Application of The Exact Muffin-Tin Orbitals Theory
Application of The Exact Muffin-Tin Orbitals Theory
Abstract
1
I. INTRODUCTION
During the last decades many attempts have been made to develop accurate and at
the same time efficient methods for solving the Kohn-Sham equations in an application
of the Density Functional Theory for condensed matter. The accuracy of the methods
is crucial e.g. when one searches for the answers given by different density functional
approximations. The full-potential techniques have been specially designed to fulfill this
requirement. Though, in principle, they give highly accurate results, they have their
own limitations. In many cases a compromise has been made between the accuracy
and efficiency, and methods based on approximate one electron potentials have been
developed. The most commonly used muffin tin approach, albeit its mathematical for-
mulation is very elegant, presents a rather poor representation of the exact potential.
Though, the Atomic Sphere Approximation (ASA) brings a real improvement to the
potential, most of the conventional methods based on the ASA use similar approxima-
tion to the one electron energies and charge density as well [1]. Therefore, using these
methods, reasonably accurate results can only be obtained for close packed systems,
and they are not suitable to treat systems of low symmetry. In order to maintain or
increase the accuracy different corrections should be included and, therefore, the ASA
based methods lose their elegance and efficiency.
A few years ago breakthrough was made by developing the Exact Muffin-Tin Or-
bitals (EMTO) Theory [2,3]. Within the EMTO Theory the one electron states are
calculated exactly for the overlapping muffin-tin potential, while the solution of Pois-
son’s equation can include certain shape approximations, if required. By separating
the two approaches used for the potential and one electron states the accuracy can be
sustained at a level comparable with that of the full-potential techniques without losing
significantly from the efficiency. The EMTO Theory can be considered as an improved
screened KKR method [4,5], within that large overlapping potential spheres can be used
for accurate representation of the exact one electron potential [3,6].
In this work we present a self-consistent implementation of the EMTO Theory within
the Spherical Cell Approximation (SCA) for the Poisson’s equation. In the first part we
review the EMTO Theory [2,3], the definition of the screened spherical waves and the
matching equation. Furthermore, we establish the expressions for the number of states
and electron density using the Green’s function formalism. In the second part of the
paper we discuss the impact of the SCA, used for the shape of the Wigner-Seitz cell,
on the total charge density and on the overlapping muffin tin potential. Finally, we
establish the accuracy of the SCA-EMTO method by performing test calculations for
some systems where reliable full-potential data are available. An approximate solution
of the kink cancellation equation in order to reduce the number of iterations needed in
a self-consistent calculation is presented in the Appendix.
In the following we review the basic concepts of the EMTO Theory developed by
O. K. Andersen and co-workers [2,3,7]. Assume that the one-electron Kohn-Sham
equations,
2
h i
−∇2 + v(r) Ψj (r) = ǫj Ψj (r), (1)
are solved within the muffin tin approximation for the effective potential,
X
v(r) ≈ v0 + [vR (rR ) − v0 ] , (2)
R
where R runs over the lattice sites. Here and in the following we use the notation
rR ≡ r − R and omit the vector notation for the index R. In (2) v0 denotes a parameter
[7] that reduces to the muffin tin zero in the case of non-overlapping muffin tins. The
spherical potentials vR (rR ) become equal to v0 outside the potential spheres of radii
sR . These radii can be chosen as the linear overlap between the spheres to be as large
as 30 − 40 % [4,3,6]. It has turned out that for a good representation of the real
full-potential in terms of overlapping muffin tin wells usually a big overlap is preferred
between the potential spheres [6].
In order to solve the Schrödinger equation (1) for the muffin tin potential (2) one
chooses different basis functions inside the potential spheres and in the interstitial
region. Inside the sphere at R the partial waves are chosen as the basis function which
are defined as the products of the regular solutions of the radial Schrödinger equation,
h i
∂ 2 rR φRl (ǫ, rR ) h l(l + 1) i
2 = 2
+ vR (rR ) − ǫ rR φRl (ǫ, rR ), (3)
∂rR rR
Here κ2 = ǫ − v0 , and in the non-overlapping muffin tins limit it denotes the interstitial
one-electron kinetic energy. The ΘR (rR ) is one inside the sphere of radius sR centered
at R and zero outside. The expansion coefficients uaRL,j and vRL,j a
as well as the en-
ergies ǫj are determined from the condition that the wave function Ψj (r) and its first
derivative should be continuous at the potential spheres. The algebraic formulation of
this matching condition in the EMTO formalism is the so the called kink cancellation
equation, which is equivalent to the KKR (Korringa-Kohn-Rostoker) equation in an
arbitrarily screened representation [5].
The screened spherical waves can be defined [2] in conjunction with hard spheres
centered at all sites R with radii aRl . They are solutions of the wave equation,
3
h i
a
∇2 + κ2 ψRL (κ2 , rR ) = 0, (6)
with the boundary condition that on their own a−spheres they behave like a pure real
spherical harmonic, while the YL′ (r̂R′ ) projections on all the other a−spheres, R′ 6= R,
vanish. They form a complete basis set in the ”a” interstitial region and may be
a a
expressed in terms of the ”value”, fRL , and the ”slope”, gRL , functions [2], whose radial
part satisfy the following boundary conditions
a
a ∂fRl (κ, r)
fRl (κ, r)|aRl = 1 and |aRl = 0, (7)
∂r
a
a ∂gRl (κ, r) 1
gRl (κ, r)|aRl = 0 and |aRl = . (8)
∂r aRl
These functions may of course be expressed in terms of the usual spherical Bessel and
Neumann functions [8]
a
fRl (κ, r) = jl (κr)Wa {f, κ n} − κ nl (κr)Wa {f, j} (9)
and
a
gRl (κ, r) = jl (κr)Wa {g, κ n} − κ nl (κr)Wa {g, j} (10)
a a
h ∂gla (κ, r) ∂fla (κ, r) a i
Wr {fRl , gRl } ≡ r 2 fla (κ, r) − gl (κ, r) = aRl . (11)
∂r ∂r
a
The screened spherical wave ψRL (κ2 , rR ) may be expanded in spherical harmonics
′
YL′ (r̂R′ ) about any site R , as
a a
(κ2 , rR ) = fRl gRa ′ l′ (κ, rR′ ) YL′ (r̂R′ ) SRa ′ L′ RL (κ2 ), (12)
X
ψRL (κ, rR ) YL (r̂R )δRR′ δLL′ +
L′
where the expansion coefficients, SRa ′ L′ RL (κ2 ), are the elements of the so called slope
matrix. The slope matrix can be derived from the bare KKR structure constant matrix
BR′ L′ ,RL (κ), by matrix inversion [2]
1 h i−1 1
S a (κ2 ) = D{j(κ, a)} − −B(κ) + κ cot α(κ) , (13)
a j(κ, a) j(κ, a)
where D denotes the logarithmic derivative, D{j(r)} ≡ r [∂j(r)/∂r] /j(r), and for sim-
plicity where we have used matrix notation. We note that this equation is equivalent
to Eq. (3.26) from Ref. [2] and Eq. (15) from Ref. [3]. The bare KKR structure con-
stants are defined as the expansion coefficients of the κ nL (κ, rR ) ≡ κ nl (κrR ) YL (r̂R )
functions around site R′ in terms of the jL′ (κ, rR′ ) ≡ jl′ (κrR′ ) YL′ (r̂R′ ) functions, i.e.
X
κ nL (κ, rR ) = jL′ (κ, rR′ ) BR′ L′ RL (κ),
L′
with
4
X
L ′′ ′
−l+l −l ′′
BR′ L′ RL (κ) ≡ 4π CLL ′ i κ nL′′ (κ, R′ − R), (14)
L′′
L ′′
and where CLL ′ are the real Gaunt numbers.
For the partial waves explicitly included in the formalism, the so called low partial
waves with l ≤ llow = 2 − 3, αRl (κ) are the hard sphere phase shifts given by
and for the remaining Rl-chanels, αRl (κ) are the proper phase shifts. For high l’s the
latter vanish, and at that point the matrix to be inverted in (13) can be truncated.
When the hard sphere radii, aR , are properly chosen and κ2 lies below the bottom of
the hard sphere continuum, the screened spherical waves have short range. Therefore,
the slope matrix can be calculated in real space and the method is suitable to treat
impurities, defects, surfaces, etc. It was shown in Ref. [2] that the shortest range of
the screened spherical waves can be achieved for non-overlapping spheres with aR ≈
0.5 − 0.85 siR , depending on the maximal orbital quantum number l of the partial waves
explicitly included in the formalism. The siR denotes the inscribed or touching sphere
radii. In the KKR community, it is customary to determine the αRl (κ)’s as the phase
shifts of repulsive potentials.
Because a screened spherical wave has pure (l, m) character only on its own a-sphere,
the matching condition in Eq. (5) should be set up at this sphere. The connection onto
the potential sphere (s) is done by introducing a free electron solution ϕRl (ǫ, rR ) YL (r̂R )
from the potential sphere back to the hard sphere, which joins continuously and differ-
entiable to the partial wave, φRL (ǫ, rR ), at sR and continuously to the screened spherical
wave at aRl . The radial part of this backwards extrapolated free-electron solution, after
normalizing it to one at its a-sphere, is given by
ϕRl (ǫ, r)
ϕaRl (ǫ, r) ≡ a
= fRl a
(κ, r) + gRl a
(κ, r) DRl (ǫ), (15)
ϕRl (ǫ, aRl )
a
where DRl (ǫ) is the logarithmic derivative of ϕRl (ǫ, r) calculated at the hard sphere aRl .
This can be determined from the matching condition between φRL (ǫ, r) and ϕRL (ǫ, r)
at rR = sR ,
a a
a fRl (κ, sR ) D{φRl (ǫ, sR )} − D{fRl (κ, sR )}
DRl (ǫ) ≡ D{ϕaRl (ǫ, aRl )} = − a a
. (16)
gRl (κ, sR ) D{φRl (ǫ, sR )} − D{gRl (κ, sR )}
The relation between the values of the free electron function at a and the partial
wave at s is
a
ϕRl (ǫ, aRl ) ϕRl (ǫ, aRl ) 1 D{φRl (ǫ, sR )} − D{gRl (κ, sR )}
= = a a a
, (17)
φRl (ǫ, sR ) ϕRl (ǫ, sR ) fRl (κ, sR ) D{fRl (κ, sR )} − D{gRl (κ, sR )}
5
smooth functions of energy, which varies more slowly than D{φRl (ǫ, s)}, because a < s.
a
The poles of DRl (ǫ) depend on the representation (a) and they are not related directly
to the band structure. The ϕRl (ǫ, aR ) from the figure was obtained for partial waves
normalized in the w-sphere. It is always a smooth function of the energy and vanishes
a
at the poles of DRl (ǫ).
The slope matrix, Eq. (13), the logarithmic derivative, Eq. (16), and the normaliza-
tion function, Eq. (17), play a central role in the present implementation of the EMTO
Theory.
Using the free electron solutions from (12) and (15) and the partial waves φRl (ǫ, rR )
we can introduce a complete basis set defined in the whole space. These exact muffin
tin orbitals or kinked partial waves may be written in the form
a
ψ̄RL (ǫ, rR ) = (φaRl (ǫ, rR ) − ϕaRl (ǫ, rR )) YL (r̂R ) + ψRL
a
(κ2 , rR ), (18)
where the radial part of the functions φaRl and ϕaRl are truncated outside the sphere of
radius sR and outside sR and inside aR , respectively. Moreover, the l ≤ llow projection
a
of the ψRl function is truncated inside the sphere of radius aR , while the high-l com-
a
ponents penetrate into the hard spheres. The ψ̄RL (ǫ, rR ) functions are continuous and
differentiable in the whole space, except at the hard spheres, where they have non zero
kinks. In Eq. (18) the partial waves are renormalized according to Eq. (15)
φRL (ǫ, rR )
φaRL (ǫ, rR ) ≡ . (19)
ϕRL (ǫ, aR )
From Eq. (17) and (19) it is immediately seen that the multiplicativ normalization of
the partial waves does not enter in the expression of the kinked partial wave. Forming
a linear combination of the kinked partial waves,
a a
X
Ψj (r) = ψ̄RL (ǫj , rR ) vRL,j , (20)
RL
and asking for the kinks be canceled we arrive to the kink cancellation or screened KKR
equations
Here we have l, l′ ≤ llow . The solutions of this equation are the one-electron energies
and eigenfunctions, which, using Eq. (5) are given by
a
vRL,j
uRL,j = . (22)
ϕRl (ǫj , aR )
6
It is worth to note that in the final expression of the wavefunction Ψj (r), Eq. (5), the
backwards extrapolated free electron solution does not enters.
In the case of translation symmetry in Eq. (21) R and R′ run over the atoms in
the primitive cell only, and the slope matrix, and thus the kink matrix KRa ′ L′ RL as
well, depend on the Bloch vector k from the first Brillouin zone. In Fig. 2 we plotted
the diagonal elements of the fcc slope matrix (symbols) calculated at the center of
the Brillouin zone as a function of the dimensionless energy parameter (κw)2 . In this
calculation the matrix inversion in (13) was performed in real space for 5 coordination
shells plus the central site using the s, p and d orbitals and 0.7w for the hard sphere
radius. The figure demonstrates the weak and smooth energy dependence of the slope
matrix up to the bottom of the continuum, (κw)2 ≈ 6. Therefore in the practical
solution of the kink cancellation equation (21) the slope matrix can be estimated using
a Taylor expansion around a fixed energy κ20 ,
1 a
SRa ′ L′ RL (κ2 ) = SRa ′ L′ RL (κ20 ) + Ṡ ′ ′ (κ2 )(κ2 − κ20 ) + ..., (23)
1! R L RL 0
where the overdot indicates energy derivative. The first and higher order energy deriva-
tives are calculated analytically as described in Ref. [9]. In equation (23) κ2 is a complex
energy not too far from κ20 . In Fig. 2 the solid lines were calculated with a fourth order
expansion around κ0 = 0. As one can observe, this expansion gives highly accurate en-
ergy dependence of the slope matrix over an energy range of approximately (−1, +1)Ry.
where the sum runs over the one-electron states below the Fermi level ǫF . In the present
implementation of the method instead of calculating explicitly the wave functions (5)
and performing the summation in Eq. (24) we introduce the path operator gRa ′ L′ RL (z, k)
defined for a complex energy z and Bloch vector k by
This function is analytic in the complex plane and it has poles at the one-electron
energies along the real axis. Therefore, using the residue theorem, for the total number
of electrons we find
1
I X Z
N(ǫF ) = gRa ′ L′ RL (z, k) K̇RLR
a
′ L′ (z, k) dk dz, (26)
2πi ǫF R′ L′ RL BZ
where the first integration is performed on a complex contour and the second one in
the first Brillouin zone. The contour is chosen in a way that it cuts the real axis below
7
the bottom of the valence band and at ǫF . In (26) l, l′ ≤ llow . The z dependent partial
a
waves, φRl (z, rR ), and logarithmic derivatives, DRl (z), are obtained by solving Eq. (3)
for complex energy. The energy derivative of the kink matrix,
h i
K̇Ra ′ L′ RL (z, k) = aR′ ṠRa ′ L′ RL (z − v0 , k) − δR′ R δL′ L ḊRL
a
(z) , (27)
is calculated by taking the derivatives of Eq. (16) and (23), where the energy derivatives
of the basis functions {f a , g a } are calculated analytically. The energy derivative of the
logarithmic derivative function is given by [2]
R sR
∂D{φRl (z, sR )} 0 φ2Rl (z, rR ) rR
2
drR
= − 2
. (28)
∂z sR φRl (z, sR )
the expression (26) gives the exact number of states at the Fermi level for the muffin
tin potential (2). In (29) the negligible terms due to the overlap between s-spheres are
omitted [3].
Inside the unit cell at R the electron density in terms of the path operator can be
expressed as
1
I Z
a a a
X
n(rR ) = ZRL ′ (z, rR ) g̃RL ′ RL (z, k)dkZRL (z, rR )dz, (30)
2πi ǫ F L′ L BZ
and where the sums over l′ and l include the high-l terms as well. These functions are
equivalent to the scattering solutions of Faulkner and Stocks [10]. In Eq. (30) we have
introduced the following matrix
a
gRL (z, k) if l, l′ ≤ llow
P RL a
′
a
if l′ ≤ llow and l > llow
PR L gRL′ R′′ L′′ (z, k)SR′′ L′′ RL (z, k)
′′ ′′
a a a
g̃RL ′ RL (z, k) ≡ ′′ ′′ S ′ ′′ ′′ (z, k)gR′′ L′′ RL (z, k) if l′ > llow and l ≤ llow (32)
PR L PRL R L a
R′′′ L′′′ SRL′ R′′ L′′ (z, k)
R′′ L′′
a a
×gR′′ L′′ R′′′ L′′′ (z, k)SR′′′ L′′′ RL (z, k) if l′ , l > llow
where the high-low and the low-high subblocks of the slope matrix are calculated by
the usual blowing-up technique [11].
In principle Eq. (26) and (30) give the exact number of states and electron density.
However, in some cases, like for the metals from the IIB and III − V A groups, where
one of the d bands is completely filled, around the top of this band the normalization
function (17) goes through zero. This happens, for example, in the case of fcc Ga
8
around the top of the 3d band, as it can be seen from Fig. 1. For this energy not only
a
the logarithmic derivative but also its energy derivative ḊRl , appearing in the diagonal
a
of the K̇R′ L′ RL matrix, has poles. In order to cancel these nonphysical poles we rewrite
the expression for the number of states as
1
I
N(ǫF ) = G(z)dz, (33)
2πi ǫF
where
a
X ḊRl
X Z (z) 1
gRa ′ L′ RL (z, k) K̇RLR
a
X
G(z) ≡ ′ L′ (z, k) dk − a − , (34)
R′ L′ RL BZ RL DRl (z) ǫD
z − ǫD
Rl
Rl
where ǫD a
Rl are the zeros of the logarithmic derivative function, DRl (ǫ). Because the
logarithmic derivative is a smooth decreasing function of energy ǫD Rl ’s can be easily
determined with high accuracy. The second and third terms from the right hand side of
a
(35) are included only for l ≤ llow . Using the fact that the residuum of the 1/DRl around
ǫRl is 1/ḊRl , it is easy to show that the poles of ḊRl (z) and those of 1/ϕRl (z, aR )2 are
D a a
Equations (21) and (33-35), derived in the previous section, constitute the basis of
the present method. In order to perform a self-consistent calculation one constructs the
electron density from the solutions of the kink cancellation equation and calculates the
new one-electron potential. In this section we describe these steps using the SCA for
the shape of the Wigner-Seitz cell.
In the SCA, for solving the Poisson’s equation, we substitute the Wigner-Seitz cells
by spherical cells with volumes equal to the volumes of the real cells. If ΩR denotes
the volume of the Wigner-Seitz cell (Voronoi polyhedron) centered at R we have ΩR =
ΩwR ≡ 4π 3
3
wR , where wR is the atomic sphere radius. Thus within the SCA, like in the
conventional ASA, the whole space is ”covered” by the ΩwR spheres.
9
A. The SCA charge density
where the site independent a constant is determined from the condition of the charge
neutrality within the whole unit cell
XZ
nSCA (rR )drR =
X
ZR . (37)
R ΩwR R
Here ZR denotes the nuclear charge at R. The sum runs over the atoms from the unit
cell, and the integrals are performed inside the SCA spheres. Throughout this section
the charge density is normalized within the SCA spheres according to (36) and (37),
however, for the sake of simplicity we neglect the SCA index for the nSCA (r).
The spherical symmetric potentials, vR (rR ), that enter in Eq. (3) have to be chosen
in a way that, together with the parameter v0 , to give the best approximation to the
full potential v(r). The original idea in Ref. [7] is to minimize the mean of the squared
deviation between the left and the right hand side of Eq. (2). This leads to a set of
integral or differential equations for vR (rR ) and v0 . In the non-overlapping muffin tins
case the equation for vR (rR ) reduce to the well known expression
1
Z
vR (rR ) = v(r)dr̂R, (38)
4π
and v0 reduces to the muffin tin zero, i.e. to the average of the full potential calculated
in the interstitial region,
4π 3
ΩI ≡ Ω −
X X
VR ≡ Ω − s ,
R R 3 R
where Ω is the volume of the region where the approximation (2) is valid (unit cell),
and VR denotes the volume of the potential sphere.
In the overlapping muffin tins case the equation for the v0 can be written in the
following simple form [7]
4π sR 1
Z Z
2
X
[vR (rR ) − v0 ] rR dr + v0 = v(r)dr, (39)
R Ω 0 Ω Ω
10
while the equation for vR (rR ) involves terms coming from the overlapping region, and
which give rise to kinks of vR (rR ) when rR touches other muffin tin spheres. In the
present implementation of the method, instead of solving the vR (rR ) equations, we all
the time, for non-overlapping and for overlapping muffin tin wells as well, fix the vR (rR )
functions to the spherical average of the full potential given by (38). In this case from
Eq. (39) we get the expression for the v0 as
"Z #
1 XZ
v0 = v(r)dr − v(r)dr , (40)
Ω− VR
P
R Ω R VR
or
P hR i
v(r)dr − v(r)dr
R
R ΩIR Ωov
v0 = R
, (41)
[ΩIR − Ωov
R]
P
R
where ΩIR is the real interstitial within a Wigner-Seitz cell centered at R with volume
ΩR , and Ωov
R is that part of the potential sphere that is outside of the cell ΩR , i.e.
Eq. (41) assumes the knowledge of the full potential v(r) in ΩIR and Ωov
R regions. How-
ever, the time consuming calculation of the full potential can be avoided by using the
SCA for the unit cell. In the non-overlapping SCA case, i.e. sR < wR , we have
Z wR
ΩIR − Ωov
R = 4π 2
rR drR ,
sR
and
Z Z Z wR Z
2
v(r)dr − v(r)dr = v(r)dr̂R rR drR , (43)
ΩIR Ωov
R sR
and
Z Z Z sR Z
2
v(r)dr − v(r)dr = − v(r)dr̂R rR drR . (44)
ΩIR Ωov
R wR
From these equations we get the expression for the parameter v0 valid within the SCA
X Z wR Z
2
X
v0 = rR v(r)dr̂R drR / WR (45)
R sR R
3
where WR ≡ 4π(wR −s3R )/3. Therefore both of the vR (rR ) function and the v0 parameter
are given in terms of the spherical symmetric part of the full potential.
The many-body part, µxc [n(r)], of the one-electron effective potential,
11
v(r) = v C (r) + µxc [n(r)], (46)
is calculated within the local density or generalized gradient approximation, while the
electrostatic part is derived solving the Poisson’s equation,
h i
∇2 v C (r) = −8π n(r) −
X
ZR δ(rR ) , (47)
R
for the electronic and nuclear charge densities. The electrostatic potential can be di-
vided into intercell and intercell component. The spherical symmetric part of the in-
tercell or Madelung potential is given by
1 X
vRM (rR ) = MRLR′ L′ QR′ L′ with L = (0, 0), (48)
w R′ L′
where MRLR′ L′ is the Madelung matrix, which can be evaluated by the usual Ewald
technique, and
√
4π r l h
Z
R
i
QRL = nR (rR ) − ZR δ(rR ) YL (r̂R ) drR . (49)
2l + 1 ΩwR w
The Hartree part of the intracell Coulomb potential can be obtained as the solution
of the Poisson’s equation using the proper boundary condition at the atomic sphere
radius. Alternatively, this term is given by
h i
′ 2
8π r1R 0rR rR nR (rR′
)drR′
+ rwRR rR
′ ′
nR (rR ′
)drR for rR ≤ wR
R R
vRI (rR ) = ′ 2
, (50)
8π r1R 0wR rR ′
nR (rR ′
)drR for rR > wR
R
that is valid inside the potential sphere sR , for sR ≥ wR as well as for sR < wR . The
total potential within the potential sphere is obtained as the sum of Eq. (48), (50),
the Coulomb potential of the nucleus and the spherical symmetric exchange-correlation
potential, namely
2ZR
vR (rR ) = vRM + vRI (rR ) − + µxcR (rR ). (51)
rR
If the spherical symmetric part of the exchange-correlation potential is approximated
by µxcR[nR (rR )] besides the higher order multipole moments from (48), which in many
cases can be neglected, all of the potential components from (51) depend only on the
spherical symmetric density nR (rR ).
Within the SCA-EMTO method the atomic and potential spheres can be and usually
they are chosen differently. The sizes of the atomic spheres, wR , are fixed by the volume,
and the ratio between them should be chosen in a way that minimizes the errors coming
from approximate solution of the Poisson’s equation. We have found that the best
representation of the potential can be achieved by choosing the potential sphere radii,
sR , larger or equal with the atomic sphere radii. For an optimal choice of the potential
spheres the potentials at sR should be the same, i.e. vR (sR ) ≈ constant for each R,
and this constant should have the maximum possible value for linear overlaps bellow
30 − 40%.
12
IV. APPLICATIONS: TEST CALCULATIONS
A. Numerical details
The hard sphere radii are chosen at aR = 0.7w. In the matrix inversion from Eq.
(13) we includ 79 sites in the case of f cc-based structures (f cc, L12 and L10 ), and 89
sites in the case of bcc-based structures (bcc, B2 and B32). The Taylor expansion of
the slope matrix is carried out for κ0 = 0 and includes terms up to the fourth to sixth
order energy derivative.
The path operator is calculated for 16 − 32 complex energy points, depending on
the band structure, distributed exponentially on a semi-circular contour. The k-point
sampling is performed on a uniform grid in the 3D Brillouin zones. All the calculations
are scalar-relativistic and employ the frozen-core approximation. The basis set include
s, p and d orbitals and the valence electrons are treated self-consistently within the
local density approximation to density functional theory using the Ceperley and Alder
[12] exchange-correlation functional and parametrized by Perdew and Wang [13]. The
atomic sphere radii, wR -s, are chosen in a way that the atomic spheres should have the
same volumes as the corresponding Voronoi polyhedra. The electrostatic and exchange-
correlation contribution to the total energy is calculated within the SCA as described,
for example, in Refs. [14,15]. The kinetic energy is given by
1 wR
I XZ
T SCA = zG(z)dz − 2
vR (rR )nR (rR )rR drR , (52)
2πi ǫF R 0
where the first term from the right hand side is the sum of the one electron energies
and G(z) is given in (34).
B. Results
Before starting on the evaluation of the results we address the question of the accu-
racy of the Taylor expansion for the slope matrix, Eq. (23). In Fig. 3 the total energy
of fcc Cu is shown for different number of terms included in the Taylor expansion.
The inclusion of the fourth order energy derivative term changes the total energy by
1.13 mRy. The effect of the fifth order term is already less than 0.2 mRy and that of
the sixth order term is about 0.04 mRy. Therefore, we conclude that for a reasonable
accuracy it is sufficient to include five terms in the expansion of the slope matrix, viz.
up to the fourth order energy derivatives. In the case of open structures, wide energy
bands or semicore states, however, more terms should be included [16].
13
The variations of the SCA-EMTO total energy with the potential sphere radius for
fcc and bcc Cu are shown on Fig. 4. For this test calculation the total potential from
(45) was weighted by the valence part of the total density according to
R wR 2
rR [ n(r)v(r)dr̂R] drR
P R
R sR
v0 = R wR 2 R , (53)
sR rR [ n(r)dr̂R ] drR
P
R
′
and the cut-off in (35) for the spherical part of the total density was lmax = lmax = 8.
i i
The inscribed sphere radii are sf cc = 0.91wf cc and sbcc = 0.88wbcc, where the theoretical
atomic sphere radii, wf cc and wbcc are shown on the figure. The total energy in both
of the fcc and bcc structures beginning from s ≈ 0.80w becomes almost flat with a
negligible slope up to s ≈ 1.20w, which means about 32 − 36% linear overlap for the
fcc and bcc structures, respectively. For s > 0.80w the further increase of the potential
sphere radius has little effect on the energy, that means the potential in the corners of
the Wigner-Seitz cell is, with a very good approximation, constant. However, for big
overlaps, s > 1.20w, the errors coming from the overlap region, and neglected in the
kink-cancellation equation and in the charge density as well, become important [3].
There is a comprehensive study of the structural stability of the transition metals
done either by full-potential or by muffin-tin or ASA based methods. In the latter
case correction terms are needed [17] for calculation of the accurate total energies. The
conventional ASA without correction terms gives, for example, with about 2 mRy lower
total energy for the Cu in the bcc phase than in the fcc phase. This underestimation of
the bcc total energy is due to the incorrect kinetic energy term, and the inclusion of the
exact Hartree energy would lower even more the bcc energy. From a more sophisticated
full-potential method [18] a structural energy difference of Ebcc − Ef cc ≈ 0.5 mRy was
obtained. This number should be compared with our results of 0.4 mRy from Fig.
4. One should note that this difference is almost constant for a wide range of linear
overlap.
The second example is the binary Lix Al1−x ordered compound in different phases.
There are two reasons of this choice: i) this system is very well studied through ac-
curate full-potential calculations [19,20], and ii) most of the experimentally observed
interesting trends, the contraction of the volume of the Al-based alloys, the asymmetric
heats of formation with respect to the equiatomic concentration etc., can not be re-
produced by an ordinary spherically symmetric calculation. Besides the three different
compositions, x = 0.25, 0.50, 0.75, we consider the pure Al and Li limits in fcc and bcc
phases as well. For x = 0.25 and 0.75 the calculations were performed only for the L12
structure, while for x = 0.50 we considered three different structures: L10 , B2 and B32.
In Fig. 5 the charge density contour plots are shown for pure fcc Al, and Al3 Li in
L10 structure, as calculated from Eq. (35) using s, p, d basis set, and the maximum
orbital quantum number l′ included in (12) was 10. The agreement between these plots
and those from Ref. [21] is very good.
The calculated equilibrium Wigner-Seitz radii and the bulk moduli are tabulated
in Table I and plotted in Figs. 6 and 7. In the case of pure Al and Li the structure
energy differences and in the case of compounds the heats of formation are included in
the Table and plotted in Fig. 8. The heat of formation is defined as
∆H ≡ ELix Al1−x − x ELi − (1 − x) EAl , (54)
14
where all the energies are obtained for the proper equilibrium volume and they are
expressed per atom. In Figs. 5-7 and Table I the full-potential values from Ref. [20] are
also included.
The mean deviations between the present and the full-potential results for the equi-
librium radii, bulk moduli and heats of formations are 9.8%, 7.5% and 17%, respectively.
Taking into account the minor discrepancies between the numerical details used in the
calculations, for example the exchange-correlation functional, the way the core elec-
trons were treated etc., and the fact that the full-potential methods have their own
error limits as well, we can conclude that the agreement between the two sets of results
is very good. One should appreciate how well the trends obtained in the full-potential
calculation are reproduced by the present method.
V. CONCLUSIONS
ACKNOWLEDGMENTS
L.V. acknowledges the interesting and helpful discussions with Prof. O. K. Ander-
sen, the assistance from Drs. C. Arcangeli and R. W. Tank, and the hospitality of the
Max-Planck Institute from Stuttgart where the first part of this work was performed.
Thanks are also due to Dr. A. V. Ruban for his valuable observations. The Swedish
Natural Science Research Council, the Swedish Foundation for Strategic Research and
Royal Swedish Academy of Sciences are acknowledged for financial support. Center for
Atomic-scale Materials Physics is sponsored by the Danish National Research Founda-
tion. Part of this work was supported by the research project OTKA 023390 of the
Hungarian Scientific Research Fund.
15
VI. APPENDIX
During the self-consistent procedure the Eq. (25), (33), (35) and (51) are solved
iteratively. In order to construct the electron density (35), as the input for the next
iteration, we have to invert the kink matrix for each complex energy z along the contour
and for each Bloch vector k from the Brillouin zone. For a reasonably high accuracy
we need at least a few hundreds of Bloch vectors in the irreducible part of the Brillouin
zone, therefore this solution means the most time-consuming step of the self-consistent
procedure. Here we apply a similar ”two-step” scheme introduced in Ref. [25] in order
to reduce the number of time-consuming iterations. Within this scheme after each
iteration an approximate charge self-consistency is achieved by solving self-consistently
the following equation written for the k-integrated path operator
h i−1
g a (z) = 1 + g a 0 (z) D a 0 (z) − D a (z) g a 0 (z), (55)
where where the index 0 denotes quantities obtained from the previous iteration, and
which are kept fixed during the solution of Eq. (55).
In the expression for the number of state (33), we need the k-integrated trace of
the product between the path operator and the energy derivative of the kink matrix,
therefore, a similar equation to (55) has to be established for the quantity
Z
GaR′ L′ RL (z) ≡ gRa ′ L′ RL (z, k) K̇RLR
a
′ L′ (z, k) dk. (56)
BZ
Using the definition of the kink matrix after some manipulations we arrive to the equa-
tion
GaR′0L′ RL (z)
" #
GaR′ L′ RL (z) = gRa ′ L′ RL (z) + δ δ
RR LL R
′ ′ a Ḋ a0
Rl (z) − Ḋ a
Rl (z) . (57)
gRa ′0L′ RL (z)
It is worth to note that in this expression we do not have matrix multiplication. Finally
we mention that as soon as the self-consistency is achieved, i.e.
a a0 a a0
DRl (z) → DRl (z) and ḊRl (z) → ḊRl (z), (58)
16
REFERENCES
[1] M. Asato, A. Settels, T. Hoshino, T. Asada, S. Blügel, R. Zeller, and P. H. Ded-
erichs, Phys. Rev. B 60, 5202 (1999).
[2] O. K. Andersen, O. Jepsen, and G. Krier, in Lectures on Methods of Electronic
Structure Calculations, edited by V. Kumar, O. K. Andersen, and A. Mookerjee,
World Scientific Publishing Co., Singapore, pp. 63-124 (1994).
[3] O. K. Andersen, C. Arcangeli, R. W. Tank, T. Saha-Dasgupta, G. Krier, O. Jepsen,
and I. Dasgupta, in Mat. Res. Soc. Symp. Proc. 491, pp. 3-34 (1998).
[4] O. K. Andersen, A.V. Postnikov, and S.Yu. Savrasov, in Applications of Multiple
Scattering Theory in Materials Science, edited by W.H. Butler, P.H. Dederichs, A.
Gonis, and R.L. Weaver, Materials Research Society, Pittsburgh, PA, (1992).
[5] L. Szunyogh, B. Újfalussy, P. Weinberger, and J. Kollár, Phys. Rev. B 49, 2721
(1994).
[6] C. Arcangeli, O.K. Andersen and R.W. Tank (unpublished).
[7] O.K. Andersen and C. Arcangeli (unpublished).
[8] Handbook of Mathematical Functions, edited by M. Abramowitz and I. A. Stegun,
Dover Publications, Inc., New York (1970).
[9] R. Tank, O.K. Andersen, G. Krier, C. Arcangeli and O. Jepsen (unpublished).
[10] J.S.Faulkner and G.M. Stocks, Phys. Rev. B 21, 3222 (1980).
[11] O. K. Andersen, O. Jepsen, and D. Glötzel, in Highlights of Condensed-Matter
Theory, edited by F. Bassani, F. Fumi, and M. P. Tosi, North-Holland, New York,
(1985).
[12] D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. 45, 566 (1980).
[13] J. Perdew and Y. Wang, Phys. Rev. B 45, 13244 (1992).
[14] L. Vitos, J. Kollár, and H. L. Skriver, Phys. Rev. B 49, 16694 (1994).
[15] H.L. Skriver, The LMTO Method (Springer-Verlag, Berlin, 1984).
[16] L. Vitos (unpublished).
[17] H.L. Skriver, Phys. Rev. B 31, 1909 (1985).
[18] T. Kraft, P. M. Marcus, M. Methfessel, and M. Sheffler, Phys. Rev. B 48, 5886
(1993).
[19] X. Q. Guo, R. Podloucky, and A. J. Freeman, Phys. Rev. B 40, 2793 (1989).
[20] M. Sluiter, D. de Fontaine, X. Q. Guo, R. Podloucky, and A. J. Freeman, Phys.
Rev. B 42, 10460 (1990).
[21] X. Q. Guo, R. Podloucky, Jian-hua Xu, and A. J. Freeman, Phys. Rev. B 41, 12432
(1990).
[22] O.K. Andersen, O. Jepsen and M. Sob, in Electronic Band Structure and its Ap-
plications, ed. M. Yussouff Springer Lecture Notes, (1987).
[23] L. Vitos, J. Kollár, and H. L. Skriver, Phys. Rev. B 55, 13521 (1997).
[24] J. Kollár, L. Vitos and H.L. Skriver in Lecture Notes in Physics, Springer Series
(1999).
[25] I. A. Abricosov, S.I. Simak, B. Johansson, A.V. Ruban, and H.L. Skriver, Phys.
Rev. B 56, 9319 (1997).
[26] D. A. Young, Phase Diagrams of the Elements (University of California Press,
Berkeley, 1991)
[27] J. M. Sanchez and C. H. Lin, Phys. Rev. B 30, 1448 (1984).
17
[28] W. Mueller, E. Bubeck, and V. Gerold, Proceedings of the 3rd International Con-
ference on Aluminium-Lithium Alloys, ed. C. Baker, P. J. Gregson, S. J. Harris,
and C. J. Peel, TMS-AIME, London, (1986).
[29] E. A. Brandes, Smithells Metals Reference Book, 6th ed., Butterworths, Boston,
MA, (1983).
[30] I. Barin, O. Knacke, and O. Kubaschewiski, Thermochemical Properties of Inor-
ganic Substances, Springer, Berlin, (1977).
18
TABLES
TABLE I. The theoretical atomic radii, bulk moduli, heats of formation and fcc-bcc struc-
ture energy differences of the ordered AlLi compounds, and pure Al and Li. The present
SCA-EMTO results are compared to the full-potential FLAPW results, and to the experi-
mental data.
Structure Present calculation Full-potential1 Experimental
Al3 Li L12 S(Bohr) 2.904 2.935 2.9343
B(GPa) 74.27 70.31 66.04
∆H(mRy) -8.21 -8.3
AlLi L10 S(Bohr) 2.881 2.917
B(GPa) 51.04 50.41
∆H(mRy) -11.53 -10.25
AlLi B2 S(Bohr) 2.837 2.876
B(GPa) 55.86 42.09
∆H(mRy) -13.94 -10.45
AlLi B32 S(Bohr) 2.863 2.910 2.9285
B(GPa) 59.91 57.75
∆H(mRy) -21.69 -16.60 -18.56
AlLi3 L12 S(Bohr) 2.932 2.965
B(GPa) 31.45 28.37
∆H(mRy) -6.71 -5.0
Al fcc S(Bohr) 2.920 2.946 2.9912
B(GPa) 91.76 82.20 72.82
E bcc − E f cc (mRy) 2.90 4.59
Al bcc S(Bohr) 2.930 2.951
B(GPa) 84.46 84.18
Li fcc S(Bohr) 3.112 3.124
B(GPa) 15.47 13.64
E − E f cc (mRy)
bcc 0.29 0.50
Li bcc S(Bohr) 3.117 3.128 3.2372
B(GPa) 15.64 15.25 12.62
1
FLAPW calculation, Ref. [20]; 2 Experimental, Ref. [26];
3
Experimental, Ref. [27]; 4 Experimental, Ref. [28];
5
Experimental, Ref. [29]; 6 Experimental, Ref. [30].
19
FIGURES
FIG. 1. The energy dependence of the logarithmic derivative DRla (ǫ) (solid line), and
normalization function ϕRl (ǫ, aR ) (dashed line) for the fcc Ga.
FIG. 2. The diagonal elements of the fcc slope matrix in the k = (0, 0, 0) point from the
Brillouin zone versus (κ w)2 . The numbers in parenthesis denote the (l, m) quantum numbers.
The Taylor expansion included terms up to the 4th order energy derivative.
FIG. 3. Self-consistent SCA-EMTO total energy of fcc Cu for different number of terms
included in the Taylor expansion for the slope matrix, Eq. (23).
FIG. 4. Total energy versus potential sphere radius, s, for the fcc and bcc Cu. The
calculations were done at the theoretical equilibrium atomic sphere radii shown in the figure.
FIG. 5. Charge density contour plots for Al3 Li in L12 structure and fcc Al in units of
0.01 electrons/Bohr 3 obtained as the output of a self-consistent SCA-EMTO calculations.
FIG. 6. The change of the atomic radii of the ordered AlLi compounds relative to the
radius of the fcc Al. The open symbols show the results obtained by the full-potential FLAPW
calculation from Ref. [20]. The present SCA-EMTO results are shown by closed symbols. The
lines connect the results for the fcc-based structures.
FIG. 7. The theoretical bulk moduli for the ordered AlLi compounds. For the notation
see caption of Fig. 6.
FIG. 8. The theoretical heat of formations for the ordered AlLi compounds. For the pure
Al and Li the fcc-bcc structure energy difference is shown. For the notation see caption of Fig.
6.
20
3
fcc Ga
Log. derivative and Norm. function
logarithmic derivative
2
normalization function
1 d
p
0
d
s
−1
p
s
−2 d
−3
−2 −1.5 −1 −0.5 0 0.5 1
Energy (Ry)
5
2
( 2, 0 )
1
0
a
−1
−2
−3
−4
−5
−8 −6 −4 −2 0 2 4 6 8
Energy * w 2
−0.08
−0.10
−0.12
−0.14
−0.16
2 3 4 5 6 7
Number of terms in the expansion
Linear overlap, fcc/bcc (%)
−0.82
2/4 18/22 35/39
−0.85
−0.86
L12 L10
6
Relative atomic radius (%)
B2 bcc
4 B32 fcc
0 fcc Al
−2
90 L12 L10
Bulk modulus (GPa)
B2 bcc
80
B32 fcc
70
60
50
40
30
−5
−10
−15
L12 L10
−20 B2 bcc
B32 fcc
−25
0 25 50 75 100