Correlated electronic structure with uncorrelated disorder

A. Östlin,1 L. Vitos,2, 3, 4 and L. Chioncel5, 1

1
Theoretical Physics III, Center for Electronic Correlations and Magnetism,
Institute of Physics, University of Augsburg, D-86135 Augsburg, Germany
2
Department of Materials Science and Engineering, Applied Materials Physics,
KTH Royal Institute of Technology, SE-10044 Stockholm, Sweden
3
Department of Physics and Astronomy, Division of Materials Theory,
Uppsala University, Box 516, SE-75120 Uppsala, Sweden
4
Research Institute for Solid State Physics and Optics,
Wigner Research Center for Physics, P.O. Box 49, H-1525 Budapest, Hungary
arXiv:1807.09582v2 [cond-mat.str-el] 5 Dec 2018

5
Augsburg Center for Innovative Technologies, University of Augsburg, D-86135 Augsburg, Germany
We introduce a computational scheme for calculating the electronic structure of random alloys
that includes electronic correlations within the framework of the combined density functional and
dynamical mean-field theory. By making use of the particularly simple parameterization of the
electron Green’s function within the linearized muffin-tin orbitals method, we show that it is possible
to greatly simplify the embedding of the self-energy. This in turn facilitates the implementation
of the coherent potential approximation, which is used to model the substitutional disorder. The
computational technique is tested on the Cu-Pd binary alloy system, and for disordered Mn-Ni
interchange in the half-metallic NiMnSb.

I. INTRODUCTION rectly. Later Győrffy9 formulated the CPA equations for

the muffin-tin potentials within the multiple-scattering
Disordered metallic alloys find applications in a large Korringa-Kohn-Rostocker (KKR) method10,11 . Con-
number of areas of materials science. Designing alloys sequently, the configurational average could be per-
with specific thermal, electrical, and mechanical proper- formed over the scattering path operator, instead of the
ties such as conductivity, ductility, and strength, nowa- Green’s function, simplifying the implementation of the
days commonly starts at the microscopic level1–3 . First- CPA for materials calculations12 . Later the CPA was
principles calculations of the electronic structure offer also implemented within the linearized muffin-tin or-
a parameter-free framework to meet specific engineer- bitals13 (LMTO) basis set14–17 . With the advent of the
ing demands for materials prediction. Advances in the third-generation exact muffin-tin orbitals18–20 (EMTO)
atomistic simulation of physical properties should to a method, and the full-charge density21 (FCD) technique,
large extent be attributed to the development of density it was possible to go beyond the atomic-sphere approxi-
functional theory (DFT)4,5 within the local density ap- mation (ASA) with CPA calculations22 , and investigate
proximation (LDA) or beyond-LDA schemes. the energetics of anisotropic lattice distortions. For in-
In solids exhibiting disorder the calculation of any teracting model Hamiltonians, dynamical mean-field the-
physical property involves configurational averaging over ory23–25 (DMFT) was also combined with the CPA26–28 .
all realizations of the random variables characterizing the Later on the methodology was extended to study real-
disorder. In the case of substitutional disorder the sym- istic materials containing significant electronic correla-
metry of the lattice is kept, but the type of atoms in the tions within the framework of a combined DFT+DMFT
basis is randomly distributed. This causes the crystal to method29,30 . To treat weak disorder within the frame-
lose translational symmetry, making the Bloch theorem work of charge self-consistent LDA+DMFT, the CPA has
inapplicable. Perhaps the most successful approach to been implemented within the KKR method31 . In an al-
solve the problems associated with substitutional disor- ternative approach, the band structures computed from
der is the coherent potential approximation6 (CPA). The DFT were mapped to tight-binding model Hamiltonians
presence of random atomic substitution generates a fluc- were the disorder was treated within the CPA32–35 .
tuating external potential which within the CPA is sub- In this paper, we introduce a method that can treat
stituted by an effective medium. This effective medium substitutional disorder effects through the CPA, and elec-
is energy dependent and is determined self-consistently tronic correlation effects through DMFT, on an equal
through the condition that the impurity scattering should footing for real materials. The method is based on the
vanish on average. The CPA is a single-site approxima- LDA+DMFT method, zMTO+DMFT, which was re-
tion, which becomes exact in certain limiting cases7 . The cently introduced by us36 . By making use of the partic-
explanation for the good accuracy of the CPA can be ularly simple parameterized form of the electron Green’s
traced back to the fact that it becomes exact in the limit function in a linearized basis set, we show that the self-
of infinite lattice coordination number, Z → ∞8 . energy can be incorporated naturally within the LMTO
The CPA was initially applied to tight-binding Hamil- formalism as a modification of a single, self-consistently
tonians7 , where for binary alloys the CPA equations determined parameter. Due to this straightforward in-
take a complex polynomial form that can be solved di- clusion of the many-body effects, the CPA within the
 2

LMTO method retains its general form, and can be used A. Linear Muffin-Tin Orbitals
almost unaltered for the LDA+DMFT approach.
The paper is organized as follows: Section II briefly The energy-independent linearized muffin-tin orbitals
reviews the muffin-tin basis sets used in this work. A χα
RL (rR ), centered at the lattice site R, are given as:
short overview of the LMTO method is given in Sec. II A, X
with additional formulas given in Appendix A, in order to χα
RL (rR ) = φRL (rR ) + φ̇α α
R′ L′ (rR )hRL,R′ L′ , (3)
introduce the most important quantities needed for the R′ L ′
CPA+DMFT implementation. A review of the EMTO φ̇α
RL (rR )= φ̇RL (rR ) + φRL (rR )oα
RL , (4)
method, which is the second basis set used in this pa- φRL (rR ) = φRL (rR )YL (r̂R ). (5)
per, is given in Appendix B. In Sec. III we present the
most important development in this paper, a combina- φRL (rR ) is the solution of the radial Schrödinger equation
tion of the CPA and the LDA+DMFT method. First, in at an arbitrary energy ǫν , usually taken to be the center
Sec. III A we discuss the principles behind the configura- of gravity of the occupied part of the band, and oα RL =
tion average used for the CPA. Sec. III B shows how the hφRL |φ̇α
RL i is an overlap integral. L ≡ (l, m) denotes the
electronic self-energy can be incorporated into the LMTO orbital and azimuthal quantum numbers, respectively.
potential functions, while Sec. III C demonstrates how The superscript α denotes the screening representation
the potential functions are used as an effective medium used in the tight-binding LMTO theory. The expan-
within the CPA. The section ends with an outline of sion coefficients hα are determined from the condition
the full computational scheme. In Sec. IV we present that the wave function is continuous and differentiable
the results of the method applied to the binary copper- at the sphere boundary at each sphere: hα ′ ′ (k) =
p α RL,R L
palladium system, treating the d-electrons of palladium α
p α α
(cRL − ǫν )δRR′ δLL′ + dRl SRLR′ L′ (k) dR′ l′ . Note that
as correlated. The half-metallic semi-Heusler compound we are now assuming a translationally invariant lattice
NiMnSb with disorder is also investigated. Section V system, so that the Bloch wave vector k is well defined.
provides a conclusion of our paper. From now on, R will denote the atoms in the unit cell
only. The coefficient hα RL,R′ L′ is parametrized by the
α α α
center of the band, cRL = ǫν − PRL (ǫν )[ṖRL (ǫν )]−1 , and
II. ELECTRONIC STRUCTURE WITH α α −1
the band width parameter dRL = [ṖRL (ǫν )] , both ex-
MUFFIN-TIN ORBITALS pressed in terms of the potential function PRL α
and its
α
energy derivative ṖRL evaluated at ǫν . The potential
The standard approach adopted in first-principles elec- function P α and the structure constant S α are expressed
tronic structure calculations is the mapping to an ef- using the conventional potential function P 0 and the con-
fective single-particle equation. The Hohenberg-Kohn- ventional structure constant matrix S 039 :
Sham density functional formalism4,5 provides a self- α
consistent description for the effective one-particle po- PRL = [P 0 (1 − αP 0 )−1 ]RL ,
α
tential Veff (r). Within the muffin-tin orbital methods, SRLR′ L′ = [S 0 (1 − αS 0 )−1 ]RLR′ L′ . (6)
the effective potential in the Kohn-Sham equations, 0
PRL is proportional to the cotangent of the phase shift
 2  created by the potential centered at the sphere at R.
∇ − Veff (r) Ψj (r) = ǫj Ψj (r), (1) Thus, the potential parameters P characterize the scat-
tering properties of the atoms placed at the lattice sites.
is approximated by spherical potential wells VR (rR ) − V0 The geometry of the lattice enters through the structure
centered on lattice sites R, and a constant interstitial constants S 0 , which is independent of the type of atoms
potential V0 , viz., occupying the sites. S 0 has a long range behavior in real
X space, but is decaying nearly exponentially in the tight-
Veff (r) ≈ Vmt (rR ) ≡ V0 + [VR (rR ) − V0 ], (2) binding β-representation38 . The so-called band distor-
R tion parameter α = (P 0 )−1 − (P α )−1 = (S 0 )−1 − (S α )−1 ,
which is also used to denote the screening representa-
where we introduced the notation rR ≡ rR r̂R = r − R. tion, gives a relation between the α-representation and
We will in the following omit the vector notation for the unscreened representation. In the nearly-orthogonal
simplicity. This form of the potential makes it possi- γ-representation, α = γ, and oγRL = 0, hence the Hamil-
ble to divide the Kohn-Sham equation (1) into radial γ
tonian HRLR ′ L′ is given by:
Schrödinger-like equations within the muffin-tin spheres,
γ γ
p p
and wave equations in the interstitial region, which can HRLR′ L′ (k) = CRl + ∆Rl SRLR ′ L′ (k) ∆R′ l′ , (7)
be solved separately. The computational scheme that
where CRl , ∆Rl and γRl are the representation-
we present in this paper is based on the muffin-tin the-
independent band center-, width- and distortion poten-
ories developed by Andersen and coworkers, namely the
tial parameters, respectively39,40 . The corresponding
LMTO37–40 , and the EMTO18–20,41,42 , methods. In the
Green’s function in the γ-representation is given by:
following, we briefly review the main ideas and notations
−1
behind the LMTO-ASA method. GγRLR′ L′ (k, z) = [(z − H γ (k))]RLR′ L′ , (8)
 3

where z is an arbitrary complex energy. Further impor- In view of the great progress achieved through the pre-
tant relations among the LMTO representations can be vious implementations of the CPA within muffin-tin or-
found in Appendix A. bitals methods9,12,14–17 we present here in detail a novel
combination of CPA+DMFT in the recently developed
zMTO+DMFT method36 .
III. ELECTRONIC CORRELATIONS AND
DISORDER: THE SINGLE-SITE
APPROXIMATION B. Self-energy-modified effective potential
parameters
In this section we present an LMTO-CPA scheme,
that allows to include local self-energies, on the level of To deal with the important question concerning the
DMFT, for the alloy components. The scheme is imple- effect of interaction, we start by observing that within
mented within the Matsubara representation, and is com- the DMFT the self-energy is local and primarily modifies
bined with the zMTO+DMFT method36 . Section III A the local parameters of the model that describes disor-
briefly discusses the configurational averaging and the der. In this section, we show that the presence of a local
CPA, while in Sec. III B we show how DMFT through self-energy, ΣRLRL′ (z), modifies the potential function P
the Dyson equation leads to a renormalization of the pa- entering in the expression of the Green’s function in the
rameters of the LMTO-ASA formalism. Section III C LMTO-ASA formalism.
combines the ideas of the previous two sections, and pro- We start from the Dyson equation used to construct
poses a combined CPA and DMFT loop. the LDA+DMFT Green’s function:
[GDMFT
RLR′ L′ (k, z)]
−1
= [GLDA
RLR′ L′ (k, z)]
−1
− ΣRLRL′ (z),
(9)
A. Configuration averaging and the CPA DMFT/LDA
where GRLR′ L′ denotes the LDA+DMFT/LDA-level
Green’s function and Σ(z) the self-energy. It is useful to
In disordered systems the configurational degrees of
define an auxiliary Green’s function, the path operator
freedom characterizing the composition are described by α
gRLR ′ L′ , as
a random variable. Consequently, the potentials at sites
are random in space as in quenched disordered solids. A α
gRLR α α
′ L′ (k, z) = [PRl (z) − SRLR′ L′ (k)]
−1
, (10)
particular realization of the random variable constitutes
a configuration of the system in discussion. According to which is valid for a general representation α. In Ap-
Anderson43 only physically measurable quantities such pendix A we present explicit expressions for the poten-
as diffusion probabilities, response functions, and densi- tial functions and auxiliary Green’s functions in differ-
ties of states should be configurationally averaged. As ent representations, as well as their connection to the
these quantities are themselves Green’s functions, elec- physical Green’s function. As is apparent in Eq. (10),
tronic structure methods using Green’s functions are fa- the full energy- and k-dependence of the path operator
vored for the study of disordered systems. is contained in the potential function and the structure
A major development of the theory of disordered elec- constants, respectively. Furthermore, the potential func-
tronic systems was achieved using the CPA. The CPA tion is fully local, i.e., it is diagonal in site index. In
belongs to the class of mean-field theories according to the following, it will prove convenient to first work in the
which the properties of the entire material are determined γ-representation, since here the Green’s function takes a
from the average behavior at a subsystem, usually taken particularly simple form (see Eq. (A2)):
to be a single site (cell) in the material. In the mul-
−1/2 γ −1/2
tiple scattering description of a disordered system, one Gγ,LDA
RLR′ L′ (k, z) = ∆Rl gRLR′ L′ (k, z)∆R′ l′ , (11)
considers the propagation of an electron through a disor-
dered medium as a succession of elementary scatterings i.e., it is the path operator normalized by the potential
at the random atomic point scatterers. In the single-site parameter ∆Rl . Using Eqs. (11) and (A1), the LDA
approximation one considers only the independent scat- Green’s function can be evaluated as follows:
tering off different sites and finally takes the average over 1/2
h i
1/2
[Gγ,LDA
RLR L (k, z)] −1
= ∆ P γ,LDA
(z) − S γ
RLR L (k) ∆R′ l′
all configurations of the disordered system consisting of ′ ′ Rl Rl ′ ′

these scatterers. One may then consider any single site in 1/2
γ
= z − CRl − ∆Rl SRLR
1/2
′ L′ (k)∆R′ l′ .
a specific configuration and replace the surrounding ma- (12)
terial by a translationally invariant medium, constructed
to reflect the ensemble average over all configurations. Hence, the LDA+DMFT Green’s function (9) can be
In the CPA this medium is chosen in a self-consistent written as
way. One assumes that averages over the occupation of
a site embedded in the effective medium yield quantities [Gγ,DMFT
RLR′ L′ (k, z)]
−1
=
indistinguishable from those associated with a site of the 1/2
γ 1/2
medium itself. z − CRl − ∆Rl SRLR ′ L′ (k)∆R′ l′ − ΣRLRL′ (z), (13)
 4

i.e., as the resolvent of the Hamiltonian (7) with an em- introduce the concentration of the respectivePalloy com-
bedded self-energy. From Eq. (13), it is obvious that ponents i, at a site R, as ciR (0 6 ciR 6 1, i ciR = 1).
the same result will follow if the potential parameter CRl In this paper, the DMFT impurity problem is solved
is replaced by an effective parameter, in which the self- within the imaginary-axis Matsubara frequency repre-
energy is embedded, viz., sentation, where the Matsubara frequencies are defined
DMFT
as iωξ = (2ξ + 1)iπT , where ξ = 0, ±1, ..., and T is
CRLRL ′ (z) ≡ CRl + ΣRLRL′ (z). (14) the temperature. In the following, we use the shorthand
DMFT iω = {iω0 , ..., iωξ , ...} to represent the set of all Matsub-
Hence the effective potential parameter CRLRL ′ will now
ara frequencies. Note that the DMFT impurity problem
in general be complex, energy-dependent, and have off-
DMFT in principle also can be solved on the real axis, depend-
diagonal elements. However, CRLRL ′ is still local. With
ing on which numerical solver is used. The corresponding
this effective potential parameter, the LDA+DMFT level
CPA equations will also hold for the real energy axis.
Green’s function can be expressed in a similar form as the
The CPA self-consistency condition requires that the
LDA-level Green’s function, viz.,
sequential substitution by an impurity atom into an effec-
tive, translationally-invariant, coherent medium should
[Gγ,DMFT
RLR′ L′ (k, z)]
−1
= produce no further electron scattering, on average. In
1/2 γ,DMFT γ 1/2 Appendix C we briefly review the CPA+DMFT algo-
∆Rl [PRLRL ′ (z) − SRLR ′ L′ (k)]∆R′ l′ , (15)
rithm for the case of a one-band model system. In
where the model Hamiltonian formalism6,7 the ensemble aver-
γ,DMFT γ,LDA −1/2 −1/2 age (coherent) Green’s function (Gc ) is constructed from
PRLRL ′ (z) ≡ PRl (z) − ∆Rl ΣRLRL′ (z)∆R′ l′ the on-site restricted averages of the componentPGreen’s
= ∆Rl
−1/2
(z − CRl − ΣRLRL′ (z))∆R′ l′
−1/2 function (Gi ), according to the formula Gc = i ciR Gi .
−1/2 −1/2
A similar formula can be written for muffin-tin potentials
DMFT
= ∆Rl (z − CRLRL ′ (z))∆ ′ ′ .
Rl (16) in the multiple scattering formalism, following Győrffy9 ,
and the averaging of the Green’s function can be trans-
Note that due to the self-energy, the effective potential fered to the path operator:
function now has off-diagonal elements.
β i,β
X
g̃RLRL ′ (iω) = ciR gRLRL ′ (iω). (18)
i
C. The combined CPA and DMFT loop
The coherent path operator in the β-representation,
For disorder calculations using the CPA, it is more con- Z
venient to use the tight-binding β-representation, since in β
g̃RLRL [P̃ β (iω) − S β (k)]−1
′ (iω) = RLRL′ dk, (19)
this case only the potential functions (and not the struc-
ture constants) are random15–17,44 (see also Appendix A).
In this case, the representation-dependent potential pa- has been integrated over the Brillouin zone (BZ). In
rameter V β (Eq. (A5)) will be modified accordingly: Eq. (19), the coherent potential function P̃ β (iω) has
been introduced. The alloy component path operators
β,DMFT ∆Rl in Eq. (18) are found through a Dyson equation,
VRLRL ′ (z) ≡ CRl + ΣRLRL′ (z) − . (17)
γRl − βl
i,β β β
X
β,DMFT gRLRL ′ (iω) = g̃RLRL′ (iω) + g̃RLRL ′′ (iω)
For LDA+DMFT calculations, this form of VRLRL ′
L′′ L′′′
should be used for the potential function (A4), the path h i
β,DMFT i,β β i,β
operator gRLR ′ L′ , and the Green’s function (A7), in or-
× PRL ′′ RL′′′ (iω) − P̃RL′′ RL′′′ (iω) gRL′′′ RL′ (iω). (20)
der for the Dyson equation (9) to be fulfilled. Note that
the transformations in Eqs. (A6) and (A7) are now no Here the potential functions P i,β (iω) are computed ac-
longer simply scaled as in the LDA case, but are matrix cording to Eq. (A4), for each type respectively. In order
multiplications due to the presence of off-diagonal terms to close the CPA equations self-consistently, a new coher-
in the self-energy. ent potential function has to be determined at each iter-
In the following, all quantities will be on the dynam- ation. This is done by taking the difference between the
ical mean-field level, and we suppress the common su- inverses of the coherent path operators from the present
perscript “DMFT” for the coherent medium path oper- iteration n + 1 and the previous iteration n, as follows:
ators g̃, the alloy component path operators g i , the co-
herent potential functions P̃ , and the alloy component P̃ β,n+1 (iω) = P̃ β,n (iω) − [g̃ β,n+1 (iω)]−1 + [g̃ β,n (iω)]−1 .
potential functions P i , which will be defined below. The (21)
representation-superscript β is kept. An additional su- The new coherent potential function can be inserted
perscript n appears to represent the iterative nature of into Eq. (19), and the cycle can be repeated until self-
the equations. The superscript i refers to the index enu- consistency has been reached. This is performed for each
merating the alloy components at a certain site. We also Matsubara frequency iω. Once self-consistency in the
 5

FIG. 1. Schematic flow diagram of the main DMFT-CPA loop, as implemented within the zMTO+DMFT formalism36 . Inputs
are the alloy component-dependent potential parameters C i , ∆i and γ i , taken from a LDA-level self-consistent EMTO-CPA
calculation. For each Matsubara frequency along the imaginary energy axis, the CPA equations are solved self-consistently.
The alloy impurity Green’s functions are supplied to the DMFT impurity problem, which is solved self-consistently, giving
alloy component self-energies as output. These self-energies in turn modify the potential parameters which is the key for the
combined self consistency of CPA and DMFT equations. To make the scheme charge self-consistent, the change in the density
due to correlation, ∆n(r), can be added to the EMTO-CPA charge density n(r), and then the Kohn-Sham equations are
iterated until convergence.

CPA equations has been achieved, the Green’s functions ure 1, we present a schematic picture of the self-consistent
for each alloy component can be obtained by normalizing loops.
the alloy component path operators in Eq. (20), using
the transformation in Eq. (A7). These (local) Green’s
functions are then used as input for the DMFT impurity IV. RESULTS
problem, with a separate impurity problem for each alloy
component. For a given interaction strength Umm′ m′′ m′′′
(defined below) on a particularly chosen alloy compo- In order to demonstrate the feasibility of our proposed
nent (i.e. an atomic impurity embedded in the CPA co- method, we apply it to investigate the electronic struc-
herent medium), we solve the interacting problem and ture of the binary Cu1−x Pdx alloy and the semi-Heusler
alloy component self-energies are produced. The average compound NiMnSb, with partially exchanged Ni and Mn
over the disorder corresponds in this case to applying the components.
CPA as described above in Eqs. (18)-(21). The coherent
self-energy is an implicit quantity, in the self-consistency
loops the alloy component self-energies and the path op- A. Computational details
erators are used.
The scheme presented above can easily be incorporated In all calculations, the kink cancellation condition was
within the formalism of the zMTO+DMFT method36 . set up for 16 energy points distributed around a semi-
Here, the EMTO method (see Appendix B for a brief re- circular contour with a diameter of 1 Ry, enclosing the
view) is employed to solve the Kohn-Sham equations for valence band. The BZ integrations were carried out on
random alloys self-consistently, within the CPA, on the an equidistant 13 × 13 × 13 k-point mesh in the fcc BZ.
level of the LDA. The DMFT impurity problem is then For the exchange-correlation potential the local spin den-
solved in the Matsubara representation, using lineariza- sity approximation with the Perdew-Wang parameteri-
tion techniques to evaluate the alloy components Green’s zation45 was used. For the studied alloys a spd-basis
functions, as presented above. This can be done both on was used. For the Cu-Pd system, the Cu 4s and 3d
the LDA-level (setting Σ = 0) and on the DMFT-level states, and for Pd the 5s and 4d states, were treated
(using the self-energy from the DMFT impurity prob- as valence. For the case of NiMnSb, the Ni and Mn 4s
lem). The charge self-consistency is achieved similarly and 3d states, and for Sb the 5s and 5p states, were
as in Ref. [36], by computing moments of the alloy com- treated as valence. The core electron levels were com-
ponent Green’s function at LDA and DMFT level. The puted within the frozen-core approximation, and were
difference between the charge densities computed in this treated fully-relativistically. The valence electrons were
way can then be added as a correction on the LDA-level treated within the scalar-relativistic approximation. Af-
charge computed within the EMTO method. In Fig- ter self-consistency was achieved for NiMnSb, the density
 6

of states (DOS) was evaluated with a 21×21×21 k-point continued by a Padé approximant55,56 .
mesh, in order to get an accurate band gap.
To solve the DMFT equations, we used the
spin-polarized T -matrix fluctuation-exchange (SPT- B. Spectral functions and the Fermi surface of
FLEX) solver46–50 . In this solver, the electron- Cu1−x Pdx random alloys
electron interaction term can be considered in a
full spin and orbital rotationally invariant form, We have previously investigated the electronic struc-
viz. 12 i{m,σ} Umm′ m′′ m′′′ c†imσ c†im′ σ′ cim′′′ σ′ cim′′ σ . Here, ture of fcc-Pd58 , within the framework of the
P

cimσ /c†imσ annihilates/creates an electron with spin σ LDA+DMFT method using the perturbative FLEX im-
on the orbital m at the lattice site i. The Coulomb purity solver59 . Recently, the properties of fcc-Pd
matrix elements Umm′ m′′ m′′′ are expressed in the usual were revisited using a lattice (non-local) FLEX solver60 .
way51 in terms of Slater integrals. In moderately corre- These recent calculations60 support our results using the
lated systems as studied here the modified multi-orbital local approximation of the self-energy. Consequently, we
fluctuation exchange (FLEX) approximation of Bickers study the electronic correlations in the CuPd alloys using
and Scalapino46 proved to be one of the most efficient the same local DMFT technique as we used before. In
approaches47–49,52 . The simplifications of the computa- particular we consider modeling correlations only for the
tional procedure in reformulating the FLEX as a DMFT Pd alloy component.
impurity solver consists in neglecting dynamical inter- Discrepancies between the measured photoemission
actions in the particle-particle channel, considering only spectra61 and the KKR-CPA spectral functions57,62 for
static (of T −matrix type) renormalization of the effec- various Cu-Pd alloys were often discussed in the litera-
tive interactions. The fluctuating potential Wσ,−σ (iω) ture. In particular LDA-CPA results for the Pd partial
is a complex energy-dependent matrix in spin space DOS of the Cu0.75 Pd0.25 alloy reveal a three-peak struc-
with off-diagonal elements W σ,−σ (iω) = U m̃ [χσ,−σ (iω)− ture (black line, Figure 2(a), peaks marked by A, B, and
χ0σ,−σ (iω)]U m̃ , where U m̃ represents the bare vertex ma- C), similar to the DOS of pure fcc-Pd58 . Experimental
trix corresponding to the transverse magnetic channel, data57 on the other hand, see also inset of Figure 2(a),
χσ,−σ (iω) is an effective transverse susceptibility ma- show a contracted band width for the partial DOS and do
not resolve the peak at the bottom of the band (marked
trix, and χ0σ,−σ (iω) is the bare transverse susceptibil-
by C). A detailed discussion concerning these discrepan-
ity. The fermionic Matsubara frequencies iω were defined
cies can be found in Ref. [61]. We note that the frequently
above and m̃ corresponds to the magnetic interaction
discussed reasons for these discrepancies are connected to
channel47,48 . In this approximation the electronic self-
matrix element effects63 , broadening by electronic self-
energy is calculated in terms of the effective interactions
energy63 , and local lattice distortions64–66 , that go be-
in various channels. The particle-particle contribution to
yond the capabilities of standard CPA. Although it is
the self-energy was combined with the Hartree-Fock and
not our intention to address all of the above inconsis-
the second-order contributions47,48 , which adds to the
(ph) σ,σ′ ′ tencies, our current implementation allows us to address
the particle-hole contribution Σ12σ = W1342 (iω)Gσ34 (iω). the possible source of discrepancy in connection to the
The local Green’s functions as well as the electronic self- combined disorder and correlation effects.
energies are spin diagonal for collinear magnetic config- In Figure 2(a) we present the spectral function (DOS)
urations. Their pole structure, when analytically con- for the Cu0.75 Pd0.25 alloy, as a function of the Coulomb
tinued to the real energy axis, produce the appearance parameter U . All curves were evaluated at the lattice
of the peaks located at specific energies determined by constant given by a linear interpolation between that of
the materials characteristics and the symmetries of the pure Cu and pure Pd (Vegard’s law), which in this case
orbitals. corresponds to a = 3.68 Å. Vegard’s law has previously
Since specific correlation effects are already included been shown to hold in a large range of concentrations
in the exchange-correlation functional, so-called “double for Cu-Pd within KKR-CPA67 . As the Coulomb inter-
counted” terms must be subtracted. To achieve this, we action is increased, the peak close to the bottom of the
replace Σσ (E) with Σσ (E) − Σσ (0)53 in all equations of band (C) shifts towards the Fermi energy, while the ma-
the DMFT procedure29 . Physically, this is related to jor peak close to EF (A) remains unchanged. The high
the fact that DMFT only adds dynamical correlations to binding energy peak (C) loses intensity with increasing
the DFT result54 . The Matsubara frequencies were trun- U , and the spectral weight is shifted to higher binding
cated after ξ =1024 frequencies, and the temperature was energy, where it builds up a satellite structure. A similar
set to T = 400 K. The values for the average Coulomb behavior in the spectral weight shift was also found for
U and the exchange J parameters are discussed in con- pure Pd58 . In Figure 2(b) we show the Pd self-energies
nection with the presentation of the results in each case. along the real-energy axis for the interaction strength of
The densities of state were computed along a horizontal U = 3 eV and J = 0.9 eV. Note that for different val-
contour shifted away from the real energy axis. At the ues of these parameters a qualitatively similar behavior
end of the self-consistent calculations, to obtain the self- of the self-energy is obtained. This behavior is that of
energy on the horizontal contour, Σ(iω) was analytically a typical Fermi-liquid frequently encountered in transi-
 7

1 U=3 eV, J=0.9 eV 0.1

(a) A (b) t2g

Cu0.75Pd0.25 C

Re(Σ)
0
0.5
B
Pd DOS (arb. units)
-0.1
-1 -0.5 0 0.5 1
0

Σ (eV)
-0.5 Cu0.75Pd0.25
0

U=0 -1 Re(Σ)-t2g

Im(Σ)-Im(Σmax)
U=1 eV
-6 -4 -2 0 2 -0.05 Im(Σ)-t2g
U=2 eV
U=3 eV t2g Re(Σ)-eg
U=4 eV -1.5 Im(Σ)-eg
-0.1
-2 -1 0 1 2
-2
-14 -12 -10 -8 -6 -4 -2 0 2 -16 -12 -8 -4 0 4 8
E-EF (eV) E-EF (eV)

FIG. 2. (a) Partial density of states of Pd in the Cu0.75 Pd0.25 alloy, as a function of Coulomb interaction U . Inset: Experimental
photoemission data taken from Ref. [57]. (b) Real parts (solid lines) and imaginary parts (dashed lines) of the Pd self-energy
along the real-energy axis, for U = 3 eV and J = 0.9 eV. The t2g -states are shown with blue boxes and the eg -states with red
disks. The insets in (b) are the real and imaginary parts of the t2g self-energy for different concentrations of palladium running
from x = 1.0 (blue) until x = 0.1 (orange).

tion metals68 : a real part (solid lines) that show a nega-

tive slope at EF and the corresponding imaginary parts
with parabolic energy dependence around EF . For the
entire concentration range, the Cu-Pd alloys therefore
exhibit a Fermi-liquid behavior, as seen in the insets
of Figure 2(b). In these calculations we kept the lat-
tice constant fixed for all concentrations to demonstrate
genuine disorder effects, while disregarding volume ef-
fects. Within the normal Fermi-liquid assumption, the
renormalization constant that measures the discontinu-
ity of the momentum distribution at k = kF can be com-
puted as Z −1 = 1 − ∂ReΣ(E)/∂E|E=EF = m∗ /mLDA .
We see that by decreasing Pd concentration (color shifts
from blue to orange), the absolute value of the slope of
the real part of the self-energy decrease monotonically
and the electronic effective mass decreases. The effective
masses are of a similar magnitude as in previous studies
on Pd58,60 .
The results of the calculations including self-energy ef-
fects shown in Figure 2(a) bring the spectral function FIG. 3. (a) Fermi surface (Bloch spectral function) for the
Cu0.60 Pd0.40 alloy with U =4 eV. (b) Difference of spectra
more in line with experimental photoemission data57 (see
between U = 4 eV and U = 0. The color maps indicate the
also inset). Since we neglect matrix element effects due to spectral weight in arbitrary units.
the photoemission process, as well as local lattice relax-
ations, we do not make a quantitative statement concern-
ing the differences between theory and experiment. How- tensities on concentration in these alloys, was traced back
ever, our calculation shows that the proposed method
to the flattening of the Fermi surface sheets with increas-
which combines correlation and alloy disorder effects, ing Pd concentration. According to their results69 the
provides the correct trend in the spectral function. Fermi surface must change from a convex shape in the
In the following we comment upon the disorder and Cu-rich alloy to a concave one for the Pd-rich limit, in
correlation induced modifications in the shape of the a continuous fashion. Consequently the Fermi surface is
Fermi surface of CuPd alloys. On the basis of KKR-CPA forced to be almost flat for some concentration, giving
calculations Győrffy and Stocks69 proposed an electronic rise to nesting phenomena. This was later confirmed by
mechanism which determines short-range order effects further experiments and CPA calculations67,70 .
experimentally seen in CuPd alloys. The experimental According to previous calculations69 a flattened Fermi
observation, namely the dependence of the scattering in- surface in the ΓXK plane was obtained for the
 8

(a) t2g (b)

Potential function (arb. units)

Potential function (arb. units)
Cu0.3Pd0.7 ~ U=3 eV, J=0.9 eV
~ eg
Im P
Im P t2g
x=0
eg
t2g
t2g x=0.3
x=0.5
x=0.7
U=3 eV eg x=1.0
t2g
eg
Cu1-xPdx
U=2 eV

U=1 eV
LDA
-16 -12 -8 -4 0 -16 -12 -8 -4 0
E-EF (eV) E-EF (eV)

FIG. 4. Imaginary part of the effective medium P̃ (E) along the real energy axis. (a) P̃ (E) for several values of the Coulomb
parameters and for a fixed concentration Cu0.3 Pd0.7 . Red lines correspond to the LDA solution (U = 0). t2g states are shown
with solid lines and eg states are shown with dashed lines. (b) P̃ (E) for several values of the concentration x and fixed Coulomb
U = 3 eV and Hund’s coupling J = 0.9 eV. The concentration ranges from pure Cu (x = 0, orange line) to pure Pd (x = 1,
blue line). Only the t2g -orbital is shown.

Cu0.60 Pd0.40 alloy. In Figure 3(a) we plot our results for responding to the bottom of the band. The connection
the Fermi surface of the same alloy. In our calculations between the DOS and the effective medium potential pa-
we used for the lattice constant the value a = 3.72 Å rameter is realized thorough the impurity Dyson equation
(from Vegard’s law), and electronic interactions on the Eq. (20). Increasing the U values in the range of 1 to 3 eV
Pd alloy component were parameterized by U = 4 eV, we see a structure at higher binding energies, related to
and J = 1.2 eV. The Fermi surface is represented in the the appearance and development of the additional pole
(010) and (110) planes of the fcc BZ. The major part of structure of the fluctuating part of the dynamical self-
the Fermi surface consists of the electron sheet centered energy described in Sec. IV A. In Figure 4(b) we study
at the Γ-point. This sheet goes from convex to concave the concentration dependence of ImP̃RLRL (E) for the t2g
with Pd-alloying, forcing parts of the sheet to be nearly orbitals, obtained for the fixed values of U = 3 eV and
flat at 40% Pd. Our result is in good agreement with pre- J = 0.9 eV. For the pure case (x = 0) no electronic corre-
vious KKR-CPA calculations67,70 . To quantify the effect lations are considered and the energy dependence of the
of correlation, we plot in Figure 3(b) the difference be- potential parameter follows the LMTO description13,44 .
tween the correlated (U = 4 eV) and the non-correlated The many-body effects, similar as seen in Fig. 4(a), gain
(U = 0) case. Note the relatively small scale, which in importance for larger concentrations and ultimately
shows that the Fermi surface is insensitive to correlation gives rise to the satellite structure in the DOS (see Fig-
effects. This is expected due to the Fermi-liquid behavior ure 2).
of the system.
The effective medium in the CPA theory is an aux-
iliary quantity that plays a similar role as the effective C. Interplay of correlation and disorder in Mn-Ni
potential in the Kohn-Sham DFT. Contrary to the ef- partially interchanged NiMnSb
fective Kohn-Sham DFT potential, the potential param-
eter P̃RLRL (E) representing the CPA effective medium Half-metallic ferromagnets (HMF)52 are ferromagnetic
is a local, complex, orbital, and energy dependent quan- systems which are metallic in one spin channel, while
tity. In the absence of electronic correlations (U = 0) for the opposite spin direction the Fermi level is situ-
the potential parameter becomes diagonal in the orbital ated in a gap. Such systems would therefore present a
index. In addition, in the absence of disorder the orbital- full spin-polarization at the Fermi level, and have con-
resolved potential parameters reduce to the form seen in sequently drawn considerable interest due to their po-
Eq. (A4), and were shown to increase monotonically44 . tential application in spintronics. One of the first sys-
In Figure 4(a) we plot its imaginary part ImP̃RLRL (E) tems to be characterized as a HMF is the semi-Heusler
for a given concentration (the case of Cu0.3 Pd0.7 ) and NiMnSb71 . The crystal structure of the NiMnSb com-
for different values of the average Coulomb parameter U . pound is cubic with the space group F 4̄3m (No. 216).
We note that at the LDA level in the presence of disor- It consists of four interpenetrating fcc sublattices equally
der (x = 0.3), ImP̃ (E) develops a similar structure to spaced along the [111] direction. The Ni lattice sites
the density of states, which drops at about −8 eV, cor- are situated at (0, 0, 0), Mn sites are at (1/4, 1/4, 1/4),
 9

5 5
(a) 0% (b) 1%
4 0.1% 5% 4
0.5% U=0, 0%
Total DOS (arb. units)

Total DOS (arb. units)

3 U=0, 0% 3

2 Mn-Ni interchange Mn-Ni interchange 2

1 U=3 eV, J=0.8 eV U=3 eV, J=0.8 eV 1

0 0

-1 -1
Impurity peak
-2 NQP NQP -2
Impurity peak
-3 -3
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
E-EF (eV) E-EF (eV)

FIG. 5. (a) Spin-resolved spectral function around the Fermi level for NiMnSb with Mn-Ni interchange, from 0% (no disorder,
light blue solid line) to 0.1% (dark blue dashed line), 0.5% (green dash-dotted line). The red solid line corresponds to U = 0 and
no disorder. (b) Same as in (a), but for the larger degrees of disorder 1% (dark blue dashed line), and 5% (green dash-dotted
line).

and Sb is situated at (3/4, 3/4, 3/4). The position at with increasing disorder. This behavior leads to a re-
(1/2, 1/2, 1/2) is unoccupied in the ordered alloy. In ex- duced minority-spin band gap and a shift of the Fermi
periment, contrary to the DFT prediction, the measured energy within the original band gap.
spin-polarization of NiMnSb is only 58%72 . Several sug- We consider the partial interchange of Ni and Mn,
gestions have been given to explain this large reduction (Ni1−x Mnx )(Mn1−x Nix )Sb, which leaves the overall sto-
in spin-polarization. Among them we mention electronic ichiometry and number of electrons constant. In Fig-
correlation effects52,73 and disorder74–76 . ure 5 we show the total DOS around the Fermi level for
Within the current implementation we have the oppor- different Mn-Ni interchange configurations and different
tunity to study the combined effect at equal footing. In concentrations. For the pure NiMnSb the LSDA minor-
the present calculations we use the experimental lattice ity occupied bonding states are mainly of Ni-d character
constant, a = 5.927 Å. To parametrize the Coulomb in- and are separated by a gap about 0.5 eV wide, while
teraction the values U = 3 eV and J = 0.8 eV were used, unoccupied anti-bonding states are mainly of Mn-d char-
which are in the range of previous studies73,77,78 . Only acter52,71,83 . It was pointed out in Ref. 71 that the open-
the Mn d states were treated as correlated. Because the ing of a gap is assisted by Sb through symmetry lowering
Ni 3d bands in NiMnSb are almost filled, these are sub- with the consequence that the distinction between Mn-t2g
ject to minor correlation effects, as shown previously78 . and Sb-p character of the electrons is lost. In the major-
The effect of electronic correlations is the appearance of ity spin channel (spin up), Ni-Mn covalency determines
nonquasiparticle (NQP) states in the minority spin gap the presence of states at EF with dominant d-character.
(spin down channel) just above the Fermi level. The The pure NiMnSb is ferromagnetic52,71 , with a total fer-
origin of these many-body NQP states is connected with romagnetic (integer) moment of 4 µB , with the main con-
“spin-polaron” processes: the spin-down low-energy elec- tribution stemming from the Mn-site (∼ 3.8 µB ). Upon
tron excitations, which are forbidden for the HMF in interchange disorder the Mn-moment at the Ni site is
the one-particle picture, turn out to be allowed as su- of opposite sign, ∼ −2 µB , while the Mn-moment at
perpositions of spin-up electron excitations and virtual the Mn-site remains positive ∼ 3.7 µB . These values
magnons52,79 . By direct computation, spin-orbit effects remain more or less unchanged by the presence of elec-
were found to be negligible80 in NiMnSb. A partially tronic correlations. The ferromagnetic states results from
filled minority spin gap was obtained but the material re- the exchange interaction of Mn spins which are situated
mains essentially half-metallic with a polarization of the relatively far form each other52,84 . The presence of inter-
DOS of about 99%80 . The interplay of spin-orbit induced change disorder induce magnetic moments of opposite
states and NQP states have been also discussed81 . In sign on the neighboring (Ni1−x Mnx ) and (Mn1−x Nix )
contrast with the spin-orbit coupling, correlation induced sites, as a consequence of magnetic couplings involving
NQP states have a large asymmetric spectral weight in both direct and mediated exchange through Ni and Sb
the minority-spin channel82 , leading to a peculiar finite- atoms. The larger the Mn-Ni interchange, the smaller
temperature spin depolarization effect. It has been shown the total magnetic moment.
that also disorder induces minority-spin states in the en- In the Figure 5(a)/(b) the LSDA(+DMFT) DOS for
ergy gap of the ordered material74 . These states widen smaller degrees of disorder x = 0.1, 0.5% and respec-
 10

0 0
(a) (b)

-1 -1

Im(Σ) (eV)
Im(Σ) (eV)

Mn-Σt (E) Mn-Σe (E)

2g g
-2 0 -2
Ni-site, down -0.2 0%
Ni-site, up 0.1%
Mn-site, down -0.4 0.5%
5% disorder Mn-site, up 5% disorder 1%
5%
No disorder, down
-3 -0.6 -3
U=3 eV, J=0.8 eV No disorder, up U=3 eV, J=0.8 eV
-1 0 1

-8 -6 -4 -2 0 2 -8 -6 -4 -2 0 2
E-EF (eV) E-EF (eV)

FIG. 6. (a) Self-energy along the real axis for the d − t2g states of Mn, at a Mn-Ni interchange disorder of 5%. Dark blue
lines correspond to Mn on the Mn-site, red lines correspond to Mn on the Ni-site. Dashed lines with triangles pointing down
correspond to the spin-down channel, while solid lines with triangles pointing up correspond to the spin-up channel. The Mn
self-energy for NiMnSb without disorder (light blue) has been plotted for comparison. (b) Same as in (a), but for the Mn d − eg
states. The inset in (b) shows the NQP states in the eg spin-down channel for different disorder concentrations.

tively for larger disorder x = 1%, 5% is seen. The results small above EF . The trend is opposite for the spin-up
for the clean, x = 0% (ideal) case, NiMnSb, are pre- channel electrons (red up-triangle). This opposite behav-
sented with red lines (non-interacting, U = 0), and light ior of the manganese self-energies at different sites reflects
blue lines (DMFT). Already at 0.1% disorder (dark blue the anti-parallel configuration of the moments.
dashed line) minority states appear below EF . These It is of interest to investigate how the effect of disor-
states are generated by the presence of Ni impurities at der, i.e. the degree of Mn-Ni interchange, influence the
the Mn site, as previously shown by Orgassa et al.74 . formation of NQP states. For this reason, in the inset of
Furthermore, the upper band edge is shifted to higher Figure 6(b), we plot the Mn-site self-energies of the eg
energy. As the disorder is increased, the width of the Ni orbitals for different disorder concentrations. For minor
impurity states are increased. With correlation, minor- degrees of Mn-Ni interchange (up to 5%), the sudden
ity spin states appear just above the Fermi level. These increase in Im Σ(E) just above EF (dashed blue lines
NQP states arise from many-body electron-magnon inter- of Figure 6), signaling the departure from Fermi liquid
actions52 . At larger degrees of disorder, see Figure 5(b), behavior, remains unaffected. It should also be noted
the impurity states and the NQP states overlap in energy, that the Mn self-energy at the Ni-site (Figure 6, red
removing the spin-down gap. Hence, the combination of lines), follows the Fermi liquid (quasiparticle) behavior
disorder due to the interchange between Ni and Mn sites Im Σ(E) ∝ (E − EF )2 . The Ni d band in NiMnSb is al-
and electronic correlation effects remove the half-metallic most fully occupied, leaving little possibility for magnons
gap in NiMnSb. to be excited, therefore weak electron-magnon interaction
exists in the Ni-sublattice and no NQP states above EF
Figure 6(a)/(b) displays the self-energy along the real are visible in the density of states.
energy axis for the Mn t2g /eg states, respectively, for a
Mn-Ni interchange of 5%. The dark blue lines correspond
to the Mn at the Mn-site, and is similar to the self-energy
for pure NiMnSb (light blue lines). The self-energy be- V. CONCLUSION AND OUTLOOK
haves as in previous calculations73 , namely: the electrons
within the spin-down channel (blue down-triangles) have In this paper we developed a calculation scheme within
a self-energy that is fairly small below EF , but starts the framework of the density functional theory, which
to increase above EF . At around 0.5 eV above EF , the allows one to study properties of disordered alloys in-
self-energy shows a hump, which gives rise to the NQP cluding electronic correlation effects. We model disorder
peak in the spectral function. The self-energy of spin- using the coherent potential approximation and include
up electrons (blue up-triangles) behaves differently, it is local but dynamic correlations through dynamical mean
relatively large below EF , while being small in magni- field theory. Similar to our previous implementation36 ,
tude above EF . The self-energy for the impurity Mn, the DFT-LDA Green’s function is computed directly on
situated at the Ni-site, is marked by the red lines in Fig- the Matsubara contour. Simultaneously the CPA is im-
ure 6(a)/(b). For the spin-down electrons (red down- plemented within the LMTO formalism also in the Mat-
triangle), the self-energy is large below EF , while it is subara representation. Within the LMTO formalism the
 11

CPA effective medium is naturally encoded in the poten- Fund (OTKA 84078). We acknowledge computational re-
tial function, which alone contains the necessary infor- sources provided by the Swedish National Infrastructure
mation about the atomic configuration (assuming that a for Computing (SNIC) at the National Supercomputer
suitable screening representation is chosen). As shown in Centre (NSC) in Linköping.
this paper, the simple parameterization of the potential
function allows us to easily embed the self-energy into
the standard LMTO potential parameters. Accordingly, Appendix A: Relations and formulas within the
the previously developed CPA schemes within the vari- LMTO method
ous muffin-tin approximations can then be used with only
minor changes. Within the nearly-orthogonal γ-representation, the po-
We presented results of the electronic structure calcula- tential function takes the simple form
tion for two disordered alloys: the Cu1−x Pdx system, and
the half-metallic NiMnSb semi-Heusler, in which correla- γ z − CRl
PRl (z) = . (A1)
tions were considered for Pd and Mn alloy components ∆Rl
respectively. For the case of the binary CuPd system, we
see that the inclusion of electronic correlation improves An insertion of this form into Eq. (10), and comparing
the agreement with the experimental spectral functions with Eq. (8), one sees that
for x = 0.25. For a Pd concentration of x = 0.4 the
γ
∆Rl GγRLR′ L′ (k, z) ∆R′ l′ ,
p p
Fermi surface, which is well captured already on the level gRL ′ RL′ (k, z) = (A2)
of the LDA, remains more or less unchanged as correla-
tion effects are turned on. In the second example, the i.e., the Green’s function is the normalized path operator.
partial exchange of Mn and Ni in NiMnSb was investi- In the case of a random alloy, the potential param-
gated, simultaneously with correlation effects. Already eters CRl , ∆Rl and γRl will be site-dependent random
γ
for low levels of disorder, impurity states appear below parameters. Hence, both the potential function PRl (z),
the Fermi level, while many-body induced nonquasipar- Eq. (A1), and the structure constants S γ , will be random
ticle states appear just above the Fermi level. For larger within the γ-representation. This can be seen from the
degree of interchange both these states contribute in clos- transformation in Eq. (6) to the γ-representation,
ing the minority-spin gap. γ 0 0 −1
In the future, the present method will be extended SRLR ′ L′ = [S (1 − γS ) ]RLR′ L′ . (A3)
to compute total energies within the full-charge density
technique21 , making it possible to study the energetics Since γ is (disorder) potential dependent, so is the struc-
of anisotropic lattice distortions22 in alloys. Since the ture constants. To avoid this, it is useful to switch
present method is implemented on the imaginary axis, to the tight-binding β-representation38 , as has been
one can also consider to change the impurity solver to pointed out previously15–17,44 . Within the tight-binding
a continuous-time quantum Monte Carlo85 algorithm. β-representation, the potential function takes the form
This would allow to compute disordered strongly corre-
lated system without any bias. Another interesting venue β ΓβRl 1
PRl (z) = β
+ , (A4)
is to change the arithmetic configuration average used in VRl −z γRl − βl
the CPA to the geometric average used in typical medium
methods86,87 . This will allow one to investigate the ef- where here the representation-dependent potential pa-
fects of Anderson localization43 in realistic materials. rameters V β and Γβ are given by

β ∆Rl ∆Rl
VRl = CRl − , ΓβRl = . (A5)
γRl − βl (γRl − βl )2
ACKNOWLEDGMENTS
The βl -parameters can be found tabulated in several
We greatly benefited from discussions with O. K. An- sources38,88 , and are independent of the (disorder) po-
dersen and D. Vollhardt, whose advice are gratefully tential. Therefore, the structure constants S β depend
acknowledged. AÖ is grateful for the discussion with only on the geometry of the underlying lattice, and only
P. Weightman concerning the experiments on the Cu- the potential function P β is random. The path operator
Pd system. AÖ also thanks W. H. Appelt, M. Seka- β
in the β-representation, gRLR ′ L′ (k, z), is given similarly
nia, and M. Dutschke for help with the graphics. Fi- as in Eq. (10). The following relation allows to transform
nancial support of the Deutsche Forschungsgemeinschaft path operators between different representations38,89 :
through the Research Unit FOR 1346 and TRR80/F6
is gratefully acknowledged. LV acknowledges financial P γ (z) P γ (z) γ P γ (z)
support from the Swedish Research Council, the Swedish g β (z) = (β − γ) β
+ β g (z) β , (A6)
P (z) P (z) P (z)
Foundation for Strategic Research, the Swedish Foun-
dation for International Cooperation in Research and where we have omitted the indices for simplicity. Using
Higher Education, and the Hungarian Scientific Research this transformation, and Eqs. (A1), (A4), and (A2), the
 12

β 88 a
Green’s function can be obtained from gRLR ′ L′ (k, z) as DRL (z) denotes the EMTO logarithmic derivative
function19,42 , and SRLR a
′ L′ (k, z) is the slope matrix
41
.
GγRLR′ L′ (k, z) = The energy dependence of the kink matrix and the sec-
√ √ ular equation poses no difficulties, since the DFT prob-
1 ΓRl Γ R′ l′
β
+ gβ
β RLR′ L′
(k, z) . (A7) lem can be solved by Green’s function techniques (see,
z − VRl z− VRl z − VRβ′ l′ for example, Ref. 91). By defining the path operator
a
gRLR ′ L′ (k, z) as the inverse of the kink matrix,
Note that the transformations in Eqs. (A6) and (A7) are
simply energy-dependent scalings of the path operator,
X
a a
KR ′ L′ R′′ L′′ (k, z)gR′′ L′′ RL (k, z) = δR′ R δL′ L , (B3)
since the potential parameters and the potential func- R′′ L′′
tions are diagonal matrices.
We here briefly mention the accuracy of the presented the poles of the path operator in the complex energy
expressions. The formulas as written above give correct plane will correspond to the eigenvalues of the system.
energies up to second order in (ǫ − ǫν ). A way to improve The energy derivative of the kink matrix, K̇RLR′ L′ (k, z),
on this is by a variational procedure90 , which produces a gives the overlap matrix for the EMTO basis set41 , and
new Hamiltonian, giving eigenvalues correct to third or- hence it can be used to normalize the path operator
der. Correspondingly, the substitution z → z + (z − ǫν )3 p gR′′ L′′ RL (k, z), which gives the EMTO Green’s func-
in Eq. (A7) gives a third-order expression for the po- tion19,42
tential function39,89 . Here, p = hφ̇γ |φ̇γ i is a (relatively X
small) potential parameter. In order to compare the spec- GRLR′ L′ (k, z) = gRLR′′ L′′ (k, z)K̇R′′ L′′ R′ L′ (k, z)
R′′ L′′
tra arising from the different orders of LMTO’s, we in-
vestigated the DOS for various systems using either sec- −δRR′ δLL′ IRL (z), (B4)
ond or third order potential functions, and comparing where IRL (z) accounts for the unphysical poles of
the result with the DOS computed from the Hamilto-
K̇RLR′ L′ (z)19,20 . The use of Green’s functions also fa-
nian through the spectral representation. We found that
cilitates the implementation of the CPA, the reader is
while at second order there was no difference between the
referred to Refs. 19, 20, and 22 for more detailed discus-
DOS, for third order there were clear differences between
sions.
the spectra. This can be attributed to the false poles
present in the third-order potential function90 , since the
energy dependence is now not linear, but cubic. In prac- Appendix C: Illustrative CPA+DMFT algorithm for
tice, we found that this lead to a loss of spectral weight in model calculations
the Green’s function of Eq. (A7), compared to the spec-
tral representation. Hence, we in this paper only consider
The algorithm presented in Sec. III C is formulated in
second order potential functions in Eq. (A7).
the language of multiple scattering for muffin-tin poten-
tials. It represents a generalization of the usual algorithm
Appendix B: Exact Muffin-Tin Orbitals method used for model calculations. In what follows we illustrate
the CPA+DMFT self-consistency loop for the one-band
Hubbard model with on-site disorder (or the so called
One choice of basis for the solution of the Kohn-Sham Anderson-Hubbard Hamiltonian):
equation (1) is the energy-dependent exact muffin-tin or-
bitals18–20 , ψ̄. They are constructed as a sum of the so
tij c†iσ cjσ +U
X X X
H=− ni↑ ni↓ + (ǫi −µσ )niσ (C1)
called partial waves, the solutions of the radial equations
ijσ i iσ
within the spherical muffin-tins, and of the solutions in
the interstitial region. Using this basis, the Kohn-Sham
Here c†iσ (ciσ ) create (annihilate) a spin-σ electron on site
eigenfunctions can be expressed as
i with niσ = c†iσ ciσ and µσ is the chemical potential of
the spin-σ electrons. The on-site energies ǫi are chosen
X
a a
Ψj (r) = ψ̄RL (ǫj , rR )vRL,j , (B1)
RL
as random, while the hopping elements tij are indepen-
dent of randomness; thus, short range order is neglected.
where the superscript a denotes the screening represen- The coherent and alloy component Green’s functions,
tation used in the EMTO theory18,19 . and the corresponding self-energies, are complex func-
a tions of the real energy E. For a multi-band case these
The expansion coefficients, vRL,j , are determined so
the Ψj (r) is a continuous and differentiable solution of quantities are matrices in orbital space. We start the
Eq. (1) in all space. This leads to an energy-dependent self-consistency loop with a guess for the self-energy Σc ,
a
secular equation, KRLR a a
′ L′ (ǫj )vRL,j = 0, where KRLR′ L′
which includes disorder and correlation effects. We em-
is the so called kink matrix, viz. phasize that the combined disorder and correlation effects
enter in oneP single self-energy Σc . The local Green’s func-
a a a −1
KRLR ′ L′ (k, z) ≡ aδRR′ δLL′ DRL (z) − aSRLR′ L′ (k, z). tion Gc = k [E + µ − ǫ(k) − Σc ] is computed from
(B2) the electronic dispersion ǫ(k) (eigenstate of the lattice
 13

Hamiltonian in the absence of disorder and electronic self-consistency loop Gc and Σc are returned into the
correlations) and the initial guess for the self-energy Σc . Eq. (C2) to produce new alloy component Green’s func-
From the coherent (local) Green’s function, alloy com- tions. On a more formal level this algorithm was pre-
ponent (i = A, B, . . . ) Green’s functions are computed sented in Ref. 27.
as
Finally, we mention here the formal equivalence be-
−1 tween the equations discussed in the present appendix
Gi = Gc 1 − (Vi + ΣDMF T

i − Σc )Gc , (C2)
with
P those shown in Sec. III C. The CPA equation Gc =
for a given (fixed) disorder realization. In the next i ci Gi and Eq. (18) are equivalent. The local Green’s
function formula (Gc = k [E + µ − ǫ(k) − Σc ]−1 ) cor-
P
step the many-body problem is solved using the DMFT
methodology: the DMFT bath Green’s function is con- responds to the Eq. (19) in the language of multiple scat-
structed as: Gi−1 = G−1 i + Σi
DMF T
. Specific DMFT im- tering. Finally, Eq. (C2) and Eq. (20) are equivalent as
purity solvers produce alloy component many-body self- they provide the alloy components computed using the
energies ΣDMF
i
T
[Gi ]. In the next step we request that Dyson equation. In our recent paper86 we have exten-
the alloy
P components should fulfill the CPA equation: sively discussed several self-consistent loop algorithms
Gc = i ci Gi . This corresponds to the averaging over for the disorder problem. These include cluster exten-
the disorder realizations. From the newly computed Gc sions and alternative effective medium theories beyond
the coherent self-energy Σc follows directly. To close the the CPA.

1 24
G. B. Olson, Science 288, 993 (2000). A. Georges, G. Kotliar, W. Krauth, and M. J. Rozenberg,
2
A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, Rev. Mod. Phys. 68, 13 (1996).
25
S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, and G. Kotliar and D. Vollhardt, Physics Today 57, 53 (2004).
26
K. A. Persson, APL Materials 1, 011002 (2013). V. Janis and D. Vollhardt, Phys. Rev. B 46, 15712 (1992).
3 27
L. Vitos, P. A. Korzhavyi, and B. Johansson, M. Ulmke, V. Janis, and D. Vollhardt, Phys. Rev. B 51,
Nature Materials 2, 25 (2002). 10411 (1995).
4 28
P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964). Y. Kakehashi, Phys. Rev. B 66, 104428 (2002).
5 29
W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965). G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko,
6
P. Soven, Phys. Rev. 156, 809 (1967). O. Parcollet, and C. A. Marianetti, Rev. Mod. Phys. 78,
7
B. Velický, S. Kirkpatrick, and H. Ehrenreich, 865 (2006).
30
Phys. Rev. 175, 747 (1968). K. Held, Adv. Phys. 56, 829 (2007).
8 31
R. Vlaming and D. Vollhardt, J. Minár, L. Chioncel, A. Perlov, H. Ebert, M. I. Kat-
Phys. Rev. B 45, 4637 (1992). snelson, and A. I. Lichtenstein, Phys. Rev. B 72, 045125
9
B. L. Gyorffy, Phys. Rev. B 5, 2382 (1972). (2005).
10 32
J. Korringa, Physica 13, 392 (1947). P. Wissgott, A. Toschi, G. Sangiovanni, and K. Held,
11
W. Kohn and N. Rostoker, Physical Review 94, 1111 Phys. Rev. B 84, 085129 (2011).
33
(1954). M. A. Korotin, Z. V. Pchelkina, N. A. Sko-
12
G. M. Stocks, R. W. Willams, and J. S. Faulkner, rikov, E. Z. Kurmaev, and V. I. Anisimov,
Phys. Rev. B 4, 4390 (1971). Journal of Physics: Condensed Matter 26, 115501 (2014).
13 34
O. K. Andersen, Phys. Rev. B 12, 3060 (1975). A. S. Belozerov, A. I. Poteryaev, S. L.
14
J. Kudrnovský, V. Drchal, and J. Masek, Skornyakov, and V. I. Anisimov,
Phys. Rev. B 35, 2487 (1987). Journal of Physics: Condensed Matter 27, 465601 (2015).
15 35
J. Kudrnovský and V. Drchal, A. S. Belozerov and V. I. Anisimov,
Phys. Rev. B 41, 7515 (1990). Journal of Physics: Condensed Matter 28, 345601 (2016).
16 36
I. Abrikosov, Y. Vekilov, and A. Ruban, A. Östlin, L. Vitos, and L. Chioncel,
Physics Letters A 154, 407412 (1991). Phys. Rev. B 96, 125156 (2017).
17 37
I. A. Abrikosov and H. L. Skriver, O. K. Andersen, Phys. Rev. B 2, 883 (1970).
Phys. Rev. B 47, 16532 (1993). 38
O. K. Andersen and O. Jepsen, Phys. Rev. Lett. 53, 2571
18
O. K. Andersen, O. Jepsen, and G. Krier, Lectures on (1984).
Methods of Electronic Structure Calculation (World Scien- 39
O. K. Andersen, O. Jepsen, and M. Sob, Electronic Band
tific, Singapore, 1994). Structure and Its Applications (Springer Verlag, Berlin,
19
L. Vitos, Phys. Rev. B 64, 014107 (2001). 1986).
20 40
L. Vitos, Computational Quantum Mechanics for Materials H. L. Skriver, The LMTO Method (Springer, Berlin, 1984).
Engineers (Springer, London, 2010). 41
O. K. Andersen and T. Saha-Dasgupta, Phys. Rev. B 62,
21
L. Vitos, J. Kollár, and H. L. Skriver, R16219 (2000).
Phys. Rev. B 49, 16694 (1994). 42
L. Vitos, H. L. Skriver, B. Johansson, and J. Kollár,
22
L. Vitos, I. A. Abrikosov, and B. Johansson, Comp. Mat. Sci. 18, 24 (2000).
Phys. Rev. Lett. 87, 156401 (2001). 43
P. W. Anderson, Phys. Rev. 109, 1492 (1958).
23 44
W. Metzner and D. Vollhardt, Phys. Rev. Lett. 62, 324 O. Gunnarsson, O. Jepsen, and O. K. Andersen,
(1989). Phys. Rev. B 27, 7144 (1983).
 14

45 69
J. P. Perdew and Y. Wang, Phys. Rev. B 45, 13244 (1992). B. L. Gyorffy and G. M. Stocks,
46
N. E. Bickers and D. J. Scalapino, Ann. Phys. (N. Y.) 193, Phys. Rev. Lett. 50, 374 (1983).
70
206 (1989). E. Bruno, B. Ginatempo, and E. S. Giuliano,
47
A. I. Lichtenstein and M. I. Katsnelson, Phys. Rev. B 57, Phys. Rev. B 63, 174107 (2001).
71
6884 (1998). R. A. de Groot, F. M. Mueller, P. G. van Engen, and
48
M. I. Katsnelson and A. I. Lichtenstein, J. Phys.: Condens. K. H. J. Buschow, Phys. Rev. Lett. 50, 2024 (1983).
72
Matter 11, 1037 (1999). R. J. Soulen, J. M. Byers, M. S. Osofsky, B. Nadgorny,
49
L. V. Pourovskii, M. I. Katsnelson, and A. I. Lichtenstein, T. Ambrose, S. F. Cheng, P. R. Broussard, C. Tanaka,
Phys. Rev. B 72, 115106 (2005). J. Nowak, J. S. Moodera, A. Barry, and J. M. D. Coey,
50
O. Grånäs, I. D. Marco, P. Thunström, L. Nord- Science 282, 85 (1998).
73
ström, O. Eriksson, T. Björkman, and J. Wills, L. Chioncel, M. I. Katsnelson, R. A. de Groot, and A. I.
Computational Materials Science 55, 295302 (2012). Lichtenstein, Phys. Rev. B 68, 144425 (2003).
51 74
M. Imada, A. Fujimori, and Y. Tokura, Rev. Mod. Phys. D. Orgassa, H. Fujiwara, T. C. Schulthess, and W. H.
70, 1039 (1998). Butler, Phys. Rev. B 60, 13237 (1999).
52 75
M. I. Katsnelson, V. Y. Irkhin, L. Chioncel, A. I. Licht- J. J. Attema, C. M. Fang, L. Chioncel, G. A. de Wijs,
enstein, and R. A. de Groot, Rev. Mod. Phys. 80, 315 A. I. Lichtenstein, and R. A. de Groot, J. Phys.: Condens.
(2008). Matter 16, S5517 (2004).
53 76
A. I. Lichtenstein, M. I. Katsnelson, and G. Kotliar, Phys. M. Ekholm, P. Larsson, B. Alling,
Rev. Lett. 87, 067205 (2001). U. Helmersson, and I. A. Abrikosov,
54
A. G. Petukhov, I. I. Mazin, L. Chioncel, and A. I. Licht- Journal of Applied Physics 108, 093712 (2010).
77
enstein, Phys. Rev. B 67, 153106 (2003). H. Allmaier, L. Chioncel, E. Arrigoni, M. I. Katsnelson,
55
H. J. Vidberg and J. W. Serene, J. Low Temp. Phys. 29, and A. I. Lichtenstein, Phys. Rev. B 81, 054422 (2010).
179 (1977). 78
C. Morari, W. H. Appelt, A. Östlin, A. Prinz-Zwick,
56
A. Östlin, L. Chioncel, and L. Vitos, U. Schwingenschlögl, U. Eckern, and L. Chioncel,
Phys. Rev. B 86, 235107 (2012). Phys. Rev. B 96, 205137 (2017).
57 79
H. Wright, P. Weightman, P. T. Andrews, W. Folkerts, D. M. Edwards and J. A. Hertz, Journal of Physics F-Metal
C. F. J. Flipse, G. A. Sawatzky, D. Norman, and H. Pad- Physics 3, 2191 (1973).
80
more, Phys. Rev. B 35, 519 (1987). P. Mavropoulos, K. Sato, R. Zeller, P. H. Dederichs,
58 V. Popescu, and H. Ebert, Phys. Rev. B 69, 054424
A. Östlin, W. H. Appelt, I. Di Marco, W. Sun,
M. Radonjić, M. Sekania, L. Vitos, O. Tjernberg, and (2004).
81
L. Chioncel, Phys. Rev. B 93, 155152 (2016). L. Chioncel, E. Arrigoni, M. I. Katsnelson, and A. I. Licht-
59 enstein, Phys. Rev. Lett. 96, 137203 (2006).
M. I. Katsnelson and A. I. Lichtenstein, Eur. Phys. J. B
82
30, 9 (2002). L. Chioncel, Y. Sakuraba, E. Arrigoni, M. I. Katsnelson,
60 M. Oogane, Y. Ando, T. Miyazaki, E. Burzo, and A. I.
S. Y. Savrasov, G. Resta, and X. Wan,
Phys. Rev. B 97, 155128 (2018). Lichtenstein, Phys. Rev. Lett. 100, 086402 (2008).
61 83
P. Weightman and R. Cole, A. Yamasaki, L. Chioncel, A. I. Lichtenstein, and O. K.
J. Electron. Spectrosc. Relat. Phenom. 178-179, 100111 (2010). Andersen, Phys. Rev. B 74, 024419 (2006).
62 84
H. Winter, P. J. Durham, W. M. Temmerman, and G. M. E. Sasioglu, L. M. Sandratskii, and P. Bruno, J. Phys.:
Stocks, Phys. Rev. B 33, 2370 (1986). Condens. Matter 17, 995 (2005).
63 85
T.-U. Nahm, M. Han, S.-J. Oh, J.-H. Park, J. W. Allen, E. Gull, A. J. Millis, A. I. Lichtenstein,
and S.-M. Chung, Phys. Rev. Lett. 70, 3663 (1993). A. N. Rubtsov, M. Troyer, and P. Werner,
64 Rev. Mod. Phys. 83, 349 (2011).
P. Weightman, H. Wright, S. D. Waddington, D. van der
86
Marel, G. A. Sawatzky, G. P. Diakun, and D. Norman, H. Terletska, Y. Zhang, L. Chioncel, D. Vollhardt, and
Phys. Rev. B 36, 9098 (1987). M. Jarrell, Phys. Rev. B 95, 134204 (2017).
65 87
J. M. C. Thornton, P. Unsworth, M. A. Newell, P. Weight- H. Terletska, Y. Zhang, K.-M. Tam, T. Berlijn,
man, C. Jones, R. Bilsborrow, and D. Norman, L. Chioncel, N. S. Vidhyadhiraja, and M. Jarrell,
EPL (Europhysics Letters) 26, 259 (1994). Applied Sciences 8 (2018), 10.3390/app8122401.
66 88
Y. Kucherenko, A. Y. Perlov, A. N. Yaresko, H. L. Skriver and N. M. Rosengaard,
V. N. Antonov, and P. Weightman, Phys. Rev. B 43, 9538 (1991).
89
Phys. Rev. B 57, 3844 (1998). O. K. Andersen, Z. Pawlowska, and O. Jepsen,
67 Phys. Rev. B 34, 5253 (1986).
I. Wilkinson, R. J. Hughes, Z. Major, S. B. Dugdale,
90
M. A. Alam, E. Bruno, B. Ginatempo, and E. S. Giu- O. K. Andersen, T. Saha-Dasgupta, R. W. Tank, C. Ar-
liano, Phys. Rev. Lett. 87, 216401 (2001). cangeli, O. Jepsen, and G. Krier, in Electronic Structure
68 and Physical Properties of Solids, edited by H. Dreyssé
A. Grechnev, I. Di Marco, M. I. Katsnelson, A. I. Lichten-
stein, J. Wills, and O. Eriksson, Phys. Rev. B 76, 035107 (Springer Verlag, Berlin Heidelberg, 1999) pp. 3–84.
91
(2007). R. Zeller, J. Deutz, and P. Dederichs,
Solid State Communications 44, 993997 (1982).