PHYSICAL REVIEW B VOLUME 53, NUMBER 7 15 FEBRUARY 1996-I

Generalized Kohn-Sham schemes and the band-gap problem

A. Seidl, A. Görling, P. Vogl, and J. A. Majewski
Walter Schottky Institut and Physik Department, Technische Universität München, 85748 Garching, Germany

M. Levy
Department of Chemistry and Quantum Theory Group, Tulane University, New Orleans, Louisiana 70118
~Received 24 August 1995!
As an alternative to the standard Kohn-Sham procedure, other exact realizations of density-functional theory
~generalized Kohn-Sham methods! are presented. The corresponding generalized Kohn-Sham eigenvalue gaps
are shown to incorporate part of the discontinuity D xc of the exchange-correlation potential of standard Kohn-
Sham theory. As an example, a generalized Kohn-Sham procedure splitting the exchange contribution to the
total energy into a screened, nonlocal and a local density component is considered. This method leads to band
gaps far better than those of local-density approximation and to good structural properties for the materials Si,
Ge, GaAs, InP, and InSb.

I. INTRODUCTION ment, in contrast to LDA. We shall call these alternative

realizations of DFT generalized Kohn-Sham (GKS) schemes.
First-principles calculations based on the Kohn-Sham1 We show that the GKS framework that we develop in this
scheme of density-functional theory2 ~DFT! have success- paper constitutes a rigorous basis for employing nonlocal
fully predicted and explained a wide range of solid-state potentials such as Hartree-Fock-like exchange potentials or
properties.3 Strictly speaking, however, this is true only for screened nonlocal exchange potentials within Kohn-Sham-
cohesive and structural properties, whereas band gaps are type approaches. An example of such an approach with
typically underestimated by a factor of 2. One approach to screened nonlocal exchange and LDA correlation potentials
solve this gap problem is to consider the energies of quasi- was first proposed by Bylander and Kleinman on empirical
particles and to calculate the electron self-energy in terms of grounds,8 and was shown to give significantly better energy
perturbation theory. This approach has been followed by sev- gaps in Si. This method was also applied to atoms in Refs. 9
eral authors4 – 6 invoking Hedin’s GW approximation.7 While and 10. In this work, we show that this method is firmly
this procedure has been quite successful, it does not allow rooted within DFT and we present results for structural prop-
one to calculate structural properties together with energy erties and band gaps in several semiconductors based on this
gaps in a self-consistent way. Since, however, principal band screened-exchange GKS scheme. We show that this ap-
gaps E g of semiconductors are differences of ground-state proach, which we shall call the screened-exchange LDA
energies of N and N61 particle systems, E g method ~sX-LDA!, compares very favorably with experi-
5E(N11)1E(N21)22E(N), they are in principle acces- ment.
sible by DFT. The paper is organized as follows. In Sec. II, the GKS
So far, DFT has been applied almost exclusively within formalism is developed, generalizing earlier work of Ref. 11.
the Kohn-Sham formalism where E g is the difference of Several examples are given in Sec. III. In particular, a
single-particle eigenvalues plus a contribution that originates scheme based on screened nonlocal exchange functionals is
in the discontinuity D xc of the exchange-correlation potential discussed. The analog of the local-density approximation of
at integer particle numbers. In practice, the Kohn-Sham this scheme is shown to lead to the sX-LDA method. In Sec.
scheme is usually carried out within the local-density ap- IV, we analyze the band gaps emerging from the various
proximation ~LDA!, whose exchange-correlation potential GKS schemes and compare them with the standard Kohn-
exhibits no discontinuity (D xc50). Indeed, the energy gaps Sham eigenvalue gaps. Based on nonlocal pseudopotentials
within LDA show a large discrepancy with experiment that is and a plane-wave representation, we present energy gaps,
caused both by the lack of the discontinuity and errors in the valence-band widths, lattice constants, and bulk moduli cal-
single-particle eigenvalues resulting from the approximative culated within the sX-LDA method for Si, Ge, GaAs, InP,
nature of the functional. and InSb in Sec. V. In addition, we develop a perturbative
In this paper we show that there are numerous exact real- version of the sX-LDA scheme that is computationally much
izations of DFT besides the standard Kohn-Sham procedure less demanding but still gives results in fair agreement with
that yield the correct total energy of the system and lead to the self-consistent version. In addition, several screening
self-consistently determined single-particle eigenvalues of models for the nonlocal exchange potential are discussed. A
the N-particle system whose differences already incorporate summary is given in Sec. VI. In Appendix A, we prove the
part of the discontinuity D xc . In particular, these schemes Hohenberg-Kohn theorem2 for the GKS formalism. Finally,
lend themselves to approximations that allow the self- some variational properties of the GKS solutions are proven
consistent calculation of both structural properties and en- in Appendix B. Atomic units are used throughout the paper
ergy gaps of semiconductors in good agreement with experi- except where noted.

0163-1829/96/53~7!/3764~11!/$06.00 53 3764 © 1996 The American Physical Society

53 GENERALIZED KOHN-SHAM SCHEMES AND THE BAND-GAP PROBLEM 3765

II. DERIVATION OF GENERALIZED KOHN-SHAM S @ F # 5 ^ F u T̂ u F & 1U H @ $ f i % # 1E sx

x @ $ f i% # . ~2.5!
SCHEMES
In Eqs. ~2.3!–~2.5!, U H is the Hartree energy and E x is the
Virtually all concrete applications of density-functional exchange energy that arises from the determinant F. Note
theory to the ground state of many-electron systems are that E x differs in general from the Hartree-Fock exchange
based on the well-known Kohn-Sham1 scheme. In this work, energy. E sx
x denotes a statically screened exchange interac-
we will develop generalized Kohn-Sham schemes that in- tion with a Thomas-Fermi screening constant k TF ,
clude the standard Kohn-Sham scheme as a special case. In

E
deriving these GKS schemes, we will invoke the N
constrained-search formulation of DFT.3,12–15 x @ F # 52
E sx (
i, j
drdr8
We start by considering the Schrödinger equation of N
electrons, 2k TFu r2r8u
f* j ~ r8 ! e
i ~ r! f * f j ~ r! f i ~ r8!
3 . ~2.6!
~ T̂1V̂ ee 1 v̂ ! C 0 @v# 5E 0 @v# C 0 @v# . ~2.1! u r2r8u
In Eq. ~2.1!, T̂ and V̂ ee are the operators of the kinetic energy In contrast to the first example, the functionals S @ F # in Eqs.
and the electron-electron interaction, respectively. The nota- ~2.4! and ~2.5! contain part of the electron-electron interac-
tion emphasizes the fact that the ground-state energy E 0 and tion besides the kinetic energy. Based on these functionals
wave function C 0 are functionals of the external potential S @ F # , we now define functionals F S @ r # of the density
v (r). Equation ~2.1! is equivalent to the minimization3,12–14 r (r)5 ( i u f i (r) u 2 ,

E 0 @v# 5 min
r ~ r! →N
H F@r#1 E J
drv~ r! r ~ r! , ~2.2a!
F S @ r # 5 min S @ F # 5 min S @ $ f i % # .
F→ r ~ r! $ f i % → r ~ r!
~2.7!

The minimization process in Eq. ~2.7! searches all Slater

F @ r # 5 min ^ C u T̂1V̂ ee u C & . ~2.2b! determinants or unitary orbitals $ f i % that yield the density
C→ r ~ r! r (r). The minimizing determinant is denoted by F S @ r # . As
The minimizations in Eqs. ~2.2! are performed within the will become clear below, the functional F S @ r # plays an
space of all densities yielding N electrons and all antisym- analogous role as the noninteracting kinetic energy in the
metric wave functions yielding the density r (r), respec- standard Kohn-Sham formalism but, in addition, contains
tively. Equation ~2.2b! defines the Hohenberg-Kohn func- parts of the electron-electron interaction energy. It depends
tional F @ r # . 2 not only on the density but also on the chosen S @ F # , and is
The basic idea of the standard Kohn-Sham scheme is to defined by a single Slater determinant rather than with the
replace the calculation of C 0 @v# by that of a single Slater N-particle wave function as is the case of the Hohenberg-
determinant that represents a noninteracting model system Kohn functional, Eq. ~2.2b!. Such functionals F S will be
and yields the same ground-state density as C 0 @v# . How- shown to lead to an exact realization of DFT, provided the
ever, the expectation value of the Hamiltonian with this de- corresponding functionals S @ F # obey in all cases of interest
terminant only gives part of the total energy whereas the the following conditions that guarantee the existence of self-
remaining exchange-correlation contribution to the total en- consistent single particle equations.
ergy is not directly accessible by the determinant. ~C1! The minimum F S @ r # defined by Eq. ~2.7! and its
Alternatively, one may try to introduce ~interacting! functional derivative with respect to r (r) exist.
model systems that take into account the electron-electron ~C2! Define the energy

E
interaction to some extent and incorporate at least part of the
exchange and correlation contribution to the total energy, but E S @ $ f i % ; v eff# 5S @ $ f i % # 1 drv eff~ r! r ~ r! , ~2.8!
can still be represented by a single Slater determinant. The
latter is important to obtain tractable single-particle equa- where v eff is an arbitrary local multiplicative potential. The
tions. In this way one may hope to obtain single-particle minimization of this energy by the usual Lagrange procedure
equations with eigenvalues that more faithfully reflect the is required to lead to a set of canonical single-particle equa-
physical excitation energies and energy gaps. tions that can be cast into the form
Let us define an energy functional S @ F # of N-electron
Slater determinants F. Throughout this work, Slater deter- Ô S @ $ f i % # f j 1 v̂ efff j 5« j f j with j51, . . . ,N, ~2.9!
minants will be denoted by F to distinguish them from gen-
eral many-electron wave functions C. S @ F # defines a func- where the operator Ô S may depend on the orbitals and the
tional S @ $ f i % # of the N unitary ~spinor! orbitals that generate functional S, but not explicitly on the potential v̂ eff , and is
invariant with respect to unitary transformations of the orbit-
F. According to its definition, the functional S @ $ f i % # is in-
variant with respect to unitary transformations of the orbitals. als. Furthermore we require that the orbitals f j that mini-
We give three examples of such functionals S @ F # that will mize E S @ $ f i % ; v eff# are the N energetically lowest eigenstates
be seen to be relevant for the further discussion, namely of the Hamiltonian Ô S 1 v̂ eff in Eq. ~2.9!. Note that the op-
erator Ô S is nonlocal in general. We denote the density that
S @ F # 5 ^ F u T̂ u F & , ~2.3! results from the minimization of the energy E S in Eq. ~2.8!
by r S0 ( @v eff# ;r)].
S @ F # 5 ^ F u T̂1V̂ ee u F & 5 ^ F u T̂ u F & 1U H @ $ f i % # 1E x @ $ f i % # , ~C3! The involved densities are v representable, i.e., ev-
~2.4! ery physically realized ground-state density r 0 ( @v# ;r) of Eq.
3766 SEIDL, GORLING, VOGL, MAJEWSKI, AND LEVY 53

~2.1! equals the density r S0 ( @v eff# ;r) for some potential the example Eq. ~2.3!, one has Ô S 5T̂, whereas Eq. ~2.4!
v eff(r) and, conversely, every density r S0 ( @v eff# ;r) is the leads to Ô S 5T̂1 v̂ NL
x 1û, where v̂ x
NL
has the form of the
ground-state density r 0 ( @v# ;r) for some external potential nonlocal Hartree-Fock exchange potential and û is the clas-
v (r) in Eq. ~2.1!. sical Coulomb potential of the density given by the orbitals
We follow the general practice in density-functional $ f i% .
theory to merely assume the first and third condition rather Next, the total energy E 0 @v# of the interacting system in
than attempting to prove them. We note, however, that the Eq. ~2.1! is divided into the total energy of the model system
third condition is less stringent than the condition that any and the remainder. In order to do this, we denote the differ-
given arbitrary density must be the ground-state density ence between F S @ r # of Eq. ~2.7! and the Hohenberg-Kohn
r 0 ( @v# ;r) of a Schrödinger equation with a suitable external functional F @ r # of Eq. ~2.2b! by the functional R S @ r # ,
potential ~which has been shown to be invalid16,17!.
In any practical application of the GKS scheme, one will F @ r # 5F S @ r # 1R S @ r # . ~2.10!
choose functionals S @ F # that lead to single-particle equa-
tions of a form that is required by the second condition. In Then, we can write

E 0 @v# 5 min
r ~ r! →N
H F S @ r # 1R S @ r # 1 E J
drv~ r! r ~ r! 5 min H min S @ F # 1R S @ r # 1
r ~ r! →N F→ r ~ r!
E drv~ r! r ~ r! J
H
5 min S @ F # 1R S @ r @ F ## 1
F→N
E J
drv~ r! r ~@ F # ;r! 5 min
$ f i % →N
H S @ $ f i % # 1R S @ r @ $ f i % ## 1 E J
drv~ r! r ~@ $ f i % # ;r! .

~2.11!

Note that the functional R S @ r # and the term * drv (r) r (r) in which follows by inserting the expression Eq. ~2.10! for
Eq. ~2.11! depend on the determinant F or on the orbitals r 0 ( @v# ;r)5 r S0 ( @v eff# ;r) into Eq. ~2.11!. The first and the
$ f i % only indirectly through the density. The latter is written third contribution on the right-hand side of Eq. ~2.15! can be
as functional r ( @ F # ;r) or r ( @ $ f i % # ;r). The ground-state calculated exactly. F S @ r S0 @v eff## follows from evaluating
density r 0 ( @v# ;r) is determined by minimizing E 0 @v# within S @ $ f i % # with the self-consistent GKS orbitals. The contribu-
the space of unitary orbitals. Because of condition ~C2!, a tion R S @ r S0 ( @v eff# ;r) # , as well as the potential v R (r) in the
Lagrange procedure leads to the GKS equations GKS equations, needs to be approximated. This procedure is
analogous to the treatment of the exchange-correlation func-
Ô S @ $ f i % # f j 1 v̂ R f j 1 v̂ f j 5« j f j with j51, . . . ,N, ~2.12! tionals in the standard Kohn-Sham scheme. In contrast to the
latter scheme, however, important portions of the exchange
where correlation energy are already contained in F S that is treated
exactly.
d R S@ r # Note that the self-consistently determined density
v R ~ r! 5 . ~2.13!
d r ~ r! r S0 ( @v eff# ;r)5 r 0 ( @v# ;r) is the minimizing density not only
of the minimization Eq. ~2.11! but also of
While the exact form of the functional R S @ r # and of its func-
tional derivative v R (r) are not known, suitable approxima-
tions can be found as will be discussed in Sec. III. E S0 @v eff# 5 min
r ~ r! →N
H F S@ r # 1 E J
drv eff~ r! r ~ r! , ~2.16!
Importantly, the orbitals resulting from these GKS equa-
tions yield the exact ground-state density r 0 ( @v# ;r) of the with v eff(r) being determined by Eq. ~2.14!. This follows
Schrödinger equation ~2.1! because they are obtained from requirement ~C2! because the preceding Eq. ~2.16! re-
through the minimization of the true ground-state energy sults from the minimization of the energy E S @ $ f i % ; v eff# .
E 0 @v# as given in Eq. ~2.11!. Note that the GKS equations We will now clarify the role of conditions ~C1! and ~C3!.
have exactly the form of Eq. ~2.9! with The first requirement ~C1! guarantees that the functional
F S @ r # and its functional derivative d F S @ r # / d r (r) exist.
v eff~ r! 5 v~ r! 1 v R ~ r! . ~2.14! Consequently, we can conclude that R S @ r # and
d R S @ r # / d r (r) exist for all densities r (r) provided the
After the GKS equations have been solved self-consistently, Hohenberg-Kohn functional F @ r # and its functional deriva-
the ground-state energy E 0 @v# of the Schrödinger equation tive exist. Indeed, the existence of F @ r # has already been
~2.1! can be evaluated according to the equation proven in Refs. 16 and 18, whereas that of d F @ r # / d r (r) is
merely assumed in the Kohn-Sham formalism.
E 0 @v# 5F S @ r S0 @v eff## 1R S @ r S0 @v eff## 1 E drv~ r! r S0 ~@v eff# ;r! ,
The first part of requirement ~C3! guarantees the existence
of an effective potential v eff(r) leading to a density
~2.15! r S0 ( @v eff# ;r) that equals the ground-state density r 0 ( @v# ;r).
53 GENERALIZED KOHN-SHAM SCHEMES AND THE BAND-GAP PROBLEM 3767

Indeed, this existence follows a posteriori whenever the and E c @ r # is the correlation energy. The corresponding func-
GKS equations lead to a self-consistent solution. tional derivatives with respect to the density are denoted by
We show in Appendix B that the second part of require- u( @ r # ;r), v x ( @ r # ;r), and v c ( @ r # ;r) for the Coulomb, ex-
ment ~C3! guarantees the calculated density r S0 ( @v eff# ;r) to change, and correlation potential, respectively. The standard
represent the minimum of F @ r # 1 * drv (r) r (r) instead of Kohn-Sham equations read
merely some stationary solution. In a similar way, one may
also show that the solution of the GKS equations is unique. @ T̂1 v̂ 1û @ r # 1 v̂ x @ r # 1 v̂ c @ r ## f i 5« i f i
This is proven in Appendix A. Thus, the condition ~C3! guar-
antees that r S0 ( @v eff# ;r) equals the ground-state density with j51, . . . ,N. ~3.3!
r 0 ( @v# ;r) of Eq. ~2.1!. Since Eq. ~3.2! does not allow one to directly calculate the
In summary, the GKS scheme replaces the original Schrö- functional derivative v x ( @ r # ;r) and the correlation energy is
dinger equation ~2.1! by a set of one-particle equations, Eqs. not known exactly, both exchange and correlation energies
~2.9!, that are much easier to handle. In this respect, this and potentials are determined from approximate density
scheme is equivalent to the standard Kohn-Sham formalism. functionals ~such as LDA!.
The key difference is that not only the noninteracting kinetic Henceforth, T @ r # refers to the kinetic energy evaluated
energy but also the part of the total energy given by F S @ r # is with the self-consistent orbitals of Eq. ~3.3!.
treated exactly. F S @ r # is determined by the choice of the
functional S @ $ f i % # . Different realizations of the GKS
B. Hartree-Fock–Kohn-Sham scheme
scheme are therefore characterized by the choice of
S @ $ f i % # and the approximations for R S @ r # @Eq. ~2.10!#. The If the functional S @ F # is chosen as the sum of the kinetic
major advantage of this GKS procedure is that suitable and the electron-electron energy as in Eq. ~2.4!, one obtains
choices of F S can result in R S to be small compared to the a GKS scheme that resembles the Hartree-Fock method. The
total energy E 0 @v# . Thus, errors in approximating R S have resulting procedure is known as the Hartree-Fock–Kohn-
only a small effect on the energy. Sham ~HF-KS! scheme.11,20,21 To refer to this case, we shall
The derivation given above does not make use of a denote the ~unknown! potential v R (r) as v HFc ( @ r # ;r) to indi-
Hohenberg-Kohn type theorem.2 For an interacting many- cate that it contains the correlation effects. The correspond-
electron system, the Hohenberg-Kohn theorem states that ing generalized Kohn-Sham equations are given by
there are not two external potentials differing by more than a
constant that give the same ground-state density. For the 2 21 ¹ 2 f i ~ r! 1 v~ r! f i ~ r! 1u ~@ r # ;r! f i ~ r!
GKS equations ~2.9!, the equivalent theorem states that there
are not two effective potentials v eff(r) differing by more than
a constant that give the same density r S0 ( @v eff# ;r) as a result
2 E x ~ r,r8 ! f i ~ r8 ! 1 v c ~@ r # ;r ! f i ~ r ! 5« i f i ,
dr8v NL HF

of minimizing E S @ $ f i % ; v eff# of Eq. ~2.8!. We present a proof ~3.4!

of this latter theorem in Appendix A.
N
f j ~ r! f *j ~ r8!
III. EXAMPLES OF GENERALIZED KOHN-SHAM x ~ r,r8 ! 52
v NL (
j51 u r2r8u
. ~3.5!
SCHEMES
Here, the correlation potential ~as well as the energy! differs
A. Standard Kohn-Sham method from the standard Kohn-Sham case since the orbitals $ f i %
By choosing S @ $ F % # to be equal to the kinetic energy of a obey different one-particle equations but it is believed that
Slater determinant, Eq. ~2.3!, one obtains the standard Kohn- this difference is small.22
Sham scheme.1,3,12–15 Strictly speaking, this is true only for Apart from the ~unknown! potential term v HF c ( @ r # ;r), Eq.
systems where the wave function of the model system can be ~3.4! corresponds to the Hartree-Fock equations. In contrast
represented by a single Slater determinant. This is the case, to the Hartree-Fock method, however, this procedure is for-
for example, in nondegenerate systems. More general situa- mally exact.
tions have been discussed in Ref. 19.
With S @ $ F % # from Eq. ~2.3!, the functional R S @ r # is C. Screened, nonlocal exchange — sX-LDA scheme
given by
Motivated by the observation that Hartree-Fock band gaps
R @ r # 5U @ r # 1E x @ r # 1E c @ r # ,
S
~3.1! in solids exceed the observed ones by a large amount, it
appears plausible to invoke a screened exchange potential in
where U @ r # is the classical Coulomb energy, E x @ r # is the the single-particle equations. Indeed, this approach has been
exchange energy, used on the basis of the quasiparticle scheme4 – 6 to derive
perturbative methods to solve quasiparticle equations. The

E E
N N
1 latter approach, however, does not easily allow one to com-
E x @ r # 52
2 ((
i51 j51
dr dr8 pute total energies.
The generalized Kohn-Sham method outlined in this pa-
f i* ~@ r # ;r! f *j ~@ r # ;r8! f j ~@ r # ;r! f i ~@ r # ;r8! per offers the possibility to split up the total energy in such a
3 , way that the variational single-particle equations contain a
u r2r8u
screened exchange potential. As will be demonstrated below,
~3.2! this procedure yields single-particle eigenvalues leading to
3768 SEIDL, GORLING, VOGL, MAJEWSKI, AND LEVY 53

more accurate band gaps than both Hartee-Fock and LDA,

while the quality of the total energy remains comparable to
LDA.
4
3
2 z2
F ~ z ! 512 z arctan 2
z 6
12
z2
4 F S D S DG
z
4
13 ln 11 2 .
~3.15!
To put these ideas into a rigorous framework, we choose
S @ r # as given in Eq. ~2.5!. The Thomas-Fermi screening Here, z5k TF /k̄ F and k̄ F is the Fermi wave vector corre-
constitutes the simplest version of screening the exchange sponding to the average density. Thus, these energy contri-
energy, particularly if one evaluates the Thomas-Fermi con- butions are essentially approximated by functionals corre-
stant with the average total density. sponding to the homogeneous electron gas, invoking the
This choice of S @ r # implies a functional R S @ r # that is spirit of the local-density approximation. Note that the usage
given by of an average Fermi wave vector is an additional approxima-
tion going slightly beyond LDA.8 The difference
R S @ r # 5E sx T @ r # 2T sx @ r # is simply set to zero ~see Ref. 22 for reasons
xc @ r # 5 $ E x @ r # 2E x @ r # 1T @ r # 2T @ r # % 1E c @ r # .
sx sx

~3.6! to justify this!.

In effect, the same type of procedure has been employed
As indicated, we shall refer to this functional as E sx xc @ r # . The by Bylander and Kleinman.8 The derivations in this section
term in the curly brackets takes into account the energy dif- put this procedure on a solid basis within the constrained
ference between the exchange energy E x @ r # in the standard search formulation of DFT.
Kohn-Sham scheme and the screened exchange energy
x @ r # as given in Eq. ~2.6! and the corresponding difference
E sx IV. BAND GAPS IN GENERALIZED KOHN-SHAM
T @ r # 2T sx @ r # in the kinetic energies. Note that the correla- SCHEMES
tion energy E c @ r # in Eq. ~3.6! is the same as in the standard
The fundamental energy gap E g in a semiconductor can
Kohn-Sham procedure.
be expressed as a total energy difference. Alternatively, it can
The resulting generalized Kohn-Sham equations ~sX-LDA
be expressed entirely in terms of KS eigenvalues:24,25
equations! are
E g 5« N11
KS
~ N ! 2« KS
N ~ N ! 1« N11 ~ N11 ! 2« N11 ~ N !
KS KS
2 12 ¹ 2 f i ~ r! 1 v~ r! f i ~ r! 1u ~@ r # ;r! f i ~ r!
5« KS
g 1D xc 5« g 1D x 1D c
KS KS KS KS
~4.1!
2 E dr8v sx,NL
x ~ r,r8! f i ~ r8! 1 v sx
xc ~@ r # ;r ! f i ~ r ! 5« i f i , with

~3.7! g 5« N11 ~ N ! 2« N ~ N !
« KS KS KS

D KS
xc 5« N11 ~ N11 ! 2« N11 ~ N ! .
KS KS
~4.2!
f j ~ r! e 2k TFu r2r8u f *j ~ r8!
N

v sx,NL
x ~ r,r8! 52 (
j51 u r2r8u
, ~3.8! « KS
i (M ) denotes the energetically ith lowest Kohn-Sham ei-
genvalue of the M particle system. Thus, « KS M (M ) is the
xc ( @ r # ;r) is the functional derivative of E xc @ r # with
where v sx sx
highest occupied and « KS M 11 (M ) the lowest unoccupied or-
respect to the density and v̂ x sx,NL
is the nonlocal screened bital energy. The contribution « KS g is the eigenvalue gap,
exchange operator. If k TF is set equal to zero, the sX-LDA given by the difference in energy of lowest unoccupied and
scheme is identical to the HF-KS scheme of the previous highest occupied level. The contribution D KS xc results from the
section III.B.. So far the formalism has been exact. For con- discontinuity of the exchange-correlation potential with re-
crete applications, the functionals contributing to R S in Eq. spect to particle number24,25 and involves the eigenvalues of
~3.6! are approximated by the (N11)-particle system. The magnitude of D KS xc in the
exact Kohn-Sham formalism is not known quantitatively26 –28
E x @ r # .E LDA
x @r#5 E drr ~ r! e x @ r # , ~3.9!
but it is generally believed that a fair fraction of the discrep-
ancy between the LDA eigenvalue gaps and the experimental
gaps originates in this discontinuity.29

x @ r # .E x
E sx sx,LDA
@r#5 E drr ~ r! e sx @ r # , ~3.10!
The crucial point in the presently introduced GKS scheme
is that it incorporates part of the discontinuity D KS
into the eigenvalue gap of the N-particle GKS system,
xc already

thereby substantially reducing the discrepancy between

E c @ r # .E LDA
c @r#, ~3.11!
theory and experiment.
In order to demonstrate this, we write the exchange part
T @ r # .T sx @ r # , ~3.12! D KS 30,31
x of the discontinuity in the form
with23
x 5 ^ N11 u D v̂ x u N11 & 2 ^ N u D v̂ x u N & ,
D KS

e x @ r # 52 SD
3 3
4 p
1/3
r 1/3, ~3.13! x @ $ f i % # 2 v̂ x @ r ~ N !# .
D v̂ x 5 v̂ NL KS

Here, u N11 & and u N & denote the Nth and (N11)th KS or-
~4.3!

x @ $ f i % # is the

SD 1/3
bital of an N-particle system and v̂ NL KS
3 3 N-particle nonlocal exchange operator specified in Eq. ~3.5!
e sx @ r # 52 r 1/3F ~ z ! , ~3.14!
4 p but evaluated with exact Kohn-Sham orbitals. In addition,
53 GENERALIZED KOHN-SHAM SCHEMES AND THE BAND-GAP PROBLEM 3769

v̂ x @ r # is the exact, local Kohn-Sham exchange potential of ~3.7!, and applied it to energy gaps and structural properties
the N-particle density r (N) entering the standard Kohn- of several semiconductors.
Sham equations, Eq. ~3.3!.
The GKS equations ~3.4! for the HF-KS scheme can be A. Numerical procedure
written in the form of standard Kohn-Sham equations plus
correction terms, The basis of our computations is the plane-wave pseudo-
potential method.32,33 Technically, the sX-LDA scheme re-
@ T̂1 v̂ 1û @ r # 1 v̂ x @ r # 1 v̂ c @ r ## f i 1D v̂ x @ $ f i % # f i 1D v̂ c f i sembles a Hartree-Fock calculation with a screened ex-
change potential. Since we are going to consider bulk solids,
5« i f i ~4.4! the orbitals in Eq. ~3.7! are Bloch states and will be denoted
by c n,k . In a plane-wave basis, the sX-LDA Hamiltonian
and we see indeed that the operator D v̂ x from Eq. ~4.3! en- matrix in Eq. ~3.7! reads
ters as a potential term in the HF-KS equations. Within a
perturbative approach, this operator simply yields an additive ^ k1Gu Ĥ u k1G8& 5 21 u k1Gu 2 d G,G81 ^ k1Gu V̂ psu k1G8&
term to the KS eigenvalues,
r ~ G2G8!
14 p 1 v sx,LDA~ G2G8!
i 1 ^ i u D v̂ x u i & .
« i .« KS ~4.5! u G2G8u 2 xc

We note that there is a subtle difference between D v̂ x in Eq. 4p

~4.4! and Eq. ~4.3!. The former operator contains HF-KS 2 ((
V n,q
G 1
orbitals instead of standard Kohn-Sham orbitals. In addition,
D v̂ c is the operator that emerges from the difference between ^ n,qu q1G1G1 &^ q1G81G1 u n,q&
the standard Kohn-Sham and the HF-KS correlation poten- 3 .
u q2k1G1 u 2 1k 2TF
tial. However D v̂ c and the differences in the operators D v̂ x
are likely to be very small22 and are neglected here. ~5.1!
Finally, the eigenvalue gap « g of the HF-KS equations is Here, the vectors k and q lie in the first Brillouin zone and
given by G,G8,G1 denote reciprocal-lattice vectors. We have em-
ployed the standard ab initio norm-conserving semilocal
« g 5« N11 ~ N ! 2« N ~ N !
pseudopotentials V̂ ps of Bachelet, Hamann, and Schlüter
.« N11
KS
N ~ N ! 1 ^ N11 u D v̂ x u N11 & 2 ^ N u D v̂ x u N & ,
~ N ! 2« KS ~BHS!.34 The numerically most demanding part is the com-
putation of the plane-wave matrix elements of the nonlocal
g 1D x .
5« KS KS
~4.6! exchange operator, given by the last term in Eq. ~5.1!. For all
investigated semiconductors, we have evaluated the k sums
Thus, the exchange contribution D KS x to the discontinuity in the electronic density and in Eq. ~5.1! with six special k
D KS
xc of Eq. ~4.1! is already contained in the eigenvalue gap in points in the irreducible wedge of the Brillouin zone.35 The
the HF-KS method. Analogously, the eigenvalue gap in the plane-wave Hamiltonian matrix has been cut off at a kinetic
sX-LDA formalism of Sec. III C can be shown to contain energy of 20 Ry for Si, Ge, and GaAs, 30 Ry for InSb and 45
parts of the discontinuity D KS
xc . Invoking perturbation theory Ry for InP, respectively. It turned out that in the case of InP
once more, one may write and InSb a smaller cutoff of 20–30 Ry for the G1 sum in Eq.
~5.1! changes the calculated band gaps by less than 1 meV.
i .« i 1 ^ i u D v̂ x u i & ,
« sx KS sx
We have used the analytic Perdew-Zunger parametrization36
of the electron gas correlation energy37 in Eq. ~3.11!. For a
x 5 v̂ x
D v̂ sx sx,NL
2 v̂ sx
x , ~4.7! finite screening wave vector k TF , the integrand in the last
term of Eq. ~5.1! is smooth, justifying the use of only six k
where v̂ sx
x is the operator to the functional derivative of the points. In the limiting case of k TF50, however, it contains a
screened exchange energy E sx x @ r #@ Eqs. ~2.6! and ~3.10!#. singularity. It can be handled efficiently by adding and sub-
The GKS single-particle equations resemble the GW tracting another term in the wave-vector integral that com-
quasiparticle equations.4 – 6 However, the GKS approach is a pensates this singularity and can be integrated exactly.38 In
ground-state formalism and yields the correct ground-state order to be able to investigate the sX-LDA method for any
density and energy. One may use the perturbative expression value of k TF , we have used this approach throughout.
Eq. ~4.7! to calculate the sX-LDA eigenvalues with the stan-
dard LDA Bloch orbitals. This grossly reduces the numerical
B. Band gaps and structural properties in sX-LDA
effort since the nonlocal screened Fock exchange operator is
not needed in the self-consistency procedure. We show in We have performed total-energy calculations for several
Sec. V D that even this perturbative approach yields very semiconductors employing the sX-LDA method of Sec.
favorable results for the energy gaps. III C. All results in this section have been obtained with the
Thomas-Fermi ~TF! dielectric function, using the average
V. IMPLEMENTATION AND RESULTS FOR BULK
valence electron density in the screening constant k TF . In
SEMICONDUCTORS
the next section we will discuss alternative screening models.
Figure 1 shows the significant increase in the principal
In order to test the proposed GKS scheme, we have band gaps of Si, Ge, GaAs, InP, and InSb obtained by the
implemented it in the form of the sX-LDA equations, Eq. sX-LDA method compared to LDA results that have been
3770 SEIDL, GORLING, VOGL, MAJEWSKI, AND LEVY 53

FIG. 1. Fundamental energy gaps in Si, Ge, GaAs, InP, and InSb
calculated in sX-LDA method compared to the LDA and experi- FIG. 2. Direct (G 1c ) and indirect (L 1c and X 1c ) energy gaps,
mental values. relative to the valence-band top, in GaAs calculated in sX-LDA
method compared to the LDA and experimental values.
obtained with the same BHS pseudopotentials and with the
same experimental lattice constants. For materials that con- linearized augmented plane-wave method41 predict a nega-
tain heavier anions ~Ge, GaAs, and especially InSb!, the tive gap at G, whereas the present nonrelativistic sX-LDA
spin-orbit interactions cannot be neglected. It is well estab- procedure gives a positive gap of 0.495 eV. Taking the ex-
lished that the spin-orbit effects can be added perturbatively perimental value for the spin-orbit splitting in InSb, one ob-
a posteriori in the LDA-pseudopotential framework. This tains an sX-LDA band gap of 0.23 eV that perfectly matches
procedure reproduces the experimental spin-orbit splitting the experimental band gap of 0.235 eV. In Ge, the calculated
energies in semiconductors very well.39,40 Since we are inter- energy gap at G is still smaller than the one at the L point by
ested only in the principal energy gaps, we have sidestepped an amount of 0.373 eV. Thus, this theory is still not able to
this calculation and simply decreased the calculated nonrel- reproduce the indirect energy gap in Ge. Table I and Fig. 2
ativistic gaps by one-third of the valence-band spin-orbit illustrate the significant overall improvement of the predicted
splitting at G ( 31 D 0 ). The energy gaps between the lowest band gaps throughout the Brillouin zone by the sX-LDA
conduction-band states in G, X, and L points and the top of method.
the valence band and the valence-band widths are shown in As a further check, we list the sX-LDA electronic energy
Table I. As a first remark, we notice that Ge and InSb have eigenvalues at high-symmetry points for the occupied and
metallic properties within the LDA. In InSb, even nonrela- lowest-lying unoccupied states of GaAs in Table II and com-
tivistic LDA calculations that use pseudopotentials39,41 or the pare them with experimental results and GW calculations of

TABLE I. Energy gaps between the lowest conduction states and the valence-band edge and valence-band
widths ~VBW! of Si, Ge, GaAs, InP, and InSb from present sX-LDA with Thomas-Fermi screening and LDA
calculations compared to available experimental results taken from Ref. 42 except where noted. Energies are
in eV.

Si Ge GaAs InP InSb

E(G) 3.37 0.28 1.11 1.60 0.21

E LDA(G) 2.54 20.06 0.44 0.94 20.32
E expt(G) 3.05a, 3.4 0.89 1.52 1.42 0.24
E(X) 1.55 1.45 2.35 2.75 1.82
E LDA(X) 0.61 0.61 1.32 1.64 1.06
E expt(X) 1.25a 1.360.2 1.98 2.38 1.79
E(L) 2.18 0.66 1.70 2.45 1.02
E LDA(L) 1.44 0.07 0.91 1.56 0.35
E expt(L) 1.6560.01 0.74 1.81 2.03 —
2.1b, 2.460.15c
VBW 12.47 13.41 13.40 11.91 11.02
VBW LDA 11.94 12.80 12.40 11.20 10.48
VBW expt 12.560.6 12.6,12.960.3d 13.1 11.0 11.7, 11.2e
a
Reference 43.
b
Reference 44.
c
Reference 45.
d
Reference 46.
e
Reference 47.
53 GENERALIZED KOHN-SHAM SCHEMES AND THE BAND-GAP PROBLEM 3771

TABLE II. Electronic energy levels without spin-orbit correc- employ BHS pseudopotentials, the core electrons are treated
tions ~in eV! for GaAs calculated at the experimental lattice con- within LDA whereas the valence electrons are subject to the
stants. Experimental values are from Ref. 48 except where noted. sX-LDA. As a consequence, we have determined only rela-
GW stands for quasiparticle band-structure calculations of Ref. 6 in tive structural energies but not the total cohesive energy. It
the GW approximation. would clearly be desirable to employ the same sX-LDA
Hamiltonian for both core and valence electrons. On the
LDA sX-LDA Expt. GW other hand, it is not clear why the same constant k TF should
work well for all electrons. To be consistent, one could use a
G 1v 212.56 213.29 213.1 213.03
screening parameter that depends on the local density rather
G 15v 0.0 0.0 0.0 0.0
than the average valence density but this procedure is com-
G 1c 0.55 1.21 1.52 1.22
putationally impractical and we have not pursued it.
G 15c 3.54 4.67 4.72 4.48 There are several known methods that model the valence-
core interactions and core polarization effects more accu-
X 1v 210.25 211.02 210.75 210.69 rately than simple LDA, such as partial core corrections,52
X 3v 26.70 26.72 26.70 27.19 mixed Hartree-Fock LDA calculations,53 model core polar-
X 5v 22.58 22.47 22.80 22.87 ization potentials,54 or self-interaction corrected pseudo-
X 1c 1.43 2.46 2.01 2.01 potentials.55 Partial core corrections eliminate a large part of
X 3c 1.64 2.55 2.58 2.24 the discrepancy between the experimental lattice constants
and the values calculated within LDA.56 –58 Another recently
L 1v 210.95 211.71 211.24 211.41 proposed scheme that leads to similar improvements is based
L 1v 26.52 26.46 26.70 26.97 on self-interaction-corrected ~SIC! pseudopotentials.59
L 3v 21.09 21.05 21.30 21.28 In order to study the effect of our choice of BHS pseudo-
L 1c 1.02 1.81 1.85 1.64 potentials on the lattice constants and the energy gaps, we
L 3c 4.52 5.63 5.45a 5.40 have constructed a SIC pseudopotential for the case of Ge,
using a recently developed rigorous procedure59 that incor-
a
Reference 43. porates the nondiagonal Lagrange multipliers of the SIC
method. The most pronounced difference in the band struc-
Ref. 6. One can see that the sX-LDA method predicts energy ture is obtained for the G point, where the direct energy gap
gaps in similar good agreement with experiment as the GW (G 82c 2G 825v ) opens by an additional 260 meV in comparison
method does. to the value obtained with the standard BHS pseudopotential.
Table III displays the calculated structural properties of Changes of other gaps are minor, e.g., the gaps L 1c 2G 825v
various bulk semiconductors as obtained by the sX-LDA and X 1c 2G 825v change by 50 and 290 meV, respectively. The
method. The relative errors in the lattice constants a are be- lattice constant increases by 2%. This is consistent with the
low 0.5%, when compared to experiment. Since these errors increase found earlier within an LDA calculation with SIC
are so small, we have calculated all energy gaps at the ex- pseudopotentials.59 Overall, the results of the present sX-
perimental lattice constants. The effect of zero point motion LDA calculations do not seem to get significantly altered by
of the ions on a is one order of magnitude smaller than the using more sophisticated pseudopotentials.
remaining discrepancies between theory and experiment and
has been neglected. Table III also shows that the bulk moduli
C. Perturbative sX-LDA eigenvalues: Results
predicted by sX-LDA underestimate the experimental values
by typically 20%, and agree less well with experiment than As pointed out in Sec. IV, one may obtain the eigenvalues
the standard LDA results. of the sX-LDA equations in a much simpler albeit approxi-
We point out that the present sX-LDA calculations con- mate way from Eq. ~4.7! with LDA wave functions by using
tain, strictly speaking, an inconsistency or at least an ambi- first-order perturbation theory. For all semiconductors inves-
guity in treating the core and valence electrons. Since we tigated in this paper, these approximate energy gaps differ by
less than 10% from the full self-consistent sX-LDA eigen-
TABLE III. Theoretical lattice constants ~in Å! and bulk moduli values. For example, in the case of Ge, the differences in
~in GPa!, both obtained from fit to Murnaghan’s equation of state perturbative and ‘‘exact’’ sX-LDA energy gaps between the
~Ref. 49!, of Si, Ge, GaAs, and InP in sX-LDA with Thomas-Fermi lowest conduction states in high-symmetry points L, X, and
screening. Experimental values are from Refs. 42 and 50. G and the valence-band edge are equal to 20.02, 20.08, and
0.09 eV, respectively. This good agreement between pertur-
Si Ge GaAs InP bation theory and the full sX-LDA calculation indicates that
the orbitals emerging from the sX-LDA and LDA proce-
a0 5.421 5.635 5.627 5.776
dures, respectively, are quite similar. This is consistent with
a LDA 5.37a 5.567a 5.51a 5.74b
0 previous results by von der Linden et al., who pointed out
a expt
0 5.43 5.657 5.652 5.869 the similarity between LDA and Hartree-Fock wave func-
B0 89.3 62.7 63.1 65.5 tions for semiconductors.60
B LDA
0 96.8a 76.2a 77.8a 76.0b
B expt
0 87.6 –97.9 73.4 –75.8 75.3–76.9 72.5 D. Alternative screening models: From HF-KS to sX-LDA
a
Reference 51. In the formulation of the sX-LDA equations @~3.7! and
b
Reference 41. ~5.1!#, we have employed a Thomas-Fermi type of screening.
3772 SEIDL, GORLING, VOGL, MAJEWSKI, AND LEVY 53

TABLE IV. Theoretical lattice constant a 0 ~in Å!, bulk modulus gaps change linearly as a function of k TF between these two
B 0 ~in GPa!, and band gaps ~in eV! of Ge from the present sX-LDA opposite limits. In particular, the calculated energy gaps in
calculations with various model dielectric functions. TF, HUBB, the HF-KS scheme, which incorporates an LDA correlation
and SEM refer to the Thomas-Fermi, Hubbard ~Ref. 61!, and semi- potential, are only marginally different from those obtained
conductor ~Ref. 62! model dielectric functions, respectively. Experi- from standard Hartree-Fock calculations for the valence elec-
mental values are from Ref. 42. trons ~also employing BHS pseudopotentials!. In Ge, for ex-
ample, the energy gap at G is 7.02 eV in HF-KS and 6.97 eV
TF HUBB SEM Expt. in Hartree Fock. Similar results hold for other k points and
a0 5.63 5.66 5.38 5.657 materials. Thus, the improvement of the band gaps calculated
B0 63 60 104 77 with sX-LDA method is due to the fact that the screening of
the exact exchange potential already takes into account es-
sential correlation effects.
G 825v →L 1c 0.76 0.63 3.76 0.74
G 825v →G 2c
8 0.38 0.23 3.39 0.89
G 825v →X 1c 1.55 1.43 4.59 1.3 VI. CONCLUSIONS

Employing a constrained-search formulation of DFT, we

More generally, we can write the screened nonlocal ex- have developed a framework to derive various exact realiza-
change contribution in Eq. ~5.1! in the form tions of DFT in addition to the standard Kohn-Sham scheme.
Such generalized Kohn-Sham ~GKS! schemes minimize not
only the kinetic energy for fixed density such as the standard
^ k1Gu H sx,NL
x u k1G8&
Kohn-Sham method but also part of the electron-electron
4p ^ n,qu q1G1G1 &^ q1G81G1 u n,q& energy. As a result, one gains more flexibility in constructing
52 ((
V n,q
G Q 2e~ Q !
generalized exchange-correlation functionals. Generally, the
1 GKS scheme leads to single-particle equations that contain
~5.2! nonlocal potentials such as the Fock exchange operator. A
key point of the GKS method is that the single-particle ei-
with Q5 u q2k1G1 u and with a general dielectric function genvalues reflect much more faithfully the lowest excitation
e (Q). energies ~such as band gaps! of the many-electron system
For germanium, we have investigated two screening mod- than the Kohn-Sham method. The reason is that the discon-
els besides the Thomas-Fermi case, namely a metallic Hub- tinuity of the exchange-correlation potential with respect to
bard model dielectric function61 and a semiconductor model the particle number is, to a large extent, already incorporated
dielectric function.62 The latter dielectric function contains in the GKS single-particle eigenvalues.
two parameters, a 51.563 and e 0 512, where e (Q)→ e 0 for We have shown that a previously suggested method based
q→0. For each of the different models, the functional de- on a total-energy functional with a nonlocal screened ex-
rivative v sx,LDA
x (r) of E sx,LDA
x @ r # @Eq. ~3.10!# has been de- change and LDA correlation energy8 is firmly rooted within
termined by numerical integration. the presented GKS framework. In particular, we have applied
The results of these calculations are summarized in Table this approach, that we call sX-LDA method, to Si, Ge, GaAs,
IV. We find that the energy gaps depend mostly on the long- InSb, and InP and obtained eigenvalue gaps in excellent
wavelength behavior of the model dielectric functions. Both overall agreement with true band gaps in these materials. The
the Thomas-Fermi and the Hubbard model yield complete improvements are comparable to the perturbative GW ap-
screening for Q→0 and give very similar results. By con- proach but the sX-LDA method allows a self-consistent de-
trast, the present model of the semiconductor dielectric func- termination of ground-state properties as well.
tion does not guarantee complete screening of a long- The lattice constants, as calculated by sX-LDA, agree
wavelength external potential in a semiconductor. This is due well with experiment, whereas the bulk moduli are underes-
to the neglect of all off-diagonal elements GÞG8 in timated by typically 20%. In part, we believe this to be
e 21 (Q1G,Q1G8) which play an essential role in a self- caused by using pseudopotentials that do not model core-
consistent calculation. Consequently, the present diagonal valence polarization effects with sufficient accuracy.
semiconductor dielectric function screens the exchange inter- We have developed a perturbative scheme for the sX-
action less efficiently than the Thomas-Fermi model and LDA single-particle eigenvalues that basically requires only
yields energy gaps that are much closer to the unscreened a standard LDA calculation and yields semiconductor band
Hartree-Fock–Kohn-Sham scheme. All of these models do gaps that agree well with the self-consistent sX-LDA eigen-
not appreciably change the k dependence of the matrix ele- value gaps.
ment in Eq. ~5.2!, but give more or less rigid shifts of the
band structure.
ACKNOWLEDGMENTS
In the calculations with the Thomas-Fermi screening, the
screening constant k TF has been determined according to the Helpful comments by Professor Del Sole and financial
average valence electron density. If one considers k TF an support by the Deutsche Forschungsgemeinschaft ~SFB 348!,
adjustable parameter, one obtains the standard LDA method Bayerische Forschungsverbund ~FOROPTO!, and Volks-
in the limit k TF5`. The opposite limit k TF50, on the other wagenstiftung are gratefully acknowledged. One of us ~A.G.!
hand, corresponds to the HF-KS scheme of Sec. III B with thanks the Deutsche Forschungsgemeinschaft for a Habilita-
the approximation E HFc @ r # 5E c @ r # . We find that the band
LDA
tionsstipendium.
53 GENERALIZED KOHN-SHAM SCHEMES AND THE BAND-GAP PROBLEM 3773

APPENDIX A: PROOF OF HOHENBERG-KOHN We note that one has to form a scalar product of Eq. ~A1!
THEOREM FOR THE GKS SCHEME with the spin part of F S0 if the latter function contains
The Hohenberg-Kohn theorem corresponding to the GKS spinors, in order to make the division well defined. Equation
equations is given by the statement that two different poten- ~A2! determines the effective potential uniquely and there-
fore establishes the Hohenberg-Kohn theorem for the GKS
tials v eff(r) cannot yield the same density r S0 (r) as a result of
equations.
minimizing E S @ $ f i % ; v eff# of Eq. ~2.8!. Its proof consists of
The given proof of the Hohenberg-Kohn theorems is a
two steps. First, it is shown that two different local potentials
direct consequence of the requirements ~C1! and ~C2!. One
v eff(r) that give the same density r S0 (r) also yield the same may therefore consider ~C1! and ~C2! as an equivalent for-
Slater determinant F S0 . Potentials that differ by a constant mulation of the Hohenberg-Kohn theorem in the constrained-
are considered equivalent. Second, we show that two differ- search formulation of the GKS formalism.
ent potentials v eff(r) cannot give the same Slater determinant
F S0 , in contradiction to the first statement. Altogether, we
APPENDIX B: VARIATIONAL PROPERTIES
establish in this way the one-to-one mapping between
OF GKS SOLUTIONS
v eff(r) and r S0 (r).
First, it follows from the minimization of E S0 @v eff# given Here, we show that the second part of requirement ~C3! in
in Eq. ~2.16! that once the ground-state density r S0 (r) is Sec. II guarantees the calculated density r S0 ( @v eff# ;r) to rep-
known, the ground-state wave function F S0 is completely de- resent the minimum of F @ r # 1 * drv (r) r (r) instead of
termined as a functional of that density by Eq. ~2.7!, where merely some stationary solution. According to ~C3!, there
r (r) is set equal to r S0 (r). This implies that two different exists some external potential ṽ (r) for the interacting Schrö-
effective potentials that yield the same ground-state density dinger equation ~2.1! that yields a ground-state density equal
r S0 (r) also give the same ground-state wave function F S0 . to r S0 ( @v eff# ;r). We now show that, up to a constant,
For the second step of the proof, we note that F S0 is built ṽ (r)5 v (r). From the Euler equation that corresponds to Eq.
of the orbitals obeying the single particle Eqs. ~2.9!. There- ~2.2a!, it follows that this external potential ṽ (r) is given by
fore, the determinant F S0 is the ground-state wave function of the functional derivative 2 d F @ r # / d r (r) evaluated at the
the N-electron Schrödinger equation density r S0 ( @v eff# ;r). The potential v eff(r) is equal to the
functional derivative 2 d F S @ r # / d r (r) evaluated at the same
$ Ô S @ $ f i % # 1 v̂ eff% F S0 @v eff# 5E Ssum@v eff# F S0 @v eff# . ~A1! density because of the Euler equation corresponding to Eq.
~2.16!. This leads, together with Eq. ~2.10!, to the identity
Here, Ô S @ $ f i % # is the N-electron equivalent of the one-
particle operator that appears in the GKS equations ~2.9! or ṽ~ r! 5 v eff~ r! 2 v R ~ r! ,
~2.12! but is denoted by the same symbol. Note that the
energy E Ssum@v eff# is nothing but the sum of the eigenvalues v R ~ r! 5 d R S @ r # / d r @ r# u r ~ r! 5 r S ~@v eff# ;r! . ~B1!
« j of the GKS equations ~2.9! or ~2.12!. 0

By dividing the preceding equation by F S0 , one obtains In the course of solving the GKS equations self-consistently,

F( G
N Eq. ~2.14! connects v eff(r) to the given real external potential
@ Ô @ $ f i % # F S0 ~ r1 . . . ri . . . rN !# v (r). Thus, once convergence of the GKS equations has
2 5 v eff~ ri ! 2E S0 .
F S0 ~ r1 . . . ri . . . rN ! i51 been achieved, the identity v (r)5 v eff(r)2 v R (r) holds and
~A2! comparison with Eq. ~2.1! leads to ṽ (r)5 v (r).

1
W. Kohn and l. J. Sham, Phys. Rev. 140, A1133 ~1965!. edited by R. M. Dreizler and J. da Providencia ~Plenum Press,
2
P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 ~1964!. New York, 1985!, Chap. 2, pp. 11–30.
3 13
For review see, e.g., R. O. Jones and O. Gunnarsson, Rev. Mod. R. G. Parr and W. Yang, Density-Functional Theory of Atoms and
Phys. 61, 689 ~1989!. Molecules, International Series of Monographs on Chemistry,
4
R. W. Godby, M. Schlüter, and L. J. Sham, Phys. Rev. B 37, No. 16 ~Oxford University Press, New York, 1989!.
10 159 ~1988!. 14
R. M. Dreizler and E. K. U. Gross, Density Functional Theory
5
M. S. Hybertsen and S. G. Louie, Phys. Rev. Lett. 55, 1418 ~Springer, Berlin, 1990!.
~1985!; Phys. Rev. B 34, 5390 ~1986!. 15
M. Levy, in Advances in Quantum Chemistry, edited by S. B.
6
X. Zhu and S. G. Louie, Phys. Rev. B 43, 14 142 ~1991!. Trickey ~Academic Press, San Diego, 1990!, Vol. 21, p. 69.
7
L. Hedin, Phys. Rev. 139, A796 ~1965!. 16
E. Lieb, in Density Functionals for Coulomb Systems, edited by
8
B. M. Bylander and L. Kleinman, Phys. Rev. B 41, 7868 ~1990!. H. Feshbach and A. Shimony ~MIT, Cambridge, 1982!.
9
F. R. Vukajlovic, E. L. Shirley, and R. M. Martin, Phys. Rev. B 17
M. Levy, Phys. Rev. A 26, 1200 ~1982!.
43, 3994 ~1991!. 18
E. Lieb, Int. J. Quantum Chem. 23, 243 ~1983!.
10
H. Yamagami, Y. Takada, H. Yasuhara, and A. Hasegawa, Phys. 19
A. Görling, Phys. Rev. A 47, 2783 ~1993!.
Rev. A 49, 2354 ~1994!. 20
S. Baroni and E. Tuncel, J. Chem. Phys. 79, 6140 ~1983!.
11
A. Görling and M. Levy, Phys. Rev. B 47, 13 105 ~1993!. 21
H. Stoll and A. Savin, in Density Functional Methods in Physics
12
M. Levy and J. P. Perdew, in Density Functional Methods in ~Ref. 12!, Chap. 7, pp. 177–207.
Physics, Vol. 123 of NATO Advanced Science Institutes Series, 22
A. Görling and M. Ernzerhof, Phys. Rev. A 51, 4501 ~1995!.
3774 SEIDL, GORLING, VOGL, MAJEWSKI, AND LEVY 53

23
J. E. Robinson, F. Bassani, R. S. Knox, and J. R. Schrieffer, Phys. by A. Goldman and E.-E. Koch ~Springer, Berlin, 1989!.
Rev. Lett. 9, 215 ~1962!. 43
J. E. Ortega and F. J. Himpsel, Phys. Rev. B 47, 2130 ~1993!.
24
J. P. Perdew and M. Levy, Phys. Rev. Lett. 51, 1884 ~1983!. 44
R. Hulthen and N. G. Nilsson, Solid State Commun. 18, 1341
25
L. J. Sham and M. Schlüter, Phys. Rev. Lett. 51, 1888 ~1983!. ~1976!.
26 45
R. W. Godby, M. Schlüter, and L. J. Sham, Phys. Rev. Lett. 56, D. Straub, L. Ley, and F. J. Himpsel, Phys. Rev. Lett. 54, 142
2415 ~1986!. ~1985!.
27 46
O. Gunnarsson and K. Schönhammer, Phys. Rev. Lett. 56, 1968 A. L. Wachs, T. Miller, T. C. Hsieh, A. P. Shapiro, and T.-C.
~1986!. Chiang, Phys. Rev. B 32, 2326 ~1985!.
28
K. Schönhammer and O. Gunnarsson, J. Phys. C 20, 3657 ~1987!. 47
D. E. Eastman, W. D. Grobman, J. L. Freeouf, and M. Erbudak,
29
J. P. Perdew and M. Levy, in Many-Body Phenomena at Surfaces, Phys. Rev. B 9, 600 ~1974!.
edited by H. S. D. Langreth ~Academic, Orlando, 1984!, Chap. 48
E. P. O’Reilly, in Properties of Gallium Arsenide, 2nd ed., EMIS
1, pp. 71– 89. Datareviews Series No. 2 ~INSPEC, London, 1990!, p. 113.
30
J. P. Perdew, in Density Functional Methods in Physics ~Ref. 12!, 49
F. D. Murnaghan, Proc. Natl. Acad. Sci. U.S.A. 30, 244 ~1944!.
50
Chap. 10, pp. 265–308. J. D. Wiley, in Semiconductors and Semimetals, edited by R. K.
31
A. Görling and M. Levy, Phys. Rev. A 52, 4493 ~1995!. Willardson and A. C. Beer ~Academic, New York, 1975!, Vol.
32
J. Ihm, A. Zunger, and M. L. Cohen, J. Phys. C 12, 4409 ~1979!. 10, p. 134.
33
W. E. Pickett, Comp. Phys. Rep. 9, 117 ~1989!. 51
V. Fiorentini, Phys. Rev. B 46, 2086 ~1992!.
34 52
G. B. Bachelet, D. R. Hamann, and M. Schlüter, Phys. Rev. B 26, S. G. Louie, S. Froyen, and M. L. Cohen, Phys. Rev. B 26, 1738
4199 ~1982!. ~1982!.
35
H. J. Monkhost and J. D. Pack, Phys. Rev. B 13, 5188 ~1976!. 53
B. M. Bylander and L. Kleinman, Phys. Rev. B 43, 12 070
36
J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 ~1981!. ~1991!.
37
D. M. Ceperly and B. J. Alder, Phys. Rev. Lett. 45, 566 ~1980!. 54
E. L. Shirley and R. M. Martin, Phys. Rev. B 47, 15 413 ~1993!.
38
F. Gygi and A. Baldereschi, Phys. Rev. B 34, 4405 ~1986!. 55
M. M. Rieger and P. Vogl, Phys. Rev. A 52, 282 ~1995!.
39
M. P. Surh, Ming-Fu Li, and S. G. Louie, Phys. Rev. B 43, 4286 56
A. Qteish and R. J. Needs, Phys. Rev. B 43, 4229 ~1991!.
~1991!. 57
A. Garcia and M. L. Cohen, Phys. Rev. B 47, 6751 ~1993!.
40 58
L. A. Hemstreet, C. Y. Fong, and J. S. Nelson, Phys. Rev. B 47, A. Dal Corso, S. Baroni, R. Resta, and S. de Gironcoli, Phys.
4238 ~1993!. Rev. B 47, 3588 ~1993!.
41
S. Massidda, A. Continenza, A. J. Freeman, T. M. de Pascale, F. 59
M. M. Rieger and P. Vogl, Phys. Rev. B 52, 6 567 ~1995!.
Meloni, and M. Serra, Phys. Rev. B 41, 12 079 ~1990!. 60
W. von der Linden, P. Fulde, and K.-P. Bohnen, Phys. Rev. B 34,
42
Numerical Data and Functional Relationships in Science and 1063 ~1986!.
Technology, edited by K.-H. Hellwege, Landolt-Börnstein, New 61
J. Hubbard, Proc. R. Soc. London Ser. A 243, 336 ~1957!.
62
Series, Group III, Vols. 17 and 22, edited by O. Madelung, M. G. Cappellini, R. DelSole, L. Reining, and F. Bechstedt, Phys.
Schulz, and H. Weiss ~Springer, Berlin, 1982!, Vol. 23a edited Rev. B 47, 9892 ~1993!.