecules: A Quantum Theory (Oxford University
131
K. Berland, V. R. Cooper, K. Lee, E. Schröder, T. Thonhauser, P. Hyldgaard, Press, New York, 1994).
and B. I. Lundqvist, Rep. Prog. Phys. 78, 066501 (2015). 172
R. Laskowski, P. Blaha, T. Gallauner, and K. Schwarz, Phys. Rev. Lett. 98,
132
J. Hermann, R. A. DiStasio, Jr., and A. Tkatchenko, Chem. Rev. 117, 4714 106802 (2007).
(2017). 173
R. Laskowski and P. Blaha, Phys. Rev. B 81, 075418 (2010).
133
S. Grimme, J. Comput. Chem. 27, 1787 (2006). 174
H. P. Koch, R. Laskowski, P. Blaha, and K. Schwarz, Phys. Rev. B 86, 155404
134
S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, J. Chem. Phys. 132, 154104 (2012).
(2010). 175
P. Blaha, H. Hofstätter, O. Koch, R. Laskowski, and K. Schwarz, J. Comput.
135
See https://www.chemie.uni-bonn.de/pctc/mulliken-center/software/dft-d3/ Phys. 229, 453 (2010).
dft-d3 for information about DFT-D3. 176
See http://www.cryst.ehu.es for the Bilbao Crystallographic Server.
136 177
J. Moellmann and S. Grimme, J. Phys. Chem. C 118, 7615 (2014). S. Ouardi, G. H. Fecher, C. Felser, M. Schwall, S. S. Naghavi, A. Gloskovskii,
137
M. Dion, H. Rydberg, E. Schröder, D. C. Langreth, and B. I. Lundqvist, Phys. B. Balke, J. Hamrle, K. Postava, J. Pištora, S. Ueda, and K. Kobayashi, Phys. Rev. B
Rev. Lett. 92, 246401 (2004); 95, 109902(E) (2005). 86, 045116 (2012).
138 178
G. Román-Pérez and J. M. Soler, Phys. Rev. Lett. 103, 096102 (2009). P. E. Blöchl, O. Jepsen, and O. K. Andersen, Phys. Rev. B 49, 16223 (1994).
139 179
F. Tran, J. Stelzl, D. Koller, T. Ruh, and P. Blaha, Phys. Rev. B 96, 054103 (2017). M. Bagheri and P. Blaha, J. Electron Spectrosc. Relat. Phenom. 230, 1 (2019).
140 180
F. Tran, L. Kalantari, B. Traoré, X. Rocquefelte, and P. Blaha, Phys. Rev. Mater. M. B. Trzhaskovskaya, V. I. Nefedov, and V. G. Yarzhemsky, At. Data Nucl.
3, 063602 (2019). Data Tables 77, 97 (2001).

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-28


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

181 219
M. B. Trzhaskovskaya, V. K. Nikulin, V. I. Nefedov, and V. G. Yarzhemsky, R. Laskowski and P. Blaha, J. Phys. Chem. C 119, 19390 (2015).
At. Data Nucl. Data Tables 92, 245 (2006). 220
R. Laskowski and P. Blaha, J. Phys. Chem. C 119, 731 (2015).
182 221
J. F. Janak, Phys. Rev. B 18, 7165 (1978). R. Laskowski and P. Blaha, Phys. Rev. B 85, 245117 (2012).
183 222
J. C. Slater, J. B. Mann, T. M. Wilson, and J. H. Wood, Phys. Rev. 184, 672 R. Laskowski, K. H. Khoo, F. Haarmann, and P. Blaha, J. Phys. Chem. C 121,
(1969). 753 (2017).
184
M. V. Ganduglia-Pirovano, M. Scheffler, A. Baraldi, S. Lizzit, G. Comelli, 223
K. H. Khoo, R. Laskowski, and P. Blaha, J. Phys. Chem. C 121, 12398 (2017).
G. Paolucci, and R. Rosei, Phys. Rev. B 63, 205415 (2001). 224
L. Kalantari, P. Blaha, K. H. Khoo, and R. Laskowski, J. Phys. Chem. C 121,
185
E. Kabliman, P. Blaha, and K. Schwarz, Phys. Rev. B 82, 125308 (2010). 28454 (2017).
186
P. S. Bagus, C. J. Nelin, X. Zhao, S. V. Levchenko, E. Davis, X. Weng, F. Späth, 225
G. A. de Wijs, R. Laskowski, P. Blaha, R. W. A. Havenith, G. Kresse, and
C. Papp, H. Kuhlenbeck, and H.-J. Freund, Phys. Rev. B 100, 115419 (2019). M. Marsman, J. Chem. Phys. 146, 064115 (2017).
187
G. B. Grad, P. Blaha, K. Schwarz, W. Auwärter, and T. Greber, Phys. Rev. B 68, 226
F. Mauri, B. G. Pfrommer, and S. G. Louie, Phys. Rev. Lett. 77, 5300 (1996).
085404 (2003). 227
C. J. Pickard and F. Mauri, Phys. Rev. B 63, 245101 (2001).
188
H. Dil, J. Lobo-Checa, R. Laskowski, P. Blaha, S. Berner, J. Osterwalder, and 228
J. R. Yates, C. J. Pickard, and F. Mauri, Phys. Rev. B 76, 024401 (2007).
T. Greber, Science 319, 1824 (2008). 229
189 M. d’Avezac, N. Marzari, and F. Mauri, Phys. Rev. B 76, 165122 (2007).
R. Laskowski, T. Gallauner, P. Blaha, and K. Schwarz, J. Phys.: Condens. Matter 230
S. Blügel, H. Akai, R. Zeller, and P. H. Dederichs, Phys. Rev. B 35, 3271
21, 104210 (2009).
190 (1987).
U. von Barth and G. Grossmann, Solid State Commun. 32, 645 (1979). 231
191 A. A. Mostofi, J. R. Yates, Y.-S. Lee, I. Souza, D. Vanderbilt, and N. Marzari,
C. Cocchi, S. Mistry, M. Schmeißer, J. Kühn, and T. Kamps, J. Phys.: Condens. Comput. Phys. Commun. 178, 685 (2008).
Matter 31, 014002 (2019). 232
192 J. Kuneš, R. Arita, P. Wissgott, A. Toschi, H. Ikeda, and K. Held, Comput. Phys.
J. Luitz, M. Maier, C. Hébert, P. Schattschneider, P. Blaha, K. Schwarz, and Commun. 181, 1888 (2010).
B. Jouffrey, Eur. Phys. J. B 21, 363 (2001). 233
193 G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko, O. Parcollet, and C.
T. Mizoguchi, I. Tanaka, S. Yoshioka, M. Kunisu, T. Yamamoto, and W. Ching, A. Marianetti, Rev. Mod. Phys. 78, 865 (2006).
Phys. Rev. B 70, 045103 (2004). 234
194 K. Held, Adv. Phys. 56, 829 (2007).
W. Hetaba, P. Blaha, F. Tran, and P. Schattschneider, Phys. Rev. B 85, 205108 235
S. J. Ahmed, J. Kivinen, B. Zaporzan, L. Curiel, S. Pichardo, and O. Rubel,
(2012).
195 Comput. Phys. Commun. 184, 647 (2013).
H. Ikeno, F. M. F. de Groot, E. Stavitski, and I. Tanaka, J. Phys.: Condens. 236
R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651(R) (1993).
Matter 21, 104208 (2009). 237
196 H. Weng, C. Fang, Z. Fang, B. A. Bernevig, and X. Dai, Phys. Rev. X 5, 011029
R. Laskowski and P. Blaha, Phys. Rev. B 82, 205104 (2010).
197 (2015).
F. Karsai, M. Humer, E. Flage-Larsen, P. Blaha, and G. Kresse, Phys. Rev. B 98, 238
D. D. Koelling and J. H. Wood, J. Comput. Phys. 67, 253 (1986).
235205 (2018). 239
198 W. E. Pickett, H. Krakauer, and P. B. Allen, Phys. Rev. B 38, 2721 (1988).
M. Nelhiebel, P.-H. Louf, P. Schattschneider, P. Blaha, K. Schwarz, and B. 240
Jouffrey, Phys. Rev. B 59, 12807 (1999). S. Bhattacharya and G. K. H. Madsen, J. Mater. Chem. C 4, 11261 (2016).
241
199
P. Schattschneider, C. Hébert, H. Franco, and B. Jouffrey, Phys. Rev. B 72, H. Yamagami, Phys. Rev. B 61, 6246 (2000).
242
045142 (2005). P. Kurz, F. Förster, L. Nordström, G. Bihlmayer, and S. Blügel, Phys. Rev. B 69,
200
C. Hébert, Micron 38, 12 (2007). 024415 (2004).
243
201
C. Hébert-Souche, P.-H. Louf, M. Nelhiebel, J. Luitz, P. Schattschneider, P. Kurz, G. Bihlmayer, S. Blügel, K. Hirai, and T. Asada, Phys. Rev. B 63, 096401
K. Schwarz, and B. Jouffrey, Ultramicroscopy 83, 9 (2000). (2001).
202 244
C. Hébert, J. Luitz, and P. Schattschneider, Micron 34, 219 (2003). L. M. Sandratskii, Adv. Phys. 47, 91 (1998).
203 245
K. S. Virdi, Y. Kauffmann, C. Ziegler, P. Ganter, B. V. Lotsch, W. D. Kaplan, J. Kuneš and R. Laskowski, Phys. Rev. B 70, 174415 (2004).
246
P. Blaha, and C. Scheu, Phys. Rev. B 87, 115108 (2013). S. Albrecht, L. Reining, R. Del Sole, and G. Onida, Phys. Rev. Lett. 80, 4510
204 (1998).
W. Khan, S. B. Betzler, O. Šipr, J. Ciston, P. Blaha, C. Scheu, and J. Minar,
247
J. Phys. Chem. C 120, 23329 (2016). L. X. Benedict, E. L. Shirley, and R. B. Bohn, Phys. Rev. Lett. 80, 4514
205 (1998).
C. Ambrosch-Draxl and J. O. Sofo, Comput. Phys. Commun. 175, 1 (2006).
206 248
S. A. Khan, P. Blaha, H. Ebert, J. Minár, and O. Šipr, Phys. Rev. B 94, 144436 M. Rohlfing and S. G. Louie, Phys. Rev. Lett. 81, 2312 (1998).
249
(2016). G. Strinati, Phys. Rev. Lett. 49, 1519 (1982).
207 250
C. Spiel, P. Blaha, and K. Schwarz, Phys. Rev. B 79, 115123 (2009). G. Strinati, Phys. Rev. B 29, 5718 (1984).
208 251
K. Schwarz and P. Blaha, “Electronic structure of solids and surfaces with M. Rohlfing and S. G. Louie, Phys. Rev. B 62, 4927 (2000).
WIEN2k,” in Practical Aspects of Computational Chemistry I: An Overview of the 252
R. Laskowski and N. E. Christensen, Phys. Rev. B 73, 045201 (2006).
Last Two Decades and Current Trends, edited by J. Leszczyncski and M. K. Shukla 253
R. Laskowski and N. E. Christensen, Phys. Rev. B 74, 075203 (2006).
(Springer Science+Business Media B.V., Berlin, Heidelberg, 2012), Vol. 7, p. 191. 254
R. Laskowski, N. E. Christensen, G. Santi, and C. Ambrosch-Draxl,
209
C. H. Gardiner, A. T. Boothroyd, P. Pattison, M. J. McKelvy, G. J. McIntyre, Phys. Rev. B 72, 035204 (2005).
and S. J. S. Lister, Phys. Rev. B 70, 024415 (2004). 255
T. Das, X. Rocquefelte, R. Laskowski, L. Lajaunie, S. Jobic, P. Blaha, and
210
A. R. Williams, V. L. Moruzzi, J. Kübler, and K. Schwarz, Bull. Am. Phys. Soc. K. Schwarz, Chem. Mater. 29, 3380 (2017).
29, 278 (1984). 256
P. Tiwald, F. Karsai, R. Laskowski, S. Gräfe, P. Blaha, J. Burgdörfer, and
211
K. Schwarz and P. Mohn, J. Phys. F: Met. Phys. 14, L129 (1984). L. Wirtz, Phys. Rev. B 92, 144107 (2015).
212
R. Sternheimer, Phys. Rev. 80, 102 (1950). 257
See http://www.chem.pku.edu.cn/jianghgroup/codes/gap2.html for informa-
213
P. Blaha, K. Schwarz, and P. Herzig, Phys. Rev. Lett. 54, 1192 (1985). tion about GAP2.
214 258
K. Schwarz, C. Ambrosch-Draxl, and P. Blaha, Phys. Rev. B 42, 2051 (1990). H. Jiang, R. I. Gómez-Abal, X.-Z. Li, C. Meisenbichler, C. Ambrosch-Draxl,
215
M. Body, C. Legein, J.-Y. Buzaré, G. Silly, P. Blaha, C. Martineau, and F. and M. Scheffler, Comput. Phys. Commun. 184, 348 (2013).
Calvayrac, J. Phys. Chem. A 111, 11873 (2007). 259
H. Jiang and P. Blaha, Phys. Rev. B 93, 115203 (2016).
216 260
P. Dufek, P. Blaha, and K. Schwarz, Phys. Rev. Lett. 75, 3545 (1995). H. Jiang, Phys. Rev. B 97, 245132 (2018).
217 261
R. Laskowski and P. Blaha, Phys. Rev. B 85, 035132 (2012). H. Jiang, R. I. Gomez-Abal, P. Rinke, and M. Scheffler, Phys. Rev. Lett. 102,
218
R. Laskowski and P. Blaha, Phys. Rev. B 89, 014402 (2014). 126403 (2009).

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-29


© Author(s) 2020
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

262 274
H. Jiang, R. I. Gomez-Abal, P. Rinke, and M. Scheffler, Phys. Rev. B 82, 045108 H. Hagemann, A. Rief, F. Kubel, J. L. M. van Mechelen, F. Tran, and P. Blaha,
(2010). J. Phys.: Condens. Matter 19, 036214 (2007).
263 275
S. K. Panda, H. Jiang, and S. Biermann, Phys. Rev. B 96, 045137 (2017). S. Ehsan, A. Tröster, F. Tran, and P. Blaha, Phys. Rev. Mater. 2, 093610
264
M. Wallerberger, A. Hausoel, P. Gunacker, A. Kowalski, N. Parragh, F. Goth, (2018).
276
K. Held, and G. Sangiovanni, Comput. Phys. Commun. 235, 388 (2019). P. Blaha, D. J. Singh, and K. Schwarz, Phys. Rev. Lett. 93, 216403 (2004).
265 277
See https://github.com/w2dynamics/w2dynamics for information about J. M. Perez-Mato, M. Aroyo, A. García, P. Blaha, K. Schwarz, J. Schweifer, and
w2dynamics. K. Parlinski, Phys. Rev. B 70, 214111 (2004).
266 278
O. Parcollet, M. Ferrero, T. Ayral, H. Hafermann, I. Krivenko, L. Messio, and J. M. Perez-Mato, P. Blaha, K. Schwarz, M. Aroyo, D. Orobengoa, I. Etxebarria,
P. Seth, Comput. Phys. Commun. 196, 398 (2015). and A. García, Phys. Rev. B 77, 184104 (2008).
267 279
See https://triqs.github.io/triqs/latest for information about TRIQS. O. Rubel, A. Bokhanchuk, S. J. Ahmed, and E. Assmann, Phys. Rev. B 90,
268
K. Haule, C.-H. Yee, and K. Kim, Phys. Rev. B 81, 195107 (2010). 115202 (2014).
269 280
See http://hauleweb.rutgers.edu/tutorials/index.html for DFT + embedded P. M. C. Rourke and S. R. Julian, Comput. Phys. Commun. 183, 324 (2012).
DMFT Functional. 281
H.-Y. Yang, J. Gaudet, A. A. Aczel, D. E. Graf, P. Blaha, B. D. Gaulin, and
270
P. Hansmann, A. Toschi, G. Sangiovanni, T. Saha-Dasgupta, S. Lupi, M. Marsi, F. Tafti, Phys. Rev. B 98, 045136 (2018).
and K. Held, Phys. Status Solidi B 250, 1251 (2013). 282
See http://www.wien2k.at/acknowledgment for acknowledgment to the
271
K. Haule and G. L. Pascut, Phys. Rev. B 94, 195146 (2016). WIEN2k contributors.
272 283
K. Haule, J. Phys. Soc. Jpn. 87, 041005 (2018). See http://www.wien2k.at/reg_user/textbooks/usersguide.pdf for the WIEN2k
273
P. Giannozzi, S. de Gironcoli, P. Pavone, and S. Baroni, Phys. Rev. B 43, 7231 user’s guide.
284
(1991). See http://www.wien2k.at/reg_user/mailing_list for the WIEN2k mailing list.

J. Chem. Phys. 152, 074101 (2020); doi: 10.1063/1.5143061 152, 074101-30


© Author(s) 2020

You might also like