Computer Physics Communications 59 (1990) 399-415 399

North-Holland

FULL-POTENTIAL, LINEARIZED AUGMENTED PLANE WAVE PROGRAMS

FOR CRYSTALLINE SYSTEMS

P. BLAHA, K. SCHWARZ, and P. SORANTIN

Inst itut für Technische Elektroche,nie, Technische Universität Wien, A -1060 WIEN, Austria

and

S.B. TRICKEY
Quantum Theory Project, Depts. of Physics and of Chemistry, University of Florida, Gainesville, FL 32611, USA

Received 7 September 1989

In solids, linearized augmented plane waves (LAPW’s) have proven to be an effective basis for the solution of the
Kohn—Sham equations, the main calculational task in the local spin density approximation (LSDA) to density functional
theory. The WIEN package uses LAPW’s to calculate the LSDA total energy, spin densities, Kohn—Sham eigenvalues, and the
electric field gradients at nuclear sites for a broad variety of space groups. Options include retention or omission of
non-muffin-tin contributions (hence WIEN is a full-potential or F-LAPW code) and relativistic corrections (full treatment for
core states, scalar-relativistic for valence states).

PROGRAM SUMMARY

Title of program: WIEN CPC Program Library subprograms used: cat, no.: AALL
(included on distribution tape); title: MFFT; ref in CPC: 42
Catalog number: ABRE (1986) 233

Program obtainable from: CPC Program Library, Queen’s Separate documentation available: “User’s Guide”; no. of
University of Belfast, N. Ireland (see application form in this pages: 36 (also included on the distribution tape)
issue)
Keywords: crystals, linearized augmented plane wave (LAPW)
Computer: IBM 3090-400E/2VF, IBM 3090-400/2VF; Instal- method, full-potential LAPW (F-LAPW), energy band struc-
lation: Universität Wien, University of Florida, resp. ture, local spin density functional calculations, electric field
gradient, electronic charge (spin) densities, total energy
Operating system: IBM VM/XA (also IBM MVS/XA)
Nature of physical problem
Programming language used: FORTRAN 77 (IBM VS compiler Calculation of spin densities, total energy, Kohn—Sham energy
used) bands, and electric field gradients at nuclear sites, for various
local density approximations in a broad variety of crystalline
High speed storage required: for the choices of FORTRAN space groups, with or without relativistic corrections, with the
PARAMETER in the program versions on the distribution full potential or in the muffin-tin approximation.
tape, the address space required is 6.9 Mbyte
Method of solution
No. of bits in a word: 64; the code is written in single precision Kohn—Sham orbitals are expanded in a linearized augmented
(32 bits) but always compiled in double plane wave basis set to construct the generalized secular equa-
tion (i.e. including the basis set overlap matrix) from a starting
No. of lines in combined program and test deck: 78228 potential which is the superposition of atomic potentials trun-

0010-4655/90/$03.50 © 1990 — Elsevier Science Publishers B.V. (North-Holland)

400 P. Blaha et al. / Plane wave programs for crystalline systems

cated at the muffin-tin radii (radii of non-overlapping spheres symmetry, Hermitian matrices would be needed in place of the
centered at each nuclear site). Diagonalization yields the first real symmetric ones now assumed. Extension to structures
N eigenvalues and eigenvectors at each k-point in the irreduci- without rectangular basis vectors would require a consistent
ble wedge of the Brillouin zone. Spin densities are then representation in Cartesian coordinates throughout the pro-
constructed. From them new potentials (as spherical plus grams, so reprogramming analogous with the existing treat-
non-spherical parts in each muffin-tin sphere and as Fourier ment of hexagonal structures would be required.
series in the interstitial region) are obtained by a combination
of multipolar. Fourier, and numerical techniques to solve Typical running lime
Poisson’s equation. A new secular matrix is then generated. Li in the bce phase with 91 k-points in the irreducible wedge of
The cycle is repeated until self-consistency is attained, the Brillouin zone, maximum angular momentum parameter
Iterative stability is enhanced by mixing densities from one or L = 12. and the longest wave vector K~a~of the plane waves
more previous iterations. Both straight mixing and the Broy- chosen such that R, x Kr,,,,~ = 8.00 (with R, the muffin-tin
den-Il scheme are provided. Relativistic corrections can be radius), requires six iterations and a total of 766 seconds on a
included fully for core states and approximately (scalar-relativ- single processor of a base model IBM 3090-400/VF under
istic corrections) for valence states. Calculation in the muffin- MVS/XA.
tin approximation may be elected either for testing or for
comparison with older literature. Unusual features of the program
The package consists of six program modules among which
Restrictions on the complexity of the problem data can be routed according to choice of EXEC’s and files to
Cubic, tetragonal, orthorhombic and hexagonal space groups achieve whatever goal (e.g. core, semi-core, or valence state,
with inversion symmetry are implemented currently. Space total energy or not, etc.) the user has at that time. File editing
groups within these restrictions may be handled by construe- between modules is not required. All modules have been tested
tion of suitable input, the procedure for which is summarized extensively in the IBM 3090 VM and MVS environments and
in the User’s Guide. Treatment of groups of lower symmetry refined to utilize IBM XA virtual addressing and the Vector
would require recoding. To handle cases without inversion Feature for sustained high performance.

LONG WRITE-UP

I. Introduction to the problem and the solution Two approximations comprise the LSDA, first.
method the assumption that ~ can be written in terms of
a local exchange-correlation energy density
The state of the art for solution of the many- ji~ (n n ) as ~
~, f dr = ( n ~ )( ~ +
~,

electron problem in a crystal with nuclei at fixed n ), and second, the particular form chosen for
positions is the local spin density approximation that p.~(,.Several forms exist in the literature, for
(LSDA) to density functional theory (DFT) [1]. In example, Xs [2], Hedin—Lundqvist [3], Von Barth
it the key quantities are the spin densities n,,(r) in and Hedin [4], and Moruzzi, Janak, and Williams
terms of which the total energy is [5].

E 1 E
tot I. ~ 1’ ~ J 101 has
familiar a variational
Rayleigh—Ritz equivalent
principle. witheffec-
The most the
= T[n ~, n ] + Eee[fl i~ n ] + ENe[fl ~ tive way known to carry through that variation is
to introduce orbitals x~constrained to construct
+ E~~[n~, n + ENN, (1) the spin densities as

with ENN the repulsive coulomb energy of the ,~ (r) = ~n~Ix”,jr) 2, (2)
fixed nuclei, and the electronic contributions are ~.k
labeled conventionally as, respectively, the elec- 0
Here, the ~ are occupation numbers such that
tron—electron repulsion, nuclear—electron attrac- 7~k~ 1 /~‘.s’ (3)
tion, and exchange—correlation energies. 0 ~ ‘
 P. Blaha et al. / Plane wave programs for crystalline systems 401

and Wk is the symmetry-required weight of the (k is a reduced wave vector in the first Brillouin
point k. zone and the K0 are the reciprocal lattice vectors)
Variation of E10~then gives the Kohn—Sham which are determined by requiring that this basis
(KS) equation (in Rydberg atomic units; we have function matches (in value) the corresponding in-
omitted the obvious spin-labeling of the potential), terstitial region basis function at the sphere

[_~2+ VNe+ l’~e+~C]XIk(r) =~~~(r),(4) boundary.

(ii) In the interstitial region each ~ is simply a
solution of which is the primary computational plane wave
task.
In atoms it is feasible to solve eq. (4) by a ~(k0, r) = ~—I/2 exp(ik0 .r), (7)
direct numerical procedure and the same has re-
cently proven true for rather small molecules [6]. with Q the unit-cell volume.
For solids and films it is almost invariably the
case that expansion of ~ in a basis is the best The basis thus defined is energy-dependent be-
route to a solution. (The apparent exception is cause E in eq. (5) must in turn be equal ~‘k in eq.
KKR methods which use a scattering solution.) (4) for each band i at each k. In fact it is this
To achieve maximum efficiency, the practice since requirement which in the limit of an infinite ex-
the beginning of computational solids state physics pansion in APW’s would yield a basis with a
has been to design basis sets which are highly continuous radial derivative at each muffin-tin
adapted to the crystalline environment. Each such boundary.
basis generates its own specific technical demands, This energy dependence causes two significant
hence each with its concomitant techniques has computational difficulties. First, the APW secular
come to be known as an “energy band method”. equation is not the usual one, but a highly non-un-
Among the most broadly successful of what ear function of the energy which must be searched
might be termed “first-generation” energy band to determine its zeros at every point k on the BZ
methods was the augmented plane wave (APW) mesh. That search is a computationally intensive
scheme [7]. Most commonly it was implemented task, particularly as the number of atoms per cell
with a muffin-tin potential, i.e. one that is spheri- rises (as is the case with technologically significant
cally averaged in non-overlapping spheres centered crystals).
upon nuclear sites and volume-averaged to a con- The job is made more difficult by the second
stant in the interstitial region. problem, namely, that as a function of E the
In those two distinct regions different basis APW secular equation exhibits singularities
functions ~ are defined in the canonical APW: whenever a node of a u1 (r, E) falls on a sphere
boundary. Taken together these two problems are
(i) Inside atomic sphere t, of radius R1, each 4. sufficiently severe that they cannot be treated by
is a linear combination of radial functions u1 brute force application of overwhelming computer
times spherical harmonics Yim(?) (the index t is power to a crystal of even moderate complexity.
omitted when unneeded) The two difficulties are also a prohibitive bar-
rier to straightforward extension of the method to
q(k0, r) = ~AI~mUI(T, E)Yim(i~), (5) include the entire potential (the so-called non-
l,m
muffin-tin potential). It is easy enough to include
where u1 (r, E) is the regular solution of the the full potential in the interstitial region since the
radial Schrodinger equation for energy E in the plane wave basis enables matrix elements of a
spherical potential inside sphere t, obtained by Fourier representation of the full potential to be
numerical integration on a radial mesh in that calculated readily there [8]. (In the jargon this is
sphere, and the coefficients A7,,~are functions of called a warped muffin-tin potential.) The situa-
tion inside the spherical regions is quite another
k0=k+K~, (6) story.
402 P. Blaha et al. / Plane wave programs for crystalline systems

On the other hand, there is a great advantage of where u,(r, E1) is (as before) the regular solution
the APW basis. It is iteratively adaptive, that is, at of the radial Schrodinger equation for the spheri-
each iteration of a self-consistent cycle a new set cal part of the potential in sphere I, but now for
of u,(r, E) is generated with respect to the input fixed energy E,. The second set of radial func-
potential for that iteration. Iterative adaptation tions, which appears in the LAPW but not the
has the striking benefit of providing much more APW, consists of the energy derivatives à,(r. E,)
rapid evolution (with respect to number of itera- = au,(r, E)/0E I L~ The coefficients A,,,, B,,~are
tions) of the scf cycle to consistency than is ob- determined by requiring that this basis function
tamed with a fixed basis, e.g. a Gaussian orbital matches the corresponding interstitial region basis
basis. For example, in one extreme case [9] a function at the sphere boundary in value and in
Gaussian code could not achieve self-consistency slope.
when started from superposed atomic densities,
but could do so when started from a self-con- As with the original APW basis, the radial
sistent APW solution which had itself been started functions u1 (r, E,) and à,(r, E,) are obtained by
from superposed atomic densities. Nevertheless, numerical integration on a radial mesh inside each
by the mid-1970’s the cumulative burden of the sphere. However, the E, parameter in the LAPW
difficulties with the canonical APW outweighed is not required to match the KS eigenvalue ~‘k but
this and other advantages sufficiently that it was is fixed instead to an expected value, for example.
well on its way to disuse. to be roughly in the middle of the occupied eigen-
Almost simultaneously, Andersen [10] intro- values which are predominantly of angular
duced the notion of linearization. The idea, which momentum type 1. The empirical Wigner—Seitz
actually traces to a widely unnoticed suggestion of rule [10] enables automated determination of E,
Marcus [11], already had been explored briefly by for those bands. The rule is that the energies of
Koelling [12] as an approximation scheme for such bands range between the values of E, for
accelerating canonical APW calculations. Al- which u(R1, E,) = 0u,(r, E)/ar I R~L, = 0 and
though Andersen’s work has come to be known u,(R~, E,) = 0. Thus, one can code a determina-
best in connection with the LMTO method, he in tion of the change of sign in value and slope of the
fact did contrive a linearization of the APW radial wave functions and then take the arithmetic
method and noted a happy result. The technical mean as the value of E1. This procedure works
barriers of the canonical APW scheme which we especially well for semi-core states which are im-
just summarized are eliminated. Koelling and portant during the first iteration cycles, when the
Arbman [13] took up the concept immediately, position of their narrow bands may change drasti-
refined it (primarily by omitting an expansion cally in energy.
technique Andersen had used) to resemble closely Region (ii) is treated just as in the original
what today is called the linearized APW (LAPW) APW method so 4 (k0, r) is given by eq. (7). The
method, and programmed it by modifying Koell- solutions to the KS equation are expanded in this
ing’s existing APW code. combined basis set in the usual linear fashion
The LAPW and APW bases differ only within
the spherical regions. Thus the LAPW replaces ~~(r) = ~c~(a, t. k)4(k0, r). (9)
possibility (i) above with:
with the coefficients determined by the Rayleigh—
(i) Inside atomic sphere I, of radius R,, each ~ Ritz variational principle.
is a linear combination of radial functions times The critical difference between the LAPW and
spherical harmonics Y~m(~ APW secular equations is that the LAPW secular

~(k0, r) = ~ [A,,,,(kju,(r, E,) equation is not energy dependent because the

LAPW basis is not. The LAPW basis removes this
dependency by fixing E1, then compensating for
+B,0,(k,,)t~,(r, E,)] YIO~)’ (8) the resulting loss of variational flexibility by inclu-
 P. Blaha et al. / Plane waveprograms for crystalline systems 403

sion in the basis of a systematic estimator (to 2. Summary of program structure

linear order) of the energy dependence, namely ii,,.
The adaptive character of the canonical APW is This section gives an overview of the structure
not lost by this procedure. of the programs, including their key features, re-
With an energy-independent basis, the barriers quirements, and limitations. Details of use and
to treatment of the full potential are also removed, operation are found in the User’s Guide.
In its general form the LAPW method can treat WIEN consists of several independent pro-
straightforwardly a potential expanded in the form grams described in more detail below. It is con-
venient, however, to have in mind a typical exam-
ple of their use. Thus, LSTART is a modified
V( r) = ~ VLM (r) ~LM (~)~
inside sphere, (relativistic) atomic LSDA code used to generate
L,M

starting potentials. From these a self-consistency

= ~ V( K) exp(i K r), outside sphere, cycle is pursued until certain convergence criteria
K (discussed in the User’s Guide) are satisfied. After
(10) completion of one cycle, which uses LSTART in
lieu of LAPWO, the iterative loop is:

with the spin (charge) densities expanded in the 1. enter with density;
same fashion. Thus no shape approximations are 2. LAPWO generate potential from density;
—

3. LAPWI compute bands (eigenvalues and ei-

—

required in general. When implemented in this

fashion the procedure is frequently called a “full- genvectors);
4. LAPW2 compute valence density from eigen-
—

potential LAPW” (F-LAPW) method. WIEN is,

in this nomenclature, an F-LAPW package. The vectors;
muffin-tin approximation corresponds to retaining 5. CORE compute core states and densities;
—

only the L = M = 0 component inside the spheres 6. MIXER mix input and output densities;
—

and only the K = 0 component outside. This op- 7. loop to 2.

tion exists in WIEN both for comparison with Here the usually intuitive distinction between
older calculations and for testing. core and valence states is made quite quantitative:
In a manner similar to the inclusion of the full core states are those whose charge is confined
potential, it is rather straightforward to include entirely to an atomic sphere. Frequently it is also
relativistic effects with the LAPW basis, at least in useful to distinguish a third type of states called
the so-called scalar-relativistic approximation [14]. semi-core. While these lie substantially below the
That is, for valence states spin-orbit coupling is Fermi energy, they are far enough above the core
omitted. Since core states are treated in atomic-like states that their charge is no longer completely
fashion (see next section), full relativistic correc- inside a sphere, though only a few percent is
tions for them can be obtained straightforwardly. outside. WIEN allows the treatment of semi-core
The formalism and techniques for the F-LAPW states straightforwardly (essentially as a distinct
method have, to our knowledge, never been re- class of band states). Details are in the User’s
viewed in one place. WIEN utilizes analyses and Guide. Since they are mostly immaterial to the
methods found in diverse references, including the present discussion, semi-core states will be men-
work of Andersen, of Koelling, and of Koelling tioned only rarely in what follows.
and Arbman already cited, as well as that of WIEN also supports spin-polarized calculations
Wimmer et al. [15], Weinert [16], Weinert et al. straightforwardly by simple replication of some
[17], Blaha and Schwarz [18], Jansen and Freeman steps for spin-up and spin-down electrons sep-
[19], Blaha et al. [20], Wei et al. [21], and Mat- arately. The procedure (again after an initial pass
theiss and Hamann [22]. We have summarized with LSTART) then becomes:
only the basic techniques with details left to these
references. 1. enter with spin densities;
404 P. Blaha et al. / Plane wave programs for crystalline systems

2. LAPWO — generate potentials from spin den- unit-cell edge-lengths; the relative position of the
sities; atoms in the unit-cell; the explicit symmetry-al-
3. LAPWI — compute valence bands for spin lowed decompositions of /m-like charge; the name
up; of each atom; the number of radial mesh points
4. LAPW1 — compute valence bands for spin on which the radial Schrodinger equation will be
down; solved in that atom’s sphere; the position of the
5. LAPW2 — compute valence density for spin first radial mesh point RO [on a logarithmic mesh
up; such that mesh points R(N) are given by R(N) =
6. LAPW2 — compute valence density for spin R0*EXP((N 1)*DX) with DX fixed internally];
—

down; R
1 = the atomic sphere radii (i.e. muffin-tin radii);
7. CORE compute core states and densities for
— Z = atomic number; a set of local rotation matrices
spin up; which transform the general coordinate system
8. CORE compute core states and densities for
— into a local one for that site which explicitly
spin down; reveals the local point group symmetry and thereby
9. MIXER mix input and output densities
— exactly cancels certain combinations of spherical
10. loop to 2. harmonics [23]; the number of symmetry oper-
ations of the space group; a series of listing of
WIEN is designed to utilize systematic file matrices of space group symmetry operations and
naming for control of program flow and passage associated non-primitive translation vectors and
of results among those programs. A consistent the indices of those symmetry operations.
naming convention for files is therefore given in Each program also has its own input file CASE
the User’s Guide. Here we will use only the input lNx A with x a systematic program specification
and output portions. In IBM VM/CMS, file names and each generates its own output as CASE OUT-
are specified by an unpunctuated triple fit ft fm PUTx D. However, the most useful file during the
which are filename, filetype and filemode, respec- scf cycle is the iteration history maintained in
tively. In the WIEN file-naming convention, fit is CASE SCF D. Because this file is updated at each
always reserved to label the case under study, iteration by appending the new information, the
hence we shall use CASE for the generic fit. A approach to self-consistency can, for example, be
specific solid, e.g. hcp Zn, then has HCPZN for its monitored conveniently. The test run output at the
fit ‘s. The filetype ft is reserved for description of end of this paper provides an abridged example of
the file content and function, hence the ft of type the actual SCF file for NaC1.
VSP is for a potential V which is SPherical, while For historical reasons of compatibility with
the ft INM signals INput for the program other operating systems, it has proven convenient
MIXER. The filemode fm distinguishes whether a to read CASE SCF into memory at each iteration,
file is defined initially for input or output, A and append the new results, write the updated file
D respectively. Thus the input to MIXER for hcp under a new (temporary) name, erase the original
Zn is found in the file HCPZN INM A. With one version, and rename the new (updated) version to
exception (the CASE SCF file discussed below), the original name. Thus, in the EXECs on the
non-temporary files with filemode D are not over- distribution tape and in the User’s Guide, CASE
written. Frequently they are read as input to steps SCF D seems to violate the convention that fm =
subsequent to the one in which they were gener- D is never rewritten in a given iteration, but this is
ated. only because of the need to have a simple, porta-
All six programs use a master input file CASE ble updating scheme for this file.
STRUCT A described in detail in the User’s Table 1, adapted from the User’s Guide, lists
Guide. It defines the data common for that CASE the conventional file names for the files required
to the entire calculation. These include: title; (in addition to CASE STRUCT A and CASE INx
lattice type; number of inequivalent atoms; A) for each program step. For simplicity, a non-
whether the calculation is to be relativistic; the spin-polarized calculation is assumed.
 P. Blaha et al. / Plane wave programs for crystalline systems 405

Table 1
Conventional file names for input and output at each program step. Input files in parentheses are optional, while output files in
parentheses are generated when requested by an input switch

Program Input files (fm = A) Output files (fm = D)

fn ft fn ft
LSTART CASE INST CASE OUTPUTST
VSP

LAPWO’0 CASEI CLMSUM CASE1 vsp

VNS
(R2V, COUL)
SCFO to SCF

LAPWI CASE VSP CASE VECTOR

(VNS)
SCF SCFI to SCF

LAPW2 CASE VSP CASE CLMVAL

(CLMSC)
VECTOR HELP3I, HELP32,
(QTL, Fl 5 temporary)
SCF SCF2 to SCF

CORE CASE VSP CASE CLMCOR

SCF SCFC to SCF

MIXER CASE CLMVAL CASE1 CLMSUM

(CLMSC), (CLMCOR), (CLMSUM)
(BROYDI) (BROYDI)
(BROYD2) (BROYD2)
SCF SCFM to SCF
a) If LAPWO completes successfully, the CASE ft D are erased and then the CASEI ft D are renamed to CASE ft D with ft being all

relevant filetypes.

2.1. LSTART is implicitly set to zero in the interstitial region.

Sometimes it is necessary to rerun LSTART with
LSTART is a modified version of a relativistic a different choice of shift in order to achieve this
atomic LSDA code written by Desclaux [24]. It is goal. When completed, however, this procedure
used to generate atomic potentials truncated at yields a crude but surprisingly effective muffin-tin
corresponding muffin-tin radii R,. The advantage starting potential.
of this technique for generating a starting poten- The output from LSTART is a file containing
tial, as compared with the more familiar superpos- the starting muffin-tin potential. The file name,
ing of atomic charge densities, is that no lattice recall table 1 above, is CASE VSP D. VSP denotes
sums are needed. Even quite complicated crystal the SPherical part of the potential V.
structures can therefore be handled straightfor- LSTART utilizes the following procedures
wardly and swiftly. which perform the indicated roles:
Usually the truncated potentials need shifting
by an additive constant (provided as input) in MAIN program manages iterations, mixing of
—

order to achieve a physically sensible starting densities, calculates shifted potentials (also total
potential. In this context “physically sensible” energies, radial moments and other quantities un-
means a potential which at R~has a value from used by WIEN).
about —0.2 to —1.0 Ry. The motivation for this SUBROUTINE INOUH establishes initial val-
—

choice is that in the muffin-tin case the potential ues for outward integrations.
406 P. Blaha et al. / Plane wave programs for crystalline systems

SUBROUTINE INSLD handles input data and

— basis there). That Fourier series represents some
starting values for potentials. incorrect density when continued into the spheres.
SUBROUTINE INTH —five point (Adams) in- So, the multipole moments of the proper
tegration used by subroutine RESLD for outward density and those Q~of the improper (plane-
integration, wave) density inside the spheres are calculated
SUBROUTINE POTSL generates KS potential.
— and moments of the difference density then result
SUBROUTINE RESLD Dirac—Slater equation
— by simple subtraction, Qdff = Q~.For
—

solver (Numerov method). these, a second plane-wave density is generated

SUBROUTINE SOMM Simpson’s rule integra-
— which both reproduces the Qd)f and has zero
tion. density in the interstitial region. When the second
FUNCTION DALP — Pratt mixing for iterative series is added to the original (interstitial) Fourier
convergence, series for the density. the sum has two useful
FUNCTION EPSXC —spin-polarized Hedin— properties simultaneous/v. It is a correct expansion
Lundqvist [3a] kernel j.t~. of the interstitial charge and it reproduces the
FUNCTION FPOT — Thomas— Fermi—Dirac multipole moments of the charge in all the spheres.
potential. Therefore the interstitial Coulomb potential fol-
FUNCTION VXCSP — spin-polarized Hedin- lows immediately, essentially by dividing by K2,
Lundqvist [3a] V~. Inside each sphere, LAPWO generates J’~
0 with a
straightforward classical Green’s function solution
The input to LSTART consists first of a title of the spherical boundary value problem.
and the number of orbitals. Three parameters for As with most7~~ other
on themethods, the non-linear
density means that V
each orbital follow, the principle quantum num- dependence of J 5~
ber, the relativistic quantum number and the oc- must be obtained from values computed on a grid.
cupation number per spin. (Definition of the Inside the atomic spheres a least squares proce-
quantum numbers is given in the User’s Guide.) dure is used to represent V~ in lattice harmonics.
Finally, there is the potential shift for that atom, with the fitting equations solved via LINPACK
For unit-cells with more than one atom, input routines (which are included on the distribution
must be provided for each inequivalent atom, tape). The radial grid in a given sphere is the same
Because the starting potential need not be very one used to solve the radial Schrodinger equation.
refined, LSTART does not have an option for while the angular mesh utilizes 173 points posi-
choosing which LSDA model to use nor whether tioned empirically to achieve the best possible fit
to choose the spin-polarized or non-spin-polarized within fixed maxima of L, M. In the interstitial
case. It simply gives the spin-polarized Hedin— region a 3-dimensional fast Fourier transforma-
Lundqvist [3a] case. When starting a non-spin- tion [25] is used to generate a Fourier representa-
polarized calculation, the extra potential is dis- tion of V,,~.The interstitial grid is the real space
carded. mesh which is reciprocal to the set of wave vectors
in the plane-wave expansion.
2,2, LAPWO To facilitate the computation of the contribu-
tion to the Hamiltonian matrix elements arising
If the calculation is spin-polarized (non-spin- from the potential in the interstitial region,
polarized), LAPWO uses the input spin densities LAPWO also calculates the so-called reanalyzed
(total electron density) to±compute
1~iOU V~. As inJ’~uI= VNe +
eq. 4) we Fourier series
consists for ~7~)t inthethat
of constraining newregion. Reanalysis
Fourier series to
ce’ V~ and
have omitted the spin-labelling of the potentials.
‘~)t = represent a total potential with the following be-
~ is calculated via a multipolar Fourier cx- havior:
pansion of the density [16]. The density in the V~~~
0(r)
= E~1(r) for r in the interstitial region,
interstitial region is readily generated as a Fourier
series (simply because of the form of the LAPW = 0, for r in any sphere. (11)
 P. Blaha et al. / Plane wave programs for crystalline systems 407

Since LAPWO manipulates the spin densities SUBROUTINE YLM — generates spherical
(or total density) it is convenient to include in harmonics.
LAPWO the contribution to E10~ which arises from SUBROUTINE SPHBES generates spherical
—

integrals with integrands of the form density- Bessel functions.

times-potential. These are calculated as in ref. [17] SUBROUTINE SUML combines spherical
—

if the option to do so is turned on. Time can be harmonics into crystal harmonics.
saved in early iterations by leaving the option SUBROUTINE STERN generates star of re-
—

turned off. ciprocal lattice vectors K.

The electric field gradient (EFG) at nuclear SUBROUTINE REANO initialization for re-
—

sites with non-cubic symmetry is computed in analysis of the Fourier series for ~
LAPWO. The calculation is based on the recogni- SUBROUTINE REANI reanalysis of the Four-
—.

tion that the EFG is given directly by the L = 2, icr series for V~a.
M = 0,2 components of the potential expansion in SUBROUTINE POISSN for testing; inverts
—

the limit that the nuclear center is approached. Poisson’s equation, i.e. generates density from
The EFG is therefore readily calculable in the given potential.
LAPW. Details of the analysis are given in refs. SUBROUTINE IFFLIM sets fast Fourier
—

[20,29,30]. transform (FFT) parameters.

For the spin-polarized case. LAPWO generates SUBROUTINE CPUTIM gets CPU time. —

the spherical part (L = M = 0) of the spin-up FUNCTION XCENER generates ~

— the cx-
potential in CASE1 VSPUP D and the spin-down change-correlation energy density.
potential in CASEI VSPDN D. The correspond- FUNCTION XCPOT generates V~. —

ing non-spherical contributions are in CASE1

VNSUP D and CASE1 VNSDN D. For the un- In addition, LAPWO utilizes the fast Fourier
polarized case the spherical part is in CASE1 VSP transform package [25] MFFT (CPC catalog no.
D and the non-spherical part is in CASE1 YNS D. AALL) and the following Linpack routines:
LAPWO utilizes the following procedures which SUBROUTINE SSICO,
perform the indicated roles: SUBROUTINE SSISL,
SUBROUTINE LATGEN defines lattice (basis
- SUBROUTINE SAXPY,
vectors). SUBROUTINE SSCAL,
SUBROUTINE ROTDEF selects symmetry op-
- SUBROUTINE SSIFA,
erations of equivalent atoms. SUBROUTINE SSWAP,
SUBROUTINE ROTATE rotates a vector by
- FUNCTION SDOT,
matrix. FUNCTION SASUM,
SUBROUTINE MULTFC combines multipole
- FUNCTION ISAMAX.
moments.
SUBROUTINE POTFAC factors for lattice
— These are used to solve the system of linear equa-
harmonics. tions which results from the least squares proce-
SUBROUTINE XCPOT1 generates V~inside a
— dure for fitting the exchange-correlation potential
sphere. in the spherical regions. On the distribution tape,
SUBROUTINE SUMFAC combines spherical
— both MFFT itself and the Linpack routines are
harmonics into cubic harmonics, included as source code in the FORTRAN mod-
SUBROUTINE XCPOT3 — generates V,,~in the ule labeled as MFFT.
interstitial region. The input to LAPWO consists simply of a
SUBROUTINE ENERGY calculates contribu-
— keyword which selects whether I’~ and E~0~ or
tions to E101 of the form of integrals of density- only ~‘ç~will be calculated (or, for testing pur-
times-potential. poses, ~ouI’ V,~,or multipole moments), a numeri-
SUBROUTINE CHARGE Simpson integration
— cal switch to select among LSDA models, and
inside sphere. some options for selecting or suppressing output.
408 P. Blaha et al. / Plane wave programs br crystalline systems

2.3. LAP Wi potential with respect to two radial functions.

SUBROUTINE SELECT selects energy param- -

In LAPW1 the matrix H of the KS Hamilto- eter E, via a Wigner—Seitz rule.

nian (4) and the overlap matrix S are constructed SUBROUTINE SETWAR generates Fourier —

[13] and the lowest N eigenvalues and eigenvectors coefficient of non-muffin-tin potential in intersti-
are obtained (in a predefined energy window) with tial region for all members of star of K from the
standard EISPACK routines [26]. These data are coefficient for K itself.
written to CASE VECTOR D. If the file CASE SUBROUTINE STERN generates star of K.--

VNS D exist (i.e. there are non-spherical terms in SUBROUTINE DVBES1 calculates derivatives
—

the potential), a full-potential calculation is per- of spherical Bessel functions.

formed automatically. SUBROUTINE LGNDRI Legendre polynomi-
—

LAPW1 itself is a very brief driver program, als.

with the actual construction and solution of the SUBROUTINE OUTWIN outward integration —

generalized secular problem performed in SUB- of radial Schrodinger equation.

ROUTINE EBND. Together they utilize the fol- SUBROUTINE RINTI3 Simpson integration
—

lowing procedures which perform the indicated inside the sphere.

roles: SUBROUTINE CBCOMB combines cubic -

harmonics.
SUBROUTINE TAPEEN writes energies for
- SUBROUTINE CPUTIM gets CPU time. -

band structure plot. SUBROUTINE SECLR4 solves secular equa-

—

SUBROUTINE LATGEN defines lattice (basis

— tion.
vectors). SUBROUTINE REBK33 Eispack. -

SUBROUTINE ROTDEF selects symmetry op-

- SUBROUTINE REDC33 Eispack. -

erations of equivalent atoms. SUBROUTINE GATHER Eispack. —

SUBROUTINE ROTATE rotates vector by ma-

— SUBROUTINE TRBAK3 Eispack. -

trix. SUBROUTINE TINVIT Eispack. -

SUBROUTINE TAPEWF writes Kohn—Sham

- SUBROUTINE BISECT Eispack. -

eigenfunctions (VECTOR file). SUBROUTINE TRED3 Eispack. —

SUBROUTINE YLM generates spherical

— FUNCTION GAUNT1 calculates Gaunt num-
—

harmonics. bers.
SUBROUTINE COORS generates reciprocal
— FUNCTION WARPIN adds non-muffin-tin
-

vectors K for each k in the irreducible wedge of term from interstitial region to H (NB. these are
the BZ such that complete stars are retained and commonly called “warping” terms).
1k + K I KMAX. FUNCTION NEBLWI Eispack. -

SUBROUTINE DEFYLM calculates spherical

— FUNCTION SDOT dot product. —

harmonics Yjm (Jc~)multiplied by a phase factor.

SUBROUTINE CBMTX3 sets up H- and S-
— The input to LAPWI consists first of a keyword
matrices, which selects output options. Next is the parame-
SUBROUTINE GAUNT2 initialization for
— ter RKMAX defined by RKMAX = R1K0~0~.
computation of Gaunt numbers. which determines the length of the plane wave
SUBROUTINE WFPNT prints Kohn—Sham
— expansion. hence the dimension of the H- and
eigenfunctions. S-matrices. This is followed by LMAX, the maxi-
SUBROUTINE USTPHX step function.
— mum /-value used inside atomic spheres. There
SUBROUTINE SPHBES spherical Bessel func-
— follows a series of entries which set the values of
tions. E, and provide an option for automatic de-
SUBROUTINE ATPAR — reads V and generates termination of those values (in SELECT) by use
u1(R,, E,), à,(R,, E,), u,’(R~,E,) = 8u,(r, E)/ of the “Wigner—Seitz rule” [10]. The remaining
I R,. E,’ and matrix elements of the non-spherical input defines the name, components, and relative
 P. Blaha et a!. / Plane waveprograms for crystalline systems 409

weight of each k-vector in the irreducible wedge of SUBROUTINE RINT13 — Simpson integration
the BZ, along with the limits of the energy window inside the sphere.
in which the search for eigenvalues will be done at SUBROUTINE HARMON — transforms K ‘s and
that vector. A utility, KGEN, is provided to gen- calls Bessel routines.
crated the k-mesh if desired; see section 2.7. SUBROUTINE DVBES1 — derivatives of spheri-
cal Bessel functions.
2.4. LAPW2 SUBROUTINE SPHBES — spherical Bessel func-
tions.
SUBROUTINE FOURIR - charge density Four-
In LAPW2 the eigenvalues and eigenvectors
(solutions to the KS equations) found by LAPW1 ier coefficients.
SUBROUTINE STERN - generates star of K.
are read from the file CASE VECTOR D, then
SUBROUTINE FERMI - determines Fermi en-
used to calculate the Fermi energy and the expan-
sions of the spin (charge) densities analogously ergy.
SUBROUTINE CPUTIM gets CPU time. -
with eq. (10) above. This is done for each occupied
FUNCTION GAUNT Gaunt numbers. -
state and at each k in the BZ scan. From those FUNCTION T3JO Gaunt numbers.—
expansions the corresponding (partial) charges are
FUNCTION T3J Gaunt numbers.
-
obtained by integration and written to files CASE
HELP31 D, CASE HELP32 D etc., where the last FUNCTION PH Gaunt numbers.
-

FUNCTION NOTRI Gaunt numbers. -

digit gives the index of inequivalent atoms in the
FUNCTION KDELTA checks whether K is -
unit cell.
member of star.
The primary output of LAPW2, however, is the
file CASE CLMVAL D for a valence calculation Input to LAPW2 consists first of a switch which
or CASE CLMSC D for a semi-core calculation. determines whether the total charge density cx-
Both contain the expansion coefficients for the pansion inside and outside the spheres or only
relevant density (spin density), again of the same certain coefficients (primarily for testing) will be
form as shown for the potential in eq. (10), above, computed. The decomposition of contributions to
LAPW2 utilizes the following procedures which the electric field gradient at the nucleus may also
perform the indicated roles: be selected (recall that the total electric field

SUBROUTINE LATGEN defines lattice (basis

— gradient was computed in LAPWO). The upper
and lower energy cutoffs for defining the range of
vectors).
SUBROUTINE ROTDEF selects symmetry op-
— occupied states follow. Then comes a parameter
for the number of electrons per unit-cell which fall
erations of equivalent atoms. in that range (e.g. the number of valence dee-
SUBROUTINE ROTATE rotates vector by ma-
-

trix. trons) and a parameter for the number of bands

for which partial charges are reported. The re-
SUBROUTINE YLM spherical harmonics.
-

SUBROUTINE CHARGE Simpson integration

— maining important input is the list of L, M values
inside sphere. for the lattice harmonics expansion, defined
according to the point symmetry of the particular
SUBROUTINE L2MAIN — calculates charge
density inside sphere. atom consistent with the local rotation matrices
defined in CASE STRUCT A.
SUBROUTINE P3SPLT splits p-like charge
—

into symmetry components.

SUBROUTINE DSSPLT splits d-like charge
- 2.5. CORE
into symmetry components.
SUBROUTINE OUTP prints partial charges
— CORE is another modified version of the Dc-
(QTL). sclaux [24] relativistic atomic LSDA code. Its sole
SUBROUTINE ATPAR reads V and generates purpose is to solve for the relativistic core states of
u,(R,,E
1), u,(R~,E1) and u(R,,E,). the spherical part of the total potential at the
410 P. Blaha eta!. / Plane wave programs for crystalline systems

current iteration, hence it delivers core eigenvalues etc.), any existing BROYDn files must be deleted
and corresponding core densities in the file CASE since the iterative history which they retain refers
CLMCOR D. to a different, incompatible calculation.
In addition to a main program substantially E
1~,1 is computed in MIXER (from data in the
identical with that for LSTART, CORE contains CASE SCF file) by adding the contributions
the procedures SUBROUTINE INOUH, SUB- calculated in preceding stages of the same itera-
ROUTINE POTSL, SUBROUTINE RESLD, tion. This causes a potentially puzzling anomaly
SUBROUTINE INTH. SUBROUTINE INSLD, for novice users if the time-saving option of not
SUBROUTINE SOMM. FUNCTION DALP, and computing the density-times-potential-integrals in
FUNCTION FPOT. Their functions are the same LAPWO has been elected (recall above), since Un-
as indicated above in the discussion of LSTART. der those circumstances the total energy reported
Input to CORE parallels that for LSTART and by MIXER clearly is wrong. The savings of CPU
consists of the number of orbitals, and the princi- time is sufficient in our view to warrant this slight
pIe quantum number, relativistic quantum number risk of confusion.
and occupation number for each. Definition of the The key output file from MIXER for beginning
quantum numbers is given in the User’s Guide, the next iteration with LAPWO is CASE1
CLMSUM D, which contains the expansion coef-
2.6. MIXER ficients of the mixed densities (and a pair of
analogously named files for spin-up and down-
In MIXER the calculated electron densities densities in the spin-polarized case).
(core, semi-core and valence as relevant) are corn- MIXER utilizes the procedures listed below to
bined into a new total output density. As is well perform the indicated functions:
known, simply combining the new densities leads
to instabilities in the iterative scf process which SUBROUTINE CHARGE - Simpson integration
can be prohibitively severe. Typically these are inside sphere.
SUBROUTINE QMIX5 — Broyden-Il routine.
dealt with by some form of iteration—iteration
FUNCTION EQIVAL - compares two K vec-
mixing. In WIEN this takes the form of density
tors.
mixing. Two schemes are provided, straight mix-
ing as originally discussed by Pratt [27] and the Input to MIXER consists first of a simple
Broyden-Il scheme as put forth by Singh et al. switch between straight and Broyden mixing. Then
[28]. comes an option used for restarting the scf cycle
Straight mixing simply takes fixed fractions of with new parameters which, if set, automatically
old and new density, with the fraction stipulated invokes straight mixing and deletes the former
from input. At the outset (zeroth iteration), no Broyden files (in a subsequent iteration the CON-
mixing is possible, equivalent to straight mixing TINUE option may be elected and the Broyden
with a factor of unity. MIXER detects this situa- scheme turned back on). Finally the value of the
tion automatically by checking for the absence of straight mixing factor ~ is read, where
the file CASE CLMSUM D as the signal.
In Broyden-Il, information about all the expan- = (1
n0~~ — Q)n,d + ~ (12)
sion coefficients of the density from preceding
and the subscripts “old” and “current” label the
iterations is used to calculate an optimal mixing
densities which came into the current iteration
fraction for each coefficient at each iteration,
Therefore, in the first two iterations MIXER is- and were generated by that iteration, respectively.
sues a warning to the effect that the iteration
history files (“Broyden files”) do not exist, 2.7. Uttltties
At the outset of a new or changed calculation
(for any changed computational parameter such as Two small utility codes are provided to facili-
lattice constants, k-mesh, matrix size, LMAX, tate preparation and checking of input. NN uses
 P. Blaha et al. / Plane wave programs for crystalline systems 411

CASE STRUCT A to define the atomic positions energy range being searched falls between two
in the unit cell, then calculates the nearest-neigh- states of the same /-value but different principle
bor distances and checks that the corresponding quantum number for a single atom. The User’s
atomic sphere radii R1 are not overlapping. In Guide also contains instructions for using the
case of an overlap, an error message alerts the EXEC’s provided on the distribution tape to run
user. The nearest-neighbor distances are recorded the eight test cases found there. Those test cases
in CASE OUTNN D as well. NN is run before include metallic, ionic and Van der Waals crystals,
starting a new case as a check on the input speci- spin- and non-spin-polarized calculations, and
fied in CASE STRUCT A. several different space groups.
KGEN generates the k-mesh in the irreducible References are given in the User’s Guide to
wedge of the BZ on a grid which can be used in a published work on each of the test cases. Some
tetrahedron integration scheme. Run interactively examples of calculations with WIEN at various
with an input that specifies the desired lattice and stages of its evolution include refs. [20,29—33].
number of k-increments, KGEN generates output In its present form, the optimization of WIEN
which can be copied straightway into the input of is for IBM 3090 systems with the Vector Feature.
LAPW1 as CASE IN1 A or CASE IN1S A. The The code makes extensive use of virtual address-
lattices treated are all cubic ones, simple tetrago- ing. Substantial testing for stability, convergence,
nal, simple orthorhombic, and hexagonal ones. etc. of the algorithms has been done in that system
environment. We have not explored parallelization
nor have we removed coding which might be
3. Brief summary of the distribution tape useful on other systems or might aid user under-
standing by indicating historical developments.
The distribution tape contains: We intend to continue to refine and improve
1. listing of contents by conventional file names WIEN as time and opportunity permit. Sugges-
as described above; tions from users would be particularly welcome.
2. the User’s Guide;
3. the seven primary FORTRAN sources
(LSTART, LAPWO, MFFT, LAPW1, LAPW2, Acknowledgements
CORE, MIXER);
4. the FORTRAN sources for the two utilities We thank Dr. Bruce Harmon for providing KS
(NN,KGEN); and PB with an early version of a muffin-tin
5. IBM VM EXECS for compilation and oper- LAPW program (originally written by Koelling
ation; and Arbman) which was used as a starting point
6. IBM MVS JCL for compilation and operation; for the development of the system presented here.
7. the input and output files (23—35 per case) for We gratefully acknowledge help in improving the
the eight test cases, grouped by case in the WIEN code by Dr. Claudia Ambrosch-Draxl of
same order as they are presented in the User’s the Technische Universität Wien and Drs. Garry
Guide. Bamford and Jan Nobel of the University of
Altogether there are 264 files on the distribution Florida.
tape. This work was made possible in part by sup-
port from the IBM Corporation through a con-
tract with the University of Florida. Additional
4. Concluding remarks support for KS and PB was provided by the
University of Florida and for SBT by the Max
The User’s Guide includes a detailed discussion Planck Gesellschaft. We thank Provost Dr. G.W.
of the most frequent problem area in an LAPW Hemp for the former and Prof. Dr. G.H.F. Di-
calculation, the so-called ghost bands. These are ercksen for the latter. Most of the original pro-
spurious eigenvalues which can arise when the gram development was done at the Interuniversi-
412 P. Blaha et a!. / Plane wave programs for crystalline systems

thres EDV-Zentrum Wien. Recent development [10] O.K. Andersen, Solid State Commun. 13 (1973) 133;
and applications also have benefited from support Phys. Rev. B 12 (1975) 3060.
[11] P.M. Marcus, Int. J. Quantum Chem. Symp. 1(1967) 567.
by IBM-Austria through the EASI project with 1121 D.D. Koelling, J. Phys. Chem. Solids 33 (1972) 1335.
the EDV-Zentrum der Universitht Wien. Part of [13] D.D. Koelling and GO. Arbman. J. Phys. F 5 (1975)
this work was also supported by the “Fonds zur 2041.
Forderung der wissenschaftlichen Forschung” [141D.D. Koelling and B.N. Harmon. J. Phys. C 10 (1977)
grants P5939C and P7063P. 3107.
[15] E. Wimmer, H. Krakauer, M. Weinert and A.J. Freeman.
Phys. Rev. B 24 (1981) 864.
References [17]
[16] M. Weinert,
Weinert. J.
E. Math.
WimmerPhys.
and22A.J.
(1981)
Freeman,
2433. Phys. Rev. B

[1] Reviews and references to the original literature of DFT 26 (1982) 4571.
and the LSDA can be found in: N.H. March and S. [18] P. Blaha and K. Schwarz, Int. J. Quantum Chem. XXIII
Lundqvist, eds., Theory of the Inhomogeneous Electron (1983) 1535.
Gas, (Plenum, New York, 1983). [191 H.J.F. Jansen and A.J. Freeman. Phys. Rev. B 30 (1984)
D. Langreth and H. SuhI, eds., Many-Body Phenomena at 561.
Surfaces, (Academic, Orlando 1984). [201 P. Blaha, K. Schwarz and P. Herzig, Phys. Rev. Lett. 54
J.P. DahI and J. Avery. eds., Local Density Approxima- (1985) 1192.
lions in Chemistry and Solid State Physics (Plenum, New [21] S.H. Wei, H. Krakauer and M. Weinert. Phys. Rev. B 32
York, 1984). (1985) 7792.
SB. Trickey, ed., Density Functional Theory for Many- [22] L.F. Mattheiss and DR. Hamann. Phys. Rev. B 33 (1986)
Fermion Systems (Academic, San Diego, 1990). 823.
[2] Reviews and references to the original Xa literature are [231 K. Kurki-Suonio, Israel J. Chem. 16 (1977) 115.
in: [241 J.P. Desclaux. Comput. Phys. Commun. 1(1969) 216: the
J.C. Slater, The Self-Consistent Field for Molecules and code we use is an apparently unpublished relativistic
Solids (McGraw-Hill, New York, 1974). version of the non-relativistic code described in this refer-
J.W.D. Connolly, in: Modern Theoretical Chemistry, GA. ence. Although widely distributed. we have not found a
Segal. ed. (Plenum, New York, 1977). formal reference for the relativistic version. The author’s
[31L. Hedin and B.I. Lundqvist. J. Phys. C. 4 (1971) 2064: treatment of the Dirac—Fock formalism is in: J.P. Dc-
J. Phys. (Paris) ColIoq. 33 (1972) 3. sclaux, Comput. Phys. Commun. 9 (1975) 31.
[4] U. von Barth and L. Hedin, J. Phys. C 5 (1972) 1629. [25] A. Nobile and V. Roberto. Comput. Phys. Commun. 42
[51V.L. Moruzzi, J.F. Janak and AR. Williams. Calculated (1986) 233.
Electronic Properties of Metals (Pergamon, New York, [261 B.T. Smith. J.M. Boyle. J.J. Dongarra, B.S. Garbow, Y.
1978). Ikehe. V.C. Klema and C.B. Moler, Matrix Eigensystem
[6] AD. Becke, J. Chem. Phys. 76 (1982) 6037; 78 (1983) Routines, Eispack Guide. Lecture Notes in Computer
4787. Science 6 (Springer. Berlin, 1976).
[7] J.C. Slater, in: Methods of Computational Physics, vol. 8. [27] G.W. Pratt. Phys. Rev. 88 (1952) 1217.
B. Alder, S. Fernbach and M. Rotenberg, eds. (Academic. [28] D. Singh. H. Krakauer and C-S. Wang, Phys. Rev. B 34
New York, 1968) p. 1. (1986) 8391.
L.F. Mattheiss, A.C. Switendick and J.H. Wood, ibid. p. [291P. Blaha and K. Schwarz. J. Phys. F. 17 (1987) 899.
64. [301P. Blaha. K. Schwarz and P.H. Dederichs. Phys. Rev. B 38
T.L. Loucks, The Augmented Plane Wave Method (Ben- (1988) 968. 9368.
jamin, New York, 1967). [311 P. Blaha, P. Sorantin, C. Amhrosch and K. Schwarz,
[8] H. Schiosser and P.M. Marcus, Phys. Rev. 131 (1963) Hyperfine Interactions 51(1989) 917.
2529. [32] C’. Ambrosch-Draxl, P. Blaha and K. Schwarz, J. Phys.
PD. DeCicco, Phys. Rev. 153 (1967) 931. CTond. Matt. 1(1989) 4491.
[9] J.C. Boettger and S.B. Trickey. Phys. Rev. B 29 (1984) [33] SB. Trickey. J.A. Nobel. P. Blaha and K. Schwarz. Phys.
6425. Rev. B, submitted.
 P. Blaha et al. / Plane wave programs for crystalline systems 413

TEST RUN OUTPUT

Abridged output of test case NaCI

ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM NA

OVERALL ENERGY PARAMETER IS -2.1000
E( 0)— —3.7600 E(BOTTOM)— —3.840 E(TOP)— —3.680
E( 1)— —1.7500 E(BOTTOM)— —1.900 E(TOP)— —1.600
ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM CL
OVERALL ENERGY PARAMETER IS -2.1000
E( 0)— —2.6000
K— 0.00000 0.00000 0.00000 GAMMA
MATRIX SIZE— 169 WEIGHT 1.00 PGR:
EIGENVALUES ARE:
—3.7418601 —1.7175804 —1.7175804 —1.7175804

NUMBER OF K-POINTS: 6
‘NUMBER OF ELECTRONS — 8.000
F E R M I — ENERGY — -1.71758
AT.NR.: 1 POSITION: 0.000 0.000 0.000 MULTIPLICITY:
LMMAX—5LM— 00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 1 — 7.906812
PARTIAL CHARGES SPHERE: 1 S,P,D,F, D-EG,D-T2G
1.992 5.915 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
AT.NR.: 2 POSITION: 0.500 0.500 0.500 MULTIPLICITY:
LMMAX—5LM— 00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 2 — 0.024575
PARTIAL CHARGES SPHERE: 2 S,P,D,F, D-EG,D—T2G
0.009 0.006 0.005 0.003 0.000 0.000 0.000 0.004 0.000 0.000 0.000 0.000
TOTAL CHARGE INSIDE CELL: 8.000000
SUM OF EIGENVALUES -
—17.796159
ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM NA
OVERALL ENERGY PARAMETER IS -0.1000
E( 0)— 0.3000
E( 1)— 0.9800 E(BOTTOM)— 0.980 E(TOP)— —200.000
ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM CL
OVERALL ENERGY PARAMETER IS -0.1000
E( 0)— —1.1600 E(BOTTOM)— —1.440 E(TOP)— —0.880
E( 1)— —0.2400 E(BOTTOM)— —0.660 E(TOP)— 0.180
Ic— 0.00000 0.00000 0.00000 GAMMA
MATRIX SIZE— 169 WEIGHT— 1.00 PGR:
ZIGENVALUES ARE:
—1.1554964 —0.1713413 —0.1713413 —0.1713413 0.4584009
0.7668801 0.7668801 0.7668801 0.9255592 0.9255592

NUMBER OF K-POINTS: 89
NUMBER OF ELECTRONS — 8.000
FERMI—ENERGY — -0.17134
AT.NR.: 1 POSITION: 0.000 0.000 0.000 MULTIPLICITY:
LMMAX—5LM— 00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 1 — 0.147766
PARTIAL CHARGES SPHERE: 1 S,P,D,F, D-EG,D-T2G
0.048 0.062 0.030 0.006 0.000 0.000 0.000 0.025 0.006 0.000 0.000 0.000
AT.NR.: 2 POSITION: 0.500 0.500 0.500 MULTIPLICITY:
LMMAX—5LM— 00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 2 — 7.176673
PARTIAL CHARGES SPHERE: 2 S,P,D,F, D-EG,D-T2G
1.920 5.246 0.009 0.002 0.000 0.000 0.000 0.002 0.007 0.000 0.000 0.000
TOTAL CHARGE INSIDE CELL: 8.000000
SUM OF EIGENVALUES —
—3.675440
CORE STATES CONTRIBUTION TO TOTAL ENERGY CALCULATED BY CORE
i.ATOM: NA 1 CORE STATES
is
—75.154262 RYD
414 P. Blaha et a!. / Plane wave programs for crystalline systems

2.ATOM: CL 4 CORE STATES

iS
—201.094712 RYD
2S
—18.097446 RYD
2P
—13.638455 RYD
2P*
—13.762843 RYD
PRATT MIXING SCHEME WITH 1.000
DENSITY AT NUCLEUS:
JATOM VALENCE SEMI-CORE CORE TOTAL
1 0.404091 49.711816 817.816402 867.932309
2 27.848612 0.048752 3551.787654 3579.685018
CHARGE IN SPHERE 1 — 10.0545779
CHARGE IN SPHERE 2 — 17.2012472
********** TOTAL ENERGY IN RYD: —692.243945

1. ITERATION

The preceding line is the top of iteration 1. The data from it

and the following five iterations are deleted.

7. ITERATION

SUBSTANCE: NaCl
LATTICE — FC
NON-SPINPOLARIZED CALCULATION
POTENTIAL OPTION 3
LATTICE CONSTANTS ARE — 9.8000000 9.8000000 9.8000000
MODE OF CALCULATION IS — RELA
ATOMNUMBER— 1 NA VCOUL-ZERO — 0.21736E+00
ATOMNUMBER— 2 CL VCOUL—ZERO — O.37078E+00
ATOM 1 AT RM’T: SIGMA OF V—XC FIT: 0.268000466821563342E—02
ATOM 2 AT RNT: SIGMA OF V—XC FIT: 0.839605323624305219E—04
DENSITY INTEGRALS—
—557.816475 1718.187925 (RYD)

ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM NA

OVERALL ENERGY PARAMETER IS -0.1000

E( 0)— 0.3000
E( 1)— 1.1600 E(BOTTOM)— 1.160 E(TOP)— —200.000
ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM CL
OVERALL ENERGY PARAMETER IS -0.1000
E( 0)— —0.9600 E(BOTTOM)— —1.260 E(TOP)— —0.660
E( 1)— —0.0500 E(BOTTOM) —0.500 E(TOP)— 0.400
K- 0.00000 0.00000 0.00000 GAMMA
MATRIX SIZE— 169 WEIGHT— 1.00 PGR:
EIGENVALUES ARE:
—0.9765704 0.0042198 0.0042198 0.0042198 0.4711303
0.8048377 0.8048377 0.8048377 0.9890194 0.9890194

NUMBER OF K-POINTS: 89
FERMI—ENERGY — 0.00422
AT.NR.: 1 POSITION: 0.000 0.000 0.000 MULTIPLICITY:
LMMAX—5LM— 00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 1 — 0.198284
PARTIAL CHARGES SPHERE: 1 S,P,D,F, D-EG,D-T2G
0.066 0.085 0.039 0.007 0.000 0.000 0.000 0.031 0.009 0.0.J0 0.000 0.000
AT.NR.: 2 POSITION: 0.500 0.500 0.500 MULTIPLICITY:
LMMAX—5 114—00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 2 — 6.877690
PARTIAL CHARGES SPHERE: 2 S,P,D,F, D-EG,D-T2G
1897 4.960 0.017 0.003 0.000 0.000 0.000 0.004 0.012 0.000 0.000 0.000
TOTAL CHARGE INSIDE CELL: 8.000000
SUM OF LIGENVALUES —
—2.401462
 P. Blaha et a!. / Plane wove programs for crystalline systems 415

ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM NA

OVERALL ENERGY PARAMETER IS —2.1000
E( 0)— —3.5600 E(BOTTOM)— —3.640 E(TOP)— —3.480
E( 1)— —1.5700 E(BOTTOM)— —1.720 E(TOP)— —1.420
ATOMIC SPHERE DEPENDENT PARAMETERS FOR ATOM CL

OVERALL ENERGY PARAMETER IS -2.1000

E( 0)— —2.6000
K— 0.00000 0.00000 0.00000 GAMMA
MATRIX SIZE— 169 WEIGHT— 1.00 PGR:
EIGENVALUES ARE:
—3.5509862 —1.5303980 —1.5303980 —1.5303980

NUMBER OF K-POINTS: 6
NUMBER OF ELECTRONS — 8.000
FERMI—ENERGY - -1.53040
AT.NR.: 1 POSITION: 0.000 0.000 0.000 MULTIPLICITY:
LMMAX—5LM— 00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 1 — 7.890221
PARTIAL CHARGES SPHERE: 1 S,P,D,F, D-EG,D-T2G
1.991 5.899 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
AT.NR.: 2 POSITION: 0.500 0.500 0.500 MULTIPLICITY:
LMMAX—5LM— 00404460640000000000000000
TOTAL CHARGE INSIDE SPHERE 2 — 0.029400
PARTIAL CHARGES SPHERE: 2 S,P~D,F, D-EG,D-T2G
0.010 0.007 0.006 0.004 0.000 0.000 0.000 0.005 0.001 0.000 0.000 0.000
TOTAL CHARGE INSIDE CELL: 8.000000
SUM OF EIGENVALUES —
—16.294031
CORE STATES CONTRIBUTION TO TOTAL ENERGY CALCULATED BY CORE
1.ATOM: NA 1 CORE STATES
is
—74.969314 RYD
2.ATOM: CL 4 CORE STATES
is
—200.800365 RYD
2S
—17.815701 RYD
2P
—13.355086 RYD
2P*
—13.479385 RYD
BROYD MIXING SCHEME WITH 0.400
DENSITY AT NUCLEUS:
JATOM VALENCE SEMI-CORE CORE TOTAL
1 0.567031 49.666534 817.839320 868.072886
2 26.270111 0.051289 3551.663513 3577.984914
CHARGE IN SPHERE 1 — 10.0885579
CHARGE IN SPHERE 2 — 16.9070901
********** TOTAL ENERGY IN RYD: —1244.061842