Computer Physics Communications 175 (2006) 114

www.elsevier.com/locate/cpc

Linear optical properties of solids within the full-potential linearized

augmented planewave method
Claudia Ambrosch-Draxl a, , Jorge O. Sofo b
a Chair of Atomistic Modelling and Design of Materials, University of Leoben, Erzherzog-Johann-Strae 3, A-8700 Leoben, Austria
b Department of Physics and Materials Research Institute, The Pennsylvania State University, 104 Davey Lab. PMB#172, University Park, PA 16802, USA

Received 29 July 2005; accepted 23 September 2005

Available online 5 May 2006

Abstract
We present a scheme for the calculation of linear optical properties by the all-electron full-potential linearized augmented planewave (LAPW)
method. A summary of the theoretical background for the derivation of the dielectric tensor within the random-phase approximation is provided.
The momentum matrix elements are evaluated in detail for the LAPW basis, and the interband as well as the intra-band contributions to the
dielectric tensor are given. As an example the formalism is applied to Aluminum. The program is available as a module within the WIEN2k code.
2006 Elsevier B.V. All rights reserved.

PACS: 71.15.Qe; 71.15.Ap; 78.20.-e; 78.40.-q

1. Introduction

Optical properties of solids are a major topic, both in basic research as well as for industrial applications. While for the former the
origin and nature of different excitation processes is of fundamental interest, the latter can make use of them in many opto-electronic
devices. These wide interests require experiment and theory to go hand in hand, and thus, in turn require reliable theoretical
concepts.
For crystalline solids, the most successful method for the calculation of materials properties is density functional theory (DFT).
Lattice properties like equilibrium volumes and lattice parameters, atomic positions, phonon frequencies, and elastic constants
differ from their experimental counterparts by a few percent only. While, however, such ground state (GS) properties, which are
based on the calculation of the total energy, are described very reliably, the treatment of excited states is not rigorously justified.
The main reasons are that the HohenbergKohn [1] theorem is exact only for the GS, and that the KohnSham eigenstates [2]
should not be interpreted as single-electron states. Moreover, approximations have to be made also in the ground state calculations
for describing exchange and correlation effects. As a consequence, the band gap problem for semiconductors is one of the most
intriguing problems in the field.
Nevertheless the interpretation of the KS states in terms of excited states has been successful for a variety of materials. Moreover,
it has been claimed that the KS wave functions hardly differ from the many-body wave functions [3,4]. It has to be pointed out
that the calculation of optical properties does not go beyond the interpretation of KS eigenvalues in terms of the band structure.
In this context, it may be hard to distinguish whether the shortcomings are due to the over-interpretation of GS properties or to
the approximations used to solve the KS equations, e.g., the local density approximation (LDA). The random phase approximation
(RPA) to the dielectric constant using KS states have been implemented in full potential all electron codes [57] demonstrating to
be a useful tool for the calculation of screening and optical response in solids.

* Corresponding author.
E-mail addresses: cad@mu-leoben.at (C. Ambrosch-Draxl), sofo@psu.edu (J.O. Sofo).

0010-4655/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.cpc.2006.03.005
2 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114

To gain insight and confidence regarding the use of KS quantities with the RPA to calculate optical properties, it is important
to have a series all-electron calculations. To this end we have chosen the full potential linearized augmented plane wave (FLAPW)
method. In it, no shape approximation for the potential is made. We have developed the formalism for treating optical properties
within the RPA taking into account inter- as well as intra-band contributions. We present an application to Aluminum, which we
selected for the following reasons: (i) Metals do not suffer from the band gap problem; (ii) metals are quite well described by the
LDA [9,10]; (iii) excitonic effects are negligible due to the effective screening; and (iv) free electron contributions are present.
The paper is organized as follows: In Section 2, the optical response within the RPA is discussed. Symmetry considerations,
and the description of other optical constants obtained via the KramersKronig relations follow hereafter. Section 3 is dedicated
to the treatment of the optical response within the LAPW method. First, the basic definitions of the needed quantities, are given,
then the momentum matrix elements are derived within the LAPW basis. The next section contains results obtained for Aluminum.
The extension of the method with localized basis functions (local orbitals) is described in Appendix A together with a summary of
optical constants and some mathematical relations which are useful in the derivation of the formulas.
All this formalism has been implemented as a module into the FLAPW code WIEN2k [11].

2. Theoretical background

The optical properties of solids are given by the response of the electron system to a time-dependent electromagnetic perturbation
caused by the incoming light. As such, the calculation of these properties is reduced to the calculation of a response function that
is the complex dielectric tensor or equivalently the polarizability. Since an exact expression for it is not known, one has to resort to
the usual techniques of many-body perturbation theory to derive approximations. The first of these is the RPA.
A particular case of the general expression for the dielectric function in the RPA is the well-known Lindhard formula [12],
2v(q)  f0 (k+q ) f0 (k )
(q, ) = 1 + , (1)
c k+q k
k
where
4e2
v(q) = (2)
|q|2
is the Coulomb interaction, c the unit cell volume, f0 the Fermi distribution function, and k the single particle energy. The
factor of two comes from the summation over spins. This dielectric function corresponds to a free electron gas in the Hartree
approximation, namely, the electronelectron interaction is reduced to the interaction of each electron with a homogeneous self-
consistent field. Many equivalent ways of deriving this expression have been given in the literature [13,14].
We start from the definition given by Hedin [15]

(r, t; r , t  ) = (r r )(t t  ) P (r, t; r , t  )v(r r ) dr , (3)

where v is the bare Coulomb interaction and P is the polarization propagator. The simplest approximation for P is the RPA that
corresponds to the Hartree approximation for the one-particle Green function G0 (r, t; r , t  ). In this approximation, the polarization
propagator P 0 is given by
P 0 (r, t; r , t  ) = i hG0 (r, t; r , t  )G0 (r , t  ; r, t). (4)
For the case of a solid with translational symmetry we can Fourier transform the equation for the dielectric function (3) to obtain:
G,G (q, ) = G,G v(q + G)PG,G
0
 (q, ), (5)
0
where PG,G (q, ) is the lattice Fourier transform of the bare polarizability defined in Eq. (4). It is given by

1  f0 (n,k+q ) f0 (n ,k ) G 
0
PG,G (q, ) = Mn ,n (k, q) MnG ,n (k, q), (6)
c  n,k+q n ,k
n ,n,k

with matrix elements

MnG ,n (k, q) = n , k|ei(q+G)r |n, k + q. (7)
The expression obtained so far is the dielectric tensor that connects the total scalar potential in the solid, V , with the potential
produced only by the external sources, V ext , through the expression

VGext (q, ) = G,G (q, )VG (q, ). (8)
G
 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114 3

If we denote by  the inverse of the dielectric tensor, we can invert the relation above to obtain an expression for the total potential

VG (q, ) = G,G (q, )VGext
 (q, ). (9)
G

When considering perturbations produced by light, only long wavelength components are present in the external perturbation
potential. Therefore, it is enough to assume that only the G = 0 component is different from zero. In this case the response is

VG (q, ) = G,0 (q, )V0ext (q, ). (10)

This expression shows that even in the case of an external potential with long wavelength variations in space, the response of the
solid will have shorter wavelength components that are commonly called local field effects. The macroscopic dielectric constant,
M , is the ratio between the average of the total potential in one unit cell, i.e. V0 (q, ), and the external field. In this way we identify
the macroscopic dielectric constant
1
M (q, ) = (11)
0,0 (q, )
by the (0, 0) element of the inverse dielectric tensor.
The expression for the macroscopic dielectric constant given in Eq. (11) is rather costly to evaluate. It involves the evaluation of
the dielectric tensor with components (G, G ) up to a certain cutoff and the subsequent inversion to obtain the (0, 0) component of
the inverse tensor. It is a common simplification to neglect the local field effects and replace the (0, 0) component of the inverse by
the inverse of the (0, 0) component to obtain
{nlf}
M (q, ) = 0,0 (q, ) = 1 v(q)P0,0
0
(q, ). (12)
We shall neglect local field effects for the rest of the paper in order to give the zeroth-order theory evaluated with the KohnSham
eigenvalues and eigenvectors from FLAPW, one of the most precise electronic structure methods available. Our results will be the
starting point for more refined theories in the sense that no basis set effect can be blamed for inaccuracies. Early attempts to include
local field corrections by Adler [13] and Wiser [14] found that these effects can be described by the LorenzLorentz formula with
a renormalized polarizability. The effect turned out to be quite small for most Fermi surfaces studied. The role of local field effects
also has been studied for Al [17], Li [7,18], and diamond [8].
The Lindhard expression for the dielectric constant of Eq. (1) can be obtained from the expression for the macroscopic dielectric
constant without local field effects (Eq. (12)) by assuming that the electron system is composed of only one band and the matrix
elements are equal to unity. This is the case for a simple parabolic band of free electrons.
Since the wave vector of light q is much smaller than any typical wave vector of electrons in the system, we need to evaluate the
polarizability P entering in Eq. (12) in the limit q 0. The matrix elements involved in the expression for P given in Eq. (6) and
defined in Eq. (7) can be evaluated for small q by k p perturbation theory giving:
h pl,n,k
0
Ml,n (k, 0) = l,n + (1 l,n ) q. (13)
m n,k l,k
It is convenient to split the sum over n and n into those terms with n = n, corresponding to intra-band electronic transitions,
and those with n = n, corresponding to inter-band transitions. In this manner, we can write
{nlf}
M (0, ) =  {intra} (0, ) +  {inter} (0, ), (14)
where the intra-band part of the dielectric function is given by
4e2  f0 (n,k+q ) f0 (n,k )  0 2
 {intra} (0, ) = 1 lim Mn,n (k, q) , (15)
q0 c |q|2 n,k+q n,k
n,k

and the inter-band part by

4e2  f0 (n ,k+q ) f0 (n,k )  0 2
 {inter} (0, ) = lim Mn,n (k, q) . (16)
q0 c |q|2 n ,k+q n,k
n ,n=n ,k

While taking the limit of q 0 we can use the expression for the matrix element evaluated in Eq. (13) and the expansion for the
band energies for small q to obtain, for the intra-band part,
   
{intra} 4 h2 e2  f0 q 2
 (0, ) = 1 lim pn,n,k , (17)
q0 c m2 2 n,k |q|
n,k
4 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114

where the derivative of the Fermi function with respect to the energy should be considered as a restriction of the summation to
only those states that are at the Fermi level. pn,l,k is the momentum matrix element between states of bands n and l with crystal
momentum k. These matrix elements are evaluated for the LAPW basis set in Section 3.2. The inter-band expression is

4 h2 e2   (pc,v,k q/|q|)2
 {inter} (0, ) = lim , (18)
q0 c m2
c,v
(c,k v,k )(c,k v,k )2
k

where, for a given k, c runs over the empty states and v over the occupied states.
From these expressions, we see that the limit is not well defined and the result depends on the direction of the vector q even
when the limit is taken. This defines the dielectric function for q 0 as a three-dimensional tensor, the dielectric tensor, given by
 
4 h2 e2  f0 4 h2 e2   pi;c,v,k pj ;c,v,k
i,j () = i,j p i;n,n,k p j ;n,n,k . (19)
2
c m 2 n,k c m 2
c,v
(c,k v,k )(c,k v,k )2
n,k k

It is clear that in the derivation of the RPA formulas presented here the unperturbed electronic states are described by the bare
one-particle propagator G0 . This procedure is also called the independent particle approximation because it neglects the electron
hole interaction during the absorption process. This should be a good approximation for metals due to the more effective screening
of the Coulomb interaction compared to semiconductors. Here, we are going to use the RPA expression for the dielectric constant
evaluated with the KS orbitals. The usage of the latter is an approximation, which has been addressed in a recent paper [16].
At the end of this section, we give the expressions which ultimately are computed. Most important is the imaginary part of the
inter-band contribution to the dielectric tensor components:

{inter} h2 e2   
ij () = 2 2
pi;n ,n,k pj ;n ,n,k f0 (n,k ) f0 (n ,k ) (n ,k n,k ). (20)
m  n,n k

The corresponding real parts are obtained by KramersKronig transformation.

As is evident from Eq. (19), the intra-band part is singular at = 0. At this point the plasma frequency pl;ij is defined by:

h2 e2 
2
pl;ij = pi;n,n,k pj ;n,n,k (n,k F ). (21)
m2 n
k

In practical calculations, a lifetime broadening is introduced by adopting a Drude-like shape for the intra-band contribution.
{intra}
ij () then reads

2
pl;ij
{intra}
Im ij () = , (22)
(2 + 2 )
2
pl;ij
{intra}
Re ij () = 1 . (23)
2 + 2
Note that the imaginary part is still singular; therefore one usually works with the optical conductivity. The BZ integrations in
Eqs. (20) and (21) are carried out by the improved linear tetrahedron method [19].
The dielectric tensor is hermitian with up to six components according to the symmetry of the crystal. For orthorhombic or
higher symmetry only diagonal components exist. In case of cubic symmetry, the optical properties are isotropic, i.e. there is only
one independent component Im xx , while for uniaxial symmetry (tetragonal, hexagonal) and orthorhombic symmetry there are two
and three independent components, respectively. In the monoclinic case non-diagonal elements occur. These materials are optically
active or birefringent. Only in the triclinic crystal are all six components different.
These symmetry considerations hold only for non-relativistic and scalar-relativistic calculations. In these cases the spin-up
and spin-down elements can be evaluated separately, where the corresponding real parts are obtained by the KramersKronig
transformation (see below). In case of magneto-optics the symmetry is reduced by the presence of a magnetic induction as well as
by the loss of time-reversal symmetry due to spinorbit coupling. The latter gives rise to antisymmetric non-diagonal components.
The corresponding imaginary parts are again obtained by KramersKronig analysis. In case of monoclinic or lower symmetry, the
off-diagonal elements can therefore have two independent contributions, one due to a non-orthogonal crystal axis, the other due to
the magneto-optical effect.
The dielectric function must obey certain sum rules that serve to check the accuracy of the calculation and also provide informa-
tion about the absorption process that can be checked experimentally:
 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114 5


Im () d = Neff ( ), (24)
0
  
1
Im d = Neff ( ), (25)
()
0
  
1 1
Im d = . (26)
() 2
0
The first two give an effective number of electrons contributing to the absorption process as a function of energy. Typically, in the
low energy region the contribution to the intra-band spectrum should sum up to the number of valence electrons.

3. Optical response within the LAPW basis

3.1. The LAPW basis set

In band structure calculations based on density-functional theory [1] the single-particle electronic states nk (r) and energies nk
are described by the solutions of the KohnSham (KS) equation [2]

h2 2
+ Veff (r) nk (r) = nk nk (r) (27)
2m
with the effective potential Veff (r) being the sum of the bare Coulomb potential of the atomic nuclei Vlatt (r), the Hartree po-
tential VH (r), and the exchange-correlation potential Vxc (r). In practical calculations, Eq. (27) is solved via the RayleighRitz
variational principle. In this procedure the KohnSham orbitals nk (r) are first expanded in terms of a physically appropriate finite
set of basis functions {k+G },

nk (r) = Cnk (G)k+G (r) (28)
G

with G and Cnk (G) denoting a reciprocal lattice vector, and the corresponding variational coefficient, respectively. To determine
Cnk (G) the ansatz (28) is inserted into Eq. (27) followed by the minimization of the total crystal energy with respect to the
variational coefficients. The eigenvectors and eigenvalues of the resulting matrix equation,

(Hk+G,k+G nk Sk+G,k+G )Cnk (G ) = 0, k BZ (29)
G
with
h2 2
Hk+G,k+G k+G | + Veff |k+G c , (30)
2m
Sk+G,k+G k+G |k+G c (31)
being Hamilton and overlap matrix, respectively, finally provide the numerical values for Cnk (G) and nk . c is the volume of the
unit cell which in the LAPW method [20,21] is partitioned into an interstitial region (I) and non-overlapping muffin-tin spheres
(MT ) centered on the atomic nuclei. The corresponding basis functions are defined as
1
k+G (r) = ei(k+G)r , rI (32)
c
and

k+G (S + r) = [Alm (k + G)ul (r, El ) + Blm

(k + G)ul (r, El )] Yl,m (r)
  
lm ,k+G
Wlm
 ,k+G
= Wlm (r, El )Yl,m (r), |r|  R , (33)
lm

where S denotes the position vector of the atomic nucleus . The radius of the corresponding muffin-tin sphere is R . The
product of the spherical harmonic Yl,m (r) and the radial function ul (r) is the solution of the Schrdinger equation for a spherically
symmetric potential in which the eigenvalue has been replaced by an appropriate energy parameter. The second radial function
6 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114

ul (r) is the derivative of the first one with respect to the one-electron energy. The coefficients Alm (k + G) and Blm
(k + G) are

determined for each atom in the unit cell by imposing the requirements that the basis function has to be continuous in value and
slope on the MT-surfaces:

4 
Alm (k + G) = i l Ylm (k + G)R2 ei(k+G)S al (k + G), (34)

4 

Blm (k + G) = i l Ylm (k + G)R2 ei(k+G)S bl (k + G), (35)

with
   
al = jl |k + G|R ul (R ) jl |k + G|R ul (R ), (36)
   
bl = jl |k + G|R ul (R ) jl |k + G|R ul (R ) (37)
and ul (r) denoting ul (r)/r. jl are the Spherical Bessel Functions. For the explicit evaluation of the Hamilton matrix elements
(30) an appropriate split representation of the effective potential is needed. Within the atomic spheres the potential is expanded in
spherical harmonics, while in the interstitial region it is represented by a Fourier series.

3.2. The momentum matrix elements

Due to the split representation of the LAPW basis functions, the momentum matrix element is a sum of contributions from the
atomic spheres as well as from the interstitial region:

n k|p|nk = n k|p|nkMT + n k|p|nkI . (38)

3.2.1. Contributions from the atomic spheres

The use of spherical harmonics in the LAPW basis set suggests that one should calculate the expressions n k|x + iy |nk and

n k|x iy |nk and derive the x- and y-component as linear combinations of the results. Taking into account the ansatz (28) for
the wave functions, the following expressions have to be evaluated:
   
k+G ,k+G k+G (S + r)x + iy k+G (S + r) ,
x+iy
(39)
   
k+G ,k+G k+G (S + r)x iy k+G (S + r) ,
xiy
(40)
   
k+G ,k+G k+G (S + r)z k+G (S + r) .
z
(41)
Expressing x + iy , x iy , and z in spherical coordinates
 
i 1 i i
x iy = sin e + e cos , (42)
r r sin
1
z = cos sin (43)
r r
we can elaborate the matrix elements (39)(41). We explicitly demonstrate the procedure for the first component:
   
k+G ,k+G = k+G (S + r) sin ei k+G (S + r)
x+iy
r
 
  1 i i  
+ k+G (S + r) e cos + k+G (S + r) . (44)
r sin
Here the first term of the operator acts on the radial coordinate only, whereas the second term acts on the angular coordinates
only. Applying this operator to the LAPW basis functions (33) one obtains:
 ,k+G  
1  ,k+G i
(x + iy )k+G (S + r) = Wlm (r) sin ei Yl,m (r) + Wlm (r)ei cos + Yl,m (r). (45)
r r sin
lm lm

Exploiting some relations between the spherical harmonics as summarized in Appendix A, and omitting their arguments for
simplicity, Eq. (45) becomes
 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114 7

 ,k+G l

(1)
(x + iy )k+G (S + r) = W (r) Wlm (r) Fl,m Yl+1,m+1
r lm r
lm
 ,k+G l+1

(2)
+ W (r) + Wlm (r) Fl,m Yl1,m+1 . (46)
r lm r
lm

The (x + iy) component will then be


x+iy
k+G ,k+G = k+G (S + r)[x + iy ]k+G (S + r) dr
MT

R   

 ,k+G l ,k+G
= r 2 dr d Wl,k+G
 m Yl ,m Wlm (r) Wlm (1)
(r) Fl,m Yl+1,m+1
r r
0 l  m lm


,k+G l + 1 ,k+G (2)
+ Wlm (r) + Wlm (r) Fl,m Yl1,m+1
r r


R

 ,k+G l ,k+G
r 2 dr Wl,k+G (r) Fl,m d Yl ,m Yl+1,m+1
(1)
=  m Wlm (r) Wlm
r r
l  m lm 0   
   l  ,l+1 m ,m+1

l ,m 
Rl,m (r)


R

 ,k+G l + 1 ,k+G
dr Wl,k+G (r) Fl,m d Yl ,m Yl1,m+1 ,
(2)
+ r 2
 m Wlm (r) + Wlm (47)
r r
l  m lm 0   
   l  ,l1 m ,m+1

l ,m
Tl,m (r)

where Tl,|m|>l is defined to be zero. Writing the sums over l, m, l  , and m explicitly and taking into account an upper limit lmax for
l we obtain:
l  
x+iy
max 
l l
max 
l

l ,m 
l
max 
l l
max 
l

l ,m 

k+G ,k+G = Rl,m (r)l  ,l+1 m ,m+1 + Tl,m (r)l  ,l1 m ,m+1
l  =0 m =l  l=0 m=l l  =0 m =l  l=0 m=l

1
l max 
l
 l+1,m+1 l,m 
= Rl,m (r) + Tl+1,m1 (r) . (48)
l=0 m=l

l+1,m+1 l,m
Evaluating Rl,m (r) and Tl+1,m1 (r) using Eqs. (47) and (33) we finally obtain:

   R R 
1
l max 
l

x+iy
Al+1,m+1 (k + G )Al,m (k + G) ul+1 (r)ul (r)r 2 dr l
(1)
k+G ,k+G = Flm ul+1 (r)ul (r)r dr
l=0 m=l 0 0
 R R 

+ Al+1,m+1 (k + G )Bl,m

(k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
 R R 

+ Bl+1,m+1 (k + G )Al,m (k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
 R R 

+ Bl+1,m+1 (k + G )Bl,m

(k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
  R R 

(2)
+ Fl+1,m1 Al,m (k + G )Al+1,m1 (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
8 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114

 R R 

+ Al,m (k + G )Bl+1,m1

(k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Bl,m (k + G )Al+1,m1 (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Bl,m (k + G
)Bl+1,m1 (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr .
0 0
(49)
Analogously we derive the other two components:
   R R 
1
l max 
l
A 
ul+1 (r)ul (r)r 2 dr
xiy (3)

k+G ,k+G = Flm l+1,m1 (k + G )Al,m (k + G)

l ul+1 (r)ul (r)r dr
l=0 m=l 0 0
 R R 

+ Al+1,m1 (k + G )Bl,m

(k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
 R R 

+ Bl+1,m1 (k + G )Al,m (k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
 R R 

+ Bl+1,m1 (k + G
)Bl,m (k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
  R R 

(4)
+ Fl+1,m+1 Al,m (k + G )Al+1,m+1 (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Al,m (k + G )Bl+1,m+1

(k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Bl,m (k + G )Al+1,m+1 (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Bl,m (k + G
)Bl+1,m+1 (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr ,
0 0
(50)

   R R 
1
l max 
l

ul+1 (r)ul (r)r 2 dr
z (5)

k+G ,k+G = Flm Al+1,m (k + G )Al,m (k + G) l ul+1 (r)ul (r)r dr
l=0 m=l 0 0
 R R 

+ Al+1,m (k + G
)Bl,m (k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
 R R 

+ Bl+1,m (k + G )Al,m (k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
 R R 

+ Bl+1,m (k + G )Bl,m

(k + G) ul+1 (r)ul (r)r 2 dr l ul+1 (r)ul (r)r dr
0 0
 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114 9

  R R 

(6)
+ Fl+1,m Al,m (k + G )Al+1,m (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Al,m (k + G
)Bl+1,m (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Bl,m (k + G )Al+1,m (k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr
0 0
 R R 

+ Bl,m (k + G )Bl+1,m

(k + G) ul (r)ul+1 (r)r 2 dr + (l + 2) ul (r)ul+1 (r)r dr . (51)
0 0
To obtain the atomic-sphere contributions to the momentum matrix elements, Eqs. (39)(41) have to be multiplied with the
variational coefficients and summed over all basis functions:
1  x+iy 
n k|x |nkMT = Cn k (G ) k+G ,k+G + k+G ,k+G Cnk (G),
xiy
2 
G ,G
1   x+iy 
n k|y |nkMT = Cn k (G ) k+G ,k+G k+G ,k+G Cnk (G),
xiy
(52)
2i 
G ,G


n k|z |nkMT = Cn k (G ) zk+G ,k+G Cnk (G).
G ,G

3.2.2. Contributions from the interstitial region

With the plane wave basis (32), the interstitial contribution to the matrix elements (38) can be worked out easily:

1  
n k||nkI = (k + G)Cn k (G )Cnk (G) ei(G G)r dr. (53)
c 
G ,G I
As is typical in the LAPW basis, the integration in Eq. (53) over the interstitial region is carried out by integrating over the whole
unit cell, then subtracting the integral over the atomic spheres (see Appendix A). This procedure finally leads to
   

i   j1 (|G G|R ) i(GG )S

n k||nkI = (k + G)Cnk (G) Cn k (G) c V 3V Cn k (G ) e ,
c 
(|G G|R )
G G =G
(54)
where V is the volume of the atomic sphere .

4. Application to Aluminum

As an example fcc Aluminum has been chosen. It has been intensively studied in literature [2228], where the most puzzling
experimental problem at that time was the distinction between inter-band and intra-band contributions to the absorption process.
Here, it serves as a test case mainly in terms of convergence with respect to the most important convergence parameters. While the
results did not turn out to be sensitive to the number of LAPWs, there is a crucial dependence on the BZ sampling as already found
by Lee and Chang [26]. Fig. 1 shows the inter-band contribution to the imaginary part of the dielectric function (Eq. (20)) for a
series of k point meshes. The peaks not only get sharper with increasing number of k points, but they also exhibit a pronounced
energy shift. Only with the most dense mesh is quantitative agreement with experimental data achieved [29]. The inset of Fig. 1
exhibits the plasma frequency (Eq. (21)) as a function of the number of k points in the irreducible part of the Brillouin zone (IBZ).
Also in this case highest accuracy is needed to reach the converged value of pl = 12.6 eV. This sensitivity can be understood in
terms of the high symmetry of the crystal structure and the simplicity of the material. Since the main contributions stem from certain
regions of the BZ [24], a refinement of the mesh there can dramatically change this contribution anddue to the high weight in the
BZthe total spectrum. In contrast, in more complex materials the optical absorption usually represents a sum of many different
contributions from several regions in the BZ and many band combinations. Therefore a change of one or some of these contributions
cannot affect the total spectrum that much.
Fig. 2 illustrates the separation of inter-band and intra-band contributions to the optical spectra. For this purpose the total loss
function is plotted (full line), together with the corresponding functions when considering only inter-band (dotted line) or intra-band
10 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114

Fig. 1. Imaginary part of the frequency dependent dielectric function of Al for a series of k point meshes. In the inset the dependence of the plasma frequency
(in eV) on the k point sampling is depicted.

Fig. 2. Loss function for Al (full line) calculated with 4735k points in the IBZ. The dashed (dotted) line shows the loss function of the pure intra-band (inter-band)
contribution.

(dashed line) contributions. Judging from the shape of the curves, all of them could be interpreted as free-electron behavior but with
very different plasma frequencies. Therefore the plasma frequency was for a long time thought to be 15.2 eV which is the peak
position of the total spectrum. For a more detailed discussion of the problem see Ref. [25].
Finally the sum rules provide another test for the quality of the calculations. To this extent, all three sum rules as given in
Eqs. (24)(26) have been tested. In Fig. 3 the results of the corresponding integrations are depicted as a function of energy, i.e.
the upper integration limit. The theoretical results for sum rules (24) and (25) excellently reproduce the region where experimental
data are available, while the third curve approaches /2 within 0.1%. A closer look at sum rule (24) shows that the kink in the
low-energy range indicates the onset of the inter-band contributions.

5. Summary

We have described a scheme for the calculation of linear optical properties of solids in the random-phase approximation. It is
worked out within the LAPW method which does not require approximations in the solution of the KS equation and thus is a good
starting point for the evaluation of the KS states in terms of band structure and excited states.
As an example we have provided results for Aluminum that is a prototype for a free-electron-like metal. It exhibits excellent
agreement with experiments in terms of plasma frequency, peak positions of the inter-band transitions, as well as sum rules. This
agreement can, however, only be achieved when the BZ integration is carried out with extreme accuracy. Thus highest precision is
in general needed to draw conclusions about the quality of the KS states.
 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114 11

Fig. 3. Sum rule check for the optical spectra of Al by carrying out the integrations given in Eqs. (24) (1), (25) (2), and (26) (3) as a function of the upper integration
limit (energy h in eV). The calculated values are given for two different numbers of k points in the BZ, namely, 165 (dashed lines) and 4735 (full lines). The
corresponding experimental data (open circles) are taken from Ref. [22]. The dotted line indicates /2 for comparison with the third sum rule (3).

Acknowledgements

We appreciate support from the Austrian Science Fund (FWF), project P16227-PHY, and by the EU research and training
network EXCITING, contract number HPRN-CT-2002-00317. The work was supported in part by the Materials Simulation Center,
a Penn State MRSEC and MRI facility.

Appendix A

A.1. KramersKronig relations and optical constants

From the imaginary part of the dielectric tensor component Im ij the corresponding real part is obtained by

2  Im ij ( ) 
Re ij = ij + d . (A.1)
 2 2
0
Given the real part, the inverse transformation has to be used.
With the knowledge of the complex dielectric tensor components all other frequency-dependent optical constants can be
obtained. The most often used ones are the real part of the optical conductivity

Re ij () = Im ij (), (A.2)
4
the loss function
 
1
Lij () = Im , (A.3)
() ij
and the reflectivity at normal incidence
(n 1)2 + k 2
Rii () = (A.4)
(n + 1)2 + k 2
with n and k being the real and imaginary part of the complex refractive index (refractive index and extinction coefficient):

|ii ()| + Re ii ()
nii () = , (A.5)
2

|ii ()| Re ii ()
kii () = . (A.6)
2
The absorption coefficient is given by
2k()
Aii () = . (A.7)
c
Note that the latter formulas only hold for the diagonal form of the tensor.
12 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114

A.2. Relations between spherical harmonics

Below, several well known relations between the spherical harmonics are given, which are needed to evaluate the momentum
matrix elements:

e+i sin Yl,m = Fl,m Yl+1,m+1 + Fl,m Yl1,m+1 ,

(1) (2)
(A.8)
ei sin Yl,m = Fl,m Yl+1,m1 + Fl,m Yl1,m1 ,
(3) (4)
(A.9)
(5) (6)
cos Yl,m = Fl,m Yl+1,m + Fl,m Yl1,m (A.10)
with

(1) (l + m + 1)(l + m + 2)
Fl,m = , (A.11)
(2l + 1)(2l + 3)

(2) (l m)(l m 1)
Fl,m = , (A.12)
(2l 1)(2l + 1)

(3) (l m + 1)(l m + 2)
Fl,m = , (A.13)
(2l + 1)(2l + 3)

(4) (l + m)(l + m 1)
Fl,m = , (A.14)
(2l 1)(2l + 1)

(5) (l m + 1)(l + m + 1)
Fl,m = , (A.15)
(2l + 1)(2l + 3)

(6) (l m)(l + m)
Fl,m = , (A.16)
(2l 1)(2l + 1)

 
+i i (1) (2)
e cos + Yl,m = lFl,m Yl+1,m+1 + (l + 1)Fl,m Yl1,m+1 , (A.17)
sin
 
i i (3) (4)
e cos Yl,m = lFl,m Yl+1,m1 + (l + 1)Fl,m Yl1,m1 , (A.18)
sin
(5) (6)
sin Yl,m = lFl,m Yl+1,m + (l + 1)Fl,m Yl1,m . (A.19)

A.3. The step function

The analytic evaluation of a plane wave integral over the interstitial region as needed in Eq. (53) is carried out by integrating
over the whole unit cell and subtracting the contributions of the atomic spheres, which are
 
V , G = G,
i(G G)r
e dr =  G|)  (A.20)
3V j1R(R|G|G G| ei(G G)S , G = G.
MT

It utilizes the Rayleigh-expansion of a plane wave in terms of spherical harmonics:

  
eiGr = 4eiGS i l jl |r S |G Ylm (G)Ylm (r
S ). (A.21)
lm

A.4. Local orbitals

The extension of the LAPW basis set by localized orbitals [30] was introduced in order to describe semi-core states, i.e. those
low-lying states which reach out of the atomic sphere, on the same footing as valence states. The corresponding basis function is
zero outside the atomic sphere; inside it reads
        
LO
k,lm (S + r) = Alm k + GLO lm ul (r, El ) + Blm k + Glm ul (r, El ) + Clm k + Glm ul r, El
LO LO LO
Yl,m (r). (A.22)
 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114 13

The coefficients Alm (k + GLO

l ), Blm (k + Gl ), and Clm (k + Gl ) are determined by choosing k,lm and its spatial derivative to
LO LO LO
LO . The additional trial energy E LO corresponds to the
vanish at the sphere boundary, and by the normalization condition for k,lm l
energy of the semi-core state, and for each localized basis function one specific GLO
lm is chosen [30]. The KS wave function then
can be written as
   LO  LO
nk (r) = Cnk (G)k+G (r) + LO
Cnk,lm Glm k,lm (r). (A.23)
G lm

Analogous to Eqs. (39)(41), matrix elements between LAPWs and LOs

 LO   
x+iy
klm,k+G k,lm (S + r)x + iy k+G (S + r) , (A.24)
 LO   
xiy
klm,k+G k,lm (S + r)x iy k+G (S + r) , (A.25)
 LO   
z
klm,k+G k,lm (S + r)z k+G (S + r) (A.26)
and between LOs and LOs have to be defined:
 LO   LO 
x+iy
kl  m ,klm k,l  
 m (S + r) x + iy k,lm (S + r) , (A.27)
 LO   LO 
xiy
kl  m ,klm k,l  
 m (S + r) x iy k,lm (S + r) , (A.28)
 LO   LO 
z
kl  m ,klm k,l  
 m (S + r) z k,lm (S + r) . (A.29)
The atomic sphere parts of the momentum matrix elements given in Eq. (52) have to be supplemented by contributions from the
local orbitals:

n k|x |nkMT

1  x+iy 
Cn k (G ) k+G ,k+G + k+G ,k+G Cnk (G)
xiy
=
2 
G ,G
1  LO  LO  x+iy xiy 
+ Cn k,lm Glm klm,k+G + klm,k+G Cnk (G)
2
lm,G
1  LO  LO  x+iy xiy  LO  LO 
+ Cn k,l  m Gl  m kl  m ,klm + kl  m ,klm Cnk,lm Glm , (A.30)
2  
l m ,lm
n k|y |nkMT
1   x+iy 
Cn k (G ) k+G ,k+G k+G ,k+G Cnk (G)
xiy
=
2i 
G ,G
1  LO  LO  x+iy xiy 
+ Cn k,lm Glm klm,k+G klm,k+G Cnk (G)
2i
lm,G
1  LO  LO  x+iy xiy  LO  LO 
+ Cn k,l  m Gl  m kl  m ,klm kl  m ,klm Cnk,lm Glm , (A.31)
2i  
l m ,lm
n k|z |nkMT
   LO  z
= Cn k (G ) zk+G ,k+G Cnk (G) + CnLO
 k,lm Glm klm,k+G Cnk (G)
G ,G lm,G
   z  LO 
+ CnLO
 k,l  m GLO LO
l  m kl  m ,klm Cnk,lm Glm . (A.32)
l  m ,lm

The evaluation of Eqs. (A.24)(A.29) is analogous to that for Eqs. (49)(51).

References

[1] P. Hohenberg, W. Kohn, Phys. Rev. B 136 (1964) 864.

[2] W. Kohn, L.J. Sham, Phys. Rev. A 140 (1965) 1133.
[3] M.S. Hybertsen, S.G. Louie, Phys. Rev. B 34 (1986) 5390.
[4] Z.H. Levine, D.C. Allan, Phys. Rev. Lett. 63 (1989) 1719; Phys. Rev. B 43 (1991) 4187.
[5] R. Abt, C. Ambrosch-Draxl, P. Knoll, Physica B 194196 (1994) 1451.
[6] M. Alounai, J.M. Wills, Phys. Rev. B 54 (1996) 2480.
14 C. Ambrosch-Draxl, J.O. Sofo / Computer Physics Communications 175 (2006) 114

[7] R.J. Mathar, J.R. Sabin, S.B. Trickey, Nucl. Instrum. Methods in Phys. B 155 (1999) 249, and references therein.
[8] R.J. Mathar, S.B. Trickey, J.R. Sabin, Adv. Quantum Chem. 45 (2004) 277, and references therein.
[9] W.-D. Schne, R. Keyling, M. Bandic, W. Ekardt, Phys. Rev. B 60 (1999) 8616.
[10] K. Sturm, E. Zaremba, K. Nuroh, Phys. Rev. B 42 (1990) 6973.
[11] P. Blaha, K. Schwarz, G.K.H. Madsen, D. Kvasnicka, J. Luitz, WIEN2k, An Augmented Plane Wave + Local Orbitals Program for Calculating Crystal
Properties, Techn. Universitt Wien, Austria, ISBN 3-9501031-1-2, 2001.
[12] N.W. Ashcroft, N.D. Mermin, Solid State Physics, Saunders College Publishing, Fort Worth, TX, USA, 1976, p. 344.
[13] S.L. Adler, Phys. Rev. 126 (1962) 413.
[14] N. Wiser, Phys. Rev. 129 (1963) 62.
[15] L. Hedin, Phys. Rev. A 139 (1965) 796.
[16] P. Romaniello, P.L. de Boeij, Phys. Rev. B 71 (2005) 155108.
[17] K.-H. Lee, K.J. Chang, M.L. Cohen, Phys. Rev. B 52 (1995) 1425.
[18] K.H. Weilacher, H. Bross, J. Phys. F: Metal Phys. 7 (1977) 2253.
[19] P. Blchl, O. Jepsen, O.K. Andersen, Phys. Rev. B 49 (1994) 16223.
[20] O.K. Andersen, Phys. Rev. B 12 (1975) 3060.
[21] D. Singh, Planewaves, Pseudopotentials and the LAPW Method, Kluwer Academic Publishers, Boston, Dordrecht, London, 1994.
[22] H. Ehrenreich, H.R. Philipp, B. Segall, Phys. Rev. 132 (1963) 1918.
[23] E. Shiles, T. Sasaki, M. Inokuti, D.Y. Smith, Phys. Rev. B 22 (1980) 1612.
[24] F. Szmulowicz, B. Segall, Phys. Rev. B 24 (1981) 892.
[25] D.Y. Smith, B. Segall, Phys. Rev. B 34 (1986) 5191.
[26] K.-H. Lee, K.J. Chang, Phys. Rev. B 49 (1994) 2362.
[27] Th. Herrmann, M. Gensch, M.J.G. Lee, A.I. Shkrebtii, N. Esser, W. Richter, Ph. Hofmann, Phys. Rev. B 69 (2004) 165406.
[28] S.R. Barman A, K. Horn, Appl. Phys. A 69 (1999) 519.
[29] This problem was avoided in Ref. [24] by using a k p interpolation for the APW results.
[30] D. Singh, Phys. Rev. B 43 (1991) 6388.