Electronic Structure and Magnetism of Complex Materials 2003

Springer Series in
MATERIALS SCIENCE 54
Springer-Verlag Berlin Heidelberg GmbH
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.
51 Point Defects 56 Si0 2 in Si Microdevices

in Semiconductors ByM.Itsumi
and Insulators
Determination of Atomic 57 Radiation Effects
and Electronic Structure in Advanced Semiconductor Materials
from Paramagnetic Hyperfine andDevices
Interactions By C. Claeys and E. Simoen
By J.-M. SpaetlI and H. Overhof 58 Functional Thin Films
52 Polymer Films and Functional Materials
witlI Embedded Metal Nanoparticles New Concepts and Technologies
By A. Heilmann Editor: D. Shi
53 Nanocrystalline Ceramics 59 Die1ectric Properties of Porous Media

Synthesis and Structure By S.O. Gladkov
By M. Winterer 60 Organic Photovoltaics
Concepts and Realization
54 Electronic Structure and Magnetism
of Complex Materials Editors: C. Brabec, V. Dyakonov, J. Parisi
Editors: D.J. Singh and and N. Sariciftci
D. A. Papaconstantopoulos
55 Quasicrystals
An Introduction to Structure,
Physical Properties and Applications
Editors: J.-B. Suck, M. Schreiber,
and P. Häussler
Series homepage - http://www.springer.de/phys/books/ssms/
Volumes 1-50 are listed at the end ofthe book.

D.J. Singh
D.A. Papaconstantopoulos
(Eds.)
Electronic Structure
and Magnetism
of Complex Materials
With 159 Figures
, 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
Professor R. M. Osgood, Jr.

Microelectronics Science Laboratory, Department of Electrical Engineering
Columbia University, Seeley W. Mudd Building, New York, NY 10027, USA
Professor Jiirgen Parisi

Oldenburg, Fachbereich Physik, Abt. Energie- und Halbleiterforschung
Universităt
Carl-von-Ossietzky-Strasse 9-11,26129 Oldenburg, Germany
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
ISBN 978-3-642-07774-6 ISBN 978-3-662-05310-2 (eBook)

DOI 10.1007/978-3-662-05310-2
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.
Typesetting by the editors

Data conversion: Le-TeX, Leipzig
Cover concept: eStudio Calamar Steinen
Cover production: design & production GmbH, Heidelberg
Printed on acid-free paper SPIN: 10844529 57/3141/ba 543210
Preface
Magnetism is today properly in the forefront of materials research. Key tech-

nologies depend on advances in magnetic materials and their manipulation.
Magnetic storage devices and media are expected to approach fundamental
materials-related limits in a few years time. Meanwhile, the development of
"spintronic" technologies, integrating magnetism with microelectronics, also
faces important challenges due to the presently incomplete understanding of
the role of atomic-scale effects in device structures, as well as the need for
better materials.
Advances in theoretical understanding and computational infrastructure
are making possible ever more reliable, material-specific modeling of mag-
netic materials and phenomena. At the same time, advances in experimental
characterization, especially at low temperatures in high fields, as well as in
the synthesis of novel magnetic phases, has revealed much rich, unanticipated
physics in the last decade. Some examples are high Curie temperature mag-
netic semiconductors, the exquisitely intricate interplay of charge, spin, and
lattice degrees of freedom in manganites, the strong renormalizations and
triplet superconductivity seen in ruthenates, and the complex phenomena
arising in non-collinear spin systems. Theory, analytic and computational, is
already playing an important role in unravelling the physics of magnetic
materials and nanostructures. Clearly, the convergences mentioned above
will lead to an increasingly important role for microscopic material-specific
theory in the development of novel magnetic materials, in the understanding
of magnetism in materials and in its exploitation in technology.
This book reviews selected new areas of magnetism from the viewpoint
of material-specific theory. It is not an attempt to cover all of the important
areas of magnetism, but rather is focused on some of the topics of most recent
current interest. The chapters are divided into two groups: (1) fundamen-
tal microseopie theory - discussing mainly magnetocrystalline and related
spin-orbit effects, non-collinear magnetism and magnetic excitations; and (2)
specific materials classes - covering magnetic semiconductors and perovskite-
derived manganites and ruthenates.
Washington, D.C. David J. Singh

December 2002 Dimitrios A. Papaconstantopoulos
Contents
1 Low-Lying Magnetic Excitations in Itinerant Systems:

SDFT Calculations
S. Halilov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Motivation: Thermodynamics of Itinerant Magnetic Systems
and Impact on Transport Phenomena ........................ 1
1.2 Magnetic Hamiltonian for Itinerant Systems. . . . . . . . . . . . . . . . . . . 2
1.3 From Postulation to Computation:
Spin-Density Matrix as Basic Variable. . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Equilibrium Configuration and Low-Lying Magnetic Excitations:
Adiabaticity Assumption and Beyond ........................ 11
1.4.1 General Formulation of the Adiabatic Spin Dynamics . . .. 11
1.4.2 Frozen Spin-Wave Method as an Approximation
to the Dynamical Transverse Spin Susceptibility ........ 14
1.4.3 Nonadiabatic Evaluation
of the Dynamical Spin Susceptibility: Practical Relevance 16
1.5 Parametrization of System-Specific Hamiltonian
Within Spin-Density Functional Approximation
and Revealation of Its Low-Energy Part. . . . . . . . . . . . . . . . . . . . .. 19
1.5.1 Iron, Cobalt, and Nickel:
Specific Choice of the Basic Variables
and the Non-Heisenberg Hamiltonian . . . . . . . . . . . . . . . . .. 20
1.5.2 Gd Through Tm:
Specific Choice of the Basic Variables
and the Hamiltonian. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 24
1.6 Spin-Wave Approach to the Low-Lying Excitations. . . . . . . . . . . .. 30
1.6.1 Magnetic Excitation Spectra of the 3d Transition Metals
and Kohn Anomalies ................................ 31
1.6.2 Various Equilibrium Phases and Magnetic Excitations
of the Heavy Rare-Earth Metals . . . . . . . . . . . . . . . . . . . . . .. 37
1. 7 Finite-Temperature Spin Dynamics:
Possible Scenario for Phase Diagram Description . . . . . . . . . . . . . .. 48
1.8 Most Valuable Problems Which Could be Addressed
Within the Adiabatic Approach ............................. 52
References .......... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 53
VIII Contents
2 Calculation of Magneto-crystalline Anisotropy

in Transition Metals
H.J.F. Jansen, G.S. Schneider, H.Y. Wang ......................... 57
2.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 57
2.2 Definition of Magneto-crystalline Anisotropy .................. 58
2.3 Examples from Model Calculations . . . . . . . . . . . . . . . . . . . . . . . . . .. 62
2.4 Magneto-crystalline Anisotropy in Density Functional Theory ... 66
2.5 Numerical Problems in the Calculations ...................... 72
2.5.1 Linear Tetrahedron Method .......................... 75
2.5.2 Gaussian Broadening Method . . . . . . . . . . . . . . . . . . . . . . . .. 79
2.5.3 Conclusions for Current Numerical Results . . . . . . . . . . . .. 83
2.6 Missing Physics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 84
2.7 The Role of k-Space.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 91
2.8 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 97
References ..................................................... 98
of Correlated Systems: Beyond LDA
A.I. Lichtenstein, V.I. Anisimov, M.l. Katsnelson ................... 101
3.1 Introduction .............................................. 101
3.2 The LDA+U Method ....................................... 103
3.3 Dynamical Mean-Field Theory .............................. 106
3.3.1 LDA+DMFT: General Considerations ................. 107
3.3.2 DMFT in Quantum Monte Carlo Approach ............. 109
3.3.3 DMFT
in Multiband Fluctuation-Exchange Approximation ..... 110
3.3.4 DMFT in Iterated Perturbation Theory ................ 111
3.3.5 DMFT in Noncrossing Approximations ................ 112
3.3.6 Cluster LDA+DMFT Scheme ......................... 113
3.4 Exchange Interactions ...................................... 114
3.5 Ferromagnetic Metals ...................................... 117
3.5.1 Localized States: Rare-Earth Metals ................... 117
3.5.2 Delocalized States: Transition Metals .................. 118
3.6 Mott-Hubbard Insulators ................................... 122
3.6.1 Electronic Structure of Transition-Metal Oxides ......... 122
3.6.2 Exchange Interactions in Transition-Metal Oxides ....... 127
3.6.3 Electronic Structure and Exchange Interactions in V 203 . 128
3.6.4 Exchange Interactions in the Ladder Vanadates CaV20 5
and MgV205 ....................................... 132
3.7 Orbital Ordering and Jahn-Teller Effect ...................... 136
3.7.1 Cooperative Jahn-Teller Distortions
in Transition-Metal Compounds: KCuF 3 ............... 136
3.7.2 Orbital Ordering in the Doped Manganite Prl_xCaxMn03138
Contents IX
3.8 Highly Correlated Metallic Compounds

and Metal-Insulator Transition .............................. 141
3.8.1 Correlation Effects in Ruthenates ..................... 141
3.8.2 Doped Mott Insulators: Lal-xSrx Ti0 3 . . . . . . . . . . . . . . . . . 147
3.8.3 Nature of Insulating State in NaV 2 0 S
Above Charge-Ordering Transition:
A Cluster DMFT Study .............................. 150
3.9 Conclusion ................................................ 155
References ..................................................... 156
4 Ferromagnetism in (III,Mn)V Semiconductors

J. König, J. Schliemann, T. Jungwirth, A.H. MacDonald ............. 163
4.1 Introduction .............................................. 163
4.2 Properties of (III,Mn)V Ferromagnets ........................ 165
4.3 Theoretical Approaches ..................................... 166
4.4 Mean-Field-Theory Predictions .............................. 170
4.4.1 Ferromagnetic Transition Temperature ................. 171
4.4.2 Magnetic Anisotropy ................................ 174
4.4.3 Anomalous Hall Effect ............................... 177
4.5 Collective Excitations Within a Continuum Picture ............ 182
4.5.1 Beyond Mean-Field Theory and RKKY Interaction ...... 182
4.5.2 Independent Spin-Wave Theory for Parabolic Bands ..... 183
4.5.3 Elementary Spin Excitations .......................... 185
4.5.4 Comparison to RKKY and to the Mean-Field Picture .... 187
4.5.5 Spin-Wave Dispersion for Realistic Bands .............. 188
4.5.6 Limits on the Curie Temperature ...................... 190
4.6 Collective Fluctuations Beyond Spin Wave Theory
and Continuum Approximation .............................. 192
4.6.1 Model Considerations ................................ 193
4.6.2 Remarks on the Monte Carlo Method .................. 194
4.6.3 Numerical Monte Carlo Results ....................... 195
4.6.4 Disorder Effects and Noncollinear Ferromagnetism ....... 200
4.7 Concluding Remarks ....................................... 206
References ..................................................... 209
5 Noncollinear Magnetism in Systems

with Relativistic Interactions
L. Sandratskii .................................................. 213
5.1 Introduction .............................................. 213
5.2 Density Functional Theory of a Noncollinear Magnet ........... 213
5.3 Relation Between Symmetry and Stability
of Magnetic Structures ..................................... 215
5.3.1 Symmetry Constraint: General Formulation ............ 215
5.3.2 Two Types of Symmetry Constraints .................. 217
X Contents
5.4 Stable Magnetic Structures .................................. 219

5.4.1 Simple Standard Cases ............................... 220
5.4.2 Magnetic Structures of U 2 Pd 2 Sn and USb .............. 220
5.5 Relativistic Instability of Collinear Ferromagnetism ............ 222
5.5.1 Magnetic Structure OfU3P 4 .......................... 223
5.5.2 Atomically Disordered Relativistic Systems ............. 224
5.6 Relativistic Instability of Collinear Antiferromagnetic Structures . 225
5.6.1 Weak Ferromagnetism in (X-Fe 2 03 ..................... 225
5.6.2 Relativistic Instability ofthe Collinear Antiferromagnetism
in UPdSn .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.7 Relativistic Instability
of a Compensated Noncollinear Magnetic Structure in Mn3Sn ... 231
5.8 Nonmagnetic Sublattices in Antiferromagnetic Systems ......... 232
5.8.1 Intersublattice Interaction in UFe4Als ................. 232
5.8.2 Magnetic Structure of UX 3 Compounds ................ 235
5.9 Helical Structures in Systems with Relativistic Interactions ...... 235
5.9.1 Magnetic Structure of UPtGe ......................... 236
5.9.2 Helices in REM ..................................... 241
5.10 Intraatomic Magnetic Noncollinearity ........................ 245
5.11 Relativistic Spectroscopy of Noncollinear Magnetic States ....... 246
5.12 Conclusion ................................................ 249
References ..................................................... 249
6 Orbital Degeneracy and Magnetism

of Perovskite Manganese Oxides
1. Solovyev and K. Terakura. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.1 Introduction .............................................. 253
6.2 Degenerate Double Exchange Model .......................... 255
6.2.1 Ferromagnetic Ordering .............................. 256
6.2.2 Antiferromagnetic Ordering .......................... 258
6.2.3 CE-type Antiferromagnetic Ordering .................. 262
6.3 First-Principles Band Structure Calculations .................. 267
6.3.1 Method of Calculations .............................. 267
6.3.2 Stability of the Ferromagnetic Ordering ................ 271
6.3.3 Jahn-Teller Distortion and A-type Antiferromagnetism
in LaMn03 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
6.3.4 Metallic Antiferromagnetism at Large x ................ 280
6.3.5 Zigzag Antiferromagnetic Ordering at x = 1/2 .......... 282
6.3.6 Optical Properties ................................... 285
6.4 Concluding Remarks ....................................... 290
References ..................................................... 291
Contents XI
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
Index ......................................................... 325

List of Contributors
Vladimir I. Anisimov J ürgen König

Intitute of Metal Physics Universitaet Karlsruhe
Russian Academy of Sciences 76128 Karlsruhe
620219 Ekaterinburg GSP-170 Germany
Russia koenig
via@ifmlrs.uran.ru @tfp.physik.uni-karlsruhe.de
Samed Halilov A.I. Lichtenstein

Center for Computational University of Nijmegen
Materials Science NL-6525 ED Nijmegen
Naval Research Laboratory The Netherlands
Washington, DC 20375, USA A.Lichtenstein@sci.kun.nl
halilov@dave.nrl.navy.mil
Allan H. MacDonald
Henri J.F. Jansen Department of Physics
Department of Physics University of Texas at Austin
Oregon State University Austin, TX 78712, USA
Corvallis, OR 97331, USA macd@physics.utexas.edu
henri@physics.orst.edu
Leonid Sandratskii
Tomas Jungwirth Max-Planck-Institut
Department of Physics für Mikrostrukturphysik
University of Texas at Austin 06120 Halle
Austin, TX 78712, USA Germany
jungw@physics.utexas.edu lsandr@mpi-halle.de
Mikhail I. Katsnelson John Schliemann

Intitute of Metal Physics Department of Physics
Russian Academy of Sciences University of Texas at Austin
620219 Ekaterinburg GSP-170 Austin, TX 78712, USA
Russia joschlie@physics.utexas.edu
mikhail.katsnelson@usu.ru
XIV List of Contributors
Guenter S. Schneider Kiyoyuki Terakura

Department of Physics JRCAT -National Institute
Brookhaven National Laboratory of Advanced Industrial Science
Upton, NY 11973, USA and Technology, AIST
guenter@bnl.gov Central 4, 1-1-1 Higashi
Tsukuba 305-8562, Japan
David J. Singh
terakura@jrcat.or.jp
Center for Computational
Materials Science
Naval Research Laboratory Haiyan Y. Wang
Washington, DC 20375, USA Department of Physics
singh@dave.nrl.navy.mil Oregon State University
Corvallis, OR 97331, USA
Igor Solovyev henri@physics.orst.edu
JRCAT-National Institute
of Advanced Industrial Science
and Technology, AIST
Central 4, 1-1-1 Higashi
Tsukuba 305-8562, Japan
igor.solovyev@aist.go.jp
1 Low-Lying Magnetic Excitations
in Itinerant Systems:
SDFT Calculations
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 Magnetic Hamiltünian für Itinerant Systems

As general practice shows, the detection and interpretation of the excited
states is of minor interest unless it is compared with a theoretical model,
which will be more reliable the fewer free parameters it contains.
Frequently the comparing is done within a semi-phenomenological ap-
proach based on the experimental data to establish the effective forces and
thus adjusting the parameters of the appropriate Hamiltonian. Adjustable
parameter free calculations of the effective forces are so far very rare but,
despite some restrictions imposed by state-of-the-art DFT, are potentially
weH suited to gain more insight into the microscopic origin of the forces, which
are in general produced by the relativistic interactions and the electrostatic
interaction between magnetic electrons and the rest of the crystal.
Any system is best described on the microscopicallevel in terms of charge
n( r) and current j (r) densities at any point in the crystal. Being averaged
over time, they describe possible states of thermodynamic equilibrium, and
those that change under the time reversal are objects of interest in the present
chapter. The non-vanishing currents j (r) produce a non-zero moment density
r x j at equilibrium, either of spin or orbital nature. If the macroscopic
moment, i.e. moment density integrated over a macroscopic domain, does
not vanish, one deals with a ferromagnet, or otherwise with an antiferro-
magnet. On the other hand, the current is not explicitly present in DFT,
which is designed for the determination of the ground state. In SDFT, only
the moment associated with the spin current is present, which couples to
the external field in the form of the Zeeman interaction. The effects of the
orbital currents are usually considered perturbatively, in terms of the minimal
single-site spin - own - orbit coupling. Prom that point of view, a clear
concept is needed to establish the sources for different type of interactions:
throughout the chapter it will be assumed that the exchange interaction is not
affected by the relativistic effects and is associated solely with the spin density
(although we understand that spin itself may systematicaHy be introduced
only within Dirac's theory). This assumption is an essential facility, since
there is no known effective current-density functional theory that would be
suitable for practical purposes. Moreover, when one deals with the energy
scale of several meV for low-Iying excitations as in the case of most of the
REMs, a detailed knowledge of the hierarchy of the magnetic forces becomes
mandatory. This plays a decisive role by establishing a feasible way to map
the system Hamiltonian onto an effective moment Hamiltonian.
The Hamiltonian for a system of electrons and nuclei, including relativistic
effects in the lowest order and coupling to an external electromagnetic field
in a semi-classical way, is written in second-quantized form as
H = Ho + Hint + Hsr + Hso + Hext + ... , (1.1)

1 Low-Lying Magnetic Excitations 3
where the contributions are as follows:
(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)
X (V' - 2V")]-J;U2(r')-J; U3(r) , (1.5)

Le. the electron spin - electron own orbit and the electron spin - other
electrons orbit relativistic interaction; and the interaction of an external field
A with the orbital and spin currents reads
Hext == -iltB J L
U1
dr-J;t1 (r)[A+(r, t)V - VA(r, t)]-J;U1 (r)
+ItB~ L J dr-J;t1 (r)A+(r, t)A(r, t)-J;U1 (r)

U1
+ItB LJ dr-J;t1 (r)aU1U2-J;U2(r)V x A(r, t). (1.6)
U1U2
Other notations are ItB == 2'!c'
Vle(r) == - '"'
~ ZR
Ir-RI' vee(lr - r'l) == Ir-r'
1 I' (1.7)
R
4 S. Halilov
stand for the Coulomb electron-nuclei and electron-electron interactions, re-

spectively, and particularly
V'2 vee (1 r - r ' I) = -I r -2r 13 .

I (1.8)
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)
where the creation-destruction operators ct,

~ are assumed to satisfy the
usual fermion equal-time anticommutation relations in space of single-particle
quantum numbers i
"+")
[Ci ,Cj + -
- 'Uij , (1.10)
Aiming at the derivation of the lower part of the Hamiltonian spectrum,
we assume that the fermionic variables ~u (t, r) may effectively be replaced by
the appropriately chosen momentum variables m(t,r). The latter will not be
postulated in the form of an integer or half-integer angular moment um, as is
usually the case for the magnetic dielectrics by transforming the Hamiltonian
(1.6) into a Heisenberg-like form. Instead, the operator of momentum density
"+ (t,r)m7f;u(t,r)
m(t,r) = 7f;u " (1.11)
with the fast electronic motion averaged out, will serve as a basic operator
in construction of the effective Hamiltonian. The expectation value of the
operator at equilibrium will be derived within SDFT, via a semi-classical
treatment. Moreover, further reduction of the degrees of freedom becomes
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
m(r) = ~ ~)mn8(r - r') + 8(r - r')mn] (1.12)

n
is contributed by individual total angular momenta m n = Sn + in built from

spin Sn and orbital in momenta, the sum runs over aH electrons n in the
system. Suppose that the momentum operators defined as
mi = r drm(r)
. lVi
(1.13)
reflect the domain structure of the effective exchange-correlation hole, and

have been verified for a given system as being adiabatic (whieh is a matter
for SDFT by present treatment), with Vi enclosing the hole domain. They
evidently obey the usual angular moment um commutation relations
(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])
H = Ho + Hexe + Hmm + Hdem + ...
(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
Hexe = -~ L Jii/SiSi/ - 4\ L Jijkz[Si X Sj][Sk X szl + ... , (1.16)

2 ii/ . ijkl
where J are the exchange parameters shown up to the fourth order, which
appears to be isotropie in our model. It is known [47], that by account of
the second-order isotropie exchange the minimum energy in systems with
identical atoms in the unit cell is attained by a simple spiral. Being comple-
mented by fourth-order terms, isotropie exchange may already stabilize the
ferromagnetie (cone) spiral. Numerieally, the fourth-order isotropie exchange
in the 3d transition and the REMs have been shown to be essentially minor
compared to that of the second order.
Even by a simple atomie-like treatment of the SOl, which became a stan-
dard way by numerieal implementations, a combination of the former with
the isotropie exchange leads to the appearance of the so-called anisotropie
exchange interaction, which, in the lowest order, couples the spin moments
as in [48,49]
iI;xc+SOI == 2: Dii,Si X Si' , (1.17)

ii'
where D ii , is a function of the lattiee sites ii'. This term is known to be

responsible for weak ferromagnetism and partieularly for its contribution to
non-collinear antiferromagnetie metastable states of the actinide systems [50],
such as divergence in orient at ions of the spin and orbital moments. In the
systems of interest, in view of the dominant effect of the direct isotropie
exchange interaction on equilibrium configuration and moment dynamies,
there was no need to consider the anisotropie exchange as an additional force
in the equation of moment motion.
1.3 From Postulat ion to Computation:

Spin-Density Matrix as Basic Variable
Early quantum theories of the low-Iying spin excitations were developed
essentially using the following assumptions: many-body Hamiltonian of elec-
trons plus lattice is replaced by an effective spin Hamiltonian, Le. taken in
various forms of the generalized Heisenberg model. That is, the magnetic
properties of the translationally ordered systems in these theories are defined
by completely localized spin moments associated with magnetie sites, and
the itinerant electrons mediating the inter-site exchange. In fact, the most
severe approximations to validate the use of the spin Hamiltonians are 1)
the interaction between the electrons in the partly filled inner shells and the
electrons in the valence shells is weak, so these two types of electrons can
be regarded approximately as two independent subsystems; 2) the magnetic
properties of the system are primarily due to the electrons in the partly filled
inner (d and f) shells. In practice, introduction of spin Hamiltonian requires
a postulat ion of the very spin moments, often fictitious, specifically in the case
of 3d magnetic metals, where the moments per site are known to be fractional
numbers. Then, the lower part of the Hamiltonian is sought with the aid of
the Holstein-Primakoff transformation to the second quantization equation
of motion and subsequent linearization. The shortcomings of the method are
weIl-known: 1) spin moments and the commutation rules for them have to be
postulated; 2) the ground state is therefore to be postulated as weIl; 3) the
equation of spin motion is valid only in the long wave-Iength limit.
When aiming at determination of the ground state within spin density
functional theory, whieh with the spin density as its basic variable is formally
exact, the relevant angular momentum operators and their permutation prop-
erties will be properly introduced, instead of being postulated.
The following equations
~
s(r, _1" ~t
t) - 2" ~ 'l/Js(r, ~
t)O"ss''l/Js,(r, t), (1.18a)
ss'
8 S. Halilov
i(r, t) = ~ 2: -J;!(r, t)i-J;s,(r, t), (1.18b)

ss'
are considered as formal definitions for the space- and time-dependent vector
spin- and orbital-density operators. The spin-density operators commute with
space-only dependent operators and are therefore well-defined, which makes
their use very suitable to describe a general situation with arbitrary equilib-
rium magnetic configuration, where spin polarization may be contributed to
by any electrons irrespective of the spatial extension of their single-particle
wave functions. Of course, that would be true if there were no effects of the
orbital polarization. Here, -J;s is the s-component of the spinor-field operator
in the Heisenberg picture, and a = (a X , a Y , a Z ) are the Pauli matrices. The
expectation value of the vector spin density is
s(r,t) = (s(r,t)) = ~tr [n(r,t)a] , (1.19)
where n(r, t) is the spin density matrix
(1.20)
and tr means the trace with respect to spin indices.

In an isolated atom, the angular moment um operator commutes with
the Hamiltonian and the orthonormal eigenstates are naturally numbered by
the angular momentum projection numbers, so the field operator -J;(r,s) =
Ll-J;l(r, s) constituents labeled by different angular numbers anticommute
with each other due to orthonormality of the orbitals. In solids, this may not
be true because of the hybridization effects, however, the anticommutation
relations might with a sufficient accuracy still be relevant, whenever the situ-
ation in asolid becomes similar to that in isolated atoms. This is, for example,
the case of the heavy rare-earth elements, where the 4f-projected wave func-
tion and that of the outer-shell electrons essentially maintain orthonormality
in the crystalline phase. We will see that, although with a limited accuracy,
the 3d-projected wave function can be considered as being orthogonal to the
highly itinerant sp-projected part in the 3d transition elements.
Each part of the field operator, which might comprise a group of angu-
lar numbers, is assumed to be spanned upon the mutually anticommuting
fermionic creation-distruction operators ak
-J;i(r, s) = l: cPk(r, s)ak, (1.21)

k
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
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
Si(t) = (Si(t) = ~ [ . tr[O'n(r, t)]d3 r. (1.23)

•
with the spin density matrix n(r, t) which in SDFT is given by the Kohn-
Sham orbitals cPkn of wave vector k and spin subband n:
(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)
where, in the presence of an external magnetic field h( r),
v(r) = ivions(r) + iJdr,~tr[11(r?? + h(r) .8-+ i~ " 8~xc ,(1.26a)

r-r 2 ~ un°
i=1,2 •
Llv(r) = ~2 (8E xc _ 8Exc ) .

8nl 8n 2
(
1.26b
)
The unitary spin marix U(r) is to be taken in such a way as to diagonalize

the density matrix n ss , (r) locally, Le. at a point r
[U11U+] ij = 8ij n i , (1.27a)
U(r) = e!e(r).'" , (1.27b)

with i = 1,2 labeling the eigenvalues of the 11 and e( r) standing for the local
moment quantization axis (LQA). This procedure ofthe local diagonalization
10 S. Halilov
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
Exc[nJ = J drn(r)E xc (nl(r),n2(r)) = J

drn(r)E xc (n(r),m(r)) , (1.28)
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
E~~(")l[nJ = J drn(r)E xc (n(r), e(r) ·1n(r)) . (1.29)
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.
1.4 Equilibrium Configuration

and Low-Lying Magnetic Excitations:
Adiabaticity Assumption and Beyond
Before getting into details, we will tacitly assume that the redueed basic
variables and forees between them are established. Onee the magnetic eonfig-
uration is settled at a metastable equilibrium, the derivation of the low-
lying excitations beeomes a problem of finding the eertain perturbations
of the eonfiguration, driven by the forees. Generally, the sort of excitation
depends on the external fields eausing the deviation from the equilibrium,
which is a matter of the theory of response functions. The most complete
description of the low-Iying magnetic excitations is given by the formalism of
the dynamical spin susceptibility X(w, r, r ' ), which describes the change of
a spin system in the presence of a magnetic field. The normal modes of the
system are therefore also imprinted in X(w, r, r ' ), since the spin susceptibility
gives, in fact, the time and space correlations between the spin variables of
the system itself. The evaluation of X(w, r, r') performed self-consistently
within a perturbation approach under an assumption of the small external
magnetie field, encompasses the issue of the present chapter, and the further
consideration eontemplates doing a feasible forecast for the spectra of the low-
lying excitations, distribution of the lifetime between the different vibration
modes and their behavior at finite temperatures.
1.4.1 General Formulation of the Adiabatic Spin Dynamics

In practice, even the lowest system state that results from an interplay be-
tween exchange and relativistic interaetions, ean hardly be described within
SDFT, since its only basic variable is the spin density matrix and therefore
is eonfined to the non-relativistic limit. This means, even a semi-classical
vision of the low-Iying excitation problem, where a quasi-classical equilibrium
part of the basic operator variables is explicitly extracted and is presently
interpreted within SDFT plus the eomplementary terms, is still too compli-
cated in practice and needs to be simplified at the expense of accuracy. As
in the theory of lattice vibrations, where the light electronic and heavy ionic
variables are weH separated, it is reasonable to build a theory of the magnetic
excitations by reduction of the electronic degrees of freedom to the fast and
slow variables. Such differentiation of the processes on the timescale is not
necessarely easy going for the magnetic variables mi, since the latter may not
generally be split off from the electronic states. We address our consideration
to the systems where the time evolution of the electronic wave function can
be factorized as a product of two parts,
(1.30)
one being fast in time and the second parametrically depending on the rel-
atively slow motion of the magnetic variables. In fact, this is the general
12 S. Halilov
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 ,
where astands for the transverse components of m.

There is a seemingly good prerequisite for establishing the adiabatically
relevant moments in magnetic metals. The timescale of the magnetic moment
dynamics in asolid is much larger than the timescale of electrons orbiting
around an atom and even larger than a/VF, the lattice spacing over the
Fermi velocity, which is the timescale of motion of aBloch electron through
the crystal. Typical magnon energies for a transition metal are a few tenths
of an eV while the d-band width is a few eV, in REMs, with the RKKY
mechanism for the exchange, the magnon energies are measured in units
of meV, thus being much smaller. Nevertheless, a systematic adiabatic ap-
proach has not been developed for this situation, simply because there is
no large mass governing this timescale. However, on a heuristic level, the
above-described frozen magnon approach can be given a physical meaning.
Suppose that an additional external potential v S8 ,(r) is applied that enforces
the ground state spin density matrix to attain the values corresponding to
the constrained-moment-directions result. Then, the corresponding energy is
the internal energy of that ground state, that is, the energy not counting the
interaction energy with that additional external potential v88' (r).
If we now allow this external potential to vary slowly in time, we may
have an adiabatic situation
(1.32)
where on the left the average is taken with the true nonstationary state, while
on the right the average is with ground states of the fast electronic motion,
parametrically depending on time.
An itinerant system is never in an eigenstate of m~, that is, generally
(mn =f (mi)2 . (1.33)
However, our central assumption will be
(mi· mj)fast ~ (mi)fast . (mj)fast = mi·mj for i =f j, (1.34)
where the average is over parametrically time-dependent ground states of the
fast electronic motion: we assume the relevant electronic correlation hole to
be essentially in the inner part 01 the atomic volume.
To decouple spin and charge excitations of a many-electron system, the

adiabatic principle can also be evolved within the Born-Oppenheimer method,
as has been recently done in [67-69]. The original idea of the relatively slow
nuclei motion has been replaced by slow motion of appropriately chosen
magnetic variables mi' In fact, it is easy to show that once the variables
are introduced, then the equation of motion for the lower part of the system
Hamiltonian iI appears to be identical with the frozen magnon equations,
the derivation and application to real systems of which is the main issue of
the chapter. Briefly, the linearized equations of spin motion, in accordance
with the Born-Oppenheimer method,
(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
E[mi] = min7ni = Iv;, dr(,pla-I,p), ('ljJ I iI 1'ljJ) , (1.37)
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
{)'ljJ '" ~ {)'ljJ '" ~'ljJo (1.41)

{)m~ m {)() m '
and therefore the Berry curvature becomes
[}X y
R
~
'"
( {){)'ljJ
mo
I~)
{) '"
~8
2 O,R,
mR
'" m (1.42)
which gives for the excitation energies

jxx _
q JXX
Wq = [}X y =m q , (1.43)
q
i.e. the same dispersion which comes out on the basis of the Heisenberg
equation of motion.
1.4.2 Frozen Spin-Wave Method as an Approximation

to the Dynamical Transverse Spin Susceptibility
The following picture is evocative of the semiclassical method, in accordance

with which the adiabatic operator of the magnetic variable mi, introduced
earlier, splits into a classical part, determined by the equilibrium configura-
tion eim?, and an operator of the quantum vibrations 8mi
(1.44)
By further consideration, the 8 will be omitted. It is assumed that the system
energy becomes minimum at a configuration given by eim? In practice, since
the local spin-density approximation to SDFT is used, the set eim? is sought
as a minimum of the exchange-correlation energy
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
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
where the following notations are used
i == R+T" == R", (1.47)

R is a lattice period andT" is a basis vector of the unit cello
At small deviation ()" « 1 for all v about a uniform ferrorp.agnetic
constraint state, the exchange energy of the configuration can always be
expanded in a Taylor series
JE = - 12:--
2
()"" -, + ...
()*,J""
q ,
(1.48)
",,'
where
(1.49)
Then the following equation may serve as adefinition and a method of

derivation of the exchange constants at the same time
_, 82 E 82E
J""
q
- - -
8(),,8()*, (Jv=(Jv'':o
+ i 8(),,8()*,
- - (J (J
v= v~=o
(1.50)
" 4>1'-4>1',-0 " 4>,,-4>1',-1</2
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
and obey the following trivial symmetry relations

jIJ-
q
v = j"l-'
-q'
Rejl-'''
q
= Rejl-'v
-q'
Imjl-'''
q
= -Imjl-'''
-q , (1.52)
since J ij are real, symmetrie, J ij = Jji, and lattice periodic. Thus, the Fourier
transforms of the exchange constants in the vicinity of a uniform moment
configuration are directly obtained from total energy calculations for a few
spin-spirals (1.48) at that wave vector q with the constraint on the LQA. As
was already mentioned, for a general case of a nonuniform configuration the
constants can be obtained by applying a supereell calculation for the energy
change. If, however, the system-specific Hamiltonian may be approximated
by a Heisenberg-like expression, which itself is a matter subject to a test to be
done within SDFT in the whole phase space, then the just-derived constants
seem to still have a reasonable accuracy.
1.4.3 Nonadiabatic Evaluation

of the Dynamical Spin Susceptibility:
Practical Relevance
In attempting to obtain a relevant interpretation of the processes in real
systems, one frequently faces a well-known situation: the more evolved the
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)
For stationary and translationally invariant systems, the Fourier transform

x( q, w) suffices to describe even those excitations whose energies are not
necessarily confined to the vicinity of the Fermi level, which is beyond the
Landau's theory of the Fermi liquid.
All apparently parameter-free models developed in the recent past can be
classified into several groups: one is using the random-phase-approximation
(RPA) [24,26,25] in the reduction of the many-body Hamiltonian; another
isbased either on time-dependent modification of SDFT [58], the result
üf which für Ni as an example displayed on Fig.1.1, or a combination of
SDFT with the RPA for decoupling in the Green function technique [59],
all differing primarily in approximation made for the dynamical exchange-
correlation kernel.
Despite the continuing progress of the various techniques used so far for
the nonadiabatic evaluation of the spin susceptibility, the basic features of
the excitation spectra associated with the poles of the transverse X+- (q, w),
have undergone only some quantitative modifications since they were first
reported in [24,26]. Those specific features are: prediction of a gap or multiple
gaps in the magnon spectrum of iron and the double-branch spectrum of
nickel. We have to emphasize that the interpretation of the complex excitation
dispersions to best of our knowledge was rather disguised by the numerical
complexity and generally was addressed to the electron band structure of the
systems in quest ion. Note that no multiple-branch spectrum for iron could
be experimentally detected [37], and there is only limited evidence for the
existence of an optical branch in nickel [35]. In contrast to the nonadiabatic
methods [24,58,59], the frozen-magnon interpretation [56,57] suggests a rather
18 S. Halilov
500
>
~ 400
g 300
,c: 200
100
o
0.2 0.4 0.6 0.8
Fig.1.1. Transverse dynamical spin suseeptibility ImX+- (q, w) of Ni along r -

X direetion, obtained by self-eonsistent treatment of the spin density as a linear
response to a transverse external field h(q,w) [58] (with permission). The light
spheres reflect the measured data [35] (see Seet. 1.5.1 for a simple interpretation of
the optical-type spin vibrations)
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.
1.5 Parametrization of System-Specific Hamiltonian

Within Spin-Density Functional Approximation
and Revealation of Its Low-Energy Part
There is an essential difference in the interaction hierarchy of the magnetic
electrons in the case of the iron group ions and those of the lanthanide group,
which is displayed in Fig.1.2. The 3d magnetic electrons are outermost and
hence are strongly affected by the electrostatic potential of all the charges
external to the ion, whereas the magnetic 4f electrons are shielded by the 5s5p
shells and are less affected. That is, the crystalline field effects in the case of
iron have an order of 2 eV and are on the same scale as the exchange splitting,
thus effectively contributing to a non-Heisenberg behavior (particularly in the
case of Co and Ni). In the case of the lanthanide group, the huge intrasite
exchange of the shielded 4f electrons of the order of 5 eV lifts up the degener-
acy in spin, and assisted by the SOl lifts up the degeneracy in the azimuthaI
numbers, thus producing a considerable orbital moment. At the same time,
the 4f-electron exchange field causes the outershell conduction electrons to
become spin-polarized. Therefore, the low-energy moment dynamics of the
REMs may be viewed as a kind of fusion between the localized 4f-projected
spin and orbital moments and the nickel-type itinerant magnetic electrons.
a <r
Fe
C sial
fie%
t2g
El<Change
-1.5eV
·1/2
Spin-orbit
~J)
..
~
3d
a~; <r ,~: ~::

(Iml=2.0) +112 ~1
-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
1.5.1 Iron, Cobalt, and Nickel: Specific Choice of the Basic

Variables and the Non-Heisenberg Hamiltonian
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]
----- X10~ -----x10),

,,
20 I' 20
Fe - - - dl Ni --- d':
I
I
1
15
I
,
I ,:
:
/
I 10 / '
I I :
I I '
I 5 I :
I
I ,
,
I
O-ft'l"';'-'-+":;'--+----+-f O+"-':"-~":"-_+-~-f
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
[SA.1,0, SA 3'ß] ~';'

_ ~ . .caß"I s 'vy' ,
bU t )\:.
A
• •
(1.56a)
[S,a, SA.] A. .
~ .' .. aß"Is
A.
Jß _ ~ tU'J E '''I' , (1.56b)
(1.56c)
22 S. Halilov
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
where the direct exchange between d-character spin momenta is comparable

to that between interstitial moments and may not be omitted, thus opposing
the case of the REMs, where in fact the RKKY model for the exchange is
relevant. The terms different from the exchange, which are primarily associ-
ated with the spin-orbit interaction and depend on the local symmetry, are
labeled by Hanis . The latter may comprise the on-site spin-orbit coupling
term, the relativistic coupling of the spin moments located on different sites
(Dzyaloshinski-Moriya term). By numerical evaluation in cubic systems, the
term will be omitted as being immaterial compared to the exchange.
Obviously, the Hamiltonian (1.57) for the magnetic transition met als has
properties different from that for the conventional Heisenberg systems, inspite
of the formally same appearance: there is no constraint on the maximum
itinerant moment projection. At this point, an important SDFT test has to
be done to demonstrate the extent of the phase space where the bilineady
approximated Hamiltonian, with the constraint on the maximum moment
projection, can still be used for the determination of the moment dynamics
with a sufficient accuracy. This test greatly simplifies the consideration of the
dynamics, allowing an effective separation of the transverse and longitudinal
vibrations.
Independent SDFT calculations of m(O, q), E(O, q) for various constrained
spin-spiral configurations have been performed to test the validity of the
1 Low-Lying Magnetie Excitations 23
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
!'ö. 0.2 fee

0.2
'"
>: 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
Fig.1.5. Site magnetie spin moment M = 2J-lBm(B, q) and total energy

6.E(B, q) = E(B, q) - E(O, 0) as functions of B for spin-spirals with q = (0,0,1/4),
(0, 0,1/ 2), (0,0,1) in units of 1r/a. Symbols mark ealculated values and dashed lines
are guides for the eye. The solid lines eorrespond to funetions j M 2 (B) sin 2 B for
fixed q . (a) bee-Fe, (b) fee-Co , (c) fee-Ni
24 S. Halilov
The comparison of the calculated values of L1E(O) of Fig.1.5 with the

m 2 (O) sin 2 0 curves gives strong support to the use of an energy expression
bilinear in moment vectors (adiabatic average of the Hamiltonian (1.57)) and
hence to the relevance of the exchange constant matrix (1.51), as long as 0 and
q are not simultaneously large. On the other hand, it also indicates that the
exchange constants J q should not be determined from 0 = 90° spin-spirals.
The analysis of the individual contributions shows that the total energy
follows essentially the exchange and correlation energy with a slight reduction
by both kinetic and Coulomb energy for larger 0 values, due to a slight
expansion of the d-shell when the moment is reduced. It shows again that
this latter intraatomic energy balance contributes little to the interatomic
magnetic interaction as long as q and 0 are not both large. This gives furt her
support to our central assumption (1.34).
1.5.2 Gd Through Tm:

Specific Choice of the Basic Variables and the Hamiltonian
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
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
Li(t) = Iv. L-0}.(r,s,t)L -0/i(r,s',t)d r,

~ ss'
i 3 (1.60)
26 S. Halilov
where fi refers to the 4f orbitals at lattice site i, and Li is the Schrödinger

operator of total orbital momentum with respect to the center of site i. The
quantum expectation of these vectors is
Si(t) = (Si(t)) , Si(t) = (Si(t)) , Li(t) = (Li(t)) , (1.61 )
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
[hJ"SjßL = i6ijfaß,Si-y, ; (1.62a)

[Sia, Sjß L = i6ijfaß, Si-y, ; (1.62b)
[Lia,LjßL = i6ijfaß'Li-y, ; (1.62c)
(1.62d)
The magnetic interactions in RE systems have been thoroughly discussed
in the literat ure (see, for instance, [62,63] and references therein). The essen-
tial interactions are:
• intersite exchange coupling between the conduction electron spin mo-

ments Si and Sj,
• intrasite Hund's rule exchange coupling between Si and Si,
• intrasite third Hund's rule coupling between Si and Li due to spin-orbit
interaction, and
• crystal-field coupling of Li to site-symmetric coordinates of the lattice
site.
To take account of these interactions, we introduce the Hamiltonian
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
______ 5 i J ißJ
/ _ Inter-site
SI ~_ excl1<11ge
I nt ra-site "!'l
excnange ,..--
I. ~ri ( ~
"=2,4,6
' )" axial anisotropy
~'YI[ ( ~ }6 + c.c,] basal 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
The onsite spin-orbit coupling constants Ai may safely be obtained from

atomic SDFT calculations. The single-electron spin-orbit parameter A which
determines the spin-orbit splitting .t1Eso = (l+ 1/2)A of single-electron states
with orbital moment um quantum number l = 3, can easily be evaluated from
the radial part cp4f(r) of the Kohn-Sham orbital and the spherical average
v(r) of the atomic Kohn-Sham potential:
A = 4ne1i 2
2m2c2
J d ",2 ( ) dv(r)
r r'i'4f r dr . (1.64)
The spin-orbit parameter A for the many-electron Russell-Saunders state,

which enters the Hamiltonian (1.63) at sites i, is then
A---~ (1.65)
•- 28i '
where Ai is (1.64) for the atom at site i.

The intraatomic exchange parameter in SDFT to first order in the spin-
dependent exchange-correlation potential, [30]
(1.66)
is obtained from the energy
.t1E = 2Is8 = J d3r.t1n4f.t1vxc = 28 Jdr4nr2cp~f.t1vxc (1.67)
needed to reverse the conduction electron polarization. Here, L1n4f is the

difference of spin-up and spin-down 4f-densities, and .t1vxc is the change of
V xc due to reversion of the conduction electron spins. Hence,
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
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
H = Iis i . Si - AiSi . Li - Li . ~i . Li (1.73)

ih
to describe the magnetic excitation spectra.
30 S. Halilov
1.6 Spin-Wave Approach to the Low-Lying Excitations
For some general situation, it is more convenient to treat all the variables as
components of a single vector operator
(1. 74)
With the Hamiltonian expressed in terms of mit,
(1. 75)
the Heisenberg equations of motion
(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
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)
are collected into a Tm Ne-dimensional complex column vector m according to
(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
nwqm(q) = D(q)m(q) , (1.82)

the low frequency mo des of which yield the adiabatic magnons. The system-
specific expressions for the dynamical matrix D (q) are rather long and not
displayed here. From now on, the matrix D(q), explicitly obtained on the
basis of the LSDA-SDFT, is the only quautity tacitly used to evaluate the
moment dynamies.
1.6.1 Magnetic Excitation Spectra of the 3d Transition Metals

and Kohn Anomalies
Throughout this section, we will consider only a single moment Si assigned
to each cell, unless the effect of the extended set is discussed explicitly.
The computed magnon spectra of bcc-F, fee-Co, and fee-Ni are shown in
Figs.1.8-1.10. For a Bravais lattice there is only one (transversal) brauch,
whose dispersion is simply
(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
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
J ij = - (~;)3 lz d3qe-iq.(Ri-Rj)j:iVj, i =I- j, (1.84)
on the site distance R ij for Fe is displayed, demonstrating that there is

indeed an oscillating long-range exchange interaction reaching beyond the
15th neighbor in the crystal. The constrained spin-spiral approach seems,
at present, to be the only technique to trace quantitatively such long-range
exchange interactions.
In order to estimate the minimum size of the real-space domain, which
would still be capable to reproduce the nesting-induced excitation spectra
features, the magnon spectrum of Fe is shown on Fig. 1.13 as a function of
the cluster size. The latter is simulated as a cutting parameter in the Fourier
transform procedure by the real-space exchange constants. Obviously, reason-
able cluster-type calculations have to comprise at least one thousand atoms,
which makes the reciprocal-space consideration of the long-range quantum
correlations more reliable.
Now we briefly discuss whether the longitudinal degrees of freedom, which
are normally interpreted in terms of the Stoner spin-flip single-particle tran-
sitions, can effectively be incorporated by extending the set of basic moment
• bcc Fe
• --->X 10
•
• •
...
• • •
•••••
•
· '"
-I --->X 10
•
2 3 4
RJalattice
Fig.1.12. Exchange constants Jij of Fe as a function of neighbor distance Rij.

Note the oscillating behavior up to at least 15 neighbors
500 R=1a
. ..... 500 65R=2aatoms
15 atoms
... ...' '

. ..........
.
...·· .....
250 250
.' ..'·
...
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
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.
1.6.2 Various Equilibrium Phases and Magnetic Excitations

of the Heavy Rare-Earth Metals
The SDFT-based specification of the various parameters of the effective
Hamiltonian (1.63) includes several steps. In a first stage, spin-orbit coupling
is neglected (A=O), and the intersite exchange constants of the conduction
spins are obtained from the total energy calculations for conical spin struc-
tures. Thereby the conduction electron polarization at each site is driven by
the f-spin. Then, the orbital interactions A and ~ are added to the equation
of motion.
The LSDA-SDFT and the so-called pseudo-core treatment of the 4f shells
have successfully been applied to the ground state properties including mag-
netic properties of REM and compounds in a large body of literature. Com-
pare for instance [73] for an early systematic DFT treatment of the lan-
thanides in the paramagnetic state, and [74J for early treatments of polar-
ized 4f sheIls, as weIl as [75J for arecent review. In this approach, the 4f
Kohn-Sham orbitals are excluded from the Kohn-Sham occupation rule, and
min(nf, 7) 4f spin-up orbitals and max(O, nf - 7) 4f spin-down orbitals are
occupied in each iteration cycle, where up and down refers to the local 4f
spin polarization direction. Therefore, we adopt the above described pseudo-
core approach and compute the state of the f-shell in each iteration step as
a single Slater determinant of Kohn-Sham (l, m, s) orbitals. These f-orbitals
are obtained from the solutions of the radial Schrödinger equation in the site
potential weIl with the boundary condition that the logarithmic derivative of
38 S. Halilov
the radial wave function at the Wigner-Seitz radius is equal to -l - 1 = -4.

The spin polarization is taken to be collinear within each atomic sphere and
pointing in a predefined direction. The calculation is first done within the
scalar relativistic approach, that is, spin-orbit interaction is neglected at this
stage while the spin-independent relativistic corrections are retained in the
radial equations. This implies slIS, and there is a unique spin-polarization
direction within each atomic sphere for both fand conduction electrons.
Table 1.1 shows the computed occupation numbers of the 6s, 6p, and 5d
Kohn-Sham orbitals and the conduction electron site spin moment s of the
constrained collinear ferromagnetic ground state of the REs Gd through Tm.
Table 1.1 shows the equilibrium lattice parameter a of the hexagonal basal
plane and the ela ratio used in the computations (and taken from experiment
since LSDA yields values of lattice parameters that are too small by 1~2%).
Also given are the calculated exchange splittings in energy of the conduction
states, averaged over all occupied Bloch states, and the total site spin moment
s + S. With the value of the 4f orbital moment of the Russell-Saunders state,
the total magnetic site moment in units of the Bohr magneton is estimated
as Mtt;;teor. = 2s + 2S + L, and in the last two columns of Table 1.1 this value
is compared to the experimental magnetic moment [63]. (The conduction
electron orbital moment is quenched by a large bandwidth and is expected to
be negligibly small.) The lattice parameters reflect the well known lanthanide
contraction from left to right through the series. The spin splitting is roughly
proportional to the 4f spin moment and hence decreases from Gd through
Tm. The conduction bands and their spin-projected densities of states are
shown in Fig. 1.15.
Next, instead of assuming a collinearly polarized ground state, the direc-
tion of the conduction electron site spin moment is constrained to be a cone
Si = si(sinOcos(q· R i ), sin 0 sin(q . R i ), cosO). Since there is no direct inter-
action between the f spins and spin-orbit interaction is neglected so far, the
site f spins point in the same directions as the conduction electron site spins
in the constrained ground state. The problem of noncollinearity of nss/(r)
Table 1.1. Lattice parameters ahcp' cl a ratios, average exchange splitting L1 xc of

the conduction states, total spin moments 8+8, theoretical total magnetic moments
M:;;teor . = 28 + 28 + L, and experimental values M:~per. [63]
ahcp' A cla L1 xc , eV s+8 M:;;teor ., MB

Mexper.
tot , MB
Gd 3.634 1.591 0.90 3.86 7.72 7.63
Tb 3.606 1.580 0.80 3.32 9.65 9.34
Dy 3.592 1.573 0.67 2.78 10.57 10.33
Ho 3.578 1.570 0.54 2.24 10.47 10.34
Er 3.559 1.569 0.36 1.68 9.35 9.1
Tm 3.538 1.569 0.20 1.10 7.20 7.14
Tb
2 2
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
0.4 0.3 , - - - -- ----, 0.2 . , - - -- ---,
....::: 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
q = 0 for Gd through Dy, while for Ho through Tm the maximum at nonzero

Q indicates instability of the collinear ferromagnetic state against a conically
polarized (or more complex) ground state, with decreasing period (increasing
Q) from Ho through Tm.
Note, that the available experimental data for the exchange constants in
the heavy RE metals, derived from the isotropie part of the magnon spectra,
were adjusted to a model Hamiltonian, in which the exchange constants cou-
pIe the localized f-moments Si. Thus, in order to compare our exchange con-
stants to the experimental values, our exchange constants are renormalized as
J(q) -+ J(q)s2 / S(S +s) (cf. (1.86) below). The calculated and experimentally
derived exchange forces are semiquantitatively weIl correlated. One should
bear in mind, however, that the experimentally derived values are based on
an oversimplified model, and what really matters is the final comparison of
calculated and measured magnon spectra given below.
One of the first electronic structure calculations of exchange constants for
Gd was undertaken in [76], where the exchange constants were derived within
the Ruderman-Kittel-Kasuya-Yosida (RKKY) model from an expression [77)
containing the electronic Bloch states. The obtained J (q) were larger than
the experimentally derived values by a factor of about 4. This discrepancy
was thought to be due to an overestimation of the exchange splitting of
".
~ Gd
5 /
- - - Tb
.. .. . Dy / \
I \.
- - Ho \
_ .. - Er \
._- T m
g
...... -5
10 Theory Experiment
0.2 0.4 0 .6 0 .8 0 .2 0.4 0. 6 0.8

Reduced wave-vector
Fig. 1.17. The exchange interaction of the magnetie heavy RE met als (see text).
Lejt panel: theoretieal curves obtained as energy derivatives at a collinear spin
configuration; Right panel : experimental data, [63] deduced from the isotropie part
of the magnon dispersion
bands in LSDA as weB as to unscreened d-f exchange matrix elements used

in the calculation. According to the present analysis, the discrepancy is rather
a matter of a correct description of the exchange correlation couplings be-
tween itinerant and localized spin moments.
The humps in both the calculated and experimental J(q) curves at q =
(0,0, Q) and q = (0,0, 21T jc - Q), with Q between 0.2 and 0.3 in units of
21T j c, are due to the Fermi surface nesting between spin-up and spin-down
Fermi surfaces. This nesting gets more pronounced and the humps sharpen
from Gd towards Tm as the spin splitting of the ban~s decreases and the spin
up and down Fermi surfaces get dose to each other. This can easily be seen
from Fig. 1.18, where the cross sections of the Fermi surfaces of both spin
directions of a collinear ferromagnetic state are depicted on the symmetry
planes. Observe in particular the nesting on the planes M - L - H - K and
r - A - L - M in the cases Ho through Tm. This nesting has been proved
experimentaBy as weB [84].
The anisotropy of the exchange interaction in the real space is weB illus-
trated by Fig. 1.19, where the coupling constants obtained from the reciprocal
space in accordance with (1.84), are plot ted as a function of the space between
neighbors Ru and R>. Obviously, the oscillating features caused by the Fermi
surface nesting, are easily traced out for those R == Ru - R~, which are heing
42 S. Halilov
Gd Tb
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
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
Table 1.3. Magnetocrystalline anisotropy parameters ~lm (theoretical [85] and

experimental [63]) in the common spherical harmonics notation. m == s + S + L
stands for the total momentum per atom (meV/atom)
60 m2 ~4om4 100 x ~6om6 ~66m6

exp. theor. exp. theor. exp. theor. exp. theor.
Tb -0.40 -1.28 -2.2 -0.38
Dy -2.27 2.35 6.3 - 1.07
Ho 1.95 -6.35 18 -4.27
Er -1.02 3.66 9 -2.42
Tm 4.74 -10.95 -31 7.29
within the DFT approach. The latter demands an independent calculation

by means of at least a fuII-potential code. Where possible, we used available
experimental data for the crystal-field parameters [63], otherwise the values
reported in [85J were used. The data used in calculations of the magnon
spectra are collected in Table 1.3. Now we are ready to solve the eigenvalue
problem (1.82) for the lowest eigenvalues, and to discuss the magnon spectra.
We begin with Gd with a collinear ferromagnetic ground state and negligible
crystal-field coupling. The experimental and our adiabatic magnon spectra
are shown on Fig. 1.20. For comparison, the result of the simple model with
rigid f-d exchange coupling (1 -+ 00), which means essentially a Heisenberg
exchange model for site spins s + S, is also shown. The latter is stiffer by
a factor of about 1.5 in the upper part of the magnon spectrum, with the
lower part of the spectrum being, however, less influenced. In the present sit-
uation, the effect of the "soft" f-d exchange coupling on the magnon spectrum
can easily be estimated by the following simple calculation. The spin-wave
Fig. 1.20. The magnon spectrum of Gd: Full lines - fuH calculation, dashed lines
- rigid s - S coupling (I = (0) assumed, dots - experiment [78]
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)
The lowest-order correction in 111 leads to
nWq = nW_q [1- InW s(8+s)

8 ]
q
. (1.87)
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
:>
, 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
to be Ba = 67°. The magnon spectrum for this configuration along high-

symmetry directions is shown on the upper panel of Fig. 1.23 (no experimental
data is available).
The most distinctive feature of the spectrum is the degeneracy removal
along the r - A direction. At small wave vectors, q ---+ 0, the magnon
dispersion law is linear in q and has no gap at q = 0. It is a well-known
case of a phason mode, which is related to a uniform change of the spiral
azimuthal angle <p. Although the rotation around the hexagonal axis in a cone
configuration costs no energy, because of the effective averaging out of the

basal-plane anisotropy, this rotation is not a proper symmetry operation.
Thus, one deals here with a situation where a continuous symmetry is spon-
taneously broken and where, therefore, the well-known Goldstone theorem
predicts the existence of collective modes with energies going to zero. Within
the basal planes, the energies retain their double degeneracy. The lower panel
of Fig. 1.23 contains the energies of excitations along the [OOq] direction for
a metastable basal-plane collinear ferromagnetic configuration in an external
magnetic field B ext = 1 T, applied in a basal plane direction.
Since the axial rotation of the spin configuration now costs energy, there
is no situation of the Goldstone theorem any more, and consequently an
anisotropy gap appears in the magnon dispersion at zero wave vector, with
a q2 dispersion low for small q.
The situation in Er is similar, where the ground state is a ferromagnetic
c-axis cone with qo ~ (O,O,O.17[11,/a]). The theoretical cone angle Bo = 25.5°
is very elose to the experimental One. The spectrum, shown on Fig. 1.24, has
no gap at the origin where the dispersion is again linear.
The removal of the degeneracy along the [OOq] direction is remarkably
more pronounced in theory than in experiment (lower panel). This discrep-
ancy has already been noted earlier [81] and might not be removed by an
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
adjustment of the crystal-field anisotropy, which appears as ascale factor in

the dispersion relation. Probably, the Hamiltonian (1.73) is not complete and
the two-ion anisotropy terms, not included in the present treatment, could
improve the situation.
The most complicated situation is met in Tm, since no spiral configura-
tion seems to satisfy the equilibrium condition for the given exchange and
crystal-field parameters. As is known, the ground state of Tm is ferrimagnetic,
with 3 subsequent moments along the c-axis aligned parallel, and 4 furt her
moments in the opposite direction. We chose to consider the computationally
more manageable case of a c-axis ferromagnetic configuration, stabilized by
an external magnetic field B ext = 2 T. The magnon spectrum along high-
symmetry directions is shown in Fig. 1.25. We are not aware of experimental
values for this case.
1. 7 Finite-Temperature Spin Dynamics:

Possible Scenario for Phase Diagram Description
Since the formal theory of the thermodynamics of magnetic systems at low

temperatures is weIl established, here we will just apply our numerical results
to the transition metals in order to justify the relevance of the adiabatic
method and to see how, in principle, the basic ideas of the collective modes
and mean-field approach could be unified to describe the area of the elevated
temperatures.
In the case of the heavy REM, the effective "masses" (spin and orbital
moments) are rather large to qualify them as classical quantities and con-
sider even the plain Langevine mean-field method as accurate enough for the
description of the thermal behavior. The details can be found elsewhere (see,
e.g. [63]), so by furt her consideration the attention will be focused primarily
on the 3d-transition metals.
If magnons are well-defined quasiparticles (harmonie regime), i.e. their
lifetime is long enough, the average mode amplitude (e~qn) is obtained by
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
L1M Z = LL1M; = gJ.tBohr L

Jl- n
(~:c)~ 1
BZ
d3 qn qn (1.90)
for the thermal reduction of the average total magnetization.

Note that the adiabatie moments have already been subjected to a quan-
tum average over the fast electronie motion and hence are no longer quantum
angular momentum operators. Consequently they are also no longer subject
to zero point quantum fluctuations that reduce the z-component of a quantum
spin S from h[S(S + 1)]1/2 to ::; hS. The equations of motion (1.77) are
already the quantum average of (1.76) and do not yield zero point motion
which was already averaged out in (1.34). This point suggests caution in using
the adiabatic approach and in particular (1.34): for instance, it will probably
not work for very weak magnetic materials.
At elevated temperature the occupation numbers and hence the ampli-
tudes (()Jl-qn) of long-wavelength modes become large. The modes start to
interact both kinematically (deviation from bosonic character) and anhar-
monically, and finally the magnon picture is to be replaced by a pieture of
strong long-wavelength transverse spin fluctuations. In 3d-metals, the (T = 0)
magnon energies for q-vectors close to the Brillouin zone boundary are much
higher than kBTc , the thermal energy at the Curie temperature, and hence
spin clusters of many atoms are expected to be aligned even at the Curie
temperature. Figure 1.5 then suggests that longitudinal spin fluctuations
may be neglected even up to Tc. We therefore divide the temperature scale
into two parts: the harmonie regime, roughly T rv T c /2, with well-defined
quantum excitations, and the anharmonie regime, roughly Tc/2 rv T rv Tc,
where one may call on semiclassical averaging schemes for the mean-field free
energy as in [17], or, since longitudinal spin fluctuations seem to be negligible
and the aligned clusters still seem to be mesoscopic, resort to the Langevin
approach with an effective Weiss field
BWeiss = L' JV[Jl-+Rj (MJl-) = L J~Jl- (MJl-) , (1.91)

Jl-,R
50 S. Halilov
acting on the moment Mv at a site of the sublattice 1/ and causing a thermally

averaged sublattice polarization
(Mv) = Mv coth [MvBweiss] _ k~T (1.92)

kBT BWeiss
in the direction of BWeiss' The prime on the sum of (1.91) means omission of
a possible on-site term (for IL = 1/), and from (1.84) one finds
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
o 200 400 600 800 1000 1200 1400

T [K)
Fig.1.26. Magnetization VS. temperature of Fe (squares, [38]), Co (juli triangles
hep, [39] open triangles fee [38]), and Ni (balls [40]). Solid lines: ealculated from
magnon speetra. Dashed lines: mean field results
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,
which simplifies for a Bravais lattice to the well-known expression
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.
Table 1.4. Curie temperatures of bee-Fe, fee-Co, and fee-Ni in Kelvin
Tc [57] Tc [41] Tc [16] Tc [17] TC, experim [42]

bee Fe 1037 1270 1015 1095 1043
fee Co 1250 1520 1012 1388
fee Ni 430 450 412 627
52 S. Halilov
An improved value for the Curie temperature of Ni, reported recently in

[89], has been obtained within the RPA treatment of the spin fluctuations
at high temperatures. We have to emphasize that this scheme has already
been discussed in [7] and leads to a wrong conclusion: the average ofthe local
magnetic moment squared completely vanishes at the phase transition, which
is obviously at odds with experimental observations. This gives more support
for cluster-based models of thermodynamics at elevated temperatures, which
we will briefly be concerned with in the last section.
1.8 Most Valuable Problems Which Could be

Addressed Within the Adiabatic Approach
Summarizing, we will not concern ourselves with a long list of the internal
problems of SDFT: elimination of self-interactions in exchange, which would
enable an aIl-electron consideration in order to, for example, investigate more
accurately the rare-earth systems; find a systematic way to include the effects
of relativity in systems with heavy atoms, such as the effects of the orbital
polarization, etc.
Here, we rather focus on some projects that could emerge from the adi-
abatic consideration of itinerant magnetism and eventually may have an
impact beyond magnetic research. We have already seen from the evalua-
tion of the Curie temperature that the standard Weiss mean-field theory
involves, by and large, the correct physics of the metallic magnets at elevated
temperatures. On the other hand, the relatively simple Weiss concept is
also a weakness. In this model each magnetic atom is in a homogenized
environment, so that the local fluctuations at elevated temperatures and
wave-like excitations at low temperatures are beyond the model's scope.
That is, the model is too coarse to distinguish the significant short-range-
order effects, which in accordance with the photoemission and the neutron
scattering measurements persist weIl above Tc, through the formation of
magnetic clusters with jluctuating size on the nanoscale, on the one hand.
On the other hand, it seems to be areal challenge to incorporate the low-
energy collective excitations into the cluster model, specifically with fluc-
tuating size. Quite recently, an attempt was made [87] to go beyond the
fixed-size cluster model by using a thermodynamical formalism known as the
"generalized ensemble" (see, e.g. [88]). That is, it was shown that the effects
of local moment correlations can be accounted for by applying the standard
mean-field theory to finite clusters with unrestricted sizes, which particularly
yields the measured critical exponents about Tc. Probably, the next step in
building the unified thermodynamics valid for the whole temperature scale,
is to combine the concept of low-temperature wave-like excitations, which are
weIl defined in the presence of a long-range order, and the mean-field cluster
model with unrestricted sizes, the latter fluctuating around the moment-
moment correlation length.
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)
45. T.A. Kaplan: Phys. Rev. 109, 782 (1957)

46. A. Yoshimori: J. Phys. Soc. Jpn. 14, 807 (1959)
47. D.H. Lyons and T.H. Kaplan: Phys. Rev. 120, 1580 (1960)
48. I. Dzyaloshinski, Sov. Phys. JETP 32, 1547 (1957)
49. T. Moria: Phys. Rev. 120, 91 (1960)
50. L.M. Sandratskii: Adv. Phys. 47, 91 (1998)
51. T.A. Kaplan: Phys. Rev. 124, 329 (1961)
52. B.R Cooper, RJ. Elliott, S.J. NetteI, and H. Suhl: Phys. Rev. 127,57 (1962)
53. M. Uhl, L.M. Sandratskii, and J. Kübler: J. Magn. Magn. Mater. 103, 314
(1992)
54. S. Fujii, S. Ishida, and S. Asano: J. Phys. Soc. Jpn. 60, 4300 (1991)
55. S. Akbar, Y. Kakehashi, and N. Kimura: J. Appl. Phys. 81, 3862 (1997)
56. S.V. Halilov, A.Y. Perlov, P.M. Oppeneer, and H. Eschrig: Europhys. Lett. 39,
91 (1997)
57. S.V. Halilov, H. Eschrig, A.Y. Perlov, and P.M. Oppeneer: Phys. Rev. B 58,
293 (1998)
58. S.Y. Savrasov: Phys. Rev. Lett. 81, 2570 (1998)
59. K. Karlsson and F. Aryasetiawan: Phys. Rev. B 62, 3006 (2000)
60. C.G. Shull and H.A. Mook: Phys. Rev. Lett. 16, 184 (1966)
61. T. Moriya: Spin Fluetuations in Itinerant Electron Magnetism (Springer,
62. P. Fulde and M. Loewenhaupt: Adv. Phys. 34, 589 (1986)
63. J. Jensen and A.R Mackintosh: Rare Earth Magnetism (Clarendon Press,
Oxford 1991)
64. A.Y. Perlov, S.V. Halilov, and H. Eschrig: Phys. Rev. B 61, 4070 (2000)
65. L. Nordström and A. Mavromaras: Europhys. Lett. 49, 775 (2000)
66. L. Steinbeck, M. Richter, H. Eschrig, and U. Nitzsche: Phys. Rev. B 49, 16
289 (1994)
67. Q. Niu and L. Kleinman: Phys. Rev. Lett. 80, 2205 (1998)
68. Q. Niu, X. Wang, L. Kleinman, W.-M. Liu, D.M.C. Nicholson, and G.M.
Stocks: Phys. Rev. Lett. 83, 207 (1999)
69. R Gebauer and S. Baroni, Phys. Rev. B 61, R6459 (2000)
70. J. Moody, A. Shapere, and F. Wilczek, in Geometrie Phases in Physies, ed.
by A. Shapere and W. Wilczek, (World Scientific, Singapore 1989)
71. O. Grotheer, C. Ederer, and M. Fähnle: Phys. Rev. B 62, 5601 (2000)
72. O. Grotheer, C. Ederer, and M. Fähnle: Phys. Rev. B 63, 100401(R) (2001)
73. H.L. Skriver, in Systematies and properties of the Lanthanides, ed. by S.P.
Sinha, (Reidel, Dordrecht 1983) p.213
74. M.S.S. Brooks, L. Nordström, and B. Johansson: J. Phys. Cond. Matt. 3, 2357;
3393 (1991); M. Richter and H. Eschrig, Physica B 172, 85 (1991)
75. M. Richter: J. Phys. Applied Physics 32, 1017 (1998)
76. P.-A. Lindgard, B.N. Harmon, and A.J. Freeman: Phys. Rev. Lett. 35, 383
(1975)
77. L.M. Roth, H.J. Zeiger, and T.A. Kaplan: Phys. Rev. 149, 519 (1966)
78. W.C. Koehler, H.R. Child, R.M. Nicklow, H.G. Smith, RM. Moon, and J.W.
Cable: Phys. Rev. Lett. 24, 16 (1970)
79. J. Jensen, J.G. Houmann, and H. Bjerrum Mreller: Phys. Rev. B 12, 303 (1975)
80. RM. Nicklow, N. Wakabayashi, M.K. Wilkinson, and RE. Reed: Phys. Rev.
Lett. 26, 140 (1971)
56 S. Halilov
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
H.J.F. Jansen, G.S. Schneider, and H.Y. Wang
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.
2.2 Definition of Magneto-crystalline Anisotropy
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
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
EMCA = loHS M . dH I - loHS M . dH I

easy hard
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
anisotropy and try to improve on it. It is tempting to define
L1E(H) = loH M .dHI -loH M .dHI '

easy hard
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,
for all reversible processes. In that case we have
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
E(8) ~ /.'-0 M . dH IH~H. '
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
hysterises can be a problem. On the other hand, torque measurements are

often faster, and care has to be taken not to heat the sampIe.
The magnetic Gibbs free energy is defined by
G(T,H) = U -TS-M ·H,

when we ignore the energy stored in the field itself. The latter is important,
because based on the definition above magnetic susceptibilities would always
be positive. For the current discussion, however, the corrections are small
and we ignore the changes in energy stored in the field. Based on the Gibbs
energy we find that
M=_8G
8H'
and hence
EMCA = G(T,H = Hs,hard) - G(T,H = Hs,easy) ,

which relates the anisotropy energy immediately to differences in the Gibbs
free energy. This is a useful formula for studying effects like domain wall thick-
ness. It also makes it possible for theorists to relate the magneto-crystalline
anisotropy energy to the internal energy U. Two points are important, though.
Note first that the internal energy has to be calculated in the presence of
a field along the easy or hard axis. This is difficult for many calculations,
and approximations have to be made. Also, when the temperature is not
zero, the entropy has to be induded, and this is impossible for electronic
structure calculations. In order to describe the electronic structure at non-
zero temperature we need to indude the motion of the atoms as weIl as the
applied field. Although we know how to do this in principle, we are not even
dose to being able to do this in practice.
In summary, we have given three definitions of the magneto-crystalline
anisotropy. Experimental data have been derived using all of them, and the
definitions are equivalent if sufficient care is taken. The next quest ion is which
definition a theorist should use to make contact with experiment. In the next
seetion we study an example in order to show some of the points that need
to be considered.
2.3 Examples from Model Calculations
In order to illustrate the merits of the various manners of obtaining the

magneto-crystalline anisotropy energy we discuss the results of a number
of very simple model calculations [19]. The model system consists of three
dimensional spin-vectors Si on a two-dimensional ribbon with width 10 and
length 50. We use periodic boundary conditions for the length of the rib-
bon and open boundary conditions for the width. The spins interact only
via nearest neighbor Heisenberg and dipole interactions. We also include an

external magnetic field H.
Model calculations along these lines are very valuable. A simple appli-
cation is the following. Suppose we have performed ab initio total energy
calculations for a material with different spin structures. This has given
us reliable and accurate values for the total energy of the material in the
ferromagnetic state and several anti-ferromagnetic and ferrimagnetic states.
These ab initio energy values are then used to find the exchange constants
J ij in the expression
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
E[{Si}] = L Cij (Si· Sj - 3Tij -2(Si . Tij)(Sj . Tij))

i<j
+ LDijTij-4(Si· Tij)2(Sj· Tij)2,

i<j
where the interaction constants C ij and D ij can be obtained from ab initio
calculations, at least in principle.
We have chosen a simple version of these model interactions, using only
nearest neighbors, and our model Hamiltonian is
1{ = -J L Si . Sj - H . L Si - K L (Si· nij)(Sj . nij) .

<ij> i <ij>
Using a simple Monte-Carlo approach the total energy of this model is eval-
uated as a function of temperature and strength of the applied field.
Figure 2.2 shows the results for J = 1, T = 1, and K = 1. The behavior
is clearly as expected, with a saturation field of order one. The values of the
parameters used in this calculation are not very realistic, however. Therefore,
we show in Fig.2.3 the results for J = 1, T = 0.25, and K = 0.05. In real
materials the ratio K / J is even smaller, but the values we have chosen here
are sufficient to demonstrate the effects we need. The magnetization curves
in Fig. 2.3 are much more noisy. The definition of the saturation field is not
as straightforward, but this field is clearly of order K. It is also obvious that
the magnetization is still increasing as a function of field. Note that this effect
would be much smaller if we had made K much smaller.
In Fig. 2.4 we show the magneto-crystalline anisotropy energy as a func-
tion of field for the parameter values corresponding to Fig.2.3. The bottom
1.2
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
data, with label Delta E, are derived from
L1E =< 1i > (H,hard)- < 1i > (H,easy) ,

while the top data, with label Int delta M, follow from
J L1M= loH M.dHI -loH M.dHI

easy hard
The second curve is what we need according to definition one of the

magneto-crystalline anisotropy energy. We see that this curves reaches a max-
imum, and then slowly decreases. Defining the saturation field to correspond
to this maximum, we obtain a value of around 0.55, which is much larger than
what we would infer from Fig. 2.3. Nevertheless, this also gives a reasonable
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
precise value of the magneto-crystalline anisotropy energy of 0.04, with error

bars on the order of one percent.
The bottom curve in Fig. 2.4 shows a very different behavior. The differ-
ence between these two curves is the entropy contribution, and we see that at
a temperature of 0.25 all of the magneto-crystalline anisotropy energy is due
to the difference in entropy! This demonstrates that obtaining data for the
magnetic anisotropy as a function of temperature will be a formidable task
indeed. Even more intriguing, the maximum in the top curve corresponds
to zero energy difference. For fields less than K the effect of the entropy is
very small, however, and perhaps it will be possible to make some predictions
based on the low field data. The most disturbing aspect of the bottom curve
is the amount of noise, due to the Monte-Carlo procedure. This indicates that
caIculated energy differences are very sensitive to numerical approximations,
an effect which has been observed in many ab initio caIculations.
CaIculating the magneto-crystalline anisotropy energy from first prin-
ciples has not been easy. The common route via differences in energy is
numerically very demanding, since integrations in the Brillouin zone have
to be done with great precision [21,22]. The obvious alternative is to base the
caIculations on magnetization curves, but this is also very demanding. First
of all, we need to repeat the caIculations for many values of the applied field,
which increases the computational time enormously. One would expect to
gain because many fewer integration points in the Brillouin zone can be used,
but it is not clear if this is actually true for very small values of the applied
field in the range we need. We have not discussed torque measurements, but
previous experience has shown that the numerical load is equivalent to that
needed to obtain energy differences [23]. The main reason is that the energy
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.
2.4 Magneto-crystalline Anisotropy

in Density Functional Theory
Density functional theory [24,25] is the appropriate tool for the evaluation of
total energies of solids. The treatment of the interactions between electrons
in solids is not straightforward, but density functional theory in the local
density approximation has had many successes in describing physical sys-
tems. For example, calculations based on density functional theory predicted
a new phase in silicon under pressure [26], and this phase was found later
in experiments. Also, calculations using this theory predicted perpendicular
anisotropy in CoNi 2 films, which was also later confirmed experimentally
[27,28]. The fact that density functional theory was used not only to describe
known data but also to predict new data successfully is the hallmark of a good
theory.
The majority of the applications of density functional theory uses the non-
relativistic Schrödinger equation, or variants where some relativistic effects
are included [29]. Magnetism, however, is a relativistic effect, and in order
to describe magneto-crystalline anisotropy we need to start with relativistic
quantum field theory [30-35]. Spin-orbit coupling is essential, and its con-
sequences have to be examined with great care. In this case it is easiest to
derive the equations using the constrained minimization procedure [36,37,9].
The only approximation we make is to describe the applied external fields by
classical, non-quantized, entities. This is in general an acceptable procedure.
The Hamiltonian density in quantum field theory is given by
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
states of the electrons and photons,
F[e,j] = min < f d3 r1-l(r) >e,j,
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
K[Q,j] = min < f d 3 r: '!f1 t (r) [~c a· V + ß)mc2 ] '!f1(r) :>o,j .
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.
The second part of the universal functional :F contains the Coulomb

interaction between electrons. After minimizing the interaction with respect
to the photon fields, which are treated in a quantum mechanical manner, we
find that the electron-electron interaction is modified [38-41], and in lowest
order we have
2
Ir-r
e 'I: '!f;!(r)'!f;l(r')WKAj.lv'!f;j.I(r')'!f;v(r):,
where we have a summation over repeated Greek indices as usual. The inter-
action kernel W is given by
1 ( [aAj.I· (r' -r)] [aKV ' (r' -r)])

WKAj.lV = OAj.lOKV - 2 aAj.I· a KV + (r' _ r)2 .
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 .
The Hartree form of the Breit interaction is interesting, and we find
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,
where m is related to the current density by
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
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
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
1iint(r) =<: 'ljJt(r) [-jLB J

d3rBexternal(r) . .E] 'ljJ(r) :>,
and leads to a term in the Dirac equation in four-vector form
-jLB Jd3rBexternal(r)· .E'ljJ(r).
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
the Dirac equations with an extra ß in the magnetic interaction, as long as

one constructs the spin up density from the first and third component of the
wave function!
We conclude this section with aremark about the exchange-correlation
potential used in the local spin density approximation. The homogeneous
interacting electron gas is a system for which we know the energy as a function
of charge and spin density. These energies are used to construct an effective
Coulomb potential and magnetic field in the Schrödinger or Dirac equation,
which only depend on the density at the same point in space. Of course, in
the usual formalism these effective potentials are combined to a spin-up and
a spin-down potential, but from a theoretical point of view it is bett er to
think in terms of magnetic fields. If we now include spin-orbit coupling into
the equations, the direction of the spins is not uniform anymore, and at each
point in space we are able to calculate the charge density e( r) and the spin
density S(r). The homogeneous electron gas gives us the local spin density
approximation for the magnitude of the effective magnetic field:
Bxc(r) = BLsDA(e(r), S(r)).

In a relativistic system it is only natural to assume that the direction of this
exchange-correlation field is along the local spin direction, and therefore we
have
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
definition of the exchange-correlation field:
The quantity Mo is interpreted as the direction of the total magnetization of

the sam pIe , and it is now easy to perform a calculation to obtain the ground
state energy as a function of this parameter, Egs(Mo).
For the usual total energy calculations of magnetic transition met als this
approximation works fine. This can be easily demonstrated. The numerical
error in the calculated spin moment per unit cell is often a few percent. This
error is a direct consequence of the limitations of the integrations in the
Brillouin zone, but also an indirect effect due to errors in the equilibrium
volume, for example. Yet this precision is almost always sufficient to find the
stable ground state structure. Since in these materials the orbital moment is
quenched, and is also only a few percent of the total moment, the inclusion of
the orbital moment will in general not affect the conclusions which are drawn
based on present local spin density calculations. The quest ion whether this
approximation is sufficient for magneto-crystalline anisotropy calculations is
much harder to answer, since the energy differences are much smaller. It is,
however, the only thing we can do at this moment! It is troublesome, though,
that in the expression for the magneto-crystalline anisotropy energy used
in the torque mode the only important quantity is the component of the
magnetization perpendicular the the applied field. This does cast doubt on
the procedure normally used, in which the perpendicular component of the
magnetization is not calculated correctly.
2.5 Numerical Problems in the Calculations
There have been many calculations of the magneto-crystalline anisotropy

of transition metals. The first calculation based on energy eigenvalues was
reported more than sixty years aga [16]. Due to the enormous increase in
computing power the quality of the calculations has improved, but many
quest ions remain. In order to understand the physics of magneto-crystalline
anisotropy in general, it will be very useful to be able to explain what happens
in the basic cubic transition metals iron and nickel, and also in hexagonal
cobalt. It turns out that the reported results for iron and nickel show a large
variation, and it is clear that issues of numerical precision playadominant
role.
The problem is related to the fact that the energy difference between
a sampIe magnetized along the <100> and along the <111> direction is
very small. This small difference is obtained by subtracting two numbers
which are about six orders of magnitude larger than the final result. This
subtractive cancellation is a well-known problem, described in all numerical
analysis text-books. At first it seemed reasonable to attribute the difference
in results to the use of different methods or different approximations for the
exchange-correlation potential. Certainly the energy values vary a lot, but

one can show that the energy differences needed in our problem are not very
strongly affected. The use of different methods mainly causes a systematic
shift and does not change the magneto-crystalline anisotropy energy very
much, at least not by as much as seen in the difference between recent
results.
Next one needs to investigate the effects of numerical approximations
within certain methods. The dependence of the results on quantities like basis
size needs to be determined. Also, one needs to investigate the convergence
to self-consistency. Methods that use a small basis sometimes have problems,
the most well-known case being hcp cobalt studied with the linear muffin tin
orbital method, where the energy changes sign when f states are ineluded in
the set of basis states [27,47]. Plane wave based methods have an advantage,
and do not suffer from basis set limitations. In general one can conelude that
convergence can be achieved in all methods, but that methods like the linear
muffin tin orbital method might start to loose their advantage in simplicity
since states with quantum number 1 of at least equal to four have to be
ineluded.
One numerical approximation is present in all calculations. In order to
obtain the total energy we have to perform an integral over the Brillouin zone.
It is precisely this integration that gives rise to large numerical problems, and
can be shown to be the cause of all problems. In this section we will address
the issue of the Brillouin zone integrations. A very complete analysis is found
in the thesis of Guenter Schneider [21], which is published on the web.
The problem we face is traced back to the Kohn-Sham approach to solve
the density functional equations. In order to find the non-interacting kinetic
energy we have to solve the equations of motion (the Dirac equation in our
case) for many different wave vectors. This is a very time-consuming step. The
Dirac equations have energy eigenvalues that are on the order of Rydbergs,
but the magneto-crystalline anisotropy energy is of order milli-Rydbergs to
micro-Rydbergs. As it turns out, the energy difference is very sensitive to the
choice of the set of wave vectors ineluded in the calculation.
A number of recent results for the magneto-crystalline anisotropy energy
for iron and nickel is summarized in Table 2.1. The first two rows show older
data that are elearly very different. The Brillouin zone integrations in these
two calculations are very approximate, the number of points used is much
smaller than in the later calculations. The results reported by Daalderop are
the first to give a detailed analysis of the convergence in k-space. Daalderop
estimates that the error bars in his results are almost equal to his values.
The results reported by Trygg are identical to Daalderop's results. A careful
analysis shows that the procedure used in this paper to perform the k-space
integrations is not converged, but accidentally gives results that are elose
to the converged values. The next three entries seem to indicate that newer
Table 2.1. Magneto-crystalline anisotropy energy in lieV per atom from various
calculations. The experimental results are included in the last row
Author year reference Fe Ni

Eckardt et al. (1987) [45] 7.4 10.0
Strange et al. (1989) [46] -9.6 10.5
Daalderop et al. (1990) [47] -0.5 -0.5
Guo et al. (1991) [48] 1.8 -3.1
-2.4
Trygg et al. (1995) [49] -0.5 -0.5
Razee et al. (1997) [50] -0.95 0.11
Halilov et al. (1998) [51] -0.5 0.04
-2.6 1.0
Beiden et al. (1998) [52] -0.78 -0.43
Schneider (1998) [21] -0.7 -0.15
expt. [53] -1.4 2.7
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
•
-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]:
E(111) - E(100) = :~.::>n(k,Mo 11< 111 » - LEn(k,Mo 11< 100 ».

occ occ
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].
2.5.1 Linear Tetrahedron Method
A typical integral that has to be evaluated in electronic structure calculations

has the form
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
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
distance between k-points ",,2
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
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
_4L-__ ~ __-L__ ~ ____L-__ ~ __-L__ ~ ____ L-~
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
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
Fig.2.8. Magneto-erystalline anisotropy energy L1E = E(001) - E(111) for fee Ni

as a function of the square of the eharaeteristic distance L1 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 = 100 meV. The number of divisions of the re-
eiproeallattice veetors is indieated at the top of the graph. Open circles denote L1E
ealculated with the regular tetrahedron method , closed circles are results ealculated
with the improved method. Lines are guides to the eye
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.
2.5.2 Gaussian Broadening Method
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
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
8(x) ~ ~ErJe (:~) .
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
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
Fermi surface smearing 0' / (meV)'
Fig.2.9. Band energy of Ni ealculated from a tight-binding model. The Brillouin

integration was performed using special points and Gaussian Fermi surfaee smear-
ing. The various lines represent different fits in order to extrapolate the result to
zero broadening
Fermi surface smearing 0/ meV
29 48 77 96 115 134 153 173 192
--- ---
0.0
\
. --
-0.1
\
-8..
E ' 10
.9 ·0.2 j ~
'"
>CI>
::l
:::::- - -- 160'
w -0.3
---0- 120]
'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
--.---
Fermi surface smearing 0/ meV
5 96 192 288 384
/ .
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
Fermi surface smearing 0' / (eV)'
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,
Fermi surtace smearina (} I meV
48 96 '35 173
0.0
-1).2
-1).4
-1).6 \ e_ e_ e- e_ --... ___•
~~/~
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
.1.6 '--__ ~ __L __ _ ~ _ _' - -_ _~_ _' - -_ _~_ _' _ __ _~___'
o '000 2000 JOOO 4000 5000
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.
2.5.3 Conclusions for Current N umerical Results

In this section we have elearly demonstrated that the numerical problems
in the ca1culation of the magneto crystalline anisotropy energy are large.
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
number of valence electrons

Fig.2.13. Magneto-erystalline anisotropy energy iJ.E = E(OOI) - E(lll) for fee
Ni (Aso = 100meV) as a function of the number of valence electrons as calcu-
lated with special points and Gaussian Fermi surfaee smearing with O"/(meV) =
(29, 48,96,135). The results for 240 3 k-points using the improved tetrahedron
method are shown as referenee (open circles)
The ealculations are tedious, time-eonsuming, and most importantly they

need large amounts of computer memory. But when we do the caleulations
carefully, a common picture emerges. For iron and cobalt one predicts the
correct easy axis, but the energy is too small by a factor of three. For nickel,
however, all converged calculations give the wrong easy axis and a value of
the energy which is too small by a factor of five. These discrepancies are
not due to numerical imprecision, and an explanation has to be found by
investigating which approximations in the description of the physics have
been made.
2.6 Missing Physics
It is always very disappointing when we have to work extremely hard to

obtain sufficient numerical precision and then to find that the results are not
1.4
1.2
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
number 01 valence eleclrons
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)
accurate. Why is the calculated value of the magneto-crystalline anisotropy

energy so small? Why does it have the wrong sign for nickel? What is missing?
In short, which changes in the electronic band structure are needed in order
for theory and experiment to agree?
An obvious quest ion relates to the Fermi surface. Since the energy dif-
ferences are small, could a small change in Fermi surface introduce large
changes in the results? In order to investigate this furt her we present in
Fig. 2.15 results for the magneto-crystalline anisotropy energy as a function
of the number of valence electrons. These data are based on a rigid band
model, the only quantity which changes is the value of the Fermi energy.
In Fig. 2.15 we show results obtained with the linear tetrahedron method,
and also include the best results from the improved tetrahedron method.
There are some strong oscillatory features, as discussed before. In the lin-
ear tetrahedron method they are easier to obtain than with the Gaussian
broadening method. For the discussion in this section we look at the changes
in the magneto-crystalline anisotropy energy as a function of band filling
near the nominal value of ten. The changes are very small, which indicates
that changing the Fermi surface is not important. We have to add at least 0.2
electrons in order to see a change of sign in the magneto-crystalline anisotropy
energy. In order to create a large change we need to move a band crossing
through the Fermi surface. This requires an unrealistically large change in
band structure. Therefore, the changes we need in order to reconcile theory
and experiment are changes that afIect the whole band structure, and not
only the Fermi surface.
2.0
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
number of valence electrons

Fig.2.15. Magneto-erystalline anisotropy energy .tJ.E = E(OOl) - E(l11) for fee
Ni (ASO = 100meV) as a function of the number of valenee eleetrons. Results are
ealculated with the linear tetrahedron method and 40 3 , 80 3 , 1203 , 1603 , and 240 3
k-points in the fuH Brillouin zone. The result for the improved tetrahedron method
and 240 3 k-points is shown as referenee (connected open circles)
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,
Homogeneous XC
Spin Moment
L-S Coupling
MAE
Fig.2.16. Without orbital polarization exchange-correlation effects based on the

homogeneous electron gas drive both the orbital moment and the magneto-
crystalline anisotropy
and in first approximation we use
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
Homogeneous XC Atomic XC
Spin Moment Orbital Moment
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
material no OP with OP experiment

iron -0.5 -1.8 -1.4
cobalt -29 -110 -65
nickel -0.5 -0.5 2.7
enough to cause the energy differences needed to describe magneto-crystalline

anisotropy. Figure 2.17 shows how the effects of spin-orbit coupling are dif-
ferent when a small orbital moment is present from the start.
Table 2.2 shows the results of calculations with orbital polarization in-
cluded [64]. For iron and cobalt the theoretical values are now larger than
the experimental values. This is encouraging, and could mean that a better
description of Hund's second rule gives the right answer. The results for nickel
are very remarkable. There is no change, and theory still predicts the wrong
easyaxis.
Orbital polarization is one way to include exchange-correlation effects
describing orbital motion. Another way is to assume that the orbital moment
a)
E
~
3>Ql
:::t
LJ.J
<l
200 400 600 800 1000
spin·orbit coupling Aso ( meV)

b) 100000
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
is mostly parallel to the spin moment and to write

Beff = a'S,
or equivalently to assume that the spin-orbit interaction has changed and can
be described by
AeffS, L,
In other words, the bare spin-orbit coupling parameter is renormalized be-
cause of the many body effects. This is certainly a good physical description.
It is also easy to incorporate in oUf model calculations [21,65].
In Fig. 2.18 we show the result of the magneto-crystalline anisotropy en-
ergy as a function of spin-orbit coupling strength for iron. The top panel
shows the data normally, while the bottom panel shows the data using a dou-
ble log scale (using the absolute value of the energy). Arguments based on the
a)
j
/
60
40
i
:>Gl
20
:::l. o ..•••••••• ~ ......................................... ;/ ..

'.'-"
..•................ I
W
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
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.
2.7 The Role of k-Space
The numerical calculations of the magneto-crystalline anisotropy energy are

demanding, because we need many integration points in k-space. The question
why we need this many points is still unresolved. It seems clear, though, that
even when we have the correct exchange-correlation potential we still need
many integration points. Therefore, it is useful to investigate the role of k-
space further. Especially, if we can find regions in k-space that dominate
the contributions to the magneto-crystalline anisotropy energy we might be
able to speed up calculations for similar materials by only focusing on such
regions.
Using a tight-binding model Hamiltonian we can investigate the role of
k-space very easily. If we change one of the parameters in the model Hamilto-
nian we effectively change the electronic structure in the whole Brillouin zone.
In addition, changes in particular parameters in the model Hamiltonian can
be related to particular interactions between atoms. If we can identify param-
eters which have a large effect on the magneto-crystalline anisotropy energy
we will be able to draw some simple conclusions on effects of impurities, for
example.
We have systematically changed all parameters in the model Hamiltonian
[66]. Changes in the on-site parameters have no large effects. We did find
certain parameters that caused large changes, though. The thesis of Haiyan
Wang analyzes three cases for both iron and nickel in detail. Here we report
only the results for changes in the parameter xy-d2-110-down. The first two
labels refer to the type of orbital, the third label to the direction between the
atoms, and the last label indicates spin. Figure 2.20 summarizes the results.
The maximal changes observed here are larger than the actual values of the
magneto-crystalline anisotropy energy at the reference value.
The shape of the plots in Fig. 2.20 is similar to what we observe when we
increase the number of electrons in such a manner that a band crossing moves
2e-06
-2e-06
~~ L-~ _ _~~_ _~~_ _~~_ _~_ _~~_ _~-"

-0.04 -0.035 -0.03 -0.025 -0.02 -0.015 -0.01
Fig.2.20. The change in magneto-crystalline anisotropy energy in Rydberg far Ni
as a function of parameter change for the matrix element xy-d2-110-down
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
0.0004 r--~-r-~-,.-~-.-~-.-~-.-----,
0.0002
-0.0002
-0.0004 l-.~_-,- _ _-,-_ _--,-_ _--,-_ _ ----,_~----'
-0,04 -0.035 -0,03 -0,025 -0,02 -0 ,015 -0.01
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
-0.0001 l-._ _L -_ _ L-_-lo:-'____~'____~'____ ____'
-0.04 -0 .035 -0.03 -0.025 -0.02 -0.015 -0.01
Fig.2.22. Change in LlE in Rydberg for fee Ni(,\so = 100 meV) as a function of
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
to the value of the magneto-crystalline anisotropy energy)_ Therefore, the

contribution of all energy bands has to be taken into account.
The previous results show that the bands are coupled, but still allow
for the Fermi surface to play an important roIe_ This we investigate with
1e·05 r---r---..------r---..,......~-_r_-__,
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
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
-Se'06 ' -_ _ _L_~_L-_ __L_~_ _. l . -_ _ _ _~
-0.06 -0.04 -0.02 0.00 0.02 0.04
Fig. 2.24. The change in magneto-crystalline anisotropy energy as a function of

the minimum absolute distance to the Fermi surface, both in Rydberg. Results are
for Ni, the change in parameter xy-d2-110-down corresponds to the largest absolute
change in magneto-crystalline anisotropy energy
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
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
We now define a correlation coefficient C(k) by
C(k) = jdpc(k,P) E(p) .

EN(k) E N
If the shapes of the curves E(k,p) and E(p) are the same, the correlation
is one, and when the shape is opposite, the correlation is minus one. Zero
correlation means that the curves are orthogonal (and hence uncorrelated).
In Fig.2.25 we show histograms of the correlation. In this case there is
a single peak in the histogram at zero correlation. For iron, on the other hand,
1~ r--------.------~--------~------_.
1000
500
o 0.5
Fig. 2.25. Histogram of correlation coefficients as defined in the text corresponding

to Fig.2.24
3e-OS r-------,---~-_,_-~-___,r_---_,
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
perhaps areal space integration. As we discussed before, this is not the

answer, since in real space the convergence is limited by the oscillations due
to the presence of the singularity in occupation at the Fermi level.
One final intriguing point remains. In each of the six cases we studied
we looked at the set of k-points that are most strongly correlated or anti-
correlated. We arbitrarily considered 10% of the points selected by requiring
that they have the largest values of IC(k)l. These points were displayed in
a three dimensional plot, which could easily be analyzed by rotating the plot
on the screen. The surprising result was that for iron the highly correlated
or anti-correlated points were along the <100> equivalent axes, while for
nickel they were along the <111> equivalent axes. Hence in k-space these
highly correlated or anti-correlated points were situated along the easy axis
directions, but now in reciprocal space. This might be coincidental, since we
only investigated this for two materials, and three parameter changes for each
material. Nevertheless, there seems to be some order in this madness after all.
2.8 Conclusions
The calculation of the magneto-crystalline anisotropy energy for cubic mate-

rials is numerically very difficult. In this chapter we have shown exactly how
difficult it is, and what needs to be done. The main problem is that the final
result is very small. Therefore, one expects that some kind of perturbation
theory should be possible. The question is, what is the right one!
There are two parameters which are smalI. First of all, the spin-orbit
coupling is smalI. In comparing energy differences with the net moment along
two different directions we find, however, that the contributions from the odd-
order terms always cancel. In cubic materials there is an extra cancellation of
the second-order terms and the lowest contribution is fourth-order. Therefore
an expansion in terms of the spin-orbit coupling parameter is not useful,
because even the second-order terms in perturbation theory are already com-
plicated.
A more useful approach is to consider the difference between the Hamilto-
nians with different directions of the applied moment. This difference corre-
sponds to a small perturbation [35J and is the reason why the Force Theorem
works in the calculations of magneto-crystalline anisotropy. In this case one
only needs the lowest order term in the perturbation expansion. The problem
with this approach is that it is essential to use relativistic wave functions. If
one uses the wave functions from a calculation without spin-orbit coupling
the final result is zero. The changes in the wave functions after including
spin-orbit coupling are the essential ingredient for the magneto-crystalline
anisotropy.
This situation is very unsatisfactory. Even if we would be able to perform
calculations with billions of integration points, we would not have gained any
physical insight. We would only have confirmed that density functional theory
indeed pro duces correct results if we use the correct exchange-correlation

potential. This is, of course, by no means a minor accomplishment, but it
would still not be of practical use. It would not tell us why magneto-crystalline
anisotropy is perpendicular in certain materials or how we can change the
anisotropy in a simple manner. In other words, at this moment we have
a quantitative picture but not a qualitative picture.
Finding the right form of a perturbation expansion in which the small
magneto-crystalline anisotropy can be obtained easily at the 10% level is an
important goal. This must be possible. There has to be some expression that
is small (in our case proportional to the square of the fine structure constant)
by nature.
In general, the results of density functional calculations of the magnetic
anisotropy in a large range of materials, bulk and films, are very encouraging.
In my opinion an improved description of exchange-correlation should be
capable of leading to the correct easy axis in nickel, just like such an im-
proved description gave the right ground state for iron [67]. If we accept this
premise, our challenge is easily formulated. We need to find an efficient way
of calculating magneto-crystalline anisotropy in density functional theory.
Acknowledgments
The material presented in this chapter is based on work performed by many

students in the group of Henri Jansen at Oregon State University, especially
Guenter Schneider (PhD 1999), Haiyan Wang (MS 2000), David Matusevich,
and John Thorndike (REU 2000). This work was made possible by the Office
of Naval Research under grant N00014-94-1-0326. The REU work of John
Thorndike was supported by NSF under grant PHY-9820483.
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)
11. S. Gasioröwicz: Elementary Parlicle Physics (Wiley, New York 1966)

12. F. Gross: Relativistic Quantum Mechanics and Field Theory (Wiley, New York
1993)
13. L. Ryder: Quantum Field Theory (Cambrige U.P., New York 1985)
14. G.S. Mahanji: Philos. Trans. R. Soc. London 228, 63 (1929)
15. J.H. van Vleck: Phys. Rev. 52, 1178 (1937)
16. H. Brooks: Phys. Rev. 58, 909 (1940)
17. E.1. Kondorskii, E. Straube: Zh. Eksp. Teor. Fiz. 63, 356 (1972) [Sov. Phys. -
JETP 36, 188 (1973)
18. N. Mori, Y. Fukuda, T. Ukai: J. Phys. Soc. Jpn. 37, 1263 (1974)
19. Summer work of John Thorndike under the NSF-REU program using codes
written by David Matusevich.
20. S.S. Peng, H.J.F. Jansen: Phys. Rev. B 43, 3518 (1991)
21. G. Schneider: Calculation of Magnetic Anisotropy. PhD thesis, Oregon State
University (1999) published at
http://www.physics.orst.edu/henri/Jansen.html
22. G. Schneider, H.J.F. Jansen: web-publication at
http://www.physics.orst.edu/henri/Jansen.html
23. R.P. Erickson: J. Appl. Phys 79, 4808 (1996)
24. P. Hohenberg, W. Kohn: Phys. Rev. 136, B864 (1964)
25. W. Kohn, L.J. Sham: Phys. Rev. 140, A1133 (1965)
26. K Chang et al.: Phys. Rev. Lett. 54, 2375 (1985)
27. G.H.O. Daalderop: Magnetic Anisotropy from First Principles. PhD thesis,
Technical University, Delft (1991)
28. G.H.O. Daalderop, P.J. Kelly, F.J.A. den Broeder: Phys. Rev. Lett. 68, 682
(1992)
29. D.D. Koelling, B.N. Harmon: J. Phys. C 10, 3107 (1977)
30. A.K Rajagopal, J. Callaway: Phys. Rev. B 7, 1912 (1973)
31. A.K Rajagopal: A Density Functional Formalism for Condensed Matter Sys-
tems. In: Density Functional Methods in Physics. ed. by R.M. DreizIer and J.
da Providencia (Plenum, New York 1985) pp. 159
32. A.H. MacDonald: J. Phys. C 16, 3869 (1983)
33. H. Eschrig, G. Seifert, P. Ziesche: Solid State Commun. 56, 777 (1985)
34. H.J.F. Jansen: Phys. Rev. B 38, 8022 (1988)
35. H.J.F. Jansen: Phys. Rev. B 59, 4699 (1999)
36. M. Levy: Proc. Natl. Acad. Sci. USA 76, 6062 (1979)
37. M. Levy: Phys. Rev. A 26, 1200 (1982)
38. G. Breit: Phys. Rev. 34, 553 (1929)
39. G. Breit: Phys. Rev. 39, 616 (1932)
40. H.A. Bethe, E.E. Salpeter: Quantum Mechanics of One- and Two-Electron
Systems (Plenum, New York 1977)
41. A.O. Barut, B.W. Xu: Ann. Phys. (ny) 148, 135 (1983)
42. R.M. Bozorth, L.W. McKeehan: Phys. Rev. 51, 216 91937)
43. J. Sticht, KH. Höck, J. Kübler: J. Phys. Condens. Matter 1, 8155 (1989)
44. M. Uhl, L.M. Sandratskii, J. Kübler: Phys. Rev. B 50, 291 (1994)
45. L. Fritsche, J. Noflke, H. Eckardt: J. Phys. F 17, 943 (1987)
46. P. Strange et al.: J. Phys. Condensed Matter 1, 3947 (1989)
47. G.H.O. Daalderop, P.J. Kelly, M.F.H. Schuurmans: Phys. Rev. B 41, 11919
(1990)
48. G.Y. Guo, W.M. Temmerman, H. Ebert: Physica B 172, 61 (1991)

49. J. Trygg et al.: Phys. Rev. Lett. 75, 2871 (1995)
50. S.S. Razee, J.B. Staunton: Phys. Rev. B 56, 8082 (1997)
51. S.V. Halilov et al.: Phys. Rev. B 57, 9557 (1998)
52. S.V. Beiden et al.: Phys. Rev. B 57, 14247 (1998)
53. K. Hellwege, A.M. Hellwege (editors): Landolt-Börnstein, New Series Vol.
iii/19a (Springer, Berlin Heidelberg New York 1987)
54. W.A. Harrison: Soild State Theory (Dover, New York 1979)
55. J.C. Slater, G.F. Koster: Phys. Rev. 94, 1111 (1954)
56. D.A. Papaconstantopoulos: Handbook 0/ the Band Structure 0/ Elemental
Solids (Plenum, New York 1986)
57. M. Weinert, R.E. Watson, J.W. Davenport: Phys. Rev. B 32, 2115 (1985)
58. J. Rath, A.J. Freeman: Phys. Rev. B 11, 2109 (1975)
59. P.E. Blöchl, O. Jepsen, O.K. Andersen: Phys. Rev. B 49, 16223 (1994)
60. P.J. Davis, P. Rabinowitz: Methods 0/ Numerical integration (Academic Press,
New York 1975)
61. N.W. Ashcroft, N.D. Mermin: Solid State Physics (Sounders College, Fort
Worth 1976)
62. R.J. Needs, R.M. Martin, O.H. Nielsen: Phys. Rev. B 33, 3778 (1986)
63. M.S.S. Brooks: Physica 130B, 6 (1985)
64. O. Eriksson, M.S.S. Brooks, B. Johansson: Phys. Rev. B 41, 7311 (1990)
65. G. Schneider, H.J.F. Jansen: J. Appl. Physics 87 5875 (2000)
66. H. Wang: Relation between Bandstructure and Magnetocrystalline Anisotropy:
Iron and Nickel. MS thesis, Oregon State University (2000) published at
http://vww.physics.orst.edu/henri/Jansen.html
67. D. Singh, W. Pickett, H. Krakauer: Phys. Rev. B 43,11628 (1991)
of Correlated Systems: Beyond LDA
A.I. Lichtenstein, V.I. Anisimov, and M.I. Katsnelson
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.
filled d (or f) shell. When applying an effective one-electron method, with

an orbital-independent potential, to transition-metal compounds, it results
in a partially filled d band with a metallic-type electronic structure and
itinerant d electrons. This is definitely not a correct picture for the insulating
antiferromagnetic transition-metal oxides and rare-earth metal compounds
with welllocalized d (f) electrons and a sizable energy separation between
occupied and unoccupied subbands such as the lower and upper Hubbard
bands in a model Hamiltonian approach [1,3].
There were several attempts to improve the LDA scheme that take into
account strong electron-electron correlations. One of the most popular ap-
proachs is the self interaction correction (SIC) method [11]. It reproduces
quite well the localized nature of the d (or f) electrons in transition-metal
(rare-earth metal) compounds, but one-electron energies calculated in the SIC
approximation are usually in disagreement with spectroscopic data (for ex-
ample, for transition-metal oxides the calculated occupied d-bands are about
10 eV below the oxygen valence band). The Hartree-Fock (HF) method [12]
is appropriate to describe the Mott insulators because it contains explicitly
a term that cancels the self-interaction. The problem of self-interaction is the
main reason why the LDA spectra for magnetic oxides are in such a strong
qualitative disagreement with experimental data. However, a serious problem
of the Hartree-Fock approximation is the unscreened nature of the Coulomb
interaction. The "bare" value of the Coulomb interaction parameter U is
rat her large for transition metal atoms (15-20eV) while the screening in
a solid leads to much smaller values of U (5-8eV) [13,14]. Due to the absence
of screening effects, the HF energy gap values are a factor of 2-3 larger than
the experimental values [12].
The problem of screening is addressed in a rigorous way in the GW ap-
proximation [15,16], which may be regarded as a Hartree-Fock theory with
a frequency and orbital-dependent screened Coulomb interaction. The GW
scheme has been applied with success to real systems ranging from simple
met als to transition metals, but applications to more complex systems are
quite rare up to now due to the large computational efforts [17]. Another
problem of using the GW approximation is that in its practical realization
[18] a response function, needed to calculate the screened interaction, is
computed with the help of the energy bands and wave functions obtained
in an LDA calculation. While such a procedure is justified for the systems
where correlation effects are small (such as semiconductors [19]), for strongly
correlated systems one may need a better starting Hamiltonian than the LDA.
This, for example, can be achieved by improving the LDA Hamiltonian using
the calculated self-energy in a self-consistent procedure [20].
In this chapter, the so-called LDA+U method [8,21-23] and the general-
ization to the dynamical LDA+DMFT method [24,25] for electronic struc-
tures of correlated magnetic systems are described, where the nonlocal and
3 Electronic Structure and Magnetism of Correlated Systems 103
energy-dependent self-energy is approximated by a moment um-independent

but nonlocal orbital and frequency dependent screened Coulomb potential.
3.2 The LDA+U Method

In these approaches, similar to the Anderson model [26], electrons are sepa-
rated into two subsystems: localized d or f electrons for which the intraatomic
Coulomb interaction should be taken into account, and delocalized s, p elec-
trons that could be described by using orbital-independent one-eleetron po-
tential (LDA). Let us consider a d ion as an open system with a fluctuating
number of d electrons describing the Coulomb interaction by the simplest ex-
pression ~U Li#j ninj (ni are d orbital occupancies). Using the Hartree-Fock
approximation we have for the energy ofthe interaction: E = UN(N -1)/2
where N = L ni is a total number of d electrons per site given by the LDA.
Let us subtraet this expression from the LDA total energy functional and add
a Hubbard-like term (negleeting for a while exchange and nonsphericity). As
a result, we would have the following functional:
1
E = ELDA - UN(N -1)/2 + 2ULninj. (3.1)
i#j
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 '
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,
EU[{n}] = ~ L {(m,m" I V ee I m',m"')n~m,n;;'':,m'"

{m},a
+((m,m" I Vee I m',m"')
- (m, m " I v.:ee I m '" ,m')) nmm,nm"m'"
a a } , (3.5)
where Vee is the screened Coulomb interactions among the nl electrons. Fi-
nally, the last term in (3.4) describes the correction for the double counting
and is given by
where Na = Tt(n~m') and N = Nt + N+.U and J are screened Coulomb

and exchange parameters [13,14].
In addition to the usual LDA potential, we have a nonlocal in the orbital-
space contribution to the effective single-particle Hamiltonian:
H = HLSDA + L I inlma) V~m' (inlm' a I , (3.7)

mm'
a
Vmm' L {( m, m " I v.:ee I m,' m ",)-a
= """ nm"m'"
rn',m"
+((m,m" I Vee I m',m"') - (m,m" I Vee I m"',m')n~"m"'}
(3.8)
The Vees remain to be determined. We again follow the spirit of the

LDA + U approach by assuming that within the atomic spheres these inter-
actions largely retain their atomic nature. Moreover, the values of screened
Coulomb (U) and exchange (J) interactions can be calculated within the
supercell LSDA approach [13]: the elements of the density matrix n':nm' have
to be constrained locally and the second derivative of the LSDA energy with
respect to the variation of the density matrix yields the wanted interactions.
In a spherical approximation, the matrix elements of Vee can be expressed in

terms of effective Slater integrals F k [27] as
(m,m" I Vee I m',m"') = Lak(m,m',m",m"')F k , (3.9)

k
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].
3.3 Dynamical Mean-Field Theory
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
idea of neglecting the spatial fluctuations, but including quantum frequency

fluctuations in the effective functional.
3.3.1 LDA+DMFT: General Considerations

The dynamical mean-field theory (for a review, see [5]) is based on the
mapping of lattice models onto quantum impurity models subject to a self-
consistency condition. The resulting multiorbital impurity model can be
solved by various rigorous approaches (quantum Monte Carlo, exact diago-
nalization) or by approximated schemes such as iterated perturbation theory
(IPT), the local fluctuating-exchange (FLEX) approximation, or the non-
crossing approximation (NCA).
In this section we describe the LDA+DMFT approach for the electronic
structure calculations. The method was first applied to Lal-xSrx Ti0 3 [24],
which is a classical example of a strongly correlated metal. A general formu-
lation of LDA+DMFT, including the justification of the effective impurity
formulation in the multiband case has been discussed in [25].
In order to implement the achievements of the DMFT model theory to the
LDA one needs a first-principle tight-binding model. The linearized muflin-
tin orbitals (LMTO-TB) method in the orthogonal representation [34,36] is
an ideal scheme for the LDA+DMFT approach and can be presented in the
tight-binding form:
HLMTO =
ilm,jl'm',a
(i - site index, Im - orbital indexes).

As we have mentioned above, the LDA one-electron potential is orbital
independent and Coulomb interaction between d electrons is taken into ac-
count in this scheme in an averaged way. We generalize this Hamiltonian for
the explicit local Coulomb correlations with the additional interaction term
for the correlated il shell:
(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
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
Um m' = < mm'IVee(r - r')lmm' >,

Jmm, = < mm'JVee(r - r')lm'm > .
Then, a new Hamiltonian will have the following form:
H = HO + H int ,
HO = L (Oilm,jllm ' E?l nilma- + tilm,jllmlcltm"Cjllmla-)' (3.15)
ilm,jl'm',u
In the reciprocal space matrix elements of the operator HO are
(3.16)
(q is an index of the atom in the elementary unit cell).

In the local, frequency-dependent dynamical mean-field theory the effect
of Coulomb correlation is described by the self-energy operator E(iw). The
inverse Green function matrix is defined as
G;/;n,qll'm/(k, iw) = iw + JL - H~lm,qll'm/(k)

-Oql,qll'Oql,idl d Em,ml (iw) , (3.17)
where JL is chemical potential, and the local Green function obtained via
integration over the Brillouin zone is:
Gqlm,qll'm/(iw) = ~B J dkGqlm,qll'm/(k,iw) , (3.18)
(VB is a volume of the Brillouin zone).

If we consider for simplicity a case of high enough symmetry (e.g. cubic
for d electrons), then the self-energy E(iw) is diagonal in orbital and spin
indices. A so-called bath Green function that defines the effective Andersen
model and preserves the double-counting of the local self-energy is obtained
as a solution of the effective impurity problem [5]:
(3.19)
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
In principle, we could use the ji parameter for the double-counted correc-

tion. Further , one has to find the self-energy E m (iw) in terms of the bath
Green function Gom(iw) and use it in the self-consistent LDA+DMFT loop
(3.18, 3.19).
3.3.2 DMFT in Quantum Monte Carlo Approach

Here we first describe the most rigorous way to solve an effective impurity
problem using the multiband quantum Monte Carlo (QMC) method [38].
In the framework of the LDA+DMFT approach it was used first in [39] for
the case of ferromagnetic iron. In this method, the local Green function is
calculated for the imaginary time interval [O,ß] with the mesh Tl = l.c:1T,
l = 0, ... , L - 1, and .c:1T = ß / L (ß = ~ is the inverse temperature ) using the
path-integral formalism [5]:
G~ = ~ 2: det[O(s)] * G~ (s), (3.21)

sI
mm'
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)
and the discrete Hubbard-Stratonovich parameters are given by the expres-

sion, Amm' = arccosh[exp(~.c:1TUmm')] [40]. Using the output local Green
function from QMC and input bath Green functions, the new self-energy is
obtained via (3.19) and the self-consistent loop can be closed through (3.18).
The main problem of the multiband QMC formalism is the large number of
the auxiliary fields s!nm'. For each time slice I it is equal to M(2M - 1),
where M is the total number of the orbitals which gives 45 Ising fields for
the d states case and 91 fields for the f states. Analytical continuations of the
QMC Green functions from the imaginary time to the real energy axis can
be done within the maximum entropy method [41].
3.3.3 DMFT in Multiband Fluctuation-Exchange Approximation

The QMC method described above is probably the most rigorous real way
to solve an effective impurity problem in the framework of DMFT theory.
However, it is rather time consuming. In addition, in the previous section we
did not work with complete four-indices Coulomb matrix:
(12 lvi 34) = J drdr''l/Ji (r)'l/J2 (r')v scr (r - r') 'l/J3(r)'l/J4(r'), (3.22)
where we define for simplicity ml == 1, etc.

For moderately strong correlations one can use an approximate scheme,
which is based on the fluctuation exchange (FLEX) approximation of Bickers
and Scalapino [42] and generalized to the multiband spin-polarized case in
[25,37,43]. The electronic self-energy in the FLEX approach is equal to:
E = E HF + E(2) + E(ph) + E(pp) , (3.23)
where the Hartree-Fock contribution has a standard form:
E~~(7 = I: [(13 lvi 24) I: n~~ - (13IvI42) n~4] , (3.24)

34 (7'
with the occupation matrix n'{2 = G 21 (r --+ -0); this contribution to E is

equivalent to a spin-polarized "rotationally invariant" LDA+U method [23].
The second-order contribution in the spin-polarized case reads:
Eg~u(r) = - I: (13IvI74) G7S (r)

{3-S}
x [(85 lvi 26) I: G6~(r)G4~( -r)

(7'
-(85IvI62)G63(r)G45(-r)] , (3.25)
and the higher-order particle-hole (or particle-particle) contribution
E~~~ (r) = I: Wf3~~2 (r) G~~ (r) , (3.26)

34,(7'
with p-h (p-p) fluctuation potential matrix:
(7(7'. _ [Wtt (iw) WH (iW)] (3.27)

W (zw) - WH (iw) W-I--I- (iw) ,
where the spin-dependent effective potentials have a generalized RPA-form

and can be found in [37]. Note that for both p-h and p-p channels the effective
interactions, according to (3.27), are nondiagonal matrices in spin space as
in the QMC scheme, in contrast to any mean-field approximation like the
LSDA.
We could further reduce the computational procedure by neglecting the

dynamical interaction in the p-p channel since the most important fluctua-
tions in itinerant electron magnets are spin-fluctuations in the p-h channel.
We take into account static (of T-matrix type) renormalization of efIective
interactions replacing the bare matrix U12 ,34 = (12 lvi 34) in the FLEX equa-
tions with the corresponding spin-dependent scattering T-matrix
(12ITCTCT'134) = (12 lvi 34) (3.28)
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.
3.3.4 DMFT in Iterated Perturbation Theory

Here, we consider a simplified version of the iterated perturbation theory
(IPT) when the Coulomb inter action is described by only one parameter U
and the self-energy E does not depend on the orbital indices [24] (a similar
scheme for a multi band case can be found in [25]). The expression for the
self-energy in terms of the bath Green function GO corresponding to the
second-order perturbation theory reads
EO(iws ) = -(N - 1)u2;2

XL L CO (iwm + iPn) CO (iwm)CO (iw s - iPn) , (3.29)
iWn ipm
where N is a degeneracy of orbitals including spin, and Ws, Pn are Matsubara

frequencies. In IPT the expression for EO is renormalized to insure a correct
atomic limit:
. AEO(iw)
E(~w) = UN(N - 1) + 1- BEO(iw) , (3.30)
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) ,
no = ~ ?= eiwnO+ GO (iwn ) . (3.33)

'Wn
n is the orbital occupation number, and the correlation function D[n] ==

< nn >OPA is calculated using the coherent potential approximation (CPA)
for the Green function with parameter 8/.L chosen to preserve n:
G (i) _ [1 - n(N - 1)] n(N - 1)

(3.34)
OPA W - iW+/.L-L1(iw)+8/.L + iW+/.L-L1(iw)-U+8/.L'
1 ~ iWnO+G
n = ß L...J e (. )
OPA ZW n , (3.35)
'Wn
(3.36)
Here L1(iw) is the "hybridization function" of the effective Anderson model

(see [5]). As weH as in the FLEX method described above the fast fourier
transform algorithm can be used for the transition from frequency variables
to the time ones and vice versa.
A serious problem is to perform integration in k-space over the Brillouin
zone. For this a generalized Lambin-Vigneron algorithm [45] can be used.
The real-axis equivalent of (3.29) is much more complicated and difficult to
implement numerically than the Matsubara frequencies version. It is much
more convenient to performanalytical continuation from imaginary energy
values to the real ones. For this continuation a Pade approximant algorithm
[46] was used. We have found that the most convenient way is to use analytical
continuation not for the Green function G but only for self-energy E, and then
to calculate G directly on the real axis through the Brillouin zone integration.
3.3.5 DMFT in Noncrossing Approximations

Another approximate scheme for the solution of the effective impurity prob-
lem, which is often used in the framework of the LDA+DMFT approach, is
the noncrossing approximation (NCA) [47]. Here, we present only the basic
idea of the method. Details can be found in [48].
The NCA is based on the perturbation theory starting from the limit
of strang intraatomic interaction. In this case, it is suitable to rewrite the
Hamiltonian of interaction in the representation of exact atomic many-body
states
(3.37)
which corresponds to the Hubbard X-operator method [49J. The one-electron

creation and annihilation operators can be written in the same basis as
c;t,u = L D73:* la) (ßI . (3.38)

a,ß
Then, the Green function of the Hubbard X-operators xaa = 10:)(0:1 is

considered. The first-order approximation in the band-energy Hamiltonian
would be equivalent to the well-known Hubbard-I approximation [50]. The
corresponding self-energy in the NCA is calculated using the second-order
perturbation theory in the hybridization between the localized and itinerant
electrons with the consequent replacement of the bare Green functions by
the exact ones. This approximation can be formally justified using the large
orbital degeneracy limit but the quantitative accuracy in real situations is
not quite. Usually, the NCA gives a correct qualitative description of such
many-body effects such as the Kondo resonance [47].
3.3.6 Cluster LDA+DMFT Scheme

When considering effects like charge-ordering or d-wave superconductivity,
which explicitly involve the electronic correlations on different sites, a cluster
generalization of the LDA+DMFT scheme would be necessary. The most
natural way to construct this generalization is to consider the cluster as a
"supersite" in an effective medium (for simplicity, consider here the case of
two-site cluster). Then, the crystal supercell Green nmction matrix can be
written as
G (k, iw) = [(iw + J.L) * 1 - h (k) - ~ (iw) ]-1 ,

where haß (k) is the effective hopping matrix, E aß (iw) is the self-energy
matrix of the N-site supercell dimension, which is assumed to be local, Le.
k-independent, and J.L is the chemical potential.
In the cluster version of the DMFT scheme, one can write the matrix
equation for a bath Green function matrix g, which describes effective inter-
actions with the rest of the crystal:
g-1 (iw) = G- 1 (iw) + ~ (iw) , (3.39)
with the local cluster Green function matrix equal to
Gaß (iw) = :EGaß (k,iw) ,

k
and the summation runningover the Brillouin zone of the lattice.

To demonstrate one of the most efficient ways to solve the impurity prob-
lem, we consider an exact diagonalisation (ED) scheme [5] to solve the cluster
DMFT problem. In this case, the lattice Hamiltonian is mapped onto a finite
cluster impurity model:
(3.40)
i,j,a i<j,t7
k,i,j,u k,i,j,CT
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
In the paramagnetic state we can transform 9, T, rk, and Ek matrices ofthe

dimension 2 x 2 for the two-site cluster to the diagonal bonding-antibonding
basis A = {b, a}. In this case TA = {-tl., tl.} and 9A = {9b,9a}, where
9b,a = (911 ± 912)/2.
The parameters {EkA17' rkA17 } are now fitted to reproduce the bath Green
function 9A17(iw,,) for bonding and anti-bonding states independently. The
next step is the solution of the cluster-impurity problem (3.41) to get the
cluster self-energies E ij17 (iw,,) that are required for the next DMFT iterations.
After solving the effective cluster problem the local Green's function matrix
Gij17 (iw,,) was determined and the cluster self-energy E ij17 (iw,,) was obtained
within the cluster-impurity scheme (3.39).
The scheme described above corresponds to the "free cluster" approach.
Alternatively, periodic modification of the DMFT in the k-space (so-called
"dynamical cluster approximation" [51]) or "periodic cluster" method in the
real space [52J can be used.
3.4 Exchange Interactions
A useful scheme for analyses of exchange interactions in the LSDA approach is

a so-called "local force theorem". In this case, the calculation of small total
energy changes reduces to variations of the one-particle density of states
[53,54J. First, it is useful to have an analog of the local force theorem in
the LDA+DMFT approach. In contrast to the standard density functional
theory, this deals with the real dynamical quasiparticles defined via Green
functions for the correlated electrons rather than with Kohn-Sham "quasi-
particles", which are, strictly speaking, only auxiliary states for the total
energy calculations. Therefore, instead of working with the thermodynamic
potential as a density functional we have to start from the general expression
for n in terms of the exact Green function [55J. One can prove [39J that the
small variation of the n-potential is equal to
8n = -8*Trln [_G- 1 ] , (3.42)
where 15* is the variation at a fixed "self-consistent" energy-dependent poten-

tial (E). This is a desired analog of the "local force theorem" in the density
functional theory [54].
Further considerations are similar to the corresponding ones in the LSDA
approach. In the LDA+DMFT scheme, the self-energy is local, i.e. is diagonal
in site indices. Let us write the spin-matrix structure of the self-energy and
Green function in the following form
where
E;c,s) = ~ (EJ ± Et), EI = Etei,

with ei being the unit vector in the direction of the effective spin-dependent
potential on site i, (T = (O"x,O"y,O"z) are Pauli matrices, Gij = ~Tra(Gij) and
Gij = ~Tra(Gij(T). We assume that the bare Green function GO does not
depend on spin directions and all the spin-dependent terms including the
Hartree-Fock terms are incorporated in the self-energy. Spin excitations with
low energies are connected with the rotations of vectors ei:
(3.43)
According to the "local force theorem" (3.42) the corresponding variation of

the thermodynamic potential can be written as J[l = 15* [lsp = ViJ<pi, where
the torque is equal to
(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]
[lspin =- L TrwL { (GkEj) (Gji En

ij
-EfGijEjG'ji - i (Ef x GijEj) Gji} (3.45)
One can show by direct calculations that
[
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.
In the nonrelativistic case one can rewrite the spin Hamiltonian for small
spin deviations away from collinear magnetic structures in the following form
ilspin = - L Jijeiej, (3.47)

ij
where
Jij = -TrwL (EtGIjEJG}i) , (3.48)
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)
(q --+ 0) in terms of exchange parameters has to be exact as the consequence

of phenomenological Landau-Lifshitz equations that are definitely correct in
the long-wavelength limit. Direct calculation based on variation of the total
energy under spiral spin-rotations (see [54]) leads to the following expression
D = -~Tr '"' (E s8Gt (k) Es8G~ (k)) (3.51)

aß M wL L....t
k
8k
0.
8k
ß
'
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}
where the spin-dependent potentials I are expressed in terms of the potentials

of (3.8),
(3.53)
while the effective intersublattice susceptibilities (i and j) are defined in terms

of the LDA + U eigenfunctions 'l/J as,
(3.54)
We will now review the applications of these new methods and computa-
tional schemes to real correlated materials.
3.5 Ferromagnetic Metals
3.5.1 Localized States: Rare-Earth Metals

Gd metal is a good test to check how physically sound are the approximations
that were used in the derivation of the LDA+U approach. The Gd3+ ion has
seven f electrons so that the majority-spin subshell is completely filled and the
minority-spin subshell is empty. The hybridization of the localized f orbitals
with the conduction bands is small and the 4f shell could be regarded, with
good accuracy, as that of the Gd3+ ion in 887 ground state, weIl separated
2
from all the other excited terms. This means that the ground state is weIl
described by a single Slater determinant wave function and the LDA+U as
a one-electron theory is valid here. The final N - 1 and N + 1 states of the
removal (XPS) and addition (BIS) spectra are also weIl described by a single
e
Slater determinant F, neglecting spin-orbit coupling, which gives splitting
of the order of less than 1 eV). As a result, the theoretical XPS and BIS are
(again neglecting spin-orbit coupling) single lines separated by U + 6J. The
calculation for Gd [57] gives the values of the parameters U = 6.7 eV and
J = 0.7 eV thus resulting in the theoretical value of the splitting between
occupied and unoccupied 4f bands, which is about 11 eV, in good agreement
with the experimental value, which is approximately 12 eV (Fig.3.1).
The advantage of the LDA+U method is the ability to treat simultane-
ously delocalized conduction band electrons and localized 4f electrons in the
same computational scheme. For such a method it is important to be sure
that the relative energy positions of these two types of bands are reproduced
correctly. The example of Gd gives us confidence in this (Fig.3.1): there is
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].
3.5.2 Delocalized States: Transition Metals

The description of correlation effects in electronic structure and magnetism
of iron-group metals is still far from the final picture and attracts continuous
interest (see, e.g. [58-61] and references therein). Despite many attempts, the
situation is still undear both theoretically and experimentally. For example,
there is no agreement on the presence of 5 eV satellite in the photoemission
spectrum of iron [62,63], and on the existence of local spin splitting above
3 Electranic Structure and Magnetism of Correlated Systems 119
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
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)
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
r Wave veclor (001) H

Fig.3.4. The spin-wave spectrum for ferromagnetic iron in the LSDA and LDA+E
approximations compared with different experiments (circles [71], squares [72], and
diamonds [73])
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].
3.6 Mott-Hubbard Insulators

3.6.1 Electronic Structure of Thansition-Metal Oxides
The advantages of the simultaneous treatment of the localized and delocal-
ized electrons in the LDA+U method and especially in the LDA+DMFT
approaches are seen most clearly for the transition met al compounds, where
3d electrons, while remaining localized, hybridize quite strongly with other
orbitals. Late transition-metal oxides, for which LSDA results strongly under-
estimate the energy gap and magnetic moment values (or even give qualita-
tively wrong metallic ground state, as for the insulators CoO and CaCu02),
are weIl described by the LDA+U [21].
As was already mentioned, sometimes it is necessary to take into account
also the intraatomic Coulomb interaction on the oxygen sites to achieve
a satisfactory agreement with the experimental data [32]. Here we present the
results of such calculations for transition-met al oxides [74]. The important
part of the LDA + U calculation scheme is the determination of Coulomb
interaction parameters U and J in (3.8): the Coulomb parameter Up for the
p orbitals of oxygen, Ud for the transition-metal ion and the Hund's parameter
J for the d orbitals of transition-metals. To get Ud and J one can use the
supercell procedure [14,75] or the constrained LSDA method [13], which are
based On calculation of the variation of the total energy as a function of the
local occupation of the d shell. We took the values of Coulomb parameters
(Ud rv 7 - 8eV and J rv 0.9 - 1eV) from the previous LDA+U calculation
[21]. The problem is how to determine the Coulomb parameter Up .
Due to the more extended nature of the 0(2p) Wannier states in compar-
ison with transition metal d states, the constrained occupation calculations
can not be implemented as easily as for the d shell of transition metals.

Nevertheless, several independent and different techniques were used for this
purpose previously by different authors. McMahan et al. [30] estimated the
value of Up in high-Tc related compound La2Cu04 using the constrained
LDA calculation where only atomic-like O(2p) orbitals within oxygen atomic
spheres were considered, instead of the more extended Wannier functions. The
corresponding value of the Coulomb interaction parameter Up was obtained
as 7.3eV. This value can be considered as the upper limit of the exact Up .
The LDA calculations gave the estimation that only 75% of the Wannier
function density lies in the oxygen atomic sphere, so that the renormalized
value of Coulomb interaction parameter for the oxygen Wannier functions is
Up = (7.3) X (0.75)2 = 4.1 eV [30].
Later, Hybertsen et al. [31] suggested the scheme to calculate Up , which
consists of two steps: (i) via a constrained-density-functional approach one
can obtain the energy surface E(Nd , Np) as a function of local charge states
and (ii) simultaneously the extended Hubbard model was solved in a mean-
field approximation as a function of local charge states Nd and Np. The
corresponding Coulomb interaction parameters are extracted as those that
give the energy surface matching the microscopic density-functional calcula-
tion results [31]. The obtained values for Up are 3 - 8eV depending on the
parameters of the calculations.
Another way to estimate Up is to use Auger spectroscopy data, where
two holes in the O(2p) shell are created in the excitation process. Such
fitting to the experimental spectra gave the value of Up = 5.geV [77]. In the
LDA+U(d+p) calculations [74] the value Up = 6eV was used. The solution of
the LDA+U problem has been done in the self-consistent tight-binding (TB)
linear muffin-tin orbitals method (LMTO) in the atomic sphere approxima-
tion (ASA) [78-80].
Comparison between the LDA+U (left column) and the LDA+U(d+p)
(right column) calculated density of states (DOS) of NiO, MnO, and La2Cu04
is presented in Figs.3.5-3.7. For all compounds, one can see that the main
difference between the LDA+U(d+p) and the LDA+U calculated densities of
states is the increased energy separation between the oxygen 2p and transition
metal3d bands. The larger value of "charge transfer" energy (O(2p)-Me(3d))
(Me = Ni, Mn, Cu) leads to the enhanced ionic and decreased covalent nature
in the electronic structure: the unoccupied bands have more pronounced 3d
character and the admixture of oxygen states to those bands becomes weaker.
The ground state is correctly described both by LDA+U and LDA+U(d+p)
calculations as antiferromagnetic insulator for all compounds. The values of
energy gaps [81] and spin magnetic moments are presented in Table 3.1 (see
the discussion of experimental data in [74]). One can see that the values
obtained in the LDA+U(d+p) calculations are, in general, in bett er agreement
with experiment than the LDA + U calculated values. While the increasing of
the energy gap values with applying Up correction was obviously expected
;;-
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)
with the increasing of "charge transfer" energy in the compounds belonging

to the dass of "charge transfer" insulators [82], the increasing of the magnetic
moments values is a more complicated self-consistency effect due to the in-
creased ionicity in the LDA+U(d+p) calculations compared with the LDA+U
results.
In Fig. 3.8, the DOS obtained by LDA+U(d+p) and LDA+U calculations
for MnO and NiO compounds are compared with the superimposed XPS and
BIS spectra corresponding to the removal of an electron (the occupied bands)
and addition of an electron (the empty bands) , respectively. The better agree-
ment with the experimental data for the position of the main peaks of the
unoccupied band relative to the occupied one is the direct confirmation of
the importance of taking into account Coulomb interactions in the oxygen
2p shell.
It is instructive to compare the results of the LDA+U calculations for
NiO with the first-principles "Hubbard I" approach [25J. First, to describe
Mott insulators in the LDA+U approach (as wen as in the SIC approach)
it is necessary to assume magnetic and (or) orbitallong-range order [9]. In
LDA+DMFT it is possible to consider the paramagnetic Mott insulators in
the framework of ab initio calculations. Moreover, it is possible to obtain not
only the Mott-Hubbard gap in the electron spectrum but also satellites and
10
8 tolal lotal
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
Fig. 3.7. NiO DOS calculated by the LDA+U (left column) and the LDA+U(d+p)
(right column) methods [74J
multiplet structure. The following effective Slater parameters, which define

screened Coulomb interaction in the d shell for NiO, have been used: FO =
8.0eV, F 2 = 8.2eV, F 4 = 5.2eV [21]. We have started from the nonmagnetic
LDA calculations in the LMTO nearly orthogonal representation [34] for
experimental crystal structures of NiO. The minimal basis set of S,p, and d
orbitals for NiO corresponds to 18x18 matrix of the LDA Hamiltonian h(k).
The occupation number for correlated electrons are 8.4 electrons in the d shell
Table 3.1. Calculated and experimental values of energy gaps (eV), spin magnetic
moments (J1-B) and intersite exchange interaction parameters J ex (meV)
LDA+U LDA+U(d+p) Experiment

Eg 0.7 2.0 2.0
La2Cu04 J1-B 0.45 0.68 0.60
Jex -82.9 -100.9 -136
Eg 3.8 4.5 3.6,3.8
MnO J1-B 4.51 4.59 4.58,4.79
J;x 5.4 -5.4 -4.8, -5.4
J;x -9.3 -5.1 -5.6, -5.9
Eg 1.8 2.8 4.0,4.3
NiO J1-B 1.50 1.64 1.64,1.77
J1x -0.8 -0.2 -1.4
J;x -23.2 -19.4 -19.0
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
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
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.
3.6.2 Exchange Interactions in Transition-Meta} Oxides
The values of the intersite exchange interaction parameters J e x depend on the

parameters of the electronic structure in a rather indirect implicit way. The
development of a good scheme for calculating exchange parameters is very
important because ab initio calculations are often the only way to describe
the magnetic properties of complicated compounds such as, for example,
"spin-gap" systems [84]. Recently Solovyev and Terakura [85] did a very
thorough analysis of the exchange interaction parameters for MnO calculated
using different methods of electronic structure calculations. They used the
positions of the Mn(3d) spin-up and Mn(3d) spin-down bands relative to
the oxygen 2p states as adjustable parameters to fit the values of exchange
interaction for the nearest and second Mn-Mn neighbors. Their results gave
nearly the same splitting between Mn(3d) spin-up and Mn(3d) spin-down
states as in standard LDA+U calculations (10.6 eV) but the position of those
states relative to the oxygen band was shifted approximately 3 eV up relative
to the LDA+U case. It is practically the same as we have in our LDA+U(d+p)
calculations, because with Up = 6eV the shift of the position Me(3d) band
relative to the oxygen O(2p)-band is equal to Up /2 = 3eV.
Comparison between LDA+U and LDA+U(d+p) calculated J ex parame-

ters and experimental data is presented in Table 3.1. J ex values were calcu-
lated from the Green function method as second derivatives of the ground
state energy with respect to the magnetic moment rotation angle [23,39,54]
as was described above. Again one can see that, in general, the LDA+U(d+p)
gives better results than the LDA+U, especially for the MnO compound.
3.6.3 Electronic Structure and Exchange Interactions in V 203
Let us discuss in more detail an interesting example of the insulating phase of

a metal-oxide system - V 203 in the monoclinic and the corundum structures,
following [86]. The V 2 0 3 system has been a topic of intense study for more
than fifty years by both theoreticians and experimentalists because of its
"rich" phase diagram. It undergoes a first-order metal-insulator transition
with a seven orders of magnitude change in the electrical conductivity [87],
which can be induced by temperature, pressure, alloying or by nonstoichiome-
try. It also exhibits an antiferromagnetic inslulator to paramagnetic insulator
transition, which is also of first-order and a first-order paramagnetic insula-
tor to paramagnetic metal transition. The calculated exchange interaction
parameters lead to a magnetic structure consistent with experiment again
without the need for orbital ordering. While the low-temperature monoclinic
distortion of the corundum crystal structure pro duces a very small effect
on the electronic structure of V 203, the change of magnetic order leads to
drastic differences in band widths and band gaps. The low-temperature mon-
oclinic phase clearly favors the experimentally observed magnetic structure,
but the calculations for the corundum crystal structure give two consistent
sets of exchange inter action parameters with nearly degenerate total energies
suggesting a kind of frustration in the paramagnetic phase. These results
strongly suggest that the phase transitions in V 203 which is so often quoted
as the example of a S = ~ Mott Hubbard system have a different origin.
The crystal structure in the low temperature AFI phase is monoclinic
[91], above Tc this changes to the corundum structure (Tc ~ 150K [92]). The
calculated densities of states for both structures are very similar, showing
that the electronic structure itself is little influenced by the lattice distortions,
so that the strong change in properties must be a rat her subtle effect with
respect to the electronic structure. In Fig. 3.10, the partial densities of states
(DOS) obtained in the LDA calculation for the monoclinic crystal structure
of V 2 0 3 are shown. They are very similar to those found by Mattheiss [93]
for the corundum lattice. As the monoclinic distortion of the lattice is not
very strong, we can plot the DOS assuming approximate trigonal symmetry.
The V ions are somewhat off-center in a slightly trigonally distorted octahe-
dron of 0 ions. This distortion causes the otherwise 3-fold degenerate t2g 3d
orbitals to split into a nondegenerate A 1g and double degenerate E g levels.
In this representation the A 1g orbital has 3z 2 - r 2 symmetry in a hexagonal
VP3 (LDA)
Monoclinie Slructure
- A " (occupalion 0.5)

E, (occupation 0.81)
--- -- cubic e,
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
VP3 (LDA+U, U=2.8, J=O.93)
2.00 r---...,....-----,----.---.---...,....---,-----,
(a)
1.00
c.~
E
~
1
(J)
8 -1.00
- A., ("",,"pation 0.26)

E, ("",,"pation 0.96)
----- cubiceu
-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
and resulting in a metallic state. The A lg band, although it shows some

structure that might be interpreted as a bonding-antibonding pseudogap,
actually exhibits a strong peaking above E F and only relatively little weight
below E F . This looks very different from a bonding-antibonding splitting,
which should have exhibited a more symmetrie structure about the center
of gravity if dimer hoping integrals were dominating the problem. We also
note that the total A lg band width is only 2.5 eV, which is considerably
sm aller than the expected value of the Hubbard U of about 3 eV including
the screening due to the strongly bonding eg electrons [94] or about 4-5 eV
without this screening channel [95]. Such a small band width would invalidate
a molecular-orbital like approach also for the A lg orbitals. So even these
results already cast doubt on the validity of the most commonly used starting
point with a strong bonding-antibonding splitting of the A lg orbitals, with
the E g orbitals in this gap ending with two spin-antiparallel electrons per
c-axis pair in the A lg bonding state leaving only one electron in an assumed
narrow doubly degenerate E g band [88]. It was this starting point that led
to the now very much used one electron per site Hubbard model for V 2 0 3 •
The above LDA results give us rough ideas that should be further checked.
The LDA+U calculations can provide us with some more details about the
electronic structure OfV 2 0 3 . We performed LDA+U calculations for different
magnetic structures, including the "real" AF (Fig.3.11). In Fig.3.12 the
LDA+U partial DOS are plotted. Here, the results for U = 2.8 (eV) and
J = 0.93 (eV) are presented. These values of U and J were calculated by
Solovyev et al. [94], taking into account the screening of t2g interactions by
eg electrons. The striking point of the LDA+U result is that the electronic
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
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.
3.6.4 Exchange Interactions in the Ladder Vanadates CaV2 0 5

and MgV 2 0 5
Here, we present the results of ab initio calculations of the electronic structure
and exchange couplings in the layered vanadates CaV20 5 and MgV 20 5.
Based on these results we provide a possible explanation of the unusual
magnetic properties of these materials, in particular the large difference in
the spin gap between CaV20 5 and MgV 20 5 [97J.
Spin-1/2 ladder models can describe the magnetic behavior of a variety of
quasi-one-dimensional systems [98J. Examples include the cuprate materials
SrCu203 [99], LaCu02.5 [100], and (Sr,Cah4Cu24041 [101]. Spin excitations
in the isolated ladders have a finite energy gap, which makes them prototype
spin liquids. This is of interest in relation to high-temperature supercon-
ductivity, since upon doping they become resonating-valence-bond liquids,
with a spin excitation gap and a dominant quasi-long-range pairing correla-
tions [98J.
Another example of the spin-1/2 ladder systems are the layered vanadate
compounds CaV20 5 and MgV 20 5. Although CaV20 5 and MgV 20 5 have
nearly identical vanadium oxygen planes, their magnetic properties are strik-
ingly different. CaV 205 has a large spin gap of about 600 K [102J, while
the spin gap in MgV 20 5 is very small, only about 20K [103]. In contrast
with the planar cuprates, where a hole in the Cu x 2 - y2 orbitals results
Fig. 3.13. Various magnetic interactions in Ca(Mg)V2Ü5 compounds. Two nearest

vanadium ladders with different z-coordinates are shown in different shades. lt , the
exchange interaction between nearest V atoms is ferromagnetic for CaV2Ü5 and
antiferromagnetic for MgV2Ü5. hand h are, respectively, the antiferromagnetic
exchange interactions along the rung and leg of the ladder. J4 is the antiferro-
magnetic exchange interaction between the V atoms along the diagonal of the
ladder. The magnetic structure used in the calculations is marked with up- and
down- arrows. Half of J1 and all of J 4 exchange interactions are frustrated in this
magnetic structure
in a strong antiferromagnetic exchange coupling for the 180° bonds and

a weak ferromagnetic one for the 90° bonds, the exchange interactions in
these vanadates can be more complicated, as shown in Fig.3.13. Even the
signs of the many exchange couplings are not obvious in these materials.
So one has to resort to ab initio numerical calculations to get information
about the relative as weH as absolute values of the exchange couplings in
these systems. The determination of the exchange couplings is crucial to
understand the markedly different spin-gap behavior in these compounds.
In this section, we discuss the computational results of the exchange
couplings that appeared to be indeed different in these two compounds, con-
sistent with their magnetic properties. As the various exchange couplings are
related to the bare hopping matrix elements, we extract them using a recently
developed systematic downfolding scheme [104].
The main building block of the crystal structures of Ca(Mg)V20 5 com-
pounds are the V ions, roughly in the center of a pyramid of oxygen ions
as can be seen in Fig.3.14. The crystal structure of CaV 20 5 is primitive
orthorhombic with space group Pmmn and lattice constants a = 11.35 A,
b = 3.60A, and c = 4.89A. As shown in Fig.3.14 (left) , the structure is
formed by a linkage of V0 5 pyramids having apex oxygens in the direction
of the c-axis. Oxygen edge- and corner-shared zigzag V chains are formed
along the b-axis, where the nearest-neighbor V- V distance is 3.03 A. Along
the a-axis, these chains are linked by sharing corners, with a V- V distance
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
of 3.49 A. It thus forms a quasi-two-dimensional ladder layer in the ab plane

with the leg along the zigzag V chains (i.e along b), while the rung is in the
perpendicular direction (i.e. along a). The Ca atoms are located between the
layers and are surrounded by eight 0 atoms.
The crystal structure of MgV 20 5 is base-centered orthorhombic with
space group Cmcmj and lattice constants a = 11.02 A, b = 3.69 A, and
c = 9.97 A. Again, the structure can be described as a linkage of V0 5
pyramids having apex oxygens in the direction of the c-axis, as can be seen
in Fig.3.14 (right). The V zigzag chains extend along the a-axis by sharing
edges and corners of the pyramids, and the nearest neighbor V-V distance is
2.98 A. They are also linked by sharing corners along the b-axis, with a V-V
distance of 3.37 A, which again leads to quasi-two-dimensionalladder layers in
the ab plane. However, in contrast to CaV 20 5, these layers stack alternately
with the distance of (c/2) and the Mg atoms are located between the layers
and each are surrounded by six oxygen atoms. As a consequence, there is
a puckering of the V 205 layers to accommodate Mg ions in the tetrahedral
coordination. It should be noted that as a consequence of the puckering of
the V 20 5 layers the tilting angle of the corner-shared pyramids is appreciably
smaller in MgV 20 5 in comparison to CaV20 5.
The main building block of the crystal structures of the Ca(Mg)V20 5
compounds are the V ions roughly in the center of a pyramid of oxygen ions.
The relevant point group symmetry is C4v . The five d orbitals ofthe vanadium
ion transform according to the following irreducible representations: 3z 2 - r 2
(Al), x 2 - y2 (Bd, xy (B 2) and (xz, yz) (E). The lowest energy orbital is
the V 3d orbital of xy symmetry (using a convention where the axes of the
coordinate system are directed towards the oxygen ions), which is the orbital
whose lobes point in the directions where the overlap with the oxygen is the
smallest. Due to the crystal-field splitting, the degeneracy of the V 3d shell
is lifted and the single d electron of the V4+ ion occupies this xy orbital,
which reminds us of the cuprates, with a single hole in the X 2 _y 2 orbital.

The important difference is that while in cuprates all copper atoms are in
the same (x, y) plane as the x 2 _y 2 orbital, in these vanadates the vertices of
the pyramids point up and down alternately with respect to the basal plane.
Thus, the V ions in their centers are correspondingly above and below the
central plane, as can be seen in Fig. 3.14. As the xy orbitals are parallel to this
plane, so the overlap (and hence the exchange couplings) is expected to be
stronger for vanadium ions situated on the same side of the ladder plane. We
will show that this is indeed the case. In addition to this alternation, a tilting
of the pyramids is present in the crystal structure of these compounds, which
we shall see seriously infiuences the interactions.
So in the problem under consideration there are two types of contributions
to the exchange interaction parameters Jij . The first is due to the xy-xy
orbital hopping. Other contributions are due to the hoppings to all other
orbitals and as the mean-field approximation is much better for a multiorbital
model, this part can be used without modification. As mentioned earlier, the
strongest inter action must be between V atoms that are situated on the same
side of the plane (above or below) (see Fig. 3.13). These atoms form ladders
with interactions along the rung and the leg of the ladder denoted as J 2 and
J3 , respectively, and the interaction between the ladders as J 1 (the notations
are chosen to refiect the interatomic distances; the shortest one is between
the atoms on different sides of the plane).
The calculated values of the exchange couplings are presented in Ta-
ble 3.2. It can be immediately seen that, indeed, the strongest interactions
are between atoms on the same side of the plane (the ladder exchanges h,
J 3 ). There is very strong anisotropy between the exchange interactions along
the rung (J2 = -608K) and the leg (J3 = -122K) for CaV 20 5 . However,
for MgV 20 5 the rung (J2 = -92K) and the leg (h = -144K) exchange
interaction parameters are comparable in size.
Our results suggest that CaV 205 is a system of weakly coupled dimers
along the rung of the ladder with a very strong interaction inside the dimer.
The analysis [105] based on fitting the results of model calculations to the
experimental susceptibility measurements for CaV 205 confirms the coupled
dimer picture and one of the obtained set of parameters (J2 = -665 K, h =
-135 K, J 1 = 25 K) is very elose to our ab initio calculated parameters values.
Table 3.2. Calculated exchange coupling parameters (in K)
CaV205 MgV2 0 5
h 28 -60
h -608 -92
h -122 -144
J4 -20 -19
However, for MgV 2 0 5 our calculations suggest ~ = -1.53 and 9- = -2.40,

which puts MgV 2 0 5 outside the scope of the lander limit, consrstent with
the helical ordered gapless phase according to the phase diagram obtained
by the Schwinger-boson mean-field theory [106]. Recently [107], the exchange
parameters for CaV2 0 5 and MgV 2 0 5 obtained in the LDA+U method were
used for the calculations of the uniform susceptibility of the Heisenberg model
by the quantum Monte Carlo method. The results agree very weIl with the
experimental measurements, and particularly very good agreement has been
found for CaV2 0 5 .
Our calculation shows that a strong anisotropy exists between the rung
and leg exchange couplings for CaV2 0 5 thus making it a system of weakly
coupled dimers along the rung with strong interaction inside the dimer, char-
acterized by a large spin gap. On the other hand the rung and leg exchange
couplings are found to be of comparable strength for MgV 2 0 5 making it
a small spin gap system.
3.7 Orbital Ordering and Jahn-Teller Effect

We must emphasize here that in spite of the model-Hamiltonian spirit in the
above derivation of the LDA+U formula, it remains a "first-principles", "ab
initio" method preserving its ability to calculate the lattice properties such
as the ground state crystal structure, equilibrium volume and even phonon
frequencies. The orbital-dependent potential of the LDA+U method (3.8)
makes it possible to treat properly the orbital polarization and corresponding
to it lattice Jahn-Teller distortions and polarons [23,108].
3.7.1 Cooperative Jahn-Teller Distortions

in Transition-Metal Compounds: KCuF 3
This was demonstrated for the example of the perovskite KCuF 3 [23]. This
compound is subject to a collective Jahn-Teller-like distortion that is more
complicated than the simple tetragonal distortion of the cuprates, involving
a staggering of quadrupolar-distorted CuF 4 units (2 short and 2 long Cu-F
bonds) in the a-b planes. In a seminal work, Kugel and Khomskii [109] pointed
out that this distortion is, in the first instance, electronically driven. They
showed that the e g (x 2 _y2, 3z 2 -1) orbital degrees of freedom are, as weIl as
the spins, subject to kinetic exchange interactions, while in addition, the spin
and orbital degrees of freedom are mutually coupled as weIl. Orbital ordering
is found from these Hamiltonians involving a staggering of the orbitals in
a-b directions (rv x 2 - Z2, y2 - z2) and a ferromagnetic ordering of the spins.
Because of this 'pre-existing', electronically driven orbital polarization, any
nonzero electron-phonon interaction then leads automatically to the observed
lattice distortion [109].
KCuF 3 has the perovskite crystal structure with a slight tetragonal dis-
tortion (c/a ratio< 1) while the planes show the quadrupolar distortion. The
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
Fig.3.16. The three-dimensional plot of the electron-spin-density distribution in

KCuF3 from the results of LDA +U calculations. Note that x 2 - y2 - and y2 - z2 -like
"d orbitals" correspond to the spin density located at the copper atoms, while the
p-like density corresponds to fluorine ions
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].
3.7.2 Orbital Ordering in the Doped Manganite Prl_",Ca",MnOa
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
localized spin (8 = 3/2). In contrast to this, e g orbitals, which have lobes

directed to the neighboring oxygen atoms, hybridize strongly with 0 2p,
producing, in the result, rather broad bands. The strong exchange interaction
with t29t subshellleads to the splitting of the eg band into unoccupied eg-1- and
half-occupied egt subbands. The half-filled egt subband is a typical example
of a Jahn-Teller system, and, indeed, LaMn03 has an orthorhombic crystal
structure [114] with distorted (elongated) Mn06 octahedra. This cooperative
Jahn-Teller effect is usually considered to be responsible for the opening of
the gap in the half-filled egt band and an insulator ground state of LaMn03.
The orthorhombic crystal structure of LaMn03 (Pbnm space group) can
be described as a perovskite with two types of distortion from a cubic struc-
ture: 1) Tilting (rotation) of the Mn06 octahedra, so that Mn-O-Mn angles
become less than 180 0 , and 2) Jahn-Teller distortion of the octahedra, with
one long Mn-O bond and two short bonds, the long bonds alternatively in
a- and b-directions. By replacing La with other trivalent ions with smaller
ionic radius, the Mn-O-Mn bond angle becomes smaller and the e g band
width W is reduced to enhance the tendency to the carrier localization and
lattice distortion. A variety of dramatic phenomena have been observed in
(Pr, Nd, Sm) (Sr,Ca)Mn03 systems. Let us discuss the application ofLDA+U
scheme to Prl/2Cal/2Mn03 [115]. This system has a very peculiar phase dia-
gram [116,117]. At low temperature, it is an antiferromagnetic insulator with
acharge ordering of MnH and MnH accompanied by an orbital ordering:
the orbital ordering and spin ordering is of CE-type. Between TN (~ 180K)
and TeD (~ 240K), there is no long-range magnetic ordering but the system
remains insulating because of the persistence of the charge ordering. Above
TeD' it is a paramagnetic insulator.
The local spin density approximation (LSDA) was used by several groups
for theoretical investigation of the electronic structure of Lal-xSrxMn03
[118-120]. Those calculations have confirmed the importance of the Jahn-
Teller distortion for correct description by the LSDA of the insulating anti-
ferromagnetic ground state ofthe undoped LaMn03' because the caIculations
with an undistorted cubic perovskite crystal structure produce a half-filled
metallic egt band. However, in order to address the charge and orbital order
observed in doped manganites it is necessary to go beyond LSDA by including
the intra d shell Coulomb interaction (LDA+U method).
In order to treat the experimentally observed CE-type of antiferromag-
netism it was necessary to quadrupie the Pbnm-type unit cell and the super-
cell had 16 formula units. The distribution of the Pr and Ca atoms in the
lattice was chosen according to a model where all Pr ions have only Ca as
their nearest neighbors in the lanthanide sublattice. No symmetry restrictions
were imposed in the caIculation, the integration in the reciprocal space being
performed for the whole Brillouin zone, and the self-consistency iteration was
started from a uniform distribution of e g electrons over both e g orbitals of
all Mn atoms.
Fig.3.17. The calculated angle distribution of the egt-electron spin density in

Prl/ 2Cal / 2Mn03 [115]
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 .
3.8 Highly Correlated Metallic Compounds

and Metal-Insulator Transition
3.8.1 Correlation Effects in Ruthenates
The detection of superconductivity in Sr2Ru04 [121] is of great importance
since this system is the only layered perovskite compound known so far that is
superconducting in the absence of copper and without requiring doping. Thus,
a critieal comparison of photoemission Fermi surface data with those derived
from dHv A measurements is feasible. Surprisingly, independent studies of the
dHvA effect [122] and angle-resolved photoemission [123-125] yield highly
contradietory Fermi surface topologies. The new photoemission results [126]
have resolved this contradietion.
Because of the layered structure of Sr2Ru04, the electronie bands dose
to the Fermi level may be qualitatively understood in terms of a simple tight-
binding pieture. These bands are derived mainly from Ru t2g states. The wide
xy band exhibits a two-dimensional character, while the narrow xz and yz
bands are nearly one-dimensional. All three bands are roughly 2/3 occupied,
giving about four Ru d electrons per formula unit. Density functional calcu-
lations based on the local density approximation (LDA) [127,129] place the
(1f,0), (0,1f) saddle point van Hove singularity ofthe xy band about 60 meV
above the Fermi energy. Taking into account gradient corrections slightly
lowers this singularity to about 50 meV above E F [129]. Figure 3.19 provides
a qualitative pieture of the t2g bands and of the Fermi surface exhibiting
one hole sheet (a) and two electron sheets (ß, 1'). Whereas the dH vA data
[122] are consistent with these results, photoemission spectra reveal a funda-
mentally different topology [123-125]: the xy van Hove singularity near M
appears below the Fermi level, converting the l' sheet from electron-like to
hole-like. Nevertheless, both experiments are reported to be in accord with
Luttinger's theorem.
According to the reduced dimensionality of Sr2Ru04, creation of a photo-
hole should be associated with highly anisotropie screening processes, which
refiect the nature of the different electronie states involved. As shown in
Fig. 3.19, the relevant bands near E F comprise a roughly 3.5eV wide band
formed by inplane hopping between Ru d xy and 0 2p orbitals, and 1.4eV
narrow d xz , d yz bands. Assuming an onsite Ru d-d Coulomb interaction
U ~ 1.5eV, we have the intriguing situation: Wxz,yz < U< W xy , where Wi
is the width of the i-th t2g band. A value U ~ 1.5eV was, in fact, deduced
from the observation of a valence band satellite in resonant photoemission
from Sr2Ru04 [132]. According to this picture, intraatomie correlations have
a much larger effect on the xz, yz bands than on the xy band, giving rise to
a strongly anisotropie self-energy. Because of the rv 2/3 filling of the xz, yz
bands, their narrowing, combined with Luttinger's theorem, leads to acharge
fiow from the xz, yz bands to the xy band. As we discuss below, for reasonable
:;- 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)
Fig.3.20. Real and imaginary parts of self-consistent second-order self-energy for

U = 1.2eV, J = 0.2eV. Solid curves: xy, dashed curves: xz
E,
(a) IOMC I
>
~
*
U)
U5 0
'0
.z..
.a; (b) ~
c
Q)
0
xy
0
-3 -2 -1 o
Energy (eV)
Fig.3.21. (a) Quasiparticle density of states Ni(W) derived from self-consistent

QMC scheme for U = 1.2 eV, J = 0.2 eV. (h) Single-particle density of states pi(W)
derived from tight-binding bands. Solid curves: xy, dashed curves: xz
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
(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
near M for 25eV photon energy), it is unavoidable to piek up spectral weight

from occupied regions near M, even when the detector is nominally set at M.
Thus, the near-degeneracy of the xy singularity with EF makes it extremely
diflicult using angle-resolved photoemission to determine the k-point at whieh
the xy band crosses the Fermi energy. Photoemission data taken with bett er
energy and angle resolution might provide a more conelusive answer.
Figures 3.21 and 3.22 also show that due to the narrowing of the XZ, yz
bands, the weakly dispersive band is shifted from -0.8eV to about -O.4eV,
in agreement with photoemission data [123-125]. For kll between M and X,
this band is observed to cross E F at about (11",0.611"), in good accord with our
calculations. In addition, the calculations indieate the existence of a satellite
below the xz, yz bands whieh might be related to the spectral feature observed
near 2.5 eV binding energy using resonant photoemission [132]. The precise
location of this satellite is diflicult to determine because of the uncertainty
of U and the approximate nature of our self-energy calculations.
Because of the proximity of the quasipartiele xy van Hove critieal point
to the Fermi level, the imaginary part of the self-consistent self-energy ex-
hibits a small linear contribution near EF, indieating that the system may
partially behave like a marginal Fermi liquid. In fact, it is only the first term
rv R ll1 (W) that gives rise to a linear term if the singularity coincides with
E F . As a result of multiband effects, however, this contribution is rapidly

dominated by stronger quadratie terms involving the narrow xz, yz bands.
Thus, we find the marginality to be rather weak.
We finally discuss the mass renormalization derived from our quasipartiele
bands. For Coulomb and exchange matrix elements in the range U = 1.2-
1.5eV, J = 0.2-0.4eV we find m* Im ~ 2.1-2.6, in agreement with photoe-
mission estimates m* Im ~ 2.5 [125], while dHvA measurements [122] and
specific heat data [134] suggest a factor of 3-4.
The multiband quasipartiele calculations for Sr2Ru04 show that the si-
multaneous existence of nearly one- and two-dimensional t2g bands near E F
leads to a highly anisotropie self-energy of the photoemission hole state.
Because of Luttinger's theorem, this anisotropy gives rise to acharge flow
from the narrow xz, yz bands to the wide xy band, thereby shifting the xy
van Hove singularity very elose to E F • As a result, in the vicinity of M,
considerable spectral weight appears below E F . These results might explain
the controversial nature of recent photoemission data that have difliculty in
determining whether or not the xy band at M is occupied.
We discuss now the electronie structures of the metallic and insulating
phases of the alloy series Ca2-xSrxRu04 [135]. The ground state is an or-
bitally nondegenerate antiferromagnetic insulator (x = 0) and a good metal
(x = 2), respectively. The most challenging is the region x ~ Xc = 0.5 where
there are strong correlations but no symmetry breaking so that both LDA and
LDA+U are inapplicable. At this concentration we employ the LDA+DMFT
scheme.
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)
Fig.3.23. Results of LDA+DMFT(NCA) calculations obtained within LDA DOS

for Sr2Ru04. The solid line is the DOS for xz, yz-orbitals and the dashed line for
(xy)-orbital. At U = 1.5 eV the xz, yz-orbitals become localized. At U = 2.5 eV
additionally the localization of xy-orbital occurs. The Fermi energy is defined to be
zero and was adjusted to conserve the number of particles (4 electrons per site)
We performed aseries of calculations using this LDA+DMFT (NCA)

approximation scheme [136] for the Sr2Ru04 structure. We increased the
value of the Hubbard U to examine how the onsite correlations grow. In
Fig. 3.23 is aseries of results for the density of states (DOS) in the xy and
(xz, yz ) subbands. Since these subbands have quite different widths, the onset
of Mott localization occurs at different critical values of U. Thus, we see that
as U is increased through a value of U ;::::j 1.5 eV there is a transfer of electrons
between the subbands so that the integer occupancy of 3 electrons and Mott
localization appears in (xz, yz) subbands, while the broader half-filled xy
band remains itinerant. This unusual behavior is driven by the combination of
the crystal-field splitting, and the narrower bandwidth ofthe (xz, yz) orbitals.
A furt her increase in the value of U to U ;::::j 2.5 eV is required to obtain Mott
localization also in the xy subband.
These results lead us naturally to the following proposal to explain the
anomalous properties in the critical concentrations x = xc. The electronic
configuration is now (3,1). The 3 electrons in the {xz,yz} subbands are
Mott localized and have a local moment of S = 1/2. The remaining valence
electrons are in the itinerant xy band and are responsible for the metallic
character. Thus, at this concentration, we have the unusual situation of 10-

calization in only part of the 4d orbitals and coexisting localized and itinerant
4d orbitals. Note that in the orthorhombic crystal structure at x = Xc the 2
subbands have different parity under reflection around a RU02 plane, sim-
ilar to tetragonal Sr2Ru04, which forbids direct hybridization between the
subbands. This proposal explains, in a natural way, the unexpected moment
of S = 1/2 of the Ru ions and the coexistence of metallic behavior and local
moments.
Note that the calculations are carried out more conveniently by increas-
ing the value of the onsite repulsion, U, which, however, should not change
appreciably with the concentration, x. In reality, it is the band width that is
changing with the decreasing x as the RU06 octahedra progressively rotate
when Ca is substituted for Sr. The key result, however, is the existence of
a parameter range where this partial localization is stable. The fact that
we calculated only for the highly symmetrie Sr2Ru04 structure, rat her than
the distorted structure is, we believe, unimportant in establishing this (3,1)
configuration as a stable electronic configuration.
3.8.2 Doped Mott Insulators: Lal_", Sr", Ti0 3

The LDA+DMFT approach was applied to the doped Mott insulator Lal-x
Sr xTi0 3 [137] LaTi0 3 is a Pauli-paramagnetic metal at room temperature
and below T N = 125 K antiferromagnetic insulator with a very small gap
value (0.2eV). Doping by a very small value of Sr (a few per cent) leads
to the transition to paramagnetic metal with a large effective mass. As
photoemission spectra of this system also show strong deviation from the
noninteracting electrons picture, Lal-xSrx Ti0 3 is regarded as an example of
strongly correlated metal.
The crystal structure of LaTi0 3 is slightly distorted cubic perovskite. The
Ti ions have octahedral co ordination of oxygen ions and t2g-eg crystal-field
splitting of the d shell is strong -enough to survive in the solid. In Fig. 3.24,
the spectral function obtained from the LDA+DMFT(QMC) calculation at
temperature T ;:::j 1000 K is compared with the noninteracting t 2g density of
states. One can see the typical features of the spectra of strongly correlated
systems: a lower Hubbard band, a well-pronounced quasiparticle peak, and an
upper Hubbard band. While for the noninteracting case 100% of the spectral
weight is located in the quasiparticle band, the LDA+DMFT spectra are
characterized by a spectral weight transfer from the quasiparticle band to
the Hubbard bands and a narrowing of the quasiparticle band.
In Fig. 3.25, we compare the results obtained within various approxima-
tions (IPT and NCA) with the numerically exact QMC simulation, all at
T ;:::j 1000 K. One not es that within IPT the shape of the upper Hubbard
band is not correct. Moreover, there is no quasiparticle peak at 1000 K, the
reason being that IPT underestimates the Kondo temperature considerably
such that the very narrow quasiparticle peak found at low temperatures (see
"
," ,
,,, ,, ,
,
-4 -2 6
Energy (eV)
Fig.3.24. Partial t2g densities of states of LaTi03 calculated with

LDA+DMFT(QMC) (solid lines) and LDA (dashed lines)
~
' _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
temperatures and/or low frequencies [139]. Similarly, the deficiencies of the

IPT results are not entirely surprising in view of the semiphenomenological
nature of this approximation, especially for a system away from half-filling.
This comparison shows that the choice ofthe method used to solve the DMFT
equation is indeed important.
Photoemission spectroscopy of the early transition metal oxides provides
a direct tool for the study of the electronic structure of strongly correlated
materials. A comparison of the experimental photoemission spectra [140]
with the results obtained from LDA and LDA+DMFT(QMC) at 1000K are
shown in Fig.3.26. To take into account the uncertainty in U, we present
the results for U = 3.2, 4.25, and 5eV. All spectra are multiplied with the
Fermi step function and Gaussian broadened with a broadening parameter
of 0.3 eV to simulate the experimental resolution [140]. The LDA band struc-
ture calculation clearly fails to reproduce the broad band observed in the
experiment at 1- 2eV below the Fermi energy [140]. Taking the correlations
between the electrons into account, this lower band is easily identified as the
lower Hubbard band whose spectral weight originates from the quasiparticle
band at the Fermi energy and increases with U. The best agreement with
experiment concerning the relative intensities of the Hubbard band and the
quasiparticle peak and, also, the position of the Hubbard band is found for
U = 5 eV. The value U = 5 eV is still compatible with the ab initio calculation
of this parameter. One should also note that the photoemission experiment is
sensitive to surface properties. Due to the reduced co ordination number at the
surface, the band width is likely to be smaller and the Coulomb interaction
to be less screened, i.e. larger. Both efIects make the system more correlated
and, thus, might also explain why better agreement is found for U = 5eV.
---..... QMC. U=3.2 eV

---- QM C. U=4.25 eV
-- QMC. U=5.0 cV
'"
..... - --- - LDA
·13 0 Exp.
;::s
~
.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
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.
3.8.3 Nature of Insulating State in NaV 2 0 5

Above Charge-Ordering Transition:
A Cluster DMFT Study
The ladder compound NaV 20 5 has been the subject of great interest for
the last five years [142-151]. It exhibits aremarkable phase transition at
Tc = 34K, now identified as charge-ordering of zigzag type [143,145-148].
Both the charge-ordered phase and the charge-disordered one are insulating
with an energy gap of the order of 0.8-1 eV [144]. The presence ofthe gap in
the ordered phase is not surprising and it was reproduced successfully, for ex-
ample, in recent LDA + U calculations [150]. The properties of the disordered
phase are much more difficult to understand. In contrast to the isostructural
ladder compound, CaV20 5 , NaV 20 5 has a quarter-filled band [143] rather
than a half-filled one and cannot be considered as a standard Mott insulator
[1]. It has been proposed in [146] that a large value of the transverse hopping
parameter in the ladder, which splits the band into sub-bands of the bonding
and antibonding states [143], could be responsible for the Mott insulator
behavior in NaV20 5 in the presence ofstrong Coulomb interactions. However,
it is not known whether this mechanism is adequate for the realistic values
of the parameters characterising the single-particle electronic structure and
the electron-electron correlations. Here, we discuss the correlation effects in
NaV 20 5 taking into account nonlocal dynamical charge fluctuations on the
rung for this two-leg ladder compound [152]. This allows us to understand
the origin of the insulating states above Tc and to estimate the relative
importance of various physical mechanisms responsible for the gap formation.
The crystal structure of NaV 20 5 pro jected in the xy plane is schematically
shown in Fig. 3.27. The results of the X-ray [147], NMR [148], and optical
[144] experiments as well as the Hartree-Fock calculations [145] support the
zigzag charge-ordering state for low temperatures (see Fig. 3.27). In this state,
one has approximately one d electron per rung of the vanadium ladder. We
start with LDA + U [8] calculations of the ordered states but in contrast to the
previous work [150] we considered several different types of charge ordering.
This gives us an opportunity to estimate the onsite and intersite Coulomb
interactions U and V respectively which in turn were used to parametrize
the model Hamiltonians to be used for the calculations taking into account
the dynamical correlation effects.
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
By mapping the results of the LDA+U calculations for different types of

charge ordering on the results of model calculations with onsite (U) and inter-
site (V) Coulomb interaction parameters, we obtained the following values:
U = 2.8eV and V = 0.17eV.
We used the extended Hubbard model for a two-Ieg ladder:
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
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,
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
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)
LDA+U Coulomb parameters about 70% of the ground state eigenfunction

corresponds to the configurations with one electron per rung and 20% related
to empty rung states. This means that the configurations with the empty
rungs in NaV 205 also have appreciable weight.
One can conclude that the insulating state of NaV 20 5 above T = 34 K
is characterized by strong dynamical charge fl.uctuations. In this respect, the
situation is close to the half-filled Hubbard chain or an pseudospin-liquid-like
phase: the tendency for the formation of the state with one electron per rung
is similar to the singlet state formation in strongly frustrated spin systems and
can be described in the limit of large U and V by an anisotropie Heisenberg-
like model [146,149,159,160]. It is important to stress that experimentally
the spin gap exists also above Tc [151]. However, this description is only
qualitative.
Our calculations demonstrate that this compound is situated near the
boundary between metal and insulator states and should be described by
itinerant-electron models rather than by localized-electron ones. Since the
metal-insulator transition in the model under consideration turns out to be
strongly first-order type, the gap is not small. The value of the gap depends
strongly on the intersite Coulomb repulsion parameter V, as is demonstrated
in Fig. 3.30. The insulating properties cannot be described correctly while
neglecting the charge fl.uctuations on the rung (too small values of Vor using
single-site DMFT approach instead of the cluster one).
To conclude, we present the cluster DMFT calculations for the insulating
phase of NaV 20 5 above the charge-ordering transition using the extended
Hubbard model with the first-principle tight-binding parameters. The non-
local charge fl.uctuations and intersite Coulomb interaction is of crucial im-
portance for the formation of insulating state.
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-
büdy theüry with the practice üf first-principle calculatiüns üf the electronic

structure and prüperties für real materials.
References
1. N.F. Mott, Metal-Insulator Transitions (Taylor and Francis, London 1974)

2. T. Moriya, Spin Fluctuations in Itinemnt Electron Magnetism (Springer,
3. F. Gebhard, Mott Metal-Insulator Transition (Springer, Berlin Heidelberg
New York 1997)
4. W. Metzner and D. Vollhardt: Phys. Rev. Lett. 62, 324 (1989); A. Georges
and G. Kotliar: Phys. Rev. B 45, 6479 (1992); M. Jarrell: Phys. Rev. Lett.
69, 168 (1992)
5. A. Georges, G. Kotliar, W. Krauth, and M.J. Rosenberg: Rev. Mod. Phys.
68, 13 (1996)
6. P. Hohenberg and W. Kohn: Phys. Rev. 136 B864 (1964); W. Kohn and L.J.
Sham: ibid. 140, A1133 (1965): R.O. Jones and O. Gunnarsson: Rev. Mod.
Phys. 61, 689 (1989)
7. P. Fulde, J. Keller, and G. Zwicknagl, in Solid State Physies, vol. 41 (Acadernie
Press, New York 1988) p.1
8. V.I. Anisimov, F. Aryasetiawan, and A.I. Lichtenstein: J. Phys.: Condens.
Matter 9, 767 (1997)
9. V.I. Anisimov and A.I. Lichtenstein, in: Strang Coulomb Correlations in Elee-
tronie Strueture Caleulations, vol. 1 (Gordon and Beach, 2000) p.97
10. A.I. Lichtenstein and M.I. Katsnelson, in: Band Ferromagnetism. Ground
State and Finite- Tempemture Phenomena (Lecture Notes in Physics)
(Springer, Berlin Heidelberg New York 2001) p.75
11. A. Svane and O. Gunnarsson: Phys. Rev. Lett. 65, 1148 (1990)
12. S. Massida, M. Posternak, and A. Baldereschi: Phys. Rev. B 48, 5058 (1993)
13. O. Gunnarsson, O.K. Andersen, O. Jepsen, and J. Zaanen: Phys. Rev. B 39,
1708 (1989)
14. V.I. Anisimov and O. Gunnarson: Phys. Rev. B 43, 7570 (1991)
15. L. Hedin: Phys. Rev. 139, A796 (1965)
16. L. Hedin and S. Lundqvist, in Solid State Physics vol. 23 (Academic Press,
New York 1969) p.1
17. P. van Gelderen, P.A. Bobbert, P.J. Kelly, and G. Brocks: Phys. Rev. Lett.
85, 2989 (2000)
18. F. Aryasetiawan: Phys. Rev. B 46, 13051 (1992)
19. R.W. Godby, M. Schlüter, and L.J. Sham: Phys. Rev. B 37, 10159 (1988)
20. F. Aryasetiawan and O. Gunnarsson: Phys. Rev. Lett. 74,3221 (1995)
21. V.I. Anisimov, J. Zaanen, and O.K. Andersen: Phys. Rev. B 44, 943 (1991)
22. V.I. Anisimov, LV. Solovyev, M.A. Korotin, M.T. Czyzyk, and G.A.
Sawatzky: Phys. Rev. B 48, 16929 (1993)
23. A.I. Lichtenstein, J. Zaanen, and V.I. Anisimov: Phys. Rev. B 52, R5467
(1995)
24. V.I. Anisimov, A.I. Poteryaev, M.A. Korotin, A.O. Anokhin, and G. Kotliar:
J. Phys.: Condens. Matter 9, 7359 (1997)
25. A.I. Lichtenstein and M.I. Katsnelson: Phys. Rev. B 57,6884 (1998)
26. P.W. Anderson: Phys. Rev. 124, 41 (1961)

27. B.R. Judd, Operator Techniques in Atomic Spectroscopy (McGraw-Hill, New
York 1963)
28. F.M.F. de Groot, J.C. Fuggle, B.T. Thole, and G.A. Sawatzky: Phys. Rev. B
42, 5459 (1990)
29. S.L. Dudarev, G.A. Botton, S.Y. Savrasov, C.J. Humphreys, and A.P. Sutton:
Phys. Rev. B 51, 1505 (1998)
30. A.K. McMahan, R.M. Martin, and S. Satpathy: Phys. Rev. B 38, 6650 (1988)
31. M.S. Hybertsen, M. Schlüter, and N.E. Christensen: Phys. Rev. B 39, 9028
(1989)
32. M. Korotin, T. Fujiwara, and V. Anisimov: Phys. Rev. B 62,5696 (2000)
33. O.K. Andersen: Phys. Rev. B 12, 3060 (1975)
34. O.K. Andersen and O. Jepsen: Phys. Rev. Lett. 53, 2571 (1984)
35. H. Sawada, Y. Morikawa, K. Terakura, and N. Hamada: Phys. Rev. B 56,
12154 (1997)
36. O. Gunnarsson, O. Jepsen, and O.K. Andersen: Phys. Rev. B 21, 7144 (1983)
37. M.1. Katsnelson and A.I. Lichtenstein: J. Phys.: Condens. Matter 11, 1037
(1999)
38. K. Takegahara: J. Phys. Soc. Japan 62, 1736 (1992); M.J. Rozenberg: Phys.
Rev. B 55, R4855 (1997)
39. M.1. Katsnelson and A.1. Lichtenstein: Phys. Rev. B 61, 8906 (2000)
40. J.E. Hirsch and R.M. Fye: Phys. Rev. Lett. 25, 2521 (1986)
41. M. Jarrell and J.E. Gubernatis: Physics Reports 269, 133 (1996)
42. N.E. Bickers and D.J. Scalapino: Annals of Physics 193, 206 (1989)
43. G. Esirgren and N.E. Bickers: Phys. Rev. B 51, 5376 (1998)
44. M. Fleck, A.1. Liechtenstein, A.M. Oles, L. Hedin, and V.1. Anisimov: Phys.
Rev. Lett. 80, 2393 (1998)
45. P. Lambin and J.P. Vigneron: Phys. Rev. B 29, 3430 (1984)
46. H.J. Vidberg and J.W. Serene: J. Low Temp. Phys. 29, 179 (1977)
47. H. Keiter and J.C. Kimbal: Phys. Rev. Lett. 25, 672 (1970); N.E. Bickers, D.
Cox, and J.W. Wilkins: Phys. Rev. B 36, 2036 (1987)
48. M.B. Zoelfl, T. Pruschke, J. Keller, A.1. Poteriaev, LA. Nekrasov, and V.1.
Anisimov: Phys. Rev. B 61, 12810 (2000)
49. J. Hubbard: Proc. Roy. Soc. A 285, 542 (1965)
50. J. Hubbard: Proc. Roy. Soc. A 216, 238 (1963); V.Y. Irkhin and Y.P. Irkhin:
Phys. Stat. Sol. B 183, 9 (1994)
51. M.H. Hettler, A.N. Tahvildar-Zadeh, M. Jarrell et al.: Phys. Rev. B 58, 7475
(1998); M.H. Hettler, M. Mukherjee, M. Jarrell et al.: Phys. Rev. B 61, 12739
(2000)
52. A.1. Lichtenstein and M.1. Katsnelson: Phys. Rev. B 62, R9283 (2000)
53. A.R. Mackintosh and O.K. Andersen, in: Electron at the Fermi Surface, ed.
M. Springford (University Press, Cambridge 1980) p. 145
54. A.1. Liechtenstein, M.1. Katsnelson, and V.A. Gubanov: J. Phys. F 14, L125
(1984); Solid State Commun. 54, 327 (1985); A.1. Liechtenstein, M.1. Kat-
snelson, V.P. Antropov, and V.A. Gubanov: J. Magn. Magn. Mater. 61, 65
(1987)
55. J.M. Luttinger and J.C. Ward: Phys. Rev. 118, 1417 (1960); see also G.M.
Carneiro and C.J. Pethick: Phys. Rev. B 11, 1106 (1975)
56. V.P. Antropov, M.1. Katsnelson, and A.1. Liechtenstein: Physica B 231-238,
336 (1997)
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
87. M. Foex: C. R. Aead. Sei. 223, 1126 (1946)

88. C. Castellani, C.R. Natoli, and J. Raninger: Phys. Rev. B 18, 4945 (1978);
18, 4967 (1978); 18, 5001 (1978)
89. T.M. Riee, in Spectroscopy of Mott Insulators and Correlated Metals, eds.
A. Fujimori and Y. Tokura (Springer, Berlin Heidelberg New York 1995)
pp. 221-229
90. J.-H. Park et al. (unpublished)
91. P.D. Dernier and M. Marezio: Phys. Rev. B 2, 3771 (1970)
92. P.D. Dernier: J. Chem. Phys. 31, 2569 (1970)
93. L.F. Mattheis: J. Phys.: Condens. Matter 6, 6477 (1994)
94. I. Solovyev, N. Hamada, and K Terakura: Phys. Rev. B 53, 7158 (1996)
95. A.T. Mizokawa and A. Fujimori: Phys. Rev. B 48, 14150 (1993)
96. W. Bao et al.: Phys. Rev. Lett. 78, 507 (1997)
97. M.A. Korotin, V.1. Anisimov, T. Saha-Dasgupta, and I. Dasgupta: J. Phys.:
Condens. Matter 12, 113 (2000)
98. E. Dagotto and T.M. Riee, Scienee 271, 618 (1996)
99. M. Azurna, Z. Hiroi, M. Takano, K Ishida, and Y. Kitaoka: Phys. Rev. Lett.
73, 3463 (1994)
100. Z. Hiroi, M. Takano: Nature 377, 41 (1995)
101. E.M. MeCarron, M.A. Subramanian, J.C. Calabrese, R.L. Harlow: Mater.
Res. Bull. 23, 1355 (1988)
102. M. Onoda and N. Nishiguehi: J. Solid State Chem. 127,359 (1996); H. Iwase
et al.: J. Phys. Soe. Jpn 65, 2397 (1996)
103. M. Onoda and A. Ohyama: J. Phys.: Condens. Matter 10, 1229 (1998); P.
Millet et al.: Phys. Rev. B 57, 5005 (1998); M. Isobe et al.: J. Phys. Soe. Jpn.
67 755 (1998)
104. T. Saha-Dasgupta, O.K Andersen, G. Krier, C. Arcangeli, R.W. Tank, O.
Jepsen, and I. Dasgupta (to be published); O.K Andersen, C. Arcangeli, R.W.
Tank, T. Saha-Dasgupta, G. Krier, O. Jepsen and I. Dasgupta, in Tight-
binding Approach to Computational Materials Science eds. P.E.A. Turehi,
A. Gonis and L. Colombo, MRS Proe. 491 (Materials Research Society,
Warrendale 1998)
105. S. Miyahara, M. Troyer, D.C. Johnston, and K Ueda: J. Phys. Soe. Jpn. 67,
3918 (1998)
106. B. Normand, K Pene, M. Albrecht, and F. Mila, Phys. Rev. B 56, R 5736
(1997)
107. M.A. Korotin, I.S. Elfimov, V.1. Anisimov, M. Troyer, D.1. Khomskii: Phys.
Rev. Lett. 83, 1387 (1999)
108. V.1. Anisimov, M.A. Korotin, J. Zaanen, and O.K. Andersen: Phys. Rev.
Lett. 68, 345 (1992)
109. KI. Kugel and D.1. Khomskii: Usp. Fiz. Nauk. 136, 621 (1982) [Sov. Phys.
Usp. 25, 231 (1982)]
110. S.K Satija, J.D. Axe, G. Shirane, H. Yoshizawa, and K Hirokawa: Phys. Rev.
B 21, 2001 (1980); M.T. Hutehings, H. Ikeda, and J. M. Milne: J. Phys. C
Solid State Phys. 12, L 739 (1979)
111. R.M. Kusters, J. Singleton, D.A. Keen, R. MeGreevy, and W. Hayes: Physica
(Amsterdam) 155B, 362 (1989); A. Urushibara, Y. Moritomo, T. Arima, A.
Asamitsu, G. Kido and Y. Tokura: Phys. Rev. B 51, 14103 (1995)
112. G.H. Jonker and J.H. van Santen: Physiea (Amsterdam) 16,337 (1950); J.H.
van Santen and G.H. Jonker: Physiea (Amsterdam) 16, 599 (1950)
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)
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
J. König, J. Schliemann, T. Jungwirth, and A.H. MacDonald
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.
interactions between Mn local moments that are mediated by holes in the

semiconductor valence band. There is, therefore, a lot of similarity between
the ferromagnetism of these materials and that of lanthanides and actinides
and their compounds in which f-electron moments are coupled by d-band
itinerant electrons. There are also similarities between these materials and
the manganite compounds that have been extensively studied [6] in recent
years in part because of the large increase in resistance that occurs when
T exceeds the Curie temperature, the so-called colossal magnetoresistance
effect. In (III,Mn)V ferromagnets, however, the local moments appear on
only a small fraction of the atomic sites arranged randomly. In addition the
itinerant electron density is also low, even lower than the Mn local moment
density. As we explain later this property is likely important in selecting
ferromagnetic over glassy magnetic order. The participation of itinerant elec-
trons in the ferromagnetism of these diluted magnetic semiconductors (DMS)
adds to their richness, leading in particular to electronic transport properties
that are very sensitive to the magnetic state of the material. The physics
of ferromagnetic semiconductors is in asense intermediate between that
of rare earth magnets and that of manganites in that the spin-splitting of
itinerant electron bands due to their exchange coupling with local moments
is comparable to their Fermi energies, rather than being much smaller than
band Fermi energies as in the rare earth case or larger as in the manganite
case.
The chapter is organized as follows. In Sect. 4.2 we briefly summarize the
main experimental properties of (III,Mn)V ferromagnets. Several different
but related approaches that have been explored in an effort to gain insight
into these materials are outlined in Sect. 4.3. The remaining sections of this re-
view chapter deal with the development of the semi-phenomenological model
we favor in which the low energy degrees of freedom are exchange-coupled
valence-band holes and S = 5/2Mn local moments that carry a negative
charge. The simplest version of this model is one in which the randomly
distributed Mn ions are replaced by a uniform continuum, thus completely
neglecting disorder. In Sect.4.4 we discuss physical predictions based on
a mean-field treatment of this disorder-free model and demonstrate that it
successfully describes a number of non-trivial properties of (Ga,Mn)As and
(In,Mn)As ferromagnets, including their anomalous Hall conductivities. In
Sect. 4.5 we discuss collective excitations of these ferromagnets within the
disorder-free model, demonstrating that the simple mean-field-theory is rea-
sonably reliable for typical parameters of current sampies but must fail at
large carrier densities and also in the limit of very strong exchange coupling.
In Sect. 4.6 we discuss the results of Monte Carlo calculations that describe
the effect of collective fluctuations of Mn moment orientations. The method
can deal with some of the complications and additional physics, including the
possibility of non-collinear ground states, that enters when disorder is added
to the theoretical model. A brief summary is given in Sect. 4.7.
4 Ferromagnetism in (III,Mn)V Semiconductors 165
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.
4.2 Properties of (III,Mn)V Ferromagnets
In this seetion we discuss some important properties of (III,Mn)V ferromag-

nets that have been established by current experiments. Thorough recent
reviews of the properties of these materials have been prepared by Dietl,
Matsukura, and Ohno [7,8]. Our objective in this section is to summarize
the observations that are most important in constraining theoretical descrip-
tions. There is at present considerable activity related to the growth and
characterization of these materials. We expect rapid progress to be achieved
in the near future in exploring the range of possible behaviors and relating
them more precisely to molecular-beam-epitaxy growth and post-growth an-
nealing protocols. We list below a number of properties that appear to be
safely established. A theoretical picture that is able to explain most of these
properties is outlined in the following sections.
• Electron paramagnetic resonance and optical experiments [9,10] demon-

strate that S = 5/2 local moments occur for dilute concentrations of
Mn in GaAs. These experiments demonstrate that the Mn local moment
model is correct.
• The Mn-induced states near the Fermi energy playa key role in the origin
of ferromagnetism and in the magnetotransport properties of (III,Mn)V
DMS's. According to photoemission studies [11-13], those states have
As 4p character, i.e., can be associated with the host semiconductor
valence band states. The angle-resolved photoemission experiment also
showed a negligible shift in the heavy- and light-hole bands with Mn
concentration x ::; 7%.
• Ferromagnetism is not observed for Mn concentrations smaller than
rv 0.01 [5]. This property demonstrates that ferromagnetism does not
occur when all valence-band holes are trapped on individual Mn ions or on
other defects. Antisite defects, for example, are common in semiconduc-
tor sampIes grown by low-temperature molecular beam epitaxy (MBE).
For very dilute Mn concentrations, electron spin resonance experiments
demonstrate that most holes are trapped not at the Mn acceptors, but
at other defects.
• The ground-state magnetization, M(T = 0), per Mn ion can exceed
4MB for larger values of x [5]. Since the magnetization contribution from
antiferromagnetically coupled valence band holes tends to partially com-
pensate the Mn local moment magnetization, ferromagnets with these
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.
4.3 Theoretical Approaches
Ferromagnetism is a collective effect due to interactions between electrons. If

it were possible to do so, we would explain its microscopic origins in a par-
ticular dass of materials by solving the many-electron Schrödinger equation
directly, given the position of all the nudeL Fortunately this uninteresting di-
reet approach is impossible, now and forever more, because of the macroscopic
number of interacting electronic degrees of freedom. Instead we must resort to
some combination of approximation and phenomenological modeling, settling
on the correct approach only after careful comparison with experiment. This
is the art of condensed matter science; an intricate tango between theory
and experiment leading to a condusion that cannot be anticipated while
the dance is in progress. In some cases, fractional quantum Hall systems
and possibly cuprate superconductors for example, the low-energy degrees of
freedom in terms of which observable physical properties are best described
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
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.
c) Antiferromagnetic exchange coupling between the Mn2+ spin and valence-

band holes, J pd 2:1 SI· s(RI ). This interaction represents virtual cou-
pling to states that have been integrated out of the model's Hilbert
space, ones in which electrons are exchanged between the Mn ion d shells
and the valence band [21,22]. The exchange interactions are isotropic to
a good approximation because the Mn 2 + ion has total angular moment um
L = O. Experimental estimates for Jpd vary from 150 ± 40meVnm3 [1],
to 68 ± 10meVnm3 [27], to 55 ± 10meVnm3 [11]. Recent experimental
work limits Jpd to a value toward the lower end of this range and fixes
its value within perhaps 20%.
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.
d) The attractive Coulomb interaction between the ionized Mn H acceptor

and a valence-band hole. In an envelope function formalism, central-cell
corrections to the interaction are necessary to capture the isolated bound-
acceptor limit accurately [28].
e) The repulsive Coulomb interaction among holes. This interaction is key
in screening the ionized Mn 2 + acceptors and cannot be neglected except
in the completely-Iocalized-hole limit. When it is included, it usually
must be approximated in a way which avoids artificial hole-hole self-
interactions.
f) The repulsive interaction between holes and ionized antisite (group-V
element on group-III site) defects. The antisite defects compensate for
the Mn acceptors and reduce the overall hole density, in addition to
providing an important additional scattering center. In the dilute Mn
limit, experiment [29] suggests that most MnH ions do not have bound
holes, possibly due to this compensation. When all holes are strongly
localized, most Mn local moments will be free and the system will not
have ferromagnetic order.
The following terms are potentially important in some circumstances.
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
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.
4.4 Mean-Field-Theory Predictions

In this and the following section we make an important approximation that
achieves a drastic simplification. We will refer to this as the continuum Mn
approximation, although it has sometimes been referred to as a virtual crystal
approximation. It is motivated by the observation that the Fermi wavelength
of the valence-band electrons is typically longer than the distance between Mn
ions, mainly because the Mn acceptors are compensated and also partially
because there are four occupied valence bands. When the Mn ion distri-
bution is replaced by a continuum with the same spin-density, randomness
is completely eliminated from the minimal model. This approximation has
many elements in common with the dynamic mean-field-theory (DMFT)
approximation, applied [30] to these ferromagnets recently by Chattopad-
hyay et al. , although the DMFT does retain some of the consequences of
randomness neglected here and in the following section. The possibility of
starting with a description based on an approximation where the random Mn
distribution is replaced by a continuum emphasizes an essential difference
between ferromagnetic semiconductors and c1assical spin gl ass systems in
which dilute Mn local moments are distributed randomly in a metallic host.
In both cases it is true that the inter action between local moments mediated
by the itinerant electrons is oscillatory and ferromagnetic for separations
smaller than the itinerant electron Fermi wavelength. However, the Fermi
wavelength is longer than the distance between local moments in the doped
semiconductor case, whereas it is shorter in the spin glass case [31]. Each Mn
ion interacts ferromagnetically with several of its neighbors. The continuum
Mn approximation will fail in the limit of dilute Mn ions and also if the
exchange interaction between band and Mn spins is too strong. This approx-
imation does not allow for the sensitivity of magnetic properties to annealing
protocols that has been established in experiment. It does, however, seem to
be reliable in the limit of principle interest, that of high Mn densities and high
critical temperatures, where the holes are metallic and their interaction with
Mn acceptors will be effectively screened. The merit of this approximation is
that it enables quantitative prediction of many physical properties. In this

section we discuss three important properties, the ferromagnetic transition
temperature, the magnetic anisotropy energy, and the anomalous Hall con-
ductance. This work described below is motivated by the view in science, it is
ultimately up to experiment to decide on the reliability of any approximation
made in modeling a physical system. As we will point out, the utility of the
continuum Mn approximation is strongly supported by observations. In fact,
it is not a surprise that this approximation is a good starting point, given its
success in describing the influence of external fields on the properties of the
closely related paramagnetic (II,Mn)VI semiconductors [20,22]. The present
section makes in addition a mean-field approximation by ignoring correlations
between Mn and band spin configurations.
Our mean-field theory is derived in the spin-density-functional frame-
work and leads to a set of physically transparent coupled equations [32].
The effective magnetic field seen by localized magnetic ions consists of an
external magnetic field and the mean kinetic-exchange-coupling contribution
from spin-polarized carriers,
(4.1)
where (s(R1 ) is the carrier spin density at Mn sites, and g is the g-factor
of the local moments. The mean spin polarization of a magnetic ion is given
by [33]
(Sh = -SBS(SgPBHeff(RI)/kBT)Heff(RI) , (4.2)
where Bs(x) is the Brillouin function and Heff(RI) is the unit vector along
the direction of the effective magnetic field defined in (4.1). The itinerant-
hole spin density is determined by solving the Schrödinger equation for holes
which experience a kinetic-exchange effective Zeeman field h( r). The field
h( r) is non-zero only in the ferromagnetic state and, in the continuum limit,
reads
= J pd NMn(r) (S)(r) ,
h(r) (4.3)
where NMn = 4x/arc is the Mn density in MnxIII 1 - xV zincblende semi-
conductors with a lattice constant alc. For inhomogeneous systems such
as quantum wells or superlattices, the itinerant holes experience also an
electrostatic potential due to heterostructure confinement and external bias
(if present). Using the local-spin-density approximation (LSDA), itinerant
hole-hole interaction can be accounted for by including an additional spin-
dependent one-particle potential in the Schrödinger equation.
4.4.1 Ferromagnetic Transition Temperature

In homogeneous DMS systems, the hole-spin density (8) and the kinetic-
exchange potential h are related at small h by
Xj
(8) = - ( )2h. (4.4)
g*PB
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)
To understand the qualitative physics implicit in this Tc-equation (4.6),

we discuss first the magnetic susceptibility expressions of a model itinerant
electron system with a single spin-split band and an effective mass m*. The
kinetic-energy contribution Et~f to the total energy gives
d2(Ef~f IV)
(4.7)
dh 2
where k F is the Fermi wavevector. The exchange energy ofthe spin-polarized
parabolic-band model adds a contribution
d2(E~~th IV)
(4.8)
dh 2
where c is the dielectric constant of the host semiconductor. At high hole
densities p, the kinetic-energy term dominates and Tc is proportional to the
Fermi wavevector, i.e., to p1/3. Equations (4.7) and (4.8) also show that the
band contribution to the mean-field Tc increases linearly with m* while the
exchange enhancement of Tc is proportional to (m*)2. Note that correlation
effects, not discussed here in detail, suppress the mean-field Tc by only rv 1%
for typical experimental hole densities (p rv 0.1 nm- 3 ) in bulk (III,Mn)As
ferromagnets.
To obtain quantitative predictions for Tc, it is necessary to evaluate the
kinetic and exchange contribution to the itinerant hole susceptibility using
a realistic six-band Kohn-Luttinger model [26], instead ofthe parabolic band
model. The Hamiltonian contains the spin-orbit splitting parameter Llso and
three other phenomenological parameters, '/'1, '/'2, and '/'3, whose values for
the specific III-V host can be found, e.g., in [7,34]. Results [7,35-37] are
plotted in Fig. 4.2 as a function of the hole density for InAs, GaAs, AIAs,
and GaN host semiconductors. In the density range considered, only the two
heavy-hole and two light-hole bands are occupied in the arsenides. However,
the mixing between these four bands and the two spin-orbit split-off bands
is strong and must be accounted for. In GaN, spin-orbit coupling is weak
and aH six bands are occupied by holes. The numerical data are consistent
with the qualitative analysis based on the parabolic band model: the band
50
InAs
25
100
50
g 0
f-"
AIAs
100
50
600
500
400
300
0.1 0.2 0.3 DA 0.5
p(nm-")
Fig.4.2. The band (kinetic energy) contribution to the mean-field ferromagnetic

critical temperature Tc for Mn concentration x = 5% is plotted as a function of
hole density p for InAs, GaAs, AIAs, and GaN host semiconductors
(kinetic-energy) contribution to Tc follows roughly the pl/3 dependence, the

exchange enhancement of rv 10% is only weakly density dependent. The Tc
values at a given density are ordered according to the heavy-hole and light-
hole masses in the arsenide hosts. For GaN, with all six bands occupied, the
simple model of a parabolic spin-split band is less instructive. Yet the large
numerical Tc's in this material are consistent with the large heavy-hole mass,
nearly twice as large as in AIAs.
The mean-field prediction for the critical temperature agrees quantita-
tively with the experimental value of 110 K measured in Mn-doped GaAs with
Mn concentration x = 5% and p = 0.35 nm- 3 . Thermal fluctuations neglected
by the mean-field theory, discussed in the following section, reduce the theo-
retical Tc estimate by less than 5% [37], explaining the quantitative success of
the mean-field theory in this sampie. The same analysis finds approximately a
Tc suppression [37] of approximately 20% compared to mean-field theory due
for (Ga,Mn)N, implying that room temperature ferromagnetism may occur
in III-V DMS.
4.4.2 Magnetic Anisotropy

Experiments [14,38] in (III,Mn)V DMS's have demonstrated that these ferro-
magnets have remarkably square hysteresis loops and that the magnetic easy
axis is dependent on epitaxial growth lattice-matching strains. The physical
origin [7,26] of the anisotropy energy in our model is spin-orbit coupling in the
valence band. Even in mean-field theory, we find that the magnetic anisotropy
physics of these materials is rich and that easy axis reorientations can occur
as a function of sam pIe parameters including hole density or epitaxial growth
lattice-matching strains.
Magnetic anisotropy in the absence of strain is weIl described by a cubic
harmonie expansion truncated at sixth order, an approximation commonly
used in the literature [39] on magnetic materials. The corresponding cubic
harmonie expansion for total energy of a system of non-interacting holes in
the presence of the effective field h is
E totV(1~f) = E tot ((100))

V + Kca(h?
1
h? + h? h,2 + h,2 h,2) + K ca h,2 h,2 h,2 (4 9)
x y y z x z 2 x y z, .
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
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
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
depends on the substrate on which the epitaxial (Ga,Mn)As film is grown.

The most important conclusion from Fig. 4.6 is that strains as small as 1% are
sufficient to completely alter the magnetic anisotropy energy landscape. For
-
Q
tCO
Q)
4
0.. 2
'-
Q)
0..
>.
a:: o
'?
o
,.... Etot <OO1>E tot< 100>
-2 Etot <111>E tot< 100>
Etot <110>E tot<100>

-4
-0.10 -0.05 0.00 0.05 0.10
90
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
example for (Ga,Mn)As on GaAs, eo = -0.0028 at x = 0.05. The anisotropy

has a relatively strong uniaxial contribution, even for this relatively mod-
est compressive strain, which favors in-plane moment orientations [7,26],
in agreement with experiment [5]. A relatively small (rv 1 kJ m- 3 ) residual
in-plane anisotropy remains which favors (110) over (100). For x = 0.05
(Ga,Mn)As on a x = 0.15 (In,Ga)As buffer the strain is tensile, eo = 0.0077,
and we predict a substantial uniaxial contribution to the anisotropy energy
which favors growth direction orientations [7,26], again in agreement with
experiment [5]. For the tensile case, the anisotropy energy changes more
dramatically than for compressive strains due to the depopulation of higher
subbands. At large tensile strains, the sign of the anisotropy changes, empha-
sizing the subtlety of these effects and the latitude which exists for strain-
engineering of magnetic properties.
4.4.3 Anomalous Hall Effect
The mean-field description of hole bands in the presence of exchange cou-

pling to the localized Mn moments provides a starting point for building
a theory of transport in (III,Mn)V ferromagnets. Here we concentrate on
the anomalous Hall effect which is an important sam pIe characterization
tool in ferromagnetic systems. The Hall resistivity of ferromagnets has an
ordinary contribution, proportional to the external magnetic-field strength,
and an anomalous contribution usually assumed to be proportional to the

sampie magnetization. In our approach [40], the anomalous Hall conductance
of a homogeneous ferromagnet is related to the Berry phase of the electronic
wavefunction acquired by a cyclic evolution along the Fermi surface.
In the standard model of the AHE in met als , skew-scattering [41 J and
side-jump [42J scattering give rise to contributions to the Hall resistivity
proportional to the diagonal resistivity p and p2 respectively, with the latter
process tending to dominate in alloys because p is larger. Gur evaluation of
the AHE in (III,Mn)V ferromagnets is based on a theory [43J of semiclassical
wave-packet dynamics which implies a contribution to the Hall conductivity
that is independent of the kinetic-equation scattering term. The interest in
this contribution is motivated in part by practical considerations, since our
current understanding of (III,Mn)V ferromagnets is not sufficient to permit
confident modeling of quasiparticle scattering. The relation of our approach
to standard theory is reminiscent of dis agreements between Smit [41J and
Luttinger [44J that occurred early in the development of AHE theory and
do not appear to have ever been fully resolved. We follow Luttinger [44J in
taking the view that there is a contribution to the AHE due to the change
in wavepacket group velo city that occurs when an electric field is applied to
a ferromagnet. The electron group velo city correction is conveniently evalu-
ated using expressions derived by Sundaram and Niu [43J:
XC = n~~ + (e/n)E x {l. (4.12)
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
importance for (III,Mn)V ferromagnets we first explore a simplified model

that yields parabolic dispersions for the two heavy-hole and two light-hole
bands, and neglect coupling to the split-off band by assuming a large spin-
orbit coupling [25,26]. Detailed numerical simulations accounting for the mix-
ing of the spin-orbit split-off bands and warping of the occupied heavy-hole
and light-hole bands [25,26] in the (In,Mn)As and (Ga,Mn)As sampIes [45,1]
will follow. Within the 4-band spherical model, the spin operator s = j /3,
and the Hamiltonian for holes in Ill- V host semiconductors can be written as
(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)
Here p = k~h/31r2 (1 + y'mlh/mhh) is the total hole density and mlh =

m/bl + 2/'2) is the light-hole effective mass. The lower bound in (4.16)
is obtained assuming mlh «mhh while the upper bound is reached when
mlh:::::; mhh·
Based on the above analysis we draw the following condusions: The
anomalous velocity due to the Berry phase can have a sizable effect on the
AHE in (III,Mn)V ferromagnets. The solid line in Fig. 4.7 shows our analytic
results for the GaAs effective masses mhh = 0.5m e and mlh = 0.08m e .
Note that in experiment, anomalous Hall conductances are in order of 1-
lOn- 1 cm- 1 and the effective exchange field h '" 10-100 meV. A large (JAH is
expected in systems with large heavy-hole effective mass and with the ratio
mlh/mhh dose to unity.
So far we have discussed the limits of infinitely strong spin-orbit coupling
and weak effective exchange field, relative to the hole Fermi energy. In the
opposite limits of zero spin-orbit coupling or large h, (JAR vanishes. This
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)
implies that the anomalous Hall conductivity is generally nonlinear in the

exchange field or the magnetization. To explore the intermediate regime we
diagonalized the six-band Luttinger Hamiltonian numerically [25,26) with
the spin-orbit gap L1 so = 1 eVasweIl as for the GaAs value L1 so = 341 meV.
Results shown in Fig. 4.7 confirm that a smaller aAR is expected in systems
with smaller L1 so and suggest that both positive and negative signs of aAR
can occur, in general. The curves in Fig. 4.7 are obtained by neglecting band
warping in lII-V semiconductor compounds. The fact that valence bands in
these materials are typically strongly non-parabolic, even in the absence of
the field h and in the large L1 so limit, is accurately captured by introducing
the third phenomenological Luttinger parameter /3 [25,26). Numerical data
including all Luttinger parameters indicates that warping tends to lead to an
increase of aAR, as seen when comparing solid curves in Fig. 4.7 and in the top
panel of Fig. 4.8. The hole-density dependence of aAR, illustrated in Fig. 4.8,
is qualitatively consistent with the spherical model prediction (4.16). Also in
accord with the outlined chemical trends, numerical data in Fig. 4.8 suggest
large positive AHE in (AI,Mn)As, intermediate positive aAR in (Ga,Mn)As,
and a relatively weaker AHE in (ln,Mn)As with the sign of aAR that may
depend on the detail structure of the sampie.
We make now a comparison between our a AH calculations and experi-
mental data in the (ln,Mn)As and (Ga,Mn)As sampies, analyzed in detail by
Ohno and coworkers [45,1,5). The nominal Mn densities in the two measured
systems are N Mn = O.23nm- 3 for the lnAs host and N Mn = 1.1nm- 3 for the
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
GaAs host, yielding saturation values of the effective field h = 25 ± 3 meV

and h = 122 ± 14 meV, respectively. The low-temperature hole density of the
(Ga,Mn)As sampIe, p = 0.35nm- 3 , was unambiguously determined [5J from
the ordinary Hall coefficient measured at high magnetic fields of 22-27 T.
Since a similar experiment has not been reported for the (In,Mn)As sampIe
we estimated the hole density, p = 0.1 nm -3, by matching the measured
ferromagnetic transition temperature Tc = 7.5 K to the density dependent
mean-field Tc.
As demonstrated in Fig.4.8, our theory explains the order of magnitude
difference between experimental AHE in the two sampIes (O"AH ~ 1 n- 1 cm- 1
in (In,Mn)As and O"AH ~ 14n- 1 cm- 1 in (Ga,Mn)As). The calculations are
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.
4.5 Collective Excitations Within a Continuum Picture

4.5.1 Beyond Mean-Field Theory and RKKY Interaction
The power and the success of the mean-field pieture employed in the previous
section lies in the fact that it is a simple theoretieal approach whieh makes it
easy to calculate many observable quantities numerically. Mean-field theory,
however, neglects correlation between local-moment spin configurations and
the free-carrier state and, therefore, fails to describe the existence of low-
energy long-wavelength spin excitations, among other things. Because of
its neglect of collective magnetization fluctuations, mean-field theory, e.g.,
always overestimates the ferromagnetic critieal temperature. There are many
examples in itinerant electron systems where mean-field theory overestimates
ferromagnetic transition temperatures by more than an order of magnitude
and it is not apriori obvious that mean-field theory will be successful in
(III,Mn)V ferromagnets. Indeed, we will find that the multi-band character
of the semiconductor valence band plays an essential role in enabling high
ferromagnetic transition temperatures in these materials.
In this section we identify the elementary spin excitations, determine their
dispersion, and discuss implications for the Curie temperature [46-50]. The
starting point of our analysis is the itinerant-carrier-mediated ferromagnetic
interaction between local magnetic moments. Such an interaction is pro-
vided by the familiar Ruderman-Kittel-Kasuya-Yoshida (RKKY) theory. The
RKKY pieture, however, only applies as long as the perturbation induced by
the Mn spins on the itinerant carriers is small. As we will derive below, the
proper condition is L1 « EF where L1 = NMnJpdS is the (zero-temperature)
spin-splitting gap of the itinerant carriers due to an average effective field
induced by the Mn ions, and EF is the Fermi energy. This condition is,
however, never satisfied in (III,Mn)V ferromagnets, partially because the
valence-band carrier concentration p is usually much smaller than the Mn
impurity density N Mn . A related drawback of the RKKY picture is that it
assumes an instantaneous statie interaction between the magnetic ions, i.e.,
the dynamies of the free carriers are neglected. We will see below that this
dynamics is important to obtain aU types of elementary spin excitations. As

a consequence, RKKY theory does not provide a proper description 0/ the
ordered state in /erromagnetic DMSs.
As in the previous section, we use here the minimal model including terms
a)-c) of Sect. 4.3 and employ the Mn continuum approximation. Extensions
to the minimal model may be important in some circumstances. They are,
however, not essential for the general discussion in the present section, which
will attempt to explain the considerations that determine when coUective
effects neglected by mean-field-theory are important.
4.5.2 Independent Spin-Wave Theory for Parabolic Bands

The main idea of our theory is to derive an effective description for the Mn
spin system by integrating out the valence-band carriers and to look for ßuc-
tuations of the Mn spins around their spontaneous mean-field magnetization
direction (which we choose as the z-axis). Using the Holstein-Primakoff (HP)
representation [51J, we express the Mn spins in terms of bosonic degrees of
freedom. We expand the effective action up to quadratic order, Le., we treat
the spin excitations as noninteracting Bose particles. From the corresponding
propagator we deduce the dispersion of aU elementary spin excitations.
To keep the discussion transparent we start with a two-band model for the
itinerant carriers with quadratic dispersion. Later, in Sect.4.5.5, we extend
our theory to a model with a more realistic band structure described by
a six-band Kohn-Luttinger Hamiltonian.
For small ßuctuations around the mean-field magnetization, we can write
the spin operators as
S+(r) ~ b(r)V2NMnS, (4.17)

S-(r) ~ bt (r)V2NMn S, (4.18)
SZ(r) = NMn S - bt(r)b(r) , (4.19)
with bosonic fields bt(r), b(r). The state with fuUy polarized Mn spins (along
the z-direction) corresponds, in the HP boson language, to the vacuum with
no bosons. The creation of a HP boson reduces the magnetic quantum number
by one.
The partition function Z can be expressed as a coherent-state path in-
tegral in imaginary time over the HP bosons and the valence-band carri-
ers, which are fermions. Since the Hamiltonian is bilinear in the fermionic
fields, we can integrate out the itinerant carriers and arrive at an effective
description in terms of the impurity spin degree of freedom labeled by the
complex number coherent state labels for the boson fields, z and z. We get
Z = I V[zzJ exp( -Seff[ZZ]) with the effective action
(4.20)
where SBP[ZZ] = ft dT f d 3 r ZÖTZ is the usual Berry's phase term. In (4.20),

we have already split the total kernel G- 1 into a mean-field part (GMF)-l
and a fluctuating part 8G- 1 ,
(GMF)ijl = (öT - p,) 8ij + (iIHolj) + NMnJpdSsfj , (4.21 )
8Gijl(ZZ) = J;d [(zSij + zst) J2NMnS - 2zzsfj] , (4.22)
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
A Debye cutoff k D with kb = 61f 2 N Mn ensures that we include the correct

number of magnetic-ion degrees of freedom, Ikl :'S k D . The kernel of the
quadratic action defines the inverse of the spin-wave propagator,
where ~ = (P.J. - Pt)/p is the fractional free-carrier spin polarization, and

Et,..j.(q) is the energy of spin-up and spin-down valence-band holes, Et,.).(q) =
Eq ±.:1/2, and Eq = n?q2/(2m*). The second term of (4.24) is the the energy for
a Mn spin excitation in mean-field-theory, [lMF = JpdP~/2 = x.:1. It differs
from the itinerant-carrier spin splitting by the ratio of the spin densities x =
p~/(2NMnS), which is always much smaller than 1 in (III,Mn)V ferromagnets.
Mean-field theory is, thus, recovered by dropping the last term in (4.24). It
is this term that describes the response of the free-carrier system to changes
in the magnetic-ion configuration.
4.5.3 Elementary Spin Excitations
We obtain the spectral density of the spin-fluctuation propagator by ana-

lytical continuation, iVm ---+ D+iO+ and A(k,D) = ImD(k,D)j7r. In the
following we consider the case of zero temperature, T = O. We find three
different types of spin excitations [46].
Goldstone-Mode Spin Waves. Our model has a gapless Goldstone-mode

branch reflecting the spontaneous breaking of spin-rotational symmetry. The
dispersion of this low-energy mode for four different valence-band carrier
concentrations p is shown in Fig. 4.9 (solid lines). At large momenta, k ---+ 00,
the spin-wave energy approaches the mean-field result Dk1) ---+ DMF (short-
dashed lines in Fig.4.9). Expansion of the T = 0 propagator for small mo-
menta yields for the collective mo des dispersion,
(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
D(1) = _x_ Ek + O(k4 ) for (4.26)

k I-x
Dk1) = P Ek (Ll)2 + O(k4) for Ll« EF. (4.27)
32NMn S EF
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.
0.004
I
p=0.01nm-3
I
s~ 0.002 ........•..........•.. ; I..•....•.....••..•...•...•.••.............................
0.""
0.000
.......................... , .................. ,.;.~.-': -- ---

.............................. .
,
s~ 0.005 p=O.035nm -3 I
I
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
Stoner Continuum. We observe that the frequency vm is not only present

in the first term of (4.24), it enters the third term, too. This is the reason
why, in addition to the Goldstone mode, other spin excitations can appear
in our model. They are absent in a static-limit description, Le., when the
frequency dependence of the third term of (4.24) is neglected.
We find a continuum of Stoner spin-flip particle-hole excitations. They
correspond to flipping a single spin in the itinerant-carrier system and, since
x « 1, occur in this simple model at much larger energies near the itinerant-
carrier spin-splitting gap ..1 (see Fig. 4.10). For..1 > f.F and zero temperature,
all these excitations carry spin SZ = +1, i.e., increase the spin polarization.
1.05
1x
0.95 '--_---L._ _ -L-~_.L.-_--L _ _-.J
0.000 0.001 0.002 0.003 0.004 0.005

klk o
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
They therefore turn up at negative frequencies in the boson propagator we

study. When ,1 < EF , excitations with both SZ = +1 and SZ = -1 contribute
to the spectral function. The continuum lies between the curves -,1 - Ek ±
2JEkEF and for ,1 < EF also between -,1 + Ek ± 2JEk(EF - ,1).
Optical Spin Waves. We find additional collective modes analogous to the

optical spin waves in a ferrimagnet. Their dispersion lies below the Stoner
continuum (see Fig.4.1O). At small momenta the dispersion is
_SPl = ,1(1 _ x) _ ~ (4EF _ 2 - (2 - 5X)/~) + O(k4). (4.28)

k 1- x 5xLl 5x
The finite spectral weight at negative frequencies indicates that, because of

quantum fluctuations, the ground state is not fully spin polarized.
4.5.4 COIllparison to RKKY and to the Mean-Field Picture

For comparison we evaluate the T = 0 magnon dispersion assuming an RKKY
inter action between magnetic ions. This approximation results from our the-
ory if we neglect spin polarization in the itinerant carriers and evaluate the
static limit ofthe resulting spin-wave propagator defined in (4.24). The Stoner
excitations and optical spin waves shown in Fig. 4.10 are then not present and
the Goldstone-mode dispersion is incorrect except when ,1 « EF, as depicted
in Fig.4.9 (long-dashed lines).
In the mean-field picture, correlations among the Mn spins are neglected.
The mean-field theory can be obtained in our approach by taking the Ising
limit, i.e., replacing S . s by SZ sZ. As mentioned before, this amounts to
dropping the last term in (4.24). The energy of an impurity-spin excitation
is then dispersionless, QMF = xL1 (short-dashed line in Fig. 4.9), and always
larger than the real spin-wave energy.
4.5.5 Spin-Wave Dispersion for Realistic Bands
For a quantitative analysis [50] of the spin-wave dispersion we extend our

parabolic-band model to a six-band Kohn Luttinger Hamiltonian. The effec-
tive action for the HP bosons describing the Mn impurity spins is given by
the same formal expression (4.20) with the contributions (4.21) and (4.22) to
the kernel. The difference is that for each Bloch wavevector i and j now label
the states in a six-dimensional Hilbert space (instead of two dimensions for
spin up and down), Ho is the Kohn-Luttinger Hamiltonian, and sij and st
are 6 x 6 matrices.
The next step is again an expansion of the effective action up to quadratic
order in z and z. In the two-band model, where spin is a good quantum
number, only z(k, vm)z(k, vm ) the combinations appear, see (4.23). Since the
coherent state labels can can be viewed as boson creation and annihilation
operators, these contributions are diagonal in total boson number. In the
presence of spin-orbit coupling, however, spin is no longer a good quantum
number, and the combinations z(k, vm)z( -k, -vm ) and z(k, vm)z( -k, -vm )
which increase or decrease the number of HP bosons, come into play.
Since our aim here is to derive the dispersion relations of the low-energy
spin waves, rather than to address the full excitation spectrum including the
Stoner continuum and the optical spin waves, we take the static limit as
discussed in the context of the two-band model. After a Bogoliubov transfor-
mation, we obtain for the spin-wave energy
(4.29)
with the definition
(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.
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)
Spin Stiffness. The quantized energy of a long-wavelength spin wave in

a ferro magnet with uniaxial anisotropy can be written as
2K 2A 2 4
fh = NMnS + NMnS k + O(k ), (4.31 )
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
- - ---
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.
4.5.6 Limits on the Curie Temperature
Isotropie ferromagnets have spin-wave Goldstone collective modes whose en-

ergies vanish at long wavelengths,
(4.32)
where k is the wavevector of the mode. Spin-orbit coupling breaks rotational

symmetry which leads to a finite gap, see (4.31). According to oUf numerical
studies, though, this gap is negligibly small as far as the suppression of
ferromagnetism by collective spin excitations is concerned and can, therefore,
be dropped for the present discussion. Each spin-wave excitation reduces the
total spin of the ferromagnetie state by 1. The coefficient D = 2Aj(NMn S)
is proportional to the spin stiffness A. These collective excitations are not
accounted for in the mean-field approximation. If the spin stiffness is small,
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
k T,coll _ 28 + IDk 2 (4.34)

B C - 6 D'
for 82: 5/2 where D is the T = 0 spin-stiffness. To obtain this equation, we

have assumed that the spin waves can be approximated as non-interacting
Bose particles, replaced the dispersion by the long-wavelength limit (4.32),
and noted that the critical temperature estimate is proportional to Dk'b,
justifying the use of the classical expression for the mode occupation number
nk ~ kBT / rlk - 1/2. These considerations set an upper bound on the critical
temperature which is proportional to the spin stiffness, abound not respected
by mean-field theory.
To get a qualitative but transparent picture we employ the two-band
model with parabolic bands, and deduce the spin stiffness from (4.26) and
(4.27) for the strong and weak-coupling regime, respectively. For strong cou-
pling, i1/EF » 1, the exchange coupling completely polarizes the valence-
band electrons, and we find (using p« 2NMn 8) the Tc bound
Tcoll,s _ 28 + 1 (~) 1/3 (4.35)

C - 128 EF N Mn
For small i1/EF' the weak-coupling or RKKY regime, exchange coupling is

a weak perturbation on the band system. In this regime we get
2/3
Tcoll,RKKY = T,MF 28 + 1 ( N Mn ) (4.36)
C C 12(8+1)~ p ,
i.e., mean-field theory is reliable only for p/NMn « 1, as expected since in

this case the RKKY interaction has a range which is long compared to the
distance between Mn spins.
10
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.
4.6 Collective Fluctuations Beyond Spin Wave Theory

and Continuum Approximation
In the preceding sections we described the magnetic properties of Mn-doped
semiconductors by a model of itinerant carriers which are exchange-coupled to
localized magnetic moments formed by the dopants. An important property,
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.
4.6.1 Model Considerations

The considerations in this section are based on the kinetic-exchange model
discussed in the previous sections. To account for the finite extent of the
Mn ions [28] in the exchange term we replace Jpds(RI) . SI by Jd 3 rJ(r-
RI)s(r)· SI with the finite-range exchange parameter
J(r) = J pd 3 e- r2 /( 2a5) (4.37)

(27ra ö) 2"
Both the strength Jpd and range ao of this interaction are phenomenological
parameters to be fixed by comparison with experiment or, ideally, to be
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),
Heff(r) = "L J (r-R I )Ss1J/(9J.1-B), (4.40)

I
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).
4.6.2 Remarks on the Monte Carlo Method

A standard way to evaluate expectation values of the form (4.39) is to use
classical Monte Carlo algorithms which perform a random walk in phase
space of the classical variables ('I?,cp). The probabilities governing this Monte
Carlo dynamics are specified by the dependence of many-fermion energy on
the 10calized-spin configuration. The many-fermion ground state is a Slater

determinant whose single-particle orbitals are the 10west energy eigenstates
of a single-band or multi-band Hamiltonian. For the case of a parabolic band,
the matrix elements of the corresponding one-particle Hamiltonian in a plane-
wave basis read
n,2 k 2
(k a 11llka) = --8k'k8u'u + L3
, , S L Jk-k,e i(k k')R
- I fh . 'T u'u,
A
4.41
( )
2m* 2
I
where k and adenote wavevector and spin indices, respectively, Jk is the

Fourier transform of J(r), and L the edge length of the simulation cube.
Periodic boundary conditions restrict the admissible values of wavevector
components to integer multiples of 27r / L. In (4.41), 'T is the vector of Pauli
spin matrices.
Since the many-particle ground state of the carrier system has to be
computed at each Monte Carlo step, the computational effort required for
the present calcu1ations is much larger than in simple classical spin models.
In the usual Metropolis algorithm, a single spin orientation is altered at each
step. If this algorithm were emp10yed here, the time required to diagonalize
the single-particle Hamiltonian each time wou1d severely limit the efficiency
of the algorithm. We therefore use the Hybrid Monte Carlo algorithm, which
was introduced in the mid 1980's in the context of lattice field theories [55].
In this method all classical variables are altered in one Monte Car10 step.
This drastically reduces the number of matrix diagonalizations required to
exp10re statistically important magnetic configurations. The Hybrid algo-
rithm is a powerful method for Monte Car10 simulations in systems containing
coupled classical and quantum mechanical degrees of freedom.
A concrete Monte Carlo simulation necessarily works in a system of finite
size. In the present case, calculation of the fermion ground state also requires
truncation of the plane-wave expansion we use for the independent particle
wavefunctions that diagonalize the Hamiltonian (4.41). We are able to ob-
tain convergence with respect to this truncation only when the range of the
exchange interaction ao is not too short. Taking into account a finite number
of plane-wave states entering the single-particle ca:r:rier Hamiltonian (4.41).
The importance of these finite-size effects in real and reciprocal space (using
periodic boundary conditions), and their interplay with the regularization
parameter ao, are discussed in detail in [49]. In the following we shallleave
aside such technica1 aspects and concentrate on the physical results.
4.6.3 Numerical Monte Carlo Results
In this subsection we present numerical Monte Carlo results. We concentrate

on the spin polarizations of the Mn ions and the carriers as a function of
temperature and address the ferromagnetic transition. We start with the
two-band model.
Two-Band Model. Figure 4.14 shows typical magnetization data as a func-

tion of temperature. These results were obtained for a Mn ion density of
N Mn = 1.0nm-3 , a carrier density p = 0.1nm- 3 in a cubic simulation volume
of V = 540nm3 , Le., the system contains 540 Mn ions and 54 carriers. The
effective band mass is half the bare electron mass, and the chosen exchange
parameter is J pd = 0.15eVnm3 . Here and in the following Monte Carlo data
the range parameter for the exchange coupling is ao = 0.1nm.
The main panel shows the average polarization of the Mn spins,
1
M = NMnV(IStotJ), (4.42)
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:
gMn = N~n V ((I StotI 2) - (I Stotl)2) , (4.44)
gp = p~ ((I (Olstot 10W) - (I (Olstot 10) 1)2) (4.45)
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).
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
0 100 150 200

Temperature [Kelvins]
0.5
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
Our objective here is not to make a quantitative prediction of the critical

temperature for particular ferromagnetic semiconductor systems. By doing
a numerically exact calculation for a model that captures much of the physics,
however, we hope to shed light on the range of validity and the sense and
magnitude of likely corrections to mean-field-theory Tc estimates.
The mean-field expression for Tc can be obtained by averaging the ion-
spin and carrier polarizations over space. The effective field which each Mn
spin experiences due to a finite carrier polarization is then constant in space
and the carrier bands are in turn rigidly spin split by L1 = JpdNMnS. (The
limit in which mean-field theory is exact can be achieved in our model by
letting ao ----t 00 in (4.37).) Mean-field theory, which neglects spatial fluctu-
ations and correlations between carriers and Mn spins, predicts that Tc is
quadratic in the exchange parameter J pd and linear in the effective band
mass m *, when correlations in the itinerant system are neglected.
Here we determine the critical temperature Tc with the help of our Monte
Carlo scheme. In Fig.4.15 we show results for Mn densities N Mn = l.Onm- 3
and a mean-field band splitting L1 = JpdNMnS = O.5eV. The left panel
shows the dependence of the critical temperature on the carrier effective
mass. This dependence is very important for the search for diluted magnetic
semiconductor systems with Tc's larger than room temperature. In the mean-
field approximation, Tc grows linear with increasing mass. The Monte Carlo
results elearly deviate from this prediction suggesting a saturation of Tc at
carrier masses elose to the bare electron mass. For even higher masses, we
500
p=O.1nm-3 NMn=1.0nm -3
NMn =1.0nm -3
'iii' m"=O.5m
c: 400 ß=O.5eV
I
ß=O.5eV
Mean Field
Mean Field Theory
I-U Theory
300
~ Simulation
Monte Carlo
Simulation
.5 1.0 o 0.1 0.2 0.3 0.4

Effective Mass m"/m Carrier Density p [nm -3]
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)
with q > O. Therefore the saturation of Tc as a function of the effective mass

at fixed Jpd corresponds to a linear dependence of Tc on Jpd at fixed m*.
This contrasts with the mean-field prediction TffF <X J;d.
In the right panel of Fig.4.15 we show Tc as a function of the carrier
density. Here the Monte Carlo approach also clearly yields a lower critical
temperatures than mean-field theory. For still higher carrier densities the
4 Ferromagnetism in (I1I,Mn)V Semiconductors 199
3.0
1.5
2.5 9Mn 1.0
0.5
c: 2.0 0.0
0
~ 0.10
~ 1.5
~ Mn ions 9p 0.05
c:
~ 1.0
o 100
0.5
0.0 L-~_,---~----,c.......>o,-=""-'_~----,_~--,
05 o 100 150 200 250

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
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
0.10 0.25 0.10 0.15 0.20 0.25

J pd [eVnm 3]
Fig.4.17. The right panel shows critical temperature Tc as a function of the

exchange parameter Jpd for the same particle densities as in Fig.4.16 and two
different system sizes. Both data sets agree within error bars and show a linear
dependence of Tc on Jpd. In the left panel, the corresponding data for the parabolic
two-band model with an effective mass of half the bare electron mass is plotted.
The latter system is a reasonable approximation to the six~band case
in the Kohn-Luttinger Hamiltonian. In the right panel of Fig. 4.17 we plot

the transition temperature as a function of the exchange coupling J pd for
two different system sizes. Both data sets agree within error bars and show
a linear dependence of Tc on J pd ' This finding is the same as for the two-
band model and contrasts with mean-field theory which predicts Tc cx: J;d.
The left panel of Fig. 4.17 shows the transition temperature as a function of
Jpd for the parabolic model for an effective mass m* = O.5m e . This value
is elose to the heavy~hole mass in the Kohn-Luttinger model for parameters
appropriate for GaAs. The data in the left panel can be obtained fram the left
panel of Fig. 4.15 via the scaling relation (4.46). Comparing the two panels
of Fig. 4.17 demonstrates that, in the range of carrier densities studied here,
a single parabolic band with an effective mass elose to that of the heavy-hole
Kohn-Luttinger-model band provides a reasonably good approximation to
the behavior of the six-band system. We expect larger differences to occur in
the strang coupling regime, which presents technical difficulties to the Monte
Carlo calculations.
4.6.4 Disorder Effects and Noncollinear Ferromagnetism
The magnetization data of Fig.4.14 were obtained by averaging over five

different realizations of the Mn ions. In fact over the range of parameters we
have explored, except for the larger values of p/NMn , results for different dis-
order realizations differ only very weakly from each other. This is illustrated
in Fig. 4.18, where the magnetization curves underlying the averaged results
3.0
Magnetisation curves lor

2.5 live different disorder
realisations
e: 2.0
~
'e:
.!!! 1.5
&.
e:
;% 1.0
0.5
0.0
05 o 100 150 200
Temperature [Kelvins)
Fig.4.18. Magnetization curves for five different realizations of Mn positions un-

derlying the averaged data of Fig. 4.14
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)
where v m = 27rm/ß is a Matsubara frequency. ZI(Vm ) stands for the bosonic

Hoistein-Primakoff field of Mn spin I that describes deviations from a fully
aligned state. The fluctuation matrix D1J (vm ) reads
D1J(Vm ) = LIJ(vm ) + KIJ(vm ) , (4.48)
with
LIJ = OIJ ( - iVm - J d3 rJ(r - RI )(sZ(r))) , (4.49)
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
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]
Fig.4.19. The disorder-averaged eigenvalue density of the matrix Di](O) de-

seribing magnetic fluetuations around the eollinear state. The data are obtained
for a simulation eube of volume V = L 3 = 400nm 3 with a Mn density of
NMn = 1.0nm- 3 and a density of p = 0.15nm- 3 of carriers having a band mass
of half the bare electron mass. The strength of the exchange interaction between
ions and carriers is Jpd = 0.05 eV nm -3 with a spatial range of aO = 0.40 nm. The
two upper panels show data for different wavevector cutoff qc with the Mn positions
chosen completely at random. The lowest panel contains data for the same situation
as the top one but with the Mn positions chosen from an fcc lattice. The peaks at
zero energy are due to the uniform rotation mode which strictly occurs in any
disorder realization
semiconductor the Mn ions are supposed to be located on the cation sites

forming an fcc lattice. In the bot tom panel of Fig. 4.19 we show data for the
same system parameters as in the top panel but with the Mn positions chosen
from an appropriate fcc lattice such that about 5% of all sites are occupied.
Both plots are practically identical indicating that our observations do not
depend on this detail of the modeling.
The shape ofthe eigenvalue distribution of the fluctuation matrix D1J (0)
is, in the model we have studied, sensitive to the Mn density NM n, the carrier
density p, and the Hamiltonian parameters m*, Jpd, and ao. Situations in
which the collinear ferromagnetic state is, for certain disorder realization,
stable can be approached most simply, for technical reasons, by letting ao be
larger. However for the value of p/NMn illustrated in Fig.4.19, correspond-
ing to a carrier density somewhat lower than measured in the highest Tc
sampIes, negative eigenvalues occur for nearly any Mn ion distribution. We
note that negative eigenvalues increase in number as the wavevector cutoff is
increased toward its converged value. These results suggest that noncollinear
ferromagnetic states are common, that they are sensitive to the distribution
of Mn ions and other defects - especially those that trap carriers, and that
the collinear ferromagnetic state tends to become unstable as mean-field band
eigenfunctions become more strongly localized around Mn ion site.
To furt her analyze the nature of this instability we consider the partici-
pation ratios for these excitations which we define by
(4.52)
where a~ is the I-th component of the j-th normalized eigenvector of D1J(0).

The ratio Pj is an estimate for the fraction of Mn sites that have important in-
volvement in the j-th spin wave. For example the zero-mode, uniform rotation
of all spins, has Pj = 1. Figure 4.20 shows the disorder-averaged participation
ratio as a function of spin-wave energy for the same situation as in Fig. 4.19.
The property that negative energy excitations have large participation ratios
shows that the instabilities of the collinear state involve correlated reorienta-
tions of many spins, rather than lone loosely coupled moments.
Figure 4.20 shows the disorder-averaged participation ratio for the same
situation as in the top panel of Figure 4.19. The negative-energy mo des have
clearly higher participation ratio than the eigenvectors at positive energy.
This shows that the instability 01 collinear state is due to long-ranged fiuctu-
ations involving a large lraction 01 the spins present in the system. Qualita-
tively the same observations are made for other values of system parameters.
The effect of a weak external magnetic field on the spin-wave excitation
spectrum of the noncollinear state is particularly simple. The field couples to
the local moment through its Lande g-factor, adding 2J1 B H ext to the energy
of a spin-wave excitation for S = 5/2, and to the spin and orbital degrees
of freedom of the band electrons. The orbital coupling leads to Landau
qe=6 '12(21t1L) V=400nm J

.Q N..,=1.0nm -3
~ 0.6 500 disorder
realizations p=O.15nm -3
c:
!o 0.4
m"=O.5m
Jpd=O.05eVnm 3
:2
1ä
Q.
0.2 ao=OAOnm
o
-0.002 o 0.002 0.004 0.006
Energy [eV)
Fig.4.20. The disorder-averaged participation ratio for the same situation as in

the top panel of Fig.4.19. The data is averaged over the sample intervals of the
histogram. The value at zero energy is enhanced due to the contribution of the
uniform rotation mode in that sample interval
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)
with gI = g} + ig;, and
g} = V2S (e'PI X ez ) • Jd3r[J(r - R I )
(( (s(r)) . e'PI) e'PI + ((s(r)) . e z ) e z ) X !h] , (4.54)
g; = V2Se z · (e'PI X J d 3rJ(r - RI)(s(r))) , (4.55)
where e'PI = (cos 'PI, sin 'PI, 0) and e z = (0,0,1).

The components gI represent the gradient of the energy with respect
to distortions parametrized by the ZI. As above in the case of fluctuations
around the collinear state, this zero-frequency contribution can also be ob-
tained via perturbation theory around the given state of Mn spin orientations.
The contribution to Saue quadratic in the Holstein-Primakoff variables in-
volves again a proper Matsubara sum and can be obtained straightforwardly.
However, its concrete form for the general case is more complicated, and
shall not be analyzed here. A given orientation of Mn spins is stationary with
respect to fluctuations if all complex coefficients gI vanish. This is the case if
and only if I d3 rJ(r-R I )(s(r)) is collinear with the direction {h of the local
ion spin. Therefore, the collinear ferromagnetic state is always a stationary
(but not necessarily stable) spin state.
We have employed the energy gradient expression (4.53) in a numerical
steepest descent procedure to search for true energy minima. aur results are
as follows: In cases where the energy minimum found by this method is dose
to the collinear state (with a magnetization of about 90% of the maximum
value or more), this minimum appears to be unique for each disorder realiza-
tion. We can therefore be confident that we have located the absolute ground
state of the system. In situations where the magnetization is reduced more
substantially, however, (by rv 20% or more) we converge to different energy
minima from different starting points. In these cases the model has substantial
spin-glass character with a complex energy landscape. This situation occurs
typically for larger ratios pi N Mn. For the system shown in Fig.4.20, for
instance, magnetization values at local energy minima are typically 30 to
40% of the collinear state value.
4.7 Concluding Remarks

III-V compound semiconductors with Mn substituted for a small fraction
of the cations are ferromagnets. In this chapter we have primarily discussed
the picture of the physical properties of these materials that follows from
a model in which each Mn pro duces a S = 5/2 local moment and acts
as an acceptor. The valence band holes and the exchange interaction that
couples them to the local moments are both treated phenomenologically.
In the simplest approximation the hole bands are taken to be those of the
host semiconductor , so that this model has a single unknown parameter,

the exchange coupling strength, whose value can be determined experimen-
tally. Many physical properties of this model are relatively simply calculated
when a virtual crystal approximation is adopted. When an envelope function
approach is used to describe the semiconductor bands this is equivalent to
replacing the Mn ions by a continuum, completely eliminating disorder. We
have shown that the mean-field theory of this model predicts critical temper-
atures in good agreement with experiment, that it can describe qualitative
features of the magnetic anisotropy energy including the effect of lattice-
matching strains between the magnetic epilayer and its substrate, and that
it implies values of the anamolous Hall effect often used to characterize
these ferromagnets that are in good agreement with experiment. We have
also described a theory of the elementary collective spin-wave excitations
of the virtual crystal model. By comparing the energy of these elementary
excitations with those of the mean-field theory, we are able to judge the
reliability of mean-field-theory critical temperature estimates. In this way
we find that we reach the wrong conclusion about mean-field theory when
we make the apparently innocent choice of using a single parabolic band
whose mass is equal to the heavy hole mass of the host semiconductor. For
such a model, we find that with typical parameters mean-field theory would
give a rat her poor estimate of the critical temperature. However when we
use a realistic band model with heavy and light holes that are coupled the
typical energy of collective magnetic excitations increases substantially. The
fact that heavy hole states are given approximately by J = 3/2 spin coherent
states with a k dependent orientation, plays an essential role in producing the
enhanced spin stiffness. We conclude that a realistic treatment of the valence
band, including its characteristic and strong spin-orbit mixing, is absolutely
necessary to achieve even a qualitative understanding of critical temperature
trends in these ferromagnets.
Some properties of (III,Mn)V ferromagnets are misrepresented by the vir-
tual crystal approximation. We have examined the effect of Mn site disorder
on these ferromagnets both by performing Monte Carlo calculations and by
evaluating the collective excitations of disordered systems. The Monte Carlo
calculations evaluate the properties of the model exactly for finite systems,
except for treating the Mn spin orientations as classical which will have
a small effect on thermal magnetization suppression at low temperatures.
These calculations confirm conclusions reached by using the virtual crystal
approximation for the most part, but also highlight some of its limitations. In
particular, while the collinear magnetic state is always stable in the virtual
crystal approximation, it can be unstable when disorder is accounted for,
leading to non-collinear but still ferromagnetic states. This property provides
an explanation to the dependence of a given sample's transition temperature
and of the magnetization on details of growth and post-growth annealing
procedures.
Research on Mn-doped semiconductors is a very active area of physics,

both theoretically and experimentally and there are many topics which we
have not been able to even touch upon. For example, studies of nanostructured
semiconductor systems with Mn-doped components represent an important
component of developments that are aimed towards future spintronic de-
vices. Nanostructures such as magnetic quantum wells are expected to be
unusual ferromagnets because of the possibility of using confinement effects
and doping profiles to manipulate their magnetic properties. The transi-
tion temperature of these systems can be tuned by external electric field
[57], as demonstrated in recent studies of (In,Mn)As field effect transistors
[58]. It is also predicted [59] that hysteresis properties of magnetic quantum
wells will be extremely sensitive to external bias voltages. As a result, the
magnetization orientation in quantum wells can be manipulated electrically
without changing the magnetic field. The possibilities for nano-engineering
of material properties have already been made apparent by the relatively
simple (Ga,Mn)As digital ferromagnetic heterostructures [60]. These systems
are grown by incorporating submonolayer planes of MnAs into a GaAs host.
They show ferromagnetic transition temperatures up to 50 K [61,62], the
anomalous Hall effect [62] and remarkably square hysteresis loops with higher
coercivities [61] than their random alloy counterparts. A confident modeling
of these alternative ferromagnetic semiconductors with controlled Mn distri-
bution would significantly contribute to our understanding of magnetic and
transport properties of (III,Mn)V ferromagnets.
In this chapter we have highlighted one particular approach to modeling
these materials. We anticipate that other approaches will also bring useful
insights. In particular dynamic mean-field-theory calculations [30] are able
to capture some of the Mn alloy disorder physics in a relatively simple way,
although they do not capture spatial correlations and would not capture
non-collinear states. Similarly first principles calculations [18] are likely to
prove extremely valuable in the future, once LDA+U or dynamical-mean-
field theory corrections have been applied to treat the Mn d-electron local
moments more accurately.
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,
A. Burkov, A. Chattopadhyay, A.1. Chudnovskiy, T. Dietl, J. Fernandez-

Rossier, J. Furdyna, E.G. Gwinn, N.A. Hill, A.J. Millis, H. Ohno, F. von
Oppen, N. Samarth, D. Sanchez, S. Sanvito, S. Das Sarma, P. Schiffer,
M. van Schilfgaarde, and C. Timm for many useful discussions. The work
was supported by the Indiana 21st Century Fund, the Welch Foundation,
DARPA, DOE, Deutsche Forschungsgemeinschaft, the EU COST program,
and the Ministry of Education of the Czech Republic.
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)
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
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
L. Sandratskii
5.1 Introduction
In recent years, noncollinear magnetism has attracted constantly increasing

interest from the density-functional-theory (DFT) community (see, e.g. [1-
19]). These studies revealed a crucial role played by noncollinear magnetism in
understanding the ground state magnetic properties, excitations, transport
properties, and spin-dynamics in various magnetic systems. In the present
chapter, we mainly focus on one aspect of the physics of noncollinear mag-
netism: the interplay between the symmetry of a magnetic system, relativis-
tic interactions and magnetic structure. The importance of the relativistic
interactions in physics of magnetism clearly follows from the fact that the
magnetocrystalline anisotropy, one of the properties most important for appli-
cations, and the orbital magnetism in solids are consequences of the spin-orbit
coupling. Analysis of the symmetry properties allows us to make predictions
important for understanding the physics of the system.
5.2 Density Functional Theory

of a N oncollinear Magnet
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
the magnetization is reproduced in the DFT calculations. In this sense, the

states with collinear magnetization are always self-reproducible and can be
considered as a possible ground state of a nonrelativistic system.
In the relativistic case, the neglect of the off-diagonal elements of the po-
tential matrix does not lead to the spin-factorization of the problem because
the SOC does not commute with the operator of spin projection. Therefore,
the matrix form of the KSE cannot be reduced to two separate equations
for the states with two opposite spin projections. The electron states have
a general spinor form
and the direction of the magnetization corresponding to these states varies

from point to point.
Physically, it is useful to distinguish between interatomic and intraatomic
magnetic noncollinearities. Under interatomic noncollinearity we understand
the noncollinearity of the atomic magnetic moments obtained by the integra-
tion of the magnetization over the atomic volumes. On the other hand, the
intraatomic magnetic noncollinearity is the noncollinearity of the magnetiza-
tion within the volume of a given atom.
Since the relativistic KSE always has a matrix form, the intraatomic mag-
netization of a relativistic system is always noncollinear [4]. The intraatomic
magnetic noncollinearity of relativistic systems does not necessary lead to
the noncollinearity of the moments of different atoms. If the intraatomic
exchange interaction is stronger than the SOC, the neglect of the intraatomic
noncollinearity of a relativistic system is a good approximation [21]. In this
chapter we will mostly be interested in the aspects of the interatomic non-
collinearity.
The calculational scheme used in the main body of the calculations dis-
cussed in this chapter is described in [11,22,23]. An important feature of this
scheme is an allowed variation of the directions of the atomic moments from
iteration to iteration: The eigenstates of the Kohn-Sham Hamiltonian (KSH),
'IjJ, are used to calculate the density matrix
(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
into account, the condition of self-consistency is imposed on the magnetic

density. According to the basic theorems of the DFT the ground magnetic
state corresponds to a self-consistent magnetic density [20].
5.3 Relation Between Symmetry and Stability

of Magnetic Structures
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.
5.3.1 Symmetry Constraint: General Formulation

We begin with the proof of the statement that the symmetry of the initial
Kohn-Sham Hamiltonian is preserved in the iterational DFT calculations. Let
us assurne that the initial KSH of the problem commutes with the operators
of group G and show that the density matrix, obtained with the use of
the solutions of the Kohn-Sham equation, is invariant with respect to the
operators of G. The concrete form of the KSH is not important here. For
nonrelativistic problems the operations are of the {as !aR!t} type where as
is a spin rotation, aR is aspace rotation, t is aspace translation [11]. (In this
chapter we do not distinguish between nonrelativistic and scalar-relativistic
cases, since from the symmetry viewpoint they are equivalent.) In the case
of relativistic problems as is always equal to aR and the operators are of the
{aR!aR!t} == {aR!t} type [11,28]. In both cases these transformations can be
accompanied by time reversal.
216 L. Sandratskii
According to the basic theorems of quantum mechanics, if operator g

commutes with Hamiltonian iI and 'l/J is an eigenfunction of iI corresponding
to eigenvalue c, then g'l/J is also an eigenfunction corresponding to the same
energy. As a consequence, all eigenstates of iI can be separated into the
subsets such that the states of one subset correspond to the same energy
and form a basis of an irreducible representation of G. The contribution to
the density matrix (5.1) of any such subset is invariant with respect to the
operations of G. Indeed,
(5.2)
LI LI
/1-1) LI LI
Here Dj is the jth irreducible representation of G.

Since
( r) =~( n(r) + mz(r) -imx(r) + my(r)) (5.3)
p 2 imx(r) + my(r) n(r) - mz(r) ,
the invariance of the p matrix immediately means the invariance of the par-
tide density n and spin magnetic density m with respect to g. Therefore the
effective potential
m(r)
v(r) = vo[n(r)] + Llv[n(r), m(r)] 0"' Im(r)1 ' (5.4)
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
J dr(p(r)(gn(r) - n(r)) + b(r)(gm(r) - m(r)).
This term contains Lagrange parameters p( r) and b( r ), which play the role
of external fields stabilizing the constrained state.
Aremarkable feature of the symmetry constraint is that the state provid-

ing the minimum of the functional under the symmetry restriction does not
need a nonzero stabilizing external field. This follows from the property that
the symmetry of the KSH and densities is preserved in calculations.
The property that a symmetry-constrained state does not need an external
stabilizing field is of exceptional importance since only such states can be the
ground state of the system. This property permits cardinal simplification of
the calculation of the ground state if the experimental data and theoretical
considerations show the presence of certain symmetry in the system. Note,
that the DFT allows, in principle, to begin calculations with a random mag-
netization and, by carrying out the iterational process to self-consistency, to
determine the magnetic state with the minimal energy. A highly symmetrical
ground state can be established in such calculations since, opposite to the loss
of symmetry, an increase of symmetry in the DFT calculations is possible.
These calculations are, however, extremely complex and time consuming even
for the simplest magnetic systems. Therefore, the symmetry constraint is an
efficient tool in the DFT studies of magnetic systems.
5.3.2 Two Types of Symmetry Constraints
We will distinguish two types of symmetry constraints. To introduce them we

consider in more detail the restrictions imposed on the magnetization by the
condition that the magnetization is invariant with respect to the operations
of group G.
Since, on the one hand, the symmetry operation g transforms the magnetic
density m( r) and, on the other hand, leaves it invariant, the magnetization
must fulfil the following condition [53]
(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
Fig.5.1. Symmetry constraints 1 and 11 (schematic picture). The continuous pa-

rameter 0 describes different magnetic configurations. (a) The state with 0 = Oe
corresponds to the symmetry constraint I. This state possesses additional regular
features compared with the states with 0 =I- Oe (b) Symmetry constraint 11. The
states with different 0 possess the same regular features. The probability that Oe
accidentally coincides with Omin is negligible
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
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.
5.4 Stable Magnetic Structures
First, we consider examples of magnetic structures distinguished by sym-

metry relative to the structures obtained by infinitesimal variation of the
directions of atomic moments.
220 L. Sandratskii
5.4.1 Simple Standard Cases

With rare exceptions, the DFT calculations reported in the literature are
performed under a symmetry constraint. Historically, the first calculations
were carried out for the nonmagnetic state of the systems. The magnetic
density was assumed to be zero at each point in the space. The study of
magnetically ordered systems began with the collinear ferromagnetism of
elementary met als, like Fe and Ni, and of the two-sublattice collinear anti-
ferromagnetism of Cr [31].
It can be easily shown that the regularities characteristic to all three
simplest magnetic states correspond to the symmetry constraint and, indeed,
must be stable in the calculations. The stability of the zero value of the
magnetic moments in the nonmagnetic state is a consequence of the invariance
of the KSH with respect to the time reversal. The stability of the equal
values and parallel directions of the atomic moments in Fe and Ni are the
consequences of the translational symmetry. The stability of the equal values
and antiparallel directions of the magnetic moments of two sublattices in
Cr are the consequence of the symmetry operation that combines a lattice
translation connecting two sublattices and the time reversal. Any disturbance
of the characteristic features of these magnetic states leads to the loss of the
invariance of the KSH with respect to the corresponding symmetry operation.
5.4.2 Magnetic Structures of U 2 Pd2 Sn and USb

The crystallographic unit cell of the U2 Pd2 Sn compound is shown in Fig. 5.2.
The magnetic state of U 2 Pd 2 Sn was established in [32J where six magnetic
structures, shown schematically in Fig. 5.2, were considered in the analysis of
the neutron diffraction data. These six structures were selected on the basis
of the symmetry arguments of the Landau theory of the second-order phase
transitions. The noncollinear magnetic configuration NC1 was found to be
the magnetic ground state of the system.
We performed the first-principles DFT calculations for all six magnetic
configurations. All of them correspond to the symmetry constraint I and are
stable in the calculations. For example, the magnetic structure NC1 possesses
the following symmetry operations: (a) the 1800 rotation about the z-axis, (b)
the 1800 rotation about the x = y-axis, (c) the 90 0 rotation about the z-axis
accompanied by inversion, (d) the 900 rotation about the z-axis accompanied
by time reversal. (Only the generators of the symmetry group are given.) If
the coordinate center is chosen at the position of the central Sn atom (Fig. 5.2)
operation (b) is accompanied by nonprimitive translation a(0.5, 0.5, 0). Other
operations are pure point transformations. Any infinitesimal deviation of the
magnetic moments destroy the invariance of the system with respect to at
least one of the symmetry operations. This additional symmetry of the NC1
structure is illustrated in Fig. 5.3, where we show the total energy of the
magnetic structures obtained by the rotation of the NC1 structure by different
...<~---- 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>
NC4 NCI NC4
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
Fig. 5.4. Triple-k structure in USb
5.5 Relativistic Instability of Collinear Ferromagnetism

In nonrelativistic systems, collinear magnetic configurations are always stable
in the calculations. The symmetry operations responsible for the stability of
this regularity are the pure spin rotations about the magnetization axis [11] .
In the relativistic case, such operations do not commute with the Hamiltonian
of the problem and cannot be symmetry operations of the system. The pure
spin rotations are not the only symmetry operations that can be responsi-
• •
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.
5.5.1 Magnetic Structure of U 3 P 4
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].
5.5.2 Atomically Disordered Relativistic Systems

Another example of the system with relativistic instability of collinear mag-
netic state is an atomically disordered relativistic system. According to the
symmetry principles formulated in Sect.5.3.1 the magnetic structure oi rel-
ativistic systems with atomic disorder is always noncollinear. Indeed, in the
presence of atomic disorder there is no spatial transformations that leave the
atomic positions invariant. Since the spin-orbit coupling connects the atomic
and magnetic subsystems, a separate transformation of these subsystems is
not allowed. Correspondingly, the system possesses no symmetry operation
that can be responsible for the collinearity of the atomic moments. This leads
to the noncollinearity of the magnetic structure.
To verify this conclusion, the following DFT calculations have been per-
formed. First, undistorted bcc Fe was considered. In this case the collinear
ferromagnetic structure is stable for both relativistic and nonrelativistic cal-
culations (see Sect. 5.4.1). At the next stage, a supereell containing 8 atoms
was constructed and the atoms were shifted from their positions in the bcc lat-
tice (Fig. 5.6) by different vectors collected in Table 5.1. These shifts destroy
the symmetry operations of the bcc structure that transform the atoms of
the super cell into one another. As a result, there is no symmetry operation
that can be responsible for the stability of the collinear directions of the
magnetic moments of any two atoms in the super cell. Therefore, according
to the symmetry analysis of Sect. 5.3.1 each of the 8 atomic moments must
deviate from the initial direction. These deviations must be different for each
of the 8 atoms.
The calculations confirmed these predictions. At the beginning, all mag-
netic moments were directed parallel to the z-axis (Fig.5.6). Already the
first iteration resulted in different deviations of the moments of all 8 atoms
from the initial direction. The self-consistent deviation angles are collected
in Table 5.1. It is important that not only the spin moments of different
atoms deviate differently but also the orbital and spin moments of the same
atom assurne different directions. This property is another consequence of the
loss of the symmetry in the system. The collinearity of the spin and orbital
moments of the same atom is a regularity that can be stable only if it is
supported by a symmetry operation. In the case of symmetry constraint II,
the directions of the spin and orbital atomic moments are always different.
Fig. 5.6. Atomic disorder leads to the noncollinearity of the magnetic structure.
Broken arrows show the initial collinear structure
Table 5.1. Noncollinear magnetic structure of distorted bcc iron: self-consistent

relativistic calculations. Atomic positions and shifts are given in the units of the
bcc lattice parameter, the deviation angles in degrees
Deviation of atomic moments

bcc position shift spin (),4> orbital (), 4>
(0,0,0) 0.81,188.8 1.63,215.6
e2'2'2
1 1) (0.01,0.02,0.03) 0.79,193.6 0.96,206.8
(0,0,1) (0,0.01,0.01 ) 0.69,191.1 0.28,78.7
e2'2'2 1 3) (-0.02,0, -0.01) 0.73,188.8 1.26,309.2
(0,1,0) (0, -0.03,0) 0.83,193.2 1.37,180.4
(i,~,i) 0.81,194.5 1.48,233.8
(0,1,1) 0.72,195.4 0.93,78.4
e2'2'2
3 3) 0.77,189.6 1.04,325.8
5.6 Relativistic Instability

of Collinear Antiferromagnetic Structures
Next we discuss two examples of the relativistic instability of collinear anti-
ferromagnetic structures.
5.6.1 Weak Ferromagnetism in IX-Fe20a

We begin with the phenomenon ofweak ferromagnetism in hematite, IX-Fe203
[34]. Dzialoshinski [35] suggested a model Hamiltonian
(5.9)
226 L. Sandratskii
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
.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
with the experimental observation of a collinear antiferromagnetic structure

with moments oriented parallel to the z-axis below 260K [38J.
The situation is changed completely when the moments are parallel to
the y-axis arranged again in the sequence (+ - -+~ [39J. Now the following
generators are left in the symmetry group of the KSH: the 180 0 -rotation
about the x-axis and the inversion. Inversion transforms the atoms of the
upper Fe2ü3 molecule into the atoms of the lower molecule, see Fig. 5.7.
Since the magnetic moments are axial vectors, they do not change under
this transformation. Hence, corresponding atoms of two molecules must keep
parallel moments and one may restrict the consideration to the lower molecule
in Fig. 5.7. The only condition imposed on the moments of the Fe atoms by
symmetry is the transformation of the moment of atom 1 into the moment
of atom 2 by the rotation through 1800 about the x-axis. However, to fulfil
this condition it is not necessary for the atomic moments to be parallel to the
y-axis nor to remain collinear. Indeed, calculations show that the magnetic
228 L. Sandratskii
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.
5.6.2 Relativistic Instability of the Collinear Antiferromagnetism

in UPdSn
Much attention was recently attracted by two magnetic phase transitions [40-
44] in UPdSn. The paramagnetic state has the hexagonal GaGeLi crystal
lattice. Below 45 K UPdSn becomes magnetic with a noncollinear antifer-
romagnetic structure. In this phase (which we will refer to as structure I)
all magnetic moments of the Uranium atoms lie parallel to a plane and
compensate one another completely. Simultaneously, orthorhombic lattice
distortions are detected. At 20 K, a second phase transition is observed. Here,
the magnetic structure (structure H) is still noncollinear and compensated,
however, the magnetic moments deviate from the plane, developing compo-
nents perpendicular to it.
The calculations [45] were started with the undistorted hexagonal lat-
tice and uranium magnetic moments directed along the z-axis and forming
a collinear antiferromagnetic structure (see Fig.5.8a). Allowed to rotate,
the moments deviated immediately from the z-axis keeping, however, their
equivalence and the compensated character of the magnetic structure. The
resulting self-consistent directions of the magnetic moments are shown in
Fig.5.8a; they form a magnetic structure that is very similar to the exper-
imental structure I. The symmetry principle, formulated above helps us to
expose the physical reasons for the instability of the initial collinear structure.
The symmetry operations of the initial collinear antiferromagnetic struc-
ture are collected in Table 5.2. The analysis of the restrictions imposed by the
symmetry operations on the directions of the magnetic moments shows that
m~=O for each atom i. For the other two components the following conditions
must be satisfied·. m y1 = m 2y = -m3y = _m 4y and m 1z = _m 2z = m 3z = _m4z·
This means that the initial collinear structure has the same symmetry as
a noncollinear structure satisfying these relations. The noncollinear magnetic
structure detected experimentally belongs to this type of structure. Thus, we
again deal with the case of symmetry constraint H.
Next, the infiuence of lattice distortions on the magnetic structure was
studied [45]. First, following the experiment [42] we introduced a small vari-
ation of the lattice parameters a and b such that the relation b = v'3a, valid
for the ideal hexagonallattice, is no longer satisfied. This distortion does not
(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
Table 5.2. Symmetry properties of the orthorhombic UPdSn
e = ( \ 1); C2z = (-1 -1 1) ; ax = (-\ 1); ay = (1 -1 1)
'Tl = ~(a,b,c); 'T2 = !(a,b,O); 'T3 = !(O,O,c) ;R time reversal operation
Operation Transposition of U atoms Restriction on U moments
c
1 {eIO} no no
2 {C2zl'TI} 1++4; 2++3

(::) •
m
-my, )
mz .
J
; i ++ j
3 {axl'T2} 1++3; 2++4

(:J (::) ; i ++ j
(:J c:')
-mz .
• J
4 {ayl'T3} 1++4; 2++3 ; i ++ j
-mz .
• J
5 {el'T2}R 1++3; 2++4 Tni = -Tnj ; i ++ j
6 {C2zl'T3}R 1++2; 3++4

(:) (3J
mz .
• J
; i ++ j
7 {axIO}R no mx =0 all i
8 {ayl'TI}R 1++4; 2++3

(:) (~)
mz .
• 2
; i ++ j
tures of UPdSn: in the monoclinic structure there is no symmetry operation

demanding the x component of the magnetic moments to be zero. This means
that a deviation of the magnetic moments from the yz plane does not change
the symmetry of the system and therefore will take place according to our
symmetry principle. Thus, the result of the calculation for the monoclinically
distorted lattice and the corresponding symmetry analysis are in agreement
with the experimental data.
5.7 Relativistic Instability of a Compensated

Noncollinear Magnetic Structure in Mn3Sn
The magnetic structure in Mn3Sn is very elose to compensated triangular.

There is, however, a weak ferromagnetic component [46J.
In an earlier paper, Sticht et al. [22J studied a number of magnetic config-
urations allowed by Landau's theory of phase transitions, but SOC was ne-
glected. These calculations showed that, compared to other magnetic states,
the triangular configurations have a distinctly lower total energy. Four of the
triangular structures are represented in Fig. 5.9. Calculations without SOC
showed that all four configurations are equivalent. Another property of all
four configurations is their stability during iterations.
The account for SOC changes the situation drastically [7J. First, the de-
generacy is lifted, i.e. all four magnetic configurations become inequivalent
and, second, for two of them (c and d) the magnetic moments deviate from
the initial directions, as depicted in Fig.5.9. The symmetry analysis shows
that structures a and b correspond to symmetry constraint I and are stable
in the calculations. Structures c and d correspond to symmetry constraint 11
0 • 0
• • 0 0 •
• ~.
q:./.. ::.::p • 0 •
0
e·:, .: .. : .:'::.
0 0 • Mn Sn
• 0 • z=1/4
z=3/4
•
0
•
0
(a) (b) (c) (d)
I~ I I
0~" ~.:. . . !.:

2~:"\~3 .... -: ........,
'>:: "-:.:' 2 ./..
~
. 3
2• . :,..:.... , ............ .
~ ··\t::·:··~ ~/. 3 "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
and are unstable. It is interesting that the weak ferromagnetism in Mn3Sn is

different from the weak ferromagnetism in Fe2ü3 and cannot be attributed to
the anisotropie exchange interaction. The symmetry principle of the magnetic
instability is, however, universal and can be applied to both cases of weak
ferromagnetism.
5.8 Nonmagnetic Sublattices

in Antiferromagnetic Systems
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.
5.8.1 Intersublattice Interaction in UFe4Als
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
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"
Operation Transposition Restriction on magnetic moments

of Fe atoms of U and Fe atoms
1++3;2++4
(::)
mz
(-~x)
-mz.
J
I no no
1++4;2++3
the nonmagnetic state of the U atoms can be illustrated by the properties of

the total energy as a function of m y (Fig. 5.11). The most important feature
of the E(m) curve is its asymmetry, which leads to an accidental position of
the minimum at a nonzero value of m y .
The orthogonality of the inducing and induced moments is also closely
connected with the symmetry properties of the system since the appearance
of a nonzero m x or m z component of the U moment would disturb at least
one of the symmetry operations of the BS.
0.5
~E
- - - mFella
ur
--------- mFe IIc
0.0 a.......:::::..-'-.-::"--~ _ _'_~ __ =...L-_ __'

0.0 0.8
m
Fig.5.11. Total energy as a function of the U spin moment for the Fe moments
collinear to the a- (basic state) and c-axes. For the Fe moments parallel to the c-axis
no induced U moment appears
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.
5.8.2 Magnetic Structure of UX a Compounds

In [49] we have shown that the relation between the inducing and induced
moments that follows from the symmetry analysis discussed in the previous
seetions can be helpful in the determination of the magnetic structure of
the inducing subsystem. In UX3 neutron diffraction did not provide com-
plete information about the magnetic structure of the U sublattice. Since,
according to our symmetry principles, different directions of the U magnetic
moments result in different properties of the induced moments of the X atoms,
the study of the induced moments provides information about the magnetic
structure of the U sublattice. Measuring the hyperfine field on the positions
of X atoms with, for example, the perturbed-angular-correlation technique
allows us to draw the conclusions about the induced moments. Figure 5.12
illustrates the properties of induced moments of the X atoms for different
orient at ions of the U moments. The reader is referred to [49] for full details
of the study of the magnetic structure of UX 3 compounds.
5.9 Helical Structures in Systems

A wide variety of systems has been reported to possess a helical magnetic

order with the periodicity incommensurate with the periodicity of the crystal
lattice (see, e.g. [50,51]). However, almost all DFT calculations of helical
magnetic configurations have been carried out for one system: I'-Fe [52,53].
(For an exception, see the recent study on the rare-earth met als [54].) The
reasons for the concentration on I'-Fe are connected with a negligible role of
the SOC in the formation of the magnetic structure in this system. Here, we
236 L. Sandratskii
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.
5.9.1 Magnetic Structure of UPtGe
In accordance with the experimental situation we consider [27J plane helical

magnetic structures of the form
e~ = (cos(qa v + cPo), 0, sin(qa v + cPo)) , (5.10)
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
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]),
which, in addition to the exchange interaction contains the Dzyaloshinski-

Moriya interaction (DMI) D(8 1 x 8 2 ). Although the DMI is a consequence of
the SOC, the simultaneous rotation of all atomic moments about the direction
of the D vector does not change the energy of the system. This property of
the DMI allows the symmetry of the system with respect to the generalized
translations to be maintained, which is a necessary condition for the stability
of helical structures. On the other hand, the DMI term disturbs the symmetry
of the E(q) curve with respect to the reversal of q and leads to the shift of
the minimum of the total energy to an incommensurate q.
The DMI is not, however, the only interaction resulting from the SOC.
Another term, the magnetic anisotropy [36,59] (MA), destroys the symmetry
of the problem with respect to the generalized translations. Therefore, to
allow the formation of the helical structure of the MnSi type the SOC must
be present, but the magnetic anisotropy must be smaller than the DM!. Since,
under certain conditions [36], the DMI is proportional to the SOC, and the
MA to the square of the SOC, these requirements can be fulfilled in the case
of weak SOC, e.g. in MnSi. The long period (180 A) of the helix in MnSi is
a reßection of the weakness of the relativistic effects.
Since for both types of helices the smallness of the magnetic anisotropy
is important, we studied this quantity and compared its value with the value
of the interatomic exchange interactions. To estimate the MA we compare
the total-energy differences between the magnetic structures with the same
relative orientation of the atomic moments but different orient at ion with re-
spect to the crystallattice (Table 5.4). The interatomic exchange interaction
is estimated by comparison of the total energies of the ferromagnetic and
antiferromagnetic configurations with atomic moments collinear to an axis.
Calculations show that the MA in the xz plane is very small. Since the
SOC is strong and there is no symmetry reason for the smallness of the
inplane anisotropy this property results from an accidental compensation
of the contributions of different electron states. An estimation of the ex-
change interaction shows that it is much stronger than the inplane magnetic
anisotropy [27].
The weakness of the inplane magnetic anisotropy is a necessary condition
for the formation of the helical structure. It does not, however, provide the
physical mechanism leading to the formation of the helix. In Fig.5.15 we
show the q dependence of the total energy that was calculated neglecting the
SOC. A elear trend to the formation of the helix with an incommensurate q
value elose to 0.5 is obtained.
On the basis ofthe Fourier transformation ofthe E(q) curve the following
qualitative picture of the magnetism of UPtGe was suggested. The U posi-
tions in UPtGe form a structure rather elose to a simple hexagonal (Fig. 5.16).
The effective exchange interaction between the magnetic moments of atoms
o and 2 is strong and antiferromagnetic. The exchange interaction of the
magnetic moments of atoms 1 and 3 with the moments of atoms 0 and 2 is
240 L. Sandratskii
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
0.0 0.5 1.0

q
Fig. 5.15. Energy as a function of q = (q, 0, 0)
much weaker. Evidently the antiparallel directions of the moments 0 and 2

lead to the frustration of the direction of the moments 1 and 3. The frustrated
magnetic interactions are the reason for the minimum of the total energy at
an incommensurate q value (Fig.5.15). This result suggests that the helical
magnetic structure in UPtGe should be considered as an exchange helix.
One furt her remarkable experimental feature of UPtGe is, however, the
observation of the domains of only one chirality [58]. This property is charac-
teristic of the relativistic helices of the MnSi type [60]. In [27] we have shown
b)
- q
Fig. 5.16. Frustrated magnetic interactions. (a) Ihl « Ihl. (b) Helical magnetic
structure as a consequence of the frustration
0.0
- 1.0 L...-_ _ _ _ _- , -_ _ _- , -_ _-...J

~
-1 .0 -0.5 0.0 0.5 1.0

q
Fig. 5.17. E(q) without DMI (dotted line) and with DMI (solid line) . See [27] for
all details
that the inequivalence of the domains of opposite chirality in UPtGe is a result

of the SOC which leads to a substantial DMI in the system. Figure 5.17 shows
the total energy as a function of q evaluated with and without the DMI
inter action. Summarizing, an accidentally small inplane magnetic anisotropy
provides a necessary condition for the formation of the incommensurate helix.
The formation and properties of the helix are determined by the frustrated
exchange interactions and relativistic DM!. The magnetic structure of UPtGe
cannot be classified either as a pure exchange or pure relativistic helix, but
possesses the features of both.
5.9.2 Helices in REM

A rich variety of complex magnetic configurations was experimentally found
in the heavy REM [51,61]. An important contribution to the understanding of
242 L. Sandratskii
the magnetic properties of heavy REM is made by Jensen and collaborators

(see the book [51] and references therein and later publications, e.g. [62])
who used a model spin-Hamiltonian to describe peculiar magnetism in these
systems. (See also [63] far an earlier phenomenological theory of the magnetic
ordering in REM.)
In contrast to the model-Hamiltonian approach, the contribution of the
DFT to the study of the complex magnetism in heavy REM is very modest.
Most of the DFT calculations for REM were performed for the collinear fer-
romagnetic structure of Gd. To the best of the author's knowledge, only two
direct first-principles DFT calculations of complex magnetic configurations in
heavy REM were reported. Nardström and Mavromaras [54] used the scalar-
relativistic approximation to study the q dependence of the total energy
of planar spiral structures. Here, q is the propagation vector of the spiral.
The E(q) curves were compared with the Fourier components of the inter-
atomic exchange parameter J(q) determined experimentally. Perlov et al. [64]
employed a scalar-relativistic approximation to calculate J(q) byexamining
the conical spiral configurations. No studies of the influence of the SOC on
the magnetic configurations of heavy REM have been performed within the
framework of the DFT. The success of the DFT in the investigation of the
magnetic properties of solids and recent developments in the computational
techniques and facilities make the complex magnetism of the heavy REM one
of the important topics for nearest-future studies. Combination of the model-
Hamiltonian and first-principles DFT approaches should provide a new level
of the theoretical description of REM magnetism.
It is not a purpose of this chapter to report a detailed DFT study of the
magnetism of concrete REM. Rather we aim to provide one example of the
usefulness of the symmetry analysis and relativistic DFT calculations in the
studies of the REM.
In the calculations, the 4f states were treated as pseudocore [65] states
and did not hybridize with the valence electron states. A scalar-relativistic
approximation was used in the description of the core states. The SOC was
considered for the valence electrons only. The neglect of the SOC in the 4f
states is a severe approximation in the physical model describing the effects of
the magnetic anisotropy in REM. For example, the SOC in the 4f states plays
an important role in the description of the magnetic properties of the 4f met-
als in terms ofthe model crystal-field Hamiltonian [51]. Neglecting the SOC in
the 4f states we can expect that the strength of the magnetic anisotropy will
be substantially underestimated. To simulate a stronger magnetic anisotropy
within the given calculational scheme in some cases we performed calculations
with the SOC enhanced by a factor of 20.
Several REM were reported to possess a helical magnetic structure. Thus,
a planar helix is observed in certain temperature intervals in Tb, Dy, Ho, and
Er. A ferromagnetic helix (cone structure) is observed in Ho and Er.
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,
type C2a: (::) (~r::b) ; type C2b: (::) (~~a) ;

mCj mCi mCj mCi
type C2c: (::)

mC j mc
( =::) ;
i
type Tnj R:-Tni. Here i and j according
to the column "Transposition". Atom i is transformed to atom j under the action

of the symmetry operation. For atoms invariant under the action of the symmetry
operation, j = i
Magnetic nat Operation Transposition Restriction on

structure magnetic moments
helix, ab-plane 8 C2b 2 ++ 8; 3 ++ 7; 4 ++ 6 ; type C 2 b

(Fig. 5.18a) R(O, 0, 2) 1 ++ 5 ; 2 ++ 6; 3 ++ 7; 4 ++ 8 ; type R
cycloid, bc-plane 8 type C 2b
(Fig. 5.18b) Ua(O, 0, 2) 1 +-t 5 ; 2 +-t 6 ; 3 +-t 7; 4 +-t 8; type C2a
R(O, 0, 2) 1 ++ 5 ; 2 +-t 6; 3 +-t 7; 4 +-t 8 ; type R
cycloid, ac-plane 8 uc(O, 0, 2) 1 +-t 5; 2 +-t 4; 6 ++ 8 ; type C2c
R(O, 0, 2) 1 +-t 5 ; 2 +-t 6; 3 +-t 7; 4 +-t 8 ; type R
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.
To understand the nature of the wobbling of the experimental vertical

structure we performed the symmetry analysis for a magnetic configuration
shown in Fig. 5.18b but, in this case, parallel to the ac plane. The symmetry
of this structure preserves (i) the directions of the atomic moments 3 and 7,
(ii) the zero c component of the moments 1 and 5, and (iii) the compensated
character of the structure as a whole. Moments 2-4 and 6-8 deviate from the
ac plane leading to the wobbling observed experimentally. Note that a model
spin-Hamiltonian that contains only the terms of the second-order with re-
spect to atomic spins: the Heisenberg exchange interaction and the single-
site magnetic anisotropy, fails to describe the wobbling. The fourth-order
'trigonal' interactions must be added [62]. In the magnetic and relativistic
DFT calculations these and higher-order interactions are automatically taken

into account.
Since the magnetic anisotropy is very sensitive to the details of the theo-
retical model, future systematic DFT studies of the REM magnetism should
consider such effects as polar magnetic interaction of atomic moments and
lattice distortion caused by magnetoelastic interactions [63]. The account
for the SOC in the 4f states is of great importance. Another important
direction for the improvement of the calculational scheme is a better account
for the correlation effects in the 4f states. Here, self-interaction corrections,
[67] orbital polarization corrections [68] or an LDA+U [69] scheme should
be considered as possible approaches. Combination of these improvements
should make possible a first-principles quantitative description of the delicate
balance of different interactions traditionally described in terms of a model
crystal-field Hamiltonian [51]. Detailed DFT study ofthe magnetism ofheavy
REM with account for the SOC and noncollinearity of the magnetic structure
is an exciting topic for the nearest-future researches. The symmetry analysis
reported here preserves its validity also for more elaborate physical models.
We hope that the present symmetry analysis and results of numerical cal-
culations will stimulate further studies of the complex magnetism in REM
systems.
5.10 Intraatomic Magnetic Noncollinearity
We will briefly comment on two important aspects of the intraatomic mag-

netic noncollinearity. The first aspect is the spatial variation of the direction
of magnetization within the volume of an atom [4,6,21,70,71] (Fig.5.19).
As already mentioned in Sect.5.2, in any relativistic magnetic system the
intraatomic spin density is always noncollinear. In terms of symmetry argu-
ments, this is an inevitable consequence of the property that the spin projec-
tion on any selected axis cannot be a good quantum number characterizing
electron states. Also, the orbital intraatomic magnetization is noncollinear
[72], though the concrete form of the orbital magnetization depends strongly
on the gauge chosen. In nonrelativistic systems, a collinear intraatomic spin
density is possible if the magnetic moments of different atoms are collinear.
Another type of intraatomic magnetic noncollinearity is the noncollinear-
ity of the spin and orbital magnetic moments of the same atom [5,56,73]. The
collinearity of both atomic moments must be considered as a regularity in
the system that, according to our symmetry principle, is possible only in the
case that there is a symmetry operation that is responsible for this regularity.
In all cases where interatomic magnetic noncollinearity is predetermined by
symmetry (symmetry constraint II) the atomic spin and orbital moments are
also noncollinear. (See, e.g. [5,56,73] for calculations of concrete systems.)
246 L. Sandratskii
If .,. ".
,
~
If ,..
~
~ -r. ~
Fig.5.19. Intraatomic spin magnetization in a U atom in U3Bi4 obtained in self-

consistent (black arrows) and nonself-consistent (grey arrows) calculations. See [21]
for all details
5.11 Relativistic Spectroscopy

of N oncollinear Magnetic States
The SOC plays a crucial role in the magneto-optical Kerr effect of magnetic
systems. Here, we briefty discuss the first DFT study of the magneto-optical
Kerr effect (MOKE) of a noncollinear system [74J. (See also [77J for a re-
cent relativistic calculation of photoemission and X-ray absorption spectra
of systems with noncollinear magnetic configuration. )
The calculation was performed for the U3P 4 compound (see Sect. 5.5.1)
for different values of the cone angle (Fig.5.5). Both optical and magneto-
optical properties of a system can be deduced from the optical conductivity
tensor a(w). The expression for a(w) contains interband and intraband con-
tributions. The interband part can be represented in the following form
inte r ( ) = ~ '" '" f(En,k) - f(Em ,k)

ao:,ß w m2 V ~ ~ E - E
n,m
kEl.BZ m,k n,k
n;>!m
< n,klpßlm, k > < m,klpo:ln,k >
x E m,k -En ,k + i8 ' (5.11)
W -
1;
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
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 '
The simple expression for the polar Kerr angle,
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
R (1) <l>K (deg)

1.0 6
- - 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
that is obtained in our calculations can be treated as a natural consequence

of the symmetry properties of the magneto-optical effect.
Still, the very high stability of the reflectivity as a function of canting and
the simple scaling of the Kerr-rotation peak with respect to the z projection
of the magnetic moment are nontrivial. The very weak dependence of the
reflectivity on the magnetic structure for the whole range studied can be inter-
preted as follows. The electron states enter the optical conductivity through
the eigenenergies and the matrix elements involving the eigenfunctions, both
of which can thus depend only weakly on the change of the directions of the
atomic moments. This means, first, the U 5f states - as seen from the loeal
atomic coordinate system having the quantization axis parallel to the atomic
moment - are almost unchanged for any magnetic configuration. Second, the
hybridization of the U 5f states with the nonmagnetic valence states, e.g.
P 3p states, does not change substantially with rotation of the U 5f states.
These properties result in the weak dependence of the energy values and near
invariance of the transition probabilities.
Thus we deal with the case of a well-defined magnetic U moment that
is formed by the itinerant 5f electrons: the moments can rotate without
substantially changing the 5f electron states as seen from the local atomic
reference system. This phenomenon is known from studies of the 3d elements
and their compounds [76]. This result is crucial for the understanding of
the properties of the compound at finite temperatures, in particular for the

Curie-Weiss behavior of the magnetic susceptibility [76].
5.12 Conclusion
Summarizing, we have shown that the spin-orbit coupling plays an important

role in the formation of the magnetic structure and properties of noncollinear
magnetic systems. The modern density functional theory allows the study of
these effects within a parameter-free calculational scheme. We have shown
that the analysis of the symmetry aspects of the problem is very helpful in
the predicting and understanding the results of the DFT calculation. On the
basis of the notion of the symmetry constraint we formulated a symmetry
principle of the stability of regular features of the magnetic configuration and
demonstrated the efficiency of this principle by application to very different
magnetic systems. It is to be expected that noncollinear magnetic states
will play an increasing role in the future of solid-state physics. Note, that
the modern engineered nano materials possess, as a rule, peculiar symmetry
properties. This gives the principles and methods discussed here an enormous
application potential in the new fields of magnetism.
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
14. A. Taga, L. Nordstrrn, P. James, B. Johansson, and O. Eriksson: Nature 406,

280 (2000)
15. M. Pajda, J. Kudrnovsky, I. Turek, V. Drchal, and P. Bruno: Phys. Rev. Lett.
85, 5424 (2000)
16. O. Grotheer, C. Ederer, and M. Fähnle: Phys. Rev. B 62, 5601 (2000)
17. B. Yaworsky, I. Mertig, A.Y. Perlov, A.N. Yaresko, and V. Antonov: Phys.
Rev. B 62, 9586 (2000)
18. P. Kurz, G. Bihlrnayer, K Hirai, and S. Blügel: Phys. Rev. Lett. 86, 1106
(2001)
19. D. Wort mann , S. Heinze, P. Kurz, G. Bihlrnayer, and S. Blügel: Phys. Rev.
Lett. 86, 4132 (2001)
20. U. von Barth and L. Hedin: J. Phys. C 5, 1629 (1972)
21. K Knöpfte, L. M. Sandratskii, and J. Kübler: J. Alloys Cornpd.: 309, 31 (2000)
22. J. Sticht, KH. Höck, and J. Kübler: J. Phys.: Condens. Matter 1, 8155 (1989)
23. L.M. Sandratskii and J. Kübler: Phys. Rev. B 55, 11395 (1997)
24. L.D. Landau and E.M. Lifshitz: Shape Statistical Physics, Part 1 (Pergamon,
New York 1980)
25. J-C. Toledano and P. Toledano: Shape The Landau Theory 0/ Phase Transi-
tions (World Scientific, Singapore 1987)
26. L.M. Sandratskii and J. Kübler: Phys. Rev. B 60, R6961 (1999)
27. L.M. Sandratskii and G. Lander: Phys. Rev. B 63, 134436 (2001)
28. L.M. Sandratskii: Sov. Phys. J. 22 941 (1979)
29. P.H. Dederichs, S. Blügel, R. Zeller, and H. Akai: Phys. Rev. Lett. 53, 2512
(1984)
30. B.L. Györffy, A.J. Pindor, J. Staunton, G.M. Stocks, and H. Winter: J. Phys.
F 15, 1337 (1985)
31. Deviation of the propagation vector of the rnagnetic structure of Cr frorn 0.5
was not taken into account in early calculations.
32. A. Purwanto, R.A. Robintson, L. Havela, V. Sechovsky, P. Svoboda, H.
Nakotte, K Prokes, F.R. de Boer, A. Seret, J.M. Winand, J. Rebizant, and
J.C. Spirlet: Phys. Rev. 50, 6792 (1994)
33. K Knöpfte and L.M. Sandratskii: Phys. Rev. B 63, 14411 (2001)
34. T. Srnith: Phys. Rev. 8, 721 (1916); L. Neel: Rev. Mod. Phys. 25, 58 (1953)
35. I.J. Dzialoshinski: Phys. Chern. Solids 4, 241 (1958)
36. T. Moriya: Phys. Rev. 120, 91 (1960)
37. L.M. Sandratskii, M. Uhl, and J. Kübler: J. Phys.: Condens. Matter 8 983
(1996)
38. F.J. Morin: Phys. Rev. 78, 819 (1950)
39. L. M. Sandratskii and J. Kübler: Europhys. Lett. 33447 (1996)
40. F.R. de Boer, E. Brück, H. Nakotte, A.V. Andreev, V. Sechovsky, L. Havela, P.
Nozar, C.J.M. Denissen, K.H.J. Buschow, B. Vaziri, M. Meissner, H. Malette,
and P. Rogl: Physica B 176, 275 (1992)
41. R.A. Robinson, A.C. Lawson, J.W. Lynn, and KH.J. Buschow: Phys. Rev. B
45, 2939 (1992)
42. R.A. Robinson, A.C. Lawson, J.A. Goldstone, J.W. Lynn, and KH.J.
Buschow: J. Magn. Magn. Mater. 128, 143 (1993)
43. R.A. Robinson, J.W. Lynn, A.C. Lawson, and H. Nakotte: J. Appl. Phys. 75,
6589 (1994)
44. R. Troc, V.H. Tran, M. Kolenda, R. Kruk, K Latka, A. Szytula, J. Rossat-
Mignod, M. Bonnet, and B. Büchner: J. Magn. Magn. Mater. 151, 102 (1995)
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
68. O. Eriksson, B. Johansson, and M.S.S. Brooks: J. Phys. Condens. Matter 1,

4005 (1989)
69. V. Anisimov, J. Zaanen, and O.K. Andersen: Phys. Rev. B 44, 943 (1991);
A.I. Liechtenstein, V.P. Antropov, and B.N. Harmon: Phys. Rev. B 49,10770
(1994)
70. D.M. Bylander and L. Kleinman: Phys. Rev. B 58, 9207 (1998)
71. H. Eschrig and V.D.P. Servedio: J. Comput. Chem. 20, 23 (1999)
72. M. Todorova, L.M. Sandratskii, and J. Kühler: Phys. Rev. B 63, 52408 (2001)
73. L.M. Sandratskii and J. Kühler: Europhys. Lett. 33,447 (1996)
74. J. Köhler, L.M. Sandratskii, and J. Kühler: Phys. Rev. B 55, R10153 (1997)
75. W. Reim and J. Schoenes: 'Magneto-optical Spectroscopy of f-electron Sys-
tems'. In: Ferromagnetic Materials, Vol. 5, eds. K.H.J. Buschow, E.P. Wohl-
farth (North-Holland, Amsterdam 1990) pp. 133-236
76. Metallic Magnetism. ed. hy H. Capellmann (Springer, Berlin Heidelherg New
York 1987)
77. H. Ehert, J. Minar, V. Popescu, L.M. Sandratskii, and A. Mavromaras: AlP
Conf. Proc. 514, 110 (2000)
6 Orbital Degeneracy and Magnetism
of Perovskite Manganese Oxides
r. Solovyev and K. Terakura
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
0.4 0.5 0.6 0.7 0.8

x
Fig. 6.1. Phase diagram ofNdl_xSrxMn03 ([10]). Abbreviations used in the figure
are defined in the text
Lal-xCaxMn03 at commensurate carrier concentrations (x = 2/3, 3/4, and

4/5) [14,15].
One of the basic microscopic mechanisms behind the fascinating proper-
ties of perovskite manganites is the double exchange (DE) , which is governed
by the strong atomic Hund's rule interaction between the localized t2g and
itinerant eg electrons. It favors the high-spin state of Mn sites and penalizes
the hoppings of the eg electrons with the change of their spin orientation
[6,16- 19]. Therefore, the DE interaction is always ferromagnetic and typically
considered in the combination with the AFM superexchange (SE) interactions
between the t2g spins [6]. The rich variety of magnetic structures observed
in manganites was one of the crucial factors that motivated many authors
to substantially revise the canonical DE picture by explicitly including other
interactions, mainly the lattice distortions [20,21], and the onsite [22] and
intersite [23] Coulomb repulsions. There is no doubt that the role played by
these ingredients in the physics of manganites is still on the agenda, especially
in the low-x and high-T regimes. It appears, however, that despite the long
history of research, many aspects of the DE physics itself are still barely
understood, especially taking into account the combination of the double
exchange mechanism with the degeneracy of the eg states [24] .
The purpose of this chapter is to discuss the low-temperature behavior
of manganites by proceeding from this very minimal model. The chapter
consists of two parts. Section 6.2 focuses on the analysis of the degenerate
DE model. In Sect.6.3 we discuss results of first-principles band structure
calculations. We will concentrate mainly on the problem of magnetic-phase
stability and the optical properties of manganites.
6 Orbital Degeneracy and Magnetism of Perovskite Manganese Oxides 255
6.2 Degenerate Double Exchange Model

In this section we discuss the physics of the degenerate DE model and argue
that the main trends of the magnetic phase diagram of doped manganites
at low T can be explained by the simple microscopic tight-binding (TB)
Hamiltonian (in the subspace spanned by the spin indices s =t, t),
~ LL' LL'~ .ilex (~ ~ )
1l ij = -tij 1 + -2- 1 - (7'. ei OijOLL', (6.1)
under the following conditions.
1. The local exchange splitting .ilex between the majority (t) and minor-
ity (t) spin states is large, reflecting the high-spin state of Mn sites in
manganites.
2. The hopping matrix elements tff' between two eg orbitals, x 2 _ y2
(L,L' = 1) and 3z 2 - r 2 (L, L' = 2), are given by the proper Slater-
Koster mIes [25].1
In (6.1), (7' is the vector of Pauli matrices, i is the 2 x 2 unity matrix, and
ei = (sin Bi cos cPi, sin fh sin cPi, cos Bi) is the direction of the spin magnetic
moment at the site i. Throughout this section, all energies are in units of
effective dda transfer integral to, unless otherwise specified.
The first condition greatly simplifies the analysis of the magnetic phase
stability in the DE model. Since .ilex is large, the local spin magnetic moments
are weIl saturated, and it is sufficient to consider the stability conditions only
with respect to their rotational degrees of freedom. If MO = {en is an
equilibrium magnetic configuration, the total energy change LlE = E(M) -
E(MO) associated with small rotations ei = e~ + [«5<Pi x e~] - ~(<<5<Pi)2e~
to the new configuration M = {ei} is given in the second order of {«5<Pi}
by [26]:
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
~ ~
4Hn" = ~i - ~(sin 2ja o:" o:

+ cos 2ja z) and na is the primitive translation in
the direction 0 of the cubic lattice (0 = 1, 2, and 3 corresponds to x, y, and z,
respectively) .
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.
In the case of collinear spin alignment, the Hamiltonian (6.1) is diagonal

with respect to the spin indices, Hi] = Hijö ss " and the parameters {J-rn}
~, ~
can be expressed analytically in terms of the one-electron Green functions

gij(e:) = (e: -H s )";/ as [26]:
(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.
6.2.1 Ferromagnetic Ordering
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)
which is proportional to to and does not depend on i1 ex , and the AFM SE

interaction
(6.6)
which is proportional to t5. The next-nearest neighbor interactions are given

by
(6.7)
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
For the cubic lattice, one can find that

-2zJP = m(l) , (6.8)
-z "'-lex
A JS1 -- m(2) , (6.9)
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).
6.2.2 Antiferromagnetic Ordering

Being guided by results of the previous section, let us turn to the analysis of
what is properly called the DE model. 1t is based on two assumptions.
1. All interactions in the limit Ll ex -+ 00 (hereafter referred to as the DE
limit) are treated rigorously and generalized to the case of arbitrary spin
arrangement. The DE limit for the Hamiltonian (6.1) can be obtained by
introducing the local coordination frame, in which the spin-quantization
axis at the site i is specified by the vector ei, and projecting out the
.J..-spin states. This yields the well-known DE Hamiltonian,
(6.11)
which defines the kinetic energy (ED ) of the eg electrons. The com-
plex multipliers ~ij = cos!? cos ~ + sin!? sin ~e-i(i-j) describe the
modulation of kinetic hoppings caused by deviations from the FM spin
alignment.
2. The remaining interactions in the first order of 1/Llex are taken into
account by the following anzatz. The effect of nn AFM SE interactions
(associated with the both eg and t2g electrons) is simulated by adding
to the total energy the term Es = -~Js E(ij) ei . ej with the isotropie
coupling constant JS < O. The longer-range interactions are formally
neglected with some additional constraints superimposed on the form of
the long-range magnetic order. 4
Despite the apparent simplicity, the model carries very rich physies, the
importance of whieh is frequently underestimated. Since the AFM spin align-
ment in the DE limit switches off all kinetic hoppings between the antifer-
romagnetically coupled atoms (~ij = 0), the e g electrons will be confined
in the regions intra-connected by FM pathes. The geometry of such regions,
and thereby the dimensionality of the e g electron system, will be determined
by the type of the magnetic ordering. So, the phases F, A, and C will be
effectively three-, two-, and one-dimensional. 5 The change of dimensionality
results in dramatic changes of the electronic band structure as shown in
Fig. 6.2. Clearly, the parameters of magnetie interactions (6.3), whieh depend
on details of the electronie structure, will be different for different magnetic
states. 6 Furthermore, the anisotropy of the AFM ordering will lead to the
anisotropy of magnetie interactions. The magnetie-state dependence and the
anisotropy of these interactions will act as a feedback effect, whieh will explain
the rich variety of magnetic structures observed in manganites. Thus, the
4 For example, sinee for the less than half-filled t-spin band 2h + h > 0, the spin
eanting will be more favorable than the ineommensurate spin-spiral ordering.
5 The AFM CE phase presents an even more interesting example, whieh will be
eonsidered separately in Seet.6.2.3.
6 Note that the situation eannot be aceounted for by the simple Heisenberg model.
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
main idea behind the low-temperature phase diagram of doped manganites

ean be aeeumulated in the following formula [32- 34]: magnetie ordering -+
details of eleetronie strueture -+ parameters of the magnetic interaetions -+
stability of this magnetie ordering.
The orbital degeneraey presents an important addition to the eanonical
DE pieture. The anisotropie AFM spin ordering breaks the eubie symmetry
also in the orbital subspaee and gives rise to the net orbital polarization sueh
that the states of the x 2 - y2 (3z 2 - r 2 ) symmetry become preferentially
occupied in the case of A(C)-type AFM ordering (Fig. 6.2). The situation is
typically referred to as the orbital ordering [11]. One of the most disputed
quest ions in the area is the role played by onsite Coulomb interactions. In the
mean-field theories based on Hubbard-type Hamiltonians, the magnitude of
the orbital polarization is controlled by an extra term in (6.1) proportional
to Ueff = U - I, where U is the onsite Coulomb repulsion between the same
orbitals [22,35]. We addopt the viewpoint that Ueff is not large in doped
manganites and a realistic description can be aehieved already in the limit
Ueff = 0 [30], i.e. starting from (6.1). The finite-Ueff effects may play some
role in the problem and modify some conclusions based on the degenerate
DE model quantitatively. However, it is not quite correct to say that Ueff is
an indispensable ingredient of the model that controls the dimensionality of

the system and the type of magnetic ordering [22]. 7
In the ground state, the AFM phase is effectively devided into FM sublat-
tices that interact with each other only via the SE term, but not the hopping
matrix elements. However, even small rotations of the spin moments away
from the AFM alignment will induce the kinetic hoppings between different
sublattices, which will contribute to the magnetic interactions (6.3) between
the antiferromagnetically coupled atoms. The corresponding parameters of
the magnetic interactions can be obtained using the perturbation theory
expansion for the hopping matrix elements, which yields [32]:8
(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:
JP(H) < IJsl < Jp(tt), (6.13)

where JP(tt) and JP(H) are given by (6.5) and (6.12), respectively. In
order to obtain similar stability conditions for the A(C)-type AFM phases
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].
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.
<:>
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
structure of (6.12) in the case of A(C)-type AFM ordering is such that

an enlargement of Lln reduces the FM interaction J D (t-Ir) and thereby
additionally stabilizes the AFM coupling between the planes (chains)
[33]. Lln can be enlarged by the onsite Coulomb interaction Ueff [22]
and the tetragonal distortion of the sample such that the contraction
(elongation) along one of the cubic axes will favor the AFM ordering of
the A(C) type [21].
An example of the phase diagram [33], obtained after minimization of the

total energy E((}xy, (}z ) with respect to the angles between spin magnetic
moments in adjacent chains and planes of the cube (correspondingly, Oxy
and (}z), is shown in Fig.6.4. 10 An approximate doping dependence of JS
is taken from (6.6), which is strictly valid only for the FM phaseY The
absolute magnitude of JS around x = 0.4 is adjusted in accordance with
the data obtained after analysis of experimental parameters of the magnetic
interactions in PrO.63SrO.37Mn03 as discussed in Sect.6.2.1. Thus, with the
increase of x we obtain the following sequence of the magnetic ground states:
FM -+ spin canting -+ A-type AFM -+ C-type AFM spin canting -+ G-type
AFM, in an overall agreement with the experimental situation [2,11- 13].
6.2.3 CE-type Antiferromagnetic Ordering
The main building blocks of the CE-type AFM structure are the FM zigzag
chains, which are coupled antiferromagnetically in the direction perpendicular
10 The states F, A, C, and G correspond to ((}xy,Oz ) = (0,0), (0,71"), (71",0), and

(71",71"), respectively.
11 In principle, the precise doping dependence of JS is not very important for the
model analysis. A qualitatively similar conclusion can be obtained by considering
the constant scan with JS ~ 0.1.
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.
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
,,--,--~----,~--r--1 - - F
o
00
-FA
si ---- AG
---, A
0.10 0.12 0.14 0.16

1I1
Fig. 6.7. Total energy of CE-type antiferromagnetic phase relative to other possible
equilibrium solutions of the degenerate double exchange model (Fig. 6.4) at x = 1/2
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
13 Note, that J;{,B , is proportional to the kinetic energy - (6.8).

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.
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.
6.3 First-Principles Band Structure Calculations

6.3.1 Method of Calculations
Most of the density-functional calculations discussed in this section have been
done in the local-spin-density approximation (LSDA), implemented in the
nearly orthogonal version of ASA-LMTO method [59].
The basie physics underlying the LSDA calculations is very similar to the
degenerate DE model considered in the previous section. Nevertheless, two
comments are in order.
1. The exchange splitting L1 ex is not large (perhaps, sometimes even insuf-
ficiently large), and the canonical DE limit L1 ex --+ 00 cannot be realized
on the level of LSDA. However, the system remains in proximity to the
DE limit, so the FM ordering for x 2: 0.3 is typically accompanied by the
fully spin-polarized (half-metallic) behavior.
2. The kinetic part of the Hamiltonian (6.1) is more complicated in the
LSDA. In partieular, the 0(2p) states are treated explicitly, whieh may
be very important in the analysis of interatomie magnetic interactions of
oxide compounds [51,52].
The validity of LSDA approach for manganites has been the subject of hot
debate [30,31,53-56]. The materials are frequently regarded as strongly corre-
lated systems, and many researchers believe that the LSDA description can be
significantly improved by combining it with the onsite Coulomb interactions,
treated on the mean-field level (the so-called LDA+U approach [57]) [55,56].
Such thinking can be very dangerous. It is true that on the present level of
first-principles band structure calculations, LDA+U presents the most ad-
vanced way of treating the orbital polarization effects [35]. However, it is also
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.
Magnetic Interactions. In the density-functional calculations, parameters

of magnetic interactions can be obtained from (6.3). However, sometimes it is
more convenient to use an alternative approach, and to find first the Fourier
images of these interactions in the reciprocal space (Jq ), by exploiting an
old idea of band structure calculations for incommensurate spin-spiral states
[60,61].17 J q can be then Fourier transformed to the real space.
In the FM states, the required (local) mapping onto the Heisenberg model
is given by the following expression for the total energy change: E(rJ, q) =
E(O,q) + ~(Ja - Jq)rJ 2 (in other words, Ja - Jq = 8~E(rJ,q)I19=a), where
E(rJ, q) is the total energy of the spin-spiral configuration with the cone-
angle rJ and the spin-spiral vector q, and Ja = Jq=a. The method is similar
°
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)
17 The spin-spiral calculations can be performed by applying the generalized Bloch

transformation and combining the real-space translations with rotations of the
spin-coordinates frame.
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)
Optical Spectra. The conductivity tensor, a(w) = areg(w) + aD(w), can

be calculated from the eigenvalues (ckn) and the eigenvectors (Ikn)) of the
ASA-LMTO method using the Kubo formula [64,65], where the interband
(regular) and intraband (Drude) counterparts are given by (in Ry units)
(6.16)
and
aaß(w)
D -_8i 1 "''''
V --.- aß
~ ~ 7rnn 7r nn t5 (cF - ckn ) , (6.17)
W + ZTJ k n
respectively, ikn is the zero-temperature Fermi distribution function for the

state Ikn), 7r~n' = (knl( -iV a)lkn') is the matrix element of the momentum
operator, Wn'n = Ckn' - Ckn, and V is the unit cell volume. The plasma
frequencies w~ are given by the identity: Re [a~a(w)] = (w~/2?t5(w).
The Brillouin zone integration has been performed on the mesh of 32 x
32 x 32k-points in the FM state, and similar density meshes for the AFM
states (in practice, the integration in (6.16) has been replaced by a summation
with the Lorentzian broadening TJ = 68 meV).
Experimentally, it is sometimes convenient to consider the integral char-
acteristic,
V r
N,ill(w) = 27r Jo dwRe [aaa(w)] , (6.18)
which gives an effective number of electrons occupying the low-Iying states

of interest (the eg electrons in the case of manganites). For w -+ 00, (6.18) is
simply the well-known f-sum rule [65].
Virtual-Crystal Approximation. In order to treat the hole doping self-

consistently, we employ the virtual-crystal (VC) approximation for the phe-
nomenological alloy Lal-xBaxMn03, in which the replacement of x atoms
of La by Ba is modeled by pseudoatoms with the fractional atomic number
Zvc = (1 - X)ZLa + XZBa = 57 - x. The validity of the VC approximation
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
as was first emphasized by experimentalists for similar compounds [67], and

typically attributed to the large spin-orbit coupling at the (heavy) Pb atoms.
This appears to be only one part of the story. Another important factor is
the specific electronic configuration of the Pb atoms, which can significantly
affect the optical properties of manganites upon doping.
On the contrary, the SWD calculated for La3/4Pb1/4Mn03 does not signif-
icantly differ from other alloys. According to the inelastic neutron-scattering
experiments [68], the form of the SWD in LaO.7Pbo.3Mn03 presents a very
special case and can be weIl described by the nearest-neighbor Heisenberg
model, contrary to other manganites like PrO.63SrO.37Mn03 [28], Ndo.7SrO.3
Mn03 [69], and Lao.7Cao.3Mn03 [70]. Such an anomalous behavior of Lao.7
PbO.3Mn03 is not yet understood (and, according to the present analysis, it
cannot be explained even in a more advanced SC model for the Ph-doping in
comparison with the VC approach considered in [27]). One possible reason
for this anomaly could be the strong disorder effects (possibly related with
the disorder of the local lattice distortions) caused by the mixture of very
different chemical elements. According to the (generalized) RKKY theory,
the disorder is expected to suppress the long-range magnetic interactions,
though the idea should be treated very cautiously [71].
The La( 4f) states play an important role in the cohesion properties of
LaMn03 [72]. However, their contribution to the magnetic and optical prop-
erties was found to be small. 18 Therefore, La( 4f) states have been treated as
the (atomic-like) core states.
6.3.2 Stability of the Ferromagnetic Ordering

An example of calculated SWD for the cubic La1_xBaxMn03 (with the lattice
parameter ao = 3.876 A is shown in Fig. 6.10. The doping dependence of wq
is determined mainly by two interactions.
1. The nn exchange interaction J 1 decreases with x, reflecting similar ten-
dency for Jf (6.8) for the less than half-filled t-spin band. Jf competes
with the AFM SE interaction Jr.
At a certain point (typically around
x = 0.5), Jrstart to prevail and the total coupling becomes antiferro-
magnetic.
2. The doping dependence of J4 is not monotonous: J4 is small for small
x and reaches the maximal value around x = 0.5, where it becomes
comparable with J 1 • Therefore, the interplay of hand J 4 has a visible
effect of the shape of wq (the so-called "softening" of the SWD [28]).19
18 The spin-polarization of La(4f) states, caused by the hybridization with other
states, is very smalI. The main contribution of the La( 4f) states to the optical
properties, dlle to the intra-atomic 5d-4f transitions, is also small becallse both
the 5d and 4f of La are located mainly in the unoccupied part of the spectrum.
19 Note, however, that contrary to several fanciflll scenarios discussed in the litera-
ture [28,69,70,73], one can argue that the softening is a very natural effect, direct1y
related with the electronic structllre of the FM state in manganites [27,52].
272 1. Solovyev and K. Terakura
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)
Fig.6.10. Spin-wave dispersion (left) and distance dependence of magnetic inter-

actions (right) in the cubic ferromagnetic Lal-xBaxMn03
The shape of the theoretical SWD is in semiquantitative agreement with

experimental data. For smaH x, it is determined mainly by Jt and can be
weH described by the Heisenberg model with the nn interactions [74]. The
increase of x gives rise to the longer-range interactions [75] and results in the
more complex shape of SWD [28,69,70]. Further increase of x leads to the
instability of the FM state (w q < 0). Since only J 1 contributes to wq at the
zone boundary (contrary to J 4 ), the combination of small Jt and finite J 4
typically leads to the very pronounced softening of the SWD as the system
approachs this instability near x = 0.5 [76].
Unlike in the TB model, it is rather difficult to separate unambiguously
the DE and SE contributions to J 1 on the level of first-principles calculations.
One possibility is to use different band width dependences of the DE and
SE interactions (6.5) and (6.6), and consider (very roughly) the following
interpolation:
(6.19)
where t(a) is an effective nn transfer integral that depends on the cubic

lattice parameter a varying around ao. By considering the canonical scaling
t(a) cx a-O: [25], and only those a that correspond to the half-metallic be-
havior, we obtain for x = 0.4: Jp(ao) = 86 meV, Jf(ao) = -71 meV, and
Cl! = 4.13. 20 The values of Jp and Jf are in reasonable agreement with the es-
timates obtained from the analysis of experimental data for Pro.63Sro.37Mn03
in Sect. 6.2.l.
Like the absolute magnitude, the volume dependence of J 1 is also deter-
mined by the competition of the FM DE and AFM SE interactions, and de-
pends on the ratio t(a)/t1 ex • As the volume changes, the DE (SE) interaction
20 An alternative interpolation with t(a) cx e- ßa yields: Jp(ao) = 84 meV,
Jf(ao) = -69 meV (Le., very similar to the previous case) and ßao = 4.28.
0
'" 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
21 The idea has received considerable attention recently [79] .

o
~xL-------~r--------~z
Fig.6.12. Variations of the spin-wave dispersion in the ferromagnetic

Lal / 2Bal / 2Mnü3 caused by the tetragonal distortion cl a. The unit-cell volume
is fixed at Vo = a~ , where ao = 3.876 A
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.
6.3.3 Jahn-Teller Distortion and A-type Antiferromagnetism

in LaMn03
There is a very special region of low hole-doping (x < 0.2) that cannot be
explained by the DE physics alone, even qualitatively. This part of the phase
diagram is still far from being completely understood. In this section we
concentrate on the undoped (parent) compound LaMn03' for which most of
the studies have been done, and try to summarize both the achievements and
failures of the first-principles calculations in this direction.
22 The values of the parameters for x = 0.5 are: Jp(ao) = 78 meV, Jf(ao) =
-71 meV, and a = 4.93.
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
Fig.6.13. Upper panel: a schematic view on the cooperative Jahn-Teller distortion

and the orbital ordering in LaMn03 (left), and the tight-binding electronic structure
of isolated xy plane (right). Lower panel: nearest-neighbor magnetic interactions
in the double exchange limit with (right) and without (left) the Jahn-Teller distor-
tions. All energies are in units of to (the effective transfer integral for the undistorted
cubic lattice )
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
23 In reality, the JTD is accompanied by substantial rotations of the Mn06 octahe-

dra [4,5].
simultaneously the experimental crystal structure and the A-type AFM spin
ordering [31,53].24 The result has a very clear explanation [84,85].
1. The A-type AFM ordering, combined with the DE physics, is responsible

for the quasi-2D behavior of LaMn03.
2. In the 2D lattice of eg electrons, any finite JTD opens the band gap.
The main results are illustrated in Fig.6.13 for a three-parameter tight-

binding model [85], in whieh the JTD dependence of the effective transfer
integrals t XY (8) and e (8), correspondingly within the xy plane and between
the planes, and the local splitting fJT(8) between the 3x 2 - r 2 (3 y 2 - r 2 )
and y2 - Z2 (Z2 - x 2 ) orbitals has been extracted from Harrison's law for
the dpO' transfer integrals in the Mn-O bonds [25] (see [85] for details). How
does the JTD affect the magnetie interactions? The situation is not simple.
In the DE limit, the nn interactions in the tt- and t .j,.-bonds, correspondingly
Jf'XY and Jf'z can be evaluated using (6.5) and (6.12), respectively. Then,
for x = 0 (n = 1 e g electrons) and without the distortion one will have that
Jf'z > Jf'XY, and therefore there will be no chance to stabilize the A-type
AFM ordering by the isotropie SE interaction in the DE model. The opening
of the band gap by the JTD pro duces the sharp dip in Jf'z around n = 1
and thus facilitates the stabilization of the AFM coupling between the planes.
The orbital ordering plays a very important role in the problem: since the
y2 _ Z2 and Z2 - x 2 orbitals, whieh produce the strongest coupling in the z-
direction, become depopulated, their contribution to Jf'z will be significatly
reduced. However, it is not enough. The problem is that the cooperative JTD
is accompanied with the tetragonal contraction of the sampie, which leads to
the inequality t Z (8) > t XY (8) and restores the "wrong" hierarchy of the DE
interactions Jf,Z(8) > Jf,XY(8) for finite 8. Therefore, the only possibility
to stabilize the nn magnetic interactions in the DE limit is to combine the
JTD effect with the anisotropy of SE interactions: IJf,Z(8)1 > IJf,XY(8)1.
Since t Z (8) > t XY (8), the required anisotropy can be produced by the same
distortion. This is the basic microscopic mechanism that explains the local
stability of the A-type AFM spin ordering in LSDA [20].
Corresponding magnetic interactions in the reciprocal space are shown
in Fig. 6.14 for two different types of the orbital ordering and the JTD: the
type C, where the pattern in the xy plane shown in Fig. 6.13 is periodically
repeated in the z-direction (the experimental situation); and the type G,
where each translation in the z-direction is accompanied by the 90° rotation
of the pattern (i.e., the 3x 2 - r 2 orbital is transformed to 3 y 2 - r 2 , etc.).25
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
--D--o- _____ -{] 0- --0- -- 8---- ---{]

fY
I
0.05 0.06 om 0.08 0.09 0.05 0.06 0.07 0.08 0.09

Ö Ö
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
in the xy plane. 26 J 2 additionally stabilizes the AFM eoupling between the

planes [20]. On the eontrary, J;Y plays a negative role in the loeal stability
of the AFM state A: sinee J;Y < 0, the AFM ordering beeomes unstable
for small 8 with respect to the ineommensurate spin-spiral ordering in the
xy plane. From the viewpoint of magnetie stability, the orbital ordering of
the type G should be preferencial, beeause it leads to larger IJ21and smaller
IJ;YI. Therefore, the type of orbital ordering realized in LaMn03 ean be
hardly understood from the viewpoint of magnetism,27 and is most probably
determined by the lattiee effects [48,49], which also drive the A-type AFM
spin ordering.
Onee the erystal struture is fixed to the experimental one, the LSDA
provides us with a rather eonsistent deseription of the electronie and magnetie
properties of LaMn03'
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].
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.
type AFM ordering, but only at the expense of spectroscopic characteristics.

For example, the gap size is substantially overestimated. There have been no
attempts at structural optimization in this regime.
A small U, combined with the experimental JTD, may improve the de-
scription of optical properties [93), but does not solve the problem of struc-
tural instability inherent in the LSDA approach [72].
To summarize this section, the LSDA-based description for LaMn03 is
by no means prefect. However, it seems that consistent improvement of it
can be achieved only by consistently including the electron correlations in
first-principles calculations, as in Hedin's GW approach [94]. This certainly
implies extremely heavy calculations, but in a broad perspective seems to be
the most realistic way to proceed. The present attempts to combine LSDA
with the model description for the onsite Coulomb interaction do not look
very promising for LaMn03 '
6.3.4 Metallic Antiferromagnetism at Large x

The appearance of the AFM spin ordering at large hole doping can be ex-
pected already from the analysis of magnetic interactions in the FM state, as
discussed in Sect. 6.3.2. However, the situation is not so straightforward. One
should not exaggerate the applicability of the Heisenberg form for the total
energy change (6.2) derived for infinitesimal rotations of the spin magnetic
moments near the FM equilibrium and try to use it for finite spin rotations.
Let us illustrate this point more in detail. According to Figs. 6.10 and 6.11,
the FM state in the cubic Lal - xBaxMn03 with x = 0.5 and ao = 3.876 A is
unstable towards the A-type AFM ordering. However, will the A-type AFM
state be stable itself? The answer is given in Fig.6.16: the A-type AFM
state appears to be also unstable (J: < 0 in the wide q-area), and the
main reason for this instability is the dramatic increase of J1 from 7 meV
in the FM state to 53meV in the A-type AFM state. Thus, although the
Heisenberg model (6.2) can be applied for infinitesimal spin rotations near
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)
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
,,
,
,,
,,
1.05 ,, c
F ,,
,, ,
,
--~~ ... ~
~ 1.00 ,,
.... { G
,,
,,
0.95 ,,
A
,
0.90 L-~_---,--_~_~ _ _--,-_------,L-.J

0.2 0.4 0.6 0.8 1.0
x
Fig.6.18. Sketch of the phase diagram of Lal-xSrxMn03 in the plane spanned

by c/a and x (pseudopotential GGA calculations - [21]). Solid lines are the phase
boundaries obtained with the experimental volumes [79]. The dashed lines are the
phase boundaries obtained after expanding the experimental volumes by 9%
GGA approach implemented in the pseudopotential code [21], is shown in

Fig.6.18. Regarding the general sequence of the magnetic ground states
FM-+A-type AFM-+C-type AFM-+G-type AFM with the increase of x, there
is a clear similarity with the experimental trend. As was already empha-
sized in Sect. 6.3.2, the validity of this picture can be checked experimentally
through the volume dependence of nn magnetic interactions: if correct, the
uniform compression of the sampie should favor the AFM coupling. However,
as it was pointed out before, the real situation can be more complicated
and some of the states shown in Fig.6.18 may be unstable with respect to
a noncollinear magnetic ordering. The total energies versus the canting angle
between spin magnetic moments in adjacent FM xy planes are shown in
Fig. 6.19, for the case of tetragonal contraction cl a = 0.95. For x = 0.5, E( B)
has a pronounced minimum at B = n, corresponding to the A-type AFM
ordering. The minimum becomes shallow at x = 0.4, and finally transforms
to a maximum at x = 0.3. In the latter case, the new minimum emerges
around B = O.4n, which corresponds to the spin-canted ground state. The
modification of the phase diagram caused by the possible spin canting (or
more general forms of the noncollinear spin arrangement) is not fully investi-
gated yet, but certainly presents an interesting problem to address. Another
interesting problem is the phase separation [7,8], which is, however, extremely
difficult to tackle on the level of first-principles calculations.
6.3.5 Zigzag Antiferromagnetic Ordering at x = 1/2

The zigzag-type AFM structures are less studied by first-principles meth-
ods, mainly because of the large size of magnetic unit cells required for
0.2 0.4 0.6 0.8

e (units of 1t)
Fig.6.19. Total energy calculations for La1/2Ba1/2Mn03 with c/a = 0.95 as
a function of angle between spin magnetic moments in adjacent xy planes
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.
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
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
change the situation dramatically. The quantitative behavior of nn magnetic

interactions in the xy plane is given by the following interpolation for small
8 [34J: JAIBI (8) ~ 43 + 3138 and JA 1 B 2(8) ~ 12 - 6408 meV (see Fig.6.5 for
notations). Even without the JTD, the magnetic interactions reveal a strong
anisotropy J A1B1 (0) > JAIB2 (0) reflecting the anisotropy of the zigzag AFM
ordering, in agreement with results of the model analysis in Sect. 6.2.3. How-
ever, it is not enough, and a small distortion 8 ~ 0.02 becomes indispensable
in order to stabilize the AFM coupling between neighboring zigzag chains in
LSDA. On the other hand, the structure optimization presents a very serious
problem also for the zigzag AFM phase. Similar to LaMn03' the optimization
tends to shrink the JTD and make the zigzag AFM state unstable [96J.
Little is known about the ability of the LSDA (or GGA) for the CE-
type AFM phase in 3D manganites. The CE-type AFM ordering cannot be
stabilized in the undistorted cubic lattice on the LSDA level. 32 Effects of the
crystal distortion have not been investigated yet, but are known to be strong
in the 3D compounds exhibiting the CE-type AFM ordering [l1J.
6.3.6 Optical Properties
A very insightful analysis of the optical conductivity in the FM phase of

manganites has been given by Shiba et al. [97J. They considered a simple
tight-binding model and argued that the existence of the off-diagonal transfer
integrals between the 3z 2 - r 2 and x 2 - y2 orbitals plays a very important role
and gives rise to the inter band optical transitions, which naturally account
for the experimentally observed absorption in the low-energy part of the
spectrum. This is the main microscopic mechanism underlying the band ap-
proach, though not the only one. Results of LSDA calculations for the virtual-
crystal alloy Lao.7Bao.3Mn03 are shown in Fig. 6.21. 33 The compound is half-
metallic: the Fermi level crosses the Mn(eg) band for the majority (t)-spin
and falls in the gap between the O(2p) and Mn(t2g ) bands for the minority
(+)-spin. Another important structure relevant to the problem is the t-spin
Mn(t2g) manifold in which the Mn states are strongly mixed with the O(2p)
states. Therefore, the low-energy absorption in the t-spin channel will be gov-
erned by both Mn(eg)---7Mn(eg) and Mn(t2g)---7Mn(eg) inter-band transitions.
The charge-transfer (CT) excitations o (2p)---7 Mn( eg ) start around 2.5 eV.
The optical absorption in the +-spin channel is gapped, and is featured by
the CT transitions ofthe O(2p)---7Mn(t2g) and O(2p)---7Mn(eg ) types starting
around 1.5 and 2.5 eV, respectively. The intraband Drude contribution is large
(the corresponding plasma frequency: nw p ~ 2.1 eV) and accounts for almost
100% of Neff at small w, being weIl consistent with the results oftight-binding
analysis of Shiba et al. [97).
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..
-
\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
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
thermally induced low-energy orbital excitations considered in [98] seem to be

an unlikely scenario, since they are expected only at very high temperatures
T rv 0.3to rv 2500 K, much higher than the Neel temperature of the A-type AFM
state around x = 0.5 (rv 220 K [12]).
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)
Fig.6.23. LSDA calculations for the C-type antiferromagnetic phase of

Lao.3Bao.7Mn03 with c/a = 1.05. See Fig. 6.22 for other notations
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].
-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)
Fig.6.24. Elements of the conductivity tensor for Yl/2Sr3/2Mn04 in the zigzag

antiferromagnetic state. Different lines correspond to different values of the Jahn-
Teller distortion 8
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
2 4
ro(eV)
Fig. 6.25. Optical conductivity for the G-type antiferromagnetic phase of CaMn03
one of the experimental reports. However, the calculated intensity at the

peak maximum (700 n-1cm- 1) appears to be smaller than the experimental
one (1100 n -1 cm -1). Another interesting feature is the anisotropy of optical
conductivity in the xy plane, caused by the anisotropy of the AFM spin or-
dering and manifested in the nonvanishing a xy component of the conductivity
tensor, in qualitative agreement with the experimental finding [109].
The optical conductivity for the G-type AFM CaMn03 is shown in
Fig.6.25. 38 The first (1.4eV) peak corresponds to the CT excitations from
the O(2p) band to the t-spin Mn(eg) and .!--spin Mn(t2g) bands (see [31] for
details of density of states). The second (3.5 eV) peak is due to the O(2p)--+.!--
spin Mn( e g ) CT excitations. The experimental position of the first peak is
around 3eV [110], i.e. higher by 1.6eV than in LSDA calculations. This
clearly indicates that the band gap is underestimated in LSDA, being actually
consistent with the fact that the nn SE interaction J 1 is overestimated in
CaMn03 (Sect. 6.2.1). However, the error is not as large as in the case of the
notorious rocksalt transition-metal oxides [58].
6.4 Concluding Remarks
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
of the first-principles band structure calculations in the local-spin-density

approximation.
Finally, we would like to point out two main unresolved problems, which
from our point of view present both the challenge and the significant interest
for the future studies in this direction:
1. Structural optimization and magnetic stability of the parent manganite

LaMn03 on the level of first-principles band structure calculations (which
interaction, missing in LSDA, makes the crystal and magnetic structure of
this compound stable?); and more generally - the quantitative description
of manganites in the region of low hole doping;
2. Understanding of the magnetic phase diagram along the temperature
axis: the nature of the insulating behavior above the magnetic transition
temperature; the existence of distinct temperatures for the charge and
orbital ordering, which is frequently different from the magnetic Neel
temperature; and, of course, the understanding of the phenomenon of
colossal magnetoresistance itself. The successful description of the mag-
netic phase diagram of doped manganites has certainly revived an interest
in the degenerate double exchange model. However, whether the same
minimal model will be sufficient in the high-temperature regime is still
an open quest ion that needs to be clarified.
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)
12. H. Kawano, R. Kajimoto, H. Yoshizawa, Y. Tomioka, H. Kuwahara, and

Y. Tokura: Phys. Rev. Lett. 78, 4253 (1997); T. Akimoto, Y. Maruyama, Y.
Moritomo, A. Nakamura, K. Hirota, K. Ohoyama, and M. Ohashi: Phys. Rev.
B 57, R5594 (1998); H. Kuwahara, T. Okuda, Y. Tomioka, A. Asamitsu, and
Y. Tokura: Phys. Rev. Lett. 82, 4316 (1999)
13. C. Martin, A. Maignan, M. Hervieu, and B. Raveau: Phys. Rev. B 60, 12191
(1999); H. Fujishiro, M. Ikebe, S. Ohshiden, and K. Noto: J. Phys. Soc. Jpn.
69, 1865 (2000)
14. S. Mori, C.H. Chen, and S.-W. Cheong: Nature 392, 473 (1998)
15. P.G. Radaelli, D.E. Cox, L. Capogna, S.-W. Cheong, and M. Marizio: Phys.
Rev. B 59, 14440 (1999)
16. C. Zener: Phys. Rev. 81, 440 (1951); Phys. Rev. 82, 403
17. P.W. Anderson and H. Hasegawa: Phys. Rev. 100, 675 (1955)
18. K. Kubo and N. Ohata: J. Phys. Soc. Jpn. 33, 21 (1972)
19. As the review, see V.Y. Irkhin and M.1. Katsnel'son: Usp. Fiz. Nauk 164,
705 (1994) (Russian). [English trans!.: Phys.-Usp. 37, 659-676]; N. Fu-
rukawa: in Physics 0/ Manganites, ed. by T.A. Kaplan and S.D. Mahanti.
(Kluwer/Plenum, New York 1999)
20. I. Solovyev, N. Hamada, and K. Terakura: Phys. Rev. Lett. 76,4825 (1996)
21. Z. Fang, LV. Solovyev, and K. Terakura: Phys. Rev. Lett. 84, 3169 (2000)
22. R. Maezono, S. Ishihara, and N. Nagaosa: Phys. Rev. B 58, 11583 (1998)
23. J.D. Lee and B.1. Min: Phys. Rev. B 55, R14713 (1997)
24. J. van den Brink and D. Khomskii: Phys. Rev. Lett. 82, 1016 (1999)
25. W.A. Harrison: Electronic Btructure and the Properties 0/ Bolids: The Physics
0/ the Chemical Bond. (Dover Publications, New York 1989)
26. A.1. Liechtenstein, M.1. Katsnelson, V.P. Antropov, and V.A. Gubanov: J.
Magn. Magn. Mater. 67, 65 (1987)
27. LV. Solovyev and K. Terakura: Phys. Rev. Lett. 82, 2959 (1999)
28. H.Y. Hwang, P. Dai, S-W. Cheong, G. Aeppli, D.A. Tennant, and H.A. Mook:
Phys. Rev. Lett. 80, 1316 (1998)
29. O. Gunnarsson: J. Phys. F 6, 587 (1976)
30. I. Solovyev, N. Hamada, and K. Terakura: Phys. Rev. B 53, 7158 (1996)
31. W.E. Pickett and D.J. Singh: Phys. Rev. B 53, 1146 (1996)
32. LV. Solovyev and K. Terakura: Phys. Rev. Lett. 83, 2825 (1999)
34. LV. Solovyev: Phys. Rev. B 63,174406 (2001)
35. LV. Solovyev, A.1. Liechtenstein, and K. Terakura: Phys. Rev. Lett. 80, 5758
(1998)
36. J. Lee, J. Yu, and K. Terakura: J. Korean Phys. Soc. 33, S55 (1998)
37. T. Hotta, Y. Takada, and H. Koizumi: Int. J. Mod. Phys. B 12, 3437 (1998)
38. J. van den Brink, G. Khaliullin, and D. Khomskii: Phys. Rev. Lett. 83, 5118
(1999)
39. T. Hotta, Y. Takada, H. Koizumi, and E. Dagotto: Phys. Rev. Lett. 84, 2477
(2000)
40. B.J. Sternlieb, J.P. Hill, U.C. Wildgruber, G.M. Luke, B. Nachumi, Y. Morit-
omo, and Y. Tokura: Phys. Rev. Lett. 76, 2169 (1996)
41. Colossal Magnetoresistive Oxides: ed. by: Y. Tokura, (Gordon and Breach,
Tokyo 2000)
42. Y. Murakami, H. Kawada, H. Kawata, M. Tanaka, T. Arima, Y. Moritomo,
and Y. Tokura: Phys. Rev. Lett. 80, 1932 (1998)
43. M. v. Zimmermann, J.P. Hill, D. Gibbs, M. Blume, D. Casa, B. Keimer, Y.

Murakami, Y. Tomioka, and Y. Tokura: Phys. Rev. Lett. 83, 4872 (1999)
44. T. Mutou and H. Kontani: Phys. Rev. Lett. 83, 3685 (1999); D. Khomskii
and J. van den Brink: Phys. Rev. Lett. 85, 3329 (2000)
45. L.E. Gontchar, A.E. Nikiforov, and S.E. Popov: J. Magn. Magn. Matter. 223,
175 (2001)
46. H. Meskine, H. König, and S. Satpathy (2000): preprint.
47. C.H. Chen and S.-W. Cheong: Phys. Rev. Lett. 76, 4042 (1996); P.G.
Radaelli, D.E. Cox, M. Marizio, and S.-W. Cheong: Phys. Rev. B 55, 3015
(1997); R. Kajimoto, H. Yoshizawa, Y. Tomioka, and Y. Tokura: Phys. Rev.
B 63, 212407 (2001)
48. J. Kanamori: J. Appl. Phys. Suppl. 31, 14S (1960)
49. A. J. Millis: Phys. Rev. B 53, 8434 (1996)
50. S.Y. Ezhov, V.1. Anisimov, D.1. Khomskii, and G.A. Sawatzky: Phys. Rev.
Lett. 83, 4136 (1999)
51. T. Oguchi, K. Terakura, and A.R. Williams: Phys. Rev. B 28, 6443 (1983);
J. Zaanen and G.A. Sawatzky: Can. J. Phys. 65, 1262 (1987)
52. P. Mahadevan, LV. Solovyev, and K. Terakura: Phys. Rev. B 60, 11439 (1999)
53. N. Hamada, H. Sawada, and K. Terakura: in Spectroscopy of Mott Insulators
and Correlated Metals, ed. by A. Fujimori and Y. Tokura, Springer Series in
Solid-State Sciences, Vol. 119 (Springer, Berlin Heidelberg New York 1995)
p.95-105
54. D.D. Sarma, N. Shanthi, S.R. Barman, N. Hamada, H. Sawada, and K. Ter-
akura: Phys. Rev. B 75, 1126 (1995)
55. S. Satpathy, Z.S. Popovic, and F.R. Vukajlovic: Phys. Rev. Lett. 76, 960
(1996)
56. V.1. Anisimov, I.S. Elfimov, M.A. Korotin, and K. Terakura: Phys. Rev. B
55, 15494 (1997)
57. V.I. Anisimov, J. Zaanen, and O.K. Andersen: Phys. Rev. B 44, 943 (1991)
59. O.K. Andersen: Phys. Rev. B 12, 3060 (1975); O. Gunnarsson, O. Jepsen,
and O.K. Andersen: Phys. Rev. B 27, 7144 (1983)
60. C. Herring: in Magnetism, Vol. IV, ed. by G. Rado and H. Suhl, (Academic
Press, New York London 1966); L.M. Sandratskii: Phys. Stat. Sol. B 136,
167 (1986); L.M. Sandratskii: Adv. Phys. 47, 91 (1998)
61. O.N. Mryasov, A.1. Liechtenstein, L.M. Sandratskii, and V.A. Gubanov: J.
Phys.: Condens. Mater 3, 7683 (1991)
62. S.V. Halilov, H. Eschrig, A.Y. Perlov, and P.M. Oppeneer: Phys. Rev. B 58,
293 (1998)
63. K. Yosida: Theory of Magnetism, Springer Series in Solid-State Sciences,
Vol. 122 (Springer, Berlin Heidelberg New York 1996)
64. Y.A. Uspenski, E.G. Maksimov, S.N. Rashkeev, and 1.1. Mazin: Z. Phys. B
53, 263 (1983); M. Alouani, J.M. Koch, and M.A. Khan: J. Phys. F 16, 473
(1986)
65. C.S. Wang and J. Callaway: Phys. Rev. B 9, 4897 (1974); Y.A. Uspenski
and S.V. Khalilov: Zh. Eksp. Teor. Fiz. 95, 1022 (1989) (Russian). [English
transI.: Sov. Phys. JETP 68, 588-596]; P.M. Oppeneer, T. Maurer, J. Sticht,
and J. Kübler: Phys. Rev. B 45, 10924 (1992)
66. C.J. Bradley and A.P. Cracknell: The Mathematical Theory of Symmetry in
Solids. (Clarendon Press, Oxford 1972)
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)
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
be at least partially magnetic in origin [5-10]. In this regard, according to

density functional calculations, lattice distortions at the Sr2Ru04 surface
are reported to be strong enough to produce surface magnetism [11], and
generalized gradient approximation calculations, which have a slightly greater
tendency to magnetism than in the standard local density approximation,
predict a magnetic ground state for this material [12]. The purpose of this
chapter is to review the electronic structure of perovskite ruthenates as it
has been understood from the interplay of experiment and calculations, and
to layout some of the unanswered questions.
7.2 Origin of Magnetism: SrRu03

Perovskite SrRu03 has been known for quite some time [13-18]. However,
it has become the subject of considerable recent attention, both because of
its relationship with the unconventional superconductor Sr2Ru04 and be-
cause of its unusual position as a 4d metallic perovskite oxide ferromagnet.
The magnetism is robust. The Curie temperature is high, Tc ~ 165 K, and
the magnetization is m ~ 1.6 MB/Ru [18,20]. There is still, however, some
uncertainty in the precise value of m because the strong magnetocrystalline
anisotropy has complicated its determination from standard measurements of
hysteresis loops [19]. Related to this Klein and co-workers [21] have measured
strong magnetooptical effects in SrRu03 films.
SrRu03 occurs in a distorted perovskite structure, specifically, of the
GdFe03 type, with orthorhombic Pbnm symmetry as is typical in perovskites
with an A-site cation that is too small relative to the B-site. The distortion
consists of rotations of the oxygen octahedra with very little change in Ru-
o bond lengths [17,22]. CaRu03' which has the same valence and differs
structurally only by a larger rotation (the space group is the same), is a para-
magnetic metal down to the lowest temperatures, though it is apparently very
dose to a magnetic instability [23-28].
The electronic structure of SrRu03 has been investigated by several au-
thors using density functional calculations [29-36]. These agree as regards
the basic features of the band structure, though there are some differences
between calculations employing spherical approximations and those with full-
potential methods. Quantum oscillations have been observed in the resistivity
of SrRu03 at low temperatures [37]. This provides strong evidence that the
ground state is a Fermi liquid. Nonetheless, transport and other data show
unconventional behavior that deserves attention. One complication is that
there is apparently a significant contribution to transport properties due to
domain walls [38], related to the spin polarization as well as substantial,
probably extrinsic, magnetoresistance effects [39]. In this context, it should
be emphasized that SrRu03' while cubic at very high temperatures, is or-
thorhombic, not cubic, at and below room temperature and therefore single
crystals and single crystal films are actually invariably twinned. Furthermore,
7 Magnetism in Ruthenates 299
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
SrAu0 3 (cubic perovskite)
·2
·4
·8 ·6 ·4 ·2 o 2
E(eV)
Fig.7.2. LSDA electronic DOS of ferromagnetic SrRu03 in the idealized cubic

perovskite structure, corresponding to the band structure of Fig. 7.1. The majority
(minority) spin channel is shown above (below) the axis. The dashed line denotes
the Ru d contribution
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
symmetry directions. This is a consequence of the B-site - 0 bonding topol-

ogy in perovskite. In ideal perovskite structure SrRu03 there are such Hat
t2g bands at E F and the Fermi surface may be described to a reasonable
approximation as the intersection of three equivalent 2-D sections, one from
each t 2g orbital. That is, three intersecting circular cylinders along k x , k y and
k z (in the actual orthorhombie structure the zone folding and antierossings
complicate this simple picture). The result is that the non-spin-polarized
DOS has a large peak almost at EF. This leads to a Stoner instability
towards ferromagnetism. As mentioned, the self-consistent magnetization is
1.17/LB/Ru. However, despite the robust magnetization, the bands are al-
most rigidly split. This is seen clearly in the DOS (Fig.7.2). In most 3d
oxides, on-site interactions, partieularly the Hund's exchange splitting, Ll,
and Coulomb repulsion, U, dominate the intersite hopping via 0, while here
the energy associated with the hopping exceeds the exchange splitting. This
along the similar hybridized characters of the bands and the fact that there is
substantial Hund's coupling on both the Ru and 0 sites are the reasons for the
unusually rigid exchange splitting of the valence bands in the ferromagnetic
state. Since the DOS near E F contains partial 0 character, the magnetization,
resulting from rigidly exchange splitting this DOS, resides partlyon O.
It is perhaps helpful to contrast the magnetism of SrRu03' whieh is repre-
sentive of ruthenates, with that found in perovskite manganites. Both systems
show important physics due to hybridization between transition metal d or-
bitals and 0 p orbitals. In the ruthenates, it is mixing of t2g - 0 P7r states,
which is strong because of the extended nature of 4d relative to 3d orbitals.
In the manganites, the hybridization might be expected to be weaker for
the same reason, but this is partly compensated by the fact that in these
materials the hybridization is of eg-pu character. However, the systems are
in fact very different. The manganites show local moment magnetism because
of three core-like t2g electrons. These t 2g moments are strongly coupled to
the itinerant eg electrons by the on-site Hund's coupling. The resulting ex-
change splitting of the e g levels places the majority e g states near the top of
the 0 p bands, while the minority states are high above the 0 bands [45].
The result is a very strongly spin dependent hybridization and rieh double
exchange physies as discussed by Solovyev and Terakura in this volume. In
the ruthenates there is no such core-like local moment. Instead, the same
electrons both provide the moments and their coupling through 0, so that
the magnetism is largely given by itinerant electron physies. One interesting
point related to this is that in manganites, as in other local moment magnetie
systems, the interesting magnetic degrees of freedom are transverse, while in
ruthenates both longitudinal and transverse magnetic degrees of freedom can
be important.
The actual crystal structure of SrRu03 is orthorhombie with four formula
units per cello The distortion from the cubic perovskite structure complicates
analysis of the band structure and broadens the DOS, but the underlying
302 D.J. Singh
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
physics is not changed. The DOS near E F is dominated by a peak derived

from hybridized Ru t2g orbitals, leading to a Stoner instability and a rigid
:::::: 0.65eV exchange splitting of the bands. This is shown in Fig.7.3. The
LSDA predicts a robust ferromagnetic ground state with a spin moment
of 1.59 J-LB in good agreement with experiment, and an energy of -0.14eV
relative to the non-spin-polarized case, both on a per Ru basis. It should
be mentioned that despite the itinerant picture of the magnetism implied by
the calculations, there are complications. In particular, optical measurements
of the exchange splitting show that it persists up to temperatures elose to
Tc and therefore that transverse spin fiuctuations play an important role
in the phase transition [36J. Interestingly, even though the majority and
minority spin DOS at EF , N(EF) are practically the same, the transport
is predicted to be highly negatively spin polarized [29J. This is because the
minority Fermi velocity is approximately twice the majority velocity (within
simple Boltzmann transport theory (Y IX N (EF )v 2 r, where r is a scattering
time). Comparisons of the LSDA value of N(EF ) with experimental data
yield a rather large mass renormalization, )' :::::: 4. The high spin polarization
is consistent with a number of transport measurements, such as the magne-
toresistance and sensitivity to grain boundaries and domain walls as noted
above. Negative spin polarization is found less frequently than positive spin
polarization in commonly studied materials, although there are no fundamen-
tal reasons why it should not occur. Thus its possible occurrence has been
of considerable interest. Recently, Worledge and Geballe [46] performed spin
polarized tunneling measurements using the Meservey-Tedrow method [47]
and found a negative spin polarization in SrRu03 - the first time a negative
spin polarization had been found by that technique. Other materials with
negative spin polarization inelude the first magnet, Fe304, which is -100%
spin polarized [48].
7.3 Importance of Lattice Degrees of Freedom:

CaRu03
CaRu03 has the same crystal structure and symmetry as SrRu03' with
the exception that the octahedral rotations are approximately twice larger,
reflecting the smaller size of the Ca2+ ion relative to Sr2+. Although initially
reported as antiferromagnetic [16], it is now established to be a paramagnetic
metal down to low temperature [23- 28]. This is exactly what LSDA calcula-
tions predict. Figure 7.4 shows the calculated LSDA fixed spin moment energy
for CaRu03' as obtained by Mazin and Singh [31] . Calculations for CaRu03,
but using the crystal structure of SrRu03 yield magnetic properties that
are practically the same as SrRu03 itself. The difference between CaRu03
and SrRu03 is therefore clearly structural in origin, which is not surprising
considering that neither Ca nor Sr are electronically active near E F . The
calculated E vs. M for CaRu03 in its actual structure is remarkably flat over
an extended range from 0 to rv 1.5IlB / Ru. This is reminiscent of fcc Pd metal,
which also shows an extended flat region in such calculations, but here the
range of flatness is even larger [49,50]. The implication is that CaRu03 should
be a very strongly enhanced paramagnet, with a high susceptibility, large
Wilson ratio, and signatures of soft spin fluctuations in transport data. Fur-
thermore, it implies that magnetic impurities may have giant moments and
that relatively small chemical perturbations may produce ferromagnetism.
High field magnetization measurements of a variety of Srl-xCax03 al-
loy samples have been reported by Kiyama and co-workers [23], confirming
that CaRu03 is indeed a strongly exchange enhanced paramagnet. He and
Cava [28] report that very low concentrations of Ti are sufficient to produce
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
ferromagnetism, placing the material on the borderline between ferromag-

netism and paramagnetism. Feiner and co-workers, using different less well
characterized sampies of CaRu03 and Fe doped CaRu03 performed magne-
tization studies and found a proximity to magnetism, though they argue in
favor of a spin-glass state [51]. However, Yoshimura and co-workers [25], while
finding robust ferromagnetic spin fluctuations in 17 0 NMR investigations
on different sampies place the ferromagnetic - paramagnetic transition in
Sr1-xCax03 near x = 0.7. Mukuda and co-workers [26], using Ru NMR
measurements, find that CaRu03 has a Stoner factor of 0.98, which is just
below the critical value of 1 for a ferromagnetic instability, and further report
that the properties are dominated by soft, long wavelength spin fluctuations.
It will be very interesting to study the low temperature Fermi liquid proper-
ties of CaRu03 in detail using highly pure, low residual resistivity sampies,
should these become available.
The absence of ferromagnetism in CaRu03 can be understood in terms
of the band structure using Stoner theory [53-56], as was discussed in [31].
Essentially, the effect of the rotational distortion of the perovskite structure
is (1) to mix Ru t2g and eg orbital symmetries, while introducing additional
small crystal field splittings, (2) at the same time to narrow the bands, and
(3) to introduce various small pseudogaps in the band structure due to zone
folding. It should also be mentioned that the zone folding and splittings
reduce the nesting of the Fermi surface. The net result is that even though
the overall t2g band width is smaller in CaRu03 than in SrRu03' the peak
at EF is not as high, as may be seen in the LSDA DOS (Fig. 7.5). The result
is that N(EF ) for CaRU03 is lower than SrRu03, and the Stoner criterion
is not exceeded. In fact, the DOS around EF is rather flat, with a value that
just reaches the Stoner criterion. This is the origin of the extended flat region
in the fixed spin moment calculation of E vs. M.
The dependence of the magnetic ground state on the octahedral rotation
in (Sr,Ca)Ru03 implies that the octahedral rotation should show sensitivity
to the magnetic state. If the characterization of the magnetism as itinerant
in nature is correct, it may be expected that there will be a noticeable lattice
signature of the onset of magnetism around Tc. Kiyama and co-workers [52]
measured the temperature dependence of the cell parameters of CaRu03
and SrRu03 using X-ray diffraction. They found that the unit cell volume of
SrRu03 was anomalously temperature independent below Tc and that the
lattice parameters have strong kinks at Tc. Meanwhile, they found normal
Debye like behavior in CaRu03.
Extended Stoner analysis helps shed some additional light on the general
features of the magnetic instabilities in ruthenates. The key parameter in
this theory is N(EF )1, where the Stoner 1 is a normally atomic-like quantity
describing the exchange enhancement. A magnetic instability occurs when
N(EF )1 exceeds 1. Generally, 1 is determined by the density distribution on
an ion, and is larger for more compact orbitals, as in 3d ions relative to 4d
8 CaRu03
E(eV)
Fig.7.5. LSDA electronic DOS of paramagnetic CaRu03 after [31]. The dashed
line denotes the Ru d contribution
ions. In compounds I is replaced by a material dependent average I. The

appropriate averaging for calculating the energetics is with the decomposed
DOS, I = IAn~ + IBn1, for two components, A and B, where IA and IB
are the Stoner I for atoms A and B, and nA and nB are the fractional
weights of A and B in N(EF ) (normalized to nA + nB = 1). The 0 2 - ion is
highly polarizable (it does not exist outside crystals) and because of this the
value of 10 may be expected to be material dependent. Nonetheless, 0 2 - is
a small ion and so 10 mayaiso be large. Mazin and Singh got IRu = 0.7eV
and 10 = 1.6eV for SrRu03. From these values and the decomposition of
N(EF) Mazin and Singh found I = 0.38eV including 0 and only I = 0.31 eV
without the 0 contribution. When inserted into the extended Stoner theory
(this uses the energy dependence of the DOS, but a fixed I to calculate the
total energy as a function of magnetization), they found good agreement
with self-consistent fixed spin moment calculations. This is not what would
be expected in a normal 3d oxide. First of all, in most but not all 3d oxides
such a large 0 contribution to I is not found because the d bands are not so
strongly hybridized. Secondly, even if they were, the energy associated with
this would not be nearly large enough to force the bands to exchange split
rigidly. In SrRu03' on the other hand, the bands are very strongly hybridized
in this sense, and the exchange splitting of the bands is very rigid, as may be
seen from the band structure.
The 0 contribution to I is generic to perovskite derived ruthenates, as it
simply reflects the hybridization of the t2g orbitals of nominally tetravalent
octahedrally coordinated Ru with O. This provides a ferromagnetic interac-
tion between Ru ions connected by a common 0 [31]. The interaction comes
about because for a ferromagnetic arrangement the 0 polarizes, and this con-
tributes to the energy, while for a strictly antiferromagnetic arrangement, 0
does not polarize by symmetry, and so in this case there is no 0 contribution
to the magnetic energy. This is local physics and so this contribution to the
paramagnetic susceptibility, while peaked at the zone center, is smooth in
306 D.J. Singh
reciprocal space. In addition to this q-dependent Stoner enhancement, which

favors ferromagnetism, there are superexchange interactions, as discussed by
Solovyev [34] and nesting effects due to the perovskite topology. The later
will be enhanced in the layered ruthenates, where, as discussed below, the
electronic structures are highly two-dimensional. All of these are controlled
by the band structure, which is sensitive to lattice distortions. In addition,
in cases where the band width is narrow, as has been suggested is the case
in Ca2Ru04, Coulomb correlations due to the Hubbard U may be important
as weIl. The importance of Ru - 0 hybridization in the electronic struc-
ture and magnetism, relative to purely on-site interactions favors stronger
coupling between lattice, magnetic and electronic degrees of freedom than
in the typical 3d case, where the strong on-site terms tend to reduce the
importance of charge ßuctuations and longitudinal spin ßuctuations. The
competing interactions and the wide variety of resulting ground states make
these ruthenates a particularly interesting playground for the interplay of
theory and experiment, as discussed in the sections below.
The layered ruthenates, (Sr,Ca)n+lRun03n+b show a very rich variety of

magnetic states. The n = 00 end-members, CaRu03 and SrRu03 have
been discussed above. The n = 1 compound, Sr2Ru04 is a paramagnetic
metal that becomes an unconventional superconductor at low temperature,
while Ca2Ru04 is an antiferromagnetic insulator [57-63]. The multilayer
Ruddlesden-Popper compounds in the series show a wide variety of magnetic
orderings, metal insulator transitions and unusual transport properties, with
sampIe dependence [59,64-70]. Not surprisingly, strong magnetoelastic effects
are found and have been emphasized both theoretically [31,71,72] (see above)
and experimentally [58,63,11]. Sr2Ru04 is the simplest layered perovskite
ruthenate and is the most extensively studied perovskite ruthenate, owing to
its suitability as a metallic substrate for growing high-Tc cuprate films, and
due to its unconventional superconductivity. Its properties as they relate to
superconductivity have been discussed in detail elsewhere [73,74], and so here
we focus on those aspects relevant to magnetism. Sr2Ru04 occurs in the ideal
body centered tetragonal (I4/mmm) structure, with no lattice distortion.
Band structure calculations have been reported by several authors [12,75-
78], and analyzed in tight-binding terms [79-83]. As seen in Fig. 7.6, showing
the band structure and in the Fermi surfaces in the bct basal plane (Fig. 7.7),
the electronic structure is highly two-dimensional. The Fermi surface has
been measured by de Haas-van Alphen (dHvA) and related techniques [84,85],
angle dependent magnetoresistance oscillations (AMRO) [86], as weIl as angle
resolved photoemission (APPES) [87-89]. These measurements confirm the
LDA predictions in detail, although they find a mass enhancement of 3-5 and
that the c-axis dispersion smaller than the already very small LDA value,
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
consistent with transport data [90J . It should be emphasized that this is

a high mass enhancement for a conventional superconductor, where the mass
enhancement would be governed by (1 + ).), ). being the electron-phonon
coupling.
The band structure, like that ofthe three-dimensional perovskite, SrRu03'
shows very strong Ru 4d - 0 2p hybridization, resulting in the same pattern,
i.e, bonding eg - pu states at the bottom of the valence bands, the correspond-
ing anti-bonding states above E F , and hybridized t 2g derived bands around
E F . There are three bands, crossing EF. These correspond to the three 4d
t2g orbitals, d xy , d xz and d yz •
The first significant difference from SrRu03 is that, in the I4/mmm
structure, the dxy orbital may have a crystal field splitting from the dxz
and d yz orbitals, and it is expected to disperse differently; this is because the
308 D.J. Singh
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.
0.4
0.3
~
g 0.2
0.1 Sr2RuO 4 Slaner Model
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
Fig.7.9. Bare band structure contribution to X in Sr2Ruü4 after [6]
fluctuations, which may suggest nearness to ferromagnetism. However, the

measurements can be interpreted in terms of a nearly q-independent sus-
ceptibility as weIl, and so it must be concluded that at present there is not
yet strong experimental support for the ferromagnetic background discussed
above.
Braden and co-workers [97] investigated the phonon dispersions of Sr2Ru04

using neutron scattering. They find, as do others, that Sr2Ru04 occurs in
the ideal bet I4/mmm structure. However, the L3 phonon branch exhibits
a sharp drop near the zone boundary at (7r, 7r, 0). This branch corresponds
to a rotation of the RU06 octahedra around the e-axis. The steepness of
this drop near the zone boundary is indicative of the rigidity of the RU06
octahedra, while the zone boundary frequency indicates the proximity to
an instability. The results also show that this branch has practically no kz
dispersion, indicating that there is very little coupling between rotations in
different layers.
In perovskite oxides, the stability of zone boundary rotational modes is
controlled by the A-site (here Sr) - 0 interaction. So substitution of a smaller
ion for Sr may be expected to destabilize this mode, and lead to a lower sym-
metry structure by condensation of rotational modes. This, in fact, happens
upon Ca substitution [58,63,98,99].
The structural and magnetic phase diagrams of Ca2-xSrxRu04 have been
presented in detail by Nakatsuji, Friedt and co-workers [63,98,99]. Only the
end-point compound Sr2Ru04 is superconducting. However, the ideal bct

14/mmm structure is retained from x = 2 to x = 1.5, at which point the
~3 phonon finally becomes unstable. Condensation of this zone boundary
mode leads to a doubling of the unit cell in spacegroup 141/ acd in the region
from x = 1.5 to x = 0.5. The octahedral rotation reaches 12.8° at x = 0.5.
Ca2-xSrxRu04 is a paramagnetic metal, with no magnetic ordering at low
temperature, throughout this range, x = 1.5 to x = 0.5, but becomes much
closer to a magnetic instability. Nakatsuju and Maeno [63] measured the
susceptibility X(O) at 2 K for several samples spanning this composition range
and found that X(O) increases strongly as xis lowered from x = 1.5 to x = 0.5
reaching a value more than 100 times X(O) for Sr2Ru04, with a high Wilson
ratio and a shape that suggests a critical point at x = 0.5. However, at x = 0.5
a structural transition occurs to a so-called "T" phase. The crystal structure
of this phase has not been fully refined, and possibly shows disorder or sample
dependence [100]. However, it is known that it has both octahedral rotations
about the c-mcis and octahedral tilting and exists below room temperature
from x = 0.2 to x = 0.5.
The onset of the tilt distortion at x = 0.5 coincides with a drop in the
susceptibility and the onset of low temperature magnetism as evidenced by
a temperature dependent susceptibility peak. However, a long range ordered
magnetic structure, corresponding to this peak, has not yet been found by
neutron diffraction. Related to this apparent lack of long range order, it
should be noted that the peak in X occurs at slightly different temperatures
depending on whether the in-plane or out-plane component is measured [63].
This is not the expected behavior for a true magnetic ordering transition and
suggests instead that dynamical fluctuations may be important in the phase
below the susceptibility peak. Friedt and co-workers [98] conclude that this
behavior of X(O) between x = 0.2 and x = 1.5 implies a coupling between
spin-fluctuations and the low lying (or condensed) tilt mo des between x = 0.5
and x = 1.5 (or x = 0.2 and x = 0.5). Another possibility is that it comes
from physics related to the unconventional low temperature properties of
Sr3Ru207, which shows a similar susceptibility peak but no magnetic ordering
down to very low temperature (see the discussion of the Ruddlesden-Popper
phases, below).
For Sr concentrations x ::; 0.2 the tilts are ordered into spacegroup Pbca,
reaching a tilt angle near 12°, with long range ordered antiferromagnetism at
low temperature. Two distinct structural phases, "S" and "L" occur between
x = 0 and x = 0.2 as shown schematically in Fig.7.1O. These differ in the
cl a ratio reflecting the geometry of the RU06 octahedra. The octahedra in
the "L" phase are like those in the "T" and 141/acd regions, while in the "S"
phase the apical oxygen is ~ 0.07 A closer to the Ru ion and the in plane
O's are ~ 0.05 A further. Importantly, the "S" phase unlike all the other
phases is an antiferromagnetic insulator, TN rv 110 K. This phase has been
characterized as a Mott insulator, based on the fact that it is insulating both
312 D.J. 8ingh
@Or-~r---~------~--------~------,
350
14/mmm
0~~~--~0.~5------~------~1~.5------~
Ca 2 Ruq X Sr 2 Ruq
Fig.7.10. 8chematic phase diagram of Ca2-x8rxRu04 following [63,98]. The re-

gion "T" is a tilted phase that is not fully refined. "8" and "L" have both rotations
and tilts in spacegroup Pbca, and differ in that "L" has a longer Ru - apical 0
bond length (and shorter Ru - plane 0) relative to "8"
above and below the magnetic ordering temperature, a fact confirmed by

transport, angle resolved phot emission (ARPE8) and optical measurements,
and shows a Curie-Weiss susceptibility above T N [57,59,63]. The gap inferred
from transport measurements is small, E g '" 0.2 eV [63]. For x » 0 there is
a sharp temperature dependent metal-insulator transition between the "8"
and "L" phases with a several orders of magnitude resistivity change [63].
8everal theoretical studies have been motivated by the richness of this
phase diagram. Density functional calculations for Ca2Ru04 were reported
by Woods [72] and Mazin and 8ingh [6]. Both of these mIthors find an antifer-
romagnetic instability within the L8DA. Fang and Terakura performed a very
detailed study of the interplay of structural, magnetic and electronic degrees
of freedom over the whole concentration range [71]. They find competing
ferromagnetic and antiferromagnetic tendencies with substantial coupling to
the lattice. In particular, they find, in their L8DA calculations, that rotation
of the RU06 octahedra favors ferromagnetism, while antiferromagnetism is
stabilized by octahedral tilts. They explained this in terms of the effects of
rotations and tilts on the d xy band, which tends to favor ferromagnetism, and
the nested d xz and dyz bands, which favor antiferromagnetism. For example,
they found that the rotation narrows the d xy more than the d xz or dyz,
increasing N (EF ) and favoring ferromagnetism.
Fang and Terakura also studied the effect of the compression of the octa-
hedra in the "8" phase, finding that this narrows the bands, as expected, and
also pro duces substantial orbital polarization, even in the L8DA, which favors
a commensurate antiferromagnetic phase due to superexchange interactions.
Significantly, they find a small band gap for antiferromagnetic Ca2 Ru04 in
their calculations. Mizokawa and co-workers, using 0 Is X-ray absorption and
spin-polarized photoemission, showed that Ca2Ru04 has a remarkably tem-
perature dependent orbital polarization and large t2g orbital moment [101].
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.
7.6 Ruddlesden-Popper Phases
As mentioned, the multilayer Ruddlesden-Popper compounds in the series,

(Sr,Ca)n+1Run 03n+l show a wide variety of magnetic orderings, metal insu-
lator transitions and unusual transport properties [59,64-70]. Unfortunately,
the strong coupling of magnetic, electronic and lattice degrees of freedom
that makes these ruthenates so interesting from a physical point of view,
also greatly complicates the study of these compounds. In particular, the
properties are know to vary substantially between different apparently high
quality sampIes. Nonetheless, chemical trends and the variety of behaviors
that can occur have been largely established. For example, the Ca compounds
tend to be more insulating, and have a greater tendency to antiferromag-
netism than the corresponding Sr compounds, and increasing the number of
layers, n, tends to produce more metallic behavior, and in the case of the Sr
compounds, ferromagnetism.
Here we focus on the bilayer Sr compound, Sr3Ru207 [103,104]. This is the
best studied of the multilayer ruthenates and is the most similar compound
to the superconductor, Sr2Ru04. Besides the obvious structural relationship,
314 D.J. Singh
.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
Fig. 7.12. Basal plane Fermi surfaces of ideal tetragonal structure Sr3Ru2Ü7 af-
ter [114]
is more k z dispersion in tetragonal Sr3Ru207 than in Sr2Ru04 , particularly

for the even bands.
The Fermi surface has a roughly square cylindrical hole pocket around
X, from odd combinations of d xz and dyz orbitals. Another, cross-shaped,
hole pocket comes from reconnected sections of the Fermi surfaces of the even
and odd parity sheets with the same orbital character. The even combination
also provides the innermost cylinder around r, which like the other three
r centered cylinders is electron-like. The second r centered electron-like
cylinder is also mainly dxz/d yz derived near the (10) directions, but is mixed
with the d xy character near (11).The two remaining r / Z centered cylinders
are derived from the d xy orbitals, and have little even-odd splitting, in ac cord
with their strongly in-plane orbital character. The three outer r centered
cylinders change topology as the electron count is increased as they touch
along the basal plane r-z lines giving rise to the van Hove singularities seen
above E F in the density of states, as in Sr2RuÜ4. It is noteworthy that even
before considering the effects of orthorhombicity, the Fermi surface nesting
is reduced relative to Sr2Ru04'
The reduced nesting relative to Sr2RuÜ4 suggests that incommensurate
antiferromagnetic spin-fluctuations seen in neutron scattering experiments
[93] on Sr2Ru04 may be less prominent in Sr3Ru207' However, tetragonal
Sr3Ru2Ü7 is much doser to a ferromagnetic instability. This is not surprising
considering the high value of the calculated density of states (DOS) at the
Fermi energy (E F ), N(EF ) = 4.5 states/eV Ru, compared with 4.1 states/eV
Ru for Sr2Ru04 [76]. According to LSDA calculations, tetragonal Sr3Ru207
is dose to a Stoner instability, but is paramagnetic. Singh and Mazin report
that orthorhombic Sr3Ru207 has a still higher N(EF) = 5.0eV-1, leading
to a magnetic ground state of itinerant character within the LSDA [114].
There are substantial 0 2p contributions to the DOS around EF , mostly from
the in-plane 0, which brings into play to Stoner mechanism for stabilizing
ferromagnetism over antiferromagnetism, discussed above for SrRu03'
316 D.J. Singh
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-
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
been reported, ferromagnetic in-plane order is strongly favored in Sr3Ru207,

unlike the situation in the alloy, (Sr,CahRu04. Further there is, of course, no
Sr/Ca disorder, as there is no Ca in Sr3Ru207. Because of this, a glassy state
due to magnetocrystalline coupling of Ru moments to disordered rotations of
the octahedra seems hardly as likely here. It also seems unlikely that a model
involving disorder in the sign of the exchange coupling of a local moment
magnet can be used for Sr3Ru207.
Finally, under applied pressures of 1 GPa, Ikeda and co-workers [107]
reported evidence for bulk ferromagnetism though with small moments (M ~
0.08 MB/Ru) starting at 70 K based on magnetization measurements. While
low moment ferromagnetism with this Curie temperature cannot be excluded,
such a high ratio of Tc / M is unusual in traditional materials.
Although these classical fluctuations due to the anisotropie magnetic in-
teractions in the Kosterlitz-Thouless scenario of Singh and Mazin may play
some role, it has recently been shown by transport data on very clean sam-
pIes in magnetic fields, that Sr3Ru207 has a novel quantum critical point
at a finite field of approximately 7T [109,117]. Because of this, one must
conclude that the explanation of Singh and Mazin is incorrect, and that in
fact Sr3Ru207 is a highly renormalized paramagnet with no ordering down
to OK. The ferromagnetic ordering predicted in the LSDA calculations, is
reflected in the susceptibility peak but long range static ordering is prevented
by quantum fluctuations that pers ist at T = O.
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
the electronic and magnetic properties of these ruthenates to lattice degrees

of freedom suggests technological applications. The key here is to get good
control of the sampie dependence in device geometries. Secondly, there has
been at least one big surprise in this area, when very clean, high quality
sampies were made, i.e. the unconventional superconductivity of Sr2 Ru04.
The recent availability of low residual resistivity samples of the highly renor-
malized paramagnet Sr3Ru207 resulted in the discovery of a novel quantum
critical point as a function of magnetic field. So one wonders what other
surprises may lurk, especially as the sampie qualities are improved in the
various ruthenate compounds.
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)
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
18. H. Kobayashi, M. Nagata, R. Kanno, and Y. Kawamoto: Mater. Res. Bull.

29, 1271 (1994)
19. A. Kanbayashi: J. Phys. Soc. Jpn. 44, 108 (1978)
20. G. Cao, S. McCall, M. Shepard, J.E. Crow, and RP. Guertin: Phys. Rev. B
56, 321 (1997)
21. L. Klein, J.S. Dodge, T.H. Geballe, A. Kapitulnik, A.F. Marshall, L. Antog-
nazza, and K Char: Appl. Phys. Lett. 66, 2427 (1995)
22. B.C. Chakoumakos, S.E. Nagler, S.T. Misture, and H.M. Christen: Physica
B 241, 358 (1998)
23. T. Kiyama, K Yoshimura, K Kosuge, H. Mitamura, and T. Goto: J. Phys.
Soc. Japan 68, 3372 (1999)
24. T. Kiyama, K Yoshimura, K Kosuge, H. Michor, and G. Hilscher: J. Phys.
Soc. Japan 67, 307 (1998)
25. K Yoshimura, T. Imai, T. Kiyama, KR Thurber, A.W. Hunt, and K Ko-
suge: Phys. Rev. Lett. 83, 4397 (1999)
26. H. Mukuda, K Ishida, Y. Kitaoka, K Asayama, R Kanno, and M. Takano:
Phys. Rev. B 60, 12279 (1999)
27. C.S. Alexander, G. Cao, S. McCall, and J.E. Crow: J. Appl. Phys. 85, 6223
(1999)
28. T. He and RJ. Cava: Phys. Rev. B 63,172403 (2001)
29. D.J. Singh: J. Appl. Phys. 79, 4818 (1996)
30. P.B. Allen, H. Berger, O. Chauvet, L. Forro, T. Jarlborg, A. Junod, B. Revaz,
and G. Santi: Phys. Rev. B 53, 4393 (1996)
31. 1.1. Mazin and D.J. Singh: Phys. Rev. B 56, 2556 (1997)
32. K Fujioka, J. Okamota, T. Mizokawa, A. Fujimori, I. Hase, M. Abbate, H.J.
Lin, C.T. Chen, Y. Takeda, and M. Takano: Phys. Rev. B 56, 6380 (1997)
33. G. Santi and T. Jarlborg: J. Phys. Conden. Matter 9, 9563 (1997)
34. LV. Solovyev: J. Magn. Magn. Mater. 177-181, 811 (1998)
35. 1.1. Mazin and D.J. Singh: J. Phys. Chem. Solids 59, 2185 (1998)
36. J.S. Dodge, E. Kulatov, L. Klein, C.H. Ahn, J.W. Reiner, L. Mieville, T.H.
Geballe, M.R Beasley, A. Kapitulnik, H. Ohta, Yu. Uspenskii, and S. Halilov:
Phys. Rev. B 60, 6987 (1999)
37. A.P. Mackenzie, J.W. Reiner, A.W. Tyler, L.M. Galvin, S.R Julian, M.R
Beasley, T.H. Geballe, and A. Kapitulnik: Phys. Rev. B 58, 13318 (1998)
38. L. Klein, Y. Kats, A.F. Marshall, J.W. Reiner, T.H. Geballe, M.R Beasley,
and A. Kapitulnik: Phys. Rev. Lett. 84, 6090 (2000)
39. L. Klein, A.F. Marshall, J.W. Reiner, C.H. Ahn, T.H. Geballe, M.R. Beasley,
and A. Kapitulnik: J. Magn. Magn. Mater. 188, 319 (1998)
40. L. Klein, J.S. Dodge, C.H. Ahn, G.J. Snyder, T.H. Geballe, M.R Beasley,
and A. Kapitulnik: Phys. Rev. Lett. 77, 2774 (1996)
41. L. Klein, J.S. Dodge, C.H. Ahn, J.W. Reiner, L. Mieville, T.H. Geballe, M.R.
Beasley, and A. Kapitulnik: J. Phys. Condens. Matter 8, 10111 (1996)
42. P. Kostic, Y. Okada, N.C. Collins, Z. Schlesinger, J.W. Reiner, L. Klein, A.
Kapitulnik, T.H. Geballe, and M.R Beasley: Phys. Rev. Lett. 81, 2498 (1998)
43. J. Okamoto, T. Mizokawa, A. Fujimori, I. Hase, M. Nohara, H. Takagi, Y.
Takeda, and M. Takano: Phys. Rev. B 60, 2281 (1999)
44. S.C. Gausepohl, M. Lee, RA. Rao, and C.B. Eom: Phys. Rev. B 54, 8996
(1996)
45. W.E. Pickett and D.J. Singh: Phys. Rev. B 53, 1146 (1996)
7 Magnetisrn in Ruthenates 321
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)
107. S. Ikeda, Y. Maeno, S. Nakatsuji, M. Kosaka, and Y. Uwatoko: Phys. Rev. B

62, 6089 (2000)
108. RS. Perry, L.M. Galvin, A.P. Mackenzie, D.M. Forsythe, S.R Julian, S.1.
Ikeda, and Y. Maeno: Physica B 284, 1469 (2000)
109. RS. Perry, L.M. Galvin, S.A. Grigera, L. Capogno, A.J. Schofield, A.P.
Mackkenzie, M. Chiao, S.R. Julian, S.1. Ikeda, S. Nakatsuji, Y. Maeno, and
C. PReiderer: Phys. Rev. Lett. 86, 2661 (2001)
110. Y. Inoue, M. Hara, Y. Koyama, S. Ikeda, Y. Maeno, and T. Fujita, in Ad-
vances in Superconductivity IX, p. 281, S. Nakajima, and M. Murakami, eds.
(Springer, Berlin Heidelberg New York 1997)
111. H. Shaked, J.D. Jorgensen, O. Chmaissem, S. Ikeda, and Y. Maeno: J. Solid
State Chem. 154, 361 (2000)
112. H. Shaked, J.D. Jorgensen, S. Short, O. Chmaissem, S.1. Ikeda, and Y. Maeno:
Phys. Rev. B 62, 8725 (2000)
113. I. Hase and Y. Nishihara, J. Phys. Soc. Japan 66, 3517 (1997)
114. D.J. Singh and 1.1. Mazin: Phys. Rev. B 63, 165101 (2001)
115. Y. Maeno, K. Yoshida, H. Hashimoto, S. Nishizaki, S.-1. Ikeda, M. Nohara,
T. Fujita, A.P. Mackenzie, N.E. Hussey, J.G. Bednorz, and F. Lichtenberg:
J. Phys. Soc. Jpn. 66, 1405 (1997)
116. J.M. Kosterlitz, and D.J. Thouless: J. Phys. C 6, 1181 (1973)
117. S.A. Grigera, RS. Perry, A.J. Schofield, M. Chiao, S.R Julian, G.G. Lon-
zarich, S.1. Ikeda, Y. Maeno, A.J. Millis, and A.P. Mackenzie: Science 294,
329 (2001)
Index
(Al,Mn)As 180 Cr 220

(Ga,Mn)As 163, 174, 175, 177, 180,
181, 188 double exchange 138, 254-261,
(Ga,Mn)N 163, 173 272-274, 276, 281, 301
(In,Mn)As 163, 180, 181 Dy 27,242
(La,Ba)Mn03 269, 271, 280-282, 285, dynamical mean field theory 106, 107,
287, 288 109-116, 119-122, 124, 145-147, 149,
(La,Ca)Mn03 138, 139, 254, 266, 270, 150, 152-154
271 Dzyaloshinski-Moria interaction 7, 22,
(La,Pb)Mn03 270,271 115, 225, 226, 239, 241, 278
(La,SrhMn04 264, 283, 284, 289
Er 27, 47, 242-245
(La,Sr)Mn03 253, 270, 281, 283, 288,
Eu 25
289
(La,Sr)Ti03 147, 149, 150 Fe 17, 20, 22, 23, 32, 34, 45, 53, 59,
(Nd,Sr)Mn03 253, 271, 288 63, 70, 72-74, 82, 84, 88-91, 96-98,
(Pr,Sr)Mn03 262, 271, 283 119-122,220,224
(Sr,CahRu04 145-147,310-313,317 Fe203 219, 225-228
(Sr,Ca)Ru03 303, 304 Fe304 302
fluctuation exchange approximation
adiabatic approximation 1, 4, 5, 110-112, 119-121
10-15,17,22,24,29-31 frozen-magnon method 12
anomalous Hall effect 177-182
Gaussian broadening method 80, 82,
Berry curvature 13, 14, 178 83
Berry phase 178, 179, 182, 184 Gd 27, 40, 43, 117, 118, 242
GW approximation 102
Ca2Ru04 306,313
CaCu02 122 Ho 27, 45, 242
CaMn03 253, 257, 290 Holstein-Primakoff theory 5, 7, 183,
CaRu03 298, 303, 304 202, 205, 206
CaV205 132-136 iterated perturbation theory 111, 112,
Ce 25
147, 149
CE-type structure 253, 262-266,
283-285 Jahn-Teller distortion 136, 265,
charge ordering 139, 265 274-278, 284, 285, 289
Co 19, 20, 32, 45, 72, 73, 84, 88
CoNi2 66 KCuF3 136-138
CoO 122 Kosterlitz-Thouless 316, 317
326 Index
La2CU04 123 quantum critical point 306, 318

LaMn03 253, 271, 274-280
LDA+U Method 103-106, 110, 117, RKKY 12, 22, 40, 116, 182, 183, 187,
118, 122-124, 127, 129-132, 136-139, 191, 192, 199, 201, 271
145, 150-155, 167, 245, 267, 279, 283
Lu 25 self interaction correction 27, 102,
124, 245
magneto-optical Kerr effect 246-248, shape anisotropy 58, 60, 69
270 Sm 25
magnetocrystalline anisotropy 6, spin-orbit 4, 6, 19, 22, 26-28, 37, 42,
59-62, 64-66, 69-77, 80, 82, 83, 57, 61, 66, 67, 69, 71, 75, 77-79, 81,
85-89,91-98, 174,239, 242, 299 82, 87-91, 96, 97, 213, 214, 219, 224,
mean-field theory 170-173, 187 226, 228, 231, 235-237, 239, 242, 243,
MgV205 132-136 260, 271, 297
Mn 273 spin-spiral 6, 16, 22, 24, 32, 34, 46, 48
Mn3Sn 219, 231, 232 Sr2Ru04 141-145, 147,297, 298,
MnO 123, 124, 127, 128 306-310
MnSi 239 Sr3Ru207 313-318
SrRu03 298-302, 305
NaV205 150-154
Tb 27,242
Ni 17-20,22,37,45, 52, 70, 72-74, 80,
tetrahedron method 76, 77, 79
82, 84, 85, 88, 90-92, 96-98, 119, 220
The LDA+U Method 103
NiO 123, 124, 127
Tm 48
non-crossing approximation 112, 113,
triplet superconductivity 297
146, 147, 149
U2Pd2Sn 220, 221
orbital moment 4, 5, 8, 19, 25, 27, 28, U3P4 223, 224, 246-248
38, 48, 70, 72, 86-88, 213, 224, 245, UFe4Al8 232-235
260 UPdSn 228-230
orbital ordering 106, 128, 132, UPtGe 236, 237, 239, 241
136-139 USb 220,221
orbital polarization 52, 87, 88, 245
Pd 303
Prl/2Cal/2Mn03 139, 140 Yb 25
Springer Series in
MATERIALS SCIENCE
Editors: R. Hull R. M. Osgood, Jr. J. Paris i
1 Chemical Processing with Lasers 15 Crystal Chemistry of

By D. Bäuerle High-Tc Superconducting Copper Oxides
By B. Raveau, C. Michel, M. Hervieu,
2 Laser-Beam Interactions with Materials
and D. Groult
Physical Principles and Applications
By M. von Allmen and A. Blatter 16 Hydrogen in Semiconductors
2nd Edition By S. J. Pearton, M. Stavola,
and J. W. Corbett
3 Laser Processing of Thin Films
and Microstructures 17 Ordering at Surfaces and Interfaces
Oxidation, Deposition and Etching Editors: A. Yoshimori, T. Shinjo,
of Insulators and H. Watanabe
By. I. W. Boyd
18 Graphite Intercalation Compounds 11
4 Microc1usters Editors: S. A. Solin and H. Zabel
Editors: S. Sugano, Y. Nishina, and S. Ohnishi
19 Laser-Assisted Microtechnology
5 Graphite Fibers and Filaments By S. M. Metev and V. P. Veiko
By M. S. Dresselhaus, G. Dresselhaus, 2nd Edition
K. Sugiliara, I. L. Spain, and H. A. Goldberg
20 Microc1uster Physics
6 Elemental and Molecular Cltisters By S. Sugano and H. Koizumi
Editors: G. Benedek, T. P. Martin, 2nd Edition
and G. Pacchioni
21 The Metal-Hydrogen System
7 Molecular Beam Epitaxy ByY. Fukai
Fundamentals and Current Status
22 Ion Implantation in Diamond,
By M. A. Herman and H. Sitter 2nd Edition
Graphite and Related Materials
8 Physical Chemistry of, in and on Silicon By M. S. Dresselhaus and R. Kalish
By G. F. Cerofolini and L. Meda
23 The Real Structure
9 Tritium and Helium-3 in Metals of High-Tc Superconductors
ByR. Lässer Editor: V. Sh. Shekhtman
10 Computer Simulation 24 Metal Impurities
ofIon-Solid Interactions in Silicon-Device Fabrication
By W. Eckstein ByK. Graff 2nd Edition
11 Mechanisms of High 25 Optical Properties of Metal Clusters
Temperature Superconductivity By U. Kreibig and M. Vollmer
Editors: H. Kamimura and A. Oshiyama
26 Gas Source Molecular Beam Epitaxy
12 Dislocation Dynamics and Plasticity Growth and Properties ofPhosphorus
By T. Suzuki, S. Takeuchi, and H. Yoshinaga Containing 111-V Heterostructures
By M. B. Panish and H. Temkin
13 Semiconductor Silicon
Materials Science and Technology 27 Physics of New Materials
Editors: G. Harbeke and M. J. Schulz Editor: F. E. Fujita 2nd Edition
14 Graphite Intercalation Compounds I 28 Laser Ablation
Structure and Dynamics Principles and Applications
Editors: H. Zabel and S. A. Solin Editor: J. C. Miller
Springer Series in
MATERIALS SCIENCE
Editors: R. Hull R. M. Osgood, Jr. J. Paris i
29 Elements of Rapid Solidification 41 Organic Electronic Materials

Fundamentals and Applications Conjugated Polymers and Low
Editor: M. A. Otooni Molecular Weight Organic Solids
Editors: R. Farchioni and G. Grosso
30 Process Technology
for Semiconductor Lasers 42 Raman Scattering
Crystal Growth in Materials Science
and Microprocesses Editors: W. H. Weber and R. Merlin
By K. Iga and S. Kinoshita
43 The Atomistic Nature
31 Nanostructures of Crystal Growth
and Quantum Effects By B. Mutaftschiev
By H. Sakaki and H. Noge
44 Thermodynamic Basis
32 Nitride Semiconductors and Devices of Crystal Growth
ByH. Morko~ P-T-X Phase Equilibrium
and Non-Stoichiometry
33 Supercarbon By J. Greenberg
Synthesis, Properties and Applications
Editors: S. Yoshimura and R. P. H. Chang 45 Thermoelectrics
Basic Principles
34 Computational Materials Design and New Materials Developments
Editor: T. Saito By G. S. Nolas, J. Sharp,
and H. J. Goldsmid
35 Macromolecular Science
and Engineering 46 Fundamental Aspects
New Aspects of Silicon Oxidation
Editor: Y. Tanabe Editor: Y. J. Chabal
36 Ceramics 47 Disorder and Order

Mechanical Properties, in Strongly
Failure Behaviour, Nonstoichiometric Compounds
Materials Selection Transition Metal Carbides,
By D. Munz and T. Fett Nitrides and Oxides
By A. I. Gusev, A. A. Rempel,
37 Technology and Applications and A. J. Magerl
of Amorphous Silicon
Editor: R. A. Street 48 The Glass Transition
Relaxation Dynamics
38 Fullerene Polymers in Liquids and Disordered Materials
and Fullerene Polymer Composites ByE. Donth
Editors: P. C. Eklund and A. M. Rao
49 Alkali Halides
39 Semiconducting Silicides A Handbook of Physical Properties
Editor: V. E. Borisenko By D. B. Sirdeshmukh, L. Sirdeshmukh,
and K. G. Subhadra
40 Reference Materials
in Analytical Chemistry 50 High-Resolution Imaging
A Guide for Selection and Use and Spectrometry of Materials
Editor: A. Zschunke Editors: F. Ernst and M. Rühle

Electronic Structure and Magnetism of Complex Materials 2003

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Electronic Structure and Magnetism of Complex Materials 2003

Uploaded by

Copyright:

Available Formats

Springer Series in

Springer-Verlag Berlin Heidelberg GmbH

51 Point Defects 56 Si0 2 in Si Microdevices

53 Nanocrystalline Ceramics 59 Die1ectric Properties of Porous Media

Series homepage - http://www.springer.de/phys/books/ssms/

Volumes 1-50 are listed at the end ofthe book.

With 159 Figures

Professor R. M. Osgood, Jr.

Professor Jiirgen Parisi

ISBN 978-3-642-07774-6 ISBN 978-3-662-05310-2 (eBook)

Typesetting by the editors

Magnetism is today properly in the forefront of materials research. Key tech-

Washington, D.C. David J. Singh

1 Low-Lying Magnetic Excitations in Itinerant Systems:

2 Calculation of Magneto-crystalline Anisotropy

3.8 Highly Correlated Metallic Compounds

4 Ferromagnetism in (III,Mn)V Semiconductors

5 Noncollinear Magnetism in Systems

5.4 Stable Magnetic Structures .................................. 219

6 Orbital Degeneracy and Magnetism

Index ......................................................... 325

Vladimir I. Anisimov J ürgen König

Samed Halilov A.I. Lichtenstein

Mikhail I. Katsnelson John Schliemann

Guenter S. Schneider Kiyoyuki Terakura

1.2 Magnetic Hamiltünian für Itinerant Systems

H = Ho + Hint + Hsr + Hso + Hext + ... , (1.1)

where the contributions are as follows:

X (V' - 2V")]-J;U2(r')-J; U3(r) , (1.5)

+ItB~ L J dr-J;t1 (r)A+(r, t)A(r, t)-J;U1 (r)

stand for the Coulomb electron-nuclei and electron-electron interactions, re-

V'2 vee (1 r - r ' I) = -I r -2r 13 .

where the creation-destruction operators ct,

m(r) = ~ ~)mn8(r - r') + 8(r - r')mn] (1.12)

is contributed by individual total angular momenta m n = Sn + in built from

reflect the domain structure of the effective exchange-correlation hole, and

H = Ho + Hexe + Hmm + Hdem + ...

Hexe = -~ L Jii/SiSi/ - 4\ L Jijkz[Si X Sj][Sk X szl + ... , (1.16)

iI;xc+SOI == 2: Dii,Si X Si' , (1.17)

where D ii , is a function of the lattiee sites ii'. This term is known to be

1.3 From Postulat ion to Computation:

i(r, t) = ~ 2: -J;!(r, t)i-J;s,(r, t), (1.18b)

s(r,t) = (s(r,t)) = ~tr [n(r,t)a] , (1.19)

where n(r, t) is the spin density matrix

and tr means the trace with respect to spin indices.

-J;i(r, s) = l: cPk(r, s)ak, (1.21)

Si(t) = (Si(t) = ~ [ . tr[O'n(r, t)]d3 r. (1.23)

where, in the presence of an external magnetic field h( r),

v(r) = ivions(r) + iJdr,~tr[11(r?? + h(r) .8-+ i~ " 8~xc ,(1.26a)

Llv(r) = ~2 (8E xc _ 8Exc ) .

The unitary spin marix U(r) is to be taken in such a way as to diagonalize

[U11U+] ij = 8ij n i , (1.27a)

U(r) = e!e(r).'" , (1.27b)

Exc[nJ = J drn(r)E xc (nl(r),n2(r)) = J

E~~(")l[nJ = J drn(r)E xc (n(r), e(r) ·1n(r)) . (1.29)

1.4 Equilibrium Configuration

1.4.1 General Formulation of the Adiabatic Spin Dynamics

where astands for the transverse components of m.

To decouple spin and charge excitations of a many-electron system, the

E[mi] = min7ni = Iv;, dr(,pla-I,p), ('ljJ I iI 1'ljJ) , (1.37)

{)'ljJ '" ~ {)'ljJ '" ~'ljJo (1.41)

which gives for the excitation energies

1.4.2 Frozen Spin-Wave Method as an Approximation

The following picture is evocative of the semiclassical method, in accordance