Springer-Science+Business Media, B.V.

Concepts in Electron Correlation

NATO Science Series
A Series presenting the results of scientific meetings supported under the NATO Science
Programme.

The Series is published by IOS Press, Amsterdam, and Kluwer Academic Publishers in conjunction
with the NATO Scientific Affairs Division

Sub-Series

I. Life and Behavioural Sciences IOS Press

II. Mathematics, Physics and Chemistry Kluwer Academic Publishers
III. Computer and Systems Science IOS Press
IV. Earth and Environmental Sciences Kluwer Academic Publishers
V. Science and Technology Policy IOS Press

The NATO Science Series continues the series of books published formerly a s the NATO ASI Series.

The NATO Science Programme offers support for collaboration in civil science between scientists of
countries of the Euro-Atlantic Partnership Council. The types of scientific meeting generally supported
are "Advanced Study Institutes" and "Advanced Research Workshops", although other types of
meeting are supported from time to time. The NATO Science Series collects together the results of
these meetings. The meetings are co-organized bij scientists from NATO countries and scientists from
NATO's Partner countries - countries of the CIS and Central and Eastern Europe.

Advanced Study Institutes are high-level tutorial courses offering in-depth study of latest advances
in a field.
Advanced Research Workshops are expert meetings aimed at critical assessment of a field, and
identification of directions for future action.

A s a consequence of the restructuring of the NATO Science Programme in 1999, the NATO Science
Series has been re-organised and there are currently Five Sub-series as noted above. Please consult
the following web sites for information on previous volumes published in the Series, as well as details of
earlier Sub-series.

http://www.nato.int/science
http://www.wkap.nl
http://www.iospress.nl
http://www.wtv-books.de/nato-pco.htm

Series II: Mathematics, Physics and Chemistry - Vol. 110

Concepts in Electron Correlation

edited by

Alex C. Hewson
Department of Mathematics,
Imperial College, London, United Kingdom

and

Veljko Zlatic
Institute of Physics,
Zagreb, Croatia

Springer-Science+Business Media, B.V.

Proceedings of the NATO A d v a n c e d R e s e a r c h Workshop on
Concepts in Electron Correlation
Hvar, Croatia
September 2 9 - O c t o b e r 3, 2002

A C.I.P. Catalogue record for this book is available from the Library of C o n g r e s s .

ISBN 978-1-4020-1419-2 ISBN 978-94-010-0213-4 (eBook)

DOI 10.1007/978-94-010-0213-4

Printed on acid-free paper

All Rights Reserved

© 2 0 0 3 Springer Science+Business Media Dordrecht
Originally published by Kluwer A c a d e m i c Publishers in 2003
Softcover reprint of the hardcover 1st edition 2003
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming,
recording or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
Contents

Preface 5
D. van del' Marel, H. J. A. Molegraaf, C. Presura, 1. Santoso/
Superconductivity by Kinetic Energy Saving? 7
R. Hlubina/ Josephson effect in the cuprates: microscopic
implications 17
M. Sigrist/ Ruthenates: Unconventional superconductivity and
magnetic properties 27
K-D. Schotte, C.T. Liang/ Real structure of perovskites looked at
from the band structure point of view 35
H. Eschrig, K Koepernik, 1. Chaplygin/ Towards models of
magnetic interactions in the cuprates 45
J. Rohler/ c-axis intra-layer couplings in the CU02 planes of
high-T c cuprates 55
P. Prelovsek, A. Ramsak / Spectral functions and pseudogap in a
model of strongly correlated electrons 65
G. Baskaran/ Mott insulator to superconductor via pressure RVB
theory & Prediction of New Systems 75
M. Lang, J. Miiller, F. Steglich, J. Schlueter, T. Sasaki/ Exploring
the phase diagram of the quasi-2D organic superconductors K,
-(BEDT-TTF)2X 85
M. Fabrizio, E. Tosatti, M. Capone, C. Castellani/ Enhancement
of superconductivity by strong correlations: a model study 95
J.K Freericks, T.P. Devereaux, R. Bulla/ Inelastic light scattering
and the correlated metal-insulator transition 115
L.F. Feiner, A.M. Oles/ Orbital physics versus spin physics: the
orbital-Hubbard model 123
J. Bah, A. M. Oles/ Quasiparticles in photoemission spectra of
manganites 133
2

H. v. Lahneysen/ Metals near a zero-temperature magnetic

instability 143
H. v. Lahneysen/ Heavily doped semiconductors: magnetic
moments, electron-electron interactions and the metal-
insulator transition 155
P. Fazekas, A. Kiss/ Competition and coexistence of magnetic and
quadrupolar ordering 169
B. Liithi, B. Wolf, S. Zherlitsyn/ Low Dimensional Spin Systems
in High Magnetic Fields: Spin-Phonon Interaction 179
N.P.Armitage, E. Helgren , G. Griiner/ 'Taxonomy' of electron
glasses 189
A.C. Hewson, D. Meyer/ Renormalization group approaches
for systems with electron-electron and electron-phonon
interactions 199
R. Bulla, M. Vojta/ Quantum phase transitions in models of
magnetic impurities 209
M. Lavagna, A. Jerez, D. Bensimon/ Instability of the Fermi-liquid
fixed point in an extended Kondo model 219
A. Lobos, A. A. Aligia/ Projection of the Kondo effect by resonant
eigenstates inside a circular quantum corral 229
P. Walfle, A. Rosch, J. Paaske, J. Kroha/ Nonequilibrium electron
transport through nanostructures: correlation effects 239
T. A. Costi/ Quantum fluctuations and electronic transport
through strongly interacting quantum dots 247
J. Spalek, E.M. Garlich, A. Rycerz, R. Zahorberiski, R. Podsiadly,
W. Wojcik/ Properties of correlated nanoscopic systems from
the combined exact diagonalization - ab initio method 257
N. Andrei, C. J. Bolech/ On the multichannel-channel Anderson
impurity model of uranium compounds 269
B. Coqblin/ Anomalous behavior in rare-earth and actinide
systems. 277
V. Zlatic, J.K. Freericks/ Describing the valence-change transition
by the DMFT solution of the Falicov-Kimball model 287
A.P. Murani/ Neutron spectroscopy of valence fluctuation
compounds of Cerium and Ytterbium. 297
 3

H. Keiter, K. Baumgartner, D. Ottoj Generalizations of DMFT,

CPA and NCA 307
J. 130nea, S. El Shawish, C.D. Batista, J.E. Gubernatisj Itinerant
ferromagnet,ism for mixed valence systems 3 I7
C. Grenzebach, G.CzychoUj Transport properties of heavy
Fermion systems 327
N. Blumer, P. G. J. van Dongenj Transport properties of correlated
electrons ;n h;gh d;mens;ons 335
J.L. Sal'l'aa, J.D. Thompsonj From Ce1n3 t.o PuCoCas: Trends in
heavy fermion Supcl'conduct.ivity 345
E. V. Sarnpathkumaran, R. Ivlallikj Do we understand electron
correlation effects in Gadolinium based intermetallic
compounds? 353
L. Degiorgij Optical properties of correlated systems 363
J.E. Hirschj Quasipart.icle undressing: a new route to collective
effects in solids 371
List of Contributors 381
Tndex 383
Preface

The NATO sponsored Advanced Research Workshop on "Concepts in

Electron Correlation" took place on the Croatian island of Hvar during
the period from the 29th of September to the 3rd of October, 2002. The
topic of electron correlation is a fundamental one in the field of condensed
matter, and one that is being very actively studied both experimentally
and theoretically at the present time. The manifestations of electron cor-
relation are diverse, and play an important role in systems ranging from
high temperature superconductors, heavy fermions, manganite compounds
with colossal magnetoresistance, transition metal compounds with metal-
insulator transitions, to mesoscopic systems and quantum dots. The aim of
the workshop was to provide an opportunity for a dialogue between exper-
imentalists and theoreticians to assess the current state of understanding,
and to set an agenda for future work. There was also a follow-up workshop
on the same topic where the presentations included more background and
introductory material for younger researchers in the field.
The papers presented in these proceedings clearly demonstrate the di-
versity of current research on electron correlation. They show that real
progress is being made in characterising systems experimentally and in
developing theoretical approaches for a quantitative comparison with ex-
periment. The more one learns, however, the more there is to understand,
and many of the contributions help to map out the territory which has yet
to be explored. We hope that the articles in this volume will be a stimulus
for such future work.

February, 2003 A. C. Hewson, V. Zlatic

5
SUPERCONDUCTIVITY BY KINETIC ENERGY SAVING?

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

D. VAN DER MAREL, H. J. A. MOLEGRAAF, C. PRE-

SURA and 1. SANTO SO
Materials Science Centre, University of Groningen, 9747 GA
Groningen, The Netherlands

Abstract. A brief introduction is given in the generic microscopic framework of super-

conductivity. The consequences for the temperature dependence of the kinetic energy,
and the correlation energy are discussed for two cases: The BCS scenario and the non-
Fermi liquid scenario. A quantitative comparison is made between the BCS-prediction
for d-wave pairing in a band with nearest neighbor and next-nearest neighbor hoppping
and the experimental specific heat and the optical intraband spectral weight along the
plane. We show that the BCS-prediction produces the wrong sign for the kink at T c of
the intraband spectral weight, even though the model calculation agrees well with the
specific heat.

Key words: Optical Conductivity, Spectral Weight, Specific Heat, Pair Correlation, Kin-
etic Energy, Correlation Energy, Internal Energy

1. Model independent properties of the superconducting state

1.1. INTERNAL ENERGY

When we cool down a superconductor below the critical temperature, the

material enters a qualitatively different state of matter, manifested by
quantum coherence over macroscopic distances. Because the critical tem-
perature represents a special point in the evolution of the internal energy
versus temperature, the internal energy departs from the temperature de-
pendence seen in the normal state when superconductivity occurs. Because
for T < T c the superconducting state is the stable equilibrium state, the
internal energy in equilibrium at T = 0 is an absolute minimum. Hence
cooling down from above the phase transition one would expect a drop in
the internal energy when at the critical temperature. This drop of internal
energy stabilizes the superconducting phase among all alternative states of

7
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 7–16.
© 2003 Kluwer Academic Publishers
8

200

sz
ro
-2' 100

~
U

~ 20 1.5
OJ
'2 E
,., 15 o
OJ
Q; 1.0 ro::J
C
U
:>
W 10
roc
Q; 5
0.5 E
C
0+-_~~~~~~~~~---10.0
o 50 100 150 200
Temperature (K)

Figure 1. Experimental internal energy and the specific heat obtained by Loram et
al (1). The extrapolations are obtained following the procedure of Ref. (3)

matter. An experimental example of this well known behavior is displayed

in Fig. 1, where the internal energy was calculated from the electronic
specific heat data (1) according to the relation Eint(T) = JOT c(T')dT'. The
broadened appearance of the phase transition suggests that the supercon-
ducting correlations disappear rather gradually when the temperature is
increased above the phase transition, an effect which remains noticeable in
the specific heat graph up to 125 K.
Understanding the mechanism of superconductivity means to understand
what stabilizes the internal energy of the superconducting state. In BCS
theory superconductivity arises as a result of a net attractive interaction
between the quasi-particles of the normal state. Note, that implicitly this
approach is firmly rooted in the paradigm of a Fermi-liquid type normal
state. On the other hand, the school based on Anderson's original work (2)
asserts, that a strong on-site repulsive interaction can also give rise to high
T c superconductivity. The latter models typically require that the material
is not a Fermi liquid when superconductivity is muted, either by raising
the temperature or by other means.

1.2. PAIR-CORRELATIONS

Without loss of generality, i. e. independent of the details of the mechanism

which leads to superconductivity, it is possible to provide a microscopic
definition of the superconducting state. For this purpose let us consider the
 9

correlation function defined as

G(r,R 1,R2) = \1Pf(R 1 +r)1PI(R1)7!Jl(R2)7/Ji(R2 +r)) (1)

where 1Pt(Rj ) is a single electron creation operator. This function defines

two types of correlation: (i) the electron pair-correlation as a function of the
relative coordinate r, and (ii)v the correlation between two pairs located
at center of mass coordinates R 1 and R 2 . In the normal state there is no
correlation of the phase of G(r, R 1, R 2) over long distances IR1 - R 2 due
1

to the finite mean free path of the electrons. As a result the integral over
the center of mass coordinates of the correlation function

(2)

averages to zero in the normal state. In contrast the superconducting state

is characterized by long range phase coherence of the center of mass coordin-
ates, implying (among other things) that the correlation function averaged
over all center of mass coordinates, g(r), is a finite number.

2. Internal energy and its decomposition using the BeS model

2.1. BCS: CORRELATION FUNCTION, AND CORRELATION ENERGY

FOR D-WAVE PAIRING

In the weak coupling scenario of BeS theory the electrons have an effective
attractive interaction, as a result of which they tend to form pairs. For the
purpose of the present discussion we will assume that the interaction is of
the form
(3)
where n(r) is the electron density operator. The interaction energy in the
superconducting state becomes lower than in the normal state, due to the
fact that the effective attractive interaction favors a state with enhanced
pair-correlations. The value of the interaction energy of the superconducting
state, relative to the normal state is

\Hi)s - \Hi)n = J d3 rg(r)V(r) = LgkVk

k
(4)

where gk and Vk are the Fourier transforms of g(r) and V(r) respectively.
Using the Bogoliubov transformation the correlation function can be ex-
pressed (4) in terms of the gap-function D..k and the single particle energies
Ek = {(Ek - p,)2 + D..k}1/2.

(5)
10

0.2

Reciprocal space (n,O)

--...
en

Real space (3,0)

Figure 2. The k-space (top panel) and coordinate space (bottom panel) representation
of the superconductivity induced change of pair-correlation function for d-wave symmetry
(bottom panel). Parameters: 6/W = 0.2, WLJ/W = 0.2. Doping level: x = 0.25

This corresponds to the conversion of a pair (q, -q) to a pair with quantum
numbers (q + k, -q - k). In the expression for the correlation energy, Eq. 4,
the transferred momentum k is carried by interaction kernel Vk .
Starting from a model expression for the single electron energy-momentum
dispersion Ek, and the gap-function D..k, it is a straightforward numer-
ical exercise to calculate the summations in Eq. 5. Adopting the nearest
neighbor tight-binding model with a d-wave gap, and adopting the ratio
D..Clf,O)/W = 0.2, where W is the bandwidth, the correlation function gk
can be easily calculated, and the result is shown in Fig. 2. We see from this
graph that a negative value of (Hi) s - (Hi)n requires either (i) Vk < 0 for
k in the neighborhood of the origin, or (ii) Vk > 0 for k in the vicinity of
Clf, Jr). The corresponding representation in real space, g(r), shown in Fig. 2,
illustrates that the dominant correlation of the d-wave superconducting
state is of pairs where the two electrons occupy a nearest neighboring site,
while the on-site amplitude is zero.
Combining the information of Fig. 2 with Eq. 2, it is clear that the strongest
saving of correlation energy is expected if the electrons interact with an
interaction of the form V(rl' r2) = Va La b(rl - r2 + a) where the vector a
 11

15

'>
Q)

.s 10

E
~
<1

-138 Input parameters for BCS model:

Vk,q = Va (cosk x - cosk)(cosqx - cosqy)
Sk = -f-I + tcosk x + tcosk y + 2t'coskXcosky
~ -139 Vc = 88 (meV)

.s t = -297.6 (meV)
t' = 81.8 (meV)
i=' -140
'::;:

-141

50 100 150 200

Temperature (K)
Figure 3. BeS prediction of the d-wave gap-function and the chemical potential.

runs over nearest neighbor sites, and Va is a negative (meaning attractive)

interaction between electrons on nearest neighbor sites.

2.2. BCS: THE GAP-EQUATION, SPECIFIC HEAT AND INTERNAL

ENERGY

To illustrate the predictions of BCS theory for the temperature dependence

of the correlation energy and the kinetic energy, we start by solving the gap
equation
A
Uk = '""""
~
Vk-q~q t ah n
( -
Ek- ) (6)
q 2Eq 2k B T
which must solved together with the constraint that the average number
of particles is temperature independent. This requires that the chemical
potential must be adjusted to keep the thermal average of L;k (nk) = N at
a constant value. In the numerical examples of this paper we have adopted
N = 0.85 corresponding to a hole doping of 0.15. This solution of the gap
equation is shown in Fig. 3. Notice, that the temperature dependence of the
chemical potential follows closely the experimental observations reported
for Y123 in Ref. (7).
To check that the band-parameters used here are reasonable, we display in
Fig. 4 the corresponding prediction for the specific heat and the internal
energy, using the same parameters as for Fig. 3. If we compare this to Fig. 1,
we see that the band-parameters adopted here quantitatively reproduce the
observed specific heat. Hence the present set of band-parameters (i,i') and
12

15

Q'
;>Q) 10
3
i='
U 5

ol:--"o"~--+-+-+--+-+-+-+-+-+--+->--t--..---.--+-+-j

>' Input parameters for BCS model:

CD Vk,q = Vo (cask, - cosk){cosqx - cosq)
S 1.5 £k = tcosk + tcosk + 2t'coskXcosk
x y y
>, V, = 88 (meV)
~ t = -297.6 (meV)
~ 10 t' = 81.8 (meV)
W
ro
E 0.5
Q)
C
o 50 100 150 200

Temperature (K)

Figure 4. Bes prediction of the internal energy and the specific heat.

the coupling parameter va

represent the best phenomenological choice for
quantitative testing of the BCS-model.

2.3. BCS: TEMPERATURE DEPENDENCE OF THE CORRELATION

ENERGY AND THE KINETIC ENERGY

The BCS prediction for the temperature dependence of the average in-
teraction energy follows from Eq. 4. We then use the BCS variational
wave-function for the statistical average of Eq. 5, resulting in

(7)

Simultaneously there is an increase of the 'kinetic' energy

(8)

In Fig. 5 this is displayed, using the same parameters as for Fig. 3.

 13

-227.0
:;- Weak coupling BCS-theory
OJ
t' / t = -0.27
-S d-wave pairing, T c = 80 K
~ -227.5
OJ
c
W
.S2 -228.0
Ql
c
S2
-228.5

Correlation Energy

o 50 100 150 200

Temperature (K)

Figure 5. BeS prediction of the kinetic energy and the correlation energy.

3. Relationship between intra-band spectral weight and kinetic

energy

A measure of the kinetic energy is provided by the following relation (8, 9,

10)
0 e2 8 2 c(k)
7r ti2V "t:(nk,(J";~
A

-0 dwReo-(w)dw = (9)
/

where the high frequency limit indicates that the integral should include
only the intra-valence band transitions, and the condensate peak at w = 0
if the material is a superconductor. The integral over negative and positive
frequencies (note that cr( w) = cr* ( -w)) avoids ambiguity about the way
the spectral weight in the condensate peak should be counted. If the band
structure is described by a nearest neighbor tight-binding model, Eq. 9
leads to the simple relation

(10)

Hence in the nearest neighbor tight-binding limit the partial f-sum provides
the kinetic energy contribution, which depends both on the number of
particles and the hopping parameter t (11, 12, 13, 14).
However, if the band-structure has both nearest neighbor hopping and next
nearest neighbor hopping, Eq. 10 is not an exact relation, and instead Eq. 9
should be compared directly to the experiments. In Fig. 6 we compare the
spectral weight, calculated directly using Eq. 9 to the result of Eq. 10, using
the same parameters as for Fig. 3. Note that the kinetic energy has to be
14

112.8

112.6

>
OJ 112.4
-S
a: 112.2

112.0

>
OJ
-S
N

10000 20000 30000 40000

2 2
T (K )

Figure 6. Beg prediction of the spectral weight function.

divided by a factor two, as we are interested in the projection along one

of the two axes in the ab-plane, which can be compared directly to the
experimental value of PL. From Fig. 6 we can conclude, that the effect of
including t f in the calculation is rather small, and it is still OK to identify
PL with the kinetic energy apart from a minus sign.

4. Experimental determination of the intra-band spectral weight

In two recent experimental papers measurements of PL in Bi2212 have been

reported (15, 16). The values of the kinetic energy change in the super-
conducting state were in quantitative agreement with each other, and both
papers arrived at the same conclusion: Contrary to the BCS prediction, the
kinetic energy of the superconducting state is lower than in the normal state
(taking into account a correction for the temperature trends of the normal
state). In Fig. 7 the data of Ref. (15) have been reproduced. Comparing
this with the BCS prediction clearly demonstrates the large qualitative
discrepancy between theory and experiment. Clearly the type of mechanism
assumed in BCS theory is not at work here.
 15

:;-
(l)
E
~

--'
Q.

142
:;-
(l)
E
--'
Q.

138

0 10000 20000 30000 40000

2 2
T (K )

Figure 7. Experimental values of the ab-plane spectral weight function, taken from
Ref. (15)

5. Implications of the experimental data

The trend seen in the experimental data has been predicted by Hirsch in
1992 (17, 18, 19). The model assumption made by Hirsch was, that the
hopping probability of a single hole between two sites becomes larger if one
of the two sites is already occupied by a hole. Although this model provides
good qualitative agreement with the optical experiments, it has one serious
deficiency: It also predicts s-wave symmetry for the order parameter, in
sharp contrast to a large body of experimental data which show that the
superconducting gap in the cuprates has d-wave symmetry.
In a recent set of calculations based on the Hubbard model, Jarrell et al (20)
obtained a similar effect as seen in our experiments, both for underdoped
and optimally doped samples. Crudely speaking the mechanism is believed
due to the frustrated motion of single carriers in a background with short-
range (RVB-type) spin-correlations, which is released once pairs are formed.

6. Conclusions

We have made a quantitative comparison between the BCS-prediction for

d-wave pairing in a band with nearest neighbor and next-nearest neighbor
16

hoppping and various experiments, in particular specific heat and meas-

urements of the optical ab-plane sumrule. We have shown that the BCS-
prediction produces the wrong sign for the kink at T c of the ab-plane
intraband spectral weight, while the model calculation is in good agreement
with the experimental specific heat data.

Acknowledgements

DvdM gratefully acknowledges M. Norman, N. Bontemps, J. Hirsch, and

M. Jarrell for stimulating discussions, and J. W. Loram for making his data
files of the specific heat of Bi2212 available.

References

1. J. W. Loram, J. Luo, J. R. Cooper, W. Y. Liang and J. L. Tallon, J. Physics and

Chemistry of Solids 62, 59-64 (2001).
2. P.W. Anderson, Science 235, 1196 (1987).
3. D. van der Marel, A.J. Leggett, J.W. Loram, J.R. Kirtley Physical Review B Rapid
Communications 66, 140501R (2002).
4. There is an additional term, due to the fact that the chemical potential of
the superconducting state is slightly shifted relative to the normal state (5, 6):
0.25 ~p IVp+kVpI2. In the numerical examples in this paper this additional term has
been taken into account. Since the effect is small, we will not discuss it further here.
5. D. van der Marel, Physica C 165 (1990) 35-43;
6. D. van der Marel, and G. Rietveld, Phys. Rev. Lett. 69 (1992) 2575-2577;
7. G. Rietveld, N. Y. Chen, D. van der Marel, Phys. Rev. Lett. 69 (1992) 2578-2581;
8. M. R. Norman, and C. Pepin, Phys. Rev. B 66, 100506(R) (2002).
9. R. Kubo, J. Phys. Soc. Japan 12, 570 (1957).
10. H. D. Drew, P. Coleman, Phys. Rev. Lett. 78, 1572 (1997).
11. Pierre F. Maldague, Phys. Rev. 16,2437 (1977).
12. J. Carmelo and D. Baeriswyl, Phys. Rev. B 37, 7541 (1988).
13. D. N. Basov, S. I. Woods, A. S. Katz, E. J. Singley, R. C. Dynes, M. Xu, D. G.
Hinks, C. C. Homes, M. Strongin, Science 283, 49 (1999).
14. S. Chakravarty, H. Kee, E. Abrahams, Phys. Rev. Lett. 82 , 2366 (1999).
15. H. J. A. Molegraaf, C. Presura, D. van der Marel, P. H. Kes , M. Li, Science 295,
2239 (2002).
16. A.F. Santander-Syro, R.P.S.M. Lobo, N. Bontemps, Z. Konstantinovic, Z.Z. Li, H.
Raffy, cond-mat/0111539 (2002).
17. J. E. Hirsch, Physica C 199, 305 (1992).
18. J. E. Hirsch, Physica C 201 (1992) 347.
19. J. E. Hirsch, and F. Marsiglio, Physica C 331 (1999), 150-156.
20. Th. A. Maier, M. Jarrell, A. Macridin, C. Slezak, cond-mat/0211298 (2002).
JOSEPHSON EFFECT IN THE CUPRATES:
MICROSCOPIC IMPLICATIONS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

R. HLUBINA
Department of Solid State Physics, Comenius University,
Mlynska dolina F2, SK-842 48 Bratislava, Slovakia

Abstract. In the tunnel limit, the current-phase relation of Josephson junctions can be
expanded as I( ¢) = h sin ¢ + h sin 2¢. Standard BCS theory predicts that hRN ~ 61 e
and hih ~ D, where R N is the resistance of the junction in the normal state, 6 is
the superconducting gap, and D « 1 is the junction transparency. In the cuprates, the
experimental value of hR N (hih) is much smaller (larger) than the BCS prediction. We
argue that both peculiarities of the cuprates can be explained by postulating quantum
fluctuations of the pairing symmetry.

Key words: High Temperature Superconductivity, Josephson Effect, Josephson Product,

Current-Phase Relation, Quantum Fluctuations, D-Wave Pairing, S-Wave Pairing, RVB

1. Introduction

The proximity to the metal-insulator transition is well known to lead to an

anomalous normal state of the high T c superconductors. It is therefore inter-
esting to ask whether, apart from the d-wave symmetry of pairing, also the
superconducting state of the cuprates can be regarded as unconventional.
Most studies attempt to answer this question by considering the properties
of quasiparticle excitations. For instance, photoemission experiments seem
to support the conventional alternative, since sharp spectral functions have
been observed at low temperatures (1). On the other hand, the effect of
strong correlations on the condensate has remained largely unexplored so
far. This is surprising, since large quantum phase fluctuations are to be
expected in the cuprates as a combined effect of the suppression of charge
fluctuations and of the uncertainty principle. In particular, such fluctuations
have been suggested to stabilize the RVB state and the presumably related
pseudogap phase of the cuprates (2). In this paper we argue that by a

17
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 17–25.
© 2003 Kluwer Academic Publishers
18

detailed analysis of the Josephson effect, new insights into the nature of
quantum fluctuations in the cuprates can be obtained.
Josephson junctions involving the cuprate superconductors have been
studied mainly because they enable phase-sensitive tests of pairing sym-
metry (3). Attention has been paid especially to two particular types of
Josephson junctions: grain boundary (4) and intrinsic Josephson junctions
(5). The best studied type of grain boundary Josephson junctions involves
junctions built on c-axis oriented films, where the weak link forms at the
boundary of grains which are rotated around the [001] axis with respect to
each other. Idealized junctions of this type are characterized by a planar
interface and two angles e1 and e2 between the interface normal and the
crystallographic directions in the grains forming the junction. The proper-
ties of the junctions depend dominantly on the misorientation angle e =
e 2 -e 1 . It is well known (4) that the transparency D of grain bound-
ary junctions decays exponentially with increasing misorientation angle e,
D(e) ex: exp(-e/eo). Thus, for e > eo;::::::; 5°, grain boundary junctions are in
the tunnel limit and (for e not too close to 45°, see Section 3) their current-
phase relation I (¢) = h sin ¢ + h sin 2¢ + ... can be well approximated by
I( ¢) = h sin ¢, neglecting the higher-order harmonics.

2. The Josephson product

A useful quantity characterizing the superconducting electrodes forming

the Josephson junction is the product of the first harmonic h with the
junction resistance in the normal state, R N . According to standard theory
(for homogeneous featureless barriers), this so-called Josephson product
is independent of the junction area and of the barrier transparency, thus
giving an intrinsic information about the superconducting banks. It is
well known that at interfaces between d-wave superconductors, anomalous
bound Andreev levels may form (6). At temperatures larger than the energy
of such Andreev levels, BCS-like theory for rough interfaces between d-wave
superconductors (7) predicts

(1)

where ,6. is the maximal superconducting gap. The measured Josephson

product of cuprate grain boundary junctions (4) can be well described by
hR N = 0 2 (hR N )BCS with a e-independent renormalization factor 0: 2 rv
10- 1 . In addition to the e > eo data of (4), this functional form can be
tested also for = a (which case can be realized in break junctions) with
e
the result that average Josephson products of such junctions (8) are fully
consistent with the grain boundary data. Moreover, in (8) it has been shown
 19

that ~ is not depressed in the junction region, thus explicitly demonstrating

the breakdown of the BCS prediction for hRN in the cuprates.
There exists no generally accepted explanation of the small renormaliz-
ation factor 0: 2. One of the reasons is that the microstructure of Josephson
junctions is typically quite complicated. In fact, it is well known that small
angle grain boundaries can be modelled by a sequence of edge dislocations,
while at larger misorientation angles the dislocation cores start to overlap
and no universal picture applies to the structure of the grain boundary.
For large-angle grain boundaries, Halbritter has proposed (9) that the
junction can be thought of as a nearly impenetrable barrier with randomly
placed highly conductive channels across it. If due to strong Coulomb repul-
sion only the normal current (and no supercurrent) is supported by these
channels, the small value of hRN follows quite naturally.
In this paper we shall argue that the smallness of the Josephson product
does not follow from the particular properties of the barrier, but is rather
an intrinsic property of the cuprates. Such a point of view has been first
advocated in (10). However, that paper did not consider alternative more
conventional explanations. In order to support our point of view, let us
begin by discussing the Josephson product for intrinsic Josephson junctions
in the c-axis direction. Such junctions can be viewed as an analogue of ab-
plane break junctions (since the misorientation angle vanishes for both),
but are preferable because of simpler geometry of the interface. Moreover,
zero energy surface bound states which may develop at ab-plane surfaces
because of the d-wave symmetry of the pairing state (6) do not form in the
c-axis direction, simplifying the analysis of intrinsic Josephson junctions.
Standard BCS theory applied to the case of tunneling between two-
dimensional superconductors (11) predicts (for coherent c-axis tunneling)
that the c-axis critical Josephson current density is]l = (2e/n)N(0)(Zkt~),
where (... ) denotes an average along the (two dimensional) Fermi line, Zk
is the wavefunction renormalization, and tk is the matrix element for c-axis
tunneling between neighboring CU02 planes. N(O) is the bare (unrenor-
malized) density of states, N(O) = .f dk( 47["2 nVk)-l, where the integration
runs along the two dimensional Fermi line and Vk is the bare Fermi velocity
at the Fermi surface point k. Note that although we have assumed that
the self-energy is only frequency dependent, for this geometry Zk enters the
expression for ]1 (and also the normal-state resistance, see below).
Let us also note that the use of ordinary perturbation theory in deriving
an expression for ]c has been criticized recently (12). However, our formula
yields the correct answer for tk < ~. This can be shown either by an
explicit solution of a 4 x 4 Bogoliubov problem for two coupled planes with
phase differences 0 and 7[" between the planes, or by considering solutions
to the gap equation in an infinite layered system with a finite c-axis total
20

momentum of the Cooper pairs (for a similar calculation, see e.g. (13)).
The conductance per square in the normal state is given by G N
(2e 2 /n)N(0)(Zktkrk1)N, where rk is the inverse lifetime ofthe quasiparticles
and the index N in the Fermi line average means that the quantities are
to be evaluated in the normal state. Therefore standard theory predicts for
the Josephson product of intrinsic Josephson junctions hRN = j1Gil ;:::::
e-l(zktk)/(Zktkrkl)N. In conventional superconductors RN can be meas-
ured at low temperatures in a sufficiently large magnetic field. This is
impossible for the cuprates and thus RN is usually defined as the c-axis
resistivity at T e .
Unfortunately, due to the unknown temperature dependence of Zk, the
theoretical hRN can not be directly tested by experiment. In order to
overcome this problem, in (14) instead of the usual Josephson product
a related characteristic of intrinsic Josephson junctions has been stud-
ied, namely the product of the critical current h and of the resistivity
Rs in the resistive mode of the junction (at low temperatures). Since
the conductance per square in the resistive mode of the superconductor
is (14) Gs ;::::: (8e 2 /h)N(O)(Zktk)node/~, standard BCS-like theory pre-
1
dicts hRs = j1G S ;::::: (7r/2)(~/e)(zktk)/(Zktk)node. In (14), hRs '" ~/e
has been found experimentally and good agreement with theory has been
claimed, since momentum-independent tk and Zk were assumed. However,
according to band structure calculations (15), tk is strongly suppressed in
the nodal directions. If this modulation of the tunnel matrix element tk is
taken into account and the presumably only moderate k-space dependence
of Zk is neglected, the experimental hRs is seen to be drastically reduced
with respect to the theoretical predictions.
Thus we have shown that although the barriers in grain boundaries and
in intrinsic Josephson junctions are of very different nature, both types of
junctions exhibit a suppressed Josephson product. Therefore we believe that
this suppression is not due to specific barrier properties as suggested in (9),
but rather due to some intrinsic property of the high-Te superconductors.

3. The second harmonic of the current-phase relation

Since the second harmonic h is not forced by symmetry to depend on the

angles ai, its Josephson product can be estimated (for temperatures larger
than the energy of anomalous Andreev levels) using the standard BCS the-
ory as hRN '" D~/e. Comparison with Eq. (1) implies that for junctions
with 45° the d-wave symmetry of pairing leads to a suppression of h,
a;: : :
and h may become comparable to h. This has in fact been observed in
two different types of 45° grain boundary Josephson junctions (16, 17).
 21

However, the results of (16, 17) are quite mysterious, if we take into
account the actual experimental setup. In fact, standard BCS theory with
ideal featureless barriers implies that in order that hi rv hi, the average
1 1

misorientation angle e would have to be given with a precision rv D, where

D rv 10- 3 . This is not realistic and therefore two alternative explanations
have been proposed, in both of which the origin of the anomalously large
Ihlhi ratio has been sought in the barrier properties.
(i) Faceted scenario has been considered as an alternative explanation
for symmetric 45° junctions (i.e. junctions with nominal geometry e1 = 0°
and e2 = 45°), in which Ih 1 > Ih 1has been found (16). It takes into
account the faceting of the grain boundary and also the twinned nature of
the (orthorhombic) YBCO thin films. Due to both of these features, the
junction can be viewed as a parallel set of 0 and 7'1 junctions (18). It has
been shown (19) that in such a case spontaneous currents are generated
along the interface, the ground state energy of the junction is minimized at
a macroscopic phase difference ±7r 12, and consequently the current-phase
relation is dominated by the second harmonic h.
In what follows we analyze quantitatively whether the faceted scen-
ario can apply to the results of (16). Let us denote the current densities
corresponding to the harmonics Ii (with i = 1,2) as ji and introduce
the Josephson penetration depth of the junction, A J = (iI>01 47'1 A/LOj2)1/2.
Moreover, let the local critical current density in the 0 and 7'1 junctions be
±jo, their typical length a, and the bulk penetration depth be A. Then,
since a ~ 0.01 - O.lp,m, A ~ 0.15p,m, and A J rv 3/Lm (estimated making
use of (4) j2 rv 10 4 A/cm2), the inequalities 7rA» a and A)>> aA are well
satisfied. Following (19) it is easy to show that these inequalities guarantee
that the spontaneously generated currents along the Josephson junction can
be calculated within perturbation theory. If we denote the total junction
length by L, then a straightforward calculation yields j21jo ~ /LojoaA 2/iI>0
and j1/jo ~ (aiL )1/2. The equation for j1 is a random walk-type formula,
indicating that j1 averages to zero in a sufficiently long junction. After
some algebra the above equations are seen to imply j2!J1 ~ JLAI(47rA)).
Therefore, standard theory predicts that h > h can be realized only in
sufficiently long junctions with L > 47'1 A) I A. This requires L > 500/Lm,
whereas in (16) much shorter junctions with L rv l/Lm were studied.
(ii) Pinhole scenario has been proposed in (17) as an explanation for
symmetric 45° junctions (i.e. junctions nominally characterized by e2 =
-e1 = 22.5°). It views the barrier as basically impenetrable, the conduction
being due to randomly placed highly conductive pinholes. This explains
quite naturally the small value of the effective barrier transmission and, at
the same time, the large value of Ihlh I. Note that in order to explain the
small value of the Josephson product, in addition to pinholes also Halbrit-
22

ter's conductive channels (9) have to be postulated, which are assumed to

be highly conductive only in the normal and not in the superconducting
channel. Because of this ad hoc nature of the pinhole scenario, and mainly
because of the absence of higher harmonics in I(¢) at 4 K (17) whose
presence it predicts, we believe that the pinhole picture should be discarded.
Thus we conclude that the large value of the second harmonic h (com-
pared with predictions of the standard BCS theory) is most probably not
an extrinsic (barrier-related) effect, but rather an intrinsic property of the
cuprates.

4. Microscopic implications

The two apparently unrelated experimental facts, namely the suppressed

Josephson product hRN and the enhanced ratio Ih/hl, can be explained
by a single assumption that in the cuprates some mechanism is operative
which leads to a suppression of h, while leaving RN and h intact. In
what follows we describe one such mechanism which we believe to be
the most promising one. Namely, we suggest that at low temperatures
the superconducting state of the cuprates supports fluctuations of pairing
symmetry towards s-wave pairing (which pairing is expected to be locally
stable within several microscopic models of the cuprates). Such fluctuations
presumably do not affect RN, while they do influence the Josephson current.
In simplest terms, if we denote the phases of the superconducting grains
forming the junction as ¢i, then the fluctuations renormalize the first and
second harmonics by the factors (e icP1 ) (e i ¢2) and (e 2icP1 ) (e 2icP2 ), respectively,
where (... ) denotes a ground-state expectation value. Thus experiment
requires that the fluctuations have to be of such type that l(eicP)1 = a ;:::j 0.3
and I(e 2icP ) I ;:::j 1. Precisely this behavior is expected if the d-wave order
parameter fluctuates towards s-wave pairing.
Now we proceed by introducing a minimal model of such fluctuations.
Unlike in standard literature on this subject (see, e.g., (20) and references
therein), we assign a phase field <Pi to each bond i of the square Cu lattice.
In other words, <Pi lives on the sites of the dual lattice. Since large phase
fluctuations are expected, the compactness of phase fluctuations is explicitly
taken into account and the model reads

H = H fluct + L [-V COS(2<pi - 2<pj) + W COS(<pi - <Pj)] , (2)

(i,j)

where H fluct = -m- 1 Li 8 2 /8<Pr. The second term in Eq. (2) (with (i,j)
denoting a pair of nearest neighbor sites) describes for 0 < W < V a
superconductor with a dominant d-wave pairing V + Wand subdominant
s-wave pairing V - W. Since we concentrate on the q ;:::j (1T,1T) fluctuations
 23

Figure 1. Ground state of the Heisenberg model on an elementary plaquette.

from d to s-wave pamng, as a first approximation we do not take into

account the coupling of phase fluctuations to electromagnetism.
In order to gain insight into the microscopies of Hfluct, let us recall that
the RVB fluctuations in an elementary plaquette (in a hole-free region) favor
a minus sign between the two different valence bond configurations (see
Fig. 1). If these configurations are thought of in terms of Bose condensates
of valence bonds, this means that energy is gained for a relative phase of
the x and y condensates arg( A ) = ±'if /2. RVB processes are thus seen
to frustrate the phase ordering dictated by the second term in Eq. (2).
Since the other singlet state of the elementary plaquette (with a plus sign
between the valence bond configurations) lies higher in energy, the effective
Hamiltonian in the singlet sector is HRVB ex .6.!.6.!+x+y.6.i+ x .6.i+Y + H.C.,
where .6.! creates a singlet on bond i. Neglecting the amplitude fluctuations
of .6. i , we can write H RVB = J Li cos (cpi - CPi+x + CPi+x+y - CPi+y). In what
follows we replace H fluct by H RVB in the quantum model Eq. (2), thereby
obtaining a completely classical toy model.
The toy model permits a simple mean field analysis: We consider two
types of solution, both of which live on two sublattices A and B, and
we explicitly disregard four-sublattice solutions, since they correspond to
translation-symmetry breaking states on the original Cu lattice. In the first
type of solution CPA = cP and cP B = 0 and there are two possibilities: either
cP = 'if (d-wave), or cP < 'if (a complex mixture of d and s, to be called d+is).
In the second type of solution we assume a disordered state of such type
that on sublattice A, the phase equals 0 and 'if with probabilities (1 + O!) /2
and (1- O!) /2, respectively, and on sublattice B the values 0 and 'if are inter-
changed. In this case the macroscopic symmetry is of the d-wave type, with
renormalized averages (ei'PA) = _(ei'PR) = O! and (e 2i 'PA) = ( e 2i'PR) = 1.
Minimization of energy (with respect to cP or O!) leads to the phase diagram
shown in Fig. 2. The renormalization factor in the disordered d-wave state
(which presumably corresponds to a homogeneous but strongly fluctuating
d-wave phase in the model Eq. (2)) is O! = (W/J)lj 2. Thus the cuprates
correspond to the region W ~ J /10 and J < 4V in Fig. 2.
24

5
d+is
4

> 3 disordered
--
"""')

2 d

OIL-----'-----------'---------'----'---------'----------'
o 2 3 4 5 6
WN
Figure 2. Mean field phase diagram of the toy model.

5. Conclusions

The anomalous effects observed in cuprate grain boundary and intrinsic

Josephson junctions can be explained by a single assumption of strong
quantum phase fluctuations at q = (7f, 7f), which leads to I(e i 4» I = 0: :::::; 0.3
and l(e 2i 4»1 :::::; 1. In addition, our picture implies that the Josephson product
for junctions between the cuprates and low-Tc superconductors is renormal-
ized by the factor 0:, in semiquantitative agreement with experiment (21). It
also may be relevant for the experiment (22), where a large second harmonic
has been found in a c-axis Josephson junction between YBCG and Nb.

Acknowledgements

I thank M. Grajcar for numerous stimulating discussions about the Joseph-

son effect. I also thank T. V. Ramakrishnan for insightful remarks on the d-s
phase fluctuations. This work was supported by the Slovak Scientific Grant
Agency under Grant No. VEGA-1/9177/02 and by the Slovak Science and
Technology Assistance Agency under Grant No. APVT-51-021602.

References

1. A. Damascelli, Z. X. Shen, and Z. Hussain, cond-mat/0208504.

2. P. W. Anderson, The Theory of Superconductivity in the High-Tc Cupmtes (Prin-
ceton Univ. Press, Princeton, 1997).
3. M. Sigrist and T. M. Rice, J. Phys. Soc. Jpn. 61, 4283 (1992).
4. H. Hilgenkamp and J. Mannhart, Rev. Mod. Phys. 74,485 (2002).
5. R. Kleiner and P. Muller, Phys. Rev. B 49, 1327 (1994).
 25

6. C. R. Hu, Phys. Rev. Lett. 72, 1526 (1994).

7. M. Grajcar, R. Hlubina, E. Il'ichev and H.-G. Meyer, Physica C 368, 267 (2002).
8. N. Miyakawa et al., Phys. Rev. Lett. 83, 1018 (1999).
9. J. Halbritter, Phys. Rev. B 46, 14 861 (1992).
10. G. Deutscher, Nature 397, 410 (1999).
11. L. N. Bulaevskii, Sov. Phys. JETP 37, 1133 (1973).
12. S. Chakravarty, Eur. J. Phys. B 5, 337 (1998).
1:3. R. Hlubina, Acta Physica Slovaca 45, 611 (1995).
14. Yu. 1. Latyshev et al., Phys. Rev. Lett. 82, 5345 (1999).
15. O. Andersen, O. Jepsen, A. Liechtenstein, 1. Mazin, Phys. Rev. B 50, 4145 (1994).
16. E. Il'ichev et al., Phys. Rev. B 60, :3096 (1999).
17. E. Il'ichev et al., Phys. Rev. Lett. 86, 5369 (2001).
18. J. Mannhart et al., Phys. Rev. Lett. 77, 2782 (1996).
19. A. J. Millis, Phys. Rev. B 49, 15 408 (1994).
20. A. Paramekanti, M. Randeria, T. V. Ramakrishnan, S. S. MandaI, Phys. Rev. B
62, 6786 (2000).
21. A. G. Sun et al., Phys. Rev. B 54, 67:34 (1996).
22. P. V. Komissinski et al., Europhys. Lett. 57, 585 (2002).
RUTHENATES: UNCONVENTIONAL SUPERCONDUCTIVITY
AND MAGNETIC PROPERTIES

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

MANFRED SIGRIST
Theoretische Physik, ETH-Honggerberg, 8093 Zurich, Switzer-
land

Abstract. A review is given of the unconventional superconductivity of Sr 2Ru04 and

the magnetic properties of Ca2-,rSr x Ru04. The electronic properties are determined by
the three 4d-t2g-electrons of Ru. It is shown that superconductivity is due to the dxy -
electrons, while the magnetic behavior is more likely a feature of the other two orbitals,
dyz and dn .

Key words: Ruthenate, Unconventional Superconductivity, Magnetism, Spin-Orbit Sys-

tems, P-Wave Superconductivity, Spin-Orbit Coupling, Multi-Band Effects

1. Introduction

For the last two decades transition metal oxides have been among the
leading topics in condensed matter physics. Best known are the cuprates
which are high-temperature superconductors and the manganese oxides dis-
playing the colossal magnetoresistence phenomena. Among these classes of
materials also ruthenates play an important role as systems with intriguing
properties. The interest in this material mainly arose from the discovery of
unconventional superconductivity in Sr2Ru04 with p-wave pairing (1, 2).
Unlike many other stoichiometric compounds Sr2Ru04 is not a Mott insu-
lator, but is a very good metal with clear Fermi liquid properties. This Fermi
liquid shows features of strong correlation and is quasi-two-dimensional due
to the layered crystal structure. It was suggested that this system could be
considered as the electronic analog of 3He with the natural implication that
the superconducting state originates from spin triplet p-wave pairing (3, 4).
The electronic properties are dominated by the nearly degenerate 4d-t2g-
orbitals of the Ru4+-ion, which introduces four electrons per Ru. These or-

27
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 27–34.
© 2003 Kluwer Academic Publishers
28

bital disperse via 1f-hybridization with the intermediate oxygen 2p-orbitals

leading to three bands with two electron-like and one-hole-like Fermi sur-
faces. The form of the orbitals suggests to distinguish two electronic sub-
systems, the electron-like I-band corresponding to the dxy-orbital and the
weakly hybridized quasi-one-dimensional a (hole-like) and (3 (electron-like)
band originate from the degenerate dyz/dzx-orbitals. There is growing evid-
ence that the superconductivity has its origin in the I-band. On the other
hand, magnetic properties are dominated by the a-(3-bands which are re-
sponsible for the strongest feature in the spin fluctuations, an incommen-
surate spin correlation due to strong nesting properties - reminiscent of
the basically one-dimensional nature of these bands. We would like to
briefly review some aspects of both, the superconducting and the magnetic
properties providing evidence for the very distinct character of the different
orbitals.

2. Chiral p-wave superconductivity

There are two key experiments determining the symmetry of the pairing
state. The measurement of the spin susceptibility via NMR-Knight shift
addresses the spin part of pair wave function, and indicates that there is
equal-spin-pairing with parallel spins in the Ru02-planes (5). Furthermore,
the zero-field muon spin relaxation shows the appearance of spontaneous
intrinsic magnetism below the superconducting phase transition (6). This
magnetism originates most likely from the net angular momentum of the
Cooper pairs, determining the orbital part of the pair wave function. For a
quasi-two-dimensional system the orbital angular momentum is naturally
oriented perpendicular to the basal plane. The resulting pairing state is
entirely consistent with the symmetry classification including spin-orbit
coupling, and can be represented in the vector notation as

d(k) = z(kx ± iky ), (1)

the analog to the A-phase of 3He (3, 7). Note that the orientation of the
d-vector is perpendicular to the equal-spin orientation. This state possesses
chirality corresponding to the angular momentum, and is also called chiral
p-wave state. This phase has two order parameter components which are
degenerate. Competing states based on the same symmetry classification
are non-chiral, with the form d = xkx + yk y , ykx - xk y , xkx + yk y and
ykx+xk y , whose equal spin-orientation is parallel to the z-axis, inconsistent
with the NMR data.
Interestingly, all these states possess the same condensation energy in
a simple weak-coupling theory, which is entirely determined by the qua-
siparticle gap magnitude Id(k) I (7). Hence, we need a mechanism to lift
 29

this degeneracy. One class of mechanisms is based on feedback effects.

Unconventional pairing implies that the attractive interaction is likely to
be of purely electronic origin. Consequently, the properties of the pair-
ing interaction would be modified, once the system undergoes the phase
transition to the superconducting state. Some pairing states may benefit
from this change others do not. Various of these feedback effects have been
investigated and exclusively favor the state d = z(kx ± iky ) (11). Feedback
is a higher order correction and is here, unlike in 3He, not so important.
Actually spin-orbit coupling plays the deciding role directly affecting the
transition temperature of the different states.
Multi-band effects: Sr2Ru04 represents a special opportunity to study
the effects of spin-orbit coupling on a spin-triplet superconducting phase.
The three 4d-t2g-bands are coupled by spin-orbit coupling, making it ne-
cessary to treat the complete multi-band problem:

H s .o. = i.\ LL Ep,vp L cLs(J~sICiVsl (2)

i f-L,V,P 8,S'

with cLs creating an electron on site i in the orbital f-l with spin s (Ep,vp is
the completely antisymmetric tensor). It is possible to fit the band structure
with a simple tight-binding model. Using a simple phenomenological model
for the pairing interaction Ng and Sigrist concluded that the i-band origin-
ating from 4dxy would have to be dominant for superconductivity in order to
stabilize the chiral p-wave phase (9, 10). This result was recently confirmed
by Yanase and Ogata (17) using a microscopic pairing mechanism based
on a perturbative approach of the extended three-band Hubbard model as
introduced by Nomura and Yamada (19, 18). The latter group had found
that within their theory indeed the i-band would play the leading role for
spin-triplet superconductivity.
The assumption that the pairing symmetry is related with the so-called
orbital-dependent superconductivity (ODS) and pairing symmetry via spin-
orbit coupling is very consistent with several experimental and theoretical
facts. Spin-triplet superconductivity is believed to be supported by fer-
romagnetic spin fluctuations. The analysis of NMR-experiments suggests
that the i-band is closer to ferromagnetism than the a-p-bands which
show strong incommensurate spin correlations due to Fermi surface nesting
features (12, 13).
ODS plays also an important role for the thermodynamics, since mul-
tiple gaps introduce significant modifications in the temperature depend-
ence (8). Power-law behavior of various quantities over a wide temperat-
ure range, which have been attributed to line nodes, can be well under-
stood also within a multi-band picture as shown by Zhitomirsky and Rice
(14), Nomura and Yamada (19) for the specific heat. Kusunose and Sigrist
30

showed that the analysis of the behavior of the London penetration depth
requires one furthermore to take non-local effects into account (15, 16).
Nevertheless, the fitting of the thermodynamic quantities does not lead to
the unique determination of gap shapes and it may remain even difficult to
settle the question whether the gaps in some bands have nodes.
3-Kelvin phase: A surprising feature has been the onset of inhomogen-
eous superconductivity in some samples of Sr2Ru04 at higher temperatures
than the bulk-Teo In accordance with the approximate onset temperature
this phase is called the "3 K-phase" (20). The analysis of the material
revealed the presence of excess-Ru in form of micrometer-sized Ru-metal
inclusion embedded into the pure Sr2Ru04-metal. Since Ru on its own
becomes superconducting below 0.5K only, it has been suggested that the
actual nucleation region of the 3 K phase is the interface between the Ru-
inclusions and Sr2Ru04. It is possible that the transition temperature of
the p-wave superconductivity is slightly enhanced at the interface due to
crystal lattice deformations. Based on this assumption a Ginzburg-Landau
theory of the nucleated phase has been developed (21). Without going into
the calculational details we would like to review here some key results of
this analysis and compare with experimental results as far as available.
1. The interface corresponds to a region of reduced symmetry, so that
the two p-wave components with momentum direction parallel and perpen-
dicular to the interface (kll and kj..) have different onset temperatures. It
is found that p-wave component parallel to the interface is more stable,
because its coherence length along the normal direction of the interface
is shorter, so that the order parameter is better confined to the beneficial
region. Thus the 3 K phase is time reversal symmetry conserving and before
the onset of the bulk phase a second phase transition has to occur breaking
time reversal symmetry.
2. The superconducting interface state corresponds to a gap function
with phase 0 and 7r depending on momentum direction. This is a condi-
tion to generate zero-energy Andreev bound states in the Ru-inclusions
analogous to the zero-energy surface states in d-wave superconductors.
These states are observable in quasiparticle tunneling experiments. Re-
cent data by Mao et al in break junctions, where tunneling is likely to
be dominated by inter-Ru-inclusion tunneling, show the presence of zero-
bias voltage anomalies in the 3 K phase, consistent with the nucleation of
the a single-component p-wave state at the interface (22).
3. The critical magnetic field for the 3 K phase is highest, if the magnetic
field lies parallel to the interface, since in this case orbital depairing is
reduced. Interestingly under this condition the temperature dependence of
H e2 is different from the standard linear bulk behavior. For an infinitely
planar interface there is a square-root behavior, H e2 ex: VT - T* with T*
 31

as the zero-field nucleation temperature. If the field is in addition parallel

to the z-axis, it couples to the angular momentum of the Cooper pair
(kll,k~ -----+ k ll + ik~). As a consequence H c2 should increase and display
an up-turn at lower temperatures. The sublinear temperature dependence
H c2 ex (T - T*)"'" with a ~ 0.6 - 0.8 have been observed as well as the
expected up-turn for the z-axis oriented field (23).
The existence of the 3 K phase together with these experiments provides
strong evidence for the scenario based on interface superconductivity and
the realization of the chiral p-wave phase in the bulk.

3. Magnetic properties of Ca2-xSrxRu04

Ca2Ru04: Replacing Sr by the isovalent Ca in Sr2Ru04 leads to drastic

changes. At temperatures below 350 K, the compound is a Mott insulator
and shows antiferromagnetic order below 110 K (24). The origin for this
contrasting behavior between the Ca- and Sr-compound lies in the fact that
the small Ca-ion induces a lattice deformation, including the rotation and
tilting of the Ru-O-octahedra as well as their flattening. The latter is essen-
tial, since it gives a decisive change of the 4d-t2g-orbitallevels arrangement.
For Sr2Ru04 the elongation of the octahedra along the z-axis put the d xy-
orbital above the other two, while the flattening in Ca2Ru04 leads to the
opposite situation. The lower dxy-orbital is then occupied by two electrons
filling the "i-band" completely and the remaining two electrons are loc-
alized on the two other orbitals forming an S = 1 spin degree of freedom
due to Hund's rule coupling, which order antiferromagnetically because of
superexchange (24). Obviously, these magnetic properties originate from
the electrons of the a-p-bands.
Intermediate doping: It is interesting to analyze the problem of the
continuously doped system Ca2-xSr2Ru04. Nakatsuji and Maeno found
experimentally that we can distinguish for 0 < x < 2 three basic regions
(24). Region I (0 < x < 0.2) corresponds to the antiferromagnetic Mott
insulator discussed above. Region II (0.2 < x < 0.5) is a metallic phase with
antiferromagnetic correlations at low temperature. Region III (0.5 < x < 2)
shows an increasing tendency towards ferromagnetism as x approaches 0.5.
There are two basic approaches to the behavior in region III. Fang and
Terakura (25), and Nomura and Yamada (26) consider the effect of Ca-
doping into Sr2Ru04 as a change of the band structure in a way that, in
particular, the i-band is narrowed and its Fermi level is pushed closer to a
Van Hove singularity. This gives rise to the increase of the spin susceptibil-
ity. All bands remain metallic. In contrast, Anisimov et al. investigated the
possibility that a redistribution of charge happens transfering 3 electrons
into the dyz / dzx-orbital which undergo Mott localization, while the single
32

electron in the i-band remains itinerant. This gives rise to a localized spin
S = 1/2 degree of freedom. This would explain the fact that for x close
to 0.5 the spin susceptibility indicates the presence of such a localized spin
per site which is difficult to explain within the picture of entirely itinerant
electrons (25, 26). On the other hand, it is not clear whether this orbital-
selective Mott insulator would give rise to the experimentally observed
degree of metallicity.
The scenario of the orbital-selective Mott insulator introduces with the
three electrons in the d yz / dzx-orbitals beside the spin also an orbital degree
of freedom (isospin 1/2). Thus the effective model of the localized degrees
of freedom corresponds to a Kugel-Khomskii type of model for coupled
spin 1/2 and isospin 1/2 (27). It has been shown that this effective model
in region III would give rise to staggered orbital order, i.e. having a single
hole in the d yz (dzx)-orbital on the sublattice A (B) of the Ru square lattice.
This in turn leads to a ferromagnetic spin interaction via the Goodenough-
Kanamori rule.
Further doping leads into region II where the crystal changes symmetry
from tetragonal to orthorhombic. This Jahn-Teller distortion introduces a
bias on the localized orbital degrees offreedom of the d yz / dzx-orbitals, sup-
pressing the staggered correlation in favor of an alignment of the orbitals.
Eventually the spin-spin interaction turns antiferromagnetic with growing
lattice deformation in the effective model. This behavior is in qualitative
agreement with experiments. The mean field phase diagram of the effective
spin-orbital model including the lattice deformation is shown in Fig.I. For
small deformation the staggered orbital order is prevailing leading to a
ferromagnetic phase at low enough temperature. There is, however, a first
order phase transition to the competing antiferromagnetic order, which
at sufficiently strong deformation truncates abruptly the staggered orbital
(antiferro-orbital) order.
The role of Ca-Sr alloying is to introduce lattice deformations through
octahedra rotation and tilt. In the alloy there is naturally disorder and the
properties show a certain degree of spatial variation. This means that in the
region 0.2 < x < 0.5 the two phases separated by the first order transition,
FM-AFO (ferromagnetic-antiferro-orbital) and AFM-FO (antiferromagnetic-
ferro-orbital), coexist in domains with rather sharp boundaries. This as-
pect allows us to connect the metamagnetic transition and the anomalous
behavior of the magnetoresistance, observed for Ca1.8SrO.2Ru04 (28). At
temperature of about 2 K a smooth metamagnetic transition has been
observed at about B ex = 2.7T and the longitudinal magnetoresistance
passes a pronounced maximum at approximately the same field value. In
a finite magnetic field the domain structure of the material is modified by
increasing the FM-AFO domains. The metamagnetic transition is associ-
 33

antiferro-orbital "
I
I
I
I
I antiferromagnetic
I
ferromagnetic

orthorhombic strain
Figure 1. Schematic phase diagram for temperature versus orthorhombic strain for
the Kugel-Khomskii type model with spin-orbital degrees of freedom: The solid (dashed)
lines represent second (first) order phase transitions (29).

ated with the percolation of the domain boundaries. Domain boundaries

influence also the transport properties of the remaining metallic band (,) as
strong scattering regions. In this way naturally the maximal impediment for
conductance occurs when the domain boundaries percolate. In this way the
metamagnetic transition and the maximal magnetoresistance are connected
(29).
Although the scenario of the orbital-selective Matt insulator with local-
ized spin and orbital degrees of freedom offers an explanation for a variety of
phenomena in region II, it is difficult at the present stage to conclude on its
validity. Further experimental tests are necessary, in particular, concerning
the distinction of localized and itinerant portions of the t2g-electrons.

4. The distinct roles of the different orbitals

The surprising contrasts of electronic behavior observed in the alloy

Ca2-xSrxRu04 are connected with the fact that the character of the 4d-
t2g-orbitals is different. There is good evidence that the chiral p-wave
superconductivity has its origin in the electrons of the ,-band, which there-
fore determines the low-temperature behavior of Sr2Ru04. On the other
hand, the tendency towards antiferromagnetism in the alloy and Ca2Ru04
is clearly dominated by the o:-,6-bands which give rise to strong finite Q
spin fluctuations.

Acknowledgements

I would like to thank to D.F. Agterberg, V. Anisimov, A. Furusaki, J.

Goryo, H. Kusunose, M. Matsumoto, H. Monien, LA. Nekrasov, D.E. Kondakov,
34

KK Ng, M. Troyer and, especially, T.M. Rice for the collaboration on the
topics summarized here. I am also very grateful to Y. Maeno, S. Nakatsuji,
H. Yaguchi and K Wada for many helpful discussions. Furthermore I thank
to M. Indergand for careful reading of the manuscript. These works had
financial support from the Swiss Nationalfonds and a Grant-in-Aid by the
Japanese Ministry of Education, Culture, Sports, Science and Technology.

References

1. Y. Maeno et al., Nature 372,532 (1994).

2. Y. Maeno, T.M. Rice and M. Sigrist, Physics Today 54, 42 (2001).
3. T.M. Rice and M. Sigrist, J. Phys. Condens. Matter 7, L643 (1995).
4. G. Baskaran, Physica B 223-224, 490 (1996).
5. K Ishida et al., Nature 396, 658 (1998).
6. G.M. Luke et al., Nature 394,558 (1998).
7. M. Sigrist et al., Physica C 317-318, 134 (1999).
8. D.F. Agterberg, T.M. Rice and M. Sigrist, Phys. Rev. Lett. 78, 3374 (1997).
9. KK Ng and M. Sigrist, Europhys. Letts. 49, 473 (2000).
10. M. Sigrist et al., J. Phys. Soc. Jpn 69 Suppl. B, 127 (2000).
11. J. Goryo and M. Sigrist, J. Phys. Condens. Matter 12, L599 (2000); Europhys.
Letts. 57, 578 (2002).
12. T. Imai, A.W. Hunt, KR. Thurber and F.C. Chou, Phys. Rev. Lett. 81, 3006
(1998).
13. K Ishida et al., Phys. Rev. B 64 100501 (2001).
14. M.E. Zhitomirsky and T.M. Rice, Phys. Rev. Lett. 87, 057001 (2001).
15. H. Kusunose and M. Sigrist, cond-mat/0205050.
16. I. Bonalde et al., Phys. Rev. Lett. 85, 4775 (2000).
17. Y. Yanase and M. Ogata, preprint.
18. T. Nomura and K Yamada, J. Phys. Soc. Jpn. 71, 1993 (2002).
19. T. Nomura and K Yamada, J. Phys. Soc. Jpn. 71, 404 (2002).
20. Y. Maenoe et al., Phys. Rev. Lett. 81, 3765 (1997).
21. M. Sigrist and H. Monien, J. Phys. Soc. Jpn. 70, 2409 (2001).
22. Z.Q. Mao et al., Phys. Rev. Lett. 87, 032003 (2001).
23. H. Yaguchi, M. Wada, T. Akima, Y. Maeno and T. Ishiguro, in preparation.
24. S. Nakatsuji and Y. Maeno, Phys. Rev. Lett. 84, 2666 (2000); Phys. Rev. B 62,
6458 (2000).
25. Z. Fang and K Terakura, Phys. Rev. B 64,020509 (2001).
26. T. Nomura and K Yamada, J. Phys. Soc. Jpn. 69, 1856 (2000).
27. V.I. Anisimov et al., Eur. Phys. J. B 25, 191 (2002).
28. S. Nakatsuji, Ph.D. Thesis, Kyoto University (2000).
29. M. Troyer and M. Sigrist, in preparation.
REAL STRUCTURE OF PEROVSKITES LOOKED AT FROM
THE BAND STRUCTURE POINT OF VIEW

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

K.-D. SCHOTTE and C.T. LIANG

Theoretische Physik, Freie Universitiit Berlin, Germany

Abstract. Although perovskites can be viewed as a dense packing of ionic spheres of

three different sizes in a cubic structure, there are several deviations from this ideal.
Most common is the tilting of the corner sharing octahedra like the Si0 6 octahedra in
MgSiO;J. For transition metal complexes like CuF 6 in KCuF;J Jahn- Teller distortions
occur. With the symmetry lowered or the unit cell enlarged, gaps in the density of
states grow or additional ones are opened. With local density band structure calculations
(LDA) the tilting angle of the octahedra can be determined by minimizing the energy,
in fair accord with the experimental values. With the same LDA Jahn-Teller distortions,
however, cannot be calculated reliably. Since the d-density of states of the transition
metal is placed above the p-density of 0 or F ions, a metal instead of an insulator is
found. With the remedy of the LDA+U method one regains a proper description of the
experimental facts. However, there are also cases for which the unmodified approach
seems to be valid, for instance the perovskite SrRuO;J with 4d-clectrons is correctly
calculated as a ferromagnetic metal. In studying the restricted perovskite class we want
to illustrate the quality of band structure calculations. The shortcomings are not so
much linked to an improper description of the transition metal ion open shell structure
but to the failure in estimating the potential differences between the ions correctly. This
is probably due to the wrong accounting of the unwanted self-interaction of the electrons
in the d-shells. We think that considering the real crystallographic structure is necessary
to sort out what is left over to be accounted for by correlated electron physics and not to
be misguided by some technical problems in the otherwise quite accurate LDA-methods.

Key words; Perovskite, Crystal Structure, Band Structure Calculation, Jahn- Teller Effect

1. What are the structural pecularities of perovskites?

The mineral CaTi0 3 found in the Ural mountains, gave its name perovskite
to a large group of compounds investigated intensively (1, 2). In solid state
physics their magnetic, ferro-electric and superconductivity properties were
the incentive. In mineralogy the high pressure conditions of the earth's

35
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 35–43.
© 2003 Kluwer Academic Publishers
36
, ~ .' -:- .. -. - ~ -.. ---.-
--
,~~-"-:""~-._-_. ~- --~ :,.. - ::;::~

, : ,
,

.. -

k
:0_---
~::::::~ -. - ..
.. - ..... "
:;., "r'
- .. -
x
Figure 1. Structure of Re03 as an ideal network of Re06 octahedra

mantle favor a denser packing and make perovskites an abundant mineral.

The silicates of the earth's crust transform there to the perovskite MgSi0 3
or Mg 1 - x Fe x Si0 3 , see ref (3).
The basic unit of the perovskite ABX 3 consists of corner-linked octa-
hedra of X anions (usually oxygen or fluorine) with B cations at their center
and cations A sitting in the empty spaces between them. In a cubic unit
cell the positive ions A are at the corners, the negative ions X in the middle
of the faces and cations B in the center of the cube (see fig.1 above). The
Goldschmidt ratio, defined as

(1)

is 1 geometrically. Using the ionic radii for the two distances d one obtains
usually smaller values, so that Mg2+ in MgSi0 3 is too small for the space
available. The tolerance ratio t < 0.9 is an indicator for quite a number of
perovskites according to the data collection of Wyckoff (1) deviating from
a simple cubic structure with one formula unit per cell.
The compound ABX 3 considered as a corner sharing network of oc-
tahedral complexes BX6 indicates that the octahedra taken as rigid units
have the freedom to rotate or tilt. Since there are many possibilities for such
displacements a large variety in the structure of perovskites is possible (4).
 37

Figure 2. One unit cell of orthorhombic SrRuO;j. The rotations can be classified as
b+ a - a - after (4), that is rotations around the x, y, z directions of an octahedron having
an opposite sense a- or the same b+ in a adjecent octahedra along the rotation axes. Ru
(smallest) and 0 are connected by bonds, the largest spheres are Sr.

The assumption of rigidity is no longer valid, if the central atom B

is Jahn-Teller active, like Cu2+ in KCuF 3 or Mn3+ in GdMn03. In the
ferro-electric substances like BaTi0 3, where Ti is no longer in the center
of 06-octahedron, the distortions of the octadron are relatively small. The
influence on the structure is also small compared to tilting of octahedra
or the Jahn-Teller effect. In the following sections we want to adress the
question whether this can be calculated in a band structure approach. The
tilting will be discussed first.

2. The tilting of octahedra

Why octadra in perovskites tilt can be answered by looking at the com-

pound Re03 which has a network of corner sharing Re06 complexes depic-
ted in fig.I. However, there is no cation A between the octahedra as in the
perovskite structure. For this substance and also for the few others listed
under ScF3 type in (1) tilting does not occur.
In a perovskite ABX 3 the positive A ions attract the negative X ion
and only the repulse forces from the core of the A ion can stabilize the
38

10

en
OJ

~
4-<
0
0
q
~
OJ
CI

-10

-20'------'-------'-------'------'------'-------'-------'-------'------'-----"'-----L--'-------'-------'------'------'-------'-------'-------'------'
-10 -8 -6 -4 -2 0 4 6 10
Enerh'Y leV]

Figure 3. Density of states of orthorhombic SrRu03. Spin-up upper curve, spin-down

lower one. F p and Ru d states are mixed, around and above E F = 0 mostly d states.

structure. However, to fix the X-ion at the high symmetry position is not
possible if the core is too small indicated by the tolerance ratio, see eq(l),
0.75 < t < 0.9. Since further the BX 6 octahedron is a rigid unit only a
collective movement of all the X-ions is possible, as classified by Glazer (4).
A typical example can be seen in fig.2, where the structure of SrRu03
is displayed as determined by neutron scattering (5). The earth's mantle
material MgSi0 3 (3) has the same structure like many others found first in
the ferrite and "parent" substance GdFe03 (6). Four molecules ABX 3 are
contained in a orthorhombic unit cell with the space group Pnma (7).
Provided one has a so called "full-potential" program that avoids spher-
ical averaging of the ionic potentials, one can determine this structure with
the present band structure codes using the local density approximation.
However, to find a minimum of the total energy close to the tilted structure
determined experimentally may be time consuming (see (3) and fig.4).
From band structure point of view a more interesting object than the
insulator MgSi0 3 is a metal like SrRu03' especially since it is a ferromagnet.
Due to the lower symmetry and larger unit cell compared to the cubic one,
diffraction effects could lead to gaps in the density of states. If such changes
are close to the Fermi energy, they influence the electronic transport and
certainly the magneto-resistance.
We repeated the calculation of Singh (8) with the flp02-program (9).
As seen in fig.3 gaps appear, which for spin-up electrons are close to the
 39

Fermi energy. The 4 d states of Ru are split by crystal field effect into t g
and ego The e g states are empty and the four d-electrons occupy only the t g
states. In an isolated ion three electrons would fill the up states completely
and one of the down states, which would give spin 1. The band structure
gives a lower value of 0.75. The small gap is due to a small hybridization
of t g and e g , which would be absent for cubic SrRu03 without tilting. We
suspect that correlation effects could pin this feature really to the Fermi
level.

3. Jahn-Teller effect in the perovskites

Unlike the tilting of octahedra discussed, the band structure approach has
difficulties to reproduce the J ahn-Teller distortions of the transition metal
perowskites with J ahn-Teller active ions like Cu 2 +, Mn3+ and Cr 2 +. For
these ions the Jahn-Teller effect is equivalent to the orbital ordering (10)
of e g states. These states are doubly degenerate in a perfect octahedral co-
ordination of the transition metal ions. Lifting this degeneracy by lowering
the symmetry is the essence of the Jahn-Teller effect (11).
Spin ordering appears similar to orbital ordering in that the two-fold
degeneracy between up and down direction is lifted. This can be treated by
modifying the local density approximation (LDA) into a local spin density
approximation (LSDA). Otherwise ferromagnetism or anti-ferromagnetism
would not occur in a band structure calculation (12). Similar modifications
seem to be necessary to generate orbital ordering.
This has been done by adding to the standard LSDA an orbital depend-
ent potential (14) which is called shortly LDA + U. That indeed an orbital
ordering and a Jahn-Teller effect in fair agreement with the structural data
occurs had been demonstrated by Liechtenstein (15).
We repeated his calculation for KCuF 3 using the "Wien2k"-code where
a LDA + U version is included (19). In order to keep the computational
effort low we have chosen the ferro-distortive KCuF 3(16) and imposed fer-
romagnetic spin ordering so that only two formula units were in a tetragonal
cell. The type of magnetic ordering has little influence on the structure.
P4/mbm (7) the Wyckoff positions are 4h: (~ - 5, ~ + 5, ~) for F in ab-
plane around the Jahn-Teller active Cu at 2c: (0, ~, ~) as shown in fig.4.
K has the position 2a: (0, 0, 0) and the other F at 2d: (0, ~' 0). We have
taken for U = 0.59, see eq(2), and J = 0.07Ryd. 1 The lattice constants are
a = 5.8543 A and c = 3.9303 A. The ordering of the orbitals can be seen in
fig.5.

1 To run the Wien2k-program the muffin-tin radii were set to Rcu = 2.0, RK = 2.5,

Rp = 1.35 (in Bohr) and the number of k-points restricted to 210.

40

-0.355
, I
" I
", II
, I

TI -0.360
'+, I
I
>. , I
~ \ I
o
t.ri
o
\, I
f
N
o
-{)365
+, I
\ I
I \ I
>.
Dl
, I
<Ii ~ I
c: \ I
w \ experimental /
-0.370 , I I
\
", :)-
I /
/

~ I /
'....... I ",,;'
-0.375 "*-----+'"
o 0.01 0.02 0.03 0.04
Distortion ,j 2S
Figure 4. Total energy as a function of the Jahn-Teller distortion (j as indicated in the
insert showing the tetragonal unit cell of KCuF a seen along the c-axis with (jcxp = 0.022.

Also with a Hartree-Fock band structure calculation orbital ordering or

cooperative Jahn-Teller leads to lowering of energy in the case of KCuF 3 as
has been shown (13). KCuF 3 analyzed by LSDA is a nonmagnetic metal,
whereas Hartree-Fock generates a gap of 1.3 Rydberg. In choosing the U in
LDA + U approach one has the means to tune the gap to a more reasonable
value and get a better agreement with the structural data.
A counterexample seems to be the ferromagnetic insulator K 2 CuF 4
where a small gap appears and ferromagnetism is found using the unmodi-
fied LSDA (17) . No minimum in energy for a distorted structure similar to
fig.4 above could be determined. Possibly the precision of the programs in
the milli Rydberg range is a limiting factor. With the Wien-code (19) we
found a shallow minimum reduced by a factor of ten. This situation may be
comparable to NiO where also a small gap is found with the LSDA (18). In
order to get a larger and realistic gap using LDA + U (14) one has a remedy
for the other transition metal oxydes where the unmodified approach fails
to produce gaps.
 41

4. Eliminating the self-interaction solves all problems?

Why does "Slater's dream machine" 2 fail so badly with the transition
metal oxides or halides? "Correlation" is the standard answer. However, in
Hartree-Fock without any additions and correlations, the orbital ordering
and generating of gaps seem to be no problem (13). Following ref. (14)
the LDA + U ansatz is constructed by adding a Hubbard-like term and
subtracting its mean value, that is the energy is

ELDA+U = ELDA - ~ U N(N - 1) + ~ UL ninj , (2)

iclj

where N = 2:i ni is the total number of electrons in the d-shell and ni the
occupation of sublevel i. One assumes here that the standard LDA leads
to a reasonable total energy but the energy of the level should be changed.
By taking the derivative one obtains an orbital dependent potential

(3)

so that an occupied level is shifted downward in energy by - ~ U with ni = 1

but an unoccupied level is shifted upward by +~ U since ni = O. Summing
the shifts of the occupied levels one arrives at -~ UN, which could be
interpreted as correcting the Coulomb self-interaction of N d-electrons.
Whether a self-interaction corrected (SIC) method (20) can also lead to
orbital ordering in perovskites is not known. Eliminating the unwanted self-
interaction one does not have the problem of a best choice for U in LDA + U,
which is still under debate. However, the computational procedure for SIC
is complicated as may be inferred from the method to solve Hartree's self-
consistent equations where the self-interaction is also excluded (21).

Acknowledgements

We thank Jurgen Kubler, Helmut Eschrig and Peter Blaha for their help.

References

1. R.W.G. Wyckoff, Crystal Structures, Vo!. 2, Inorganic Compounds RX n , R n MX 2 ,

RnMX:J , Interscience Pub!. (1964).

2 P.W. Anderson in his talk at the "International Conference of Theoretical Physics"

in Paris 2002 referring the significant developments of solid state physics in the past.
42

Figure 5. Density plots: above the sum and below the difference of spin up and down.
The contour lines have values 0.2, 0.4 A-~ .. The horizontal axis is in the ab-plane along
the Jahn-Teller distorted Cu-F bonds, vertical is the c-axis. The e g hole orbitals alternate
as described in ref. (10). Above the Cu ions shrink where the hole orbitals are largest.

2. J.B. Goodenough & J.M. Longo, Crystallograhpic and Magnetic Properties of Per-
ovskites and Perovskite-Related Compounds, Landolt-Bornstein Neue Serie III/4a,
p.126, Springer Verlag (1970)
3. Lars Stixrude and R.E. Cohen, Nature 364, 613 (1993).
4. A.M. Glazer, Acta Cryst. B 28, 3384 (1972), see also Acta Cryst. A31,756 (1975).
5. C.W. Jones, P.T. Battle, P. Lightfoot and W.T.A. Harrison, Acta Cryst. C 45, 365
(1989).
6. S. Geller, J. Chern. Phys. 24, 1236 (1956).
7. International Tables for X-Ray Cr·ystallography, Kynoch Press, Birmingham (1969).
8. D.J. Singh, Electronic and IvIagnetic properties of the 4d itinerant ferromagnet
SrRuO:J, J. Appl. Phys. 79, 4818 (1996).
9. K Koepernik and H. Eschrig, Full-Potential Nonorthogonal Local-Orbital
minimum-basis band structure scheme, Phys. Rev. B 59, 1743 (1999).
10. KL Kugel and D.L Khomskii, The Jalm-Teller Effect and Magnetism: Transition
metal compounds, Sov. Phys. Usp. 25, 231 (1982).
11. M.D. Sturge, Jalm-Teller Effect in Solids, Solid State Physics 20, 91 (1967).
12. J. Kubler, Theory of Itinerant Electron Magnetism, Clarendon Press (2000).
13. M.D. Towler, R. Dovesi and V.R. Saunders, Phys. Rev. B 52, 10150 (1995).
14. V.L Anisimov, LV. Solovyev, M.A. Korotin, M.T.Czyzyk and Sawatzky, Density-
Functional Theory and NiO Photoemission Spectra, Phys. Rev. B48, 16929 (1993).
15. A.L Liechtenstein, V.L Anisimov, J. Zaanen, Phys. Rev. B 52, R5467 (1995).
16. M.T. Hutchings, E.J. Samuelsen, G. Shirane & K Hirakawa, Phys. Rev. BI88, 919
(1969).
17. V. Eyert and K-H. Hock, J. Phys.: Condens. Matter 5, 2987 (1993).
 43

18. K. Terakura, A.R. Williams, T. Oguchi and J. Kiibler, Transition-Metal Monoxides:

Band or Matt Insulators, Phys. Rev. Lett. 52, 18:30 (1984).
19. P. Blaha, K. Schwarz, G.K.H. Madsen, D. Kvasnicka and J. Luitz, WIEN2k, An
Augmented Plane Wave + Local Orbitals Program for Calculating Crystal Proper-
ties, (Karlheinz Schwarz, Techn. Univ. Wien, Austria), ISBN 3-9501031-1-2 (2001),
see also P. Novak, $WIENROOT/SRC/novak-lecture_on_ldau.ps (2001).
20. A. Svane, Electronic Structure of La2 CU04 in the Self-Interaction-Corrected
Density Functional Formalism, Phys. Rev. Lett. 68, 1900 (1992).
21. C. Edmiston and K. Ruedenberg, Rev. Modern Phys. 35, 457 (1963).
TOWARDS MODELS OF MAGNETIC INTERACTIONS IN
THE CUPRATES

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

H. ESCHRIG, K. KOEPERNIK and 1. CHAPLYGIN

IFW Dresden, P.O.B. 270016, D-01171 Dresden, Germany

Abstract. The electronic structure of undoped cuprates is reanalyzed with the help of the
LDA+U approach of density functional theory. It is found that in-plane oxygen orbitals
forming 'if-bonds with copper are in energetic concurrence to those forming the Zhang-
Rice singlet. The corresponding bands display a surprisingly large c-axis dispersion due
to hybridization with unoccupied cation d-states in the adjacent layers. Hence, they must
have relevance to c-axis magnetic coupling. In view of this situation, the Emery model is
incomplete for the description of magnetism of undoped cuprates.

Key words: Electronic Structure, Undoped Cuprates, Density Functional Theory, LDA
+ U, Strong Correlations, Hubbard Model, Emery Model, Zhang-Rice Singlet, Magnetic
Coupling, t-J Model

1. Density Functional Theory and Strong Correlation

The variational principle by Hohenberg and Kohn for the ground state
energy E and spin density n of electrons in a static external field u:

E [u, N] = mJn {H [n]

n
+ (u In) I (1 In) = N} , (1)

where we use the short-hand notation

(u I n) = L Jd ru 3
88 m 8 '8 = Jd 3 r(un - B· m), (1 In) = L Jd rn
3
88 ,

88' 8

(2)
holds true in any case of arbitrarily strong correlation. It is based on many-
particle quantum theory by rigorous mathematics. Of course, the density
functional H[n] is unknown.
Since the external field u must be static, it does not cope with non-
adiabatic electron-lattice interaction.

45
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 45–53.
© 2003 Kluwer Academic Publishers
46

The electronic quasiparticle excitations (if they exist) are obtained from
the coherent part (pole term) of the single particle Green's function:

G ss' (r,r I.) _ "Xs(r)rl;,(r ' ) + Gincoh(

,W - ~ ss' r,r I.)
,W , (3)
k W - Ek

the pole Ek of which is obtained from Dyson's equation for W ----7 E:

L Jd3r ' [o(r-r l

) ( - ~2 +ll(r)+llH(r)) +I;ss,(r, r' ;Ek)] Xs,(r' ) = Xs(r)Ek
s'
(4)
Here, the Hartree potential llH and the self-energy I; are functionals of the
ground state density n (we consider the case of temperature T ----7 0).
Recall that the assumption of a pole in the Green's function is a working
approximation of the theory. In principle from the full I;ss' (r, r ' ;w) the total
energy might also be obtained. We have only model theories for both H[n]
and I;[n].

2. Model Functionals

In the standard modeling of H[n], the Kohn-Sham (KS) ansatz

(5)

is used to split the density functional into an orbital variation expression

K[n] and a (possibly generalized by gradient terms) local density expression
L[n]:
H[n] K[n] +L[n]
K[n] min {k[¢i' nil I
{<p"n,}
L ¢ini¢; = n}
i
(7)

L[n] = J d3 rn(r)1(nss ,(r), Vn, ... ).

The well known LSDA model is obtained with

and 1(n) taken to be the exchange and correlation energy of the homogen-
eous electron liquid.
 47

K[n] works well for weakly interacting particles in an arbitrary external

potential. L[n] works well for weak external potential (nearly homogeneous
system) with arbitrarily strong interaction provided the total energy of the
homogeneous system as a function of the density is known.
The generalized KS equation

(9)

is obtained by variation of cP'k, variation of the orbital occupation number

nk yields the aufbau principle.

2.1. THE ORBITAL FUNCTIONAL AND L(S)DA+U

In the L(S)DA+U model, the k-functional is extended by an eu-functional

which is expressed through occupation numbers iiJLa of correlated local
orbitals IRi~(T) (centered in the unit cell at ~):

(10)
The corresponding U-potential is an operator containing a projector:

1 6
(11)
nk 6cP'k eu =

2.2. THE ORBITAL OCCUPATION MATRIX

The L(S)DA+U models depend on the basis set of the solver of the KS
equation. We will use for the latter a nonorthogonal local-orbital basis
(1), since a local orbital representation is mandatory for considering strong
correlations. For an LMTO solver see (2), for an LAPW solver (3).
Consider KS orbitals Ik; = IcPk; and orbital occupation numbers nk
as previously; they need not be eigenstates of spin. Let {Il)} 3 Ii) be a
possibly non-orthogonal basis for KS orbitals: Ik; = 2: z ll)Czk' Sll' = (lll').
The occupation matrix ii = ii[cPk' nk] of correlated orbitals Ii) at site R i in
an orthogonal form is introduced as (see Fig. 1)
 48

Ii} ~i)
Ik)n1/2 k
Ii} = ~Z Il)(S-l )Zi, (lli} = 6Zi
Il)
Figure 1. Orthogonal projection on the contragredient orbital.

The orbital occupation matrix may be diagonalized with respect to

m, m' at each lattice site R i and for each spin value (J" independently:

(14)
Averages are:
(15)

2.3. THE ORBITAL POLARIZATION LSDA+U FUNCTIONAL

This functional was introduced under the name 'around the mean field'
(AMF) in Ref. (4). It is zero if the orbitals of an atomic shell are equally
occupied, hence it depends on orbital polarization. It is given by

l(nss,(r), ... ) = kSDA,

e&MF } L {(fLafL'-alwlfLafL'-a)(n/La- na)(n/L'-a - n-a) +

IW/L/L' }
+ [(ItaIt~ 11Vl fJ·a fJ·~) - (Ita It~ Iiii IfJ·~ fJ·a)] (n/La - na)(filL'a - na)

} L {(fLafL'-alwlfLafL'-a)nILan/L'-a + (16)
IWILIL~ +[(fLafL~lwlfLafL~) - (fLafL~lwlfL~fLa)] nlLanlLa}
- 2 L { U(N - na) - J ( Na - na) } Na,
IW
Na = L nlLa = (2l + l)na.
/L
The corresponding U-potential is

onlLa
L{ (ItaILalwl/taILa)(fI/L'-a- n-a) +
/L'
+ [(fLafL~ Iw IfLafL~) - (fLafL~ Iw IfL~fLa)] (n/L'a - na)} (17)
 49

o·
x
x

Figure 2. Unit cell of antiferromagnetic CaCu02

2.4. THE 'ATOMIC LIMIT' LSDA+U FUNCTIONAL

The AMF-functional yields nothing for a completely spin-polarized half-

filled shell which is not so bad since the LSDA gives nearly the right
spin polarization energy of Gd, although there is a problem with the right
magnetic ground state. For this reason and aiming at a better description of
the photoelectron spectra, Czyzyk and Sawatzky introduced in (4) another
functional which they labeled 'atomic limit' (AL). At least regarding its
relation to photoemission it should rather be considered a model for the
quasiparticle self-energy L; instead of being related to eu. Nevertheless it
was given in an eu-form as (again l(nss,(r), ... ) = lLSDA)

~ L {({LU{L'-uliiJl/J,U{L'-u) iiJLuiiJL'-U +
&JLJL'
1 + [( {Lu{Lu/ 1-I / ) - ({Lu{Lu/ 1-I
W {Lu{Lu / )] nJLUnJLU
W {Lu{Lu - - }
-2 ~{UN(N -1) - JJ;Nu(Nu -I)} = (18)

e&MF + ~ L(U - J)(1 - iiu)Nu .

2&

(19)

The structure of the 'infinite layer' cuprate CaCu02 is shown in Fig. 2.

There are two relevant molecular orbitals (MO) which are candidates for
the highest occupied molecular orbital (HOMO) in the electronic structure
of cuprates, that is, those MOs which are relevant for the valence band
edge, see Fig. 3.
The LSDA+U KS bands of CaCu02 with the AMF and AL functionals
are shown in Fig. 4. On Fig. 5 the orbital weight to the AMF bands for
50

Figure 3. Left: Antibonding (k = (1f,1f,0)) dpo--orbital commonly assumed as the

HOMO that forms the Zhang-Rice singlet together with the nominal Cu-d hole (5, 6).
Right: 0-0 antibonding (k = 0) in-plane p1f-orbital, lifted up by crystal field and weakly
hybridized with Ca-d orbitals in adjacent layers: the true HOMO of the LSDA+U model
(7, 8).
AL
CaCu021 LSDA+U (U=8.1 eV, J=1eV) MCu =O.71
10.0

:> 50 ~ 5.0
~
~

'J 00
>-
~
~
W
-5.0

M A

Figure 4. LSDA+U band structure of CaCu02. Left: AMF, right: AL.

several orbitals is shown, that is, the square of the coefficient in the expan-
sion of the KS orbital into local basis orbitals is given by the linewidth. It
is clearly seen that both the 0 2pu and 0 2p7r orbitals hybridize equally
strongly with the Cu 3dx L y 2 orbital at the valence band edge. Thereby,
the considerable dispersion in z-direction of the 0 2p7r dominated bands
comes from hybridization with the unoccupied Ca 3dx L y 2 orbitals.
Sr2Cu02Cb is probably the most two-dimensional of all cuprates. The
antiferromagnetic unit cell is shown in Fig. 6 and the same orbital weights
as in the last section are shown on Fig. 7. The situation is quite similar
except that there is practically no dispersion in z-direction.

4. Implications for Magnetic Interactions

Gross magnetic couplings may be obtained from total energy differences of

LSDA+U results for ferromagnetic and antiferromagnetic order; in more
detail they may be obtained from calculated energies of spin spiral states
(9).
 51
AMF AMF
CaCu021 LSDA+U (U=8.1 eV, J=1 eV) MCu =O.71 CaCu021 LSDA+U (U=8.1 eV, J=1 eV) MCu =O.71
100 • Cu3dx 2 Y2 I 10.0 :--I-I--I~.'......"O."'2P-"--" -

~ 5.0 r-'----'-::-----4...._Ji""""~-
0.0 ~ 0.0 -
~
~-~---
.-
-50

rXM r
AMF
ZRA
- Z rXM r
AMF
ZRA Z
CaCu021 LSDA+U (U=8 1 eV J=1 eV) MCu=O 71 CaCu021 LSDA+U (U=8.1 eV, J=1 eV) MCu =O.71

10.0 ~ __"_~~" L 10.0

~ 50 ~ 5.0_

M M A

Figure 5. Upper panels: orbital weight of the Cu 3d x 2 _y2 orbital and the 0 2p" orbitals,
resp. Lower panels: the same for the 0 2p" and the 0 2pz orbitals.

z
z

x x

Figur'e 6. Unit cell of antiferromagnetic Sr2Cu02Cb

For a more detailed understanding of the physics , a tight-binding model

of the kind of Emery's model should be extracted from the LSDA+U res-
ults , which then may be down-mapped to a kind of a t - J model. For the
ferromagnetic order such a tight-binding model was derived in Ref. (8).
The conclusion from the present orbital analysis is that in a large part
of the Brillouin zone there is a strong hybridization of 0-2pa with 0-2p1r
orbitals in bands hybridized with the Cu-3d x L y 2 orbital. Hence, the 0-2p1r
orbitals must be included in the Emery model in order to correctly describe
the t-terms which determine the magnetic coupling.
In CaCu02 there is a sizable dispersion of the in-plane 0-2p1r bands
52
AMF AMF

--
Sr2Cu02CIZ LSDA+U (U=8.16 eV, J=1 eV) MCu =O.748 Sr2Cu02CIZ LSDA+U (U=8.16 eV, J=1 eV) M Cu =O.748

6.0 • Cu3d X2 "; I

6.0 . '"
• On

4.0 4.0 - - - - ,.•._ .. _-+---

~ 2,0 ~ 2.0

~ 00
>- I-. ..
~ -2.0
~ ....
W -4.0

-6.0

r X p X, z r r X P Xj Z r
AMF AMF
Sr2Cu02CIZ LSDA+U (U=8.16 eV, J=1 eV) MCu =O.748 Sr2Cu02CIZ LSDA+U (U=8.16 eV, J=1 eV) M Cu =O.748

60 ~ 6.0 -
• 0,

40 4.0 _

~ 2,0 ~ 2.0_

-60 -6.0 =-

in z-direction mediated by some hybridization with Ca-3d x L y2 orbitals.

This is in accordance with the experimental finding that CaCu02 has the
highest Neel temperature, TN = 540 K, of all layered cuprates indicating
3D magnetism (10, 11).
By contrast, in Sr2Cu02Cb there is no dispersion of the corresponding
bands in z-direction due to the Cb buffer layers. There is only dipole-
dipole coupling of the planes, compatible with the experimental findings,
TN = 256 K (12).

Acknowledgements

We acknowledge helpful discussions with R. Hayn, S.-L. Drechsler and H.

Rosner as well as financial support by the DFG (SFB463) and by the G1F
(project 1-614-13.14/1999).

References

1. K. Koepernik and H. Eschrig, Phys. Rev. B59, 174:3 (1999).

2. A. 1. Liechtenstein, V. 1. Anisimov and J. Zaanen, Phys. Rev. B52, R5467 (1995).
3. A. B. Shick, A. 1. Liechtenstein and W. E. Pickett, Phys. Rev. B60, 10763 (1999).
4. M. T. Czyzyk and G. A. Sawatzky, Phys. Rev. B49, 14211 (1994).
5. V. J. Emery, Phys. Rev. Lett. 58, 2794 (1987).
6. F. C. Zhang and T. M. Rice, Phys. Rev. B37, 3759 (1988).
7. J. J. M. Pothuizen et al., Phys. Rev. Lett. 78, 717 (1997).
8. R. Hayn et al., Phys. Rev. B60, 645 (1999).
 53

9. A. N. Yaresko et al., Phys. Rev. B65, 115111 (2002).

10. D. Vaknin et al., Phys. Rev. B39, 9122 (1989).
11. R. Pozzi et al., Phys. Rev. B56, 759 (1997).
12. M. Greven et al., Phys. Rev. Lett. 72, 1096 (1994).
C-AXIS INTRA-LAYER COUPLINGS IN THE CU0 2 PLANES
OF HIGH-To CUPRATES

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

J. ROHLER
UniversitCit zu Koln
Ziilpicher Str. 77, D-50937 Koln, Germany

Abstract. We discuss the static deformations of the doped CU02 lattices in YBa 2Cu 3 0x.
Intrinsic lattice effects driven by the strong correlations of the electron system are separ-
ated from extrinsic chemical and bandstructure effects. c-axis displacements of the planar
copper atoms seem to be a generic property of the metallic CU02 lattices.

Key words: Superconductivity, Cuprates, YBa 2Cu 3 0x, CU02 Lattice, C-Axis, Buckling,
Strongly Correlated Electrons, Doping Mechanism, X-Ray Diffraction, EXAFS

1. Introduction

Unconventional metals exhibit a plethora of unusual lattice effects not

observed in "simple" band metals. In the physics of strongly correlated
electron systems these puzzling "lattice instabilities" are usually ignored
or relegated to the high energy scale of the crystal chemistry. This lecture
focusses on the static distortions of the nearly square planar CU02 lattices
in the under- and overdoped regimes of YBa2Cu30x, the superconducting
cuprate with the best known lattice structure. We shall discuss the dop-
ing dependence of the basal lattice parameters, the CU02 dimples, and
the orthorhombicity. The structural data suggest that doping into a Mott
insulator leads to weak but significant deformations of the metallic CU02
lattice. Our focus is on the structural aspects of the normal state; it is
not intended to suggest the inclusion of electron-lattice interactions into a
specific model of high-Tc superconductivity.

55
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 55–63.
© 2003 Kluwer Academic Publishers
56

2. Planar and perpendicular orbitals

The pioneering work of Muller and Bednorz (1) has been guided by the
idea of high-Tc superconductivity (HTSC) in transition-metal chalcogen-
ides with strong J ahn-Teller (JT) effects. In fact, the strong JT effect is
confirmed to playa crucial role for the electronic structure of the super-
conducting cuprates selecting from the e g doublet the antibonding 3dx 2_ y 2
as the only relevant d orbital in the plane of the Cu-O polyhedron. It
generates the enormous stability of the square-planar CU02 structure. An
electronically active role of the perpendicular 3d z 2 orbital is however not
confirmed.
But superconductivity is not a purely 2-dimensional effect. T c depends
strongly on interactions of the planes with the doping reservoir, and a
mechanism linking the plane with its perpendicular surroundings is expec-
ted to be involved in the formation of the electronic ground state. Recent
calculations of the one-electron bandstructure of YBa2Cu307 (2) show that
the generic low-energy layer-related features have to be described by the
planar orbitals Cu3d x L y2, 02px,y directing in the plane, and the isotropic
Cu4s orbital. The chemical trend of T cmax is found to be controlled by
the energy of a "perpendicular orbital" , a hybrid between Cu4s, Cu3d z 2,
the apical 02pz, and farther d z 2 or pz orbitals. Interestingly materials with
the perpendicular orbital more localized onto the CU02 layers, i.e. with
pure Cu4s character, exhibit the highest T cmax. It seems that the strong
hybridization between Cu3d x L y 2 and Cu4s allows for atomic displacements
in the rigid covalent CU02 lattice of the parent Mott insulator such that
doped holes are optimally accomodated in the background of the quantum
liquid of spin singlets.
Notably the dependence of T cmax from the structure and the chemistry
perpendicular to the planes enters a single-band many electron Hamiltonian
simply by the ratio of the nnn and nn hopping integrals, t f It. It is worthwile
to stress that the (rare) acknowledgement of the chemistry outside the
planes as an integral part of the HTSC problem does not justify dynamically
independent treatments of the Cu3d and 02p electrons in any kind of multi-
band many electron Hamiltonians.

3. Dimpled CU02 planes

The CU02 planes of the superconducting cuprates are not perfectly flat
but dimpled. Many diffraction work, e.g. Ref. (3,4,6,8), and some EXAFS
studies (9, 10) yield evidence for dimples in the relevant CU02 planes of the
multi-layer cuprates, i.e. the planes adjacent to the doping reservoirs, cf.
table 1. "Buckled" planes are also observed in single-layer materials, most
 57

TABLE I. CU02 dimplings in typical HTSC, see text.

Compound Pl.. NL Dimpling (A) Tc Ref.

YBa 2Cu;jOx 3+ - 2+ 2 0.28 92 (3, 9)

YBa 2Cu 4 OS 3+ - 2+ 2 0.24 80 (8)
Pb 2Sr2 YCU;jOS 3+ - 2+ 2 0.23 70 (6)
HgBa2(Ca,Y) Cu20 x 2+/3+ - 2+ 2 0.10 85-110 (4)
HgBa2 CaCu2 Ox 2+ - 2+ 2 0.02 120 (4)
HgBa 2Ca 2Cu:JOx 2+ - 2+ 3 0.05 132 (4)
HgBa 2Ca:JCu4 Ox 2+ - 2+ 4 0.05 127 (4)

clearly in the doped La2Cu04 systems, but so far not in the single-layer Hg-
and Tl-cuprates (4). As we shall see, the dimplings in the latter materials
are expected to be small, and since their CU02 planes are crystallographic
inflection planes the usual crystallographic refinements tend to average
them to zero.
We distinguish "chemical" dimples driven by external electrostatic po-
larization along c, mixed valence of the dopants, lattice mismatches and
other bandstructure effects, from "correlation-driven" dimples generic of
the strongly correlated electron system. We show that the former define a
large length-scale of rv 0.3 A, and the latter a small length-scale of rv 0.03
A, corresponding to buckling angles of rv 10°, and rv 10, respectively. The
contribution of the chemical dimples in many of the doped metallic phases
can be traced back to the lattice of their antiferromagnetic parent phase,
and thus be differentiated from the correlation-driven dimples. However in
concentrated alloys, doped by varying ratios of one or two combinations of
heterovalent cations, the correlation-driven dimples may be masked by the
varying external electrostatic polarizations (12). Table I compares the CU02
dimplings in some typical multi-layer cuprates with each other: as a function
of the static charge contrast polarizing the CU02 layer perpendicularly, Pl.,
the number of CU02 layers, N L , and T e . The large length-scale of rv 0.3
A shows clearly up in all compounds with Pl. = 2+ - 3+, but is absent
in compounds with P1.. = 2+ - 2+. Materials alloyed with heterovalent
cations, e.g. Ca2+ / y3+, exhibit intermediate values. Notably CU02 layers
sandwiched between isovalent layers seem to exhibit higher T emax than
those between heterovalent layers, cf. Ref. (7).
58

4. Perpendicular and planar CU02 deformations in YBa2Cu30x

Figure Ib sketches a metallic CU02 layer of YBa2Cu30x as a stack of

copper and oxygen layers. The horizontal dashed lines in Figure la indicate
a "chemical" offset of "':' 0.23 A, typical for Ba+ 2/y+ 3 polarization. The
spacing between the Cu2 and 02,3 layers depends clearly on doping. While
the Cu2 layer starts to move towards the Ba-O layer by about 0.05 A
between the onset of the metallic phase (x "':' 6.38) and Xopt, the 02,3
layers seem to be almost unaffected. We show in figure 3 that the scatter
in the Y-02, 03 data points does not behave arbitrarily, and that the
relatively weak perpendicular displacements of 02 and 03 are correlated
with the planar deformations.

4.1. LATTICE PARAMETERS AND ORTHORHOMBIC DEFORMATIONS

The relatively strong orthorhombicity of its unit cell puts YBa2Cu30x

somewhat outside the lattice systematics of the most prominent high-Tc ma-
terials, which have mostly tetragonal unit cells. Weak orthorhombic strain
might be however operative also on the nominally tetragonal lattices, if a
correlation-driven anisotropy of the 2-D Fermi surface is a generic feature of
HTSC, as theoretically suggested by renormalization group methods (11).
Figure 2a displays the doping dependence of the lattice parameters
a, b, c/3 in the metallic phases of YBa2Cu30x' a i- b in the under- and
overdoped regimes, but the inequality results clearly from different types
of orthorhombic strain ellipsis. The quadrupolar a-ortho stresses a while
straining b, and the monopolar p-ortho stresses both axis. "Plateaus" of a

~ 1.65--------------
co Ba'+·O
N Cu2
o 1.6
>-
8 1.55
YBa Cu 0 Cu2
2 3,
>- 5 K
~ 1.5 (O.20)+O.o.BA
'u 02 0,0211 I --- I
g. 1.45 02,3
03
'-----------------!!-~--~------:--;;;----:-;-;;-----i-
1.4 '--'-~~~~~~~~~

6 6.2 6.4 6.6 6.8

a. Y-Cu2, Y-02,3 spacings b. CU02 intralayer spacings

Figure 1. Layer spacings in YBa 2 Cu 3 0x. a.: Y-Cu2, and Y-02,3 interlayer spacings
as a function of oxygen concentration, x, at 5 K. After Cava et al. (3). Open triangles:
02. Open circles: 03 . b.: CU02 plane sketched as a stack of copper and oxygen layers
sandwiched between polarizing adjacent layers (arbitrary scale).
 59

and b between x ':::::' 6.8 and 6.9 connect the two regimes suggesting a thermo-
dynamical instable regime. If the orthorhombic strain originated exclusively
from Cul-04 hybridization in the I-dimensional charge reservoir, we expect
the b-, and the a-axis to exhibit similar orthorhombic deformations in both,
the under- and overdoped regimes. The abrupt change from the quadrupolar
,G-ortho to the monopolar ,G-ortho deformation at Xopt points however to
an additional and different mechanism to be operative.
Figure 2b displays the doping dependence of the basal area, B(x) = a·b.
Evidently accomodation of oxygen atoms in the reservoir layer does not
increase B(x) as might be expected from simple stereochemical grounds,
but reduces the basal area. Hence the doping dependence of B(x) is most
likely controlled by the ground state of the quantum liquids in the metallic
planes, and not by the anisotropic oxygen diffusion processes in the non-
stoichiometric chain layer. To discuss B (x) in more detail we find it useful
to associate "'p ex: -0B /onh with a 2-dimensional "electronic compress-
ibility", eliminating the almost doping independent bandstructure effects.
Here nh ex: x denotes the hole concentration.
"'p ':::::' 0 in the insulating phase. The plane of the spin lattice in the
lightly doped Mott insulator seems to be incompressible as indicated by
the arrow for x = 6.0 - 6.38. The magnetic exchange energy J of order
1500 K determines the electron-electron interactions in the Cu2 planes.
"'p > 0 in the metallic phases. While weakly decreasing in the under-
doped a-ortho regime, B (x) starts to collapse in the overdoped ,G-ortho
regime. The plateau between x = 6.8 and 6.9 points to phase segregation,
possibly into incommensurate stripe phases. In the metallic phase the doped

c/3 _x=6.0 ~ 6.38

3.9
14.9 (J.

I
3.88 b ~ I

~ I
I ~ 14.88
1 ro
'3 3.86 (J. I
1
~
D I
ro I ~ 14.86
3.84 <Il

a~' I.
I
I
I
•
\.0
\
YBaCuO I· . 14.84 YBa Cu 0
3.82 2 3 x I 2 3, I "'\

300 K I • ~
300 K : opt. opt.
3.8 L.-~~~~~...L....-~~~-....J 14.82 L.-~~~~~~~~~-...J
6.4 6.5 6.6 6.7 6.8 6.9 6.4 6.5 6.6 6.7 6.8 6.9

a. Lattice parameters b. Basal area

Figure 2. a.: Lattice parameters 'Vs. oxygen concentration in near-equilibrium samples of
metallic YBa 2 Cu 3 0," synthezised by direct oxydation of the elements (DO), after Conder
et al. (5). Drawn out lines are guides to the eye. a and f3 label different orthorhombic
strain ellipsis, see text. b.: Basal area, a· b, from the data in a. Dashed line: smoothed
average. The step indicates the onset of a possible two-phase regime.
60

I
02 I
YBa Cu 0 YBa CU 0 :
1.435 2 3, 1.435 2 3 x
~ 02_'\".
5K ~ 5 K
I
I

'"c: 1.43
'"o 1.43
I
I
I
>- >- ____ I
• .JI _
N
~ ! ' ~

o 1.425 0- 1.425 J.' I

<02,03"'"- - - - - - - : I"
>- >-
~

:? : ~
...... : ....
1.42
I .1 "-
.~ I I
<f) I I
1.415 I I
I I
~~
I I
03 I I
1.41 L . . . . . ...~~~~~~~~~ 1.41 L........~~~~-L.-..-~~...L.L'--'--"-'
6.5 6.6 6.7 6.8 6.9 6.5 6.6 6.7 6.8 6.9

a. Without c-axis minimum (3) b. With c-axis minimum (13, 14)

Figure 3. Spacings Y-02, Y-03 VB. oxygen content in YBa 2 Cu:jOx at 5 K from neutron
diffraction in different samples. Triangles: 02 (a-axis). Dots: 03 (b-axis). All lines are
guides to the eye. a.: In quenched samples from the carbonate route after Cava et al. (3).
b.: In near equilibrium samples from the BaO route after Kaldis (13), Hewat et al. (14).

holes hop with a nn matrix element t ex: d- n , n 2:> 2, and thus shorter
nn distances, d, increase strongly the kinetic energy of the holes. But the
physics of the doped Mott insulator is that of competition between the
exchange energy J and the kinetic energy per hole nht. The doped holes in
the underdoped regime appear only as vacancies in the background of a spin
singlet liquid. The lattice of such a strongly correlated t - J type electron
system is expected to be much harder than that of a nearly noninteracting
electron liquid.
Hence the a- and p-ortho deformations may be identified as charac-
teristic lattice responses to fundamentally different types of electron li-
quids in the metallic phase: a weakly compressible t - J like in the under-
and optimum doped, and a strongly compressible Fermi liquid-like in the
overdoped regimes.

4.2. PERPENDICULAR 02,03 DISPLACEMENTS

Figure 3 displays the doping dependence of the interlayer spacings Y-02,

Y-03 from samples without (3) and with (14) a c-axis minimum. At the
onset of the metallic phase the degenerate tetragonal positions of the planar
oxygens are split into 02 along a, and 03 along b. Both are also displaced
along c: 02 by rv -0.05 A below, and 03 by rv +0.05 A above the tetragonal
reference value, cf. figure 4b. The anisotropic displacement of the planar
oxygens along c is usually attributed to the anisotropic Coulomb repulsion
between the I-dimensional charge reservoir and the 2-dimensional planes.
 61

a
a
b

a. Tetragonal insulator b. a-ortho underdoped metal

02

a a

b
b

d. ;3-ortho overdoped metal

Figure 4. Sketches of the CU02 lattices in YBa 2Cu 3 0x. The frame indicates the
dimpling due to the surrounding chemistry and bandstructure effects.

This is yet another example for Coulomb repulsion and hybridization

are not able to correctly describe the doping-induced displacements in the
CU02 layer: increasingly perfected chains in the reservoir layer relax the c-
axis anisotropy of the planar 02, 03 instead of increasing them. Thus the
average spacing Y-<02,03> matches the degenerate tetragonal reference
value throughout the underdoped a-ortho regime (thick drawn out line
in figures 3a,b ). Meaningful crystallographic studies of the overdoped ;3-
ortho regime require samples with a c-axis minimum around Xopt (13).
Figure 3b displays the 02,03 c-axis displacements in such samples (13, 14).
Herein the transition into the overdoped regime seems to be connected with
weak unidirectional displacements of both oxygens 02, 03 towards the Y-
layer, cf. figure 4d. Studies of the local atomic structure by Y-EXAFS
(9, 13) confirm this unidirectional c-axis shift of the planar oxygens in the
overdoped regime also in standard samples without c-axis minimum (5).
We summarize the doping-induced displacements of Cu2 and 02,03 in
the schematic figures 4a-d:
i. Metallic hole concentrations increase the two-dimensional density of
Cu2 by dimpling the planes. There is some similarity with the mechanism
collapsing an umbrella (15). As Cu2 moves out of the plane the basal area
shrinks.
ii. The underdoped regime is governed by the quadrupolar a-ortho
strain. 02 and 03 shift oppositely along c such that the basal Cu2 area
may adjust to a relative maximum.
iii. Close to optimum doping the quadrupolar a-ortho strain vanishes.
02 and 03 may achieve nearly degenerate c-axis positions.
62

iv. In the overdoped regime the deformations have changed from the
quadrupolar 00- to the monopolar ,6-ortho type stressing the a-, b-axes in
the same direction therewith collapsing the basal Cu2 area. Both oxygens
02,3 shift perpendicularily in the same direction along e.

5. Concluding Remarks

We have undertaken an attempt to identify among the many doping-driven

lattice effects in YBa2Cu30x those which are most likely connected with the
generic low energy electronic structure. Fermi surface driven lattice effects
are well known from the classical metals. For instance about 3/4 of the
metallic elements tend to maximize the gain of kinetic energy by crystalliz-
ation in most closely packed structures hep, fcc, bee. In electron compounds
and Hume-Rothery alloys the Fermi energy is well known to avoid maxima
in the density of states and to drive structural transformations changing
appropriately the symmetry of the Brillouin zone. These mechanisms are
expected to be operative also in low dimensional and strongly correlated
electron systems. The competition between exchange and kinetic energies
in strongly correlated electron systems however gives rise to new and more
intricate lattice effects which, once disentagled from bandstructure effects,
may yield new insights into the many electron ground state.

Acknowledgements

I thank E. Kaldis for a stimulating and fruitful cooperation. The ESRF

supported this work partially through project HE731 at BM29.

References

1. J. G. Bednorz and K. A. Muller, Z. Phys. B 64, 189 (1986).

2. E. Pavarini, 1. Dasgupta, T. Saha-Dasgupta, O. Jepsen and O. K. Andersen, Phys.
Rev. Lett. 87, 047003 (2001).
3. R. Cava, , A. Hewat, E. Hewat, B. Battlog, M. Marezio, 1. K. K.M. Rabe, W. Peck
Jr. and L. Rupp Jr., Physica C 165,419 (1990).
4. H. Schwer and J. Karpinski, In: A. Narlikar (ed.): Studies of High Temperature
Superconductors, Vol. 24. p. 49 (1997).
5. K. Conder, D. Zech, C. Kruger, E. Kaldis, H. Keller, A. Hewat and E. Jilek: In: E.
Sigmund and K. A. Muller (eds.): Phase Separation in Cuprate Superconductors. p.
210 (1994).
6. R. Cava, M. Marezio, J. Krajewski, W. Peck Jr., A. Santoro and F. Beech: Physica
C 157, 272 (1989).
7. S. Kambe, and O. Ishii, Physica C 341-348, 555 (2000).
8. P. Fischer, J. Karpinski, E. Kaldis and S. R. E. Jilek, Solid State Comm. 69, 531
(1989).
 63

9. J. Rohler, , P. Loeffen, K. Conder and E. Kaldis, Physica C 282-297, 182-195

(1997).
10. J. Rohler, and R. Crusemann, In: E. Kaldis. K. A. Muller, D. Mihailovic, G. Ruani
(ed.), Anharmonic Properties of high-Tc C1tpmtes. pp. 86-94 (1995).
11. C. J Halboth and W. Metzner, Phys. Rev. B 61, 7364 (2000).
12. O. Chmaissem, J. D. Jorgensen, S. Short, A. Knizhnik, Y. Eckstein and H. Shaked,
Nature 397, 45 (1999).
1:3. E. Kaldis, In: K. A. Gschneidner Jr., L. Eyring and M. B. Maple (eds.), Handbook
on the Physics and Chemistry of Rare Earths, Vol. 31 (2001).
14. A. W. Hewat, E. Kaldis, S. Rusiecki and E. J. P. Fischer, unpublished. Data sheets
are included in Ref. 13 (1991).
15. Rohler, J.: 2000, Physica B 284-288, 1041.
SPECTRAL FUNCTIONS AND PSEUDOGAP IN A MODEL
OF STRONGLY CORRELATED ELECTRONS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

P. PRELOVSEK and A. RAMSAK

Faculty of Mathematics and Physics, University of Ljubljana,
1111 Ljubljana, Slovenia
J. Stefan Institute, University of Ljubljana, 1111 Ljubljana,
Slovenia

Abstract. The theoretical investigation of spectral functions within the single-band t-

J model, as relevant for superconducting cuprates, is presented. For spectral functions
the method of equations of motion is used, where for the self energy the decoupling of
spin and single-particle fluctuations is performed. Longer-range spin fluctuations induce
a pseudogap showing up at low doping in the effective truncation of the Fermi surface
and in reduced electron and quasiparticle density of states at the Fermi level.

Key words: Cuprates, Superconductivity, t-J Model, Strong Correlations, Pseudogap,

Spectral Functions, Fermi Surface, ARPES, Spin Fluctuations

1. Introduction

One of the central questions in the theory of strongly correlated electrons

is the nature of the ground state and of low energy excitations. Exper-
iments in many novel materials with correlated electrons reveal even in
the 'normal' metallic state striking deviations from the usual Fermi-liquid
universality as given by the phenomenological Landau theory involving qua-
siparticles (QP)(l). The attention in the last decade has been increasingly
devoted to the underdoped cuprates, where experiments reveal character-
istic 'pseudogap' temperatures T > T e , which show up crossovers where
particular properties change qualitatively. There seems to be an indication
for two crossover scales T* and T sg . T* scale (2) shows up clearly as the
maximum of the spin susceptibility Xo(T = T*), the kink in the in-plane
resistivity p(T), in the anomalous Hall constant RH(T < T*) and in the
specific heat r = C V IT (1). It seems plausible that the T* crossover is

65
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 65–73.
© 2003 Kluwer Academic Publishers
66

related to the onset of short-range antiferromagnetic (AFM) correlations

for T < T*, since this is clearly the case for an undoped AFM.
The single-particle spectral function A(k, w) and its properties are of
crucial importance. In recent years there has been an impressive progress
in the angle-resolved photoemission spectroscopy (ARPES) experiments, in
particular for cuprate materials, which in principle yield a direct inform-
ation on A(k, w). In most investigated Bi2Sr2CaCu202+5 (BSCCO) (3)
ARPES shows quite well defined large Fermi surface in the overdoped and
optimally doped samples at T > T e . On the other hand, in the underdoped
materials the quasiparticles (QP) dispersing through the Fermi surface
(FS) are resolved by ARPES in BSCCO only in parts of the large FS,
in particular along the nodal (0,0)-(1f,1f) direction (4, 3), indicating that
the rest of the large FS is truncated (5). At the same time near the (1f, 0)
momentum ARPES reveals a hump at rv 100 meV (4), which is consistent
with large pseudogap scale T*.
The prototype single-band model relevant for cuprates which takes
explicitly into account strong correlations is the t-J model,
1
H =- 2:tijC}sCis+ J2:(Si' Sj - -ninj), (1)
..
2,J,S (ij)
4

where fermionic operators are projected ones, ct = (1 - ni,-S)ct. We

consider besides tij = t for the n.n. hopping also tij = t f < 0 for the
n.n.n. hopping on a square lattice.
There have been so far numerous analytical and numerical studies of the
t-J model and related Hubbard model on a square lattice. The importance
of AFM spin correlations for the emergence of the (large) pseudogap was
found in some of these numerical studies (6, 7, 8) and in phenomenological
model calculations (9, 10).
First, we describe some evidence for the pseudogap within the t-J model
obtained via finite-size studies using the finite-temperature Lanczos method
(FTLM) (8). A most straightforward evidence for a pseudogap appears in
the uniform static spin-suceptibility Xo(T). The maximum T* being related
to the AFM exchange T* rv J in an undoped AFM gradually shifts to
lower T with doping and finally disappears at 'critical' Ch = ch. rv 0.15.
Obtained results are qualitatively as well as quantitatively consistent with
experiments in cuprates (1). Another quantity relevant for comparison with
the anaytical approach is the single-particle density of states (DOS) N(w),
which shows at smallest dopings a pronounced pseudogap at w rv 0 which
closes with increasing T rv J. This again indicates the relation of this
pseudogap with the AFM short-range correlations which dissolve for T > J.
On the other hand, the pseudogap closes also on increasing doping being
barely visible even at Ch rv 0.12.
 67

2. Spectral functions: Equation-of-motion approach

In our anaytical approach we analyse the electron Green's function for

projected fermionic operators,

(2)

In an equations-of-motion (EQM) method one uses relations for general

correlation functions w((A; B))w = ({A, B}+)+(([A,H]; B))w, applying them
to the propagator G (k, w) (11),
0:
G (k, w) = ---------,-,--------,-- (3)
w + If, - (k - I;(k,w)
We notice that the renormalization 0: = (1 + ch)/2 < 1 is already the con-
sequence of the projected basis, while (k represents the new 'free' propaga-
tion term

(4)

where Tlj = 0: + (So·Sj) /0: and rk = (cos k x + cos k y)/2, r~ = cos k x cos kyo
The central quantity for further consideration is the self energy

(5)

where only the 'irreducible' part of the correlation function should be

taken into account in the evaluation of I;. We express EQM in variables
appropriate for a paramagnetic metallic state

(6)

where f~ is the bare band energy, ni = ni - Ce and mkq is the effective

spin-fermion coupling,
mkq = 2J rq + q· fL (7)
One important achievement of EQM method is that it naturally leads to
an effective coupling between fermionic and spin degrees of freedom, which
are essential for the proper description of low-energy physics in cuprates.
Such a coupling is e.g. assumed as input in phenomenological models (9, 10)
as the spin-fermion model. The essential difference in our case is that mkq
is strongly dependent on k just in the vicinity of most relevant 'hot' spots.
We assume that spin fluctuations remain dominant at the AFM wavevector
Q with the characteristic inverse AFM correlation length K, = l/~AFM.
68

It is sensible to divide spin fluctuations into two regimes with respect to

q = q - Q: a) For q > '" spin fluctuations are paramagnons, they are
propagating like magnons and are transverse to the local AFM ordering.
b) For q < '" spin fluctuations are critically overdamped and the deviations
from the long range order are essential.
For q > '" the decoupling of spin and fermion degrees of freedom
reproduces for an undoped AFM the evaluation of ~(k, w) within the
self-consistent Born approximation (12) which we generalise within the
linearized magnon theory into a paramagnon contribution at finite doping
(at T = 0), (13)

~pm(k,w) = ~ ~ [M~qG-(k - q,w +wq ) + M~+q,qG+(k + q,w - wq )],

q,q>",
(8)
where G±(k, w) refer to the electron (w > 0) and hole-part (w < 0) of the
propagator, respectively. We are dealing with a strong coupling theory due
to t > wq and a selfconsistent calculation of ~pm is required.
At Ch > 0 besides the paramagnon excitations also the coupling to
longitudinal spin fluctuations becomes crucial. The latter restore the spin
rotation symmetry in a paramagnet and EQM (6) introduce such a spin-
symmetric coupling. Within a simplest approximation that the dynamics
of fermions and spins is independent, we get (13)

(9)

where 9 = (1/2)[th(pwl/2) + cth(pw2/2)] and X is the dynamical spin

susceptibility. Quite analogous treatment has been employed previously in
the Hubbard model (14) and more recently within the spin-fermion model
(9, 10).
If we want to use the analogy with the phenomenological spin-fermion
model the effective coupling parameter mkq should satisfy mk,q = mk-q,-q
therefore we use instead the symmetrized coupling mkq = 2Jrq + (E~_q +
E~) /2. In contrast to previous related studies of spin-fermion coupling (9,
10), mkq is strongly dependent on k, but also quite modest along the AFM
zone boundary ('hot' spots), here determined solely by J and t'. For A we
first insert the unrenormalized A 0 , i.e. without ~lf.
In the present theory spin response X( q, w) is taken as input. The system
is close to the AFM instability, so we assume the overdamped form

"() W (10)
X q,w CX(-2
q + '" 2)(W 2 + w'"2)'
 69

Nevertheless, the appearance of the pseudogap and the form of the FS

are not strongly sensitive to the particular form of X" (q, w), provided that
characteristic"" and w" are fixed.

3. Pseudogap

We first establish some characteristic features of the pseudogap and the

development of the FS following a simplified analysis (13). ~pm induces a
large incoherent component in the spectral functions and renormalizes the
effective QP band, leading to ckf and reduced QP weight Zkt. Neverthe-
less, the pseudogap can appear only via ~lf' We restrict ourselves here to
T = 0 and to the regime of intermediate (not too small) doping, where ckf
defines a large FS. The simplest case is the quasi-static and single-mode
approximation (QSA) which is meaningful if w" « t, "" « 1, where we get

QSA _ a Z kef(w - ck_Q

ef)
G (k,w)-( W ef)( ef) 2' (11)
- ck_Q w - ck -,6.k

The spectral functions show in this approximation two branches E±, sep-
arated by the gap which opens along the AFM zone boundary k = kAFM
and the relevant (pseudo)gap scale is

(12)

,6.CO does not depend on t, but rather on smaller J and in particular t'.
For t' < 0 the gap is largest at (7r, 0), consistent with experiments.
Within the simplified effective band approach, it is not difficult to eval-
uate numerically ~lf beyond the QSA, by taking explicitly into account
X"(q,w), Eq. (10), with"" > 0 and w" 2J"". In Fig. 1 we present results
;v

for A(k, w = 0) at T = 0 for a broad range of "" = 0.01 - 0.6. Curves in fact
display the effective FS determined by the condition G- 1 (k F , 0) = O. At
the same time, intensities A(k, w = 0) correspond to the renormalization
factor ZF. At very small "" = 0.01 we see the hole-pocket FS which follows
from the QSA. Already small "" ;v 0.05 destroys the 'shadow' side of the
pocket. On the other hand, in the gap emerge now QP solutions with very
weak ZF « 1 which reconnect the FS into a large one. We are dealing
nevertheless with effectively truncated FS with well developed arcs. The
effect of larger "" is essentially to increase Z F in the gapped region, in
particular near (7r, 0) .
Note that within present theory the low-energy excitations corresponds
to a Fermi liquid, although a very strange one, where QP exist (at low
T) everywhere along the FS. It is quite remarkable to notice that in spite
κ0.01 κ0.04

κ0.2 κ0.6
 71

κ0.1

0_2

wit 0 1---------

(n/8,5n/8) (3n/8,7n/8)
k

Figure 2_ Contour plot of spectral functions A(k, w = 0) across the FS in the pseudogap
regime_

decreases with decreasing '"' i_e. approaching an undoped AFM, consistent

with experiments in cuprates (15, 16). We can as well define the QP DOS
N QP ex 1 dSF/v(k). It is quite important to notice a decreasing," leads
also to a decrease of N QP ' This is consistent with the observation of the
pseudogap also in the specific heat in cuprates (17), since N QP ex f. We
note here that such a behavior is not at all evident in the vicinity of a
metal-insulator transition. Namely, normally in a Fermi liquid one would
drive the metal-insulator transition by Zav ---+ 0 and within an assumption
of a local character I; (w) this would lead to v F ---+ 0 and consequently to
N QP ---+ 00.
It is also important to understand the role of finite T > O. The most
pronounced effect is on QP in the pseudogap regime. The main conclusion
is, that weak QP peak with ZF « 1 at T = 0, as seen e.g. in Fig. 2, is
not just broadened but entirely disappears (becomes incoherent) already
by very small T > T s « J (13).

4. Conclusions

We have presented our results for spectral functions and pseudogap within
the t-J model, which is the prototype model for strongly correlated elec-
trons. The physics of the t-J model at lower doping is determined by the
interplay of the magnetic exchange and the itinerant kinetic energy of fer-
mions. At intermediate doping the system is frustrated, the effect showing
up in large entropy, pronounced spin fluctuations, non-Fermi liquid effects
72

etc. Evidently, this is one path towards the metal-insulator transition, but
definitely not the only one possible. In our case, fermionic and spin de-
grees of freedom coexist but are coupled and both active and relevant for
low-energy properties.
Within the present theory (13) the origin of the pseudogap feature is in
the coupling to longitudinal spin fluctuations near the AFM wavevector Q.
It is important to note that apart from extremely small /'l, we are still dealing
with large FS. Still, at /'l, « /'l,* rv 0.5 parts of the FS in the nodal direction
remain well pronounced while the QP weight within the pseudogap (zone
corners) region of the FS are strongly suppressed. QP within the pseudogap
have small weight Z F « 1 but due to nonlocal character of ~ (k, w) not
diminished (or even enhanced) VF. This gives an explanation for a well
known theoretical challenge that approaching the magnetic insulator both
electron and QP DOS decrease and vanish.
We presented results for T = 0, however the extension to T > 0 is
straightforward. Discussing only the effect on the pseudogap, we notice that
it is mainly affected by /'l,. So we can argue that the pseudogap should be
observable for /'l,(Ch' T) < /'l,* rv 0.5. This effectively determines the crossover
temperature T*(Ch) approximately as T* rv To (l- ch/ciJ where To rv 0.6J
and ci, rv 0.15.

References

1. for a review see, e.g., M. Imada, A. Fujimori, and Y. Tokura, Rev. Mod. Phys. 70,
1039 (1998).
2. B. Batlogg et al., Physica C 235 - 240, 130 (1994).
3. J.C. Campuzano and M. Randeria et al., in Proc. of the NATO ARW on Open
Pr'oblems in Strongly Cor'related Electron Systems, Eds. J. Bonca, P. Prelovsek, A.
Ramsak, and S. Sarkar (Kluwer, Dordrecht, 2001), p. 3.
4. D.S. Marshall et al., Phys. Rev. Lett. 76,4841 (1996); H. Ding et al., Nature 382,
51 (1996).
5. M.R. Norman et al., Nature 392, 157 (1998).
6. R. Preuss, W. Hanke, C. Grober, and H.G. Evertz, Phys. Rev. Lett. 79, 1122 (1997).
7. J. Jaklic and P. Prelovsek, Phys. Rev. B 55, R7307 (1997); P. Prelovsek, J. Jaklic,
and K. Bedell, Phys. Rev. B 60, 40 (1999).
8. for a review see J. Jaklic and P. Prelovsek, Adv. Phys. 49, 1 (2000).
9. A.V. Chubukov and D.K. Morr, Phys. Rep. 288, 355 (1997).
10. .J. Schmalian, D. Pines, and B. Stojkovic, Phys. Rev. Lett. 80, 3839 (1998); Phys.
Rev. B 60, 667 (1999).
11. P. Prelovsek, Z. Phys. B 103, 363 (1997).
12. C.L. Kane, P.A. Lee, and N. Read, Phys. Rev. B 39, 6880 (1989).
13. P. Prelovsek and A. Ramsak, Phys. Rev. B 63, 180506 (2001); Phys. Rev. B 65,
174529 (2002).
14. A. Kampf and J.R. Schrieffer, Phys. Rev. B 41, 6399 (1990).
15. A. Ino et al., Phys. Rev. Lett. 81, 2124 (1998).
 73

16. Ch. Renner, B. Revaz, K. Kadowaki, 1. Maggio-Aprile, and O. Fischer, Phys. Rev.
Lett. 80, 3606 (1998).
17. J.W. Loram, K.A. Mirza, J.B. Cooper, and W.Y. Liang, Phys. Rev. Lett. 71, 1740
(1993); J.W. Loram, J.L. Luo, J.B. Cooper, W.Y. Liang, and J.L. Tallon, Physica
C 341 - 8, 831 (2000).
MOTT INSULATOR TO SUPERCONDUCTOR VIA PRESSURE
RVB THEORY & PREDICTION OF NEW SYSTEMS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

G. BASKARAN
Institute of Mathematical Sciences
Madras 600 113, India

Abstract. This paper summarizes my new work, on the nature of Mott transition in
orbitally non degenerate spin-~ system, motivated by observation of Mott insulator
superconductor transition in real systems. The structure of superexchange perturbation
theory for a repulsive Hubbard model at half filling is analyzed and it suggests, un-
der some conditions, a resonating valence bond (RVB) mechanism of Mott insulator to
superconductor transition; charge 'deconfinement' is accompanied by electron pair delo-
calization aided by a preexisting spin singlet correlations in the insulator. An RVB mean
field theory at half filling illustrating our mechanism of Mott insulator-Superconductor
transition is sketched. We identify some family of compounds as potential candidates for
Mott insulator to superconductor transition under pressure: CuO, BaBi03, thin films of
La 2 CuD4 or CaCu02 (infinite layer compound) under pressure or an effective ab-plane
pressure/epitaxial compressive stress; synthesizing new compounds such as La 2CuS202 ,
La 2CuS4 or CaCuS2 or compounds containing CUS2 or CuSe2 planes to mimic large
ab-plane chemical pressure.

Key words: High Temperature Superconductivity, Mott Insulators, Pressure Induced Su-
perconductivity, Resonating Valence Bond (RVB) Theory, Strongly Correlated Electron
Systems, Organic Superconductors, Slave Boson Theory

1. Introduction

Following Bednorz-Muller's discovery (1) of high temperature supercon-

ductivity in doped La2Cu04 in 1986, Anderson's resonating valence bond
(RVB) theory (2), brought out deep connections between the Mott insu-
lating quantum state and high T c superconductors. In RVB theory the
pre-existing singlet correlations among the electron spins in a spin- ~ Mott
insulator readily become the superconducting correlations on doping. The
RVB mean field theory (3), gauge theory (4) and the developments that

75
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 75–84.
© 2003 Kluwer Academic Publishers
76

followed (5) over nearly 15 years have given a picture that is in qualitative
agreements with a variety of experimental results.
Inspired by cuprates, RVE theory has focussed on metallization of Mott
insulating state by means of hole or electron doping. However, it is also
known that undoped commensurate systems such as the family of quasi-
1 dimensional Eechgaard salts (6, 7), quasi-2 dimensional ET salts (7),
and 3 dimensional fullerites (8, 9) do exhibit unconventional and high
T c superconductivity. For these narrow tight binding systems, a half filled
repulsive Hubbard model provides a correct starting point and various the-
ories (10, 11) show possibility of superconductivity near the Mott transition
point. Recent theoretical work by Capone and collaborators (12) bring this
out in a general context. Experimentally one does see, what may be termed
as Mott insulating spin Peierls or antiferromagnetically ordered states that
undergo a transition to a superconducting state as a function of pressure
or chemical pressure. A striking example (9) is K 3 (N H 3 )C60 , a spin-~ an-
tiferromagnet that becomes a high T c superconductor with a T c (:::::; 32 K)
under pressure.
This has prompted us to reanalyze the issue of Mott insulator metal
transition at commensurate fillings from the point of view of RVE theory.
We focus on orbitally non degenerate spin-~ Mott insulators. We argue
that strong local singlet correlations induced by higher order superexchange
processes in the vicinity of Mott transition play a key role in the Mott
insulator superconductor transition and RVE theory is the most natural
way to understand the transition and the superconducting state itself.
Our paper is organized as follows. In section 2, starting from the large
U repulsive Hubbard model at half filling we look at the higher order terms
of the superexchange perturbation theory. The divergence of the perturba-
tion theory suggests correlated singlet pair delocalization in the conducting
state. In section 3, the above argument is substantiated by an RVE mean
field theory directly applied to the repulsive Hubbard model at half filling,
which shows a Mott insulator superconductor phase transition. In section 4
we suggest that some known Mott insulating systems should become high
Tc superconductors under pressure. We also propose some new systems;
they are predicted to be potential high temperature superconductors, either
on their own or under pressure.
 77

2. Mott Insulator Superconductor transition without Doping

- a new RVB Route to high T c Superconductivity

The low energy physics of an odd electron (per unit cell) system such as
La2Cu04 is well described by a single band Hubbard model:

H = - L tijcIa·Cja + h.c. + U L nnnil (1)

ij
Here tij are the hopping matrix elements and U is the on site repulsion.
Our major focus will be on the half filled band situation. First we show
that above a critical U, there is an insulating state with a true charge gap
and then show how superconducting correlations develop as we decrease U,
leading to an eventual Mott insulator to superconductor transition.
We work in a slave boson representation which helps us to separate
charge and spin degrees of freedom (5, 4) in a manifest way: cIa == SIadi +
CTsia-el. Here the chargeons dI's and eI's are hard core bosons that create
doubly occupied sites and empty sites respectively. The fermionic spinon
operators sIcr's create singly occupied sites with a spin projection CT. The
local constraint, dI di + eI ei + 2:a sIcrsa = 1, removes redundant Hilbert
space.
In this representation Hubbard Hamiltonian takes a suggestive form:

H = - L tij(dIdj L SiaS}a + eie} L sIcrsja) + h.c. + U L(dIdi + eIei)

ij a a

Ltij(dIe} LCTSiaSja- + eidj LCTSIas}a-) + h.c. (2)

ij a a

In the limit It ij I >> U one derives an effective spin Hamiltonian for

the low energy subspace by a degenerate perturbation theory - hopping
parameter expansion, or equivalently a superexchange perturbation theory.
The lowest energy eigenstates of the interaction Hamiltonian have one elec-
tron per site and a 2N fold spin degeneracy. For large U It we can treat the
hopping, kinetic energy term (second line of equation 2) as a perturbation
and get an effective Hamiltonian that lifts the 2N fold spin degeneracy. The
effective spin Hamiltonian is generated through virtual mixing to excited
states having two, three etc. number of doubly occupied sites. The explicit
form of the effective Hamiltonian up to fourth order in tij in the single
occupancy subspace is shown to be (13):
 n
78

• • •
•

• i~1
Figure 1. A higher order superexchange process that signals co-operative delocalization
of charge ±2d singlets.

h
were t = V2
bij 1 (t8 8 t - 8 t 8jT t)·IS a spm .. smgIeteIect ron pair. creat'IOn oper-
ir j1 i1
ator at the bond ij. In the singly occupied subspace the first term becomes
the familiar superexchange term: bIjbij == -(Si . Sj - ininj)' Similarly all
the higher order terms also can be written purely in terms of the spin
operators. The prime in the summation avoids double counting.
I have written the effective Hamiltonian in terms of spinon operators,
rather than the customary Pauli spin operators, as it has some advantage
in proving a theorem and helps in seeing the physics of Mott insulator
to superconductor transition more transparently. Further a dynamically
generated local U(l) gauge symmetry of Heff, in the low energy subspace
of the Mott insulator is also manifest (4): bIj ----t eieibIjeiej.
As far as I know there is no rigorous proof for the convergence of
the above perturbation theory; such a proof should be simple as charge
localization is a rather robust phenomenon in a Mott insulator. I present
a rigorous proof for the convergence by using some of Takahashi's results
(13) for half filled band repulsive Hubbard model.
Theorem: For the repulsive half filled band Hubbard model the hopping
parameter expansion converges for U It >> l.
Proof: From Takahashi's result it is clear that (13) the nth order terms of
the spin Hamiltonian corresponds to closed graphs drawn on the lattice,
connected by n non-zero matrix elements tij; and n has to be even. A
loop may intersect and also may self overlap (figure 1). As the effective
Hamiltonian is also an operator cumulant expansion, there are no diagrams
with disconnected graphs.
It can be shown that the most general term in the expansion has the
. h
lorm, 1·t'1"2
r b b bt b bt b ffi . ti1i2ti2i:, ...tin_1intini1
~ 1·'1'"~2 ~"23
1" 1" ~4'"
3"
~'-'n-l'-'n
~ ~"n1'1' WIt a coe clent rv un 1
They are closed loops on the lattice connected by the hopping parameters
tij'S. It is also easy to see that there are prefactors of individual terms which
 79

are bounded by 2n . Since the hopping matrix elements are short ranged the
number of loops of size n is bounded by (ad)n, where a is a number of the
order of unity and d is the spatial dimension.
The matrix elements of the operator bl, b, , bl ,. b, , ... bl"n-lCJn
/Jlu2 "1"2 u2lJ,j "3"4
,b",
"11,"1
in
our single occupancy subspace is bounded by 1. Thus the maximum value
of the sum of n-th order terms is bounded by u n - 1 2n (ad)n. This term
t(max)n

clearly becomes exponentially small for If: > > 1. QED.

Mott transition is signalled by the divergence of the above expansion.
Higher order terms that become important as we approach the Mott trans-
ition point correspond to very specific physical processes, namely delo-
calization of bond singlet pairs over large closed loops, as given by an
n-th order term, bl"1"2
, b,"l{J2
, bl"2"3
, b,.,
(;.5,,4
,b"lJnul (figure 1). Such higher order
...bl"n-l"n
electron singlet pair delocalization processes establish a coherent supercon-
ducting state, if there are no other competing instabilities such as metallic
antiferromagnetism.
The spin singlet correlations existing in the Mott insulator are the ana-
logue of the preexisting neutral singlets of RVB theory for a doped Mott
insulator. Across a Mott transition induced by pressure in real systems,
the Mott-Hubbard gap does not close continuously - there is a first order
transition to a state with a finite density of free carriers. Free carriers are
produced in the ground state on top of the existing enhanced spin singlet
correlations by a kind of cooperative singlet pair delocalization process, as
described in the last paragraph. In other words, the charge delocalization is
governed and guided by the background spin singlet correlations. This ac-
tually means a transition to a superconducting state as we will demonstrate
within a mean field theory in the next section.
From weak coupling point of view, by avoiding Fermi surface nesting
we can discourage umklapp, SDW and CDW instabilities. Nesting can be
avoided by appropriate choice of the band hopping parameters tij. From
various works in RVB theory (14) it is clear that even an ordered antifer-
romagnet has sufficient local singlet correlations and is indistinguishable
from a spin liquid state as far as short range singlet correlations are con-
cerned. Thus the ordered antiferromagnetic state of a Mott insulator has
sufficient local spin singlet correlations to be able to induce electron pair
delocalization and hence superconductivity in a Mott transition.

3. RVB mean field theory for Mott insulator to Superconductor

Transition under Pressure: a direct analysis of Hubbard Model

As we mentioned before, early works on Mott transition focussed more on

the charge degree of freedoms. In the present section we summarize the
results of our recent work, where we give equal importance to spin and
80

charge degree of freedom. We find that this leads to the possibility of Mott
insulator to superconductor transition (without doping) as a function of
pressure, in a natural way as superconductivity has been argued to arise in
a Mott insulator on doping. From our work, which analysis the Hubbard
model directly, we also reproduce known RVB mean field theory results
(3, 5) for Heisenberg and t-J models as interesting and non-trivial limiting
cases.
Superexchange processes in the Mott insulator is kinetic in origin; second
term of equation 2, a part of the kinetic energy term of the Hubbard model,
generates superexchange process. This enables us to perform a Hartree-
Fock type of factorization of the kinetic energy term of the Hubbard model
(equation 2),

eidj L asL-s}o- ---t 6.:j ei dj + rJij L asL-s}o- etc. (4)

a a

involving singlet spinon pair amplitude, 6.:j == (La asIas}o-) , Xij == (La sIasja) ,
and charge amplitudes rlij == (eidj),eij == (eIe) and dij == (dId j ).
The resulting slave boson mean field Hamiltonian has a charge part:

ij ij

and a spin part

In the slave boson mean field theory we diagonalize the above Hamilto-
nian and determine the Hartree-Fock parameters by minimizing the total
energyjfree energy with the global constraint that total number of spinons
eI
and chargeons Li (dI di + e i + La SIaSia) = N, number of lattice sites.
To establish the principle of an RVB mechanism of superconductivity
without doping at a Mott transition point we quote three general results
from our mean field theory: i) Hch , the boson Hamiltonian has a charge gap
E c which vanishes at aU = Uc , as we decrease U from a large value - this is
the desired Mott transition, ii) the RVB mean field transition temperature
Tc(RV B) starts from a value zero at U = 00 and increases continuously as
we decrease U and iii) the chargeons develop anomalous expectation value
(e i ) = (d i ) i- 0 below Uc with a Tc(BE) i- O.
To see the possibility of superconductivity we express the Cooper pair
order parameter in terms of the slave boson variables in the conducting
phase: (La aCiaCjo-) rv (eI) (e})6.ij + (ei) (d})rJij + otherterms. As (ei), (d i )
 81

(a) repulsive hubbard model Mott insulator

at haljjilling ( b ) (AFM order)
TdBE)
U
(e), (d) charge Ec
T g~

l~~.?(~~~ t super

o u U C O n ________

Figure 2. (a) T-U Phase diagram (Schematic). Tc(BE) is the chargeon Bose condens-
ation temperature, T c (RVB) is the mean field transition temperature of the RVB order
parameter and E c is the charge gap in the Mott insulating state. (b) Schematic Ground
state phase diagram in the U-n plane

are nonzero in the conducting phase and 6.ij and TJij continue to remain
finite, we have superconductivity within the above mean field theory.
Physically we can understand the above as follows. Coherent electron
singlet pair delocalization naturally leads to a finite density of doubly oc-
cupied and empty sites present in a phase coherent fashion. In terms of
the slave boson picture, presence of real empty and doubly occupied sites
in the ground state in a phase coherent fashion corresponds to the bose
condensation of chargeons.
Figure (2a) illustrates the region where superconductivity is likely to
occur in the T vs U plane. Figure (2b) sketches the ubiquitous supercon-
ducting region in the U vs n plane.
Slave boson mean field as applied directly to the Hubbard model in the
present paper has to be understood carefully. First, as in the RVB mean
field theory applied to Heisenberg Hamiltonian, we get electron pair con-
densation in the Mott insulating state. This has to be properly interpreted,
in view of the presence of charge gap present even in our mean field theory.
Use of gauge fluctuations eliminates this fictitious superconducting order in
the Mott insulating state, consistent with Elitzur's theorem. Secondly, our
mean field theory produces superconductivity even as U ---7 O. This is an
artifact of Hartree-Fock factorization. However, we believe that our results
are qualitatively correct close to the Mott transition point. As pointed
out earlier, nesting instabilities in the conducting phase can be avoided by
appropriate choice of tij'S, the band parameters.
In this paper we do not go into the detailed energetics of mean field
solutions corresponding to different symmetries, for example extended-s
and d symmetry of the spinon pair order parameter 6.ij and the charge
pair order parameter rlij.
82

4. Predicting New Superconductors and the Scale of T c

A practical problem with the 'doping of Mott insulator', an RVB route to

high T c superconductivity is the chemical complications/difficulty in doping
Mott insulators. The pressure route does not have this complication.
CuO is the mother compound of the cuprate high T c family. CU02
ribbons form a three dimensional network, each oxygen is shared by two
ribbons mutually perpendicular to each other. The strong square planar
character from the four oxygens surrounding a Cu isolates out one non
degenerate valence d-orbital with one valence electron. This makes CuO
an orbitally non-degenerate spin-~ Mott insulator, a potential candidate
for our pressure route to RVB high T c superconductivity. The frustrated
(rv 1200 ) superexchange leads to a complex three dimensional magnetic
order with a Neel temperature rv 200K. These frustrations should help in
stabilizing short range singlet correlations, which will help in singlet Cooper
pair delocalization on metallization.
The CU02 ribbons give CuO a character of coupled 1d chains. This
makes it some what similar to quasi-one dimensional Bechgaard salts, which
has a Mott insulator to superconductor transition, via an intermediate
metallic antiferromagnetic state as a function of physical or chemical pres-
sure. The intermediate metallic antiferromagnetic state represents a suc-
cessful competition from nesting instabilities of flat Fermi surfaces arising
from the quasi-one dimensional character. Once the quasi-one dimensional
character is reduced by pressure nesting of Fermi surface is also reduced
and the RVB superconductivity takes over.
If other oxides in perovskite family are any guidance, the metallization
should take place in tens of CPa's. Unless a crystallographic transition
intervenes and changes the valence electron physics drastically, CuO should
undergo a Mott insulator superconductor transition, perhaps with an inter-
mediate antiferromagnetic metallic state. The superconducting T c should
be a finite fraction of the Neel temperature, as is the case with Bechgaard
salts or K 3 (N H 3 )C60 . Thus an optimistic estimate of T c will be 50 to 100
K.
Similar statements can be made of the more familiar La2Cu04 , insu-
lating YBCO and the CaCu02, the infinite layer compound or the family
of Mott insulating cuprates such as Hg and Tl based insulating cuprates.
Infinite layer compound has the advantage of absence of apical oxygen and
should be less prone to serious structural modifications in the pressure
range of interest to us. The quasi-2d Hubbard model describing the CU02
planes does have an appreciable t', making nesting magnetic instabilities
weaker. Thus we expect that on metallization a superconducting state to
be stabilized with a small or no antiferromagnetic metallic intermediate
 83

state. The superconducting T c could be fairly high rv 200K, as the intrinsic

antiferromagnetism scale of the CU02 layer is very high rv 500K. The
quasi-2d cuprates have a special advantage in the sense we may selectively
apply ab-plane pressure in thin films by epitaxial mismatch and ab plane
compression.
One way of applying chemical pressure in cuprates is to increase the
effective electron band width by increasing the band parameters such as t
and t' in the Hubbard model. This can be achieved by replacing oxygens in
the CU02 planes by either sulfur or selenium, which because of the larger
size of the bridging 3p or 4p orbitals increase the band width and at the
same time should reduce the charge transfer or Mott-Hubbard gap. So if
nature allows us to make a complete or partial replacement of oxygen in
the CU02 planes one might achieve metalization/superconductivity without
doping. Some possible compounds are La2CuS202 , La2CuS4 and CaCuS2
or their Se versions.
Superconductivity in doped BaBi03 was one of the sources of inspiration
for the RVB theory of Anderson (2). On closer look this system reveals itself
as a Hubbard model with repulsive interaction (15). High pressure is likely
to induce a high T c superconducting state without any external doping.

Acknowledgements

I thank Erio Tosatti for bringing to my attention Mott insulator supercon-

ductor transition under pressure in (N H 3 )K3 C 60 family.

References

1. A. Bednorz and A. Muller, Z. Phys. B 64 189 (1986)

2. P.W. Anderson, Science 235 1196 (1987); The theory of high temperature super-
conductivity (Princeton University Press, NY, 1996); P. W. Anderson, G. Baskaran,
T. Hsu and Z. Zou, Phys. Rev. B 582790 (1987)
3. G. Baskaran, Z. Zou and P.W. Anderson, Sol. State Commm. 63973 (1987)
4. G. Baskaran and P.W. Anderson, Phys. Rev. B37 580 (1988); G. Baskaran, Physica
Scripta, T27 53 (1989);
5. S. Kivelson, J. Sethna and D. Rokhsar, Phys. Rev. B 38 8865 (1987); G. Kotliar,
Phys. Rev. B 373664 (1988); 1. Affleck and J. B. Marston, Phys. Rev. B 373774
(1988); S. Sorella et al., Phys. Rev. Lett. 88 117002 (2002); A. Paramekanti, Mohit
Randeria and Nandini Trivedi, Phys. Rev. Lett. 87, 217002 (2002)
6. D. Jerome, Science 252 1509 (1991); C. Bourbonnais, D. Jerome, in Advances in
Synthetic Metals (Elsevier, 1999) page 206, Eds. B. Bernier, S. Lefran and G. Bidan;
T. Vuletic et al., E. Phys. J B25 319 (2002)
7. T. Ishiguro, K. Yamaji and G. Saito, Organic Superconductor (Springer, Berlin
1998)
8. A. Ramirez, Superconductivity Rev. 1 1 (1994)
84

9. O. Zhou et al., Phys. Rev. B 52 483 (1995)

10. D. Jerome and H.J. Schulz, Adv. Phys. 31 229 (1982);also reprinted in Adv. Phys.
51 293 (2002); L. N. Bulaevski, Adv. Phys. 37443 (1988)
11. H. Kino and H. Fukuyama, J. Phys. Soc. Jap. 642726 (1995); K. Kanoda, Physica
C282 820 (1997); R. H. Mackenzie, Science 278 820 (1997); J. Schmalian, Phys.
Rev. Lett. 81 42:32 (1998) K. Kuroki and H. Aoki, Phys. Rev. B60 3060 (1999)
12. M. Capone, M. Fabrizio, C. Castellani and E. Tosatti, Science 296 2364 (2002)
1:3. Minoru Takahashi, J. Phys. C 10 1289 (1977) used Kato's method, T. Kato, Prog.
Theor. Phys. 4 514 (1949)
14. T. Hsu, Phys. Rev. B 41 11:379 (1990); G. Baskaran, Phys. Rev. B 64 092508
(2001)
15. G. Baskaran and P.W. Anderson (1987, unpublished)
EXPLORING THE PHASE DIAGRAM OF THE QUASI-2D
ORGANIC SUPERCONDUCTORS K, -(BEDT-TTFhX

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

MICHAEL LANG
Johann Wolfgang Goethe- University, FOR412
D-60054 Frankfurt am Main, Germany
JENS MULLER and FRANK STEGLICH
Max-Planck Institute for Chemical Physics of Solids
D-01187 Dresden, Germany
JOHN SCHLUETER
Material Science Division, Argonne National Laboratory,
Argonne, Illinois 60439, U.S.A.
TAKAHIKO SASAKI
Institute for Materials Research, Tohoku University,
Sendai, Japan

Abstract. By means of high-resolution thermal expansion and specific heat measure-

ments on the title compounds, we explore various aspects of their unified phase diagram.
Particular attention has been paid to the interrelation of high-temperature anomalies,
i.e. a glass transition associated with the ethylene end groups at T g c::::: 50 - 80 K and a
density-wave-type transition at T* c::::: 40 K, with the superconducting state.

Key words: Organic Superconductors, Thermodynamic Properties, Thermal Properties

1. Introduction

The organic charge-transfer salts of the (BEDT-TTF)2X family (BEDT-

TTF or simply ET denotes bis(ethylenedithio)-tetrathiafulvalene and X a
monovalent anion) represent fascinating model systems for exploring the
interplay of strong electron-electron and electron-phonon interactions in
reduced dimensions. Of particular interest are the K,-phase (EThX salts
with various monovalent anions such as X = Cu(NCS)2, Cu[N(CN)2]Br

85
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 85–93.
© 2003 Kluwer Academic Publishers
86

and Cu[N(CNh]Cl. They exhibit unusual normal (N)- and superconducting

(SC)-state properties which resemble in some aspects those of the layered
high-T c cuprates. Apparent similarities include the quasi-two dimensional
(quasi-2D) electronic structure and the proximity to an antiferromagnetic-
ally (AFM) ordered state, d. Fig. 5. Moreover, the presence of strong AFM
spin fluctuations above T c as inferred from various magnetic and thermal
properties have been associated with a magnetically mediated non-s-wave
type of SC not only for the cuprates but also for the present organic systems
(1). For the latter, however, such a scenario is at variance with the results
of numerous other studies, see, e.g. (2) and references cited therein.
In this paper we summarize our recent results of a detailed thermodynamic
study employing thermal expansion and specific heat measurements on the
above K:-type salts. Particular attention is paid to the anomalous properties
at elevated temperatures and their interrelation to the SC state. A full
account of the work is given in (2, 3).

2. Classification of anomalies and discussion

Figure 1 compiles the results of the coefficient of thermal expansion meas-

ured perpendicular to the conducting planes, a~, for the various K:-(EThX
salts investigated: X = CU(NCS)2 (in short K:-Cu(NCSh), Cu[N(CNh]Br
(K:-Br) and Cu[N(CN)2]Cl (K:-Cl). For all compounds, a variety of anomalies
can be discerned and classified as (A) step-like changes in a at temperatures
T g around 50 - 80 K, (B) phase-transition-like anomalies at an intermediate
temperature T* around 40 K only for the SC salts and (C) phase-transition
anomalies into the SC (K:-Cu(NCSh, K:-Hs-Br) and AFM (K:-Cu[N(CNh]Cl)
ground states. In what follows, we will discuss all three kinds of anomalies
and their implications for the SC state.
(A) Glass-like transitions at T q c::,: 50 - 80 K
Figure 2 shows on expanded scales the anomalies in a~ for all three com-
pounds near T g . Both the distinct hysteresis between heating and cooling
curves as well as their characteristic cooling-rate dependence (not shown)
are clear manifestations of the glassy nature of these anomalies (3). It is
well known that in positionally and/or orientationally disordered systems,
relaxation processes can lead to glass-like transitions where below the glass
temperature T g , a short-range order characteristic for this temperature
becomes frozen, see (3) and references cited therein.
As discussed in detail in ref. (3), the relevant entities involved in the
freezing process for the present K:-(ET)2X salts are the terminal ethylene
groups [(CH)2] of the ET molecules. From these results together with earlier
x-ray-diffraction studies (4), we infer that upon cooling from room temper-
ature, the gradual ordering in the relative orientation of the ethylene groups
 87

T (K)
o 20 40 60 80 100 120 140

80 • K-Cl
o K-Ds-Br 20
60
80
60
15 50
~
60
~ 40
10 40
''b
-
'-'
1:t 20 40 5 30
T(K)
26 28 30 32

r- o T 20

·,.r"·
o TN

L. . L- - '- -'- -'a- - '- -'-~_ '_,: -5 b D K-Cu(NCS)2

10

• K-Hs-Br
-20 - o
-10 I-.t-,----'-----,--"--,---'---r---r---.----r---!
o 40 60 80 100 120
T(K)
Figure 1. Cross-plane thermal expansion coefficient for various K;-(EThX salts; H s (D s )
indicate protonated (deuterated) ethylene endgroups. Inset shows the phase transition
into the AFM ground state for K;-Cl.

in one of two possible conformations becomes interrupted at T g leaving

a residual frozen disorder at low temperatures. The amount of disorder
depends on the cooling rate employed at T g and is expected to be at a
10 - 20 % level. Its implications for the ground-state properties can be
discerned by recalling the results of magnetic and transport measurements
on protonated (H s ) and deuterated (D s ) Ii:-Br: rapid cooling through a
temperature around 80 K (c:::' T g ) was found to cause a strong reduction of
the SC volume fraction and may even transform a SC into an insulator (5).
(B) Phase-transition anomalies at T* c:::' 40 K
For the SC salts, i.e. Ii:-Hs-Br and Ii:-Cu(NCSh, the Cl:.l data (Fig. 1b)
reveal distinct local maxima at T* = 38 and 45 K, respectively. In addi-
tion, directional-dependent measurements (not shown) disclose a strongly
anisotropic lattice response at T*. In an attempt to extract the anomalous
contributions to the uniaxial coefficients of thermal expansion Cl:i at T*, b"Cl:i,
we use a smooth interpolation of the data outside of the Cl:i anomalies. The
so-derived Cl:bi serve as rough estimates of the background expansivities.
In Fig. 3 we display the quantities b"Cl:i(T) = Cl:i(T) - Cl:ib(T) over an
extended temperature range which also covers the SC transition. We find
pronounced 2nd-order phase-transitions-like anomalies at slightly different
temperatures T* for both compounds. Replacing the somewhat broadened
features by idealized sharp jumps employing an "equal-areas" construction
88

70 K-(ETj,Cll[N(CN),lCI 70 K-(ET),Cu[N(CN)JH, K-(ET),Cll(NCS),

tY
65 t
+

r
65
i #'
60
JI
o.j
.~ 55

·~60

j II
50
ij' ~ 20
~
iI
D
I
45
55
18

a 35 b 16 c

55 60 65 70 75 80 85 60 65 70 75 80 85 90 40 45 50 55 60 65 70 75
T~ T~ T~

Figure 2. Blow-up of the cross-plane thermal expansion coefficient around the glass-like
transition for various K;-(EThX salts. Arrows indicate cooling and heating runs.

in an alT vs T plot yields transition temperatures T* = (41 ± 2) K and

(47 ± 2) K for the K;-Hs-Br and K;-Cu(NCS)2 salts, respectively. We note
that the overall shape of these anomalies, i.e. their width and asymmetry,
but not the peak itself, depends somewhat on the interpolation procedure
used to determine the background expansivities. An important piece of
information enclosed in Fig. 3 is a striking interrelation between the a(T)
anomalies at T* and those at T e for both compounds: for each axis, a large
(small) feature at T* is accompanied by a large (small) discontinuity at
T e , while the signs of both anomalies are just reversed! According to the
Ehrenfest relation

the uniaxial-pressure dependence of a 2nd-order phase transition at To

is related to the discontinuities at To in the coefficient of thermal expansion,
~ai, and specific heat ~C. Thus, the data of Fig. 3 imply that for both
compounds the uniaxial-pressure coefficients of T* and those of T e are
strictly anticorrelated in their signs but correlated in their magnitudes.
In the case of the K;-Cu(NCSh salt, for example, we infer that uniaxial
pressure applied perpendicular to the planes causes a substantial shift of
T* to higher temperatures and, at the same time, a strong reduction of T e .
Likewise, uniaxial pressure along the in-plane b-axis neither affects T* nor
does it have any effect on T e . The anomaly at T* is particularly interesting
because it coincides with the temperature where magnetic, transport, and
elastic properties exhibit anomalous behavior: a sharp peak in the spin-
lattice relaxation rate (T1T)-1 (7,6) accompanied by a reduction of the spin
susceptibility as determined by Knight-shift (7) or ESR (8) measurements
as well as pronounced softening of ultrasonic modes (9, 10). In addition,
 89

,...
1'. K-Cu(NCS),

•• c-axis
.- --
.....
b-axis

~\
x
~
c-axis jT,
~-

a -axis U- layers) rP I \
1T .
o 10 20 30 40 50 60 0 10 20 30 40 50 60

T (K) T (K)
Figure 3. Anomalous contributions to the uniaxial thermal expansion coefficients at T c
and T* derived by subtracting a smooth background contribution.

our analysis yields for both compounds an increase of T* under hydrostatic

pressure conditions, in accordance with the results of pressure studies (7, 9).

Concerning the nature of the anomaly at T* we note that it is inde-

pendent of the crystal structure as K:-Cu(NCSh is not isomorphic to K:-Br
but it appears to be related to the electronic structure which is very similar
for the two SC compounds (4).
Based on the above results, we propose that T* marks a density-wave
instability that involves only the minor 1D parts of the Fermi surface while
leaving the major 2D portions unaffected. This implies that cooling through
T* is accompanied by the formation of a real gap on a small fraction of
the Fermi surface as opposed to a pseudogap on the major parts (1). The
above interpretation is consistent with recent results of both magnetic and
transport measurements on the SC salts yielding the onset of a small but
distinct anisotropy below T* (11).
(C) Transitions into the magnetic and superconducting ground-
states
The phase transitions into either the AFM ordered (K:-Cl),(cf. Fig.1a) or
SC (K:-Hs-Br, K:-Cu(NCSh), (cf. Figs. 1 and 3) ground state manifest them-
selves in distinct and strongly anisotropic phase-transition anomalies in
O:i. As for the transition at TN, our directional-dependent measurements
(not shown) reveal a finite discontinuity only along the axis perpendicular
to the conducting planes (6.0:..l < 0), cf. inset Fig. la, while 6.0:11 ~ 0
for both in-plane expansion coefficients (3). Applying the Ehrenfest re-
lation, we thus conclude that uniaxial pressure applied perpendicular to
the conducting planes causes a reduction of TN whereas in-plane pressure
90

leaves TN unaffected. Taking these uniaxial-pressure coefficients together

results in a negative pressure dependence of TN under hydrostatic-pressure
conditions in accordance with the results of pressure studies (12). As for the
nature of the AFM state, our results seem difficult to reconcile with existing
models. For neither the nesting-induced itinerant magnetism (13, 14), nor
for a Mott-Hubbard-type magnetic state of localized spins (15, 1) - even
when next nearest neighbor couplings are included - is a vanishingly small
in-plane pressure coefficient expected.
Applying the same thermodynamic analysis to the directional-dependent
(li(T) anomalies at the SC transition for the K:-(EThX series clearly demon-
strates the lack of a simple form of systematics regarding the in-plane
pressure coefficients of T c : While for the K:-Hs-Br salt both in-plane pressure
coefficients of T c are negative, an either zero or even positive in-plane pres-
sure effect is found for K:-CU(NCS)2 (16, 3). Such a non-universal behavior
is in clear contrast to the predictions of the purely 2D electronic models
proposed by Kino et al. (17) and Kondo et al. (18) yielding in all cases a
suppression of T c under in-plane stress. On the contrary, our studies reveal
that it is the extraordinarily large negative inter-layer pressure coefficient of
T c which is common to all K:-(EThX systems investigated so far including
the X = 13 salt (16). Apparently, it is this component which predominates
the reduction of T c under hydrostatic pressure.
As for the symmetry of the SC state, arguments in favor of an uncon-
ventional order parameter with d-wave symmetry for the present K:-(ET)2X
family have been derived from temperature dependent studies and, more
recently, from orientational-dependent measurements aiming at a direct
determination of the gap anisotropy. Conversely, there are numerous exper-
imental investigations which indicate a SC state characterized by a finite
gap. For a discussion on the present status of this controversy, see e.g. (2).
A decisive technique to probe certain aspects of the gap structure -
in particular, the question whether gap zeroes exist or not - is provided
by specific heat measurements. In case this integral thermodynamic probe
were to detect an electronic quasiparticle contribution Ces(T) for T « T c
that varies exponentially weakly with the temperature, the existence of gap
zeroes on parts of the Fermi surface could be definitely ruled out. In Fig.
4 we display the quantity tlC(T) = C(T, B = 0) - C(T, B = 8T), where
C(T,8T) is the specific heat in the N-state, measured on a tiny high-quality
single crystal of K:-Cu(NCSh with a mass of only 0.72 mg.
Given a B-independent specific heat above T c as proved experimentally
(19,20), the analysis of tlC(T) has the advantage that the unknown phonon
contribution and all other extraneous contributions cancel each other out.
Figure 4 demonstrates that tlC(T) deviates markedly from a weak-coupling
BCS-behavior (broken line) but is in excellent agreement with the so-called
 91

O.72mg
0.8
Tc =9.15K

0.6

Q 00

"0 0.4 0.01

S
.....,
---- 1.0 1.5 2.0 2.5
'-'
TiT
U 0.2
<:l - - - - - fJ. ~ 1.764
--fJ.~2.8
0

o 2 4 6 8 10 12

T(K)
Figure 4. Specific heat difference !:::.C = C(OT) - C(8T) = C(OT) - Cn, where C n
denotes the N-state specific heat. Dotted and solid thick lines represent BCS curves
for weak and strong coupling, respectively. Inset shows the electronic contribution in a
BCS-plot.

a-model, an empirical extension of the BCS formalism to strong-coupling

superconductors (21). As the inset of Fig. 4 clearly shows, C es = I:1C +
IT, where I is the Sommerfeld coefficient, reveals an exponentially weak
temperature dependence at low temperatures incompatible with ad-wave
order parameter, d. inset of Fig. 4. The same conclusions have been drawn
recently from similar measurements on the K;-Hs-Br salt (20).

3. Summary and outlook

The present study discloses important new aspects which impose clear
constraints for both experimentalists and theorists attempting to unravel
the nature of the states below and above T c in these K;-(EThX salts. At
elevated temperatures, a glass transition has been identified that defines
the boundary between an ethylene liquid at T > T q and a glassy state
at T < T g where a certain amount of disorder in the terminal [CH 2 b
units becomes frozen. With increasing cooling rate through T g , the level of
disorder (10 - 20 %) increases and may substantially influence the electronic
properties at low temperatures, especially the SC state.
At intermediate temperatures T*, our studies disclose a second-order
phase transition for the metallicjSC salts. We propose that instead of a
pseudogap on the major 2D fractions of the Fermi surface, T* is associated
with the formation of a density wave, i.e. a real gap, on the minor 1D parts
which competes with superconductivity for stability.
92

ethylene liquid

100

g T~ I PMIDW I
E-< 10 1--~

I '

X~
Cu[N(CN),]Cltt Ds Hs
- Cu[N(CN)2]Br CU(NCS)2 I kbar

Figure 5. Compilation of thermal expansion anomalies for the various /'i;-(EThX salts in
a pressure-temperature phase diagram. Arrows indicate the location of the different salts
at ambient pressure. AFI, PM/DW and SC denotes antiferromagnetic insulator, para-
magnetic metal in coexistence with a density-wave state and superconductor, respectively.
Solid lines represent the hydrostatic pressure dependencies of TN and T c taken from the
literature.

As for the response of T c to pressure, our finding of an extraordinarily

large interplane pressure effect but a non-universal behavior in respect to
the in-plane pressure coefficients demonstrates that attempts to understand
these quasi-2D organic superconductors solely on the basis of 2D electronic
models are not appropriate. It is a combination of electronic correlations,
electron phonon-interactions as well as interlayer-coupling effects which
have to be considered in order to provide a realistic modelling of this family
of superconductors.

References

1. A. Kanoda, Hyperfine Interact. 104, 235 (1997)

2. J. Muller, M. Lang, R. Helfrich, F. Steglich, T. Sasaki, Phys. Rev. B 65 140509
(2002)
:3. J. Muller, M. Lang, F. Steglich, J. A. Schlueter, A. M. Kini, T. Sasaki, Phys. Rev.
B 65, 144521 (2002)
4. A.M. Kini, U. Geiser, H. H. Wang, K. D. Carlson, J. M. Williams, W. K. Kwok,
K.G. Vandervoort, J. M. Thompson, D. L. Stupka, D. Jung, M.-H. Whangbo, Inorg.
Chern. 29, 2555 (1990)
5. A. Kawamoto, K. Miyagawa, K. Kanoda, Phys. Rev. B 55, 14 140 (1997)
6. A. Kawamoto, K. Miyagawa, Y. Nakazawa, K. Kanoda, Phys. Rev. Lett. 74, 3455
(1995)
 93

7. H. Mayaffre, P. Wzietek, C. Lenoir, D. Jerome, P. Batail, Europhys. Lett. 28, 205

(1994)
8. V. Kataev, G. Winkel, D. Khomskii, D. Wohlleben, W. Crump, K.F. Tebbe, J.
Hahn, Solid State Commun. 83, 435 (1992)
9. K. Frikach, M. Poirier, M. Castonguay, K.D. Truong, Phys. Rev. B 61, R6491 (2000)
10. T. Simizu, N. Yoshimoto, M. Nakamura, M. Yoshizawa, Physica B 281&282, 896
(2000)
11. T. Sasaki, N. Yoneyama, A. Matsuyama, N. Kobayashi, Phys. Rev. B 65, 060505(R)
(2002)
12. S. Lefebvre, P. Wzietek, S. Brown, C. Bourbonnais, D. Jerome, C. Meziere, M.
Fourmigue, P. Batail, Phys. Rev. Lett. 85, 5420 (2000)
13. P. Wzietek, H. Mayaffre, D. Jerome, S. Brazovskii, J. Phys. I 6, 2011 (1996)
14. M.A. Tanatar, T. Ishiguro, H. Ito, M. Kubota, G. Saito, Phys. Rev. B. 55, 12 529
(1997)
15. K. Miyagawa, A. Kawamoto, Y. Nakazawa, K. Kanoda, Phys. Rev. Lett. 75, 1174
(1995)
16. J. Miiller, M. Lang, F. Steglich, J. A. Schlueter, A. M. Kini, U. Geiser,
J. Mohtasham, R. W. Winter, G. L. Gard, T. Sasaki, N. Toyota, Phys. Rev. B 61,
11739 (2000)
17. H. Kino and H. Fukuyama, J. Phys. Soc. Jpn. 65, 2158 (1996)
18. H. Kondo and T. Moriya, J. Phys. Soc. Jpn. 67, 3695 (1998)
19. B. Andraka, J. S. Kim, G. R. Stewart, K. D. Carlson, H. H. Wang, J. M. Williams,
Phys. Rev. B 40, 11 345 (1989)
20. H. EIsinger, J. Wosnitza, S. Wanka, J. Hagel, D. Schweitzer, W. Strunz, Phys. Rev.
Lett. 84, 6098 (2000)
21. H. Padamsee, J. E. Neighbor, C. A. Shifman, J. Low Temp. Phys. 12, 387 (1973)
ENHANCEMENT OF SUPERCONDUCTIVITY BY STRONG
CORRELATIONS: A MODEL STUDY

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

MICHELE FABRIZIO, ERIO TOSATTI

International School for Advanced Studies (SISSA-ISAS) and
Istituto Nazionale per la Fisica della Materia (INFM) ,
Via Beirut 2, 34014 Trieste, Italy
and International Center for Theoretical Physics (ICTP),
Strada Costiera 11, 34014 Trieste, Italy
MASSIMO CAPONE, CLAUDIO CASTELLANI
Universita di Roma "La Sapienza" and
Istituto Nazionale per la Fisica della Materia (INFM) ,
Piazzale A. Moro 2, 00185 Roma, Italy.

Abstract.
High-temperature superconductivity in doped Mott insulator cuprates denies the
conventional belief that electron repulsion is detrimental to superconductivity. We discuss
a parallel situation in the alkali fullerides which, even if lacking the cuprate's spectacular
Te's, are no lesser puzzle. The repulsive electron correlations in fullerides are of com-
parable strength to the cuprates: yet trivalent fullerides superconduct, and that with an
s-wave order parameter. We discuss theoretical studies of a model for alkali fullerides
which reveals, in presence of a weak pairing attraction (here of Jahn Teller origin), an
unsuspected superconducting pocket near the Mott insulating state. The peculiar pairing
mechanism in this model shows that close to an ideal, singlet Mott state, superconducting
pairing can manage not just to avoid frustration by strong electron-electron repulsion,
but actually to take vast advantage from it. Although derived by solving a very specific
tetravalent fulleride model, we argue that this mechanism is in fact common to a wider
class of strongly correlated models, including those proposed for cuprates (1).

Key words: High Temperature Superconductivity, Strongly Correlated Systems, Mott

Transition, Fullerenes

95
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 95–113.
© 2003 Kluwer Academic Publishers
96

1. Introduction

Among the unsolved questions raised by the discovery of high T c supercon-

ductivity in cuprates, a challenging one is why these good superconductors
are such bad metals in the normal phase. It is in fact conjectured in many
quarters that the main reason for the poor metallic behavior, namely the
strong electron-electron correlation, is also the key ingredient to understand
the high T c' That conjecture, even if supported by experimental evidence,
is somehow heretical with respect to the Migdal-Eliashberg theory for con-
ventional superconductors, consisting of phonon mediated attraction acting
among Landau quasiparticles.
Within Landau Fermi liquid theory, strong repulsive correlations between
electrons have two main effects on quasiparticles: first of all they increase
the effective mass, m* > m, and secondly they reduce the quasiparticle
weight, Z < 1, of the single particle excitations. The mass enhancement in-
creases the quasiparticle density of states (DOS) at the chemical potential,
p*, with respect to the uncorrelated one, p, according to
m*
p* = -po
m
On the other hand, the quasiparticle attraction, parametrized for instance
by the coupling constant ~, is renormalized with respect to the bare value
Vas
V* = z2rvv,
rv being the vertex correction. In the standard situation of an attraction
provided by coupling to phonons, vertex corrections are small, rv ': : :'
1.
Therefore the dimensionless parameter A*, which controls the strength of
the attraction, is renormalized down with respect to the uncorrelated value
A = pV according to
m m
A* *
= p*V* ':::::' Z 2 -pV * < A,
= Z 2 -mA
m
since usually Z ::; mjm*. It seems therefore hardly conceivable within
Landau Fermi liquid theory that strong correlations might help supercon-
ductivity, except in the fortunate situation where vertex corrections, rather
than being negligible, would be large. For instance, if vertex corrections
were to compensate for the wave-function renormalization, say rv rv 1jZ2,

then
m*
A* ':::::' -A> A,
m
which might indeed produce a T c enhancement.
This hypothetical possibility is based on the assumption that Landau
Fermi liquid theory is valid well below T c . However, when the mass en-
hancement pushes A* rv 1, T c becomes of the order of the quasiparticle
 97

bandwidth and a description of the excitations above T c in terms of Landau

quasiparticles may not be correct. This situation is compatible with the
behavior of the cuprates. In fact, while their normal phase shows anom-
alous non-Fermi-liquid properties, there are spectroscopic evidences that
single particle excitations well below T c can still be represented by Landau-
Bogoliubov quasiparticles (2).
There are models for cuprates where the above hypothetical scenario
seems to be realized. For instance the t-J model, where strong repulsive
correlations cause the single occupancy constraint, and yet superconduct-
ivity does arise. The on-site constraint strongly suppresses the coherent
electron motion, without affecting the pairing mechanism, because the lat-
ter is provided by the intersite spin exchange J. So even if the normal
phase of the t-J model will show some exotic properties, for instance spin-
charge separation within the RVB phase and similar features (3), the
superconducting phase seems still describable in terms of quasiparticles
(spinon-holon bound states) subject to a strong attraction J which causes
them to pair off, with a loss of kinetic energy given by a coherent bandwidth
of similar magnitude (4, 5, 6).
Here we discuss another class of superconductors, the alkali doped fullerenes,
or fullerides. While these compounds lack the spectacular Tc's of the cuprates,
they nonetheless appear to share some aspects with them. The fullerides
are molecular conductors with a narrow bare bandwidth of about W C"-' 0.5
eV (for a broad review see Ref. (7)). The relative strength of correla-
tions is comparable with that of cuprates, with a large intramolecular
Coulomb repulsion U rv 2/3W (8, 9), which undoubtedly makes them
strongly correlated superconductors, like the cuprates. Unlike the cuprates,
which condense with a d-wave order parameter so as to minimize on-site
Coulomb repulsion, fullerides are s-wave superconductors. Here an efficient
intramolecular pairing mechanism and the orbital degeneracy, are evidently
able to induce superconductivity avoiding the relatively strong U. Histor-
ically, because of a consensus that the pairing mechanism is a conventional
electron-phonon coupling, they never attracted quite the same attention
as the cuprates. We actually find instead that this circumstance is an
advantage when one wants to understand the truly revolutionary effects
of extreme correlations on superconductivity.
C 60 molecules possess a threefold degenerate tlu LUMO which forms
in a crystal lattice three narrow bands. These bands are occupied by the
n valence electrons provided by the alkali ions M in M n C 60 . According to
band structure calculations, alkali doped fullerenes should be metallic for
any 0 < n < 6. On the contrary however, only M 3 C 60 are metallic and
superconducting, while for instance M 4 C 60 are non magnetic narrow gap
insulators. The phonons which are believed to be mostly involved in super-
98

conductivity of M 3 C 6o are the eigth fivefold degenerate molecular vibrations

of H g symmetry, Jahn-Teller coupled to the tlu orbitals (10). The supercon-
ducting critical temperature in cubic fullerenes varies from 18 K in K 3 C 60
to 28 K in Rb 3 C 6o , and appears to be a monotonically increasing function
of the lattice spacing a. Because this grows parallel to the density of states
(11), this increase is qualitatively consistent with standard BCS theory,
which is therefore often used to discuss superconductivity in fullerides in
spite of very strong correlations (7). In some trivalent compounds however
the behavior of T c versus the lattice spacing a is opposite. CS 3 C 60 , a bct
compound with a larger unit cell volume per C 60 than either K 3 C 60 and
Rb 3 C 6o , shows an increasing T c with pressure, reaching at 15 Kbar the
highest T c rv 40 K among all fullerides (12). Unless the difference between
cubic and noncubic lattices happened to play here a fundamental role (13),
one might speculate that T c actually decreases when the volume is too large.
An even more compelling suggestion in the same direction is provided by the
ammoniated n=3 compound NH 3 K 3 C 60 , which can be thought of as a very
expanded anisotropic K 3 C 60 , and is an 5=1/2 antiferromagnetic insulator,
certainly of the Mott type. Under pressure it undergoes an insulator-to-
metal transition, and superconductivity is recovered with a large T c rv 28
K at the transition (14), decreasing for increasing pressure. This result is
very important. It shows clearly that superconductivity in fullerides does,
like in cuprates, occur close to a Mott insulator. A corollary is that ignoring
strong correlations cannot in the end be right in the superconducting phase
either.
If the behavior of these compounds could be accepted as evidence that
T c varies nonmonotonically with volume, the family of fullerides under
pressure would realize the whole T c versus>" curve - first increase, then
decrease - within the narrow span of unit cell volumes per C 60 ranging
from 670 A3 in K 3 C 60 at 28 Kbar (T c rv 6 K) to something of order 800
A3 and 830 A3 in CS 3 C 60 at 15 Kbar (T c rv 40 K) and at residual pressure
(T c rv 18 K), respectively. In other words, in the unit cell volume interval
6. V /V rv 0.2, fulleneres could explore almost the full region from weak
to strong coupling superconductivity. This in itself suggests that a simple
BCS-type of approach will not be appropriate. We will argue here that
strong correlations are the clue ingredient to explain such a drastic changes
in a narrow unit-cell volume interval.
Evidence of non-BCS behavior of superconductivity in trivalent fullerides
is widespread. Most strikingly, NMR measurements have shown the exist-
ence of a spin gap 6. sp in c:::' 0.07 ~ 0.1 eV even in metallic K 3 C 60 (15). The
spin gap magnitude found in n=3 fullerides is moreover quite comparable
with the spin gap of an isolated c~t molecular ion, suggesting that molecu-
lar features do survive in the metallic phase, as expected close to a Mott
 99

transition, but not otherwise explainable within band theory.

Turning to tetravalent fullerides such as K4C60 and Rb 4C 6o , they should,
as was previously said, also be metallic according to DFT-LDA calculations.
In reality, they appear on the contrary to be on the nonmagnetic insulating
side of an insulator-to-metal transition that can be observed under pressure
(16). The zero pressure insulating phase could at first sight be explained
away (say within Hartree-Fock theory) as a band insulator. By assum-
ing for example a large static J ahn-Teller effect (17), the four electrons
would occupy two out of three Jahn Teller split tlu orbitals. The mean
field treatment of Coulomb repulsion in Hartee-Fock simply adds to the
J ahn-Teller distortion, amplifying its effect and leading in this case to an
exceedingly wide insulating gap. However, the gap in the real materials is
by far not that large, and in fact about zero in K4C60 and nearly zero in
Rb 4C 6o (16). Moreover, there is no experimentally detectable static Jahn-
Teller distortion (18). Finally, and most strikingly, the magnetic response
is dominated here too by a 0.1 eV spin gap (19, 16, 15). This spin gapt is
totally unaccounted for in the band insulator model provided by Hartree
Fock, or for that matter by any other non-strongly correlated theory. All
of this indicates that K 4C 60 and Rb 4C 6o are not really band-insulators but
most likely singlet, nonmagnetic ideal Mott insulators, systems possessing
no spin entropy and devoid of any breaking of symmetry.
Our earliest analysis of the role of correlations in fullerides started pre-
cisely with K 4C 60 , in an attempt to understand how and why an ideal Mott
insulator could be favored instead of the Hartree-Fock band insulator (17).
In order to treat strong correlations we resorted to the Dynamical Mean
Field Theory (DMFT) that permits an analysis of both Fermi Liquid states
and Mott insulating states. We indeed found that the Mott insulator is the
most stable phase for realistic parameters. That analysis however also led
to unexpected developments (20), which are the subject of these lecture
notes. In addition to the Mott state it was found that, by increasing con-
tinuously the bandwidth W of our model for K 4C 60 , the Mott insulator
developed a direct and continuous transition not to a simple metal, but to
a superconductor, which only subsequently turned into a normal metal. A
careful study of this superconducting region, showed that a) the unexpected
superconducting phase shares many common features with high T c super-
conductors; and that b) the Fermi liquid hypothetical scenario mentioned
above seems to describe quite well the onset of superconductivity as well as
the approach to the Mott insulating phase. The analysis of that model also
allows to identify the condition under which the scenario applies: namely
when the vertex correction of the attractive channel compensates for the
wave function renormalization, rv rv 1/Z2. This situation can only occur
when the pairing attraction involves other degrees of freedom than the
100

charge, because charge fluctuations get completely suppressed close to the

Mott transition. In fullerides the pairing attraction is in fact a result of
J ahn-Teller coupling, that involves orbital and spin degrees of freedom.
The related tendency to stabilize J ahn Teller distorted low spin states for
the molecular ion persists for all charge states (10, 21). This is parallel to
the situation in the t-J model of cuprates, where the attraction is provided
by the spin exchange J, again unrelated to charge.

2. A model for alkali doped fullerenes

After this long introduction, let us describe the model for alkali doped
fullerenes we studied. The Hamiltonian is

(1)

where ck ia is the electron creation operator at site R in the orbital i

(i = 1,2, '3) of the tlu multiplet with spin (J" and nR = L:ia Ck,iaCR,ia the
occupation number at site R. t~RI are the hopping matrix elements giving
rise to three narrow bands of width W '"'-' 0.5 eV. For convenience we will
assume nearest neighbor hopping to be diagonal in the orbital index. U is
the on-site Coulomb repulsion which tends to suppress valence fluctuations
around the average electron density (nR) = n.
The last term HHund is the multiplet exchange splitting required by
Hund's rules. Within the tlu subspace, we can define angular momentum
operators
LR,i = LLCk,ja (i~i) jk cR,ka' (2)
jk a

where (L i ).]k = -iCijk' as well as spin operators

(3)

a-ibeing the Pauli matrices. In terms of these operators the Hund term
reads

The bare JH is of course positive, favoring high spin states in artificially

frozen high symmetry molecular ions. However, there is no such freezing in
real C~o, and the Jahn-Teller (JT) coupling to the H g molecular vibrations
 101

can and does (10, 21) reverse Hund's rule, generally favoring instead low
spin and angular momentum configurations. We include approximately this
effect by assuming JH < O. This is formally equivalent to treating the
electron-vibron coupling within an antiadiabatic approximation, where it
can be shown to renormalize JH ----+ JH - 3EJT/4, EJT being the Jahn-
Teller energy gain. The antiadiabatic approximation is more justified than
it may seem in strongly correlated fullerides, where the vibron frequencies
of order 0.1 eV, is not in fact larger, but rather comparable, with the
bare electron hopping Itl. Vibrational frequencies are in fact safely larger
than the renormalized hopping Ziti close to the Mott state, where Z « 1.
In any case, the neglect of retardation involved in replacing the true JT
coupling with a negative JH does disfavor superconductivity, by preventing
high energy screening of the Coulomb interaction U. Therefore, adoption
of the JH < 0 approximation in place of the fully retarded JT effect is
conservative, and makes physical sense in this problem.
Let us for a moment consider the isolated molecule with JH < 0 and
with two valence electrons, or, equivalently, four (having in mind K 4 C 60
type compounds). The ground state has quantum numbers 8 = L = O.
The spin gap to the spin triplet state with quantum numbers 8 = L = 1 is
5IJHI. We may therefore fix an approximate JH magnitude by comparing
this value of the spin gap with the exact singlet-triplet gap for the isolated
molecular ion. The n=4 C 60 ion has of course a more favorable Jahn Teller
distortion in the 8=0 ground state than in the 8=1 excited state, and this
gives rise to a spin gap tlspin r-v 0.1 eV. This value is quite comparable with
that observed by magnetic resonance experiments in K 4 C 60 . The compar-
ison leads to a crude estimate JH c:::: -0.02 eV, namely JH/U c:::: -0.02,
the value we adopt hereafter. While this will constitute a very weak pairing
attraction, we shall see that it may still give rise to a remarkably important
consequences.
Since all relevant interactions, namely U and JH, are local, the model
described by the Hamiltonian (1) is suitable to be treated by the so-called
Dynamical Mean Field Theory (DMFT) (22), which is a very efficient tool
to deal with strong local correlations. We studied (1) by DMFT with an
average number of electrons (holes) per site n = 2 at fixed JH /U = -0.02
but at various ratios of U /W. Before presenting the results, let us analyse
the behavior of (1) in the two extreme limits U /W « 1, weak coupling,
and U/W » 1, strong coupling.
In the weak coupling limit, the Hamiltonian (1) describes a metal with
three 1/3-filled degenerate bands. Since JH < 0 is an explicit attraction
present in the model, one needs to check the stability of the metal to-
wards Cooper pairing. In the absence of U, JH would indeed lead to a
102

superconducting phase with an s-wave order parameter

3
(!:::.R) = L(c1,i r c1,it) -I- O. (5)
i=l

However, for our choice of JH /U = -0.02, the dimensionless superconduct-

ing coupling A = 10p1JHI/3 = 0.2pU/3, p being the DOS at the Fermi
energy per spin and band, is much smaller than the Coulomb pseudopo-
tential fL* = pU. Therefore the weakly correlated metal is stable against
superconductivity.
In the strong coupling limit, U /W » 1, the system is a Mott insulator.
Each site is occupied by two electrons which, due to the inverted Hund's
rules, form a spin as well as an orbital singlet. Since in this limit the
intermolecular superexchange generated by virtual hopping processes is
smaller than the on-site I JH I the molecular singlet state remains stable
against the triplet. Therefore the Mott insulator is nonmagnetic and has a
gap in both spin and orbital excitations, of the same order of magnitude as
the molecular gaps, in agreement with the observed behavior in tetravalent
fullerides. This ideal Mott insulating phase - or rather recalling that JH < 0
really mimicks the J ahn-Teller effect, this " Mott-Jahn-Teller" insulating
phase (23) - is a kind of on-site version of a valence-bond spin liquid phase.

3. Dynamical Mean Field results

The above discussion shows that, as U /W increases, there should be in

our model a transition from a weakly correlated metal to a non-degenerate
singlet Mott insulator. However, such a transition cannot be continuous.
In fact the only way a metal which satisfies Luttinger's theorem can turn
smoothly into an insulator is by flattening the energy vs. momentum dis-
persion, namely by a diverging effective mass. This would imply that a
finite entropy per site is released at the Mott transition, which is not
compatible with the insulator being non-degenerate, and hence with zero
residual entropy. Therefore either the transition is discontinuous, or else a
third phase must intrude between the metal and the Mott insulator. This
third phase will most likely break a symmetry of the Hamiltonian so to
avoid Luttinger's theorem, and allow the large metallic Fermi surface to
shrink continuously as one goes from the metal phase towards the Mott
phase.
This is precisely what we find by DMFT: a narrow but well defined
superconducting region with s-wave order parameter (5) intrudes between
the metal and the Mott insulator (see Fig. 1). Of course for the much
larger realistic value of U /W c:::' 2, we do find an insulating phase, quite
 103

Mon
METAL SC INSULATOR

I I
I I
o 0.8 0.9 UIW
Piyu.re 1. Pha:;e diagram a:; a funct.ioll of U flV. SC st.ands for superconductor

compatible with the behavior of realllcro-pressure tetravalent fullerides. It

is in fact a very good question whether in view of our results the pressme-
induced insulator-metal transition which was demonstrated in these com-
pounds (16) could be re-examined closely, searching for a possible narrow
superconducting region.
Theoretically, the existence of a Supcl'conducting pocket. in our model
is a surprise. At weak coupling our model has no supcl'conducLing instabil-
it,y, and we would inwitively llot expect the s·wave pairing tendency to
increase by increasing intra~sit,e electron repulsion. Even morc st.l'iking arc
the properties of t,he supel'conducLing phase.
Tn Fig. 2 \ve plot the superconducLing single particle gap 6. as function
of !JHI/HI. The points correspond to our case with U = 50!JHI, where
superconductivity occurs only within a. small region just before the Mott
transition. The gap is ext.remely large for such a weak attraction. For com-
parison, we also show t.he BCS gap calculated in the same model but. for
U = 0 : only after multiplication of the latter by 1000 are they comparable.
Therefore, in the presence of a large U, we find that superconductivity
appears where it would seem less probable, and with gap values thousand of
times larger than at U = 0 ! \:\"e thus have found a soluble model (at least
as far as DMFT can be regard<..--d as a wlution) showing explicitly that
strong correlations are not detrimental but may on the contrary enorm-
ously enhance superconductivity, which is of course suggestive of high T c
superconductors.
The analogy gets even closer by calculating the spin gap 6 spill c.xtracted
by the main peak values of the dynamical spin susceptibility, also drawn
in Fig. 2 reduced by a factor 20. The two gaps, the superconductillg gap
and the spin gap follow each other at the onset, of superconductivity, but
quite soon they depart, the sllperconducting gap disappearing in the insu-
lator, and the spin gap smoothly merging with the large one of the singlet
IIv[ott insulator. This behavior is reminiscent of the pseudo-gap region in
underdoped cuprates.
Finally \ve note that superconductivity occurs in the small region 61'VjW ("v

0.12 around W 1.18U, reaching at this value almost the largest possible
("v
104

0.06

IU "" -50 J1
0.06 t
Ll sPin /20
-<
...........
<I
r:e
0.04

METAL
\
103x~U=O

0.02 MI

O-......t===--~ ....- ......-~---....

0.012 0.02

Figure 2. Superconducting gap (square points) and spin gap reduced by a factor 20
(circles) as functions of IJH I/W at fixed U = 50lJH I. Also drawn is the superconducting
gap at U = 0, 6u=o, as function of IJHI/W, multiplied by one thousand.

superconducting gap at the given attraction 10IJHI/3. This is compatible

with the behavior of T c vs. unit cell volume in fullerenes, as discussed in
the introduction.
Since Fermi liquid theory seems to apply all the way up to the Mott
transition in the single band Hubbard model, we can also try to interpret
our results within the same framework. Indeed our model (1) has many fea-
tures which suggest it as a possible condidate for the Fermi liquid scenario
for correlation enhanced superconductivity discussed in the introduction.
The Hund's rules, even if inverted, are known to be most likely violated in
a metal but satisfied in an insulator. Hence we reasonably expect that a
strong U will not impede the working of JH but rather favor it.
Let us assume that the vertex corrections compensate exactly for the
wavefunction renormalization so that the renormalized

is practically unchanged. We do expect a reduction of the Coulomb pseudo-

potential U* felt by the quasiparticles, which is equal to U only at weak
coupling. The DMFT solution of the single band Hubbard model suggests
that a wrong quasiparticle charge configuration, namely a quasiparticle
double occupancy, costs energy ZU and not U, which also agrees with the
 105

behavior of the Gutzwiller wavefunction. Let us assume the same holds in

our multiband model, namely the Coulomb pseudopotential renormalizes
like U* = ZU. Then the quasiparticle scattering amplitude in the s-wave
channel A* should modify into

0.2) U.
(Z-T (6)
While at weak coupling, Z "':' 1 the scattering amplitude A* is repulsive,
at sufficiently strong correlation, such that Z <::: Zc "':' 0.067, the repulsion
may turn into attraction and the system becomes unstable towards super-
conductivity. The conjecture (6) for the quasiparticle scattering amplitude
also explains the large values of the gap compared with BCS values. Since
the Mott transition occurs for W "':' U, when Z is further reduced down
to rv Zc/2, the effective attraction A* gets of the same order as the quasi-
particle bandwidth, W* = ZW "':' ZcW/2 "':' ZcU/2 = 10IJHI/6. In the spirit
of Landau Fermi liquid theory, this should correspond to the maximum of
the superconducting gap. The actual gap value, ~ rv W*/10 = IJHI/6
linear in the bare attraction, is in fact compatible with that expected at
the maximum of the ~ vs. >. curve for a generic unretarded attraction at
U = 0 (24). The notable difference is that this value is here obtained for
a bare bandwidth W rv 601JHI, namely for a bare>. rv 0.056, where BCS
theory would rather predict ~BCS rv W exp (-1/ >.) much smaller than the
actual value we obtain ~ "':' W >./20.

3.1. FERMI LIQUID THEORY

So far the Fermi liquid scenario is just a conjecture, which needs to be

supported by calculations. DMFT allows to extend the normal metal solu-
tion inside the region of superconducting instability by simply preventing
gauge-symmetry breaking in the self-consistency equations. This provides a
full description of the normal metal phase up to the Mott transition, which
can be compared with Landau Fermi liquid theory. Our model possesses a
spin SU(2) and an orbital 0(3) symmetry so that the Landau functional
contains 7 parameters: the mass enhancement, which in DMFT is simply
m*/m = l/Z, and six j-parameters. Indeed we can define six density
operators in momentum space, the charge density
3
nk = L L c1,io- ck,io-,
i=l 0-

the spin density

3
O"k = L L c1,io:O"o:;3Ck ,i;3'
i=l 0:;3
106

the angular momentum density in the symmetric and antisymmetric chan-

nels, with components (a, b = 1,2,3)
3

lk = L L 4,io- (La)
i,j=l 0-
ij Ck,jo-'

and
3
(la (J"b) k = LL 4,ia (La) ij o-~(jCk,j(j'
i,j=l a{3

In addition we can introduce five independent bilinears of the angular mo-

mentum operators, namely WI = V3(L 2 L3 + L3 L2 ), W2 = V3(L 3 L l +
L l L3 ), W3 = V3(L l L2 + L2 Ld, W4 = V3(L l Ll - L2 L2 ), and finally
W5 = (LlL l +L2 L2 -2L 3 L3 ), in terms of which we define, with d = 1, ... ,5,
3

w~ = L L 4,io- (Wd)
i,j=l 0-
ij Ck,jo-'

as well as its antisymmetric partner (a = 1, ... ,3)

t=LL
3

(wd(J"a 4,ia (Wd) ij o-~(jck,j{3'

i,j=l a{3

Through all the above definitions, the Landau functional reads

3
E = L L L Ek4,io- ck,io- +
k i=l 0-

~ L (is 7Ik7lk' + jA(J"k' (J"k') +

kk'
~~~ [lllklk'+~hA (la(J"b)k (la(J"b)k'] +
V1 E
5~ [
gS w~ 3
w~, + ~ gA (w d(J"b) k (w d(J"b) k' ] . (7)

Here Ek are the single-quasiparticle energies with respect to the chemical

potential. We introduce the F-parameters FS(A) = 6pjS(A) /Z, HS(A) =
4ph S(A) /Z and CS(A) = 12pgS(A) /Z, where p is the bare DOS, through
which the susceptibilities X to each field which couples to one of the above
densities are given by the standard expression
1 1
--- (8)
Zl+F
 107

0.6
•
0.2ITJ
0.1

0.6
•
• 0
0.7 0.8 0.9

0.4

0.2

0.2 0.4 0.6 0.6

U/W

Figure 3. Quasiparticle residue for the metallic solution. The inset shows the enlarged
region where superconductivity appears, which is signalled by a vertical line at U c::: 0.8W.
The line is the conjectured A* of Eq. (6) which indeed crosses zero at U c::: 0.8W when
Z ~ 0.067.

Here F = FS(A) , HS(A), GS(A) depending on the form of the external field.
By calculating within DMFT the quasiparticle residue Z as well as all the
six susceptibilities, we obtain all F-parameters.
Fig. 3 shows the decrease of Z in the metallic solution on approaching
the Mott transition (MIT). Superconductivity sets in at Z very close to our
conjectured value Zc = 0.67.
The charge compressibility, "', is a decreasing function of U, in agree-
ment with the system approaching the incompressible insulator, see Fig.
4. Through Eq. (8) this implies that F S is always positive and diverges
faster than l/Z at the MIT, actually like 1/Z 2 . All other susceptibilities,
(see for example Fig. 4 where the spin susceptibility is shown), initially
increase due to Stoner enhancement, then turn around at U /W rv 0.7, and
finally vanish at the Mott transition, consistent with the spin and orbital
gaps. This implies that FA, HS(A) and GS(A) all start negative and decrease
roughly like -U/W, as can be perturbatively checked, until U rv ZW, when
they turn upward, cross the zero and finally diverge as + 1/ Z2 at the MIT.
The Landau F-parameters are related to the quasiparticle scattering
amplitudes strictly only at low frequencies and momentum transfers, hence
it is not rigorously justisfied to use them to estimate the scattering amp-
litudes in the Cooper channel, primarily a high momentum transfer process.
Yet the F-parameters have been often used to that purpose, for instance
to justify the triplet p-wave pairing in He 3 . In our case, this approximation
108

6
0.6

~g
.~
"-
~
><
g
0.'
2

0 0
0 0.' 0.6

U/W

Figure 4. Charge cOlllpressibili~y K, and spin susceptibility X normalized to the uncor-

related value:; /'4) and xu, n..>spL,<:tivdy. All other slJ.';Ceptibilit.i~ behave qualitat.ively like
x·

leads to t.he following estimate of the quasiparicle scattering amplitude in

the s+wave channel of Eq. (5):

Z [ FS pi 11 5 lJA CS CA ]
3
A= 12p 1+FS- 1+pA -31+HS+91+HA +51+Cs-151+GA .
(9)
Fig. 5 shows the scattering I:unplitude A a,s calculated through the above
formula.
At weak coupling, A = U -\- lOJH/3 > 0 is repubive and increases with
UjW until at UIW '" 0.5 it turns downwards and crosses zero signalling
a sllperconducting instability: one which we actually find when allowing
gauge-symmetry breaking. As the l..,lIT is approached, all F-parameters
diverge hence
Z
A-----jo--Z·
p.
saturat.es to a value about equal to half the quasipart.icle bandwidth W.. =
ZlV. For comparison we have also drawn in Fig. 4 our conjectured A.. of
Eq. (6). Up to the metal-to-superconductor transition, A.. seems \.0 agree
quite well wit.h A, alt.hough close to the ~i[1T A.. would remain constant
while A vanishes as Z. If the estimate of A through t.he F-parameters is
at least qualitatively correct, this behavior would imply that the attraction
among the quasiparticles cannot exceed t.heir bandwidth.
 109

0.08 .----------r--.--------,,....-----,

0.06

~
<I 0.04
o
Q.
METAL

0.02 MI

O ......c=....---c::F--a____-....----4I
0.012 0.02

Figure 5. Quasiparticle scattering amplitude A in units of the quasiparticle bandwidth

ZW as calculated through Eq. (9) (circles) and as conjectured in Eq. (6) (line).

Although the expression (9) seem to correctly reproduce the onset of su-
perconductivity, its validity needs further justification. As Nozieres pointed
out to us (25), the same scheme applied to the single band Hubbard model
would lead to an incorrect result. In the single band Hubbard model, the
oS-wave quasiparticle scattering amplitude within the same approximation
is

(10)

Close to the MIT, both F-parameters diverge, F S 1/Z2 and FA ;v liZ, ;v

respectively. Hence one would still predict A ZW near the MIT,

;v -

namely an instability towards an oS-wave superconductor, which seems to

be very unlikely. However the single band Hubbard model is pathological.
While the uniform magnetic susceptibility is finite across the MIT, im-
plying FA ;v liZ, the local susceptibility diverges like liZ. Therefore
the quasiparticle interaction in the spin channel must behave in a com-
pletely different way depending whether the momentum transfer is zero
or is finite, which correspond within DMFT to either uniform or local
response, respectively. Since the typical momentum transfer involved in
superconductivity is finite rather than zero, one would be tempted to use
in the expression (10) for A the analogous of the F-parameters as naively
110

extracted by local susceptibilities

Xloeal 1 1
(0)
Xloeal
Z 1 + Floeal

This would still imply Fl~cal rv 1/Z2, but instead Fl~eal = canst. < 0,
namely a repulsive A > 0 at the MIT. This is obvioulsy a more realistic
prediction.
In our model there is in fact no such ambiguity since the uniform and the
local susceptibilities behave exactly in the same way, hence both procedures
lead to the same conclusion A rv -ZW/2 near the MIT. Moreover the use of
local Landau parameters might be justified within DMFT where the lattice
problem is mapped onto an Anderson impurity model where the coupling
to the conduction bath is determined self-consistently by imposing that the
conduction-electron local Green's function coincides with the impurity one.
At self-consistency one can indeed imagine that a Fermi liquid theory for
the effective Anderson impurity model should be valid.
We conclude this overview of the Fermi liquid properties of the Hamilto-
nian (1) close to the Mott transition, by studying another quantity which
sheds some light on the real physical behavior. For our choice of a very small
JH /U = -0.02, superconductivity arises when the Mott-Hubbard bands
are already well separated from the narrow quasiparticle resonance. This
suggests that the effective Anderson impurity model has already flowed in
the Kondo regime, where a two-component description might be used (26).
Let us consider the probability P( n) for a site to be occupied by n electrons
in the ground state. In Fig. 6 we draw P(n) for the superconductor (panels
B to D) and for the nearby Mott insulator (panel A), which are quite
similar. In spite of the large superconducting gap, there is no evidence of
preformed pairs or bipolarons as underlined by the strong steady peaking
of P(n) at n = 2. We assume the two-component form of P(n)

where Pins (n) is the probability distribution in the nearby Mott insulator.
Since Z, P(n) and Pins(n) are all known, this allows to extract the quasi-
paricle probability distribution pq;C)(n), which is drawn in Fig. 6A'-6D'.
It shows oscillations between even and odd n's, as expected for a supercon-
ductor, with no relevant variations as a function of U /W upon approaching
the Mott transition (U/W c:::' 0.9), in agreement with an intermediate
coupling regime as implied by A rv - W*. To check the consistency of the
(METAL)
two-component form, we also extracted the Pqp (n) from the (meta-
stable) metallic solution obtained by preventing gauge-symmetry breaking.
The results are very similar to the probability distribution of the free Fermi
 111

P(n) P::)(n) P~BTAL)(n)

u• O.
0.6
SC
0
0 2 . 80 2 .. 80 2 . 8
n n n
Figure 6. Occupation probabilities for the particles, P(n), and for the quasiparticles in
the superconducting and in the metallic phases, n (n) and pJ;;c
I
pJi:
ET AL) (n), respectively.

gas, U = JH = 0, drawn in Fig. 6D". The agreement is quite remarkable,

as PqCJ:IETAL)(n) is but a tiny fraction of order Z of P(n) and there is no
free parameter.
These results strongly support to a two-componentdescription where
fermion-like quasiparticles of small weight Z get strongly paired while float-
ing in a prevailing Mott insulator background. That background slows down
quasiparticle motion while taking away their Coulomb repulsion but not
their pair attraction. As a remark we notice that the behavior of P(n) is also
consistent with our conjecture that the quasiparticle repulsion renormalizes
as U* = ZU.

4. Conclusions

We have shown in a simplified model for alkali doped fullerides that strong
correlations may enhance superconductivity just close to the Mott trans-
ition. The detailed analysis of the results indicates that such effect appears
when the pairing occurs in a channel not involving the charge degrees of
freedom which are most affected by the repulsion. Then superconductivity
is automatically pushed to an intermediate coupling regime where the qua-
siparticle bandwidth, strongly suppressed by correlations, gets of the order
of the attraction, which stays unrenormalized.
112

In our model (1) this newer pairing mechanism, able not only to avoid
U but to take advantage from U, is provided by an inverted Hund's rule
mimicking the J ahn-Teller coupling to molecular vibrations. However the
general mechanism is quite general and seems valid in other cases too, as
for instance in t - J models for cuprates.

Acknowledgements

This work was sponsored through MIUR COFIN, by INFM/G, and by the
EU, through contract ERBFMRXCT970155 (FULPROP). We are grateful
to P. Nozieres and to G. Santoro for illuminating discussions.

References

1. Based on lectures presented at Hvar, and at Varenna (2002).

2. A. Kaminski et al., Phys. Rev. Lett. 84, 1788 (2000).
3. P. W. Anderson, Science 235, 1196 (1987).
4. G. Kotliar and J. Liu, Phys. Rev. B 38, 5142 (1988).
5. S. Sorella et al., Phys. Rev. Lett. 88, 117002 (2002).
6. A. Paramehanti, M. Randeria, and N. Trivedi, Phys. Rev. Lett. 87, 217002 (2001).
7. O. Gunnarsson, Rev. Mod. Phys. 69, 575 (1997).
8. R.L. Martin and J.P. Ritchie, Phys. Rev. B 484845 (1993);
9. R.W. Lof, M.A. van Veenendaal, B. Koopmans, H.T. Jankman, G.A. Sawatzky,
Phys. Rev. Lett. 68, 3924 (1992).
10. N. Manini, E. Tosatti, and A. Auerbach, Phys. Rev. B 49, 13008 (1994)
11. O. Zhou et al., Science 255,833 (1992).
12. T.T.M. Palstra et al., Solid State Commun. 93, 327 (1995).
13. N. Manini, G.E. Santoro, A. Dal Corso, and E. Tosatti, Phys. Rev. B 66, 115107
(2002)
14. O. Zhou et al., Phys. Rev. B 52, 483 (1995); S. Margadonna, K. Prassides, H.
Simoda, y. Iwasa, and M. Mezouar, Europhys. Lett. 56, 61 (2001).
15. V. Brouet, H. Alloul, T.N. Le, S. Garaj, and L. Forro, Phys. Rev. Lett. 86, 4680
(2001).
16. R. Kerkoud et al., J. Phys. Chern. Solids 57,143 (1996).
17. M. Capone, M. Fabrizio, P. Giannozzi, and E. Tosatti, Phys. Rev. B 62,7619 (2000).
18. C.A. Kuntscher, G.M. Bendele, P.W. Stephens, Phys. Rev. B 55, R3366 (1997).
19. G. Zimmer et al., in Physics and Chemistry of Fullerenes and Derivatives, edited
by H. Kuzmany, O. Fink, M. Mehring, and S. Roth (World Scientific, Singapore,
1995).
20. M. Capone, M. Fabrizio, and E. Tosatti, Phys. Rev. Lett. 86, 5361 (2001); M.
Capone, M. Fabrizio, C. Castellani, and E. Tosatti, Science 269, 2364 (2002).
21. M. Lueders et. al., Phil. Mag. B 82, 1611 (2002).
22. A. Georges, G. Kotliar, W. Krauth, and M.J. Rozenberg, Rev. Mod. Phys. 68, 13
(1996).
23. M. Fabrizio and E. Tosatti, Phys. Rev. B 55, 13465 (1997).
24. R. Micnas, J. Ranninger, and S. Robaszkiewicz, Rev. Mod. Phys. 62, 113 (1990).
 113

25. P. Nozieres, private communication and preprint (2002, to appear in Jour. of Stat.
Phys.).
26. G. Kotliar, Em. Phys. J. B 11, 27 (1999), and references therein.
INELASTIC LIGHT SCATTERING AND THE CORRELATED
METAL-INSULATOR TRANSITION

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

J.K. FREERICKS
Department of Physics, Georgetown University, Washington,
DC 20057, U.S.A.
T.P. DEVEREAUX
Department of Physics, University of Waterloo, Canada
R. BULLA
Theoretische Physik III, Elektronische Korrelationen und Mag-
netismus, Institut fur Physik, Universitiit Augsburg, D-86135,
Augsburg, Germany

Abstract. Inelastic light scattering is an important probe of the two-particle charge

excitations in a correlated material. We show how to determine the nonresonant response
to inelastic light scattering exactly in the limit of large spatial dimensions by employing
dynamical mean field theory. We examine the optical photon case of Raman scattering
and the X-ray photon scattering case which exchanges both energy and momentum with
the charge excitations. A number of formal details that have not appeared elsewhere are
included here.

Key words: Metal-Insulator Transition, Inelastic Light Scattering, Dynamical Mean-Field

Theory, Infinite Spatial Dimensions, Exact Solution, Hubbard Model, Falicov-Kimball
Model, Electronic Raman Scattering, Charge Dynamics

1. Introduction

In correlated materials, two-particle properties, such as charge fluctuations,

may be quite different from single-particle properties (such as the interact-
ing density of states measured in photoemission). In particular, one expects
significant renormalization effects of the charge excitations near the correl-
ated metal-insulator transition. There are a number of direct experimental
probes of the charge excitations in a material. The most common probe is

115
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 115–122.
© 2003 Kluwer Academic Publishers
116

an optical conductivity measurement, which is either performed directly in

a transmission geometry in the THz range, or involves a Kramers-Kronig
analysis of reflectivity data for higher frequencies. Another important ex-
perimental probe involves the inelastic scattering of light from the charge
excitations. If optical photons are used, the process is called electronic
Raman scattering, and only energy is exchanged with the correlated charge
excitations. If X-rays are employed, then both energy and momentum are
exchanged with the charge excitations.
The theoretical description of inelastic light scattering is complicated,
and has only been solved recently in the limit of large dimensions (1, 2, 3).
This solution involves the exact evaluation of the two-particle diagrams for
nonresonant inelastic light scattering. Nonresonant scattering implies that
there is no feedback effect from the energy of the incident photon, and only
the transferred photon energy (and momentum) enters into the scattering
response function. While it is well known that many resonant scattering
effects exist, and indeed, resonant effects are necessary for some experiments
to even be possible, the main qualitative features of the inelastic scatter-
ing process are contained in the nonresonant formalism. Furthermore, the
computations are significantly more tractable.

2. Formal Development

We employ dynamical mean-field theory (DMFT) to solve the inelastic light

scattering problem. In DMFT, the electronic self energy is local, and only
the local piece of the irreducible vertex function enters into any physical
response function. We describe below how to formulate the inelastic light
scattering problem on a hypercubic lattice in d ----+ (X) dimensions. It turns
out that much of the formalism is independent of the explicit form of the
Hamiltonian, but the interactions must all be local. For concreteness, we
consider the Hubbard Hamiltonian (4)

(1)

with rescaled hopping integral t* (5) (between nearest neighbors-the sum-

mation is over nearest-neighbor pairs) and screened Coulomb integral U.
The cL- (Cia) operators create (destroy) an electron with spin (]" at site i. The
noninteracting bandstructure is E(k) = -limd---+= 2:.1=1 t* cos(ki)/Vd and
the noninteracting DOS becomes p( E) = exp( _( 2) /t* ft. The inelastic light
scattering response function is given by an effective density-density correl-
ation function S( q, v) = -~ [1 + n(v)]ImX(q, v) for transferred momentum
 117

CDO CID ~O',

(0) (b)
'Ya =::::::::
§§ 'Ya -- 'Ya 'Ya
~ 1'0
=:::::::::

- T 'Ya 'Ya -T 1'0

Figure 1. Coupled Dyson equations for the inelastic light scattering density-density
correlation functions described by the scattering amplitude fa. Panel (a) depicts the
Dyson equation for the interacting correlation function, while panel (b) is the supple-
mental equation needed to solve for the correlation function (the difference in the two
equations is the number of fa factors). The symbol r stands for the local dynamical
irreducible charge vertex. In situations where there are no charge vertex corrections (like
B lg scattering along the zone-diagonal), the correlation function is simply given by the
first (bare-bubble) diagram on the right hand side of panel (a).

q and energy v with

x(q,v) = ([p(q),p(-q)])(v) (2)

formed with an "effective" density operator given by

p(q) = L ia(k)ct(k + q/2)c a (k - q/2), (3)

k,a

where the 4(k) and ca(k) operators create or destroy an electron with spin
(J" and momentum k and n(v) = 1/[1 - exp(v/T)] is the Bose factor. The

strength of the scattering ia is determined by the curvature of the band as

(4)

Here ei,s denote the incident, scattered photon polarization vectors, re-
n
spectively, and we have chosen units kB = c = = t* = 1 and have set the
hypercubic lattice constant equal to 1. We can classify the scattering amp-
litudes by point group symmetry operations. If we choose e i = (1,1,1, ... )
and e S = (1, -1, 1, -1, ... ), then we have the BIg sector, while e i = e S =
(1,1,1, ... ) projects out the A Ig sector since the B 2g component is identically
zero in our model due to the inclusion of only nearest-neighbor hopping (1).
Hence iAl g (q) = -E( q) and iBl g (q) = t* L~I cos qj(-I)j / /d.
The Dyson equation for the density-density correlation function takes
the form given in Figure 1. Note that there are two coupled equations
illustrated in Figures 1 (a) and (b); these equations differ by the number
of ia factors in them. The irreducible vertex function r is the dynam-
ical charge vertex which is known explicitly only for the Falicov-Kimball
118

model (6). If the scattering amplitude r does not have a projection onto
the full symmetry of the lattice, then there are no vertex corrections from
the local dynamical charge vertex (7). This is the only case that can be
analyzed for the Hubbard model.
Let's begin our discussion on the imaginary axis in the BIg sector. If we
restrict ourselves to the zone diagonal, then q = (q, q, q, ... , q). Examining
the diagrams in Figure 1 (a), we see that there is a bare response plus a
vertex correction term. The vertex correction term, however, must vanish,
because the leftmost piece of that diagram is

1
x (5)
iW n + iVI + fL - I;(iw n + iVI) - f(k + ~q)
for "energy" transfer iVI = 2in-Tl and momentum transfer q [in Eq. (5)
iW n = iJrT(2n + 1) is the fermionic Matsubara frequency and fL is the
chemical potential]. It vanishes, because each term indexed by j is equal
in magnitude, but opposite in sign, so the overall summation is equal to
zero. Hence the BIg response on the zone diagonal (including the Raman
response) is given by the bare bubble.
The bare bubble on the zone diagonal is simple to calculate directly (we
shift k ---+ k + q/2):

",",",",'
xo(q, iVI) = -T ~~ 11m ~
"'"' cos(ki
d
+ 2q)cos(k
1
j + 2 q )(-1) J
1 i+"

d---+oo" " 1 d
n k ~,J=

1
x
iW n + fL - I;(iw n ) - f(k)
1
x (6)
iW n + iVI + fL - I;(iw n + iVI) - f(k + q)'
Now, the terms with i # j are all equal in magnitude, but there are as many
positive as negative, so they vanish-only the terms with i = j survive. If
we assume that nand n + l are both larger than 0, and define Zn = iW n +
fL - I; (iw n ), then we can rewrite the fractions as integrals of exponentials

d cos2(k + lq)
XO(q,iVI) = TLL lim L ~ 2
n k d---+oo j=1

X 1 d>.l00 OO

d>" exp[i>'Zn + i>" Zn+l - i>'f(k) - i>" f(k + q)]. (7)

To evaluate this integral, we first expand the functions f(k) and f(k + q)
in terms of the Cartesian momentum components. Then it is obvious that
 119

each j term is equal in magnitude, so the sum over j is trivial. The next step
is to expand each exponential factor that has a l/vId prefactor in a Taylor
series expansion, and keep the lowest nonvanishing terms in the multiple
integrals (8). The multiple integral over momentum then becomes

limd---+oo J J J
dk 1 dk2... dk d cos (k 1 + ~)
2

d. . ~2

II (
j=1
1 + _L_~ cos k·
vId J
+ _L_~ cos(k + q) -
vId J
- cos 2 k·
2d J

~X X2
- cos kj cos(kj + q) - - cos 2(k j + q) + ...). (8)
d 2d
Here we have chosen the extra prefactor to lie in the 1 direction. Each
integral over kj , except the first, is equal to each other and equal to 1 -
(~2 + 2~~/X(q) + ~/2)/4d + .... The first integral becomes ~[1 _ (~2 +
2~X X(q)+X2)/4d-(~X +X2 cos q) sin 2 q/8d+ ...]. Here we use the notation
X (q) = limd---+oo "'L1=1 cos qj/ d (= cos q for a zone-diagonal momentum).
The next step is to rewrite each factor as an exponential, and then take the
infinite product. The result for the integral over k is

1 exp [_ ~2 + 2~~/:(q) + ~/2] (9)

since the terms proportional to sin 2 q coming from the I-direction are just
a 1/ d correction. The end result for the susceptibility is then

XO(q,iVl) = -1 J 1
dEp(E) Zn _ E vI ~ X2Foo (Z~:l_-:;E), (10)

with F00 the Hilbert transformation of the DOS: F00 (z) = .f dEp( E) / (z - E).
The analytic continuation of this expression is straightforward, and
produces the final result for the BIg response

~ ;'00 dw{f(w)xo(w; X, v) - f(w + v)xO(w; X, v)

47f
+ v)]xO(w; X, vn
-00

[f(w) - f(w
(11)

with

j
oo 1 1
Xo(w; X, v) - dEp(E) ( )
. -00 W + It - I: w - E VI - X2
x F (w+V+f.l-I:(W+V)-XE) (12)
00 VI _ X2 '
120

0.3
OJ
(J)
e 0.25
0
0...
(J)
OJ
0.2
l-

e 0.15
0 -- 0.566
E 0.1 -- 0.424
0 -- 0.283
0::: -- 0.141
0.05 -- 0.071

0
0 2 4 6 8 10
Frequency [t*]
Figure 2. Nonresonant BIg Raman response (X = 1) for different temperatures at
U = 4.24 and half filling. The numbers in the legends label the temperature.

and
00 1 1
Xo(w; X, v) - dEp(E)
( )
/ -00 W + JL - I;* W - E VI - X2
x F (W+V+JL-I;(W+V)-XE). (13)
00 VI _ X2
Here f(w) = 1/[1 + exp(wIT)] is the Fermi factor.
A similar, but more complicated analysis can be performed for the A Ig
response, or the BIg response off of the zone diagonal, but we don't have
enough space to report those results here, and they cannot be analyzed
numerically for the Hubbard model.

3. Results

We employ a numerical renormalization group analysis to determine the

self energy and Green's function of the Hubbard model on the real axis (9).
We begin by showing Raman scattering results at half filling for U = 4.24
and a variety of temperatures in Figure 2. At this value of U, the system
is a correlated insulator for all temperatures. The results display all of
the behavior seen on correlated insulators like FeSi (10), or 5mB 6 (11), or
the high-temperature superconductors (12). In particular, we see a charge-
transfer peak at high-energy and the onset of low-energy spectral weight
at a low (but nonzero) temperature. The curves also cross at the so-called
isosbestic point (near v;::::; 3.2).
 121

~
X=-1
(f)
-+-'
C
::J

...0
L
0
L........I

Q)
(f)
C
0
0...
(f)
Q)
L

>.
0
L
I
x

o 2 4 6
Frequency [t *]
Figure 3. Nonresonant Big inelastic X-ray response for different temperatures at
U = 3.54 and pc = 0.9. Five values of the transferred photon momentum are plotted,
each shifted by an appropriate amount, and running from the zone center (X = 1) to
the zone boundary (X = -1) along the zone diagonal. The temperature decreases with
decreasing thickness of the lines and ranges from 0.503 to 0.114 to 0.042 to 0.026.

We show the inelastic X-ray scattering at four different temperatures

for a slightly smaller value of U and at n = 0.9 in Figure 3. Here the
behavior is quite different because the system is metallic for all temperat-
ures. In particular, we see a Fermi-liquid peak form and evolve toward zero
frequency as T is lowered for momentum transfer near the zone diagonal.
But for finite momentum transfer, the peak never fully evolves. In addition,
there is significant "mid-IR" spectral weight occuring at energies below the
charge-transfer peak but not corresponding to the Fermi peak.
122

4. Conclusions

We have presented a number of new results for the inelastic scattering

of light with correlated materials. On the insulating side of the metal-
insulator transition, Raman scattering results agree well with experiments
that have been performed on a wide variety of different materials. We find a
number of interesting features for correlated metals as well, and it would be
interesting to experimentally measure both Raman scattering and inelastic
X-ray scattering for these materials. We expect this behavior might be able
to be seen in a variety of different heavy-fermion compounds.

Acknowledgements

J.K.F. acknowledges support of the National Science Foundation under

grants DMR-9973225 and DMR-0210717. T.P.D. acknowledges support from
the National Research Council of Canada. R.B. acknowledges support by
the Deutsche Forschungsgemeinschaft, through Sonderforschungsbereich 484.

References

1. J.K Freericks and T.P. Devereaux, J. Condo Phys. (Ukraine) 4, 149 (2001); Phys.
Rev. B 64, 125110 (2001).
2. J.K Freericks, T.P. Devereaux, and R. Bulla, Acta Phys. Pol. B 32, 3219 (2001);
Phys. Rev. B 64, 233114 (2001); unpublished.
3. T.P. Devereaux, G. E. D. McCormack, and J.K. Freericks, submitted to Phys. Rev.
Lett. (cond-matj0208049).
4. J.C. Hubbard, Proc. Royal Soc. London, Ser. A 276, 238 (1963).
5. W. Metzner and D. Vollhardt, Phys. Rev. Lett. 62, 324 (1989).
6. A. M. Shvaika, Physica C 341-348, 177 (2000); J. K Freericks and P. Miller, Phys.
Rev. B 62, 10022 (2000); A. M. Shvaika, J. Phys. Stud. 5, 349 (2002).
7. A. Khurana, Phys. Rev. Lett., 64, 1990 (1990).
8. E. Muller-Hartmann, Int. J. Phys. B 3, 2169 (1989).
9. R. Bulla, Phys. Rev. Lett. 83, 136 (1999); R. Bulla, T. A. Costi, and D. Vollhardt,
Phys. Rev. B 64, 045103 (2001).
10. P. Nyhus, S.L. Cooper, and Z. Fisk, Phys. Rev. B 51, R15626 (1995).
11. P. Nyhus, S.L. Cooper, Z. Fisk, and J. Sarrao, Phys. Rev. B 52, R14308 (1995);
Phys. Rev. B 55, 12488 (1997).
12. X. K Chen, J.G. Naeini, KC. Hewitt, J.C. Irwin, R. Liang, W.N. Hardy, Phys. Rev.
B 56, R513 (1997); J.G. Naeini, X.K Chen, J.C. Irwin, M. Okuya, T. Kimura, and
K Kishio, Phys. Rev. B 59, 9642 (1999); M. Rubhausen, O.A. Hammerstein, A.
Bock, U. Merkt, C.T. Rieck, P. Guptasarma, D.G. Hinks, M.V. Klein, Phys. Rev.
Lett. 82, 5349 (1999); S. Sugai and T. Hosokawa, Phys. Rev. Lett. 95, 1112 (2000);
M. Opel, R. Nemetschek, C. Hoffmann, R. Philipp, P.F. Muller, R. Hackl, 1. Tutto,
A. Erb, B. Revaz, E. Walker, H. Berger, and L. Farro, Phys. Rev. B 61, 9752
(2000); F. Venturini, M. Opel, T. P. Devereaux, J. K Freericks, 1. TuttO, B. Revaz,
E. Walker, H. Berger, L. Forro and R. Hackl, Phys. Rev. Lett. 89, 107003 (2002).
ORBITAL PHYSICS VERSUS SPIN PHYSICS:
THE ORBITAL-HUBBARD MODEL

Proceedings oj the ARW NATO Workshop Hvar, Croatia, October 2002

LO UIS FELIX FEINER

Institute jor Theoretical Physics, Utrecht University,
Leuvenlaan 4, NL-3584 CC Utrecht, The Netherlands
Philips Research Laboratories, Prof. Holstlaan 4,
NL-5656 AA Eindhoven, The Netherlands
ANDRZEJ M. OLES
Marian Smoluchowski Institute oj Physics, Jagellonian
University, Reymonta 4, PL-30059 Krakow, Poland
M ax-Planck-Institut jur Festkorperjorschung,
Heisenbergstrasse 1, D-70569 Stuttgart, Germany

Abstract. To elucidate the similarities and differences between orbital and spin degrees
of freedom, we analyze an orbital-Hubbard model with two orbital flavors, corresponding
to pseudospin 1/2, and contrast its behavior with that of the familiar (spin-1/2) Hubbard
model. The orbital-Hubbard model describes a partly filled spin-polarized e g band on a
cubic lattice, as occurs in ferromagnetic manganites.
We demonstrate that the absence of SU(2) invariance in orbital space has important
implications - superexchange contributes in all orbital ordered states, the Nagaoka
theorem does not apply, and the kinetic energy is enhanced as compared with the spin
case. As a result orbital-ordered states are destabilized by doping, and instead a strongly
correlated orbital liquid with disordered orbitals is realized.

Key words: Orbital-Hubbard Model, Hubbard Model, Superexchange, Orbital Liquid,

Orbital Ordering, SU(2) Symmetry, Nagaoka Theorem, Gutzwiller Factor

Recently there has been renewed interest in orbital degrees of freedom

in Mott insulators (1). In undoped perovskites such as KCuF 3, LaMn03,
LaTi0 3, and LaV0 3, strong on-site Coulomb repulsion U eliminates charge
fluctuations and replaces them by effective low-energy interactions of super-
exchange (SE) type. When the electrons occupy partly-filled degenerate e g
or t2g orbitals, it is necessary to consider the orbital degrees of freedom on
equal footing with the electron spins (2, 3). The SE interactions are there-

123
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 123–131.
© 2003 Kluwer Academic Publishers
124

fore strongly frustrated - the quantum fluctuations are then enhanced, and
might even lead to a spin liquid state in LiNi0 2 (4). Another possibility,
that an orbital liquid (OL) is stabilized and coexists with long-range spin
order, was pointed out recently for t2g systems (5). In contrast, the SE in
undoped e g systems, KCuF 3 (2) and LaMn03 (6), favors alternating orbital
(AO) order which coexists with the antiferromagnetic spin (AS) order.
In this paper we study a generic model of correlated electrons in a spin-
polarized e g band, with two orbital flavors described by a pseudospin 1/2
in the orbital Hilbert space, and consider its relation to the (spin) Hubbard
model for electrons with spin s = 1/2. We investigate: (i) in what respect
long-range order in orbital systems is different from that in spin systems,
and (ii) whether the orbitals are ordered or rather form a disordered OL.
These questions are both of fundamental nature and of immediate interest
for understanding the metallic ferromagnetic manganites (3, 7). Our pur-
pose here is to elucidate the physical mechanisms which operate in the eg
band by contrasting them with those known in spin systems.
So we consider spinless eg electrons [in a ferromagnetic spin (FS) state]
on a cubic lattice with kinetic energy

(1)

where hopping t between sites i and j occurs only for the directional orbitals
I(a; oriented along the bond (ij;, i.e., I(a; ex: 3x 2 - r 2 , 3 y 2 - r 2 , and 3z 2 -
r 2 , when (ij; is along the cubic axis Q = a, b, and c, respectively. To
describe the local electron interactions one needs to choose an orthogonal
basis for the two orbital flavors. One possibility is Ix; == x 2 - y2 and Iz; ==
(3z 2 -r 2 )/J3, called real orbitals, whereupon the local Coulomb interaction
becomes Unixniz. The disadvantage is that the expression for the kinetic
energy then takes a different form for each axis (8). We thus prefer to
use instead the basis of complex orbitals 1+; = ~ (Iz; - ilx;) and 1-; =
~(Iz; + ilx;), corresponding to "up"
and "down" pseudospin flavors, and
write the eg orbital-Hubbard model in the form

H = -~ L L [t( cI+cj+ + cLcj_) + rt( e-ixncI+cj_

a (ij)lla

+e+iXn Ci_Cj+
t )] + U '""
L.J ni+ni-' (2)
i

with Xa,b = ±27f /3, Xc = 0, and r = 1. The phase factors e±ixn are
characteristic of the orbital problem - the orbitals have an actual shape
in real space so that each hopping process depends on the bond direction.
 125

The representation in Eq. (2) has several advantages: (i) It displays

manifestly the cubic symmetry, as the transformation cl± ---7 cl±e±2i7f/3,

which corresponds to cyclic permutation of the cubic axes, leaves the Hamilto-
nian Eq. (2) invariant. (ii) It exhibits clearly the difference between the spin
case and the orbital case, and in fact allowed us to introduce the parameter
, by which one can turn the eg-band orbital-Hubbard model h = 1) into
what looks formally like a spin-Hubbard model h = 0). (iii) It shows
explicitly that rotational SU(2) symmetry for the pseudospins is absent
(2). The total pseudospin operator yz = 2:i Tiz , with Tiz = ~(ni+ - ni-), is
conserved only at , = 0 (i.e., [yz, 'H] = 0), while the terms ex: , commute
instead with the staggered pseudospin operator Yq = 2:i exp(iQ . R i )7iz ,
where Q = (]f,]f,]f).
Because the electrons interact by the local Coulomb interaction U, they
are prone to instabilities towards orbital order, which we compare with
magnetic instabilities in the spin case (9). At half-filling (n = 1) the simplest
possibility to reduce the interaction energy ex: U would be to polarize the
system completely into ferro orbital (FO) states, I<I>FO) = I1i cl ('ljJ, B) 10)
(uniform order (10)). As in the spin case, another possibility is AO order,
I<I>Ao) = I1iEA cl ('ljJA, BA)fl jE B C}('ljJB' BB)IO), i.e., with orbitals alternating
between two sublattices A and B. If the band is partly filled (n < 1), such
states must involve a coherent mixture of occupied and empty sites.
It is an important feature of the I+)-polarized (FO+) and 1+)/1-)-
staggered (AO±) complex states, that they retain cubic symmetry (11).
By contrast, the FO and AO real states, such as Ix)-polarized (FOx), Iz)-
polarized (FOz), or (Ix) + Iz) )/(Ix) - Iz) )-staggered (AOxz), with either
quasi-one-dimensional (FOz and AOxz) or two-dimensional (FOx) disper-
sion, break cubic symmetry (and are thus favored in lower dimensional
systems, e.g. FOx in a 2D square lattice (12)). This nonequivalence between
real and complex states is a manifestation of the broken SU(2) symmetry.
We focus here on the orbital-ordered states with complex orbitals (13),
which occur in Hartree-Fock (HF) approximation for large enough U (11).
In the case of the FO+ state, the electron bands split above a critical
value of U, and lead (14) to a finite order parameter T Z = (Tn of- 0, but the
mechanism of the instability is different from that known in the spin case.
At , = 0 one recovers the Stoner criterion UoN(EF) = 1 for the existence
of FS order, with the FS states becoming saturated only at larger U [see
Fig. l(a)]. By contrast, at, > 0 the FO states appear as a global property
of the band rather than as a Fermi surface instability [Fig. 1(b)]. For large
U (;2: 6t) a gap opens, only the lower band Eko = -tAk + U(~n - T E k )
Z
126

10 , 10
\

,,
\
8 \ 8
,,
-
......
::J
6
4
I
.................. .I\ .... _~;I'
6
4
2 2
(a) (b)
0 0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
n n
Figure 1. HF instabilities towards FO states (full lines) for: (a) 0, and (b) 1 ,= ,=
[dotted line shows N- 1 (w)]. Saturated FS states occur only in (a) above the dashed line.

is occupied, and
It) 2 2] 1/2
E k = [1 + ( UTz B k , (3)

where A k = Cx + cy + Cz , B~ = c;, + c~ + c; - (cxcy + cyc z + czcx ), with Cx =

cos kx , etcetera, and the sum is over the occupied part of the Brillouin zone
(BZ). Unlike in the spin case, T Z = n/2 only at U = 00, since the saturated
FO+ state is not an eigenstate of H. Thus the FO+ state resembles the AS
phase in the spin-Hubbard model.
In the AO phase one has four subbands in the reduced BZ. For large
U only the lowest two with dispersion E:t,~ = ±ltBk + U(~n - T Z Fk) are
occupied, and the order parameter, T Z = (TtEA) = -(TjEB)' is given by

(4)

rather similar to the FO case Eqs. (3), but with the interchange A k +--7
±IBk' The reason is readily recognized from Eq. (2): for FO order, the
diagonal hopping ex cJ±cj ± that gives A k is order-preserving, while the off-
diagonal terms ex I are order-perturbing and reduce T Z • For AO order this
is reversed: the off-diagonal hopping ex cJ±cj:f that gives Bk is compatible
with the order, while the diagonal one disturbs it. The similarity between
the FO and AO states at I '::::::' 1 becomes even more transparent at large U,
where, at x = 1 - n > 0,

TFO "21 { (1 - x) - (1 _3 x)2 (It)2}

U ' (5)

TAO -1 { (l-x)- 3 - 2x - 2.} (t) (6)

2 (1 - x)2 U
 127

0.0

-0.2

-0.4

-0.6
0.0 0.2 0.4 0.6 0.8 1.0
Y
Figure 2. Energies per site at n = 0.75 and U = 00: E in the KR approach for OL
(solid line), FO+ (dashed line), AO± state (long-dashed line), and ED for: the ground
state (filled squares) and excited states (empty symbols) of a single plaquette (P).

Note that a SE contribution ex (rt? /U appears also in the FO+ state,

because the off-diagonal hopping permits virtual charge fluctuations. In
the genuine orbital case (r = 1) the reduction of the order parameter by
SE is the same for FO and AO at x = 0, but at x 2: 0 it is slightly larger
for the FO phase, and surprisingly the energy per site of the FO phase is
lower than that of the AO phase. Thus near half-filling FO+ order is more
stable than AO± order at any U, because the FO phase not only gains more
kinetic energy ex -3tx than the AO phase ex -2t(x due to the difference in
band edge between -tA k and -,tBk , but also has lower SE energy (15).
Instead, AO± order dominates at larger doping x 2: 0.27 (11, 16), as a
consequence of its peculiar density of states with large weight close to the
band edges. Note that this is opposite to the spin case (r = 0), where the
Neel (AS) state has lower energy near n = 1 and the FS state takes over
above a critical doping Xc ':::::' t/2U.
Another striking difference is that the Nagaoka theorem, stating that at
U = ex) a single hole in the half-filled Hubbard model gives a spin-polarized
(FS) ground state (17), does not hold for the orbital-Hubbard model at , >
0, because it requires conservation of TZ [see, e.g., the proof in Ref. (17)].
Thus polarized states are harder to stabilize, as is illustrated explicitly by an
exact calculation for a four-site plaquette with three electrons. Whereas at
, = 0 the ground state is fourfold degenerate, corresponding to maximum
spin S = ~ as required by the Nagaoka theorem, at , > 0 it splits into
a nondegenerate ground state and three excited states (Fig. 2), none of
which can be classified by a pseudospin quantum number. One notes the
large energy gain in the ground state (from ED = -0.25t at , = 0 down to
ED ':::::' -0.44t at , = 1) when the orbitals get disordered and full advantage
128

is taken of the pseudospin non-conserving hopping.

We will now argue that indeed orbital (FO or AO) order is not robust at
r> 0 and gets replaced by a disordered (OL) phase, if one goes beyond the
HF approximation and includes electron correlation effects in the disordered
phase as well. We consider specifically the U = 00 limit, where the OL
competes with fully saturated FO (5) and AO (6) states. Needing a reliable
variational method to calculate the correlation energy, we adapted the slave-
boson approach introduced by Kotliar and Ruckenstein (KR) for the spin-
Hubbard model (19) to the orbital case. The introduction of slave bosons
(b!+ and bL for occupied, e! for empty sites) and pseudofermions Uit+ and
fL) was done for the complex {I+),I-)} orbitals, which is essential to
arrive at a gauge (cubic) invariant formulation (20). Treating the bosons in
mean-field approximation yields

Hu=oo = - LJLiAniA - ~t L [q+fit+fj+ + q-fLfj -

iA (ij)

+ ry!q+q- (e+
iXa
fl+fj - + e- ixa fLfj+)] , (7)

with niA = flAf iA , yI(i± = [x / (1 - \ ni±) )p /2, and the local constraints
(19) implemented by means of Lagrange multipliers {JLH, JLi-}' The self-
consistent solution corresponds to an OL state with \nH) = \ni-) = ~(1­
x) and Gutzwiller renormalization factor q(x) = q±(x) = 2x/(1 + x) (21).
The fermion bands, c~~ = -tq(x) [Ak±rBk]' interpolate correctly between
the uncorrelated (x c:::.' 1) and Mott insulator (x = 0) limits. Since the OL
state is incoherent (it is described by the density matrix Pi = ~ Ii at every
site), it is SU(2) symmetric: random complex or random real orbitals are
equivalent, and indeed the same correlated disordered OL state is obtained
using real orbitals (22).
The ordered states can be obtained too within the present KR slave-
boson formalism by a suitable choice of the Lagrange multipliers, e.g. JL+ =
0, JL- = -00 in Eq. (7) gives the FO+ state. Such states do not experience
any band narrowing, as double occupancy is eliminated at U = 00, and
the correlation energy vanishes (23). As a result, only the cEO
= -tA k
band (ct~ = ±rtBk bands) is (are) partly filled in the FO+ (AO±) state.
Therefor~, the bands in the OL state represent formally a (renormalized)
superposition of the FO and AO bands.
It is instructive to consider the variation with r of the total energy E
of ordered and disordered states at fixed doping (Fig. 2). In the spin model
(r = 0) the FS phase has somewhat lower energy than the disordered state
close to half-filling (24). When r is increased, EFo does not change, whereas
EAO± decreases from zero ex r, but at x = 0.25 still does not surpass the
FO+ state for r = 1. However, in spite of the band narrowing ex q(x),
 129

1.0 ......
,,,,
,
...

0.9 "
FO ,,,/
,,
,,
c 0.8 ,,
,,/ OL
,
0.7 ... ,,

0.6
0.0 0.2 0.4 0.6 0.8 1.0
Y
Figure 3. Phase diagram as function of r: OL versus real FOx(z) (full line) and complex
FO+ (dashed line) states at U = 00.

considerably more (kinetic) energy is gained in the OL state, because both

hopping channels contribute (25).
Finally, we compare at U = 00 the energies of all states, both complex
and real, varying nand f. One finds that AO states are never stable, while
FO states are stable only at small r (Fig. 3). At r = 0 the FO+ and
FOx(z) states are degenerate, but at any r > 0 the real phases have lower
energy, with FOz (FOx) being more stable at n < 0.71 (n > 0.71). The
range of FO order shrinks gradually with increasing r, and above r ': : :' 0.94
the OL phase is stable in the entire range of n. At finite U the kinetic
energy will become more dominant and favor disorder even more, except
near n ':::::' 1 where SE stabilizes real-orbital AO order (2, 3, 4, 6, 16). We
thus argue that for the eg orbital-Hubbard model (r = 1) doping induces a
crossover to the OL state at any U, supporting earlier conjectures that
such a disordered state is realized (21, 26). Indeed, the disordered OL
state provides a natural explanation why the magnons in the ferromagnetic
metallic state are isotropic. As we have shown elsewhere (27), the stiffness
constant (determined mainly by the double exchange) is proportional to
the Gutzwiller band narrowing factor q(x) and increases with hole doping
x - thus it measures the kinetic energy of strongly correlated eg electrons
released by doping.
In conclusion, the Nagaoka theorem does not apply to the orbital-
Hubbard model of correlated e g electrons, and ordered states are harder
to realize than in the spin case. This is manifest in the inverted stability
of the ordered phases with complex orbitals, with ferro (staggered) orbital
order favored at small (large) doping. Yet, the exciting suggestion that
such complex-orbital ordered states could be stable at finite doping (11)
is not confirmed, because of the inherent tendency of e g systems towards
orbital disorder due to the enhancement of the kinetic energy when SU(2)
130

symmetry is absent.

Acknowledgements

We thank P. Horsch, G. Khaliullin, D. 1. Khomskii, P. Walfle, and particu-

larly K. Rosciszewski for valuable discussions. This work was supported by
the Committee of Scientific Research (KBN) Project No.5 P03B 055 20.

References

1. Y. Tokura and N. Nagaosa, Science 288, 462 (2000).

2. K. 1. Kugel and D. 1. Khomskii, Usp. Fiz. Nauk 136, 621 (1982) [ Sov. Phys. Usp.
25,231 (1982)]; A. M. Oles, L. F. Feiner, and J. Zaanen, Phys. Rev. B 61, 6257
(2000).
3. A. M. Oles, Acta Phys. Polon. B 32, 3303 (2001).
4. L. F. Feiner, A. M. Oles, and J. Zaanen, Phys. Rev. Lett. 78, 2799 (1997).
5. G. Khaliullin and S. Maekawa, Phys. Rev. Lett. 85, 3950 (2000); G. Khaliullin, P.
Horsch, and A. M. Oles, ibid. 86, 3879 (2001).
6. L. F. Feiner and A. M. Oles, Phys. Rev. B 59, 3295 (1999).
7. A. P. Ramirez, J. Phys.: Condens. Matter 9, 8171 (1997).
8. A. Takahashi and H. Shiba, Eur. Phys. J. B 5, 413 (1998); J. van den Brink and
D.1. Khomskii, Phys. Rev. Lett. 82, 1016 (1999).
9. P. Fazekas, Lecture Notes on Electron Correlation and Magnetism (World Scientific,
Singapore, 1999).
10. Electron creation operators C!(1!J.i,tt;) depend on two angles {\bi,Bd and create e g
electrons in orbital coherent states 10i) = cos(\bi/2)e-i8ili+) +sin(\b;/2)e+ i8i li-).
11. A. Takahashi and H. Shiba, J. Phys. Soc. Jpn. 69, 3328 (2000); J. van den Brink
and D. 1. Khomskii, Phys. Rev. B 63, 140416 (2001).
12. F. Mack and P. Horsch, Phys. Rev. Lett. 82, 3160 (1999).
13. Such states would however cost Jahn-Teller energy which we neglect here, see: R.
Englman, The Jahn-Teller Effect in Molecules and Crystals (Wiley, London, 1972);
Y. Motome and M. Imada, Phys. Rev. B 60, 7921 (1999).
14. In the absence of SU(2) symmetry only the decoupling Uni+ni- ~ U((ni+)ni- +
ni+(ni-) - (ni+) (ni-)) gives complex states with finite T = (Tn.
15. The reduction of T in the FOx (FOz) state is only half of that in the FO+ state. As
the SE energy gain is halved as well, these states are unstable at finite U for n ~ 1.
For the real states in HF: Uni+ni- ~ -U((Ti+)cLci+ + c!+ci-(Ti-) - (T/) (Ti-)).
16. S. Maezono and N. Nagaosa, Phys. Rev. B 62,11 576 (2000).
17. Y. Nagaoka, Phys. Rev. 147, 392 (1966).
18. Q. Yuan, T. Yamamoto, and P. Thalmeier, Phys. Rev. B 62, 12 696 (2000).
19. G. Kotliar and A. E. Ruckenstein, Phys. Rev. Lett. 57, 1362 (1986).
20. Like the SU(2) invariant formulation for spins: R. Fresard and P. Wolfle, Int. J.
Mod. Phys. B 6, 685 (1992).
21. The nonvariational slave-fermion approximation gives a different band renormaliza-
tion ex x [So Ishihara, M. Yamanaka, and N. Nagaosa, Phys. Rev. B 56, 686 (1997)],
and underestimates the stability of the OL phase.
 131

22. The cubic invariant KR approach for real states consists in substituting bI± f--+
(bIz ± ib!x)/V2, and treating the amplitudes (biz) and (b ix ) in mean field.
23. Equivalent results are therefore obtained for the orbital ordered states by a single
slave-fermion approach.
24. At I = 0 the present method is believed to give an upper bound for the stability of
FS states;(9) they are stable below x c::: 0.33 in the cubic lattice, very close indeed to
x c::: 0.32 found for a single spin-flip in the Gutzwiller wave function [B. S. Shastry,
H. R. Krishnamurthy, and P. W. Anderson, Phys. Rev. B 41, 2375 (1990)].
25. At I = 1 and n = 0.75 the energy gain in the OL state comes close to that in the
exact ground state for a plaquette (Fig. 2).
26. R. Kilian and G. Khaliullin, Phys. Rev. B 58, Rll 841 (1998).
27. A. M. Oles and L. F. Feiner, Phys. Rev. B 65, 052414 (2002).
QUASIPARTICLES IN PHOTOEMISSION SPECTRA OF MANGANITES

Proceedings oj the ARW NATO Workshop Hvar, Croatia, October 2002

J. BALA and ANDRZEJ M. OLES

Marian Smoluchowski Institute oj Physics, Jagellonian
University, Reymonta 4, PL-30059 Krakow, Poland
M ax-Planck-Institut jur Festkorperjorschung,
Heisenbergstrasse 1, D- 70569 Stuttgart, Germany

Abstract. We compare the spectral functions of a hole moving in the orbital-ordered un-
doped LaMn03, obtained using a self-consistent Born approximation. The quasiparticle
(QP) dispersion and spectral weight depend critically on the type of orbital ordering. If
a hole scatters on orbital excitations, the QP dispersion on the orbiton energy scale is
modified by orbital polarization and by electron-lattice coupling. A lower QP dispersion
and quantum decoherence are obtained due to the inter-planar hole-magnon scattering
in the A-AF phase which disproves the classical concept of a ferromagnetic polaron.

Key words: Manganites, Orbital Ordering, Born Approximation, Spectral Functions,

Orbital Polaron, Quasiparticle Dispersion, Hole Scattering, Orbiton

1. Orbital t-J model for manganites

It is well known from the cuprates that a hole propagating in an an-

tiferromagnetic (AF) CU02 plane of a high temperature superconductor
scatters on magnon excitations and thus its spectral properties are com-
pletely different from those of a free hole. Stronger, the propagation is only
possible due to the processes in which a single magnon is either excited
or absorbed, and without them (Ising model), a hole would be confined
by a string potential (1). Here we show that although richer models are
needed to describe a propagation of a hole in LaMn03' a parent compound
for 'colossal magnetoresistance', the essential feature of such propagation,
dressing of a hole by low-energy excitations (2), remains the same.
LaMn03 is an AF Mott insulator due to strong on-site Coulomb re-
pulsion ex U, with ferromagnetic (FM) (a, b) planes which stagger along
the c-axis in the so-called A-AF phase. This kind of order is stabilized by

133
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 133–141.
© 2003 Kluwer Academic Publishers
134

the anisotropic superexchange HJ due to the charge excitations for Mn3+-

Mn3+ pairs with singly occupied degenerate e g orbitals, described within
the spin-orbital model (3). If the electron correlations could be ignored,
the hole propagation on the energy scale ex t (t is the hopping element
between Iz) orbitals along the c axis) would follow from the tight-binding
Hamiltonian H t . However, the double-exchange hinders the motion along
the c-axis (4), and a hole can move mainly within a single FM (a, b) plane.
In the present paper we analyze possible scenarios of hole dynamics when it
couples either to orbital excitations (5), or phonons (6), or to magnons (7).
First we consider a two-dimensional (2D) model for a single FM (a, b)
plane when the oxygen atoms are frozen in their optimal positions dictated
by the Jahn-Teller (JT) effect. As U» t, one has to use the strong-coupling
limit represented by the following extended orbital t-J model (8),

H = Ht + HJ + Hf':,. + H,lT + Ez· (1)

It includes, in addition to H t and HJ which give the orbital (in analogy to

spin) t-J model (5), also the polarization of orbitals around a hole (Hf':,.)
which induces the splitting of eg orbitals next to the hole (9),

Hf':,. = -~L L niT], (2)

I (ij)1IJ

with r = a, b, c labelling cubic axes, Tj(b) = -~ (Tj =f V3Tj) and Tj = Tj

given by two pseudospin operators {Tj, Tj}, and the JT interaction,

(3)

using the notation of Kanamori (10). The last term in Eq. (1) stands for
the elastic energy of the distorted lattice (E z ) (11).
The orbital ordering within the (a, b) planes is stabilized even by the
superexchange interaction alone (3, 12), and one finds alternating occupied
e g orbitals, as observed experimentally (13),

lif-L) = cos (~ - ¢) liz) ± sin (~ - ¢) lix), (4)

where + (-) refers to i E A (i E B) orbital sublattice, and lix) (liz)) stands
for the local basis orbital Ix 2 - y2) (13z 2 - r 2)) at site i, respectively. In
contrast to the spin t-J model, a single hole can now propagate coherently
without disturbing the orbital order (5).
The lattice distortion is described by three independent parameters: c5x
(c5z ) - uniform deformation along the a and b (c) cubic directions, and u
- oxygen displacement along the Mn-O-Mn bond in the (a, b) plane; all
 135

(a) (b)

(n,n) A I
j
I
I
,
I
"
(00)
-
-4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4
mit mit

Figure 1. The hole spectral functions A(k, w) for the orbital ordering given by ¢ = 0
in the absence of the electron-phonon interaction (.\ = 0) for: (a) 6 = 0, and (b) 6 = t,
as obtained along the (0,0) - ('if, 'if) direction in the 2D Brillouin zone.

distortions are in the units of the Mn-Mn distance (d = 1) (11). The JT

term (3) acts then at both sublattices as the staggered field (14),

HJT = -2A [(6x - 6z ) sin 2¢ - 2v3u cos 2¢] (LTiz - LTt), (5)
iEA iEB

with the pseudospin operators rJiz referring to the ground state, Eq. (4).
In the linear orbital-wave (LOW) theory (12) the effective Hamiltonian
represents a coupled hole-orbiton problem in momentum space,

HLOW = LEk(¢)hLh k + Lwq(¢)cx&cxq

k q

+ IN Lk,q {hLqh k [.L\!Ik,qcx& + Nk,qCX~+Q] + H.c.}, (6)

with N being the number of lattice sites and two hole-orbiton vertices,

Mk,q = 2tcos(2¢) (Uqrk-q+Vqrk) + ~cos(2¢) (rq -1) (uq+Vq ) , (7)

Nk,q = -v3t (Uq+Q7]k-q+Vq+Q7]k)-v3~sin(2¢)7]q( uq+Q+Vq+Q) ,(8)

where Q = (Jr,Jr), rq = ~(cosqx + cosqy), 7]q = ~(cosqx - cosqy), and

{uq,V q } follow from the respective Bogoliubov transformation for the or-
bital waves. The free hole, Ek(¢) = 4t~brk with t~b = t(l - 2sin2¢), and
the orbiton, wq (¢) = 3J{A(A)[A(A) + ~(2cos4¢ - 1)rq]}1/2, dispersion
depend on the type of orbital ordering ¢. Here A(A) = 1 + 2A 2 /(JK),
with K being the Mn-O spring constant. The Green's function, G(k, w) =
[w - Ek(¢) - ~(k,w)]-l, is determined in the self-consistent Born approx-
136

1.0 ~ 1.0
I:- 0.8 .A> oj(0.8
co O6
. (/v
/V<> S 0.6
~ 0.4 ~~r 0.4
oj( -

co 0.2 .¢.q(/ ta) .§ 0.2

o. 0 ~""- .......................................................---' E 0.0 ~.........................................................:a......l---'
0.0 0.4 0.8 1.2 0.0 0.4 0.8 1.2
J/t J/t
Figure 2. QP's in the orbital t-J model (~ = A = 0) as functions of J /t for ¢ = (5): °
(a) spectral weights at k = (7f,7f) (aM, full circles) and k = (0,0) (aI', empty circles);
(b) inverse effective mass m/m* and the bandwidth W* /W (full and empty circles).

imation (15) by the hole self-energy,

~(k,w) = L {M~,qG[k - q,w - wq (¢)] + N~,qG[k - q,w - wq+Q(¢)]}'

q
(9)
to obtain the spectral functions, A(k,w) = -*ImG(k,w + iO+).
In the numerical calculations we use the parameters: K = 200 eV, the
Mn-Mn effective hopping t = 0.4 eV, and the orbital superexchange J ~
50 meV, as estimated using the basic electronic parameters for LaMn03
(3). Consider first the simplest case with ,\ = ,6, = O. One finds a large
redistribution of spectral weight with respect to the free bands, a more
drastic effect than for the hole-phonon interaction (6). The spectral function
A(k, w) consists of a quasiparticle (QP) at low energy, with a minimum at
the M = (1f,1f) point, and a broad incoherent part [Fig. l(a)], similar to
the spin case (1). In the realistic regime of J It ~ 0.12, the QP spectral
weight is large at the M (aM), but almost negligible at the r point (ar)
[Fig. 2(a)].
The orbital polarization ex: ,6, enhances the hole localization and again
modifies the spectral properties. For instance, for the (I x) ± Iz) ) I orbital v"2
ordering (¢ = 0), the polarization changes the spectrum into the ladder-
like one with practically vanishing QP dispersion [Fig. 1(b)]. This leads to a
large effective mass m* 1m of a hole due to its strong confinement. Moreover,
the binding energy of the hole at k = (1f,1f) (the bottom of the QP band)
increases by energy I5E ;:::j 0.5t « 2,6, due to the polarization, suggesting
that the rearrangement of e g orbitals around the hole is small. In contrast,
the effective mass m* 1m is much less enhanced when ,6, = 0, and decreases
with increasing J [Fig. 2(b)]. At the same time the QP bandwidth W* IW
is about 2.2J at J It < 0.4, and approaches t at large J It > 1.
The changes of the QP dispersion with ,6, and ,\ become more trans-
parent by comparing the QP's obtained for two orbital orderings shown in
Fig. 3. The shape and position of the QP band strongly depends on both
 137

, ,
i:>.i:>.AA1:>.A~A+Ai:>.8.AAAAL;.i:>.+1:>.AAAL::..AAA8
-3.0
(0,0) (n,n) (n,O) (0,0) (0,0) (n,n) (n,O) (0,0)

Figure 3, Dispersion of the QP band for: (a) ¢ = 0, and (b) ¢ = -1f/12, with 6 =)., = 0
(0),6 = 0, )., = lOt (e), 6 = t, )., = 0 (D), and 6 = t, )., = lOt (0),

orbital polarization (~), and on the JT coupling (,\), With polarized e g

orbitals around a hole (~ = t) and no JT interaction (,\ = 0), the QP
band calculated for ¢ = 0 (¢ = -Jr /12) is narrowed by a factor of 50
(20), respectively, in comparison with the free hole dispersion and moves to
much lower energies, The increasing JT interaction leads to a rigid orbital
ordering and consequently to the hole motion with almost no scattering off
orbital waves (8, 16), As a result, the binding energy gets small, and the
hole undresses, resulting in large dispersion set by the hopping element t.

2. Dynamical electron-phonon coupling

To describe the dynamical JT effect we have to quantize the oscillations of

the oxygen atoms around their mean-field positions (ex: u) and derive the
dynamical part of the JT Hamiltonian of the following form (17),

KL L QT'~+2~L L Pi:1;
dyn _
H JT -
i 1;=1,2,3 i 1;=1,2,3

+ x
2V6,\ LTi (sin 2¢ Qi,2 - cos 2¢ Qi,3) , (10)
i

where M stands for the mass of an oxygen ion, with the JT phonon modes:
(11)
where Uiv (with v = x, y, z) is the oxygen vibration along the respective
Mn-O bond and Pi,1; is the conjugate momentum vector corresponding to
the lattice distortion Qi,I;' The breathing mode, Qi,l = (Uix + Uiy + Uiz), Js
does not couple to the orbital excitations. In the LOW order the total
effective Hamiltonian [derived from Eqs. (1) and (10) at ~ = 0] represents
a coupled hole-orbiton-phonon problem in momentum space,

Heff = LEk(¢)hLh k + L L n&O(¢)p~,I;Pq,~ + WQ L B~,/"Bq,/"

k q 1;=1,2 q,/"=1,3
138

(a) (b)

I I l~JHlulvV' !l! UI
-2.0 -1.0 0.0 1.0 2.0 3.0 -2.0 -1.0 0.0 1.0 2.0 3.0
mit mit

Figure 4. The local hole spectral functions A(w) obtained for the Ix 2 - z2)/l y 2 - Z2)
orbital ordering (¢ = 7[/12) of LaMn03 with: (a) Alt = 0 and (b) Alt = 8.

+ L {hLqhk[Mo (hBL + Bl,3)

k,q

+ "(
~ (i;)(3t
Mk,q q,~ + Nk,q
(i;)(3t
q+Q,i;
)]
+ H.c. } , (12)
i;=1,2

with the energy of the phonon mode Wo = J2K/M and the hole-phonon
vertex M o = J3/NA/(2KM)1/4. O&i;)(¢) and M~:~ are the energies of
mixed electron-phonon excitations and the respective vertices, which de-
pend on A and ¢ (for more details see Ref. (17)). Moreover, Bq,/L are phonon
operators representing modes which do not couple to orbitons, while (3q,i;
represent mixed orbiton-phonon excitations (18).
In the presence of both orbital ordering and lattice distortions, a hole
can scatter off both orbitons and phonons, already renormalized by each
other. The simplest situation occurs for the Ix 2 - z2) /l y 2 - z2) orbital
ordering with Gk (¢) = O. In this case, similar to a hole in a quantum anti-
ferromagnet (1), a hole can only propagate making mixed orbital-phonon
excitations on its path. Such mixed excitations have already been observed
in FM LaMn03H (19). As the orbitons and phonons having no momentum
dependence in this case, one finds k-independent spectra and the infinite
hole effective mass. In the absence the JT effect, the model reduces to the t-
jZ model (1) with ladder-like spectral functions shown in Fig. 4(a). When
the orbiton-lattice coupling ex: A increases, the QP peak moves to lower
energies and the spectrum looses gradually its coherence; for large values
of A c:::: 8t it separates into a number of subspectra consisting of a series
of lattice vibrations [Fig. 4(b)]. Each hopping of the hole (leading to an
excitation of an orbiton) is accompanied by a few vibrational quanta. This
spectral function well exemplifies the structure of the electron dressing after
a sudden photoemission process. For other orbital orderings the vibrational
peaks merge into a single broad maximum next to the QP peak.
 139

(a)

/~
_ (n,O,O)
(b) (n,0,nI2)
(c)
("::1
~.~
~
8'
d
~~
<t:: -~

-~-
~-~
~-
~----
(0,0,0)
=\~-- ~~o,O,nI2) (0,0,0)
-

-1.0 0.0 1.0 -1.0 0.0 1.0 -1.0 0.0 1.0

mit mit mit

Figure 5, The hole spectral functions obtained in the hole-magnon 3D model for the
A-AF phase with cP = 1f/24 along high symmetry lines in the 3D Brillouin zone (7),

3. Scattering on magnons in the A-AF phase and discussion

In this section we consider the limit of rigid orbital ordering (>.. » t) for
~ = O. Then the model given by Eq. (1) leads to free propagation of an eg
hole in the (a, b) plane. However, the A-AF order leads to hole scattering
on spin excitations when it hops out off FM planes. Assuming that the
orbital ordering repeats itself in (a, b) planes, the hopping part of our three-
dimensional (3D) t-J model, H = H t + H.], reads (7),

(13)

where cIa = cIa(l - ni(j). Here (ij)ab and (ij)c represent the nearest-
neighbors in the (a, b) plane and along c direction, and tt = ~t(l + sin 2¢).
The interactions between spins S = 2 of Mn3+ ions are given by:

HJ = -Jab L Si' Sj + Jc L Si . Sj, (14)

(ij)ab (ij)c

with the superexchange constants, Jab> 0 and J c > O.

The scattering off magnons strongly depends on the orbital ordering
and can considerably change the hole propagation when tt > t~b' Here, we
present the spectral functions for the orbital ordering given by ¢ = 7r /24,
where ttlt~b c:::: 5.2 and the hole-magnon coupling is strong (see Fig. 5). We
find weak QP states at small momenta with incoherent structureless con-
tinuum developing above them, which absorbs most of the spectral weight.
At larger momenta the spectra widen with thresholdlike peaks developing at
w c:::: -O.St. Consequently, the whole spectrum bears again little resemblance
with the free-hole dispersion. On the contrary, for the 13x 2 - r 2 ) /13 y 2 - r 2 )
140

orbital ordering, where ttlt~b = 0.5 the hole hardly feels the AF ordering
in the c direction and moves almost coherently in the (a, b) plane (7).
Summarizing, we have studied a single hole propagation in undoped
LaMn03 with orbital and magnetic ordering. The hole scatters strongly on
low-energy excitations: phonons, mixed orbiton-phonon, or magnons. The
orbital polarization around a carrier leads to its additional confinement.
Our results predict a large redistribution of spectral weight with respect to
the bands found in local density approximation (LDA) or in LDA+U, and
can have important implications on the angular resolved photoemission
spectroscopy (ARPES). In all cases we found low-energy QP's, but the
minimum of the QP band falls at k = ('if, 'if, kz ) when the hole-orbiton
scattering dominates, while it is at k = (0,0, 'if /2) for the hole-magnon
scattering. Which of the considered processes dominates and which disper-
sion is realized might be answered only by the ARPES experiments. In
fact, the orbitons are gapped, and were found recently at energies larger
than rv 150 meV (20). The magnons are gapless, and have much lower
energies rv 30 meV, so the hole scattering on them might dominate, but
only if the orbital ordering is close to Ix 2 - z2) /l y 2 - z2) (¢ = 'if /12) rather
than to 13x 2 - 7'2) /13 y 2 - 7'2) (¢ = -'if /12). Thus, the results of future
ARPES experiments could also help to establish which type of orbital
ordering is realized in LaMn03, the issue difficult to resolve by resonant
x-ray scattering (13). We also argue that the spectral features measured in
the highly doped La1.2Sr1.sMn207 (21) should change substantially when
magnetic order changes to the A-AF phase under doping.
This work was supported by the Polish State Committee of Scientific
Research (KBN), Project No.5 P03B 055 20.

References

1. G. Martinez and P. Horsch, Phys. Rev. B 44, 317 (1991).

2. J. Zaanen and A. M. Oles, Phys. Rev. B 48,7197 (1993).
3. L. F. Feiner and A. M. Oles, Phys. Rev. B 59, 3295 (1999).
4. C. Zener, Phys. Rev. 82, 403 (1951).
5. J. van den Brink, P. Horsch, and A. M. Oles, Phys. Rev. Lett. 85, 5174 (2000).
6. V. Perebeinos and P. B. Allen, Phys. Rev. Lett. 85, 5178 (2000).
7. J. Bala, G. A. Sawatzky, A. M. Oles, and A. Macridin, Phys. Rev. Lett. 87, 067204
(2001).
8. .J. Bala, A . .lVI. Oles, and P. Horsch, Phys. Rev. B 65, 134420 (2002).
9. R. Kilian and G. Khaliullin, Phys. Rev. B 60, 13458 (1999).
10. J. Kanamori, J. Appl. Phys. 31, 14S (1960).
11. A. J. Millis, Phys. Rev. B 53 8434 (1996).
12. J. van den Brink, P. Horsch, F. Mack, and A. M. Oles, Phys. Rev. B 59, 6795 (1999).
13. Y. Murakami et al., Phys. Rev. Lett. 81, 582 (1998).
14. J. Bala and A. M. Oles, Phys. Rev. B 62, R6085 (2000).
15. C. L. Kane, P. A. Lee, and N. Read, Phys. Rev. B 39, 6880 (1989).
 141

16. W.-G. Yin, H.-Q. Lin, and C.-D. Gong, Phys. Rev. Lett. 87, 047204 (2001).
17. J. Bala, A. M. Oles, and G. A. Sawatzky, Phys. Rev. B 65,184414 (2002).
18. J. van den Brink, Phys. Rev. Lett. 87, 217202 (2001).
19. Y. G. Pashkevich et al., unpublished (2002).
20. E. Saitoh et al., Nature 410, 180 (2001).
21. D. S. Dessau et al., Phys. Rev. Lett. 81, 192 (1998).
METALS NEAR A ZERO-TEMPERATURE MAGNETIC INSTABILITY

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

H.V.LOHNEYSEN
1 Physikalisches Institut, Universitat Karlsruhe, D-76128 K arls-
ruhe, Germany and
Forschungszentrum Karlsruhe, Institut fur Festkorperphysik,
D-76021 Karlsruhe, Germany

Abstract. Zero-temperature magnetic phase transitions exhibit an abundance of nearly

critical magnetic fluctuations leading to unusual low-temperature properties that have
become known as non-Fermi liquid (NFL) behavior. For the prototypical heavy-fermion
compound CeCu6-,rAu x this transition may be tuned by either Au concentration, hy-
drostatic pressure, or magnetic field. In the system UCu5-xPdx, strong disorder affects
the NFL properties. The d-electron Laves phase ZrZn2 was recently found to display
superconductivity in coexistence with itinerant ferromagnetism.

Key words: Heavy Fermions, Non-Fermi Liquid, Antiferromagnetic Order, Quantum

Phase Transition, Ferromagnetic Superconductor, Ce-Cu-Au, Zr-Zn, U-Cu-Pd

1. Introduction

Metals with strong electronic correlations are often close to a magnetic in-
stability. In a number of these systems, the transition temperature between
a paramagnet and a magnetically ordered phase can be tuned to absolute
zero by some externally controlled parameter such as chemical composi-
tion, pressure, magnetic field, or charge carrier concentration. This offers
the possibility to induce a T = 0 magnetic-nonmagnetic quantum phase
transition (QPT). In the vicinity of this transition, non-Fermi-liquid (NFL)
behavior (1) may occur in thermodynamic and transport properties: the
linear specific-heat coefficient I = C /T acquires an unusual temperature
dependence, often I rv -In(T/To), and the T-dependent part of the elec-
trical resistivity 6.p = P- Po where Po is the residual resistivity, often varies
as 6.p rv T m with Tn < 2, in contrast to the Fermi-liquid (FL) predictions
I = canst and Tn = 2.

143
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 143–153.
© 2003 Kluwer Academic Publishers
144

It is generally believed that the NFL behavior observed in a number

of heavy-fermion systems (HFS) at the magnetic-nonmagnetic QPT arises
from a proliferation of magnetic excitations (2, 3, 4). If the transition is con-
tinuous, it is driven by quantum fluctuations instead of thermal fluctuations
in finite-T transitions.
CeCu6-xAux provides one of the best studied examples of NFL behavior
at a QPT where macroscopic (thermodynamic and transport properties
(5, 6, 8)) as well as microscopic measurements (elastic (7) and inelastic
neutron scattering (9, 10, 11)) have been performed. This system presents
very unusual spin dynamics which will be briefly reviewed. We will also
discuss how hydrostatic pressure or magnetic field are operative in tuning
the system through a QPT. Another NFL system that has been studied
early on (12) is UCu5-xPdx. Since this system is strongly disordered, it
presents an example for a Griffiths phase scenario (13) near a magnetic
instability (14, 15). Very recently, the coexistence of ferromagnetism and
superconductivity was demonstrated for UGe2 for pressures close to the
critical pressure where ferromagnetism is suppressed (16). In addition, the
prototype weak itinerant ferromagnet ZrZn2, long considered to be a prime
candidate for p-wave superconductivity (17), was finally found to be su-
perconducting in sufficiently pure samples (18). The purpose of this article
is to review briefly the salient features of these three different systems, all
being at the verge of a magnetic instability.

2. Non-Fermi liquid behavior and magnetic fluctuations in

CeCu6-xAux

Pure CeCu6 shows no long-range magnetic order down to very low T due
to the quenching of Ce 4f magnetic moments by the Kondo effect (19, 20),
with some evidence for magnetic ordering (either electronic or nuclear)
occurring at a few mK (21, 22, 23). With r = 1.6 J jmoleK 2 it is one of the
"heaviest" HFS. CeCu6 exhibits a pronounced magnetic anisotropy with
the magnetization ratios along the three axes Me : l'vla : Mb ;:::j 10 : 2 : 1
at low T (20). We use the orthorhombic notation, in spite of the (small)
monoclinic distortion.
Already at relatively high T, i.e. around 1 K, does CeCu6 exhibit in-
tersite antiferromagnetic fluctuations as observed with inelastic neutron
scattering (INS) by peaks in the dynamic structure factor S (q, w) for energy
transfer nw = 0.3 meV at Q = (100) and (0 1±0.15 0) (24,25). The rather
large widths of these peaks correspond to correlation lengths extending
roughly only to the nearest Ce neighbors. Recently, additional features in
the a*c* plane at an energy transfer of 0.1 meV were found (26).
 145

Q
2.0 CeCu6 Au -/ '--
-x x / _\

! ~
g 1.5
z
I- 1.0

0.5 single crystals

l "'.
// 6 0
polycrystals
O'c::-'~~-"-:::-'-;:-'-~~-'---;-''-;:-'-~~-'---;-'
0.0 0.5 1.0 1.5
x

Figure 1. Neel temperature TN of CeCu6- x Au x versus Au concentration x.

x p (kbar)
• 0.1 0
,,0.2 0
• • 0.3 0
• o 0.1 6.0
.. 0.2 4.1
o 0.3 8.2

T(K)

Figure 2. Specific heat C of CeCu6-xAux (x = 0.1,0.2,0.3) plotted as CIT versus log

T at ambient pressure and at the critical pressure (x = 0.2: pc ~ 4 kbar, 0.3: pc ~8 kbar).
For x = 0.1 which is critical at ambient pressure, data for 6.0 kbar are shown.

Upon alloying with Au, the CeCu6 lattice expands (27), thus weak-
ening the hybridization between conduction electrons and Ce 4f electrons.
Hence the conduction-electron-4f-electron exchange constant J decreases,
leading to a stabilization of localized magnetic moments which can now
interact via the RKKY interaction, with ensuing antiferromagnetic order
(28). Fig. 1 shows the Neel temperature TN of CeCu6-xAux vs x. The
magnetic structure of CeCu6-xAux (0.15 :S x :S 1) as determined with
elastic neutron scattering (7, 29, 30) shows magnetic Bragg peaks in the
a*c* plane (see Fig. 3 below).
The onset of magnetic order in CeCu6-xAux is observed as a kink in
CIT, ef. Fig. 2. For the critical concentration Xc = 0.1 the linear specific-
146

heat coefficient depends logarithmically on T, CiT = a In(ToIT) with a =

0.58 J ImolK 2 and To = 6.2 K, between 0.06 and 2.5 K, instead of being
constant as for a Fermi liquid and roughly obeyed for CeCu6.
The abundance of low-energy magnetic excitations when TN is just
tuned to zero, has been suggested early on to cause the NFL behavior at the
magnetic instability (5). However, the -lnT dependence of CiT and the
unusual linear T dependence of p (not shown), also signaling NFL behavior
in CeCu6-xAux at the magnetic instability, are at variance with spin-
fluctuation theories for three-dimensional (3D) itinerant fermion systems
which predict (3, 4) CiT = /0 - (3vT and 6.p rv T 3 / 2 for antiferromag-
nets. On the other hand, 2D critical fluctuations coupled to quasiparticles
with 3D dynamics do indeed lead to the observed behavior CiT rv -lnT,
6.p rv T, and a linear dependence of TN or x on p (31).
A detailed INS investigation at the critical concentration x = 0.1 (10)
showed that the critical fluctuations as measured with an energy transfer
nw = 0.10 meV extend into the a* c* plane. Hence the dynamical structure
factor S(q,nw = 0.10meV), where nq is the transferred momentum, has
the form of rods as indicated by the shaded regions in Fig. 3. A quasi-ID
dynamic feature in reciprocal space corresponds to quasi-2D fluctuations
in real space. The 3D ordering peaks for x = 0.15,0.2 and 0.3 fall on the
rods for x = 0.1 which therefore can be viewed as precursors of 3D ordering
(7). From the width of the rods in reciprocal space, the prefactor a of the
logarithmic CIT dependence could be calculated to within a factor of two
of the experimental value (10).
The spin fluctuations also develop specific dynamics at x = 0.1 (9).
The scattering function S(q,E, T) or the susceptibility X" = S· (1 -
exp( - E IkBT)) exhibit E IT scaling (where E = nw) in the critical q region,
e.g. at Qc = (0.8, 0, 0), which can be expressed by

(1)
with 0: = 0.75 (9). The anomalous non-Lorentzian response ( 0: # 1) does
not change for other q away from the critical region (11). For all q the
susceptibility can be expressed as

(2)
In particular, the T dependence of the static uniform susceptibility
X(q = 0, E = 0) for x = 0.1 yields the exponent 0: >:::; 0.8 to a high degree
of accuracy. The simple form of Eq.(2) separates static spatial correlations
from the specific temporal correlations, the latter being independent of q.
These local fluctuations at the quantum critical point have received con-
siderable theoretical attention, although a detailed model is not available
 147

I J, d
a
II ~ IJ II
•
.--...
I r ~Ilj
::>

.
"'i::

ffliIII
'-'"

u 0 ••-m---

0 x=O.1 t
C::.Ax=O.2
••
-0.5
x=O.3
x=O.5 CeCu 6 _xAu x
-1
• x=1.0

0 0.5 1 1.5 2
a' (rlu)
Figure 3. Position of the dynamic correlations (x = 0.1, = 0.1 meV, T < 100 mK)
nw
and magnetic Bragg peaks (0.15 x s: s:
1.0) in the a*c* plane in CeCu6- x Au x . Closed
symbols for x = 0.2 represent short-range order peaks. The vertical and horizontal bars
indicate the Lorentzian linewidths for x = 0.1. The four shaded rods are related by the
orthorhombic symmetry (we ignore the small monoclinic distortion). The inset shows a
schematic projection of the CeCu6- x Au x structure onto the ac plane where only the Ce
atoms are shown. The bars in reciprocal space correspond to planes in real space spanned
by b and the lines in the inset.

yet (32, 33). The evolution of the ordered moment with increasing x > Xc
(7), may provide a valuable input to test the different models.
The onset of magnetic order in CeCu6-xAux is attributed to a weak-
ening of J because of the increase of the molar volume upon alloying
with Au as mentioned above. This is confirmed by the observation that
TN of CeCu6-xAux decreases roughly linearly under hydrostatic pressure P
(6,34). At the critical pressure Pc where TN ----t 0, i.e. Pc ~ 4 kbar for x = 0.2
and ~ 8kbar for x = 0.3, CIT exhibits NFL behavior, i.e., CIT rv -lnT,
with the same coefficients a and To as for x = 0.1 at p = 0 (see Fig. 2).
Consequently, p =6 kbar drives the ambient-pressure NFL alloy x = 0.1
into the FL regime.
An inducement of NFL behavior by a magnetic field in the related sys-
tem CeCu6-xAgx was reported previously (35, 36). The low-T properties at
the critical field are compatible with the conventional 3D antiferromagnetic
spin-fluctuation scenario (4). A detailed comparison of the specific heat and
electrical resistivity of CeCu5.sAuo.2 revealed distinctive differences depend-
148

ing on whether the QPT is tuned by B or p, presenting strong evidence

for pronounced differences in the fluctuation spectra (37). The pressure-
tuning results suggest that the strongly anisotropic fluctuation spectrum
observed for x = 0.1 at ambient pressure, which can be modeled by quasi-
2D fluctuations as discussed above, prevails. One may expect that likewise
the unexpected energy-temperature scaling of the dynamic susceptibility
survives at the QPT under pressure. On the other hand, magnetic field ap-
pears to drive the system towards a more isotropic 3D fluctuation spectrum.
Indeed, preliminary experiments suggest that the dynamical susceptibility
at the field-induced QCP exhibits a Lorentzian behavior (0: = 1).

3. Phase transitions and non-Fermi-liquid behavior in UCu5-xPdx

at low temperatures

The antiferromagnetic order of pure UCU5 below TN = 16 K is suppressed

quickly upon Cu substitution by Pd, reaching TN = 0 around x = 1
(12, 38) (see Fig. 4 for the magnetic phase diagram). Beyond x = 1.5, a
spin-glass state is found up to x = 2.3 which is the stability limit for the
cubic AuBe5 structure. Several different mechanisms have been proposed to
account for the observed NFL behavior (12) in UCu5-xPdx. One possibility
is the proximity to magnetic order, in particular for x = 1 and 1.5. On
the other hand, the strong structural disorder has been suggested to lead
to a distribution P(TK) of Kondo temperatures TK (39) mimicking NFL
behavior. Inelastic neutron scattering data on UCu5-xPdx, again for x = 1
and 1.5, revealed an unusual frequency temperature-scaling of the dynamic
magnetic susceptibility, i.e., X" (w, T) . T 1 / 3 rv (Tlw)1/3 f(wIT) (40). The
independence of X" (w, T) of the transferred neutron momentum suggests
a single-ion mechanism for this scaling behavior which was observed at
intermediate T > 10 K and tiw > 1 meV. Note that the above scaling
for X"(w, T) is similar to that in CeCu5.gAuo.l. Finally, in an attempt to
combine disorder and magnetic instability, the unusual CIT behavior and
the similarly diverging static magnetic susceptibility X(T) were interpreted
(14) in terms of a Griffiths phase scenario which was proposed to lead to
weak power-law divergences, i.e. CIT rv X rv T-!+A (13).
The proximity to a magnetic instability in UCu5-xPdx for x = 1 was dir-
ectly demonstrated through the observation of a spin-glass-like maximum
in the ac susceptibility X ac and a maximum in CIT around 0.2 K (41).
Figs. 5( a) and (b) show CIT and X ac for UCu5-xPdx vs. T on log-log
plots (15). The high-temperature data can be reasonably well described
by the algebraic T dependences suggested by the Griffiths phase model
(13), with a fitting range centered around 1 K in order to exclude the
low-T spin-glass order not explicitly treated in the model. The spin-glass
 149

• TN this work
15 - 0 o TN [3]
!1 TN [5]
v TN [12]
g10
~ • Tf this work
z
f-
5 \ 4n
o Tf
D Tf
[5]
[3]

0-
UCu 5 _xPd x IV
I
,~~----
0
0 0.5 1 1.5 2
x

Figure 4. Magnetic phase diagram T vs. x of UCu5- x Pd x . TN indicates the Neel

temperature, Tf the spin-glass freezing temperature.

x=1
A = 0.66 0.1
x=1
oCO
000

03
0.03

0.1 ]

0.03 ~

0.1

03 0.Q3

0.1 0.1 5
T(K) T(K)

Figure 5. (a) In CIT vs. In T for UCu5- x Pd x with x = 1,1.25,1.5; solid lines are fits of
the form CIT = T-l+ A . (b) In Xac vs. In T for x = 1, 1.25, 1.5; solid lines are fits of the
form X = T- 1 + A .

ordering is clearly visible as a maximum in Xac(T) for x = 1 and 1.5. The

A values obtained from the specific heat and the linear susceptibility are
in reasonable agreement with each other, with systematically larger values
for CIT. Interestingly, the A values are largest for x = 1.25, i.e., for the
concentration furthest away from the magnetic instabilities. Note that A =
1 corresponds to Fermi-liquid behavior.
A further prediction of the Griffiths phase scenario concerns the non-
linear susceptibility which should have the form Xnl ex: T- 3 +), (13). Approx-
150

imate data of Xne above the spin-glass freezing temperature are consistent
with this prediction (15).

4. Superconductivity and ferromagnetism in ZrZn2

The compound ZrZn2 is ferromagnetic (42) despite being made from non-
magnetic and even superconducting elements. The magnetic properties are
believed to derive primarily from the Zr 4d orbitals that have a significant
direct overlap (43). Ferromagnetism develops below the Curie temperature
TFl'vI = 28.5 K with an ordered moment jLs = 0.17jLB per formula unit.
ZrZn2 has a large electronic heat capacity at low temperatures, CiT ;:::::
47 mJ I molK 2, signaling the presence of many low-energy magnetic excita-
tions in addition to spin waves (44). The low TFl'vI and small ordered moment
make ZrZn2 unique among stoichiometric ferromagnetic metals and indicate
that the compound is close to a ferromagnetic QPT. This proximity has
led to numerous proposals that ZrZn2 might be a superconductor (17, 46).
For the highest-quality sample with a low- T residual resistivity of Po
= 0.62 jLOcm, we observe a rapid drop in the electrical resistivity p(T)
below Tsc = 0.29 K at ambient pressure (inset of Fig. 6), suggesting an
incomplete transition to a zero-resistance state. Application of a field of
0.2 T suppresses the drop as would be expected for a superconducting
transition. In addition, a clear diamagnetic signal in the ac susceptibility
associated with superconducting screening is observed below Tsc. For the
lowest excitation amplitudes, Xac approaches -0.65 as T ----t 0, comparable
with the ideal value of -1 (18). This diamagnetic signal rides on top of a
large ferromagnetic background.
It has been known for a long time that ferromagnetism in ZrZn2 is
rapidly suppressed under pressure P (47). Fig. 6 summarizes the effect of
pressure on TFl'vI and Tsc. p suppresses both ferromagnetism and supercon-
ductivity above a critical pressure of Pc = 21 kbar. In view of its sensitivity
to the quality of the sample, the superconductivity in ZrZn2 is likely to be
unconventional. The fact that superconductivity in ZrZn2 only occurs in
the presence of ferromagnetism where Tsc is relative insensitive against
pressure, and is hence promoted by the ferromagnetic state, may arise
naturally in scenarios where the Cooper pairs are in a parallel-spin (triplet)
state, which is already favored in the ferromagnetic state. Such behavior
could well be universal for itinerant ferromagnets in the limit of small Curie
temperature and long electron mean free path. Further work has to establish
the microscopic relation between ferromagnetism and superconductivity
and to investigate if and how the ferromagnet is affected by the onset of
superconductivity.
 151

30
_0.6
E
()
22 kbar
1.6 kbar
~0.5 mbient
20
SZ
~

I-
10

ol
D

n
u
TFM
-I'I.. .~

10*Tsc
D
a. 0.4

~O

<x~'"C,~~>------()
, 0.2 0.4
T (K)

i'- .J
0.6

a 5 10 15 20 25
P (kbar)

Figure 6. Pressure dependence of the ferromagnetic temperature T FM and of Tsc. A

few typical p(T) curves are shown in the inset. Note that Tsc for clarity is magnified by
a factor of ten.

Acknowledgements

This work reviewed here was done in collaboration with C. Pfleiderer,

A. Schroder, O. Stockert, and R. Vollmer We thank our coworkers and
colleagues for help and discussions: F. Huster, A. Neubert, T. Pietrus,
M. Sieck, U. Tutsch, M. Waffenschmidt and B. Will; G. Aeppli, N. R.
Bernhoeft, R. Chau, P. Coleman, J. Flouquet, S. M. Hayden, A. D. Hux-
ley, M. Loewenhaupt, G. G. Lonzarich, M. B. Maple, G. Miiller-Vogt, N.
Pyka, A. Rosch and P. Wolfle. We would like to acknowledge support
of the European Science Foundation within the program on Fermi-liquid
instabilities in correlated metals (FERLIN), and the Deutsche Forschungs-
gemeinschaft (DFG).

References

1. See, e.g.: Proceedings of the Conference on Non-Fermi-Liquid Behavior in Metals,

Santa Barbara (1996), published as special issue of .J. Phys.: Condo Matt., 8, 9689,
(1996).
2. .l. A. Hertz, Phys. Rev. B, 14, 1165, (1976).
3. A . .l. Millis, Phys. Rev. B, 48, 7293, (1993).
4. T. Moriya, T. Takimoto, .l. Phys. Soc. .lpn., 64, 960, and refs. therein, (1995).
5. H. v. Li:ihneysen et al., Phys. Rev. Lett., 72, 3262, (1994).
6. B. Bogenberger, H. V. Li:ihneysen, Phys. Rev. Lett., 74, 1016, (1995).
7. H. v. Li:ihneysen et al., Eur. Phys. .l. B, 5, 447, and refs. therein, (1998).
152

8. H. v. Lohneysen, J. Magn. Magn. Mat., 200, 532, (1999).

9. A. Schroder, G. Aeppli, K Bucher, R. Ramazashvili, P. Coleman, Phys. Rev. Lett.,
80, 5623, (1998).
10. O. Stockert, H. v. Lohneysen, A. Rosch, N. Pyka, M. Loewenhaupt, Phys. Rev.
Lett., 80, 5627, (1998).
11. A. Schroder et al., Nature, 407, 351, (2000).
12. B. Andraka and G. R. Stewart, Phys. Rev. B, 47, 3208, (1993).
13. A. M. Castro, G. Castilla, B. A. Jones, Phys. Rev. Lett., 81, 3531, (1998).
14. M. C. de Andrade et al., Phys. Rev. Lett., 81, 5620, (1999).
15. R. Vollmer, T. Pietrus, H. v. Lohneysen, R. Chau, M. B. Maple, Phys. Rev. B, 61,
1218, (2000).
16. S. Saxena et al., Nature, 406, 587. (2000).
17. D. Fay, J. Appel, Phys. Rev. B, 22, 3173, (1980).
18. C. Pfleiderer et al., Nature, 412, 58; ibid 412, 660, (2001).
19. Y. Onuki, T. Komatsubara, J. Magn. Magn. Mat., 63 & 64, 281, (1987).
20. A. Amato et al., J. Low Temp. Phys., 68, 371, (1987).
21. K A. Schuberth, J. Schupp, R. Freese, K. Andres, Phys. Rev. B, 51, 12892, (1995).
22. L. Pollack et al., Phys. Rev. B, 52, R15707, (1995).
23. H. Tsujii et al., Phys. Rev. Lett., 84, 5407, (2000).
24. G. Aeppli et al., Phys. Rev. Lett., 57, 122, (1986).
25. J. Rossat-Mignod et al., J. Magn. Magn. Mat., 76 & 77, 376, (1988).
26. O. Stockert et al., to be published.
27. T. Pietrus, B. Bogenberger, S. Mock, M. Sieck, H. v. Lohneysen., Physica B, 206
& 207, 317, (1995).
28. A. Germann, A. K. Nigam, J. Dutzi, A. Schroder, H. v. Lohneysen, J. Physique
CoIl., 49, C8-755, (1988).
29. A. Schroder et al., Physica B, 199 & 200, 47, (1994).
30. H. Okumura et al., J. Magn. Magn. Mat., 177-181, 405, (1998).
31. A. Rosch, A. Schroeder, O. Stockert, H. v. Lohneysen., Phys. Rev. Lett., 79, 150,
(1997).
32. Q. Si, S. Raballo, K. Ingersent, J. L. Smith, Nature, 413, 804, (2001).
33. P. Coleman, C. Pepin, Q. Si, R. Ramazashvili, J. Phys.: Condens. Matt., 13, R723,
(2001).
34. A. Germann, H. v. Lohneysen, Europhys. Lett., 9, 367, (1989).
35. K. Heuser, K-W. Scheidt, T. Schreiner, G. R. Stewart, Phys. Rev. B, 57, R4198,
(1998).
36. K. Heuser, K-W. Scheidt, T. Schreiner, G. R. Stewart, Phys. Rev. B, 58, R15959,
(1998).
37. H. v. Lohneysen, C. Pfleiderer, T. Pietrus, O. Stockert, B. Will, Phys. Rev. B, 63,
134411, (2001).
38. R. Chau and M. B. Maple, J. Phys.: Condens. Matt., 8,9939, (1996).
39. O. O. Bernal, D. K McLaughlin, H. G. Lukefahr, B. Andraka, Phys. Rev. Lett., 75,
2023, (1995).
40. M. C. Aronson et al., Phys. Rev. Lett., 75, 725, (1995).
41. R. Vollmer et al., Physica B, 230-232, 603, (1997).
42. T. B. Matthias, R. M. Bozorth, Phys. Rev., 109, 604, (1958).
43. T. Jarlborg, A. J. Freeman, D. D. Kolling, J. Magn. Magn. Mat., 23, 291, (1981).
44. N. R. Bernhoeft et al., Phys. SCL, 38, 191, (1988).
45. A. P. J. van Deusen et al., J. Magn. Magn. Mat., 54, 1113, (1986).
46. K. B. Blagoev, J. R. Engelbrecht, K. S. Bedell, Phys. Rev. Lett., 82, 133, (1998).
 153

47. T. F. Smith, J. A. Mydosh, E. P. Wohlfahrt, Phys. Rev. Lett., 27, 1732, (1971).
48. E. B. Sonin, 1. FeIner, Phys. Rev. B, 57, R14000, (1998).
49. P. Fulde, R. A. Ferrell, Phys. Rev. A, 135, 550, (1964).
50. A.1. Larkin, Y. N. Ovchinnikov, Sov. Phys. JETP, 20, 762, (1975).
HEAVILY DOPED SEMICONDUCTORS: MAGNETIC MOMENTS,
ELECTRON-ELECTRON INTERACTIONS AND THE METAL-
INSULATOR TRANSITION

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

H.V.LOHNEYSEN
Physikalisches Institut, Universitiit Karlsruhe, D-76128 Karls-
ruhe, Germany and
Forschungszentrum Karlsruhe, Institut fur Festkorperphysik,
D-76021 Karlsruhe, Germany

Abstract. Electron localization in a metal, ultimately leading to a metal-insulator trans-

ition (MIT), can occur because of disorder (Anderson transition) or electron-electron
interactions (Mott-Hubbard transition). Both effects playa role in heavily doped semi-
conductors which have become prototype systems for the study of MIT. In this review
we focus on phosphorus-doped Si. The MIT in Si:P can be tuned by varying the P
concentration or - for barely insulating samples - by application of uniaxial stress
S. The continuous stress tuning allows the observation of dynamic scaling of CT(T, S)
and hence a reliable determination of the critical exponent f-l of the extrapolated zero-
temperature conductivity CT(O) ~I S - Se 11f, i.e. f-l = 1, and of the dynamical exponent
z = 3.

Key words: Doped Semiconductors, Metal-Insulator Transition, Electron-Interactions,

Anderson Transition, Hubbard Splitting, Localized Magnetic Moments, Si:P, Thermo(electric)
Power, Critical Behaviour, Stress Tuning

1. Introduction

Heavily doped semiconductors have become prototype systems for the study
of a metal-insulator transition (MIT) driven by the combined effects of
disorder (Anderson transition) and electron-electron interactions (Mott-
Hubbard transition). In a strict sense, the MIT viewed as a transition from
extended to localized states at the Fermi level, occurs only at T = 0 because
at finite temperature T thermally activated carrier transport may occur.
Hence a metal is defined by a finite dc conductivity O"(T) for T ---+ 0, while

155
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 155–167.
© 2003 Kluwer Academic Publishers
156

for an insulator CT(O) = O. The determination of the critical exponent JL

describing how CT(O) vanishes at T = 0 as a function of a control parameter
t at a critical value t e , i.e. CT(O) "'I t - t e Ill, has been a subject of consider-
able controversy. However, using both carrier concentration N and uniaxial
stress S to tune the MIT, identifying the critical region of t where critical
behavior is expected to occur, and employing a dynamic scaling analysis,
a consistent picture is beginning to emerge.
In this review, we will discuss a few aspects of electron-electron inter-
actions and the critical behavior at the MIT focusing on three-dimensional
P-doped Si. In section 2 we treat the local magnetic moments on the metal-
lic side and evidence for a Kondo effect will be presented. The evidence for
a Hubbard gap will be discussed in section 3. In section 4 we present results
on the critical behavior of the conductivity at the MIT. In particular, we
will discuss dynamic scaling properties. A more comprehensive review can
be found elsewhere (1).

2. Metallic Si:P: formation of local magnetic moments and Kondo

effect

It has been known for a long time that a small fraction of localized moments
derived from the spin-degenerate groundstate of the hydrogen-like donor
or acceptor wave function on the insulating side, survive on the metal-
lic side of the MIT. These localized magnetic moments have been mostly
detected with magnetization (2, 3), magnetic resonance (4, 5) and specific-
heat measurements (6, 7). The dependence of the concentration of localized
moments on the P concentration N has been mapped out systematically
from specific-heat measurements for Si:P (8, 9) and more recently also for
Si:(P,B) (10).

The data for Si:P are shown in figure 1 with the overall behavior of
Si:(P,B) being quite similar. The concentration of local moments has been
determined (i) from zero-field excess specific heat 6.C = C -IT - {3T 3 , i.e.
conduction electron and phonon contributions subtracted from the meas-
ured specific heat C, via the entropy S = J(6.C / T) dT = N S ln2, and (ii)
from the Schottky anomaly observed in applied magnetic fields B = 1.5 and
5.7 T, yielding N S eh. In both cases, S = 1/2 is assumed for the spin. N sand
N Seh vary smoothly through the critical concentration N e = 3.52 ·10 18 cm- 3
and localized moments survive in the metallic state due to the disorder and
on-site Coulomb repulsion as will be discussed in detail below.
The random distribution of dopant atoms, which has been demonstrated
convincingly by scanning tunneling microscopy (12), leads to a wide dis-
tribution of nn distances and, hence, nn exchange couplings. This wide
 157

, ,

..
<:T

, ,, 00 .0
,,
00

, ,, ~ 1>.1
, ,, M

,, I>.
,,

\
en Q I>.
E , ,,
~
£10 17
,, •
-i;
en
Z
in
Z
• B=0
I>. B= 1.5T
oB=5.7T I>.

10 '6 L..----L---L....................u..LL_....................................LU
0.1 10

Figure 1. Concentration of magnetic moments N s determined experimentally in zero

field and NSch in finite magnetic fields 1.5 and 5.7T versus doping concentration. Dashed
line indicates N Sch = N expected in the dilute insulating limit. Solid line gives the
concentration of local moments NM in the metallic phase as calculated theoretically (9).

distribution of exchange couplings led Bhatt and Lee (11) to a description of

the magnetic properties of doped semiconductors in terms of a hierarchical
coupling of antiferromagnetic pairs of localized moments. In this model,
only free spins which are not coupled to pairs at a given T (with density
Njree (T)) contribute to the specific heat at that temperature with

Cl oc = kBT ·ln2 . dNjree(T)/dT (1)

while the magnetic susceptibility is given by

(2)
In order to test the validity of the Bhatt-Lee model one can calculate C loc
from Xloc with the above equations and compare it to tlC as measured
directly (8, 9). The overall good agreement found on the insulating side (2,
13) indicates the validity of the Bhatt-Lee model. However, with increasing
N 2 3.3 . 10 18 cm -3, i.e. on the metallic side of the MIT, the agreement
quickly deteriorates (13), indicating that the Bhatt-Lee model is unsuited
for the metallic state.
The origin of localized moments in the metallic state has been investig-
ated in a number of approaches (14, 15, 16, 17,9), most of them starting
with an Anderson-Hubbard model which is the simplest model Hamilto-
nian featuring the essential elements of disorder, through random hopping
integrals tij, and on-site Coulomb interaction U:

(3)
158

S2" -0.5
~
(f) -1

• B=0
-1.5 oB=1.5T
• B = 3T
6 B = 6T
-2 '-_...L..-_-'-_--'-_---'-_----'
o 0.2 0.4 0.6 0.8
T(K)

Figure 2. Thermoelectric power S vs. temperature T for different magnetic fields B

for a metallic Si:P. Solid line is the best fit of a Kondo thermoelectric power with TK =
0.8 K to the zero-field data (18).

Cia (ct) is the annihilation (creation) operator for an electron with SpIll
projection (]" in the Is(Ad groundstate of the donor atom at site i, and
nia = ctcia' It is usually assumed that the five excited states generated by
valley-orbit splitting are sufficiently far away (experimentally r-v 10 meV,
see below) that a single-band model is applicable. The host semiconductor
properties enter only through the fact that the positions i are random sites
of the Si lattice and that the effective mass and interaction are renormalized.
The various approaches differ in the type of mean-field model employed.
In the approach taken by Langenfeld and Wolfle (17) an isolated moment
("impurity") forming in an effective homogeneous medium is considered.
The calculated concentration N M of local moments is in good agreement
with the experimentally determined values Ns and NSch on the metallic
side (Fig. 1).
The coupling of the localized moments to the itinerant electrons by
an effective exchange interaction JeI I gives rise to the Kondo effect. Be-
cause of disorder, both the density of states at the Fermi level N(EF)
and the effective exchange interaction Jeff ~ [2/U where t is the average
hopping amplitude from the local-moment site to the neighboring sites,
will fluctuate. This leads to a wide distribution P(TK ) of Kondo temper-
atures T K r-v exp( -(I/N(EF )Jeff). An approximate power-law behavior
P(TK ) r-v Ti/XK is found, with aK ~ 0.9, which may account for flC in the
metallic state of Si:P (9).

The thermoelectric power S clearly gives evidence for scattering by

localized magnetic moments (18) because it is particularly sensitive to the
Kondo effect. Fig. 2 shows S(T) for a Si:P sample with N = 4.1 . 10 18 cm- 1 .
 159

The maximum of 8(T) is attributed to magnetic scattering since it is

observed only for N slightly above N c where an appreciable density of
localized moments exists. Even more convincing is the suppression of the
8(T) maximum in large magnetic fields, also shown in Fig. 2. In B = 6 T we
recover the negative diffusion thermoelectric power 8 rv - T observed for
N » N c in zero field (18). A very similar behavior of 8(T) has been found
for compensated Si:(P,B) (19). Assuming a Kondo-derived thermoelectric
power one can compare the 8(T) maximum to a corresponding single-ion
expression derived by Maki (20), d. Fig. 2. The deviations between data
and best fit might be due to the neglect of a TK distribution in the fit
where a single-valued TK = 0.8 K is assumed.
Despite a number of important differences, transport processes of com-
pensated vs. uncompensated Si:P on the metallic side not too close to the
MIT are overall similar. For instance, we have just seen that the scattering
by magnetic moments leads to a similar behavior of the thermoelectric
power in both types of materials. It has been demonstrated with infrared
reflection measurements that the MIT in Si:P occurs for electronic states
in the impurity band which even for P concentrations N :::::; 2Nc is still
energetically separated from the conduction band (21, 22). This was in-
ferred from features at 10 and 45 meV in the reflection spectra that could
be attributed to optical transitions within the valley-orbit split impurity
multiplet, i.e. from the ls(A 1 ) state to the (nearly degenerate) ls(E) or
ls(T) states, and to the conduction band, respectively. That the impurity
band is energetically separated from the conduction band at the MIT was
actually suggested long time ago by Mott, based on some more indirect
evidence (23). Again, overall similar behavior is found for the infrared
reflection of compensated Si:(P,B) (21, 22).

3. Insulating Si:P: evidence for Hubbard splitting from transport

properties

Marked differences in transport properties between Si:P and Si:(P,B) oc-

cur well on the insulating side (24). Fig. 3 shows the thermoelectric power
plotted as -8· N vs. T on a log-log plot. (Here the T range is above 1 K
so that the data should not be confused with the low-T Kondo anomaly
in metallic samples, i.e. the positive 8(T) maximum superimposed on an
overall negative diffusion thermopower with a concomitant sign change of
8 (T).) For sufficiently high T (~ 15 K) 8 is always negative. While the
-8· N curves for compensated Si:(P,B) fall on an almost universal curve
for carrier concentration N between 1.5 and 2.95·10 18 cm- 3 , such a scal-
ing is seen for uncompensated Si:P only in the concentration range above
No = 2.78.10 18 cm -3. Below that concentration 8 (T) exhibits a sign change
160

" '--'-'-~~~,':-o~-L..J40· 1 10 40
T{K1 T(K)

Figure 3. Negative thermoelectric power times carrier concentration -SN vs. tem-
perature T on a log-log plot for insulating samples of (a) Si:P and (b) Si:(P,B) (24)

Figure 4. Electrical resistivity p plotted vs. inverse temperature for (a) Si:P and (b)
Si:(P,B). Straight lines in (a) indicate fits to obtain the activation energy E 2 . Inset in (b)
shows the Si:(P,B) data plotted as log p vs T- 1 / 2 . Straight lines indicate Efros-Shklovskii
hopping (24).

at a temperature Ts=o which rapidly shifts to higher values with decreasing

N. This sign change from 5 < 0 to 5 > 0 with decreasing T is visible in
Fig. 3 as a precipitous drop of log( -5· N) with decreasing T.

A further strong difference between uncompensated and compensated

samples is seen in the electrical resistivity p(T) (Fig. 4). For N < No, p
rises much faster with decreasing T for Si:P than for Si:(P,B), while the be-
havior is similar for both types of material N > No. The strong qualitative
difference in 5(T) and p(T) in Si:P upon crossing No points to different
dominant transport processes above and below No. Interpreting the steep
p(T) increase of Si:P below No as an activated process, we can extract an
activation energy E 2 .

Fig. 5 shows a comparison of E 2 and Ts=o. Despite the order-of-magnitude

difference in absolute values of E 2 /k B and Ts=o, the similarity of the con-
 161

...............

......
15 .............. (a)
......

" 'e,
5 \
\
O'--_ _----0------01>..
...J.-_ _........_ _----L----"....

150

",,,
I I I

(b)
- -
\
~
\
- \ -
\

o I 0 I
o .~.
o 3

Figure 5. (a) Temperature Ts=o of the thermoelectric power zero and (b) activation en-
ergy E 2 vs. carrier concentration N for uncompensated Si:P (closed circles) and Si:(P,B)
(open circles) (24).

centration dependence is striking. Hence we interpret the sign change of

S(T) as the onset of an activated process. If one were forced to assume an
E 2 process also for Si:(P,B) - although the data actually suggest Efros-
Shklovskii variable range hopping (25) with an exponent p = 1/2, see inset
of Fig. 4 - one would obtain the open circles in Fig. 5, i.e. no feature appears
at No.

The sudden appearence of a hard gap only in Si:P at No well below

N c suggests that we are observing the Hubbard gap due to the on-site
Coulomb repulsion. It has been proposed already many years ago that the
E 2 process is indeed due to the Hubbard gap (26). While the negatively
charged isolated P donor is barely stable, i.e. U is of the order of the
ionization energy, it is very likely that the on-site electron repulsion weakens
progressively as the P concentration (and carrier concentration) increases,
until at No = 2.78 . 1Q18 cm -3 ;:::j 0.8 N c the two Hubbard bands start to
overlap. Because of disorder, however, metallic behavior does not incur
immediately since the tail states of the Hubbard bands are localized. It is
only at the critical concentration N c that the Hubbard bands are so close
that the chemical potential is within the range of extended states. This
scenario is schematically depicted in Fig. 6, yielding a physical picture of
how disorder and electron-electron interactions drive the MIT in Si:P. Of
162

N«N o

.. -
_ _Wll
N~Nc

occupied states localized states

Figure 6. Qualitative sketch of the density of states of the impurity band of uncom-
pensated Si:P for several P concentrations N, indicating the splitting into lower and
upper Hubbard band around the chemical potential. See text for details.

course, the Hubbard features are absent in Si:(P,B) as they should because
compensated semiconductors are away from half-filling. It is interesting to
note that the density of localized magnetic moments in Si:P is not maximal
at N c but peaks rather precisely at No where the two Hubbard bands in
our scenario are suggested to merge, see Fig. 6.

4. Critical behavior of the conductivity at the metal-insulator

transition

While early suggestions considered the MIT to be discontinuous (23), scal-

ing approaches for noninteracting electrons suggested the existence of a con-
tinuous second-order phase transition for three-dimensional systems (27)
where the de conductivity at T = 0 is expected to vary as 0-(0) rvl t - t c IfL.
Theoretically, JL is usually inferred from the correlation-length critical expo-
nent v via Wegner scaling JL = v( d - 2) where d is the spatial dimension. In
a self-consistent theory of Anderson localization (neglecting interactions),
an exponent JL = 1 for d = 3 is suggested (28). Field theoretical approaches
are discussed in Ref. (29). Values of v derived from numerical studies of
noninteracting systems range between 1.3 and 1.6 (30, 31, 32).
 163

T (K)
0.01 0.1 0.2 0.5

16 N = 3.2110 18 cm- 3

12

E
C)
----
========
g 8
b

ill
4

oo~-=~=
0.2
.....
0.4
"""==::~=:::r::::'-------J.
0.6 0.8
T1/2 (K1/2)

Figure 7. Conductivity a of a Si:P sample with P concentration N = 3.21 . 1Q18 cm -3

versus VT for several values of uniaxial stress S applied along the [100] direction. From
top to bottom: S = 3.05, 2.78, 2.57, 2.34, 2.17, 2.00, 1.94, 1.87, 1.82, 1.77, 1.72, 1.66,
1.61, 1.56, 1.50, 1.41, 1.33, 1.26, 1.18, 1.00 kbar. Solid lines are connecting the very finely
spaced individual data points.

Experimentally, uncompensated semiconductors, were reported to show

fL = 0.5, in contrast to fL = 1 generally found for compensated semiconduct-
ors and most amorphous metals (33). The exponent fL = 0.5 was largely
based on experiments (35,34,36) where uniaxial stress was used to drive an
initially insulating Si:P sample metallic. We suggested a few years ago, on
the basis of an observed crossover in concentration-tuned uncompensated
Si:P from an exponent fL ;:::j 1.3 close to N c where do-jdT > 0, to fL = 0.64
above a crossover concentration N CT ;:::j 1.1 N c where dO"/ dT changes sign,
to limit the critical region of the MIT on the metallic side to concentrations
where 0" decreases with decreasing T, i.e. to the range N c < N < N CT (37).

Fig. 7 shows the electrical conductivity O"(T) of a barely insulating sample

(N = 3.21 . 10 18 cm- 3 ) for uniaxial stress S between 1 and 3.05 kbar (38).
The data are plotted vs. y'T which is the T dependence expected due to
e-e interactions and indeed observed well above the MIT (36). The curves
are in fact polygons connecting adjacent data points. For S between 1
and 2.57 kbar the O"(T) curves evolve smoothly from insulating to metallic
behavior with m > 0, and O"(T) becomes nearly independent of T with
a value O"CT ;:::j 120- 1 cm- 1 at ;:::j 2.7kbar. Our data do not exhibit the
164

T (K)
0 0.01 0.05 0.1 0.2
6

4
E
u

S- 2
o

0
0
20

10 0

E
u
glO -1 0 1
0

0" (S - Sc) (kbar)

'6
(b)
0
3

Figure 8. (a) Low-temperature data of u of Fig. 1 for stress in the immediate vicinity of
the metal-insulator transition plotted against T 1 / a. Dashed line indicates the conductivity
at the critical stress (see text). (b) Extrapolated conductivity u (0) for T ----> 0 versus
uniaxial stress S for two P concentrations N = 3.21 and 3.43.10 18 cm -:3 (open and closed
circles, respectively). The inset shows earlier u(O) versus S - Se data (triangles) together
with the reported I-" = 0.5 (dashed line) from (35) in comparison to our data for sample
1 (circles) (38).

precipitous drop of O"(T) below rv 40 mK for B close to the MI transition,

in distinction to the earlier stress-tuning data on Si:P extending to 3 mK
(35, 34).

Closer inspection shows that the data near the MIT are actually better
described by a T 1/ 3 dependence for low T, see Fig. 8a. 0"(0) obtained from
the T 1/ 3 extrapolation to T = 0 is shown in Fig. 8b, together with data for
a sample closer to the critical concentration, yielding Be = 1.75 kbar and
1.54 kbar for 3.21 and 3.43 . 1Q18cm -3, respectively. Note that the critical
stress Be is quite well defined, as 0"(0) breaks away roughly linearly from zero
within less than 0.1 kbar, i.e. f-L ;::::; 1. This behavior contrasts with the earlier
stress-tuning data (35) reproduced in the inset of Fig. 8b, where appreciable
rounding close to N e is visible. However, those 0"(0) data between 4 and
16 n-1cm- 1 are compatible with a linear dependence on uniaxial stress.
In order to analyze the scaling behavior of 0" at finite temperatures using
the data of the sample with N = 3.21 . 1Q18cm -3, we employ the scaling
 165

SI:P
N ,,3.2110 18 cm-3

metallic

10-2

10-3

10"4 insulating

Figure 9. Scaling plot of u/u, vs. IS - s, I /ScTY for Si:P with N = 3.21·1Q18 cm -3
at different uniaxial pressures S, with Se = 1.75kbar and y = 0.34 (38).

relation (39)
(4)
where CJe(T) CJ(t e,T) is the conductivity at the critical value t e of the
parameter t (in our case S) driving the MIT. If the leading term to CJe(T) is
proportional to T X , one obtains x = Itlvz and y = I/vz from a scaling plot.
Fig. 7 shows that CJ for S close to Se does not exhibit a simple power-law
T dependence over the whole T range investigated. We therefore describe
CJe(T) by the function CJe(T) = aT X (1 + dT W ) with a = 6.01 O-lcm-l, x
= 0.34, d = -0.202, w = 0.863, and T is expressed in K. Here the dT w
term presents a correction to the critical dynamics. All CJ(S, T) curves with
1.00 kbar < S < 2.34 kbar where dCJ I dT ;::: 0 are then used for the scaling
analysis.

Fig. 9 shows the scaling plot of CJ(S,T)/CJe(T) vs. I S - Se I ISeTY.

The data are seen to collapse on a single branch each for the metallic and
insulating side, respectively. The best scaling, as shown, is achieved for
y = II zv = 0.34. Together with It = 1.0 as obtained from Fig.8b and
assuming Wegner scaling v = It for d = 3, we find z = 2.94, consistent
with CJe T 1/ z
roJ T 1 / 3 for T --+ 0 used to obtain CJ(O), see Fig.8a. It is
roJ
166

the consistent combination of both approaches, 0"(0) scaling and dynamic

scaling, that lends confidence to the results.
From an analysis of O"(T) of Si:P for different P concentrations N we
had previously inferred JL = 1.3 from 0"(0) vs. N (37) and z = 2.4 from a dy-
namic scaling analysis of metallic samples only (40), broadly consistent with
the values obtained from the present stress-tuning study. In addition, the
Mott temperature evaluated for insulating samples from the variable range
hopping conductivity, vanishes at the same concentration N c as determined
for metallic samples (41). Recently, Bogdanovich et al. (39) demonstrated
that O"(S, T) for Si:B obeys very nicely dynamic scaling on both metallic and
insulating sides, yielding JL = 1.6 and z = 2, while concentration tuning of
0"(0) on the same system had suggested JL = 0.63 (42). This large difference
is not understood at present. For Si:P, the exponents JL and z are in broad
agreement with the expectation for a noninteracting system. This is in
line with our above schematic scenario for the MIT (d. Fig. 6) where the
transition proper is driven by disorder although interactions are vital in
inducing the Hubbard splitting.

5. Acknowledgments

The work reviewed here grew out of a very fruitful collaboration with
students, post-docs and colleagues. Their contributions can be identified
from the references cited. In particular, I would like to acknowledge M.
Lakner, X. Liu, C. Pfleiderer, H. G. Schlager, C. Siirgers, T. Trappmann
and S. Waffenschmidt. I am grateful to P. Wolfle for numerous enlightening
discussions on the theoretical aspects of the metal-insulator transition.

References

1. H. v. Li:ihneysen, Advances in Solid State Physics (Ed. by B. Kramer), Vieweg

BraunschweigjWiesbaden (2000), p. 143.
2. Y. Ootuka and N. Matsunaga, J. Phys. Soc. Jpn., 59, 1801 (1990).
3. A. Roy and M. P. Sarachik, Phys. Rev. B, 37, 5531 (1988).
4. M. A. Paalanen, S. Sachdev, A. E. Ruckenstein, Phys. Rev. Lett., 57, 2061 (1986).
5. H. Alloul and P. Dellouve, Phys. Rev. Lett., 59, 578 (1978).
6. J. R. Marko, J. P. Harison, J. D. QUirt, Phys. Rev. B, 10, 2448 (1974).
7. N. Kobayashi, S. Ikehata, S. Kobayashi, W. Sasaki, Solid State Commun., 24, 67
(1977).
8. M. Lakner and H. v. Li:ihneysen, Phys. Rev. Lett., 83, 648 (1989).
9. M. Lakner, H. v. Li:ihneysen, A. Langenfeld, P. Wi:ilfle, Phys. Rev. B 50, 17064
(1994).
10. S. Wagner, M. Lakner, H. v. Li:ihneysen, Phys. Rev. B 55, 4219 (1997).
11. R. N. Bhatt and P. A. Lee, Phys. Rev. Lett., 48, 344 (1982).
12. T. Trappmann, C. Siirgers, H. v. Li:ihneysen, Europhys. Lett., 38, 177 (1997).
 167

13. H. G. Schlager and H. v. Lohneysen, Europhys. Lett., 40, 661 (1997).

14. M. Milovanovic, S. Sachdev, R. N. Bhatt, Phys. Rev. Lett., 63, 82 (1989).
15. R. N. Bhatt and D. S. Fisher, Phys. Rev. Lett., 68, 3072 (1992).
16. V. Dobrosavljevic and G. Kotliar, Phys. Rev. Lett., 71, 3218 (1993).
17. A. Langenfeld and P. Wolfle, Ann. Phys. (Leipzig), 4, 43 (1995).
18. M. Lakner and H. v. Lohneysen, Phys. Rev. Lett., 70, 3475 (1993).
19. P. Ziegler, M. Lakner, H. v. Lohneysen, Europhys. Lett., 33, 285 (1996).
20. K. Maki, Prog. Theor. Phys., 41, 586 (1969).
21. A. Gaymann, H. P. Geserich, H. v. Lohneysen, Phys. Rev. Lett., 71, 3681 (1993).
22. A. Gaymann, H. P. Geserich, H. v. Lohneysen, Phys. Rev. B, 52, 16486 (1995).
23. N. F. Mott, Metal-insulator transitions, London Taylor & Francis (1990).
24. X. Liu, A. Sidorenko, S. Wagner, P. Ziegler, H. v. Lohneysen, Phys. Rev. Lett., 77,
3395 (1996).
25. A. L. Efros and B. I. Shklovskii, J. Phys. C, 8, L49 (1975).
26. N. F. Mott and E. A. Davis, Electronic processes in non-crystalline materials, 2nd
edn. Oxford: Clarendon (1979).
27. E. Abrahams, P. W. Anderson, D. C. Licciardello, T. V. Ramakrishnan, Phys. Rev.
Lett., 42, 673 (1979).
28. D. Vollhardt and P. Wolfle, in: Electronic phase transitions, ed. by W. Hanke and
Yu. V. Kopaev, Elsevier B.V. (1992), 1.
29. D. A. Belitz and T. R. Kirkpatrick, Rev. Mod. Phys., 66, 261 (1994).
30. B. Kramer, K. Broderix, A. MacKinnon, M. Schreiber, Physica A, 167, 163 (1990).
31. B. Kramer and A. MacKinnon, Rep. Prog. Phys., 56, 1469 (1993).
32. K. Slevin and T. Ohtsuki, Phys. Rev. Lett., 82, 382 (1999).
33. G. A. Thomas, Phil. Mag. B, 52, 479 (1985).
34. G. A. Thomas, M. A. Paalanen, T. F. Rosenbaum, Phys. Rev. B, 27, 3897 (1983).
35. M. A. Paalanen, T. F. Rosenbaum, G. A. Thomas, R. N. Bhatt, Phys. Rev. Lett.,
48, 1284 (1982).
36. T. F. Rosenbaum, R. F. Milligan, M. A. Paalanen, G. A. Thomas, R. N. Bhatt, W.
Lin, Phys. Rev. B, 27, 7509 (1983).
37. H. Stupp, M. Hornung, M. Lakner, O. Madel, H. v. Lohneysen, Phys. Rev. Lett.,
71, 2634 (1993).
38. S. Waffenschmidt, C. Pfleiderer, H. v. Lohneysen, Phys. Rev. Lett., 83, 3005 (1999).
39. S. Bogdanovich, M. P. Sarachik, R. N. Bhatt, Phys. Rev. Lett., 82, 137 (1999).
40. H. Stupp, M. Hornung, M. Lakner, O. Madel, H. v. Lohneysen, Phys. Rev. Lett.,
72, 2122 (1994).
41. M. Hornung, M. Iqbal, S. Waffenschmidt, H. v. Lohneysen, phys. stat. sol. (b), 218,
75 (2000).
42. P. Dai, Y. Zhang, M. P. Sarachik, Phys. Rev. Lett., 66, 1914 (1991).
COMPETITION AND COEXISTENCE OF MAGNETIC AND
QUADRUPOLAR ORDERING

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

P. FAZEKAS and A. KISS

Research Institute for Solid State Physics and Optics,
P.O.B. 49, H-1525 Budapest 114, Hungary

Abstract. The large number of low-lying states of d- and i-shells supports a variety of
order parameters. The effective dimensionality of the local Hilbert space depends on the
strength, and kind, of intersite interactions. This gives rise to complicated phase dia-
grams, and an enhanced role of frustration and fluctuation effects. The general principles
are illustrated on the example of the effect of a magnetic field on quadrupolar phase
transitions in the Pr-filled skutterudite PrFe 4P 12 .

Key words: Quadrupolar Order, Multipolar Order, PrFe4P12, Skutterudites, Rare Earth
Compounds, Pr Compounds, Mean Field Theory, Tricritical Point, Metamagnetic Trans-
ition, Quadrupolar Phase Transition in External Magnetic Field

1. Introduction

Transition metal and rare earth compounds show a rich variety of collective
behavior: various kinds of ordered phases as well as strongly fluctuating
states (spin and orbital liquids). The basic reason is that d- and f-shells
have a relatively large number of low-energy states. Crystal field splitting
usually reduces this number (the dimensionality D of the local Hilbert
space) considerably below the free ion value, but complicated physics can
arise even from D = 3 or 4.
Let us briefly consider some examples. D = 3 is, in one interpretation,
the case of S = 1 spin models which turned out to have unanticipated
phases like the spin nematics (1). A different realization is offered by in-
teracting f-electron models based on the low-lying quasi-triplet of Pr ions
in PrBa2Cu307-6 where the nature of Pr ordering is still an open issue
(2, 3). A literal realization of D = 4 is offered by the r s ground state
of Ce ions in CeB 6 which has a rich phase diagram (4). Alternatively, we

169
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 169–177.
© 2003 Kluwer Academic Publishers
170

may think of the fourfold quasidegeneracy arising from the combination

of twofold spin degeneracy with twofold orbital degeneracy, which is the
simplest model of d-electrons which is capable of supporting either spin
or orbital order, or a combination of both. Twofold orbital degeneracy
may occur in cubic, tetragonal, and hexagonal environments, and there
are quite different versions of the eg-model. In the cubic e g model, orbital
momentum is completely quenched, but orbital order may still break time
reversal invariance in the octupolar phase (5). In contrast, the trigonal
e g states sustain permanent orbital momentum along the threefold axis;
such a model should be relevant for understanding the complex behavior of
BaVS 3 (6, 7). The possibilities of 1) > 4 models are largely unexplored, but
we should mention the 1) = 6 t2g-models of LaTi0 3. There is an intricate
relationship between orbital ordering and spin ferromagnetism which would
be remarkably difficult to explain without due consideration of the orbital
degrees of freedom (8).
As implied above, the definition of 1) is not straightforward: it is usually
higher than the degeneracy of the ground state level (which would often
be small because of small low-symmetry components of the crystal field),
but it is not so high as the free-ion value. In CeB 6 , the fourfold degenerate
r s level is well separated from the higher-lying r 7 which still arises from
the Hund's rule ground state; in PrBa2Cu307-J, a low-lying doublet and a
singlet can be lumped together to give a quasi-triplet which would have r s
character if the symmetry were cubic; but it is, in fact, only tetragonal. In
any case, the relevant dimensionality 1) = 3 is much smaller than 9 which
would be the Hund's rule value. It depends primarily on the strength of
intersite interaction, which splittings should be considered small.
The highest symmetry of the the 1) = 3 models would be SU(3), etc.
Clearly, the exact realization of a high-symmetry model is more than im-
probable, and if it were really required, we should forget about it. However,
there are indications that the domain of influence of such a seemingly arti-
ficially high symmetry point in parameter space extends over a substantial
portion of the phase diagram (7). It stands to reason that SU(1)) models
(which have symmetries connecting spin and orbital axes in Hilbert space)
are more quantum fluctuating than the pure SU(2) spin models: there are
more transverse directions to fluctuate to. For instance, the SU(4) model
on the triangular lattice has a plaquette resonating ground state (the SU (4)
version of the resonating valence bond idea (9)), and the influence of this
spin-orbital liquid state may extend to physically relevant regions of the
parameter space (10).
Limitation of space forbids us to present more than one concrete ex-
ample of spin-orbital models. We consider Pr-filled skutterudites, in par-
ticular PrFe4P12 which we model as a 1) = 4 system.
 171

2. The case of PrFe4P12

Pr-fmed skutterudites show varied behavior: PrRu4P12 has a metal-insula-

tor transition (13), PrOs4Sb12 is thought to be an exotic superconductor
(14), while PrFe4P12 remains a normal metal in the entire temperature
range studied so far. Our interest lies in PrFe4P12 which in a certain
parameter range can be characterized as a heavy fermion system with
exceptionally high electronic specific heat (11). PrFe4P12 has a phase trans-
ition at Ttr ;::::: 6.5K to an ordered phase which had first been thought to
be antiferromagnetic, but mounting evidence indicates that it is, in fact,
antiferroquadrupolar (AFQ) (11, 15, 12).
Our purpose is to model the AFQ transition by a crystal field model
of the 4f electrons. We assume that Pr is trivalent (4P). This can not
be literally true, since the driving force of the formation of a heavy band
is presumably the admixture of other valence states. However, high-field
studies show that the heavy fermion state competes with AFQ ordering
(11), so a localized f-shell description should be acceptable within, or
adjacent to, the AFQ phase. Besides, at low fields H and high temperatures
T, thermal dehybridization acts to obviate the need to consider interband
coherence effects.

2.1. THE CRYSTAL FIELD MODEL

The J = 4 manifold of Pr3+ is split by the approximately cubic crsytal

field into the r 1 singlet, the r 3 doublet, and the r 4 and r 5 triplets 1. Group
theory does not tell us the sequence of the states, but fitting the measure-
ments narrows the choice. Analyzing the anisotropy of the magnetization
curves, it was concluded that the likely possibilities are: a r 1 ground state
and a low-lying exited state r 4 (the r 1-r 4 scheme); or the r 1-r 5 scheme;
or the r 3 -r 4 scheme (11). Similar schemes were suggested for PrOs4Sb12
(14, 16).
The assumption that the low-T ordered state is AFQ, seems to speak in
favour of the r 3 -r 4 scheme, since then the ionic ground state r 3 posesses a
(permanent) quadrupolar moment. It was also pointed out that the choice of
the r 3 ground state is consistent with a symmetry analysis of the structural
distortion accompanying the AFQ ordering (12). This latter argument relies
only on the assumption of the r 3 ground state, and does not consider the
effects of the low-lying excited state. Here we show that the assumption of

1 The symmetry group is really not Oh, but the tetrahedral T h . We nevertheless use

the cubic classification, which is an approximation at zero field, but when H # 0, the
symmetry will be in any case substantially lowered.
172

the r l-r 4 scheme is also capable to account for most of the observed static
properties of PrFe4P12.
We now discuss the consequences of assuming a r l-r 4 level scheme.
Since the singlet ground state

(1)

does not carry any kind of moment, the ordered quadrupolar moment has to
be induced by intersite interactions, assuming that the local Hilbert space
contains also the triplet

Irt) }{13)+1-3)+v7(11)+1-1))} (2)

Ir~) ~ (I - 4) - 14)) (3)

lei) = }{13)-\-3)+v7(1-1)-11))}, (4)

where we have chosen the basis of quadrupolar eigenstates. Choosing the

energy of (1) as the zero, the states (2)-(4) lie at the level ~.
The possible moments in the four-dimensional local Hilbert space spanned
by (1)-(4) are given by the decomposition

(5)

Evidently, the system could support either dipolar (r 4), or either of two
kinds of quadrupolar (r 3 or r 5 ) order 2. The quadrupolar order parameters
are the same that appear in the decomposition of a purely r 4 system

(6)
i.e., they are not sustained by inter-level matrix elements 3 . - It may be
of some interest to mention that the r l-r 4 scheme does not offer the
possibility of octupolar order (but the r 3-r 4 scheme would).

2 This is a classification of order parameters which can be defined purely locally. Q # 0

order needs further discussion.

:J In contrast, matrix elements between r 1 and r 4 would bring extra possibilities of
dipolar ordering. This does not seem to be relevant for PrFe4P12.
 173

2.2. THE EFFECT OF EXTERNAL MAGNETIC FIELD

Our decomposition (5) allows to seek dipolar and/or quadrupolar ordering

in the system. Experiments give the clue that we should, in fact, look for
(antiferro)quadrupolar order. We may rather arbitrarily assume that it is
of the r 3 kind 4 , i.e., the possible order parameters are og = 31'; - J(J + 1)
and O~ = 1';: - J;. Furthermore, since the total energy expression for a pair
of sites has only tetragonal (as opposed to cubic) symmetry, we need not
assume that the og and o~ couplings would be equal, and we may seek,
say, O~-type order. Using a mean field decoupling, our problem would be
rather similar to a four-state Blume-Emery-Griffiths model.
It is a well-known feature of quadrupolar ordering that its phenomen-
ology closely imitates that of antiferromagnetic transitions, though the
underlying order parameter is non-magnetic. The phase diagram in the H-
T plane (H: magnetic field) was mapped in (11). The salient features are the
following: A sufficiently strong field applied in any direction will suppress
AFQ ordering completely. On the phase boundary, a low-field regime of
continuous transitions is separated by a tricritical point (H* : : : ; 2Tesla,
T* ::::::; 4K) from the high-field regime of first-order transitions. This change
in the character of the phase transition is shown in the field dependence of
the specific heat. The nature of the magnetization curve changes drastically
at T*. For T < T*, there is a steplike metamagnetic transition corres-
ponding to the first-order transition from the low-T ordered phase to the
disordered phase. For T* < T < Ttr(H = 0) ::::::; 6.5K, there is a kind of
a smooth metamagnetic transition, where the system crosses the second-
order part of the phase boundary. For T > Ttr (H = 0), the magnetization
curve is completely smooth. We will show that a mean field treatment of
the AFQ transition in the r l-r4 scheme accounts for these observations
quite well.
In the absence of an external magnetic field, r 3 and r 4 type order
parameters (i.e., quadrupolar moment and magnetization) are decoupled
because the former is invariant under time reversal, while the latter changes
sign. Switching on the magnetic field breaks time reversal invariance, allow-
ing that quadrupolar moment and magnetization get mixed. We can also
say that magnetic field, though it couples directly to the angular momentum
J, may also induce quadrupolar moment.
This is best illustrated by looking at the matrix which contains the
matrix elements of the crystal field, the Zeeman energy for a field in the
x-direction -hxJ x , and also a term containing the quadrupolar moment

4 Assuming Oa,y-type r,s ordering would give similar results.

174

AO§, within the basis (1)-(4)

0 -2J5/3h x
M(A h ~) = -2J5/3h x ~ + 7A (7)
, x, 0 0
(
o 0

The field couples the singlet ground state Ir 1; to the O§-moment bearing
excited state Irt;. Therefore in the presence of a magnetic field, uniform
quadrupolar moment is no longer" spontaneous". If at H = 0, we had to do
with a transition to a ferroquadrupolar state, it would be smeared out in
H i- 0, and we no longer had a phase boundary to speak about 5 . However,
for antiferroquadrupolar coupling, the appearance of the staggered quad-
rupolar moment is still symmetry breaking, and therefore a sharp phase
transition remains possible also in an external magnetic field. Therefore, if
we had no other evidence than that PrFe4P12 has sharp phase transitions
in external magnetic field, and we adopted the r l-r 4 scheme, we would
have to conclude that the ordered state could not be ferroquadrupolar, but
only antiferroquadrupolar.

2.3. MEAN FIELD RESULTS

Doing the mean field theory (3) for AFQ ordering involves diagonalizing
matrices like (7), and we do not give the details here.
Fig. 1 illustrates that an AFQ transition, though of non-magnetic nature,
may give a susceptibility which looks very much like what you expect from
an antiferromagnet. The cubic (001) and (100) directions are equivalent in
the para phase but the appearance of O§-type AFQ order makes the x-field
susceptibility appear as "transverse", while the z-field susceptibility looks
"longitudinal". Of course, instead of O§ = f; - J;
we might have chosen
J; - fj; or fj; - f;, so in a crystal one would expect an equal mixture of
the corresponding AFQ domains, and the susceptibility suitably averaged.
The experiments may be taken to correspond to this.
Fig. 2 (left) gives the phase diagram. The crystal field splitting ~ and
the quadrupolar coupling A were chosen so as to get at least a rough
numerical agreement with the experimental phase diagram (11). There
is still some freedom in the parameters, but we found that a rather low
~ >:::; 4 - 6K has to be chosen (with z = 8, A >:::; 0.08k B ), if we want to get
both Ttr(H = 0) and the tricritical point right. These estimates are likely

5 Purely in symmetry terms: applying a field in one of the cubic (100) directions, the
symmetry would be lowered to C 4 h, and the decomposition of r 4 of Oh in terms of the
irreps of C 4h would contain the identity which also comes from r 1 ground state.
 175

H//<100>
0.61- ~

04
X

0.2

H//<OOI>

Figure 1. The temperature dependence of the magnetic susceptibility. The onset of

()~-type quadrupolar order makes the two H II z and H II x behavior inequivalent.

H//<100>

0.5
CIT

2 3 4 5 6 7 o 6
TIKI

Figure 2. Left: The boundary of the antiferroquadrupolar phase in the H-T plane.
The curve is drawn in black for first-order transitions, and in grey for continuous phase
transitions. Arrow indicates the tricritical point. Right: The T-dependence of the tem-
perature coefficient of the specific heat for H = 0, 2.5, and 4Tesla (in order of decreasing
transition temperatures).

to be subject of some revision when further (especally dipolar) couplings

are allowed for. - In spite of an overall similarity to the phase diagram
based on experiments, we note that the low-T, high-H up curving part of
our present phase boundary represents a deviation, the reason for which
remains to be clarified.
176

T~2K T~4.2 K
T~7K
11//<100> IIII<J(X»
11//<100>

II II II

Figure 3. Antiferroquadrupolar order underlies the sharp metamagnetic transition at

low T (left). With increasing T, we pass through second order transitions (middle) to
smooth behavior (right).

The changeover to a regime of first-order transitions in higher fields is

evident in the field dependence of the specific heat; the curves shown in
Fig. 2 (right) bear a close resemblance to the measured ones. The same is
true of the magnetization curves (Fig. 3) where we see a change from the
regime of sharp metamagnetic transitions at low temperatures to continu-
ous phase transitions at intermediate T's, and eventually smooth behavior
in the para phase.

Acknowledgements

The authors profited much from cooperation with K. Penc, L. Farro, G.

Mihaly and 1. Kezsmarki, and from discussions with H. Shiba. This work
was supported by the grants OTKA T038162, and AKP 2000-123 2,2.

References

1. C.D. Batista, G. Ortiz, and J.E. Gubernatis, Phys. Rev. B65, 180402(R), (2002),
and references therein.
2. A.T. Boothroyd, J. Alloys and Compounds 303-304, 489, (2000).
3. A. Kiss and P. Fazekas, cond-mat/0212345, and to be published.
4. R. Shiina, H. Shiba, and P. Thalmeier, J. Phys. Soc. Japan 66, 1741, (1997).
5. A. Takahashi, H. Shiba, J. Phys. Soc. Japan 69, 3328, (2000).
6. G. Mihaly et aI., Phys. Rev. B61, R7381, (2000); G. Mihaly et aI., at this conference.
7. K. Penc et aI., to be published.
8. P. Fazekas, Found. Phys. 30, 1999, (2000).
9. P.W. Anderson, Mat. Res. Bull. 8, 173, (1973); P. Fazekas and P.W. Anderson, Phil.
Mag. 30, 423, (1974).
10. K. Penc, M. Mambrini, P. Fazekas, and F. Mila, cond-mat/0212211.
11. Y. Aoki et aI., Phys. Rev. B 65 064446, (2002); J. Phys. Chern. Sol. 63, 1201, (2002).
12. S.H. Curnoe, K. Ueda, H. Harima, and K. Takegahara, J. Phys. Chern. Sol. 63,
1207, (2002); S.H. Curnoe, H. Harima, K. Takegahara, and K. Ueda ,Physica B
312-313, 837, (2002).
 177

13. C. Sekine, T. Uchiumi, 1. Shirotani, and T. Yagi, Phys. Rev. Lett. 79, 3218, (1997).
14. E.D. Bauer et al., Phys. Rev. B 65, 100506(R), (2002); K. Miyake et al.,
unpublished; D.E. MacLaughlin et al., Phys. Rev. Lett. 89, 157001, (2002).
15. K. Iwasa et al., Physica B 312-313, 8:34, (2002).
16. Y. Aoki et al., J. Phys. Soc. Japan 71, 2098, (2002).
LOW DIMENSIONAL SPIN SYSTEMS IN HIGH MAGNETIC
FIELDS: SPIN-PHONON INTERACTION

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

B. LUTHI and B. WOLF

Physikalisches Institut, Universitiit Frankfurt, D-60054 Frank-
furt
S. ZHERLITSYN
Forschungszentrum Rossendorf, PO Box 51 01 19, D-01314
Dresden

Abstract. We discuss ESR- and ultrasonic experiments for quasi- one and twodimensional
spin-systems. We found several effects for (VOhP207 and SrCu2(B03h. With ESR we
determined the magnetic excitation spectrum and with ultrasonic measurements we were
able to follow these excitations to lowest frequencies. The spin-phonon interaction is of
the exchange-striction type and with ultrasonic experiments we are able to determine the
strain dependence of the magnetic coupling constant.

Key words: Low Dimensional Spin Systems, Pulse Field Experiments, Spin Phonon
Interaction, Ultrasonic Experiments, ESR Experiments, Magnetization Plateaus

1. Introduction

It is possible today to generate pulsed magnetic fields up to 60T in coils with

20 mm inner diameter and of special design. In our Frankfurt laboratory we
specialized on developing measuring techniques for electron spin resonance
(ESR) and ultrasonics in such high fields (1, 2). For the ultrasound we used
a quadrature system capable of measuring simultaneously phase and amp-
litude in the frequency range of 10-200 MHz. From these data we calculate
the change in sound velocity and the relative ultrasonic attenuation.
Such experiments (ESR and ultrasound) are very important for high
magnetic fields, because they constitute the only spectroscopic tool for
B > 20T, where neutron scattering experiments cannot be performed so
far. In this article we concentrate on experiments on low dimensional spin

179
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 179–187.
© 2003 Kluwer Academic Publishers
180

systems. There are several reviews dealing with this topic (2, 3, 4). Here
we analyze in some detail the interaction between magnetic excitations and
sound waves for the different spin systems investigated.
We discuss one- and two-dimensional dimerized spin-l/2 compounds,
such as (VOhP207 and SrCu2(B03h. Here V4+ and Cu2+ have spin-
1/2 and form either chains or two dimensional dimer arrays. The lowest
energy excitations are usually of the singlet-triplet type or modifications
of it. These can be determined either by ESR excitations from the ground
state or within the excited triplet. For B = 0 inelastic neutron scattering
give the excitation spectrum for the whole k-space.

2. Spin-Phonon interaction in low dimensional spin systems

2.1. COUPLING MECHANISM

Usually the spin-phonon interaction for transition metal ions is of the single
ion magnetoelastic type. The strain component couples to the electric quad-
rupole moment of the magnetic ion. However for spin 1/2 systems, involving
e.g. Cu2+ or V 4 + and for lower point symmetry than cubic (where the
orbital moment is quenched) this magnetoelastic interaction is negligible.
In this case two ion interactions of the form of exchange striction becomes
important. A soundwave or more generally a phonon can modulate the
exchange interaction thus creating a spin-phonon coupling as given by

E exs = ~iJlJ(c5 + Ui - Uj) - J(c5)]Si • Sj . (1)

Here c5 = R i - R j measures the distance between two magnetic ions and

Ui is the displacement vector for the ion R i . With a sound wave given by a
plane wave U = U e ei(q.r-wt) with e the polarization vector, U amplitude,
w frequency of the sound wave the exchange striction term reads:

(2)

In eq.(2) we have expanded the exponential to first order because qc5 «

1 for sound waves of typically 100 MHz. Eq.(2) tells us that shear waves
propagating along symmetry directions do not couple to the spins with
this mechanism whereas longitudinal waves do. This is seen in many ex-
periments discussed below. It is worth pointing out that this coupling is
also responsible for sound wave anomalies (attenuation and velocity) at
magnetic phase transitions for magnetic ions with orbital moment L = 0
(Mn2+, Fe3+, Gd3+)(5).
 181

2.2. TEMPERATURE AND MAGNETIC FIELD DEPENDENCE OF

ELASTIC CONSTANTS IN DIMERIZED SPIN SYSTEM

One can describe the effective elastic constant in such systems by consid-
ering the strain susceptibility Xd for an isolated dimer and introducing the
dimer-dimer interaction Kin RPA- approximation. One gets in this way
for a symmetry elastic constant:

(3)
Here N is the number of dimers per unit volume and G is a strain-
dimer coupling constant. For a simple singlet-triplet dimer we have for the
free energy F = -kTlnZ with Z = 1 + 3exp( -fl/kT) and fl the singlet-
triplet splitting. For Xd we get with G = dfl/dcxx: Xd = G- 2d 2F/dc 2 =
3C b./kT /[kTZ 2 ].
In the next section we will give some examples where such effects have
been observed. Note that Xd is the strain susceptibility analogue of the
magnetic susceptibility. Analogously to the case of the temperature de-
pendence we can calculate the field dependence of the symmetry elastic
constant. Instead of the free energy used above we have to take the field
splitting of the excited triplet state into account: F = -kTlnZ with Z =
1 +exp( (-fl + gP,BB)/kT) + exp( -fl/kT) + exp(( -fl- gP,B)/kT). Since in
the systems investigated fl ~ 30-60K and T < 2K, we have only to consider
the lowest triplet level which cuts the singlet branch for gP,BB c = fl. With
this simplification we get for the single dimer strain susceptibility

and the elastic constant c[' is given again by eq.(3) with Xd taken from
eq.(4). The coupling constant G is given again by G = dfl/dc.
There is another way to calculate the field dependent elastic constant us-
ing the Landau theory of phase transitions (6). For a longitudinal wave with
strain cxx one can write for the spin-phonon interaction F me = G' cxxm;
with G' rv dJ/ dcxx. One gets with Cn = d2 F / dc;x with F = FL + Fme the
expression (here F L is Landau-order parameter free energy)

(5)
with X(B) the differential magnetic susceptibility and m/m o the re-
duced magnetization. For (VOhP207 a comparison of the magnetic field
dependence of the elastic constant we are using either the thermodynamic
expression (eq.4) or the Landau formula (eq.6), the latter gives a much
better fit.
182

2.3. THERMODYNAMIC TREATMENT VERSUS EQUATION OF MOTION

Above we have given a thermodynamic treatment of the temperature and

field dependence of the symmetry elastic constant. This treatment should
be applicable if one is in the hydrodynamic limit, i.e. WT << 1, with W the
sound wave frequency and T a typical spin fluctuation relaxation time. In
the opposite limit WT 2': lone has to take the coupled equations of motion
for the magnetic excitations and for the sound waves and to calculate the
sound velocity change from them. Close to the crossover of the sound wave
frequency W = vk and the magnetic excitation W m one expects a coupled
oscillator formula

(W - wm)(w - vk) = f(w) (6)

From sound wave velocity and attenuation experiments one can estimate
the value of WT (7). For WT < lone obtains

a = -T' (w2/2vO) . (~c/co) (7)

Here a denotes the attenuation and ~c/ Co the relative elastic constant
change. From eq.(7) one can determine the spin fluctuation relaxation time
and therefore one can determine the regime (hydrodynamic(wT < 1) or
non-hydrodynamic (WT > 1)), in which the experiment is performed.

3. Examples for Spin-Phonon Effects:

In the following we discuss some compounds where ESR and ultrasonic

measurements have been performed in high magnetic field. We will dis-
cuss also the temperature dependence of the elastic constants and some
attenuation data.

3.1. DIMERIZED SPIN 1/2 CHAIN:

(VOh P2 0 7
This compound exists in two modifications, an ambient pressure phase
and a high pressure phase (8). ESR and ultrasonics have been performed
only in the former phase, therefore we concentrate on this one. A review on
this compound has been given recently (9). It consists of strongly dimerized
chains along the c-axis in alternate layers along the b-axis. These two chains
have energy gaps at the r point of 35K and 60K respectively. The gaps
have been determined with inelastic neutron scattering (10) and with ESR
(11, 3). From the latter experiment we get by extrapolation a critical field
value of Be = 26T, where the lower triplet branch meets the singlet ground
 183

0
0

-2
-2
.".
'0
..--
-4
~

~ L
2. C11 <0
>~ -6 143 MHz -4 >~
...:::] ...:::]

-8
-6
a.) b.)
-10
0 10 20 30 40 0 10 20 30 40 50
B (T)

Figure 1. a: theoretical curves of the sound velocity for VOPO using eq.5 and the
same temperatures as in fig.lb. b: field dependence of the longitudinal sound velocity in
VOPO for temperatures from 1.6K to 11.0K measured in pulsed fields

state. Precisely around this field value we observe a sound velocity anomaly
as shown in fig.1.
The velocity anomaly observed is of the order of t1v/v o ;:::j 5 . 10- 4 .
This rather small value for the Cll mode is due to the small exchange (and
correspondingly small dJ/ de) in the a-direction. For C33 along the c-axis we
would expect a much larger effect, because in this case the propagation is
along the dimerized chain. Nevertheless we have measured a distinct tem-
perature dependence of this elastic anomaly. The effect being largest at low
temperature of 1.6K and disappearing already at 11K. We notice that eq.(5)
describes the experimental data of fig.1 much better than eq.(4). As seen
from a comparison of fig.1a and 1b the shift of the sound velocity minima to
higher fields for higher temperatures is well described by fig.1a and eq.(5).
Preliminary sound attenuation data corroborate our thermodynamic data:
Using eq.(7) we obtain WT ;:::j 0.1 so this approach is justified.
184

3.2. PLATEAU SUBSTANCE:

SrCu2(B03)2
This substance forms two dimensional orthogonal Cu- dimers as shown
in fig.2a. High field magnetization data exhibit clear plateaus at low tem-
peratures for m/m o = 1/8,1/4,1/3 (12, 13) as shown in fig.2c. It was
shown theoretically that this dimer arrangement corresponds to the Shastry
- Sutherland model, with an exact singlet ground state wavefunction (14).
The plateaus were calculated using two methods. One was the model of the
square unit cells and the second was a hard core boson model (14).
It is interesting to notice that one can describe the temperature depend-
ence of the magnetic susceptibility with an RPA-formula Km = x o /(l- jxo)
with XO being the magnetic susceptibility of a single dimer and j being the
interdimer coupling (15). The fit is very good and j = 273K gives exactly
4 times the dimer-dimer exchange J'indicated in fig.2a. In a similar way
we can interpret the temperature dependence of the elastic constants (16).
The polarization and propagation directions for the different modes (cn,
C44, C66) are indicated in fig.2a. Note that we obtain an exchange striction
coupling also for the shear wave C66 because the propagation direction
makes an angle of 45 degree with the dimer axis. We can interpret the
cij(T) using eq.(3). Using the lowest singlet-triplet splitting as 6. 1 = 30K,
the same value used for the magnetic susceptibility fit, we obtain a very
good fit for Cn (15, 16). For the C66 mode (polarization and propagation in
the dimer plane), which exhibits the strongest temperature dependence, we
have to include a further coupling to the next higher triplet level 6. 2 = 58K
(taken from inelastic neutron scattering) (17). Again we get a very good
fit with coupling parameters as listed in ref. (16). The C44 mode exhibits a
very small temperature dependence because the polarization vector points
along the c-axis out of the dimer plane and the exchange striction coupling
is therefore of higher order. Nevertheless we can fit also this mode satis-
factorily (16). In ref. (16) we have given a table with the different coupling
constants G 1 , G 2 , K for the three modes Cn, C44 and C66. It is seen that K
is ferrodistortive for Cn and antiferrodistortive for C66.
We turn now to the magnetic field dependence of the elastic modes.
ESR experiments determined the lowest singlet-triplet transition (18). We
extended these measurements to lower frequencies and found still a linear
field dependence down to 55GHJI. In fig.2b,c we show the 3 elastic modes
together with the magnetization taken from ref. (13). Similar to the tem-
perature dependence the C66 mode exhibits the strongest field effect, a total
softening of more than 25%, whereas the C44 mode shows no softening at all.
We observe clear minima for Cn and especially for C66 exactly at the field
values where new plateaus appear, indicated by dashed lines in fig.2b,c.
These plateaus occur at 27T, 36T and 43T for m/m o = 1/8,1/4 and 1/3.
Note that the next plateau 1/2 should occur high above 60T according
theoretical predictions.
 0,01------

o 10 20 30 40 50
B(T)

Figure 2. a: The polarization and propagation directions for the different modes (Cll'
C44, C66) together with the two coupling constants J and J'.b : The field dependence of
the elastic modes Cll, C44 and C66 at 1.5K. c : The magnetization taken from ref(13). The
broken lines indicate the the positions of the different plateaus

How can one interpret these results of fig.2b,c? Attenuation measure-

ments have been performed for the Cn mode. With eq.6 we get WT rv 1 which
means that one can no longer take thermodynamic expressions, eqs.3,4, but
one has to calculate sound velocity changes from the coupled equations of
motion, eq.6. In this case we have therefore a resonant interaction between
the sound waves and the magnetic excitations, expressed by the coupled
dispersion expression of eq.6. The important C66 mode has B 2g -symmetry
for this tetragonal compound. Therefore the strong coupling to the mag-
netic excitation for the plateaus 1/4 and 1/3 implies that these magnetic
modes have the same B 2g -symmetry. Consequently the condensed triplet
states in the plateaus should have also the same order parameter symmetry.
Therefore we can select from the calculations of this triplet condensation
the one with the same symmetry. For the 1/4 plateau it is the calculated for
the square unit cell (14). For the 1/8 plateau Cn shows also a considerable
effect. We therefore conclude that a mixture of BIg and B2g holds for this
plateau.
186

4. Conclusion and outlook

This short review shows that the combination of ESR and ultrasonics gives
an efficient spectroscopic tool for soft magnetic modes. This spectroscopy
was applied in field regions where other experimental tools like elastic
and especially inelastic neutron scattering no longer work (> 15T). ESR-
and ultrasonic measurements were also performed on the low dimensional
spin-plateau systems NH4CuCb and CsCuCl 3 (21, 23, 22). But our ex-
periments have a much wider range of applications. We have performed
ultrasonic experiments to determine the B-T -diagram for a mixed valence
compound Ybln1-xAgxCU4(24) and the high field ultrasonic investiga-
tion of the heavy fermion compound URu2Si2 (25). Other pulsed field
experiments with ESR and ultrasonics are possible, e.g. an investigation of
semiconductors, especially heterostructure materials exhibiting quantum
Hall effect.

Acknowledgements

The crystals for these investigations were given to us by A. V. Prokofiev and

W. Assmus ((VOhP207) and H. Kageyama and Y. Ueda (SrCu2(B03h).
We thank H. Schwenk and S.Schmidt for the collaboration.

References

1. B. Wolf, B. Liithi, S. Schmidt, H. Schwenk, M. Sieling, S. Zherlitsyn, 1. Kouroudis,

Physica B 294-295, 612 (2001).
2. B. Wolf, B. Liithi, S. Zherlitsyn, in: Very High Magnet-Ie Fields and their' Applica-
tions, ed. by F.Herlach and N.Miura, World Scientific.
3. B. Liithi, B. Wolf, S. Zherlitsyn, S. Schmidt, H. Schwenk, M. Sieling, Physica B
294-295, 20 (2001).
4. B. Wolf, S. Zherlitsyn, S. Schmidt, B. Liithi, Phys. stat. sol. (a) 189, 389 (2002).
5. B. Liithi in: Dynamical properties of Solids, Vol.3, North-Holland (1980).
6. B. Wolf, S. Schmidt, H. Schwenk, S. Zherlitsyn, B. Liithi. J. Appl. Phys. 87, 7055
(2000).
7. B. Liithi, W. Rehwald in: Structural Phase Transitions I, Topics in Current Physics
(1981).
8. .lVI. Azuma, T. Saito, Y. Fujishiro, Z. Hiroi, M. Takano, F. Izumi, T. Kamiyama,
T. Ikeda, Y. Narumi, K. Kindo,Phys. Rev. B 60, 10145 (1999).
9. D.C. Johnston, T. Saito, M. Azuma, M. Takano, T. Yamauchi, Y. Ueda, Phys. Rev.
B 64, 134403 (2001).
10. A.W. Garrett, S. E. Nagler, D.A. Tennant, B.C. Sales, T. Barnes, Phys. Rev. Lett.
79, 745 (1997).
11. A.V. Prokofiev, F. Biillesfeld, W. Assmus, H. Schwenk, D. Wichert, U. Low, Eur.
Phys. J. B 5, 313 (1998).
 187

12. H. Kageyama, Yoshimura, R. Stern, N.V. Mushnikov, M. Kato, K. Kosuge,

C.P. Slichter, T. Goto, Y. Veda, Phys. Rev. Lett. 82, 3168 (1999).
13. K. Onizuka, H. Kageyama, Y. Narumi, K. Kindo, Y. Veda, T. Goto, J. Phys. Soc.
Jpn. 69, 1016 (2000).
14. S. Miyahara and K. Veda, Phys. Rev. B 61, 3417 (2000).
15. S. Zherlitsyn, S. Schmidt, B. Wolf, H. Schwenk, B. Liithi, K. Onizuka, Y. Veda,
K. Veda, Phys. Rev. B 62, R6097 (2000).
16. B. Wolf, S. Zherlitsyn, S. Schmidt, B. Luthi, H. Kageyama, Y. Veda, Phys. Rev.
Lett. 86, 4847 (2001).
17. H. Kageyama, M. Nishi, N. Aso, K. Onizuka, T. Yosihama, K. Nukui, K. Kodama,
K. Kakurai, Y. Veda, Phys. Rev. Lett. 84, 5876 (2000).
18. H. Nojiri, H. Kageyama, K. Onizaka, Y. Veda, M. Motokawa, J. Phys. Soc. Jpn.
68,2906 (1999).
19. H. Tanaka, W. Shiramura, T. Takatsu, B. Kurniawan, M. Takahashi, K. Kamishima,
K. Takizawa, H. Mitamura, T. Goto, Physica B246-247, 230 (1998).
20. B. Kurniawan, H. Tanaka, K. Takatsu, W. Shiramura, T. Fukuda, H. Nojiri,
M. Motokawa, Phys. Rev. Lett. 82, 1281 (1999).
21. S. Schmidt, S. Zherlitsyn, B. Wolf, H. Schwenk, B. Luthi, H. Tanaka, Europhys.
Lett. 53, 591 (2001); addendum 54, 554 (2001).
22. B. Wolf, S. Zherlitsyn, S. Schmidt, B. Luthi, Europhys. Lett. 48, 182 (1999).
23. S. Schmidt, B. Wolf, M. Sieling, S. Zvyagin, I. Kouroudis, B. Luthi, Solid state
Comm. 108, 509, (1998).
24. S. Zherlitsyn, B. Luthi, B. Wolf, J.L. Sarrao, Z. Fisk, V. Zlatic, Phys.Rev.B 60,
3148 (1999).
25. B. Wolf, S. Zherlitsyn, H. Schwenk, S. Schmidt, B. Luthi, J.Magn.Magn.Mat. 226-
230, 107 (2001).
'TAXONOMY' OF ELECTRON GLASSES

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

N.P. ARMITAGE, E. HELGREN and G. GRUNER

Department of Physics and Astronomy,
University of California, Los Angeles, CA 90095

Abstract. We report measurements of the real and imaginary parts of the AC conduct-
ivity in the quantum limit, hw > kBT of insulating nominally uncompensated n-type
silicon. The observed frequency dependence shows evidence for a crossover from inter-
acting Coulomb glass-like behavior at lower energies to non-interacting Fermi glass-like
behavior at higher energies across a broad doping range. The crossover is sharper than
predicted and cannot be described by any existing theories. Despite this, the measured
crossover energy can be compared to the theoretically predicted Coulomb interaction
energy and reasonable estimates of the localization length obtained from it. Based on a
comparison with the amorphous semiconductor NbSi, we obtain a general classification
scheme for electrodynamics of electron glasses.

Key words: Long Range Interaction, Electron Glass, Silicon, Si:P, NbSi, Electrodynamics,
Localization, Insulator, Disorder

1. Introduction

Strong electronic interactions are known to playa central role in disordered

solids, of which Coulomb glasses are a canonical example. The lack of
metallic screening on the insulating side of the metal-insulator transition
(MIT) enables long-range Coulomb interactions (1). Efros and Shklovskii
(ES), following the original considerations for the non-interacting Fermi
glass case of Mott (2), derived a form for the T = 0 K photon assisted
frequency dependent conductivity describing the crossover from interact-
ing Coulomb glass-like behavior to Fermi glass-like behavior (3). These
derivations were based on a theory of resonant absorption (4) and take into
account the mean Coulomb interaction between two sites forming a resonant
pair U(r w ) = e 2 /Glr w , where r w = ~[ln(2Io/nw)] is the most probable hop
distance between pairs and Gl is the dielectric constant. The real part of
the ES crossover form for the frequency dependent conductivity is:

189
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 189–197.
© 2003 Kluwer Academic Publishers
190

(1)
where (3 is a constant of order one, go is the non-interacting single particle
density of states (DOS), 10 is the pre-factor of the overlap integral and ~
is the localization length. The concentration dependent localization length
is predicted to diverge as (1 - x/xc)-V as the MIT is approached, where x
is the dopant concentration, Xc is the critical dopant concentration of the
MIT (xc = 3.52 x lO I8 cm- 3 in Si:P (5)) and v is the localization length
exponent.
Neglecting logarithmic factors, Eq. (1) predicts a gradual crossover from
linear to quadratic behavior as the incident photon energy exceeds the inter-
action energy of a typical charge excitation. For the case where the photon
energy, fiuJ > U(r w ), one recovers the quadratic frequency dependence, plus
logarithmic corrections, that Matt originally derived for the non-interacting
Fermi glass case (2). In the opposite limit, fiuJ < U(r w ) the conductivity
shows an approximately linear dependence on frequency, plus logarithmic
corrections, and the material is called a Coulomb glass. We should note
that Eq. (1) was derived for the case where fiuJ > ,6" the Coulomb gap
width. However a quasi-linear dependence (albeit with a different pre-
factor) and an eventual crossover to Matt's non-interacting quadratic law
is still expected even for the case where fiuJ < ,6"

1 WeI
CTI ;::::: 10 In(21o/fiuJ)·
(2)

There is a lack of experimental evidence to either corroborate or dis-

prove Eq. (1) due to the difficulties associated with performing frequency
dependent measurements in the so-called quantum limit, i.e. fiuJ > kBT, but
at small enough photon energies so as to not be exciting charge carriers to
the conduction band. Moreover, in order to study the possible crossover
from Matt to ES type behavior, one must measure across a broad enough
bandwidth centered about the characteristic crossover energy scale for in-
stance the Coulomb interaction energy U or the Coulomb gap width (6),
,6,.
There have been some very recent experiments that have attempted to
address these issues. M. Lee et al. found that for concentrations close to the
MIT the expected linear to quadratic crossover occurs, but is much sharper
than predicted (7). They proposed that this sharp crossover was controlled
not by the average interaction strength U as in Eq.(l) (3), but instead by
a sharp feature in the density of states, i.e. the Coulomb gap (8). They
postulated that this Coulomb gap was not the single particle one measured
in tunneling, but rather a smaller" dressed" or renormalized Coulomb gap
 191

that governs transport. There is some evidence from DC transport that

such a feature exists, at least close to the MIT (9).

2. Experiment

Nominally uncompensated n-type silicon samples were obtained from Recticon

Enterprises Inc. A Czochralski method grown boule with a phosphorous
gradient along its length was cut into 1 mm thick discs. Room temperature
resistivity was measured using an ADE 6035 gauge and the dopant concen-
tration calibrated using the Thurber scale (10). The Si:P samples discussed
here span a range from 39% to 69%, stated as a percentage of the sample's
dopant concentration to the critical concentration at the MIT. A number of
samples were measured before and after etching with a 4 %HF + 96%HN 0 3
solution; this resulted in no difference in the results.
In the millimeter spectral range, 80 GHz to 1000 GHz, backward wave
oscillators (BWO) were employed as coherent sources in a transmission
configuration (11). The transmitted power through the Si:P samples as
a function of frequency was recorded. For plane waves normally incident
on a material, resonances occur whenever the thickness of the material is
an integer number of half wavelengths. Both components of the complex
conductivity can be uniquely determined for each resonance. The real part
of the conductivity was evaluated at microwave frequencies from the meas-
ured loss of highly sensitive resonant cavities at 35 and 60 GHz via the
perturbation method. This is a common technique and is described in the
literature (12). The conductivity as determined from the resonant cavity
data was normalized to the DC conductivity at higher temperatures, at
above approximately 25 K. The resonant cavity data confirmed the linear
dependence on frequency of the real part of the complex conductivity into
the microwave regime for the samples closest to critical.

3. Results

In Fig. 1, we show the T---+O frequency dependent conductivity for two

samples. This data, representative of all samples in our range, shows an
approximately linear dependence at low frequencies and then a sharp cros-
sover to an approximately quadratic behavior at higher frequencies. This is
the qualitatively expected behavior from Eq. (1). However, as seen by the
overlayed fits, Eq. (1) provides only a rough guide. The solid lines are linear
and quadratic fits to the low frequency and high frequency data respectively.
The dotted line is a fit to the form of the ES crossover function achieved
by summing the separately determined linear and quadratic fits. As can
be seen, the crossover between linear and quadratic portions is much more
192

/
Si:P ..-' /
1
10
? / " /.• / / / /

,x
()
,/

S
b 10'

69%
3
10

10- 1
Si:P

E() 10-'
S
b
10-3

100 1000
Frequency (GHz)

Figure 1. Low temperature frequency dependent conductivity data plotted as a function

of frequency. The Si:P samples shown are at 50% and 69% dopant concentration relative
to the critical concentration. The solid lines are linear and quadratic fits to the lower and
upper portions of the data respectively. The dotted and dashed lines are fits following
the form of Eq. (1) using two different methods described in the text.

abrupt than the ES function predicts. The dashed line is a fit using the same
method as Ref. (7), namely forcing the linear portion to pass through the
low frequency data, as well as the origin and leaving the pre-factor of the
quadratic term as a free variable. The fit is not satisfactory in either case.
A sharp crossover as such is observed over our entire doping range and has
been observed previously in an analogous system, Si:B, for samples closer
to the MIT (7). Note, that a linear dependence is seen in the imaginary
part of the conductivity (J"2 over the whole measured frequency and doping
range (13). This is consistent with theoretical predictions (14).
Because our data spans a large range of concentrations, the doping
dependence of the crossover energy scale can be analyzed to see whether
its dependence is consistent with other energy scales, e.g. the Coulomb
interaction energy U or the Coulomb gap width 6. as per Ref. (7). Recall
that the Coulomb interaction energy between two sites forming a reson-
ant pair is U(r w ) = e 2 /clr w which is dependent on concentration via the
dielectric constant (measured, but not shown) and the localization length
 193

dependent most probable hop distance. By equating the crossover energy

scale to the expected functional form for this Coulomb interaction energy
we are able to determine the magnitude of the localization length and its
scaling exponent. With an appropriate pre-factor in the overlap integral
(15), fa = 10 13 s- 1 for the expression for the most probable hop distance
term, T w , we get a localization length dependence of ~ ex (1 - x/x c )-0.83
with a magnitude of 21.2, 19.9, 20.1, 14.5, 14.3 and 13.0 nm for the 69%,
62%, 56%, 50%,45% and 39% samples respectively. The localization length
exponent is close to unity, the value originally predicted by McMillan in
his scaling theory of the MIT (16), and the magnitude of the localization
length is reasonable. Due to the fact that we obtain reasonable estimates
for the relevant physical parameters over the whole doping range, we do not
favor the previous speculation that it is in fact the Coulomb gap energy
that creates the sharp crossover and hence sets its energy scale (7).
The approximately linear power law seen in the Coulomb glass regime
at low w in Fig. 1 of the conductivity can be expressed with the imagin-
ary part as a simple Kramer-Kronig compatible form, 0-( w) = A(iw)c> =
AwC>cos (1T2C» + iAwC>sin (1T2C» • In order to determine the power 0: one can
take the ratio of I0-21 versus 0-1 (with the frequency as a variable). The
power 0: is given by,

0: = ~tan-l (10-21 ) . (3)

7r 0-1
Fig. 2a shows the ratio mentioned above of the imaginary to the real part
of the dielectric constant for Si:P. Similar data from amorphous NbSi is
included for comparison purposes (1). We note that this ratio for Si:P is
large and essentially constant across the entire doping range. From Eq.
(2), one expects 10-11 to be approximately equal to 10-21 to within a factor
of 2-5 (with a reasonable estimate for fa) because 10-21 ex Cl . w. Applied
to Si:P, the theory correctly predict a linear correspondence between 0-1
and 10-21, but incorrectly predicts the proportionality by at least a factor
of thirty. The proportionality is near the predicted value for NbSi, but
has a dependence on the doping concentration which is presumably due
to entering the quantum critical (QC) regime as discussed below. Here
we have used the susceptibility 47rX of the dopant electrons (i.e. with the
background dielectric constant of silicon subtracted) in the expression for
the magnitude of the imaginary component of the conductivity in Eq. 3.
The middle panel in Fig. 2 shows the power 0: as determined by Eq.
(3). The values for Si:P are approximately equal to, but slightly less than
one, consistent with Fig. 1. This indicates that the prefactor of the real
and imaginary components of the complex conductivity have the same
concentration dependence. The situation is different for NbSi. Near the
194

10
2 ! !
II! II! II! ,
\:)

--N
\:)
10'
1;1 2 ~
D i;i

1.0 • • • • • •
~D

ij

0.8

8 ! ! •
0
Si:P
NbSi

\:)
! t
! I

0.3 0.4 0.5 0.6 0.7

1 - x/x c

Figure 2. The upper panel shows the ratios of the imaginary to the real part of the
complex conductivity for samples of Si:P and amorphous NbSi. The NbSi data is adapted
from Ref 1. The middle panel shows the calculated powers of a as determined from Eq.
3. The dashed line through the NbSi data is a guide to the eye. The bottom panel shows
the divergence of the prefactor of the real part of the conductivity, and the dotted line
is a simple power law fit.

MIT, CT(W) is expected to cross over to the QC dynamics (17, 18), i.e.
CTI ex: w 1 / 2 when ~, the localization length, is of the same scale as £w, the
dephasing length (the characteristic frequency dependent length scale) (19).
This should be a smooth crossover and therefore looking at a fixed window
of frequencies, a continuous change from w ----t w 1 / 2 is expected, similar to
that measured for NbSi shown in the middle panel of Fig 2.
Setting the relations for localization length and dephasing length equal
(19), one finds the crossover condition for the frequency in terms of the
normalized concentration, w ex: £0 (1 - x/xcYv where z is the dynamic
exponent. As the prefactor can vary from system to system, the fact that we
see an 0: >:::; 1 across our entire doping range in Si:P, but an 0: that approaches
1/2 in NbSi indicates that the critical regime in Si:P is much narrower and
out of our experimental window. This is consistent with simple dimensional
arguments (20) that show the crossover should be inversely proportional to
the dopant density of states. The much smaller dopant density in Si:P vs.
 195

NbSi (a factor of 103 ) is consistent with a narrower QC regime as compared

to NbSi.
The bottom panel in Fig. 2 shows magnitude of the real part of the
conductivity as the MIT is approached. This demonstrates that the pre-
factor A can be written as a function of the normalized concentration, i.e.
A ex f(l - x/xc) for Si:P.

4. Discussion

In typical interacting systems, the effects of correlations become simpler

as one goes to lower energies and/or lower temperatures. The canonical
example of this is a Fermi liquid where at T=O and w=O one recovers the
non-interacting theory, but with parameters that are substantially mod-
ified (renormalized) from the free electron ones. The Coulomb glass is a
fundamentally important example in solid state physics because it belongs
to a class of systems where this does not occur and the non-interacting
functional forms are not recovered at asymptotically low energies.
As predicted in the crossover function Eq. 1, we have observed in Si:P
that the non-interacting functional form is recovered in the high-frequency
limit and the low-frequency response shows interactions. Within the theory
this is a result of the additional internal excitation structure of a resonant
pair caused by interactions and the fact that anyone pair can be thought
of as a distinct entity, well separated in energy from other spatially nearby
pairs because of the large disorder induced energy spread. This internal
structure enables the excitation of pairs relatively deep within the Fermi
sea and changes the factor in the initial state phase space from w to U (r ) +
w. In contrast, in amorphous NbSi a smooth crossover is observed as the
MIT is approached from a linear frequency dependence to a power law
characterized by an exponent smaller than unity. This is consistent with the
eventual frequency dependence (]" ex w 1 / 2 expected from quantum critical
scaling arguments. No crossover to w 2 was observed in NbSi.
Since it is the exponent a that distills the important physics (it indicates
the phase space of initial states for a = 1 or 2 and the presence of critical
dynamics for a = 1/2), we propose that one can classify the electrodynam-
ics of electron glasses based on their a value. A schematic showing the
parameter space for a is shown in Fig. 3. Here one has a's close to 1/2
near the MIT. There is a smooth crossover to Coulomb glass-like a = 1 at
lower doping levels and an intervening non-interacting Fermi glass regime
at even higher energies and lower dopings. We expect that these general
considerations are valid, despite the fact that some of the parameter space
may not be accessible in certain systems. For instance, one may begin to
excite structural modes at energies high enough to see w 2 behavior in NbSi.
196

Coulomb Glass
a=l

a
a( (i)) ex; (i)

Figure 3. A schematic showing the parameter space for values taken by a for (J" ex: we>.
Note that the boundaries drawn on the plot are smooth crossovers and not sharp onsets.
A classification based on a gives a taxonomy for the electrodynamics of electron glasses.

For dopings not close enough to critical in Si:P, excitation to the conduction
band may be observed before critical dynamics are.

5. Conclusion

In summary, we have observed a crossover in the frequency dependence

of the conductivity from Coulomb glass-like behavior to Fermi glass-like
behavior across our entire range of doping concentrations in Si:P. The
existence of a crossover is consistent with theoretical predictions, but it
is sharper than predicted. The fact that we see the same functional form
over the whole doping range (even deep into the insulating regime, where
Eq. 1 is expected to be more valid) shows that the nature of the low energy
charge excitations is qualitatively the same over the whole doping range;
the inadequacy of Eq. 1 in describing the frequency dependent conduct-
ivity quantitatively is not limited to concentrations close to critical. In
the amorphous semiconductor NbSi, we observe a gradual crossover from
a = 1 Coulomb glass-like behavior to quantum critical-like dynamics. This
comparison allows us to obtain a general classification scheme for the elec-
trodynamics of electron glasses based on the exponent of the frequency
dependence a. We expect that this classification or 'taxonomy' will be valid
even when certain regimes are not experimentally accessible.
 197

Acknowledgements

We wish to thank Phu Tran for assisting with the cavity measurements and
Barakat Alavi for assisting with the sample preparation. We would also
like to thank Steve Kivelson and Boris Shklovskii for helpful conversations.
This research was supported by the National Science Foundation grant
DMR-Ol02405.

References

1. Erik Helgren, George Gruner, Martin R. Ciofalo, David V. Baxter, and John P.
Carini, Phys. Rev. Lett. 87, 116602 (2001).
2. N. F. Mott and E. A. Davis, Electronic Processes in Non-Crystalline Materials,
Second Edition, (Oxford University Press, Oxford, 1979).
3. A. L. Efros and B. 1. Shklovskii, in Electron-electron Interactions in Disordered
Systems, edited by A. L. Efros and M. Pollak (Elsevier New York, 1985), p. 409-482.
4. S. Tanaka and H. Y. Fan, Phys. Rev. 132, 1516 (1963).
5. H. Stupp, M. Hornung, M. Lakner, O. Madel, and H. v. LiShneysen, Phys. Rev.
Lett. 71, 2634 (1993).
6. A. L. Efros and B. 1. Shklovskii, J. Physics C 8, L49 (1975).
7. M. Lee and M. L. Stutzmann, Phys. Rev. Lett. 30, 056402 (2001).
8. M. Lee and J. G. Massey, Phys Rev. B 60,1582(1999).
9. J. G. Massey and M. Lee, Phys. Rev. B 62, R13270 (2000).
10. W. R. Thurber et al., J. Electrochem. Soc. 127,1807 (1980).
11. A. Schwartz et al., Rev. Sci. lnstrum. 66, 2943 (1995).
12. G. Gruner, Millimeter and Submillimeter Wave Spectroscopy of Solids (Springer
Verlag, Berlin, 1998).
13. E. Helgren, N.P. Armitage, and G. Gruner, Phys. Rev. Lett. 89, 246601 (2002).
14. A. L. Efros, Sov. Phys. JETP 62, 1057 (1985). An analagous three-dimensional form
of the two-dimensional theory described in this paper can be derived.
15. B. 1. Shklovskii. pr'ivate comm'Un·ication. The In used is the Bohr energy of
phosphorous.
16. W. L. McMillan, Phys. Rev. B 24, 2739 (1981).
17. H.-L. Lee, J. P. Carini, D. V. Baxter, and G. Gruner, Phys. Rev. Lett. 80, 4261
(1998).
18. H.-L. Lee, John P. Carini, David V. Baxter, W. Henderson, and G. Gruner, Science
287, 633 (2000).
19. S. L. Sondhi, S. M. Girvin, J. P. Carini, and D. Shahar, Rev. Mod. Phys. 69, 315
(1997).
20. S. Kivelson, private communication.
RENORMALIZATION GROUP APPROACHES FOR SYSTEMS
WITH ELECTRON-ELECTRON AND ELECTRON-PHONON
INTERACTIONS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

A.C. HEWSON and D. MEYER

Department of Mathematics, Imperial College,
London SW7 2AZ, UK.

Abstract. We present results of numerical renormalization group calculations for an

impurity Anderson model with a coupling to a local phonon mode. The calculations are
then extended, by incorporating the self-consistency condition of dynamic mean field
theory, to consider the behaviour of the infinite dimensional Holstein model.

Key words: Holstein Model, Anderson-Holstein Model, Numerical Renormalization Group

(NRG), Dynamical Mean-Field Theory Electron-Phonon Coupling Metal-Insulator Trans-
ition

1. Introduction

The application of the renormalization group approaches to strongly correl-

ated systems has now a long history stretching back to the seminal work of
Wilson on the Kondo problem (1), and before that to the work of Anderson
(2). The Wilson numerical approach (NRC) has been particularly useful in
providing a very reliable way of calculating the thermodynamic behaviour
of magnetic impurity models, over the whole parameter range from weak
to strong coupling (3). Its later extension to the calculation of dynamic
response functions has considerably increased the range of physical prop-
erties that can now be calculated for comparison with experiment (4, 5).
It has also enabled the method to be applied to lattice models, such as the
Hubbard and periodic Anderson models, within the framework of dynamical
mean field theory (DMFT), where the lattice problem is mapped into an
effective impurity problem with an additional self-consistency constraint
(for a comprehensive review of DMFT see reference (6)). Though this

199
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 199–207.
© 2003 Kluwer Academic Publishers
200

mapping is only approximate, except in the limit of infinite dimensionality

d -----+ 00, there are indications that it provides a significant first step to
understanding the behaviour of these lattice models in the difficult strong
correlation regime. For an introduction to the NRG approach and a survey
of the applications to both impurity and lattice models we refer the reader
to recent review articles (7, 8).
In this paper we concentrate on using the NRG method to calculate the
effects, within a generalized Anderson model, of including a coupling of the
impurity with a local phonon mode. The generalization to the lattice case,
using DMFT, is then used to calculate the behaviour of the Holstein model
(9) in the strong and intermediate coupling regimes.

2. Anderson-Holstein impurity model

The Hamiltonian for the Anderson-Holstein model (10) is

IJ IJ

+L Vk(C}IJCkIJ + cLcfIJ) + L EkCLckIJ + wobtb, (1)

k,IJ kIJ

which corresponds to an Anderson model with an additional coupling of the

impurity occupation number to a local optical phonon mode of frequency
wo, as in the Holstein model.
We can get some insight into the behaviour of this model by considering
the large frequency limit Wo -----+ 00. In this limit the model can be mapped
into a Anderson model without the phonon term but with a renormalized
level Ef = Ef + v and an interaction U eff = U - 2v, where v = g2/ wO and
the limit 9 -----+ 00 is taken such that v remains finite. In the particle-hole
symmetric case with U = 0, and Ef = 0, the model then becomes a negative-
U Anderson model. The behaviour of the negative U model is essentially
that of the positive U model with spin and charge interchanged. This means
that if we do move away from particle-hole symmetry for the negative U
model by switching on a finite value for Ef' then this is equivalent to adding
a magnetic field to the positive U model.
The NRG technique as used to calculate the response functions for the
standard Anderson model can be generalized to this model also (10), as
the phonon term interacts with the electrons only at the impurity site. The
number of phonon excitations has to be truncated and this has negligible
effect provided that it exceeds a value of the order 4g 2 / w5.
In practice it
is easy to take many more than this and about 200 phonon excitations are
retained.
 201

100
20 - g=O.O 90
---- g=0.025
...... g=0.05 80
._._. g=0.071 70
15
60
1m Xs 50
1m Xc 10
40
30
5: 20
, I

-_.--------. 10
"
O~'c..::.........~;±--~-~===== 00 0.02 0.04
0.02 0.04
0) 0)

Figure 1. Imaginary part of the dynamic charge (Xc) and spin susceptibilities (Xs)
for the symmetric parameters, U = 0.1, Ef = -0.05, 6. = 0.04 and values of
Ucff = 0.1,0.075,0.0, -0.1.

The change in the nature of the charge and spin excitations can be seen
in the spectral densities of the corresponding charge and spin susceptibilities
which are shown in figure 1. For a particle-hole symmetric model with
U = 0.1, Ej = -0.05, and 9 = 0, the spin susceptibility has a low energy
peak, while the peak for the charge susceptibility is suppressed and pushed
to higher energies due to the local repulsive interaction U. As the coupling
strength 9 to the phonons is increased these features are gradually reversed.
For the largest value of the phonon coupling shown, 9 = 0.071, the effective
local interaction is negative U eff = U - 2v = -0.1, and now the spin
excitations are suppressed and the charge excitations develop a low energy
peak. The susceptibilities shown in figure 1 are for Wo = 0.05, and the low
energy peak in the charge susceptibility at Ueff = -0.1 (g = 0.071) can be
seen to be much narrower than that of the corresponding spin susceptibility
at Ueff = 0.1 (g = 0). As WQ ----t 00 the peaks are identical as one would
expect from our earlier discussion.
The narrowing of the low energy peak in the charge susceptibility with
decrease in WQ is reflected in the spectral densities of the local electron
Green's function for the particle-hole symmetric model, which are plotted
in figure 2 for three values of WQ = 0.05,0.1,0.5 with U = 0 and a fixed value
of U eff . The first plot has U eff = -0.05 and so is in the weak to intermediate
coupling regime. The narrowing of the resonance as a function of Wo is
quite pronounced and distinct shoulders on either side of the resonance
can be seen in the plot with the smallest value of W00 The second plot is
for U eff = -0.1 which is in the strong coupling regime. Again the central
resonance can be seen to narrow as Wo is decreased. There are also distinct
202

8 - (00=0.05 8 - (00=0.05
.... (00=0.1 .... (00=0.1
-- (00=0.5 - - (00=0.5
6 6

4 4

2 2

--- ---
0
0 -0.2 0 0.2

Figure 2. In these two figures the spectral density of the local Green's function for the
particle-hole symmetric model with 6 = 0.04, and U = 0, are plotted for three values of
Woo The value of Ucff is maintained as constant in each plot with Ucff = -0.05 for the

first figure and Ueff = -0.1 for the second.

upper and lower peaks in all three plots. The lower energy peak is associated
with the adding or removing of an electron from a double or zero occupied
impurity state, which are degenerate in the particle-hole symmetric case.
The corresponding peak above the Fermi-level is associated with the adding
or removing an electron from a singly occupied impurity level. These peaks
are the broadened counterparts of those that exist in the 'atomic limit'
V = O. Further shoulders on these peaks can be seen in the case of the
smallest value of WQ taken, which correspond to additional excitations in
which phonons are emitted or absorbed. In the opposite limit WQ ----t 00 such
excitations are pushed out to very high energies, and in the strong coupling
regime the central peak can be interpreted as a Kondo resonance, with
a Kondo temperature TK given by TK = (IUeffl~/2)1/2exp( -1TIUeffl/8~ +
1T~/IUeffl). The Kondo physics in this case is associated with charge or
pseudo-spin degeneracy and not spin degeneracy.
The effects of the narrow peak in the charge susceptibility for the
particle-hole symmetric model with U = 0 and Ueff < 0 can be seen the
spectrum for the phonon Green's function D(w) = ((b + bt : b + bt )). A
relation between the two can be derived from the equations of motion (10),

(2)

The low energy peak in Xc( w) induces a low energy peak in the phonon
spectrum, which survives even if we take the limit WQ ----t 00.
 203

It is also possible to interpret the results of these calculations in terms of

the development of a double potential well. At large coupling strengths, or
in the atomic limit, there are two equivalent potential energy minima (for
U = 0) as a function of x = (b + bt )/ y/2wo, which is treated as a classical
variable, corresponding to double or zero occupation of the impurity site.
The tunnelling between these two minima is responsible for the central
resonance. In this strong coupling limit the interaction can be described
in terms of a pseudo-spin s-d type of model (12). In the limit Wo ----+ 00
this s-d model is isotropic and leads to the usual Kondo physics in terms
of a pseudo-spin. However, for finite values of Wo the transverse coupling
terms in the pseudo-spin model are reduced by a factor of e- 4g2 / w 6 over the
longitudinal terms, as they are associated with tunnelling between the two
minima, and the model then corresponds to an anisotropic s-d model. It
would be of interest to compare the results with those of this anisotropic
s-d model in detail in the appropriate parameter regime.
These results for the spectra of the one-electron Green's function for
the particle-hole symmetric model throw some light on the corresponding
results for the infinite dimensional lattice Holstein model which we consider
in the next section, but before considering that case we will discuss a few
more results for the Anderson-Holstein model in other parameter regimes.
Away from particle-hole symmetry the results for the Anderson-Holstein
model are quite straightforward to interpret. For finite U and Ueff > 0 the
Kondo resonance first broadens as the coupling strength 9 increases from
zero, because this reduces Ueff, and then rapidly disappears. In the strong
coupling regime (Ueff < 0) the Kondo resonance is not present as there is
no longer any degeneracy of the doubly and zero occupied states. If the
double and zero occupation states are regarded as the up and down states
of a pseudo-spin then in this regime there is an effective field acting on the
pseudo-spin and the Kondo resonance is suppressed.

3. The infinite dimensional Holstein model

The NRG calculations for the impurity models can be extended to the
calculation of the response functions for lattice models in the framework
of dynamical mean field theory (DMFT). The DMFT has been applied
widely and is extensively reviewed in reference (6). The lattice problem is
mapped onto an effective impurity problem, which is then subject to a self-
consistency constraint. There are many ways of carrying out calculations for
the effective impurity model, and the Monte Carlo approach is one that has
been used extensively (13). However, the Monte Carlo calculations cannot
be performed for T = 0, or for very low temperatures, so the NRG approach
has a distinct advantage in this regime. The NRG method is also capable
204

of resolving very low energy scales. Most applications so far of the NRG-
DMFT technique have been to the Hubbard and periodic Anderson models.
Using the approach developed in the previous section, we are in a position
to carry out calculations which also include a coupling to phonons, and we
have focussed our attention so far on the Holstein model (9). In this model
the electron occupation number at each site is linearly coupled to a local
boson mode of frequency woo The corresponding Hamiltonian has the form,

H = L EkCLcka + 9 L(b! + bi)(L ni" - 1) + Wo L b!bi, (3)

ka a

where i is a site index. In the limit Wo ----7 00 with Ueff = -2g2/wo finite the
model maps into an effective 'negative-U' Hubbard model. The dynamic
mean field theory approach is strictly speaking only valid for the infinite
dimensional model but should constitute a good starting point for under-
standing the physics of this model, particularly in the intermediate and
strong coupling regime where most methods break down. The effective im-
purity model in this case is the U = 0 Anderson-Holstein model considered
in the previous section. The effective density of states of the host, however,
has to be calculated self-consistently from the DMFT constraint. Results
of applying the DMFT-NRG approach to this model are shown in figure 3.
The first plot gives the spectral density of the local one-electron Green's
function for the particle-hole symmetric model. As the coupling strength 9
is increased from zero, a narrow peak develops at the Fermi-level, which is
quite similar to that seen in the results for the impurity model in figure 2
in the same parameter regime. The results are also quite similar to those
obtained by application of the Migdal-Eliashberg approach (14) in which
all perturbational theory diagrams are summed subject to the neglect of
all vertex corrections.
As the coupling is further increased a pseudo-gap with a very nar-
row central peak begins to develop. At a critical coupling strength gc the
central peak disappears and a gap in the excitation spectrum develops
continously. The quasi-particle spectral weight z decreases monotonically
with increasing coupling strength and goes to zero at 9 = gc. Due to the
attractive effective interaction the electrons pair to form bipolarons as the
interaction becomes stronger, and the lower peak is associated with the
bipolaron formation. The Migdal-Eliashberg (ME) approach breaks down
before a pseudogap develops and the results are limited to the weak coupling
regime. If the ME approach is applied to the Anderson-Holstein model its
limitations can be clearly seen as it does not develop the higher and lower
atomic-like peaks.
It is interesting to contrast these results with those for the positive U-
Hubbard model where a metal-insulator gap develops at a critical coupling
 205

- g=0.03
._._. g=0.08
....... g=0.098
---- g=0.12

,, cr(m)
pew) ,,
,,
0.5
, \

\
,,
,,
,,

o 0.01 0.02 0.03 0.04 0.05

00

Figure 3. The spectral density for the local one-electron Green's function p( w) and that
for the phonon Green's function O"(w) for the infinite dimensional Holstein model plotted
for various values of the coupling strength g.

strength Uc (6, 15). In the Hubbard case a central quasi-particle peak

occurs at the Fermi-level but on the approach to Uc this central peak is
largely isolated from the upper and lower Hubbard bands, and at the metal
insulator transition when the central peak disappears there is a finite gap.
We can also calculate the corresponding spectral density for the phonon
spectrum, either directly or from an equation of motion similar to that
given in equation 2 in terms of the effective impurity charge susceptibil-
ity. As the coupling strength increases a lower excitation mode develops
in the spectrum in addition to the broadened original mode at woo This
additional low energy peak softens as the critical value of 9 is approached
from below, and diverges in the limit 9 ----+ gc. A similar low energy peak
is seen in the calculations for the Anderson-Holstein model but this mode
never completely softens, as no gap develops and there is always a central
resonance. In the impurity case the induced lower phonon peak can be seen
clearly to reflect the low energy peak in the local charge susceptibility. The
picture of the change in the phonon spectrum in this case is quite different
from that found in the ME calculations, where there is only a single peak
which softens monotonically with increasing 9 (14). In the ME case there is
no complete softening, as the method breaks down before that can occur.

4. Conclusions

The combination of the numerical renormalization group and dynamical

mean field theory gives an effective way of tackling a number of difficult
problems relating to strongly correlated systems. Here we have shown that
206

it can be used to provide insights into models with coupling to phonons

in the intermediate and strong coupling regime, which have so far proved
to be intractable. These calculations are only a start, and there are many
other interesting questions that one should be able to address using this
approach. The Holstein model neglects the Coulomb interaction and so
over emphasizes the tendency for bipolaron formation. Introducing a short
range repulsive interaction, as in the Hubbard model, should enable strong
polaron physics to be studied. Polaronic physics described by the spinless
Holstein model (which has no induced attractive electron-electron interac-
tion) with one or two electrons present has been extensively investigated
(see for example (16)). There has however been very little work on systems
with finite electron density, which is the most relevant for applications (for
example to the fullerides (17)). The extension of the DMFT-NRG to the
Holstein-Hubbard model should be a good way of investigating the physics
in the strong coupling, finite electron density regime. It should also be
possible to extend the technique to include a coupling to a local spin as a
model to investigate the rich physics of manganite systems (18). Both these
possibilities are currently being investigated.

Acknowledgements

We thank the EPSRC for the support of a research grant (GR/J85349),

Ralf Bulla for cooperation on aspects of this project, and David Edwards
for helpful and stimulating conversations.

References

1. KG. Wilson, Rev. Mod. Phys. 47, 773 (1975).

2. P.W. Anderson, J. Phys. C 3, 2349 (1970).
3. H.R. Krishna-murthy, J.W. Wilkins, and KG. Wilson, Phys. Rev. B 21, 1003 and
1044 (1980).
4. O. Sakai, Y. Shimizu and T. Kasuya, J. Phys. Soc . .lap. 58, 3666 (1989).
5. T. A. Costi and A. C. Hewson, Phil. Mag. B 65, 1165, (1992) : J. Phys. Condo Mat.
30, L361 (1993): T.A. Costi, A.C. Hewson and V. Zlatic, J. Phys. Cond Mat 6,
2519 (1994).
6. A. Georges, G. Kotliar, W. Krauth and M.J. Rozenberg, Rev. Mod. Phys. 68, 13
(1996).
7. R. Bulla, Adv. in Solid State Physics, 40, 169 (2000).
8. A.C. Hewson, S.C. Bradley, R. Bulla, and Y. Ono, Int. J. Mod. Phys. B, 15, 2549
(2001).
9. T. Holstein, Ann. Phys. 8, 325 (1959).
10. A.C. Hewson and D. Meyer, J. Phys.:Cond.Mat., 14, 427, 2002.
11. D. Meyer, A.C. Hewson and R. Bulla, Phys. Rev. Lett. 89, 196401 (2002).
12. J.K Freericks, Phys. Rev. B 48, 3881 (1993).
 207

13. J.K. Freericks, M. Jarrell and D.J. Scalapino, Phys. Rev. B 48, 6302 (1993).
14. J.P. Hague and N. d'Ambrumenil, cond-mat/0l06355 (2001).
15. R. Bulla, Phys. Rev. Lett. 83, 136 (1999).
16. S.A. Trugman, J. Bonca, and L.-C. Ku, Int. J. Mod. Phys. B 15, 2707 (2001).
17. O. Gunnarsson, Rev. Mod. Phys. 69, 575 (1997).
18. A.J. Millis, P.B. Littlewood and B.!. Shraiman, Phys. Rev. Lett. 74,5144 (1995).
QUANTUM PHASE TRANSITIONS IN MODELS OF MAGNETIC
IMPURITIES

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

R. BULLA and M. VOJTA

Theoretische Physik III, Elektronische Korrelationen und Mag-
netismus, Universitiit Augsburg, 86135 Augsburg, Germany

Abstract. Zero temperature phase transitions not only occur in the bulk of quantum
systems, but also at boundaries or impurities. We review recent work on quantum phase
transitions in impurity models that are generalizations of the standard Kondo model
describing the interaction of a localized magnetic moment with a metallic fermionic host.
Whereas in the standard case the moment is screened for any antiferromagnetic Kondo
coupling as T ----> 0, the common feature of all systems considered here is that Kondo
screening is suppressed due to the competition with other processes. This competition
can generate unstable fixed points associated with phase transitions, where the impurity
properties undergo qualitative changes. In particular, we discuss the coupling to both non-
trivial fermionic and bosonic baths as well as two-impurity models, and make connections
to recent experiments.

Key words: Quantum Phase Transition, Kondo Effect, Local Criticality, Numerical Renor-
malization Group, Two-Impurity Kondo Model, Two-Channel Kondo Model

1. Introduction

Quantum mechanical systems can undergo zero-temperature phase trans-

itions upon variation of a non-thermal control parameter (1), where order
is destroyed solely by quantum fluctuations. Quantum phase transitions
occur as a result of competing ground state phases, and can be classified
into first-order and continuous transitions. The transition point of a con-
tinuous quantum phase transition, the so-called quantum-critical point, is
typically characterized by a critical continuum of excitations, and can lead
to unconventional behavior - such as non-trivial power laws or non-Fermi
liquid physics - over a wide range of the phase diagram (see Fig. 1a).

209
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 209–217.
© 2003 Kluwer Academic Publishers
210

An interesting class of quantum phase transitions are so-called bound-

ary transitions where only the degrees of freedom of a subsystem become
critical. In this paper we consider impurity transitions - the impurity can
be understood as a zero-dimensional boundary - where the impurity con-
tribution to the free energy becomes singular at the quantum critical point.
Such impurity quantum phase transitions require the thermodynamic limit
in the bath system, but are completely independent of possible bulk phase
transitions in the bath.
Our model systems are built from magnetic moments which show the
Kondo effect (2). Originally, this effect describes the behavior of localized
magnetic impurities in metals. The relevant microscopic models are the
Kondo model and the single-impurity Anderson model. In the standard case
(i.e. a single magnetic impurity with spin ~ coupling to a single conduction
band with a finite density of states (DOS) near the Fermi level) screening of
the magnetic moment occurs below a temperature scale TK. The screening
is associated with the flow to strong coupling of the effective interaction
between impurity and host fermions. The Kondo temperature TK depends
exponentially on the system parameters.
In this paper we want to give a brief summary of impurity models
where the flow to strong coupling is prevented by the competition with
other processes, and, depending on the system parameters, weak and/or
intermediate-coupling fixed points can be reached. We are particularly
interested in intermediate-coupling fixed points, which typically show a
finite ground-state entropy and are unstable, which means that to reach
them requires either fine tuning of the couplings (typically associated with
an impurity phase transition) or the presence of special symmetries (e.g.,
channel symmetry in multi-channel models). In the following sections we
discuss single-impurity models with a coupling to a non-trivial fermionic
bath (Sec. 2), to multiple fermionic baths (Sec. 3), to both fermionic and
bosonic baths (Sec. 4), and finally systems of coupled magnetic impurities
(Sec. 5). In quoting results, we refer to perturbative renormalization group
(RG) treatments, to numerical RG (NRG) calculations, and to exact results
if available. We restrict ourselves to models with spin-~ impurities.

2. Pseudogap Kondo and Anderson models

A straightforward possibility to suppress Kondo screening is to reduce the

electron bath DOS at the Fermi level to zero - in a superconductor this
can also be interpreted as competition between Kondo effect and super-
conducting pairing. Two cases have to be distinguished: so-called hard-gap
and pseudogap (soft-gap) systems.
 211

0.06
~
.3 a b
~
OJ
D-
E 0.04
2
SC
<1

symmetric case
0.02
asymmetric case

fixed point A B LM
\
0.00
0.00 0.25 0.50 0.75

Figure 1. (a) Schematic phase diagram near a quantum phase transition between fixed
points A and B upon variation of a control parameter x; the T = 0 critical point controls
the dynamics in the T > 0 quantum critical region whose crossover boundaries are given
by T ~ Ix - where l/ and z are the correlation length and dynamical critical
exponents, respectively. (b) T = 0 phase diagram for the pseudogap Anderson model in
the p-h symmetric case (solid line, U = 10- 3 , E f = -0.5 . 10- 3 , conduction band cutoff
at -1 and 1) and the p-h asymmetric case (dashed line, Ef = -0.4·10-:1); 6 measures
the hybridization strength, .6.(e) == 7fV2 p(e) = 6lel r .

In the hard-gap case, the DOS p( c) is zero in a finite energy interval

around the Fermi level. Then, the absence of low-energy states prevents
screening for gap values exceeding the energy gain due to Kondo screen-
ing. The resulting transition between a local-moment (LM) phase without
Kondo screening, realized at small Kondo coupling J, and a screened strong-
coupling (SC) phase, reached for large J, is of first order, and it occurs only
in the presence of particle-hole (p-h) asymmetry. In the p-h symmetric case,
the local-moment state persists for arbitrary values of the coupling (3).
The pseudogap case, first considered by Withoff and Fradkin (4), cor-
responds to a bath with p(c) ex: Icl T (r > 0), i.e., the DOS is zero only at
the Fermi level. The corresponding Kondo and single-impurity Anderson
models interpolate between the metallic case (r = 0) and the hard-gap case
(r ----t ()()). The pseudogap case 0 < r < ()() leads to a very rich behaviour, in
particular to a continuous transition between a local-moment and a strong-
coupling phase. Figure Ib shows a typical phase diagram for the pseudogap
Anderson model. In the p-h symmetric case (solid) the critical coupling
,6., measuring the hybridization between band electrons and local moment,
diverges at r = ~, and no screening occurs for r > ~ (5, 6). No divergence
occurs for p-h asymmetry (dashed) (5).
We now briefly describe the properties of the fixed points in the pseudogap
Kondo problem (5). Due to the power-law conduction band DOS, already
the stable LM and SC fixed points show non-trivial behavior (5, 6). The LM
phase has the properties of a free spin ~ with residual entropy Simp = k B In 2
212

L1=L1c L1>L1c
local moment quantum critical strong coupling

a b c

2.0
z
W
~
z
<::
1.0

50 100 0 50 100 0 50 100

N N N

Figure 2. Flow diagrams for the low-energy many-body excitations obtained from
the numerical renormalization group for the three different fixed points of the soft-gap
Anderson model. N is the number of iterations of the NRG procedure, A the NRG
discretization parameter.

and low-temperature impurity susceptibility Ximp = 1j(4kB T), but the

leading corrections show r-dependent power laws. The p-h symmetric SC
fixed point has very unusual properties, namely Simp = 2rk B in 2, Ximp =
rj(8k BT) for 0 < r < ~. In contrast, the p-h asymmetric SC fixed point
simply displays a completely screened moment, Simp = TXimp = O. The
impurity spectral function follows a w r power law at both the LM and the
asymmetric SC fixed point, whereas it diverges as w- r at the symmetric SC
fixed point - this "peak" can be viewed as a generalization of the Kondo
resonance in the standard case (r = 0), and scaling of this peak is observed
upon approaching the SC-LM phase boundary (6, 7). At the critical point
non-trivial behavior corresponding to a fractional moment can be observed:
Simp = kBCs(r), Ximp = Cx(r)j(kBT) with Cs, Cx being universal functions
of r (5, 8). The spectral function displays a w- r power law (for r < 1) with
a remarkable "pinning" of the critical exponent.
We note that the critical point at small r is perturbatively accessible
in a double expansion in rand J. However, the NRG results suggest that
the physics is different for r > ~; in particular, r = 1 appears to play
the role of an upper critical "dimension" (5, 8, 9). The universal critical
theory of the transitions in the pseudogap Kondo model is not yet known.
A piece of information, provided by the numerical renormalization group
method, is shown in Fig. 2. Here we plot the energies of the many-body
states as a function of the iteration number N of the NRG procedure.
Increasing values of N correspond to decreasing temperature, N ex -in T,
 213

so we can easily recognize the three different fixed points for ,6. < ,6.c (Fig.
2a), ,6. = ,6.c (Fig. 2b), and ,6. > ,6.c (Fig. 2c). The structure of the LM
and SC fixed points can be easily understood as that of a free conduction
electron chain (10, 11). The combination of the single-particle states of the
free chain leads to the degeneracies seen in the many-particle states of the
LM and SC fixed points (Fig. 2a and Fig. 2c). In contrast, the structure of
the quantum critical point is unclear. Degeneracies due to the combination
of single-particle levels are missing, probably because the quantum critical
point is not build up of non-interacting single-particle states.
The pseudogap Kondo model has been proposed (8) to describe impurity
moments in d-wave superconductors (r = 1), where signatures of Kondo
physics have been found in NMR experiments (12). Furthermore, a large
peak seen in STM tunneling near Zn impurities (13) can be related to the
impurity spectrum in the asymmetric pseudogap Kondo model.

3. Multi-channel Kondo model

Kondo screening is strongly modified if two or more fermionic screening

channels compete. Nozieres and Blandin (14) proposed a two-channel gen-
eralization of the Kondo model, which shows overscreening associated with
an intermediate-coupling fixed point and non-Fermi liquid behaviour in
various thermodynamic and transport properties. In general, such behavior
occurs for any number of channels K > 1 coupled to a spin ~, and does not
require fine-tuning of the Kondo coupling, however, it is unstable w.r.t. a
channel asymmetry.
The anomalous static properties at the two-channel non-Fermi liquid
fixed point are (15) a residual entropy Simp = k; In 2 (indicating that 'half'-
fermionic excitations playa crucial role for the structure of the fixed point),
a logarithmic divergence of the susceptibility Ximp ex: In T and of the specific
heat ratio, = Cimp/T ex: In T, and an anomalous Wilson ratio R = xl, =
8/3, in contrast to the result for the standard Kondo model R = 8/4 = 2.
Analogous to the discussion in the previous section the many-particle
excitation spectrum of this intermediate-coupling fixed point cannot be
decomposed in terms of usual free fermions (16, 17), however, a description
in terms of non-interacting Majomna fermions is possible (18). Many of
the low-energy properties of the two-channel and related models have been
studied using conformal field theory techniques (19, 20). Interestingly, the
multi-channel Kondo fixed point is perturbatively accessible in the limit
of large channel number (K » 1) (14, 20). Experimental realizations have
been discussed in the context of rare-earth compounds (15); furthermore,
proposals based on quantum-dot devices have been put forward.
214

4. Bose-Fermi Kondo model

Novel phenomena occur for magnetic impurities coupled to both a fermionic

and a bosonic bath, where the bosons represent collective host spin fluc-
tuations. In the resulting Bose-Fermi Kondo model, the two interactions
compete in a non-trivial manner (21, 22). Particularly interesting is the
case of a bosonic bath with zero or small gap, corresponding to the vicinity
to a magnetic quantum critical point in the host. In (3 - E) dimensions,
the bosonic spectral density then follows a power law ex: WI-E. For E > 0,
the interaction between the impurity and the bosonic bath can completely
suppress fermionic Kondo screening. The resulting phase corresponds to
an intermediate-coupling fixed point w.r.t. the impurity-boson interaction.
Here, the impurity moment shows universal fluctuations, with local spin
correlations characterized by a power law, (S(T)S) ex: T- E , and a Curie
contribution to the impurity susceptibility equivalent to a fractional spin,
Ximp = Cx(E)j(kBT) (21, 22, 23). On the other hand, large fermionic
Kondo coupling leads to a strong-coupling phase with conventional Kondo
screening.
The resulting phase diagram for the Bose-Fermi Kondo model thus
shows a Kondo-screened phase, a bosonic fluctuating phase, and a continu-
ous quantum phase transition in between. The boundary quantum critical
point has magnetic properties similar to the bosonic fluctuating fixed point
(22). Both intermediate-coupling fixed points are perturbatively accessible
for small E; it is likely that the structure of the phase diagram also applies
to E = 1, however, no accurate numerical calculations are available to date.
Both the critical and the bosonic fluctuating fixed points are unstable w.r.t.
breaking of SU(2) symmetry, but the structure of the phase diagram is
similar for both XY and Ising symmetries (22).
The Bose-Fermi Kondo model has recently received a lot of interest
in the context of the extended dynamical mean-field theory (24) where a
lattice model is mapped onto a self-consistent impurity model with both
fermionic and bosonic baths. In particular, based on neutron scattering
experiments on the heavy-fermion compound CeCu5-xAux (25), a self-
consistent version of the Bose-Fermi Kondo model has been proposed to
describe a "local" critical point in Kondo lattice models.
We further note that the Bose Kondo model with Ising symmetry and
an additional external field is related to the spin-boson model (26), which
also shows phase transitions upon variation of the coupling between spin
and dissipative bath. Spin-boson models have been studied extensively in
the context of dissipative two-level systems, and have applications in many
fields like glass physics, quantum computation etc.
 215

5. Two-impurity Kondo models

Models of two impurities offer a new ingredient, namely the exchange in-
teraction, I, between the two impurity spins which competes with Kondo
screening of the individual impurities. This inter-impurity interaction, which
can lead to a magnetic ordering transition in lattice models, arises both from
direct exchange and from the Ruderman-Kittel-Kasuya-Yosida (RKKY)
interaction mediated by the conduction electrons.
In the simplest model of two S = ~ impurities, a ground state sing-
let (Stat = 0) can be realized either by individual Kondo screening (if
I < TK) or by formation of an inter-impurity singlet (if I > TK)' It has
been shown that these two parameter regimes are continuously connected
(without a T = 0 phase transition) as I is varied in the generic situation
without particle-hole symmetry. Notably, in the particle-hole symmetric
case one finds a transition associated with an unstable non-Fermi liquid
fixed point (27, 28, 29).
Quantum phase transitions generically occur in impurity models show-
ing phases with different ground state spin. For two impurities, this can be
realized by coupling to a single conduction band channel only (30). In this
case, a Kosterlitz-Thouless-type transition between a singlet and a doublet
state occurs, associated with a second exponentially small energy scale in
the Kondo regime (30). The physics becomes even richer if multi-channel
physics is combined with multi-impurity physics - here, a variety of fixed
points including such with local non-Fermi liquid behavior can be realized.
Experimentally, quantum dots provide an ideal laboratory to study
systems of two (or more) "impurities" - note that the local "impurity"
states can arise either from charge or from spin degrees of freedom on each
quantum dot. In particular, a number of experiments have been performed
on coupled quantum dot systems which can be directly mapped onto mod-
els of two Kondo or Anderson impurities (31). In addition, experimental
realizations of two-impurity models using magnetic adatoms on metallic
surfaces appear possible.

6. Summary

We have reviewed a variety of interesting zero-temperature critical points,

which exist in impurity models where conventional Kondo screening is
suppressed by competing physics. Significant progress has been made in
recent years, both in analytical and numerical work; however, in a number
of cases our understanding concerning, e.g., the nature of the quantum
critical points is rather limited. Clearly, a further development of theoretical
techniques, both perturbative and non-perturbative, is essential.
216

Acknowledgements

It is a pleasure to acknowledge fruitful discussions and collaborations with

A. Hewson, W. Hofstetter, D. Logan, Th. Pruschke, and S. Sachdev, as well
as financial support by the DFG through SFB 484.

References

1. S. Sachdev, Quantum Phase Transitions, Cambridge University Press, Cambridge

(1999).
2. A. C. Hewson, The Kondo Problem to Heavy Fermions, Cambridge University Press,
Cambridge (1993).
3. K. Chen and C. Jayaprakash, Phys. Rev. B 57, 5225 (1998).
4. D. Withoff and E. Fradkin, Phys. Rev. Lett. 64, 1835 (1990).
5. C. Gonzalez-Buxton and K. Ingersent, Phys. Rev. B 57, 14254 (1998).
6. R. Bulla, T. Pruschke and A. C. Hewson, J. Phys.: Condens. Matter 9, 10463 (1997);
R. Bulla, M. T. Glossop, D. E. Logan and T. Pruschke, ibid 12, 4899 (2000).
7. D. E. Logan and M. T. Glossop, J. Phys.: Condens. Matter 12, 985 (2000).
8. M. Vojta and R. Bulla, Phys. Rev. B 65, 014511 (2002).
9. K. Ingersent and Q. Si, Phys. Rev. Lett. 89, 076403 (2002).
10. K. G. Wilson, Rev. Mod. Phys. 47, 773 (1975).
11. H. R. Krishna-murthy, J. W. Wilkins and K. G. Wilson, Phys. Rev. B 21, 1003
(1980).
12. J. Bobroff, W. A. MacFarlane, H. Alloul, P. Mendels, N. Blanchard, G. Collin and
J.-F. Marucco, Phys. Rev. Lett. 83, 4381 (1999).
13. E. W. Hudson, S. H. Pan, A. K. Gupta, K. W. Ng and J. C. Davis, Science 285, 88
(1999).
14. P. Nozieres and A. Blandin, J. Physique 41,193 (1980).
15. D. L. Cox and A. Zawadowski, Adv. Phys. 47, 599 (1998).
16. D. M. Cragg, P. Lloyd and P. Nozieres, J. Phys. C, 13, 803 (1980).
17. H.-B. Pang and D. L. Cox, Phys. Rev. B 44, 9454 (1991).
18. R. Bulla, A. C. Hewson and G.-M. Zhang, Phys. Rev. B 56, 11721 (1997).
19. 1. Affleck and A. W. W. Ludwig, Nuc!. Phys. B 352, 849 (1991) and 360, 641
(1991), Phys. Rev. B 48, 7297 (1993).
20. O. Parcollet, A. Georges, G. Kotliar and A. Sengupta, Phys. Rev. B 58, 3794 (1998).
21. J. L. Smith and Q. Si, cond-mat/9705140, Europhys. Lett. 45, 228 (1999);
A. M. Sengupta, Phys. Rev. B 61, 4041 (2000).
22. L. Zhu and Q. Si, Phys. Rev. B 66, 024426 (2002); G. Zarand and E. Demler, Phys.
Rev. B 66, 024427 (2002).
23. S. Sachdev, C. Buragohain and M. Vojta, Science 286, 2479 (1999); M. Vojta, C.
Buragohain and S. Sachdev, Phys. Rev. B 61, 15152 (2000).
24. Q. Si, S. Rabello, K. Ingersent and J. L. Smith, Nature 413, 804 (2001).
25. A. Schroder, G. Aeppli, R. Coldea, M. Adams, O. Stockert, H. v. Lohneysen, E.
Bucher, R. Ramazashvili and P. Coleman, Nature 407, 351 (2000).
26. A. J. Leggett, S. Chakravarty, A. T. Dorsey, M. P. A. Fisher, A. Garg, and W.
Zwerger, Rev. Mod. Phys. 59, 1 (1987).
27. B. A. Jones and C. M. Varma, Phys. Rev. Lett. 58, 843 (1987); B. A. Jones,
C. M. Varma and J. W. Wilkins, ibid. 61, 125 (1988).
 217

28. O. Sakai, Y. Shimizu and T. Kasuya, Solid State Comm. 75, 81 (1990); O. Sakai
and Y. Shimizu, J. Phys. Soc. Jpn 61, 2333 (1992), ibid, 61, 2348 (1992).
29. 1. Affleck, A. W. W. Ludwig and B. A. Jones, Phys. Rev. B 52, 9528 (1995).
:30. M. Vojta, R. Bulla and W. Hofstetter, Phys. Rev. B 65, 140405(R) (2002).
31. W. G. van der Wiel, S. De Franceschi, J. M. Elzerman, S. Tarucha, L. P. Kouwen-
hoven, J. Motohisa, F. Nakajima and T. Fukui, Phys. Rev. Lett. 88, 126803
(2002).
INSTABILITY OF THE FERMI-LIQUID FIXED POINT IN AN
EXTENDED KONDO MODEL

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

M. LAVAGNA*
Commissariat al'Energie Atomique, DRFMC/SPSMS, 17, rue
des Martyrs, 38054 Grenoble Cedex 9, France
A. JEREZ
European Synchrotron Radiation Facility, 6, rue Jules Horow-
itz, 38043 Grenoble Cedex 9, France
D. BENSIMONt
Department of Appled Physics, Hongo 7-3-1, University of Tokyo,
Tokyo 113-8656, Japan

Abstract. We study an extended SU(N) single-impurity Kondo model in which the

impurity spin is described by a combination of Abrikosov fermions and Schwinger bosons.
Our aim is to describe both the quasiparticle-like excitations and the locally critical
modes observed in various physical situations, including non-Fermi liquid (NFL) behavior
in heavy fermions in the vicinity of a quantum critical point and anomalous transport
properties in quantum wires. In contrast with models with either pure bosonic or pure
fermionic impurities, the strong coupling fixed point is unstable against the conduction
electron kinetic term under certain conditions. The stability region of the strong coupling
fixed point coincides with the region where the partially screened, effective impurity repels
the electrons on adjacent sites. In the instability region, the impurity tends to attract
(N -1) electrons to the neighboring sites, giving rise to a double-stage Kondo effect with
additional screening of the impurity.

Key words: Non-Fermi Liquid, Strong versus Intermediate Coupling Fixed Point, Exten-
ded SU(N) Kondo Model

* Also at the Centre National de la Recherche Scientifique (CNRS)

t Also at the Commissariat a
l'Energie Atomique, DRFMCjSPSMS, 17, rue des
Martyrs, 38054 Grenoble Cedex 9, France

219
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 219–227.
© 2003 Kluwer Academic Publishers
220

1. Introduction

One of the most striking properties discovered in Heavy-Fermion systems

in recent years is the existence of a non-Fermi-liquid behavior (11, 21, 12)
in the disordered phase close to the magnetic quantum critical point with a
temperature dependence of the physical quantities which differs from that
of a standard Fermi-liquid. Recent results obtained in CeCu5.9Auo.l by
inelastic neutron scattering experiments (19, 20) have shown the existence
of anomalous wiT scaling law for the dynamical spin susceptibility at the
antiferromagnetic wavevector which persists over the entire Brillouin zone.
This indicates that the spin dynamics are critical not only at large length
scales as in the itinerant magnetism picture (7, 13) but also at atomic length
scales. It strongly suggests the presence of locally critical modes beyond the
standard spin-fluctuation theory. Alternative theories have been proposed
to describe such a local quantum critical point. In this direction, we will
mention recent calculations (18) based on dynamical mean field theory
(DMFT) which seem to lead to encouraging results such as the prediction
of a scaling law for the dynamical susceptibility. The other approach which
has been put forward to describe the local QCP is based on supersymmetric
theory (6, 17, 3, 4) in which the spin is described in a mixed fermionic-
bosonic representation. The interest of the supersymmetric approach is
to describe the quasiparticles and the local moments on an equal footing
through the fermionic and the bosonic part of the spin respectively. It
appears to be specially well-indicated in the case of the locally critical
scenario in which the magnetic temperature scale TN, and the Fermi scale
T K (the Kondo temperature) below which the quasiparticles die, vanish at
the same point 6c.
Let us now emphasize another aspect in the discussion of the break-
down of the Fermi-liquid theory. It has to do with the question of the
instability of the strong coupling (SC) fixed point. A stable SC fixed point
is usually associated with a local Fermi-liquid behavior. On the contrary,
an instability of the SC fixed point is an indication for the existence of
an intermediate coupling fixed point with non-Fermi-liquid behavior. The
traditional source of instability of the SC fixed point in the Kondo model
is the presence of several channels for the conduction electrons with the
existence of two regimes, the underscreened and the overscreened ones with
very different associated behavior. In the one-channel antiferromagnetic
single-impurity Kondo model, it is well known from renormalization group
arguments, that the system flows to a stable SC fixed point (2) with a
behavior of the system identified with that of a local Fermi liquid (14). The
situation is rather different in the case of the multichannel Kondo model
with a number K of channels for the conduction electrons (K > 1) (15). In
 221

the underscreened regime when K < 28 (where 8 is the value of the spin
in 8U(2)), the SC fixed point is stable. It becomes unstable in the other
regime K > 28, known as the overscreened regime. The underscreened
regime corresponds to the one-stage Kondo effect with the formation of an
effective spin (8 - 1/2) resulting from the screening of the impurity spin
by the conduction electrons located on the same site. The system described
by the SC fixed point behaves as a local Fermi liquid. The instability of
the SC fixed point in the overscreened regime is associated with a multi-
stage Kondo effect in which successively the impurity spin is screened by
conduction electrons on the same site, and then the dressed impurity is
screened by conduction electrons on the neighboring site and so forth. The
overscreened regime leads to the existence of an intermediate coupling fixed
point with non-Fermi-liquid excitation spectrum and an anomalous residual
entropy at zero temperature. It has been recently put forward (3, 4) that
other sources of instability of the SC fixed point may exist other than
the multiplicity of the conduction electron channels. Recent works have
shown that the presence of a more general Kondo impurity where the spin
symmetry is extended from 8U(2) to 8U(N), and the impurity has mixed
symmetry, may also lead to an unstability of the SC fixed point already
in the one-channel case. In the large N limit, Coleman et al. (3) have
found that the SC fixed point becomes unstable as soon as the impurity
parameter q (defined below) is larger than N /2 whatever the value of 28
(defined below) is, giving rise to a two-stage Kondo effect. This result opens
the route for the existence of an intermediate coupling fixed point with
eventually non-Fermi-liquid behavior.
It is worth noting that the supersymmetry theory, or more specific-
ally the taking into account of general Kondo impurities appears to offer
valuable insights into the two issues raised by the breakdown of the Fermi
liquid theory that we have summarized above, i.e. both the existence of
locally critical modes and the question of the instability of the SC fixed
point. Somehow it seems that the consideration of general Kondo impurities
captures the physics present in real systems with the coexistence of the
screening of the spin by conduction electrons responsible for the formation
of quasiparticles , and the formation of localized magnetic moment that
persists and eventually leads to a phase transition as the coupling to other
impurities becomes dominant.
The aim of this work is to study the 8U(N), extended single-impurity
Kondo model in the one-channel case (10). We would like to investigate
how the system behaves when not only the values of the parameters (28, q)
of the representation vary, but also the number nd of conduction electrons
on the neighboring site does. We want to discuss the effect of nd on the
stability of the SC fixed point and to further understand the nature of the
222

screening with the possibility of achieving either a one-stage, two-stage or

multi-stage Kondo effect depending on the regime considered. Implications
for the behavior of physical quantities will be given.

2. Extended 8U(N) single-impurity Kondo model

We consider a generalized, single-impurity, Kondo model with one channel

of conduction electrons and a spin symmetry group extended from 8U(2)
to 8U(N). An impurity spin, S, is placed at the origin (site 0). We deal
with impurities that can be realized by a combination of 28 bosonic (b&)
and q - 1 fermionic (Jl) operators. The hamiltonian describing the model
is written as

H = LEkCL,Ck,a + JLS A Lc1(0)T~;3c;3(0) , (1)

k,a A a,;3

where ckt ,a is the creation operator of a conduction electron with momentum

k, and 8U(N) spin index a = a, b, ... , TN, c&(O) = k ~k cta , where Ns
is the number of sites, and T~;3 (A = 1, ... , N 2 1) are the generators
-

of the 8U(N) group in the fundamental representation, with TT[TAT B ] =

(jAB/2. The conduction electrons interact with the impurity spin SA (A =
1, ... ,N2 - 1), placed at the origin, via Kondo coupling, J > O. When
the impurity is in the fundamental representation, we recover the Coqblin-
Schrieffer model (5, 8) describing conduction electrons in interaction with
an impurity spin of angular momentum j, (N = 2j + 1, a = j, b = j -
1, ... ,TN = -j), resulting from the combined spin and orbit exchange
scattering.

2.1. STRONG-COUPLING FIXED POINT

We consider the case where J -----+ 00 and we can neglect the kinetic energy
in Eq. 1. In this limit the model can be solved exactly in terms of the
invariants associated to the spin of the electrons and of the impurities (9).
The eigenvalues are of the form

(2)

where I denotes the impurity spin, Y the spin of the n c conduction electrons
coupled to the impurity at the origin, and Rsc the spin of the resulting SC
state at the impurity site. The quantities C2 are the 8U(N) generalization
of the 8U(2) eigenvalues, 8(8+1), and can be readily evaluated (for details,
see Ref. (10)).
 223

The ground state corresponds to having n c = (N - q) electrons at the

origin, partially screening the impurity. It can be written explicitly as the
action of (28 - 1) bosonic operators on a singlet state. For instance the
highest spin state can be written as

with, :::::: )(28 + N - l)Cir-_\. Here, cl :::::: cl(O). We denote the ground
state energy by Eo.
The effect of the kinetic term in Eq. 1 is, to lowest order in perturbation
theory, to mix the ground state with excited states where the number of
electrons changes by one. There are three such states, which we denote
by IG8 + l)S, IG8 + l)A and IG8 - I). The labels S(A) indicate that
the additional electron is coupled symmetrically(antisymmetrically) to the
ground state. The states are readily obtained by deriving the relevant SU(N)
Clebsch-Gordan coefficients (10).

2.2. STABILITY OF THE STRONG COUPLING FIXED POINT

In order to understand better the low-energy physics of the system, we

should consider the finite Kondo coupling, allowing virtual hopping from
and to the impurity site. These processes generate interactions between the
composite at site 0 and the conduction electrons on neighboring sites, that
can be treated as perturbations of the SC fixed point. The energy shifts
due to the perturbation can be reproduced by introducing an interaction
between the spin at the impurity site and the spin of the electrons on the
neighboring site, with effective coupling Jeff. Applying an analysis similar
to that of Nozieres and Blandin (15) to the nature of the excitations, we
can argue whether or not the SC fixed point remains stable once virtual
hopping is allowed.
Thus, if the coupling between the effective spin at the impurity site and
that of the electrons on site 1 is ferromagnetic we know, from the scaling
analysis at weak coupling, that the perturbation is irrelevant, and the low
energy physics is described by a SC fixed point. That is, an underscreened,
effective impurity weakly coupled to a gas of free electrons with a phase shift
indicating that there are already (N - q) electrons screening the original
impurity.
If, on the contrary, the effective coupling is antiferromagnetic, the per-
turbation is relevant, the SC fixed point is unstable and the low-energy
physics of the model corresponds to some intermediate coupling fixed point,
to be identified.
224

We explicitly calculate the effects of hopping on the SC fixed point

to the lowest order in perturbation theory, that is, second order in t. We
consider the case with an arbitrary number nd of conduction electrons on
site 1 generalizing the case nd = 1 considered in ref. (4).
Adding nd electrons on site 1 leads to two different states, that we will
call symmetric (S) and antisymmetric (A). These states, degenerate in the
SC limit, acquire energy shifts, b..Etf and b..E6 due to the perturbation
given, in the large-N limit, by
2
b..E A = _ (2t ) [( N - q) _ nd (N (N - 2q ))] , (4)
o J q N q(N - q)

_ (25 + nd - 1) (~) (N(N - 2q )) (5)

25+N-q IN q(N-q)
Jeff ( 25
-2- + nd - 1) .

Notice that the behavior of both Eq. 4 and Eq. 5 are controlled by the
same factor. This result has the immediate following physical consequence.
The change of sign of Jeff (Fig. 1, Right) -and hence of the stability of
the SC fixed point- is directly connected to the change in the behavior of
b..Etf rv b..E6 with nd. In particular, when b..E6 = b..Etf, b..Etf = -(2t 2 I J)
independently 1 of nd, q or 25. In the regime where the SC fixed point
is stable, qlN < 1/2, Jeff < 0, the lowest energy corresponds to nd =
1, whereas for qlN > 1/2, Jeff> 0, the energy expressed in Eq. (4) is
minimized for nd = (N - 1) (Fig. 1, Left). This is precisely the mechanism
behind the two-stage quenching. The accumulation of electrons on site 1
is not related to Jeff which is independent of nd, but results from the
dependence of b..Etf rv b..E6 with nd.
We finish by making some remarks on the physical properties of the
model in the different regimes. As is common to all models with an antifer-
romagnetic Kondo coupling, there will be a crossover from weak coupling
above a given Kondo scale, T K , to a low-energy regime. When the SC
fixed point is stable, we should expect for T « T K a weak coupling of
the effective impurity at site-O with the rest of the electrons. The physical
properties at low temperature are controlled by the degeneracy of the ef-
fective impurity, d([25 -1]) = C~+iS-2' Thus, we should expect a residual
entropy Si rv In C~+iS-2 and a Curie susceptibility, Xi rv C~+is_2IT, with
logarithmic corrections (9, 16). This is the result that we would expect for
a purely symmetric impurity. Contrary to the purely bosonic case, only

1 This is true for arbitrary values of N, and 28.

 225
)
I
I
-",,,,
I
I
-1 I
I
I

_ -2 ~ . / /

~g ~ I,'
oz °f---==~:;-:f'':'''''''''''==------1
,,/."
,.""
<~
<I -3 -,~ ...
/' .'
-2 /
I
-4 1- - nd~
- n =N-l
d
1 I I
I

-4 I
I
:
0,25 0,5 0,75 ••• (125 0,5 (J,75
gIN q/N

Figure 1. (Left) Leading term in the energy shift, 6.E(i ~ 6.E5, as a function of q/N,
for 1 < nd < N - 1 (shaded region), and in the limiting cases nd/N « 1 (dashed line),
and nd/N ~ 1 (solid line). Notice that the value at q/N = 1/2 is equal to J, for any -2e/
nd. (Right) Energy difference, (6.E[~ - 6.E((), as a function of q/N, for different values
of 28 in the large-N limit.

(N - q) electrons are allowed at the ongm in the SC limit, instead of

(N - 1). Thus, we would expect to find different results for quantities that
involve the scattering phase shift of electrons off the effective impurity.
In the q > N /2, we do not have access to the intermediate coupling
fixed point that determines the low-energy behavior. Nevertheless, it is
reasonable to think that there would be a magnetic contribution to the en-
tropy, and a Curie-like contribution to the susceptibility, since the impurity
remains unscreened. This behavior is different from that of the multichannel
Kondo model, which is also characterized by an intermediate coupling fixed
point, but where the impurity magnetic degrees of freedom are completely
quenched. The degeneracy of the true ground state is an open question, but
we can assume that the entropy will be smaller than that of the SC fixed
point (1). It is in the scattering properties that we might be able to see the
anomalous features of this new fixed point more clearly.

3. Conclusions

We have studied a Kondo model where the spin of the impurity has mixed
symmetry, as a way of incorporating the phenomena of local moment screen-
ing and magnetic correlation that is observed in some heavy fermion com-
pounds. Such model is naturally realized by extending the spin symmetry
to SU(N), and it displays a rich phase diagram. We find that as long as
the bosonic component of spin is of order N, there is a transition around
226

the point where the fermionic component of the impurity is q = N /2. At

this particular point, the energy shift is, to lowest order in perturbation
theory around the SC fixed point, equal to -2t 2 / J, independent of the
impurity parameters, q, Sand N. When q < N /2, the low-energy physics
corresponds to the SC fixed point. For q > N /2 the SC fixed point is
unstable and anomalous behavior is expected, in particular, the two-stage
quenching effect. This phase diagram is not accidental, but is due to the
relation of the effective impurity site in the SC regime to the conduction
electrons in neighboring sites, as our study of the dependence of the energy
shifts on nd reveals. If q < N /2, the energy is minimized when the dressed
impurity repels the electrons on the next site. That is, when nd = 1. At
q = N /2, the energy shift is also independent of nd. Finally, the lowest
energy shift for q > N / q corresponds to a maximal nd, indicating the
accumulation of electrons leading to the two-stage Kondo quenching.

Acknowledgements

We would like to thank V. Zlatic and A. Hewson for organizing this very
interesting meeting. We are most grateful to N. Andrei for continuous en-
couragement and discussions. We also thank S. Burdin, P. Coleman, Ph.
Nozieres, and C. Pepin for helpful discussions.

References

1. 1. Affleck and A. W. W. Ludwig, Phys. Rev. B 48, 7297 (1993).

2. P. Anderson, Phys. Rev. B 164, 352 (1967).
3. P. Coleman, C. Pepin, and A. M. Tsvelik, Phys. Rev. B 62, 3852 (2000).
4. P. Coleman, C. Pepin, and A. M. Tsvelik, Nucl. Phys. B 586, 641 (2000).
5. B. Coqblin and J. R. Schrieffer, Phys. Rev 185, 847 (1969).
6. J. Gan, P. Coleman, and N. Andrei, Phys. Rev. Lett. 68, 3476 (1992).
7. J. Hertz, Phys. Rev. B 14, 1165 (1976).
8. A.C. Hewson, The Kondo problem to heavy fermions. (Cambridge University Press,
1993).
9. A. Jerez, N. Andrei, and G. Zanind, Phys. Rev. B 58, 3814 (1998).
10. A. Jerez, M. Lavagna, and D. Bensimon. Preprint.
11. H. v. Lohneysen, J.Phys.: Condo Mat. 8,9689 (1996).
12. N. Mathur, F. Grosche, S. Julian, 1. Walker, D. Freye, R. Haselwimmer, and G.
Lonzarich, Nature 394, 39 (1998).
13. A. Millis, Phys. Rev. B 48, 7183 (1993).
14. P. Nozieres, J. Phys. (Paris) 37, CI-271 (1976).
15. P. Nozieres and A. Blandin, J. Phys. (Paris) 41, 193 (1980).
16. O. Parcollet, A. Georges, G. Kotliar, and A. Sengupta, Phys. Rev. B 58, 3794
(1998).
17. C. Pepin and M. Lavagna, Phys. Rev. B 19, 12180 (1999).
 227

18. Q. Si, S. Rabello, K. Ingersent and J.L. Smith, Nature 413, 804 (2001).
19. A. Schroder, G. Aeppli, E. Bucher, R. Ramazashvili, and P. Coleman, Phys. Rev.
Lett. 80, 5623 (1998).
20. A. Schroder, G. Aeppli, R. Coldea, M. Adams, O. Stockert, H. v. Lohneysen, E.
Bucher, R. Ramazashvili, and P. Coleman, Nature 407, 351 (2000).
21. Steglich, F., B. Buschinger, P. Gegenwart, M. Lohmann, R. Helfrich, C. Langham-
mer, P. Hellmann, L. Donnevert, S. Thomas, A. Link, C. Geiber, M. Lang, G. Sparn,
and W. Assmus, J. Phys.: Condo Mat. 8, 9909 (1996).
PROJECTION OF THE KONDO EFFECT BY RESONANT
EIGENSTATES INSIDE A CIRCULAR QUANTUM CORRAL

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

A. LOBOS and A. A. ALICIA

Centro At6mico Bariloche and Instituto Balseiro, Comisi6n
Nacional de Energia At6mica, 8400 Bariloche, Argentina

Abstract. We calculate the resonant eigenstates inside a circular quantum corral, mod-
eling the confining potential by a continuous radial 6-function. In order to describe the
corral with a magnetic impurity inside, we use the impurity Anderson model and second
order perturbation theory in the Coulomb repulsion U. The effect of the impurity in
the conduction electron density of states reveals a mirage of the Kondo resonance at a
position different from that of the impurity.

Key words: Quantum Corrals, Mirage, Surface States, Magnetic Impurity, Kondo Effect,
Anderson (Impurity) Model, Perturbation Theory, Differential Conductance, Nanoscale
Structures, Scanning Tunneling Microscopy

1. Introduction

Nanoscale structures called "quantum corrals" have been constructed using

recent developments in scanning tunneling microscopy (STM) techniques.
These corrals are assembled by depositing a close line of atoms on Cu or
noble metal (111) surfaces (1, 2). The Fermi surface in these metals is
nearly spherical, with eight necks at the (111) and equivalent directions.
For small wave vectors parallel to the (111) surface, these metals have a
nearly parabolic band of two dimensional (2D) surface states, which are
uncoupled to bulk states and cross the Fermi energy Ep (3).
A circular corral of radius 71.3 A was constructed depositing 48 Fe atoms
on a Cu(111) surface (1). Measured differential conductance dI/ dV can be
fitted by a combination of the density of eigenstates close to Ep, of a 2D
electron gas inside a hard wall circular corral (1). In another experiment,
a Co atom acting as a magnetic impurity was placed at one focus of an
elliptical quantum corral, and the corresponding Kondo feature in dI/ dV

229
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 229–237.
© 2003 Kluwer Academic Publishers
230

was observed not only at that position, but also with reduced intensity at
the other focus (2). This "mirage" is the projection of the Kondo effect,
transmitted from the impurity to distant points by the corral eigenstates.
Similar experiments were done in circular corrals (4). The corral confine-
ment is needed in order to separate the conduction electronic energy levels
(otherwise forming a continuum) and reduce the destructive interference
produced by incoherent superposition of standing waves (5). Among these
eigenstates, the density of the one closest to Ep is clearly displayed in the
change of differential conductance b.dI/dV after adding the impurity (2).
Some theories of the Kondo mirage have attempted to fit or explain the
experimental data assuming the existence of the Kondo resonance, without
actually calculating it (5,6,7). The line shape of b.dI/dV, in the many-
body approaches which can explain it, is very sensitive to the width of
the resonance levels, and this width was taken as a parameter (8, 9, 10).
If this width is reduced to zero, as in ordinary Lanczos calculations (11),
the Kondo resonance splits into two delta functions out of Ep and the
experimental line shape cannot be reproduced. This problem persists if
hopping between impurity and bulk is considered, in spite of the broadening
of the delta functions (12). On the other hand, if the width of the resonance
is very large, the mirage disappears (8).
In this work we calculate the widths of the conduction states 8i for a
circular corral and solve the impurity Anderson model which describes the
mirage experiment using perturbation theory in the Coulomb repulsion U.

2. Resonances in a circular corral

Here we consider first the corral without impurity. We model the boundary
of an empty circular corral by a continuous potential W 8(r - ro), where r is
the distance to the center of the circle and ro its radius. The approximation
of a continuous boundary (instead of discrete adatoms forming the corral)
simplifies considerably the mathematics and is justified by the fact that the
Fermi wave length for surface electrons 2Jr/ kp ':::' 30 A is larger than the
average distance between adatoms rv 10 A.
The angular momentum projection perpendicular to the surface lz =
m is a good quantum number. For each m and energy E = (nk)2 /2m~,
where m~ is the effective mass, the eigenstates can be written in the form
!.pkm(r, e) = ?/Jkm(kr) e ime in polar coordinates. The solutions of the 2D
radial Schrodinger equation with this potential can be written in the form
?/Jkm = ?/J::-me(ro - r) + ?/Jkme(r - ro), where e(r) is the step function and

?/J::-m = f Jm(kr), (1)

 231

where Jm (Ym ) is the Bessel function of the first (second) kind. We nor-
malize the wave functions CPkm(r, e) inside a hard wall box of large radius
R and take the limit R ---7 00. In this limit, the eigenstates of the problem
form a continuum in the k variable. However, inside the corral, resonances
are formed for finite W (13, 14). The resonant eigenstates of the problem
can be obtained by scattering theory methods, specifically, by finding the
complex poles of the scattering matrix S(k) (13, 14).
The essential information on the spectral properties observed by STM
is contained in the surface conduction electrons Green's function. In the
absence of the impurity, it can be written in the form

GO(. e ' e') = '"" CPkm(r, e)Zj5km(r', e') (2)

sZ,r"r, ~ z-(kn)2/2m~'

where m~ is the effective electron mass. The sum can be converted to an

integral by standard methods using the density of allowed k. Inside the
corral we obtain:

GO(Z; r, e, r', e') = L J dk k Jm(kr)Jm(kr') eim(e-e'). (3)

s m 27f (A~ (k) + B~ (k)) (z _ ~::;~2)

This expression for G~ can be written as a sum of contributions from

the resonances (12, 13):

J (k m )J (k m ,) im(e-e')
GO(z'r e r' e') =
s , , , ,
Le m
n m n r m
Z -
n r e
E~~ + io~~ ,
(4)
n,m,

where k:';: are the poles in complex k-space of the scattering matrix (m
and n are angular and radial quantum numbers respectively) (15), with
( nknm )2/2m*r 2 = Em - io m and om > 0 Here Em is the energy of the
eo n 71,' n . 'n
resonant eigenstate, 0:';: > 0 is its width, and e~ are the residues at the
poles of the integral in Eq. (3) (12). We have taken an effective mass m~ =
0.38 m e (1, 16). Jm(kr) has no singularities in the complex k-plane. The
poles appearing in Eq. (4) are poles of the normalization factor f (when
continued into the complex k-plane). From the properties of 1/;;;n and 1/;f:rn'
and its derivatives at the boundary ro, one obtains for f:

2 k
(5)
f = 2R(A~(k, W) + B~(k, W))'
where the coefficients Am(k, W) and Bm(k, W) depend on Jm(krO) and
Ym(krO) and the dependence on W is now explicit. In Fig l(a) we show
the adimensional product f2 Rro, proportional to the density of states, as
a function of k (real) for two different values of Wand for states with
232

50
(a) (b)
>' 150
40
.............. W = 1.1geV.A
Q)
.§. 120
,,

\,
30 - - W = 2.38 ev.A
....
0
90
0::: 20
60
~
10
[ ... 30
------.

0
./ ~(,Q

0
0 3 6 9 12 15 18 21 0.0 0.5 1.0 1.5 2.0 2.5 3.0
k [rt W [eV.A]

Figure 1. (a)f2 Rro as a function of radial momentum k, for m=O states. (b) Width of
the resonant state n = 4, m = 1, 5l as a function of the confinement parameter W.

quantum number m=O. As k approaches a pole in the complex plane k~,

P Rro tends to diverge, and the weight of the resonant eigenfunction outside
the corral becomes very small in comparison to that inside (J depends
essentially on 'l/Jfm' since the area of the corral compared to that of the hard
wall box is of order (rol R)2) . The confinement of the resonant eigenstate
within the corral, the width 6i and the energy eigenvalue Ei depend on the
magnitude of the parameter W. In the limit W -----+ 00, P Rro becomes a
sum of delta functions at zeros of Jm(a) and the hard wall eigenstates are
recovered (1). This can be seen in Fig l(a). When W is increased, the poles
move towards the real k axis in the complex plane, while the weight of the
states with k # Ile k~ decreases, leading to the formation of sharp peaks
in the conduction electron density of states for k = Re kr;:.
As a first approximation, the resulting 6~ I E~ is a constant, independent
of m and n. As W decreases, the confinement properties of the corral
decrease and the width 6i must increase. This is shown in Fig l(b) where
we have studied the dependence of 6l. Another effect of confinement is that
E~ decreases slightly with increasing W. This effect can be interpreted as
a reduction of the effective mass, plus smaller corrections.

3. The many-body problem

In this section we introduce the magnetic impurity in the form of one non-
degenerate localized d orbital. We consider surface states, described inside
the corral by the Green's function G~, hybridized with the d orbital, for
which there is an important intra-orbital repulsion U (5, 8, 9, 10, 11).
We also include a hybridization of the d orbital with bulk states. The
 233

61~-----'--~-----'---~~~~
--pd(m)
5
···········pdo(m)
4
5:'
~3

~2
0:
1
o -_.__._.._...-_...-... ~. ..- _--- _- .

-1.0 -0.5 0.0 0.5 1.0

Figure 2. Impurity density of states for u=o (dotted line) and U=l eV (solid line).
Parameters are OF = 20 meV, Ob = 35 meV and Vs = 0.45 eV, and E~ff = EF.

Hamiltonian can be written as:

H = L EjS}iTSjiT + L E~b}iTbjiT + Ed L dLdiiT + U d~dTdl d 1

jiT jiT liT

Neglecting small distortions very near the impurity, dI / dV is propor-

tional to Ps(w, T, e) = -1m G s /1f, where G s is the Green's function for
conduction electrons, in presence of the impurity (9). Using equations of
motion it can be shown that:

where Gd is the impurity Green's function, R t and R i are the position of

the STM tip and the impurity respectively, and ).2 is the area per eu atom
in the surface. To calculate Gd we used perturbation theory up to second
order in U (17). We assume independence of spin projection (no magnetic
field applied) and a situation in which the effective one-particle d level
E~ff = Ed + U(dtdiTJ is very near the Fermi level (18). We took U = 1 eV.
Then:
(8)
where G~ is the impurity Green's function for U = 0 and is also obtained
from the equations of motion (12):

GO(w) = 1 (9)
d w-E~ff +i(5b-(V).)2G~(Ri,Ri,W)·
The term (5b represents here the hybridization between bulk states and the
impurity d level, and is a parameter in our calculations. The function I;(w)
234

/
0.9-r-r~~~---.--.-~~~~---,--"
0.0 +--::,------~~-----,.._~~~_____1
0.6 ••.....
~
>
-0.3
:>
~ 0.3
.....

~ -0.6

8' -0.9 ~ _::: + - - - -....-/-Jljf-+--li-------j

H -1.2
E ~ -0.6
-1.5
-0.9-L...,-~~~-,-!--c~~~---,---J
-0.9 -0.6 -0.3 0.0 0.3 0.6 0.9 -0.9 -0.6 -0.3 0.0 0.3 0.6 0.9
0) leV] 0) leV]

Figure 3. Real and imaginary parts of the self-energy for sl = 20 meV (solid line) and
sl = 40 meV (dotted line).

is the self-energy of the problem up to second order in U (17):

"E(iwz, T) (10)
m

-TL G~(iwn)G~(iwn + ivm ), (11)

n

where T is the temperature, W n and V m are fermionic and bosonic Matsub-

ara frequencies respectively. A more detailed explanation of this calculation
is in Ref. (12).
In Fig. 2 we show the impurity spectral densities of states, for two
different values of U, and 6~ = 20 meV. In Fig 3 we display a self-energy
calculation for these parameters and another calculation for a different set
of parameters. The origin of energies was set at EF = O. We see that, as
a consequence of the structure of the real part of "E (w) near EF, a narrow
peak in Pd develops there (see Fig. 2). This is the Kondo resonance. The
imaginary part of "E (w) is proportional to -w 2 , as it should be for a Fermi
liquid. For U = 1 eV, also the usual broad peaks at energies near Ed
and Ed + U are clearly seen. For the results presented here, as in recent
experiments (4), we have taken TO = 63.5 A, so that the energy level of the
state with m = 1, n = 4 falls at EF. We took W = 7fj,2 /2m~To ':::::' 1.19 eV A
in order that the width of this state would be 6~ ':::::' 40 meV (8). This leads
to a good agreement with experiment in the case of the elliptical corral
(8, 9). The point of observation T = 0.14 TO was taken at the maximum
of Jl(k~T). As 6F decreases (see Fig. 3), the structure of the resonances in
G~(w) are transmitted to Gd(w) and "E(w), and several peaks appear. For
6F ----+ 0 the Kondo resonance is destroyed (8, 9, 10).
In Fig. 4 we show the space dependence of tlps( EF) and the density of
the corresponding hard wall eigenstate with quantum numbers m = 1 and
n = 4. We used ~ = 0.48 eV, 6b = 0.032 eV and the broadening of the
 235

0.03
6.:0.02
....,..0.01 1.0
"'.0.00 0.5
cc
<1- 0.01 o.o~c
• .>;
-1.0 -0.5 -0.5
-1.0

Pigl/,re 4. (a) Space dependence of -6.(:-.(£1") and (b) dell~i1.y of the llurd wall cigcn~tatc
with m = 1 and n = 4,

eigenstate in €p is bp = 6J = 40 meV. These parameters lead to a good

description of the reported line shape of !:J.(U !dV for the elliptical corral
(2).The largest peak occurs at the position of the impurity and the other,
weaker in intensity, is the so called mirage. The correspondence between
both figures is clearly seen. However, the mirage is somewhat distorted
because of destructive interference of other resonant states which energy
eigenvalues lies llcar fp. An analysis of the voltage dependence of .6.pA€p)
is in Refs.(9, 12).

4. Summary and discussion

\Ve have studied the resonances produced inside a circular quantum conal,
assuming a continuous W6(1' - TO) confining potential in a 20 electron
gas. These resonant. eigenstates are very similar to the eigenstal.es of a hard
wall potential and t.hey can be obtained from the poles of the normalization
factor f. This fact. can be easily interpret.ed, since a pole in f indicates that
the state has little weight outside the corral, leading to a "quasi-confined"
state. The introduction of a width iJ.; for each resonanl.:e is the main differ-
ence with a calculation assuming a hard wall corral. Our model leads to a
resonance &., proportional to its energy lSi, in a. first approximation.
236

We used the Anderson model to describe the physics when a magnetic

impurity is introduced inside the corral. At T < TK, the Kondo effect is
observed in the impurity spectral density of states as a sharp peak at EF.
The resonant conduction state which lies at EF can be observed because of
two factors: the low energy scale of the Kondo effect (the width of the Kondo
peak is 2kB TK ':::::' 10 meV), and the energy level separation of resonant
states which hybridizes significatively with the impurity, due to the corral
confinement. In our calculations, this separation ranges from 50 to 100
meV. This avoids an incoherent superposition of waves. Nevertheless, the
broadening introduced by finite W increases the destructive interference
because at EF the density of states has non-negligible components of other
states, different to that we want to project. A reduction in 6i leads to a
stronger intensity at the mirage point (because of reduction of destructive
interference), and the space dependence tends to that of the conduction
state at EF. However, for very small 6i the voltage dependence, is strongly
distorted and two peaks of positive weight of tldI/dV appears at moderate
non-zero voltages (9).
So far, the theories of the quantum mirage have neglected the effect of
sand p states brought by the impurity atom. The introduction of these
states should affect the intensity of dI/ dV near the impurity. Therefore, a
study of this effect is desirable.

Acknowledgements

One of us (AA) is partially supported by CONICET, Argentina. This work

benefited from PICT 03-06343 of ANPCyT.

References

1. M.F. Crommie, C.P. Lutz, and D.M. Eigler, Science 262, 218 (1993).
2. H.C. Manoharan, C.P. Lutz, and D.M. Eigler, Nature (London) 403, 512 (2000).
3. S.L. Hulbert, P.D. Johnson, N.G. Stoffel, W.A. Royer, and N.V. Smith, Phys. Rev.
B 31, 6815 (1985).
4. H.C. Manoharan, PASI Conference, Physics and Technology at the Nanometer Scale
(Costa Rica, June 24 - July 3, 2001).
5. D. Porras, .J. Fermindez-Rossier, and C. Tejedor, Phys. Rev. B 63, 155406 (2001).
6. O. Agam and A. Schiller, Phys. Rev. Lett. 86, 484 (2001).
7. G.A. Fiete, J. S. Hersch, E. J. Heller, H.C. Manoharan, C.P. Lutz, and D.M. Eigler,
Phys. Rev. Lett. 86, 2392 (2001) ..
8. A.A. Aligia, Phys. Rev. B 64, 121102(R) (2001).
9. A.A. Aligia, Phys. Status Solidi (b) 230,415 (2002).
10. G. Chiappe and A.A. Aligia, Phys. Rev. B 66, 075421 (2002).
11. K. Hallberg, A.A. Correa, and C.A. Balseiro, Phys. Rev. Lett. 88, 066802 (2002).
 237

12. A.Lobos and A.A.Aligia, cond-mat/0208533

1:3. G. Garcia Calderon, Nuc!. Phys. A 261, 1:30 (1976).
14. See for example J.R. Taylor, Scattering Theory: the quantum theory of non-
relativistic collisions (Wiley, New York, 1972).
15. The Green's function and the scattering matrix have the same distribution of poles
in the complex k-plane (1:3).
16. A. Euceda, D.M. Bylander, and L. Kleinman, Phys. Rev. B 28, 528 (1983).
17. B. Horvatic, D. Sokcevic, and V. Zlatic, Phys. Rev. B 36, 675 (1987); references
therein.
18. M. Weissmann and A.M. Llois, Phys. Rev. B 63, 11:3402 (2001).
NONEQUILIBRIUM ELECTRON TRANSPORT THROUGH
NANOSTRUCTURES: CORRELATION EFFECTS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

P. WOLFLE, A. ROSCH, J. PAASKE and J. KROHA

Institut fur Theorie der Kondensierten Materie, UniversitCit
Karlsruhe, 76128 Karlsruhe, Germany

Abstract. We consider electron transport through a quantum dot in the Coulomb block-
ade regime, when the dot has a total spin ~. The low energy physics of this system is
modeled by a Kondo-type Hamiltonian describing spin-dependent tunneling and exchange
interaction of the electron spins in the leads and the local spin. We show that in the regime
of large transport voltage V and magnetic field B with max(V, B) » T K , where T K is the
Kondo temperature, renormalized perturbation theory is valid. Physical quantities such
as the differential conductance G and the local magnetization 1\;1 may be calculated in a
controlled way by summing the leading logarithmic terms in each order of perturbation
theory. We develop a poor man's scaling renormalization group treatment for frequency
dependent coupling functions, which allows one to calculate G and 1\;1 to leading order
in l/£n[(V, B)/TK].

Key words: Nonequilibrium Transport, Quantum Dot, Kondo Effect, Conductance, Poor
Man's Scaling, Renormalization Group Method

1. Introduction

The transport of electrons through a quantum dot in the limit of weak

coupling to the leads is dominated by Coulomb interaction effects, forcing
integral electron charge on the dot (1) (Coulomb blockade). In the case
that the total spin of the dot is nonzero, however, the antiferromagnetic
exchange interaction of this local spin with the conduction electron spins
in the leads gives rise to a Kondo resonance in the local density of states
at the Fermi level. Electron transport may then take place via resonance
tunneling, thus removing the Coulomb blockade (2, 3). This has been seen
in a number of experiments (4, 5, 6, 7, 8).

239
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 239–246.
© 2003 Kluwer Academic Publishers
240

Powerful methods have been developed to describe this strong coup-

ling problem, among them methods based on the renormalization group
idea, in a simplified form (9) ("poor man's scaling") and in a full nu-
merical implementation by Wilson (10), exact methods for calculating the
thermodynamic quantities using the Bethe ansatz (ll), conformal field the-
ory (12) and resummations of perturbation theory using auxiliary particle
representations (13) to access dynamical quantities.
Necessary prerequisites for the existence of the Kondo effect are (i) the
spin degeneracy of the ground state of the quantum dot, and (ii) quantum
coherence over sufficiently long time periods. The characteristic energy scale
against perturbations have to be measured is the Kondo temperature TK ~
D)JNo exp(-1/2No J), where J is the exchange coupling constant, No is
the conduction electron density of states per spin at the Fermi level and D
is the conduction band width.
Nonlinear transport through a Kondo system has been considered by
several authors. Early on, Appelbaum (15) calculated the differential
conductance G(V, B) for large bias voltage V in a finite magnetic field
B, assuming erroneously, however, that the local spin remains in thermal
equilibrium. Konig et al. (16) calculated G(V, B) for an Anderson model,
employing a certain resummation of perturbation theory. Brief accounts of
the work to be presented below have been published elsewhere (14, 17).

2. Magnetization and Conductance in Perturbation Theory

We study the following Kondo-type model Hamiltonian of a quantum dot:

H = L (Ek - Ma)C:ko-Cako- - BSz

k,o-,a=L,R

+-1 "
L...J Ja /a S· t
(ca/k/o-,To-'o-Cako-), (1)
~,k' ,0-,0-' ,a,o:' =L,R
where S is the local spin operator on the dot (assumed to be S = ~) and
T is the vector of Pauli matrices. The c~ko- create conduction electrons
with momentum k and spin (]" in lead a. The chemical potential shifts
induced by the bias voltage are given by ML,R = ± V /2 for the left (L) and
right (R) lead. The exchange energies Jaa /, a = L, R, will be assumed
to have the symmetry J RR = J LL and J LR = J RL . We shall use the
dimensionless coupling constants gd = NOJRR and gLR = NOJLR , with
No the local density of states at the Fermi energy, assumed to be flat in
the accessible regime I w I~ V, B. A magnetic field is included, inducing
a Zeeman splitting of the magnetic levels of the local spin = -,B /2, w,
,= ±1. The effect of the magnetic field on the conduction electrons gives
 241

only rise to particle-hole asymmetry terms, which are usually small and
may be neglected.
In lowest order perturbation theory the current through the dot is given
by the Golden Rule expression

1= :~ Jdw L n, [gUw-ILL (1 -
,=±l
IW-ILR)

+ 2gUw-ILL(I- IW-ILR-,B)] - (L ~ R) (2)

where gl and g2 are the bare dimensionless coupling constants for spin-
nonflip and spin-flip interaction, gl = g2 = NOJLR, IW-lLu is the Fermi
function in the lead 0: = L, R, and IB is the Zeeman energy transfer taking
place in a spin flip process. At low temperatures, T « V, B, the product of
Fermi functions limits the energy integration to the window ILR < w < ILL
and ILR + IB < w < ILL, respectively. The current is seen to depend on
the occupation n, = ~ (1 + 1M) of the local spin states, where M is the
magnetization.
For sufficiently small voltage, V « T, the local spin system is in thermal
equilibrium. In the opposite limit, V » T, the stationary current through
the dot drives the system out of equilibrium. The occupation numbers n,
gt
are then determined by the rate equation n, = C,{n,'} = 0, where C
is the collision integral. This leads to the condition (assuming JLR = hL
here)

nr L
a,a'=L,R
J dwg§Iw-lLu ( 1 - IW-lLu,-B) =

= nl L
a,a'=L,R
J
dwgUw-lLu (1 - IW-lLu,+B) (3)

The spin-flip coupling constant g2 cancels out of this equation, yielding the
nonequilibrium magnetization as given by (5) in the limit J -----+ O.
In the limit V -----+ 0, the equilibrium result M = tanh 2~ is recovered,
whereas in the limit V » B, T, M = 2(/, independent of temperature. This
result has been obtained independently in (18, 20).
In order to calculate higher order contributions in perturbation theory
in J we switch now to a more formal description in terms of nonequilibrium
Green's functions. We find it convenient to represent the local spin operator
in pseudofermion (PF) language (19), S = ~ L", 1," where 1:;
I:;T",
creates a pseudofermion of spin I =1, 1 at the dot. The projection onto the
physical sector of Hilbert space, with pseudofermion occupation number
Q = L, 1:;1, = 1, is done by adding a term AQ to the Hamiltonian and
242

a c

E)+ YX
....
.,. ...... - -... ...,

Figure 1. Feynman diagrams for (a) PF self-energies, (b) current and (c) vertices
entering the I-loop RG equation. PF (electron-) propagators are displayed as dashed
(full) lines.

taking the limit A ----+ 00. This means that the PF system is taken in the
low density limit. The Feynman diagrams of perturbation theory in the
exchange interaction will therefore have one PF loop at most.
The local magnetization can be calculated in terms of the PF Green's
function G~(w) as M = L:f=±1 I J :f:iG~(w), where G< is found by solving
a quantum kinetic equation. In steady state one finds

G~(w)r,(w) = ~~(w)A,(w) (4)

where A, (w) is the PF spectral function, ~~ (w) is the lesser component of
the self-energy and r ,(w) its imaginary part. In general Eq. (4) is an integral
equation for G~(w) = in,(w)A,(w), which, however, can be approximated
in a controlled way by making use of the sharply peaked form of A, (w)
to give an algebraic equation for n,(O). In this form Eq. (4) corresponds
to the rate equation discussed above, with higher order processes included
in the transition amplitudes. The diagrams of the self-energy up to third
order in J are depicted in Fig. la. Note that the lines represent Keldysh
matrix Green's functions dressed with arrows in all topologically different
ways. The result including leading logarithmic corrections is found as

M = Z (5)
coth 2~ [~+ g~B(l + £(B))] + gIR(c(B) + c(-B))
where

z= 2g~B(1+£(B))+29IR[2B+(V +B)£(V+B)-(V-B)£(V -B)] (6)

and we use the abbreviations g~ = gJ ± gIR' £(B) = 2gd£nl~I' and

c(B) = coth V2j,B[(V + B)(l + £(V + B)) + V£(V) + B£(B)]. We note
that logarithmic corrections of order g3 £nD to the self-energies lead to
 243

corrections of order ginD to M. These corrections are larger than the

leading logarithmic corrections in equilibrium (V -----+ 0), which are of order
g2inD. A corresponding result of the current I is given in (17).

3. Resummation of Perturbation Theory in Nonequilibrium: a

Poor Man's Scaling Approach

Even for small coupling constants g and for sufficiently large V and B,
such that V, B » TK, but still in the scaling regime V, B « D, such that
gin(D IV) « 1, bare perturbation theory converges slowly. It is necessary
to sum the leading logarithmic contributions in all orders of PT. In the
equilibrium state a powerful method is available to perform this resumma-
tion in a controlled way: the perturbative renormalization group method
(9). It makes use of the fundamental idea that a change of the cut-off D
can be fully absorbed into a redefinition of the coupling constant g. As
long as the running coupling constant g(D) is small, the change of g under
an infinitesimal change of D, 8g/mnD, may be calculated in perturbation
theory. It is well known that in the equilibrium Kondo problem the coup-
ling constant is found to grow to infinity, thus leaving the perturbative
regime beyond g rv 1. A nonperturbative treatment is then necessary, as
shown by Wilson (10) in his pioneering work on the numerical RG. In
the nonequilibrium situation, the RG flow is cut-off by inelastic processes
already within the weak coupling regime, such that perturbation theory is
valid (14). This is true at least for the case hR = JLL, JRR, considered here.
While the discussion of the RG flow in (14) was a qualitative one, motivated
by results obtained in the so-called "Non-crossing approximation" for the
Anderson model, and applied only to the limit B = 0, here we consider the
RG formulation on a more fundamental level. A different and considerably
more involved real-time RG scheme has been developed by Schoeller and
Konig (21).
An analysis of the PT result Eq. (5) suggests that the frequency de-
pendence of the couplings becomes important. In order to understand how
these frequency dependencies are generated on the lowest level, i.e. in one-
loop order, it is sufficient to analyze the behavior of the vertex corrections
shown in Fig. lc under a change of cut-off. Logarithmic corrections in
perturbation theory are generated by PF - conduction electron bubbles
containing the product of a real part of a PF Green's function, 1/(w ±
B /2), and the Keldysh component of the local conduction electron line,
-27fiNo tanh[(w - J1o;)/2T]. If the energy of the PF - conduction electron
intermediate state is within the interval [- D, D] the process will contribute
to the renormalization of the coupling function g (w), otherwise it will not. In
general the coupling functions depend on three frequencies (taking energy
244

conservation into account). Using the fact that the spectral function of the
PFs is sharply peaked at w = ~B /2, we set two of the three frequencies to
w = ~B /2, keeping only one frequency variable. In addition, the frequency
dependence may be neglected within the running band-width, I w 1< D. The
spin structure of the general coupling functions is given by two invariant
amplitudes, fh for spin flip and fh for spin non-flip processes

where 9~.;;' denotes the coupling function for conduction electrons of spin (J"
and ene;gy w in lead 0:( a = (0:, (J", w)) interacting with a pseudofermion in
state b = (" w f) and going into states a', b'. The two frequency-dependent
running coupling functions .ih(w) and fh(w) obey the following flow equa-
tions,

8fh(w)
8lnD
8fh(w)
(8)
8inD

with initial condition .ih(w) = fh(w) = JNo and 8 w = 8(D- 1 w I). In the
limit V, B ----7 0, Eq. (8) reduces to the well-known scaling equations (6).
There is, however, still one additional effect missing: the finite relaxation
of the spins even in the limit T ----7 0 leading to inelastic broadening of the
local spin levels, which we identify with the transverse spin relaxation rate
r = 1/T2 . Within the PF representation this effect shows up as a finite
imaginary part of the PF self-energy.
Assuming the RG flow to be stopped at the scale r, we replace the step
functions 8 w in Eq. (8) by 8(D - vw
2 + r 2 ). The decay rate r has to

be calculated self-consistently with the solution of the flow equations (8).

Starting from the golden rule expression for the transverse relaxation rate,

r :12 L jdw[91(W)2fw-lLa(1-fw-lLa')
a,a'=L,R,,=i,l
+92(W - I B /2)2 fW-/La (1 - fW-/La,-,B)] (9)
the renormalized r is obtained by replacing 91,2 by 91,2(W) as determined
from the solution of (2).
We are now ready to calculate further physical quantities. The renor-
malized value of the magnetization is obtained by substituting 91,2 (w) in
 245

a - RG c 0.15
._. o(i)
0.1 -- o(i) 0.1
o
(9
. I ·~·~_--.j~~r
(5 --------1 i------ __
L-'-'-l
a h-h-+-r-+---,--r---r--h----+--.-+---r-r----r-+-,H-
1
b
0.8
~
0.6

Figure 2. a) Conductance G(V/B) calculated in leading (dashed line) and next-to-

-leading (dot-dashed line) order PT compared to the result of perturbative RG (solid line)
for B = 100TK, D = 104 TK, TK = Dy/rje- 1 / 29 ). b) Local magnetization M(V/B,B/TK)
of a symmetric dot for fixed magnetic field B = 20,500 TK. c) Experiments by Ralph
and Buhrman (22) (symbols) on transport through metallic point contacts in magnetic
fields 0.85,1.7,2.55 T (B = 36, 72, 104 T K where T K "" 30mK (22)).

place of 91,2 in the Golden Rule expression (3). In Fig. 2b we show the fully
renormalized result for M as a function of V / B for different values of B.
The charge current I is calculated from Eq. (2) inserting the renor-
malized coupling functions. Fig. 2a shows a comparison of the differential
conductance G(V) = dI/dV obtained in this way, with the bare result (2)
and the PT result including leading logarithmic corrections.
The peak structures in G(V, B) appearing at V = ±B have been detec-
ted in experiment. In order to reach large values of B /TK , it is necessary to
have relatively low TK. This happened to be the case in transport through
metallic point contacts, containing a magnetic impurity (22). In Fig. 2c the
result of our theory using the value of TK ':::::' 30mK given in (22) is compared
to the experimental data, after subtracting a background contribution (see
(17) for details). The agreement is seen to be excellent.

4. Conclusion

We considered the transport of electrons through a Kondo dot in the regime

of large transport voltage V and in the presence of a magnetic field B,
such that max(V, B) » T K is satisfied. A finite difference of the chemical
potentials in the two leads, /JL -/JR = V, changes the RG flow substantially.
246

Furthermore, it opens an energy window for inelastic processes even at T =

0, leading to a broadening r of the spin sublevels of the dot and providing
a cut-off for the scaling towards the Kondo fixed point. A finite magnetic
field induces a local magnetization M at the dot, which for V » TK, T is
independent of temperature, being solely controlled by the voltage V. We
derived a set of RG equations within a poor man's scaling approximation,
which for once reproduce the bare PT result, but then may be integrated
to give the fully renormalized result. The integration is done including the
self-consistently determined cut-off r. The results obtained in this way are
valid up to corrections of order l/£n[(V, B)/TK].

5. Acknowledgments

We would like to thank S. De Franceschi, J. Konig, O. Parcollet, H. Schoeller

and A. Shnirman for helpful discussions and especially L. Glazman, who
suggested investigating the case of finite B. Part of this work was suppor-
ted by the Center for Functional Nanostructures and the Emmy-Noether
program (A.R.) of the DFG.

References

1. r. Aleiner, P. Brouwer, and L. Glazman, Phys. Rep. 358, 309 (2002).

2. L. Glazman and M. Raikh, JETP Letters 47, 452 (1988).
3. T. Ng and P.A. Lee, Phys. Rev. Lett. 61, 1768 (1988).
4. D. Goldhaber-Gordon, H. Shtrikman, D. Mahalu, D. Abusch-Magder, U. Meirav
and M. Kastner, Nature 391, 156 (1998).
5. S. Cronenwett, T. Oosterkamp and L. Kouwenhoven, Science 281, 540 (1998).
6. J. Schmid, J. Weis, K Eberl, and K v. Klitzing, Physica B 258, 182 (1998).
7. J. Nygard, D. Cobden and P. Lindelof, Nature 408,342 (2000).
8. S. De Franceschi et al., cond-mat/0203146.
9. P. W. Anderson, J. Phys. C 3, 2436 (1966).
10. KG. Wilson, Rev. Mod. Phys. 47, 773 (1975).
11. N. Andrei, Phys. Rev. Lett. 45, 379 (1980); N. Andrei, K Furuya and J.H.
Lowenstein, Rev. Mod. Phys. 55, 331 (1983).
12. r. Affleck and A. W. W. Ludwig, Nue!. Phys. B 360, 641 (1991).
13. J. Kroha and P. Wolfle, in press (Springer), cond-mat/0105491.
14. A. Rosch, J. Kroha and P. Wolfle, Phys. Rev. Lett. 87, 156802 (2001).
15. J. Appelbaum, Phys. Rev. Lett, 17, 91 (1966); Phys. Rev. 154,633 (1967).
16. J. Konig, J. Schmid, H. Schoeller, and G. Schon, Phys. Rev. B 54, 16820 (1996).
17. A.Rosch, J. Paaske, J. Kroha and P. Wolfle, cond-mat/0202404.
18. L.r. Glazman, private communication.
19. A.A. Abrikosov, Physics 2, 21 (1965).
20. O. Parcollet, and C. Hooley, cond-mat/0202425.
21. H. Schoeller and J. Konig, Phys. Rev. Lett. 84, 3686 (2000); see also M. Keil, Ph.D.
thesis, U. Gottingen (2002).
22. D.C. Ralph and R.A. Buhrman, Phys. Rev. Lett. 72, 3401 (1994).
QUANTUM FLUCTUATIONS AND ELECTRONIC TRANSPORT
THROUGH STRONGLY INTERACTING QUANTUM DOTS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

T. A. COSTI
Universitat Karlsruhe, Institut fur Theorie der Kondensierten
Materie 16128 Karlsruhe, Germany

Abstract. We study electronic transport through a strongly interacting quantum dot

by using the finite temperature extension of Wilson's numerical renormalization group
(NRG) method. This allows the linear conductance to be calculated at all temperatures
and in particular at very low temperature where quantum fluctuations and the Kondo
effect strongly modify the transport. The quantum dot investigated has one active level
for transport and is modeled by an Anderson impurity model attached to left and right
electron reservoirs. The predictions for the linear conductance are compared to available
experimental data for quantum dots in heterostructures. The spin-resolved conductance
is calculated as a function of gate voltage, temperature and magnetic field strength and
the spin-filtering properties of quantum dots in a magnetic field are described.

Key words: Quantum Dot, Numerical Renormalization Group, Anderson (Impurity)

Model, Kondo Effect, Linear Conductance, Spin-Filtering

1. Introduction

Recent experimental work (1, 2, 3, 4, 5) has demonstrated the importance

of correlations in determining the low temperature transport properties of
nanoscale size quantum dots. These dots consists of a confined region of
electrons, of typical diameter 100nm, "weakly" coupled to leads via tunnel
barriers. A "weak" coupling, r, means that the quantized levels of the dot
are broadened into resonances, but are not completely washed out. A gate
voltage, Vg , controls the position of the quantized levels relative to the
chemical potentials of the leads and thereby the total number of electrons
on the dot. The charging energy U for adding electrons to the dot from the
surrounding electron reservoirs (leads) is typically the largest energy scale,
and the dot is strongly correlated provided U jr » 1. Typical values for U

247
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 247–256.
© 2003 Kluwer Academic Publishers
248

and r for the dots in (1, 5) are U rv 0.5 - 2.0 meV and r rv 0.2 - 0.3 meV,
so these dots are strongly correlated.
For temperatures r « T « U quantum fluctuations are small, and
transport is dominated by charging effects. This regime is well understood
(6). The conductance, G, exhibits a series of approximately equidistant
peaks as a function of Vg, with spacing U. The peaks correspond to a
fractional number of electrons on the dot and alternate "Coulomb blockade"
valleys to either an even or an odd number of electrons.
In this paper we shall be interested in the regime T ;S r « U, where
the strong quantum fluctuations can lead to a dramatic modification of the
above picture of Coulomb blockade. In particular, for an odd number of
electrons on the dot, the quantum dot can have a net spin 1/2, and a Kondo
effect can develop. It has been predicted (7, 8), that, at low temperature,
this enhances the conductance in the odd electron valleys, turning them
instead into plateaus of near perfect transmission.
The outline of the paper is as follows. Sec. 2 describes the model and
Sec. 3 the NRC method used to solve it. The results (9, 10, 11) for the
conductance are described in Sec. 4-5 and these are used to compare with
experiment over a range of gate voltages and temperatures in Sec. 6. Ex-
perimental investigation of the regime T ;S r has only recently become
possible as a result of better control of the electrode geometry, allowing
parameters like r to be tuned to values of 1-3K so that T ;S r is accessible
(1, 2, 3, 4, 5). Sec. 7 describes the effect of a magnetic field on the transport
through a quantum dot. Influencing transport via spin effects is of great
current interest (12). The results presented in Sec. 7 should prove use-
ful for interpreting magnetotransport experiments on strongly interacting
quantum dots.

2. Model

At sufficiently low temperature, transport through a strongly interacting

quantum dot will be mainly determined by the partially occupied level of
the dot, denoted Ed, which lies closest to the chemical potential of the leads.
Its occupancy, nd, can be varied from nd = 0 to nd = 2 by varying the gate
voltage V g, where -eVg = Ed. The resulting model of a single correlated
level Ed with Coulomb repulsion U, coupled to left and right free electron
reservoirs can be written as

+ L Ek,ic1,i, CT c k,i,CT + L 1!i(c1, ip d CT + d~Ck,.ip). (1)

k,CT,i=L,R kp,i=L,R
 249

We assume energy independent lead couplings rL,R = 21fPL,R(EF)Vf,R'

where PL,R( EF) is the Fermi level density of states (per spin) of the left/right
electron reservoir. The first two terms in H represent the quantum dot,
the third term is a magnetic field coupling only to the dot's spin S~ =
~(4dT - dl dl) (we set 9 = ME = 1), the fourth term represents the free
electron reservoirs and the last term is the coupling between the dot and the
reservoirs. This model can be reduced to the standard Anderson impurity
model of a single reservoir attached to the dot with strength r = r L + r R
(7). Note that r = 2~, where ~ is the hybridization strength as usually
defined in the Anderson model (13). We use r throughout. In experiments
(1, 5), r is extracted by analyzing the high temperature (T » r) behaviour
of the conductance peaks (see figure: 2a below).
We assume, from here on, symmetric coupling to the leads, r L = r R.
The linear magnetoconductance, G(T, H) = ~a Ga(T, H), is written as a
sum of spin-resolved magnetoconductances, G a , where

(2)

Aa(w, T, H) is the equilibrium spectral density and is expressed in terms of

the local level Green function, gd,a = 1/(w - cd + ir /2 - I;u), with I;u the
correlation part of the self-energy, by
1
Aa(w,T,H) = --Imgda(w+iE,T,H), (3)
1f '

3. Method

We calculate Aa(w, T, H) by using Wilson's NRG method (14) extended

to finite temperature dynamics (9, 15), with recent refinements (16, 17)
which improve the high energy features. The NRG procedure for finite
temperature dynamics is described in (9, 15). Here, we make a few remarks
concerning the above refinements.
The first refinement (16) uses the correlation part of the self-energy, I;u,
to evaluate the spectral density A in the Anderson impurity model (16).
This improves the spectra around the single-particle excitations cd and
cd + U since the single particle broadening r /2 is put into the Green func-
tion, gd, exactly, thereby making the excitations at Cd and Cd + U slightly
sharper than in the earlier procedure (9) which evaluated A directly from its
spectral representation. The description of the low energy Kondo resonance
is equally accurate within both approaches and close to exact (16).
The second refinement uses the reduced density matrix in the NRG
procedure for dynamical quantities (17) in place of the grand canonical
250

0.2 0.2
a b

~'.:E:J
0.2
-Pc
----- P, 0.15
-Pc
~ 0.15 0.15 -----
0 P,
II
::r: 0.1
TK £
0"
~- 0.7
0" II ' 0.5
II 0.1 f-< 0.1 0 4 8
f-<
i
0.05 i H

<
<c-

< 0.05 0
-10 10
0.05

0 0
-200 0 200 400 -10 -5 0 5 10 15 20
m/TK m/TK

Figure 1. Comparison of methods for AT (W, 0, H), of the Kondo model, using canonical
(Pc, solid line) (9) and reduced density matrices (Pr, dashed line) (17). TK = 3.97x 10- 5
(in units of half bandwidth D = 1) is the HWHM of the T = H = Kondo resonance: °
(a) at zero field, with the inset showing the Kondo resonance in more detail, and, (b) in
finite field, with the curves from left to right corresponding to H = 0,1,2,5,10 in units
of T K . The inset shows the peak position, Eo (H), as a function of H for the canonical
(circles) and reduced density matrix (squares) approaches. Both Eo and H are measured
in units of T K . The limit (18) limH~o Eo(H)/H = 2/3 is recovered in both approaches.

density matrix of the usual procedure (9). This reduces finite-size effects
in the spectra. The latter are usually small on all energy scales in the case
of zero magnetic field, but can be large in the case of a finite magnetic
field, particularly for the high energy parts of the spectra (since these are
calculated from the shortest chains and are therefore subject to the largest
finite-size effects). Figure: 1a shows that both procedures give equally ac-
curate results for the Kondo model in zero magnetic field on all energy
scales. The same holds for the Anderson model in zero magnetic field. At
finite magnetic field, the reduced density matrix is required for a correct
description of the features at cd and cd+U in the spin-resolved spectra (17).
In contrast, the low energy Kondo resonance in a magnetic field is well
described by either procedure for H;S 10TK, as shown in figure: 1b for the
Kondo model.
The position, Eo(H), of the spin-resolved Kondo resonance in a weak
magnetic field, H « TK, is known from an exact Fermi liquid result due
to Logan (18) (see also (19) showing the low field quasiparticle resonance).
This states that limH-to Eo(H)/H = 2/3. This indicates that correlations
reduce Eo(H)/ H from its expected high field limit of 1 to 2/3. The inset
to figure: 1b shows that both procedures recover this result. The use of the
reduced density matrix has also proven useful for studying magnetic states
in the Hubbard model within the dynamical mean field theory (20).
 251

1.5
,,",0.8
o
Me.
Q,l
is
>, 0.6
C ------ Fit
G" c;I 0.4
0.5
0.2

0.0 0.5 1.0 1.5 ~o 3 10

2
10 ' 10'
V/U T/T K

Figure 2. (a) G(T, H = 0) versus 11;1 in the regime, T -s: r, of strong quantum fluctu-
ations, for U /r = 4.712. The symbols are the T = 0 limit, Eq. (4). Temperatures decrease
from bottom to top and correspond to TN = 0.64rA -N, N = 0,1,2, ... ,A = 1.5. The
dashed curve has T = 1.2TK, where TK ~ 0.05r, is the HWHM of the T = 0 Kondo
resonance at mid-valley gate voltage. The regions marked 1,2 and 3 and separated
by vertical dashed lines correspond to the Kondo (Ed :s -0.5r, i.e. nd ~ 1), mixed
valent (IEdl :s 0.5r, i.e. nd ~ 0.5, or by particle-hole symmetry, lEd + UI :s 0.5r, i.e.
nd ~ 1.5), and empty (full) orbital regimes (Ed ~ 0.5r, i.e. nd ~ 0, or, by particle-hole
symmetry, Ed + U :s -0.5r, i.e. nd ~ 2). (b) The universal conductance curve G(T)/G(O)
for a quantum dot in the Kondo regime at mid-valley, Vg/U = 0.5, (circles) (9, 10).
The dashed line is the fit formula G(T)/G(O) = (T'l/(T 2 + T'l))' with s = 0.22 and
T~ = T K /V2 1 / 8 -1 used in (1) to interpolate the NRC results up to lOTK (with T K as
in figure: 2a).

4. Linear conductance for H = 0

The gate voltage dependence of the linear conductance of a quantum dot

is shown in figure: 2a for a decreasing set of temperatures. There are three
distinct physical regimes, corresponding to three ranges of the gate voltage
controlling cd relative to the Fermi level (see also figure: 3a-c). In the empty
(full) orbital regime (region 3) the conductance shows the expected activ-
ated behaviour as a function of temperature. In the mixed valent regime
(region 2), the behaviour of G versus T is approximately that corresponding
to tunneling through a resonant level close to the Fermi level. The most
dramatic behaviour is in the Kondo regime (region 1), where one sees an
anomalous enhancement of the conductance with decreasing temperature.
The conductance continues to increase and eventually reaches the unitarity
limit of 2e 2 /h (see below) at mid-valley (Vg/U = 0.5 or nd = 1). This
remarkable enhancement of the conductance results from the formation,
with decreasing temperature, of the many-body Kondo resonance at the
Fermi level of the leads. The three types of behaviour have been observed
in experiments on quantum dots carried out at T ;S r (1, 5, 21). The
252

unitarity limit of the conductance, 2e 2 / h, has also recently been observed

(5). The T -----+ 0 conductance curve depends only on nd via the Friedel sum
rule (7, 8, 22)

(4)

and is shown in figure: 2a (circles), with nd deduced from the spectral

densities as a check on the calculations.

5. Scaling of the linear conductance at H =0

In the Kondo regime, G(T)/G(O), is a universal function of T/TK starting

from low temperatures and extending up to some high temperature which
depends on other details such as the precise value of r and Ed. The latter
energy scales cut off the universal behaviour of the conductance on the
high temperature side. The universal conductance curves for the Kondo
and Anderson models have been calculated for the experimentally inter-
esting temperature range 0 :S T :S 10TK via the NRG (9, 10, 11). This
is shown for the Kondo model in figure: 2b together with an approximate
interpolating formula used in (1). The scaling of G(T)/G(O) with T /TK
shown in figure: 2b persists in the Anderson model throughout the Kondo
regime (region 1) as long as T « r. At higher temperatures and in the
other regimes, the conductance curves deviate from this universal shape
(see Sec. 6 below). The Kondo scaling of the conductance in figure: 2b
agrees well with measurements for both quantum dots in heterostructures
(1, 5) and and carbon nanotubes (21).

6. Comparison with experiment

For a comparison of theory with experiment in all regimes, the complete set
of conductance curves for the Anderson model is required. These are shown
in figure: 3a-c. G(T) in the empty orbital and mixed valent regimes has been
calculated in (23) and used to compare with experimental data in (1). Here
we make a parameter free comparison to similar data from Reference (5)
for all three regimes of interest. The results are shown in figure: 3d. In
making the comparisons, we estimated G(T = 0) from G(T = 30mK),
close to the lowest effective electron temperature T = 40mK (5), and used
it to determine nd from Eq:4 and hence the appropriate Ed to use in the
NRG calculation for G(T) at all T. The calculations also used the values
of r = 231.4JLeV and U ~ 0.5meV from the experiments. It is remarkable
that this zero parameter comparison yields the agreement seen. The Kondo
scale automatically comes out correctly as seen for the Vg = -414mV
 253

1.4
- Ed~-O.51 - Ed~1.01
Ed~-O.25r Ed~]·51
1.2 2 d
--- Ed~O.O --- Ed~2.01

1.0 -- Ed~O.251

..
Vg = -420mV
N";g -- Ed~O.51 ~ 1.5 77
0.8 N
rE
~ ~
P 0.6
E 1 Vg = -422mV
C5 0.4 '-'
a 0.5
0.2

0.0 0
10-'10-1 10" 10- 1 10" 10-1 10 100 1000
Ttr T/r Ttr T(mK)

Figure 3. (a-c) G(T) for gate voltages Vg = -Eede in, (a), the Kondo (K) regime, (b),
the mixed valent (MV) regime (Ed = -0.5r, .. , +0.5r), and, (c), the empty orbital (EO)
regime (Ed = +r, .. , +2.0r), and parameters as in figure: 2a. (d) Comparison of theoretical
(lines) to experimental (symbols) conductance curves, G(T), for the quantum dot in (5)
with U = 0.5meV, r = 0.231meV and 13mK :s; T :s; 900mK. The values used for Ed in the
calculations were Ed = -0.85r (Vg = -414mV, K), Ed = -0.25r (Vg = -420mV, MV),
Ed = +0.375r (Vg = -422mV, MV), Ed = +1.00r (Vg = -424mV, EO), Ed = +1.50r
(Vg = -426mV, EO). We find a linear dependence Vg/e = -QEd for Vg :s; -420mV,
which is a consistency check for determining Ed from G(T -+ 0) using Eq. 4.

comparison, and the agreement with the theoretical conductance curve is

very good up to 700mK. The conductance curve used here includes the
non-universal corrections discussed above and therefore differs slightly from
that used in (5) (notably a slope change at 500mK due to corrections
from charge fluctuations). The general trends of the experimental data in
going to the mixed valent (Vg = -420mV, Vg = -422mV) and empty
orbital cases (Vq = -424mV, Vg = -426mV) are well reproduced by the
calculations. In particular the expected finite T peak in G(T) (23) develops
and becomes increasingly more pronounced on entering the empty orbital
regime. As described earlier, transport in this regime is likely to involve
additional neighbouring levels and a larger conductance at higher temper-
atures, as observed in the experiments. In this light, the main discrepancy
remaining between theory and experiment appears to be the dip in G(T)
at 200-300mK in the measurements. No signature of this is present in our
model calculations. This could be due to interference effects associated with
more than one level. It would therefore be interesting to investigate this
possibility further.

7. Effect of a magnetic field

A magnetic field suppresses the Kondo effect on a scale of T K (13) and

consequently it is expected to have a drastic effect on the conductance of
254

a
b
1.5
0.8
::c: 53'
N-
~
t t:: 0.6
I.-'
r5 0.4
0.5
0.2

o 0.5 o 0.5 1.5

V/U V/U
Figure 4. (a) T and Vg dependence of G(T, H) for H = T K and for temperatures
and parameters as in figure:2a. (b) T and Vg dependence of the spin-resolved conduct-
ance G r (T, H) for H = 5TK and for temperatures and parameters as in figure:2a. The
dot-dashed curve is for the down spin conductance, G 1 (T, H), at the lowest temperature.

a quantum dot (10, 11). Indeed, figure: 4a shows that a field of H = TK

suffices to reduce the T = 0 conductance at mid-valley to nearly half its
value at H = 0 (cf. figure: 2a). This can be understood from the splitting
of the Kondo resonance in a magnetic field and its strong reduction at the
Fermi level (see figure: 1b and (10)). The suppression of the conductance
is large in the whole Kondo regime (region 1) as long as T ;S T K . At higher
temperatures and in the other regimes, the effect of a field on the total
conductance is less pronounced. However, in these regimes too, there can
be a large effect in the spin-resolved conductances (see below).
The effects of a magnetic field become even more apparent in spin-
resolved quantities such as Gi in figure: 4b (by particle-hole symmetry G 1
is the mirror reflection of Gi about Vg/U = 0.5). As in G, a magnetic field
has a large effect on the spin-resolved conductance in the Kondo regime. In
addition, there is quite a dramatic spin-filtering effect in the mixed valent
regime, with IGi (0, H) - G 1(0, H) I being largest in this regime. In contrast,
both up and down spin conductances are approximately equally suppressed
in the other regimes. A quantum dot in a field is seen to act as a spin-filter
as discussed in (24) for dots very weakly coupled to leads (G a «e 2 /h).
The field dependence of the conductance of quantum dots defined in car-
bon nanotubes has been studied (21) and a preliminary comparison between
our results and the experimental G(B, T ;S T K ) shows good agreement (25).
 255

8. Conclusions

We considered electronic transport through a strongly interacting quantum

dot in zero and finite magnetic fields and in the low temperature regime
T ;s r where quantum fluctuations strongly modify the transport for an
odd number of electrons on the dot.
The NRG conductance curves, G(T, Vg ), were used to make a para-
meter free comparison with experimental results of Reference (5) and we
found good agreement. Discrepancies in the empty orbital regime where
attributed to the neglect, within our model, of neighbouring levels
The results for the magnetoconductance in the Kondo regime exhibited
the strong suppression of the Kondo effect by a magnetic field. A large
spin-filtering effect was found in the mixed valent regime.
The field of non-equilibrium transport through quantum dots remains
largely unexplored. Perturbative methods valid in the limit of large trans-
port voltage, and magnetic field, V, H » TK, are being developed (26), but
non-perturbative techniques, such as extensions of NRG, will be required
to address strong coupling.

Acknowledgements

I would like to thank the authors of Reference (5) for sending me the
experimental data used in figure: 3b and W. van der Wiel for useful discus-
sions. Financial support from the Deutscheforschungsgemeinschaft through
SFB 195, and in part by the National Science Foundation under Grant No.
PHY99-07949 during the writing of this paper at the KITP of UCSB, is
acknowledged.

References

1. D. Goldhaber-Gordon et al., Phys. Rev. Lett. 81, 5225 (1998).

2. S. M. Cronenwett, T. H. Oosterkamp and L. Kouwenhoven, Science 281, 540 (1998).
3. J. Schmid, J. Weis, K. Eberl and K. von Klitzing, Physica B 258, 182 (1998).
4. F. Simmel, R. H. Blick, J. P. Kotthaus, W. Wegscheider and M. Bichler, Phys. Rev.
Lett. 83, 804 (1999).
5. W. G. van der Wiel et al., Science 289, 2105 (2000).
6. C. W. J. Beenakker and H. van Houten, Solid State Physics, 44, 1 (1991).
7. L. I. Glazman and M. E. Raikh, Sov. Phys. JETP Lett. 47, 454 (1988).
8. T. K. Ng and P. A. Lee, Phys. Rev. Lett. 61, 1768 (1988).
9. T. A. Costi, A. C. Hewson and V. Zlatic, J. Phys. Condo Matt. 6, 2519 (1994).
10. T. A. Costi, Phys. Rev. Lett. 85, 1504 (2000).
11. T. A. Costi, Phys. Rev. B 64, 241310(R) (2001).
12. Y. Ohno et al., Nature 402,790 (1999).
256

13. A. C. Hewson, The Kondo Problem to Heavy Fermions (Cambridge University Press,
Cambridge, England 1993)
14. K. G. Wilson, Rev. Mod. Phys.47, 773 (1975); H. R. Krishna-murthy, J. W. Wilkins
and K. G. Wilson, Phys. Rev. B21, 100:3 (1980).
15. T. A. Costi, in Density Matrix Renormalization, edited by 1. Peschel, X. Wang, M.
Kaulke and K. Hallberg (Springer, Berlin, Germany 1999).
16. R. Bulla, A. C. Hewson and Th. Pruschke, J. Phys. Condo Matt. 10, 8365 (1998).
17. W. Hofstetter, Phys. Rev. Lett. 85, 1508 (2000).
18. D. E. Logan and N. L. Dickens, J. Phys.: Condens. Matter 13, 9713 (2001).
19. A. C. Hewson, J. Phys.: Condens. Matter, 13, 10011 (2001).
20. R. Zitzler, Th. Pruschke and R. Bulla, European Phys. Journal, B27, 47:3 (2002).
21. J. Nygard, D. H. Cobden and P. E. Lindelof, Nature 408, 342 (2000).
22. D. C. Langreth, Phys. Rev. 150, 516 (1966).
23. J. Konig and H. Schoeller, Phys. Rev. Lett. 84, 3686 (2000).
24. P. Recher, E. V. Sukhorukov and D. Loss, Phys. Rev. Lett. 85, 1962 (2000).
25. J. Nygard, private communication.
26. A. Rosch, J. Paaske, J. Kroha, P. Wolfle, cond-mat/0202404.
PROPERTIES OF CORRELATED NANOSCOPIC SYSTEMS
FROM THE COMBINED EXACT DIAGONALIZATION - AB
INITIO METHOD

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

J. SPALEK, E.M. eORLICR, A. RYCERZ, R. ZAHORBENSKI

and R. PODSIADLY
Marian Smoluchowski Institute of Physics, lagiellonian Uni-
versity, ulica Reymonta 4, 30-059 Krakow, Poland
W. WOJCIK
Institute of Physics, Tadeusz Kosciuszko Technical University,
ulica Podchorl}zych 1, 30-084 Krakow, Poland

Abstract. We review briefly our recent exact results concerning the electronic properties
of atomic, molecular, and nanoscopic systems by combining the exact diagonalization and
ab initio methods (EDABI). Particular emphasis is put on determining the microscopic
parameters. The multiatom system behavior is discussed as a function of interatomic
distance. A transformation from a nanometal (for small lattice constant) to an atomic
system is observed with increasing distance between the atoms.

Key words: Nanoscopic Systems, Correlated Electrons, Exact Diagonalization, Nano-

chains, Nanoclusters, Ab Initio Approach

1. Introduction

Theoretical determination of the electronic and atomic properties of nano-

scopic systems containing N rv 10 - 100 atoms is very important from the
two principal perspectives. First, with the advancement of nanotechnology
we begin utilizing the electronics on the atomic scale. Second, it is crucial to
determine in a precise manner electronic properties of such small systems
to see the evolution from a collection of atoms to a piece of a solid with
the delocalized (Bloch-type or Fermi-liquid-like) states. The latter aspect is
reviewed briefly here after introducing a novel method of approach (EDABI)
developed in our group recently (1, 2, 3), in which we combine the exact

257
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 257–268.
© 2003 Kluwer Academic Publishers
258

diagonalization providing us with rigorous treatment of interparticle inter-

action, with an ab-initio single-particle-wave-function optimization in the
system ground state. In this manner, the interaction and the single-particle
terms in the Hamiltonian (or Lagrangian) are treated on the same footing.
Hence, the method is particularly useful for the strongly correlated sys-
tems when the kinetic (band) energy of electrons propagating throughout
the system is at best comparable to the interaction among the particles.
One can also say that the approach combines the particle and the wave
aspects of the electron-state characterization into a single formalism, since
the interaction is treated in the occupation number representation, whereas
the shape of matter wave for a single particle is found from the renormalized
(self-adjusted) wave equation or from a related to it variational procedure.
In Sec.2 we summarize briefly our method of approach and its relation to
multiconfiguration-interaction (Mel) approach from quantum chemistry.
In Secs.3-5 we provide the examples of applications ranging from atomic
physics to correlated chain and cluster systems. Sec.6 contains conclusions.

2. The Method

2.1. RENORMALIZED (SELF-ADJUSTED) WAVE EQUATION, SWE

We start with the Hamiltonian in the standard second-quantized form

H = J d3 x\ift (x)H1(x) \if (x)

+~ JJ d3 xd3 x'\ift (x)\ift (x')V(x - x')\if(x')\if(x), (1)

where \if(x) = LiO- Wi (X)XCJaiCJ is the field operator, H1(x) and V ==

V(x-x') are respectively the single-particle and the pair-interaction terms,
and {Wi(X)} is a complete single-particle basis (arbitrary at this point). A
complete solution for N indistinguishable particles would involve determin-
ation of \if(x) via the Heisenberg equation of motion. Such a solution does
not seem to be executable for N-fermion system, so we proceed as follows.
We substitute the expression for \if(x) int~ (1) to obtain

(2)

with the microscopic parameters defined by

(3)
 259

and

Vijkl ==< wiwjlVlwkWI >= J d 3xd 3x'w7(x)wj(x')V(x - X')Wk(X)WI(X').

(4)
In this form, the single and many-particle aspects of the problem are separ-
ated in the sense that the calculation of the parameters tij and Vijkl, both
containing the single-particle basis {Wi (X)}, is separated from the diagonal-
ization procedure of the Hamiltonian in the Fock space (occupation-number
representation). This two-step procedure can be seen explicitly when we
calculate the system ground state energy

_ H >-
Ee =< _~' " tij < aicyajCY
t > +-1 '~
" t ajCY2alCY2akCYl
Vijkl < a iCY1 t >,
ijCY 2 ij klCY I CY 2
(5)
where the averaging < ... > is for the ground IWe > or excited states;
we also take into account all accessible occupancies of the states {Iicy >},
{Ijcy' >}, etc. for fixed values of tij and Vijkl regarded as parameters (if
we consider grand canonical ensemble, we diagonalize H - fLN, with N =
Jd3 xW~t (x)w(x)
~
= ~ijCY Jd3 XW i* (X)Wj(X) < aicyajCY
t
> ).
To close the solution we treat the expression (5) as a functional {Wi (X)}.
In the most general case of the grand canonical ensemble and with a
nonorthogonal basis {Wi (X)} this functional can be written as

F{Wi(X)} = Ee{Wi(X)}- ?=(fL+Aij )

ZJCY
J d3xw7(x)Wj(X) < aIcyajCY >. (6)

In effect, the Lagrange-Euler equation, which plays the role of the stationary
self-adjusted wave equation (SWE) for Wi(X), takes the form

(7)

where Aij and fL play the role of Lagrange multipliers (if we use explicitly
orthogonal basis and particle-conserving diagonalization procedure in the
Fock space, then Aij = fL == 0 and Eq.(7) reduces to the usual Euler
equation). Also, when the class of {Wi(X)} is selected variationally (as
in the following), Eq.(6) is then minimized with respect to trial function
parameters (orbital radii ai in the case when the orthogonal basis Wi (x) is
composed of superposed atomic-like wave functions).
260

2.2. RELATION TO THE MULTICONFIGURATION-INTERACTION (MCI)

METHOD

With the help of the field operator \IJ(x) we can define N-particle wave
function Wa (X1' .. ,XN) for particles distributed among M states as follows

(8)

where l<I>a > is the N-particle state in the Fock space

(9)

where Cj1 ... jN are the expansion coefficients (note that: ]1 == ]10'1). Within
our method we determine C j1 ... jN with the help of Lanczos- or other exact-
diagonalization method (4). Substitution (9) to (8) and using the decom-
position of \IJ(Xi) in terms of {Wi (X)} we obtain that

where (A, S) are respectively (-l)P, where P is the permutation sign of

the indices (iI, i2, ... , iN) for fermions and a symmetrization operation in
case of bosons. Therefore if we are able to perform the diagonalization in
the Fock space, we have an exact many-particle wave function within the
subspace selected by the single-particle basis {Wi (X) }. In the next Sections
we discuss properties of different atomic and nanoscopic systems.

3. Application to atomic and molecular systems

We construct a nanoscopic system out of atoms. Therefore, as a first step

we apply our method first to simple atomic and molecular systems.

3.1. LIGHT ATOMS AND IONS

He is the lightest many-electron atom. In the simplest approximation the

Hilbert space is defined only by a single orbital <I>(r). Under these conditions
\IJ(x) = <I>(r)Xrar + <I> (r)x1 al' The ground state energy is Ea = ta + U,
with

J
2 2
3 * 1t 2 2e
ta = d r<I> (r)[--\7 - -]<I>(r) (11)
2m r
 261

TABLE I. Optimized Bohr-orbit radii of Is, 2s, and 2p

orbits (in units of ao), the overlap S between renormal-
ized Is and 2s states, and the ground state energy for the
lightest atoms and ions.

als a2s a2p S Ec (Ry)

H 1 2 2 0 -1

H- 0.9696 1.6485 1.017 -0.1 -1.0487

He 0.4274 0.5731 0.4068 -0.272 -5.79404

He- 1.831 1.1416 0.4354 -0.781 -5.10058

Li 0.3725 1.066 0.2521 0.15 -14.8334

Be+ 0.2708 0.683 0.1829 0.109 -28.5286

being the atomic energy, and

U = J d3rd3r/I<I>(r)12Ir
2
~ r/I I<I>(r/)1 2 (12)

being the magnitude of the Coulomb interaction between electrons. Con-

sidering Ec as a functional of <I> (r) we obtain Eq. (7) in the form (/f' =
0):

(13)

This means <I>(r) is determined from the Hartree equation. Assuming that
<I>(r) has Is-like form <I>(r) = (7fa 3)1/2 exp( -ar), and adjusting variationally
the size a = a-I of the renormalized Is orbit we obtain the well known
results (5): Ec = -5.695Ry and a = (16/27)ao, where ao is the Bohr
radius. We can systematically improve on this result by enriching our single-
particle basis. Including Is- and 2s-wave functions, we obtain 6-dimensional
Fock space and by repeating both the diagonalization procedure and the
variational optimization of Ec with respect to Is- and 2s-like state radii
we obtain an improved value of Ec = -5.7549Ry. Finally, taking Is-, 2s-,
and 2p-like states (with the corresponding adjustable radii a1s, a2s, a2p) we
obtain the value Ec = -5.794Ry, which is close to the accepted "exact"
value -5.S074Ry. The values of radii are displayed in Table I for the light
atoms and ions. The ground-state 2-particle wave function has in the Fock
space the form for He atom:

IG >':::' {O.Sa tIs ratls-).I - O.4(a tIs rat2s-).I + at2s ratls-).I) + 0.135a 2s
t rat I
2s-).
262

TABLE I!. Microscopic parameters (in Ry) of the selected atoms and ions
all quantities are calculated for the orthogonalized atomic states.

t U1 U2 U3 Up K 12 K 13 K 23
H- 0.057 1.333 0.369 0.77 0.728 0.519 0.878 0.457

He 1.186 3.278 1.086 1.924 1.821 1.527 2.192 1.289

He- -1.1414 1.232 0.764 1.798 1.701 0.929 1.421 1.041

Li -0.654 3.267 0.533 3.105 2.938 0.749 3.021 0.743

Be+ -0.929 4.509 0.869 4.279 4.049 1.191 4.168 1.175

(14)
Thus He atom is indeed the smallest atom in the Universe, and we have
a substantial (0.32) probability of having the two electrons in the singlet
Is-2s configuration. Nota bene, we can determine the first excited states
in this case: the Is-2s triplet with energy E t = - 2.3707 Ry, and the next
singlet of primary Is-2s character with E 2s = -1.3737Ry.
Important are the values of the microscopic parameters, which are
displayed for those systems in Table II. They comprise the Is-2s overlap
integral 5, the Is-2s hopping integral t, the magnitudes of Is (Ud, 2s (U2 ),
2p (U3 for m = 0 and Up for m = ±1), and 2p (Up) Coulomb interaction.
Additionally, K 12 , K 13 , K 23 represent respectively Is-2s, Is-2p, and 2s-2p
Coulomb interactions.

3.2. H 2 AND Hi SYSTEMS

The next step in building up step by step a model of nanoscopic system

is to consider H 2 and Hi systems. Here we consider the situation with
one orbital per atom, which will be taken as Is-like. For this case the Fock
space for H 2 is spanned on (~) = 6 states, whereas that for Hi contains
(~) = 4 basis states. In both cases one can perform the diagonalization
in Fock space analytically, whereas the Is orbit readjustment represents a
simple numerical minimization of EG. In Fig.l we display the interatomic-
distance (R/ao) dependence of the ground-state energy for those systems.
Obviously, within one-orbital-per-atom basis Hi state is unstable. How-
ever, consideration of those systems provides us also with the values of
microscopic parameters such as the hopping integral t, intraatomic (and
interatomic) Coulomb interaction U(K), as well as the magnitudes of the
correlated hopping parameter V, Heisenberg (J) and the kinetic exchange
 263

,
\ , '.... AH,- states
\

\, ---->-'--"---'---------------------
~
L........J -2.0

§ Q)

Jj
H 2 ground state

Distance, R/ao

Figure 1. Energies of the ground state for H 2 molecule (solid line), as well as the two
lowest laying states for H:; ions, all quantities as a function of interatomic distance R/ao
(au - Bohr radius).

TABLE III. Microscopic parameters (in Ry) of H 2 system vs. R/ao.

R/au EG/N t U K V [mRy] J[mRy] 4(t+V)2 [R]

U-K m y

1. -1.0937 -1.1719 1.8582 1.1334 -13.55 26.254 7755.52

1.5 -1.1472 -0.6784 1.6265 0.9331 -11.687 21.252 2747.41
2. -1.1177 -0.4274 1.4747 0.7925 -11.577 16.921 1130.19
2.5 -1.0787 -0.2833 1.3769 0.6887 -12.054 13.149 507.209
3. -1.0469 -0.1932 1.3171 0.6077 -12.594 9.8153 238.939
3.5 -1.0254 -0.1333 1.2835 0.5414 -12.812 6.9224 115.143
4. -1.0127 -0.0919 1.2663 0.4854 -12.441 4.5736 55.8193
4.5 -1.006 -0.0629 1.2579 0.4377 -11.441 2.8367 26.9722
5. -1.0028 -0.0426 1.2539 0.3970 -9.9894 1.6652 12.9352
5.5 -1.0012 -0.0286 1.2519 0.3623 -8.3378 0.9334 6.1455
6. -1.0005 -0.01905 1.251 0.3327 -6.7029 0.5033 2.8902
6.5 -1.00024 -0.0126 1.2505 0.3075 -5.2242 0.2626 1.3452
7. -1.0001 -0.0083 1.2503 0.2856 -3.9685 0.1333 0.6197

amplitudes which are listed in Table III. Those values will be compared
with those for the chains containing up to N = 14 atoms to draw some
conclusions concerning a similar behavior of both atomic and nanoscopic
systems.

4. N anochains

We now consider a nanoscopic linear chain of up to N = 14 atoms, each con-

taining a single valence electron (hydrogenic-like atoms) and including all
overlap integrals, all hopping integrals, as well as all interactions (including
264

\
\

R;j N=8 6 I
I

Figure 2. Schematic representation of the finite chain with periodic boundary conditions
(or planar cluster) used in the calculations.

TABLE IV. Microscopic parameters for a linear-chain configuration and the

Slater orbitals taken.

t2 U Ka Vi V2
I R/ao I
[mRy] [Ry] [Ry] [mRy] [mRy]

2.0 -0.5851 86.98 2.301 1.077 0.676 0.450 -18.07 33.58

2.5 -0.3302 44.08 1.949 0.843 0.499 0.331 -17.45 19.58
3.0 -0.2000 23.60 1.717 0.692 0.391 0.259 -16.08 11.95
4.0 -0.0825 7.56 1.452 0.508 0.269 0.179 -12.92 4.49
5.0 -0.0366 4.29 1.327 0.403 0.206 0.138 -9.64 1.56

3- and 4-site terms in the Gaussian STO-3G basis). The method and the
principal results for the case of Slater orbitals have been published before
[2]. Here we present also the first results obtained for the Gaussian (STO-
3G) basis, which we will try to apply for realistic systems containing Li and
Na atoms in the near future. In Fig.2 we present schematically a ring of
N atoms. We analyze its basic properties in a linear-chain configuration
with periodic boundary conditions (i.e. overlaps, hoppings, interactions
only along the ring), whereas in the next Section we discuss a multidi-
mensional (d = 2,3) configurations. By adopting EDABI to this situation
we can calculate all hopping integrals and therefore, calculate the band
energy Ek, as it evolves as a function of interatomic distance R/ao (2). For
R/ao :2: 4 the tight-binding approximation works very well. The bandwidth
W = 2z Lj(i) Itij I falls with growing R systematically reflecting reduced
overlap. This illustrates the famous Mott remark (6) that by looking at
the single-particle (band) energy one can never observe a transition from
atomic to metallic behavior, only a systematic evolution towards the atomic
 265

TABLE V. Microscopic parameters (in Ry) vs. R/ao for N = 4 planar cluster.
The star marks the distance R with the lowest system energy.

R/ao tl t2 [mRy] U KI K2 VI [mRy] J I [mRy]

0,5 -3,8651 1510 2,7792 1,6847 1,2008 -37,30 49,04

1,0 -1,4691 304,4 2,1831 1,2618 0,9237 -12,54 32,54

1,5 -0,7657 87,17 1,8310 1,0020 0,7354 -8,151 24,03

1,8 -0,5563 41,40 1,6877 0,8918 0,6532 -7,884 20,28

1,9* -0,5043 31,74 1,6482 0,8604 0,6296 -7,960 19,17

2,0 -0,4587 23,91 1,6128 0,8314 0,6077 -8,098 18,12

2,5 -0,2959 1,930 1,4701 0,7101 0,5156 -9,134 13,57

3,0 -0,1991 -5,710 1,3767 0,6187 0,4461 -10,29 9,910

3,5 -0,1368 -7,640 1,3186 0,5467 0,3917 -11,12 6,930

4,0 -0,0945 -7,140 1,2845 0,4879 0,3477 -11,31 4,584

N = 6+
N = 8- N = 6-- 1
1. N = 1()D R/ct =~2 5
N = 12 8
N = 14 6

o.

O. O---"=='---~~~~~~----":C= __---l
-1 -0.5 0 0.5
MOMENTUM I kR/n

Figure 3. Momentum distribution nkO" for electrons in linear chain of N = 6-;-12 atoms;
the interatomic distance is specified in units of ao. The solid line represents the parabolic
fit, which is of the same for both k < k F and k > k F . The discontinuity at kFR/Tr = 0.5
is also marked (STO-3G basis is taken).

states with the increasing R. In Fig.3 we display the statistical distribution

function n(k) =< aLO"akO" > obtained by using the STO-3G basis to repres-
ent the starting atomic wave function, out of which we build the Wannier
functions {wi(r)h=l, ... ,N. These functions are optimized to minimize the
ground state energy, as before. We see that for the interatomic distance
R c:::: 3.8ao the Fermi discontinuity tlnF at the Fermi level vanishes; this
is regarded as a quantitative criterion for the electron localization. Further
266

:>
~
-12.0

-12,5

.
-130
,
~
~

-.
~- N
N~

N-
-

---+-- N =
4, Telrahedron
3, Triangle
4. Square
2, H,

~
z
J' -13,5
,;;
.... _..
11. --------~fiit..··-·...········1II
~
c:
<D
-14,0
. ".:."
~A'
l~
\1 _-:1I.1t.
'"
§ -14,5 i\/It
1

'"
'C \• ;11i"
'\ ~
R •• 2.2 ~-].()jtE"N--]4.40<V I
§ -15.0 I
e , .,'1 ~;j!J.,=2.l u= 1.l2E.:.N=·15.0l cV
CJ
; :
-15,5 1 ].9 OFI.]4 g,.iN: -15.11 ,v
~~.=

Ria.;.- 1.45 cc.- 1.19 E.,/N- -lj.62.eV

-16.0
023 689
Interatomic distance. Ria,

Figure 4. Total ground state energy (per atom, including the ionic contribution) of
several cluster systems vs. R/ao. The optimal values of the parameters for the minimal
energy are provided in the insets.

features of this Mott localization crossover are discussed elsewhere (7). To

compare the values of the microscopic parameters with those for H 2 and
Hi ions we have shown in Table IV the interatomic distance dependence
of those parameters (for N = 10) when using the Slater orbitals also for the
linear-chain calculations.The conclusion is that the values of the parameters
obtained for the optimized basis are only weakly dependent on the system
SIze.

5. Nanoclusters

We have also applied our method to study various spatial arrangements

of clusters containing up to N ::; 4 hydrogenic-like atoms and a planar
(honeycomb) cluster containing N = 6 such atoms. In Fig.4 we display the
interatomic distance dependence of the ground-state energy (per atom),
together with the values characterizing the minimal point. As before, the
size of the atomic-like orbitals (a -1 ao) is smaller than that for the hydrogen
atom. Hence, the repulsive Coulomb interaction always causes the atomic
orbit shrinking. Also, the planar configuration for N = 4 is the stable
one with respect to tetrahedron configuration. However, it is unstable with
respect to the dissociation into two H 2 molecules. Therefore, these sys-
tems will be stable only when adsorbed on the surface of other materials.
 267

-0,25

III subband
>:
i.
0::
-0,50
W
ul
<lJ
'OJ
Qj
a3 -0,75
.$
.s
<J)
c
<lJ
Ol
Qj I subband
-1,00 1--\------~~==:"'------'--=='_==_1
E n=16
<lJ
1ii
>-
(f)

Interatomic distance, R/a o

Figure 5. Energies of the excited states for the square configuration of H 4 cluster. The
states group into the Hubbard subbands with increasing R.

Nevertheless, interesting for us is the nature of the correlated states in

given configuration. The microscopic parameters for this case are listed in
Table V. They can be compared with those for H 2 systems and the linear
chains discussed earlier. A clear consistence of their values is observed.
Probably the most interesting result at this stage is the R dependence of
the spectrum of excited states displayed in Fig.5. The excitation spectrum
for N = 4 separates into well defined Hubbard sub bands located at U /4
and U /2 in the atomic limit (note that we display the energy per atom).
The lowest manifold contains n = 24 = 16 spin arrangements of roughly
singly occupied sites. The second subband reflects n = 48 = 32 + 16 states
with a single double occupancy, whereas the topmost subband represents
n = 2 states with two doubly occupied sites across the square diagonal and
n = 4 states with doubly occupancies located at the nearest neighboring
sites. Note a substantional manifold-energy reduction within the topmost
subband. The additional splitting within the subband is caused by the
difference between n.n and n.n.n. values of intersite Coulomb interaction
K 1 and K 2 , respectively. The parameters have dependence K n rv R- 1 at
large distances and this leads to a slow convergence of the results to the
atomic-limit values marked by horizontal lines.
268

6. Concluding remarks

In this article we have sketched our method of approach (EDABI) to the

correlated electronic states that combines the exact diagonalization in the
Fock space and a simultaneous ab initio readjustment of single-particle wave
functions within a selected subspace of their full Hilbert space. The method
is particularly useful for small systems, as illustrated on the examples ran-
ging from simple atoms, ions and molecules to nanoscopic chains, rings, and
clusters. The method is related to the multiconfiguration-interaction (MCI)
method known in quantum chemistry. However, here we put emphasis on
questions related to the physics of a systematic evolution from an atomic to
a condensed (solid-state) type of state. Additionally, we provide the values
of microscopic parameters for the hydrogenic-like systems, as well as set up
a self-adjusted wave equation (SWE) for single-particle states in the milieu
of correlated particles.
We believe that such an exact analysis within a subspace contain-
ing selected orbitals is valuable for a systematic development of reliable
methods of approach to nanoscopic systems which represent the systems
intermediate between the molecules and the solids.

Acknowledgements

The work was supported by the Committee for Scientific Research (KBN) of
Poland, Grant No. 2P03B 050 23. The discussions with Profs. J.E. Hirsch
(UCLA) and G. Czycholl (Bremen) at the NATO workshop in Hvar are
appreciated.

References

1. J. Spalek, R. Podsiadly, W. W6jcik, and A. Rycerz, Phys. Rev. B 61, 15676 (2000).
2. A. Rycerz and J. Spalek, Phys. Rev. B63, 073101 (2001); ·ibid. B 65, 035110 (2002).
3. For a brief overview see: J. Spalek et al., Acta Phys. Polonica B 32, 3189 (2001);
ib·id. B31, 2879 (2000).
4. Within the MCI method one determines the Ci., ... i.M coefficients variationally, see
e.g. R.A. Shavitt, in H. Schaeffer (editor), Methods of Electronic StmctuTe Theor'y
(Plenum Press, New York, 1977).
5. See e.g. H.A. Bethe and E.E. Salpeter, Quantum Mechanics of One- and Two-
Electron Atoms (Academic Press, New York, 1957), pp.154-156.
6. N.F. Mott, Metal-InsulatoT Tmnsitions (Taylor & Francis, London, 1990).
7. J. Spalek and A. Rycerz, Phys. Rev. B 64,161105 (2001)(R); d. also: A. Rycerz, J.
Spalek, R. Podsiadly, and W. W6jcik, in Lect'uTes on the Physics of Highly Con'elated
Electron Systems VI: Sixth Training CouTse, edited by F. Mancini, AlP Conf. Proc.
No. 629 (2002).
ON THE MULTICHANNEL-CHANNEL ANDERSON IMPURITY
MODEL OF URANIUM COMPOUNDS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

N. ANDREI and C. J. BOLECH

Center for Materials Theory, Serin Physics Laboratory, Rut-
gers University, Piscataway, New Jersey 08854-8019

Abstract. In this talk we will present the solution of the two-channel Anderson impurity
model, proposed in the context of the heavy fermion compound UBe13, and discuss
briefly the more general multi-channel case. We will show results for the thermodynamics
in the full range of temperature and fields and make the connection with the current
experimental situation.

Key words: Heavy Fermions, Kondo Model, Anderson (Impurity) Model, Multi-Channel
Models, Non-Fermi Liquid, Uranium Compounds, UBe13, Bethe Ansatz, Integrability, In-
tegrable Models, Mixed Valence, Crystal Field, Thermodynamic Bethe Ansatz, Impurity
Entropy, Impurity Specific Heat

1. Models of Uranium ions

During the last two decades, there has been growing interest in materials
whose desciption falls outside the framework of Landau's Fermi Liquid
Theory. Examples are the high T c superconductors, heavy fermions and
quasi-ID conductors.
Uranium heavy fermion compounds provide very interesting examples,
exhibiting unusual specific heat temperature dependence, I = CV IT rv
in T. This behavior is often related to multichannel Kondo physics, occa-
sioned in the case of uranium by the ground state configuration which is
degenerate and non-magnetic, so that one is led to consider a quadrupolar
Kondo scenario (1, 2).
The starting point is to consider the energies of the uranium ion in
its different states of valence. They fall on a parabola depicted in Fig. 1.
The ground state corresponds to a U4+ ionization state, with a 5f 2 shell-
configuration. Considering the spin-orbit Hund's coupling, the lowest mul-

269
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 269–276.
© 2003 Kluwer Academic Publishers
270

E fl f2 f3 f4 f5

\
\
\
\
\
\

~
/-

J 5/2 4 9/2 4 5/2

Figure 1. Schematics of the energy and lowest angular momentum multiplet for different
states of valence of a uranium ion.

tiplet corresponds to J = 4. Taking, further, into account the splitting due

to a cubic crystal field (as is the case for the heavy fermion compound
UBe13) one finds that the lowest multiplet corresponds to a non-Kramers
r 3 doublet (see Fig. 2).

X-field
.",---

J=4
L-S'----;;""""';'--<"

Figure 2. Schematic depiction of the spin-orbit and crystal field splittings for the ground
state valence configuration of uranium in a cubic medium.

A similar analysis for the neighboring states of valence indicates that

they corresponds to: (i) U3+: a 5f 3 shell-configuration that splits giving a
J = 9/2 lowest multiplet configuration, that further splits giving a lowest r 6
Kramer's doublet and (ii) U5+: a 5f 1 shell-configuration that splits giving
a J = 5/2 lowest multiplet configuration, that further splits giving a lowest
r 7 Kramer's doublet. These three valence states are in fact very close in
energy and their relative positions are not completely resolved, since the
experimental evidence is not conclusive and sometimes contradictory (3, 4).
The most accepted scenario is that of a r 3 quadrupolar doublet ground state
coming from the tetravalent state and next a r 6 magnetic doublet coming
from the trivalent state of uranium. Neglecting the contributions from the
pentavalent state of the uranium as well as those from excited crystalline
electric field states we can write down the following Hamiltonian, the two-
channel Anderson impurity model (see Fig. 3 for a schematic depiction).
 271

---a=t,. Es
E
a=+,- Eq

Figure 3. Schematics of the electronic configurations and symmetry multiplets retained

for the Hamiltonian of UBela.

+ V L [f!baCk,a,a + 4,a,a blfa]

k,a,a

with Q = Lblba+ Lf!fa = 1

where the bar on top of the index 0: indicates that it transforms according
to the complex conjugate representation. We used slave-boson language and
denoted Cq == Era and Cs == ErG (let us also define C == Cs - Cq and ,6. ==
V 2 /2). Notice that the model has full SU(2) Q9 SU(2) internal symmetry.
In the parameter regimes when Ici » ~, we can map it via a Schrieffer-
Wolff transformation into a quadrupolar two-channel Kondo model (with
a localized quadrupolar moment in the case when Cq « cs), or a magnetic
two-channel Kondo model (with a localized magnetic moment in the case
when Cq » cs), in both cases with a coupling constant given by J =
V 2 / Ics - cql (assuming zero chemical potential).
We will also consider a generalized version of the Hamiltonian, the multi-
channel Anderson impurity model. In slave-boson language the Hamiltonian
looks exactly as in the two-channel case, but this time the indices take
values according to: (i) (]" = 1, ... ,N and (ii) 0: = 1, ... , M. If we follow
the standard nomenclature we would call (]" the spin index and 0: the channel
index, and the model will be the SU (N) (2) SU (M) version of the multi-
channel Anderson model (5).
As in the two-channel case we can perform Schrieffer-Wolff transform-
ations to derive effective models in any of the two local moment regimes
of the problem. When Cq « Cs (assuming always zero electronic chem-
ical potential) we obtain an 'N-channel SU(M) Coqblin-Schrieffer model',
whereas when Cs « Cq we obtain an 'M-channel SU(N) Coqblin-Schrieffer
model'.
This model has been extensively studied over the past fifteen years by
a variety of methods: Large-N methods (6), NCA (4), conserving T-matrix
approximation (7), Monte Carlo (4) and Numerical Renormalization Group
272

(8) among others. All these techniques have difficulties accessing the mixed
valence regime.
We showed that this model is integrable and carried out a full analysis
by means of the Bethe Ansatz technique. That allowed us to obtain exact
results characterizing the ground state of the model as a function of its
parameters and very accurate numerical results for the thermodynamics of
the model for all temperatures and fields, as well as all valence regimes,
including the mixed valence. We stress that is in the mixed valence re-
gime where the Bethe Ansatz solution is most crucial, since all the other
techniques have difficulties accessing it.

2. The Thermodynamics of the Model

In this section we present the results of the numerical solution of the

Thermodynamic Bethe Ansatz equations for the two-channel case (i.e. N =
M = 2). This is the first model proposed in the context of UBe13(1). We will
study the physics of the model on its own right and also briefly discuss its
virtues and inadequacies to describe the physics of the uranium compound
that motivated it. The conclusion that we will reach is that one needs to
go beyond the simplest scenario given by the two-channel model.

In Fig. 4 we show the behavior of the impurity entropy as a function of

temperature for different values of (e - fL), and in Fig. 5 the reader can see
the effect of switching on an external field.
At high temperatures the entropy is Simp = k B In 4 in agreement with
the size of the impurity Hilbert space. For Ie - fLl » 6. the impurity entropy
is quenched in two stages. The degrees of freedom corresponding to the
higher energy doublet are frozen first. The entropy becomes Simp = k B In 2
and the system is in a localized magnetic or quadrupolar moment regime
depending on the sign of (e - fL). As the temperature is further decreased,
the remaining degrees of freedom undergo frustrated screening leading to
entropy Simp = k B In yI2. On the other hand, for values of Ie - fLl « 6.,
the quenching process takes place in a single stage. The initial stage of the
quenching process happens when the system makes the transition from a
high-energy state of valence n c = ~ (i.e. when both states of valence are
equally likely) to the state of valence that will be present at low-energies,
n~(e - fL). This is further illustrated in Fig. 6.
As (e - fL) is varied, the behavior interpolates continuously between
the magnetic (n~ = 1) and the quadrupolar (n~ = 0) scenarios. The zero
temperature entropy is found to be independent of e in accordance with
our analytic results (9). External fields, either magnetic or quadrupolar,
constitute relevant perturbations that drive the system to a Fermi Liquid
fixed point with zero entropy.
 273

2
h ~ 0
q
£-J.l= 0
£-J.l= ±2'"
1.5 £-J.l= ±4'"
£-J.l= ±6'"
£-J.l= ±s'"

0.5 r----~=--~~c:::.:...--~-.=-.~--------

T/'"

Figure 4. Impurity contribution to the entropy at zero field and as a function of

temperature for different values of E - f.L. As the temperature goes to zero all curves
approach the universal value kB In -12. Positive and negative values of E - f.L fall on top
of each other.

In Fig. 7 we show the impurity contribution to the specific heat. The

two stage quenching process gives rise to two distinct peaks. The lower
temperature peak is the Kondo contribution centered around a temper-
ature Tz (e - p,), whereas the higher temperature peak - often referred to
as the Schottky anomaly - is centered around a temperature Th (e - p,).
Approximate expressions for Th,Z can be read off from the curves:

with 1 < a < 4. For large Ie - p,1 the two peaks are clearly separated and
the areal under the Kondo peak is kB in V2 while that under the Schottky
peak is kB in 2.

As mentioned earlier, the model was proposed as a description for

the uranium ion physics of UBe13. It is expected to describe the lattice
above some coherence temperature. We provide in the inset of Fig. 7 the
experimental data for the 5f-derived specific heat of the compound. It is
obtained by subtracting from its total specific heat, the specific heat of the
isostructural compound ThBe13 containing no 5f electrons (10, 11). This
way one is subtracting the phonon contribution as well as the electronic
1 Here by area we mean the following integral: JC v (T) d (In T). In other words, the
area as plotted, i.e. in a logarithmic scale.
274

E-fl = 0
E-fl = ±2'"
1.5 E-fl = ±4'"
E-fl = ±6'"
E-fl = ±S'"

E-fl<O

T/'"

Figure 5. Impurity contribution to the entropy in the presence of an applied quadrupolar

field and as a function of temperature for different values of c - jL. As the temperature
goes to zero all curves approach zero. Positive and negative values of c - jL fall on top of
each other only at high temperatures and, asymptotically, for low temperatures.

contribution from electrons in s, p and d shells (the procedure is quite

involved and we refer the reader to the cited articles for full details). The
sharp feature at rv 0.8K signals the superconducting transition of UBe13
and falls outside the range where this compound might be described by
considering a single impurity model.
In order to be certain of having eliminated lattice effects it would be
better to carry out the measurements on U 1- xThxBe13. For x > 0.1 the
compound has no longer a superconducting transition and the lattice coher-
ence effects are largely suppressed. Further, there are several experimental
indications that support the idea of an impurity model description of the
thoriated compound for a wide range of temperatures (12).
Concentrating on the temperature range containing the Kondo and
Schottky peaks we conclude that no values of EO and ,6. of the 2-channel
model yield a good fit. Further, the entropy obtained by integrating the
weight under the experimental curve falls between kB in 4 and kB in 6. This
suggests that a full description of the impurity may involve another high
energy multiplet (possibly a r 4 triplet, d. with the work of Koga and Cox
(8)) close to the r 3 to yield the peak for the Schottky anomaly, with the
r 6 doublet falling outside the range of measurements. The nature of the
multiplet could be deduced from further specific heat measurements. For
an n-plet degenerate with the r 3 , one has an SU(2) Q9 SU(n + 2) Anderson
model and the area under ctrnp IT is then given by kB In[nt 4 sec n~4], while
 275

-"-"-"-"-"-"-"-"-"-"-'" E-f.l= 0
-'-'-'-'-'-'-'-'-'-'-'-'-.-., " E-f.l=±2~
" "
0.8 " '. E-f.l=±M
.......................................
" ., E-f.l=±lO~
".
,", . E-f.l=±20~

.:. ---------------------------,-~~ft"*~,:·j3;-~
0.6 ',',

/ .
/' ,/
....................................... /' /
0.2 /
./
./
_.. _.. _.. _.. _.. _.. _.. _.. _.. _ _.. '
-'-'-'-'-'-'-'-'-'-'-'-'-'-'~
..
/

ot::::::r====:::r=====r=====r:::::::=-L-_----.L__U
10- 3 10- 2 10- 1 10° 10] 10 2 10 3
T/13.

Figure 6. Charge content of the impurity site as a function of temperature for different
values of E - p,. At high temperatures all curves start from n c = 1/2 and flow, as the
temperature is lowered to specific values n~) (E - p,).

if the n-plet is slightly split off the doublet, the area is given by kB In[n~4]
(13). Also the scale of energy involved is then determined by crystal field
splitting and corresponds to the scale observed. The split off non-magnetic
triplet (i.e. n = 3 in the p configuration seems, indeed, to be the most
reasonable scenario for U Be13. We are currently working in this direction.

Acknowledgements

We thank the organizers for the invitation to participate in the NATO

Advanced Research Workshop on Concepts in Electron Correlation.

References

1. D. L. Cox, Phys. Rev. Lett. 59, 1240 (1987).

2. A. P. Ramirez et al., Phys. Rev. Lett. 73, 3018 (1994).
:3. F. G. Aliev et al., Europhys. Lett. 32, 765 (1995).
4. A. Schiller, F. B. Anders and D. L. Cox, Phys. Rev. Lett. 81, 3235 (1998).
5. D. L. A. Cox and A. Zawadowski, Exotic Kondo Effects in Metals: Magnet·ic ·ions
in a crystaline electric field and tunnelling centres (Taylor & Francis, London UK,
1999).
6. D. L. Cox and A. E. Ruckenstein, Phys. Rev. Lett. 71, 1613 (1993).
7. J. Kroha and P. Wolfle, Act. Phys. Pol. B 29, 3781 (1998).
8. M. Koga and D. L. Cox, Phys. Rev. Lett. 82, 2575 (1999).
276

0.6
£-J.l= 0

.,I. E-J.l=±2li
0.25
. ,
0.4 I.
£-J.l=±M
I . £-J.l=±lOli
l /Ii."i. ' 0.2
r. ...- 0.2
" .:\. i
.i\....·················.. .· ,. :' \:. i
I . I ..
•••.,l I . . \. 1
1---"=...LU-LllL-----'---'--'-'-'-'-ilL----'---'--'-~~---'--'0
I : I I·.. i 0.15
I . I I·
I :. I:.i i
I : . I: i
I : I I:. 0.1
I : . \: 1
I . I ,..
I .: i i· \
,. .......
... .···t·· i \\ 0.05
/
... .
.
I
I i \.'.,,,
:-: / I \-
.... .....; t'-.-._ ..... . ~

4
0
10

Fig1tre 7. Main plot: impurity contribution to the specific heat as a function of tem-
perature for different values of E - IL. As this parameter approaches zero the Kondo
contribution (left) and the Schottky anomaly (right) collapse in a single peak. Inset:
experimental data for the 5f-derived specific heat of UBela.

9. C. J. Bolech and N. Andrei, Phys. Rev. Lett. 88,2:37206 (2002).

10. R. Felten et al., Europhys. Lett. 2, 323 (1986).
11. R. Felten, G. Weber and H. Rietschel, J. Magn. Magn. Mat. 63 & 64, 38:3 (1987).
12. F. G. Aliev, S. Viilar and V. V. Moshchaikov, J. Phys. Condo Matt. 8, 9807 (1996).
1:3. A. Jerez, N. Andrei and G. Zanind, Phys. Rev. B 58, 3814 (1998).
ANOMALOUS BEHAVIOR IN RARE-EARTH AND ACTINIDE
SYSTEMS.

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

B. COQBLIN
Laboratoire de Physique des Solides, Bat. 510, Universite Paris-
Sud, 91405-0rsay, France

Abstract. Many different anomalous behaviours (intermediate valence, heavy fermions,

competition between magnetic order and Kondo effect) have been experimentally ob-
served in alloys and compounds containing anomalous rare-earths, ee, Yb, Eu and also
Pr, Sm and Tm, as well as in systems with non-magnetic actinides U, Np and Pu.
This paper reviews theoretical models describing several interesting cases: the transitions
accompanied by a valence change such as for example in cerium metal and Eu(Pd 1 - a,Aua,hSi2
alloys; the "normal" Kondo effect and the heavy fermion behaviour; the "underscreened"
Kondo effect which can be applied to actinide compounds; the occurrence of antiferromag-
netic, Kondo, Non Fermi Liquid and spin glass phases in cerium or uranium disordered
alloys.

Key words: Kondo Effect, Heavy Fermions, Rare-Earth, Uranium, Kondo Lattice, Un-
derscreened Kondo Effect, Spin Glass

1. Introduction

In the series of rare-earth metals, most of them are magnetic with a mag-
netic moment corresponding to the 4fn (with integer n values) configuration
and a valence of 3 corresponding to three electrons in the conduction band;
these metals are called "normal rare-earth metals" (1). There are three ex-
ceptions : cerium which has a peculiar phase diagram with two face-centered
cubic phases r and a which have different valences (respectively almost 3
and clearly intermediate between 3 and 4)(2), europium and ytterbium
which are divalent at normal pressure. On the other hand, many systems
containing rare-earths have been extensively studied and, in addition to
cerium, ytterbium and europium, three other rare-earths have been found
to have "anomalous" behavior in alloys and compounds, namely Pr, 8m

277
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 277–286.
© 2003 Kluwer Academic Publishers
278

and Tm. For example, PrSn3 (3) and TmS (4) present a Kondo effect, while
SmS (5) and TmSe (6) present an intermediate valence behavior. Moreover,
volume-collapse transitions have been observed at extremely high pressures
(7) in praseodymium at 20 CPa (8)and gadolinium at 59 CPa (9); thus, we
can imagine that gadolinium changes valence at 600 kbar !
On the other hand, the 5f electrons are less localized at the begin-
ning of the actinide series than the 4f ones in rare-earths and it results
that uranium, neptunium and plutonium metals are not magnetic, while a
strong magnetism occurs for curium and berkelium. Thus, a strong heavy
fermion character or even magnetism can occur in compounds or alloys
with uranium or also in a few known cases with neptunium and plutonium.
It is well known that different anomalous behaviors have been observed
and theoretically studied in mainly ee, Vb, Eu, U but also in other an-
omalous rare-earth systems (10, 11). In the present paper, we will review
briefly the intermediate valence or the Kondo behavior associated with the
heavy fermion character and we will present some recent features of these
two topics, with a special emphasis on the underscreened Kondo effect and
the effect of disorder in ternary cerium or uranium compounds.

2. Intermediate Valence transitions

The first extensively studied transition accompanied by a valence change

is the ,-a transition of cerium metal (2). The phase diagram of cerium
metal has two face-centered cubic phases a and, and, if one extrapolates
the valence from the atomic radius, one obtains a valence very close to 3
for , - Ce and a valence of roughly 3.6 for a - Ceo The ,-a transition
is first-order at low pressures and temperatures; but the discontinuity in
the different physical properties decreases along the transition and the
transition becomes a second-order one above a critical point located at
roughly 20 kbar and 600 K. At normal pressure and low temperatures,
cerium undergoes a transition to the double hexagonal close-packed (3 phase
which orders antiferromagnetically below 13 K and finally cerium becomes
superconducting at very low temperatures above 20 kbar (10).
A similar transition has been observed in two systems containing europium
: Eu(Pd 1 - x Au x hSi 2 (12) and more recently Eu(Gel-xSixhNi2 (13). For
example, in the first case, when gold concentration decreases, there is firstly
an antiferromagnetic order and then a first-order transition between an
almost divalent Eu to an almost trivalent Eu, which ends up also at a
critical point.
We can also cite the case of the compound YbInCu4, which undergoes
at normal pressure an isostructural valence transition from a high temper-
ature magnetic phase to a low temperature non magnetic one; ytterbium
 279

is trivalent in the magnetic phase, but the valence change to the non mag-
netic one is very small, of order 0.1 (14). This transition has been studied
extensively under pressure (15) or in related materials (14, 16).
These isostructural valence transitions have been extensively studied
theoretically. The first theoretical explanation starts from the orbitally de-
generate Anderson Hamiltonian and accounts for the magnetic-nonmagnetic
transition accompanied by a large valence change and the existence of the
critical point (17). Then, the Ramirez-Falicov model starts from an ionic
description of the 4f electrons and can also account for the phase diagram
of cerium (18). The main point of these two first theoretical models is
the explanation of a large valence change. Another theoretical explana-
tion of the i-a transition of cerium was based on the Kondo effect and
assumes a large change of the Kondo temperature when going from the
non magnetic phase to the magnetic one (19, 20). More recently, many
calculations involving detailed band calculations and based in particular on
the dynamical mean-field theory (DMFT) were performed to account for
the phase diagram of cerium and to give in particular a better description
of the photoemission experiments (21, 22, 23). Several theoretical models,
including DMFT calculations within the Falicov-Kimball model (24), have
been developed to account for the experimental phase diagrams of Eu or Yb
systems such as for example YbInCu4 compound; but in addition, there
often exists an antiferromagnetic phase in these phase diagrams and at
present there is not a full theoretical model for explaining such a complete
phase diagram.
In actinide metals at the beginning of the series, the 5f electrons are
less localized than the equivalent 4f electrons in rare-earth metals and the
situation is clearly different : there is a strong d-f hybridization and it
results that U, Np and Pu metals are not magnetic, while these actinides
can be magnetic in alloys or compounds where this hybridization is smaller.
Then the 5f electrons become more localized in the middle of the series and
a strong magnetism occurs for curium and berkelium. The phase diagrams
of the first actinides are complicated and in particular plutonium has five
phases at normal pressure (25, 26). The curve of the atomic volumes lies
between the corresponding one for 5d transition metals up to Pu and that of
lanthanides starting at Am, which is a good indication of the rapid increase
of the 5f localization between Pu and Am. On the other hand, there is a
huge volume expansion (26percent) between the low temperature a phase of
plutonium and the high temperature <5 phase, which is of the same order as
the volume expansion occurring at the i-a transition of cerium. The volume
expansion experiments, as well as the analysis of transport and magnetic
properties suggest a strong localization of 5f electrons and a Kondo effect
in <5 - Pu (27, 28). On the other hand, plutonium is close to be magnetic
280

at low temperatures and the spin fluctuation model has been used a long
time ago in the 70's to account fairly well for the resistivity which is large
and passing through a maximum as a function of temperature (29) and for
the thermoelectric power (30) of 0: - Pu.

3. The Kondo effect

The long story of the Kondo effect starts 40 years ago with the third-
order calculation of J. Kondo (31) who explains theoretically the resistivity
minimum observed in dilute alloys such as Cu-based alloys with Fe or Mn
impurities and lanthanum-based alloys with Ce impurities. The properties
of many cerium, ytterbium or also uranium compounds or alloys are well
understood in the theoretical framework of the Kondo effect and we will
describe here some features of this difficult problem. Cerium and ytter-
bium systems are well described by the Kondo effect, when the number
of 4f-electrons is close to 1 in the case of cerium or to 13 (or 1 hole) in
the case of ytterbium. First of all, the single-impurity Kondo effect has
been exactly solved(32): at low temperatures, the system has a "Fermi
Liquid" behavior with a T 2 law for the electrical resistivity and very large
values of both the electronic constant of the specific heat and the magnetic
susceptibility, which were at the origin of the name "heavy fermions" given
to these systems (33, 34, 35). At high temperatures with respect to the
Kondo temperature, the magnetic contribution to the electrical resistivity
passes generally through a maximum corresponding to the overall crystal
field splitting and decreases then as Log T (36). The other transport and
magnetic properties have been extensively studied in the low temperature
and high temperature regimes.
The second extensively studied problem is the Kondo-lattice one. In
the case of a lattice, there exists a strong competition between the Kondo
effect and magnetic ordering, arising from the RKKY (Ruderman-Kittel-
Kasuya-Yosida) interaction between rare-earth atoms at different lattice
sites. This situation is well described by the Doniach diagram(37), which
gives the variation of the Neel temperature and of the Kondo temperature
with increasing antiferromagnetic intrasite exchange interaction JK(> 0)
between localized spins and conduction-electron spins. Usual theories of
the one-impurity Kondo effect and of the RKKY interaction yield a Kondo
temperature TKo, that is proportional to exp( -1 / pJK ), and an ordering
temperature (Neel or, in some cases, Curie), T No , proportional to pJ'k, p
being the density of states for the conduction band at the Fermi energy.
Thus, for small pJK values, T No is larger than TKo and the system tends to
order magnetically, with often a reduction of the magnetic moment due to
the Kondo effect; on the contrary, for large pJK , TKo is larger than T No and
 281

the system tends to become non magnetic. The actual ordering temperature
TN, therefore, increases initially with increasing pJK, then passes through
a maximum and tends to zero at a critical value pJ'K corresponding to
a "quantum critical point" (QCP) in the Doniach diagram. Such a be-
havior of TN has been experimentally observed with increasing pressure
in many Kondo compounds, such as for example in CeAb (Ref. (38)) or
in CeRh 2Sid39) Thus, we can conclude that the variation of the Neel
temperature predicted by the Doniach diagram is well observed experiment-
ally in many cerium compounds. We also know that the Neel temperature
starts from zero at a given pressure, and increases rapidly with pressure
in YbCu2Si2 (Ref. (40)) or in related ytterbium compounds, in agreement
with the Doniach diagram.
The one-impurity model predicts an exponential increase of the Kondo
temperature with pJK. This means that the Kondo temperature should
increase with increasing pressure in cerium compounds and with decreasing
pressure in ytterbium compounds, which agrees well with many observa-
tions. However, deviations seem to occur in some cerium compounds, such
as CeRh 2Si 2 (Ref. (39)), CeRu2Ge2(41, 42), or Ce2Rh3Ges (43), where
the actual Kondo temperature observed in a lattice can be significantly
different from the one derived for the single-impurity case. Thus, in order
to account for such an effect, we have studied in detail the Kondo-lattice
model within a mean-field approximation and with both intrasite Kondo
exchange and intersite antiferromagnetic exchange, treating successively
the half-filled case (corresponding to a number of conduction electrons
n = 1) (44) and then the general case n < 1 (45, 46, 47, 48) which
gives a much better description of the metallic cerium systems. We have
studied in detail this problem in Ref. (48) and we have shown that the
Kondo temperature tends to decrease with an increasing intersite exchange
interaction and a decreasing number of conduction electrons corresponding
to the so-called "exhaustion" limit (49). This calculation addresses sev-
eral questions : first, it would be very interesting to understand better
the conditions yielding a Kondo temperature for the lattice much different
than the single-impurity one and further experiments are certainly needed.
Second, the discussion on the Kondo-lattice problem addresses again the
difficult issue of the nature of the ground state and screening in the Kondo-
lattice problem. We have shown here that, as the number of conduction
electrons is reduced, exhaustion may be compensated by formation of in-
tersite singlets of localized spins and exact calculations for small clusters
would be interesting (50, 51). The third interesting question concerns the
derivation of a correlation temperature below which short-range magnetic
correlations appear, in good agreement with neutron scattering experiments
in cerium compounds. Finally, it is interesting to notice that taking into
282

account lattice effects seems to be important for describing the properties of

cerium or other anomalous rare-earth compounds at low temperatures, as
shown, for example, in photoemission experiments (52, 53). Thus, further
theoretical and experimental developments are certainly necessary to better
understand the Kondo-lattice problem.
Then, up to now, we have described the Kondo compounds within the
"Fermi liquid" picture. But, it is now well established that an anomalous
effect, called "Non Fermi Liquid" (NFL), occurs very often at or around the
QCP in many cerium compounds. This NFL behavior is characterized by
laws for the physical properties different from those derived within the clas-
sical Fermi Liquid case: for example, the electronic specific heat constant is
no more constant with temperature and the electrical resistivity is no more
behaving as T 2 . Moreover, there are clearly different NFL behaviors accord-
ing to the different compounds studied. We can distinguish at least three
possible origins for these NFL behaviors: the first one is the proximity to
the QCPi for example, some cerium compounds are antiferromagnetic, have
then an NFL just above the QCP before recovering a Fermi Liquid behavior,
when the pressure (for example in Ce7Ni3 (54)) or the relative concen-
tration is varied (for example in the well studied case of Ce(Cul-xAux)6
alloys (55)). Then, the disorder can produce an NFL behavior especially in
ternary alloys (56) and finally the multichannel Kondo effect can also be
at the origin of another NFL behavior. An extensive review discusses the
present status of knowledge of NFL, including a compilation of data in over
50 systems containing mainly cerium or uranium citeStewart.
Moreover, the Kondo effect is simply defined in the case of cerium or
Ytterbium systems, where there is one 4f electron or hole screened by one
conduction electron. But, in the case of uranium systems, the situation is
even more complicate, because there are several f-electrons, corresponding
to the so-called "multi-channel" Kondo effect. If we call S the f-electron
spin and n the number of channels for conduction electrons, the normal
Kondo effect corresponds to n = 28, while the "underscreened" Kondo
effect corresponds to n < 28 (where there are too many f-electrons with
respect to the conduction electrons) and the" overscreened" case to n > 28
(where there are not enough f-electrons). The multichannel model for a
single impurity has been completely solved analytically (58). The multi-
channel is more appropriate for uranium systems where clearly there is a
number of 5f electrons larger than 1. In fact, in uranium compounds such
as UPt3 (59), UNi 2 Ah or UPd 2 Ah (60), there exist both a large heavy-
fermion character and a magnetic ordering and the underscreened model
can account for this "coexistence" between the Kondo effect and magnetic
order, in contrast to the "competition" observed in cerium systems(61). A
remarkable example of underscreened Kondo effect is the case of uranium
 283

monochacogenides, as evidenced by Schoenes (62): for example, the com-

pound UTe is ferromagnetic with a large Curie temperature of 102 K and
has a Kondo behavior characterized by a LogT decrease of the resistivity
at higher temperatures. Thus, there is clearly here coexistence between the
Kondo effect and a ferromagnetic ordering, which can be accounted for by
the underscreened Kondo effect.
Then, an unusual interplay between Kondo effect and spin-glass beha-
vior has been recently observed in several other disordered cerium alloys
such as CeNi1-xCu x (63, 64), CeCoGe3-xSix,(65) Ce2Aul-xCoxSi3 (66)
or Ce2Pdl-xCoxSi3 alloys (67). In particular, a spin glass-Kondo trans-
ition occurs in CeNi1-xCu x for x = 0.2 and for x > 0.2 there exists a
ferromagnetic phase at low temperatures and a spin glass phase at higher
temperatures. In the case of Ce2Aul-xCoxSi3 alloys (66) , the sequence of
SG-AF-Kondo phases is obtained with increasing x at low temperatures.
NFL and spin glass phases have been also observed in disordered uranium
alloys (68), as for example a sequence of AF-NFL-SG phases with increas-
ing x in UCu5-xPdx alloys (69) or the opposite sequence AF-SG-NFL
in U1 - x La x Pd 2 Ab alloys (70). We have studied theoretically firstly the
Spin Glass-Kondo transition (71) by considering a Kondo lattice with the
classical intrasite Kondo exchange interaction and a randomly distributed
Ising intersite exchange interaction and we have obtained theoretically a
SG-Kondo transition as a function of the Kondo exchange parameter JK,
in agreement with the experiment in CeNil-xCUx alloys. Then, we have
taken a random Ising intersite exchange interaction with a non zero mean
value in order to describe a ferromagnetic phase and we have obtained a
Ferromagnetic-Spin Glass-Kondo phase sequence (72), in better agreement
with some experimental results. However, we have found a first-order trans-
ition without any QCP between the SG and Kondo phases and then we have
not obtained any "mixed" SG-Kondo phase. The existence of a QCP in the
phase diagram can be solved by considering an additional transverse field
which mimics the spin-flipping in the spin glass exchange interaction(73),
but the unsolved problem of the absence of a "mixed" SG-Kondo phase is
certainly connected to the mean-field approximation used to describe the
Kondo intersite exchange interaction.
Finally, a very fascinating subject concerns the occurrence of supercon-
ductivity in some cerium or uranium compounds at low temperatures. The
first evidence of superconductivity in cerium compounds has been obtained
in CeCu2Si2 at normal pressure below 0.6 K by Steglich et al (74). At
present, there are relatively many cerium compounds which become super-
conducting generally at high pressure close to the QCP(75, 76). There are
also several uranium compounds such as UPt3 (59) or UPd 2 Ah (60), which
become superconducting at low temperatures generally at normal pressure;
284

these uranium compounds present, therefore, the particularity of having

an heavy-fermion character and becoming magnetically ordered and super-
conducting at low temperatures. We can finally cite the case of the newly
studied compound UGe2 which is ferromagnetic at normal pressure and
becomes superconducting at high pressure just before the QCP (76). This
competition between superconductivity, the heavy-fermion character and a
magnetic order is clearly a new challenge which is not solved theoretically
at present.

4. Conclusion

Thus, we have reviewed some features of the anomalous behaviors found in

many rare-earth and actinide systems. In the case of intermediate valence
systems, it is interesting to note that the isostructural phase transition
with a valence change observed in cerium metal has been also found in
Eu or Yb alloys or compounds. On the other hand, we have reviewed
several features of the Kondo effect which are extensively studied at present,
namely the Kondo-lattice problem, the multichannel Kondo effect, the spin
glass-Kondo competition and the superconductivity in cerium or uranium
compounds.

References

1. B. Coqblin, The electronic structure of Tare-earth metals and alloys: the magnetic
heavy Tare-earths, Academic Press, (1977).
2. A. Jayaraman, Phys. Rev., 137, A179 (1965).
3. P. Lethuillier and P. Haen, Phys. Rev. Lett., 35, 1391 (1975).
4. F. Lapierre, P. Haen, B. Coqblin, M. Ribault and F. Holtzberg, Proceedings of LT
16, Los Angeles (Aug. 1981).
5. M.B. Maple and D. Wohlleben, Phys. Rev. Lett., 27, 511 (1971).
6. M. Ribault, J. Flouquet, P. Haen, F. Lapierre, J.M. Mignot and F. Holtzberg, Phys.
Rev. Lett., 45, 1295 (1980).
7. A.K. McMahan, C. Huscroft, RT. Scalettar and E.L. Pollock, Journal of Computer-
Aided Materials Design, 5, 131 (1998).
8. G.S. Smith and J. Akella, J. App!. Phys. 53, 9212 (1982).
9. H. Hua, Y.K. Vohra, J. Akella, S. Weir, R Ahuja and B. Johansson, Rev. High
Press. Sci. Techno!., 4, 233 (1998).
10. B. Coqblin, in Magnetism of metals and alloys, ed. by M. Cyrot, (North-Holland,
1982).
11. A. C. Hewson, The Kondo problem to Heavy Fermions, (Cambridge University
Press, Cambridge, 1992).
12. C.U. Segre, M. Croft, J.A. Hodges, V. Murgai, L.C. Gupta and RD. Parks, Phys.
Rev. Lett., 49, 1947 (1982).
13. H. Wada, M.F. Hundley, R Movshovich and J.D. Thompson, Phys. Rev. B, 59,
1141 (1999).
14. J.L. Sarrao, Physica B, 259-261, 128 (1999) and references therein.
 285

15. A. Uchida, M. Kosaka, N. Mori, T. Matsumoto, Y. Uwatoko,J.L. Sarrao and J.D.

Thompson, Physica B, 312-313, 339 (2002)
16. M. Ocko and J.L. Sarrao, Physica B, 312-313, 341 (2002).
17. B. Coqblin and A. Blandin, Adv. in Phys., 17, 281 (1968).
18. R Ramirez and L.M. Falicov, Phys. Rev. B, 3, 2425 (1971).
19. M. Lavagna, C. Lacroix and M. Cyrot, J. Phys.F, 13, 1007 (1983).
20. J.W. Allen and RM. Martin, Phys. Rev. Lett., 49, 1106 (1982).
21. A. Svane, Phys. Rev. B, 53, 4275 (1996).
22. M.B. Zolfl, LA. Nekrasov, Th. Pruschke, V.L Anisimov and J. Keller, Phys. Rev.
Lett., 87, 276403 (2001).
23. K Held, A.K McMahan and RT. Scalettar, Phys. Rev. Lett., 87, 276404 (2001).
24. V. Zlatic and J.K Freericks, Acta Phys. Pol. B, 32, 3253 (2001).
25. D.R Stephens, J. Phys. Chern. Solids, 24, 1197 (1963).
26. Challenges in Plutonium Science, Los Alamos Science, 26 (2000).
27. S. Meot-Reymond and J.M. Fournier, J. of Alloys and Compounds, 232,119 (1996).
28. S.Y. Savrasov, G. Kotliar and E. Abrahams, Nature, 410, 793 (2001).
29. R. Jullien, M.T. Beal-Monod and B. Coqblin, Phys. Rev. B, 9, 1441 (1974).
30. J.R. Iglesias-Sicardi, R Jullien and B. Coqblin, Phys. Rev. B, 17, 2366 (1978).
31. J. Kondo, Progr. Theoret. Phys. (Kyoto), 32, 37 (1964).
32. KG. Wilson, Rev. Mod. Phys., 47, 773 (1975).
33. P. Nozieres, J. Low Temp. Phys., 17, 31 (1974).
34. K Andres, J.E. Graebner and H.R Ott, Phys. Rev. Lett., 35, 1779 (1975).
35. B. Coqblin, J. Arispe, A.K Bhattacharjee and S.M.M. Evans, Selected Topics in
Magnetism, Frontiers in Solid State Sciences, vol. 2, ed. by L.C. Gupta and M.S.
Murani, (World Scientific, Singapore 1993), p.75.
36. B. Cornut and B. Coqblin, Phys. Rev. B, 5, 4541 (1972).
37. S. Doniach, Proceedings of the Int. Con/. on Valence Instabilities and Related
Narrow-band Phenomena, ed. by R D. Parks, Plenum Press, p.168 (1976).
38. B. Barbara, H. Bartholin, D. Florence, M.F. Rossignol and E. Walker, Physica B,
86-88, 177 (1977).
39. T. Graf, J. D. Thompson, M. F. Hundley, R. Movshovich, Z. Fisk, D. Mandrus, R
A. Fischer and N. E. Phillips, Phys. Rev. Lett., 78, 3769 (1997).
40. K Alami-Yadri, H. Wilhelm and D. Jaccard, Physica B, 259-261, 157 (1999).
41. S. Sullow, M.C. Aronson, B.D. Rainford and P. Haen, Phys. Rev. Lett., 82, 2963
(1999).
42. H. Wilhelm, K Alami-Yadri, B. Revaz and D. Jaccard, Phys. Rev. B, 59, 3651
(1999).
43. K. Umeo, T. Takabatake, T. Suzuki, S. Hane, H. Mitamura and T. Goto, Phys.
Rev. B, 64, 144412 (2001).
44. J. R. Iglesias, C. Lacroix and B. Coqblin, Phys. Rev. B, 56, 11820 (1997).
45. A. R. Ruppenthal, J. R Iglesias and M. A. Gusmao, Phys. Rev. B, 60, 7321 (1999).
46. B. Coqblin, M. A. Gusmao, J. R Iglesias, C. Lacroix, A. Ruppenthal and Acirete
S. Da R Simoes, Physica B, 281-282, 50 (2000).
47. C. Lacroix, J. R. Iglesias and B. Coqblin, Physica B, 312-313, 159 (2002).
48. B. Coqblin, C. Lacroix, M. A. Gusmao, J. R Iglesias, to be published.
49. P. Nozieres, Eur. Phys. J. B, 6, 447 (1998).
50. H. Tsunetsugu, M. Sigrist, and K Ueda, Rev. Mod. Phys., 69, 809 (1997).
51. C. Busser, E. V. Anda, J. R. Iglesias, and B. Coqblin, J. Magn. Magn. Mater.,
226-230, 134 (2001).
52. A. N. Tahvildar, M. Jarrell and J. K Freericks, Phys. Rev. B, 55, R3332 (1997).
286

53. A.J. Arko, J.J. Joyce, A.B. Andrews, J.D. Thompson, J.L. Smith, D. Mandrus,
M.F. Hundley, A.L. Cornelius, E. Moshopoulou, Z. Fisk, P.C. Canfield and Alois
Menovsky, Phys. Rev.B, 56, R7041 (1997).
54. K. Dmeo, T. Takabatake, H. Ohmoto, T. Pietrus, H. von Lohneyssen, K. Koyama,
S. Hane and T. Goto, Phys. Rev. B, 58, 12095 (1998).
55. H. von Lohneyssen, J. Mag. Mag. Mater.,200, 532 (1999).
56. E. Miranda, V. Dobrosavljevic and G. Kotliar, Phys. Rev. Lett., 78, 290 (1997).
57. G.R Stewart, Rev. Modern Phys., 73, 797 (2001).
58. P. Schlottman and P. D. Sacramento, Adv. in Phys., 42, 641 (1993).
59. G.R Stewart, Z. Fisk, J.O. Willis and J.L. Smith, Phys. Rev. Lett., 52, 679 (1984).
60. C. Geibel et al., Z. Phys. B, 83, 305, 1991 and 84, 1 (1991).
61. Karyn Le Hur and B. Coqblin, Phys. Rev. B, 56, 668 (1997).
62. J. Schoenes, private communication and J. Schoenes, Journal Less-Common Met.,
121, 87 (1986).
63. J. C. Gomez Sal, J. Garcia Soldevilla, J.A. Blanco, J.I. Espeso, J. Rodriguez
Fernandez, F. Luis, F. Bartolome and J. Bartolome, Phys. Rev. B, 56, 11741 (1997).
64. J. Garcia Soldevilla, J.C. Gomez Sal, J.A. Blanco, J.I. Espeso and J. Rodriguez
Fernandez, Phys. Rev. B, 61, 6821 (2000).
65. D. Eom, M. Ishikawa, J. Kitagawa and N. Takeda, J. Phys. Soc. Japan, 67, 2495
(1998).
66. S. Majumdar, E.V. Sampathkumaran, St. Berger, JVI. Della Mea, H. Michor, E.
Bauer, JVI. Brando, J. Hemberger and A. Loidl, Solid State Comm., 121,665 (2002).
67. E.V. Sampathkumaran, to be published in the proceedings of SCES02 Conference,
Acta Physica Polonica.
68. M.B. Maple, M.C. de Andrade, J. Herrmann, Y. Dalichaouch, D.A. Gajewski, C.L.
Seaman, R. Chau, R. Movshovich, M.C. Aronson and R. Osborn, J. Low Temp.
Phys., 99, 223 (1995).
69. R. Chau and M.B. Maple, J. Phys. C: Condens. Matter, 8, 9939 (1996).
70. V.S. Zapf, RP. Dickey, E.J. Freeman, C. Sirvent and M.B. Maple, Phys. Rev. B,
65, 024437 (2002).
71. Alba Theumann, B. Coqblin, S.G. Magalhaes and A.A. Schmidt, Phys. Rev. B, 63,
054409 (2001).
72. S.G. Magalhaes, A.A. Schmidt, Alba Theumann and B. Coqblin, to appear in Eur.
Phys. J. B.
73. Alba Theumann and B. Coqblin, to be published.
74. F. Steglich, J. Aarts, C. D. Bredl, W. Lieke, D. Meschede and W. Franz, Phys.
Rev. Lett., 43, 1892 (1979).
75. F. Steglich, J. Mag. Mag. Mater., 226-230, 1 (2001).
76. S.S. Saxena, P. Agarwal, K. Ahilan, F.M. Grosche, RK.W. Haselwimmer, M.J.
Steiner, E. Pugh, I.R Walker, S.R Julian, P. Monthoux, G.G. Lonzarich, A. Huxley,
I. Sheikin, D. Braithwaite and J. Flouquet, Nature, 406, 587 (2000).
DESCRIBING THE VALENCE-CHANGE TRANSITION BY
THE DMFT SOLUTION OF THE FALICOV-KIMBALL MODEL

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

v. ZLATIC
Institute of Physics, Bijenicka c. 46, 10 001 Zagreb, Croatia
J. K. FREERICKS
Department of Physics, Georgetown University, Washington,
DC 20057, U.S.A.

Abstract. The anomalous properties of the YbInCu4 family of intermetallic compounds

are discussed and the experimental data are compared with the dynamical mean field
theory (DMFT) of the Falicov-Kimball model.

Key words: Valence-Change Transition, Falicov-Kimball Model, Dynamical Mean-Field

Theory

1. Introduction to Valence-Change Materials

Yb-based intermetallic compounds have an interesting isostructural valence-

change transition (1, 2, 3, 4, 5, 6). Here, we briefly describe the most
characteristic features of these systems, taking Yb 1 - xYx1nCu4 as an ex-
ample, and show that the Falicov-Kimball model provides a qualitative
description of the experimental data.
At zero doping and ambient pressure the properties of YblnCu4 are
dominated by a first-order valence-change transition at about 40 K. The
valence of the Yb ions changes from Yb 3 + above Tv to Yb 2 .85 + below
Tv (1, 7), where Tv is the transition temperature, and the lattice expands
by about 5 %. The crystal structure remains in the CI5(b) class and
the volume expansion estimated from the atomic radii of the Yb 3 + and
Yb 2+ ions is compatible with the valence change estimated from LIII-edge
data (1, 8). Specific heat data shows a first-order transition at Tv with an
entropy change 6.S c:::' R ln 8 corresponding to a complete loss of magnetic
degeneracy in the ground state (4). Neutron scattering does not provide any

287
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 287–296.
© 2003 Kluwer Academic Publishers
288

evidence for long range order below Tv (9). Y-doping reduces Tv until a crit-
ical concentration of 15 % of Y ions is reached, where the high-temperature
phase extends down to T=O K (5, 6, 10). The experimental results for the
resistivity, susceptibility and thermopower (6) of Yb 1- xYx1nCu4 are shown
in Fig. 1.

,-.....
.:Y

0
0 0.8 (0)
n ~0.15
.
'-..../
()
-0 0.6 E . x=O
"- 0.1 o x=0.3
Q x=O :J
E * x=O.5
"'-...
,-..... 0.4 x=0.1 ~0.05 n x=O.7

.
!"'•••••••• . x~0.9
f-
'-..../ * x=0.5 P
>< 0
()
-0
0.2 x=0.9
0 100 200
Q x x=1 Temperature [K]
0
0 100 200 300
Temperature [K]

6 (b)
4
2
o
-2
• x=0.1
-4
.. x=0.3
fJ) -6
* x=0.5
-8
• x=0.9
-10
o 100 200 300
Temperature [K]

Figure 1. Panel (a) shows the resistivity and the magnetic susceptibility of
Yb 1 - x Y xlnCu4 as function of temperature for various concentrations of Y ions (6).
Note, all the "high-temperature" data can be collapsed onto a single universal curve, by
normalizing the susceptibility with respect to an effective Yb-concentration (not shown).
Panel (b) shows the thermopower of Yb 1 - xY xlnCu4 as a function of temperature for
various concentrations of Y ions (6).

The low-temperature phase (T ::; Tv) shows anomalies typical of an

intermetallic compound with a fluctuating valence. The electronic specific
heat and the susceptibility are enhanced (4); the ESR data (11) also indic-
ates a large density of states at the Fermi level E F . The electrical resistance
and the Hall constant are small and metallic, and the low-temperature slope
of the thermoelectric power is large (6). The optical conductivity is Drude
like, with an additional structure in the mid-infrared range which appears
 289

quite suddenly at Tv (3), indicating a mixing of the f-states with the con-
duction band. Neither the susceptibility, nor the resistivity (6), nor the Hall
constant (2) show any temperature dependence below Tv, i.e., the system
behaves as a fermi liquid with a characteristic energy scale TFL » Tv. The
magnetic moment of the rare earth ions is quenched in the ground state
by the f-d hybridization but the onset of the high-entropy phase cannot
be explained by the usual Anderson model in which the low- and high-
temperature scales are the same and the spin degeneracy is not expected
to be recovered below TFL. In these valence-change systems, however, the
f-moment is recovered for Tv « TFL.
The high- T phase that sets in at Tv is also anomalous. In doped systems,
the susceptibility data above Tv can be represented by a single universal
curve, provided one scales the data by an effective concentration of magnetic
f-ions, which is smaller than the nominal concentration of f-ions. The
functional form of the magnetic response agrees well with the "single-ion"
crystal field (CF) theory for all values of the field. The Yb ions seem to
be in the stable 3+ configuration with one f-hole and with the magnetic
moment close to the free ion value nv J(J + l)JLB = 4.53{lB (9L = 8/7 is
the Lande factor and J = 7/2 is the angular momentum of the 4f13 hole).
The dynamical susceptibility obtained from neutron scattering data (12) is
typical of isolated local moments, with well resolved CF excitations (13).
The resistivity of Yb 1 - x YxInCu4 alloys exhibits a weak maximum and
the thermopower has a minimum above 100 K but neither quantity shows
much structure at low temperatures, where the susceptibility drops below
the single-ion CF values. The discontinuity of the thermoelectric power at
the valence transition is a trivial consequence of the different thermoelectric
properties of the two phases: the thermopower of the valence-fluctuating
phase has an enhanced slope and grows rapidly up to Tv, where it suddenly
drops to values characteristic of the high-temperature phase. The resistivity
is not changed much by a magnetic field up to 30 T (14). In typical Kondo
systems, on the other hand, one expects a logarithmic behavior on the scale
T /TK and a large negative magnetoresistance. Here, despite the presence
of the well defined local moments, there are no Kondo-like anomalies. The
Hall constant of the x = 0 compound is large and negative in the high-
temperature phase, typical of a semi-metal (2); the optical conductivity (3)
shows a pronounced maximum of the optical spectral weight at a charge-
transfer peak near 1 eV and a strongly suppressed Drude peak.
The hydrostatic pressure and the magnetic field give rise, like the tem-
perature and the doping, to strong and often surprising effects. The critical
temperature decreases with pressure (14) but the data cannot be explained
with the Kondo volume collapse model (4). We mention also that doping
the Yb sites with Lu3+ ions (5) reduces Tv despite the fact that Lu has a
290

smaller ionic radius than Yb or Y; doping the In sites by smaller Ag ions

enhances Tv in YbInl-xAgxCu4 (8, 15) without changing appreciably the
lattice parameter. Thus, the effects of doping cannot be explained in terms
of a chemical pressure. An external magnetic field of a critical strength
Hc(T) destabilizes the low-temperature phase and induces a metamagnetic
transition which can be seen in the magnetoresistance and the magnetiza-
tion data (4). The experimental values of Hc(T = 0) and Tv(H = 0) are of
the same order of magnitude.

2. Theoretical description and the DMFT solution

A qualitative description of the properties described above is provided by

the Falicov-Kimball (FK) model (16) which takes into account the interac-
tion between a 2-fold degenerate conduction band and a lattice of Yb and Y
ions. Each lattice site can be occupied either by a Yb 2 +, Yb3+ or y3+ ion.
The Yb 2+ ion has a full f-shell and is non-magnetic, the Yb 3 + is magnetic
with one f-hole in a J=7/2 spin-orbit state, and the y3+ is non-magnetic,
with one additional hole with respect to the Yb 2 + ion. The number of y3+
ions is fixed in each alloy, while the concentration of Yb 3 + and Yb 2+ ions is
a thermodynamic variable. The f-holes are localized and the state of a given
Yb ion cannot change in time but the relative number of Yb 2 + and Yb 3 +
changes due to thermodynamic fluctuations. The conduction electrons can
hop between nearest-neighbor sites on the D-dimensional lattice, with a
hopping matrix -tij = -t* /2VD; we choose a scaling of the hopping matrix
that yields a nontrivial limit in infinite-dimensions (17). We assume that
the magnetic f -hole on Yb3+ and the spinless hole on y3+ interact with
the holes in the conduction band by a Coulomb repulsion Uf and Uy,
respectively. Averaging over all possible random distributions of Yb 2 + ,
Yb 3 + and y 3 + ions restores the translational invariance and leads to the
Falicov-Kimball model for the lattice of Yb-Y ions,
(1)
where
H~ = 2)-tij -poij)dladja, (2)
ij,a
Hf = "E}Ef - P)fi~/i77' (3)
i,77

(4)

and
(5)
 291

Spin-l/2 conduction holes are created or destroyed at site i by dJ<7 or di<7' the
8-fold degenerate localized f-holes are created or destroyed at site i by f itrt or
firt, and the spinless Y-hole is created or destroyed at site i by cJ or Ci. We
use (J" and TJ labels to denote the angular momentum state of the d- and f-
holes, respectively. The d-, f- and Y-number operators at each site are n~ =
2:<7 nd<7' nj = 2: rt njrt' and ny, respectively, and we have the local constraint
nj +nir ::; 1. The Y-doping reduces the number of f-holes in the conduction
band and provides additional Coulomb scattering for conduction-holes. For
a given concentration x of Y ions, the chemical potential JL is employed to
conserve the total number of remaining d- and f-particles, nd(T) + nf(T) =
ntot - x. In the presence of a magnetic field the magnetic degeneracy of the
f-holes is lifted and the Hamiltonian (1) is supplemented by a Zeeman term.
Using the basis that diagonalizes simultaneously the zero-field Hamiltonian
and the J;/2 component of the angular momentum operator, we can write,

Hz = gdJLBH L (J"dJ<7 dj<7 + gfJLBH L TJfitrtJi rt , (6)

i<7 irt

where gd and gf are the corresponding Lande g-factors. The numerical

calculations are performed for a hyper-cubic lattice with a Gaussian non-
interacting density of states p(c) = exp[-c 2/t*2]j(ftt*); and t* is taken
as the unit of energy (t* = 1). We consider only the homogeneous phase,
where all quantities are translationally invariant.
The DMFT reduces the problem on an infinite-dimensional lattice to
the problem of an atomic d-state coupled to an atomic f-state by the same
Coulomb interaction as on the lattice (18), and perturbed by an external
time-dependent field, ),( T, T/). The field is taken in such a way that the
local d-electron Green's function of the lattice coincides with the Green's
function of the atomic d-state, Gfoc(z) = G~t(z). The atomic problem is
solved by defining the generating functional (the partition function of the
FK atom) in the interaction representation (19),

(7)

where the statistical sum runs over all possible quantum states of the system
and depends on ),<7 ( T, T/) for T, T/ E (0, (3). The Hamiltonian of the FK atom,

Hat = -JL L dt d<7 + (Ef - JL) LfJfrl + (Ey - JLy)ctc

<7 rt
+ Uf Ldtd<7f~frt + Uy Ldtd<7ctc, (8)
<7rt <7rt
292

defines the time evolution of the operators, and the external field defines
the time-evolution operator for the state vectors,

(9)

In the presence of the magnetic field, we add to (8) a Zeeman term that is
obtained from (6) in an obvious way. The Hilbert space can be decomposed
into invariant subspaces with respect to nf and n c , and the matrix elements
in (7) can be calculated within the n f-invariant subspace by replacing
L;TJ fJfTJ in Hat by its eigenvalue (0 or 1) and setting n c = O. Within the
nc-invariant subspace we use nf = 0 and n c = 1. This gives,

(7/2)
where, Zf = L;TJ e-(3(E'1 -/1) and Zy = e-(3(EY-/1Y) are the partition func-
tions of the Yb 3 + and y3+ holes decoupled from the d-states, and ZO(/f-, A)
is the partition function of the Uf = Uy = 0 atomic d-state coupled to the
A-field only. We have,

(11)

where
(12)

and
(13)

To calculate Zo(/L, A) we introduce the d-electron propagator for this sim-

plified atomic problem,

(14)

which is determined by the equations of motion (EOM) and the boundary

condition GO(T, T') = GO(T + (3, T'). The EOM's reduce, in the Matsubara
representation, to a set of decoupled linear algebraical equations, such that
[GOn ]-1 = [gon]-1 - A~, where [gon]-1 = iW n + /L - a/LBgdH is the propag-
ator of the trivial Uf = Uy = A = 0 atomic d-state. Using Zo(/L, A) =
det I [G ]-11 and det I [go]-11 = 1 + exp (/L - a/LBgdH), we can write the
o
partition function of the Uf = Uy = 0, A # 0 atomic model as,

zg(/L, A) = (1 + e(3(/1-a!LI3gd H )) II(1- A~gon)' (15)

n
 293

The partition functions in the nf = 1 (ny = 0) and the ny = 1 (nf = 0)

subspaces are obtained from the nf = ny = 0 solution by replacing /L in
Zo (/L, A) by /L - Uf and /L - Uy, respectively.
The fully renormalized Green's function of the FK atom is, by definition,

IJ( ') 1 6Zat(/L, A)

(16)
G T,T =- Zat(/L,A) 6AIJ (T',T) '

and can be calculated using Eqs.(lO) and (14). In the Matsubara repres-
entation, where GOn = -6InZo/6A~, we obtain

IJ(') IJ(') Nf Ny (17)

Gat LWn = NoGo LWn + [Go(iwn)]-l_ Uf + [Go(iwn)]-l- Uy'

where No = ZO(/L, A)/Zat, Nf = ZfZo(/L-Uf,A)/Zat, and Ny = ZyZo(/L-

Uy, A)/ Zat are the respective average numbers of the Yb 2 +, Yb 3 + and y3+
ions in a Y-doped system. The concentration of Y-ions, Ny = x, is kept
constant at each temperature, and G~t is calculated using No = 1- Nf - x.
Since Go is the solution of the Uf = Uy = 0, Ai- 0 problem, the self energy
of the full atomic problem is given by the Dyson equation,

(18)

The DMFT solution for the FK lattice is obtained from the atomic solution
for A-field such that

G IJ (iwn )
at
= J
iWn + /L
p(c)
+ CJ/LBgdH- I;IJ(z) - c
de. (19)

The equations (17)-(19), together with the expressions (10) and (15) for the
partition function, can be solved by iteration. One starts from some trial
self energy and finds G~t using Eq.(19). Then, one finds Go
from Eq.(18),
finds Zat(/L, A) using (10) and (15), calculates G~t by functional derivatives,
recalculates I;IJ using (18), and continues until the fixed point is reached.
Once the numbers No and N f are obtained we can iterate (17), (18) and
(19) on the real axis and find the retarded quantities. In what follows, we
use the DMFT to calculate the thermodynamic and transport properties
of the model corrsponding to the Yb 1 - xYxInCu4 alloys.

3. Numerical results

The calculations are performed assuming that doping by y3+ ions removes
holes from the conduction band. In an undoped sample the total number of
conduction holes and the holes on the Yb 3+ ions is nd+nf = 1.5, while for a
concentration x ofY ions we assume nd+nf = 1.5-x. The parameter space
294

of the model is very large and a quantitative comparison would require a

fine-tuning of the parameters. Here, we only show the results for static
correlation functions obtained for Uf = Uy = 2 t*, and Ef = -0.6 t*.

0.4

-- x=0.2
0.2 ---- x=0.15
----- x=O.l
---. x=0.05
----. x=O
(a)

005 01 015 02
Temperature [t"]

10
- - x=0.2
_....-.-- x=0.15
x=O.l
x=0.05
---- x=O

(b)

O'-L-------~------------'
o 0.05 0.1
Teperature [t·]

Figure 2. Panel (a) shows localized electron filling for eightfold degenerate doped
Falicov-Kimball model with UJ = Uy = 2t*, EJ = -O.6t*, and various doping levels x.
Panel (b) shows spin susceptibility normalized to the nominal concentration of Yb ions
in a Yb 1 - xY xlnCu4 alloy.

Panel (a) of Fig.2 shows the effect of temperature and doping on the
average concentration of Yb3+ ions. Panel (b) shows the f-electron contri-
bution to the spin susceptibility, which vanishes below Tv, and is Curie-like
for T » Tv. Above the transition temperature, defined by the inflection
point of the susceptibility curves, the concentration of Yb 3 + ions becomes
significant. The effective Curie constant decreases with doping.
Panel (a) of Fig. 3 shows the result for the DC resistivity, Pdc(T),
normalized by Po = 2fLdad - 2 /e 2 Jr, where a is the lattice spacing and d
is the dimensionality of the system (po;::::; 2.3 X 10- 4 Dcm in d = 3 with
 295

2 I
• x=O.2
•
• x=O.15
A x=O.1
•
A
l' x=O.05 • l'

• X=O
• •
l'
0:
:;:;
t 'l
• •
• • • • l'
Q. A
•
•
• •
l'

A
A • (a)
•'"
A

l' l'

0
0 001
Temperature [t·]

1.5 ,-------~-----_=_"
,./'
.//-
0.5 //////

-0.5

i=' -15
Cii
- x=o().2
-2.5 x=o().15
----- x=O.1
x=O.05
-35 ---- x=O (b)

-4.5 L -_ _~_ _~ ~_-----.J

o 0.1 0.2
Temperature [t·)

Figure 3. Panel (a) shows low-temeperature DC resistivity of the Falicov-Kimball

model for the same parameters as in Fig.(2). Panel (b) shows thermopower on a largere
temperature scale.

a ::::::; 3 x 10- 8 cm). In an undoped sample we have Pdc(O) = 0 and the

resistivity becomes significant only around Tv, where the concentration
of Yb 3 + ions is large. In doped (x > 0) samples, Pdc(O) : : : ; x, but the
relative importance of the doping diminishes for T ;::: Tv. At high tem-
peratures, the resistivity of all samples has a maximum (not shown here).
The thermopower results are shown in panel (b), using a 10 times larger
temperature scale than in panel (a). The data reveal the minimum (at about
the same temperature as the resisitivty maximum), which is followed by a
sign-change. The theoretical results shown in Figs. 2 and 3 have a number
of qualitative features that one finds in the experimental data. A detailed
comparison will be published elsewhere.
296

Acknowledgements

We acknowledge useful discussions with C. Geibel and J. Sarrao. We ac-

knowledge the support by the National Science Foundation under grant
DMR-0210717. V.Z. acknowledges the support by the Swiss National Sci-
ence Foundation, grant no. 7KRPJ65554.

References

1. 1. FeIner and 1. Novik, Phys. Rev. B 33, 617 (1986).

2. E. Figueroa, J. M. Lawrence, J. Sarrao and Z. Fisk, M. F. Hundley and J. D.
Thompson, Solid State Commun. 106, 347 (1998).
3. S. R. Garner, J. N. Hancock, Y.W. Rodriguez, Z. Schlesinger, B. Bucher, Z. Fisk
and J. L. Sarrao, Phys. Rev. B 63, R4778 (2000).
4. J. L. Sarrao, Physica B 259 & 261, 129 (1999).
5. W. Zhang, N. Sato, K. Yoshimura, A. Mitsuda, T. Goto and K. Kosuge, Phys. Rev.
B 66, 024112-1 (2002).
6. M. Ocko and J. L. Sarrao, Physica B 312 & 313, 341 (2002).
7. C. Dallera, M. Grioni, A. Shukla, G. Yanko, J. L. Sarrao, J. P. Rueff and D. L. Cox,
Phys. Rev. Lett. 88, 196403 (2002).
8. A. L. Cornelius, J. M. Lawrence, J. Sarrao, Z. Fisk, JV1. F. Hundley, G. H. Kwei, J.
D. Thompson, C. H. Booth and F. Bridges, Phys. Rev. B 56, 7993 (1997).
9. J. M. Lawrence, S. JV1. Shapiro, J. L. Sarrao and Z. Fisk, Phys. Rev. 55, 14467
(1997)
10. A. Mitsuda, T. Goto, K. Yoshimura, W. Zhang, N. Sato, K. Kosige and H. Wada,
Phys. Rev. Lett. 88, 137204 (2002).
11. C. Rettori, S. B. Oseroff, D. Rao, P. G. Pagliuso, G. E. Barberis, J. Sarrao, Z. Fisk
and M. F. Hundley, Phys. Rev. B 55, 1016 (1997).
12. E. A. Goremychkin and R. Osborn, Phys. Rev. B 47,14280 (1993).
13. A. Murani, D. Richard and R. Bewley, Physica B 312 & 313, 346 (2002).
14. C. D. Immer, J. Sarrao, Z. Fisk, A. Lacerda, C. Mielke and J. D. Thompson, Phys.
Rev. B 56, 71 (1997).
15. J. M. Lawrence, R. Osborn, J. L. Sarrao and Z. Fisk, Phys. Rev. 59, 1134 (1999).
16. L. M. Falicov and J. C. Kimball, Phys. Rev. Lett. 22, 997 (1969).
17. W. Metzner and D. Vollhardt, Phys. Rev. Lett. 62, 324 (1989).
18. U. Brandt and C. Mielsch, Z. Phys. B 75, 365 (1989).
19. L.P. Kadanoff and G. Baym, Quantum Statistical Physics (W. A. Benjamin, Menlo
Park, CA, 1962).
20. J. Freericks and V. Zlatic, Phys. Rev. B 58, 322 (1998); V. Zlatic and J. Freericks,
in Pmc. NATO ARW Conference, Bled 2000, edited by P. Prelovsek, S. Sarkar, J.
Bonca (North-Holland, Amsterdam, 2001).
NEUTRON SPECTROSCOPY OF VALENCE FLUCTUATION
COMPOUNDS OF CERIUM AND YTTERBIUM.

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

A. P. MURANI
Institut Laue Langevin, 6 Rue Jules Horowitz, 38042 Grenoble.

Abstract. At low temperatures, the magnetic response of Ce and Yb-based valence

fluctuation systems consists of a broad inelastic lorentzian distribution centered on a
characteristic energy T K which varies from ~ 10 meV e.g. YbAgCu4 (1), CeIn1.5Sn1.5
(2) to ~ 35 - 55 meV, e.g. CeSna (3), YbAb (4), YbInAu2 (5), CePd a (6,7), going up
to ~ 120 - 140 meV, e.g. CeRh 2 (8), CeNi 2 (9), and ~ 170 - 180 meV as in a-Ce (10)
and YbAb (11). In strongly hybridized Ce systems such as CeFe2 (12), CeRu2 (13) very
broad magnetic responses with characteristic energies of ~ 500 meV have been observed.
In all these systems the variation of the spectral intensity with Q follows, or is consistent
with, the single ion form factors of Ce H or Yb H .

,--+
Measurements of excitations to the upper spin-orbit state in Ce show a dramatic shift
at the a transition from ~ 260 meV in the, phase to around ~ 500 meV in the
a phase (10, 14). A similar enhancement of the spin-orbit excitation energy by about ~
50 ± 30 meV is observed in the compound YbInCu4 as it undergoes a similar transition
from a low-T K (~ 2 meV) heavy fermion state to one with a significantly higher T K (~
32 meV) (15). In CePd a with a ground state TK of ~ 55 meV, the spin-orbit excitation
is observed at ~ 370 meV representing an enhancement of ~ 100 meV (16). Also, a
progressive evolution of the spin-orbit excitation energy from ~ 275 meV to ~ 340 meV
is observed in CeIn:J-xSnx as x is increased form 0 to 3 (2). Remarkably, in all these
systems the enhancement of the spin-orbit energy relative to the free-ion value is about
twice the characteristic energy TK of the ground state.

Key words: Neutron Spectroscopy, Paramagnetic Scattering, Valence Fluctuation Sys-

tems, Heavy Fermions, Cerium Systems, Ytterbium Systems, Kondo Effect

1. Introduction

Metals, alloys and intermetallic compounds containing the end members

of the 4f electron series of the periodic table, viz. Ce and Vb, very often
show interesting physical properties that deviate significantly from those

297
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 297–305.
© 2003 Kluwer Academic Publishers
298

expected for a localized 4f ion having a Russel-Saunders ground state and

interacting lightly with the surroundings (crystalline field and inter-ion
couplings). The latter is the most common feature amongst the rare earth
series, except for a number of deviant Ce and Yb systems as well as a few
systems containing Sm, Eu and Tm.

Two principal characteristic properties of the 'anomalous' Ce and Yb

based systems are the enhanced magnitude of the linear coefficient of the
specific heat, and the constant low temperature (T -----+ 0 K) susceptibility
X(O). Intermetallic compounds in which the Sommerfeld constant { ~ 100
mJ mole- 1 K- 2) are commonly classified as heavy fermion systems, while
those with { :s 50 mJ mole- 1 K- 2 are called valence fluctuation systems.
The low temperature susceptibility X(O) and the linear specific heat coef-
ficient { are related via a common characteristic energy parameter T K
as expressed by the relations X(O) = p,2(nf) /3TK and { = ]f2k 2 (nf) /3TK'
where (nf) (<::: 1) is the occupancy of the 4f state. For heavy fermion systems
the occupancy factor (nf) is close to unity while in valence fluctuation
systems (nf) can deviate from unity by a significant amount. Some of the
early theories have shown that the Wilson ratio R = ]f2k2X(0)/p.2{ is,
ideally, close to unity for heavy fermion/valence fluctuation systems (17).

There is no fundamental difference between heavy fermions and valence

fluctuation systems except in the degree of hybridization, and the relative
strength of inter-site correlations. In a heavy fermion system hybridiza-
tion is relatively weak and spin-spin correlations are usually quite strong,
while hybridisation is much stronger in valence fluctuation systems and
inter-site correlations almost negligible. If a heavy fermion system does
not order magnetically or become superconducting at low temperatures,
its bulk magnetic susceptibility tends to a constant value X(O) as T -----+
o K. In general, hybridization acts to counter the tendency for a system
to order magnetically. Increasing hybridization thus favors paramagnetic
states with increasing T K and progressively decreasing susceptibility X(O)
and specific heat coefficient {. In the following we review and discuss results
of neutron scattering measurements on these relatively strongly hybridized
(valence fluctuation) systems in which the inter-site magnetic correlations
are relatively weak or negligible compared with T K.

2. Formalism

In the following we show that if a characteristic temperature T K is

defined via the low temperature susceptibility relation X(O) = p,2/3TK
then T K can be identified with the energy WQ representing the centroid
 299

of the broad lorentzian spectral distribution observed in neutron scattering

measurements on valence fluctuation systems.
Following Marshall and Lowde (18) we express XI/(w) in terms of a lorent-
zian spectral function as,

(1)

where A = 7fX(O) = 7f2JL2/6r, and r is assumed Q-independent.

Using the Kramers-Kronig relation

X(O) = ~]oo XI/(w)dw = ~ roo XI/(w)dw (2)

7f -00 w 7f Jo w

and substituting equation 1, we obtain

JL2 roo 1 JL2 7f 1 Wo

X(O) = 3 Jo r2 + (w - wo)2dw = 3r ["2 - tan- ( r)] (3)

Thus, for r /wo <::::1, (generally, r /wo lies within the range 1/2 to 2/3) we
obtain,
(4)

hence, TK ':::::' woo However, for a quasi-elastic spectral distribution we obtain

7f JL2
X(O) = - - (5)
23r
hence, T K = ~r. Note that the relation (5) is implicit via the constant
A in the definition of XI/(w), in relation (1). Furthermore it can be easily
shown that relation (1) satisfies the moment sum rule

00 S(w)dw 2/7f ]00 [ 1 2 2

]-00
= 22
9 JLB -00 1 - exp -
(W )]XI/(w)dw = -JLo
kT 3
(6)

where JLo = ---.l!:..-, and g = 2. Unlike the susceptibility integral, the integral
91LI3
of S (w), of course, diverges logarithmically. Hence a cut-off needs to be
applied around We rv 10 - 12wo, or rv lOr for a quasi-elastic distribution.

3. Experimental Results

In heavy fermion systems owing to the relatively weak hybridization one

can still observe crystal field excitations which are, nevertheless, energy
broadened. As the hybridization increases the crystal field states broaden
further and eventually merge into a single broad distribution of lorentzian
300

1
2 Celn 3
·c
::J

...m
.D 0

8 1
-
cD
(/)
0
Celn
1.5
Sn
1.5

4>

-40 -30 -20 -10 0 10 20 30 40

Energy Transfer (meV)
Figure 1. Paramagnetic spectral response from CeIna at 20 K and CeIn1.5Sn1.5 at 5
K. The continuous curves represent fits to the data using two lorentzian components for
CeIn;J corresponding to the r7 ground state and r s excited state. For CeIn1.5Sn1.5 the
curve represents a broad lorentzian distribution centered on ~ 10 meV

shape. This phenomenon is well illustrated by the system CeIn3-xSnx which

evolves continuously from the heavy fermion state in CeIn3 to a valence
fluctuation system in CeSn3. The data in fig.1 show that the centroid of
the broad distribution observed in CeIn1.5Sn1.5 (2) has roughly the same
energy as the overall crystal field splitting in the pure CeIn3 sample. The
broad spectrum observed for CeIn1.5Sn1.5 could also be interpreted in terms
of broadened crystal field states as proposed by Zwicknagel, Zevin and
Fulde (19). For example, the observed broad spectral distribution with
a characteristic energy of rv 9 meV in YbAgCu4 has been re-interpreted
assuming the crystal field scheme observed in iso-structural YbAuCu4 and a
Kondo temperature To of rv 6 meV (23). However, if one accepts to define a
characteristic temperature via the relation X(O) = /12/3T K then, as shown
above, the centroid of the broad lorentzian spectral distribution represents
T K directly. The question of whether the broad lorentzian distribution can
be decomposed into broadened crystal field states is then less relevant.

A classic example of a valence fluctuation system is the compound CePd 3

which shows the characteristic lorentzian inelastic spectral distribution,
figure 2 (16). The measurements performed with high energy neutrons (E i
= 900 meV) on the HET spectrometer at ISIS permits a substantial energy
range to be covered with, however, a moderately poor energy resolution.
The non-magnetic scattering has been subtracted out with reference to the
data taken on the iso-structural compound YPd 3. The continuous curve
 301

1.2

~
c 0.8
::J
.ci
l-
co

r
8 0.4 CePd 3
T = 12 K
~
if)

-0.2
-200 0 200 400 600 80
Energy Transfer (meV)
Figure 2. Paramagnetic spectral response from CePd;J at T = 12 K measured with
high energy neutrons. The continuous curve represents a fit to the data using the Kur-
amoto-Muller Hartmann (KMH) spectral form (20) with a characteristic temperature of
~ 55 meV. The inset shows the additional scattering relative to the fitted curve, on an
expanded vertical scale and re-binned on the energy scale. It represents the spin-orbit
excitation at ~ 370 meV.

shows a fit to the KMH function which is similar to the lorentzian fit over
a very wide energy range. In the region above rv 200 meV one observes
some additional scattering, lying above the fitted curve, which is shown in
the inset on an expanded vertical scale. It corresponds to the spin-orbit
excitation which is centered on rv 370 meV, representing an enhancement
over the free ion value of rv 100 meV.

The r ----+ 0: transition in Ce which occurs on cooling below rv 110 K

(21) provides one of the best illustrations of the changes in the magnetic
spectral response induced via a change in the hybridization with temper-
ature. A similar transition occurs in the system YbInCu4 below 42 K (22).
In their high temperature phases both can be described as heavy fermion
systems showing a central quasi-elastic distribution together with broad
but well defined crystal field excitations. Both systems transform abruptly
below their respective transition temperatures and show valence fluctuation
character, in particular broad, lorentzian spectral distributions centered on
WQ rv 32 meV in YbInCu4 (15) and rv 170 meV in o:-Ce (10).

Following Johansson's interpretation of the r ----+ 0: transition in Ce m

terms of the transformation of the localized 4f state into an itinerant 4f
302

E,= 150 meV

6
,,
I

,,
,
,
,

1:-
'iii
c
4
~ ? o 2 4 6 8 10
Q)
-+-'
c
2
.?,
~
o ~rr2dJ~L...L.......L.-L...J.........L~~~bJ
-50 o 50 100 150
Energy Transfer (meV)
Figure 3. Paramagnetic spectral response from YblnCu4 at 35 K. The continuous
and the dashed curves, which are practically indistinguishable, represent fits to the data
above 10 meV to a lorentzian distribution centered on ~ 32 meV and the KMH spectral
function. The inset shows the variation of the intensity around the peak as a function of
Q. The thick solid line represents a fit to the Yb H form factor.

band (24), as well as the observations of 'coherence' effects in a number

of physical properties including heavy band masses on the Fermi surface
of CeSn3 (25) and many similar systems, a number of band theoretical
calculations (LDA+U) have been performed assuming, at the outset, 4f
band states. One such calculation (26) for CeFe2' shows clearly that while
both the spin and orbital angular momenta are quenched substantially
due to band formation, the reduction is much greater in proportion for the
orbital moment with dramatic consequences for the magnetic form factor.
While the observations in U-based systems as for example UFe2 (27) appear
consistent with theory (28), the data for a-Ce, YbInCu4 shown in figure 3,
and many others indicate, on the contrary, a remarkably single ion-like form
factor dependence. In the case of CeFe2 inelastic scattering measurements
(12) are not of adequate statistical quality to permit an explicit verification
of the form factor dependence, but the data can be fitted over a wide angular
range 8 to the single ion spectral forms (lorentzian and KMR distributions)
assuming the Ce3+ form factor variation of intensity with w, hence Q.

Finally, in figure 4 we present data showing the spin-orbit excitations in

r and a Ce measured at very large Q-vectors, using the Be-filter technique
on the IN1 spectrometer at ILL (14), They substantiate our earlier obser-
vations of the same at lower Q's using the RET time-of-flight spectrometer
 303

~
c
600 400
::J

....
.ri
~ 400 200
.c
·m
c
C
Q)
200 o

0 ....L...1. '--'-' .........

a 100 200 300 400 500 600 700

Energy (meV)
Figure 4. Paramagnetic scattering from ,-Ce showing the spin-orbit excitation at ~
260 meV which transforms to a broad excitation centered on ~ 500 meV in the a-phase.

at ISIS (10). The well defined excitation seen around 260 meV in the I
phase is found to have broadened and shifted to rv 500 meV in the 0:
phase, as found earlier. However, the most remarkable aspect of these later
measurements is that, with the Be-filter technique, at energy transfers of rv
500 meV the momentum transfers Q of rv 15 A-I are attained. Hence, the
presence of measurable intensity in the spin-orbit excitations at such large
Q-vectors further testifies to the single-ion character of the 4f state in the
0: phase. If, the enhancement of the excitation energy by about rv 250 meV
can be interpreted as representing the characteristic energy of the upper
2F7/2 spin-orbit state, then T K 7 /2 is roughly rv 1.5 times the characteristic
energy T K 5 / 2 (rv 170 meV) of the 2F5/2 ground state. As mentioned earlier
the data in figure 2 for CePd 3 indicate T K 7 /2 c:::: 2 x T K 5 / 2 . A very similar
ratio for the excited to ground state characteristic energies is observed for
the series of compounds CeIn3-xSnx over the full range 0 ::; x ::; 3.

4. Conclusions

We have shown that a lorentzian spectral form centered on an energy WQ

gives an adequate description of the observed spectra in valence fluctuation
systems where WQ can be identified with the characteristic energy or
temperature T K defined via the relation, X(O) = /12/3T K . The question
whether the characteristic temperature represents the Kondo temperature
or corresponds, in reality, to some spin fluctuation or moment fluctuation
temperature needs still to be addressed. However, what is abundantly clear
from the experimental observations is that the magnetic response shows
304

the localised, single ion, character via the Q dependence of the spectral
intensity as well as its single-ion spectral form.

Observations of the energy-enhanced spin-orbit excitations in both Ce

and Yb-based valence fluctuation systems are rather intriguing. A simple
interpretation would suggest that the characteristic energy of the upper
spin-orbit state is about r-v 1.5 to 2 times that of the ground state. In
view of the large range in the TK'S of the ground states from r-v 10 to r-v
170 meV (for systems on which these observations are made) such a small
and relatively constant ratio would appear surprising on the basis of the
Anderson model which indicates an exponential variation of TK'S with the
various physical parameters including the degeneracy as well as the energy
of a state relative to the Fermi level (29). Perhaps other hybridization
processes such as with the ligand electrons may also be operative in giving
rise to the observed phenomena (30). Clearly much further progress is still
to be achieved to reach a coherent understanding of these exciting systems.

References

1. A. Severing, A.P. Murani, J.D. Thompson, Z. Fisk, C.-K, Loong, Phys. Rev. B 41,
1739 (1990).
2. A.P. Murani, A.D. Taylor, R.Osborn and Z.A.Bowden, Phys. Rev. B48, 10606
(1993).
3. A.P. Murani, J. Phys. C 33 6359, (1983); Phys. Rev. B 28, 2308 (1983).
4. A.P. Murani, Phys. Rev. B 509882, (1994).
5. A.P. Murani and J. Pierre, Physica B 206-207,329 (1995).
6. R.M. Galera, A.P. Murani, J. Pierre and K.R.A. Ziebeck, J. Magn. Magn. Mater.
63-64, 594 (1987).
7. A.P. Murani, A. Severing and W.G. Marshall, Phys. Rev. B 53, 2641 (1996).
8. A.P. Murani and RS. Eccleston, Physica B 230-232, 126 (1997).
9. A.P. Murani and R.S. Eccleston, Phys. Rev. B 53, 48 (1996).
10. A.P. Murani, Z.A. Bowden, A.D. Taylor, R. Osborn and W.G. Marshall, Phys. Rev.
B 48, 13981 (1993).
11. A.P. Murani, Z.A.Bowden, A.D. Taylor and R Osborn, Phil. Mag. B65, 3092 (1992).
12. A.P. Murani, B. Ouladdif and R.S. Eccleston, Physica B 259-261, 1167 (1999).
13. A.P. Murani and RS. Eccleston, Physica B 241-243, 850 (1998).
14. A.P. Murani, J. Reske, A.S. Ivanov and P. Palleau, Phys. Rev. B 65, 094416 (2002).
15. A.P. Murani D. Richard and R Bewley, Physica B 312-313, 346 (2002).
16. A.P. Murani, R.Raphel, Z.A.Bowden and R.S.Eccleston, Phys.Rev. B53, 8188
(1996).
17. H. Lustfeld and A. Bringer, Solid State Comm. 28, 119 (1978).
18. W. Marshall and RD. Lowde, Rep. Prog. Phys. 31, 705 (1968).
19. G. Zwicknagl, V. Zevin and P.Fulde, Physica B 163, 577 (1990).
20. Y. Kuramoto and E. Muller-Hartmann, J. Magn. Magn. Mater. 52, 122 (1985).
21. A.W. Lawson and T.-Y. Tang, Phys. Rev. 76,301 (1949).
22. 1. FeIner et al. Phys. Rev. B 35, 6956 (1987).
23. G. Polatzek and P. Bonville, Z. Phys. B 88, 189 (1992).
 305

24. B. Johansson, Phil. Mag. 30, 469 (1974).

25. W.R. Johanson, G.W. Crabtree, A.S. Edelstein and O.D. McMasters, Phys. Rev.
Lett. 46, 504 (1981).
26. O. Eriksson, L. Nordstrom, M.S.S. Brooks and B. Johansson, Phys. Rev. Lett. 60
2523 (1988).
27. M. Wulff, G.H. Lander, B. Lebech and A. Delapalme, Phys. Rev. B39, 4719 (1989).
28. M.S.S. Brooks, O. Eriksson, B. Johansson, J.J.M. Franse and P.H. Frings, J. Phys.
F 18, L:3:3 (1988).
29. N.E. Bickers, D.L. Cox and J. W. Wilkins, Phys. Rev. B 36, 2036 (1987).
:30. Q.G. Sheng and B.R. Cooper, Phil. Mag. Lett. 72, 123 (1995).
GENERALIZATIONS OF DMFT, CPA AND NCA

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

H. KEITER, K. BAUMGARTNER and DIRK OTTO

Institute of Physics, Dortmund University, D-44221 Dortmund

Abstract. Local and non-local generalizations of dynamical mean-field theory (DMFT)

are presented. They come from a self-avoiding walk in a hypercubic lattice, in which
the principle of exclusion and inclusion is used. The coherent potential approximation
(CPA) can be treated in a similar way. CPA and DMFT can be mapped onto each
other. The numerical corrections to the DMFT are relatively weak, with the non-crossing
approximation (NCA) used for the single impurity part. NCA-generalizations similar to
those of the WCilfle group are set up with direct self-consistent perturbation theory.

Key words: Dynamical Mean-Field Theory, Coherent Potential Approximation, Non

Crossing Approximation, Self-Avoiding Walk, Generating Functional, Scattering Matrices,
Anderson Lattice Model, Anderson (Impurity) Model

1. Self-avoiding loop and DMFT generalizations

The competition between itinerant and local moment behavior of electrons

determined 40 years of research on strongly correlated electron systems,
starting from a magnetic impurity which could form a local moment (1)
and interact via hybridization with a conduction band. The strength of
the latter interaction (hybridization V or a corresponding hopping term)
competes with the strong on-site part of the screened Coulomb interaction
U. Usually U is the largest energy in the system, so perturbation expansions
in the hybridization seemed to be an adequate tool for the impurity case
(2) and for the lattice (3). For reviews see (4, 5, 6, 7). In general, the
following picture emerges: Physical processes contributing to the (grand
canonical) partition function start from a lattice site" 0", perform all "on-
site" interactions there, then migrate to a (nearest neighbor) site Xl with
on-site interactions there, etc. and finally return to the initial lattice site.
In such a process a certain site may be visited several times (3). One of the

307
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 307–315.
© 2003 Kluwer Academic Publishers
308

Hamiltonians which give rise to this behavior is the Anderson lattice model

HpAM L txy cl,m Cy,m +L E j f1,mfx,m + V L (cx,mfl,m + h.c.) + Hu

xY,m x,m x,m
Hu L uf1,mfx,mf~,m,jx,ml (1)
x,m<m'

N-fold degeneracy of the f- and band-electrons, dispersionless hybridiza-

tion, discrete sums on the (hypercubic) lattice sites x in d dimensions are
assumed. U will be infinite in the following, and txy = t/Vd is the scaling
suggested in (8) for nearest neighbor hopping. With Ek = l:x tox exp ikx =
~t l:J=lcos kj one obtains for the unperturbed propagators
eik(x-y) 1 {':xJ (2 1
gxy = ( z _ Ek); goo = iJ dseiSZ(Jo(y dt))d = ~(1- v2dtglQ(z))
o
(2)
Here, the average denotes the sum on k, normalized by the number of
lattice sites. Jo is the standard Bessel function, the lattice constant is 1.
The unperturbed propagators are needed to set up the self-avoiding walk
through the lattice:

+L L
00

GOy(z) = gOy(z) gOxl (z)T(Z)gxlX2(Z)'" T(z)gxny(z) (3)

n=l Xl,···,X n

The prime on the sum indicates, that all sites have to be pairwise different
from each other and (up to y) from the initial site. The scattering matrix
T(z) = V 2 Gj(z) is related to the local f-Green's function to be calculated
from a single impurity theory. Setting y = 0 one obtains a self-avoiding loop
(SAL). Corrections to the SAL especially by multiple visits at a site vanish
for large spatial dimensions (9). At finite dimensions the SAL is known from
the XNCA (10). In (11) some of its mathematical properties were studied.
Here we use a self-avoiding walk to derive DMFT and generalizations of it.
Defining a self-energy via

60 y
GOy(z) = T(z)
v
+ ""'"
~ gox(z)T(z)Gxy(z) + ~ ~ox(z)Gxy(z)
v ""'" v
(4)
x x

In order to find the first contribution to ~ox, start from site 0 and go in
a first step to x. But then in going to site y one must exclude the return
to site O. This correlation in the walk generates the self-energy. For, if one
includes the site 0 into the walk from x to y, one needs a subtraction of the
term which contains site O. In this term we have a walk from x to 0 to y.
Correlations on this walk arise, because in the first part the same site may
 309

•o
:- -- - - --- - - -- ~

•o : 2

Figure 1. Principle of exclusion and inclusion for GOy (z). If the correlation line is left
off, one obtains I:~lJ and a subtraction with 4 correlation lines. If these are left off, the
first non-local contribution to I:~2J is obtained and another subtraction etc ..

be visited as in the second. It is easier to see this from the graphics in fig. 1.
Anyway, the first contribution to the self-energy I;W(z) = -6xy T(z)G oo (z),
if it is inserted into the eq. 4 which is multiplied from the right by g gives
the self-consistency equation of DMFT

Goo(z) = ( gk(Z) ) (5)

1 - T(z)(gdz) - Goo(z))

with gk(Z) = (z - Ek)-l. The second approximation to the self-energy

contains all second nearest neighbor contributions and has a diagonal as
well as a non-diagonal term

Here Gxy(z) = Gxy(z) for x oj y and 0 otherwise. The corrections to the

DMFT are displayed in fig. 2.

2. Equivalence of DMFT and CPA

From the matrix equations of standard many-body theory, involving the full
single p~rticle Green's function G, the unperturbed one g, the scattering
matrix T and the self-energy I;

G = g + gTg = g + gI;G (7)

one obtains
(8)
We next cast the DMFT equation (5) into a similar form by multiplying
eq. (5) with 1 + T(z)Goo(z). This yields

g(z) _ -
Goo(z)(l + T(z)Goo(z)) = (1 _ g(z)T(z) ) = goo(z - I;(z)) = G(z) (9)
l+T(z)Goo(z)
310

0.45 r--------,--------r--------.--------,

-- a=2.0
---- a=2.2
--- a=2.4
0.4 --- a=2.6

o
&. 0.35

0.3

0.25 '-----~-----'--~-----'---~-----'---~------'
-0.04 -0.02 o 0.02 0.04
z

Figure 2. Conduction band density of states with eq. (6). T(z) is approximated by
two Lorentzians T(z) = 1V12C_~~;ibJ + z~bib)' Parameters are 1V1 2 = 0.4, Ej = -2,
aj = b = 0.2, bj = 1.

where I;(z) = T(z)(1 + T(z)GoO(z))-l was introduced. This allows us to

write the DMFT eq. (5) in the equivalent form

G(z) = Goo(z) + Goo(z)T(z)Goo(z) = Goo(z) + Goo(z)I;(z)G(z) (10)

In contrast to the general matrix equations (7) of many-body theory the

eq. (5) or the equivalent form eq. (10) are scalar equations containing only
the diagonal elements of all quantities.
The equation for the single site CPA
-----c;Vc-;-----conf

I;(z) = --~-,-,on---;-f- - - - (11)

1- G(z) (V - I;(z))
contains two configuration averages over the random potentials V and over
~ """":::;"'"conf _
G(z). Then G enters into all the equations above, replacing G. Within the
DMFT the scattering matrix T(z) is calculated via the effective propagator
Goo(z), which in the CPA describes the influence of other scattering centers.
Thus
V conf

T(z) = -1------,G-----,(Z----,)V- (12)

oO
 311

Given the local quantity T(z) one obtains the self-energy ~(z) via

T(z)
~(z) = 1 + T(z)Goo(z)
----------V~---------'conf

(13)
1 - V(Goo(z) + Goo(z)T(z)Goo(z)) + T(z)Goo(z)
~ -~-conf

Using eq. (10) with G(z) ----+ G(z) we obtain the CPA equation (11). In
view of this, we may interprete the CPA as a special case of the DMFT
for disordered systems. This completes the equivalence of both approaches.
conf
We note in passing, that also the non-local scattering matrix T of the
CPA can be obtained in a closed form from eq. (10)

Tconf(z) = T(z) (14)

1 - T(z)(g - Goo(z))

3. Improvements of the NCA

For the numerical results obtained in section 1 we used either model scatter-
ing matrices T(z) (see fig. 2) or the NCA. The latter, however, is restricted
to temperatures above the single impurity Kondo-temperature. Then the
corrections to the DMFT results by ~1~(z) from eq. (6) are only a few
percent. This is why one has to use another single impurity input. Besides
numerical renormalization group methods or density matrix renormaliza-
tion group, NCA-extensions have been set up. The extension used by the
Wolfie group, the CTMA, can be found in (13), with earlier references
therein. Their derivation via auxiliary bosons can be avoided, if one uses
direct perturbation methods reviewed in (4, 5, 6, 7). In the generating
functionals used before in the direct scheme one can also vary with respect
to the conduction electrons (i.e. the lines crossing the circles in fig. 3). This
can be easily be done, using the formalism in the last appendix of (4). The
first diagram for the generating functional in fig. 3 then reads

(3 is the inverse temperature, the contour integral surrounds all singularities

of the integrand in a counter-clockwise direction, W n = (2n+ 1)71-j (3, Po(z) =
(z - ~o(z) )-1, Pa (z) = (z - f{ - ~a(z) )-1 are the auxiliary propagators for
the empty f-level (dotted lines in the diagrams) and the occupied one (full
lines on the circumference in the diagrams of fig. 3; both are related to their
self-energies, of which the lowest orders are depicted in figs. 4 and 5. There
312

is energy- and spin-conservation at every vertex. Varying 2,6<I> with respect

to the bare band-electron Green's function F(iw n ) (this amounts to cutting
lines crossing the circles in the diagrams for the generating functional) a
formal self-energy ~c,a(iw n ) is obtained, which is related to the full f-Green-
function G f(T fJ (iw n ) as follows:

ZfGf(TfJ (iwn) = ~c,a(iwn) (16)

Here Zf is the full partition function divided by the band part. In

(17)

A a (w) denotes the density of states, which is needed in the diagrammatic

contributions for ~o(z), ~a(z) and ~c,a(z). The rules are similar to the ones
in (4, 5, 7) with the following changes: The complex energy z enters into the
lowest vertex in the diagrams of figs. 4 and 5. A curved line entering a vertex
carries an energy wand a spin tJ and stands for A a (w )f (w ), where f (w) is
Fermi's function. A curved line leaving a vertex stands for A a (w)(I- f(w)).
There is a factor (-1 )cr, where cr is the number of crossings of curved lines
(= 3 for the first diagram in fig. 4). The f-level (straight lines) on the left of
the diagrams in figs. 4 and 5 is alternatively occupied (lines with an arrow)
or empty (dotted lines). It carries the auxiliary propagators. The essentials
can be seen from the analytic expression for the first diagram in fig. 4.

~6a)(z) = L V 4 JdWIdw2dw3f(WI)f(W2)(1 - f(W3))Pa (Z + wd x (18)

a
xPo(z + WI - w3)TI (z - W3, WI, W2)PO(Z + W2 - W3)Pa (Z + W2)
The one-dimensional integral equation for T I is depicted in fig. 6 and reads

The scattering matrix T I can also be used for the self energy ~~a), as shown
in fig. 6 and for the first part of the conduction band self energy.

~~aJ(
,
J
iw n ) = - V 2 dz. e-(3zjdw2dW3Aa(W2)Aa(W3)f(w2)(1 - f(W3) )Po(z)
27rz
Pa(z + iwn)Po(z + iW n - w3)TI (z + W3, iw n ,W2)PO(Z + W2 - W3)Pa (Z + W2)
(20)
The case with two spin sums is more complicated. In the 3rd diagram
of fig. 4 the curved line entering the lowest vertex carries the energy iW n
and the dashed dotted box is replaced by (L a' T 2,a' (z + iW n + W4, WI, W2)) +
 313

Figure 3. Generating functional for improvements of the NCA. The diagrams have the
inverse number of vertices as symmetry factor. All lines are dressed (skeleton diagrams)
and the propagators of the corresponding lines are given in the main text.

Figure 4. Self-energy contributions to ~o, obtained by cutting a dotted line in the

generating functional. Note that there is an additional sum on the spins in the 3rd
diagram.

T 3(z+iw n +W4, wI, W2). Multiplying by the corresponding spectral functions

Aa(w), by Fermi's functions and by the auxiliary propagators, times an
external Po(z), and integrating all read off from the 3rd diagram of fig. 4
we obtain the contribution ~~~~(iwn). The final equation then is

"uc,a (.ZW )
n = "NCA(.
uc,a ZW n ) + ,,(a)('
uc,a ZW n ) + ,,(b)('
uc,a ZW n ) (21 )

together with the contribution from the 1st diagram in fig. 3

~c
NCA -
,
a - J
dz -(3z PO(z)Pa(z + zw
-.e
21rZ
.
n) (22)

The contributions for the self-energies of the auxiliary propagators follow

from the 1st diagram in fig. 3 and from the 3rd diagram of fig. 4 with the
dashed dotted box replaced in the same way as for ~~~~(iwn), which then
gives ~6b)(z). The self-energy ~~)(z) follows with the same substitution
from the third diagram in fig. 5. Finally the equations for T 2 and T 3 follow
from fig. 6 and will not be written here because of lack of space.

References

1. P.W. Anderson, Phys. Rev. 124, 41 (1961).

314

Figure 5. Self-energy contributions to '2:. j , obtained by cutting an arrowed line on the

circumference in the generating functional. Note that there is an additional sum on spins
in the 2nd diagram.

0
Figure 6. Scattering matrix T 1 and self-energies '2:.6 ) and '2:.~0).

2. H. Keiter, J.C. Kimball, Phys. Rev. Lett. 25, 672 (1970).

3. N. Grewe, H. Keiter, Phys. Rev. B 24, 4421 (1981).
4. H. Keiter, G. Morandi, Phys. Rep. 109, 227 (1984).
5. N.E. Bickers, Rev. Mod. Phys. 59, 845 (1987).
6. A. Georges, G. Kotliar, W. Krauth, M.J. Rozenberg, Rev. Mod. Phys. 68,13 (1996).
7. A.C. Hewson, The Kondo Problem to Heavy Fermions, Cambridge University Press
(1993).
8. W. Metzner, D. Vollhardt, Phys. Rev. Lett. 62, 324 (1989).

Figure 7. Scattering matrices T 2 and T 3 . The complication is related to the additional

spin summation.
 315

9. G. Hiilsenbeck, Q. Qin, Solid State Comm. 90, 195 (1994).

10. C. Kim, Y. Kuramoto, T. Kasnya, J. Phys. Soc. Japan 59, 2414 (1990).
11. H. Keiter, T. Lenders, Europhys. Lett. 49, 801 (2000).
12. N. Madras, G. Slade, The self-avoiding walk, Birkhanser Verlag, Basel (1993).
13. S. Kirchner, J. Kroha, JLTP 126, 1233 (2002).
ITINERANT FERROMAGNETISM FOR MIXED VALENCE
SYSTEMS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

J. BONCA and S. EL SHAWISH

J. Stefan Institute, 1000 Ljubljana, Slovenia
C.D. BATISTA
Center for Nonlinear Studies and Theoretical Division, Los
Alamos National Laboratory, Los Alamos, NM 87545
J.E. GUBERNATIS
Theoretical Division, Los Alamos National Laboratory, Los
Alamos, NM 87545

Abstract. We investigate the Periodic Anderson model in the regime of itinerant fer-
romagnetism. We compare Quantum Monte Carlo (QMC) results with results obtained
using the Gutzwiller approximation (GA). As expected, the energy of the paramagnetic
state is overestimated by the GA in comparison with QMC results; however, the partially
saturated ferromagnetic (FM) state energies obtained by both methods are surprisingly
similar. While the GA gives the ferromagnetic instability too easily, its results for the
FM ground state, once its existence is confirmed by more dependable methods, can be
used for further examination. This thesis is confirmed by a direct comparison of the mean
values of the occupation numbers nk obtained using both methods.

Key words: Gutzwiller Approximation, Ferromagnetism, Periodic Anderson Model, Quantum

Monte Carlo, Mixed Valence Systems

Introduction

Recently, a novel mechanism for itinerant ferromagnetism has been sugges-

ted Refs.(l, 2). The simplest model, containing this mechanism, is a simple
two band model consisting of a narrow correlated band hybridized with
a dispersive, uncorrelated one. The hybridization between bands and the
particular band structure playa crucial role in this mechanism by generat-
ing a multi-shell structure for the correlated orbitals. This structure, when

317
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 317–326.
© 2003 Kluwer Academic Publishers
318

combined with the local Coulomb repulsion, favors a partially saturated

ferromagnetic state (FM) which forms as a consequence of a generalized
Hund's rule. The most important physical mechanism that leads to FM is
associated with the fact that spin-polarized electrons are more successful in
avoiding the Coulomb interaction due to Pauli principle than unpolarized
ones.
In the past, many Gutzwiller approximation approaches were developed
to investigate the ground state properties of the periodic Anderson model
(PAM), Refs. (3, 4, 5, 6). Most of these approaches were focused on the
Kondo limit of the PAM while the itinerant ferromagnetic state has been
found in the mixed valence regime, Refs.(l, 2). In this Paper we focus on a
direct comparison between QMC results and GA calculations in the mixed
valence regime. We derive GA equations for the spin-polarized PAM by
closely following the work by Fazekas and Brandow Ref.(5). We focus on
the direct comparison of energies as functions of magnetization using both
above mentioned methods. For a closer inspection of the nature of the FM
state in comparison to the energetically higher paramagnetic (PM) state
we compute average particle densities.

Model

The PAM was originally introduced to explain the properties of the the
rare-earth and actinide metallic compounds including the so called heavy
fermion compounds. The basic ingredients of this model are a narrow and
correlated f band hybridized with a dispersive and uncorrelated d band.
The Hamiltonian associated with this model is:

H = - t L (dtudr'u + d~'udru) + Ef L ntu

(r,r'),u r,u

+ V LUJudru + dtufru) + ~ L ntunta , (1)

r,O" r,O"

where flu and dbu create an electron with spin (J' in f and d orbitals at lattice
site rand ntu = flu fru. The hopping t is limited to the nearest-neighbor
sites. We will work in units where t = 1.

Gutzwiller method

We extend the two-band version of the Gutzwiller U ----+ 00 method proposed

by Fazekas and Brandow Ref.(5) to the spin-polarized ground state. After
replacing the determinant expressions by their average values we make no
further approximations in optimizing the ansatz with respect to a large
 319

number of independent variational parameters. Laborious and exacting

derivations not presented here can be found in Refs.(3, 5).
We construct a magnetic trial state by allowing both d- and f-band
polarization, as well as spin-dependence of the mixing amplitudes. The
average electron density per site n is assumed to be sufficiently small, i.e.
n < 1, so that only the lower hybridized band needs to be used for either
spin, Ref.(5). Our ansatz for a magnetic ground state is thus obtained by
projecting out the p component from an optimally chosen free-electron
hybridized ground state,

17/)} -_ II (1- niln

f f
it ) II II (1 t
+ aa(k)fkadka)IFS a} . (2)
a kEFSo-

In the spin-up Fermi sea IFS r }, the d-band is filled up to the Fermi level
EFT, and similarly for down-spins, leaving all f-states empty. The number
of independent variational parameters aa(k) equals the number of k-states
within the Fermi volume.
As in the single-band Gutzwiller method we seek the term dominating
the normalization factor ('If!I'lf!) to set as a replacement for the sum over all
possible configurations. After the average treatment of the determinants
is carried out, Ref.(5), the combinatorics brings us to the self-consistency
condition
(3)

where
1- nf
qa = (4)
1 - nfa'
with nf = nfa + nf(j· With the help of Eq.3 we obtain

IH\= " (Gf - Gk)a;(k) - 2Va a (k)

\ I L.J Gk
k,a
+ 1 2
q;; + aa(k)
' (5)

where Gk = -2t Li cos(kxJ is the d-band dispersion relation. Here, as in

all subsequent summations over k, the sums are understood to be taken
such that k E FS a .
The minimization of the expression Eq.5 for (H) is complicated by the
fact that the self-consistency equation Eq.3 for nfa connects all variational
parameters aa(k). Taking the partial derivative of Eq.3 with respect to
aa(k) we obtain
8nfa 2
8 qa-1
8a a (k) -A (1- nf) 8a (k) +Ca ,
a (6)
a
8nf(j 2
8 q(j-1
-A(j(1- nf) 8a (k) ' (7)
8a a (k) a
320

where
1 a;(p)
L(1 - nf)2 ~ (q;;l + a;(p))2'
1 L (cf - cp)a;(p) - 2Va a (p)
L(I-nf)2 p (q;;l+a;(p))2
2a a (k)
(8)

Similarly, by recalling Eq.4, we arrive at the second set of equations

onfa nfij onfij 1 - nfa

(9)
oaa(k) (1 - nf)2 + oaa(k) (1 - nf)2'
onfa nfa onfa 1 - nfa
(10)
oaa(k) (1 - nf)2 + oaa(k) (1 - nf)2'

from which a straightforward expression for all derivatives follows

onfa C 1 + Aanfa
(11)
oaa(k) a 1 + Aanfa + Aanfij + A a A a (1- nf)'
onfij _ C Aanfij- Aa
(12)
oaa(k) - a 1 + Aanfa + Aanfij + A a A a (1- nf)·

We next minimize the energy, Eq.5

o(H)
oaa(k)

where the renormalized i-level is given by

- Ba + Banfa - Banfa - AaBa (1- nf) (14)

cfa = cf - .
1 + Aanfij + Aanfa - A a A a (1- nf)

Using :a~~~) = 0 we get from Eq.13 the corresponding variational paramet-

ers
2V
aa(k) = (15)
Efa - Ck + V(ck - Efa)2 + 4V2 qa
Comparing Eq.15 to the U = 0 solution, Ref.(5), the effect of switching on
U ----+ 00 is: a) to replace Cf by the renormalized i-level Efa and b) to replace
the bare hybridization V by the renormalized V,;q;; in the denominator of
aa(k), but not in the numerator. As a simple test to verify the relation
Eq.14 we compute the non-magnetic solution, i.e. (J = 0-, which leads to
 321

the correct result, Ref.(5), Efu = Efa- = cf - B u/(1 + Au). The Eq.5 is
just a formal relation, since the determination of Efu and nfu still needs
to be done. For a general set of model parameters, this can be achieved
numerically using iterative methods.

Quasiparticle density
Contrary to the Hubbard model, Refs.(7, 8), the number of the strongly
correlated f-electrons in the two-band model is not conserved. In order to
control this number in lower and upper band, we calculate the quasiparticle
density as a function of k. Following the steps in Ref.(5) we find the d-
electron distribution

(16)

for k's contained in the FS u ' The f-band occupation number does not obey
a simple relationship like Eq.16. Instead, just as in the Gutzwiller solution
of the Hubbard model Refs. (7,8), correlation causes the f-state occupation
to spread over the entire Brillouin zone, rather than being confined to the
k-values contained in the FS u ,

(1 )n + qua;(k) . k E FS
/ f )-/ft
\n f)- -qu fu qu-1 + au2(k) , u
. (17)
ku - \ ku ku -
{
(l-qu)nfu ikctFSu
The above relation suggests that the f-electron distribution experiences a
jump at the Fermi wave vector. Analogous procedure, Ref.(5), brings us to
the mixing terms,

(18)

which holds only for k's contained in FS u ' Instead of presenting the actual
dispersion of f - and d-electrons in the k-space, we decide to investigate
the quasiparticle density of hybridized bands defined with the following
operators
t
a ku u(k)flu +v(k)dtu'
13L -v(k)flu + u(k)dtu' (19)
with
Ek - cf
u(k) ,
J (Ek - Cf )2 + V2
-V
v(k) , (20)
J (Ek - Cf )2 + V2
322

-3.08
-1.952
-1.954 V=0.05 (}--{)OMC'Uy
~ Q----O OMC, U=8
~ -1.956 GA -3.10
W -1.958
-1.960 8x8
-3.12 t;=-2.5
z a) V=O.l ~
-1965
~
W -1.97 =--=----- w
-------
~
W -3.14
V=0.5
n=9/32s0.28
-1.975
~
~/_----
-2 -3.16
~
W -2.01 - ,~,

-2.02 -3.18
o 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
M M

Figure 1. a) E/N for 1D PAM vs. AI. Other parameters of the system are N = 50,
n = 0.3, Ef = -1.5. Results are presented for various V = 0.05,0.1,0.2 and U = 4,8.
Gutzwiller results are presented with full lines. b) E/N for 2D PAM vs. lVI. Other
parameters are presented in the inset. QMC results were calculated on a 8 x 8 square
lattice, Gutzwiller results were calculated on a 8 x 8 lattice (full line) and 100 x 100
(dashed lines).

J
and Ek = ~ (ck + Cf + (ck - Cf)2 + 4V2 ). We finally arrive at desired
relations for the upper (0:) and lower ((3) band, respectively,

u2(k)(nL) + v2(k)(n~a) + u(k)v(k)(n[~ + nff'a),

v2(k)(nL) + u2(k)(n~a) - u(k)v(k)(n[~ + nff'a)' (21 )

Results

Our numerical method, the constrained-path Monte Carlo (CPMC) method,

is extensively described and benchmarked elsewhere, Refs. (9, 10). For de-
tails about the implementation of the method to PAM we refer the reader
to our recent publication Ref.(2). Most of our QMC calculations were per-
formed on N = 50 and 70 lattice sites for 1D cases and on 2D square
lattices with N = N x x N x lattice sites with N x = 8 and 12. Calculations
were done at various dopings n = N e / N where N e is the number of elec-
trons N e = Nt + Nl and various magnetizations M = (Nt - Nl)/2N. In
Fig.1.a. we show E/N for 1D PAM as a function of the magnetization JIll
for various values of hybridization V and two coupling strengths U. In all
cases E(M)/N reaches minimum at finite magnetization 0.3 :::; M :::; 0.35.
Comparing QMC results with GA, we observe that the difference between
energies obtained using the two methods systematically decreases with in-
creasing M. This is simple a consequence of the fact that with increasing
 323

the magnetization the system becomes less correlated which renders the
GA results more accurate. It is less straightforward to understand why
lowering of V also leads to a similar effect. The main effect of lowering V is
diminishing the crossover region in k-space where the d- and f - bands
hybridize. We would also like to stress that even though GA overestimates
the paramagnetic energy, it predicts reasonably well the value of magnet-
ization of the ground state. This is particularly evident at small V = 0.05.
Comparing systems with substantially different U = 4,8 we notice that at
small V = 0.05 we obtain almost identical results (the two plots for U = 4
and 8 nearly overlap) while at larger values of V = 0.1, and 0.2 differences
are visible nevertheless they remain small. This confirms our hypothesis
from Refs.(l, 2) that the relevant energy scale leading to ferromagnetism
is the hybridization gap ,6,. When U » ,6, the system essentially reaches
U ---t 00 limit.
Results from the 2D system show similar trends as the 1D results, see
Fig.l.b. The main difference is that in the contrast to the 1D case, the
system size used to obtain QMC results in 2D is not large enough to reach
the thermodynamic limit. While finite-size scaling has been performed in
our previous work Refs.(l, 2), we only show QMC results for 8 x 8 system
and GA results for 8 x 8 and 100 x 100 systems. The most prominent
finite-size effect in the GA (comparing 8 x 8 and 100 x 100 systems) is a
slight increase of the ground state energy around M rv 0.42. As in the 1D
case, the GA overestimates the paramagnetic energy in comparison to QMC
calculation. We should nevertheless stress that even in the paramagnetic
case the relative difference between GA and QMC energies is no more than
2%.
Next we compare the mean value of the occupation numbers (n~;(3) for
the 1D PAM as obtained using GA, Eqs.21 and QMC results for the para-
magnetic (Fig.2.a.c.) and ferromagnetic solutions (Fig.2.b.d.) as obtained
using V = 0.05 and V = 0.2. GA calculation captures surprisingly well
many details of (n~~,(3) in both the PM and FM cases. Starting with the
lower band (nk(u,d) for both PM solutions, presented in Fig.2.a.c., we first
see a downturn around k rv 0.27r, which is more pronounced in the V = 0.05
case, followed by the plateau-like k-dependence. Even though the position
of the downturn can be easily predicted by the position of the crossover
from the d-like to the f -like nature of the hybridized 0: band, the actual
hight of the plateau for k > 0.27r is set entirely by correlation effect. Even
more surprising is the finding that the GA nearly perfectly predicts the size
of the jump at the Fermi wave vector k = kp = 0.67r and the magnitude of
(nk(u,d) for k > kp .
Good agreement persists in the FM regime, Fig.2.b.d., where upper
band majority (n ku ) and minority (n kd ) spin occupancies are equally well
324

1.5 1.5
• <n~"(k» 1D U=4 M=0.3
10 U=4 • <n~"(k»
N=70 N=70 ~=-15
~=-15 o <n~,(k» <n~IoI>=O.025(4)
n=O.3 n=0.3 V=0.2

...
V=0.2 0 <n~d{k»

1\
------ GA: <n"U{d)(k» 1\ • <n .(k»
:;z - - GA :<n~"I,){k» :;z • <nud(k»
C a) C
v v o <n'.(k»
,I ~
,I
M=O o <n',(k»

r
0.5 I 0.5 ----- GA:<n".(k»
<n~ ".>=0.044(5) b)
--- GA:<n",(kj>
I -- GA:<n~,,(k»
I
""'-~-'
I
I ---- GA:<n' ,(k» ,
1__ -
I
0 0
0 02 0.4 0.6 0.8 0 02 04 0.6 08
kIn) kIn)

15 15
(k):>
e <n~
10 U=4 M=033
10 U=4 • <n":(k):>
N=70 N=70 £,=-1.5
£,=-1.5 o <n~"{k» <n~",>=0.015(8)
n=O.3 n=0.3 V=0.05
V=0.05 0 <n',(k»
------ GA: <n"U{d)(k):> ,
1\
:;z c) :f - - GA :<n~"I.)(k»
1\
:;z d)
'\\ ~ e<n".(k»
• <nl;\.{k}> ---!
"E'
v M=O
"E'
v tI o <n',(k» i
05 <n~",>=0.050(11)
i
I

I
05 I
\
~ ,
o
-----
---
<n',(k»
GA:<n",(k»
GA:<n",(k»
i
i
- GA:<n~.(k»

~
I I
~!I ---- GA:<n~d(k»

0
0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
kIlt) kIlt)

Figure 2. Mean value of the occupation numbers of the non-interacting band states of
the PAM in the ID lattice for the paramagnetic state .M. = 0 a) and c) and for the ground
state with partially saturated ferromagnetic solution b) and d). Legends: (n~u) repres-
ents lower-band occupation number of up-spins, (n~d) represents upper-band occupation
number of down-spins, etc.

reproduced using both methods. Furthermore, we notice that the minority-

spin upper band occupancy (nfd) reaches a peak with increasing k that is
located around k rv O.21T. On the other hand, the majority (nfu) occupancy
monotonously decreases with increasing k. It seems as if the spin-up and
spin-down occupancies are avoiding each other in the k-space. This effect
is captured by the QMC as well as with the GA calculations. Finally, we
would like to emphasize yet another interesting result. Comparing the total
occupancy of the upper p-band for the PM and FM solution

(nfot)= L (nfcr) , (22)

k,cr
we find, that (nfot) for the PM solution is always larger than (nfot) for the
FM solution (actual numbers are given in the inserts of Figs.2). This agrees
 325

1.5 1.5
12x12 V=-05 M=O 12x12 U=10.0 M=0.42
n=0.29 £,=-3 n=0.29
U=10.0 • <n~"(k»
b) V=-05
£,=-3 • <n~ik»
O<n~"(k»
"
:;<
C
a) o <n~d(k»
- - GA:a
"
:;<
C
• <n"u(k»
• <n"d(k»
v v o <n~,(k»
----- GA:13
0.5 0.5 , 0 <n'd(k»
: ' - - GA:o.
,"6 ------ GA:13

0 0
r M X r r M X r
Figure 3. Mean value of the occupation numbers of the non-interacting band states of
the PAM in the 2D 12 x 12 lattice for the a) paramagnetic state M = 0 and b) for the
ground state with partially saturated ferromagnetic solution with !vI = 0.42.

with our general picture of itinerant ferromagnetism Refs.(l, 2) where a

FM ground state develops as a consequence of a generalized Hund's rule
where electrons in the partially saturated FM state feel weaker Coulomb
interaction in comparison to electrons in the PM state. Since the upper
band occupancy develops as a consequence of electron interactions, it is
then reasonable to expect the inequality (n~ot(M = 0))> (n~ot(M > 0)).
In Fig.3 we present the occupation numbers for the 2D system. While
agreement between QMC and GA results is close in the PM case, it is rather
surprising that in the FM case agreement is not as close as in the ID calcu-
lation. The unusual up-turn of the minority upper-band occupancy seems
to be well captured by both methods. However the majority upper-band
occupation numbers seem to show a large discrepancy. In part we attribute
the disagreement to finite-size effects that are much more pronounced in
2D due to a smaller linear dimension of the lattice and in part to different
parameters of the model.

Conclusions

We have presented a quantitative comparison of the QMC and GA res-

ults in the mixed valence regime of PAM. Following work by Fazekas and
Brandow Ref. (5) , we have extended their two-band version of the Gutzwiller
method to the spin-polarized state. Apart from the so-called Gutzwiller
approximation we have made no further simplifications to the model. We
have solved Gutzwiller equations numerically using iterative procedures.
As naively expected, GA overestimated the energy of the PM solution;
however captured reasonably well the energy of the partially saturated
FM state. For this reason, GA was not suitable to determine the phase
326

diagram; nevertheless it can be used to investigate physical properties of

the partially saturated FM ground state as long as its existence has been
established by more reliable methods. Contrary to naive expectations, we
have shown, that despite overestimated PM energy, GA predicted well the
unsaturated magnetization of the FM ground state and furthermore, it gave
the quasiparticle occupation numbers with surprising accuracy in the FM
as well as in the PM case.

Acknowledgements

J. B. acknowledges the support of Slovene Ministry of Education Science

and Sports, FERLIN, and the hospitality and financial support of the Los
Alamos National Laboratory where part of this work has been performed.

References

1. C. D. Batista, J. Bonca and J. E. Gubernatis, Phys. Rev. Lett. 88, 187203 (2002).
2. C. D. Batista, J. Bonca and J. E. Gubernatis, cond-mat/0208604.
3. T. M. Rice, K. Deda, Phys. Rev. Lett., 55, No.9, 995 (1985); Phys. Rev. B 34,
6420 (1986).
4. C.M. Varma, W. Weber, and L.J. Randall, Phys. Rev. B 33, 1015 (1986).
5. P. Fazekas, B. H. Brandow, Physica Scripta, 36, 809 (1987).
6. P. Fazekas and E. Muller-Hartmann, Z. Phys. B - Condensed Matter 85,285 (1991).
7. M. C. Gutzwiller, Phys. Rev., 137, No. 6A, A1726 (1965).
8. D. Vollhardt, Rev. Mod. Phys., 56, No.1, 99 (1984).
9. S. Zhang, J. Carlson and J. E. Gubernatis, Phys. Rev. Lett., 74, 3652 (1995).
10. S. Zhang, J. Carlson and J. E. Gubernatis, Phys. Rev. B 55, 7464 (1997); J. Carlson,
J. E. Gubernatis, G. Ortiz, and Shiwei Zhang, Phys. Rev. B 59, 12788 (1999).
TRANSPORT PROPERTIES OF HEAVY FERMION SYSTEMS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

C. GRENZEBACH and G. CZYCHOLL

Institut fur Theoretische Physik, Universitiit Bremen, D-28334
Bremen, Germany

Abstract. Heavy fermion systems (HFS) are described by the periodic Anderson model
(PAM), which is studied within the dynamical mean-field theory (DMFT), mapping the
PAM on an effective single-impurity Anderson model (SIAM) to be determined selfcon-
sistently, which becomes exact in the limit of high spatial dimensions, d ---+ 00. We use the
modified perturbation theory (MPT) as approximation for the effective SIAM. The MPT
is exact up to second order in the Coulomb correlation U and simultaneously reproduces
the exact results for the atomic limit and the lowest moments. Within this approximation
we have calculated the temperature dependence of the resistivity, the thermopower and
the frequency and temperature dependence of the dynamical conductivity. For all these
quantities the typical HFS-behavior is qualitatively well reproduced within our treatment.

Key words: Heavy Fermions, R.esistivity, Thermo(electric) Power, Temperature Depend-

ence of Transport Quantities, Periodic Anderson Model, Dynamical Mean-Field Theory,
Modified Perturbation Theory, Dynamical Conductivity

1. Introduction

Besides the specific heat (,-coefficient) and the magnetic susceptibility the
unusual transport properties are most characteristic for heavy fermion sys-
tems (HFS). Pure metallic HFS, for example CePd 3 (1), CeCu6 (2), have
a small zero temperature resistivity R(T = 0), a sharp increase of R(T) for
low temperature T, a maximum at some characteristic temperature T*, and
a R(T) decreasing with increasing T, i.e. a negative temperature coefficient
(NTC), for higher temperatures. There are, however, also some HFS (for
instance U Pt3, YbCuAl) for which no maximum but only a flattening and
a plateau behavior of R(T) is observed for higher T.
Another interesting transport quantity is the thermoelectric power (TEP).
In HFS the TEP S(T) may have a sign change at low T (6, 7, 8, 9, 10) and
it can assume giant values; much of the recent interest in the heavy fermion

327
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 327–334.
© 2003 Kluwer Academic Publishers
328

thermo-electricity is due to the belief that some HFS with a thermopower

larger than 150 JLV/K might be useful for application.
In the frequency dependent (dynamical) conductivity O"(w, T) one usu-
ally observes a narrow Drude peak at low T and an additional peak at higher
frequency. With increasing T the Drude peak quickly disappears and the
finite frequency peak becomes less pronounced and there is a crossover to
a behavior with one small, but broad peak around zero frequency (ll).
Finally, we mention the Kondo insulators like 5mB6 and Ce3Bi4Pt3
which show clear evidence of a gap formation at low T resulting in an
activation behavior of R(T) for very low T and an (exponentially) vanishing
conductivity O"(w, T) for very low T and low w, whereas for higher T the
behavior is similar as in the metallic HFS. The thermopower can be very
large also for the Kondo insulators.
In this contribution we present an oversimplified model and approxim-
ation within which all mentioned experimentally observed transport prop-
erties being typical for HFS can be reproduced and explained.

2. Model

It is well accepted that the basic electronic properties of HFS are described
by the periodic Anderson model (PAM), which contains a conduction band,
strongly correlated f-levels localized at the sites of a lattice simulating
the rare-earth or actinide ions, and a hybridization between f-states and
conduction electron states. For real rare earth or actinide atoms or ions
the f-shells are highly degenerate, namely 14-fold degenerate in the bare
(hydrogen like) atom, which is split into a 6- and an 8-fold degenerate
level due to spin-orbit coupling, and this degeneracy is further reduced in
a crystal due to crystalline electric fields so that the lowest f-level may be
only two-fold degenerate. But the crystal-field (CF) split level is thermally
accessible so that taking into account only a two-fold degenerate f-level is
probably not sufficient to account for the full temperature dependence of
the relevant physical quantities.
Nevertheless for the actual calculations presented in this paper the
standard PAM with only a two-fold degeneracy of the f-levels has been
used, i.e. (with nftlT := f~ITfRIT)

H = LEkCLITCklT + L [Efn ft lT + ~ nftITnft-1T + V(ckITfRIT+C.C.)] (1)

kIT RIT

where Ek = 2t 2:.1=1 cos(kza) denotes the band electron dispersion (t nearest

neighbor hopping, a lattice constant), for which a tight-binding model on
a simple-cubic lattice in dimension d is assumed. To calculate transport
 329

quantities one needs also the current operator, and for the current operator
being consistent with the PAM the component in x-direction is given by

8Vk t
8k UkaCka + c.c.) (2)
, x v
~

= 0, if V k-independent
So for k-independent hybridization V one has in site representation

(3)

(i imaginary unit). Being interested in transport quantities two-particle

Green functions for the PAM have to be calculated. To calculate the con-
ductivity we start from the Kubo formula for the (diagonal element of
the) frequency (w-) dependent conductivity (tensor), which (according to
standard linear response theory (12, 13, 14)) is given by the current-current
response function:
1 .
o-(w) ---ImX' .
w + iO Jx,Jx (w+zO) (4)

-i ~= dt'e izt' < [jx(t/),jx(O)]- > (5)

The static conductivity can be obtained as the zero frequency (w ---t 0)

limit of the dynamical conductivity and then the TEP follows by means
of the standard Mott relation (15). So one must determine two-particle
Green functions of the kind « ctacka; ct'alCklal »z. Using standard rela-
tions of many-body theory (12) the two-particle Green functions can be
obtained from (products of) one-particle Green functions and additional
vertex corrections. The one-particle Green functions are defined by

Gk~(Z) = «aka;bL »z= -i ~= dt/e izt' < [aka(t'),bL(O)]+ > (6)

(with a, b E {c, j}). The f-electron selfenergy is defined by:

«
(«
Cka; cta »z «Cka; ita »z ) _ (z - Ek - V ) -1 (7)
Aa; cta »z «Aa; ita »z - - V Z - E f - ~kf(Z)

3. Approximations

We study the PAM within the dynamical mean-field theory (DMFT) (19),
which becomes correct in the limit of infinite dimension d ---t 00, t ---t 0
330

ntotal = 1.5; U = 1.0; Ef = -0.5 ntotal = 1.5; V = 0.2; Ef = -0.5

0.25r;======~----',-------, _=-=U===1=.0~
0.08,-----=~---?
- V=0.10
- V=0.15 ---- U = 1.5
0.2 ---- V=0.20 --- U 2.0 =
-- - V = 0.30
%,0.15 - V=OAO ." _
~ ,/#,~~;::.,: ~ - .
&. 0.1 /;::~'-"
0.05 ,<?-------------- 0.02 If
"
I ?--- "
It.lL__
0.05 0.1 0.15 0.1 Q2 0.3
Temperature T Temperature T

Figure 1. T-dependence of resistivity for Ef = -0.5,nLol = 1.5 and for U = 1 and

different hybridizations V (left panel) and fixed hybridization V = 0.2 and different U
(right panel)

keeping dt2 = constant (16). Within DMFT certain simplifications be-

come exact, which extremely simplify practical calculations. It has been
shown (17) that these simplifications are already a reasonable additional
approximation for realistic dimension d = 3, i.e. 3-dimensional systems
behave already similar as systems in the mathematical limit d -----t 00. The
most important simplification is that the selfenergy becomes site-diagonal
(k-independent), i.e. the so called "local approximation" becomes exact
within DMFT and is a reasonable approximation for d = 3. Consistent with
the site diagonality of the self energy is the vanishing of vertex corrections
for the current-current response functions. Then the frequency dependent
conductivity is simply given by (18)
2 2-d
f
o-(w,T)=ea n t 2 !dE (E)-f(E+W) L(E,E+w) (8)
2~ w
where f(E) is the Fermi function and the function L(E1 , E 2 ) is defined by

L(E1 , E 2 ) = ~ L(ImGko-(E1 + iO)ImGko-(E2 + iO)) (9)

N ko-
where
Gko-(Z) = «Cko-; cto-»z =
Z -
v~ - Ck
(10)
z-Ef-L.f(z)
is the band electron Green function ofthe PAM and ~.f(z) the (k-independent)
f-electron selfenergy. The static conductivity (and the resistivity R(T) as
its inverse) follow from the limit w -----t 0, and the TEP is also determined
by the function L(E1 , E 2 ) according to (15)

J dE ( - :~) (E - fL)L(E, E)
S = ----'-------'--;----,--------- (11)
eT J dE (--£) L(E,E)
 331

V = 0.2; £1 = -0.5; ntotal = 1.5

u = 1.0; £1 = -0.5; nlotal = 1.5 - V=0.10
---- V = 0.15 - U=1.0
sz 100 \_., --- V = 0.20 SZ ---- U 1.5 =
::> .,'" ,"'.,-r,.,.-....-----.. -+- V = 0.30 ~ -- - U 2.0 =
2, ,
~ V=OAO
i=' 50/ i='
Cii ' Cii 0
~
oCl.
0 ~
o
Cl.
o ~ -50
E
iii -50 Q)
.!::
f- F_ 100
0.05 0.1
Temperature T
0.15 0.2 o 0.05 0.1
Temperature T
0.15 0.2

Figure 2. T -dependence of thermopower, otherwise as in Fig.1

Therefore, in the limit d ---t 00 the transport quantities of interest can

be obtained from the one-particle Green functions or the (k-independent)
selfenergy alone.
But a full exact solution of the PAM is not possible even in infinite
dimensions and one needs additional approximations. Within DMFT a self-
consistent mapping on an effective single-impurity Anderson model (SIAM)
is possible. (19) The DMFT selfconsistency relation reads

G~SIAM(z)
1
z - Ef - ~f(z) - ~(z)

~L 1 (12)
N k
z - E f - ~f (z) - ~
Z-Ek

Here the "bath" Green function ~(z) (i.e. the effective SIAM conduction
band Green function) has to be determined selfconsistently. But the selfen-
ergy for the effective SIAM is not known rigorously and an additional
suitable approximation for the SIAM has to be used. We have applied the
"modified perturbation theory" (MPT), which starts from the following
ansatz (20, 21, 22)

(13)

where ~1°C (z) is the second order contribution in U to the SIAM selfen-
ergy relative to the Hartree-Fock solution and the parameters Ct, j3 can be
determined by the condition that the atomic limit (of vanishing V) and an
additional criterion (Fermi liquid sum rule, reproduction of the first four
moments) are fulfilled. The MPT has the advantage that it is exact up to
order U 2 and the atomic limit is fulfilled simultaneously.
332

ntotal = 1.5; V = 0.15; U = 1.0 ntotal = 1.5; V = 0.20; U = 1.0

4~.--n--~-~===="il 40rr--"';=~---r========­
--e- T = 0.001 --e- T = 0.001
---- T = 0.020 ---- T = 0.020
-+- T = 0.040 -+- T = 0.040
30
- T=0.100 - T= 0.100
i="
8 20
0"

10

--0=-'".2=---=0"'3~~0~.4~~0.5·
00:----=-'0."'"1
m OJ

Figure 3. Frequency dependence of dynamical conductivity for different T,

Ef = -0.5, That = 1.5, U = 1 and hybridization V = 0.15 (left panel) and V = 0.2
(right panel)

4. Results

For practical calculations we choose a semielliptical model density of states

for the unperturbed conduction band, i.e. we use the assumption

~
1
2:-- = 8z·
N k Z - Ek
(1- )1- (4Z)-2) , (14)

which becomes exact for a Bethe lattice in the limit of infinite coordination
number. Thereby we have chosen the unperturbed conduction band width
as our energy (and temperature) unit. The chemical potential JL has to
be determined selfconsistently for a given total number ntot of electrons
per site. Choosing ntot = 1.5 JL does not fall into a possibly existing
hybridization gap, i.e. the model describes a metallic situation. Typical
results are shown in Figs. 1-3. The resistivity R(T) is small for T ----+ 0,
has a rapid increase with increasing T, a maximum at T*, and an NTC for
higher T > T*. With increasing hybridization V T* also increases and for
larger V one can get a behavior with no maximum but a plateau behavior.
Changing U for fixed V only the absolute value of R(T) varies whereas
T* remains fixed. In the thermopower S(T) we observe the large absolute
value (of the magnitude 100JLV / K), a low T extremum and a sign change
at a temperature rv T*. In the frequency dependent (optical) conductivity
O"(w, T) (Fig.3) we observe a Drude peak and an additional (mid infrared)
peak at finite frequency for low T. With increasing T we obtain a crossover
to a behavior with only one rather broad peak around zero frequency.
The position of the mid infrared peak obviously increases with increasing
hybridization V.
Choosing ntot = 2, E f = -0.5, U = 1 we have the "symmetric PAM"
and the chemical potential JL falls into a hybridization gap at zero T; this
 333

ntotal = 2; U = 1; £1 = -0.5 U = 1.0; £1 = -0.5; ntotal =2

0.251:---,------"='=='==:==:==,---1
- V-0.20 I SZ 0
0.2
-. -_. V = 0.30
--- V = 0040
~
0-2
o

fO.15
, ~-4
~
'iii , --
, :: ...... --
if)
~ -6
& 0.1 ' .....
~
0--8
o
0.05 E ~ V=0.20
",-10 --+-V = 0.30
..c
f- - - V =0040
-12
0.1 0.2 0.3 0 0.05 0.1
Temperature T Temperature T

Figure 4. T-dependence of R(T) (left panel) and S(T) (right panel) for ntot = 2,
Ef = -0.5, U = 1, V = 0.2

ntotal = 2; V = 0.20; U = 1.0

40,------,-"'~----,--.-r========;l
-- T = 0.001
---- T=0.020
-+- T = 0.040
30
- T=0.100

o 0.1 0.2 0.3 004 0.5

OJ

Figure 5. w- and T-dependence of o-(w, T) for ntot = 2, Ef = -0.5, U = 1 and V = 0.2

situation may, therefore, model a Kondo insulator. Results for R(T), S(T)
and O'(w, T) are shown in Figs. 4, 5. The hybridization gap (of width 6.)
manifests itself in an activation behavior, i.e. an exponential increase of
R(T) for very low T and in a vanishing of O'(w, T = 0) for w < 6.. With
increasing T the hybridization gap disappears and we get a crossover to a
behavior similar as in metallic HFS for sufficiently high T. The thermopower
is also extremely large, even one magnitude larger than in the metallic HFS.

5. Conclusion

The most characteristic features observed in the T-dependence of the res-

istivity and the thermopower and in the T- and frequency-(w- )dependence
of the dynamical (optical) conductivity of HFS can qualitatively be repro-
duced within simple approximations of the PAM, in particular within the
DMFT/MPT. This investigation can be extended into several directions.
The effects of impurities and of alloying (and of "chemical pressure") on
R(T), S(T) and O'(T, w) can be modelled using additionally the CPA for
334

disordered systems, similarly as it has been done for R(T) in (23) using
the SOPT. The treatment can also be applied to a more realistic version
of the PAM with a more realistic (higher) orbital degeneracy of the f-levels
and more realistic assumptions for the hybridization (dispersion); then also
crystal field effects can be studied and included. Furthermore, using a two-
channel Anderson model it should be possible to study also non-Fermi liquid
systems within this approach.

References

1. P. Scoboria, J. E. Crow, T. Mihalisin, J. Appl. Phys. 50, 1895 (1979).

2. Y. Onuki, T. Komatsubara, J. Magn.& Magn. Matter 63 & 64, 281 (1987).
3. F.J.Blatt et al., Thermoelectric Power of Metals, Plenum New York (1962); R.D.
Barnard, Thermoelectricity in Metals and Alloys, Taylor & Francis, London (1962).
4. F. Steglich, Festkorperprobleme XVII, 319 (1977).
5. D. Jaccard et al., J. Magn. Magn. Mater.47 & 48, 23 (1985).
6. E. Bauer, Adv. Phys. 40, 417 (1991).
7. P. Link, D. Jaccard and P. Lejay, Physica B 225, 207 (1996).
8. M. Ocko, B. Bushinger, C. Geibel and F. Steglich, Physica B 259-261, 87 (1999).
9. D. Huo, K. Mori, T. Kuwai, S. Fukuda, Y. Isikawa, J. Sakurai, Physica B281 &
282, 101 (2000).
10. J. Sakurai, D. Huo, D. Kato, T. Kuwai, Y. Isikawa, K. Mori, Physica B281 & 282,
98 (2000).
11. B.C.Webb, A.J.Sievers, T.Mihalisin, Phys. Rev. Letters 57, 1951 (1986).
12. G. Mahan, Many-Particle Physics, Ch.3.8 and 7, Plenum New York (1990).
13. O. Madelung, Introduction to Solid-State Theory, Springer Heidelberg (1978).
14. G. Czycholl, Theoretische Festkorperphysik, Vieweg Wiesbaden (Springer Heidel-
berg) (2000).
15. G. Mahan in: Solid State Physics, Vol. 51, p.82, H.Ehrenreich, F.Spaegen (eds.),
Academic Press (1997).
16. W. Metzner and D. Vollhardt, Phys. Rev. Letters 62, 324 (1989).
17. H.Schweitzer, G.Czycholl, Solid State Commun. 74, 735 (1990); Z. Physik B 83, 93
(1991).
18. H. Schweitzer and G. Czycholl, Phys. Rev. Letters 67, 3724 (1991).
19. A. Georges, G. Kotliar, W. Krauth, M.J. Rozenberg, Rev. Mod. Phys. 68,13 (1996).
20. A. Martin-Rodero, E.Louis, F. Flores, C.Tejedor, Phys. Rev. B 33, 1814 (1986).
21. H. Kajueter, G. Kotliar, Phys. Rev. Letters 77, 131 (1996).
22. D. Meyer, W. Nolting, Phys. Rev. B 62, 5657 (2000).
23. S. Wermbter, K. Sabel, G.Czycholl, Phys. Rev. B 53, 2528 (1996).
TRANSPORT PROPERTIES OF CORRELATED ELECTRONS
IN HIGH DIMENSIONS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

N. BLUMER and P. G. J. VAN DONGEN

Institut fur Physik, Johannes Gutenberg- UniversitCit,
55099 Mainz, Germany

Abstract. We develop a new general algorithm for finding a regular tight-binding lattice
Hamiltonian in infinite dimensions for an arbitrary given shape of the density of states
(DOS). The availability of such an algorithm is essential for the investigation of broken-
symmetry phases of interacting electron systems and for the computation of transport
properties within the dynamical mean-field theory (DMFT). The algorithm enables us
to calculate the optical conductivity fully consistently on a regular lattice, e.g., for the
semi-elliptical (Bethe) DOS. We discuss the relevant f-sum rule and present numerical
results obtained using quantum Monte Carlo techniques.

Key words: High Dimensions, Infinite Dimensions, Correlated Electrons, Optical Con-
ductivity, Quantum Monte Carlo, Transport, Density of States, General Dispersion Method,
F-Sum Rule

1. Introduction

Microscopic studies of strongly correlated electron systems require methods

which take the Coulomb repulsion between electrons explicitly into ac-
count. Many features of such systems can be modeled using the single-band
Hubbard model (1)

H = Ho + Hint = L tij CkaCRja + U L nR;rnR; 1, (1)

ij,a i

where the operators cka and cR;a create and destroy electrons of spin (]"
on site R i , respectively; nRw measures the corresponding occupancy. A
general nonperturbative treatment of this model is only possible in the
limit of infinite dimensionality where the dynamical mean-field theory be-
comes exact: due to a local self-energy the model reduces for d -----+ 00 to a

335
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 335–343.
© 2003 Kluwer Academic Publishers
336

single impurity Anderson model plus a self-consistency equation. For ho-

mogeneous phases, local properties then only depend on the lattice via the
noninteracting DOS p(E) := -kLk6(E-Ek). In the simplest case of uniform
nearest-neighbor hopping on a hypercubic (hc) lattice, the dispersion reads
E~c = - 2t L~=l cos( koJ which for the proper scaling t = t* I yI2d leads to a
Gaussian DOS phc(E) = exp[-E 2 1(2t*2)]/(yl27ft*).
Vertex corrections to the optical conductivity (T(w) vanish (2) in the
limit d ----7 00 so that it may be expressed in the isotropic case as (3)

Here, Ac(w) is the "momentum" dependent spectral function, nf(w') is the

Fermi function, (TO = 27fe 2 11t2 N IV, and

(3)

Note that the frequency- and interaction-dependent part in (2) depends on

the lattice only via the DOS while the explicitly lattice-dependent part p( E)
is universal, i.e., independent of the interaction U, filling, and temperature.
In the hypercubic case, the momentum dependence of the Fermi velocity Vk
becomes irrelevant in (3) for d ----7 00; for unit hopping and lattice spacing,
one observes phc(E) = phc(E). As a consequence, the optical f-sum is then
proportional to the kinetic energy:

1
00 (TO
dw (T(w) --(Hal' (4)
A

=
a 4
However, the hc DOS is unbounded which is hardly compatible with the
single-band assumption. In fact, no regular lattice model with sharp band
edges in d ----7 00 could be constructed so far. In this situation, many DMFT
studies have focussed on the so-called Bethe lattice which is not a regular
lattice, but a tree in the sense of graph theory as shown in Fig. la. The
semi-elliptic DOS p( E) = vi 4 - E2 I (27f) of this model (for Z ----7 (0) fixes
the local properties of the model; transport, however, is a priori undefined.
A derivation of (T (w) directly for the Bethe tree (using the level-picture
Fig. la) by Chung and Freericks (4) is still incomplete (5); up to a factor
of 3, the same expression p(E) ex (4 - E2 ) p( E) was obtained (6) in a heuristic
scheme by enforcing the hc f-sum rule (4). An alternative direct approach
(7) fails to describe the coherent transport expected in the metallic regime.
The local DMFT problem is unchanged when a finite number of hopping
bonds per site are added. Therefore, the periodically stacked Bethe lattice
(Fig. 1b) is still a Bethe lattice in the DMFT sense; potentially coherent
 337

a) c)
/ ./ /
,/ ,,/ ../

... ...
.. /
"

... "
./'

.. /
"

// ./' //

/" /' /"

Figure 1. Bethe lattice: a) conventional tree level picture for coordination Z = 4, b)

stacked Bethe lattice, c) redefined Bethe lattice as a regular lattice with cubic symmetry:
here, symbols denote equal hopping matrix elements relative to a fixed site (central circle).

transport is then well-defined (only) in stacking direction (8). A semi-elliptic

DOS also results from fully disordered hopping on lattices of arbitrary
topology; in this case, O"(w) is incoherent (9).
We will in the following construct and evaluate a new definition for
0"( w) compatible with a semi-elliptic DOS. This definition is unique by
being based on a regular lattice as illustrated in Fig. lc and by leading to
an isotropic conductivity which is coherent in the noninteracting limit.

2. General Dispersion Method

We rewrite the translation-invariant noninteracting Hamiltonian,

fIo = LLt-r ckWcR;+-r,o- = LE(k)nkCY' (5)

i,O" T k,O"

where contributions to the dispersion may be classified by the taxi-cab

hopping distance IITII = L~=1 ITo:!:
00

E(k) = L ED(k), ED(k) = L t-rei-r.k. (6)

D=1 11-rII=D
In high dimensions, only vectors of the form T = L~1 e a , with pairwise
different directions ai need to be considered. This follows from the fact
that the fraction of neglected vectors (with IT· eal > I for some direction
a) vanishes as lid. Furthermore, the considered vectors are of minimal
Euclidean length ITI hinting at maximal overlap, i.e., largest It-rl for fixed
taxi-cab distance IITII and fixed D.
By deriving a recursion relation for ED(k) we have established that
00 t* hC
E(k) = '" -----.!.L HeD(E ) =: F(EhkC). (7)
(;:1 v75f k
338

Using the orthogonality of the Hermite polynomials, one may express the
hopping matrix elements in terms of the transformation function F(x):

tt = ~joo
2Jr D!
dEF(E)HeD(E)e-E2/2.
-00
(8)

Specializing on the case of a monotonic transformation function F(x) (with

derivative F'(x)), we can write

(9)

which leads to

(10)

Furthermore, the Fermi velocity Vk = \7 Ek can be computed:

_ F/(F- 1 ( )) he _ phe(v'2erf-l(2J~oodE/p(E/) -1)) he

Vk- EVk- p(E) Vk' (11)

A practical application of the general formalism proceeds as follows:

1. compute F- 1(E) from arbitrary target DOS p(E) using (10)
2. invert function (numerically or analytically) to obtain F(E)
3. evaluate transport properties, e.g., (IVkI2)(E) or p(E) using (11)
4. optionally determine microscopic model parameters tt using (8)
The only choice inherent in this procedure beyond the usual assumptions
for large dimensions is contained in step 1 which by construction produces
a monotonic transformation function F. The optical i-sum rule reads

roo dW(Txx(W) = (TO

Jo
(P/(E)) = (TO / [F"(F- 1(E)) _ F-1(E)F/(F-1(E))]) .
4d p(E) 4d \
(12)
Here, the first equality follows from (2), i.e., is generally valid within the
DMFT while the second expression in terms of the transformation is specific
to the formalism developed within this section.

Example: Flat-band System One interesting limiting case of a monotonic

transformation function which can be treated analytically is the step func-
tion F(x) = 28(x) -1 corresponding to a fiat band DOS of the form
p(E) = (b(E-1) + b(E+1))/2. For this case, the hopping matrix elements
read (trivially, t 2n = 0):

* ()n (2 (2n - I)!!

t 2n+ 1 = -1V-; v!(2n + I)! (13)
 339

a) 2 r-.------,--,------,------,--,------,------,--,--, b) 0.5 r-.------,--,------,----:7'T~-.---___,_______,-"

hypercubic I stacked - exact -
Millis I Freericks d=5 -----
this work - d=6 .
0.4
1.5 d=100

0.3
A

~1
V
0.2

0.5
0.1

-2 -1.5 -1 -0.5 0 0.5 1.5 -2 -1.5 -1 -0.5 0.5 1.5

Figure 2. a) The average squared Fermi velocity (Ivk 2)(E) is constant for the hypercubic
1

lattice (or for the x-component of a stacked lattice); in contrast it vanishes for the isotropic
lattice defined in this work. For comparison, the form suggested by Millis is also shown.
b) Resulting function jJ( E) of the full isotropic model (solid line) in comparison with
truncated models (D max = 3 or D max = 5), evaluated in finite dimensions 5 :S d :S 100.

The asymptotic exponent -3/4 is only slightly smaller than the threshold
value -1/2 required for a finite variance JC"X)CXJ du 2 p( f) = 2:;D tj}. For a
rectangular model DOS, t'D rv 2- n n- 3 / 4 already decays exponentially fast.

3. Application to the "Bethe" semi-elliptic DOS

In this section, we will apply the new formalism to the Bethe semi-elliptic
DOS in order to determine a corresponding tight-binding Hamiltonian
defined on the hypercubic lattice with the same local properties as the
Bethe lattice (with NN hopping) in the limit d ---+ 00. From (11), we derive
the average squared Fermi velocity defined in (3) in closed form:

(I Vk
2)() 27f exp [
f = ---2 -2(er f_l(fV1-f2/4+2arCSin(f/2))~2]. (14)
1
4-f 7f

Here, we have used the fact that (IVkI2)(f) is effectively constant (and equals
1 for unit variance and lattice spacing) in the hypercubic case. The result
(solid line in Fig. 2a) has all the qualitative features expected for this
observable in any finite dimension: (IVkI2)(f) is maximal near the band
center, strongly reduced for large (absolute) energies and vanishes at the
band edges: states at a (noninteracting) band edge do not contribute to
transport. The violation of this principle in the stacked case (dashed lines
in Fig. 2a), which corresponds to an application of the hc formalism to
the Bethe DOS with (IVkI2) constant up to the band edges, is clearly
340

U=4.0 - U=4.0 -
U=4.6 ----- U=4.6
U=50 - U=5.0
U=5.5 U=55
0.2 0.2

'8
b

0.1

Figure 3. Numerical results for the half-filled Hubbard model with semi-elliptic DOS for
00 in the paramagnetic phase at T = 0.05. a) Local spectral function A(w) obtained
d ---+
from QMC (using a discretization 6T = 0.1) and MEM. b) Optical conductivity u(w) for
w
the isotropic "redefined Bethe lattice". The inset shows the partial f-sum Jo dw' u(w').

pathological. Therefore, our method has not only the merit of yielding
isotropic transport, but also of avoiding unphysical behavior.
In order to determine the microscopic model, we have to apply (8) to
the numerically evaluated transformation function F. Again, the scaled
hopping matrix elements fall off exponentially fast: only a fraction 10- 3
of the total energy variance arises from hopping amplitudes beyond third
nearest neighbors and only a fraction 10- 6 results from hopping beyond gth_
nearest neighbors. This result suggests that properties of the model should
be robust with respect to truncation. In fact, p( E) (and consequently the
definition of CT( w)) hardly changes when hopping is cut off beyond 3rd or
5th nearest neighbors, even when evaluated in finite dimensions as seen
in Fig. 2b. This behavior is very general so that results for CT(W) of a
local theory in finite dimensions will depend on d predominantly via the
interacting DOS A(w) and only very little via p( E).
The local spectral functions for T = 0.05, i.e., slightly below the critical
temperature T* ~ 0.055, are shown in Fig. 3a as obtained from QMC /MEM
(5). In the metallic phase, the spectral density at the Fermi level (w = 0)
is approximately pinned at the noninteracting value p(O) = II-IT ~ 0.32
for U ;oS 4.4. The quasiparticle weight decreases drastically and a shoulder
develops for U 2: 4.6 before a gap opens for U 2: 4.8. An application of
(2) to these spectra for the isotropic model characterized by (14) yields
the estimates for the optical conductivity CT( w) shown in Fig. 3b. A low-
frequency Drude peak (of Lorentzian form) and a mid-infrared peak at
w ~ U /2 are present in the metallic phase and decay towards the metal-
insulator transition at U ~ 4.7. For large U, the optical spectral weight
 341

a) b) 0.5 ,..,------,_.------,------,_-,------,--------,------,
isotropic --e-
U=4.0 disordered - -A--
1.5
stacked ··0··
0.4 """
'-.....
G>·····B.~"'''',
8' 0.3
8'
>....,
. '"0-.>~.,.,
b
b 8
"0
.~ 0.2
-1[/2 Ekin
0.5 isotropic -
disordered - 0.1
stacked -

4 4.2 4.4 4.6 4.8 5.2 5.4 5.6

U

Figure 4. a) Optical conductivity a(w) for T = 0.05 and U = 4.0 for the new iso-
tropic model with semi-elliptic DOS in comparison with disordered and stacked models
consistent with the same DOS. b) Optical i-sum for T = 0.05.

concentrates in incoherent peaks at w ;::::: U. The inset of Fig. 3b shows the

partial optical f-sums. As expected, both the contribution of the Drude
peak and the total f-sum decrease for increasing U.
Figure 4a shows the impact of the definition for 0'( w) on the results
for U = 4.0. Our isotropic model yields by far the largest contributions
at small w. While the stacked model leads to otherwise similar results,
the low-frequency form of O'(w) is qualitatively different in the disordered
case (where a Drude peak is absent even for U ----7 0). As seen in Fig. 4b,
the f -sum is generally larger for the isotropic than for the stacked Bethe
lattice, in particular in the metallic phase (while the disordered case is
in-between). These differences can be attributed to the enhanced squared
Fermi velocity in the isotropic model: The enhancement is largest near the
Fermi surface, at E = 0 by a factor of 1r /2;::::: 1.57. A corresponding increase
is expected of the Drude peak for small enough U and T when transport
is dominated by states with E ;::::: O. Since energy eigenstates spread out in
momentum space at large U, the enhancement reduces (in general) to the
integral J~CXJ dE p( E) = L~=l D tj} which evaluates here to 1.05406.
In all three cases, the proportionality (4) of the f-sum to the kinetic
energy as characteristic for the hc lattice is clearly violated. This is true even
for the stacked case (which is otherwise similar to the hc case): while the f-
sum is here proportional to the contribution to the kinetic energy associated
with hopping in current direction (8), JoCXJdwO'xx(w) = -0'0 (Tx )/4, this
contribution (which is negligible in the limit Z ----7 CX)) is not proportional
to the total kinetic energy in this anisotropic case. A more relevant sum
t2 2
rule is derived from (12): JoCXJdwO'xx(w) = ~(4=~2)'
342

4. Conclusion

We have presented a new general method for constructing regular lat-

tice models with hypercubic (hc) symmetry, i.e. isotropic optical transport
properties, in large dimensions. Previously, calculations of the optical con-
ductivity O"(w) of the Hubbard model in the limit d -----+ CX) had been restricted
to the hypercubic lattice (using NCA (3) or QMC (10)) or have ignored the
lattice dependence: Applying the hc formalism to the Bethe semi-elliptic
DOS, Rozenberg et. al (11) overlooked violations of the hc f-sum rule.
All previous approaches specific to the Bethe DOS were associated with
anisotropic or incoherent transport; the most interesting of these (4, 6)
have not yet been linked rigorously to microscopic models.
Our method yields the first derivation for 0"( w) consistent with a semi-
elliptic DOS that implies isotropic transport which is fully coherent in the
noninteracting limit. This reinterpretation of the "Bethe lattice" (in the
DMFT sense) as an isotropic, regular and clean lattice and the demon-
stration that the associated transport properties are robust (with respect
to finite dimensionality or hopping range) removes, finally, the pathologies
previously associated with the DMFT treatment of transport in connection
with non-Gaussian DOSs. At essentially no additional cost, the method
can also be used for computing properties such as transverse conductivities
and thermopower; these vanish, however, in the particle-hole symmetric
case considered in this paper. Our numerical results have shown that the
precise definition of 0"( w) does matter, in particular within the metallic
phase where transport is potentially most coherent. We have also found a
general DMFT expression for the f-sum rule as well as a form specific to
our new approach.

Acknow ledgements

We gratefully acknowledge useful discussions with J. Freericks, D. Logan,

A. Millis, and D. Vollhardt.

References

1. J. Hubbard, Proc. Roy. Soc. London A276, 238 (1963); M. C. Gutzwiller, Phys.
Rev. Lett. 10, 59 (1963); J. Kanamori, Prog. Theor. Phys. 30, 275 (1963).
2. A. Khurana, Phys. Rev. Lett. 64, 1990 (1990).
3. T. Pruschke, D. L. Cox and M. Jarrell, Phys. Rev. B 47, 3553 (1993).
4. W. Chung and J. K. Freericks, Phys. Rev. B 57, 11955 (1998). J. K. Freericks,
private communication (2000, 2002).
5. N. Blumer, Ph.D. Thesis, Universitiit Augsburg (2002).
 343

6. A. Chattopadhyay, A. J. Millis and S. Das Sarma, Phys. Rev. B 61, 10738 (2000);
A. J. Millis, private communication (2002).
7. M. P. H. Stumpf, Ph.D. Thesis, University of Oxford (1999).
8. G. S. Uhrig and R Vlaming, J. Phys. Condo Matter 5, 2561 (1993).
9. V. Dobrosavljevic and G. Kotliar, Phys. Rev. Lett. 71, 3218 (1993).
10. M. Jarrell, J. K. Freericks and T. Pruschke, Phys. Rev. B 51,11704 (1995).
11. M. J. Rozenberg, G. Kotliar, H. Kajiiter, G. A. Thomas, D. H. Rapkine, J. M.
Honig, and P. Metcalf, Phys. Rev. Lett. 75, 105 (1995).
FROM CEIN 3 TO PUCOGA 5 : TRENDS IN HEAVY FERMION
SUPERCONDUCTIVITY

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

J.L. SARRAO and J.D. THOMPSON

Los Alamos National Laboratory, Los Alamos, NM 87545

Abstract. We review the serendipitous discovery of superconductivity in PuCoGas and its

relationship to CeMlns (M=Co, lr, Rh). Outstanding issues in our understanding of the
phase diagrams of CeMln,5 are also discussed. Finally, we speculate about the relationship
between pressure-induced superconductivity in Celna at 200 mK and ambient-pressure
superconductivity in PuCoGas at 18.5 K and the prospects for other high-temperature
unconventional superconductors.

Key words: PuCoGas, CeColns, Celna, Plutonium, Heavy Fermion Superconductivity

1. Crystal Chemistry

The connection between crystal structure and superconductivity has long

been recognized if not understood rigorously (1). This is also true of the
known heavy fermion superconductors, of which a large number crystalize
in the tetragonal ThCr2Sb crystal structure. The materials to be dis-
cussed here further confirm this trend. The CeJ\!IIn5 compounds as well
as PuCoGa5 crystalize in the tetragonal HoCoGa5 structure type (2, 3).
This structure is a layered derivative of the cubic CU3Au structure in which
units of HoGa3 alternate with layers of CoGa2. As such, CeIn3 and PuGa3
are the 'parent' compounds of the superconductors. In addition to these
materials, there are a large family of UMGa5 compounds reported; however,
none of these are known to superconduct (4). The extent to which this is
attributable to the properties of UGa3 compared to PuGa3 and CeIn3 is
an important outstanding question.

345
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 345–351.
© 2003 Kluwer Academic Publishers
346

2. Heavy fermion superconductivity in CeMln5

The discovery of superconductivity in CeMIn5 (M=Co,Ir,Rh) has increased

significantly the number of known heavy fermion superconductors, and
these materials are currently the subject of much experimental and the-
oretical effort. The three variants that have been reported - CeRhIn5 (5),
CeIrIn5 (6), and CeCoIn5 (2)- are isovalent, and each displays heavy Fer-
mion superconductivity. Here, we summarize the basic properties of these
materials and point out trends and relationships between and among them.
At ambient pressure, CeRhIn5 orders antiferromagnetically at 3.8 K.
TN increases weakly with applied pressure, an unconventional dependence,
until it vanishes near 16 kbar, at which point superconductivity, with aTe
of 2.1 K is observed (5). The superconductivity is bulk as evidenced by heat
capacity (7) and nuclear magnetic resonance (8, 9) measurements, and the
observed temperature dependencies are consistent with unconventional, d-
wave pairing. The value of CiT just above TN and T c is rv 400 mJ/molK 2
in both cases, confirming the heavy Fermion character of CeRhIn5.
CeIrIn5 displays bulk superconductivity at 0.4 K that develops out of
a state with r rv 700 mJ ImolK 2 . Again, heat capacity, thermal transport,
and nuclear magnetic resonance measurements provide strong evidence for
a nodal superconducting gap symmetry (10, 11, 12). An interesting wrinkle
regarding CeIrIn5 is that its resistivity vanishes near 1 K, well above the
bulk transition temperature. The nature of this presumably filamentary
state remains unclear, but it is robust as a function of applied magnetic
field and 'indicates' the value of bulk T c that can be approached with either
applied pressure (13) or chemical substitution (14).
With an ambient-pressure T c of 2.3 K, CeCoIn5 displays the highest
transition temperature of any heavy Fermion superconductor. There is also
strong evidence for CeCoIn5 being a d-wave superconductor (10, 12, 15).
Applied pressure can further increase T c to 2.6 K (at a pressure of 15
kbar) (16). The observed CIT of 300 mJ/molK 2 at T c is rather modest,
but the application of a magnetic field greater than HC2 reveals that r is not
fully developed at Tc(H=O), and, in fact, the field-induced normal state r rv
1000 mJ ImolK 2 is quite large and displays a non-Fermi-liquid temperature
dependence. The H-T phase diagram of CeCoIn5 is also anomalous in the
exceptionally large values of HC2 observed (rv 120 kOe H II [100], rv 50 kOe
H II [001]) and the observation of a first-order superconducting transition
in the H-T plane (17, 18).
An additional surprising feature of the CeMIn5 materials is the "ro-
bustness" of T c of intermediate alloys, e.g., CeMl-xM~In5 (M or M' =
Co, Ir, Rh) (19). Single crystals can be grown for the full range of x
for all M and M', and superconductivity is observed over wide ranges of
 347

Co 0.5 Rh 0.5 Ir 0.5 Co

Figure 1. Generalized doping phase diagram for CeMIn5. The left-hatched regions
indicate antiferromagnetic order (AFM), while the right-hatched regions indicated super-
conductivity (SC). Note the substantial regions of coexistence of both superconductivity
and antiferromagnetism. After (14, 19).

concentration. Thus, one is presented with the opportunity of exploring the

competition and interaction of diverse ground states, e.g., magnetism and
superconductivity (14, 20, 19). The range of x for which superconductivity
can be observed is remarkable, rv 0.7 for Rh-Ir alloys, and is rivaled only
by the cuprates in terms of the flexibility of chemical tuning (See Figure
1). This suggests that the doping may be "out of plane" in analogy with
the cuprates and emphasizes the layered nature of the crystal structure of
CeMIn5.
The role of crystallographic anisotropy and the extent to which CeMIn5
is a two-dimensional version of CeIn3 has been an issue since the discov-
ery of these materials (5, 21). The evolution of superconductivity with
doping reveals a surprising relationship between T c and cia, the ratio of
tetragonal lattice constants (19). Taking the full range of Ce(Rh,Ir,Co)In5
superconductors including stoichiometric compounds and doped alloys, one
finds a linear relationship between T c and cia that is so steep that a two
percent change in cia increases T c by a factor of five. The origin of such
a strong coupling between these quantities remains unclear. In a general
sense a correlation between two dimensionality and superconductivity is
expected for magnetically mediated superconductors (22); however, the
functional dependence here is remarkably strong. If this mechanism could
be generalized to other materials, the potential consequences are obviously
348

12 r----r---..,.-----,r------,

1(y~
o
o
8

o
6 o
o
4

0L...-_ _
0'
TN
o

~ ,&l,aMDA I".
.- TI".l".

...I'_-_ _.. . . _ _
c 1".4,
....L._ _......I

o 20 40 60 80

P (kbar)

Figure 2. Generalized pressure phase diagram for CeMIn5. The open(closed) circles
indicate the evolution of antiferromagnetism(superconductivity) for CeIn3. The open
and closed squares have the same meaning for CeRhIn51 with the pressure offset by
13 kbar from the measured data to account for chemical pressure effects. Similarly, the
triangles represent the evolution of superconductivity with pressure in CeCoIn5 including
an overall shift of 28 kbar relative to CeIn3. See (16, 23) for details.

enormous.
The extent to which proximity to a quantum critical point influences the
properties of CeMln5 has also been the topic of much discussion. Non-Fermi
liquid behavior has been observed in both CeColn5 and CelrIn5 (6, 2, 24),
and the extent to which magnetism and superconductivity are proximate
to each other would suggest the the presence of one or more quantum
critical points. Furthermore, it can be argued that CeColn5 is equivalent to
CeRhln5 under 16 kbar of pressure (16, 23). This supposition yields a phase
diagram (see Figure 2) that not only produces a smooth evolution of su-
perconducting transition temperature with pressure, Tc(P), but also yields
a normal state phase diagram quite reminiscent of the high-T c cuprates.
To take an alternative view of these data, one could interpret the gen-
eralized phase diagram for Ce(Rh,Ir,Co)In5 of Figure 1 as revealing a more
typical competition between magnetism and superconductivity. The ob-
served ground states can be rationalized in terms of the deleterious effects
of magnetism on T c convolved with the beneficial effects of increasing aniso-
tropy on T c' Further measurements of the spin fluctuation spectra of these
materials, as well as an understanding of their combined pressure-doping
phase diagram, are needed to clarify this issue.
Finally, the extent to which the f-electrons in CeMln5 are localized or
itinerant remains an issue. Reasonable agreement between Fermi surfaces
 349

deduced from band structure calculations that consider Ce's f-electrons to

be itinerant (25, 26, 27, 28) can be found; however, de Haas- van Alphen
studies of dilute alloys, at least for Cel-xLaxRhIn5, indicate localized f-
electrons (29). More generally, the observation of coexistence of magnetism
and superconductivity in Ce(Rh,Ir)In5 (11, 30) raises the issue of whether
a one-band or two-band model is required to explain the properties of
CeMIn5.

3. Superconductivity in PuCoGa5

Quite recently, superconductivity has been reported in PuCoGa5 at the

rather high T c of 18.5 K (3). As discussed above, PuCoGa5 is isostruc-
tural with CeMIn5' and in a simple sense can be viewed as beginning
with CeCoIn5' replacing In with isovalent Ga and replacing Ce with Pu.
Examining this simple logic in terms of the importance of dimensionality,
proximity to magnetism, and degree of f-electron localization is worthy of
discussion.
Dimensionality would seem to be the simplest to examine. The cia
ratio in PuCoGa5 is 1.603, somewhat smaller than that of, e.g., CeCoIn5
(1.638). Naively, this would suggest that even higher Tcs could be found
by tuning the crystallographic anisotropy of PuCoGa5. Uniaxial pressure
measurements would certainly be valuable in this regard; however, Onuki
et al. have argued (31) that the underlying electronic structure becomes in-
creasingly two dimensional in the series LaMIn5, CeMIn5, UMGa5 without
regard for cia ratio. Until Fermi surface measurements can be performed
on PuCoGa5, its degree of structural optimization will remain unknown.
Too few isostructural transuranic compounds are known to assess how
close PuCoGa5 is to ordered magnetism in a generalized phase diagram;
however, the known UMGa5 materials provide some insight, at least with
respect to the related issue of degree of f-electron localization. UMGa5
compounds have been reported for many transition metals (4). Only a small
number, UNiGa5 and UPtGa5 order magnetically, and even these do not
show local-moment susceptibility. The temperature-independent paramag-
netism displayed by UMGa5 indicates itinerant f-electron character. The
observed value of C IT at low temperature for UMGa5 is typically on the
order of 20 mJ/molK 2 , consistent with this observation. Further, there are
no reports of superconductivity in UMGa5. PuCoGa5, on the other hand,
displays both a somewhat enhanced value of r rv 80 mJ ImolK 2 and a local-
moment susceptibility consistent with trivalent Pu. Thus, at least at this
simplest level, the extent to which PuCoGa5 has more localized f-electron
character and therefore is closer to ordered magnetism is beneficial for
superconductivity.
350

In fact, PuCoGa5 would seem to be intermediate between CeCoIn5

and UI'vIGa5, and may represent a rather optimized balance of degree of
localization and proximity to ordered magnetism. If one accepts this spec-
ulation, then the two orders of magnitude difference in T c between CeIn3
(rv 200 mK) and PuCoGa5 (rv 20 K) may be just the simple optimization
of magnetically mediated superconductivity with dimensionality and band-
width (32, 22). Only further elaboration of the physical properties of other
transuranic materials will provide a definitive answer to this speculation.

Acknowledgements

We thank our many colleagues and collaborators, especially M. Nicklas,

P.G. Pagliuso, N.O. Moreno, L.A. Morales, and Z. Fisk, for important
contributions to this work. Work at Los Alamos was performed under the
auspices of the U.S. Department of Energy.

References

1. B. T. Matthias, Phys. Rev. 97, 75 (1995).

2. C. Petrovic, P. G. Pagliuso, JVI. F. Hundley, R. Movshovich, J. L. Sarrao, J. D.
Thompson, Z. Fisk and P. Monthoux, J. Phys. Condens. Matter 13, L337 (2001).
3. J. L. Sarrao, L. A. Morales, J. D. Thompson, B. L. Scott, G. R. Stewart, F. Wastin,
J. Rebizant, P. Boulet, E. Colineau and G. H. Lander, Nature 420, 297 (2002).
4. Yu. Grin, P. Rogl and K. Hiebl, J Less Common Met. 121, 497-505 (1986).
5. H. Hegger, C. Petrovic, E. G. Moshopoulou, M. F. Hundley, J. L. Sarrao, Z. Fisk
and J. D. Thompson, Phys. Rev. Lett. 84,4986 (2000).
6. C. Petrovic, P. G. Pagliuso, M. F. Hundley, R. Movshovich, J. L. Sarrao, J. D.
Thompson and Z. Fisk, Europhys. Lett. 53, 354 (2001).
7. R. A. Fisher, F. Bouquet, N. E. Phillips, M. F. Hundley, P. G. Pagliuso, J. L. Sarrao,
Z. Fisk and J. D. Thompson, submitted to Phys. Rev. Lett. (2001).
8. T. Mito, S. Kawasaki, G.Q. Zheng, Y. Kawasaki, K. Ishida, Y. Kitaoka, D. Aoki,
Y. Haga and Y. Onuki, Phys. Rev. B 63, 220507 (2001).
9. Y. Kohori, Y. Yamato, Y. Iwamoto, and T. Kohara, European Physical Journal B
18, 601 (2000).
10. R. Movshovich, M. Jaime, J. D. Thompson, C. Petrovic, Z. Fisk, P. G. Pagliuso and
J. L. Sarrao, Phys. Rev. Lett. 86, 5152 (2001).
11. G. Q. Zheng, K. Tanabe, T. Mito, S. Kawasaki, Y. Kitaoka, D. Aoki, Y. Haga and
Y. Onuki, Phys. Rev. Lett. 86, 4664 (2001).
12. Y. Kohori, Y. Yamato, Y. Iwamoto, T. Kohara, E. D. Bauer, M. B. Maple and J.
L. Sarrao, Phys. Rev. B 64, 134526 (2001).
13. R. Borth, E. Lengyel, P. G. Pagliuso, J. L. Sarrao, G. Sparn, F. Steglich and J. D.
Thompson, Physica B 312, 136 (2002).
14. P. G. Pagliuso, C. Petrovic, R. Movshovich, D. Hall, M. F. Hundley, J. L. Sarrao,
J. D. Thompson and Z. Fisk, Phys. Rev. B 64, 100503 (2001).
15. K. Izawa, H. Yamaguchi, Y. Matsuda, H. Shishido, R. Settai and Y. Onuki, Phys.
Rev. Lett. 87, 057002 (2001).
 351

16. M. Nicklas, R. Borth, E. Lengyel, P. G. Pagliuso, J. L. Sarrao, V. A. Sidorov, G.

Spam, F. Steglich and J. D. Thompson, J. Phys. Condens. Matter 13 (2001) L905.
17. T. P. Murphy, D. Hall, E. C. Palm, S. W. Tozer, C. Petrovic, Z. Fisk, R. G. Goodrich,
P. G. Pagliuso, J. L. Sarrao and J. D. Thompson, Phys. Rev. B 65, 100514 (2002).
18. A. Bianchi, R. Movshovich, N. Oeschler, P. Gegenwart, F. Steglich, J. D. Thompson,
P. G. Pagliuso and J. L. Sarrao, Phys. Rev. Lett. 89, 137002 (2002).
19. P. G. Pagliuso, R. Movshovich, A. D. Bianchi, M. Nicklas, N. O. Moreno, J. D.
Thompson, M. F. Hundley, J. L. Sarrao and Z. Fisk, Physica B 312, 129 (2002).
20. V. S. Zapf, E. J. Freeman, E. D. Bauer, J. Petricka, C. Sirvent, N. A. Frederick, R.
P. Dickey and M. B. Maple, Phys. Rev. B 65, 014506 (2001).
21. Y. N. Grin, Y. P. Yarmolyuk and E. 1. Giadyshevskii, Sov. Phys. Crystallogr.24,
137 (1979).
22. P. Monthoux and G. G. Lonzarich, cond-mat/0207556.
23. V. A. Sidorov, M. Nicklas, P. G. Pagliuso, J. L. Sarrao, Y. Bang, A. V. Balatsky
and J. D. Thompson, Phys. Rev. Lett. 89, 157004 (2002) .
24. J. S. Kim, J. Alwood, G. R. Stewart, J. L. Sarrao and J. D. Thompson, Phys. Rev.
B 64, 134524 (2001).
25. D. Hall, E. C. Palm, T. P. Murphy, S. W. Tozer, C. Petrovic, E. Miller-Ricci, L.
Peabody, C. Q. H. Li, U. Alver, R. G. Goodrich, J. L. Sarrao, P. G. Pagliuso, J. M.
Wills and Z. Fisk, Phys. Rev. B 64, 064506 (2001).
26. Y. Haga, Y. Inada, H. Harima, K. Oikawa, M. Murakawa, H. Nakawaki, Y. Tokiwa,
D. Aoki, H. Shishido, S. Ikeda, N. Watanabe and Y. Onuki, Phys. Rev. B 63, 060503
(2001).
27. D. Hall, E. C. Palm, T. P. Murphy, S. W. Tozer, Z. Fisk, U. Alver, R. G. Goodrich,
J. L. Sarrao, P. G. Pagliuso, and T. Ebihara, Phys. Rev. B 64, 212508 (2001).
28. R. Settai, H. Shishido, S. Ikeda, Y. Murakawa, M. Nakashima, D. Aoki, Y. Haga,
H. Harium and Y. Onuki, J. Phys. Condens. Matter 13, L627 (2001).
29. U. Alver, R. G. Goodrich, N. Harrison, D. Hall, E. C. Palm, T. P. Murphy, S. W.
Tozer, P. G. Pagliuso, N. O. Moreno, J. L. Sarrao and Z. Fisk, Phys. Rev. B 64,
180402 (2001).
30. A. D. Christianson, A. Llobet, et al., unpublished.
31. Y. Onuki, R. Settai, D. Aoki, H. Shishido, Y. Haga, Y. Tokiwa and H. Harium,
Physica B 312, 13 (2002).
32. N. D. Mathur, F. M. Grosche, S. R. Julian, 1. R. Walker, D. M. Freye, R. K. W.
Haselwimmer and G. G. Lonzarich, Nature 394, 39 (1998).
DO WE UNDERSTAND ELECTRON CORRELATION EFFECTS
IN GADOLINIUM BASED INTERMETALLIC COMPOUNDS?

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

E.V SAMPATHKUMARAN and R. MALLIKt

Tata Institute of Fundamental Research, Homi Bhabha Road,
Mumbai-400005, India

Abstract. Recognising the difficulties in systematic understanding of the physical charac-

teristics of strongly correlated f-electron systems, we considered it worthwhile to subject
the so-called "normal" f-electron systems like those of Gd to careful investigations. We
find that the spin-disorder contribution to electrical resistivity (p) in the paramagnetic
state, instead of remaining constant, apparently shows an increase as the temperature
(T) is lowered towards magnetic ordering temperature in some of the Gd alloys. In some
cases, the T -dependence of "excess resistance" is so large that a distinct minimum in the
plot of p versus T can be seen, mimicking the behaviour of Kondo lattices. This excess
resistance can be suppressed by the application of a magnetic field, naturally resulting
in large magnetoresistance. In addition, these alloys are found to exhibit heavy-fermion-
like heat-capacity behavior. These unusual findings imply hither-to-unexplored electron
correlation effects even in Gd-based alloys.

Key words: Gd Compounds, Electron Correlations, Kondo Effect, Heavy Fermions, Spin
Disorder, Electrical Resistivity, Kondo Lattice, Magnetic Polaron, Heat Capacity, Mag-
netoresistance

1. Introduction

The physical characteristics of the materials containing some of the f-

elements like ee, Eu, Yb and U, arising from strong electron correlation
effects, have been of great interest during last two decades in condensed
matter physics and the identification of new materials containing these ions
often resulted in discovery of novel phenomena. But, the fact remains that,
in spite of intense theoretical and experimental activities, there are con-

t PRESENT ADDRESS: DEPARTMENT OF DEVELOPMENTAL AND CELL

BIOLOGY, UNIVERSITY OF CALIFORNIA, IRVINE, USA.

353
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 353–361.
© 2003 Kluwer Academic Publishers
354

siderable difficulties in systematic understanding of these properties even

at a qualitative level. However, it has been recognised that the observed
anomalies are in some way related to the tendency of the f-orbital to get
delocalised, in other words, to the capability of these f-ions to exist in
more than one valence state. Naturally, the f-ions with deeply localised
f-orbital, like Gd, are expected to behave normally without any unusual
electron correlation effects known for the above-mentioned class of f-ions
and therefore careful investigations on Gd systems by and large do not
exist in the literature. During the course of our investigations of strongly
correlated electron systems (SCES), we came across some Gd alloys in
which the observations are so puzzling that this fundamental assumption in
this field is questionable. This article, surveying our experimental findings
with few examples, raises a question whether the understanding of the
electron correlation effects in "normal" rare-earths is truly complete.

2. Unusual results in ternary Gd compounds derived from A1B 2 -

derived hexagonal structure

A systematic investigation of Gd based alloys was actually motivated by

our original observation of heavy-fermion-like heat-capacity (C) anomalies
in Gd-substituted LaCu2Si2 (1), similar to that known for Ce alloys in
the low-coupling limit (2). Subsequently, we noticed that the electrical
resistivity (p) of GdNi 2Si 2 (3) and GdNi (4) over a wide temperature
(T) range above respective magnetic ordering temperatures (To) could be
suppressed to relatively smaller values by the application of a magnetic
field (H). In other words, these Gd compounds are found to exhibit negative
magnetoresistance (MR), the magnitude of which increases with decreasing
temperature. This finding implies that there is aT-dependent p-component
in excess of the lattice contribution present in these materials. While all
these features, new as far as Cd alloys are concerned, are typical of Kondo
alloys, no distinct minimum in the T dependence of p could be observed in
the raw experimental data. It is therefore of interest to look for Gd alloys
which can directly show this minimum in the raw data. In this respect, the
ternary Gd compounds, crystallizing in the AlB 2-derived hexagonal struc-
ture, provided an unique situation to demonstrate this aspect as well. While
the reader may refer to all our published articles for various measurements
and discussions (5, 6, 7, 8, 9, 10, 11, 12, 13, 14), here we summarise the key
experimental findings.
The behaviour of the compound Gd 2PdSi 3 (5) by far remains as the
best example to demonstrate our point. We have confirmed by various
measurements including 155Gd Mi::issbauer effect studies that the compound
Gd 2PdSi 3 undergoes long range antiferromagnetic ordering at (T N=) 21 K.
 355

09
sz
0
0
C'0
'-"
0.8
3.3

3.0 -
~
if]

'§
--
E
Q.

Q.
0.7

0.6
2.7
..0
@
'-"
2.4 Q.
0.5 '--~--'-~_'--~--'-~----' '--~----L_~----'-_~--'-----'
0.040 50 100 150 0 20 40 60

+
E r,. F/'\\T 20 g
u 0.03 I GdldSi3 K
16

I
~ 02'
,~"
_ 0.02 I-"~ " . .;d) 12 ]
~.,.,
j 0.01
N -......

............-.
(b)
... ....
>.. . . . .
8 0
Q.
.................. 0.8
-:---. .... ...• ".
-..;...:..,.-A•. 4
~ 0.00
~ ;.
" o
20 40 60 0 10 20 30 40 50
Temperature (K)
Figure 1. (a) The electrical resistivity (p) of the alloys, (Gd 1 - xY x,hPdSi3 , normalised to
respective 300 K values, as a function of temperature (T). (b) An estimate of "excess p"
as a function of T for Gd 2 PdSi 3 , obtained as described in the text. (c) The p as a function
of T for Gd 2 PdSi;J in the absence of a magnetic field and in the presence of 50 kOe. (d)
The 4f-contribution to heat-capacity as a function of T for the alloys, (Gd 1 - xY xhPdSi 3 .

As far as p(T) behaviour is concerned, what one would therefore expect is

that the T -coefficient of p is positive above 21 K. Looking at figure 1(a and
c), p(T) of Gd 2 PdSi 3 gradually decreases as the T is lowered down to 60
K, however followed by an upturn below about 45 K, thus giving rise to a
resistivity minimum at about (T min ) 45 K. The value of p before the onset
of magnetic ordering (say, at 22 K) is about 5% higher compared to that
at T min .
Such a behaviour of p with a distinct minimum in the paramagnetic state
was not previously observed for any "normal" (that means, with deeply
localised f-orbital) f-ion based intermetallic compound. [Further lowering of
T below TN does not result in a drop in p(T), as expected due to the loss
of spin-disorder contribution, and this finding is not relevant to the present
discussion; this is understandable in terms of opening of a pseudogap at
some portions of the Fermi surface due to the onset of antiferromagnetic
ordering]. Substitution of Y for Gd results in a depression of TN as ex-
pected, but the minimum in p(T) persists (see Fig. 1a). The fact that
the minimum is observable also in the sample with the highest Y content
(x= 0.8) excludes any explanation for the observed p minimum at 45 K
in Gd 2 PdSi3 in terms of the opening of an energy gap above 21 K. We
have also estimated the 4f-contribution (P4f) to P by subtracting the p of Y
analogue; for this purpose, we adjusted the slope of p(T) plot with that of
Gd 2 PdSi 3 at 100 K and then matching the p value at 100 K; though such
an analysis to subtract out lattice contribution is not free from ambiguities,
356

one gets an idea about T -dependence of" excess p" (P4f= p(T)-Plattice) from
the derived data (shown in Fig. 1b). Irrespective of the nature of reference
for lattice contribution, the temperature coefficient of P4f is found to be
distinctly negative in the paramagnetic state.
We have also measured P in the presence of an externally applied H.
For x= 0.0, it is found that MR, defined as [p(H)-p(O)]/ p(O), is about -
10% at 40 K in the presence of, say, H= 70 kOe, which is very large for a
normal intermetallic compound in this temperature range (5). Interestingly,
the minimum in p(T) vanishes in the presence of H (say at 50 kOe) and p
exhibits a positive T coefficient in the entire range (Fig. 1b). The magnitude
of MR keeps increasing with decreasing T for a given H. These features are
characteristics of Kondo alloys as well. It is to be noted (5) that a large MR
was not observed even for the analogous Kondo compound, Ce2PdSi3, which
emphasizes the importance of the finding in the Gd compound. Another
noteworthy finding is that the T at which MR starts taking noticeably large
values (with a negative sign) decreases with decreasing Gd concentration
(5), following the trend in TN. This implies that the observed anomalies
are magnetic in origin.
With respect to the C data (Fig. 1d), apart from the peaks attributable
to magnetic ordering at TN, the 4f contribution (C rn ) to C exhibits a tail
extending over a wide T range above TN. For instance, for x= 0.8, for
which TN is lower than 2 K, the tail in Crn(T) extends to T as high as 20
K, with the value of CIT as large as 1 J/mol K 2 at 2 K, comparable to the
behaviour of prototype heavy-fermion system, CeCu2Si2.
It is clear from the above results that the compound, Gd 2PdSh, exhibits
all the characteristics of (antiferromagnetically) ordering Kondo lattice
systems in p, MR and C data, including a resistivity minimum. These
features could be confirmed even in single crystals of this compound (8).
In addition, the thermopower also exhibits a large value at 300 K, typical
of Kondo systems (6). In order to explore whether these anomalies are
due to an artifact of spin-glass freezing setting in around 45 K, we have
probed the frequency-dependence of the ac susceptibility as well as pos-
sible irreversibilities in zero-field-cooled and field-cooled dc susceptibility
behaviour carefully (6, 9). The results obtained, apart from establishing
that the magnetic ordering below 21 K is of a long-range type, rule out
any type of magnetic ordering above 21 K. Therefore, a question arises
whether this compound could be the first Gd-based Kondo-lattice. Our
photoemission data on the other hand revealed (14) that there is no 4f-
derived feature at the Fermi level (Ep) and that 4f-level is placed about
8 eV well below Ep, thereby conclusively establishing that this compound
can not be categorised as a Kondo alloy. We will return to further discussion
on this" magnetic precursor effect" (T-dependent" excess" p and Crn above
 357

To) in Section 4.
Before closing this section, it is worth stating that two more isostruc-
tural Gd compounds, viz., Gd 2PtSi 3 (13) and Gd2CuGe3 (12), have been
subsequently identified to exhibit properties resembling those of Gd 2PdSi 3.
This suggests that the anomalies discussed here must be more common
among Gd intermetallics.

3. Behaviour of some other Gd compounds

The observation of unusual magnetic precursor effects in the ternary Gd

compounds mentioned above raises many questions: (i) Are these anomalies
specific to this crystal structure? (ii) Is it something to do with the layered
nature of Gd ions in this crystal structure? (iii) Is it due to polarization of
the d-band of some of the transition metal ions, like Pt or Pd, which are
prone for induction of moments by the large moment-carrying neighbouring
ions (like Gd)? (iv) Is this magnetic precursor effect anything to do with
the type of magnetic structure at lower temperatures? (v) Is there relation-
ship between C and p anomalies? In order to tackle these issues, we have
investigated a large number of Gd compounds (7) with different crystal
structures. The results clearly established that there is no straightforward
relationship between the observation of the" T -dependent excess p" on the
one hand, and the crystal structure or the type of transition metal and s-p
ions present in the compound, on the other. It was also established that
possible onset of magnetic correlations within a layer before long-range
magnetic order sets in can not be offered as the sole reason for excess
p, as many of the compounds with Gd layers are not characterized by
this behaviour of "excess p". Also, Gd compounds exhibiting this excess p
anomaly fall into the category of both ferromagnets and antiferromagnets
and therefore the nature of the magnetic structure below To is not the
determining factor.
A further careful look at the features above To in the p and C data
revealed (7) that the Gd alloys can be classified into three categories: (i)
Magnetic precursor effects appearing in both p and C; (ii) "T-dependent
excess p" and hence" MR anomaly" are negligible, but C tail persists over
a wide temperature range; (iii) Both "excess p" and "excess C" are absent.
The behaviour of the second category of Gd compounds is demonstrated
in figure 2 by showing the data for GdCu2Ge2, crystallizing in the layered
ThCr2Sb-type crystal structure, for which TN is about 13 K. It is obvious
from this figure that, apart from small magnitude, the sign of MR is not
even negative (unlike first category of Gd alloys) in the entire T range.
Figure 2. Electrical resistivity in zero magnetic field and in the presence of a mag-
netic field of 50 kOe and magnetic contribution (C 4 f) to heat-capacity, for GdCu2Ge2,
demonstrating that this compound belongs to Category-II as described in the text.

4. Discussion

The central experimental observation is that, in some Gd alloys, the spin-

disorder contribution to p in the paramagnetic state apparently is not a
constant value, but varies as though it increases gradually with decreasing
T as one approaches magnetic ordering temperature, resulting in Kondo-
lattice-like characteristics. At this juncture, it may be mentioned that sim-
ilar observations have been made also for EU2CuSi3 (11) and Tb 2PdSi 3 (17)
and therefore these anomalies must be more common among other rare-
earths as well with deeply localised f-orbitals. Since this feature does not
arise from the Kondo effect, the origin of this unusual magnetic precursor
effect remains a puzzle. One can not attribute it to Fisher-Langer type of
critical spin-fluctuations at To (15, 16), considering that the T-range over
which these features above To appear is too large to be explained by this
idea; in addition, the absence of a relationship between C and p anomaly
in second category of Gd compounds is convincing enough to rule out this
possibility. It therefore appears that the "T-dependent excess p" in first
category Gd compounds arises from a hitherto unrecognised phenomenon
in moment-carrying metals above To. We speculate that the formation
"magnetic polarons" (localisation or trapping of the conduction electron
cloud polarised by the s-f exchange interaction) above To could be the
root-cause of these anomalies. The proposed magnetic localisation could be
triggered by Gd 4f short range magnetic order and consequent spatial mag-
netic fluctuations, just as crystallographic disorder reduces the mobility of
charge carriers. We have observed (8) distinct anomalies in Hall coefficient
around 100 K in single crystals of Gd 2PdSi 3; also, on approaching magnetic
order, the linewidth and the resonance field in ESR measurements show a
minimum as a function of T (18). It is not clear at present whether this
 359

signals possible magnetic localisation effects. The application of a magnetic

field suppresses the spin fluctuations or alters the localization length result-
ing in an enhancement of the mobility of the electron cloud. This explains
the suppression of the p minimum in the paramagnetic state in a magnetic
field. Below T e, in ferromagnetic Gd compounds like GdNi (4), the polarons
are anyway itinerant as they are coupled ferromagnetic ally with the Gd 4f
moment; as a result polaronic contribution can be turned off below Te. It
must however be stated that the concept of magnetic polarons have been
generally ignored in metals in the literature (17) due to the fact that large
carrier density is expected to make these polarons delocalised. Possibly, in
poor metals, one may have to still invoke this concept. If this proposal is
confirmed in the compounds under discussion, one should explore various
factors determining the presence or the absence of such effects. Some of
the decisive factors could be the relative magnitudes of mean free path,
localization length and short range correlation length, in addition to the
strength of polarization of the conduction band and carrier density.
Another outcome of the studies on Gd alloys is the observation of negat-
ive MR even at temperatures far above To (till 3T o), peaking at To in some
Gd alloys (not presented here). These materials turned out to be one of the
first few intermetallics exhibiting giant magnetoresistance behavior in the
paramagnetic state at rather high temperatures. This finding established
that neither double-exchange nor Jahn-Teller effects (proposed for by-now
well-known "manganites") is crucial to observe GMR behavior. Similar
conclusions were drawn based on the observation of GMR behaviour in the
pyrochlore, TbMn207 (19, 20), for which the concept of magnetic polarons
has been discussed (21). It is to be stressed that Gd 2PdSi 3 turned out to
be the first normal intermetallic compound exhibiting ThlVln207-like p(T)
behaviour.

5. Conclusion

The discussion presented in this article suffciently demonstrates that, Gd

intermetallics, which have been traditionally considered uninteresting from
the magnetism point of view, exhibit novel behaviour in some compounds
with rich features in their physical properties, even in the paramagnetic
state. Our results on Gd alloys bring out for the first time that the un-
derstanding of electron correlation effects even in such" normal" materials
is far from complete. We hope that our data on several Gd compounds
will motivate new theoretical approaches in magnetism. Whatever be the
microscopic origin of the anomalies, the results offer a message: One has
to be a bit cautious while attributing low-temperature upturns in p and
C in alloys of Ce, U or Yb non-Fermi liquid and Kondo effects, since
360

these features may also be as a precursor effect to the magnetic transitions

occuring at further lower temperatures. It is of interest to explore, both
theoretically and experimentally, whether magnetic polarons can give rise
to non-Fermi liquid phenomena.

Acknowledgements

One of us (E.V.S) would like to acknowledge M. Blanco, D. Eckert, 1. Das,

A. Handstein, J. Hemberger, C. Laubschat, A. Loidl, Subham Majumdar,
S.L. Molodtsov, K.-H. Muller, V. Nagaraj an, P.L. Paulose, S.R. Saha, H.
Sato, H. Sugawara, and G. Wortmann for fruitful collaborations.

References

1. E.V. Sampathkumaran and 1. Das, Phys. Rev. B, 51, 8178 (1995).

2. W.P. Beyermann et aI, Phys. Rev. Lett. 66, 3289 (1991); W.P. Beyermann et aI,
Phys. Rev. B, 43, 13130 (1991).
3. E.V. Sampathkumaran and 1. Das, Phys. Rev. B, 51, 8631 (1995); Physica B 223-
224, 149 (1996)
4. R. Mallik, E.V. Sampathkumaran, P.L. Paulose and V. Nagaraj an, Phys. Rev. B,
55, R8650 (1997).
5. R. Mallik, E.V. Sampathkumaran, M. Strecker and G. Wortmann, Europhys. Lett.
41, 315 (1998).
6. R. Mallik, E.V. Sampathkumaran, P.L. Paulose, H. Sugawara and H. Sato, Pramana
- J. Phys. 51, 505 (1998).
7. R. Mallik and E.V. Sampathkumaran, Phys. Rev. B 58, 9178 (1998).
8. S.R. Saha, H. Sugawara, T.D. Matsuda, H. Sato, R. Mallik and E.V. Sampathku-
maran, Phys. Rev. B 60, 12162 (1999).
9. Subham Majumdar, E.V. Sampathkumaran, D. Eckert, A. Handstein, K.-H. Muller,
S.R. Saha, H. Sugawara and H. Sato, J. Phys.: Condens. Matter. 11, L329 (1999).
10. Subham Majumdar, M. Mahesh Kumar and E.V. Sampathkumaran, J. Alloys and
Compounds, 288, 61 (1999).
11. Subham Majumdar, R. Mallik, E.V. Sampathkumaran, K. Rupprecht and G.
Wortmann, Phys. Rev. B 60, 6770 (1999).
12. Subham Majumdar and E.V. Sampathkumaran, Phys. Rev. B 61, 43 (2000).
13. Subham Majumdar, E.V. Sampathkumaran, M. Brando, J. Hemberger and A. Loidl,
J. Magn. Magn. Mater. 236,99 (2001).
14. A.N. Chaika, A.M. ronov, M. Busse, S.L. Molodtsov, Subham Majumdar, G. Behr,
E.V. Sampathkumaran, W. Schneider and C. Laubschat, Phys. Rev. B 64, 125121
(2001).
15. S.-K. Ma, Modern Theory of Critical phenomena Benjamin, London, 1976.
16. M.E. Fisher and J.S. Langer, Phys. Rev. Lett. 20, 665 (1968).
17. R. Mallik, E.V. Sampathkumaran and P.L. Paulose, Solid State Commun. 106,
169 (1998); see references cited herein for a proposal of "magnetic polarons" for
rare-earth-based semiconductors.
18. J. Deisenhofer, H.-A. Krug von Nidda, A. Loidl and E.V. Sampathkumaran, Solid
State Commun., in press; arXiv: Cond-mat/0211587.
 361

19. Y. Shimakawa, Y. Kubo and T. Manako, Nature (London) 379, 53 (1996).

20. M.A. Subramanian, B.H. Toby, A.P. Ramirez, W.J. Marshall, A.W. Sleight, and
C.H. Kwei, Science 273, 81 (1996).
21. P. Majumdar and P.B. Littlewood, Nature 395, 479 (1998).
OPTICAL PROPERTIES OF CORRELATED SYSTEMS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

L. DECIORCI
Laboratorium fur Festkorperphysik, ETH Zurich, CH-8093 Zurich,
Switzerland

Abstract. We review selected optical results on correlated systems. This review mainly
focuses the attention on some prototype Kondo materials, characterized by the heavy
electron behaviour, and on the linear chain Bechgaard salts, as representative systems
of "real" one-dimensional materials. We will address a variety of relevant problems and
concepts associated with the physics of an interacting electron gas in three and low
dimensions, as Kondo behaviour, dimensionality crossover, and Luttinger liquid state.

Key words: Optical Properties, Correlated Systems, Heavy Electron Materials, Low
Dimensional Systems, Kondo Problem, Mott-Hubbard

1. Introduction

Highly correlated states of condensed matter have opened new chapters

in physics. Examples of particular interest include the so-called Kondo or
heavy-electron (HE) materials, which were discovered in the late seven-
ties. The class of strongly correlated systems also encounters the transition
metal oxides, including the d-electron (Mott-Hubbard) systems as well as
the high-temperature superconducting cuprates, the quasi-one dimensional
materials, as the organic Bechgaard salts or carbon-nanotubes, and possibly
the carbon-fullerenes. Due to space limitation, the present review will be
focused on selected examples of heavy-electron systems and on the quasi-
one dimensional Bechgaard salts, only.

HE systems are electrically conducting materials, with peculiar low

temperature physical properties distinguishing them from ordinary metals
(1, 2). In fact, the conduction electron specific heat is typically some 100
times larger than that found in most metals. Similarly, the magnetic sus-
ceptibility can be two or more orders of magnitude larger than the temper-

363
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 363–369.
© 2003 Kluwer Academic Publishers
364

ature independent Pauli susceptibility observed in conventional conducting

materials. The prototype HE materials include actinide and rare earth al-
loys, mostly containing U and Ge, respectively, like GeAl 3 , GeGu6, U Be13
and U Pt3 etc.
The name "heavy-electron" stems from the observation that enorm-
ously large specific heat coefficient I implies an effective mass m* of the
quasiparticles which is several hundred times the free electron mass. The
"heavy-electron" can be considered as the 4f or 5f electron interacting with
the bands of the delocalized electrons (d or s). Such an interaction occurs
via the Kondo spin fluctuations (3, 4). The final result is the formation of
the sharp and narrow Abrikosov-Suhl (ASR) resonance (width of about TK
for T << TK, TK being the Kondo temperature) in the density of states
N(E) at Ep (3). The electrons sitting in this narrow ASR are the new
heavy quasiparticles , responsible for the low temperature transport and
thermodynamic properties.

Strictly one-dimensional (ID) interacting electron gas systems are of

interest because the Fermi liquid (FL) state is replaced by a state where
interactions play a crucial role, and which is generally referred to as a
Luttinger liquid (LL). The ID state predicted by the LL theory (5, 6, 7)
is characterized by features such as spin-charge separation and the absence
of a sharp edge in the momentum distribution function n(k) at the Fermi
wave vector kp. The first immediate consequence of the absence of the
discontinuity at kp in the momentum distribution function is the power-
law behaviour of the density of states (DOS) p(w) rvl w la(w=Ep - E).
The exponent a in this expression reflects the nature and strength of the
interaction. The non-Fermi liquid nature of the LL is also manifested by
the absence of single-electron-like quasiparticles and by the non-universal
decay of the various correlation functions.
Nature has recently provided us with remarkable materials, which have
considerably boosted the experimental and theoretical understanding of
electrons with a dimensionality less than three. Low dimensional organic
conductors (like, e.g., the quasi one-dimensional Bechgaard salt chain or
the so-called quasi two-dimensional BEDT systems) can be described as
having units of strongly linked molecules, which are chains and planes for
one-dimensional (ID) and two-dimensional (2D) conductors, respectively,
with weak interactions (i.e., t~J between the units. They thus form a three-
dimensional lattice solid, but the electrons are confined either along one
direction or within a plane. Nevertheless, at low temperatures, when kBT is
smaller than t~, one should rather consider the low-dimensional conductor
as a three-dimensional anisotropic conductor (8, 9). One can readily see
that dimensionality crossovers should exist in real solids.
 365

(a) (a)

Figure 1. (left) (a) Real part O"l(W) of the optical conductivity obtained from the
Kramers-Kronig analysis of the absorptivity of CeAI 3 , and directly from the surface im-
pedance measurements (Z,,). (b) An enlargement of the dynamical conductivity showing
the development of the low-frequency resonance at low temperatures (10, 11). (right)
Sketch of the calculated frequency dependent conductivity O"I(w) after Millis and Lee
(12).

2. Optical response in heavy electron systems

Figure 1 (10) shows the real part of the optical conductivity 0"1 (w) of CeAl 3 ,
a prototype HE compound. The general trend of the optical conductivity
is rather common in all HE materials (11). In fact, at high temperatures
we can easily recognize a rather broad metallic-like behaviour (i.e., Drude),
indicative of a scattering rate r rv liT (i.e., the width of the Drude term)
of the order of 0.1 eV. Moreover, the conductivity drops below its dc value
at frequencies of the order of 1 IT and is starting to flatten again above
3000 em-I. In contrast, at relatively low temperatures we observe the
gradual emergence of a narrow mode centered at zero frequency in O"l(W),
which merges into the high temperature 0"1 (w) at FIR frequencies. This
dramatic temperature dependence does, nevertheless, preserve the good
agreement between the O"dc values and the zero energy limit of 0"1 (w).
366

The development of the low frequency narrow mode bears a striking

similarity with the theoretical prediction of O"l(W) (see Fig. 2) by Millis and
Lee (12). Millis and Lee (12) developed the most comprehensive theoretical
calculation of the HE electrodynamic response. They performed a study of
the low-temperature properties of the lattice Anderson Hamiltonian, in the
Kondo limit, where nearly free electrons hybridize with a correlated band
of f-electrons. The model is believed to contain the essential physics of the
currently interesting "heavy electron" metals: a non-magnetic ground state
that behaves as a Fermi liquid with a large effective mass. It turns out
that the narrow low-frequency part of the spectrum at low temperatures is
very much reminiscent of a Drude term with renormalized scattering rate
(11, 12). In fact, in the spirit of the Lorentz-Drude model (13), one can
rewrite the Drude term with the plasma frequency (w;? = w~(mblm*) and
the scattering relaxation time T* = Tm*1mb, which correspond to the bare
quantities wp and T being renormalized by the enhanced effective mass m*
of the heavy quasiparticles at low temperatures. One can easily realize that
spectral weight argument (i.e., the integral of the narrow (renormalized)
Drude resonance in 0"1 (w)) leads directly to the estimation of the effective
mass of the heavy quasiparticles (11).

Another major issue and current topic of debate in the field of highly-
correlated electron systems is the fundamental question whether these sys-
tems, in their normal state, may be described as simple Fermi liquids. For
some HE compounds (e.g., U Pt3) this seems indeed to be the appropri-
ate picture. More recent experimental work (11) has indicated, however,
that several HE compounds and related alloys display quite remarkable
properties, which may be much less well related with" conventional" Fermi
liquid behaviour. We refer to our review (11) for a broader experimental
and theoretical perspective on these issues.

3. Optical response of Bechgaard salts

Figure 3 displays the temperature dependence of the real part 0"1 (w) of the
optical conductivity in different Bechgaard salts for Ella, the chain direction
(14, 15, 16, 17, 18, 19). The optical conductivity of the TMTTF salts
displays several absorption features mostly ascribed to lattice vibrations
(phonon modes), mainly due to the inter and intramolecular vibrations
of the TMTTF unit. The optical properties of the (TMTSF)2PF6 ana-
logs (Fig. 3b), for which the dc conductivity gives evidence for metallic
behaviour down to low temperatures, are markedly different from those
of a simple metal. A well-defined absorption feature around 25 me V and
a zero-frequency mode (15, 16, 17, 18) are observed at low temperatures.
 367

10
'
1000 (TMITF),Br -10K
-··90K
Ella - - 300K
Mott insulator
} 500 ~
~
"

i"j---
o ,,-;-,,-~-"-;,...-..-~'-;:-"-~-"-;,...-..-=-~
10-4 10- 3 10- 2 10- 1
co
Photon Energy (eV)

Figure 2. (left) On chain optical conductivity of (a) (TMTTFhBr, and (b)

(T J'vlT S FhP F6 at temperatures above the transitions to the broken symmetry ground
states. The arrows indicate the (charge correlation) gaps (E gap ) observed by dielectric (0)
response and photoemission (ph). A simple Drude component is also shown in part (b).
Note that photoemission measures the quantity E gap /2, assuming that the Fermi level
is in the middle of the gap (14, 15). (right) Optical conductivity of the Mott insulator
and of the doped Mott semiconductor (20). The simple Drude behaviour is shown for
comparison.

This latter mode is responsible for the large metallic conductivity. Figure
3 also displays the temperature evolution of O"l(W) (14).
There is an ample theoretical literature, which is hopeless to review here
in great detail (14, 15, 17). As introduction to the theoretical expectation for
the electrodynamic response in one dimension, we briefly sketch the optical
response as suggested by Ref. (20,21), which catches the essential features.
Due to the commensurate filling, a strictly one-dimensional Luttinger liquid
with Umklapp scattering effects transforms into a Mott insulating state,
which is dominated by the charge correlation gap excitation. For a strict
one-dimensional Mott insulator, the charge correlation gap E gap , corres-
ponding to the excitation between the lower and upper Hubbard bands,
appears as a singularity in the real part 0"1 (w) of the optical conductivity
(Fig. 4). If the interchain coupling t.d rv tb) between chains is relevant (i.e.,
t.1. > E gap ), small deviations from commensurate filling due to the warping
of the Fermi surface exist, and should lead to effects equivalent to real
doping on a single chain. Then, it is energetically favorable for the charge
carriers to hop between parallel chains (20, 21). The scenario calculated for
a doped one-dimensional Mott semiconductor consists of a Mott (pseudo)
gap (as reminder of the originallD limit with Umklapp scattering process)
and a zero-energy mode (i.e., theoretically a D!5(w) function at w = 0,
representing the Drude resonance of the effective metallic contribution with
368

scattering rate r = 0) for small doping levels (Fig. 4). The spectral weight D
encountered in the" Drude resonance" is proportional to the effective charge
doping, induced by the interchain coupling t~, in the upper Hubbard band.
Therefore, by increasing t~ there is a dimensionality crossover, which is
accompanied by the evolution from a one-dimensional Mott insulator to a
"doped" semiconductor. Indeed, the (Tl\IITTFhX salts, with X = P F 6 or
BT, are insulators at low temperatures (8) with a substantial (Mott) charge
gap (see arrows for different experimental estimations of E gap in Fig. 3a).
For these compounds the correlation gap E gap is so large that the interchain
hopping (t~) is not relevant. Charge carrier hopping on parallel chains is
here strongly suppressed leading to a truly ID insulating phase (20, 21). In
analogy to the TMTTF salts, the strong FIR excitation of the TMTSF
salts (Fig. 3b) is ascribed, within the scenario depicted in Fig. 4, to the
so-called pseudo charge correlation gap. Additionally, there is the narrow
Drude peak, originating from deviations of commensurability due to t~.
The existence of a gap or pseudogap in the charge excitations (Fig. 3) with
the absence of a gap for spin excitations indicates, moreover, spin-charge
separation in the metallic state (14, 18).
Furthermore, at frequencies greater than t ~, the interchain electron
transfer is irrelevant and calculations based on the ID Hubbard model
should be appropriate (20, 21). The theoretical expectation (Fig. 4) consists
in a powerlaw of the frequency-dependent optical conductivity (71 (w)rvw-'
for frequencies greater than t~ and E gap but less than the on-chain band-
width 4t a . The theory also predicts that the exponent r = 5 - 4n 2 K p , K p
being the so-called Luttinger liquid parameter and n the degree of commen-
surability (20, 21, 22). Our results on the powerlaw behaviour (15, 18, 23)
are very robust and allow us to discriminate among different regimes and
type of correlations. This issue is thoroughly reviewed elsewhere (14).

4. Conclusion

This short review only touched a few issues and problems, which recently
attracted a lot of interest in strongly correlated systems. Signatures of
heavy quasiparticles in Kondo materials and of the Luttinger phenomeno-
logy in quasi-ID systems have been observed. However, the interpretation
of the spectra remains problematic on various issues and await further
experimental and theoretical work.

5. Acknowledgements

Research at ETH Zurich was supported by the Swiss National Foundation

for the Scientific Research.
 369

References

1. H.R. Ott and Z. Fisk, in Handbook on the Physics and Chemistry of the Actinides,
Eds. A.J. Freeman and G.H. Lander (Elsevier, Amsterdam, 1987), p.85.
2. Z. Fisk, D.W. Hess, C.J. Pethick, D. Pines, J.L. Smith, J.D. Thompson and J.O.
Willis, Science 239, 33 (1988).
3. N.B. Brandt and V.V. Moshchalkov, Adv. Phys. 33, 373 (1984).
4. P. Fulde, J. Phys. F.: Met. Phys. 18, 601 (1988).
5. J. Luttinger, J. Math. Phys. 4, 1154 (1963).
6. S. Tomonaga, Prog. Theor. Phys. 5, 554 (1950).
7. for a review, see H.J. Schulz, Int. J. Mod. Phys. B 5, 57 (1991).
8. D. Jerome and H.J. Schulz, Adv. Phys. 31, 299 (1982).
9. D. Jerome, in Strongly interacting fermions and high Tc superconductivity, Eds. B.
Doucot and J. Zinn-Justin (Elsevier Science, 1995), p. 249.
10. A.M. Awasthi et al., Phys. Rev. B 48, 10692 (1993).
11. for a review, see L. Degiorgi, Rev. Mod. Phys. 71, 687 (1999).
12. A.J. Millis and P. A. Lee, Phys. Rev. B 35, 3394 (1987).
13. F. Wooten, in Optical Properties of Solids, (Academic Press, New York, 1972).
14. for a review, see L. Degiorgi, in Strong interactions in low dimensions, Eds. D.
Baeriswyl and L. Degiorgi (Kluwer, 2003).
15. V. Vescoli et al., Em. Phys. J. B 13, 503 (2000).
16. M. Dressel et al., Phys. Rev. Lett. 77, 398 (1996).
17. A. Schwartz et al., Phys. Rev. B 58, 1261 (1998).
18. V. Vescoli et al., Science 281, 1181 (1998).
19. V. Vescoli et al., Solid State Commun. 111,507 (1999).
20. T. Giamarchi, Physica B 230-232, 975 (1997).
21. S. Biermann et al., cond-mat/0201542.
22. J. Voit, Em. Phys. J. B 5, 505 (1998) and Proceedings of the NATO Advanced
Research Workshop on The Physics and Mathematical Physics of the Hubbard Model,
San Sebastian, October 3-8, 1993, Eds. D. Baeriswyl, D. K. Campbell, J. M. P.
Carmelo, F. Guinea, and E. Louis (Plenum Press, New York, 1995).
23. F. Zwick et al., Solid State Commun. 113,179 (2000).
QUASIPARTICLE UNDRESSING: A NEW ROUTE TO COLLECTIVE
EFFECTS IN SOLIDS

Proceedings of the ARW NATO Workshop Hvar, Croatia, October 2002

J.E. HIRSCH
Department of Physics, University of California, San Diego
La Jolla, CA 92093-0319

Abstract. The carriers of electric current in a metal are quasiparticles dressed by electron-
electron interactions, which have a larger effective mass m* and a smaller quasiparticle
weight z than non-interacting carriers. If the momentum dependence of the self-energy
can be neglected, the effective mass enhancement and quasiparticle weight of quasi-
particles at the Fermi energy are simply related by z = m/m* (m=bare mass). We
propose that both superconductivity and ferromagnetism in metals are driven by qua-
siparticle 'undressing', i.e., that the correlations between quasiparticles that give rise to
the collective state are associated with an increase in z and a corresponding decrease in
m* of the carriers. Undressing gives rise to lowering of kinetic energy, which provides the
condensation energy for the collective state. In contrast, in conventional descriptions of
superconductivity and ferromagnetism the transitions to these collective states result in
increase in kinetic energy of the carriers and are driven by lowering of potential energy
and exchange energy respectively.

Key words: Quasiparticle, Dressing, Undressing, Quasiparticle Weight, Effective Mass,

Holes, Optical Sum Rule, Dynamic Hubbard Model

1. Particles and quasiparticles

Quasiparticles are 'dressed' bare particles, and they have a smaller quasi-
particle weight and a larger effective mass than bare particles. In several
superconductors and ferromagnets of current interest there is experimental
evidence that quasiparticles 'undress', and resemble more free particles,
when correlations build up and the system orders. Associated with this,
that the kinetic energy, that is supposed to be optimal in the Fermi liquid
normal state, decreases rather than increases in the ordered state. This
behavior is counterintuitive, since in a normal Fermi liquid description it
is expected that quasiparticles should become further dressed and less like

371
A.C. Hewson and V. Zlatić (eds.), Concepts in Electron Correlation, 371–380.
© 2003 Kluwer Academic Publishers
372

free particles when they develop the correlations leading to the collective
state, and that they should pay, rather than gain, kinetic energy. Several
'unconventional' theories have been proposed to explain these phenomena.
Instead we propose here that in fact quasiparticle undressing is a unifying
concept that can describe these collective effect in both new and conven-
tional materials. The only difference is that only in the newer materials
is the 'undressing phenomenology' strong enough that it is easily seen in
experiments.

2. Dressing and undressing

The single particle Green's function for an electron in an electronic energy

band is
1
G(k, w) = --------:-------,- (1)
W - Ek - 'L,(k,w)

Expanding the self-energy 'L, (k, w) around w = 0

0'L,
'L,(k,w) = 'L,(k, 0) + wOw (2)

the Green's function can be written as

G(k,w) = ~ + G/(k,w) (3)

w - Ek

where the first term is the quasiparticle part and the second term is the
incoherent part, and the 'quasiparticle weight' Zk is given by

(4)

If the momentum dependence of the self-energy can be neglected,

'L,(k,w) rv'L,(w) (5)

then we have simply Zk = Z and Ek = ZEk which implies that the quasi-
particle weight and effective mass m* are simply related by
m
-=Z (6)
m*
hence a highly dressed particle will have a small quasiparticle weight and
a large effective mass. For the models of interest in this paper the 'local
approximation' Eq. (5) is reasonable. Finally, the 'kinetic energy' of the
system is defined by
(7)
 373

where the occupation number nk is obtained from the single particle Green's
function
00 1
nk = -00 dwf(w)(-;ImG(k,w)) (8)
/

with f the Fermi function. The discontinuity of nk at the Fermi surface is

the quasiparticle weight Zk. If Zk was to increase for example as the tem-
perature is lowered, Eq. (7) predicts that the kinetic energy of the system
would decrease. The existence of this general relation between quasiparticle
weight increase and kinetic energy lowering was pointed out by Norman et
al (1).
In the presence of interactions bare particles become dressed with qua-
siparticle weight Z and effective mass m* rv 1/ z. The residual weak inter-
actions between these quasiparticles may cause the system to undergo a
transition to a collective state. In this paper we discuss a description of
superconductivity and ferromagnetism that predicts that when the collect-
ive state develops quasiparticles undress, namely, that z increases and m*
decreases. The resulting kinetic energy lowering provides the condensation
energy stabilizing the ordered state.

3. Low energy effective Hamiltonians

The low energy effective Hamiltonians in our theory can be derived from
the single band generalized Hubbard model

H = - L tij(ct·Cjo- + h.c.) + L (ijII/rlkl)cIo-c}o-ICzo-ICko- (9)

<ij>o- ijklo-o-'
where (ijII/rlkl) are matrix elements of the Coulomb interaction involving
Wannier orbitals at sites i, j, k, l. Keeping only two center integrals, we have
shown that the off-diagonal interactions (iiII/rlij) == (,6.t)ij and (ijII/rlji) =
(iiII/rljj) == Jij lead to simple descriptions of superconductivity (2) and
ferromagnetism (3) respectively, that have many features in common with
the phenomena seen in real materials. It should be pointed out however
that these off-diagonal interactions are not simply calculable by comput-
ing matrix elements of the Coulomb interaction between fixed Wannier
orbitals (4).
The Hamiltonian that gives rise to (hole) superconductivity is

H sc = Hex Hub + L (,6.t)ij(ni,-o- + nj,-o-)[cIo-Cjo- + h.c.] (10)

<ij>o-
and the one giving rise to ferromagnetism is

(11)
374

The 'extended Hubbard' Hamiltonian HexHub includes the kinetic energy

in Eq. (9) and the ordinary density-density Coulomb interactions U =
(iilljrlii), Vij = (ijlljrlij). The 'bond charge repulsion' term in Eq. (11)
describes both exchange and pair hopping processes (3), arising from the
matrix elements (ijlljrlji) and (iilljrljj) respectively.
These 'off-diagonal' interactions lead to a decrease of the effective mass
and associated with it a decrease in the kinetic energy as the collective
states develop. For ferromagnetism, the effective hopping for a carrier of
spin (J" is
t
tij - tij - Jij < Ci,_aCj,-a + h.c. >
eff _
(12)
and it increases when spin polarization develops because the bond charge
C!
< _aCj,-a > decreases. For superconductivity the effective hopping for
an electron is
eff _ (A )
tij - tij - n Dot ij (13)
and it decreases monotonically as the number of electrons in the band
increases. For holes instead the effective hopping amplitude is
eff _
tij -
h
tij + nh (ADot )ij (14a)
with
t7j = tij - 2(~t)ij (14b)
and it increases as the hole concentration nh increases. Because the local
hole concentration around a given hole increases when holes pair, the ef-
fective hopping Eq. (14a) will increase when pairing occurs (6).
These Hamiltonians describe changes in the quasiparticle effective mass
when the system enters the collective state, and also as function of doping
in the normal state. However they do not properly describe the expected
relation between effective mass and quasiparticle weight discussed in the
previous section. This is most clearly seen for the effect of ~t in the normal
state. According to the model Hamiltonian Eq. (10), the effective mass of a
single hole in a filled band can be much larger than that of a single electron
in an empty band, their ratio is
rn 'wle t
(15)
rn*electron t- 2~t

On the other hand, the quasiparticle weights for the single electron and the
single hole in this single band model are simply Zel = Zhole = 1, hence the
expected relation between quasiparticle weight and effective mass Eq. (6)
is strongly violated.
In addition, the optical conductivity sum rule is violated. The integral
of the optical conductivity in a tight binding model is given by

1 00

dW(J"l(W) rv< -nin > (16)

 375

(proportionality factors are omitted). For the Hamiltonian Eq. (10), the
average kinetic energy for holes from the ij bond is

< Tkin >ij= -(t~ + nh(~t)ij) < c!a.Cja > + < T kin >ij (17a)

< T kin >ij= -(~t)ij[< cl,-acla >< CjaCi,-a > + < c},_acla >< CjaCj,-a >]
(17b)
and it decreases below T c as the anomalous expectation values in Eq. 17(b)
become nonzero. Hence the integrated optical spectral weight (left side of
Eq. (16)) increases. A similar situation occurs for ferromagnetism. However,
in a real system the total optical spectral weight is conserved (optical sum
rule), hence the optical sum rule is 'violated' if kinetic energy lowering
occurs. The resolution of both this violation and the unphysical relation
between m* and z arises from consideration of other degrees of freedom
not contained in the effective Hamiltonian Eq. (9).

4. Relation between particle and quasiparticle operators

The low energy effective Hamiltonian Eq. (9) should be understood as

describing the dynamics of quasiparticles rather than bare particles. For the
description of superconductivity (5) the quasiparticle operator (which we
denote by Cia) is related to the coherent part of the bare particle operator
Cia by
(18a)
with 0 < 8 :s; 1. Here, Cia is an electron operator. The corresponding
relation for hole operators is

Cia = 8[1 + lni,-a]cia (18b)

1
1= - - 1 (19)
8
In the kinetic energy operator for bare electrons,

H kin = - L tij [c!aCja + h.c.] (20)

<ij>a
upon replacement of the bare electron operator in terms of the quasiparticle
operators using Eq. (18), a 'correlated hopping' term of the form Eq. (10)
results. Hence we can identify the correlated hopping amplitude as

(21 )

The quasiparticle weight for holes in a filled band is Zh 8 2 . When Zh

is small, holes are heavily dressed in the normal state low hole concen-
tration regime, and the 'undressing parameter' 1 is large. When the hole
376

dressed _----'::....::...ce...::..-_[>
s. t. undressed
carrIers carrIers
,.. ~ ~ ~

f-

H
undressed
paIrs

n
Figure 1. Undressing phenomenology in the cuprates. Both when hole carriers pair
and when the hole concentration increases by doping, there is an increase in the local
hole density around a given carrier that gives rise to undressing. This is accompanied by
spectral weight transfer from high to low frequencies

concentration nh increases, the quasiparticle weight and the quasiparticle

bandwidth increase according to the relations
nh 2 (
z=zh(I+Y2) 22a)
nh
D=D h (l+y )2. (22b)
2
As a consequence, the system becomes increasingly more coherent as the
hole concentration increases.
The relation between bare particle and quasiparticle operator Eq. (18)
gives rise to hole superconductivity when the parameter Y is large, as well
as to the 'undressing' phenomenology observed in experiments whereby
quasiparticles undress both when the temperature decreases and the system
becomes superconducting and when the hole concentration is increased
in the normal state. This behavior qualitatively shown in Figure 1 was
predicted (7) based on microscopic considerations for the cuprates well
before it was observed experimentally.
For ferromagnetism, a parallel analysis results when in the relation
between bare and quasiparticle operators Eq. (18a) the site charge nia is
replaced by the bond charge. Replacement in the kinetic energy Eq. (20)
gives rise to exchange and pair hopping terms that give rise to ferromag-
netism and to undressing (increase in quasiparticle weight and decrease in
quasiparticle mass) as the system develops spin polarization (8).
 377

5. Dynamic Hubbard models

That the conventional single band Hubbard model is fundamentally flawed

is seen as follows: in the Hubbard model, destruction of an electron in a
doubly occupied orbital yields the single electron state, ie

cia-I Tl>= 1 - iJ > (23)

This relation implies that the doubly occupied state 1> is a single Slater 1 r
determinant. The very fact that electrons interact makes this an incor-
rect assumption. Hence the conventional Hubbard model fails to describe
the most basic aspect of the electronic correlation problem it purports to
embody, namely correlation of electrons in the same Wannier orbital.
Recognition of the fact that the doubly occupied orbital I 1> is a r
correlated state rather than a single Slater determinant leads to dynamic
Hubbard models (9). The correct form of Eq. (23) is

Cia-I Tl>= 1 - iJ > S+L 1 - iJ >n Sn (24)

n>O

where 1- iJ >n are excited state of the singly occupied orbital, and 1- iJ >::::::
1 - iJ (24) is
>n=O the ground state. Because electrons interact, S in Eq.
never unity, and the second term on the right side of Eq. (24) is never zero.
This leads to the relation between bare particle and quasiparticle operators
Eq. (18).
We have discussed various realizations of dynamic Hubbard models,
involving either an auxiliary boson degree of freedom at each site or more
than one orbital per site (10). A new energy scale enters, given by the
excitation energies of the states iJ >n. This is the energy range from
1 -

which the high frequency spectral weight gets transfered from. In dynamic
Hubbard models the Hubbard U becomes a dynamical variable, which can
take more than one value depending on the relative state of the two elec-
r
trons in the correlated state I 1>, and destruction of an electron in that
correlated state never yields the singly occupied state liJ > with its full
amplitude. The study of dynamic Hubbard models is only in its beginning
stages but it is clear already that they exhibit very rich physics absent in
the conventional Hubbard model.

6. Conclusions and summary

We are proposing that there is a single unifying concept behind the two most
common collective effects in metals, superconductivity and ferromagnetism:
quasiparticle undressing. Our proposal rests on four pillars, namely: (1)
378

Theoretical consistency, (2) Experimental evidence, (3) Microscopic justi-

fication, (4) Philosophical considerations. We summarize arguments in each
category in the following.

(1) Theoretical consistency

The theory is based on the relation beween bare particle and quasi-
particle operators,

ct,. = [1 - (1 - S)iil oea z]2t,. + incoherent part (25)

with S < 1. For the description of superconductivity and ferromagnetism

iiloeal is the site charge or the bond charge respectively (normalized to
unity). This relation gives rise to low energy effective Hamiltonians with
off-diagonal interaction terms b.t and J which drive hole superconductivity
and ferromagnetism respectively. Inclusion of both the local site and bond
charge in Eq. (25) will yield a description of both instabilities and their
competition with the same Hamiltonian.
As the collective state develops, the term iiloeal in Eq. (25) decreases: for
ferromagnetism, spin polarization reduces the bond charge density, and for
superconductivity hole pairing reduces the electronic site charge density. As
a consequence the coefficient of cL-
in the first term of Eq. (25) increases,
and the incoherent part correspondingly decreases. This leads to an increase
in the quasiparticle weight and a decrease in the quasiparticle mass, as well
as lowering of kinetic energy. Spectral weight in the optical conductivity is
transfered to low frequencies from the high frequency processes giving rise
to the incoherent terms in Eq. (25). This framework then maintains the
simple relation between quasiparticle weight and quasiparticle mass expec-
ted on general grounds, as well as explains the optical sum rule violation
that occurs if only the low energy effective Hamiltonians are considered.
These low energy effective Hamiltonians are seen to give rise to ferromag-
netism and superconductivity both within mean field theory as well as in
exact treatments.
For superconductivity, the theory predicts that it cannot occur when
only electron carriers exist at the Fermi energy, because electron carriers
are already undressed and will not undress further by pairing. It can only
occur when hole carriers exist at the Fermi energy because holes are dressed.
When the local hole concentration increases either by pairing or by hole
doping in the normal state, holes undress and turn into electrons. Pairing
of holes is especially favored when holes propagate in negatively charged
structures, since in that case the dressing of the hole is highest and the
undressing associated with hole pairing is strongest (4, 9).
 379

(2) Experimental evidence

Certain ferromagnets exhibit clear evidence of undressing in optical
spectroscopy: manganites (11), EuB 6 (12), TlMn207 (13), and some fer-
romagnetic semiconductors (14). Furthermore the anomalous lowering of
resistivity below T c and the negative magnetoresistance observed in all
ferromagnets may be interpreted as originating in lowering of effective mass
upon spin polarization.The undressing in optical spectroscopy may be too
weak in certain metallic ferromagnets to be directly observed. The undress-
ing has not yet been seen in photoemission experiments in the manganites,
but we expect that it will be seen in the future. The reduction in the bond
charge density associated with spin polarization that the theory requires
(see the previous subsection) manifests itself in the anomalous thermal
expansion behavior of ferromagnets below T c (15).
High T c cuprates exhibit abundant experimental evidence that quasi-
particles undress when they pair, in photoemission (16) and optical spectro-
scopy (17, 18). Furthermore, optical (19) and photoemission (20) spectro-
scopy as well as transport (21) show that holes undress in the normal state
when the hole concentration increases by doping. The fact that by pairing
dressed holes turn into undressed electrons is seen directly in experiments
in both conventional as well as high T c superconductors (22, 23, 24) . The
fact that superconductors are frequently prone to lattice instabilities is
naturally explained by the fact that many anti bonding electronic states
need to be occupied in order for the charge carriers to be hole-like. The
observed isotope effects and phonon signatures in tunneling are expected
to result from the fact that ionic displacements will affect the magnitude of
the interaction flt that drives superconductivity. The empirical observation
that superconductors have hole carriers in the normal state was made by
Chapnik (25).

(3) Microscopic justification

Analysis of the physics of electrons in atomic orbitals shows that the
basic relation Eq. (18) is valid, with the parameter S decreasing as the net
ionic charge Z decreases (9). Also, first principles calculation of hopping
amplitudes in simple diatomic molecules show that indeed the hopping
amplitude for holes is smaller than the one for electrons in certain parameter
ranges (4). The effect becomes large for small interatomic distance and
small Z, as expected. For ferromagnetism, first principles calculations of
the relevant quantities have not yet been reported.

(4) Philosophical considerations

According to the philosophical principle known as Occam's rallor, 'plur-
alitas non est ponenda sine necessitas', plurality is not to be assumed
without necessity. If a single principle can explain a variety of observations it
380

should be preferred over multiple explanations. The principle of undressing

provides a single explanation for phenomena for which a very large number
of different explanations have been proposed, hence it should be preferred
over the other explanations unless clearly proven wrong.
Another requirement on scientific theories is that they can be falsified by
experiment. The present theory offers plenty of opportunity for falsification.
A few examples of possible observations that would disprove the theory
are: finding of a single superconductor that does not have hole carriers at
the Fermi energy; observation of transfer of spectral weight in one or two-
particle spectral functions from low to high frequencies as the collective
state (superconductor or ferromagnet) develops; finding that in manganites
the observed effective mass reduction is not accompanied by a substantially
enhanced quasiparticle peak in photoemission; finding that the gap slope
in superconductors does not have universal sign (2); etc.

References

1. M.R. Norman, M. Randeria, B. Janko and J.C. Campuzano, Phys. Rev. B 61,14742
(2000).
2. J. E. Hirsch and F. Marsiglio, Phys. Rev. B 62, 15131 (2000) and references therein.
3. J. E. Hirsch, Phys. Rev. B 43, 705 (1991) and references therein.
4. J.E. Hirsch, Phys.Rev. B48, 3327 (1993).
5. J.E. Hirsch, Phys. Rev. B 62, 14487 (2000), 14498 (2000).
6. J. E. Hirsch and F. Marsiglio, Phys. Rev. B 45, 4807 (1992) .
7. J.E. Hirsch, in Folarons and Bipolarons in high-Tc Superconductors and Related Ma-
terials, ed. by E.K.H. Salje, A.S. Alexandrov and W.Y. Liang (Cambridge University
Press, Cambridge, 1995) p. 234 ; Physica C 201, 347 (1992).
8. J.E. Hirsch, Phys. Rev. B 62, 14131 (2000).
9. J.E. Hirsch, Phys. Rev. Lett. 87, 206402 (2001); Phys.Rev.B 65, 184502 (2002).
10. J.E. Hirsch, Phys.Rev. B 65, 214510 (2002); Phys.Rev. B 66, 064507 (2002); cond-
mat/0207369 (2002).
11. Y Okimoto et aI, Phys.Rev. B57, 9377 (1998).
12. L. Degiorgi et aI, Phys. Rev. Lett. 79, 5134 (1997); Phys.Rev. B65, 121102 (2002).
13. H. Okamura et aI, Phys.Rev. B64, 180409 (2001).
14. E.J. Singley et aI, Phys.Rev.Lett. 89, 097203 (2002) .
15. J.F. Janak and A.R. Williams, Phys.Rev. B14, 4199 (1976).
16. H. Ding et aI, Phys. Rev. Lett. 87, 227001 (2001).
17. H. J. A. Molegraaf et aI, Science 295, 2239 (2002).
18. A.F. Santander-Syro et aI, cond-mat/0111539 (2001).
19. S.Uchida et aI, Phys. Rev. B 43, 7942 (1991).
20. Z.M. Yusof et aI, Phys. Rev. Lett. 88, 167006 (2002).
21. YAndo et aI, Phys. Rev. Lett. 87, 017001 (2001).
22. LK. Kikoin and S.W. Gubar, J.Phys. USSR 3,333 (1940).
23. A.F. Hildebrand and M.M. Saffren, Fmc. 9th Int. Conf. on Low Ternp.Fhys., ed. by
J.G. Daunt et al (Plenum, New York, 1965) p.459.
24. A.A. Verheijen et aI, Nature 345, 418 (1990).
25. LM. Chapnik, Sov.Phys. Dokl. 6, 988 (1962).
 381

List of Contributors
A. Aligia [email protected]
N. Andrei [email protected]
N. P. Armitage [email protected]
J. BaLa [email protected]
G. Baskaran [email protected]
C. D. Batista [email protected]
K. Baumgartner [email protected]
D. Bensimon [email protected]
N. Blümer [email protected]
C. J. Bolech [email protected]
J. Bonča [email protected]
R. Bulla [email protected]
M. Capone [email protected]
C. Castellani [email protected]
I. Chaplygin [email protected]
B. Coqblin [email protected]
T. A. Costi [email protected]
C. Czycholl [email protected]
L. Degiorgi [email protected]
T. P. Devereaux [email protected]
S. El Shawish [email protected]
H. Eschrig [email protected]
M. Fabrizio [email protected]
P. Fazekas [email protected]
L.-F. Feiner [email protected]
J. K. Freericks [email protected]
E. M. Görlich [email protected]
C. Grenzebach [email protected]
G. Grüner [email protected]
J. E. Gubernatis [email protected]
E. Helgren [email protected]
A. C. Hewson [email protected]
J. E. Hirsch [email protected]
R. Hlubina [email protected]
A. Jerez [email protected]
H. Keiter [email protected]
A. Kiss [email protected]
K. Koepernik [email protected]
J. Kroha [email protected]
M. Lang [email protected]

.
382

M. Lavagna [email protected]
C.T. Liang [email protected]
A. Lobos [email protected]
H. v. Löhneysen [email protected]
B. Lüthi [email protected]
R. Mallik [email protected]
D. Meyer [email protected]
H. J. A. Molegraaf [email protected]
J. Müller [email protected]
A. P. Murani [email protected]
A. M. Oleś [email protected]
D. Otto [email protected]
J. Paaske [email protected]
R. PodsiadLy [email protected]
P. Prelovšek [email protected]
C. Presura [email protected]
A. Ramšak [email protected]
J. Röhler [email protected]
A. Rosch [email protected]
A. Rycerz [email protected]
E. V. Sampathkumaran [email protected]
I. Santoso [email protected]
J. L. Sarrao [email protected]
T. Sasaki [email protected]
J. Schlueter [email protected]
K.–D. Schotte [email protected]
M. Sigrist [email protected]
J. Spalek [email protected]
F. Steglich [email protected]
J. D. Thompson [email protected]
E. Tosatti [email protected]
D. van der Marel [email protected]
P. G. J. van Dongen [email protected]
M. Vojta [email protected]
W. Wójcik [email protected]
B. Wolf [email protected]
P. Wölfle woelfl[email protected]
R. Zahorbeński [email protected]
S. Zherlitsyn [email protected]
V. Zlatić [email protected]
 383

Index

Ab Initio Approach, 253 Effective Mass, 367

Anderson (Impurity) Model, 225, 243, 265, Electrical Resistivity, 349
303 Electrodynamics, 185
Anderson Lattice Model, 303 Electron Correlations, 349
Anderson Transition, 151 Electron Glass, 185
Anderson-Holstein Model, 195 Electron-Interactions, 151
Antiferromagnetic Order, 139 Electron-Phonon Coupling, 195
Antiferroquadrupolar Order, 167 Electronic Raman Scattering, 111
ARPES, 65 Electronic Structure, 45
Emery Model, 45
Band Structure Calculation, 35 ESR Experiments, 175
Bethe Ansatz, 265 Exact Diagonalization, 253
Born Approximation, 129 Exact Solution, 111
Buckling, 55 EXAFS, 55
Extended SU(N) Kondo Model, 215
C-Axis, 55
Ce-Cu-Au, 139
F-Sum Rule, 331
CeCoIn5 , 341
Falicov-Kimball Model, 111, 283
CeIn3 , 341
Cerium Systems, 293 Fermi Surface, 65
Charge Dynamics, 111 Ferromagnetic Superconductor, 139
Coherent Potential Approximation, 303 Ferromagnetism, 313
Conductance, 235 Fullerenes, 93
Correlated Electrons, 253, 331
Correlated Systems, 359 Gd Compounds, 349
Correlation Energy, 7 General Dispersion Method, 331
Critical Behaviour, 151 Generating Functional, 303
Crystal Field, 265 Gutzwiller Approximation, 313
Crystal Structure, 35 Gutzwiller Factor, 119, 124, 125
CuO2 Lattice, 55
Cuprates, 55, 65 Heat Capacity, 349
Current-Phase Relation, 17 Heavy Electron Materials, 359
Heavy Fermion Superconductivity, 341
D-Wave Pairing, 17 Heavy Fermions, 139, 265, 273, 293, 323,
Density Functional Theory, 45 349
Density of States, 331 High Dimensions, 331
Differential Conductance, 225 High Temperature Superconductivity, 17,
Disorder, 185 73, 93
Doped Semiconductors, 151 Hole Scattering, 129
Doping Mechanism, 55 Holes, 367
Dressing, 367 Holstein Model, 195
Dynamic Hubbard Model, 367 Hubbard Model, 45, 111, 119, 120
Dynamical Conductivity, 323 Hubbard Splitting, 151
Dynamical Mean-Field Theory, 111, 195,
283, 303, 323 Impurity Entropy, 265
384

Impurity Specific Heat, 265 Nanochains, 253

Inelastic Light Scattering, 111 Nanoclusters, 253
Infinite Dimensions, 331 Nanoscale Structures, 225
Infinite Spatial Dimensions, 111 Nanoscopic Systems, 253
Insulator, 185 NbSi, 185
Integrability, 265 Neutron Spectroscopy, 293
Integrable Models, 265 Non Crossing Approximation, 303
Internal Energy, 7 Non-Fermi Liquid, 139, 215, 265
Nonequilibrium Transport, 235
Jahn-Teller Effect, 35 Numerical Renormalization Group, 195,
Josephson Effect, 17 205, 243
Josephson Product, 17
Optical Conductivity, 7, 331
Kinetic Energy, 7 Optical Properties, 359
Kondo Effect, 205, 225, 235, 243, 273, 293, Optical Sum Rule, 367
349 Orbital Liquid, 119, 120, 124, 125
Kondo Lattice, 273, 349 Orbital Ordering, 119, 124, 129, 130, 134–
Kondo Model, 265 136
Kondo Problem, 359 Orbital Polaron, 129
Orbital-Hubbard Model, 119, 120
LDA + U, 45
Orbiton, 129, 136
Linear Conductance, 243
Organic Superconductors, 73, 83
Local Criticality, 205
Localization, 185
P-Wave Superconductivity, 27
Localized Magnetic Moments, 151
Pair Correlation, 7
Long Range Interaction, 185
Paramagnetic Scattering, 293
Low Dimensional Spin Systems, 175
Periodic Anderson Model, 313, 323
Low Dimensional Systems, 359
Perovskite, 35
Magnetic Coupling, 45 Perturbation Theory, 225
Magnetic Impurity, 225 Photoemission, 136
Magnetic Polaron, 349 Plutonium, 341
Magnetism, 27 Poor Man’s Scaling, 235
Magnetization Plateaus, 175 Pr Compounds, 165
Magnetoresistance, 349 Pressure Induced Superconductivity, 73
Manganites, 129 PrFe4 P12 , 165, 167
Mean Field Theory, 165 Pseudogap, 65
Metal-Insulator Transition, 111, 151, 195 PuCoGa5 , 341
Metamagnetic Transition, 165, 169 Pulse Field Experiments, 175
Mirage, 225
Mixed Valence, 265 Quadrupolar Order, 165, 167
Mixed Valence Systems, 313 Quadrupolar Phase Transition in External
Modified Perturbation Theory, 323 Magnetic Field, 165
Mott Insulators, 73 Quantum Corrals, 225
Mott Transition, 93 Quantum Dot, 235, 243
Mott-Hubbard, 359 Quantum Fluctuations, 17
Multi-Band Effects, 27 Quantum Monte Carlo, 313, 331
Multi-Channel Models, 265 Quantum Phase Transition, 139, 205
Multipolar Order, 165 Quasiparticle, 132, 367
Quasiparticle Dispersion, 129
Nagaoka Theorem, 119, 123 Quasiparticle Weight, 367
 385

Rare Earth Compounds, 165 U-Cu-Pd, 139

Rare-Earth, 273 UBe13 , 265
Renormalization Group Method, 235 Ultrasonic Experiments, 175
Resistivity, 323 Unconventional Superconductivity, 27
Resonating Valence Bond (RVB) Theory, Underscreened Kondo Effect, 273
73 Undoped Cuprates, 45
Ruthenate, 27 Undressing, 367
RVB, 17 Uranium, 273
Uranium Compounds, 265
S-Wave Pairing, 17
Scanning Tunneling Microscopy, 225 Valence Fluctuation Systems, 293
Scattering Matrices, 303 Valence-Change Transition, 283
Self-Avoiding Walk, 303
Si:P, 151, 185 X-Ray Diffraction, 55
Silicon, 185
Skutterudites, 165, 167 YBa2 Cu3 Ox , 55
Slave Boson Theory, 73 Ytterbium Systems, 293
Specific Heat, 7
Zhang-Rice Singlet, 45
Spectral Functions, 65, 129
Zr-Zn, 139
Spectral Weight, 7
Spin Disorder, 349
Spin Fluctuations, 65
Spin Glass, 273
Spin Phonon Interaction, 175
Spin-Filtering, 243
Spin-Orbit Coupling, 27
Spin-Orbit Systems, 27
Stress Tuning, 151
Strong Correlations, 45, 65
Strong versus Intermediate Coupling Fixed
Point, 215
Strongly Correlated Electron Systems, 73
Strongly Correlated Electrons, 55
Strongly Correlated Systems, 93
SU(2) Symmetry, 119, 121
Superconductivity, 55, 65
Superexchange, 119, 123, 130, 135
Surface States, 225

t-J Model, 45, 65, 130, 135

Temperature Dependence of Transport Quant-
ities, 323
Thermal Properties, 83
Thermo(electric) Power, 151, 323
Thermodynamic Bethe Ansatz, 265
Thermodynamic Properties, 83
Transport, 331
Tricritical Point, 165
Two-Channel Kondo Model, 205
Two-Impurity Kondo Model, 205