Professional Documents
Culture Documents
Electronic Structure and Magnetism of Complex Materials 2003
Electronic Structure and Magnetism of Complex Materials 2003
MATERIALS SCIENCE 54
ONLINE LlBRARY
Physics and Astronomy
http://www.springer.de/phys/
Springer Series in
MATERIALS SCIENCE
Editors: R. RuH R. M. Osgood, Jr. J. Parisi
The Springer Series in Materials Science covers the complete spectrum of materials physics,
including fundamental principles, physical properties, materials theory and design. Recognizillg
the increasing importance of materials science in future device technologies, the book titles in this
series reflect the state-of-the-art in understanding and controlling the structure and properties
of all important classes of materials.
Electronic Structure
and Magnetism
of Complex Materials
, Springer
Dr. David J. Singh
Dr. Dimitrios A. Papaconstantopoulos
Code 6391, Naval Research Laboratory
Washington, DC 20375, USA
E-mail: sing@dave.nrl.navy.mil
papacon@dave.nrl.navy.mil
Series Editors:
Professor Robert HulI
University of Virginia, Dept. of Materials Science and Engineering, Thornton HaU
Charlottesville, VA 22903-2442, USA
ISSN 0933-033X
Library of Congress Cataloging-in-Publication Data: Electronic structure and magnetism of complex ma-
terialsl D.J. Singh, D. A. Papaconstantopoulos (eds.). p. cm. - (springer series in materials science ; v. 54)
Includes bibliographical references and index. 1. Magnetic materials. 2. Magnetism. 1. Singh, David ).
(David Joseph), 1958- II. Papaconstantopoulos, D. A. III. Series. TK 454.4.M3 E43 2002 621.34-dc21
2002030474
This work is suhject to copyright. AlI rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or
parts thereof is permitted only under the provisions of the German Copyright Law ofSeptember 9, '965, in its
current version, and permission for use must always he obtained from Springer-Verlag. Violations are liable
for prosecution under the German Copyright Law.
http://www.springer.de
© Springer-Verlag Berlin Heidelberg 2003
Originally published by Springer-Verlag Berlin Heidelberg New York in 2003
Softcover reprint ofthe hardcover lst edition 2003
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
7 Magnetism in Ruthenates
D.J. Singh ..................................................... 297
7.1 Introduction .............................................. 297
7.2 Origin of Magnetism: SrRu03 ............................... 298
7.3 Importance of Lattice Degrees of Freedom: CaRu03 ............ 303
7.4 Sr2Ru04 .................................................. 306
7.5 CaxSr2-xRu04 ............................................ 310
7.6 Ruddlesden-Popper Phases .................................. 313
7.7 Conclusion ................................................ 318
References ..................................................... 319
S. Halilov
1.1 Motivation:
Thermodynamics of Itinerant Magnetic Systems
and Impact on Transport Phenomena
There are two main areas of applications, where the concept of low-energy
magnetic excitations seems to be of major importance: these are thermody-
namics and spin-dependent transport. It is well-established that the wave-like
vibrations govern these magnetic properties, at least at low temperatures.
On the other hand, low-Iying magnetic excitations are a necessary ingre-
dient for transport phenomena because the spin dependence of the inelastic
mean free path of an excited electron is essentially determined by spin-flip
exchange scattering at these excitations (see, e.g. [1,2]). There is more and
more evidence for the strong support of spin fluctuations as a possible me-
diator in the superconducting pairing mechanism (see, e.g. [90]). Magnetism
in reduced dimensionality has become extremely important because of the
very rapid development of thin-film technologies for basic research and for
applications in magneto-transport. A quantitative analysis of the various
phenomena requires information ab out the spectra of low-Iying excitations,
which is often difficult to obtain experimentally.
This chapter is focused mainly on two problems: 1) establishment of a fea-
sible system-specific Hamiltonian that will adequately describe the perturba-
tions of the ground or metastable magnetic configuration, and 2) derivation
of the low-Iying magnetic excitations within the adiabatic frozen-magnon
method, which seems to be the simplest but still practically relevant way for
the evaluation of the lower part of the system Hamiltonian.
All calculations are performed on the basis of spin-density functional
theory (SDFT), thus leaving no space for adjustable parameters.
The attention will be primarily focused on interpretation of the theoretical
results for 3d transition met als and heavy rare-earth metals (REM), as weIl
as their complications for the correct treatment of the experimental data.
Important predictions lead to some new types of application, particularly
showing a route towards a new type of excitation.
2 S. Halilov
(1.2)
i.e. the kinetic energy of electrons plus their interaction with the nuc1ei;
Hint == e; L
U1 U2
JJ drdr' -J;t 1(r)-J; t 2(r')vee (I r - r' I)-J;U2 (r')-J;U1 (r) ,
(1.3)
Le. the Coloumb interaction between electrons;
-~It~ L
U1 U2
JJ drdr'-J;t1 (r)-J;t2(r')V'2
(1.4)
Le. the spin-independent (scalar) part of the relativistic electron-electron
interaction;
Hso == ilt~ L
U1U2
J dr-J;t1 (r)aU1U2 [V'Vle x V']-J;U2(r)
-ilt~ L
U1 U2U3
JJ
drdr'-J;t1 (r)-J;t2(r')a U1U3 [V'vee(1 r - r' I)
Hext == -iltB J L
U1
dr-J;t1 (r)[A+(r, t)V - VA(r, t)]-J;U1 (r)
Vle(r) == - '"'
~ ZR
Ir-RI' vee(lr - r'l) == Ir-r'
1 I' (1.7)
R
4 S. Halilov
The spin-orbit interaction Hso [3-5) comes mainly from two sources: in
classical terms, there is magnetic interaction between spin magnetic moment
JLs = JLBa and magnetic field B = BIe + B ee , which is effectively seen by
an electron i due to its relative motion in electric fields EIe = -V'VIe of
nuclei, and E ee = - V'V ee of all other electrons, respectively. The magnetic
field components from the nuclei and the electrons are BI e = .! V lex EIe
1 . c
and B ee = cV ee X E ee, respectIvely.
Other relativistic terms are not shown, among them the most important
is probably the magnetic dipole interaction between the relative orbital mo-
mentums L == -i( r - r') X V' r and L' == -i( r - r') x V' r', which will,
however, be effectively supressed as the exchange between spin moments and
the spin- own orbit interaction increases. This interaction is spin-independent
and can be classified as a scalar-relativistic effect. On similar arguments, the
effects of dipole spin-spin interaction (not shown) will be assumed immaterial
compared to the exchange between spins and so further consideration of it
will be omitted.
The field operators ~u(t, r) are expanded in terms of single-particle wave
functions rPiu (r )
(1.9)
possible even for the itinerant systems, if the structure of the exchange-
correlation hole is used by building the adiabatie operators, whieh obviously
are to be system-specific. That is, if the spatial distribution of the moment
density at equilibrium shows a weH-defined domain structure, which persists
against orientational momentum density motion, then the new momentum
operators obtained as integrals of the continuous density operators over the
domains, open a way for an adiabatie treatment of the momentum dynam-
ies in a sense similar to that weH knowni, for instance, from the Holstein-
Primakoff [6] theory for magnetic dielectries. The key asset of the latter, by
approapriate choiee of the variables, may be applied also to the itinerant
systems. The operator of the density of the total angular moment um m( r)
at a point r
mi = r drm(r)
. lVi
(1.13)
(1.14)
Once the adiabatie momentum operators mi are established, then the ef-
fective Hamiltonian as an adiabatie approximation to (1.6), in view of the
quite general symmetry consideration such as time and space inversion, can
be written as (see e.g. [7])
(1.15)
where Ho stands for the dynamieaHy stiff part of the Hamiltonian and is
assumed to be stable against the moment transverse motion. The second
term describes the on- and inter-site exchange and is determined by the
parameters Jij ... n , whieh remain unchanged against arbitrary rotation of
the moments as a single entity in respect to the lattiee, i.e. are isotropie.
6 S. Halilov
Obviously, the exchange interaction can not, on its own, change the squared
angular momentum and one of its projections.
The relativistic anisotropy part Hmm is caused by moment-moment dipole
interaction (MMI), with tij ... being the tensorial parameter of the n-th rank.
Particularly, the latter will be reduced by a simplified system-specific treat-
ment to an on-site spin-orbit interaction (SOl) and an on-site electrostatic
orbit-orbit coupling, as presumably leading terms for the magnetocrystalline
anisotropy in transition and REMs, respectively.
The simplified structure of the SOl, taken in atomic-like form, is imposed
by its short-range character and by the supposition that the density of mag-
netic moment is a weak function of distance on the scale of the SOl change.
Recall that magnetic dipole (not shown) and SOl terms are responsible for
the establishment of the thermal equilibrium in the system complementary to
the interaction of the spin subsystem with the lattiee vibrations. The shape-
dependent part of the magnetic dipole interaction, whieh is usually described
via the tensor of demagnetizing coefficients, will be omitted, since we focus
only on bulk systems.
Of course, the legitimacy of the expression (1.15) is yet to be verified for
the systems of interest. Investigation within SDFT shows that the assumption
of conservation of squared angular momentum in transition met als is a very
crude approximation [56], whieh qualifies them as non-Heisenberg systems.
However, this observation does not matter much as long as the amplitudes of
transverse moment vibrations in the vicinity of equilibrium are small enough
to treat the squared angular moment um as constant. This fact remarkably
simplifies the derivation of low-lying magnetie excited states in itinerant
systems, but, as was mentioned, it is to be proved for each system. More
has to be added regarding the exchange interaction as a driving force for an
equilibrium and moment dynamies. It is easy to verify the following expansion
for the exchange part
as in [48,49]
(1.20)
specifying the details for every system under consideration. The annihilators
ak annihilate an electron in a k-shell.
Since we aim at the magnetic structure and dynamics performed at a fea-
sible level, the relevant magnetic variables are needed. The operator of the
1 Low-Lying Magnetic Excitations 9
conduction electron spin moment vector (in units of fi) attached to lattice
site i is introduced (in the Heisenberg picture) as
(1.22)
where ass' is the vector of Pauli matrices and the integration domain is the
site cell Vi, the whole r-space being partitioned into mutually disjoint cells
Vi. The quantum expectation of this vector is
(1.24)
The vector 2J-tBSi(t) (with Bohr magneton I-tB) is the quantum expectation of
the conduction electron site magnetic moment. We will not concern ourselves
with the details of the Kohn-Sham formalism generalized for the case of
nonuniformly magnetized systems here, since a treatment may be found,
e.g., in arecent monograph [86] devoted to the theory of itinerant electron
magnetism. Instead, our attention will be focused only on specific points
to make clear how to evaluate the force constants of the dynamic problem
in a general case of nonuniform spin distribution. The orbitals cPkn are the
eigenvectors of the spinor equation:
[
-1
fi2
2 - 'iJ
A
m
2
+ v(r) + Llv(r)U+(r)O'zU+]
A A
(r)
A A
88'
_
cPkns' - f.knscPkns, (1.25)
is introduced in order to enable the use of the results weIl known from the
many-body theory for uniform systems, Le. the LSDA-SDFT method. The
external magnetic field h( r) is a neccessary ingredient in a moment response
calculation, when the self-consistent Kohn-Sham equation is constrained by
a certain orientation and magnitude of the external field.
Obviously, in self-consistent calculations the local quantization axis e( r)
will not necessarily assume the orientation of the magnetic field h(r), un-
less the latter is not in contradiction with the local system symmetry. In
fact, (1.27a) and (1.27b) provide a straightforward way to obtain the local
quantization axis e( r) at every iteration of a self-consistent LSDA-SDFT
calculation. In accordance with the latter, the exchange-correlation energy is
a functional of the local eigenvalues of the spin-density matrix
with Exc being a function of ni, interpolated from many-body calculations for
a homogeneous electron system, n( r) = nl (r) + n2 (r) is total charge density,
and m( r) = nl (r) - n2 (r) is moment density.
There is another way of doing constrained LSDA-SDFT calculations, more
straghtforward for the purposes of the force constant calculations, when an
arbitrary moment configuration, given by a set of the LQA vectors e(r), is
needed in the vicinity of the ground state. That is, instead of performing
a self-consistent calculation as a response to an external fixed field h(r),
one keeps a needed set e( r) by "hand", which is obviously equivalent to the
applying of an external field h(r), where this field is now getting effectively
changed at every iteration. This change is caused by a constraint upon the
LQA vectors, which means a forced setting to zero of the non-diagonal com-
ponents with an effective meaning of the adjustment of the external field.
Formally, it corresponds to the account of only the projection of the local
moment 1n( r) onto the LQA vectors in the LSDA-SDFT expression for the
exchange-correlation energy
Since via this constraint, the orientation of 1n(r) need not necessarily be
the same as that of e(r), there is a certain amount of inaccuracy because
of this deviation between the intentional moment configuration e (r) and the
calculated one. The amount of the mismatch has been thoroughly investigated
in [72J. Here, we note that this type of "preprogrammed" numerical error,
by ordinary choice of spin moments as the site-associated muffin-tin integrals
over the sphericaIly averaged spin density, increases as the the criterion for
adiabaticity is less weIl fulfiIled.
1 Low-Lying Magnetic Excitations 11
demand for the adiabatic wave function, where the variables mi(t) evolve
with time in such a way that kinetic processes associated with f[t] are effec-
tively averaged out from the equation ofmotion for mi(t). The corresponding
approach will be labeled as the frozen spin-wave (magnon) method. Once the
adiabatic variables are established, the general formulation of the low-lying
excitations will be done in terms of the free system eigenstate energy as
a constrained functional of the given variables and the coupling constants as
its second derivatives with respect to the deviations of the variables
J~;X' = {)2 E[m]
(1.31)
U - J;I ",J;I ","
um i umi ,
(1.35)
are given in terms of the second derivatives of the constrained system eigen-
functions and eigenvalues with respect to the moment deviations in the vicin-
ity of a given magnetic configuration
(1.36a)
(1.36b)
respectively. The first quantity is known in the literature as the Berry curva-
ture of the adiabatic coherent state [70], i.e. the constrained magnetic ground
state 'ljJ[mi] as a solution of the variational problem
and the second matrix is to be identified with the force constants Jii' within
the frozen magnon scheme. To demonstrate the equivalence of the methods
when the variables mi are adiabatically really well-defined, let us consider
a periodic system uniformly magnetized along z and specified by a single
adiabatic variable mi on the i-th site. In this case, the Born-Oppenheimer
equation (1.35) in the reciprocal space can be cast into the form
(1.38)
where
nXY(q) = L n:le iqR . (1.39)
R
Suppose that spin densities centered on different sites do not overlap, the wave
function depends on the change of the moment orientation in the lowest order
of the perturbation theory as
'ljJ(J~ ( 1-((}/2)2)
(}/2 'ljJo, (1.40)
14 S. Halilov
where () stands for the deviation angle between z and the moment. Then
[}X y
R
~
'"
( {){)'ljJ
mo
I~)
{) '"
~8
2 O,R,
mR
'" m (1.42)
i.e. the same dispersion which comes out on the basis of the Heisenberg
equation of motion.
E xc = L
/-tR
1V!'R
drn(r)EXc[n(r) , e/-tR· m(r)] , (1.45)
where e/-tR stands for the local spin quantization axis at the (JLR)-site, and Exc
is the density of the exchange-correlation energy considered as a functional of
charge n( r) and vector spin m( r) densities, respectively. Note that as a result
of the constraint on the orient at ion of the exchange-correlation field in the
course of SDFT calculation, the local quantization axis e/-tR is parallel to
the local exchange-correlation field, but deviates from the expectation value
of the magnetic variable m/-tR. As was mentioned above, constraint LSDA-
SDFT calculation can be performed in two ways: either one holds the LQA
vectors e/-tR, ignoring the change of the charge density n( r) caused by the
deviation of the local vector of spin density m(r) from the orientation of
the local exchange-correlation field e/-tR, or one performes a self-consistent
calculation of the m( r) as a response to an external static transverse field
1 Low-Lying Magnetic Excitations 15
h( r). At first glance, the latter method seems to be more legitimate and more
accurate. But in practice it turns out that the more time-consuming static
response method deviates significantly from the constraint-on-exchange-field
method only at those parts of the phase space where the adiabatic approach
itself is a poor approximation. This is usually the case for the high-energy part
of the collective excitation spectrum, where the electron transitions between
opposite-spin states cause the longitudinal change of the moment with an
amplitude large enough on the scale of transverse vibrations. Consequently,
the frozen-magnon method works, for example, with much higher accuracy
in Fe, than in Ni, since the large exchange-correlation field in Fe makes the
moments in Fe essentially stiffer than that in Ni.
Determination of the force constants (1.31) when a system is near its
equilibrium is, even by adiabatic treatment, a tricky procedure: it may be
done only when the equilibrium itself is well defined. The latter is generally
given by a nonuniform set of the LQA evR and the variable amplitudes
mvR, which is an interplay between exchange and relativistic forces. Even
the nonrelativistic part of the problem, which is to be specified within the
by definition nonrelativistic constraint LSDA-SDFT, is a kind of journey full
of mishaps: if the equilibrium LQA for some nonuniformly polarized system
is not known, then it is not clear what kind of constraint is to be used in
order to self-consistently determine the equilibrium. In practice, of course,
the problem could be solved iteratively: starting from a relatively simple
uniform consraint on the LQA, i.e. assuming all the LQA are the same,
one can obtain the force constants for that constraint equilibrium, then by
adding the single-site relativistic terms into the single-particle LSDA-SDFT
equations perturbatively, one can recalculate the new LQA, then proceed
with redetermination of the new equilibrium constraint by the just-obtained
LQA, and so on. From that point ofview, the frozen magnon method, in fact,
becomes as involved as the response-function formalism. On the other hand,
it still keeps the interpretation of the moment dynamics on a feasible level.
To complete this section, we give here the details of the helix technique as
probably the easiest one for an approximate evaluation of the force constants
(1.31) within the frozen magnon approach, which parametrize the exchange
energy in the vicinity of a uniformly (collinearly) polarized phase. The tech-
nique is applied in reciprocal space, which has a great advantage against
the real-space methods whenever the long-range effects like the Friedel oscil-
lations have to be taken into account. To cover the full phase space of spin
configurations, it would, in a Fourier representation, be necessary to introduce
two independent wave vectors for the position dependence of () and cp. Since,
however, in linear spin wave modes () is constant, we keep it constant on each
sublattice 1/ (if there are several sublattices) and consider only configurations
(1.46)
16 S. Halilov
JE = - 12:--
2
()"" -, + ...
()*,J""
q ,
(1.48)
",,'
where
(1.49)
for v #- v'. The force constants j~'" are related to the real space by the
Fourier transform
J-",,'
q
(;;;;-J"[,,'+Rjym,,,e
= L~ ym" ~ iq(Tv-T~-R)
, (1.51)
R
theoretical method is, the less suitable it becomes for elucidation of the
experiment. Apparently, the dynamical spin susceptibility derived directly
from the equation of motion for the electronic wave function is better than
that obtained by the equation of motion for the effective adiabatic moment
variables, on account lack of an established method to verify the adiabaticity
of the variables and check the accuracy of the effective moment Hamiltonian
projected out from the original many-particle one.
Nevertheless, we have to say something in favor of the adiabatic methods.
The obvious advantage of them is that, first they deliver a Hamiltonian which
is much easier to analyze and apply for practical purposes, for example,
for spin-dependent transport and thermodynamics, and second they make
it much easier to interpret experimental data, for example, the results of the
inelastic neutron scattering. The latter is best described in terms of the spin-
spin correlation function, or equivalently the imaginary part of the system
dynamical spin susceptibility X, which is a functional of an external field
applied to the system and describes the response of the system to that field
h(r, t),
8m a (r,t) = J
dr'dt'Xaß(r,r',t,t')hß(r',t'). (1.53)
500
>
~ 400
g 300
,c: 200
100
o
strong Kohn anomaly in iron instead of the gaps. What is still not clear
is whether the claim is in favor or against the adiabatic model, since the
measured excitations at high energies are highly damped [37]. To see whether
the double-branch mode in nickel may be reproduced within the adiabatic
model, we will briefly refer to the initiating paper [24], where the many-body
Hamiltonian is adjusted with just two band parameters Weg and W t29 ) for
the screened Coloumb matrix element, as evidently preimposed by crystalline
field effects. This means if the exehange-eorrelation hole had an adequate do-
main structure causing the spin density to effectively split into two adiabatic
moments, each for instance of predominantly the same symmetry character eg
and t2g, then the double-branch mode in the spin excitation spectrum would
be rather a trivial matter provided a relatively weak exchange force linking
those moments. This hypothesis has been checked within the LSDA-SDFT
and the results are briefly discussed in Sect. 1.6.1.
Applications of the nonadiabatic methods to the systems with nonuniform
magnetic configurations and remarkable effects of the orbital polarization, as,
for example, in the REM, have not yet been done and are expected to be very
tedious, since there is no accurate all-electron scheme treating the exchange-
correlation and relativistic effects for the itinerant and localised electrons on
the same basis.
1 Low-Lying Magnetic Excitations 19
a <r
Fe
C sial
fie%
t2g
El<Change
-1.5eV
·1/2
Spin-orbit
~J)
..
~
3d
-2 'V
s m
Spin-orbit Crystal field
Gd -+3
_+2
~ EJ<Change _+1
-1/2 _
_ _ 10
~
_-2
--3
s m
....i..
-+2
+3
_+1
+1/2 -_ _10
--2
T-3
Fig.1.2. Various sources of forces: hierarchy of interactions of the magnetic elec-
trons in the case of the transition met als and REMs
20 S. Halilov
A relevant choice of the variables in the 3d systems has to take into account
the specifics of the spatial extension of the partly fiHed 3d orbitals. The latter
are essentiaHy more extended than open-shell 4f-electrons, and therefore the
direct exchange between spin moments of the 3d electrons becomes compara-
ble to that mediated by conduction s electrons. That there are spatial regions
of positive and negative spin polarization was weH known long ago from the
neutron Bragg scattering measurements [60]. The Fourier transform of the
data shown in Fig. 1.3, is just an illustration of the fact that the magnetization
density has rather a domain structure, with positively polarized domains elose
to nuelei and negatively polarized domains in the interstitial.
In fact, the specific structure of the exchange interaction in 3d-met als,
where the spatial distribution of the spin polarization seems to have two
backbones, serves as a good reason to introduce at least two moments per
site. The question is which of the following separation methods is more rele-
vant. The first method uses the fact, that the radial distribution of the spin
density has, by and large, two maxima, one attributed to the relatively heavy
elose-to-nueleus electrons and a second contributed to by the highly itinerant
interstitial electrons. This is what we leam from Fig. 1.4. This way is rather
coarse and disregards the local angular symmetry. Another way, which seems
to be more relevant for the description of wave-like excitations, is to use
the angular symmetry decomposition of the spin density caused by crystal
field effects and verify whether this decomposition correlates with the spatial
structure of the density. The quest ion will be discussed later, and whatever
the relevant choice is, we ass urne that the set of basic variables contains two
Fig. 1.3. Magnetization distribution in iron retrieved on the basis of the neutron
Bragg scattering [60]
1 Low-Lying Magnetic Excitations 21
2
0.5
Fe Ni
o 2 3 o 2 3
r (a.u.) r (a.u.)
Fig. 1.4. Radial spin density distribution in Fe and Ni, resolved in angular character
(upper panel) and spin (lower panel). The solid lines denote the d-projected density;
the dashed lines stand for the remaining (all other but d-state) contributions
moments [Si, Si] per site. In the second quantized picture, they are on the
i-th site introduced as
(1.54a)
(1.54b)
where 0" ss' is the vector of Pauli matrices. The volumes ~(1,2) enclose the
different parts of spin density along with either of the above-mentioned two
concepts, and are spatially disjoint from each other, as weIl as from other
sites. The quantum expectation of these vectors
(1.55)
will be used by determination of both the ground and the excited states. The
adopted picture is apparently beyond the atomic Hund's rules.
The equal time commutation relations of the introduced spin operators
can, with limited accuracy, be shown to be
within either of the two methods. In these relations, a, ß, and 'Y represent
Cartesian indices, and Eo.ß'Y is the antisymmetric fundamental form.
The first separation method is illustrated by Fig.l.4, obtained on the basis
of the LSDA-DFT calculations, which shows that the angular-moment um
projected spin density is spatially weIl separable for the sp and d-character
states, whose maxima are in interstitial and intraatomic regions, respectively.
This spatial separation is rat her weIl defined in Ni, since there is a well-defined
minimum in the spatial distribution of the total spin density. In the case of Fe,
a similar partitioning of the spin density is less relevant. Obviously, the sp-
electrons in the case of Ni are more screened by the almost completely filled
d-shell than in Fe, which also makes the spin density distribution appreciably
different in Fe and Ni. The total spin moment um associated with almost free
electrons and induced by the essentially less itinerant electrons, is antipar-
allel to the moment of the latter, which obviously is an interplay between
intraatomic exchange and the effects of crystal field. The motivation for the
second separation method is not known to the author, being a matter for
a different investigation, which could probably be based on a full-potential-
like technique.
We suppose that the adiabatic Hamiltonian for spin motion has a bilinear
form
(1.57)
ij ij
Heisenberg-like expression and to analyze the behavior of m(O, q). The spirals
have been intentionally used to cover the whole phase space of the moment
configurations. The results are presented in Fig. 1.5.
What is most striking in these results is the remarkable stability of the site
moment m, particularly for bcc-Fe. The moment is significantly reduced only
if both q and 0 are large: only if large q (and hence large wq ) magnons have
occupation numbers rv N (that is, rv 1 per site) , they induce longitudinal
spin fiuctuations and reduce the moment. That longitudinal magnetization
density waves should have higher energies than transversal ones has been
repeatedly stated in the literat ure [31,32]. It is particularly this stability of
the moment that justifies the whole approach of Sect. 1.4.2, above all (1.34).
125
.=, 2.0 +-....... ---4-_. . . 25
--t::::::t--""""'*==t 2.0
1 15
::.
-12
;;
1.5
1.2
~ 15 15 ~ 0.9 0.9
~ 10 bcc Fe 10 ~ 0.6 fee Co
·~o_s 05 -~O.3
'" '"
o '1_(0.0.1) o q_(O.O.I )
'" '1-(0.0.112) '" '1-(0.0.112)
o '1-(0.0.113) o q - (O.O.II~ )
12
20
10
0 0 o ~~~~~~~----~~o
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) Ang le 0 {degl (b) Angle 0 {degl
1
.=,0.6 0.6
"11 0,4 OA
0
'"
>: 3 0 '1-(0.0.1)
~ .. q _(O.O. II2)
- 0 '1-(0.0.114)
~ 2
IJJ
Ö I
~
o ~....,-;a:::::::~~~~---=:::=_----l 0
(c) 0 10 20 30 40 50 60 70 80 90
Angle 0 Idegl
The radial distributions of the charge and spin densities are nearly the same
in all heavy rare-earth metals, which is clear from Fig. 1.6. In constrast to the
partly filled orbitals of 3d metals, the 4f orbitals centered on the nearest sites
are significantly less overlapped, which makes the indirect exchange between
them a dominant mechanism for the low-lying excitations, along with the
anisotropie terms of the relativistic plus electrostatics nature.
I'
\
- - x1 60 , \ - - x1
-
/
N Gd /
\ f Tm I
\
\
f
~ 40 \
40
...0. \
\
;t
~ 20
,,
I 20
..9: /\ I
"-
0 0
~6 2
"2
~4
--
<-0. 2
0 0
0 1 2 3 0 1 2 3
r (a.u.) r (a.u.)
Fig. 1.6. Radial spin density distribution in Gd and Tm, split in angular momen-
turn and spin space
1 Low-Lying Magnetic Excitations 25
The electronic ground state and the low-lying excited states of element al
rare-earth (RE) met als are characterized by delocalized conduction electrons
of predominantly 5d orbital character, but hybridized with 6s and higher
orbital contributions, and a very localized partially filled 4f shell in its Russell-
Saunders state with maximum total 4f spin Sand maximum compatible with
S total 4f orbital momentum L. The only exception, Q-Ce, which is stable
under pressure with delocalized f-electrons, is not considered here. In most
cases, the 4f shell is that of a RE 3 + ion with three electrons donated into
the conduction states. Here, the exceptions are Eu and Yb, the 4f shells of
which are those of RE 2 +. Up to room temperature, most of the 4f shells
are stable in the lowest fine-structure level J. Although Sm and Eu are the
only cases where an admixture of the next higher fine-structure level J + 1
plays a major role in the ground state, and these two more complex cases
are again not considered here, we have to consider J - J' transitions in
magnetic excitations and hence will treat Sand Lasindependent variables
not fixing J. Finally, Yb2+, like LuH , has a filled f shell and hence no f
moment. In what follows, the heavier magnetically ordering RE met als Gd
through Tm, which crystallize in the hexagonal close-packed (hcp) structure,
are considered. Their 4f shell is essentially in the lowest fine-structure level
J of the REH ion in the ground state.
Having in mind the situation just described, we consider a suitable or-
thonormal basis of conduction electron orbitals, 4f orbitals, and deeper core
orbitals (which we will not refer to explicitly), and split the fermionic field
operator ofthe electrons -0(r, s) = -0c(r, s) +-0f(r, s) +-0core(r, s) into the con-
duction electron part, and the core electron part. Due to the orthonormality
of the considered orbitals, the operators -0c, -0f, and -0core anticommute with
each other. Therefore, a set of the relevant magnetic variables consists of the
operators of the conduction and the open-shell electrons. The operator of the
conduction electron spin moment vector (in units of h) attached to lattice
site i is introduced (in the Heisenberg picture) as
(1.58)
where a 88' is the vector of Pauli matrices and the integration domain is the
site cell Vi, the whole r-space being partitioned into mutually disjoint cells
Vi, e.g. the Voronoi cells.
The operators of the 4f electron spin and orbital moment vectors, respec-
tively, on lattice site i are introduced as
(1.59)
and
respectively, while (S;) = S(S + 1) and (Lr) = L(L + 1) are assumed fixed
at the values Sand L according to the first and second Hund's rules.
With the commutation relations of the involved operators it is easily
shown that the introduced quantities at equal time obey the algebraic re-
lations
H = H({Si})
+ L {Hs(S;) + HdL;)
"
-fiSi · Si - AiSi · Li - L~in
n
rr
a
(Liara }. (1.63)
The first line, the site off-diagonal part of the Hamiltonian, describes the
intersite exchange coupling of the conduction electron site spin moments.
The rest of our Hamiltonian is site diagonal. It contains terms that establish
the first and second Hund's rules for Sand L in the 4f shells and that we
will not give explicitly since they are constant operators in the sector of
the Hilbert space of states relevant for our purposes. The last line contains
in turn the on-site Hund's rule couplings between conduction electron and
4f spins with coupling constants 1;, the spin-orbit coupling within the 4f
1 Low-Lying Magnetic Excitations 27
______ 5 i J ißJ
/ _ Inter-site
SI ~_ excl1<11ge
I nt ra-site "!'l
excnange ,..--
I. ~ri ( ~
"=2,4,6
' )" axial anisotropy
Fig.1.7. Variables and forces for heavy rare-earth metals taken at some instan-
taneous moment: basement of the orbital moments symbolize the electrostatic
coupling of the latter to the rest of the crystal
sheHs with coupling constants Ai , and the crystal-field terms with crystal-
field parameters ~in, where n is a vector of integer components n x , n y, n z ,
the sum of which yields the order of the corresponding crystal field term. Not
aH ~in are independent, they are related to the point symmetry of each site
i . To visualize the model, a diagram in Fig. 1.7 shows the basic variables and
the forces at some instant snapshot.
The parameters of the model Hamiltonian have been yielded by con-
trasting with SDFT calculations, which are still very seldom for the REMs
[64,65], and whose reliability might still suffer from the limited accuracy
of the LSDA. It is weH established that the moments in Gd, Dy, and Tb
stabilize at a uniform ferromagnetic phase, whereas Er and Ho prefer a conical
helix. The situation correlates weH with the LSDA-SDFT calculations [64]
performed within the atomic sphere approximation (ASA) , as well as within
the fuH-potential calculations [65], the latter, however, for reasons that are
yet unclear, prefer a collinear ferromagnet configuration for the ground state
ofHo.
The crystal-field parameters have been calculated [66] from a self-interac-
tion corrected SDFT self-consistent crystal calculation with spherically aver-
aged 4f charge clouds and spin-averaged conduction electron charge densities
on the one side, and the asphericity of the RusseH-Saunders 4f charge density
on the other side.
28 S. Halilov
A = 4ne1i 2
2m2c2
J d ",2 ( ) dv(r)
r r'i'4f r dr . (1.64)
A---~ (1.65)
•- 28i '
(1.66)
1= 8n
3s
Jdrr2cp~f(r) nct(r) (" ~c-l-(r) A[n(r)] ,
n r
(1.68)
where n cs is the conduction electron spin density, and n is the total density.
If we put all Ai = 0, then the spin-dependent parts of the Hamiltonian de-
couple from the orbital angular-momentum-dependent parts, and the former
are independent of rotations of r-space, a property called magnetic isotropy.
Under this assumption, Ai = 0, which is underlying the scalar relativistic
approximation in relativistic electronic structure theory, one may calculate
the constrained SDFT ground state energy for fixed spin-moment config-
urations and Taylor expand it around the unconstrained minimum energy
configuration. By doing so, we find that the resulting energy surface on the
space of {Sio;} configurations is steep in longitudinal directions with respect
to the ground state moments and comparatively shallow in transversal di-
rections. As a consequence, magnons are essentially transversal excitations.
The longitudinal steepness comes from the strong intraatomic exchange (in
the Hamiltonian (1.63) described by the Hs(fff) term in connection with
1 Low-Lying Magnetic Excitations 29
the Iis i . Si term), which, due to isotropy, does not produce a transver-
sal force. Hence, if considerations are restricted to transversal spin motion
s; = (s? +OSi)2 = (s?)2, then the ground state configuration {s?",} must obey
the condition OSi'" 2: j (#i) Ji"'/ s~ß = 0 for all OSi"" Therefore, the Taylor term
osJ8s may be cast into the form sJs + const., resulting in
E SDFT = E o - "21L.J
",
Si"':!;j"'ß Sjß + ... , (1.69)
ih
moreover, the above-mentioned magnetie isotropy implies the scalar form
.7/;ß = o"'ßJ;j of the exchange constants.
The special adiabaticity assumption for itinerant spins, [56,57] meaning
that fast electronie intersite correlations are to be neglected,
(1. 70)
implies that to second order the intersite Hamiltonian of (1.63) can be cast
into the Heisenberg form: H({Si}) = -~ 2: i#j Si",Ji"'/Sjß' with J~ß = :!;~ß
for the effective exchange constants.
There are two situations possible for an adiabatie scenario
(1.71)
One is for weak I i and Ai compared to electron-ion coupling (which is at least
of the order of a Rydberg), and the other is for I i and Ai strong compared
to crystal fields and S ~ 1/2. For RE both may be weIl realized.
For a discussion of the constraint SDFT treatment of non-collinear sit-
uations in this context and references to related work see [57]. It has been
stressed there that nss/(r) and vss/(r) will not be collinear unless self-consist-
ency is reached. As was already mentioned above, this implies an extra error,
if the direction of the vector spin density is treated as constant inside atomic
spheres, as is usually done. This error can be minimized by a special treatment
of spin and field directions [72].
In cases where the site symmetry is not too high, the second order crystal
field constants usually dominate the crystal-field part of the Hamiltonian.
Although higher order crystal-field terms do not pose any problem, in the
following, for the sake of keeping the simplicity of the expressions, we restriet
all formulas to second order crystal-field terms only. The crystal-field part of
the Hamiltonian (1.63) can then be cast into the symmetrie bilinear form
(1.72)
n '"
and we are essentially left with the Hamiltonian
1",
-"2 L.J JijSi . Sj -
A A A A A A A
For some general situation, it is more convenient to treat all the variables as
components of a single vector operator
(1. 74)
(1. 75)
(1. 76)
is averaged over the fast electronic motion. Then, observing the adiabaticity
assumptions yields the adiabatic equations of motion:
(1. 77)
The adiabatic ground state configuration m?t is obtained from these equa-
tions by putting the time derivatives to zero.
Linearization with respect to small-amplitude transversal motions, intro-
duction of separate local coordinates for the mit, with z-axes pointing in the
directions of m?t, and complexification of these equations of motion leads to
the linear equation system for the motion of the moments in with respect to
the local coordinates (it abbreviated by i and i't' by j):
(1. 78)
where the notation means a = x, Yi ö: = =
1, (-l)Y = -li and
y, Xi (-1)x
Je' = eiß' Jij . ell , ß, 'Y = X, y, z with local (and possibly t dependent)
co ordinate unit vectors eiß, and no summation over repeated a on the r.h.s.
If the magnetic ground state forms a periodic structure, then Fourier
transformation on this lattice with lattice vectors Rand basis vectors TI-" that
is, site positions R i = RI-' = TI-" R j = R[,,+R] = R+T", and with a plane-wave
form of the phase, cPi = wt - q . R i , leads to Fourier transformed exchange
constants
J1-''' (q) -- - '~
" JI-'[,,+R] eiq'(T!'-Tv-R) . (1. 79)
R
1 Low-Lying Magnetic Excitations 31
The subscripts i, j, and likewise f.J" v, are still meant as multi-indices, the
above introduced vectors aud phases are only site dependent but not type
dependent. The rescaled complex amplitudes
(1.80)
(1.81 )
with f.J, = 1,2, ... ,Ne, t = 1,2, ... ,Tm, Tm and Ne being the number of dif-
ferent moment um variablesand the number of sites per unit cell, respectively
(T means transposed). With these definitions, the linearized aud complexified
equations of motion are finally cast into the form
(1.83)
In the cases of Fe aud Ni, where experimental parts of the spectra are available
from the literature, the agreement is excellent in the long-wavelength region
and good even for shorter wavelengths, when not considering the interference
of au additional excitation brauch around 150 meV in the case of Ni, which
will be accounted for later in this section by use of the extended set of the spin
variables. The 150-meV branch in Ni has been earlier theoretically derived
both within a many-body treatment of the dynamic transverse magnetic
susceptibility [26] and within the self-consistent SDFT calculations [58] men-
tioned above, and has been interpreted as an efIect of the Stoner continuum
excitation, though called au "optical magnon" brauch in the transitive sense.
Note that the results shown in Figs.1.8-1.10 do not contain auy adjustable
parameter.
32 S. Halilov
400
~ 200
CI)
t::
I.il JOO
r H N r P NO 1
Fig. 1.8. Adiabatic magnon dispersion relations on high-symmetry lines and
magnon densities of states (in states/(meV*cell)xlO- 2 ) of bcc-Fe. Solid circles
mark calculated frequencies, lines are guides for the eye. For comparison, experi-
mental data, for pure Fe: 0 [33], crossed circles [37], and for Fe(12%Si) at room
temperature: 0 [34] are given. Note the Kohn anomalies in the adiabatic spectrum.
Note also that wave vector resolution in [37] obtained by a time-of-flight technique
is very poor at higher energies
As an example with a true optical branch (at least in the adiabatic treat-
ment), we have also calculated the magnon spectrum of hcp-Co, which is
shown in Fig.1.9b. Both the maximum frequency and the stiffness are not
much different for fcc- and hcp-Co. However, the neglect of anisotropy terms
in the adiabatic Hamiltonian is more problematic in the hcp case, where the
true spectrum has a gap at zero frequency.
The oscillating character of the exchange interaction in metals, includ-
ing its long-range part, is most easily done in the reciprocal space as the
direct SDFT calculation of the q-dependent exchange constant matrix via
constrained spin-spirals.
By analogy to phonon spectra, where in metals long-range interactions
mediated by Friedel oscillations due to charge density perturbations with
Fermi surface nesting vectors lead to Kohn [36] anomalies in the spectra, such
anomalies at the same q-positions and with the same cause are expected in
the magnon spectra.
The calculated adiabatic magnon spectrum of Fe, displayed in Fig. 1.8,
shows prominent anomalies around point Hand weaker ones in both r - N
and r-
P directions. In Fig. 1.11 we present Fermi surface cross sections with
symmetry planes for Fe, which indicate particularly strong nesting behavior of
the majority and minority spin Fermi surface pockets around r. Specifically,
the plane part perpendicular to the r-H direction can explain the strong
anomalies in the magnon dispersion around H both in r - Hand H - N
directions. A magnon, as a kind of many-electron excitation, is formed by
electron transitions between opposite spin projected states, which is related to
the spin-moment flip in the entire system. Therefore, specific nesting vectors
1 Low-Lying Magnetie Excitations 33
600
~ 500
5 400
~ 300
[)
c: 200
u.l
100
(a) X WK L WUXO
600
>
~
500
§ 400
:>.. 300
g.o
<I.) 200
&S 100
0
(b) r M K r A L H AO 1 2
Fig.1.9. As Fig. 1.8, but for (a) fee Co, (b) hep Co
r x W K r L W U XO
Fig.1.lO. As Fig.1.8, but for fee Ni. The experimental data (0) are those of (35)
q for the spin-wave softening must be sought among those that connect the
Fermi surface fragments with the opposite spins. The most essential softening
of the spin-wave modes occurs at q ~ 2/37r /2a along r-H type directions.
The study of the magnetic susceptibility in [26] indicates large windows
free of spin-flip continua for Fe even at higher excitation energies, but lacking
theoretical predictions, no particular experimental search for Kohn anomalies
has been undertaken up to now, although several investigations of the high-
energy magnetic excitation spectrum were reported [33,37] . Although the
statistics are obviously not sufficient, the experimental points along the line
34 S. Halilov
p
r k----+--a-;-----l~ H
p p
Fig.1.11. Cross sections of the Fermi surface of Fe: solid lines: up spin, dashed
lines: down spin. Vectors ql, q2 stand for the wave vectors of a considerable nest-
ing between opposite-spin r-centered Fermi surfaces, which particularly leads to
aremarkable softening of the spin-wave modes
r- P of Fig. 1.8 are not unlikely of showing the anomaly. More precise
measurements in this area are strongly encouraged by our results.
In Fig.1.12 the dependence of J ij , which is obtained from the reciprocal
space by the Fourier transform as
• bcc Fe
• --->X 10
•
• •
...
• • •
•••••
•
· '"
-I --->X 10
•
2 3 4
RJalattice
500 R=1a
. ..... 500 65R=2aatoms
15 atoms
.' ..'·
...
0 0
500 R=3a 500 R=4a
.....
.......·
259 atoms 537 atoms
:> .
.·..'.'
' .'
...'.'
Q)
250 250
....·
E_
8
~ 0 .. .'·
' 0
500 R=5a • 500
whole space •
........ ...........·
1067 atoms
250
.'.' 250
0 ..·· 0
.
..'.·
r H r H
Fig. 1.13. Magnon spectrum of Fe as a function of the cluster size: the number of
neighbors is simulated by simply making the appropriate choice of the maximum
lattice vector R in the Fourier transform of the exchange constants
variables, i.e. regarding the vibrations along the local equilibrium moments
as a superposition of the transverse precession. Presumably, nickel is the best
transition metal candidate to fit the situation, where the fluctuations of the
interstitital part of the spin density manifest themselves as rather one more
spin-wave-type excitation branch instead of a continuum. To get to the point,
one has to admit that the choice of the adiabatic moment made above was too
36 S. Halilov
coarse and by averaging out the fast electronic degrees, we have accidentally
smeared those fluctuations that belong to the spin-wave scale. If we adopt the
Hamiltonian (1.57), then the standard eigenstate problem yields two modes,
whose eigenvectors are constrained by the normalization condition (1.88) to
the Plank distribution n qm and describe the thermal amplitudes of the spin
vibrations.
Interestingly, these results are very similar to that of the LSDA-SDFT
frozen-magnon calculations with the ASA-charge and spin densities and the
extended variable's set, exhibited in Fig.1.14, and are very evocative of
the dynamical spin susceptibility (Fig. 1.1) obtained nonadiabatically [58],
at least regardless of the lifetime effects. We emphasize that the picture has
been found to be very stable against the arbitrarily chosen variables, obtained
by various division of space into domains with the subsequent integration
of the spin-density over the domains and establishing the exchange-force
constants between them, with the arbitrariness available at least within the
ASA method.
In fact, even in the case where there is the uniform exchange force J(r-
r/) = J(I r - r ' I) within a volume Vi, where the spin density is integrated
to build a moment Tni, any further subdivision of the Tni, i.e. the extension
of the variable set, would not change the wave modes, contributing just to
the Stoner continuum. Note also that since the cross section of the inelastic
neutron scattering on a given mode is proportional to the ImX, i.e. to the
moment squared of the mode, the observable mode may not be associated
with those moments whose amplitudes are immaterial compared to the others.
In that sense, the association of the moments with the areas of the positive
spin polarization, primarily of d-character, and negative spin polarization,
mainly of highly itinerant sp-character, would not be sufficient to reproduce
the observed double-branch picture in nickel, since the net sp-moment is
.. ...
hm r - - - - - - - - - - - ,
meV Ni ,
extended
300 basis set
200
100
o .
r x
Fig. 1.14. Magnon spectrum of Ni derived within the frozen-magnon LSDA-SDFT
approach with the extended set of the adiabatic variables, T = O. The slight
degenracy removal of the optic branch is due to the nonequivalent spheres used
in the ASA calculation
1 Low-Lying Magnetic Excitations 37
one order smaller than that of d-character and the corresponding excitations
would not survive thermodynamically.
This is why we would rather be inclined in favor of the symmetry-related
decomposition of the spin density caused by the crystalline field effects, and
consider this way as the most motivated for the extension of the variable
set in transition metals. In nickel, for example, the total per unit cell spin
moment is 1/5 of eg character, whereas in iron eg and t2g states contribute
equally [24 J. The spin density, because of the different angular dependence of
eg and t2g, is also spatially resolved, what lead us to consider two domains Ve g
and "Vt 29 of integrations instead of a single one Vi enclosed the whole atom.
It is clear that, in the limiting case of the most extended possible set, where
each moment is associated with the spin density at each point of the given
spatial mesh, the frozen-magnon scheme becomes the formally exact spin
susceptibilty method. In particular, the Stoner excitations may be viewed
as the ill-defined transverse collective mode vibrations. It is remarkable that
the similar multiple-sphere calculations for iron revealed no new wave modes.
Apparently, the points outlined here briefly in connection with the extended
class of the vibration modes, are not weIl established and require a furt her
analysis of the intracell exchange-correlation hole structure within the full-
potential-based methods.
o o
- 2 -2
, MK ,A LH AO 2 MK ,A LH AO 2
Ho
2
o
-2
MK ,A LH AO 2
Er Tm
, MK ,A LH AO 2 , MK ,A L H AO 2
Fig. 1.15. Exchange-split conduction bands [eV] (left panel) and densities of states
[st/atom/eV] (right panel) ofthe heavy RE metals, calculated for the ferromagnetic
configuration. Hybridization due to spin-orbit coupling between spin-majority (solid
line) and spin-minority (dashed line) bands is neglected
with vss,(r) mentioned before does not playa role here due to the dominance
of the f spin density. The ()- and q-dependent total energies and conduction
electron site spin moments are shown in Fig.1.16. First, it is seen that the
conduction electron site spin moments are nearly independent of () and q. The
f spin moments are strictly independent of () and q by construction. Hence,
the low-Iying spin excitations are expected to be nearly perfectly transversal,
particularly at low magnon occupation numbers, that is, at small ().
Next, one observes that for alt q the energy behaves elose to the Heisen-
berg form ~J(q)S2 sin 2 (). Hence, (1.69) is justified with the J-values fitted
to these results, even in the cases of Ho through Tm where the true ground
state is either a cone with q = qo and 0 = 00 or even more complex.
The Fourier-transformed exchange constants J(q) for the heavy RE metals
together with values extracted from experimental magnon spectra are plot ted
in Fig. 1.17. The theoretical values are directly obtained in reciprocal space
for each q along the line r- r
A - in the double zone, as the second derivatives
of the scalar relativistic total energy with respect to conical deviations from
a collinear state. The lowest energy state (maximum J(q)) corresponds to
40 S. Halilov
....::: 0.3
0.2
Gd Dy
S 0 .2
0 - q=0.5
0.1
S er-&-<> q= 0.8 0. 1
:::
.s. 0.1
r/)
1. 5
...-
3
0.1
;>,
1.0
~ 2
0.0
;;;.
e!l
~
0.5
::: - 0. 1
~
0
0 20 40 60 80 0 20 40 60 80
Angle, (Deg) Angle, (Deg) Angle, (Deg)
Fig. 1.16. Conduction electron site spin moment s(O, q) and total energy E(O, q) of
cone configurations as functions of the cone angle 0 (deviation from aglobaI c-axis)
and of the modulation vector q = (0,0, q), given in units of [7r ja], for heavy RE
metals. The dashed lines give fits to the Heisenberg expression ~J(q)s2 sin 2 0
".
~ Gd
5 /
- - - Tb
.. .. . Dy / \
I \.
- - Ho \
_ .. - Er \
._- T m
g
...... -5
10 Theory Experiment
A A
Dy Ho
A A
Er Tm
A A
Fig.1.18. The cross sections of the minority (dashed lines) and majority (solid
lines) Fermi surfaces of the heavy REM on the symmetry planes. The calculation
is performed for collinear spin configurations. Note the increasing proximity effect
between the opposite-spin Fermi surface sheets on going from Gd through Tm,
which explains the enhancement of the exchange constant J( Q) at the ground
state modulation vector Q through the REMs
directed dose to the hexagonal axis. The most prominent dips correspond to
R rv Q-l rv 1.6a, which are mostly profi ted from the maximum J(Q). This
oscillating out-of-plane behavior is opposed by a rather smoothly decaying
in-plane exchange, which reflects the fact that the moments throughout the
hexagonal plane prefer to be aligned uniformly.
To proceed, the intraatomic f-d exchange constants land the spin-orbit
inter action constants Aare straightforwardly calculated from the expressions
(1.68) and (1.64), respectively. The results are shown in Table 1.2. While the
values of land the one-electron spin-orbit constants >. only weakly depend
1 Low-Lying Magnetic Excitations 43
Tb
Ho
Tm
1 2 3 2 3
Distance (Iatt. par.)
Fig. 1.19. Exchange force constants of the heavy REMs J(R - R / ) as a function
of distance between R v and R v ' sites. Full circles: all out-of-basal-plane neighbors,
trianges: neighbors within the basal plane. Friedel-like oscillations are clearly notice-
able only along the hexagonal (out-of-plane) axis. This is obtained as the Fourier
trans form of J(q) derived in the vicinity of the constrained uniform state
Table 1.2. Calculated exchange and spin-orbit parameters: Spin splitting of the
f-levels .:10: as being imposed by spin-polarized valence states only, s - Sexchange
constant I, and multiplet spin-orbit coupling parameter A4f
.:10:, eV I, eV A4 f, eV
Gd 0.062 0.17 -0.03
Tb 0.055 0.17 -0.04
Dy 0.047 0.16 - 0.05
Ho 0.037 0.16 -0.07
Er 0.026 0.15 -0.11
Tm 0.014 0.14 -0.18
on the RE element, the increase of A within the series reflects the rescaling
(1.65) .
Since Gd has a spherical f-shell, the crystal-field parameters vanish in this
case. As already stated, the crystal-field parameters of the other RE metals
could either be taken from experiment or - with limited accuracy - calculated
44 S. Halilov
Fig. 1.20. The magnon spectrum of Gd: Full lines - fuH calculation, dashed lines
- rigid s - S coupling (I = (0) assumed, dots - experiment [78]
1 Low-Lying Magnetic Excitations 45
energies of a Bravais lattice (or along the c-axis of the hexagonal double zone)
can easily be shown to be
nWq = 2 1{ J(q)s [
+ (s + 8)1 - (J(q)s + l(s + 8)) 2- 4IJ(q)s2 ]1/2} ,
(1.85)
if there is no effect of spin-orbit coupling. The limit I -+ 00 yields
S2
nWq = J(q)-S .
s+
(1.86)
Clearly, the effect of softening the f-d exchange (I < 00) vanishes for W -+ 0,
and reduces the frequencies in the higher region, where the f-spin 8 does
not follow the fast motion of the conduction electron site spin moments any
longer.
At least part of the persisting discrepancy between the experimental and
the calculated magnon spectra, in particular along the hexagonal axis r -
A, should be attributed to the neglect of intersite spin correlations in the
adiabatic approach, which is less justified for the more extended conduction
states of the REMs as compared to the 3d-states of Fe, Co, and Ni [57] .
Figures 1.21 and 1.22 display the magnon spectra of Tb and Dy above
their basal-plane ferromagnetic ground state. The opening of the energy gap
is due to the easy-plane anisotropy. The experimental data for Dy also contain
some traces of magnon-phonon hybridization.
The ground state of Ho is a shallow c-axis cone with qo >:::; (0,0,0.1[71' la]) ,
the equilibrium angle of which, on account of the CF parameters [63], appears
15
.... :
:>
0
10
.§.
;;...
e!l
0
!:: 5
~
Fig. 1.21. The magnon spectrum of Tb: Lines - calculation, dots - experiment [79]
46 S. Halilov
10
:>
<U
8
;>, 5
e!l
<U
I'::
~
0
r K M r A L H A
Fig.1.22. The magnon spectrum of Dy: Lines - calculation, dots - experiment
[80J. Note that the experimental magnon spectrum is distorted by the interaction
with phonons
O~ ____L - - L_ _ _ _L - - L_ _ _ _L - - L_ _ _ _ ~
r K M r A L H A
O+--------------+--------------~
r A r
Fig. 1.23. The magnon spectrum of Ho: Upper panel - calculated spectrum for
the conical magnetic ground state. Lower panel - for the ferromagnetic state sta-
bilized by an external magnetic field in the hexagonal plane; the calculated and
experimental [82J dispersion along the (OOq)-direction in the double-zone is shown
4,----,--,------.-,,----,--,----,
:>
! 2
M K r A L H A
4
..
..
:>
! 2 .:.r .. ....
>.
e.o
<U
&j
A r
Fig. 1.24. The magnon spectrum of Er: Upper panel - calculated spectrum for the
conical magnetic ground state. Lower panel - the calculated and experimental [83]
dispersion along the (OOq)-direction in the double-zone is shown
48 S. Halilov
>'
!
,.,
~
"""
MK r A L H A
Fig. 1.25. Theoretical magnon spectrum of Tm, ferromagnetic phase
putting the mode energy equal to hwqnnqn(T) in accordance with the Planck
distribution. For small deviations from the equilibrium, the moment vibra-
tions are normalized as
2 N L MJl- (()Zqn) = nqn , (1.88)
gJ.tBohr Jl-
and the thermal reduction of the average sublattice magnetization is ob-
tained as
L1M; = MJl- L[l- (COS()Jl-qn)] ~ ~Jl- L(()Zqn). (1.89)
qn qn
Angular brackets denote thermal averaging in this section, which comprises
averaging over the slow magnetic dynamies. Summation of (1.89) over J.t and
consideration of (1.88) as weIl as (l/N) L: q = (Vu.c./(21T)3) J d3q yields
J,VIl-
o
= _jvll-
q=O
+J
Vll-
Vu .c .
(27r)3
r
JBZ
d3
q
JVIl- -iq'(Tv-TI')
q e . (1.93)
For temperatures for which (1.92) has a nonzero solution the material is
magnetically ordered and the thermal average of the total magnetization
(moment per unit ceIl) is now obtained by summation of (1.92) over 1/. This
approach assurnes that on the fast timescale the site moment Mil- is practically
temperature independent up to Tc. The Langevin formula is used because
again this site moment is already a quantum average and hence may vary
continuously.
In Fig. 1.26 the magnetization of 3d ferromagnets vs. temperature T cal-
culated according to (1.90) and (1.92) for T "-' Tc/2 and Tc/2 "-' T ::; Tc,
respectively, is compared with experimental values. In particular, the very
satisfactory agreement between our calculation and experiment at low tem-
peratures is due to our correct use of the Planck distribution of magnons
instead of the semiclassical averaging of [17]. This indicates that the low-
temperature excitation spectrum is indeed weIl represented by magnons.
The mean-field treatment for higher temperatures is only semiquantitatively
correct.
Within this mean-field treatment, the Curie temperature itself is obtained
as the boundary of solubility of (1.92). This is the temperature at which the
Weiss field vanishes, and therefore (1.92) may be replaced by the lowest order
of an expansion in powers of (MvBWeiss/kBT) for T ~ Tc:
(1.94)
The solubility condition ofthis homogeneous linear equation system for (M) is
•
det ( Uvp, M; vp,)_ (1.95)
- 3kB T c Ja - 0,
T. _ M 2 Jo (1.96)
c - 3kB
for the dassical Langevin situation.
The Curie temperatures obtained within the adiabatic method together
with experimental values and with other theoretical estimates from the lit-
erature are given in Table 1.4. Probably the first SDFT calculation of Curie
temperatures of 3d-met als on the basis of (1.96) was reported in [43]. The
deviation between the calculated Curie temperatures of [41,43] and [57] is
most probably due to differences in the calculation of Ja, which was obtained
as either a sum over a finite number of neighbors in real space [41,43] or
a complete Fourier space treatment [57]. Arecent improvement [16] of the
Korringa-Kohn-Rostoker coherent-potential-approximation approach [15] to
the paramagnetic phase above Tc, which is based on similar adiabaticity
assumptions, resulted in Tc values very dose to that obtained by the frozen-
magnon method.
We stress once more that due to our analysis, in particular the results of
Fig.1.5 in connection with the magnon energy scale related to Tc, the site
moments remain fully developed on the fast timescale even above the Curie
temperature. This relates nicely to very recent experimental findings [44] of
strong short-range ferromagnetic correlations in magnetic neutron scattering
results for fcc-Fe-Ni alloys at high temperatures far above ordering temper-
atures and even for concentrations (rv 20% Ni) where there is practically no
low-temperature magnetic order. On average, over the slow timescale there
is, of course, no moment.
Here is another problem that is in the limelight and might be worth consid-
ering in the near future: determination of the excitation spectra of magnetic
systems with a strong nesting between the opposite-spin sheets of the Fermi
surface in the nonlinear limit. On this route, one could face a new domain-wall
type collective excitation. Iron can serve as a suitable prototype for the system
in question: the equation of motion keeps its validity for arbitrary deviations
from equilibrium, since basic variables in iron, at least in accordance with
the SDFT calculations, keep their integrity practically in the whole phase
space, i.e. are almost perfect adiabatic variables. The presence of the strong
Fermi surface nesting may cause the unbounded oscillatory solutions with low
energies different from the ordinary linear spin waves. The latter would most
likely, by enhancing the temperature, gradually be replaced by the former,
since the amplitude of the moment precession would also adequately increase
dragging the vibrations into the nonlinear regime. These softened soliton-type
excitations could be of great importance for thermodynamics and transport
phenomena of other systems, particularly of the high Tc superconductors,
where the strong spin fluctuations are believed to play an essential role in the
mechanism of the very phenomenon of the superconductivity (see e.g. [90]).
Acknowledgements
The results presented in this chapter are essentially a product of the author's
cooperation with A. Perlov and H. Eschrig during his stay at Max-Planck
Institute, Dresden. The author is thankful to everyone who was involved in
mutual discussions of the problems raised in the chapter.
References
1. T. Kasuya: Prog. Theor. Phys. 16, 58 (1956)
2. A. Fert and LA. Campbell: Phys. Rev. Lett. 21, 1190 (1968)
3. G. Breit: Phys. Rev. 34 553 (1929); Phys. Rev. 36 383 (1930); Phys. Rev. 39
616 (1932)
4. H.A. Bethe and E.E. Salpeter: Quantum Mechanics o/One- and Two-Electron
Atoms (Aeademic Press Ine, New York 1957) p. 181
5. A.K Rajagopal and M. Moehena: Phys. Rev. B 57 11582 (1998)
6. T. Holstein and H. Primakoff: Phys. Rev. 58, 1098 (1940)
7. S.V. Tyablikov: Methods in the Quantum Theory 0/ Magnetism (Plenum Press,
New York 1967)
8. V.L. Moruzzi and P. Mareus, in Handbook 0/ Magnetic Materials, ed. by KH.J.
Busehow (Elsevier Scienee Publishers, Amsterdam 1993) p.97
9. M.S.S. Brooks and B. Johansson, in Handbook 0/ Magnetic Materials, ed. by
KH.J. Busehow (Elsevier Scienee Publishers, Amsterdam 1993) p.139
10. J.-H. Cho and M. Seheffier: Phys. Rev. B 53, 10685 (1996)
11. L.M. Sandratskii: Phys. Stat. Sol. B 136, 167 (1986)
12. J. Kübler, K-H. Höek, J. Sticht, and A.R. Williams: J. Phys. F 18, 469 (1988)
13. T. Oguehi, K Terakura, and M. Hamada: J. Phys. F 13, 145 (1983)
54 S. Halilov
14. A.J. Pindor, J. Staunton, G.M. Stocks, and H. Winter: J. Phys. F 13, 979
(1983)
15. B.L. Gyorffy, A.J. Pindor, J.B. Staunton, G.M. Stocks, and H. Winter: J. Phys.
F 15, 1337 (1985)
16. J.B. Staunton and B.L. Gyorffy: Phys. Rev. Lett. 69, 371 (1992)
17. M. Uhl and J. Kübler: Phys. Rev. Lett. 77, 334 (1996)
18. N.M. Rosengaard and B. Johansson: Phys. Rev. B 55, 14975 (1997)
19. V.P. Antropov, M.1. Katsnelson, M. van Schilfgaarde, and B.N. Harmon: Phys.
Rev. Lett. 75, 729 (1995)
20. V.P. Antropov, M.1. Katsnelson, B.N. Harmon, M. van Schilfgaarde, and D.
Kuznecov: Phys. Rev. B 54, 1019 (1996)
21. S. Akbar, Y. Kakehashi, and N. Kimura: J. Appl. Phys. 81, 3862 (1997)
22. T. Moriya: Spin Fluctuations in Itinerant Electron Magnetism (Springer,
Berlin Heidelberg New York 1985)
23. H. Capellmann: Metallic Magnetism (Springer, Berlin Heidelberg New York
1987)
24. J.F. Cooke, J.W. Lynn, and H.L. Davis: Phys. Rev. B 21,4118 (1980)
25. J. Callaway, A.K. Chatterjee, S.P. Singhal, and A. Ziegler: Phys. Rev. B 28,
3818 (1983)
26. J.F. Cooke, J.A. Blackman, and T. Morgan: Phys. Rev. Lett. 54, 718 (1985)
27. J.A. Blackman, T. Morgan, and J.F. Cooke: Phys. Rev. Lett. 55, 2814 (1985)
28. E. Stenzel and H. Winter: J. Phys. F 16, 1789 (1986)
29. L.M. Sandratskii: J. Phys. Cond. Matter 3, 8565 (1991); W. Brinkman and
R.J. Elliott: Proc. Roy. Soc. A 294, 343 (1966)
30. U. von Barth and L. Hedin: J. Phys. C 5, 1629 (1972)
31. J. Hubbard: Phys. Rev. B 19, 2626 (1979); ibid. 20, 4584 (1979); J. Appl.
Phys. 52, 1654 (1981)
32. M.V. You, V. Heine, A.J. Holden, and P.J. Lin-Chung: Phys. Rev. Lett. 44,
1282 (1980)
33. C.-K. Loong, J.M. Carpenter, J.W. Lynn, RA. Robinson, and H.A. Mook: J.
Appl. Phys. 55, 1895 (1984)
34. J.W. Lynn: Phys. Rev. B 11, 2624 (1975)
35. H.A. Mook and D. McK. Paul: Phys. Rev. Lett. 54, 227 (1985)
36. W. Kohn: Phys. Rev. Lett. 2, 393 (1959)
37. T.G. Perring, A.T. Boothroyd, D. McK. Paul, A.D. Taylor, R Osborn, RJ.
Newport, J.A. Blackman, and H.A. Mook: J. Appl. Phys. 69, 6219 (1991); M.
Yethiraj, R.A. Robinson, D.S. Sivia, J.W. Lynn, and H.A. Mook: Phys. Rev.
B 43,2565 (1991); A.T. Boothroyd, T.G. Perring, A.D. Taylor, D. McK. Paul,
and H.A. Mook: J. Magn. Magn. Mater. 104-107,713 (1992)
38. J. Crangle and G.M. Goodman: Proc. Roy. Soc. A 321, 477 (1971)
39. R. Pauthenet: J. Appl. Phys. 53, 2029 (1982)
40. P. Weiss and R Forrer: Ann. Phys. 5, 153 (1926)
41. R.F. Sabiryanov, S.K. Bose, and O.N. Mryasov: Phys. Rev. B 51, 8958 (1995)
42. N.W. Ashcroft and N.D. Mermin: Solid State Physics, (Holt, Rinehart and
Winston, 1976)
43. A.1. Liechtenstein, M.1. Katsnelson, V.P. Antropov, and V.A. Gubanov: J.
Magn. Magn. Mater. 67,65 (1987)
44. M. Acet, E.F. Wassermann, K. Andersen, A. Murani, and O. Schärpff: Euro-
phys. Lett. 40, 93 (1997)
1 Low-Lying Magnetic Excitations 55
81. R.A. Cowley and J. Jensen: J. Phys. Condens. Matter 4, 9673 (1992)
82. C.C. Larsen, J. Jensen, and A.R. Mackintosh: Phys. Rev. Lett. 59, 712 (1987)
83. R.M. Nicklow, N. Wakabayashi, M.K. Wilkinson, and R.E. Reed: Phys. Rev.
Lett. 27, 334 (1971)
84. H. Bjerrum Mooller, J.C. Gylden Houmann, and A.R. Mackintosh: Phys. Rev.
Lett. 19, 312 (1967)
85. J. Kuriplach and P. Novak, in Physics of Transition Metals, ed. by P.M. Op-
peneer and J. Kübler (World Scientific, Singapore 1993) p.609
86. J. Kübler: Theory of Itinerant Electron Magnetism (Oxford Science Publica-
tions 2000)
87. R.V. Chamberlin: Nature 408, 337 (2000)
88. T.L. Hill: Thermodynamics of Small Systems (Parts land 11) (Dover, New
York 1994)
89. M. Pajda, J. Kudrnovsky, I. Turek, V. Drehal, and P. Bruno: Phys. Rev. Lett.
85, 5424 (2000)
90. 1.1. Mazin and D.J. Singh: Phys. Rev. Lett. 79, 733 (1997)
2 Calculation
of Magneto-crystalline Anisotropy
in Transition Metals
2.1 Introduction
It is hard to underestimate the importance of magnetic anisotropy. Without
this property we would not be able to generate electricity efficiently or use
electricity to do work. The basic premise of an electro-motor is that the
currents in a coil generate a magnetic moment that pushes against an external
magnetic field. Without magnetic anisotropy this force would simply result
in a change in direction of the external field, and no rotation of the coil would
occur. The shape of the yoke, however, fixes the direction of the external field
due to magnetic anisotropy.
The shape of a material is not the only cause of magnetic anisotropy. The
non-isotropie environment of atoms in a crystal is the source of magneto-
crystalline anisotropy. This type of anisotropy governs the thickness of do-
main walls. Without this magneto-crystalline anisotropy domain structures
would be very different, which in turn would change the way we use magnets
in practical applications. Many basic text-books discuss magnetism in solids.
Some examples are the books by Culllity [1], Bozorth [2], Chikazumi [3],
and Morrish [4]. An example of theoretical books are Mattis [5] and White
[6]. Density functional theory is also described in various places. One of
my favorites is still the early work edited by Lundqvist and March [7]. An
excellent book is by DreizIer and Gross [8]. A fun mathematieal description is
published by Lieb [9]. Quantum field theory is explained in books like Mandl
and Shaw [10], Gasiorowiez [11], Gross [12], and Ryder [13]. Of course, there
are many more books on all these topics, the list given here is only meant as
an example.
The earliest explanation of the origin of magneto-crystalline anisotropy
is probably due to Mahanji [14] in 1929. Van Vleck in his review(!) in 1937
already pointed at the importance of spin-orbit coupling [15]. The first at-
tempt of calculating the magneto-crystalline anisotropy energy was by Brooks
[16] in 1940. The early calculations were obviously very unprecise, but did
point out the relation between changes in the band structure as a results
of changes in the direction of the applied magnetic field. The importance
of band crossings and anti-crossings was recognized early [17,18], but even
today a complete understanding of the role of k-space is stilllacking. Although
many researchers have published data for the basic transition metals, in this
58 H.J.F. Jansen et al.
chapter we will only give some examples from the last ten years. Due to
limitations in computer speed and disk space older data are in general not
weIl converged.
In this chapter we give an overview of the current status of the calculations
of the magneto-crystalline anisotropy energy. We discuss what we need to
calculate, what the numerical problems are, and what possible explanations
are for the difference between theory and experiment. We use the simple
transition metals nickel, iron, and cobalt as illustrations. Our goal is to pro-
vide a theoretical basis for the understanding of numerical results in general.
Armed with the material discussed in this chapter we hope that one will
be able to judge the quality of results published in the literature. It is not
the intent of this chapter to provide an extensive overview of all results for
transition metals, pure or alloys.
The outline of this chapter is as follows. In Bect. 2.2 we discuss magneto-
crystalline anisotropy in the context of its experimental definitions. There are
three ways of obtaining this quantity and we discuss the relations between
them. In Beet. 2.3 we show some results from model calculations to give more
clarity to the discussion in Beet. 2.2. These two sections set the stage for
the theoretical approach in Beet. 2.4, where we derive expressions for the
magneto-crystalline anisotropy energy from quantum field theory. In Beet. 2.5
we discuss the numerical problems related to the evaluation of the magneto-
crystalline anisotropy energy. These problems are due to the integration in
the Brillouin zone and the need for high precision in these calculations. Once
we have established weIl converged results, we see a difference between theory
and experiment, and in Beet. 2.6 we discuss the possible cause of these differ-
ences and the physics related to this. Because the calculation of the magneto-
crystalline anisotropy energy is very time-consuming, it would be useful to
have a better understanding of the numerical problems and in Bect.2.7 we
analyze the integration over the Brillouin zone in further detail. Finally, we
draw some conclusions in Bect. 2.8.
The words magnetic anisotropy simply mean that the properties of a certain
material depend on the direetion of the applied magnetic field. If a sam pIe has
the shape of a disk, there is a clear distinetion between applying an external
field in the plane of the disk or perpendicular to the plane of the disko If the
material in the sampIe is ferromagnetic, then there is a net magnetization even
at zero applied field, and we expect that the direction of this magnetization
is related to the shape of the sam pIe. This is indeed true, and as we all
know the magnetization of a bar magnet is along the axis of the bar, while
the magnetization of a disk shaped magnet is in the plane of the disko This
shape anisotropy can easily be described if the magnetization in the sam pIe
2 Calculation of Magneto-crystalline Anisotropy 59
is uniform [1], and after solving Maxwell's equations one finds indeed that
the situations described above correspond to the lowest energy states.
Even in a perfectly spherical sampie we still find magnetic anisotropy. One
reason could be that the material is not uniform and that there are variations
in density or other materials properties. This kind of anisotropy plays an
important role in practical materials, where there are always stress and strain.
Annealing will typically reduce this anisotropy. But even in a perfect material
there is anisotropy due to the atomic structure. In iron, for example, the
atoms are arranged in a bcc lattice and we expect a difference whether the
magnetization is parallel to the edge of the cubes or to the body diagonal.
This effect is called magneto-crystalline anisotropy and is the topic of our
paper.
Magneto-crystalline anisotropy plays an important role in the physics of
magnetic materials. For example, the thickness of domain walls depends on
the strength of the magneto-crystalline anisotropy. Also, in layered mate-
rials it is possible that magneto-crystalline anisotropy forces the preferred
direction of the magnetization to be perpendicular to the layers. If the layers
are then combined into a disk, it is possible that the preferred direction
of the magnetization is perpendicular to the disk if the magneto-crystalline
anisotropy is stronger than the shape anisotropy. Such materials could be
very important for magnetic recording and detection, and the recent interest
in the theory of magneto-crystalline anisotropy is driven by the search for
these materials with so-called perpendicular anisotropy.
It is important to understand how magneto-crystalline anisotropy is de-
fined experimentally, before we can try to describe it theoretically. There are
three main approaches, which under the right conditions are all equivalent.
It is possible to define magneto-crystalline anisotropy using
• Magnetization curves
• Torque measurements
• Free energy differences
and we shall discuss all three.
In Fig. 2.1 we show an idealized picture of magnetization curves in a cubic
material with an applied field along the <100> direction and along the <111>
direction. Iron is an example of a material that has similar magnetization
curves. Along the easy axis, for iron the <100> direction, the magnetization
reaches a saturated value already for a very small applied field. If the field
is applied along the hard axis, <111> for iron, the magnetization reaches
about half the maximal value very rapidly, but only reaches the saturated
value when the field is stronger than a value H s , called technical saturation.
There are two important approximations made in constructing this ide-
alized figure. First of all, the top line, representing the magnetization along
the easy axis, is supposed to be horizontal. This is not correct, this line has
a very small slope. The effect of this slope is, however, not important for all
practical applications. We do know, though, that theoretically at an infinite
60 H.J.F. Jansen et al.
1.2
1.1
easy
IAR EA = MAEI
1.0
"t:I 0.9
äi
:;::: 0.8
Ol
c: 0.7
0
(ij 0.6
'E
Q)
0.5
E 0.4
0
::E 0.3
0.2
0.1 Hsal
0.00
10 12 14
External field
Fig.2.1. Idealized magnetizatiün curves für a cubic material
applied field all electron spins will point in the same direction, giving rise
to an enormous magnetization. In order for this effect to be noticeable, the
difference in energy between spin up and spin down electrons has to be larger
than the bandwidth, or larger than a few eV. This corresponds to external
fields on the order of millions of Teslas, a bit higher than we can produce in
our laboratories.
The second approximation is that at H s the magnetization in the easy
and in the hard direction are the same. This is also not true, the magnetiza-
tion along the hard axis is slightly smaller. For all practical materials these
effects can be ignored, but they are important in order to arrive at a correct
theoretical description.
The magneto-crystalline anisotropy energy is now easily defined, since the
product of magnetization and applied field corresponds to thermodynamic
work. The gray area in Fig. 2.1 corresponds to this energy and we have
This definition has been used for many years. It is now easy to study the
magneto-crystalline anisotropy as a function of temperature, for example.
Also, this definition fits in very weIl with all standard notions of thermody-
namics. A small aside is that in thermodynamics we also have to include the
energy stored in the magnetic field itself. This gives a small correction, as
long as we have treated the shape anisotropy right. To eliminate the shape
anisotropy one prefers to use a disk-like sampie, cut to have all three main
directions, <100>, <011>, and <111>, in the plane of the disko This is
possible if the axis of the disk is along the <oll> direction.
In a real experiment the value of H s is not always easy to find precisely and
therefore a theorist might change this definition of the magneto-crystalline
2 Calculation of Magneto-crystalline Anisotropy 61
and define
E MCA = lim L1E(H).
H--toc
This is wrong, however, because the magnetization curves are not exactly
identical. This definition does give a result, but now the result does not
depend on temperature anymore. When the field becomes infinitely large,
the field energy dominates any thermal energy contribution, and is the only
important quantity in the problem. If we study the curve L1E(H) we see
that for fields less than H s the curve rises rapidly, but that for fields larger
than H s the curve is almost Hat, until it reaches a point where the field is
strong enough that lower core states are mixed in. Again, this happens at
unreasonably large fields, and a practical description is to take the value of
L1E(H) for H within a factor of two above H s • In general, the difference
between L1E(Hs ) and L1E(2Hs ) is much less than the experimental error
bars.
The definition of magneto-crystalline anisotropy using torque measure-
ments is based on the fact that
fM.dH=O,
EMCA = l::: Y
M· dH IH=H s
In general, we can write a formula that gives the energy for a rotation over
an arbitrary angle () from the easy axis, and the corresponding energy has
the form
which dearly gives rise to a torque. The functional form of E(()) turns out to
be very simple, a few sines are sufficient to describe this relation. The reason
that higher order terms become small very rapidly is related to the fact that
the spin-orbit coupling is a small parameter, on the order of the fine structure
constant, and that terms in order n scale with this coupling to the power n.
Several experimental setups to use this formula to measure the magneto-
crystalline anisotropy exist. It is important to note, though, that the results
are the same to the first definition only if the process is reversible. Mag-
netization curves can easily be measured in a reversible manner, although
62 H.J.F. Jansen et al.
M=_8G
8H'
and hence
E[{Si}] = E o - L JijSi . Sj .
i<j
Once we have determined the values of the exchange constants J ij we then
can use standard statistical mechanics to derive quantities like the critical
temperature. This procedure has shown very promising results when applied
to bcc iron [20].
Van Vleck already showed how to extend this model to include higher
order interactions [15]. He studied energy terms of the form
0.8 L
c I
Q)
§ 0.6 !
1
~
0.4 I
;
I
0.2
i
o
o 2 3 4
H-external
Fig_ 2.2. Magnetization curves for a simple model Hamiltonian with J = 1, T = 1,
and K = 1
1
0.9
................ ,,, ..
0.8 'r'T"
0.7
r
0.6 1
C
Q)
E 0.5
Iv
0
~ 0.4
I; I •-M~aSY I
) M ·hard
0.3
i
0.2
0.1
0
o 0.2 0.4 0.6 0.8 1.2
H-external
Fig.2.3. Magnetization curves for a simple model Hamiltonian with J = 1, T =
0.25, and K = 0.05
0.04
0.03 r .. a _ • •
>.
e>
0.02 l
Q)
c ~~,o
.. .
0.01
.: tj"•.•..
Q)
~
• ,... -1'*. ••
,. , . ... ...••
, ...
., .. .....
~
- ... ..
0
o. 0
'
-0.01
-0.02
o 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
H-external
Fig. 2.4. Magneto-crystalline anisotropy energy as a function of applied field for
the parameter values used in Fig. 2.3
derived from the torque only depends on the component of the magnetization
perpendicular to the applied field. This component is small, and not easy to
obtain with sufficient precision.
1l(r) =: 'l/Jt(r) [~c a· V + ßmc2 - e</J(r) + ea· A(r)] 'I/J(r) : +1l rad(r).
The electromagnetic fields in this Hamiltonian density are separated into two
parts. The internal fields describing the interaction between the electrons
are quantized and included in the radiation part in quantized form, but the
external fields due to the nuclei are classical fields and are omitted in the
radiation part.
The charge density of the electrons in a given quantum state follows from
e(r) = -e <: 'l/Jt(r)'I/J(r) :>,
and the current density from
j(r) = -e <: 'l/Jt(r)a'I/J(r) :>
We now define a universal functional of the charge and current density by
minimizing the total energy of the electron system over all possible quantum
2 Calculation of Magneto-crystalline Anisotropy 67
where the minimization is constrained in the sense that it only uses states
with the prescribed charge and current density. Note that normal ordering
is essential, else the minimum would not exist. Without normal ordering the
positrons would give negative energy states, which used to be a problem in
interpretation in the old days. The universal functional F represents the
energy of system of interacting electrons with the prescribed charge and
current density when no external fields are present.
The total energy in the presence of the external fields is
and the charge density, the current density, and the total energy of the system
of electrons are found by minimizing this expression as a functional of e and
j. Of course, one of the constraints on the charge density is that the total
charge of the electron fields is equal to the total charge of the nucleL This is
not a constraint on the total number of electrons, as long as excess electrons
are created in equal amounts as positrons. With the electron energies involved
in solids, however, this pair creation is not a likely event, perhaps with the
exception of the small region near a heavy nucleus.
The universal functional F is obviously impossible to evaluate, and has
to be approximated. The main observation needed in such an approximation
is that this universal functional consists of two parts, kinetic energy and
potential energy. Therefore, we define a kinetic energy functional by
This is simply the relativistic analogue of the kinetic energy defined in stan-
dard density functional theory. It includes a term mc2 per particle, but this
term can and should be subtracted in the formaliE!m before doing any cal-
culations. The equation of motion that has to be solved in order to obtain
this kinetic energy is now the relativistic Dirac equation for non-interacting
particles in stead of the standard Scrödinger equation. This introduces the
relativistic effects into the equation of motion, and hence these effects can
be called dynamical effects. One can expand the effects in terms of the fine-
structure constant and in lowest order terms like mass-velocity and Darwin
appear. Also, in this order we introduce spin-orbit coupling. The so-called
semi-relativistic approximation [29] includes alilowest order terms except the
spin-orbit coupling. For a description of magnetic anisotropy the latter term
is essential, though, and it is better to work with the complete relativistic
expression.
68 H.J.F. Jansen et al.
where we have a summation over repeated Greek indices as usual. The inter-
action kernel W is given by
The first part of this kernel corresponds to the standard Coulomb inter action
as used in all applications. The second part is called the Breit [38,39] interac-
tion and is the lowest order relativistic correction to the Coulomb interaction.
It represents the relativistic effects in the electron-electron interaction.
In order to approximate the Coulomb inter action we use the Hartree form
and for the first term this simply gives the usual
~
2
Jd Jd 3
r
3 ,g(r)g(r')
r Ir _ r'l .
11 3 3' 1
d rd r 21r _ r'1 3
() . (') _ 3 [m(r) . (r' - r)] [m(r') . (r' - r)]]
X [m r m r (r-r
')2
-3
47r J d 3rlm(rW,
j(r) = cV X m,
and is the sum of the magnetic moments related to the spin and orbital
angular momentum density [34]. Using this form of the current we find that
the interaction with the external fields can be rewritten in the form
-J d3rBexternal(r) . m(r),
which suggest that m is the magnetization density indeed. One can also show
that approximately
2 Calculation of Magneto-crystalline Anisotropy 69
which is indeed the sum of orbital and spin angular momentum. The factor
in front is the Bohr-magneton, which is related to the fine structure constant,
and in Rydberg atomic units it has the value 5.2 x 10- 3 . Therefore, magnetic
effects are inherently small, which leads to important issues of numerical
precision in the calculations.
The Hartree form of the Breit interaction leads to an important physical
effect. If we replace the magnetization density by its average value, the in-
tegral is conditionally convergent, and leads to the energy related to shape
anisotropy. Therefore, shape anisotropy is included in density functional the-
ory, as long as we. do it right! The standard calculations that leave out the
Breit interaction will need to include shape anisotropy after the fact, which
is normally done.
In density functional theory the universal functional :F is written in the
form
where the first term represents the kinetic energy of a system on non-inter-
acting electrons with the same charge and current density as the interacting
system and the second term is the Hartree approximation to the electron-
electron interaction. The last term, which makes the equation an identity, is
(incorrectly) called exchange-correlation energy [9].
The dynamical relativistic correction in the non-interacting kinetic energy
functionallead to the inclusion of spin-orbit coupling in the equations for the
single-particle equations, and this is the driving force for magneto-crystalline
anisotropy. Relativistic corrections in the Coulomb interaction lead to shape
anisotropy and dipole-dipole interactions. These dipole-dipole interactions
do give a contribution to the magneto-crystalline anisotropy [42], but the
effects are generally small by three orders of magnitude [15]. The exchange-
correlation energy also has to be corrected for relativistic effects. This has
been done, but turns out to be less important for our problem. In the standard
local density approximation for the exchange-correlation energy there are still
large errors related to the orbital motion of the electrons. The relativistic
corrections are smaller than the errors which are present in the local density
approximation, and these errors have to be minimized before we can address
the effects of relativistic corrections to the exchange-correlation energy. We
address this point further in Sect. 2.6.
The electronic structure of atoms is well described by Hund's rules [5,6].
The first rule states that the spin moment of a partially filled shell is maximal.
This is an atomic many-body effect. A uniform electron gas can also be spin-
polarized, though, and these are the spin-polarization effects that are included
in the local density approximation, which is based on results for the homoge-
nous electron gas. Electronic structure calculations using the local density
approximation give very good results for itinerant magnets like nickel and
iron. Therefore the exchange-correlation effects present in the homogeneous
70 H.J.F. Jansen et al.
electron gas are already sufficient to explain the spin moment in itinerant
metals.
Hund's second rule focusses on orbital moments and states that the orbital
moment of a partially filled sheH is maximal, within the constraint of Hund's
first rule. This is a purely atomic effect, since a homogeneous electron gas
has zero angular momentum, by definition. Any finite rotation of an infinite
homogeneous gas would always lead to infinite angular momentum. This is
again one of those cases where taking the thermodynamic limit early leads
to special consequences. If we want to discuss rotations in the homogeneous
electron gas we should first perform the calculations for a finite rotating
system and then take the limit to an infinite system. In itinerant magnets,
however, the orbital moment is always quenched, and the local density ap-
proximation does in general give a good description. We will discuss later,
however, that for the problem of magneto-crystalline anisotropy this might
not be sufficient.
Conventional wisdom states that the local density approximation works
weIl because materials like iron and nickel are itinerant, but it is actually
better to turn this around. Magnetism is called itinerant if a local density
calculation describes the material weH, or in other words if the exchange-
correlation effects present in the homogeneous electron gas are sufficient to
describe the magnetic properties. This implies that the orbital moment has
to be quenched. Therefore, magnetism in rare earth materials and in atoms
is localized, since the local density approximation fails to describe the spin
and orbital moment correctly.
The term in the Hamiltonian density describing the interaction with an
external magnetic field is found be combining the expressions for the inter-
action and the approximate form of the magnetization density. We get
If we survey the literature, we see many places where people use the same
term, but with an extra factor ß, which in effect changes the sign of the lower
components. Both equations seem to be derived correctly, and one wonders
why there is a difference. The explanation is simple [12]. In field theory, the
components in the four-vector 'ljJ refer to the following spin states, starting
at the top: up-down-down-up. The operator for the spin angular momen-
turn is ~1'5'Y. This is the formulation used here. In a theoretical framework
closer to the older first-quantization formulation is is customary to order
the components as: up-down-up-down. This seems more natural. In this case
the spin operator is ~1'5a and in the Dirac equation the magnetic term
acquires an additional four matrix ß. Hence it is perfectly legitimate to use
2 Calculation of Magneto-crystalline Anisotropy 71
This is the expression used in non-collinear calculations [43]. The driving force
for non-collinear magnetic structures in this case is the spin-orbit coupling
in the equation of motion, and hence is a dynamical effect. Calculations
along these lines are definitely more tedious, but have produced interesting
results [44].
The disadvantage of the formulation above is that we have no direct
control over the direction of the spin and total magnetization. This makes cal-
culations of the magneto-crystalline anisotropy energy very difficult, because
we now need to introduce a external saturation field in the system. This
leads to all kinds of complications, theoretical and numerical. An applied
external field can only be defined for a finite sampie, but calculations are
often done for bulk solids using periodic boundary conditions. What is the
screened applied field in a unit cell in such a calculation? Again, taking the
thermodynamic limit early causes problems. But even if we could define this
screened field, we still have trouble with the numerical precision, since the
field energies involved are on the order of micro-Rydbergs. These numbers
are much smaller than the values used in the exchange-correlation field, and
this approach is probably not feasible due to the numerical problems caused
by this imbalance.
All calculations of the magneto-crystalline anisotropy in density func-
tional theory have taken a more practical approach and use the following
72 H.J.F. Jansen et al.
Table 2.1. Magneto-crystalline anisotropy energy in lieV per atom from various
calculations. The experimental results are included in the last row
calculations give results doser to the experimental values, but it turns out
that again the numerical convergence is not sufficient.
The next to last entry in Table 2.1 is very interesting. The work reported
by Beiden uses a real-space method, and therefore does not need Brillouin
zone integrals. Beiden's results are reproduced in Fig.2.5 as a function of
inverse distance to the third power. These results still have large error bars,
but they are clearly consistent with the results of Daalderop. Using a real-
space method is a very attractive idea for a problem where the convergence in
k-space is very slow. Unfortunately, the convergence in real space is also slow,
this time due to the presence of a Fermi surface in k-space. This discontinuity
in k-space gives Friedel type oscillations in real space [54], and the amplitude
decays with the inverse third power of the cluster size only. This is why we
presented Beiden's data in the form chosen in Fig.2.5.
The conclusion is that at the moment the best theoretical results for the
magneto-crystalline anisotropy energy for iron and nickel are -0.5 ± 0.5j.leV
for both. Hence the result for iron is too small by a factor of three, and
for nickel it is too small by a factor of five and has the wrong sign (wrong
easy axis). In the next section we will discuss the significance of this results
in terms of physics, in this section we will give details ab out the numerical
problems in k-space.
Our analysis of the k-space integrations is based on the results for a model
tight binding Hamiltonian [55,56]. The basis states include s, p, and d states
with spin up and down, leading to a matrix of size 18 by 18. The model
parameters in this Hamiltonian are obtained from fits to ab initio band
structure data. The difference between the results for the model Hamiltonian
2 Calculation of Magneto-crystalline Anisotropy 75
•
-0.2
>CI>
-0.4 •
••
ll;l -0.6
e
g
(.)
-0.8
>- -1.0
~
•
CI>
c: -1.2
w
<I:
ü -1.4
•
::2:
-1.6
-1.8
0.0 0.2 0.4 0.6 0.8 1.0
Inverse volume cluster
Fig. 2.5. Convergence of the magneto-crystalline anisotropy energy with the size
of the real-space cluster (adapted from [52])
and for the "real" system are equivalent to errors introduced when the density
functional calculations are not fully self-consistent. Although the overall shifts
in energy values are large, the effect is mainly systematic and the values for
the magneto-crystalline anisotropy we obtain should be elose to the "real"
results. This is exactly what we find.
The spin-orbit interaction is included in our model Hamiltonian by assum-
ing that all spins point in a direction Mo as described before. The magnitude
of the spin-orbit coupling parameter is a free parameter, and we use the
values reported in the literature [47] for most of our calculations. Finally,
the magneto-crystalline anisotropy energy is obtained by using the force
theorem [57,27]:
In this section we report the results for the two most commonly used
procedure to perform the Brillouin zone integrations, first the linear tetrahe-
dron method and next the Gaussian broadening method. Other methods are
discussed in [21,22].
F = ( d 3 kj(k)8(E F - c(k)),
JBZ
where j(k) is either 1 or c(k) and the corresponding interpretation of F
is N or U. The main problem with this integral is the discontinuity at the
76 H.J.F. Jansen et al.
Fermi energy EF. The linear tetrahedron method was developed to deal with
this singularity in the easiest manner [58]. Space can always be divided into
tetrahedrons, and inside each tetrahedron it is always possible to approximate
the energy e(k) bya linear function. Once we know the values ofthe energies
at the corner points of the tetrahedron, we have sufficient information to
construct the linear interpolate. It is easy to see that this linear interpolate is
described by four parameters (one value and three for the gradient) and the
tetrahedron has four corners. Inside each tetrahedron the linear interpolate
is integrated analytically, and the contribution of this tetrahedron to the
integral is approximated by this analytical value of the integral of the linear
interpolate.
The error analysis of the tetrahedron method is straightforward. If the
average length of the edge of each tetrahedron is ..1, the error in the procedure
sketched above is proportional to ..1 2 • The main contribution to this error is
due to the Fermi surface. For completely filled bands errors cancel in lowest
order, due to the opposite effects of regions of positive and negative curvature.
It is possible to estimate the effects of the lowest order error term, and the
improved tetrahedron method takes this into account [59]. The resulting error
in the improved tetrahedron method is proportional to ..1 4 •
In Fig. 2.6 we show the result of the application of the linear tetrahedron
method to the total energy of fcc copper. The precise answer is subtracted,
and the plot represents the numerical error in the calculations. The solid dots
show indeed that the error is proportional to ..1 2 • The improved tetrahedron
method gets much closer to the precise answer faster, as expected. Both meth-
ods contain similar information, however. This is clearly seen by comparing
the deviations from a linear fit to the tetrahedron method results (crosses)
to the improved tetrahedron method results. This situation is completely
analogous to the comparison of the trapezoidal rule, Simpson rule, and the
Romberg extrapolation in numerical integration.
One never performs one calculation for one k-mesh only, since it is never
known apriori what mesh needs to be used. The best procedure is to use two
different mesh sizes and calculate both the results for the linear tetrahedron
method and for the improved tetrahedron method. These four data points
together will then give a reliable extrapolation to the real result. Also, if the
results of the regular tetrahedron method and the improved method do not
agree one knows that a larger set of k-points has to be used. Finally note
that the number of k-points involved in the irreducible zone is around one
hundred.
In Fig. 2.7 we show a similar plot for the magneto-crystalline anisotropy
energy. There is some indication that the results for the linear tetrahedron
method follow a straight line for small tetrahedrons. The results for the
improved tetrahedron method are more or less constant. The curves seem
to extrapolate to a similar value. This value, however, is close to zero, and
relative errors are enormous. Also on the top of the graph we indicate the
2 Calculation of Magneto-erystalline Anisotropy 77
irredueible k-points
364 120 5647 35 29 20 16 10
0.025
195 84
0.020
•
0.015 •
>-
CI:
-.
>- 0.Q10
~
Q)
cQ) •
(ij
0.005
:§
0
c!>
+
+
0
-0.005
0 10 20 30 40 50 60 70
Fig.2.6. Comparison of the regular linear tetrahedron method (filled circles) and
the improved tetrahedron method (open eireles) using the total energy of fee cop per
as an example. The energy is plotted against the square of the eharacteristic distanee
between k-points with the k-point distanee of a 643 mesh set to 1. Corresponding
numbers of irredueible k-points are indicated along the top axis. A least square fit
to the values for ..1 2 :::; 10 is indieated by a straight line, eorresponding residual
errors by crosses. The eonverged value for the energy is shifted to zero
number of divisions of the edges of the Brillouin zone, and the last data point
corresponds to 2403 points in the total Brillouin zone. Since the presenee of
spin-orbit coupling breaks the symmetry of the cubic crystal, the number of
points at which the calculations has to be performed is somewhat smaller,
but not by a factor of 48 as expected in a cubic crystal. Because we need to
perform the calculations with magnetization along the <100> and <111>
direction there is a reduction of a factor of two only. Hence the calculational
effort is tremendous.
At first the qualitative difference between the results presented in Figs. 2.6
and 2.7 seems incomprehensible. But there is an easy clue to understand the
origin of the problem. The magneto crystalline anisotropy energy is very
small, and we need numerical convergence to a value which is about four
orders of magnitude smaller than the value for regular total energy calcula-
tions. This requires small tetrahedrons, and the basic premises of the error
analysis in the linear tetrahedron method have to be reevaluated. We need
to investigate the convergence in more detail, and in Fig. 2.8 we present data
78 H.J.F. Jansen et al.
k-poinl. I reciprocal lattice veclor
1 240 120 80 60 40
160
o ._. ___.~~
•
-1 o~ •
E
~
0
(;j
:>GI -2
::I.
W
<I
-3
o 4 8 12 16
k-point spacing lJ.2
Fig. 2.7. Magneto-erystalline anisotropy energy LlE = E(OOl) - E(111) for fee Ni
as a function of the square of the eharaeteristic distanee Ll. between two k points
along a reciproeallattice veetor, with the distanee for the 1603 mesh set to 1. The
spin-orbit eoupling strength is ASO = 100meV. The number of divisions of the re-
ciproeallattice veetors is indieated at the top of the graph. Open circles denote LlE
ealculated with the regular tetrahedron method, closed circles are results ealeulated
with the improved method. Lines are guides to the eye
similar to those in Fig. 2.7, but now we have divided the edges ofthe Brillouin
zone in all possible integer values up to 160.
A striking feature of the results in Fig. 2.8 is the presence of oscillations
in the data. Less noticeable is a change in slope of the envelope for small
tetrahedrons. These results can be explained easily. The oscillations are due
to the presence of points in the band structure where the gradient is zero.
These are the points that give rise to van Hove singularities in the density of
states. Guenter Schneider [21,22] has shown that such points give oscillations
in the calculated quantities on a J.!eV scale in all calculations. This effect is
related to the difference between such a singular point being in the center of
a tetrahedron or being on the edge of a tetrahedron. In standard calculations
with convergence on the meV level this is not essential, but on the extreme
convergence level we need here it shows.
In the linear tetrahedron method points in the Brillouin zone where bands
are (anti-)crossing give a large error to the numerical integration. When the
tetrahedrons are large, the effect of a band crossing is limited to one tetrahe-
dron, and does not playa role in the error analysis. The region in the Brillouin
zone where the effect of a band crossing is important is not a singular point,
however, but a finite region. This region has a two-dimensional character in
our problem, since the width along the gradient of the energies is very small.
Once the tetrahedrons become smaller than this finite region the number of
2 Calculation of Magneto-crystalline Anisotropy 79
0,.'. . . .
k-points I reciprocallattice vector
240 120 80 70 60 50 40
2~~r---,,-.---.------~-----------r---,
160
•
0
4iIt ••• • •• • •
.~ 1' ... " •• • •• • • • • • •
·1 • •
E ep><:po
0 ·2 Olo'" 0 00 0 0
6> 0
~ 00 0 0 0 0
0 0
°00 0 0
Q) 0
:::t ·3 0 0 0
000
W
<l -4 0 0
0 0
0
0
·5
-6
0 6 8 10 12 14 16 18
k-point spacing 1'12
tetrahedrons with errors grows like .:1- 2 and the net effect is an extra error
term proportional to .:1 2 , which represents the change of slope of the envelope
in Fig. 2.8. Also, if we push the convergence even further, the region involved
becomes three dimensional, and the dominant error term is proportional to
.:1 itself. This is beyond out observations, however.
In summary, the linear tetrahedron method works as expected, as long
as we do a complete analysis of all possible error terms. Unfortunately, the
oscillating error due to the points with zero gradient is difficult to extrapolate,
since we do not have a good mathematical model for this, and since the
effects of several of these points interfere with each other. The only solutions
is to go to large numbers of k-points. It is also clear that results based on
calculations using a few divisions only can easily give wrong results. For
example, extrapolating data based on two open circles only in Fig.2.8 gives
a range of values one hundred times the real answer. This easily explains the
difference in published data.
Energy bands are nice periodic functions. The integrations we need to per-
form are over a periodic cell, and in such cases the extended trapezoidal rule
converges very rapidly, if the integrant is wen behaved [60]. That is a problem
in OUf case, however, since the integrant contains a step function, which gives
80 H.J.F. Jansen et al.
rise to singularities in the lowest order derivatives. Therefore, one would like
to approximate the step function by a smooth function. In that case the
numerical integral is best obtained by a simple sum of the contributions at
uniformly spaced points in the Brillouin zone.
There are several ways to approximate the step function by a smooth
function. If the width in which the function changes from zero to one is
denoted by ()", one could use a spline interpolation between zero and one in
that range. That is a fast procedure, which shifts the singularities to higher
order derivatives. One can also use analytic functions without singularities.
One example is a Fermi function, another is a hyperbolic tangent (related to
a Lorentzian shaped peak). The most commonly used smooth interpolation
in electronic structure calculations is a Gaussian broadening of the delta
function peaks, which gives after integration
The complementary error function Er Je has limits zero at infinity and two at
minus infinity. The value at x = 0 is one, which means that the approximated
!
step function always has the value at the position of the step.
The advantage of the Gaussian broadening method is that the error term
as a function of ()" can easily be derived using a Sommerfeld expansion [61].
The lowest order error is proportional to ()"2, which is in complete analogy
with the T 2 behavior of the total energy in the presence of a Fermi function.
Only even order terms survive.
The Gaussian broadening method has been applied in many cases. In
general, many quantities are rat her insensitive to the choice of broadening
[62]. We give some typical results in Fig. 2.9. The total band energy of nickel
is calculated as a function of ()"2. Of course, when ()" becomes smaller we need
more points in the Brillouin zone, but the number of points needed for the
calculations presented in Fig. 2.9 are comparable to those used in the linear
tetrahedron method. It is clear from the results in Fig. 2.9 that one has to
be careful in using the Gaussian broadening data, and that the higher order
terms are important even on a scale of a few meV. On the other hand, once
we are careful it is easy to extrapolate results with the precision which is
needed in standard calculations.
In Fig. 2.10 we present the results for the magneto crystalline anisotropy
energy of nickel. Three sets of results are shown, all for rat her large numbers
of k-points in the Brillouin zone. When the broadening is large, the three
curves overlap, which is an indication of the rapid convergence of the Brillouin
zone integrations. When the broadening becomes smaller, more integration
points are needed. The curvature seems much larger, though, and it is clear
that we need to use a smaller value of the broadening parameter ()". This
is, of course, completely expected. Relevant features in the changes in the
band structure are of order >'so and hence we expect that we need a value
of ()" which is much smaller than >'so. This is shown in Fig. 2.10. When the
2 Calculation of Magneto-erystalline Anisotropy 81
Fermi surface smearino 0/ meV
38 58 77 9tl 115 135 154 173
63_355 ,...:..r....:.;..--,--'-i----r----T'----------r------,--~r_:I'.,
63.350
63.345
E &3_340
~
>CI> &3.335
~
CI>
63.330
c
CI>
63.325
&3.320
0 10000 30000 40000
--- ---
0.0
\
. --
-0.1
\
-8..
E ' 10
.9 ·0.2 j ~
'"
>CI>
::l
:::::- - -- 160'
w -0.3
---0- 120]
<I
- 0- 80'
-0_'
0 10000 20000 30000
Fig. 2.10. Magneto-erystalline anisotropy energy i1E = E(OOI) - E(ll1) for fee
Ni (>'SO = lOOmeV) as a function of the square of the Gaussian width used for
Fermi surface smearing. The number of k-points used in the fuH Brillouin zone is
indicated. Lines are guides to the eye
broadening is larger than the spin-orbit coupling strength, the curve is almost
linear. A large change is seen when the broadening is similar to the spin-
orbit coupling strength. It is also clear that the individual curves start to
deviate sharply when the broadening is too small for the mesh in k-space. It
82 H.J.F. Jansen et al.
--.---
Fermi surface smearing 0/ meV
/ .
0 j. •
/
·5
E
-
~
>(I) · 10
/
-
::l
~.
............--/ I
w
<l ·15
·ro ~~~~~~~~~-i~~~~~-L~~~~
0.00 0.04 0.08 0.12 0.16 O.ro 0.24 0.28 0.32 0.36 0.40
Fig.2.11. t1E for fee Ni (ASO = 340meV) as a function of the square of the
Gaussian width used for Fermi surfaee smearing. 40 3 k-points were used in the fuH
Brillouin zone and aH ealculated points are eonverged. Lines are guides to the eye
is tempting to conelude from Fig. 2.10 that the value for nickel is -0.15 f.leV.
We do have confidence in this result, because it is consistent with the result
obtained by the linear tetrahedron method.
The relation between the values of the broadening parameter and the spin
orbit coupling strength can be furt her investigated by artificially increasing
the latter. Figure 2.11 shows the results. There is a very elear difference in
behavior when the broadening is large or small. It is also elear that we need
a value of the broadening about a factor of four smaller than the spin orbit
coupling strength in order to be get elose to the converged value. We also
need calculations for several values of the broadening parameter in order to
be able to extrapolate reliably, because of the oscillations in the data. The
higher order terms do not disappear rapidly.
In Fig. 2.12 we present the results for iron. The spin orbit coupling strength
is somewhat smaller, and as a result the extrapolated result has a somewhat
larger uncertainty. A value of - 0.75 f.le V seems the optimal when reconciling
these results with the data from the linear tetrahedron method.
Both the linear tetrahedron method and the Gaussian broadening method
need large numbers of integration points. In general, there is no advantage
of one method over the other. There are, however, some disadvantages of the
Gaussian method. In Fig. 2.13 we show the results for the magneto crystalline
anisotropy energy as a function of band filling.
The pronounced oscillations near N = 9.5 and N = 10.5 are related to
band crossings at the Fermi surface. The reference data for the improved
tetrahedron method are weIl converged. The Gaussian broadening method,
2 Calculation of Magneto-erystalline Anisotropy 83
Fermi surtace smearina (} I meV
48 96 '35 173
0.0
-1).2
-1).4
~~/~
E
0 -1).8
1iJ - 160'
> - A- 120'
'" ·1.0
-
::t
-0- 80'
".2 -'Q- 60]
W
<1
--<>- 40]
" .4
·1.6
0 10000 20000 JOOOO
10'9 29 38 48 58 67
00n-~--.-----r-----,-------~--------~--,
·0.2 ~ - e- '60'
::.~:\
-6- 120'
-0- 80l
-'Q- 60 3
40 3
._e~~O~~
-0-
-1).8
,',2/
·'.0 -~--=:::::I======:;;;=li==1
W
<1
·1.4
Fig.2.12. L1E for bee Fe (.Aso = 60 meV) as a funetion of the square of the
Gaussian width used for Fermi surfaee smearing. The number of k-points used in
the fuH Brillouin zone is indicated. Lines are guides to the eye
however, has a hard time reproducing these peaks. In Fig. 2.14 we show the
results for a very small value of the broadening parameter and a k-mesh
which is too smalL This combination of parameters leads to large oscillations
as a function of band filling, and hence to large errors. Note that (J = 10 meV
is sufficiently small to describe the parts of the curve with large contributions
from the states elose to the Fermi surface, but, even with Nd = 160 the results
are not converged around q = 10.
1.4
a(meV)
1.2 - _.- 135
96
1.0 48
29
0.8 -<>- (ITM) 24 0'
0.6
E 0.4
~
-..
> 0.2
Q)
::1. .,
0.0 .
:. ~
w
<I -0.2
·0.4
-0.6
-0.8
-1 .0
1.0
0.8
0.6
E 0.-
~ 0.2
>Q) 0 .0
-
::1.
-0.2
W -0.-
<l
-0.6
-0.8
·1.0
9.4 10.6
Fig. 2.14. Magneto-erystalline anisotropy energy LlE for fee Ni (ASO = 100 meV)
as a function of the number of valenee eleetrons as ealculated with special points
and a Gaussian Fermi surfaee smearing using 1603 k-points with (J" = 10 meV (open
squares) and with (J" = 10 meV (solid line) . The results for 240 3 k-points using the
improved tetrahedron method are shown as referenee (open circles)
1.5
1.0
0.5
0.0
-0.5
E
0
-1 .0
1il
>Q) -1.5
::I..
-2.0
W
<l
-2.5
·3 .0
·3.5
The next question is which physical effects can change the electronic
structure globally. In the context of Density Functional Theory this means
a change in the exchange-correlation energy. The standard approximation
used in the calculations of the magneto-crystalline anisotropy energy is the
local spin density approximation. In this approximation the exchange-correla-
tion potential is spin dependent, but there is no dependency on the orbital
moment. An infinite homogeneous electron gas has zero orbital momentum,
because any rotation needs infinite energy.
The correction term in Density Functional Theory that describes many
body effects related to the orbital momentum has the form [34]:
similar to the coupling of the exchange field to the spin direction. The form of
this effective field coupling to the orbital momentum is not known, however,
2 Calculation of Magneto-crystalline Anisotropy 87
Homogeneous XC
Spin Moment
L-S Coupling
MAE
B eff = aL,
locally on each atom, where the pre-factor a is calculated using the results
of atomic physics for open shells. In this formulation the pre-factor is de-
termined by the usual Slater integrals. This approximation is called orbital
polarization [63].
Orbital polarization should play an important role. In the calculations
that are based on the local spin density approximation we start with a model
for exchange-correlation efIects that give a good description of the spin mo-
ment in itinerant magnets. In fact, as we mentioned before, one can argue
that the definition of itinerant magnetism should be a system where the local
spin density approximation gives the right results. In other words, all many
body efIects needed to describe an itinerant magnet are already present in
the homogeneous electron gas. If we now include spin-orbit coupling in the
calculations, we create both an orbital moment and anisotropy because the
orbital moment is coupled to the crystallattice. This is shown in Fig. 2.16.
Orbital polarization includes Hund's second rule into Density Functional
Theory. Atoms with incompletely filled shells have a non-zero orbital moment.
The hybridization with neighboring atoms reduces this orbital moment, but
there is still a small amount present. This small orbital moment is large
88 H.J.F. Jansen et al.
Homogeneous XC Atomic XC
MAE
Fig. 2.17. With orbital polarization, the orbital moment is already present at the
start and the magneto-crystalline anisotropy is caused by the interaction of the
existing spin and orbital moments
Table 2.2. Magneto-crystalline anisotropy energy calculated without (no OP) and
with (with OP) orbital polarization [64]. Energy values are in IJ-eV /atom
a)
E
~
3>Ql
:::t
LJ.J
<l
10000
1000
100
E
~ 10
3>Ql
:::t
0.1
0.01
10 100 1000
spin-orbit coupling \0 ( meV)
Fig. 2.18a,b. L1E = E(OOl) - E(111) for bee Fe as a function of spin-orbit eoupling
strength. The number of eleetrons/atom is n = 8. Results shown are eomputed with
80 3 points in the fuH Brillouin zone using a Gaussian broadening of 96 meV
a)
j
/
60
40
i
:>Gl
20
-40
0 100 200 300 400 500 600 700 600 900 1000
spin-orbit coupling Aso ( meV)
1000
100
b)
- ').4.
..
10
E
.9
-!!!
>Gl
:::l.
ur
& 0.1
0.01
100 1000
spin-orbit coupling A", ( meV)
Fig. 2.19a,b. flE = E(OOl) - E(111) for fee Ni as a function of spin-orbit eoupling
strength. The number of eleetrons/atom is n = 10. Results shown are eomputed
with 1203 points in the fuH Brillouin zone using a Gaussian broadening of 96 meV
eubie symmetry of the crystal structure show that the energy should vary with
the fourth power of the spin-orbit coupling strength, and our results show
that this is true_ If we compare the theoretical value and the experimental
value of the magneto-crystalline anisotropy energy in iron, we find a ratio of
3.6 between these values. Hence increasing the spin-orbit coupling strength
by a factor of 1.4 will reconcile theory and experiment. A change of 40%
in the spin-orbit coupling parameter due to many body effects is certainly
reasonable and compares very weH with other data for iron.
In Fig. 2.19 we present the results for nickel. Again we observe the fourth
order dependence on spin-orbit coupling strength. The ratio of the magnitude
of the experimental and theoretical results is 5.4 and therefore an increase
of the spin-orbit coupling strength by a factor of 1.52 would give similar
magnitudes. It is weH known that many body effects in nickel are larger
2 Calculation of Magneto-crystalline Anisotropy 91
than in iron (Fermi surface and photo emission data) , and hence a larger
percentage increase is expected and 52% is a very reasonable number. The
problem, of course, is that the sign is still wrong. We need to increase the
spin-orbit coupling strength by a factor of around eight in order to obtain
the correct easy axis. This is unreasonable and something is not right.
What are other possible mechanisms that could influence the results? One
term that is neglected in the interactions is the coupling between the spin
and orbital angular moment um of electrons in different orbits. This could play
a possible role, but chemistry calculations seem to indicate that the effect is
too small. Another mechanism is the presence of spin-waves in the ground
state. It has been speculated that this is more important in nickel than in
iron, and hence could be important in the discussion of magneto-crystalline
anisotropy.
2e-06
-2e-06
through the Fermi energy. Therefore, one expects to see a clear indication in
the band structure of what is happening. It turns out, though, that there is
no visible change in electron eigenvalues. It is not possible to relate a visible
change in band structure to a change in magneto-crystalline anisotropy en-
ergy. This is quite surprising and we do not yet have a good explanation for
this phenomenon.
An observation that has been made many times is that bands near the
Fermi surface should play the most important role. Therefore we integrate
the changes in the energy eigenvalues band by band. In Fig.2.21 we show
the contribution of bands 11 and 12, the highest partially occupied bands
in nickel, to the magneto-crystalline anisotropy energy as a function of the
change in the parameter xy-d2-110-down. The values were obtained using a
603 mesh in the Brillouin zone, which in this case gave converged results.
The energy values in this plot are a factor one hundred larger that the
total changes in the corresponding plot in Fig.2.20. There is also a large
cancellation of the contributions of these bands, which is not surprising since
one is mostly spin up, while the other is mostly spin down.
In Fig. 2.22 we show the combined contributions of bands 11 and 12, and of
bands 9 and 10. Again, the shapes of the two curves are similar but opposite,
leading to a large cancellation. One might argue that this is not surprising,
since all these bands are partially occupied and hence go through the Fermi
surface. In Fig. 2.23 we show the combined contributian of bands 5 through
12 (all partially occupied) and of bands 1 through 4 (all fully occupied). The
basic shape of the final result is clearly due to the bands that cross the Fermi
surface, but the overall shift due to the fully occupied bands is large (similar
2 Calculation of Magneto-crystalline Anisotropy 93
0.0004 r--~-r-~-,.-~-.-~-.-~-.-----,
0.0002
-0.0002
Fig.2.21. Change in LlE in Rydberg for fee Ni('\so = 100 meV) as a function of
the change in parameter xy-d2-110-down using Gaussian Fermi surface smearing
with a = 49 meV and division 60 along a reciprocal lattice vector. The diamonds
indicate the contribution from band 12, the filled circles represent the contribution
from band 11, and the stars describe the sum of those two bands
00001
Fig.2.22. Change in LlE in Rydberg for fee Ni(,\so = 100 meV) as a function of
the change in parameter xy-d2-110-down using Gaussian Fermi surface smearing
with a = 49 meV and division 60 along a reciprocal lattice vector. The diamonds
are the sum of band 9 and band 10, the filled circles are the sum of band 11 and
band 12, and the stars are the sum of all four bands
Se·06
·5e-06 '---.l.-~--'-----'-~--'--~--'------'
-0.04 -0.035 -0.03 -0.025 -0.02 -0.015 -0.0 1
Fig.2.23. Change in L1E in Rydberg for fee Ni(.\so = 100 meV) as a function of
the change in parameter xy-d2-110-down using Gaussian Fermi surface smearing
with u = 49meV and division 60 along a reciprocallattice vector. The diamonds
are the sum of band 1 through band 4, the filled circles are the sum of band 5 to
band 12, and the stars give the final contribution of all the bands
le-05 r-------.----....---r------.------,
Se·06
)"
",/.:
o
the following procedure. For all k-points we sum the contributions from
all bands. Also, we calculate the minimal distance between the bands and
the Fermi energy. This value is negative if the band nearest to the Fermi
level is unoccupied. In Fig.2.24 we show the contribution to the magneto-
crystalline anisotropy energy of the k-points as a function of the minimal
2 Calculation of Magneto-crystalline Anisotropy 95
distance. Although points that are far from the Fermi level tend to have the
smallest contributions, it is clear that the majority of k-points has a large
contribution, and that the final value of the magneto-crystalline anisotropy
energy is obtained by adding these large but opposite values. This is why the
numerical integrations are so difficult.
It is possible to analyze the role of k-space furt her. Suppose the contribu-
tion to the change in magneto-crystalline anisotropy energy at a point k for
a parameter value pis c(k,p) and the integrated energy is E(p). These two
contributions can be normalized using
and
1~ r--------.------~--------~------_.
1000
500
o 0.5
2e-OS
l e-OS
:. " .
Fig. 2.26. The total absolute energy change for as a function of the correlation
index corresponding to Fig. 2.24
the interpretation of the histograms is more difficult [66], and there are peaks
at highly (anti-)correlated values. In all cases, however, the contributions
of many k-points are not correlated with the final answer of the magneto-
crystalline anisotropy energy as a function of parameter change.
The previous analysis omits the absolute value of the contribution. AI-
though the results presented in Fig. 2.24 already give a good indication that
all points play a role, we show in Fig. 2.26 the absolute values of the con-
tribut ions of the k-points to the magneto-crystalline anisotropy energy as
a function of the correlation. It is clear that large contributions occur every-
where, and that all points in the Brillouin zone have to be included.
One has to keep in mind that the results presented in this section are for
the cubic materials iron and nickel, where symmetry reduces the value of the
magneto-crystalline anisotropy energy greatly. Changes in energy eigenvalues
are proportional to the spin-orbit coupling strength near band crossings and
proportional to the square of the spin-orbit coupling strength everywhere else.
The final answer for cubic materials is only proportional to the fourth power
of the spin-orbit coupling strength. Therefore, one expects a large amount of
cancellation in the contributions of the points in k-space. Surprisingly, there
seems to be no structure in the contributions of the points in k-space. The
slow convergence of the numerical integrals, as discussed in a previous section,
shows that nearby points can have vastly different contributions. Here we also
show that there is no correlation with individual bands or with the change
in electronic structure.
Since all bands are important, and all k-points are important, and since
the contribution to the change in magneto-crystalline anisotropy energy as
a function of parameter change is uncorrelated for most k-points, we need
2 Calculation of Magneto-crystalline Anisotropy 97
2.8 Conclusions
Acknowledgments
References
1. B.D. Cullity: Introduction to Magnetic materials (Springer-Verlag, Berlin
1983)
2. R.M. Bozorth: Ferromagnetism (D. van Nostrand, Toronto 1951)
3. S. Chikazumi: Physics 0/ Magnetism (Krieger, Malabar, Florida 1964)
4. A.H. Morrish: The Physical Principles 0/ Magnetism (Wiley, New York 1965)
5. D.C. Mattis: The Theory 0/ Magnetism (Springer-Verlag, New York 1965)
6. R.M. White: Quantum Theory 0/ Magnetism (Springer-Verlag, Berlin 1983)
7. S. Lundqvist, N.M. March (eds.): Theory 0/ the Inhomogeneous electron Gas
(Plenum, New York 1983)
8. R.M. DreizIer and E.K.U. Gross: Density Functional Theory (Springer, Berlin
Heidelberg New York 1990)
9. E.H. Lieb: Density Functionals tor Coulomb Systems. In: Density Functional
Methods in Physics. ed. by R.M. DreizIer and J. da Providencia (Plenum, New
York 1985) pp.31
10. F. Mand, G. Shaw: Quantum Field Theory (Wiley, New York 1984)
2 Calculation of Magneto-crystalline Anisotropy 99
3.1 Introduction
The first-principle calculations of electronic structure, different ground state
properties and excitation spectra of strongly correlated materials are very im-
portant in the modern microscopic theory of magnetism. The magnetic prop-
erties of parent high-temperature superconducting oxides, nondoped man-
ganites and different ladder compounds are closely related to the electronic
structure of these systems, which appeared to be the Mott insulators [1], due
to strong electron-electron interactions. A standard way to investigate the
excitation spectrum and magnetic properties of strongly correlated electronic
systems is the model-Hamiltonian approach, such as the Hubbard model with
several adjustable parameters [2,3). The physics of the metal-insulator tran-
sition and properties of the Mott insulators are weH studied in the one-band
model with the strongly simplified bare electron spectrum, which is realized
for the models with infinite lattice connectivity (d = 00) [4,5). To investigate
strong interactions in real materials we have to develop a first principles
approach that takes into account a complicated crystal structure with several
atoms per unit cell, band degeneracy, and other important features of known
correlated electron systems. Unfortunately, except for small molecules, it is
impossible to solve many-body problem without severe approximations. The
most successful first principles method for investigations of electronic struc-
ture and magnetism of weakly correlated materials is the density functional
theory (DFT) within the local (spin) density approximation (L(S)DA) [6],
where the many-body problem is mapped into a noninteracting system with
an effective one-electron exchange-correlation potential that is approximated
by that of the homogeneous electron gas. The LSDA has proven to be very
efficient for many extended systems, such as large moleeules and solids. How-
ever, it cannot describe adequately correlation effects such as the Mott tran-
sition, charge and orbital ordering, heavy fermion physics, and many others
[1,3,5,7-10).
Strongly correlated magnetic materials where dynamical spin and charge
correlations are of crucial importance cannot be described satisfactory in the
LSDA, although an exact (but unknown) non-loeal DFT should be capable,
in principle, of finding the ground-state properties. These systems usuaHy
contain magnetic transition-metal or rare-earth metal ions with partiaHy
102 A.1. Lichtenstein et al.
The orbital energies Ei are derivatives of (3.1) with respect to orbital occu-
pations ni:
Ei = oE = ELDA + U (~ - ni ) • (3.2)
oni 2
This simple formula shifts the LDA orbital energy by -U/2 for occupied
orbitals (ni = 1) and by +U/2 for unoccupied orbitals (ni = 0). This orbital-
dependent LDA+U scheme gives upper and lower Hubbard bands with energy
separation between them equal to the Coulomb parameter U, thus reproduc-
ing qualitatively the correct physics for Mott-Hubbard insulators, at least
for the narrow-band limit. To construct a quantitatively sound calculational
scheme, one needs to define in a more general way an orbital basis set and
to take into account properly the direet and exchange Coulomb interactions
inside partially filled d (or f) atomic shell.
We need the identification of regions in space where the atomic char-
acteristics of the electronic states have largely survived ("atomic spheres"),
which is not a problem for the d or f eleetrons. Within these atomic spheres
one can expand an eleetron wave function into a localized orthonormal ba-
sis I inlma) (i denotes the site, n is the main quantum number, l- orbital
quantum number, m- magnetic number and a- spin index). Although it is
not strictly necessary, let us restrict ourselves to the usual situation where
only a particular nl shell is partially filled. The density matrix is defined by:
n':nrn' = --
1 JEF ImGfnlrn inlrn,(E)dE, (3.3)
7r '
104 A.1. Lichtenstein et al.
where Gatnm,'lnm
I . I ,(E) = (inlma I (E - H)-l I inlm' a) are the elements of
the Green function matrix in this localized representation, while H will be
defined later. In terms of this density matrix {na}, the generalized LDA+U
functional [23] is defined as follows:
ELDA+U[pa(r), {na}] = ELSDA[pa(r)] + EU[{n a }]_ Edc[{n a }]. (3.4)
Here pa (r) is the charge density for spin-a electrons and E LSDA [pa (r)] is the
standard LSDA functional. Equation (3.4) asserts that the LSDA suffices in
the absence of orbital polarizations, while the latter are driven by,
a
Vmm' L {( m, m " I v.:ee I m,' m ",)-a
= """ nm"m'"
rn',m"
(3.8)
where 0 ~ k ~ 2l and
k
ak(m,m',m",m"') = k47r '"' (lm I Ykq I lm')(lm" I Ykq Ilm"').
2 +1 ~
q=-k
For d electrons one needs FO, F 2 , and F 4 ; they can be linked to the Coulomb-
and Stoner parameters U and J via U = FO and J = (F 2 +F4 )/14, while the
ratio F 2 / F 4 is, to a good accuracy, a constant rv 0.625 for the 3d elements
[28,22]. For f electrons the corresponding expression is J = (286F 2 + 195F4 +
250F6)/6435.
If one neglects the nonsphericity of the Coulomb and exchange interaction
(which is exact in the case of the fuHy occupied or empty band) the energy
functional and potential correction will have the simpler form [29]:
ELDA+U[peT(r), {neT}] = ELSDA[peT(r)] (3.10)
U - J '"'
+-2- ~ ('"' eT - '"'
~ n mm eT eT ) ,
~ nmm,nm'm
t7 m m,m'
and
This form is useful for the full-potential calculations within LDA+U formal-
ism, since one needs to take care of non-spherical double counting in the LDA
part of the nonspherical formulation (3.8).
In the LDA +U approach the Coulomb interactions are taken into account
conventionaHy only on d orbitals of transition metals. However, it is known
that the Coulomb interactions between electrons on p orbitals of oxygen
are comparable in order of magnitude with the corresponding d-d Coulomb
interactions [30,31] and therefore should be taken into consideration on the
same footing as for d orbitals. The usual justification for omitting of U on
oxygen p sheH is that the oxygen sheH is fuHy occupied and the correlation
effects between electrons (or rather holes) in it can be neglected due to the
smaH number of holes in the ground state. However, the LDA+U scheme will
give nonzero correction for the fuHy occupied oxygen band:
(3.11)
This potential correction must be applied to the orbitals forming the oxygen
band, but the corresponding Wannier functions (in contrast to d states)
106 A.I. Lichtenstein et al.
are far from being of pure O(2p) character because they have very strong
admixture of sand p states of transition metal ions and other extended
orbitals. Since the main influence on the electronic structure is the change of
the energy separation between the oxygen p band and the transition metal
d band, the upward shift in energy of the transition metal d band on Up /2
will be equivalent to the shifting down of the oxygen p band on the same
value. Thus, in our calculations we added Up /2 term to the diagonal matrix
elements of the LDA+U potential correction (3.8). We call this extension of
the LDA+U method in this chapter the LDA+U(d+p).
Recently, the modified LDA+U(d+p) method was used by Korotin et al.
[32] for investigation of charge and orbital-ordering effects in the compound of
La7 /sSrl/SMn03. The inclusion of Coulomb interactions in the oxygen p shell
was found to be crucial in that calculation, since it controls the value of charge
transfer energy between Mn(3d) and O(2p) valence states and significantly
enhances the tendency of localization in this system.
The new Hamiltonian in (3.7) contains an orbital-dependent potential
in (3.8) in the form of a projection operator. This means that the LDA+U
method is essentially dependent on the choice of the set of the localized
orbitals in this operator. This is a consequence of the basic Anderson-model-
like ideology of the LDA+U approach: the separation of the total variational
space into a localized d (f) orbitals subspace, with Coulomb interaction be-
tween them treated with a Hubbard-type term in the Hamiltonian, and the
subspace of all other states, for which the local density approximation for
Coulomb interaction is regarded as sufficient. The arbitrariness of choice of
the localized orbitals is not as crucial as might be expected. The d (f) orbitals,
for which Coulomb correlation effects are important, are indeed welllocalized
in space, and retain their atomic character in asolid. The experience of
using the LDA+U approximation in various electronic structure calculational
schemes shows that the results are not sensitive to the particular form of the
localized orbitals.
Due to the presence of the projection operator in the LDA+U Hamil-
tonian in (3.7), the most straightforward calculational schemes would be to
use atomic-like orbitals type basis sets, such as LMTO (linear muffin-tin
orbitals) [33,34]. However, as soon as localized d (f orbitals) are defined, the
Hamiltonian in (3.7) could be realized even in schemes using plane waves as
a basis set, such as pseudopotential methods [35].
In spite of many successes [8] the LDA+U approach has obvious limit at ions
as a one-electron method with a single Slater determinant as a trial func-
tion. There are various ways to go beyond such a mean-field approximation.
The most convenient scheme for the large-scale first-principle calculations
seems to be the dynamical mean field theory (DMFT) [4] with the similar
3 Electronic Structure and Magnetism of Correlated Systems 107
HLMTO =
ilm,jl'm',a
(3.13)
where i is the site index and m is the orbital quantum numbers; a =t, t is
the spin projection; c+, c are the Fermi creation and annihilation operators
(n = c+c); E and t in (3.12) are effective one-electron energies and hopping
parameters obtained from the LDA in the orthogonal LMTO basis set. To
avoid the double-counting of electron-electron interactions one must in the
same time subtract the averaged Coulomb interaction energy term, which is
108 A.I. Lichtenstein et al.
present in the LDA. In the spirit of the LDA+U scheme we introduce new
E~ where the d-d Coulomb interaction is excluded:
°
Eda- (1) 1
= Eda- - U nd -"2 +"2 J (a-
nd -"2 ' 1) (3.14)
where U and J are the average values of Umml and Jmm, matrices and nd is
the average number of d electrons.
The screened Coulomb and exchange vertex for the d electrons are de-
fined as
H = HO + H int ,
HO = L (Oilm,jllm ' E?l nilma- + tilm,jllmlcltm"Cjllmla-)' (3.15)
ilm,jl'm',u
(3.16)
where JL is chemical potential, and the local Green function obtained via
integration over the Brillouin zone is:
(3.19)
3 Electronic Structure and Magnetism of Correlated Systems 109
For the bath Green function, one can introduce a different chemical potential
ji to satisfy the Luttinger theorem [5]:
1 '"' eiw 0+ G( . ) d
ßL n
~(. ) 0
~Wn d(iwn)LI ~Wn = . (3.20)
'Wn
where we redefined for simplicity m == {m, a}, Z is the partition function and
the so-called fermion-determinant det[O(s)] as weH as the Green function for
an arbitrary set of the auxiliary fields G(s) = 0-1(S) are obtained via the
Dyson equation [40] for imaginary-time matrix (Gm(s) == G~(s))
Gm = [1- (G~ _l)(e Vm -1)]-1G~.
Here, the effective fluctuation potential from the auxiliary Ising fields s!nm' =
±1 is defined as:
< m'
v;' = 2: Amm,S!nm,amm" whereamm, = {
1,m
-l,m>m,
,
m'(#m)
(12 lvi 34) = J drdr''l/Ji (r)'l/J2 (r')v scr (r - r') 'l/J3(r)'l/J4(r'), (3.22)
-(85IvI62)G63(r)G45(-r)] , (3.25)
JdTG~dT) G7~
ß
-L (12IvI56) (T) (78ITCTCT'134) .
5678 °
A similar approximation has been used for the Hubbard model [44] and
appeared to be accurate enough for a U that is not too large. Finally, in
the spirit of the DMFT-approach E = E[Go], and all the Green functions in
the self-consistent FLEX equations are, in fact, the bath Green functions Go.
where
B _ U[1 - (N - 1)n] - p, + ji
(3.31)
- U2(N - 1)no(1 - no) ,
A _ n[1- (N -1)n] + (N - 2)D[n]
(3.32)
- no(1- no) ,
(3.36)
(3.37)
(3.40)
i,j,a i<j,t7
k,i,j,u k,i,j,CT
114 A.1. Lichtenstein et al.
where T ij are the hopping parameters inside the cluster (for example, in
the case of a two-site cluster this is only intersite hopping tl.), E ij17 (k), and
rij17 (k) are the effective energies and hybridization matrix for finite-chain
bath orbitals k = 1, ... , nb.
For the iterative solution of the effective impurity model (3.41) one can
use the Lanczos version of ED method [5J. The orbital energy matrix for
the conduction band E ij17 (k), and the corresponding hybridization elements
rij17 (k), are the effective parameters that reproduce the bath Green function:
9- 1(iw n ) = (iw n + J.L) * 1 - T - L rk[iw n - Ekr 1rt . (3.41)
k
where
(3.43)
(3.44)
Using the spinor structure of the Dyson equation one can write the Green
function in this expression in terms of pair contributions (a similar trick has
been proposed in [56] in the framework ofthe LSDF approach). As a result, we
represent the total thermodynamic potential of spin rotations or the effective
Hamiltonian in the form [39]
[
J[lsPin] _ V. (3.46)
15 - ".
<Pi G=const
This means that [lspin {ei} is the effective spin Hamiltonian. The last term
in (3.45) is simply the Dzyaloshinski-Moriya interaction term. It is nonzero
only in the relativistic case where Ej and Gji can be, generally speaking,
"nonparallel" and Gij i' G ji for the crystals without an inversion center.
116 A.1. Lichtenstein et al.
In the nonrelativistic case one can rewrite the spin Hamiltonian for small
spin deviations away from collinear magnetic structures in the following form
where
are the effective exchange parameters [39]. This formula generalizes the LSDA
expressions of [54] to the case of correlated systems.
Spin-wave spectra in ferromagnets can be considered both directly from
the exchange parameters or by consideration of the energy of the correspond-
ing spiral structure (see [54]). In the nonrelativistic case when the anisotropy
is absent one has
4 4
wq = MLJOj(l-cosqRj ) = M[J(O)-J(q)], (3.49)
j
where M is the magnetic moment (in Bohr magnetons) per magnetic ion.
It should be noted that the expression for the spin-stiffness tensor Daß is
defined by the relation
(3.50)
where k is the quasimomentum and the summation is over the Brillouin zone.
The expressions (3.48) and (3.49) are reminiscent ofthe usual RKKY indirect
exchange interactions in the s-d exchange model (with ES instead of the s-d
exchange integral). One can prove [10] that the expression for the stiffness
is exact within the DMFT approximation. At the same time, the exchange
parameters themselves, generally speaking, differ from the exact response
characteristics defined via static susceptibility since the latter contains vertex
corrections. The derivation of approximate exchange parameters from the
variations of thermodynamic potential can be useful for the estimation of Jij
in the different magnetic systems.
Using our Green-function method, we can obtain the effective exchange
interaction parameters in the case of the static self-energy, or, equivalently,
3 Electronic Structure and Magnetism of Correlated Systems 11 7
in the LDA+U scheme in combination with (3.3-3.8), one obtains the well-
known expression [23]
_ ~ i ij j
Jij - ~ Imm'Xmmlmllmllllmllmlll' (3.52)
{m}
(3.53)
(3.54)
We will now review the applications of these new methods and computa-
tional schemes to real correlated materials.
Gd LDA+U
60 .---,---------------------------,
XPS BIS: .
Er'I
40
I
>CI)
I
./
." I :
'-- ~~
... ... ....... ............... j ....... . ........... .
I
I
o +.~~~~~~~~~~~
~~
....,,~
-1 0 -8 -6 -4 -2 0 2 4 6
Energy, eV
Fig.3.1. Density of states for ferromagnetic Gd metal from LDA+U calculation
and results of BIS (Bremstrahlung Isochrornat spectroscopy) and XPS (X-ray Pho-
toemission Spectroscopy) experiments
good agreement between the calculated and experimental spectra not only
for the separation between 4f bands but also for the position of the 4f peaks
relative to the Fermi energy. Gd is usually presented as an example where the
LSDA gives the correct electronic structure due to the spin-polarization split-
ting of the occupied and unoccupied 4f bands (in all other rare-earth met als
the LSDA gives unphysical 4f peak on the Fermi energy). In the LSDA, the
energy separation between 4f bands is not only strongly underestimated (the
exchange splitting is only 5 eV instead of the experimental value 12 eV) but
also the unoccupied 4f band is very dose to the Fermi energy, thus strongly
infiuencing the Fermi surface and magnetic ground state properties (in the
LSDA calculation, an antiferromagnetic state is lower in total energy than
the ferromagnetic one, in contradiction to experiment). The LDA+U method
solves both of these problems [57].
the Curie temperature of nickel [64]. The experimental data on the absence
of spin-polarization in the thermo emission from cesiated iron [65] are still
not understood completely [66]. From the theoretical point of view, different
approaches, such as the second-order perturbation theory [67,60], the three-
body Faddeev approximation [68], and the moment expansion method [69]
were used. Unfortunately, the conditions of applicability of these schemes are
not dear. We have investigated [37,39] the ferromagnetic properties of bcc
iron using the LDA+DMFT scheme in QMC and FLEX approximations and
argued that for moderately strong correlations (the case of iron-group met als )
one of the most efficient approximate ways for solving the quantum-impurity
problem in DMFT would be the fluctuation-exchange approximation.
First, we compare the density of states for ferromagnetic iron calculating
in the LSDA and LDA+DMFT scheme with the numerically exact QMC
solution for DMFT impurity problem (Fig.3.2). We computed the sum over
the auxiliary fields in (3.21) using an important sampling QMC algorithm
and performed 12 self-consistent iterations over the self-energy (3.18, 3.19,
3.21). The number of QMC sweeps was of the order of 105 on the CRAY-T3e
supercomputer . The final Gm (T) has very little statistical noise. We used the
maximum entropy method [41] for analytical continuations ofthe QMC Green
functions to the real axis. Comparison of the total density of states (DOS)
with the results of LSDA calculations (Fig. 3.2) shows a reasonable agreement
für single-partide properties of "moderately correlated" ferromagnetic iron.
4r-~---------..---------,
O~~~~-r~-4~~r=~
2
(j)
oCl
-4 -2 o 2 4
Energy (eV)
Fig. 3.2. Spin-resolved density of d states far ferramagnetic iran in the LSDA and
the LDA+QMC calculations for U = 2.3eV and J = O.geV
120 A.I. Lichtenstein et al.
We calculated the case of bcc iron at the experimental lattice constant with
256 k-points in the irreducible part of the Brillouin zone. The Matsubara
frequencies summation corresponds to a temperature of about T = 850 K.
The average magnetic moment is about 1.91l-B, which corresponds to a small
reduction of the LSDA-value of 2.21l-B for such a high temperature. The
DOS curves in the LDA+DMFT approach with exact QMC solution of on-
site multiorbital problem is similar to that obtained within the perturbative
fluctuation-exchange (FLEX) approximation.
The energy dependence of self-energy (calculated in the FLEX approach)
in Fig. 3.3 shows characteristic features of moderately correlated systems. At
low energies lEI< 1 eV we see a typical Fermi-liquid behavior ImE (E) '"
-E2 , 8ReE (E) /8E < o. At the same time, for the states beyond this
interval within the d bands the damping is rather large (of the order of
1 eV), so these states correspond to ill-defined quasiparticles, especially for
occupied states. This is probably one of the most important conclusions of our
calculations. Qualitatively, it was already pointed out in [67] on the basis of
a model second-order perturbation theory calculations. We have shown that
>CD
W-
H
-8 ·4 o 4 8
Energy, eV
Fig. 3.3. Total spin-polarized density of states and the d part of self-energy for iron
with U = 2.3eV and J = 0.geV for the temperature T = 750K. Two different self-
energies for t2g and eg d states in the cubic crystal field symmetry are presented
and the four different lines correspond to the imaginary part spin-up (Jull line)
and spin-down (dashed line) as weil as the real part spin-up (dashed-dot line) and
spin-down (dashed-double-dot line)
3 Electronic Structure and Magnetism of Correlated Systems 121
this is the case of realistic quasiparticle structure of iron with the reasonable
value of Coulomb interaction parameter.
One of the most unexpected experimental results concerning the electronic
structure of iron was obtained by spin-polarized thermoemission for cesiated
iron [65]. In this case, the thermal current is determined by the states with
the energy W = 1.37 eV above the Fermi level, which are in the region of the
quasiparticle DOS peak for minority spin. One might expect a strong negative
spin polarization of the current (polarization ratio P as estimated from LSDA
DOS is about -85% ). A more accurate estimation, which takes into account
the group velocities [66], results in P = -34% for the polarization ratio.
Experimentally it was found to be zero within the experimental error. To
clarify the situation, we considered this effect on the basis of our LDA+DMFT
calculations [37] . We obtained P = -12%, which is to be compared with
the value of -34% from LSDA calculations [66]. The decrease of P is not
a pure effect of damping of the quasiparticle states, but is the result of rather
complicated cancellations of s, p, and d electron contributions. Therefore,
one may conclude that there is no drastic discrepancy between experimental
results [65] and the theoretical description of the electronic structure of iron,
in spite of the approximate character of our treatment of the thermoemission
problem. For a more accurate description one needs to consider the surface
effects as weIl as an infiuence of the cesium layer on the electronic structure
of iron according to the experimental conditions.
500
0
Fe spin-wave 0
400
300
>CI)
E
;>:.
e>
CI) 200
c
W
100
0
0,0 0.5 1,0
Using the self-consistent values for Em(iw) computed by the QMC tech-
nique, we calculate [39] the exchange interactions (3.48) and spin-wave spec-
trum (3.49) using the four-dimensional fast Fourier transform (FFT) method
[70] for (k,iw) space with the mesh 203 x 320. The spin-wave spectrum for
ferromagnetic iron is presented in Fig. 3.4 in comparison with the results of
LSDA-exchange calculations [54] and with different experimental data [71-
73]. These room-temperature neutron scattering experiments have a sam-
pIe dependence (Fe-12%Si in [71,73] and Fe-4%Si in [72]) due to problems
with the bcc-Fe crystal growth. Note that for high-energy spin-waves the
experimental data [73] has large error bars due to Stoner damping (we show
one experimental point with the uncertainties in the q space). On the other
hand, the expression of magnon frequency in terms of exchange parameters
itself becomes problematic in that region due to breakdown of the adiabatic
approximation, as discussed above. Therefore, we think that comparison of
theoretical results with the experimental spin-wave spectrum for the large
energy needs additional investigation of Stoner excitation and required cal-
culations of dynamical susceptibility in the LDA+DMFT approach [5].
;;-
Ql 0
<n
!!l
<U
:§.
In
0 2
0
0
0.5
0
0.5
Fig.3.5. La2Cu04 DOS calculated by the LDA+U (left column) and the
LDA + U( d+p) (right column) methods [74]. (In all figures the total DOS is presented
per formula unit, the DOS of particular states are per atom. The Fermi energy
corresponds to zero)
6
4
2
0
;;;- 4
~ 2
!
CI)
0
2
4
0
Cl 3
o-8 -8 -4 - 2 0 2 4 4 8 10 -8 -4 - 2 0 2 4 6 8 10
Energy (eV) Energy (eV)
Fig. 3.6. MnO DOS calculated by the LDA+U (left column) and the LDA+U(d+p)
(right column) methods [74J
8
6
4
2
0
;;;- 4
~ 2
*
~
CI)
0
Cl
0
2
4
2
-8-4 - 2 0 2 4 -8-4 - 2 0 2 4
Energy (eV) Energy (eV)
Fig. 3.7. NiO DOS calculated by the LDA+U (left column) and the LDA+U(d+p)
(right column) methods [74J
Table 3.1. Calculated and experimental values of energy gaps (eV), spin magnetic
moments (J1-B) and intersite exchange interaction parameters J ex (meV)
NiO MnO
10
8
~
'e
:::l
~ 6
~
i
(f)
0 4
0
-2 0 2 4 6 -10 -8 -8 -4 -2 0 2 4 6 8
Energy (eV) Energy (eV)
Fig. 3.8. DOS calculated by the LDA+U (dashed line) and the LDA+U(d+p) (solid
line) Ni{3d) and Mn(3d) in comparison with superimposed XPS and BIS spectra
for MnO [76] and NiO [83]
of Ni. Using the corresponding atomic self-energy for a Ni atom, the total
DOS for NiO has been calculated. In Fig. 3.9, we compare the paramagnetic
LDA results with the HIA scheme. It is weIl known that paramagnetic LDA
calculations can not produce the insulating gap in nickel oxide: the Fermi level
located in the middle ofthe half-filled eg bands [9]. In the HIA approximation
there is a gap (or pseudogap in Fig. 3.9 due to temperature broadening) of the
3 Electronic Structure and Magnetism of Correlated Systems 127
15
El
E,
10
20
-o~ 15
>
~ 10
Cf)
0 5
Cl
0
20 g)
10
0
-10 -5 0 5
Energy (eV)
Fig.3.9. Density of states for paramagnetic nickel oxide in the LDA and HIA
approximations as weH as Ni-atom Green function
order of 3.5 eV, even in this "nonmagnetic" state. This gap and the satellit es
at -5 and -8 eV are related to the structure of the atomic Green function
shown in the lower panel of the Fig. 3.9.
VP3 (LDA)
Monoclinie Slructure
Fig. 3.10. Partial densities of state for 3d states of V obtained in the LDA calcu-
lation. üccupation numbers are given per one orbital (and both spins), i.e. each of
the E g orbitals has occupation of 0.81 electrons. The Fermi level is at zero energy
Fig. 3.11. The AF structure of the low-temperature AFI phase of V2Ü3. The gray
and filled circles correspond to spin-up and spin-down orientations of the local mag-
netic moments on V ions. The definitions of notations used for the superexchange
interactions along the various paths are also shown
coordinate system, i.e. with the z-axis along c-direction (V-V pairs), and the
E g orbitals are directed more towards the V ions in the basal plane.
At first glance, the LDA picture looks very simple and one may conclude
that the conventionaIly used ideas concerning the splitting of the A 1g orbitals
into bonding and antibonding partners because of a hopping integral between
the V ions in the pairs along the c-axis with two electrons per pair in the
bonding orbital are confirmed. However, upon closer examination a different
picture emerges, which is further supported by the LDA +U calculations be-
low as weIl as the recent experimental XAS results [90] . First, we see indeed
a rather broad A 1g band with a total width of a little more than 2 eV and
an E g band that is about ~ eV narrower, both stradling the Fermi energy
130 A.I. Lichtenstein et al.
VP3 (LDA+U, U=2.8, J=O.93)
2.00 r---...,....-----,----.---.---...,....---,-----,
(a)
1.00
c.~
E
~
1
(J)
8 -1.00
-2.00_2~.0-~-_~1.-=-
0 - - - - :0:':.0- - - - :1.'="
' 0 - - -2=".0: - - - - - :3'":.0- - - 4'.0
Fig. 3.12. Partial densities of states for d states of V obtained in LDA+U calcula-
tion in the antiferromagnetic structure
structure strongly depends on the magnetic structure. The band gap, for
instance, in the real AF magnetic structure is 0.6 eV, elose to the experimental
value (Fig.3.12).
There are two problems arising from the present results. First, the spin
should be considered to be 1 per V atom rather than 1/2 and, secondly,
the orbital occupation is consistent with a E~ configuration of the ground
state, which is in-plane symmetric and orbitally nondegenerate, but still with
the complex "real" magnetic structure of the AFI phase. In this magnetic
structure shown in Fig.3.11, every atom has three elosest neighbors in the
basal plane, one of which is ferromagnetically aligned (Ja) and the other two
antiferromagnetically (Jßl and J ß2)' and also there is one neighbor along the
hexagonal c-axis that is ferromagnetically aligned (J"j) (see Fig. 3.11 for def-
initions of exchange inter action parameters). To check the above-mentioned
consistency the exchange interaction parameters (EIP) have been calculated
using the method described above. It was found that according to EIP cal-
culation only one stable magnetic structure exists in the monoelinic phase.
("Stable" (or "consistent") means that if in the EIP calculations a certain
pair of spins was parallel, the corresponding exchange parameter came out
as ferromagnetic, and if antiparallel, as antiferromagnetic.) It is the "real"
AF magnetic structure with the following values of EIPs: Ja = 48 K (fer-
romagnetic), Jß1 = -214K, Jß2 = -90K, J"j = 47K. For the corundum
crystal structure we obtained two stable magnetic configurations from the
EIP calculation point of view: "real" AF (Ja = 55K, J ß1 = Jß2 = -120K,
J"j = 44 K) and magnetic structure with uniform AF exchange in the basal
plane (Ja = Jßl = Jß2 = -65 K) and small frustrated exchange along the
hexagonal c axis (J"j ~ 0). From these results one can say that the mono-
elinic distortion of the crystal structure does stabilize the real AF magnetic
structure and the E~ configuration of the d electrons is consistent with this
magnetic structure. The values of EIPs depend strongly on the magnetic
structure, which tells us that one cannot adequately model the magnetic
interactions in V 203 with only nearest-neighbor Heisenberg exchange. This
is most probably connected with the fact that V 2 0 3 is elose to being metallic,
which is also reflected in the strong dependence of the electronic structure on
the magnetic one (and vice versa) that we saw in the LDA+U calculation. The
polarized neutron scattering experiments [96] show the qualitative change of
magnetic interactions at the transition from the antiferromagnetic insulator
with the monoelinic crystal structure to both the metallic phase and param-
agnetic insulator with the corundum structure. lnstead of a peak in reciprocal
space corresponding to an AFI magnetic structure ("real" AF) it has a peak
corresponding to a magnetic structure with all three V-V interactions in
the basal plane antiferromagnetic ("layered" AF). Another peculiarity of the
neutron scattering results is a very large width of the peak.
Our calculations also give a value of the magnetic moment per V rv 1.7 I-lB,
nearly the same in all the structures studied. This value is somewhat larger
132 A.I. Lichtenstein et al.
than the value of 1.2 I-lB obtained from neutron scattering for the antiferro-
magnetic phase, but is consistent with the value (1. 71-lB) obtained from the
high-temperature susceptibility. This relatively large value of I-l is evidently
a consequence of a strong Coulomb interaction on V3+ ion, which tends to
destroy the formation of the molecular orbital singlet state on A1g-orbitals
of the V-V pair, which was assumed in most previous studies. The fact that
this is independent of the magnetic or crystal structure suggests that this
should be treated as a high-energy scale parameter in any model.
Summarizing this section, we have presented the LDA+U calculations
of the electronic structure and exchange constants of V 20 3 in both the
monoclinic and corundum structures and obtained a consistent description
of the main properties of the antiferromagnetic insulating phase and of the
paramagnetic insulating one. In contrast to the previous assumptions, in both
these phases the electronic configuration is predominantly an E; one, i.e. the
two d electrons of V3+ occupy the doubly degenerate Eg-orbitals. In addition,
the spins of the two electrons are parallel leading to a high spin S = 1 local
moment. As a result there is no orbital degeneracy left and correspondingly
no orbital ordering of the kind that was invoked previously to explain the
magnetic properties of V 203 [88,89J. Despite this, we are able to obtain
the correct magnetic structure of V 203: the signs of the exchange constants
in the monoclinic phase are consistent with the observed antiferromagnetic
structure.
Fig. 3.14. On the left, the crystal structure of CaV205 and on the right, the crystal
structure of MgV205. Ca and Mg atoms are not shown. Large balls represent V
and small - 0 atoms. In each figure the oxygen atoms, which are linked by edge
and corner sharing, constitute a pyramid, as described in the text
CaV205 MgV2 0 5
h 28 -60
h -608 -92
h -122 -144
J4 -20 -19
136 A.1. Lichtenstein et al.
250
200
150
:>
<I.l
§. 100
<0
W
W
50
-50
Quadrupolar distortion (% of a)
Fig. 3.15. The dependence ofthe total energy of KCuF3 on the quadrupolar lattice
distortion obtained in calculations with LSDA and LDA +U functionals
spins are ferromagnetically ordered in the planes while the unit cell is doubled
in the c-direction by antiferromagnetic spin ordering, so that the resulting
unit cell contains four formula units. The prime subject of the investigation
[23] was the quadrupolar distortion in planes, which is directly connected with
the peculiar orbital ordering. The total energy as a function of the shifts of
the fiuorine ions in the CuF 2 plane was calculated with the standard LSDA
and LDA+U functionals (Fig. 3.15). The striking difference between the two
calculations is that the LSDA solution has no instability against quadrupolar
distortion, while the LDA+U curve has a minimum at Q=2.5% of a compared
with the experimental value of 4.4%. This means that exchange-only and
lattice-electron ("electron-phonon") interactions in LSDA are not enough to
drive the observed orbital polarization and collective Jahn-Teller distortion.
In order to be able to reproduce them, the orbital-dependent interaction
terms must be included in the functional as is the' case in the LDA+U. It
is possible to directly observe the orbital ordering in KCuF 3 by plotting the
3-dimensional spin density obtained in the LDA+U calculation (Fig.3.16).
As there is only one hole in the d shell of the Cu2+ ion, this spin density
distribution gives the charge density of the holes. The picture agrees quite
weIl with the orbital ordering of the alternating x 2 - z2 and y2 - z2 Cu3d
orbitals. We notice that the charge distribution changes only very little under
the infiuence of the lattice distortion, emphasizing that this ordering is, in
the first instance, of an electronic origin.
As expected, the electronic properties come out essentially correct. The
"Koopman theorem" gap in the LDA+U band structure should give an order
138 A.1. Lichtenstein et al.
of magnitude estimate for the single particle gap and was found to be similar
to that in the cuprates (2 eV) [108]. Furthermore, the magnetic ordering is
reproduced and to test the method more severely the magnetic exchange
interactions were calculated as weIl.
It was found that the antiferromagnetic exchange in the CuF "chains"
amounts to Je = -20.7 meV, while the ferromagnetic exchange in the a- b
planes is much smaBer (Jab = 0.52 meV), emphasizing the quasi-1D character
of this S = 1/2 spin system. This compares quite weB with the neutron
scattering measurements, showing the Luttinger liquid nature of the spin
system, with the 1D exchange estimated to be Je = -17.5, -17.0 meV and
J ab = 0.17, 0.27 meV [110].
The doped rare-earth manganites Rel - xAxMn03 (Re is a rare earth such as
La and A is a divalent element such as Sr or Ca) due to their peculiar cor-
relation between magnetism and conductivity have been extensively studied
during the 1950s and 1960s [112]. The most thoroughly investigated was the
Lal-xSrxMn03 system. Undoped (x=O), LaMn03 is an antiferromagnetic
insulator. Upon doping with Sr, this perovskite oxide becomes a ferromag-
netic metal; the connection between metaBicity and ferromagnetism was weB
explained by the double exchange hopping mechanism [113]. The discovery of
colossal magnetoresistance phenomena in sampies with Sr dopant densities
in the 0.2 ::; x ::; 0.4 regime [111] brought a revival of the interest in these
systems.
The Mn+ 3 ion in the hole-undoped compound LaMn03 has the high-spin
d electron configuration t~gte~t . The t2g orbitals hybridize with 0 2p orbitals
4
much more weakly than the eg orbitals and can be regarded as forming the
3 Electronic Structure and Magnetism of Correlated Systems 139
Fig.3.18. The scheme of the spin, charge and orbital order in Prl/ 2Cal / 2Mn03
deduced from the neutron difraction data [116k The open circles with the lobe of
the eg electron density distribution denote Mn + and the filled circles Mn 4 +. The
arrows denote the magnetic moments
The result of the self-eonsistent ealculation was acharge and orbital or-
dered insulator. The eg electron spin-density-plot is presented in Fig. 3.17.
This is in striking agreement with the orbital order derived in [116] from
the neutron diffraetion measurements (Fig.3.18) . The unexpeeted result is
that the total number of d eleetrons at all types of Mn sites are nearly equal
(4.99 and 5.01) , so that formally MnH ions (in the Fig. 3.17 the ones with
symmetrie in-plane density distribution) and MnH ions (the ones with the
density strongly anisotropie) have nearly the same number of 3d eleetrons.
However, the difference in the magnetie moment values is more pronouneed:
3.34 MB for MnH and 3.44 MB for MnH .
3 Electronic Structure and Magnetism of Correlated Systems 141
:;- M M
~ -)
»
e."
""
~
·2
X r
r M r
Fig.3.19. (a) Dispersion of t2g bands of Sr2Ru04 in simplified two-dimensional
Brillouin Zone (EF = 0). (h) Solid lines: Fermi surface consistent with LDA band
structure and dHvA measurements, with hole sheet Q and electron sheets ß, I (after
accounting for hybridization) . Dashed line : approximate xy Fermi surface derived
from photoemission, indicating that I is hole-like
values of U this charge transfer is large enough to push the xy van Hove
singularity elose to or even below the Fermi level.
Since we are concerned with the qualitative influence of multiband corre-
lations on quasipartiele spectra, we have considered [128], for simplicity next-
nearest-neighbor tight-binding bands of the form e(k) = -co - 2t x cosak x -
2ty cos aky+4t' cosak x cos aky, where (co, t x , t y, t') = (0.50,0.44, 0.44, -0.14),
(0.24,0.31 , 0.045,0.01), (0.24,0.045,0.31,0.01) eV for xy, xz, yz, respectively
(see Fig. 3.19). These parameters ensure that the xy band has edges at -2.8
and 0.7 e V, with a van Hove singularity at 0.05 e V, and the xz, yz bands have
edges at -0.9 and 0.5eV, with van Hove singularities at -0.80 and 0.26eV,
in agreement with the LDA band structure [129] .
Re 1: ,,"' ...
u/ E,
Re 1:.,
>
~ 0 .. - .... _--
>-
e>
Q)
c: .......
'!' ·1
,.- Im I:.,:'
'äi
CI)
/
Im r",
·2
,-
·3 -2 ·1 0
Energy (eV)
E,
(a) IOMC I
>
~
*
U)
U5 0
'0
.z..
.a; (b) ~
c
Q)
0
xy
0
-3 -2 -1 o
Energy (eV)
Next we specify the onsite Coulomb and exchange integrals that we use
in the self-energy calculations discussed below. In the present case involv-
ing only t2g states, there are three independent elements (i -I- j) [133]:
U = (iillii), U' = (ijllij), and J = (ijllji) = (iillJj) = (U - U')/2,
where i = 1 ... 3 denotes xy, xz, yz. Thus, the Hartree-Fock energies are
EPF = n1U + 2n2(2U' - J) and E~r = nl(2U' - J) + n2(U + 2U' - J).
As the band occupations ni are rather similar, it is convenient to define
the average occupation n, so that nl = n - 28, n2,3 = n + 8, and EPF =
5n(U - 2J) + 28(U - 5J), E~r = 5n(U - 2J) - 8(U - 5J).
For a more accurate desc~iption of charge transfer among quasiparticle
bands, we include self-consistency in the dynamical mean-field theory [5].
A typieal frequency variation of Ei is shown in Fig.3.20. Near EF, the
imaginary parts vary quadratically with frequency and the real parts satisfy
Exz,yz » E xy , i.e. the energy shift of the narrow xz, yz bands is much
larger than far the wide xy band. Mareover, the difference Exz,yz - E xy
at E F is much larger than the difference between the Hartree-Fock energies
HF
E xZ,yz -
EHF
xy·
Qualitatively similar results are derived from more refined treatments of
on-site Coulomb correlations using multiband self-consistent quantum Monte
Carlo (QMC) methods [5,38]. The temperature oft he simulation was 15 meV
with 128 imaginary time slices and '" 300000 Monte Carlo sweeps. Figure 3.21
shows the quasiparticle density of states Ni (w) = - ~ Im Gi (w), obtained
via maximum entropy reconstruction [41], together with the bare density of
states Pi(W). The van Hove singularities near the edges of the xz, yz bands
are shifted towards E F , causing a sizeable band narrowing. Because of the
'" 2/3 filling of these bands, this effect is not symmetrie, giving astronger
relaxation shift of the occupied bands than for the unoccupied bands. There
144 A.1. Lichtenstein et al.
(al
:;- M M
~ 0
>-
e>
Q)
c:
.
..:: ....... :....... .
W
.,
'
r M X r M r
Fig.3.22. (a) Quasiparticle bands along r M and MX derived from self-consistent
second-order self-energy. Symbols : tight-binding bands. (b) Quasiparticle Fermi
surface after accounting for energy broadening and resolution (see text)
is also some band narrowing of the xy bands, but since U < Wxy this effect
is much smaller than for the xz, yz bands.
A crucial point is now that in order to satisfy the Luttinger theorem the
more pronounced band narrowing of the xz, yz bands requires a transfer of
spectral weight to the xy bands. Thus, the xy van Hove singularity is pushed
towards the Fermi level. In the example shown in Fig.3.22, it lies about
10 meV above E F , compared to 50 meV in the single-particle spectrum. We
emphasize that this result is a genuine multiband effect, where the filling of
a relatively wide quasiparticle band is modified by correlations within other
narrow bands of a different symmetry. Since the values of U and J are not
weIl known, and considering the approximate nature of our single-particle
bands and self-energy calculations, it is not possible at present to predict
the exact position of the xy singularity. It is conceivable, therefore, that this
saddle point might lie even closer to or below EF.
As indicated in Fig.3.19, the topology of the Fermi surface of Sr2 Ruü4
depends critically on the position of the xy van Hove singularity with respect
to E F . It is evident therefore that the charge transfer from xz, yz to xy due
to the creation of the photohole must be taken into account when using
angle-resolved photoemission to determine the shape of the Fermi surface.
To compare our results with photoemission spectra, we show in Fig.3.22
r
a the dispersion of the t2g quasiparticle bands along M and MX derived
from the spectral function Ai(k, w) = - ~ Im [w + /1- ci(k) - L'i(W)] -1. The
xy van Hove singularity at M lies lOmeV above EF, so that considerable
spectral weight appears below E F in the immediate vicinity of M. To account
for the finite energy resolution, and following the experimental procedure
for determining the spectral weight near E F [125], we show in Fig.3.22b
the Fermi surface obtained from the partially integrated spectral function
Ai(k) = J~~dwAi(k,w + iLl) with Ll = 25meV. Considering, in addition,
the finite aperture of the detector (typically ±1°, corresponding to ±5% of k ll
3 Electronic Structure and Magnetism of Correlated Systems 145
0.8
0.6 U=1.9
0.4
0.2
0.8
0.6 U=2.1
CI)
U:1 .3
Q)
0.4
1ii
üi 0.2
Ö
",
Z' 0.8
'e;;
c:
Q) 0.6 U=1.5 U=2.3
Cl
0.4
0,2
0,8
0,6 U: 1.7 U=2.5
0.4
0,2
Energy (eV)
"
," ,
,,, ,, ,
,
-4 -2 6
Energy (eV)
~
' _LO i ".,
.j-... ..... \
.- ,... .... .....
.... .
..... -
.
------- - .'
~,3 o 0,3 -3 3 6
- - QMC T=JOOOK
.,...... ' NCA
- - - - (PT
-3 o 3 6
Energy (eV)
Fig.3.25. Comparison of the spectral densities of Lal-xSrx Ti03 (x = 0.06) as
calculated by LDA+DMFT using the approximations IPT and NCA, with the
numerically exact QMC-result at T = O.leV, i.e. approximately 1000K, and U =
4 eV. Inset Zeft: behavior at the Fermi level including the LDA DOS. Inset right:
NCA and IPT spectra for a temperature of 80 K
right inset of Fig. 3.25) disappears already at about 250 K. A similarly narrow
IPT quasiparticle peak was found in a three-band model study with Bethe-
DOS by Kajueter and Kotliar [138]. While the NCA performs much bett er
than the IPT, it still underestimates the width of the quasiparticle peak by
a factor of two. Furthermore, the position of the quasiparticle peak is too close
to the lower Hubbard band. In the left inset of Fig. 3.25, the behavior at the
Fermi level is shown. At the Fermi level, the NCA yields a spectral function
that is too small by almost a factor two. The shortcomings of the NCA re-
sults appear to result from the well-known problems that this approximation
scheme encounters already in the single-impurity Anderson model at low
3 Electronic Structure and Magnetism of Correlated Systems 149
~
.5
,q
'"
.@
..s '.
, ,.."..
-3 -2 -1 o
Energy (eV)
Fig. 3.26. Comparison ofthe experimental photoemission spectrum [140], the LDA
result, and the LDA+DMFT(QMC) calculation for LaTi03 with 6% hole doping
and different Coulomb interaction U = 3.2, 4.25, and 5 eV
150 A.1. Lichtenstein et al.
Besides, the polycrystalline nature of the sampie and, also, spin and orbital
[141] fluctuation, not taken into account in the LDA+DMFT approach, could
further reduce the quasiparticle weight.
The LDA+DMFT approach not only explains the existence of the lower
Hubbard band in doped LaTi03' but also, in contrast to LDA, reproduces
the qualitative picture of the spectral weight transfer from the quasiparticle
band to the lower Hubbard band, the position of the lower Hubbard band,
and the narrowing of the quasiparticle band.
DMFT
Fig. 3.27. Schematic representation of the crystal structure of the vanadium layers
in NaY205 and the hopping matrix elements. The vanadium ions are denoted
by filled circles. The ellipse shows the cluster that plays the role of an effective
impurity in the DMFT calculations. A zigzag charge ordering of the y 4 + and y5+
ions, obtained from our LDA +U calculations as a ground state is shown on the left
ladder. Bold arrows are the translation vectors
ij i<j,a
where the tijS are the effective hoppings, Ui and Vij are local and intersite
Coulomb interactions, respectively, nia = ctcia' The proper choice of hop-
ping parameters is not simple, and the most widely used set (t-L = 0.38eV,
tll = 0.18eV, h = 0.012eV, h = 0.03eV) was obtained by the fitting to the
LDA bands [143J. Recently, a rigorous procedure of massive downfolding of
LDA bands to a few-band description and the subsequent Fourier transforma-
tion ofthe resulting Hamiltonian from the reciprocal to direct space to'extract
the single-electron parameters has been developed within the framework of
a linear-muffin-tin-orbital (LMTO) description [156J. This method applied
to NaV 20 5 gave the following set of hopping parameters: t-L = 0.398eV,
tll = 0.084 eV, t 1 = 0.025 eV, t2 = 0.022 eV which is rather elose to the stan-
dard one presented above, but, in addition, the diagonal (Fig.3.27) hopping
parameter td = 0.083 eV is appreciable and was not considered before. We
have found that these diagonal hopping processes are very important . As
can be seen in Fig. 3.28, the inelusion of diagonal hopping td in the single-
electron part of the model Hamiltonian provides a much better agreement of
152 A.I. Lichtenstein et al.
5
"
"
"
4 "
," ,
äi ,,
u
> 3
:'
~
!
cn
2
0
0 ./
,:
.....
I ,:
( .'
.;
0
-1 ~. 5 o 0.5
ENERGY(eV)
Fig. 3.28. Bare densities of states (DOS) . Solid lines correspond to the LDA DOS.
The dashed and lang dashed lines represent the first-principles tight-binding pa-
rameterizations without and with td hopping, respectively
the bare DOS with the LDA DOS. It also results in the insulating state of
the charge-disordered systems for the realistic values of U and V obtained
from LDA+U calculations, while the other TB parametrization, excluding
the diagonal hopping, results in the metallic state for the same values of U
and V (see Fig.3.29).
It is natural to assurne that the tendency to keep the number of d electrons
per rung close to unity also takes place above the transition temperature
leading to strong short-range order and well-developed dynamical charge
fluctuations. This is confirmed by the temperature dependences of the spin
gap and the entropy measurements [151]. Usual LDA as weIl as the mean-
field theories like Hartree-Fock or LDA+U methods are insufficient to take
into account these essential many-body processes. The simplest reliable way
to consider such short-ranged correlation effects is the use of the dynamical
mean-field theory (DMFT) [5], which can be combined with realistic LDA
band structure calculations (LDA+DMFT) [39,153,154]. The DMFT maps
the initial many-body problem for a crystal onto a self-consistent quantum-
impurity problem. To consider the phenomena such as charge ordering or
fluctuations , the intersite correlations are of crucial importance, therefore we
need to use a cluster generalization of the DMFT method [51 ,52,155]. The
most reasonable choice of the cluster in our case is a pair of vanadium atoms
at the rung (see Fig.3.27). The unique geometry of the ladder compounds
makes the choice of proper cluster simpler and the free-cluster consideration
is the most natural choice [52,155].
The phase diagram obtained in our DMFT calculations is shown in
Fig. 3.29. One can see that for large enough U and realistic values of V
the disordered state turns out to be insulating. This arises from the two
physically different mechanisms. The first is the spectral density transfer,
3 Electronic Structure and Magnetism of Correlated Systems 153
6 ,,
I
,,
~4
::l
,,
X ,
2
M
0
0 0.1 0.2 0.3 0.4 0.5 0.6
V (eV)
Fig. 3.29. Calculated phase diagram for the model with hopping parameter from
[143J (dashed line) and the parameters obtained from our TB-LMTO calculation
(solid line). The lines demark the metallic and insulating phases in two calcula-
tions. The cross corresponds to the values of U and V obtained from our LDA+U
calculation
which can effectively change the quarter-filled system into the half-filled one.
For large enough U the energy band splits into Hubbard subbands with the
average spectral weight 1/2 for each of them and therefore the system appears
to be an insulator for half-filling instead of complete filling [157]. In this
large-U limit the Hubbard model reduces to the so-called t-J-V model and
can explain insulating properties and optical spectra of NaV 2 0 5 [158]. Our
calculations do show the formation of the lower and upper Hubbard bands,
which leads to the spectral density transfer and moves the Fermi energy into
the pseudogap between bonding and antibonding states in the lower Hubbard
band, as was proposed qualitatively in [146]. However, the gap that can be
obtained due to this mechanism is very small. To increase its value (consistent
with the experiment [144]) the inclusion of the intersite Coulomb repulsion
(V) appears to be important (Fig. 3.30). Our calculations show that the broad
enough gap arises with the increase of V suddenly, as a result of a first-order
phase transition. With V = 0 the insulator is stable only for U :::::; 4 eV and
above, while already the smaH value of V = 0.1 eV decreases the critical
U value below 3eV. The results of LDA+U calculations (U = 2.8eV and
V = 0.17eV) gave a point in the phase diagram (the cross on Fig.3.29)
which is weH above the metal-insulator transition line. It is important to
stress that if one carries out the usual single-site DMFT calculation instead
of the cluster DMFT an adequate description of the electronic structure is
not obtained for the same values of the parameters, and the insulating states
appears only for U > 12 eV! This demonstrates the crucial importance of
the charge fluctuations on the rung for the formation of the insulating state
in NaV 2 0 5 with realistic values of U. We also analyzed the structure of the
ground state in the effective cluster model (3.41) and found that for our
154 A.1. Lichtenstein et al.
1.5 r----~--~--..,.......-__;__,__,
~ 1
>
:llCD
!
~ 0.5
Cl
o ~~~~-~~~~~-~~
1 0.5 0 0.5
ENERGY(eV)
Fig.3.30. The renormalized densities of state (only the lower Hubbard bands)
obtained in our DMFT calculation with parameters U = 2.8eV and V = O.17eV
(solid line); U = 2.8eV and V = O.5eV (dashed line)
3.9 Conclusion
The LDA+U method was proven to be a very efficient and reliable tool in
calculating the electronic structure of systems where the Coulomb interaction
is strong enough to cause localization of the electrons. It works not only for
nearly core-like 4f orbitals of rare-earth ions, where the separation of the
electronic states on the subspaces of the infinitely slow localized orbitals and
infinitely fast itinerant ones is valid, but also for such systems as transition
metal oxides, where 3d orbitals hybridize quite strongly with oxygen 2p-
orbitals. In spite of the fact that the LDA+U is a mean-field approximation,
which is, in general, insufficient for the description of the metal-insulator
transition and strongly correlated metals, in some cases, such as the metal-
insulator transition in FeSi and LaCo0 3 , LDA+U calculations gave valuable
information by giving insight into the nature of these transitions. However, in
general, LDA+U overestimates the tendency to localization as is well-know
for Hartree-Fock type methods. The main advantage of the LDA+U method
over model approaches is its "first principles" nature with a complete absence
of adjustable parameters. Another asset is its fully preserved ability of LDA-
based methods to address the intricate interplay of the electronic and lattice
degrees of freedom by computing the total energy as a function of lattice
distortions. When the localized nature of the electronic states with Coulomb
interactions between them is properly taken into account, this ability allows
us to describe such effects as polaron formation and orbital polarization. As
the spin and charge densities of the electrons are calculated self-consistently
in the LDA+U method, the resulting diagonal and off-diagonal matrix ele-
ments of the one-electron Hamiltonian could be used in more complicated
calculations where many-electron effects are treated beyond the mean-field
approximation. The main idea of the LDA+U method: the mapping of the
LDA Hamiltonian on the multiband lattice Anderson model, can be used for
constructing "ab-initio" calculating schemes based on the achievments of the
Anderson and Hubbard model studies.
At the same time, all the most subtle and interesting many-body effects
(such as spectral weight transfer, Kondo resonances, and others) are beyond
the LDA+U approach. To describe these effects, a dynamical character of
the effective potential acting on the electrons should be taken into account,
or, in other words, we have to work with the Green function instead of the
density matrix and with the self-energy instead of the effective exchange-
correlation potential. The LDA+DMFT method seems to be an effective and
useful form of such approaches. In particular, in contrast with the LDA+U
method it is not necessary to consider only the magnetically or orbitally
ordered phases to describe the Mott insulator states, spectral weight effects
are taken into account, etc. - see the results for NaV 2 0 5 , Ca2-xSrxRu04,
and Lal-xSrx Ti0 3 presented here. These first results demonstrate that the
dynamical mean field theory does give us an opportunity to unify the many-
156 A.I. Lichtenstein et al.
References
57. B.N. Harmon, V.P. Antropov, A.1. Lichtenstein, LV. Solovyev, and V.1. Anisi-
mov: J. Phys. Chem. Solids 56, 1521 (1995)
58. A. Liebsch: Phys. Rev. Lett. 43, 1431 (1979); Phys. Rev. B 23, 5203 (1981)
59. J. Staunton, B.L. Gyorffy, A.J. Pindor, G.M. Stocks, and H. Winter: J. Phys.
F 15, 1387 (1985)
60. M.M. Steiner, R.C. Albers, and L.J. Sham: Phys. Rev. B 45,13272 (1992)
61. V.Y. Irkhin, M.1. Katsnelson, and A.V. Trefilov: J. Phys.: Condens. Matter
5, 8763 (1993); S.V. Vonsovsky, M.1. Katsnelson, and A.V. Trefilov: Phys.
Met. Metallogr. 76, 247, 343 (1993)
62. D. Chanderis: J. Lecante, and Y. Petroff, Phys. Rev. B 27, 2630 (1983)
63. RE. Kirby, B. Kisker, F.K. King, and E.L. Garwin: Solid State Commun.
56, 425 (1985)
64. T. Greber, T.J. Kreuntz, and J. Osterwalder: Phys. Rev. Lett. 79, 4465
(1997); B. Sinkovich et al.: Phys. Rev. Lett. 79, 3510 (1997)
65. A. Vaterlaus, F. Milani, and F. Meier: Phys. Rev. Lett. 65, 3041 (1990)
66. R Monnier, M.M. Steiner, and L.J. Sham: Phys. Rev. B 44, 13678 (1991)
67. G. Treglia, F. Ducastelle, and D. Spanjaard: J. Phys. (Paris) 43, 341 (1982)
68. J. Igarashi: J. Phys. Soc. Jpn. 52, 2827 (1983); F. Manghi, V. Bellini, and C.
Arcangelli: Phys. Rev. B 56, 7149 (1997)
69. W. Nolting, S. Rex, and S.M. Jaya: J. Phys.: Condens. Matter 9, 1301 (1987)
70. S. Goedecker: Comp. Phys. Commun. 76, 294 (1993)
71. J.W. Lynn: Phys. Rev. B 11, 2624 (1975)
72. H.A. Mook and RM. Nicklow: Phys. Rev. B 7, 336 (1973)
73. T.G. Peerring, A.T. Boothroyd, D.M. Paul, A.D. Taylor, R Osborn, RJ.
Newport, and H.A. Mook: J. Appl. Phys. 69, 6219 (1991)
74. LA. Nekrasov, M.A. Korotin, and V.1. Anisimov: cond-mat/0009107
75. LV. Solovyev, P.H. Dederichs, and V.1. Anisimov: Phys. Rev. B 50, 16861
(1994)
76. G.A. Sawatzky and J.W. Allen: Phys. Rev. Lett. 53, 2239 (1984)
77. M.L. Knotek and P.J. Feibelman: Phys. Rev. Lett. 40, 964 (1978)
78. O.K. Andersen, Z. Pawlowska, and O. Jepsen: Phys. Rev. B 34, 5253 (1986)
79. O.K. Andersen, C. Arccangeli, RW. Tank, T. Saha-Dasgupta, G. Krier, O.
Jepsen, and I. Dasgupta: cond-mat/9804166
80. All calculations were done without downfolding, within the orbital basis of
Me(4s,4p,3d) (Me = Ni,Mn,Cu) and 0(3s,2p,3d). Logarithmic derivatives of
0(3s), 0(3d) and Me(3p) bands were fixed to make the EI/ and c parameters
of the LMTO method equal
81. One should say that we look at the energy gap between the highest edge of
the occupied and the lowest edge of the empty part of Me(3d)-band (Me =
Ni,Mn,Cu), because in total DOS the value of energy gap is '" 1 eV due to
very low intensive 0(3s) band
82. J. Zaanen, G.A. Sawatzky, and J.W. Allen: Phys. Rev. Lett. 55, 418 (1985)
83. J. van Elp, R.H. Potze, H. Eskes, R Berger, and G.A. Sawatzky: Phys. Rev.
B 44, 1530 (1991)
84. M.A. Korotin, I.S. Elfimov, V.1. Anisimov, M. Troyer, and D.1. Khomskii:
Phys. Rev. Lett. 83, 1387 (1999)
85. LV. Solovyev and K. Terakura: Phys. Rev. B 58, 15496 (1998)
86. S.Y. Ezhov, V.I. Anisimov, D.L Khomskii, and G.A. Sawatzky: Phys. Rev.
Lett. 83, 4136 (1999)
3 Electronic Strueture and Magnetism of Correlated Systems 159
113. C. Zener: Phys. Rev. 82,403 (1951); P.W. Anderson and H. Hasegava: Phys.
Rev. 100, 675 (1955); P.G. de Gennes, Phys. Rev. 118, 141 (1960)
114. J.B.A.A. EIernans, B. van Laar, K.R. van der Veen, and B.O. Loopstra: J.
Solid State Chern. 3, 238 (1971)
115. V.I. Anisirnov, I.S. Elfirnov, M.A. Korotin, K. and Terakura: Phys. Rev. B
55, 15494 (1997)
116. Z. Jirak, S. Krupicka, V. Nekvasil, E. PolIert, G. Villeneuve, and F. Zounova:
J. Magn. Magn. Mater. 15-18, 519 (1980); Z. Jirak, S. Krupicka, Z. Sirnsa,
M. Dlouha, and Z. Vratislav: ibid. 53, 153 (1985)
117. Y. Tornioka, A. Asamitsu, H. Kuwahara, Y. Moritomo, and Y. Tokura: Phys.
Rev. B 53, R1689 (1996)
118. N. Hamada, H. Sawada and K. Terakura, in Spectroscopy 01 Mott Insulators
and COrTelated Metals, eds. A. Fujirnori and Y. Tokura (Springer, Berlin
Heidelberg New York 1995) pp. 95-105
119. W. Pickett and D. Singh: Phys. Rev.B 53, 1146 (1996)
120. S. Satpathy, Z. Popovic, and P. Vukajlovic: Phys. Rev. Lett. 76, 960 (1996)
121. Y. Maeno, H. Hashimoto, K Yoshida, S. Nishizaki, T. Fujita, J.G. Bednorz,
and F. Lichtenberg: Nature 372, 532 (1994)
122. A.P. Mackenzie, S.R. Julian, A.J. Diver, G.J. MacMullan, M.P. Ray, G.G.
Lonzarich, Y. Maeno, S. Nishizaki, and T. Fujita: Phys. Rev. Lett. 76, 3786
(1996); A.P. Mackenzie, S.R. Julian, G.G. Lonzarich, Y. Maeno, and T. Fujita:
Phys. Rev. Lett. 78, 2271 (1997)
123. T. Yokoya, A. Chainani, T. Takahashi, H. Katayama-Yoshida, M. Kasai, and
Y. Tokura: Phys. Rev. Lett. 76, 3009 (1996); ibid. 78,2272 (1997); T. Yokoya,
A. Chainani, T. Takahashi, H. Ding, J.C. Campuzano, H. Katayama-Yoshida,
M. Kasai, and Y. Tokura: Phys. Rev. B 54, 13311 (1996)
124. D.H. Lu, M. Schmidt, T.R. Cummins, S. Schuppler, F. Lichtenberg, and J.G.
Bednorz: Phys. Rev. Lett. 76, 4845 (1996)
125. A.V. Puchkov, Z.X. Shen, T. Kimura, and Y. Tokura: Phys. Rev. B 58,
R13322 (1998). According to these data, the intensity near M varies strongly
with photon energy, suggesting that at 22 eV the xy van Hove singularity is
occupied while above 25eV it is unoccupied
126. A. Damaseelli, D.H. Lu, KM. Shen et al.: Phys. Rev. Lett. 85, 5194 (2000);
KM. Shen, A. Damascelli, D.H. Lu et al., cond-mat/0105487
127. T. Oguchi: Phys. Rev. B 51, 1385 (1995)
128. A. Liebsch and A. Lichtenstein: Phys. Rev. Lett. 84, 1591 (2000)
129. LI. Mazin and D.J. Singh: Phys. Rev. Lett. 79, 733 (1997); D.J. Singh: Phys.
Rev. B 52, 1358 (1995)
130. A. Fang and K Terakura, private communication
131. A.P. Mackenzie, S. Ikeda, Y. Maeno, T. Fujita, R. Julian, and G.G. Lonzarich:
J. Phys. Soc. Japan 67, 385 (1998)
132. T. Yokoya, A. Chainani, T. Takahashi, H. Katayama-Yoshida, M Kasai, Y.
Tokura, N. Shanthi, and D.D. Sarma: Phys. Rev. B 53, 8151 (1996)
133. J. Kanarnori: Progr. Theoret. Phys. 30, 275 (1963); J. Igarashi, P. Unger, K
Hirai, and P. Fulde: Phys. Rev. B 49, 16181 (1994)
134. Y. Maeno et al.: J. Low Temp. 105, 1577 (1997)
135. V.I. Anisimov, LA. Nekrasov, D.E. Kondakov, T.M. Rice, and M. Sigrist:
cond-mat/0l07095
136. M.B. Zölfl, T. Pruschke, J. Keller, A.I. Poteryaev, LA. Nekrasov, and V.I.
Anisimov: Phys. Rev. B 61, 12810 (2000)
3 Electronic Structure and Magnetism of Correlated Systems 161
137. LA. Nekrasov, K. Held, N. Blümer, A.I. Poteryaev, V.I. Anisimov, and D.
Vollhardt: Eur. Phys. J B 18, 55 (2000)
138. H. Kajueter and G. Kotliar: Int. J. Mod. Phys. 11, 729 (1997)
139. E. Müller-Hartmann: Z. Phys. B 57, 281 (1984)
140. A. Fujimori et al.: Phys. Rev. Lett. 69, 1796 (1992). A. Fujimori et al.: Phys.
Rev. B 46, 9841 (1992)
141. B. Keimer et al.: Phys. Rev. Lett. 85, 3946 (2000)
142. M. Isobe and Y. Ueda: J. Phys. Soc. Japan 65, 1178 (1996)
143. H. Smolinski et al.: Phys. Rev. Lett. 80, 5164 (1998)
144. C. Presura, D. van der Marel, A. Damascelli, and R.K. Kremer: Phys. Rev.
B 61, 15762 (2000)
145. H. Seo and H. Fukuyama: J. Phys. Soc. Japan 67, 2602 (1998)
146. M.V. Mostovoy and D.I. Khomskii: Solid State Commun. 113, 159 (2000)
147. H. Nakao et al.: Phys. Rev. Lett. 85, 4349 (2000)
148. T. Ohama et al.: J. Phys. Soc. Japan 69, 2751 (2000)
149. D. Sa and C. Gros: Eur. Phys. J. B 18, 421 (2000)
150. A.N. Yaresko, V.N. Antonov, H. Eschrig, P. Thalmeier, and P. Fulde: Phys.
Rev. B 62, 15538 (2000)
151. D.C. Johnston et al.: Phys. Rev. B 61, 9558 (2000)
152. V.V. Mazurenko, A.I. Lichtenstein, M.I. Katsnelson, I. Dasgupta, T. Saha-
Dasgupta, and V.I. Anisimov: cond-matj0107200
153. V.I. Anisimov, A.I. Poteryaev, M.A. Korotin, A.O. Anokhin, and G. Kotliar:
J. Phys.: Condens. Matter 9, 7359 (1997)
154. LA. Nekrasovet al.: Eur. Phys. J. B 18, 55 (2000); M.B. Zoelfl et al.: Phys.
Rev. B 61, 12810 (2000)
155. G. Kotliar, S.Y. Savrasov, and G. Palsson: cond-matjOO10328
156. O.K. Andersen and T. Saha-Dasgupta: Phys. Rev. B 62, R16219 (2000)
157. J. Hubbard: Proc. Roy. Soc. (London) A 281, 401 (1964)
158. M. Cuoco, P. Horsch, and F. Mack: Phys. Rev. B 60, R8438 (1999)
159. V.Y. Irkhin and M.I. Katsnelson: JETP Lett. 49, 576 (1989); Phys. Lett.
A 150, 47 (1990)
160. A. Langari, M.A. Martin-Delgado, and P. Thalmeier: Phys. Rev. B 63, 054432
(2001)
4 Ferromagnetism
in (III,Mn)V Semiconductors
4.1 Introduction
Ferromagnetism occurs when Mn is randomly substituted for more than
about 2 percent of the cations of several III-V compound semiconductors.
Although only a few host materials have been explored at present, this prop-
erty is likely shared by most III-V semiconductors. In this chapter we will
discuss some of the theoretical pictures that are being developed to explain
the magnetic and transport properties of these materials. Our development
will be based on a phenomenological model that has been used with great
success to explain the sensitivity of bulk and layered (II,Mn)VI semiconduc-
tor optical properties to. external magnetic fields. (Ferromagnetism does not
occur for Mn in undoped lI-VI hosts.) The low energy degrees of freedom
in this model are holes in the semiconductor valence band and one S = 5/2
local moment for each Mn ion.
Interest in these ferromagnets was heightened by the demonstration sev-
eral years aga that ferromagnetic transition temperatures [1] in excess of
100 K can be achieved in (Ga,Mn)As. It has been further heightened recently
both by the demonstration of long spin-coherence times 'in semiconductors
[2] and by the dramatic and rapid development of new information storage
technology based on magnetotransport effects in ferromagnetic met als [3]. It
seems elear that semiconductors have many potential advantages over met als
for devices based on the magnetotransport effects that occur in itinerant
electron ferromagnets, principally because they present a wider canvas for
their creative manipulation by some combination 'of impurities, gates and
optical excitation. It is likely that important applications for these materials
will be found only if ferromagnetism at room temperature can be achieved.
The very recent discovery [4] of ferromagnetism at temperatures elose to
1000 K in (Ga,Mn)N has fueled hopes that these materials will indeed have
technological impact. Our focus here, however, is on the physics of these fer-
romagnets; we therefore concentrate mainly on the properties of (Ga,Mn) As
and (In,Mn)As which have been studied most extensively [5].
It is generally accepted that Mn acts as an acceptor when it substitutes
for a cation in a I II-V semiconductor lattice, leaving a MnH ion which has
a half-filled d-shell with angular moment um L = 0 and spin S = 5/2. It is also
generally accepted that ferromagnetism OCCUfS in these materials because of
164 J. König et al.
In Sects. 4.4-4.6 we have been able to present only a small fraction of the-
oretical data obtained using our approach. An extensive survey of predicted
physical properties of DMS's will be available at http://unix12 . fzu. cz/
ms/index. php web-pages, launched by the authors of this chapter in collab-
oration with Jan Kucera, Byounghak Lee, and Jairo Sinova.
large values of M(T = 0) likely have ground states with (nearly) fully
aligned Mn local moments.
• To date the largest ferromagnetic transition temperatures in (Ga,Mn)As
occur for x rv 5%. The current record is Tc rv lIOK [1]. The drop in
critical temperatures at higher x values may be related to Mn dustering
or may have a more fundamental origin. There is a correlation between
large values of M(T = 0) and high Tc's.
• (III,Mn)V thin film ferromagnets grown under compressive strain have
their magnetic easy axis in the plane, while ferromagnets grown under
tensile strain have their magnetic easy axis in the growth direction [5].
External magnetic fields rv 100 mT are sufficient to align the magnet i-
zation along the hard axis [14,7]. These properties can be explained by
weIl understood strain effects in the spin-orbit coupled valence bands and
demonstrate that the macroscopic properties of these ferromagnets are
sensitive to details of the valence band electronic structure [5].
• The ferromagnetic critical temperature and the temperature dependence
of the magnetization are altered by post-growth annealing and are sensi-
tive to the details of the annealing protocol [5].
• These ferromagnets have large anomalous Hall resistivities [5], demon-
strating that the itinerant valence bands are full participants in the
magnetism. The large anomalous Hall resistivities reflect the strong spin-
orbit coupling that is present at the top of the valence band in zincblende
semiconductors.
• The semiconductor valence bands are spin-split in the ferromagnetic
state. These semiconductor ferromagnets exhibit strong magnetoresis-
tance effects, like tunnel magnetoresistance [15], that are characteristic
of itinerant electron ferromagnets, again demonstrating that the itinerant
electrons are full participants in the magnetism.
are not simply related to the bare eleetrons which appear in the many-eleetron
Hamiltonian. Often, however, the low-energy degrees of freedom are more
obvious.
A practical approach to many-eleetron physies that is often sueeessful is
spin-density-functional (SDF) theory, in whieh many-body effeets appear in
exehange-eorrelation potential eontributions to effeetive independent-particle
Hamiltonians. SDF theory has the advantage that it is a first principles ap-
proach without any phenomenological parameters. Among its many aehieve-
ments is a generally satisfaetory deseription of itinerant electron ferromag-
netism in transition metals. SDF theory has been applied [16J to (III,Mn)V
ferromagnetism by Sanvito et al. and by van Sehilfgaarde and Mryasov. To
date these ealculations have been performed using the loeal density approx-
imation (LDA) of SDF theory, an approximation that is not reliable when
loeal moments are formed, i.e., when strong eorrelations suppress fluctuations
in the number of eleetrons in the d-shell or the f-shell of a particular atom.
LDA-SDF theory supereell [16J ealculations predict that majority spin d-
eleetrons of a Mn atom substituted on a eation site of GaAs lie at the Fermi
energy, rat her than lying well below the Fermi energy as they would if the
half-filled d-shell formed a S = 5/2 loeal moment. Similar results have been
obtained in eoherent potential approximation band-strueture ealeulations [17J
for (In,Mn)As, and are clearly in disagreement with experiment in both eases.
In loeal moment systems, it is neeessary to aeeount for the inerease in instan-
taneous cite energy when the oeeupaney of a loealized orbital is inereased.
The LDA+U method has been developed to mitigate this deficieney of SDF
theory and has reeently [18J been applied to (III,Mn)V ferromagnets and
finds a dominant As 4p orbital weight at the Fermi energy, eonsistent with
the photoemisson experiment.
It appears likely that a lot of detailed information on the electronic proper-
ties of (III,Mn)V ferromagnetie semieonductors ean be reliably obtained from
LDA+U SDF ealculations and we expect this approach will play an important
role in the modeling of these materials in the future. Our work, however,
follows a semi-phenomenological strategy, starting from a model in which the
loeal-moment eharaeter of the Mn d-orbitals is asserted rather than derived.
These models are adapted from ones used [19,20J to deseribe the optieal
properties of (II,Mn)VI semiconduetors. The low-energy degrees of freedom
in the kinetic-exchange model [21,22J we employ are S = 5/2 loeal moments
representing half-filled Mn d-shells, and holes in the Mn valenee band (see
Fig.4.1). Sinee Mn2+ ions should act as aeeeptors when substituted on the
eation sites of III-V semiconductors, we would expect one hole per Mn if no
other eharged defects were present in the system. However, antisite defects
are eommon when III-V semiconductor films are grown by MBE at the low
temperatures required to prevent Mn segregation. The hole density and the
Mn density are therefore taken as separate sample-dependent quantities, to
be determined experimentally. The Hamiltonian of this model is specified in
168 J. König et al.
t "t'\ t
-
t
\ \
\ \
t
\
"-/
t
Fig.4.1. Model for (III,Mn)V semiconductors: local magnetic moments (Mn 2 +)
with spin S = 5/2 are antiferromagnetically coupled to itinerant carriers (holes)
with spin s = 1/2
detail in the following paragraph. There has also been theoretical work on
these materials based on a still simpler model [23J where holes are assumed
to hop only between Mn acceptor sites, where they interact with the Mn
moments via phenomenological exchange interactions. These models have
some advantages in getting at the physics of the dilute Mn limit, and can also
easily be adapted to include the holes that are localized on ionized antisite
defects rather than Mn acceptors [24J.
Like any phenomenological model, the one we use is defined most fun-
damentally by its low-energy degrees of freedom. Also important, however,
is the Hamiltonian that acts in the implied Hilbert space. The length scales
associated with holes in these compounds are still long enough that a k . p,
envelope function, description [25J of the semiconductor valence bands is
appropriate and we take that approach here. The operators in terms of which
the phenomenological Hamiltonian is expressed include the spin operator SI
for the S = 5/2 local moment on site land the multi-band envelope function
hole spin density operator s(r). The following key terms are included in the
minimal version of the model Hamiltonian:
a) The coupling of the Mn spin to the external magnetic field, gj.LB LI SI·
Hext·
b) The band Hamiltonian of the host III-V semiconductor, usually described
using a multi-band envelope function formalism [25,26J. For many prop-
erties it is necessary to incorporate spin-orbit coupling in a realistic way.
Six- or eight-band models that include the 'split-off' band and/or the con-
duction band are sometimes desirable. Unlike the local moment models,
the Hilbert space includes all host lattice sites for each hole rather than
only localized orbitals centered on the Mn sites. This band Hamiltonian
should include the strain effects due to lattice matching between the
epitaxially grown (III,Mn)V films and the substrate on which they are
grown.
4 Ferromagnetism in (III,Mn)V Semiconductors 169
The terms d)-f), listed below, are necessary to describe the crossover to
the localized limit in which holes are bound either to Mn acceptors or to other
defects. Note that simpler, impurity-band models assurne that the system is
in this limit from the outset, much as our model assurnes from the outset
that the Mn d-shells form local moments. There are sometimes technical
difficulties in describing the localized limit with our higher-Ievel model, so
that considerable simplification arises from using the impurity-band model.
There is however, a penalty to pay since the model does not apply to the
regimes of greatest interest in which the band electrons are not localized.
Even when impurity models do apply, it is difficult to guess at appropriate
distribution functions for the inter-site hopping parameters that play a key
role.
g) The scalar scattering potential that represents the energy difference be-
tween a valence-band p electron on a host site and a valence-band p
170 J. König et al.
electron on a Mn site. This effect has normally been exc1uded since its
size and sign is not yet know.
h) Direct exchange interactions between Mn ions on neighboring sites. These
terms result from microscopic pro ces ses in which exchange of electrons
between the valence band and two nearby Mn d-shells is correlated. Terms
of this type are known to be important in (II,Mn)VI semiconductors, but
appear to be less important in (III,Mn)V semiconductors.
i) Direct coupling of band electrons to external magnetic field.
In the rest of this chapter we discuss only pictures of (III,Mn)V ferro-
magnetism that follow from the minimal model that inc1udes only the a)-c)
Hamiltonian terms.
Here, g* is the hole g-factor and Xi is the interacting hole magnetic suscep-
tibility,
Xi
(4.5)
where Etot/V is the total energy density of the itinerant-hole system. The
Curie-Weiss transition temperature, obtained from (4.1)-(4.4) in the Hext = 0
limit, is
(4.6)
25
100
50
g 0
f-"
AIAs
100
50
600
500
400
300
0.1 0.2 0.3 DA 0.5
p(nm-")
where h, is the unit vector along the field h. The cubic anisotropy coefficients
K'r and K 2a are related to total energies for h, along the high symmetry
crystal directions by following expressions:
K~a = 4[Etot ((1l0)) - E tot ((100))] ,
V
K~a = 27Etot ((111)) - 36Etot ((11O)) + 9Etot ((100)) (4.10)
V .
MBE growth techniques produce (III,Mn)V films whose lattices are locked
to those of their substrates. X-ray diffraction studies [5] have established that
the resulting strains are not relaxed by dislocations or other defects, even for
thick films. Strains in the (III,Mn)V film break the cubic symmetry assumed
in (4.9). However, the influence of MBE growth lattice-matching strains on
the hole bands of cubic semiconductors is wen understood [25] and we can use
the same formal mean-field theory as in the previous subsection to account
for strain effects on magnetic anisotropy.
We turn now to aseries of illustrative calculations intended to closely
model the ground state of (Ga,Mn)As. For Mn density N Mn = 1 nm- 3
(x >:::! 5%), h >:::! 140meV at zero temperature. This value of h is not so
much smaIler than the spin-orbit splitting parameter in GaAs [34,7] (.1 80 =
341 meV), so that accurate calculations require the six-band Luttinger model
[26]. Even with N Mn fixed, our calculations show that the magnetic anisotropy
of (III,Mn)V ferromagnets is strongly dependent on both hole density and
strain. The hole density can be varied by changing growth conditions or by
adding other dopants to the material, and strain in a (Ga,Mn)As film can
be altered by changing substrates. The cubic anisotropy coefficients (in units
of energy per volume) for strain-free material are plot ted as a function of
4 Ferromagnetism in (III,Mn)V Semiconductors 175
5.0
2.5
~
'"
I
E
~
0.0
~
e
öcn
'c 2.5
ro
5.0
0.0 0.1 0.2 0.3
P (nm -3)
Fig. 4.3. Cubic magnetic anisotropy coefficients K 1a and K~a as a function of hole
density p
hole density in Fig.4.3. The easy axis is nearly always determined by the
leading cubic anisotropy coefficient K 1a , except near values of p where this
coefficient vanishes. As a consequence, the easy axis in strain free sampIes is
almost always either along one of the cube edge directions (K1a > 0), or along
one of the cube diagonal directions (K1a < 0). Transitions in which the easy
axis moves between these two directions occur twice over the range of hole
densities studied. (Similar transitions occur as a function of h, and therefore
temperature, for fixed hole density.) Near the hole density p = 0.01 nm -3,
both anisotropy coefficients nearly vanish and a fine-tuned nearly perfeet
isotropy is achieved. The slopes of the anisotropy coefficient curves vary as the
number of occupied bands increases from 1 to 4 with increasing hole density.
This behavior is clearly seen from the correlation between oscillations of the
anisotropy coefficients and onsets of higher band occupations.
Six-band model Fermi surfaces are illustrated in Figs.4.4 and 4.5 by
plotting their intersections with the k z = 0 plane at p = 0.1 nm- 3 for the
cases of (100) and (110) ordered moment orientations. The dependence of
quasiparticle band structure on ordered moment orientation, apparent in
comparing these figures, should lead to large anisotropie magnetoresistance
effects in (Ga,Mn)As ferromagnets. We also note that in the case of cube edge
orientations, the Fermi surfaces of different bands intersect. This property
could have important implications for the decay of long-wavelength collective
modes.
In Fig. 4.6 we present mean-field theory predictions for the strain-depend-
ence ofthe anisotropy energy at h = 140 meV and hole density p = 0.35nm- 3 .
According to our calculations, the easy axes in the absence of strain are along
176 J. König et al.
0.10
0.05
"i
0.00
!!.
::t?"
-0.05
-0.10
-0.10 -0.05 0.00 0.05 0.10
k. (ao-')
Fig.4.4. Six-band model Fermi surface intersections with the k z = 0 plane for
p = 0.lnm- 3 and h = 140 meV. This figure is for magnetization orientation along
the (100) direction
0.10
0.05
•
"i
0.00
!!
::t?"
-0.05
-0.10
-0.10 -0.05 0.00 0.05 0.10
k, (ao-')
Fig.4.5. Six-band model Fermi surface intersections with the k z = 0 plane for the
parameters of Fig. 4.4 and magnetization orientation along the (110) direction
the cube edges in this case. The relevant value of the in-plane strain produced
by the substrate-film lattice mismatch,
a s - af
eo = , (4.11)
af
-
Q
tCO
Q)
4
0.. 2
'-
Q)
0..
>.
a:: o
'?
o
,.... Etot <OO1>E tot< 100>
-2 Etot <111>E tot< 100>
Fig.4.6. Energy differences among (001), (100), (110), and (111) magnetization
orientations vs. in-plane strain eo at h = 140 meV and p = 0.35 nm -3. For com-
pressive strains (eo < 0), the system has an easy magnetic plane perpendicular to
the growth direction. For tensile strains (eo > 0), the anisotropy is easy-axis with
the preferred magnetization orientation along the growth direction. The anisotropy
changes sign at large tensile strain
The first term on the right-hand-side of (4.12) is the standard Bloch band
group velocity. Gur anomalous Hall conductivity is due to the second term,
proportional to the Berry curvature {l, defined below. It follows from sym-
metry considerations that for a cubic semiconductor under lattice-matching
strains and with maligned by external tIelds along the (001) growth direction,
only the z-component of {l is nonzero:
(4.13)
Here ju n ) is the periodic part of the n-th Bloch band wavefunction with the
mean-field spin-splitting term included in the Hamiltonian. The anomalous
Hall conductivity that results from this velo city correction is
O"AH
e2
= -r; L n
J dk fn
(27r)3 ,k[lZ(n,k) , (4.14)
where f n,k is the equilibrium Fermi occupation factor for the band quasi-
particles. We have taken the convention that a positive 0" AH means that the
anomalous Hall current is in the same direction as the normal Hall current.
This Berry phase contribution to the anomalous Hall conductance occurs
in any itinerant electron ferromagnet with spin-orbit coupling. To assess its
4 Ferromagnetism in (III,Mn)V Semiconductors 179
(4.15)
where j is the total angular momentum operator and /'1 and /'2 are the Lut-
tinger parameters [25,34]. In the unpolarized case (h = 0), the total Hamilto-
nian, H = Ho - hjz/3 (the external magnetic field is assumed to be in the +2
direction), is diagonalized by spinors Ih) where, e.g., Jt" == j.k = ±3/2 for the
two degenerate heavy-hole bands with the effective mass mhh = m/bl -2/'2)'
The corresponding Berry phase, J d 2kil(±3/2, k) = ±3/2(cosBk - 1), is
largest at the equator (cosB k == kz/k hh = 0) and vanishes at the poles
(I cos Bk I = 1) of the spherical Fermi surface of radius khh. Because of the
band degeneracy, the anomalous Hall conductivity (4.14) vanishes in the
h = 0 limit. The effective Zeeman coupling present in the ferromagnetic state
both modifies the Fermi surface shapes and renormalizes the Berry phases.
Up to linear order in h we obtain that k~h = khh±hmhh/(2n?khh) cosB k and
that the Berry phase is reduced (enhanced) by a factor [1=f2mh/(9/'2n2k~h)]'
A similar analysis for the light-hole bands leads to the total net contribution
to the AHE from the four bands whose lower and upper bounds are:
(4.16)
80
_._._._. ßso"-+co
60 . ....
- - - -- ß so =1eV .;
,--
,.'
,.'
40 -- ß so=341 meV ,.'
E
u
........ '
'9- 20
..
J:
t:)
.~.:::............
0
20
0 50 100 150 200
h (meV)
Fig. 4.7. Illustrative calculations of the anomalous Hall conductance as a function
of the band-splitting effective Zeeman field for hole density p = 0.35 nm -1. The
dotted-dashed curve was obtained assuming infinitely large spin-orbit coupling and
the decrease of theoretical aAR with decreasing spin-orbit coupling strength is
demonstrated for Llso = leV (dashed line) and Ll so = 341 meV (solid line)
80
p=O.1 nm-3
60
-
0.2 nm-3
,- 40 0.35 nm-3
E
,-u
9- 20
I
..:
\::)
0 (Ga,Mn)As
60 .,_ 80
,- 40 (In,Mn)As
(AI,Mn)As
E
u
,- o 100 200
9- 20 h(meV)
I
..: ...•.................................................
\::)
0
20
o 50 100 150 200
h (meV)
Fig.4.8. Full numerical simulations of (}" AH for GaAs host (top panel), InAs host
(bottom panel), and AlAs host (inset) with hole densities p = 0.1 nm -1 (dotted
lines), p = 0.2nm- 1 (dashed lines), and p = 0.35nm- 1 (solid lines). The filled
circles in the top and bottom panels represent measured ARE [45,5J values. The
saturation mean-field h values for the two points were estimated from nominal
sampie parameters [45,5J. Horizontal error bars correspond to the experimental
uncertainty of the Jpd coupling constant. The measured hole density in the
(Ga,Mn)As sampie is p = 0.35nm-\ for (In,Mn)As, p = 0.1 nm- 1 was determined
indirectly from the sample's transition temperature
also consistent with the observed positive sign and monotonie dependence of
GAH on sampIe magnetizations [5].
We take the agreement in both magnitude and sign of the AHE as a strong
indieation that the anomalous velo city contribution dominates the AHE in
homogeneous (III,Mn)V ferromagnets. This Berry phase term, which is in-
dependent of quasiparticle scatterers, is relatively easily evaluated with high
accuracy, enhancing the utility of the Hall measurement in sampIe charac-
terization. The success of this model also supports the use of the simple
mean-field approximation discussed in this section, in which Mn ions are
represented by a uniform density continuum, to describe at least the ground
state of these ferromagnets.
where p, denotes the chemical potential, and i and j range over a complete set
of hole-band states (i.e., here, for the model with two parabolic bands, i and j
label band wavevectors and spin, t,t), and sfj and s~ are matrix elements of
the itinerant-carrier spin matrices. The combination .:1 = NMnJpdS defines
the mean-field energy to flip the spin of an itinerant carrier. The physics of
the itinerant carriers is embedded in the effective action of the magnetic ions.
It is responsible for the retarded and non-local character of the interactions
between magnetic ions.
So far we have made no approximations. The independent spin-wave
theory is obtained by expanding (4.20) up to quadratic order in Z and z,
i.e., spin excitations are treated as noninteracting HP bosons. This is a good
approximation at low temperatures, where the number of spin excitations per
Mn site is small.
We obtain (in the imaginary time Matsubara and coordinate Fourier rep-
resentation) an action that is the sum of the temperature-dependent mean-
field contribution and a fluctuation action. The latter is
Sefdzz] =
1
ßV L z(k, vm)n-1(k, vm)z(k, vm). (4.23)
Ikl::::kD,m
(4.25)
where EF is the Fermi energy of the majority-spin band. In strong and weak-
coupling limits, Ll » EF and Ll « EF, respectively, (4.25) simplifies to
We note that the dependence of the spin-wave energy on the system param-
eters, namely the exchange interaction strength Jpd, hole concentration p,
local-impurity density N Mn , and effective mass m* is different in these two
limits, indicating that the microscopic character of the gapless collective exci-
tat ions differs qualitatively in the two limits. The energy of long-wavelength
spin waves is determined by a competition between exchange and kinetic ener-
gies. To understand this in more detail one can impose the spin configuration
of a static spin wave on the Mn spin system, evaluate the ground-state energy
of the itinerant-carrier system in the presence of the generated exchange field,
and compare this with the ground-state energy of a uniformly polarized state.
The results of this calculation are explained briefly below; for details see
[47]. Given the Mn spin configuration, the valence-band carriers can either
follow the spatial dependence of the Mn spin density in order to minimize the
exchange energy, as they do in the strong-coupling limit Ll » EF, or minimize
the kinetic energy by forming astate with a homogeneous spin polarization,
as they do in the weak-coupling limit Ll « EF. The corresponding energy
scales are provided by Ll and EF, i.e., the crossover from one regime to the
other is governed by the ratio LljEF.
186 J. König et al.
0.004
I
p=0.01nm-3
I
0.""
0.000
0.""
0.000
0.01
p=0.1nm-3
0.00
0.02
.... ...... mean field n MF
-~ ---- RKKY p=0.35nm-3
::::'0."" 0.01
- - spinwave nk
0.00 L_--,-==o=::===~~_-"---~~
0.0 0.2 0.4 0.6 0.8 1.0
kIk o
Fig.4.9. Spin-wave dispersion (solid lines) for Jpd = 0.06eVnm3 , m* = 0.5me,
NMn = 1 nm -3, and four different itinerant-carrier concentrations p = 0.01 nm -3,
0.035nm- 3 , 0.1 nm- 3 , and 0.35 nm- 3 . The ratio l1/€F is 2.79, 1.21,0.67, and 0.35,
which yields the fractional free-carrier spin polarization as 1, 1, 0.69, and 0.31. e
The short wavelength limit is the mean-field result gMF = xl1 (short-dashed lines),
and the long-dashed lines are the result obtained from an RKKY picture
1.05
1x
Fig. 4.10. Stoner excitations and optical spin-wave mode in the free-carrier system
for Jpd = O.06eVnm3 , m* = O.5me, NMn = 1 nm- 3 , and p = O.35nm- 3 . In an
RKKY picture these modes are absent
is then dispersionless, QMF = xL1 (short-dashed line in Fig. 4.9), and always
larger than the real spin-wave energy.
(4.29)
(4.30)
for (1, (1' = ±. The indices a and ß label the single-particle eigenstates for
valence-band carriers at a given wavevector q and q+k, and s!ß = (als±Iß).
The remaining task is to evaluate the fractional itinerant-carrier polarization
~ and the quantities Et-
and Et+ numerically.
In Fig.4.11 we show the spin-wave dispersion for wavevectors k along
the easy axis obtained using parameters valid for (Ga,Mn)As [5]. We observe
that the effect of Et+ in (4.29) is negligibly small and can, therefore, be
dropped. Furthermore, we find that the dispersion is fairly independent of
its wavevector direction, a property that is usually implicitly assumed in
micromagnetic descriptions of magnetic materials.
4 Ferromagnetism in (III,Mn)V Semiconductors 189
0.06
- - - - mean field n MF
0.04
- - spin wave n.
<I
cl
0.01
0.02
0.cXl1
0.00 ........:::..-----'--~--'--~--'--~----'-~--'
0.00 .2 0.4 0.6 0.8 1.0
k I ko
Fig.4.11. Main panel: Spin-wave dispersion for the 6-band model for itinerant-
carrier density p = 0.35 nm -3, impurity-spin concentration NMn = 1.0 nm -3 and
exchange coupling Jpd = 0.068eVnm- 3 . Inset: Spin-wave dispersion on a log-log
plot (eircles) and the parabolic fit (solid line)
where K is the anisotropy energy constant, and Adenotes the spin stiffness
or exchange constant. While the anisotropy constant can be obtained from
the mean-field energy for different magnetization orientations (see previous
section), the virtue of the spin-wave calculation is to extract the spin stiffness
as weIl.
In Fig. 4.12 we show the spin stiffness A as a function of the itinerant-
carrier density for two values of Jpd for both the isotropie two-band and the
full six-band model. We find that the spin stiffness is much larger for the six-
band calculation than for the two-band model. Furthermore, for the chosen
range of itinerant-carrier densities the trend is different: in the two-band
model the exchange constant decreases with increasing density, while for the
six-band deseription we observe an inerease with a subsequent saturation.
To understand this behavior we recall that the two-band model predicts
a different dependenee of A on p in the strong and weak-coupling limits with
a crossover near .d '" EF, see (4.26) and (4.27). The difference in the trends
seen for the two- and six-band model in Fig. 4.12 is explained in part by the
observation that, at given itinerant-carrier concentration p, the Fermi energy
EF is mueh smaller when the six-band model is employed, where more bands
are available for the carriers, than in the two-band case. Furthermore, we
emphasize that, even in the limit of low carrier concentration, it is not only
the (heavy-hole) mass of the lowest band which is important for the spin
190 J. König et al.
- - ---
1.2
- _0--
--
..0-
1.0
D'
,,
,,
0.8
,0" ° -() 6band model, large Jpd
'E ,,
0--0 6band model, small Jpd
""") 0.6
, ,, o - .[J 2band model, large Jpd
..e,
,, 0--0 2band model, small Jpd
«
0.4
0.2 ~
--- - - - - 0 - - - - - - - 0 - - - 0 - - - 0 _______
0.0
0.1 0.2 0.3 0.4 0.5
P [nm-3)
Fig.4.12. Exchange constant A as a function of itinerant-carrier density p for the
six-band and the two-band model for two different values of Jpd = 0.068 eV nm- 3
(solid lines) and 0.136eVnm- 3 (dashed lines). The inipurity-spin concentration is
chosen as NMn = 1.0nm- 3 , which yields.:1 = 0.17eV (solid lines) and.:1 = 0.34eV
(dashed lines), respectively
stiffness. Instead, a collective state in which the spins of the itinerant carriers
follow the spatial variation of a Mn spin-wave configuration will involve the
light-hole band, too. Our calculations show that accounting for the presence
of this second more dispersive band is essential to understanding the suc-
cess of mean-field theory. Crudely, the large mass heavy hole band dominates
the spin-susceptibility and enables loeal magnetie order at high temperatures,
while the dispersive light hole band dominates the spin stiffness and enables
long range magnetie order. The multi-band character of the semiconductor
valence plays an essential role in the ferromagnetism of these materials.
(4.32)
they will dominate the suppression of the magnetization at all finite tem-
peratures and limit the critical temperature. In this case, the typical local
valence-band carrier polarization remains finite above the critical tempera-
ture. Ferromagnetism disappears only because of the loss oflong-range spatial
coherence.
A rough upper bound on the critical temperature Tco ll can be obtained by
the following argument which accounts for the role of collective fluctuations
[48]. The magnetization vanishes at the temperature where the number of
excited spin waves equals the total spin of the ground state.
N Mn 8 = --;
27r Jo
r kD
dk k 2 n(fh) , (4.33)
where n(rlk ) is the Bose occupation number and the Debye cutoff, k D =
(67r 2 N Mn ) 1/3, ensures the correct number of magnetic ion degrees of free-
dom. We therefore find that the critical temperature of a ferromagnet cannot
exceed
I mean field
<I
0.1
0.001 0.01 0.1
pi NMn
Fig. 4.13. Critical-temperature-limit regimes for the two-band model. In the mean-
field regime Tc is limited by individual Mn spin fluctuations. In the collective
regimes, the critical temperature is limited by long-wavelength fluctuations with
a stiffness proportional to the bandwidth for weak (RKKY) exchange coupling and
inversely proportional to the bandwidth for strong exchange coupling. At the solid
line TlfF = Tg ll . Dashed lines: expansions for large and small .::lIEF, (4.35) and
(4.36), and the crossover from the RKKY to the strong coupling collective regime
We expect that the qualitative picture derived from the two-band model
will persist for the six-band model. The actual values of the boundaries
between the regimes indicated in Fig. 4.13 will, however, be shifted because
of the important differences in the microscopic physics that determines the
spin stiffness of the two models discussed above. We expect that the Tc
estimates derived in the preceding paragraph will be directly applicable to
n-type carrier-mediated ferromagnets . As a consequence we expect that it will
be impossible to achieve Zarge jerromagnetic transition temperatures in n-type
semiconductors with carrier mediated jerromagnetism. The rv 5% reduction
of (Ga,Mn)As mean-field Tc due to spin fluctuations, mentioned in Sect. 4.1,
was obtained using the six-band model and solving self-consistently (4.34) and
D(TcO Il ) = D(T = 0) (S)(TCOll)jS. Larger reductions compared to mean-field-
theory estimates are expected in some hosts, but mean-field-theory retains
a qualitative validity.
resulting from the growth process of such materials, is that the Mn acceptors
are distributed at random on the cation sites of the underlying crystallattice.
To this point in this chapter, we have used a continuum approximation for
the Mn ion distribution that yields a disorder-free problem. The continuum
approximation makes it possible to obtain some results analytically, and also
crucially simplifies numerical calculations. However, it neglects substitutional
disorder in the Mn positions which can have a substantial impact on the
ferromagnetism. That this is true is evident from the fact that magnetic
properties of presently available sampIes are often sensitive to the conditions
of their fabrication, and reproducibility is achieved only if the growth pa-
rameters are carefully controlled [5]. Moreover, recent studies of post-growth
annealed (Ga,Mn)As sampIes have revealed that the magnetic [52] as well as
the structural [53] properties can depend crucially on the type of defects and
disorder present in the system.
The spin-wave theory presented in the previous section describes collective
excitations of the ion spin system (modeled as a continuum) in terms of
Gaussian fluctuations around the ferromagnetic ground state in the many-
body path integral. This non-interacting spin-wave theory is exact at low
temperatures, where deviations from the ordered ground state are small,
but is less reliable when the temperature is raised toward the ferromagnetic
transition temperature.
In this section we complement the theoretical approaches described above
by methods which (i) take disorder effects due to the randomly chosen ion
positions into account and (ii) are able to address directly the region of larger
deviations of the spin configuration from the ferromagnetically ordered state.
Specifically, we present results of Monte Carlo simulations [48,49] which treat
the minimal model without approximation. Finally we report on a rigorous
stability analysis of the perfectly ferromagnetically ordered collinear state
of Mn ions in the presence of disorder[54]. We predict that noncollinear
jerromagnetism is common in (III, Mn) V semiconductors; the robust collinear
ferromagnetic states that have the highest ferromagnetic transition temper-
atures occur only when the carriers are relatively weakly localized around
individual Mn ions.
extracted from first principles electronic structure calculations. Note that the
exchange-coupling range parameter ao in (4.37) is required in our calculations
once the discreteness of the Mn ions is acknowledgedj exchange-coupling shifts
of quasiparticle energies would diverge otherwise. Given this finite range, the
only approximation we make below in treating the minimal model is that we
treat the Mn spins classically. Because of the relatively large value of the Mn
ion spins, this approximation should have minimal consequence except for
the leading low temperature magnetization suppression.
We are interested in thermal expectation values of the form
1= ~ 1 1"
2
" dcp d'l? sin 'I? Tr {!( 'I?, cp )e- ß1l } , (4.38)
where ß is the inverse temperature, Z the partition function, and 'I?, cp are
shorthand notations for the whole set of classical spin coordinates. The quan-
tity !('I?, cp) is a function of the ion spin angles and an operator with respect
to the quantum mechanical carrier degrees of freedom over which the trace is
performed. In practice we replace the fermion trace by a ground state expec-
tation value, since the temperatures of interest will always be much smaller
than the Fermi energy. For typical carrier densities p of order 0.1 nm -3, the
Fermi temperature for the carriers is typically larger than 1000 K, compared
to ferromagnetic critical temperatures rv 100 K. Thermal effects in the carrier
system are therefore negligible. Thus,
1= ~ 1 1"
2
" dcp d'l? sin 'I? (01!('I?,cp)10)e- ß (OI1l 10) , (4.39)
where 10) denotes the groundstate of non-interacting fermions with the ap-
propriate band Hamiltonian and a Zeeman-coupling term h whose effective
magnetic field Heff is due to exchange interactions with the localized spins,
h = gJ.1-B J
d 3 rs(r). Heff(r),
where s1I = (sin (h cos <PI, sin (h sin <PI, cos l'h) is the direction of the classical
spin at RI. In the following we denote thermal expectation values of quant i-
ties defined in terms of classical spin orientation variables by (.) and quantum
mechanical expectation values within the carriers ground state by (01·10).
4.41
( )
2m* 2
I
Le. the thermally averaged modulus of the total ion spin, along with the
carrier magnetization,
1
m = pV(I(OlstotIO)I), (4.43)
which is the ensemble average of the modulus of the total ground-state carrier
spin. Both quantities are divided by the number of partieles and are elose to
their maximum values at low temperatures. At higher temperatures they
show the expected transition to a paramagnetic phase. The critical tempera-
ture of this ferromagnetic transition is most readily estimated from numerical
results for the magnetization fluctuations:
These two fluctuations per partiele are plotted in the insets of Fig.4.14.
They both show a pronounced peak at a temperature T rv 100 K, defining
the finite-system transition temperature for these model parameter values. In
fact, in a region around this transition the two data sets differ by a factor of
approximately 25, which is the square of the ratio of the two spin lengths en-
tering the expressions (4.44) and (4.45), respectively. This observation shows
explicitly that the correlation length is the same for Mn ions and the carrier
system near the transition, that is both approach the finite system size of the
simulation.
Our Monte Carlo approach elearly reproduces the expected ferromagnetic
transition. Ferromagnetism still occurs when mndomness in the Mn positions
is accounted for. The transition temperature Tc can be determined unam-
biguously and consistently from the positions of very pronounced peaks in
total magnetization fluctuations of both the Mn ions and the carriers.
Results for Tc- We now turn to the transition temperature Tc for the
two-band model. Within mean-field theory this quantity is given by (4.6).
4 Ferromagnetism in (III,Mn)V Semiconductors 197
3.0
3
2.5 9Mn 2
c: 2.0 0
0
~
(/) 0.15
.~
1.5 Mn ions 9p 0.10
~
c:
&1.0 0.05
Temperature [Kelvins]
Fig.4.14. Magnetization curves for Mn ions and carriers. The upper and lower
inset show the magnetic fluctuations for the Mn ions and the carriers, respectively.
Both differ by a factor of approximately 25 reflecting the square of the ratio of
spin lengths. The density of Mn ions is NMn = 1.0 nm -3, the carrier density is
p = 0.1 nm -3 in a cubic volume of V = 540 nm 3 . The band mass is half the bare
electron mass with an exchange parameter of Jpd = 0.15 eV nm 3
~ Simulation
Monte Carlo
Simulation
Fig. 4.15. The critical temperature Tc as a function ofthe carrier mass (left panel)
and the carrier density (right panel). The exchange parameter is in both cases
Jpd = 0.2eVnm3 leading to a zero temperature mean-field spin splitting of Ll =
JpdNMnS = 0.5 eV. The results of the Monte Carlo runs are compared with the
mean-field predictions
expect the electrons to behave more classically and localize around individual
Mn spins, suppressing long-range order further. In this limit the electronic
energy cast of changes in the relative orientations of nearby Mn spins will get
smaller causing Tc to decline, and eventually ferromagnetism will disappear.
This is consistent with the observation that, within the continuum model, the
spin stiffness declines as 1/m* for large band masses [46-48]. As mentioned
earlier, the spin stiffness tends to be remain substantially larger when coupled
light and heavy holes are retained in the calculation. Still, this calculation
demonstrates that mean-field-theory must be regarded with some caution
and its validity must be checked in each new circumstance.
To discuss the critical temperature as a function of the exchange coupling
parameter Jpd, we observe that the Hamiltonian of itinerant carriers satisfies
the scaling relation
(4.46)
~ 0.10
~ 1.5
~ Mn ions 9p 0.05
c:
~ 1.0
o 100
Temperature [Kelvins]
0.5
0.0 L-~_,---~----,c.......>o,-=""-'_~----,_~--,
Fig.4.16. Magnetization curves for Mn ions and carriers in the six-band model
for n exchange coupling of Jpd = O.15eVnm 3 . The carrier density is p = O.lnm- 3
with an Mn ion density of NMn = l.Onm- 3 in a volume of V = 280nm3 . As in
the case of parabolic bands, the ferromagnetic transition in clearly and consistently
signaled by pronounced peaks in the magnetic fluctuations shown in the insets
typical distance between nearby Mn ions will become larger than the band
electron Fermi wavelength, causing the sign of the typical exchange coupling
to oscillate in an RKKY fashion. As we shall see in the next subsection, the
resulting frustration makes the ferromagnetic state unstable, possibly leading
to a regime of spin-glass order. Disordered states can also occur when the
exchange coupling becomes strong.
Six-Band Model. We now turn to the six-band model in which the kinetic
energy part is given by the Kohn-Luttinger model. Figure 4.16 shows typical
magnetization data for the (Ga,Mn)As system, assuming exchange coupling
Jpd = O.15eVnm3 , carrier density is p = O.1nm- 3 , and Mn ion density
NMn = l.Onm- 3 in a volume of V = 280nm3 . As in the case of parabolic
bands, a ferromagnetic transition is c1early signaled by pronounced peaks
in the magnetic fluctuations of both Mn ions and carriers. We find that,
in contrast to the parabolic two-band model, the carrier magnetization is
already reduced at temperatures well below Tc. In fact, because of strong
spin-orbit coupling in the valence band, full polarization of the carrier spins
never occurs. Another difference compared to the parabolic-band model con-
cerns the shape of the magnetic fluctuations for the Mn ions and the carriers
as a function of temperature. Although both curves indicate the same value
for Tc for these parameter values , their shape in the vicinity of Tc is slightly
different and the ratio of their fluctuations is smaller than 25 (the square
of the ratio of spin lengths involved). These differences arise because of
the more complicated band structure and the spin-orbit coupling present
200 J. König et al.
N M"=1.0nm -3 NM"=1.0nr -3
p=O.1nm~3 p=O.1nm
W m*=O.5m
:2 150
~
/-"
Sixband
model
parabolic
two-band model
_V=280nm 3
_ _ V=360nm 3
0.5
0.0
05 o 100 150 200
Temperature [Kelvins)
of Fig. 4.14 are plotted. Those five datasets are hardly distinguishable from
each other. Positional disorder in the localized magnetic moments of Mn ions
can however affect the nature of the ground state itself in some circumstances.
As seen from Fig. 4.14, the ion spin magnetization seems to saturate at low
temperatures at a value slightly smaller than its maximum, given by the
Mn spin length of 5/2. Moreover, as shown in Fig. 4.18, this behavior occurs
quite consistently for different disorder realizations. As we see below, the
effect is due to the randomness in Mn positions which, combined with the
RKKY-like oscillations in the effective coupling between Mn spins, makes the
perfectly ferromagnetically ordered collinear state state unstable and leads
to noncollinear jerromagnetism.
We now outline a theory of magnetic fluctuations around a given state
within a path integral formalism similar to the spin-wave approach presented
in Sect.4.5.2 for fluctuations around the collinear ferromagnetic state. In
contrast to the spin-wave calculations, here we do not approximate the Mn
magnetic moments by a continuum but retain them, as in the Monte Carlo
approach, as individual spins of length S = 5/ 2 placed at arbitrary locations.
Let us first consider magnetic fluctuations around the perfectly ferromagnet-
ically ordered collinear state having all Mn spins oriented in parallel.
Repeating the steps described in Sect.4.5, but keeping the Mn spins as
individual objects, one obtains the following expression for the fluctuation
part of the effective action in up to second order in the bosonic spin variables:
(4.47)
202 J. König et al.
K IJ -- ~ 'L
" [f(1}oJ
· - f(1}ß) Fa-l-,ßtFßt,a-l-]
I J . (4.50)
2 a, ß m + 1}a - 1}ß
ZV
Here (s(r)) is the expectation value of the carrier spin density, f the Fermi
function, and
(4.51)
with 'Waa(r) being the spin component U of the carrier wave function with
label 0: and energy Ca = 1}a + Jk, where Jk is the chemical potential for the
carriers All quantities referring to the carrier system are to be evaluated for
the collinear orientation of Mn spins.
The zero-frequency (m = 0) contribution to the effective action (4.47)
describes the energy of static fluctuations around the collinear state. For this
state to be stable, the matrix D1J(0) must have non-negative eigenvalues
only, while the occurrence of negative eigenvalues of this matrix indicates
that the perfectly collinear state is not the ground state. This ihterpretation
of the zero-frequency fluctuation term is confirmed by the observation that
this contribution is also obtained (at zero temperature ) by a standard per-
turbation theory for the carrier ground state energy with respect to small
deviations of the ion spins from collinear orient at ion. The formalism given
above embeds this finding in a more general theory of dynamic fluctuations at
finite temperature. However, we shall concentrate in the following on static
ground state properties, i.e. on T = 0, where the Fermi functions become
step functions.
We note that for any arrangement of the Mn positions R I , the matrix
D1J (0) contains a zero eigenvalue corresponding to a uniform rotation of
all spins. The eigenvalues of D1J(0) are proportional to magnetic excitation
energies. In this sense the eigenvalue distribution of D1J(0) can be interpreted
as a density of states for magnetic excitations.
We have evaluated the spectrum of D1J(0) in systems given by a sim-
ulation cube with periodic boundary conditions averaging over different re-
alizations of the Mn positions. The single-particle wavefunctions 'Waa(r) are
computed in a plane-wave basis taking into account wavevectors q with length
up to an appropriate cutoff qc. The same truncated plane-wave basis is used
to compute the quantities (4.51) entering (4.50). Note that, far fluctuations
4 Ferromagnetism in (III,Mn)V Semiconductors 203
around the collinear ferromagnetic state, Di](iw) is always real and symmet-
rie for real w since a11 carrier wavefunctions have, for a given spin projection
Cf, the same coordinate-independent phase. This follows from the fact that
the single-particle Hamiltonian describes for each spin projection just the
problem of a spinless particle in a potential landscape provided by the Mn
ions. Since Di] (iw) is real and symmetrie, the components of each of its
eigenvectors a11 have the same phase (and can be chosen to be real). Physica11y
this corresponds to the invariance of the system under rotations around the
magnetization axis of the collinear state.
The two upper panels of Fig. 4.19 show results for typical system param-
eters for two different values of qc. Comparison of the panels shows that
the effects of the wavevector cutoff on the low-lying excitations have already
saturated for the sma11er qc. Almost a11 eigenvalues of Di](O) lie at positive
energies, but there is often a small fraction of negative eigenvalues, indicating
an instability of the co11inear state.
In the calculations discussed so far, the Mn positions were chosen com-
pletely at random with a uniform distribution, while in areal (III,Mn)V
8.0
V=400nm -3
6.0 N~= 1. 0 nm-3
,. 4.0
p= .15nm
m"=O.5m
>Q) 2.0 Jpd=0 .05eVnm
3
E
0 0.0
~
<fl ~=3(2 7t1L)
Q) 6.0
:::J
~
c::
4.0
Q)
Ol 2.0
iIi 0.0
Ö
~ 6.0
'üi fee laltice
c::
Q) 4.0
Cl
2.0
0.0
--(l.OO2 0.000 0.002 0.004 0.006
Energy [eV]
(4.52)
o
-0.002 o 0.002 0.004 0.006
Energy [eV)
levels that do not play an important role at weak fields in these highly
disordered sampies. Zeeman coupling is also unimportant, yielding a con-
tribut ion to the spin-splitting that is negligible compared to the mean-field
splitting..1 = JpdNMnS '" 0.1 eV. It follows that the most negative eigenvalue
of DIJ is the value of gLJ.LBHext necessary to force full spin alignment of
a noncollinear state. The experiments of Potashnik et al. [52] demonstrate
that the maximum value of M(T = 0) is achieved over a certain range of
annealing histories. We associate this maximum value with the fully aligned
collinear Mn configuration state; indeed the maximum moment per Mn is
consistent with full alignment partially compensated by band electrons. For
other annealing histories, M(T = 0) is reduced, corresponding to noncollinear
order of the Mn spins. Calculations like those described above show that
these spins gradually align as an external field is added to the Hamilto-
nian. We expect full alignment to be indicated experimentally by a kink in
the M(T = 0, H exd curve. At this point, we predict that the system will
still have gapless excitations and power-law temperature dependence of the
magnetization, in sharp contrast to the gapful excitations and exponentially
suppressed temperature dependence that holds for conventional ferromagnets
in an external magnetic field. Spin-resonance experiments will nevertheless
see a gaped spin-wave spectrum in this regime, since they couple only to the
zero mode.
Finally we comment on the degree of spin alignment in the noncollinear
state. We generalize our formalism by expanding around and defining Hol-
stein-Primakoff bosons with respect to a spin coherent state configuration
n
with orientations I = (sin ß I cos 'PI, sin ß I sin 'PI, cos ß I). In this general case
the effective action has a static contribution (at zero Matsubara frequency
only) linear in the Holstein-Primakoff variables:
(4.53)
206 J. König et al.
Acknowledgements
The work presented in this chapter was done in collaboration with a num-
ber of researchers. M. Abolfath, W.A. Atkinson, J. Brum, and Byounghak
Lee made important contributions to the mean-field theory of ferromagnetic
critical temperature and magnetic anisotropy in DMS's, H.H. Lin has been
active in the development of the spin wave theory, Qian Niu's expertise
on the semiclassical transport theory has been essential for the anomalous
Hall effect study, J. Sinova has been working on correlation effects in the
itinerant carrier system, and S.-R.E. Young contributed to the study of
disordered DMS's. We thank D.D. Awschalom, D.V. Baxter, R.N. Bhatt,
4 Ferromagnetism in (III,Mn)V Semiconductors 209
References
1. H. Ohno: Science 281, 951 (1998); F. Matsukura, H. Ohno, A. Shen, and Y.
Sugawara: Phys. Rev. B 57, R2037 (1998)
2. J.M. Kikkawa and D.D. Awschalom: Nature 397, 139 (1999); D.D. Awschalom
and N. Samarth: J. Magn. Magn. Mater. 200, 130, (1999); I. Malajovich, J.J.
Berry, N. Samarth, and D.D. Awschalom: Nature 411, 770 (2001)
3. J.M. Daughton: J. Appl. Phys. 81, 3758 (1997); J.S. Moodera and G. Mathon:
J. Magn. Magn. Mater. 200, 248 (1999); J. Bass and W.P. Pratt: J. Magn.
Magn. Mater. 200, 274 (1999); K. O'Grady and H. Laidler: J. Magn. Magn.
Mater. 200, 616 (1999)
4. S. Sonoda, S. Shimizu, T. Sasaki, Y. Yamamoto, and H. Hori: cond-
mat/0108159
5. H. Ohno: J. Magn. Magn. Mater. 200, 110 (1999)
6. Y. Tokura and Y. Tomioka: J. Magn. Magn. Mater. 200, 1 (1999); E. Dagotto,
T. Hotta, and A. Moreo: Physics Reports 344, 1 (2001)
7. T. Dietl, H. Ohno, and F. Matsukura: Phys. Rev. B 63, 195205 (2001)
8. F. Matsukura and T. Dietl, to be published
9. J. Szczytko, A. Twardowski, K. Swiatek, M. Palczewska, M. Tanaka, T.
Hayashi, and K. Ando: Phys. Rev. B 60, 8304 (1999)
10. M. Linnarsson, E. Janzen, B. Monemar, M. Kleverman, and A. Thilderkvist:
Phys. Rev. B 55, 6938 (1997)
11. J. Okabayashi, A. Kimura, O. Rader, T. Mizokawa, A. Fujimori, T. Hayashi,
and M. Tanaka: Phys. Rev. B 58, R4211 (1998)
12. J. Okabayashi, A. Kimura, T. Mizokawa, A. Fujimori, T. Hayashi, and M.
Tanaka: Phys. Rev. B 59, R2486 (1999)
13. J. Okabayashi, A. Kimura, O. Rader, T. Mizokawa, A. Fujimori, T. Hayashi,
and M. Tanaka. Phys. Rev. B 64, 125304 (2001)
14. H. Ohno, F. Matsukura, A. Shen, Y. Sugawara, A. Oiwa, A. Endo, S. Kat-
sumoto, and Y. Iye, in Proceedings 01 the 23rd International Conlerence on
the Physics 01 Semiconductors, Berlin 1996, edited by M. SchefHer and R.
Zimmermann (World Scientific, Singapore 1996) p.405
15. T. Hayashi, H. Shimada, H. Shimizu, and M. Tanaka: J. Cryst. Growth
201/202, 689 (1999)
16. S. Sanvito and N.A. Hill: Phys. Rev. B 62, 15553 (2000); S. Sanvito, P. Or-
dej6n, and N.A. Hill: Phys. Rev. B 63, 165206 (2001); S. Sanvito and N.A.
Hill: cond-mat/0108406; M. van Schilfgaarde and O.N. Mryasov, Phys. Rev.
B 63, 233205 (2001)
17. H. Akai, Phys. Rev. Lett. 81, 3002 (1998)
210 J. König et al.
18. J.H. Park, S.K Kwon, and B.1. Min: Physica B 281-283, 703 (2000)
19. J. Kossut: Phys. Stat. Solidi 78, 537 (1976)
20. J.K Furdyna and J. Kossut, Diluted Magnetic Semiconductors, Vol. 25 of Semi-
conductor and Semimetals (Academic Press, New York, 1988)
21. B.E. Larson, KC. Hass, and H. Ehrenreich: Phys. Rev. B 37, 4137 (1988)
22. T. Dietl, Diluted Magnetic Semiconductors, Vol. 3B of Handbook of Semicon-
ductors, (North-Holland, New York 1994)
23. M. Berciu and R.N. Bhatt: Phys. Rev. Lett. 87, 107203 (2001)
24. A.1. Chudnovskiy, cond-matj0108396
25. W.W. Chow, S.W. Koch, and M. Sargent III, Semiconductor Laser Physics,
(Springer, Berlin Heidelberg New York 1999) p. 179-192
26. M. Abolfath, T. Jungwirth, J. Brum, and A.H. MacDonald: Phys. Rev. B 63,
054418 (2001)
27. T. Omiya, F. Matsukura, T. Dietl, Y. Ohno, T. Sakon, M. Motokawa, and H.
Ohno: Physica E 7, 976 (2000)
28. A.K Bhattacharjee and C. Benoit a la Guillaume, Solid State Comm. 113, 17
(2000)
29. O.M. Fedorych, E.M. Hankiewicz, Z. Wilamowski, and J. Sadowski: cond-
matj0106227
30. A. Chattopadhyay, S. Das Sarma, and A.J. Millis: cond-matj0106455
31. The role ofrandomness in a variety of models where spins are coupled to band
electrons has been reviewed by A.A. Abrikosov: Advances in Physics 29, 869
(1980)
32. T. Jungwirth, W.A. Atkinson, B.H. Lee, and A.H. MacDonald: Phys. Rev. B
59, 9818 (1999)
33. A. Aharoni, Introduction to the Theory of Ferromagnetism (Oxford University
Press, New York 1996)
34. I. Vurgaftman, J.R. Meyer, and L.R. Ram-Mohan: J. Appl. Phys. 89, 5815
(2001)
35. T. Dietl, H. Ohno, F. Matsukura, J. Cibert, and D. Ferrand: Science 287,1019
(2000)
36. T. Jungwirth and A.H. MacDonald: Physica E 10, 153 (2001)
37. T. Jungwirth, J. König, and A.H. MacDonald, to be published
38. H. Munekata, A. Zaslavsky, P. Fumagalli, and R.J. Gambino: Appl. Phys. Lett.
63, 2929 (1993)
39. R. Skomski and J.M.D. Coey, Permanent Magnetism (Institute of Physics
Publishing, Bristol 1999)
40. T. Jungwirth, Q. Niu, and A.H. MacDonald, cond-matjOll0484
41. J. Smit: Physica 23 39 (1958)
42. L. Berger: Phys. Rev. B 2, 4559 (1970)
43. G. Sundaram and Q. Niu: Phys. Rev. B 59, 14915 (1999)
44. J.M. Luttinger: Phys. Rev. 112, 739 (1958)
45. H. Ohno, H. Munekata, T. Penney, S. von Molnar, and L.L. Chang: Phys. Rev.
Lett. 68, 2664 (1992)
46. J. König, H.H. Lin, and A.H. MacDonald: Phys. Rev. Lett. 84, 5628 (2000);
Physics E 10, 139 (2001)
47. J. König, H.H. Lin, and A.H. MacDonald, in Interacting Electrons in Nanos-
tructures. Eds. R. Haug and H. Schoeller, Lecture Notes in Physics 579
(Springer, Berlin Heidelberg New York 2001) p.195-212
4 Ferromagnetism in (III,Mn)V Semiconductors 211
48. J. Schliemann, J. König, H.H. Lin, and A.H. MacDonald: Appl. Phys. Lett.
78, 1550 (2001)
49. J. Schliemann, J. König, and A.H. MacDonald: Phys. Rev. B 64,165201 (2001)
50. J. König, T. Jungwirth, and A.H. MacDonald: Phys. Rev. B 64,184423 (2001)
51. A. Auerbach, Interacting Electrons and Quantum Magnetism (Springer, New
York 1994)
52. S.J. Potashnik, K.C. Ku, S.H. Chun, J.J. Berry, N. Samarth, and P. Schiffer:
Appl. Phys. Lett. 79, 1495 (2001)
53. G.M. Schott, W. Faschinger, and L.W. Molenkamp, cond-mat/0105562
54. J. Schliemann and A.H. MacDonald, cond-mat/0107573
55. S. Duane, A.D. Kennedy, B.J. Pendleton, and D. Roweth: Phys. Lett. B 195,
216 (1987)
56. J. König, J. Schliemann, and A.H. MacDonald, unpublished
57. Byuonghak Lee, T. Jungwirth, and A.H. MacDonald: Phys. Rev. B 61, 15606
(2000)
58. H. Ohno, D. Chiba, F. Matsukura, T. Omiyama, E. Abe, T. Dietl, Y. Ohno,
and K. Ohtani: Nature 408, 944 (2000)
59. Byounghak Lee, T. Jungwirth, and A.H. MacDonald, cond-mat/0106536
60. S.A. Crooker, D.A. Tulchinsky, J. Levy, D.D. Awschalom, R. Garcia, and N.
Samarth: Phys. Rev. Lett. 75, 505 (1995)
61. R.K. Kawakami, E. Johnston-Halperin, L.F. Chen, M. Hanson, N. Guebels,
J.S. Speck, A.C. Gossard, and D.D. Awschalom: Appl. Phys. Lett. 77, 2379
(2000)
62. H. Luo, B.D. McCombe, M.H. Na, K. Mooney, F. Lehmann, X. Chen, M.
Cheon, S.M. Wang, Y. Sasaki, X. Liu, J.K. Furdyna, Proceedings of the 14the
International Conference on the Electronic Properties of Two-Dimensional
Systems, Prague (2001)
5 N oncollinear Magnetism in Systems
with Relativistic Interactions
L. Sandratskii
5.1 Introduction
In an early formulation of the DFT by von Barth and Hedin [20], the energy
of the system is treated as a functional of the two-by-two density matrix.
Correspondingly, the Kohn-Sham equation (KSE) and the effective electron
potential have matrix form. The consequent consideration of the matrix char-
acter of the density and effective potential allows the DFT description of sys-
tems with noncollinear magnetization. However, for many years the concrete
calculations were performed for collinear magnetic systems only. In this case
the density and potential matrices are assumed to have a diagonal form. In
the problems where the SOC is not taken into account, the neglect of the
off-diagonal elements of the potential matrix leads to the spin-factorization
of the problem: the spin projection of the electron states on the global z-
axis is a good quantum number and there are two separate equations for
the electron states with opposite spin projections. Such states being used to
calculate new density and potential matrices lead automatically to vanish-
ing off-diagonal elements. Thus, if the SOC is negligible, the collinearity of
214 L. Sandratskii
(5.1)
i,occ
After integration of (5.1) over atomic spheres one gets the atomic matrices
[22]. Diagonalization of the atomic matrices gives the directions of the atomic
magnetic moments corresponding to these matrices. In general, calculated
directions of the atomic moments are different from the directions used in
the formulation of the KSH at the beginning of the iteration. In the self-
consistent state, the initial and calculated directions of atomic moments
coincide. If the intraatomic noncollinearity of the spin magnetization is taken
5 Noncollinear Magnetism in Relativistic Systems 215
The relation between the symmetry of a system and its magnetic structure
has attracted much attention in the history of solid-state magnetism. Much
consideration was given to the prediction of the magnetic structures that
can appear in the system as the result of a continuous phase transition (see,
e.g. [24,25]). We will focus on another aspect of the relation between the
symmetry and magnetic structure: the stability of a given magnetic configu-
ration, independent of the kind of phase transition into the magnetic state.
Applications of the DFT to the studies of complex noncollinear magnetic con-
figurations have shown that the magnetic structure chosen at the beginning
of the DFT calculation is, in general, unstable: the magnetic moments deviate
in the course of iterations from the initial directions, tending to form another
magnetic state. On the other hand, in some cases the magnetic moments,
although allowed to move, keep their initial directions. The ability to predict
to which of the two types of structures a given magnetic configuration belongs
is an important capability for the study of the magnetism of the system.
In [3,11,26,27] it was shown that there exists an intimate connection be-
tween the stability of the magnetic structure in the DFT calculations and the
symmetry of the system. Here, we will generalize the approach suggested in
[3,11,26,27] by using the not ion of symmetry constraint and apply it to the
studies of the properties of various magnetic structures.
/1-1) LI LI
calculated with the use of densities n and m is also invariant with respect
to g. As a result, the KSH for the next iteration, which uses the calculated
effective potential (5.4), is again, as the initial one, invariant with respect to
operations of G.
Thus we have shown that the densities obtained in the calculations are
invariant with respect to the symmetry operations of the initial KSH and
any symmetry operation of the initial KSH is preserved in the calculations.
Since only the densities invariant with respect to operations of G appear
in the calculations, one deals with a constrained minimization of the total
energy considered as a functional of the densities. We will refer to this type
of restrictions on the densities as symmetry constraint.
A general approach to a constrained minimization of the energy as a func-
tional [29,30] of the charge and magnetic densities requires adding to the
functional the following term
This term contains Lagrange parameters p( r) and b( r ), which play the role
of external fields stabilizing the constrained state.
5 Noncollinear Magnetism in Relativistic Systems 217
(5.5)
After integration of the magnetization over atomic spheres we get the restric-
tion
(5.6)
imposed on the atomic magnetic moments, where i and j label the atoms
defined by the relation
(5.7)
Therefore the atoms that are transformed one into another by g possess
magnetic moments of equal magnitude, and the direction of one moment is
transformed into the direction of another under the action of g. In the case
when the position of an atom is unchanged under the action of g, (5.6) takes
the form
218 L. Sandratskii
(a) (b)
symmetry constraint I symmetry constraint II
and imposes a restriction on the moment of this atom, which consists of the
invariance of the moment with respect to g. If operation 9 contains time
reversal, (5.6) is modified as follows
(5.8)
The restrictions (5.6-5.8) on the lengths and directions of the atomic
magnetic moments can be considered as regular features (regularities) of the
magnetic structure that are the necessary consequences of a given symmetry
constraint. Two different situations can follow from relations (5.6-5.8). In
the first case, the symmetry constraint determines the magnetic structure
uniquely. This means that any deviation of the magnetic moments from the
initial directions disturbs, at least, one of the symmetry operations. Since all
symmetry operations must be preserved the structure cannot change in the
course of calculations. We will refer to this type of constraint as symmetry
constraint I (Fig.5.1).
In the second case (symmetry constraint II) there is an infinite set of
magnetic configurations that, first, satisfy the conditions (5.6-5.8) imposed by
the invariance with respect to G and, second, can be continuously transformed
into one another without disturbing the symmetry of the system. Let (j be
a continuous parameter that describes this set of magnetic configurations.
(The number of parameters can be larger than one, but this does not change
the essence of the arguments.) As all magnetic configurations are described
by the same symmetry none of the (j values is distinguished. The purpose of
the DFT calculations in this case is to find the value of (j that corresponds
to the state with the lowest energy. Since all (j values are equivalent, this
minimum cannot be predicted without calculations (Fig.5.1).
To begin the DFT calculation a value (je of parameter (j is selected. Since
it is improbable that (je accidentally equals (jmin, providing the minimum of
5 Noncollinear Magnetism in Relativistic Systems 219
the total energy, the initial state in the case of symmetry constraint 11 is
unstable. In the iterational process, the magnetic structure deviates from the
state described by 8 tending to assurne the state with the lowest energy.
0
Note that a self-consistent DFT calculation for the state with arbitrary 8
is possible. This calculation needs, however, an additional (nonsymmetry)
constraint on the system [29]. This additional constraint requires application
of an external stabilizing field.
The situation described by symmetry constraint 11 is related to many
interesting physical phenomena. In a typical case, the neglect of apart of the
interactions leads to the ground state of the system that belongs to constraint
I and, therefore is uniquely determined by symmetry. With account for the
fuH Hamiltonian this state corresponds, however, to symmetry constraint 11.
Therefore, in the full- Hamiltonian study it becomes unstable and a variation
of the state must take place. Examples of such systems are, e.g. Fe203 and
Mn3Sn, where the SOC leads to the phenomenon ofweak ferromagnetism [11].
Summarizing this section we can formulate a number of conclusions. First,
a given magnetic structure is stable in the DFT calculations only in the
case when it corresponds to symmetry constraint I. Second, if the structure
corresponds to symmetry constraint 11 its variation is subjected to restrictions
imposed by the relations (5.6-5.8). Thus, although the structure itself is
unstable, the regularities in the magnetic state that foHow from (5.6-5.8)
are preserved features of the magnetic structure. On the other hand, an
assumed regularity in the initial magnetic structure that is not supported
by a symmetry operation is not a stable feature of the magnetic state of the
system.
It is important to distinguish between the stability in the DFT calcula-
tions and the stability in the nature. Magnetic configurations stable in the
calculations may not necessarily be the physical ground state, since random
fluctuations characteristic of real systems are absent in the DFT calculations.
Therefore the symmetry constraint is not efficient in the real systems. On the
other hand, the instability of a magnetic state in the DFT calculations can
be directly related to the instability in nature, because this instability is
a consequence of the interactions in the system. The latter property is of
primary importance for the discussion of concrete physical systems in the
next section. In the next sections we apply the symmetry principles to the
analysis of various magnetic systems.
...<~---- a -------:>~
t
;;'.
c
~
... : : ", ___ . ---- I
,
,.:", ·..~· ~ · O
".
• Pd
Sn
" ® ,,' ®,
® FM "
,, ,® AFM '
® ®
® " ®"" "
Fig.5.2. Unit cell of the U2Pd2Sn crystal structure, The six magnetic structures
studied are indicated
angles about the z-axis. The strueture NCI ean be transformed into NC4 by
a rotation by 90° about the z-axis.
All structures lying between the NCI and NC4 structures have the same
symmetry and are invariant with respect to operations (a), (e), and (d). They
are less symmetrie al than the NCI and NC4 eonfigurations. This property
results in the extrema of the total energy for NCI and NC4 struetures. The
form of the eurve shows clearly that in the ease of the NCI eonfiguration
we deal with symmetry eonstraint I (Fig.5.3). Note that all intermediate
magnetie states in Fig. 5.3 are unstable and the nonsymmetry eonstraint was
used in the ealculations of these states.
Another example of a stable magnetie state is a so-ealled triple-k mag-
netie structure in USb (Fig. 5.4). The list of the symmetry operations of this
strueture ean be found in [33].
In the next seetions, we eonsider relativistic instabilities of the regular
features of magnetie structures.
222 L. Sandratskii
5 3
.s
=
~Ei
2
iil
AFM
0
-4S 4S 13S
<1>
Fig.5.3. Total energy of U2Pd2Sn as a function of 1> calculated with SOC (solid
line) and without SOC (dashed line). On the right-hand side the energies of the six
magnetic structures (see Fig. 5.2) are indicated
• •
3 4
EB 11111 0 U • X
o
-10 -5 o 5 10
e
Fig.5.5. The total energy of U3 P 4 as a function of the deviation of the magnetic
moments from the (111) axis. Solid / dashed line shows the result of the calculation
with/without the spin-orbit coupling. The inset shows the projection of the crystal
and magnetic structure onto the (111) plane. The magnetic moments form a cone
structure
ble for the stability of a collinear magnetic state. For example, the lattice
translations and combined spin-space rotations about the axis parallel to the
magnetic moments are examples of such operations. If, however, the system
possesses none of the symmetry operations responsible for the stability of
a collinear magnetic configuration, this structure must be unstable. Below,
we consider two examples of the system with instabilities of the collinear
ferromagnetic structure.
The inset in Fig. 5.5 shows the experimentally determined magnetic structure
of U 3 P 4. The atomic moments form a cone with the axis parallel to the
(111) axis. The system possesses a strong ferromagnetic component. The
DFT calculation started with all U moments parallel to the (111) axis results
immediately in the noncollinear magnetic structure similar to the structure
found in the experiment [3]. Analysis of the symmetry of the system shows
224 L. Sandratskii
that the deviation of the magnetic moments does not influence the symmetry
of the magnetic structure. Therefore, we deal here with symmetry constraint
II and the initial collinear structure must be unstable. The minimum of the
total energy is at an accidental point.
With the SOC being neglected, the curve of the total energy as a function
of the deviation angle () is symmetrical with respect to the change of the sign
of () (Fig.5.5). Therefore, the collinear configuration corresponding to () = 0
is distinguished by symmetry and is stable [3].
Fig. 5.6. Atomic disorder leads to the noncollinearity of the magnetic structure.
Broken arrows show the initial collinear structure
that gave a basis for most of further work on weak ferromagnetism. In (5.9)
the indices i and j number the atoms in the lattice, I ij and D ij are the
symmetrie and antisymmetrie exchange constants and the tensor K i contains
information about the single ion magnetocrystalline anisotropy. The first term
of the Hamiltonian (5.9), the symmetric exchange, is supposed to lead to
a compensated magnetie configuration. The next two terms, the anisotropie
exchange and the magnetocrystalline anisotropy terms, respectively, can lead
to a small ferromagnetie moment in an otherwise antiferromagnetie crystal.
Moriya [36] showed that depending on the type of crystal structure either
of two mechanisms, antisymmetrie exchange or magnetocrystalline anisot-
ropy, can be the origin for the canting of magnetic moments. Thus, in the
case of IX-Fe203 it is the antisymmetrie exchange that plays the dominant
role, whereas in the case of NiF 2 antisymmetrie exchange is ruled out in
favor of the magnetocrystalline anisotropy, whieh here is responsible for the.
appearance of the ferromagnetic component.
IX-Fe203 is a classical example of the weak ferromagnetism caused by the
antisymmetric exchange. The triangular antiferromagnet Mn3Sn is considered
in Sect. 5.7 and is a system where the antisymmetric exchange contributions
from different atoms cancel and cannot be a reason for the observed weak
ferromagnetism. In this case, the magnetocrystalline anisotropy term is seen
to be responsible for weak ferromagnetism.
The crystal structure of Fe203 is shown in Fig. 5.7. This is a rhombohedral
lattice with a basis of two formula units per unit cell. The following operations
are the generators of the group characterizing the symmetry of the atomic
positions: rotation by 1200 about the z-axis, rotation by 180 0 about the y-
axis, and inversion.
Calculations with the SOC neglected [37] have shown that the magnetic
state with the lowest energy is a collinear antiferromagnetic structure of the
type (+ - -+). Here, + and - designate up and down directions of the Fe
moments with respect to some chosen axis. This result is in agreement with
experiment.
If we next choose the z-axis as the direction of the magnetic moments and
switch on SOC, the following generators remain in the symmetry group of
the Kohn-Sham Hamiltonian: rotations about the z-axis and the inversion
[39]. The symmetry operations that are of special importance for us are
the rotations about the z-axis: none of them change the position of any
of the four Fe atoms lying on the axis of rotation. The directions of the
magnetic moments are parallel to this axis and are not changed either. It
is clear that any deviation of the magnetic moments from the z-axis will
destroy the invariance of the crystal with respect to this operation. As the
symmetry of the Hamiltonian cannot become lower during iterations this
change is forbidden by symmetry and the magnetic structure will remain
collinear during iterations even in the presence of SOC. This result agrees
5 Noncollinear Magnetism in Relativistic Systems 227
.Fe
00
1.
3 I
o-----~
---T.1 -----.
~l
1 2
Fig. 5.7. The unit cell of Fe203. Cross on the diagonal of the rhombohedron shows
the point of inversion. The solid line passing through the 1st oxygen atom indicates
a two-fold symmetry axis. The collinear (dashed arrow) and canted (solid arrow)
directions of the Fe atoms are shown. The canting of the Fe moments in the xy
plane is illustrated differently in the lower right corner of the figure
moments move and deviate from their collinear directions toward the direc-
tion of the x-axis until an "accidental" self-consistent magnetic structure will
be achieved. Since, without SOG, the magnetic structure remains collinear,
the canting of the magnetic moments is a direct result of this interaction.
Because of the small value of SOG compared with the exchange coupling that
is responsible for the anti parallel directions of atomic moments, the canting
of the moments is rat her small and leads in the calculation to a ferromagnetic
moment of about 0.002 JLB per Fe atom [39]. This estimate is in quite good
agreement with the experimental value.
(a) (b)
b tb
c
- a
o u 0 Sn 0 Pd
Fig. 5.8. Projections of the crystal and magnetic structure onto the xy and yz
planes. (a) Projection of the orthorhombic unit cell onto the yz plane. Dotted arrows
show the initial magnetic structure used to start the calculation, thick arrows show
the resulting self-consistent directions of the magnetic moments. Thin arrows show
the experimental magnetic structure. (b) Projection of the orthorhombic unit cell
onto the xy plane. Both experimental and theoretical projections of the magnetic
moments are parallel to the y-axis. (c) Projection of the monoclinic unit cell onto
the xy plane. Arrows show schematically the deviations of the magnetic moments
from the yz plane
affect the symmetry of the system because the magnetic structure has already
lowered the symmetry of the crystal from hexagonal to orthorhombic. As
a result, no qualitative changes of the magnetic structure were observed due
to the orthorhombic lattice distortion and quantitative changes also appeared
to be very small.
A basically different response was obtained to the monoclinic distortions
[45]. In agreement with experiment the b-side of the basal rectangle (Fig.5.8)
was rotated by 0.4° about the c-axis. We started the calculations with the
magnetic structure I. Already after the first iteration all uranium magnetic
moments deviated from the yz plane staying, however, mutually equivalent.
The compensating character of the magnetic structure was also preserved.
Again, a symmetry analysis helps to understand this process. The mon-
oclinic distortion decreases the symmetry of the system such that only one
half of the symmetry operations of the orthorhombic structure are left over
in this case. These are the operations numbered in Table 5.2 as 1, 2, 5, and 6.
Operation 5 demands equivalence of atom 1 to atom 3 and atom 2 to atom
4. Simultaneously, the moments of the equivalent atoms must be antiparallel:
ml = -m3 and m2 = -m4. Operation 2 is responsible for the equivalence
of atoms 1 and 4 and the following relation between the components of the
magnetic moments: m~ = -m!, mt = -mt,
m; = m;.
Thus, we see the
important difference between the orthorhombic and the monoclinic struc-
230 L. Sandratskii
c
1 {eIO} no no
(:J c:')
-mz .
• J
-mz .
• J
5 {el'T2}R 1++3; 2++4 Tni = -Tnj ; i ++ j
7 {axIO}R no mx =0 all i
0 • 0
• • 0 0 •
• ~.
q:./.. ::.::p • 0 •
0
e·:, .: .. : .:'::.
0 0 • Mn Sn
• 0 • z=1/4
z=3/4
•
0
•
0
I~ I I
Fig. 5.9. Crystal and magnetic structure of Mn3Sn. Rotations of the magnetic
moments leading to weak ferromagnetism in structures (c) and (d) are shown only
for atoms in the z = 0.25 plane (thin arrows). Moments ofthe atoms in the z = 0.75
plane are parallel to the moments of the corresponding atoms of the z = 0.25 plane
232 L. Sandratskii
In this section we show that symmetry principle is useful in the study of the
state of the nonmagnetic sublattices in antiferromagnetic systems. Although
the atoms of a nonmagnetic sublattice do not develop an independent mag-
netism, the properties of these atoms are influenced by the interaction with
the electron states of the atoms of the antiferromagnetic subsystem.
For a number of years the magnetic structure of UFe4Als was the matter of
much controversy. The suggestions made on the basis of different experimen-
tal investigations ranged from simple one-sublattice ferromagnetism up to un-
usual spin-glass state [47]. Recent investigation of a single crystal of UFe4Alg
with the use of unpolarized and polarized neutron diffraction revealed an
ordered magnetic structure with a number of unique features [48] (Fig.5.10).
Two magnetic sublattices were detected. The magnetic moments of the U
sublattice form a collinear ferromagnetic structure. A strong noncollinearity
is, however, observed between the U and Fe magnetic moments, which are
almost orthogonal to one another. The magnetic structure of the Fe sublattice
is elose to a collinear antiferromagnetic (so called G-type antiferromagnetic
structure). Additionally, there is a canting of the Fe moments, which leads
to the second type of noncollinearity in UFe4Alg. The noncollinearity within
the Fe sublattice results in a weak ferromagnetic moment parallel to the U
moments.
In [26] we have shown that the complex noncollinear magnetic structure
of UFe4Als is a necessary consequence of the properties of a much simpler
magnetic state with collinear antiferromagnetism of the Fe sublattice and
nonmagnetic U sublattice (Fig.5.10). To stress the importance of this state
for understanding of the magnetism of UFe4Alg we refered to it as the basic
state (BS).
The BS possesses a number of regular features: the Fe atoms have equal
magnetic moments, these magnetic moments are collinear, the magnetic mo-
ment of the U atoms has the value of zero. We emphasize that the property
that the U moment is equal to zero must be considered as a regular feature of
the system, which can be stable only in the case that there exists a symmetry
5 Noncollinear Magnetism in Relativistic Systems 233
J-
b
f u
~
~ Fe
f 4 basic state
~
~
a
C
Fig. 5.10. Experimental magnetic structure and basic magnetic state of UFe4AIs
operation that is responsible for the stability of this feature in the sense of
our symmetry criterion.
The DFT calculation started with the BS [26] shows immediately that two
of these three regular features are not preserved by the system: An induced
moment appears at the U site. It is orthogonal to the initial directions of the
Fe moments. Simultaneously, the Fe moments deviate from their collinear
directions and form a canted noncollinear configuration. Thus, the BS is
transformed into the complex magnetic state observed experimentally. Sym-
metry analysis helps to understand the origin of this transformation.
In Table 5.3 we collect the generators of the symmetry group of the BS
of UFe4AIs. One of the interesting quest ions to answer is about the physical
mechanism of breaking, by the planar magnetic state of the BS, the symmetry
between two opposite directions of the b-axis. Indeed, the ferromagnetism
of the U sublattice distinguishes one of the two directions. Two opposite
directions of the U magnetic moments can be equivalent only in the case
when there is a symmetry operation of the system that reverses the direction
of the magnetic moment. In the present case it must be an operation that
reverses m y . Analysis ofTable 5.3 shows that this condition is not fulfilled. All
symmetry transformations leave m y invariant. As a direct consequence of this
property the appearance of a magnetic moment parallel to the y-axis at the U
sites does not disturb any of the symmetry operations. Therefore we deal here
with symmetry constraint 11 where the value of the U magnetic moment plays
the role ofthe variable parameter. The symmetry predetermined instability of
234 L. Sandratskii
Table 5.3. Generators of the symmetry group of the basic state of UFe4Als C2x
and C2y are 180 0 rotations about the x- and y-axis, respectively; I inversion; R
time revers al. In the last column, for the U sublattice i = j; for the Fe sublattice i
and j according to the column "Transposition of the Fe atoms"
1++3;2++4
(::)
mz
(-~x)
-mz.
J
I no no
1++4;2++3
0.5
~E
- - - mFella
ur
--------- mFe IIc
The same symmetry analysis shows that the collinear directions of the
Fe moments are also not distinguished by symmetry and cannot be stable.
Both m y and m z components of the moments must appear. To preserve
the symmetry of the BS the m z components of different atomic moments
have different signs and compensate one another. On the other hand, the
m y components of all atomic moments are equal and, in agreement with
experiment, result in a weak ferromagnetic moment along the b-axis.
Summarizing, if the appearance of the induced moments on nonmagnetic
atoms in an antiferromagnetic system does not destroy any of the symmetry
operations of the system, these moments must appear. The directions of
the induced moments do not necessary coincide with the directions of the
induced moments and are determined by the condition that the symmetry of
the magnetic system must be preserved.
We emphasize that the symmetry analysis and DFT calculations reported
in this section were performed taking account of the SOC. With the SOC
neglected, none of the effects discussed here appear and the BS is stable.
1 2 3
u
o Pb
o In
5 6 7
Fig.5.12. Models for the magnetic structure of UX3 compounds. The induced
magnetic moments on sites that follow from the symmetry analysis are shown. See
[49] for all details
consider two examples of the systems where helical magnetic structures were
reported to be observed and the role of the SOC cannot be neglected.
where the a v are the positions of the U atoms and q is the helix vector.
Although not periodic with the underlying lattice periodicity, the helical
structure is very regular. This regularity can be described by the generalized
translations [l1J that combine lattice translations Rn with spin rotations by
5 Noncollinear Magnetism in Relativistic Systems 237
qRn about the y-axis. The scalar-relativistic Hamiltonian of the helix com-
mutes with the generalized translations. This property allows us to reduce the
consideration of the incommensurate helical magnetic structure to a problem
in one chemical unit ceH of the crystal and thereby drastically simplifies the
calculations.
The role of the symmetry with respect to the generalized translations
is not, however, restricted to the technical convenience. This symmetry is
crucial for the stability of the regularities characteristic of a helix. The spin-
orbit coupling term of the Hamiltonian does not commute with the general-
ized translations. Therefore, the relativistic Hamiltonian cannot be invariant
with respect to the generalized translations. This leads to the instability of
the incommensurate helical structures for the systems where the SOC plays
an important role in the formation of the magnetic structure. The physical
reason for the distractive role of the SOC can be easily understood. Indeed,
regularity inherent in the helices assumes that only the relative directions of
the magnetic moments are important, but not their directions with respect to
the crystal lattice. The SOC connects the space and spin degrees of freedom
and makes inequivalent different orientations of the moments with respect to
the lattice. The destructive role of the SOC provides an explanation for the
fact that the main body of the DFT studies of helical structures is restricted
to the case of the helical structure in ,/,-Fe where the SOC can be neglected.
A much more complex situation exists in the case of rare-earth met als
(REM) [51]. Here, the exchange interaction favors the formation of a helical
structure. However, the influence of the SOC is stronger than in ,/,-Fe and
cannot be neglected.
A further increase of the role of the SOC can be expected in the case
of actinide systems. The considerable spatial extent of the 5f electron states
enhances the influence of the SOC on the directions of the atomic moments.
This property is confirmed by the observation of a large magnetic anisotropy
(see, e.g. [55]) in almost aH actinide compounds. Taking into account the
influence of the SOC on helical structures, the formation of a helix in U
compounds can be considered as highly improbable [11,56]. This conclusion
seems to be in good agreement with the experimental observations. There is,
however, one exception: UPtGe (Fig.5.13) was reported to possess a helical
ground-state structure [57,58]. An understanding of the nature of the unique
magnetism of UPtGe is essential for the physics of incommensurate magnetic
structures. On the other hand, the unique magnetic structure makes UPtGe
an important test case for the verification of the capability of DFT to explain
the complex magnetism in 5f systems. Below, we discuss the explanation of
the properties of UPtGe suggested by Sandratskii and Lander [27].
Two different types of helices are known. The first type is formed, für
example, in fee-Fe and REM and results from the properties of the exchange
interactions. Neglecting the SOC and considering the total energy of the helix
as a function of q, Eex(q), one obtains the minimum at an incommensurate
238 L. Sandratskii
y
~ z
z
Fig. 5.13. The U positions in UPtGe. The magnetic structure shown corresponds to
q = (0.5,0,0) and is a helix with the moments in the xz plane with the propagation
along x. Experimental value of q is (0.55-0.57,0,0) [57,58]
a) b)
,,
,
,
o o
q q
Fig. 5.14. Schematic picture of the total energy as a function of q for (a) exchange
and (b) relativistic helices
q value (Fig. 5.14a). As stated above, such a helix can be formed only in the
case when the MA is much weaker than the interatomic exchange interactions
determining the form of the E ex (q) function.
In apparent contradiction to the arguments above, the second type of
helices is formed because of the influence of SOC, e.g. the helix in MnSi
[59,60]. The exchange interaction in MnSi favors a collinear ferromagnetic
ordering. This means that the minimum of the total energy as a function
of q, calculated nonrelativistically, is at q = 0 (Fig.5.14b). To explain the
formation of the helix in MnSi a model Hamiltonian was used (see, e.g. [59]),
5 Noncollinear Magnetism in Relativistic Systems 239
Table 5.4. Important energy differences in mRydjU atom. The energy of mag-
netic anisotropy is estimated as the total-energy difference between the magnetic
structures with the same relative orientation of the atomic moments but different
orientation with respect to the crystal lattice. The interatomic exchange interac-
tion is estimated as the difference of the total energies of the ferromagnetic and
antiferromagnetic configurations with atomic moments collinear to the same axis
Magnetic anisotropy
E FM (010)-E FM(100) 1.5
EFM (OlO)-EFM (001) 1.4
EFM(001)-E FM (100) 0.1
EAFM(010)-EAFM(100) 1.3
EAFM (OlO)-EAFM (001) 1.2
EAFM(OOl)-EAFM(lOO) 0.1
Exchange interaction
E AFM (lOO)-E FM (100) -0.1
E AFM (0l0)-EFM(010) -0.3
EAFM(OOl)-EFM(OOl) -0.2
E q =(0.5,0,0)-EFM(100) -0.8
0.0
•
•
E
~ -{).5
• •
::J
&-
s •
>- -1.0
~
•
• .,,-. •
Q)
c:
Q)
-1.5
b)
- q
Fig. 5.16. Frustrated magnetic interactions. (a) Ihl « Ihl. (b) Helical magnetic
structure as a consequence of the frustration
0.0
Table 5.5. Generators of the symmetry groups for three magnetic states in hcp
metals. Number of atoms in the magnetic unit cell nat characterizes the periodicity
of the magnetic structure along the c-axis. C 2 b is the 1800 rotation about the b-
axis; U a and U C are the reflections in the plane orthogonal to the axes a and c,
respectively; R time reversal. Vectors in the column "Operation" give the nonprim-
itive translations entering the symmetry operations. Vectors are given in units of c.
Atoms not presented in the column "Transposition" are invariant with respect to
the given symmetry operation. In the last column,
Here, using the example of Er, we will consider the influence of the SOC
on the helical magnetic structures in REM. A more extensive discussion can
be found [66].
In the case of Er, the structures with q = ~ are of interest [51]. First, we
consider the influence of the SOC on a planar helix with q = ~ (Fig.5.18a)
The generators of the symmetry group are given in Table 5.5. There are
three groups of equivalent atoms: {1,5}, {3,7}, {2, 4, 6, 8}. The moments of
atoms 1 and 5 must keep their initial directions parallel to the b-axis. Atomic
moments 3 and 7 deviate within the ac plane, no b component can appear.
Moments 2, 4, 6, 8 move both within the ab plane and out of the ab plane.
No ferromagnetic component can appear.
Numerical calculations started with this helical structure gave an inter-
esting result that differs drastically with the results obtained in the calcu-
lations for Ho. For the SOC scaled by a factor of 20 the moments deviate
244 L. Sandratskii
(a) (b)
1 c 3
* *-
3 7 5 1
4 6 6 8
7
b
Fig.5.18. The 8-layer magnetie eonfigurations in hep Er. (a) The initial planar
helical strueture. All moments are parallel to the ab plane. (b) The ealculated
planar magnetie strueture. All moments are parallel to the bc plane
strongly from the ab plane and result in the magnetic configuration shown
in Fig.5.18b. Thus, the initial planar magnetic configuration with moments
parallel to the horizontal ab plane is replaced by a planar magnetic structure
parallel to the vertical bc plane. This transformation of the magnetic config-
uration is not forbidden by symmetry since all the symmetry elements of the
initial structure are preserved. The final magnetic state is more symmetrical
than the initial one since the symmetry group contains one additional gener-
ator (Table 5.5). This example illustrates the property that the symmetry of
the state of the system can increase in the calculations. The planar vertical
structure obtained in the calculations is in good agreement with a vertical
planar cycloidal structure found experimentally in Er. Two structures are,
however, not identical: The calculations resulted in a structure parallel to the
bc plane. The experimental structure is parallel to the ac plane. A wobbling
of the vertical structure found experimentally is also not reproduced in this
calculation. The reason for this disagreement is, again, connected with the
symmetry ofthe initial state. Indeed, the structure shown in Fig. 5.18a cannot
transform within the DFT calculations into the planar structure parallel to
the ac plane since this transformation leads to a loss of symmetry operations.
If .,. ".
,
~
If ,..
~
~ -r. ~
where w is the frequency of the incident photons, En,k is the energy of the
electron eigenstate In, k > labeled with the band index n and vector k in
the first Brillouin zone. Furthermore, Po: is the a-component of the electron
moment um operator, fee) is the Fermi-Dirac function, and 8 is a phenomeno-
logical parameter describing finite-lifetime broadening. The comparison with
5 Noncollinear Magnetism in Relativistic Systems 247
experiment was earried out with the normal-ineidence refiectivity and the
polar Kerr rotation [74]. Assuming the z-axis to be normal to the surface of
the erystal the refiectivity takes the following form
n . V+
+ zk = 1 47ri
--CTx x'
W '
is applicable only in the ease when the z-axis is a symmetry axis of at least
third order (see, e.g. [75]). This is so for U 3 P 4 where the magnetization is
parallel to the crystallographic [111] axis, which is a three-fold symmetry axis.
Although the presence of magnetie ordering can substantially decrease the
symmetry, for the noncollinear ground-state magnetic structure of U 3 P 4 the
symmetry with respect to the [111] axis stays intact. Thus, by directing the
z-axis along the crystallographic [111] axis we obtain the geometry necessary
for the calculation of the polar magneto-optical Kerr rotation. Note that
the expression (5.11) for the optical conductivity tensor is quite general and
applicable to both collinear and noneollinear magnetic configurations. The
difference between both cases is contained implicitly in the electron eigen-
energies and eigenfunctions entering this formula.
To study the sensitivity of the optical characteristics to the canting angle
we performed the calculation for different cone angles (Fig. 5.20). The depen-
dence of the refiectivity on the canting angle is seen to be surprisingly weak,
whereas the photon-energy dependence of the Kerr rotation practically scales
with the macroscopic magnetization given by the projection of the magnetic
moments onto the [111] axis. To understand the essential difference in the
angular dependenee of the refiectivity and the Kerr angle note that the first
eharacteristic is determined by the diagonal component of the eonductivity
tensor (5.11) in contrast to the seeond characteristic, which depends crucially
on the off-diagonal component of the tensor. Although the same electron
transitions contribute to both components the weight of the contribution is
different [75]: the absorption part of the diagonal component can be repre-
sented as a sum of the corresponding components for the right- and left-
circularly polarized light, opposite to the absorption part of the off-diagonal
component that ean be seen as a difference of the corresponding character-
istics for the right- and left-eircular polarized light. As a direct consequence
of this property, the off-diagonal component of the conductivity tensor must
be zero for a canting angle of 90° because in this case all atomic moments
are parallel to the xy plane and the right- and left-circularly polarized waves
become equivalent. Simultaneously, the Kerr rotation becomes zero. Thus,
the monotonie decrease of the Kerr rotation with increasing canting angle
248 L. Sandratskii
- - 0 =1° - - 0=1° 5
0.8 +---+ 0=15° +---+ 0 =15°
~ 0=30° - 0 =30°
. . . 0=50° - - _. 0=50° 4
0.6
3
OA 2
0.2
o
0.0
o 2 4 o 2 4 6
E (eV) E (eV)
Fig.5.20. Calculated [74] reflectivity (left) and polar Kerr rotation spectra (right)
of U3 P 4 for different deviations B of the uranium magnetic moments from the
crystallographic [111] direction
5.12 Conclusion
Acknowledgement
The author is very grateful to all colleagues in collaboration with whom the
studies cited in this chapter have been performed.
References
1. R. Lorenz, J. Hafner, S.S. Jaswal, and D.J. Sellmyer: Phys. Rev. Lett. 74, 3688
(1995)
2. V.P. Antropov, M.L Katsnelson, M. van Schilfgaarde, and B.N. Harmon: Phys.
Rev. Lett. 75, 729 (1995)
3. L.M. Sandratskii and J. Kübler: Phys. Rev. Lett. 75, 946 (1995)
4. L. Nordström and D.J. Singh: Phys. Rev. Let. 76,4420 (1996)
5. 1. Solovyev, N. Hamada, and K. Terakura: Phys. Rev. Lett. 76, 4825 (1996)
6. T. Oda, A. Pasquarello, and R. Car: Phys. Rev. Let. 80, 3622 (1998)
7. L.M. Sandratskii and J. Kübler: Phys. Rev. Lett. 76, 4963 (1996)
8. M. Uhl and J. Kübler: Phys. Rev. Lett. 77, 337 (1996)
9. Y. Wang, G.M. Stocks, D.M.C. Nicholson, and W.A. Shelton: J. Appl. Phys.
81, 3873 (1997)
10. O.N. Mryasov, R.F. Sabiryanov, A.J. Freeman, and S.S. Jaswal: Phys. Rev.
56, 7255 (1997)
11. L.M. Sandratskii: Adv. Phys. 47, 91 (1998)
12. Q. Niu, X. Wang, L. Kleinman, W.-M. Liu, D.M.C. Nicholson, and G.M.
Stocks: Phys. Rev. Lett. 83, 207 (1999)
13. M. van Schilfgaarde, LA. Abrikosov, and B. Johansson: Nature 400, 46 (1999)
250 L. Sandratskii
45. L.M. Sandratskii and J. Kübler: J. Phys.: Condens. Matter 9, 4897 (1997)
46. S. Tomiyoshi and Y. Yamaguchi: J. Phys. Soc. Jpn. 51, 2478 (1982)
47. B. Ptaciewicz-Bak, A. Baran, W. Suski, and J. Leciejewicz: J. Magn. Magn.
Mater. 76-77,439 (1988); J. Gal, I. Yaar, E. Arbaboff, H. Etedgi, F.J. Litterst,
K Aggarwal, J.A. Pereda, G.M. Kalvius, G. Will, and W. Schäfer: Phys. Rev.
B 42, 237 (1989); A.V. Andreev, H. Nakotte, and F.R de Boer: J. Alloys
Compd. 182, 55 (1992)
48. J.A. Paixäo, B. Lebech, A.P. Gon<;alves, P.J. Brown, G.H. Lander, P. Burlet,
and A. Delapalme, J.C. Spirlet: Phys. Rev. B 55, 14370 (1997)
49. S. Demuynck, L.M. Sandratskii, S. Cottenier, J. Meersschaut, and M. Rots: J.
Phys.: Condens. Matter 12, 4629 (2000)
50. J.M.D. Coey: Can. J. Phys. 65, 1210 (1987)
51. J. Jensen and A.R Mackintosh, Rare Earth Magnetism (Clarendon Press,
Oxford 1991)
52. O.N. Mryasov, A.I. Liechtenstein, L.M. Sandratskii, and V.A. Gubanov: J.
Phys.: Condens. Matter 3, 7683 (1991); M. Uhl, L.M. Sandratskii, and J.
Kübler: J. Magn. Magn. Mat. 103, 314 (1992); M. Körling and J. Ergon:
Phys. Rev. B 54, 8293 (1996); D.M. Bylander and L. Kleinman: Phys. Rev. B
59, 6278 (1999) and 60, 9916 (1999)
53. K Knöpfte, L.M. Sandratskii, and J. Kübler: Phys. Rev. B 62, 5564 (2000)
54. L. Nordström and A. Mavromaras: Europhys. Lett. 49, 775 (2000)
55. V. Sechovsky and L. Havela: 'Magnetism ofternary intermetallic compounds of
uranium'. In: Handbook of Magnetic Materials, ed. KH.J. Buschow (Elsevier,
Amsterdam 1998) pp. 1-289
56. L.M. Sandratskii and J. Kübler: Mod. Phys. Lett. B 10, 189 (1996)
57. A. Szytula, M. Kolenda, R Troc, V.H. Tran, M. Bonnet, and J. Rossat-
Mignod: Solid State Commun. 81, 481 (1992); RA. Robinson, A.C. Lawson,
J.W. Lynn, and KH.J. Buschow: Phys. Rev. B 47, 6138 (1993); S. Kawamata,
K Ishimoto, Y. Yamaguchi, and T. Komatsubara: J. Magn. Magn. Mater.
104-107, 51 (1992)
58. D. Mannix, S. Coad, G.H. Lander, J. Rebizant, P.J. Brown, J.A. Paixäo, S.
Langridge, S. Kawamata, and Y. Yamaguchi: Phys. Rev. B 62, 3801 (2000)
59. O. Nakanishi, A. Yanase, A. Hasegawa, and M. Kataoka: Solid State Commun.
35, 995 (1980)
60. M. Ishida, Y. Endoh, S. Mitsuda, Y. Ishikawa, and M. Tanaka: J. Phys. Soc.
Jpn. 54, 2975 (1985)
61. W.C. Koehler: 'Magnetic structure of Rare Earth Metals and Alloys'. In: Mag-
netie Properties of Rare Earth Metals, ed. RJ. Elliott (Plenum Press, London
1972) pp. 81-128
62. J. Jensen: Acta Phys. Pol. A 91, 89 (1997)
63. B.R Cooper: 'Phenomenological Theory of Magnetic Ordering: Importance of
Interactions with the Crystal Lattice'. In Magnetic Properties of Rare Earth
Metals, ed. R.J. Elliott (Plenum Press, London 1972), pp. 17-80
64. A.Y. Perlov, S.V. Halilov, and H. Eschrig: Phys. Rev. 61, 4070 (2000)
65. M.S.S. Brooks, L. Nordström, and B. Johansson: J. Phys.: Condens. Matter
3, 2357 (1991); M. Richter and H. Eschrig: Physica B 172, 85 (1991)
66. L.M. Sandratskii: Phys. Rev. 64, (2001)
67. P. Strange, A. Svane, W.M. Temmerman, Z. Szotek, and H. Winter: Nature
399, 756 (1999)
252 L. Sandratskii
6.1 Introduction
The perovskite manganese oxides (the manganites) with the general chemical
formula (Rl-xDx)n+lMnnÜ3n+l (R and D being the trivalent rare-earth and
divalent alkaline-earth ions, respectively) have been the focus of enormous
research attention during the past few years. The renewed interest in these
systems has been been spurred mainly by the rediscovery of the phenomenon
of colossal magnetoresistance, that is a gigantic suppression (by more than
100 000%) of the resistivity of the sampIe by the external magnetic field near
the Curie temperature, known from the early 1950s and very intensively
studied in magnetic semiconductors [1 J.
Another important feature of perovskite manganites is the rich magnetic
phase diagram as a function of the hole doping x and the temperature T. Like
the colossal magnetoresistance, the main trends of the phase diagram were
established in the 1950s by the work by Wollan and Koehler [2], and first
discussed by Goodenough [3J. As an example, let us focus on the behavior
of three-dimensional manganites (n = 00) at low T. The parent material
LaMnü3 is an insulator, which reveals a very strong orthorhombic distortion
accompanied by the layered (A)-type antiferromagnetic (AFM) spin ordering
[4,5J. The characteristic region of the phase diagram spreads up to x ~ 0.17
(in the case of Lal-xSrxMnü3). The exact nature of the magnetic ordering
in this region is the subject of much controversy between theories of spin
canting [6J and the inhomogeneous phase separation [7-9J. A furt her increase
of x leads to rather characteristic sequence of ferromagnetic (F or FM),
layered (A)- and chainlike (C)-type AFM ground states [2,1O-13J (shown
in Fig.6.1 for the case of Ndl-xSrxMnü3 [10]). The end-member materials,
such as CaMnü3, form the simple two-sublattice (G)-type AFM ordering.
The phase diagram mayaIso contains the regions of canted AFM ordering
(i.e. CAF in Fig.6.1) and/or the phase separation. x = 1/2 presents a very
special region of the magnetic phase diagram, where the manganites may
form (with the exception of the wide-band materials such as Lal-xSrxMnü3,
for example) the CE-type AFM structure [2,3], which is composed of the one-
dimensional FM zigzag stripes and the existence of which is typically ascribed
to the charge ordering. The stripe structures in manganites appear to be
more generic and a new family of such phases has been observed recently in
254 I. Solovyev and K. Terakura
350
PM
300 insulator
52
-- 250
~
::l
...
iii
Q)
200
a. 150
E
~ 100
50
LlE = - ~2 L...J
~ J.Tn [e·.
~
e·+
~ Tn
- e~t . e~ ]
~+Tn· (6.2)
iTn
In this sense LlE is exactly mapped onto the Heisenberg model. However,
one should not exaggerate the generality of such a mapping.
1. It is applicable only for small rotations {~<pd. Therefore, the mapping is
local and can be used only for the analysis of the loeal stability of a mag-
netic equilibrium. Attempts to use the same parameters of the Heisenberg
model for the description of finite-temperature effects associated with the
spin disorder generally have no solid theoretical justification.
1 In the subspace spanned by the orbital indices, the 2 x 2 matrices 4j == IIt{jL' 11 can
",3
be written in the compact form: tij = L...a=l tiHn" (Oj,Hn" + c5j,i-n,,), where
~ ~
2. Since E(M O ) =I- -~ L:i-rn J-rne~ . e~+-rn' (6.2) is applicable only to i1E,
but not for the absolute values of the total energy.
(6.3)
where ilij and gij stand for the 2 x 2 matrices in the orbital subspace and
Tr L is the trace over the orbital indices.
In the case of FM ordering, e~ = (0,0,1) and the Hamiltonian (6.1) takes the
simple form: ilij = -4j + i1exÖs,.J.Öij. If i1ex is large enough, and the system
is fully spin-polarized, all poles of g.J.(e:) are located in the unoccupied part
of the spectrum and one can use the following expansion for e: ::; e:F:
(6.4)
Since 4j are restricted to the nearest neighbors, each term in the expansion
(6.4) has a limited radius equal to the maximal cllstance between two sites
separated by the n number of hops. In the second order of 4j / i1 ex , the
nonvanishing interactions are limited by the nearest neighbors (J1 ) and the
next-nearest neighbors (h and J 4 in the cubic lattice) [27]. h is given by
the combination of the FM DE interaction
(6.5)
(6.6)
(6.7)
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 257
Using the identity L .... gr.... (c)tTnj + cgrj(c) = 8ij, which follows from the
definition ofthe Green function, the interactions (6.5)-(6.7) can be expressed
through the moments of Iocal density of states
and
(6.10)
where z = 6 is the coordination number. Then, one can verify that Jp > 0,
Jr < 0, and 2J2 + J4 > 0 « 0) for the less than half- (more than half-)
filled t-spin band. Furthermore, m(O) = 1 - xis the number of eg electrons
and m(l) = E o is the kinetic (double exchange) energy of the eg electrons.
In order to get a rough idea about the magnitude of DE and SE interac-
tions in realistic compounds, one can use experimental parameters of mag-
netic interactions reported in [28] for Pro.63Sro.37Mn03 (x = 0.37): J 1 = 10.1,
J 2 = -0.6, and J4 = 2.7meV;2 and the parameters ofthe electronic structure
in LSDA: to ~ 0.7eV [27] and .1ex ~ IM, where I is the intraatomic Hund's
rule coupling (I ~ 0.geV for Mn atoms [29,30]) and M = 4 - x is the
spin magnetic moment corresponding to the formal configuration t~gteltX.
Then, the e g contribution to the SE interaction can be estimated from (B.9)
and (6.10) as Jr(eg) ~ -50 meV. It will be combined with the nearest-
neighbor (nn) SE interaction between the localized t2g electrons. The latter
can be estimated from the Neel temperature of CaMn03 (TN = 130 K)
as Jr(t2g) ~ -5.6 meV. The LSDA calculations yield somewhat overes-
timated value: Jr(t2g) ~ -18meV. 3 In any case, one can conclude that
the main contribution to the nearest-neighbor SE interaction in manganites
comes from the eg electrons. Then, the DE interaction is given by Jp =
J 1 - Jr(eg) - Jr(t2g) ~ 66-78 meV. Finally, the kinetic energy of the eg
electrons can be obtained from (6.8) as 788-937meV (at x = 0.37).
2 According to the form of (6.2), the experimental parameters have been multiplied
by 28 2 = (4 - x)2/2.
3 The value has been obtained as the second derivative of the total energy near the
G-type AFM ground state (Sect.6.3.1). For CaMn03, it agrees with the total
[31]). Jr
energy difference between the collinear states F and G (-116/6 ~ -19 meV,
(t2g) is substantially reduced in LaMn03 ( -13 meV, [20]), for which
both M and the unit-cell volume are larger (and therefore Ll ex is larger, while to
is smaller).
258 I. Solovyev and K. Terakura
N
ci
O r-------------~
FM
1\,-;:~)1
<t <t C-lype
ci ci
N N
ci ci
°LL~~~L-~~~ o
ci -3 -2 -I 0 1 2 3 ciO~-OL.2--0L.4~O~.6~O~.-8~
Energy x
Fig.6.2. Left panel: local densities of states of the degenerate double exchange
model for the ferromagnetic, A-, and C-type antiferromagnetic states. Right panel:
parameters of orbital polarization Lln = n3z2 _r2 - n x 2 _y2 (the difference of atomic
populations of the two eg orbitals) as a function of hole doping
(6.12)
where G(c) is the t-spin Green function for the isolated FM sublattice in
the ground state of the DE model, the sites 0 and i belong to one FM
sublattice, and the sites m and j to another sublattice, different from the
first one. Magnetic interactions within the FM sublattice (tt) consist of two
contributions. The first one is the DE interaction for the isolated sublattice,
given by (6.5) with gt
= G. The second one takes into account the interaction
between different sublattices caused by the kinetic hoppings induced by small
deviations from the AFM alignment. It is given by (6.12) with the substitution
i +--+ m in the curly brackets. Since tim,
and 40
are restricted by the nearest
neighbors, the real number of terms in the sum over i and j in (6.12) is very
limited.
The interactions (6.12) are long ranged. The contribution of the long-
range part to the stability of the AFM ordering depends on the actual scenario
of how the perfect AFM ordering is broken. For the single-spin rotations, the
contributions of intra- and intersublattice long-range interactions of the form
(6.12) are cancelled out [32]. The condition of local stability is given by the
requirement that the nn coupling, J D + J s, is positive (negative) in the FM
(AFM) bonds:
7 In this respect one can find some similarity to the problem of orbital magnetism
in solids. Formally, the combination of the (relativistic) spin-orbit interaction and
the spin polarization is sufficient to induce the orbital magnetization. However,
the magnitude of this efIect is typically too smalI, leaving some room for the
additional enhancement driven by the onsite Coulomb interactions [35]. The
parameter Ueff, controlling this enhancement, varies from one dass of materials
to another. For metallic compounds it is typically smalI.
8 As an alternative, one can consider the angular dependence of the kinetic energy
ED for the Hamiltonian (6.11), which yields the same result for J~(t+) [33].
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 261
0
N
G
ci
CE
V>
ci
F
8u-__~~__~__~__~=-__~
ci 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x
Fig.6.3. Areas of local stability of the main collinear states in the degenerate
double exchange model
with respect to the canting of all spins in adjacent planes (chains), JP(H)
in (6.12) will be replaced by JD(H) = L.-n J~(t.!.) with m running over all
atoms in the plane (chain) [33]. The corresponding areas of the local stability
of the main collinear phases are shown in Fig.6.3. The main results of the
model analysis can be summarized as follows.
1. The AFM spin ordering can be a stable equilibrium state of the de-
generate DE model and compatible with the metallic behavior if it ac-
companies the cubic symmetry breaking (the AFM phases A and C).
The conclusion is in drastie contrast with the simplified DE model con-
sidered by de Gennes [6], who assumed that all carriers are localized
at the single (r) point of the Brillouin zone, which led to the incorrect
J(l + ei . ej )/2 == I~ij I dependence of the DE energy on the directions of
the spin magnetic moments. 9 The arguments by de Gennes are applicable
only to the highly symmetrie AFM phase G, which can be stabilized only
at x = 1 (i.e. when there is no e g electrons).
2. The states F, A, and C can coexist at certain electron densities, being
different local minima ofthe total energy E(M) = t'v(M)+t's(M) . Such
regions are featured by first-order phase transitions between different
collinear states and can be potential candidates for the phenomenon of
phase separation [7,8].
3. There are several distinct regions, especially for large IJsl and x, where
none of the collinear states is locally stable. Within such regions one can
expect a number of spin-canted ground states, which will compete with
the inhomogeneous phase separation [33].
4. The orbital degeneracy plays a very important role in the stability of
the anisotropic AFM structures [24]. The stability conditions can be al-
tered through the parameter of orbital polarization .1n (see Fig. 6.2). The
9 Note, that due to the J(l + ei· ej)/2 term, the AFM spin alignment is unstable
with respect to the spin canting and cannot be realized as the ground state at
finite carrier concentrations.
262 I. Solovyev and K. Terakura
<:>
N
ci
.,..
ci
<:>
s: ci
on
0
ci
F
8
ci 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x
Fig. 6.4. Phase diagram of the degenerate double exchange model including the
collinear states F, A, C, and G, and the intermediate spin-canted states (denoted
as XY). Remaining areas correspond to the spin-canted ground states of a general
type (see [33] for details) . An approximate doping dependence of JS is shown by
dashed line
The main building blocks of the CE-type AFM structure are the FM zigzag
chains, which are coupled antiferromagnetically in the direction perpendicular
z=O z=a o
Fig. 6.5. Spin and orbital structure in two adjacent planes of the CE-type AFM
phase. ao is the cubic lattice parameter. The "orbital structure" means the prefer-
entially occupied orbitals at the bridge sites. The orbitals at the corners sites are
not shown for simplicity
to the chain propagation. Thus, in the DE limit, the chains can be considered
as separated objects interacting only via the SE term. The geometry of the
CE phase makes the Mn sites nonequivalent and divides them into the two
types: the "bridges" (A and A' in Fig.6.5) and the "corners" (B and B'),
alternating in the xy plane. Historically, the behavior has been attributed to
the charge order of the MnH and MnH ions, occupying correspondingly the
bridge and the corner sites [3], and the CE phase of manganites is frequently
called the "charge ordered" phase. Although the issue of the charge ordering
is still on the agenda, especially in the high-temperature regime (we will
return to this problem later), the appearance of the CE-type AFM phase can
be understood exclusively from the viewpoint of magnetism [32,34,36-39].
The crucial observation along this line is that the isolated zigzag chain is the
band insulator, as a consequence of both the topological aspects [37,39] and
the peculiarities of kinetic hopping of the eg electrons [25]. The basic idea is
very simple.
1. The translation invariance requires an equal number of the right (B) and
left (B') corner sites in the chain.
2. At the bridge sites A (A'), only the 3x 2 _r 2 (3 y 2 -r 2 ) orbitals participate
in the kinetic hoppings, while the states y2 - z2 (z2 - x 2 ) are inert.1 2
At the corner site, the electron hops via two eg orbitals, 3z 2 - r 2 and
x 2 - y2, when it preserves and changes its phase, respectively. In the one-
dimensional case, the gap opens both in the center and at the boundary
of the Brillouin zone, which explains the insulating behavior.
The TB electronic structure of the single zigzag chain consists of the bonding
and antibonding states (ct = ±J2 - cos k) and two groups of nonbonding
12 Therefore, in the DE limit , the CE-type AFM phase will automatically acquire
the orbital-order pattern shown in Fig. 6.5.
264 I. Solovyev and K. Terakura
o
v 1~
S11
I
I
~ M
.'-1 ~
~
v
o 1. 1I~ v
;!; ~
ii
11"1 1
N
~
o
-2 -I o
Energy
Fig. 6.6. Densities of states of the zigzag ehains specified by different numbers of
bridge and eorner atoms in the unit eell: 2NB and 2Ne, respeetively. The CE phase
eorresponds to NB = Ne = 1. The nonbonding y2 - z2 and z2 - x 2 orbitals at the
bridge sites are not shown. The arrows show positions of the Fermi levels for the
hole eoneentrations, whieh leads to the insulating behavior
states (.s2 = 0), which include the atomic y2_ z 2 (z2- x 2) orbitals ofthe bridge
sites A (A') and a nonbonding combination of the eg states at the corner sites
Band B'. Clearly, if there is exactly one electron per two sites, A and B (the
"charge-ordering" regime), the band .s k will be fully occupied and the system
will be a band insulator. The corresponding density of states is shown in
Fig.6.6. Note that the electron or hole doping away from x = 1/ 2 will place
the Fermi level at the high peaks of the density of states, and the system
will be unstable. Therefore, the pure CE phase may exist only at x = 1/2.
The corresponding stability conditions can be obtained by considering the
DE limit for magnetic interactions in the tt-bond Al-BI (6.5), and three H-
bonds: A I -B2 , A I -A 3 , and B I -B3 (6.12) , see Fig. 6.5. Their numerical values
are given by [34]: Jf1Bl = 0.174, Jf1B2 = 0.106, Jf1A3 = 0.076, and Jg1B3 =
0.117, which reveals a strong anisotropy reflecting the anisotropy of the AFM
spin ordering. Hopefully, the FM interaction in the tt-bond Jf1Bl appears
to be the largest, and the correct (FM or AFM) coupling in all nn bonds
may be stabilized by setting the isotropie SE interaction as JK B 3 < IJ8 1<
Jf1B 1 (the corresponding CE segment is shown in Fig.6.3). In the layered
perovskites, such as Lal / 2Sr3/2Mn04 [40], the interplane interactions in the
bonds A I - A 3 and B I - B 3 are supressed due the 2D geometry of the crystal
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 265
,,--,--~----,~--r--1 - - F
o
00
-FA
si ---- AG
---, A
structure, and the stability of the zigzag AFM ordering is given by a softer
eondition: J;{lB2 < IJ8 1 < J;{lBl' Therefore, the appearenee of the CE-type
AFM phase is eonsiderably faeilitated in the 2D ease.
The CE-type AFM state appears to be not only the loeal, but also the
global minimum of the total energy and ean be realized as the ground state
of the degenerate DE model at x = 1/2 (Fig. 6.7). Therefore, the existence of
the CE phase, at least at low temperatures, ean be readily explained by the
DE physics.
Even a small Jahn-Teller distortion around the bridge sites significantly
reduees J;{lB2' J;{lA3' and JJ?lB3' and additionally stabilizes the CE-type
AFM ordering [34].
Ironically, the role of the charge ordering, which was originally thought
to be the main ingredient (if not the source) of the CE phase in manganites,
is still rather controversial. One of the most puzzling aspects is the existence
of distinct "charge ordering" transition temperature in many compounds
(though with some exceptions - see [41]), which is typically higher than the
magnetic Neel temperature, as observed in the elastie neutron [40] and the
resonant X-ray [42,43] scattering experiments. These data seem to suggest
that the charge ordering should be the trigger that drives the AFM spin order-
ing. Naturally, the charge ordering can be obtained by including the intersite
Coulomb repulsion V explieitly into the model, whi:ch might be an explana-
tion for the high-temperature behavior above TN [44]. However, the charge
disproportionation induced by V will also reduce the kinetic energy of the
eg electrons. Therefore, the same approach applied to the low-temperature
regime will eventually suppress the FM DE coupling J;{lB 1 ,13 and the CE-
type AFM phase will be unstable (a quantitative account of the situation is
given in [34]). Thus, even if the charge disproportionation does take place, it
cannot be large. On the other hand, the theories of magnetic stability of the
CE phase [32,34,38,39] rely on the existence of the long-range AFM order and
eannot be easily reconcilled with the high-temperature behavior. A unique
theory that would account for both temperature regimes is not available at
the present stage. In any case, the behavior of manganites around x = 1/2
seems to be at odds with extreme points of view, based on either V» to or
U » to limits. 14
The idea of insulating zigzag chains can be generalized [37,39]. The objects
consisting of the straight-line segments (the "bridges") periodically connected
via the right- and left-turn corner-sites (Fig.6.6) will lead to the insulating
behavior at the hole concentrations x = 1 - NB~Nc' where NB and Ne
are the number of bridge- and corner-sites, respectively, and N is an integer
number of e g electrons per the complex of NB + Ne atoms. The finding
may have a direct implication to the insulating stripe structures observed in
La1-xCaxMn03 at commensurate carrier concentrations [14,15]. Note, how-
ever, that in the pure DE picture for x 2: 2/3, the zigzag stripes have a higher
energy than the simple straight stripe correponding to the AFM phase C
[39].15 Therefore, the stabilization of the zigzag strutures for x 2: 2/3 should
invoke an extra mechanism, presumably related to the lattice distortions [14].
An alternative scenario is a canted AFM structure [15]. In this case, however,
the zigzag chains will not be fully separated from each other, even in the DE
limit. The consequences of this spin canting on the insulating behavior and
the magnetic stability of the system have not been investigated yet.
One can also speculate that some of one-dimensional zizgag structures,
examples of which are shown in Fig.6.6, are not realized in the ground state
but may carry a considerable weight in the thermodynamics averages and
therefore contribute to the high-temperature behavior. 16 This might shed
some light on the change of periodicity of the orbital-order pattern observed
in some experiments [47]. If correct' the crucial issue in the understanding of
the high-temperature behavior of the "charge-ordered" manganites seems to
be the shape evolution of the ne-dimensional zigzag chains.
It is curious to note that the insulating zigzag structure can be expected
even for x = 0 (Fig. 6.6). Why is it not realized in practice? One reason is that
it would require very large SE coupling J S , beyond any realistic estimates.
So, the values of nn magnetic interactions in the DE limit are 0.123 and 0.193
for the t -.l--bonds within the xy plane and between the planes, respectively,
14 Formally, the structure of nn magnetic interactions in the CE-type AFM state can
be understood from the viewpoint of the canonical SE physics, Le. starting from
the atomic limit and treating the Coulomb U as one of the largest parameters
of the problem [45,46]. However, in a broader perspective, the picture is hardly
relevant to the realistic situation realized around x = 1/2, which is extremely
flexible and can be controlled by various means, switching not only between
different types of the magnetic ordering, but also between metalllic and insulating
states of the system [41].
15 For comparison, the phase C is unstable (Fig.6.3) and does not enter into the
problem at x = 1/2.
16 For example, at x = 1/2 the structure with NB = 1 and Ne = 3 is higher than
the CE-type ground state (NB = Ne = 1) only by 0.021to (rv 160 K) per Mn site.
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 267
and 0.234 for the tt-bond. Therefore, the three-dimensional structure can be
stable only if 0.193< IJsl < 0.234. From this point of view, there are much
bett er perspectives to design this structure in the two-dimensional case, which
is expected to be stable ifO.123 < IJsl < 0.234, Le. for considerably lower JS.
Another reason is that the magnetic energies at x = 0 may not playa primary
role, being much smaller than the energy gain caused by the lattiee distortion
[48,49], whieh will be iscussed in Sect.6.3.3.
Finally, we would like to note that the physies of zigzag chains may have an
interesting implication to another famous oxide compound - V 20 3 . The low-
temperature magnetic structure OfV20 3 can be also viewed as a combination
of antiferromagnetically coupled FM zigzag chains. Some aspects of magnetie
stability of this phase can be understood on the mean-field level, and related
with the anisotropy of electronie structure of V20 3 caused by the anisotropie
AFM spin ordering [50]. However, contrary to the manganites, Ueff plays
a more important role in V20 3 and contributes to the localization of t2g
electrons in the zigzag chains.
Fig. 6.8. Two diagonal modes describing small deviations of magnetic moments
near the antiferromagnetic ground state
true that due to the ad hoc form of the LDA + U functional, some important
details of the electronic structure of strongly correlated materials can be very
wrong in LDA+U, which may lead to an incorrect description of interatomic
magnetic interactions [58]. Our experience shows that the LSDA, though not
perfect, presents the most reliable reference point for the analysis of magnetic
properties of manganites. Some details will be discussed in Sect. 6.3.3.
°
to that proposed in [62], except that in order to calculate the total energy
change near rJ = we additionally apply the local-force theorem and express
8~E(rJ, q)I19=a through the change ofthe single-particle energies [26,58]. Using
J q , the spin-wave dispersion (SWD) in the FM state can be evaluated from
it
the well-known expression for the Heisenberg model [63]: w q = (Ja - J q ).
Similar analysis for the two spin sublattices in the (simple) AFM state
yields the following expression for the total energy change with respect to
the two diagonal modes shown in Fig. 6.8:
1 + 2 1 _ 2
E(rJ+,rJ_,q) = E(O,O,q) + 2Jq rJ+ + 2Jq rJ_. (6.14)
In the first case CI? +), the phase of the perpendicular component of the spin
magnetic moments changes as qRö, where Rö is the radius-vector of the site
i. In the second case ('!9 _), the phase gains an extra factor 7r when moving
from one spin sublattice to another. Then, the SWD in the simple AFM state
can be evaluated as [63]:
wq -- ~(J+J-)1/2
M q q . (6.15)
(6.16)
and
aaß(w)
D -_8i 1 "''''
V --.- aß
~ ~ 7rnn 7r nn t5 (cF - ckn ) , (6.17)
W + ZTJ k n
V r
N,ill(w) = 27r Jo dwRe [aaa(w)] , (6.18)
r x M R r DOS
8N
0
> -'"
"
55:
0
8"
0
'"
0
E ..,eil
u C! - - LaSr 2.
a
0
---LaCa _ c
b .!2
- LaPb ):;: ;:;
:::::"1
80 -r~~
J •
e ":'
t:
0-
"
;,.:
"i
2 3 4 2 3 4
ro (eV) ro (eV)
Fig. 6.9. Results ofvirtual-crystal (VC, dotted line) and super-cell calculations for
the R3/4Dl/4Mn03 alloy with RD = LaBa (dot-dashed line) , LaSr (long-dashed
line), LaCa (dashed line) and LaPb (solid line). Upper panel shows the spin-wave
dispersion (Zeft) and density of states (right). The notations of the high-symmetry
points are taken from [66]. Lower panel shows the diagonal optical absorption (left)
and the magneto-optical Kerr rotation (e, right figure )
has been tested by comparing it with direct supereell (SC) calculations for
X = 1/4 and RD = LaBa, LaSr, and LaCa. The results of the VC and
SC ealculations of the SWD, optieal and magneto-optical spectra in the FM
state are summarized in Fig.6.9. Onee the erystal strueture is fixed (eubie
with V = 60.884A3 in the present ealculations), the VC and SC methods
yield praetieally identieal results, which only weakly depend on the ehemical
nature of divalent s-element used for the doping (D = Ba, Sr, or Ca).
An interesting situation oeeurs when the LaMn03 is doped by the divalent
p-element Pb. The 6p and 6d states of Pb eonsiderably admix into the wave-
functions of manganites, eorrespondingly in the oeeupied and unoeeupied
parts of the spectrum, and through the intraatomie dipole 6p--+6d transi-
tions signifieantly enhanee intensities of the optieal absorbtion. The effect
is clearly seen already in the diagonal part of eonductivity. An even more
dramatie enhaneement is expeeted for the off-diagonal eonductivity and the
magneto-optieal properties. So, La3/4Pbl/4Mn03 may exhibit rather large
Kerr rotation [65]:
CI
OK
+.ZEK = -uxy(w)
,
u xx (w)J1 + 47riuxx (w)/w
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 271
Ox=O.1
+ .\"=0.2
~ r------,,-----,-~,_~ CI>
0
'" ....
8 0
'"
o
~'"
> - '"0 ....
•
I~ IV
0 3'
3~ 0
on
J,
iI Ci .s"
J, Ö
i I:}-()e- 0
0
2 3
IRml (units of ao)
(6.19)
'" F
a-
..;
o~ .:1
" A
00
..; 'I :'i il ! !~ !
Fig. 6.11. The minima points of Wq in the ferromagnetic state of Lal- xBaxMn03'
Depending on x and a the ferromagnetic ordering can be stable (F area) and
unstable with respect to the AFM states of the A- , C-, and G-type
scales linearly (quadratically) with t(a) . For the t(a)/ß. ex ratio corresponding
to the LSDA approach, the change of the SE interaction will always prevail
[27]. Thus, we have an interesting situation, where the sign of J 1 is determined
by the FM DE, and the volume dependence by the AFM SE interaction. The
compression of the sampIe should make the FM ordering unstable, which
suggests a possible experimental verification of the LSDA-based picture for
manganites.
The main directions in which this instability tends to resolve, obtained af-
ter analysis of global minima of wq for different x and a, are shown in Fig. 6.11.
This strongly resembles the magnetic phase diagram of doped manganites.
However, the similarity should be taken very cautiously, because the magnetic
interactions for DE systems are magnetie-state dependent (Seet.6.2.2) , and
the fact that the FM state is unstable with respect to some AFM ordering
does not neeessary mean that this AFM ordering will be stable and realized
as the ground state. It is true that the FM ordering ean be easily destroyed
in some eases, but it is mueh more diffieult to find out what will be the new
ground state of the system after destruction of the ferromagnetism. Gener-
ally, one ean propose a lot of collinear [21], noneollinear [33], and also the
inhomogeneous phase-separated [8] scenarios of the magnetic ordering in such
cases. The problem is still far from being fully investigated. In so me sense, it
has many things in common with the competition between different magnetic
structures on the very narrow energy range realized in other famous materials,
like bcc (Ö)-Mn [77] and fcc (y)-Fe [61,78].
The stability of the FM ordering around x = 0.5 can be very effieiently
controlled by the tetragonal distortion of the sampIe (cl a). 21 It has a dramatic
effect on w q , which becomes strongly anisotropie and unstable at one of the
zone boundaries (Fig.6.12). The instability tends to resolve towards the A-
type AFM ordering (wz < 0, while Wx > 0) for cla < 1, and the C-type
o
~xL-------~r--------~z
AFM ordering (wx < 0, while Wz > 0) for cla > 1. The effect is directly
related with the anisotropy of nn interactions .!J.J1 = J: Y - J{ , induced by the
distortion. For x = 0.5, the numerical values of these parameters are: J: Y =
20 and J{ = -36 meV for cla = 0.95, J: Y = -12 and J{ = 29 meV for cla =
1.05, to be compared with J: Y = J{ = 7 meV for cl a = 1. There are two main
contributions to .!J.J1 . The first one comes from the distance dependence of
effective transfer integrals t(a), in some analogy with the exchange striction
encountered long aga in MnO [80], and can be evaluated from (6.19).22 It
accounts for up to 30% of .!J.J1 . Another 70% should be related with the
orbital polarization .!J.n, induced by the tetragonal distortion (.!J.n = -0.07
and 0.06 for cla = 0.95 and 1.05, respectively). For example, the additional
population (depopulation) of one of the e g orbitals increases (decreases) the
kinetic energy and the DE coupling involving this type of orbitals [21].
Similar to the undistorted cubic case, the instability towards the A- and
C-type AFM ordering shown in Fig. 6.12 does not necessary guarantee that
these states will be realized as the ground states [33]. We will return to this
problem again in Sect. 6.3.4.
0=0.00 0=0.06
g'"
.~ ~
l! 0
.5
.g N
~ ci
~
~
O~~L-~~~~~
0.5 1.0 1.5 2.0 'ä.O 0.5 1.0 1.5 2.0
number of eg electrons number of eg electrons
LaMn03 is an insulator with a rather small energy gap ("" 1.1 eV, [81,82]).
At low temperatures, it exhibits the A-type AFM ordering. The crystal
structure is orthorhombic and featured by the large frozen-in cooperative
Jahn-Teller distortion (JTD) of the Mn06 octhedra (schematically shown in
Fig.6.13, in the xy plane).23 The magnitude of the JTD is characterized
by the parameter 8 = (d L - ds)/(d L + ds ), with dL(ds) being the long
(short) Mn-O bondlength. The experimental situation (roughly) corresponds
to 8 = 0.06 [4,5].
The cooperative JTD gives rise to the orbital ordering of (preferentially
occupied) 3x 2 - r 2 and 3y 2 - r 2 orbitals in the xy plane [83]. The JTD, the
orbital ordering, and the energy gap persist both below and above the Neel
temperature (TN = 140 K), being the main reason why LaMn03 is typically
categorized as a Mott insulator.
On the level of LSDA calculations, the insulating behavior can be ob-
tained (though with somewhat smaller band gap) by taking into account
simultaneously the experimental crystal structure and the A-type AFM spin
ordering [31,53].24 The result has a very clear explanation [84,85].
24 If one neglects the distortion or changes the type of spin ordering, the LSDA
solution turns out to be metallic.
25 Unfortunately, it is still very demanding computationally to treat spin sublattices
corresponding to two different orbital sublattices in the xy plane as nonequivalent,
and in the present calculations we map the total energy change on the Heisenberg
6 Orbital Degeneraey and Magnetism of Perovskite Manganese Oxides 277
C-type G-type
0
:>
~
0
g
" N
R Z rr X R Z r
LSDA ealculations of magnetie interactions Jt
for LaMn03 in the
reeiproeal spaee as a function of J ahn-Teller distortion and for two different types
of the of the orbital ordering in the z-direetion. The unit-eell volume is fixed to the
experimental one. The A-type AFM spin ordering was used as the reference point
The orbital ordering of the C-type stabilizes the A-type AFM spin ordering
around 8 = 0.06. For smaller 8, the state A becomes unstable with respect
to the spin rotations described by the mode '!9+ (Fig.6.8) in which the spin
canting between two AFM sublattices is combined with the incommensurate
spin-spiral ordering in the xy plane. It is interesting to note that alternative
J:
G-type orbital ordering stabilizes the A-type AFM spin ordering even for
smaller 8. The reason can be easily seen by transforming into the real
space. The behavior of the key parameters is shown in Fig. 6.15. In addition to
J: Y and J 1, there are the following important interactions: J 2, the second-nn
exchange between neighboring xy planes; and J:
Y , the third-nn interaction
model for the simple antiferromagnet (6.14). The mapping would be exaet if
the magnetie interactions were limited by the nearest neighbors. However, this
appears not to be the ease, at least in LSDA, being the main reason why we eannot
ealculate Wq. The experimental Wq in LaMn03 is typically interpreted in terms
of two parameters: Jt y and J 1 [86]. However, taking into aecount the existenee
of four different spin sublattices, and the low erystal symmetry of LaMn03, the
analytieal expression for Wq beeomes rather eomplieated, and it is quite possible
that there is an alternative interpretation of the experimental data involving
the longer-range interaetions (as for the A-type AFM ordering around x = 0.5
diseussed in Seet. 6.3.4). There are many examples when, even in insulators, the
range of magnetie interactions is not neeessarily limited by the nearest neighbors
[87]. In order to clarify the situation in LaMn03, it would be very useful to have
Wq measured throughout the whole Brillouin zone.
278 L Solovyev and K. Terakura
C-type G-type
Fig. 6.15. Dependence ofmagnetic interactions in the real space on the Jahn-Teller
distortion for two types of the orbital ordering in LaMn03' The unit-cell volume is
fixed to the experimental one
1. The A-type AFM state has the lowest energy amongst the eollinear states
F, A, C, and G [31,53]. Fine details of the magnetie ordering eaused by
Dzyaloshinsky-Moriya interaetions ean also be understood on the LSDA
level [20].
2. The optical spectra are in reasonable agreement with the experimental
data [30], including the anisotropy of optieal eonductivity.28 The rela-
tivistie effect may lead to the strong optieal nonreeiproeity, even in the
A-type AFM state [90].
26 Since two sublattices in the xy plane were treated as equivalent, J;Y in fact
represents the average of two interactions in the bonds 1-3 and 1-4 shown in
Fig.6.13.
27 Note also that the experimental orbital ordering cannot be stabilized by the
Coulomb correlations alone [22,88]. According to this scenario, the lattice effects
playa very important role. They determine the direction of the orbital ordering
(through the anharmonicity of the JTD) [88] and suppress the quantum fiuctu-
at ion effects of the Coulomb interaction [89].
28 K. Terakura, LV. Solovyev, and H. Sawada, Chap.4 in [41].
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 279
A very serious problem occurs if one tries to optimize the crystal structure
in the LSDA (or in the generalized gradient approximation - GGA). The
optimization substantially reduces 8 and makes the A-type AFM ordering
unstable [72J. This, of course, makes the situation very uncomfortable and
clearly indicates that something important is missing in the existing methods
of electronic structure calculations. However, so far no consistent solution of
this problem has been found. Much attention has been paid to the problem
of onsite Coulomb interactions in the framework of the LDA+U method
[30,55,56J. For LaMn03' the main idea of LDA+U calculations has been
inspired by results of the Kugel-Khomskii theory of the orbital ordering in
the degenerate Hubbard model [88J. However, the model analysis cannot be
easily transferred to the first-principles calculations.
The effect of the Coulomb U is two-fold. On the one hand, it enters the
denominator of the SE interaction, and decreases the absolute magnitude
of the latter. On the other hand, it is expected to enhance (through Ueff)
the orbital polarization and the tendencies towards the orbital ordering,
with all natural consequences for the interatomic magnetic interactions: the
FM DE interactions will be further suppressed by Ueff; and the SE interac-
tions will eventually obey the conventional Goodenough-Kanamori-Anderson
(GKA) rules, underlying the Kugel-Khomskii theory in the limit of small
tofU. However, does the LDA+U work in that way? The problem is that
by including a large U (rv 8eV, and adjusting the charge-transfer energy
so as to reproduce the small experimental band gap), the occupied t-spin
Mn(e g ) states will be pushed downwards into the O(2p) band. As a result,
the Mn(eg )-O(2p) hybridization will be increased, and the orbital ordering
will be very incomplete. 29 Therefore, a large U in the LDA + U will suppress
the SE interactions (both t2g and eg ), but the FM DE interactions will be
practically intact. This leads to the incorrect FM ground state for LaMn03,
even with experimental lattice parameters [30,72J. Thus, the irony is that
despite the apparent superiority, the LDA + U gives a much worse description
for the magnetic properties of LaMn03 than the regular LSDA.
In order to restore the GKA picture of magnetic interactions in LaMn03'
the Coulomb U should be huge, as in the first-principles Hartree-Fock (HF)
calculations [91J. However, the HF picture is in strong disagreement with the
spectroscopic data, both optical [81J and photoemission [82J, and cannot be
applied to the metallic state of doped manganites.
The LDA+U description can be improved to some extent by shifting all
Mn(3d) states upwards relative to the O(2p) band (in other words, by increas-
ing the charge-transfer energy) [58], in the analogy with the picture obtained
in the model HF calculations [92J. In the case of LaMn03' this gives a better
description of the magnetic interactions and the relative stability of the A-
29 Do not confuse this with the spin-magnetization densities plotted in [56]. The
spin-magnetization density is not the correct characteristic for the analysis of the
orbital ordering of t-spin Mn(eg) states in LaMn03.
280 I. Solovyev and K. Terakura
00-
0
X
....
0
3.....
N~
X 03
"
-:5
0
N
r 2 3 4
0
IRml (units of a)
Fig.6.16. Magnetic interactions in the A-type antiferromagnetic state of cubic
Lal /2 Bal / 2MnÜ3: the behavior in the reciprocal space (Zeft) and the distance
dependence in the real space (right)
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 281
IRml (unilS of a)
Fig.6.17. A-type antiferromagnetic ordering in Lat/ 2Bat/ 2Mn03 stabilized by
the tetragonal distortion c/a = 0.95. The unit-cell volume is fixed at Vo = ag
with ao = 3.876 A. The lower panel shows the behavior of magnetic interactions in
the reciprocal (left) and real (right) space. The upper panel shows the spin-wave
dispersion (left) and density of states (right)
the both FM and A-type AFM states, the parameters entering this model
can be very different for these two states. This difference should by no means
be considered as a surprise. The behavior is very natural for the DE system
and reflects the magnetic state dependence of magnetic interactions discused
in Sect.6.2.2. It presents a very interesting problem for first-principles band
structure calculations. We know that neither FM nor A-type AFM ordering
is locally stable around x = 0.5. Then, what will the magnetic ground state
look like: a simple canting or something else?
The A-type AFM ordering at x = 0.5 is stabilized by the tetragonal
contraction cja = 0.95, which significantly reduces J 1 up to 1 meV (Fig. 6.17).
In this case one can calculate wq using (6.15) for the simple antiferromagnet.
An interesting observation is that although the magnetic interactions spread
beyond the nearest neighbors, their contribution to r,.<}q is not so trivial in the
AFM case, and the SWD can still have a simple form along some directions
of the Brillouin zone (r-X, for example). This casts some doubts on the in-
terpretation of inelastic neutron scattering measurements for the doped AFM
manganites, which is typically done in terms of only nn interactions [95]. The
apparent simplicity of the SWD along some directions of the Brillouin zone
does not necessary mean that the magnetic interactions in the real space
should be restricted by the nearest neighbors.
On the contrary, the C-type AFM ordering is additionally stabilized by
the tetragonal elongation of the sampIe. An example of the phase diagram
of Lat - xSrxMn03 in the plane of x and cja, obtained from the total en-
ergy calculations for the collinear FM, A-, C-, and G-type AFM states in
282 I. Solovyev and K. Terakura
,,
,
,,
,,
1.05 ,, c
F ,,
,, ,
,
--~~ ... ~
~ 1.00 ,,
.... { G
,,
,,
0.95 ,,
A
,
such calculations. 30 The first more or less serious attempt to investigate the
electronic structure and relative stability of different magnetic states in the
2D Rl/2D3/2Mn04 compounds has been undertaken by Lee et al. [36], using
LSDA and LDA+U methods. 31 Below, we discuss this story in more detail
using the phenomenological virtual-crystal alloy Yl/2Sr3/2Mn04 as a proto-
type compound (with the experimentallattice parameters ofLal/2Sr3/2Mn04
[40]). The most striking result was that even in LSDA the system becomes
a band insulator in the zigzag AFM state (contrary to the half-metallic
behavior in the normal FM state). The basic idea of local stability of the
anisotropie AFM structures in the DE model, discussed in Sects.6.2.2 and
6.2.3, was largely inspired by this finding.
The distribution of the Mn(3de g ) and Mn( 4p) states, obtained for the
zigzag AFM ordering, is shown in Fig. 6.20. They reveal many similarities to
the TB model discussed in Sect. 6.2.3. Indeed, there is a clear splitting of the
t-spin 3de g states into the bonding band located around -0.5eV (without
the additional JTD around the bridge sites, J = 0), the nonbonding bands
around 0.7 e V, and the antibonding band around 1. 7 eV. Two nonbonding
bands are formed by the e g states of the corner atoms (B and B' in Fig. 6.5)
and the strongly localized y2 - Z2 and x 2 - Z2 states of the bridge atoms (A
and A', respectively). Even without the JTD, the system is insulating and
the band gap is about 0.3 eV. The JTD significantly widens the band gap
(0.7eV for J = 0.03). The partial densities of states in the bonding part of
30 80 atoms for the CE-type AFM ordering in the 3D perovskite and 56 atoms for
the 2D analog of the CE phase realized in La1/2Sr3/2Mn04 [40J.
31 Historically, the first LDA+U calculations for the CE-phase of Pr1/2Cal/2Mn03
were done by Anisimov et al. [56J. However, they did not analyze the relative
stability of magnetic states, which is presumably incorrect in LDA+U, similar to
the problem of LaMn03 discussed in Sect. 6.3.3.
284 I. Solovyev and K. Terakura
0=0 0=0.03
~ ro~-'~~--~A~:~3?x~1~_rl2----------~.-~~-.-,
i
2 2
A: y-z
2 2 2 2
B: 3z -r and X -y
~ 0 i-"r:.,..........-b;~~~...".,~~__q-..c.:,.--"---b:~~~......-::~~
o
V)
o
o
CI) V)
00
Cl 0
o
o B: X and y
-I 0 234 -I 0 234
Energy (eV) Energy (eV)
Fig.6.20. Partial densities of Mn(3deg) (top) and Mn(4p) (bottom) states for
y 1/2Sr3/2Mn04 in the zigzag antiferromagnetic state. Two panels show results
of calculations with (right) and without (Zeft) the Jahn-Teller distortion around
the bridge sites. The Fermi level (the top of occupied band) is at zero
the spectrum clearly shows the characteristic peaks at the band edges, which
is a signature of (quasi-) one-dimensional behavior.
The distribution of the Mn( 4p) states is very important for the analysis
of resonant x-ray scattering near the Mn K-absorption edge, which is consid-
ered as a very powerful tool for studying the charge and orbital ordering in
manganites [42]. The t-spin Mn( 4p) states repeat the characteristic bonding-
nonbonding-antibonding splitting of the Mn(3de g ) states around the Fermi
level. Such a behavior can be easily understood by considering the hybridiza-
tion between the Mn(3de g ) and Mn(4p) orbitals in the zigzag chain [34]. So,
the nonbonding part of the spectrum, which contributes to the lowest-energy
part (pre-edge) ofthe K -absorption, is composed mainly of the x (y) Mn( 4p)
orbitals of the bridge A (A') atoms, and carries the information about the
distribution of the eg states at the corner atoms.
Without the JTD, the zigzag AFM state appears to be unstable and
has a higher energy than the FM state [36]. However, even small JTD may
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 285
32 The band gap does not open, and the AFM solution becomes unstable.
33 In this section, all calculations for 3D compounds correspond to the unit-cell
volume Vo = 58.231 A..
286 I. Solovyev and K. Terakura
-
\0
.--
'2 "<t
=='
'"
'3
E 0
..8
>
~ N
,- I
V)
"<t ,/
0 totaL_'--
Cl
\0
J, -"
_---- 0
-------=~~--~~--~~4°
-4 -2 0 2 4 0 2 3
Energy (eV) ro (eV)
Fig. 6.21. LSDA calculations for the ferromagnetic phase of Lao. 7 Bao.3Mn03 with
c/a = 1. Left panel: total and partial densities of states (the Fermi level is at zero).
Right panel: the optical conductivity total as weil as partial contributions of t-
and t-spin states. The inset shows the integrated conductivity (see Sect. 6.3.1 for
details)
' lg
-
.--
'2
=='
2
x .y
3z 2 _/
2
~:,' _/ j..JO Cl
'"
'3 - -"", 9
0 ~
E
..8
0 I 2 3 4°
'" 0
> 0
~
::t.:..
~ (")
3
'-'
V) 0
'-'
0 --- N
Cl "<t
e> J- O'r,l.
-2 0 2 4 0 2 3 40
Energy (eV) ro (eV)
Fig.6.22 . LSDA calculations for the A-type antiferromagnetic phase of
Lao.5Bao.5Mn03 with c/a = 0.95. Left panel: partial densities of Mn(3d) states.
Right panel: corresponding optical conductivities Uxx (solid line) and Uzz (dotted
line). The inset shows the integrated conductivities
As was discussed in Sect. 6.2 .2, the eg-band dispersion will be (quasi-) two-
and one-dimensional in the case of A- and C-type AFM ordering, respectively.
In the 2D case, the eg-band structure is characterized by a direct gap between
eigenstates with the same k [98]. The gapsize depends on x, and for x = 0.5
it is of the order of to. Therefore, the interband optical absorption a~e,Hw)
will be gapped in the A-type AFM state of Lao.5Bao.5Mn03 (Fig. 6.22) . The
Mn(eg)-+Mn(eg) transitions, mainly between the states of the x 2 - y2 sym-
metry, give rise to the well-defined peak ofaxx near 1.5 eV, in good agreement
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 287
with w rv 2to expected from the tight-binding analysis [98].34 The existence
of the direct gap does not contradict with the metallic behavior and the
existence of the Fermi surface: the A-type AFM ordering is accompanied by
the large Drude conductivity (nw; ': : ' 2.1eV, while nw~ ': : ' 0.6eV) and large
N:ff(O) in the FM xy planes. There is a elear deviation from the canonical
DE behavior. The -J.-spin 3z 2 - r 2 states are located just above CF due to the
strong mixing with the t-spin 3z 2 - r 2 states via the interplane hoppings,
allowed for finite Ll ex . Another factor that affects the position of the -J.-spin
3z 2 - r 2 states is the -J.-spin Mn( t2g) manifold. It is also located just above
cF and may admix the 3z 2 - r 2 states via the next-nn hoppings between the
3z 2 - r 2 and xy orbitals in the xy plane. The transitions between the t- and
t-spin 3z 2 -r 2 subbands give rise to the 1.3eV peak of uzz(w). Accidentally,
both the position of this peak and its intensity is elose to those of the u xx
peak. Therefore, the anisotropy of optical conductivity is expected to be small
in the metallic A-type AFM phase realized around x = 0.5. The situation is
drastically different from the undoped LaMn03' where the A-type AFM or-
dering is accompanied by the pronounced anisotropy uxx(w) > uzz(w) in the
low-energy region of the spectrum [41]. The difference is primarily caused by
two factors. (i) Since the local spin magnetic moment m is larger in LaMn03'
Ll ex will also be larger. Therefore, the weight of the -J.-spin 3z 2 - r 2 states
near CF and their contribution to uzzCw) will be smaller. (ii) In LaMn03'
the A-type AFM ordering exists due to the cooperative JTD, which splits
the atomic Mn(e g ) levels into the low-Iying group of 3x 2 - r 2 and 3 y 2 - r 2
orbitals (see Fig.6.13), and the higher-Iying y2 - Z2 and Z2 - x 2 orbitals.
Since the contribution of the 3z 2 - r 2 character to the states of 3x 2 - r 2 and
3y 2 - r 2 symmetry is significantly reduced (only 25% of the total weight),
the intensity of uzz(w) in the low-energy region will also be reduced.
The optical conductivity for the C-type AFM phase of Lao.3Bao.7Mn03
(Fig.6.23) is in drastic contrast with the canonical DE behavior, for which no
interband transitions are expected in the low-energy region. (i) Contrary to
the DE picture (Fig. 6.2), the x 2 _y2 states form a rather broad band, which is
partly occupied. There is a strong interchain interaction of the t -spin x 2 - y2
states with the -J.-spin x 2 - y2 states, so the latter have a significant weight
near CF. This gives rise to the low-energy absorption in U XX ' (ii) The U zz
conductivity also displays a large peak near 1 eV, in the region of (forbidden in
the TB model) 3z 2 -r 2 -+ x 2 _y2 interband transitions. The latter transitions
become possible only if the x 2 - y2 band accumulates a significant weight of
the 3z 2 - r 2 states due to the interchain interactions, which is apparently
34 A realistic estimate for the effective transfer integral is to
rv 0.7 e V [27]. The
v - tzg
c 0
·S
;:I Cl
1
N
0 '§"""'
0
0- '"""'
..2 00 0
[0'"
-
:> 0
:.:..
a""
~ N ("l
'"-' 0 --(}"
rIl xx 0 '"-'
0 --- (}" N
0 v ,j, zz
0
0
-2 0 2 4 0 2 3 4°
Energy (eV) co (eV)
the case for modest ..1ex . There is also an appreciable Drude conductivity,
especially in the z-direction (w~ ~ 2.0eV, while w~ ~ 0.6eV).
Experimentally, the optical properties have been studied in detail mainly
for the FM state of manganites [99,100].35 The most striking feature of the
experimental conductivity for x = 0.3 compounds is the broad Drude-like
peak, which implies some kind of disorder existing in the system down to
the very low temperature. 36 For the thin-film sampIes, the experimental
magnetization is considerably smaller than the value expected for the fully
spin-polarized FM state, and the broadening of the conductivity spectrum
can be ascribed to the spin disorder [102].37 However, a similar broadening
35 Very recently, Okimoto et al. [101] have measured the optical conductivity of the
thin Lal_",Sr",Mn03 films. Being based on the analysis of the spin magnetization
and the resistivity in the plane of the film, and also on the analogy with band-
structure calculations, some regions of the phase diagram have been ascribed to
the AFM phases ofthe A- and C-type [79]. However, exact details ofthe magnetic
phase diagram as wen as the validity of the present classification for the appar-
ently AFM phases are still unclear. It would be very useful to conduct similar
optical measurements for the bulk Ndl_",Sr",Mn03, for which the magnetic phase
diagram is weIl classified [11].
36 Note also that although the experimental spectral shape differs substantially from
the LSDA calculations for the homogeneous virtual-crystal alloys (Fig. 6.21), the
integrated conductivities Neff agree reasonably wen (Neff rv 0.2 at w = 2eV
according to the LSDA calculations for Lao.7Bao.3Mn03 and the experimental
data for x = 0.3 sampIes [99]).
37 In fact, the spin-disordered phase may emerge as one of the possible solutions of
the DE model around x = 0.5, as was suggested by recent dynamical mean-field
studies [103]. For the purposes discussed in [103], the approach is equivalent to the
weIl-known coherent-potential approximation (CPA) employed for the analysis of
optical conductivity in [102].
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 289
-0.00
:!I---f-..., ........... 0.0 I !--If-----If----i
9 ---- 0.02
E 0 -- - 0.03
_u ....::
c:
"0 '"
:::"0
t> "
~0~====~~-2~---3~---4~-~5
(j) (eV)
has been seen also in the bulk Lal-xSrxMn03 samples, for which the spin
magnetization was well saturated [100], suggesting that there should be yet
another source of disorder in the system. Possible scenarios discussed in the
literat ure, from the strong correlation viewpoint on these materials, are the
(quantum) orbital disorder [104] and the fiuctuating Jahn-Teller distortions
[105]. Another possibility is the disorder in the La/ Sr sublattice combined
with the local crystal distortions. If correct, the spectral shape is expected to
depend not only on the averaged radius of the RH and D 2 + ions, but also
on the second moment of the ionic radii distribution [106].
The temperature dependence of the optical conductivity presents an even
more challenging problem. As the temperature increases, the spectral weight
transfers from the low-energy Drude-like part to the broad midinfrared peak
centered at w ~ 1 eV [99- 101], creating a pseudogap. The effects are believed
to be due to the dynamic Jahn-Teller distortion [107] and/or the phase sep-
aration [8].
The optical conductivity tensor for Yl / 2Sr3/ 2Mn04 (Fig.6.24) reveals
a strong anisotropy between the inplane (o-xx) and out-of-plane (o-zz) ele-
ments, expected for the layered perovskite structures [108]. The zigzag AFM
ordering gives rise to the large peak of O-xx in the midinfrared region [34].
Two experimental groups reported somewhat different values for the peak
position at low temperatures: 1.0eV [108] and 1.3eV [109]. Both are larger
than the 0.7eV obtained in the LSDA calculations without the JTD. The
JTD J = 0.03 brings the peak position to 1.0eV, in good agreement with
290 I. Solovyev and K. Terakura
2 4
ro(eV)
Fig. 6.25. Optical conductivity for the G-type antiferromagnetic phase of CaMn03
We discussed the electronic properties and main trends of the magnetic phase
diagram of doped perovskite manganites. We argued that at least at low
temperatures, many aspects of the apparently complex behavior of these
systems can be understood from the viewpoint of double exchange physics
by taking into account the realistic degeneracy effects for the itinerant eg
electrons. The main ideas have been illustrated by considering the analytical
solutions of the degenerate double exchange model and supported by results
38 The calculations has been performed for the cubic lattice with the experimental
volume.
6 Orbital Degeneracy and Magnetisrn of Perovskite Manganese Oxides 291
Acknowledgments
The present work is partly supported by New Energy and Industrial Tech-
nology Development Organization (NEDO).
References
1. J.M.D. Coey, M. Viret, and S. von Molmir: Adv. Phys. 48, 167 (1999)
2. E.O. Wollan and W.C. Koehler: Phys. Rev. 100, 545 (1955)
3. J.B. Goodenough: Phys. Rev. 100, 564 (1955)
4. G. Matsurnoto: J. Phys. Soc. Jpn. 29, 606 (1970)
5. J.B.A.A. EIernans, B. van Laar, K.R. van der Veen, and B.O. Loopstra: J.
Solid State Chern. 3, 238 (1971)
6. P.-G. de Gennes: Phys. Rev. 118, 141 (1960)
7. E.L. Nagaev: Usp. Fiz. Nauk 166, 833 (1996) (Russian). [English trans!.:
Phys.-Usp. 39, 781-805]
8. A. Moreo, S. Yunoki, and E. Dagotto: Science 283, 2034 (1999)
9. M.O. Dzero, L.P. Gor'kov, and V.P. Kresin: Eur. Phys. J. B 14, 459 (2000)
10. H. Kuwahara, T. Okuda, Y. Tornioka, A. Asarnitsu, and Y. Tokura in Science
and Technology of Magnetic Oxides, ed. by M. Hundley, J. Nickel, R. Rarnesh,
and Y. Tokura. MRS Syrnposia Proceedings No. 494, (Materials Research
Society, Pittsburg, 1998)
11. R. Kajirnoto, H. Yoshizawa, H. Kawano, H. Kuwahara, Y. Tokura, K.
Ohoyarna, and M. Ohashi: Phys. Rev. B 60, 9506 (1999)
292 I. Solovyev and K. Terakura
67. E.A. Balykina, E.A. Ganshina, G.S. Krinchik, A.Y. Trifonov, and 1.0. Troy-
anchuk: J. Magn. Magn. Matter. 117, 259 (1992); Y. Yokoyama et al. (1996):
private communication.
68. T.G. Perring, G. Aeppli, S.M. Hayden, S.A. Carter, J.P. Remeika, and S.-W.
Cheong: Phys. Rev. Lett. 77, 711 (1996)
69. J.A. Fernandez-Baca, P. Dai, H.Y. Hwang, C. Kloc, and S.-W. Cheong: Phys.
Rev. Lett. 80, 4012 (1998)
70. P. Dai, H.Y. Hwang, J. Zhang, J.A. Fernandez-Baca, S.-W. Cheong, C. Kloc,
Y. Tomioka, and Y. Tokura: Phys. Rev. B 61,9553 (2000)
71. J.A. Blackman and R.J. Elliott: J. Phys. C: Solid St. Phys. 3, 2066 (1970);
L.N. Bulaevskii and S.V. Panyukov: Pis'ma Zh. Eksp. Teor. Fiz. 43, 190
(1986) (Russian). [English transl.: JETP Lett. 43, 240-243]; P.M. Levy, S.
Maekawa, and P. Bruno: Phys. Rev. B 58, 5588 (1998)
72. H. Sawada, Y. Morikawa, K. Terakura, and N. Hamada: Phys. Rev. B 56,
12154 (1997)
73. G. Khaliullin and R. Kilian: Phys. Rev. B 61, 3494 (2000)
74. L. Vasiliu-Doloc, J.W. Lynn, A.H. Moudden, A.M. de Leon-Guevara, and A.
Revcolevschi: Phys. Rev. B 58, 14913 (1998)
75. L. Vasiliu-Doloc, J.W. Lynn, Y.M. Mukovskii, A.A. Arsenov, and D.A. Shuly-
atev: J. Appl. Phys. 83, 7342 (1998)
76. R. Kajimoto, H. Yoshizawa, H. Kawano-Furukawa, H. Kuwahara, Y. Tomioka,
and Y. Tokura (2000): e-print cond-mat/0005497
77. P. Mohn, K. Schwarz, M. Uhl, and J. Kübler: Solid State Commun. 102, 729
(1997)
78. V.P. Antropov, M.1. Katsnelson, B.N. Harmon, M. van Schilfgaarde, and D.
Kusnezov: Phys. Rev. B 54, 1019 (1996); Y. Kakehashi and N. Kimura: Phys.
Rev. B 60, 3316 (1999)
79. Y. Konishi, Z. Fang, M. Izumi, T. Manako, M. Kasai, H. Kuwahara, M.
Kawasaki, K. Terakura, and Y. Tokura: J. Phys. Soc. Jpn. 68, 3790 (1999)
80. D.S. Rodbell and J. Owen: J. Appl. Phys. 35, 1002 (1964); M.E. Lines and
E.D. Jones: Phys. Rev. 139, A1313 (1965); M. Kohgi, Y. Ishikawa, and Y.
Endoh: Solid State Commun. 11,391 (1972); G. Pepy: J. Phys. Chem. Solids
35, 433 (1974)
81. T. Arima, Y. Tokura, and J.B. Torrance: Phys. Rev. B 48,17006 (1993)
82. T. Saitoh, A.E. Bocquet, T. Mizokawa, H. Namatame, A. Fujimori, M. Ab-
bate, Y. Takeda, and M. Takano: Phys. Rev. B 51, 13942 (1995)
83. Y. Murakami, J.P. Hill, D. Gibbs, M. Blume, I. Koyama, M. Tanaka, H.
Kawata, T. Arima, Y. Tokura, K. Hirota, and Y. Endoh: Phys. Rev. Lett.
81, 582 (1998)
84. L.P. Gor'kov and V.Z. Kresin: Pis'ma Zh. Eksp. Teor. Fiz. 67, 934 (1998)
(Russian). [English transl.: JETP Lett. 67, 985-990]
85. LV. Solovyev and K. Terakura: J. Korean Phys. Soc. 33, 375 (1998); LV.
Solovyev and K. Terakura (1999): in Proceedings of Moscow International
Symposium on Magnetism, Moscow State University - Faculty of Physics,
Moscow, 155-158
86. K. Hirota, N. Kaneko, A. Nishizawa, and Y. Endoh: J. Phys. Soc. Jpn. 65,
3736 (1996); F. Maussa, M. Hennion, J. Rodriguez-Carvajal, H. Moudden, L.
Pinsard, and A. Revcolevschi: Phys. Rev. B 54, 15149 (1996)
87. E.J. Samuelsen, M.T. Hutchings, and G. Shirane: Physica 48, 13 (1970)
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 295
88. KI. Kugel and D.I. Khomskii: Zh. Eksp. Teor. Fiz. 64, 1429 (1973) (Russian).
[English transl.: Sov. Phys. JETP 37, 725-730]; KI. Kugel and D.I. Khomskii:
Usp. Fiz. Nauk 136, 621 (1982) (Russian). [English transl.: Sov. Phys.-Usp.
25, 231-256]
89. Y. Motome and M. Imada: Phys. Rev. B 60, 7921 (1999)
90. I.V. Solovyev: Phys. Rev. B 55, 8060 (1997)
91. F. Freyria Fava, P. D'Arco, R. Orlando, and R. Dovesi: J. Phys.: Condens.
Matter 9, 489 (1997); Y.-S. Su, T.A. Kaplan, S.D. Mahanti, and J.F. Harrison:
Phys. Rev. B 61, 1324 (2000)
92. T. Mizokawa and A. Fujimori: Phys. Rev. B 54, 5368 (1996)
93. KH. Ahn and A.J. Millis: Phys. Rev. B 61, 13545 (2000)
94. F. Aryasetiawan and O. Gunnarsson: Rep. Prog. Phys. 61, 237 (1998)
95. H. Yoshizawa, H. Kawano, J.A. Fernandez-Baca, H. Kuwahara, and Y.
Tokura: Phys. Rev. B 58, R571 (1998)
96. P. Mahadevan, J. Lee, and K Terakura (2000): unpublished.
97. H. Shiba, R. Shiina, and A. Takahashi: J. Phys. Soc. Jpn. 66, 941 (1997)
98. F. Mack and P. Horsch: Phys. Rev. Lett. 82, 3160 (1999)
99. M. Quijada, J. Cerne, J.R. Simpson, H.D. Drew, KH. Ahn, A.J. Millis, R.
Shreekala, R. Ramesh, M. Rajeswari, and T. Venkatesan: Phys. Rev. B 58,
16093 (1998)
100. E. Saitoh, A. Asamitsu, Y. Okimoto, and Y. Tokura: J. Phys. Soc. Jpn. 69,
3614 (2000)
101. Y. Okimoto, Y. Konishi, M. Izumi, T. Manako, M. Kawasaki, and Y. Tokura
(2001): submitted to Phys. Rev. B.
102. P.E. de Brito and H. Shiba: Phys. Rev. B 57, 1539 (1998)
103. A. Chattopadhyay, A.J. Millis, and D. Das Sarma (2000): e-print cond-
mat/0004151
104. S. Ishihara, M. Yamanaka, and N. Nagaosa: Phys. Rev. B 56, 686 (1997)
105. H. Nakano, Y. Motome, and M. Imada: J. Phys. Soc. Jpn. 69, 1282 (2000)
106. L.M. Rodriguez-Martinez and J.P. Attfield: Phys. Rev. B 54, R15622 (1996)
107. A.J. Millis, R. Mueller, and B.I. Shraiman: Phys. Rev. B 54, 5405 (1996)
108. J.H. Jung, J.S. Ahn, J. Yu, T.W. Noh, J. Lee, Y. Moritomo, I. Solovyev, and
K Terakura: Phys. Rev. B 61, 6902 (2000)
109. T. Ishikawa, K Ookura, and Y. Tokura: Phys. Rev. B 59, 8367 (1999)
110. J.H. Jung, KH. Kim, D.J. Eom, T.W. Noh, E.J. Choi, J. Yu, Y.S. Kwon,
and Y. Chung: Phys. Rev. B 55, 15489 (1997)
7 Magnetism in Ruthenates
D.J. Singh
7.1 Introduction
Investigations of magnetic transition metal oxides with metal-insulator tran-
sitions in their phase diagrams have revealed aremarkable range of interest-
ing, complex and often unanticipated phenomena, including high temperature
superconductivity, various charge, spin and orbital ordered states, triplet
superconductivity, giant magnetoelastic effects, and heavy Fermion behavior
[1]. A good example is provided by the perovskite manganites [2]. Although
known for several decades, these compounds were re-investigated in detail
during the last few years because of interest in the colossal magnetoresistive
effect, which is basically a metal-insulator transition occurring at or near
the magnetic ordering temperature for some compositions. What has been
revealed is a complex rich phase diagram including charge, spin and orbital
ordered phases resulting from an interplay of strong correlations, strong
lattice coupling, and band structure effects, as discussed by Solovyev and
Terakura in this volume.
Magnetism is less common in 4d and 5d compounds than in 3d and 4f
materials because the on-site Stoner and Coulomb parameters, land U are
lower, while the band widths, W tend to be larger, both because of the
more extended nature of 4d and 5d orbitals relative to the 3d case. However,
when it does occur it may be more interesting than typical 3d magnetism.
In particular, the more extended active orbitals make it much more likely
for intinerant electron physics to play an important role in such materials,
leading to interesting new physics, such as strong coupling to lattice degrees
of freedom. Furthermore, much stronger spin-orbit effects may be expected
compared with 3d systems. These may manifest themselves in unusually
strong magnetocrystalline and magnetooptical effects.
Perovskite derived ruthenates provide a particularly interesting example
of the new physics that can arise from these effects. Although they are all
based on octahedrally coordinated RuH, they displaya fascinating variety of
magnetic and electronic states, often with experimental signatures of strong
coupling to the lattice. These include ferromagnetism, antiferromagnetism,
paramagnetism, with both metallic and insulating ground states. Further-
more, the layered perovskite Sr2Ru04 is an unconventional, probably triplet
paired, superconductor [3,4]. The pairing mechanism is strongly believed to
298 D.J. Singh
because of the high magnetocrystalline anisotropy, these twins will have mag-
netizations in different directions. The possible contributions of the resulting
twin boundaries to transport properties should be kept in mind when inter-
preting data on "single crystal" bulk and thin film sampIes of SrRu03 (see
also [34]). In any case, based on transport data, it was concluded by Allen
and co-workers [30], that SrRu03 should be characterized as a "bad metaI" ,
since e.g. it does not show saturation of the resistivity p(T) at the Ioffe-
Regel limit. These "bad metaI" properties have also been seen in a number
of other experiments [40], including the same high quality sampIes for which
quantum oscillations were observed [37]. The Hall coefficient also behaves
unusually, changing sign around 50 K [44]. Strong signatures of the Curie
temperature are seen in p(T) implying strong magnetic scattering of carriers
[40,41]. Even at low T, the infrared optical conductivity behaves unusually in
its frequency dependence [42]. Okamoto and co-workers [43] reported a com-
bined photoemission, inverse photoemission and density functional study of
SrRu03. They found dispersive band features corresponding to the calculated
electronic structure, but also a rather large depletion of the spectral weight
near E F implying incoherency of the electronic structure, presumably due
to correlation effects. Here, these low energy correlations are presumed k-
dependent, and may very weIl be mediated by soft spin fluctuations.
In discussing the electronic structure of oxides, it is generally helpful to
begin with an ionic picture. SrRu03 then contains nominal RuH ions. The
4d states are divided into t2g and eg manifolds. In the octahedral crystal
field, appropriate to the perovskite structure, the t2g manifold is lower than
the eg , so one expects for the low spin arrangement four t2g electrons per
Ru. This would at first sight seem to provide a very reasonable description
of the properties of SrRu03' as the nominal low spin magnetization would
then be 2 f.JB per Ru. This is quite in ac cord with the experimental value
of ;:;::; 1.6 MB when one allows for some reduction due to hybridization. The
valence electronic structure from low energy to high should then consist of
o 2p derived states, followed by exchange split Ru t2g levels around E F ,
and finally unoccupied Ru e g levels. However, the density functional band
structure is not in accord with this simple picture.
In order to understand the basic features of the electronic structure, we
begin with the ideal cubic perovskite structure (note that the actual crystal
structure involves a rotation of the octahedra). The calculated local spin
density approximation (LSDA) ground state is ferromagnetic with a spin
magnetization of 1.17 f.JB. Interestingly, according to [31] simple antiferro-
magnetic configurations cannot be stabilized in the LSDA, implying that the
magnetism is itinerant, as opposed to local moment, in character. This is very
different from most 3d oxides. The energy bands are shown in Fig.7.1. As
may be seen, there is a single continuous manifold of hybridized Ru 4d - 0
2p bands extending from ;:;::; -8 eV to ;:;::; 6 e V relative to E F with substantial
300 D.J. Singh
. 10r~----t
X---:M';-----*
A ----~r:------!M'
Fig. 7.1. LSDA band structure ofSrRu03 with the ideal cubic perovskite structure,
after [29]. Majority (minority) bands are shown as solid (dashed) lines. The self-
consistent spin moment for this idealized structure is 1.17J-tB per formula unit
·2
·4
·8 ·6 ·4 ·2 o 2
E(eV)
exchange splittings through this range. The density of states (DOS) and Ru
4d projection, as given in [29], are shown in Fig. 7.2.
Qualitatively, the bands near the bottom (~ -8 eV upto -5 eV have
mainly 0 2p-e g bonding character, those from ~ -5eV upto -2eV, are
mainly 0 p7r like with an add-mixture of t2g character, those from ~ -2 eV
to 1 eV are t2g - 0 p hybrids, and those above are antibonding eg - 0 pa
in character. However, it should be emphasized that the bands have mixed
character throughout. Significantly, near E F the bands have approximately
2/3 Ru t2g character, with the remaining 1/3 being 0 p in origin. These are
formally, antibonding t2g - p7r bands.
As is common in perovskites with A-site cations, like Sr, that do not
participate significantly in the hopping, there are non-dispersive bands along
7 Magnetism in Ruthenates 301
4 SrRu0 3
~O
Z
-2
-4
-8 -6 -4 -2 0 2
E(eV)
Fig.7.3. LSDA eIectronic DOS of SrRu03 in its actual Pbnm structure. The DOS
is on aper formula unit basis, with the majority (minority) spin channel shown
above (below) the axis. The dashed line clenotes the Ru cl contribution
0.04 ,------,-----~------,----..,
0.02
S"
~ 0 ~--+-<>-....-.,..--t--...-....
~
W
-0.02
-0.04 Ca R u~ in Sr struct.
-0.06 0'------,OL
.5----"""'---------,J 5 =-='''------!2
1 .~
M (ILB/Ru)
Fig. 7.4. Fixed spin moment LSDA calculations of the energy of CaRu03 as a func-
tion of magnetization in its actual crystal structure and in the crystal structure of
SrRu03, after Mazin and Singh [31]
304 D.J. Singh
8 CaRu03
E(eV)
Fig.7.5. LSDA electronic DOS of paramagnetic CaRu03 after [31]. The dashed
line denotes the Ru d contribution
z x r z
Fig. 7.6. LSDA band structure of Sr2Ru04 after [76]. The long r-z direction is in
the bct basal plane, while the short r-z direction is along the c-mds. E F is at 0
z
Fig. 7.7. LSDA basal plane Fermi surface of Sr2Ru04 after [83]. The solid lines
are from direct calculations [76], while the dashed lines are a tight binding fit. The
corners of the plot are rand Z points, while the center is X. The weak c-axis
dispersion is refiected in the slight deviation from 4-fold symmetry in this plot
d xz and dyz hybridize with one in-plane 0 and the apical 0 2p states, which
are further away and do not have a direct hopping channel to a neighboring
Ru, whereas the d xy hybridizes with only in-plane 0 2p states, which are of
course coupled to neighboring Ru atoms. In this simple picture [5,6], the dxy
band would have twice the width of the d xz and dyz bands, because it has
hopping via both neighboring in-plane 0 as opposed to only the one along x
or y, respectively. This, in fact, is very dose to the actual situation, as may
be seen from the band structure. It may also be noted that because of the
tetragonal crystal field, the d xy is centered lower than the d xz and dyz bands,
with the result that the band maxima at X are quite dose.
The validity of the simple tight-binding picture, discussed above, is also
manifested in the Fermisurfaces. As mentioned, they are highly two-dimen-
sional, and consist of cylindrical sections along the c-axis. These are a nearly
circular section centered around r (denoted y) and two nearly square sec-
tions, cx and ß centered around rand X, respectively.
The y section is the two-dimensional surface that comes from the d xy
orbital. In the simple nearest neighbor approximation, it should be a circular
cylinder, which it is to a remarkably good approximation. The d xz and dyz
bands yield one-dimensional sections - flat sheets perpendicular to x and y,
respectively. As may be seen, the cx and ß are quite dose to this, allowing for
some second neighbor hopping and minor reconnection at the intersections.
Mazin and Singh have discussed the magnetic fluctuations that may be
expected based on this band structure from the point of view of a possible
pairing interaction [5,6]. Noce and co-workers [91] investigated the normal
state properties of Sr2Ru04 within a the framework of near criticallity of
a magnetic character. They showed that a semi-quantitative explanation of
the experimental normal state properties is possible, within a self-consistent
spin fluctuation theory provided that the interactions are mostlyon the d xy
sheet. Mazin and Singh identified two contributions to X(q). The first is
a weakly ferromagnetic tendency due to the 0 contribution to the Stoner I,
as discussed above for (Sr,Ca)Ru03' This factor is important because of the
high Stoner renormalization, 1/(1- IN) = 9, somewhat larger than deduced
from the experimental susceptibility [3], X/Xband = 7.3. The difference in the
denominator makes X/Xband very sensitive to the precise value of I. Mazin
and Singh estimated the q dependence of I, I(q) by using Stoner model,
discussed for (Sr,Ca)Ru03, with values of the partial Ru and 0 contributions
to N(EF) and the atomic-like I Ru and 10 from the calculations. In particular,
they smoothly interpolated between the full value for Sr2Ru04, i.e. I(q) =
0.43eVat r and the 14% smaller Ru only value at the zone corner, (7l',7l').
This resulted in the variation shown in Fig. 7.8. Using only the ferromagnetic
fluctuations, they obtained a mass renormalization due to spin fluctuations,
(1+).8) ~ 3. This is large, but still significantly smaller than the experimental
values of 3.4 to 5 depending on the Fermi surface sheet.
7 Magnetism in Ruthenates 309
0.4
0.3
~
g 0.2
00 (rt,rt)
q
Fig. 7.8. Variation of l(q) from r to the zone boundary in Sr2Ruü4, according to
the model of Mazin and Singh [5J. The model sets 1(0) to the calculated Stoner I
and l(q = (7r,7r) to the Ru only value, 1Ru
The second contribution is from the Fermi surfaces. As may be seen, the (X
and f3 sheets are elose to one-dimensional, and therefore are strongly nested.
Calculation of the band structure contribution to X involves matrix elements
and I (q). However, lacking a proper linear response, calculation, Mazin and
Singh [6] set the matrix element to its full value between bands of the same
character and zero otherwise and used the Stoner renormalization from the
model discussed above:
Xo(q) Xo(q)
(7.1)
X(q) = 1 - I(q)Xo(q)
~ ----~~~~~~--~~
1- I(q)N(O) - I(q)Xn(q) ,
where Xn is the nesting dependent contribution. The bare susceptibility Xo(q)
is shown in Fig. 7,9,
In spite of the ferromagnetic tendency of l(q), X(q) is slightly higher at
the zone corner than at rand in addition shows strong ridges corresponding
to the nesting of the d xz and dyz Fermi surfaces, The peaks at the inter-
section of these ridges, occurring at q ~ (27r /3a, 27r /3a), are particularly
pronounced. A similar feature emerges from arecent RPA calculation based
on an empirical tight binding band structure [92],
From an experimental point of view, the nesting related enhancement of
X(q) at q ~ (27r/3a,27r/3a) has been elearly seen in spin polarized neutron
experiments, The temperature dependence consistent with an itinerant band
structure origin [93-95]. Ishida and co-workers [95], using NMR, find that
the spin fluctuations associated with this peak are strongly anisotropie, with
the out-of-plane component of the dynamical susceptibility larger by a fact ar
of ~ 3 than the in-plane. However, these measurements do not find evidence
for a ferromagnetic background in the susceptibility, even when searched for
at small scattering angles and in wide energy ranges, although it should be
noted that a weak q dependence of X may be hard to observe with neutrons.
17 0 and Ru NMR measurements [26,95,96] do show exchange enhanced spin
310 D.J. Singh
350
14/mmm
0~~~--~0.~5------~------~1~.5------~
Ca 2 Ruq X Sr 2 Ruq
From this, they inferred a coupling between the octahedral compression and
the orbital polarization, at least qualitatively, supporting the conelusions of
Fang and Terakura. Anisimov and co-workers, very recently, extended the
work of Fang and Terakura by performing LDA+U and dynamical mean
field calculations, also finding a competition of ferromagnetic and antifer-
romagnetic interactions and orbital polarization controlled by the structure
[102]. Also, as expected, with a suitable choice of U, they produce a Mott
insulating state for Ca2Ru04'
The GGA calculations of Fang and Terakura, which show energetically
elose antiferromagnetic and ferromagnetic states and a small gap, seem to
present an alternative to the widely held view that Ca2Ru04 is a Mott
insulator. In particular, with the proximity of ferromagnetic and antiferro-
magnetic instabilities, the magnetism will take on a local moment character,
explaining the Curie-Weiss behavior above TN, while the insulating gap would
explain the transport and specific heat data for the antiferromagnetic phase.
One problem with this interpretation is that GGAs appear to enhance the
tendency towards magnetism in ruthenates (e.g. producing a ferromagnetic
ground state in Sr2Ru04), while the LSDA does not produce a gap. It should
be noted, however, that Fang and Terakura argue, based on the narrow band
width, that the GGA may be better in Ca2Ru04' However, another difficulty
with this band interpretation is that within a band picture, one would expect
structure in p(T) at T N perhaps crossing over to a less resistive, variable range
hopping form above TN, which is not observed. Thus the preponderance of
the evidence supports the view that the "S" phase of Ca2Ru04 is a Mott
insulator.
.10r
):------~z;------b-
x-----:r~z
Fig. 7.11. Band structure of ideal tetragonal structure Sr3Ru2ü7 after [114]
it is highly two dimensional and metallic. Cao and co-workers reported ferro-
magnetism at 104 K with additionallower temperature transitions [65], while
other sampies were not ferromagnetic at these temperatures but exhibited
transitions below 20K [68,69J. Meanwhile, Huang and co-workers found no
evidence for any long range magnetic order down to 1.6 K [105J; Perry, and
Ikeda and co-workers report that single crystals grown by a floating zone
technique are paramagnetic, with strongly enhanced susceptibility and high
Wilson ratio, leading to the conclusion that the material is on the verge of
ferromagnetism [106- 109J.
Unlike Sr2Ru04, Sr3Ru207 sampies have at least disordered rotations of
the RU06 octahedra [110,105,111]. Recently, Shaked and co-workers [111,112]
have shown that the distortions are ordered and refined the crystal structure
into spacegroup Bbcb with a rotation of approximately 7°.
The electronic structure of Sr3Ru207 has been studied using ARPES
[64J and density functional calculations both for the ideal tetragonal struc-
ture [113J and for the refined Bbcb structure [114J. The band structure and
Fermi surface for the tetragonal structure are shown in Figs.7.11 and 7.12,
respectively. As mentioned, the three Fermi surfaces of Sr2Ru04, which has
one RU02 layer per cell, may be regarded as arising from the three Ru t2g
orbitals. The the d xy orbital gives rise to a round cylindrical electron-like
sheet centered at r (Z) and the d xz and dyz orbitals provide flat sheetlike
sections perpendicular to k y and k z , respectively, that after reconnection
become square cylindrical sections around X and r along with strong nesting.
To a first approximation, the Fermi surfaces of tetragonal Sr3Ru207 may be
thought of as deriving from the six same bands (three from each RU02 layer)
with bonding - antibonding (odd - even) splittings due to the interaction
between the RU02 sheets comprising the bi-layer. However, as may be seen
from the lack of four-fold symmetry in the basal plane Fermi surface there
7 Magnetism in Ruthenates 315
Fig. 7.12. Basal plane Fermi surfaces of ideal tetragonal structure Sr3Ru2Ü7 af-
ter [114]
Singh and Mazin also find large changes in the Fermiology relative to the
tetragonal structure, and in particular, beyond the zone folding effects, they
note the introduction of Fermi surface sections that change topology with
very small shifts in energy. In this regard, Perry and co-workers reported
transport and thermodynamic evidence for unusual critical fiuctuations that
may be associated with Fermi surface instabilities [109]. However, the LSDA
ferromagnetic ground state needs to be reconciled with the experimentally
observered lack of long range ordering.
The LSDA calculations yielded a self-consistent ferromagnetic (FM) solu-
tion with a spin magnetization of 0.80 f.lB/Ru and an energy of -23 meV /Ru
relative to the non-spin-polarized case. Calculations were also reported for
antiferromagnetic configurations with a c(2 x 2) in-plane ordering, and having
adjacent Ru ions in the two planes of the bilayer polarized parallel. However,
no self-consistent magnetic configuration was found. Thus it may be con-
cluded that the LSDA magnetic character within each plane is itinerant.
However, calculations in which the Ru ions in a layer were ferromagnetically
aligned, but the layers were stacked antiferromagnetically (so each bi-layer
had one spin up and one spin down RU02 layer) did yield a stable magnetic
solution (denoted AF-A in the following), in this case with an energy of
- 20 meV/Ru and a Ru moment (as measured by the moment in a Ru sphere)
only 14% smaller than the ferromagnetic solution.
The energy difference between the FM and AF-A solutions contains two
parts within the most simple model. The first is the inter action between
the two planes comprising the bi-layer. This, in turn, has a ferromagnetic
part, originating from the Hund rule energy on the bridging oxygen, and
an antiferromagnetic superexchange interaction. Based on the FM - AF-
A energy difference, it may be concluded that the former is slightly stronger.
In any case, it is likely much larger than the inter bi-layer coupling through
the rock-salt layers. This expectation is based on the geometry (hopping via
two oxygens with unfavorable bond angles) and the fact that the k z dispersion
is considerably less than the anti-symmetrie, symmetrie band splittings due
to the interaction between the planes comprising the bi-layer. (Note that
the exchange constants are usually quadratic in hopping and the dispersion
linear.) Singh and Mazin speculated about the consequences as they may
relate to experiment in a likely scenario. Supposing that the inter-bilayer
interaction is anti-ferromagnetic (as expected for a superexchange coupling)
and much weaker than the intra bi-layer coupling, one may conclude that the
latter is ferromagnetic with a strength of order 3 meV /Ru. In this scenario,
the ground state would be antiferromagnetic, consisting of ferromagnetic bi-
layers, stacked antiferromagnetically. On the other hand, direct exchange
interactions, if dominant, would yield ferromagnetism.
In any case, Sr3Ru207 was discussed [114] as a possible case for the
Kosterlitz-Thouless theory of quasi-2D magnetic phase transitions [116]. The
difference from the textbook case is that the in-plane magnetism is itiner-
7 Magnetism in Ruthenates 317
anti the same qualitative picture applies, still, and so the 3D long range
order (LRO) transition temperature should be logarithmieally suppressed
compared to the ferromagnetic-paramagnetic energy difference. One may
safely assume that the inter-bilayer coupling is less than 30 K (the interplane
coupling within a bilayer), probably very much less, thus bringing about
a logarithmic Kosterlitz-Thouless suppression. One would, however, expect
strong itinerant spin fluctuations of ferromagnetic character in planes for tem-
peratures weIl above the LRO transition temperature; these should manifest
themselves, for instance, in specific heat and magnetie susceptibility. Within
such a pieture, classieal fluctuations due to the softness of the magnetie
interactions in the c-direction suppress the transition temperature.
Ikeda and co-workers reported a detailed study of single crystal floating-
zone Sr3Ru207 including magnetie susceptibility, specific heat and resistivity
plus magnetic measurements under pressure [107). They report a susceptibil-
ity maximum at T max = 16K accompanied by structure in the resistivity,
but conclude that long range antiferromagnetie order does not set in, based
on the high nearly isotropie X(T) below Tmax and thus that the ground state
is a paramagnetic Fermi liquid on the verge of ferromagnetism. In fact, the
measured X(T) is "" 15 times larger than that of Sr2Ru04 [115) and at least as
isotropie at low T. Certainly, this is not expected with simple local moment
ordered magnetism, especially in a material with strong magnetocrystalline
anisotropy as is the case for magnetic ruthenates [19).
However, things are less clear cut in the itinerant metallic scenario above.
First of aIl, in the ordered itinerant case, there is a Stoner continuum that
contributes to X(T) along the direction of the moments (the low X(T) direc-
tion); normally this is a very small effect, but here the magnetocrystalline
anisotropy is expected to be very large [19) and the Fermi liquid very soft
as evidenced by the strong Stoner renormalizations. Secondly, it should be
noted that as long as the anisotropy is not in the trivial c-axis direction, the
twinning of orthorhombic sampIes will mean that there will be no orienta-
tion where the applied field is aligned with the moments, again presumably
lowering the anisotropy of X in the magnetic state. However, both of these
mechanisms for lowering the anisotropy in X would require fortuitous strong
numerical coincidences to explain the observed isotropy of the measured X in
a magnetieally ordered state. On the other hand, the measured temperature
dependence of the two components of X is notably different, whieh is not
simply anticipated for a standard enhanced paramagnet either.
As mentioned, there is a magnetic metal phase at low temeprature in the
SrxCa2-xRu04 phase diagram for Sr concentrations between x = 0.2 and
x = 0.5. This phase shows similarities to the low temperature behavior of
Sr3Ru207 in that it shows a susceptibility peak but not long range order.
One explanation of that phase is that disorder in the octahedral tilts results
in a glassy state due to disorder in the interatomie exchange or simply due to
magnetocrystalline anisotropy. According to the LSDA calculations that have
318 D.J. Singh
7.7 Conclusion
Perovskite derived ruthenates display a remarkably rich variety of magnetic
and electronic states considering that they are all isovalent oxides based on
the same octahedrally coordinated RuH building block. As discussed above,
much can be learned from the interplay of density functional calculations and
experiment. Still, these materials display a lot of interesting physics beyond
this level, with evidence for important k-dependent electron correlation ef-
fects (e.g. strong scattering by various sorts of spin-fluctuations), and, at least
in Ca2Ru04, Hubbard-type correlations. Much of this variety comes about
because of the closely competing interactions - magnetic vs. non-magnetic,
via itinerant Stoner physics; ferromagnetic vs. antiferromagnetic, via the 0
contribution to I vs. nesting and superexchange - all coupled strongly to
the lattice through band structure effects (band width and crystal field), and
the strong magnetocrystalline anisotropy due to the 4d magnetic ion. These
competitions lead both towards a sensitivity to sampIe quality as weIl as
a tendency towards ground states dominated by frustration and/or fluctua-
tions, like the non-Iong-range ordered magnetic state in the (Ca,SrhRu04.
As discussed above, there mayaIso be some quite interesting two-dimensional
physics at play in the layered ruthenates. In any case, two things are worth
bearing in mind about the future. First of all , the extreme sensitivity of
7 Magnetisrn in Ruthenates 319
Acknowledgments
I am very grateful for many helpful discussions and collaboration with Igor I.
Mazin. I have enjoyed fruitful discussions with G. Cao, J.E. Crow, Z. Fang,
R.P. Guertin, S. Ikeda, S.R. Julian, A.P. Mackenzie, Y. Maeno, S.E. Nagler,
S. Nakatsuji, D.A. Papaconstantopoulos, W.E. Pickett, A.J. Schofield and K.
Terakura. Work at the N aval Research Laboratory is supported by the office
of Naval Research.
References
1. M. Imada, A. Fujirnori, and Y. Tokura: Rev. Mod. Phys. 70, 1039 (1998)
2. J.M.D. Coey, M. Viret, and S. von Molnar: Adv. Phys. 48, 167 (1999)
3. Y. Maeno, H. Hashirnoto, K. Yoshida, S. Nishizaki, T. Fujita, J.G. Bednorz,
and F. Lichtenberg: Nature 372, 532 (1994)
4. T.M. Rice and H. Sigrist: J. Phys. Condens. Matter 7, L643 (1995)
5. 1.1. Mazin and D.J. Singh: Phys. Rev. Lett. 79, 733 (1997)
6. 1.1. Mazin and D.J. Singh: Phys. Rev. Lett. 82, 4324 (1999)
7. Y. Kitaoka, K. Ishida, K. Asayarna, S. Ikeda, S. Nishizaki, Y. Maeno, K.
Yoshida, and T. Fujita: Physica C 282-287, 210 (1997)
8. L. Tewordt: Phys. Rev. Lett. 83, 1007 (1999)
9. H. Mukuda, K. Ishida, Y. Kitaoka, K. Asayama, Z. Mao, Y. Mori, and Y.
Maeno: J. Phys. Soc. Jpn. 67, 3945 (1998)
10. E.V. Kuz'rnin, S.G. Ovchinnikov, and 1.0. Baklanov: Phys. Rev. B 61, 15392
(2000)
11. M. Matzdorf, Z. Fang, Isrnail, J. Zhang, T. Kirnura, Y. Tokura, K. Terakura,
and E.W. Plurnmer: Science 289, 746 (2000)
12. P.K. de Boer and R.A. de Groot: Phys. Rev. B 59, 9894 (1999)
13. J.J. RandaH and R. Ward: J. Am. Chern. Soc. 81, 2629 (1959)
14. R.J. Bouchard and J.L. Gillson: Mater. Res. BuH. 7, 873 (1972)
15. A. Callaghan, C.W. MoeHer, and R. Ward: Inorg. Chern. 5, 1572 (1966)
16. J.M. Longo, P.M. Raccah, and J.B. Goodenough: J. Appl. Phys. 39, 1327
(1968)
17. C.W. Jones, P.D. Battle, P. Lightfoot, and W.T.A. Harrison: Acta Cryst. C
45, 365 (1989)
320 D.J. Singh
46. D.C. Worledge and T.H. Geballe: Phys. Rev. Lett. 85, 5182 (2000)
47. R. Meservey and P. Tedrow: Phys. Rep. 238, 173 (1994)
48. Z. Zhang and S. Satpathy: Phys. Rev. B 44, 13319 (1991)
49. J.F. Janak: Phys. Rev. B 16, 255 (1977)
50. D.J. Singh and J. Ashkenazi: Phys. Rev. B 46, 11570 (1992)
51. I. FeIner, I. Nowik, I. Branaric, and M. Gospodinov: Phys. Rev. B 62, 11332
(2000)
52. T. Kiyama, K. Yoshirnura, K. Kosuge, Y. Ikeda, and Y. Bando: Phys. Rev.
B 54, 756 (1996)
53. J.C. Slater: Phys. Rev. 49, 537 (1936)
54. E.C. Stoner: Proc. Roy. Soc. Lond., Sero A 165, 372 (1938)
55. O.K. Andersen, J. Madsen, U.K. Poulsen, O. Jepsen, and J. Kollar: Physica
B&C 86-88, 249 (1977)
56. G.L. Krakso: Phys. Rev. B 36, 8565 (1987)
57. S. Nakatsuji, S. Ikeda, and Y. Maeno: J. Phys. Soc. Jpn. 66, 1868 (1997)
58. M. Braden, G. Andre, S. Nakatsuji, and Y. Maeno: Phys. Rev. B 58, 847
(1998)
59. A.V. Puchkov, M.C. Schabei, D.N. Basov, T. Startseva, G. Cao, T. Tirnusk,
and Z.-X. Shen: Phys. Rev. Lett. 81, 2747 (1998)
60. G. Cao, S. McCall, J. Bolivar, M. Shepard, F. Freibert, P. Henning, J. Crow,
and T. Yuen: Phys. Rev. B 54, 15144 (1996)
61. G. Cao, S. McCall, M. Shepard, J.E. Crow, and R.P. Guertin: Phys. Rev. B
56, 321 (1997)
62. G. Cao, S. McCall, M. Shepard, J.E. Crow, and R.P. Guertin: Phys. Rev. B
56, 2916 (1997)
63. S. Nakatsuji, and Y. Maeno, Phys. Rev. Lett. 84, 2666 (2000)
64. A.V. Puchkov, Z.-X. Shen, and G. Cao: Phys. Rev. B 58, 6671 (1998)
65. G. Cao, S. McCall, and J.E. Crow: Phys. Rev. B 55, 672 (1997)
66. G. Cao, S. McCall, J.E. Crow, and R.P. Guertin: Phys. Rev. Lett. 78, 1751
(1997)
67. G. Cao, S.C. McCall, J.E. Crow, and R.P. Guertin: Phys. Rev. B 56, 5387
(1997)
68. S. Ikeda, Y. Maeno, and T. Fujita: Phys. Rev. B 57, 978 (1998)
69. R.J. Cava, H.W. Zandbergen, J.J. Krajewski, W.F. Peck, Jr., B. Batlogg, S.
Carter, R.M. Flerning, O. Zhou, and L.W. Rupp, Jr.: J. Solid State Chern.
116, 141 (1995)
70. G. Cao, K. Abboud, S. McCall, J.E. Crow, and R.P. Guertin: Phys. Rev. B
62, 998 (2000)
71. Z. Fang and K. Terakura: Phys. Rev. B 64, 020509 (2001)
72. L.M. Woods: Phys. Rev. B 62, 7833 (2000)
73. Y. Maeno: Physica B 281, 865 (2000)
74. Y. Maeno, T.M. Rice, and M. Sigrist: Physics Today 54,42 (2001)
75. T. Oguchi: Phys. Rev. B 51, 1385 (1995)
76. D.J. Singh: Phys. Rev. B 52, 1358 (1995)
77. I. Hase and Y. Nishihara: J. Phys. Soc. Japan, 65, 3957 (1996)
78. G.J. McMullan, M.P. Ray, and R.J. Needs: Physica B 224, 529 (1996)
79. T.M. Mishonov, LN. Genchev, R.K. Koleva, and E.S. Penev: Czech. J. Phys.
46, 953 (1996)
80. C. Noce and T. Xiang: Physica C 282, 1713 (1997)
322 D.J. Singh
81. C. Noce and M. Cuoco: Int. J. Mod. Phys. B 13, 1157 (1999)
82. C. Noce and M. Cuoco: Phys. Rev. B 59, 2659 (1999)
83. 1.1. Mazin, D.A. Papaconstantopoulos, and D.J. Singh: Phys. Rev. B 61, 5223
(2000)
84. A.P. Mackenzie, S.R. Julian, A.J. Diver, G.J. McMullan, M.P. Ray, G.G.
Lonzarich, Y. Maeno, A. Nishizaki, and T. Fujita: Phys. Rev. Lett. 76,3786
(1996)
85. Y. Yoshida, R. Settai, Y. Onuki, H. Takei, K. Betsuyaku, and H. Harima: J.
Phys. Soc. Japan 67, 1677 (1998)
86. E. Ohmichi, H. Adachi, Y. Mori, Y. Maeno, T. Ishiguro, and T. Oguchi: Phys.
Rev. B 59, 7263 (1999)
87. A.V. Puchkov, Z.X. Shen, T. Kimura, and Y. Tokura: Phys. Rev. B 58, 13322
(1998)
88. A. Damaseelli, D.H. Lu, K.M. Shen, N.P. Armitage, F. Ronning, D.L. Feng,
C. Kim, Z.X. Shen, T. Kimura, Y. Tokura, Z.Q. Mao, and Y. Maeno: Phys.
Rev. Lett. 85, 5194 (2000)
89. K.M. Shen, A. Damaseelli, D.H. Lu, N.P. Armitage, F. Ronning, D.L. Feng,
C. Kim, Z.X. Shen, D.J. Singh, 1.1. Mazin, S. Nakatsuji, Z.Q. Mao, Y. Maeno,
T. Kimura, and Y. Tokura: Phys. Rev. B 64, 180502 (2001)
90. C. Bergemann, S.R. Julian, A.P. Mackenzie, S. NishiZaki, and Y. Maeno:
Phys. Rev. Lett. 84, 2662 (2000)
91. C. Noce, G. Busiello, and M. Cuoco: Europhysics Lett. 51, 195 (2000)
92. D.K. Morr, P.F. Trautman, and M.J. Graf: Phys. Rev. Lett. 86, 5978 (2001)
93. Y. Sidis, H. Braden, P. Bourges, B. Hennion, S. NishiZaki, Y. Maeno, and Y.
Mori: Phys. Rev. Lett. 83, 3320 (1999)
94. F. Servant, S. Raymond, B. Fak, P. Lejay, and J. Flouquet: Solid State
Commun. 116,489 (2000)
95. K. Ishida, H. Mukuda, Y. Minami, Y. Kitaoka, Z.Q. Mao, H. Fukazawa, and
Y. Maeno: Phys. Rev. B 64, 100501 (2001)
96. T. Imai, A.W. Hunt, K.R. Thurber, and F.C. Chou: Phys. Rev. Lett. 81,
3006 (1998)
97. M. Braden, W. Reichardt, S. Nishizaki, Y. Mori, and Y. Maeno: Phys. Rev.
B 57, 1236 (1998)
98. O. Friedt, M. Braden, G. Andre, P. Adelmann, S. Nakatsuji, and Y. Maeno:
Phys. Rev. B 63,174432 (2001)
99. S. Nakatsuji and Y. Maeno: Phys. Rev. B 62, 6458 (2000)
100. Friedt et al. [98] report that powder and single crystal "T" phase sampIes
near x = 0.2 show different structures.
101. T. Mizokawa, L.H. Tjeng, G.A. Sawatzky, G. Ghiringhelli, O. Thernberg,
N.B. Brooks, H. Fukazawa, S. Nakatsuji, and Y. Maeno: Phys. Rev. Lett. 87,
077202
102. V.1. Anisimov, LA. Nekrasov, D.E. Kondakov, T.M. Rice, and M. Sigrist:
cond-mat/0107095 (2001)
103. H.K. Muller-Buschbaum, and J. Wilkens: Z. Anorg. Allg. Chem. 591, 161
(1990)
104. M. Itoh, M. Shikano, and T. Shimura: Phys. Rev. B 51, 16432 (1995)
105. Q. Huang, J.W. Lynn, R.W. Erwin, J. Jarupatrakorn, and R.J. Cava: Phys.
Rev. B 58, 8518 (1998)
106. S.1. Ikeda and Y. Maeno, Physica B 259-261, 947 (1999)
7 Magnetism in Ruthenates 323
Pd 303
Prl/2Cal/2Mn03 139, 140 Yb 25
Springer Series in
MATERIALS SCIENCE
Editors: R. Hull R. M. Osgood, Jr. J. Paris i