Kouwenhoven - Mesoscopic e - Transport PDF

Mesoscopic Electron Transport
NATO ASI Series

Advanced Science Institutes Series
A Series presenting the results of activities sponsored by the NA TO Science Committee,

which aims at the dissemination of advanced scientific and technological knowledge,
with a view to strengthening links between scientific communities.
The Series is published by an international board of publishers in conjunction with the NATO
Scientific Affairs Division
A Life Sciences
B Physics
Plenum Publishing Corporation

London and New York
C Mathematical and Physical Sciences

D Behavioural and Social Sciences
E Applied Sciences
Kluwer Academic Publishers

Dordrecht, Boston and London
F
G
H
I
Computer and Systems SCiences

Ecological Sciences
Cell Biology
Global Environmental Change
Springer-Verlag
Berlin, Heidelberg, New York, London,
Paris and Tokyo
PARTNERSHIP SUB-SERIES
1.
2.
3.
4.
5.
Disarmament Technologies
Environment
High Technology
Science and Technology Policy
Computer Networking

Springer-Verlag / Kluwer Academic Publishers
The Partnership Sub-Series incorporates activities undertaken in collaboration with NATO's

Cooperation Partners, the countries of the CIS and Central and Eastern Europe, in Priority Areas of
concern to those countries.
NATo-PCo-DATA BASE
The electronic index to the NATO ASI Series provides full bibliographical references (with keywords
and/or abstracts) to more than 50000 contributions from international scientists published in all
sections of the NATO ASI Series.
Access to the NATO-PCO-DATA BASE is possible in two ways:
- via online FILE 128 (NATO-PCO-DATA BASE) hosted by ESRIN,
Via Galileo Galilei, 1-00044 Frascati, Italy.
- via CD-ROM "NATO-PCO-DATA BASE" with user-friendly retrieval software in English, French
and German ( WTV GmbH and DATAWARE Technologies Inc. 1989).
The CD-ROM can be ordered through any member of the Board of Publishers or through NATOPCO, Overijse, Belgium.
Series E: Applied Sciences' Vol. 345

edited by
Lydia L. Sohn
Department of Physics,
Princeton University,
Princeton, U.S.A.
Leo P. Kouwenhoven
Department of Applied Physics and DIMES,
Delft University of Technology,
Delft, The Netherlands
and
Gerd Schon
Institut fOr Theoretische Festkorperphysik,
Universitat Karlsruhe,
Karlsruhe, Germany
Springer-Science+Business Media, B.V.
Proceedings of the NATO Advanced Study Institute on

Curac;ao, Netherlands Antilles
25 June - 5 July 1996
A C.i.P. Catalogue record for this book is available from the Library of Congress
ISBN 978-90-481-4906-3
ISBN 978-94-015-8839-3 (eBook)
DOI 10.1007/978-94-015-8839-3
Printed on acid-free paper
All Rights Reserved
1997 Springer Science+Business Media Oordrecht

Originally published by Kluwer Academic Publishers in 1997.
Softcover reprint of the hardcover 1st edition 1997
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any information storage and retrieval system, without written
permisSion from the copyright owner.
This book contains the proceedings of a NATO Advanced Study Institute held within the programme of
activities of the NATO Special Programme on Nanoscale Science as part of the activities of the NATO
Science Committee.
Other books previously published as a result of the activities of the Special Programme are:
NASTASI, M., PARKING, D.M. and GLEITER, H. (eds.), Mechanical Properties and Deformation Behavior of
Materials Having Ultra-Fine Microstructures. (E233) 1993 ISBN 0-7923-2195-2
VU THIEN BINH, GARCIA, N. and DRANSFELD, K. (eds.), Nanosources and Manipulation of Atoms under
High Fields and Temperatures: Applications. (E235) 1993 ISBN 0-7923-2266-5
LEBURTON, J.-P., PASCUAL, J. and SOTOMAYOR TORRES, C. (eds.), Phonons in Semiconductor Nanostructures. (E236) 1993 ISBN 0-7923-2277-0
AVOURIS, P. (ed.), Atomic and Nanometer-Scale Modification of Materials: Fundamentals and Applications. (E239) 1993 ISBN 0-7923-2334-3
BLOCHL, P. E., JOACHIM, C. and FISHER, A. J. (eds.), Computations for the Nano-Scale. (E240) 1993
ISBN 0-7923-2360-2
POHL, D. W. and COURJON, D. (eds.), Near Field Optics. (E242) 1993 ISBN 0-7923-2394-7
SALEMINK, H. W. M. and PASHLEY, M. D. (eds.), Semiconductor Interfaces at the Sub-Nanometer Scale.
(E243) 1993 ISBN 0-7923-2397-1
BENSAHEL, D.
c., CANHAM, L. T.
and OSSICINI, S. (eds.), Optical Properties of Low Dimensional Silicon
Structures. (E244) 1993 ISBN 0-7923-2446-3

HERNANDO, A. (ed.), Nanomagnetism (E247) 1993. ISBN 0-7923-2485-4
LOCKWOOD, DJ. and PINCZUK, A. (eds.), Optical Phenomena in Semiconductor Structures of Reduced
Dimensions (E248) 1993. ISBN 0-7923-2512-5
GENTILI, M., GIOVANNELLA, C. and SELCI, S. (eds.), Nanolithography: A Borderland Between STM, EB,
IB, and X-Ray Lithographies (E264) 1994. ISBN 0-7923-2794-2
GUNrHERODT, H.-J., ANSELMETII, D. and MEYER, E. (eds.), Forces in Scanning Probe Methods (E286)
1995. ISBN 0-7923-3406-X
GEWIRTH, A.A. and SIEGENTHALER, H. (eds.), Nanoscale Probes of the Solid/Liquid Interface (E288)
1995. ISBN 0-7923-3454-X
CERDEIRA, H.A., KRAMER, B. and SCHON, G. (eds.), Quantum Dynamics of Submicron Structures (E291)
1995. ISBN 0-7923-3469-8
WELLAND, M.E. and GIMZEWSKl, J.K. (eds.), Ultimate Limits of Fabrication and Measurement (E292)
1995. ISBN 0-7923-3504-X
EBERL, K., PETROFF, P.M. and DEMEESTER, P. (eds.), Low Dimensional Structures Prepared by Epitaxial
Growth or Regrowth on Patterned Substrates (E298) 1995. ISBN 0-7923-3679-8
MARTI, O. and MOLLER, R. (eds.), Photons and Local Probes (E300) 1995. ISBN 0-7923-3709-3
GUNTHER, L. and BARBARA, B. (eds.), Quantum Tunneling of Magnetization - QTM '94 (E301) 1995.
ISBN 0-7923-3775-1
PERSSON, B.NJ. and TOSATTI, E. (eds.), Physics of Sliding Friction (E311) 1996. ISBN 0-7923-3935-5
MARTIN, T.P. (ed.), Large Clusters of Atoms and Molecules (E313) 1996. ISBN 0-7923-3937-1
DUCLOY, M. and BLOCH, D. (eds.), Quantum Optics of Confined Systems (E314). 1996.
ISBN 0-7923-3974-6
ANDREONI, W. (ed.), The Chemical Physics of Fullereness 10 (and 5) Years Later. The Far-Reaching
Impact of the Discovery ofC60 (E316). 1996. ISBN 0-7923-4000-0
NIETO-VESPERINAS, M. and GARCIA, N. (Eds.): Optics at the Nanometer Scale: Imaging and Storing with
Photonic Near Fields (E319). 1996. ISBN 0-7923-4020-5
RARITY, J. and WEISBUCH, C. (Eds.): Microcavities and Photonic Bandgaps: Physics and Applications
(E324). 1996. ISBN 0-7923-4170-8
LURYI, S., XU, J. and ZASLAVSKY, A. (Eds.): Future Trends in Microelectronics: Reflections on the Road to
Nanotechnology (E323). 1996. ISBN 0-7923-4169-4
JAUHO, A. and BUZANEVA, E.V. (Eds.): Frontiers in Nanoscale Science for MicronlSubmicron Devices
(E328). 1996. ISBN 0-7923-4301-8
ROSEl, R. (Ed.): Chemical, Structural and Electronic Analysis of Heterogeneous Surfaces on Nanometer
Scale (E333). 1997. ISBN 0-7923-4489-8
JOACHIM, C. (Ed.): Atomic and Molecular Wires (E341). 1997. ISBN 0-7923-4628-9
CONTENTS
PREFACE
L.P. Kouwenhoven, G. Schon, and L.1. Sohn
ix
INTRODUCTION TO MESOSCOPIC ELECTRON TRANSPORT

L.P. Kouwenhoven, G. Schon, and L.L. Sohn
GEOMETRIC PHASES IN MESO SCOPIC SYSTEMS
- FROM THE AHARONOV-BOHM EFFECT TO BERRY PHASES
A. Stern
45
DELOCALIZATION, INELASTIC SCATTERING AND TRANSPORT

DUE TO INTERACTIONS
Y.Imry
83
ELECTRON TRANSPORT IN QUANTUM DOTS

L.P. Kouwenhoven, C.M. Markus, P.L. McEuen, S. Tarucha, R.M. Westervelt, and N.S. Wingreen
105
MAGNETOTUNNELING SPECTROSCOPY: STUDYING SELF-ORGANIZED

QUANTUM DOTS AND QUANTUM CHAOLOGY
L. Eaves
215
SHOT NOISE IN MESOSCOPIC SYSTEMS
M.J .M. de Jong and C.W.J. Beenakker
225
ADMITTANCE AND NONLINEAR TRANSPORT IN QUANTUM WIRES,

POINT CONTACTS, AND RESONANT TUNNELING BARRIERS
M. Biittiker and T. Christen
259
TRANSPORT THEORY OF INTERACTING QUANTUM DOTS

H. Schoeller
291
TRANSPORT IN A ONE-DIMENSIONAL LUTTINGER LIQUID

M.P.A. Fisher and L.I. Glazman
331
THE PROXIMITY EFFECT IN MESO SCOPIC DIFFUSIVE CONDUCTORS

D. Esteve, H. Pothier, S. Gueron, N.O. Birge, and M. H. Devoret
375
viii
MESOSCOPIC EFFECTS IN SUPERCONDUCTIVITY

R. Fazio and G. Schon
407
ULTRASMALLSUPERCONDUCTORS
D.C. Ralph, C.T. Black, J.M. Hergenrother, J.G. Lu, and M. Tinkham
447
THE SUPERCONDUCTING PROXIMITY EFFECT IN SEMICONDUCTORSUPERCONDUCTOR SYSTEMS: BALLISTIC TRANSPORT, LOW DIMENSIONALITY AND SAMPLE SPECIFIC PROPERTIES
B.J. van Wees and H. Takayanagi
469
SCANNING PROBE MICROSCOPES AND THEIR APPLICATIONS

L.L. Sohn, C. T. Black, M. Eriksson, M. Crommie, and H. Hess
503
QUANTUM POINT CONTACTS BETWEEN METALS

J .M. van Ruitenbeek
549
CONDUCTANCE QUANTIZATION IN METALLIC NANOWIRES

N. Garcia, J.L. Kosta-Kramer, A. Gil, M.l. Marques, and A. Correia
581
QUANTUM OPTICS
Y. Yamamoto
617
TOPICS IN QUANTUM COMPUTERS

D.P. DiVincenzo
657
Preface
The major advances in lithographic technology have enabled researchers to

fabricate nano-scale structures with great control. This created an enormous interest in the investigation of these "mesoscopic" systems for novel
phenomena, especially those relating to electronic transport. Metallic and
superconducting tunnel junctions, quantum dots and multi-dot systems in
semiconductor heterostructures, as well as semiconductor-superconductor
hybrid nanostructures have been studied. This research has led to the discovery of a variety of new phenomena. Quantum mechanical interference
effects and level quantization, as well as single-electron charging effects
influence the electron transport. This is demonstrated, for instance, by
the tunnel spectroscopy of few-electron quantum dots, resonant tunneling
phenomena, parity effects in superconducting tunnel junctions, mesoscopic
proximity effects near normal-superconductor interfaces, and quantum fluctuations in Josephson junction arrays. New mesoscopic topics are emerging and evolving where the overall understanding is still in progress. Such
topics include scanning-tunneling microscopy studies of few-atom metallic
clusters, Kondo-type physics in the transport properties of quantum dots,
time-dependent effects, and properties of interacting systems, e. g. of Luttinger liquids.
The primary goals of the NATO-Advanced Study Institute "Mesoscopic
Electron Transport" was to highlight selected areas in the field, to provide
a comprehensive review of such systems and to serve as an introduction to
the new and developing areas of mesoscopic electron transport.
Areas which have been addressed in this ASI included:
nano-fabrication and spectroscopy using scanning probe
phase-coherent transport
time-dependent phenomena, noise, and chaos
transport through quantum dots, grains, clusters, and molecules
interaction effects in quantum systems, Luttinger liquids
mesoscopic superconductivity
some applications and extensions.
The following NATO ASI proceedings is a compilation of the views and
ideas of the ASI speakers on the above subjects. We hope that they will
provide a stimulus for new directions of the research in the rapidly evolving
field of mesoscopic electron transport.
ix
x
Acknowledgements
We would like to thank all lecturers for their care in delivering their courses
and in preparing their manuscripts. Special thanks go to the members of
our Organizing Committee: Daniel Esteve, Leonid I. Glazman, and Hans
Mooij. Their advise in the selection of topics is highly appreciated.
We also would like to thank the Sonesta Beach Hotel in Curacao, The
Netherlands Antilles for hosting our ASI for two weeks. Their efficiency
and help in all matters made it possible for us to manage the rather loaded
program.
Finally we acknowledge the generous grant by the Scientific Affairs Division of the North Atlantic Treaty Organization (NATO) - without which
this ASI would not have been possible. We acknowledge furthermore the
support of the Office of Naval Research (GRANT No. N00014-96-1-0724),
the US National Science Foundation, NEC Research Institute, as well as
the Universities of Delft and Karlsruhe and Princeton University.
Leo P. Kouwenhoven, Delft University
Gerd Schon, University of Karlsruhe
and Lydia L. Sohn, Princeton University
June 1997
INTRODUCTION TO MESOSCOPIC ELECTRON TRANSPORT
LEO P. KOUWENHOVEN
Dept. of Applied Physics and DIMES

Delft University of Technology, 2600 GA Delft, The Netherlands
GERD SCHON
Institut fur Theoretische Festkorperphysik

Universtitiit Karlsruhe, 76128 Karlsruhe, Germany
AND
LYDIA L. SOHN
Dept. of Physics, Princeton University

Princeton, NJ 08544 - 0708, USA
In this introductory chapter several basic concepts, relevant for mesoscopic electron transport, will be described. The aim is to provide a basis
for several of the following Chapters of this volume. We, therefore, describe
in the first Section various aspects of electron quantum transport in twodimensional electron gases. This includes an estimate of typical material
parameters as well as comments on the fabrication. We then describe the
quantization of the conductance in point contacts and the edge state picture of the Quantum Hall effect. In the second Section we describe the
theory of single-electron tunneling in systems with strong charging and
Coulomb-blockade effects. Here we restrict ourselves to the simplest case
where low order perturbation theory is sufficient. We first discuss metallic
low-capacitance junction systems and then indicate the relevant extensions
when dealing with transport through quantum dots with discrete levels.
Many further, equally important aspects of mesoscopic transport can not
be covered here. Exam pIes are interference and weak localization effects,
level statistics or the many body description of solids. Fortunately, some of
those will be covered in the more specialized Chapter of this volume.
L. L. Sohn et al. (eds.), Mesoscopic Electron Transport, 1-44.
1997 Kluwer Academic Publishers.
(0)
(b)
Figure 1. (a) Schematic cross section of a GaAs/ AIGaAs heterostructure. The 2DEG is
located at the interface between the GaAs substrate and the AIGaAs top layer. (b) Hall
bar with six Ohmic contacts (shaded squares).
1. Electron Transport
1.1. 2DEG SYSTEMS AND THE QUANTUM HALL EFFECT
In this Section we introduce basic properties of a two-dimensional electron

gas (2DEG) in a GaAs/ AIGaAs heterostructure (see Fig. Ia). On the GaAs
substrate a layer of typically 100 nm AIGaAs is grown. Somewhere halfway
in the AIGaAs layer there is a thin layer where the Ga atoms are replaced by
Si donor atoms. With a proper amount of Si one finds that at low temperature the only mobile electrons are located at the GaAs/ AIGaAs interface.
These free electrons are attracted by the GaAs since they can lower their
energy in this smaller band gap material. They are also held as close as
possible to their ionized Si+ donors and thus they form a thin conducting
layer near the GaAs/ AIGaAs interface (for a review on growth of GaAs
heterostructures see Ref. [1]). Since GaAs and AIGaAs can form a nearly
perfect interface on the atomic scale and since the Si donors are spatially
separated, the electrons experience very little scattering. Typical mean free
paths are 10 f.Lm and the record is close to 100 f.Lm. Before we discuss ballistic mesoscopic devices we first review a few important electron transport
properties of 2DEG's. In Table 1.1 we have summarized a number of useful
relations and some typical values for GaAs/ AIGaAs heterostructures.
For transport experiments one first defines a so-called Hall-bar of typically O.Immx Imm. At the edge ofthe Hall-bar one then locally evaporates
a number of metal squares of roughly 50f.Lmx50f.Lm in size. By heating the
whole sample this metal diffuses into the semicondu.ctor where at some
point it makes electrical contact to the 2DEG. Good contacts are characterized by linear current-voltage traces and are therefore called Ohmic
contacts. Fig. Ib shows a Hall-bar with 6 Ohmic contacts in a configuration
that allows measurements of the Hall resistance Rxy and the longitudinal
3
2.5
CII
.....
s:.
r::=
" 4 3
1.5
0 .5
II I
us
213
I
":['I
'11
31S
1 ~11l5lt
Iii" ""l
SI1
4/5
71S
10
'-'-
15
20
MAGNETIC FiElD {T]
25
Figure 2. Measurement of the Hall resistance Rxy and the longitudinal resistance Rxx
at 0.1 K. The mobility is J.l = 1.5 1Q6 cm 2/VS, the temperature is T = 80mK. (Figure
provided by R. L. Willett.)
resistance Rxx. The sam pIe is current biased and resulting voltages can be
measured as a function of magnetic field.
Fig. 2 shows a remarkable measurement of Rxy and Rxx as a function
of magnetic field measured at a temperature of 80mK. The Hall resistance
shows plateaus where the longitudinal resistance has minima of virtually
zero resistance. The striking aspect of this data is that the plateaus are
precisely quantized at integer and fractional multiples of hi e2 The integer
plateaus are known as the integer quantum Hall effect (QHE), and correspondingly, the fractional plateaus are referred to as the fractional QHE.
The precision of the quantization of the former is so accurate that it now
forms the international standard of resistance. An important quantity in
the quantum Hall regime is the filling factor v = hnsleB. It is equal to
the number of electrons divided by the number of flux quanta <[>0 = hie.
For filling factor v = 1 the system is in the center of the first plateau at
hi e2 , for v = 2 in the center of the second plateau at h12e 2 , and so forth.
The filling factor is, for instance, convenient for determining the electron
density. The oscillations in the longitudinal resistance, which have minima
at the same magnetic fields where the Hall resistance shows plateaus, are
called Shubnikov-de Haas oscillations. The temperature dependence of the
oscillation minima is an accurate determination of the mobility and mean
free path of the 2DEG.
Table 1.1. Useful relations and system parameters (including spin degeneracy). We have chosen typical values for J1e, and n s , from which the
other values are deduced.
-
electron mobility
scattering time
effective mass
electron density
Fermi energy
Fermi velocity
Fermi wavelength
elastic mean free path
cyclotron radius at EF
angular cyclotron frequency
magnetic length
J1e = 106 cm 2/Vs (typical value)
= m*J1e/e = 38 ps
m* = 0.067 mo
ns = 2.810 15 m-=2(typical value)
EF = 7rTi 2 ns /m* = 10 meV

VF = (2EF/m*)1/2 = 2.310 5 m/s
AF = (27r/ns) 1/2 = 47 nm
1= VFT= 8.7 J1m
re = m*vF/eB = 88 nm at B = 1 T
We = eB/m* = 2.610 12 rad/s
IB = (Ti/eB)I/2 = 8.1 nm at B = 10 T
1.2. E-BEAM FABRICATION OF A SUBMICRON SEMICONDUCTOR

DEVICE
We briefly outline a standard procedure for fabricating small devices in a

2DEG. We start from a 2DEG confined in a GaAs/ AlGaAs heterostructure. To laterally confine the electrons one must define a pattern on top of
the heterostructure. As an example, we describe e-beam lithography, which
is a technique also used in the fabrication of chips. Depending on the desired pattern, one can choose between many variatio~s in the lithography
process steps. One procedure is shown schematically in Fig. 3. An organic
resist film (~100nm thick) is spun onto the substrate. Exposing the resist with an electron beam results in a molecular-mass difference between
the exposed and unexposed parts (see Fig. 3a). An appropriate developer
removes only the exposed resist, resulting in the mask pattern shown in
Fig. 3b. Evaporated material now sticks only at the substrate where the
resist has been removed (see Fig. 3c). The mask itself can be removed by
dissolving the remaining resist (lift-off), leaving a small pattern on top of
the substrate (see Fig. 3d). The minimum resolution of such a pattern with
present day electron-beam lithography facilities is about 20 nm.
(a)~
W4
+ ++ + + ++ +
(d)
f%?M
.
I,
,
electron-beam
->O'nm o"an',
reslst
substrate
(b)~
(C)
evaporation
Wd?4
-w
mask after
development
submicron
structure
after lift-off
Figure 9. Outline of the electron-beam lithography procedure for fabricating submicron

structures.
. !
(a) Cross-section
~
(b) Top-view
split-gate
50 nm
n-A1GaAs
--7-=:;;:=""""c:---
20 nm
A1GaAs
GaAs
.. ......
cont ined 2DEG
Figure 4. (a) Cross-section of a GaAs/ AIGaAs heterostructure with typical layer thicknesses. A negative voltage VG applied to the metal split-gate confines the electrons laterally in the 2DEG.
(b) Top-view of a QPC. The dotted line indicates the depletion region in the 2DEG,
which is tuned by VG. The two wide 2DEG regions act as reservoirs, emitting electrons
through the QPC with energies up to their electrochemical potentials PI and P2. A
voltage difference V = (PI - P2)/e results in a net current I through the QPC.
From this point there are basically two ways to transfer the pattern to
the 2DEG. One way is to use the pattern as an etch mask. Etching removes
the portion of the 2DEG not protected by the pattern. The boundaries of
the etched pattern cause a depletion region, such that the conducting width
in the 2DEG is unknown and often much smaller than the defined width [2].
Another, more flexible way to transfer the pattern to the 2DEG is to use it
as a gate [3, 4]. Applying a negative voltage to the gate depletes the electron
gas beneath it, thereby confining electron motion to the ungated region. For
a split-gate geometry, shown in Fig. 4, this results in a narrow conducting
channel. The advantage of the split-gate technique is that the conducting
width of the point contact in the 2DEG can be tuned from the defined
lithographic width of the pattern to zero, by making the gate voltage more
negative. Transport between the two wide 2DEG regions in Fig. 4b occurs
only through the point contact and can be studied !'l.8 a function of the
width by changing the gate voltage. The width W of the constriction can
be made comparable to the Fermi wavelength, so this device is called a
quantum point contact (QPC). The actual induced potential in the 2DEG
is unknown, but self-consistent calculations [5] indicate that it has a saddleshape (see Fig. 5a). In the constriction, electrons are confined in the lateral
x-direction and slowed down by the presence of a potential barrier in the
y-direction. Making the gate voltage more negative, simultaneously reduces
the width and increases the barrier height. For zero-width or a barrier which
is higher than the Fermi energy EF of the 2DEG, the QPC is pinched-off and
electron transport between the wide 2DEG regions is impossible. Fig. 5b
shows a scanning electron micrograph of a double point contact device.
(a)
..
(b)
,,"
!!.;
,
o
Figure 5.
(a) Saddle-shaped potential induced in the 2DEG upon application of a
negative gate voltage, resulting in lateral confinement in the x-direction and a potential
barrier in the longitudinal y-direction (from Beenakker, and van Houten in Ref. [6]).
(b) Scanning electron micrograph of a double-QPC device. The white areas are the Au
gates, and the marker is 1 IJm long. The QPCs are 250 nm wide and are separated by
1.5 IJm.
1.3. QUANTIZED CONDUCTANCE OF A POINT CONTACT
The resistance of a point contact in the classical ballistic regime is known as

the Sharvin resistance [7]. The Sharvin resistance is entirely due to elastic
backscattering at the geometrical narrowing of the ballistic point contact.
Dissipative processes, which bring the electron system into thermodynamic

equilibrium, take place far away from the point contact (Le. several times
the inelastic mean free path). Therefore, the cause of the resistance in this
system is spatially separated from its corresponding Joule heating. The
classical conductance of a Sharvin point contact in a 2DEG is [8]
Gs =
e2 dN 2D
-;;d"E VFW .
(1)
The quantum mechanical 2D density of states per unit area, including

a factor 2 for spin degeneracy, is dN 2D IdE = m* /7rn, while VF = nkF/m*
is the Fermi velocity. We can rewrite the classical equation (1) to a semiclassical version which includes the conductance quantum 2e 2 /h
G _ 2e 2 kF W _ 2e 2 2W
s - h-7r- - h
AF .
(2)
The Fermi wave-vector kF or wavelength AF are related to the 2D electron density ns by kF = 27r / AF = (27rns) 1/2. The semi-classical form of the
Sharvin conductance is continuous and linear in the width W. However,
Eq. 2 suggests that we can expect deviations due to the wave nature of
electrons whenever AF is of order W. We give a derivation of the quantum
version of the Sharvin conductance below, but first discuss the experimental
results.
The conductance of a point contact is measured by passing a current
I through the sam pIe and measuring the voltage V between the current
source and drain (see Fig. 4b). Fig. 6 shows the conductance G in units
of 2e 2 /h = (12906 0)-1 versus gate voltage Vo measured at B = O. Assuming that the width varies linear with gate voltage, we indeed see that
on average the conductance decreases linearly when the constriction is narrowed. However, around this classical dependence we see that G changes
in quantized steps of 2e 2 /h. At Vo = -2.2V, the conductance becomes
zero, corresponding to a pinched-off point contact. Fig. 6 further shows
that on increasing the temperature the conductance quantization gradually disappears [9]. The conductance G(VG) is roughly linear at 4.2 K, in
accordance with the classical dependence of Eq. 1. Although the classical
result cannot explain the quantization, we note that the plateau values are
obtained in Eq. 2 whenever the width W is an integer multiple of AF/2.
Eq. 2 predicts that an increase in W of AF/2 (which is 21 nm in this sample) increases G by 2e 2 /h. In total 16 steps were observed between pinch-off
and VG = -0.3V were the constriction is just formed in the 2DEG. The 16
steps give an estimate of the width W of about 340 nm, somewhat larger
than the lithographic width of 250 nm, but consistent with the schematic
depletion profile shown in Fig. 4( a). These considerations are reminiscent
-2
-1 . B
GA TE VOLTAGE
-1.6
IV)
Figure 6. Conductance versus gate voltage at B = 0 and different temperatures. Increasing the temperature thermally averages the higher plateaus first (from Ref. [9]).
of the states of a particle-in-a-lD-box, which, as we show below, is the basic

idea behind the conductance quantization.
We note that the conductance quantization is not as exact as the quantum Hall effect. First, a series resistance (:::::i 100 f2) ~riginating from the
wide 2DEG regions has been subtracted to line up the plateaus at their
quantized values [8]. Furthermore, the plateaus are not completely flat.
This may be due to scattering at impurities in the vicinity of the QPC or,
as we discuss below, the abruptness of the constriction.
We now discuss that the conductance quantization for transport through
ID subbands. If the potential which describes the transition from the
wide 2DEG regions to the narrowest point in the QPC varies sufficiently
smoothly (i.e. adiabatically), the potential variation in the x- and y-directions
may be decoupled [10] (see Fig. 5a). The narrowest point forms the bottleneck of the QPC in the sense that it completely determines the transport
properties. In this case we can calculate transport through a QPC from the
Hamiltonian
2
H = 2Px
m*
+ eV(x) + 2Py ,
m*
(3)
where V (x) describe the transport at the bottleneck. For the confinement in
the lateral x-direction, we follow Berggren et al. [11] and choose a parabolic
confining potential V(x) = 1/2m*w6x2. Self-consistent calculations of Laux
et al. [5] have shown that small split-gate samples have a confinement close
to such a parabola. The advantage of using a parabolic potential is that the
resulting Schrodinger equation can be written in the form of a harmonic

oscillator having energy eigenvalues
1
2
n2 k 2
E = (n - -)nwo + - y
n
2m*
(n = 1,2, ... )
(4)
which contains a free-electron kinetic energy dispersion in the longitudinal

y-direction. In the lateral x-direction the energy states, indexed by n =
1,2, ... , are quantized and separated in energy by nwo.
Because the electron motion is free in one direction .only, Eq. 4 describes
ID subbands. Fig. 7 shows the ID subband dispersion versus longitudinal
wave-vector kyo The right-going electrons, with a velocity nVn = dEn/dky,
originate from the left 2DEG reservoir, which at zero temperature populates
all the states up to its electrochemical potential J.l1. Similarly, the left-going
electron states are occupied up to J.l2, the electrochemical potential of the
right 2DEG reservoir (see Fig. 4b).
A voltage difference V = (J.l1 - J.l2) / e between the two reservoirs results
in a net current I, which is carried by the (uncompensated) electron states
in the energy interval between J.l1 and J.l2. Note that we define the Fermi
energy as EF = J.l1 = J.l2 when V = o. The net current I at zero temperature
IS
(5)
which includes the transmission probability of the n-th subband Tn(E) to
describe possible scattering events. Here N denotes the number of occupied
subbands, i.e. the largest number for which EN(ky = 0) < EF. The ID spindegenerate density of states is dNn/dE = 2/rr(dEn/dk y)-1. The important
aspect of ID transport is the cancelation of the energy dependence in the
product of velocity and density of states (dNn/dE)v n = 4/h. For small
voltages (eV ~ EF), one can take Tn(E) = Tn(EF). Substituting this
in Eq. 5, one finds that the conductance G = I/V = eI/(J.l1 - J.l2) is
independent of energy
(6)
Eq. 6 is known as the 2-terminal Landauer formula [12]. If no backscattering
takes place, so that 2:;:=1 Tn (EF) = N, Eq. 6 reduces to
(7)
10
Figure 7. Energy En versus longitudinal wave-vector k y from Eq. 4 at the bottleneck of

a QPC assuming a parabolic confinement potential. The ID subbands are separated by
nwo. A net current results from the uncompensated occupied electron states in the interval
between /Jl and /J2, the electrochemical potentials of the two wide 2DEG reservoirs.
demonstrating that each occupied subband contributes 2e 2 /h to the conductance. The subbands are called 1D current channels to emphasize that
each channel carries the same amount of current.
In the experiment, a decreasing VG increases the barrier in the QPC,
and simultaneously increases the lateral confinement ~nd consequently the
energy splitting. Both effects increase the subband energies. As long as EF
is between two subband bottoms, N is constant and G is quantized. If a
subband bottom moves through EF, N changes by 1 and G by 2e 2 /h.
Several numerical calculations [13] have shown that Eq. 7 gives an accurate description of a QPC with the assumptions that impurity scattering
is absent and that the potential variations are smooth. Sharp potential
variations, possibly present at the entrance and exit of the QPC [14] or
originating from impurities [15, 16] can give rise to backscattering and
therefore destroy the quantization. The assumption of an adiabatic constriction in Eq. 3 is therefore no longer valid. At T > 0 the reservoirs inject
electrons with a Fermi-Dirac distribution, which averages the conductance
G(T) = f dE( -df /dE)G(E). Comparing this with the temperature dependence of the curves in Fig. 6, it is found that the sub band separation gradually increases from about 1 meV at VG = -IV to 3 meV at VG = -2.1V.
This illustrates that thermal averaging has a stronger effect on the higher
plateaus, as observed in Fig. 6.
1.4. DEPOPULATION OF 1D MAGNETO-ELECTRIC SUBBANDS
We now turn to the case of a QPC with an applied magnetic field in the
z-direction. In the Hamiltonian of Eq. 3, the magnetk field B is included
by substituting (p - eA) for the momentum p. In the Landau gauge for
the vector potential A = Ay = Bx, the Schrodinger equation is once again
11
En
En
E - --
-------
llObt
tiro
;;'
~n
-~~I
ky
(a) electric
8=0
(b)
ky
hybrid
8;tO
-----------
ky
(c) magneto
C4:: >> COo
Figure 8. ID subband dispersion for three values of the magnetic field, illustrating
magnetic depopulation. The energy splitting is electric (hwo) in (a), and hybrid (hw,
with w2 = w~ + w~) in (b). For large magnetic fields (c) the 1D subbands are Landau
levels with an energy splitting of hw c
that of a harmonic oscillator, but now with energy eigenvalues [11]
(8)
These eigenvalues describe hybrid magneto-electric subbands. With w 2 =
w6 +w~, where the cyclotron frequency is We = eBjm*, the energy separation is now a combination of the electrical confinement and the magnetic
field. mB = m*w 2 jW6 is a magnetic field dependent effective mass yielding
a smaller dispersion for larger magnetic fields. The influence of the magnetic field on the subband dispersion is shown schematically in Fig. 8. For
B = 0 the subbands are determined by the electrical 'confinement only. A
small magnetic field increases the subband splitting and reduces the dispersion. For large magnetic fields, the subbands have the magnetic energy
separation and a vanishing dispersion. In this case, the subbands are the
well-known Landau levels. It can be seen from Fig. 8 that on increasing the
magnetic field, the number of occupied subbands decreases. This process is
known as the depopulation of magneto-electric subbands.
One can show that the velocity and density of states also cancel in a
magnetic field and that Eqs. ( 6) and ( 7) are still valid [17, 18]. From
the above analysis it follows that a gradual transition exists between the
quantized conductance G = N2e 2 jh at zero magnetic field (with N the
number of occupied electric subbands in Fig. 8a) to the quantum Hall
conductance GH = NL2e 2 jh at a high magnetic field (with NL the number
of occupied Landau levels in Fig. 8c). We note that our particular choice
of a parabolic confinement does not affect the general conclusions, such as
the cancelation of velocity with the density of states, and the conductance
quantization at zero and non-zero magnetic field.
12
. C:
6=0
10
-E.a
(lI
~ 6
ow
~
~
Z 0
80
o
-2
-1. 8 - 1 . 6 -1. 4 -1 2
-,
GATE VOLTAGE (V)
Figure 9. QPC conductance versus gate voltage at 0.6 K for several values of magnetic
field. The increasing width of the plateaus demonstrate the increasing energy splitting in
a magnetic field. The curves have been offset for clarity (from Ref. [18]).
Fig. 9 shows the conductance of a QPC versus gate voltage for several
values of the magnetic field [18]. Af, can be seen, the quantization is preserved in a magnetic field. Above B = 1 T, spin-resolved plateaus develop
at odd multiples of e2 /h. The depopulation can be seen from the fact that
at a fixed gate voltage, the number of plateaus (or, equivalently, the number of occupied subbands), decreases with increasing magnetic field. From
the measurements of Fig. 9, one can deduce subband splittings of about
1 meV at VG = -IV and 3 meV at VG = -2V, in agreement with the
values obtained from the temperature dependence [9]. A third independent
way to determine the subband splittings is by measuring the non-linear
current-voltage characteristics of a QPC [19].
1.5. ELECTRON TRAJECTORIES IN A LOW MAGNETIC FIELD
In the previous Section, we discussed the influence of a magnetic field on

the subband dispersion in momentum space. We now consider the electron
motion in real space, which yields a simple physical picture of the QHE
and associated effects. To elucidate the quantized electron motion in a
high magnetic field, we first discuss classical electron trajectories in a low
magnetic field.
13
(~~
~
~l ~~m~'E'~--~1 ~2
eV(x,y)
Figure 10. (a) Schematic classical electron trajectories in a magnetic field.

(b) Corresponding quantum picture of the energy states of Eq. (10) along a cross-section
of the 2DEG, illustrating the formation of edge channels at the boundary of the 2DEG
(formed by the electrostatic potential energy eV(x, y)) where the Landau levels intersect
the Fermi energy EF.
In the absence of an electric field E, the balance 'of the Lorentz force
FL
= evB and the centripetal force F = m*v2/r leads to a cyclotron motion
of the electrons, with an angular frequency We = eB/m* and at the Fermi

energy a cyclotron radius re = m*vF/eB (see Fig. lOa). When the electric
field E = - \7V(x, y) '# 0, the electrons have a net drift velocity VD = E / B.
At the boundary of the sample, where E is large, the collisions at the edge
result in skipping orbits. The electrons skip with the drift velocity along
the edge of the sample. The direction of the velocity is opposite for the two
opposite edges (see Fig. lOa).
The skipping orbit motion of electrons along a 2DEG boundary in a
small magnetic field has been observed in an electron focusing experiment
by van Houten et al. [20]. The geometry shown in Fig. 5b with two adjacent QPCs with a separation of L = 3JLm was used, where one QPC
injects electrons into the 2DEG and the second QPC collects them. The
injected electrons are focused by the magnetic field on the boundary between injector and collector at distances 2p re(P = 1,2, ... ). Focusing into
the collector occurs when 2p re = L. In the experiment the collector voltage
is measured as a function of magnetic field B. From the condition 2p r e = L
it follows that the largest number of electrons reaches the collector when
Bfoe = 2pm*vF / eL which leads to periodic oscillations in the collector signal. This is shown in Fig. 11 together with a calculation. On average the
voltage increases linearly with magnetic field, which'is expected for the
classical Hall resistance VHall/ I = B / ens. For curve (a), however, we see
14
1.0
ec 0.5
~
O~----------~~~--------~
-o.5~
-0.3 -0.2 -0.1
~~
B (T)
0.1
0.2
0.3
Figure 11. Bottom: Electron focusing (T = 50 mK, L = 3.0JJm) in the configuration

depicted in the inset. The two traces a and b are measured with interchanged current
and voltage leads, and demonstrate the injector-collector reciprocity as well as the reproducibility of the fine structure. Top: Calculated classical focusing spectrum corresponding
to the experimental trace a (50 nm wide point contacts were assumed). The dashed line
is the extrapolation of the classical Hall resistance seen in reverse fields. (From Ref. [20].)
additional large oscillations with fine structure around the average Hall resistance for positive magnetic field. These oscillations occur at the expected
focusing fields. For a negative magnetic field no focusing signal is observed
since now the injected electrons are deflected away frQm the collector.
Curve (b) is taken with the current and voltage probes interchanged.
The relation between (a) and (b) demonstrates Biittiker's reciprocity relation [21],
(9)
It implies that upon interchange of the current probes i and j with the
voltage probes k and 1, one obtains the same resistance at the opposite
magnetic field. This fundamental symmetry relation is clearly confirmed
in Fig. 11, including the symmetry in the fine-structure. The latter has
been explained by extending the classical focusing calculation of the top
section to include quantum interference effects between trajectories being
injected at different angles [20]. This focusing experiment demonstrates
that the collisions at the boundary between injector and collector are highly
specular, since diffusive boundary scattering would average the oscillations.
In a subsequent experiment by Spector et al. [22] focusing signals were
observed up to a distance between injector and detector as large as 64 J1m.
15
1.6. THE QUANTUM HALL EFFECT ON MESOSCOPIC LENGTH SCALES
1.6.1. Edge Channels

In a high magnetic field the electron motion is quantized. The flux <P enclosed by an electron in a cyclotron orbit equals an integer times the flux
quantum <Po = hie, and the quantized electron energies are:
(10)
where n = 1,2, ... is the spin-degenerate Landau level index, and we have
ignored the Zeeman energy splitting g/1BB/2. We assume that the electrostatic potential V (x, y) is flat in the interior of the sam pIe and rises at
the boundary. Electrostatic variations due to impurities are ignored, because we are dealing with ballistic samples. Fig. lOb .shows schematically
the Landau energy levels of Eq. 10. The electron states at the left boundary
are occupied up to /11, the electrochemical potential of the current source,
and at the right boundary up to /12, the electrochemical potential of the
current sink (see Fig. 4b). At the two sample boundaries, the electron states
have opposite velocity directions, similar to the classical case of Fig. lOa.
The relevant electron states for linear transport are only those at the
Fermi energy. As can be seen, these are located at the sample boundaries,
where the Landau levels intersect the Fermi energy (En = EF), and they
extend in the direction perpendicular to the cross-section of Fig. lOb. The
intersections are the current-carrying states, which are known as edge channels [17, 23, 24]. The net current I only results from the uncompensated
states in the interval between /11 and /12, The total current carried by
the states below /L2 is zero. The transport through edge channels is 1dimensional [17]. Edge channels can therefore also be viewed as lD current
channels, each carrying a current In = 2e/h(/L1 - J.L2). With the Hall voltage VH = (J.L1 - J.L2)/e measured between the two sample boundaries, this
directly gives the quantized Hall conductance GH = NLIn/VH = NL2e 2 /h.
Up to now we have ignored all scattering processes. Biittiker [25] has
pointed out that due to the spatial separation of the electron states with opposite velocity, backscattering requires scattering from one sample boundary to the other. Backscattering is therefore suppressed when the edge
states between /11 and /12 are not connected by extended electron states.
This is the case in Fig. lOb, where the Fermi energy is between two bulk
Landau levels.
1.6.2. Selective Probing of Edge Channels
The above description of transport in the quantum Hall effect (QHE)
regime, known as the Landauer-Biittiker formalism, is reviewed in Ref. [26].
This edge channel description gives an appealing physical picture of the
16
QHE. The question now arises about how to prove the existence of edge
channels and whether they can really be viewed as independent current
channels. We now discuss an experiment involving two adjacent QPCs
which directly probes the transport through a particular edge channel. The
theory for this experiment provides a simple illustration of Biittiker's multiprobe formalism which has been used very successfully to describe different
kind of mesoscopic transport phenomena.
In Section 1.5 we discussed the electron focusing from one QPC to an
adjacent second QPC by small magnetic fields (B ::; IT). These fields are
too small to quantize the electron motion, and the focusing can be explained
in terms of classical cyclotron motion. We consider now the same geometry
in the high field regime (B 2 IT). Fig. 12a shows schematically two adjacent QPCs A and B defined in a 2DEG with an applied magnetic field such
that two edge channels are occupied. The 2-terminal conductances GA and
GB of the individual QPCs measure the number of transmitted channels
and are quantized in multiples of 2e 2 /h (see also Fig. 9). The Hall conductance GH is normally thought to be independent of the characteristics of
the current and voltage probes and to correspond directly to the number of
occupied (spin-degenerate) Landau levels NL in the 2DEG, GH = NL2e 2 /h.
This is not true for the situation in Fig. 12a and we discuss here that for
describing transport through mesoscopic conductors one has to include the
properties of the measurement contacts.
B_3.ST
Va _ " .86 V
OL.~2.6L-----'2.~2--~~.1~
.B--~--~1.4~--~
GATE VOLTAGE VA
(V)
Figure 12. Left: Geometry used to observe an anomalous quantization of the Hall
conductance, for which QPC A is used as current probe and QPC B as voltage probe.
Right: Comparison between the Hall conductance GH and the 2-terminal conductance
of the current probe GA. The voltage probe conductance is kept fixed at GB = 2e 2 /h.
Although the number of occupied Landau levels in the 2DEG is unchanged, GH follows
the largest probe conductance (from Ref. [27])
We first assume that the edge channels shown in Fig. 12a are independent, i.e. no scattering events occur between different edge channels
or, equivalently, electrons travel with conservation of quantum-subbandnumber. In this case the transport is adiabatic. The edge channels are oc-
17
cupied up to the electrochemical potential of the last Ohmic contact they

have left. (We assume ideal Ohmic contacts, meaning that all incoming
electrons are absorbed and all outgoing states are occupied up to an average electrochemical potential.) Consequently, in the region between the
two QPCs, the two edge channels have an unequal population. Channel-2 is
occupied up to f.11, while current probe 5 populates channel-1 up to f.15. This
process is called selective population of edge channels. If the voltage probe
B detects all edge channels like an ideal Ohmic contact, the regular quantum Hall conductance is measured. However, this is not the case in Fig. 12a
where the voltage probe selectively detects only the first channel. The second channel is neither populated by the current probe nor detected by the
voltage probe and, therefore, is not measured. The Hall conductance G 54 ,61
equals 2e 2 jh instead of 4e 2 jh which a regular Hall measurement would give.
The current and voltage contacts do not measure all the 2DEG properties,
but only those properties they "see" , or couple to. In general one can show
that in the absence of inter-edge channel scattering this Hall conductance
is given by [27]
(11)
implying that GH is completely determined by the characteristics of the
probes and is independent of the number of occupied Landau levels in the
2DEG.
Fig. 12b compares the measured probe conductances G A and GB with
the Hall conductance GH. The magnetic field is kept fixed at 3.8 T, corresponding to NL = 2 in the bulk 2DEG. The voltage on QPC B, defining the
voltage probe, is also fixed such that only the first edge channel is transmitted, and therefore GB = 2e 2 jh. The voltage VgA on QPC A, defining the
current probe, is varied, resulting in a decreasing GA. Again we note that a
normal Hall measurement would give a constant conductance GH = 4e 2 jh.
However, Fig. 12b shows a Hall conductance which virtually follows the
largest probe conductance in agreement with Eq. 11 (for VgA > -1.5V,
GH = GA > 2e 2 jh; for -2.1V < VgA < -1.5V, GH = GA = GB = 2e 2 jh
corresponding to the edge channel flow of Fig. 12a; and for VgA < -2.1 V,
GH = GB = 2e 2 j h). The experiment demonstrates that on short distances
of order f.1m the transport through edge channels is adiabatic, implying
that they can be viewed as independent 1D current channels. Subsequent
experiments have shown that a non-equilibrium population can persist up
to larger distances (several tens of f.1m 's) [28]. In particular, it is found that
the top most channel is virtually decoupled from the lower edge channels
even over macroscopic distances of several times 100 f.1m [28,29,30,31,32].
These experiments have clarified the important role of measurement probes
in the QHE regime.
18
1
8=7T
,...
.!! 2
3'
o~
__
____
-1.5
______
-1
-D.5
____
GATE VOLTAGE IV)
Figure 13. Two-terminal conductance of a quantum point contact at B = 7T. At zero

gate voltage the conductance is G = e 2 /h implying that the 2DEG is spin-polarized at
filling factor 1. On decreasing the gate voltage fractional plateaus are observed (from
Ref. [37].
1.6.3. The Importance of Electron-Electron Interactions for the QHE on

Mesoscopic Length Scales
In our discussion we neglected Coulomb interactions between the electrons.
For many experimental observations it is sufficient to give a description
in terms of a non-interacting electron model. There are several exceptions,
however. First of all, in very high mobility samples not only are plateaus
observed at integer filling factors, but also at fractional filling factors (see
Fig. 2). The theory for this fractional QHE needs to incorporate electronelectron interactions [33]. Recently, a new appealing formulation for the
fractional QHE has been given in terms of composite particles [34]. Another
recent development is the prediction [35] and observation [36] of so-called
skyrmions.
Also, on mesoscopic length scales it has recently become appreciated
that electron-electron interactions can be important under certain circumstances. First of all, the FQHE can also be observed at short ballistic length
scales [37]. Fig. 13 shows the conductance of a QPC at 7 T. At zero gate
voltage the conductance is equal to e2 /h showing that the filling factor of
the bulk 2DEG is equal to one (i.e. spin resolved). On making the gate
voltage more negative the conductance decreases. However, at fractional
values 2/3 and 1/3 times e2 / h it shows quantized plateaus. The QPC used
in this experiment shows quantized conductance at B = 0 implying that
fractional plateaus can exist in a ballistic sub-micron device [37,38]. Also,
selective population and detection experiments of fractional edge channels
have been performed [39, 40] in a similar way as in Section 1.6.2 for the
case of integer edge channels. These experiments on the fractional QHE in
19
the ballistic regime can not be explained with the non-interacting electron
models of Sections 1.6.1 and 1.6.2. Instead interacting models have been
proposed by Beenakker [41], MacDonald [42] and Chang [43]. These models
are the basis of recent models on self-consistent calculations of edge channels [44]. From these electrostatic calculations it follows that edge channels
have a finite width. Regions where the Fermi energy is pinned in a Landau
level are called a compressible liquid whereas regions where the Fermi energy is between Landau levels are called an incompressible liquid. Although,
Coulomb interactions give edge channels a finite width the selective population experiments suggest that they remain 1D channels. In 1D Coulomb
interactions are known to give rise to a new type of electron liquid. This
so-called Luttinger liquid is discussed by Fisher and Glazman in this volume including predictions for correlated electron transport in fractional
edge channels. Altogether, we can conclude this Section with the statement
that the QHE is still a very interesting system for studying new electronic
properties.
2. Single-electron tunneling
2.1. REVIEW
The electron transport in two-dimensional electron gases displays the quantum coherence of the electronic states, as described in the previous Section.
Another important aspect of mesoscopic systems is the role of interactions.
We mentioned already their importance in connection with the fractional
QHE. Interaction effects also influence strongly electron tunneling through
mesoscopic systems, and they lead to strong correlation effects. In this
Section we will discuss situations where it is sufficient to approximate the
Coulomb interaction by an effective capacitance model. This applies for
metal junction systems where after a relaxation of the space distribution
of the electrons the remaining Coulomb interaction is well described by
the geometric capacitance of the junctions. The capacitance model also
works remarkably well for the transport through small quantum dots. (See
the Chapter of Kouwenhoven et al. for many facts on quantum dots.) The
"charging effects" allow us to control single electron charges, which leads
to a variety of single-electron effects, e. g. to the suppression of tunneling,
a phenomenon known as the Coulomb blockade.
Single-electron effects have been studied for more than a decade, and
a large number of papers, incl. several reviews have been devoted to the
topic. Kulik and Shekhter [45] and Averin and Likharev [46] developed
the perturbation theory of single-electron tunneling in metal junctions and
discussed several consequences. Initial scepticism against the new concepts
20
was overcome when experiments proved to be successful. After early experiments by Fulton and Dolan [47], important breakthroughs were achieved in
Delft by Mooij and Geerligs and in Saclay by Devoret, Esteve, Urbina and
further members of these groups [48, 49]. Their work is well summarized in
the proceedings of an earlier NATO ASI Single Charge Tunneling [50].
In this Section we introduce the concepts and basic description of singleelectron tunneling in systems with strong charging effects. For definiteness
we first consider metallic electrodes with a large density of electron states.
We study how the charging energy depends on the number of electrons
and the transport and gate voltages applied to various parts of the system.
The simplest model systems demonstrating these features are the "singleelectron box" and the "single-electron transistor". We then derive within
perturbation theory the single-electron tunneling rates. In low capacitance
systems it is crucial to account for the change in the charging energy associated with the tunneling process. A master-equation description accounts for
the large-scale features of the current-voltage characteristic of the singleelectron transistor. In the Coulomb-blockade regime, where single-electron
tunneling is suppressed, higher-order processes such as coherent "cotunneling" of electrons through several junctions become observable. Finally we
discuss single-electron tunneling through quantum dots, where the level
quantization becomes observable as well ..
Several extensions have been described in the literature. We mention
only a few:
- The mesoscopic junction systems studied here are small such that charging effects and higher-order quantum processes play a role. On the other
hand, they are large enough such that macroscopic current and voltage
probes and sources can be coupled to the system. This makes the mesoscopic system susceptible to the influence of the electric circuit. A complete
description has to include this circuit. The influence of the electrodynamic
environment on single-electron tunneling has been reviewed in the article
by Ingold and Nazarov in Ref. [50].
- In this introduction we study tunnel junctions with two normal-conducting
electrodes (NN). If the system or part of it is superconducting the combination of single-electron tunneling, Cooper pair tunneling and Andreev
reflection leads to further highly interesting effects [51]. For a review see
the Chapter of Fazio and Schon in this volume.
- Many of the single-electron effects can be described within simple perturbation theory. A necessary requirement is that the resistance of the tunnel
barriers is high compared to the quantum resistance RK = h/e 2 :::::: 25.8 kQ.
For tunnel junctions with lower resistance a more general formulation is
needed [52, 53, 54]. A systematic description of tunneling in systems with
strong charging effects is presented later in the Chapter of Schoeller.
21
(a)
(b)
---I[IJf----
Figure 14. a) An overlap junction with an oxide layer (hatched region), b) schematic
diagram for a tunnel junction.
2.2. CHARGING ENERGY AND SINGLE-ELECTRON DEVICES
2.2.1. The Energy Scale

Modern lithographic techniques allow the fabrication of narrow metallic
lines with widths down to approximately 20 nm, as well as tunnel junctions
in overlap regions of such lines, as illustrated in Fig. 14. The structures are
grown by evaporation of the metal, e. g. Aluminum, through masks onto
the substrate. Tunnel junctions can be produced by shadow evaporation
techniques, which involves two stages of evaporation at different angles.
Between the two stages the first layer is allowed to oxidize. The junction is
formed in the overlap region. They can be produced reliably with areas of
S = (100 nm)2 and below. The oxide layer is roughly d = 10 A thick, and
the dielectric constant of the oxide is E ~ 10. Using the classical expression
for the capacitance we arrive at C = ES/(47rd) ~ 10- 15 F.
The capacitance introduces an energy scale, the charging energy corresponding to a single-electron charge (-e),
e2
Ec == 2C'
(12)
which characterizes all charging effects. It is of the order of Ec ~ 1O-4 eV

if the capacitance is C = 1O- 15 F , which corresponds to a temperature
Ec/kB ~ 1K. In a tunneling process the electrostatic energy changes by an
amount of the order of magnitude of Ec. Hence we expect in the sub-Kelvin
regime electron transport to be affected by charging effects. Similar properties have been observed in semiconductor nanostructures, for instance
in quantum dots in 2-dimensional electron gases. The Coulomb energy in
these systems again can be characterized by a capacitance which depends
on the size of the dot and also may lie in the range of 10- 15 F or less.
Charging effects play a role in granular materials and ultimately even in
molecular systems. Here the capacitance may be as low as 1O- 18 F, making
single-electron tunneling observable even at room temperatures.
2.2.2. Single-Electron Box
We now analyze in more detail the charging energy of simple systems of
tunnel junctions. It depends on the electron number in various parts of
the particular system and the applied voltages. The first example is the
22
Figure 15.
The single-electron box.
single-electron box, shown schematically in Fig. 15. It consists of a small

metallic island, coupled via a tunnel junction with capacitance CJ to an
electrode and via a capacitor CG to a gate voltage source VG. We choose
the reference such that for VG = 0 the lowest energy state of the system is
charge-neutral, i. e. the electrons on the island compensate the charge of the
ions; consequently there are n = 0 excess electrons on the island. If a gate
voltage is applied the number of excess electrons on the island can change
due to tunneling across the junction in discrete steps to n = 1, 2, ....
While the total number of electrons on an island is integer, the charge
is spatially distributed and in general shifted relative to the positive background. If a voltage is applied the surface charges on the capacitor plates,
which are of equal magnitude but opposite sign on the two sides of each
junction, are in general non-quantized. They are determined by the integer
n and the non-quantized applied voltage. We obtain the charging energy
from the following elementary arguments: the total excess charge of the
box splits into two parts on the left and right capacitor plate -ne =
QL + QR. The corresponding voltage drops add to the applied voltage
VG = QL/CJ - QR/CG, and the charging energy is Ql!2CJ + Q~/2CG.
The relevant free energy is a Legendre transform of this energy, which also
contains the work done by the voltage source - VGQR. Elimination of QL
and QR in favor of nand VG yields, up to a contribution which does not
depend on the variable n, the result
E ( Q) _ (ne - QG) 2
ch n, G 2C
(13)
Here C = CJ + CG is the total capacitance of the island. The effect of the

voltage source is contained in the "gate charge" defined as QG = CG VG.
The charging energy Ech is plotted in Fig. 16 as function of the gate
charge for different excess electron numbers n. With increasing gate voltage,
the electron number corresponding to the lowest energy state increases. It
does so in discrete steps from n to n + 1 at the degeneracy points QGle =
23
Figure 16. The charging energy of a single-electron box as a function of the gate voltage
for different numbers n of electron charges on the island.
1.5
0.5
0
-0.5
-1
-1.5
-1.5
-1
-0.5
0.5
1.5
Figure 17. The average number of electron charges (n) on the island of a single-electron
box as a function of the gate charge (voltage) for different temperatures T / Ee = 0
(dashed steps), 0.02, 0.05, 0.1, 0.2, 0.4, and 1 (nearly linear).
n + 1/2. Under the same conditions, the voltage of the island, Visland =
BEch/ BQG, displays a sawtooth dependence on the applied voltage.
At finite temperatures the steps and sawtooth dependence are washedout, as follows from the classical statistical average
(n(QG)) = -
L
00
ne-ECh(n,QG)/kBT,
(14)
Zch n=-oo
where Zch is an obvious normalization. The result is displayed in Fig. 17 for

different temperatures. The stepwise increase has been observed experimentally, e.g. by the Saclay group (see results in Ref. [50]). Their measurement
24
CR
CO
r--_ _l-+_v_G_--i1 r-V_R--J
Figure 18.
The SET transistor.
procedure will be discussed below. The experiments agree well with theoretical expectations. However, it is crucial to control heating and the noise
from the measurement setup, which usually is at a temperature higher than
that of the cryostat.
2.2.3. Single-Electron Transistor

Another fundamental example is provided by the single-electron transistor
shown in Fig. 18. Here an island is coupled via two tunnel junctions to
a transport voltage source V = VL - VR such that a current can flow.
The island is, furthermore, coupled capacitively to a gate voltage Vc. The
charging energy of the system depends again on the integer number of
electrons n on the island and the continuous voltages. Some algebra along
the lines outlined for the electron box produces again Ech(n, Qc) = (ne Qc)2/2C. For the transistor C = CL + CR + Cc is the total capacitance
of the island, i.e. the sum of the two junction capacitances and the gate
capacitance, and all three voltage sources define the gate charge
(15)
In a tunneling process, increasing the island charge from n to n + 1, the
charging energy changes by
(16)
The second energy differences are equally spaced and can be tuned by the
gate voltage. The situation is illustrated in the energy scheme shown in
Fig. 19. The differences in charging energy are plotted in the center. We
further display the Fermi levels of the two leads which are shifted by the
applied potentials VL / R .
25
Ech (2,QJ - E ch (1,Ou)

Ech (1,QJ - Ech(O,QJ
Ech(O,QJ - Ech(-l,QJ
Ech(-l,QJ - E ch(-2,QJ
eVa
Figure 19. In the island the energy differences corresponding to the addition or removal
of an electron charge are shown. They can be shifted by the gate voltage VG. The electrochemical potentials of the leads are shifted relative to each other by the transport
voltage V = VL - VR.
For definiteness we assume that the energy of the electrons in the left
lead is higher than that in the right lead. Then at low temperature tunneling
from the left lead to the island (transition from n to n + 1) is possible if the
electrochemical potential in the left lead eVL is high enough to compensate
for the increase in charging energy of the island
(17)
Similarly tunneling from the island (transition from n + 1 to n) to the right
lead is possible at low temperature only if
(18)
Both conditions have to be satisfied simultaneously in order for a current
to flow through the transistor. It is obvious from the figure that at low
transport voltages, depending on the gate voltage VG we may be either in
a Coulomb blockade regime or have a finite current. By varying the gate
voltage we produce the Coulomb oscillations, i.e. the e-periodic dependence
of the conductance on QG.
Additional devices can be examined (for a review see Esteve's article
in Ref. [50]): (i) The electron trap is similar to the electron box except
that it contains at least two junctions in series. In contrast to the box
the trap has metastable charge states. (ii) Two traps can be combined to
build the electron turnstile, which can serve as a current source [48]. A
suitable ac-gate voltage with frequency f allows the controlled transfer of
26
a single-electron per cycle. Hence the current is J = ef. (iii) Finally, in

single-electron pumps a current is driven by two phase-shifted ac-voltages
applied to different islands [49]. In this case a current J = e/ is transported
even at vanishing transport voltages. The turnstile and pump can serve
as a current standards, if one manages to minimize the effect of missed
cycles, of thermal fluctuations, and of quantum fluctuations. This !requires
low frequencies, low temperatures, and a design (many junctions) which
minimizes higher order quantum tunneling processes ..
Many properties of single-electron systems can be understood by considering only the energy of different charge configurations. However, a detailed understanding of the J- V characteristics requires knowledge of the
tunneling rates of the electrons, which will be next topic.
2.3. TUNNELING RATES AND I-V CHARACTERISTICS
In this Subsection we introduce the Hamiltonian of the SET transistor.

Using simple golden-rule arguments we derive the rate for the transfer of a
single electron charge across the tunnel barriers. It depends crucially on the
change in the charging energy. The transition rates enter a master equation,
from which we obtain the current-voltage characteristic. If a tunneling process would increase the charging energy it is suppressed at low temperature.
This phenomenon is called "Coulomb blockade". This "orthodox theory"
was developed in Refs. [45,46].
2.3.1. The Single-electron Tunneling Rate

For definiteness we consider a SET transistor, shown in Fig. 18, which
consists of a metallic island coupled via tunneling barriers to two leads and
capacitively to an external gate voltage. Its Hamiltonian is
(19)
Here, HL = I:k,a Ekck,aCk,a describes the noninteracting electrons with wave
vector k in the left lead, with similar expressions for the island HI (with
states denoted by q) and the right lead HR. We allow that the leads have different electrochemical potentials. The Coulomb interaction Hch is assumed
to depend only on the total charge on the island, as discussed above,
H
_ (he - QG)2
2C
.
ch -
(20)
The number operator of excess electrons on the island is given by h =

I:q,a ctaCq,a - N+, where the number of positively charged ions of the
island has been subtracted. Charge transfer processes are described by the
27
standard tunneling Hamiltonian, for instance tunneling in the left junction
between the states k and q by
Ht,L
Tk,qCLcq,(7
+ h.c ..
(21)
k,q,(7
We determine the transition rates by golden-rule arguments. The tunneling rate of an electron in the left junction is
It describes the tunneling from one of the many states k in the left lead
to one of the available states q in the island, In this process the electron
number is increased from n to n + 1. The crucial point is that the energy,
which is conserved as expressed by the 15-function, contains the energies of
the electron states Ek/q, but also the charging energy. The latter depends on
the electron number and the applied voltages VG and VL/ R . In the process
considered it changes by
(23)
We further introduced the tunnel conductance of the left junction
1
41re 2
R = -h-N1(O)nINL(O)fkITI
t,L
(24)
It depends on the tunnel matrix elements Tk,q, which at this stage can
be considered as constants, as well as the densities of states at the Fermi
level, N1/dO) , and the volumes, nI / L , of the island and lead. Equivalent
expressions apply for the reverse process IL" (n + 1), decreasing the island
charge from n+ 1 to n, and for the tunneling processes in the other barriers.
In equilibrium the distribution functions fI/L are Fermi functions, and
the integrals over the electron states in Eq. (22) can be performed explicitly.
The resulting "single-electron tunneling" (SET) rate is [46]
I
15Ec h
+( n) - -1- ---:-:,-----:-:-----::--
e2 Rt,L exp[8Ech/ kBT] - 1
(25)
At low temperatures, kBT <t: 115Echl, if the charging energy would increase
in a tunneling process, the tunneling is suppressed, I --t O. This phenomenon is called "Coulomb blockade" of electron tunneling. If charging
energy is gained the rate is
1 I15EchI for 15Ech:<::; 0 ,T --t 0 .
IL+ (n) -_ ~R
t,L
(26)
28
At finite temperatures all processes are allowed. The forward and backward
rates satisfy detailed balance, ,t(n)/'L(n + 1) = exp[-8Ech/kBT].
A familiar limit of what is described above is a single voltage-biased tunnel junction, where 8Ech is replaced by -eV, independent of n. In this case
(25) yields a linear current- voltage relation, It = e[,+ - ,-] = V/ R t . We
can also reverse the argument. The two requirements - (i) a linear characteristic in the voltage-biased case and (ii) detailed .balance - uniquely
determine the expression for the rate to be of the form (25).
2.3.2. Master Equation for Sequential Tunneling

Given the electron tunneling rates we can set up a master equation for
the probability P( n, t) to find the island in a state with n electrons. The
probability changes by tunneling in the left and right junctions. Hence
d
dt P(n, t) =
[,t(n) + 'L(n) + ,~(n) + 'R(n)] P(n, t)

+
+
[,t(n - 1) + ,~(n - 1)] P(n - 1, t)

[,L(n+1)+,R(n+1)]P(n+1,t).
(27)
The rates and probabilities also determine the current. In the left junction
the current is
h(t) = -e :l),t(n) -'L(n)] P(n, t) .
(28)
In most ~ases we apply dc-voltages and are interested in the dc-current. In

this case we need only the stationary solution of the master equation, and
the currents in the left and right junctions are equal J = h = JR.
As an example we consider a junction with symmetric bias VL = - VR =
V /2. At low temperatures and low transport voltages (except at symmetry
points) only two different charge states - and those transitions which connect both - have an appreciable probability. For instance, if ne < QG <
(n+ l)e we need to consider only P(n) and P(n+ 1) and the two transitions
,t (n) and l~ (n) increasing the island charge from n to n + 1 electrons, as
well as the two reverse transitions 'L (n + 1) and 'R (n + 1). The energy
(n) and 'L (n + 1) are
changes determining the rates
,t
8Ech
QG) eC2-""2
ev] '
= [( n + '12 - -;-
(29)
respectively, while for the transitions in the right junction eV is replaced

by -eV. In the 2-state limit the stationary probability and current become
P(n)
= 'L(n+1)+'R(n+1);
'L
P(n+1)=1-P(n)
29
Figure 20. The current of a symmetric transistor is shown as a function of gate and
transport voltage. At low temperatures and low transport voltages VOle < 1 only two
charge states playa role, and the Coulomb oscillations are clearly demonstrated. At larger
transport voltages more charge states are involved.
-e it(n)'yR:(n + 1) -i~(n) iL(n + 1)

i~
(30)
where i~ = it(n) + i~(n) + iL(n + 1) + iR:(n + 1).

This expression is readily analyzed by inspection of (29). At low temperatures the tunneling process in the left junction from n to n + 1, with
rate it(n), is allowed when QG - (n+1/2)e ~ -VC/2. On the other hand,
the transition which carries on the charge to the right electrode with rate
iR:(n + 1) is allowed when QG - (n + 1/2)e S; VC/2. Both coexist in a
window of width CV around QG = (n+ 1/2)e. The other two processes are
not allowed simultaneously, in fact they are suppressed in the window just
mentioned. Therefore, at low temperature the current is
[v _~
(QG - n C2V
e
~)2l
2
for _ VC < QG -n- ~ < VC

2e - e
2 - 2e '
(31)
while it vanishes outside the window. For simplicity we have assumed in (31)
that the two junctions have the same tunneling resistance R t = Rt,L = Rt,R.
The current through a symmetric SET transistor is plotted as a function
of transport and gate voltages in Fig. 20. For gate voltages such that QG/e
is close to an integer, the current vanishes below a threshold bias voltage
Vth(QG = ne) = e/C. This is a manifestation of the Coulomb blockade.
At non-integer values of QGle the threshold voltage is lower vth(QG) =
1=_1
4Rt
30
3
2.5
--
r:i+
1.5
..J
~
......
0.5
00
1.5
2
2.5
VC/e
3.5
Figure 21. Coulomb staircase: The current of an asymmetric transistor with different
tunneling resistances in the two junctions Rt,R = 10Rt,L is shown as a function of the
transport voltage for QG = 0 (pronounced Coulomb blockade, QGle = 0.25 (intermediate), and Q G Ie = 0.5 (linear at low voltage).
minn {2IQG - (n + 1/2)el/C}. One finds a series of evenly spaced peaks

centered around half-integer values of QG/e = n + 1/2, each of parabolic
shape as given by Eq. (31). These are called "Coulomb oscillations".
The strong dependence of I(QG, V) on the gate voltage makes the SET
transistor a highly sensitive "electrometer". Small changes of polarization
charges by fractions of an electron charge influence a macroscopic measurement current. It has been used, for instance, to measure the charge in the
electron box (n(QG)) in Fig. 15.
For larger transport voltages more charge states play a role even at
low temperatures. For illustration we consider a junction with symmetric
bias VL = - VR = V /2 and QG = 0, where the lowest energy state has
n = 0 electrons in the island. At transport voltages exceeding a threshold
yth,O = e/C tunneling sets in to the state with n = 1. Above this voltage
the electrochemical potential in the left lead is sufficient to compensate
the increase in charging energy of the island. Since this state with n = 1
is unstable against a tunneling process in the right junction, a current is
transported through the system. At the same voltage tunneling processes
involving the state with n = -1 are possible. At still higher voltages further
charge states Inl 2 1 playa role. This leads to a series of voltages yth,n =
(2n+l)e/C, each marking the threshold above which another pair of charge
states becomes populated and a new channel for the conductance opens.
The increase in conductance is limited, however, due to the normalization
condition for the P( n). Still for suitable parameters (differing conductances
of the two junctions or different capacitances) the current increases in the
31
shape of a staircase. The phenomenon got accordingly the name "Coulomb

staircase" [55]. The behavior is demonstrated in the plot of Fig. 21.
2.4. HIGHER-DRDER TUNNELING PROCESSES
If sequential single-electron tunneling is suppressed by the Coulomb blockade, higher-order processes such as coherent "cotunneling" through several
junctions becomes crucial (Averin and Nazarov in Ref. [50]). As a specific example we consider a SET transistor, biased slich that the current
in lowest-order perturbation theory vanishes (see Fig. 19). At low temperatures sequential tunneling is exponentially suppressed in this regime
since the energy of a state with an excess charge on the island lies above
the Fermi levels of the leads. On the other hand, if a transport voltage is
applied, a higher-order tunneling process transferring an electron charge
coherently through the total system is energetically allowed. In this case
the state with an excess electron charge in the island exists only virtually.
Standard second-order (or fourth, depending on the counting) perturbation
theory yields the rate
(32)
The energy of the intermediate virtual state lies above the initial one,
E1jJ - Ej > 0, but it enters only into the denominator rather than into
the exponent of the sequential tunneling rate. Hence the higher-order rate
is nonzero even at very low temperatures.
When analyzing the process we have to pay attention to the following:
(i) There are actually two channels which add coherently. Either an electron
tunnels first from the left lead onto the island, and then an electron tunnels
from the island to the other lead. In this case the increase in charging
energy of the intermediate state compared with the initial one is OEL =
ECh(n + 1, QG) - Ech(n, QG) - eVL. Or an electron tunnels first out of the
island to the right lead, and another electron from the left lead replaces
the charge. In this case the increase in energy of the intermediate state is
OER = Ech(n - 1, QG) + eVR - Ech(n, QG). Both amplitudes have to be
added before the matrix element is squared.
(ii) The leads contain a macroscopic number of electrons. Therefore, with
overwhelming probability the outgoing electron will come from a different
state than the one which the incoming electron occupies. Hence, after the
process an electron-hole excitation is left in the island, which explains why
it is called "inelastic" cotunneling. This scenario is visualized in Fig. 22.
32
o
o
Figure 22. In an inelastic cotunneling process two electrons tunnel coherently, i.e. in a
single qutnum process, in the left and right junction. The result is the transfer of charge
through the system even in the Coulomb blockade regime. A particle-hole excitation is
left in the island.
Transitions involving different excitations are added incoherently. The

resulting rate for inelastic cotunneling is
{cot
JkEL
dEk
JqEI
dEq
x[ + 8~
Eq
L - Ek
JqlEI
dEql
21l"e
Jk/ER
+ Ek' + 8~R -
4R t LR
,
t ,R
dEklf(Ek)[l-f(Eq)]f(Eql)[l-f(Ekl)]
EQI
j28(eV+Ek-Eq+EQI-Ekl).
(33)
At T = 0 the integrals can be performed analytically. The result for in the

Coulomb blockade regime (eV ~ 8EL, 8ER) is
Icot
12uRt ,LRt ,R
(1- - + --1)2 V
8EL
8ER
3 .
(34)
At finite temperatures forward and backward processes occur. They obey

a detailed balance relation Icot(-V) = exp(-eV/kBThcot(V). The current
then is
In the Coulomb blockade regime of a SET transistor the V 3 dependence of the cotunneling current has been observed. In systems with N
junctions a corresponding N-th order process (or 2N-th order, depending
on the counting) leads to a current I ex V 2N -1. As an exam pIe we consider N=4 junctions with C = 1O- 15 F and tunneling resistance R t . In this
33
case (see D. Esteve in Ref. [50]) Icot = (2.5 x 1O-3/sec) (v/J.lVf (kO/R t )4.
These cotunneling processes limit the accuracy of the single-electron turnstile even under the most favorable situations, i.e. low T and low frequency,
where thermally activated multi-electron transfer processes and missed cycles play little role.
The expression for the cotunneling rate presented above displays several important properties: (i) The expansion parameter is the dimensionless
tunneling conductance RK/ Rt , where the quantum resistance RK serves as
reference. (ii) The approximate expression given diverges when the intermediate and initial or final states are degenerate. This divergence is removed by
life-time broadening effects. The complete cotunneling theory, recently analyzed in Ref. [56], describes well the logarithmic temperature dependence
observed in th experiments of the Saclay group on junctions with tunneling
resistances comparable to the quantum resistance [57]. (iii) Higher order
processes and eventually resonant tunneling processes are most essential
near the points of degeneracy of the charging energy, QG/e = n + 1/2. The
Chapter of Schoeller in this volume presents the theoretical framework to
describe tunneling beyond perturbation theory.
There exists also the process of "elastic cotunneling" where one electron
tunnels through the total system, leaving no excitations in the island. It is
the dominant process in the Coulomb blockade regime of tunneling through
a single-level quantum dot. However, in the metallic junction it is usually
not important, since its rate is smaller by a prefactor ex 1/0INI(0) (i.e.
inversely proportional to the number of states of the island) compared to the
inelastic cotunneling rate. The exception is the range of very low voltages
and temperatures kBT, eV ~ JEc/ON(O), since elastic cotunneling yields
a current which is linear in the applied voltage.
Single-electron tunneling is also influenced by the response and the fluctuations of the electromagnetic circuit where it is embedded. This is particularly important in single junctions. In this case Coulomb blockade effects
can only be observed if the junction is in series with a nearby large resistor
of the order of the quantum resistance or larger. In systems of junctions the
tunneling resistance of one junction usually provides the required decoupIing of the remaining junctions and single-electron effects are observable.
Here we do not have the space to present this theory. It is reviewed in the
article by Ingold and Nazarov in Ref. [50].
2.5. SINGLE-ELECTRON TUNNELING THROUGH QUANTUM DOTS
In this Subsection we describe the influence of Coulomb blockade phenomena on single-electron tunneling through ultrasmall quantum dots. The
important difference compared to the metallic case is the quantization of
34
the energy levels inside the quantum dot with typical separation 8. It can be
resolved in transport experiments, when the level spacing exceeds the temperature 8 > kBT. In this case one encounters the phenomenon of resonant
tunneling through discrete levels.
Many of the experiments showing single-electron effects in quantum
dots can be explained by lowest order perturbation theory in tunneling. As
in the metallic case, the theory is based on a classical master equation with
golden-rule rates which describe incoherent transport through the device.
This means that the electrons tunnel sequentially, i.e. they loose their phase
memory before the next tunneling process occurs. The "orthodox theory" ,
initially developed for metallic islands [45, 46], was later generalized to
quantum dots with discrete spectra [58,59,60,61]. In this Section we describe this golden-rule theory for the simplest systems. Quantum dots with
exact many-body wave functions in the few electron limit have been studied in Ref. [62, 63, 64], while coupled quantum dots have been considered
in Ref. [65, 66]. The effect of time-dependent fields have been described in
Refs [67, 68] for metallic junctions and in Ref. [69] for the Coulomb blockade model, and in Refs. [70, 71, 72] for the metallic case in the presence of
a heat bath. Various extensions, incl. experimental results and further references, will be reviewed in the Chapter of Kouwenhoven et at. later in this
volume. A powerful theory which allows a consistent treatment of higher
order tunneling processes and further generalizations will be presented in
the Chapter by Schoeller.
The electron tunneling through a quantum dot is described by the

Hamiltonian H = H res + HD + Ht , where
H res
L: [L:
<"al."a'," + eV,n'l
k,u
(36)
r=L,R
HD
I: Esls >< sl,

I: ~ Tk1al,ural,uD + h.c.,
(37)
Ht
(38)
r=L,R k,l,u
describe the reservoirs, the dot, and the tunneling, respectively.

The reservoirs are assumed to have noninteracting single-particle states
labeled by the reservoir index r, wave vector k and spin (j. The voltage of
reservoir r is denoted by Vr, and nr denotes the total particle number.
The eigenstates of the isolated dot are denoted by Is > with energy Es.
For the Coulomb blockade model, the states Is > of the dot are specified
by the set of all occupation numbers for the single particle states: Is >=
35
l{nl,aD}
>.
In this case, the dot energy is given by

Es
= I>lDnl,aD + Ec(n -
nG)2,
(39)
I,a
n is the particle number operator of the dot. The charging energy

= e2/2C and the gate charge QG = -enG = Li=L,R,G Ci Vi coincide
where
Ee
with the expressions (13) and (15) introduced in the metallic case. The
general notation in terms of the many-body wave functions Is > is introduced here to include cases where the states of the dot cannot be described
by single particle states, see e.g. Refs. [73, 62, 63, 74].
The tunneling part describes charge transfer processes between the
reservoirs and the dot. The tunneling matrix elements are conveniently
combined in the spectral function
rr,II'(E) =
2:
LTk;Tk!,O(E - Ekr) ,
(40)
which depends on the energy and the single-particle. states [, [' involved.
A typical value of the spectral function defines an energy scale r which
characterizes the intrinsic broadening of the single-particle states of the dot
by quantum fluctuations from tunneling. If r <t:: kBT, thermal fluctuations
dominate over quantum fluctuations, and we can use golden-rule theory.
The reservoirs are treated as large systems in equilibrium described by
the grand canonical density matrix p~~s. The tunneling rate for a transition
of the dot from state Is' > to Is > when P = 1 charges have been added
from reservoir r to the dot is given by the golden-rule expression
p
"ir,ss' =
27r
p~~(x')1
< xslHtlX's' > 12o(Es-Es,+Ex-Ex'-PJLr).
xx'
nr(x')=nr(x)+p
(41)
Here X denote the states of the reservoirs with particle numbers nr(x).
The energy conserving o-function includes the change JLr = eVr of the electrostatic energy, which is regarded here as the effective electrochemical
potential of reservoir r. The change of the electrostatic energy of the dot is
included in Es - Es'.
Inserting the tunneling Hamiltonian H t from Eq. (38) into the goldenrule rate (41) yields
"i~ss' =
L rr,II'(Es-Es'-JLr) < slalDls' >< s'laz'Dis > g(Es-Es') '
(42)
II'
where f1(w) = f(w - JLr) is the Fermi distribution of reservoir r, while

fr- = 1 - fr+
36
The rates can be used as an input for a master equation. Consequently,

the stationary dc-probability distribution Ps for the dot and the stationary
dc-tunneling current in reservoir r can be calculated from
I}YssIP(s') -ls'sP(s)] ,
Sl
(43)
e I}Y:ssIP(s') -1~s'sP(s)] ,
(44)
SSI
with Iss'
property
'"
L.Jrp
r,ssl' Current conservation "'r

L.J Ir = 0 follows from the
"V P
I
(45)
r
where ns denotes the number of particles in the dot in state s.

The rates satisfy the detailed balance relation I;s'sll:ss'
,
, = exp[,6(EsES' - J-lr)]' As a consequence, the equilibrium solution of the master equation
(43) is the grand canonical distribution
(46)
which applies when all electrochemical potentials are the same J-lr = J-l. In
this case the dc-current (44) is zero.
Using detailed balance we can write the tunneling rates as
(47)
where
~
+
Ir,ss'
-- Ir,ss'
+ Ir,s's
-
(48)
is the sum of tunneling 'in' and tunneling 'out' rates. As a consequence the
current (44) can be expressed as
Ir = e
L I;'ss' [J:(Es SSI
ES')P(S') - fr-(Es - ES')P(S)]
(49)
A current can flow through the structure if both the tunneling 'in' and
tunneling 'out' rate are non-zero. At low temperatures we consider the
transition between two dot states Sn H sn+1, with nand n + 1 particles
in the dot, respectively. For tunneling 'in' from reservoir r we need tlE =
ESn+l - Esn < J-ln and a similar relation for tunneling 'out' to reservoir r'.
Both conditions are satisfied simultaneously if the excitation energy tlE lies
in the window of the effective potentials of the reservoirs, J-lrl < tlE < J-lr.
The energy- and state-dependence of the spectral function r r,lI' (w) accounts for the energy-dependence of the density of states in the reservoirs,
37
and mesoscopic fluctuations of the conductance peak heights. Here we concentrate on the simplest case and set
(50)
In this case we find from (42) that the sum of the tunneling 'in' and 'out'
rates defined in (48), differs from this scale
I;'ss' = fr
L 1 < slaiDls' > 12
(51)
only by a constant factor, since the Fermi functions have canceled out.
Inserting this relation in (49), and using current conservation Lr Ir = 0 to
eliminate the term which is independent of the Fermi functions, we find
Ir =
r'ss'
fr;r'
< sla!Dls' > 12[P(s) + P(s')]
x [Jr(Es - E s') - fr,(E s - E s')] ,
(52)
with f = Lr f r' This formula is frequently used in the literature. In linear

response, it reduces to a form first derived by Beenakker [59]. In this case
we set /1r = /1 + e8Vn and find
Ir
= L Grr ,(8Vr - 8Vr,) ,
(53)
r'
where the conductance is
G rr, = _e 2
ss'
rr;r,
< slaiDls' > 12[p(s)eq + p(s'rq]!,(Es - Es' -
/1) .
(54)
In the general case, many excitation energies Es - Es' can lie between
/1r and /1r' and are relevant for transport. However, only those transitions
s' -1 s will occur for which the initial probability P( s') is not too small.
For temperatures and bias voltages smaller than the level spacing 8 and the
charging energy Ech, only the ground states s~ of the dot will have nonzero
occupation probability. This means that only the energies
(55)
are relevant. The transition from s~ to an excited state S~tl does not occur
since, upon increasing the gate voltage, the transition sn -1 s~+1 occurs
sooner and afterwards the dot is already in the n + 1-particle ground state.
Thus, we obtain the same physical picture as shown in Fig. 19 with the
38
0.10
0.08
Cl
0.06
0.04
0.02
0.OQ1.0
0.0
\ J.
1.0
2.0
1JI(2Ecl
/.
3.0
4.0
Linear conductance versus I-' for two doubly degenerate levels with (1 = 0,
Ec = 75r, and rL = rR = r /2. The distance between the second and
third main resonance is larger due to the finite level spacing. All resonances involving
excited states are hidden.
Figure 23.
(2
= 25r, T = 5r,
only difference that the distance ~ between adjacent excitation energies is

no longer a constant.
This behavior is reflected in the formula (54) for the conductance matrix
in linear response. Due to the derivative of the Fermi function, the conductance will be maximal when f.L coincides with one of the excitation energies
within a range set by the temperature. The energy difference Es - ESI is
varied experimentally by the gate voltage. Thus, t'he conductance shows
a series of resonances with varying distance between the peaks and a line
shape which is approximately given by the derivative of the Fermi distribution function. Between the resonances, transport is 'not possible and the
system is in the Coulomb blockade regime. As an example, we show the
Coulomb oscillations in Fig. 23 for the Coulomb blockade model with two
spin-degenerate levels with energies E1 < E2. According to (39) the excitation energies, describing the energy changes of the dot when a particle is
added in level 1 to a state with n particles, are given by
~nl
= /D + 2nEc .
(56)
Here /D = E/D + Ec(l- nG) describes the effective level position, which is
tuned by the gate voltage. As a consequence, we observe four resonances
corresponding to the excitation energies ~Ol = 1, ~11 = 1 + 2Ec, ~22 =
E2 + 4Ec, and ~32 = E2 + 6Ec. As explained above, all other excitation
energies ~02 = 2, ~12 = 2 + 2Ec, ~21 = 1 + 4Ec, and ~31 = 1 + 6Ec
are hidden because they involve excited states.
39
6.0
5.0
~ 4.0
NQ)
b. 3.0
2.0
1.0
0.0 L........-....-L~_~~~~_~_-.J
0.0
2.0
4.0
6.0
8.0
10.0
eV/(2Ecl
Figure 24.
The dc current in nonlinear response versus eV = e(VL - VR), with
VL = - VR = V /2 and CG VG / e = 1 fixed, for two doubly degenerate levels with 1 = 0,
2 = SOr. The other parameters are T = sr, Be = 7Sr, and rL = rR = r /2. All
one-particle excitations of the dot are visible.
At finite bias voltage all excitations are in principle visible since the
excited states acquire a finite occupation probability. This holds at least
in the absence of certain selection rules arising from the matrix element
< slarDls' > in (52). For a constant density of states of the leads the I-Vcharacteristic shows steps each time a new excitation becomes relevant. This
result is shown in Fig. 24 for the same example as before. Equivalently, the
differential conductance dI/ dV shows peaks as function of the bias voltage.
As can be seen, all eight excitation energies mentioned before are visible.
The effects of strong correlations on the dot are not only reflected in
the increase of the distance between adjacent resonances but also in the
line shape of an individual peak as a function of the applied gate voltage.
To show this explicitly, we consider the Coulomb blockade model for a
single spin-degenerate state with energy f on the dot. For Ec ~ T we can
restrict ourselves to the transition between an empty and a singly occu pied
dot, n = 0, 1. According to relation (56), the excitation energy for this
transition is given by ~ = E. From the master equation and (52) we find
2e
fr;r'
r'
2frfr'
1 + Lrll
1
- 2e - f - 1 + f (~
=-f frll (~ )
'(
[Jr(~) - fr'(~)],
_ I-l) f ~ -
)
I-l .
(57)
(58)
The current contains an asymmetry factor 1/(1 + Lr fr(~)) which is

absent either for a nondegenerate level or for the noninteracting case Ec =
40
o.
This factor arises from correlations since double occupancy of the dot
is forbidden. This gives rise to particle-hole asymmetry and, consequently,
to an asymmetric line shape of the differential conductance as a function
of ~ as shown in Fig. 25 for finite bias voltage. The maximal value of the
conductance in linear response is given by
cmax _ 4 e2 frfr' ~
rr' - 7r h f 3T
(59)
For a nondegenerate level or for the noninteracting case with one degenerate level, the factor 2/3 has to be replaced by 1/2 or 1, respectively.
This can be easily understood. At the maximum of the conductance several states of the dot have the same probability. For large charging energy
the doubly occupied state can be excluded, and we have two degenerate
excitations which can be used for transport, and three possible states of
the dot (the empty dot and two degenerate states with one electron). Each
excitation contributes equally to the current but has to be multiplied with
the probability 1/3 of the initial state. This explains the factor 2/3. For
a nondegenerate level we have only one excitation and two states, resulting in a factor 1/2. For a noninteracting model with one degenerate level
we have four excitations (two for each transition n = 0 -t n = 1 and
n = 1 -t n = 2) and four possible states, giving a factor 1. The reduction of the current by Coulomb repulsion is obvious, since certain processes
are blocked. In contrast to the noninteracting case, we have seen that the
presence of degenerate states does not give rise to a pure multiplicative
factor of the degeneracy. The reason is that Coulomb interaction induces a
correlation between the levels. When one level is occupied, the other is not
allowed to be occupied due to the strong on-site Coulomb repulsion.
Selection rules occur due to the matrix element I < slarDls' > 12 in (52).
Spin conservation allows only transitions where the total spin of the states
sand s' differs by 1/2. For the discussion of spin blockade effects and
related negative differential conductances we refer to Refs. [62,63, 74].
Conclusions
Many further fundamental concepts of mesoscopic electron transport

could not be included in this Introduction, but will be covered in following Chapters. Interference and localization effects and extensions will be
discussed in the Chapters of Stern and of Imry. The Chapter by Kouwenhoven, Markus, McEuen, Tarucha, Westervelt, and Wingreen will cover the
wide field of electron transport through quantum dots. Eaves includes in
his Chapter concepts related to chaos. De Jong and Beenakker review noise
properties of electron transport. Also Biittiker and Christen's Chapter deals
41
0.4
:E'
0.3
CJ 0.2
0.,)
0.0 b=::::==-~~""::::::====-----_----':~d
-15.0 -10.0 -5.0
0.0
5.0
10.0 15.0
IJ.
Figure 25. The differential conductance as a function of ~ = for a two-fold degenerate level with large charging ener~y ECR so that double occupancy can be neglected.
T = O.25f, PL = -/-IR = 15f, and f = f = f /2.
with extensions to time-dependent phenomena. A systematic discussion of

tunneling beyond perturbation theory is presented in Schoeller's Chapter,
while Fisher and Glazman discuss transport in lD interacting systems.
Superconductivity adds further degrees of freedom to mesoscopic electron transport. The properties of normal metal-superconductor heterostructures are described in the Chapter of the Saclay group, while Fazio and
Schon describe single-charge tunneling in superconducting junction systems
and further review the theory of quantum transport in NS heterostructures.
The very existence of superconductivity in ultrasmall particles is investigated in the article of Ralph et al., while van Wees and Takayanagi address
transport through semiconductor - superconductor sy.stems.
The field of scanning probe microscopy has advanced substantially in
recent years. It is reviewed by Sohn et al. The transport through quantum
point contacts still reveals new results as described in two Chapters by van
Ruitenbeek and by Garcia et al.
We included in this Book also two peripheral Chapters. Yamamoto describes concepts of quantum optics to solid state physicists, while DiVincenzo reviews the novel field of quantum computing.
References
1.
2.
Molecular Beam Epitaxy: Fundamentals and Current Status, M. A. Herman and

H. Sitter, Springer-Verlag, NY (1985); The Technology and Physics of Molecular
Beam Epitaxy, ed. E. H. C. Parker, Plenum Press, NY (1985).
H. van Houten, B. J. van Wees, M. G. J. Heijman, and J. P. Andre, Appl. Phys. Lett.
49, 1781 (1986).
42
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
T. J. Thornton, M. Pepper, H. Ahmed, D. Andrews, and G. J. Davies, Phys. Rev.

Lett. 56, 1198 (1986).
H. Z. Zheng, H. P. Wei, D. C. Tsui, and G. Weimann, Phys. Rev. B 34, 5635 (1986).
S. E. Laux, D. J. Frank, and F. Stern, Surf. Sci. 196, 101 (1988).
H. Ehrenreich and D. Turnbull, eds. , Solid State Physics (Academic Press, New
York, 1991).
Yu. V. Sharvin, Zh. Eksp. Teor. Fiz. 48, 984 (1965) [Sov. Phys. JETP 21, 655
(1965)].
B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. G. Williamson, L. P. Kouwenhoven, D. van der Marel, and C. T. Foxon, Phys. Rev. Lett. 60, 848 (1988).
B. J. van Wees, 1. P. Kouwenhoven, E. M. M. Willems, C. J. P. M. Harmans,
J. E. Mooij, H. van Houten, J. G. Williamson, and C. T. Foxon, Phys. Rev. B 43,
12431 (1991).
1. I. Glazman, G. B. Lesovik, D. E. Khmel'nitskii, and R. I. Shekter, JETP Lett
48,238 (1988); 1. I. Glazman, and A. V. Khaetskii, J. Phys. : Condens. Matter 1,
5005 (1989).
K. F. Berggren, T. J. Thornton, D. J. Newson, and M. Pepper, Phys. Rev. Lett.
57, 1769 (1986).
R. Landauer, IBM J. Res. Dev. 1, 223 (1957); Phys .. Lett. 85A, 91 (1981);
J. Phys. Condens. Matter 1,8099 (1989).
A. Szafer and A. D. Stone, Phys. Rev. Lett. 62,300 (1989); E. G. Haanappel and
D. van der Marel, Phys. Rev. B 39, 5484 (1989); G. Kirczenow, Phys. Rev. B 39,
10452 (1989).
Several authors have obtained transmission resonances in the calculated conductance of narrow constrictions, see Ref. [13].
J. Nixon and J. Davies, Phys. Rev. B 41, 7929 (1990).
J. G. Williamson, C. E. Timmering, C. J. P. M. Harmans, J. J. Harris, and
C. T. Foxon, Phys. Rev. B 42, 7675 (1990).
B. I. Halperin, Phys. Rev. B 25, 2185 (1982).
B. J. van Wees, 1. P. Kouwenhoven, H. van Houten, C. W. J. Beenakker, J. E. Mooij,
C. T. Foxon, and J. J. Harris, Phys. Rev. B 38,3625 (1988).
1. P. Kouwenhoven, B. J. van Wees, C. J. P. M. Harmans, J. G. Williamson, H. van
Houten, C. W. J. Beenakker, C. T. Foxon, and J. J. Harris, Phys. Rev. B 39, 8040
(1989).
H. van Houten, B. J. van Wees, J. E. Mooij, C. W. J. Beenakker, J. G. Williamson,
and C. T. Foxon, Europhys. Lett. 5,721 (1988); C. W. J. Beenakker, H. van Houten,
and B. J. van Wees, Europhys. Lett. 7, 359 (1988); H. van Houten et al., Phys. Rev. B
39, 8556 (1989).
M. Biittiker, Phys. Rev. Lett. 57, 1761, (1986).
J. Spector, H. 1. Stormer, K. W. Baldwin, 1. N. Pfeiffer, and K. W. West,
Surf. Sci. 228, 283 (1990).
P. Streda, J. Kucera, and A. H. MacDonald, Phys. Rev. Lett. 59, 1973 (1987).
J. K. Jain and S. A. Kivelson, Phys. Rev. Lett. 60, 1542 (1988).
M. Biittiker, Phys. Rev. B 38, 9375 (1988).
M. Reed, ed. , Semiconductors and Semimetals (Academic Press, New York, 1990).
B. J. van Wees, E. M. M. Willems, C. J. P. M. Harmans, C. W. J. Beenakker, H. van
Houten, J. G. Williamson, C. T. Foxon, and J. J. Harris, Phys. Rev. Lett. 62, 1181
(1989).
G. Miiller, D. Weiss, S. Koch, K. von Klitzing, H. Nickel, W. Schlapp, and R. Losch,
Phys. Rev. B 42, 7633 (1990).
S. Komiyama, H. Hirai, S. Sasa, and S. Hiyamizu, Phys. Rev. B 40, 12566 (1989).
B. J. van Wees, E. M. M. Willems, L. P. Kouwenhoven, C. J. P. M. Harmans,
J. G. Williamson, C. T. Foxon, and J. J. Harris, Phys. Rev. B 39, 8066 (1989).
B. W. Alphenaar, P. L. McEuen, R. G. Wheeler, and R. N. Sacks, Phys. Rev. Lett.
64, 677 (1990).
43
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
P. L. McEuen, A. Szafer, C. A. Richter, B. W. Alphenaar, J. K. Jain, A. D. Stone,

R. G. Wheeler, and R. N. Sacks, Phys. Rev. Lett. 64, 2062 (1990).
For a review see T. Chakraborty and P. PietiUiinen, The fractional quantum Hall
eHect, Springer series in Solid-State Sciences 85 (Springer-Verlag 1988).
For a review see S. C. Zhang, Int. J. of Mod. Phys. B, 6, 25 (1992).
S. L. Sondhi, A. Karlhede, S. A. Kivelson, and E. H. Rezayi, Phys. Rev. B 47, 16419
(1993); and H. A. Fertig, L. Brey, R. Cote, and A. H. MacDonald, Phys. Rev. B 50,
11018 (1994).
S. E. Barrett, G. Dabbagh, L. N. Pfeiffer, K. W. West, and R. Tycko,
Phys. Rev. Lett. 74, 5112 (1995).
L. P. Kouwenhoven, unpublished.
G. Timp, R. Behringer, J. E. Cunningham, R. E. Howard, Phys. Rev. Lett. 63,
2268 (1989).
L. P. Kouwenhoven, B. J. van Wees, N. C. van der Vaart, C. J. P. M. Harmans
C. E. Timmering, and C. T. Foxon, Phys. Rev. Lett 64, 685 (1990).
A. M. Chang and J. E. Cunningham, Phys. Rev. Lett. 69, 2114 (1992).
C. W. J. Beenakker, Phys. Rev. Lett. 64, 216 (1990).
A. H. MacDonald Phys. Rev. Lett. 64, 220 (1990).
A. M. Chang, Solid State Commun. 74, 871 (1990).
P. L. McEuen, E. B. Foxman, J. Kinaret, U. Meirav, M. A. Kastner, N. S. Wingreen,
and S. J. Wind, Phys. Rev B 45, 11419 (1992); D. B. Chklovskii, B. 1. Shklovskii,
and L. 1. Glazman, Phys. Rev. B 46, 4026 (1992).
1. O. Kulik and R. 1. Shekhter, Zh. Eksp. Teor. Fiz. 68, 623 (1975)
[Sov. Phys. JETP 41, 308 (1975)].
D. V. Averin and K. K. Likharev, J. Low Temp. Phys. 62, 345 (1986); and in
Mesoscopic Phenomena in Solids, B. L. Altshuler, P. A. Lee and R. A. Webb, eds. ,
p. 173 (Elsevier, Amsterdam, 1991).
T. A. Fulton and G. J. Dolan, Phys. Rev. Lett. 59,109 (1987).
L. J. Geerligs et al. , Phys. Rev. Lett. 64, 2691 (1990).
.
H. Pothier et al. , Physica B 169, 573 (1991), Europhys. Lett. 17, 249 (1992)
Single Charge Tunneling, NATO ASI Series, Vol. B 294, eds. H. Grabert and
M. H. Devoret, (New York, Plenum Press 1992).
Mesoscopic Superconductivity, Proceedings of the NATO ARW, edited by
F. W. J. Hekking, G. Schon, and D. V. Averin, Physica B 203 (1994).
G. Schon and A. D. Zaikin, Phys. Rep. 198, 237 (1990).
L. 1. Glazman and K. A. Matveev, Sov. Phys. JETP. 71, 1031 (1990); K. A. Matveev,
Sov. Phys. JETP 72, 892 (1991).
J. Konig, H. Schoeller, G. Schon, and R. Fazio, in Quantum Dynamics of Submicron
Structures, NATO ASI Series B 291, edited by H. Cerdeira et al. (Kluwer 1995),
p. 221; J. Konig, H. Schoeller, and G. Schon, Europhys. Letters 31, 31 (1995).
K. Mullen, E. Ben-Jacob, R. C. Jaklevic, and Z. Schuss, Phys. Rev. B 37, 98 (1988).
J. Konig, H. Schoeller, and G. Schon, Phys. Rev. Lett. 78, 4482 (1997).
P. Joyez, V. Bouchiat, D. Esteve, C. Urbina, and M. H. Devoret, preprint.
D. V. Averin and A. N. Korotkov, Zh. Eksp. Teor. Fiz. 97, 1661 (1990) [Sov.
Phys. JETP 70, 937 (1990)]; D. V. Averin, A. N. Korotkov, and K. K. Likharev,
Phys. Rev. B 44, 6199 (1991).
C. W. J. Beenakker, Phys. Rev. B 44,1646 (1991).
Y. Meir, N. S. Wingreen, and P. A. Lee, Phys. Rev. Lett. 66, 3048 (1991).
H. van Houten, C. W. J. Beenakker, and A. A. M. Staring, in Ref. [50].
D. Weinmann, W. Hausler, W. Pfaff, B. Kramer, and U. Weiss, Europhys. Lett. 26,
467 (1994); D. Weinmann et al Phys. Rev. Lett. 74, 984 (1995).
K. Jauregui, W. Hausler, D. Weinmann, and B. Kramer, Phys. Rev. B 53 Rl713
(1996).
D. Pfannkuche and S. E. Ulloa, Phys. Rev. Lett. 74, 1194 (1995).
R. H. Blick, R. J. Haug, J. Weis, D. Pfannkuche, K. von Klitzing, and K. Eberl,
44
66.
67.
68.
69.
70.
71.
72.
73.
74.
Phys. Rev. B 53, 7899 (1996).

G. Klimeck, G. Chen, and S. Datta, Phys. Rev. B 50, 2316 (1994); Phys. Rev. B
50, 8035 (1994).
L.P. Kouwenhoven, S. Jauhar, K. McCormick, D. Dixon, P.L. McEuen,
Yu.V. Nazarov, N.C. van der Vaart, and C.T. Foxon, Phys. Rev. B 50, 2019 (1994).
I. A. Devyatov and K. K. Likharev, Physica B 194-196, 1341 (1994).
C. Bruder and H. Schoeller, Phys. Rev. Lett. 72, 1076 (1994).
G. Ingold and Yu. V. Nazarov, in Ref. [50].
A. A. Odintsov, G. Falci, and G. Schon, Phys. Rev. B 44, 13089 (1991).
K. Flensberg, S. M. Girvin, M. Jonson, D. R. Penn, and M. D. Stiles, Phys. Scripta
T 42, 189 (1992).
D. Pfannkuche, V. Gudmundsson, and P. A. Maksym, Phys, Rev. B 47, 2244 (1993).
W. Hausler, Ann. Physik 5,401 (1996).
GEOMETRIC PHASES IN MESOSCOPIC SYSTEMS - FROM

THE AHARONOV-BOHM EFFECT TO BERRY PHASES
ADY STERN
Weizmann Institute of Science
Department of Condensed Matter Physics
Rehovot 76100, Israel
1. Introduction
Contents Here is what this chapter is about, in a few sentences. In quantum mechanics, an electron interacting with a time-independent vector potential exhibits the Aharonov-Bohm effect. This effect manifests itself time
and again in the thermodynamics and transport properties of mesoscopic
systems. It results from the observation that a vector potential induces a
geometric shift in the phase of a quantum mechanical wave packet. We
review this rather well known effect in Section (2).
A vortex moving in a two-dimensional ideal fluid is subject to a classical force, Magnus force, proportional and perpendicular to its velocity. Just
like Lorenz force. Consequently, its quantum mechanical behavior should
be similar to that of an electron interacting with a magnetic field - an
Aharonov-Bohm type behavior. Josephson junctions and Josephson junction arrays seem to be most suitable to a study of such a behavior. This is
the subject of Section (3).
An electron in a magnetic field whose direction varies in space accumulates a geometric (Berry) phase due to the spin's Zeeman interaction with
the magnetic field. There are some similarities between this phase and the
Aharonov-Bohm phase. Consequently, this phase should manifest itself in
electronic properties of mesoscopic systems in non-uniform magnetic fields.
Some of these manifestations are quantum mechanical, and some, surprisingly, are classical. This is the subject of Section (4).
45
46
Tips for the reader. If you are a graduate student - this chapter should
be accessible to you. Familiarity with the Josephson effect is important for
Sections (3.3) and (3.4). If you are an expert in mesoscopic physics, you
are probably familiar with much of the material presented. However, you
might find the discussion in Section (3.3) interesting.
2. The Aharonov-Bohm effect - review

Let us review a few basic facts regarding the interaction of electrons with
electric and magnetic fields.
2.1. CLASSICAL MECHANICS
A classical electron in an electric and a magnetic field experiences a force

given by
e
F = eE+ -v x B
(1)
c
(the notation is obvious). This is conceptually simple - the force is determined by the fields at the position of the electron, and is independent of the
fields in other places. The classical force can be derived from a Lagrangian
or an Hamiltonian. Both describe the interaction of the electron with a
gauge field, composed of a scalar and a vector potential. The Lagrangian
is,
L(x(t), v(t))
e
= -mv 2 + -v
A-
eV(x)
(2)
2.2. QUANTUM MECHANICS
A quantum mechanical electron exhibits the Aharonov-Bohm effect. In its

most striking manifestation, depicted in Fig. (1), the Aharonov-Bohm effect deals with the effect of a solenoid carrying a magnetic flux on the
interference pattern of a quantum mechanical electron in a double-slit
experiment. In its mesoscopic manifestation (to be discussed below), the
Aharonov-Bohm effect deals with the dependence of various properties of
a mesoscopic ring on a flux threading it (Fig. 2.3). Let us first understand
the double slit experiment[l].
2.2.1. The double slit experiment
Fig. (1) describes a double slit interference experiment in which a small,
impenetratable, solenoid is placed between the two slits, and in which one
measures the number of electrons coming from a source, going through the
slits and hitting a screen at a point r.
47
Figure 1.
The Aharonov-Bohm effect. The interference pattern of an electron in a
double slit experiment is shifted by its interaction with the vector potential induced by
the solenoid.
First, let us disregard the solenoid. In a semiclassical description of this

experiment, the wave function of an interfering particle is composed of a
sum of two waves,
(3)
where al(r),a2(r) are the amplitudes corresponding to the particle going
through the first and the second slit, respectively. The squared modulus of
this wave function, which is the measured and observed quantity, is then
composed of two classical terms, lal(r)1 2 and la2(r)1 2 , and two quantum mechanical interference terms, that sum up to 2Reai(r)a2(r) cos (k1 - k 2 ) . r.
The latter terms are a contribution to the probability distribution which
is not associated with a well defined trajectory, but rather with the interference of two distinct ones. These terms depend on the relative phase
between the two interfering waves.
Now, let us consider the way the solenoid affects the relative phase between the waves. The relative phase between two wave packets following trajectories Xl(t),X2(t) is given by the action difference [S(Xl(t)) - S(X2(t))],
divided by n. The action is,
S(x(t)) ==
rtf
iti
dtL(x,:ic)
(4)
In the absence of the solenoid, the relative phase of the two waves is
determined by the paths' lengths. The solenoid shifts the relative phase of
the two waves by the Aharonov-Bohm phase shift, </JAB = 21[10, where </J is
48
:c
the magnetic flux it encloses, and CPo ==

is the flux quantum. To see this,
we observe (see (2)) that the contribution of a time-independent vector
potential term to the action difference is
CPAB =
it
ti
f dt-XlAe
lic
it
ti
f
e
X2 Ae
dt=lic
lic
Xf
Xi
e
dhA-lic
Xf
d12 A (5)
xi
where the integrals dh, dl 2 go along the respective paths. As these integrals
demonstrate, this contribution is geometric, i. e., it depends on the trajectory, but not on its time parameterization. As the name of this chapter
indicates, a geometric phase is an important concept to us.
Using Stokes theorem, this relative phase can be converted to,
CPAB
= -e
lic
da B
= -27f
CPo
da B
(6)
where J da indicates a surface integral of the area bounded by the two

paths, and CPo is the flux quantum.
Since the solenoid is impenetratable, the flux enclosed by the two paths
is the flux enclosed by the solenoid times the the number of times the
two paths wind around it. The phase in this case is not only geometric
but also topological - a continuous change of the paths leaves the phase
unchanged. The general expression for the phase, Eq. (6), was geometric
due to the vector potential being time-independent. In the particular case of
the solenoid the phase was also topological since the solenoid has effectively
made space multiply-connected.
The loop created by the two classical paths in the double slit experiment winds the solenoid once, and therefore the solenoid shifts their relative
phase by 27f1o. Consequently, if one measures the number of electrons hitting a particular point r on the screen as the flux in the solenoid is varied,
one observes a periodic pattern, with a period of CPo.
To conclude this review, it is important to emphasize the strikingly
quantum mechanical nature of the effect: the solenoid is impenetratable,
no electron ever gets into it, and yet, the magnetic flux it encloses does
affect the motion of the electron to the interference screen.
2.3. THE AHARONOV-BOHM EFFECT IN MESOSCOPIC PHYSICS
So this is, on a nut shell, the story of the Aharonov-Bohm effect in its
double-slit realization. What does this have to do with mesoscopic physics?
A few things[2]:
Double-slit type mesoscopic experiments One may try and fabricate a
mesoscopic Young double-slit apparatus, patterned, e.g, in a two- or three-
49
Figure 2. In mesoscopic physics, the Aharonov-Bohm effect is manifested in the flux

sensitivity of transport and thermo-dynamical properties of small rings, such as the conductivity and magnetization.
dimensional semi-conducting or metallic device. Practically, it is simpler to

fabricate devices in the forms of rings, and replace the screen by a single
collector. What is measured in the experiments is then the current flowing
from the electron's source to its collector, as a function of the magnetic
flux threading the ring. The next chapter in this book reviews these experiments.
When working ideally, this kind of experiments should reproduce the
results of experiments done in vacuum. The deviation from ideal behavior, however, is very interesting, since it results from the electron's interaction with its environment, an interaction that is negligible in vacuum
experiments, and important in solid state devices. The main effect of this
interaction, dephasing, is described in Section (2.4).
Linear response theory as a complicated interference experiment Linear
response measurements constitute, in a non-trivial way, a measurement of
a quantum interference pattern. To see this, we recall linear response theory,
in its Kubo formula representation[3][4]. Consider a sample described by
the Hamiltonian Ho, subject to an external weak perturbation given by an
electro-magnetic vector potential A(x, t). In the presence of the perturbation, the expectation value of the current density operator in the direction
C, Ja , has a component proportional to the vector potential in the direction
{3, Ap, and the proportionality factor is the linear response function Kap.
It is a matter of straightforward perturbation expansion to .show that this
component is
50
where 0 denotes an average over the quantum state of the sample, a and
f3 are Cartesian indices, f3 is summed over, and the operators J are interaction picture operators, whose time evolution is determined by Ho.
When Eq. (7) is Fourier transformed to momentum-frequency representation, the second-quantized form of the current operator is used, and a
non-interacting electrons model is employed, the linear response function
becomes (for the case k = 0, for simplicity) [3][4],
e2 n
Ka,8(k = O,w) = --oa,8
mc
+ -me2c
x { nF(w') [Gt:'+w. (p', p )Gt:'. (p, p') -
dw'
-2 LPaP~
1f2
p,p
G~. (p', p )G~. -w (p, p')1
+[nF(w + w') - nF(w')I . Gt:'+w' (p', P )G~. (p, p') }
(8)
where e, m and n are the electrons' charge, mass and density, respectively, c
is the speed of light, np is Fermi's function, the summation is over momentum states and GR,GA are retarded and advanced propagators in energy momentum representation. As seen in Eq. (8), the linear response function
is a sum of two-propagator products. By Feynman's path integral formulation, each propagator can be written as a sum over paths of interfering
waves. The linear response function is then proportional to the square of
such a sum, like the intensity of light on an interference screen, and like the
electron interference patters discussed above. Thus, although Eq. (8) is not
going to be used below, it is important to remember what it reflects: a measurement of a linear response function is a measurement of a complicated
interference pattern.
Unlike the double slit experiment, the square sum over paths appearing
in the expression for the linear response function (8) includes, even in the
semiclassical approximation, a vast number of (usually diffusive) paths to
be summed on. Generally, the relative phase between these paths is random,but there are two important exceptions. The first is the interference
of a path with itself, which gives a classical contribution, just as in the
two slit experiment. The second is the interference of two time reversed
paths. For a time reversal symmetric system, time reversed paths accumulate identical phases, and thus interfere constructively. Two time reversed
paths can interfere only if they start and end approximately at the same
point, and thus their constructive interference enhances the probability for
closed paths, i.e., enhances resistance. This enhancement is commonly referred to as weak localization. When a weak magnetic field is applied to
the system, the Aharonov-Bohm phases accumulated by each pair of time
51
reversed paths destroy the constructive interference, and thus the magnetic
field decreases the resistance. Both weak localization and the resulting negative magneto-resistance are observed experimentally, as reviewed in the
next chapter of this book.
Persistent thermodynamic currents in normal mesoscopic rings [5] So far
we discussed only transport, non-equilibrium, properties. The AharonovBohm effect manifests itself also in the occurrence of a thermodynamic,
equilibrium, persistent electronic current in a normal (non super-conducting)
mesoscopic ring threaded by a magnetic flux. To understand this effect,
think of an hydrogen atom. In the presence of a magnetic field, an hydrogen atom has an orbital magnetic moment, due to a persistent motion of
the electron around the nucleus. The persistence of that motion is surprising from a classical point of view, since a classical electron, under such
circumstances, would radiate energy and crash on the nucleus. Quantum
mechanically, of course, this does not happen, since the electron has no
states to decay to. Even at finite temperatures, when the electron has some
probability of being at excited states, the magnetic moment does not decay
in time. The reason is that each of the excited states the electron spends
time in carries a persistent magnetic moment.
The persistent current in normal rings is of the same nature - a thermodynamical motion which would classically decay in time, but quantum mechanically stays forever. It is, perhaps, more surprising, since an electron in
a mesoscopic ring has additional ways of losing its energy or momentum,
such as impurities, interactions with phonons etc.
The next chapter of this book reviews experimental evidence for the
existence of persistent currents in mesoscopic rings.
2.4. DEPHASING - THE LOSS OF QUANTUM INTERFERENCE
The conductance of micron-sized rings at low temperatures is sensitive to

an Aharonov-Bohm flux, but this sensitivity is lost when the ring's size is
macroscopic, or when the temperature is not low enough. A micron-sized
ring carries a persistent current, but this current is, again, suppressed when
the ring's size is too large or the temperature too high. Put in other words,
the ring shows various Aharonov-Bohm effects only if it is smaller than a
certain size, called the phase breaking (or dephasing) length, and denoted
by l. The phase breaking length depends on temperature, as well as on
other parameters of the ring.
Quantum interference is lost (or the electron phase is "broken") due
to the electron's interaction with other degrees of freedom (the "environment"). Phase breaking is a broad subject[6][7] [9], and we describe here
52
only "the tip of the iceberg". Consider an Aharonov-Bohm interference
experiment on a ring. This experiment starts by a construction of two electron wave packets, l(x) and r(x) (l, r stand for left, right), crossing the ring
along two opposite sides. The interference is examined after each of the two
wave packets traverses half of the ring's circumference. During their motion,
the electron partial waves may interact with the environment, whose set of
coordinates is denoted by 'r/, and whose initial state is denoted by XO('r/).
Therefore, the initial wave function of the system (electron+environment)
is:
(9)
'ljJ(t = 0) = [l(x) + r(x)] XO('r/)
At time TO, when the interference is examined,and after the electron waves
have interacted with the environment, the wave function is, in general,
'ljJ( TO) = l(x, TO) Xl('r/)
+ r(x, TO) Xr('r/)
(10)
and the interference term is,

(11)
The effect of the interaction is then to multiply the interference term by
Jd'r/xi ('r/)xr ('r/).
There are two ways to understand this expression. First, it is the scalar
product ofthe two environment's states coupled to the two partial waves. At
t = 0 these two states are identical. During the time of the experiment, each
partial wave has its own interaction with the environment, and therefore
the two states become different. Since the environment is not observed in
the interference experiment, its coordinate is integrated upon, i.e. the scalar
product of the two states is taken. When the two states do not overlap at
all, the final state of the environment identifies the path the electron took,
and quantum interference between the two paths is lost. Thus, l</> is the
length after which the two waves have shifted the environment into two
orthogonal states.
The second way to understand the expression (11) regards it from the
point of view of how the environment affects the partial waves, rather
than how the waves affect the environment. When a static potential V(x)
is exerted on one of the partial waves, this wave accumulates a phase,
=
J V(x(t))dt, and the interference term is multiplied by ei </>. "A static
potential" here is a potential which is a function of the particle's degrees of
freedom only. For a given particle's path, the value of a static potential is
well defined. When V is not static, but created by environment degree(s)
of freedom, its value is not well defined any more. The uncertainty in its
value results from the quantum uncertainty in the state of the environment.
-k
53
Therefore, is not definite, too. In fact, becomes a statistical variable, a

quantum operator, described by a distribution function P().
The effect of the environment on the interference is then to multiply the
interference term by the average value of ei </>, i.e.,
(12)
Since ei </> is periodic in , (e i </ tends to zero when the width of the phase
distribution, the phase uncertainty, is of the order of 21r, one period. The
phase breaking length, then, is the distance over which the uncertainty in
the phase accumulated by the two interfering waves becomes of order of
21r.
It can be shown, and is shown in great detail in Refs. [9],[8],[7], that
the two descriptions given above are equivalent, i.e., that when the two
environment states coupled to the two partial waves are orthogonal, the
relative phase between the two waves is completely uncertain.
A natural, and subtle, question to raise here is the relation between the
phase breaking length l</> (or time 7</ and other scattering times with which
we are familiar, such as the electron-electron scattering time, the electronphonon scattering time, the momentum relaxation time etc. A detailed
discussion of this question is outside the scope of this chapter. It is partially
given in [9]. Here, we just mention two important points. First, elastic
collisions with heavy impurities relax the electron's momentum but do not
break its phase, since the quantum state of the impurity is not changed. And
second, while electron-electron and electron-phonon collisions do break the
phase of the interfering electron, identifying 7</> with the collision times
corresponding to these processes is frequently not the right thing to do.
2.5. CONCLUSIONS FROM THE AHARONOV-BOHM EFFECT
Here is the moral we should take from the above review of the AharonovBohm effect, in order to proceed to the next sections:
1. A particle whose classical equation of motion includes a "Lorentztype" force, i.e., a force proportional and perpendicular to the particle's
velocity, is a particle that interacts with a vector potential. Quantum
mechanically, the interaction with the vector potential leads to the
accumulation of geometric phases which can be seen in double-slit
type experiments.
In the case discussed above, the particles were electrons and the Lorentztype force was simply Lorentz force. In the next section, the particles
are vortices and the Lorentz-type force is Magnus force.
2. The converse is also true: if a particle's wave function accumulates a
geometric phase, it is due to the particle's interaction with a vector
54
potential. Then, its classical equation of motion includes a "Lorentztype" force.
In Section (4) we study the quantum and classical mechanics of a spin
in a magnetic field whose direction varies in space. Our starting point
in this case will be quantum-mechanical, and it will lead us to quite
surprising classical conclusions.
3. Vortices and geometric phases

3.1. A GENERAL DISCUSSION
The dynamics of vortices in two dimensional (x - y plane, say) fluids carries

some resemblance to that of electrons in a magnetic field[lO]. This resemblance, well known in classical hydrodynamics, is most clearly demonstrated
by Magnus force. A vortex that moves relative to the fluid it is in experiences a force which is proportional and perpendicular to the velocity, like
Lorentz force.
In this section we carry this resemblance from classical physics to quantum phenomena, focusing on interference effects in which the interfering
degree of freedom is that of a vortex, and, most particularly, on geometric
phases accumulated by the wave function of the vortex.
We start by recalling what a vortex is, and what Magnus force is.
What is a vortex?
a 2D fluid is this:
A classical theorist recipe of constructing a vortex in
1. Pick up a point to be the center of the vortex. Call it Ro.

2. Give each and everyone of the fluid particles an angular momentum
li relative to Ro. This gives rise to a velocity field v(x) = ~ z~~R~g)
(where m is the mass of a fluid particle) that decays as l/lx - Rol,
and to a hydro-dynamical vorticity f v . dl = 2~1.
3. Deplete the density near the vortex center to avoid a short distance
divergence of the kinetic energy. This point is not crucial to our argument, and we will not dwell on it here.
The outcome of this recipe is an excitation in which the fluid moves

around the center of the vortex.
What happens when the vortex' center Ro moves? [11] Then, the velocity
field at any given point x is the sum of the vortex velocity relative to the
laboratory frame, and the fluid velocity relative to the vortex center, i.e.,
~ zl:~R~g) + Ro. Suppose that the vortex moves in the x-direction. The
velocity field at points along the y-axis is directed in the x-direction, and is
55
different on the vortex' two sides. Consequently, there's a pressure gradient
along the y-axis, and, as a consequence of Bernoulli's theorem, there's a
force acting on the vortex, Magnus force. Magnus force is both proportional
and perpendicular to the vortex velocity, Ro. It is given by,
FMagnus
= 27rlnRo
xz
(13)
where n is the fluid density. This is a "Lorentz-type" force!

Analogies to Lorentz force The role played by the product ~B in Lorentz
force is played, in Magnus force, by the product 27rln. Given that similarity,
we will refer to 27rln as the "magnetic field". While the fluid density plays
a role analogous to that of a magnetic field, a fluid current plays a role
analogous to that of an electric field. To see that, note that an electron
subjected to a magnetic field in the laboratory frame, sees this field as
an electric field in its rest frame. This electric field is given by ~Ro x z.
Similarly, in a frame of reference in which the vortex center is at rest, the
force it is subjected to seems to arise from the motion of the fluid. Since
the fluid current density is J = nRo, in the vortex rest frame the Magnus
force is FMagnus = 27rlJ x Z, and J x z plays a role analogous to that of an
electric field [12][15J.
What happens quantum mechanically? Our discussion in section (2.5) has
prepared us to the present case. Like Lorentz force, Magnus force can be
formulated as resulting from the vortex' interaction with a vector potential, and this interaction must give rise to a quantum mechanical geometric
phase. Based on the mapping of 27rln onto ~B, we expect that the geometric
phase is 27rk Ida zn, i.e., 27r times the number of fluid particles encircled
by the vortex trajectory, times the ratio
This expression is even further
simplified by another quantum ingredient, the quantization of angular momentum, or, alternatively, the quantization of hydro-dynamical vorticity.
These require that 1 be quantized in units of Ti [16J. In fact, in most known
examples, it is precisely Ti, leading to a geometric, "Aharonov-Bohm-type"
phase, given by
k.
27rx the number of fluid particles encircled by the vortex path
(14)
The analog of the magnetic flux is then the number of fluid particles. The
analog of a flux quantum is a single fluid particle, a quantity independent
of Til [17]. This geometric phase was first discovered by Arovas, Schrieffer
and Wilczek [18], in the context of the Quantum Hall Effect, using a Berry
phase calculation. It was studied in depth by Ao, Thouless and Niu[12]
56
in the context of super-conductors, and by van Wees, in the context of

Josephson junctions [13].
There are a few important comments to be made regarding this quantummechanical geometric phase.
Firstly, let us be more specific as to who precisely accumulates this
geometric phase. Unlike the electron, which is a "fundamental" particle,
the vortex is a many-body" composite" excitation. The assumption made
in the classical derivation of Magnus force, reviewed above, is that the
motion of the vortex is decoupled from other degrees of freedom in the
fluid (e.g., sound waves). Under that assumption, the quantum mechanics
of the vortex can be described by a wave function whose argument is Ro,
and it is this wave function that accumulates the geometric phase discussed
above. Generally, the vortex is not perfectly decoupled from other degrees
of freedom, and their coupling constitutes a mechanism for dephasing of
the vortex wave function. This coupling is of particular importance for
vortices in super-conductors, where impurities and quasi-particles play an
important role [14].
Secondly, in some sense, the vortex' wave function describes a macroscopic excitation, made of a huge number of fluid particles. Thus, interference effects involving the vortices are, in some sense, macroscopic interference effects. However, one should exercise caution when thinking in these
terms, as our discussion in Section (3.3).
Thirdly, we note that the analogy between vortex dynamics in a fluid
and electron dynamics in a magnetic field does not depend on the fluid
being charged, like super-conductors, or neutral, like Helium in its superfluid state.
Fourthly, we should point out that our discussion has so far covered
only one aspect of the dynamics of vortices - their interaction with a vector potential. We have not discussed the vortex mass and kinetic energy,
interaction between vortices etc. To a large extent, those aspects depend
on the vortex and fluid studied, and we will briefly discuss them below for
what might be the simplest case, that of vortices in Josephson junctions.
To summarize - general conclusions regarding vortices and geometric phases

Our discussion of vortices has so far led us to conclude that under conditions in which the motion of a vortex is decoupled from other degrees of
freedom in the fluid, its wave function accumulates a geometric phase due
to its interaction with a vector potential. The phase accumulated by a vortex traversing a closed trajectory is 21f times the number of fluid particles
enclosed by the trajectory.
57
3.2. PHYSICAL REALIZATIONS
We now turn to look for a physical system in which vortices can behave
quantum mechanically and the geometric phases they accumulate can be
observed. The system we look for should have features similar to those of the
mesoscopic rings in which the Aharonov-Bohm effect is observed. First, we
should be able to confine the vortices to move only in multiply-connected
parts of space, and confine the "magnetic field" they interact with to the
region in space they are excluded from. Second, the vortex classical motion
should be only weakly damped, indicating that the interaction of the vortex
with its environment would not destroy its quantum character. And third,
we would like to be able to control the strength of the "magnetic field"
the vortex interacts with, i.e., to control the number of the fluid particles
encircled by the vortex' trajectory.
These requirements are best complied with by Josephson junctions.
First, vortices in Josephson junctions are confined to have their core in
the insulating layers separating the super-conductors. On the other hand,
the "magnetic field" that induces the geometric phase in the vortex' wave
function, namely the Cooper-pairs, is all confined to the super-conducting
island. In a Josephson junction, then, the vortex and the "magnetic field"
are separate in space.
Second, the damping of vortex motion in Josephson junctions is weaker
than that of vortex motion in super-conductors. A vortex in a superconductor has a core full of normal electrons. Its motion creates an electromotive force that acts on these electrons, and leads to dissipation of energy.
In contrast, the core of Josephson vortices is always in the insulating regions. When it moves and creates an electro-motive force, this force almost
does not lead to normal currents.
And third, the controllability of the" magnetic field" they interact with,
i.e., the number of fluid particles encircled by the vortex. For this number
to be controllable, it is essential that the vortex encircles an "island" of the
fluid, an island in which the fluid particles are well localized. This is the
structure we discuss below, that of a long, circular, Josephson junction, i.e.,
a super-conducting island surrounded by an insulating layer, surrounded
by another super-conductor.
More generally, similar conditions can be created in a Josephson junction array, an array of weakly coupled super-conducting islands. In such
an array, the vortex moves only between the islands, while the "magnetic
field" , the Cooper-pairs, is confined to within the islands. The vortex can
then be viewed as a particle on a tight-binding lattice in a "magnetic field".
In the next section we study in details one effect associated with the
geometric phase accumulated by a vortex, the persistent motion of a vortex
58
in a long, circular, Josephson junction. This effect is a simplified version
of a similar effect in super-conductors, discussed first by Delin and Orlando[15], and a similar effect in Josephson junction arrays, discussed first
by van Wees[19]. It is analogous to the persistent current in AharonovBohm rings. In the literature, this effect is sometimes referred to as the
"Aharonov-Casher" effect, although this name might be better reserved
for the geometric phase accumulated by a magnetic moment due to the
spin-orbit interaction ([20]).
3.3. PERSISTENT VOLTAGE IN A LONG, CIRCULAR, JOSEPHSON
JUNCTION
The physical system considered [21] Consider a circular Josephson junction (see Fig. 3) made of two thick, super-conducting, concentric cylinders
separated by a thin insulating layer. The strength of the Josephson coupling between the two super-conductors can be characterized either by 1],
the critical Josephson current density, or by A J , the Josephson penetration
depth (1] == 4~ 4~~2L~' with)..L the London penetration depth). The height
of the junction, hz, is much smaller than the Josephson screening length,
A J , but much larger than )..L. Thus, the ring is effectively one-dimensional,
i.e., the phase difference between the two super-conductors is approximated
to be independent of z. We denote the polar coordinate (the only relevant
dimension) by x. The circumference of the ring is denoted by L and the
width of the insulating layer by d.
Super-conductor
Super
Conductor
+Q
-Q
Figure 3. The Josephson junction considered. Two thick concentric superconducting

cylinders separated by a thin insulator.
59
In the first step we create a vortex in the

junction. In the second step we adiabatically charge the junction. This is
done by slowly switching on a small current to flow across the junction,
and then slowly switching it off, such that the total charge flowing across
the junction is finite and controlled. We refer to this charge as "the bias
charge". The current flowing across the junction accelerates the vortex,
and the acceleration is proportional to the current. Therefore, assuming
the absence of any dissipation or spatial disorder, the velocity of the vortex
is proportional to the bias charge. In the third step, we analyze the way
the energy of the system and the voltage across the junction depend on the
bias charge. Then, in the fourth step we use the mapping of our problem
onto that of an electron on an Aharonov-Bohm ring to briefly address the
effects of disorder and dissipation.
Technically, the main step we have to carry out is the derivation of
an effective description of the junction that describes it in terms of the
vortex degrees of freedom. Before carrying out this derivation, we remind
the reader of the description of a long Josephson junction in terms of a
phase field and a Sine-Gordon Hamiltonian.
The scenario to be analyzed
In
an Hamiltonian description the junction is characterized by the field O(x),
describing the gauge-invariant phase difference between the two superconductors (O(x) == <P2(X) - <P1(X) - ~~ J12 A(x, y) dy, where the <P's are
the phases of the super-conducting order parameter on the two sides and A
is the electro-magnetic vector potential), and its conjugate field, nhzn(x).
The field n(x) describes the two-dimensional number density (number per
unit area) of Cooper-pairs on the two sides of the junction. The magnetic
field in the junction is directed in the z-direction (neglecting edge effects),
and is given by
Ox' In the presence of a uniform bias charge density,
denoted by 2eo-, the electric field is 81fe(n(x) - a). The Hamiltonian of
the junction is then given by (unless otherwise stated, all integrals are one
dimensional, taken from x = 0 to x = L),
The Hamiltonian of the junction and the classical equation of motion
4tt
=- /
lldx = he /
{rph; (n - a)' + ;, [~O~ + ;} (1- COSO)]}
dx
(15)
where the Sine-Gordon coupling constant {J2 == 161f~: ,j~:Ld and c == cJ 2t
is the speed of light in the junction. For the clarity of the expressions, the xdependence of 0, a and n is suppressed. The first term in the Hamiltonian
is the electrostatic, capacitive, energy, the second term is the magnetic,
inductive, energy, and the third term is the Josephson energy.
60
The classical equation of motion derived from the Hamiltonian (15) is
the Sine-Gordon equation,
(16)
This equation of motion is nothing but a continuity equation for the electric current. The external bias current density, 2ea, is split into three parts.
h sinO is the Josephson current flowing across the junction. 4~
is the
displacement current charging the junction. 4~ 4~fL Oxx is the difference between the inductance current (flowing along the junction) to the right and
to the left of the point x. Since the junction is circular, the phase difference
satisfies O(x) = O(x + L) + 21Tm, with m being an integer. A single vortex enclosed in the junction is described by the solution to Eq. (16), with
m=l.
In the absence of a bias current, i.e., when a = 0, the solution of the classical equation of motion, (16) , describing a vortex trapped in the junction
;!
moving in a velocity
sin
v c == cJ 2t is, [22]
}
{"2I
[O(x, t) 1T]
= sn
(X - kAJ
Xo -
vt)
(17)
where sn(z) is the Jacobi elliptic function and k is implicitly given by the
complete elliptic integral of the first kind (see, for example, [24]) L/ AJ =
2kK(k). The parameters Xo and v describe the initial position and velocity
of the vortex, respectively. Below, we abbreviate X(t) = Xo + vt. In the
limit L AJ (long junction), the phase changes from 0 to 21T along a
narrow strip of length AJ , and is approximately constant everywhere else.
The magnetic and electric fields are non-zero only within that strip. In the
limit of L AJ (short junction), the phase changes almost linearly along
the junction, and the magnetic and electric fields are spread uniformly along
the junction.
From infinitely many to one degree of freedom - how to describe the system
in terms of the vortex alone [23] We now turn to reduce our description of
the system from a field description, including an infinite number of degrees
of freedom, to a description in terms of one pair of canonically conjugate
variables, the vortex' position and momentum. The need to perform such a
reduction characterizes most problems involving vortices, and the method
we present here is quite easily generalizable.
61
We firstly write the Lagrangian of the junction in terms of the field O( x)

and its time derivative:
(18)
Secondly, we substitute the classical solution (17) in the Lagrangian, and
perform the integral over x. This step transforms the Lagrangian from being
a functional of the field O(x) to being a functional of the vortex position
X(t) and velocity X(t). The resulting Lagrangian is,
(19)
where Q = 2ea Lhz is the bias charge, M r c2 is the sum of the magnetic
energy and the Josephson energy needed to create a static vortex in the
junction, written in the suggestive form of a rest mass energy. Md is the
dynamical mass involved in the motion of the vortex. For a long junction
(fJ -+ 00),
2h 2 h z
Mr = Md = 7rAJd(2e)2
(20)
For a short junction (fJ -+ 0),
(21)
The rest mass term is of no interest to us, since (for the junction we
specified) it is independent of the vortex coordinate. From now on, we omit
it. The other two terms are of interest to us. The first describes a kinetic
energy of the vortex. The second describes the interaction of the vortex with
a vector potential given by ~.
Finally, we change the way we regard X(t). So far, we thought of it
as representing a pair of parameters in the classical solution (17). From
now on, we regard it as our new dynamical variable, the vortex timedependent position. Putting it differently, although the classical solution
(17) we substituted in (18) was valid only for a vortex moving at constant
speed, we use it also to describe a vortex whose speed is not constant. This
is, of course, an approximation, valid when the vortex acceleration is small
enough.
Having done that, we completed out reduction of the problem to that
expressed in terms of the vortex only. Two consequences of the vortex
62
Lagrangian (19) are immediate: the Hamiltonian corresponding to the Lagrangian (19) is
Hvortex
= 2Md
P -
21fn Q)2
(22)
2e
where P is the vortex momentum, and its quantum mechanical eigenenergies are characterized by the vortex momentum quantum number N,
= _1_
2Md
(21fn)2 (2 N _ Q)2
2eL
e
(23)
Discussion Well, the vortex Hamiltonian (22) and eigenvalues (23) are
exactly what the previous Section would lead us to expect. What they tell
us about the vortex is:
1. It is a massive particle whose dynamical mass is determined by the
electrostatic energy. The physics of that mass is simple. A moving vortex implies a time dependent phase configuration O(x-X(t)), implying
an electric field proportional to X(t), implying an electrostatic energy
proportional to X2 , defining a vortex mass.
2. It interacts with a vector potential. The number of flux quanta associated with the vector potential is
and the analog for a flux quantum
is a Cooper-pair.
3. Since the junction is ideal, the vortex Hamiltonian commutes with the
vortex momentum, whose eigenvalues are 27riN.
4. The Hamiltonian (22) is, apart from the replacement
-+
the
Hamiltonian of an electron on a one dimensional Aharonov-Bohm ring.
The analogy to the persistent current problem implies then that for
~ which is not an integer or an half integer, a vortex in the N'th
eigenstate of the Hamiltonian (22) moves around the ring persistently
across the
in a velocity ~L n, and this motion induces a voltage
ring.
S,
S 10,
ate!
Now, before we consider the effects of disorder and dissipation, there

are some important points of interpretation we should dwell on:
The first point is the physical meaning of the integer N, characterizing
the eigenvalues (23). It is the vortex momentum quantum number, with
the momentum being 27riN. We know that from the Hamiltonian (22).
However, let us look closely at the eigenvalues (23). They seem to describe
a capacitor of effective capacitance Gef f == Md (~~~) 2 charged by a charge
Q - 2eN. This charge is the total charge on the junction, the sum of the
externally imposed bias charge Q and the Cooper-pairs charge imbalance
2eN. Thus, N also counts the number of Cooper-pairs that tunneled across
63
the junction. The quantum number that quantizes the vortex momentum
also counts the number of Cooper-pairs that tunneled across the junction.
Moreover, N has another interpretation. The collective coordinates X
and P can be represented in terms of the fields 0, n. Their representations
are given by [23]
= 2~
and
P = -nhz
f
f
xOx(x) dx
(24)
n(x)Ox(x) dx
(25)
As seen from Eq. (24), X is the "center of mass" of the vortex, and P is
the total momentum of the field. The electric field in the junction, which
points in the radial direction, is made of one part due to the Cooper-pairs
density, given by En = 87ren and one part due to the externally imposed
charge density Ecr = 87reO". The first is a dynamical field in the classical
description, and an operator in the quantum description. The second is
a c-number. The magnetic field in the junction points in the z-direction,
and is B =
Ox' Substituting these expressions into Eq. (25), we observe
that the field momentum P is the electro-magnetic momentum associated
with the electric field En, i.e., induced by the Cooper-pairs in the junction.
It is,
4tt
C
P = 47rc
2
En
Bd3 x,
(26)
The quantization of P is then a quantization of the dynamical electromagnetic field momentum.

The second point is the effective capacitance of the junction. Being
proportional to the vortex mass, it depends on the ratio fJ' Using the
expressions for the mass (Eqs. (20) and (21)) we find that the effective
junction capacitance is identical to the geometric capacitance ~ for a
short junction (L AJ ), but differs from it for a long junction (L AJ).
In the latter case the effective capacitance is given by 2~::AJ'
For short junctions, the sole effect of the vortex is to turn off the Josephson coupling, and make the junction a capacitor. The charging is uniform
along the junction, and so are the electric and magnetic fields. Therefore,
the effective capacitance is the geometric one.
The long junction case is more interesting. The presence of the vortex
shuts off the Josephson coupling, and makes the junction a capacitor. However, while we assumed the bias charge density to be uniform, we find that
the total charge density (bias charge + Cooper-pairs) is non-uniform, as
we see by examining the electric field. The electric field (ex: O(x)) is non-zero
only within the vortex, in a region of order AJ around the moving center
of the vortex. In the rest of the junction, the externally put bias charge
64
is screened by the Cooper-pair density n(x). Thus, the unscreened part
of the charge is (Q - 2eN)
the size of the charged part of the capacitance is AJ x hz, and the electrostatic energy it carries is of the order of
[( Q - 2eN)
Afh z ' Electro-statically, then, a moving vortex is a moving
capacitor. This last part of the discussion applies more generally: the size
of the vortex determines the effective size of the capacitor, and thus affects
the capacitance of the junction (or junction array).
(J
A comment on the Elion et al. experiment The quantum mechanics of single vortices, described theoretically above, was studied in an experiment by
Elion et.al.[25] In this experiment, a "generalized Aharonov-Casher" effect
was observed for a vortex in a ID array. As we explain below, a vortex
in a ID periodic array is very similar to a vortex in a long junction. The
Elion et. al. experiment is essentially a double slit interference experiment
where the interfering degree of freedom is that of a vortex. In an oversimplification, the geometry of the experiment could be described as that of a
circular Josephson junction, to which two semi-infinite Josephson junctions
(" leads") are symmetrically attached. The circular junction is charged by a
bias charge Q: a charge Q is put on the inner super conductor, and a charge
-Q /2 on each of the two outer super-conductors. A vortex is created at the
far end of one of the leads, and is accelerated to a velocity v before it gets
to the circular junction. The presence or absence of quantum interference is
analyzed by examining the dependence of system's resistance on the charge
Q.
In our discussion above we used the mapping of the vortex momentum

on the number of Cooper-pairs charging the junction to describe the Qdependence of the junction's energy. This mapping clarified the suppression
of the Josephson current in the junction and the onset of capacitive behavior, but was not very useful in a calculation of the junction's capacitance
when it differed from the geometric one. The same statement holds for the
present case. The superposition of a vortex wave packet going along the
right half of the circular junction with a vortex wave packet going along
the left half can be described as a superposition of a charging of the right
half and a charging of the left half. The number of Cooper-pairs involved
in that charging is again proportional to the momentum of the vortex. This
mapping should perhaps be kept in mind when one attempts to think of
vortex interference in terms of a macroscopic quantum effect, and may hint
that the latter is an ill defined concept.
Disorder The main source of disorder relevant to our case is a spatial
disorder in the Josephson coupling energy EJ(X). In terms of the vortex
description, such disorder makes the vortex rest mass space-dependent,
65
i.e., it amounts to the introduction of a disordered potential. Consequently,

its effect can be understood using the mapping of our problem onto that of
an electron in a disordered Aharonov-Bohm ring.
The energy levels given by Eq. (23) are a set of parabolas (as a function
of Q), each centered at an integer multiple of 2e. The parabolas intersect
one another at Q = (2j + l)e (j being an integer). Spatial disorder opens
gaps in the energy levels at the intersections of the parabolas. Then, if the
bias charge Q is adiabatically turned on, whenever it becomes (2j + l)e,
the vortex momentum is changed by 21ft. This can be better understood
in terms of the Cooper-pairs number. Whenever the value of Q becomes
(2j + l)e, a Cooper-pair tunnels across the junction, and the momentum
of the vortex is changed by 21ft. The junction is not a perfect capacitor
anymore.
When the disorder is strong enough such that the vortex mean free path
is much smaller than the ring's circumference, the vortex is localized. Then,
the sensitivity of junction's energy to the bias charge is exponentially small,
and so is also the capacitance of the junction.
Dissipation and dephasing Quantum effects involving the phase of the
vortex' wave function may be suppressed by the interaction of the vortex
with other degrees of freedom. There are two such interactions. The first is
the vortex interaction with other excitation modes of the super-conducting
phase field O(x), i.e., with other excitations of the Sine-Gordon Hamiltonian
(15). This interaction is discussed in Ref. ([28]). Here we confine ourselves
to a qualitative discussion of the second interaction, with degrees of freedom
that are not included in the Sine-Gordon description. The main relevant
degree of freedom we have in mind is that of Bogolubov quasi-particles.
Our discussion of dephasing, in Section (2.4), has prepared us to discuss
the dephasing of the vortex' phase by its interaction with quasi-particles. As
we discussed before, dephasing can be understood in two ways. Within one
point of view, the vortex wave function is dephased when the uncertainty
in its phase becomes of the order of 21l". The phase accumulated by the
vortex' wave function is proportional to the bias charge it encircles. When
quasi-particles are present, and tunnel across the junction, the bias charge
fluctuates, and the phase becomes uncertain. By the fluctuation dissipation
theorem, the charge fluctuations are proportional to the product of the
temperature and the dissipative conductance of the junction. Consequently,
we expect the phase breaking rate to follow the same behavior.
The other point of view of dephasing considers the effect of the vortex
on the quasi-particles. When a vortex moves along the Josephson junction,
it induces an electric field between the two sides of the junction. If there are
quasi-particles in the super-conductors, an electric field leads to a normal
66
current. The vortex wave function is dephased when the number of quasiparticles that tunnel due to this electric field is large enough to shift the
quasi particles to a state orthogonal to their original one. Again, the rate
at which this orthogonality develops is proportional to the product of the
temperature and the conductance of the system.
3.4. GEOMETRIC PHASES ASSOCIATED WITH VORTICES IN ARRAYS

OF JOSEPHSON JUNCTIONS
Vortices in Josephson junction arrays are intensively studied, both theoretically and experimentally [29]. We do not intend to cover this rich field
of research here, but rather point out how the concepts outlined above are
useful for understanding the dynamics of such vortices.
One dimensional arrays A one dimensional Josephson junction array is a
slight variation on the theme studied above. The variation is a spatial modulation of the Josephson coupling J, making it a function of position, J(x).
Since the Josephson coupling determines the potential energy of the vortex,
a spatial modulation of that coupling induces a potential energy term in
the vortex Hamiltonian (22). In particular, if the Josephson coupling varies
periodically, the vortex on a ring, studied in the previous section, becomes
a particle on a ring under the effect of a periodic potential. The spatial
dependence of the Josephson coupling does not change, however, the interaction of the vortex with the vector potential and "magnetic field". Thus,
most of our study of the long Josephson junction remains valid, with some
important differences, which we now discuss.
On top of exerting a potential energy on the vortex, the periodic modulation changes the structure of the vortex. In a long Josephson the vortex
size, AJ, is determined by the interplay between Josephson and inductance
energies. In contrast, in the case of an array the inductive energy is usually
negligible, since the Josephson currents are small. The size of the vortex,
the scale over which the phase winds approximately by 211", is the size of a
unit cell of the array. This is so since the energy cost for having cos () =1= 1
is minimal where the Josephson coupling is minimal, so that most of the
phase winding happens there. When the vortex moves in an array, then, it
charges only a small part of the array, one Josephson junction at a time.
In that sense, then, a ID Josephson junction array is analogous to the long
junction case studied above.
As for the dynamics of a vortex in a ID Josephson junction array, it can
indeed be mapped onto a particle in a ID periodic potential. Consequently,
we may expect the energy of a ring shaped array to depend periodically on
67
the bias charge, just as in the long Josephson junction case, but one expects the vortex mass (and hence the junction's capacitance) to depend on
the periodic potential (i.e., on EJ). Moreover, one may expect phenomena
unique to particles in a periodic potential, such as Bloch oscillations, to occur. Such effects were recently studied experimentally by van Oudenaarden
et.al.[30]
Two dimensional arrays Schematically, a two dimensional Josephsonjunction array is a checker board pattern in which the black squares are superconductors and the white squares are voids. It allows a study of a wealth
of classical and quantum phenomena of vortices which are distinctively
two-dimensional. Examples include Kosterlitz-Thouless transitions, superconductor to insulator transitions, quantized Hall states and more. A survey
of this intensively studied field can be found, e.g., in [29]. Here, we confine
ourselves to a short introductory description of the array's Hamiltonian in
terms of phase and number variables, and its reduction to vortex degrees
of freedom.
The dynamical degrees of freedom describing a 2D Josephson junction
array are two sets of conjugate variables, the phase of each island {<Pi} and
the number of Cooper-pairs on each island {nd. In a quantum description,
both become quantum operators. The phase is defined only modulo 27[, and,
consequently, its conjugate variable is an operator with integer eigenvalues.
The Hamiltonian of the array consists of two terms. The first term is the
Josephson energy, describing the coupling of nearest neighboring islands.
This coupling is characterized by an energy scale EJ, and depends also on
an externally applied magnetic field B z (or a vector potential A). This
magnetic field controls the number of vortices in the array.
The second term in the Hamiltonian is an electrostatic energy term. The
electro-static energy depends quadratically on the number of Cooper-pairs
on each island {ni}. The set of charges needed to minimize this quadratic
function is denoted by {nx ,d.
The Hamiltonian describing the array is [31],
2
(22e ) 2::(ni - nX,i)C{/(nj - nx,i) +EJ

ij
2::(1- cos (<Pi - <Pj _1 A dl))

.
W)
(27)
where I: (ij) denotes a sum over nearest neighbors. The integral over A
is taken between the sites i and j. A factor of
is absorbed in A. The
matrix 6- 1 is the inverse of the capacitance matrix C. Generally, the matrix
6 includes elements coupling an island to its nearest neighbors, to the
substrate and to neighbors further away. The dependence of CiJ 1 on the
distance between i and j depends crucially on the nature of the substrate.
2:
68
One important feature of the Hamiltonian (27) is simple to observe: its
spectrum is unchanged when any of the nx,i is varied by an integer number.
Vortices in Josephson junction arrays The reduction of the Hamiltonian
(27) to a vortex Hamiltonian is done by a method similar to that employed
in the previous Section, although some of the details are more complicated.
The resulting Hamiltonian describes the vortices as massive, interacting
particles subject to a periodic potential, and interacting with a vector potential.
The "magnetic field" associated with this vector potential is proportional to bias charge {nx,i}' The bias charge {nx,i} plays here the role played
by Q in the case of the long junction. The flux associated with the vector
potential is carried by the fluid particles, the Cooper-pairs, and those reside on the super-conducting islands, where the vortex never enters. Thus,
the Josephson junction array is a realization of the Azbel-Hofstadter problem, that of an electron on a lattice in a magnetic field. In that problem,
the spectrum of electronic states (the "Hofstadter butterfly") is periodic
with respect to adding one flux quantum to any of the lattice plaquettes.
Similarly, here, the spectrum is unchanged when nx is varied by one.
As for the other dynamical properties of the vortices: their mass is determined by the charging energy scale Ec. Their interaction is logarithmic,
and its magnitude is determined by the Josephson energy EJ. The periodic potential is induced by the array itself, and its magnitude is, again,
determined by E J .
4. A Spin in a Textured Magnetic Field - Berry's Phase, Mesoscopic Transport and Classical Forces
This section describes a study of the geometric phase, also called Berry's
phase[32], and its effect on the electronic transport of a mesoscopic system
in a magnetic field whose direction varies in space. We start by a general
discussion of the concept of the geometric phase, as it was presented by
Berry in 1984. Then, we concentrate on a particular example of a geometric phase, the phase accumulated by an electron in a non-uniform magnetic
field due to the Zeeman interaction of its spin. Mostly, we study how this
phase affects the electronic transport of mesoscopic and macroscopic conducting rings.
In Section (2.4) we discussed how the interaction of an interfering electron with its environment leads to dephasing. Berry's phase is also a result
of the interaction of an interfering electron with its environment, but a very
different result. It happens when the dynamics of the electron is much slower
than that of its environment. Then, we may use the Born-Oppenheimer ap-
69
proximation, in which the position of the slow degree of freedom is regarded
as a parameter in the Hamiltonian of the fast degree of freedom. When the
electron moves, the eigenstates of the "fast Hamiltonian" change. By the
adiabatic theorem [33], if the fast degree of freedom is initially at an eigenstate of its Hamiltonian, it follows the time evolution of that state. Then,
when the slow degree offreedom (the electron) traverses a closed trajectory,
the final state of the fast degree of freedom is identical to the its state, up
to a phase factor.
It is this phase factor which is of interest here. To calculate that phase
we write down the Schroedinger equation governing the evolution of the

fast environment,
i~(t) = H(t)~(t)
(28)
where H(t) = Henv+ V (x(t)), and x(t) is the coordinate of the slow degree
of freedom. If the environment follows adiabatically an eigenstate IXn(t))
of H(t), its state is,
(29)
where H(t)IXn(t)) = fn(t)IXn(t)). Substituting Eq. 29 in Eq. 28, and multiplying from the left by (Xn(t)1 we obtain an equation for the phase ,(t),
h(t) = fn(t)
+ (Xn(t)1 :t IXn(t))
(30)
The rate of accumulation of phase is then composed of two terms: the first
term is the term we are familiar with from time-independent Hamiltonians
- the energy. The second term arises from time dependence of the Hamiltonian. Since this second term is a total derivative with respect to the
time, its contribution to the total phase accumulated along a given path
depends on the geometry of the path, and not on its time-dependence. Just
like the Aharonov-Bohm effect, with all the consequences following, as we
see below. Note that the effect of the environment here is different from
the one discussed in Section(2.4). The environment does not de phase the
interference pattern, it shifts it.
4.1. A CONDUCTING RING IN A SPACE-DEPENDENT MAGNETIC
FIELD
The simplest example that illustrates the concept of Berry's phase is that
of a spin-~ that follows adiabatically a magnetic field whose direction varies
in time. When the magnetic field returns to its initial direction, the spin
wave function is found to have acquired a geometric phase factor, given by
half the solid angle subtended by the magnetic field during its variation.
As we see below, this phase can be regarded as induced by a geometric
flux, similar to the phase shift induced by an electro-magnetic flux in the
70
Figure 4. The physical problem considered. A ring is put in a uniform external magnetic
field B z , and a tangential magnetic field Bq, created by the current carrying wire. The
ratio between the two fields define the angle Q.
Aharonov-Bohm effect. [32] [34] Thus, we might expect Berry's phase analogies to thermo-dynamical effects like the persistent currents, and transport
effects, like the classical Faraday's law for time dependent flux, and like the
quantum effect of time-independent flux on the conductivity of a mesoscopic ring (through the Aharonov-Bohm effect) [35] [36]. The thermodynamical effect was studied by Loss, Goldbart and Balatzky. [37] Here, we
discuss transport effects[38]. In all these analogies, the electron's spin plays
the role played by the electric charge in the electro-magnetic effects.
We consider a quasi-one dimensional ring, whose radius is a. The ring
lies in the xy plane, and its center is in the origin. A non-uniform magnetic
field is applied on the ring (we use a cylindrical coordinate system): first,
a magnetic field B tangent to the ring is induced by a current carrying
wire lying along the z-axis. Second, a uniform field, Bzz, is applied on
the system. Along the ring, the magnitude of the field is constant, but the
direction varies. In fact, it follows a cone shaped path, where the angle
between the cone and the z-axis, denoted by a, satisfies tan a = ~ (See
Fig. 4.1). The angle a is regarded as a controlled variable in the experiment.
The spin of an electron that slowly encircles the ring is then expected to
follow the direction of the magnetic field and thus accumulate a geometrical
phase of
(31)
The t, + and .}, - refer to the spin being parallel and anti-parallel to the
field, respectively.
71
In the following Sections, we analyze the dependence of various physical

properties on the geometric phase. By mapping that phase onto an effective vector potential, we show that when the phase is time-independent, it
affects the ring's conductance. When the phase varies in time, it induces a
current in the ring. By discussing the analogies to the electromagentic phenomena, we point out that the effect of a time-independent geometric flux
is observable only in mesoscopic rings, while the effect of a time-dependent
geometric flux should be observed also in macroscopic rings, since it does
not depend on phase coherence. We discuss the conditions under which the
adiabatic approximation is justified, and what happens when it is not. We
end with comments on the relations of these effects to spin-orbit coupling.
While for practical reasons our discussion is concentrated on the electric
properties of the ring, we nevertheless stress that the electric charge of the
electron plays no role in our analysis.
4.2. THE ADIABATIC APPROXIMATION
In this section we use the Born-Oppenheimer approach to separate the

Hamiltonian of the system into two parts, one (the adiabatic part) in which
the spin follows adiabatically the direction of the magnetic field, and one
(the non-adiabatic part) which is purely non-diagonal with respect to the
eigenstates of the adiabatic part. We show that the adiabatic part includes
a geometric vector potential that couples to the electron's spin.
Assuming that the ring is one dimensional, and denoting its azymuthal
coordinate by </J, its Hamiltonian is
II2
H = 2M
+ V(</J)
(32)
-/-lB(</J) . a
where II = -~ d~ - eB2zc1ra is the generalized momentum (a system of units

where n = 1 is utilized), V(</J) is the impurity potential along the ring, /-l
is the magnetic moment, M is the mass of an electron, and a is the Pauli
matrices vector. We diagonalize the spin dependent part of the Hamiltonian,
treating the angle </J as a parameter. The two eigenstates corresponding to
the spin being parallel (anti-parallel) to the magnetic field are
It( </J)) =
(i cos-sm~e:i
2
(33)
and
The corresponding eigenvalues are ":fpB where B == JB~
+ B;. Defining
now I</J) as the eigenstate of the operator ei1 , the two sets of states {It( </J))
I</J)
10 ~ </J < 21f}
and {IH</J)) I</J)
10 ~ </J < 21f}
constitute together a
72
basis of the Hilbert space of the Hamiltonian (32) . Each one of these sets
span a subspace in which the spin is either parallel or anti-parallel to the
magnetic field. The impurity potential is spin independent, and hence, it
is diagonal in that basis. However, the kinetic part of the Hamiltonian
has both adiabatic matrix elements and matrix elements that induce spinflips. A simple calculation shows that the matrix elements connecting states
within the first sub-space are,
(t( 4 I
(4)12~ 14>') It( 4>')) = (4)1
[IT
2~
Ot]2
9
14>')
(34)
The corresponding matrix element in the second subspace has O~ rather

than
These matrix elements demonstrate that within the adiabatic approximation, the spatial variation of the magnetic field induces a vector potential
[39] whose magnitude is independent of the electron's charge, but is rather
determined by the direction of the spin being parallel or anti-parallel to the
field. An alternative way of writing the matrix elements (34) is[40]
0b.
Ho
[IT - A ]2
2M 9
+ V(4)) - /-LB(4)) . a + 8Ma2 sin2 a
(35)
where Ho is the adiabatic part of the Hamiltonian, the projection of the

full Hamiltonian onto the subspaces in which the spin follows adiabatically
the magnetic field, and Ag = 21a sin a[cos aa . - sin aaz] Note that Ag has
non-zero matrix elements only between states of opposite spin directions.
In these terms, the non-adiabatic part of the Hamiltonian, HI, is given by
(36)
By construction, Ho has a set of eigenstates In, t) = It(4))) 'l/J"A(4))
in which the spin is parallel to the field, and a set of eigenstates In,.}) =
1+(4))) 'l/J*(4)) in which the spin is anti-parallel to the field. The wave
functions 'l/J"A(4)) and 'l/J*(4)) satisfy the Schroedinger equations HJq,)'l/J~(+) =
E~W'l/J~W, where the Hamiltonians HJW are given by,
1 [IT - _otU)
1 ] 2 + V(4)) T /-LB + --sin
1
2 a}
HJU) == { 2M
27ra 9
8M a2
(37)
The significance of the induced vector potential becomes clearer when

one considers space translation transformations. The momentum operator, Pcp, is the generator of such a transformation, i.e., for any state w,
73
. ppa4>o
Ii
Iw) = (<p + <Polw). In such a transformation, the electron is
translated spatially, but the direction of the spin is kept constant. On the
contrary, the generalized momentum appearing in the adiabatic Hamiltonian, II - Ag is the generator of a different translation transformation, a
transformation in which the electron is translated spatially, and the direction of the spin follows the direction of the field.
(<ple-~
We conclude this section by emphasizing its main conclusion: Under

conditions in which HI can be disregarded, the ring can be viewed as composed of two uncoupled electron gases. Those gases are subject to the effect
of different geometric vector potentials and opposite constant potential energy, originating from the Zeeman interaction. They are also subject to
the effect of identical electro-magnetic flux B z 7ra 2 and identical impurity
potential. Each of the two gases obviously does not have a spin degeneracy.
4.3. NON-LOCAL AND LOCAL EFFECTS OF THE GEOMETRIC FLUX
ON ELECTRONIC TRANSPORT
We now assume that the magnetic field is strong enough for the adiabatic
limit to be applicable. Assuming that the Zeeman energy J.LB is smaller than
the Fermi energy, EF, our ring consists of the two uncoupled electron gases
described above. The electric conductance of the ring is then the sum of
the conductances of the two gases. The conductance of a meso scopic ring
depends on a magnetic flux threading the ring, through the AharonovBohm effect[35][36]. For rings in the diffusive regime, the flux dependence
of the conductance is manifested in the average conductance of an ensemble
of macroscopically identical rings as well as in sample-specific fluctuations.
The flux-dependent part of the average conductance was calculated by
Al'tshuler, Aronov and Spivak[41] and shown to be,
8a
= _ e2 a
sinh(r)
f cosh(f) - cos(~)
(38)
where <P is the flux threading the ring, f == 2~a and l> is the phase breaking
length. In the configuration we discuss, the flux threading the sample is
a sum of an electro-magnetic flux <Pem = B z 7ra 2 , and the geometric flux
<pg = ~o (1 cos a). The sum of the two geometric fluxes corresponding to
the two gases equals a flux quantum, so that one can view the two electron
gases as subject to the influence of geometric fluxes of equal magnitude and
opposite signs. The total quantum correction to the conductivity is given
74
by
fJ(J
= _ e2 a [
sinh(r)
cosh(r) - COS(47f(>em+>g))
>o
sinh(r)
]
r
cosh(r) _ COS(47f(>em->g))
>o
(39)
The ~-periodic component of the Aharonov-Bohm oscillations of the conductance is then multiplied by a geometrical factor, cos(4~~g). Note that
the difference between the Fermi wavelengths of the two spin directions is
not reflected in the expressions above, since the quantum correction to the
conductivity is independent of kplel'
The effect of the geometric flux on the sample-specific fluctuations of the
conductance is best understood when the periodicity of those oscillations
with respect to Bz is considered. In the absence of geometric flux (B> = 0),
the field periodicity is tlB z =
irrespective of the spin direction. In
the presence of geometric flux, a variation of B z varies both the electromagnetic and the geometric fluxes. Thus the periodicity with respect to Bz
is changed, and is no more independent of the spin direction. Specifically,
when B z B> (i.e., a --+ ~), the geometrical flux is approximately ~a:z,
and the B z period becomes,
!!:X'
(40)
where the +( -) sign refers to the spin being parallel (anti-parallel) to the
field. The magnitude of the sample-specific fluctuations is not affected by
the geometric flux, i.e., it is of the order of ~ .
Eqs. (39) - (40) summarize our predictions for the effect of Berry's phase
on the conductivity of a mesoscopic ring. Its effect on the thermodynamics
of such a ring is discussed in details in [37]. We now turn to discuss the
case of a time-dependent geometric flux, and, in particular, the currents it
induces in the ring. We consider the case in which the tangential magnetic
field is B> = Bg cos wt. For simplicity, we limit ourselves to the case in which
the electron gas is completely spin-polarized, a case which is realized when
Ep + W fLB z . For semi-conducting rings, this condition may be fulfilled
at fields of the order of 1 Tesla. By passing, we note that another way to
realize a completely spin polarized electron gas is by an injection of spin
polarized electrons through a ferromagnetic-metallic interface. [42] Under
the assumption of complete spin polarization, the electron gas in the ring
75
is subject to the effect of a time-dependent geometrical flux
~ ~o (1+ msa(t)) ~ ~ {1+
1+
EO
Gt; cos wt)2
Consequently, this gas is subject to a motive force E, given by E = -~,

and this motive force induces a current in the ring, according to Ohm's law.
Assuming that B~ B z , the motive force induced by the time dependence
of g is
(41)
The frequency of the induced current is twice as large as that of B, so
that it can experimentally be distinguished from currents induced due to
the wire being not exactly perpendicular to the ring. For B z = 1 Tesla,
B = 0.2 Tesla and w = 1 GHz, this motive force has an amplitude of 10- 7
Volts. Similarly to the electro-motive force, the geometric motive force can
be amplified if the ring is replaced by a solenoid.
There are a few points that should be stressed regarding the case of a
time dependent geometrical flux. Firstly, contrary to the effect of a timeindependent flux, the time dependent geometric flux exerts a force on the
electron [43]. Thus, similar to the observation of currents induced due to
Faraday's law, the observation of currents induced by the time dependent
geometric flux does not depend on the electron phase being coherent along
the ring. Those currents should be observed in macroscopic rings, as well as
in mesoscopic ones. In fact, the force accelerating the electrons in the case
of a time-dependent geometric flux is classical, and can be derived from
Newton's equations[44]. Secondly, the motive force induced in the ring is not
electric, since if the electrons were replaced by neutrons, the picture would
not have changed. The field, given by the derivative of the vector potential
with respect to the time, does not couple to the electric charge, but rather to
the direction of the spin. Thirdly, the origin of the motive force exerted on
the electron can be understood by noting that in our symmetrical structure
the sum of the orbital and spinor angular momenta in the z direction is
time-independent even when the angle a is time dependent. Thus, a change
in a transfers angular momentum from the spin to the orbital motion of
the electron.
Finally, the geometric flux, motive force and current all depend on the
direction of the spin. Therefore, if the ring includes two electron gases with
opposite spin directions, the currents induced in the two gases are opposite
in direction, and the net current is proportional to the difference between
76
the conductances of the two electron gases in the ring. Such a difference
arises from the 2p,B difference between the kinetic energy of electrons in
the Fermi levels of the two electron gases.
4.4. WHAT HAPPENS WHEN THE ADIABATIC CONDITION IS NOT
MET?
An instructive starting point for answering this question is the ballistic

case, the case in which V() = 0, and the Hamiltonian (32) can be exactly
diagonalized. The exact eigenstates are,
In, t( )) = e in >
(i cos-sm~e~i
2
and
In H)) = e in >
,
(i sincos~e-i>
1.
2
(42)
The angle, is implicitly given by

cot, = cot a
n,2(2n' - 1)
(43)
+ 4M a2 p, B sma
.
where n' == n - e~zca2 . The corresponding eigenvalues are
E(n)
n, 2n 2
2Ma 2
n,2(2n' - 1)
4Ma 2 (1 cos,) =f p,Bcos b
a)
(44)
The significance of the angle , is understood by looking, say, at the eigenstate In, t). The expectation value of O"z for that state is cos" i.e., , is the
angle to which the spin bends relative to the z-axis. The expectation value
of the spin projection onto the direction of the magnetic field is cos (, - a).
Two other spin projections of interest are the projection onto two directions
perpendicular to the magnetic field, the direction of ~~, which is here the
radial direction, and that of B x ~~. The former is zero, while the latter is
sin b - a)[44].
As seen from the exact solutions Eqs. (42)-(44), when the magnetic
field is not strong enough to force the spin to bend in an angle a, the
spin bends to an angle, < a. The Zeeman energy is then proportional to
the projection of the spin onto the magnetic field, and the induced vector
potential is still of the form found in the adiabatic case, but with the angle
a replaced by ,.
However, in the adiabatic limit the vector potential was determined only
by a and the direction of the spin. Thus, it deserved the name" geometric" .
In the non-adiabatic case the vector potential depends, through the angle
" on the magnitude of the magnetic field and the velocity of the electron.
It is no more purely geometric.
77
The observations discussed above in the context of the ballistic case

allow for a qualitative understanding of the non-adiabatic limit of the diffusive case. Diffusive eigenstates are built out of superposition of many
momentum (or velocity) components. If the magnetic field is too weak to
force adiabaticity, each of these components is subject to a different vector
potential, and thus also to a different flux. If the range of fluxes induced in
the different momentum components is of the order of a flux quantum, the
energy of the diffusive eigenstate loses its sensitivity to the direction of the
magnetic field, and the geometric effects are lost.
4.5. CONDITIONS FOR THE VALIDITY OF THE ADIABATIC
APPROXIMATION
For the adiabatic approximation to be valid, the Zeeman precession frequency g~B should be large, but large compared to what? In the ballistic
case, Eq. (43) tells us that I ::::; a when gp,B ;;~2. For electrons at the
Fermi level, which are the electrons responsible for transport, this condition
implies,
gp,B
VF
(45)
-n a
The spin precession frequency should be larger than the frequency at which
the electron revolves around the ring.
In the diffusive case the frequency ~ plays, presumably, no role, and
there are two new frequency scale of interest. The first, and larger, is the
frequency of elastic collisions, ~, and the second is ~ (where D is the diffusion constant), the typical frequency at which an electron diffuses around
the ring.
Two calculations were carried out to examine the condition for the validity of the adiabatic approximation in the diffusive regime, with two contradicting outcomes. The first, carried out by the present author[38], led to
the condition g~B ~. That calculation is based on the following picture.
The non-adiabatic part of the Hamiltonian, HI of Eq. (36), induces scattering between opposite spin eigenstates of the adiabatic part Ho. Thus, it
leads to a finite lifetime T of the latter, and constitutes a dephasing mechanism. For non-local geometric phase effects to be observed, T should be
larger than the diffusion time around the ring ~. The lifetime T is calculated by perturbation theory, and the condition T ~ is found to lead to
!l.I!:.!i
.1T
Ii
The second calculation, carried out by Loss et. al., was based on a semiclassical analysis, and led to the less restrictive condition g~B ~ [45].
The precise condition for the validity of the adiabatic approximation in the
diffusive case is then still under debate.
78
4.6. HOW IS THE GEOMETRIC FLUX RELATED TO SPIN-ORBIT
COUPLING?
Some of the phenomena discussed in this chapter are similar to the phenomena resulting from a one-dimensional ring of spin-orbit scatterers [46].
It is instructive, then, to devote this section to the relation between the
geometric phase and spin-orbit coupling. This relation becomes clear when
spin-orbit coupling is expressed as an interaction with a vector potential.
The origin of the spin-orbit coupling lies in the coupling of a moving magnetic moment p, = 2::c(J to an electric field E. In the frame of reference
in which the magnetic moment is at rest, the electric field is Lorentztransformed to a magnetic field. If the velocity of the magnetic moment is
slow compared to the speed of light, the magnetic field in the rest frame is
given by ~ x E. The magnetic moment couples to that magnetic field via the
Zeeman interaction, thus yielding an interaction term p,. ~ x E = v ~ x E.
Having in mind the interaction term of an electron with an electro-magnetic
vector potential v . A, we find that ~ x E =
(J x E can be identified
as the spin-orbit vector potential. When the magnetic moment arises from
the internal spin of a charged particle, as in the case of an electron, the
acceleration of the particle due to the interaction of the charge with the
electric field has to be taken into account, and this leads to a correction factor of ~ to the above expressions. This factor of ~ is known as the Thomas
precession factor. [47]
Similar to the Berry geometric vector potential discussed above, the
spin-orbit vector potential is, in principle, space and spin-dependent, and
its values at different points in space do not necessarily commute. It is
important, however, to note the differences between the vector potential
resulting from the spin-orbit coupling and the one resulting from the Zeeman interaction with a space dependent magnetic field. The first difference
has to do with the symmetry with respect to time reversal. While spin-orbit
interaction gives rise to a vector potential, it does not break time-reversal
symmetry - it does not induce a =rp,B term. Thus, for each eigenstate for
which the effective spin-orbit flux is c!>, there is another state, degenerate in
energy, for which the effective flux is -c!>. This is Kramers' degeneracy. On
the contrary, the effective flux induced by the space-dependent magnetic
field is accompanied by the Zeeman energy, that removes the degeneracy.
The second difference is a difference in magnitudes. Being inversely
proportional to mc2 , spin-orbit interaction term is very small, unless it involves very strong electric fields. In the context of condensed matter physics
such fields result from microscopic molecular charge distributions, which
are strong enough to make spin-orbit coupling significant. The geometric
flux resulting from Berry's phase, on the other hand, is determined by the
externally controllable magnetic field.
2:2
79
This relation between spin orbit coupling and geometric phases has led
to two interesting suggestions. The first, by Lyanda-Geller and Aronov, is to
use the well ordered spin orbit coupling in semiconductors for observation of
geometric phases in mesoscopic systems[48]. The second suggestion, made in
a beautiful work by Lyanda-Geller, Loss and Goldbart, is that the coupling
of He 3 in its super-fluid magnetic phase to the walls of its container may
lead to a thermo-dynamical persistent current, provided that the container
is shaped appropriately[49].
5. Summary
This chapter presents an overview of the application of the concept of geometric phase to the physics of meso scopic systems. The most striking manifestation of this concept has so far been the Aharonov-Bohm effect, whose
observation in mesoscopic systems is by now commonplace. We described
two other manifestations, in the physics of vortices and in the physics of
spins in a magnetic field whose direction varies in space. In all subjects covered, we confined ourselves to theory, and covered only a few of the topics
involved. We hope, however, that the chapter is useful as an introduction
to the subject, and that the reference list given can be useful for further
studies.
Acknowledgments This work was supported by the US-Israel Binational
Science Foundation. I am indebted to Y. Aharonov, Y. Imry, E. Ben-Jacob
and Z. Hermon for collaborations on works described in this chapter, and
to G. Schoen, L. Kouwenhaven and L. Sohn for inviting me to write this
chapter.
References
1.
2.
3.
4.
5.
6.
7.
8.
Most quantum mechanics textbooks would remind you of the double slit experiment,
in case a deeper reminder is needed. Feynman's lecture on physics would be one
example.
There are many good reviews available. See for example, "Introduction to Mesoscopic Physics", by Y. Imry (to appear in Cambridge Press (1997), or the review
by B.L. Altshuler and A. Aronov in "Electron-electron interactions in disordered
systems" A.L. Efros and M. Pollak (eds.), North Holland (1985).
G. D. Mahan, Many Particle Physics New York: Plenum Press, 1990,
J. Rammer and H. Smith, Rev. Mod. Phys. 58, 323 (1986).
See, e.g., Y. Imry in "Quantum Coherence in Mesoscopic Systems", B. Kramer ed.
Plenum 1991.
B.L.Altshuler, A.G. Aronov and D.E. Khmelnitskii J. Phys. C15, 7367 (1982) .
A. Stern, Y. Aharonov and Y. Imry, Phys. Rev. A 41, 3436 (1990).
M. Biittiker, Y. Imry and R. Landauer, Phys. Lett. 96A, 365 (1983).
80
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
Y. Imry and A. Stern, Semicond. Sci. Techno!. 9, 1879 (1994).

Lamb, Horace,Hydrodynamics, New York: Dover, 1945.
Batchelor, G. K An introduction to fluid dynamics. Cambridge: Cambridge University Press, 1973.
D.J. Thouless, P. Ao and Q. Niu, Physica A, 200, 42 (1993). P. Ao and D.J. Thouless, Phys. Rev. Lett. 2158 (1993).
B. J. van Wees Phys. Rev. B 44, 2264 (1991).
E. Sonin, a review paper to be published.
Orlando T.P. and Delin KA., Phys. Rev. B 43,8717 (1991).
Onsager L., Nuovo Cimento 6 Supp. 249 (1949),
M.P.A. Fisher, Phys. Rev. Lett. 65, 923 (1990).
Arovas D, Schrieffer J.R. and Wilczek F., Phys. Rev. Lett. 53, 722 (1984).
B. J. van Wees, Phys. Rev. Lett 65, 255 (1990). See also R. Fazio, U. Geigenmuller
and G. Schoen in the Proc. Adriatico Res. Conference on quantum fluctuations in
mesoscopic and macroscopic systems (1990).
Y. Aharonov and A. Casher, Phys. Rev. Lett. 53, 319 (1984).
Z. Hermon, A. Stern and E. Ben-Jacob, Phys. Rev. B Phys. Rev. B 49, 9757 (1994).
P. Lebwohl and M. J. Stephen, Phys. Rev. 163, 376 (1967).
D.J. Bergman, E. Ben-Jacob, Y. Imry and K Maki, Phys. Rev. A 27, 3345 (1983);
A.C. Scott, F.Y.F. Chu and D.W. McLaughlin Proc. IEEE 61, 1443 (1973); R.
Jackiw, Rev. Mod. Phys. 49, 681 (1977).
A. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, Dover
Publications, 1965.
W. J. Elion, J. J. Wachters, L. L. Sohn and J. E. Mooij, Phys. Rev. Lett. 71, 2311
(1993); Physica B 203, 497 (1994).
D. J. Thouless, Phys. Rev. Lett. 39, 1167 (1977).
P. W. Anderson, D. J. Thouless, E. Abrahams and D. S. Fisher, Phys. Rev. B 22,
3519 (1980).
Z. Hermon, A. Shnirman, E. Ben Jacob Phys. Rev. Lett. 74, 4915 (1995).
Proceedings of the NATO Advanced Research Conference on Mesoscopic on Mesoscopic Superconductivity, edited by F. W. J. Hekking, G. Schon and D. V. Averin
(North Holland, Amsterdam, 1994). Proceedings of the ICTO workshop Physica
B 222 issue 4, 1996. See also: H.S.J. van der Zant et al in Quantum coherence in
mesoscopic systems, B. Kramer (ed.) Plenum Press (1991).
Van Oudenaarden A., Vardy SJK arid Mooij J.E., Cszechoslovak J. of Phys. 46, 707
(1996).
U. Eckern and A. Schmid, Phys. Rev. B 39 6441 (1989).
M. V. Berry, Proc. R. Lond. A392, 45 (1984).
A. Messiah, Quantum Mechanics, Chapter 17, J. Wiley & Sons, New York (1962)
A. Shapere and F. Wilczek, Geometric Phases in Physics, World Scientific (1989).
Y. Imry in Directions in Condensed Matter Physics eds. G. Grinstein and G.
Mazenko World Scientific, Singapore (1986).
B.L. Altshuler, A.G. Aronov, D.E. Khmlenitskii and A.1. Larkin
in: Quantum theory of solids, ed. I.M. Lifschits, MIR publishers,
Moscow (1982).
D. Loss, P. Goldbart and A.V. Balatsky Phys. Rev. Lett. 65, 1655(1990).
81
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
A. Stern, "Berry's phase, motive forces and mesoscopic conductance" Phys. Rev.
Lett., 68,1022 (1992).
M. Stone Phys. Rev. D 33, 1191(1986); J. Moody, A. Shapere and F. Wilczek Phys.
Rev. Lett. 56, 893(1986).
Y. Aharonov, E. Ben-Reuven, S. Popescu and D. Rohrlich Phys. Rev. Lett.
65, 3065(1990).
B.L. Altshuler, A.G. Aronov and B.Z. Spivak Pisma Exsp. Teor. Fiz. 33, 101(1981),
JETP Lett. 33, 94(1981).
S. Datta and M. McLennan, Rep. Prog. Phys. 53, 1003 (1990); A. Aronov, U. Sivan,
A. Yacoby, private communications.
M.V. Berry The quantum phase, five years later in Ref.[34]
Y. Aharonov and A. Stern, Phys. Rev. Lett. 69, 3593 (1992).
D. Loss, H. Schoeller and P.M. Goldbart, Phys. Rev. B 48, 15218 (1993).
Y. Meir, Y. Gefen and O. Entin-Wohlman Phys. Rev. Lett. 63, 798(1989).
J. D. Jackson, Classical electromagnetism, Chapter 11 J. Wiley & Sons, New-York
(1975); G. Baym, Lectures on Quantum Mechanics, Chapter 23, W.A. Benjamin
Inc. (1969).
Lyanda Geller Y., Surf. Sci. 362, 692 (1995) and references therein.
Lyanda Geller Y., Goldbart P.M. and Loss D., Phys. Rev. B 53, 12395 (1996).
DELOCALIZATION, INELASTIC SCATTERING AND

TRANSPORT DUE TO INTERACTIONS
YOSEPH IMRY
Weizmann Institute of Science,
Department of Condensed Matter Physics,
IL-76100 Rehovot, Israel
Abstract.
The enhancement of electronic transport in the localized regime by the
electron-electron interactions is discussed both in the incoherent channel,
leading to transport via inelastic scattering, and in the coherent channel,
via delocalization. For the former channel, a new prediction for the temperature dependence of the conductivity in a weak granular insulator is given.
The conditions for an effective decay of a given state weakly coupled to a
quasicontinuum are obtained. This is applied to both the decay of a singleelectron excited state due to interactions and to interblock transport in the
scaling theory of localization. Thus the latter theory can include some of the
effects of the interactions. Two-electron delocalization is a good example
for that, which is also treated in this paper.
83
84
1. Introduction
The electron-electron Interactions are by now appreciated to have a much

more profound effect in disordered systems than in ideal, pure and ordered
ones. Schmid[1] discovered the enhanced inelastic electron-electron scattering in the disordered case. Important corrections to both the density of
states and to the transport have been obtained by Altshuler and Aronov[2].
The very nontrivial issue of the dephasing rate as compared with the simple inelastic one, which has a nonphysical divergence at low dimensions,
was first understood by Altshuler, Aronov and Khmelnitskii[3j.1 A simple
physical picture for the last effect, based on phase uncertainty differences,
was given by Stern et al.[5], relying on the semiclassical picture[6]. The important case of the metal-insulator transition is discussed by Finkelstein[7].
Ideas for the enhancement of the equilibrium orbital magnetic response in
the mesoscopic case by the interactions were given by Miiller-Groeling et
al.[8]. It is of interest to extend these studies to the localized phase and to
granular and quantum-dot systems.
The general theory for the Inelastic scattering and associated level
broadening due to the electron-electron interaction, will be reviewed in section 2, emphasizing the isolated zero-dimensional (OD) quantum dot[9, 10].
Motivated by several intriguing experiments, this inelastic scattering rate,
,,(, and the resulting transport are calculated, following ref. [11], for a system
of weakly coupled quantum dots. Being due to longitudinal electromagnetic
fluctuations, "( is determined by the conductivity of the medium. The latter
is proportional to ,,(, as shown by Thouless[12] and in refs.[13, 14]. Thus,
a self-consistent expression for "( is obtained. A linear temperature dependence of"( (and hence of the conductivity) follows for this "granular insulator", for a wide temperature range above the single-particle level spacing
in the metallic quantum dot. The scale dependence of the parameters (such
as the conductivity) used in this calculation is of a rather general applicability. It is suggested that it may be used in further situations, such as that
of "anomalous diffusion" , as well.
The subtle issue of what is the effect of the discreteness of the final
states on the decay and the level broadening of a resonance state[15] weakly
coupled to a noninteracting quasicontinuum is discussed in section 3. It is
shown that when the nominal width, r, becomes comparable to the level
spacing of the quasicontinuum, a crossover between a weak admixture of a
small number of states to a more complicated stuation, resembling a decay,
occurs. This leads to a simple interpretation of an extremely interesting
transition in Fock space, recently found by Altshuler et al.[16].
1 An
excellent review of these topics is given in ref.[4]
85
The generality of the Thouless picture[12] for two coupled "blocks" (refs.
[17, 18, 19, 20]) is then discussed in section 4. Two coupled quasicontinua
(refs. [17, 18]) are considered, the crossover discussed above now implies
going from localization in the separate systems to delocalization. Thus, the
Thouless picture for localization is very general and may include interactions as well as disorder[19, 20]. This yields a simple and general explanation
for the delocalization of two-electron states by the interactions[21, 19, 20],
reviewed in section 5.
2. Inelastic scattering, level broadening and transport in the

granular insulator
In the presence of the interactions, the single-particle excitations of a manybody system are not only shifted in energy, but are also no longer exact
eigenstates of the system. If the system is prepared in a single-quasiparticle
state of energy E (much lower than the Fermi energy and the bandwidth),
it will decay within a time Tee(E), by creating electron-hole pairs, into the
many-particle quasicontinuum. This decay can be understood as due to the
finite lifetime of the single-paricle state, which gives it a width of order
'Y = li/Tee(E). An alternative, and in principle a more rigorous approach
(see the next section) is to regard this quasiparticle resonance lineshape as
a superposition of the numerous many-body excitations at the given energy.
Each eigenstate appearing with its particular weight (strength function[15])
in the initial" quasiparticle" wavepacket. This width increases with E, and
as long as
'Y
= li/Tee ;S E,
(1)
which we are going to assume in this paper, the Landau Fermi-liquid theory
is believed to be approximately valid. For electrons in a disordered system,
with a given mean free path and diffusion constant D, this width was
first calculated by Schmid[I], and found to be much larger than in the
ballistic ( --+ 00) case. Altshuler, Aronov and Khmelnitskii[3] considered
this problem further, and found that at low dimensions, d ~ 2, when'Y is
calculated straightforwardly, it diverges at finite temperatures. The physical
scattering time is then the dephasing time, T, which was discussed and
calculated in Ref.[3]. It is the time for which interference between any two
different electron paths is lost (see 2 ref.[5]) due to inelastic scattering. While
this circumstance is of crucial importance at d = 1,2, the times Tee and T
are basically the same at d = 3 and also for the quantum dot case. For
simplicity and definiteness, we consider here only the latter situation, but
2In the semiclassical picture[6] the divergence of ree is cured by taking the difference[5]
in the phase uncertainties of the two paths.
86
our results can easily be applied to the case where the underlying system
is one- or two-dimensional as well.
The general expression for Tee, derived first by perturbation theory and
later from the semiclassical picture, for electrons in the diffusive metallic
regime is[4]:
W -
E]
coth 2kBT - tanh 2kBT
(2)
The integration variable W is the energy transfer in the scattering. It

is usualy limited to be between 0 and E for E k B T because of the Pauli
exclusion principle, and by energies on the order of kBT otherwise. In the
main calculation to be presented in this section, a new and different cutoff,
W max , see eq.7, will be applied. The integrand is composed ofthe imaginary
part of the screened Coulomb potential, ~ w2+(D~~+47r0")2 = ~Im C(q~w))'
multiplied by the appropriate matrix-element squared, related to the "diffusion pole", Re [iW+~q2 ], in diffusive systems. Here D is the appropriate
diffusion coefficient, see below. eigenstates. Thus, eq.2 is the lowest order
contribution to the imaginary part of the electron's self energy, where the
perturbation is the complex potential q2~(:,~). Similar remarks apply to a
related calculation by Giuliani and Quinn[22] for 2D ballistic systems.
, was recently calculated by Sivan et al[9] for a single compact, isolated,
quantum dot of volume L3 in the diffusive metallic regime, L ki/,
where is the elastic transport mean free path. The dot was taken to
have a negligible coupling to outside particle reservoirs, hence a constant
number of electrons, as deemed appropriate for the experimental situation
in Ref[10]. quantum A dynamically screened Coulomb interaction was used,
as in eq.6 below. Under ordinary metallic circumstances, which we shall
qTF, where kF is the Fermi wavenumber and
assume for simplicity, kF
qTF the Thomas- Fermi one. The result in the dirty (/L;S I/JkFL) zerodimensional (d = 0) regime can be written as:
f'J
(3)
where D..L is the level spacing of the quantum dot and Ec = D / L2 its Thouless energy. An important ingredient in that calculation was the inclusion
in the summation over q (cf. eq. 2) only q's that are larger than 1/ L. This is
valid for a perfectly isolated quantum dot. When the coupling of the latter
87
to the outside world is introduced, that issue must be reconsidered. We
shall return to this later.
One of the main physical results of Ref.[9] was, from eq.3, that the
single-particle level width becomes comparable to the level separation at
the Thouless energy, Ec. Thus, individual single-quasiparticle levels cannot
be resolved for E 2: Ec. This fundamental and quite universal result was
motivated by and agrees with the experimental findings of Ref.[lO].
Following ref. [11] we are going to consider now a large system of metallic
quantum dots weakly coupled by small hopping matrix elements so that the
states are localized within the individual dots and the localization length
of the coupled system is comparable to the dot size, L. This model applies
in principle also to a homogenous bulk weak Anderson insulator. By that
we mean that L is taken to be the localization length, ~, assumed to satisfy
kF~ 1 so that the localization volume plays the role of the quantum
dot. We shall present a physically motivated but still tentative evaluation
of Tee, resulting in a new linear temperature dependence for I/Tee in this
regime. For metallic quantum dots this result is relevant at least in the
whole temperature range
(4)
The dimensionless conductance of the metallic quantum dot, 9 "" Eel flL,
is much larger than unity (this assumption is strictly valid for the granular
case only, for the homogenous Anderson insulator 9L "" 1). The quantities
f, D, Ec "" D / L2 and the conductivity, 0', pertain to the dot's material.
The macroscopic conductivity, O'M, and diffusion constant, DM, are much
smaller. For temperatures below 6. L , the interdot transport will be exponenially activated[12]. Using the Thouless idea[12] (see also ref.[13] for the
ID case and ref. [14]) , for conduction in this regime being determined by
the inelastic scattering, we predict anew, linear, temperature dependence
of the inelastic rate and the conductivity, in the above regime.
According to Thouless[12] and Gogolin et al.[13, 14], O'M is given by:
(5)
where v is the single-particle density of states (strictly speaking, v should
be replaced by the thermodynamic quantity an/aM, but a more complete
treatment in ref. [11] including this, produces similar final results. Eq.5 uses
the Einstein relation and the fact that the macroscopic diffusion constant
is given by L2/Tee.
Our task is to evaluate eq.2 for an inhomogenous system, where the
conductivity is 0' on scales smaller than Land O'M on scales larger than
L and the diffusion constant behaves similarly. Alternatively, we shall use
88
these two values of 0" for q larger and smaller than 1/ L respectively, and
the integrals will turn out to be dominated by q "" 1/ L. The small-scale
contribution, inside the dot, is the one evaluated in ref. [9] (i. e. eq.3 above).
Thus, only the large scale (small q, below 1/ L) contribution remains to
be calculated. For the screened Coulomb interaction, we note from the
Einstein relation that D M/ L2 0" M, as long as qTF L 1, and 0" M W
as long as W I/Tee . The latter is the conventional condition of validity
(see eq.l) of the Landau Fermi-liquid picture, which we shall assume and
self-consistently confirm from the final results. Thus, the imaginary part of
the dynamically screened Coulomb interaction, ~ 1m C(q~w))' in eq.2 will
be given by
W
41l'q 20" ,
(6)
in all cases relevant for our purposes. We emphasize that the ratio between
Dq2 and w is immaterial for this screening.
We shall consider the "thermal" case, E kBT. The most important
modification in the large scale regime is that the Thouless relationship for
DM and O"M is valid only for frequencies lower than the inverse hopping
time I/Tee . Thus, the effective cutoff in the frequency integration in eq.2 is:
Wmax
= I/Tee .
(7)
We now address the estimation of the matrix elements squared in eq.2.

Treating the intergrain tunneling, v, perturbatively and noting that v predominantly couples a level in one grain to the one closest to it in the nearest
neighbor grain, they are given by (v/I::1)2(qL)2, times the intragrain matrix
element squared. The latter is a Lorentzian with a width, E e , much larger
than w max . The small q contribution is then approximately given by
1
small q
Tee
= 2e 2
rO(l/L) d3qL2 kBT (v/l::1) 2

lna
41l'O"M T ee E e
The self-consistent equation for
Tee
(8)
is solved by,
(9)
In the last approximate equality we used eq. 5 and vL 3 = 1/I::1 L , where
I::1L is the level-spacing on scale 1. We note that the integration over q is
dominated by q "" 1/ L. This will not be the case for lower dimensions of the
underlying space, but the form of the final result is unchanged. We also note
that the W integration is dominated by W "" Wmax "" DM / L2 = I/Tee . This
is in contrast to the application of Eq. (2) to an infinite uniform system,
89
where the dominant momentum transfer is q '" JTlTiD, the dominant

energy transfer is T and the resulting T~l is proportional to T 3/ 2 . It is
also in contrast to the application of (2) to a small grain (Eq. (3)), where
the dominant momentum transfer is II L, the dominant energy transfer is
kBT, and the resulting Te~l is proportional to T2. We emphasize that the
scattering rate we obtain is much smaller than kBT, as required for the
validity of Fermi liquid theory. This is due to both Ec being much larger
than !:l.L and (v I!:l.d being small.
Even though this result for I/Tee due to large scales can be much smaller
than the one for the isolated dot and small scales, it is the one relevant for
interdot electron transfer and hopping conductivity. In fact, substituting
Eq. (9) in Eq. (5) we then find that within that range of temperatures, the
conductivity of the granular system is
aM
an !:l.L
~ e OM Ec (vl!:l.d kBT.
(10)
This is the central new result of this section - the conductivity of a granular
system follows, within a certain range of temperatures, a linear temperature dependence. Its generality might be, at least at first glance, surprising,
since it does not depend, for example, on the dimensionality of the granular
system. The source of the linear temperature dependence is the combination of Eq. (5), stating that aM '" Te~l, with Eq. (2), stating (within the
approximations used) that T~l '" cr;;Tee Eq. (5) is a consequence of the
system being an Anderson insulator at zero temperature. The T and aM dependences of Eq. (2) are direct consequences of the fluctuation-dissipation
theorem, once we assume that the dominant momentum transfer, II L, is
smaller than qTP, and that the dominant energy transfer, I/Tee, is much
smaller than the conductivity aM. Both assumptions are satisfied if a single
grain charging energy e 2 I L is much larger than !:l.. Our approach originates
from Fermi liquid theory, that requires T T~l(T), which is justified by
Eq. (9).
Our prediction for a linear dependence of the conductivity on temperature can be tested both on self-assembled granular systems and on arrays
of fabricated quantum dots. We mention that such "soft insulator behavior" - the conductivity increasing linearly with T - was in fact observed
in ref.[24]. On the other hand, even softer, sub-linear dependence of the
conductivity on temperature was observed in the past, for example, near
the metal-insulator transition[25]' in disconnected granular lead films[26]
and in high Tc oxides[27, 28] (where superconductivity was quenched by
high magnetic fields). The present work is not able to explain these latter
experiments.
The issue of the coupling of the quantum-dot with the environment was
recently discussed by Kamenev and Gefen[23]. This is related to the k = 0
90
mode, which does not appear in the isolated dot case[9]. Ref. [23] used an
appropriately chosen R-C circuit to model this coupling (which vanishes in
the isolated-dot limit, RC -+ 00). An interesting inhomogenous broadening
was found for the single quasiparticle levels. This broadening is due to the
time-dependent fluctuations of the "floor" of the potential energy of the dot,
because of thermal fluctuations in the above circuit. Thus, this is different
from the level broadening due to decay, discussed here. For example, the
long-time average of the inhomogenous broadening vanishes, while there is
no way to get rid of the decay broadening.
In summary, we considered in this section the temperature dependence
of the conductivity in a granular system in which single-electron states are
localized over individual grains. We found that while for temperatures T
lower than the single grain effective energy level spacing the conductivity is
exponentially small, for higher T it is proportional to T for a wide range of
temperatures. The division of the contributions to tau ee to different scales
can be extremely useful. This will be used in subsequent work.
3. Coupling of a single-particle state to the quasicontinuum, the

crossover between decay and oscillation
The result in eq.3 at T = 0 can be qualitatively understood as follows:
We calculate the golden-rule rate of a quasiparticle with energy E to scatter
with another particle below the Fermi enrgy, into two quasiparticles. The
DOS of the created electron-hole pair excitation with energy E' ::; E is E' / ilr
where ill is the single-particle level spacing, denoted before by ilL. By the
energy conservation the energy of the second quasiparticle created is E - E'.
Thus, the total final DOS for the above process is:
(11)
The matrix element for this process, v, is seen to be given by il1/g. theory of the This resu1t[4, 9, 23, 29, 16] is the matrix element for processes
dominated by small wavenumber or energy exchanged, where regime) the
wavefunctions are correlated by diffusion. For larger energy transfers, a
different result should hold, as mentioned in section 5.
It is well-known that a true decay can strictly occur only into a continuum of final states. However even the exact many-body states of a finite
quantum dot are, in principle, discrete. Thus, we have to face the fundamental question of how one can talk about a decay process when the set of
final states is discrete (albeit rather dense). To this end[20], we discuss some
elementary, but not widely appreciated, properties of a state, la), coupled
by matrix elements of order v to a dense discrete set of states, Icy), with
91
a typical separation fl, which do not interact among themselves [15, 30].
This model can be solved essentially exactly. Discussions equivalent to the
one below have been presented in the theory of the single-particle strength
function in nuclear physics[15]. We will review these discussions and present
some hopefully new results based on them, towards the end of this section,
but we first start with the perturbative treatment.
It is important to realize that a profound change occurs in the physics
of any two coupled systems upon crossing-over between the above regimes
(v fl and v fl, which are equivalent to r fl and r fl, respectively). For v fl, the weak admixtures between the initial state and the
quasicontinuum are describable by the usual discrete-states perturbation
theory. On the other hand, for v fl, such a treatment becomes useless
and it is instructive to revert to time-dependent perturbation theory. Here
the situation is as follows, there are three relevant energy scales:
The microscopic energy, which we denote by 11,ITv, over which v retains
its order of magnitude. This energy can often be taken to be of the order
of the bandwidth, V.
The Thouless-type energy, Ee rv v 2jfJ rv 11,ITe (where, for v fl, liTe is
the golden-rule rate of decay from the state la) into the final states).
The level spacing, fl rv 11,ITH of the final states. TH is referred to, in some
circumstances, as the Heisenberg time.
In the case of golden-rule decay (v fl, or 11,1 Te fl), the orders of magnitude in terms of the times associated with the above energy scales are
taken to satisfy:
(12)
Let us remind ourselves of the theory of the decay. At the time t = 0 the
system is prepared in the state la). For times in the range Tv t T e ,
the states whose energy differences, 11,w, from the initial state are much
smaller than lilt have each a population of the order of v 2t 2lli 2. The states
with energy separation from the initial state that are much larger than lilt,
have populations oscillating with time with amplitudes of order (v lliw)2,
and hence long-time averages of the same order. (This decrease like w 2 is
related to the" Breit-Wigner" type structure [15, 32, 33]). Thus, the initial
state basically decays into a band of width lilt, containing
li I tfl
states.
(13)
The total depletion of the initial state is on the order of tlTe, as it should
for Te to be the decay time. It should be noted that the depletion caused
by the states with w 2: lit, is of the same order of magnitude as the total.
For TH t Te, the depletion of the initial state becomes important and
the decay becomes the well-known exponential. For times longer than TH,
92
where the number of populated states (eq.13) above would become smaller
than unity, the situation becomes more complicated. Generalized Rabitype oscillations occur and the above golden-rule treatment is not strictly
valid. However, in the regime TH t Te , most of the decay of the initial
state has already occurred and the long-time details, while very interesting,
should practically not change the essential behavior. 3 This is easy to see
using the equivalent approach to the problem from the lineshape point of
view, where a "strength function" measures the weights of the exact states
of the combined system, in the initial-state wavepacket[15]. The linewidth
is basically hlTe, and the long-time phenomena (on scales;::: TH) determine
the fine structure of the line (on scales ;S 8 hITe). This will be discussed
further at the end of this section.
Let us now consider the special case Te TH. Here, at t Te , the above
treatment is roughly valid, and at longer times, the populations oscillate,
where the dominant term has a frequency equal to the smallest separation
between the initial level and the energies of the final states. The finite
depletion changes to an oscillation for times around TH.
Finally, we observe that for:
f'V
f'V
(14)
the oscillations start before the population has significantly decayed, and
one is back in the range of validity of discrete-states perturbation theory.
Thus, we see that when
(15)
the crossover between the two behaviors occurs.
Very recently, the more general hierarchy of the coupling by the interactions of the single-particle excitation, to states with higher and higher
numbers of quasiparticles, was considered in Ref. [16] (see also Ref.[23]). The
extremely interesting idea of delocalization in Hilbert space (in the sense of
the exact many-body states becoming strong mixtures of all quasiparticlenumber states) was introduced in that work[16]. The energy scale characterizing the above process was found in Ref[16] to be:
*
Note that Ee
J 1:l1Ee
(16)
1:l 1 .
3To eliminate this subtle t ~ TH Tc regime, the final states, Io:}, can be taken, in
turn, to be weakly coupled to some much larger environment. Once the level broadening
due to this additional coupling is comparable to 8, the states 10:) become in effect a
continuum and full decay is established (e.g. Ref.[30]). We shall not directly employ this
device, and stay in the t ;S TH regime, being aware of the fact that at long times the
population of the initial state does not fully decay.
93
This energy has an extremely simple interpretation from the point of

view of the present paper. It is just the energy at which from Eqs.3 (with
~ = ~l) and 11, the golden-rule decay rate becomes comparable to the
energy spacing of the final states:
(17)
The energy at which the above occurs is, in fact, just the energy E* mentioned (eq.16) above!
Physically, one has a hierarchial strucure of levels of n quasiparticles
coupled to the denser levels of n + 2 quasiparticles, etc.. Eventually, the
multiparticle level spacing is so small that an incredibly weak coupling to
the rest of the world will suffice to give a practically continuous spectrum
with an irreversible decay of the initial state ..
Thus, E* is the energy of the crossover (eq.15) into the regime where
the golden-rule decay (eq.3) is practically valid, and the single-particle
states really develop a width[20, 29]. This energy scale was not mentioned
in Ref. [9]. However, the main result of that work, that the single-particle
states become a continuum at E e, obviously remains true (since Ee E*).
We remark that from the discussion of the present paper, a weak enough
coupling to the outside world is not very relevant for the gross features discussed here. LFrom the present discussion (see also Ref.[30]), one is led to
expect that as long as this coupling causes level broadenings much smaller
than 8 and h / Tee (E), it will only affect features of the lineshapes on these
small scales. This is expected to be true also for further stages in the hierarchy as well. coupled to the quasicontinuum of the latter, due to coupling
of 'Y. 'Y separation of the many-body energies, especially ~l small scales.
We now review the" strength function" treatment and present some further consequences of that model. The model is again of a state, la), coupled
by matrix clements of order v to a dense discrete set of states, la), with
a typical separation 8, which do not interact among themselves[15]. The
phases of the latter can be chosen to make the v's positive. The initial energy of la) is within the spectrum of the final states. This problem is exactly
solved formally and, as already proven by Rayleigh, in a related vibrational
problem, the energies of the exact resulting states, Ii), are straddled by
those of the unperturbed spectrum. The amplitude of the exact state, Ii),
in the initial state, la), is given by (ila) = [1 +~a (Ei~~a)d-l/2. The strength
function is defined as Pa(i) = l{ila)12/8. For the particular case of uniform
v and 8, the strength function can be given in closed form as
(18)
94
where r = 21fV2 16 = E e , and we took r 6. The assumptions of uniform
v and 6 can be dropped if appropriate averaging over the final spectrum is
performed[15]. If the system is prepared at t = 0 in the initial state la), the
probability that it will remain in that state at time tis:
(19)
Analyzing eqs.18 and 19, one arrives at the following conclusions:
1. For short times, the population of the initial state, Pa , decays exponentially with a decay constant r In.
2. Without further coupling of the states Ii) to a larger reservoir the population Pa will have a small long-time average of order 6/r, and extremely
interesting oscillations.
3. Looking at the above time-dependence with a good enough time resolution, will give a measurement of the "spectral form factor" (the Fourier
transform of the level density correlation function), for energy scales much
smaller than r.
In the energy domain, the above observations simply mean that the
state la) is a wavepacket in the exact states Ii), with weights, given by
the strength function, that form approximately a Lorentzian with width r,
around the initial energy. The fine structure of this Lorentzian, on scale 6,
has the detailed information on the exact spectrum. In the case of interest
to us here, the state la) is a single-particle excited state of the system. It can
be thought to provide the dominant coupling to most experimental probes,
such as single-electron tunneling and electromagnetic absorption. Thus,
such experiments usually measure a lineshape given by the above strength
function. If further coupling of the final states to a larger reservoir[30] is
assumed, giving them a width larger than 6, a continuous Lorentzian and
a strict decay will follow. In practice this may often occur via hierachical
couplings to larger and larger systems, as in the example mentioned above.
In the next section we apply similar considerations to transport in the
scaling theory of localization.
4. The Thouless Scaling Picture with Interactions

We recall the Thouless [12] block picture: The macroscopic system is mentally divided to blocks of linear size L. The typical level separation in each
block is .6., and levels in neighboring blocks are coupled by matrix elements
of typical absolute value v. It is obvious that the system's wavefunctions
are localized (extended) on scale L for v .6. (v .6.). The ratio ~ rv ~
[17, 18] gives the dimensionless conductance, g, on scale L, at least as long
as 9 2: 1. We emphasize that the discussion below is for finite, perhaps large,
L. It may form the basis for extrapolations of varying degrees of correctness
95
to the L ~ 00 limit. But here we discuss only finite-L crossover phenomena,

and not directly the possibilty of a phase transition at the L ~ 00 limit.
This picture was originally developed for independent electrons (where
.6. was the single-particle level spacing, .6. L , at the energy of interest) and
it has been the basis for the modern scaling theory of localization [34].
One may quite easily be convinced, however, that this description is really
very general and that it holds as a measure of the localization of the states
whatever the levels in the blocks are. Only in the range where the decay
applies, Ee 2:.6., where Ee rv ~, will there exist interblock transport. The
conductance on the scale of the block size is then given [17] by the Thouless number, Ec/.6.. This can also be seen from the fact that the diffusion
constant on the scale of the block-size, L, is of the order of L2/Te where
n/Te =
Ee
lt is obvious that all the above holds whatever the two blocks and their
levels are. 4 The "blocks" can be large molecules, atomic clusters, or any
relevant object. The levels need not be single-particle ones, an example
might be the translational motion of a molecular species made from the
constituent particles under discussion. The identification of the ratio
with the conductance can be demonstrated if these levels characterize some
entity of charge q (this charge would then be reckoned in the conductance
unit and in the flux quantum). The diffusion constant of that entity on scale
L will be given, as above, by L2 times the rate of transitions, l/Te between
two such blocks [17]. This includes the diffusion constant of the molecular species in the above example. If the entity is charged, the appropriate
Einstein relation will yield its conductivity, thence the conductance. In the
next section we take a pair of electrons as the entity under consideration.
fr
5. Two-electron coherent delocalization

A mechanism for enhancement of the coherent delocalization by electronelectron interactions has recently been suggested theoretically[21, 19]. Section 2 employs a related mechanism to incoherent conduction. The fact that
(attractive or repulsive) electron-electron interactions enhance the conductivity in insulators is quite straightforward and appears to be rather general.
We review first the case of two electrons in an empty band.
4Note that we are considering here only the case where the two coupled systems
have a level separation of the same order of magnitude. The asymmetric case, where the
densities of states of the coupled systems are vastly different, will be discussed in more
detail elsewhere. When the level separations are ,6, and ,6,', the conductance is given by
~ v 2 j(,6,,6,')[18]. For ,6, ,6,', the condition for an effective decay of the states of the
small system into those of the large system is v 2 ;::: ,6.'2. This has a number of interesting
ramifications. Some of those are mentioned in section 3.
96
Recently,[21] Shepelyansky considered strong short-range interactions
between two electrons in a one dimensional (ID) random potential having
a one-particle localization length ~ = 6. He found that the interactions can
result in a larger localization length, 6, for certain two-electron states. An
Anderson model with a lattice constant a, transfer matrix element V and
disorder range W, and a Hubbard-type on-site interaction U, was used. U is
initially taken to satisfy I U I ;S V. By establishing a correspondence of the
above problem with a certain class of banded random matrices and numerical computations on the latter (and on several other models), Shepelyansky
obtained:
(20)
This is an approximate expression, obtained by fitting numerical results.
The ratio? can be arbitrarily large for weak enough disorder and it is
independent of the sign of the interaction.
This result is of interest, since it has been believed that strong repulsive
interactions always tend to strengthen localization. The present author [19]
gave a physical argument, based on the generalization of the Thouless [12]
block-scaling picture of section 4, substantiating the above result. This scaling picture, discussed below, was powerful enough to be straightforwardly
generalizable to an arbitrary dimensionality, d, and to suggest a useful picture for two quasiparticles in many-Fermion systems. Further support for
this generalization is provided by a recent formulation of the interacting
two-electron problem in terms of an effective nonlinear lJ-model[35].
For the attractive interaction case, these results in ID are in agreement
with an extremely interesting previous study by Dorokhov [36] of the propagation of a weakly bound pair in a disordered ID potential. The qualitative
result that interactions of either sign can cause delocalization has been by
now confirmed by a number of studies [37, 38, 39, 32, 40, 33, 41, 42, 43, 44,
?]. A particularly interesting aspect is the h/2e period of the AB-flux depenendence [38], in a ring geometry, of the special delocalized two-particle
states. It should be noted that the exponent for the ~ dependence appears
numerically not to be exactly equal to 2 in ID, but roughly 20% smaller. A
problem also exists with regard to the U dependence, which is still under
active study [39, 45]. Much of the cited literature has considered shortrange interactions, often just the simple Hubbard interaction. One should
be aware that the results may well be different for the physical case with
Coulomb interactions [48].
We start the scaling discussion by considering the noninteracting twoelectron states in the blocks, on scale L = where is the single-electron
e,
97
localization length. 5 The single-electron conductance is then of order unity
and the interblock tunneling matrix elements can be ignored to lowest approximation. We denote the set of all single-electron orbitals on a block, including disorder, by n(x). The noninteracting two-electron states, 'lI nm =
n(Xdm(X2), are a complete basis set. The number of sites in a block,
(Lja)d (or Nc~ja, for a quasi-1D wire with Nc channels), is denoted by
N. The interaction matrix element for scattering between 'lI nm and 'lin' m' ,
denoted by U2, is of a typical6 absolute value[21]
(21)
This follows since one has to sum N U's, each multiplied by a product of
four 's, the latter being each of order N- 1/ 2 and are assumed to have
random phases. Next, we consider the matrix element for the scattering of
the pair between neighboring blocks. Here, of the four orbitals, two will be
on one block and two on the neighboring one. The wavefunction product
will therefore be governed by the spread of each localized wavefunction
into the nearest neighbor block. Thus the result will be of the same order
of magnitude but somewhat smaller than the intrablock one. We denote it
by AU2, where A == exp( -a), and a is expected to be of the order of but
somewhat larger than unity. 7
To apply block scaling to the present problem, we notice that the typical
separation of two neighboring two-electron states on a block, not-too-close
to the band edge, is of the order of:
Ll2
rv
B
N2.
(22)
B being the bandwidth, of order V for the case of interest W ;S V, U;S V.

To get the effect considered, we need ~ to be large enough compared with
5This discussion works in an arbitrary dimensionality. In fact, the estimates used are
at their worst in strictly ID systems that lack a proper diffusive regime.
6In special cases mentioned in section 3, where low-energy processes are important,
correlations of the states due to quantum diffusion come into play and the matrix elements
are changed. This is irrelevant in the case discussed here where the energies are high.
It may be important for the two-quasiparticle case discussed later. Important remarks
on the matrix elements, mainly in ID, have been made in ref.[31]. These do not seem
however to substantially affect the final results.
7The intrablock interaction mixes strongly the states wnm within a given block. One
might worry that these interactions could modify the interblock matrix elements. However
it is straightforward to check that the order of magnitude of the latter does not, in fact,
typically change. In the simplest case, let us assume that Nl wnm'S are thoroughly mixed
by the intrablock interaction. In an eigenstate of a given block, this implies that each of
these Wnm's will appear with an amplitude'" If,fJli; and a random phase. Thus, the
interblock U2 will be modified by a sum of Nf random terms, each of order 1fNl, which
is of order unity.
98
the microscopic length a. This necessitates W V at d=1,2 and W /V not

too far above the critical value for delocalization, at d=3. We take a large
enough N so that ~ 1. Thus U2 will thoroughly mix the two-particle
states around the given energy on a given block. This can be easily seen,
as discussed in footnote 7, to not modify drastically either the interblock
interaction-dominated matrix element AU2 (see footnote 6) or the level
separation L12. Thus the interblock dimensionless two-particle conductance
IS
(23)
92 on scale can be much larger than unity, even though for indepemdent
electrons, 9 rv 1 on the same scale!
At high d (three dimensions) this implies that the two-electron (correlated)
states can even be de localized when the single-electron ones are localized.
For d =1,2, this just means that the localization length for the relevant
two-electron states will be much larger than that of the single-electron
ones. The effect at 2D is exponentially large. For a quasi 1D wire, we find
6 by considering by how much L has to be increased to get 92 rv 1 and
using eq. (23) and N = Nee/a. This yields, in agreement with [21J:
(24)
Here 6 is the localization length for the wire. In the single-channel case,
Ne = 1 and for U;S V, this reduces to Shepelyansky's result (eq.20). Obviously we are not able to find the numerical coefficient from our general
discussion. The constant in eq.20 will correspond to A '" .17, which appears
resonable.
Physically, the delocalized entity is a pair of electrons that are within 6
of each other. The" center of mass" of this pair has a wavefunction which
is extended over a length 6. This is only a rough way of describing what
happens, since the problem does not separate into relative and center of
mass coordinates, as the disorder breaks the translational symmetry. The
correct quantum-mechanical way vf describing these states is that they are
coherent mixtures of products of two single-particle states that are within 6
of one another. The coefficients in this superposition (" entangled state") are
such that the pair, as a whole, extends over the range 6. A very unusual
feature of this picture is that each transfer of the pair by a localization
length occurs via a "scattering" process due to the interaction. The kinetic
energy, represented here by V, only plays a role in determining 6, which
has to be large enough.
It should be emphasized that we only demonstrated that there exist
correlated two-electron delocalized states, in which the two electrons are
within 6 of each other. Obviously, there exist many more, roughly by a
99
factor of order (L / ~) d, uncorrelated localized two-electron states where the
two electrons are at least several 6 's away from each other and are therefore not sensitive to the interaction. However, in the interesting case, where
6 L ~ 6, the latter states should make an exponentially small contribution to physical properties that necessitate delocalization. Among those
are the transport coefficients and, for example, the persistent currents in
a ring geometry. The increasing dilution of the interesting states when the
system-size increases means that more care is needed to find them in calculations that look at the whole spectrum. This may well be the reason for
the results announced in ref[46] discussing the thermodynamic limit based
on numerical calculations. This reference is entitled "No enhancement of
the localization length for two interacting particles in a random potential" .
The above obvious interpretation of this, along with more technical issues
is discussed in the rebuttal comment by Frahm et al[47]. It is important
to emphasize that the belief that the size of localized states may shrink
when the system size increases much beyond the localization length is obviously against the basic idea that localization is related to insensitivity to
boundary conditions.
The above consideration was for just two particles in a random potential in a large volume. It is qualitatively valid regardless of the statistics of
these particles[21]. It is of great interest to find out to what extent these
considerations also apply to degenerate (finite density, low temperature)
many-Fermion systems. This would appear possible since the interaction
changes the states of two particles only, in the noninteracting basis. However, this issue is far from being obvious. On a rather naive and tentative level, one may consider exact many body states with, say, Nand N'
electrons on neighboring blocks. The coupling via the interaction to other
states, with e.g. N - 2 and N' + 2 electrons on these blocks, will be U2, as
above (eq.(21)). The separation between states differing by the excitation
of two particles on the same block, whose smallness (being on the order
of ~2) was the crucial point in the above consideration, is actually of the
order of ~1, the single-particle spacing, for the lowest-lying states. It then
increases quickly with excitation energy above the ground state and for a
total excitation energy E > > ~1' it becomes: 8
(25)
This can be seen to become similar to eq.(22) for E rv V. ~2(E) becomes
8To obtain eq.25 more formally, one may note that the two-particle density of states
(DOS) is given by the convolution of the single-particle DOS with itself.
100
small enough to cause delocalization as above, for excitation energies of
2
m
rv
B N!/2!::..
U
rv
B2 (Em! )vd/2
U V
'
(26)
to get the last approximate equality we used N = (6/a)d rv (E{71 )-vd.

The energy Em2 delineates the energies for which the two-electron states
are all localized and those for which there exist de localized two-electron
states. It thus plays the role of an effective two-electron mobility edge. This
new concept of Em2, should have an important role at least close enough
to the usual, one-electron mobilty edge, Em!. In fact, Em2 becomes much
closer to the ground state than Em!, when the ordinary delocalization transition is approached (i.e. when Em! -+ 0, generalizing the noninteracting
electrons condition Em! -+ EF) This is easy to see, employing the wellknown [50] fact that the critical exponent v around the one-electron mobility edge satisfies v :2: 2/d. Therefore, when 6 increases, Em2 approaches
the Fermi energy faster than the single-electron mobility edge. A further
issue that should be discussed here is the possible decay of such extended
two-particle states to lower-energy excitations. Such a decay, iffast enough,
would make the observaton of physical effects associated with these delocalized pairs more difficult.
Both the interaction U2 and the small level separation (eq. (22)) apply to any quasiparticle pair, obviously including an electron-hole one. We
thus reach the important conclusion that the pair de localization as discussed
above, applies to electron-hole excitations as well. Such an excitation consisting of an electron and a hole moving together does not carry charge but
it carries entropy and will thus directly contribute to thermal transport.
We believe that the naive argument above for the two-electron delocalization edge provides at least a sufficient condition for the delocalization
of two-quasiparticle states in a Fermi system. The possibility of existence
of two-quasiparticle extended states at a different excitation-energy in a
degenerate system is thus an interesting, but speculative, issue. The abundance of two-quasiparticle states whose density strongly increases with E,
and the fact that such states are connected to the ground state with matrix
elements of order U2 , implies that the second-order interaction correction
to, for example, the ground-state energy "does not converge". This means
that energies of order B dominate this contribution. Thus a serious rearrangement of the ground state by the interactions is an interesting issue.
An important aspect is the independence of these effects of the sign of
the interaction. It has to be borne in mind that the state with two electrons
within a localization volume, has an interaction energy on the order of U / N.
This increase or decrease of the energy, depending on whether U is positive
or negative, will play some role as well.
101
Our picture implies that both for weak disorder at d=1,2 and in the
localized regime near the mobility edge at 3D, the delocalization of two
correlated electrons becomes relevant. Such a physical system whose singleelectron states are localized, but some of its two electron states are delocalized, should exhibit an interesting behaviour, especially if the speculation
on the two-electron mobility edge were valid. For example, physical properties which are due only to extended states (such as the dc conductivity,
or quantities that are sensitive to an Aharonov-Bohm phase) might behave
under some conditions as if the charge carrier were a two-electron entity.
An interesting situation is that of A-B oscillations in a system in which the
single-electron states are localized but there exist extended states of two
electrons. The basic A-B oscillation period should then be h/2e, to exponential accuracy for L 6. Evidence for that has recently been obtained
in Ref[38]. Much more work is needed however to find out whether this is
related to the results of ref [51]. The coupling between such a normal conductor and a superconductor, including the SNS case, is another question
of interest. the Fermi level.
While an on-site Hubbard-type interaction has been considered here,
these ideas should apply to other short-range interactions as well. On the
other hand, further effects of the interaction, especially when it is longranged[52J, such as changes in the density of states [48] and the involvement
of magnetism [7] are neglected in this simple picture. Whether this is related
to interaction effects in the metallic regime [2, 4, 8], is a question that needs
a further serious consideration.
6. Summary
In this paper the enhancement of transport in the localized phase was discussed both in the incoherent channel, via inelastic scattering and in the
coherent channel, via two-particle delocalization. For the first problem, the
inelastic electron-electron scattering was evaluated for the "granular insulator", yielding a novel, linear, temperature dependence of the conductivity.
Then the crossover between oscillation and decay was considered, first for
a single state coupled to a quasicontinuum, then for two coupled quasicontinua. The latter situation allows the extension of the Thouless scaling
picture to include interactions. This was used to treat two-particle delocalization, first in an otherwise empty band, and then, on a more speculative
level, for quasiparticles in a Fermi sea. The experimental study of these issues in solid-state quantum dots offers many possibilities that do not exist
for studies of other finite quantum systems. The size, shape and particle
density can be controlled and meaningful magnetic fields can be applied.
A large variety of materials exists as well. It may therefore be hoped that
102
these systems, will shed more light on the interesting and nontrivial physics
of strongly interacting quantum systems. The question of the coupling the
pairs discussed here with Cooper pairs in, for example, adjacent superconductors, promises to be of much interest as well.
Acknowledgements: This research was partially supported by the

Israeli Academy of Sciences, by the German-Israeli Research Foundation
(GIF), by the US-Israel Binational Science Foundation and by the Minerva foundation. The work reported in section 2 started when the author
was participating in the program on quantum chaos in mesoscopic systems
at the Institute for Theoretical Physics, University of California, Santa Barbara, supported in part by the National Science Foundation under Grant
No. PHY94-07194. The author is grateful to his colleagues at the Lorentz
Institute and the Kammerlingh Onnes Laboratory, Universityof Leiden, for
their hospitality when the work was concluded and the paper written. The
author is indebted to U. Sivan and A. Stern for collaboration on the work
reported in section 2 and to Y. Blanter, G. S. Boebinger, E. Brezin, V. F.
Gantmakher, Y. Gefen, A. Gerber, T. Geszti, B.I. Halperin, A. Kamenev,
D. E. Khmelnitsky, A. Miiller-Groeling, P. Nozieres, F. von Oppen, Z.
ovadyahu , J.L Pichard, M. Pollak, H. Weidenmiiller, D. Weinmann and
P. Wolfie for instructive discussions. publication.
References
1.
2.
3.
4.
A. Schmid (1974) Z. Phys. 271, 251.

B. L. Altshuler and A. G. Aronov, SOy. Phys. JETP, 30, 968 (1980).
B.L.Altshuler, A.G. Aronov and D.E. Khmelnitskii J. Phys. CI5, 7367 (1982).
For a general reference see: B.L. Altshuler and A.G. Aronov, in: Electron-Electron
Interactions in Disordered Systems, A.L. Efros and M. Pollak, eds., North Holland
(1985).
5. A. Stern, Y. Aharonov. and Y. Imry. in Quantum Coherence in Mesoscopic Systems,
B.Kramer, ed, Plenum (1990), p. 99.
6. S. Chakravarty and A. Schmid (1986) Phys. Rep. 140, 193.
7. A. Finkelstein, SOy. Sci. Rev. A, Phys. 14, 1 (1990); J. de Physique Colloque C8,
Suppl n12 49, 1173 (1988).
8. A. Miiller-Groeling, H. A. Weidenmiiller and C. H. Lewenkopf, Europhys. Lett. 22,
193 (1993).
9. U. Sivan, Y. Imry and A. G. Aronov, Europhys. Lett., 28, 115 (1994).
10. U. Sivan, F.P. Milliken, K. Milkove, S. Rishton, Y. Lee, J.M. Hong, V. Boegli, D.
Kern and M. deFranza, Europhys. Lett. 25 605 (1994).
11. Y. Imry, U. Sivan and A. Stern, to be published.
12. D. C. Thouless, Phys. Rev. Lett. 39, 1167 (1977).
13. A. A. Gogolin, V. I. Mel'nikov and E.I. Rashba, SOy. Phys. JETP 42, 168 (1975)
14. A. A. Gogolin, Phys. Rep. 86, 2 (1982)
15. A. Bohr and B.R. Mottleson, Nuclear Structure, Benjamin, New York (1969). See
vol. I, chapter 2 and especially appendix 2-D.
16. B.L. Altshuler, Y. Gefen, A. Kamenev and L. Levitov, Phys. Rev. Lett 78, 2803
(1997).
17. Y. Imry, Phys. Rev. B21, 2042 (1980).
103
18.
19.
20.
J. Bardeen, Phys. Rev. Lett. 6, 57, (1961).

Y Imry, Europhysics Lett. 30, 405 (1995).
Ylmry, in the proceedings of the 1996 Rencontres de Moriond, Correlated Fermions
and Transport in Mesoscopic Systems, T. Martin, G. Montambaux and J. Tran Thanh
Van, Editions Frontieres 1996, p.211.
21. D. L. Shepelyansky, Phys. Rev. Lett. 73, 2607 (1994).
22. G.F. Giuliani and J.J Quinn (1982) Phys. Rev. B26, 4421.
23. A. Kamenev and Y Gefen, in the proceedings of the 1996 Rencontres de Moriond,
as in ref[20], p.503.
24. Z. Ovadyahu and Y. Imry, J. Phys. C18, L19 (1985).
25. V.F. Gantmakher, M.V. Golubkov, J.G.S. Lok and A.K. Geim, JETP 109,1765
(1996).
26. A. Gerber, J. Phys. Condo Mattter 2, 8161 (1990) and unpublished results.
27. Y Ando, G.S. Boebinger, A. Passner, T. Kimura and K. Kishio, Phys. Rev. Lett.,
75, 4662 (1995).
28. Y. Ando, G.S. Boebinger, A. Passner, N. L. Wang, C. Geibel and F. Steglich Phys.
Rev. Lett. 77, 2065 (1996).
29. Y Blanter, Phys. Rev. B54, 12807 (1996)
30. M. Shechter, M.Sc. thesis (unpublished), Weizmann Institute (1995).
31. LV. Ponomarev and P.G. Sylvestrov, preprint cond-mat/9702241.
32. F. Borgonovi and D.L. Shepelyansky, Nonlinearity 8, 877 (1995); J. de Phys. 6, 287
(1996).
33. P. Jacquod and D. L. Shepelyansky, Phys. Rev. Lett. 75, 3501 (1995).
34. E. Abrahams,P. W. Anderson, D. C. Licciardello, and T. V. Ramakrishnan, Phys.
Rev. Lett. 42, 673 (1979).
35. K. Frahm, A. Miiller-Groeling, and J.-L. Pichard, Phys Rev Lett 76, 1509 (1996)
36. O.N. Dorokhov, Sov. Phys. JETP 71(2), 360 (1990).
37. K. Frahm, A. Miiller-Groeling, J.-L. Pichard and D. Weinmann, Europhys. Lett.
31, 405 (1995).
38. D. Weinmann, A. Miiller-Groeling, J.-L. Pichard and K. Frahm, Phys. Rev. Lett.
75, 1598 (1995).
39. F. von Oppen, T. Wettig and J. Miiller, Phys. Rev. Lett. 76, 491 (1996).
appear).
40. F. von Oppen and T. Wettig, Europhys. Lett. 32, 741 (1996).
41. K. Frahm and A. Miiller-Groeling, Europhys. Lett. 32, 385 (1995).
42. Y. V. Fyodorov and A. D. Mirlin, Phys. Rev. B52, R11580 (1996).
43. D.L.Shepelyansky, in the proceedings of the 1996 Rencontres de Moriond, as in
ref[20]' p.201.
44. D.L.Shepelyansky, Physica D 28, 103 (1987).
45. J.-L.Pichard, in the proceedings of the 1996 Rencontres de Moriond, as in ref[20],
p.221.
46. R.A. Romer and M. Schreiber, Phys. Rev. Lett. 78, 515 (1997).
47. K. Frahm, A. Miiller-Groeling, J.-L. Pichard and D. Weinmann, preprint condmat/9702084.
48. M. Pollak, Disc. Faraday Soc. 50, 7 (1970); A. L. Efros and B. L Shklovskii J. Phys.
C8, 49 (1975).
49. J.-L.Pichard, D. Weinmann and Y.Imry, unpublished (1996).
50. A. Brooks Harris, J. Phys. C, Solid State Physics 7,1671 (1974); N. F. Mott, Comm.
Phys. 1, (1976) 203.
51. Ya. B. Poyarkov, V. Ya. Kontarev, L P. Krylov and Yu. V. Sharvin. JETP Lett.
44, 373 (1987).
52. J. Talamantes, M. Pollak and L. Elam, Europhys. Let. 35, 511 (1996).
ELECTRON TRANSPORT IN QUANTUM DOTS.

LEO P. KOUWENHOVEN,l CHARLES M. MARCUS,2
PAUL L. MCEUEN,3 SEIGO TARUCHA,4 ROBERT M.
WESTERVELT,5 AND NED S. WINGREEN 6 (alphabetical order).
1. Department of Applied Physics, Delft University of Technology,
P.O.Box 5046,2600 GA Delft, The Netherlands.
2. Department of Physics, Stanford University, Stanford, CA 94305, USA
3. Department of Physics, University of California and Materials
Science Division, Lawrence Berkeley Laboratory, Berkeley, CA
94720, USA.
4. NIT Basic Research Laboratories, 3-1 Morinosoto Wakamiya, Atsugishi; Kanagawa 243-01, Japan.
5. Division of Applied Sciences and Department of Physics, Harvard
University, Cambridge, Massachusetts 02138, USA.
6. NEC Research Institute, 4 Independence Way, Princeton, NJ 08540,
USA
1. Introduction
The ongoing miniaturization of solid state devices often leads to the question:
"How small can we make resistors, transistors, etc., without changing the way
they work?" The question can be asked a different way, however: "How small
do we have to make devices in order to get fundamentally new properties?" By
"new properties" we particularly mean those that arise from quantum
mechanics or the quantization of charge in units of e; effects that are only
important in small systems such as atoms. "What kind of small electronic
devices do we have in mind?" Any sort of clustering of atoms that can be
connected to source and drain contacts and whose properties can be regulated
with a gate electrode. Practically, the clustering of atoms may be a molecule, a
small grain of metallic atoms, or an electronic device that is made with modem
chip fabrication techniques. It turns out that such seemingly different structures
have quite similar transport properties and that one can explain their physics
within one relatively simple framework. In this paper we investigate the
physics of electron transport through such small systems.
105
106
One type of artificially fabricated device is a quantum dot. Typically,

quantum dots are small regions defined in a semiconductor material with a size
of order 100 nm [1]. Since the first studies in the late eighties, the physics of
quantum dots has been a very active and fruitful research topic. These dots
have proven to be useful systems to study a wide range of physical phenomena.
We discuss here in separate sections the physics of artificial atoms, coupled
quantum systems, quantum chaos, the quantum Hall effect, and time-dependent
quantum mechanics as they are manifested in quantum dots. In recent electron
transport experiments it has been shown that the same physics also occurs in
molecular systems and in small metallic grains. In section 9, we comment on
these other nm-scale devices and discuss possible applications.
The name "dot" suggests an exceedingly small region of space. A
semiconductor quantum dot, however, is made out of roughly a million atoms
with an equivalent number of electrons. Virtually all electrons are tightly bound
to the nuclei of the material, however, and the number of free electrons in the
dot can be very small; between one and a few hundred. The deBroglie
wavelength of these electrons is comparable to the size of the dot, and the
electrons occupy discrete quantum levels (akin to atomic orbitals in atoms) and
have a discrete excitation spectrum. A quantum dot has another characteristic,
usually called the charging energy, which is analogous to the ionization energy
of an atom. This is the energy required to add or remove a single electron from
the dot. Because of the analogies to real atoms, quantum dots are sometimes
referred to as artificial atoms [2]. The atom-like physics of dots is studied not
via their interaction with light, however, but instead by measuring their
transport properties, that is, by their ability to carry an electric current.
Quantum dots are therefore artificial atoms with the intriguing possibility of
attaching current and voltage leads to probe their atomic states.
This chapter reviews many of the main experimental and theoretical results
reported to date on electron transport through semiconductor quantum dots.
We note that other reviews also exist [3]. For theoretical reviews we refer to
Averin and Likharev [4] for detailed transport theory; Ingold and Nazarov [5]
for the theory of metallic and superconducting systems; and Beenakker [6] and
van Houten, Beenakker and Staring [7] for the single electron theory of
quantum dots. Recent reviews focused on quantum dots are found in Refs. 8
and 9. Collections of single electron papers can be found in Refs. 10 and 11.
For reviews in popular science magazines see Refs. 1, 2, 12-15.
The outline of this chapter is as follows. In the remainder of this section
we summarize the conditions for charge and energy quantization effects and we
briefly review the history of quantum dots and describe fabrication and
measurement methods. A simple theory of electron transport through dots is
outlined in section 2. Section 3 presents basic single electron experiments. In
107
section 4 we discuss the physics of multiple dot systems; e.g. dots in series,
dots in parallel, etc. Section 5 describes vertical dots where the regime of very
few electrons (0, 1, 2, 3, etc.) in the dot has been studied. In section 6 we
return to lateral dots and discuss mesoscopic fluctuations in quantum dots.
Section 7 describes the high magnetic field regime where the formation of
Landau levels and many-body effects dominate the physics. What happens in
dots at very short time scales or high frequencies is discussed in section 8.
Finally, applications and future directions are summarized in section 9. We
note that sections 2 and 3 serve as introductions and that the other sections can
be read independently.
1.1. QUANTIZED CHARGE TUNNELING.
In this section we examine the circumstances under which Coulomb charging
effects are important. In other words, we answer the question, "How small and
how cold should a conductor be so that adding or subtracting a single electron
has a measurable effect?" To answer this question, let us consider the
electronic properties of the small conductor depicted in Fig. 1.1(a), which is
coupled to three terminals. Particle exchange can occur with only two of the
terminals, as indicated by the arrows. These source and drain terminals
connect the small conductor to macroscopic current and voltage meters. The
third terminal provides an electrostatic or capacitive coupling and can be used
as a gate electrode. If we first assume that there is no coupling to the source
and drain contacts, then our small conductor acts as an island for electrons.
(b) Vertical
(a) Lateral
SOURCE
Quantum
DRAIN
~Dot~
Quantum
Dot
Figure 1.1. Schematic of a quantum dot, in the shape of a disk, connected to source and drain
contacts by tunnel junctions and to a gate by a capacitor. (a) shows the lateral geometry and (b)
the vertical geometry.
108
The number of electrons on this island is an integer N, i.e. the charge on the
island is quantized and equal to Ne. If we now allow tunneling to the source
and drain electrodes, then the number of electrons N adjusts itself until the
energy of the whole circuit is minimized.
When tunneling occurs, the charge on the island suddenly changes by the
quantized amount e. The associated change in the Coulomb energy is
conveniently expressed in terms of the capacitance C of the island. An extra
charge e changes the electrostatic potential by the charging energy Ec = e2/C.
This charging energy becomes important when it exceeds the thermal energy
kBT. A second requirement is that the barriers are sufficiently opaque such that
the electrons are located either in the source, in the drain, or on the island. This
means that quantum fluctuations in the number N due to tunneling through the
barriers is much less than one over the time scale of the measurement. (This
time scale is roughly the electron charge divided by the current.) This
requirement translates to a lower bound for the tunnel resistances Rt of the
barriers. To see this, consider the typical time to charge or discharge the island
Lit = RtC. The Heisenberg uncertainty relation: L1ELit = (e 2/C)R tC> h implies
that Rt should be much larger than the resistance quantum h/e2 = 25.813 kn in
order for the energy uncertainty to be much smaller than the charging energy.
To summarize, the two conditions for observing effects due to the discrete
nature of charge are [3,4]:
R t hle2
(l.la)
e2/CkBT
(l.1b)
The first criterion can be met by weakly coupling the dot to the source and
drain leads. The second criterion can be met by making the dot small. Recall
that the capacitance of an object scales with its radius R. For a sphere, C =
47rEre,ft, while for a flat disc, C = 8ere,ft, where er is the dielectric constant of
the material surrounding the object.
While the tunneling of a single charge changes the electrostatic energy of
the island by a discrete value, a voltage Vg applied to the gate (with capacitance
Cg) can change the island's electrostatic energy in a continuous manner. In
terms of charge, tunneling changes the island's charge by an integer while the
gate voltage induces an effective continuous charge q = CgVg that represents, in
some sense, the charge that the dot would like to have. This charge is
continuous even on the scale of the elementary charge e. If we sweep Vg the
build up of the induced charge will be compensated in periodic intervals by
tunneling of discrete charges onto the dot. This competition between
continuously induced charge and discrete compensation leads to so-called
109
Coulomb oscillations in a measurement of the current as a function of gate

voltage at a fixed source-drain voltage.
An example of a measurement [16] is shown in Fig. 1.2(a). In the valley
of the oscillations, the number of electrons on the dot is fixed and necessarily
equal to an integer N. In the next valley to the right the number of electrons is
increased to N+ 1. At the crossover between the two stable configurations N
and N+1, a "charge degeneracy" [17] exists where the number can alternate
between N and N+ 1. This allowed fluctuation in the number (i.e. according to
the sequence N ~ N+ 1 ~ N ~ .... ) leads to a current flow and results in the
observed peaks.
An alternative measurement is performed by fixing the gate voltage, but
varying the source-drain voltage Vsd' As shown in Fig. 1.2(b) [18] one observes
in this case a non-linear current-voltage characteristic exhibiting a Coulomb
staircase. A new current step occurs at a threshold voltage (- e2jC) at which an
extra electron is energetically allowed to enter the island. It is seen in Fig.
1.2(b) that the threshold voltage is periodic in gate voltage, in accordance with
the Coulomb oscillations of Fig. 1.2(a).
1.2. ENERGY LEVEL QUANTIZATION.

Electrons residing on the dot occupy quantized energy levels, often denoted as
OD-states. To be able to resolve these levels, the energy level spacing L1E >>
kBT. The level spacing at the Fermi energy EF for a box of size L depends on
0.15
(a)
"~
~
0.1
0.05
-0.5
-0.45
GATE VOLTAGE
-0.4
(V)
-O.51-J.'-'-"---~----'---l
VOLTAGE (mV)
Figure 1.2(a). An example of a measurement of Coulomb oscillations to illustrate the effect of

single electron charges on the macroscopic conductance. The conductance is the ratio IlVsd and
the period in gate voltage Vg is about e/Cg. (From Nagamune et al. [16].) (b) An example of a
measurement of the Coulomb staircase in I- Vsd characteristics. The different curves have an
offset for clarity (I = 0 occurs at Vsd = 0) and are taken for five different gate voltages to illustrate
periodicity in accordance with the oscillations shown in (a). (From Kouwenhoven et al. [18].)
110
the dimensionality. Including spin degeneracy, we have:
L1E=( N 14)llhr2 ImL2

= (1hr )li 27r 2 I mL2
= (1I37r 2N )1I3 li 27r 2 I mL2
1D
(1.2a)
2D
(1.2b)
3D
(1.2c)
The characteristic energy scale is thus Ii)(ImL2. For a 1D box, the level
spacing grows for increasing N, in 2D it is constant, while in 3D it decreases as
N increases. The level spacing of a 100 nm 2D dot is - 0.03 meV, which is
large enough to be observable at dilution refrigerator temperatures of -100 mK
= - 0.0086 meV. Electrons confined at a semiconductor hetero interface can
effectively be two-dimensional. In addition, they have a small effective mass
As a result, dots made in
that further increases the level spacing.
semiconductor heterostructures are true artificial atoms, with both observable
quantized charge states and quantized energy levels. Using 3D metals to form
a dot, one needs to make dots as small as -5 nm in order to observe atom-like
properties. We come back to metallic dots in section 9.
The fact that the quantization of charge and energy can drastically
influence transport through a quantum dot is demonstrated by the Coulomb
oscillations in Fig. 1.2(a) and the Coulomb staircase in Fig. 1.2(b). Although
we have not yet explained these observations in detail (see section 2), we note
that one can obtain spectroscopic information about the charge state and energy
levels of the dot by analyzing the precise shape of the Coulomb oscillations and
the Coulomb staircase. In this way, single electron transport can be used as a
spectroscopic tool.
1.3. HISTORY, FABRICATION, AND MEASUREMENT TECHNIQUES.
Single electron quantization effects are really nothing new. In his famous 1911
experiments, Millikan [19] observed the effects of single electrons on the
falling rate of oil drops. Single electron tunneling was first studied in solids in
1951 by Gorter [20], and later by Giaever and Zeller in 1968 [21], and Lambe
and Jaklevic in 1969 [22]. These pioneering experiments investigated transport
through thin films consisting of small grains. A detailed transport theory was
developed by Kulik and Shekhter in 1975 [23]. Much of our present
understanding of single electron charging effects was already developed in
these early works. However, a drawback was the averaging effect over many
grains and the limited control over device parameters. Rapid progress in device
control was made in the mid 80's when several groups began to fabricate small
systems using nanolithography and thin-film processing.
The new
111
technological control, together with new theoretical predictions by Likharev

[24] and Mullen et al. [25], boosted interest in single electronics and led to the
discovery of many new transport phenomena. The first clear demonstration of
controlled single electron tunneling was performed by Fulton and Dolan in
1987 [26] in an aluminum structure similar to the one in Fig. l(a). They
observed that the macroscopic current through the two junction system was
extremely sensitive to the charge on the gate capacitor. These are the so-called
Coulomb oscillations. This work also demonstrated the usefulness of such a
device as a single-electrometer, i.e. an electrometer capable of measuring
single charges. Since these early experiments there have been many successes
in the field of metallic junctions which are reviewed in other chapters of this
volume.
The advent of the scanning tunneling microscope (STM) [27] has renewed
interest in Coulomb blockade in small grains. STMs can both image the
topography of a surface and measure local current-voltage characteristics on an
atomic distance scale. The charging energy of a grain of size -10 nm can be as
large as 100 meV, so that single electron phenomena occur up to room
temperature in this system [28]. These charging energies are 10 to 100 times
larger than those obtained in artificially fabricated Coulomb blockade devices.
However, it is difficult to fabricate these naturally formed structures in selfdesigned geometries (e.g. with gate electrodes, tunable barriers, etc.). There
have been some recent successes [29,30] which we discuss in section 9.
Effects of quantum confinement on the electronic properties of
semiconductor heterostructures were well known prior to the study of quantum
dots. Growth techniques such as molecular beam epitaxy, allows fabrication of
quantum wells and heterojunctions with energy levels that are quantized along
the growth (z) direction. For proper choice of growth parameters, the electrons
are fully confined in the z-direction (i.e. only the lowest 2D eigenstate is
occupied by electrons). The electron motion is free in the x-y plane. This
forms a two dimensional electron gas (2DEG).
Quantum dots emerge when this growth technology is combined with
electron-beam lithography to produce confinement in all three directions.
Some of the earliest experiments were on GaAs/AIGaAs resonant tunneling
structures etched to form sub-micron pillars. These pillars are called vertical
quantum dots because the current flows along the z-direction [see for example
Fig. 1.1(b)]. Reed et al. [31] found that the I-V characteristics reveal structure
that they attributed to resonant tunneling through quantum states arising from
the lateral confinement.
At the same time as the early studies on vertical structures, gated AIGaAs
devices were being developed in which the transport is entirely in the plane of
the 2DEG [see Fig. 1.1(a)]. The starting point for these devices is a 2DEG at
112
the interface of a GaAs/AIGaAs heterostructure. The only mobile electrons at

low temperature are confined at the GaAs/AIGaAs interface, which is typically
- 100 nm below the surface. Typical values of the 2D electron density are ns (1 - 5).1015 m- 2 To define the small device, metallic gates are patterned on the
surface of the wafer using electron beam lithography [32]. Gate features as
small as 50 nm can be routinely written. Negative voltages applied to metallic
surface gates define narrow wires or tunnel barriers in the 2DEG. Such a
system is very suitable for quantum transport studies for two reasons. First, the
wavelength of electrons at the Fermi energy is AF = (2rr1ns)112 - (80 - 30) nm,
roughly 100 times larger than in metals. Second, the mobility of the 2DEG can
be as large as 1000 m 2y- I s-I, which corresponds to a transport elastic mean free
path of order 100 ~m. This technology thus allows fabrication of devices
which are much smaller than the mean free path; electron transport through the
device is ballistic. In addition, the device dimensions can be comparable to the
The
electron wavelength, so that quantum confinement is important.
observation of quantized conductance steps in short wires, or quantum point
contacts, demonstrated quantum confinement in two spatial directions [33,34].
Later work on different gate geometries led to the discovery of a wide variety
of mesoscopic transport phenomena [35]. For instance, coherent resonant
transmission was demonstrated through a quantum dot [36] and through an
array of quantum dots [37]. These early dot experiments were performed with
barrier conductances of order e2/h or larger, so that the effects of charge
quantization were relatively weak.
The effects of single-electron charging were first reported in
semiconductors in experiments on narrow wires by Scott-Thomas et al. [38].
With an average conductance of the wire much smaller than e2/h, their
measurements revealed a periodically oscillating conductance as a function of a
voltage applied to a nearby gate. It was pointed out by van Routen and
Beenakker [39], along with Glazman and Shekhter [17], that these oscillations
arise from single electron charging of a small segment of the wire, delineated by
impurities. This pioneering work on "accidental dots" [38,40-43] stimulated the
study of more controlled systems.
The most widely studied type of device is a lateral quantum dot defined by
metallic surface gates. Fig. 1.3 shows an SEM micrograph of a typical device
[44]. The tunnel barriers between the dot and the source and drain 2DEG
regions can be tuned using the left and right pair of gates. The dot can be
squeezed to smaller size by applying a potential to the center pair of gates.
Similar gated dots, with lithographic dimensions ranging from a few !-lm down
to -0.3 ~m, have been studied by a variety of groups. The size of the dot
formed in the 2DEG is somewhat smaller than the lithographic size, since the
2DEG is typically depleted 100 nm away from the gate.
113
Figure 1.3. A scanning electron microscope (SEM) photo of a typical lateral quantum dot device
(600 x 300 nm) defined in a GaAs/AIGaAs heterostructure. The 2DEG is -100 nm below the
surface. Negative voltages applied to the surface gates (i.e. the light areas) deplete the 2DEG
underneath. The resulting dot contains a few electrons which are coupled via tunnel barriers to the
large 2DEG regions. The tunnel barriers and the size of the dot can be tuned individually with the
voltages applied to the left/right pair of gates and to the center pair, respectively. (From
Oosterkamp et al. [44].)
We can estimate the charging energy e21C and the quantum level spacing
fill from the dimensions of the dot. The total capacitance C (i.e. the
capacitance between the dot and all other pieces of metal around it, plus
contributions from the self-capacitance) should in principle be obtained from
self-consistent calculations [45-47]. A quick estimate can be obtained from the
formula given previously for an isolated 2D metallic disk, yielding e21C =
e21(8er e,ft) where R is the disk radius and Er = 13 in GaAs. For example, for a
dot of radius 200 nm, this yields e21C = 1 meV. This is really an upper limit for
the charging energy, since the presence of the metal gates and the adjacent
2DEG increases C. An estimate for the single particle level spacing can be
obtained from Eq. 1.2(b), LiE = 1i 2Im*R 2, where m* = O. 067me is the effective
mass in GaAs, yielding fill = 0.03 meV.
To observe the effects of these two energy scales on transport, the thermal
energy kBT must be well below the energy scales of the dot. This corresponds
to temperatures of order 1 K (kBT = 0.086 meV at lK). As a result, most of the
transport experiments have been performed in dilution refrigerators with base
temperatures in the 10 - 50 mK range. The measurement techniques are fairly
standard, but care must be taken to avoid spurious heating of the electrons in
the device. Since it is a small, high resistance object, very small noise levels
can cause significant heating. With reasonable precautions (e.g. filtering at low
114
temperature, screened rooms, etc.), effective electron temperatures in the 50 100 mK range can be obtained.
It should be noted that other techniques like far-infrared spectroscopy on
arrays of dots [48] and capacitance measurements on arrays of dots [49] and on
single dots [50] have also been employed. Infrared spectroscopy probes the
collective plasma modes of the system, yielding very different information than
that obtained by transport. Capacitance spectroscopy, on the other hand, yields
nearly identical information, since the change in the capacitance due to electron
tunneling on and off a dot is measured. Results from this single-electron
capacitance spectroscopy technique are presented in sections 5 and 7.
2. Basic theory of electron transport through quantum dots.

This section presents a theory of transport through quantum dots that
incorporates both single electron charging and energy level quantization. We
have chosen a rather simple description which still explains most experiments.
We follow Korotkov et al. [51], Meir et al. [52], and Beenakker [6], who
generalized the charging theory for metal systems to include OD-states. This
section is split up into parts that separately discuss (2.1) the period of the
Coulomb oscillations, (2.2) the amplitude and lineshape of the Coulomb
oscillations, (2.3) the Coulomb staircase, and (2.4) related theoretical work.
2.1. PERIOD OF COULOMB OSCILLATIONS.
Fig. 2.1(a) shows the potential landscape of a quantum dot along the transport
direction. The states in the leads are filled up to the electrochemical potentials
/1left and /1right which are connected via the externally applied source-drain
voltage Vsd = (/1left - J1,ight)le. At zero temperature (and neglecting co-tunneling
[53]) transport occurs according to the following rule: current is (non) zero
when the number of available states on the dot in the energy window between
/1left and J1,ight is (non) zero. The number of available states follows from
calculating the electrochemical potential /1dol N). This is, by definition, the
minimum energy for adding the Nth electron to the dot: /1dolN) == UrN) U(N-i), where UrN) is the total ground state energy for N electrons on the dot
at zero temperature.
To calculate UrN) from first principles is quite difficult. To proceed, we
make several assumptions. First, we assume that the quantum levels can be
calculated independently of the number of electrons on the dot. Second, we
parameterize the Coulomb interactions among the electrons in the dot and
between electrons in the dot and those somewhere else in the environment (as
115
in the metallic gates or in the 2DEG leads) by a capacitance C. We further

assume that C is independent of the number of electrons on the dot. This is a
reasonable assumption as long as the dot is much larger than the screening
length (i.e. no electric fields exist in the interior of the dot). We can now think
of the Coulomb interactions in terms of the circuit diagram shown in Fig. 2.2.
Here, the total capacitance C = CI + Cr + Cg consists of capacitances across the
barriers, CI and C" and a capacitance between the dot and gate, Cg This simple
model leads in the linear response regime (i.e. Vsd < < L1Eie, e/C) to an
electrochemical potential /ldolN) for N electrons on the dot [18]:
e~V
(2.1)
Coulomb Blockade
(a)
g.
6E+/ C
"
N ___ N+1 ___ N -+N+ 1-+
Figure 2.1. Potential landscape through a quantum dot. The states in the 2D reservoirs are filled
up to the electrochemical potentials lileft and liright which are related via the external voltage Vsd =
(lileft - liright)le. The discrete OD-states in the dot are filled with N electrons up to lido,(N). The
addition of one electron to the dot would raise lido/N) (i.e. the highest solid line) to lido/N+1)
(i.e. the lowest dashed line). In (a) this addition is blocked at low temperature. In (b) and (c) the
addition is allowed since here lido/N+1) is aligned with the reservoir potentials lileft' liright by
means of the gate voltage. (b) and (c) show two parts of the sequential tunneling process at the
same gate voltage. (b) shows the situation with N and (c) with N+l electrons on the dot.
116
--.....---+~ V }-------'
Figure 2.2. Circuit diagram in which the tunnel barriers are represented as a parallel capacitor
and resistor. The different gates are represented by a single capacitor ICg The charging energy
in this circuit is il(CI + C, + ICg ).
This is of the general form /1do,(N) = J.lciN) + e({>N, i.e. the electrochemical
potential is the sum of the chemical potential J.lch(N) = EN, and the electrostatic
potential e({>N. The single-particle state EN for the Nth electron is measured
from the bottom of the conduction band and depends on the characteristics of
the confinement potential. The electrostatic potential ({>N contains a discrete and
a continuous part. In our definition the integer N is the number of electrons at a
gate voltage Vg and No is the number at zero gate voltage. The continuous part
in ({>N is proportional to the gate voltage. At fixed gate voltage, the number of
electrons on the dot N is the largest integer for which /1do,(N) < /1left == J1,ight.
When, at fixed gate voltage, the number of electrons is changed by one, the
resulting change in electrochemical potential is:
(2.2)
The addition energy /1do,(N+ 1) - /1do,(N) is large for a small capacitance and/or a
large energy splitting .till = EN+1 - EN between OD-states. It is important to note
that the many-body contribution e2/C to the energy gap of Eq. (2.2) exists only
at the Fermi energy. Below /1do,(N), the energy states are only separated by the
single-particle energy differences .till [see Fig. 2.l(a)]. These energy
differences .till are the excitation energies of a dot with constant number N.
A non-zero addition energy can lead to a blockade for tunneling of
electrons on and off the dot, as depicted in Fig. 2.l(a), where N electrons are
localized on the dot. The (N+ l)th electron cannot tunnel on the dot, because
the resulting electrochemical potential /1do,( N + 1) is higher than the potentials of
117
the reservoirs. So, for /1dolN) < /1le.ft>/1right < /1dolN+ 1) the electron transport is
blocked, which is known as the Coulomb blockade.
The Coulomb blockade can be removed by changing the gate voltage, to
align /1dolN+1) between /1left and /1righl' as illustrated in Fig. 2.1(b) and (c). Now,
an electron can tunnel from the left reservoir on the dot [since, /1left >
/ldolN+1)]. The electrostatic increase eqJ(N+1) - eqJ(N) = e2/C is depicted in
Fig. 2.1(b) and (c) as a change in the conduction band bottom. Since /1dolN+l)
> /1righl' one electron can tunnel off the dot to the right reservoir, causing the
electrochemical potential to drop back to /1dol N). A new electron can now
tunnel on the dot and repeat the cycle N ~ N+l ~ N. This process, whereby
current is carried by successive discrete charging and discharging of the dot, is
known as single electron tunneling, or SET.
C!l
1lc:
'C
c:
o "
N+2
~~~
",m
0;
5j
.!o
.,c.
~1'-4--;:=~;-----J-.
tNg
gate voltage V g
Figure 2.3. Schematic comparison, as a function of gate voltage, between (a) the Coulomb
oscillations in the conductance G, (b) the number of electrons in the dot (N+i), (c) the
electrochemical potential in the dot J1do,( N + i), and (d) the electrostatic potential rp.
118
On sweeping the gate voltage, the conductance oscillates between zero

(Coulomb blockade) and non-zero (no Coulomb blockade), as illustrated in Fig.
2.3. In the case of zero conductance, the number of electrons N on the dot is
fixed. Fig. 2.3 shows that upon going across a conductance maximum (a), N
changes by one (b), the electrochemical potential J1dot shifts by fill + e21C (c),
and the electrostatic potential eq> shifts by e21C (d). From Eq. (2.1) and the
condition J1dolN, Vg) = J1dolN+J, Vg+..1Vg), we get for the distance in gate
voltage ..1Vg between oscillations [18]:
(2.3a)
and for the position of the Nth conductance peak:
YiN)
C
= -eCg
( EN +(N--)1 e )
2 C
(2.3b)
For vanishing energy splitting fill == 0, the classical capacitance-voltage relation

for a single electron charge ..1Vg = elCg is obtained; the oscillations are periodic.
Non-vanishing energy splitting results in nearly periodic oscillations. For
instance, in the case of spin-degenerate states two periods are, in principle,
expected. One corresponds to electrons N and N+ 1 having opposite spin and
being in the same spin-degenerate OD-state, and the other to electrons N+ J and
N+2 being in different OD-states.
2.2. AMPLITUDE AND LINESHAPE OF COULOMB OSCILLATIONS.
We now consider the detailed shape of the oscillations and, in particular, the
dependence on temperature. We assume that the temperature is greater than the
quantum mechanical broadening of the OD energy levels hT << kBT. We return
to this assumption later. We distinguish three temperature regimes:
(1) e21C kBT, where the discreteness of charge cannot be discerned.
(2) fill << kBT << e21C, the classical or metallic Coulomb blockade
regime, where many levels are excited by thermal fluctuations.
(3) kBT fill < e21C, the quantum Coulomb blockade regime, where
only one or a few levels participate in transport.
In the high temperature limit, e 21C < < kBT, the conductance is independent
of the electron number and is given by the Ohmic sum of the two barrier
conductances JIG = JIGoo = JIG/eft + lIGright . This high temperature
119
conductance Goo is independent of the size of the dot and is characterized

completely by the two barriers.
The classical Coulomb blockade regime can be described by the so-called
"orthodox" Coulomb blockade theory [4,5,23]. Fig. 2.4(a) shows a calculated
plot of Coulomb oscillations at different temperatures for energy-independent
barrier conductances and an energy-independent density of states. The
Coulomb oscillations are visible for temperatures kBT < 0.3e 2/C (curve c). The
lineshape of an individual conductance peak is given by [6, 23]:
Goo : :
8/kBT
2sinh(8/ kBT)
"'"
.!..COSh- 2(
8
2.5kBT
for hr, L1E kBT e2/C
(2.4)
8 measures the distance to the center of the peak in units of energy, which
expressed in gate voltage is 8 = e(CIC)IVg,res - Vgl, with Vg,res the gate voltage at
resonance. The width of the peaks are linear in temperature as long as kBT
e2/C. The peak maximum Gmax is independent of temperature in this regime
(curves a and b) and equal to half the high temperature value Gmax = Goo /2.
This conductance is half the Ohmic addition value because of the effect of
correlations: since an electron must first tunnel off before the next can tunnel
on, the probability to tunnel through the dot decreases to one half.
0.75
e
0.75
(b)
0.50
C!'
.......
C!'
(a)
0.50
C!'
.......
c:J
0.25
0.25
0.00 '--_ _.-<:...<.......J...--"--""___ _- '
0.75
1.25
AEF / (e 2 /C)
Figure 2.4. Calculated temperature dependence of the Coulomb oscillations as a function of
Fermi energl in the classicr regime (a) and in the quantum regime (b). In (a) the parameters are
,1E :::: O.Ole Ie and kBTI(e Ie) :::: 0.075 [a], 0.15 [b], 0.3 [C], 0.4 [d], 1 [e], and 2 [f]. In (b) the
2
parameters are ,1E :::: O.Ole Ie and kBTI(,1E) :::: 0.5 [a], 1 [b], 7.5 [c], and 15 [d]. (From van
Houten, Beenakker and Staring [7].)
120
In the quantum Coulomb blockade regime, tunneling occurs through a

single level. The temperature dependence calculated by Beenakker [6] is
shown in Fig. 2.4(b). The single peak conductance is given by:
G
G~
fill
= 4kBT cosh
-2( 2kBT
8 )
(2.5)
with the assumption that fill is independent of E and N. The lineshape in the
classical and quantum regimes are virtually the same, except for the different
'effective temperatures'. However, the peak maximum Gmax = G oo(i1EI4kBT)
decreases linearly with increasing temperature in the quantum regime, while it
is constant in the classical regime. This distinguishes a quantum peak from a
classical peak.
The temperature dependence of the peak height is summarized in Fig. 2.5.
On decreasing the temperature, the peak maximum first decreases down to half
the Ohmic value. On entering the quantum regime, the peak maximum
increases and starts to exceed the Ohmic value. Thus, at intermediate
temperatures, Coulomb correlations reduce the conductance maximum below
the Ohmic value, while at low temperatures, quantum phase coherence results
in a resonant conductance exceeding the Ohmic value.
Above we discussed the temperature dependence of an individual
conductance peak and how it can be used to distinguish the classical from the
quantum regimes. Comparing the heights of different peaks at a single
temperature (i.e. in a single gate voltage trace) can also distinguish classical
from quantum peaks. Classical peaks all have the same height Gmax = Goo 12.
3,----------------------.
Cl
Cl
10
ks T IAE
Figure 2.5 Calculated temperature dependence of the maxima and minima of the Coulomb
oscillations for hr kBT and .1E =o.olilc' (From van Houten, Beenakker and Staring [7].)
121
(In semiconductor dots the peak heights slowly change since the barrier
conductances change with gate voltage.) On the other hand, in the quantum
regime, the peak height depends sensitively on the coupling between the levels
in the dot and in the leads. This coupling can vary strongly from level to level.
Also, as can be seen from Gmax = Goo (&J4kBT), the Nth peak probes the
specific excitation spectrum around /1dolN) when the temperatures kBT - L1E
[45]. The quantum regime therefore usually shows randomly varying peak
heights. This is discussed in section 6.
An important assumption for the above description of tunneling in both the
quantum and classical Coulomb blockade regimes is that the barrier
conductances are small: G'ejt,right e2/h. This assumption implies that the
broadening hT of the energy levels in the dot due to the coupling to the leads is
much smaller than kBT, even at low temperatures. The charge is well defined in
this regime and quantum fluctuations in the charge can be neglected (i.e. the
quantum probability to find a particular electron on the dot is either zero or
one). This statement is equivalent to the requirement that only first order
tunneling has to be taken into account and higher order tunneling via virtual
intermediate states can be neglected [53]. A treatment of the regime kBT - hT
involves the inclusion of higher order tunneling processes. Such complicated
calculations have recently been performed [54-56]. For simplicity, we discuss
this regime by considering non-interacting electrons and equal barriers. Then
the zero temperature conductance is given by the well-known Breit-Wigner
formula [57]:
for T = 0, e2/C < < hT, L1E
(2.6)
The on-resonance peak height (i.e. for 8 = 0) is equal to the conductance

quantum 2e2/h; the factor 2 results from spin-degeneracy. The finite
temperature conductance follows from G = !dEGBW(-a.tlaE), wheref(E) is the
Fermi-Dirac function. Although the electron-electron interactions are ignored,
it will be shown in the experimental section that Coulomb peaks in the regime
kBT - hThave the Lorentzian line shape of Eq. (2.6).
2.3. NON-LINEAR TRANSPORT.
In addition to the linear-response Coulomb oscillations, one can obtain
information about the relevant energy scales of the dot by measuring the nonlinear dependence of the current on the source-drain voltage Vsd. Following the
rule that the current depends on the number of available states in the window
e Vsd = /1lejt - J1,ight, one can monitor changes in the number of available states
122
when increasing Vsd' To discuss non-linear transport it is again helpful to

distinguish the classical and the quantum Coulomb blockade regime.
In the classical regime, the current is zero as long as the interval between
/lleft and f.l,ight does not contain a charge state [i.e. when /1dolN) < f.l,ight> /11eft <
/1dolN+ 1) as in Fig. 2.1(a)]. On increasing Vsd, current starts to flow when
either /11eft > /1dol N + 1) or /1dol N) > f.l,ight> depending on how the voltage drops
across the two barriers. One can think of this as opening a charge channel,
corresponding to either the N ~ (N+ 1) [this example is shown in Fig. 2.6] or
the (N-I) ~ N transition. On further increasing Vsw a second channel will open
up when two charge states are contained between /11eft and f.l,ight. The current then
experiences a second rise.
For an asymmetric quantum dot with unequal barriers, the voltage across
the device drops mainly across the less-conducting barrier. For strong
asymmetry, the electrochemical potential of one of the reservoirs is essentially
fixed relative to the charge states in the dot, while the electrochemical potential
of the other reservoir moves in accordance with Vsd. In this asymmetric case,
the current changes are expected to appear in the J- Vsd characteristics as
pronounced steps, the so-called Coulomb staircase [3, 4, 23]. The current steps
M occur at voltage intervals L1 Vsd ::::: e/C. In a region of constant current, the
topmost charge state is nearly always full or nearly always empty depending on
whether the reservoir with the higher electrochemical potential is coupled to
the dot via the small barrier or via the large barrier, respectively. In the
example depicted in Fig. 2.6 the (N+l) charge state is nearly always occupied.
Coulomb Staircase
N electrons
N+ 1 electrons
Figure 2.6. Energy diagram to indicate that for larger source-drain voltage Vsd the empty states
above the Coulomb gap can be occupied. This can result in a Coulomb staircase in currentvoltage characteristic [see Fig. 1.2(b)]
123
Reversing the sign of V. d would leave the (N+l) charge state nearly always
empty.
In the quantum regime a finite source-drain voltage can be used to perform
spectroscopy on the discrete energy levels [51]. On increasing V. d we can get
two types of current changes. One corresponds to a change in the number of
charge states in the source-drain window, as discussed above. The other
corresponds to changes in the number of energy levels which electrons can
choose for tunneling on or off the dot. The voltage difference between current
changes of the first type measures the addition energy while the voltage
differences between current changes of the second type measures the excitation
energIes. More of this spectroscopy method will be discussed in the
experimental section 3.2.
2.4. OVERVIEW OTHER THEORETICAL WORKS.
The theory outlined above represent a highly simplified picture of how
electrons on the dot interact with each other and with the reservoirs. In
particular, we have made two simplifications. First, we have assumed that the
coupling to the leads does not perturb the levels in the dot. Second, we have
represented the electron-electron interactions by a constant capacitance
parameter. Below, we briefly comment on the limitations of this picture, and
discuss more advanced and more realistic theories.
A non-zero coupling between dot and reservoirs is included by assuming
an intrinsic width hT of the energy levels. A proper calculation of hT should
not only include direct elastic tunnel events but also tunneling via intermediate
states at other energies. Such higher order tunneling processes are referred to
as co-tunneling events [53]. They become particularly important when the
barrier conductances are not much smaller than e2/h. Experimental results on
co-tunneling have been reported by Geerligs et ai. [58] and Eiles et ai. [59] for
metallic structures and by Pasquier et ai. [60] for semiconductor quantum dots.
In addition to higher order tunneling mediated by the Coulomb interaction, the
effects of spin interaction between the confined electrons and the reservoir
electrons have been studied theoretically [61-67]. When coupled to reservoirs,
a quantum dot with a net spin, for instance, a dot with an odd number of
electrons, resembles a magnetic impurity coupled to the conduction electrons in
a metal. "Screening" of the localized magnetic moment by the conduction
electrons leads to the well-known Kondo effect [61-67]. This is particularly
interesting since parameters like the exchange coupling and the Kondo
temperature should be tunable with a gate voltage. However, given the size of
present day quantum dots, the Kondo temperature is hard to reach, and no
124
experimental results have been reported to date. Experimental progress has

been made recently in somewhat different systems [68,69].
The second simplification is that we have modeled the Coulomb
interactions with a constant capacitance parameter, and we have treated the
single-particle states as independent of these interactions. More advanced
descriptions calculate the energy spectrum in a self-consistent way. In
particular, for small electron number (N < 10) the capacitance is found to
depend on N and on the particular confinement potential [45-47, 70]. In this
regime, screening within the dot is poor and the capacitance is no longer a
geometric property. It is shown in section 7 that the constant capacitance
model also fails dramatically when a high magnetic field is applied.
Calculations beyond the self-consistent Hartree approximation have also been
performed. Several authors have followed Hartree-Fock [71-73] and exact
[74,75] schemes in order to include spin and exchange effects in few-electron
dots [76]. One prediction is the occurrence of spin singlet-triplet oscillations
by Wagner et ai. [77]. Evidence for this effect has been given recently [50,78],
which we discuss in section 7.
There are other simplifications as well. Real quantum dot devices do not
have perfect parabolic or hard wall potentials. They usually contain many
potential fluctuations due to impurities in the substrate away from the 2DEG.
Their 'thickness' in the z-direction is not zero but typically 10 nm. And as a
function of Vg , the potential bottom not only rises, but also the shape of the
potential landscape changes. Theories virtually always assume effective mass
approximation, zero thickness of the 2D gas, and no coupling of spin to the
lattice nuclei. In discussions of delicate effects, these assumptions may be too
crude for a fair comparison with real devices. In spite of these problems,
however, we point out that the constant capacitance model and the more
advanced theories yield the same, important, qualitative picture of having an
excitation and an addition energy. The experiments in the next section will
clearly confirm this common aspect of the different theories.
3. Experiments on single lateral quantum dots.

This section presents experiments which can be understood with the theory of
the previous section. We focus in this section on lateral quantum dots which
are defined by metallic gates in the 2DEG of a GaAs/AIGaAs heterostructure.
These dots typically contain 50 to a few hundred electrons. (Smaller dots
containing just a few electrons N < 10) will be discussed in section 5.) First,
we discuss linear response measurements (3.1) while section 3.2 addresses
experiments with a finite source-drain voltage.
125
3.1. LINEAR RESPONSE COULOMB OSCILLATIONS.

Fig. 3.1 shows a measurement of the conductance through a quantum dot of the
type shown in Fig. 1.1 as a function of a voltage applied to the center gates
[79]. As in a normal field-effect transistor, the conductance decreases when the
gate voltage reduces the electron density. However, superimposed on this
decreasing conductance are periodic oscillations. As discussed in section 2, the
oscillations arise because, for a weakly coupled quantum dot, the number of
electrons can only change by an integer. Each period seen in Fig. 3.1
corresponds to changing the number of electrons in the dot by one. The period
of the oscillations is roughly independent of magnetic field. The peak height is
close to e2/h. Note that the peak heights at B = 0 show a gradual dependence on
gate voltage. This indicates that the peaks at B = 0 are classical (i.e. the single
electron current flows through many OD-Ievels). The slow height modulation
is simply due to the gradual dependence of the barrier conductances on gate
voltage. A close look at the trace at B = 3.75 T reveals a quasi-periodic
modulation of the peak amplitudes. This results from the formation of Landau
levels within the dot and will be discussed in detail in section 7. It does not
necessarily mean that tunneling occurs through a single quantum level.
-::2
'a,
1.5
= 100 mK
en
.~
..3.
<I>
c
!9
u
0.5
::I
'0
C
-0.7
-0.6
-0.5
-0.4
Gate voltage (V)

Figure 3.1. Coulomb oscillations in the conductance as a function of center gate voltage
measured in a device similar to the one shown in Fig. 1.3, at zero magnetic field and in the
quantum Hall regime. (From Williamson et al. [79].)
126
10-5
280
285
290
295
300
10- 1 , - - - - - - - - ,
(b)
.. ..
,,'
.."...........
~\
10-5 l--..IJ'-----'~-'
282.5
282.9 291.0
292.2
Vg (mV)
Figure 3.2. Coulomb oscillations measured at B = 2.53 T. The conductance is plotted on a

logarithmic scale. The peaks in the left region of (a) have a thermally broadened lineshape as
shown by the expansion in (b). The peaks in the right region of (a) have a Lorentzian lineshape
as shown by the expansion in (c). (From Foxman et ai. [80].)
The effect of increasing barrier conductances due to changes in gate

voltage can be utilized to study the effect of an increased coupling between dot
and macroscopic leads. This increased coupling to the reservoirs (i.e. from left
to right in Fig. 3.1) results in broadened, overlapping peaks with minima which
do not go to zero. Note that this occurs despite the constant temperature during
the measurement. In Fig. 3.2 the coupling is studied in more detail in a
different, smaller dot where tunneling is through individual quantum levels
[80,81]. The peaks in the left part are so weakly coupled to the reservoirs that
the intrinsic width is negligible: hT kBT. The expanded peak in (b) confirms
that the lineshape in this region is determined by the Fermi distribution of the
electrons in the reservoirs.
On a logarithmic scale, the finite temperature Fermi distribution leads to
linearly decaying tails. The solid line in (b) is a fit to Eq. (2.5) with a
temperature of 65 mK and fit parameter e 2/C = 0.61 meV. The peaks in the
127
right part of (a) are so broadened that the tails of adjacent peaks overlap. The
peak in (c) is expanded from this strong coupling region and clearly shows that
the tails have a slower decay than expected for a thermally broadened peak. In
fact, a good fit is obtained with the Lorentzian lineshape of Eq. (2.6) with the
inclusion of a temperature of 65 mK (so kBT = 5.6 f.leV). In this case, the fit
parameters give hr= 5 f.leV and e2/C = 0.35 meV. The Lorentzian tails are still
clearly visible despite the fact that kBT "" hr. An important conclusion from
this experiment is that for strong Coulomb interaction the lineshape for
tunneling through a discrete level is approximately Lorentzian, similar to the
non-interacting Breit-Wigner formula (2.6).
Fig. 3.3(a) shows the temperature dependence of a set of selected
conductance peaks [52]. These peaks are measured for barrier conductances
much smaller than e2/h where hr kBT. Above approximately lK, the peak
heights increase monotonically, but at low temperature, a striking fluctuation in
peak heights is observed. Moreover, some peaks decrease and others increase
on increasing temperature. Random peak behavior is not seen in metallic
Coulomb islands. It is due to the discrete density of states in quantum dots.
The randomness from peak to peak is usually ascribed to the variations in the
nature of the energy levels in the dot. The observed behavior is reproduced in
the calculations of Fig. 3.3(b), where variations are included in the form of a
random coupling of the quantum states to the leads [52]. The origin of this
random coupling is the random speckle-like spatial pattern of the electron wave
function in the dot resulting from a disordered or "chaotic' confining potential.
We return to chaos in dots in section 6.
02
lal experiment
-.:
...
.......
- .----
T : 1.25 K
T :0.8K
T :0.4K
T :0.2K
300 (bllheory
~200
"U
C
~:0 OJ
...g
~
....
0
gllle voltage
(10 mV full scale)
....... kaT :0.2 U

.. kaT =O.12U
kaT: 0.06U
kaT: 0.03U
_
-
tOO
0
7.0
Chemical potential (normalized to lJ)
Figure 3.3. Comparison between (a) measured and (b) calculated Coulomb oscillations in the
quantum regime for different temperatures at B = O. For the calculation, the level spacing was
taken to be uniform: AE = 0.1 U = O.1iIC, but the coupling of successive energy levels was
varied to simulate both an overall gradual increase and random variations from level to level
(From Meir, Wingreen and Lee [52].)
128
The temperature dependence of a single quantum peak at B = 0 is shown in

Fig. 3.4 [81]. The upper part shows that the peak height decreases as inverse
temperature up to about 0.4 K. Beyond 0.4 K the peak height is independent of
temperature up to about 1 K. We can compare this temperature behavior with
the theoretical temperature dependence of classical peaks in Fig. 2.4(a) and the
quantum peaks in Fig. 2.4(b). The height of a quantum peak first decreases
until kBT exceeds the level spacing LlE, where around 0.4 K it crosses over to
the classical Coulomb blockade regime. This transition is also visible in the
width of the peak. The lower part of Fig. 3.4 shows the full-width-at-halfmaximum (FWHM) which is a measure of the lineshape. The two solid lines
differ in slope by a factor 1.25 which corresponds to the difference in "effective
temperature" between the classical line shape of Eq. (2.4) and quantum
lineshape of Eq. (2.5). At about 0.4 K a transition is seen from a quantum to a
classical temperature dependence, which is in good agreement with the theory
of section 2.2.
The conclusions of this section are that quantum tunneling through
discrete levels leads to conductance peaks with the following properties: The
maxima increase with decreasing temperature and can reach e2/h in height; the
lineshape gradually changes from thermally broadened for weak coupling to
Lorentzian for strong coupling to the reservoirs; and the peak amplitudes show
random variations at low magnetic fields.
300
:2 200
"'-...
~
'j a.
CJ
>
100
0
70 2.0
- 1.0
T'"
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Temperature (K)
Figure 3.4. Inverse peak height and full-width-half-maximum (FWHM) versus temperature at B
= O. The inverse peak height shows a transition from linear T dependence to a constant value.
The FWHM has a linear temperature dependence and shows a transition to a steeper slope. These
measured transitions agree well with the calculated transition from the classical to the quantum
regime. (From Foxman et ai. [81].)
129
3.2. NON-LINEAR TRANSPORT REGIME.

In the non-linear transport regime one measures the current I (or differential
conductance dlldVsd ) while varying the source-drain voltage Vsd or the gate
voltage Vg The rule that the current depends on the number of states in the
energy window e Vsd = (/llejt - flright) suggests that one can probe the energy level
distribution by measuring the dependence of the current on the size of this
window.
The example shown in Fig. 3.5 demonstrates the presence of two energies:
the addition energy and the excitation energy [82]. The lowest trace is
measured for small Vsd (Vsd M?). The two oscillations are regular Coulomb
oscillations where the change in gate voltage L1 Vg corresponds to the energy
necessary for adding one electron to the dot. In the constant capacitance model
this addition energy is expressed in terms of energy in Eq. (2.2) and in terms of
gate voltage in Eq. (2.3).
The excitation energy is discerned when the curves are measured with a
larger Vsd . A larger source-drain voltage leads not only to broadened
oscillations, but also to additional structure. Single peaks clearly develop into
double peaks, and then triple peaks. Below we explain in detail that a single
peak corresponds to tunneling through a single level (when eVsd < M?), a double
peak involves tunneling via two levels (when LiE < eVsd < 2M?), and a triple
peak involves three levels (when 2LiE < eVsd < 3M?). The spacing between
mini-peaks is therefore a measure of the excitation energy with a constant
number of electrons on the dot. These measurements yield to an excitation
energy LiE ::::: 0.3 meV which is about ten times smaller than the Coulomb
CENTER GATE VOLTAGE [V]

Figure 3.5. Coulomb peaks at B = 4 T measured for different source-drain voltages Vsd. From
the bottom curve up Vsd = 0.1, 0.4, and 0.7 meV. The peak to peak distance corresponds to the
addition energy i Ie + fill. The distance between mini-peaks corresponds to the excitation
energy fill for constant number of electrons on the dot. (From Johnson et al. [82].)
130
charging energy in this structure.

To explain these results in more detail, we use the energy diagrams [83] in
Fig. 3.6 for five different gate voltages. The thick vertical lines represent the
tunnel barriers. The source-drain voltage Vsd is somewhat larger than the
energy separation & = EN+l - EN, but smaller than 2&. In (a) the number of
electrons in the dot is N and transport is blocked. When the potential of the dot
is increased via the gate voltage, the OD-states move up with respect to the
reservoirs. At some point the Nth electron can tunnel to the right reservoir,
which increases the current from zero in (a) to non-zero in (b). In (b), and also
in (c) and (d), the electron number alternates between N and N-J. We only
show the energy states for N electrons on the dot. For N-J electrons the single
particle states are lowered by e2/C. (b) illustrates that the non-zero source-drain
voltage gives two possibilities for tunneling from the left reservoir onto the dot.
A new electron, which brings the number from N-J back to N, can tunnel to the
ground state EN or to the first excited state EN+1 This is denoted by the crosses
in the levels EN and EN+1 Again, note that these levels are drawn for N
electrons on the dot. So, in (b) there are two available states for tunneling onto
the dot of which only one can get occupied. If the extra electron occupies the
excited state EN+lo it can either tunnel out of the dot from EN+lo or first relax to
EN and then tunnel out. In either case, only this one electron can tunnel out of
the dot. We have put the number of tunnel possibilities below the barriers.
If the potential of the dot is increased further to the case in (c), the
electrons from the left reservoir can only tunnel to EN' This is due to our choice
for the source-drain voltage: & < eVsd < 2&. Now, only the ground state
level can be used for tunneling on and for tunneling off the dot. So, compared
to (b) the number of tunnel possibilities has been reduced and thus we expect a
somewhat smaller current in (c).
Continuing to move up the potential of the dot results in the situation
depicted in (d). It is still possible to tunnel to level EN, but in this case one
electron, either in EN or in EN-lo can tunnel off the dot. So, here we have one
possibility for tunneling onto the dot and two for tunneling off. Compared to
(c) we thus expect a rise in the current. It can be shown that for equal barriers
the currents in (b) and (d) are equal [83].
Increasing the potential further, brings the dot back to the Coulomb
blockade and the current drops back to zero. Note that in (e) the levels are
shown for N-J electrons on the dot.
Going through the cycle from (a) to (e) we see that the number of available
states for tunneling changes like 0-2-1-2-0. This is also observed in the
experiment of Fig. 3.5. The second curve shows two maxima which
correspond to cases 3.6(b) and 3.6(d). The minimum corresponds to case
3.6(c). For a source-drain voltage such that 2& < eVsd < 3& the sequence of
131
(a)
(b)
(e)
N"' N- 1
(d)
N"'N-1
N"'N - 1
----
r ---
(e)
EN
~
~
-""
:~ E
N-1
e2
C ----
EN
ij
Figure 3.6. Energy diagrams for five increasingly negative gate voltages. The thick lines
represent the tunnel barriers. The number of available tunnel possibilities is given below the
barriers. Horizontal lines with a bold dot denote occupied OD-states. Dashed horizontal lines
denote empty states. Horizontal lines with crosses may be occupied. In (a) transport is blocked
with N electrons on the dot. In (b) there are two states available for tunneling on the dot and one
state for tunneling off. The crosses in level EN and E N+1 denote that only one of these states can
get occupied. In (c) there is just one state that can contribute to transport. In (d) there is one state
to tunnel to and there are two states for tunneling off the dot. In (e) transport is again blocked,
but now with N-l electrons on the dot. Note that in (a)-(d) the levels are shown for N electrons
on the dot while for (e) the levels are shown for N-l electrons (From van der Vaart et ai [83].)
4.0
..,
c::
(a)
2.5
3
?;,~
......
:a
1.0
-0.5
-2
-I
V. (mV)
~~.306:. .:. . .:. . :>.307

::.u...':""""::....l>..;~"""";'~~~~
306
309
310
311
'S- 0.70
305
V, (mY)
Figure 3.7. (a) dUdVsd versus Vsd at B = 3.35 T. The positions in Vsd of peaks from traces as in
(a) taken at many different gate voltages are plotted in (b) as a function of gate voltage. A factor
ef3 has been used to convert peak positions to electron energies. The diamond-shaped addition
energy and the discrete excitation spectrum are clearly seen. (From Foxman et ai. [80].)
132
contributing states is 0-3-2-3-2-3-0 as the center gate voltage is varied. Here,

the structured Coulomb oscillation should show three maxima and two local
minima. This is observed in the third curve in Fig. 3.5.
Fig. 3.7 shows data related to that in Fig. 3.5, but now in the form of
d/1dVsd versus Vsd [80]. The quantized excitation spectrum of a quantum dot
leads to a set of discrete peaks in the differential conductance; a peak in d/ldVsd
occurs every time a level in the dot aligns with the electrochemical potential of
one of the reservoirs. Many such traces like Fig. 3.7(a), for different values of
Vg, have been collected in (b). Each dot represents the position of a peak from
(a) in Vg - Vsd space. The vertical axis is multiplied by a parameter that converts
Vsd into energy [80]. Fig. 3.7(b) clearly demonstrates the Coulomb gap at the
Fermi energy (i.e. around Vsd = 0) and the discrete excitation spectrum of the
single particle levels. The level separation ,1E "" 0.1 meV in this particular
device is much smaller than the charging energy e21C"" 0.5 meV.
The transport measurements described in this section clearly reveal the
addition and excitation energies of a quantum dot. These two distinct energies
arise from both quantum confinement and from Coulomb interactions. Overall,
agreement between the simple theoretical model of section 2 and the data is
quite satisfactory. However, we should emphasize here that our labeling of ODstates has been too simple. In our model the first excited single-particle state
EN+1 for the N electron system is the highest occupied single-particle state for
the ground state of the N+l electron dot. (We neglect spin-degeneracy here.)
Since in lateral dots the interaction energy is an order of magnitude larger than
the separation between single-particle states, it is reasonable to expect that
adding an electron will completely generate a new single-particle spectrum. In
fact, to date there is no evidence that there is any correlation between the
single-particle spectrum of the N electron system and the N+ 1 electron system.
This correlation is still an open issue. We note that there are also other issues
that can regulate tunneling through dots. For example, it was recently argued
[84] that the electron spin can result in a spin-blockade. The observation of a
negative-differential conductance by Johnson et al. [82] and by Weis et al. [85]
provides evidence for this spin-blockade theory [84].
4. Multiple Quantum Dot Systems.

This section describes recent progress in multiple coupled quantum dot
systems. If one quantum dot is an artificial atom [2], then multiple coupled
quantum dots can be considered to be artificial molecules [86]. Combining
insights from chemistry with the flexibility possible using lithographically
defined structures, new investigations are possible. These range from studies
133
of the "chemical bond" between two coupled dots, to OD-state spectroscopy, to

possible new applications as quantum devices, including use in proposed
quantum computers [87,88].
The type of coupling between quantum dots determines the character of
the electronic states and the nature of transport through the artificial molecule.
Purely electrostatic coupling is used in orthodox Coulomb blockade theory, for
which tunnel rates are assumed to be negligibly small. The energetics are then
dominated by electrostatics and described by the relevant capacitances,
including an inter-dot capacitance Cint However, in coupled dot systems the
inter-dot tunnel conductance Gint can be large, i.e. comparable to or even
greater than the conductance quantum GQ = 2e2/h. In this strong tunneling
regime the charge on each dot is no longer well defined, and the Coulomb
blockade is suppressed. So, both the capacitance and the tunnel conductance
may influence transport.
All experiments described in this section are performed on lateral coupled
quantum dots. Experiments on electrostatically-coupled quantum dots are
described in 4.1, followed by experiments and theory of tunnel-coupled dots in
4.2. A new spectroscopic technique using two quantum dots is described in
4.3, and the issue of quantum coherence is discussed in 4.4.
4.1. ELECTROSTATICALLY-COUPLED QUANTUM DOTS.
Electrostatically-coupled quantum dots with negligible inter-dot tunnel
conductance fall within the realm of single electronics and are covered by the
orthodox theory of the Coulomb blockade [4]. In this regime, the role of
tunneling is solely to permit the incoherent transfer of electrons through an
electrostatically-coupled circuit. Single electronics, so called because one
electron can represent one bit of information, has been extensively studied for
possible future applications in electronics and computation. Many interesting
applications have been developed, including the electron turnstile [89,90], the
electron pump [91,92], and the single electron trap [93]. Electron pumps
fabricated from metal tunnel junctions are closely related to coupled quantum
dot systems. An analysis of the electron pump using the orthodox theory of the
Coulomb blockade was originally done by Pothier et al. [91], who found good
agreement with experiment.
We confine the discussion here to
electrostatically-coupled semiconductor dots.
Parallel coupled dot configurations permit measurements using one dot to
sense the charge on a second neighboring dot. If the first dot is operated as an
electrometer the charge on a neighboring dot can modulate the current through
this electrometer. Dot-electrometers are extremely sensitive to the charge on a
neighboring dot. Even small changes of order 10-4e can be detected. A change
134
of a full electron charge can completely switch off the current through the
electrometer. Narrow channels have also been employed as electrometers [94].
A parallel quantum dot configuration which demonstrates switching [95]
is shown by the atomic force micrograph in Fig. 4.1. The dot on the right (dot
1) is operated as an electrometer controlled by gate Cl, while the dot on the left
(dot 2) is a box without leads whose charge is controlled by a second gate C2.
The voltages on gates Fl, F2, Ql, and Q2 were used to define the double dot
structure and to adjust the coupling between the dots. The coupling here is
primarily capacitive, since the tunnel probability between dots 1 and 2 is made
very small. Plots of the peak positions in the linear response conductance were
used to map out the charge configurations (NbN2) vs. gate voltages Vg1 and Vg2 ,
with N; the number of electrons on dot i.
Fig. 4.2(a) is a plot of the measured conductance vs. gate voltages Vg1 and
Vg2 (labeled Vel and Vd. The peak locations jump periodically as the gate
voltage on dot 2 is swept. This is clear evidence of the influence of the charge
of dot 2, and it provides a demonstration of the switching function of
capacitively-coupled dots. Fig. 4.2(b) is a comparison of the measured peak
positions with a capacitive coupling model based on Coulomb blockade theory,
described below. As shown, the agreement between theory and experiment is
excellent.
The honeycomb pattern in Fig. 4.2 is characteristic of electrostaticallycoupled double dot systems. Using Coulomb blockade theory, one can
calculate the total electrostatic energy UNl,NlVgb Vg2 ) of the coupled dot system
for a given charge configuration (NJ,N2) in the limit of negligible inter-dot
tunneling [91, 95-97]. The energy surface UNl.NlVgb Vg2) is a paraboloid with a
minimum at the point (VgJ, Vg2) for which the actual charge configuration
(Nb N 2) equals the charge induced by the gates. As Nl or N2 is incremented, the
charging energy surface U repeats with a corresponding offset in gate voltage.
The set of lowest charging energy surfaces forms an "egg-carton" potential
defining an array of cells in gate voltage. Within each cell, Nl and N2 are both
fixed. The borders between cells are defined by the intersections of charging
energy surfaces for neighboring cells.
Along the border between cells the dot charge can change and transport
can occur. For parallel coupled dot systems, current can flow through dot 1
only when the number of electrons on dot 1 can change. Referring to Fig. 4.2,
this occurs along the lines joining cells (NJ,N2) and (Nll,N2)' Changes in the
charge of dot 2 are possible along the boundaries between cells (NJ,N2) and
(NJ,N21). However, these changes do not produce current through dot 1.
Also, no current flows as charge is swapped from one dot to the other along the
boundaries between cells (NJ,N2) and (Nl+l,Nz-l) or (Nr l,N2+l) because the
135
Figure 4.1. An atomic force image of the parallel-coupled dot device. The outer electrodes
define both the quantum point contacts coupling the main dot to the reservoirs as well as the
inner quantum point contact which controls the inter-dot coupling. The geometries of the two
dots can be independently tuned with the two center gates. The right dot is about 400 x 400 nm
(From Hoffman et at. [95].)
.710 _
..
<.>
>
730
740
.750
540
520
500
V el (mV)
540
500
Figure 4.2. (a) Typical conductance measurements for the parallel dot structure shown in Fig.
4.1. Each trace corresponds to a fixed value of second center-gate voltage Ve2 ; the curves are
offset vertically. The conductance scale is arbitrary but constant over the whole range of
measurements. (b) The conductance maxima of (a) are plotted as dots as a function of the two
center-gate voltages. The phase diagram calculated from an electrostatic model is plotted as solid
lines. (From Hoffman et al. [95].)
136
total number N = N] + N2 stays constant. These two borders are not predicted
to lead to observable current peaks, and the corresponding sections of the
honeycomb pattern in Fig. 4.2(b) do indeed not contain experimental data
points. Note that for series coupled double dot systems, like those discussed
below, the constraints on transport are different. Current can flow through both
dots only when the charges on both dots can change. This occurs at the set of
points where the line boundaries between cells (NbN2) and (N]1,N2) and
between cells (NbN2) and (NbN21) intersect. The resulting conductance peak
pattern is just a hexagonal array of points, as demonstrated in section 4.2.
Two dimensional scans of double dot conductance peak data are necessary
in order to see the simplicity of the pattern of hexagonal cells. If only one gate
is swept, the one dimensional trace cuts through data such as Fig. 4.2(a) along a
single line, which may be inclined relative to both axes due to capacitive
coupling of a single gate to both dots. The resulting series of peaks is no longer
simply periodic, but shows evidence of both periods. With series coupled
double dots, for which transport occurs only at an array of points, one
dimensional conductance traces typically miss many points. The resulting
suppression of conductance peaks predicted [98,99] and observed [100,101] in
series coupled dot systems has been called the stochastic Coulomb blockade.
4.2. TUNNEL-COUPLED QUANTUM DOTS.
When electrons tunnel at appreciable rates between quantum dots in a coupled
dot system, the system forms an artificial molecule with "molecular" electronic
states which can extend across the entire system. In this regime the charge on
each dot is no longer quantized, and the orthodox theory of the Coulomb
blockade can no longer be applied. Nonetheless, the total charge of the
molecule can still be conserved. The reduction in ground state energy brought
about by tunnel-coupling is analogous to the binding energy of a molecular
bond. Analogies with chemistry and biochemistry may permit the design of
new types of quantum devices and circuits based on tunnel-coupled quantum
dots. Perhaps the most adventurous of these would be a quantum computer
with electronic states which coherently extend across the entire system [87].
The issue of coherence is addressed in section 4.4.
Early experiments [37] on fifteen tunnel-coupled dots in series
demonstrated evidence for coherent states over the entire sample. Below, we
focus on more recent experiments on smaller two and three dot arrays, which
make use of the ability to calibrate and continuously tune the inter-dot tunnel
conductance during the experiment [96,101-107]. These experiments are
analogs of textbook "gedanken" experiments in which the molecular binding
energy is studied as a function of inter-atomic spacing.
137
The central issues in the theory of tunnel-coupled dots are the destruction
of charge quantization by tunneling and the resulting changes in the ground
state energy. These issues were originally addressed theoretically in the
context of small metal tunnel junctions with many tunneling channels [108111], unlike semiconductor dots with few channels.
The theory of
semiconductor dot arrays initially focused on models with one-to-one tunnel
coupling between single particle levels in neighboring dots [112-114] with
charging energy Ec = e2/C small or comparable to the energy level spacing Lill.
This approximation is inappropriate for strongly coupled many electron
quantum dots like those discussed below, for which the charging energy is
much greater than the level spacing Ec Lill. (For 2D dots with constant EF
the charging energy varies with N as Ec - EF IYN, because C is proportional to
the dot radius, while the level spacing varies as Lill - EF IN. In addition for N =
1 we have EF - Ec - Lill, producing the above inequality.) The destruction of
charge quantization requires inter-dot tunneling rates comparable to Ec I Ii ,
large enough to couple to a broad range of single-particle levels on each dot.
The many-body theory of tunnel-coupled quantum dots in this limit has
recently been developed [115-118] and is summarized below.
We begin this section with a description of an experiment on a double
quantum dot with tunable inter-dot tunneling by Waugh et ai. [101,102]. The
experimental geometry is shown in Fig. 4.3(a), which is an SEM photograph of
fourteen gates used to define a series array of nominally identical coupled
quantum dots inside the 2DEG of a GaAs/AIGaAs heterostructure. By
energizing only some of the gates, systems with one, two, or three dots can be
created, and the conductance of each of the four quantum point contacts can be
recorded and calibrated individually. For the data discussed below, two dots in
series were formed, isolated by tunnel barriers from the leads, and traces of the
linear response conductance peaks at low temperature were recorded vs. gate
voltage. In this experiment two side gates with equal gate capacitance Cg were
tied together so that the induced charges on each dot were nominally identical.
This unpolarized gate arrangement corresponds to a trace taken along the
diagonal VCI = V C2 in Fig. 4.2(b).
Fig. 4.4 presents the central experimental result. As the inter-dot tunnel
conductance increases from Gint = 0.03e 2/h in (a) to Gint = 1.94e2/h in (d), the
conductance peaks split by an increasing amount LiV., saturating in (d) at the
spacing LiVsar ' The peak spacing in (a) is the same as for a single dot, whereas
the spacing in (d) is half as large, indicating that the two dots have combined
into a single large dot with twice the gate capacitance. Data similar to Fig. 4.4
were obtained independently by van der Vaart et ai. [103].
We can understand the origin of peak splitting by examining theoretical
plots of the electrostatic energy associated with the ground state of the array for
138
gate voltage (eICg )

Figure 4.3. (a) SEM micrograph of three coupled quantum dots with tunable barriers. The dots
are 0.5 x 0.8 Ilm2. (b) Double dot charging energy vs. gate voltage for (NJ,N2) electrons on each
dot for identical dots. Without inter-dot coupling, parabolas with NJ, #: N2 are degenerate (solid
curves). Inter-dot coupling removes the degeneracy, shifting the lowest parabola down by .Ii =
Eint (dotted curves). Schematic double dot conductance vs. gate voltage without coupling (c),
and with coupling (d). Peaks occur at open markers in (b). (From Waugh et ai. [102].)
gate voltage V s
Figure 4.4. Double dot conductance Gdd vs. gate voltage V5 for increasing inter-dot coupling.
Coupling splits conductance peaks, with split peak separation .liV. proportional to the interaction
energy Eint Inter-dot conductance in units of ilh is in (a) 0.03, in (b) 0.88, in (c) 1.37, and in (d)
1.94. (From Waugh et ai. [102].)
139
different charge configurations (NJ,N2 ). Plots of the electrostatic energy for

zero inter-dot coupling are shown as the solid lines in Fig. 4.3(b). Because
equal charges are induced by the gates in the experiment, unpolarized
configurations with equal numbers of electrons, (0,0) and (1,1), have lowest
energy. If only one electron is added to the array, and we assume the absence
of inter-dot tunneling, it must necessarily reside on one dot or the other. This
results in the polarized configurations, (1,0) and (0,1), with excess electrostatic
energy. A conductance trace for this case shows the periodicity associated with
a single dot, as illustrated in (c). Inter-dot coupling, either tunnel coupling or
electrostatic, acts to reduce the energy of the polarized configurations, as
indicated by the dashed curve in (b), causing conductance peaks to split, as
indicated in (d). When the polarization energy is completely destroyed, the
dashed curve in (b) touches the horizontal axis, and the peak spacing saturates
at half the original value.
By calibrating the inter-dot tunnel conductance vs. gate voltage Waugh et
ai. [102] demonstrated that the peak splitting was strongly correlated with the
inter-dot tunnel conductance Gint , increasing from a small value for Gint == to
saturation for Gint == 2e 21h. Thus the quantization of charge on individual dots is
completely destroyed by one quantum of inter-dot conductance. Intuitively,
one might expect that the inter-dot capacitance Cint diverges as the constriction
joining the dots is opened, and this approach has been used to analyze coupled
dot data with orthodox charging theory. However, for lateral dots the inter-dot
capacitance increases more slowly than logarithmically with inter-dot spacing.
Thus the inter-dot capacitance remains essentially constant even when the
inter-dot tunneling changes exponentially fast. Numerical simulations [119] of
the actual experimental geometry confirm these arguments. (Note that the
capacitance per unit length L between two thin co-planar rectangular metal
strips of width wand separation d in a medium with dielectric constant e is CIL
= (2t:1n)ln(2wld) for L d.)
In the experiments discussed in this section we can ignore effects from the
single-particle states. The central issue here is the destruction of charge
quantization by tunneling. For single dots, the destruction of charge
quantization by increasing the tunnel conductances was addressed early on by
Kouwenhoven et ai. [18], and Foxman et ai. [80). As the tunnel conductance G
to the leads increases, the linewidth of the peaks becomes lifetime broadened
(e.g. see Fig. 3.2) and the Coulomb blockade is destroyed when G == 2e 21h [18].
To analyze data such as these Matveev [115] extended earlier theoretical work
on charge fluctuations in metal junctions [56,108-111] to the few channel case
appropriate for dots. Matveev considered a single dot with many electrons
connected to the surrounding reservoir by a single 1D lead. Using a Luttinger
liquid approach he calculated the interaction energy vs. tunneling rate for the
140
strong tunneling limit and confirmed that the quantization of charge on the dot
and the associated energy were completely destroyed when the lead
conductance reached exactly 2e 2/h.
Matveev et al. [117] and Golden and Halperin [116,118] extended the
single dot results to address the case of two quantum dots connected by a 1D
lead but isolated from the surrounding reservoir. The initial work by both
groups [116,117] achieved similar results, and both found good agreement with
the experiment [102]. Later the calculations were extended to include higher
order terms [118]. To compare theory with data like Fig. 4.4, Golden and
Halperin [116,118] express their results for the interaction energy Eint, the
analog of the molecular binding energy, in terms of a normalized energy shift:
(4.1)
where 0 < p ~ 1 is the difference in induced charge between the two dots in
units of e, and Ee2 =e21(C+2Cint) is the charging energy of one of the two dots
including the effect of the inter-dot capacitance Cint For unpolarized dots p =
1, as discussed in Ref. 116, andf(p) is equal to the fractional peak splittingf =
L1V.IL1V. at In the weak tunneling limit the fractional peak splitting is [118]:
(4.2)
where Nch is the number of channels, Nch = 2 for one spin degenerate mode, and
g = Gint I(Nche 2/h) is the dimensionless inter-dot conductance per channel. In
the strong tunneling limit 0 < l-g 1, the peak splitting for Nch = 2 becomes
[118]:
== 1+0.919(1- g)lg(l- g)-0.425(1- g)
(4.3)
Below we compare these functions to measurements.

Tunnel-coupled double dots are controlled by three parameters, the two
gate voltages Vg1 and Vg2 and the inter-dot tunnel conductance Gint Data on
tunnel-coupled double dots in series [102,103,106,107] initially consisted of
2D slices through this 3D parameter space. In order to map out the full 3D
pattern Livermore et al. [104] conducted new experiments on a coupled dot
structure with two nominally identical dots, otherwise similar to Fig. 4.3(a).
The two gate voltages Vg1 and Vg2 were swept in a raster pattern while the series
conductance Gdot through the double dot was measured to create an image of
141
10
>
->
E
5
0
~10
5
0
10
150
10
150
10
15
Vg1 (mV)
Figure 4.5. Logarithm of double dot conductance as a function of gate voltages VgJ and V g2 ,
which are offset to zero. Dark indicates high conductance; white regions represent low
conductance. Panels are arranged in order of increasing inter-dot conductance Gin" which in
units of 2ilh is in (a) 0.22, in (b) 0.40, in (c) 0.65, in (d) 0.78, in (e) 0.96, and in (f) 0.98. (From
Livermore et at. [104].)
the conductance pattern for fixed Gint . This procedure was repeated for a series
of values for Gint
Figures 4.5(a) to (f) present conductance images for inter-dot tunnel
conductances increasing from a small value Gint = O.22GQ in (a) to Gim =
O.98GQ in (f). This series of images presents a complete picture of how the
pattern of Gdot evolves with inter-dot tunneling. In (a) the effects of inter-dot
conductance are small, and the pattern is a hexagonal array of points, as
expected for two capacitively coupled dots in series. Close examination
reveals a small splitting of each point due to the inter-dot capacitance C int
Within each hexagon, NJ and N2 are well defined. Compared with data from
capacitively-coupled parallel dots in Fig. 4.2, the pattern for series dots differs
in that conductance only occurs at the points where two sides of each hexagon
intersect. As discussed above, this follows from the requirement that the
number of charges on both dots must be able to change simultaneously in order
for transport to occur.
As the inter-dot tunnel conductance increases, the dot conductance pattern
in Fig. 4.5 changes dramatically. At Gim == GQ in (f) the conductance pattern
142
becomes an array of lines corresponding to the Coulomb blockade for a single

large dot. It is convenient to change gate voltage variables from Vgl and Vg2 to
the average gate voltage Vav = (Vgl +Vg2 )/2, and the difference Vd(ff = Vgr Vg2 .
The average Vav induces charge equally on both dots and increases the total
charge, whereas the difference Vd(ff induces polarization in the double dot
system and leaves the total charge unchanged. The pattern in (f) shows an
array of lines perpendicular to the Vav direction which separate regions defined
by integer values of the total double dot charge Ntot = Nl + N2. The pattern in
(f) is insensitive to Vd(ff because the dots have effectively joined into one large
dot. The evolution of the conductance patterns between the weak and strong
tunneling limits shows how inter-dot tunneling changes transport. As the interdot tunnel conductance increases, the condition that both Nl and N2 are
quantized relaxes into a single condition that the total charge Ntot = Nl + N2 is
quantized. Between these two extremes the conductance grows steadily out
from the array of points in (a) along the boundaries between configurations
with different total charge Ntol> and the shape of these boundaries changes from
the zigzag pattern for weak tunneling to straight lines in (f).
The splitting between the lines of conductance in Fig. 4.5 measures the dot
interaction energy predicted by theory [116-118] and thus determines an analog
of the molecular binding energy. The fractional splitting f = 2L1Vs IVp is the
ratio of the minimum separation L1 Vs between lines of conductance along the
Vav direction to the saturated splitting Vp 12 where Vp is the period in (a) along
the Vav direction.
The inter-dot tunnel conductance G int was independently
1.0
0.8
lL
0.6
0.4
0.2
0.0
0.0
L-._",--_",,--_..I-_~_--'--
0.2
0.4
0.6
2
0.8
1.0
G1nt (2e /h)

Figure 4.6. Measured fractional splitting F (circles), theoretical fractional splitting (solid lines),
and theoretical interpolation (dotted line) plotted as a function of inter-dot tunnel conductance
Gint Theoretical splitting includes both splitting due to inter-dot tunneling and the small splitting
due to interdot capacitance. (From Livermore et al. [104].)
143
determined by separate measurements of the center point contact with the two
outer point contacts open. Fig. 4.6 plots measurements of the fractional
splittingfvs. Ginr , together with theory [118] in the weak and strong tunneling
limits (solid curves) and an interpolation between these limits (dashed curve).
As shown, the agreement between experiment and theory is excellent,
providing strong support for the charge quantization theory. Similar agreement
was also found for measurements by Adourian et al. [105] for two tunnelcoupled quantum dots in parallel. A somewhat different approach was used by
Molenkamp et al. [120] to analyze their experiments on a parallel dot system.
So far we have discussed only linear response transport, measured for very
small source-drain voltages. Crouch et al. [121] have measured the Coulomb
blockade for series double dots via non-linear J- Vsd curves. The devices used
were essentially a double dot version of the device in Fig. 4.3(a). The two side
gates were tied together to induce equal charges on both dots, and the
differential conductance Gdif = dJldVsd was measured as both the source-drain
voltage Vsd and gate voltage were swept. Figs. 4.7(a) to (d) present gray scale
plots of Gdif vs. Vsd and vs. gate voltage as the inter-dot tunnel conductance is
increased from Gint == 0 in (a) to Gint == 2e2/h in (d). The diamond pattern in (a)
750
750
~5OO
500
250
250
f
.i-~
o
250
500
1-500
750
-0.632
-0.630
-0.628
-0.626
(b)
gw=O.69
750
~5OO
.ll.6.14
.ll.6.12
.ll.630
.ll.628
750
500
t2S0
250
1.~
.ll.6.16
-0.624
o
-250
-500
500
750
-0.580
-0.578
-0.576
gate voltage V1 (V)
-0.574
-0.572
gale voltage VI (V)
Figure 4.7. Differential conductance Gdif on a logaritmic scale as function of source-drain

voltage and side gate voltages Vg1 = Vg2 for inter-dot conductances gint (units of 2ilh) in (a) 0.28,
in (b) 0.69, in (c) 0.89, and in (d) 0.95. (From Crouch et al. [121].)
144
is essentially that measured for a single dot (see section 3.2), showing a single
well developed region of Coulomb blockade. The period in gate voltage
corresponds to adding one electron to each dot. As the inter-dot tunnel
conductance increases, a second smaller region of Coulomb blockade develops.
This second region grows with Gint until the evolution saturates in (d) at a
pattern for a single large dot with half the gate voltage periodicity. This
evolution is qualitatively in accord with what one would expect from linearresponse data above and from charge quantization theory. A quantitative
comparison of theory and experiment [121] requires an additional selfconsistent calculation [96] to include the effects of inter-dot polarization
induced by Vsd, when this is done very good agreement is found.
In summary, the quantization of electronic charge produces frustration in
electrostatically-coupled dots which is relaxed by inter-dot tunneling, resulting
in a reduction in the ground state energy analogous to a molecular binding
energy. Frustration arises because the optimal charges induced by gates on
coupled dots are not generally integral multiples of the electronic charge.
Forcing integral occupation increases the electrostatic energy. Inter-dot
tunneling destroys charge quantization on individual dots and allows the charge
distribution to reach its optimal configuration. This picture is quantitatively
confirmed by comparison of theory with linear-response measurements on
series and parallel coupled quantum dots and nonlinear current-voltage
measurements on series coupled dots.
4.3. SPECTROSCOPY WITH COUPLED DOTS.
Single quantum dots display well defined discrete energy states as
described in sections 2 and 3. For semiconductor dots with Fermi energy EF lOmeV and N - 100 electrons the separation between the OD-states is L1E 100 j..leV which exceeds kBT for T < 1 K. The values of energy states are of
interest for specific quantum dot structures and their statistics are of interest for
comparison with quantum chaos theory (see section 6). The resolution of
tunnel measurements of the OD-states in single dots is limited by thermal
broadening of the Fermi energy in the leads.
Van der Vaart et al. [103] have demonstrated an improved spectroscopic
technique using OD-states in the second dot of a coupled dot structure to
remove the effects of thermal broadening in the leads. Fig. 4.8(a) shows the
coupled dot device, which is also defined in the 2DEG of a GaAslAIGaAs
heterostructure. The dots are a few times smaller than the ones discussed in
section 4.2 such that here the OD-states are well resolved. In contrast to the
experiments in section 4.2, the tunneling rate between dots is made very weak,
145
Figure 4.8. (a) SEM micrograph of the double dot with lithographic dimensions of 320 x 320
nm2 (left dot) and 280 x 280 nm2 (right dot). (b) Schematic potential landscape of the double
quantum dot, where J.Lle!t and J.Lright denote the electrochemical potentials of the left and right
reservoirs and V the bias voltage across the double dot. The OD-states in dot I are denoted by
levels I to 5 and in dot II by levels a and /3. (From van der Vaart et ai. [103].)
Double dot
o
012
VOLTAGE (mY)
Figure 4.9. /- V curve of the double dot, showing sharp resonances in the current when two ODstates line up. Upper inset: /- V curve of dot I. Lower inset: /- V curve of dot II. Both insets
show a suppression of the current at low voltages due to Coulomb blockade and a stepwise
increase of the current due to the discrete energy spectrum of a single dot. (From van der Vaart et
ai. [103].)
146
as well as tunneling to the leads. The lifetime in each dot is therefore long and
the OD-states have a broadening much smaller than their separations.
Although, the Coulomb blockade also exists for both dots, we here describe
only the additional influence of the OD-states. As indicated in Fig. 4.8(b)
transport through the series double dot begins as an electron tunnels in from the
left lead to the OD-state 3 in the left dot. If one of the unoccupied states of the
right dot, state a in (b), lines up with state 3, then the electron can tunnel
elastically to the right dot and eventually leave through the right barrier. So,
even if the conditions for Coulomb blockade are fulfilled, if none of the ODstates in the right dot line up with a OD-state in the left dot the tunneling rate is
strongly suppressed. Thus transport through the dot should show sharp features
limited only by the lifetime of the single-particle states, not the temperature.
The J- Vsd curve in Fig. 4.9 demonstrates how narrowing occurs. The two
insets show J- Vsd traces for dots I and II individually; both show steps which
occur when the Fermi level in the lead passes through a OD-state. The average
spacing between levels is found to be L1E = 125 ~V for dot I and L1E = 225
~V for dot II. When the gates forming both dots are energized, the series
double dot shows a very different J- Vsd characteristic. Instead of steps we now
see sharp peaks which occur at source-drain voltages for which a OD-state in
the right dot and one in the left dot line up; the current is suppressed at other
source-drain voltages. Analogous structure is seen in traces of conductance vs.
gate voltage at finite Vsd, for which each conductance peak breaks up into a
series of peaks, each associated with a specific pair of OD-states in the left and
right dots. Fitting of one of these resonances shows that the lineshape is
Lorentzian with an additional broadening corresponding to a temperature of 35
mK [103]. The fact that this temperature is several times smaller than the
electron temperature in the leads demonstrates that the discrete states in one dot
act to remove thermal broadening due to the leads [86,122].
4.4. COHERENT TUNNELING THROUGH QUANTUM DOTS
The coherence of electronic states becomes an important issue whenever wavelike aspects of electrons are important. In the early 1980's researchers were
surprised to find that electron waves traveling through disordered metallic
conductors exhibited wave interference phenomena at low temperatures, in the
form of weak localization and universal conductance fluctuations, over length
scales much greater than the elastic scattering length [123-125]. This
interference demonstrated that coherent transport can occur in a Fermi liquid
even though the particles interact strongly via Coulomb forces. A considerable
body of experimental and theoretical work led to a detailed quantitative
understanding of coherence in disordered conductors. Quantum dots also show
147
conductance fluctuations and analogs of weak localization, which are

associated with coherent interference inside the dots (see section 6), but our
understanding of the coherence times is not yet as well developed as for open
structures.
The issue of coherence is critical for applications of quantum dots to
possibly construct quantum computers [87,88]. The theoretical speed of
quantum computers follows from the fact that they coherently process
superpositions of wavefunctions, and the original approaches required that
processing remain coherent across the computer during the entire calculation.
The resulting experimental constraints are truly daunting for useful calculations
and threaten to place the implementation of quantum computers as hardware far
into the future. Recent advances in quantum error correction relax these
requirements and create hope for somewhat nearer-term implementations.
Tunneling in experiments, especially in the weak tunneling regime, is
generally considered to be an incoherent process. Theoretically this need not
be true, but the electrons in quantum dots interact with many other particles:
charged impurity atoms, electrons in surrounding gates, phonons in the
substrate, and photons from the environment, for example. If the dynamics of
these particles are not included in the description of the many-body system
which comprises the quantum dot (or the quantum computer), then the
interaction between the electrons on the dot with these "external particles"
effectively destroys the coherent behavior of the electrons on the dot.
Currently, the issue of coherence of electron states inside quantum dots and the
coherence of tunneling processes is actively discussed but relatively few data
are available [36,82].
Recent experiments [126,127] have made an important step in our
understanding of coherence for tunneling experiments on quantum dots by
demonstrating that part of the tunnel current is coherent. Figs. 4.1O(a) shows
an SEM photograph of the device of Yacoby et al. [126], which consists of a
GaAs/AIGaAs quantum dot '.vith two adjustable quantum point contacts placed
inside a ring which forms two arms of an electron interferometer. The quantum
dot was placed in the tunneling regime with an electron temperature -80 mK.
At this temperature for this size dot, tunneling normally takes place through a
single quantum level of the dot. The coherence length of electrons in the ring
was larger than its circumference, and interference between electron waves
tunneling through the dot and passing along the open arm of the interferometer
was detected via Aharonov-Bohm (AB) oscillations in the ring current as a
perpendicular magnetic field B was swept.
Coherence in the tunneling current is demonstrated by the data of Fig.
4.1 O(b), which is a trace of the ring current vs. plunger gate voltage Vp for
148
(b)
Figure 4.10. (a) An SEM micrograph of an Aharonov-Bohm ring with a quantum dot placed in
the left arm. The light regions are the metal gates. The central metallic island is biased via an air
bridge (B) extending to the right. (b) One of the ring's current peaks as a function of the gate
voltage applied to P in (a). At the top left the current is plotted as a function of magnetic field
(the magnetic field increases to the left) at the fixed gate voltage Vm showing Aharonov-Bohm
oscillations. The large current of the right arm is subtracted. (From Yacoby et al. [126].)
small source-drain voltage. The Coulomb blockade for the dot creates sharp
conductance peaks in the ring current for fixed magnetic field. When the gate
voltage is fixed on the side of a conductance peak and the magnetic field is
swept instead, periodic AB-oscillations are observed, with a period
corresponding to the ring area, demonstrating the existence of a coherent
component of the current tunneling through the dot. The phase of the ABoscillations was the same for different Coulomb blockade conductance peaks,
but changed abruptly at the top of each peak as expected for a resonant
phenomenon like tunneling through a single level.
The observation of coherence in tunneling through a single dot is an
important step, but many issues remain open in the study of coherence in single
and coupled quantum dots and quantum dot circuits. These include a
quantitative understanding of the coherence time of electron states inside single
dots and an understanding of coherence of molecular electronic states inside
tunnel-coupled quantum dots and quantum dots circuits. This knowledge will
permit reasonable assessments of the feasibility of quantum computation using
coupled dot circuits in the future.
149
5. Vertical quantum dots.

This section deals with the transport properties of a single vertical
semiconductor quantum dot which is weakly coupled to the reservoirs via
heterostructure tunnel junctions. We focus on Coulomb interactions and
quantum mechanical effects, both of which become enhanced in a small dot
containing just a few electrons (N:::; 20). First, in section 5.1, we describe the
vertical dot tunnel structures and the technique of single electron spectroscopy,
while section 5.2 addresses atom-like properties observed in a quantum dot
with a two-dimensional harmonic potential.
5.1. SINGLE ELECTRON SPECTROSCOPY OF A FEW-ELECTRON DOT.
Coulomb oscillations are usually periodic. However, in a small dot containing
just a few electrons, both the electron-electron interactions and quantum
confinement effects become sufficiently strong that the spacings between the
Coulomb oscillation peaks become irregular [50,128-131]. Lateral dot
structures, discussed in sections 3 and 4, are not suitable for defining a system
of only a few electrons because the tunnel barriers are formed by a depletion
potential. This potential is significantly affected by the center or plunger gate
potential close to 'pinch-off'. In practice, when N < -25 the tunnel barriers
become too large for observing a current, even at resonance. In vertical dot
structures, this problem is overcome by the use of heterostructure barriers.
These tunnel barriers are abrupt and thin, and are only weakly affected by a
gate potential. Furthermore, the lateral geometry in a vertical dot is sufficiently
well defined to allow for the formation of systematically degenerate sets of ODstates in the dot [31,132]. However, there are inherent technical difficulties in
squeezing the vertical dot, i.e. varying the number of electrons. To date there
only have been a few reports of how these difficulties have been solved. These
include transport through two-terminal dots [133-137], transport through threeterminal gated dots [130,131,138-142], and capacitance measurements on twoterminal gated dots [15,50].
Fig. 5.1 shows a schematic diagram of a two-terminal quantum dot
structure and the Coulomb staircase observed by Schmidt et al. [136]. The
GaAs dot is located between the two AIGaAs barriers. The two barriers have
different widths, which promotes the accumulation of electrons in the dot for
negative Vsdand the escape from the dot for positive Vsd [51,143]. For negative
Vsd a current step occurs whenever an extra electron is trapped in the dot, while
for positive Vsd a sharp current peak appears due to resonant tunneling through
a single-particle OD-state [133-137,144,145]. Since the dot is depleted (N = 0)
at Vsd = 0 the first step occurs when Vsd is increased sufficiently such that the
150
c:
-0,5
-0,4
-0,3
-0,2
-0,1
0,0
-40
-SO
-60
V (mV)
Figure 5.1. (a) Schematic diagram of a two-terminal quantum dot device and (b) the Coulomb
staircase in the current vs. source-drain voltage at B = 0 T. The fabricated dot diameter is 0.35
f1m, but due to depletion effects the electrons experience a dot which is several times smaller.
(From Schmidt et ai. [136].)
~ r-----------------------------,
(b)
O ~~~~~~~_1~
~~~~~~~
GATE VOlTAGE M
Figure 5.2. (a) Schematic diagram of the gated quantum dot device and (b) the Coulomb
oscillations in the current vs. gate voltage at B = 0 T observed for a 0.5 J1m diameter dot. (From
Tarucha et al. [131].)
10
-1 0
sourcodrain Vollage (m V)
Figure 5.3. Measured size of the Coulomb gap versus gate voltage. The diamonds represent the
source-drain voltage value at which current starts to flow. The sizes of the diamonds directly
correspond to the peak spacings in Fig. 5.2(b). (From Kouwenhoven et al. [146].)
151
electrochemical potential of one of the reservoirs aligns with the N = 1 energy

state. At Vsd = -40 mV in Fig. 5.1(b) the first electron enters the dot. The
subsequent steps each represent the addition of one more electron. This
charging process is in the nonlinear transport regime (see section 3.3). From
the voltage interval Ll Vsd between the Nth and (N+ J)th steps, an addition energy
needed to put an extra electron in the N-electron dot, Llj1, can be deduced since
Llj1 = f3eLl Vsth where f3 < 1 is a factor to convert source-drain voltage to energy.
The large values of Ll Vsd, compared to lateral dots, arise from the strong
Coulomb interactions and quantum mechanical effects in the few-electron
system.
Fig. 5.2 shows a schematic diagram of a vertical dot and the Coulomb
oscillations observed at B = 0 T [131]. This dot has a circular Schottky gate
placed on the side of the mesa close to the dot region, and is used to squeeze
the dot. The dot is made from InGaAs, which has a narrower band gap than the
GaAs in the contacts. The inclusion of In reduces the conduction band bottom
in the dot below the Fermi level in the contacts. This means that electrons are
accumulated in the dot even when no voltages are applied. This allows to study
linear transport properties. (Note that for the device in Fig. 5.1 both the dot and
contacts are made from GaAs.) The current J vs. gate voltage Vg measured at
small Vsd shows clear Coulomb oscillations.
In Fig. 5.3 the measured size of the Coulomb gap (i.e. the source-drain
voltage at which current starts to flow) is plotted as a function of gate voltage
[146]. The vertical Vsd = 0 axis corresponds directly to the data in Fig. 5.2(b).
Note that, for instance, the size of the peak spacings along the Vsd = 0 axis are
the same as the peak spacings for the corresponding N in Fig. 5.2(b). These
irregular peak spacings at Vsd = 0 lead to correspondingly irregular sizes for the
diamond shaped Coulomb gaps. This is in contrast to the regularly shaped
diamonds in Figs. 3.7(b) and 4.7, which were measured on lateral dots. As we
discuss in the next section these irregularities can be described to the small
number of electrons in vertical dots. Note that the Coulomb diamond in the N
= 0 region never closes when we continue to make the gate voltage more
negative, implying that the dot is indeed empty. From this observation the
absolute values of N can be identified and used to label the spaces between the
current peaks.
Single electron effects are observed also in ac capacitance measurements
[15,50]. Fig. 5.4 shows a schematic diagram of the vertical dot structure and
the capacitance signal observed by Ashoori et al. [15]. The dot is located in the
GaAs quantum well and its size is controlled by the top gate voltage. Electron
exchange occurs only through the barrier to the bottom electrode. The ac
capacitance measurement shows a sequence of peaks as a function of top plate
voltage. Note that although this is a two terminal device, the capacitance
152
Top plate voltage (mV)
Figure 5.4. Schematic diagram of the dot structure prepared for capacitance spectroscopy, and
the capacitance data vs. gate bias (top plate voltage) characteristic at B = 0 T. The dot is the
black disk between the two AIGaAs barriers. (From Ashoori et ai. [15].)
measurements are in the linear response regime. The first capacitance peak on
the left represents the first electron entering the dot. Thus, the Nth peak
position in top plate voltage is directly related to Jl(N). The capacitance peaks
are more widely spaced as the dot approaches the "pinch-off' point. This also
reflects the increase in both the Coulomb interactions and the quantum
mechanical effects when N approaches zero. We discuss the B-field
dependence of the capacitance signal in section 7.
5.2. SHELL STRUCTURE AND HUND'S RULE.
If a dot has the shape of a circular disk and the confining potential is harmonic,
then we have a system with a high degree of symmetry. This symmetry leads
to sets of degenerate single-particle states which form a shell structure. Such a
shell structure for circular 2D potentials was predicted by self-consistent
calculations [147]. Here, we discuss the formation and electron filling of a
shell structure by comparing a non-interacting model with models that include
Coulomb interactions and exchange interactions.
153
N= 2
.. 10.0
(b)
+- +- +(0.2)
(1.0)
(0, I) f i s
(.. I}-(o.o)
(0.2)
~(O" I )
~
Q)
5 .0
10
15
20
Figure 5.5. Addition energy vs. electron number. (a) Hartree-Fock calculation for a circular dot
with harmonic lateral potential (filled circles). The open circles show the single-particle
excitation energy, and the open triangles give the charging energy from a classical selfcapacitance modeL (From Tamura et al. [148].) (b) The filled circles are the measured addition
energies derived from the Coulomb oscillations in Fig. 5.2. The inset depicts the filling of the
shells for N = 9 in line with Hund's rule.
Fig. 5.5(a) shows a comparison between energies of non-interacting

single-particle states, classical charging energies derived from a selfcapacitance model, and those derived from a Hartree-Fock (HF) calculation.
The results are shown in the form of addition energies with respect to N [148];
see Eq. (2.2) for the addition energy in the constant interaction modeL For a
circularly symmetric 2D confinement potential the quantum numbers for the
single-particle states are conveniently described by a radial quantum number n
= 0, 1,2,. and an angular momentum quantum number f = 0, 1, 2,.. We
consider a harmonic confinement potential V( r) = lhm' rool for which the noninteracting Schr6dinger equation has the following analytic solutions for the
eigenergies Enl :
(5.1)
The calculation in Fig. 5.5(a) is performed by taking nOJ o = 2Ry* (= 11.5 meV
for the InGaAs dot in Fig. 5.2), where Ry* is the effective Rydberg constant
which is a measure for the interaction energy. It follows from Eq. (5.1) that Enl
has degenerate sets of states, which are separated by nOJ0 from each other and
which are completely filled for N = 2, 6, 12, 20, etc. For these special values of
N the addition energy is maximaL These maxima persist when interactions are
included in a HF approximation. However, in addition, the HF calculation also
154
reveals maxima in the addition energy at N = 4, 9 and 16. For these N values,
respectively, the second, third and fourth shells are half filled with parallel
spins in accordance with Hund's rule. Half filled shells correspond to a
maximum spin state, which, due to exchange interaction, has relatively low
energy. These atom-like features persist as long as nwo is comparable to, or
larger than, the interaction energy.
Fig. 5.5(b) shows the addition energy as a function of N obtained from the
data shown in Fig. 5.2. The spacing between the Nth and (N+l)th current
peaks reflects the energy to add one more electron to a dot containing N
electrons. For example, the energy to add the third electron to an N = 2 dot can
be derived from the spacing between the second and third peak. For each value
of N a factor to convert gate voltage to addition energy (i.e. the factor eCIC in
Eq. 2.1) can be determined from the Coulomb diamonds in Fig. 5.3. The
addition energy generally becomes larger as N decreases and has large maxima
for N = 2, 6 and 12, and also relatively large maxima for N =4,9 and 16. This
N-dependence of the addition energy is consistent with the HF calculation in
Fig. 5.5(a) and reflects the complete filling of the first, second and third shells
as well as the half filling of the second, third and fourth shells with parallel
spins. The inset in (b) shows schematically the occupation of the lowest three
shells for N = 9. The actual value of the addition energy is smaller than that in
the HF calculation, because nwo derived from the experiment (see below) is
smaller than that used in the calculation
The electronic states are expected to be significantly modified by a
magnetic field, B, applied parallel to the tunneling current. Fig. 5.6 shows the
B-field dependence of the position of the current peaks. The positions of the
first three peaks depend monotonously on B, whereas the other peaks oscillate
back and forth a number of times. The number of "wiggles" increases with N.
Close inspection of the figure reveals that the current peaks generally shift in
pairs with B. The spacing between the peaks when N is odd is nearly constant,
whereas the peak spacing varies strongly with B when N is even. This evenodd effect is particularly clear around 3.5 T, where all the peaks are evolving
smoothly with B. Here the peak spacing alternates between "large" for even N
and "small" for odd N. Most of these features can be explained within a singleparticle framework.
If we assume a parabolic confinement, the non-interacting SchrMinger
equation can be solved in the presence of a magnetic field. This calculation
was first performed in 1928 and the eigenenergies are known as Darwin-Fock
states [149,150]:
(5.2)
155
-0.4
-r---- -
I-'""
=--
-- --- -
~
~
i"""
--
r--
18
16'
14
12
10
f-
r-
8
6
r-4
2
-1.7 0
.
1
2
Magnetic field (T)
N=O
Figure 5.6. Plot of the gate voltage positions of the current peaks in Fig. 5.2 vs. magnetic field,
(From Tarucha et al. [131].)
'"c
..
W6
>-
Cl
c 3
W
(0,3)
II)
N=O
~(Do =3 meV
(n,1) =(0,0)
Magnetic field (T)
limo
Figure 5.7. Calculated single-particle energy vs. magnetic field for a parabolic potential with
= 3 meV. Each state is two fold spin degenerate. The dashed and dot-dashed lines are discussed
in the text. (From Tarucha et al. [131].)
156
where tUJJ c = lieBlm' is the cyclotron energy. Fig. 5.8 shows Enl vs. B
calculated for hID0 = 3 meV. Spin is neglected so each state is twofold
degenerate. The degeneracies at B = 0 T are lifted in the presence of a B-field
such that a single-particle state with a positive or negative f shifts to lower or
higher energy, respectively. The B-field dependence of these states gives rise
to an addition energy for even N that is strongly dependent on B. On the other
hand, the addition energy for odd N is determined only by the effect of
Coulomb repulsion. This leads to a pairing of the conductance peaks, which is
evident in Fig. 5.6. In Fig. 5.7 the energy curve for the fifth and sixth electrons
(dot-dashed line) predicts that these electrons undergo a transition in their
quantum numbers: (n,) changes from (0,-1) to (0,2) at 1.3 T. The energy
curve for the seventh and eighth electrons (dotted line) predicts that these
electrons undergo transitions in (n,) from (0,2) to (0,-1) at 1.3 T and then to
(0,3) at 2T. These transitions are also seen in Fig. 5.6. However, one should
keep in mind that the charging energy separates the non-interacting states of
Fig. 5.7 by a value which is roughly constant in magnetic field. In a similar
fashion, the quantum numbers can be identified for the other electron states.
Above the B-field where the last "wiggle" occurs, the single-particle states
merge into the lowest spin-degenerate Landau level (i.e. for n = 0, f :? 0). The
single-particle excitation energy calculated, for example, at B = 3.5 T, is still
large (between 1 and 1.5 meV in Fig. 5.7) and significantly contributes to the
addition energy for even N. This leads to the alternating peak spacings
observed around 3.5 T in Fig. 5.6 (i.e. the spacings where the even numbers
have been placed are larger than the neighboring spacings).
Fig. 5.8(a) shows the B-field dependence of the third, fourth, fifth and
sixth current peaks, i.e. peaks belonging to the second shell. The pairing of the
third and fourth peaks, and the fifth and sixth peaks above 0.4 T is clearly seen.
However, this pairing is rearranged for B < 0.4 T. In this region the third and
fifth peaks, and the fourth and sixth peaks are paired. The evolution, as a pair,
of the third and fifth peak for B < 0.4 T is continued by the third and fourth
peak for B > 0.4 T. Similarly, the evolution, as a pair, of the fourth and sixth
peak for B < 0.4 T is continued by the fifth and sixth peak for B > 0.4 T. For B
> 0.4 T, following the arguments related to Fig. 5.7, the third and fourth peaks
are identified by the quantum numbers (n,f) = (0,1) with anti-parallel spins.
The fifth and sixth peaks are identified by (n,f) =(0,-1) with anti-parallel spins.
The rearrangement of the pairing for B < 0.4 T can be understood in terms of
Hund's rule, which is well known in atomic physics [151]. Hund's rule says
that degenerate states in a shell are filled first with parallel spins up to the point
where the shell is half filled. This is modeled in the calculation of /l(N) vs. B
shown in Fig. 5.8(b). In this figure the quantum numbers (n,f) help to identify
the angular momentum transitions, and the diagrams illustrate the spin
157
(b)
ffijDJ
II
(ij
Gl
0
~L'.
(II
'0
>
'E
E -1.3
Q)
(0,2)
N=5
N=6
(0,2)
N=4
~L'.
co
C!>
(0,1)
...0
tEP
t)
Gl
iii
-1.4
0.0
Magnetic field
0.5
N=3
1.0
(0,1)
1.5
2.0
Magnetic field (T)
Figure 5.8. (a) Evolution of the third, fourth, fifth and sixth current peaks with magnetic field
from 0 to 2 T. The original data consists of current vs. gate voltage traces for different magnetic
fields, which are offset and rotated by 90 degrees. (b) Calculated electrochemical potential vs.
magnetic field for the model described in the text and parameters U = 3 meV, L1 =0.7 meV, and
!t0J0 = 3 meV. The quantum numbers (n, f ) are shown of the Nth electron and the diagrams show
the spin configurations. (From Tarucha et al. [131].)
configurations. In the constant interaction model, [see Eqs. (2.1) and (2.2)]
/1(N) can be written as a constant interaction energy U added to Enl [152,153].
To include Hund's rule in the calculation we introduce an energy L1, which
represents the energy reduction due to the exchange interaction between
electrons with parallel spins. Specifically, for N = 4, the ground state energy is
reduced if the outer two electrons have parallel spins with different angular
momenta rather than anti-parallel spins with the same angular momentum.
/1(4) is thus reduced by an amount L1 and there is a corresponding increase in
/1(5) by L1. This exchange effect is canceled in the presence of a B-field when
the (0,1) states, which are degenerate at B = 0 T, are split by an energy
exceeding L1. This is a simple way to include exchange effects in a constant
interaction model. However, for small N we find a remarkable agreement
between what is seen in Fig. 5.8(a) and that predicted in (b) with U = 3 meV
and L1 = 0.7 meV. In this model, the addition energy for N = 4 at B = 0 T is
expected to be larger by 2L1 than that for N = 3 and 5, and this is indeed
observed in Fig. 5.5(b). This simple Hund's rule model is a first correction to
the constant interaction model. A more rigorous Hartree-Fock approach, or
exact diagonalization of the N-electron Hamiltonian, as outlined in Refs. 73-75,
154-159, are required for a more quantitative comparison. Very recently Eto
[160] has actually been able to calculate a B-field dependence of the addition
158
spectra that very closely duplicates the data in Fig. 5.9. These calculations thus
confirm the simple model of Fig. 5.8(b).
In this section we have described that the linear transport characteristics
through a 2D artificial atom reflect a shell structure, and the filling of electrons
is in line with Hund's rule. The atom-like energy spectrum of the dot states is
systematically modified by a magnetic field, which allows the identification of
the quantum numbers of the single-particle states. Note that the observation of
orbital degeneracy implies that the system is non-chaotic, which is very
unusual for solid state systems. In fact, on the level of single-particle states the
vertical dots are the first non-chaotic solid state devices. In the next section we
return to lateral dots and discuss chaos in dots without symmetry.
6. Mesoscopic Fluctuations of Coulomb Blockade.

This section concerns the rather specific subject of meso scopic fluctuations of
conductance in the Coulomb blockade regime. After briefly reviewing
universal conductance fluctuations in open quantum dots (6.1), we tum to
discuss the newer and experimentally more challenging problem of mesoscopic
fluctuations of Coulomb blockade peak heights (6.2 and 6.3), peak positions
and spacings (6.4 and 6.5), and elastic co-tunneling in the valleys between
Coulomb blockade peaks (6.6). At the end, some open questions and
conclusions are given (6.7). Whereas transport in open quantum dots with
highly conductive leads can be described in terms of quantum interference of a
non-interacting electron traversing the dot via multiple diffusive or chaotic
paths, in nearly isolated, Coulomb blockaded quantum dots, interactions have a
dominant role in transport, coexisting with large non-periodic fluctuations due
to quantum interference.
Nonetheless, many experimental aspects of
mesoscopic fluctuations in blockaded dots can be understood quantitatively
within the constant interaction model where fluctuations arise from the spatial
structure of single-particle OD states.
6.1. CONDUCTANCE FLUCTUATIONS IN OPEN QUANTUM DOTS.
Mesoscopic conductance fluctuations typically refer to quasi-random
fluctuations of the conductance of small open conductors with large
conductance G > e2/h [161-163]. These fluctuations are ubiquitous at low
temperatures when the size of the system L becomes comparable to the phase
coherence length rti,.1), which can grow to several microns at temperatures
below - 1 K. Mesoscopic fluctuations are distinct from noise in that they do
not depend on time, but rather depend on experimentally controllable
159
parameters such as magnetic field or gate voltage. Because these parameters

can be swept back and forth, it is readily seen that although the fluctuations are
unpredictable, like noise, they are perfectly repeatable within a single cooldown of the device - a striking instance of deterministic randomness in a
quantum system. An example of mesoscopic fluctuations in an open quantum
dot is shown in Fig. 6.1.
Conductance fluctuations in open conductors can be understood as
interference of phase-coherent electrons traversing the sample via a number of
interfering paths. The influence of an external magnetic field is to alter the
relative phase of the various interfering paths, scrambling the interference
"speckle" pattern and thus causing the conductance to fluctuate in a
complicated, essentially random way. Less obvious from this trajectory-withphase picture is the fact that these fluctuations exhibit universal statistical
properties. For instance, measured in conductance their magnitude is always of
order e2/h, independent of material and the average conductance, giving them
their name: "universal conductance fluctuations" (UCF) [164-167]. A vast
theoretical effort over the past decade has shown that the universal aspects of
mesoscopic phenomena are associated with universal spectral properties of
random matrices [168-172] as well as the universal properties of the quantum
manifestations of classical chaos [173-175].
UCF in disordered metals and semiconductors has been widely
investigated over the last fifteen years (for reviews of the experimental
literature, see [163,35]). More recently, experiments in high-mobility GaAs
quantum dots have shown that gate-confined ballistic structures (i.e. devices in
which the bulk elastic mean free path exceeds the size of the dot) also exhibit
UCF. This ballistic UCF is similar to UCF in disordered systems [163,176178] with the same universal statistics [170,172] as long as several conducting
channels per lead are open, so that G dot > e2/h. The applicability of UCF
concepts to ballistic quantum dots draws particular attention to the fact that
disorder is not a requirement for UCF, but is only one means of generating the
universal features of quantum transport. The universality of UCF applies
whenever, but only when, the quantum dot has an irregular shape that gives rise
to chaotic scattering from the walls of the device. Fortunately, this chaoticshape condition is easily met in practice; with sufficiently large number of
electrons (N > -50) nearly any irregular shape will yield chaos at sufficiently
low magnetic field. The non-chaotic character of the vertical dots discussed in
section 5.2 is possible since they contain a small number of electrons. (Here,
we sidestep the fact that in classical dynamics the generic situation is a mixed
phase space, with some trajectories executing regular motion and others chaotic
motion. A mixed phase space can lead to fractal conductance fluctuations
[180]).
160
3.7
3.6
;? 35
NO
0 3.4
3.3
3.2
3.1
(c)
-20
-10
B (mT)
10
20
Figure 6.1. (a) Chaotic classical trajectory entering, then bouncing within, and finally exiting an
open stadium billiard. The semiclassical model of transport accounts for all such classical
trajectories and includes phase interference between trajectories. Signatures of classical chaos
therefore appear in the quantum transport. (b) A quantum wave function for the open stadium
billiard (the size compared to AF corresponding to the stadium device in Ref. [176], suggests an
alternative approach more applicable to nearly-isolated devices, in which transport is described
by the coupling of the dot wave function to the electron states in the leads. (From Akis and Ferry
[181].) (c) An example of experimental conductance fluctuations in an open (N - 3) quantum dot
from the device used in Ref. 179.
Much of the recent progress in understanding UCF and other mesoscopic

effects such as weak localization in quantum dots has been gained through the
application of random matrix theory (RMT) [182]. (A powerful alternative
approach based on supersymmetry has also provided many breakthrough
results [183].) To treat the case of an open quantum dot with two leads, each
transmitting M channels per lead, one introduces a scattering matrix of the
form:
=(r r'tl)
t
(6.1)
where r,t and r' ,t' are M x M complex reflection and transmission matrices for
particles approaching the dot from the right or left, respectively. In order to
apply RMT, the matrix S is assumed to be as random as possible given the
physical constraints of the system, which are that S be unitary, sst = I, in order
to conserve current (all particles must be either transmitted or reflected), and S
is symmetric (S = ST) for the case of time-reversal symmetry (B - 0), or
161
Hermitian (S = st) for broken time-reversal symmetry (IBI > - tPO/L2). We will
not discuss the case of strong spin-orbit scattering, which introduces additional
structure to S [182]. The conductance G of the dot can be related to S through
the Landauer formula:
(6.2)
Through Eq. (6.2), the statistics of conductance fluctuations may be related to
the spectral statistics of the random scattering matrix if we assume that changes
to the impurity configuration or external parameters applied to the dot are
equivalent to choosing another member of the random matrix ensemble. From
this point of view, generic statistical properties of random matrices [184], in
particular, level repulsion and spectral rigidity, can be seen to be intimately
connected to the universal statistics observed in transport through disordered or
chaotic dots [167,169,171,185]. (For a collection of articles on this subject, see
Ref. [162]). The harder question, of course, is why the random matrix
assumption should work at all in describing even single-particle transport, let
alone transport in a strongly interacting electron liquid? Without attempting an
answer, we simply note that while the rough connections between UCF and
RMT, and between RMT and quantum chaos have been appreciated since the
early days of meso scopic physics, a rigorous theoretical web tying these
subjects together has emerged only in the past year or so. The reader is
referred to Refs. [183,186,187] for discussions of this fascinating subject.
From an experimental viewpoint, it seems miraculous that such an abstract
approach succeeds in quantitatively describing quantum transport in real
materials.
6.2. FLUCTUATIONS OF COULOMB BLOCKADE PEAK HEIGHTS.
The random scattering matrix approach described above applies to conductance
fluctuations in open quantum dots. When the leads form tunnel barriers with
low conductance, Gleft,Gright < e2/h, Coulomb blockade appears at moderately
low temperatures, kBT < -Ee. For lower temperatures, kBT < L1E, discrete ODstates are resolved and conduction is mediated in this case by resonant
tunneling through the quasi-bound state of the dot, which is lifetime-broadened
by hT. In this regime, conductance fluctuations as large as the average
conductance itself will result as the electron states in the leads couple better or
worse to the quasi-bound state of the dot, as shown in the numerical results of
Fig. 6.2. For disordered or chaotic-shaped quantum dots, conductance
fluctuations in the resonant tunneling regime appear random, as seen in Fig. 6.2
162
1.0
/I
~l. 1I
~IJ
1.5
J ~
2 .0
Figure 6.2. Numerical calculation of resonance conductance geE) for a disordered and
desymmetrized stadium billiard with single-mode tunneling leads, as a function of Fermi energy,
E = EF . Resonances are fits to Lorentzian lineshapes, and have large amplitude fluctuations due
to the coupling of the wave function to the leads. (From Jalabert, Stone, and Alhassid [188].)
but with different statistical properties than UCF in open dots. In this case, the
origin of the random fluctuations can be understood as resulting from the
spatial structure of the quasi-bound wave function, In particular, the amplitude
of the wave function in the vicinity of the leads determines the fluctuations,
rather than the spectral properties of the scattering matrix. These differing
views of the origin of meso scopic fluctuations can be reconciled by the socalled R-matrix formalism, originally developed to address similar problems in
compound nuclear scattering. R-matrix theory relates the Hamiltonian of the
isolated system to the scattering matrix of the corresponding open system.
The effects of finite temperature and charging energy can be readily
accounted for in the quantum Coulomb blockade regime, hT < kBT < i1E
e2/C. Recalling the discussion in section 2, Coulomb blockade conductance
peaks in this regime are approximately uniformly spaced in gate voltage Vg ,
and have a thermally broadened line shape,
(6.3)
where 8 = e(C/C)lVg,,.. - Vgl. The peak height Gmax is related to the tunneling
rates through the left and right leads, I/eft and T,.igh" by:
G
max
= ~( r 1eft rr;ght
4k B T r 1eft + rr;ght
J== 4ke rT
2
(6.4)
163
where
(6.5)
is a dimensionless peak height and hT = hI/eft + hrright is the total lifetime

broadening of the OD-state.
An RMT approach to fluctuations in peak height was developed by
Jalabert, Stone and Alhassid [188] based on the assumption that the OD-states
of the dot can be described as typical large-quantum-number eigenstates (1fIn
with n 1) of a quantum chaotic system, and thus can be characterized by an
RMT which is appropriate to the symmetry of the system. In this case, it is the
Hamiltonian of the isolated dot rather than the scattering matrix that is modeled
as a random matrix. The required symmetry of the random matrix ensemble is
again confined to two classes (in the absence of strong spin-orbit scattering):
symmetric matrices for B = 0, when the system obeys time-reversal symmetry
(the ensemble of such random matrices is known as the Gaussian orthogonal
ensemble, or GOE, because of the invariance of the spectrum under orthogonal
transformation) or Hermitian matrices for B"* 0 (Gaussian unitary ensemble, or
GUE).
The resulting model of transport in the quantum Coulomb blockade regime
closely resembles the statistical theory of compound nuclear scattering, with
peak height distributions analogous to Porter-Thomas distributions of
resonance widths. The assumption that the overlap integrals of the wave
functions in the dot with the wave functions in the lead are Gaussian distributed
implies that the tunneling rates into and out of the dot Iieft and Fright (which are
proportional to the square of the overlap) obey Porter-Thomas statistics, that is
x3 distributed with v = 1 degree of freedom for GOE and v = 2 for GUE. If
one further assumes that the leads are statistically independent (valid when
their separation greatly exceeds AF ) and have the same average tunneling rates,
F1eft = Fright = F I 2 , the distribution of dimensionless peak heights P( a) have the
following forms, depending only on the presence or absence of time-reversal
symmetry:
(6.6a)
~B;tO) (a) = 4a[ Ko (2a) + Kl (2a)] e-2a
(6.6b)
164
5
(a)
C?3
(b)
C?3
.......
.......
~2
~2
1
ex
0
0.0 0.5 1.0 1.5 2.0
ex
Figure 6.3. Numerical distributions of the dimensionless conductance ex defined in Eq. (6.5),
based on resonance data similar to that in Fig. 6.2, though for a different billiard shape (the
Robnik billiard), along with RMT results. (a) For the case of time-reversal symmetry, with zero
magnetic flux; solid curve is the GOE result, Eq. (6.6a). (b) For the case of broken time-reversal
symmetry, with an applied magnetic flux; solid curve is GUE result Eq. (6.6b). (From Bruus and
Stone [192].)
where Ko and KJ are modified Bessel functions [188,189]. Note that the
average peak height in zero and nonzero field are different, !aP(B=ola)da = 114
and !aP(B'rola)da = 113. This effect is related to weak localization in open
mesoscopic systems. The above results have been extended using both RMT
and nonlinear sigma-model approaches to include nonequivalent, multi-mode,
and correlated leads [189-191], dot shapes undergoing distortion across the
transition from integrable to chaotic classical dynamics [192], and partially
broken time-reversal symmetry [193,194]. In each case the results were found
to agree well with direct numerical simulations of tunneling through chaotic
dots. These numerical studies of peak height fluctuations are based on a noninteracting picture of electronic wave functions in confined hard-wall 2D
chaotic cavities with tunnel-barrier leads. An example comparing numerics to
RMT, Eq. (6.6), is shown in Fig. 6.3 [192].
We now discuss the experiments. Earlier measurements of transport
through blockaded, gate-confined quantum dots demonstrated significant
height fluctuations among Coulomb blockade peaks at low temperatures and
low magnetic fields [80,82,195], as seen for instance in Fig. 3.3. These
fluctuations were not the main focus of these works and were not studied in
detail. Recently, two groups have directly checked the RMT predictions, Eq.
(6.6), using gate-defined GaAs quantum dots [196,197]. Representative series
of peaks showing large height fluctuations as a function of gate voltage are
shown for the data of Chang et al. [196] in Fig. 6.4(a) and Folk et ai. [197] in
Fig. 6.4(b). Both experiments found excellent agreement with the RMT
predictions, as shown in Fig. 6.5. The consistency with theory in the two
165
8
(a)
:-.
eu
I"",
O.~~~~~~~~~-J
-770
-760
-750
-740
-730 - -720
Vg (mV)
.:E
~ O.Ol
50
100
Vgl (mV)
150
200
Figure 6.4. Coulomb blockade peaks as function of gate voltage, showing fluctuations in peak
height, including some peaks of zero height. (a) From the experiment of Chang et al. [196] for
very small quantum dots, N-50-JOO. Both low temperature (Tdol - 75 mK) and highertemperature (Tdol - 600 mK) data are shown. Inset shows micrograph of multiple devices used to
gather ensemble statistics. (b) From the experiments of Folk et al. [197] for larger dots, N-JOOO
which use two shape distorting gates to create an effective ensemble of dots (inset). Data for base
temperature Tdol - 90 mK is shown. Both data sets (a) and (b) are for B =O.
0.'
(0)
&-0
(e)
9.0
0.4
r
r~
0.'
I'
0.4
0.2
Figure 6.5. Experimental distributions of Coulomb blockade peak heights. (a,b) From Chang et
al. [196] and (c,d) from Folk et al. [197]. Distributions for B = 0 (a,c) and B:t- 0 (b,c) in units of
Gmax for (a,b) and dimensionless conductance a for (c,d), with units related by Eq. (6.4). Both
experiments find reasonably good agreement with RMT, Eq. (6.6), shown as solid lines (An
alternative fitting procedure, allowing variation in lead conductance, is shown as a dashed line in
(a,b .
166
experiments is noteworthy considering the differences in device design. The

dots in Ref. 196 were relatively small, containing only -50-100 electrons, and
easily satisfied the requirement of the theory that kBT till, with till - 620
mK compared to an estimated electron temperature of -75 mK. The dots used
in Ref. 197 were considerably larger, containing of order 1000 electrons. This
allowed a greater number of peaks to be observed in a single dot, but had the
disadvantage that the inequality kBT < till was only satisfied by a factor of 2,
with till -200 mK, and an electron temperature of - 90 mK.
An interesting feature of the data in Fig. 6.4(b) is the correlation in height
between neighboring peaks, with a correlation length of -4 peaks for these
dots. Within a single-particle picture, one would expect the spatial pattern of
adjacent wave functions (associated with the ~ and (N+ i)SI states of the dot) to
be essentially uncorrelated [198]. The nature of this correlation, and whether it
can be wholly attributed to finite temperature effects or whether it is an
intrinsic feature remains unanswered and is a subject of current investigation.
6.3. PARAMETRIC CORRELATION OF COULOMB PEAK HEIGHT.
Besides investigating the height distribution of a set of Coulomb peaks by
sweeping Vg, one may also investigate mesoscopic peak height fluctuations as a
continuous function of an external parameter such as magnetic field
[196,197,199]. This procedure gives a peak height function Gmax(B) which in
many respects resembles traditional UCF measurements of conductance G(B)
in open systems but with different statistics. Peak height fluctuations are also
closely related to universal correlations of the level velocities, i.e. the rates of
change of energy of a single-particle level [200,201] with respect to a
parametric change in boundary conditions. Experimentally, the measurement
of Gmax(B) is complicated by the fact that the peak position also depends on
magnetic field, as seen for instance in Fig. 5.3 and Fig. 7.3. In practice, one
needs a two-dimensional raster over both Vg and B, with peak height
information extracted by following a peak in the Vg-B plane. An example of
random fluctuations in Gmax(B) measured using this raster method is shown in
Fig. 6.6(a) for the dot in Fig. 6.4(b).
Sensitivity of the Coulomb peak height to changes in magnetic field can be
characterized by an autocorrelation function
where Gmax = (Gmax - (Gmax is the deviation of peak height from its average.
For disordered or chaotic dots, Cg(L1B) has been calculated within RMT by
167
0.04
;e
N~
'.
0.03
"" 0.02
0.01
1.0
0.8
-..
0.6
0.2
1.0
o RMT numerics
D.
Pert. sol'n (xl
x1fit to tail
- . 0.6
~ 0.4
.......,
Eq. 6.9
Experiment
0.4
0.2
.....
L.....,.,.........:~~!!B~~...I
0.0
0.0
OJ
1.0
I.S
2.0 X 2.S
~.... - - - - - - -~_-........F1
0.0
-0.2
(b)
10
20
30
6B (mT)
40
50
Figure 6.6. (a) Parametric fluctuations in peak height, Gmax as a continuous function of magnetic
field B measured from a 2D raster over gate voltage and B, as described in text. Dashed curve is
peak height at -B showing symmetry and repeatability. (b) Experimental auto-correlation of peak
height fluctuations, defined in Eq. (6.7) (diamonds). Approximate theoretical form for peakheight auto-correlation, Eq. (6.9) suggested in Ref. [46] (dashed line) gives Be - 16 mT. Inset:
Full theoretical (RMT) distribution (open circles) along with perturbative solution Cg(x) - 1 1t2X2 (solid curve) and large-x asymptotic form, Cg(x) - O.735(1tXr 2 (dashed curve), where x =
MIBe . (From Folk et al. [41].)
Alhassid and Attias [202] and Bruus, et al. [203] and found to be universal
when plotted as a function of the scaled variable x == ifBIB e The characteristic
magnetic field Be is smaller than the field corresponding to one flux quantum
through the dot:
(6.8)
where Er is the Thouless energy, defined for ballistic dots as the inverse time of
flight across the dot, Er = hv).J Adot , N is the number of electrons on the dot,
168
and 1(is a geometrical factor (1-1) [168,192,203,204]. Measurements of Be

based on fits of the experimental Ci~B) to the scaled theoretical curve for
gate-confined dots have found values considerably larger than this prediction
[197,199], close to, or even exceeding cp/A dot Without a direct measurement of
1( these results cannot be said to be inconsistent with theory. Parametric
correlations of density-of-states fluctuations have been measured in vertical
transport by Sivan et al. [205]. In their system the characteristic magnetic field
was quite close to the value expected from theory. Their system differed from
the present one in several ways: transport was vertical rather than lateral (and
therefore only sensitive to fluctuations in density of states, not the coupling to
the leads), the dot was 3D rather than 2D, and disordered rather than ballistic.
The importance of these experimental differences remains to be sorted out.
The theoretical auto-correlation function of peak height fluctuations in the
quantum Coulomb blockade regime, hr < kBT < Lill < < e2/C, is shown as open
circles in the inset of Fig. 6.6(b) for the case of broken time-reversal symmetry
and statistically equivalent single-channel leads. This curve is found
numerically using R-matrix theory assuming a Hamiltonian for the dot of the
form H (x) =HI cos(x) + H 2 sin(x) with H1 and H2 independent random
matrices of the appropriate symmetry, i.e. GOE or GUE [202,203,206]. Near x
= 0, perturbation theory gives Cix) - 1 - rr:~, while the x 1 tail is found
numerically to be Cg(x) - 0.735(1tXy2. To facilitate comparison to experiment,
Alhassid and Attias [202] find numerically that the whole auto-correlation
function is reasonably well estimated by a Lorentzian squared,
(6.9)
for statistically equivalent leads and broken time reversal symmetry.
Experimental results for the auto-correlation of peak height fluctuations
from Ref. 197 are shown in Fig. 6.6(b) along with a fit to the form of Eq. (6.9).
The theory misses the dip below the Cg(M) = 0 axis usually seen in the
experimental data. This dip presumably result from some short-trajectory
effect and so would not be expected to show up in RMT, though perhaps would
appear in new approaches that go beyond RMT by considering the specific
phase-space dynamics for a particular ballistic billiard [187]. An outstanding
problem is the role of finite temperature on both the distribution functions and
the correlation functions. In open quantum dots, finite temperature effects on
mesoscopic fluctuations were analyzed by Efetov [207]. No similar analysis
for Coulomb blockade peak height fluctuations has been given to date.
169
6.4. MESOSCOPIC FLUCTUATIONS IN PEAK POSITION AND SPACING

While peak heights in the quantum Coulomb blockade regime show large
fluctuation as a function of peak number or magnetic field (i.e. fluctuations on
the order of the height itself), the spacing between peaks fJ. Vg generally appears
quite uniform once the dot contains many electrons, N 1, as seen in Fig. 3.3
and 6.4(b). In very small dots, N < 20, nonuniform spacing reveals shell
structure as the first few quantum states are filled, as discussed in section 5. At
high B, in the quantum Hall regime, regular peak position oscillations as a
function of magnetic field have been observed by McEuen and coworkers
[43,152,208], as discussed in section 7. An important conclusion of McEuen's
high-field experiments is that in order to adequately explain the data, a selfconsistent model of the confined electrons in a field is needed. Whether this
continues to hold in the low field regime, where OD quantization rather than
Landau level quantization modifies the classical electrostatics problem, is not
known. To start off, however, we will discuss the simplest model of random
fluctuations in peak spacing as a function of the number of electrons on the dot
(as set by a gate voltage). This model assumes a constant classical charging
energy e2/C which can be separated out from the level spacing !lE between
non-interacting OD states. In this picture, fluctuations in peak spacing are
purely associated with fluctuations in spacing between the OD-states (see also
Eq. (2.3) and below):
L1vt = e/Cg
(N odd)
(6.10a)
(N even)
(6.10b)
As discussed by Sivan et at. [209], if one further assumes L1E to be distributed

according to RMT statistics (assuming the dot is disordered or chaotic) the
resulting fluctuations in spacing L1e = L1E/(L1E) should then be distributed
according to the famous "Wigner surmise" [210] for the distribution of
eigenvalue spacings in random matrices,
(B = 0; GOE)
(6.11a)
(B:;{: 0; GUE)
(6.11b)
170
This "constant interaction plus random matrix theory" (CI+RMT) model yields
an alternating average peak spacing given by the averages of Eq. (6.10) with
rms fluctuations in the spacing of peaks that bracket an even-N state given by:
CAE (3rt/8-1)1/2 :::0.42-CAE

8(.1Vg )=-eCg
eCg
(B = 0; OOE)
(6.12 a)
(B"* 0; OOE)
(6.12 b)
independent of N, as long as kBT < till. Since the level spacing is typically
much smaller than the charging energy, Eq. (6.12) implies relatively small
fluctuations in peak spacing, consistent with experiment. A more detailed
comparison, however, reveals both quantitative and qualitative disagreement
between CI+RMT and experiment. At zero or small magnetic field, no
even/odd behavior has been reported in dots with N 1 (although wellunderstood spin effects are seen in tunneling and capacitance spectroscopy for
small N as discussed in section 5). In fact, Sivan et al. [209] find that
fluctuation statistics in peak spacing in small gate-defined OaAs quantum dots
at low temperature (-100 mK) disagree significantly from the CI+RMT
prediction. They find peak spacing fluctuations larger by a factor of up to five
from the predictions of the CI+RMT model, with an insensitivity to factor-oftwo changes in till as N ranges from -60 to -120 as seen in Fig. 6.7.
Moreover, the observed distribution of fluctuations does not appear similar in
form to Eq. (6.11), but is symmetric about its average. These observations
have lead Sivan et al. to suggest a picture of peak spacing fluctuations that is
essentially classical in origin, closely related to the problem of packing charges
onto a finite volume, with spacing fluctuations resulting from random "magic
numbers" in which better and worse packings of charge depend on N. Their
picture is supported by a numerical calculation of the ground state energy of a
lattice model of the dot which shows that as interactions are turned on,
fluctuations in ground state energy transform from the Wigner statistics of Eq.
(6.11) to roughly gaussian fluctuations with an rms amplitude of -0.10 - 0.15
e2/C, independent of till.
A recent self-consistent calculation of ground state energy fluctuations in
2D and 3D quantum dots beyond the CI+RMT picture predicts fluctuations in
peak spacing due to capacitance fluctuations (i.e. in addition to single-particle
effects) that does depend on till:
171
1000
--B=O
GaAsdots
T=50mK
- - B=O (shifted)
---- B=1T (shifted) CI+RMT predicti
800
:~b!LLJ_
SCI)
.;!, 600
01
!l
400
8.
200
20
40
electrons added
60
80
Figure 6.7. Spacing between neighboring Coulomb blockade peaks (solid squares, in energy
units, including capacitance lever arm) versus number of electrons added to a GaAs quantum dot.
Bottom two traces have been shifted by 200 IN and 400 /lV, respectively. Overall slope results
from gradual increase in capacitance as N increases, solid line is a linear fit. CI+RMT prediction
(dashed lines) is based on Eq. (6.12). (From Sivan et at. [209].)
~~ a 2 ( 27tN rI/4
(2D)
(6.13)
O(LlVg ) -
~4/3 C till
eC
(
) -1/4
a 3 27tN
(3D)
for weakly disordered dots [211]. (Eq. (6.13) applies in the case of quasiballistic motion of electrons; if the dot is strongly diffusive, f L, the term
(27tN),1I4 should be replaced with g,ll2, where g is the dimensionless
conductance of the material.) The parameter ~(3) ::::: 1 for weak interactions, for
which the screening length is much less than the Fermi wavelength, and is of
order the gas parameter, ~(3) - e2/EvF' for strong interactions. For GaAs
quantum dots, one expects ~(3) ::::: 1. The analysis in Ref. [211] suggests that
additional contributions to the fluctuations unaccounted for in the CI+RMT
come from unscreened charge at the edge of the dot (roughly analogous to the
packing picture). On the other hand, Eq. (6.13) suggests fluctuations in peak
spacing should of order LlE for weak interactions. If the value of ~(3) were to
become 1, the fluctuations would indeed be large according to Eq. (6.13),
but why this should occur in GaAs gate-defined dots is not apparent. Clearly
more experiments are needed to sort out this interesting problem.
172
A related issue concerns the parametric motion of a single Coulomb

blockade peak in a magnetic field. Such motion is discussed in section 5 for
few-electron dots, and in section 7 for dots in the quantum Hall regime.
Parametric fluctuations of peak position at low field in larger gate confined
dots [197,199] have an rms amplitude 8 [AVlB)] ::::: (e/Cg) [L1E/(e 2/C)],
corresponding to energy fluctuations of order AE. Since peak spacing
distributions over an ensemble of peaks are similar to, but certainly not the
same as parametric fluctuations of a single peak, it may not be appropriate to
compare this result directly to the experimental and theoretical work on peak
spacing statistics gathered over many peaks.
6.5. PARAMETRIC PEAK MOTION AND ORBITAL MAGNETISM.
The fluctuations of Coulomb blockade peak position, as distinct from peak

height, as a function of B is closely related to the universal parametric motion
of quantum levels [186,200,201] as well as to the magnetic properties of
mesoscopic samples. Connections between the statistics of peak position and
peak height fluctuations have been addressed within RMT by Alhassid and
Attias [202]. Peak position fluctuations are particularly important because they
can be related to mesoscopic fluctuations of orbital magnetism in small
electronic systems, a subject of great interest in the last few years as the result
of a provocative handful of technically difficult direct measurements of the
magnetic response of meso scopic structures. By definition magnetization M =
-au(N,ByaH and magnetic susceptibility X = aM/aH are the first and second
derivatives of the ground state energy of a system with respect to B. So, at zero
temperature, M and X are respectively the sums of parametric level velocities
and level curvatures of all states below the Fermi surface [212]. (Remember
the definition /1dolN) UrN) - U(N-i).) Experiments measuring the magnetic
moment (or, alternatively, the persistent current, expressing derivatives in
terms of flux rather than field, I = -aU(N,cpyacp in metallic rings have found
dramatically enhanced magnetic response, one to two orders of magnitude
larger than expected for non-interacting electrons, both for large ensembles
[213] and individual rings [214]. In contrast, the persistent currents measured
in a single ballistic GaAs ring [215] was also found to be large, but in this case
was consistent with theory (for reviews see [212,216]). The susceptibility of
105 ballistic 2D GaAs squares showed a dramatically enhanced paramagnetic
response around zero field, roughly 100 times the Landau diamagnetic
susceptibility Xo = -e2/121tm*c 2 [217]. This effect has been interpreted as the
result of threading Aharonov-Bohm flux through nonchaotic families of
trajectories in the square billiard, emphasizing the importance of the underlying
classical dynamics on mesoscopic magnetic properties [204,216,218-220].
173
More generally, one expects typical mesoscopic fluctuations in X for an

isolated ballistic 2D dot to exceed the Landau susceptibility by powers of kFL
depending on whether the shape of the dot is chaotic or integrable
[204,219,220]:
(chaotic dynamics)
(6.14)
(integrable dynamics)
Many of the unanswered questions concerning mesoscopic magnetism can
be recast in terms of the B dependence of Coulomb blockade peak position Vg*
In particular, the derivative of the peak position with respect to magnetic field
is proportional to the difference between the magnetizations for subsequent
values of N:
av; laB - (CjeC MN - MN+l]
(6.15)
g )[
assuming the ratio of capacitances in the prefactor is not field dependent.

Theoretically, fluctuation statistics of
can be calculated by the same
methods used to obtain Eq. (6.14). An important difference between Coulomb
peak position and magnetization, however, concerns fluctuating particle
number. Whether or not the number of particles on the dot is a fixed quantity
affects orbital magnetization and susceptibility. For instance, the zero field
susceptibility of a chaotic-ballistic 2D dot has zero average (over an ensemble
of dots or over shape distortions of a single dot) when particle number is not
fixed (grand canonical ensemble), \X) GCE = 0, but is paramagnetic for fixed
particle number (canonical ensemble), \X) CE - -kFLXo. Transport through a
Coulomb blockade peak, on the other hand, represents a hybrid ensemble in
which particle number may fluctuate by 1 but no more on the conductance
peak, and can undergo quantum fluctuations (co-tunneling) between peaks.
The rules of magnetic response in this case have not been established.
av; laB
6.6. FLUCTUATIONS IN ELASTIC CO-TUNNELING

At moderately low temperature and small voltage bias (kBT, eVsd) < (AE,e 2fC)tn.,
the residual conductance between Coulomb blockade peaks is dominated by
elastic co-tunneling in which an electron (or hole) virtually tunnels through an
energetically forbidden charge state of the dot lying at an energy
above
174
(below) the Fermi energy in the leads, where 8 equals e2/2C at the center of the
valley between peaks and decreases to zero on the peak. As discussed by
Averin and Nazarov [221], elastic co-tunneling is a coherent virtual process
that occurs on a short time scale, 'feot -h/8 , consistent with the time/energy
uncertainty relation. Average transport properties for elastic as well as inelastic
co-tunneling were given in Ref. 221 and experimental aspects in Ref. 222.
Aleiner and Glazman recently extended this work to include mesoscopic
fluctuations of elastic co-tunneling [223]. Unlike on-peak conduction which
can be described as a one-electron resonant tunneling process, co-tunneling
properties are strongly affected by electron-electron interactions in the form of
the charging energy.
The co-tunneling current for weakly coupled leads is usually very small
and therefore difficult to measure. However, once the tunneling point contacts
are sufficiently open, say GI,r > -0.5(2e2/h), fluctuations in the valleys can be
measured quite easily, allowing co-tunneling fluctuations Gmin(B) at valley
minima to be studied along with the resonant tunneling fluctuations Gmax(B) at
peak tops. Figs. 6.8(a) and (b) show co-tunneling and resonant tunneling
fluctuation for an adjacent peak and valley in a -0.3 Jim GaAs quantum dot
with Ee ::::: 600 IleV and AE ::::: 20 JleV [224]. Again, because the gate voltage
positions of the peaks and valleys depend on B, a 2D raster over the B-Vg plane
is needed to follow peaks and valleys.
The autocorrelation functions C(MJ) for both Gmax(minlB), (defined by Eq.
6.7) shown in Fig. 6.8(c) illustrate the primary difference between resonant
(peak) and co-tunneling (valley) fluctuations: the characteristic magnetic field
Be is significantly larger for the valleys than for the peaks [197]. The
difference in Be can be understood from a semiclassical point of view as
follows: On resonance, the characteristic time during which an electron
diffusively accumulates Aharonov-Bohm phase is the so-called Heisenberg
time, or inverse level spacing, 'fH - h/I1E, the same as for an isolated billiard.
Because co-tunneling is a virtual process, the time over which phase may
accumulate is much shorter, 'feot -h/8, limited by the uncertainty relation. This
suggests a characteristic field in the valleys defined in analogy to Eq. (6.8):
(6.16)
giving a ratio of characteristic fields:
(6.17)
175
0.05
""'
0.8
rE
--l:r- 'peak'
T 'valley'
Valley
'c
Il)
'-"
OJ)
0.00 '------'-,-::----,=-,"---.1----1
'c
0.6
C(~B)
0.4
0.5
""'
N~
on
0.0
0.2
(b) Peak
-150 -100 -50
B (mT)
B (mT)
12
Figure 6.8. (a) Mesoscopic fluctuations of elastic co-tunneling in the valley and (b) resonant
conductance on the adjacent peak for a -0.4 J11rI GaAs gate-defined quantum dot. (Note the
different vertical scales in (a) and (b).) (c) Normalized auto-correlation of peak and valley
fluctuations, showing the factor of -2 larger correlation field for the Valley. (From Cronenwett et
al, [224].)
For the gate-defined GaAs dots studied in Fig. 6.8, the expected ratio of
characteristic fields is ((300 JleV)I(20 JleV)1I2 "" 4. This estimate appears
inconsistent with the experimentally observed ratio of - 2 in Fig. 6.8(c). A
possible explanation for this large discrepancy is that on the peak some time
scale shorter than hl!3.E is acting as the characteristic time for phase
accumulation in resonant tunneling.
A proper theoretical treatment [223] of co-tunneling fluctuations accounting
for virtual processes through all excited levels above the Coulomb gap
reproduces the semiclassical results Eqs. (6.16) and (6.17), and for the case ET
< 21t8, predicts explicit universal forms for the autocorrelation of valley
conductance (Fig. 6.9) as well as the full distribution of co-tunneling
fluctuations for arbitrary magnetic field [223]. One interesting feature of the
analysis is that although the full distribution is sensitive to time-reversal
symmetry breaking by a small magnetic field, its first moment, the average cotunneling conductance (\ g) cot ,is independent of field and therefore (unlike peak
conductance) does not show an analog of weak localization.
176
Finally we point out that the increased field scale of the valleys is a direct
reflection of the "tunneling time" of an electron through the dot [225].
Mesoscopic fluctuations of co-tunneling therefore provide a novel tool for
measuring full distributions of such times in a much simpler way than can be
realized in time-domain tunneling experiments. Experimental work in this
direction is in progress.
6.7. CONCLUSIONS AND OPEN PROBLEM.
The coexistence of quantum interference, quantum chaos (leading to universal
statistics of wave function and scattering statistics), and electron-electron
interaction makes the problem of transport through quantum dots at low
temperatures both complicated and very rich, experimentally and theoretically.
This is true for both open quantum dots and Coulomb blockaded dots, the
subject of the present section. As in the nuclear scattering problem, the
strongest justification for the use of RMT in mesoscopics has been agreement
with experiment.
Recent experiments described here [196,197] have
highlighted an important new success: a correct description of the peak height
fluctuations in the quantum regime, hT < kBT < fill e2/c. We have also
seen how mesoscopic fluctuations in virtual tunneling are observed
experimentally and understood qualitatively within relatively simple models.
However, many phenomena remain unaddressed theoretically, for instance the
effects of finite temperature, scarred wave functions, dephasing and mixed
dynamics on the distribution and correlation of peak heights. Other phenomena
disagree quantitatively with a "constant interaction plus single-particle
quantum chaos" model. Such outstanding disagreements include correlations
between the heights of neighboring Coulomb peaks, the magnetic field scale
for peak height fluctuations, ratios of peak to valley correlation fields, and peak
spacing distributions.
Much of the theoretical story relating mesoscopic fluctuations, quantum
chaos, and random matrix theory has been worked out only in the last year or
so, and is only now beginning to be tested experimentally. As the focus in
mesoscopic physics continues its shift toward the influence of interactions and
coupling to the environment, new difficulties and challenges will certainly
continue to appear. Two directions of interest that will further expand the
palette of mesoscopic phenomena in microstructures in the coming years are
the inclusion of superconducting contacts and high-frequency excitation.
Present successes motivate a statistical approach to these problems as well, and
indeed significant theoretical inroads have been made. Here, too, experiments
remain very far behind theory.
177
7. Quantum Dots in High Magnetic Fields.

In this section, we examine the addition spectra of quantum dots when
large magnetic fields are applied, and compare the experimental results to
theoretical predictions. In 7.1, we address few electron dots, where exact
calculations can be performed. Interesting predictions, such as singlet-triplet
oscillations in the spin state of the two-electron dot, are compared with
experiment. In 7.2 and 7.3, we discuss many-electron dots, where the
quantization of the electron orbits into Landau levels is important. 7.2
addresses results that can be understood at a Hartree level, while 7.3 looks at
Hartree-Fock and beyond.
7.1. FEW-ELECTRON DOTS AT HIGH MAGNETIC FIELDS.
As was discussed in the previous sections, the simplest model of a quantum dot
consists of non-interacting electrons residing in a parabolic confining potential.
The classical motion is then a periodic oscillation with a characteristic
frequency mo' The addition of a magnetic field alters the motion, leading to
orbits of the type shown in Fig. 7.1. An electron at the center of the dot rotates
in a cyclotron-like orbit, which becomes the cyclotron frequency me = eBlm * at
high magnetic fields. Electrons away from the center slowly precess around
the dot as they perform their cyclotron motion. This is due to the drift velocity
vD = ExB of the cyclotron orbit in the electric field of the confinement
potential. Quantum mechanically, this model can be easily solved [149,150];
the result is given in Eq. (5.2). If we include spin then at high magnetic fields
(mc m), Eq. (5.2) simplifies to:
E(n,m,Sz)=(n+lf2)1iwc + (2n+II+l)nw; fwc +g/1BBS z
(7.1)
where n = 0, 1, 2, ... is the radial or Landau level (LL) index, I! labels the
angular momentum of the drifting cyclotron orbit, and Sz = 1I2 is the spin
index. Roughly speaking, the LL index n labels the number of magnetic flux
quanta hie enclosed by the electron orbit during its cyclotron motion, while II! I
labels the number of flux quanta enclosed by the drifting orbit. Since each
successive I!-state encloses one more flux quantum, each (spin-degenerate) LL
within the dot can be occupied by one electron per flux quantum penetrating
the area of the dot. Increasing B causes both types of orbits to shrink in order
to encircle the same number of magnetic flux quanta, making more states fit in
the same area and increasing the LL degeneracy.
178
Figure 7.1. Top: Classical electron orbits inside a parabolically confined quantum dot. The orbit
in the center exhibits cyclotron motion, while the orbit away from the center also drifts in the
electric field of the confining potential. Bottom: Schematic energy level diagram of a quantum
dot in a high magnetic field. The n = 0 and n = 1 orbital LLs are shown, each of which is spinsplit. The dots represent quantized states within a LL that encircle m flux quanta in their drifting
cyclotron motion, where m is linearly related to t, the angular momentum of the state.
Eqs. (5.2) and (7.1) ignore electron-electron interactions. Nevertheless,

they should be valid for the first electron occupying a dot, since there are no
other electrons with which to interact. The solution from Eq. (5.2) with n = f =
o should thus describe the ground-state addition energy of the first electron. At
B = 0 this is the zero-point energy of the harmonic oscillator, /iro o l2. At high
B it is the energy of the lowest LL /iro)2, and the electric to magnetic
crossover occurs when ro ~ ro0 .
Measuring a one-electron dot in the lateral gated geometry has proven to
be difficult. Vertical dots with as few as one electron have been studied by
both linear transport measurements and nonlinear /- V characteristics and by
capacitance spectroscopy, as we discussed in section 5. Results from the latter
technique are shown in Fig. 7.2, taken from Ashoori et ai. [50]. The change in
the capacitance due to a single electron tunneling on and off a dot is plotted in
grey scale as a function of energy, which was deduced from an applied gate
voltage, along the y-axis and magnetic field along the x-axis. The first line at
the bottom of Fig. 7.2 represents the addition energy for the first electron as a
function of B. The addition energy is constant for low B and grows linearly for
high B. Fitting to Eq. (7.1) allows the determination of the bare harmonic
oscillator frequency: /iro o = 5.4 meV.
C
179
Magnetic Field (Tesla)
Figure 7.2. Gray scale plot of the addition energies of a quantum dot measured as a function of
magnetic field. Each successive light gray line corresponds to the energy for adding an
additional electron to the dot. (a) Addition spectrum for the first few electrons. The dot on the
curve for the second added electron marks the singlet-triplet transition discussed in the text. (b)
Addition spectrum for 6 through 35 electrons. The triangles mark the filling factor v = 2 (From
Ashoori et al. [90]).
The situation gets more interesting for more than one electron on the dot.
To describe the addition energy for larger number of electrons, the simplest
approach is to use the non-interacting electron spectrum, Eq. (7.1), combined
with the Coulomb-blockade model for the interactions. This model is
discussed in section 2. In this approximation, the second electron would also
go into the n = f = state, but with the opposite spin, creating a spin singlet
state. This spin singlet state remains the ground-state configuration until the
Zeeman energy is large enough to make it favorable for the second electron to
flip its spin and occupy the n =0, f = 1 state. From Eq. (7.1), this occurs when
nwo / we = gJ1j3. The two electron ground state is then an Sz = 1 spin-triplet
state. For GaAs the spin splitting is quite small (g = -0.4), and the Zeemandriven singlet-triplet transition would occur at a very large B of around 25 T for
the dot in Fig. 5.2. The data, however, shows something quite different. The
180
addition energy for the second electron has a feature at a much lower field
(marked by a dot) that has been attributed to the singlet-triplet transition [50].
A more realistic model of the Coulomb interactions can explain this
discrepancy [77]. The size of the lowest state (i.e. n = 0, f = 0) shrinks in size
with increasing B. As a result, the Coulomb interaction between the two spindegenerate electrons grows. At some point, it becomes favorable for the
second electron to occupy the f = 1 single-particle state, avoiding the first
electron and reducing the Coulomb interaction energy. Now the electrons are
in different single particle states, the Pauli exclusion principle no longer
requires that their spins point in opposite directions. Both the exchange
interaction and the external magnetic field favor an alignment of their spins,
and the two-electron system thus switches to a triplet state. This transition is
driven predominantly by Coulomb interactions, since the spin splitting is still
quite small.
Many other features are also observed in the addition energies of the first
few electrons as a function of B, as seen in Fig. 7.2(a). These features can also
be interpreted by comparison with microscopic calculations [227]. The
agreement between experiment and theory is not always perfect, which
indicates the need for further study.
7.2. MANY-ELECTRON DOTS IN THE QUANTUM HALL REGIME.
At larger number of electrons on the dot (N) 20), the capacitance spectroscopy
measurements begin to show very organized behavior, as is seen in Fig. 7.2(b).
This large N regime has been extensively explored by transport spectroscopy in
lateral structures [152,195,208]. An example is shown in Fig. 7.3, where the
addition energy for the Nth electron (N - 50) is measured as a function of B
[152]. This plot is made by measuring a Coulomb oscillation and plotting the
position in gate voltage (a) and height (b) of the peak as a function of B. The
behavior is very regular in the regime between 2 T and 4 T. The peak positions
drop slowly, and then rise quickly, with a spacing between rises of
approximately 60 mT. At the same time that the peak position is rising, the
peak amplitude drops suddenly. Regularities can also be seen in the peak
amplitudes measured at a fixed B, but with changing V,g i.e. for adding
successive electrons. For example, the data presented in Figs. 4.1 and 4.2 are
plots of a series of peaks in the ordered regime above 2 Tesla [152]. A close
examination reveals that the peak heights show a definite modulation with a
period of every-other peak.
To understand these results, a theoretical model of the many-electron dot
is needed. Unfortunately, for dots containing more than -10 electrons, exact
calculations cannot easily be performed and approximation schemes must be
181
131.0 r------.----..--------r------,
(a)
130.5
v= 2
jl30.0
~
CJ
129.5
129.0
WI
10- 1
10-'
10-4 L -_ _- - ' -_ _ _- ' -_ __
1.50
2.25
........_
3.00
3.75
KapeUc ne1d (T)
---.J
4.50
Figure 7.3. (a) Position in gate voltage and (b) peak height of a conductance peak measured as a
function of magnetic field. The filling factors v in the dot are as marked. The quasi-periodic
structure reflects single-electron charge rearrangements between the two lowest LLs. (From
McEuen et al. [152].)
p(r)
E
J
21tli
1
21t1i
Figure 7.4. (a) Self-consistent model of a dot with two Landau levels occupied. (a) Filling of the
LLs that would yield the classical electrostatic charge distribution. (b) Electrons redistribute
from the higher to the lower LL to minimize their LL energy. (c) Resulting self-consistent level
diagram for the dot. Solid circles: fully occupied LL, i.e. an "insulating" region. Open circles:
partially occupied LL, i.e. a "metallic" region. (From McEuen et al. [152].)
used. Again, the simplest approach is to assume the electrons fill up the noninteracting electron states, given by Eq. (7.1), and to use the Coulomb blockade
model to describe the Coulomb interactions [7,208]. This model was used to
interpret early experiments [208], but later work showed it to be seriously
inadequate [152], for essentially the same reasons that we discussed above for
the two-electron dot. In a high magnetic field, Coulomb interactions cause
rearrangements among the states that cannot be understood from the behavior
of non-interacting levels.
182
An improved description of the addition spectrum treats the Coulomb

interactions in a self-consistent manner [152,228,229]. This proto-Hartree
approach is essentially the Thomas-Fermi model, but with the LL energy
spectrum replacing the continuous density of states that is present at B = O. In
this model, one views the quantum dot as a small electron gas with a
nonuniform electron density. Classically, this density profile would be
determined by the competition between the Coulomb interactions and the
confinement potential. For example, for a parabolic confinement potential, the
result is an electron density that is maximal at the center and decreases
continuously on moving away from the center, as shown in Fig.7A(a).
We now include the effects of Landau level quantization in this picture. In
a first approximation, the electrons fill up the requisite number of Landau
levels to yield the classical electrostatic distribution. For simplicity, we
concentrate exclusively on the case where only two LLs are occupied (n = 0; Sz
= l12, i.e. the spin-resolved lowest orbital LL). This is shown in Fig. 7A(a).
Note, however, that the states in the second (upper) LL have a higher spin
energy than those in the first (lower) LL. As a result, some of these electrons
will move to the lower LL. This continues until the excess electrostatic energy
associated with this charge redistribution cancels the gain from lowering the LL
energy. The resulting (self-consistently determined) charge distribution for the
island is shown in Fig. 7 A(b), and the electrochemical potentials for electrons
added to the two LLs are shown in Fig. 7A(c). Note that partial occupation of a
LL implies that there are states at the Fermi energy available to screen the bare
potential. If we assume perfect screening then the resulting self-consistent
potential is flat. This is analogous to the fact that in the interior of a metal no
electric fields are present. For example, in the center of the island, where the
second LL is partially occupied the self-consistent electrostatic potential is flat.
Similarly, near the edge, where the first LL is partially occupied, the potential
is also flat. In between, there is an insulating region where exactly one LL is
occupied.
The result is that we have two metallic regions, one for each LL, separated
by an insulating strip. Electrons added to the dot are added to one of these two
metallic regions. If the insulating strip is wide enough, tunneling between the
two metallic regions is minimal; they will effectively act as two independent
electron gases. The charge is separately quantized on each LL. Not only is the
total number N of electrons in the dot an integer, but also the numbers of
electrons N1 in LL, and N2 in LL2 are integers. In effect, we have a two-dot, or
"dot-in-dot" model of the system, very much similar to the parallel dot
configuration in Fig. 4.1.
This schematic picture of a quantum dot in high magnetic fields is
supported by a number of simulations [152,228-230]. Fig. 7.5 shows a contour
183
Figure 7.5. Contour plot of the self-consistent electrostatic potential for a (1 Ilm x 1 Ilm)
quantum dot in a high magnetic field. In the regions labeled #1 and #2, the first and second LLs
are partially occupied. The electrons can thus rearrange themselves to screen the external
potential, and the resulting self-consistent potential is constant. In between, where one LL is
fully occupied and no screening occurs, the potential rises sharply. (From Stopa [230].)
map of the electrostatic potential for a quantum dot with two occupied LLs, as
calculated by Stopa [230]. In the center of the dot (region #2) where the second
LL is partially occupied, the potential is flat. Similarly, the first LL creates a
ring of constant potential where it is partially occupied (region #1). Electrons
tunneling onto the dot will go to either one of these metallic regions.
We now discuss the implications of this model for transport
measurements. First, as additional electrons are added to the dot, they try to
avoid each other. As a result, successively added electrons tend to alternate
between the two metallic regions. Note, however, that electrons will most
likely tunnel into the outer LL ring, as it couples most effectively to the leads.
Peaks corresponding to adding an electron to the inner LL should thus be
smaller. If electrons are alternately added to the inner and outer LLs with
increasing gate voltage, the peaks should thus alternate in height. The
measurements of Fig. 3.2 show this behavior. Measurements [231] for higher
numbers of LLs occupied give similar results (i.e. a periodic modulation of the
peak amplitudes), with a repeat length determined (approximately) by the
number of LLs occupied [83,232].
To understand the peak-position structure in Fig 7.3(a), we again note that,
as B increases, the electrons orbit in tighter circles to enclose the same
magnetic flux. In the absence of electron redistribution among the LLs, the
charge density therefore rises in the center of the dot and decreases at the
184
edges. This bunching causes the electrostatic potential of the second LL to rise
and that of the first LL to drop. Therefore, the energy for adding an electron to
the first LL, JI/N],N2 ), and hence the peak position, decreases with increasing
B. This is illustrated schematically in Fig. 7.6. This continues until it becomes
energetically favorable for an electron to move from the second to the first LL.
This electron redistribution, which we call internal Coulomb charging, causes
the electrostatic potential of the first LL to jump from JI/N], N 2) to JI/N] +1,
N 2-1) with N = N] + N 2 The energy difference [JI](N], N 2 ) - JI/N]+l, N 2 -1)] is
equal to the interaction energy between LL\ and LL2 minus the single electron
energy of LL\. These jumps are clearly observable in the data of Fig. 7.3,
occurring every 60 mT. Note that these electron redistributions are a manyelectron version of the two-electron singlet-triplet transition. In both cases,
Coulomb interactions push electrons into states at larger radii with increasing
B.
The peak height data shown in Fig. 7 .3(b) can be similarly explained. The
peak amplitude for adding the Nth electron is strongly suppressed at B fields
where it is energetically favorable to add the electron to the inner LL. This
corresponds to the magnetic field where the peak position is rising. A dip in
the peak amplitude thus occurs at every peak position where an electron is
transferred from the second to the first LL. The period of the oscillation, 60
mT, roughly corresponds to the addition of one flux quantum to the area of the
dot This period implies an area of (0.26 Ilm)2, a size which is consistent with
the dimensions of the dot.
J.1 1(N)
~~-------+~------ B
Vg
Vg
Figure 7.6. Schematic illustration of charge redistribution within a dot with increasing magnetic
field. When a single electron moves from the 2nd to the 1st LL, the electrochemical potential for
adding an additional electron to the 1st LL increases. As a result, the peak position shifts.
185
Care must be taken in interpreting the peak height, however. Other

experiments show [208,231,233] that the heights of the smaller peaks do not
directly reflect the tunneling rate into the inner LL. The tunneling rate into the
inner LL is typically too small to produce a significant current. The observed
peak is actually due to thermally-activated transport through the outer (first)
LL. Since all of the observed current corresponds to tunneling through the first
LL, the position of a peak is proportional to the electrochemical potential
Il/N],N2) for adding an electron to the first LL. This potential is a function of
both N] and N 2 , the number of electrons in the first and second LL, respectively.
The jumps in the peak position with increasing B thus represent a
redistribution of electrons between the LLs. In experiments by van der Vaart et
ai. [233], the peaks were actually observed to jump back and forth in time.
This is shown in Fig. 7.7. Fig. 7.7(a) shows that with two LLs occupied a peak
that corresponds to N electrons in the dot can appear as a double peak. The
double peak has a resonance when either Il/N],N2) or 1l](N]+1,N2-1) aligns with
the Fermi energy of the reservoirs. Fig. 7.7(b) shows that the conductance
measured as a function of time at a fixed gate voltage switches between two
discrete levels. This peak-switching is due to a single electron hopping
..:
~~ ~:
::
:: :
~ 0.2
~
o
u
:; :~
.0
o'
~ ~~ ~
! -:f
:.
!
~
~ ~
~
. ..,
:. :" .
:! \
~ t~
~
.
i)
o ......
..
.
o
: :
\~.
.....
-0.58 t
-0.57
GATE VOLTAGE (V)
.'
(b)
(a) B=5.2T
~: ~
.: ::
f3
.-'
".
.0a
50
100
TIME (s)
Figure 7.7. (a) Conductance through a quantum dot as a function of gate voltage, measured in a
regime where 2 LLs are occupied inside the dot. The Coulomb peaks are observed to switch back
and forth between two positions. (The dotted lines are a guide to the eye.) (b) Conductance
versus time with the gate voltage fixed at the value denoted by the arrow in (a). The switching
behavior results from the hopping of a single electron between the 1st and 2nd LL. (From van
der Vaart, et al. [233].)
186
between the inner and the outer LL. At this magnetic field, the time for
hopping was on the order of 10 seconds. The tunneling rate between the inner
and outer dot is thus incredibly small. This corroborates the point made earlier
that the coupling to the inner Landau level is very weak and all of the
measurable current is carried by tunneling through the outer LL. Note that this
justifies viewing two LLs in a single dot as effectively a double, parallel dot
system.
7.3. HARTREE-FOCK AND BEYOND.
The model and experiments discussed above indicate that much of the
behavior of quantum dots in magnetic fields can be understood based on LL
quantization and self-consistent electrostatics. Recently, however, a number of
measurements have demonstrated the importance of Coulomb interactions
beyond the Thomas-Fermi approximation. For example, the Hund's rule
behavior discussed in section 5 is most easily understood within the HartreeFock approximation.
In the quantum Hall regime, the Hartee-Fock
approximation [155,234] yields an effective short-range attractive interactions
between electrons of the same spin that leads to larger incompressible regions
than in the model above. For example, this significantly alters the rate at which
electrons move from the second to the first LL with increasing B in the regime
2 < v < 1. In particular, it is predicted that the transition from a spinunpolarized dot at v = 2 to a spin-polarized dot at v = 1 can be described as a
second-order phase transition between a magnetic and nonmagnetic state. The
magnetization (i.e. the spin polarization of the dot) is predicted to vary as: M (B - By12 [235], where Be is the magnetic field at v = 2. This implies a rate of
change of M with B, i.e. a spin susceptibility, of the following form:
X=dMldB - (B -BJl12
(7.2)
The diverging spin susceptibility near B = Be indicates that the spins flip very
rapidly with increasing B near the transition. This is driven by the exchange
interaction making it desirable to create a region of spin polarized electron gas
around the perimeter of the dot.
This prediction is borne out by experiments of Klein et ai. [235]. Fig. 7.8
shows measurements of the addition spectrum, and Fig. 7.9 the spin
susceptibility. The latter is measured by extracting the discrete derivative of M
with respect to B from the data: dMldB = (1 spin)/.:::1B between successive spinflips). As Fig. 7.9 shows, the HF theory closely resembles the experimental
data, while the self-consistent theory does not produce the diverging
susceptibility seen in the experiment.
187
(A)
1151
l
1
(8)
25
<l>
25
20
15
10
i-.
fQ~
e
159
(a)
20
15
10
(b)
2tI
20
SC
(e)
U
1
Magnetic field (T)
10
'..00"
xlO
5 ___ _ _._~ SI. ~

0
2.0 2.5 3.0 3.5 4.0 4.5
DD (T)
Figure 7.8. (A) Position of the Nth conductance peak as a function of magnetic field. The filled
black circle marks the magnetic field at which all electrons are in the lowest orbital LL. The
arrows indicate spin flips of individual electrons within the dot. (From Klein et al. [235].) Inset:
Schematic diagram of the spin-flipping process. (From Ashoori [15].) (B) Plots of the spin
susceptibility of the dot versus B. (a) Experimental data from (A). (b) Predictions of HarteeFock model. (c) Predictions of self-consistent model. (From Klein et at. [235].)
At higher B, in the regime v < 1, Hartree-Fock models also make

interesting predictions. In the self-consistent model for v < 1, the charge
density simply retains its classical electrostatic profile, since the kinetic energy
of the electrons are quenched. However, the exchange interaction and
correlations beyond the exchange interaction favor different possibilities. If the
electron gas is assumed to remain spin-polarized, then theory predicts an edge
reconstruction with increasing B where the charge density no longer
monotonically decreases with increasing radius [155,234,236,237].
More recently, people have considered the possibility of non-spinpolarized ground states, motivated by the observation of spin textures, or
"skyrmions" in bulk 2DEGs at filling factors near v = 1 [238]. In this case, the
exchange interaction favors a slow variation of the spin of the 2DEG in space
to accommodate an extra, or a missing, electron in a full LL. Recent work
indicates that such spin textures will form at the edge of a 2DEG [239], or at
the periphery of a quantum dot [240], under the right experimental conditions.
Experimentally, jumps in the addition spectra are observed for v < 1
[50,235]; see, for example, the jump marked by a triangle in Fig. 7.8. These
jumps have been interpreted at resulting from edge reconstructions [15,235]. It
188
is very difficult, however, to delineate between the two types of reconstructions

discussed above from measurements of the addition spectrum only. Recent
experiments on the excitation spectra [241] of dots give evidence that spin flip
excitations are important, but the evidence is indirect. A direct measurement of
the spin polarization of the dot would be very helpful, but performing such a
measurement remains an unsolved experimental challenge.
Also of potential interest are many-body effects on the tunneling rates of
single electrons on and off the dot. If tunneling on the dot requires a complex
rearrangement of all other electrons, its rate is predicted to be dramatically
suppressed [242]. This "orthogonality catastrophe" may be contributing to the
extremely slow tunneling rates between the inner and outer LL regions found in
the experiment of Fig. 7.7. More experiments are necessary to fully explore
these possibilities.
8. Time-dependent transport through quantum dots.

This section presents a brief review of some of the experiments and theory on
time-dependent transport in quantum dots. ill practice, "time-dependent
transport" means that an ac signal is applied to a single dot or a multiple dot
system and the time-averaged current is measured. ill this sense, the process is
simply rectification, although the effects can be both non-linear in the driving
signal and also non-adiabatic in the driving frequency. illdeed, the application
of external frequencies comparable to internal energies of the dot (e.g. level
spacings) can be thought of as a form of spectroscopy. The following topics
are addressed here: (8.1) adiabatic driving of electrons "the electron turnstile",
(8.2) non-adiabatic driving and the Tien-Gordon picture of time-dependent
transport, (8.3) spectroscopy of a single dot, and (8.4) time-dependent transport
through a double dot.
8.1. ADIABATIC REGIME; THE ELECTRON TURNSTILE.
Because of the Coulomb blockade the current through a quantum dot is limited
to one electron at a time. This property can be exploited to create an electron
turnstile, a device which passes one electron in every cycle of an external
driving field. Such a device was first realized by Geerligs and coworkers [89]
using a series of metal dots. Here, we discuss a simpler realization of the
quantum-dot turnstile by Kouwenhoven et al. [90,243]. The device is shown
schematically in Fig. 8.1. Electrons are moved one at a time through the dot by
two sinusoidal signals applied to the two tunneling barriers, 180 degrees out of
phase. The rf frequency of the applied signal,f = 10 MHz, is much slower than
189
(c)
Figure 8.1. Schematic potential landscape for a quantum-dot electron turnstile. (a)-(d) are four
stages of an rf cycle. The solid lines indicate the electrochemical potential /ldot for the number of
electrons that are actually on the dot [i.e. N in (a) and (d) and N+ 1 in (b) and (c)]. The dashed
lines indicate /ldot for one extra electron on the dot. The probability for tunneling is large when
the barrier is low (solid arrows), and small when the barrier is high (dashed arrows). During one
cycle an integer number of electrons are transported across the quantum dot. (From
Kouwenhoven et al. [90].)
I
. . ... .. ....... .. ..............
.
I
I
::( 2 .... . .... . . . ....... .1 .

,eo
I
. ....... + . . .. .. . .... . .... . . .

I
I
.... ... .. "1" .....
4~~
- 3----~2----~1--~O--~--~2--~3~
VOLTAGE (mV)
Figure 8.2. Current-voltage characteristics of a quantum-dot electron turnstile. Current plateaus
occur at integer multiples of ef (dotted lines) where the driving frequency f = 10 MHz. (From
Nagamune et at. [16].)
190
the tunneling rate of electrons on and off the dot when the barriers are low.
The driving signal is therefore in the adiabatic limit in which the state of the dot
is fully determined by the electrochemical potential on the side of the low
barrier. As depicted in Fig. 8.1, in one cycle exactly one electron is transferred
across the quantum dot: The cycle begins with N electrons on the dot. The
barrier to the left lead is then lowered allowing an additional electron to enter
the dot. The barrier to the left is then raised, preventing the extra electron from
escaping back to the left. The right barrier is then lowered and the electron
escapes into the right lead. Raising the barrier to the right lead completes the
cycle and returns the dot to its initial configuration with N electrons. By
applying a larger source-drain bias to increase the number of extra electrons
allowed on the dot when the left barrier is lowered, two electrons, or three
electrons, and so on, can be transferred in each cycle. As a result the timeaveraged current passing through the dot is just an integer times the single
electron charge times the driving frequency, 1= nef This current quantization
is clearly observed in Fig. 8.2. Each plateau corresponds to an integer number
of electrons passing through the quantum dot in each cycle.
Recent work by Keller et al. [244] on electron turnstiles has focussed on
the possibility of creating a current standard. A high precision experimental
connection between current and frequency would complement the standards of
voltage and resistance provided by the Josephson and quantum Hall effects.
This in turn would provide a new measurement of the fine structure constant.
Experiments on a series of four metal dots subjected to precisely phased rf
signals have demonstrated a current locked to the rf frequency to an accuracy of
15 parts in 109 [244].
8.2. NON-ADIABATIC REGIME; TIEN-GORDON THEORY.
When the driving signal frequency exceeds the rate at which electrons tunnel
on and off the dot, the state of the dot is no longer simply determined by the
instantaneous values of the applied voltages [245]. In this non-adiabatic regime
it is essential to take into account the phase coherence in time of the electrons
on the quantum dot [246-248]. As an instructive example, consider an isolated
dot containing a single non-degenerate level whose energy is oscillated up and
down in time with respect to the rest of the device. According to Schrodinger's
equation, the electron's wavefunction is given by:
VJ(X,t) = VJ(x) exp[-i!dt' E(t')IIi]
(8.1)
where E(t) = Eo + ev cos(211[t), and VJ(x) is the electron's fixed spatial

wavefunction. From the point of view of the rest of the device the oscillating
191
level does not have a definite energy. Instead it has energy components at 0,
0 /if, 0 2hf, etc. This is simply seen by expanding the phase factor into its
spectral components:
(8.2)
where the weights of the spectral peaks are given by the Bessel functions
Ji ev/hf). Note that one cannot obtain a spectrum with discrete sidebands as
in Eq. (8.2) by the adiabatic procedure of averaging the instantaneous spectrum
over a cycle of the oscillation. Conceptually, the presence of sidebands in the
energy spectrum of a level corresponds to the absorption and emission of
photons from the ac field. Therefore transport involving the sidebands of the
electronic level is commonly referred to as photon-assisted tunneling (PAT).
Many of the experiments on photon-assisted tunneling in quantum dots
[249,250], and in quantum wells [251-254], can be understood in terms of the
theory developed by Tien and Gordon for time-dependent tunneling into a
superconductor [245]. Tien and Gordon's theory assumes two things: First, the
time-dependence must appear entirely through rigid level shifts as in Eq. (8.1).
That is, all oscillating electric fields must be confined to the tunnel barriers.
Second, transitions between regions with different time dependences must
occur only to lowest order in perturbation theory, i.e., according to Fermi's
Golden Rule. In practical application of the theory, the Golden-Rule tunneling
rates across a barrier are simply modified to reflect the changed spectral
densities due to the relative time dependence. For sinusoidal signals, this
corresponds to including the sidebands in Eq. (8.2) into the tunneling rates.
An example in which the Tien-Gordon theory was applied successfully to
transport through a quantum dot is shown in Fig. 8.3. The usual peaks in
current as a function of gate voltage are modified by the application of a
microwave-frequency ac bias across the dot. This modification of the current
can be quantitatively understood within the Tien-Gordon picture. The ac bias
causes an oscillating energy difference between the dot and the leads. The
tunneling rate of electrons on and off the dot are therefore modified according
to [249,255]:
~
eV
r (e) =n~!J( hf)' f(e + nhf)
where ITe) is the tunneling rate in the absence of microwaves.
(8.3a)
192
i
"
~
!z
(el
THEORY
e
~
.!.
a:
a:
a:
a:,
ffi
::>
~,
.~
.s
l?,a
tif
-100
-,10.0
2905
-29.0
-28.5
G...TE VOLTAGE (mV)
3.5
-3.0
2.5
2.0
GATE VOlTAGE (artl.uniIB)
Figure 8.3. Comparison between measurement and Tien-Gordon theory for transport through a
quantum dot. The parameters in the calculation are taken from the experiment; only the ac
amplitudes are adjusted. The conversion of gate voltage to energy in units of hf is indicated by
the arrows where!:::; 27 GHz. (From Kouwenhoven et al. [249].)
Eq. (8.3a) is a special, discrete case of a general description of the

interaction between tunneling electrons and photons in the environment:
(E)
= j d( hf )P(hf )r( E+hf )
(8.3b)
Here the weight function P(hf) is the spectral density function describing the
fluctuations in the environment. These fluctuations include the black body
radiation of the environment [256], the electrical noise that is coupled into the
measurement wires [257], and excitations such as plasmons that can exist in the
current and voltage leads due to their finite impedance [258]. These
fluctuations are broad-band in frequency. One needs to create a special,
resonating environment like an LC-oscillator [259] or apply a microwave signal
at a single frequency to get a photocurrent containing sharp, discrete features.
8.3. PHOTOCURRENT SPECTROSCOPY OF A QUANTUM DOT.
In the experiment of Fig. 8.3, the density of states in the dot is effectively
continuous and one does not see evidence of OD-states. In contrast, a similar
experiment performed on a smaller dot by Oosterkamp and coworkers [44]
clearly reveals the OD-states of the dot. For this case, Fig. 8.4 shows
schematically the processes which lead to peaks in the current vs. gate voltage
193
.-
Figure 8.4. Diagrams of six processes which can lead to a current through a quantum dot with
discrete OD-states driven by microwaves. Eo denotes the groundstate (lower dashed lines) and E}
the first excited state (upper dashed lines) ofthe N-electron system. Without microwaves only the
upper-center diagram can contribute to the current. With microwaves, the indicated inelastic
tunnel processes lead to photon-induced current peaks which occur at distinguishable positions in
gate voltage. (From Oosterkamp et al. [44] .)
20r-------------~--~------------~
61 .SGHz
42GHz
540
Gate Voltage (mV)
520
Figure 8.5. Measured, time-averaged current as a function of center-gate voltage for different
microwave powers at 61.5 and 42 GHz using the device shown in Fig. 1.3. The dashed curves are
without microwaves. The peaks at Eo and E} remain fixed while the photon-assisted-tunneling
sidebands at Eo - hf and E} hf shift proportionally to the applied microwave frequency. (From
Oosterkamp et al. [44].)
194
for a dot driven by microwaves. The Tien-Gordon picture continues to apply

because the ac fields are confined to the barriers and there are no oscillating
fields to cause direct transitions within the dot or leads. Fig. 8.5 shows the
measured current vs. gate voltage for different microwave frequencies and
amplitudes. There are two types of peaks in the current: those associated with
transport through the bare levels Eo and 1, which remain fixed as the
microwave frequency is changed, and those associated with transport through
the sidebands of the levels (PAT) which shift as expected with microwave
frequency. Note that the peak at 1 is only made visible by microwave
excitation of electrons out of Eo, as shown in the bottom center diagram of Fig.
8.5. The observation of the excited-state energy level 1 represents a
spectroscopy of the quantum dot. This spectroscopy requires both the presence
of the microwave field and the measurement of the time-averaged current, so it
is best called a "photocurrent spectroscopy" of the dot.
8.4. RABI OSCILLATIONS IN A DOUBLE QUANTUM DOT.
One interesting example of a system which cannot be treated by Tien-Gordon
theory is a pair of strongly coupled quantum dots connected in series. A
sinusoidal signal of the proper frequency applied to this system will result in a
coherent oscillation (Rabi oscillation) of electrons between the two dots. This
effect lies beyond a Golden-Rule description of transitions between the dots,
and so is not accounted for in the Tien-Gordon model. For the same reason the
Shapiro steps in irradiated Josephson junctions do not follow from a GoldenRule description [97].
In the time-independent case, coherence between the dots is treated
theoretically by solving for the eigenstates of the coupled dot system. In the
time-dependent case, the equivalent approach is to solve for the quasi-energy
eigenstates of the system [260]. As a simple example, which is also relevant to
experiment, consider two coupled dots each of which has a single nondegenerate energy level [261-263]. The Hamiltonian is simply:
H = IA(t)d/d i + w(did! + H.c.)
(8.4)
i=!
where the energies of the states are driven by an external sinusoidal signal, 1 =
0, 2 = ev cos(ro t). Since the Hamiltonian is a periodic function of time H(t +
2rr1ro) = H(t), one can diagonalize the system into eigenstates of the one-period
evolution operator U(t + 2rrJro, t) = T{exp[-(ilh)f:+ 21r/ro dt' H(t')]}. For the
double-dot system, these states have the form [260]:
195
(8.5)
where Ej is thejth quasienergy, and cpF) (t + 2rrJm) = cp}j) (t) is the time periodic
Floquet function whose components give the time-dependent amplitudes on the
two quantum dots. The quasi-energies are plotted in Fig. 8.6. Qualitatively,
each avoided crossing occurs when the levels on the two dots differ by an
integer number of photon energies hm. The gap at each crossing is given by
z2wJn(ev Ihm), which corresponds to the usual symmetric-anti symmetric
splitting, 2w, for the time-independent case weighted by the amplitude of the
nth sideband, In(e V I hm). As in the time-independent case, the wavefunctions
are delocalized at the avoided crossings. At these resonances, if an electron
were placed on one of the dots, it would oscillate back and forth between the
dots at a frequency Q R = 2wJn(ev I hm). For the avoided crossings involving a
nonzero number of photons, this is just the Rabi oscillation familiar from
atomic physics [151].
Time-dependent transport through the double-quantum-dot system coupled
to leads can be characterized by the ratio of the Rabi frequency Q R to the
tunneling rate to the leads r. If Q R is large compared to r then electrons will
perform many coherent oscillations between the dots before each tunneling
event to the leads. The rate-limiting step in transport will therefore be
tunneling to the leads, and so the current will be proportional to r. In the
opposite limit, tunneling to the leads will be fast and only rarely will electrons
tunnel between the dots (in the latter case the Tien-Gordon picture still applies
to tunneling between the dots). These effects are apparent in the left part of
2
:3
...
...........
0
0
E 2/hr.;
Figure 8.6. Calculated quasi-energies of two coupled quantum dots vs. detuning energy 2. Here
fit) = eV cos(rot), with eV = 1; OJ = lOw, where w is the hybridization matrix element between
the two dots. The quasienergies are defined mode 1; OJ). The electronic states on the dots
hybridize and split by 2wJn(eV I( 1; OJ), becoming delocalized, when 2 crosses the nth sideband
of }. (From Stafford and Wingreen [261].)
196
0.8
1.0
0.6
N=l
S 0.5
':l
.....e 0.4
I~
a1
II
""':l
0.2
0.0
2
8
ftc.;
10 12 14
1...;
0.0
-6
1.. ........ ....
":
:E+
C\j:
............ :
..... 1
:
+:I.
~:
':1
C\1
l......t
;:=>
:++
I
:~
-4 J-lL -2
Figure 8.7. Time-averaged current J (in units of Jmox = (eD211) through a double quantum dot
with Ej= -5, E2= 5, r= 0.5, and ac amplitude eV =2,4,6 (increasing J). Energies are given in
units ofw, the tunneling matrix element between the dots. With }.lL=}.lR =0, the system functions
as an electron pump due to coherent n-photon-assisted tunneling. Inset: Time-averaged current at
the one-photon resonance versus dc bias }.lL, with e V = 6. Solid curve: U12 = 0; dotted curve: U12
= 2. The jumps allow one to resolve the Rabi splitting IE+ + EJ and the inter-dot interaction U12
(From Stafford and Wingreen [261].)
Fig. 8.7 where the time-averaged current through a double-quantum-dot system

is plotted vs tiro for different ac driving amplitudes V . Since in Fig. 8.7 the dc
bias is large compared to the coupling to the leads r, the current at the photonassisted-tunneling peaks is given by [261,264].
(8.6)
As shown on the right in Fig. 8.7, the Rabi splitting can be observed directly via
transport measurements, although care must be taken to distinguish it from the
Coulomb interaction Ul2 between electrons on the two dots [262].
Experimentally, time-dependent transport through a double quantum dot
has been studied by Blick et al. [265] and by Fujisawa and Tarucha [266].
These results are best understood by first considering the charging diagram of a
double quantum dot as shown in Fig. 8.8(a) (see also Fig. 4.2). The vertices,
e.g., V and V', correspond to conditions where a pair of electron levels, one on
each dot, become degenerate in energy (Fig. 8.8(b) central panel). Resonant
transport can therefore occur through the two dots in series and one expects a
peak in the current [103]. Such a peak is shown in the bottom panels of Fig.
8.9(a) and (b). If one applies microwaves of energy hfto the double dot, one
also expects enhanced current due to the photon-assisted tunneling processes
shown in the side panels of Fig. 8.8(b). This enhancement is clearly observed
in the top panels of Fig. 8.9. It is natural to expect that the Rabi splitting, and
197
possibly time-resolved Rabi oscillations, will be observed m such a doublequantum-dot structure in the future.
(n.1 , m)
(n1, m1)
(b)
' on tholino PLA '
' at a point V'
' on tholino PRL'
Figure 8.8. (a) Schematic charging diagram of the coupled dot system. (n, m) gives the number
of electrons on the left and right dots, respectively. The OD-OD resonant-tunneling peaks occur at
the vertices, e.g. V and V'. The thick lines,PLR, PRL , and so on, indicate the conditions for
resonant photon-assisted tunneling. (b) Energy diagram for photon-assisted tunneling on the line
PLR , for ordinary resonant tunneling at the point V, and for photon-assisted tunneling on the line
PRL (From Fujisawa and Tarucha [266].)
640 642 644 646 648 -650 -652

V GR (mV)
Figure 8.9. (a) Current vs. two gate voltages for increasing microwave power from bottom to top
panel. On applying microwaves the photon sideband becomes visible. (b) Contour plot of the
observed current near point V' in Fig. 8.8. Note the clear signature of photon-assisted tunneling
along the segment P'LR. (From Fujisawa and Tarucha [266].)
198
9. Conclusions and future directions.

The field of electron transport through quantum dots has progressed from its
first tentative steps to maturity in less than ten years. This startling rate of
progress might be attributed to a confluence of fabrication, refrigeration, and
measurement technologies. It may be more honest, however, to attribute the
rapid progress to the fundamental simplicity of the behavior of electrons in
quantum dots. New experimental results have rarely waited more than a year or
two to find satisfactory theoretical explanation. "What is that makes quantum
dots so simple?" The answer is the strong separation of energy scales in dots.
The largest relevant energy is the Coulomb interaction energy, -1 meV in
lateral dots and -10 mV in vertical dots. (All energies larger than this, say the
bandgap or intervalley energies of GaAs, are frozen out and play no role in the
dynamics of the dot.) The next relevant energy scale is the single particle level
spacing, -0.1 me V in lateral dots and -1 mV in vertical dots. Last is the
coupling energy between the dot and the leads which for opaque tunnel barriers
is -0.01 meV. The energy scale set by the temperature merely determines
which of these other scales can be resolved in transport of electrons through the
dot.
As a result of the separation of energy scales, the behavior of electrons in
dots can often be understood in a simple hierarchical way: First, the number of
electrons on the dot is determined by minimizing the direct Coulomb
interaction energy. Second, the state of these electrons on the dot is determined
by balancing their kinetic energy against the residual parts of the Coulomb
interaction, including correlation and exchange effects. Finally, the transition
rates among such states are determined by the small hybridization energy to the
leads. When this hierarchical scheme applies, the agreement between
experiment and theory is often startlingly good. The few outstanding
experimental puzzles in transport through dots correspond to those cases when
two or more energy scales are brought into competition. Examples that we
have discussed include: charge fluctuations or co-tunneling events between dots
and leads or between two dots when the tunnel coupling energy is equal to or
larger than the single-particle energy separation; the formation of Landau levels
at high magnetic field with a Zeeman energy or cyclotron energy of order either
the single-particle energies or the Coulomb energy.
The overall simplicity of transport through dots may in the long run prove
to be the field's greatest blessing. This simplicity has certainly permitted the
accumulation of a core of well explored and well understood phenomena.
While no single great discovery has characterized the study of electron
transport in dots, many small discoveries coming in rapid succession have
added up to a big advance. Today, researchers armed with fabrication
199
techniques developed over the past ten years, and also armed with a good
understanding of the basic phenomenology of dots, are pursuing many new
directions both technological and scientific. Dots are being used in the study of
other systems, dots and other mesoscopic structures are combined into minilaboratories on a chip, and the complex regimes where several energy scales
compete in a dot are under exploration. Some of this current research has been
touched on in the previous chapters. It seems appropriate in the remaining
space to point to a few directions which seem most promising in the near and
not so near future. The following sections briefly address (9.1) technological
and (9.2) novel scientific applications of transport through quantum dots, and
(9.3) quantum dot physics in other systems.
9.1. TECHNOLOGICAL APPLICATIONS.
The ability to measure and control current at the single-electron level has a
number of potential uses, ranging from metrology to electrometry to computing
[267]. In fact, both metal and semiconductor quantum dots are already finding
niche applications, though their utility is limited because of the low
temperatures required. To broaden their usage, devices must be developed that
operate under ambient conditions, i.e. at room temperature. Ways of
accomplishing this will be discussed in section 9.3.
One of the most important Coulomb blockade application is singleelectrometry - the detection of single charges. As discussed already, these
devices are very sensitive to small changes in their local electrostatic
environment. Sensitivities of 10-5 elHz l12 are possible [268]. In other words,
the electrometer can detect a charge e in one second if 10-5 of the field lines
leaving the charge terminate on the dot. These devices are the electrostatic
counterpart to the SQUID, a superconducting device which is sensitive to
extremely small magnetic fluxes. There are important differences, however
[267]. SQUIDs can be used to measure macroscopic magnetic fields by
utilizing flux transformers to couple the macroscopic magnetic field into the
SQUID. No such transformer exists to date for electric charge, so the change in
the charge over a large object cannot be carefully measured. Nevertheless, as a
local electrometer, semiconductor as well as metallic dots may find many uses.
Already, they have been used in scientific applications, mainly to monitor the
behavior of single electrons in other circuits. We have discussed the
semiconductor electrometers in section 4 (e.g. see Fig. 4.1), discussions of the
metallic electrometers can be found in Refs. 93,268-270.
Another application is in the field of metrology. The single electron
turnstile, and related devices in metal dots, are being investigated as current
standards. They produce a standardized current from a standardized RF
200
frequency, with the conversion factor being the electronic charge e. Accuracies
of 15 parts in 109 have been obtained in multi-dot metallic circuits [244].
These turnstiles would complete the solid-state device "metrology triangle"
relating frequencies, currents and voltages [24]. Already, the quantum Hall
effect is used to relate current to voltage, and the Josephson effect to relate
frequency to voltage. The turnstile would fill in the last leg of the triangle by
relating frequencies to currents.
Another application is the measurement and regulation of temperature.
As discussed in section 3, the Coulomb blockade peak widths are proportional
to kBT, and can, once calibrated, be used to measure the temperature of the dot
or its surroundings. Even at higher temperatures, where most of the Coulomb
structure has been washed out, there are slight non-linearities in the /- V
characteristic that can be used to measure T [271]. Temperature gradients can
also be detected, as thermopower measurements of dots have shown [272].
Quantum dots may be able to control the temperature as well as measure it. A
quantum dot "refrigerator" that can cool a larger electronic system has been
proposed [122]. The idea is to use tunneling through single quantum levels to
skim off the hot electrons above EF , thereby cooling the electron system.
The experiments discussed in section 8 showed that photon-assisted
tunneling over the Coulomb gap can induce DC currents through a quantum
dot. This suggests applications for dots as photon detectors in the microwave
regime. The tunability of the dot potential relative to the source and drain
means that the detector can be frequency-selective. It is even possible for a
single photon to lead to a current of many electrons [273]. Photon-detection
applications are not limited to the microwave region. For example, a metallic
dot operating as a single-electrometer has been utilized to (indirectly) detect
visible photons. The dot was fabricated on a semiconductor substrate, and was
then used to electrostatically detect the presence of photoexcited electrons
within the semiconductor [273].
One can also contemplate electronics applications for these devices - a
field sometimes called single-electronics. It is in principle possible to perform
calculations using quantum dot circuits, based on either charging [274] or
quantum-coherent phenomena [88], although little experimental work has been
done in this direction. Multi-dot circuits can also serve as static memory
elements. This has been tested in the laboratory; for example, a single-electron
memory with a hold time of several hours (at millikelvin temperatures) has
been demonstrated [93]. One must exercise extreme caution in extrapolating
these successes to a useful product, however. The technological barriers to
creating complex circuits that work in the real world are enormous.
201
9.2. SCIENTIFIC APPLICATIONS.

One of the most promising scientific directions in quantum dot research is the
use of dots as part of on-chip laboratories. The first steps have already been
taken in this direction, with encouraging results. As discussed in section 4,
combining two or more dots in close proximity has allowed an exploration of
the crossover from a double dot to a single dot as the barrier between the dots is
removed. An important question is: "How do charge fluctuations drive this
crossover?" The systematic control offered by the double dot structure is a
powerful tool for answering this question. Another phenomenon susceptible to
study in double-dot structures is the coherent delocalization of single-particle
levels between dots. In the presence of ac fields, this delocalization
corresponds to the Rabi oscillations discussed in section 8. By extending
delocalization to multiple dots, the formation of coherent bands is possible
[37].
Multiple dot structures are only one possibility for on-chip laboratories.
Dots, wires, rings, and gates can be integrated into more complex structures. A
beautiful example of this kind of integrated structure was employed in a series
of experiments on quantum coherence by researchers at the Weizmann Institute
[126,127]. In the experiments, a quantum dot was embedded in one arm of an
Aharonov-Bohm ring; see Fig. 4.10. By measuring the amplitude and phase of
the resulting Aharonov-Bohm oscillations the coherent transmission amplitude
of the dot, including the phase-shift, was determined. In addition to proving
that transmission through dots can be coherent, the research uncovered an
unexpected phase-slip between Coulomb-blockade conductance peaks. Perhaps
most importantly the experiments have opened up the possibility of studying, in
a controlled way, dephasing of quantum transport by the environment.
Fig. 9.1 contains a schematic of such an integrated on-chip laboratory for
studying dephasing. The Aharonov-Bohm ring plus quantum dot is augmented
by a quantum point contact in close proximity to the dot. This quantum point
contact forms a controllable "environment" for electrons on the dot. An extra
electron on the dot changes the transmission amplitude through the point
contact. Hence the point contact acts as an electrometer for the number of
electrons on the dot [94]. Since number and phase are conjugate variables, the
quantum point contact results in dephasing of electron transport through the
dot, and suppresses the Aharonov-Bohm oscillations [275,276].
The technological and scientific applications are of course connected. For
example, we have mentioned in sections 4 and 9.1 the possibility of using
quantum dots as elements in a quantum computer [87,88]. The construction of
even a simple prototype quantum computer out of solid state elements is
technologically extremely complicated and will not be accomplished during
202
this century. Nevertheless, the ideas around quantum computation do generate
new scientific directions. One direction is the measurement and control of
dephasing. The idea is that even though a quantum dot may be a nondissipative system, the electrons on the dot interact with electrons, or more
generally, with other degrees of freedom in the environment such as the nearby
point contact in Fig. 9.1. In the ring-dot-point-contact geometry the interaction
collapses the wavefunction and the interference between the amplitudes
traveling along the two arms of the ring gets suppressed. Since a quantum
computer should be fully coherent, dephasing simply implies an error.
Therefore, control over the environment is a necessary requirement for
successfully building a quantum computer. We foresee in the near future a
research direction which could be described as mesoscopic environmental
engineering. Another direction stimulated by the recent proposals on quantum
computers is the control in time of bits. (For a quantum computer the bits are
called qubits.) This control in time is called handshaking in ordinary computers
and could be called quantum handshaking in quantum computers since the
control needs to occur within the phase coherence time. For quantum dots it
means that the single electron tunneling events are regulated on times scales as
short as 1 ns to 1 ps. Experiments such as observing the predicted Rabi
oscillations [261,262] and the control of tunneling using short pulses [246,247]
would be a first step to accomplishing quantum handshaking.
ope
----------~----------Figure 9.1. Schematic view of the "Which Path?" interferometer [275]. A quantum dot is built
in one arm of an Aharonov-Bohm ring. The transmission amplitude of the nearby quantum point
contact depends on the occupation number of the dot. Since number and phase are conjugate, the
quantum point contact produces dephasing of electrons passing through the dot, and suppresses
the Aharonov-Bohm oscillations. (From Aleiner et al. [276].)
203
9.3. OTHER SYSTEMS.

Quantum dots are really just a generic example of a small, confined structure
containing electrons. There is no fundamental physical discontinuity between a
quantum dot and a large molecule or even an atom. There should be no
surprise then, that the physics of dots applies as well to small metallic particles,
clusters, and molecules [277]. In this section, we point out a few recent
examples where the ideas and measurement techniques developed in the study
of quantum dots have been applied to ever-smaller systems.
The analogy to quantum dots is particularly clean in the case of metal
nanoparticles. In experiments at Harvard, Al particles of a few nanometer size
were studied in a Coulomb blockade geometry [29]. The charging energy -10
meV, and level spacings -0.1 meV, appear in J- V traces in exactly the same
fashion as in semiconductor dots. The larger separation between charging
energy and level spacing and the Fermi liquid nature of the states on the
nanoparticle in fact make the metallic case somewhat easier to understand in
detail [278]. In addition, the rich behavior introduced by superconductivity in
the dots and/or the leads makes these nanoparticles a topic of ongoing interest.
[29,279]
Another promising approach utilizes metal or semiconductor nanoparticles
made by synthetic chemistry and subsequently incorporated into electrodes.
Fig. 9.2 shows an example taken from Klein et al. [280]. Six nm diameter
CdSe nanocrystals are bound to electrodes using a molecular linker. The
conductance versus gate voltage shows Coulomb oscillations; nonlinear
measurements reveal a charging energy of - 30 meV.
The use of molecules as Coulomb blockade structures is not merely
theoretical speculation. For example, Porath et al. [281], has recently used an
STM to explore transport through C60 molecules deposited on a gold substrate.
These measurements clearly show features associated with Coulomb blockade
and level quantization [281]. Fig. 9.3(a) shows a schematic of the measurement
geometry, and Fig. 9.3(b) shows dJldV,d as a function of V,d. A large gap is
observed, followed by a series of peaks associated with tunneling into the
excited states of the molecule. The large gap is a combination of Coulomb
charging (-0.4 eV) and the energy-level difference between the lowest
unoccupied molecular orbital (LUMO) and highest occupied molecular orbital
(HOMO) level. This HOMOILUMO band gap is easily incorporated in
standard models of the Coulomb blockade, and agreement with the experiment
is good. Joachim and Gimzewski [282] have recently shown that single C60
molecules can operate as an amplifier through their electromechanical
properties. Other molecules have also been studied, including a recent report
of a gated single-electron transistor operating at room temperature [283].
204
(b)
10
10
v, (V)
Figure 9.2. Single nanocrystal transistor. (a) SEM micrograph of 5.5 nm diameter CdSe
nanocrystals bound to lithographically patterned gold electrodes. (b) Inset: Schematic of the
device, showing a single nanocrystal bridging the leads. A gate voltage can be applied to the
conducting substrate. Main Panel: Conductance versus gate voltage measured at 4.2 K showing
three Coulomb oscillations. (From Klein et al. [280].)
1.0
~
V
C6C)
1-1
Insulating "-...: Y 1-2
layer ~
.C2
O.B
.!.
0.0
Of
0.2
/-
gold
0.0
-1.0
0.'
0.0
0.'
1.0
&8s[VJ
Figure 9.3. Left: Schematic diagram and equivalent circuit of a double-junction system realized
by a C60 molecule weakly coupled to an STM tip and to a gold substrate. Right: Tunneling
spectroscopy d//dVsd as a function of Vsd at 4.2 K. The first discrete state observed for negative
bias corresponds to the HOMO and the first state for positive bias corresponds to the LUMO.
(From Porath et at [281].)
Carbon nanotubes, the extended cousins of Cw have also proven to be a

system that can be understood using the ideas developed for dots [284,285].
The nanotube is predicted to act as a one-dimensional quantum wire, and a
finite length turns it into a one-dimensional quantum dot. Fig. 9.4(a) shows a
bundle of single walled carbon nanotubes to which electrical leads have been
205
patterned. The conductance on a 200 nm segment between two of the contacts

versus gate voltage shows Coulomb oscillations, as is seen in Fig. 9.4(b). The
inset shows the temperature dependence; the peak height decreases with
increasing temperature, indicating resonant tunneling though a single quantum
level delocalized over the entire length of the tube. Nonlinear transport
measurements indicate that the charging energy is - lameV and the level
spacing is - 3 meV, consistent with estimations for a ID conductor of - 200
nm in length.
Clearly, these molecular systems offer many exciting options for future
research. Since the charging and level spacings are quite large, it is possible to
investigate physics that lies at lower energy scales than is accessible in
lithographically patterned quantum dots. For example, the long-standing
prediction of a Kondo resonance between a localized spin on a quantum dot
and the Fermi seas in the leads may finally succumb to experimental
investigation. Of course, there are new phenomena in these systems as well:
superconductivity in metal particles with level spacings as large as the
superconducting gap, surface states and bandgap pinning in clusters, and strong
electron-lattice interactions in molecules, etc. If the history of the field has
been any guide, new surprises also await us in these systems in the years to
come.
0.12
(b)
0.10
!!.
\.:)
! ;
~0.02
0.06
T!K)
: : 1.1
ffc
0.0&
.~
0.04
0.00
Q.IO
0,(14
0. 15
V. (V)
0.Q2
0.00
0.4
0.6
0.8
1.0
1.2
1.4 1.6
Y,(V)
1.8
2.0
2.2
Figure 9.4. (a) AFM image of a single-walled nanotube bundle to which multiple electrical leads
have been attached. A gate voltage can be applied to the conducting substrate to change the
number of electrons on the tubes. (b) Main Panel: Measurement of the conductance versus gate
voltage of the 200 nm segment between the leftmost leads. Dramatic Coulomb oscillations are
observed. Inset: Temperature dependence of a selected peale. The peak height increases as the
temperature is lowered, indicating coherent transport though a single quantum level. (From
Bockrath et al. [284].)
206
Acknowledgements:
We gratefully acknowledge our (numerous!)
collaborators at UC Berkeley, Delft, Harvard, MIT, NEC, NTT, Philips,
Stanford, and the University of Tokyo who, with the authors, performed most
of the work presented here. We also thank our colleagues R. Ashoori, R. Akis,
H. Bruus, C.W.J. Beenakker, A.M. Chang, D. Ferry, R.A. lalabert, O. Klein,
D.A. Wharam, T. Schmidt, U. Sivan, M. Stopa, H. Tamura, and A. Yacoby who
graciously provided figures for this review.
References.
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[to]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
M. Reed, Scientific American 268, 118 (1993).

M. Kastner, Physics Today 46, 24 (1993).
Single Charge Tunneling, edited by H. Grabert and M.H. Devoret (Plenum Press, 1991).
D.V. Averin and K.K Likharev, J. Low Temp. Phys. 62, 345, (1986); and in: Mesoscopic
Phenomena in Solids, edited by B.L. Altshuler, P.A. Lee, and RA. Webb (Elsevier,
1991).
G. SchOn and A.D. Zaikin, Physics Reports 198, 237 (1990); G.-L. Ingold and Yu. V.
Nazarov, in Ref. 3; R. Fazio and G. SchOn, in this volume.
C.W,J. Beenakker, Phys. Rev. B 44, 1646 (1991).
H. van Houten, C.W,J. Beenakker, A.A.M. Staring, in Ref. 3.
U. Meirav and E.B. Foxman, Semicond. Sci. Technol. 10, 255 (1995).
L.P. Kouwenhoven and P.L. McEuen, in Nano-Science and Technology, edited by G.
Timp (AlP Press, 1997). Parts of the present introduction sections can also be found in
this review.
Special issue on Single Charge Tunneling, edited by H. Grabert, in Zeitschrift fur Physik
B 85 (1991).
Special issue on Few-Electron Nanostructures, edited by L,J. Geerligs, C,J.P.M.
Harmans, and L.P. Kouwenhoven, in Physica B 189 (1993).
C.J.P.M. Harmans, Physics World 5,50 (1992).
KK Likharev and T. Claeson, Scientific American 266, 50 (1992).
M.H. Devoret, D. Esteve, and C. Urbina, Nature 360,547 (1992).
RC. Ashoori, Nature 379, 413 (1996).
Y. Nagamune, H. Sakaki, L.P. Kouwenhoven, L.c. Mur, C.J.P.M. Harmans, J. Motohisa,
and H. Noge, Appl. Phys. Lett. 64, 2379 (1994).
L.1. Glazman and R1. Shekhter, J. Phys.: Condens. Matter 1, 5811 (1989).
L. P. Kouwenhoven, N.C. van der Vaart, A.T. Johnson, W. Kool, C,J.P.M. Harmans, J.G.
Williamson, A.A.M. Staring, and C.T. Foxon, Z. Phys. B 85,367 (1991).
RA. Millikan, Phys. Rev. 32, 349, (1911).
C,J. Gorter, Physica 17, 777 (1951).
1. Giaever and H.R Zeller, Phys. Rev. Lett. 20,1504 (1968).
J. Lambe and RC. Jaklevic, Phys. Rev. Lett. 22,1371 (1969).
1.0. Kulik and R.L Shekhter, Zh. Eksp. Teor. Fiz. 68, 623 (1975) [SOy. Phys. JETP 41,
308 (1975)].
KK Likharev, IEEE Trans. Magn. 23,1142 (1987); IBM J. Res. Dev. 32, 144 (1988).
K. Mullen, E. Ben Jacob, RC. Jaklevic, and Z. Schuss, Phys. Rev. B 37, 98 (1988); M.
Amman, K Mullen, and E. Ben-Jacob, J. Appl. Phys. 65, 339 (1989).
207
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
T.A. Fulton and G.I. Dolan, Phys. Rev. Lett. 59,109 (1987).
L.L. Sohn, M. Crommie, and H. Hess, in this volume.
C. Schonenberger, H. van Houten, H.C. Donkersloot, Europhys. Lett. 20, 249 (1992).
D.C. Ralph, C.T. Black, J.M. Hergenrother, J.G. Lu, and M. Tinkham, in this volume.
D.L. Klein, P.L. McEuen, lE. Bowen Katari, R Roth, A.P. Alivisatos, Appl. Phys. Lett.,
68, 2574 (1996).
M.A. Reed, J.N. Randall, R.I. Aggarwall, R.I. Matyi, T.M. Moore, and A.E. Wetsel,
Phys. Rev. Lett. 60, 535 (1988).
L.P. Kouwenhoven, G. Schon, and L.L. Sohn, in this volume.
B.I. van Wees, H. van Houten, C.W.I. Beenakker, lG. Williamson, L.P. Kouwenhoven,
D. van der Marel, and C.T. Foxon, Phys. Rev. Lett. 60, 848 (1988).
D.A. Wharam, T.l Thornton, R Newbury, M. Pepper, H. Ahmed, lE.F. Frost, D.G.
Hasko, D.C. Peacock, D.A. Ritchie, and G.A.C. Jones, l Phys. C 21, L209 (1988).
C.W.I. Beenakker and H. van Houten, Solid State Physics 44, 1 (1991). This reviews
transport in meso scopic systems in which single electron charging is not important.
B.I. van Wees, L.P. Kouwenhoven, C.I.P.M. Harmans, lG. Williamson, C.E.
Timmering, M.E.1. Broekaart, C.T. Foxon, and J.J. Harris, Phys. Rev. Lett. 62, 2523
(1989).
L.P. Kouwenhoven, F.W.I. Hekking, B.I. van Wees, C.I.P.M. Harmans, C.E. Timmering,
and C.T. Foxon, Phys. Rev. Lett. 65, 361 (1990).
lH.F. Scott-Thomas, S.B. Field, M.A. Kastner, H.I. Smith, and D.A. Antonadis, Phys.
Rev. Lett. 62, 583 (1989).
H. van Houten and C.W.I. Beenakker, Phys. Rev. Lett. 63, 1893 (1989).
U. Meirav, M.A. Kastner, H. Heiblum, and S.I. Wind, Phys. Rev. B 40, 5871 (1989).
S.B. Field, M.A. Kastner, U. Meirav, lH.F. Scott-Thomas, D.A. Antonadis, H.I. Smith,
and S.I. Wind, Phys. Rev. B 42, 3523 (1990).
A.A.M. Staring, H. van Houten, C.W.I. Beenakker, and C.T. Foxon, High magnetic fields
in semiconductor physics III, edited by G. Landwehr (Springer, Berlin, 1990).
M.A. Kastner, Rev. Mod. Phys. 64, 849 (1992).
T.H. Oosterkamp, L.P. Kouwenhoven, A.E.A. Koolen, N.C. van der Vaart, and
C.J.P.M. Harmans, Phys. Rev. Lett. 78, 1536 (1997).
M. Stopa, Phys. Rev. B 48,18340 (1993); and Phys. Rev. B 54,13767 (1996).
M. Macucci, K Hess, and G.I. Iafrate, Phys. Rev B 48,17354 (1993).
D. Jovanovic, and J.P. Leburton, Phys. Rev. B 49, 7474 (1994).
B. Meurer, D. Heitmann, and K. Ploog, Phys. Rev. Lett. 68, 1371 (1992).
W. Hansen, T.P. Smith, III, KY. Lee, lA. Brum, C.M. Knoedler, J.M. Hong, and D.P.
Kern, Phys. Rev. Lett. 62, 2168 (1989).
R.C. Ashoori, H.L. StOrmer, lS. Weiner, L.N. Pfeiffer, KW. Baldwin, and KW. West,
Phys. Rev. Lett. 68, 3088 (1992); RC. Ashoori, H.L. StOrmer, lS. Weiner, L.N. Pfeiffer,
SJ. Pearton, KW. Baldwin, and KW. West, Phys. Rev. Lett. 71, 613 (1993).
A.N. Korotkov, D.V. Averin, and K.K. Likharev, Physica B 165&166,927 (1990); D.V.
Averin, A.N. Korotkov, and KK Likharev, Phys. Rev. B 44, 6199 (1991).
Y. Meir, N.S. Wingreen, and P.A. Lee, Phys. Rev. Lett. 66, 3048 (1991).
D.V. Averin and Yu.V. Nazarov, in Ref. 3.
Yu.V. Nazarov, l Low Temp. Phys. 90, 77 (1993).
D.V. Averin, Physica B 194196,979 (1994).
H. Schoeller, Physica B 194196, 1057 (1994); H. Schoeller and G. SchOn, Phys.
Rev. B 50, 18436 (1994)
208
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
AD. Stone, and PA Lee, Phys. Rev. Lett. 54, 1196 (1985).
L.I. Geerligs, D.V. Averin, and J.E. Mooij, Phys. Rev. Lett. 65, 3037 (1990); and L.I.
Geerligs, M. Matters, and J.E. Mooij, Physica B 194-196, 1267 (1994).
T.M. Eiles, G. Zimmerli, H.D. Jensen, and J. Martinis, Phys. Rev. Lett. 69, 148 (1992).
c. Pasquier, U. Meirav, F.I.B Williams, D.C. Glattli, Y. Jin, and B. Etienne, Phys. Rev.
Lett. 70,69 (1993); and D.C. Glatdi, Physica B 189, 88 (1993).
L.1. Glazman and M.E. Raikh, Pis'ma Zh. Eksp. Teor. Fiz. 47, 378 (1988) [JETP Lett. 47,
452 (1988)].
T.K. Ng and PA Lee, Phys. Rev. Lett. 61, 1768 (1988).
A Kawabata, J. Phys. Soc. Jpn. 60, 3222 (1991).
S. Hershfield, J.H. Davies, and J.W. Wilkins, Phys. Rev. Lett. 67, 3720, (1991); Phys.
Rev B 46, 7046 (1992).
Y. Meir, N.S. Wingreen, and PA Lee, Phys.Rev. Lett. 70, 2601 (1993).
T. Inoshita, A Shimizu, Y. Kuramoto, H. Sakaki, Phys. Rev. B 48, 14725 (1993).
J. Konig, H. Schoeller, and G. Schon, Phys. Rev. Lett. 76, 1715 (1996); J. Konig, J.
Schmid, H. Schoeller, and G. Schon, Phys. Rev. B 54,16820 (1996); H. Schoeller, in this
volume.
D.C. Ralph and RA Buhrman, Phys. Rev. Lett. 72, 3401 (1994).
I.K. Yanson, V.v. Fisun, R Hesper, AV. Khotkevich, J.M. Krans, J.A Mydosh, and
J.M. Ruitenbeek, Phys. Rev. Lett. 74, 302 (1995).
A. Kumar, S.E. Laux, F. Stern, Phys. Rev. B 42,5166 (1990).
G.W. Bryant, Phys. Rev. B 39, 3145 (1989).
D. van der Mare!, in Nanostructure Physics and Fabrication, eds. M.A Reed and W.P.
Kirk (Academic Press 1989).
L. Wang, J.K. Zhang, and A.R Bishop, Phys. Rev. Lett. 73, 585 (1994).
U. Merkt, J. Huser, M. Wagner, Phys. Rev. B 43, 7320 (1991).
D. Pfannkuche, R.R. Gerhardts, P.A. Maksym, and V. Gudmundsson, Physica B 189, 6
(1993); see also references therein.
N.F. Johnson, J. Phys.: Condens. Matter 7,965 (1995).
M. Wagner, U. Merkt, and AV. Chaplik, Phys. Rev. B 45, 1951 (1992).
Bo Su; V.I. Goldman; J.E. Cunningham, Surface Science 305,566 (1994).
J.G. Williamson, AAM. Staring, L.P. Kouwenhoven, H. van Houten, C.W.J. Beenakker,
C.E. Timmering, M. Mabesoone, and C.T. Foxon, in: Nanostructures and Mesoscopic
Systems, edited by M.A Reed, and W.P. Kirk (Academic Press, 1991).
E.B. Foxman, P.L. McEuen, U. Meirav, N.S. Wingreen, Y. Meir, P.A. Belk, N.R Belk,
MA Kastner, and S.I. Wind, Phys. Rev. B 47, 10020 (1993).
E.B. Foxman, U. Meirav, P.L. McEuen, MA Kastner, O. Klein, PA Belk, D.M.
Abusch, and S.I. Wind, Phys. Rev B 50, 14193 (1994).
A.T. Johnson, L.P. Kouwenhoven, W. de Jong, N.C. van der Vaart, C.I.P.M. Harmans,
and C.T. Foxon, Phys. Rev. Lett. 69, 1592 (1992).
N.C. van der Vaart, A.T. Johnson, L.P. Kouwenhoven, D.I. Maas, W. de Jong, M.P. de
Ruyter van Steveninck, A. van der Enden, C.I.P.M. Harmans, and C.T. Foxon, Physica B
189, 99 (1993).
D. Weinmann, W. Hausler, W. Pfaff, B. Kramer, and U. Weiss, Europhys. Lett. 26,467
(1994); D. Weinmann, W. Hausler, B. Kramer, Phys. Rev. Lett. 74, 984 (1995).
J. Weis, R.I. Haug, K. v. Klitzing, and K. Ploog, Phys. Rev. Lett. 71, 4019 (1993).
L.P. Kouwenhoven, Science 268, 1440 (1995).
RP. Feynman, Foundations of Physics 16, 507 (1986).
209
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
[98]
[99]
[100]
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]
[110]
[111]
[112]
[113]
[114]
[115]
S. Lloyd, Science 261, 1569 (1993).

LJ. Geer1igs, V.F. Anderegg, P.A.M. Ho1weg, lE. Mooij, H. Pothier, D. Esteve, C.
Urbina, M.H. Devoret, Phys. Rev. Lett. 64, 2691 (1990).
L.P. Kouwenhoven, A.T. Johnson, N.e. van der Vaart, A. van der Enden, and C.J.M.P.
Harmans, Phys. Rev. Lett. 67, 1626 (1991).
H. Pothier, P. Lafarge, P.F. Orfi1a, C. Urbina, D. Esteve and M.H. Devoret, Physica B
169,573 (1991).
H. Pothier, P. Lafarge, C. Urbina, D. Esteve, and M.H. Devoret, Europhys. Lett. 17,249
(1992).
P.D. Dresse1haus, L. Ji, S. Han, lE. Lukens, and KK Likharev, Phys. Rev. Lett. 72,
3226 (1994).
M. Field, e.G. Smith, M. Pepper, D.A. Ritchie, lE.F. Frost, G.A.C. Jones, and D.G.
Hasko, Phys. Rev. Lett. 70, 1311 (1993).
F. Hoffmann, T. Heinzel, D.A. Wharam, lP. Kotthaus, G. Bohm, W. Klein, G. Trankel,
and G. Weimann, Phys. Rev. B 51, 13872 (1995).
e.H. Crouch, Ph.D. Thesis, Harvard University (1996).
M. Tinkham, Superconductivity, 2nd Ed. (McGraw-Hill, New York, 1996).
L.I. Glazman and V. Chandrasekhar, Europhys. Lett. 19,623 (1992).
I.M. Ruzin, V. Chandrasekhar, E.I. Levin, and L.I. Glazman, Phys. Rev. B 45, 13469
(1992).
M. Kemerink, and L.W. Mo1enkamp, Appl. Phys. Lett. 65, 1012 (1994).
F.R. Waugh, MJ. Berry, C.H. Crouch, C. Livermore, DJ. Mar, R.M. Westervelt, KL.
Campman, and A.C. Gossard, Phys. Rev. B 53, 1413 (1996).
F.R. Waugh, MJ. Berry, DJ. Mar, R.M. Westervelt, KL. Campman, and A.e. Gossard,
Phys. Rev. Lett. 75,705 (1995).
N. C. van der Vaart, S.F. Godijn, Y.V. Nazarov, CJ.P.M. Harmans, lE. Mooij, L.W.
Molenkamp, and C.T. Foxon, Phys. Rev. Lett. 74,4702 (1995); and N. C. van der Vaart,
Ph.D. Thesis, T.U. Delft (1995).
e. Livermore, C.H. Crouch, R.M. Westervelt, KL. Campman, and A.C. Gossard, Science
274, 1332 (1996).
A.S. Adourian, C. Livermore, R.M. Westervelt, KL. Campman, and A.C. Gossard,
Superlattices and Microstructures 20, 411 (1996); and submitted to Appl. Phys. Lett.
(1997).
R.H. Blick, R.J. Haug, J. Weis, D. Pfannkuche, K v. Klitzing, K Eberl, Phys. Rev. B 53,
7899 (1996).
D. Dixon, L.P. Kouwenhoven, P.L. McEuen, Y. Nagamune, l Motohisa, N. Sakaki,
Phys. Rev. B 53, 12625 (1996).
S.V. Panyukov and A.D. Zaikin, Phys. Rev. Lett. 67, 3168 (1991).
D.S. Golubev and A.D. Zaikin, Phys. Rev. B 50, 8736 (1994).
G. Fa1ci, l Heins, G. Schon, and G.T. Zimanyi, Physica B 203,409 (1994); G. Falci, G.
SchOn, and G.T. Zimanyi, Phys. Rev. Lett. 74, 3257 (1995); l Konig, H. Schoeller, and
G. Schon, Europhys. Lett. 31, 31 (1995).
H. Grabert, Phys. Rev. B 50, 17364 (1994).
C.A. Stafford and S. Das Sarma, Phys. Rev. Lett. 72, 3590 (1994).
G. Klimeck, Guanlong Chen, and S. Datta, Phys. Rev. B 50, 2316 (1994).
Guanlong Chen, G. Klimeck, S. Datta, Guanha Chen, and W.A. Goddard III, Phys. Rev.
B 50, 8035 (1994).
KA. Matveev, Phys. Rev. B 51, 1743 (1995).
210
[116]
[117]
[118]
[119]
[120]
[121]
[122]
[123]
[124]
[125]
[126]
[127]
[128]
[129]
[130]
[131]
[132]
[133]
[134]
[135]
[136]
[137]
[138]
[139]
[140]
[141]
[142]
[143]
[144]
[145]
[146]
[147]
lM. Golden and B.I. Halperin, Phys. Rev. B 53, 3893 (1996).
K.A Matveev, L.I. G1azman, and H.D. Baranger, Phys. Rev. B 53,1034 (1996).
J.M. Golden and B.I. Halperin, Phys. Rev. B 54, 16757 (1996).
M. Stopa, private communication (1995).
L.W. Mo1enkamp, K. Flensberg, and M. Kemerink, Phys. Rev. Lett. 75,4282 (1995).
C.H. Crouch, C. Livermore, RM. Westervelt, K.L. Campman, and A.C. Gossard,
submitted to App!. Phys. Lett. (1997).
H.L. Edwards, Q. Niu, and AL. de Lozanne, App!. Phys. Lett. 63,1815 (1993).
P.A. Lee and T.V. Ramakrishnan, Rev. Mod. Phys. 57, 287 (1985).
B.L. Altshuler and AG. Aronov, in Electron-Electron Interaction in Disordered Systems,
eds. AL. Efros and M. Pollak (Elsevier, Amsterdam, 1985).
S. Washburn and RA Webb, Adv. Phys. 35, 375 (1986).
A Yacoby, M. Heiblum, D. Mahalu, and Hadas Shtrikman, Phys. Rev. Lett. 74,4047
(1995).
R Schuster, E. Buks, M. Heiblum, D. Mahalu, V. Umansky, H. Shtrikman, Nature 385,
417 (1997).
C.lB. Ford, M. Field, P.I. Simpson, M. Pepper, D. Popovie, D. Kern, lE. Frost, D.A.
Ritchie, and G.A.C. Jones, Inst. Phys. Conf. Ser. 127,235 (1992).
T. Fujisawa, T. Bever, Y. Hirayama, and S. Tarucha, l Vac. Sci. Techno!. B 12(6),3755
(1994).
S. Tarucha, D.G. Austing, and T. Honda, Superlattices and Microstructures 18, 121
(1995).
S. Tarucha, D.G. Austing, T. Honda, R.I. van der Hage, and L.P. Kouwenhoven, Phys.
Rev. Lett. 77, 3613 (1996).
S. Tarucha, Y. Tokura, and Y. Hirayama, Phys. Rev. B 44, 13815 (1991).
Bo Su, V.I. Goldman, and lE. Cunningham, Science 255,313 (1992).
M. Tewordt, L. Martin-Moreno, J.T. Nicholls, M. Pepper, M.J. Kelly, V.J. Law, D.A.
Ritchie, lE.F. Frost, and G.A.C. Jones, Phys. Rev. B 45, 14407 (1992).
S. Tarucha, T. Honda, T. Saku, and Y. Tokura, Surf. Sci. 305, 547 (1994).
T. Schmidt, M. Tewordt, RH. Blick, R.I. Haug, D. Pfannkuche, K. von Klitzing, A
Foerster, and H. Lueth, Phys. Rev. B 51, 5570 (1995).
J.W. Sleight, E.S. Hornbeck, M.R Deshpande, R.G. Wheeler, M.A. Reed, RC. Bowen,
W.R Frensley, IN. Randall, and R.I. Matyi, Phys. Rev. B 53, 15727 (1996).
M.W. Dellow, P.H. Beton, M. Henini, P.C. Main, and L. Eaves, Electron. Lett. 27, 134
(1991).
P. Gueret, N. Blanc, R Germann, and H. Rothuizen, Phys. Rev. Lett. 68,1896 (1992).
C.I. Goodings, 1.RA. Cleaver, and H. Ahmed, Electron. Lett. 28,1535 (1992).
D.G. Austing, T. Honda, and S. Tarucha, Semicond. Sci. Techno!. 11,388 (1996).
V.R. Kolagunta, D.B. Janes, G.L. Chen, K.I. Webb, and M.R. Melloch, Superlattices and
Microstructures 17, 339 (1995).
A Groshev, T. Ivanov, and V. Valtchinov, Phys. Rev. Lett. 66, 1082 (1991).
H. Liu and G. Ayers, lApp!. Phys. 65, 4908 (1989).
S. Tarucha, Y. Hirayama, T. Saku, and Y. Tokura, in Science and Technology of
Mesoscopic Structures, edited by S. Namba, C. Hamaguchi, and T. Ando (SpringerVerlag, Tokyo 1992).
L.P. Kouwenhoven, R.I. van der Hage, S. Tarucha, D.G. Austing, and T .Honda,
(unpublished, 1996).
M. Macucci, K. Hess, and G.I. Iafrate, 1. App!. Phys. 77, 3267 (1995).
211
[148]
[149]
[150]
[151]
[152]
[153]
[154]
[155]
[156]
[157]
[158]
[159]
[160]
[161]
[162]
[163]
[164]
[165]
[166]
[167]
[168]
[169]
[170]
[171]
[172]
[173]
[174]
[175]
[176]
[177]
[178]
[179]
[I 80]
[181]
[182]
[183]
[184]
[185]
H. Tamura (private communications).

V. Fock, Z. Phys. 47, 446 (1928).
e.G. Darwin, Proc. Cambridge Philos, Soc. 27, 86 (1930).
L.1. Schiff, Quantum Mechanics (MacGraw-Hill, New York, 1949).
P.L. McEuen, E.B. Foxman, J. Kinaret, U. Meirav, M.A. Kastner, N.S. Wingreen, and
S.l Wind, Phys. Rev. B 45, 11419 (1992).
R.H. Silsbee and R.e. Ashoori, Phys. Rev. Lett. 64, 1991 (1990).
G.W. Bryant, Phys. Rev. Lett. 59, 1140 (1987).
e. de C. Chamon and X.G. Wen, Phys. Rev. B 49, 8227 (1994).
U. Palacios, L. Martin-Moreno, G. Chiappe, E. Louis, and e. Tejedor, Phys. Rev. Lett.
50,5760 (1994).
Y. Tanaka and H. Akera, Phys. Rev. B 53, 3901 (1996).
P.A. Maksym and T. Chakraborty, Phys. Rev. B 45, 180 (1992).
P.A. Maksym and T. Chakraborty, Phys. Rev. Lett. B 65, 108 (1990).
M. Eto, submitted to Jpn. l Appl. Phys. (1997).
Mesoscopic Phenomena in Solids, edited by B. L. Altshuler, P. A. Lee, and R. Webb
(North-Holland, Amsterdam, 1991).
Mesoscopic Quantum Physics, edited by E. Akkermans, G. Montambaux, l-L. Pichard,
and J. Zinn-Justin (Elsevier, Amsterdam, 1995).
S. Washburn and R. A. Webb, Rep. Prog. Phys. 55,1311 (1993).
B. L. Altshuler, JETP Lett. 41, 648 (1985).
P. A. Lee, A. D. Stone, and H. Fukuyama, Phys. Rev. B 35, 1039 (1987).
P. A. Lee and A. D. Stone, Phys. Rev. Lett. 55,1622 (1985).
B. L. Altshuler and B. I. Shklovskii, Sov. Phys. JETP 64, 127 (1986).
K. B. Efetov, Adv. in Phys. 32, 53 (1983).
Y. Imry, Euro. Phys. Lett. 1,249 (1986).
H. U. Baranger and P. A. Mello, Phys. Rev. Lett. 73,142 (1994).
e. W. l Beenakker, Phys. Rev. B 47,15763 (1993).
R. A. Jalabert, l-L. Pichard, and C. W. l Beenakker, Europhysics Letters 27, 255 (1994).
R. A. Jalabert, H. U. Baranger, and A. D. Stone, Phys. Rev. Lett. 65, 2442 (1990).
R. V. Jensen, Chaos 1,101 (1991).
H. U. Baranger, R. A. Jalabert, and A. D. Stone, Chaos 3, 665 (1993).
e. M. Marcus, A. J. Rimberg, R. M. Westervelt, P. F. Hopkins, and A. C. Gossard, Phys.
Rev. Lett. 69, 506 (1992).
M. W. Keller, O. Millo, A. Mittal, and D. E. Prober, Surf. Sci. 305, 501 (1994).
J. P. Bird, A. D. C. Grassie, M. Lakrimi, K. M. Hutchings, l J. Harris, and e. T. Foxon,
l Phys. Condo Matter 2, 7847 (1990).
I. H. Chan, R. M. Clarke, C. M. Marcus, K. Campman, and A. e. Gossard, Phys. Rev.
Lett. 74, 3876 (1995).
R. Ketzmerick, Phys. Rev. B 54, 10841 (1996); H. Hegger, B. Huckestein, K. Hecker, M.
Janssen, A. Freimuth, G. Reckziegel, and R. Tuzinski, Phys. Rev. Lett. 77, 3885 (1996).
R. Akis and D. Ferry, unpublished (1996).
C.W.J. Beenakker, cond-matl9612179 (1996); to appear in Rev. Mod. Phys.
K. Efetov, Supersymmetry in Disorder and Chaos (Cambridge University Press,
Cambridge, 1997).
M. L. Mehta, Random Matrices (Academic Press, Boston, 1991).
A. D. Stone, P. A. Mello, K. A. Muttalib, and J. Pichard, in Mesoscopic Phenomena in
Solids, edited by B. L. Altshuler, P. A. Lee, and R. A. Webb (Elsevier, Amsterdam, 1991).
212
B. L. Altshuler and B. D. Simons, in Mesoscopic Quantum Physics, edited by E.
Akkermans, G. Montambaux, J.-L. Pichard, and J. Zinn-Justin (Elsevier, Amsterdam,
1995).
[187] A. V. Andreev, O. Agam, B. D. Simons, and B. L. Altshuler, Phys. Rev. Lett. 76,3947
(1996).
[188] R. A. Jalabert, A. D. Stone, and Y. Alhassid, Phys. Rev. Lett. 68, 3468 (1992).
[189] V. N. Prigodin, K. B. Efetov, and S. Hda, Phys. Rev. Lett. 71,1230 (1993).
[190] E. R. Mucciolo, V. N. Prigodin, and B. L. Altshuler, Phys. Rev. B 51, 1714 (1995).
[191] Y. Alhassid and C. H. Lewenkopf, Phys. Rev. Lett. 75, 3922 (1995).
[192] H. Bruus and A. D. Stone, Phys. Rev. B 50, 18275 (1994).
[193] V.l. Fal'ko and K. B. Efetov, Phys. Rev. B 50,11267 (1994).
[194] Y. Alhassid, J. N. Hormuzdiar, and N. D. Whelen, cond-matl9609115 (1996).
[195] P. L. McEuen, N. S. Wingreen, E. B. Foxman, J. Kinaret, U. Meirav, M. A. Kastner, Y.
Meir, and S. J. Wind, Physica B 189,70 (1993).
[196] A. M. Chang, H. U. Baranger, L. N. Pfeiffer, K. W. West, and T. Y. Chang, Phys. Rev.
Lett. 76, 1695 (1996).
[197] J. A. Folk, S. R. Patel, S. Godijn, A. G. Huibers, S. Cronenwett, C. M. Marcus, K.
Campman, and A. C. Gossard, Phys. Rev. Lett. 76, 1699 (1996).
[198] Y. M. Blanter and A. D. Mirlin, cond-matl9604139 (1996).
[199] S. R. Patel, S. M. Cronenwett, A. G. Huibers, M. Switkes, J. A. Folk, C. M. Marcus, K.
Campman, and A. C. Gossard, Superlattices and Microstructures, to be published (1997).
[200] B. D. Simons and B. L. Altshuler, Phys. Rev. Lett. 70,4063 (1993).
[201] A. Szafer and B. A. Altshuler, Phys. Rev. Lett. 70, 587 (1993).
[202] Y. Alhassid and H. Attias, Phys. Rev. Lett. 76,1711 (1996).
[203] H. Bruus, C. H. Lewenkopf, and E. R. Mucciolo, Phys. Rev. B 53, 9968 (1996).
[204] O. Agam, J. Phys. I (Paris) 4, 697 (1994).
[205] U. Sivan, F. P. Milliken, K. Milkove, S. Rishton, Y. Lee, J. M. Hong, V. Boegli, D. Kern,
and M. DeFranza, Europhys. Lett. 25, 605 (1994).
[206] Y. A1hassid and H. Attias, Phys. Rev. B 54, 2696 (1996).
[207] K. B. Efetov, Phys. Rev. Lett. 74, 2299 (1995).
[208] P. L. McEuen, E. B. Foxman, U. Meirav, M. A. Kastner, Y. Meir, N. S. Wingreen, and S.
J. Wind, Phys. Rev. Lett. 66,1926 (1991).
[209] U. Sivan, R. Berkovits, Y. Aloni, O. Prus, A. Auerbach, and G. Ben-Yoseph, Phys. Rev.
Lett. 77, 1123 (1996).
[210] O. Bohigas, in Chaos and Quantum Physics, edited by M. Giannoni, A. Voros, and J.
Zinn-Justin (Elsevier, Amsterdam, 1991).
[211] Y. M. Blanter, A. D. Mirlin, and B. A. Muzykantskii, Phys. Rev. Lett. 78, 2449 (1997).
[212] G. Montambaux, in Les Houches, Session LX/ll, Quantum Fluctuations, edited by E.
Giacobino, S. Reynaud, and J. Zinn-Justin (Elsevier, Amsterdam, 1996).
[213] L. P. Levy, G. Dolan, J. Dunsmuir, and H. Bouchiat, Phys. Rev. Lett. 64, 2074 (1990).
[214] V. Chandrasekhar, R. A. Webb, M. J. Brady, M. B. Ketchen, W. J. Gallagher, and A.
Kleinsasser, Phys. Rev. Lett. 67, 3578 (1991).
[215] D. Mailly, C. Chapelier, and A. Benoit, Phys. Rev. Lett. 70, 2020 (1993).
[216] K. Richter, D. Ullmo, and R. A. Jalabert, Physics Reports, (in press) (1996).
[217] L. P. Levy, D. H. Reich, L. Pfeiffer, and K. West, Physica B 189,204 (1993).
[218] K. Nakamura and H. Ishio, J. Phys. Soc. Japan 61,3939 (1992).
[219] F. von Oppen, Phys. Rev. B 50,17151 (1994).
[220] D. Ullmo, K. Richter, and R. A. Jalabert, Phys. Rev. Lett. 74, 383 (1995).
[186]
213
[221]
[222]
[223]
[224]
[225]
[226]
[227]
[228]
[229]
[230]
[231]
[232]
[233]
[234]
[235]
[236]
[237]
[238]
[239]
[240]
[241]
[242]
[243]
[244]
[245]
[246]
[247]
[248]
[249]
[250]
[251]
D. V. Averin and Y. Nazarov, Phys. Rev. Lett. 65, 2446 (1990).

D. C. Glattli, C. Pasquier, U. Meirav, F. 1. B. Williams, Y. Jin, and B. Etienne, Z. Phys. B
85,375 (1991).
1. L. Aleiner and L.1. Glazman, Phys. Rev. Lett. 77, 2057 (1996).
S. M. Cronenwett, S. R. Patel, A. G. Huibers, C. M. Marcus, K Campman, and A. C.
Gossard, unpublished (1996).
E. H. Hauge and J. A. Stovneng, Rev. Mod. Phys. 61, 917 (1989).
R. M. Clarke, I. H. Chan, C. M. Marcus, C. I. Duruoz, J. S. Harris, K. Campman, and A.
C. Gossard, Phys. Rev. B 52, 2656 (1994).
P. Hawrylak, Phys. Rev. Lett. 71, 3347 (1993).
I. K. Marrnorkos and C. W. J. Beenakker, Phys. Rev. B 46, 15562 (1992).
J.M. Kinaret and N.S. Wingreen, Phys. Rev. B 48, 11113 (1993).
M. Stopa, unpublished. Note that in this simulation, the two LLs are two orbital LLs, n =
o and n = 1. The spin splitting is too small to be clearly seen.
A.A.M. Staring, B.W. Alphenaar, H. van Houten, L.W. Molenkamp, OJ.A. Buyk,
M.A.A. Mabesoone, and C.T. Foxon, Phys. Rev. B 46, 12869 (1992).
T. Heinzel, D.A. Wharam, J.P. Kotthaus, G. Bohm, W. Klein, G. Trankle, and G.
Weimann, Phys. Rev. B 50, 15113 (1994).
N.C. van der Vaart, M.P. de Ruyter van Steveninck, L.P. Kouwenhoven, A.T. Johnson,
Y.V. Nazarov, C.PJ.M. Harmans, and C.T. Foxon, Phys. Rev. Lett. 73, 320 (1994).
s. R. Eric Yang, A.H. MacDonald, and M.D. Johnson, Phys. Rev. Lett. 71, 3194 (1993);
A.H. MacDonald, S.R. Eric Yang, and M.D. Johnson, Aust. J. Phys. 46, 345 (1993).
O. Klein, C. de C. Chamon, D. Tang, D.M. Abusch-Magder, U. Meirav, X.-G. Wen,
M.A. Kastner, and SJ. Wind, Phys. Rev. Lett. 74, 785 (1995).
J. H. Oaknin, L. Martin-Moreno, J.J. Palacios, and C. Tejedor, Phys. Rev. Lett. 74, 5120
(1995).
J.K. Jain and T. Kawamura, Europhys. Lett. 29, 321 (1995).
S.E. Barrett, G. Dabbagh, L.N. Pfeiffer, KW. West, R. Tycko, Phys. Rev. Lett. 74, 5112
(1995).
A. Karlhede, S.A. Kivelson, K. Lejnell, and S.L. Sondi, Phys. Rev. Lett. 77, 2061 (1996).
I.H. Oaknin, L. Martin-Moreno, and C. Tejedor, Phys. Rev. B 54, 16850 (1996).
O. Klein, D. Goldhaber-Gordon, C. de Chamon, and M.A. Kastner, Phys. Rev. B 53,
4221 (1996).
Y.V. Nazarov and A.V. Khaetskii, Phys. Rev. B 49, 5077 (1994).
L.P. Kouwenhoven, A.T. Johnson, N.C. van der Vaart, A. van der Enden, C.J.P.M.
Harrnans, and C.T. Foxon, Z. Phys. B 85, 381 (1991).
M. W. Keller, l. M. Martinis, N. M. Zimmerman, A. H. Steinbach, Appl. Phys.
Lett. 69,1804 (1996).
P.K Tien, and J.R. Gordon, Phys. Rev. 129,647 (1963).
N.S. Wingreen, A.P. lauho, and Y. Meir, Phys. Rev. B 48, 8487 (1993).
A. P. lauho, N. S. Wingreen, and Y. Meir, Phys. Rev. B 50, 5528 (1994).
Y. Fu and S. C. Dudley, Phys. Rev. Lett. 70, 65 (1993).
L.P. Kouwenhoven, S. lauhar, K. McCormick, D. Dixon, P.L. McEuen, Yu. V. Nazarov,
N.C. vanderVaart, andC.T. Foxon, Phys. Rev. B 50,2019 (1994).
L.P. Kouwenhoven, S. Jauhar, J. Orenstein, P.L. McEuen, Y. Nagamune, J. Motohisa,
and H. Sakaki, Phys. Rev. Lett. 73, 3443 (1994).
H. Akiyama, H. Sugawara, Y. Kadoya, A. Lorke, S. Tsujino, and H. Sakaki, Appl. Phys.
Lett. 65, 424 (1994).
214
[252]
[253]
[254]
[255]
[256]
[257]
[258]
[259]
[260]
[261]
[262]
[263]
[264]
[265]
[266]
[267]
[268]
[269]
[270]
[271]
[272]
[273]
[274]
[275]
[276]
[277]
[278]
[279]
[280]
[281]
[282]
[283]
[284]
[285]
H. Drexler, A. 1. S. Scott, S. J. Allen, Jr., K. L. Campman, and A.c. Gossard, Appl. Phys.
Lett. 67, 2816 (1995).
B. J. Keay, S. 1. Allen, Jr., 1. Galan, J. P. Kaminski, K. L. Campman, A. C. Gossard, U.
Bhattacharya, and M. J. W. Rodwell, Phys. Rev. Lett. 75, 4098 (1995).
B. J. Keay, S. Zeuner, S. 1. Allen, Jr., K. D. Maranowski, A. C. Gossard, U. Bhattacharya,
and M. 1. W. Rodwell, Phys. Rev. Lett. 75, 4102 (1995).
1.M. Hergenrother, M.T. Tuominen, 1.G. Lu, D.C. Ralph, and M. Tinkham, Physica B
203,327 (1994).
D. Vion, P.F. Orfila, P. Joyez, D. Esteve, and M.H. Devoret, 1. of Appl. Phys. 77, 2519
(1995).
M.H. Devoret, D. Esteve, H. Grabert, G.-L. Ingold, H. Pothier, and C. Urbina, Phys. Rev.
Lett. 64, 1824 (1990).
T. Holst, D. Esteve, C. Urbina, M.H. Devoret, Phys. Rev. Lett. 73, 3455 (1994).
Y.B. ZeJdovich, Sov. Phys. JETP 24, 1006 (1967).
c.A. Stafford and N.S. Wingreen, Phys. Rev. Lett. 76, 1916 (1996).
T. H. Stoof and Yu. V. Nazarov, Phys. Rev. B 53, 1050 (1996).
M. Holthaus and D. Hone, Phys. Rev. B 47, 6499 (1993).
M. Yu. Sumetskii and M. L. Fel'shtyn, JETP Lett. 53, 24 (1991).
RH. Blick, R.I. Haug, K. v. Klitzing, and K. Eberl, Surf. Science 361,595 (1996).
T. Fujisawa and S. Tarucha, Superlattices and Microstructures, 21, 247 (1997).
D.V. Averin and K.K. Likharev, in Ref. 3.
E.H. Visscher, S.M. Verbrugh, 1. Lindeman, P. Hadley, and J.E. Mooij, Appl. Phys. Lett.
66,305 (1994).
P. Lafarge, H. Pothier, E.R Williams, D. Esteve, C. Urbina, and M.H. Devoret, Z. Phys.
B 85, 327 (1991).
J.M. Martinis, M. Nahum, and H.D. Jensen, Phys. Rev. Lett. 72, 904 (1994).
J.P. Pekola, K.P. Hirvi, J.P. Kauppinen, and M.A. Paalanen, Phys. Rev. Lett. 73, 2903
(1994).
A.M.M. Staring, L.W. Molenkamp, B.W. Alphenaar, H. van Houten, O.J.A. Buyk,
M.A.A. Mabesoone, C.W.1. Beenakker, C.T. Foxon, Europhys. Lett. 22, 57 (1993).
A.N. Cleland, D. Esteve, C. Urbina, M.H. Devoret, Appl. Phys. Lett. 61, 2820 (1992).
C.S. Lent, P.D. Tougaw, and W. Porod, Appl. Phys. Lett. 62, 714 (1993).
E. Buks and M. Heiblum, unpublished (1997).
I.L. Aleiner, N.S. Wingreen and Y. Meir, cond-matl9702001 (1997).
L.P. Kouwenhoven, Science 275,1896 (1997);
O. Agam, N.S. Wingreen, B.L. Altshuler, D.C. Ralph, and M. Tinkham, Phys. Rev. Lett.
78, 1956 (1997).
1. von Delft, A.D. Zaikin, D.S. Golubev, and W. Tichy, Phys. Rev. Lett. 77, 3189 (1996).
D.L. Klein and P.L. McEuen, unpublished (1997).
D. Porath and O. Millo, J. Appl. Phys. 81, 2241 (1996).
C. Joachim, and J.M. Gimzewski, Chern. Phys. Lett. 265, 353 (1997).
E.S. Soldatov, V.V. Khanin, A.S. Trifonov, S.P. Gubin, V.V. Kolesov, D.E. Presnov, S.A
Iakovenko, G.B. Khomutov, and A.N. Korotkov, cond-matter 9610155 (1996).
M. Bockrath, D.H. Cobden, P.L. McEuen, N.G. Chopra, A. Zettl, A. Thess, and R.E.
Smalley, Science 275, 1922 (1997); and unpublished.
S.I. Tans, M.H. Devoret, H. Dai, A. Thess, RE. Smalley, L.I. Geerligs, and C. Dekker,
Nature 386, 474 (1997).
MAGNETOTUNNELING SPECTROSCOPY: STUDYING SELF-ORGANIZED

QUANTUM DOTS AND QUANTUM CHAOLOGY
L. EAVES
Department of Physics,
University of Nottingham,
Nottingham NG7 2RD, UK
1.
Introduction
This paper reports investigations of two topical areas of semiconductor physics research:
the electronic properties of zero-dimensional quantum dots, and the effect of chaotic
motion on the quantum transport properties of semiconductor nanostructures. Section
2 examines the electronic properties of self-organized InAs quantum dots, grown by
Stranskii-Krastanov epitaxy. The dots are incorporated in the AlAs barrier of a singlebarrier GaAs/ AIAs/GaAs heterostructure. Section 3 of the paper examines the quantum
states of electrons moving in a quantum well in the presence of a strong, tilted magnetic
field. Under these conditions, the electron motion in the well is classically chaotic. In
both cases, magnetotunneling spectroscopy is employed as the experimental technique
to study the electronic states.
2.
Magnetotunneling of Self-Organized Quantum Dots
The self-organized Stranskii-Krastanov mode of epitaxial growth provides a relatively

simple and highly efficient way of producing quantum dot (QD) structures in which the
carriers are strongly confined on a length scale of = 10 nm in all three dimensions [IS] . In the case of the InAs on GaAs system, the lattice mismatch is = 7 %, and the
InAs QDs have a typical base length of 10-20 nm and height =4-10 nm. Their
electronic states have been investigated by optical [2,6-8] and capacitance [6,9,10]
spectroscopy. The electron and hole energies in the dots are strongly influenced by
both the large strain in the region of the dot and by size quantization. The dot ensemble
has a fairly wide distribution of eigenenergies due to variations in size, shape and strain.
Typically, an optical emission line from the dot ensemble has energy
1. 1-1.3 eV
and a full width half maximum (FWHM) linewidth of Ll = 40-60 meV [2,5,7,8].
Recently, Itskevich et at have developed a new approach for studying selfassembled QDs, embedding them in the barrier layer of a n-i-n single-barrier tunneling
nw =
215
216
heterostructure device [11]. The device for magnetotunneling studies was grown on a
(100) n+ -GaAs substrate. The 1.8 monolayers of lnAs which forms the QDs were
embedded in the centre of a 10 nm AlAs barrier which is surrounded by 100 nm buffer
layers of undoped GaAs. These were sandwiched between two 1 JLm layers of GaAs
with graded Si doping. Circular mesas, from 30 JLm to 400 JLm diameter, were
produced by optical lithography , and AuGe was layered into the n + GaAs to form the
ohmic top contact. Full details of the device are given in reference [11]. When a
voltage V is applied between the contacts, a two-dimensional electron gas (2DEG)
accumulates in the undoped GaAs layer near the tunnel barrier. Figure 1 shows the
conduction band profile of the device. Resonant tunneling occurs when an electronic
state in the barrier is resonant with a state in the 2DEG.
20
..... 15
GaAs
AlA
n-GaAs emitter
2DEG
eV
I
I
1.8 monolayers
of InAs
B=OT
4.2 K
Q)
'-
:; 10
0.3 K
GaAs
n-GaAs
Figure 1. A schematic conduction band diagram

under an applied voltage V of the device used for
magnetotunneling studies of quantum dots.
0.04 K
o L...:......:........:~:::.:::::::::r....-'---.J
o
0.1
0.2
0.3
Voltage (V)
Figure 2. I(V) characteristics of the quantum dot

tunneling device at different temperatures.
At low bias and low carrier density in the 2DEG the non-resonant tunneling
current through the barrier is much smaller than the resonant single electron tunneling
current through one QD state. Using the effective mass model, the tunnel current
through a single dot is estimated to be a few pA [11], in agreement with the typical
resonant peak height in Figure 2, which shows I(V) traces recorded at T = 4.2 K,
300 mK and 30 mK. An isolated peaked feature can be seen between 100 mV and 150
mVat all temperatures due to electron tunneling through an individual QD [11]. At low
T the resonance peak develops a kT -broadened sharp onset at the low voltage edge Vp,
i.e. when the localized state coincides with the Fermi energy in the 2DEG [12]. The
high voltage cut-off of the peak, Yo, indicates alignment of the localized state with the
subband bottom of the 2DEG.
Figure 3 shows examples of recorded I(V) traces at different constant magnetic
fields, B = 1.2 T, 0.6 T and 0.4 T applied parallel to the current, showing one, two
and three components indicated by arrows, respectively. The splitting is due to Landau
quantization (n = 0,1,2 ... ) in the emitter 2DEG. In order to obtain information over
a large range of magnetic fields B the I(B) curves are recorded at constant V values
217
between 100 mV and 150 mY. These results are summarised in the fan chart plot in
Figure 4. The onset of the current peak corresponds to the chemical potential in the
emitter aligning with the single dot level. Therefore, as B is swept, the variation of the
current onset VF is effectively the variation in EF and shows liB periodic oscillations.
The Fermi energy of the 2DEG at B = 0 T is found to be EF = 2 meV and the
corresponding electron density ns = 5.6 X 1010 cm-2 at B = 0 T. In addition, above
VF (i.e. below EF) current peaks in the I(B) curves occur at magnetic field values where
the peak of the DOS of a certain LL aligns with the QD energy level for a given voltage
V. The symbols represent these points in the B-V plane. The straight lines labelled n
= 0,1... ,5 represent a Landau fan obtained from Vn = V0 - (n + 1I2)fheB/m*, where
f = 6.5 is the electrostatic leverage and m* is the GaAs conduction band effective mass.
130
B
4
=O.6T
=1.2T n = 0
125
:;-
n=1
.s
Q)
120
Ol
.l!!
>
n =O
.-r'
0.1
0.12
0.14 0.1
0.12
Voltage (V)
0.14
115
0.2
0.4
0.6
0.8
Magnetic Field (T)
Figure 3. I(V) characteristics of the current peak
Figure 4. Fan diagram of peaks in I(B) taken at
at 0.12 V at different magnetic fields B showing

the Landau level splitting.
constant V around the resonance.
Applying B .1 I causes the diamagnetic shift of peaks to lower voltage. Figure 5

shows the lowest voltage peak splitting into two components with increasing B. This
can be interpreted as a breaking of the spin-degeneracy of the InAs QD. Using the
leverage factor f = 6.5, as estimated from the LL splitting at B II I, the QD electron gfactor is estimated to be Ig * I = 1.2 0.3, much smaller than the value for bulk InAs.
The different voltage widths of the spin-split components at B = 10 T may be evidence
for partial spin polarization of the 2DEG, with the QD acting as a spin filter. The
difference in voltage width::::: gllBB gives for 2DEG electrons g ::::: -0.5, with the sign
opposite to that for QDs.
In summary, this section of the paper has shown that electron tunneling through
a single quantum dot state is a sensitive probe of the Landau-quantized density of states
in an adjacent emitter accumulation layer. In principle, it appears that tunnelling
218
T=350 mK
B ..ll
,........
10
a.
..........
.......
spectroscopy using a single

discrete quantum dot level can
provide new information about
the states of the 2DEG below
the 2DEG Fermi energy. In
addition, information about the
spin-splitting of the electronic
states can also be obtained .
c:
0,)
lI-
::J
OW-__- L_ _ 8=0
__
~
0.10
~~
0.12
Voltage (V)
_ _ ~_ _ ~
0.14
Figure 5. I(V) characteristics at various B.11. Curves are offset.
3.
Probing Scarred Wavefunctions of Chaotic Electron Motion in the Quantum

Well
Berry [13] has defined quantum chaology as the "study of semiclassical but non-classical
behaviour of systems whose classical motion exhibits chaos". Of crucial importance in
these studies are the periodic, but unstable, classical orbits which occur within the
chaotic phase space of the system [13,14]. These periodic unstable orbits give rise to
periodically spaced clusters of levels in the energy spectrum of chaotic systems, with
a cluster energy spacing given by the well-known trace formula [14]. The periodic
unstable orbits also influence the wavefunctions of a subset of eigenstates of the system.
These wavefunctions are said to be "scarred" by the orbits, since regions of enhanced
probability density occur along the paths of the orbits [15].
Recent theoretical studies by Fromhold et al [16] of the electronic levels in a
quantum well of a resonant tunneling diode (RTD) in the chaotic regime have revealed
subsets of eigenstates strongly scarred by unstable periodic orbits [17]. The calculations
also indicated that the current through the RTD was dominated by tunneling via the
scarred states of the quantum well. The first experimental work by Fromhold et ai,
using an RTD incorporating a 120 nm quantum well, could not resolve discrete levels
due to broadening by scattering processes involving the emission of longitudinal optic
(LO) phonon. These wide well RTDs were therefore not ideally suited to probing the
wavefunction scarring of individual states.
In order to probe individual states, Wilkinson et al [18] carried out a new set of
experiments and theoretical analysis on an RTD with a narrower quantum well width
219
of 22 nm. In order to maintain chaotic electron motion in this narrower well, it was
necessary to ensure that the tilted magnetic fields were in excess of 35 T. Such fields
give Poincare sections with the "dusty" appearance, characteristic of classical chaos.
The experiments were therefore carried out using pulsed magnetic fields at the Institute
of Solid State Physics, University of Tokyo. A schematic band diagram of the device
under an applied bias V is shown in Figure 6. Electrons resonantly tunnel into the
quantum states of the well through the left-hand barrier from a degenerate emitter
2DEG which forms close to the barrier under bias. Full details of the experimental
arrangements are given in reference [18].
A typical set of current-voltage
characteristics at various fixed magnetic field values which was obtained using this
technique is shown in Figure 7.
B=3~
/\
barriers
emitter
well
collector
B=~N
B=~
B=~
B=~
B=~M
0.8
1.0
Figure 6. Schematic variation of the potential
Figure 7. Plots of conductance dI/dV versus V
energy of an electron at the conduction band edge

in the RTD with position x normal to the layer
interfaces.
Inset: orientation of the tilted
magnetic field B relative to the co-ordinate axes.
for the 22 nm wide QW RTD at a temperature of

4 K in various magnetic field strengths, with (J =
40.
The most striking feature of the data is that the oscillations observed in dI/dV
remain approximately periodic over the field range of 15-37 T, with the period
increasing gradually with increasing B. Note that the small high frequency noise at the
tops of the conductance peaks is an experimental artifact, due mainly to vibration of the
sample in the pulsed fields.
220
Cl
200
(c)
250
300
e (meV)
En=280.90 meV, v=9
350
En=309.25 meV, v=10
~:
z
En calculated for the QW with V = 668 mV, B = 37
T, (j = 40. Details of the calculations are given in ref. [18]. Solid curve: density of levels D(E) obtained
by broadening each energy level to a width r = 6 meV consistent with the finite lifetime of "" 0.1 ps
imposed by the emission of longitudinal optic phonons. Arrows show successive minima in D(e) associated
with Gutzwiller fluctuations produced by periodic unstable orbits in the quantum well. These orbits also
produce strong scars in the wavefunctions corresponding to the subset of energy levels shown as vertical
dashed lines. Orbits and scar patterns are shown in (c). (b) Open circles: squared matrix elements ~2 for
tunneling transitions into quantized energy levels shown in (a). Closed circles indicate matrix elements for
transitions into a subset of individual scarred states, two of which are shown in (c).
(c) Probability density plots (white background = 0) of scarred QW eigenfunctions corresponding to energy
levels and quantum numbers v shown. Similar scar patterns are found in the wavefuncticins corresponding
to all the energy levels shown as dashed lines in (a). The thick black vertical lines indicate positions of
barriers. Inset: orientation of magnetic field B in x-z plane. In our chosen gauge the probability density
depends only on the x- and z-coordinates. The trajectories of three distinct unstable periodic orbits which
contribute to scar patterns are shown projected on the x-z plane as solid continuous curves.
Figure 8. (a) Vertical lines: quantized energy levels
221
In order to interpret these data, Wilkinson et al [18] consider the energy levels
and wavefunctions of the system. Fig. 8(a) shows a quantised energy level spectrum
En of the QW in the high magnetic field regime corresponding to predominantly strong
classical chaos. Note that the distribution of energy levels is highly irregular. This
type of spectral complexity is characteristic of classically chaotic systems [13,14]. The
mean energy level spacing is sufficiently large that close to the energy of the injected
electrons ( -280 meV), most individual energy levels produce well-resolved peaks in
the density of levels plot D(E). The arrows in Figure 8(a) correspond to regularlyspaced minima in D(e) corresponding to the period of the Gutzwiller oscillations.
Fig. 8(b) shows Mn2 for tunnelling transitions into the energy levels En' The
matrix elements for transitions into a subset of almost equally-spaced energy levels
(indicated by broken vertical lines in Fig. 8(a are much larger than for the remaining
states. The reason for this can be seen by examining the probability distributions of the
corresponding wavefunctions in the QW, two of which are shown in Fig. 8(c). These
wavefunctions reveal remarkably clear scars of unstable closed orbits, each of which has
a single bounce per period on the left-hand barrier, with three bounces per period on
the right-hand barrier. The energies of the scarred states can be accurately located
using a simple quantisation rule S(E) ::: (p + <Jh, where S(E) is the classical action of
the scarring orbit, <J> :::: 1, and the quantum number P gives the number of antinodes in
the scar pattern along the classical path [16]. Each scarred state has a single antinode
close to the LH barrier which is almost aligned in the z-direction with the occupied
(ground state) Landau state in the emitter accumulation layer. Consequently the emitter
states overlap strongly with the scarred states in the QW. This quasi-selection rule is
analogous to the conservation of Landau index [19] in tunnelling transitions at () ::: 0,
and corresponds to classical momentum conservation. In a classical picture, electrons
enter the QW with a small lateral velocity component vz ' due to the low Fermi energy
of the emitter 2DEG. Consequently, only periodic orbits in the QW which impinge on
the LH barrier with small Vz are accessible in momentum conserving transitions from
the emitter accumulation layer. These orbits produce scarred wavefunctions with a
single antinode close to the LH barrier (as shown in Fig. 8(c, which overlap strongly
with the occupied Landau state in the emitter accumulation layer. By contrast, periodic
classical orbits which impinge on the LH barrier with large Vz are inaccessible to the
tunnelling electrons. These orbits produce scarred states with several antinodes close
to the LH barrier, which couple weakly to the occupied emitter state.
The way in which the dominant transitions into the scarred states shown in Fig.
8(b) influence the tunnelling characteristics of the device can be seen in Figure 9(a) ,
which shows a detail of the experimental dI/dV versus V plot. The corresponding
calculated I(V) is shown in the upper curve of Fig. 9(b) for B ::: 37 T, () ::: 400
together with the current contributions due to transitions into individual states in the QW
(middle and lower curves in Fig. 9(b [20]. The dominant current contributions
(middle curves, Fig. 9(b originate from transitions into scarred states with P ::: 7, 8
and 9, and generate strong periodic resonant peaks in I(V) (arrowed). The probability
densities of these three states are also shown in Figure 9(b). The contributions to the
tunnel current of the other states in the QW (lower curves, Fig. 9(b are relatively
222
small. The measured voltage

of these peaks, !!.V =:::
spacing
(a)
87 mV, is in good agreement
--.
with the value of =:::94 mV
I
calculated for the dominant
current contributions in Fig.
9(b). This provides clear
evidence that the observed
"0
::::
resonant peaks originate from
"0
transitions into the subset of
scarred states. We stress that
the large voltage spacing and
0
periodic distribution of the
--.
observed resonant peaks exclude
en
the possibility that they
:t:::
c:
originate from transitions into
=s
adjacent energy levels in the
.c
QW.
Quasi-selection rules
Lbased on periodic scarring of
eo
individual wavefunctions are
essential to account for the
experimental data.
In conclusion, this section
has
described
how resonant
200 300 400 500 600 700 tunnelling spectroscopy
can be
used
to
investigate
wavefunction
V (mV)
scarring in a QW with strongly
chaotic electron dynamics.
Figure 9. (a) Experimental dI/dV versus V plot measured for a
Detailed calculations of the
100 ILm diameter mesa RTD at B = 37 T and (J = 4<1'. (b)
tunnelling rates provide clear
Calculated I(V) characteristic (top curve) and current contributions
evidence that for voltage ranges
(middle and bottom curves) due to transitions into individual
where electrons tunnel into
eigenstates of the QW when (J = 40, B = 37 T. Strong periodic
strongly-scarred wavefunctions,
resonant peaks in I(V) (arrowed) originate from the dominant
current contributions (middle curves) produced by tunneling
the resonant peaks observed in
transitions into individual scarred states with probability
the I(V) characteristics
distributions in the x-z plane shown inset. These scarred states
originate
from dominant
are calculated at the peak of the current contributions and belong
transitions
into individual
to the same sequence shown in Figure S(c). The remaining
current contributions (lower curves) generate only weak features
scarred states which occur
in I(V).
periodically with the same
energy spacing as the
Gutzwiller oscillations. These studies highlight the potential of RTDs containing narrow
quantum wells in very high magnetic fields for experimental study of wavefunction
scarring. This quantum well system is, to the author's knowledge, the only quantum
system in which an oscillatory effect originating from individual scarred eigenstates is
0.2
-->a
--
223
observed in experiment.
These experiments therefore complement recent
magnetoresistance measurements performed under ohmic conditions of chaotic electron
motion in antidot superlattices [21] and of Bunimovich stadia [22].
Acknowledgements. This work is funded by the UK Engineering and Physical
Sciences Research Council (EPSRC). The author is funded by an EPSRC Senior
Fellowship. The work on self-organised InAs quantum dots was carried out in
collaboration with Dr. T. Ihn, Dr. I. E. Itskevich, Prof. P. C. Main and Dr. M. Henini
at the University of Nottingham. The work on quantum chaology was done in
collaboration with Mr. P. B. Wilkinson, Dr. T. M. Fromhold, Prof. F. W. Sheard
(University of Nottingham) and Prof. N. Miura and Dr. T. Takamasu (Institute for
Solid State Physics, University of Tokyo).
224
4.
References
l.
Moison, J. M., Houzay, F., Barthe, F., Leprince, L., Andre, E. and Vatel, O. (1994) Appl. Phys.
Lett. 64, 196.
Marzin, J.-Y., Gerard, J.-M., Izrael, A., Barrier, D. and Bastard, G. (1994) Phys. Rev. Lett. 73,716.
Petroff, P. M. and DenBaars, S. P. (1994) Superlatt. & Microstruct. 15, 15.
Gossard, A. C. and Fafard, S. (1994) Solid St. Commun. 92, 63.
Bimberg, D., Ledentsov, N. N., Grundmann, M., Kirstaedter, N., Schmidt, O. G., Mao, M. H.,
Ustinov, V. M., Egorov, A. Yu., Zhukov, A. E., Kop'ev, P. S., Alferov, Zh. I., Ruvimov, S. S.,
Giisele, U. and Heydenreich, J. (1996) Phys. Stat. Sol. b 194, 159.
Drexler, H., Leonard, D., Hansen, W., Kotthaus, J. P. and Petroff, P. M. (1994) Phys. Rev. Lett.
73,2252.
Heitz, R., Grundmann, M., Ledentsov, N. N., Eckey, L., Veit, M., Bimberg, D., Ustinov, V. M.,
Egorov, A. Yu., Zhukov, A. E., Kop'ev, P. S. and Alferov, Zh.1. (1996) Appl. Phys. Lett. 68, 36l.
Grundmann, M., Ledentsov, N. N., Stier, 0., Bimberg, D., Ustinov, V. M., Kop'ev, P. S. and
Alferov, Zh. I. (1996)Appl. Phys. Lett. 68,979.
Medeiros-Ribeiro, G., Leonard, D. and Petroff, P. M. (1995) Appl. Phys. Lett. 661767.
Brunkov, P. N., Konnikov, S. G., Ustinov, V. M., Zhukov, A. E., Egorov, A. Yu., Maksimov,
M. V., Ledentsov, N. N. and Kop'ev, P. S. (1996) Semiconductors 30,492.
Itskevich, I. E., Ihn, T., Thornton, A., Henini, M., Foster, T. J., Moriarty, P., Nogaret, A., Beton,
P. H., Eaves, L. and Main, P. C. (1996) Phys. Rev. B 54, to be published December 1996.
Thornton, A., Ihn, T., Itskevich, I. E., Main, P. C., Eaves, L. and Henini, M. (1996) 12th Int. Conj.
on the Application o/High Magnetic Fields, Wiirzburg, Germany, July/August 1996, to be published.
Berry, M. V. (1987) Proc. R. Soc. Lond. 413, 183-198.
Gutzwiller, M. C. (1990) Chaos in Classical and Quantum Mechanics, Springer, New York.
Heller, E. J. (1984) Phys. Rev. Lett. 53, 1515-1518.
Fromhold, T. M., Wilkinson, P. B., Sheard, F. W., Eaves, L., Miao, J. and G. Edwards (1995)
Phys. Rev. Lett. 75, 1142-1144.
Fromhold, T. M., Eaves, L., Sheard, F. W., Leadbeater,M. L., Foster, T. J. and Main, P. C. (1994)
Phys. Rev. Lett. 72,2608-2611.
Wilkinson, P. B., Fromhold, T. M., Eaves, L., Sheard, F. W., Miura, N. and Takamasu, T. (1996)
Nature 380, 608-610.
Leadbeater, M. L., Sheard, F. W. and Eaves, L. (1991) Semicon. Sci. Tech. 6 1021-1024.
Nonparabolicity of the GaAs conduction band is included in the quantum calculations by taking the
effective mass of electrons in the QW to be m* = 0.067 m/l + aK(V)), where K(V) is the kinetic
energy of electrons at the centre of the QW under bias voltage V, and a = 2 eV-I. This value of a
was obtained by fitting the positions of resonant peaks in the I(V) characteristics of
GaAs/(AIo.4Gao.6)As RTDs containing 22 nm, 60 nm, and 120 nm wide QWs with (J = 0 and (J =
90, and gives m* values consistent with multiband k.p calculations (see, for example, Shi, J.M.,
Peeters, F.M. & Devreese, J.T. (1993) Phys. Rev. B 48, 5202-5216).
Weiss, D., Richter, K., Menschig, A., Bergmann, R., Schweizer, H., von Klitzing, K. and Weimann,
G. (1993) Phys. Rev. Lett. 70,4118.
Marcus, C. M., Rimberg, A. 1., Westervelt, R. M., Hopkins, P. F. and Gossard, A. C. (1992) Phys.
Rev. Lett. 69, 506.
2.
3.
4.
5.
6.
7.
8.
9.
10.
1l.
12.
13.
14.
15.
16.
17.
18.
19.
20.
2l.
22.
SHOT NOISE IN MESOSCOPIC SYSTEMS
M. J. M. DE JONG
Philips Research Laboratories

5656 AA Eindhoven, The Netherlands
AND
C. W. J. BEENAKKER
Instituut-Lorentz, University of Leiden

2300 RA Leiden, The Netherlands
Abstract. This is a review of shot noise, the time-dependent fluctuations in

the electrical current due to the discreteness of the electron charge, in small
conductors. The shot-noise power can be smaller than that of a Poisson process as a result of correlations in the electron transmission imposed by the
Pauli principle. This suppression takes on simple universal values in a symmetric double-barrier junction (suppression factor ~), a disordered metal
Loss of phase coherence has no
(factor i), and a chaotic cavity (factor
effect on this shot-noise suppression, while thermalization of the electrons
due to electron-electron scattering increases the shot noise slightly. SubPoissonian shot noise has been observed experimentally. So far unobserved
phenomena involve the interplay of shot noise with the Aharonov-Bohm
effect, Andreev reflection, and the fractional quantum Hall effect.
i).
1. Introduction
1.1. CURRENT FLUCTUATIONS
In 1918 Schottky [1] reported that in ideal vacuum tubes, where all sources
of spurious noise had been eliminated, there remained two types of noise
in the electrical current, described by him as the Wiirmeeffekt and the
Schroteffekt. The first type of noise became known as Johnson-Nyquist
noise (after the experimentalist [2] and the theorist [3] who investigated it),
or simply thermal noise. It is due to the thermal motion of the electrons
and occurs in any conductor. The second type of noise is called shot noise,
225
@ 1997 Kluwer Academic Publishers.
226
caused by the discreteness of the charge of the carriers of the electrical

current. Not all conductors exhibit shot noise.
Noise is characterized by its spectral density or power spectrum P(w),
which is the Fourier transform at frequency w of the current-current correlation function [4, 5],
00
P(w)
=2
dt eiwt(fl.I(t + to)fl.I(to)) .
(1)
-00
Here fl.I(t) denotes the time-dependent fluctuations in the current at a

given voltage V and temperature T. The brackets (... ) indicate an ensemble
average or, equivalently, an average over the initial time to. Both thermal
and shot noise have a white power spectrum - that is, the noise power
does not depend on w over a very wide frequency range. Thermal noise
(V = 0, T i= 0) is directly related to the conductance G by the fluctuationdissipation theorem [6],
(2)
P =4kBTG,
as long as liw kBT. Therefore, the thermal noise of a conductor does not
give any new information.
Shot noise (V i= 0, T = 0) is more interesting, because it gives information on the temporal correlation of the electrons, which is not contained
in the conductance. In devices such as tunnel junctions, Schottky barrier
diodes, p-n junctions, and thermionic vacuum diodes [4], the electrons are
transmitted randomly and independently of each other. The transfer of
electrons can be described by Poisson statistics, which is used to analyze
events that are un correlated in time. For these devices the shot noise has
its maximum value
(3)
P = 2eI == .?poisson ,
proportional to the time-averaged current I. (We assume I > 0 and V > 0
throughout this review.) Equation (3) is valid for w < 7-1, with 7 the
width of a one-electron current pulse. For higher frequencies the shot noise
vanishes. Correlations suppress the low-frequency shot noise below PPoisson'
One source of correlations, operative even for non-interacting electrons, is
the Pauli principle, which forbids multiple occupancy of the same singleparticle state. A typical example is a ballistic point contact in a metal,
where P = because the stream of electrons is completely correlated by the
Pauli principle in the absence of impurity scattering. Macroscopic, metallic
conductors have zero shot noise for a different reason, namely that inelastic
electron-phonon scattering averages out the current fluctuations.
Progress in nanofabrication technology has revived the interest in shot
noise, because nanostructures allow measurements to be made on "mesoscopic" length scales that were previously inaccessible. The mesoscopic
227
Figure 1. Schematic representation of the transport through the conductor. Incoming

states (1) are scattered into outgoing states (0), by a scattering region (dashed). A cross
section in lead 1 and its coordinates are indicated.
length scale is much greater than atomic dimensions, but small compared
to the scattering lengths associated with various inelastic processes. Mesoscopic systems have been studied extensively through their conductance
[7, 8, 9, 10]. Noise measurements are much more difficult, but the sensitivity of the experiments has made a remarkable progress in the last years.
Some theoretical predictions have been observed, while others still remain
an experimental challenge. This article is a review of the present status of
the field, with an emphasis on the theoretical developments. We will focus
on the scattering approach to electron transport, which provides a unified
description of both conductance and shot noise. For earlier reviews, see
Refs. [11, 12, 13]. For brief commentaries, see Refs. [14, 15].
1.2. SCATTERING THEORY
In his 1957 paper [16] Landauer discussed the problem of electrical conduction as a scattering problem. This has become a key concept in mesoscopic
physics [7, 8]. The conductor is modeled as a scattering region, connected
to electron reservoirs. The electrons inside each reservoir are assumed to be
in thermal equilibrium. Incoming states, occupied according to the FermiDirac distribution function, are scattered into outgoing states. At low temperatures the conductance is fully determined by the transmission matrix
of electrons at the Fermi level. The two-terminal Landauer formula [7, 17]
and its multi-terminal generalization [18, 19, 20, 21] constitute a general
framework for the calculation of the conductance of a phase-coherent sample. A scattering theory of the noise properties of mesoscopic conductors
was derived in Refs. [22, 23, 24, 25, 26, 27, 28]. The basic result is a relationship between the shot-noise power and the transmission matrix at the
Fermi level, analogous to the Landauer formula for the conductance. Here
we review the derivation of this result, following closely Biittiker's work
[26, 28].
Two leads are connected to an arbitrary scattering region (see Fig. 1).
Each lead contains N incoming and N outgoing modes at energy E. We
assume only elastic scattering so that energy is conserved. The incoming
228
and outgoing modes are related by a 2N x 2N scattering matrix S
(4)
where 11,01, h, O2 are the N-component vectors denoting the amplitudes
of the incoming (1) and outgoing (0) modes in lead 1 and lead 2. The
scattering matrix can be decomposed in N x N reflection and transmission
matrices,
S=
( 5 11
521
5 12 )
522
==
(r t;) ,
t
(5)
where the N x N matrix 5ba contains the amplitudes Sbn,am from incoming
mode m in lead a to outgoing mode n in lead b. Because of flux conservation
S is a unitary matrix. Moreover, in the presence of time-reversal symmetry
S is symmetric.
The current operator in lead 1 is given by
i(t)
=~L
JJ
00
00
dc' Io:{3(c, c') al(c) a{3(c') eit(c-c')/fi
de
o:,{3 0
(6)
where al(c) [ao:(c)] is the creation [annihilation] operator of scattering state

'lfJo:(r, c). We have introduced the indices a == (a, m), f3 == (b, n) and the
coordinate r = (x, y). The matrix element Io: f3 (E, E') is determined by the
value of the current at cross section Sl in lead 1,
Here, fix is the velocity operator in the x-direction. At equal energies, Eq.
(7) simplifies to [26, 28]
Iam,bn(c, c)
= 6a16ab 6mn -
L Slp,am(c) Sip,bn(C) .
p=l
(8)
The average current follows from
(9)
where fa is the Fermi-Dirac distribution function in reservoir a:
II (c) = f (c - E F - eV) ,
h(c) = f(c - EF) ,
f(x) = [1 + exp(x/kB T)t 1 ,
(lOa)
(lOb)
(10c)
229
with Fermi energy EF. The result is
00
(i(t))
~L
00
!dffa(f)Ioo(f, f)
~ !ddh(f) -
h(f)] Trt(f)tt(f) , (11)
0 0
where we have substituted Eq. (8) and used the unitarity of S. The linearresponse conductance, G == limv~o(I)/V, becomes
G=
he2
/00
df ( - 88cf ) Trt(f)tt (f) ,
(12)
o
which at zero temperature simplifies to the Landauer formula
e2
e2 N
(13)
G= -Trttt = - LTn.
n=l
Here t is taken at EF and Tn E [0,1] is an eigenvalue ofttt. The conductance

is thus fully determined by the transmission eigenvalues. Knowledge of the
transmission eigenstates, each of which can be a complicated superposition
of incoming modes, is not required.
In order to evaluate the shot-noise power we substitute the current
operator (6) into Eq. (1) and determine the expectation value. We use the
formula [28]
(at a2al (4) - (at ( 2) (ala4) = Ow523h (1 - h) == ~1234 ,
(14)
where e.g. 012 stands for 00j30(f - f'). Equation (14) shows that there are
cross correlations between different scattering states. Although this bears
no effect on the time-averaged current, it is essential for the current fluctuations. For the noise power one finds
P(w)
~:
00
00
00
00
! dt! df ! df' ! df" / df"'ei(fiw+E-E')t/fi
a,{3;y,o
X
00
-00
A
( f, f , ,f II ,f IJI)
I a{3 (f, f ')1,0 ("
f ,f"') Ua{3,o
00
2 ~ L ! dda j3(f, f
a,j3 0
+ nw)Ij3a(f + nw, f)fa(f)[l - fb(f + nw)] .

(15)
The low-frequency limit is found by substitution of Eq. (8),
00
2 ~ ! df {[h(l - h)
o
+ 12(1- h)] Trttt(l- ttt)
+ [h (1 - h) + 12 (1 - h)] Tr tt tt t t}
(16)
230
where we have again used the unitarity of S.

Equation (16) allows us to evaluate the noise for various cases. Below
we will assume that eV and k B T are small enough to neglect the energy
dependence of the transmission matrix, so that we can take t at c = EF.
Let us first determine the noise in equilibrium, i.e. for V = O. Using the
relation f(1 - f) = -kBTof jOe we find
e2
= 4kBThTrttt = 4kBTh I: Tn ,
e2
(17)
n=l
which is indeed the Johnson-Nyquist formula (2). For the shot-noise power
at zero temperature we obtain
= 2eV h
Trttt(l - ttt)
= 2eV h
I: Tn(1- Tn) .
(18)
n=l
Equation (18), due to Biittiker [26], is the multi-channel generalization

of the single-channel formulas of Khlus [22], Lesovik [24], and Yurke and
Kochanski [25]. One notes, that P is again only a function of the transmission eigenvalues.
It is clear from Eq. (18) that a transmission eigenstate for which Tn = 1
does not contribute to the shot noise. This is easily understood: At zero
temperature there is a non-fluctuating incoming electron stream. If there is
complete transmission, the transmitted electron stream will be noise free,
too. If Tn decreases, the transmitted electron stream deviates in time from
the average current. The resulting shot noise P is still smaller than PPoisson,
because the transmitted electrons are correlated due to the Pauli principle. Only if Tn 1, the transmitted electrons are uncorrelated, yielding
full Poisson noise (see Sec. 1.3.1). Essentially, the non-fluctuating occupation number of the incoming states is a consequence of the electrons being
fermions. In this sense, the suppression below the Poisson noise is due to
the Pauli principle. On the other hand, one must realize that the noise
suppression is not an exclusive property of fermions. It occurs for any incoming beam with a non-fluctuating occupation number, for example a
photon number state [29].
The generalization of Eq. (18) to the non-zero voltage, non-zero temperature case is [28, 30]
2
P = 2~
[2kBTT~ + Tn(1- Tn)eV coth(eVj2k B T)]
(19)
n=l
The crossover from the thermal noise (17) to the shot noise (18) depends
on the transmission eigenvalues.
231
As a final remark, we mention that in the above derivations the absence

of spin and valley degeneracy has been assumed for notational convenience.
It can be easily included. For a two-fold spin degeneracy this results in the
replacement of the e2 / h prefactors [such as in Eqs. (13) and (18)] by 2e 2 / h.
l,From now on, we simply use Go == degeneracy factor x e2 / h as the unit of
conductance and Po == 2eVGo as the unit of shot-noise power.
1.3. TWO SIMPLE APPLICATIONS
The above results are valid for conductors with arbitrary (elastic) scattering. If the transmission eigenvalues are known, the conduction and noise
properties can be readily calculated. Below, this is illustrated for two simple
systems. More complicated conductors are discussed in Secs. 2-4.
1.3.1. Tunnel barrier
In a tunnel barrier, electrons have a very small probability of being transmitted. We model this by taking Tn 1, for all n. Substitution into the formula for the shot noise (18) and the Landauer formula for the conductance
(13) yields P = .?poisson at zero temperature. For arbitrary temperature we
obtain from Eq. (19),
P = coth(eV/2kBT) .?poisson'
(20)
This equation, due to Pucel [31], describes the crossover from thermal noise
to full Poisson noise. For tunnel barriers this crossover is governed entirely
by the ratio eV/kBT and not by details of the conductor. This behavior has
been observed in various systems, see e.g. Refs. [32, 33]. Electron-electron
interactions can lead to modification of Eq. (20), see Ref. [34, 35, 36].
1.3.2. Quantum Point Contact
A point contact is a narrow constriction between two pieces of conductor.
If the width W of the constriction is much smaller than the mean free
path of the bulk material, but much greater than the Fermi wave length
AF, the conductance is given by the Sharvin formula [37], which in two
dimensions reads G = G o2W/ AF. In such a classical point contact the shot
noise is absent, as found by Kulik and Omel'yanchuk [38]. In a quantum
point contact W is comparable to AF. Experimentally, a quantum point
contact can be formed in a two-dimensional electron gas in an (AI,Ga)As
heterostructure [8]. The constriction is defined by depletion of the electron
gas underneath metal gates on top of the structure. Upon changing the gate
voltage Vg , the width W is varied. The conductance displays a stepwise
increase in units of Go as a function of Vg [39, 40]. This is caused by the
discrete number No = Int[2W/ AF] of modes at the Fermi energy which fit
232
(a)
o
,,
0..
0..
.- 3
(b)
,,
0.4
0.2
0
0.4
0
C..':l
0..
C..':l
0..
C..':l
~
0.2
C..':l
-0.6
-0.4
Vg (V)
Figure 2. (a) Conductance G (dashed line) and shot-noise power P (full line) versus
Fermi energy of a two-dimensional quantum point contact, according to the saddle-point
model, with Wy = 3w x . (b) Experimentally observed G and P versus gate voltage Vg
(unpublished data from Reznikov et ai. similar to the experiment of Ref. [49], but at a
lower temperature T = 0.4 K).
into the constriction width. As a result, No transmission eigenvalues equal

1, the others 0, yielding a quantized conductance according to Eq. (13).
Lesovik has predicted that the shot noise in a quantum point contact
is distinct from a classical point contact [24]. At the conductance plateaus
the shot noise is absent, as follows from Eq. (18). However, in between the
plateaus, where the conductance increases by Go, there is a transmission
eigenvalue which is between 0 and 1. As a consequence, the shot noise
has a peak. We illustrate this behavior with a model by Biittiker [41] of a
two-dimensional saddle-point potential,
(21)
where Vo is the potential at the saddle point, and Wx and Wy determine the
curvatures. The transmission eigenvalues at the Fermi energy are [41]
Tn = [1
+ exp( -21fen/liwx)rl, en == EF - Vo - (n -
~)liwy .
(22)
Results for the conductance and the shot-noise power are displayed in Fig.
2a. The shot noise peaks in between the conductance plateaus and is absent
on the plateaus. For large N, the peaks in the shot noise become negligible
with respect to the Poisson noise, in agreement with the classical result
[38]. More theoretical work on noise in quantum point contacts is given in
Refs. [42, 43, 44, 45].
The prediction by Lesovik of this quantum size-effect in the shot noise
formed a challenge for experimentalists. An early experiment was done
by Li et al. [46]. However, a difficulty in the interpretation was that the
frequency was not high enough to distinguish between shot noise and resistance fluctuations, which are also quite sensitive to changes in the width of
233
the point contact [47, 48]. Recent experiments at much higher frequencies
by Reznikov et al. [49] (see also Ref. [50]) and with very elaborate shielding
by Kumar et al. [51] have unambiguously demonstrated the occurrence of
suppressed shot noise on the conductance plateaus. Experimental data of
Reznikov et al. are shown in Fig. 2b.
1.4. KINETIC THEORY
The scattering theory of Sec. 1.2 fully takes into account the phase coherence of the electron wave function. If phase coherence is not essential,
one can use instead a semiclassical kinetic theory. The word "semiclassical"
means that classical mechanics is combined with the quantum-mechanical
Pauli principle. A semiclassical kinetic theory for shot noise has been developed by Kulik and Omel'yanchuk for a point contacts[38]' by Nagaev for
a diffusive conductor [52], and by the authors for an arbitrary conductor
[53, 54]. (Refs. [53, 54] correct Ref. [55].)
The theory is based on an extension of the Boltzmann equation to include fluctuations of the distribution function [56, 57, 58]. By analogy with
the Langevin equation in the theory of stochastic processes, this fluctuating
Boltzmann equation is called the Boltzmann-Langevin equation. We give a
brief summary of the method.
The fluctuating distribution function f(r, k, t) in the conductor equals
(211")d times the density of electrons with position r, and wave vector k, at
time t. The average over time-dependent fluctuations (I) := 1 obeys the
Boltzmann equation,
(:t + s) f( r, k, t)
d
= 0,
-:=-+v-+F-.
dt
at
ar
liak
(23a)
(23b)
The derivative (23b) (with v = lik/m) describes the classical motion in the
force field F(r) = -ea(r)/ar+ev x B(r), with electrostatic potential (r)
and magnetic field B (r). The term S1 accounts for the stochastic effects of
scattering. In the case of impurity scattering, the scattering term equals
S f(r, k, t) =
Jdk' Wkk' (r)[f(r, k,
t) - f(r, k', t)].
(24)
The kernel Wkk,(r) is the transition rate for scattering from k to k', which
may in principle also depend on r.
We consider the stationary situation, where J is independent of t. The
time-dependent fluctuations 8f := f -1 satisfy the Boltzmann-Langevin
234
equation [56, 57],
(! +8) 6f(r,k,t) =j(r,k,t),
(25)
where j is a fluctuating source term representing the fluctuations induced

by the stochastic nature of the scattering. The flux j has zero average,
(j) = 0, and covariance
(j(r, k, t) j(r', k', t')) = (27f)d 6(r - r') 6(t - t')J(r, k, k') .
(26)
The delta functions ensure that fluxes are only correlated if they are induced by the same scattering process. The flux correlator J depends on
the type of scattering and on J, but not on 6f. Due to the Pauli principle
the scattering possibilities of an incoming state depend on the occupation
of possible outgoing states. As a consequence, J is roughly proportional to
1(1- 1'). The precise correlator J for the impurity-scattering term (24) has
been derived by Kogan and Shul'man [57]. Scattering by a tunnel barrier
corresponds to another correlator [53, 54].
The kinetic theory can be applied to calculate various noise properties,
including the effects of electron-electron and electron-phonon scattering
[59, 60, 61]. In Refs. [53, 54] a general formula for the shot-noise power
has been derived from Eqs. (25) and (26). Further discussion of the kinetic
theory is outside the scope of this review. In the following Section, we
discuss an alternative method to calculate the effects of phase breaking
and other scattering processes.
1.5. PHASE BREAKING, THERMALIZATION,
AND INELASTIC SCATTERING
Noise measurements require rather high currents, which enhance the rate
of scattering processes other than purely elastic scattering. The phasecoherent transmission approach of Sec. 1.2 is then no longer valid. The
effects of dephasing and inelastic scattering on the shot noise have been
studied in Refs. [52,54,59,60,61,62,63,64,65,66,67]. Below, we discuss a
model [54, 64] in which the conductor is divided in separate, phase-coherent
parts connected by charge-conserving reservoirs. This model includes the
following types of scattering:
- Quasi-elastic scattering. Due to weak coupling with external degrees
of freedom the electron-wave function gets dephased, but its energy is
conserved. In metals, this scattering is caused by fluctuations in the
electromagnetic field [68].
- Electron heating. Electron-electron scattering exchanges energy between the electrons, but the total energy of the electron gas is con-
[1()
:::::'1::::.
y-
235
;' 2 . f 2 ()
f I2 ( )
Figure 3. Additional scattering inside the conductor is modeled by dividing it in two
parts and connecting them through another reservoir. The electron distributions in the
left and the right reservoir, h() and h(), are Fermi-Dirac distributions. The distribution /12() in the intermediate reservoir depends on the type of scattering.
served. The distribution function is therefore assumed to be a FermiDirac distribution at a temperature above the lattice temperature.
- Inelastic scattering. Due to electron-phonon interactions the electrons
exchange energy with the lattice. The electrons emerging from the
reservoir are distributed according to the Fermi-Dirac distribution, at
the lattice temperature T.
The model is depicted in Fig. 3. The conductors 1 and 2 are connected
via a reservoir with distribution function h2(E). The time-averaged current
1m through conductor m = 1,2 is given by
(Gde) JdE [idE) - h2(E)] ,
(27a)
(G2/e) JdE [h2(f) - h(E)].
(27b)
The conductance Gm == 1/ Rm = Go ~;:=1 T~m) , with T~m) the n-th transmission eigenvalue of conductor m. We assume small eV and kBT, so that
the energy dependence of the transmission eigenvalues can be neglected.
Current conservation requires that h = 12 == I. The total resistance of
the conductor is given by Ohm's law,
(28)
for all three types of scattering that we consider. Our model is not suitable
for transport in the ballistic regime or in the quantum Hall effect regime,
where a different type of "one-way" reservoirs is required [69, 70].
The time-averaged current (27) depends on the average distribution
h2(f) in the reservoir between conductors 1 and 2. In order to calculate
the current fluctuations, we need to take into account that this distribution
varies in time. We denote the time-dependent distribution by 112(E, t). The
fluctuating current through conductor 1 or 2 causes electrostatic potential
fluctuations 812(t) in the reservoir, which enforce charge neutrality. In
Ref. [64], the reservoir has a Fermi-Dirac distribution h2(E, t) = i[E EF - eV12 - e812(t)], with EF + eV12 the average electrochemical potential
236
in the reservoir. As a result, it is found that the shot-noise power P of the
entire conductor is given by [64]
(29)
In other words, the voltage fluctuations add. The noise powers of the two
segments depend solely on the time-averaged distributions [26, 28],
where Sm == Go ~;:==1 T~m) (1 - T~m)). The analysis of Ref. [64] is easily

generalized to arbitrary distribution h2' Then, we have i12(c, t) = h2[c e81/>12(t)]. It follows that Eqs. (29) and (30) remain valid, but h2(c) may be
different. Let us determine the shot noise for the three types of scattering.
Quasi-elastic scattering. Here, it is not just the total current which must
be conserved, but the current in each energy range. This requires
(31)
We note that Eq. (31) implies the validity of Eq. (28). Substitution of Eq.
(31) into Eqs. (29) and (30) yields at zero temperature the result [54]:
Electron heating. We model electron-electron scattering, where energy

can be exchanged between the electrons at constant total energy. We assume
that the exchange of energies establishes a Fermi-Dirac distribution h2(c)
at an electrochemical potential EF + eV12 and an elevated temperature T 12.
From current conservation it follows that
V12
= (RdR) V.
(33)
Conservation of the energy of the electron gas requires that T12 is such that
no energy is absorbed or emitted by the reservoir. This implies
(34)
with the Lorentz number 0 == ~(7rkB/e)2. At zero temperature in the
left and right reservoir and for R1 = R2 we have kBT12 = (v'3/27r)eV c:::::
237
0.28eV. For the shot noise at T

(30) the result [54]:
P
{Rf 8 1 + R~82
= 0,
we thus obtain using Eqs. (29) and
+ ~y'3R1R2 [ R1 (1 -
(1 + e+ 2R~821n (1 + e-
+ 2Ri 8 1 In
7r y'Rl/ 3R
7r y' R
R18d
+ R 2(1
- R28 2)
2)
2/ 3Rl) ]}R-
2 PPoisson .
(35)
Inelastic scattering. The distribution function of the intermediate reservoir is the Fermi-Dirac distribution at the lattice temperature T, with an
electrochemical potential /-t12 == EF + eV12, where V12 is given by Eq. (33).
This reservoir absorbs energy, in contrast to the previous two cases. The
zero-temperature shot-noise power is given by [64]:
(36)
This model will be applied to double-barrier junctions, chaotic cavities, and disordered conductors in Sees. 2-4. Quite generally, we will find
that quasi-elastic scattering has no effect on the shot noise, while electron
heating leads to a small enhancement of the shot noise. Inelastic scattering suppresses the shot noise in most cases, but not in the double-barrier
junction.
1.6. STATISTICS OF TRANSMITTED CHARGE
The conductance is a measure for the average number of electrons transmitted per unit time. The shot noise quantifies the variance of the transmitted
charge. Levitov and Lesovik have studied the full distribution function of
charge transmitted through a meso scopic conductor [71, 72]. This function
Pq(t) gives the probability that exactly q electrons have been transmitted
during a given time interval t. An alternative way to describe this distribution is through its characteristic function X(>', t). They are mutually related
according to [73]
00
X(\ t)
=L
Pq(t)e iqA , Pq(t)
J
7r
= 2~
d>' e-iqAx(\ t) .
q=O
(37)
-7r
The average number of electrons transmitted during a time t is given by

_
q(t) =
L qPq(t) = >--+02
lim "0' X(>', t) .
q=O
00
(38)
238
More generally, one can express the k-th moment !-Lk(t) of the distribution
by
!-Lk(t) == qk(t)
= l~ (8)k
i8)"
X().., t) .
(39)
The average current is simply I = e!-Ll(t)/t and the noise power equals
P = 2e2limt-too varq(t)/t = 2e 2 limHoo[!-L2(t) -!-Li(t)]/t.
Levitov and Lesovik [71] have computed the characteristic function at
zero temperature and at small voltage V. The result
N
XN().., t)
= II [(e iA -
1)Tn
+ 1]GoVt/e
(40)
n=l
is the characteristic function of the binomial or Bernoulli distribution: In

scattering channel n, the charge transmitted in a time interval t is due to
Go V t / e independent attempts to transmit an electron, each time with a
probability Tn. The fact that only one electron (within a single channel)
can be transmitted during a time e/GoV is due to the Pauli principle.
Only if Tn 1 for all n, Eq. (40) reduces to the characteristic function
of a Poisson process. Otherwise, the electrons are transmitted according to
sub-Poissonian statistics.
The distribution function of transmitted charge has been determined for
a normal-metal-superconductor point contact by Muzykantskii and Khmelnitskii [74], for a disordered conductor by Lee, Levitov, and Yakovets [75],
and for a double-barrier junction by one of the authors [76].
2. Double-Barrier Junction
2.1. RESONANT TUNNELING
In 1973 Tsu and Esaki predicted the occurrence of a negative differential

resistance due to resonant tunneling through two tunnel barriers in series
inside a semiconductor heterostructure [77]. The experimental observation
[78] opened a large field of research. The study of noise in resonant tunneling is a recent development, sparked by the demonstration by Li et at. [79]
that the shot noise in an (Al,Ga)As double-barrier junction may vary between one-half (for equal barrier heights) and the full Poisson noise (for very
unequal barrier heights). This observation, confirmed by other experiments
[80, 81, 82], has inspired many theoreticians [83, 84, 85, 86, 87, 88, 89].
Below we will only consider the zero-frequency, low-voltage limit, in order
to treat the double-barrier junction on the same footing as the other systems described in this review. We assume high tunnel barriers with modeindependent transmission probabilities fl' f2 1.
239
The transmission eigenvalues through the two barriers in series are given
by a Fabry-Perot type of formula,
T _
r 1r 2
n - 2 - rl - r 2 - 2J1 - r 1 - r2 cos rPn '
(41)
where rPn is the phase accumulated in one round trip between the barriers.
The density p(T) == (2:n 8(T - Tn)) of the transmission eigenvalues follows
from the uniform distribution of rPn between 0 and 27r [90],
Nr 1r 2
1
+ r 2) JT3(T+ _ T)' T
p(T) = 7r(rl
p(T) = 0 otherwise, with T_ = r1r2/7r2 and T+

density (42) is plotted in Fig. 4a.
The average conductance,
[T_, T+] ,
= 4r 1r 2/(r 1 + r2)2.
(42)
The
(43)
is just the series conductance of the two tunnel conductances. The resonances are averaged out by taking a uniform distribution of the phase
shifts rPn. Physically, this averaging corresponds either to an average over
weak disorder in the region between the barriers, or to a summation over
a large number of modes if the separation between the barriers is large
compared to the Fermi wave length, or to an applied voltage larger than
the width of the resonance.
For the shot-noise power one obtains
(44)
using Eqs. (42) and (43). This result was first derived by Chen and Ting
[83]. For asymmetric junctions, one barrier dominates the transport and the
shot noise equals the Poisson noise. For symmetric junctions, the shot noise
gets suppressed down to (P) = 1PPoisson for r 1 = r 2. The theoretical result
(44) is in agreement with the experimental observations [79, 80, 81, 82].
The suppression of the shot noise below PPoisson in symmetric junctions
is a consequence of the bimodal distribution of transmission eigenvalues, as
plotted in Fig. 4a. Instead of all Tn's being close to the average transmission
probability, the Tn's are either close to 0 or to 1. This reduces the sum
Tn(1-Tn). A similar suppression mechanism exists for shot noise in chaotic
cavities and in disordered conductors, see Sees. 3 and 4.
240
0.4
~ ~
r
fi
0.2
0.0
N'O N'
........
(a)
e;:
l-
_L_
(b)
(c)
o.oL-~==:::;:==:::::::;~:::::;::;:::::::::::~
0.0
0 .2
0.4
0.6
0.8
1.0
Figure 4. The distribution peT) of transmission eigenvalues T for (a) a double-barrier

junction, according to Eq. (41) with fl = f2 = 0.01; (b) a chaotic cavity, according
to Eq. (46) with Nl = N2 == N; and (c) a disordered wire, according to Eq. (51) with
L = 20. Each structure has a bimodal distribution.
Phase coherence is not essential for the occurrence of suppressed shot

noise. Davies et at. obtained the result (44) from a model of incoherent
sequential tunneling [84]. The method of Sec. 1.5 (with Gm = 8 m = GoNr m
for m = 1,2) shows that both quasi-elastic scattering [see Eq. (32)] and
inelastic scattering [see Eq. (36)] do not modify Eq. (44). Thermalization
of the electrons in the region between the barriers enhances the shot noise,
as follows from Eq. (35). For r 1 = r 2 we find
= [~+ ~ In (1 +e-1l'/v3)]
PPoisson
~ O.58PPoisson
(45)
which is slightly above the one-half suppression in the absence of thermalization. More theoretical work on the influence of internal scattering and
of dephasing on the shot noise in double-barrier junctions is contained in
Refs. [91, 92, 93, 94, 95].
2.2. COULOMB BLOCKADE
The suppression of the shot noise described in the previous Section is due
to correlations induced by the Pauli principle. Coulomb interactions are
241
another source of correlations among the electrons. A measure of the importance of Coulomb repulsion is the charging energy Ee = e2 /2C of a
single electron inside the conductor with a capacitance C. In open conductors, where C is large, charging effects are expected to be negligible
[96]. For closed conductors, such as a double-barrier junction, Ee can be
larger than kBT, in which case charging effects have a pronounced influence on the conduction [97]. If eV < Ee, conduction through the junction
is suppressed. This is known as the Coulomb blockade. At eV > E e , one
electron at a time can tunnel into the junction. The next electron can follow, only after the first electron has tunneled out of the junction. This is the
single-electron tunneling regime. The theory of shot noise in single-electron
tunneling devices has been developed by Korotkov et al. [98], Hershfield et
al. [99], and others [100, 101, 102, 103, 104, 105, 106].
Experiments have been reported by Birk, de Jong, and Schonenberger
[107]. Here, the double-barrier junction was formed by a scanning-tunneling
microscope positioned above a metal nanoparticle on an oxidized substrate.
Due to the small size of the particle, Ee 2: 1000kBT, at T = 4 K. The
relative heights of the two tunnel barriers can be modified by changing
the tip-particle distance. Experimental results for an asymmetric junction
are plotted in Fig. 5. The I-V characteristics display a stepwise increase
of the current with the voltage. (Rotating the plot 90 yields the usual
presentation of the 'Coulomb staircase. ') At small voltage, I ~ 0 due to
to the Coulomb blockade. At each subsequent step in I, the number of
excess electrons in the junction increases by one. The measured shot noise
oscillates along with the step structure in the 1-V curve. The full shot-noise
level P = PPoisson is reached at each plateau of constant I. In between, P
is suppressed down to ~ PPoisson. The experimental data are in excellent
agreement with the theory of Ref. [99].
A qualitative understanding of the periodic shot-noise suppression caused
by the Coulomb blockade goes as follows: On a current plateau in the I-V
curve, the number of electrons in the junction is constant for most of the
time. Only during a very short instance an excess electron occupies the
junction, leading to the transfer of one electron. This fast transfer process
is dominated by the highest tunnel barrier. Since the junction is asymmetric, Poisson noise is expected. The situation is different for voltages where
there is a step in the I-V curve. Here, two charge states are degenerate
in total energy. If an electron tunnels into the junction, it may stay for a
longer time, during which tunneling of the next electron is forbidden. Both
barriers are thus alternately blocked. This leads to a correlated current,
yielding a suppression of the shot noise.
An essential requirement for the Coulomb blockade is that G;S e2 / h.
For larger G the quantum-mechanical charge fluctuations in the junction
242
0.25
0.50
I (nA)
Figure 5.
Experimental results by Birk et al. [107] in the single-electron tunneling
regime. The double-barrier junction consists of a tip positioned above a nanoparticle on
a substrate. (a) Experimental voltage V versus current I. (b) Shot-noise power P versus
I. Squares: experiment; solid line: theory of Hershfield et al. [99].
become big enough to overcome the Coulomb blockade. The next Section
will discuss shot noise in a quantum dot, without including Coulomb interactions. This is justified as long as G ~ e 2 /h. For smaller G, the quantum
dot behaves essentially as the double-barrier junction considered above.
3. Chaotic Cavity
A cavity of sub-micron dimensions, etched in a semiconductor is called a

quantum dot. The transport properties of the quantum dot can be measured
by coupling it to two electron reservoirs, and bringing them out of equilibrium. We consider the generic case that the classical motion in the cavity
can be regarded as chaotic, as a result of scattering by randomly placed
impurities or by irregularly shaped boundaries. Then transport quantities
are insensitive to microscopic properties of the quantum dot, such as the
shape of the cavity and the degree of disorder.
A theory of transport through a chaotic cavity can be based on the single
assumption that the scattering matrix of the system is uniformly distributed
in the unitary group [108, 109J. This is the "circular ensemble" of randommatrix theory [110, 111J. The assumption of a uniform distribution of the
scattering matrix is valid if the coupling to the electron reservoirs occurs
via two ballistic point contacts, with a conductance Gm = GoNm, m =
1,2. (Otherwise a more general distribution, known as the "Poisson kernel"
applies [112J.) The presence of time-reversal symmetry is accounted for
243
by restricting the scattering matrix to the subset of symmetric unitary

matrices. This is known as the circular orthogonal ensemble (labeled by
the index (3 = 1). If any unitary matrix is equally probable, the ensemble
is called circular unitary ((3 = 2).
To compute the statistics of transport properties in a quantum dot one
needs to know the distribution of the transmission eigenvalues in the circular ensemble. In the most general case Nl #- N 2 , the transmission matrices
t12 and t21 are rectangular. The two matrix products t12tL and t21 t~l contain a common set of min(Nl' N 2 ) non-zero transmission eigenvalues. Only
these contribute to the transport properties. For Ni 1 the distribution
p(T) of the transmission eigenvalues is [113J
(46)
p(T) = 0 otherwise, with T_ = (Nl - N 2)2/(Nf.
+ Ni).
This density is
plotted in Fig. 4b.

The average conductance,
(47)
is the series conductance of the two point contacts. The average shot-noise
power,
(48)
is smaller than the Poisson noise. For two identical point contacts the suppression factor is one quarter [109], to be compared with the one-half suppression in a double-barrier junction (see Sec. 2.1) and the one-third suppression in a disordered wire (see Sec. 4.1).
The result (48) does not require phase coherence, as follows from Eq.
(32) using Sm = 0 for m = 1,2. However, it is affected by thermalization
of the electrons and also by inelastic scattering. l,From Eq. (35) we find in
the case of complete thermalization,
(49)
For Nl = N 2 , this yields P = (v'3/27f)PPoisson c::: O.28PPoisson. As follows
from Eq. (36), inelastic scattering suppresses the shot noise completely.
244
4. Disordered Metal
4.1. ONE-THIRD SUPPRESSION
We now turn to transport through a diffusive conductor of length L much

greater than the mean free path R, in the metallic regime (L localization
length). The average conductance is given by the Drude formula,
(50)
up to small corrections of order Go (due to weak localization). The mean
free path R = ad Rtr equals the transport mean free path Rtr times a numerical
coefficient, which depends on the dimensionality d of the Fermi surface
(a2 = 7f/2, a3 = 4/3).
From Eq. (50) one might surmise that for a diffusive conductor all the
transmission eigenvalues are of order R/ L, and hence 1. This would imply
the shot-noise power P = .?poisson of a Poisson process. This surmise is
completely incorrect, as was first pointed out by Dorokhov [114], and later
by Imry [115] and by Pendry, MacKinnon, and Roberts [116]. A fraction
R/ L of the transmission eigenvalues is of order unity (open channels), the
others being exponentially small (closed channels). For R L Nf, the
density of the Tn's is given by [114]
(T) - N
- 2L TV1 - T '
[T 1]
-,
(51)
p(T) = 0 otherwise, with T_ = 4e- 2L / l . The density p(T), plotted in Fig.

4c, is again bimodal with peaks near unit and zero transmission. Dorokhov
[114] obtained Eq. (51) from a scaling equation, which describes the evolution of p(T) on increasing L in a wire geometry [117, 118]. A derivation for
other geometries has been given by Nazarov [119].
One easily checks that the bimodal distribution (51) leads to the Drude
conductance (50). For the average shot-noise power it implies
NR
1
(P) = Po 3L = "3PPoisson
(52)
This suppression ofthe shot noise by a factor one-third is universal, in the

sense that it does not depend on the specific geometry nor on any intrinsic
material parameter (such as f). The one-third suppression was discovered
by Beenakker and Biittiker [64] in the way described above, and by others
using different methods [52, 119, 120, 121]. Nagaev's theory [52] is based
on the semiclassical Boltzmann-Langevin equation, see Sec. 1.4. One might
therefore infer that there is also a semiclassical derivation of the bimodal
245
0.5
"
~
0
0.4
P..
0.3
0.2
P-.
P-.
.....
..
-_.......
(1/4)V3
1/3
'
0.1
0.0
0
lee
lep
Figure 6.
The shot-noise power P of a disordered metallic wire as a function of its
length L, as predicted by theory. Indicated are the elastic mean free path f, the electron-electron scattering length lee and the electron-phonon scattering length lep. Dotted
lines are interpolations (after Ref. [127]).
distribution of transmission eigenvalues, which is the key ingredient of the

quantum-mechanical theory. Such a derivation is given in Ref. [122].
4.2. DEPENDENCE ON WIRE LENGTH
The one-third suppression of the shot noise breaks down if the conductor
becomes too short or too long. Upon decreasing the length of the conductor, when L becomes comparable to f!, the electron transport is no longer
diffusive, but enters the ballistic regime. Then the shot noise is suppressed
more strongly, according to [120]
P = ~ [1 - (1
+ L/f!)-3] PPoisson .
(53)
For L f! there is no shot noise, as in a ballistic point contact [38]. Equation (53) is exact for a special model of one-dimensional scattering, but
holds more generally within a few percent [54]. A Monte-Carlo simulation
in a wire geometry [123] is in good agreement with Eq. (53). The crossover
of the shot noise from the ballistic to the diffusive regime is plotted in Fig.
6. Upon increasing L at constant cross section of the conductor, one enters
the localized regime. Here, even the largest transmission eigenvalue is exponentially small [114], so that P = PPoisson' Shot noise in one-dimensional
chains for various models of disorder has been studied in Ref. [124].
Experimentally, the crossover from the metallic to the localized regime is
usually not reached, because phase coherence is broken when L is still much
smaller than the localization length N f!. In the remainder of this Section, we
apply the method of Sec. 1.5 to determine the effect of phase breaking and
other inelastic scattering events on the shot noise in a disordered metal
[54]. We divide the conductor into M segments connected by reservoirs,
taking the continuum limit M -+ 00. The electron distribution at position
x is denoted by f(f,X). At the ends of the conductor f(f,O) = h(f) and
246
1(f, L)
= !2(f),
i.e. the electrons are Fermi-Dirac distributed at temperature T and with electrochemical potential jL(O) = EF +eV and jL(L) = EF,
respectively. It follows from Eqs. (29) and (30) that the noise power is given
by
JJ
L
;L
00
dx df 1(f, x)[l - 1(f, x)] ,
(54)
a formula first obtained by Nagaev [52]. We evaluate Eq. (54) for the three
types of scattering discussed in Sec. 1.5.
Quasi-elastic scattering. Current conservation and the absence of inelastic scattering requires
1(f,X)
L-x
= -L- 1 (f, 0) + 1(f,L).
(55)
The electron distribution at x = LI2 is plotted in the inset of Fig. 7.

Substitution of Eq. (55) into Eq. (54) yields [52]
= ~
[4kBTG + eI coth(eVI2k BT)] .
(56)
At zero temperature the shot noise is one-third of the Poisson noise. The
same result follows from the phase-coherent theory [Eqs. (19) and (51)],
demonstrating that quasi-elastic scattering has no effect on the shot noise.
The temperature dependence of P is plotted in Fig. 7.
Electron heating. The electron-distribution function is a Fermi-Dirac
distribution with a spatially dependent electrochemical potential jL(x) and
temperature Te(x),
f(f,X)
jL(x)
Te(x)
{1 + exp [fkBTe(X)
- /t(x)]}-l
(57a)
L-x
EF+-L-eV,
(57b)
JT2 + (xl L)[l - (xl L)] V2 I LO ,
(57c)
cf. Eqs. (33) and (34). Equations (54) and (57) yield for the noise power
the result [59, 60, 125]
J3
= 2kBTG + 2eI [ 27r
J3]
(kBT) 2
eV
+ 27r arctan
plotted in Fig. 7. In the limit eV
P=
iJ3
(J3
eV )
27r kBT '
(58)
kBT one finds
.?poisson
~ 0.43 PPoisson
(59)
247
8
7
N 1 o~
_____,\
3<.i
':i:;"
05
".,I"~
A
','
""
00
. 0.0
" .........
"
"'0
"--;'
0.5 1.0
..-'~
(c-EF)/eV,A ~
20
eV / kBT
30
40
Figure 7. The noise power P versus voltage V for a disordered wire in the presence
of quasi-elastic scattering [solid curve, from Eq. (56)] and of electron heating [dashed
curve, from Eq. (58)]. The inset gives the electron distribution in the middle of the wire
at kBT = -doeV. The distribution for inelastic scattering is included for comparison
(dash-dotted). Experimental data of Steinbach, Martinis, and Devoret [127] on silver
wires at T = 50 mK are indicated for length L = 1j.tm (circles) and L = 30j.tm (dots).
Electron-electron scattering increases the shot noise above ~ ?Poisson because

the exchange of energies makes the current less correlated.
Inelastic scattering. The electron-distribution function is given by
(60)
with J.L(x) according to Eq. (57b). We obtain from Eqs. (54) and (60) that
the noise power is equal to the Johnson-Nyquist noise (2) for arbitrary
V. The shot noise is thus completely suppressed by inelastic scattering
~9, 60, 62, 64, 65, 6~.
The dependence of the shot-noise power on the length of a disordered
conductor is plotted in Fig. 6. The phase coherence length (between f and
lee) does not play a role. An early experimental demonstration of subPoissonian shot noise in a wire defined in an (Al,Ga)As heterostructure
was reported by Liefrink et al. [126]. The measurements were in agreement
with theory, but lacked the precision needed to discriminate between the
elastic and the hot-electron value. More accurate experiments by Steinbach,
Martinis, and Devoret [127] on silver wires are shown in Fig. 7. The noise in
a wire of L = 30 J.Lm is in excellent agreement with the hot-electron result
(58). For the L = 1 J.Lm wire the noise crosses over to the elastic result (56),
without quite reaching it.
248
Figure 8. Feynman paths enclosing a magnetic flux <1>. The paths in (a) correspond to
the same electron, the paths in (b) to two different electrons.
5. Aharonov-Bohm Effect
Since the current operator is a one-particle observable, the shot-noise power
(given by the current-current correlator) is a two-particle transport property. This is an essential difference with the conductance, which is a oneparticle property. The distinction between one-particle and two-particle
properties is relevant even without Coulomb interactions between the electrons, because of the quantum-mechanical exchange interaction. A striking demonstration of the two-particle nature of shot noise, discovered by
Biittiker [128, 129], occurs in the Aharonov-Bohm effect.
The Aharonov-Bohm effect in electrical conduction is a periodic oscillation of the conductance of a ring (or cylinder) as a function of the enclosed
magnetic flux <P. (For reviews, see Refs. [130, 131].) The fundamental periodicity of the oscillation is hie, because a flux increment of an integer
number of flux quanta changes by an integer multiple of 211" the phase difference between Feynman paths along the two arms of the ring (see Fig.
8a). Since the conductance is a one-particle property, the two interfering
Feynman paths must belong to the same electron, which on entering the
ring has a probability to traverse the ring either clock-wise or counterclock-wise. For a maximal amplitude of the conductance oscillations the
two probabilities should be approximately equal. The Lorentz force causes
an electron to traverse the ring preferentially in one of the two directions.
This is why the Aharonov-Bohm effect is suppressed by a strong magnetic
field [8].
Shot noise can exhibit Aharonov-Bohm oscillations which persist in a
strong magnetic field. The Lorentz force guides the right-moving electron
along the upper arc of the ring, and the left-moving electron along the
lower arc (see Fig. 8b). The absolute value squared 1'1/111 2, 1'1/1212 of each of
the two single-particle wave functions '1/11, '1/12 does not depend on <P, hence
the conductance is <P-independent. However, the absolute value squared of
the two-particle wave function 11l1(r1' r2) 12 = 1'1/11 (r1)'I/I2(r2) -'1/11 (r2)'I/I2(r1)12
does depend on <P. The flux sensitivity is an exchange effect, which vanishes
only if the two single-particle wave functions do not overlap.
249
To measure the flux sensitivity in the shot noise due to the exchange
effect one can not simply use the ring geometry of Fig. 8: Because each
electron is fully transmitted the shot noise vanishes completely in a strong
magnetic field. (The case of weak magnetic field has been studied in Ref.
[132].) Biittiker [129] has suggested a four-terminal configuration, where the
correlator of the current at two terminals is measured using the other two
terminals as current sources. Lesovik and Levitov [133] have proposed a twoterminal configuration, but with a time-dependent magnetic field. Ideally,
the shot noise should show a flux sensitivity while the time-averaged current
should not. Observation of this effect remains an experimental challenge.
6. Cooper Pairs
6.1. NORMAL-METAL-SUPERCONDUCTOR JUNCTIONS
If a normal metal is connected to a superconductor, the dissipative normal

current is converted into dissipationless supercurrent. This conversion goes
through a process called Andreev reflection [134]: Incoming electrons are
reflected into outgoing holes, with the transfer of a Cooper pair into the
superconductor. Since the elementary charge transfer now involves a charge
2e instead of e, one might expect a doubling of the shot noise in an NS
junction. Let us see how this follows from the theory [22, 74, 135].
We assume low temperatures and an applied voltage eV smaller than the
excitation gap 6. in the superconductor, so that the electrons and holes are
confined to the normal metal. The scattering from incoming into outgoing
states is described by the 2N x 2N reflection matrix R,
(61 )
where Ie, h, Ge, Gh are the N-component vectors denoting the amplitudes
of the incoming (I) and outgoing (G) electron (e) and hole (h) modes. The
reflection matrix R can be decomposed in N x N submatrices, where e.g.
rhe contains the reflection amplitudes from incoming electrons into outgoing
holes. The conductance [136, 137, 138] and the shot-noise power [135] are
given by
N
2GoTr rhert
= 2G o
L Rn ,
(62)
n=l
N
4PoTr rhert(l - rhert) = 4Po
Rn(1- Rn) ,
n=l
with Rn an eigenvalue of rhert, evaluated at the Fermi energy.
(63)
250
. . ~.::::: Js
c:
o
VI
VI
'0
-L-
a..
0...
.............
/\
Vl
z
"'- ... . . . . - ....... .2
0...
V
....,......,,--_-0-'
0.0'=-=_----,:-1-::-_ _
0.0
0.5
1.0
1.5
rLie
Figure 9. The shot-noise power (PNS) of a disordered NS junction with a barrier at the
NS interface (shown in the inset) as a function of its length L, for barrier transparencies
r = 1,0.9,0.8,0.6,0.4,0.2, 1 from bottom to top. For L = 0, (PNS) varies with r
according to Eq. (67) . If L increases it approaches the limiting value (PNS) =
for
each r (after Ref. [135]).
tel
The eigenvalue Rn can be related to the scattering properties of the

normal region through the Bogoliubov-de Gennes equation [139], which is
a 2 x 2 matrix Schrodinger equation for electron and hole wave functions.
In the presence of time-reversal symmetry, Rn can be expressed in terms
of the transmission eigenvalue Tn of the normal region [140]:
Rn = T;(2 - Tn )-2 .
(64)
Substitution into Eqs. (62) and (63) yields [135, 140]
E
N
Go
2T2
(2 - ;n)2 '
R .;... 16T~(1 - Tn)

o L.,.. (2 _ 1:)4 .
n=l
(65)
(66)
As in the normal state, scattering channels which have Tn = 0 or Tn = 1

do not contribute to the shot noise. However, the way in which partially
transmitting channels contribute is entirely different from the normal state
result (18).
For a planar tunnel barrier (Tn = f for all n) one finds [22]
P
_ R N I6f2 (1- f)
NS -
(2 _ f)4
8(1 - f)
= (2 _ f)2 .?poisson,
(67)
251
which for r 1 simplifies to PNS = 4eI = 2.?poisson. This can be interpreted

as an uncorrelated current of 2e-charged particles.
Since Eq. (66) is valid for arbitrary scattering region, we can easily
determine the average shot-noise power for a double-barrier junction in
series with a superconductor,
(68)
for a chaotic cavity in series with a superconductor,
x2
(PNS) = (1
with X = 4NIN2/(Nl
+ N2)2,
+ x)2( v"f+X _
1) .?poisson,
(69)
and for a disordered NS junction [135],
(~s) =
3" .?poisson
(70)
The average was computed with the densities (42), (46), and (51) of transmission eigenvalues, respectively. The shot noise in a disordered NS junction
with a barrier at the NS interface is plotted in Fig. 9. It makes the connection between the results (67) and (70). We have indeed found that for
a high tunnel barrier and for a disordered NS junction the shot noise is
doubled with respect to the normal-state results. For other systems, the
relation is more complicated.
More theoretical work on shot noise in NS systems is given in Refs.
[141, 142, 143]. The effects of the Coulomb blockade on the shot noise in
low-capacitance NSN junctions are described in Refs. [144, 145]. With one
inconclusive exception [146], no experimental observation of shot noise in
NS junctions has been reported, yet.
6.2. JOSEPHSON JUNCTIONS
A Josephson junction contains two normal-metal-superconductor interfaces, with a phase difference of the superconducting order parameter.
Such an SNS junction sustains a current I() in equilibrium, i.e. even
if the voltage difference V between the superconductors vanishes. Since
this supercurrent is a ground-state property, it can not fluctuate by itself. (It exhibits no shot noise.) Time-dependent fluctuations result from
quasiparticles which are excited at any finite temperature T. Their zerofrequency power density P() is related to the linear-response conductance
G() = limv--+o 8I()/8V of the Josephson junction in the same way as in
the normal state,
(71)
252
cf. Eq. (2). A remarkable difference with Johnson-Nyquist noise in a normal metal is that P( ) may actually increase with decreasing temperature,
because of the rapid increase of G () when T -+ O. Because thermal noise
falls outside the scope of our review, we do not discuss this topic further.
The interested reader is referred to Refs. [147, 148J.
7. Quantum Hall Effect

In a strong magnetic field the scattering channels of a two-dimensional
electron gas consist of edge states. Edge states at opposite edges propagate
in opposite directions. In the absence of scattering from one edge to the
other, each of the scattering channels at the Fermi level is transmitted with
probability Tn = 1. This is the regime of the (integer) quantum Hall effect.
(For reviews, see Refs. [8, 149J.) The conductance G = (e 2 /h) L:n Tn shows
plateaus at integer multiples of e2 / h as a function of magnetic field. The
implication for the shot-noise power P ex: L:n Tn(l - Tn) is that it should
vanish on the plateaus, similar to the situation in a quantum point contact,
see Sec. 1.3.2.
Biittiker [26, 28J considered the noise in the four-terminal conductor
depicted in Fig. 10. It is assumed that the transmission probability of all
edge channels but one is reduced to zero by means of a gate across the conductor. The remaining non-zero transmission probability is denoted by T.
A current flows between contacts 1 and 2 (voltage difference V), while contacts 3 and 4 are voltage probes. This four-terminal configuration requires
a generalization of the two-terminal formulas of Sec. 1.2. The current fa in
contact a is related to the voltages l'b at the contact b by the scattering
matrix [18J,
L TrS~bSabl
fa = e [Va Tr (1- slasaa) l'b

h
ba
2
'
(72)
where we have assumed zero temperature. The correlator
J
00
Pab
=2
dt (!1fa(t + to)!1fb(tO))
(73)
-00
The prime in the summation over c and d means a restriction to terms with
> Vd The two-terminal formula (18) is recovered if c, d = a, b. Application to the four-terminal geometry of Fig. 10 shows that Pab vanishes if a
Vc
253
4
Figure 10. Four-terminal conductor in the regime of the quantum Hall effect. The spatial
location of the edge states is indicated, as well as their direction of motion (solid lines
with arrows). A barrier (shaded region) causes scattering from one edge to the other
(dashed lines). The corresponding scattering probabilities are indicated.
or b equals 1 or 3, while [26, 28]

(75)
Equation (75) assumes that the voltages on all terminals are fixed, while
the current fluctuates in time. Usually, one measures voltage fluctuations at
fixed currents. Current and voltage are linearly related by Eq. (72), which
at low frequencies holds both for the time-average and for the fluctuations.
The resulting voltage-noise power measured between contacts 1 and 4 or
between contacts 2 and 3 is zero. The noise power measured between any
other pair of contacts equals 2eV(hle 2 )(1 - T)IT [26, 28]. The voltage
fluctuations diverge as T -+ 0 with increasing barrier height. Experiments
in support of the edge-channel description of shot noise in the quantum
Hall effect have been reported by Washburn et al. [150].
If the magnetic field becomes so strong that only a single edge channel
remains at the Fermi level, one enters the regime of the fractional quantum
Hall effect. Plateaus in the conductance now occur at e2 /mh, m = 1,3,5, ...
(and at other odd-denominator fractions of e2 I h as well) [149]. The quasiparticle excitations of the electron gas have a fractional charge e* = elm.
Since the shot noise, in contrast to the conductance, is sensitive to the
charge of the carriers, one might hope to be able to find evidence for fractionally charged quasi-particles in shot-noise measurements. The theory
has been developed in Refs. [151, 152, 153]. For very low barrier heights,
Poisson noise with a fractional charge e* is expected in the backscattered
current (Imax - It), with Imax = (e* Ih)eV the current in the absence of a
barrier. This yields for the shot noise
Pn
= 2e*(Imax -
It) .
(76)
254
For high barriers the usual Poisson noise P

ments remain to be done.
2eh is recovered. Experi-
Acknowledgements
We would like to thank our collaborators in this research: H. Birk, M.

Biittiker, J. I. Dijkhuis, H. van Houten, F. Liefrink, L. W. Molenkamp, and
C. Schonenberger. Permission to reproduce experimental results was kindly
given by M. Reznikov and A. H. Steinbach. Research at Leiden University
is supported by the Dutch Science Foundation NWO /FOM.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
W. Schottky, Ann. Phys. (Leipzig) 57, 541 (1918).

M. B. Johnson, Phys. Rev. 29, 367 (1927).
H. Nyquist, Phys. Rev. 32, 110 (1928).
A. van der Ziel, Noise in Solid State Devices and Circuits (Wiley, New York, 1986).
W. B. Davenport and W. L. Root, An Introduction to the Theory of Random
Signals and Noise (IEEE Press, New York, 1987).
H. B. Callen and T. W. Welton, Phys. Rev. 83, 34 (1951).
Y. Imry, in Directions in Condensed Matter Physics, Vol. 1, edited by G. Grinstein
and G. Mazenko (World Scientific, Singapore, 1986).
C. W. J. Beenakker and H. van Houten, Solid State Phys. 44, 1 (1991).
Mesoscopic Phenomena in Solids, edited by B. L. Altshuler, P. A. Lee, and R. A.
Webb (North-Holland, Amsterdam, 1991).
Mesoscopic Quantum Physics, edited by E. Akkermans, G. Montambaux, J.-L.
Pichard, and J. Zinn-Justin (North Holland, Amsterdam, 1995).
Th. Martin, in Coulomb and Interference Effects in Small Electronic Structures,
edited by D. C. Glattli, M. Sanquer, and J. Tran Thanh Van (Editions Frontieres,
France, 1994).
R. Landauer, Ann. N. Y. Acad. Sc. 755, 417 (1995); Physica B 227, 156 (1996).
M. Biittiker, J. Math. Phys. 37, 4793 (1996).
L. P. Kouwenhoven, Science 271, 1689 (1996).
M. J. M. de Jong, Phys. World 9 (8), 22 (1996).
R. Landauer, IBM J. Res. Dev. 1, 233 (1957); 32, 306 (1988).
D. S. Fisher and P. A. Lee, Phys. Rev. B 23, 6851 (1981).
M. Biittiker, Phys. Rev. Lett. 57, 1761 (1986); IBM J. Res. Dev. 32, 317 (1988).
A. D. Stone and A. Szafer, IBM J. Res. Dev. 32, 384 (1988).
H. U. Baranger and A. D. Stone, Phys. Rev. B 40, 8169 (1989).
K. Shepard, Phys. Rev. B 43, 11623 (1991).
V. A. Khlus, Zh. Eksp. Teor. Fiz. 93, 2179 (1987) [Sov. Phys. JETP 66, 1243
(1987}).
R. Landauer, Physica D 38, 226 (1989).
G. B. Lesovik, Pis'ma Zh. Eksp. Teor. Fiz. 49, 513 (1989) [JETP Lett. 49, 592
(1989}).
B. Yurke and G. P. Kochanski, Phys. Rev. B 41, 8184 (1990).
M. Biittiker, Phys. Rev. Lett. 65, 2901 (1990).
R. Landauer and Th. Martin, Physica B 175, 167 (1991).
R. C. Liu and Y. Yamamoto, in Quantum Dynamics of Submicron Structures,
edited by H. A. Cerdeira, B. Kramer, and G. Schon, NATO ASI Series E291
(Kluwer, Dordrecht, 1995).
255
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
Th. Martin and R. Landauer, Phys. Rev. B 45,1742 (1992).

R. A. Pucel, Proc. IRE 49, 1080 (1961).
S. T. Liu and A. van der Ziel, Physica 37, 241 (1967).
G. Lecoy and L. Gouskov, Phys. Stat. Sol. 30, 9 (1968).
T.-L. Ho, Phys. Rev. Lett. 22, 2060 (1983); E. Ben-Jacob, E. Mottola, and G.
Schon, ibid. 22, 2064 (1983).
G. Schon, Phys. Rev. B 32, 4469 (1985).
H. Lee and L. S. Levitov, Phys. Rev. B 53, 7383 (1996).
Yu. V. Sharvin, Zh. Eksp. Teor. Fiz. 48, 984 (1965) [Sov. Phys. JETP 21, 655
(1965)].
1. O. Kulik and A. N. Omel'yanchuk, Fiz. Nizk. Temp. 10, 305 (1984) [Sov. J. Low
Temp. Phys. 10, 158 (1984)].
B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. G. Williamson, L. P.
Kouwenhoven, D. van der Marel, and C. T. Foxon, Phys. Rev. Lett. 60, 848 (1988).
D. A. Wharam, T. J. Thornton, R. Newbury, M. Pepper, H. Ahmed, J. E. F. Frost,
D. G. Hasko, D. C. Peacock, D. A. Ritchie, and G. A. C. Jones, J. Phys. C 21,
L209 (1988).
S.-R. Yang, Solid State Commun. 81, 375 (1992).
T. Kuhn, L. Reggiani, and L. Varani, Superlatt. Microstruct. 11, 205 (1992).
C. W. J. Beenakker and J. A. Melsen, Phys. Rev. B 50, 2450 (1994).
L. Y. Chen and S. C. Ying, Mod. Phys. Lett. B 9, 573 (1995).
Y. P. Li, D. C. Tsui, J. J. Heremans, J. A. Simmons, and G. W. Weimann, Appl.
Phys. Lett. 57, 774 (1990).
G. Timp, R. E. Behringer, and J. E. Cunningham, Phys. Rev. B 42, 9259 (1990).
C. Dekker, A. J. Scholten, F. Liefrink, R. Eppenga, H. van Houten, and C. T.
Foxon, Phys. Rev. Lett. 66, 2148 (1991).
M. Reznikov, M. Heiblum, H. Shtrikman, and D. Mahalu, Phys. Rev. Lett. 75,
3340 (1995).
R. C. Liu, B. Odom, J. Kim, and Y. Yamamoto, in Proc. 23rd Int. Con/. on the
Physics of Semiconductors, edited by M. Scheffler and R. Zimmermann (World
Scientific, Singapore, 1996).
A. Kumar, L. Saminadayar, D. C. Glattli, Y. Jin, and B. Etienne, Phys. Rev. Lett.
76, 2778 (1996).
K. E. Nagaev, Phys. Lett. A 169, 103 (1992).
M. J. M. de Jong and C. W. J. Beenakker, Phys. Rev. B 51, 16867 (1995).
M. J. M. de Jong and C. W. J. Beenakker, Physica A 230, 219 (1996).
C. W. J. Beenakker and H. van Houten, Phys. Rev. B 43, 12066 (1991).
B. B. Kadomtsev, Zh. Eksp. Teor. Fiz. 32, 943 (1957) [Sov. Phys. JETP 5, 771
(1957)].
Sh. M. Kogan and A. Ya. Shul'man, Zh. Eksp. Teor. Fiz. 56, 862 (1969) [Sov.
Phys. JETP 29, 467 (1969)].
S. V. Gantsevich, V. L. Gurevich, and R. Katilius, Rivista Nuovo Cimento 2 (5),
1 (1979).
K. E. Nagaev, Phys. Rev. B 52, 4740 (1995).
V. 1. Kozub and A. M. Rudin, Phys. Rev. B 52, 7853 (1995).
V. L. Gurevich and A. M. Rudin, Phys. Rev. B 53, 10078 (1996).
A. Shimizu and M. Ueda, Phys. Rev. Lett. 69, 1403 (1992).
M. Ueda and A. Shimizu, J. Phys. Soc. Jpn. 62, 2994 (1993).
C. W. J. Beenakker and M. Biittiker, Phys. Rev. B 46, 1889 (1992).
R. Landauer, Phys. Rev. B 47, 16427 (1993).
R. C. Liu and Y. Yamamoto, Phys. Rev. B 49, 10520 (1994); 50, 17411 (1994);
53, 7555(E) (1996).
D. Zheng, Z. D. Wang, and Y. Liu, Phys. Lett. A 208, 375 (1995).
B. L. Altshuler, A. G. Aronov, and D. E. Khmelnitskii, J. Phys. C 15, 7367 (1982);
256
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
A. Stern, Y. Aharonov, and Y. Imry, Phys. Rev. B 41, 3436 (1990).

M. Biittiker, in Noise in Physical Systems and 1/f Fluctuations, edited by V.
Bareikis and R. Katilius (World Scientific, Singapore, 1995).
L. S. Levitov and G. B. Lesovik, Pis'ma Zh. Eksp. Teor. Fiz. 58, 225 (1993) [JETP
Lett. 58, 230 (1993)].
L. S. Levitov, H. Lee, and G. B. Lesovik, J. Math. Phys. 37, 4845 (1996).
H. Cramer, Mathematical Methods of Statistics (Princeton University Press,
Princeton, 1946).
B. A. Muzykantskii and D. E. Khmelnitskii, Phys. Rev. B 50, 3982 (1994); Physica
B 203, 233 (1994).
H. Lee, L. S. Levitov, and A. Yu. Yakovets, Phys. Rev. B 51, 4079 (1995).
M. J. M. de Jong, Phys. Rev. B 54, 8144 (1996).
R. Tsu and L. Esaki, Appl. Phys. Lett. 22, 562 (1973).
L. L. Chang, L. Esaki, and R. Tsu, Appl. Phys. Lett. 24, 593 (1974).
Y. P. Li, A. Zaslavsky, D. C. Tsui, M. Santos, and M. Shayegan, Phys. Rev. B 41,
8388 (1990).
E. R. Brown, C. D. Parker, A. R. Calawa, and M. J. Manfra, IEEE Trans. Electron
Devices 38, 2716 (1991).
P. Ciambrone, M. Macucci, G. Iannacone, B. Pellegrini, M. Lazzarino, L. Sorba,
and F. Beltram, Electron. Lett. 31, 503 (1995).
H. C. Liu, J. Li, G. C. Aers, C. R. Leavens, M. Buchanan, and Z. R. Wasilewski,
Phys. Rev. B 51, 5116 (1995).
L. Y. Chen and C. S. Ting, Phys. Rev. B 43, 4534 (1991).
J. H. Davies, P. Hyldgaard, S. Hershfield, and J. W. Wilkins, Phys. Rev. B 46,
9620 (1992).
L. Y. Chen and C. S. Ting, Phys. Rev. B 46, 4714 (1992).
H. Sheng and S.-J. Chua, J. Phys. D 27, 137 (1994).
S.-J. Xiong, J. Appl. Phys. 78, 6079 (1995).
0. L. Bo and Yu. Galperin, J. Phys. Condens. Matter 8, 3033 (1996).
G. Iannacone, M. Macucci, and B. Pellegrini (preprint, cond-mat/9607134).
J. A. Melsen and C. W. J. Beenakker, Physica B 203, 219 (1994).
M. A. Alam and A. N. Khondker, IEEE Trans. Electron Dev. 39, 2184 (1992).
J. Carlos Egues, S. Hershfield, and J. W. Wilkins, Phys. Rev. B 49,13517 (1994).
Yu. M. Galperin, N. Zou, and K A. Chao, Phys. Rev. B 49, 13728 (1994).
J. H. Davies, J. Carlos Egues, and J. W. Wilkins, Phys. Rev. B 52, 11259 (1995).
Yu. V. Nazarov and J. J. R. Struben, Phys. Rev. B 53, 15446 (1996).
M. Biittiker, in Noise in Physical Systems and 1/f Fluctuations, edited by P. H.
Handel and A. L. Chung (AlP, New York, 1993).
D. V. Averin and K K Likharev, in Ref. [9].
A. N. Korotkov, D. V. Averin, K K Likharev, and S. A. Vasenko, in SingleElectron Tunneling and Mesoscopic Devices, edited by H. Koch and H. Liibbig
(Springer, Berlin, 1992).
S. Hershfield, J. H. Davies, P. Hyldgaard, C. J. Stanton, and J. W. Wilkins, Phys.
Rev. B 47, 1967 (1993).
A. Levy Yeyati, F. Flores, and E. V. Anda, Phys. Rev. B 47, 10543 (1993); E. V.
Anda and A. Latge, ibid. 50, 8559 (1994).
K-M. Hung and G. Y. Wu, Phys. Rev. B 48, 14687 (1993).
U. Hanke, Yu. M. Galperin, K A. Chao, and N. Zou, Phys. Rev. B 48, 17209
(1993); U. Hanke, Yu. M. Galperin, and K A. Chao, ibid. 50, 1595 (1994).
A. Imamoglu and Y. Yamamoto, Phys. Rev. Lett. 70, 3327 (1993).
W. Krech and H. O. Miiller, Z. Phys. B 91, 423 (1993).
A. N. Korotkov, Phys. Rev. B 49, 10381 (1994).
H. O. Miiller, A. Hiidicke, U. Hanke, and K A. Chao (preprint, cond-mat/9603098).
H. Birk, M. J. M. de Jong, and C. Sch6nenberger, Phys. Rev. Lett. 75, 1610 (1995).
257
108.
109.
H. U. Baranger and P. A. Mello, Phys. Rev. Lett. 73, 142 (1994).

R. A. Jalabert, J.-L. Pichard, and C. W. J. Beenakker, Europhys. Lett. 27, 255
(1994).
110. M. L. Mehta, Random Matrices (Academic, New York, 1991).
111. C. W. J. Beenakker, Rev. Mod. Phys. (to be published).
112. P. W. Brouwer, Phys. Rev. B 51, 16878 (1995).
113. Yu. V. Nazarov, in Quantum Dynamics of Submicron Structures, edited by H. A.
Cerdeira, B. Kramer, and G. Schon, NATO ASI Series E291 (Kluwer, Dordrecht,
1995).
114. O. N. Dorokhov, Solid State Commun. 51, 381 (1984).
115. Y. Imry, Europhys. Lett. 1, 249 (1986).
116. J. B. Pendry, A. MacKinnon, and P. J. Roberts, Proc. R. Soc. London Ser. A 437,
67 (1992).
117. O. N. Dorokhov, Pis'ma Zh. Eksp. Teor. Fiz. 36, 259 (1982) [JETP Lett. 36, 318
(1982)].
118. P. A. Mello, P. Pereyra, and N. Kumar, Ann. Phys. (N. Y.) 181, 290 (1988).
119. Yu. V. Nazarov, Phys. Rev. Lett. 73, 134 (1994).
120. M. J. M. de Jong and C. W. J. Beenakker, Phys. Rev. B 46, 13400 (1992).
121. B. L. Altshuler, L. S. Levitov, and A. Yu. Yakovets, Pis'ma Zh. Eksp. Teor. Fiz.
59, 821 (1994) [JETP Lett. 59, 857 (1994)].
122. M. J. M. de Jong and C. W. J. Beenakker, in Coulomb and Interference Effects
in Small Electronic Structures, edited by D. C. Glattli, M. Sanquer, and J. Tran
Thanh Van (Editions FronW~res, France, 1994).
123. R. C. Liu, P. Eastman, and Y. Yamamoto (preprint).
124. V. I. Melnikov, Phys. Lett. A 198, 459 (1995).
125. J. M. Martinis and M. H. Devoret (unpublished).
126. F. Liefrink, J. I. Dijkhuis, M. J. M. de Jong, L. W. Molenkamp, and H. van Houten,
Phys. Rev. B 49, 14066 (1994).
127. A. H. Steinbach, J. M. Martinis, and M. H. Devoret, Phys. Rev. Lett. 76, 3806
(1996).
128. M. Biittiker, Physica B 175, 199 (1991).
129. M. Biittiker, Phys. Rev. Lett. 68, 843 (1992).
130. S. Washburn and R. A. Webb, Adv. Phys. 35, 375 (1986).
131. A. G. Aronov and Yu. V. Sharvin, Rev. Mod. Phys. 59, 755 (1987).
132. M. A. Davidovich and E. V. Anda, Phys. Rev. B 50, 15453 (1994).
133. G. B. Lesovik and L. S. Levitov, Phys. Rev. Lett. 72, 724 (1994).
134. A. F. Andreev, Zh. Eksp. Teor. Fiz. 46, 1823 (1964) [Sov. Phys. JETP 19, 1228
(1964)].
135. M. J. M. de Jong and C. W. J. Beenakker, Phys. Rev. B 49, 16070 (1994).
136. G. E. Blonder, M. Tinkham, and T. M. Klapwijk, Phys. Rev. B 25, 4515 (1982).
137. C. J. Lambert, J. Phys. Condens. Matter 3, 6579 (1991).
138. Y. Takane and H. Ebisawa, J. Phys. Soc. Jpn. 61, 2858 (1992).
139. P. G. de Gennes, Superconductivity of Metals and Alloys (Benjamin, New York,
1966).
140. C. W. J. Beenakker, Phys. Rev. B 46, 12841 (1992).
141. J. P. Hessling, V. S. Shumeiko, Yu. M. Galperin, and G. Wendin, Europhys. Lett.
34, 49 (1996).
142. M. P. Anantram and S. Datta, Phys. Rev. B 53, 16390 (1996).
143. Th. Martin, Phys. Lett. A 220, 137 (1996).
144. U. Hanke, M. Gisselfiilt, Yu. Galperin, M. Jonson, R. I. Shekhter, and K. A. Chao,
Phys. Rev. B 50, 1953 (1994); u. Hanke, Yu. Galperin, K. A. Chao, M. Gisselfiilt,
M. Jonson, and R. I. Shekhter, ibid. 51, 9084 (1995).
145. W. Krech and H.-O. Miiller, Mod. Phys. Lett. B 8, 605 (1994).
146. A. N. Vystavkin and M. A. Tarasov, Pis'ma Zh. Tekh. Fiz. 9, 869 (1983) [Sov.
Tech. Phys. Lett. 9, 373 (1983)].
258
147.
148.
149.
150.
151.
152.
153.
A.Martin-Rodero, A. Levy Yeyati, and F. J. Garcia-Vidal, Phys. Rev. B 53, 8891

(1996).
D. Averin and H. T. Imam, Phys. Rev. Lett. 76, 3814 (1996).
R. E. Prange and S. M. Girvin, The Quantum Hall Effect (Springer, New York,
1990).
S. Washburn, R. J. Haug, K. Y. Lee, and J. M. Hong, Phys. Rev. B 44, 3875
(1991).
C. L. Kane and M. P. A. Fisher, Phys. Rev. Lett. 72, 724 (1994).
P. Fendley, A. W. W. Ludwig, and H. Saleur, Phys. Rev. Lett. 75, 2196 (1995).
C. de C. Chamon, D. E. Freed, and X. G. Wen, Phys. Rev. B 51, 2363 (1995).
ADMITTANCE AND NONLINEAR TRANSPORT IN QUANTUM

WIRES, POINT CONTACTS, AND RESONANT TUNNELING
BARRIERS
M. BUTTIKER AND T. CHRISTEN
Department of Theoretical Physics,

University of Geneva,
24, Quai E. Ansermet, CH-1211 Geneva
4,
Switzerland
1. Introduction
The admittance determines the current response of an electrical conductor to oscillating voltages applied to its contacts. The admittance and the
nonlinear dc-transport have in common that they require both an understanding of the charge distribution which is established as the conductor
is driven away from its equilibrium state. The dynamic and the nonlinear
transport can thus be viewed as probes of the charge response of the conductor. In contrast, linear dc-transport is determined by the equilibrium
charge distribution alone. Thus the dynamic conductance and the nonlinear transport provide information which cannot be gained from a linear
dc-measurement.
It is the purpose of this article to provide a discussion of ac-transport
and nonlinear transport in small mesoscopic conductors. Our discussion
leans heavily on the scattering approach to electrical dc-conductance and
can be viewed as an extension of this approach to treat transport beyond the
stationary ohmic regime. Our emphasis is on the low-frequency admittance
for which expressions can be derived which are very general and can be
applied to a large class of problems. Similarly our discussion emphasizes the
departure from linear ohmic transport in the case when nonlinearity first
becomes apparent. The principles which govern the low-frequency behavior
and the weakly nonlinear behavior apply of course as well for high-frequency
transport and for strongly nonlinear transport. We underline this fact by
treating the simple example of a resonant tunneling barrier for its entire
frequency range and by calculating the nonlinear I-V characteristic for a
large range of applied voltages.
259
260
The general guiding principles of our discussion are charge and current conservation and the need to obtain gauge invariant expressions. The
conservation of charge becomes a fundamental requirement under the following condition [1]: Suppose that it is possible to find a volume n which
encloses the mesoscopic part of the conductor including a portion of its
contacts such that all electric field lines remain within this volume. Clearly,
if the mesoscopic conductor is formed with the help of gates the volume n
must be large enough to enclose also a portion of the gates. The electric
field is then localized within n. Such localized electric (and magnetic) field
distributions are always implicitly assumed when a circuit is represented
simply in terms of R, C and L elements. If the field is localized in n the
electric flux through the surface of that volume is zero and according to
the law of Gauss the total charge inside this volume is zero. Consequently,
application of a time-dependent or stationary voltage to the conductor (or
a nearby gate) leads only to a redistribution of charge within n but leaves
the overall charge invariant. Therefore, a basic ingredient of a reasonable
discussion of transport of an electrical conductor is a theory of the charge
distribution with the overall constraint that the total charge is conserved.
If the total charge is conserved then the currents measured at the contacts of the sample are also conserved. As we will show, this leads to sum
rules for the admittance coefficients of a mesoscopic conductor and leads
to sum rules for the coefficients which govern the nonlinear dc-transport.
The conservation of the total charge is also connected with the fact that
an electrical conductor does not change its properties if we shift all potentials by an equal amount. The principle of gauge invariance states that the
results can depend only on voltage differences. This leads to a second set
of sum rules for the admittance coefficients and for the nonlinear transport
coefficients [2].
The three guiding principles which we have taken for the development of
a theory are thus very closely related and hinge on the localization of electric
and magnetic fields. The existence of a Gauss volume is not, however, a
trivial requirement. A different point of view is taken in a recent article
by Jackson [3]. In his circuit the potential distribution depends on the
location of the battery vis-a.-vis the resistor. Such a point of view would
make it necessary to develop a theory for the entire circuit including its
components used to drive it out of equilibrium. We believe that the notion
of localized fields is a more fruitful one and is closer to the experimental
reality as it is encountered in mesoscopic physics. In what follows, it is not
only assumed that the electric fields are localized, but that the localization
volume n is in fact of the same dimension as a phase breaking length. This
permits us to formulate a theory for the mesoscopic structure alone and to
treat the completion of the circuit as a macroscopic problem.
261
A charge and current conserving approach to mesoscopic ac-conductance

was developed by one of the authors in collaboration with Pretre and
Thomas [4]. Similar work, without an effort to achieve a self-consistent
description, was also presented by Pastawski [5] and Fu and Dudley [6].
In Ref. [4] charge and current conservation was achieved by attributing a
single self-consistent potential to a conductor. A discussion which treats
the potential landscape in a Hartree-like approach was given in Ref. [7] for
metallic screening and in Ref. [1] for long range screening appropriate for
semiconductors. These two discussions can be extended to include an effective potential as it occurs in density functional theory[8]. Both the limit
of a single potential and the case of a continuous potential are important
but for many applications the first is to crude and the later is to complex.
In this work we will treat an approach which is in between these two: In
this generalized discrete potential model the conductor is divided into as
many regions as is necessary to capture the main features of the charge and
potential distribution. This permits the description of charge distributions
which are mainly dipolar or of a more complex higher order multi-polar
form. The authors applied such an approach to the low-frequency behavior
of quantized Hall samples [9], quantum point contacts [10] and to nonlinear
transport [2].
We will only briefly review the scattering approach to electrical conductance, mainly to introduce the notation used in this article. For a more
extended discussion of some of the basic elements of the scattering approach
we refer the reader to the introductory chapter of this book or to one of
the reviews [11, 12, 13, 14]. The advantage of the scattering approach, as
formulated by Landauer [15], Imry[ll] and one of the authors [16, 17] lies
in the conceptually simple prescription of how to model an open system,
i.e. how to couple the sample to external contacts which act as reservoirs of
charge carriers and provide the source of irreversibility. The self-consistent
potential in the context of transport was already of interest to Landauer [18]
and is in fact the main issue which distinguishes his early work from preceding works on transmission through tunnel contacts [19]. A more recent
discussion of the potential distribution in ballistic mesoscopic conductors
is provided by Levinson [20] and technically is closely related to our approach [1]. For the discrete potential model discussed here, the potential
distribution can be discussed in a purely algebraic formulation which we
will present in some detail. A large portion of this article is devoted to an
explicite illustration of our approach. We discuss the low frequency admittance of wires with barriers and the low frequency admittance of quantum
point contacts. In particular, we generalize published work on the low frequency admittance of the quantum point contact to investigate the effect
of the gates used to form the quantum point contact. The last section of
262
this work presents a discussion of ac-transport and nonlinear transport over
a large frequency and voltage regime under the assumption that the only
important displacement current is that to a nearby gate.
The discussion of potentials also sheds light on the limits of validity of
the scattering approach as it is used to describe dc-transport: The scattering approach is often classified as a noninteracting theory: That is incorrect.
Since scattering states can be calculated in an effective potential the range
of validity of the scattering approach is at least the same as density functional theory. That makes it possible to include exchange and correlation
effects and makes the scattering approach a mean-field theory. As is known,
for instance from the BCS-theory of superconductivity, mean-field theories
can be very powerful tools which render any remaining deviations very hard
to detect.
The low-frequency behavior discussed here should experimentally be
accessible with much of the same techniques as those that are used for
dc-measurements. We mention here the work of Chen et al. [21] and Sommerfeld et al. [22] on the magnetic field symmetry of capacitances. We
mention further a recent experiment by Field et al. [23] who used a quantum dot capacitively coupled to a two-dimensional electron gas to measure
the density of states at the metal-insulator transition. In contrast, experiments at high-frequencies require special efforts to couple the signal to
the mesoscopic conductor and to measure the response. As examples, we
cite here the work by Pieper and Price on the admittance of an AharonovBohm ring [24], investigations of photo-assisted tunneling by Kouwenhoven
et al. [25], noise measurements at large frequencies [26], and transport in
supedattices [27]. We emphasize that already the low-frequency linear admittance presents a very interesting characterization of the sample. Indeed,
we hope, that this article demonstrates that there is considerable room for
additional theoretical and experimental work in this area.
Nonlinear dc-effects of interest are asymmetric current-voltage characteristics and rectification [28], the evolution of half-integer conductance plateaus
[29, 30], the breakdown of conductance quantization [31], and negative
differential conductance and hysteresis [32]. Like in frequency-dependent
transport most of the theoretical work shows no concern for self-consistent
effects. An exception is a numerical treatment of a tunneling barrier by
Kluksdahl et al. [32].
The emphasis on the role of the long-range Coulomb interaction in this
work should be contrasted with the recent discussions of mesoscopic transport in the framework of the Luttinger approach. The Luttinger approach
treats the short-range interactions only. However, since a one-dimensional
wire cannot screen charges completely, a more realistic treatment should
include the Coulomb effects. Without the long-range Coulomb interaction
263
the results based on Luttinger models only are not charge and current
conserving and are not gauge invariant. In this context we note that the
theory of the Coulomb blockade has been very successful in describing the
important experimental aspects. The Coulomb blockade is a consequence
of long-range and not of short-range interactions. In any case the Hartree
discussion given here is useful as a comparison with other theories which
include interactions. To distinguish non-Fermi liquid behavior from those
of a Fermi liquid it will be essential to compare experimental data with a
reasonable, i. e. self-consistent, Fermi liquid theory.
2. Theory
In this section we recall briefly the theory of ac-transport and weakly nonlinear transport which has been developed in Refs. [1,2,4, 7, 10]. A review
of these works is provided by Ref. [33].
2.1. MESOSCOPIC CONDUCTORS
Consider a conducting region which is connected via leads and contacts

a = 1, ... , N to N reservoirs of charge carriers as shown in Fig. 1a. In
order to include into the formalism the presence of nearby gates (capacitors), some parts of the sample are allowed to be macroscopically large
and disconnected from other parts. A magnetic field B may be present. It
is assumed that the leads of the conducting part are so small that carrier
motion occurs via one-dimensional subbands (channels) m = 1, ... , MO' for
each contact. Moreover, the distance between these contacts is assumed to
be short compared to the inelastic scattering length and the phase-breaking
length, such that transmission of charge carriers from one contact to another one can be considered to be elastic and phase coherent. A reservoir
is at thermodynamic equilibrium and can thus be characterized by its temperature TO' and by the electrochemical potential/1O' of the carriers. While
we assume that all reservoirs are at the same temperature T, we consider
differences of the electrochemical potentials which is usually achieved by
applying voltages Va to the reservoirs. We fix the voltage scales such that
Va == 0 corresponds to the equilibrium reference state where all electrochemical potentials of the reservoirs are equal to each other, /10' == /10, The
problem to be solved is now to find the current IO'(t) entering the sample at
contact a in response to a (generally time dependent) voltage V,o( t) at contact {3. The general time-dependent nonlinear transport problem is highly
nontrivial and we consider mainly weakly nonlinear dc-transport and linear
low- frequency transport.
264
b)
-, ,
,"-----~
,, ,
/'dn(l,x,l)
",
dE
dn(l,x,2) dn(2,x,2)
',dE
dE"
"
,,~
......
Figure 1.
.Q
'
_--_ ...
a) Multi-terminal sample including a gate; the voltages Va induce charges
(q(x)) which influence the potential (U(x)), and drive currents (If3). b) Decomposition of
the local density of states in a two-terminal sample into local partial densities of states.
2.2. THE SCATTERING APPROACH
The sample is described by a unitary scattering-matrix S [11, 15, 16, 34].

Unitarity together with micro-reversibility implies that under a reversal
of the magnetic field the scattering matrix has the symmetry ST (B) =
S( -B) [16,17]. Furthermore, the scattering matrix S for the conductor can
be arranged such that it is composed of sub-matrices scxf3(E, [U(x)]) with
elements scx(3nm(E, [U(x)]) which relate the out-going current amplitude in
channel n at contact a to the incident current amplitude in channel m at
contact (3. The scattering matrix is a function of the energy E of the carrier
and is a functional of the electric potential U(x) in the conductor [1]. For
the dc-transport properties, all the physical information which is needed,
is contained in the scattering matrix.
The electric potential U(x, {V-y}) is a function of the voltages V-y = (J.l-y J.lo)/e applied at the contacts and at the nearby gates. Thus the scattering
matrix depends not only on the energy of the scattered carriers but also on
the voltages, scx(3(E, {V-y}). With the help of the scattering matrix, we can
find the current d1cx(E) in contact a due to carriers incident at contact (3
in the energy interval (E, E + dE). It is convenient to introduce a spectral
conductance [2]
265
such that the current at energy E becomes d1a(E) = gaf3(E)(dE/e). The
unity matrix la lives in the space of the quantum channels in lead a
with thresholds below the electrochemical potential. Note that this matrix is also a (discontinuous) function of energy and of the potential. It
changes its dimension by one whenever the band bottom of a new subband passes the electrochemical potential. The current through contact
a is the sum of all spectral currents weighted by the Fermi functions
!(z) = (1 + exp(z/kBT)t I of the reservoirs [35] at temperature T
N
1a= Lj(dE/e)!(E-J-lO-eVf3)gaf3(E,{V"!}).
(2)
f3=1
Linear transport is determined by an expansion of Eq. (2) away from the

equilibrium reference state to linear order in VT The linear conductance is
found to be
(3)
where the G a f3 are evaluated at VI = ... = VN = o. Consequently, the linear
conductances depend on the equilibrium electrostatic potential u(e q )( x) ==
U(x, {O}) only. In contrast, both the ac-transport and the nonlinear transport depend explicitly on the nonequilibrium potential. A discussion of the
nonequilibrium potential requires knowledge of the local charge distribution in the conductor. Within linear response, the charge response is related
to the DOS of the conductor at the Fermi energy. The local DOS can be
obtained from the scattering matrix by a functional derivative with respect
to the local potential [1]
dn(x) = --1-LTr
dE
47ri af3
[staf3 e8U(x)
8s af3 - 8s~f3
1== Laf3 dn(a,x,(3)
e8U(x) f3
dE
Sa
. (4)
As indicated in Eq. (4), the local DOS can be understood as a sum oflocal
partial densities of states (PDOS) dn(a,x,(3)/dE [36]. The sum is over all
injecting contacts (3 and all emitting contacts a. The meaning of a local
PDOS dn( a, x, (3) / dE is then obvious: it is the local density of states associated with carriers which are scattered from contact (3 to contact a. The
local PDOS are illustrated in Fig. lb. The global DOS, dN/dE, follows by
a spatial integration of Eq. (4) over the sample region n which encloses the
mesoscopic structure with a boundary deep inside the reservoirs. Integration of the local PDOS over the whole sample leads to the global PDOS,
dNaf3 / dE. Clearly, it holds dN / dE = L-af3 dNaf3 / dE. Note that Eq. (4)
counts only scattering states. Pure bound states, e.g. trapped or pinned at
an impurity, are not included.
266
The local PDOS dn(x,(3)/dE which contains information only on the contact from which the carriers enter the conductor (irrespectively of the contact through which the carriers exit) is called the injectivity of contact
(3 and is given by dn(x,(3)/dE = L.,exdn(a,x,(3)/dE. The local PDOS
dn( a, x)/ dE which contains information only on the contact through which
carriers are leaving the sample but contains no information on the contact
through which carriers entered the conductor is called the emissivity of
contact a and is given by dn( a, x)/ dE = L.,(3 dn( a, x, (3)/ dE. Due to reciprocity, the injectivity at a magnetic field B is equal to the emissivity at
the reversed magnetic field, dnB(x,a)/dE = dn_B(a,x)/dE.
Throughout this work we use the discrete potential approximation in which
the conductor is divided into M regions Ok, k = 1, ... , M. The charge
and the potential in region k are denoted by qk and Uk. Since we are
are interested in the charge variation in response to a variation in voltage, we introduce the DOS of Ok. For later convenience, we express these
DOS in units of a capacitance by multiplication with e2 Thus, Dk =
e2 Ink dx (dn(x)/dE) is the DOS of region k. The PDOS of region k are
Dexk(3 = e2 Ink dx( dn( a, x, (3)/ dE). Obviously, the injectivities and the emissivities areDk(3 = e2 Ink dx(dn(x,(3)/dE)andD ex k = e2 Ink dx(dn(a,x)/dE),
respecti vely.
2.3. LINEAR LOW-FREQUENCY TRANSPORT
We are interested in the admittance G ex (3 ( w) which determines the Fourier

amplitudes of the current, Mex(w), at a contact a in response to an oscillating voltage b'V(3( w) exp( -iwt) at contact (3
Hex = :LGex(3(w) b'V(3
(5)
(3
To investigate the low-frequency limit, we expand the admittance in powers

of frequency
(6)
The dc-conductance G~oJ has already been derived in Eq. (3). The first order
term is determined by the emittance matrix Eex(3 == i(dG ex (3/dw)w=o. The
emittance governs the displacement currents of the mesoscopic structure.
The second order term, J(ex(3, contains information on the charge relaxation
of the system. As an example, consider for a moment a macroscopic capacitor C in series with a resistor R. For this purely capacitive structure with
vanishing dc-conductance, G = 0, the emittance is the capacitance, E = C,
267
and J( = R(q)C2 contains the RC time with the charge relaxation resistance R(q) == R. More generally, for any structure consisting of N metallic
conductors, each connected to a single reservoir, the emittance matrix is
just the capacitance matrix. These simple results must be modified for
mesoscopic conductors and conductors which connect different reservoirs.
Firstly, it is not the geometrical capacitance but rather the DOS dependent
electro-chemical capacitance which relates charges on mesoscopic conductors to voltage variations in the reservoirs [37J. Secondly, conductors which
connect different reservoirs allow a transmission of charge which leads to
inductance-like contributions to the emittance [1,4, 10J. Thirdly, the charge
relaxation resistance cannot, in general, be calculated like a dc-resistance
[37, 38J.
To illustrate our approach, we derive now a general expression for the
emittance. We first notice that for the purely capacitive case the current (displacement current) at contact Q is the time derivative of the total charge 8QO/ on the capacitor plates connected to this contact. Hence,
810/ = -iw8Q 0/ and the emittance is given by EO/(3 = fJQ 0/ / fJV(3. If transmission between different reservoirs is allowed, the charge 8Q 0/ scattered
through a contact can no longer be associated with a unique spatial region.
The charge at a given point is rather injected at different contacts and
ejected at different contacts. However, the charge emitted at a contact can
still be calculated within the scattering approach. We decompose it in a
bare and a screening part 8QO/ = 8Q~) +8Q~). The bare part of the charge
corresponds to the charge which is scattered through the contact Q for fixed
electric potential and is thus given by
8Q~) =
L DO/(38V(3
(7)
(3
where DO/(3 = I:k DO/k(3 is the global PDOS. The screening part, on the
other hand, is associated with the charge which is scattered through contact Q for a variation in the electric potential only. Since a shift of the
band bottom contributes with a negative sign and since the potential is
in general spatially varying, 8Q~s) is connected to the local PDOS by
8Q~) = - I:k DO/k8Uk, where DO/k is the emissivity of region k into contact Q. If we introduce the vector of emis sivities D~ = (DO/b .. ,DO/M) and
write the potential variation as a vector, 8U = (8Ub .... , 8UM), the charge
emitted into contact Q is
(8)
268
+BV1
t
LEAD
LEAD
1----\
....
--
u 1(x)8V1
Figure 2. Change (dashed) of the equilibrium potential (solid) in a two-terminal sample
due to a voltage variation 8V1 (dotted) in the left reservoir.
In linear response the potential variation 6U is linearly related to the potential variations 6V,e at the contacts of the sample. We write
(9)
where the response coefficients UO' = (UIO', , UMO') are called the characteristic potentials of contact Il. The emittance can be written as
(10)
To complete the calculation of the emittance, we next need a discussion of
the characteristic potentials.
2.4. PILED-UP CHARGE AND SELF-CONSISTENT POTENTIAL
The variation of the charge 6qk is the sum of a bare charge and a screening
charge. The bare charge can be expressed with the help of the injectivity
6q(b) =
L D~ 6V,e
,e
(11)
The screening charge is connected to the electrostatic potential [20] via the
Lindhard polarization function, which is here a matrix II with elements
Ilkl'
6q(S) = -IT 6U .
(12)
269
The Lindhard matrix can be expressed in terms of the scattering states.
It is in general not a diagonal matrix, i.e. the charge response is in general nonlocal. Non-local effects are, however, small quantum mechanical
effects which can be neglected deep inside a conductor with a sufficiently
large electron density. In a quasiclassical treatment the Lindhard matrix
becomes local, Ilkl = t5kIDk. Here Dk is the local DOS in region k.
The total charge, t5q = t5q(b) + t5q(S) , acts now as the source of the none quilibrium electric potential in the Poisson equation. For our discrete model
we also have to discretize the Poisson equation. We introduce a geometrical capacitance matrix C which relates the charges to the electrostatic
potentials, t5q = Ct5U. If one uses this matrix C and expresses the electric
potential in terms of the characteristic potentials, one can write the Poisson
equation in the form (C +II)u a = D~) which determines the characteristic
potentials U a ,
1 D~) .
(13)
U a = (II +
ct
Going back to Eq. (10) we find for the emittance
(14)
We notice that the bare contribution to the emittance occurs with a positive
sign and the Coulomb induced contribution occurs with a negative sign.
Depending on which contribution dominates we call the emittance element
capacitive (Eaa > 0, Ea(3 < 0 for a # f3) or inductive-like (Eaa < 0,
Ea(3 > 0 for a # f3).
The nonequilibrium charge distribution becomes
t5q
= Ct5U = L
D~) t5V(3 - IIt5U
(15)
(3
This determines the nonequilibrium steady-state to linear order in the applied voltages. The nonequilibrium charge-distribution can be written in
terms of an electrochemical capacitance matrix CjL,k(3 [2, 9, 10] which determines the net charge variation in region k in response to a potential
variation at contact f3. In vector notation, we have
(16)
where
(17)
We do not present here a general discussion of the charge relaxtion term
Ka(3. This requires a dynamical screening theory, and we have results only
270
for special cases. We discuss one of these special cases in Sect. 2.7 and Sect.
3.3.2.
2.5. WEAKLY NONLINEAR DC-TRANSPORT
The nonequilibrium potential distribution is not only of importance in actransport but also in nonlinear transport. Knowledge of the nonequilibrium
potential distribution to first order in the applied voltages permits to find
the nonlinear I-V -characteristic up to quadratic order in the voltages,
(18)
The coefficients G~oJ and G~oJ')' are obtained from an expansion of Eq. (2)
with respect to the voltages VO" One obtains for the linear conductance Eq.
(3). An expansion of Eq. (2) up to O(V2) yields G~oJ')' = (1/2) f dE (-OE!)
(2ov-ygO'(3 + eOEgO'(38(3')')' Writing ov-ygO'(3 in terms ofthe derivatives dgO'(3/ dUk
and the characteristic potentials yields
(19)
Expressing the characteristic potentials in terms of the injectivities we find
that the rectification coefficient is given by
For a quantum point contact, Eq. (20) has been discussed in Ref. [2]. In
Sect. 3.3 we will apply this result to the resonant tunneling barrier.
2.6. CHARGE CONSERVATION, GAUGE INVARIANCE, AND MAGNETIC
FIELD SYMMETRY
Since the system under consideration includes all conductors and nearby
gates, the theory must satisfy charge (and current) conservation and gauge
invariance. Due to micro-reversibility the linear response matrices must
additionally satisfy the Onsager-Casimir symmetry relations.
Let us first discuss charge conservation, which states that the total charge
in the sample remains constant under a bias. This implies also current
conservation, L0'10' = O. Imagine a volume n which encloses the entire
conductor including a portion of the reservoirs which is so large that at
the place were the surface of n intersects the reservoir all the characteristic
potentials are either zero or unity. According to the law of Gauss, one
271
concludes that the total charge remains constant. Application of a bias

voltage results only in a redistribution of the charge. If the conductor is
poor, i.e. nearly an insulator, the contacts act like plates of capacitors. In
this case long-range fields exist which run from one reservoir to the other
and from a reservoir to a portion of the conductor. But if we chose the
volume n to be large enough then all field lines stay within this volume.
Charge and current conservation imply for the response coefficients the sum
rules
LCILN'
= LG~oJ = LEa(3 = LJ(a(3 = LG~oJ'Y = 0

a
(21)
Second consider the fact that only voltage differences are physically meaningful. Gauge invariance means that measurable quantities are invariant
under a global voltage shift OV in the reservoirs, oVa f-+ oVa +OV. A global
voltage shift corresponds, of course, only to a change of the global voltage
scale. Consequently, the characteristic potentials satisfy [37]
LUa = 1
(22)
For the response coefficients the following sum rules hold
L C IL ,k(3
(3
= L G~oJ = L
(3
(3
Ea(3
=L
(3
J(a(3
= L(G~oJ'Y + G~o~(3) = 0
. (23)
(3
Note that due to charge conservation and gauge invariance, the geometrical
capacitance matrix C has the zero mode 1 = (1,1, .... 1), corresponding to
a unit potential in each region. Thus C cannot be inverted. However, the
Green's function (II + C)-l which solves the Poisson equation exists (see
Eq. (13)). Expressing the characteristic potentials with the help of the
Green's function and the injectivities gives
(24)
where D = (Db D 2 , ... , DM) is the vector of the local DOS. Equation (24)
is the key property of the Green's function needed to show that our final
results are charge and current conserving. From Eq. (24) it follows immediately that the sum over the elements of a row of the Lindhard matrix is
equal to the DOS in region k, i.e. Dk = LI Ilkl. Since the Lindhard matrix
is symmetric, it holds also Dk = LI Illk. The above mentioned statement,
that in the quasiclassicallocal case Ilkl = DkOkl holds, is now clear.
In the presence of a magnetic field, the admittance matrix must additionally
satisfy the reciprocity relations
(25)
272
which is a consequence of time-reversal symmetry of the microscopic equations and the fact that we consider transmission from one reservoir to another [1, 16, 17]. These Onsager-Casimir relations are only valid for the
linear response coefficients, Le. close to equilibrium. Of course, the symmetry relations (25) hold individually for G~oJ, Eo:(3, and 1(0:(3' Note that the
emittance is symmetric, Eo:(3 = E(3o:, in the purely capacitive case.
2.7. FREQUENCY-DEPENDENT TRANSPORT: SINGLE POTENTIAL
APPROXIMATION
In this section we show that it is possible to discuss the full frequencydependence of the admittance, if the electric potential in the mesoscopic
conductor can be approximated by a single variable. The external response
is defined as the response for fixed electrostatic potential (Le. the 'bare'
response). A general result for the external response has been derived in
Ref. [4]. If the differences of wave vectors k(E) and k(E + 1iw) deep in the
reservoirs are neglected, the external response can be expressed in terms of
the scattering matrix only,
0:(3 (w)=~JdETr[18
h
0: 0:(3 _st0:(3 (E)s 0:(3 (E+1iw)](J(E)-f(E+1iw))
1iw'
Ge
(26)
The scattering matrices and the Fermi functions are here evaluated at the
equilibrium reference state. The real part of the ac-conductance, (26) is
related via the fluctuation-dissipation theorem to the current-current fluctuation spectra derived in Ref. [39]. An expansion of Eq. (26) to linear order
in w gives [4]
where the dN0:(3/ dE are the global PDOS, in which the functional derivative
with respect to the local potential (and the integration over the volume n)
is replaced by a (negative) energy derivative. Asymptotically, for a large volume n the global PDOS obtained from an energy derivative is identical to
the integral of the local PDOS associated with a functional derivative with
respect to the potential according to Eq. (4). However, for a finite volume,
there are typically small differences of the order of the Fermi wavelength
divided by the size of the volume n. For a more detailed discussion, we
refer the reader to Ref. [36].
In order to obtain the full admittance, we have still to find the internal
response (the 'screening' part of the admittance). A general result for the
internal response is not known. In the simple case, where the conductor
is treated with a single discrete region (M = 1), the external response
273
defines, however, also the internal response. We assume that the sample
is in close proximity to a gate which couples capacitively to the conductor with a capacitance coefficient C. The current response at contact O! is
Ma(w) = L G~~(w)8V~+G~(w)8Ul where G~(w) is the (unknown) internal
response of the conductor generated by the oscillating electrostatic potential 8Ul . The current induced into the gate is Mg(w) = -iwC(8Vg - 8Ul ).
Now we can determine G~(w) from the requirement of gauge invariance. In
particular, if all potentials are shifted by -8Ul , it follows immediately that
G~(w) = - L~G~~(W). Using current conservation, LaMa + Mg = 0, we
find for the conductances [4J of the interacting system (O!, (3, 1,8 =I g),
G;~(w)
G!~(w)
G~~(w)
(i/wC) L-r G~"((w) Lo G8~(w)

,
1 + (i/wC) L-ro G~o(w)
LoG8~(w)
1 + (i/wC) L"(o G~o(w)'
(28)
(29)
G;g(w)
L"( G~"((w)
1 + (i/wC) L-ro G~o(w)'
(30)
G!g(w) =
L"(o G~o(w)
1 + (i/wC) L-ro G~o(w)
(31 )
The single-potential approximation provides a reasonable description only if

the conductor has a well-defined interior region, which might be described
by a single uniform potential. Furthermore, it is assumed that the only
relevant capacitance is that of the sample and the gate. Examples for which
this approach is reasonable are quantum wells or quantum dots or cavities
for which the Coulomb interaction is so weak that single electron effects can
be neglected. In Section 3.3 we discuss the case of a double barrier with a
well which is capacitively coupled to a gate.
3. Examples
In a first example we discuss the emittance of a wire which contains a single

barrier. The wire is capacitively coupled to a gate. As a second example we
discuss the capacitance and emittance of a quantum point contact formed
with the help of gates [10J. It turns out that steps in the capacitance and
the emittance occur in synchronism with the well-known conductance steps.
As a third example, we treat the resonant tunneling barrier close to a resonance, for which we calculate the nonlinear current-voltage characteristic
and the frequency-dependent conductance [2J with the single-potential approximation. For all examples, we use explicit expressions for the Lindard
function (matrix) II given within the local Thomas-Fermi approximation,
274
~1
gate
Figure 3.
Quantum wire containing an impurity and coupled to a gate.
where II is diagonal with elements determined by the local DOS. In all

examples we consider finally the zero-temperature limit.
3.1. ADMITTANCE OF A WIRE WITH A BARRIER
Our first example is a mesoscopic wire which connects two contacts and is
capacitively coupled to a nearby gate (see Fig. 3). The wire is assumed to
be perfect with a uniform potential along its main direction, except for a
single scatterer, a barrier, or an impurity. The discussion is carried out in
the semiclassical limit where Friedel oscillations are neglected. We take the
potential of the impurity to be of short-range and assume that the potential
to left and right of the scatter can be assumed to be uniform. We assume
here that the gate couples to the wire only and neglect scattering at the
interface of the wire and the reservoir. Furthermore, to simplify the discussion we assume in this section that the capacitance between the charges to
the left and the right of the barrier can be neglected (Le. vanishingly small
C12 ). Consider first a perfect wire (T = 1) of length L. The gate couples
with a geometrical capacitance c per length. The DOS per length for a
one-dimensional wire is
dn/dE
= 2/hv
(32)
where v = (2/m)1/2(E - eU)1/2 is the velocity of carriers with energy E. If

the wire is filled up to the electrochemical potential J-L of the reservoirs the
total number n of states per length is
n(J-L)
= rp, dE(dn/dE) = e- 1(2/7r 2aB)1/2(J-L leU
eU)1/2.
(33)
275
Here we have introduced the Bohr radius aB = h 2 /me 2 . To determine this

density self-consistently we consider the Poisson equation. The difference
of the electron charge density en and the ionic background charge density
-en+ is equal to the capacitive charge,
en(p,) - en+ = e(U - Ug) ,
(34)
where Ug is the electrostatic potential of the gate. Similarly, near its surface,
the deviation of the electronic gate charge away from charge neutrality is
equal to -e( U - Ug ). Now we take the gate to be a macroscopic conductor.
Therefore, the electrostatic potential and the electrochemical potential are
everywhere related by p,g = EFg +eUg, where EFg is the chemical potential
(Fermi energy) of the gate. We eliminate the gate potential Ug in Eq. (34)
and find
en(p,) = (e/e)(e 2n+ /e + EFg - p,g
+ eU)
(35)
Measuring energies from the band bottom of the one-dimensional wire and
setting eU = 0 gives the electrochemical potential as a function of the gate
potential,
(36)
In the limit e
---+ 0 the electrochemical potential of the wire is simply determined by charge neutrality, p, = (7r 2aB/2)e 2 (n+)2. For p,g = P,o ==
e2n+ / e + EFg the electrochemical potential vanishes and the wire is empty
for p,g ~ P,o. The DOS of left moving carriers, dnz/dE = l/hv, is directly related to the phase accumulated by the electrons traversing the wire,
Ldnz/dE = L/hv = (l/7r)d/dE. The phase accumulated by a traversing
electron (at the Fermi energy) is thus given by
(37)
The scattering matrix element for the transmitting channel is given by
821 = exp( i) which shows that the scattering matrix and thermodynamics
are intimately related. At the Fermi energy of the wire, the DOS is given
by
(38)
The true electron density in a narrow energy interval at the Fermi energy
is found by considering a small variation of the electrochemical potentials
bp, and bp,g and the electric potential bU away from a reference state (p,(e q ) ,
p,1eq ), u(e q )) which obeys Eq. (35). Variation of this equation gives
bq
= e(dn/dE)(bp,- ebU) = e(bU -
bp,g/e).
(39)
276
Solving for 6U gives for the charge 6q = cg(6f-l/e - 6f-lg/e) with an electrochemical wire-to-gate capacitance
(40)
Here the DOS appears in series with the geometrical capacitance It depends
on the reference state and is given by Eq. (38) with f-lg = f-l~q.
Next, we discuss a wire containing a symmetric barrier. At equilibrium,
and if we can treat the potential created by the barrier as short-range, the
DOS to the left and right of the barrier remains unchanged. Hence the
consideration given above for the electrostatic equilibrium potential also
characterizes the wire to the left and right of the barrier. Similarly, for a long
wire the gate-to-wire capacitance is essentially unaffected by the presence
of the barrier. In the case of transport, on the other hand, the chemical
potentials of the reservoirs connected to the wire differ by a small amount
implying a variation of the charge distribution. We introduce two regions 0 1
and O2 of size L/2 to the left and the right ofthe barrier, respectively. With
the help of Fig. 1b, the semiclassical local PDOS, D a k(3, can be constructed
with simple arguments. For example, D211 is given by the transmission
probability times the DOS of 0 1 associated with carriers with positive
velocity, hence D211 = T DI/2, with D1 = D2 = D /2 due to symmetry.
To determine D 212 , we note that in the semiclassical limit considered here,
there are no states in 0 1 associated with scattering from contact 2 back to
contact 2, hence it holds D212 = o. With similar arguments one finds for
the semi-classical PDOS
Form Eq. (41) we obtain for the emissivities and injectivities DI1 = D~l =
(1/4)(1 + R)D and DI2 = Di2 = (1/4)T D. The integrated geometrical capacitance is denoted by C = cL. With these quantities the Poisson equation
reads
(D /2)(1 + R)6V1 + (T D /2)6V2 - D6U1

(TD/2)6V1 + (1 + R)(D/2)6V2 - D6U2
C(6U1 - 6U3 )
C(6U2 - 6U3 )
(42)
These equations can also be derived by noticing that the variations 6Fk
of the quasi-Fermi levels [34] in the regions k = 1 and 2 are given by
6F1 = ((1+R)6V1+T6V2)/2 and 6F2 = ((1+R)6V2+T6V1)/2, respectively.
For a local Lindhard function, the charge in region k is then 6qk = Dk(6Fk-
6Uk) = (C /2)(6Uk - 6U3 ).

For the charge neutral case, C = 0, Eqs. (42) are familiar from Landauer's
discussion of the potential drop across a barrier [15, 18, 34]. We assume
277
again that the gate is a massive conductor for which we have /-l3 = EFg+eU3
everywhere. We use Eqs. (42) to evaluate the characteristic potentials and
get
Un
1 (1 + R)D
C +D '
=2
1 TD
U12
= 2C + D'
U13
= C +D
(43)
Due to symmetry, it holds U22 = Un, U21 = U12 and U23 = U13. Defining
Ll V == 8V1 - 8V2 , the potential drop across the barrier can be written in
the form
(44)
where Cg = Leg is the electro-chemical gate-to-wire capacitance given by
Eq. (40). The voltage difference LlU is proportional to the reflection probability R and the chemical potential difference of the reservoirs. The total
charges to the left or to the right of the barrier are the sum of the injected
charges and the induced charges
(45)
8qk = 2)Di/3 - Dk U k/3)8V/3

/3
We find, e.g., for 8ql,

(46)
The difference in the charge density on the left and the right hand side
becomes
( 47)
In the limit of vanishing gate-to-wire capacitance, the charge difference

vanishes also. Away from the barrier the wire is charge neutral. In this
charge-neutral limit the potential difference is determined by the reflection
coefficient only, 8U = RLl V.
From Eq. (46) we find the electrochemical capacitance coefficients C j.t,kcx.
We write this capacitance matrix in the form
-C;
,
2C
-C )
(48)
where
Cj.t = (1/2)(1 + R)Cg
(49)
278
TABLE 1. Gate capacitance and capacitance, emittance, polarization,
and voltage drop of the quantum point contact with a grounded gate for
limiting cases
C-+O
I Cg
I CI'
I Ell
I !:::.q/!:::.V
I !:::,U/!:::.V
II
I
II
I
I
Co = canst.
C-+oo
C = canst.
D/2
R/(C;l
Co -+ 0
l/((D /2)-1
+ D- 1 ) I (D/4)(1 + R) I
RCI' - DT2/4
2CI'
CI'/Co
I
I
I
RD/2
RD/2
0
(1
+ C- 1 )
+ R)Cg/2
I (Cg/C)(RC - DT2/4) I
RCg
I
I
RCg/C
I
I
It follows CJ.L,12 = (1/2)TCg. Using the characteristic potentials and the

PDOS given by Eq. (41) we can calculate the emittance,
E22 = (Cg/C)(RC - DT2/4)

E21 = (Cg/C)(TC + DT2/4)
E31 = E23 = E32 = -E33/2 = -Cg
(50)
If the transmission probability vanishes, the emittance matrix is purely ca-
pacitive. In the case where the geometric gate capacitance C tends to zero,
the electrochemical gate capacitance and the electrochemical capacitance
across the barrier vanish Cg = CJ.L = 0 but the ratio Cg/C tends to 1. The
wire is then charge neutral and the charge imbalance t:lq vanishes. In the
single-channel case considered here, the voltage drop is t:lU = Rt:l V. In
this case the wire responds inductively for all values of the transmission
probability E = - T2 D /4. In the absence of a barrier (R = 0) the wire has
also inductive emittances E = Ell = En = -E12 = -E21 = -D /4. The
results of this section are summarized in the rightmost column of table 1.
3.2. EMITTANCE OF A QUANTUM POINT CONTACT IN THE
PRESENCE OF GATES
A quantum point contact (QPC) is a small constriction in a two-dimensional

electron gas which allows the transmission of only a few conducting channels [40,41]. We consider a symmetric QPC with two'gates as shown in Fig.
4a. and ask for its capacitance and low-frequency admittance. To discretize
the QPC, we define to the left and to the right of the constriction two
regions f21 and f22 with sizes of the order of the screening length. If only a
279
b)
a)
:---- r
--"'\..
~
3
0
'iN,
-3
-6
0
10
eUo
(meV)
15
Figure 4. a) Quantum point contact with gates. The regions 0 1 ,2 at the left and right
of the point contact can be charged. b) Capacitance CiJ and emittance Ell of the QPC.
Capacitances and emittances are in units of femto farads.
few channels are open, the equilibrium potential of the constriction has the
form of a saddle [42]. Away from the saddle towards the two-dimensional
electron gas, the potential has the form of a valley which rapidly deepens
and widens. In contrast to the previous example, nl and n2 are now regions immediatley to the left and right of the saddle, where the equilibrium
potential and the equilibrium density are not uniform.
Consider a symmetric QPC. The two gates are taken to be at the same
voltage V3. Thus in effect, the two gates act like a single gate. As in the
previous section, the gate will be treated as a macroscopic conductor. For
the QPC contact in the presence of the gates the charge distribution is
not dipolar. It is the sum of the charges in nl , n2 and at the gates which
vanishes, 8ql + 8q2 + 8q3 = O. Assume a single open channel. Due to the
symmetry of the problem, the geometric capacitance matrix can be written
in the form
C= (
Co + C -Co -C)
-Co Co + C -C
-C
-C
2C
(51)
where Co is the geometric capacitance between 1 and 2 and where C is

the geometric capacitance between these regions and the gates. The electrochemical capacitance has again the form (48), with CJ1, equal to the
electrochemical capacitance across the QPC and with Cg being the electro-
280
chemical gate capacitance. With the help of Eq. (17) and the characteristic
potentials Uk{3 = (Dk{3 - CIl-,k(3)/ Dk one finds
1
C-l
Gil- =
+ (D /2)-1
RCo + (C /2)(1 + R) + 2CoCg D- 1

1 + 2(2Co + C)D-l
(52)
(53)
The emittance elements (14) become
E21 = Cg - En
E31 = E23 = E32 = -E33/2 = -Cg
(54)
with D = 2D 1 The charge difference !1q and the electrostatic voltage drop
!1U = 8U1 - 8U2 across the QPC are, respectively, given by
!1q
!1V
!1U
!1V
DR(2Co + C)
D + 2(2Co + C) ,
DR
D + 2(2Co + C)
(55)
(56)
The discussion in the previous section is a limiting case of the general results given here. In the limit Co --t 0 we obtain the results discussed for
the quantum wire with a barrier collected in column three of table 1. In
the limit C --t 0 overall charging of the QPC is prohibitive. The charge
distribution is dipolar and the results of column 1 of table 1 apply. In the
limit C --t 00 (corresponding to a geometrical capacitance of the QPC to
the gates which is much larger than the capacitance across the QPC) an
electrostatic voltage difference across the QPC does not develop, even so a
charge imbalance exists. The results for this limiting case are collected in
the middle column of table 1.
To generalize these results to the many channel case, we mention that for
the potential considered the contributions of each individual channel just
add. Each occupied subband j contributes with a transmission probability T(j) to the total transmission function T (= T11 ) = Lj T(j). The dcconductance (3) is then given by Gi~) = G~~ = -Gi~ = -G~~) = (2e 2 / h)T.
Clearly, G~oJ vanishes whenever one of the indices corresponds to the gate
index 3. Similarly the local PDOS (41) D~) of each individual channel simply add. Hence, the local DOS becomes Dk = Lj D~), and T = 1- R is now
an average transmission probability defined by T == Tk = D;1
Lj D~)T(j)
281
where (k = 1,2). Note that the average transmission probability T (1o T)

has nothing to do with the dc-conductance. If only a few channels are open
the potential of the QPC has the from of a saddle point [42]. As a specifi.c example we consider a quadratic potential U(x) = UO(b 2 - x 2 )/b 2 if
Ixl ~ b, and U(x) = 0 if b < Ixl ~ l. The transmission probabilities T(i) and
the PDOS can be calculated analytically from a semiclassical analysis [43].
For simplicity, we assume a constant electrostatic capacitance Co = 1 f F
between n1 and n2 and a fixed number of occupied channels in these regions. Both the height Uo and the spatial variation of the potential should
in principle be found by minimizing the grandcanonical potential for the
system [44]. Here we consider Uo as an independent control parameter. We
assume that no additional channels enter into the regions nk during the
variation of Uo. In Fig. 4b. we show the results for a constriction with
b = 500 nm, l = 550 nm, and with three equidistant channels separated by
EF/3 = 7/3 meV. The dotted curve represents the transmission function
which determines the dc conductance. At each step a channel is closed.
The dashed and solid curves correspond to a gate with coupling C = 0
and C = Co, respectively. The curves represent the capacitance CJ.I and
the emittance En. For the two-terminal QPC (C = 0) the capacitance
vanishes and the emittance is negative for small Uo where all channels are
open. At each conductance step, the capacitance and the emittance increase
and eventually merge when all channels are closed. Due to a weak logarithmic divergence of the WKB density of states at particle energies E = eUo
(where WKB is not appropriate), the emittance shows steep edges between
the steps. In the presence of the gate (C # 0), the curves are shifted upwards due to a capacitive contribution of the gate.
3.3. THE RESONANT TUNNELING BARRIER
In the framework of the single-potential approximation introduced in Sect.

2.7 we discuss now the fully nonlinear I-V characteristic and the fully
frequency-dependent admittance of a resonant tunneling barrier (RTB)
with a single resonant level. For a comparison of the results of our Hartreelike discussion (which to be realistic needs to be extended to a many level,
many channel RTB) with theories that treat single electron effects, we
refer the reader to the works by Bruder and Scholler [45], Hettler and
Scholler [46], Stafford and Win green [47] and Stafford [48]. Consider the
RTB sketched in Fig. 5a, where two reservoirs are coupled by a double barrier structure. The well between the barriers is considered to be the single
relevant region. Additionally, the well may couple electrically with a capacitance C to an external gate. We label quantities associated with the well
with an index O. The band bottom of the well between the barriers and the
282
energy of the resonant level are denoted by eUo and by Eo = eUo + E r ,
respectively. The electrochemical potentials of the particles which are scattered at the RTB are assumed to be close to the resonant level, such that
the energy dependence of the scattering properties is described by a BreitWigner resonance with a width r. The asymmetry of the barrier is denoted
by .6.r = r 1 - r 2, where rOt /h is the escape rate of a particle trapped in
the well through the barrier a . The scattering matrix elements have then
a pole associated with a denominator .6.(E) = E - Eo + ir /2,
s (J = ei('P,,+'P,B)
Ot
(0 Ot
(J -
~)
.6.(E)
(57)
Here, the <POt are arbitrary phases. We assume that the only energy dependence occurs in the resonant denominator, while the rOt and the <POt are
energy independent. One finds a transmission probability
(58)
with a maximum value Tmax = 4r1r 2/r2. The injectivities and emissivities
are equal to each other due to the absence of a magnetic field and are
(59)
Summing up injectivities (or emissivities) yields the total DOS in the well
(60)
Again, the (P)DOS are here represented in units of a capacitance.
3.3.1. Nonlinear transport

Let us discuss the nonlinear J- V characteristic [2] for the charge neutral
case (C = 0). Note that only the effect of the resonant level is considered
and that transport is still phase coherent. First, we determine the selfconsistent potential. The relation between the electrostatic potential shift
.6.Uo == Uo - u~eq) and the voltage shifts Vot = (/-LOt - /-L(eq)/e is obtained
from the charge-neutrality condition
J.Ll
-00
DOl(E, Uo) dE +
1M
-00
D02(E, Uo) dE -
1J.L(e
-00
(eq)
Do(E, Uo ) dE == 0 ,
(61)
283
b)
a)
III
0.5
~
E,
'\
"\-
0.0
.~
AU. Ie
-15 -1 0 -5
10
15
et'N/ r
Figure 5.
a) Resonant tunneling barrier with a single level and biased by voltages
= (ILl - IL(eq/e and V2 = O. The well may be coupled capacitively with a capacitance
C to a gate (dark region). b) Nonlinear current-voltage characteristic with r l = 0.5 meV,
r 2 = 1 meV, t:.E = 0 (solid), t:.E = -r (dashed), t:.E = -2r (dotted). Thin dotted
lines indicate the quadratic approximation .
YJ.
which can be integrated analytically. The result can be written as a function

which determines implicitly W == ,6.Uo - (VI + V2 )/2 as a function of ,6.V
only,
r 1 arctan [,6.E - e(W/ - ,6.V/2)]

r 2
,6.E - e(W + ,6.V/2)]
+ r 2 arctan [
r/2
Equation (2) yields for the current I == II
I
= - Iz
2e r 1 r 2 (
[,6.E - e(W h-r- arctan
r /2
arctan
[,6.E - e(W + ,6.V/2)])
r/2
(62)
,6. V /2)]
'
(63)
where ,6.E = p(e q ) - (eUJe q ) + Er) is the equilibrium distance between

the Fermi energy and the resonance [2]. Without taking into account the
self-consistent shift ,6.Uo one would get a wrong result which is not gauge
invariant. The current given by Eq. (63) saturates at a maximum value
284
proportional to 11" /2 - arctan(2~E /r). The conduction is optimal for ~E =

o and r 1 = r 2 when I = (e/h)rarctan(e~V/r). In Fig. 5b we have plotted
the characteristic for an asymmetry ~r /r = -1/3 and for various values of
~E. Due to the complete screening, the resonant level fioates up or down
to keep the charge in the well constant. An expansion of the current yields
[2] Gi~~ = -( e3 / h)( ~r /2r)8ET (thin dotted lines in Fig. 5b). The case of
incomplete screening can similarly be treated with our approach. At large
voltages, the resonance can then eventually fall below the conductance band
bottom of the injecting reservoir as is known from semiconductor doublebarrier structures. Finally we mention that, in general, even an elastically
symmetric resonance can be rectifying if screening is asymmetric.
3.3.2. AC-Response
In order to calculate the admittances (28) we must know the external response G~(J defined for fixed potential. Using the specific scattering matrix
elements (57), Eq. (26) gives
G~(J(w) =
I dE f(E)-{E+nw)
(8 a(J -
s~(J(E)sa(J(E + nw))
= aa(J 9I( w)
with
aa(J(w) - f(1iw
+ if)
( r1(nw + ir 2 )
-iflfz
(64 )
-ir 1 r 2
fz(1iw
+ if l )
(65)
and where
I(W) = e 2 irjdE f (E)-f(E+nw) (

1
_
1 ) . (66)
9
h 4
nw
~(E + nw) ~*(E)
Here, we used
1
~(E + nw)~*(E) = nw + if
(1
~(E + nw) - ~*(E)
(67)
It turns out that gI (w) is the ac-admittance for the symmetric barrier
without gate (C = 0). This is is readily verified since for C = 0 and for
all = a22 it holds
In the case of C = 0 it is thus sufficient to discuss the symmetric barrier.

We consider first this case and assume zero temperature. The integral over
the difference of the Fermi functions reduces then to a simple integral from
285
j1(e q) _
nw to
j1(e q)
and can be carried out analytically. We find for the real

part and the imaginary part of the admittance
e2
Re[gI(w)]
Im[l(w)]
f (
~E + nw
~E - nw )
4nw arctan( (f/2) ) - arctan( (f/2) ) (69)
e 2 ~ In
h 4nw
(1~12 - (nw)2)2 + (fnw)2

1~12
(70)
where I~I == 1~(j1(eq))I. Our results for gI(w) are equivalent to earlier results
of Fu and Dudley [6]. However, these authors neglect Coulomb interaction
and, coincidentally, for a symmetric RTB, obtain the right expression only
since they apply an antisymmetric bias ~ V /2 at the contacts.
An expansion of Re[G(w)] and Im[G(w)] with respect to frequency yields
the coefficients defined in Eq. (6)
e2
hT(EF) ,
G(O)
E
J(
max
Do
4
-r' /3 )
C~I'1~12r'/2)
he'T( EF) C~I'

(27l"1~1)2
1~12
(71)
(72)
(73)
We see, that E and J( change sign. In particular, for the symmetric case
this occurs if the value of the transmission probability T equals 1/2 and
3/4, respectively. In Fig. 6a we plotted the real part and the imaginary part
ofthe normalized admittance gI(w)/gI(O) as a function of nw/I~I for three
different cases T = 1 (solid), T = 0.5625 (dashed), and T = 0.25 (dotted).

For T < 3/4 there is a peak in the conductance which belongs to the
excitation of the resonance by an energy quantum nw ~ I~I. We mention
that the normalized admittance is determined by a single parameter I =
f/I~I
We discuss now the effect of the capacitance C. In principle, by using
the Eqs. (64) it is possible to derive expressions for the conductances (28).
Here, we do not display the lengthy formulas but we rather present the
results in a figure. A comparison of the admittance for C = 0 (thin dotted
curves) and for finite C (thick curves) is shown in Fig. 6b for the resonant
case T = 1. One sees that for the small-capacitance case shown in the figure
the Coulomb-coupling to the external gate enhances the capacitive part of
the admittance. Indeed, an expansion with respect to the capacitance yields
(74)
286
a)
2.0 , - - - - - - - - - - - ,
1.5
- - '1-1
---- '1-1.5
- - '1-2
S 1.0
is
0.0 H - - - - - - ----1
0.0
10
lS
--
\ Im(G 12)
\."~~:.....-- -- -- - ---
.(l.S
, //
-1 .0
nrol~1
Figure 6.
1.0
0.5
0.5
b)
Re(G Il)
2
tTroI~1
Frequency dependent admittance of the symmetric RTB. a) charge neutral

weak gate: G l l (solid) and G 12 (dashed) for
(C = O~ case, "y = r/161; b) Influence of a

C161/e = 0.01, T = 1 (dotted: C = 0).
In particular, one concludes La{3 Ga{3 = -iwC + O( C 2 ) which follows also

directly from the fact that the admittance La{3 G~{3 can be understood as
a resistor (with an external admittance La{3 G~{3) in series with a capacitor
(with a capacitance C).
The characteristic time-scales of tunneling problems[49] are a subject
of considerable interest. With the analysis given above we are now able to
identify characteristic times for the electric problem. We remark that the
electrical problem investigated here leads to answers not for single electron
motion, but for the collective charge dynamics. Consider the case of a RTB
in the zero capacitance limit. From the expansion of the conductance up to
quadratic order in frequency we can identify two characteristic frequencies
or time-scales. There exists a frequency WI at which the magnitude of the
dc-current and the displacement current are equal. The frequency WI is
thus determined by 71 = lEI/GO, where E is the emittance and GO the
dc-conductance. The time-scale 71 is thus a generalization of the RC-time.
For the RTB we find that this time is equal to 71 = h /r at resoance
and is also equal to 71 = h/r far away from resonance. It vanishes for a
Fermi energy at which the transmission of a symmetric barrier is equal
to 1/2. A second characteristic frequency is obtained by comparing the
287
displacement current determined by the emittance E with the second order

dissipative term K. The second characteristic time 72 is thus determined
by 72 = IKI/IEI. At resonance this time is given by 72 = (l/3){li/r). This
time tends to zero for Fermi energies far away from resonance. It is singular
at the Fermi energies at which the emittance vanishes and it is zero at the
Fermi energies for which K is zero. We conclude by mentioning that for the
case in which the capacitance is not zero, the above consideration has to
be applied to each element of the admittance matrix separately. In analogy
to the tunneling time problem, where there exist characteristic times for
traversal and reflection, from left and from right, the electrical problem will
then be characterized by a set of characteristic frequencies or time scales
for each admittance element, related only by sum rules due to current
conservation and gauge invariance.
4. Conclusion
In this work we presented a theory for the admittance and the nonlinear
transport of open (i.e. connected to different reservoirs) mesoscopic conductors. We have emphasized the application of this theory to a number of
systems of current interest, like quantum wires, quantum point contacts,
and resonant tunneling barriers. Our emphasis has been to derive results,
which even so they might not be realistic in detail, nevertheless capture the
essential physics. Our results should be useful both for comparison with
additional theoretical work and with experimental work.
Due to the limitations of space, we have not reviewed a number of
closely related subjects. The approach discussed here has been applied to
ac-transport in two-dimensional electron gases in high magnetic fields [9,38]
which is particularly interesting in the regime where transport is dominated
by carrier motion along edge states and by charging and de-charging of edge
states. Of fundamental interest are Aharonov-Bohm effects in capacitance
coefficients if one of the capacitor plates has the form of a ring [50] or for
rings between capacitor plates [51]. As in dc-transport these interference
effects are closely related to the sample specific fluctuations away from
an average behavior. For a chaotic cavity capacitance fluctuations away
from an ensemble averaged capacitance have been investigated by Gopar
and Mello and one of the authors [52]. The ac-response and its fluctuations of a chaotic cavity connected to two leads and coupled capacitively
to a back gate has been investigated by Brouwer and one of the authors.
This system is particularly interesting since the ensemble averaged quantities exhibit weak localization corrections [53]. Since the weak-localization
effect is not associated with a net charge accumulation the ac-response
exhibits in addition to a Coulomb charge relaxation pole also a pole for un-
288
charged excitations. Scholler [54] has investigated the dynamic capacitance

of a one-dimensional wire. The dynamic capacitance exhibits interesting
structure due to the plasma-modes of the one-dimensional wire. This list
illustrates that there are many avenues to extend the work presented here.
The application of the theory to more realistic models is a quite challenging
undertaking but very likely also full of rewards.
ACKNOWLEDGMENTS
This work was supported by the Swiss National Science Foundation under
grant Nr. 43966.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
Biittiker, M., (1993) J. Phys.: Condens. Matter 5, 9631.

Christen, T., and Biittiker, M., (1996) Europhys. Lett. 35, 523.
Jackson, J. D, (1996) Am. J. of Physics 64, 855.
Biittiker, M., Thomas, H., and Pretre, A., (1993) Phys. Rev. Lett. 70, 4114; Pretre,
A., Thomas, H., and Biittiker, M., (1996) Phys. Rev. B, Oct.
Pastawski, H., (1992) Phys. Rev. B46, 4053.
Fu, Y. and Dudley, S. C., (1993) Phys. Rev. Lett. 71, 466.
Biittiker, M., Thomas, H., and Pretre, A., (1994) Z. Phys. B 94, 133.
Biittiker, M., (1995) in "Quantum Dynamics and Submicron Structures", edited by
H. Cerdeira, G. Schon, and B. Kramer, (Kluwer Academic Publishers, Dordrecht)
p. 657 -672.
Christen, T., and Biittiker, M., (1996) Phys. Rev. B 53, 2064.
Christen, T., and Biittiker, M., (1996) Phys. Rev. Lett. 76, 143.
Imry, Y., (1986) in Directions in Condensed Matter Physics, edited by G. Grinstein
and G. Mazenko, (World Scientific Singapore) p. 101.
Beenakker, C. W. J., and van Houten, H., (1991) Quantum transport in semiconductor nanostructures, edited by. H. Ehrenreich and D. Turnbull (New York Academic
Press).
Buot, F. A., Phys. Rep. (1993) 234, 73.
Datta, S., (1993) Electronic Transport in Mesoscopic Conductors, Cambridge University Press, 1995; Buot, F. A., Phys. Rep. 234, 73.
Landauer, R., (1970) Philos. Mag. 21, 863.
Biittiker, M., (1986a) Phys. Rev. Lett. 57, 1761.
Biittiker, M., (1988b) IBM J. Res. Develop. 32,317.
Landauer, R., (1957) IBM J. Res. Develop. 1, 223.
Frenkel, J., (1930) Phys. Rev. 36, 1604.
Levinson, I. B., (1989) Sov. Phys. JETP 68, 1257.
Chen, W., Smith, T. P., Biittiker, M., and Shayegan, M., (1994) Phys. Rev. Lett.
73, 146.
Sommerfeld, P. K. H., van der Heijden, R. W., and Peeters, F. M., (1996) Phys.
Rev. B 53, 13250.
Field, M., et aI., (1996) Phys. Rev. Lett. 77, 350.
Pieper, J. B. and Price, J. C., (1994) Phys. Rev. Lett. 72, 3586.
Kouwenhoven, L. P., et aI., (1994) Phys. Rev. Lett. 73, 3443.
Reznikov, M., Heiblum, M., Shtrikman, H., and MahaIu, D., (1995) Phys. Rev. Lett.
75, 3340.
289
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
Hofbeck, K, Genzer, J., Schomburg, E., Ignatov, A. A., Renk, K F., Pavel'ev,
D. G., Koschurinov, Yu., Melzer, B., Ivanov, S. Schaposchnikov, S., Kop'ev, P. S.,
(1996) Phys. Lett. A218, 349.
Taboryski, R., et al., (1994) Phys. Rev. B 49, 7813.
Glazman, 1. I., and Khaetskii, A. V., (1989) Europhys. Lett. 9, 263.
Patel, N. K, et al., (1990) J. Phys.: Condens. Matter 2, 7247.
Kouwenhoven, L. P., et al., (1989) Phys. Rev. B 39, 8040.
Kluksdahl, N. C., et al., (1989) Phys. Rev. B 39, 7720.
Biittiker, M., and Christen, T., (1996) in Quantum Transport in Semiconductor Submicron Structures, edited by B. Kramer, (Kluwer Academic Publishers,Dordrecht);
NATO ASI Series, Vol. 326, 263 -291.
Biittiker, M., Imry, Y., Landauer R., and Pinhas, S., (1985) Phys. Rev. B 31, 6207.
Biittiker, M., (1992b) Phys. Rev. B 46, 12485.
Gasparian, V. M., Christen, T., and Biittiker, M., (1996) Phys. Rev. A Oct.
Biittiker, M., Pretre, A., and Thomas, H., (1993) Phys. Lett. A 180, 364.
Biittiker, M., and Christen, T., (1996) Dynamic Conductance in Quantum Hall
Systems, to appear in 'The application of high magnetic fields in semiconductor
physics', edited by G. Landwehr, (unpublished). cond-mat/9607051
Biittiker, M., (1992) Phys. Rev. B 45, 3807.
van Wees, B. J., et al., (1988) Phys. Rev. Lett. 60, 848.
Wharam, D. A., et al., (1988) J. Phys. C: Solid State Phys. 21, L209.
Miller, S. C., and Good, R. M., (1953) Phys. Rev. 91,174.
Chklovskii, D. B., Shklovskii, B. I., and Glazman, 1. I., (1992) Phys. Rev. B 46,
4026.
Bruder, C., and Scholler, H., (1994) Phys. Rev. Lett. 72, 1076.
Hettler, M. H., and Schoeller, H., (1995) Phys. Rev. Lett. 74, 4907.
Stafford, C. A., and Wingreen, N. S., (1996) Phys. Rev. Lett. 76, 1916.
Stafford, C. A., (1996) Phys. Rev. Lett. 77, 2770.
Landauer, R., and Martin, Th., (1994) Rev. Mod. Physics 66, 217.
Biittiker, M., (1994) Physica Scripta, T 54, 104.
Biittiker, M. and Stafford, C. A., (1996) Phys. Rev. Lett. 76, 495.
Gopar, V. A., Mello, P. A., and Biittiker, M., (1996) Phys. Rev. Lett. 773005.
Brouwer, P. W., and Biittiker, M., (unpublished).
Scholler, H., (1996) (unpublished correspondence).
TRANSPORT THEORY OF INTERACTING QUANTUM DOTS
H. SCHOELLER
Institut fur Theoretische Festkorperphysik,

Universitat Karlsruhe,
76128 Karlsruhe, Germany
1. Introduction
In the last few years the progress in microfabrication technology has led
to an enhanced interest in transport properties of ultrasmall conducting
islands coupled weakly to leads (for reviews see Refs.[1-6]). Quantization of
charge and tunneling through zero-dimensional states lead to many interesting phenomena in these systems. Adding a single charge to a small island
costs the charging energy Ec rv e2 / (EL) rv e2 / (2C) (L being the length
scale of the island, E the dielectric constant, and C the self-capacitance),
and, second, the level spacing bE of the single-particle states. For system
lengths in the nanoscale regime, charging energies can be reached of order
1 - 10K. For temperatures below 1K this implies that electron transport
can be completely blocked (Coulomb blockade) or being restricted to a
small number of possible charge states. In the same way electron transport
can be influenced by the discrete level structure on the island. Especially
in 2d semiconductor quantum dots the level spacing is large (typically 1/10
of the charging energy).
The sensitivity to adding a single charge can be used for measurement
applications, e.g. for the detection of single charges or for setting up current
standards. Electronic applications are the subject of intensive research and
could become of technological interest if the operating temperature of these
devices can be increased. Experimentalists can use single electron phenomena as spectroscopic tools. For theoreticians quantum dots are important
systems for studying models of strongly correlated systems in equilibrium
or nonequilibrium. Quantum dots represent various realizations of generalized Kondo and Anderson models. Arrays of quantum dots can be used to
model Hubbard chains.
291
L. L. SOM et al. (eds.), Mesoscopic Electron Transport, 291-330.

292
Many phenomena in single electron devices can be understood within
golden rule theory. This means that tunneling to the particle reservoirs is so
weak that the spectral density of the island remains unchanged and transport can be described by classical master equations, the so-called orthodox
theory [1]. A crucial assumption in justifying perturbation theory is a small
intrinsic broadening of the island excitations compared to temperature T
(we always set kB = 1). Experimentally this can easily be achieved by
using tunneling barriers with tunneling resistances RT much higher than
the quantum resistance RK = hi e2 Thus, there exists a well-defined experimental regime where perturbation theory can describe single-electron
tunneling through zero-dimensional states [2].
It is important to notice that a master equation with golden rule tunneling rates is a perturbative approach in the coupling to the reservoirs but
not in the interaction within the island. Therefore, this approach has to be
distinguished from the well-known scattering formalism [7] which can describe coherent transport through mesoscopic devices for arbitrary tunneling barriers and temperatures but is restricted to noninteracting systems.
It is therefore very important to formulate theories which can interpolate
between both limits. It is the purpose of this report to present a technique
which is capabable of describing coherent transport through interacting
islands.
There are several experimental motivations to study coherent transport
through strongly interacting quantum dots. First of all there are regimes
where sequential tunneling is exponentially suppressed. This happens in the
Coulomb blockade regime where the current is dominated by higher order
processes such as coherent" cotunneling" processes of electrons through several junctions [8]. In interference geometries where quantum dots are part
of an Aharonov-Bohm ring, only higher order processes beyond sequential
tunneling show a flux dependence and lead to Aharonov-Bohm oscillations
[9]. Experiments can be performed in the limit where the tunneling barriers
are so low that even the case of perfect transmission can be reached without destroying the effect of Coulomb blockade. This leads to a significant
deviation from" orthodox theory" even in regimes where sequential tunneling contributes. The spectral density of the island will be strongly affected
by the coupling to the leads, and the broadening of levels will approach
temperature or level spacing upon continously increasing tunneling. In the
presence of interactions the broadening can be a complicated function of
energy, temperature and bias voltage. This induces strong renormalization
effects of the levels and the system parameters. For quantum dots described
by one degenerate low-lying level it can even lead to new resonances in the
spectral density in the form of Kondo resonances. They show up in various
anomalies in the differential conductance as function of the bias voltage.
293
Quantum dots with continuous level spectra are, in the two charge-state
approximation, equivalent to multichannel Kondo models. Again, this gives
rise to anomalous temperature dependences of the conductance as function
of gate or bias voltage. By varying the level spacing, level position or using
multi-dot systems an enormous variety of interesting many-body systems
can be realized. Their low-temperature scaling behaviour is still not known
for most cases.
Here we are interested in the case where the transmission per channel
of the barriers is still much less than unity so that a well-defined description via a tunneling Hamiltonian is justified. One should recognize that,
for large channel number, this includes the possibility of total transmission being larger than unity. Experiments in this regime have recently been
performed in metallic dots with clear signs for deviations from classical behaviour [10]. Furthermore, quantum fluctuations become visible by lowering
the temperature. Especially vertical quantum dots [11], ultrasmall metallic
particles [12] or molecules [13], are promising candidates for the observation of quantum fluctuations in the weak transmission limit at realistic
tem peratures.
The transport theory presented here is based on a recently developed
real-time diagrammatic approach [14-17]. It is closely related to path-integral
methods using the Feynman-Vernon technique [18] formulated in connection with dissipation [19,20] or tunneling in metallic junctions [5,21]. The
idea is to integrate out the reservoir degrees of freedom and to set up a
formally exact kinetic equation for the reduced density matrix of the dot.
The kernel of this integro-differential equation is represented as a sum over
all irreducible diagrams and can be calculated in a systematic perturbation
expansion in tunneling. In this way the strong correlations on the island
are fully taken into account. Furthermore, the golden rule theory, which
is reproduced by using the kernel in lowest order perturbation theory, can
be systematically generalized to higher orders. We will formulate an approximation for an explicit calculation of the kernel which reproduces the
Landauer-Biittiker theory in the noninteracting limit but provides also a
good description for coherent transport in the strongly interacting case.
2. Single-electron devices
2.1. MOTIVATION: THE COULOMB BLOCKADE MODEL
In this section we discuss the basic physical properties of quantum dots.

We introduce a simplified model and discuss the conditions for various
energy scales when Coulomb blockade phenomena and tunneling through
zero-dimensional states are observable.
We consider a small island containing interacting electrons in a uni-
294
Figure 1. The SET transistor. All three terminals are coupled capacitively to the island.
Two tunnel junctions allow transport from the left reservoir to the right one.
form positive background charge. The island is coupled electrostatically

to macroscopic metallic reservoirs with different electrochemical potentials
P,r = eVr , r = L, R. A current can flow by tunneling of electrons across
the tunnel junctions. A schematic view of such an arrangement is shown
in Fig. 1. The total charge on the island is given by Q = eN, where N
denotes the excess electron number. By means of a third terminal, called
the gate, which is coupled electrostatically to the island, one can change
the electrochemical potential of the island independent of VL and YR. In
this way it is possible to control the particle number on the island. Such a
system is called a single-electron transistor (SET) in the general nonequilibrium situation where VL #- VR, or a single electron box (SEB) for the
equilibrium case where VL = YR.
The length scale L of the island is typically of order 0.1 - Ip,m. This is
large compared to atomic scales and it is possible to couple the island to
macroscopic voltage sources. On the other hand, the system size is so small
that single charge-transfer processes can be measured on a meV voltage
scale. Adding one single charge to the neutral island will cost a charging energy Ec rv e2 j(fL) rv 0.1 - 1meV rv 1 - 10K where we have assumed f rv 10 for typical semiconductor quantum dots. The level spacing
5E rv (kFL)2-dJi2Jr2j(m*L 2) defines the second energy scale for adding one
electron. Here, kF is the Fermi wave vector, d the dimension, and m* the
effective electron mass. To achieve 5E rv 1K, one has to reduce the dimensionality or use smaller system sizes. For a 3d metallic system with Fermi
wave length AF rv lOA, one needs L rv 10nm. For a 2d electron gas it is
sufficient to take L rv 100nm. Furthermore, the level spacing is increased
in systems with small effective mass.
We start with an analysis of the concept of charging energy. To calculate
the electrostatic work Epot to build up an arbitrary charge distribution on
the island we use the so-called Coulomb blockade model. It means that the
295
island is treated like a metal, i.e., the electrostatic potential on the island
is assumed to be homogeneous. For 3d metallic systems this is usually a
good approximation except for system sizes L smaller than 10nm where
the Thomas-Fermi screening length ATF = (AFaB/8)1/2 starts to become
comparable to L and the potential is no longer homogeneous over the island.
In 2d semiconductor quantum dots there is no exponential screening and
the screening length is given by the Bohr radius aBo Here it depends on
the particle number and the distance to the gates whether the Coulomb
blockade model can be used.
Within the capacitive model the electrostatic work Epot(Q) to build up
the total charge Q on the island is given by Epot(Q) = foQ dQ'V(Q'), where
V(Q) is the electrostatic potential of the island for given island charge Q.
It depends on the fixed voltages ~, i = L, R, 9, of the metallic reservoirs
and follows from Ci(~ - V) = Qi, where Qi is the screening charge on
capacitor i (see Fig. 1 for notations). Using -Q = QL + QR + Qg together
with the definitions C = CL + CR + Cg and qx = -enx = Li=L,R,g Ci~,
we obtain V(Q) = (Q + qx)/C and thus
(1)
where we have added the irrelevant constant Ecn;. The charging energy
Ec is given by Ec = e2 /(2C). For typical lengths L rv 0.1 - 1J.Lm and a
dielectric constant E rv 10, the capacitance is of order C rv 10- 16 - 10- 15 F.
The system tries to minimize its electrostatic energy. Therefore, the
integer particle number N tends to be as close as possible to the continuous variable n x . As a consequence, the particle number on the island can
be controlled in discrete units by varying nx via the gate voltage Vg. For
half-integer values of n x , two adjacent particle numbers N = nx 1/2 lead
to the same electrostatic energy and transport is possible. Away from the
degeneracy points, transport is suppressed up to smearing due to temperature, bias voltage and quantum fluctuations. This is the phenomenon of
Coulomb-blockade.
So far we have considered only the Coulomb interaction. The total energy E of the island is given by
=L
EkDnkD
+ Ec(N -
n x )2 ,
(2)
where IkD > are single-particle states of the dot with occupation nkD and
energy EkD. k is the wave vector numerating the states. Furthermore, the
total excess particle number is given by N = Lk nkD - No, where No is
the number of electrons on the neutral island. If the particle number increases from N to N + 1, the ground state energy changes by the amount
296
t --
I!L
N+2
0 ... _ _ ON+I
- - ON
- - ONI
eVg
Figure 2. One-particle excitation energies of the Coulomb blockade modeL For simplicity
it is assumed that the level spacing is a constant. If an excitation AN falls into the window
of the electrochemical potentials of the reservoirs, transport can occur. The position of
AN depends linearly on the gate voltage Vg
1::1N = ERr+1 - ERr = E~+No+1 + 2Ec(N - n x ) + Ec It describes a oneparticle excitation energy of the island corresponding to a transition between ground state energies with different particle numbers. The quantity
1::1N can also be regarded as the definition of the electrochemical potential
of the island. Of course there are other excitations involving excited states,
which become important if the level spacing oE is smaller than temperature
or bias voltage.
We are now able to set up the conditions when transport is possible.
In Fig. 2 we have shown an energy profile of the double barrier structure
indicating all electrochemical potentials of the reservoirs and the excitation
energies of the island. For constant level spacing oE, all excitation energies
are equidistant 1::1 = 1::1N+1 - 1::1N = oE + 2Ec. Furthermore, their absolute
position can be shifted linearly by the electrochemical potential J.lg = eVg
of the gate o1::1N/OJ.lg = Cg/C. In lowest order perturbation theory in the
tunneling barriers, energy conservation and the Pauli principle restrict tunneling. This means that one of the excitations 1::1N has to lie within the
window of the electrochemical potentials of the reservoirs J.lR < 1::1N < J.lL.
For finite temperatures, this condition has to be fulfilled only within the
smearing defined by the Fermi distribution function. If no excitation lies
between J.lR and J.lL, transport is suppressed. Thus, in order to observe a
significant modulation of the current as function of the gate voltage, we
need T, eV = e(VL - VR) ~ 1::1 which implies T, eV ~ oE or T, eV ~ Ec.
The first condition guarantees transport through zero-dimensional states,
whereas the second one implies the occurence of Coulomb-blockade phenomena.
Within golden rule theory it is sufficient to consider the excitation spectrum of the isolated dot as shown in Fig. 2. This means that we have neglected so far the fact that the spectral density of the dot itself can be
297
changed by the presence of the reservoirs. Due to the finite life-time T of

the excitations there will be a corresponding broadening rv tilT and via
Kramers-Kronig also a renormalization. We denote the temperature where
the renormalization becomes significant by TK and call it "Kondo temperature" since the model is similiar to Kondo and Anderson models. The
broadening and renormalization has two important consequences. First, in
the low-temperature region where T < tilT or T < TK, golden rule theory
breaks down, higher order processes become important and nonperturbative
methods have to be applied. This is the region where quantum fluctuations
are important but single-electron tunneling still persists. Secondly, if the
broadening tilT approaches the spacing ~ of the excitations, single electron
phenomena will no longer be visible. This is the regime of strong tunneling.
Let us start with the case of large level spacing 8E ~ T. Although the
life-time of an excitation involving many-body states is strongly influenced
by interactions (see section 4.3), a rough estimate for the energy scale of the
broadening can be obtained by comparing with the noninteracting case. A
single state in a dou ble barrier has a Breit-Wigner broadening r of the order
r == It1 2 8E, where IW is the transmission probability of a single barrier
[22]. For the Kondo temperature TK, no general estimate is possible since
it depends on the spectrum of the dot (see section 4.3). As already stated
above, deviations from golden rule theory occur in the low-temperature
region T < r or T < TK (see section 4.2 and 4.3). The regime of strong
tunneling tilT rv 8E cannot be achieved here since, for high tunneling
barriers, IW ~ 1, and consequently tilT rv r ~ 8E.
For 3d metallic systems, where the level spacing 8E is very small, the
situation is more complicated. Here, tunneling can happen through many
excited states and the broadening of the charge excitations turns out to be
r multiplied with the number of available states for tunneling into or out
of the island (see section 4.4) tilT rv (\10 max(~N, T, eV). Here,
(\10
1 RK
1
2
=- = -Ziti
411"2 RT
411"2
rv
Z8E
(3)
is, up to a conventional factor 1/(411"2), the dimensionless conductance of a

single barrier, and Z the number of transverse channels. RK = hi e2 is the
quantum resistance and GT = llRT = Z(e 2 Ih)IW the tunneling conductance of a single barrier. For ~N rv Ee ~ T, eV, the broadening is given
by tilT rv (\IoEe rv hl(RTC) which agrees with the classical relaxation time
for a charge on a capacitor in a RC-circuit. Single electron phenomena persist if the broadening tilT is much less than the distance ~ rv Ee between
the excitations. This is fulfilled for (\10 ~ 1 or, equivalently, zr ~ 8E.
In contrast to the case of large level spacing, this condition is not automatically fulfilled for large tunneling barriers. For large transverse channel
298
number Z, ao can be of order unity even if r ~ 8E. This is the regime of

strong tunneling where quantum fluctuations are enhanced by lowering the
tunneling barriers. When the condition ao ~ 1 is fulfilled, single electron
phenomena are visible, but, due to renormalization of charge excitations,
golden rule theory again has to be improved in the low-temperature regime
(see section 4.4).
2.2. HAMILTONIAN AND CURRENT OPERATOR
In this section we will set up the Hamiltonian and the current operator. We
distinguish between two different cases: Quantum dots with discrete quantum states and metallic islands with a continuous single-particle spectrum.
We use the convention = kB = 1 and e < O.
2.2.1. Quantum dots

We consider a small island coupled to several metallic reservoirs and to an
external heat bath. The bath can be represented by an environment or by
internal bosonic degrees of freedom like, e.g., phonons or plasmons. For the
general theory we need no assumption for the island Hamiltonian and include the possibility that the voltages on the reservoirs are time-dependent.
The coupling to the reservoirs includes an electrostatic interaction as well
as tunneling of electrons through high barriers. Let us first state the obvious form of the Hamiltonian and the current operator. For the interested
reader, the explicit derivations are presented at the end of this section.
The model Hamiltonian reads H(t) = Ho+HT(t) with Ho = HR+HB+
HD. Here, HR, HB and HD denote the Hamiltonians for the reservoirs, the
heat bath, and the dot, respectively, and HT(t) describes the tunneling
between dot and reservoirs. Explicitly, we have
Ho
HT(t)
HR+HB+HD
Ekakr akr + wqb!bq + EJ>s ,
r=L,R k
q
s
t
'J + (h.c.) .
Tk,ss,(t)akrPss,e-t
L L
= L L
r=L,Rk,ss'
(4)
(5)
All terms have an obvious interpretation. Ikr > denote the single particle
states in reservoir r with energy Ek' Wq are the frequency modes of the heat
bath, Es are the energy eigenvalues of the many-body states Is > of the
isolated dot, and Ps = Is >< sl is the projector on state Is >. For the
Coulomb blockade model (2), the states Is > ofthe dot are specified by the
set of all occupation numbers for the single particle states: Is >= I{ nkD} >.
The more general notation is introduced here since we want to include cases
where the states of the dot cannot be described by single particle states, see
299
e.g. [23, 24, 33]. Furthermore, the states Is > can represent charge states
(see section 2.2.2.), spin states, or states of multiple dots. This allows a
unified treatment of many possibilities.
The tunneling part (5) describes charge transfer processes where the
tunneling matrix element T/;,ss' corresponds to a transition of the dot state
Is' > to Is >. Therefore, we have introduced the operators Pss' =

Is >< s'l. Due to particle number conservation, we have Tkr ,ss, = 0 unless
Ns = N s' - 1, where Ns is the particle number on the dot for state Is >.
from
The electrostatic interaction between dot and reservoirs is described by the

effective time dependence
r (t) = T r (t) ie It: dt'Vr(t')
Tk,ss'
k,ss'
e
,
(6)
where Vr(t)
Vr(t) - VD(t) is the change of the electrostatic energy of
a particle entering reservoir r. Vr (t) denotes the time-dependent voltage
on reservoir r, and VD(t) is the spatial average of the external electrostatic
potential taken over the dot. The part of the electrostatic interaction which
remains for zero voltage on all reservoirs is included in HD. E.g. for the
Coulomb blockade model (2), we have VD(t) = -2Ecnx(t) whereas the
part ECN2 of the electrostatic interaction is included in HD. A possible
explicit time dependence of the tunneling matrix elements T in (6) accounts
for a modulation of the barriers.
Finally, the bosonic phase factor exp (-i) in (5) describes the energy
exchange with the heat bath due to absorption or emission of bosonic
modes. The linear bosonic field is defined by = i L. q9q / Wq (bq - b~),
where 9q is the coupling constant to the heat bath for mode q. This model
has been used widely in the literature, either to describe optical phonons
in semiconductor quantum dots [25] or voltage fluctuations in metallic systems [26]. In the latter case, the relation between the coupling constants
9q, their spectral function J(w), and the impedance Z(w) of the external
circuit is given by [26]
J(W) =
7r
L9;8(w - wq ) = e2 wReZ(w) ,
(7)
where W > 0, since the bosonic modes Wq are all positive. For the special
case of ohmic dissipation J(w) rv W, we obtain the Caldeira-Leggett model
[19]. For an extended discussion of various kinds of possible environments
we refer the reader to Ref. [27].
The physical observable which can be measured experimentally is the
current Ir flowing in reservoir r. This current consists of two contributions:
a tunneling current I;un(t) from electrons hopping on or off the island and a
300
displacement current I~iS(t) = ftQr(t) arising from the change of the timedependent screening charge Qr (t) on reservoir r. Let us show how the latter
can be calculated for the simplified Coulomb blockade model introduced in
section 2.1. For given charge Q(t) on the island and given potentials Vr(t),
r = L, R, g, on the reservoirs and the gate we get for the screening charge
Qr = Cr(Vr - V) with V = (Q + qx)/C being the potential on the island.
Inserting the definition qx = L:r CrVr and taking the time derivative we get
for the displacement current
(8)
The time-derivative of the island charge Q = L:r I;un is known after we
have calculated the tunneling currents. Summing (8) over r we find total
current conservation L:r Ir(t) = L:r[I~iS(t)+I;un(t)] = O. The displacement
currents are only important for the calculation of AC-currents since the time
average of I~is is usually zero except for cases where f Vr #- O.
The tunneling current operator i;un(t) is given by the time derivative of the particle number operator in reservoir r, i;un(t) = -eftNr =
-ie[H(t), Nr]. Inserting for H(t) from (4) and (5) we find
i;un(t) = ie
L T'k,ssl(t)alJ>ssle- i, + (h.c.) ,
k,ss'
(9)
where the explicit time dependence stems only from the time dependent
tunneling matrix elements.
Let us now turn to the derivation of the Hamiltonian (4) and (5). The
microscopic starting point is
H(t) = HD(t)
+ HR(t) + HB + VDB + HT(t) ,
(10)
where HD(t), HR(t) and HB denote the Hamiltonians for the dot, the reservoirs and the heat bath, respectively. VDB describes the interaction between
dot and heat bath and HT(t) the tunneling between dot and reservoirs.
The general form of the dot Hamiltonian is HD(t) = HB + eVex(t),
where eVex(t) is the potential energy from the fixed voltage distribution
on the reservoirs whereas the electrostatic energy for zero voltage on all
reservoirs is included in HB. Screening effects and nearby gates often justify
the form Vex(t) = VD(t)N, where VD(t) is the spatial average of the external
electrostatic potential taken over the dot. We denote the normalized and
orthogonal many-body eigenfunctions of HB by Is > with energy Es and
obtain HD(t) = L:s EsPs + eVD(t)N.
301
For the reservoir Hamiltonian HR(t) we take a noninteracting Fermi

liquid with perfect screening properties like in an ideal metal HR(t) =
Lkr El::akrakr + e Lr Vr(t)Nn where Vr(t) is the electrostatic potential of
reservoir r, and Nr is the particle number operator.
Tunneling between reservoirs and island is described by
HT(t) = LTkZ(t)akralD
r,kl
+ (h.c.) ,
(11)
where TkZ(t) are possibly time-dependent tunneling matrix elements and

aID is a field operator corresponding to a set of single particle states on the
dot. The tunneling matrix elements are described by the spectral function
fZl,(t, t'; w) = 27r Lk TkZ (t)*Tkl' (t')o(w - EI::). In the time-independent case,
one often uses the approximation rzz'(w) :::::: oll'f r This assumes constant
density of states in the reservoirs as well as the neglect of interference
phenomena in higher order perturbation theory in f. Expressed in the basis
.
t
Is > we have HT(t) = Lr,k,ss' Tk,ss,(t)akrPss' + (h.c.). The transformed
tunneling matrix elements Tk ss' (t) = Lz TkZ(t) < slalDls' > involve matrix
elements of the field operator' aID between many-body states of the island.
They can lead to exclusion rules [23, 24]. The function f in the basis Is >
reads
A
The heat bath HE is modelled by a set of harmonic oscillators HE =

N of the island by
the interaction term
Eq wqb~bq which couple to the particle number operator
(13)
The second term is a counter-term which is necessary to avoid an unphysical
renormalization of the energies Es (see below). The first term describes a
fluctuating electrochemical potential on the island.
Finally we perform a standard time-dependent unitary transformation
U(t) to bring the Hamiltonian into the most convenient form. We choose
U(t) = exp {ie Jt~ dt'(Lr Vr(t')Nr + VD(t')N)}exp (iN), where to is the
initial time and the hermitian bosonic field is defined by = i Lq ~: (b qb~). The transformed Hamiltonian H = UHUt + i(ftU)Ut reads H(t) =
flo + iJr(t) where flo and flT(t) are given by (4) and (5), respectively. We
see that the last term of (13) has been cancelled. Furthermore, we define
302
the coupling function
f'r,
,(tl,t2;W) after the unitary transformation
S1 S1 ,S2 S2
by replacing T by f' in Eq. (12).

For convenience we drop finally the bar on all operators and imply that
the Hamiltonians and all observables A(t) == A(t) = U(t)A(t)U(t)t are the
transformed ones after the unitary transformation. The states Is > together
with the projectors PSS' are kept unchanged.
2.2.2. Metallic island

A metallic island is characterized by a very small level spacing <5E ~ T
which means that a very large number of excitation energies are relevant.
Therefore, following the standard approach, we introduce two approximations from the very beginning. First, like the reservoirs, we treat the island as a Fermi liquid with perfect screening. This means that we use the
Coulomb blockade model (2) for the dot Hamiltonian
HD(t)
= L EkDakDakD + EC N2 + eVD(t)N,
(14)
with eVD(t) = -2Ecnx(t). The total Hamiltonian is again of the form (10)
with all other parts given as in the previous section. Secondly, we separate
the charge degrees of freedom of the island (described by N) from the degrees of freedom describing how the particles on the island are distributed
among the single particle states (described by nf). Furthermore we fix the
distribution function on the island by a Fermi distribution. These approximations are justified since the time scale for the change of the distribution
function is much larger than the time scale for the variation of the total particle number. To formulate this more precisely we enlarge first our Hilbert
space by introducing formal charge states IN > with N ranging from plus
to minus infinity. We define the operator N in (14) by NIN >= NIN >
and the projectors PNN ' = IN >< N'I. We demand that each time an
electron changes its position from some reservoir to the island or vice versa
via tunneling, the charge state has to change simultaneously from IN >
to IN 1 >. This is achieved by introducing the projectors PN 1,N into
the tunneling Hamiltonian HT(t) = Lr,kZ,N TkZ(t)akralDPN-l,N + (h.c.).
By construction, the new Hamiltonian is exactly equivalent to the old one
provided we use the constraint N = Lk
No to restrict ourselves to the
original physical Hilbert space. The approximation formulated above corresponds to the neglect of the latter constraint. Although the corrections
have never been analysed systematically it appears reasonable that they
are finite-size corrections and scale like the inverse volume of the island.
We now apply the same unitary transformation U(t) as in the previous
section. The result for the transformed Hamiltonian is fI(t) = fIo + fIT(t)
nf -
303
with
Ho
fIT(t)
fIR + fIB + fIe

L
L Ekatr akr
r=L,R,D k
~
L.J
L.J
r=L,RkZ,N
+ L wqb~bq + LEN PN ,
(15)
.J,
(16)
t'
Tkl(t)akra/DPN-l,Ne-t
+ (h.c.) ,
where TkZ(t) is defined analog to (6). As indicated we decomposed the dot

Hamiltonian (14) into a "reservoir" part Lk EkDatDakD, which has been
included in fIR, and a charge part fIe = LN ENPN , with EN = EeN2.
Following the previous section, we drop from now on all bars and identify A == A = U AUt for all observables A. Furthermore, using the analog
derivation to the quantum dot case, we obtain for the tunneling current
operator
i;un(t) = ie L TkZ(t)akra/DPN_l,Ne-iJ,
kI,N
+ (h.c.) ,
(17)
whereas the displacement current can be calculated from (8).

3. Real-time transport theory
3.1. GENERAL CONCEPT
In this section we will explain the general structure of the theory without
going into details of technical derivations. The full microscopic approach
together with explicit expressions for various quantities introduced here will
be presented in the next section 3.2.
The following considerations refer to the quantum dot case but hold
as well for the metallic island by the replacement of dot states by charge
states (formally D -+ C, s -+ N).
The tunneling part HT(t) describes the coupling between the environment (reservoirs and heat bath) and the dot. It will drive the dot system
out of equilibrium. For t :S to, we assume HT(t) to vanish, and the environment to be in equilibrium. This means that the initial density matrix can be
written in factorized form p(to) = pe;{P'Jj p(to), where pe;{p';j is the grandcanonical equilibrium density matrix of the environment, and p(t) is the
reduced density matrix of the dot p(t) = TrRBP(t) with TrRB = TrRTrB
being the trace over the reservoir and heat bath degrees of freedom.
At time to we switch on the tunneling between dot and reservoirs. For
to -+ -00 this is performed adiabatically. Our first aim is to study the
time evolution of p(t). This will be performed in section 3.2 by integrating
304
out the reservoirs and the heat bath with the result of an effective theory
in terms of the dot degrees of freedom. We will obtain the formally exact
kinetic equation
where PSSI(t) =< sl?(t)ls' >. The second term on the l.h.s. of this equation
is a flow term which describes the time evolution of the density matrix in the
absence of tunneling. This is not a dissipative source and, in the absence of
tunneling, would lead to a coherent time evolution of the dot. Dissipation is
described by the r.h.s. of Eq. (18). It forces the dot to approach a stationary
state and is due to tunneling.
The kinetic equation (18) can be written in a more familiar and transparent form by eliminating the nondiagonal matrix elements of the probability distribution. This leads to
(19)
Using the property Ls ESSI (t, t') = 0, which is proven in section 3.2 and
guarantees the conservation of probability Ls Ps (t) = 0, we can rewrite the
kinetic equation as
dd Ps(t)
t
L rt dt' {ESSI(t,t')PSI(t') -
S'::pS lto
ESIS(t,t')Ps(t')} .
(20)
We have obtained the structure of a master equation with a gain and loss
term on the r.h.s .. The kernel ESSI(t, t') can be interpreted as a generalized
and formally exact transition rate from the state s' at time t' to the state
s at time t. In second order in HT, we obtain the lowest order expression
for the rate but for arbitrary time-dependent situations. In the asymptotic
limit to -+ -00 it reduces to the golden rule rate when integrated over
the time difference t - t' (see at the end of this section and section 4.1).
In the context of Coulomb blockade phenomena, this term is called the
transition rate of "sequential tunneling". It corresponds to the physical
situation where all tunneling processes are incoherent. The next term, which
is of forth order in HT, is called the cotunneling transition rate. It means
that at least two tunneling processes are correlated allowing for coherent
transport through the dot from one reservoir to the other. The higher order
terms contain processes where the electron tunnels coherently back and
forth between the dot and the reservoirs. We will see in section 4 that
this can lead to renormalization and broadening effects. Except for special
305
systems which are exactly solvable (see section 4.2 for an example), it is
not possible to calculate ~ exactly. However, we will at least formulate
a systematic and very general approximation in section 3.2 which will be
applied to specific examples in section 4. We call the kernel ~SSI(t, t') within
this approximation the resonant tunneling transition rate.
For the special case of a diagonal density matrix PSSI(t) = OSSIPS(t),
the kernel is given by ~SSI (t, t') = ~SS,SI S' (t, t'). As we will see in section
4, there are special systems with the property that p(t) will be diagonal
for all times t if it is diagonal at the initial time to. To give a concrete
example we note the following property of PSSI (t) which follows from particle number con-servation PSSI(t) '" ONs,Nsil which is fulfilled for all times
if it is fulfilled initially. Here, Ns =< slNls > denotes the particle number on the dot for state Is >. This follows directly from the definition
PSSI(t) =< sITrRBP(t)ls' > and the fact that the total particle number
N tot = Lr=L,R N r + N is a conserved quantity. Thus, for metallic systems,
we find PNNI(t) = ONNIPN(t) if this property holds initially. A similiar
proof can also be given for quantum dots with a single spin 1/2 state where
spin conservation can be used (see section 4.3).
Another quantity of interest is the tunneling current given by the average of the corresponding operator I;un(t) = Trp(t)i;un(t). Inserting the
form (9) or (17) for the operator and again integrating out the reservoir and
heat bath degrees of freedom, we will show in section 3.2 that the tunneling
current can be written as
(21)
where we have already eliminated the nondiagonal elements.
The physical interpretation of (21) is very obvious. To obtain the tunneling current at time t, one has to multiply the current rate Ls ~~Sl(t, t'),
corresponding to the sum over all processes starting at t' in state s' and
ending at time t in any state, with the appropriate initial probability PSi (t')
and finally integrate over all initial times t' and sum over all initial states
Is' >. The index r indicates that during these processes the particle number
in reservoir r has changed. As a minor remark we note that, just for formal
reasons, only the sum over s of ~~Sl(t, t') is allowed to be interpreted as the
current rate.
The current rate includes all possible processes, i.e. the change of the
particle number in reservoir r can take any value. Therefore it is natural to
decompose the current rate in the form
L:~:SI(t,t')
s
=-
L: L:
00
s p=-oo
p~:~/(t,t'),
(22)
306
where ~:~, (t, t') corresponds to that part of the total transition rate ~ss' (t, t')
where in sum p particles are taken out of reservoir r. This allows a decomposition of the tunneling current into a tunneling "in" and a tunneling" out"
contribution
I;un(t)
= eL
00
p=l
it
ss' to
dt' {~:~(t, t')Ps,(t') - ~:~~P(t, t')Ps, (t')} .
(23)
Finally we show how the kinetic equation and the tunneling current can
be written in Fourier space. We are interested in the stationary solution
and set the initial time to = -00. We assume that the time-dependence
of the voltages and the tunneling matrix elements is periodic in time with
period T = 27r /0,. This implies ~ss' (t + T, t' + T) = ~ss' (t, t') and the
periodicity of the stationary probability distribution and the tunneling
current. Thus, we use the Fourier expansion Ps(t) = L~=-oo P-;-e innt ,
" (t t') -- ,\,00
I r (t) -- ,\,00
L.tn=-oo Inr einnt ,Llss',
L.tn=-oo "n
Llss' (t - t') einnt' ,an d a corresponding representation for ~~s' (t, t'). Inserting these expansions in the
kinetic equation (19) and the tunneling current (21), we find
'"""'
L...J '"""'
L...J ~mn
ss' pn-m
s'
,
s'
(24)
_ e '"""'
pn-m
L...J '"""'
L...J ~r,mn
ss'
s'
sst
(25)
where we have defined the Laplace transform ~~~ = fooo dT~~s'(T)e-imn'T,

and analog for ~:~~m.
If the period T is much smaller than the characteristic memory time
TE of the kernels, the n = 0 component of ~~s' (t - t') will give the most
dominant contribution to (19). The reason is that, for n i= 0, the factor
einnt ' will oscillate very strongly for t' varying on a range 71; ~ 0,-1. With
the same argument one can also neglect the components of ~~~ (t, t') for
m i= o. The AC-components of the probability distribution are then much
smaller than the DC-components and we obtain
L ~ss,Ps' = L (~ss,Ps' s'f.s

-e L ~~s,Ps"
~s'sPs) ,
(26)
s'
Ir
(27)
ss'
where, by convention, we imply always that we mean the DC-Fourier component n = m = 0 if no time argument and no Fourier index is written.
For time-translational invariant systems, the kernels depend only on the
relative time argument t - t' and Eqs. (26)-(27) hold exactly.
307
SI
tJ
si
,
s'----~,----~----~----
ss'
r\,
s'
+ ...
t.
Figure 3.
An example for a diagram contributing to the matrix element P ,(t) of
the reduced density matrix of the dot. Reservoir (boson) lines are indicated by dashed
(wiggly) lines.
3.2. MICROSCOPIC THEORY
In this section we provide the microscopic basis for the equations set up in
the previous section. We start from the definition PSSI(t) = Trp(t)Ps's of
the matrix elements of the reduced density matrix of the dot and find in
interaction picture with respect to Ho = HR + HB + HD
Pss' ( t ) = Trp (to) T, { e-i f 'Y dt' HT(t')J Ps's t

A
()}
(28)
where HT(th and Ps's(th are operators in interaction picture, and 'Y denotes the usual closed Keldysh contour which runs from to to t on the real
axis and then back again from t to to. T, denotes the time ordering along
this closed time path. Using the initial condition for the density matrix and
taking matrix elements we obtain
Pss,(t)
= ~ Ps's(to) < slTrRB p~p~T, {e-i I'Y dt'HT(t')J Ps's(th} Is' >
ss
(29)
The next step is to expand (29) in HT(th and insert the form (5) or (16)
for the tunneling Hamiltonian. The tunneling vertices are arranged along
the closed time path as indicated in Fig. 3. The upper line corresponds to
the forward propagator and the lower line to the backward propagator. To
each vertex we assign a time variable ti and, from the tunneling Hamiltonian, a projection operator Ps's;, where Si is the ingoing state and s~
the outgoing state at each vertex'(see Fig. 3). There is one external vertex
emerging from the projector Ps's in Eq. (29), which is the rightmost vertex
at time t in Fig. 3. It is the only vertex which does not contain any reservoir
or heat bath field operator.
The procedure is now to perform the trace over the reservoirs and the
heat bath, and finally calculate the matrix element with respect to the dot
states. The trace can be calculated exactly since Ho is a bilinear form in the
reservoir and boson field operators, and p~ B are equilibrium density matrices. What is left for each term is a c-numher multiplied with the matrix
308
element < 81 ... 1S' > of a product of dot projection operators in interaction picture. We note that the three steps, i.e. calculating TrR, TrB and the
matrix element of the dot operators, can be performed independently since
Ho = HR + HB + HD contains no coupling between reservoirs, heat bath
and dot. Furthermore, the reader can convince himself that Fermi statistics
does not give any minus sign during the factorization of reservoir from dot
field operators if both are kept in the same sequence separately. This is due
to the quadratic structure akraLD or a;Dakr of the tunneling vertex. In our
convention, the time-ordering operator T"( does not introduce any change
of sign.
Let us start with the calculation of TrR. It can be performed using
Wick's theorem with the result that all reservoir field operators are contracted in pairs of creation and annihilation operators. In our convention, a
single contraction for the quantum dot case gives the contribution ( refers
to tl ;t2 with respect to the Keldysh time path)
whereas for the metallic case we get
with j+ = j, j- = 1 - j, n+ = n, n- = 1 + n, and j(w), n(w) being the

Fermi and Bose distribution function. Furthermore, we have defined
and D by replacing r --+ f'.

For the metallic case we have used the fact that each loop of Wick
contractions is proportional to the transverse channel number Z. Therefore,
for large channel number, the loops will contain the minimal number of
vertices, i.e. they have the form of Eq. (31). Experimentally, the channel
number is usually of order 103 .
Eq. (32) can be written in a more explicit form if we assume a constant
density of states VD in the island, take tunneling matrix elements of the
309
form TkZ(t) = Fr(t)Tkz , and use approximately rrz(w) ~ rr independent of

1 and w. We obtain Dr(t,t'iW) = Fr(t)*Fr(t')Dr(w) with
(33)
and o!o = 1/(271')rrVD = RK /(471'2 RT) is proportional to the conductance
GT = 1/ RT of a single barrier connecting the island to reservoir r = L, R
in units of the quantum conductance GK = I/RK = e2/h.
For the quantum dot case, we get a minus sign for each crossing of contractions due to Fermi statistics. Diagrammatically, a contraction between
reservoir field operators is indicated by a dashed line (see Fig. 3). The direction of the line is chosen in such a way that it leaves the vertex where
a particle is annihilated on the dot. The time argument of this vertex has
to be chosen as the second time argument of the functions I and O!, i.e.
corresponds to t2 in Eq. (30) and (31). The states 81,2 (8~,2) refer to the
ingoing (outgoing) dot states at both vertices.
The calculation of TrB proceeds in a different way since the tunneling
vertex contains an exponential exp (iJ of a linear bosonic field. Here we
can use path integral methods or Feynman's disentangling method [28] to
get
(T"( { e-iJ(tJ)I eiJ(t~h ... e-iJ(tmh eiJ(t:-"h }) P~q
II p (ti, t
j)
-1
II p (t~, tj (1 II p (ti , tj) ,
(34)
i<j
i<j
where, for t1 ~ t2 with respect to the Keldysh path, we have defined the
(T"({e-iJ(tdleiJ(t2h})p~q
P(t) = e-W(-t), W(t) = S(t) + iR(t), and [20,27]
correlators p(t1,t 2)
S(t)
R(t)
10
P((t1 - t2)) with
J(w)
(JBW
dw,
2-(I- c08(wt))coth(-)
W
2
00
J(w)
dw-1
2-sin(wt) ,
71' 0
W
-1
71'
00
1
0
(35)
(36)
where J(w) is defined in (7). The Fourier transform P(E) = 2~ J dte iEt P(t)
describes the probability that an electron absorbs the energy E from the
bosonic environment [26, 27]. We write (34) formally as a sum by defining
L~(t1,t2) = p(t1,t2) -1 and L;(h,t 2) = p(t1,t2)-1 -1. Here, Ld
corresponds to a pair of vertices with different signs of the bosonic phase
fields, whereas Ls refers to a pair with the same sign. Both Ld and Ls
are zero if the coupling to the environment is absent. Diagrammatically,
310
we represent the bosonic contributions Ld,8 by wiggly lines (see Fig. 3).
In contrast to reservoir lines, an arbitrary number of bosonic lines can be
attached to a single vertex.
The matrix element < 81 ... IB'
in interaction picture is given by
> of products of dot projection operators
i=l
i=O
< 81 II 1\:8i(ti)JiB' >= II < si+lIUD(ti+!,ti)ls~ >,
(37)
where UD(t, t') is the evolution operator of HD, and we identified Sm+! = 8,
s~ = 8', and tm+l = to. This result means that each segment of the Keldysh
contour in Fig. 3, which connects two vertices, corresponds to a matrix
element of the dot evolution operator starting from the outgoing state of
the initial vertex to the incoming state of the final vertex. Since HD is
diagonal in the states Is >, the matrix element of the evolution operator is
given by < sIUD(t, t')ls' >= b88,e-iEs(t-t'). This means that we can assign a
certain dot state to each segment of the Keldysh contour. However, for more
general dot Hamiltonians, which include transitions between the dot states
Is >, one has to use the above description in terms of the dot evolution
operator.
Finally we have to assign the factor (_i)nim to each diagram which
arises from the expansion of the exponentials in Eq. (29). Here, n (m) is
the number of vertices on the forward (backward) propagator. The time
integrations are then all performed on the real axis from to to t. Assigning
a factor (_i)2 to each reservoir line, we can alternatively say that each
reservoir line and each vertex on the lower line of the Keldysh contour
gives rise to a minus sign. Furthermore, each diagram is multiplied with
the matrix element Ps',s(to) of the initial distribution.
We mention that each reservoir line can be dressed by a bosonic contribution. This means that instead of ,r,, the contribution of a reservoir
,(tl' t2) = ,r,,
,(tl' t2)P(tl , t 2), where we
line is replaced by..:yr,,
81 8 1' 8 2 8 2
81 8 1' 828 2
have added the two contributions of the two vertices being connected by
a reservoir and a boson line (giving ,Ld), and the term where they are
only connected by a reservoir line (giving ,). In the same way we replace
a;(t l ,t2) by &;(tl,t2) = a;(tl,t 2)P(tl,t2). In the presence of the heat
bath we will automatically imply from now on that a dashed line corresponds to a dressed reservoir line.
The dressing can be written very elegantly in Fourier space if we consider
the case of time-independent tunneling matrix elements Tkl and periodic
voltages of the harmonic form eVr(t) = eVro + ~r sin (fU). We define the
311
/ " +
+ ...
Figure 4. The kernel ~SlS~,S2S~(tl' t 2 ) which contains all irreducible diagrams in the
sense that an arbitrary vertical line will always cut through some reservoir or boson line.
Fourier expansion of A =
A(tl,t2) =
1', a by
f Jdwein~Ule-iw(tl-t2)An(w),
(38)
n=-(X)
which is possible since A(h

obtain
-r,,n
Is 1 s'l' s2 s'2
()
W
J
~J
~
2rr
+ T, t2 +T) =
A(h, t2). After some algebra we
dw'r: 1 s'l' s 2 s'2 (w - w' - e1i,.)f!(w - w')Prn(w')

, (39)
'
dw'Dr(w - w' - eVrO)n;(w -
w')Pr~n(w') ,
Pr~n(w) = in LJn+m(~)Jm(~ )P(w + mn).
(40)
(41)
This shows very clearly the effect of dressing a reservoir line with index r
and Fourier component n. The heat bath and the time-dependent fields
supply the energy w' for absorption/emission with probability Prn(w').
Without heat bath we obtain the usual Tien-Gordon theory for the 'n = 0
Fourier-component [29].
We can now proceed to derive the kinetic equation (18). Looking at
an arbitrary diagram we distinguish between two different time segments.
There are "free" time segments in the sense that a vertical line drawn
through the diagram will not cut through any reservoir or boson line. These
parts correspond to the free evolution of the density matrix of the dot without any coupling to the external environment. All the other time segments
are "irreducible", i.e. a vertical line cuts either through a reservoir or a boson line. They reflect the influence of the environment. We denote the sum
of all irreducible diagrams by the kernel ESlS~,S2S~(tl,t2)' with arguments
as shown in Fig. 4. The summation of sequences of irreducible blocks with
free parts in between can be written in the style of a Dyson equation (see
312
pet)
P(t,) (
=N
+
t
Figure 5. The Dyson-like equation for the probability distribution. L; includes all irreducible diagrams in the sense that any vertical line will at least cut one reservoir or
boson line.
Fig. 5)
SSI
(42)
where 11(0)
I
I (tl' t2) =< sllUD (tl, t2) I S 2 > < s~ IUD(t2, tt) lsi> describes
S1Sl'S2S2
the evolution of the density matrix in the free segments. Differentiating
(42) with respect to t, we arrive at the kinetic equation (18).
For the diagonal kinetic equation (19) we have to define the kernel ESS'
in a different way. We allow for free segments in the kernel as well but with
the restriction that the dot states associated with the lower and upper line
of the contour are different in the free segments.
The property 2:s ESS,(t, tf) = 0, needed for the derivation of (20), can
be easily proven by attaching the rightmost vertex of each diagram E to
the upper and lower proppagator. The minus sign for each vertex on the
backward propagator cancels both contributions if we sum over all states
s.
To calculate the tunneling current, we have to replace the projector PSiS
in (29) by the tunneling current operator (9) (quantum dot case) or (17)
(metallic island). This means that the rightmost vertex of each diagram
will be the tunneling current vertex which has the same structure as the
other tunneling vertices from HT. Therefore, the first irreducible block ~r
to the right is part of the total kernel E which enters the kinetic equation.
Here r is the index for the reservoir for which we want to calculate the
tunneling current. Accounting correctly for the sign of the tunneling current
vertex, we find immediately that Er is that part of E, where the reservoir
line attached to the rightmost vertex corresponds to reservoir r and is an
outgoing (ingoing) line if the rightmost vertex lies on the upper (lower)
313
(a)
;'
II
;'
~
1
10k
ok
Ie
/1
(b)
;'~
;'
II
;'
ok
;'
1 /
1
1/
.r
;'
k'
ok ek'
Figure 6. Diagrams contributing to (a) sequential and (b) resonant tunneling. At each
reservoir line we have indicated which state k of the reservoir is involved at the tunneling
vertices. This creates holes (open circles) or particles (filled circles) on the propagators.
propagator. The other irreducible blocks which follow ~r to the left are
identical to ~. Thus, after summing over all sequences of ~ which gives the
probability distribution P, we obtain (21). The proof of (22) requires some
more technical considerations and can be found in [17].
For a given model it is straightforward to calculate the lowest orders of
the kernels ~ and ~r. However, as we will see in section 4, renormalization
and broadening effects due to quantum fluctuations can only be derived by
considering an infinite series of higher order diagrams. We select this series
by allowing the total density matrix to be nondiagonal with respect to the
reservoir states up to a certain degree. For this let us disregard the bosonic
heat bath and consider first the lowest order contribution to the kernels.
This is the contribution to the sequential tunneling or golden rule rate and
consists diagrammatically of one single reservoir line. An example is shown
in Fig.6a. If the reservoir field operator at the tunneling vertices is a~~,
we see that one hole in reservoir r is present on the backward propagator
whereas the forward propagator remains unchanged (we use the states to
the left of the diagram as reference). This means that we are considering
a matrix element of the total density matrix which is offdiagonal up to
one hole excitation. If we consider all diagrams in lowest order, we find
that sequential tunneling can be characterized by offdiagonal elements up
to one hole or one electron excitation. This shows that the density matrix
tries to be as close as possible to a diagonal matrix with respect to the
reservoir states. Therefore it is natural to improve sequential tunneling by
considering the next possibility of nondiagonal matrix elements, namely
those which are offdiagonal up to one electron-hole, electron-electron or
hole-hole excitation. An example is shown in Fig.6b. It shows that this
approximation can be characterized diagrammatically by the condition that
any vertical cut can cut at most two reservoir lines.
Without the heat bath it can be shown that the sum over all diagrams
within this approximation can be written in the form of a self-consistent
integral equation. For special models this integral equation can be solved
314
s'
L :;, (t,t')
;'
s' ..
//~r
s
S
t'
.. "
~"
""
..
Figure 7. The diagrams for the golden rule rate to tunnel from reservoir r to the dot.
The rate to tunnel from the dot to reservoir r is obtained from the same diagrams by
inverting the direction of the reservoir lines.
analytically, otherwise one has to find the solution numerically. For the technical details we refer the reader to [14, 17]. In the presence of a heat bath,
one can use the same solution by dressing the reservoir lines. The inclusion
of bosonic lines between vertices which are not connected by reservoir lines
is very difficult and is still an open problem.
4. Applications
In this section we will describe several applications using the formalism we

have developed in section 3. We start with two well-known limits which
are standardly used in the literature to describe most of the experiments
dealing with transport through small devices: golden rule theory (sequential
tunneling) and the non interacting case (Landauer-Biittiker theory). Golden
rule theory treats the tunneling in lowest order whereas interaction effects
are incorporated in all orders. The noninteracting case disregards interaction effects whereas the tunneling is treated in all orders. In section 4.3 we
describe resonant tunneling in a quantum dot with large charging energy
and two possible spin excitations, and in section 4.4 resonant tunneling for
the infinite-Z metallic island in the two state approximation. Here tunneling is considered in all orders within the resonant tunneling approximation
set up in section 3, and interaction effects are taken fully into account.
4.1. SEQUENTIAL TUNNELING
The rates I;:~, (t, t') in second order in H T are shown in Fig. 7 for 8 =I=- 8'.
In the current formula (23) we need in lowest order only I;:; == I;:~71 with
8 =I=- 8'. Furthermore, for 8 =I=- 8', we get for the kernels entering the kinetic
equation (20) I;SS,(t, t') = Lr L p =1 I;:~, (t, t').
The diagrammatic rules give
I;r+
(t t')
ss' ,
+ (t H t') ,
e-i(Es-Es,)(t-t').:yr(t t') + (t H t') ,
s's,ss' ,
ei(Es-Es,)(t-t').:yr+
I
ss',s's
I
(t ' t')
(43)
(44)
315
for the quantum dot case, and
NN" (t t') --
I;r
ei(EN-EN, )(t-t') (i (t
r
t')b N,N'l + (t "-'---'- t')
(45)
for the metallic case with EN = ECN2.

Using these results one can, in principle, calculate the full time dependent solution starting from an arbitrary initial state. For the stationary
state one needs only the quantities I;:~;nm used in (24) and (25). From (43)
and the Fourier expansion (38) of 1', we find for periodic voltages
"r+,nm
LJ ss'
d - r,+,n ( )
w'Yss',s's w
{E s - Es' - w: (n - m)r2 + iry - Es - Es' -
~ + mr2 -
iry} ,
(46)
and a corresponding equation for I;:;,nm if we interchange Es H E~ and

" The metallic case follows from ;yr,,n
;yr;-s,n
H (ir,nbN N'l'
Iss, s
I . . . ,
Without heat bath and using the Coulomb blockade model for the quantum
dot, the stationary current has been calculated from these rates in Ref. [35].
Let us discuss some limiting cases which are usually treated in the literature. As discussed in section 3, we can restrict ourselves to the DCcomponents I;:~ == I;:~'oO if the time-dependent fields are absent or have
a very high frequency r2 ::?> r. From (46) we get
;yr,:,~
H
Iss,ss
(47)
and analog for the metallic case by 1'~,.,n H (i~,n bN,N'l.
In the absence of time-dependent fields and the heat bath this gives for
the quantum dot case
rr r:(Es -
E s')
2:1 < sla;Dls' > 12 ,
(48)
I;r-
ss'
rr f; (Es' -
Es)
2: I < slawls' > 12 ,
where we have used (39) together with rZI'(w)

we get
(49)
= bll'rr. For metallic islands
where we used (40) together with (33).

Eqs. (48)-(51) are the usual golden rule rates which are standardly used
in the literature. In the presence of time-dependent voltages and the heat
316
bath they have to be convoluted with the probability function p;' == Pr~O
defined in (41). Physically, they express what we have already discussed
qualitatively in section 2.1. For a current to flow through the dot we need
that both the tunneling "in" and tunneling "out" rates are present. For
this let us consider a transition between two dot states SN f+ SN+I, where
SN corresponds to a state with N particles on the dot. For tunneling "in"
we have s' = SN and S = SN+l in (48). This means that E SN +1 - ESN < eVr
according to the Fermi function in (48). For tunneling" out" we have s' =
SN+l and S = SN in (49). This gives E SN +1 - ESN > eVrl. Both conditions
can only be fulfilled simultaneously if the excitatio~ energy E SN +1 - ESN l~es
in the window of two different effective potentials eVrl < E SN +1 -ESN < eVr .
This expresses energy conservation from golden rule and the Pauli principle
as already explained in section 2.1. and illustrated in Fig. 2.
The master equation with golden rule rates has been studied extensively
in the literature. We mention Ref. [1] for the metallic case, Refs. [30-32] for
the Coulomb blockade model, Ref. [23, 33] for the quantum dot case with
exact many-body wave functions in the few electron limit, Ref. [34] for the
metallic case in time-dependent fields, Ref. [35] for the Coulomb blockade
model in time-dependent fields, and Refs. [26, 27] for the metallic case in
the presence of a heat bath.
Finally we want to mention that the kernels in lowest order perturbation
theory remain the same for the metallic case even if the channel number Z
is finite, since the lowest order can contain only one loop with two tunneling
vertices.
4.2. "NONINTERACTING" QUANTUM DOT
In this section we consider the special case of a quantum dot consisting

of one single-particle state in the absence of time-dependent fields and the
heat bath. This means that we consider the Hamiltonian
kr
kr
where c, ct are the field operators of the dot, and the time dependence of
the tunneling matrix elements involves only the static effective potentials
of the reservoirs tk(t) = TJ. exp (ieVr(t - to)). Obviously the Hamiltonian
has the form of a noninteracting system which can be solved exactly. Only
the presence of the effective potential Vr = Vr - VD within the tunneling
matrix elements reminds of the Coulomb interaction. Here, the latter has
only the effect of shifting the band buttom of the reservoirs and the dot.
The above Hamiltonian can be thought of as a special case of our general
Hamiltonian set up in section 2.2.1. To show this let us consider the case
317
where only one excitation energy ESI - Eso of the quantum dot is relevant,
with Iso> and lSI> being two ground states of the dot corresponding to
particle n urn bers Nand N + 1, respectively (e.g. N = 0). In this case, the
Hamiltonian from (4) and (5) reads
kr
kr
where we have already used Pso + PSI = 1 and omitted an overall constant.
This form is equivalent to the above Hamiltonian since we can identify
PSOSI = C, Iso >= 10 >, lSI >= 11 >, TL, 0 S I = TJ;" and f = ESI - Eso' This
means that there is a well-defined limit where an interacting quantum dot
can effectively be described by a noninteracting Hamiltonian [36]. However,
in a realistic situation degeneracies of excitations can hardly be excluded
due to spin and orbital effects, at least in the absence of high magnetic fields.
It is only this case where interaction effects really become important and
will change the qualitative behaviour of the noninteracting case completely
in the whole temperature regime (see section 4.3).
The nonequilibrium problem corresponding to the Hamiltonian (52) has
been solved exactly by many authors. We mention the Landauer-Biittiker
formalism [7,22,37], Keldysh technique [36,38], equation of motion methods [31], and golden rule theory with lorentzian broadening of the energy
conservation [35]. Here we will rederive the solution by using the resonant
tunneling approximation which turns out to be exact in the noninteracting
limit. We only show the analytical result here. For the technical details the
reader is refered to Refs. [17].
We first introduce the quantities 'Y;(w) = 'Y~llQ(w) which are, up to a
factor 27r, identical to the golden rule tunneling "in" and "out" rates given
by (47) if we set W= f. With (39) we obtain 'Y;(w) = (1/(27r))fr(w)f(=(w),
where we defined fr(w) = f Ol ,lQ(w - eVr )
The transition rates within the resonant tunneling approximation are
given by
(54)
where
()-f
aw -
d I Lr 'Yr (w')
w,"
W - W + try
= 'Y;:r(w) + 'Y;(w) = (1/(27r))fr(w),
(55)
and A = I dwlw - f - a(w)I-2.

In the numerator of Eq. (54) we recognize the golden rule transition
rates. The denominator describes a renormalization and a broadening of
'Yr(W)
318
the dot excitation energy
by the real and imaginary part of o-(w). We get
Ipfd ,r(w')
ReO' ()
w = -2
w
,
7r
(56)
w-w
where r = Lr r rand P J denotes a principal value integral. The renormalization and the broadening are independent of temperature and bias voltage. This is the reason why quantum fluctuations in noninteracting systems
do not result in anomalies in the zero-temperature limit. Furthermore, for
nearly constant density of states in the reservoirs the energy dependence of
r(w) is weak. This results in a small renormalization and a nearly constant
broadening.
If r is energy independent we have 0' = ir /2 and A = 27r /r. We obtain
L: 1o = J dw Lr rrfj(w)<5r(w - E) and L:01 = J dw Lr rrfr- (w)<5r(w - E) for
the transition rates. Here, the function <5r(w) = (r / (27r)) / (w 2 + (r /2)2)-1
has a lorentzian form with half-width r. If we replace this function by a
Dirac delta function we would obtain the golden rule theory. This result
expresses a very important feature of noninteracting systems with constant
r. One can just use elementary golden rule theory and obtains the exact
solution by smearing out the energy conservation by r! It is remarkable
that this property even holds when time-dependent fields are present [35].
It is basically due to the fact that the broadening of the dot excitation
energy is a constant and does not depend on energy, temperature or bias
voltage. We will see in the next section that the behaviour is very different
in interacting systems.
Finally, we find for the stationary tunneling current
Ir
=~ L
r'i-r
f dwTrr,(w) Ur(w) -
fr'(w)] ,
(57)
with the one-particle transmission probability given by
(58)
This formula agrees with the well-known Landauer-Biittiker formalism [7,
37] and is discussed in detail in [22]. In linear response, we have eVr =
f-l + e<5Vr with e<5Vr ~ T, r. This gives Ir = Lr' Grr , (<5Vr - <5Vr,) with the
conductance matrix given by the Breit-Wigner formula
e2 -rrrr'
-
G rr , = -27r h
dw<5r(w)f'(w + E - f-l),
where we have neglected the energy dependence of r r (w).
(59)
319
For T
(incoherent or sequential tunneling limit), we obtain

2 rrrr' r
max -2 ~--
c rr'
- IT h r 2 T'
(60)
i.e. a symmetric line shape of the resonance around J.l = f with exponential
tails. With decreasing temperature the line width decreases'" T and the
height of the resonance increases rv liT.
For T ~ r (coherent or resonant tunneling limit), we obtain
(61)
i.e. the line shape saturates at zero temperature to a lorentzian form reflecting the energy dependence of the transmission probability. For the special
case of two reservoirs which couple symmetrically to the dot, the height
of the resonance is given by the quantum conductance e2 lh. Compared to
the incoherent limit we see that quantum fluctuations tend to suppress the
cond uctance and broadens the line shape. The latter behaviour will also
be obtained qualitatively in the interacting case described in the following sections. However, we will see that the line shape has no longer to be
symmetrically, there can be logarithmic temperature or bias voltage dependences of peak position, peak height and broadening, and we will find
interesting anomalies for the differential conductance as function of the
bias voltage. All these features are completely absent in the noninteracting
case, since the renormalization and broadening of the dot level have no
interesting structure.
4.3. INTERACTING QUANTUM DOT
In this section we will study a more realistic and interesting case, namely
the presence of two relevant excitation energies f<7 = Esu - E so ' with ()' =t, t,
in the dot. This means that we consider two possible transitions when a
particle tunnels into the dot. If the incoming electron has spin up or down
we consider the transition So -+ St or So -+ s..!.' respectively. Due to spin
conservation the corresponding Hamiltonian is given by
(62)
where we have assumed spin independent tunneling matrix elements and
used Pso = 1 - 2:<7 Psu ' Each reservoir line carries now a spin index in
addition to the reservoir index.
320
This model has a very interesting analog in the theory of strongly correlated fermions, namely the so-called infinite-U Anderson impurity model
in nonequilibrium which is described by the Hamiltonian
kO'r
kO'r
with U --+ 00 being assumed to be the largest energy scale in the system.
The role of the dot is here taken over by the role of a local impurity with
one single spin 1/2 state. U is the on-site Coulomb repulsion and takes
over the role of the charging energy. Since U is assumed to be large, double
occupancy of the impurity level is suppressed and only the three states
10 >, I t>, and I ~> are possible. They are identified with the states Iso >,
1st >, and Is.j.. >, corresponding to the Hamiltonian (62), respectively. This
gives CO' == [>80.S", and we can see that the two Hamiltonians are equivalent.
The significance of this equivalence lies in the fact that it is known from
equilibrium theory that the Anderson model reveals a very interesting lowtemperature behaviour. For degenerate energies f = ft = f.j.. and in the
Kondo regime f ~ - f, the system shows resonant transmission at zero
temperature although the level position is far away from the Fermi level
(defined at zero energy). The reason is that the transmission probability
develops a Kondo resonance at the Fermi level for temperatures below the
Kondo temperature Tl{
(Uf)1/2 exp (n/f) [28,39]. The height of this
resonance increases'" In (TK /T) and saturates for very low temperatures.
At zero temperature the Kondo resonance is decreasing when the level
f approaches - f from below since the system leaves the Kondo regime.
However, for experimentally accessible temperatures, the Kondo resonance
is only visible for f '" - f due to the exponential dependence of the Kondo
temperature on f. This is the cross-over from the Kondo regime to the
mixed valence regime and corresponds roughly to the optimal value for the
height of the resonance at the Fermi level.
The idea to test these features by measuring zero-bias anomalies of the
differential conductance has a long history and many experiments have
been performed [40]. The disadvantage here is that the current is measured
through an ensemble of impurities and the control over physical parameters
like coupling constants or impurity level positions is weak. Therefore the
idea was formulated to test such features by measuring the conductance
through quantum dots [41]. Various calculations were performed for the
differential conductance as function of the bias voltage [42, 43] with the result of a zero-bias anomaly in the form of a maximum in the Kondo regime.
It was predicted that the Kondo resonance splits by an applied bias and is
shifted by Zeeman splitting [43]. The latter leads to a splitting of the zerobias maximum. These features have been observed experimentally by Ralph
f'.I
321
& Buhrman [44]. They measured the differential conductance through single charge traps in a metallic quantum point contact. Although this system
does not allow a controlled variation of the level position, the appearance of
a zero bias maximum with a peak height varying logarithmically with temperature clearly demonstrates the mechanism of Kondo assisted tunneling.
A detailed com parism of the line shape between experiment and theory can
be found in Refs. [17, 45]. The influence of external time dependent fields
or bosonic environments was studied in Refs. [16, 17, 46] with the result
of side band anomalies in the differential conductance and pump effects.
A closer investigation of the zero-bias anomaly reveals a cross-over of the
zero-bias maximum to a zero-bias minimum by shifting the level position
of the dot through the Fermi level [16, 17]. Further studies of the Kondo
effect in quantum dots involve the AC-conductance in linear response [47]
and Aharonov-Bohm oscillations [9].
To understand some of these results let us apply the resonant tunneling approximation. It can be evaluated analytically for the degenerate case
which we will consider from now on. First we note that due to spin conservation the reduced density matrix of the dot is diagonal once it is diagonal
at the initial time. The solution is identical to the one for the noninteracting dot, given by (54) and (55) for the transition rates and (57) and (58)
for the tunneling current, but with an important change of the definition
of Ir (w) = 2,t (w) + ,; (w). The golden rule tunneling "in" rate It has to
be multiplied with a factor 2 since there are 2 possibilities for an electron
to tunnel onto the dot. Adding the golden rule tunneling" out" rate I;'
we obtain Ir which is an estimate for the inverse finite life-time of the dot
excitation. This is expressed by the imaginary part of cr(w) which is given
by Ima(w) = -lr 2:r Ir(w) = -~ 2:r f r (w)(l + fr(w)). We see that the
broadening depends now on the Fermi functions and is therefore energy,
temperature and voltage dependent. When energy increases the broadening decreases, i.e. we expect quantum fluctuations to become weaker if we
mcrease f.
From the Kramers-Kronig relation we have necessarily also a renormalization which is given by the real part of cr(w). We obtain
Recr(w)
=
f3 E)
c
. [1jJ( -1 + 2
2lr
f3 - - Re1jJ( -1
+.z-(Vr
2
2lr
w))
- Vr 1
+ lr w2Ec
,
(64)
where 1jJ is the digamma function and we have chosen a lorentzian form
for fr(w) = fr Ebl((w - Vr )2 + Eb). The cut-off will be of the order of
the charging energy Ec since we do not allow for two electron to tunnel
322
onto the dot. For T ~ IVr - wi ~ Ee, the renormalization depends logarithmically on energy Reo-(w) ~ 1/(2rr) Lr rr In (Ee/IVr - wI). This leads
to a logarithmic increase of the renormalization when w approaches the
effective potentials eVr of the reservoirs. As a consequence the transmission
probability (58) has a maximum near w ~ eVr since there is a solution of
w - E - Reo-(w) = 0 near these values. This indicates the occurence of the
Kondo resonance and explains the splitting when the potentials Vr are not
equal (see inset of Fig. 8a).
To illustrate the consequences for the current let us start with the incoherent limit T ~ r. In this case we can neglect the renormalization and
the transmission probability is a sharp function around w E. Neglecting
the energy dependence of r r (w) we can replace the transmission probability
in formula (57) by Trr/(w) -+ -2rrrrrrl/Imo-(E)o(w - E) This gives for the
conductance in linear response
f'.J
(65)
As expected the line shape is assymmetric since the broadening of E depends on VD. This result shows a clear difference to the noninteracting case
where the line shape is symmetric. It shows up already in the high temperature regime and can be calculated also from the golden rule approach.
The assymmetry was first predicted by Beenakker [30] but has never been
identified experimentally.
In the coherent regime T ~ r, the real part of o-(w) becomes important.
As already mentioned above, the resonance of the transmission probability at the Fermi levels is only significant for Vr - E
r since the Kondo
temperature depends exponentially on Vr - E. In this regime the relevant
energy scale for the onset of quantum fluctuations is r. In Fig. 8 we show
the differential cond uctance G = dI/ dV (I = I R = - h) as function of the
bias voltage V = VL - VR for E - f1
-rand E
f1. Thereby we have
chosen VL = - VR = V/2 and used eVD = e Li=L,R,g(Ci/C)"Vi with symmetric capacitances CL = CR. ':'his gives eVD = (Cg/C)eVg independent
of the bias voltage. For a low lying level a pronounced zero bias maximum
is developed which is due to the fact that the Kondo resonances of the
transmission probability at w = Vr , r = L, R, are split by the bias voltage
and decrease in magnitude (see inset of Fig. 8a). In contrast, for E + VD
near the electrochemical potentials of the reservoirs, a zero bias minimum
is observed although the Kondo resonances are absent. This is due to the
fact that the nontrivial structure of the real part of o-(w) is still present and
influences the differential conductance always for T ~ r independent of
whether the transmission probability shows Kondo resonances or not. The
striking difference of the zero-bias anomaly for different values of VD or
f'.J
f'.J
f'.J
323
G[e2/h]
0.6
0.2r-~~---,1
TIE)
"MJ
0.5
0.5
0.4
0.1
0.3
0.4
-0.5
0.0
0,5
f
r
1.
0.2
0.1
0.0
0,3
0.2
0.5
0.2 '--~-'-~-'--~-'-~----'
-0.50
-0.25
0.00
0.25 eV 0.50
(a)
0.4
-1.0
-0.5
0.0
(b)
f
0.5
eV 1.0
-
Figure 8.
(a) The differential conductance vs. bias voltage for r L = rR = r/2,
T = O.Olr, I-l = 0, t = -4r and Ee = 100r. The curve shows a maximum at zero bias.
Inset: increasing voltage leads to an overall decrease of the transmission probability in
the range lEI < eV. (b) The differential conductance vs. bias voltage for rL = rR = r/2,
T = o.osr, VD = 0, t = 0 and Ee = 100r. The curve shows a minimum at zero bias.
Inset: increasing voltage leads to an overall increase of the transmission probability in
the range lEI < eV.
Vg motivates an interesting experiment which can only be performed with

devices where the effective positions of the dot excitations can be varied by
an external gate voltage.
4.4. METALLIC ISLAND
In this section we study the resonant tunneling approximation for the

infinite-Z metallic island. In addition, we assume that only one excitation
energy t1N = EN+l - EN with EN = EcN2 lies within the relevant energy
range of the effective potentials eVr = eVr - eVD of the reservoirs. This
means that the charging energy Ec is assumed to be much larger than
temperature and bias voltage so that the other excitations are irrelevant.
Without loss of generality we can set N = O.
In the absence of time-dependent fields and the heat bath, the Hamiltonian follows from (15) and (16)
H(t) =
t
(Tkl(t)akralDPOl
A
r=L,R,D;k
r=L,R;kl
+ h.c.) ,
(66)
where the time dependence of the coupling constants is only due to the
324
static voltages Tkl(t) = Tkl exp (ieVr(t - to)). This Hamiltonian looks very
similiar to (53) where we considered a quantum dot with one excitation energy. However, the important difference here is that the vertex akralD leads
to bosonic contractions in the infinite-Z case whereas in (53) we had to deal
with fermionic contractions. Therefore, the resonant tunneling approximation does not turn out to be exact here. We obtain the same solution as
in the fermionic case but with the replacements rr(w) -+ 2Dr (w - eVr ),
if -+ n~, "Y;' -+ a~) and "Yr -+ ar = at + a;, where Dr (w) = 7ra ow and
a~ = (l/7r)Dr(w - eVrO)n~(w) follow from (33) and (40).
The tunneling current is given by (57) but the Fermi functions in this
expression are replaced by Bose distributions. Using (58) together with the
above mentioned replacements, and performing some elementary manipulations, we can rewrite the current as Ir = elh
f dwT:;',(w) Ur(w) - ir'(w)]
where T:;', is the transmission probability between the original Fermi reservOlrs
I>
Trr,(w) =
47r
ar(w)ar,(w)
(w _ ~o _ ReO"(w))2 + (ImO"(w))2 .
(67)
Renormalization and broadening effects are described by the real and

imaginary part of O"(w) = f dw'L-r ar(w')/(w - w' + iTJ). We see that in
contrast to the fermionic case the bosonic distribution functions n~ occuring in a~ do not cancel in the sum ar. Like in the quantum dot case with
two excitations, this gives rise to a broadening which depends on energy,
temperature and voltage, and via Kramers Kronig to a nontrivial renormalization. Explicitly, we get ImO"(w) = - L-r Dr(w - eVr)(l +2n r (w)) and
ReO"(w)
=
(68)
where we have chosen a lorentzian form for Dr(w) = 7ra owEb/(w 2 + Eb).
aD is the dimensionless conductance of barrier r defined after (33). For
very low temperatures we get ImO"(w) ~ -7r L-r aolw - eVrl and ReO"(w) ~
- 2 L-r aD (w - Vr ) In Ec I Iw - Vr I. The broadening is proportional to energy
since the number of available states for tunneling on or off the island is
also proportional to energy. In contrast to the interacting quantum dot in
the previous section, the renormalization is zero for w '" Vr Therefore, no
additional resonances occur here for the transmission probability but we
still have a logarithmic shift of the excitation energy ~o.
The renormalization of ~o is determined by finding the maximum of the
transmission probability (67) which is approximately determined by solving
the self-consistent equation Ao = ~o + ReO"(Ao). In a first approximation
325
we use Lio for the value of w inside the 7/'-function of the real part of a
given by (68). For w ~ Ee we obtain w - ~o - Rea(w) = Z-l(w - Lio)
with the renormalization factor Z defined by
Z-l = 1 +
L a~[7/'( 27rT
Ee ) + 7/'(1 + Ee ) _ 2Re7/' (i ILio - eVrl)] .
27rT
27rT
(69)
Within this approximation the transmission probability reads

(70)
where &r and (j are defined as before but multiplied with Z. This can be
interpreted as a renormalization of the dimensionless conductance &0 =
Zao. What we mean by renormalization becomes clear when we neglect
the broadening in (70) which is described by the imaginary part of (j. This
is justified if &0 ~ 1. We obtain
F ( )
Trr'
W
'"
'"
7r
2 &r
(Lio)&r'(Lio)
i'( _
_
U W
I::r ar(~O)
A )
LlO.
(71)
This is precisely the golden rule result for the transmission probability but
with renormalized parameters.
In certain limits we can estimate the renormalized parameters. We take
VL = - VR = V, VD = 0 (otherwise one has to shift the excitation energy
~o by VD), and a~ = a~ = ao. If one of the energy parameters Li o, T,
or eV is large compared to the other two ones but small compared to the
charging energy, we obtain for the renormalization factor
Z=
1 + 4ao In (Eelmax(l~ol, 27rT, leVI/2))
(72)
We note that ao is the dimensionless conductances of a single barrier. For

the derivation we have used the asymptotic expansion 7/'(z) = In (z), valid
for Izl -+ 00. The renormalized parameters follow from ~o = Z~O and
&0 = Za o These equations agree with the renormalization group results
performed for the equilibrium case Vr = 0 [48]. This shows that the leading logarithmic terms are included within the resonant tunneling approximation. However, we have achieved more than renormalization group here
since we do not need all the approximative steps used so far. We can handle
all intermediate regimes for the three energy parameters described before
and can account for the broadening of the charge excitations by not neglecting the imaginary part of (j in (70). The latter can be estimated to
be of the order hiT'" Ima(Lio). Within the same limits discussed before
326
this gives hiT '" 7r&omax(ILi ol, 2T, leVI/2) (compare with the discussion
in section 2.1). We see that broadening effects start to become important for no > 0.1 which means that they can only be enhanced by lowering the tunneling barriers. Renormalization effects become significant for
max(ILiol, 27rT, leVI/2) < ECe-1/(2o:o). This means that they can be enhanced either by lowering the tunneling barriers or by lowering all the
other energy parameters.
Let us demonstrate the influence of quantum fluctuations on the differential conductance as function of the gate voltage. Again we set VL =
-VR = V/2 and a& = a~ = ao. We study G = dIldY, with 1= IR = -1,
as function of D.o and set VD = 0 (equivalently we could study G as function of eVD = (Cg/C)eVg and keep D.o fixed). We insert the transmission
probability (70) including the broadening into the current formula. Using
the result (72) for the renormalization factor, we find in the two limits
T leVI and leVI T that the differential conductance at D.o = Lio = 0
is given by
2G(D.o = O)RT
= "2 = 21 + 4ao In (Ec/max(27rT, leVI/2)) ,
(73)
where RT = R!t = R is the resistance of a single barrier. The golden

rule result is 2G(D.o = O)RT = 1/2 and corresponds to 1/2 of the ohmic
resistance since all the other excitation energies D.N (N i=- 0) are suppressed
by the Coulomb blockade. We see that due to quantum fluctuations, the
differential conductance is no longer a constant at the symmetry point
but decreases logarithmically with bias voltage or temperature. We note
that the qualitative effect of quantum fluctuations is again a suppression
of the differential conductance like it was the case for quantum dots. It
is not suprising that the differential conductance for the noninteracting
quantum dot case saturates at low temperatures whereas it decreases for
the metallic island since the temperature dependence of the golden rule
results are already different for the two cases.
The broadening of the line shape of the differential conductance can be
estimated by noting that the integral of G(D.o) over D.o is not influenced
by quantum fluctuations and can be calculated to be f dD.oG(D.o)RT =
(l/3)leVI for T leVI, and f dD.oG(D.o)RT = (7r 2/8)T for leVI T.
Together with the value at the symmetry point we conclude that quantum
fluctuations lead to a broadening that increases logarithmically with bias
voltage or temperature if we measure D.o in units of leVI or T.
Both features, the logarithmic decrease of G(D.o = 0) and the logarithmic increase of the broadening with bias voltage or temperature is demonstrated in Fig. 9. In Fig. 9b we furthermore observe a splitting of the resonance due to nonequilibrium effects and a shift of the individual resonances
327
0.5
1.0
GR T
GRT
0.4
0.8
0.3
0.6
0.2
0.4
0.1
0.2
-5.0
0.0
(a)
5.0
d, 10.0
Q"
0.0
(b)
1.0
~ 2.0
eV
Figure 9. (a) The differential conductance in linear response (V =0) for the metallic
island as a funciton of the excitation energy ~o normalized to the temperature kBT with
at = a{; = 0.05 and (1) kBT/Ee = 0.1, (2) kBT/Ee = om, (3) kBT/Ee = 0.001. For
comparison, (0) shows the golden rule result, which is independent of the temperature
kBT. (b) The differential conductance in nonlinear response for the metallic island as
a function of the excitation energy ~o normalized to the transport voltage e V with
at = a{; = 0.05, T = 0 and (1) eV/ Ee = 0.1, (2) eV/ Ee = 0.01, (3) eV/ Ee = 0.001.
For comparison, (0) shows the golden rule result, which is independent of the transport
voltage eV.
due to quantum fluctuations. The logarithmic decrease of G(~o = 0) has

been observed experimentally [10] with a good fit to the theoretical predictions [49].
5. Conclusions
Within this paper we have analysed a very fundamental problem of statistical mechanics, namely the interaction between a large environment and
a small mesoscopic system. To describe experimentally realizable systems,
we concentrated on particle exchange via tunneling and energy exchange
by considering a fluctuating voltage. For the environment we have chosen
metallic electronic reservoirs with different electrochemical potentials, and
the system was realized by a strongly interacting quantum dot. Macroscopic
systems being in equilibrium with large particle reservoirs are described by
a grandcanonical ensemble. For a mesoscopic system we have to consider
the following new aspects. First, the energy scale associated with the coupling between system and environment can be so large that quantum fluctuations lead to a complete deviation from a grandcanonical ensemble. In
macroscopic systems, the coupling to the environment is always a surface
effect which can be negleted in the thermodynamic limit. Second, the en-
328
ergy scale characterizing the distance between the one-particle excitation

energies of the system can be so large that the discreteness of the density of
states becomes visible on experimentally controllable voltage scales. This
demands the consideration of finite size effects and strong capacitive interactions. Third, the nonequilibrium stationary state induced by different
electrochemical potentials on the reservoirs can no longer be described by
a local equilibrium distribution. Therefore, we have aimed at presenting
a nonequilibrium theory which provides a nonperturbative analysis in the
coupling between an environment and a strongly correlated finite system.
We have demonstrated that the measurement of the differential conductance G as function of the gate voltage Vg or the bias voltage V can
reveal all aspects decribed above. The discreteness of the dot excitation
spectrum leads to resonances in G(Vg) separated by the sum of level spacing and charging energy. Strong nonequilibrium effects can be observed by
comparing the line shape of an individual peak for V = 0 and T = O.
E.g. for a metallic island we have shown these two cases in Fig. 9 where a
splitting of the resonance occurs at finite bias voltage. Quantum fluctuations set on by lowering temperature or increasing tunneling. Without spin
degeneracies, they lead to a renormalization and broadening of the excitation energies of the island. Whereas for noninteracting systems the effects
on G(Vg) are already well-known from Landauer-Biittiker theory, the presence of interactions can lead to an anomalous temperature dependence of
height, broadening or position of the resonances. This is demonstrated in
Fig 9 for the case of metallic islands. For spin degenerate excitation energies quantum fluctuations can create zero bias anomalies of G(V) at fixed
gate voltage as is demonstrated in Fig. 8. They can occur in the form of
zero bias maxima or minima dependent on the postion of the excitation
energies relative to the electrochemical potentials of the reservoirs. Due
to the enormous variety of possible arrangements of island systems and
the experimental progress in realizing such devices, we expect that future
research will reveal many more motivations for studying quantum fluctuations induced by strong coupling between mesoscopic systems and particle
reserVOIrs.
6. Acknowledgements
It is a pleasure to thank all my collaborators C. Bruder, R. Fazio, M.

Hettler, J. Konig, J. Schmid and G. Schon. This work was supported by the
Deutsche Forschungsgemeinschaft through Sonderforschungsbereich 195.
References
1.
D.V. Averin and K.K. Likharev, in Mesoscopic Phenomena in solids, edited by B.
329
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
Altshuler, P.A. Lee, and RA. Webb (North-Holland, Amsterdam, 1991).

P.L. McEuen and L.P. Kouwenhoven, to be published in Nano-Science and Technology, G. Timp ed. (AlP Press, New York).
M. A. Kastner, Rev. Mod. Phys. 64, 849 (1992); Physics Today 46, 24 (1993).
H. Grabert and M.H. Devoret, Single Charge Tunneling, NATO ASI Series, Vol.294
(New York, Plenum Press 1992).
G. Schon, Single-Electron Tunneling, to be published in Quantum Transport and
Dissipation, Chapter 2, eds. T. Dittrich, P. Hiinggi, G. Ingold, B. Kramer, G. Schon,
and W. Zwerger, (VCH Publishers).
Z.Phys.B-Condensed Matter 85.
M. Biittiker, Phys. Rev. Lett. 57, 1761 (1986); Phys. Rev. B46, 12485 (1992).
D.V. Averin and Y.V. Nazarov, Phys. Rev. Lett. 65 (1990) 2446; ibid. in Ref. [4].
C. Bruder, R Fazio, and H. Schoeller, Phys. Rev. Lett. 76, 114 (1995).
P. Joyez, V. Bouchiat, D. Esteve, C. Urbina, and M.H. Devoret, submitted to Phys.
Rev. Lett.
S. Tarucha, D.G. Austing, T. Honda, RJ. van der Hage, and L.P. Kouwenhoven,
Phys. Rev. Lett. 77, 3613 (1996).
D.C. Ralph, C.T. Black and M. Tinkham, Phys. Rev. Lett. 74, 3241 (1995); D.C.
Ralph, C.T. Black, M. Tinkham, and RC. Tiberio, preprint, submitted to Science.
D.L. Klein, P.L. McEuen, J.E. Bowen Katari, R Roth, A.P. Alivisatos, Appl. Phys.
Lett. 68, 2574 (1996).
H. Schoeller and G. Schon, Phys. Rev. B50, 18436 (1994).
J. Konig, H. Schoeller and G. Schon, Europhys. Lett. 31, 31 (1995).
J. Konig, H. Schoeller and G. Schon, Phys. Rev. Lett. 76, 1715 (1996).
J. Konig, J. Schmid, H. Schoeller, and G. Schon, Phys. Rev. B54, 16820 (1996).
RP. Feynman and F.L. Vernon, Ann. Phys. (N.Y.) 24 (1963) 118.
A. O. Caldeira and A. J. Leggett, Physica A 121, 587 (1983); Ann. Phys. (N.Y.)
149, 374 (1983); 153, 445(E) (1983).
U. Weiss, Quantum Dissipative Systems, Series in Modern Condensed Matter
Physics, Vol. 2 (World Scientific, 1993).
U. Eckern, G. Schon and V. Ambegaokar, Phys. Rev. B30, 6419 (1984); G. Schon
and A.D. Zaikin, Phys. Rep. 198, 237 (1990).
M. Biittiker, IBM J.Res. Dev. 32, 317 (1988).
D. Pfannkuche and S.E. Ulloa, Phys. Rev. Lett. 74, 1194 (1995).
K. Jauregui, W. Hausler, D. Weinmann, and B. Kramer, Phys. Rev. 53 Rl713
(1996).
N.S. Wingreen, K.W. Jacobsen, and J.W. Wilkins, Phys. Rev. Lett. 61,1396 (1988);
Phys. Rev. B40, 11834 (1989); L.I. Glazman and RI. Shekhter, Zh. Eksp. Teor. Fiz.
94, 292 (1988 [sov. Phys. JETP 67, 163 (1988)]; M. Jonson, Phys. Rev. B39, 5924
(1989).
M.H. Devoret, D. Esteve, H. Grabert, G.-L. Ingold, H. Pothier, and C. Urbina,
Phys. Rev. Lett. 64, 1824 (1990); A.A. Odintsov, G. Falci, and G. Schon, Phys.
Rev. B44, 13089 (1991); K. Flensberg, S.M. Girvin, M. Jonson, D.R Penn, and
M.D. Stiles, Phys. Scripta T42, 189 (1992); H.T. Imam, V.V. Ponomarenko, and
D.V. Averin, Phys. Rev B50, 18288 (1994).
G. Ingold and Yu.V. Nazarov, in Ref. [4].
G.D. Mahan, Many-Particle Physics, (Plenum, 1990).
P.K. Tien and J.R Gordon, Phys. Rev. 129,647 (1963).
C.W.J. Beenakker, Phys. Rev. B44, 1646 (1991).
Y. Meir, N.S. Wingreen, and P.A. Lee, Phys. Rev. Lett. 66, 3048 (1991).
1.0. Kulik and R.L Schekhter, Sov. Phys. JETP 41 (1975) 308; L.I. Glazman and
K.A. Matveev, Pis'ma Zh. Eksp. Teor. Fiz. 48, 403 (1988), [JETP Lett 48, 445
(1988)]; H. van Houten, C.W.J. Beenakker, and A.A.M. Staring, in Ref.[4]; D.V.
Averin, A.N. Korotkov, and K.K. Likharev, Phys. Rev. B44, 6199 (1991).
D. Weinmann, W. Hausler, W. Pfaff, B. Kramer, and U. Weiss, Europhys. Lett. 26,
330
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
467 (1994); W. Hausler, Ann. Physik 5, 401 (1996).

L.P. Kouwenhoven, S. Jauhar, K. McCormick, D. Dixon, P.L. McEuen, Yu. V.
Nazarov, N.C. van der Vaart, and C.T. Foxon, Phys. Rev. B50, 2019 (1994); I.A.
Devyatov and K.K. Likharev, Physica B194-196, 1341 (1994).
C.A. Stafford, Phys. Rev. Lett. 77, 2770 (1996).
R. Landauer, IBM J. Res. Dev. 1,233 (1957);
C. Caroli, R. Combescot, P. Nozieres and D. Saint-James, J.Phys. C4, 916 (1971);
ibid. J.Phys. C5, 21 (1972); N.S. Wingreen, A.P. Jauho, and Y. Meir, Phys. Rev.
B48, 8487 (1993).
N.E. Bickers, Rev. Mod. Phys. 59, 845 (1987); A.C. Hewson, The Kondo Problem
to Heavy Fermions, (Cambridge University Press, 1993).
J. Appelbaum, Phys. Rev. Lett. 17,91 (1966); A.G.M. Jansen, A.P. van Gelder, P.
Wyder, and S. Strassler, J. Phys. C13, 6073 (1980); A.M. Duif, A.G. Jansen, and
P. Wyder, J.Phys.: Condo Matt. 1, 3157 (1989); E.L. Wolf, Principles of Electron
Tunneling Spectroscopy (Oxford Univ. Press, New York, 1989), Chap. 8.
T.K. Ng and P.A. Lee, Phys. Rev. Lett. 61, 1768 (1988); L.I. Glazman and M.E.
Raikh, Pis'ma Zh. Eksp. Teor. Fiz. 47, 378 (1988) [JETP Lett. 47, 452 (1988)]; A.
Kawabata, J. Phys. Soc. Japan 60, 3222 (1991).
S. Hershfield, J.H. Davies, and J.W. Wilkins, Phys. Rev. Lett. 67, 3720 (1991);
Phys. Rev. B46, 7046 (1992); T.K. Ng, Phys. Rev. Lett. 70, 3635 (1993); A.L.
Yeyati, A. Martin-Rodero, and F. Flores, Phys. Rev. Lett. 71, 2991 (1993).
Y. Meir, N.S. Wingreen, and P.A. Lee, Phys. Rev. Lett. 70, 2601 (1993); N.S.
Wingreen and Y. Meir, Phys. Rev. B49, 11040 (1994).
D.C. Ralph and R.A. Buhrman, Phys. Rev. Lett. 72, 3401 (1994).
N. Sivan and N.S. Wingreen, Phys. Rev. B54, 11622 (1996).
M.H. Hettler and H. Schoeller, Phys. Rev. Lett. 74, 4907 (1995).
T.K. Ng, Phys. Rev. Lett. 76, 487 (1996).
K.A. Matveev, Sov. Phys. JETP 72, 892 (1991); G. Falci, G. Schon and G.T. Zimanyi, Phys. Rev. Lett. 74, 3257 (1995).
J. Konig, H. Schoeller, and G. Schon, submitted to Phys. Rev. Lett., condmat/9702196.
TRANSPORT IN A ONE-DIMENSIONAL
LUTTINGER LIQUID
MATTHEW P. A. FISHER
Institute for Theoretical Physics, University of California,

Santa Barbara, CA 93106-4030
AND
LEONID I. GLAZMAN
Theoretical Physics Institute, University of Minnesota,

Minneapolis, MN 55455
Abstract. In this paper we review recent theoretical results for transport

in a one-dimensional (ld) Luttinger liquid. For simplicity, we ignore electron spin, and focus exclusively on the case of a single-mode. Moreover,
we consider only the effects of a single (or perhaps several) spatially localized impurities. Even with these restrictions, the predicted behavior is very
rich, and strikingly different than for a 1d non-interacting electron gas. The
method of bosonization is reviewed, with an emphasis on physical motivation, rather than mathematical rigor. Transport through a single impurity
is reviewed from several different perspectives, as a pinned strongly interacting "Wigner" crystal and in the limit of weak interactions. The existence
of fractionally charged quasiparticles is also revealed. Inter-edge tunnelling
in the quantum Hall effect, and charge fluctuations in a quantum dot under the conditions of Coulomb blockade are considered as examples of the
developed techniques.
1. Introduction
Landau's Fermi liquid theory is a beautiful and successful theory which accounts for the behavior of the conduction electrons in conventional metallic
systems [1]. The central assumption of this theory, is that the low energy excited states of the interacting electron gas can be classified in the same way
as a reference non-interacting electron gas. Upon adiabatically "switching
331
332
on" the electron interactions, the excited states are still composed of free
quasiparticles, labelled by momentum, which obey Fermi statistics. The low
energy excitations above the ground state, consist of creating quasiparticles (and quasiholes) with momenta just above (below) the Fermi surface.
The single-particle spectral function for an electron is presumed to have a
6-function peak at the Fermi surface, A(kF' E) = 21fZF6(E), with a non-zero
coefficient, ZF. For the non-interacting electron gas ZF = 1, but with interactions Z F < 1, reflecting a smaller overlap between the electron and the
quasiparticle. However, provided ZF is non-zero, there exist well defined
low energy quasiparticle excitations, which are in a one-to-one correspondence with the bare electron excitations of the reference non-interacting
electron gas. It is then possible to build a transport theory of Fermi liquids
in direct analogy with Drude theory of the free electron gas, by focussing
on the scattering (by impurities, say) of the low energy quasiparticle excitations. For example, a quasiparticle will scatter off an impurity with a
finite cross-section, leading to a metallic resisitivity for low impurity concentration.
For a weakly-interacting electron gas, deviations in the spectral weight
ZF from one can be computed perturbatively in the interaction strength,
provided the spatial dimensionality, d, is greater than one. However, for
the one-dimensional interacting system, low order terms in the perturbation theory are divergent. This signals the breakdown of Landau's Fermi
liquid theory in the Id electron gas, for arbitrarily weak interactions [2].
This divergence can be traced to the large phase space in Id, for an electron to relax by creating an electron-hole excitation. For d > 1, both energy
and momentum conservation highly constrain the available phase space, but
in Id for a single branch with linear dispersion, momentum conservation
automatically implies energy conservation, so the phase space is less constrained. This leads to a divergent rate for such scattering processes, and
the electron becomes "dressed" by a large number of electron-hole pairs.
This in turn drives the electron spectral weight to zero, ZF = 0, even for
weak electron interactions. In the limit of weak interaction, it is still possible to extract the resulting properties of the 1d electron gas by a clever
resummation of the most divergent terms in each order of perturbation
theory [3J. However, the physics is more readily revealed by the method of
bosonization, as discussed below.
Tomonaga [4J (and more recently Luttinger [5]) considered a special
class of interacting 1d electron models, which were linearized around the
Fermi points, and could be diagonalized in terms of boson variables. These
boson variables described collective plasmon excitations, or density waves,
in the electron gas. Some years later, Haldane [6J argued that this bosonized
decription was valid more generally, giving an appropriate description for
333
the low energy excitations of a generic Id interacting electron gas. Haldane

coined the term "Luttinger liquid" to describe this generic behavior, although sometimes it is referred to as a Tomonaga-Luttinger liquid. In such
a Id Luttinger liquid, creation of a real electron is achieved by exciting an
infinite number of plasmons. Because of this, the space and time dependence of the electron correlation function is dramatically different than in
a non-interacting electron gas, or in a Fermi liquid [7J. This manifests itself
in various kinetic quantities, in particular the Drude conductivity, which is
predicted to vary (as a power law) with temperature [8, 9, 10J. The Luttinger liquid approach to transport in Id systems was popular in the 1970's,
being employed in attempts to understand the behavior of quasi-ld organic
conductors.
Recent advances in semiconductor technology have renewed interest in
transport of Id electron systems. By cleverly "gating" a high mobility twodimensional (2d) electron gas, it is possible to further confine the motion
of the electrons, so that they can only move freely along a Id channel.
Moreover, it is possible to make a very narrow channel, with width comparable to the electron's Fermi wavelength. In this case, the transverse
degrees of freedom are quantized, and only one, or perhaps several, Id
channels are occupied at the Fermi level. This "quantum wire" provides
an experimental realization of a Id electron gas, which should reveal the
signatures of Luttinger liquid behavior without the complicated crossovers
to three-dimensional behavior inherent in the quasi-ld organic conductors.
Edge states formed at the boundaries of the 2d electron gas when placed
in strong magnetic field in the quantum Hall regime, provide another important example of a one-dimensional electron system [11]. These edge excitations are chiral, moving only in one direction along the edge. For a
quantum Hall bar geometry, the two edge modes confined to the edges
move in opposite directions. Together they comprise a non-chiral system,
which resembles a Id electron gas. For the integer quantum Hall effect,
these edge excitations are equivalent to a Id non-interacting electron gas,
(despite the presence of electron interactions). However, in the fractional
quantum Hall effect, the edge excitations are believed to be isomorphic to
a Id interacting Luttinger liquid [12]. Moreover, due to the spatial separation between the modes on opposite edges, impurities cannot backscatter,
so that localization effects important in a quantum wire are absent here.
A Id quantum wire is appropriately characterized by a conductance,
which can be measured in a transport experiment. In the absence of interactions, the conductance of an ideal single-mode channel wire, adiabatically
connected [13J to leads, is quantized, G = 2e 2 j h - where the factor of 2
accounts for spin. If a scatterer is introduced into the channel, the conductance drops, with G = 2Te 2 jh - where T < 1 is the transmission coefficient
334
for electrons at the Fermi level. For edge states in the integer quantum Hall
effect, a quantized conductance occurs when there is no backscattering between modes on opposite edges, and corresponds to the quantized Hall
conductance. Inter-edge scattering is equivalent to a reduced transmission
coefficient, and reduces the conductance.
As discussed above, electron interactions should dramatically modify
the low energy excitations in the quantum wire. As we shall see, this leads to
striking predictions for the transport in a quantum wire with one or several
impurities. Likewise, inter-edge backscattering in the fractional quantum
Hall effect is predicted to be very different than in the integer Hall effect.
In this paper we review recent theoretical results for transport in a ld
Luttinger liquid. For simplicity, we ignore electron spin, and focus exclusively on the case of a single-mode. Moreover, we consider only the effects
of a single (or perhaps several) spatially localized impurities. As we shall
see, even with these restrictions, the predicted behavior is very rich, and
strikingly different than for a ld non-interacting electron gas.
The paper is organized as follows. In Section II the method of bosonization is reviewed, with an emphasis on physical motivation, rather than
mathematical rigor. After discussing briefly the conductance of an ideal
channel, we reveal the existence of fractionally charged excitations in the
Luttinger liquid. These correspond to Laughlin quasiparticles for fractional
quantum Hall edge states, but should also be present in quantum wires. In
Section 3, we consider the tunnelling density of states for adding an electron
into a ld Luttinger liquid, which is suppressed, vanishing as a power law
of energy. The exponent is extracted for both tunnelling into the middle
of an infinite ld wire, and into the "end" of a semi-infinite wire. Section
4 is devoted to a detailed analysis of transport in a Luttinger liquid with
a single barrier or impurity. We consider two limiting cases, a very large
barrier in IVA and a very small barrier in IVB. In Section IVC we show how
the crossover between these two limits can be understood by considering
general barrier strengths, but weak electron interactions. A general picture of this crossover is described in IVD, and the special case of resonant
tunnelling is discussed in IVE.
In Section V, we consider briefly two particular applications. In VA
we consider tunnelling between edge states in the fractional quantum Hall
effect, which is a rather straightforward application of the general theory.
In Section VB we employ Luttinger liquid theory to describe Coulomb
blockade in a quantum dot. Section VI is devoted to a very brief summary,
and list of other related topics.
335
2. Bosonization
Consider the Hamiltonian for spinless electrons hopping on a 1d lattice,
1-
-t
2:. c}Cj+1 + h.c. + V2 2:. C}CjC}+1 CHI,

J
(1)
with hopping strength t and near-neighbor interaction strength V. When

V = 0 one can diagonalize the problem in terms of plane waves with energy Ek = -tcos(k) for momentum k satisfying Ikl < 7r. The low energy
excitations consist of particle/hole excitations across the two Fermi points,
at kp. Consider a single particle hole/excitation about the right Fermi
point, where a single electron is removed from a state with k < kp and
placed into an unoccupied state with k + q > kp. For small momentum
change q, the energy of this excitation is Wk = vpq, with vp the Fermi
velocity. Together with the negative momentum excitations about the left
Fermi point, this linear dispersion relation is identical to that for phonons in
one-dimension. The method of Bosonization exploits this similarity by introducing a phonon displacement field, (} to decribe this linearly dispersing
density wave. When interactions are turned on, the single particle nature of
the excitations is not retained, but this linearly dispersing mode survives,
with a renormalized velocity.
The method of Bosonization focusses on the low energy excitations,
and provides an effective theory. We follow the heuristic development of
Haldane [6], which reveals the important physics, dispensing with mathematical rigor. To this end, consider a Jordan-Wigner transformation which
replaces the electron operator, Cj, by a (hard-core) boson operator, Cj =
exp( i7r Lj<i ni)b j , where nj = c}Cj is the number operator. One can easily verify that the Bose operators commute at different sites. Moreover, the
above Hamiltonian when re-written is of the identical form with c's replaced
by b's. This transformation, exchanging Fermions for Bosons, is a special
feature of one-dimension. The Boson operators can be (approximately) decomposed in terms of an amplitude and a phase, b -+ .,j'i'fj exp( i</>j). We now
imagine passing to the continuum limit, focussing on scales long compared
to the lattice spacing. In this limit we replace, </>j -+ </>( x) and nj -+ p( x),
where p(x) is the 1d electron density. This can be decomposed as p = Po+p,
where the mean density, Po = kp/7r, and p is an operator measuring fluctuations in the density. As usual, the density and phase are canonically
conjugate quantum variables, taken to satisfy
[</>(x),p(x')]
= i8(x -
x').
(2)
Now we introduce a phonon-like displacement field, (}(x), via p(x) =

ox(}(x)/7r. The factor of 7r has been chosen so that the full density takes the
336
simple form: p(x) = (kF

satisfied if one takes,
+ oxO)/1f.
[(X),O(X')]
The above commutation relations are
= ~; sgn(x -
x').
(3)
Notice that ox is the momentum conjugate to O. The effective (Bosonized)

Hamiltonian density which describes the 1d density wave takes the form:
H
V
2
2
= 21f[g(Ox)
+ g-1 (oxO)].
(4)
This Hamiltonian describes a wave propagating at velocity v, as can be

readily verified upon using the commutation relations to obtain the equations of motion, olo = v 2 o;O, and similarly for . For non-interacting electrons one should equate v with the (bare) Fermi velocity, VF, but with
interactions v will be modified (see below). The additional dimensionless
parameter, g, in the above Hamiltonian also depends on the interaction
strength. For non-interacting electrons one can argue that g = 1, as follows.
A small variation in density, p will lead to a change in energy, E = p2/2K"
where K, = op/op, is the compressibility. Since oxO = 1fP, one deduces from
H that K, = g/1fv. But for a non-interacting electron gas, 1fVK, = 1, so that
g=1.
To see why g shifts with interactions, it is convenient first to identify the

electron creation operator. On physical grounds, the appropriate operator,
denoted 'ljJt(x), should create a unit charge (e) excitation at point x, which
anti-commutes with the same operator at a different point x'. Consider the
Bose creation operator, bt '" exp(i). To see that this indeed creates a unit
charge (e) excitation, one can write,
iA.(x)
e'f'
= ei7r JX
-00
dx' P(x') ,
(5)
where P = ox/1f is the momentum conjugate to O. Since the momentum

operator is the generator of translations (in 0), this creates a kink in 0
of height 1f centered at position x. This corresponds to a localized unit of
charge, since the density p = oxO / 1f.
To construct an electron operator requires multiplying the Bose operator
ei by a Jordan-Wigner "string": ei7r~j<ini --+ e i7r p. Rewriting in terms
of 0, this corresponds to ei(kpx+O). The most general Fermionic charge e
operator can thus be written,
'ljJ(x)
L
modd
'ljJm ~
L
modd
eim(kpx+O(x))ei(x).
(6)
337
The requirement that the integer m is odd is dictated by Fermi statistics,

since one can show using the commutation relations, that
(7)
The terms with m = 1 have a very simple interpretation. The electron
field operator has been expanded into a right and left moving piece, corresponding to the right and left Fermi points at kF :
(8)
with the definition, ipR/L = e. Here ipR/L describe the slowly varing
piece of the electron field. These two fields commute with one another, and
satisfy:
Notice that these unusual commutation relations - referred to as KacMoody - imply that the field conjugate to ip is the derivative of ip itself.
These fields are simply related to the right and left moving electron densities, denoted N R/ L, via
(10)
Note that these densities sum to give the total density, NR + NL = p.
It is instructive to re-write the Hamiltonian (4) in terms of the right
and left electron densities,
(11)
with
Vo = v (9
+ 9-1) /2
.\ = (1 - 92 )/(1 + 92 ).
(12)
This Hamiltonian descibes the system of right- and left-moving electrons

with the interaction strength .\ between the two species. Notice that for
9 = 1 the interaction vanishes, and Vo = v; Hamiltonian (11) then gives the
Bosonized decsription of the non-interacting electron gas, and Vo == VF, the
bare Fermi velocity. Repulsive interactions, (.\ > 0) correspond to 9 < 1,
whereas attractive interactions give 9 > 1.
Since Hamiltonians (4) and (11) give only an effective low energy description, it is generally difficult to relate the velocity v, and the interaction
parameter 9, to the bare interaction strength in the original lattice model,
338
say Eqn. (4). However, for a model with long-range electron-electron interactions (Tomonaga model, see e.g., Ref. [7]) these relations can be obtained analytically. In the Tomonaga model, only Fourier components with
k kF of the interaction potential V(k) are taken into account, and the
parameters in Eqs. (4) and (11) can be expressed [7] in terms of V(O).
We consider here a generalized Tomonaga model, which allows for different
values of interaction between the electrons moving in the same direction
(Vll(O) = V;-r(0) == VI) and in opposite directions (Vlr(O) = Vrl(O) == V2).
The velocity v and interaction parameter 9 can be related to VI and V2 as
follows. In the generalized Tomonaga model, the energy associated with a
density fluctuation in the right- or left-moving mode (NR or NL) is a sum
of two terms, representing compressibility of the non-interacting Fermi gas,
and the interaction potential VI, respectively. Equating this with the coefficient of
in 11, gives: 7rVo = 7rVF + VI. The interaction between the
modes V2 is responsible for the last term in Eq. (11), so that 7rVOA = V2.
These two relations can be used to solve for the velocity v and parameter
g, giving [15]
Nk.
V;1
v = VF ( 1 + --
7rVF
TT2)
v2
2 2
TT2
VI -
7r
vF
1/2
,
9=
(1 + VI/27rVF - V2/27rVF) 1/2

1 + VI/27rVF
+ V2/27rVF
(13)
In the conventional case VI = V2 > 0, and v is simply the plasmon velocity, increasing above VF with repulsive interactions which reduce the
compressibility of the electron gas. Notice, moreover, that the sign of 1 - 9
is determined by the sign of the interaction, with 9 < 1 for repulsive interactions. Edge states propogating on opposite edges of an integer quantum
Hall bar, constitute a 1d system with VI > V2. In particular, for a wide
Hall bar, the inter-edge interaction V2 vanishes, so that 9 = 1 .
2.1. MEANING OF 9
No matter what the underlying microscopic model is, the parameter 9 of

the Luttinger liquid (4) can be given a simple physical meaning as a dimensionless conductance. To this end we define new variables R/L = g ()
which decouple the right and left moving sectors, and at the same time
diagonalize the Hamiltonian (4). The corresponding right and left moving
densities are
(14)
Although nR/L again sum to give the total density, nR + nL = p, for 9 f:. 1
they do not correspond to the right and left pieces of an electron, but
rather involve contributions from electrons propagating in both directions.
339
The right and left fields commute with one another and satisfy,
[R(X), R(X')] = -[L(X), dx')] = i7rgsgn(x - x'),
(15)
the same as for N R/ L except with an important factor of 9 on the right side.
In terms of these fields, the Bosonized Hamiltonian decouples into right and
left moving sectors:
(16)
This innocent looking decoupling, leads to a number of remarkable predictions for the Id electron gas, as discussed below. Firstly, from the Hamiltonian one can generate the equations of motion, which are
(17)
These chiral wave equations have the general solutions, nR/L = !(X":f vt ), for
arbitrary!, and describe a density disturbance which propagates, without
scattering (or dispersion), to the right or left. This, despite the fact that
the electrons, which are interacting, do scatter off one another.
The physical meaning of 9 as a conductance can now be revealed. Imagine an external probe (or contact) which raises the chemical potential of the
chiral mode nR, by an amount }-lR. This can be incorporated by adding to
the Hamiltonian density a term of the form, 8H = -e}-lRnR. This leads to
a shift in the right moving density, since the total Hamiltonian is no longer
minimized by nR = o. Rather, minimization with respect to nR, gives
nR = (ge/27rv)}-lR. This extra density carries an additional (transport) current to the right, of magnitude, IR = enRV. One thus has, IR = Gil- with a
conductance,
(18)
For non-interacting electrons (g = 1), we recover the result of Landauer
transport theory - e2 / h conductance per channel. But with interactions G
is modified.
Unfortunately, in a quantum wire, it is essentially impossible to selectively couple to the right moving mode nR, and not the left. Rather, in a
typical transport experiment, the quantum wire is connected at its ends to
bulk metallic contacts, which can be modelled as free electron gases (Fermiliquids). In this case, the dc transport will be sensitive to the contacts. This
can be seen by considering an alternative derivation of Eq. (18), based on
linear response theory. Consider an ac electric field E(x, t) = -U'(x) sinwt,
applied along an infinite wire, which is non-zero for x near the origin, say
Ixl < L/2. This leads to a term of the form geE(x, t) on the right hand
side of (17). This field will cause an excitation of density waves n R/ L (x, t),
340
which propogate along the wire. For w < vi L, the wavelength of the excitation becomes much larger than L, so that one may then substitute
U'(x) = -U 8(x), with U the amplitude of the applied ac voltage across the
length L segment. The equations of motion can now be readily solved, giving for the current response, J(x, t) = ge 2 U cosw[t - xsgn(x)lv]. Although
the current depends on x, this dependence vanishes in the w -+ 0 limit, and
gives a dc conductance G = J IU = ge 2 lh, which reproduces Eq. (18). The
conductance is clearly related to the energy radiated away from the length
L segment. Indeed, the radiated energy per unit time is exactly equal to
the power U 2 G12 absorbed from the field. The resulting conductance is
finite, even in the absence of any disorder. It is clear from this arguemnt
that the dc conductance, proportional to the power radiated away, depends
on the value of 9 in the segment of wire away from the length L region. For
a wire of length L attached (adiabatically) to Fermi liquid leads, one can
take 9 = 1 in the leads [16), and one then expects a dc conductance given
by G = e2 1h. The value of 9 =1= 1 in the wire, can be revealed by an a.c.
conductance measurement for frequencies w > vi L, which has magnitude
[17, 18] G(w) = ge 2 lh.
There is a physical system however, in which the right and left modes,
n R/ L can be coupled to selectively. As discussed in more detail in Section 5.1
below, in the fractional quantum Hall effect (FQHE) at such filling factors
v < 1, that v-I is an odd integer, a single current carrying mode is predicted along the sample edges. In a Hall bar geometry, the right and left
moving modes are localized on the top and bottom edges of the sample,
respectively. These modes are mathematically isormorphic to the right and
left moving modes of the 1d interacting electron gas, nR/L, provided one
makes the identification 9 = v. In the FQHE geometry, a chemical potential
difference between the right and left moving modes, corresponds to a Hall
potential drop, transverse to the transport current. The above conductance,
corresponds directly to the quantized Hall conductance, G = ve 2 1h.
2.2. CHARGE FRACTIONALIZATION
The above analogy with the FQHE, suggests the possibility of fractionally charged excitations in the 1d electron gas. To see these, it is convenient to consider adding a small impurity potential, localized at the origin
x = o. This potential will cause scattering between the right and left moving modes, nR/L. We shall now show that the "particle" which is backscattered is not a (bare) electron. Rather, the total backscattered charge is
non-integral, with magnitude ge. For repulsive interactions 9 < 1, this corresponds to the backscatering of a quasiparticle with fractional electron
charge. We shall also show, that this fractionally charged quasiparticle,
341
also has fractional statistics. A localized impurity potential, U(x) = w5(x),

couples to the local electron density, 1P t (O)1P(O). Inserting the expression
(8), gives the dominant 2kF backscattering contribution:
t
Himp = u5(x) (1PR1PL
+ h.c.) = u5(x)(e i2B + h.c.).
(19)
Although, this perturbation backscatters a (bare) electron, there is a "backflow" term due to the electron-interactions. To reveal this, one should reexpress the above perturbation in terms of the fields, <P R/ L, which propogate
freely (to the right or left) even in the presence of electron interactions. One
gets simply,
(20)
so that the impurity hops an eiR/L quasiparticle between the two modes.
The charge created by the operator eiR can be deduced from the conjugate
momentum, PR = ox<PR/21fg which follows from the commutation relations
(15). One has,
(21)
which creates a kink in <PR of magnitude 21fg centered at x. Since nR =
ox<P R/21f the charge associated with this (kink) excitation is fractional,
Q =ge.
Thus, a localized impurity potential in an interacting Id electron gas,
causes backscattering of fractionally charged quasi particles (Q = ge) between the free-streaming right and left moving modes, nR/L. In the FQHE,
a localized impurity is equivalent to a "point contact" in which the right and
left moving edge modes on opposite sides of the Hall bar are "pinched" together, to allow for inter-mode backscattering. In this case, these backscattered quasiparticles have a natural physical interpretation as Laughlin quasiparticles with charge Q = ve. The fractional charge can perhaps be revealed
in a shot noise type experiment, as discussed in Ref. [19J.
Not surprisingly, these quasiparticle excitations also have fractional
statistics. To see this we can use the commutation relations (15) to show
that,
(22)
so that under exchange they pick up a phase factor exp(i1fg). For the noninteracting electron gas, 9 = 1, the quasiparticles have charge e and Fermi
statistics (the electron!), but with interactions will have fractional statistics.
3. Tunnelling into a Luttinger liquid

As discussed in the previous section, elementary excitations in the Luttinger
liquid are significantly different from bare electrons. This difference reveals
342
itself in the behavior of the density of states for tunnelling an electron into
the Luttinger liquid. In subsection A below we will consider specifically the
local tunnelling density of states for an infinitely long Luttinger liquid. Another situation of interest, analyzed in subsection B, consists of tunnelling
an electron through a large barrier separating two semi-infinite systems.
This process is sensitive to the density of states at the "end" of a semiinfinite Luttinger liquid, which, in contrast to the non-interacting electron
gas, behaves differently than the "bulk" density of states.
Since the effective Hamiltonian density (4) is quadratic in the boson
fields, local density of states can be readily extracted. Perhaps the simplest
way to proceed, is to represent the partition function,
= Trexp(-(3 Ix H),
(23)
as an imaginary time path integral over the boson fields:
z=
D1>D(}exp( -S),
(24)
where the integration is over classical fields 1>(x, T), and similarly (), with
imaginary time T running from 0 to (3. The (Euclidian) action can be written
in terms of the Lagrangian, S = I dxdT .co, with
(25)
with H given in (4). It is instructive to perform the (Gaussian) functional
integration over, say 1>, which gives:
(26)
describing, via the "displacement" field (), 1d phonons propagating with
velocity v. Likewise, integration over () gives:
(27)
This can be interpreted as a wave propagating in the phase-field 1>. Notice
that 9 -+ g-l upon transforming between the two representations, () -+ 1>.
At the special non-interacting point, 9 = 1, there is a self-duality between
these two representations. Physically, for strong repulsive interactions corresponding to 9 1, (zero-point) fluctuations in the displacement field ()
are greatly suppressed. In this limit () becomes a "good" classical variable,
and the 1d electron system is well described as a "solid" - or equvalently
343
a Wigner crystal. In the opposite extreme of very strong attractive interactions, 9 > > 1, fluctuations in the phase c/> are strongly suppressed. The
system is well described as a (almost) condensed superfluid of the (JordanWigner) bosons, b.
In the presence of a single impurity, the Hamiltonian H(c/>, 0) in Eq. (25)
should be replaced by H + Himp . Using Eq. (19) and integrating out the
field C/>, we find
In the language of 1d displacements, the impurity plays the role of a pinning

center, which favors a periodic set of values of the displacement field at the
pinning site. A strong center (large luI) effectively cuts the infinite system
in two semi-infinite pieces with independent excitation spectra. Each piece
is then described by the Lagrangian (26), defined on the appropriate semiinfinite space with boundary condition O(x = 0) = O.
3.1. TUNNELING INTO A CLEAN LUTTINGER LIQUID
Consider now the local electron tunnelling density of states for adding an
electron at energy E,
p(E)
= 21f L
l(nl~t(x)IO)128(En - Eo - E).
(29)
Here In) are exact eigenstates of the full interacting Hamiltonian, and En
are the corresponding energies. The summation in Eq. (29) is performed
over a complete set of states, and therefore with the help of the identity
8(En - Eo - E) =
~Re
1f
roo dtei(E+Eo-En)t,
10
(30)
can be related to the electron Green's function,
p(E)
= -Re
1 10
1f
00
dteiEt(~(x, t)~t (x, 0)).
(31)
This can be evaluated by computing the imaginary time correlator,

(32)
and then performing an analytic continuation, (~(t)~t(O)) = 9(T -+ it).
Upon re-expressing the electron operator in terms of the boson fields,
using (8), the average in (32) can be readily performed for a system with
344
a quadratic Lagrangian. For a clean infinite Id system, one can use the
Lagrangian (25). With a frequency cutoff, Tc-I, one finds,
(33)
with exponent a = (g + g-l) /2. After analytic continuation and Fourier
transformation, one thereby obtains,
p
()
E
27r 8( ) a a-I
= r(a)
- E Tc E
.
(34)
Notice that for non-interacting electrons (g = 1), one recovers the expected constant density of states. However, with interactions present, 9 =1= 1,
the electron tunnelling density of states vanishes as E --+ O. This striking
feature of the Id electron gas is intimately related to the fractional charge.
For 9 =1= 1 the excitations of the system are not free electron like, and there
is an orthogonality catastrophe when one tries to add in an electron. The orthogonality catastrophe occurs because the accomodation of an added electron requires modification of the wavefunctions of all the electrons forming
the liquid.
The tunneling conductance through a large barrier separating two semiinfinite Luttinger liquids, is proportional to the product of the "end" density
of states for the two pieces. This can be evaluated with the help of the Lagrangian (27) defined, say, for x > 0, together with the boundary condition
O(x = 0) = O. The result is similar to Eq. (34) with a = 1/g. This result
can be understood in rather simpler physical terms in the limit of 9 1,
as we now discuss.
3.2. TUNNELING THROUGH A BARRIER IN THE LIMIT
In the limit of strong electron-electron interactions, the fluctuations in the

"displacement" field are small, and the Id system with a single barrier can
be treated as a Wigner crystal pinned by an impurity. At low energies E,
the process of tunneling consists of several stages.
If the barrier created by the impurity is narrow and high enough, tunneling of the electrons close to the barrier occur on a fast time scale, set by
the Wigner crystal Debye frequency, WD == kFv. This fast process results in
a sudden (Itl;:::l/wD) creation of an electron vacancy-interstitial pair separated by the barrier. The energy of such a configuration is large, '" W D / g,
and therefore the corresponding state of the Wigner crystal is classically
forbidden. Gradual relaxation of the electron localized near the barrier,
constitutes the slow stage of tunneling.
345
Complete relaxation requires that all the electrons of the Wigner crystal
shift from their initial positions by one crystalline period. Since the overlap
between the initial and final wave functions for each electron is suppressed,
in the thermodynamic limit the initial and final many-body ground states
are orthogonal to one other - the orthogonality catastrophe. This implies
that the tunneling amplitude at E = 0 vanishes. At E > 0, a complete
relaxation is not required, since the system can end up in an excited state.
The initial ground state should have a non-vanishing overlap with the final
excited state, and the tunneling amplitude is finite [20J.
The density of states is directly related to the amplitude of the tunneling
process for a semi-infinite Wigner crystal, after an interstitial is introduced
at its edge, see Eq. (31). In the semiclassical approximation this amplitude
can be expressed in terms of the action accumulated along the optimal
classical trajectory,
('ljJ( x, t)'ljJ t (x, 0)) '" exp[ -8( t) J.
(35)
As we now show, the slow stage of tunneling contributes a logarithmically

divergent contribution [21 J as It I -+ 00. This divergence justifies the use of a
(26) harmonic in deformations, and also enables one to treat the low-energy
tunneling semiclassically, not only in the limit g 1, but at any value of
g < l.
Let us consider the evolution of the right half of the pinned Wigner
crystal, x> O. To connect the initial local deformation p(x, t = 0) = 28(x)
with the final relaxed state p(x, t = (0) = 0, the optimal trajectory must
run in imaginary time, t = iT. The appropriate solution of the equations of
motion gives
2
VT
p(x, T) = -7r X 2 + (VT )2'
x> O.
(36)
This solution is applicable for sufficiently large x 2 + (VT) 2 , so that the

harmonic description, see Eqs. (4), (26) is valid. Note that the dimensionless
displacement at the origin, O(x = +0, t) = -7r, remains constant in time,
which allows us to use consistently the limit of a 8-function in the initial
condition. Eq. (36) describes the spread and relaxation of the deformation
created by the tunnelling electron. The associated potential energy, Vdef (T),
decreases monotonically with time. Specifically, with the help of (36), one
finds:
Vdef(T)
= -v
27rg
loT dx(ox O)2 = -2v loT dx

trg
(VT)2
[x 2 + (VT)2J
1
= -2
.
gT
(37)
The kinetic energy is equal to the potential energy (virial theorem), and
thus we find
8(T) =
Vdef(T) = ~ In(wDT)
(38)
wZ;l
g
21T
346
for the tunneling action. Here we have used 1/WD as the short-time cut-off
for the slow part of the evolution. Finally, upon using Eqs. (31) we can
extract the tunnelling density of states, p( E) ex E Cl - 1 with a = 1/g.
4. Electron transport in a Luttinger liquid with a barrier

In the preceding Section we obtained the tunnelling density of states for
adding an electron at energy E into a Luttinger liquid. We considered
two cases: Tunnelling into an infinite Luttinger liquid and tunnelling into
the end of a semi-infinite system. In both cases the DOS vanished as a
power law of energy for an electron gas with repulsive interactions. Here
we consider an infinitely long interacting electron gas with a single defect
or barrier, localized at the origin. Of interest is electron transport through
the barrier. We first consider two limiting cases, a very large barrier in
Subsection A, and a very small barrier in Subsection B. In Subsection C
we show how the crossover between these two limits can be understood
by considering general barrier strengths, but weak electron interactions. A
general picture of the crossover is discussed in Subsection D. Finally, we
consider the special case of resonant tunneling in Subsection E.
4.1. LARGE BARRIER
An infinitely high barrier breaks an electron gas into two de-coupled semiinfinite pieces. Each piece can be described by either of the quadratic Lagrangians, (26) or (27). For a very high, but finite barrier, we can consider
electron tunnelling from one semi-infinite piece into the other as a perturbation. The appropriate tunnelling term to add to the Hamiltonian, is of
the form
(39)
H tun = to[V;{ (x = O)~h(x = 0) + h.c.],
where'l/Jl ('l/J2) is the electron operator iIi the left (right) semi-infinite Luttinger liquid. Here to denotes the (bare) tunnelling amplitude. Using (6)
these operators can be readily expressed in terms of the boson fields, Band
cp. However, with an infinitely high barrier the displacement field B(x) is
pinned at the origin, so one can take 'l/J(x = 0) = exp[icp(x = 0)].
The two-terminal conductance through the point contact can now be
computed perturbatively for small tunneling amplitude to. In the presence
of a voltage V across the junction, the tunneling rate to leading order can
be obtained from Fermi's Golden rule:
21l'e 'Lt
" snl(nIHtunIO}1 2 8(En - Eo - sneV ).
1= T
(40)
The sum on n is over many-body states in which an electron has been transferred across the junction in the Sn = 1 direction. It is straightforward to
347
re-express this as
et 2
1= 27r~
dE [pi(E)pi(E - eV) - p((E - eV)pi(E)]
(41)
where p~ (p;{) is the tunneling densities of states for adding (removing) an

electron at energy E. These are related by pE) = p>(-E).
Upon using the expression (34) for the tunnelling DOS into the end of
a semi-infinite Luttinger liquid, one readily obtains,
G(V) == dI ex t6JVI(2/ 9 )-2.

dV
(42)
For repulsive interactions (g < 1), the linear conductance is strictly zero!
This is a simple reflection of the suppressed density of states in a Luttinger
liquid. When 9 = 1 a linear I - V curve is predicted, consistent with
expectations for non-interacting electrons which are partially transmitted
through a barrier. At finite temperatures the density of states is sampled
at E ~ kT, and a non-zero (linear) conductance is expected. Generalizing
Fermi's Golden rule to T-/:-O gives the expected result for the (linear
response) conductance:
(43)
It is instructive to re-cast this result in the language of the renormalization group (RG). Specifically, the vanishing conductance for 9 < 1 indicates that the tunneling perturbation, to, is irrelevant. The RG can be
implemented using the Bosonized representation, in which the tunnelling
term takes the form,
(44)
with 1(T) = 11(X = O,T) - 12(X = O,T). This tunnelling term is aded to
the quadratic Lagrangians (27) for the two semi-infinite Luttinger liquids.
Since the perturbation to acts at a single space point, x = 0, it is useful
to imagine "integrating out" the fields 1a(x) for x away from the origin,
leaving only the time dependence, 1a(x = 0, T). The RG then proceed as
follows. In the frequency domain, 1(w), one integrates over modes in a shell
wclb < W < We, with We a high frequency cutoff of order the Fermi energy.
This can be done by splitting the field into "slow" and "fast" modes, below
and inside the shell, respectively: 1 = 1s + 11' To lowest order in to one
must average over the fast modes:
(e i )!
eis(eif)!
b-6. eis.
(45)
348
Here ~ is the scaling dimension of the operator ei . The scaling dimension

is most easily deduced from the two point correlation function,
(46)
From (35) and (38) one obtains ~ = l/g. The RG transformation is completed by rescaling time, 7' = 7/b, to restore the cutoff to We' The resulting
action is then equivalent to the original one with to replaced by t~ = tob 1-[)..
Upon setting b = e, and denoting the renormalized tunneling amplitude
as t(/!), one thereby obtains the leading order differential RG flow equation,
dt
d/!
= (1 -
(47)
~)t,
with ~ = l/g. The perturbative results (42) and (43) can be obtained by
integrating this RG flow equation until the cutoff is of order kT (or eV),
giving teff t oT(1!g)-l and G t~ff'
For a channel of finite length L coupled to Fermi liquid leads, renormalization in (47) should be stopped at the level spacing, i.e., at e kFL. The
conductance is temperature and voltage independent for T, eV liVF / L.
f",J
f",J
f",J
4.2. WEAK BACKSCATTERING LIMIT
Having established that the conductance of a repulsively interacting Luttinger liquid vanishes at T = 0 for weak tunneling , we now turn to the
opposite limit in which the barrier is very weak. In this limit we can treat
the barrier as a small perturbation on an ideal Luttinger liquid. The Lagrangian (28) is a particularly convenient representation, with the barrier
strength u assumed small. As discussed in Section lIB, the perturbation
proportional to u backscatters fractionally charged quasiparticles between
the de-coupled right and left moving Luttinger modes. What effect does
this weak backscattering have on transport?
Consider first a simple RG. As descibed in the previous section, to
leading order in u the RG flow equation is,
du
d/!
with the scaling dimension
= (1 -
(48)
~)u,
defined via the correlation function,
(ei2IJ(X=O,'T)e-i2IJ(x=O,O)) ex:
171- 2[)'.
(49)
Evaluating this using the quadratic part of the Lagrangian (28) gives ~ =
g. With repulsive interactions (g < 1) the backscattering strength grows
at low energies. Cutting off the flow equations when the temperature T
349
is comparable to the cutoff We, gives an effective backscattering stength

diverging at low temperatures as Ueff rv uTg-i. Eventually, at very low
temperatures the backscattering becomes sufficiently strong that treating
it as a perturbation is no longer valid. Nevertheless, one expects that upon
cooling the backscattering strength will continue to grow, until the system
scales into the large barrier regime discussed in Subsection A. This implies
that at T = 0 even a very small barrier will be inpenetrable, and effectively
break the Luttinger liquid into two decoupled pieces.
It is also possible to directly calculate the conductance through a weak
barrier, perturbatively in u. To extract a two-terminal conductance for
an infinite wire (ignoring Fermi liquid leads), one can reason as follows.
With no barrier present, the right and left moving Luttinger modes differ
in potential by the bias voltage eV. This results in a transport current
I = g( e2 / h) V. Quasiparticle backscattering between the two modes will
tend to reduce this current. The reduction can be computed perturbatively
using Fermi's Golden rule as in Subsection A, but with two differences.
First, the charge e in (40) must be replaced by the quasiparticle charge,
e* = ge. Second, the electron tunneling operator (39) must be replaced by
the quasiparticle tunneling term proportional to u. At zero temperature
one obtains,
(50)
as could have been anticipated from the RG flow equation. Notice that 9
has been replaced by 1/9 in going from the strong to weak barrier result.
Likewise, at temperature T, the backscattering contribution to the (linear)
conductance is given by
(51)
4.3. THE LIMIT 1 - 9
We have seen in Section 4.2 that backscattering off a barrier is enhanced

dramatically when the electron gas is repulsively interacting, with 9 < l.
Indeed, even a weak scatterer is expected to cause full reflection at zero
excitation energy. In this section, we show how this surprising result can
be understood in terms of the physical electrons. Specifically, we show that
the scattering rate for an electron off an impurity is renormalized by the
electron-electron interaction, due to the formation of a Friedel oscillation
near the barrier. This results in singular reflection amplitude at q = 2kF.
By considering the limit of very weak interaction, 1 - 9 1, it is possible
to treat the scattering for arbitrary barrier strength [22]. This allows us to
350
describe the crossover between the perturbative results for large and small
barrier described above.
Consider first a Id gas of spinless non-interacting electrons scattering
on a potential u(x) localized near the origin. For simplicity, we assume
the barrier is symmetric, u(x) = u( -x). This allows us to use the same
transmission and reflection amplitudes, to and ro, to describe both sets of
asymptotic wave functions far from the barrier:
(/Jk(x)
1 { eikx + r e- ikx x < 0,

'2=
t eikx 0
,
x> 0,
YL,'rr
0
,
(52)
for the states incoming from the left, and
1 {t e- ikx
-k(X) =..j'im e~ikX ~ roeikx,
< 0,
x> 0,
(53)
for the states incoming from the right. The wave vector k is defined to be
positive.
Scattering from the barrier is modified by electron-electron interactions,
and can be considered perturbatively in the interaction strength. To lowestorder we neglect inelastic processes in which electrons above the Fermi level
lose coherence by exciting electron-hole pairs. Within this Hartree-Fock
approximation, the many-body electron state can be described by a Slater
determinant of single-electron wave functions. Each electron is affected by
an extra average potential produced by other electrons in the Fermi sea.
This potential and the barrier potential, u(x), act together as an effective
barrier for electron scattering. The single-electron wave functions can be
found as a solution of the Schr6dinger equation with this effective barrier.
Then the transmission coefficient can be calculated.
The extra potential consists of two parts: the Hartree potential VH (x)
determined by the electron density in the system, and a non-local exchange
potential Vex (x, y) which accounts for Fermi statistics of the electrons.
Within this Hartree-Fock approach, the single-electron wave functions can
be found by the Green's function method. The equation for a single-electron
Hartree-Fock state 'ljJk is:
The Hartree and exchange potentials are defined as:

VH(X) =
Vex(x,y)
dyV(x - y)n(y),
= -V(x -
y)
Iql<kF
'ljJ~(y)'ljJq(x),
(55)
(56)
351
where V(x - y) is the electron-electron interaction potential, and n(y) =

I:lql<k F l1Pq(y)1 2 is the electron density.
In a first-order Born approximation, 1Pk on the right-hand side ofEq. (54)
and in Eqs. (55) and (56) is replaced by the unperturbed wave function <Pk.
The unperturbed electron density has the form:
n(x)
e- 2ikx
}
= { no + 1 JOrkF dkRe{r0
.,
no + ~ IokF dkRe{ro*e- 2tkX },
11"
x < 0,
x > O.
(57)
LFrom Eq. (57), at large distances Ixl k"Fl the disturbance of density
= n(x) - Po caused by a symmetric barrier decays as
8n(x)
8n(x)
~ ~III
sin(2kplxl + argro).
27r x
(58)
It follows from Eq. (55) that the oscillations of density (57) produce an
oscillating Hartree potential- commonly referred to as a Friedel oscillation;
see Fig. 1. In contrast to the 3d case, where the density oscillation a distance
R from an impurity decays as 1/ R 3 ], in Id it decays only as 1/lxl. As we
shall see, the 7r / k p periodicity and slow decay of the Hartree potential, gives
a contribution to to and ro which is logarithmically divergent as k -t kp.
To extract the contribution to to, consider an incoming wave from the
left with wave vector k. The modified wave function 1Pk(X) must have the
following asymptotics:
x -t
+00,
(59)
where tk is the modified transmission amplitude. Thus to find the correction to the transmission amplitude, we only need the asymptotic form of
the Green function Gk(x, y) as x -t +00. Calculating it with free wave
functions, we find:
Gk(x, y)
1 {t eik(x-y)
= -.Wk-
~k(x-y)
+' roeik(x+y) ,
< 0,
> 0,
(60)
where Vk is the velocity of an electron with wave vector k.

The transmission amplitude resulting from Eq. (54) is then,
(61)
where d is a characteristic spatial scale of the interaction potential V (x) (if
the range of the interactions is shorter than the Fermi wave length, then
352
V(x)
Figure 1. Total scattering potential. The central peak is the bare potential of the barrier.
The wings represent the Friedel oscillation induced by the barrier.
d should be replaced by 27r / k F)' Here , is a dimensionless parameter that

characterizes the strength of the interaction:
,= V(O) - V(2kF)
27rVF
'
(62)
with V(q) denoting the Fourier transformation of the interaction.

The zero-momentum contribution V(O) originates from the exchange
term, whereas V(2k F ) comes from the Hartree term. Notice that, vanishes for a contact interaction, for which V(q) is a constant. Due to Pauli
exclusion, such an interaction can play no role. For a finite range repulsive
353
interaction, , is positive, and the transmission is suppressed, in accord with

the results of Sections 4.1,4.2. Indeed, we can relate 9 and, by calculating
the lowest-order interaction correction to the compressibility of a 1d Fermi
gas. For weak interaction we find
1- 9
- -
1 ~ ,.
(63)
Thus we see that Eq. (61), upon taking the proper limit (Itol 1 or
Ira I 1), can be viewed as the solution of the lowest order RG equation
(Eq. (47) or Eq. (48)respectively).
The first-order result (61) for the transmission amplitude consists of
two equal contributions. The first one corresponds to a plane wave coming
from the left and reflected by the barrier with amplitude ro. It is then
scattered back to the barrier by the Friedel oscillation on the left-hand side
with amplitude -~,ro In(l/lk - kFld). Finally the electron penetrates the
barrier with amplitude to. The second contribution is the product of the
amplitudes of the following processes: an electron first penetrates the barrier
with amplitude to, then it is reflected back to the barrier by the Friedel
oscillation on the right-hand side with amplitude -~,ro In(l/lk - kFld),
and eventually reflected by the barrier to the right with amplitude roo The
total first-order contribution to the transmission amplitude is the sum of
these two coherent processes.
The result (61) has a logarithmic divergence as k --+ kF, no matter
how small the coupling constants (62). This indicates the inadequacy of
the first-order calculation at smallik - kFI. The second order contribution
can be extracted by using the first-order 7/Jk(X) (59) as a new wave function
in the right-hand side of Eq. (54), and repeating the previous calculation.
The result is,
1
1-2"
1 to lral 2 (2l t ol 2 -lral)
2 [
1
I] 2
tk=to-tolral 2 ,In 1(k-kF)d
,In 1(k-kF)d
(64)
Here we have only kept the most divergent terms, which at second order
has the form [r In(l/lk - kFld)f At n-th order we expect terms of the
form [r In(l/lk - kFld)Jn. Since all these terms are divergent, a straightforward perturbative approach is clearly inadequate. Instead, we adopt a
renormalization group approach, as described below.
The Hartree and exchange potentials depend on the reflection amplitudes, and they are modified along with these amplitudes. In a region (-l, l)
close to the origin, the electrons are scattered by the bare barrier with transmission amplitude to and produce an extra potential that is proportional
to lral. Perturbative calculation for the transmission amplitude is carried
354
out with the bare amplitudes. Such a calculation is justified as long as the
correction from the perturbative calculation is indeed small. This is true
for 1 not too large, such that alnUld) 1. Beyond this distance, the whole
region (-1, l) enclosed should be considered as an effective barrier to the
electrons outside. This effective barrier is characterized by the now renormalized amplitudes, denoted rand t. With these new amplitudes, we can
find the new Hartree and exchange potentials in the outer region. Then
the perturbative calculation can be carried out to a larger spatial scale. To
ensure that perturbation theory is valid in every step, the above renormalization procedure is done repeatedly for larger and larger scales.
This idea leads to the following formulation. We start with a region of
length 21 centered around the barrier. The scale 1 is chosen to be much
larger than d but not too large, so that 1 In( 1I d) ,-1. The modified transmission amplitude due to the electron-electron interaction in the
region (-1, l) can then be found by perturbation theory,
it
= to -
,to(1-ltoI2)f,
(65)
with f = In1ld 1.
We then go to a larger scale, taking the region ( -1,1) as a composite scatterer. Using the renormalized transmission amplitude and correspondingly
renormalizing the additional Hartree and exchange potentials, we repeat
the calculation for this next scale 1exp(f). Then to the next larger scale,
which is 1exp(2f), and so on. In general, the iterative renormalization of
the transmission amplitude after n steps of scaling to larger distances can
be found from
(66)
This iteration procedure should be stopped at a length scale l/lk - kFI,
beyond which the scattered electron loses phase coherence with the Friedel
oscillation, and the transmission amplitude is not renormalized any further.
In the continuous limit, Eq. (66) becomes
dt
df
= -,t(l -It I ),
(67)
where f is the logarithm of the length scale. Integrating equation (67) from
f = 0 to f = In(lilk - kFI) and using the boundary condition tlf=O = to,
we find a renormalized transmission amplitude
tk
tol(k - kF)dl'Y
JIro1 2+ Itol21{k - kF)dI 2'Y
-r.;==c;r:~:==i'ii~====,=;==c=;:;::;c
The transmission coefficient T
= Itl 2 is then
ToIEI Do12'Y
T(E)
(68)
= no + ToIEI DoI2'Y'
(69)
355
where To = 1 - no = Itol2 is the bare transmission coefficient, and Do =

vF/d. The expansion ofEq. (68) up to the second order in 'Y coincides with
(64).
The renormalized transmission coefficient (69) allows us to find the temperature dependence of the linear conductance of a 1d spinless interacting
electron system with a single barrier. At high temperatures T > Do the
conductance is given by the Landauer formula for an ideal Fermi gas,
Go = (e 2/27rn)To. At lower temperatures the transmission coefficient is
renormalized. Because of the smearing of the Fermi surface, E in Eq. (69)
should be replaced by T, and the following temperature dependence of the
linear conductance is found:
GT = ~
()
To(T / Do)2'Y
27rn no + To(T / Do)2'Y
(70)
Formula (70) describes explicitly the crossover between the limits of weak
reflection and weak tunneling considered above in the framework of Luttinger liquid theory. The proper expansion of Eq. (70) and the use of
Eq. (63) yields the limiting results, Eqs. (43) and (51). The crossover temperature depends on the strength of the barrier, T* = Do(no/To)1/2'Y. The
differential conductance G(V) at a high voltage eV > T may be obtained
by substitution T -+ eV.
4.4. CROSSOVER BETWEEN THE WEAK AND STRONG
BACKSCATTERING
The preceding results can now be pieced together to form a global picture
of the behavior of a scattering defect in a Luttinger liquid. The perturbative
results in IV.A and IV.B describe the stability of two renormalization group
fixed points. For repulsive interactions (g < 1), the "perfectly insulating"
fixed point, with zero electron tunneling to = 0, is stable, whereas the "perfectly conducting" fixed point, with zero backscattering u = 0, is unstable.
(For attractive interactions, 9 > 1, the stability conditions are reversed.)
Provided these are the only two fixed points, it follows that the RG flows
out of the conducting fixed point eventually make their way to the insulating fixed point. In the limit of very weak repulsive interactions, 1- 9 < < 1,
we showed this crossover explicitly in Section IV.C, but it is true more generally. This is a very striking conclusion, since it implies that a Luttinger
liquid with arbitrarily weak scatterer, with amplitude u, will cause the conductance to vanish completely at zero temperature. Of course, for u very
small, very low temperatures would be necessary to see this. In this scenario, the conductance as a function of temperature will behave as follows.
At high temperatures, the system does not have "time" to flow out of the
perturbative regime, so the conductance is given by G ~ (ge 2/ h) - u 2T2g-2.
356
As the temperature is lowered below a scale T* <X u1/(1-g), perturbation

theory breaks down. Eventually, the system crosses over into a low temperature regime in which the conductance vanishes as T(2/g)-2.
The validity of this scenario rests on the assumption that no other
fixed points intervene. This assumption has been verified both by quantum
Monte Carlo simulations [23], and more recently by exact non-perturbative
methods based on the thermodynamic Bethe ansatz [24].
4.5. RESONANT TUNNELLING
We now briefly consider the phenomena of resonant tunnelling through a

double barrier in a Id Luttinger liquid. Our reasons are two fold: (i) Experiments on quantum Hall edge states, discussed in the next Section, have
measured resonances which can be compared with Luttinger liquid theory,
(ii) The nature of the crossover between the weak and strong backscattering
limit, determines the lineshape and temperature dependence of resonance
tunnelling peaks in a Luttinger liquid.
As a point of reference, we first review resonant tunneling theory for a
Id non-interacting electron gas. Consider then Id electrons incident on a
double barrier structure, with a (quasi-) localized state between the barriers. As the chemical potential J-t of the incident electron sweeps through
the energy of the localized state, EO, the conductance will exhibit a peak
described by,
e2
I
rLrR
(71)
G= h
dE! (E - J-t\: _ Eo)2 + r 2
Here r Land r R are tunneling rates from the resonant (localized) state to
the left and right leads and r = (r L + r R) 12. The Fermi function is denoted
! (E). At high temperatures, T > r, the resonance has an amplitude r IT
and a width T. At low temperatures, the lineshape is Lorentzian, with
a temperature independent width r. Moreover, when the left and right
barriers are identical, the on-resonance transmission at zero temperature is
perfect, G = e2 lh.
How is this modified when the electron gas is an interacting Luttinger
liquid? Since arbitrarily weak backscattering causes the zero temperature
conductance to vanish, one might expect that resonances are simply not
present at T = o. As we now show, this is not the case. Rather, perfect
resonances are possible, but in striking contrast to (71) for non- interacting
electrons, they become infinitely sharp in the zero temperature limit.
To see this, it is convenient to consider the limit of a very weak doublebarrier structure. For non-interacting elecrons, resonant tunneling is not
normally studied for weak backscattering, since in this limit the transmission is large even off resonance, which tends to obscure the resonance.
357
However, in a Luttinger liquid, the off-resonance conductance vanishes at

zero temperature, leaving an unobscured resonance peak, as we now argue.
Consider then scattering of a pure Luttinger liquid from a small barrier, denoted u(x), which is non-zero only for x near zero. For resonant
tunnelling, the potential should be taken to have a double-barrier structure. This potential couples to the electron density, via an additional term
in the Hamiltonian,
1limp
Jdxu(x)V;t (x)V;(x).
(72)
This can be re-expressed in terms of the boson fields by inserting the expression (2.6) for the electron operator. We assume that (}(x) is slowly varying
on the scale of the potential, and so replace it by (}(x = 0). The integration
over x can then be performed to give,
1limp
="2
L
00
unei2nB(x=O) ,
(73)
n=-oo
where the coefficients, Un = u~n = u(2nkF), are proportional to the Fourier

transform of u(x) at momenta given by n times 2kF. For a symmetric
barrier, u(x) = u( -x) the coefficients Un are real.
In Equation (19) we retained only the first term in this sum. The higher
order terms correspond to processes where n-electrons are simultaneously
backscattered, each by a momenta 2k F, from one Fermi point to the other.
Alternatively, (73) can be viewed as an effective potential Ueff((}(X = 0)),
which is invariant under the transformation (} --+ (} + 7r. Since (} / 7r is the
number of particles to the left of x = 0, Ueff may be regarded as a weak
pinning potential in the Wigner crystal picture.
As we shall now see, for weak backscattering, the single electron process,
ie. the 2kF backscattering term Ul, is the most important. This follows readily from a perturbative RG calculation, as in Section 4.2. Specifically, the
scaling dimension, .6.n , of the perturbation Un, defined via the correlation
function,
(74)
with imaginary time T, can be readily evaluated using the quadratic part
of the Lagrangian (28), to give .6. n = n 2g. Thus the leading order RG flow
equations for the renormalized coupling u n () are simply,
dUn
)
= ( 1- n 2gUn
(75)
With increasing n, the scaling dimension increases, and the perturbation, Un, becomes less relevant (or more irrelevant), suggesting we need
358
only focus on Ul. However, imagine fine-tuning Ul to zero. As we shall
see, this corresponds to tuning to a resonance. The next most important
backscattering process is U2. But notice, that for 9 > 1/4 this backscattering amplitude scales to zero, as do all the higher order processes, Un with
n > 2. Thus, if Ul is tuned to zero, provided 9 > 1/4 one expects perfect
transmission at T = O. On the other hand, as we have shown earlier when
Ul is nonzero, it grows under RG (for 9 < 1), and the conductance vanishes at T = O. Thus at zero temperature, there will be an infinitely sharp
resonance peak as Ul is varied through zero!
How easy is it to achieve such a resonance? For 9 > 1/4, the criterion
is that the renormalized value of Ul vanishes. In general, Ul is complex,
so that the resonance condition requires the simultaneous tuning of two
parameters. However, for symmetric barriers, Ul is real and only a single
parameter need be tuned.
A resonance with no width is in striking constrast to the conventional
result for non-interacting electrons. In that case, the width is set by the
tunnelling rate r from the localized state between the barriers, into the
leads. The higher the barrier, the smaller the decay rate, and the narrower
the resonance. An electron in a localized state between two barriers in a
Luttinger liquid will be unable to decay (at T = 0), since the tunnelling
density of states into the "leads" vanishes. The electron remains localized
forever, with an infinite "lifetime" - which gives a simple explanation of
the infinitely sharp resonance. At a finite temperature, the electron will be
able to decay, since the tunnelling DOS into a Luttinger liquid is non-zero
away from the Fermi energy. One thus expects that the resonance will be
thermally broadened, as we now confirm.
Consider tuning through such a perfect resonance by varying a parameter. It is convenient to denote by 8 the "distance" from the peak position
in the control parameter. Close enough to the resonance one has Ul <X 8.
For very small 8 the RG flows will thus pass very near to the perfectly
conducting fixed point, since all of the other irrelevant operators will scale
to zero before Ul has time to grow large. Eventually, Ul does grow large
and the flows crossover to the insulating fixed point. Temperature serves as
a cutoff to the RG flows, as usual. This reasoning reveals that for both 8
and temperature small, the conductance will depend only on the universal
crossover trajectory which joins the two fixed points. The uniqueness of the
RG trajectory implies that the conductance will be described by a universal
crossover scaling function. Moreover, since Ul grows with exponent 1 - g,
the conductance, which is generally a function of both 8 and T, will only
depend on these parameters in the combination, 8/T 1-g. Thus, for small T
359
and 0, the resonance lineshape is given by a universal scaling function,

_ ge
G(T, 0, g) - TGg(cgO/T
l-g
),
(76)
where cg is a (non-universal) constant.

The scaling function Gg(X) depends only on the interaction parameter
g, but is otherwise universal, independent of all details. The limiting behavior of Gg(X), may be deduced from the perturbative limits. For small
argument X, the perturbation theory result (51) implies
(77)
For large argument, corresponding to the limit T --+ 0, the scaling function
must match on to the low temperature regime (43), which gives a T(2/g)-2
dependence. This implies that for X --+ 00,
(78)
The scaling form for the conductance near resonance, reveals that the
width of the resonance scales to zero with a power of temperature, T 1-g. Notice that in the non-interacting limit, 9 --+ 1-, this reduces to the expected
temperature independent linewidth. Moreover, the finite temperature resonance lineshape is non-Lorentzian, with tails decaying more rapidly, as
X- 2 /g. Again, for 9 --+ 1 this reduces to the expected Lorentizian form for
the non-interacting electron gas.
The exact scaling function which interpolates between the small and
large X limits, can be computed for 9 = 1/2 by re-Fermionizing the Luttinger liquid. More recently, Fendley et. al. [24J have computed Gg(X) for
arbitrary g, using the thermodynamic Bethe Ansatz.
5. Applications
Here we demonstrate two applications of the theory reviewed in the previous

sections. The first one relates the general theory to the experimentally
observable transport phenomena in a mesoscopic quantum Hall effect. The
second one uses the mathematical tools described above in the theory of
Coulomb blockade.
5.1. TUNNELING BETWEEN QUANTUM HALL EDGE STATES
5.1.1. Edge states in the regime of quantum Hall effect
Despite the firm theoretical basis upon which the Luttinger liquid theory rests, there has been precious little compelling experimental evidence
360
that real one-dimensional electron gases are anything but Fermi liquids.
This, despite the fact that in recent years it has become possible to fabricate single channel quantum wires. Several recent experiments [25, 26]
have reported interesting transport data on such quantum wires, and offered possible interpretations in terms of Luttinger liquid theory. But the
analysis is complicated by several factors. Firstly, it is difficult to eliminate
unwanted impurity scattering along the wire. When this is strong, it causes
backscattering and localization, destroying the Luttinger liquid phase. But
even in a very clean wire, the Luttinger liquid parameter g, which determines the power law of tunnelling density of states, is unknown, depending
on the details of the Coulomb interaction strength, which complicates the
interpretation. Fortunately, there is another experimental system which is
expected to exhibit Luttinger liquid behavior, and does not suffer from the
above difficulties - namely edge states in the quantum Hall effect.
The quantum Hall effect occurs at low temperatures in two-~imensional
electron gases with low carrier density, when placed in a strong perpundicular magnetic field. The key experimental signature consists of quantized
plateaus in the Hall conductance as the magnetic field is varied. In the
plateaus, the Hall conductance, GH, is "locked" to values which are simple
rational numbers of the quantum conductance, G H = ve 2 I h, with rational
v. In the integer quantum Hall effect, v is an integer, whereas in the fractional quantum Hall effect, v = plq with integer p and odd integer q. In
the plateaus the longitudinal conductivity, a xx , vanishes rapidly as T -+ o.
The integer quantum Hall effect (IQHE) can be understood in terms of
2d non-interacting electrons in a magnetic field. When the Fermi energy
lies between Landau levels, there is an energy gap of order the cyclotron
energy, We = eB 1m, which accounts naturally for the vanishing a xx . But at
the edges of the sample, there are gapless current carrying states. In a semiclassical picure, these can be thought of as orbits which "skip" along the
edge due to the magnetic field and edge confining potential. The direction
of the skipping, clockwise or counter-clockwise, is determined by the sign of
the magnetic field. In a full quantum treatment, these edge orbits become
quantum modes. For n full Landau levels in the bulk, there are n edge
modes, one for each Landau level.
The IQHE edge modes are equivalent to the right moving sector of a 1d
non-interacting electron gas. In a Hall bar geometry, the modes on the top
and bottom edges move in opposite directions. For v = 1 there is a single
right moving mode on the top edge, and a left mover on the bottom. These
two modes can be described by the Hamiltonian (16) with g = v = 1. That
is, they are mathematically equivalent to a 1d non-interacting electron gas.
However, since the two modes are spatially separated - on opposite sides of
the Hall bar - impurities cannot cause backscattering and localization.
361
Quantization of the Hall conductance can be understood very simply in

terms of the edge states. As discussed in Section II.A, raising the chemical
potential of the right moving mode with respect to the leftmover by an
amount p" leads to a transport current, 1= Gp, with G = ge 2 /h from (18).
In the Hall bar geometry, p, corresponds to the transverse (Hall) voltage
drop, so G is a Hall conductance - appropriately quantized for 9 = 1.
The fractional quantum Hall effect occurs when a single Landau level is
partially filled, with a fractional filling v. For non-interacting electron there
would be an enormous ground state degeneracy in this case. At rational filling fractions, v = p/q, the electron Coulomb repulsion lifts this degeneracy,
leading to a unique ground state with an energy gap to excited states. The
resulting FQHE state is predicted to have very interesting properties, such
as quasiparticle excitations with fractional charge and statistics.
In pioneering work, Wen [12] argued that FQHE states should also have
gapless current carrying edge modes. But in contrast to the IQHE, these
edge modes were argued to be Luttinger liquids. Specifically, for the Laughlin sequence of fractions, at filling v = l/q with odd integer q, a single
chiral Luttinger liquid mode was predicted. For a Hall bar geometry, as in
the IQHE with v = 1, there would be one right moving mode on the top
edge, and one left mover on the bottom. Again, these two modes can be described by the Hamiltonian (16), but now with fractional Luttinger liquid
parameter, 9 = v = l/q. Remarkably, the Luttinger liquid parameter for
FQHE edge states is universal, determined completely by the bulk FQHE
state, and independent of all details.
For hierarchical FQHE states, at fillings v =1= l/q, multiple edge modes
are predicted [12], in some cases moving in both directions along a given
edge. In the following we will focus for simplicity on the Laughlin sequence
of states, and in particular on the most robust state at v = 1/3.
Although FQHE edge modes at filling v = 1/3, and a Luttinger liquid
with 9 = v are essentially equivalent, there are some subtle differences which
must be kept in mind. Specifically, in the Luttinger liquid the fundamental
(right moving) charge e Fermion operator (the electron) is given from (6) by
'l/JR rv ei (8+. This can be re-written in terms of the right and left moving
Boson modes, CPR/L = gcpfJ, as 'l/JR rv ei (2>R+>L). This operator thus adds
charge 2/3 into the right moving Boson mode, and 1/3 into the left. In
the FQHE, this corresponds to adding 2 fractionally charged quasiparticles
(charge e/3) to the top edge of the Hall bar, and 1 to the bottom edge. But
this is a non-local operator, and hence unphysical in the FQHE, whereas it
is nice local operator in a quantum wire.
362
Figure 2. Schematic portrait of a point contact, in which the top and bottom edges of
a Hall fluid are brought together by an electrostatically controlled gate (G), allowing for
the tunneling of charge between the two edges. Here S and D denote source and drain,
respectively.
5.1.2. Inter-edge tunneling at a Point Contact
As discussed in Sections 3 and 4, the most remarkable property of a Luttinger liquid is the vanishing density of states for tunnelling electrons. To
allow for intermode edge tunnelling in the FQHE, it is necessary to bring
together the opposite edges of the Hall bar. This can be achieved by gating the electron gas, as depicted schematically in the Figure 2. Consider
specifically, a gate which brings the opposite edge modes together at one
point. This is equivalent to putting a single defect in an otherwise clean
Luttinger liquid. In this geometry, charge can tunnel between the top and
bottom edge modes, through the narrow strip of FQHE fluid (see Fig. 2).
The appropriate term to add to the Hamiltonian is given in (20), and corresponds to the transfer of a fractionally charged Laughlin quasiparticle
(charge e* = ve = e/3) from top to bottom, with amplitude u. In a Luttinger liquid, this same perturbation corresponds to a weak 2kF electron
backscattering, with "backflow".
Since g = v < 1 this weak backscattering amplitude, u, grows at low
temperatures, and the system crosses over into a large barrier regime, as
discussed in Section 4. The large barrier limit corresponds to Figure 3a,
in which the incident top edge mode is almost completely reflected. Weak
tunnelling from left to right can be treated perturbatively, in the amplitude,
to. In this case, it is an electron which is tunnelling, with charge e. As shown
in Section 4.1, (Eqn. (43)), this leads to a conductance which vanishes as
363
(a)
(b)
Figure 3. A quantum Hall point contact in the (a) weak tunneling limit and (b) the
weak backscattering limit. The shaded regions represent the quantum Hall fluid with
edge states depicted as lines with arrows. The dashed line represents a weak tunneling
matrix element connecting the two edges.
a power of temperature,
(79)
for v = 1/3.
Fig. 4 shows data [27] for the conductance as a function of temperature
through a point contact in an IQHE fluid at v = 1 and a FQHE fluid at
v = 1/3. The difference in behavior is remarkable. For v = 1 the conductance approaches a constant at low temperatures, whereas for v = 1/3 the
conductance continues to decrease upon cooling. Moreover, the low temperature behavior for v = 1/3 is consistent with the T4 dependence predicted
364
--:c
8
)(
........
0.20
r-----_,__----r------,
11=1
Vpc=-0.S53 & -0.S99 V

0.10 ....~~~..'!I&I'+<P'~G.
..
"G
'-"
0.00 '-----'-----'---~
160
40
120
SO
T (mK)
(a)
(b)
Figure 4. Conductance of a quantum point contact as a function of temperature for (a)
v = 1, and (b) v = 1/3. Taken from Ref. [27].
in (79). This data provides experimental evidence for the Luttinger liquid,
a phase discussed theoretically over 30 years earlier.
In this same experiment, the conductance through the point contact was
measured as a function of a gate voltage, controlling the pinch-off. The data
in Fig. 5, shows a sequence of reproducible conductance peaks. A natural
interpretation is that these are resonant tunnelling peaks, through a state
localized in the vicinity of the point contact. As discussed in Section 4.5,
resonant tunnelling peaks are expected in a Luttinger liquid whenever the
amplitude for the dominant backscattering process (Ul) is tuned to zero. In
the FQHE, this process corresponds to tunnelling a single e/3 quasiparticle
from the top to bottom edge. Since v > 1/4, the higher order processes
which involve multiple quasiparticle tunnelling are irrelevant. Thus, for a
symmetric scattering potential, finding a resonance requires tuning only a
365
1.0
0.8
(b)
11= 1
.J::.
0.6
~
(1)
C,!)
l\,'1 ~'
:
, I
0.2
0.0
,'\ I'
I V
0.4
r I
I\ I
I \ I
J \1
T=42 mK
"
-0.90
"
-0.70
-0.50
Vpc (V)
-0.30
Figure 5. Two-terminal conductance as a function of gate voltage of a GaAs quantum

Hall point contact taken at 42 mK. The two curves are taken at magnetic fields which
correspond to v = 1 and v = 1/3 plateaus. Taken from Ref. [27].
single parameter, such as the gate potential.

Fig. 6 shows a scaling plot of one of the resonances for 11 = 1/3 in
Fig. 5 from the data of Webb et al. The widths of the resonances at
several different temperatures have been rescaled by T 2/3, as suggested by
(76). Since the peak heights were also weakly temperature dependent (and
roughly one third of the quantized value (1/3)e 2 /h) the amplitudes have
also been normalized to have unit height at the peak. The temperature
scaling of the peak widths is indeed very well fit by T2/3. Also shown in
Fig. 6 are quantum Monte Carlo data and an exact computation from
Bethe Ansatz for the universal scaling function in (4.38). The agreement is
striking. Although the experimental lineshape does drop somewhat faster
in the tails, the shape is distinctly non Lorentzian with a tail decaying with
a power close to that predicted by theory. It should be emphasized that the
experimental data does not represent a "perfect resonance" , since the peak
amplitude is dropping (slowly) upon cooling, rather than approaching the
quantized value, (1/3)e 2 /h. By varying an additional parameter besides the
gate voltage (such as the magnetic field) though, it should be possible to
find a perfect resonance for 11 = 1/3.
In a very recent experiment, Chang et. al. [28] have succeeded in tunnelling into a FQHE edge from a bulk metallic system. The 2DEG was
exposed, in situ, by cleaving a bulk GaAs sample. A thin insulating barrier
was grown onto the cleaved face. Heavily doped material was deposited on
top of the insulating layer. The experiment measured the tunnelling con-
366
Conductance
0.1
0.01
exact curve Monte Carlo IJ

experimental data x
0.001
0.D1
0.1
10
X=.74313(T_BfT)A(2/3)
Figure 6. Log-log scaling plot of the lineshape of resonances at different temperatures.

The x axis is rescaled by T2/3. The crosses represent experimental data of Ref. [27] at
temperatures between 40mK and 140mK. The squares are the results of the Monte Carlo
simulation, and the solid line is the exact solution from Ref. [24].
ductance from the heavily doped "metal" through the insulating barrier
into the FQHE edge. The tunnelling conductance from a Fermi liquid into
a 9 = v chiral Luttinger liquid, is predicted to vanish as, G '" T(1/v)-l.
Since the DOS in a Fermi liquid is constant, this power is 1/2 as large as
for tunnelling between two 9 = v chiral Luttinger liquids. Similarly, at zero
temperature, the I-V curve is predicted to vary as J '" VI/v. Figure 7 shows
data for the J- V curve which is consistent with this power law.
5.2.
COULOMB BLOCKADE OF STRONG TUNNELING
In this section we apply the mathematical tools developed for a 1d Luttinger

liquid, to an apparently unrelated physical problem - namely the Coulomb
blockade. Consider a conducting metallic grain (or "dot") connected to
an infinite lead by a junction, see Fig. (8a). The grain is also coupled
capacitively to another gate electrode. The electrostatic energy of the grain
367
v=1/3
T=25mK
Figure 7. Current-voltage characteristics for tunneling from a bulk-doped n+ GaAs

into the edge of a 1/ = 1/3 fractional quantum Hall fluid for two samples: Sample 1 at
magnetic field B = 13.4T (crosses), and sample 2 at B = 10.8T (solid circles). The solid
curves represent fits to the theory; the extracted values of the exponents for the two
curves are a = 2.7 and a = 2.6 respectively. (Taken from Ref. [28].)
IS
_ (Q - eN)2
Q -
2CO
'
(80)
where Q is the charge of the grain and Co is its capacitance. The parameter
N is proportional to the gate voltage Vg. Variation of the gate voltage Vg
changes the equilibrium charge of the grain. If the channel to the lead
is almost depleted, and its conductance G --+ 0, then the charge Q can
only take on discrete values Q = N e. The equilibrium charge (Q) should
minimize the energy (80) within the set of integer N, and therefore changes
368
a)
b)
y
2DEG
2DEG
Figure 8. (a): Metallic grain connected to a lead by a channel (taken from Ref. [30]).
The charge on the grain is controlled by the gate voltage, Vg , applied to an electrode
which is coupled capacitively to the grain. (b): One-dimensional channel connecting the
grain to the lead
in a stepwise fashion with the gate voltage, (Q) = IntNe. On the other
hand, if G is large, one expects the grain charge to vary linearly with the
gate voltage, (Q) = Ne. Of interest is the behavior in the crossover region
between these two limiting cases, as the conductance to the lead is varied.
This problem is non-trivial, because the electron-electron interaction
term (80) must be treated non-perturbatively. As we shall see, if the junction is modelled as a single-mode channel with tunable reflection amplitude
369
r, it is possible [29, 30] to apply the methods described in Section 2. Indeed,

using bosonization, Matveev [30] has shown that (Q) = Ne at r = 0, and
found analytically the function q(N) == (Q) - N e for weak reflection. (Here
and after, (... ) denotes a ground-state average.) The average charge is directly related to the ground state energy, q(N) = -(Co/e)8E/8N. Below
we follow closely the line of arguments presented in Ref. [30].
To find the ground state energy E, we need to supplement the Hamiltonian (SO) by a part describing non-interacting electrons. For the case
of a grain with size L large compared to the Bohr radius, aB = ;,,2 /me 2 ,
this can be simplified. If the motion within the grain is chaotic, then the
dwelling time for an electron entering the grain through a single-mode channel is Td rv 8-1, where 8 is the electron level spacing in the grain. On the
other hand, the quantum charge fluctuations which destroy the discreteness of (Q) occur on the time scale Tq rv 1/ Ee, where Ee rv e2 /Co is the
single-electron charging energy. The latter time scale is relatively short,
Tq/Td rv aB/ L 1 (the estimate is performed for a two-dimensional grain).
Over this time, electrons participating in the charge fluctuations (i.e., those
that pass through the channel) do not get back into the channel after "exploring" the grain. It means that we need to add to Eq. (SO) a Hamiltonian
describing only the electrons moving through a one-dimensional channel.
In the absense of a barrier within the channel, this Hamiltonian is
where 1j11(x) and 1j1i(x) are the creation operators for right- and leftmovers, respectively. We have used here a linearized electron spectrum,
since the energy scale Ee is much smaller than the Fermi energy, Ee / EF rv
(aB/L)(kFaB)-2 l.
The definition of the charge operator Q in Eq. (SO) depends on where we

"draw" the boundary between the channel and the grain. This arbitrariness
in convention only affects the phase of the oscillations of q(N). But this
phase does not carry any physical meaning, provided the gate voltage is
small enough not to deplete the grain entirely. Therefore, Q can be viewed as
the charge passed e.g. through the middle (x = 0) of the channel, Fig. (Sb)
into the dot,
(S2)
Now, with the help of Eq. (8), we can bosonize the Hamiltonian (81),
which gives Eq. (4) with 9 = 1,
(83)
370
Bosonization of the interaction part (80) is also straightforward, after we

put 'l/Jk(x)'l/m(x) + 'l/J1(x)'l/JL(x) -* -8x B(x)/1r into (82),
He
[~B(O) -
= Ee
Nf '
(84)
In the absence of a barrier in the channel, it is easy to show that q(N)

O. Indeed, after the transformation
B(x) -* B(x) + 1rN,
(85)
the Hamiltonian Ho + He does not depend on N; there is no Coulomb

blockade at r = O. The ground state energy starts to depend on N, if a barrier causes backscattering in the channel. The corresponding Hamiltonian
(cf Eq. (19)) in the boson variables is
Himp
= --Dcos[2B(0)],
1rVp
(86)
where u is the 2k p -component of the scattering potential, and D is a high

energy cut-off. The transformation (85), makes evident the periodic dependence of the full Hamiltonian, Ho + Himp + He, on N.
To find the correction to the ground state energy to first order in Irl,
denoted ElY, the perturbation (86) must be averaged in the ground state
of the Hamiltonian Ho + He, Eqs. (83), (84). It then follows directly from
Eqs. (83), (84) and (85), that El ex: cos(21rN). The proportionality coefficient can be estimated with the help of the RG equation (48). In the
absence of the local interaction (84), the renormalization (48) is valid for
all energy scales and can be taken to the limit -* 00, which results in a
vanishing effective scattering potential, El = O. With a finite Ee, the renormalization should be stopped at exp (- ) '" Ee / D, which at 9 = 1 leads to
El = -const (u/vp)Ee cos 21rN. To leading order in the backscattering,
Irl = u/vp, so that
El
= -const 'IrlEe cos 21r N.
(87)
An exact result for El obtained [30] in the framework of the Debye-Waller

theory for the quadratic Hamiltonian Ho + He, yields const = eC /1r 2 , with
C ~ 0.5772 being Euler's constant. Finally,
q(N)
eC
-e- = --Irl
sm21rN.
1r
(88)
371
6. Conclusion
In this paper we have reviewed recent theoretical results for transport in a
one-dimensional (1d) Luttinger liquid. For simplicity, we have ignored electron spin, and focussed exclusively on the case of a single-mode. Moreover,
we have considered only the effects of a single (or perhaps several) spatially localized impurities. Even with these restrictions, the predicted transport behavior is qualitatively different than for a non-interacting electron
gas. Specifically, for repulsive interactions, even a weak impurity potential
causes complete backscattering, and the conductance vanishes completely
at zero temperature. This can be understood in terms of the vanishing
density of states, to tunnel an electron throught the barrier from one semiinfinite Luttinger liquid to another. At finite temperature the tunnelling
electron has finite energy, and the conductance is non-zero, varying as a
power of temperature. The precise power law depends on the dimensionless
parameter (conductance) 9 which characterizes the Luttinger liquid.
For a very weak barrier, at elevated temperatures, the backscattering
is weak, but grows rapidly with cooling. This growth can be understood
physically in terms of the fact that the discrete backscattered charge in this
limit, is less than the electron charge e, but rather given by ge with 9 < 1
for a repulsively interacting electron gas (see Eqn. (2.21)). For inter-edge
tunnelling in the fractional quantum Hall effect, this process corresponds to
the backscattering of a Laughlin quasiparticle (with 9 = l/ = 1/3, say), but
such fractionally charged excitations are in fact a generic property of the 1d
Luttinger liquid, and should be present, for example, in narrow quantum
wires. The fractional charge might be directly measureable via shot-noise
experiments through a single impurity.
The transport behavior in a 1d Luttinger liquids is of course much richer
once one relaxes the restrictions of a single impurity and a single channel.
The case of a single channel with many impurities has been considered by a
number of authors [31 J. In the absence of electron interactions, all states are
localized in 1d, and the system is an insulator. With repulsive interactions,
localization effects are enhanced, and the system is insulating, even though
the notion of single particle localized eigenstates is no longer operative.
However, for spinless electrons with sufficiently strong attractive interactions, the system is predicted to undergo an insulator-to-metal transition.
In the metallic phase, the conductivity is predicted to diverge as a power
law of temperature (with a variable power greater than one), in contrast to
a normal 3d metal, in which there is a finite residual resistivity at T = o.
While the electron interaction in a quantum wire is clearly repulsive, a long
skinny gate across a fractional quantum Hall fluid at, e.g., filling l/ = 1/3,
creates a system which is isomorphic to a (spinless) 1d electron gas with
372
strong attractive interactions [32J. By varying the gate potential along such
a line junction, it should be possible to tune through this Id localization
transition.
For a real one-channel quantum wire, it is of course necessary to take
into account the spin of the electron [22, 33]. The method of bosonization
can readily accomodate the spin degree of freedom. The spinful interacting
electron gas has two modes, one which carries the charge and the other the
spin [3, 7J. These two modes will generally propagate at different velocities, a phenomena known as charge/spin separation. The effects of impurity scattering on the spinful electron gas are qualitatively similar to that
without spin. Specifically, with repulsive interactions, a single impurity will
typically be sufficient to completely backscatter both the charge and spin
modes. The conductance will be driven to zero as a power law of temperature, with possible logarithmic corrections. Resonant tunnelling is also
possible for a spinful electron gas, but can occur in several different guises,
depending on the charge and spin state of the localized state.
In a wider quantum wire, several transverse modes will co-exist at the
Fermi energy. This can also be treated with bosonization. Generally, associated with each channel is one gapless charge mode and one gapless spin
mode. The tunnelling density of states to add an electron, will generally
still be singular, as for a single channel, but the associated exponents become smaller as the number of channels increase [18, 34J. Multiple channels
are also present at the edges of hierarchical fractional quantum hall states
[35], such as at filling 1/ = 2/3 . Multi-component models are also necessary
to treat charge fluctuations under the Coulomb blockade conditions (Section 5.2) in the realistic case of electrons with a spin degree of freedom.
Indeed, a recent theory of Coulomb blockade effects for coupled quantum
dots [36J, has yielded predictions which are in quantitative accord with
experiment [37J.
7. Acknowledgements
We have both benefitted from extensive interactions with numerous friends
and collaborators over the past years. We are particurly indebted to Igor
Aleiner, Harold Baranger, Albert Chang, Steve Girvin, Bertrand Halperin,
Charlie Kane, Anatoly Larkin, Patrick Lee, Hsiuhau Lin, Andreas Ludwig,
Allan MacDonald, Konstantin Matveev, Kyungson Moon, Joe Polchinski,
Igor Ruzin, Boris Shklovskii, M. Stone, Richard Webb and Hangmo Vi. We
gratefully acknowledge the National Science Foundation for support under grants DMR-9400142, PHY89-04035 and DMR-9528578 (University of
California at Santa Barbara), and No. DMR-9423244 (University of Minnesota).
373
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
"Methods of Quantum field theory in statistical physics", by A.A. Abrikosov, L.P.

Gorkov and I.E. Dzyaloshinski, (Dover, New York, 1963).
See R. Shankar, Rev. Mod. Phys. 66, 129 (1994) and references therein.
For a review, see e.g., J. Solyom, Adv. Phys. 28, 201 (1979).
S. Tomonaga, Prog. Theor. Phys. (Kyoto) 5, 544 (1950).
J.M. Luttinger, J. Math. Phys. N.Y. 4, 1154 (1963)
F.D.M. Haldane, J. Phys. C14, 2585 (1981); Phys. Rev. Lett. 47, 1840 (1981).
See, e.g., G.D. Mahan, Many-Particle Physics, 2nd ed. (Plenum Press, New York,
1990), section 4.4.
D. C. Mattis, Phys. Rev. Lett. 32, 714 (1974).
A. Luther and I. Peschel, Phys. Rev. Lett. 32, 992 (1974).
A. Luther and V.J. Emery, Phys. Rev. Lett. 33, 589 (1974).
B.1. Halperin, Phys. Rev. B 25, 2185 (1982).
X.G. Wen, Phys. Rev. B 43, 11025 (1991); Phys. Rev. Lett. 64, 2206 (1990); Phys.
Rev. B 44, 5708 (1991).
L.1. Glazman, G.B. Lesovik, D.E. Khme1'nitskii, and R.1. Shekhter, Pis'ma Zh. Eksp.
Teor. Fiz. 48, 239 (1988) [JETP Lett. 48, 238 (1988)J.
R. Landauer, IBM J. Res. Dev. 1, 223 (1957).
I.E. Dzyaloshinskii and A.1. Larkin, ZhETF 63, 411 (1973).
D.L. Maslov and M. Stone, Phys. Rev. B 52, 5539 (1995); I.Safi and H.J. Schulz,
Phys. Rev. B 52,17040 (1995); V.V. Ponomarenko Phys. Rev. B 52,8666 (1995).
C.L. Kane and M.P.A. Fisher, Phys. Rev. Lett. 68, 1220 (1992).
K.A. Matveev and L.1. Glazman, Physica B, 189, 266 (1993).
L.1. Glazman, I.M. Ruzin, and B.1. Shklovskii, Phys. Rev. B 45, 8454 (1992).
A similar logarithmic divergency with the system size at It I = 00 was found in the
problem of the charge density wave tunneling, see A.1. Larkin and P.A. Lee, Phys.
Rev. B 17, 1596 (1978).
K.A. Matveev, D. Vue, and L.1. Glazman, Phys. Rev. Lett. 71, 3351 (1993) ; D.
Vue, K.A. Matveev, and L.1. Glazman, Phys. Rev. B 49, 1966 (1994).
K. Moon, H. Vi, C.L. Kane, S.M. Girvin, and M.P.A. Fisher, Phys. Rev. Lett. 71,
4381 (1993).
P. Fendley, A.W.W. Ludwig, and H. Saleur, Phys. Rev. Lett. 74, 3005 (1995).
S. Tarucha, T. Honda, and T. Saku, Solid State Commun. 94, 413 (1995).
A. Yacoby, H.L. Stormer, L.N. Pfeiffer, and K.W. West (unpublished).
F.P. Milliken, C.P. Umbach, and R.A.Webb, Solid State Commun. 97, 309 (1996).
A.M. Chang, L.N. Pfeiffer, and K.W. West, Phys. Rev. Lett. 77, 2538 (1996).
K. Flensberg, Phys. Rev. B 48, 11156 (1993); Physica B 203, 433 (1994).
K.A. Matveev, Phys. Rev. B 51, 1743 (1994); Physica B 203, 404, (1994).
See T. Giamarchi and H. Schulz, Phys. Rev. B 37, 325 (1988), and references therein.
S. Renn and D.P. Arovas, Phys. Rev. B 51, 16832 (1995).
C.L. Kane and M.P.A. Fisher, Phys. Rev. B 46, 15233 (1992).
K.A. Matveev and L.1. Glazman, Phys. Rev. Lett., 70, 990 (1993).
C.L. Kane, M.P.A. Fisher and J. Po1chinski, Phys. Rev. Lett., 72,4129 (1994); C.L.
Kane and M.P.A. Fisher, Phys. Rev. B 51, 13449 (1995).
K.A. Matveev, L.1. Glazman, and H.U. Baranger, Phys. Rev. B 53, 1034 (1996).
J.M. Golden and B.1. Halperin, Phys. Rev. B 53, 3893 (1996).
F.R. Waugh, M.J. Berry, D.J. Mar, R.M. Westervelt, K.L. Campman, and A.C.
Gossard, Phys. Rev. Lett. 75, 705 (1995).
THE PROXIMITY EFFECT IN MESOSCOPIC DIFFUSIVE

CONDUCTORS
D. ESTEVE, H. POTHIER, S. GUERON, NORMAN O. BIRGE*,

AND M. H. DEVORET
Quantronics group, Service de Physique de l'Etat Condense,
CEA-Saclay, 91191 Gif-sur- Yvette, France
1. Introduction
The "proximity effect" is a loose term describing the presence of a superconducting order in a piece of normal metal (N) placed in electrical contact
with a superconducting one (8). Because of an action-reaction like principle, the order induced on the N side of the N8 interface due to the presence
of 8 is correlated with a weakening of the superconducting order on the
8 side due to the presence of N. Thus, the proximity effect is actually the
dilution of superconducting order at an N8 interface.
The understanding of the proximity effect is of interest because it is
a simple example of a quantum many-body problem in real space, with
imposed spatial boundary conditions. In this chapter, we consider the following question: what is the exact nature of the superconducting order in
a disordered normal metal which is in contact with a BC8 superconductor?
As we will see, the answer is subtle because the order is not BC8-like. For
instance, even though a short N diffusive wire connecting two large 8 electrodes can sustain a supercurrent, the same wire, but connecting a large
N electrode to a large 8 one, is resistive and has, at zero temperature, the
same resistance as in the normal state.
The proximity effect was analyzed within the framework of the GinzburgLandau theory in the late 60s [2]. In this phenomenological approach, the
superconducting order is described by a single complex order parameter
W(r) whose modulus is related to the amount of pair correlations and whose
argument corresponds to the superconducting phase. The spatial variations
of w(r) are determined theoretically by minimizing the Ginzburg-Landau
free energy functional. A more rigorous microscopic approach was later pro375

376
vided by the Bogoliubov-de Gennes (BdG) equations [3]. These equations

are the Schrodinger equations for the wavefunctions of the electron-like and
hole-like electronic fluid excitations, which correspond to doubly occupied
and empty electron states, respectively. In the case of NS samples, these
equations are coupled, and the eigenstates describe quasiparticles which
are a coherent superposition of electron-like and hole-like excitations of the
normal state. Although the BdG theory formally solves the problem of the
proximity effect, the BdG equations can be explicitly solved only in the socalled ballistic regime, in which the electron eigenstate wave-functions in
the normal state are plane waves. In disordered metals, where the electron
propagation is diffusive, the BdG equations can only be solved numerically
[4]. The first part of this chapter shows how a more comprehensive understanding of the proximity effect in the diffusive regime, which is well
adapted to metallic NS systems, has emerged in the past years. This recent
understanding is based on the theory of non-equilibrium superconductivity
[5], developed in the 70s, and on recent methods of mesoscopic physics, such
as random matrix theory. The second part of this chapter discusses how the
experimental results on the proximity effect in the diffusive regime fit into
this framework and compare with the somewhat non-intuitive predictions
of the theory. It is worthwhile to emphasize that the recent mesoscopic
revival of the proximity effect is a by-product of the quest for quantum
coherence effects in mesoscopic systems. The proximity effect is in fact one
of the most spectacular mesoscopic quantum phenomena since it can modify strongly transport properties. Roughly speaking, the proximity effect
involves all conduction channels whereas standard mesoscopic phenomena
such as weak-localization and conductance fluctuations are said to be one
effective conduction channel effects.
1.1. PROXIMITY EFFECT AND MESOSCOPIC PHYSICS
The proximity effect involves an intermediate (=meso) length scale was

found well before the word "mesoscopic" was coined. It had been calculated [3] that the Ginzburg-Landau order parameter decays exponentially
in the N side of a good NS contact with a characteristic thermal length
LT =
k~T' where D and T are the diffusion constant and the temperature respectively. The observed temperature dependence of the critical
currents of SNS bridges was found to be in agreement with this prediction in the regime when the sample length L is longer than the thermal
length LT. In the opposite L ~ LT regime which has been probed only
recently, the Ginzburg-Landau description of the pair correlations by a single order parameter is in fact oversimplified because the pair correlations
at different energies have different spatial dependences, as shown by the
377
following argument. In a bulk superconductor, the ground state wave function is the product of coherent superpositions of empty and doubly occupied
time-conjugated electron states of different energies E with respect to the
Fermi level. The relative phases between the empty and doubly occupied
state amplitudes are the same at all energies because the effective pair interaction "aligns" them in order to minimize the interaction energy. This
common phase is called the superconducting phase. When a pair enters the
normal metal, the empty and doubly occupied states become eigenstates
with an energy difference 2E. Their amplitudes dephase in time by a factor
exp (-i2Etjh) , where t is the time elapsed after the pair has left the superconductor. The pair correlations induced in the normal side at a distance
L from the interface are obtained by summing the contributions of all the
diffusive paths reaching this point. In a diffusive sample, in which paths
with end-to-end length L have a duration L2 j D, the dephasing factors of
the pair correlations are spread out at distances larger than VhD j E. Thus,
the pair correlations decay away from a NS interface with an energy dependent characteristic length VhD j E. At small energies, the propagation of
the superconducting order should however be limited by the loss of electron
phase coherence due to spin-flip or other inelastic processes. The formalism
described in this chapter substantiates this simple diffusive picture for the
propagation of energy-dependent superconducting correlations.
1.2. THE GREEN FUNCTION DESCRIPTION OF PROXIMITY EFFECT
The Green function description of superconductivity provides a comprehensive framework which contains the BCS theory of bulk systems but
which is also capable, as is the Ginzburg-Landau approach, of handling
situations where the superconducting order varies in space. It also has the
great advantage of merging in a unified formalism both equilibrium and
non-equilibrium aspects of NS systems.
In the theory of non-equilibrium superconductivity [5, 6, 7], the space
and energy-dependent correlations between two time-conjugated electron
states are described by Green functions which are 2 x 2 matrices, called
il, ..4, K . These Green functions are statistical quantities averaged over
the different sample realizations, as indicated by their full name "impurity
averaged Green functions". They depend on position and energy, and are
related to the annihilation and creation operators 'ltt(r, t) and 'It i(r, t) of
an electron with spin t at position r in the Heisenberg representation by
the following relations:
378
(1)
(2)
K(r,c:) =F
[-i (([w[t~r,t),w+~r,t')~)])
-( w.).(r,t),wt(r,t)
(([w[t~r,t),w.).(r,t')'~l))]'
w.).(r,t),w.).(r,t)J
(3)
where F denotes the Fourier transform, {O, O'} = 00'+0'0 and [0,0'] =
00' - 0'0. The Green functions il and A are called respectively the retarded and advanced Green function because of the () function which selects
t > t' for the retarded one t < t' for the advanced one. These Green functions describe the ground state and the quasiparticle excitation spectrum.
The Green function K is called the Keldysh function, and contains the information on the filling factors of the quasiparticle states. At equilibrium,
K is proportional to the difference il - A. In out of equilibrium situations,
when the quasiparticle filling factors are not given by a Fermi distribution, K is related to il and A by the relation K = ill - lA, where 1 is
the quasiparticle filling matrix. The goal of the theory is to determine the
Green functions il, A, K and to calculate afterwards the physical quantities
of interest.
The general relations A2 = i, il2 = i and ilK + KA = 0 allow to
arrange these Green functions to form a 4 x 4 matrix noted G:
-.
G(r,c:)
(il0 K)
A .
(4)
In the diffusive regime, the impurity averaging technique directly leads to

the following central equation for G(x, c:) :
where lio
(floo Ho0)
A '
~.
,t
0)
= (Ti.0 Ti
379
with
fio
A
Ty
=(
=
(0 -i)
0
;
i
Tz
A
(10
E is the energy; ~ is the effective pair potential that must be determined

self-consistently; the fi are the usual Pauli matrices; D is the diffusion constant; Ts! is the spin-flip scattering time. We assume here that the inelastic
scattering rate is negligible (a finite rate can be readily incorporated by
adding an imaginary part to the energy E).
The self-consistent equation for ~ is [7]:
{'twD
~=noVe!! 10
1
dE 4i Tr[(fx+ify)K]
A
(6)
where no, Vel! and WD are the normal metal density of states at the Fermi
level, the strength of the pair interactions and the Debye frequency, respectively. The central equation (5) must be supplemented with the boundary conditions for the various Green functions in the different electrodes
of the circuit. The equations obeyed by the Green functions R, A, K,
called the Usadel equations [8], are obtained by projecting the central equation (5). In presence of a magnetic field, the central equation is the same as
Eq. (5) but with the derivative operator '\7 replaced by the covariant form
('\7 - 2ieA/n).
Compared to the BdG equations, this formalism does not require solving the Schrodinger equation for the microscopic wavefunctions but directly
leads to a set of diffusion-like equations for averaged quantities which, as
will be seen, are directly related to physical properties. "Dirty" NS structures made of diffusive metallic thin films are, in this sense, perfect simple
systems for comparison with theory.
2. Green function description of the proximity effect at equilibrium

In this section we discuss the physical meaning of the Green functions
and summarize the main theoretical results at equilibrium [9]. In a normalsuperconducting (NS) nanostructure, the electron pair correlations described
by R and A vary in space and interpolate between those of the Sand N
reservoirs to which the structure is connected. We will see that the equilibrium Usadel equations for R and A derive from a variational principle
which generalizes the Ginzburg-Landau variational approach to an energy
380
dependent order parameter. This approach, previously used by Eilenberger

to obtain a simpler formulation of the self-consistent BeS equations [10],
extends the framework developed by Nazarov [11] to finite energy and finite
magnetic field [12, 13]. It also provides physical insight into the proximity
effect, as illustrated by simple examples of NS structures combining diffusive wires and tunnel junctions.
2.1. PARAMETRIZATION OF GREEN FUNCTIONS BY ANGLES ON THE
COMPLEX UNIT SPHERE
We use the following angular parametrization for the retarded

vanced A Green functions:
R
+ sin O( cos cpfx + sin cpfy)

- cos jJfz + sin jJ( cos cpfx + sin cpfy)
cos (}fz
R and ad(7)
where O(x, E) and cp (x, E) are the complex superconducting order angle
and the real superconducting phase. We note that the superconducting
phase cp (x, E) depends on energy in contrast with the case of a bulk superconductor. We define the complex unit sphere as the set of points with
polar coordinates 0 and cp, for complex 0 and real cpo The proximity effect in
a NS structure can thus be described by assigning to each position x along
the circuit and to each energy E a representative point P on the complex
unit sphere. At a given energy, these representative points P form a trajectory which is continuous except at tunnel junction interfaces. At a contact
with a large normal electrode, which acts as an N reservoir, P is located at
the north pole of the sphere (0 = 0). At a contact with a large superconducting electrode (i.e. a S reservoir), P is located at an angle 0 = OBGS (E)
with an energy independent phase cp equal to the superconducting phase of
the reservoir. The identity of the phases at different energies is imposed in
the reservoirs by the pairing interaction. The superconducting order angle
corresponding to the BeS theory is given by sinOBGs (E) = 1:::../v'1:::.. 2 - E2
where I:::.. is the BeS gap. The parametrization takes a particularly simple
form at E = 0, where all trajectories lie on the real unit sphere (see Fig. 1)
and join the north pole N to the equator 0 = OBGs(E = 0) = 7r /2.
2.2. DENSITY OF STATES
The equilibrium properties of the proximity effect are completely determined by the set of points P. The density of states n (x, E) at position x
and at energy E is related to the superconducting order angle O(x, E) by
the relation:
381
E=oo . -_ _E_=_O-+_-.
1'f2
Re 8
Figure 1.
Left: Representation of the parametrization of the advanced and retarded
Green functions describing the proximity effect on the unit sphere. At zero energy, the
point P follows a trajectory going from a normal reservoir (N) to a superconducting reservoir with super conducting phase cp(S). Right: Variations of the complex superconducting
order angle () with energy in the case of a BCS superconductor with gap ~.
n (x, E)
= noRe [cos 8(x, E)]
(8)
where no is the normal density of states at the Fermi level. At thermal

equilibrium, the quasiparticle state filling factor f (E) which represents the
probability of finding a quasiparticle in an excitation state with energy E
is given by the Fermi function
temperature.
f (E) = (1 + exp k:T) -1 , where T is the
2.3. SUPERCURRENT DENSITY
If a NS structure contains either S reservoirs biased at different phases or

loops threaded by a magnetic field, equilibrium supercurrents will flow in
the sample. As in a BeS superconductor, the supercurrent density js is
obtained from the covariant phase gradient [7]:
where 0', e, and A are respectively the N state conductivity, the electron
charge and the vector potential. At high temperature, the dominant contribution to the integral arises from small energies and one can neglect the
382
energy dependence of the phase. If we identify then the above expression

with the Ginzburg-Landau result is = -~ 1'1112 (Vcp + ~ A), one finds:
1'1112
4~1i aD 10+
00
dE tanh
(2k~T) 1m [sin20] .
(10)
The Ginzburg-Landau order parameter is a weighted average ofIm [sin2 0] ,

which can be understood as an effective energy dependent "pair density".
2.4. THE EQUILIBRIUM USADEL EQUATIONS
We consider here wires with lateral dimensions small enough to ensure that
the proximity effect depends only on one space coordinate x along each wire.
A sufficient condition is that these dimensions are kept smaller than the
London penetration length in the superconductor. In this one-dimensional
regime, the Usadel equation for R deduced from Eq. (5) reads:
a (A a A) [ A A] 1 [ A A]
D ax R ax R + i Ho, R - Ts! TzRTz, R
With the parametrization (7), the Usadel equations for
following form:
a [( acp
2e ) . 2 ]
ax + h Ax sm 0
ax
= O.
(11)
0 and cp take the
(13)
where Ax is the vector potential component along the wire and D.(x) the
pair potential. The self-consistency equation for D.(x) is:
D. (x)
= noVe!! fohwD dE tanh (2k~T) 1m [sinO (x, E)]
expicp (x, E).
(14)
We note that this equation does not contain the effective pair density appearing in the expression (9) of the supercurrent, and cannot be expressed
as a function of the Ginzburg-Landau order parameter.
2.5. THE PROXIMITY EFFECT IN A N WIRE IN CONTACT AT ONE END
TO A SWIRE
We first discuss the simple case of two semi-infinite normal and superconducting wires in contact at x = O. In this sample geometry, the reservoirs
383
are located at x = oo. We assume that a tunnel barrier with normal

state resistance RT is located at the interface. The Usadel equations (1213) in the different electrodes are supplemented with boundary conditions
at the reservoirs and at the interface. The phase is equal at all energies to
the superconducting phase of the single S reservoir. Far from the interface,
ON = in the normal metal, and Os = OBCS in the superconductor; at the
interface, the discontinuity of 0 is related to the gradient of 0 on both sides
through the relation:
ax
[JON s)
-1 .
aN,S (
x=o = (aRT) sm (OS (0, E) - ON (0, E))
(15)
where ax is the normal state conductivity of electrode X and a is the

contact area [21]. In the case of a good contact, Os (0, E) = ON (0, E) . The
resolution of the Usadel equation is greatly simplified if b. is assumed to
be independent of x in the superconductor, in which case Eq. (12) can be
integrated once:
tiD
4
([J(}) 2
Ti
[Jx
- iE cos 0 + 4TsJ cos 20 + b. sin 0 = F (E) .
(16)
The superconducting order angle 0 is obtained by a second integration

performed numerically. The angle 0 significantly departs from its BeS value
over a distance of order Lb. = JTiD / b., which is the coherence length
entering into the Ginzburg-Landau theory. Further from the interface, 0
decays down to zero except at small energies. Analytical expressions have
been obtained in this regime [9]. In particular, the superconducting order
angle takes a simple universal form at distances x from the contact such
that Lb. x Lcp = JDTsf :
(17)
where Ec = tiD / x 2 is often called the Thouless energy associated to the
length x . In this regime, the reduced density of states n (x, E) /no only
depends on the reduced energy E = E / Ec (see Fig. 2) and is gapless. The
N character of the metal is fully recovered at distances x Lcp. A complete
numerical resolution of the self-consistent Eqs (12, 14) has recently been
performed [14]. It shows that the BeS gap is recovered in the superconductor at a distance of the order of Lb.. Finally, let us mention that the
density of states in the wire develops a gap of order Ec when the length of
the normal wire is finite and short compared to Lcp, a situation similar to
that of NS bilayers [15].
384
./
p;e} /
1
E
Figure 2. Reduced density of states p (f) in the universal regime. f = E / Ec is the
reduced energy. The density of states in the normal wire at distance x and at energy E
is n(x,E} = nop(E/Ec}.
3. A variational principle for the Usadel equations

3.1. THE EFFECTIVE POTENTIAL U
We define the following dimensionless potential:
u=
1,
vol.
(18)
U dv
with the density U given by:
no [liD
4
(dS)
dx
2 _
!!.... sin2 (} + D. (x) sin (}].
iE cos (} -
TsJ
(19)
The gauge invariant metric is defined by:

ds 2 = d(}2
+ (drp -
drpA)2 sin2 (}
(20)
with drpA = -~ Ax.dx. The integral is taken over the volume of the sample.
The phase field rp A (x) results from the applied magnetic field but also from
the field produced by the S currents in the structure. In the following, we
will assume that the superconducting currents are too small to significantly
screen the applied magnetic field and that rp A (x) is an externally applied
lip
phase field in a given gauge. The first term

(~!) 2 in U can be understood as an elastic energy term for the trajectory, while the other ones are
potential energy terms.
3.2. THE VARIATIONAL PRINCIPLE
It is straightforward to check that the variational equation 8U = 0 with

respect to (} and rp yields the Usadel equations (12-13). The set of Us-
385
adel equations at all energies generalize the G-L equation for 'l1. As in
the Ginzburg-Landau energy functional whose minimization leads to the
Ginzburg-Landau equation, the density U given by Eq. (19) is the sum of
a covariant squared gradient term and potential terms. At a given energy,
the global potential U is a functional of the trajectories followed by the representative points P on the complex unit sphere. This treatment provides
physical insight into the different terms appearing in the Usadel equations
and is a direct way to numerically solve the Usadel equations.
3.3. THE PROXIMITY EFFECT AT ZERO ENERGY
The Usadel equations (12-13) are more easily solved at zero energy where
their form is simpler. The solution is further simplified if the length of
the mesoscopic structure is smaller than the phase coherence length Lrp =
JDTsf. The variational principle is particularly useful in this case because
the contribution to the global potential U of each circuit element (diffusive
wire or tunnel junction) is a simple function of the position of its representative end points on the unit sphere. The problem is thus reduced to
the determination of the positions on the unit sphere of the representative
points of the nodes of the structure.
In the case of a N wire with length L and with representative end points
P and Q, the minimal potential trajectory is the geodesic of the unit sphere
with the metric (20). The trajectory thus follows the circular arc joining P
to Q with the shortest possible length C, and ~; = C/ L is constant along
it. The contribution u to the global potential of the diffusive wire is then
easily calculated:
(21)
where RK = h/e 2 and R is the wire resistance. In the absence of a magnetic
field, C is simply the length of the arc PQ. A magnetic field acts as an extra
rotation field which affects the length of a trajectory in the same way that
winds modify the effective distances airplanes must travel around the earth.
The length C is in this case the length of the arc PQ' where Q' is at the same
latitude as Q but at a longitude increased by (CPA (Q) - CPA (P)) . If instead
of a wire we consider a tunnel junction of normal state tunnel resistance
RT, the contribution u' to the global potential takes the form:
(22)
,2
= 2 (1 - cos C) is the square of the geometrical distance between

where
the representative points of both sides of the junction.
386
The global potential for the whole structure is simply the sum of the
contributions of the different elements. Finding the minimum of the global
potential is equivalent to finding the equilibrium positions of a set of springs
with the following correspondence rules:
a) Each wire R (resp. each tunnel junction RT) is represented by a
spring of stiffness R- 1 (resp. IC[.1).
b) The springs are connected in the same way as the elements they
correspond to. The arcs of springs forming a loop are modified as explained
above.
c) A wire-type spring is constrained to lie on the unit sphere.
d) A junction-type spring lies along the cord joining its end points.
These rules are equivalent to the conservation law of the spectral current
at the nodes found by Nazarov [11]. The determination of the angles () and
c.p at zero energy gives the density of states at zero energy (8) and allows
one to calculate the transport properties at zero temperature.
A simple circuit with a loop and its representation on the unit sphere are
shown in Fig. 3. When the loop encloses a flux 1>, the two springs associated
with the two branches of the loop correspond to different arcs on the unit
sphere because the vector potential is different on the two branches . The
trajectories corresponding to the two branches are shown for the gauge
choice c.p = 7rx1>/L1>o, where x is the running position along each branch
with length L, and 1>0 = h/2e is the flux quantum. The effective length
of the two arcs is A' B= A" B' The total potential U of the circuit is the
sum of the potential (21) of each of the four branches. The superconducting
order angles at nodes A and B can easily be determined by minimizing the
potential U. The variations of (}A and (}B with flux 1> shown in Fig. 3 have
been calculated in the particular case of equal resistances RNA, RAB and
RBs
3.4. THE PROXIMITY EFFECT AT FINITE ENERGY
Even in the simple case of a short wire between two Nand S reservoirs (see
Fig. 4), the direct integration of the Usadel equations is not straightforward
because the boundary conditions are given at two different positions. The
variational calculation provides an alternative method in order to circumvent this difficulty. The variations of the complex superconducting order
angle () with the position x along a wire of length L = 5Lfl. are shown in
Fig. 4 for different values of the energy, in the absence of spin-flip scattering.
4. Transport in presence of proximity effect
Let us suppose now that the equilibrium problem in a general NS structure

has been solved, i.e. (}(E,x) and c.p(E,x) are known. What happens if the
387
0.5
Figure 3. Simple NS circuit with a loop (enclosing a flux ) connected to Nand S

reservoirs through equal resistors. The flux determines the opening angle <p = 27l' /o
between the representative points A and A' obtained by following the two arms of the
loop. The two curves show the variations of the polar angle (J at points A and B as a
function of the flux , calculated using the variational principle.
2------~----------~--~
I
I
1.1
_ _ _~10.9
- - - - - - - - { 0.8
~
~04
____
.. N
.'
o
-1t/4
0.6
1t/4
0.2
1t/2
Re(8)
Figure 4.
Variations of the proximity angle (J along a N wire of length L = 5Lt>,
connected at its ends to a normal and to a superconducting reservoir, for different energies. The distance between successive points is 0.2Lt>,. The energies are
E/b.. = 0,0.2,0.4,0.6,0.8,0.9,1.1. We have assumed here L Lop. Each curve goes
from (J = 0 to the BeS angle at that energy. At E / b.. = 0, (J is real and is given by a
linear interpolation between the N value (J = 0 and the S value (J = 7l' /2.
388
potentials of the N electrodes are changed? The equilibrium Usadel equations are supplemented with equations for the out of equilibrium Green
function K = ilj - j A. The population of the quasiparticle states is no
longer given by a thermal equilibrium distribution and is space dependent.
It is imposed by the reservoirs at the boundaries of the system. The Keldysh
Green function K, which describes departure from the equilibrium situation, obeys the following equation derived from Eq. (5):
axa (AR axa KA+ KAaxa AA) + i [HAo, KA] -
1 [A
A]
TsJ fzK fz, K
= o.
(23)
Once this equation is solved, the current at any place in the structure
can be calculated using the following expressions [7]:
o In a wire with cross section area S and conductance a :
I =
as
&
1+
00
-00
dE Tr {fz(R aK
ax
A
+ K aA)
ax } .
A
(24)
o At a tunnel junction of normal state conductance GT between right r

and left 1 electrodes :
Irl = GT ~
8e 2
1+
00
-00
dE Tr {(fz
The filling matrix

under the form[5, 6]:
+ i) (-ilrKI - KrAI + RIKr + KIA r }.
j , which relates K
to
il
and
.4,
(25)
can be expanded
(26)
The filling factors fo and h are respectively the asymmetric and symmetric
parts of the filling function. Their values in a reservoir at potential V are:
fo
1(
= 2"
(E + V)
tanh 2kBT
+ V)
f 1 -- ~2 ( tanh (E2kBT
+ tanh
_
tan
(E - V))
2kBT
(27)
h (E - V))
2kBT
.
(28)
In the case of a wire with constant superconducting phase, the equations

for fo and h deduced from Eq. (23) read:
(cos 2 01 V fo)
(cosh2 (0 2) V h)
=0
- 2~Re [sinO] h
(29)
=
0,
(30)
389
where (h and (h denote the real and imaginary parts of the superconducting
angle (). In a normal wire with constant phase, the current is simply given
by:
as
1=-2
e
1+
00
-00
dE cosh2 ()2 \7 h
(31)
This expression is similar to the result obtained in the absence of the

proximity effect but with a renormalized diffusion constant D cosh 2 ()2 which
is energy dependent [17, 18]. The current is obtained by integrating a "spectral" current ~~ cosh2 ()2 \7 h which is conserved at each energy. The spectral current is not proportional to the gradient of the quasiparticle density
at the energy considered, as one could expect for a dilute gas of diffusive
quasiparticles, but to the gradient of the filling factor. The quasiparticles
which represent excitations of the condensate cannot be considered as local
objects and the current cannot therefore be identified to the quasiparticle
flux. We now discuss the predictions of this theory for some simple NS
structures.
4.1. SUPERCURRENT IN SNS STRUCTURES
When the different superconducting electrodes are at the same potential,

the phases are time independent and the supercurrent density is readily
calculated once the angles () and <p are known at all energies using Eq.
(9). Supercurrents can flow in the normal metal between S reservoirs with
different phases, or around a loop enclosing a magnetic flux as in the case
shown in Fig. 3. The current-phase relation has been derived in some cases
of simple SNS structures. In the particular case of a short normal wire of
length L and normal resistance RN in good contact with two superconducting electrodes at its ends, the predicted critical current Ie, i.e. the
maximum supercurrent that can be sustained by the wire, is different from
the de Gennes result [3]. More precisely, the RNIe product at zero temperature is not proportional to the gap energy .D. but to the Thouless energy :
RNIe ~ Ee/e. The predicted temperature dependence of Ie [19,20] is also
qualitatively different. When the different superconducting electrodes are
at different potentials, the phases vary in time and an alternating supercurrent flows between the superconducting electrodes. We do not consider
this situation here.
4.2.
TRANSPORT IN NS STRUCTURES AT ZERO TEMPERATURE
4.2.1. Circuit theory in the presence of proximity effect
Nazarov has derived the relation between the conductance of a NS circuit

at zero temperature and the angles () and <p at zero energy along the various
390
branches of the circuit [11]. His derivation is based on the low energy limit
of the non-equilibrium Usadel equations given above. At zero temperature,
and for small applied voltages, Eq. (30) is a simple diffusion equation with
the normal state diffusion constant since 02 = 0 at zero energy. The introduction of the fictitious potential C; = J!":: dE iI, which coincides at the
reservoirs with the electric potential, further simplifies the resolution of the
current distribution in the circuit. Eq. (31) reduces to I = - (aB/2e) 'Vc;,
which implies that the current in a wire AB with normal state conductance 9 is simply I = g(C;A - C;B). In presence of the proximity effect, the
usual rules of circuit theory apply to the effective potential C; instead of the
usual electric potential. A similar rule does not apply to tunnel junctions,
which have a different behavior in presence of proximity effect. The normal
current through a tunnel junction with normal state conductance GT, calculated using expression (25), is I = GT cos I:- (C;A - C;B) where I:- is the angle
between representative points A and B of both sides of the junction on the
unit sphere. The discontinuity of the representative point on crossing the
tunnel barrier results in the reduction of the conductance by a geometrical
factor determined by the equilibrium positions of the representative points
at zero energy. In particular, the effective tunnel conductance of a tunnel
junction between perfect Nand S reservoirs is zero. The conductance of
any circuit can be obtained by applying the usual rules of circuit theory
but with renormalized elements:
a) Shunts are installed between all the superconducting electrodes.
b) The diffusive wire conductances are unchanged.
c) The tunnel junction conductances are renormalized according to
G'T n = GT cos 1:-, where I:- is the angle between representative points at
zero energy of both sides of the junction on the unit sphere.
4.2.2. Proximity effect and Andreev reflection

We note that the case of diffusive wires is markedly different from the case
of ballistic NS point contacts, for which Andreev reflection results in a
doubling of the conductance compared to the NN case.
The renormalization of the tunnel resistances can be understood without resorting to the arsenal of the Usadel equations. At temperatures much
less than the critical temperature, single electron tunneling does not contribute to the subgap conductance G which is entirely due to two-electron
tunneling. This phenomenon is known as Andreev reflection of an electron
as a hole, a pair being added to S. Let us consider the simple circuit consisting of a small resistance r between a N reservoir and a NS tunnel junction of
normal state conductance GT. A direct golden rule calculation of the twoelectron current yields the surprising result G = rGf for the conductance
of the whole circuit [22]. In contrast with the NN case, the conductance
391
Figure 5.
Example of diffusive NS structure in which the resistance is strongly
modified by the proximity effect. The resistance R between the Nand S reservoirs
R = R1 + (R; 1 + R:;1 ) -1 depends on the resistance R3 of the dangling arm.
is a global property of the NS structure and not of the tunnel barrier itself. This result explains why the observed Andreev conductance in NS
junctions is larger than the ballistic model prediction of order RKG?r/N,
where N is the number of transverse channels. This effect was first found
in semiconducting-S systems in the form of an excess zero voltage conductance [23]. In the proximity effect language, the interpretation is completely
different [24]: the proximity effect on the normal side n of the junction is
weak because electrons on that side are more coupled to the N reservoir at
"distance" r than to the superconducting condensate at a larger "distance"
G"Tl. The perturbative calculation of the proximity effect angle On at zero
energy on both sides of the junction yields On ~ rGT, Os ~ 7r /2. Finally,
the result G = rG?r is obtained from the general expression 9 = GT cos"
where, = (On - Os) ~ (rG T - 7r/2) is the angle between representative
points of both sides of the junction. In this regime, Andreev reflection is
simply another description of the proximity effect.
4.2.3. Does proximity effect modify the conductance of a N wire?
In the simple case of a N structure in good contact with a single large S

reservoir, the resistance at zero temperature is the same as in the normal
state. This surprising prediction agrees with the extension to NS structures of the random matrix theory, widely used to describe transport in
mesoscopic N structures [25]. However, one should not think that proximity effect does not change the resistance of all NS structures! The rules
given above can indeed lead to curious results, as in the case of the circuit
shown in Fig. 3 a: the resistance between the Nand S reservoirs depends on
the diffusive resistance connected to the dangling S' reservoir. Although no
net current flows in the dangling arm, a supercurrent and a quasiparticle
current flow in opposite directions. The shunt between the superconducting
electrodes is effective as long as the bias current is smaller than the maximal
supercurrent that can flow between both superconducting electrodes.
392
4.3. RESISTANCE OF NS STRUCTURES AT FINITE TEMPERATURE
4.3.1. Resistance of a normal wire placed between Nand S reservoirs
The simplest structure one can think of is a normal wire with length L, in
good contact at its ends to Nand S reservoirs. What is the resistance measured between the two reservoirs? The solution of Eq. (30) in this geometry
is:
(32)
The conductance of the wire is then given by [16, 18]:
G = GN
00
dE
-00
2.f1.
2T cosh 2T
L
L
fo
dx'
(33)
COSh292(X')
This expression predicts a weak temperature dependence with a reentrant

behavior at low temperature. At higher temperatures, the resistance behaves as if a length of order of the thermal length LT = J1iD j k B T was
effectively superconducting. The minimum occurs when LT is of the order of the sample length. At zero temperature, the resistance recovers its
normal state value, in agreement with the result given previously. The variations of the resistance with the temperature are shown in Fig. 6 for a
wire of length L = 5Lt:,., for which the largest reduction is 14% and occurs at T ~ 0.25 !:l.jkB' i.e. at L ~ 0.4L. Although the quasi-particle
density of states in the wire can be strongly affected by the proximity effect, the transport, which is entirely due to the diffusive motion of the
quasiparticles, remains almost unaffected! However, the proximity effect
does modify the electric field close to the NS interface: As intuitively expected, the electric field is expelled from a part of the normal metal, close
to the contact with the superconductor [17]. The full electric potential
profile V (x) can be deduced from the function h (.::, x) using the relation
eV (x) = f dE h (E, x) Re cos (} (E, x). Detailed calculations have been performed in the case of more complex geometries [18].
4.4. SUMMARY OF THE THEORETICAL PREDICTIONS ON TRANSPORT
IN NS STRUCTURES
We have considered NS structures connected to normal reservoirs biased at

potentials which are different from the common reference potential of all the
superconducting reservoirs. The theory makes very different predictions for
diffusive wires and for tunnel junctions : the resistance of a normal wire is
only weakly affected at intermediate temperatures by the proximity effect
393
1.0
r-----------------,
0.8
'----~---.L..1---~----'2
keTf /).
Figure 6.
Temperature dependence of the resistance of a normal wire with length
L = 5LC!. placed between a normal and a superconducting reservoir.
whereas the conductance of a tunnel junction is strongly renormalized.

When several superconducting electrodes are in good contact with the same
piece of normal metal, there is furthermore an effective shunt between all
these electrodes.
We now discuss a few basic experiments which test the theoretical predictions on proximity effect in equilibrium and out-of-equilibrium situations. The first experiment is reported in detail in order to provide a description of a typical experiment in this field, whereas only the results of
the other ones are given.
5. Experiments on proximity effect at equilibrium

5.1. DENSITY OF STATES
In this subsection, we report an experimental test of the predictions of the

Usadel equation (12) for the density of states n (x, E) = noRe [cos (} (x, E)]
in the simple case of a long normal wire in good contact at one end with a superconducting wire [26]. We have obtained the density of states n (x, E) by
measuring the differential conductance dI/ dV (V) of a weak transparency
tunnel junction between the wire at position x and another normal metal
electrode. In this simple situation, the Eq.(25} indeed predicts that the differential conductance dI/ dV (V) of the junction is simply proportional to
Re [cos (} (x, eV)] and therefore to n (e V) . This general result is at the root
of tunneling spectroscopy [27].
5.1.1. Experimental setup
Figure 5 shows a photograph of our sample, which consists of two similar circuits. On the bottom one, two copper electrodes (called "fingers"
in the following, and labeled Fl and F 3 ), are in contact through large re-
394
sistance tunnel junctions with a normal wire N, whose left end makes an
overlapping contact with a superconductor S. On the top circuit, a single
finger, labeled F 2, is placed at an intermediate distance from the NS contact, between Fl and F 3. The three fingers, positioned 200,300 and 800 nm
from the left end of the normal wire, constitute the tunneling spectroscopy
probes. Since the quality of the NS contact is known to be a critical parameter in the proximity effect [3], all the layers were deposited through
a suspended mask in a single vacuum process [29]. The mask, made of
germanium, was fabricated bye-beam lithography followed by reactive ion
etching. We first evaporated 20 nm of aluminum perpendicular to the mask
in order to obtain the S superconducting electrode. We then immediately
evaporated 25 nm of copper at an angle to obtain the'N normal wire. The
angle was chosen so as to produce an overlap with the aluminum electrode
on the left, presumably making a good contact. The insulating barrier was
grown from two 1.4 nm-thick layers of aluminum oxidized in a 80 mbar O2
(10%) Ar (90%) mixture for 10 minutes. Lastly, we evaporated 30 nm of
copper at an angle to produce the fingers F l ,2,3. In order to separate the
three shadows of the mask, the MAA resist layer carrying the germanium
mask was overetched. This was obtained with a low-dose pre-exposure of
the sample around the normal wires and the fingers. The parasitic replicas on both sides of the superconducting electrode produced by the angle
evaporations were lifted off in the non-overetched regions. A reference NS
tunnel junction and a long and narrow NS sandwich were simultaneously
fabricated on the chip. The former was used to measure the unperturbed
density of states in the superconducting film while the critical temperature
of the latter provided a lower limit for the transparency of the NS contact.
The sample was mounted in a copper box thermally anchored to the mixing
chamber of a dilution refrigerator. Measurements were performed through
coaxial lines and miniature RC-transmission line filters [30].
5.1.2. Measurements of the tunneling density of states

Using lock-in detection, we measured the differential ' conductance dI/dV
of each of the three probe junctions as a function of the voltage V applied
between the finger and the right end of the normal wire. The differential
conductance displayed a V-shaped groove at low voltages, which became
less pronounced at larger distances from the interface. This behavior is
shown in Fig. 8 where we plot the dI/dV (V) characteristic of the F l , F2
and F3 junctions, taken at 20 mK. We have multiplied each trace by the
resistance Ri = (dI/dV)-l measured at V = 0.3 meV.
The differential conductance of the reference NS tunnel junction (inset
of Fig. 8) is well fitted by a BCS density of states for the superconducting
electrode and yields the energy gap ~ = 0.212 meV.
395
Figure 7. SEM photograph of the sample: a normal (copper) wire N, horizontal, is in

good contact with a superconducting (aluminum) wire S, diagonal on the left, at their
overlap. Two normal (copper) fingers, vertical, labelled Fl and F3, are connected to the
wire through very opaque tunnel barriers. The density of states in the normal wire is
given by the differential conductance of the tunnel junction as a function of voltage. On
a similar device, a third finger, labelled F 2 , is placed at an intermediate distance.
We repeated the differential conductance measurement of the three fingers with an external magnetic field perpendicular to the chip. In Fig. 7
we present the Fl data taken at T = 30 mK for H = 0, 0.06, and 0.1
T. As the field is increased, the groove structure progressively disappears
and, above 0.1 T, only a weak, broad-winged, field-independent structure
remains. This structure, which extends to 3 m V, is the same for the three
fingers. As explained below, we attribute it to single-electron charging effects. When the temperature was increased (data not shown), the groove
structure was progressively washed out, whereas the high-field shape was
unaffected.
5.1.3. Calculation of the tunneling density of states
We have calculated the tunneling density of states with the assumption

that we have a good NS contact. Although the conductance Gint of the NS
interface is not measured, the absence of superconductivity in the sandwich
down to 18 mK provides a lower limit: Gint > 2 S [15]. With such a high
conductance, the perfect contact approximation Os (0, E) = ON (0, E) is
396
1.0
HQ
-1
experiment
>
""C
::::::
""C
0.5
a::
1.0
theory
0.5
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
V (mV)
Figure 8. Top panel: differential conductance of the tunnel junctions at F 1 , F2 and F3
as a function of the applied voltage V, taken at 20 mK (from ref. [26]). The data were
normalized by the differential conductance at V = 0.3 mV : Gl = 0.19 J.1.S, G2 = 0.38 J.1.S,
G3 = 0.27 J.1.S. Inset, differential conductance of the reference NS tunnel junction. Bottom panel: predicted differential conductance at the three distances to the NS contact
obtained from the convolution of the density of states calculated from the Usadel equation (Eq. (12)) with the function P (E) which describes the Coulomb blockade at the
junctions. We used ~ = 0.212 meV for the gap of aluminum, D = 70 X 10- 4 m 2 /s for the
diffusion constant of copper, and "fs/ = 1.5 X 1010 S-1 for the spin-flip scattering rate.
valid. We used in the numerical calculations the value of jj. given by the
measurement of the reference NS tunnel junction and the diffusion constant
DN = 70 X 10-4 m 2 /s in copper deduced from the conductivity of the
was taken as an
wire between Fl and F 3 . The spin-flip scattering rate
adjustable parameter and the inelastic rate was assumed to be negligible. To
account for the finite extension of the overlapping NS contact, we assumed
that the effective NS interface (x = 0) is positioned 20 nm away from the
extremity of the normal wire, in the overlap region, and calculated the
density of states at the center position of each finger.
r;/
397
0.5
~--~----~----~--~--~
0.0
0.1
V (mV)
0.2
Figure 9. Differential conductance as a function of the voltage V measured at 30 mK

and in a magnetic field H = 0 (curve a), 0.06 T (curve b), and 0.1 T (curve c), taken from
ref. [26]. The thin solid line is a fit of curve c using Eq. (34), in which the DOS n (x, E)
was taken constant. It accounts for the influence of single-electron charging effects on the
conductance of a tunnel junction between normal electrodes.
For quantitative comparison of the Usadel theory with the experimental

data, we must take into account the influence of single-electron charging
effects on the conductance of the tunnel junction. At zero temperature, the
differential conductance of the probe tunnel junction at a finger is related
to the density of states through:
rev
dI
1
dV = R t Jon (x, E) P (e V - E) dE
(34)
where R t is the tunnel resistance of the junction and P (E) is the probability for the electromagnetic environment of the tunnel junction to absorb an energy E [28]. Finite but low temperatures can be accounted for
by convolving expression (34) with the derivative of the Fermi function.
For a tunnel junction of capacitance C in series with a resistance R such
that a = 2R/(h/e 2 ) 1, P (E) = a/Eo (E/ Eort - l for E smaller than
Eo = e2 /1faC. The high field data for F l , F2 and F3 are well fitted by
Eq. (34) with n (x, E) constant (see fit of curve (c) in Fig. 7), and yield
a = 0.022. The fit corresponds to R = 300 nand C = 1 fF, in good
agreement with the estimated values.
5.1.4. Comparison between experiment and theory

The comparison between the zero field data taken at 20 mK for the three
fingers F l , F2 and F 3, and the prediction of Eq. (34) using the calculated
density of states n (x, E) is shown in the bottom panel of Fig. 8. The calculation was performed with the value Tsjl = 1.5 X 1010 S-l which provided
398
Theoretical curve
Experimental data
200.0
400.0
600.0
800.0
T[mK)
Figure 10. Comparison between the measured and predicted temperature dependences
of a silver bar periodically covered by aluminum islands (taken from ref. [20] ).
the best overall agreement. As seen in the figure, the theoretical curves
reproduce the general features of the experimental data, especially the evolution of the characteristic energy scale with distance from the NS interface.
An even better agreement can be obtained by solving the Usadel equation
(12) together with the gap self-consistency equation (14) [14]. This exact
calculation shows that the density of states is modified on the S side over
a distance of the order of the coherence length.
5.2. SUPERCURRENT IN SNS STRUCTURES
The critical current of various SNS structures have been measured as a

function of temperature. The comparison between the experimental results
[31] and the theoretical predictions [20] is shown in Fig. 10 in the case of a
silver bar regularly covered by narrow aluminum islands. The agreement is
satisfactory but one should notice that the calculation has been performed
assuming a 1.2 J..t spacing between the islands in place of the stated 800 nm
value. On the other hand, the Ginzburg-Landau theory would not reproduce
the flattening of the curve at low temperature behavior and would fail to
predict the magnitude of the zero temperature critical current.
399
6. Experiments on proximity effect out of equilibrium

We now discuss three experiments on transport in NS structures placed in
out-of-equilibrium situations.
6.1. REENTRANT BEHAVIOR OF THE RESISTANCE OF A NORMAL
METAL WIRE
Charlat et aL [32] have measured the four-wire resistance of a 400 nm-Iong

copper wire with a 200 nm-Iong lateral branch in good contact with a single
aluminum superconducting reservoir. The width and the thickness of the
copper wire were 80 and 50 nm respectively. This side-branch geometry
has the advantage of avoiding any change of resistance due to a shunt of
the normal wire by an overlapping superconducting electrode. The normal
state resistance was 10.1 n,corresponding to an elastic mean free path of
6 nm. The measured temperature dependence of the resistance shown in
Fig 11 exhibits a reentrant behavior, similar to that predicted in Fig. 6,
and is in good agreement with the theoretical predictions for the precise
geometry of the sample. At the minimum, which occurs at a temperature of
about 500 mK, the resistance is reduced by 2.5 %. The interpretation that
the temperature dependence results from proximity effect and not from
weak localization is confirmed by the voltage dependence of the differential
conductance. The reentrant behavior found at low temperature is destroyed
if the applied voltage is too large, as predicted by Eqs. (32 and 33). This
feature thus discards an interpretation in terms of weak localization which
is insensitive to voltage.
6.2. MODULATION OF THE RESISTANCE OF A NORMAL WIRE
Various transport experiments have been carried out on NS structures in

which the proximity effect was modulated either by threading a magnetic
field through a loop or by phase biasing the sample at two different points
[33, 32]. All the resistance modulation patterns found in these experiments
are now attributed to the intermediate temperature effect discussed above.
In particular, Petrashov et a1.(34] have measured the four wire resistance of
a silver wire in good contact with two superconducting reservoirs Sand S'
with a phase difference if> (see Fig. 12). They have observed a 211" periodic
modulation of the resistance which is in quantitative agreement with the
theoretical prediction[18] taking into account the detailed geometry of their
sample, as shown in Fig.13. Although the agreement is satisfactory, the
interpretation of these experiments has been questioned because the phase
difference if> is not fully controlled experimentally [35].
400
10.1
dY IdI
(n)
10
I~I
t
Y-
400
800
y+
9.9
1200
T (mK)
Figure 11. Comparison between the measured temperature dependence of the longitudinal resistance of a normal metal wire side-connected to a superconductor[32]' and
the theoretical prediction (smooth curve). The calculation is based on an approximate
solution of the Usadel equations (P. Charlat, private communication).
S
N
SI
Figure 12. Schematics of the mesoscopic NS structure measured by Petrashov et al.
[34] .The proximity effect induced in the normal metal wire is modulated by the phase
difference between the superconductors Sand S'.
6.3. RENORMALIZATION OF TUNNEL RESISTANCES
In contrast with diffusive resistors, the conductance of tunnel junctions is

strongly affected by the proximity effect. This phenomenon, used in the
density of states measurement reported above, already shows up in the
current-voltage characteristic of NS junctions. In the absence of induced
401
0,0
-0,2
-0,4
3R
13Rmaxl -0,6
-0,8
-I,D
-21t
-1t
1t
21t
31t
41t
Figure 13. Comparison between the measured (Open squares taken from ref. [34]) and
the predicted variations (continuous line) of the resistance of the circuit sketched in Fig.
12, as a function of the phase difference cf> between the superconducting electrodes (taken
from ref. [18]). The maximum magnitude corresponds to a change of the resistance by
9.7%. The theoretical prediction has been calculated at the temperature for which the
effect is maximum.
proximity effect through the tunnel barrier, the sub gap conductance of a
NS junction should be zero. Usually, the subgap conductance is found to
be much smaller than the normal state conductance but is definitively nonzero. The residual subgap current provides a direct measurement of the
weak proximity effect induced on the normal side of the junction. We have
carried out an experiment in which this induced proximity effect can further be modulated [36]. For that purpose, we have measured "NS-QUIDs"
(normal superconducting quantum interference devices) consisting of a normal wire forming two neighboring junctions with two superconducting electrodes whose phase difference 8 can be varied (see Fig. 14). The phase difference 8 is controlled by applying a magnetic field perpendicular to the
plane of the fork: 8 = 21f/o, where is the magnetic flux threading the
loop formed by the fork and the wire.
The behavior of the NS-QUID is easily understood from the zero temperature circuit theory developed in section 4. Neglecting the resistance of
the short segment of normal wire between both junctions with respect to
the resistance ref f between the junctions and the effective normal reservoir, the device conductance 9 is readily calculated using the circuit theory.
402
s
N
Figure 14. NS-QUID layout: a normal metal wire overlaps an oxidized superconducting
electrode to form a split tunnel junction. The weak proximity effect induced in the normal
wire through the tunnel junctions is modulated by the flux J enclosed by the loop.
In the case when the tunnel conductance of each junction GT/2 is much
smaller than the wire conductance r;/f' the expression of 9 reduces to :
9
= reff
G(1 + 2COSO) .
2
(35)
This simple model thus predicts a full modulation of the zero voltage conductance at zero temperature, with maxima at fluxes corresponding to an
integer number of flux quanta in the loop. This modulation of course extends to finite temperatures and finite voltages. The variations of the zerovoltage conductance with flux are shown in Fig 15 for a sample with normal
state resistance GT = 640 I-lS at a temperature T = 27 mK [36]. The shape,
phase and contrast of the modulation pattern are in good agreement with
the predictions of the effective circuit model provided that small offsets
of the conductance and phase are incorporated. The offset in conductance
is due to the finite wire resistance between the junctions not taken into
account in the model, and the offset in the phase is, due to the residual
magnetic field. The value ref f is in good agreement with the value estimated from the geometry of the sample.
The modulation of the current appears when the temperature is decreased below 300 mK and extends up to the gap voltage, with the same
phase at all voltages. We define the modulated current Imod as the difference between the maximum and minimum currents. The measured variations of Imod with the voltage are shown in Fig 16 at various temperatures
together with the full theoretical prediction using measured parameters, an
adjusted value of Tsf and a scaling factor. The calculation is based on the
direct perturbative calculation of the Andreev current through the tunnel
junction[22]. In the case of low conductance tunnel junctions, this approach
is equivalent to the approach based on proximity effect, as previously shown
403
>
......
T =30mK, V=4.1IlV
z
a: 4
("')
,-
2
0
-0.4
-0.2
0.0
8 (mT)
0.2
0.4
Figure 15. Current modulation in a NSQUID at low voltage and at low temperature.
The data have been fitted by a cosine function with appropriate offsets (taken from ref.
[36]).
in a simpler situation. This latter formalism is more complex to handle but

has the advantage of calculating directly the total current, without splitting
it in a pair Andreev current and in a quasiparticle current. Its validity thus
extends above the gap voltage where the quasiparticle current exceeds the
Andreev current and where the perturbative approach diverges.
The theoretical predictions concerning the renormalization of the conductance by the proximity effect have thus been tested in two simple situations:
a) Both junction electrodes are normal but the proximity effect has been
induced in one of them by contacting it to a superconductor close to the
junction (density of states measurement).
b) One electrode is superconducting and induces a weak proximity effect
in the other normal one through the junction itself (NS-QUID experiment).
7. Conclusion
A large variety of experiments on the proximity effect in metallic NS structures are now well accounted for by a unified formalism derived from
the theory of non-equilibrium superconductivity. A diffusive normal metal
in contact with a superconductor behaves as an "exotic" superconductor
404
30~--~~----~----~-----'
:?"
20
c..
100
Figure 16.
Comparison between the measured (open symbols) and predicted (solid
lines) bias voltage dependence of the peak to peak modulation ~9d of the current in
the NS-QUID with the magnetic field at different temperatures ~ taken from Ref. [36])
. From top to bottom: T = 27,82,137, 177,and 233 mK. Inset: temperature dependence
of the maximum and minimum conductances. The full and dotted lines are theoretical
predictions taking into account the quasiparticle conductance (clashed line).
whose properties differ profoundly from those of usual BCS superconductors:

- Its density of states is depleted at the Fermi level. However, the spectrum is gapless when the normal metal is also in contact with a normal
reservoir, or when it extends from the superconductor to "infinity", i.e.
distances larger than the coherence length. The spectrum shows a true
gap only when all electron trajectories in the normal metal touch the
superconducting electrode.
- Like an ordinary superconductor, it can sustain supercurrents ifit contains a loop enclosing an applied magnetic flux. However, it develops
no potential difference when submitted to a current source only if the
current is applied through superconducting leads. If one or both leads
are normal, a voltage will appear.
- When connected to a single superconductor, the resistance of a wire
is the same as in the normal state except for a small reduction at
intermediate temperatures. The electric field is nevertheless predicted
to be expelled close to the contact with the superconductor.
405
The resistance of a tunnel junction embedded in a diffusive proximity

effect superconductor is strongly modified, in contrast with the resistance of a wire.
In conclusion, the proximity effect develops in a diffusive normal metal
some properties found in superconductors but does not turn it simply into
an ordinary BCS superconductor with a small gap. The manifestation of
these properties depends crucially on the geometry of the connections between the normal metal and the various superconducting or normal reservoirs.
7.1. ACKNOWLEDGMENTS
We thank Cristian Urbina, Laurie Calvet, and Frank K. Wilhelm for discussions and helpful suggestions on the manuscript.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
On leave from Michigan State University, East Lansing, MI 48824, USA.

M. Tinkham, Introduction to Superconductivity (Mc Graw Hill, 1985), chapter 4.
G. Deutscher and P. G. de Gennes, in Superconductivity, edited by R. D. Parks
(Marcel Dekker, New York, 1969), p. 1005.
C. J. Lambert, J. Phys.:Condensed matter 3, 6579 (1991); C.J. Lambert, V.C. Hui
and S.J. Robinson, J. Phys.:Condensed matter 5, 4187 (1993).
A. I. Larkin and Yu. N. Ovchinikov, in Nonequilibrium Superconductivity, edited by
D. N. Langenberg and A. I. Larkin (North Holland, Amsterdam, 1986), p. 493.
A. Schmid, in Nonequilibrium Superconductivity, Phonons, and Kapitza Boundaries,
edited by K. E. Gray (Plenum Press, New York, 1981), p. 423.
J. Rammer and H. Smith, Rev. Mod. Phys. 58, 323 (1986).
K. D. Usadel, Phys. Rev. Lett. 25, 507 (1970).
A. F. Volkov, A. V. Zaitsev, and T. M. Klapwijk, Physic a (Amsterdam) 210C, 21
(1993); A. F. Volkov, Phys. Rev. Lett. 74, 4730 (1995).
G. Eilenberger, Z. Phys. 214, 195 (1968).
Yu. V. Nazarov, Phys. Rev. Lett. 73, 1420 (1994).
D. Esteve, H. Pothier, S. Gueron, N. O. Birge, and M. H. Devoret in Correlated
fermions and transport in mesoscopic systems, edited by T. Martin, G. Montambaux
et J. Tnin Thanh Van (Editions Frontieres, Paris, 1996).
T. H. Stoof and Yu. Nazarov Phys. Rev. B54, R772 (1996).
W. Belzig, C. Bruder, and G. Schon, Phys. Rev. B54, 9443 (1996).
A. A. Golubov, in Superconducting Superlattices and Multilayers, edited by I. Bozovic, SPIE proceedings Vol. 215 (SPIE, Bellingham, WA, 1994), p353.
S. N. Artemenko, A. F. Volkov, and A. V. Zaitsev, Solid State Comm. 30, 771
(1979).
F. Zhou, B. Spivak, and A. Zyuzin, Phys. Rev. B52, 4467 (1995).
Yu. V. Nazarov and T. H. Stoof, Phys. Rev. Lett. 76, 823 (1996); T. H. Stoof and
Yu. Nazarov Phys. Rev. B53, 14496 (1996).
A. Volkov and H. Takayanagi, Phys. Rev. Lett. 76, 4026 (1996).
F. Wilhelm, A. Zaikin and G. Schon, Czech. J. Phys. 46, Supplement S4, 2395
(1996) and J. Low. Temp. Phys. 106, 305 (1997).
S. Yip, Phys. Rev. B52, 15504 (1995).
406
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
W. J. Hekking and Yu. V. Nazarov, Phys. Rev. Lett. 71, 1625 (1993) and Phys.
Rev. B49 6847 (1994).
A. KastaIsky, A. W. Kleinsasser, L. H. greene, R. Bhat, F. P. Milliken, and J. P.
Harbison, Phys. Rev. Lett. 67, 3026 (1991).
A. F. Volkov, Physica B203, 267 (1994) and refs. therein to earlier work.
C. W. J. Beenakker, Phys. Rev. B46, 12841 (1992) and in Mesoscopic Quantum
Physics, edited by A. Akkermans et aI., (Elsevier, Amsterdam, 1995), course 5.
S. Gueron, H. Pothier, N. O. Birge, D. Esteve, and M. H. Devoret, Phys. Rev. Lett.
77, 3025 (1996).
J. M. Rowell, in Tunneling Phenomena in Solids, edited by E. Burstein and S.
Lundqvist (Plenum, New York, 1969), p. 385; T. Claeson, ibid., p. 443.
M. H. Devoret, D. Esteve, H. Grabert, G.-L. Ingold, H. Pothier, and C. Urbina,
Phys. Rev. Lett. 64, 1824 (1990); G.-L. Ingold and Yu. V. Nazarov, in Single Charge
Tunneling, edited by H. Grabert and M.H. Devoret (Plenum Press, New York, 1992),
p.21.
G. J. Dolan and J. H. Dunsmuir, Physica B152, 7 (1988).
D. Vion, P.F. Orfila, P. Joyez, D. Esteve, and M. H. Devoret, J. Appl. Phys. 77,
2519 (1995).
H. Courtois, Ph. Gandit, and B. Pannetier, Phys. Rev. B52, 1162 (1995); Phys.
Rev. Lett. 76, 130 (1996).
P. Charlat, H. Courtois, Ph. Gandit, D. Mailly, A. F. Volkov and B. Pannetier,
Czech. J. Phys. 46, Supplement S6, 3107 (1996).
V. T. Petrashov, Czech. J. Phys. 46, 3303 (1996).
V. T. Petrashov, V. N. Antonov, P. Delsing and T. Claeson, Phys. Rev. Lett. 74,
5268 (1995).
The phase difference is obtained by applying a magnetic field through the loop
formed by connecting the reservoirs S and S' by a superconducting lead. In this
situation, the phase difference is = 211" (ezt - LIs()) /<po, where L is the loop
inductance and Is() is the screening supercurrent (not measured) flowing in the
loop. See B. J. van Wees, S. G. den Hartog, and A. F. Morpurgo, Phys. Rev. Lett.
76, 1402 (1996), and V. T. Petrashov, R. Sh. Shaikhadarov, and 1. A. Susin, JETP
Lett., 64, 840 (1996) for a detailed discussion.
H. Pothier, S. Gueron, D. Esteve, and M. H. Devoret, Phys. Rev. Lett. 73, 2488
(1994) and Physica B203, 226 (1994).
MESOSCOPIC EFFECTS IN SUPERCONDUCTIVITY
ROSARIO FAZIO
Istituto di Fisica, Universita di Catania

viale A. Doria 6, 95129 Catania, Italy
AND
GERD SCHON
Institut fur Theoretische Festkorperphysik

Universtitiit Karlsruhe, 76128 Karlsruhe, Germany
1. Introduction
Several chapters in this book elaborate on the concepts of mesoscopic

physics. This includes phase-coherent quantum transport combined with
concepts from the macroscopic world such as reservoirs and dissipation, as
well as single-electron effects. Mesoscopic physics is displayed in the electronic transport properties of small systems with spatial dimensions in the
range of a few nanometers to micrometers, at low temperatures typically
below 1 K. The progress in nano-fabrication allowed the controlled fabrication of these structures and led to an increased interest in this physics.
Characteristic for superconductivity is the macroscopic phase coherence
of the order parameter and the supercurrent flow, as well as the modifications of quasiparticle properties by the energy gap. Superconductivity adds
new degrees of freedom and makes the description of. mesoscopic electron
transport richer. On the other hand, typical superconducting properties are
influenced by mesoscopic effects, e.g. by charging effects, and the question
arises whether superconductivity persists in ultrasmall systems (see e.g. the
chapter by Ralph et al. in this volume).
In this chapter we will investigate mesoscopic superconducting systems
and heterostructures of normal metals and superconductors. We will first
discuss in Section 2 in a few illustrative examples the single-electron and
charging effects in superconducting tunnel junction systems. We show how,
on the one hand, the superconducting gap influences single-electron tunneling and how, on the other hand, charging effects influence Andreev reflec407
@ 1997 Kluwer Academic Publishers.
408
tion processes. Cooper pairs can tunnel coherently; associated with their
quantum dynamics is the 'macroscopic quantum tunneling' of the phase in
low capacitance junctions (see e.g. the chapter of Devoret and Grabert in
this volume). In junctions with even lower capacitance quantum mechanical superposition of charge states playa role. The combination of coherent
Cooper pair tunneling, Andreev reflection, and quasiparticle tunneling leads
to richly structured dissipative I-V characteristics.
We then turn in Section 3 to the properties of superconductor-normal
metal heterostructures. The key words here are proximity effect and, again,
Andreev reflection. It has been known for a long time that the proximity
effect and the conversion between normal and supercurrents modifies the
system properties over a finite length near the interfaces. Recent experiments could spatially resolve these properties either by probes placed close
enough to the interface, or in samples with small spatial dimensions L,
such that the Thouless energy D / L2 becomes comparable to the temperatures in the experiment. We present some examples and several theoretical
approaches to these physical questions.
2. Charging effects in low-capacitance superconducting junction
systems
Modern technology has made it possible to fabricate, in a controlled way,
metallic tunnel junctions with capacitances in the range of C = 1O-15F and
below. In this case the charging energy associated with a single-electron
charge, Ec == e2 /2C, is of the order of 1O-4eV or larger, which corresponds
to a temperature scale Ec/kB 2: 1K. This implies that electron transport
in the sub-Kelvin regime is strongly affected by charging effects (see the
introductory chapter and Refs. [1,2]).
If part of the system is superconducting further interesting effects are
found: at subgap voltages single-electron tunneling (SET) is suppressed.
This makes higher-order processes such as Andreev reflection in normalsuperconductor (NS) junctions a dominant transport process. Here we discuss how this process is affected by the charging effects.
The charging energy allows the control of the electron number of small
islands. Adding one electron to a small superconducting island necessarily
puts it into an excited state with an energy exceeding the gap. Only when
a second electron is added, can both recombine to form a Cooper pair.
If this happens in a coherent way, the energy of the excitation created
in the first tunneling process can be regained in the second. This leads to
"parity effects", which distinguish between states with "even or odd electron
number in the superconducting island. As an example we analyze the IV characteristics of a NSN SET transistor with a superconducting island
409
VL
1 - - 1_
Figure 1.
CR
CG
l-f---VG--I
The SET transistor.
between normal conducting leads.

Also the coherent tunneling of Cooper pairs is influenced by charging
effects. The charge and the phase difference in a Josephson junction, although macroscopic degrees of freedom, are quantum mechanical conjugate
variables. The eigenstates in general are superpositions of different charge
states. We discuss the consequences for the dissipative I-V characteristics
of superconducting SET transistors. In an NSS transistor the Andreev reflection process in the NS junction can be used to probe the eigenstates
emerging from coherent Cooper pair tunneling in the SS junction. In SSS
transistors we analyze the combination of coherent Cooper pair tunneling
and quasiparticle tunneling, which leads to richly structured I-V characteristic.
Reviews of single-electron effects in normal metal systems can be found
in the book Single Charge Tunneling [2]. Tinkham's Introduction to Superconductivity [3] includes some topics of the present chapter. Recent work is
presented in the proceedings of the workshop Mesoscopic Superconductivity
[4] and reviews by Bruder [5] and Schon [6].
2.1. THE CHARGING ENERGY
The charging energy of systems of tunnel junctions depends on the electron number in various parts of the system and the applied voltages. An
important example is the single-electron transistor shown in Fig. 1. An
island is coupled via two tunnel junctions to a transport voltage source,
V = VL - VR, such that a current can flow. The island is, furthermore, coupled capacitively to a gate voltage VG. The charging energy of the system
depends on the integer number of excess electrons n = l, 2, ... on the
island and the continuously varied voltages. Elementary electrostatics [2]
410
yields the "charging energy"
E ( Q) _ (ne - QG) 2
2C
ch n, G -
(1)
Here C = CL +CR +CG is the total capacitance of the island. The effect of
the voltage sources is contained in the "gate charge" QG = CG VG + CL VL +
CR YR. Similar expressions hold for the "single-electron box", an even simpler system which consists of one junction only and a gate capacitance.
In a tunneling process, which changes the number of excess electrons in
the island from n to n + 1, the charging energy changes. Tunneling in the
left junction is possible at low temperatures only if the energy in the left
lead, eVL, is high enough to compensate for the increase in charging energy
eVL > ECh(n + 1, QG) - Ech(n, QG). Similarly, tunne.Jing from the island
(transition from n + 1 to n) to the right lead is possible at low T only if
Ech(n + 1, QG) - Ech(n, QG) > eVR. Both conditions have to be satisfied
simultaneously for a current to flow through the transistor. If this is not the
case the current is exponentially suppressed, which is denoted as "Coulomb
blockade". Varying the gate voltage produces the Coulomb oscillations, i.e.
an e-periodic dependence of the conductance on QG.
Many properties of the SET transistor and its extensions can be understood by considering only the energy of the different charge configurations.
A further understanding of the I-V characteristic requires the knowledge
of the tunneling rates of the electrons, which will be next topic.
2.2. SINGLE-ELECTRON TUNNELING RATES
The SET transistor, shown in Fig. 1, is described by the Hamiltonian
(2)
Here, HL = Lk,O' fkckackO' describes noninteracting electrons with wave
vector k in the left lead, with similar expressions for the island (with
states denoted by q) and the right lead. The Coulomb interaction, Hch =
(ne - QG)2/(2C), is assumed to depend only on the total (net) charge
n = Lk,O' CkaCkO' - n+ on the island (electronic and ionic background), as
discussed above. Charge transfer processes are described by a tunneling
Hamiltonian, for instance tunneling in the left junction by
Ht,L =
k,q,O'
TkqCkaCqO'
+ h.c ..
(3)
We determine the transition rates by Golden-rule arguments. An electron tunnels from one of the states k in the left lead into one of the available
411
states q in the island, thereby changing the electron number from n to n+ 1,

with rate
~t'L
I'LI(n) = e2
i: i:
dE
dE' NL(E)NI(E')
xfL(E)[l- JI(E')] o(E' - E + OEch). (4)
The crucial point is that the conservation of energy, expressed by the 0function, includes apart from the energies of the electron states Ek/q also the
charging energy. The latter depends on the change of the electron number
and the applied voltages. In the process considered it changes by OEch =
Ech(n+ 1, QG) - Ech(n, QG) - eVL. We further introduced the normal state
tunnel conductance of the junction
= (4rre 2/n)NI(O)Q 1NL(O)fkITI2.
At this stage, the tunnel matrix elem~nts Tkq can be considered as constants; NI/dO) and Q1/ L denote the normal densities of states and volumes
of the island and lead. If the electrodes are superconducting we have to
account for the reduced densities of states. In ideal systems they take the
R;L
BCS form Ni/L(E) = 8(IEI- ~I/d IEI/ E 2 - ~i/L'

Usually the distribution functions fI/L can be chosen to be Fermi functions. If both electrodes are normal conducting the integrals over the electron states in Eq. (4) can be performed, resulting in. the "single-electron
tunneling" (SET) rate [1]
1
OEch
I'LI(n,Qg) = ~R
e t,L exp [oEch /k BT]
(5)
At low temperatures, kBT ~ \8Ech\, a tunneling process which would increase the charging energy is suppressed, I' -+ O. This phenomenon is called
"Coulomb blockade" of electron tunneling.
If one or both electrodes are superconducting the rate still can be expressed in a transparent way
I'LI(n,Qg)
= ~Iqp
(OECh)
1
-e- exp[oEch/kBT] -1 .
(6)
The function Iqp (V) is the well-known quasiparticle tunneling characteristic (see e.g. Ref. [3]), which is suppressed at voltages below the superconducting gap(s). Charging effects reduce the quasiparticle tunneling further.
At zero temperature the rate is nonzero only if the gain in charging energy
compensates the energy needed to create the excitations OEch+~I+~L :S O.
The rates describe the stochastic time evolution of the charge of the
junction system. For the theoretical analysis Monte Carlo schemes or in small systems - a master equation approach can he used. Examples of
412
the resulting J- V characteristics of normal metal junctions are presented

in the introductory chapter of this volume. Characteristic is the e-periodic
dependence of the current and conductance on the applied gate charge QG.
Examples of superconducting junction systems will be presented below.
2.3. TWO-ELECTRON TUNNELING, ANDREEV REFLECTION
In the regime where quasiparticle tunneling is suppressed by the superconducting gap higher-order processes involving multi-electron tunneling
playa role. Cooper-pair tunneling is such a process, and will be discussed
later. If only one of the electrodes is superconducting there still exists a 2electron tunneling process, denoted as Andreev reflection 1 In this process
an electron approaching from the normal side with energy below the gap is
reflected as a hole, while a Cooper pair propagates into the superconductor.
We will determine now the rate of Andreev tunneling taking into account charging effects, as discussed in Ref. [7]. For this purpose we consider
a SET transistor with a superconducting island and normal leads (NSN).
The tunneling Hamiltonian is rewritten in terms of the Bogoliubov creation
and annihilation operators for the excitations in the superconducting island
Ht,L =
Here, Uq,O' and
k,q,O'
Tkq[U q,0'1'L,. + v;,O'I-q,-O']Ck,O'
+ h.c ..
(7)
are the standard BeS coherence factors with magnitudes
J(1 q/ Eq)/2, and Eq = JE~ + ~2 is the energy of the quasiparticles.

Vq,O"
Andreev reflection is a second-order coherent process. In the first part of

the transition one electron is transferred from an initial state, e.g. k t of the
normal lead, into an intermediate excited state q t of the superconducting
island. In the second part of the coherent transition. an electron tunnels
from k' t into the partner state -q t of the first electron, such that both
form a Cooper pair. The final state contains two excitations in the normal
lead and an extra Cooper pair in the superconducting island. The amplitude
for this process, to which we add the amplitude of the process in reverse
order, is given by [7]
Here spin indices have been suppressed and the relation Vq,t = v;,.j.. has
been used. The change in the charging energy 8Ec h,1 == ECh(n + 1, QG) 1 Andreev considered a normal metal and a superconductor in good metallic contact,
but his phy;:;icru picture can be generalized to tunnel junctions.
413
Ech(n, QG) - eV corresponds to the virtual intermediate state where one

electron has tunneled from the lead (at voltage V) to the island. Finally,
the rate for the Andreev reflection process is
ItI
= 2: L
k,kl
IAkk'12 fL(Ek)fL(EIe)c5(Ek + Ele + c5Ech,2) .
(9)
Here, the change in the charging energy c5Ech,2 = Ech(n+2, QG)-Ech(n, QG)2eV corresponds to the real final state where two electron charges have been
added to the superconducting island.
If we approximate the product of tunneling matrix elements by its average the q-summation in (8) can be performed, with the result Akk' =
7rNI(O)a (ll./c5Ec h,l) (TkqTk'-q)q, where a(x) == ~ VX~-l arctan J~+~. Andreev reflection is most important if the gap ll. is larger than the relevant energy differences Ic5Ech ,ll. In this limit the function a reduces to
a(x ~ 1) ~ 1. Henceforth, we disregard this weak energy dependence: As
such the integrations in (9) can be performed, resulting in
CA
c5Ech,2
ILI(n, QG) - 4e2 exp(c5Ech,2/ k BT) - 1 .
_
(10)
Note that the functional dependence of this rate coincides with that for
single-electron tunneling in a normal junction, Eq. (5). Hence Andreev reflection is subject to Coulomb blockade like normal-state single-electron
tunneling [8] with the exception that:
(i) The charge transferred in an Andreev reflection is 2e, and the charging
energy changes accordingly.
(ii) The effective Andreev conductance is of second-order in the tunneling
conductance
(11)
(iii) We have to account for the number of independent parallel channels

for both the normal state conductance, 1/ Rt = Nch/ Rt,o, and the Andreev
conductance, C A ex NchRK/ R~,o. (Note that in Eq. (11) we express the
latter by 1/ R t . Hence the factor Nch appears in the denominator.) In the
tunneling Hamiltonian approach Nch is expressed by the correlations between the matrix elements [7]
1 _ (l(n qTk'_ q)qI2)k,k'
Nch
(ITkqI2)kq)2
(12)
A more detailed analysis [9] show~ that the second-order Andreev process is
sensitive to spatial correlations in the normal metal, which can be expressed
414
JeR/Ec
0.8
0.6
0.4
0.2
Figure 2.
J- V characteristic of an NNS transistor. Both junctions have the same
normal-state conductance. The ratio of Andreev and normal-state conductance is
C A R t = 0.02, and ~ = 4Ec. From Ref. [10].
by the Cooperon propagator. For the moment we consider Nch as a fit

parameter; a comparison of the Andreev and the normal state conductance
shows that even in small junctions it is much larger than one.
As an example we show in in Fig. 2 the I-V characteristic of a NNS
SET transistor. The structure observed there with two characteristic scales
arises due to a combination of single-electron tunneling with rate (6) and
Andreev reflection with rate (10).
2.4. PARITY EFFECTS
2.4.1. The Superconducting Electron Box

In a normal-metal electron box, sweeping the applied gate voltage increases
the electron number on the island in unit steps, and the voltage of the island
shows a periodic saw-tooth behavior. The periodicity in the gate charge Qc
is e. If the island is superconducting and the gap ~ smaller than the charging energy Ee, the charge and the voltage show at low temperatures a
characteristic long-short cyclic, 2e-periodic dependence on Qc. This effect
arises because single-electron tunneling from the ground state, where all
electrons near the Fermi surface of the superconducting island are paired,
leads to a state where one extra electron - the "odd" one - is in an excited state [11]. In a small island, as long as charging effects prevent further
tunneling, the odd electron does not find another excitation for recombination. Hence the energy of this state stays (at least metastable) above that
of the equivalent normal system by the gap energy. Only at larger gate
415
()
UJ
.........
-.
(!J
c
.........
..r::
0
UJ
3
2
1\
-1
-1
Figure 3. The charging energy of a superconducting single-electron box as a function of

the gate voltage shows a difference between even and odd numbers n of electron charges
on the island. Accordingly the average island charge (n) is found in a broader range of
gate voltages in the even state than in the odd state.
voltages can another electron enter the island, and the system can relax
to the ground state. This scenario repeats with periodicity 2e in QG, as
displayed in Fig. 3.
At low temperatures the even-odd asymmetry has been observed in the
electron box [12] as well as in the I-V characteristics of superconducting
SET transistors [13, 14, 15]. However, at higher temperatures, above a
cross-over value Tcr ~ ~, the e-periodic behavior typical for normal-metal
systems is recovered. We can explain this cross-over as well as the structure
in the I-V characteristics by analyzing the rate of tunneling of electrons
between the lead and the island, paying particular attention to the fate of
the "odd" electron [16, 17].
We first consider an electron box with a superconducting island and
a normal lead. If the distribution functions of lead and island are Fermi
functions, the rate of tunneling is given by Eq. (6). At low temperature the
rate ILl is finite only at voltages where the gain in charging energy (i.e.
8Ech < 0) exceeds the energy of the excitations (Ek ?: 0, Ep ?: ~) created
in the lead and island, i.e. for 8Ech + ~ < O. It is exponentially suppressed
otherwise. The assumption of equilibrium Fermi distributions is sufficient
when we start from the even state. For definiteness let us assume that we
started from n = 0 at gate voltage 0 ~ QG ~ e. As such, the relevant
change in charging energy is 8Ech = E ch(1, QG) - Ech(O, QG) and the rate
416
of tunneling from an even to an odd state is given byeq. (6)

'Yeo
= 'YLI(n = O,QG) .
(13)
In the odd state the quasiparticle distribution differs from an equilibrium Fermi function. There is extra charge in the normal component.
After thermalization the excitations in the island can be described by a
Fermi function, fs!-,(E) = [e(-O!-')/kBT + It l , but with a shifted chemical
potential JlN = Jls + OJl relative to the condensate. The shift in chemical potential is fixed by the constraint of one excess electron charge 1 =
NI(O)OI J~oo dEAti(E) [fs!-'(E) - fo(E)]. This reduces at low temperatures
to
(14)
where
(15)
is the number of states in the island available for quasi particles near the
gap [13]. Parity effects are observable as long as the shift of the chemical
potential is relevant OJl > kBT. This is the case for temperatures below the
cross-over temperature
(16)
The tunneling rate back from the odd state (here n = 1) to the even
sta.te (n = 0), 'Y 0e = 'YIL.5J.'(n = 1, QG), is given by (4) with the island
distribution function replaced by fs!-,(E). For exp(-~/kBT) : 1 the ratio
of the rates of the two transitions is
In other words, they obey a detailed balance relation, depending on a "free

energy" difference, which, in addition to the charging energy, contains the
shift of the chemical potential OJl. This free energy difference coincides with
that introduced in Ref. [13].
For the following discussion it is useful to decompose the rate as
(18)
where 'YIL is given by the equilibrium form, equivalent to (6), and
O'Y(QG) =
e2~t
i: i:
dEk
dENI(E)
x [Jo!-'(E) - fo(E)] [1 - fO(Ek)]O(Ek - E - oEch)(19)
417
describes the rate of tunneling of the odd, excited electron [16]. In the
important range of parameters ~+bEch > kBT this "odd-electron tunneling
rate" reduces to
1
(20)
b"((QG) = 2e2 RtNl(O)Q l '
whereas it is' exponentially suppressed otherwise. It contains a small prefactor l/Nl(O)Ql as compared to "(lL. On the other hand, in the considered
range of gate voltages - since the energy of the excitation in the island
is regained in the tunneling process - b"( is not exponentially suppressed.
Hence it may be larger than "(lL.
In the range 0 ~ QG ~ e tunneling connects the island states n = 0
and n = 1. The range e ~ QG ~ 2e can be treated analogously. The
tunneling now connects the states n = 1 and n = 2. In this case, except
for the single-electron tunneling processes which create further excitations
with rate (6), one electron can tunnel into one specific state (-k, -0"),
the partner state of the excitation (k,O") which is already present. Both
condense immediately; the state with two excitations only exists virtually.
The latter process occurs again with rate b"((QG). The symmetry implies
,,(e%e(QG) = "(e%e(2e - QG). Since the properties of the system are 2eperiodic in QG, we have provided a complete description for all gate voltage.
The sequential tunneling of charges between the island and the lead is
described by a master equation for the occupation probabilities of the even
and odd states Pe(QG) and Po(QG),
dPe~~G)
= _,,(eo(QG)Pe(QG)
+ ,,(oe(QG)Po(QG)
(21)
with Pe(QG)+Po(QG) = 1. With "(~(QG) = ,,(oe(QG)+,,(eo(QG) the equilibrium solution follows to be Pe(o)(QG) = "(oe(eo)(QG)h~(QG). For "(oe ~ "(eo
we have Pe(QG) ~ 1, i.e. the system occupies the even state, while for
"(eo ~ "(oe the island is in the odd state.
The cross-over value Qcr of the gate charge, where. the system switches
between the even and the odd state, is determined by the condition Pe ~ Po,
i.e. ,,(oe(Qcr) ~ ,,(eo(Qcr). At low temperatures this condition coincides with
the condition that the energy is minimal, see Fig. 3. At finite, but low
temperature we find
Qcr(T)
= -2 + -e
[~- kBTln Neff (T)] ,
(22)
where Neff(T) was introduced in (15). This means the short plateaus in
Fig. 3 get longer until, above TCfl we have Qcr = e/2, and the e-periodic
behavior known from normal systems is recovered.
418
2.4.2. I -V Characteristics of NSN Transistors

The analysis presented above can be extended such that we can derive
the I-V characteristics of SET transistors with a superconducting island.
We first consider an NSN transistor with an energy gap smaller than the
charging energy scale ~ < Ee. In this system the important processes are
single-electron tunneling processes in the left and right junction, causing
transitions between even and odd states, with rates I~%e and I~%e which
are obvious generalizations of Eq. (13) and (18). At low T it is sufficient to
consider only one even and one odd state of the island. The solution of the
corresponding master equation yields the single-electron tunneling current
I - e(
eo p.
IL e
oe P
)_ e
IL -
'Ro/Le
+ ILoe + IRoe
'Lo/Re eo
eo
IL
+ IR
(23)
At high temperatures, T > Ten this current (23) shows the Coulomb oscillations known from normal systems with parabola-shaped maxima at the
points Qc = e/2+ne with integer n. At low temperatures, T < Ten the current is limited by the odd-electron tunneling rate I in one of the junctions.
In the window Qer(T) < Qc < ~ + ~C /e + Qer/2 < e it is
Iplateau
1
= eli"( = 2eR t N r(0)nr '
(24)
while it is exponentially small outside. A second current plateau exists in

the window e < 3e/2 - ~C/e - Qcr/2 < Qc < 2e - Qcr. Both plateaus
create a dou ble structure which repeats 2e-periodically. For ~ +eV /2 > Ee
the two plateaus merge to form a 2e-periodic single plateau structure. The
resulting I-V characteristic is visualized in Fig. 4.
In NSN transistors with a larger superconducting gap ~ > Ee the odd
states have a large energy. Hence a mechanism which transfers two electrons
between the normal metal and the superconductor becomes important. Andreev reflection with rate (10) provides such a mechanism [7]. The master
equation description can be generalized to include also this process. At low
temperatures a set of parabolic current peaks is found centered around the
degeneracy points Qc = e, 3e, ... [7]
IA(5Q c, V)
= C A (V -
4 8Qb
VC2 )
e (V _ 4 VC2
8Qb )
(25)
Here 5Qc is 5Qc = Qc - e for Qc close to e, and similar near the other
degeneracy points.
At larger transport voltages, single-electron tunneling sets in, even if
~ > Ee, and Andreev reflection gets "poisoned" [7]. This occurs for
V ~
Vpoison
eQc
= -;2 ( Ee - C
+~) .
(26)
419
Figure 4. The current I( QG, V) through a NSN transistor with
< Ee. From Ref.
[17].
II fA
2500
2000
1500
1000
500
300
o
0.0
0.5
100
1.0
Qa/ e
1.5
200 V /
JlV
2.0
Figure 5.
The current I( QG, V) through a NSN transistor with ~ > Ee. The parameters correspond to those of the experiments [15]' Ee = 100/JeV, ~ = 245/JeV,
Rtl/R = 43kO, I/G A ~ 1.2(2.4)108 0 for the left (right) junction. From Ref. [17].
The rate for this transition, from the even to the odd state, is of the order
of I'eo rv (V - Vpoison)/eR t . It puts the system into an excited state, making
it energetically favorable that a second electron tunnels into the partner
state of the excitation created in the first process. The rate for the second
process is given by 61', which in the considered range of parameters takes
the value given in Eq. (20). Typically the rate for the second transition,
from odd to even, is smaller than that of the first processes and, hence, creates the bottleneck in the sequence of SET processes. The same inequality
420
also implies that above Vpoison the system is most likely in the odd state,
Pol Pe = ,eo / fyy' ~ 1. Hence the current produced by the cycle is given by
Eq. (24).
Fig. 5 shows the current-voltage characteristic of a NSN transistor with
~ > Ec. At small transport voltages the 2e-periodic peaks due to Andreev
reflection dominate; they get poisoned above a threshold voltage. The peaks
at larger transport voltages arise from a combination of single-electron tunneling and Andreev reflection. The shape and size of the even-even Andreev
peaks and some of the single-electron tunneling features at higher transport
voltages agree well with those observed in the experiments of Hergenrother
et al. [15].
2.5. COOPER PAIR TUNNELING
2.5.1. Macroscopic Quantum Effects

In "classical" Josephson junctions, Cooper pairs can tunnel free of dissipation between the superconducting electrodes. The coupling is described
by the Josephson energy - EJ cos r.p, which depends on r.p, the phase difference across the barrier. The energy scale EJ = nlcr /2e is related to the
critical current of the junction, which in turn can be expressed by the tunneling resistance of the junction and the energy gap of the superconductor,
Icr(T = 0) = 7r~/(2eRt).
Charging effects introduce quantum dynamics: The phase difference and
the charge on the electrodes, Q, are quantum mechanical conjugate variables. An ideal Josephson junction is governed by the Hamiltonian
Q2
Ho = 2C - EJcosr.p, Q =
T8(nr.p/2e)
(27)
(For simplicity, we first describe a single junction; generalizations are presented below.) An important question, addressed in Refs. [18, 19, 20, 21],
is how to account for dissipation due to the flow of normal currents and/or
quasiparticle tunneling. The so-called "macroscopic quantum effects" like
macroscopic quantum tunneling of the phase, or quantum coherent oscillations are derived from the Hamiltonian (27). Macroscopic quantum tunneling has been observed in tunnel junctions with small capacitances of the
order of 10- 12 F. These values are still orders of magnitude to large for
single-electron effects to playa role.
2.5.2. Superposition of Charge States

We now turn to mesoscopic Josephson junctions or junction systems, where
the number of electrons or Cooper pairs in small islands is the relevant
degree of freedom. The charging energy has been discussed above. The
421
Josephson coupling describes the transfer of Cooper-pair charges forward

or backward, and can be written in a basis of charge states as
(28)
Below, we will consider situations where Cooper pairs tunnel coherently,
which shows features known from the phenomenon of resonant tunneling.
Coherent Cooper pair tunneling is non-dissipative and strongest near points
of degeneracy. First we will show that these quantum fluctuations broaden
the steps in the expectation value of the charge on the island of a superconducting electron box. Then we will discuss how coherent Cooper-pair
tunneling can be probed by Andreev reflection and observed in the dissipative I-V characteristic of a NSS transistor [22]. Finally we describe how the
combination of coherent Cooper-pair tunneling and dissipative quasiparticle tunneling leads to a dissipative I-V characteristic of SSS transistors
[23, 24, 13, 25, 26]. Further examples of coherent tunneling of Cooper pairs
can be found in the literature, e.g. the gate-voltage dependence of the critical current of SSS or SNS transistors [27, 28].
We first consider an electron box with superconducting island and lead
with large energy gap at low temperatures, ~ > Ec ~ kBT. In this case,
at low voltages, quasiparticle tunneling is suppressed, and the island charge
can change only by Cooper-pair tunneling in units of 2e as described by
Eq. (28). The tunneling is strong near points of degeneracy. For instance
for QG ~ e the charging energies of the states with n = 0 and n = 2 are
comparable, and we can restrict our attention to these two charge states.
The coherent tunneling between both is described by the 2 X 2 Hamiltonian
(29)
This Hamiltonian is easily diagonalized: the eigenstates and energies are
+ ,812) , 'l/Jl = ,810) - (12) ,

~ [1 + 8Ech ] = 1- ,82 ,
2
J8Ech2 + E2J
'l/Jo
odO)
~ [ECh(O) + ECh(2) =f J8E~h + El]
(30)
Here we have introduced the difference in charging energy 8Ech == Ech(2) Ech(O) = 4Ec (QGle - 1). The coefficient a approaches unity if the charging energy of the state 12) lies far above that of 10), i ..e. for 8Ech > 0, and
vanishes in the opposite limit, while ,8 has the complementary behavior.
422
The expectation value of the charge on the island in the ground state
is given by
(31)
It changes near QG = e from to 2 over a width of order 8QG

This has recently been observed experimentally [29].
EJ/ Ee.
2.5.3. NSS Transistors

Next we consider a NSS transistor. In this system the coherent tunneling
of Cooper pairs in the Josephson (SS) junction can be probed by the dissipative current due Andreev reflection across the NS junction [22]. We
restrict ourselves to low temperatures, kBT ~ EJ. In order to describe coherent Cooper-pair tunneling in a situation with nonzero transport voltage
we have to account in the Hamiltonian for the work done by the voltage
sources during the transitions. We, therefore, keep track also of the number of electrons NL and NR in the left and right electrode. A basis set of
states is denoted by INL, n, NR), and the corresponding charging energy
(for symmetric bias VL = - VR = V /2) is
In a situation where only two charge states get appreciably mixed the
eigenstates and energies of the corresponding 2 X 2 Hamiltonian are
1/J0
EO/1
alO, 0, 0) + ,610, 2, -2) , 1/J1 = ,610,0,0) - alO, 2, -2) ,

~ [ECh(O, 0, 0) + Ech(O, 2, -2) =f J8E;h + EJ] .
(33)
The coefficients coincide with those of the box discussed above, except for
the obvious change of notation, and 8Ech = Ech(O, 2, ...:..2) - Ech(O, 0, 0).
In the low-bias regime, the dominant mechanism of transport in the NS
junction of the transistor is Andreev reflection. Starting from a state 10,0,0)
we are led in such a process to the state 1- 2, 2, 0). The Josephson coupling
mixes this state with the state I - 2,0,2). Hence we have to consider a
second set of eigenstates
1/Jb
= al -
2,0,2) + ,61- 2,2,0) , 1/J~
= ,61 -
2,0,2) -
al -
2,2,0). (34)
The coefficients a and ,6 are the same as for the other pair, but the corresponding energies are shifted Eb/1 = EO/ 1 - 2eV.
Andreev reflection causes transitions between the two set of eigenstates
1/J0 ---7 1/Jb. The rate for this process can be derived along the lines described in an earlier. Compared to Eq. (8) a modification arises since the
charge transfer operators pick from the initial state the component with
423
zero charge on the island, which has amplitude a, and select from the final
state the component with two extra charges, which has amplitude (3. Hence
the amplitude for a Andreev reflection process between the states 'l/Jo and
'l/Jb with two electrons tunneling from the states k, t and k', of the normal
electrode is
The energy of the virtual intermediate state 1- 1k, 1q , 0), with one electron
added to the island and two excited quasi particles with energies Ek and
Eq = JE~ + ~2 in the normal and superconducting electrode, is given by
Ekq = Ech( -1,1,0) - Ek + Eq.
The summation in Eq. (35) can be performed, and the rate for the
Andreev reflection process is obtained by the Golden rlJ.le. After summation
over the initial states k and k' one finds for Eb - Eo = - 2e V ~ 0
A
= (a{3 )2 ao2 G
- 2 2e V
4e
'Y ('l/Jo -+ 'l/Jo')
(36)
The rate is proportional to the product
El
2 2
1
a {3 = 4" (8Ech) 2 +
El '
(37)
which displays a typical resonance structure. Here G A is the Andreev conductance (l1),and the function ao = a(~/[Ech(-l,l,O)-Eo]) has been
defined below Eq. (9). We further assumed that the energy 6.+Ec h( -1,1,0)
of the intermediate state lies above Eo. If ~ ~ Ec the function ao reduces
to ao ~ l.
Andreev reflection processes can also lead to transitions between the
other states introduced above, with rates
a 4 a6 G 2 [Eo + 2eV - E 1] 8[Eo + 2eV - Ed] ,

4e
A
GA
(3 a1 4e 2 El
2
GA
+ 2eV -
(a{3) a1 4e 2 2e V .
Eo] ,
(38)
The function al is defined similar as ao, but the energy of the initial state
Eo is replaced by E 1 .
Below the threshold voltage V < vth = (El - Eo) /2e the only transition
at low temperatures is Andreev reflection between the states 'l/Jo and 'l/Jb.
424
I
0.3
0.2
0. 1
VC/e
Figure 6. I-V characteristic of a NSS transistor. A resonant structure due to Cooper pair
tunneling is visible in the dissipative current due to Andreev reflection. From Ref. [22].
The resulting current, I = 2e-y( 'l/Jo -r 'l/Jb), shows a pronounced resonant

structure due to the overlap of the functions Il' and (3. At higher voltages
Andreev reflection can take the transistor to the excited state 'l/Ji. A master
equation yields the probabilities for the ground and excited states
Po
,('l/Jl -r 'l/Jb)
= , ('l/J0 -r 'l/J'I ) +, ('l/J I -r 'l/Jo')
,H
=1-
Po
..J.
T
0 for V
> lith . (39)
The current then is
;e
= [,( 'l/Jo -r 'l/Jb) +,( 'l/Jo -r 'l/JD] Po+ [,( 'l/Jl -r 'l/JD +'Y.( 'l/Jl -r 'l/Jb)] PI. (40)
A plot of the current-voltage-characteristic, as a function of the gate and

bias voltage is shown in Fig. 6 for the case where the superconducting gap
is larger than the charging energy ~ ~ Eo.
2.5.4. SSS Transistors

Next we consider the case of an SSS-SET transistor with superconducting
electrodes and island below the crossover temperature Tcr where parity
effects can be observed. The charging energy and coherent Cooper pair
tunneling in this system are described by the model Hamiltonian [24]
Ho
,,([(ne-2CQG )2 - 21 enV1In, n)(n, nl

=~
_ EJ
2
LLln2,n2)(n,nl)
(41)
425
Here we shortened the notation as compared to the previous subsection Eq.

(32) and introduced n == (NR - NL), the number of electrons which have
tunneled through the transistor. The eigenstates of Ho are linear combinations of different charge states
Iwa)
= L a~,n In, n)
n,n
(42)
and the energies are Ea.

Quasiparticle tunneling can cause transitions between different eigenstates IWa). It is accounted for by
H~P
T~~)ln + 1, n + 1)(n, nlc~ck + h.c.
kEL,qEI
(43)
T~~)ln - 1, n + 1)(n, nlck,cq + h.c ..
qEI,k'ER
If the junction resistances are large compared to the quantum resistance

Rt,L/R > RK = hi e2 the transition rates can be calculated by the Golden
rule. A quasiparticle tunneling process in the left junction gives rise to a
transition with rate
(L)
'Ya-+(3 =
Iqp
(W (ca(3)le
L- ( 1 _exp
(_ ca(3 Ik BT) + "I)
I(W(3ln 1, n 1)(n, nlWa)1 .
n,n,
(44)
Here I~~) is the I-V characteristic for quasiparticle tunneling in the left
junction [3], and ca(3 = Ea - E(3 is the energy difference between initial and
final state. We describe parity effects by including the escape rate 6"1 of an
odd quasiparticle in the island. It is given by an expression similar to Eq.
(20), modified by the density of state in the superconducting electrode. It
IS
'"
u'Y -
( )
2e RtNI 0
V(c
ca(3 + ~
O()
ca(3
a (3 + ~)2 _ ~2
(45)
if n is odd and vanishes in the even state.

In order to determine the dc-current we follow the procedure described
in Ref. [26] and first determine the eigenstates of H o, either in an expansion in the Josephson coupling or numerically taking into account a
sufficient number of charge states. This procedure converges for not too
large Josephson coupling energies, EJ < Ec. Given the eigenstates IWa)
we calculate the rates in Eq. (44), which then enter a master equation
8t Pa = 'L.(3=/:.a(P(3'Y(3-+a - Pa'Ya-+(3) for the probabilities Pa to find the system in the a-th eigenstate. The stationary solution 8t Pa = 0 is sufficient
426
IRC/e
0.0002
0.0001
VC/e
Figure 7. I-V characteristic of an SSS transistor. The parameters are

EJ = O.17Ec, Rlfr = R ~ RK, 'Y = 2.5 .1Q-5(RC)-1. From Ref. [26].
= 1.3Ec,
to evaluate the dc-current

I =
cx,(3tcx
PCXJcx-+(J (W(JlnIW(J) - (WcxlnIWcx))
(46)
The combination of coherent Cooper pair tunneling and single-electron tunneling leads to a dissipative I-V characteristic. Results are shown in Fig. 7
with parameters corresponding to those in Ref. [13]. We note that the I-V
characteristic is 2e-periodic and observe a rich structure deep in the subgap
region. For transport voltages eV 2:2.5Ec the 2e-periodic features disappear
and the current becomes e-periodic in QG again. This is not surprising since
on a current scale I ~ e8J the unpaired quasiparticle in the island looses
its importance.
For the parameters chosen at low transport voltages only few (two or
three) states Iwcx) are noticeably populated. Therefore, we can calculate
the eigenstates of H a, i.e. the coefficients a~,n in Eq. (42), by expanding in
EJ. Away from certain resonant situations, the a-th eigenstate has only one
coefficient a~ n of order unity, whereas all other coefficients are considerably
smaller. To fix ideas, let us consider the state IWa) in the range of gate
charges QG E [0, e/2]. In this eigenstate the most likely charge state is
In = 0, n = 0), i.e. ag a ~ 1. Due to coherent tunneling of one Cooper pair,
there is a non-zero a~plitude a~2,2 <X EJ/ Ec for the system to be in the
charge states In = 2, n = 2). Higher order Cooper pair tunneling leads to
427
a population of higher charge states with smaller amplitude. Off resonance

the system is in the charge state 12,6) with amplitude ag 6 ex (EJ/ EC)3.
At resonance these amplitudes are much larger. For i~stance along the
line
3eV = 4Ec(1 - Qc)
(47)
the charge states 10,0) and 12,6) have the same energy, and a three-Cooperpair tunneling process is in resonance. As a result the amplitude is drastically increased ag 6 ex (EJ / Ec).
A transition f~om Iwo) to another eigenstate can occur if it is energetically favorable and the matrix element Eq. (44) is nonzero. When analyzing
the energies we find that the process
(process a)
is possible. Off resonance the rate of process a) is of the order J(a) ex
(EJ/ EC)6. However, in a narrow strip of width is proportional to EJ around
the resonance line (47) we find
J(a)
ex
EJ/2
4Ec(1 _ Qc)
+ eV
ex
(EJ )
Ec
(48)
This process leads to the most significant resonance in the I-V characteristic. We are, thus, led to the conclusion that the dominant transport process
in the subgap region is tunneling of a quasiparticle accompanied by simultaneous tunneling of 3 Cooper pairs. This combination provides enough
energy to overcome the quasiparticle tunneling gap 2~. The importance of
this type of transport mechanism was first noted by Fulton et ai. [23].
So far we have studied the conditions for the system to leave the initial
state. However, a dc charge transport through the system requires cycles,
after which the island returns to a state equivalent to the initial one. The
simplest version is a two-step cycle of subsequent transitions of the same
type in the left and right junction. Such cycles dominate in NNN or NSN
transistors at low bias voltages. The cycle which leads to the pronounced
feature in Fig. 7, at 3eV = 4Ec(1 - Qc), arise due to two-step cycles
as well, but the second step is different from the first one. The transition
completing the cycle which starts with process a) is
(process b)
This means a quasiparticle transfer is accompanied by 2 Cooper-pair tunneling processes. The latter process is not in resonance and, therefore, the
rate is J(b) ex (EJ/Ec)4. Whereas off resonance the process a) is the bottleneck for the current, at resonance the process b) has the smaller rate. This
explains the value of the current at the resonance.
428
For further discussions of the structures manifest in Fig. 7, including

extensions such as the influence of fluctuations of the electromagnetic environment, as well as a comparison with experiments on SSS transistor [13, 28]
we refer to Ref. [26].
2.6. EXTENSIONS
In the examples discussed above, the charging energy is the dominant energy, while tunneling could be described in low order perturbation theory
or - in the case of coherent Cooper pair tunneling - by diagonalization of a
simple Hamiltonian. The expansion parameter is the dimensionless tunneling conductance RK/(47r 2 R t )' where RK = h/e 2 = 25.8kO is the quantum
of the resistance. In situations where this parameter is not small a more
general approach is required. H. Schoeller describes in his Chapter of this
volume a diagrammatic expansion to account for strong tunneling through
quantum dots [30, 31]. Strong tunneling in normal metal junctions has been
studied by several authors [32, 33, 34, 35].
A formulation in terms of path integrals displays in a transparent way
the interplay of charging effects and tunneling phenomena [31, 34]. Here
we would like to draw attention to the equivalent path-integral description
of superconducting junction systems, presented in Refs. [5, 6, 36]. In these
articles applications to selected problems have been Qiscussed, such as (i)
the influence of'charging effects on the Josephson current through a SNS
system, where earlier results of Bauernschmitt et at. [37] have been reproduced, (ii) the influence of charging effects on Andreev reflection, and the
proximity effect, which extends earlier results of Aslamazov et at. [38].
3. Hybrid Normal-Metal/Superconductor Structures
3.1. REVIEW
In the last few years new experiments revived the interest in equilibrium
and non-equilibrium properties of superconductor-normal metal (SN) hybrid structures. Two key words in this context are: proximity effect and
Andreev reflection. The hybrid structures can be grouped in two classes
depending on the transparency of the interface between superconductor
and normal metal. If they are separated by an insulating barrier with low
transparency the process of two-electron tunneling is the relevant transport mechanism at low bias. If they are in good metallic contact nearly all
particles are transmitted; here the dominant process is Andreev reflection.
Various excellent reviews [39, 40, 5] discuss many aspects of SN structures.
Our aim here is to introduce the basic concepts and theoretical techniques,
and to review some of the current literature. Some examples are discussed
429
explicitly to demonstrate the physics involved.

When a superconductor is put in contact with a normal metal, Cooper
pairs can leak across the interface. As a result there exists a non-vanishing
pair amplitude in the normal metal, defined by
(49)
where 'lj;(! (f) is the annihilation operator for an electron with spin (7. The
pair amplitude is a two-particle property, related to the probability of finding two time-reversed electrons at position r. The decay of F(f') away from
the interface depends strongly on properties - diffusive vs. ballistic, non interacting vs. interacting - of the normal metal [41]. At finite temperature
it decays in the normal metal exponentially on a scale ~T given by
c _ fiVF
27fT
<"T -
or
/nD
V2d'
(50)
depending on whether the metal is in the clean or diffusive limit. Here D

is the diffusion constant. (Henceforth, we use units where fi = kB = 1.)
At zero temperature, if interaction effects can be be disregarded, the decay
follows a power law, F(f') ex: l/r. The appearance of the pair amplitude on
the normal side of the interface is accompanied by a depression of the order
parameter on the superconducting side.
A nonvanishing pair amplitude implies the coherence of two electrons
in the normal metal induced by the coupling to the superconductor. It does
not necessarily lead to a gap in the spectrum, ~(f') = >"F(f') , since both are
related by the interaction strength >., which may vanish in normal metals
in the absence of an attractive or repulsive interaction.
The proximity effect is intimately related to the microscopic mechanism
which governs the transport through SN interfaces. At voltages and temperatures below the superconducting gap single particle tunneling is exponentially suppressed. The dominant process is then An4reev reflection [42],
where an incoming quasi-electron from N is reflected at the interface as
a quasi-hole, as a result of which a Cooper pair is injected into the superconductor. The reflected hole has a momentum which is opposite (to
order Ik - kFl/kF) to the one of the incident electron. The small difference
in the momentum implies that the particle and the hole maintain their
phase coherence up to distance of the order of L rv J DIE where E is the
energy of the particle relative to the Fermi energy. If E is the thermal energy, this length coincides with the correlation length given in Eq. (50) [43].
This demonstrates that the proximity effect and Andreev reflection, though
seemingly different concepts, are closely related. Also in the presence of a
tunnel barrier at the NS boundary the dominant mechanisms of transport
430
is the transfer of two electrons across the barrier. We call also this process
Andreev tunneling, although the momentum perpendicular to the interface
is not conserved. Andreev processes are also responsible for the Josephson
effect in S-N-S sandwiches [38, 44]. If the thickness of the normal region is
comparable to or less than its coherence length ~T,N' phase coherence can
be maintained and a supercurrent can flow through it, depending on the
phase difference of the two superconductors.
Although many properties of hybrid SN system have already been studied in the past, the interest in proximity devices has been renewed recently.
The reason is that it became possible to study mesoscopic hybrid systems
with dimensions smaller than ~T. In this case the particle and the hole preserve their phase coherence across the entire conductor. Another relevant
length scale, the phase-coherence length L> of single electrons in a normal
metal, might well be larger than ~T'
In mesoscopic proximity systems the interplay between phase-coherent
electron propagation in N and macroscopic phase coh~rence in S gives rise
to interesting new physics [4]. For instance, Andreev reflection in mesoscopic N-S tunnel junctions is strongly influenced by electronic interference. The transport through NS-QUIDS (Normal Metal-Superconductor
QUantum Interference DeviceS) was studied theoretically by Hekking and
Nazarov [9] and experimentally by the Saclay group [45], showing the existence of a modulated current as a function of the flux piercing the device.
Nakano and Takayanagi [46] considered a different type of interferometer
where the phase difference is created by a current which passes through
the superconductor. Petrashov et ai. [47] and Courtois et ai. [48] performed
a series of experiments on interference effects in transport through mesoscopic samples containing superconducting arms. Proximity systems with
clean N-S interfaces show a remarkable non-monotonic temperature dependence [49,43]. In these systems the presence of the superconductor renders
the diffusion constant of the normal metal effectively energy dependent.
Since electrons from the normal metal can enter the superconductor
and then return to the normal metal not only the off-diagonal properties of
the metal are modified, but also the single particle properties (diagonal in
the Nambu space). Very recently, the local electron density of states (DOS)
of a normal metal in contact with a superconductor has been studied at
mesoscopic distances from the N-S interface [50, 51]. Close to the Fermi
energy, a suppression of the DOS below its normal valu has been observed.
Due to the development of superconductor-semiconductor (S-Sc) integration technology, it is now possible to observe the transport of Cooper
pairs through S-Sc meso scopic interfaces as well [4]. Examples are the supercurrent through a two-dimensional electron gas (2DEG) with Nb contacts
(S-Sc-S junction) [52, 53] or through quantum point contacts [54, 55]. The
431
critical current was predicted to be quantized in units of e!:l/fi analogously

to the quantization of the normal state conductance in ordinary quantum
point contacts. Another example is the excess low-voltage conductance due
to Andreev scattering in Nb-InGaAs (S-Sc) junctions [56].
Electron-electron interactions in the normal metal modify the proximity
effect, both qualitatively and quantitatively [41]. If the interactions between
the electrons are repulsive, the induced pair amplitude in N decays faster
than in the noninteracting case. This is because interactions scatter the two
electrons out of their initial, time-reversed state. If the interaction is attractive, e.g. if N becomes superconducting at a lower transition temperature,
TeN < T < Tes, the pair amplitude decays slower because of the presence of
superconducting correlations, and ~T diverges at TeN. A perturbative treatment of the interactions [57] shows that an additional contribution to the
supercurrent arises. Its sign depends on the nature of the interactions in
the slab (attractive or repulsive), and its phase-dependence has period 7r (in
contrasts to 27r in the non-interacting case). In the tunneling regime, if the
dimensions of the normal metal and its electric capacitance are small, the
phenomenological capacity model described in section 2 can be used. In this
case the critical current of an S-N-S system depends strongly on charging
effects and can be tuned by a gate voltage applied to the island [37].
In low-dimensional semiconductor nano-structures with low electron
concentration the Coulomb interactions cannot be treated as a weak perturbation. Rather a non-perturbative, microscopic treatment of interactions is
required. For 1D systems, e.g. in a 2DEG gated to form a quantum wire,
this can be done in the framework of the Luttinger liquid (LL) model [58].
Hybrid superconductor - Luttinger liquid (S-LL) systems are interesting
since they enable one to study how the Coulomb interaction influences
the phase-coherent propagation of two electrons through a 1D normal region. The Josephson current through a S-LL-S device has been evaluated
in Refs. [59, 60]. Due to the interactions the Andreev current in a junction
between a superconductor and a chiral Luttinger liquid depends in a nonlinear way of the voltage [61]. Recently also the single particle properties
(DOS) of a LL connected to a superconductor have been studied, where the
combined effect of interaction and Andreev tunneling leads to a behavior
compared which differs qualitatively from that of an isolated LL [62].
Various theoretical approaches have been employed to study SN heterostructures. One school generalizes the scattering approaches of Landauer
to include Andreev tunneling. In this case the Bogoliubov-de Gennes equations are used to construct the scattering matrix (the interested reader is
invited to read the reviews on the topic [39, 40]). Another school uses quasiclassical methods starting from the Eilenberger equations (or the Usadel
equations for dirty metals) with the inclusion of the appropriate boundary
432
conditions for the Green's functions at the interface. The two complementary methods provide a framework to tackle various problems involving
hybrid structures. The quasiclassical methods have been useful to extract
analytic results in the diffusive limits. On the other hand the scattering approach is more appropriate in multi-terminal geometries or in the regimes
where neither the ballistic nor the diffusive limit are appropriate. In this
case numerical solutions have been worked out. The next sections are devoted to a summary of the two approaches. In the final part of this chapter
we discuss the influence of interactions on the proximity effect in superconductor - Luttinger liquid systems.
3.2. SCATTERING THEORY
Transport properties of mesoscopic systems have been described successfully within the scattering (Landauer) formalism [63]. The conductance
is related to the transmission, and the transport theory is reduced to
an analysis of the properties of the scattering matrix. This approach has
been generalized to systems containing SN interfaces by Lambert [64] and
Beenakker [65].
We consider an n-terminal geometry where the n-th reservoir is superconducting (the case of two or more superconducting reservoirs can be
described in the same fashion). Each lead contains N incoming and outgoing modes (for simplicity we assume here that N is the same for each
reservoir). In the scattering approach, one needs to evaluate the S-matrix,
defined as
(51)
Here (lI == (a,p) and j3 == (b,q), a,b = 1, ... n refer to the reservoirs while
q,p = 1, ... N refer to the channel indices. The superconducting reservoir
is characterized by the pair (n,l = 1, ... N) and it will be denoted by the
index s. In Eq. (51) 0 and I are the amplitudes of Qutgoing and incoming channels, respectively. The underline denotes the two components in
particle - hole space (in the absence of the superconductor the S-matrix is
block-diagonal in this space).
The aim of this section is to express the scattering matrix SOi(3 as a
function of the scattering in the mesoscopic region and the scattering which
takes place at the SN-interface. In order to pursue this scheme, it is conceptually simpler to separate the scattering in the normal region, which is
determined by the geometry and disorder in the the mesoscopic conductor,
from the scattering at the SN interface, where the Andreev processes occur. For this purpose it is assumed that a normal region, free of disorder,
lies between the scattering region and the SN boundary, as illustrated in
433
BaUistic region
a =(a,p=I,... N)
cr*=(n,r= I ... N)
Superconducting lead
s =(n,I=l...N)
Scattering region
Normal leads
~
Figure 8. Sketch of the scattering region including the small ballistic region in contact
with the superconducting reservoir.
Fig. 8. This ballistic region can be thought of as arbitrarily small, and its
properties do not appear in the physical results.
If the superconductor were not present the scattering matrix is determined exclusively by the geometry of the mesoscopic region
(52)
On the other hand, the scattering at the SN interface is described by the
2 x 2 matrix
(
~: ) =
(A)
S(7*S
(53)
(A)
Sss
A
where the index (7* == (n, r = 1, .. N) refers to the intermediate ballistic

region , separating the scattering region from the superconducting reservoir (see Fig.8). The components of the Andreev scattering matrix are
constructed by solving the Bogoliubov - de Gennes equations [66 , 67]. Note
also that the outgoing states in the previous equation are L* and 0 s, since
the incoming wave in the reservoir n (as defined in Eq. (51)) is outgoing
with respect to the SN interface.
Using Eq. (52) and Eq. (53) it is possible is to eliminate
L*
and
~*,
434
,
'(0)
'(A)
which allows us to express Sa{3 in terms of S a{3 and S O'.{3
(a,b=l=n)
(a
=1=
n)
(b=l=n)
(54)
The previous expressions for the S-matrix have a simple physical meaning:
By expanding the denominators one can identify each term of the series
as a sequence of scattering processes (reflections and transmissions) at the
various reservoirs.
The final step is to express physical quantities in the scattering formalism. Let us first consider the current operator in the (normal) lead a
(55)
Here a trace is performed in Nambu and spin space, and the matrix O'z
accounts for the different signs of the current in the electron and hole channels. The integration is over the transverse coordinate yO'. in lead a. Using
scattering states [68, 69] as a (more convenient) basis, with destruction
and creation operators a and at of incoming scattering states, the current
can be expressed as
Since the reservoirs are in thermal equilibrium, the occupation probabilities of the scattering states are given by Fermi distributions. Combining
Eq. (56) with the expressions Eq. (54) one arrives at the desired result
for the transport properties in terms of the geometric properties of the
scattering region and the Andreev scattering at the boundary with the
superconducting lead.
In a two terminal geometry, where a == (1, q = L.N) is the index for the
normal contact and s == (2, p = L.N) for the superconductor, the average
current is
A trace over the channels is implied. The current depends on the reflection
coefficients. Note that the normal reflection (ee) and Andreev reflection
435
(he) enter with opposite sign. If there is no potential barrier at the NS

interface, at energies below the gap there is no normal reflection but only
Andreev reflection. The linear conductance in this regime has been obtained
by Beenakker [65]
(58)
This result is the multi-terminal generalization of the formula obtained by
Blonder, Tinkham and Klapwijk [66] and by Shelankov [67]. The amplitudes
Tq are the eigenvalues of the geometrical transmission matrix (S~~\ i.e.
the same coefficients which enter in the multichannel Landauer formula
G = (2e 2 /h) Lq Tq, and the index q runs over the transverse channels in the
normal lead. Various applications of the previous expression and extensions
can be found in Ref. [39]. The general formula for the current beyond the
Andreev approximation and at finite voltage has been discussed in Refs. [70,
71], extensions to d-wave superconductors have been considered in Ref. [72].
The use of the scattering approach is not limited to the study of the average
current. Eq. (56) also allows the evaluation of the current noise (see the
chapter by de Jong and Beenakker in this volume).
3.3. QUASICLASSICAL APPROACH
3.3.1. Equilibrium Theory

A complementary approach, developed to study hybrid structures, employs
Green's functions
9(1",;', t)
0)), :F(r,;', t) = -i(T7f;(r, t)7f;e, 0)) .

(59)
Despite their apparent simplicity, the Gor'kov equations governing the dynamics of 9 and F are almost impossible to handle in inhomogeneous situations. On the other hand, the information contained in these equations
is redundant, since usually only properties close to the Fermi energy are
interesting. It is possible in these cases to reduce the Gor'kov equations
to transport-like equations which are much easier to study. These are the
Eilenberger [73] and Usadel [74] equations for clean and dirty systems, respectively.
The main steps are as follows. It is convenient to introduce the center
and to consider
of mass R = (f + r') /2 and relative coordinates if = f the Fourier transform of the Green's functions with respect to the latter
(since we are dealing with time-independent situations we use the energy
representation). The Green's function show strong oscillations as a function
of the relative coordinate on the scale of the Fermi wavelength AF. If one is
interested only in variations on scales much larger than AF it is sufficient to
= -i(T7f;(r, t)7f;t(rl,
r'
436
consider the quasiclassical Green's functions obtained by integrating g,:F

over ~P = p2/2m - f.1,
...
{ 9
f } (E,R,VF)
f d 3P {
=;:i f
d~p
g} (E,R,[f)e. . .~ ~
:F
1p . p
(60)
Further simplifications are possible if the system is dirty and the dependence on the direction of VF is weak (the system is nearly isotropic). In
this case 9 and f can be expanded in spherical harmonics, g(E, R, VF) =
G(E, R) + VF . G(E, R) and f(E, R, VF) = F(E, R) + VF . F(E, R). An expansion yields the Usadel equation
.
-IEF
- 6.G
D [G\1
~2 F ="2
F\1 2 G]
(61)
Inelastic interactions can be accounted for by the shift -iE -+ -iE +

1/ (2Tin), by the inelastic scattering rate. Pair-breaking effects add a further
term (1/Ts )GF on the left hand side. The magnetic field is introduced
through the gauge invariant derivative, \7 -+ \1 - 2ieA, acting on F.
The diagonal and off-diagonal component satisfy a normalization condition, G2 + F2 = 1, which is automatically guaranteed if we choose a
parameterization F = sin () and G = cos ().
The formalism is completed by the self-consistency equation for the gap
(62)
We further have to specify the boundary conditions at the SN interface
(which we assume to be located in the x = 0 plane). In the absence of an
extra boundary potential these read [75]
(63)
Hence, the parameter which describes the properties of the interface is the
ratio of the coherence lengths over the ratio of the conductivities in the two
materials r = aN~Ts/ aS~TN.
This semiclassical approach has recently been applied to study the local
density of states (DOS) in hybrid structures [51]. Earlier theoretical treatments of this problem can be found in Refs. [76, 77]. Experimentally the
DOS is studied by attaching several tunnel junctions at certain distances
from the interface and measuring the I-V characteristics [50].
437
(b)
(a)
1.0
N(E)/No
0.5
5
bulk
...... x=-1.5~
-- x=-0.75~
-- x=o
N(E)/NO
- x=0.75~
-- x=1.5 ~
-- x=3~
3
2
- ..
0.0 '----_~....I..--~~--'-~
0.0
0.5
1.0
'~~'~~'~'~'~'='~.'.'/'.
~~~~~~~-~
0.0
E/L\
0.5
E/L\
1.0
1.5
Figure 9. The density of states on the normal (a) and superconducting side (b) of a NS
heterostructure at different distances from the interface. The length scale is ~ = JDI2!!l..
A nonzero pair breaking strength liT. = O.03~ and r = 1 have been assumed. From
Ref. [51].
The local DOS is defined through the retarded one-electron Green's

function gR(x, x'; t) == -i( {1jJ(x, t),-1j7(x', O)t}) O(t) as
N(x, E)
= --1 1m
1("
00
dt eiEt gR(x, x; t)
-00
= N(O) ReG(E) .
(64)
If the metal is a Fermi liquid the DOS is almost featureless", N (0) while in
the superconductor it behaves like N(E) = N(0)E/JE2 - ~2. This raises
the question how the DOS behaves close to an NS interface to interpolate
between these two very different limits.
Due to the proximity effect, the DOS indeed acquires nontrivial structure. Results are shown in Fig. 9 (a) for the normal side of the interface.
It shows a subgap structure (a bump) and a depression close to the Fermi
energy. These features tend to disappear when one moves away from the
interface. In the absence of pair breaking the DOS vanishes at the Fermi
level. On the superconducting side the singularity at ~ is suppressed and
a finite DOS appears also at low energies, as shown in Fig. 9 (b).
If the dimensions of the normal metal are finite (a slab of thickness L), a
true gap Eg appears in the DOS of the normal metal. This mini-gap scales
with the length and is related to the Thouless energy D / L2. A fit to the
numerical results is
2 ,
Eg '" (~+
Lr
where ~ = v'D/2ll., implying that the effective diffusion length is
f'V
+ L.
438
3.3.2. Nonequilibrium Situations

To describe systems with a finite applied voltage, the formalism of nonequilibrium superconductivity [78, 79, 80] should be used. It is based on the
real-time Keldysh technique [80, 81], which involves matrix Green's functions
G=
AR AK)
( G GAA
o G
(65)
having retarded, advanced and Keldysh (R,A,K) components. Each of these

are 2 x 2 matrices in Nambu space typical for superconductivity
G
A
(Gpt -GP) '
(66)
whose entries are quasiclassical Green's functions, which in the dirty limit
satisfy the Usadel equation (61). The boundary conditions for Keldysh
Green's functions at NS-interfaces have been derived by Zaitsev [82]. Applications to diffusive NS heterostructures have been discussed by Volkov
et al. [83].
As an example, and application of the Keldysh technique, we consider
the transport through a diffusive wire of length L connected to a normal
and a superconducting reservoir via metallic contacts. One of the striking
effects in the transport of this system is the non-monotonic temperature
dependence when the temperature is of the order of the Thouless energy
EL = D / L2. The resistance of this system initially decreases when the
temperature is lowered but approaches again the normal state resistance
at T = O. This effect was analyzed theoretically in Refs. [49, 84, 85] and
experimentally in Ref. [43].
The differential conductance, normalized to its value if the superconductor was not present, can be expressed as
GN =
J.2T
(X> dE
io
D(E)
cosh2(E/2T)
(67)
where D(E) is the effective energy-dependent transparency to be determined microscopically from the quasiclassical equations. It is the presence
of the electric field combined with the proximity effect which renders this
situation a nonequilibrium one.
At temperatures much lower than the Thouless energy EL the conductance increases quadratically with temperature
(68)
439
1.0
I /)
c:
,'-
.~
E
.... 0.5
-w
c:
I ",
(J)
,~,
-.:---~,t
~!
.............. \ ",,~ I
:J
"'C
(ij
"
", 'I
-,
:!::
- - T/EL=0.1
- - T/EL=1
------ T/~=10
---- T/E L=100
----- TIEl =1 000
I,
II
I'
\ II
\ \I
I I'
II I
I I
\I
I
I
I
I
I
0.0 '---~---'--~-'-~--'--~-'---~--"
0.0
0.2
0.4
0.6
0.8
1.0
x/L
Figure 10. Electric field in the normal bridge between a bulk superconductor (x
and a bulk normal metal (x ~ 0). From Ref. [85].
L)
where A is a constant. Although the low-temperature conductance coincides

with the normal state value, the wire is influenced by the superconductor,
as can be seen in the local density of states [85]. At higher temperatures
T > EL the conductance decreases with rising temperature
(69)
where B is a constant. In this regime the coherence length eT in the normal
wire is shorter than L. Hence the resistance of the structure is determined
by the portion of the wire, rv L - eT, which is still normal [85].
In the non-equilibrium situation considered here, it is essential to analyze the penetration of the electrical field in the wire. This problem was
considered in Ref. [86] for the infinite wire case and in Ref. [85] for the case
where the wire is attached to two reservoirs. At high temperatures, the field
is essentially constant, but at low temperatures it has a non-monotonic
behavior. This in turn is responsible for the non-monotonic temperature
dependence of the conductance. The electric field in the wire is plotted in
Fig. 10.
In the presence of tunnel barriers with resistance larger than the Drude
resistance of the wire, the electric field is confined to the barrier. In this
440
case the nonequilibrium effects related to the electric field in the wire, and
responsible for the non-monotonic temperature dependence disappear.
Recent experiments [47, 48] have also studied more complex NS structures where supercurrents can flow between at least two superconducting
reservoirs [47] or are induced by a magnetic flux throu'gh a loop within the
structure [48]. In the picture of Andreev reflections, interference between
quasiparticles acquiring a superconducting phase during the reflection process occurs. As the Usadel equations (61) describe the modulation of the
Green's functions by possible gradients of the superconducting phase, the
influence of these currents on the system conductance (67) can be easily
calculated within the quasiclassical approach. These effects remain pronounced even at higher temperatures, when the coherence length ~TN is
smaller than the geometrical lengths of the system. In this case one could
expect that superconducting correlations are destroyed before interference
occurs, hence the effect should be absent and supercurrents are exponentially small. However, low energy channels (E ~ EL) can still interfere
and contribute to the conductance with a relative weight of EL/T. Their
contribution remains pronounced. This is characteristic for linear response
quantities, in contrast to thermodynamic ones like the supercurrent.
3.4. SUPERCONDUCTOR-LUTTINGER LIQUID SYSTEMS
As discussed already in the first part of this chapter, electron-electron interaction plays an important role in systems of reduced dimensionality. The
interplay of proximity effect and charging was discussed in [36]. Another
class of systems in which interaction is of fundamental important is that
of quantum wires. In this case the capacitance model cannot applied any
longer, instead a paradigm model for interacting one-dimensional electron
systems is the Luttinger model (see the chapter by Fisher and Glazman
in this volume for an introduction) In this last section we briefly review
some properties of hybrid systems of superconductors and a Luttinger liquid. In 1D, interactions have drastic consequences. For instance, there are
no fermionic quasiparticle excitations, and the transport properties cannot
be described in terms of the conventional Fermi-liquid approach. Instead
the low-energy excitations of the system are independent long-wavelength
oscillations of the charge (p) and spin density (a), which propagate with different velocities. For a quantum wire with an arbitrarily small barrier this
leads to a complete suppression of transport at low energies [87,88,89].
Hybrid S-LL have been studied in the two extremes of tunneling junction and of perfectly transparent interfaces. In the first case the tunneling
Hamiltonian is used [59], while in the second case a new type of bosonization developed in Ref. [60] is employed. In this section we consider only
441
the tunneling density of states in a LL with a highly transparent S- LL

interface [62].
The Hamiltonian of a LL can be written in bosonized form as
IfF
~~Vj Jdx [~(V8j)2+ :P</>j)2] ,
(70)
where j = p, a, and Vj = (2/9j )VF are the renormalized interaction-dependent Fermi velocities for charge and spin density excitations. For repulsive,
spin-independent interactions we have 9p < 2 and 9(J = 2. The Fermi field
operators are decomposed in right- and left-moving Fermion operators 1/J+,s
and 1/J-,s, respectively, 1/Js = eikFX 1/J+,s + e- ikFX 1/J_,s, where kF is the Fermi
wave vector. The fields 1/J,s in turn can be expressed through Boson operators
(71)
where ()s = ,fi(()p + s()(J) and s = ,fi(p + s(J)' The density of electrons
per spin in the LL is Po = k F /27r. The fields (), can be decomposed in a
normal mode expansion which incorporates the boundary conditions at the
S-LL interfaces. For a LL coupled to two superconductors at a distance L,
Maslov et at. [60] obtained the result
()p(X)
~2(J + X)Ix + {~.

'j
.
/,qsm(qx)(bp,q
-'
bp,q),
Z
1 (0)
yf1i()(J
()(J(x)
7r
'j
.
+~
- ~ /,qcos(qx)(b(J,q
+'b(J,q),
9(J q>O
~
't 2M 1x + Z'j
2 L/,q 'sm(qx)(b(J,q
(J(x)
p(x)
9p q>O
q>O
1 (0)
yf1ip
+ [if
2L
q>O
'
.,
b(J,q)
' t+'bp,q) .
/,qcos(qx)(bp,q
(72)
(73)
(74)
(75)
Here, b;;d are Bose operators and /,q = exp{ -qa/27r} / y'qL where a is a
short range cut-off. The expansion (72) - (75) is valid at energies smaller
than the superconducting gap ~. The phase difference between the two
superconductors is X; J and M describe topological excitations satisfying
the constraint J +M = odd. Finally, ()~O) and ~O) are canonically conjugate
to M, J. The local density of states (per spin) of the LL measured at a
distance x from the superconducting contact is related to the retarded oneelectron Green's function of the LL by Eq. (64).
As an example we discuss the space and frequency dependent DOS of a
LL contacted at x = 0 with a superconductor. This corresponds to the limit
442
L -7 00 in the mode expansion given by Eqs. (72) - (75). In this case only
the non-zero modes (q > 0) contribute to the local DOS. The correlation
function ("p!(x, t)"ps(x, 0)) can be evaluated using the boson representation
Eq. (71), with the result
("p!(x, O)"ps(x, t))
= 2po II
j=p,u
(76)
at distance x from the LL-S interface. Here Ij = (gj/16 - 1/(4gj)) and
'fJj = (gj/16 + 1/(4gj)). At small energies w ~ .6., the DOS behaves as
(77)
The exponent of the DOS is negative (gp < 2), which implies a strong
enhancement at low energies whereas in the absence of the superconductor
the DOS of the LL vanishes at the Fermi energy
(78)
Thus the presence of the superconductor changes the properties of the
Luttinger liquid in a qualitative way. Although we consider a clean S-LL
interface, backscattering is induced by the superconducting gap, which reflects low-energy electrons either directly or via (multiple) Andreev processes. The enhanced DOS as a function of frequency, -Eq. (77), is schematically drawn in Fig. 11; for comparison we also show the vanishing DOS in
absence of the superconductor, Eq. (78).
On the other hand, at low energies w the enhancement of the DOS
persists over large distances x(w) '" vp/w from the interface. On the other
hand, the induced pair amplitude in the LL, which is characteristic of the
presence of the superconductor, decays as a power [60] of the distance x.
This profound difference in the space dependence demonstrates that the
DOS provides different information compared to the proximity effect. The
reason why the DOS does not approach the well-known behavior of an
Luttinger liquid far from the superconducting contact is in part related to
the fact that we are considering a clean wire. In this case the states in the
LL are extended and the DOS enhancement does not depend on x.
Acknowledgments
We would like to thank our colleagues with whom we had been working
on the problems reviewed in this article, W. Belzig, C. Bruder, G. Falci,
F. W. J. Hekking, A. Odintsov, J. Siewert, L.L. Sohn, F. K. Wilhelm, and
443
N(oo)
~S
~-xx-- " \
00
Figure 11. Schematic dependence of the DOS on frequency for a pure LL (dashed line)
and for a LL connected to S (solid line). Inset: Luttinger liquid, connected adiabatically
to a superconductor. The shaded area indicates a tunnel junction with a normal metal
used to measure the DOS in the LL at a distance x from the interface. From Ref.[62]
A. D. Zaikin. The work has been supported by the 'Sonderforschungsbereich' 195 of the 'Deutsche Forschungsgemeinschaft'. Also the support by
the A.v.Humboldt award of the Academy of Finland (G.S.) is gratefully
acknowledged.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
D. V. Averin and K. K. Likharev, in Mesoscopic Phenomena in Solids, B. L. Altshuler, P. A. Lee and R. A. Webb, eds., p. 173 (Elsevier, Mnsterdam, 1991).
Single Charge Tunneling, NATO ASI Series, Vol. B 294, eds. H. Grabert and M. H.
Devoret, (New York, Plenum Press 1992).
M. Tinkham, Introduction to Superconductivity, 2nd Edition, McGraw Hill (1996).
Mesoscopic Superconductivity, Proceedings of the NATO ARW, F. W. J. Hekking,
G. Schon, and D. V. Averin, eds., Physica B 203 (1994).
C. Bruder, in Superconductivity Review 1, 261 (1996).
G. Schon, in Quantum Transport and Dissipation, VCH Publishers, Chapter 4, to
be published.
F. W. J. Hekking, L. 1. Glazman, K. A. Matveev, and R. 1. Shekhter, Phys. Rev. Lett.
70, 4138 (1993).
F. Guinea and G. Schon, Physica B 152, 165 (1988).
F. W. J. Hekking and Yu. V. Nazarov, Phys. Rev. Lett. 71, 1625 (1993); Phys. Rev.
B 49, 6847 (1994).
W. Tichy, Diplomthesis, University Karlsruhe (1996).
D. V. Averin and Yu. V. Nazarov, Phys. Rev. Lett. 69, 1993 (1992).
P. Lafarge, P. Joyez, D. Esteve, C. Urbina, and M. H. Devoret, Phys. Rev. Lett. 70,
994 (1993).
M. T. Tuominen, J. M. Hergenrother, T. S. Tighe, and M. Tinkham, Phys. Rev. Lett.
69, 1997 (1992); Phys. Rev. B 47, 11599 (1993).
T. M. Eiles, J. M. Martinis, and M. H. Devoret, Phys. Rev. Lett. 70, 1862 (1993).
444
15.
36.
J. M. Hergenrother, M. T. Tuominen, and M. Tinkham, Phys. Rev. Lett. 72, 1742

(1994); J.M. Hergenrother et al., page 327 in Ref. 4.
G. Schon and A. D. Zaikin, Europhys. Lett. 26, 695 (1994).
G. Schon, J. Siewert, and A. D. Zaikin, p. 340 in Ref. [4], and in Quantum Dynamics
of Submicron Structures, NATO ASI Series B 291, eds. H. Cerdeira, B. Kramer,
and G. Schon, p. 489 (1995).
A. O. Caldeira and A. J. Leggett, Ann. Phys. (NY) 149, 374 (1983).
U. Weiss, Quantum Dissipative Systems, Series in Modern Condensed Matter
Physics, Vol. 2 (World Scientific, 1993).
U. Eckern, G. Schon, and V. Ambegaokar, Phys. Rev. B 30, 6419 (1984).
G. Schon and A. D. Zaikin, Phys. Rep. 198, 237 (1990).
F. W. J. Hekking, L.1. Glazman, and G. Schon, Phys. Rev. B 51, 15312 (1995).
T.A. Fulton, P.L. Gammel, D.J. Bishop, L.N. Dunkleberger, and G.J. Dolan, Phys.
Rev. Lett. 63, 1307 (1989).
A. Maassen v.d. Brink, G. Schon, and L. J. Geerligs, Phys. Rev. Lett. 67, 3030
(1991); A. Maassen v.d. Brink et al., Z. Phys. B 85, 459 (1991).
D. B. Haviland, Y. Harada, P. Delsing, C. D. Chen, and T. Claeson, Phys. Rev. Lett.
73, 1541 (1994).
J. Siewert and G. Schon, Phys. Rev. B 54, 7421 (1996).
KA. Matveev, M. Gisselfa.J.t, L.1. Glazman, M. Jonson, and RI. Shekter, Phys.
Rev. Lett. 70, 2940 (1993); K A. Matveev, L.1. Glazman, and RI. Shekter, Mod.
Phys. Lett. B 8, 15 (1994).
P. Joyez, Ph.D. Thesis, Universite Paris 6 (1995)
M. H. Devoret, private communication.
H. Schoeller and G. Schon, Phys. Rev. B 50, 18436 (1994).
J. Konig, H. Schoeller, G. Schon, Europhys. Letters 31,31 (1995); J. Konig et al.
in Quantum Dynamics of Submicron Structures, NATO ASI Series B 291, eds. H.
Cerdeira, B. Kramer, and G. Schon, p. 221 (1995).
L.1. Glazman and K. A. Matveev, Sov. Phys. JETP. 71, 1031 (1990); K A. Matveev,
Sov. Phys. JETP 72, 892 (1991).
S. V. Panyukov and A. D. Zaikin, Phys. Rev. Lett. 67, 3168 (1991); D. S. Golubev
and A. D. Zaikin, Phys. Rev. B 50, 8736 (1994).
G. Falci, G. Schon, and G. T. Zimanyi, Phys. Rev. Lett. 74, 3257 (1995); and p.
409 in [4].
H. Grabert, Phys. Rev. B 50, 17364 (1994); X. Wang and H. Grabert, Phys. Rev.
B 53, 12621 (1996).
C. Bruder, R Fazio, and G. Schon, Phys. Rev. B 50, 12766 (1994); and p. 240 in
37.
R Bauernschmitt, J. Siewert, A.A. Odintsov, and Yu.V. Nazarov, Phys. Rev. B
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
38.
39.
40.
41.
42.
43.
44.
45.
46.
[4].
49, 4076 (1994).

L. G. Aslamazov, A. I. Larkin, and Yu. N. Ovchinnikov, Zh. Eksp.Teor. Fiz. 55,
323 (1968) [Sov. Phys. JETP 28, 171 (1969)].
C.W.J. Beenakker in Mesoscopic Quantum Physics, edited by E. Akkermans, G.
Montambaux, and J.-L Pichard (North Holland, Amsterdam) 1995.
C.J. Lambert and R Raimondi, J. Phys. Condo Matt. (to be published).
G. Deutscher and P. G. de Gennes, in Superconductivity, edited by R D. Parks
(Marcel Dekker, New York, 1965), VoLlI, p. 1005.
A. F. Andreev, Zh. Eksp. Teor. Fiz. 46, 1823 (1964) [Sov. Phys. JETP 19, 1228
(1964)].
P. Charlat, H. Courtois, Ph. Gandit, D. Mailly, A.F. Volkov, and B. Pannetier,
Phys. Rev. Lett., to be published.
1.0. Kulik, Zh. Eksp. Teor. Fiz. 57, 1745 (1969) [Sov. Phys. JETP 30, 944 (1970)].
H. Pothier, S. Gueron, D. Esteve, and M. H. Devoret, in Ref[4]; Phys. Rev. Lett.
73, 2488 (1994).
H. Nakano and H. Takayanagi, Sol. St. Comm. 80, 997 (1991).
445
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
V.T. Petrashov, V.N. Antonov, and M. Persson, Physica 'Scripta 42, 136 (1992);
V.T. Petrashov, V.N. Antonov, P. Delsing, and, T. Claeson, Phys. Rev. Lett. 70,
347 (1993); Phys. Rev. Lett. 74, 5268 (1995).
H. Courtois, Ph. Gandit, D. Mailly, and B. Pannetier, Phys. Rev. Lett. 76, 130
(1996).
Yu. V. Nazarov and T. H. Stoof, Phys. Rev. Lett. 76, 823 (1996); Phys. Rev. B 53,
14496 (1996).
S. Gm!ron, H. Pothier, N. O. Birge, D. Esteve, and M. Devoret, Phys. Rev. Lett.
77, 3025 (1996) .
W. Belzig, C. Bruder, and G. Schon, Phys. Rev. B 54, 9443 (1996).
J. Nitta, T. Azaki, H. Takayanagi, and K Arai, Phys. Rev. B 46,14286 (1992).
A. Dimoulas, J. P. Heida, B. J. van Wees, T. M. Klapwijk, W. v.d. Graaf, and G.
Borghs, Phys. Rev. Lett. 74, 602 (1995).
C. W. J. Beenakker and H. van Houten, Phys. Rev. Lett. 66, 3056 (1991).
A. Furusaki, H. Takayanagi, and M. Tsukada, Pllys. Rev. Lett. 67, 132 (1991).
A. Kastalsky, A. W. Kleinsasser, L. H. Greene, R. Bhat, F. P. Milliken, and J. P.
Harbison, Phys. Rev. Lett. 67, 3026 (1991).
B. L. Altshuler, D. E. Khmelnitskii, and B. Z. Spivak, Solid State Comm. 48, 841
(1983).
J. Voit, Rep. Prog. Phys. 58, 977 (1995).
R. Fazio, F.W.J. Hekking" and A.A. Odintsov, Phys. Rev. Lett. 74, 1843 (1995).
D.L. Maslov, M. Stone, P.M. Goldbart, and D. Loss, Phys. Rev. B 53, 1548 (1996).
M.P.A. Fisher, Phys. Rev. B 49, 14550 (1994).
C. Winkelholz, R. Fazio, F.W.J. Hekking, and G. Schon, Phys. Rev. Lett. 77, 3200
(1996).
see e.g. the textbook by S. Datta, Mesoscopic Electron Transport, Cambridge University Press (1995).
C.J. Lambert, J. Phys. C 3, 6579 (1991).
C.W.J. Beenakker, Phys. Rev. B 46, 12481 (1992).
G.E. Blonder, M. Tinkham, and T.M. Klapwijk, Phys. Rev. B 25, 4515 (1982).
.
A.L. Shelankov Sov. Phys. Solid St. bf 26, 981 (1984).
M.P. Anantram and S. Datta, Phys. Rev. B 53, 16390 (1996).
G.B. Lesovik, A.L. Fauchere, and G. Blatter, Phys. Rev. B 55, 3146 (1997).
J. Sanchez-Canizares and F. Sols, Phys. Rev. B 55, 531 (1997).
P. Cook, R. Raimondi and C.J. Lambert, Phys. Rev. B 54,9491 (1996).
K Usadel, Phys. Rev. Lett. 25, 507 (1970).
M.Y. Kupriyanov and V.F. Lukichev, Zh. Eksp.Teor. Fiz. 94, 139 (1988) [Sov. Phys.
JETP 67, 1163 (1988)].
W.L. McMillan, Phys. Rev. 175, 537 (1968).
A.A. Golubov and M.Y. Kupriyanov, J. Low. Temp. Phys. 70,83 (1988).
A.I. Larkin and Yu.N. Ovchinnikov in Nonequilibrium Superconductivity, eds. D.N.
Langenberg and A.I. Larkin, (Elsevier, Amsterdam, 1985).
A. Schmid and G. Schon, J. Low. Temp. Phys. 20, 207 (1975).
A. Schmid in Nonequilibrium Superconductivity, Phonons, and J<apitza Boundaries,
NATO ASI Series B 65, ed. KE. Gray, (Plenum, New York 1981)
J. Rammer and H. Smith, Rev. Mod. Phys. 58,323 (1986).
A.V. Zaitsev, Sov. Phys. JETP 59, 1015 (1984); M.Yu. Kuprianov and V.F. Lukichev, Sov. Phys. JETP 94, 139 (1988).
A.F. Volkov, A.V. Zaitsev, and T.M. Klapwijk, Physica C 59, 21 (1993).
A.F. Volkov and C.J. Lambert, J. Condo Mat., 8, L45 (1996).
A.A. Golubov, F.K. Wilhelm, and A.D. Zaikin, to be published in Phys. Rev. B.
F. Zhou, B. Spivak, and A. Zyuzin, Phys. Rev. B 52,4467 (1995)
446
88.
89.
K.A. Matveev and L.I. Glazman, Phys. Rev. Lett. 70, 990 (1993).
A. Furusaki and N. Nagaosa, Phys. Rev. B 47,4631 (1993).
ULTRASMALLSUPERCONDUCTORS
D. C. RALPH, C. T. BLACK, J. M. HERGENROTHER,

J. G. LV AND M. TINKHAM
Department of Physics and Division of Engineering
and Applied Sciences
Harvard University, Cambridge, MA 02138, USA
1. Introduction
In recent years, advances in techniques for fabricating small electrical devices have allowed the investigation of several open questions in the theory
of superconductivity. In this chapter we will focus on two series of experiments which have probed the superconducting properties of small metal
grains. The first issue that we will discuss is the nature of superconductivity in a sample that is electrically isolated, so as to have a fixed number of
electrons. The BCS theory of superconductivity is most conveniently formulated in terms of a grand canonical ensemble. That is, within the BCS
theory one imagines that a sample of superconducting material is in electrical contact with a bath of electrons, so that the number of electrons within
the superconductor may fluctuate. We have investigated the experimental
consequences of having, instead, a constant number of electrons, by using
devices capable of controlling the number of electrons on a single metal
grain with unit precision. Superconducting particles with even numbers of
electrons have been found to have properties that are dramatically different
from odd-electron particles, on account of the pairing nature of superconductivity. Remarkably, this is true even when there are as many as 109
conduction electrons within a superconducting grain.
The second series of experiments probes the properties of even smaller,
nm-scale, superconducting samples. For particles on the order of 10 nm in
diameter, it is no longer appropriate to use the usual approximation that
the electronic energy levels within the metal sample effectively form a conCurrent address for DCR: Laboratory of Atomic and Solid State Physics, Cornell
University, Ithaca, NY, 14853 USA. Current address for CTB: IBM T. J. Watson Research
Laboratory, Yorktown Heights, NY 10598 USA.
447
L. L. SOM et al. (eds.), Mesoscopic Electron Transport, 447-467.

448
tinuum of states. In fact, the individual, discrete energy levels for electrons
in the metal "box" can be resolved experimentally at low temperatures. By
examining the spectrum of individual energy levels within single superconducting particles, we have been able to view in detail how these levels are
modified in a superconducting system, as compared to simple models for
non-interacting electrons. We have also begun investigations into the consequences of the level discreteness on the existence of superconductivity.
P. W. Anderson predicted in 1959 [1] that the formation of a superconducting electronic state should no longer be energetically possible within
metal particles small enough that the average energy-level spacing, 6, is
larger than the bulk superconducting gap ~o. We have found, thus far,
that superconductivity is present in particles where 6 is comparable to, but
smaller than ~o. How to determine experimentally whether or not superconducting correlations are present in smaller particles, with larger 6, IS
still an unresolved question.
2. Measuring the Presence of Superconductivity

The traditional experimental hallmarks of superconductivity are zero electrical resistance and the Meissner effect - the exclusion of an applied magnetic field from the interior of a superconducting sample. In the devices
that we will describe, neither of these effects are present. We will investigate superconducting particles using electron tunneling, so there will be a
non-zero (in fact, quite large) resistance. The particle size (particularly for
the nm-scale samples) will be smaller than the magnetic field penetration
depth, meaning that there will also be little appreciable Meissner effect.
It is therefore worthwhile to begin our discussion with the basic question,
"What, exactly, do we mean by the existence of superconductivity in a
small particle with fixed electron number?"
A good discussion of this issue has been given by von Delft et al. [2]. The
key idea is that the essence of superconductivity is the presence of a paircorrelated electronic ground state. When there is an attractive interaction
between electrons, the lowest energy state of the overall electronic system
is not given simply by filling a set of independent-electron states up to the
Fermi energy. Instead, the electronic system can lower its total energy by
allowing the virtual population of some electronic pair states with energy
above the non-interacting Fermi level, and depopulation of states below the
Fermi level. This creates phase space for pair scattering, and thus allows the
attractive electronic interaction to lower the overall ground state energy.
The experimental signature of the resulting pair-correlated superconducting
ground state is an energy gap, ~, for the excitations of quasiparticles above
the ground state. In our experiments, we will use tunneling techniques to
449
measure the effects of this energy gap, and how it varies as a function of
external parameters such as magnetic field.
In all of the experiments to be discussed here, it is essential that one
be able to control the number of electrons on the metallic grain with unit
precision. Since the number of electrons may be as large as 109 in our
micron-scale islands, this might appear to be a daunting task. In fact, only
two key requirements must be met. Firstly, the island must be sufficiently
isolated that the number of electrons is a "good quantum number", despite quantum fluctuations of electrons tunneling on and off the island.
This requires that the tunneling resistance RT between the island and all
electron reservoirs must be substantially larger than the quantum resish/e 2
26 kO. Secondly, to prevent smearing of the electron
tance RQ
number by thermal fluctuations, the energy difference associated with a
change in charge by e must be large compared to the thermal energy kBT.
Starting with a neutral grain, the energy required to add or subtract one
electron is Ec = e2 /2C"" where C", is the total capacitance from the grain
to its environment (which is dominated by the capacitance of the tunnel
junctions). This Ec will exceed thermal energies if C", e2 /kBT, e.g., if
C", < 10- 15 F = 1 fF and T < 1 K. These two requirements on RT and
C", can both be satisfied simultaneously by using tunnel junctions of sufficiently small area A, since this reduces C", ex: A while increasing RT ex: l/A.
The tunnel junctions in the "micron-scale" devices that we will discuss have
A < 10- 10 cm 2 , C", < 1 fF, and Rr > 100 kO, while the nanoscale samples
have A < 10- 12 cm2 , C", < 10 aF, and Rr > 1 MO. Both types of samples
are studied at dilution refrigerator temperatures < 1 K.
Our experiments to probe small superconducting particles are carried
out using a "single-electron tunneling transistor" configuration (Fig. 1).
This consists of the metal particle under study, coupled to two leads by high
resistance, low capacitance tunnel junctions, and electrostatically coupled
to a gate electrode. The gate voltage Vg can be parameterized in terms of
a continuous polarization charge induced on the island, Qo = Cg Vg , and
n discrete electrons will enter the island through the tunnel junctions to
neutralize this charge as closely as possible (i.e., to within e/2). The
experimentally measured quantity is the current I(V, Vg ), or equivalently
I(V, Qo), which results if a bias voltage V is applied across the two tunnel
junctions.
I'V
I'V
3. Even-Odd Effects in Micron-Scale Aluminum Islands

The first experiments that we will consider were performed with aluminum
islands having dimensions about 20 nm thick by 70 nm wide by 2 microns
long. The corresponding energy spacing between electronic states inside the
450
+V/2
Figure 1.
- V /2
Schematic diagram of a single-electron tunneling transistor.
particle is on the order of 30 neV, or 0.3 /l-K. At experimental temperatures

near 20 mK, the levels can therefore safely be considered a continuum. For
ease of presentation in this section, we will display a sample of experimental
data taken by the Harvard group. However, the development of this field
was the work of many different groups [e.g., 3-12].
The detailed interpretation of the J(V, Qo) data for a single-electron
transistor made from a micron-scale island requires a computer solution of
the many coupled kinetic equations describing all possible tunneling and
relaxation processes (the "orthodox theory" [13]). However, we can account
for the qualitative features of the observations at low bias voltages by a statistical model based on near-equilibrium conditions [3, 14]. For simplicity we
first consider a device with a normal-state metal island and electrodes. The
(n-dependent part of the) Coulomb energy resulting from having an excess
electronic charge ne which inexactly cancels Qo is E = (Qo - ne)2 /2Cy:,. As
shown in Fig. 2(a), for a normal-state metal island, the lowest electronic
energy for n excess electrons can therefore be represented by a parabola
on an energy vs. Qo graph, with the parabolas for different values of n
displaced successively by e along the Qo axis. Where the parabolas cross
(at half-integer values of Qo, where the energy is the same for nand n + 1
electrons), one electron can enter and then leave the island with no energy
barrier, and a current can flow via the metal particle in response to a small
bias voltage. At other values of Qo, there is a Coulomb barrier to charge
transfer to the island, and no current flows at low bias voltage. This is
the "Coulomb blockade". Thus, J(V, Qo) as a function of Qo at small fixed
V consists of a series of current peaks separated by e along the Qo axis;
451
3.0
2.5
2.0
a) Normal
State
1.5
1.0
0.5
3.0
2.5
2.0
1.5
1.0
0.5
3.0
2.5
2.0
ill
1.5
1.0
0.5
0.0
-2
-I
Qje
Figure 2. The n-dependent part of the electronic energy in a micron-scale metal particle,
as a function of the gate charge Qo = Cg Vg . (a) Normal island. (b) Superconducting island
with ~ < Ec. (c) Superconducting island with ~ > Ec. Hatched lines denote energy
states which border a continuum of levels. Unhatched lines denote non-degenerate fully
Cooper-paired states.
i.e., the current is e-periodic. This is what is observed experimentally for

normal-metal devices.
For a superconducting island, quite different behavior is seen. Representative data of I(Qo) at a fixed value of V is shown in Fig. 3 for an
aluminum island in tunnel contact to normal-metal (gold) electrodes [15].
At low temperature, current flow is 2e-periodic as a function of Qo, not
e-periodic as for the normal-metal island. This means that the current is
periodic in the addition of 2 (but not one) electrons to the Al island. As
the temperature is slowly raised, the pattern of current flow becomes more
complicated, but still with 2e periodicity in the Qo dependence. Finally,
at temperatures above about 280 mK, the current becomes periodic with
a period of Ie. The Al island is still superconducting at this temperature,
since the critical temperature of the Al grain is about 1.6 K.
The data for the superconducting island can be understood by consider-
452
240mK
225mK
2
130mK
-6
-5
-4
-3
-2
-I
Qcle
Figure 3_ Current vs. gate charge Qo for a tunneling transistor with a micron-scale Al
island and Au electrodes. The bias voltage V= 125 p,V, and the temperature ranges from
100 mK to 300 mK. Data for different temperatures are artificially offset on the current
axis. The 2e periodicity disappears above T* ~ 280 mK.
ing how the energy diagram is modified by the presence of superconducting

pairing in the island (Fig_ 2(b,c)). Because of pairing, even and odd values of n are not equivalent in the superconducting state. If n is odd, one
electron in the particle must be occupy an unpaired quasiparticle state,
which increases the system energy by the BeS energy gap b.. This causes
the odd-n parabolas to be lifted up by b. relative to those for even values
of n, as shown in Fig. 2(b,c). Two cases arise. If b. < Ec [Fig. 2(b)], then
as Qo is swept the ground state alternates between even and odd n, but
with even n predominating. If b. > Ec [Fig_ 2(c)], then the ground state
always has n = even whatever the value of Qo. It is this second case which
applies to the data of Fig. 3 at the lowest temperatures. In this case, electron tunneling at low temperatures and low bias voltages must occur by
means of the simultaneous transfer of two electrons on or off the island, a
process known as Andreev tunneling. In either of the two cases, the pattern
453
of crossing points or degeneracies repeats only after a period of 2e, so that

I (Qo + 2e) = I (Qo) -=J. I (Qo + e), and the current is said to be 2e-periodic
in Qo, or to show an even-odd effect.
Since the even-odd effect does not persist to the macroscopic limit, we
have studied what factors limit its observability. As first shown by Tuominen et al. [16]' and subsequently demonstrated in greater detail by Tinkham
et al. [14], the even-odd effect fades out at a characteristic temperature T*
(280 mK in Fig. 3). This is the temperature at which the free energy difference between states with even and odd electron numbers goes to zero,
because the larger entropy of the odd n state balances its greater energy.
Accordingly, T* ~ (l1/kB)ln(Neff), where Neff is essentially the number of
quasiparticle states in the island within kBT above the energy gap 11. Upon
increasing the temperature from 0, the average number of thermally-excited
quasiparticles in an even-electron grain goes from zero to a very large number in a narrow range of temperature around T*. (In the same range of temperature, the number of thermally-excited quasiparticles in an odd-electron
particle goes from 1 to many.) Above T*, parity-independent quasiparticle current dominates transport, and even-odd effects are swamped. For a
typical island length (~ 2 p,m), Neff ~ 104 , and thus T* ~ T e/5. Since
T* depends only logarithmically on sample volume through Neff, it appears superficially that the even-odd effect should be observable even in
macroscopic samples, but this is not correct. T* is a only a measure of the
temperature up to which the gap 11 is sufficient to favor an even number of
electrons; this effect is independent of Vg . To get an observable dependence
on the discrete electron number upon sweeping Vg , it must also be true
that the Coulomb energy Ee kBT, as noted earlier. Since Ee depends
inversely (not logarithmically) on the capacitance C and hence on the physical size of the island, this requirement cannot be satisfied in a macroscopic
sample.
Given that the presence of thermally excited quasiparticles is sufficient
to wash out the even-odd effect, it is not surprising that the effect can also
be destroyed by application of a magnetic field which sufficiently reduces
the energy gap relative to kBT [16, 15]. The even-odd effect is also destroyed
by tunnel injection of quasiparticles by bias voltages above the gap, or by
pair-breaking by absorption of photons with energy above 211 (e.g., from 4K
blackbody radiation) . The latter effect provides the basis for a microwave
photon detector of great sensitivity [17].
4. Discrete Electronic Levels in Nanoscale Grains

We now describe a new series of experiments on much smaller aluminum
grains (diameter 3-30 nm), in which the quantum mechanical confinement
454
AI electrode
AI
particle
~ 10 nm size scale
Figure
4. Schematic cross-sectional view of a nanoparticle transistor.
energy of electrons "in a box" leads to resolvable discrete electronic energy

levels. In order to add an electron to the grain, one must not only overcome
the classical Coulomb blockade, but also supply the extra kinetic energy
needed to occupy one of the available quantum states. This makes possible
for the first time a detailed spectroscopy of these discrete states in a metal
[20, 21, 22]. (Similar "artificial atom" levels had been seen earlier in semiconductor quantum dots [18].) Because the average spacing 0 ofthe discrete
energy levels in these samples can be either greater or less than the bulk
superconducting energy gap b. o, depending on the sample volume, these
experiments have also allowed us to begin probing predictions [1, 2, 19]
that superconductivity should be destroyed when 0 > .6. 0 .
Just as in the experiments on "micron-scale" Al particles, we measure
the properties of the nm-scale particles by attaching them to electrodes
via high-resistance tunnel junctions in a single-electron transistor configuration [23]. In order to make electrical contact between the electrodes and
a nanoparticle, however, we make the devices in a rather unusual vertical
geometry. A schematic diagram is shown in Fig. 4. We use electron-beam
lithography and reactive-ion etching to fabricate a bowl-shaped hole in an
insulating Si3N4 membrane, with the opening on the lower edge having
diameter 3-10 nm. We then form a gate electrode by depositing 12 nm of
Al on the flat side (lower side in Fig. 4) of the device. Plasma anodization
of the Al and deposition of insulating SiO are used to provide electrical isolation between the gate electrode and the other parts of the device which
will be deposited on top of it. The formation of the gate electrode does
not block the nm-scale hole. We then make a second (source) electrode by
evaporating 100 nm of Alan the top side of the membrane so as to fill the
bowl, and oxidize to form a tunnel barrier near the lower edge of the Si3N4
membrane. Next we evaporate 2-3 nm of Al on the lower side to form a
layer of electrically isolated Al particles. In about 25% of the samples, a single particle nucleates in contact with the nm-scale tunnel junction. We can
455
20
E-
O
-20
-60
-40
-20
20
40
60
-5~~
____
-40
____
-20
____
____
20
~~
40
V (mV)
Figure 5. Current-voltage curves displaying Coulomb-staircase structure for two different

Al nanoparticle transistors, at equally-spaced increments of Vg . Data for different Vg
values are artificially offset on the current axis.
identify these "good" devices by the presence of "Coulomb-staircase" structure in their J- V curves, as described below. Following a second oxidation,
we finally deposit a final Al (drain) electrode to cover the particles. The
experiments consist of measuring the tunneling current which flows vertically from the top electrode, via a nanoparticle, to the bottom electrode,
as a function of source-drain voltage and gate voltage.
The samples are initially characterized by measuring J(V, Vg ) at 4.2 K.
Fig. 5 displays J-V curves at a sequence of different values for Vg , for two
different samples, over a large range of source-drain voltage. At this high a
temperature, the discreteness of the level spectrum is not yet apparent for
these devices. The form of the J- V curves as a sequence of equally-spaced
sloping steps is a signature that current flow is due to tunneling via a single
nanoparticle. The step-like structure is a consequence of the electrostatic
energy needed to change the charge on the metal nanoparticle by first one,
then two, etc., additional electrons. The fact that, at most values ofVg , there
is negligible current at low source-drain voltages, is a manifestation of the
Coulomb blockade of tunneling. The extent of the Coulomb blockade regime
456
and the positions of the Coulomb staircase steps are continuously tunable
by adjusting Vg. When Vg is used to set the range of the Coulomb blockade
to 0, this corresponds to a Qo value such that the energy of the ground state
with n electrons is equal to the ground state with n + 1 (or n - 1) electrons
(compare Figs. 2 and 3). From the positions of the voltage thresholds for
steps in the J- V curve, the capacitances of the tunnel junctions within the
device can be determined directly [24]. For Fig. 5(a), the lead-to-particle
capacitances are C1 = 3.5 aF and C2 = 9.4 aF, and the gate-to-particle
capacitance is Cg = 0.09 aF; and for Fig. 5(b) the capacitances are 3.4,8.5,
and 0.23 aF. The Coulomb charging energies in all of our nanoscale devices
are much larger than the bulk superconducting gap in Al (Ee = e2 /20'2:, =
3-50 meV, compared to ~o = 0.175 meV), placing the devices in the regime
of Fig. 2(b).
The size of the nanoparticle in each device can be estimated roughly
using a value for the capacitance per unit area, 0.075 aF Inm 2 , determined
from larger tunnel junctions made using our oxidation process. If we make
the crude assumption that our particles are roughly hemispherical, and base
the size estimate on the larger lead-to-particle capacitance, we estimate
radii of 4.5 nm and 4.3 nm for Fig. 5(a,b).
To obtain spectroscopic information about the discrete electronic states
within the nanoparticle, we cool the devices to mK temperatures and measure the J- V curve in the range of the first Coulomb-staircase step, with
Vg held constant so as to fix the classical Coulomb-charging energy. (Note
that we therefore will plot the data using a different procedure than was
followed in studying even-odd effects, where we fixed a small bias voltage and swept Qo = Og Vg looking for e vs. 2e periodicity in the current
through the device.) When the bias voltage is such that the Fermi level
in one of the leads becomes equal to the threshold energy for an electron
to tunnel either into or out of one of the discrete electronic states within
the Al particle (taking the Coulomb charging energy into account), then
electrons can tunnel one at a time through the particle between the source
and drain electrodes, causing a step increase in J vs. V. Additional step
increases occur when the voltage is raised sufficiently to allow tunneling
alternatively via other, higher-energy discrete levels. Thus, a plot of dJ1dV
vs. V displays a series of peaks [see Fig. 6(b)], corresponding to the opening
of these successive resonant tunneling channels. The spectrum of discrete
electronic states within the nanoparticle can be read directly from this
curve, after converting from the bias voltage scale to electron energy, by
taking into account the capacitive division of the applied voltage across the
two tunnel junctions. We will limit the range of bias voltage displayed in
this chapter to include only those discrete levels which together produce
the first Coulomb-staircase step - in other words the range of voltage where
457
-
c.
-a
-:g>
...,
superconducting leads
(0 Tesla)
80
40
(a)
S-Iead data shifted -0.28 mV
200
(!J
100
"'C
8
V (mV)
10
Figure 6. (a) I-V curves near the onset above the Coulomb blockade at 320 mK, reflecting the contribution from several discrete electronic states within an Al nanoparticle.
The trace with superconducting leads shows BCS density of states peaks, and is offset
by 10 pA for clarity. (b) dI / dV curves for the same data, showing peaks associated with
each level in the spectrum. The S-lead data are shifted to compensate for displacement
due to the superconducting energy gap.
the charging energy forces electrons to traverse the device one at a time.
Therefore all the discrete peaks we show will correspond to states with the
same number of electrons within the nanopartic1e - either one more or one
less than in the particle's ground state.
In our earliest experiments, we used grains which were so small that the
typical level spacing 5 was quite large, rv 1 meV. For temperatures from
500 mK down to rv 80 mK, the full width at half maximum of the resonant
tunneling peaks followed that expected from the derivative of the Fermi
functions in the leads, namely 3.5 kBT, or 0.15 meV down to 0.025 meV
(see inset in Fig. 7), so the peaks were always well separated. At the lowest
temperatures, the peak widths were independent of refrigerator temperature, but the peaks at higher bias voltage levels showed increasing width,
perhaps due to nonequilibrium effects (discussed below). The aluminum
electrodes were superconducting in zero magnetic field, but could be made
normal in a field of 300 Gauss. With normal leads, each level was marked
by a step rise in current as the bias voltage reached the resonant condition,
as described above. With superconducting leads, the square step changed
to the singular peak shape of the BeS density of states, suitably shifted in
458
I"'0
:::; 0.1
~ 0.0
:;(
Eo
0.0
0.5
..
T (I<)
2.0
2.2
2.4
2.6
2.8
3.0
V (mV)
Figure 7. Points are tunneling current via one electronic state at 30 mK for superconducting and normal leads. The line is a fit to the BCS density of states. Inset: Temperature
dependence of the full width at half maximum of dI / dV for a low-voltage resonance peak.
voltage to reflect the energy gap D. in the leads and the capacitance ratio
for the two tunnel junctions in the device [20]. (See Figs. 6 and 7.) This
is exactly what would be expected from the Fermi golden rule for the tunneling probability between a single state on the island and the continuous
density of states in the leads [26]. Identical values of the threshold shift for
different steps in each device confirmed that all the steps were due to discrete states within the same nanoparticle, because they were governed by
the same capacitance ratio. For devices with gates, the identification of all
the conductance peaks with a single nanoparticle could also be confirmed
by checking that all the states shifted together in a consistent way as a
function of Vg .
One question that is often asked is whether it is possible to relate the
measured level spectrum to the shape of the nanoparticle. The answer is
almost certainly no for a metal particle. (In this respect, metal particles
are quite different than few-electron semiconductor quantum dots [25].)
For a metal particle, the electronic states that are measured near the Fermi
level correspond to very high quantum numbers - in the range of the tenthousandth conduction electron on the particle. The corresponding wavelength of the electronic eigenstate is close to atomic dimensions. This means
that the movement of even a single atom along the surface of the nanopartide is capable of altering the energy of the eigenstates and shuffling their
spectrum. Infact, it is expected that random matrix theories should provide a good statistical description of the energy levels in metal particles. We
shall comment briefly on the application of random matrix theory below.
In magnetic fields of a few Tesla, the resonant tunneling peaks were
459
1.0
>
:g
0.5
"0
0.0
4.0
400
4.2
4.4
4.6
V (mV)
4.8
5.0
(b)
>
:g
200
"0
0
0.6
0.8
1.0
V (mV)
1.2
1.4
Figure 8. dI/dV vs. Vat 50 mK and H = 0.03,1,2, and 3 Tesla from bottom to top,
for the lowest-energy tunneling resonance of two different samples. (a) Sample identified
as having an initially even number of electrons because this first transition exhibits a
Zeeman spin splitting. (b) A sample identified as having an odd number of electrons
because the first transition does not show Zeeman splitting
found to split on account of breaking the electronic spin degeneracy of the

levels. The splitting corresponded to g-values of typically "-' 1.85 - 2.0. In
addition there was also an underlying shift of the center of gravity of the
peaks with applied magnetic field, presumably due to an orbital contribution to the energy levels. In some cases, an "avoided crossing" was observed,
indicating the effect of spin-orbit coupling in mixing spin eigenvalues [22].
It is noteworthy that we could distinguish grains with even vs. odd
numbers of electrons by observing the spin splitting, as follows: If n = even
in the ground state, all levels are doubly occupied spin-degenerate pairs
or else empty, and an electron with either sign of spin can tunnel via the
lowest-energy unoccupied orbital state. In an applied magnetic field, these
two spin states will exhibit Zeeman splitting, so that the first tunneling
level will split in two. On the other hand, if n = odd in the ground state,
only the lower spin state of the top orbital level is occupied, so that a
tunneling electron can enter this orbital state only in the higher-energy
spin orientation. Therefore the lowest-energy tunneling resonance peak will
not split, but will simply shift upward in voltage with increasing H. This
difference is illustrated in Fig. 8. As a result, there is an observable "evenodd effect" even in the normal state, provided the grain is small enough
460
to have well-resolved electronic energy levels. By adjusting gate voltage,
we have successfully swept samples from even to odd and then to even
again. We note that the explanation of this even-odd effect which we have
just stated assumes non-interacting electrons, but the result is in fact more
general. The different magnetic-field behaviors of the first tunneling state
for even- vs. odd-electron particles can also be viewed as a consequence of
Kramers' Theorem, even for a system of interacting electrons.
One subtlety should be mentioned. When electrons tunnel through a
nanoparticle biased with a source drain voltage larger than the discrete
level spacing (divided bye), it is energetically allowed that an electron may
enter the particle by tunneling through one junction into a high-energy
empty state, but then an electron may leave the particle by tunneling
through the second junction out of a lower-energy filled state. This leaves
an electron-hole excitation on the nanoparticle. If this excitation does not
relax before the next electron tunnels into the nanoparticle, the presence
of the excitation can shift the energy of the electronic state into which the
next electron can tunnel, because of electron-electron interactions within
the nanoparticle. If several different non-equilibrium electron configurations are energetically accessible within the particle, each may produce a
slightly different shift. The final time-integrated result is that the signal
from one single-electron quantum state in the particle may appear as a
cluster of different peaks in the differential conductance, with each peak
associated with a different non-equilibrium occupancy configuration of the
other single-electron states of the metal particle. The width of the cluster directly reflects the variance in electron-electron interaction energy between
the quantum-chaotic wavefunctions of electrons in the nanoparticle. Recently, this variance has been estimated numerically [27]. The results indicate that in very small particles, with diameters less than'" 5 nm, electronic
interactions are sufficiently strong that non-equilibrium effects may cause
an individual independent-electron energy level to be shifted about to produce many separate, well-resolved conductivity peaks. This non-equilibrium
effect therefore explains why conductivity peaks are sometimes observed in
clusters [22], rather than exhibiting energy-level repulsion as predicted for
independent-electron states by random matrix theory. The magnitude of
the predicted non-equilibrium shifts decreases with increasing particle size,
so that in larger particles (including the ones that exhibit superconductivity) the individual resonance peaks within a non-equilibrium-induced
cluster are unlikely to be resolved. Nevertheless, the non-equilibrium effect
may still be sufficiently strong to be the dominant source of resonance-peak
broadening at low temperature and large source-drain voltage.
In our initial experiments, the Al nanoparticles were small enough that
the discrete level spacing was far larger than the bulk gap of AI, so that
461
4000
';"
3000
>
:g
2000
"0
1000
0.5
1.0
Energy (meV)
Figure 9. Thnneling spectra at 4 different values of Vg , for a nanoparticle transistor

formed from a superconducting Al grain roughly 9 nm in diameter. The transitions are
labeled according to the number of electrons in the initial and final states, where no is
the number of electrons in the ground state at Vg = O. no is odd.
by Anderson's argument [1], one would expect the particles to be normal. Experimentally, we had no way to distinguish the possible presence
of superconductivity because we had no means to differentiate between
superconductivity-induced gaps between states and simple "electron-in-abox" level spacings. Therefore, more recently we have probed the properties
of somewhat larger grains, in which the level spacing 0 < ~o, and superconductivity was expected to be observable.
These experiments were made somewhat more difficult to interpret by
the fact that the required level separations of < 0.1 meV were also comparable with the observed resonance widths. Thus the spectra were no
longer completely resolved (see Fig. 9), and it was sometimes difficult to
track the H-dependence of individual levels when they were partially overlapping. This difficulty was compounded by the fact that, as a function of
magnetic field, the upward-trending branches of the spin-split levels lose intensity (for reasons poorly understood), leaving an apparent predominance
of downward-trending levels. To obtain the best data, it was necessary to
tune Vg so as to shift the conductance signals to low bias voltage, where
the resonances are best resolved.
Data for the magnetic field dependence of the resolved energy levels in
a representative sample are shown in Fig. 10. Advantage was taken of the
gate electrode to controllably tune the equilibrium number of electrons on
the island, so that we could compare the spectra observed when the equi-
462
(a)
1 .5
Sal
.s>-
1 .0
CI
Qj
c
- -------
0 .5
transitions from even n to odd n
0 .0
H (Tesla)
(b)
1 .5
Sal
.s>CI
Qj
c::
w
trans itions from odd n t o even n
0 .0
H (Tesla)
Figure 10. Magnetic field dependence of the energies of the resonant tunneling peaks
for the same sample as Fig. 9, for (a) Vg = 181 mV where n, the number of electrons in
the ground state of the particle, is even and (b) Vg = 110 mV where n is odd.
librium number was even vs. when it was odd. In both cases, the individual
levels split and trend up and down with increasing H with slopes corresponding to 9 ~ 2, just as with the very small grains. Unlike the spectra in
smaller particles, however, the levels in Fig. 10 display "gaps" near H =0,
which are much larger than the average energy level spacing. For the case of
an even number of electrons in the ground state of the particle (Fig. 10(a)),
the lowest-lying tunneling levels near H =0 have significantly greater energies than the levels at larger fields. The result for the odd-electron particle
is even more dramatic. Near H =0, one lowest energy level is split off from
all the others by a very large energy difference. This lowest level has an
energy near H =0 that is much lower than the threshold energy required
for tunneling at large H.
These gaps can be explained as a consequence of super conducting pairing within the Al nanoparticle. Consider first the case of an initially evenelectron particle (Fig. 10(a)). The ground state will be the fully Cooperpaired superconducting state. The addition or subtraction of a single elec-
463
tron to the particle will result in the creation of an unpaired quasiparticle,

with excess energy D.. The threshold for tunneling near H = 0 is therefore
increased by D., relative to the threshold at large H where superconducting correlations are suppressed. The numerical value of .60 in Fig. lO(a),
0.30 meV, is consistent with previous measurements in granular aluminum
[28]. For an initially odd-electron particle (Fig. 10(b)), the lowest-energy
tunneling process takes an initial state with an unpaired quasiparticle and
forms a fully-paired even-electron final state. The presence of superconducting correlations therefore aids this tunneling process, and allows it to
proceed at an energy threshold that is .60 less than the high-field threshold.
Tunneling via any even-electron final state other than the fully-paired state
requires the presence of at least two unpaired quasi particles. This explains
the gap between the first tunneling state and all the others for the initially
odd-electron particle. This large gap has a magnitude of approximately
2.60, though the measurement is reduced slightly from this nominal value,
perhaps by gap suppression due to the presence of the two quasiparticles.
The process by which superconducting correlations in a nanoparticle are
disrupted by an applied magnetic field is a subject of our current research.
In all of the samples studied to date, the energies of the discrete states
evolved in a continuous manner as a function of field. For example, the
lowest-energy state in Fig. 10(a) simply Zeeman-shifts with a g-factor of
approximately 2 for H up to 4 Tesla. At that point, that downward-trending
empty level crosses with an upward-moving (opposite spin) filled level, and
the ground state of the nanoparticle changes from spin 0 to spin 1 n. This
process is repeated again at higher fields, producing a continuous zig-zag
pattern in the energy of the tunneling threshold, as the ground state of the
particle increases its spin by one unit of 11, at a time. The curious thing about
this process is that it is continuous. Superconductivity is apparently being
destroyed one Cooper pair at a time, as single electrons in the ground state
of the nanoparticle are flipped individually. This is contrary to predictions
for the superconducting transitions in small particles, driven by spin pairbreaking in an applied field [29]. These predict an abrupt, discontinuous
jump in the tunneling threshold, at a field where many pairs are broken
simultaneously. We are currently investigating whether these theories do
not properly take into account the effect of discrete electronic energy levels
in the particle, or perhaps whether the experimental transitions are made
continuous by the contribution of orbital pair breaking [21].
In Fig. 11, we plot the values of .60 (filled dots) which we have measured
in a variety of superconducting Al particles, as a function of estimated particle size. The size estimate is made by determining the area of the larger
tunnel junction in the device from its measured capacitance and the known
capacitance per unit area of our oxidation process, and then assuming that
464
0.6
:>
.s
Q)
xl
I
0.4
XX
XI
>-
measured!1
X mean level spacing
Q)
Q)
0.2
0.0
0
10
5
radius (nm)
Figure 11. Dots: Measured values of the superconducting gap ~ in single Al nanoparticles as a function of estimated particle radius. Crosses: Measured mean discrete level
spacing. Solid lines: Predictions of von Delft et al. [2] for the dependence of ~, assuming
even or odd numbers of electrons.
the particle is roughly a hemisphere. The radii are accurate to within perhaps a factor of 2. The result to date is that we do not see any strong
dependence of the gap on particle size in the range where 8 < ~. Attempting to test Anderson's prediction for the elimination of superconductivity
when 8 > ~ , by extending these studies to smaller particles, is made difficult by the fact that we do not know of any way to distinguish the presence
of a superconducting gap from energy gaps due to simple electron-in-a-box
physics in this smaller size regime. (The measured values of 8 as a function
of particle size are plotted in Fig. 11 as crosses.) Theoretically, some of our
devices are in the size range where ~ is expected to be different for even
versus odd numbers of electrons [2, 19]. The solid lines in Fig. 11 depict
the predicted dependence of ~even and ~odd on sample size for a particle
with evenly-spaced discrete levels [2]. We do not yet know how to make
experimental contact to these predictions, either. The energy levels that we
measure by tunneling correspond to the difference in energy between an
even-electron state and an odd-electron state. Both ~even and ~odd therefore contribute to each of our measured levels, and we do not currently
know how to separate these contributions.
5. Concluding Discussions
The experiments that we have described demonstrate that electron tunneling is a powerful and sensitive probe into the physics of small superconducting samples. Even in relatively large, micron-scale samples, containing on
the order of 109 conduction electrons, tunneling measurements have been
able to measure in detail the differences between even and odd numbers
of electrons that are a consequence of superconducting pairing. The nature
465
of the superconducting state appears to undergo little qualitative change

as sample dimensions shrink from the micron scale to the few nm scale.
Even though the sample diameter crosses from being larger than both the
superconducting coherence length and magnetic field penetration depth to
being much smaller, a Cooper-paired superconducting ground state is still
possible as long as the discrete level spacing, 8, is smaller than the superconducting gap, .6... What consequences pairing interactions may have
in still smaller samples, where 8 is comparable to and larger than .6.., is
still an open question, requiring further experimental work and theoretical insights. In particular, it is unclear in principle whether there are any
techniques by which one can take an experimentally-measured discrete electronic spectrum and separate out the contribution of superconducting interactions from simple independent-electron-in-a-box physics, when 8 > .6...
New techniques to distinguish the effects of .6.. even and .6.. odd in nm-scale superconducting particles would also be welcome, as would an understanding
of the nature of the superconducting to normal transition in a nanoparticle
as a function of applied magnetic field.
Are there any other experimental techniques which can aid the effort
to understand the nature of the electronic state in ultrasmall superconductors? Measurements of the diamagnetism of single nm-scale particles,
perhaps with a scanning SQUID or a scanning Hall bar, would likely provide
useful insights. (The existence of superconducting diamagnetism should be
unaffected by the fixed particle number constraint, since supercurrents depend only on phase gradients, not the overall phase which is conjugate to
the fixed particle number.) However, accurate measurements of the diamagnetism in single nm-scale particles will be an experimental challenge,
because the signals will be tiny, at best. For particles in the critical size
range where 8 '" .6.., the total diamagnetic moment of a grain due to superconductivity can be estimated to be ::; l/-LB even in a field of a few Tesla.
The resulting magnetization has the same scale as the orbital magnetization of a normal-metal particle, and also the spin paramagnetism of an
odd-electron particle.
As a final remark, we note that our new ability to measure in detail
the spectrum of energy levels inside a nm-scale sample is a very powerful
technique for understanding all the forces which act on the electrons inside
the metal. We have shown how superconducting pairing interactions, spinorbit scattering, Zeeman splitting, and direct electron-electron interactions
can all be studied quantitatively by observing the separate changes that
they induce in the level spectra. In general, any force or interaction which
influences electronic structure in a metal can be studied, from a most fundamental point of view, by examining the effects on the discrete electronic
levels. We look forward to using this technique to investigate a broad range
466
of interacting-electron systems, including other superconductors, magnetic
materials, and the physics of quantum impurities in metals.
Acknowledgments - This research was supported by NSF Grant No.
DMR-92-07956, ONR Grant No. N00014-96-1-0108, JSEP Grant No. N0001489-J-1023, and ONR AASERT Grant No. N00014-94-1-0808, and was performed in part at the Cornell Nanofabrication Facility, funded by the NSF
(Grant No. ECS-9319005), Cornell University, and industrial affiliates.
References
1.
2.
P. W. Anderson, J. Phys. Chern. Solids 11, 28 (1959).

J. von Delft, A. D. Zaikin, D. S. Golubev, and W. Tichy, Phys. Rev. Lett. 77, 3189
(1996).
3. D. V. Averin and Yu. V. Nazarov, Phys. Rev. Lett. 69, 1993 (1992).
4. M. T. Tuominen, J. M. Hergenrother, T. S. Tighe, and M. Tinkham, Phys. Rev. Lett.
69, 1997 (1992).
5. P. Lafarge, P. Joyez, D. Esteve, C. Urbina, and M. H. Devoret, Phys. Rev. Lett. 70,
994 (1993).
6. T. M. Eiles, J. M Martinis, and M. H. Devoret, Phys. Rev. Lett. 70, 1862 (1993).
7. K. A. Matveev, M. Gisselfalt, L. I. Glazman, M. Jonson, and R. I. Shekhter, Phys.
Rev. Lett. 70, 2940 (1993).
8. J. M. Hergenrother, M. T. Tuominen, and M. Tinkham, Phys. Rev. Lett. 72,1742
(1994).
9. P. Joyez, P. Lafarge, A Filipe, D. Esteve, and M. H. Devoret, Phys. Rev. Lett. 72,
2458 (1994).
10. A. Amar, D. Song, C. J. Lobb, and F. C. Wellstood, Phys. Rev. Lett. 72, 3234
(1994).
11. D. B. Haviland, Y. Harada, P. Delsing, C. D. Chen, and T. Cleason, Phys. Rev.
Lett. 73, 1541 (1994).
12. W. J. Elion, M. Matters, U. Geigenmuller, and J. E. Mooij, Nature 371,594 (1994).
13. D. V. Averin and K. K. Likharev, in Mesoscopic Phenomena in Solids, eds. B. L.
Altshuler, P. A. Lee, and R. A. Webb (Elsevier, New York, 1991) p. 169.
14. M. Tinkham, J. M. Hergenrother, and J. G. Lu, Phys. Rev. B 51, 12649 (1995).
15. J. G. Lu, J. M. Hergenrother, and M. Tinkham, Phys. Rev. B 53,3543 (1996).
16. M. T. Tuominen, J. M. Hergenrother, T. S. Tighe, and M. Tinkham, Phys. Rev. B
47, 11599 (1993).
17. J. M. Hergenrother, J. G. Lu, M. T. Tuominen, D. C. Ralph, and M. Tinkham,
Phys. Rev. B 51, 9407 (1995).
18. See, for example, M. A. Kastner, Phys. Today 46, No.1, 24 (1993).
19. R. A. Smith and V. Ambegaokar, Phys. Rev. Lett. 77, 4962 (1996).
20. D. C. Ralph, C. T. Black, and M. Tinkham, Phys. Rev. Lett. 74, 3241 (1995).
21. C. T. Black, D. C. Ralph, and M. Tinkham, Phys. Rev. Lett. 76,688 (1996).
22. D. C. Ralph, C. T. Black, and M. Tinkham, Physica B 218, 258 (1996).
23. Some of the data shown in this chaper was taken with devices made without gate
electrodes.
24. A. E. Hanna and M. Tinkham, Phys. Rev. B 44, 5919 (1991).
25. S. Tarucha, D. G. Austing, T. Honda, R. J. van der Hage, and L. P. Kouwenhoven,
Phys. Rev. Lett. 77, 3613 (1996).
26. Regardless of whether the Al nanoparticle is superconducting or normal, a field of
300 Gauss should not produce a significant shift in the energy of the discrete electronic
states within the nanoparticle. The only effect of the field is on the superconductivity
in the electrodes. This is true because the nanoparticle is much smaller than the
467
magnetic-field penetration depth in AI, so that any field-induced superconducting
transition within the particle should occur only at fields of several Tesla.
27. O. Agam, N. S. Wingreen, B. L. Altshuler, D. C. Ralph, and M. Tinkham, condmat/9611115.
28. R. W. Cohen and B. Abeles, Phys. Rev. 168, 444 (1968).
29. A. M. Clogston, Phys. Rev. Lett. 9, 266 (1962); B. S. Chandrasekhar, Appl. Phys.
Lett. I, 7 (1962).
THE SUPERCONDUCTING PROXIMITY EFFECT

IN SEMICONDUCTOR-SUPERCONDUCTOR SYSTEMS:
BALLISTIC TRANSPORT, LOW DIMENSIONALITY
AND SAMPLE SPECIFIC PROPERTIES
BART J. VAN WEES

Department of Applied Physics and Materials Science Centre
University of Groningen
Nijenborgh 4, Groningen, The Netherlands
AND
HIDEAKI TAKAYANAGI
Basic Research Laboratories
Nippon Telegraph and Telephone Corporation
3-1, Morinosato- Wakamiya
Atsugi-shi, Kanagawa, 243-01 Japan
1. INTRODUCTION
The superconducting proximity effect describes the interaction between a

superconductor and the electronic properties of a normal conductor. Its
manifestations for transport are twofold: First, the presence of superconducting electrodes can modify the normal (dissipative) transport. Second,
the presence of two (or more) superconductors can allow a dissipationless
supercurrent to flow through a normal conductor. In both cases coherent
Andreev reflection is the key underlying phenomenon. As we will show in
this chapter, its effect on the transport is directly linked to the scattering
properties of the normal region.
We will review both aspects of the superconducting proximity effect for
systems based on superconductors coupled to semiconductors in general,
and the two-dimensional electron gas (2DEG) present in these in particular. The emphasis will be on those aspects which distinguish degenerately
doped, or "metallic", semiconductors from their true metal counterparts.
The following aspects can be mentioned:
469
470
* Semiconductors
allow the study of the proximity effect in systems

of low-dimensionality. Starting from a 2DEG, one-dimensional wires and
quantum point contacts (QPC's) can be studied, and one can proceed to
zero-dimensional systems, e.g. with the use of quantum dots.
* High mobility 2DEG systems offer a unique way to study the ballistic
aspects of Andreev reflection. Here (quantum) point contacts can be used
as injectors, as well as detectors of Andreev reflected particles.
* Due to the relatively large Fermi wavelength AF, the number of quantum channels can be made low, as a result of which sample specific behaviour can be relatively strong. This should be compared to metal systems
where the effect of superconductors on the sample specific properties is almost always dominated by the much stronger ensemble averaged behaviour
of the proximity effect.
We will base our discussion on the scattering description of electron
transport, as formalized by the Landauer-Biittiker description. We will restrict us to a basic overview. For more elaborate discussions of the various
aspects of the proximity effect we refer to the reviews [1, 2, 3, 4].
2. ANDREEV REFLECTION
The Bogolyubov-de Gennes equations [5] form the basis for the description of quantum transport in normal/superconductor (N/S) systems. They
describe the coupled evolution of the electron 'l/;e(r) and hole 'l/;h(r) wavefunctions at an energy E relative to the Fermi energy EF:
Here H = (p-eA)2/2m-eV(r) is the single particle Hamiltonian, and the

superconducting pairpotential a = tlexp( i</, with tl the superconducting
gap, and </> the superconducting phase.
The elementary scattering problem to be considered is shown in fig. 1. A
voltage V is applied between the normal reservoir N and the superconductor
S. For eV < 0 electrons are injected by the normal reservoir, for eV > 0
holes. Here we focus on an injected electron. Its wavefunction has to be
matched to that of a reflected electron, a reflected hole, and transmitted
electron-like and hole-like quasiparticles in the superconductor.
Throughout this chapter we will be mostly interested in energies E below tl. For these energies the electron-like and hole-like quasiparticle wave
functions decay inside the superconductor. At E = 0 this happens on the
scale of the superconducting coherence length ~o = nVF/7rtl. Throughout
we will assume that the superconductor is much longer than ~o, so that
quasiparticle transmission through the superconductor is excluded.
471
N
x=O
Figure 1. Reflection and transmission of an incoming electron by a step in the pair

potential t::. at an N/S interface. The electron wave function has to be matched to an
Andreev reflected hole wave function with amplitude a, a reflected electron wavefunction
(b), and transmitted electron-like (c), and hole-like (d) wavefunctions. The latter decay
on a length scale k;l =
eo
Matching of the wavefunctions and their derivatives at the N/S interface yields for the amplitude of the normal reflection: b = 0, and for the
amplitude of Andreev reflection[6]:
aeh(E) = -exp(i</exp(iarccos(E/il)) for e -+ h

ahe(E) = -exp(-i</exp(iarccos(E/il)) for h-+e
(2)
Apart from a shift </> due to the superconducting phase, which is opposite for both AR processes, the Andreev reflected wave function acquires an
additional energy dependent phase shift. This phase is -11"/2 at the Fermi
energy, which turns out to be important later when multiple coherent Andreev reflections have to be taken into account.
During the Andreev reflection process the particle momentum p=1ik is
conserved. Classically this results in retroreflection, where the hole is reflected back in the same direction as the incoming electron. Retroreflection
has been studied in metal systems[7], where electrons were injected with a
point contact in a normal metal/superconductor bilayer, and retroreflected
holes returned to the point contact, thus reducing its resistance. With two
adjacent point contacts the deflection of the retroreflected holes by a magnetic field was studied in transverse focussing experiments[8].
Phase conjugation is the quantum aspect associated with retroreflection.
At the Fermi energy the phase f k:dl acquired by an electron on its way to
the superconductor is exactly cancelled by the phase f k~dl of the retroreflected hole, independent of the total path length L. In diffusive conductors
472
(a)
(b)
ri
s
p- lnAs
Figure 2. Experimental systems. a) Nb/Si/Nb system, showing superconducting contacts S fabricated on either side of a thin p++ doped Si membrane. b) Superconducting
contacts fabricated to an AIGaAs/GaAs heterostructure by alloying superconducting
contacts at elevated temperatures. c) Superconducting contacts to InAs based quantum
wells. d) Superconducting contacts on an inversion layer present at the InAs surface. The
dashed regions in b, c, and d indicate the location of the 2DEG
this phase conjugation property is destroyed on the scale ofthe Thouless energy ET (= TiD / L2), in which L is the length ofthe conductor, and D is the
diffusion constant. In semiconductors the relatively long Fermi wavelength
makes it possible to investigate phase conjugation in ballistic systems. Here
phase conjugation is preserved on an energy scale Ee = TivF/ L
3. EXPERIMENTAL SYSTEMS
Compared to metallic systems, semiconductors are more flexible in the sense
that their carrier concentrations can be controlled by doping, or by electrostatic control with gate electrodes. The major complication for the study of
the proximity effect in semiconductor/superconductor systems arises from
the formation of a Schottky barrier at the N/S interface. To obtain an
sizeable probability for Andreev reflection requires that the interface transmission T has to be of the order of unity (see section 4.1). The role of
the Schottky barrier can be reduced by heavily doping the semiconductor
close to the N/S interface. In this way the width of the depleted region is
reduced, and tunneling through the Schottky barrier is made possible. Fig.
2a shows a p++ doped Si membrane, which has been employed to study
super current and quasiparticle transport[9]. Also lateral systems based on
superconducting contacts to heavily doped semiconductors layers have been
studied[10, 11, 12, 13].
Due to its high electron mobility, as well as the ease of defining struc-
473
tures with gate electrodes, the GaAsl AIGaAs 2DEG system presents itself
as an obvious candidate. Here the Schottky barrier problem was tackled by
fabricating contacts by alloying low-melting point superconductors like Sn
or In at elevated temperatures[14, 15, 16, 17]. Results on Andreev reflection[16], and as well as supercurrents[17], have been reported. However, the
interpretation of the experiments is hindered by the lack of understanding
of the contact morphology and the properties of the various alloys (some
of which can become superconducting), which are formed in the alloying
process.
The most intensively studied systems are based on the III/V compound
InAs. Due to the Fermi energy pinning in the conduction band, InAs and
InGaxAs 1 - x, with x < 0.2, do not form Schottky barriers when brought in
contact with a metal electrode. Bulk InAs based systems have been studied,
as well as 2DEGs present in quantum well structures[18, 19, 20] (fig. 2c),
and surface inversion layers[21](fig. 2d). We will focus our discussion on the
latter two systems.
4. TWO-TERMINAL NORMAL-SUPERCONDUCTOR STRUCTURES

The basic geometry to study experimentally is a superconductor S coupled
to a normal reservoir N. A scattering region (with length L, and width
W) may be present in between. Coherent transport through this region
can be described by a ( complex) scattering matrix Se( E) for electrons,
and a corresponding one Sh(E) for holes. Phase conjugation requires that
Se(E) = S'h(-E). At the Fermi energy therefore Se = S'h. This property
allows to express the Andreev conductance as[22, 23]:
(3)
Here the sum n is over the eigenvalues Tn of the transmission matrix ttt of
the scattering region. Similar to the Landauer formula G N = 2e 2 I h Ln Tn
for the normal conductance, this formula is a linear statistic on the transmission eigenvalues. This implies that once the distribution of the transmission eigenvalues is known, the Andreev conductance can be obtained
directly. We note that eq. 3 in principle holds at zero voltage, zero magnetic
field, and zero temperature. This effectively restricts its validity to magnetic fields B < <poIW L, with <1>0 = hl2e, energies E < ET (= TiD IL2),
and temperatures kT < ET. We will give three examples of this formula
applied to experimental systems: a tunnel barrier, a diffusive conductor,
and a ballistic point contact.
474
Z0.3
Z'O
1
0.5
0.5
..................~.
ZolO
"'~""
c
"8--'
Z '3.0
i' .....
0.5
c
E
61T)
Figure 9. BTK result for a tunnel barrier, for several values of the barrier strength
parameter Z. A gives the probability for Andreev reflection, B the probability for normal
reflection, C the probability for transmission as a electron-like quasiparticle, and D the
probability for transmission as a hole-like quasiparticle (from ref. 24)
4.1. TUNNEL BARRIER
In the scattering formalism a tunnel barrier is represented by an electrostatic potential in the form of delta function Vo6(x), located at the N/S
interface. In the normal state the transmission probability through the barrier is given by T = 1/(1+Z 2 ), with Z = Vo/TivF1.' Although T depends on
VF 1., the velocity normal to the interface, and therefore on the mode index
n, it can for all practical practical purposes be assumed to be mode independent. Eq. 3 then gives for the Andreev conductance of a tunnel barrier
GA = 4e 2/hNT2/(2 - T)2. This expression covers the whole range from
complete transparency T = 1, to the tunnel junction limit T < 1, where
the Andreev conductance is proportional to T2.
The energy dependence of Andreev and normal reflection and transmission of a tunnel barrier has been obtained by Blonder, Tinkham and Klapwijk(BTK)[24]. This was done by taking into account the energy dependent
phase shift due to AR, given in eq. 2. Due to the gradual destruction of the
phase conjugation property, the Andreev reflection probability slowly increases with energy. Interestingly, 100% probability for Andreev reflection
is obtained at E = ~, irrespective of T (see fig.3).
Numerous experiments have shown the validity of these results, and we
475
'-' 1
O'-"'&:"'--'---'---'--~--L-L---'-------'
Figure 4. Comparison between the Andreev conductance G A of a quantum point contact

(solid line) and twice its normal state conductance GN (dotted line). In between the
plateaux the Andreev conductance is less than 2GN (from ref.22)
will not discuss them here. We conclude with a brief remark about the
physical requirements for a tunnel barrier. In order to have a transmission
T independent of energy, the barrier should be located close to the interface.
Bagwell et al.[25] have calculated the conductance of an N/I/N /S system,
where the tunnel barrier I is located a distance d away from the interface.
Due to the formation of bound electron-hole states, a gradual crossover
was obtained from tunnel junction behaviour (for d < ~o) to Ohmic linear
behaviour (for d > ~o), when the tunnel barrier was moved away from the
interface.
4.2. BALLISTIC POINT CONTACTS
A ballistic point contact (PC) is characterized by its transmission eigenvalues Tn being either 0 or 1. These are not affected by the presence of a
wide region in between the PC and the superconductor, provided that no
backscattering occurs in this region. Eq. 3 then predicts that GA = 2G N .
Note that this conductance doubling is slightly more intricate than that
expected on classical grounds, as a result of retrorefiection of the Andreev
holes back through the PC. This is shown in fig. 4 which shows that the
conductance doubling is only exact at the quantized plateaux, where Tn is
either 0 or 1[22].
The experimental observation ofthe conductance doubling requires that,
in addition to ballistic transport, the interface between the superconductor and normal conductor is free of disorder as well. This is usually not
476
8 .-----------------------~--~
O~~~L_~~L_~~~~~L_~~
- 1.8
-1.7
-1.6
v~
- 1.5
- 1.4
- lJ
(V)
Figure 5. Measured Andreev conductance of a quantum point contact in between superconductors. The conductance steps are enhanced above 2e 2 /h (from ref.27)
the case however[26]. Striking results on a possible experimental realization have been reported by Lenssen et al.[16]. Here we show data (fig.5)
from Takayanagi et al. [27]. This experiment shows that the conductance
steps are enhanced above 2e 2 / h. The discrepancy with the expected 4e 2 / h
is most likely due to the presence of disorder at the 2DEG/superconductor
interfaces, as well as residual scattering in or near the point contact.
4.3. DIFFUSIVE CONDUCTORS
A diffusive conductor can be characterized by a conductivity a which can be

expressed in terms of a local diffusion coefficient D. The proximity effect in
a diffusive system can be described by ensemble averaged Green's functions
methods, in terms the Usadel equations. In this description the proximity
effect is accounted for by an (energy dependent) renormalization of the
diffusion coefficient D(E).
Here we discuss its Andreev conductance from a scattering point of view.
The transmission eigenvalue distribution peT) of a diffusive wire in the
metallic regime (L ~ W, G ~ e2 / h, number of transverse modes N ~ 1)
was found to be universal[28]:
1
p(T) -- GoT~'
. h G0 = hi 2e.
2
WIt
(4)
It has a typical bimodal shape, with a large density peT) for high transmission (open channels) and low transmission (closed channels). When
477
placed against a superconductor, one finds with eq. 3 that GA = GN. This
result is made possible by the fact that Andreev reflection enhances the
conductance of the open channels whereas the conductance of the closed
channels is reduced. The transmission eigen value distribution for a diffusive wire given by eq. 4 is such that its (zero energy) conductance is not
affected by the presence of a superconductor[29].
Note that this is not fully correct. First, in the regime L 2': I the resistance is made up by the diffusive resistance of the wire in series with a
Sharvin contact resistance h/2e 2 N. The resistance of the latter is affected
by the presence of the superconductor (see sections 4.2 and 4.6). Also quantum corrections to the conductance such as weak localization and universal
conductance fluctuations are altered by the presence of the superconductors
(see section 6.3).
4.4. REFLECTIONLESS TUNNELING
In the sections above we have shown how a superconductor modifies the

transport properties of three basic conducting elements: a tunnel barrier, a
ballistic point contact, and a diffusive conductor. One can now also study
the Andreev conductance of composite systems, where the normal region
contains different elements, connected in series. In this respect the phenomenon of reflectionless tunneling is one of the most striking discoveries in
relation to phase coherent AR. Kastalsky et al.[30] reported experiments on
N-I-S junctions based on Nb contacts to a (doped) InGaAs semiconductor
layer. A suppression of the differential conductance for voltages below b../ e
was observed, typical of N-I-S systems where a tunnel barrier is present.
Surprisingly however, it was found that the differential conductance rose
again at low bias voltages. Similar increases of the zero-bias conductance
were subsequently observed by others, both in semiconductor[ll, 31, 32],
as well as in metal systems[33]. Here we show experimental results from
Magnee et al.[31]' who studied normal metal-Si-Nb junctions. Despite of
the heavily p++ -doped Si membrane, a residual Schottky barrier was still
present, and the system could be described as an N-I-S system, where N is
a disordered normal region. The experiments show two characterics of the
zero- bias anomaly, namely its destruction by a finite voltage (or temperature) and by a magnetic field (fig. 6).
Both features suggested that quantum interference should be the underlying mechanism. The crucial role of multiple phase coherent Andreev
reflections was recognized by Van Wees et al.[34]' who described the transport through a tunnel barrier in series with a disordered region with a
semi-classical transport model, taking into account the quantum phase of
the electrons and holes. They found that due to the phase conjugation of
478
1.3,..,.............,..........,....,..,..........,....,..,...,..............,...,...............,...,....,~
B (mT)
1.25
~a
>
:-g
=0 1.15
1.1
-3
-2
-1
0
V (mV)
Figure 6. Destruction of the zero-bias conductance anomaly in a W jIjNb junction at

T=100 mK by an applied magnetic field and voltage (from ref.31).
electrons and holes at the Fermi energy, multiple Andreev reflections enhance the probability for the incoming electron to be Andreev reflected.
This makes it possible to understand a surprising outcome of this description, namely that the Andreev conductance can be enhanced by the
addition of scatterers in front of the tunnel barrier. It was found that G A
reached a maximum when the resistance of the diffusive region was approximately equal to that of the tunnel barrier (see fig. 7). The resistance of
the composite system is then about equal to that of the disordered region.
In this way the tunnel barrier resistance is suppressed, hence the name
reflectionless tunneling.
Beenakker et al.[35] showed that this surprising result could be under-
stood as a result of the broadening of the sharply peaked distribution of

transmission eigenvalues Tn ~ 1, corresponding to a tunneljunction, toa
distribution with finite density for Tn ~ 1. It is this generation of open quantum channels which accounts for the increase in the Andreev conductance.
Alternatively the reflectionless tunneling has been studied with numerical
techniques[36]. Analytical quasi-classical techniques have proven to be quite
fruitful for the calculation of transport in disordered systems[37], and also
predict an enhanced zero-bias conductance. These were the building blocks
of Nazarov's circuit theory for Andreev conductance[38, 39]. Here reflectionless tunneling emerges from the renormalization of the tunnel junction
conductance (section 4.6). A detailed description of the circuit theory is
given elsewhere in this book.
A further refinement can be obtained by studying the Andreev conduc-
479
5rn,-----.--------r------~
......... 4
..c
~
~3
z
C\l
........ 2
x
0::
N~S
O~------L-------~------~3
rL/1
Figure 7. Dependence of the resistance RN s on the length L of the disordered normal
region (see inset), for different values of the transmittance r of the NS interface. Solid
curves are for r=l, 0.8, 0.6, 0.4, 0.1 from bottom to top. For r <. 1 the dashed curve of
ref.37 is approached (from ref.3S)
tance of two tunnel barriers in series. It was predicted that a maximum

conductance is obtained when the two tunnel barriers have equal conductances[40]. Preliminary experimental results[41] have been obtained by creating artificial tunable tunnel barriers with narrow gate electrodes. They
reveal a resistance minimum when the tunnel resistances are approximately
equal.
4.5. "GIANT" ANDREEV REFLECTION
A second composite circuit to be analysed is a ballistic point contact in

series with a diffusive conductor. In the case of a ballistic region, with a
length L smaller than the mean free path 1 in between the PC and the
superconductor, the conductance is doubled because of the transmission
eigenvalues being either zero or one (section 4.2). A surprising result was
obtained which revealed that the enhancement of the PC conductance persisted, even when L > l, and the transport in between the PC and the
superconductor is diffusive[42]. This result was attributed to diagonal Andreev reflection, which was shown to be enhanced relative to off-diagonal
AR, even in the case L > l. Fig. 8 shows that the enhancement of the
Andreev conductance is a function of the parameter NoL/Nl, where No
and N are the number of modes in the point contact and disordered region
respectively. In the next section we will apply Nazarov's circuit theory to
this system.
Experimental observation ofthis "giant" Andreev reflection is problematic, since it requires the combination of a disordered region with a ballistic
480
1
o
t-'o
............
t-'
<l
L---
Figure 8. Excess Andreev conductance t::.G of a ballistic point contacts in series with
a diffusive region. No, N indicate the number of quantum channels in the PC and wide
region respectively, Go = 2e 2 /h (from ref. 42).
point contact. Heslinga et al.[43], studied AR with a metal point contact

on a diffusive Si membrane with a superconducting counter electrode. Despite of L > > I, they observed a large enhancement of the conductance for
E < ~, which may well be a manifestation of the physics discussed here.
We will give a further experimental example in section 6.2.
4.6. COMPARISON WITH CIRCUIT THEORY
Based on quasiclassical impurity averaged Green's function techniques,

Nazarov has developed a circuit theory for Andreev conductance[38]. In
this theory the various conducting elements: tunnel barriers, diffusive conductors and ballistic point contacts, are treated as two-terminal circuit
elements, connected at nodes. The nodes are characterized by spectral vectors, defined on a unit hemisphere. For a given circuit the spectral vectors
can be determined from the condition of conservation of spectral current
at the nodes. After determination of the spectral vectors, the renormalized
Andreev conductances of the circuit can be calculated, since the Andreev
conductance of the each element is a function of the spectral vectors on
either side of the element.
The circuit theory can be extended to finite energy[39, 44]. Here we
restrict ourselves here to the zero energy case, applied to a single normal
reservoir and a single superconductor. In this case the spectral vector can
be parametrized with an angle variable 0, which is a real scalar, which describes the angle between the spectral vector and the north pole on the
unit hemisphere. The boundary conditions are that the spectral vector of
a normal reservoir points towards the north pole (0 = 0), and that of the
superconductor is in the plane ofthe equator (0 = 7r /2). In Table 1 we sum-
481
marize the expressions for the spectral current and Andreev conductances
as a function of difference f).(} across the elements (The general case is from
ref.[44]). By evaluating the renormalized conductances for f).(} = 7r 12, the
TABLE 1. Circuit theory parameters
circuit element
spectral current I(~O)
Andreev conductance G A
diffusive conductor
tunnel junction
ballistic point contact
GN~O
GN
GNsin(~O)
GNcOS(~O)
2GNtan(~O/2)
general case
G '\'
GN/cos2(~O/2)
G '\' 1', cOS(.c.0)+Tnsin2(.c.0~2)
Tnsin~.c.O)
N i...J n l-Tn sin (.c.0/2)
N i...J n
(l-Tnsin2(.c.0/2))
circuit theory reproduces the results obtained with the scattering formalism: The conductance of a diffusive conductor is not affected by a superconductor, the conductance of a tunnel junction is suppressed, and the ballistic
point contact conductance is doubled.
For a circuit consisting of a diffusive region in series with a tunnel
junction the reflectionless tunneling of section 4.4 can be derived. Here we
will illustrate the case of a PC in series with a diffusive conductor, with
normal conductances Gp and G D respectively. The spectral vector () at the
node can be obtained by the conservation of spectral current:
7r
2 Gp tan ((}/2) = G D ("2 - ())
(5)
From the obtained value of (J the total renormalized Andreev resistance can
then be calculated:
(6)
The Andreev resistance is reduced below its normal state value, given by
RN = l/Gp + l/GD. The functional dependence of this excess Andreev
conductance corresponds to that shown in fig. 8.
5. MULTITERMINAL SCATTERING DESCRIPTION OF COHERENT ANDREEV REFLECTION

The above deals with two-terminal systems where a single normal reservoir
is coupled to a single superconductor. Due to its straightforward recipe,
the circuit theory of section 4.6 lends itself for a description of more complex geometries, including multiple normal and superconducting contacts.
However, being based on impurity averaging techniques, a full description
including ballistic and sample specific properties is beyond its scope.
482
U2
A
-~"-~-'_~-:---""1
I~l.
\
I
'i \
~ L
;L
____.~
""'~'~'. irs.:'.' -'
1:.1
n
B
'..
.-
-~
..
- O~t;lt
"".
,'1,:"
.! ; .
'J
.: ' - I~. 3~
Figure 9.
(a) (a) Geometry for multiterminal measurements in an Andreev interferometer, showing a cross-wire structure connected to normal contacts 1,2,3, and 4. The
superconducting contacts 0 form the ends of a superconducting loop. (b) Scanning electron micrograph of the experimental structure. The white bar indicates 1 micrometer
(from ref.53).
The extension of the multiterminal scattering description (LandauerBiittiker formalism[45]) to include superconductors is relatively straightforward[46]. We will restrict ourselves to the static case, where all superconductors have the same electrochemical potential J.lo. Allowing the superconductors to have different electrochemical potentials requires a time
dependent formulation, to account for multiple phase coherent Andreev reflections at different energies[47]. Another complication arises due to the
possibility for supercurrent flow between pairs of superconducting contacts. These supercurrents can also be dealt with using a scattering formalism[46](see section 7.4). However, in contrast to the normal (dissipative)
transport, which can be expressed in transport coefficients at, or close to
EF, the supercurrent is determined by the states in an energy range L\
around EF. Therefore one generally does not expect supercurrent and dissipative transport to be directly related. Also, from an experimental point
of view, the superconducting electrodes in the experiments are usually coupled by a superconducting loop, so that only the net current leaving or
entering the superconductor is relevant. A possible supercurrent only gives
rise to a circulating current, and therefore does not affect the transport
measurements[ 48].
We describe the general case for N normal contacts and M superconducting contacts with the same electrochemical potential. We take the electrochemical potential of the superconductors J.lo = O. The following transport
coefficients are needed: Rii the (generalized) probability for an electron
entering lead i to be scattered back as an electron, R~e(= Tr(rherte)): the
probability for an electron entering lead i to be scattered back into lead
i as a hole due to Andreev reflection, Ttl, and Ti e : the probabilities for
483
normal and Andreev transmission from lead j into lead i respectively. In

principle another four coefficients are required to describe the reflection and
transmission of injected holes. However, due to the electron-hole symmetry at or near the Fermi energy, we can put them equal to their electron
counterparts.
The (charge) current flowing in lead i can now be written as:
Ii =
2: ((Ni - Rii + R~e)/1i - ~(Tt/ - Ti}e)/1j,
(7)
where Ni indicates the number of quantum channels in lead i. Particle

conservation imposes the conditions:
Rii
+ R~e + 2)TJt + TN) = Ni
(8)
We can now apply these equations to obtain the expressions for the various
multiterminal resistances of fig. 9. For convenience we define Rii = Rit R~e , and Tij = Ttl - Ti}e. We obtain the following expressions:
_ h
N - RBB
2e2 N - RAA)(N - RBB) _ TABTBA
(9)
h
N - RAA
2e2 (N - RAA)(N - RBB) - TABTBA
(10)
10,20 -
_
30,40 -
h
TBA
RIO 30 = ,
2e 2 (N - RAA)(N - RBB) - TABTBA
h
(11)
TAB
= (12)
2e 2 (N - RAA)(N - RBB) - TABTBA
,
Here N is the number of quantum channels in the wires. From the timereversal symmetry of the transport coefficients one can obtain [46]:
R30 10
(13)
These are the generalized reciprocity relations for phase coherent transport
in normal-superconductor structures. Changing the sign of magnetic field
and superconducting phase, accompanied by the interchange of current and
voltage leads leaves the four-terminal resistance unchanged.
6. PHASE DEPENDENT TRANSPORT IN ANDREEV INTERFEROMETERS

In an Andreev reflection process the phase of the superconductor is communicated to the reflected particle as a phase shift in the particle wavefunction. This implies that when an electron wave is split in two partial
484
2400
2300
:~w
-300
- 200
- 100
B (Gauss)
100
200
300
Figure 10. Multiterminal resistance measurements on a diffusive interferometer of fig.

9. a) and b) show oscillations of the "two-terminal" resistances R lO ,20 and R 30 ,40. c) and
d) show oscillations of the "three-terminal" resistances R 10 ,30 and R 30 ,lO (from ref.53).
waves, which are reflected from two superconductors with two different
phases, the Andreev reflected hole waves can interfere, and the Andreev
conductance will be a 211" periodic function of the superconducting phase
difference D..<jJ. This principle was exploited by Nakano and Takayanagi[49],
who theoretically studied the phase dependence of a one-dimensional wire
connected to two superconductors via a V-splitter.
Here we will illustrate two types interferometers, one based on a diffusive
transport, the other on ballistic transport. These interferometers allow the
study of the phase dependence of both ensemble averaged as well as sample
specific conductance.
6.1. PHASE DEPENDENT DIFFUSIVE TRANSPORT
Several experiments have been performed on the phase dependent conductance of diffusive conductors, both in metals[33, 50] as well as semi conductors[51, 52, 53, 54]. Fig. 9 shows a SEM micrograph of the geometry
employed by den Hartog et al.[53]. The superconducting phase difference
D..<jJ is changed by applying a perpendicular magnetic field to an interrupted
superconducting loop, with an area A. It is given by D..<jJ = 211" if! / if! 0, with
if! = BA
Fig. 10 shows a comparison between the various measured multiterminal resistances. Due to the relatively low number of quantum channels
485
Figure 11. Left panel: SEM micrograph of a ballistic PC connected to a cavity. Right
panel: Three different sets of phase conjugated trajectories. The two superconductors
(with phases 1/>1 and 1/>2) form the ends of a superconducting loop (not shown) (from
ref. 54)
(N ~ 35), the data shows a superposition of ensemble averaged and sample specific behaviour. As predicted from eq.13, the resistances which are
effectively "two- terminal"(fig.10a,10b) are symmetric under reversal of B.
A pair of "three-terminal" resistances (fig.10c,10d) is symmetric under reversal of B, provided that current and voltage leads are interchanged, in
agreement with eq. 13.
We do not discuss the ensemble averaged behaviour here. The modulation of the sample specific behaviour by the superconducting phase will be
discussed in section 6.3.
6.2. BALLISTIC ANDREEV INTERFEROMETER
We discuss experimental results from Morpurgo et al.[54], who studied AR

in a ballistic point contact connected to a ballistic 2DEG cavity (fig. 11).
Electrons are injected by a ballistic PC, can be Andreev reflected by superconductor 1 and 2, and return through the PC, thus lowering its resistance.
This experiment therefore allows a direct study of both classical and quantum aspects of Andreev retroreflection of holes.
The analysis of the magneto-resistance of this device has shown that no
classical retroreflection occurs at the 2DEG/superconductor interface, but
that the Andreev reflected holes are more or less uniformly distributed over
all angles. This diffusive nature of Andreev reflection was attributed to the
presence disorder at the superconductor/2DEG interface[26].
Strictly speaking the required conditions for applying the quantum description of section 4.5 are not satisfied, since the disorder is not uniformly
distributed throughout the cavity, but is located at the 2DEG/superconductor
interface. Nevertheless, the experiment shows a enhanced subgap conductance which was shown to be due to constructive quantum interference
486
1020 I"' ..._...., ..-...." .........,""~,.'... : .. ..".''''
:!2
1000
980
960
-100
B (Gauss)
100
Figure 12. Phase dependent resistance of a ballistic Andreev interferometer. (1) CalcuO. (3)
lation with eq. 14 (see text). (2) Measured resistance at zero voltage bias VDC
Measured resistance at VDC = O.8mV. (from ref. 54)
of Andreev reflected holes, which return to the point contact along phase
conjugated trajectories.
The magnetic field dependence is shown in fig 12. It illustrates the
modulation of coherent Andreev reflection by the superconducting phase.
Assuming a uniform injection of electrons in an interval -()o < () < ()o, one
can write the total Andreev reflection probability as:
A(E) =
J(
1160
dOcosO e i J(ke(E)-kh(E))di+ J
A(T')dlH(E))
12
(14)
-60
with K an arbitrary constant. The calculated modulation of the resistance

obtained from this expression is shown in fig.12. The beating pattern visible
in the amplitude of the oscillations arises from the additional AharonovBohm phase shift due to the magnetic flux through the cavity. The Fourier
transform of the data (not shown) reveals, in addition to the central peak,
side peaks which can be associated with the trajectories which hit the
sidewall of the cavity first before being Andreev reflected, as illustrated in
fig. 11.
6.3. SUPERCONDUCTING PHASE MODULATED CONDUCTANCE
FLUCTUATIONS AND WEAK LOCALIZATION
Both the RMS amplitude of the (universal) conductance fluctuations, as

well as the magnitude of the weak localization correction to the conductance
are affected by the presence of superconductors[55, 56]. In contrast to the
normal case, the variance of the UCF is not reduced by a factor of two
in high magnetic fields. Also in contrast with the normal case, where a
magnetic field B ~ ipo/W L suppresses weak localization, it was found
487
8
.-..
g
....
~
4
0
"'-4
-8
1000
2000
B (Gauss)
3000
0.6
b)
~
<l
'-'
0.0
50
100
6B (Gauss)
150
-1
Figure 13. Top panel: Envelope of the phase dependent conductance modulations in a
T-shaped interferometer: For B;:::100 Gauss the ensemble averaged oscillations are suppressed, and only the phase dependence of the sample specific conductance survives.
Bottom panel: Autocorrelation functions of the conductance oscillations and background
conductance fluctuations (from ref. 53)
that in order to completely suppress weak localization in NIS systems, the

particle-hole symmetry has to be broken by as well by a finite energy[56].
Here we will focus on the modulation of both effects by the superconducting phase[57]. In metals the effect of a superconductor on the sample
specific behaviour is difficult to observe, because it is usually obscured by
the ensemble averaged behaviour of the proximity effect, which can be much
larger than e2 lh. However, De Vegvar et al.[58] have claimed the observation of the modulation of UCF by the superconducting phase.
We will show here experiments by den Hartog et al.[53], who studied
the phase dependence of the conductance fluctuations in a T- geometry,
similar to the one illustrated in fig. 9. The phase difference was tuned
with a magnetic field, which suppresses the phase modulations of the average conductance when the flux though the conductor exceeds q,o Fig.
13 shows that for B ~ 100 Gauss the relatively large ensemble averaged
oscillations are suppressed, but the modulation of the sample specific conductance persists. Its amplitude (~ 0.01e 2 I h )is however much smaller than
the amplitude (~ e2 I h) of the conductance fluctuations themselves.
These results can be compared with calculation from Brouwer et al.[59],
who calculated the phase dependence of the sample specific modulations in
a chaotic Josepson junction coupled to a superconductor with M quantum
488
/\
o
v
-1
0.5 1 1.5 0 0.5 1 1.5 2

o~/2TI
o~/2TI
Figure 14. Top panels: Conductance minus ensemble average (in units of 2e 2 /h ) as a
function of the phase difference between the superconductors. Bottom panels: normalized
correlator c(~<p). (a) is for N=120, M=60, (b) is for N=10, M=160 (from ref.59)
channels and to normal reservoir with N quantum channels. Depending on

the ratio N1M, a phase dependence is found which is more or less sinusoidal
(for N ~ M), or has a high harmonic content (for M ~ N). In the
experiment a sinusoidal phase dependence was found, which is in agreement
with the experimental set up where N ~ M.
7. SUPERCURRENT IN S/N/S JUNCTIONS

In this section we discuss supercurrent transport in S/N /S junctions. Electrons and holes in the normal region generate discrete bound states due
to repeated Andreev reflections between two interfaces. Since the electrons
and holes acquire the macroscopic phase of the superconductor at each Andreev reflection, a coherent connection between the two superconductors is
established and supercurrent (dc Josephson current) can be carried by the
quantized levels of the excitation spectrum.
7.1. ANDREEV BOUND STATES
The quantum states of the elementary excitations in an S/N /S junction

are calculated by the Bogolyubov-de Gennes (BdG) equations (eq. 1). For
simplicity, we assume a one-dimensional model in which the pair-potential
A(1') has only an x dependence A(x) = Llexp(i4>/2) for Ixl > L/2 and
A = 0 for Ixl < L/2 as shown in fig. 15. This assumption is good for
junctions which use a semiconductor as the normal conductor. We also
assume that there is no barrier at the interfaces and that the normal region
489
Lt.1
t:.e-il/J 12
t:.e il/J /2
+
+
-Ll2
Figure 15.
Ll2
Pair potential configuration and Andreev bound states.
is ballistic.
When E < L1, from the continuity conditions for the wavefunctions and
their derivatives at x = L 12 we obtain the dispersion relation exp( i20:( E))
exp[i(k+ - k- )L] exp(i = 1, where nk is the momentum of the electron
(hole) and o:(E) = arccos(EIL1) [60]. Since the considered energy E is
usually much less than the Fermi energy, the energy eigenvalues can be
calculated readily by using the relation Mk == n(k+ - k-) ~ 2ElvF,
nVF
En = 2L [2(7rn+ 0:) >J,
n=0,1,2,,
(15)
where VF is the Fermi velocity in the N region. This equation is equal to the
condition that the phase gain of the round-trip of a particle is 2n7r. If we
consider the energy spectrum for E ~ L1, we obtain the following equation
for (positive) energy levels (see figs. 15 and 16):
nVF
En = 2L [2rr(n+ 1/2) </>].
(16)
One can see that the lowest levels are evenly spaced and there are many
levels for a long junction with L ~ ~o, since the number of levels is roughly
given by L I ~o. The most important issue of eqs. 15 and 16 is that the energy
levels depend on the phase difference >. This indicates that there is a phase
coherent state in the whole SIN IS junction and a supercurrent can flow via
these quantized levels.
In eqs. 15 and 16, E corresponds to the state for the electron moving in the positive- (negative-) direction. The total super current through
the discrete levels is determined by the sum of these currents. This can be
understood in a very simple way: The supercurrent through the discrete
levels is obtained from I = (2eln) Ln dEnld> at T = 0 (see eq. 19). This
indicates that each level carries a supercurrent of positive (negative) value
eVF I L. For a short junction with L ~ ~o, one can obtain the simple dispersion relation arccos(EIL1) = (>/2). This indicates that there is a single
bound state in a short junction [61] (see fig. 16).
490
When E > ~, the excitation spectrum becomes continuous. The wavefunctions and the energy spectrum can be calculated from BdG equations,
but we do not discuss them in detail. Worth mentioning though is that the
spectrum density p(E) depends on </> due to partial Andreev reflection and
this term also contributes to the supercurrent.
Here, however, we should discuss the decay length of the supercurrent
for the understanding of later sections. The average length for which the
difference between the electron and hole phase becomes 1(" determines the
decay length of the supercurrent in the N region and is called the (thermal)
coherence length in the normal region ~n' In the clean limit ~n = nVF 121("kT.
If the normal region is diffusive (or dirty), we have ~n = v'fin 121(" kT.
7.2. LONG JUNCTIONS
The current-phase relation in a long (L ~ ~o) and pure (L <{:: I) junction

with no barrier at the interface was calculated by Kulik [60]. The supercurrent is obtained from the derivative of the temperature Green's function
which is constructed from the wavefunctions obtained by solving BdG equations in the system shown in fig. 1. Ishii[62J, following Kulik, obtained the
Green's function for the whole SIN IS system and discussed the crossover
from a short to a long junction.
Bardeen and Johnson [63] showed that the supercurrent can be derived
by Galilean invariance arguments. They considered a supercurrent flow with
velocity Vs and quasiparticles with zero velocity, since they are in equilibrium with the lattice. One may calculate the total super current current
density from the sum of these contributions, taking into account that the
quasiparticles have an occupation probability f( En + vsnkF), where f is
the Fermi function and En is given by eq. 16.
The above results give the same current-phase relation at T = 0, i.e.,
1= aG nvF P..,
eL 1("
-1("
< </> < 1(".
(17)
Here, G is the conductance of the junction and coefficient a is of order

unity. It shows that the supercurrent in a long and pure junction increases
linearly with </> and jumps discontinuously at </> = 1(", as shown in fig. 16.
This has to be compared with the purely sinusoidal current-phase relation
in a tunnel-type Josephson junction [64].
In the three-dimensional case and at </> = 1(", eq. 17 yields a critical current Ie'" (evFIL)(SI>"}), where S is the NIS contact area and >"F is the
Fermi wavelength. This shows that one transmission mode carries a supercurrent eVF I L as discussed in the previous section. Alternative calculations
of the supercurrent in long clean junctions are given in refs.[65] and [66].
491
(C) , / 1
(d)7 J
Figure 16. Phase dependence of Andreev bound states for (a) a long and (b) a short
junction. Current-phase relation for (c) a long and (d) a short junction.
7.3. SHORT JUNCTIONS
The current-phase relation for a short (L <t: ~o) and ballistic (L <t: f)
junction with no barrier at the N-S interface was obtained by Kulik and
Omel'yanchuk [67]. They used the Eilenberger theory [68], which is valid for
arbitrary temperatures, and obtained the following current-phase relation
for a superconducting constriction with a radius much smaller than f and
~o:
1( '+'"') --
7r
G~(T) . ("'/)
e
8m '+'
2 tan
h ~(T) cos(<Pj2)
2kT
'
-7r
<
</J
< 7r.
(18)
Similar to a long and pure junction, the supercurrent shows a discontinuous

jump at </J = 7r. Very recently a nonsinusoidal current-phase relation has
been directly observed by placing a break junction in a superconducting
loop [69].
The supercurrent in an SIN /S junction can be obtained from the free energy F of the junction by using the fundamental relation I = (2e / h )dF/ d</J
[70]. Therefore, by calculating F from the energy spectrum of the quasipartides in the junction, one can obtain the supercurrent. From the free energy
derived from the eigenvalues of the BdG equations [71] [72], the Josephson
current is
1=
2e "" dEn
En)
-r;
~ d</J tanh ( 2kT
-
4ekT
-h-
00
II
[
(
E )] dp
dEln 2cosh 2kT d</J (19)
where p( E, </J) is the density of states of the continuous spectrum. The first
term describes the current carried by the discrete levels (E < ~) and the
492
second that carried by the scattering states (E > ~) with a continuous

spectrum.
Beenakker described the excitation spectrum of the quasiparticles in
terms of the scattering matrix [73] and showed that the contribution of
the continuous spectrum becomes zero for a very short junction. This can
be seen from another calculation, in which this contribution is given by
N (e~ / 1i ) L / ~o and becomes zero in the limit L / ~o -+ 0, where N is the
number of the transmission modes [74].
According to the scattering matrix method [73], En is given by En =
~[1- Tnsin2(</>/2)P/2 (n = 1,2 .. N), where Tn(O :::; Tn :::; 1) are the
eigenvalues of the transmission matrix tt t of the normal region. Inserting
these En into the first term in eq. 19, we obtain the expression for the
supercurrent in a very short junction:
e~ 2
Tn
( En )
I = 21i sm </> ~ En tanh 2kT
(20)
Eq. (18) can be derived from eq. 20 by making Tn = 1 and using the relation
G = N ( e 2 /7r 1i). This equation will be used again in the discussion of the
quantized critical current in section 7.2.
7.4. SUPERCURRENT IN TERMS OF ANDREEV REFLECTION
PROBABILITY
Alternatively the dc Josephson current can be calculated from the scattering amplitudes of Andreev reflection for the quasiparticles at the N/S
interface [75]. Let us assume a SIN /S junction with the pair-potential configuration shown in Fig. 15. As discussed before, there is a continuous spectrum for energies E > ~ where the eigenfunctions are scattering states.
Furusaki and Tsukada[75] derived the temperature Green's function from
the retarded Green's function, which was constructed from the eigenfunctions. They obtained the DC Josephson current:
where tr is the sum over the conductive channels in the superconductor

and Wn = 7rkT(2n + 1). Here, a(</>,iwn ) is the probability amplitude of
the process in which an electron coming from the left side (x < -L/2) is
reflected as a hole, and a( -</>, iw n ) is that of the reverse process in which a
hole is reflected as an electron. These amplitudes are obtained by solving
the BdG equations, in which E is replaced by iwn . It should be noted
that once the probability amplitude a at an energy E > ~ is obtained,
493
the contribution to the supercurrent from all a with the energy E > can
be calculated by this equation. This equation looks very different from eq.
19, but coincides with results obtained from eq. 20. Indeed this expression
reduces to eq. 20 in the case of a very short junction with no barrier at the
interface.
7.5. DIFFUSIVE JUNCTIONS
We will describe shortly the supercurrent in a diffusive junction (L ~ f)
(for details, see [2]). One of the most effective methods of calculating the
supercurrent in a diffusive junction with arbitrary lengths and at arbitrary
temperatures is based on the Usadel equations (i.e., the ensemble-averaged
Green's functions method) [76]. For a very short and dirty junction with
f ~ L ~ ~n' Kulik and Omelyanchuk give an exact solution of the currentphase relation that is somewhat distorted from a sinusoidal one at a low
temperature.
The Usadel equation has also been solved numerically and the result
was that the current-phase relation is sinusoidal and the critical current Ie
decays exponentially with L as Ie ,. . ., exp( -L/~n) at a temperature close
to the critical temperature [77]. The former result agrees with the theory
based on the Ginzburg-Landau equations [78].
8. SUPERCURRENT IN LOW-DIMENSIONAL SYSTEMS

This section discusses the supercurrent in two-, one-, and zero-dimensional
systems based on the theories described in the previous section. Lowdimensionality leads to new quantum effects like quantization of the critical
current and resonant super current flow through discrete states.
8.1. SUPERCURRENT FLOW IN A 2DEG
A supercurrent in an S/2DEG/S junction was first observed in a p-type

InAs-coupled junction (see fig. 2( d)) [21]. It was shown that the critical
current has a dependence of Ie""'" exp(-L/~n) with D = vFf/2. A Josephson junction coupled with a high mobility 2DEG was first realized using
an InAs/ AISb heterostructure [82], and subsequently with an InAs-inserted
InAIAs/InGaAs heterostructure [83]. A ballistic S/2DEG/S junction with a
gate was achieved using the latter material[84] and showed a high gate controllability. (fig. 17). This junction had a voltage gain over 1, and enabled
the first demonstration as a Josephson field effect transistor (JOFET). It
was also used for the study of Fabry-Perot interference of the supercurrent[85,86].
494
(a)
Nb
InAIAs
InGaAs
InGaAs
InAIAs
n-f:. InAiAs
InAiAs
InP Substrate
8-r------.......-------".,
(b)T= 1 K
~are
(Al)
<'
Nb
-0.8 V
.( InAs
(2DEG)
8~
i==-_ _-JF-_ _-=:=:S-0.9 V

-1.1 V
-4
-8~-~----~--~--~
-0.1
-0.05
0
0.05
0.1
Voltage (mV)
Figure 17. (a) Schematic view of a gated junction. (b) Current-voltage characteristics
with changing gate voltage (from ref.84).
8.2. SUPERCONDUCTING QUANTUM POINT CONTACT
8.2.1. Theory
If a ballistic PC is located between two superconductors, one might ask
the simple question: Is the critical current of this so-called superconducting
quantum point contact (SQPC) quantized like its conductance?
The supercurrent in a very short SQPC (L < ~o) was studied by
Beenakker and van Houten [61]. They assumed a ballistic PC which has
a special set of transmission eigenvalues; i.e., Tn = 1 if 1 $ n $ No and
Tn = 0 if n > No, where No is the number ofthe occupied one-dimensional
subbands. Then the energy eigenvalues of the Andreev states (section 7.3)
are given by En = ~I cos(</>/2)1 for 1 ~ n $ No, and (eq. 20) the supercurrent at T = 0 becomes I = No( e~/1i )sin( </>/2). This shows that the critical
current is quantized in units of e~/1i. Comparing this equation with eq.
18, one finds that its current- phase relation is the same as of the classical
short junction. (fig. 16) The theory in section 7.4 can be used also for the
calculation of the supercurrent in an SQPC with arbitrary lengths. In the
case of adiabatic transport, Furusaki et al. obtained an analytic equation
for the supercurrent that provides the same result as that by Beenakker
and van Houten for a very short junction. For a long junction (L > ~o) at
T = 0 they find[89] [90]:
1= 1reL
No
2: vn </>,
-1r
< </> < 1r,
(22)
n::;;l
where Vn is the electron velocity in the nth channel (v n ~ VF). It is found

that one mode carries a supercurrent of eVF 1L and the critical current is
discretized in this unit. Note that this current-step height is not univer-
495
1.0
(a)
1)(1
x=L
x=O
(b)
<'~
.5; 0.5
0.00
100
Figure 18. (a) Schematic drawing of an SQPC. (b) Critical current as a function of
constriction width. Z value is 0, 0.2, 0.5, and 1.0 from top to bottom (from ref.89).
sal, but rather depends on parameters in the N region (fig. 18). So far we
have assumed that the N IS interface is ideal. In actual interfaces, however,
we have to take into account the normal reflection due to a Schottky barrier or Fermi velocity difference at the interface. Furusaki et al. described
the supercurrent in terms of a Z value (see section 4.1) and showed that
the stepwise change of the critical current becomes blurred as the Z value
increases. In addition, they obtained resonant peaks caused by the interference due to the normal reflection, as shown in fig. 18
8.2.2. Experiment
A supercurrent jump on the order of e6.lh was observed by using a mechanically controllable break junction made from Nb [91]. This result suggests
that the critical current changes discretely when the contact diameter is
increased. A more systematic investigation was done using an SQPC in a
ballistic semiconductor heterostructure [100]. Figure 19 shows the measured
critical current and conductance as a function of gate voltage. The semiconductor heterostructure was the same as that in fig. 17. Thise data indicate
that the critical current in an SQPC is approximately quantized by the
number of one-dimensional subbands in the QPC. Because ~o( >::J 0.14/.Lm)
is much less than L = 0.3 /.Lm, the step height of the critical current depends on the 2DEG parameter. Resonance structures were also observed at
the edge of some current steps. They can be explained by the interference
of the quasiparticles discussed above (see fig. 20). Also important is that
the conductance steps are not yet fully developed. This is probably due
to the limited mobility of the 2DEG, and implies that the ideal situation
of transmission eigenvalues being either 0 or 1 is not yet achieved. To further enhance the quality of the steps in the critical current, experiments
496
12
"
r---------------~~T50
II =IO ,~;
'.
(b)
"
t'.
1m
I'
I
..,
:..>
<=t
:'!"'<:I
:..>
""
U
;~
4,.
/
- . _.' :". -
.~
4 " .:.:,-:.....
1.6
- II
- 2
40 ~
-5
C
30 ~
::l
,:." - . 5
' .. :'" - 4
~'2..wt/ "'-:~-3
:>
8 ~.
/. ..-' 7- "~
7;
'.
.- .........( ..........
6-J'.",::'
~:J~"
20 ~
'P
Ie =>
'I:
10
=1
-1.4 . 1.3 1.2

Gate Voltage (V)
1.5
Figure 19. (a) SEM micrograph of an SQPC, (b) Measured approximate quantization
of the critical current and the conductance as a function of gate voltage at 10 mK (from
ref, 100).
are needed on devices with higher critical currents, which are closer to the
theoretical values. Such experiments will also help to clarify several aspects
of the ac Josephson effect and the new type of noise in the SQPC [92] [93].
8.3. O-DIMENSIONAL SYSTEMS
A supercurrent through a O-dimensional system has been discussed for a

semiconductor quantum dot weakly coupled by two tunnel barriers to two
superonductors [74]. The supercurrent flowing through a resonant state in
the quantum dot is calculated from the quasiparticle excitation spectrum
for both discrete and continuous ones (which are obtained by the scattering
matrix method) and eq. 19. The resonances can be divided into two asymptotic regimes; i.e., a wide resonance (r ~ ~) and a narrow one (r ~ ~),
where r = r 1 + r2 and r 1,2 is the tunnel rate through each tunnel barrier. The critical current was obtained for both cases at T=O and for the
crossover between these two cases as
(23)
In the calculation, the Coulomb repulsion U (or charging energy) in the
dot is neglected. Glazman and Matveev calculated the resonant Josephson
current by taking into account U [94] and showed that the critical current
is suppressed by a factor of r / ~ for narrow resonance (r ~ u, ~). This
is a very large reduction of Ie. On the other hand, for wide resonance Ie
is not affected by U, but the resonance width is broadened due to the
Kondo effect by a factor of In(r / ~)( U, r ~ ~). For a further discussion of
497
24
17
23
..::
16
22
15 ___
21
14 ~
20
13
19
12
18
17
-20
-19.8
- 19.6
-19.4
- 19.2
- 19
11
Vg (V)
Figure 20. Mesoscopic fluctuations of the critical current Ie and the conductance G n
measured with changing gate voltage Vg at 20 mK (from ref. 101)
superconductor induced Andreev states in confined systems and quantum

dots we refer to ref. 59, and references therein.
9. SAMPLE SPECIFIC FLUCTUATIONS OF THE SUPERCURRENT

The phase of the electron wave function in a normal metal is maintained
over the phase breaking length. This provides quantum interference effects
like UCF in a mesoscopic-size disordered sample at low temperatures. As
discussed in previous sections, the supercurrent in an S IN IS junction flows
through the N region by coherent multiple Andreev refelction. Therefore,
it is natural to expect sample-to-sample (or mesoscopic) fluctuations of the

critical current in a diffusive SIN IS junction (L ~ l) with a similar origin
as that of UCF.
9.1. THEORY
In a short junction with ~ <t: ET, Beenakker showed that the mesoscopic
fluctuations bIe of the critical current at T = 0 have a universal value of
'" e~/n, independent of C or L [73]. The opposite case (a long junction
with ET <t: Ll was studied by Al'tshuler and Spivak [96]. They showed that
He depends on the parameters of the N region and the Thouless energy
ET = nD I L2 plays the main role instead of ~. When the temperature
satisfies the condition ~n ~ L , He is given by C( eET In) in the case of
L ~ W, d. Here C ~ 0.6 , and Wand d are the width and the thickness of
the N region respectively. Numerical simulations [95Jgive C = 0.46 0.06.
498
9.2. EXPERIMENT
Observations of sample-specific fluctuations of the critical current can be

achievd by changing the Fermi energy of the 2DEG. A gated semiconductorcoupled Josephson junction is particularly suited for this purpose. Here we
will show experimental results that show the sample-specific fluctuations of
the critical current [101]. The sample is a p-type InAs-coupled junction (see
fig. 2( d)) with an metal-insulator-semiconductor gate structure. Figure 20
shows the measured critical current Ie and the conductance Gn(measured
by suppressing the supercurrent with a voltage bias). Both Gn and Ie show
fluctuations as a function of gate voltage, i.e., with varying the Fermi energy. Moreover, the behaviour of the Ie fluctuations follows that of the G n
fluctuations. It was clarified that Gn showed a very clear UCF as a function of a magnetic field. These results indicate that the critical current
fluctuations have the same physical origin as that of UCF.
The measured He was about 15 nA in the weakly, and 2-4 nA in the
strongly localized regime, respectively. The experimental results should be
compared with the theory by Al'tshuler and Spivak [96], since ET < < ~ in
the used 2DEG. However, they did not calculate the fluctuation amplitude
in the case of W ~ L ~ d, which is the exact case for the measured
sample. Koyama et al. showed that the experimental value in the weaklylocalized regime could be explained based on AI'tshuler and Spivak's theory
by taking into account the effects of temperature, sample geometry, and
reduced Andreev refelction probability [98, 99].
In a strongly-localized regime the problem of the fluctuation amplitude
has not been solved yet. In this regime there is another interesting phenomenon: the critical current decreases with decreasing temperature [102]
[103]. This effect should be discussed from the viewpont of interaction effects associated with Anderson localization.
10. Acknowledgements
The authors thank their collaborators from Nippon Telegraph and Telephone Corporation and the University of Groningen for their contributions
to this work. B.J .van Wees acknowledges the Dutch Foundation for Fundamental Research on Matter and the Dutch Academy of Sciences for financial
support. H. Takayanagi thanks prof. J.B. Hansen from the Technical University of Denmark, and prof. A. Furusaki from Kyoto University for their
contributions to this work.
499
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
T.M. Klapwijk, Physica B197, 481 (1994).

K.K. Likharev, Rev. Mod. Phys. 101 (1979).
C.W.J. Beenakker, in Mesoscopic Quantum Physics, eds. E. Akkermans, G. Montambaux, J. Zinn Justin, and J.-L. Pichard, North Holland, Amsterdam (1994);
C.W.J. Beenakker, in Transport Phenomena in Mesoscopic Systems, eds. H.
Fukuyama, and T. Ando, Springer, Berlin (1992); C.W.J. Beenakker, Rev. Mod.
Phys., to be published.
C.J. Lambert et al., in preparation.
P.G. de Gennes, Superconductivity of Metals and Alloys, Benjamin, New York (1966)
A.F. Andreev, J. Exptl. Theoret. Phys. 46, 1823 (1964)[Sov. Phys. JETP 19, 1228
(1964)].
P.A. Benistant et al., Phys. Rev. B32, 3351 (1985).
P.C. van Son et al., Phys. Rev. Lett. 59, 2226 (1987); Phys. Rev. B37, 5015 (1988).
W.M. van Huffelen, T.M. Klapwijk, and L. de Lange, Phys. Rev. B45, 535 (1992);
Phys. Rev. B47, 5170 (1993)
A. Kastalsky, L.H. Greene, J.B. Barner, and R. Bhat, Phys. Rev. Lett. 64, 958
(1990).
S.J.M. Bakker et ai., Phys. Rev. B49, 13275 (1994).
A. Kleinsasser et al., Appl. Phys. Lett. 57, 1811 (1990); A. Chrestin, T. Matsuyama,
and U. Merkt, to be published.
M. Hatano et al., J. Vac. Sci. Technol. B7, 1333 (1989).
Z.G. Ivanov, Phys. Stat. Sol. (a) 113,439 (1989).
J.R. Gao et al., Semicond. Sci. Technol. 11, 621 (1996); Proc. Int. Conf. Electron
Microscopy, 681 (1994).
K.-M.H. Lenssen et ai., Appl. Phys. Lett. 63, 2079 (1993); Proc. EP2DS 1993 Newport, Surf. Sci. 305, 476 (1994).
A.M. Marsh, D.A. Williams, and H. Ahmed, Phys. Rev. B50, 8118 (1994); Physica
B203, 307 (1994); Phys. Rev. B50, 8118 (1994); R. Taboryski et al., Appl. Phys.
Lett. 69, 656 (1996).
C. Nguyen, H. Kroemer, and E.L. Hu, Phys. Rev. Lett. 69, 2847 (1992)
L. Mur et al., Phys. Rev. B54, 2327 (1996).
F. Rahman et al., Phys. Rev. B54, 14026 (1996); D. Uhlisch et al., Physica B225,
197 (1996); A. Yu. Kasumov et al., Phys. Rev. Lett. 77, 3029 (1996); J.P. Heida et
al., in preparation.
H. Takayanagi, and T. Kawakami, Phys. Rev. Lett. 54, 2449 (1985).
C.W.J. Beenakker, Phys. Rev. B46, 12841 (1992).
Y. Takane, and H. Ebisawa, J. Phys. Soc. Jpn. 61,3026 (1992).
G.E. Blonder, M. Tinkham, and T.M. Klapwijk, Phys. Rev. B25, 4515 (1982).
R.A. Riedel, and P.F. Bagwell, Phys. Rev. B48, 15198 (1993).
P.H.C. Magnee et al., Appl. Phys. Lett. 67, 24 (1995) .
H. Takayanagi et al., Phys. Rev. Lett. 75, 3098 (1995).
Yu. V. Nazarov, Phys. Rev. Lett. 73,134 (1994).
S.N. Artemenko, A.F. Volkov, and A.V. Zaitsev, Solid State Comm. 30, 771 (1979)
A. Kastalsky et al., Phys. Rev. Lett. 67, 3026 (1991).
P.H.C. Magnee et al. Phys. Rev. B50, 4594 (1994),
W. Poirier, D. Mailly, and M. Sanquer, in Proceedings Rencontres de Moriond, Les
Arcs (1996).
H. Pothier et aI., Phys. Rev. Lett. 73, 2488 (1994).
B.J. van Wees et ai., Phys. Rev. Lett. 69, 510 (1992).
C.W.J. Beenakker, B. Rejaei, and J.A. Melsen, Phys. Rev. Lett. 72, 2470 (1994).
N.R. Claughton, R. Raimundi, and C.J. Lambert, Phys. Rev. B, to be published.
A.F. Volkov, A.V. Zaitsev, and T.M. Klapwijk, Physica 210C, 21 (1993); A.F.
Volkov, Phys. Lett. A174, 144 (1993); Phys. Lett. A187, 404 (1994); A.V. Zaitsev,
500
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
Pis'ma Zh. Eksp. Teor. Fiz. 51,35 (1990)[JETP Lett.51, 41 (1990)]; F.W.J. Hekking
and Yu.V. Nazarov, Phys. Rev. B49, 6847 (1994); A.A. Golubov, F.K. Wilhelm,
and A.D. Zaikin, Phys. Rev. B55, 1123 (1997).
Yu. V. Nazarov, Phys. Rev. Lett. 73, 1420 (1994).
Yu. V. Nazarov, unpublished.
J.A. Melsen, and C.W.J. Beenakker, Physica B203, 219 (1994).
H. Takayanagi et al., in preparation.
C.W.J. Beenakker, J.A. Melsen, and P.W. Brouwer, Phys. Rev. B51, 13883 (1995);
Y. Takagaki, and H. Takayanagi, Phys. Rev. B53, 14530 (1996).
D. Heslinga et al., Phys. Rev. B49, 10484 (1994).
N. Argaman, and A. Zee, Phys. Rev. B54, 7406 (1996).
M. Biittiker, Phys. Rev. Lett. 57, 1761 (1986).
C.J. Lambert, J.Phys. Condo Matt. 3, 6579 (1991); B.J. van Wees, K.-M.H. Lenssen,
and C.J.P.M. Harmans, Phys. Rev. B44, 470 (1991); A. Kadrigrobov et al., Phys.
Rev. B52, 8662 (1995); S. Datta, P.F. Bagwell, and M.P. Anantram, in Physics. of
Low-Dimensional Structures, to be published.
M. Octavio et al., Phys. Rev. B27, 6739 (1983).
A circulating supercurrent may however affect the measurement indirectly, see B.J.
van Wees et al., Phys. Rev. Lett. 76, 1402 (1996).
H. Nakano, and H. Takayanagi, Phys. Rev. B47, 7986 (1993); Solid State Comm.
80,997 (1991).
V.T. Petrashov et al., Phys. Rev. Lett. 74, 5268 (1995); Pis'ma Zh. Eksp. Fiz. 60,
589 (1994)[JETP Lett. 60, 607 (1994)].
A. Dimoulas et al., Phys. Rev. Lett. 74, 602 (1995).
S.G. den Hartog et al., Phys. Rev. Lett. 76, 4592 (1996) .
S.G. den Hartog et al., Phys. Rev. Lett. 77, 4954 (1996).
A. Morpurgo et al., Phys. Rev. Lett., to be published.
P.W. Brouwer, and C.W.J. Beenakker, Phys. Rev. B52, 3868 (1995); Y. Takane,
and H. Ebisawa, J. Phys. Soc. Jpn. 61, 2858 (1992).
P.W. Brouwer, and C.W.J. Beenakker, Phys. Rev. B52, 16772 (1995); Y. Takane,
and H. Otani, J. Phys. Soc. Jpn. 63, 3361 (1994).
B.Z. Spivak, and D.E. Khmelnitskii, Pis'ma Zh. Eksp. Teor. Fiz. 35, 334
(1982)[JETP Lett. 35,412 (1982)].
P.G.N. de Vegvar et al., Phys. Rev. Lett. 73, 1416 (1994).
P.W. Brouwer, and C.W.J. Beenakker, Phys. Rev. B54, 12705 (1996).
I. O. Kulik, Sov. Phys. JETP 30, 944 (1970).
C. W. J. Beenakker and H. van Houten, Phys. Rev. Lett. 66, 3056 (1991).
C. Ishii, Prog. Theor. Phys. 44, 1525 (1970).
J. Bardeen and J. L. Johnson, Phys. Rev. B5, 72 (1972).
B. D. Josephson, Phys. Lett. 1, 251 (1962).
B. J. van Wees et al., Phys. Rev. B44, 470 (1991).
R. Kiimmel, Phys. Rev. B16, 1979 (1977); U. Schiissler and R. Kiimmel, Phys. Rev.
B47, 2754 (1993); A. Martin-Rodero, F.J. Garcia-Vidal, and A. Levy Yeyati, Phys.
Rev. Lett. 72,554 (1994).
I. O. Kulik and A.N. Omel'yanchuk, Sov. J. Low Temp. Phys. 4, 142 (1978).
M. C. Koops, G. V. van Duyneveldt, and R. de Bruyn Ouboter, Phys. Rev. Lett.
77, 2542 (1996).
P. W. Anderson, in Ravello Lectures on the Many-Body Problem, ed. by E. R.
Gianello (Academic, New York, 1963).
J. Bardeen, R. Kiimmel, A. E. Jacobs, and 1. Tewordt, Phys. Rev. 187, 556 (1969).
C. W. J. Beenakker and H. van Houten, in Nanostructures and Mesoscopic Systems,
ed. by W. P. Kirk and M. A. Reed (Academic, Boston, 1992).
C. W. J. Beenakker, Phys. Rev. Lett. 67, 3836 (1991), 68 1442 (1992).
C. W. J. Beenakker and H. van Houten, in Single-Electron Tunneling and Meso-
501
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
scopic Devices, ed. by H. Koch and H. Liibbig (Springer, Berlin, 1992).

A. Furusaki and M. Tsukada, Solid State Commun. 78, 299 (1991).
K. D. Usadel, Phys. Rev. Lett. 25, 507 (1970).
K. K. likharev, Sov. Phys. Tech. Phys. Lett. 2, 29 (1976).
L. G. Aslamazov, A. I. Larkin, and Yu. N. Ovchinikov, Sov. Phys. JETP 28,171
(1968).
V. Kresin, Phys. Rev. B34, 7587 (1986).
M. Yu. Kupriyanov and V. F. Lukichev, Sov. Phys. JETP 67, 1163 (1988).
A. F. Volkov and H. Takayanagi, Phys. Rev. B35, 15162 (1996).
C. Nguyen, J. Werking, H. Kraemer, and E. L. Hu, Appl. Phys. Lett. 57, 87 (1990).
J. Nitta, T. Akazaki, H. Takayanagi, and K. Arai, Phys. Rev. B46 14286 (1992).
T. Akazaki, H. Takayanagi, J. Nitta, and T. Enoki, Appl. Phys. Lett. 68, 418 (1996).
H. Takayanagi, T. Akazaki, and J. Nitta, Phys. Rev. B51, 1374 (1995).
H. Takayanagi and T. Akazaki, Jpn. J. Appl. Phys. 34, 4552 (1995).
L. Y. Gorelik, V. S. Shumeiko, R. I. Shekhter, G. Wendin, and M. Jonson, Phys.
Rev. Lett. 75, 1162 (1995).
R. Fazio, F. W. J. Hekking, and A. A. Odintsov, Phys. Rev. Lett. 74, 1843 (1995).
A. Furusaki, H. Takayanagi, and M. Tsukada, Phys. Rev. Lett. 67, 132 (1991).
A. Furusaki, H. Takayanagi, and M. Tsukada, Phys. Rev. B45, 10563 (1992).
C. J. Mulier, J. M. van Ruitenbeek, and L. J. de Jongh, Phys. Rev. Lett. 69, 140
(1992).
D. Averin and A. Bardas, Phys. Rev. B53, R1705 (1996).
D. Averin and H. T. Imam, Phys. Rev. Lett. 76, 3814 (1996).
1. I. Glazman and K. A. Matveev, JETP Lett. 49, 659 (1989).
Y. Takane, J. Phys. Soc. Jpn. 63, 2668 (1994).
B. L. Al'tshuler and B. Z. Spivak, Sov. Phys. JETP 65, 343 (1987).
Y. Takane, J. Phys. Soc. Jpn. 63,4310 (1994).
Y. Koyama, Y. Takane, and H. Ebisawa, J. Phys. Soc. Jpn. 64, 1466 (1995).
Y. Koyama, Y. Takane, and H. Ebisawa, J. Phys. Soc. Jpn. 65, 683 (1996).
H. Takayanagi, T. Akazaki, and J. Nitta, Phys. Rev. Lett. 75,3533 (1995); Surface
Sci. 361/362, 289 (1996).
H. Takayanagi et aI., Phys. Rev. Lett. 74, 166 (1995); Physica B203, 291 (1994).
H. Takayanagi, et al., Phys. Rev. Lett. 74, 162 (1995).
H. Takayanagi and T. Akazaki, Solid State Commun. 96, 815 (1995).
SCANNING PROBE MICROSCOPES AND THEIR APPLICATIONS
L. L. SOHN
Dept. of Physics, Princeton University, Princeton, NJ 08544,

USA
C. T. BLACK AND M. ERIKSSON
Dept. of Physics, Harvard University, Cambridge, MA 02138,

USA
M. CROMMIE
Dept. of Physics, Boston University, Boston, MA 02215, USA

AND
H. HESS
Bell Laboratories, Lucent Technologies, Murray Hill, NJ 07974,

USA
Recent advances in nanofabrication technology have provided access to
continually smaller length scales. Consequently, a number of novel quantum
phenomena-some of which have already been highlighted in this bookhave been observed in fabricated structures oflength scale ",50-100 nm. Because of the clear possibility that additional novel effects will become manifest at even smaller dimensions, there is strong motivation to develop: (1)
fabrication techniques which can routinely access sub-50 nm length scales in
a variety of materials and (2) measurement techniques which can spectroscopically probe the local electronic properties of "ultra" nanometer-scale
devices.
It is within this context that we now discuss scanning probe microscopes.
No longer just a tool for imaging the local or atomic structure of conductors,
insulators, or optically-active materials, scanning probes-including the
scanning tunneling microscope (STM), the atomic force microscope (AFM),
and more recently, the near field scanning optical microscope (NSOM)are now being developed for sub-50nm lithography and local spectroscopy
probes. That scanning probe microscopes have the capability to both fabricate and noninvasively probe devices in situ heightens their attractiveness
503
L. L. Sohn et al. (eels.), Mesoscopic Electron Transport, 503-547.
504
as a new class of instruments for next-generation experiments in mesoscopic
physics.
In order to appreciate fully the unique capabilities of scanning probe
microscopes, it is necessary to understand (1) the fundamental basis; (2)
the hardware operation; and (3) the limitations of each individual probe. In
the following sections, we outline the basic operation of 3 scanning probes:
STM, AFM, and NSOM. In addition, we also highlight a few choice applications of each scanning probe: from STM-fabricated quantum corrals, to
AFM-fabricated atomic point contacts, to NSOM-based spectroscopy of
individual quantum wells.
1. The Scanning Tunneling Microscope

Invented by Binnig and Rohrer in 1981 [1], the STM is based on electron
tunneling. Using piezodrives, one can position the STM's conducting tip to
within a few angstroms of the surface of a metallic substrate (see Fig. 1a). At
such close distances, the wavefunctions of both will overlap. Consequently, if
a bias voltage is applied between the tip and the metallic surface, electrons
from the tip have a finite probability of tunneling across the vacuum space
to the sample, resulting in a measurable current on the order of 1 nA [2].
Because the tunnel current is confined to a narrow beam underneath the
tip, the STM can image the surfaces of metals, semiconductors, and superconductors with extremely high resolution [2, 3]. In addition, it can probe
local adsorbate properties and eptiaxial growth phenomena. Moreover, the
STM can fabricate and spectroscopically probe atomic-scale structures.
Below, we discuss the basic principles of the STM, the elements ofinstrumentation and operation, and finally a few of its ever-growing applications.
1.1. BASIC PRINCIPLES
In this section, we highlight the basic operating principles of the STM.

For a more thorough discussion, we refer the reader to Black [2] and to
Chen [4]-both of which we follow closely here.
As we have stated above, there is a finite probability for an electron to
tunnel between the STM tip and the metallic surface because the wavefunctions of both overlap. In Fig. 1b, the tip and the metallic surface are
modeled as free-electron metals which are separated by a high potential
barrier (the vacuum gap between the two). As shown in the figure, when a
voltage (eV) is applied, the electrochemical potential of one electrode shifts
with respect to that of the other. If the barrier height is larger than the
applied bias voltage, the resulting tunneling current can be calculated using perturbation theory [5]. Using Fermi's Golden Rule, we determine the
505
(a)
--d
-I
lev
'-'
=>
(b)
Position
..
Figure 1. Schematic of STM operation. (a) When a sharpened metal tip electrode is placed
in close proximity to a substrate, electrons in the tip tunnel with a finite probability to
the sample, resulting in a measurable electric current. (b) The STM junction modeled as
a one-dimensional potential barrier. [From Ref. [2].]
tunneling rate of electrons from electrode 1 to electrode 2 to be

(1)
Here PI and P2 are two-fold, spin-degenerate densities of states in the two

electrodes, f( e) is the Fermi distribution function, and \T( e)\2 is the matrix
element coupling electrons from electrode 1 to electrode 2. IT normal metals
are employed, Pb P2, and \T(E)12 are constant over eV. Eq.(l) is thus
471"
r I _ 2 (eV) = h
2
eV
PIP2 \T\ 1 _ e-~eV
(2)
where,B == l/kBT. Similarly, reverse tunneling from electrode 2 to electrode

1 is
506
r2-+I(eV)
41r
= TPIP2 1T I
eV
e{jeV -1 .
(3)
Combining Eqs.(2) and (3), we find that the total current is
(4)
or
41r 2
2
I = TITI PIP2 e V .
(5)
Note that Eq.(5) is Ohm's law where (41r/1i)ITI2pIP2e2 is the tunneling

resistance.
For small applied voltages, the tunneling current depends linearly on
ITI2. Although the exact form of ITI2 is dependent on the shape of the potential barrier [6), it is dominated by the exponential decay of the electronic
wavefunction into the barrier [2), i.e.
(6)
where K. = v'(2m4/1i is dependent on the work function of the metal,
4> (I"V 3-12 eV), and where d is the width of the potential barrier. For a
typical value 4> 1"V4eV, the ITl2-dependence leads to a tunnel current which
roughly decays an order of magnitude for every Angstrom the barrier width
d is increased.
1.1.1. Region of Tunneling Contact
To determine how localized the tunnel current is, i.e. the lateral resolution,
we refer to the tip-sample model shown in Fig. 2.
Sample
Figure 2. STM resolution. Schematic of the STM geometry, with the tip modeled as a
sphere with radius R above a flat, conducting surface. [From Ref. [2].]
507
Here, the tip with a radius of curvature R is positioned a distance distance d from a fiat, conducting sample and emits electrons with all possible
k-vectors. As shown in Fig. 2, an electron with wave-vector kl can tunnel
through the sample with a probability'" e- 2k1d , as given by Eq.(6). IT we
define the region of significant electron tunneling to be roughly where the
transmission probability for an electron with wave-vector k2 is 1/ e2 of that
for an electron with wave-vector kl [2], then by geometry,
e -2 e -21td -_ e -21t[v(d+R)2+P -R] .
(7)
Solving this equation for l, we find that

l=
2(R +d)
~
+ ~ '"
~2
2(R +t!J
~
(8)
where the last equality holds for (R + d) ~ ~-I. Eq.(8), although a rough
approximation, agrees well with more complicated theories of tunnelingbeam widths [7,8]. Taking R=1000 A, d=5 A, and ~=1 A-I, we find that
the tunnel current is confined to a beam width l of only 45 A. Thus, it is the
very exponential decay of the electronic wavefunctions into the tunneling
barrier [Eq.(6)] which makes it possible to form highly localized beams of
current and in turn, to have very high lateral resolution.
1.1.2. Imaging
STM imaging is accomplished by attaching a conducting-probe tip to a
piezodrive of z-, y-, and i-piezoelectric transducers-all of which are capable of moving on the Angstrom-scale. [By convention, the zy plane is
taken to be the plane of the sample and z, perpendicular to the surface.]
Mter the tip and the sample are coarsely brought together, a constant
voltage bias is applied between the two. The resulting tunnel current is
monitored while the tip and sample are then finely brought together. Once
the current becomes appreciable, it is compared to a reference signal and
the tip is then rastered across the sample surface (see Fig. 3). Any changes
in surface topography are registered as variations in the measured tunnel
current and are compensated for by an electronic feedback circuit which
adjusts the tip-sample spacing so that the current is kept constant at each
point during the surface scan (see Black [2] and Chen [4]). By recording the
voltage signals sent to the i-transducer as a function of tip position on the
surface, one can topographically image a surface. Again, the lateral resolution of an STM image is roughly given by the width of the probe current
beam, Eq.(8). An example of an atomically-resolved image obtained with
an STM in shown in Fig. 4.
508
Figure 3. STM imaging process. (a) The STM tip is positioned a distance above the
sample, resulting in a tunnel current. (b) When the tip is moved over a higher region of
the sample surface, the measured current increases. (c) In response to increased tunnel
current, the electronic feedback circuit retracts the tip a distance I:l.z in order to return
the current to its initial value. Changes in tip height I:l.z are recorded as a function of tip
position to obtain an image of surface topography. [From Ref. [2].]
Figure 4. Height deflection image of 2H~NbSe2 taken over 5 nm x 5 nm area at 4 K. Both

atomic and charge density wave corrugations are visible. The vertical deflection in the
tip to maintain constant current conditions is 0.02 nm.
509
1.2. ELEMENTS OF STM INSTRUMENTATION
Although the concept of electron tunneling is straightforward, the construction of an STM proves to be relatively difficult because of problems associated with maintaining the Angstrom-scale gap between the tip and the
surface. A number of sophisticated mechanical, electronic, and feedbackcontrol technologies have evolved in order to maintain this gap [4]. In this
section, we address the necessary technologies, including piezodrives, electronic feedback, and instrument vibration isolation, to construct an STM.
1.2.1. Piezodrives
The motion of an STM tip is controlled by piezodrives made from piezoelectric crystals (typically lead-zirconate-titanate ceramics) having expansion
coefficients on the order of a few A/v. Although a wide variety of geometries
exist fo~ piezodrives, including the "tripod" [9] and the "bimorph" [10], the
most WIdely employed geometry is the "tube" [11]. Compared to other geometries, the tube is simple in both mechanical and electrical design [4, 11].
In Fig. 5, we show a piezodrive in the tube geometry. Five nickel-plated
electrodes are attached to the tube: one on each of four outer quadrants of
the tube and a single electrode lining the inner wall [2, 4]. When a voltage
bias is applied across the electrodes, strain is induced in the piezocrystals
which, in turn, leads to a change in the length of the tube. For an applied
voltage in the i-direction, the length change (measured in units of A/V)
in the i-direction is
(9)
where Sz == 5z/ z is the strain in the i-direction. Note that by convention,
the i-, y-, i-directions are labeled 1, 2, and 3, respectively.
1.2.2. Vibration Isolation

Because of the compact and rigid design of an STM (see Fig. 6, highfrequency vibrations (> 1kHz) can disrupt STM operation. These frequencies can couple to the lowest resonant mode of the piezodrive and cause
the tip to move relative to the sample. As we have shown with Eq.(6),
the tunneling current and lateral resolution are exponentially dependent
on the distance between the tip and the sample-any vibrations present
510
(a)
x-
x+
y+
1.T fly
+
0
(b)
-4-
6
+
(c)
flz
Figure 5. Pielodrive operation. (a) The pielodrive mechanism conmts of a tube of pieloelectric material with quartered electrodes attached to the exterior of the tube, and
a single electrode lining the inside wall. Motion in the three orthogonal directions is
achieved by applying a voltage between a particular outer quadrant of the tube and the
single electrode lining the tube's inner wall. (b) A voltage diJ[erence between a single
outer quadrant and the inner electrode causes the tube to bend, thus generating lateral
motion. (c) A voltage applied to the inner electrode (relative to all outer quadrants)
causes the tube to expand or contract. [From Ref. [2].]
will significantly affect both.

While many vibration isolation schemes exist, such as hanging springs,
rubber bands, viton stacks, air pistons, and magnetic eddy-current dampers,
all share two common characteristics:
1) the mechanical link between the STM and the environment (i.e. the
ground and/or pumps) is soft,
2) the mechanical link from tip to sample surface is rigid.
511
+-1"----+
Figure 6. STM schematic. illustration of major components in the STM. The I" mark
demonstrates the compact nature of the instrument. [From Ref. [2].]
As shown in Fig. 7, the STM can be thought of as two coupled oscillators: one is the spring with resonant frequency Wo1 and damping 11 upon
which the STM platform (mass ml) is connected to the ground; the other
is the tip-sample mechanical link which has a mass m2, resonant frequency
Wo2, and damping coefficient 12.
tip
surface
first stage
Figure 7. Schematic of STM vibration isolation. The first stage is often an optical table
with resonant frequency 1001 and damping 'Yl. Scan-head resonance occurs at 100 2, with
damping 'Y2.
512
For just a single oscillator whose base is shaken with vibration amplitude
A(w), the response amplitude, X(w), of the mass is
X(w)/ =
/ A(w)
(10)
Low-frequency vibrations (w << wo ) propagate, while high-frequency vibrations (w wo ) are damped. As such, the design goals for an STM are:
(1) to have Wol very low (a few Hz) so that the STM loop sees only low
frequency vibrations from the ground; and (2) to have Wo2 very high (1-2
kHz) so that the surface and tip move together in response to whatever remaining low-frequency vibrations pass through the first stage of isolation.
Overall, if the tip and surface move together in response to a vibration
from the ground, then the critical tip-surface separation is not significantly
affected.
1.2.3. Feedback and Control

The tip-sample separation in an STM is controlled by an integral feedback [2, 12, 13]. Error signals generated by changes in tunnel current are
integrated in time and fed to the z-piezo for correction, i.e.
V out
= G j(Vref - Vsig )dt

Ti
(11)
where Vsig is a voltage proportional to the tunnel current, Ti is the time constant of the integrator, and G is the gain of other amplifiers in the feedback
circuit. Because the current depends exponentially on tip-sample separation, Vsig is a highly non-linear function. As such, the control circuit will
often oscillate when attempting to compensate for changes in surface topography. Black et al. [2] have achieved a more stable feedback circuit with
a logarithmic amplifier which anticipates and accounts for the exponential
nature of Vsig. The error signal is then a linear function of the change in
tip-sample separation az,
V out
~ j(2K(az)dt.
(12)
1.3. MODES OF OPERATION
Upto now, we have only discussed the constant current mode of STM operation. In this mode, the tip height is adjusted during lateral scanning so
that the tunnel current remains constant. What results is a topographic
image, Z(z, y), of the sample surface.
513
z-
Another STM operation mode is constant-height mode. Here, the

feedback is disconnected so that the tip height is kept constant during
lateral scanning. Surface topography, 1(z, y), is measured by recording the
change in the tunnel current as the tip is moved. Because the z-piezo feedback is disconnected, the STM system can no longer correct for unexpected
changes in sample topography. Consequently, there is greater probability
of the STM tip "crashing" in this mode than in constant-current mode,
where the feedback is active. As a result, constant-current mode is the
more popular of the two.
1.4. SURFACE PREPARATION
The STM is primarily a surface probe, and as such, it is extraordinarily sensitive to the condition of the top-most layer of atoms on the sample surface.
Surface preparation is thus a crucial aspect of STM-based investigations.
Preparation of a clean, atomically-flat surface can be accomplished via
a number of ways. Perhaps the most desirable method is to cleave a crystal
in situ; thereby exposing a perfectly fresh, clean surface for investigation.
Unfortunately, problems with cleaving do exist: (1) only a limited number
of materials can easily cleave (such as TaSe2, NbSe2, and other transitionmetal dichalcogonides, as well as III-V semiconductors)j and (2) the cleaving process can result in a high concentration of defects. For those materials which cannot be cleaved, such as most elemental metals, the standard
cleaning procedure is to perform cycles of sputtering and annealing, the
former to remove dirty surface layers and the latter to re-order the rough,
sputtered surface. There exists a huge number of idiosyncratic recipes for
cleaning different surfaces-the ability to clean surfaces for STM imaging
is an "art."
1.5. APPLICATIONS
Having established a broad-based foundation of the STM, we now discuss a

number of exciting STM applications as they apply to mesoscopic electron
transport.
1.5.1. STM Spectroscopy
That the STM tip-sample contact region is nothing more than a nanometerscale tunnel junction suggests that one can employ the STM to perform local tunneling spectroscopy measurements. Indeed, such measurements via
the STM have been performed on a number of systems including GaAs [14],
Au particles [15], and atomically-fabricated Fe structures on Cu(I11) [18].
Here we discuss two recent examples of using the STM to perform local
spectroscopy measurements. [Other examples can be found in Chen [4], for
514
instance.] First, Black et al. used a low-temperature STM to measure the

electrical properties of a single Au particle of size ",1.5nm [19, 20]. They
obtained temperature-dependent IV curves consisting of Coulomb staircase steps. Such steps reflected electronic charging in the tunnel junction
system [19]. Although, in theory, they could have measured with the STM
the energy-level spectrum of one of the single Au particles, Black et al.
were unable to do so because the fabricated system was not adequately
stable [2].
A second example of STM-based spectroscopy is the work of DuBois
et al. in which the discrete density of states of a ligand-stabilized metal
cluster, Pt30gPhen36 0 30 , were measured [21]. By tunneling into a metal
cluster of size '" 1 nm, DuBois et al. measured an energy level splitting of
20 meVj theory predicted a splitting of 6.E ~ 4EF/3N ~8 meV (EF is the
Fermi energy of the cluster and N =14 70, the calculated number of electrons
on the cluster, itself [22]). Discrepancy between experiment and theory may
have been due to the actual value of N and to the tunnel barrier between
the cluster and the substrate (the latter of which varied depending upon
how the cluster lied on the substrate).
Although briefly summarized here, the experiments and results of the
Black et al. [19] and DuBois et al. [21] clearly demonstrate that the STM
is a powerful spectroscopic tool with which one can probe the energy levels
of nanometer-scaled conductors and devices.
1.5.2. Atomic Manipulation
A great source of recent excitement in the scanning probe community is
the STM's ability to position controllably a single adsorbed atom onto
a metal surface [23]. The ability to arrange matter atom-by-atom opens
many new possibilities for studying the electronic and structural properties
of clusters, defects, and chemical reactants on surfaces. In addition, it offers
new opportunities for exploring the ultimate limits of nanofabrication, that
of atomic-scale devices.
Current STM literature contains a multitude of studies involving the
atomic-scale modification of surfaces [24]. However, documented examples
of controlled and reproducible manipulation of single atoms are few. The
first truly controlled manipulation of single atoms reported was by Eigler
and Schweizer in 1989 [23]: they used a low-temperature STM to arrange
Xe atoms on Ni(llO). Since then, atomic manipulation has been performed
on only a select number of adsorbate-surface systems, including rare gases
and 3d-metal adsorbates on low-index metal surfaces [18, 23, 25] and, to
a lesser extent, on Si surfaces [26, 27, 28, 29, 30].
Single-atom adsorbate manipulation is performed using two main techniques: the sliding process and the (vertical) transfer process [23, 31, 32].
515
The sliding process involves bringing the STM tip so close to the adsorbate
that the attractive forces between the tip and the adsorbate are greater
than those between the adsorbate and the substrate surface. When this
occurs, the absorbate will slide with the tip as it laterally scans along the
surface. Note that at no point during the sliding process is the atom lifted
from the surface (as it does with the transfer process, to be described below) [31]. Because it requires stable, isolated monomers, the sliding process
can only be performed at low temperatures.
As opposed to the sliding process, the transfer process involves lifting
an adsorbate from the surface with a voltage-biased STM tip and then
depositing it to a selected region with the same STM tip, now biased with
opposite polarity [31]. The transfer process is species-dependent, i.e. only
Xe, and to a lesser extent Si(111) or Si(100), adsorbates have demonstrated
a willingness to be manipulated in this manner [26,27,28,29,30,32]. The
reason for this species-dependence is unknown at this time.
In general, it is not difficult to slide controllably a wide variety of adsorbates on a substrate surface. This opens the possibility of fabricating and
studying 2D atomic-scale structures; fabrication and study of 3D atomicscale structures, however, is presently very limited because the transfer
process is too species-dependent.
1.5.3. Quantum Corrals

An example of the power of STM imaging, atomic manipulation, and spectroscopy-all in situ-is the quantum corral work of Crommie, Lutz, and
Eigler [18]. Here, Crommie et al. used the sliding process to position Fe
atoms into a corral-like geometry on top of a Cu(111) surface. Fig. 8 shows
the process by which 48 Fe atoms were arranged into a circular quantum
corral with average radius 71.3 A, and Fig. 9, a clearer image of the corral
with surface-state electrons scattering within it.
To probe the energy dependence of the local density of states (LDOS)
inside their corral, Crommie et al. performed STM spectroscopy measurements at the center of the ring. As shown in Fig. 10, prominent peaks appeared in the dI/ dV spectrum. Crommie et al. have attempted to explain
qualitatively these peaks by modeling their corral as a 2D-hardwall circular
box with radius 71.3 A [18,33]. The peaks then correspond to the zero angular momentum eigen-energies of that box, and the quantum-interference
pattern observed within (see Fig. 9, the particle-in-a-box eigenstates [33,
34].
516
Figure 8. Construction of a quantum corral. These four images show the process of constructing a 48-atom Fe ring on Cu(I11), &I described in Ref. [18]. Each Fe atom is
individually positioned with an STM tip. The radius of the ring is 71.3 A.
Figure 9. Magnified view of the completed quantum corral shown in Fig. 8, lower right
quadrant. The "ripples" within the corral correspond to the Cu surface-state electrons
scattering of[ the corral walls and interfering with one another.
517
-....
....,
c:
0
....
,
1=0,
n= 1 2
Center of
48 Atom
Ring
>
"C
:::=
"C
L -_ _
-0.8
~~
-0.6
__
~~~~~
-0.4
-0.2
__
____
____
0.2
~~~
0.4
0.6
Voltage (V)
Figure 10. dI/dV spectrum (solid lines) taken with an STM tip held stationary over the
center of the corral shown in Fig. 9. Vertical, dotted lines correspond to the theoretical
eigenenergies of the 1=0 states of a 2D hard-wall box having a width of 71.3 A.
1.5.4. Summary
In summary, the STM is a powerful new tool with which we can both fabricate and noninvasively probe ultra-nanoscale devices in situ. While we
have concentrated on the positive aspects of the STM, we must comment
on some of its weaknesses. Perhaps the most limiting factor for STM operation is the need for a conducting substrate or surface. Imaging, atomic
manipulation, and spectroscopy all rely on electron tunneling between the
STM tip and the surface. In addition, lateral resolution is very much dependent on the sharpness of the tip. Finally, tip crashes are common because of
the STM's inability to adapt quickly to abrupt topography changes, despite
active feedback. Nonetheless, the STM is an unique probe which allows us
to control matter at extremely small length scales and to explore many new
exciting possibilities for next-generation experiments in mesoscopic physics.
2. Atomic Force Microscopy
In contrast to the STM where a tunneling tip is used to image a conductor's

surface, the AFM employs a mechanical tip to image an arbitrary surfacebe it conducting or insulating, hard or soft. Invented by Binnig and Quate
518
in 1986 [35], the AFM consists of a flexible, force-sensing cantilever tip

under which a sample is rastered. The tip interacts with the surface of the
sample via a variety of forces including short- and long-range interactions
and van der Waals forces. Using an active-feedback mechanism (much like
that found in an STM system), one can maintain a constant force between
the tip and the sample surface. Consequently, the cantilever tip will defiect
with respect to the changing sample topography and a topographic image
of the sample surface can be constructed.
Although the process by which the AFM images is, to first order, mechanical, atomic resolution is possible. Atoms are imaged, not pushed or
destroyed, by the tip during rastering because the cantilever is made with
a spring constant (k ...... 0.0006 to 2 N/m) that is far less than that of an
equivalent "spring" which binds atoms together in most substances (k ...... 10
N/m) [36].
Below we discuss the basics of the AFM and its various operation modes.
2.1. AFM BASICS
As shown in Fig. 11, all AFMs have five essential components [36]:
(i) A sharp tip mounted on a soft cantilever spring
(il) A way of sensing the cantilever's deflection

(iii) A feedback system to monitor and control the deflection (and hence
the interaction force)
(iv) A mechanical scanning system (based on piezodrivers) that moves the
sample with respect to the tip in a rastering fashion
(v) A display system that converts the measured data into an image.
Comparisons between the AFM and STM reveal striking similarities: the
piezodrives for zy rastering and z positioning, the active feedback control,
and the vibration isolation. Yet, the AFM and the STM do differ in one
crucial aspect: the STM is based on electron tunneling between the tip and
a conducting surface, and the AFM, the force interaction between the tip
and an arbitrary sample surface.
519
""""/; - - - - - - - {
'--
""""
Position-sensitive Photodetector
I
\
I
\
I
\
I
\
sample
feedback loop
0(
Image
PZTScanner
Figure 11. Schematic of an AFM. A sample is mounted on top of a tube scanner and
rasters beneath a cantilever tip. The force between the tip and the sample is detected
by optical means and is compared to a reference value. The signal difference is amplified
and used to drive a feedback circuit. A constant-force topographic image of the surface
is thus obtained. [Diagram courtesy of Park Scientific Instruments.]
The AFM can operate in three different modes: contact, noncontact,

and tapping.
(i) Contact Mode

In this mode, the tip directly touches the sample at all times during rastering. Consequently, strong, repulsive short-range interatomic forces ("" 10- 7
to 10- 11 N) dominate tip-sample interactions. Because the tip is always in
contact with the sample, the obtained images have extremely high resolution. Unfortunately, compression and shear forces generated between the
tip and surface can and do damage those samples which are soft and only
weakly attached to the substrate [37].
(ii) Noncontact Mode
In noncontact mode, the tip is held ,,"50-100 A. above the sample surface
during a scan. At these distances, weak long-range forces, such as magnetic,
electrostatic, and van der Waals forces, dominate tip-sample interactions.
In order to detect these forces, one vibrates the cantilever at a constant
frequency near its mechanical resonant frequency (wo= Jkolm ",100-400
kHz), with an amplitude of 10-20 A.. As the sample rasters beneath the tip,
both the cantilever resonant frequency and vibration amplitude will change
520
in response to force gradients, dF j dz, which vary with the tip-sample distance. For example, if the force gradient is positive, then the cantilever's
effective spring constant will decrease, keff = ko-(dFjdz), and correspondingly so will the resonant frequency and amplitude. An image representing
surface topography is obtained by monitoring these changes in vibration
amplitude.
Overall, because the cantilever tip never touches the sample surface in
noncontact mode, soft samples such as polymers and biological materials,
can be imaged. It is important to note, however, that this mode is difficult
to employ because the tip is easily captured by adhesive forces at the surface
(see below) [36]. In addition, because of the relatively large tip-sample distance, the resolution is not comparable to that obtained with contact mode.
(iii) Tapping Mode

This particular mode is similar to noncontact mode, except that the vibrating cantilever tip intermittently touches the sample. As in noncontact mode,
in tapping mode, the cantilever vibration amplitude changes in response to
force gradients which vary with tip-sample distance; by monitoring these
changes, one can image the surface topography. For some samples, tapping
mode is preferable to full contact mode because it eliminates lateral forces,
such as friction and drag, that might damage the tip or sample [37].
Based on our outline of the three different AFM operation modes, tipsample interactions may appear to be straighforward: they are either dominated by strong, repulsive short-range forces or by weak long-range forces.
This, however, is not at all the case. Many environmental factors can complicate tip-sample interactions. For instance, when any sample is scanned
in ambient environment, it will have a few monolayers of adsorbed water.
Such monolayers can generate a large, attractive capillary force (f'V 100-300
nN) that can dominate any short- or long-range force. In contact mode,
the capillary force can pin the tip to the surface, placing a limit on the
lowest possible stable imaging force. In tapping mode, it can put a lower
limit on the amplitude of oscillations required to prevent the capture of the
tip. Finally, in noncontact mode, the capillary force can prevent imaging
altogether. One way to minimize the capillary force 10 nN) is to decrease
the ambient humidity or to image in liquid, in which case, tip-sample interactions are dominated by smaller van der Waals and electrostatic forces
(typically 0.1-1 nN) [37].
Another factor which can greatly affect tip-sample interactions is the
cantilever's own thermal vibrations [36]. Such vibrations can limit the detectable force gradient dFjdz by as much as '" 10- 5 N. In addition, the
thermal-vibration amplitude, determined by the equipartition theorem and
521
given by !kO((~z)2) = (1/2)kBT, can significantly affect atomic imaging. Consider a 1 N/m cantilever at room temperature: its rms thermalvibration amplitude is a rather large ",,0.6 A [36]. Clearly, thermal vibrations are significant and should be avoided by operating the AFM at low
temperatures.
2.2. CANTILEVER DEFLECTION DETECTION
Crucial to the AFM is the ability to detect cantilever tip deflections. Although a number of detection methods have been devised, the most common is optical beam deflection (see Fig. 11). This method uses a laser beam
reflected off the back side of the cantilever onto a position-sensitive foursegment photodetector. A cantilever deflection of 0.01 nm translates to a
laser beam displacement of 3-10 nm at the photodetector, large enough to
generate a measurable voltage.
Although very sensitive to cantilever deflections, optical beam deflection does require the use of precisely aligned optical components. This
introduces complications in the design and operation of any AFM, particularly those operating at low temperatures or under UHV conditions. Consequently, an alternative method-that of using a piezoresistive cantileverhas been developed [38]. Fabricated out of boron-doped silicon, a piezoresistive cantilever takes advantage of the piezoresistive effect of the material
from which it is made: the resistance will vary according to the strain induced (by the cantilever deflections). By running a current through the
cantilever, one can monitor the change in resistance and in turn, the cantilever deflection. The sensitivity of piezoresistive cantilevers is equivalent
to those used in optical beam deflection.
2.3. RESOLUTION
Despite their ability to achieve high resolution, AFMs are not free of scanning probe-sample interaction problems that bedevil other metrology tools
[39]. Because of uncertainties in tip position and tip geometry, errors in
measurement are quite common. Ideally, cantilever tips should reflect the
topography of the surface to be scanned; this, however, rarely happens.
Fig. 12 demonstrates in detail how tip shape can affect the scan trace
and, therefore, the apparent shape of the surface. As shown in Fig. 12a,
the tip is so blunt that it cannot reach the bottom of the trench. Consequently, a cusp is imaged, giving the impression that the tip is resolving
much smaller features than is actually the case. Because convolution is a
linear operation and an AFM scan is not, the actual trench shape, and in
turn the trench depth, cannot be deconvolved from the AFM image [39].
522
Scan Line
(a)
(b)
Fipre 12. Two wayl in which probe lhape can affect the Icanned imqe. (a> The tip
is too blunt to reach the bottom of the trench, and thus, the trench feature. cannot be
extracted from the image. (b) The cone angle of the tip hide. the true aqIe. of the lide
walls. [Adapted from Ref. [39].]
Problems of mapping inaccessible areas such as that shown in Fig. 12a have
been addressed by Niedermann and Fischer [40] and Reiss et al. [41].
In Fig. 12b, we show a tip that is sharp enough to reach the bottom of
the trench; the image obtained accurately corresponds to the trench depth.
Unfortunately, the apparent sidewalls of the image re:H.ect the tip cone angle and not the sample. For cases such as this, a special tip whose shape is
optimized for a particular part of the sample is necessary. Various fabrication techniques, including focused-ion milling and electrochemical etching,
are currently being explored to sharpen and shape already existing siliconnitride tips (see Griffith et al. [39], Ximen et al. [42], and Nyssonen et
al. [43]).
Figure 13. Atomic re.olntion of mica in ambient environment. The field of view is 80
[Original image courte.y of Park Scientific Instrument-.]
A.
523
Despite these limitations, atomic resolution using the AFM is still possible (see Fig. 13). The probe must be extremely sharp with a well-defined
tip at its end. [Paradoxically, it is usually a small cluster of atoms on an
irregular tip that leads to atomic resolution [39].] In addition, it must be
light and flexible and have a resonant frequency high enough to follow
the contour of the surface. Finally, tip stiffness must differ in the vertical
and horizontal directions so that frictional lateral forces, which can lead
to image artifacts, can be minimized. Exact details for achieving atomic
resolution can be found in Chen [4]. That the AFM can achieve atomic
resolution on a number of different materials, including both insulators and
conductors, sets it apart from the STM which can only image conductors.
2.4. APPLICATIONS
Although the AFM is a relatively new instrument, its simplicity and its
ability to examine a wide range of materials including both insulators and
conductors have made its applications far more reaching than those of the
STM. In the following sections, we highlight only a few of the growing AFM
applications.
2.4.1. Lithography
As previously mentioned, scanning probes in general are capable of performing sub-50 nm lithography. We have already seen how the STM can perform
the "ultimate" lithography-that of atomic manipulation. The AFM can
also perform precise lithography, but over a much larger area extent on a
number of insulating or conducting materials. It therefore is not suprising that attention has focused on the AFM, as opposed to the STM, for
lithography.
There are a number of current AFM-based lithography techniques, including direct deposition/chemical modification [44, 45], low-voltage resist
exposures [46], plowing [47], machining [48], and tapping [49]. Each one of
these techniques are capable of producing structures that are on the sub-50
nm length scale (see Marrian [50]). Yet a number of distinct advantages
hinder each one of these technqiues from general use. First, a conductive
tip and/or substrate is needed for direct deposition/chemical modification
or low-voltage resist exposure [44, 45, 46]. Second, extremely thin resists
(",25 nm-thick), subject to pinholes and nonuniformity on a topographically varying substrate, are necessary for low-voltage resist exposures [46].
Third, for all methods listed, the resulting nanostructure's height-width
aspect ratio is extremely poor (at best 1:1). Such poor aspect ratios place
524
severe limitations on device design.

Below, we highlight two very recent AFM-based lithography techniques
which have attempted to circumvent these problems.
(i) Bilayer-Resist Scheme

Sohn and Willett [51] have developed an AFM-based lithography technique which continues to demonstrate its flexibility in directions that the
aforementioned techniques fall short. Their technique, which employs the
bilayer-resist scheme commonly used in electron-beam lithography [52], involves plowing through the first of two resist layers (the second resist has
been previously flood-exposed with e-beam lithography) with a standard
silicon nitride (ShN 4 ) tip and performing subsequent development, metaldeposition, and lift-off steps. Fig. 14 displays a characteristic result: 40
nm-wide, 50 nm-thick chromium (Cr) wires. The lines are very straight
and parallel (this holds true for the entire length of the wires, ",15 pm),
thus indicating that the tip is not subject to any appreciable noise during
plowing. In addition, all lines have smooth edges and are uniform in width,
suggesting that an appropriate undercut forms during the development step
and that the tip maintains a constant force over the substrate and neither
degrades nor accumulates any appreciable PMMA residue while plowing
through the top-resist layer.
Figure 14. ScamUng electron micrograph of ",,40 nm-wide, ",,50 nm-thiclt Cr wes fabricated on a GaAs substrate. Fabrication parameters include: F=1100-1300 nN, development time r=5s, top-resist layer thiclmess=25 nm. [From Ref. [51].]
525
A unique aspect of the Sohn and Willett technique is its ability to create
continuous structures over varying topography and over large distances [51].
This is demonstrated in Fig. 15a which shows 40 nm-wide, 50 nm-thick
Cr wires fabricated between 50 nm-thick gold (Au) contact pads set 3 p.m
apart. [Alignment of the wires and pads is achieved by positioning the AFM
tip with an optical microscope.] As shown in Fig. 15b, all the wires for this
simple device are continuous and uniform in width over the edge of the Au
contact pads (continuity was confirmed by measuring a resistance R=2 kO
across the wires). How is this possible? It is the thick bottom-resist layer
which planarizes the substrate and thus makes fabrication over varying topography successful.
Figure 15. 15 I'm-long, ",40 mn-wide, ",50 mn-thick Cr wirel. (a) Fabricated with the
lame parameters as thole in Fig. 14, wirel traverle 50 mn-thick Au contact pads let 3
I'm apart on a GaAs substrate. Resistance of wires R=2 kO. (b) Magnified SEM of wirel
at the boundary of the Au contact pads and GaAs substrate. [From Ref. [51].]
Overall, the AFM-based lithography technique developed by Sohn and

Willett [51] has a number of distinct advantages. The generality of the presented technique is important; the good resist profile obtained allows for
many materials now to be deposited with small lateral length scales. By
planarizing the substrate surface with a thick bottom-resist layer, one can
use an arbitrary, topographically varying, substrate. With the bilayer-resist
scheme, one can fabricate nanostructures with excellent height-width aspect ratios (roughly 3:1). This particular advantage allows one to consider
practical device architecture. Finally, because plowing is performed only on
the top-resist layer, the substrate is not damaged. Likewise, the tip is left
undamaged because plowing is performed on unbaked resist; the lifetime of
the tip is indefinite so long as catastrophic failures such as tip "crashes" do
not occur. At present, line width is limited to ",40 nm; smaller line widths
are possible if sharper tips, such as ion-milled ones or even carbon nanotubes [53], are used.
526
(il) Local Anodization

The bilayer-resist scheme we have just outlined is simple and flexible;
yet, as we have just stated, the line widths obtained are limited to ",40
nm. In contrast, we now describe a very recent AFM-based lithography
technique-that of anodic ozidation-which can produce nanostructures
far smaller: ",10 nm. Anodic oxidation is a process in which a voltage
is applied to an AFM tip in contact with a metallic surface in ambient
humidity, resulting in local surface oxidation [54]. By performing real-time
in situ measurement of device properties during critical fabrication steps,
one can control the amount of oxidation to be applied.
Snow and Campbell [55] have used anodic oxidation to fabricate singleatom point contacts. Their method is the following: An AFM tip, continuously biased with increasing voltage, is scanned repeatedly across a 40
nm-wide, 500 nm-Iong constriction in the middle of a 1 pm. AI wire (see Fig.
16. As the tip voltage is increased, the oxide which forms penetrates deeper
into the AI film, shutting off parallel conduction paths and increasing the
wire resistance. When a predetermined resistance value (for instance 2-2.5
x 103 n) is achieved, the tip voltage is set to zero. Fig. 17 shows an AFM
image of a finished device and Fig. 18 plots the time dependence of the conductance for the device immediately after the tip voltage is set to zero. As
shown in the time record of the conductance, anodization leaves the sample
in a nonequilibrium state [55]. Additional time is necessary for the device
to stabilize; it is during this time that further oxidation takes place and
closes the remaining conduction paths [55]. These remaining paths form
the atomic-sized point contacts which are indicated by the discrete steps
in conductance '" 2e 2 / h shown in Fig. 18.
Anodic oxidation is a novel and effective means of fabricating sub-10
nm length-scaled devices. Successive anodization with an ion-milled tip
can produce structures as narrow as 3 nm. Unfortunately, nonuniformities
dues to the grain size of metals is a major drawback to this technique [55].
Nonetheless, as is particularly germane to mesoscopic electron transport,
anodic oxidation can be used to control the tunnel barrier resistance and
size of tunnel junctions [54]. Nakumura et al. [56] have already shown that
gross anodization of an entire SET device can shrink the dimensions of the
device and consequently raise its operating temperature. If selective AFM
anodization were applied to the same system, the junction areas could be
reduced independently, producing two ultrasmall, low-capacitance tunnel
junctions [55]. Selective anodization of these structures via the AFM may
thus provide an additional level of control in fabricating nanodevices.
527
Feedback Control
Figure 16. Diagram of the fabrication circuit. The electrical bias on the AFM tip anodically oxidizes the Al metal film. The current is measured through the device and used
as a feedback to control the anodization. [From Ref. [55].]
10
Figure 17. (a) A 1.5 p.m by 1.5 p.m AFM image of a point-contact device. The image
shows a 1 p.m-wide Al wire, the 40 nm-wide constriction defined by the AFM-patterned
oxide, and the oxide that forms the point contact. [From Ref. [55].]
528
T=300K
4 ............................................................................................................................
Three Atom Channel
~~
3 r".,-.......'"-.::J-~;;,;.:.:.:.:~c ...............................................
B
Two Atom Channel
~::s
'a
o
(.)
1 ......................................................
OneAlomChannef:=;: .............
O~~--~--~--~--~~~~--~--~~
00
00
1001~1~1OO100200
Time (Sec)
Figure 18. Conductance as a function of time immediately after the Up voltase ia let to
sero. The large conductance steps correspond to the closing or atomic-siaed conduction
paths. [From R.ef'. [55].]
2.5. CHEMICAL FORCE MICROSCOPY
Perhaps one of the most clever applications of the AFM has been its use in
identifying the spatial arrangement and interactions between chemical functional groups on the molecular level. Employing the AFM in such a manner
is now referred to as chemical force microscopy. Although optical tweezers
can probe the interactions between two molecules [57] and surface force
probes, the adhesive and frictional force between molecular groups [58],
these methods do not have the ability to map the spatial amzngement of
the very functional groups which give rise to site-specific interactions and
reactions [59, 60]. Such information plays an important role in molecular
assemblies-a future direction of mesoscopic electron transport.
(i) Spatial Mapping of Functional Groups
One clear demonstration of chemical force microscopy is the work of
Frisbie et al. [59]. Here, the spatial arrangement of hydrophilic (-COOH)
and hydrophobic (-CH3) functional groups of an organic monolayer was
529
mapped by measuring the frictional forces between the organic monolayer

and a self-assembled monolayer (SAM)-coated AFM tip. The ability to
distinguish between hydrophilic and hydrophobic functional groups is important since the interactions between and among such groups correspond
to the must fundamental type of forces that can occur between molecules
and molecular assemblies [59, 60].
In detail, Frisbie et al. used standard Si3N4 tips that were coated with
a SAM terminated with either -COOH or -CH3 functional groups. Once
coated, the tip was placed in contact with a patterned sample (see Fig. 19),
itself terminated with -COOH and -CH3 groups. Adhesive forces between
the tip and the sample were measured by monitoring the deflection of the
cantilever as the sample approached, contacted, and withdrew from the
probe tip. The forces between the functional groups of the tip and sample varied in magnitude, COOH/COOH > CH3/CH3 > COOH/CH3 (see
Fig. 20), reflecting that fact that interaction via hydrogen-bonding between
hydrophilic groups is stronger than interactions between hydrophobic-hydrophilic groups and hydrophilic-hydrophobic groups [59].
By knowing and distinguishing the adhesive forces between functional

groups, one can then spatially map functional groups. Fig. 21 demonstrates
this [59]. Here a SAM has been lithographically patterned with -CH3 and
-COOH functional groups (see Fig. 21a). Surface topography via an AFM
shows relatively featureless topography (Fig. 21b )-this is expected since
the -CH3 and -COOH regions of the sample are structurally similar. As
shown in Fig. 21c, when a CH3-terminated tip is used for imaging, the
CH3-terminated regions on the sample exhibit more friction (because of
the high CH3-CH3 adhesive forces) than those regions terminated with
COOH. In contrast, when a COOH-terminated tip is used to image the
same sample (Fig. 21d), the friction contrast is reversed: a lower frictional
force is observed over the CH3-terminated regions of the sample, and a
high frictional force over those regions terminated with -COOH functional
groups.
The results of Frisbie et al. demonstrate that it is possible to map,

with fair accuracy, the spatial distribution of chemically-distinct functional
groups on a sample surface when functionalized tips are used [59]. The
limiting resolution of this technique can be approximated using the JKR
model [61] where the contact radius at "pull-ofi"", as, for surfaces terminating in the same functional group is
530
TIp Preparatlon
Self
Assembly . .
Chemical Force Microscopy
.cH s
Figure 19. (above) Modification of the cantilever-tip assembly with a specific functional
group. Gold-coated S~N4 cantilever-tip assemblies were immersed in a SAM terminated
with either -CH3 or -COOH functional groups. The specific case of a tip terminating
with -COOH groups is shown. (below) Schematic views of the experiments. Inset: Interactions between a tip terminating in COOH groups and a patterned sample terminating
in both -CHa and -COOH groups. [From Ref. [59].]
COOH-C~
C~-C~
COOH-COOH
1 nN
o
5
10
15
20
Z displacement (nm)
25
Figure 20. Typical force vs. displacement curves Frisbie et 01. obtained between a COOHterminated tip and sample, a CB:,-terminated tip and COOH sample, and a CHaterminated tip and sample in ethanol. [From Ref. [59].]
531
Figure 21. Optical condensation iJnase of a patterned IWIlple. The bright, raised areas in
this image correspond to H2 0 that haa condenaed on the COOH-terminated regions of
the sample. (b) Force microscope image of the topography of the sample surface pattern
as in (a) recorded with a probe tip terminated with CHa functional groups. As expected,
there is little difrerence in the topography observable between the CHa- and COOHterminated regions of the sample. (c) Image of the friction force recorded simultaneously
with the topography in (b). The bright regions correspond to a high friction force, whereas
the dark regions correspond to a lower friction force. In this image, which was recorded
with a CHa-terminated tip, high friction is observed over CHa-terminated regions of the
sample. (d) Image of the friction force recorded between a COOH-terminated tip and a
similar region of the sample as in (a) through (c). In this image, high friction is observed
over COOH-termianted regions of the sample. Images (b) through (d) were acquired
under ethanol with an applied load of 4 nN and a scan rate of 3 HI. [From Ref. [59].]
532
(13)
Here, K is the elastic modulus of the tip and sample. Both Frisbie et al. [59]
and Noy et al. [60] have estimated that the adhesive and frictional forces
they have observed using the just-described technique correspond to only
15 molecules on the tip and sample. If the AFM tip radius were reduced
to ",,10 nm, then the contact area at pull-off would correspond to only a
single molecular pair [60].
(ii) Local Chemical Catalysis

Recently, Miiller et al. [62] have demonstrated that an AFM tip can
be used to selectively modify molecular groups to generate more complex
structures. In particular, they haved used aPt-coated AFM tip to induce
catalytically hydrogenation of local azide groups to amines on a SAMcoated substrate. Fig. 22 demonstrates the process undertaken, and Fig.
23, the results.
Figure 22. Scanning with a platinum-coated AFM tip over a SAM surface containing
terminal azide groups in the presence of H2 leads to the reduction of azide groups to
. primary amino acids. [From Ref. [62].]
533
Figure 23. Florescence microsraphs of surface uide groups derivatiled after scannjng
under different conditions in the AFM. (a) A 10 pm by 10 pm area wu scanned with a
Pt-coated AFM tip in hydrogen-aaturated isopropanol and derivatiled with fluorescent
aldehyde-modified latex beads for labeling. The latex beads will only bind to primary
amino groups. (b) A 10 I-'m by 10 I-'m area scanned as in (a) but derivatiled with ATTOTAG reagent which also binds to only primary amino groups. (c) A 10 I-'m by lOl-'m
area scanned with a silicon AFM tip under the same conditions as in (a) and (b) and
derivatized with ATTO-TAG reagent. [From Ref. [62].]
The significance of Miiller et al. 's process is largely two-fold: first, because catalysis rather than electrochemistry is employed, a large and diverse
number of surfaces can be used for local chemistry; second, by varying the
nature of the catalyst and the chemical composition of the surface, one can
synthesize molecular assemblies not readily produced by existing nanofabrication techniques [62]. Clearly this process opens new pathways toward
studying mesoscopic electron transport in novel materials and device architecture. The ultimate resolution of the Miiller et al. technique is unknown;
however, improvements may be made with tip morphology, reaction conditions, and the nature ofthe surfaces employed [62].
2.6. AFM ELECTRON SPECTROSCOPY
Thus far, we have reviewed AFM applications within the context of fabrication, spatial imaging, and local chemical catalysis. In this section, we
will highlight AFM electron spectroscopy. In particular, we will examine
how Eriksson et al. employ an AFM to study the current flow through a
ballistic point contact [64].
Eriksson et al. use an AFM in two modes simultaneously. In one mode,
the surface topography is measured; in the other, the resistance of the
point contact is monitored while a conducting AFM tip locally modifies
the sheet density beneath the tip via capacitive coupling. Measuring the
534
change in resistance of the point contact in response to this perturbation

allows details concerning the current flow through the point contact to be
extracted from the images. The topographic and electronic data are easily
correlated because the two measurements are made simultaneously. Because
the AFM is functioning in a standard mode throughout the experiment, the
tip position is well controlled at all times.
2.6.1. E:r:perimental Details
As with all semiconductor samples, the carrier concentration in semiconductor microstructures is easily modified by exposing the sample to light.
Although useful in many circumstances, this property precludes the use
of either optical beam deflection or laser-based interferometry as methods
for detecting the deflection of the cantilever. Thus, Eriksson et al. employ
commercially available piezoresistive cantilevers [65] to sense tip deflection.
(a)
Figure 24. Schematic of the experiment. (a) Sample is a balliatic point contact in a
GaA8/AlGaAs heterostructure, as described in the text. (b) Circuit diagram: b is used
to measure the cantilever rcaiatance, and h is used to modulate the 2DEG sheet density
beneath the tip.
535
So that a long mean free path in a degenerate 2DEG is achiever, the
experiment is performed at 4.2 K. This imposes additional constraints
on the experimental apparatus [66]-details of which can be found elsewhere [64, 67].
Fig. 24a illustrates the experiment. By applying a voltage between the
cantilever tip and the 2DEG, Eriksson et al.locally modify the sheet density
directly beneath the tip. A simultaneous measurement of the resistance of
the point contact determines the effect of the sheet density modification
on transport through the point contact. The width of the change in sheet
density determines the resolution of this method. It has been previously
shown that for low tip-sample voltages the half-width at half-maxjmum.
(HWHM) of the modulation should equal the electron gas depth d, as long
as the contact area between the tip and the surface is smaller than d (d=520
A in this experiment) [67]. For voltages larger than the depletion voltage,
the HWHM of the perturbation will be larger than d. An effective depletion
width of 600 A fits the data in this experiment. This length compares
favorably with characteristic lengths such as the Fermi wavelength AF=420
A in this material.
The sample used in this experiment consists of a single ballistic point
contact formed by a mesa etch in a GaAsl Alo.3Ga.o.7. A heterostructure
containing a 2DEG in a remotely doped accumulation layer is located 520
Abeneath the surface. The sheet density is ns =3.0xlO ll em-2, the mobility,
1'=540,000 cm2 /Vs, and the electron mean free path, 1=5.0 pm. The width
of the point contact is 2.95 I'm, and its resistance, after subtraction of the
lead resistance, Rpc 100 n.
Fig. 24b shows the actual circuit used to make the measurement. The
cantilever serves the dual purpose of monitoring the tip deflection and applying a voltage to the tip. To measure the cantilever resistance (and monitor the deflection), Eriksson et al. place the cantilever in a Wheatstone
bridge, ac-biased at frequency II. To apply a voltage to the tip relative to
the sample, they hold the bridge and thus the cantilever at a second voltage by applying Vae to the center tap on the transformer at frequency h.
The point contact is dc biased with a current Ide' By detecting the voltage
across the sample at frequency h only the change in resistance is measured.
2.6.2. Results and Analysis

Fig. 25 shows an image of the change in point-contact resistance t1R induced by the voltage-biased AFM tip as it is scanned over the point contact.
The surface topography is obtained simultaneously and the outline of the
point contact is shown as the white lines in Fig. 25. As expected, the sheet
density modulation has a much larger effect when it is inside the point
536
contact than when the tip is far away.
2Q
OQ
Figure 25. Image of the change in point contact resistance. The white lines indicate the
edge of the etch trenches defining the point contact. Tip voltage is 0.5 V rm., and the full
color scale range is 2 O.
In order to analyze the data in Fig. 25, Eriksson et al. have presented
the following model. Because the mean free path is l = 5.0 /Lm, electron
transport is in the ballistic limit, and a scattering model is appropriate.
The model assumes that ballistic electrons are scattered by the localized
change in sheet density directly beneath the tip. When the tip is positioned
such that the scattering decreases the transmission coefficient of the point
contact, a large change in point contact resistance 6.R is observed. When
the tip is positioned so that the additional scattering of electrons does not
affect the transmission through the point contact, 6.R is negligible. A large
signal is expected only when a large fraction of the current flows near the
tip, which is the case only in the point contact itself.
The change in resistance 6.R is found to be proportional to Vac. From
the measured peak resistance change in the point contact Ro ,. . ., 0.02Rpc
the estimated effective change in diameter of the density perturbation is
wRo/ Rpc ,.....,600 A, where w = 2.95p.m is the point contact width, assuming
that the electron gas is fully depleted under the tip.
Fig. 26a is a semi-log plot of the normalized resistance change I:lR/ Ro
vs. distance :I: along the line perpendicular to the point contact (see inset).
As shown, the resistance change caused by the density perturbation decays
exponentially with distance away from the point contact. The decay length
is 2.8 p.m. In order to change the transmission coefficient of the point con-
537
tact, electron trajectories passing under the tip must also traverse the point
contact. Therefore the decay length should be one half the mean free path l,
because the round trip distance between the density perturbation and point
contact is 2:1:. The mean free path inferred via this argument from the decay
length is 5.6 JLIn, which is comparable with the mobility measured mean free
path l = 5.0 JLIn, a sample-averaged value. This simple argument neglects
an algebraic correction to fiR due to angular spreading of the electron flow.
(a)
1111iiii:::-----,r-----,--"-T---,..--_
~ 0.1
<l
0.01
10
x (I-I.m)
(b)
# ....
0.5
0.0
-60
,,
1.0

-40
-'
-20
-...
e (degrees)
20

40
60
Figure 26. (a) Line Ican extracted from Fig. 25, as indicated in the inset. (b) Data along
a circular arc from Fig. 25, again as indicated in the inset.
Fig. 26b is a plot of the normalized resistance change fiR/ Ro vs. angle ()
along an arc 4 p.m away from the center of the point contact (see inset). The
resistance change shows a pronounced collimation of the ballistic electron
flux perpendicular to the point contact: fiR is strongly peaked at () = 0,
and has a HWHM of fi() = 20. Collimation of the ballistic electron flux
through point contacts due to flaring of the entrance and exit is a well
538
known phenomenon previously studied by using multiple point contacts

and by steering electron trajectories with a magnetic field [68, 69, 70].
In summary, scanning probe microscopy offers a direct probe of the spatially varying features of transport through mesoscopic devices. By studying
the ballistic electron current through a point contact, Eriksson et al. provide images of the point contact collimation and the local effect of a finite
mean free path, properties difficult to measure directly without scanning
probe techniques.
2.6.3. Summary
As we have seen in this section, the AFM is a much more versatile instrument than the STM. The ability to examine, probe, and manipulate both
conductors and nonconductors truly sets the AFM apart from the STM.
We comment that despite these positive aspects, the AFM does have an
"Achilles Heel." Like the STM, the AFM needs extremely sharp tips to
perform well; at present, the technology for fabricating sharp tips is not as
reliable as one would like it to be. Nonetheless, the AFM is still a powerful
and versatile tool for lithography, local chemistry, and spectroscopy.
3. N ear Field Scanning Optical Microscopy
A variety of applications of near field scanning optical microscopy, NSOM,

are reviewed in this section. NSOM's ability to optically resolve features
smaller than the diffraction limit presents a potential for wide ranging impact both for local modification and high-resolution characterization. In
finding the practical applications for NSOM, however, it is equally important to take into account its limitations of slowness, of its serial process,
fragility of the probe, and short depth of focus.
3.1. NSOM BASICS
Near field scanning optical microscopy (NSOM) has recently proved itself
through a number of applications to be a powerful new tool in the laboratory. NSOM exploits the optical interaction arising from a sharp probe
in close proximity to a sample in order to image surfaces on a scale well
beyond the classical diffraction limit. Typically, this probe consists of an
optical fiber that terminates into a fine truncated cone. The sides are made
optically opaque with an aluminum coating, but, the tip itself has a small
uncoated aperture through which the light can pass. Aperture diameter
commonly ranges from A/2 to A/40, depending on the desired tradeoff between resolution and light throughput. This locally confines the radiation
field to a spot smaller than that achievable with far-field diffraction-limited
optics. The aperture is freely positioned or scanned across the surface allow-
539
ing highly localized exposures of the sample or generation of high resolution

images. Fig. 27 shows a typical configuration of the NSOM which has been
used for local spectroscopy of quantum wells. Here the tip was 'A/4.5 allowing a throughput of a few nanowatts of light .
"---...~.
.....
100 nm
_-...---
" -
.. -
- - -
.-
- .
_ _ "
.- '.
- _
0 -
0 .'
Figure 27. Cutaway schematic of NSOM system used for near-field photoluminescence
microscopy and spectroscopy of quantum wells. The probe and the sample portions are
drawn to scale.
As with conventional far-field optics, nanotechnological applications fall

into two broad classes: surface modification/lithography and characterization/metrology. Below, we will consider each of these two classes of applications as they apply to NSOM. In addition, several topics from a more
comprehensive review [71] are described here along with a sampling of more
recent highlights.
3.2. SURFACE MODIFICATION AND NANOLITHOGRAPHY
The potential advantages of NSOM as a modification tool include its use
with existing optical resists, its ability to image simultaneously with exposure, and the possibility of closed-loop control of this exposure by monitoring the resultant latent image. Fig. 28 (from [71]) shows lines with ",100 nm
line widths and spaces that have been written by NSOM in a 50 nm-thick
film of conventional photoresist and developed with a standard photoresist developer. The contrast produced is produced solely by changes in the
coupling of light into the aperture caused by the changing refraction in-
540
dex. The primary disadvantages of NSOM are its limited depth of focus
and its limited speed, arising from both the mechanical scanning aspects
and the relatively low-photon flux involved. Line width is limited to resistthickness, graininess, and probe height and diameter. Therefore, barring
the development of massively parallel arrays of efficient near-field probes,
the potential uses of NSOM for modification will be in the laboratory research and development rather than production setting.
Figure 28. NSOM-written patterns on photoresist.
One possible exception is in the area of high-density near-field optical data

storage [72], which in many aspects represents a marriage of magnetic and
optical storage technologies. Fig. 29 (from [72]) shows how NSOM has been
used to both image and record domains in thin-film magneto-optical materials such as CoPt. Writing occurs when the radiation from the aperture
locally warms the magneto-optical material to the Curie point, allowing
magnetic domains to flip. Polarized light incident on these areas is rotated by the local magnetization, enabling NSOM readout. Data densities
of ",45 Gbits/in2 have been achieved, well in excess of current technologies.
If NSOM is to become a viable technology a crucial issue ofread/write speed
must be addressed. Recent development of a high-flux near-field fiber laser
probe [73] further highlights the potential of this particular application.
541
Figure 29. Magnetic domains recorded by near-field optics (top), compared with bits
recorded by conventional means.
3.3. CHARACTERlZATION
In the area of nanostructure characterization, the applications are more
broadly based, because NSOM can take advantage of the numerous and
powerful contrast mechanisms of conventional optical microscopy. For instance, the example of the previous section illustrates the use of refraction
index and polarization contrast. Reflectivity, absorption, and spectroscopy
are on a growing list of contrast methods. Use of the NSOM probe as a
subwavelength optical detector has permitted the characterization of mode
profiles of semiconductor lasers [74J, and near-field photoconductivity measurements on such lasers have resulted in maps of the p-njunctions therein.
Recent demonstration of single molecule detection with NSOM [75] and in
particular the ability to determine the dipole orientations of individual
flourescing chromophore molecules, as illustrated in Fig. 30 (from [75])
may be of some aid in the growing field of nanofabrication via molecular
self assembly. The NSOM probe might also offer interesting opportunities
for metrology, where position monitoring can be accomplished via the opti-
542
cal signal [76], possibly using the optical path as a leg of an interferometer.
... .
\
'\
'II
..,.
"
.,
'~
.
..
"
.
,
..
,.
" "
", I
..
'"
....
"
.
\ ...
Figure 30. Near-field optical image of individual dye molecules (left) with orientations
inferred from the observed patterns (right),
Finally, low-temperature near field scanning optical microscopy j spectroscopy has proved useful in characterization of quantum-confined semiconductor structures. In a study of cleaved-edge overgrown quantum wire laser
structures, the NSOM spatially resolved the three regions of different luminescent spectra and imaged how local exciton diffusion influences their spatial extent [77]. This has recently been extended to revealing the individual
optically-active quantum constituents of quantum wells [78]. Specifically,
sharp 0.07 meV), spectrally-distinct emission lines of a GaAsj AlGaAs
quantum well can be imaged at a specific spatial location, or as a spectral
evolution image as the probe is scanned along a line across the surface, or
as a real space image at a specific luminescence wavelengths, see Fig. 31
(from [78]).
543
00
averaged ....."... uulll'l
80 - near-ffeld spectru
... far-field spectrum
60
",
"
40
I
" .... ~."
4
,', it
....., "'... -:.... . '

-
"
I'.' ... . . . '
IT'~. , ....... ' ;
't
;.
..
: . .
;
."
,='
g
'e
..,.
cD
Figure 31. (Top) Near-field and spatially- averaged photoluminescence spectra from a
23 A single quantum well. (Middle) Spatial evolution of the near-field spectrum with
linear position, presented as the image L(i, z )111. (Bottom) Real-space near-field images
L(z,1I)li at the five wave lengths labeled above. Luminescence is found to originate only
from certain spatially and spectrally discrete sites, which appear as rings.
Temperature, magnetic field, and line width measurements establish

that these luminescence centers arise from excitons localized at interface
fluctuations. In other words the excited volume is so small that one is resolving the individual quantum components of an inhomogeneously broadend distribution of photoluminescing states. For sufficiently narrow wells,
virtually all emission originates from such centers. One can think of such
centers as bulges or caverns of a thicker quantum well that can capture the
excitons and define its energy. The local spectral shifts can allow effective
well thickness variations to be imaged. This is possibly the most immediately useful development in the quest for a new monitors of optimal crystal
growth conditions. Quantities such as diffusion can be measured on a site
from the size and shape of the luminescent ring surrounding the center (see
Fig. 31 bottom). Larger rings, (Le. diffusion lengths) are seen in thicker
quantum wells and indicate that the position of the tip where the exciton
is created can be further away from the center and still have the exciton
migrate to the center for luminescence. Another quantity, the lateral confinement, can be extracted from the magnetic field dependences of each
exciton energy, where a quadratic term measures the diamagnetism of each
state. Even the linewidth or lifetime of the excitation can now be measured
544
at the quantum level. Near field microscopy /spectroscopy provides a means

to access energies and homogeneous line widths for the individual eigenstates of these centers, and thus allows the luminescent components to be
identified and characterized with the extraordinary detail previously limited to the realm of atomic physics. Such techniques may soon be extended
to map quantum well thicknesses more accurately using a NSOM transmission absorption setup. Likewise when dopant spacings are larger than the
NSOM resolution then each donor or acceptor might soon be individually
spatially mapped.
3.4. SUMMARY
Despite the impressive potential of this new technology, it is important
not to forget its fundamental limitations. The serial nature of the scanning process means that scanning speed is a severe constraint, commonly
from photon flux or mechanical considerations. As an extreme example,
to map out all of the eigenstates of a wafer using near-field spectroscopy
would take a few millennia. A short depth of focus and the requirement
that the probe be much closer to the sample than the required resolution,
makes it useful only for exposed surface or near-surface systems. Only flat
surfaces are suited to maintaining uniform resolution. Furthermore, the
near-field probes are extremely delicate. It is very easy to damage them by
abrasion with the surface or excessive laser power, thereby degrading their
resolution. Many "near-field" experiments have been published using such
probes where the images are no higher in resolution than one that could
be more easily obtained from a standard diffraction-limited optical microscope. As with all microscopies, interpretation of the NSOM images can be
easily misleading if careful scrutiny is lacking. For example, topography can
influence the optical images, and especially so if supplemental tip height
control techniques (such as shear force) are utilized. Sharp topographical
signals can then give misleading sharp optical signals. Even if the data is
well understood, there remains the tremendous challenge of using it constructively and cost-effectively. Consider that all of the energy states of a
device such as a laser can be individually catalogued. (This might be true
only for narrow quantum wells 10 nm thickness).) Then how, if at all, can
this highly, possibly overly detailed information be utilized to improve device performance than knowledge of a simpler far-field photoluminescence
measurement, where only the overall distribution is known? Knowing and
finding the answer to such questions in addition to an honest appreciation
of NSOM's limitations will determine the niche that NSOM finds amongst
competing technologies.
545
4. Conclusions
In this chapter, we have provided a survey of three different kinds of scanning probes-the STM, AFM, and NSOM-and a brief foray of diverse
experiments made possible with them. In particular, we have shown how
each can be used to fabricate and probe spectroscopically ultra-nanoscale
devices. At present, there is an ever-evolving variety of scanning probe
microscope uses and consequently, a constant flux of new and emerging
insights into mesoscopic systems.
ACKNOWLEDGEMENTS. We gratefully acknowledge R. Fitzgerald for

careful reading of this manuscript and for providing valuable suggestions.
We also gratefully acknowledge P. Campbell, D. Klein, A. Noy, E. Snow,
and Park Scientific Instruments for providing us with figures. Finally, we
would like thank E. S. Zeisky and C. Quigg for valuable assistance in preparing figures for this manuscript. This work was supported in part by NSF
DMR-96-24536, AT&T/Lucent Technologies Foundation, and the DuPont
Foundation.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
G. Binnig, H. Rohrer, Ch. Gerber, and E. Weibel, Appl. Phys. Lett. 40, 178 (1992).
C. T. Black, Ph.D. Thesis, Harvard University (1996).
J. A. Dagata and C. R. K. Marrian, in: Technolo!l1J of Prozimal Probe Lithography,
ed. C. R .. K. Marrian (SPIE-Institutes for Advanced Optical Technologies, 1993).
C. J. Chen, Introduction to Scanning Thnneling Micro.copy, Oxford University
Press, NY (1993).
J. Bardeen, Phys. Rev. Lett. 6, 57 (1961).
C. B. Duk, Tunneling in Solid., Academic Press, NY (1969).
J. Tersoff and D. R. Haman, Phys. Rev B 31, 805 (1985).
E. Stoll, Surf. Sci. 148, L411 (1984).
G. Binnig and H. Rohrer, Rev. Mod. Phys. 56, 615 (1987).
The "bimorph" is a trade name of Morgan Matroc, Inc., Vernitron Division, Bedford,
Ohio.
G. Binnig and D. P. E. Smith, Rev. Sci. Instrum. 57, 1688 (1986).
D. P. DiLella, J. H. Wandaas, and R. J. Colton, Rev. Sci. Inst. 60, 997 (1989).
Sang-il Park and C. F. Quate, Rev. Sci. Instrum. 58, 2004 (1987).
R. M. Feenstra, J. A. Stroscio, J. Tersoff, and A. P. Fein, Phys. Rev. Lett. 58,1192
(1987).
Y. Hasegawa and P. Avouris, Phys. Rev. Lett. 71, 1071 (1993).
H. F. Hess, R. B. Robinson, R. C. Dynes, J. J. M. Valles, and J. V. Wassczak, Phys.
Rev. Lett. 62, 214 (1989).
R. M. Feenstra and J. A. Stroscio, in Scanning Tunneling Micro,copy, p. 251, ed.
by J. A. Stroscio and W. J. Kaiser (Academic Press, Inc., San Diego, 1993).
M. F. Crommie, C. P. Lutz, and D. M. Eigler, Science 262, 218 (1993).
C. T. Black, M. T. Thominen, and M. Tinkham, Phys. Rev. B 50, 7888 (1994).
For other similar experiments and results, see P. J. M. van Bentum, R. T. M.
546
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
Smokers, and H. van Kempen, Phys. Rev. Lett. 60, 25443 (1988); R. Wilkens, E.
Ben-Jacob, and R. C. Jaklevic, Phys. Rev. Lett. 6S, 801(1989); and J. G. A.
DuBois, E. N. G. Verheijen, J. W. Gerritsen, and H. van Kempen, Phys. Rev. B
48, 11 260 (1993).
J. G. A. DuBois, J. W. Gerritsen, G. Schmid, and H. van Kempen, Physica B 218,
262 (1996).
F. M. Mulder, T. A. Stegink, R. C. Thiel, L. J. de Jongh, and G. Schmid, Nature
S67, 716 (1994).
D. M. Eigler and E. K. Schweizer, Nature S44, 524 (1990).
See for instance Ref. [4] and the references therein.
G. Meyer, S. Zophel, and K.-H. Rieder, Phys. Rev. Lett. 77, 2113 (1996).
I.-W. Lyo and P. Avouria, Science 25S, 173 (1991).
H. Uchida, D. H. Huang, and J. Yoahinobu, Surf. Sci. 287/288, 1056 (1993).
D. H. Huang, F. Grey, and M. Aono, Surf. Sci. SSl/SSS, 365 (1995).
J. W. Lyding, T. -C. Shen, and J. R. Tucker, Nanotechnology 7, 128 (1996).
C. T. Salling and M. G. Lagally, Science 265, 502 (1994).
J. A. Stroscio and D. M. Eigler, Science 254, 1319 (1991).
D. M. Eigler, C. P. Lutz, and W. E. Rudge, Nature S52, 600 (1991).
M. F. Crommie, C. P. Lutz, D. M. Eigler, and E. J. Heller, Physica D 8S, 98 (1995).
A number of rigorous analyses now exist which describe Crommie et al. 's quantum
corral. See E. J. Heller, M. F. Crommie, C. P. Lutz, and D. M. Eigler, Nature S61,
464 (1994); S. Crampin, M. H. Boon, and J. E. Inglesfield, Phys. Rev. Lett. 7S,
1015 (1994); G. Hoermandinger and J. B. Pendry, Phys. Rev. B 50, 18607 (1994);
H. K. Harbury and W. Porod, Phys. Rev. B 5S, 15445 (1996); and S. Crampin and
O. R. Bryant, (preprint) (1996).
G. Binnig, C. F. Quate, and C. Gerber, Phys. Rev. Lett. 56, 930 (1986).
D. Rugar and P. Hansma, Physics Today, October, 23 (1990).
C. Bustamante and D. Keller, Physics Today, December, 32 (1995).
M. Tortonese, H. Yamada, R. C. Barrett, and C. F. Quate, in Proceeding. of 7hm.ducer. '91, (IEEE, New York, 1991), 91CH2817-5, p. 448.
J. E. Griffith, D. A. Grigg, G. P. Kochanski, M. J. Vasile, and P. E. Russel, in:
Technology of Prozimal Probe Lithography, ed. C. R. K. Marrian (SPIE-Institutes
for Advanced Optical Technologies, 1993).
Ph. Niedermann and O. Fischer, J. of Microscopy, 152, 1988.
G. Reiss, F. Schneider, J. Vancea, and H. Hoffmann, Appl. Phys. Lett. 57, 1991.
H. Ximen and P. E. Russell, Ultramicroscopy, 42-44,1992.
D. Nyysonen, L. Landstein, and E. Coombs, J. Vac. Sci. Technol. BI, 1991.
M. A. McCord and R. F. W. Pease, J. Vac. Sci. Tech. B4, 86 (1986).
J. A. Dagata, W. Tseng, J. Bennett, E. A. Dobisl, J. Schneir, and H. H. Harary,
J. Vac. Sci. Technol AID, 2105 (1992); E. S. Snow, P. M. Campbell, and P. J.
McMarr, Appl. Phys. Lett. 6S, 749 (1993); and E. S. Snow, P. M. Campbell, and
B. V. Shanabrook, Appl. Phys. L-:tt. 6S, 3488 (1993).
M. A. McCord and R. F. W. Pease, Appl. Phys. Lett. 50, 569 (1987); A. Majumddar,
P. I. Oden, J. P. Carrejo, L. A. Nagahara, J. J. Graham, and J. Alexander, Appl.
Phys. Lett. 61, 2293 (1992).
T. A. Jung, A. Moser, M. T. Gale, H. J. Hug, and U. D. Schwarl, in Technology of
Prozimal Probe Lithography, ed. C. R. K. Marrian (SPIE-Institutes for Advanced
Optical Technologies, 1993).
M. A. McCord and R. F. W. Pease, Appl. Phys. Lett. 50, 569 (1987); J. Gobrecht
and J. B. Pethica, Microcircuit Eng. 5,471 (1986); Y. Kim and C. M. Lieber, Science
257, 375 (1992).
M. Wendel, S. KUhn, H. Lorenz, J. P. Kotthaus, and M. Holland, Appl. Phys. Lett.
65, 1775 (1994).
C. R. K. Marrian, Technology of Prozimal Probe Lithography, (SPIE-Institutes for
Advanced Optical Technologies), 1993.
547
51. L. L. Sohn and R. L. Willett, Appl. Phys. Lett. 87, 1552 (1995).
52. This method has been extended to a trilayer system. See V. Bouchiat and D. Esteve,
Appl. Phys. Lett. 89, 3098 (1996).
53. H. Dai, J. H. Hafner, A. G. RimIer, D. T. Colbert, and R. E. Smalley, Nature 384,
147 (1996).
54. E. S. Snow and P. M. Campbell, Science 270, 1639 (1995).
55. E. S. Snow, D. Park, and P. M. Campbell, Appl. Phys. Lett. 89, 269 (1996).
56. Y. Nakumura, D. L. Klein, and J. S. Tsai, Appl. Phys. Lett. 88, 275 (1996).
57. A. Ashldn, J. M. Dliedsic, J. E. Bjorkholm, and S. Chu, Opt. Lett. 11, 288 (1986).
58. J. Israelachvili, Intermolecular and Surface Force. (Academic Press, New york,
1992); J. Israelachvili, Acc. Chem. Res. 20, 415 (1987); H. Yoshilawa, Y.-L. Chen,
J. Israelachvili, J. Phys. Chem. 97, 4128 (1993); Y.-L. Chen, C. A. Helm, J. N.
Iraelachvili, ibid. 95, 10736 (1991).
59. C. Daniel Frisbie, Lawrence F. ROlSnyai, Aleksandr Noy, Mark S. Wrighton, and
Charles M. Lieber, Science 285, 2071 (1994).
60. Aleksandr Noy, C. Daniel Frisbie, Lawrence F. ROlSnyai, Mark S. Wrighton, and
Charles M. Lieber, J. A. Chem. Soc. 117, 7943 (1995).
61. K. L. Johnson, K. Kendall, and A. D. Roberts, Proc. R. Soc. London A 324, 301
(1971).
62. W. T. Miiller, D. L. Klein, T. Lee, J. Clarke, P. L. McEuen, and P. G. Schultl,
Science 288, 272 (1995).
63. P. N. Rylander, Organic Synthe.i. with Noble Metal Catalylt., (Academic Press,
New York, 1973).
64. M. A. Eriksson, R. G. Beck, M. A. Topinka, J. A. Katine, R. M. Westervelt, K. L.
Campman, and A. C. Gossard, Appl. Phys. Lett. 89, 671 (1996).
65. Park Scientific Instruments, Sunnyvale, CA 94089.
66. For a review of scanned probe microscopy techniques, see Scanning Thnneling Micro,copy II, ed. by R. Wiesendanger and H.-J. Guntherodt (Springer-Verlag, NY,
1995).
67. M. A. Eriksson, R. G. Beck, M. A. Topinka, J. A. Katine, R. M. Westervelt, K. L.
Campman, A. C. Gossard, Superlattices and Microstructures 20, 435 (1996).
68. H. Baranger and D. A. Stone, Phys. Rev. Lett. 83,414 (1989).
69. L. W. Molenkamp, A. A. M. Staring, C. W. J. Beenakker, R. Eppenga, C. E.
Timmering, J. G. Williamson, C. J. P. M. Harmans, and C. T. Foxon, Phys. Rev.
B41, 1274 (1990).
70. K. L. Shepard, M. L. Ronkes, and B. P. van der Graag, Phys. Rev. Lett. 88, 2660
(1992).
71. E. Betlig and J. K. Trautman, Science 257, 189 (1992).
72. E. Betlig, J .K. Trautman, R. Wolfe, E. M. Gyorgy, P. L. Finn, M. H. Kryder and
C.-H. Chang, Appl. Phys. Lett. 81, 142 (1992).
73. E. Betlig, S.G. Grubb, R.J. Chichester, D.J. DiGiovanni, and J.S. Weiner, Appl.
Phys. Lett. 83, 3550 (1993).
74. S.K. Buratto, J.W.P. Hsu, J.K. Trautman, E. Betlig, R.B. Bylsma, C.C. Bahr, and
M.J. Cardillo, J. Appl. Phys. 72, 7720 (1994); S.K. Buratto, J.W.P. Hsu, E. Betlig,
J .K. Trautman, R.B. Bylsma, C.C. Bahr, and M.J. Cardillo, Appl. Phys. Lett. 85,
2654 (1994).
75. E. Betlig and R.J. Chichester, Science 282, 1422 (1993).
76. H. M. Marchman, J. E. Griffith, J. K. Trautman, J. ofVac. Science and Tech. (1995)
77. R. D. Grober, T. D. Harris, J. K. Trautman, E. Betlig, W. Wegscheider, L. Pfeiffer,
and K. West, Appl. Phys. Lett. 84, 1421 (1994).
78. H. F. Hess, E. Betlig, T.D.Harris, L.N. Pfeiffer, and K. W. West, Science 284, 1740
(1994).
QUANTUM POINT CONTACTS BETWEEN METALS
J.M. VAN RUITENBEEK
Kamerlingh Onnes Laboratorium, Leiden University

Box 9506, NL-2300 RA Leiden, The Netherlands
1. Introduction
In the chapter by Kouwenhoven, Schon and Sohn, the phenomenon of quantization of the conductance is introduced. The phenomenon is a consequence
of the wave nature of the charge carriers. In a conductor for which the width
is not very much larger than the Fermi wavelength, it is useful to describe
the conductance, G, in terms of the Landauer-Biittiker expression
(1)
where the Tn describe the transmission probability for each of the modes
at the entrance of the conductor and the sum runs over all occupied modes.
If we can contrive our experiment in such a way that the Tn are equal to
1, up to a mode number N, and equal to zero for all other modes, then the
conductance assumes values which are an integer times the quantum unit of
conductance, Go = 2e 2 / h. It turns out to be possible to fabricate conductors
which have precisely this property, as was beautifully demonstrated in the
seminal experiments by van Wees et al. [1] and Wharam et al. [2] on a
two-dimensional electron gas (2DEG) semiconductor device.
The question that will be addressed here is: is it possible to observe
this quantization effect also in metallic point contacts? As we will see, the
answer is yes, but the problem is more complicated, due to an interesting
interplay between the atomic structure of the contacts and the quantization
of the electronic wave functions. The problem is of importance for fundamental and applied research. This can be judged from the fact that the
Fermi wavelength in metals is much shorter than that in the semiconductor devices, namely it is of the order ofthe atomic diameter, )..F ~ 5 A , in
metals compared to rv 100 times larger values in the 2DEG devices. This
549
L. L. 80hn et al. (ells.), Mesoscopic Electron Transport, 549-579.
550
implies that, where the 2DEG experiments require cooling to helium temperatures in order to be able to resolve the splitting of'" 1 me V between the
modes, in metals the mode splitting is ",1 eV, which is sufficiently high to
allow experiments to be performed at room temperature. There are various
links from the research in this field to problems such as friction and wear
at surfaces [3], since the conductance of the contact is intimately related to
the atomic scale dynamics of formation of the contact. On the fundamental
side, there are a number of properties of the electron transport in metals,
which are of great interest to study under conditions of a limited number of
quantized modes. At the end of this chapter a few sections are devoted to
such problems, with emphasis to superconductivity and Kondo scattering
in atom size contacts. We begin by first describing the experimental techniques and outlining the theory for conduction in metallic quantum point
contacts. The central body of the chapter is devoted to a discussion of the
experimental investigation of conductance quantization in metals.
2. Experimental techniques for the study of metallic point contacts

Many years before the advent of nanofabrication, ballistic metallic point
contacts were widely studied, and many beautiful experiments have been
performed [4, 5]. The principle was discovered by Yanson [6] and later
developed by Jansen et al. [7]. The technique has been worked out with
various refinements for a range of applications, but essentially it consists of
bringing a needle of a metal gently into contact with a metal surface. This is
known as the spear-anvil technique. Usually, some type of differential-screw
mechanism is used to manually adjust the contact. With this technique
stable contacts are typically formed having resistances in the range from
",0.1 to ",10 0, which corresponds (see below) to contact diameters between
d ~ 10 and 100 nm. The elastic and inelastic mean free path of the charge
carriers can be much longer than this length d, when working with clean
metals at low temperatures, and the ballistic nature of the transport in such
contacts has been convincingly demonstrated in many experiments. The
main application of the technique has been to study the electron-phonon
interaction in metals. Here, one makes use of the fact that the (small but
finite) probability for back-scattering through the contact is enhanced as
soon as the electrons acquire sufficient energy from the electric potential
difference over the contact, that they are able to excite the main phonon
modes of the material. The differential resistance, dV/ dI, of the contact
is seen to increase at the characteristic phonon energies of the material,
and a spectrum of the energy-dependent electron-phonon scattering can be
directly obtained by measuring the second derivative of the voltage with
551
6.0
Au
4.0
2.0
0.0
~;"':':';;;;"""='::-"""'-I:l:----'-~-l..-~---I
30
40
voltage ImYI
Figure 1. An example of an electron-phonon spectrum measured for a gold point contact

by taking the second derivative of the voltage with respect to the current. The long-dashed
curve represents the phonon density of states obtained from inelastic neutron scattering
(From: [7]).
current, d2 V/dI 2 , as a function of the applied bias voltage. An example is

given in Fig. 1. Peaks in the spectra are typically observed between 15 and
30 mY, and are generally in excellent agreement with spectral information
from other experiments, and with calculated spectra.
The ballistic character of the transport has been exploited in even more
ingenious experiments such as the focusing of the electron trajectories
onto a second point contact by the application of a perpendicular magnetic field [8] and the injection of ballistic electrons onto a normal metalsuperconductor interface for a direct observation of Andreev reflection [9].
Such type of experiments have become very popular nowadays, using far
more sophisticated sample preparation, where the level of control over the
various aspects of the experiment has, of course, increased enormously.
Contacts of the spear-anvil type are not suitable for the study of the
quantum regime, which requires contact diameters comparable to the Fermi
wavelength, i.e. contacts of the size of atoms. For smaller contacts (higher
resistances) the above described technique is not sufficiently stable for measurement. What is more important, most of the experiments in the quantum
regime need some means fine control over the contact size.
The first of these requirements, a high stability, can be met by nanofabrication of point contacts. This approach was introduced by Ralls and
Buhrman [10] and is also discussed in the chapter by Ralph and Tinkham.
Briefly, e-beam lithography is used to fabricate a nanometre size hole in a
free standing thin film of Si3 N4 , and a metal film is evaporated onto both
sides of the silicon nitrate film, filling up the hole and forming a point contact between the metal films on opposite sides. These structures are very
552
stable, and contacts only several nanometres wide can be produced. The
great advantage, here, is that the point contact can be cycled to room temperature and be measured as a function of field or temperature without
influence on the contact size. The drawback is that one cannot control the
size of the contact during the experiment and that the contacts are still
fairly large compared to atomic dimensions.
Two intimately related techniques have been used recently to study
metallic point contacts in the quantum limit. For measuring the conductance, the Scanning Tunnelling Microscope (STM) is frequently used, which
is described in the chapter by Crommie and Sohn. The experiment, once
we have the STM installed, is simple enough: the metal probe tip is driven
into the sample surface to form a large contact, and it is subsequently
pulled back. The contact resistance is measured during pull-off. These cycles typically take 1 msec. The first experiment of this type was reported
by Gimzewski and Moller [11], who observed an exponential dependence
of the tunnel resistance with distance, which tended to saturate at close
proximity. The transition to contact was seen to manifest it self as a distinct jump to contact!. The first contact value was found to be near 10 kfl.
This value is close to the quantum resistance, but this work was published
in the year before the discovery of conductance quantization and it was
(correctly) interpreted to correspond to a first contact of the size of one or
two atoms.
In 1992 in our group we introduced a slightly different method for the
study of atom size contacts, which is a modified version of the breakjunction technique developed by Moreland [13], and which Chris Muller
baptised the Mechanically Controllable Break-junction (MCB) technique
[14, 15, 16]. The principle of the technique is illustrated in Fig. 2. By
breaking the metal, two clean fracture surfaces are exposed, which remain
clean due to the cryo- pumping action of the low-temperature vacuum can.
This method circumvents the problem of surface contamination of tip and
sample in an STM experiment, where a DRV chamber with surface preparation and analysis facilities are required to obtain similar conditions. The
fracture surfaces can be brought back into contact by relaxing the force on
the elastic substrate, while a piezo-electric element is used for fine control.
The roughness of the fracture surfaces results in a first contact at one point,
and experiments usually give no evidence of multiple contacts. In addition
to a clean surface, a second advantage of the method is the stability of the
two electrodes with respect to each other. ~From the noise in the current
in the tunnelling regime we obtain an estimate of the vibration amplitude
of the vacuum distance, which is typically less than 10- 3 A. The stability
IThe sharp transition and hysteresis at this jump-to-contact have recently been proposed for applications as an atomic switch mechanism [12].
553
.......................... (
L :::X3 . . .:
,... 4
~::::::::=l 5
Figure 2. Schematic top and side view of the mounting of a MeB, where the metal to
be studied has the form of a notched wire (1), which is fixed onto an insulated elastic
substrate (3) with two drops of epoxy adhesive (4) very close to either side of the notch.
The substrate is mounted in a three-point bending configuration between the top of a
stacked piezo element (5) and two fixed counter supports (2) . This setup is mounted
inside a vacuum can and cooled down to liquid helium temperatures. Then the substrate
is bent by moving the piezo element forward. The bending causes the top surface of the
substrate to expand and the wire to break at the notch. Typical sizes are L ::: 20 mm
and u ::: 0.1 mm.
results from the reduction of the mechanical loop which connects one contact side to the other, from centimetres, in the case of an STM scanner, to
'" 0.1 mm in the MCR
In the first experiments, aimed at a study of conductance quantization
in metals [15J, distinct steps were observed in the conductance of Pt and Nb
contacts. Although the steps are of order of the conductance quantum, the
authors cautioned against a direct interpretation in terms of conductance
quantization. This point will be discussed in some detail below. Fig. 3 shows
some recent examples of recordings of the conductance as a function of the
voltage Vp on the piezo element, which is a measure of the displacement
of the two wire ends with respect to each other. In the experiment, when
coming from the tunnelling regime, a large contact is formed by pressing
the electrodes together, reducing the resistance to a value below 100 n
(conductance above 100 Go). The contact is then slowly elongated while
recording the conductance. The scans as given in Fig. 3 are recorded in
about 10 minutes each, using low amplitude (0.1 /-LA, 180 Hz) current modulation and lock-in detection. The conductance is observed to decrease in a
step-wise fashion, which is different each time a new contact is made. Only
the last plateau in the conductance before the jump to tunnelling shows a
reproducible value of 1 x 2e 2 jh, for these simple metals [17J.
The STM has obviously the advantage over the MCB technique that by
554
.
-------10A
7
6
5
4
3
2
Copper
:c
;::;--
cu
~
cu
v
...=
<:II
'e
=
0
25
50
100
75
125
7
6
5
4
3
2
1
0
50
100
200
150
250
7
6
5
4
3
2
1
10 V
[V]
15
20
25
Figure 3. Typical recordings of the conductance G , measured using the MCB technique
for the simple metals Cu, Au and Na, at temperatures of 4.2 K and below. The electrodes
are pulled apart by increasing Vp An estimate for the corresponding displacement in
angstrom is given on the upper axis. For each of the metals three traces are shown,
obtained by making a large contact before starting each curve. The curves are displaced
along the Vp axis for clarity. Note that sodium is difficult to handle, and cannot be
mounted in the simple fashion described in Fig. 2. (From: [18]).
scanning the tip over the sample surface, information on the topography
of the contact area can be obtained. This drawback can be remedied to
some extent by introducing two additional thin piezo-plates for X and
Y displacement on the substrate of the MCB [19]. Other applications of
the MCB principle include a nanofabricated break-junction [20, 21], which
involves the breaking of a bridge only 2 /-lm wide. This has the advantage
of a further increase in stability, and the possibility of incorporating the
controllable junction into more complicated circuitry on the same chip. As
a first example of the latter an adjustable RF -SQUID using a MCB as the
weak link was recently fabricated and applied to the study of the currentphase relation in superconductors [22].
Many experiments on the conductance of atom size contacts have recently been performed using an STM. The first experiment of this type
was reported by Agralt et at. [23], Fig. 4. Further experiments have been
555
ON-:.
'"
'"uc:
'!.
____
-L~
_ _ _ _- L L -_ _ _ _
____
'"
U
~ O
=>
c:
o
50
100
150
200
250
time (ms)
Figure 4. STM experiment of the conduction of atom size contacts of Au at 4.2 K.

In the upper trace the z-piezo voltage is kept constant and the conductance is seen to
oscillate between two well defined values. In the lower traces the z piezo is driven by a
triangular voltage of 16 Hz, corresponding to a tip displacement of 14 A. The bias voltage
was fixed at 0.1 V. (From: [23])
performed at room temperature in air [24, 25, 26, 27] and ultra high
vacuum (URV) [28, 29, 30], and at cryogenic temperatures [31, 32, 33].
Refs. [24, 29, 31, 52] argued that, at room temperature but also at helium
temperatures, a neck is formed between the tip and the surface, which is
elongated and finally breaks. After breaking, the sample-side remnant of
the neck was observed as a hillock, by scanning over the surface. Often,
tip and sample are formed from the same elemental metal, but in other
experiments a tip and surface are made of dissimilar metals, where it has
been argued that the surface metal will wet the tip material resulting after the first few indentations in contact formation between identical metals
[28,29].
Costa-Kramer et at. [37] reported that conductance steps can even be
observed when having two common wires loosely touching each other and
measuring the contact resistance on a fast oscilloscope while the wires vibrate in and out of contact. This shows that the phenomena described here
can be very general and robust.
The measured conductances are usually presented as a plot of the conductance as a function of either time or voltage on the z-piezo along the
abscissa (Fig. 3, Fig. 4), both of which can be taken as a measure for the
displacement of the bulk of the electrodes with respect to each other. It is
useful to note that the plots have a misleading similarity with plots of the
conductance versus gate voltage which we know from the research in 2DEG
devices. It is misleading in the sense that the gate voltage directly controls
556
20
:c;;-
..
15
<>
c 10
'"
U
::J
"
C
0
"
0
2
(b)
Z -2
..
.E
4
6
8
0.5
1.0
tip displacement (nm)
1.5
Figure 5. Simultaneous measurement of force and conductance on atom scale point

contacts for Au. The sample is mounted on a cantilever beam and the force between tip
and sample is measured by the deflection of the beam using an Atomic Force Microscope
(AFM). The measurements are done in air at room temperature. (From: [34])
the width of the point contact in the 2DEG, whereas for the metal point
contacts pulling the tip away from the surface reduces the contact diameter
in a fashion which we have no direct control over. In stead of heaving two
parameters that directly relate to the properties of the contact, we have
only one. As we will see, it requires some careful analysis of the data to
draw conclusions from the measurements. Important progress was recently
made by Rubio and co-workers [34, 35] and Stalder and Diirig [36] who
made a simultaneous measurement of the conductance of the contact and
the force between tip and sample. This was achieved by mounting the sample on a flexible cantilever and measuring the deflection of the cantilever
during contact pull-off. They were able to measure the force down to a
single atom contact simultaneously with its conductance (Fig. 5).
In the next section some of the theoretical concepts are briefly summarised, after which we return to a discussion of the results of the experiments.
557
3. Theory of conductance quantization in metals

It is instructive to first consider the semiclassical result for conduction
through a point contact, which was first derived by Sharvin [38]2. In this
approximation, one integrates the contribution to the current density, dJx =
2ev x / L3, over all occupied states in k- space. The right-moving states are
occupied to an energy eV higher than the left-moving states, resulting in a
net current density. The total current is then obtained by integration over
the cross-sectional area of the contact, and by dividing out the potential
difference V we can express the conductance of the contact as
G = 2e
S
2(k Fa)2
2
'
(2)
where kF is the Fermi wave vector. Quantum mechanics enters only through
the density of states in k-space, and Fermi statistics. The factor 2e 2 / h
appears because this is the natural unit for conductance and it does not
imply that the conductance is quantized. Indeed, we have not made explicit
use of the wave-nature of the electrons in the contact, and Gs in Eq.2
increases smoothly with increasing contact radius a. For small contacts
deviations from this relation are expected, even on the semiclassical level,
because the result becomes sensitive to the fact that the density of states
in k-space drops to zero near k = 0 [39]. Torres and Saenz derived the
following, corrected form for the Sharvin resistance, for a cylindrical contact
[40]
(3)
where a' = a + 8a is an effective radius correcting the hard wall potential for the finite height in real systems. For metals kF8a =0.72~1.04 and
a =0.13~0.025 [40]. Using this approximation, the conductance of a contact
having the size of a single atom is obtained as 1.1~1.8 Go for monovalent
metals. 3 This should be regarded as an estimate only, since the boundaries
of an atom are not sharply defined and contributions from back scattering
around the contact will lower the actual value. In any case, the value for
a single atom contact is very close the quantum unit of conductance. We
have not made use of the quantization of the wave functions in the contact.
Therefore, in the experiment, any change of conductance of order of the
2In the discussion of conductance quantization we limit ourselves to the zero bias
conductance, i.e. G in the limit for V --+ 0
3We take the nearest neighbour distance to correspond to the atomic diameter. For
monovalent metals the density of atoms na e~als the density of electrons, ne. With
ne = k}/37r 2 , na = Vi/a 3 for FCC and na = 3y3/4a 3 for BCC, we obtain kFa/2 = 0.868
and 0.844, respectively.
558
Radius
10
9
2 ~
ru""
Cll
~ 4
'"
3
2
1
OL-____
L -_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
m: 0
n: 1
2 0
1 2
1 2
Figure 6. The Bessel functions 1m bmnr / R) for m = 0, 1,2, and 3, and the conductance
of a perfect cylinder as a function of the radius, R, which is assumed to be a continuous
parameter. The conductance increases by one unit at each zero of 10 and by two units
at each zero of all other Bessel functions.
quantum unit, may simply be the result of a change in contact size by one
atom (i.e. a semiclassical effect), rather than a manifestation of the quantum nature of conductance. We will discuss below how these two effects
can be distinguished.
We will now proceed to a full quantum description of point contacts.
Let us first consider a free electron gas confined to a cylindrical conductor
of radius R, with hard wall boundaries [41, 44J. The wave functions are
described by
(4)
where the z co-ordinate is taken along the cylinder axis. The radial part
of the wave function is described by the Bessel functions Jm , which are
restricted to have a zero at the wall of the cylinder. This imposes the quantization condition, and the quantum numbers m and n correspond to the
n-th zero of the Bessel function of order m. Since J_m(r) = (-1)mJm(r),
all states for m f= 0 are two-fold degenerate (in addition to spin degeneracy). The degeneracy is evident from the expression for the energies of the
eigenstates,
(5)
The conductance of the cylinder as a function of its radius can now be
simply deduced by plotting the Bessel functions and recording the sequence
of the zero's, as shown in Fig. 6.
559
Figure 7. Model calculation of the conductance (right) of a curvilinear contact (left),

both as a function of the area of the contact and as a function of the opening angle eo
(From: [44]).
This model of a cylindrical wire is a highly idealised representation

of a metallic point contact. A point contact can have a cross-section which
deviates from perfect cylindrical symmetry, which will lift the degeneracy of
the m modes, and split the double conductance steps into two single steps.
The effect of deformations on the degeneracy was analysed by Bratkovsky
and Rashkeev [43]. In addition, the perfectly sharp steps in Fig. 6 become
smeared when we consider a finite length for the wire. This was modelled
by Torres and Saenz [44] who used a hyperbola of revolution to describe the
contact. The opening angle 2eo of the hyperbola determines the effective
length of the contact: for (Jo approaching zero the length of the wire tends to
infinity, whereas for 2(Jo = 7r the model collapses to an infinitely thin barrier
separating the two electrodes, pierced by a circular hole, corresponding to
an infinitely short wire. The result of their calculation is reproduced in
Fig. 7. The sharp steps obtained for a long cylindrical wire ((Jo = 0)
gradually smear out as (Jo approaches 7r /2. The steps at high conductances
= 7r /2 the result coincides with the semiclassical
disappear first and for
behaviour, Eq. 3. The smearing of the steps is readily understood when
we assume a sufficiently slow variation of the radius of the conductor as
a function of the position z along the direction of current flow (adiabatic
approximation) [42]. With this assumption the eigenstates are described by
Eq. 4, with the sole modification that R is now a function of z. The energies
are similarly described by Eq. 5, where the second term depends on position
z through R(z). This term can be considered as an effective potential energy
Vmn(z), which depends on the mode numbers m,n (see Fig. 8). For each
mode this potential energy has a maximum at the centre of the contact,
and the mode is transmitted when this maximum falls below the Fermi
eo
560
z
Figure 8. For a slowly varying radius of the conductor (top) the motion parallel and
perpendicular to the axis of the contact can be decomposed, and the energy required for
the perpendicular motion in mode {m, n} can be considered as a potential energy Vmn(Z)
for an otherwise freely propagating wave in the Z direction. When the Fermi energy falls
above the maximum in Vmn(Z), EF = EA, the mode {m,n} is conducting, when it falls
below the maximum, EF = EB, the mode is blocked.
energy, EF, and the mode is reflected when it is higher than EF. However,
when EF falls just below the maximum of Vmn(z), part of the mode will be
transmitted by tunnelling through the barrier. When EF falls just above
the maximum of Vmn(z), part of the mode will be reflected due to the
variation of the potential height (over-the- barrier scattering). The latter
corrections are most important just around the step in the conductance,
when a new mode starts to conduct, which explains the rounding of the
steps. Tunnelling and over-the- barrier scattering are more important the
thinner is the barrier Vmn(z), i.e. the shorter is the contact.
Further, back scattering of the conduction electrons in the contact degrades the definition of the conductance plateaux and the accuracy of the
quantization. Surface roughness can form an important cause for back scattering, and its effect was analysed in Ref. [43] as well as in several molecular
dynamics based calculations which will be discussed below [28, 29, 45, 46].
Back scattering further results from a non-adiabatic connection of the
atomic size wire to the bulk electrodes, and from disorder and impurities
in the contact. Also, the scattering on the regular periodic crystal potential may introduce complications. The scattering on the periodic lattice in
bulk metals results in an anisotropy of the Fermi surface or the breaking-up
of the Fermi sphere into multiple components, depending on the material
we are studying. The Fermi surface has no direct relevance for the point
contacts, since the boundaries break the necessary translational invariance,
but Bragg reflection should still affect the modes in the contact. This is
implicitly taken into account in the tight-binding calculations by Todorov,
561
Sutton, and co-workers [45, 46, 47], although it is not important at half
filling.
Note that we started our discussion in this section from models based
on plane waves and hard wall boundaries, which is inspired by the work on
point contacts in 2DEG's, where this is a very good approximation. In point
contacts of real metals, however, the wave functions may be more closely
described by starting from atomic orbitals. Early work towards a realistic
description was reported by Lang [48, 49J. He describes a local density
type calculation for a single atom sandwiched between two flat "jellium"
half spaces, and finds for the conductance through this one-atom contact
geometry a value which depends on the type of atom: 32kO for Na, 18kO
for Ca and 6.6kO for AI. In his 1987 paper he observes that these values
are close to 2e 2 / h, which he discusses in terms of the Landauer formula,
and as an alternative, in terms of resonant tunnelling. In view of the recent
experiments, however, it is striking that the deviations from the quantum
value are so large. Exact correspondence of one atom with one quantum is
obtained in the somewhat crude tight binding calculations [47, 45, 50J but
only for a symmetric position of the atom between the electrodes, and for
a single orbital per atom [33J.
4. The nature of the steps in the conductance

Let us return to the experiments, and discuss the problem of the interpretation of the steps which are observed in the conductance (Figs. 3-5). As
was pointed out already after the first observation of the steps [15], the
dynamics of the contact conductance around the steps strongly favours an
interpretation in terms of atomic rearrangements, which result in a step
wise variation of the contact diameter. This interpretation has been corroborated by a number of recent experiments and molecular dynamics simulations. We will first review the experimental evidence.
In the low temperature experiments, it was observed [15, 17, 18, 23,
31, 51J that the conductance traces are different in each experiment, but
that it is possible to make reproducible cycles over a small range in piezo
voltage, Vp. In such cycles, the steps often show hysteresis in the position
at which they occur for the forward and backward sweeps. The hysteresis
is sensitive to the bath temperature of the experiment, and could in some
cases be removed by raising the temperature by only a few Kelvin's (see
Fig. 9). For other steps, such hysteresis was not observed, but in stead the
conductance showed spontaneous fluctuations of a two-level type, between
the values before and after the step. This phenomenon is observed only in a
very narrow range of Vp around the steps: at the plateaux the conductance
assumes stable values. In some cases it is even possible to tune the "duty
562
~ 0.3 A
-=
y
:::
o
v.[V]
"
Figure 9. Measurement of the conductance for an atom-sized Au contact while sweeping

the piezo voltage forward and backward around a single step. Clear hysteresis of the order
of 0.1 A is observed at 2.2 K. At 3.1 K the width of the hysteresis is reduced, and at
4.0 K it is entirely suppressed. (From: [18])
cycle" of the two level fluctuation (TLF), i.e. the relative portion of the time
spent in the upper compared to the lower state, by fine adjustment of Vp,
as shown in Fig. 10. Although the fluctuations have not yet been studied
as a function of temperature, it was observed that the average period of
the fluctuations reduces exponentially with increasing current through the
contact, which was interpreted as the result of heating by the electron
current [15]. Whether hysteresis or TLF is observed, the steps are always
steep, the slope being limited by the experimental resolution.
All these observations are clearly not consistent with a smoothly varying
contact radius and steps resulting directly from the quantization of the
conductance. A natural interpretation, rather, is formed by a model which
describes the contact as having a stable atomic geometry over the length of
a plateau in the conductance, where the total energy finds itself in a local
minimum. At the jumps in the conductance, the local energy minimum for
a new geometry drops below that of the present state as a result of the
stress applied to the contact. When the energy barrier between the old and
the new minimum is large compared to the thermal energy, we will observe
hysteresis, and when it is small enough thermally activated fluctuations
between the two states will be observed.
The height of the steps is of the order of the conductance quantum,
but no systematic correlation between the position of the plateaux and
563
0 _1l00
0 _006
0.000
0 .005
time [s J
time [sJ
0 .010
O.OH
0 .010
0.015
Figure 10. Time traces of the resistance of a Na contact at 4.2 K. From the lower
plot upward Vp is increased in increments of 0.1 V. The dominant occupation state is
gradually inverted upon increase of Vp. The conductance of the contact fluctuates between
approximately Go and 2Go. (From: [18])
quantized values could be discerned, except for the last conductance plateau
before the jump to the tunnel regime, and only for simple metals [17, 23].
Again, this is consistent with the picture of atomic rearrangements, where
the contact changes size by approximately the area of one atom, and as
we have seen, one atom contributes rv1 Go to the conductance, even in the
semiclassical approximation.
This should be different for semimetals, such as Sb and Bi, where the
density of electrons is more than three orders of magnitude smaller than
in ordinary metals. From the electron density in Sb one estimates a Fermi
wavelength of 55 A, which is much larger than the atomic diameter. Experiments on point contacts for this semimetal [53J show indeed that the
jump from vacuum tunnelling to contact is found at rv1 MO, or rvO.01 Go.
Continuing to increase the contact, steps in the conductance were again
observed, with a step height which was also of order 0.01 Go. This is consistent with the notion of steps resulting from atomic rearrangements and
with a semiclassical estimate for the conduction of one atom of Sb. An interpretation of the steps in terms of conductance quantization is definitely
ruled out. At still larger contact diameters, no evidence was found for quan-
564
tization around the unit values of conductance, illustrating that the various
mechanisms which can destroy quantization, as discussed in section 3, can
be effective down to the last quantum value.
The first evidence that atom size contacts should change size in a step
wise fashion came from an early molecular dynamics calculation by Landman et ai. [52]. In a model of a metal tip contacting a metal surface at room
temperature, it was found that when the tip is driven into the surface and
then pulled back, a neck is formed between the two metals, which shows a
surprisingly regular atomic stacking. During elongation of the neck a periodic evolution was observed, where the increasing stress first results in
elastic energy build up by increasing the distance between the layers along
contact axis. A maximum in the (negative) contact pressure is then followed by a disordering of the atomic positions in the contact and a release
of elastic energy. After this phase is completed the contact shows again a
regular layered stacking of the atoms along the contact, where a new layer
has been introduced compared to the start of the cycle. The maximum
pressure in the contact can be an order of magnitude higher than in bulk
metals.
Todorov and Sutton [47] were the first to combine a molecular dynamics
simulation with a calculation of the conductance. At every 10th time step of
the simulation they placed hydrogen-like orbitals at each of the momentary
atomic positions, and using a tight binding scheme evaluated the conductance of the tip-sample geometry. The calculation clearly demonstrated
steps in the conductance, which coincide with the atomic rearrangement
events of the contact. Similar results have been obtained from calculations
of the conductance for a free electron gas, bounded by a surface which is
determined at each time step from the atomic positions [28, 29, 46]. Landman et ai. [27] calculated the conductance from the minimal cross sectional
area, and the corrected Sharvin expression, Eq. 3. The result of their calculation reproduces the experiment very well, and is a clear demonstration
that the steps can be described by a semiclassical theory, without taking
quantization of the conductance into account.
Recently, the experiments of Refs. [34, 35, 36] (see Fig. 5) which combine
measurement of the conductance of a contact with simultaneous measurement of the force between tip and sample, established that the interpretation of the steps as resulting from atomic rearrangements, also applies at
room temperature and for the STM geometry. The steps in the conductance
are found to be systematically accompanied by sudden relaxations of the
force on the contact. The results are in excellent agreement with several
model calculations [54, 55, 56].
565
5. Evidence for conductance quantization in metals

In view of the discussion in the previous sections we are not satisfied with
observing steps in the conductance of order 2e2 /h to be convinced of the
existence of quantization effects in the conductance of metallic point contacts. On the one hand, many experimental curves show plateaux in the
conductance at very close proximity to quantized values, on the other hand
there are many plateaux which are evidently not at a quantum number.
A systematic analysis is required to disentangle effects due to the atomic
structure from those due to the electronic wave functions in the contact.
The solution which has been found relies on a statistical analysis of
many conductance curves. It is based on the observation that every individual measurement of a conductance curve such as shown in Fig. 3, is
different as a result of the many choices for the atomic structure which the
contact assumes. For each atomic geometry there is an effective cross section area of the contact, and the contact jumps between successive values
for this area. If the properties of the contact are favourable, we can find
the conductance for each of these cross sections at the given values for the
area from one of the curves at small eo in Fig. 7. In the ideal situation,
for eo = 0, the conductance assumes only quantized values, whatever the
value of the contact area. However, for realistic parameters of the contact,
which can in this simplified model be represented by finite values of the
angle eo, the conductance is a smooth function of the area. The various
contact areas can, therefore, result in any value for the conductance, and
these conductances remain nearly stable until the contact jumps to a new
geometry. The structure in the relation between the conductance and the
area of the contact can then be uncovered by measuring for each conductance value how often it is found in the experiments. If the conductance
curve is nearly structureless, as in the semiclassical expression, Eq. 3, or
for eo ~ 0 in Fig. 7, then a distribution will be observed which is constant
or some smooth function of G. If, on the other hand, in the conductance as
a function of area is influenced by quantization of the wave functions, then
the distribution should demonstrate peaks at the quantum numbers.
A first attempt at such an analysis [17] only considered the values for
the first conductance plateau in the experimental curves obtained with the
low temperature MCB technique. For a simple monovalent metal (Cu) the
distribution for the conductance value of the first plateau showed a fairly
narrow peak near 1 Go; for aluminium the distribution was centred around
Go, but much wider, and platinum showed a peak near 1.6 Go.
A full statistical analysis for a wider range of conductance values was
applied, independently, to two different experiments, with some very interesting differences in the results. The Aarhus/Lyngby co-operation [57]
566
4,----,----,-----,----,----,
::J
>-
a:
iii
a:
I-
<t
O~~~~~~2~~~3~~=+4=---~5
CONDUCTANCE (2e 2/h)
Figure 11. Histogram representing the relative weight which each conductance value has
in the experiments. The histogram is constructed from 227 conductance curves recorded
while breaking Au contacts, using and STM under UHV at room temperature. (From:
[29])
analysed their previous experiment on Pt contacts in a DRV-STM at room

temperature, and observed peaks at 1,2,3 and 4 Go. Soon after this, they
reported very convincing peaks at similar values for Au at room temperature [29] (Fig. 11). The latter results were accompanied by a calculation
using molecular dynamics simulations for the geometry of the contact, and
freely propagating waves to calculate the conductance. The calculations
reproduced the relative height, as well as the width of the peaks in the
histogram.
In a study of sodium contacts at liquid helium temperatures using the
MCB technique [58] a similar histogram for the distribution of conductance
values (Fig. 12) displays clear peaks at 1 and 3, and smaller peaks near 5
and 6 Go. Peaks at 2, 4 and 7 Go are absent. This sequence of conductance
values, coincides exactly with the sequence obtained for a cylindricaly symmetric contact, as shown in Fig. 7. The symmetry of the contact produces
a degeneracy of the modes, which results in a characteristic series of single
and double conductance steps. The fact that this sequence can be observed
experimentally makes a very clear case for the existence of quantization of
the conductance in metals. Torres and Saenz [55] used a combination of
a mechanical atomic scale model and free electron waves, to calculate the
histograms. The relative height and width of the peaks, as well as a small
shift relative to the perfect quantum numbers, are in very good agreement
with the sodium experiment, with the exception of the peak at 5, which has
a smaller amplitude in the simulation. It should be noted that the cylindrical symmetry has been put into the model, and that for Au it also predicts
peaks at 1 and 3, no peaks at 2 and 4 Go, whereas the room temperature
STM experiments show peaks at 1,2,3 and 4 Go. On the other hand, low
567
Na
4.2
1000
750
:l
.5
C>
!:I.
500
250
34567
Conductance (WIh)
Figure 12. Histogram of conductance values, constructed from G(Vp)-curves measured

for Na at 4.2 K using an MeB device. A total of 105 individual measurements on two
samples were used. The characteristic sequence of peaks (G = 1,3,5,6) forms a clear
signature for conductance quantization. (From: [58])
temperature MCB experiments for Cu show a peak at 1, a much smaller

peak at 3, and no other quantum numbers, and for Au only a peak at 1 Go is
observed [58, 18]. The next section compares these results to several other
recent experiments and forms an attempt to find an explanation for the
differences in the results.
6. Origin of differences in the experiments

This section is necessarily somewhat speculative, because most of the experimental results are very new, and have not been confirmed by other
groups. Ideas about the origin of the differences in the results are emerging
at this moment, and have mostly not been tested. However, in view of the
striking differences between the results of the various experiments, a few
remarks must be made.
6.1. MATERIAL DEPENDENCE
The choice of metal used in the experiment can be expected to play an

important role. The fact that Na gives results which are closest to the idealised free electron approximation of a symmetric contact are most likely
related to the free electron character of bulk Na. Sodium has one valence
electron which is very loosely bound to the atom core, which results in (1)
nearly free electron waves in bulk Na metal, where the electrons hardly
feel the presence of the nuclei, and the Fermi surface is very well approximated by a free electron sphere; and (2) the atomic scale corrugation at
the surface is almost completely smeared out. The latter is illustrated by
568
the electron cloud for a Na2 molecule, which instead of showing a dumbbell
shape, is nearly spherical at a distance of half the inter-atomic spacing [59].
This is already different for the monovalent noble metals Cu and Au, for
which the atomic corrugation at the surface can be resolved with an STM,
and for which the Fermi surface is distorted to the extent that the Fermi
spheres in adjacent Brillouin zones are connected by necks. These effects
are expected to introduce real-space and k-space scattering, respectively,
which may suppress the quantum features in the conductance. For metals
having a still more complicated electronic structure, various parts of the
Fermi surface can give inter-dependent contributions to the conductance.
The observation [17] that the last contact for Pt has an average value of
1.6 Go was attributed to the contributions of s- and d- electrons, giving rise
to more than one conductance mode for a single atom. This observation was
recently confirmed in a low temperature STM experiment [33], where the
authors added a tight-binding calculation for one-atom point contacts of
Au, Pt and Ni. The calculation confirms that a single atom of Au has a
conductance close to 1 Go, but that for Ni and Pt this value can be a factor ,,-,2 higher. On the other hand, room temperature STM histograms for
Pt show a clear peak at 1 Go. In addition to the electronic structure, the
atomic scale mechanical properties of the materials will be of influence on
the histograms. Of particular relevance is the melting point of the material,
in comparison to the temperature of the experiment. This is the subject of
the next paragraph.
6.2. TEMPERATURE DEPENDENCE
One of the most striking differences in the experiments is the appearance

of peaks at the first 4 quantum numbers in the histogram for Au and
Cu, obtained in the room temperature STM experiments [29, 30, 32, 37,
61]' whereas a histogram for Au measured at low temperature using the
MCB technique showed only a peak at 1 Go [18] and peaks at 1 and 3 Go
for Cu. The latter suggests that for Cu the same degeneracy of modes
exists as was observed for Na. One of the possibilities being investigated
is the temperature dependence of the results. Using a room-temperature
UHV version of the MCB Muller et al. [60] observed peaks at the first 5
quanta for Cu and Au, demonstrating that similar results are obtained for
different techniques at the same temperature. Also, at 77K with a standard
MCB set-up, peaks were observed for Au at 1, 2 and 3 Go [18]' showing
that temperature does indeed influence the peak structure. On the other
hand, STM experiments at helium temperature [32, 61] show no influence
of temperature on the peak structure. A possible explanation is given in
section 6.3.
569
A second example of the influence of temperature is found in the investigations on Pt mentioned in the previous section. At 4.2 K and below,
the last conductance value before the jump to tunnelling shows an average
value of 1.6 Go, both in MCB [17J and in STM [33J experiments, whereas
room temperature STM results give 1 Go [57J.
A mechanism for the temperature dependence of the conductance steps
was suggested by Bratkovsky et al. [46J. They argued that an elevated
temperature should assist the contact in finding a favourable shape. The
higher mobility of the atoms would allow them to reduce irregularities at
the surface, and make the shape of the contact longer and closer to adiabatic. Molecular dynamics simulations for various temperatures [46J are
consistent with this idea. Although it is likely that this mechanism is of importance, it does not explain the value of 1 Go for Pt at room temperature
if a single atom contact gives already 1.6 Go, and it makes it even more
difficult to explain why degenerate modes are observed for Na and Cu at
low temperatures, and not at room temperature.
6.3. SPEED OF CONTACT BREAKING
An aspect which is related to the temperature dependence, is the speed at

which the contacts are formed or broken. At each step in the conductance
the elastic strain energy, which has been built up during the elongation of
the contact on a conductance plateau, is suddenly released. This energy
is released in the form of lattice vibrations, which may take some time to
diffuse into the banks due to the restricted geometry and lattice mismatch.
The speed of contact breaking ranges from less than 10- 1 Als in the MCB
experiments to '" 105 Als in the STM experiments and", 3 . 1010 Als in
the molecular dynamics simulations. In the latter the problem is probably
circumvented by the inclusion of a numerical thermostat on each atom.
In the fast experiments, it is possible that non-equilibrium lattice vibrations generated at previous steps, influence the behaviour at the next step.
Experiments to test this conjecture are under way.
6.4. INDENTATION DEPTH AND CONTACT 'TRAINING'
Experiments at various indentation depth suggest that longer necks are

pulled between the two electrodes when indenting them deeper into each
other, and after repeated indentation ('training' of the contact) [62J. The
length of the neck can be estimated from the global variation of the conductance with displacement. Longer necks may be favourable for quantization
as a result of a shape which is close to adiabatic.
This concludes our discussion of conductance quantization, and we will
now turn to a brief review of other properties of quantum point contacts
570
of metals.
7. Atom size superconducting junctions

When the cross section of a conductor is reduced to the order of the Fermi
wavelength, many properties are expected to be modified, notably the superconducting properties. As an example, Beenakker and van Houten [63]
predicted that the critical current, Ie, of a quantum point contact should
show quantization, in the sense that the product of Ie and the normal state
resistance Rn is constant, IeRn = 7r!:1/e, where !:1 is the energy gap of the
superconductor. When the width of the contact increases such that a new
quantum mode causes to the conductance to increase by 2e 2 / h, the supercurrent increases by !:1e/n. In contrast to the conductance, the quantum
increments of the supercurrent have no universal value, but depend on the
properties of the material.
One approach to study the quantum properties of superconducting contacts, is using a 2DEG in intimate contact with a metallic superconductor
at close proximity to the point contact. The three dimensional superconductor induces superconductivity in the 2DEG by the proximity effect, and
one has direct control over the width of the point contact by means of a
gate electrode, just as for the conductance quantization measurements. Despite the great difficulty in sample preparation, experiments of this type
are producing the first results, and this is discussed in the chapter by van
Wees and Takayanagi.
Using the MCB technique, Muller et al. [14, 15] measured the IeRn
product for Nb point contact as a function of the contact resistance, from
,.,.,1 n up into the tunnel regime. The product was found to be smaller than
the theoretical value, and especially above 100 n a systematic decrease was
observed, which extrapolates to zero at a resistance close to the quantum
value. Similar behaviour was observed for a number of other bulk superconductors [64]. The value of Ie in these studies was determined from the
maximum supercurrent, i.e. the current at which a very small threshold
voltage was detected. However, the present view is that the true supercurrent is hidden by thermal fluctuations, and that what is measured is the
thermally activated switching current out of the zero voltage state [65].
Measurement of the true supercurrent would require cooling down to much
lower temperatures and introducing a damping of the junction by mounting
a resistor at close proximity to the contact. Experiments with this aim have
been started by C. Urbina in Saclay. Nevertheless, it is interesting to note
that for the smallest contacts, at which the conductance changes in jumps
due to the atomic rearrangements, simultaneous jumps in the supercurrent
can be observed [14, 15, 66]. As illustrated in Fig. 13 the steps in Ie and in
571
~
. . .~
3.5
'/jJA) r
3
2.5
~ - - -"- - - - -l
.r/.
~.-~.~~
R (0) ,
;00
600
500
lt
J~
:~'.;, ~
',': I
1.65 .
\000
~
-'-----L........_, .....
timers)
1500
Figure 13. Critical current and normal resistance measured for a Nb contact at 1.2
K using a MCB device. The piezo voltage was swept over a small range such that the
curves are reproducible. During one cycle the normal resistance is measured at bias
voltages above the gap, and during a next cycle the critical current is measured. The
product feRn is nearly constant; note the expanded scale. (From: [15])
Rn are such that the product IeRn remains nearly constant, albeit at a value
which is about 3 times smaller than the theoretical value, 1fD./e = 4.5 mY.
In a recent experiment Koops et at. [22] used a MCB device shunted by a
superconducting loop. The shunting makes the junction much less sensitive
to thermal fluctuations, and by simultaneous measurement of the current
through the leads and the flux through the loop it is possible to directly
observe the current-phase relation of the junction. For a point contact, in
contrast to a tunnel junction, this relation is not sinusoidal, and the results
extrapolate to the theoretical prediction for small sizes of the loop, and for
point contacts down to a few atoms in cross section.
For standard junctions in the vacuum tunnelling regime the supercurrent is almost completely suppressed, and the current-voltage characteristics is dominated by quasiparticle tunnelling. Apart from the usual sharp
increase in the current at eV = 2D., smaller current steps are observed
at eV = 2D./2 and 2D./3 [14]. The latter features are known as subgap
structure, and have been observed in planar oxide barrier tunnel junctions.
They have been measured as a function of the tunnelbarrier for Nb and Pb
junctions by van der Post et al. [67] (see Fig. 14). The subgap structure
is believed to result from higher order tunnel processes, involving n quasi-
572
40
145kQ
Nb
T = 1.4 K
30
r--I
<~
'---'
......
20
10
0
0
V [mY]
Figure 14. Current-voltage characteristics for a Nb vacuum tunnel junction, using an
MCB device at 1.4 K. In addition to the usual steep rise of the current at eV = 2~ there
are smaller current onsets at 2~/n, as is clear on the expanded scale for n = 2 and 3.
(From: [67))
particles for the current step at 2!:1/n. The theory has been developed in
the perturbative regime by Schriefi'er and Wilkins [68], known as Multiple
Particle Tunnelling, and for SNS junctions by Klapwijk, Blonder and Tinkham [69], where it is known as Multiple Andreev Reflection. In the planar
junction experiments, however, the subgap current steps were found to be
orders of magnitude larger than predicted, which has been attributed to
the non-uniformity of the tunnelbarrier [70].
One can argue that the vacuum tunnel current in and STM or MCB
experiment is carried by a single quantum mode. The last contact conductance value before the jump to tunnelling is between 1 and 2 Go for
most superconductors as a result of contributions of several orbital atomic
wave functions (see section 6.1). However, as soon as contact is broken,
tunnelling takes place between the two front- most atoms, and the tunnelling electrons experience a saddle point in the potential profile midway
between the atoms. As a result, a single mode, having the highest tunnelling
probability, will dominate the tunnel current. This simplifies the theoretical
description to such an extent, that direct comparison with experiment is
possible. A nearly perfect agreement with theory [71, 72, 73] was indeed
obtained [67]. Here the only free parameter is the tunnelling probability,
and this can be fixed by measuring the normal state resistance.
573
8. Size effects in the scattering on magnetic impurities

The scattering of conduction electrons on magnetic impurities gives rise to
a characteristic minimum in the resistivity at low temperatures followed
by a logarithmic increase when the temperature is lowered further. This is
known as the Kondo effect, and it is caused by the resonant scattering of the
conduction electron spin with the impurity spin. In the limit of T TK,
where TK is the characteristic energy scale of the Kondo problem, the conduction electrons form a cloud around the impurity with a net polarisation
such that the magnetic moment of the impurity is effectively screened. In
point contacts of dilute magnetic alloys Kondo scattering is observed as a
maximum in the differential resistance, dV/ dI, as a function of bias voltage [74]. The predictions for the maximum at V = 0 in the differential
resistance of point contacts are completely analogous to the theory for the
maximum at T=O in the resistivity of bulk materials.
We can now study the question of how the Kondo scattering is affected
by the size of the point contact. Experiments with this aim have been
performed on a alloy of Cu with 0.1 % Mn impurities, using an MCB device
[75]. The surprising result of this study is that the effective scattering cross
section per impurity increases for smaller contacts. When the results are
analysed using the standard theory the Kondo temperature TK is observed
to increase by five orders of magnitude between the bulk metal and a contact
of 2 nm in diameter.
An explanation for this size effect has been proposed by Zarand and
Udvardi [76J in terms of the strong variation of the density of states in
the contact resulting from the limited number of modes available. The
density of states fluctuates both as a function of energy and as a function of
position in the contact. The Kondo temperature depends on the density of
states N F through the relation TK = TF exp( -1 / J N F), with TF the Fermi
temperature and J the exchange interaction energy between the conduction
electrons and the impurity spin. The fluctuations in N F are thus amplified
in variations in the Kondo temperature for an impurity, depending on its
position in the contact. The impurities obtaining a high Kondo temperature
will dominate the differential resistance, resulting in a high effective Kondo
temperature for the contact.
The size dependence predicted by this mechanism would be most dramatic for small J, i.e. low bulk Kondo temperatures [76]. For Cu(Mn) we
have TK ~ 10- 2 K. Experiments on Cu(Fe), having a bulk Kondo temperature of about 30 K, showed indeed a much more modest size effect [77].
Although this observation is in qualitative agreement with the prediction,
quantitatively there are large discrepancies, and for small contact diameters much larger scatter in the experimental data would be expected as a
574
o
....
o
....
before onneol
o
o
N
~
.J::
....
'-0
0
NCl)
'-' 0
0
....
<':>0
!Xl
Ol
1'1
to
Ol
1'1_ 20
-10
10
20
V (mV)
Figure 15. Differential conductance for a nanofabricated eu point contact at 4.2 K,
measured within a few hours after fabrication. The minimum at zero bias, and the excess
conductance, which is suddenly destroyed at about 10 m V, are attributed to the scattering
of the conduction electrons on a TLF associated with a defect in the contact. The lower
trace is measured on the same device after allowing it to anneal for 2 days at room
temperature, and the anomalies seen in the first curve have disappeared. (From: [79])
result of the random distribution of the impurities. More work is clearly

needed, but the results demonstrate that the quantization of the modes in
a contact may dramatically modify properties other than the conductance.
9. Two-level fluctuations
It turns out to be possible to observe a maximum in the differential resistance at zero bias, even for a pure metal point contact without magnetic
impurities [78, 79, 80]. An example is given in Fig. 15. The features have
been observed in nanofabricated point contacts, MCB devices and spearanvil type point contacts. Experiments on the first type of contacts shows
that the anomaly can be removed by annealing the device at room temperature for a few hours. The results have been interpreted as being due to
two-level fluctuations (TLF). The microscopic origin of the TLF should be
a lattice defect, most probably a dislocation kink [79].
We have encountered TLF already in section 4, where they are seen as
random switchings between two values of the conductance on a laboratory
time scale. They result from a switching of the contact between two atomic
scale structures. Similar TLF have been studied in larger contacts [10],
where they are observed as a relatively small (",,0.1%) fluctuation in the
resistance, which are attributed to defect motion in the contact. At higher
temperatures more and more TLF become active and a large ensemble
of these TLF is believed to be responsible for the ubiquitous l/f noise in
metals. The transitions between the two states are thermally activated and
slow enough, so that individual transitions can be resolved.
575
These slow TLF should be distinguished from fast TLF which lead to
the Kondo-like zero bias anomaly in the differential resistance. For the
latter, transitions between the two states are induced by (nearly) elastic scattering by conduction electrons. The correspondence to the Kondo
problem is intuitively appealing: in Kondo scattering conduction electrons
induce transitions between two spin states of an impurity, and for the fast
TLF transitions between two atomic configurations are induced. For a system with two nearly degenerate levels there is a formal analogy with the
Kondo problem [81]. However, when taking spin degeneracy into account
the low temperature properties are described by a multi-channel Kondo
model leading to a non-Fermi liquid ground state [82]. The fact that the
results obtained by Ralph and Buhrman coincide with the predictions for
a multi-channel Kondo model has attracted a lot of attention. In addition,
the experiments give evidence for a many-body ground state of the conduction electrons interacting with the TLF, which gives a contribution which
lowers the resistance. A first-order phase transition out of this many body
state is induced by a finite current through the contact, and some analogies
with superconductivity have been pointed out [79].
10. Conclusions
Experiments on atomic size contacts are surprisingly simple to perform,
and, what I find even more surprising, they demonstrate that conductance
quantization can be observed. The mechanics on the atomic level is apparently such that quantization effects are not completely washed out. There
is a clear challenge to understand the atomic scale dynamics, in relation
to the electronic wave functions in the contact. Where the most realistic
conductance calculation for a single atom contact of a monovalent metal
produces values which are distinctly different from 2e 2 jh, the experiment
most explicitly shows a strong preference for a single quantum of conductance. This problem, together with the differences in behaviour for different
elements, is among the most interesting to be solved.
Apart from the problem of conductance quantization, many other properties of point contacts remain to be explored. The last few sections give an
(incomplete) overview of the properties studied up to this moment and in
each case there are lots of open questions. And many properties have not
even been investigated. I believe that the unsolved problems in quantum
point contacts between metals offer perspectives for many years of fruitful
investigations, with implications for a wide range of fundamental fieldsof
research.
I have profited from stimultating discussions with L.J. de Jongh, J.M.
Krans, C.J. Muller, N. van der Post, LK. Yanson, M.H. Devoret, D. Esteve
576
and C. Urbina. This work was supported by the "Stichting FOM".
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
van Wees, B.J., van Houten, H., Beenakker, C.W.J., Williamson, J.G., Kouwenhoven, L.P., van der Marel, D. and Foxon, C.T. (1988) Quantized conductance of
point contacts in a two-dimensional electron gas, Phys. Rev. Lett. 60, pp. 848-850 .
Wharam, D.A., Thornton, T.J., Newbury, R., Pepper, M., Ahmed, H., Frost,
J.E.F., Hasko, D.G., Peacock, D.C., Ritchie, D.A. and Jones, G.A.C. (1988) Onedimensional transport and the quantization of the ballistic resistance, J. Phys. C 21,
pp. L209-L214 .
Gimzewski, J.K and WeIland, M.E., eds. (1995) The ultimate limits of fabrication
and measurement, NATO-ASI Series E: Applied Sciences, 292 Kluwer Academic
Publishers, Dordrechtj Singer, I.L. and Pollock, H.M., eds. (1992) Fundamentals
of friction:macroscopic and microscopic processes, NATO-ASI Series E: Applied
Sciences, 220 Kluwer Academic Publishers, Dordrecht.
Khotkevich, A.V. and Yanson, I.K(1995) Atlas of point contact spectra of electronphonon interactions in metals. Kluwer Academic Publishers, Dordrecht.
van Ruitenbeek, J.M. and Jansen, A.G.M., eds. (1996) Proceedings of the Second
International Conference on Point Contact Spectroscopy, Physica B 218.
Yanson, I.K (1974) Nonlinear effects in the electric conductivity of point junctions
and electron-phonon interaction in metals, Zh. Eksp. Teor. Fiz. 66 pp. 1035-1050
(1974 Sov. Phys.-JETP 39 pp. 506-513).
Jansen, A.G.M., van Gelder, A.P. and Wyder, P. (1980) Point-contact spectroscopy
in metals, J. Phys. C: Solid St. Phys. B 13 pp. 6073--6118.
Tsoi, V.S. (1974) Focussing of electrons in a metal by a transverse magnetic field,
ZhETF Pis. Red. 19 pp. 114--116 (1974 JETP Lett. 19 pp. 70-71).
Benistant P.A.M., van Kempen, H. and Wyder, P. (1983) Direct observation of
Andreev reflection, Phys. Rev. Lett. 51 pp. 817-820.
Ralls, KS. and Buhrman, R.A. (1988) Defect interactions and noise in metallic
nanoconstrictions Phys. Rev. Lett. 60 pp. 2434-2438.
Gimzewski, J.K, and Moller, R. (1987) Transition from the tunneling regime to
point contact studied using scanning tunneling microscopy, Phys. Rev. B, 36,
pp. 1284-1287.
Smith, D.P.E. (1995) Quantum point contact switches, Science, 269, pp. 371-373.
Moreland, J and Ekin, J.W. (1985) Electron tunneling experiments using Nb-Sn
"break" junctions, J. Appl. Phys. 58 pp. 3888-3895.
Muller, C.J., van Ruitenbeek, J.M., and de Jongh, L.J. (1992) Experimentalobservation of the transition from weak link to tunnel junction, Physica C 191, pp. 485-504.
Muller, C.J., van Ruitenbeek, J.M. and de Jongh, L.J., (1992) Conductance and
supercurrent discontinuities in atomic-scale metallic constrictions of variable width,
Phys. Rev. Lett., 69, pp. 140-143.
Muller, C.J (1996) An experimental study on mechanically controllable break junctions PhD. thesis, Leiden, The Netherlands.
Krans, J.M., Muller, C.J., Yanson, I.K, Govaert, Th.C.M., Hesper, R. and van
Ruitenbeek, J.M. (1993) One-atom point contacts, Phys. Rev. B, 48, pp. 1472114724.
Krans, J.M. (1996) Size effects in atomic-scale point contacts PhD. thesis, Leiden,
The Netherlands.
van der Post, N. and van Ruitenbeek, J.M. (1996) High stability STM made of a
break junction Proceedings of the 21st Int. Conf. on Low. Temp. Phys., Chech. J.
Phys. 46- suppl. pp. 2853-2854.
Ruitenbeek, J.M. van, Alvarez, A., Piiieyro, 1., Grahmann, C., Joyez, P., Devoret,
M.H., Esteve, D., and Urbina, C. (1996) Adjustable nanofrabricated atomic size
577
contacts, Rev. Sci. Instrum., 67, pp. 108-111.
Zhou, C., Muller, C.J., Deshpande, M.R., Sleight, J.W. and Reed, M.A. (1995)
Microfabrication of a mechanically controllable break junction in silicon, Appl. Phys.
Lett., 67, pp. 1160-1162.
22. Koops, M.C. and de Bruyn-Ouboter, R. (1996) Direct observation of the currentphase relation of an adjustable superconducting point contact Phys. Rev. Lett. in
print.
23. Agrai:t, N., Rodrigo, J.G. and Vieira, S. (1993) Conductance steps and quantization
in atomic-size contacts, Phys. Rev. B 47, pp. 12345-12348.
24. Pascual, J.I., Mendez, J., Gomez-Herrero, J., Baro, A.M. and Garda, N., (1993)
Quantum contact in gold nanostructures by scanning tunneling microscopy, Phys.
Rev. Lett., 71, pp. 1852-1855.
25. Pascual, J.I., Mendez, J., Gomez- Herrero, J., Baro, A.M., Garcia, N., Landman,
V., Luedtke, W.D., Bogachek, E.N. and Cheng, H.-P., (1995) Properties of metallic
nanowaires: From conductance quantization to localization, Science, 267, pp. 17931795.
26. Dremov, V.V. and Shapoval, S.Yu. (1995) Quantization of the conductance of metal
nanocontacts at room temperature, JETP Lett., 61, pp. 336-339.
27. Landman, V., Luedtke, W.D., Salisbury, B.E. and Whetten, R.L. (1996) Reversible
manipulations of room temperature mechanical and quantum transport properties
in nanowire junctions, Phys. Rev. Lett., 77, pp. 1362-1365.
28. Olesen, L., Lcegsgaard, E., Stensgaard, I., Besenbacher, F., Schi0tz, J., Stoltze, P.,
Jacobsen, K.W. and N0rskov, J.K., (1994) Quantized conductance in an atom-sized
point contact, Phys. Rev. Lett., 72, pp. 2251-2254.
29. Brandbyge, M., Schi0tz, J., S0rensen, M.R., Stoltze, P., Jacobsen, K.W., N0rskov,
J.K., Olesen, L., Lcegsgaard, E., Stensgaard, I. and Besenbacher, F. (1995) Quantized conductance in atom-sized wires between two metals, Phys. Rev. B 52,
pp. 8499-8514.
30. Gai, Z., He, Y, Yu, H., and Yang, W.S. (1996) Observation of conductance quantization of ballistic metallic point contacts at room temperature, Phys. Rev. B 53,
pp. 1042-1045.
31. Agrai:t, N., Rodrigo, J.G., Sirvent, C. and Vieira, S. (1993) Atomic-scale connective
neck formation and characterization, Phys. Rev. B 48, pp. 8499-8501.
32. Sirvent, C., Rodrigo, J.G., Agrait, N. and Veira, S. (1996) STM study of the atomic
contact between metallic electrodes, Physica B 218, pp. 238-241.
33. Sirvent, C., Rodrigo, J.G., Vieira, S., Jurczyszyn, L., Mingo, N. and Flores, F. (1996)
Conductance step for a single-atom contact in the scanning tunneling microscope:
noble and transition metals, Phys. Rev. B 53, pp. 16086-16090.
34. Rubio, G., Agrait, N., and Vieira, S. (1996) Atomic-sized metallic contacts: mechanical properties and electronic transport, Phys. Rev. Lett. 76, pp. 2302-2305.
35. Agrai:t, N., Rubio, G., and Vieira, S. (1995) Plastic deformation of nanometer-scale
gold connective necks, Phys. Rev. Lett. 74, pp. 3995-3998.
36. Stalder, A., and Diirig, V. (1996) Study of plastic flow in ultrasmall Au contacts,
J. Vac. Sci. Technol. B, 14, pp. 1259-1263.
37. Costa-Kramer, J.L., Garda, N., Garda-Mochales, P., and Serena, P.A. (1995)
Nanowire formation in macroscopic metallic contacts: quantum mechanical conductance tapping a table top, Surface Science342, pp. L1144-L1149.
38. Sharvin, Yu.V. (1965) A possible method for studying Fermi surfaces, Sov. Phys.JETP 21 pp. 655-656.
39. Weyl, H. (1911) Veber die asymptotische Verteilung der Eigenwerte Nachr. Akad.
Wiss. Gottingen, pp. 110-117.
40. Garda-Martin, A., Torres, J.A. and Saenz, J.J. (1996) Finite size corrections to the
conductance of ballistic wires, Phys. Rev. B, in print.
41. Bogachek, E.N., Zagoskin, A.N. and Kulik, 1.0. (1990) Conductance jumps and
21.
578
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
magnetic flux quantization in ballistic point contacts, Sov. J. Low Temp. Phys. 16,
pp. 796-800.
Glazman, L.L, Lesovick, G.B., Khmel'nitskii, D.E. and Shekhter, R.L (1988) Reflectionless quantum transport and fundamental ballistic-resistance steps in mesoscopic
constrictions, JETP Lett., 48 pp. 238-241.
Bratkovsky, A.M. and Rashkeev, S.N. (1996) Electronic transport in nanoscale contacts with rough boundaries, Phys. Rev. B 53, pp. 1-12.
Torres, J.A, Pascual, J.L and Saenz, J.J. (1994) Theory of conduction through narrow constrictions in a three-dimensional electron gas, Phys. Rev. B 49, pp. 1658116584.
Todorov, T.N., Briggs, G.A.D. and Sutton, A.P. (1993) Elastic quantum transport
through small structures, J. Phys.: Condo Matter 5, pp. 2389-2406.
Bratkovsky, A.M., Sutton, A.P. and Todorov, T.N. (1995) Conditions for conductance quantization in realistic models of atomic-scale metallic contacts, Phys. Rev.
B 52, pp. 5036-5051.
Todorov, T.N. and Sutton, A.P. (1993) Jumps in electronic conductance due to
mechanical instabilities, Phys. Rev. Lett. 70, pp. 2138-2141.
Lang, N.D. (1987) Resistance of a one-atom contact in the scanning tunneling microscope, Phys. Rev. B 36, pp. 8173-8176.
Lang, N.D. (1995) Resistance of atomic wires, Phys. Rev. B 52, pp. 5335-5342.
Ferrer, J., Martin-Rodero, A. and Flores, F. (1988) Contact resistance in the scanning tunneling microscope at very small distances, Phys. Rev. B 38, pp. 1011310115.
Krans, J.M., van Ruitenbeek, J.M. and de Jongh, L.J. (1996) Atomic structure and
quantized conductance in metal point contacts, Physica B 218, pp. 228-233.
Landman, V., Luedtke, W.D., Burnham, N.A. and Colton, R.J. (1990) Atomistic
mechanisms and dynamics of adhesion, nanoindentation and fracture, Science 248,
pp. 454-461.
Krans, J.M. and van Ruitenbeek, J.M. (1994) Subquantum conductance steps in
atom-sized contacts of the semimetal Sb, Phys. Rev. B50, pp. 17659-17661.
Torres, J.A., and Saenz, J.J. (1996) Conductance steps in point contacts: quantization or cross-section jumps? Physica B, 218, pp. 234-237.
Torres, J.A. and Saenz, J.J. (1996) Conductance and mechanical properties of
atomic-size metallic contacts: a simple model, Phys. Rev. Lett., in print.
Todorov, T.N., and Sutton, A.P. (1996) Force and conductance jumps in atomicscale metallic contacts, Submitted to Phys. Rev. B, Rapid Commun ..
Olesen, L., Lcegsgaard, E., Stensgaard, 1., Besenbacher, F., Schi0tz, Stoltze, P.,
Jacobsen, K.W. and N0rskov, J.K. (1995) Reply on Comment on Quantized conductance in an atom-sized point contact, Phys. Rev. Lett., 74, p. 2147.
Krans, J.M., Ruitenbeek, J.M. van, Fisun, Yanson, LK. and Jongh, L.J. de (1995)
The signature or conductance quantization in metallic point contacts, Nature, 375,
pp. 767-769.
Moore, W.J. (1972) Physical Chemistry. Longman Publishers, London, p. 692.
Muller, C.J., Krans, J.M., Todorov, T.N., and Reed, M.A. (1996) Quantization
effects in the conductance of metallic contacts at room temperature, Phys. Rev. B
53, pp. 1022-1025.
Garcia, N., this volume.
Rubio, G. and Agrait, N., private communication.
Beenakker, C.W.J. and van Houten, H. (1991) Josephson current through a superconducting quantum point contact shorter than the coherence length, Phys. Rev.
Lett., 66, pp. 3056-3060.
Muller, C.J., Koops, M.C., Vleeming, B.J., de Bruyn Ouboter, R. and Omelyanchouk, A.N. (1994) Reduction of the Andreev scattering processes in ultra-small
superconducting point contacts with direct conductivity, Physica C 220, pp. 258264.
579
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
Urbina, C., Esteve,D. and Devoret, M.H. (1996), private communications.

Vleeming, B.J., Muller, C.J., Koops, M.C. and de Bruyn Ouboter, R (1994) Singleatom point contacts in the superconducting state, Phys. Rev. B, 50, pp. 1674116744.
van der Post, N., Peters, E.T., Yanson, I.K. and van Ruitenbeek, J.M. (1994) Subgap
structure as a function of the barrier in atom-size superconducting tunnel junctions,
Phys. Rev. Lett. 73, pp. 2611-2613.
Schrieffer, J.R and Wilkins, J.W. (1963) Two-particle tunneling processes between
superconductors, Phys. Rev. Lett. 10, pp. 17-20.
Klapwijk, T.M., Blonder, G.E. and Tinkham, M. (1982) Explanation of sub harmonic
energy gap structure in superconducting contacts, Physica 109 & nOB, pp. 16571664.
Kleinsasser, A.W., Miller, RE., Mallison, W.H. and Arnold, G.B. (1994) Observation of multiple Andreev reflections in superconducting tunnel junctions, Phys. Rev.
Lett. 72 pp. 1738-1741.
Arnold, G.B. (1987) Superconducting tunneling without the tunneling Hamiltonian.
II. Subgap harmonic structure, J. Low Temp. Phys., 68 pp. 1-27.
Bratus, E.N., Shumeiko, V.S. and Wendin, G. (1995) Theory of subharmonic gap
structure in superconducting mesoscopic tunnel contacts, Phys. Rev. Lett., 74,
pp. 2110-2113.
Averin, D. and Bardas, D. (1995) ac Josephson effect in a single quantum channel
Phys. Rev. Lett., 75, pp. 1831-1834.
Naidyuk, Yu.G., Shklyarevskii, 0.1. and Yanson, I.K. (1982) Microcontact spectroscopy of dilute magnetic alloys CuMn and CuFe, Sov. J. Low Temp Phys., 8,
pp.362-365
Yanson, I.K., Fisun, V.V., Hesper, R, Khotkevich, A.V., Krans, J.M., Mydosh,
J.A. and van Ruitenbeek, J.M. (1995) Size dependence of Kondo scattering in point
contacts Phys. Rev. Lett. 74, pp. 302-305
Zarand, G. and Udvardi, L. (1996) Role of the local density of states fluctuations
in metallic point contatcs, Physica B 218 pp. 68-72. and preprint.
van der Post, N., Mettes, F.L., Mydosh, J.A., van Ruitenbeek, J.M. and Yanson,
I.K. (1996) Size dependence of Kondo scattering in point contacts: Fe impurities in
Cu, Phys. Rev. B, 53 pp. R476-R479.
Ralph, D.C. and Buhrman, RA. (1992) Observations of Kondo scattering without
magnetic impurities: a point contact study of two-level tunneling systems in metals,
Phys. Rev. Lett., 69 pp. 2118-2121.
Ralph, D.C. and Buhrman, RA. (1995) Kondo scattering from atomic two-level
tunneling systems in metals: Enhanced conductance, critical-bias transitions, and
the non-Fermi-liquid electronic state, Phys. Rev. B, 51 pp. 3554-3568.
Keijsers, RJ.P., Shklyarevskii, 0.1. and van Kempen, H. (1995) Two-Ievel-systemrelated zero-bias anomaly in point-contact spectra, Phys. Rev. B, 51 pp. 5628-5634.
Zawadowski, A. (1980) Kondo-like state in a simple model for metallic glasses, Phys.
Rev. Lett., 45 pp. 211-214.
Muramatsu, A. and Guinea, F. (1986) Low-temperature behavior of a tunneling
atom interacting with a degenerate electron gas, Phys. Rev. Lett., 57 pp. 23372340.
CONDUCTANCE QUANTIZATION IN METALLIC NANOWIRES
N. GARCIA, J.L. COSTA-KRAMER, A. GIL, M.1. MARQUES and

A.CORREIA
Laboratorio de Fisica de Sistemas Pequenos (CSJC-UAM)
Universidad Aut6noma de Madrid, C-JX
Cantobianco, E-28049-Madrid, Spain
Abstract.
Recent investigations carried out in our group concerning the conductance quantization
of metallic nanowires are reviewed. These include: i) The formation of metallic
nanowires between macroscopic electrodes, including liquid metals, demonstrating that
at the last stages of the contact breakage, a nanowire exists, independently of the initial
contact size. ii) A statistical study of the conductance using thousands of consecutive
contact breakage experiments, both at room and at liquid helium temperatures. These
histograms, totally reproducible, present clear peaks close to integer values of the
quantum of conductance Go=2e 2/h for diamagnetic metals like Gold, Silver, Copper,
Sodium, Platinum.... Ferromagnetic metals, Iron, Cobalt and Nickel, exhibit a flat
conductance histogram. This effect is attributed to the combination of the lifting of the
spin degeneracy in the ferromagnetic nanowires and the effect of geometry and disorder.
The measured conductance histograms are basically independent of the temperature. iii)
A discussion of the position and width of the observed peaks. Just geometrical effects
can not explain the large conductance peak shifts observed experimentally, and disorder,
behaving as a residual resistance, has to be invoked to explain them. iv) First realization
of conductance quantization in Bi at 4K. Conductance plateaus lasting 20-100 nm
electrode separation are presented; the histogram displays also clear peaks. v) A
statistical study of the conductance plateau duration, demonstrating a broad distribution
of this duration, 0.05-0.4 nm, with an average value that decreases as conductance
increases. vi) A discussion of force and energy quantization within a resonant energy
model of two reservoirs connected by a ballistic channel. vii) Experiments performed in
ultra high vacuum, where we manage to stabilize the nanowires for hours and study
switching and current voltage characteristics for different quantum conductance
channels with remarkable accuracy. viii) Visualization inside scanning and transmission
electron microscopes of the metallic contact between two macroscopic electrodes at the
micron and nanometer scales. These experiments provide experimental evidence of the
formation of a connective neck between the electrodes. ix) Experiments on light
emission from breaking nanowires. A plausible explanation for this phenomenon is
presented.
581
582
1. Introduction
The conductance through a small constriction in a two-dimensional electron gas (2DEG)

is quantized [1] when its width varies, being the quantum of conductance Go=2ilh,
where e is the electron charge and h the Planck constant. The study of electronic
transport through small metallic and non-metallic contacts has been a topic of increasing
interest since this first experimental evidence. The future of communication and
information processing technologies depends on the use of nanometric and submicron
integration as an essential tool [2]. Theoretical studies on the electronic conductance
through nanometric sized systems have been already carried out, including its
relationship with quantization and localization phenomena [3-13], as well as their
mechanical properties [14-17].
With the advent of the scanning tunneling microscope (STM) [18] and several related
techniques, it became possible to study with extreme accuracy the transition from tunnel
to contact regimes when two metallic electrodes (tip and sample) are approached. Since
then, STM has been widely used to study the phenomenon of conductance quantization
(CQ). This was usually performed driving a STM tip into a metallic sample and then
pulling it out. During this process a nanowire is formed, being possible to measure both
electrical [17,19,20,21] and mechanical [22, 23] properties. Conductance experiments
have been performed with the mechanically controllable break junction (MCBJ)
technique [24-27] as well. Switching behavior has been also induced between tunneling
and ballistic transport regimes by repeatedly bringing a sharpened nickel wire into
contact with a gold surface [28]. These approaches conclude that, at low and high
temperature, CQ occurs for several metallic species forming the contact.
Recently [29] it has been found that the same quantization features appear in
conductance experiments where two macroscopic wires are placed in contact, and the
formation and breakage of the contacts is produced by making them vibrate. With this
simple experimental technique, CQ has been observed for a wide variety of metallic
macroscopic contacts under different environments (air, liquids, ultra-high-vacuum
(UHV. These results, revealing the quantum nature of the conductance at the latest
stages of the contact breakage, obtained with the use of macroscopic electrodes, are
indistinguishable from those obtained with other experimental techniques based on the
control (via piezodisplacement techniques) of the separation of a sharp electrode. This
proves that, independently on the initial contact area, at the latest stages of the breakage
of a metallic contact, only a small bridge of nanometric dimensions (nanowire) is
present. This process is due probably to the formation of nanoscopic threads at the last
stages of the contact thin-out process, only one remaining at the very end.
In the present review we have analyzed more thoroughly the main properties of
nanoscopic contacts formed between two macroscopic wires using piezo-displacement
techniques based in STM methods as well as table top experiments. We have constructed
conductance histograms at low and room temperatures, using thousands of consecutive
experiments. Results are presented showing that we have stabilized these nanometric
583
sized contacts during time scales three orders of magnitude (several hours) higher than
previously reported measurements, characterizing its electronic behaviour with
unprecedented accuracy. We observe and study the nonlinear contribution to the
1-V characteristic curves. New statistical results on nanocontacts formed with STM are
presented. A statistical analysis of the length of the conductance plateaus provides
information about the correlation between force jumps due to atomic rearrangements
and conductance jumps due to the change of the number of electronic levels in the
constriction. Visualization inside a scanning electron microscope of the contact areas
between two macroscopic electrodes at the micron and nanometer scales have been
performed providing an experimental evidence of the formation of connective necks
between the electrodes. Experiments on light emission from breaking nanowires have
also been performed and a plausible explanation for this phenomenon is presented.
2. Table-top experiments
As mentioned above, it has been found [29] that the quantization features appear in
conductance experiments where two macroscopic wires vibrate in and out of contact. In
this section, this kind of experiments are briefly described. Figure la shows the
experimental setup: two wire ends with some tens of mV potential difference between
them are placed close to each other; the other ends are connected to a current-voltage (IV) converter working in the microamp range (10 5 gain). This simple setup is placed on a
table and the table tapped with the hand; the wires vibrate, getting in and out of contact.
During this process, the current signal is measured in a digital oscilloscope.
Independently of the setup (table-top, STM-like experiments, under UHV or in air) the
conductance measurements have been performed with two different techniques. One
measures the current through the contact with an I-V converter with 105 gain, I Ils rise
time and 3 IlS settling time. The other is even simpler; the current through the contact is
measured as a voltage drop across a 1 kO resistance placed in series with the contact.
Both measurement techniques give identical conductance results.
When the macroscopic wires separate, the contact area decreases and several
microscopic contacts are formed (a description of the initial stages of these microscopic
contacts will be addressed in Section 6). The separation process produces the stretching
of these microcontacts, increasing their length and decreasing their cross section. At the
latest stages of the process, only one contact remains, the cross section corresponds to
few atoms, and conductance quantization appears. In Figure I b, we show the result of a
typical conductance measurement when a contact between two macroscopic copper wires
is broken. In principle, this curve is indistinguishable of those obtained with more
sophisticated methods. This kind of experiments demonstrates that nanowires can be
formed by driving two metals of macroscopic dimensions and any size in and out and
contact [29].
584
lSJ
.0.0...
40
5
30
I-V
onverter
..... a
<"
2:
105 galn
. . . 1
:2
3 ~
20
2
10
(a)
(!)
0
0.0
0.2
0.4
time (ms)
0.6
Figure 1. (a) Experimental setup for table-top CQ experiments: two metallic wires are touching loosely
with a potential difference offew tens ofmV. The other ends are connected to the input leads of an I-V
converter. An oscilloscope traces the conductance of this system during the wires vibration.
(b) Conductance measurement during a table top experiment using two macroscopic copper wires
(0.6 mm diameter) and a potential difference of90.4 mV.
3. Liquid Metallic Nanowires

The table-top experiment has opened new interesting possibilities in the study of the
conductance of nanowires. The use of macroscopic electrodes allow us to study the
formation of nanocontacts in liquid systems, driving in and out of a liquid metal
(mercury or liquid tin) a macroscopic metallic copper tip until a nanocontact is formed.
Figure 2 depicts schematically the experimental setup.
(~~I~~'
~
.....
:J
Oscilloscope
"'
.,
.,
II
.-- Micrometric screw
.+
.11+---;--
Liquid nanowire
Liquid metal
Figure 2. Scheme of the experimental setup, based in results from table-top experiments, used for the
study of conductance quantization in liquid metals.
A micrometric screw is used in order to control the displacement of the tip inside the
liquid metal. In Figure 3a we show conductance measurements for mercury
585
nanocontacts, where clear conductance steps appear when driving the tip into contact
with the liquid metal. Same studies have been performed in liquid tin (at 300C) where
similar characteristics are obtained in the conductance measurements, as shown in
Figure 3b.
- - _ . Microscopic displacement - - -
70 ~;:::::======:::;--~--::;;J\,;M;n 10
63
56
49
42
35
6
5
28
21
()
5.
~
&)
14
In
0
0.0
0.2
0.4
0.6
0.8
1.0
Time(ms)
- - _ . Microscopic displacement - - -
70t,===~~~~--~-:-JJl
63
56
49
_
Co
!l
III
"
28
21
14
0.0
()
"c:
42
35
0.4
0.6
0.8
&)
N
In
1.0
Time(ms)
Figure 3. Conductance experiments showing the evolution of the current and the conductance at the first
stages of the formation of a liquid metal contact. The contact forms between a cooper wire and (a)
mercury ( at room temperature) and (b) liquid tin (at 300C). The applied bias voltage between tip and
the metallic liquid reservoir is 90.4mV.
The general features are very similar those obtained in the case of the experiments
performed on solid metallic contacts at the nanometer scale, however we find that liquid
contacts present less plateaus but more pronounced than in the solid case.
4. STM Nanowire Experiments: Massive Conductance Statistics with No Sample
Selection
These experiments are performed in a home built Scanning Tunneling Microscope,
STM. Both tip and sample are high purity metals. Typically, a few tens ofmV potential
difference is applied between them. The current is measured while the tip is repeteadly
586
crushed into and retracted from the sample, driving the tip piezoelectric actuator with a
triangular wave. The tip speed is controlled by varying the amplitude and/or frequency
of this signal. The current is measured with a current-voltage converter working at
105 gain (100 mV/IlA), with a 90% rise time of 1 IlS and a settling time of about 3 IlS.
The current signal is measured with a LeCroy 9354AM digital oscilloscope with
500 MHz bandwidth and 2 Gsample/s sampling rate. The data acquisition is triggered
when the current crosses a predetermined value with a predetermined slope. In the case
of a breaking contact we trigger the data acquisition when the current crosses the trigger
level with a negative slope (current decreases as contact breaks). The conductance
histogram is built with all the measured curves in real time. The raw histogram can be
filtered slightly or corrected in order to avoid problems due to the Non-Ideal Differential
Linearity (NIDL) of the digital oscilloscope [30, 31]. Both approaches give similar
results. However, the peak structure is better resolved when correcting the NIDL, since
this method corrects for a systematic error and the filtering is a smoothing procedure. A
schematic drawing of the experimental setup is shown in Figure 4.
Oscilloscope
Piezo
00
..........
Sample
Figure 4. Schematic drawing of the experimental setup used to measure conductance and conductance
histograms. A metallic tip is crushed repeatedly onto a metallic surface, driving the tip piezo with a
triangular signal. The conductance is measured at the last stages of the contact breakage process. The
conductance histogram is built with all the measured curves and is displayed in real time. In RT studies
the STM operates at atmospheric conditions. The STM unit is inside a dipstick working under a low
pressure of He gas and immersed in a liquid helium bath for the low temperature studies.
4.1 ROOM TEMPERA TURE CONDUCTANCE HISTOGRAMS

The experiments presented in this section are performed in a home-built STM operating
at RT and in air. The tip is mounted in a quartered piezotube that allows movement in
the vertical (about 8 nmN) and horizontal (about 46 nroN) directions. Typically,
587
90.4 mV potential difference is applied between tip and sample, this corresponds to a
quantum current jump of7 IlA (Go = 2e% -77.5 IlS - 1112907 Q).
4.1.1. Gold Nanowires Conductance Histogram at RT
Conductance histograms for gold nanowires at RT, when the electrodes separate at
8.9 Ilm/s, are displayed in Figure 5.
--l
1x10 6
8x10 5
.....Cen
6x10 5
ico "" I
!:!
. ...
I" ...
11 j
s ...
1000
- "1ciOOO '
1~
Speed AI
:::J
0
()
4x10 5
2x10 S
0
Conductance (2e 2 /h
Als, the
potential difference 90.4 mY. Histograms built with 3000, 6000, 9000 and 12000 consecutive
conductance curves are shown, demonstrating the reproducibility of the method. The raw histograms
have been smoothed, averaging three bins to the side, and 490 n have been subtracted. Notice clear
peaks overimposed to the NIDL noise. The inset shows the dependence of the first conductance peak,
without subtracting the residual resistance, on the electrode speed.
Figure 5. Conductance histogram of gold nanowires at RT. The electrode speed is 89000
Notice that the features of the histogram do not change when the number of consecutive
samples change from 3000 to 12000. These histograms are obtained from the raw data,
averaging three bins to the side, and subtracting 490 n. This resistance arises most
probably from electron backscattering [12]; its value will then quantify the average
disorder in the nanowire. Evidence for this statement is two folded: Firstly, tight-binding
calculations demonstrate that as disorder grows, the amplitude of the quantum
conductance step decreases [12]. This decrease is larger the larger the conductance
value. Secondly, the electrode speed affects the conductance histogram. In Figure 5
inset, the position of the first conductance peak is plotted as a function of the electrode
speed. Notice that as the electrodes separate slower, the position of the first conductance
peak approaches the value Go=2e 2/h corresponding to a perfectly ordered nanowire. The
588
peak position is determined from a Gaussian fit to the data between 0.8 and 1.2 Go; the
line is drawn as a guide to the eye. Observe in the histogram that the peak height
decreases as the conductance increases, and that the peak width increases as the
conductance increases. This agrees very well with the data reported by Besenbacher et al
[30]. In addition, theoretical calculation support this [10, 21]. Notice however, that we
have subtracted 490 n from the original data. Just geometrical effects [10, 21] can
explain the observed width of the conductance peaks, but not the large shifts. Disorder
has to be invoked in order to account for the measured large conductance peak shifts.
4.1.2. Nickel Nanowires Conductance Histogram at RT
The conductance histogram for nickel nanowires at RT, when the electrodes separate at
8.9 !lm/s, is displayed in Figure 6c.
400000
300000
100000
(a)
Ni 3000 consecutive curves
II)
c 200000
::::I
1.6
c 1.4
::::I 1.2
0
0 1.0
"C 0.8
Q)
N 0.6
(\j 0.4
E 0.2
....
0 0.0
II)
(b)
Normalized NIDL
400000
300000
C
::::I
200000
100000
II)
(c)
Ni 3000 consecutive curves

corrected for the NIDL
Conductance (2e2/h)
Figure 6. Conductance histogram for Ni nanowires at RT. The raw histogram is shown in Figure 6a, the
NIDL in Figure 6b and the raw histogram corrected for the NIDL in Figure 6c. Notice that the histogram
is basically flat. A similar result is observed for Co and Fe nanowires.
589
In Figure 6a the raw data is displayed, the normalized NIDL for the oscilloscope channel
under the same conditions used in the Ni nanowire conductance experiments is shown in
Figure 6b, and, the raw data corrected for the NIDL is shown in Figure 6c. No residual
resistance has been subtracted from the histogram. As observed, the Ni histogram is
basically flat. However, the measured conductance curves show a clear stepped
behaviour (Figure 7). This is also the behaviour found for Cobalt and Iron nanowires.
56
49 -42 ___
<"
2=
--'1NI nanowire I
a
7
S
(')
0
35
::I
28
~
::I
21
14
Q.
}'"
0
0.000
Time(s)
Figure 7. Typical Ni conductance curve. Notice the stepped behaviour of the conductance.
Ni is a RT ferromagnet, and as such, the transmitivity across the nanowire should

depend on the spin direction. That is, G = e2/h ( L j=LM Tjt + L j=LN Tj,), ). In the case of a
diamagnet Tjt = Tj.J. and the 2 in Landauer's formula is recovered, i.e., the transmitivity
is independent of the spin direction. In the case of a ferromagnet that is not the case. In
our opinion the combined effect 0 disorder, geometry, and the spin dependent
transmitivity produces a flat histogram [32], even though the conductance shows clear
steps. The fact that Pt, a similar metal from the mechanical point of view, shows peaks
in the conductance histogram supports also this idea [32].
4.1.3. Copper and Silver Nanowires Conductance Histogram at RT

The conductance histograms for Cu and Ag are shown in Figure 8a and 8b.
, OCOODO
......,
Cu 12000 consecutive curves

NIDL corrected
Ag 12000 consecutive curves

NIDL correc1ed
ICODOO
!'J
"
200000
(a)
Conductance (2e21h)
(b)
Conductence (2e 21h)
Figure 8. Conductance histograms for (a) Cu and (b) Ag nanowires at RT. The histograms are corrected
for the NIDL; the electrode separation speed is 8900 Als and no residual resitance has been subtracted.
590
The histograms are corrected for the NIDL and no residual resistance has been
subtracted. Notice for both metals a first clear peak, and some structure between 2 and 3
Go. The presented histograms agree very well with the ones reported by Besenbacher et
al [30] obtained with relays. Actually, we can resolve two peaks, one at 2 Go and another
at about 2.5 Go, being this last one larger than the one located at 2 Go.
4.1.4. Sodium Nanowires Conductance Histogram at RT
(work done in collaboration with H.Olin, Chalmers Univ. of Technology, Goteborg,
Sweden)
The conductance histogram for Na is shown in Figure 9. This histogram is obtained
using a Tungsten tip and a freshly cleaved piece ofNa immersed in parafin oil.
Na 5167 consecutive curves

NIDL corrected
sooo
4000
Conductance (2e 2/h)

Figure 9. Conductance histograms for Na nanowires at RT.
The histogram is corrected for the NIDL and no residual resistance has been subtracted.
Notice clear peaks at 1,2 and 3 Go. Similarly to what is observed in Cu and Ag, the Na
conductance histogram second peak is smaller than the third one. Notice also that the Na
histogram is not as clean as those for Cu and Ag. This is most probably due to oxide.
Actually, a freshly cleaved Na sample produces a flat conductance histogram after 10-15
minutes in the oil.
Summarizing, conductance histograms taking all consecutive conductance experiments
show the quantized nature of transport through nanowires at RT. These histograms are
totally reproducible. Departures from a perfect 2e2/h quantization can be explained in
terms of the effect of disorder. On the other hand, conductance experiments in Ni, aRT
ferromagnet, show staircase quantized behaviour, but the histograms do not show clear
quantized peaks most probably due to the lifting of the spin degeneracy. Copper and
Silver show a very similar conductance histogram, with peaks at 1,2, and 2.5 Go. These
histograms are very similar to the ones obtained by F. Besenbacher group [30] using
relays. Na histograms at RT show clear conductance peaks at 1,2 and 3 Go.
591
4.2. LOW TEMPERATURE CONDUCTANCE HISTOGRAMS
(work done in collaboration with H. Olin, Chalmers Univ. of Technology, Goteborg,
Sweden)
In this section, nanowire conductance experiments performed at liquid helium
temperatures are presented and discussed. A home-built STM unit is placed inside a
dipstick, working at a low He gas pressure environment. The dipstick is evacuated and
filled with He gas several times before placing it inside the liquid helium bath. The tip is
mounted in a quartered piezotube that allows movement in the vertical (Z, up to 1.4 ~m
with 280 V) and horizontal (X-Y, up to 8 ~m with 280 V) directions.
4.2.1. Gold Nanowire Conductance
50000
(a)
II)
c:
::I
()
40000
30000
20000
10000
0
50000
(b)
II)
c:
::I
0
()
40000
30000
20000
10000
0
(c)
1.6
1.4
1.2
II)
1.0
C 0.8
::I
0 0.6
()
0.4
0.2
0.0
Conductance (2e21h)
Figure 10. Conductance histogram for gold nanowires at 4K built with 5000 consecutive curves. The
applied bias between tip and sample is 90.4 mY, the tip speed is 20 l1m/s. (a) displays raw data;
(b) shows the data corrected for the Non-Ideal Differential Linearity of the oscilloscope (NIDL);
(c) shows the normalized NIDL of the oscilloscope channel, measured under the same conditions as
the conductance experiments.
592
Figure lOa shows the raw conductance histogram data, for a series of 5000 consecutive
indentations experiments using Au electrodes at 4.2 K and with an electrode speed of
about 20 f.1m/s. Figure lOb shows the data corrected for the NIDL. The measured NIDL
signal, normalized, is shown in Figure 10c. Observe in Fig.l Ob at least 5 peaks, slightly
separated from integer values of GO' The histogram shown in Figure lOb is displayed in
a vertical logarithmic scale in Figure lla, in order to magnify the detail. Notice that up
to 7 conductance peaks are resolved. In Fig. I I b the conductance values are displayed
after subtracting a series resistance of 390 n.
(a)
NIDL corrected
C/)
C
:::J
()
10000
(b)
C/)
:::J
a
()
10000

Figure 11. The data shown in Fig. lOb is displayed in Figure II a in a vertical logarithmic scale in order
to magnify the peak structure. Notice that up to 7 peaks are resolved. Figure II b displays the result of
subtracting 390 n to the conductance. Notice that the first five peaks faIl on integer values of
GO=2e 2/h=I112907 n.
593
This has been a common procedure in all quantized conductance reports, and has been
justified by the resistance of the leads, i.e., what is not the nanowire itself. As shown
recently in a theoretical work, this resistance could arise also from disorder within the
nanowire [12]. Recent experimental results [32], including the ones presented here,
support also this view. This series resistance varies with experimental conditions, but is
usually in the range 200-1000 O. Cleaner systems usually produce lower series
resistance values.
A conductance histogram with 5 clear peaks at integer values of the quantized
conductance value is observed in gold nanowires at 4K. The height of the conductance
peaks decreases and the width of the peaks increases with increasing conductance, in
agreement with room temperature STM-experiments [20, 21, 33]. It has been argued that
the higher speed used in our STM experiments as compared to MCBJ experiments might
result in a higher nanowire temperature. However, a simple estimation, based on
macroscopic heat transport theory [31], produces a temperature rise in the nanowire of
the order of ilK, ruling out this explanation. Therefore, conductance quantization in gold
nanowires is basically the same at high, low and intermediate temperatures.
4.2.2. Bismuth Nanowire Conductance
Previous experiments in antimony have failed to observe CQ plateaus using the MCBJ
technique at low temperature [34]. Furthermore, the existence of subquantum
conductance plateaus was interpreted [34] as evidence of a conductance determined by
atomic rearrangements, but no conductance histogram was presented. While CQ can be
observed at RT for metals with Fermi energies of 6eV, for semimetals this value is tens
of meV (AF~lOnm) and CQ can be observed only at low temperatures, like in GaAs
devices [35]. For Bi the relevant values are
EF~25
meV,
AF~24nm.
In Figure 12 we display several current decays (conductance in the vertical right axis)
during the Bi tip retraction at 4K. The tip separation speed is 17 Ilm/s. The lower
horizontal axis shows the time lapsed from the trigger point and the upper one displays
this time multiplied by the electrode speed, i.e., the separation distance between
electrodes measured from the trigger point. The data presents remarkable conductance
plateaus at about 1 and 2 GO (GO=2e2/h~ 77.5IlS~ 11129070). Notice the step durations.
In the case of metals, the length of these plateaus is typically between 0.05 and 0.4 nm
(the Fermi wavelength, AF' is about 0.5nm) under the same experimental conditions
[36]. As shown in Figures 12a-c, for Bi these step durations range between 20 and 100
nm, also of the order of the electron AF in Bi, about 30 nm. The conductance plateaus
can be explained in terms of the cross section change needed in order to introduce (or
remove) electronic quantum levels in the constriction. This cross section changes should
be of the order of AF' requiring elongations of that order of magnitude.
594
Electrode Displacement (nm)
56.0
33.9
67.8
101 .7
49
(a)
42
..- ;
35
28
2;
-
16g 5
135.6
.- -.-
21
49
42
---
14
7
00
0.
58. 0
CJ
0
0.
C
"
61::>
~
33.9
35
33.9
87.B
;(
2;
28
21
0
::>
0.
6l
::>
Ii!
Electrode
101.7
(C)
35
14
CJ
Time (5)
42
21
6
5
O. 10
49
-2:
c;
2B
IOV
(b)
Time (s)
0.0
87.8
(nm)
67.8
(d)
8
5
33.9
4
3
CJ
0
::>
0.
61
Ii!"
CJ
0.
c:
6l::>
Ii!
00
0.
Time (5)
Figure 12. In Figure I2a-c selected conductance decay curves, when the Bi tip separates at 17 Ilmls with
an applied bias of90 mV are displayed. The upper horizontal axis displays the electrode separation (time
lapsed from the trigger point multiplied by electrode speed). Notice the large conductance step durations,
20-100 nm. Fig. 12d shows a magnification of the conductance plateau in Figure 12b.
Observe in the plateaus displayed in Figure 12a and 12b, that the conductance is not
perfectly horizontal but is slightly inclined, i.e., the current is continuously decreasing
before the abrupt step. This effect may be due to an increase in disorder, both in the
bulk and at the surface [12, 10], as the wire elongates, decreasing the transmitivity of the
channel. These irregularities are also reflected as current fluctuations in the plateau
conductance, shown in Figure 12d which is a magnification of Figure 12b plateau. The
frequency of the fluctuation (-20kHz) is well within the time resolution of our IV
converter (-3 J.ls), and could arise from the natural vibration of the tip producing a
lateral mechanical vibration of the nanocontact. The conductance fluctuations are also
consistent with theoretical calculations performed with configurational disorder included
in the wire [12, 10].
595
100000
90000
80000
70000
II)
60000
......
c
50000
::J
0
40000
()
30000
20000
10000
3004 consecutive curves

1736
lSi 4KI

Figure 13. Conductance histogram for Bi nanowires at 4K built with 1736 and 3004 consecutive curves.
The applied bias between tip and sample is 90.4 mY, the tip speed is 1.4 f.lrnls. The histogram is
corrected for the Non-Ideal Differential Linearity of the oscilloscope (NIDL).
The bismuth histogram (Figure 13) is corrected for the non-ideal differential linearity of
the oscilloscope [30] and no residual resistance has been subtracted. As observed, there
are two clear conductance peaks at about 1 and 2 GO' and some structure at about 3 and
4 GO' Although the statistics average out some information, these histograms are
experimental evidence of a large number of curves with plateaus at integer values of the
quantum of conductance, Go.
Summarizing, the conductance of Bi nanocontacts at 4K exhibits plateaus at quantized
values of GO=2e 2/h, remaining basically constant during electrode separations of about
50 nm. This is the first time that such plateaus have been observed in semimetals. The
histogram of conductance values, constructed with thousands of consecutive contact
breakage conductance experiments, exhibits clear peaks at 1 and 2 GO' The IV
characteristics [37] present conductance breaks at half and/or integer quantum values.
This kind of behaviour has not been observed in similar experiments with Au electrodes
[36]. The data seems to indicate that the addition or removal of electronic states to the
constriction can occur both when the size of the nanowire changes at constant bias or
when the bias change at constant size; i.e., the applied bias plays the role of a gate
voltage.
596
4.3. CONDUCTANCE PLATEAU LENGTH, FORCE JUMPS AND ATOMIC

REARRANGEMENTS
4.3.1. Statistical Study a/the Conductance Plateau Length
In the previous sections we have analyzed some properties concerning nanocontacts

formed between macroscopic electrodes, showing that these nanostructures are present
when a macroscopic contact breaks, and that they exhibit well marked conductance
features. In this section, we address a different problem: the length of the conductance
plateaus. To carry out the measurements, we have used the STM setup, with controlled
electrode approach and separation, at 89.000 A/s, at RT and in air. We stored 3200
curves, characterized by well defined quantized conductance steps (those conductance
curves with clear plateaus at G/G o=I,2,3). Since the separation speed is known, we
determine the elongation corresponding to each plateau, and analyze statistically the
distribution of such lengths. In Figure 14 we depict the histogram of elongations
corresponding to the first (a), second (b) and third (c) conductance plateaus.
300
(a) G
200
<t.s>=1.28
III
150
c..
100
<l
=1'Go
250
=3'Go
(b) G = 2Go
(c) G
<t.s>=1 .13 A
<t.s>=1 .04
50
o -+~~~~~~nTn+~~~~~~~~~~~~~~~~
o 1 234 5 0 1 234 5 0
2 345 6
L'l s(A)
Figure 14. Distribution of conductance plateau lengths for the first (a) second (b) and third (c)
conductance plateau in gold nanowires. Notice that as the conductance value decreases the average
plateau length increases. The nanocontacts are formed between a gold macroscopic wire
(1 mm diameter) and a gold plate, at RT and with a electrode separation speed of 89.000 Als. Solid lines
correspond to a Wigner distribution fit.
It is clear that the elongation distribution is very broad. The average value of the first
plateau length is 0.13 nm, close to gold interatomic distance. This average value
decreases to 0.11 nm for the second conductance quantization level and to 0.10 nm for
the third conductance quantized level. These results are at variance with previous
experiments based on STM setups [22, 23] in which atomic rearrangements during
nanowire pulling processes are detected at well defined elongation values, with narrow
lengths distribution. Furthermore, the dependence of the elongation distribution (or
elapsed times) on the conductance value seems to be consistent with conductance jumps
597
originated by the decrease of the nanocontact section under a uniform pull, instead of
conductance jumps caused by atomic rearrangements.
To show that this kind of histogram can be interpreted in terms of the separation
between electronic levels of the nanoneck, we propose a very simple model. Consider a
cylindrical neck with volume V = S x 1(S and 1 are the section and length, respectively),
and assume that the electrodes separate with constant velocity (in our experiments a
linear voltage ramp is fed the piezo). The section varies as S oc 111 assuming a constant
neck volume [17]. For the two dimensional section, the energy of the last allowed
electronic level is roughly E oc liS. Thus, in this simple model, the variation of the new
energy level with length is E oc I. In other words, by measuring conductance plateaus
(the length between two subsequent conductance jumps), we measure the distribution of
the separation in energies of nearest-neighbour quantum levels (aE). This simple model
predicts also that the higher the conductance value, the shorter the elapsed time in its
associated plateau, as experimentally observed. If the results shown in Figure 14 are
interpreted in terms of the proposed model, the distribution can be viewed as the
distribution of the energy jumps between two consecutive levels. Figure 14a would then
represent the distribution of energy differences between the first and second, (2nd and
3rd (b), 3rd and 4th (c electronic eigenstates formed within the nanowire during the
elongation process. We have fitted length histograms, like the one presented in Figure
14, to different functions in order to study them from this new point of view. We have
used a Wigner distribution (solid line in Figure 14),
P(x) = a (x/2) e-a (X/2)2
(1)
where a is a fitting parameter. This law is followed by the distribution of energy changes
in quantum chaos problems (as the Stadium [38] and Sinai's billiard [39] problems). On
the other hand, we know that for integrable systems, with more than one degree of
freedom, the distribution should be described by a Poisson-like law [40],
P(x) = be -bx
(2)
being b the corresponding fitting parameter. In order to determine the parameters a and
b associated to both distributions (Wigner and Poisson) we have fitted our data to the
cumulative probability, since this function has a smooth behaviour and does not depend
on the bin size. Clearly, the experimental distribution does not correspond to the Poisson
type, as expected for circular constriction sections [38, 40). However, the Wigner
distribution fits well the experimental histograms shown in Figure 14. This fact supports
the idea that the energy level spectrum of the nanocontacts corresponds to that of a
quantum chaotic system. Furthermore, the plateau length histogram shows a broad
distribution of lengths, in contrast with the force jumps [14] that are at a constant length.
Our data suggests then that the conductance steps are not only due to sudden changes in
cross section but also to the pinching off of states at the constriction in a continuous way
[16,21].
598
4.3.2. Force Jumps and Atomic Rearrangements
It has been suggested [34, 41], that quantized conductance is simply a consequence of
atomic rearrangements, because subquantum conductance plateaus are observed for
antimony [34] and force jumps occur simultaneously with conductance jumps in gold at
RT [41]. The Bi experiments presented above were carried out to check the hypothesis
that conductance jumps occur only when there are abrupt atomic rearrangements. The
data presented above seems to contradict this hypothesis, because it is difficult to
imagine that there are no atomic rearrangements when a Bi tip is pulled 20-100 nm.
Remember that for Au the force jumps were observed at intervals of - 0.2 nm [41].
While it is clear that the atoms in the contact have to rearrange in order to change the
contact cross section, that does not imply that this cross section changes abruptly;
specially from the electrons point of view, since they move much faster than atoms.
Actually, molecular dynamic simulations by Landman et at. [14] do not show these
abrupt force jumps at the last stages of the contact breakage. The measured force jumps
are attributed to abrupt atomic rearrangements (abrupt section changes), although this
hypothesis has not been verified. These abrupt atomic rearrangements have not been
observed yet, and, the force jumps might very well be due to the changes in conductance
themselves, i.e., the conductance jump itself produces the force jump.
Due to the fact that there is a conductance in the system formed by two reservoirs
connected by a ballistic channel of length L and width W (see work by Chen [42] for the
tunneling case) there is a resonant energy, Le., electrons are transferred back and forth
between the reservoirs. This energy could be written for L > Af as,
(3)
where E and T are the bandwidth and the transmitivity of the electrons between reservoir
1 and 2. Eq. 3 will be the metallic binding exchange energy when L - Af; however, for
larger L, Ex (L,W) - III - 1/t because this energy decreases as the traveling time t for
the electron between the two reservoirs, even if T does not change. Notice that Ex is
nothing but the matrix elements for electron transfer back and forth through the
nanocontact. On the other hand:
IT 12 = G(L,W)
(4)
where the conductance G is expressed in 2e2/h units. Notice that the value of Ex exists
even for an applied voltage V=O; this is a consequence of the conductance.
Forces on the nanocontact can be due to the variation of L, W, or both simultaneously.
For example, in 2D GaAs devices L is constant while W is changed by a gate voltage
[34]. In 3D metallic nanowires formed by breaking contacts [17,19,20,21,27,29,30,
31, 36] Land W change simultaneously; Le., as the nanowire becomes longer the section
is reduced. Three cases can be distinguished:
599
1) W=constant as L varies;
(5)
2) L=constant and W varies, case of the 2D GaAs contacts [35];

(6)
3) L and W vary; W = f(L), the case for metallic nanowires or nanocontacts [17, 19,20,
21,27,29,30,31,36];
Considering only the part of T due to conductive ballistic channels in the contact,
(neglecting the tunneling part of T because it is orders of magnitude smaller than the
conductive one) we have,
(8)
where N is the number of channels in the nanowire connecting the two reservoirs. It is
clear that the above F(L), eq.5, has a lIL2 dependence without any more structure
because G does not depend on L. The F(W) at L=constant, eq.6, will have a set of
"comblike" peaks in the force whenever the conductance has a change by a new level as
W changes (like GaAs 2D devices); i.e., a pulse of force appears when a new channel
opens or closes. However, the case of metallic nanowires, in which L is a function of W,
FQ(L) has two terms. The first one is quantized as the squared root of the conductance,
and the second one recovers the comblike structure of F(W) in eq.6. Therefore the total
force consists of a non-universal quantized part plus a set of pulses happening at the
conductance transitions.
To quantify this behaviour we simulate the conductance by a function
G=N/2 + ~i=l..N tanh l3(iLo-L),
(9)
where 13 is an inverse length parameter denoting the sharpness of the conductance step
between two consecutive channels, i and i+1, and N is the number of channels in the
contact. This function is plotted in Figure 15a and the energy, Ex(L), in Figure ISb.
Note that the energy is quantized as - G 1I2. In Figure ISc we present FQ(L), eq.7,
showing the quantized - (EIL2)G 1I2 and the comb-like -(EIL)oG 1I2/oL values as a
function of L. The values LO=O.lS nm, E = SeV, "'F = O.S nm, and, ~ = SOO nm- l ,
typical values for metallic nanowires where conductance is quantized, have been used.
The values of the quantized force are not universal and depend on L, "'F and E but there
is a clear stepped behaviour ofG.
600
A typical value for Ex is eV (Figure 15a) and for the step height in force nN
(Figure 15c) which are characteristic of the metallic binding.
6+-----,
(a)
5
4 ------------,
..r::.
.....
"to
N
....... "!""
3
2 " ...... .
0
0.0
(b)
>
.s
C1I
0.2
0.6
0.8
1.0
0.8
1.0
L-A F (nm)
22
20
18
16
14
12
10
8
6
4
2 ....
0
-2
0 _0
0.2
0.6
0.4
L-AF (nm)
:I
(c)
0.4
j' 0.12 1
I I~ . Ii
I
- - 0.2/
~;
-1
.s
:J
-2
Vi
1-
0.41
'0
LL
--- ?
, 0.23
-,---
-3
0 .0
..
0.2
.. _-- .............
---
0.4
0.6
-....
0 .8
-.-
1.0
L-A F (nm)
Figure 15. Resonant or exchange energy model for two reservoirs connected by a ballistic channel.:
(a) shows a model of the conductance versus length of the connective neck, eq. 9. (b) shows the energy
associated with this conductance variation, eq.4, and, (c) shows the force, Le., the derivative of the
energy, eq.7. Notice force steps and sharp force pulses at the position of the conductance steps.
601
There is also an important contribution to the force, the pulse (EIL)dGII2/dL that
appears at the transition region between i and i+ 1 channels and that is very large (tens of
nN). This term should be measurable in metallic nanowires, creating force pulses in the
contact that will produce instabilities. When this force pulse appears, the system will
respond; and due to the large force value during a short time, it will create mechanical
instabilities in the contact. This is the contribution that exists also in GaAs devices [35].
However, these force pulses in 2DEG systems should be about 300 times smaller than in
metallic nanowires when the length of the tube is twice the Fermi wavelength, that is,
about tenths ofnN.
Another way to visualize the forces is that for the same EF in the reservoirs and the
nanocontact, the electron density in the contact changes abruptly each time that a new
channel opens, Le., the electron filling in the contact has discontinuities. If the density
changes abruptly, the energy, being a function of this density, will change abruptly as
well. This is a natural way to understand the problem because the force is nothing but a
functional of the electron density. It should be stressed that these forces appear even if
the section varies smoothly as a function of L. For example, if the volume of the
nanowire is constant (W2 ~ IlL) then for certain values of W a new channel opens and
the energy and the force jump. In particular, these forces exist for a jellium model of the
nanowire.
There are additional forces acting on the contact; these are:
i) capacitive forces: when there is an applied bias voltage the capacitance of the system
will change abruptly every time that a new channel opens.
ii) van der Waals forces due induced dipoles [43]. These are canceled if the tip and
sample are immersed in water [43, 44]. If the experiment is performed in air this force
can produce large values that might be erroneously assigned to the force needed to break
the contact.
Summarizing, we have shown that a natural consequence of conductance quantization
through ballistic systems is energy and force quantization. The forces are large enough
to be measured and they might produce mechanical instabilities in the contact, due to
large pulsed forces due to the abrupt conductance jumps. The force pulse should be
measurable in experiments with fast time response.
Linear conductance quantization using Landauer [45] approach is easy to justify [46]
and calculations predicted conductance quantization [6]. However, a non-linear theory
for the conductance is not a settled issue. There has been predictions [47] of the
development of a new step of half quantum (e2/h) as the voltage is increased, and
experimental work by Patel et al. [48] claim agreement with Glazman. However, there
has been also criticism [49] based on numerical calculations for the conductance when
eVE f . This is quite a complicated problem and the solution depends on whether
602
adiabatic transport takes place or not, and how is the behaviour of the electron and its
density in the high field region created at the constriction. Possible scenarios are:
i) electrons accommodate in the constriction to the high fields, due to the applied bias
voltage. In this case the relevant wavelength is that of the electrons at eV.
ii) the field produces band bending for a constriction narrower than Af and redefine the
electron density at the constriction.
iii) the Fermi energy of Bi depends on subtleties in the L-point of the band structure in
the bulk [50, 51]. Perhaps when Bi is under stress and defining a narrow wire the Fermi
energy is modified and then a new gas density exists at the constriction [50].
iv) the conduction is not totally ballistic and electrons travel ballistically from one bulk
Bi reservoir to the constriction, accommodating in it, and then from the constriction gas
into the other reservoir. This phenomena also produces CQ [52].
5. Experiments in UHV: stability and switching behaviour

In order to avoid the influence of impurities on the nanocontact formation and achieve
full control on its formation, we have used a new experimental approach depicted in
Figure 16. In this configuration we have tried to get maximum control of the formation
of nanocontacts between two macroscopic wires. We have placed two macroscopic
metallic wires (having diameters larger than 100 /lm) within an ultra high vacuum
(UHV) chamber (with a typical pressure of 10- 10 Torr.). The separation between the
macroscopic is controlled by a commercial Inchworm and conventional STM
piezoelectrics, which allows to control the movement along the z direction, whereas it is
possible to control vertical and lateral displacements (from A to cm) with the combined
use of two quartered piezotubes and a step motor. This UHV system has been suitably
isolated from external mechanical vibrations. Previous to the measurements, both wire
surfaces were cleaned, heating the wires up to 600C in UHV. All the experimental data
presented in this work have been acquired at room temperature (RT).
lIoscope
Figure J6.
Experimental setup used to study the nanowire formation between macroscopic metallic
electrodes in UHV at RT.
603
5.1. HIGH STABILITY NANOCONTACTS AND SWITCHING BEHAVIOUR
As shown in the previous section, when two Au wires, initially separated, are brought
into contact after a controlled approximation between them, it is possible to form a small
contact corresponding to a relative low number of eigenstates through which the
electronic ballistic transport takes place. In this section we describe how the UHV setup
is used to study the evolution of the current during the approach and separation of two
macroscopic wires. Once the contact is established, we have observed that it remains
stable for a long time (up to hours), and that, we need large inchworm displacements
("" I 11m) to break it. In our opinion, the high stability is due to two extrinsic factors: the
excellent mechanical isolation of the UHV chamber, and, the intrinsic isolation provided
by the configuration of the macroscopic electrodes, where one electrode is a wire free to
bend (see Figure 16). This is an important factor, since we have observed that contacts
formed between rigid wires are less stable.
At constant voltage, the current flowing through the contact is continuously monitored
during all the approach and retraction processes. On approaching the wires to contact
with inchworm steps, we are able to reach a position very close (about nm) to the
formation of nanocontact. Under this situation we observe conductance switching
between 0 and I (in units of Go) as shown in Figure 17a. This switching behavior has
been observed also between 0 and 2 conductance channels (Figure 17b).
--
--
--- -
(a)
Q)
(\I
Q)
1
0
0
10
15
20
time (s)
Figure 1 7. Switching behaviour observed in gold contacts in UHV at RT. a) Transitions between 0 and I
quantum channels. b) Transitions between 0 and 2 quantum channels.
604
Notice that the time scale in which this phenomenon occur is of the order of seconds,
and, that these measurements are performed at RT. Similar switching behaviour has
been observed using a Ni tip and a Au sample in a STM configuration at 4.6-8.6 K, in
the ms time scale [28]. After this switching behaviour, a new inchworm step usually
produces the formation of a stable contact involving a large number of conduction
channels. In our experiments, the current window is limited to 140 !lA. The current
versus displacement characteristics measured in the excursions in-out of contact are
hysteretic, similarly to what has been observed in STM experiments. This hysteresis
made somehow difficult to stabilize the current inside the experimental window,
although after several approaching-separation processes using smaller inchworm steps (2
nm) and the quartered piezo tubes, we finally managed to obtain a stable current.
Concerning the approach procedure it should be mentioned that the lower the voltage
difference between the macroscopic wires, the lower the hysteresis in the contact-non
contact transitions, and the easier to stabilize the current within the experimental
window. Once stabilized, we have observed during several hours a constant current. This
corresponded to a single metallic contact involving few conductance levels! This means
we improve by more than three orders of magnitude the stability reached by standard
STM techniques [17, 19, 20, 2 I] at RT. In general, only external sources of mechanical
noise (for instance, a small Inchworm displacement (2 nm), door bangs ... ) change the
conductance once stabilized. In these cases, either the contact breaks totally, or the
conductance presents a clear jump between two quantized values, indicating a change in
the contact section. In Figure 18 we display the time evolution of the current for two
selected nanocontacts with a clear evidence of this kind of transitions due to mechanical
vibrations. These experiments are evidence of the high control achieved in the formation
and breaking processes of nan ow ires when created between macroscopic contacts.
5.8
__________
____
__
_ .____WIIl (a)
~--,
()
"c
a.
2 ~
2.9
"
00
34.6
31.7
28.8
25.9
23.0
20.2
17.3
14.4
11 .5
8.6
5.8
2.9
0.0
15
10
Time (s)
(b)
l
f
20
40
60
Time(s)
60
24
22
2o
18
16
14
12
1
8o
6
4
2
118
()
g
a.
c
';;J
Figure 18. Conductance transitions due to mechanical instabilities for gold nanocontacts in UHV at RT.
a) Transition from 3 to I quantum channels; b) Transition from 22 to 6 quantum channels.
605
5.2 INTENSITY -VOL TAGE CHARACTERISTICS

The high stability nanocontacts allows the measurement of their IV characteristic curves.
A triangular potential difference is applied to the nanocontact with a time period ranging
from 0.1 s to 50 s, recording simultaneously the current. We have performed extensive
measurements and analysis of the IV characteristic dependence on the quantum channel
number. In Figure 19 we plot a series of IV characteristic curves obtained for the same
contact at different stages (different quantized values) of the pulling process.
150
4
100
-c(
:L
50
n=1
0
-50
-100
-150
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
Voltage (V)
Figure 19. IV characteristics for different quantum channels, measured as the dynamical slope at zero
bias, denoted by n.
All these curves correspond to different breaking stages of a single nanoneck. At a first
glance, the IV characteristic curves present a well defined linear behaviour for contacts
having large conductance, whereas for contacts involving only few conductance
channels non-linear effects become evident. In all cases, the dynamical conductance (the
slope of the IV characteristic at zero bias) agrees within a few percent with the expected
quantized conductance value n x Go.
We have fitted the measured current dependence on the applied voltage to this
expression
where I(V) is the value of the current in !lA, V is the applied voltage in Volts, and, IL(V)
and INL(V) are the linear and non-linear contributions, respectively. The non-linear
component is represented in Figure 20 for some of the IV curves shown in Figure 19
(1,2,3 and 6 channels), showing that the non-linear character does not depend strongly
on the conductance channel.
606
30
20
10
1
...J
_z
0
-10
-20
-30
30
20
1
...J
_z
-0,3 -0,2 -0,1 0,0 0,1 0,2 0,3 -0,2 -0,1 0,0 0,1 0,2 0,3
Voltage (V)
Voltage (V)
Figure 20. Non-linear part of the IV characteristic of gold nanowires in UHV for a) n=l, b) n=2, c) n=4
and d) n=6 quantum channels.
In Figure 21 we plot the coefficients a(o a2 and aJ for another series of measurements.
2000
...en
...
QI
QI
...
til
til
Q.
CI
c;
1750
1500
750
tf'
500
..
,
1000
E
I;:
.t
1250
250
. ..
..
.. ... ....
.
2
t.
,
6
Conductance
'"
r
-
..
.. ..
..
.
. - .. ..
8
10
(2e2/h)
Figure 21. Coefficients corresponding to a third degree polynomial fit of several IV curves for gold
nanowires in UHV at RT. The linear coefficient, a" is represented by solid squares, the quadratic, a2 by
solid circles and the cubic, a3, by solid triangles.
607
As observed, a2a3 with a3 nearly independent on the conductance channel. We find
that a3 = 1500 150 (averaging over all the measurements). These considerations
indicate a more cubic-like behaviour of the non-linear component of the IV curve.
Notice that the IV curves do not reflect the existence of changes in the slope with a value
of ilh as reported in STM measurements using a tungsten tip and a gold substrate [53].
In order to analyze in detail the non-linear component of the IV characteristic curves, we
have assumed that INL(V) can be fitted to a potential law pV q Figure 22 shows this
exponent q for a wide set of measurements.
5,0
4,5
..
IT
4,0
c
II 3,5
c
0
~ 3,0
2,5
~
.
.. . , . .
..,
'"
2,0
0
10
Figure 22. Power law fit result of the non-linear part of several IV characteristics of gold nanowires in
UHV at RT. The inset shows the histogram ofthe exponents, revealing a cubic dependence of this nonlinear part.
Notice that the value of the exponent q spreads out over a range of values between 2 and
4; however, an statistical analysis of these data reveals a more cubic-like behaviour,
being the average exponent <q> = 2.93 0.15.
Therefore, since the non-linear component is almost independent on the nanocontact
conductance, whereas the linear part obviously increases with the number of propagating
channels, the linear behaviour quickly becomes dominant, and only for contacts with
low conductance values, the non-linear contribution becomes noticeable. The most
plausible origin of this behaviour is connected with Coulomb blockade phenomena in
small conducting systems, or the special conducting properties of Luttinger liquids [54].
608
6. Scanning electron microscopy visualization of metallic nanocontacts.
Conductance measurements provide useful insight when studying nanocontacts and, as
mentioned above, point out to the formation of connective necks between the electrodes.
A direct observation of the contact structure is then of considerable interest. The
experimental set-up consists of two high purity metallic wires twisted together in a helix
as depicted in the SEM micrograph (Figure 23).
Figure 23. SEM image showing two macroscopic silver wires twisted together in a helix. The arrow
indicates the region studied.
As time passes, one would expect that the torque will break some of the contact areas.
After a contact is produced, the surface modifications appearing at the contact region
during the electrodes separation are visualized by scanning electron microscopy (SEM).
In this series of experiments, our experiments have been performed at room temperature
and silver wires, 0.25 mm diameter, have been used.
Our SEM analysis reveals the existence of small nanocontacts formed between the
macroscopic metallic electrodes. We have observed many of these nanostructures in Ag
contacts, presenting smooth shapes. However, the fine structure of the observed
nanocontacts only allows the visualization of the nanoneck at early evolution stages on
breaking the contact. However, the obtained images point out unambiguously to the
formation of a connective micro-nanoneck between electrodes. In Figure 24a we present
one SEM micrograph of a silver microcontact with a characteristic hyperbolic shape.
609
Macroscopic silver wire
(b)
5 !-lm
Figure 24. a): SEM image showing a connective neck between two macroscopic silver wires; b): SEM
image showing a silver nanocontact ",300 nm width (see arrow).
610
Other contacts have been observed during our investigations as shown in Figure 24b.
This image shows a nanocontact with a length of 800 nm, and a minimum diameter of
about 300 nm (see arrow).
A chemical analysis has been performed on the bigger metallic bridge formed between
macroscopic wires (see Figure 24b) using an ultrathin window energy dispersive x-ray
spectrometer, detecting a large Ag signal (Figure 25). No traces of contaminants like
carbon, oxygen, etc. were found, indicating that the nanocontact material comes from
the macroscopic wire electrodes.
200
Ag
150
~
c 100
::::I
.ci
cu 50
...
Energy (keV)
Figure 25. EDX chemical analysis with an ultrathin window energy x-ray spectrometer showing a large
Ag signal. No other contaminants as oxygen, carbon, etc ... have been observed. Other signals are due to
the sample holder. The probe size used during this analysis was ",,200 nm.
Similar nanocontacts' have been observed with Au and eu electrodes [55]. This series of
experiments clearly demonstrates the existence of nanocontacts appearing when two
macroscopic metals separate. Preliminar transmission electron microscopy (TEM)
images reveal also the existence of these nanocontacts [56] with in some cases features
down to width scales of the order of 1-2 nm which certainly lead to nanowires formation
in the last stages of the contact breaking [55].
Our investigations have been performed on different metals, soft (gold, copper and
silver) and hard (tungsten and platinum). Experiments show clearly a different structural
behaviour depending on the metal hardness. The nanocontacts between macroscopic
metallic wires are easier to produce with soft metals [56], for example we did not
observe them with tungsten. The metallic "contacts" exhibit in general an hyperbolic
shape; however some weird morphologies are observed in several cases which could be
due to the breaking process [55, 56].
611
7. Light emission from breaking nanocontacts
We have studied the metallic breaking contact in the STM configuration from the point
of view of the light emission.
The experimental setup is shown in Figure 26. Light was detected by a water cooled
photomultiplier, which works in the range of 350 nm to 880 nm (3.5 eV- 1.4 eV), a
single photon counter and an oscilloscope. Tip-sample conductance was measured with
the resistance method and registered in the oscilloscope, so current decay and light
emission signals could be observed simultaneously to study their correlation in time. The
experiments were performed at room temperature with different atmospheres, air,
helium and oxygen and different metals, Au, Ag, Pt, W and AI, obtaining similar results
for all of them.
PHOroN SIGNAL
------""---
TIP
TSAMPLE
OSCILLOSCOPE
Figure 26. The photon emission signal goes from the photomultiplier (PMT) to the photon counter and
the oscilloscope. The tip current is measured by the resistor method (R=5000) and registered at the
oscilloscope. Tip and sample are of the same metal: gold, silver, platinum, tungsten or aluminium.
At first, voltages in the range of 50 to 100 mV were applied and conductance

quantization was observed, but no light emission was detected. The bias voltage was
then raised and a threshold of 1.5 volts was found over which light emission occurred.
In Figure 27, three examples of current decay for contacts of different metals with the
corresponding light emission are shown. In all of them, the applied voltage is over 1.5
volts, and light emission is observed at the end of the breaking contact process, but
always before the current goes to zero. This means that the photon emission takes place
before the contact breaks, when the wire section is very small (below 20 channels of
conductance).
612
a)
PI-PI
2 volls applied
c
CI
t:
'
il
time, sec
b)
WoW
~i!
3 volls applied
,
,
0
.,
~
..m
"
...
....
0.01
0.112
0."
0."
lime, sec
C)
Au-Au
3.3 volls applied
C~
il
,
o
.,
" .00
.....
..."'
....,
0.00
0.112
0."
lime, sec
Figure 27. Current decay and photon emission signals correlated in time for the breaking contact process
for three different metals and applied Voltages. a) Pt-Pt, 2 volts, b) WoW, 3 volts, c) Au-Au, 3.3 volts.
In Figure 28, photon counts (normalized) are shown for different applied voltages and
different atmospheres. The threshold voltage of 1.5 volts for the photon emission is
clearly seen, and the increase in photon counts with voltage is similar for the three
atmospheres.
613
Au-Au
1,0
'0
.gj
~0
.s
!'l
c:
::>
0,8
air
'"
He
- [J - oxygen
0,6
:
:
0.4
c:
:Q
0,2
i.
0-
00
0.8
1.0
1.2
1.4
1,6
1.8
2.0
2.2
2.4
2.6
applied voHage, volHs
Figure 28. Normalized photon counts versus applied voltage for contact breaking processes for Au tip
and sample in three different atmospheres: air, helium and oxygen.
In order to explain this photon emission, surface plasmons had to be ruled out because
platinum, tungsten and aluminium can not excite them. The idea of gas ionization can
not be considered either, because similar results were obtained for different atmospheres
and Weiland et al [57] have observed this light emission in vacuum. A third idea ruled
out is the electric arc; in case it happened, there would be a great quantity of material
from the tip which would evaporate making the current evolution highly irreproducible,
and that is not what is observed experimentally. The mechanism we propose to explain
this phenomena is a radiative hot electron gas in a cold lattice. In small systems the
electron-phonon coupling is weak compared with the one for bulk [58], our estimation
gives a ratio (nanowirelbulk)-10-5 . With this ratio, we assume that the electron-phonon
coupling is small in our system and that we have hot electrons and cold phonons. From
the experimental data it is possible to estimate the temperature of the electron gas [59].
For emitted photons of energies in the range of our photomultiplier, 1.4 eV to 3.5 eV,
estimated temperatures for the electron gas range from 4000K to 9980K. It is clear that
the lattice can not be at this temperatures because the contact would blow away
immediately, and then the photon emission will be simultaneous to the current
disappearance. That is not observed experimentally; notice in Figure 27 that the current
does not have any abrupt change simultaneously with the photon emission signal.
614
8. Concluding remarks
Electrical and mechanical properties of metallic nanocontacts have been studied with
microscopic and macroscopic electrodes. All the experimental setups involve the
controlled approach and separation between metallic electrodes. Conductance
histograms constructed without sample selection (up to thousands of consecutive
conductance experiments are used) show the quantized nature of transport through
nanowires at RT. These histograms are totally reproducible. Departures from a perfect
2e2/h quantization can be explained in terms of the effect of disorder. Conductance
experiments in Ni, a RT ferromagnet, show staircase quantized behaviour, but the
histograms do not show quantized peaks most probably due to the lifting of the spin
degeneracy. The low temperature histograms are basically the same than those measured
at RT. The conductance of Bi nanocontacts at 4K exhibits plateaus at quantized values
of GO=2e2/h, remaining basically constant during electrode separations of about 50 nm.
This is the first time that such plateaus have been observed in semimetals. The histogram
of conductance values, constructed with thousands of consecutive contact breakage
conductance experiments, exhibits clear peaks at 1 and 2 GO' The distribution of plateau
lengths for Au at RT, with the corresponding conductance steps and force jumps, has
been presented. It is not simple, in our opinion, to imagine a mechanical model, based
on the introduction or removal of new atomic layers, that accounts for the observed
distributions. In addition, a resonant energy model for two reservoirs connected by a
ballistic channel has been presented. This model shows that a natural consequence of
conductance quantization through a ballistic systems is energy and force quantization.
The forces are large enough to be measured and they might produce mechanical
instabilities in the contact, due to large pulsed forces due to the abrupt conductance
jumps. The force pulse should be measurable in experiments with fast time response
Controlled experiments with gold wires in vacuum conditions are presented. The
nanocontacts formed between macroscopic electrodes can be also visualized by SEM.
These nanocontacts exhibit a high stability, allowing measurements of currents flowing
through nanocontacts during hours. This high stability feature has allowed
measurements of the IV characteristic curves, showing the importance of non-linear
effects in contacts involving few conduction channels. The study of these IV curves
opens new theoretical challenges, that probably will be understood in terms of electronelectron interaction effects.
Acknowledgments
The authors thank P. Garcia-Mochales and P.A. Serena for their contribution. This work
has been supported partially by the Spanish DGICyT PB94-0 151 and CICyT MA T941456-CE projects, and the HCM, BRITE-EURAM, and ESPRIT programs from the
European Union.
615
References
I.
2.
3.
4.
5.
6.
7.
8.
9.
10.
II.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
van Wees, B.1., van Houten, H., Beenakker, C.W.1., Williamson, 1.G., Kouwenhoven, L.P., van der
Marel, D. and Foxon, C.T . (1988) Phys. Rev. Lett. 60, 848;. Wharham, D.A.,. Thornton, T.1.,
Newbury, R., Pepper, M., Ahmed, H., Frost, 1.E.F., Hasko, D.G., Peacock, D.C., Ritchie, D.A. and
Jones, G.A.C. (1988) 1. Phys. C 21, L209.
Joint European-American-Japanese Conference on Future Information Technologies, Helsinki, Finland
(1995).
Anderson, P.W. (1958) Phys. Rev. 109, 1492.
Lang, N.D. (1987) Phys. Rev. B 36,8173.
Landauer, R. (1989) 1. Phys. Condens. Matter 1, 8099.
Garcia, N. and Escapa, L. (1989) Appl. Phys. Lett. 54, 1418; Garcia, N. (1987) presentation at the STM
Workshop, International Centre/or Theoretical Physics, Trieste.
See for the 2D electron gas the review by Beenaker, C.W.1. and Houten, H.V. (1991) in Solid State
Phys. 44, edited by Ehrenreich, H. and Turnbull, D.,Academic, New York.
Kander, E., Imry, Y., and Sivan, U. (1990) Phys. Rev. B 41,12941.
Garcia, R. and Garcia, N. (1991) Surf Science 2511252, 408.
Torres, 1., Pascual, J.1. and Saenz, 1.1. (1994) Phys. Rev. B 49,16581.
Bogachek, E.N., Zagoskin, A.M. and Kulik, 1.0. (1990) Sov. 1. Low Temp. Phys. 16,796.
Garcia-Mochales, P., Serena, P.A., Garcia, N. and Costa-Krlimer, 1.L. (1996) Phys. Rev. B 53,10268.
Todorov T.N. and Briggs, G.A.D. (1994) 1. Phys. Condens. Matter 6,2559.
Landman, U., Luedtke, W.D., Burnham, N.A. and Colton, R.1. (1990) Science 248, 454; Landman U.
and Luedtke, W.D. (l991)J. Vac. Sci. Technol. 9,414.
Todorov, T.N. and Sutton, P. (1992) Phys. Rev. B 70,2138.
Bratkovsky, A.M., Sutton, A.P. and Todorov, T.N. (1995) Phys. Rev. B 52,5036.
Pascual, 1.1., Mendez, J., G6mez-Herrero, J., Bar6, A.M., Garcia, N., Landman, U., Luedtke, W.D.,
Bogacheck, E.N., Cheng, H.-P. (1995) Science 267, 1793.
Binnig, G., Rohrer, H., Gerber, Ch. and Weibel, E. (1982) Phys. Rev. Lett. 49, 57.
Pascual, 1.1., Mendez, J., G6mez-Herrero, 1., Bar6, A.M., Garcia, N. and Binh, V.T. (1993) Phys. Rev.
Lett. 71, 1852.
Olesen, L. Laegsgaard, E., Stensgaard, I., Besenbacher, F., Schiotz, 1., Stoltze, P., Jacobsen, K.W. and
Norskov, 1.K. (1994) Phys. Rev. Lett. 72,2251.
Brandbyge, M., Schiotz, 1., Sorensen, M.R., Stoltze, P., Jacobsen, K.W., Norskov, J.K., Olesen, L.,
Laegsgaard, E., Stensgaard, I. and Besenbacher, F.(I995) Phys. Rev. B 52,8499.
Agran, N., Rodrigo, J.G. and Vieira, S. (1993) Phys. Rev. B 47, 12345.
Agrart, N., Rubio, G. and Vieira, S. (1995) Phys. Rev. Lett. 74,3995.
Muller, C.1., van Ruitenbeek, 1.M. and de Jongh, L.1. (1992) Phys. Rev. Lett. 69,140.
Krans, 1.M., Muller, C.1., Yanson, I.K., Govaert, Th.C.M., Hesper, R. and van Ruitenbeek, J.M. (1993)
Phys. Rev. B 48, 14721.
J.M. Krans, 1.M., Muller, C.J., van der Post, N., Postma, F.R., Sutton, A.P., Todorov, T.N. and van
Ruitenbeek, 1.M. (1995) Phys. Rev. Lett. 74,2146; See Comment by Olesen L. et ai, (1995) Phys. Rev.
Lett. 74,2147.
Krans, 1.M., van Ruitenbeek, 1.M., Fisun, V.V., Yanson, I.K. and de longh, L.1. (1995) Nature
(London) 375,767.
Smith, D.P.E. (1995) Science 269, 371.
Costa-Krlimer, J.L., Garcia, N., Garcia-Mochales P. and Serena, P.A. (1995) Surf. Sci. 342, Ll144.
Besenbacher, F., Olesen, L., Hansen, K., Laegsgaard, E. and Stensgaard, I. (1996), to be published in
NATO ASI Series E (eds. P.A. Serena and N. Garcia, Kluwer Academic Publishers, Dordretch).
Costa-Krlimer, J.L., Garcia N. and Olin, H. "Conductance Quantization Histograms of Gold Nanowires
at 4K", submitted to Phys. Rev. B.
Costa-Krlimer 1.L. and Garcia, N. Phys. Rev. B (in press).
Costa-Krlimer 1.L. and Garcia, N. (1996) Europhysics News 27,89.
Krans 1.M. and van Ruitenbeek 1.M. (1994) Phys. Rev. B 50,17659.
Wharam, D.A., Thornton, T.1., Newbury, R., Pepper, M., Ahmed, H., Frost, J.E.F., Hasko, D.G.,
Peacock, D.C., Ritchie, D.A. and Jones, G.A.C. (1988) J. Phys. C 21, L209; van Wees, B.1., van
Houten, H., Beenakker, C.W.J., Williamson, 1.G., van der Marel, D. and Foxton, C.T. (1988) Phys.
Rev. Lett. 60,848.
616
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53..
54.
55.
56.
57.
58.
59.
Costa-Kr!imer, J.L., Garcia, N., Garcia-Mochales, P., Serena., P.A., Marques, M.1. and Correia, A.
"Conductance Quantization in Nanowires Formed Between Micro and Macroscopic Metallic
Electrodes", Phys. Rev. B (in press).
Costa-Kr!imer, J.L., Garcia N. and Olin, H. "Conductance Quantization in Bismuth Nanowires at 4K",
submitted to Phys. Rev. Lett.
McDonald S.W. and Kaufman, A.N. (1979) Phys. Rev. Lett. 42, 1189; (1988) Phys. Rev. A 37, 3067
and references therein.
Bohigas, 0., Giannoni, M.J., and Schmit, C. (1983) Phys. Rev. Lett. 52,1.
Berry, M.V. and Tabor, M. (1977) Proc. Roy. Soc. London 356, 375; Garcia-Mochales P., Serena,
P.A., Garcia N. and Costa-Kramer, J.L. "Quantum states ofnanocontacts formed between macroscopic
wires", submitted to Europhysics Letters.
Rubio, G., Agrart N. and Vieira, S. (\996) Phys. Rev. Lett. 76,2302.
Julian Chen, C. (1993) Introduction to Scanning Tunneling Microscopy, New-York, Oxford university
Press.
Garcia N. and Binh, V.T. (1992) Phys. Rev. B 46,7946; Goodman F.O. and Garcia, N. (1991) Phys.
Rev. B 43,4728.
Ohnesorge F. and Binning, G. (1993) Science 260,1451.
Landauer, R. (1989).1. Phys. Condens. Matter I, 8099.
Imry J. (1986) Directions in Condensed Matter Physics, (ed. G. Grinstein and G. Masenko, World
Scientific PUbl., Singapore), 101.
Glazman L.1. and Khaetskii, A.V. (1989) Europhys. Lett. 9(3),263.
Patel, N.K., Martin-Moreno, L., Pepper, M., Newbury, R., Frost, J.E.F., Ritchie, D.A., Jones, G.A.C.,
Jansses, J.T.M.B., Singleton J. and Perenboom, J.A.A.J. (\990).1. Phys. Condens. Matter 3,7247.
Castano E. and Kirczenow, G. (1990) Phys. Rev. B 41,3874.
Fukuyama H. and Kubo, R. (1970) .I. Phys. Soc. of Jap(Jn 28(3), 570; Fukuyama, H. private
communication.
Wolff, P.A. (1964).1. Phys. Chern. Solids 25, 1057.
Escapa L. and Garcia, N. (1990) Appl. Phys. Lett. 56,901.
Dremov, V.V. and Shapoval, S.Yu. (1995) JEPT Lett. 61,337.
Kane C.L. and Fisher, M.P.A. (1993) Phys. Rev. B 46,15233; Maslov D.L. and Stone, M. (1995) Phys.
Rev. B 52, R5539.
Correia A. and Garcia N., Phys. Rev. B (in press).
Correia A., M.1. Marques and Garcia N, submitted to J. ofVac. Sc. and Techn. B.
Weiland M. et al (1996) To be published. The suggestion that there should be light emission in this
experiments at low voltage, presumably due to plasmon relaxation, was made by J. Sass from Berlin.
Belotskii, E.D., Luk'yanets S.N and Tomchuk PM. (1992) Soviet Physics JEPT74(1), 88.
Bischoff, M. and Pagnia, H. (1975) Thin solidfilrns 29,303.
QUANTUM OPTICS
Y. YAMAMOTO
ERATO Quantum Fluctuation Project,

Edward L. Ginzton Laboratory, Stanford University,
Stanford, CA 94305, U.S.A.
and
NTT Basic Research Laboratories,
Atsugi-shi, Kanagawa, 243-01, Japan
617
618
Abstract.
Several topics of current interest in quantum optics, such as electron
optics vs photon optics, squeezed states of light, cavity quantum electrodynamics, and quantum non demolition measurements are reviewed.
Electron optics and photon opitics are essentially identical in single particle interference effects. However, when two entangled particles are involved
in the interferometry, the fermionic character of electrons and bosonic character of photons produce remarkably different results. The suppression of
partition noise due to dissipation in fermionic (electron) systems is in sharp
contrast to the enhancement of partition noise due to dissipation in bosonic
(photon) systems.
Electromagnetic waves, with the noise on one quadrature amplitude reduced to below the quantum noise of a coherent state and the noise on
the other quadrature amplitude enhanced to above it, are called squeezed
states of light. Such nonclassical states of light have been studied extensively for the last decade because of their potential applications to various
precision measurements. Generation of number-phase squeezed states by
constant current driven semiconductor lasers is emphasized in this article.
The extension of this technique to mesoscopic systems for regulated single
photon generation is also described.
Spontaneous emission of an atom is either enhanced or suppressed by
modifying a vacuum field fluctuation by a cavity wall. In a high-Q cavity,
spontaneous emission is made even a reversible process. Such a technique of
manipulating spontaneous emission is called cavity quantum electrodynamics. A new class of optical micro cavity is making possible the exploration of
cavity quantum electrodynamic phenomena in condensed matter systems.
Control of spontaneous emission in a semiconductor quantum well microcavity and its application to a low-threshold microlaser are emphasized in
this article.
Quantum non demolition measurements are the repeated measurements
of an observable without changing the free evolution of the observable. The
measurement back action is confined to the conjugate observable of the measured observable and the Heisenberg uncertainty principle is still satisfied.
This new measurement scheme can improve the measurement sensitivity of
a quantum object and has recently been demonstrated.
1. Introduction
When a classical particle is incident on a 50-50% beam splitter as shown in
Fig. lea), the particle is detected in either one oftwo detectors. The average
619
~-D-o
J
50-50% B.S.
Counts
Counts
(1,0)
(0, 1)
Probability
2"
Classical
2"
Quantum
(Boson)
(fermion)
(a)
(2,0)
(1, 1)
4"
(0,2)
2"
2"
2"
(b)
Figure 1. (a) A single particle incident on a 50-50% beam splitter and the probabilities
of counting (1, 0) and (0, 1) particles by the two detectors for classical, boson and fermion
particles. (b) Two particles simultaneously incident on a 50-50% beam splitter and the
probabilities of counting (2, 0), (1, 1) and (0, 2) particles by the two detectors for classical,
boson and fermion particles.
counts of the two detectors for N incident particles are equal to one-half of
the total particle number, (n) =
The variance of each detector count is
half the Poisson noise, (6.n 2 ) = ~(n). That is, division of incident particles
by a 50-50% beam splitter is a stochastic process and introduces partition
noise. In a general case of the beam splitter with a transmission coefficient
T, the average count and the variance of the transmitted particles are (n) =
TN and (6.n 2 ) = (1 - T)T N = (1 - T)(n), respectively. If an incident
particle is either a boson (like a photon) or a fermion (like an electron), the
quantum mechanically calculated partition noise is identical to the above
classical partition noise. (1)(2) A constant photon or electron stream acquires
a full shot noise (Poisson noise) if it encounters high attenuation.
When two particles are incident on a 50-50% beam splitter simultaneously from two input ports as shown in Fig. l(b), classical, boson and
fermion particles feature different partition properties. (3) For the classical
If.
620
particles, the probabilities of detecting (2,0), (1,1) and (0,2) particles by
the two detectors are 1/4, 1/2 and 1/4, respectively. This result reflects the
fact that two particles are divided by the beam splitter "independently" .
For the boson particles, the probabilities of detecting (2,0), (1,1) and (0,2)
particles by the two detectors are 1/2, and 1/2, respectively. The joint
probability (1,1) of each detector counting one particle is identically equal
to zero. This is because the probability amplitudes of the two particles simultaneously transmitting through the beam splitter and simultaneously
reflecting from the beam splitter, both of which result in the same result
(1,1) count, destructively interfere to nullify the total probability of (1,1)
count result. This theoretically prediction was confirmed experimentally.(4)
Such a destructive interference mathematically originates from the "symmetric" nature of the two boson particles wavefunction for exchanging their
coordinates. Physically, this is a clear demonstration of the bunching property of boson particles. For the fermion particles, the probabilities of detecting (2,0), (1,1) and (0,2) particles in the two detectors are 0, 1 and
0, respectively. The joint probability (1,1) of each detector counting one
particle is now equal to one. This is because the probability amplitudes
of the two particles simultaneously transmitting through the beam splitter
and simultaneously reflecting from the beam splitter now constructively
interfere. A completely opposite result compared to a bosonic case is due
to the "anti-symmetric" nature of the two fermion particles wavefunction
for exchanging their coordinates. Physically, this anti-bunching property is
a clear demonstration of Pauli's exclusion principle.
It is well known that shot noise is generated by ballistic division or partial tunneling of electrons in mesoscopic systems. This shot noise originates
from a stochastic single particle partition process shown in Fig. 1( a). "Electron optics" are essentially identical to "photon optics" in such a coherent
and ballistic regime. It is also known that such shot noise is partially suppressed by distributed elastic scattering and fully suppressed by distributed
inelastic scattering of electrons in mesoscopic systems. This suppression of
shot noise has the physical origin as a deterministic two fermion particles
partition process shown in Fig. l(b). "Electron optics" now departs from
"photon optics" when two electrons are involved in the scattering process
and Pauli's exclusion principle plays a role. Squeezing in photon optics is
always destroyed by dissipation, but squeezing in electron optics can be
"enhanced by dissipation".
A coherent state of light has identical noise on two quadrature components, which are cosine and sine components of the oscillating field. The
product of the two quadrature noise components satisfy a (Heisenberg)
minimum uncertainty product, (~a~)(~a~) = 116 , Thus, a coherent state
of light is a minimum uncertainty state. A laser pumped at far above the
621
oscillation threshold produces light which is not far from a coherent state.
Indeed, the limits of various optical precision measurements are all attributed to the quantum noise of a coherent state of light mentioned above.
For instance, when a laser intensity or phase is measured, so-called shot
noise appears and determines a signal-to-noise ratio. The origin of this
shot noise is the in-phase component noise (~ai) = ~ for an intensity measurement and the quadrature-phase component noise (~a~) = ~ for a phase
measurement, if only cosine component a1 carries a coherent (dc) amplitude. Such performance limits are often referred to as standard quantum
limits (SQL).
A ground state of light ("darkest state") is also a coherent state, whose
average field is zero, (ih) = ((2) = 0, but quadrature noise still satisfies
(~a~) = (~a~) = ~. This quadrature noise of a ground state is often
referred to as a quantum mechanical zero-point fluctuation and the ground
state itself is called a vacuum state.
Electromagnetic fields, with the noise on one quadrature component reduced to below the quantum mechanical zero-point fluctuation level and
the noise on the other quadrature component enhanced to above it, are
currently of great interest in quantum optics because of their potential
applications to various precision measurements. Such quadrature squeezed
states oflight are usually produced by imposing (nonlinear) unitary evolution on coherent (or vacuum) states.
A coherent state of light is also a minimum uncertainty state for photon number and phase. The photon number noise (~n2) = (n) and the
phase noise (~;p) = 4(~} of a coherent state satisfy a (Heisenberg) minimum uncertainty product, (~n2){~J2) = ~. Electromagnetic fields, with
the photon number noise reduced to below that of a coherent state and
the phase noise enhanced to above it or vice versa are called number-phase
squeezed states. Number-phase squeezed states with reduced photon number noise and enhanced phase noise are generated directly by a constant
current-driven semiconductor laser.
A quadrature squeezed state has been discussed in detail in the recent
monographs (5)-(7), so we will not reproduce it in this article. We will
concentrate here on a number-phase squeezed state.
Cavity quantum electrodynamics (QED) are also important subjects
of quantum optics which have been studied intensively for the last two
decades. Spontaneous emission of an atom is not an immutable property
of an atom but is a consequence of atom-vacuum field coupling. The quantum mechanical zero-point fluctuation of a vacuum state is constant over
a relevant frequency range for an atom and also isotropic in a propagation
direction in a free space. This is why the spontaneous emission of an atom
in a free space is an irreversible process with a unique decay time. If an
622
atom is enclosed by a cavity, the distributions of the zero-point fluctuations
in frequency and space are modulated. For instance, if the atomic transition frequency is below a cut-off frequency of the cavity, the vacuum state
which is the origin for spontaneous emission is evacuated from the cavity
and spontaneous emission is inhibited. If the atomic transition frequency is
just on resonance of the cavity, the vacuum state is resonantly trapped in
the cavity and spontaneous emission is enhanced. Moreover, when the cavity
has an enough high-Q value (long photon lifetime), spontaneous emission
can be made even a reversible process. Such cavity QED effects have been
demonstrated for a two-level atomic system and excitonic system.
A new class of optical microresonators is making possible the exploration of such quantum electrodynamic phenomena in condensed matter
systems and providing microlasers with a wide range of potential applications. The coupling efficiency of spontaneous emission into a lasing mode
and the spontaneous emission rate can be modified by various semiconductor microcavity structures. By increasing the coupling efficiency, semiconductor lasers with a very low threshold current, and semiconductor lasers
and light emitting diodes with a high quantum efficiency, broad modulation
bandwidth and low noise can be realized.
Atomic cavity QED effects have been discussed in the recent monograph
(8), so we will concentrate here on semiconductor cavity QED effects and
microlasers.
Quantum non demolition (QND) measurements are the repeated measurements of an observable without changing the free evolution of the observable. The measurement error of the observable and the back action
noise imposed on the conjugate observable satisfy the Heisenberg uncertainty relationship. The back action noise on the measured observable can
be eliminated by choosing an observable and constructing a measuring device appropriately. The concept of QND measurements was originally created for improving the sensitivity of gravitational wave detection, which
has been discussed in the recent monograph (9), so we will concentrate here
on QND measurement schemes based on quantum optics.
2. Electron optics vs. photon optics
2.1. EQUILIBRIUM AND NON-EQUILIBRIUM NOISE IN MESOSCOPIC
ELECTRON TRANSPORT AND TUNNELING
Let us consider a single-mode electron waveguide shown in Fig. 2, in which

an electron wavepacket is emitted by one reservoir electrode and propagates
toward another reservoir electrode without scattering. A conductance of
such a ballistic single electron channel is given by GQ = e2 / h, if an electron
spin degeneracy is not taken into account. The reason why the ballistic
623
0(
0(
'--------L----.f1
(:'f\/\['~n/\
AMI\AM . ~.
1
Figu.re 2. Equilibrium noise in a ballistic electron channeL The stochastic emission

of electrons near the Fermi energy due to Fermi-Dirac distribution results in thermal
equilibrium noise and the beating between high-energy empty electron wavepacket and
low-energy occupied electron wavepacket results in quantum equilibrium noise.
channel without scattering has a finite conductance is that only one electron
can be accommodated for each electron wavepacket (degree-of-freedom)
which satisfies the minimum time-energy uncertainty product, /}.Ellt = ~,
and the number of degrees-of-freedom for a given time internal Ilt is finite
for a given chemical potential difference, J-l2 - J-ll = t::.eE , where J-ll and J-l2
are the chemical potentials of the two electrodes.
If there is no chemical potential difference (zero external bias voltage)
between two reservoir electrodes as shown in Fig. 2, an average current is
zero but there is a finite noise current. The electron energy distributions
in both electrodes obey a Fermi- Dirac distribution 1 and electrons in both
electrodes near a Fermi energy are stochastically emitted according to the
partition theorem with an emission rate of 0 < 1 < 1. The variance of emitted electron number for N electron wavepackets is (/}.N 2 ) = N 1(1 - 1).
This stochastic emission of electrons due to the Fermi-Dirac distribution
at a finite temperature 0 results in thermal noise with a current spectral
density Sfh(w) = 4k B OGQ. There is another noise mechanism which exists
even at zero temperature (0 = 0); When the electron wavepackets above
and below the Fermi energy (one empty and one occupied) are absorbed by
the receiving reservoir, beating between these two wavepackets results in a
noise current at a frequency determined by the energy difference between
the two wavepackets. This beating between an occupied low-energy electron wavepacket and an empty high-energy electron wavepacket results in
quantum noise with a current spectral density Sf(w) = 21iwGQ. The total
noise current spectral density is identical to "generalized Nyquist noise"
624
spectral density:(lO)
(1)
If there is a finite chemical potential difference between two reservoir electrodes, an average current increases with a bias voltage (i) = GQ(!-ll - !-l2)
but a noise current remains constant. This is because the stochastic emission of electrons due to Fermi-Dirac distribution (thermal noise) and the
beating between high-energy empty electron wavepacket and low-energy
occupied electron wavepacket (quantum noise) occur in the two reservoir
electrodes independently, so the chemical potential difference does not affect.
Next let us consider an electron channel with a single elastic scattering
center between two reservoir electrodes as shown in Fig. 3. An elastic scattering center with a finite transmission T and reflection (1- T) coefficients,
is physically implemented into an electron channel by unintentional ionized
impurity located either inside or outside of the channel, artificially introduced channel constriction and tunnel barrier. When there is no chemical
potential difference between the two electrodes (!-l1 = !-l2), the generalized
Nyquist formula (1) still holds if the quantum unit of conductance GQ is
replaced by TGQ.(ll) As far as a system is in thermal equilibrium, the generalized Nyquist formula (1) holds irrespective of the details of a dissipative
element, i.e. independent of its material, shape and size. The scattering-free
ballistic conduction, diffusive conduction with many elastic scatterings and
dissipative conduction with many inelastic scatterings in mesoscopic systems are not an exception of this general principle.
However, when there is a chemical potential difference between the two
electrodes (!-l1 > !-l2) as shown in Fig. 3, non-equilibrium excess noise is
generated. This is because a transmitted occupied electron wavepacket
(from the left electrode) and reflected empty electron wavepacket (from
the right electrode) beat with each other, just like the quantum partition
noise in the electron beam splitter situation discussed in the previous section. Similarly, a reflected occupied electron wavepacket and transmitted
empty electron wavepacket produce independent excess beat noise. This
non-equilibrium excess noise has the similar quantum mechanical origin as
the high-frequency quantum noise 21iwGQ discussed above but is frequency
independent white noise(12).
Figure 4 shows the noise current spectral density at 1 J( Hz normalized
by the equilibrium thermal noisse 4 kB(} . TGQvs. normalized bias voltage
eV / kB(}. (12) The transmission coefficeint T of an elastic scatterer is assumed
to be 0.5. When a chemical potential differencee (V = !-l1 - !-l2) is smaller
than the thermal voltage kB(} / e that determined the region of stochastic
625
elastic scatterer
T
~~ if
Jb ~ F:rr't(. . . . . . . . . . . . .
I-------+-~fl
e~2
f2+----'1-------'.0
Figure 3.
Non-equilibrium noise in a ballistic electron channel with a single elastic
scatterer. A transmitted (reflected) occupied electron wave-packet from the left electrode
and reflected (transmitted) empty electron wave-packet from the right electrode beat
with each other to produce the quantum non-equilibrium noise.
emission of electrons due to Fermi-Dirac distribution, the non-equilibrium

excess noise of quantum mechanical origin is smaller than thermal noise
and the total noise current spectral density is dominated by thermal noise.
However, at a chemical potential difference larger than the thermal voltage, V > KBT / e, the non-equuilibrium excess noise exceeds thermal noise
and the total noise current spectral density is just one-half of full shot
noise Si(W) = e(i) = ~eGIQV, As the transmission coefficient T decreases,
the total noise current spectral density approaches to a full shot noise, i.e.
S4(W) = 2e(1 - T)TGQ V ~ 2e(i)(T ~ 1). Figure 5 shows the noise
current spectral density at a bias voltage e V / KBf) = 602v8. normalized
frequency nw/kB().(12). In a low-frequency region (nw:::; eV), the noise current spectral density is dominated by the non-equilibrium quantum partition noise with a spectral density of S4(W) = 2e(1 - T)TGQV = e(i) (half
shot noise). On the other hand, in a high-frequency region (nw 2: eV), the
noise current spectral density is dominated by the equilibrium quantum
beat noise with a spectral density of Si(W) = 2nwTGQ = nwGQ.
The shot noise of a quantum point contact fabricated in a GaAs twodimensional electron gas system has been measured in an ac lockin technique similar to Reznikov. (13) As shown in Fig. 6, the noise peaks are observed when a new transverse mode of the point contact opens.
2.2. MICROSCOPIC MODEL FOR SUPPRESSION OF ELECTRON
NUMBER PARTITION NOISE AND LOSS OF PHASE COHERENCE
The inelastic scattering process in a mesoscopic single electron channel is

microscopically modeled by a Monte-Carlo numerical simulation. An ex-
626
Beam Splitter, T
:s
=R =0.5
'N'
102
~
C'I
;:r.
en
.......
:s
N
10
C'I
"
-' Dissi~ativ~.. M1
Ul
..
1
100
1()3
102
10 1
Equilibrium
Noise Regime
Figure 4. For an electron 50-50% beam splitter, the current noise at 1 kHz (normalized by
the equilibrium noise, eq.(I) with G = (GQ )/(M +1) is plotted as a function of normalized
bias voltage. As inelastic scattering increases (increasing the number of voltage probe
reservoirs M), the excess noise is suppressed to the generalized Nyquist noise limit.
103 Beam Splitter, T =R =0.5

1
1 tiro> 2kBE>
~
~
+
::;E 102
--
'-"
e,:)
__
M=4
.!>I:
'=t
8'
~M~ ~_B~ruli= 'S~ti~C=L;Uru;'t~ .!___~_~!

J.
<Dt:Q
......
__
M=l
1
______________
----1--
CY
'-"
eVIkBE> = 602
10
,~
1_ ""
-----------~
,~
I.
..
II
1..,':
M=20
----------------------,1 .
'-"
CIj
1I
1
10-6
. >?} .1)isiplltiY~.I,;jII!it ~
1()4
10-2
-100
102
tlroIkBE>
Thennal Noise Limit
Quantum
Noise Limit
Figure 5. The normalized current noise for an electron 50-50% beam splitter is plotted
as a function of normalized frequency. For no inelastic scattering (M = 0), quantum
non-equibrium noise induces a large deviation from the generalized Nyquist noise. This
is suppressed with increasing M to the Nyquist noise limit.
627
3.0
2.5
....
2.0
$2
U
ItS
u.
0
c:
ItS
U.
V . =
- --___ - _ 1.5V
-- 3.5V
45 x
larger
..
.,,~-
~-
4.0
...... 1 V
__ 2.5V __..
3.5
- . - 4.5V
3.0
..-
.- .. -
1.5
_. . _ . _. .__._. -"--"- "...
2.0
1.5
1.0
1.0
0.5
0.07
0.5
0.0 ~C.L_~t!L---+-~~~=t::::~~~~~i- 0.0

-1.00V -0.95
-0.90
-0.85
-0.80
-0.75
-0.70
Gate Voltage (V)
Figure 6. Measured channel conductance normalized by the quantum unit of conductance (G Q ) and current noise spectral density normalized by the full shot noise value
(2eI) vs. gate voltage in a GaAs quantum point contact.
ternal bias voltage V uniquely determines the chemical potential difference

between two terminal electrodes. The corresponding electron wavenumber,
which contributes a current transport at T = 0, is divided into small segments with a wavenumber interval of !:l.k. A minimum uncertainty electron wavepacket with a wavenumber spread b.k and position spread b.x
(!:l.k!:l.x = ~) is assumed to propagate in the channel. The system is divided into many degrees of freedom, as shown in Fig. 7, which accommodate
a single electron per DOF. At T = 0, every electron wavepacket emitted from the left electrode is occupied. We put an electron beam splitter
with a transmission coefficient T to introduce a non-equilibrium partition
noise. When an electron moves from one DOF to next DOF with the same
wavenumber, a computer-generated random number q between 0 and 1 is
compared with a prescribed phonon scattering probability p. If the random
number is greater than the scattering probability and the electron final
state is empty, the electron emits an acoustic phonon with an energy of
about 0.5 meV and is back-scattered (the process A in Fig. 7). Otherwise
the electron moves to a next DOF with keeping its wavenumber, which
includes the two cases: the random number q is smaller than the scattering
probability (the process B in Fig. 7) or q is greater than p but the final
state is occupied (the process C in Fig. 7). A noise current is measured at
628
C(q > p)
I~
electrode
beam splitter
Figure 7.
tering.
r+
A(q> p)
8(q < p)
~ ...~
~x
++
I~
l..t.<n
Monte-Carlo simulation model for an electron transport with inelastic scat-
an input plane of either one terminal electrodes.

Figure 8 shows the non-equilibrium noise current normalized by full shot
noise value vs. channel conductance with distributed inelastic scattering,
obtained by such numerical simulation. In the case of zero inelastic scattering probability, with decreasing the beam splitter transmission coefficient,
the channel conductance decreases according to TGQ and the normalized
noise current increases according to Si(w)/2e(i) = (1 - T). In the case of
high inelastic scattering probability, the non-equilibrium noise is completely
suppressed. This noise suppression originates from the final-state occupancy
dependent scattering rate. In a very high inelastic scattering probability,
both forward and backward propagating electrons obey thermal equilibrium
Fermi-Dirac distribution rather than non-equilibrium distribution originally
produced by the beam splitter. In this high inelastic scattering limit, the
(macroscopic) voltage probe model and the (microscopic) Monte-Carlo simulation model give the same result. It is interesting to note that the noise
current without elastic scattering center (T = 1) increases with an inelastic scattering probability initially but decreases with further increase in an
inelastic scattering probability.
Figure 9 shows the non-equilibrium noise current normalized by full
shot noise value vs. channel conductance with distributed elastic scattering.
Irrespective of an initial noise value, the final noise current is always reduced
to one-third of full shot noise in the limit of strong elastic scattering. The
Pauli exclusion principle is effective to suppress partially the nonequilibrium
noise even if there is no energy dissipation associated with electron back
scattering.
Since the non-equilibrium partition noise introduced, by a 50-50% beam
splitter originates from Heisenberg uncertainty relation between the elec-
629
0.9
<I>
0.8
C\J
.....
0.7
II
0.6
t)
0.5
0.3
(ff
I\S
U.
0
I\S
U.
0.4
0.2
0 .1
0
=100%
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Conductance (units of 2e 2/h)
Figure 8. Normalized noise current spectral density S;(w)/2eI vs. channel conductance
in a meso scopic electron channel with a beam splitter with a transmission coefficient T
and distributed inelastic scattering.
0.9
0.8
<I>
C\J
.....
(f)-
II
0.7
0.6
0.5
~
U.
0.4
0.3
I\S
U.
0.2
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
Conductance (units of
0.7
0.8
0.9
2e 2/h)
Figure 9. Normalized noise current spectral density S;(w)/2eI vs. channel conductance
in a mesoscopic electron channel with and without a 5050% beam splitter and distributed
elastic scatterings.
tron number difference and electron phase difference between two arms, suppression of the non-equilibrium partition noise must accompany increase in
the electron phase difference noise (loss of coherence). Figure lO(a) shows an
electron wave (Mach-Zehnder) interferometer model for quantum MonteCarlo wavefunction study, in which an upper arm introduces only phase
shift but a lower arm couples to phonon reservoirs. Ten electrons are injected one by one into the interferometer and electron numbers are counted
630
(8)
(b)
"0"
0.500
0.100
0.00.
0.000
,........-~,..........-~,........~--,
....0
0.'00
+N
~:
0.100
-0.500
1.0
L......~~"...-...~.......,z!-,,~~-=
0.'00
f...
0.100
0.400
0.400
0.200
0.200
>
0.000
o;--""'"'7~.............-
......."'-"----.....]'00.000
L (ub. uni.. )
Figure 10.
Suppression of the electron wave intereference by the dissipative electron-phonon coupling in an electron-interferometer: (a) Schematic structure, (b) the interferometer output vs. phase shift of one arm for the two values of the length L of
the dissipative region: (i) L = 0.lIikr- 1 (k)/m*, and (ii) L = 0.5nkr- 1 (k)/m*, (c) the
visibility as a function of L. Each data point is obtained by averaging over 10 injected
electrons.
in the two output channels as a function of acoustic phonon emission probability. Figure 10(b) shows the oscillatory behavior of the electron wave
interferometer output (N1 - N 2 )/(N1 + N 2 ) for small and large phonon
emission probabilities. (14) Figure 10( c) shows how visibility is degraded
with increasing system-reservoir interaction length. A discrete drop in the
visibility by 1/10 is due to the emission of a single phonon in the lower
arm, which localizes one electron out of total ten electrons into the lower
arm. A continuous decrease in the visibility is due to no emission result of
a phonon in the lower arm, which localizes an electron into the upper arm.
In the example of Fig. 10(c), six electrons are found in the lower arm and
four electrons must have propagated in the upper arm.
The same mechanism of phonon emission simultaneously suppresses the
electron number partition noise through Pauli exclusion principle and increases the electron phase partition noise (loss of coherence). The complementarity between D..N and D..<J? is preserved in this way.
2.3. COULOMB BLOCKADE IN MESOSCOPIC ELECTRON TUNNELING
AND SINGLE PHOTON TURNSTILE DEVICE
In a mesoscopic pn junction with junction capacitance C < e2 /2kBfJ, driven

by a high-impedance constant current source, a single electron Coulomb
631
blockade effect regulates individual electron tunneling or thermionic emission process. Figure 11 shows the junction voltage oscillation at a frequency
f = I Ie due to regulated single electron thermionic emission with a time
separation of T = ell, where I is a drive current.(15) A similar behavior is
expected in a pn mesoscopic tunnel junction.(15) As shown in this figure,
a source resistance Rs must be large compared to a differential resistance
R = !f =
of a pn junction (constant current operation) to regulate
an individual electron emission or tunneling process. However, a parasitic
capacitance (so-called electromagnetic environment effect) usually shunts
such a small junction capacitance, which makes an ideal constant current
operation of such a mesoscopic junction very difficult.
A double-barrier mesoscopic p-i-n tunnel junction shown in Fig. 12(a)
has a unique feature of regulating a single electron and single hole injection
process even under constant voltage operation. (16) At a junction voltage
V = Va, the resonant tunneling condition for an electron sub-band state is
satisfied and a single electron tunnels through a potential barrier. When a
single electron is captured by a quantum well, the electron resonant tunneling peak is shifted to Va+eIC1 , due to a Coulomb blockade effect where C1
is an electron tunnel barrier capacitance, and subsequent electron tunneling
is inhibited. Then we switch a junction voltage to Va + ~V, at which the
resonant tunneling condition for a hole sub-band state is satisfied. Once a
single hole tunnels through a potential barrier, the hole resonant tunneling
peak is shifted to Va + ~ V + e1C2 , where C2 is a hole tunnel barrier capacitance, and subsequent hole tunneling is inhibited. In this way, a single
electron and single hole can be injected into the quantum well per a period
of junction voltage modulation. If an electron-hole pair radiative recombination lifetime is much shorter than the period of voltage modulation,
a single photon is generated per a period. Figure 12(b) shows the MonteCarlo simulation result, in which a voltage spike in each junction voltage
corresponds to a single electron or single hole tunneling event and each
cross indicates a time of photon emission.
Such a regulated single photon stream with a well-defined time clock
may find applications in high precision current standard technologies and
quantum cryptography communications.
k:l
3. Squeezing
Quantum statistical properties of laser light have been extensively studied

during the last 30 years. A laser is a nonequilibrium open system. The cavity
internal field and population inversion, which are referred to as "systems",
are coupled to equilibrium external fields and pump source, etc, which are
referred to as "reservoirs." Seperation of system dynamics from reservoir
632
..." , - - - - - - - - - - - ,
(0)
UDG
time
(liormalbed I. ell)
,-.----------,
(b)
1._
1. 7
II ...
(nor.... IiHd to ell)
(e)
.;
1.410
.....
lime
(normalized 10 ell)
"
Figure 11. The junction voltage of a pn junction driven by a high impedance constant
current source as a function of time for Rs = 300 Mn, Cs = 0, Yin = 1.6 V (e/(I) = 500
ps), and (a) T = 0.3 K, (b) T = 3 K, and (c) T = 30 K.
dynamics is usually employed in the theoretical analysis of such an open

system. Indeed, a quantum mechanical reservoir theory successfully applied
to a laser (17L(19). Coherent states of light, which feature quantum mechanical zero-point fluctuations around its average field excitation, has also been
extensively studied in quantum optics(20).
633
(a)
...... ----- ~
PAlo.A.
(b)
_=-f---'~ Iv,:v...v
1-+-==1
(c)
1.11515
1.1115
1.110
1.1150
1.1115
1.1'5
1.1'0
1.110
1.175 '--_~"""""_~....L...~~"""'~~.........J 1.175

o
1
2
3
4
lime (normalized to TIC )
Figure 12. The energy-band diagram of the p - ip - i-in - n AIGaAs-GaAs heterojunction, with applied voltage (a) Vi = Va and (b) Vi = Vo+~V. For
= Va(~ = Vo+~V),
the Fermi energy of the N - (P-) type AlGaAs layer is at least e /2Cn i(e /2Cp i) higher
(lower) than the energy of the quantum-well electron (hole) subband. (c) The junction
voltage as a function of time shown together with the accompanying photon emission
events. The voltage spikes correspond to electron and hole tunneling events.
'i
Recent theoretical and experimental studies on quantum noise properties of a semiconductor laser have revealed that the semiconductor pn junc-
634
tion laser produces photon-number squeezed states with photon-number

noise reduced to below the standard quantum limit (SQL) (~n2) = (n)(21).
3.1. BRIEF REVIEW OF SQUEEZED STATES
The Heisenberg uncertainty principle for two conjugate observables 0 1 and

02 is the direct consequence of the commutation relation between them. It
can be derived by the Schwartz inequality
(2)
where ~Oi = Oi - (Oi), i = 1,2. A state that exactly satisfies the equality
in Eq. (2) is called a minimum-uncertainty state and is mathematically
defined as an eigenstate of the operator
(3)
where r is a squeezing parameter. That is, r determines the distribution of
quantum noise between 0 1 and O2:
(4)
A quadrature squeezed state of the electromagnetic field is a minimumuncertainty state, in which 0 1 and O2 correspond to the two quadrature
phase components ih = 1/2(a + at) and a2 = (1/2i)(a - at). 0 3 is then a
c number (= 1/2). The squeezed state features reduced quantum noise in
one quadrature and enhanced quantum noise in the other quadrature. In
the special case of r = 0, the two quadrature components share the same
amount of quantum noise, namely, (~an = (~a~) = 1/4. This is a characteristic of a coherent state. The electromagnetic field generated by a classical oscillating current is a coherent state(20). Its quadrature-component
quantum noise is sometimes referred to as the standard quantum limit
(SQL). In order to ultimately reduce the noise in one of the quadratures to
zero, the electromagnetic-field mode must have an infinite photon-number
and thus an infinite energy. This is because the enhanced quadrature noise
component will require part of the total photon-number
(5)
635
That is, there is a trade-off relation between quantum-noise reduction and

the required photon-number. A number-phase squeezed state of the electromagnetic field is defined as
(6)
(7)
and
(8)
where n, S, and C are the photon-number operator, the sine operator, and
the cosine operator, respectively. When r is equal to -(1/2) In(2( n)) and the
average photon number is much larger than one, the photon number noise
and the sine-operator noise (which corresponds to the classical phase noise)
are reduced to ((~n)2) = (n) and ((~S)2) = [((C)2)/(4(n))] ~ 1/(4(n)).
This photon number and sine-operator noise is also referred to as the SQL.
A number-phase squeezed state with r greater than -(1/2) In(2(n)) features
smaller photon-number noise and larger sine-operator (phase) noise than
the SQL, while the minimum-uncertainty relationship ((~n)2)((~S)2) =
(1/4)(C)2 is preserved. The photon-number noise can ultimately be reduced
to zero (photon-number state) without requiring an infinite photon-number,
because the enhanced phase noise does not consume an energy at all.
An unmistakable mark of squeezed state generation is the observation
of quadrature amplitude squeezing. That is, a homodyne-detector output
current will feature a reduced noise level below that for vacuum fields.
An unmistakable mark of number-phase squeezed state generation, on the
other hand, is the observation of amplitude squeezing. That is, a photoncounting detector output will feature a noise below the shot-noise level and
thus sub-Poissonian photon statistics.
3.2. PRINCIPLE OF SQUEEZED STATE GENERATION IN
SEMICONDUCTOR LASERS
The macroscopic coherence of light and matter in a laser is established by

the balance between ordering force of "systems" (gain saturation in the
case of a laser) and fluctuating forces from external "reservoirs." The conventional quantum mechanical reservoir theory for a laser treats only three
degrees of freedom as "systems": cavity internal field, population inversion,
and dipole moment. All of the other degrees of freedom are considered as
"reservoirs" and eliminated by use of the quantum mechanical fluctuationdissipation theorem.
636
A fluctuating force pertinent to our discussion is that from an external pump source. The pump fluctuating force for optically pumped lasers
is usually shot-noise-limited, because the pump light absorption by a gain
medium is considered as a random Poisson point process. Of course, this is
not a fundamental limit imposed by quantum mechanics. We may consider
a thought experiment in which a gain medium is excited by photon-number
states of light with a perfect absorption efficiency. The pump fluctuating
force determined by the photon-number noise of pump light is reduced to
zero for such a case (22). The phase noise of pump light does not contribute
to the pump noise of a laser at all, because a laser oscillator is an incoherently pumped device in contrast to a parametric oscillator, which is a
coherently pumped device (similarities).
The photon-number noise spectra for a pump-noise-suppressed laser are
reduced to below SQL at frequencies below the cavity cut-off frequency at
pump rates well above the threshold. The physical interpretation for the
result is rather straightforward. Every absorbed pump photon is emitted
sooner or later as a coherent lasing photon from the cavity, in which the
stimulated emission by a strong laser field is much stronger than a random spontaneous emission at high pump rates. Therefore, if the emitted
photon-number is counted for a measurement time interval longer than the
population lifetime and the photon lifetime, the photon statistics of the
laser output should be the same as the photon statistics of the pump light.
A high conversion efficiency is indeed only one criterion for such preserva-
tion of the absorbed pump photon statistics to the emitted coherent photon
statistics.
However, the phase noise spectra are increased to above SQL at frequencies below the cavity cut-off frequency because of the so-called SchawlowTownes phase diffusion process. (17)-(19) The product of reduced photonnumber spectrum SAN(n) and increased phase spectrum SA~(n) satisfies
the Heisenberg uncertainty principle,
(9)
In the limit of high pump rates, the product is higher by only a factor of 2
over the minimum uncertainty product. A finite photon lifetime introduces
the residual photon-number noise SAN(n) by a stochastic photon output
process and, at the same time, it stores coherent photons inside the cavity
which leads to the stabilization of the phase noise SA~(n) by a stimulated
emission. This result suggests that the output field from a pump-noisesuppressed laser is not far from the number-phase squeezed state.
The junction current noise of a pn diode driven by a constant voltage
source is shot-noise-limited. This fact has been erroneously identified as a
"shot-noise-limited pump fluctuating force" for a semiconductor laser. The
637
!Ill ttl ttt 1., IllII11IlIIlIl

I
tnrr-rrr-ri~--'Y+H'
I
....
CR
++.
-:--- ...... ~C?~s .. .
(a)
.. t
... t
(b)
Figure 13. The junction-voltage fluctuation and the relaxation-current pulse caused by
a photon emission event for (a) R ~ Rs (constant-voltage driving case) and (b) R ~ Rs
(constant-current driving case). C is the diffusion capacitance of a pn junction. The
time constant C R corresponds to the electron lifetime, and CR. is the external circuit
relaxation time constant.
mistake in this argument is twofold. First, the junction current noise is not
a pump fluctuating force but a relaxation current pulse for the radiative
recombination (photon emission) event in a junction. Second, a semiconductor laser usually operates under a constant current mode rather than a
constant voltage mode.
The difference between constant-voltage-mode and constant-currentmode is schematically shown in Fig.13. When a pn junction is driven by a
constant voltage source with a source resistance Rs much smaller than the
diode's differential resistance R, that is, Rs ~ R =
where kB is the
Boltzmann constant and T is temperature, the junction-voltage drop resulting from the radiative recombination of an electron-hole pair is quickly
eliminated by an external circuit relaxation current pulse. By the time of
next radiative recombination event occurs, the junction-voltage (number of
electron-hole pairs in a junction) is recovered to the original value. In such a
memoryless system, photon emission events follow a Poisson-point-process
and the junction current carries full-shot noise. On the other hand, when
a pn junction is driven by a constant current source with a source resistance much larger than the diode's differential resistance, that is, Rs ~ R,
the junction-voltage is not pinned by a source but is free to fluctuate. The
k:l,
638
fluctuation in the junction-voltage caused by the radiative recombination

is not recovered at the time of next radiative recombination event. In such
a system with memory, the photon emission events follow a sub-Poissonian
process and the junction current carries noise below the shot noise level
because of self-stabilization.
The pump fluctuating force of a semiconductor laser is identified as
the thermal noise generated in the source resistance Rs whose spectral
density is 4~~T. The thermal noise is much higher than the shot noise
value 2e1 for a constant voltage source with Rs <t: R, but is much smaller
than the shot noise value for a constant voltage source with Rs ~ R.
The thermal-noise-limited pump fluctuating force and the dissipation rate
of the junction-voltage through the pump circuit satisfy the fluctuationdissipation theorem exactly. In the case of a constant current source, the
dissipation rate (C Rs)-l of the junction-voltage through a pump circuit is
very slow and the fluctuating force from the pump circuit 4kBT / Rs is also
small. The two conditions necessary for the generation of photon-number
squeezed states in semiconductor lasers, that is, the free voltage fluctuation
and the pump noise suppression, are satisfied simultaneously by Rs ~ R.
3.3. SQUEEZED STATE GENERATION EXPERIMENTS
Figure 14A shows a dual-balanced detector setup for measuring the photon
number noise spectrum from a semiconductor laser. A semiconductor laser
output is split equally into the two arms by a half-wave plate and polarization beam splitter. One detector output current It is delayed by a coaxial
cable and combined with the other un delayed detector output current h
by a differential amplifier or 180 0 hybrid. For a fluctuation frequency nin
satisfying nin T = 2N 7r, where T is the delay time and N is an integer, the
differential amplifier output, It - 12 , calibrates the shot noise value. For a
fluctuation frequency n out satisfying noutT = (2N + 1)7r, the differential
amplifier output It + h measures the photon number noise of the laser
output. Thus, the differential amplifier output simultaneously displays the
laser noise and corresponding shot noise values with a frequency period of
~n = ~ on a spectrum analyzer.
The measured noise spectra at two different pump rates for a GaAs
transverse junction stripe semiconductor laser are shown in Fig. 14B. The
noise spectra are normalized by the shot noise value, which is independently
calibrated by a light-emitting diode with the same wavelength as the laser.
The noise spectrum at the pump rate just above the threshold, 1/ 1th = 1.03,
shows a lower noise value at nin and a higher noise value at nout, where
1th is the threshold current. This is a characteristic of anti squeezed (superPoissonian) light. The noise spectrum at the pump rate well above the
639
HWP
SlgnalL
wave1:r""
J);j--" . f\T' . .
Shot noise level
~
n
2n:)
Antlsqueezed light
O~
2ft
:)01:
Squeezed light
(a)
4
iii' 2
~
1
J
!
-1
-2
0
100
200
300
Frequency (MHz)
400
500
(b)
Figure 14. (A) A dual-balanced-detector setup for measuring the intensity noise of a
semiconductor laser: HWP, half-wave plate; PBS, polarization beam splitter; and ATT,
attenuator and equalizer. (B) The measured intensity noise spectra normalized by the
shot noise value for two pump rates, I / Ith = 1.03 (upper curve) and I / Ith = 13.6 (lower
curve). The solid line represents the standard quantum limit (SQL).
threshold, 1/ Ith = 13.6, shows a higher noise value at Din and a lower noise
value at nout, which is a characteristic of squeezed (sub-Poissonian) light.
The degree of squeezing is degraded by the Poissonian partition noise
640
1100.0
10.0
11.O.f---=..=..--------'~--------t
0.1
0.02+-_-.-_ _ _ _ _ _ _-.-_ _ _ _ _ _ _40.5
10
Normallzlacl pumping IeVII (f)
100
Figure 15. The theoretical (lines) and experimental (.) intensity noise values normalized
by the shot noise value versus the normalized pump rate R == J / J lh -1. The experimental
results are corrected for a detection quantum efficiency of 89%. The three curves correspond to different interval absorption losses (the dotted line represents an ideal case of
no absorption).
associated with optical loss. To increase the light collection efficiency, a

GaAs transverse junction stripe laser, with less than % front facet reflectivity Rl and more than 90% rear facet reflectivity R 2 , is operated at 66
K to minimize the free carrier absorption loss and is directly coupled to a
Si photodetector to minimize the coupling loss. Figure 15 shows theoretical
and experimental photon number noise values normalized by the shot noise
value versus the normalized pump rate 1/ Ith - 1. The experimental results
are corrected for a detection quantum efficiency of about 89%, so the degree
of squeezing shown in Fig.15 corresponds to that of the laser output.
4. Cavity Quantum Electrodynamics

Spontaneous emission is not a fixed property of an atom but a consequence
of atom-vacuum field coupling. It has been known for some time that the
decay rate, transition energy, and radiation pattern of spontaneous emission can be altered by modifying vacuum field fluctuations by a cavity wall.
The principle, often referred to as cavity quantum electrodynamics (cavity
QED), is a classic theoretical problem. Cavity enhanced and suppressed
spontaneous decay rates were predicted by Purcell in 1946 (23). A cavity
induced radiative energy shift was pointed out by Casimir in 1948 (24). The
coupling of atom and vacuum fields was formulated by Jaynes and Cummings in 1963 (25), in which it was predicted that spontaneous emission
becomes even reversible if an atom is put in a high-Q single mode cav-
641
ity. However, those effects have been observed only recently using either
Rydberg atoms in microwave superconductor cavities or atoms in optical
micro-cavities. The only exception is the pioneering work by Drexhage in
1974(26), in which he demonstrated that the decay rate and radiation pattern of spontaneous emission is modified by a metal mirror.
Since spontaneous emission is a major source of energy loss, speed limitation and noise in semiconductor optical devices, the possibility of employing cavity QED to control spontaneous emission is expected to improve
the device performances fundamentally.
4.1. BRIEF REVIEW OF CAVITY QUANTUM ELECTRODYNAMICS
There are two distinct regimes in cavity QED, i.e., low-Q regime and highQ regime. A decisive parameter separating the two regimes is the vacuum
Rabi frequency,
f! -
R-
PI2 [
'Ii
vac-
PI,
'Ii
Iiw
2EoVo
'
(10)
where P12 is an atomic dipole moment, Evac is a vacuum field amplitude

and Vo is an optical cavity volume. When nR is much smaller than the
escape rate of a photon from the cavity and the decay rate of an atomic
dipole moment, i.e., nR ~ 'photon, ,dipole, the atom-vacuum field coupling
is only weakly perturbed by a cavity and spontaneous emission remains
an irreversible and incoherent process. Nevertheless, a cavity can modify
the radiation pattern and decay rate. Spontaneous emission rate is either
reduced or enhanced.
On the other hand, when OR ~ 'photon, ,dipole, the atom-vacuum field
coupling is strongly perturbed by a cavity and spontaneous emission becomes a reversible and coherent process. A coupled atom-field system, often
referred as a "dressed state" has two-split eigen-frequencies. This vacuum
Rabi splitting corresponds to a periodic energy exchange between the atom
and field.
When N atoms collectively couple with the cavity field, the atom-field
coupling strength is enhanced in proportion to VN. In a low-Q regime,
the cavity enhances a so-called cooperative superradiance decay (27). The
decay rate is enhanced and the emission pattern is highly concentrated
due to the formation of a gigantic dipole moment. In a high-Q regime, it
corresponds to a collective Rabi splitting(8). The difference between the two
eigen-frequencies is increased to nRVN from nR for a single atom case.
Enhanced and reduced spontaneous emission rates at microwave frequencies have been demonstrated with either Rydberg atoms(28)(29) or cyclotron electrons(30). Enhanced and inhibited spontaneous emission at optical frequencies have also been demonstrated with either dye molecules(31)
642
or Rb atoms(32). Collective Rabi splitting has been demonstrated both at

microwave (33) and optical frequencies(34)(35). Cavity QED experiments in
semiconductors require a micro-cavity structure with an optical wavelength
size. This is because the spontaneous emission spectral width in semiconductors is much broader than that of an atom in vacuum due to a very short
dephasing time and a very large inhomogeneous broadening of an electronhole pair dipole. The cavity structures used for cavity-QED experiments in
semiconductors are quite similar to those utilized in vertical cavity surface
emitting lasers. In fact, a major thrust in the study of cavity-QED effects
in semiconductors is the development of small-sized, ultra-low threshold,
high speed diode lasers and light emitting diodes for applications in optical
communication, parallel optical signal processing and display.
Because spontaneous emission is a major source of energy loss, speed
limitation and noise in lasers, the capability to control spontaneous emission
is expected to improve laser performance. New phenomena are expected
if the radiation pattern is highly directed toward a lasing mode and the
fraction of spontaneous emission coupled into the lasing mode is made
close to 1. The fraction of the total spontaneous emission that is coupled
into the single lasing mode at low excitation rates is defined to be the
spontaneous emission coefficient (3. An example of the new phenomena
expected as (3 approaches 1 is the "thresholdless laser, (36) in which the light
output increases nearly linearly with pump power instead of exhibiting a
sharp turn-on at a pump threshold.
Consider a single atom centered in an optical cavity between two confocal mirrors, each with reflectivity R. This cavity radiates into a mode that
subtends a solid angle ~n. If the atom is located at an antinode of the
cavity field, the spontaneous emission rate into ~n is enhanced by a factor
4/(1- R) because the vacuum field at that point is strongly enhanced relative to its free-space value. Correspondingly, when the atom is located at a
node of the ~avity field, the spontaneous emission rate into ~n is reduced
by a factor of (1 - R)/4. If atoms are distributed over many wavelength,
strong spatial modulation of the spontaneous decay rates is not expected.
Note that although the spontaneous emission rate into the cavity mode can
be strongly modified, the total spontaneous emission rate integrated over
all angles may change only a modest amount if ~n ~ 411".
The most important effect of modified spontaneous emission for possible applications is the altered radiation patterns and the enhanced spontaneous emission coefficient (3. The strong reflectivity dependence of the
spontaneous emission rate into a cavity mode clearly indicates that (3 can
be increased when an atom is located at the antinode of the cavity field. As
the reflectivity approaches unity the (3 value approaches the ideal limit of 1.
Such high (3 values can be used to enhance the efficiency of light-emitting
643
diodes and lower the thresholds of lasers.

4.2. MICROCAVITY LASER PHYSICS
The spontaneous emission coefficient 13 for the relatively large-volume lasers

in common use today can be approximated by
A4
13 = -47r--:2'-V-~-A-n-3
(11)
Here ~A is the emission linewidth, n the refractive index and V the mode
volume. This expression accounts for the number of modes in the system,
which is roughly equal to the number of cubic wavelengths in the system,
and the number of modes within the emission linewidth, which is proportional to ~A/ A. For example, even a semiconductor laser the size of a grain
of salt has a 13 value of only 10- 5 . One can reduce both V and ~A to obtain a larger 13 value. In principle one can increase the 13 value to close to 1.
For this purpose, one does not necessarily require a true single-mode cavity.
One needs only to enhance the spontaneous emission rate into a single mode
relative to the many existing modes over the spectral emission bandwidth,
which is typically less than a few percent of the transition frequency.
One of the most important consequences of the increased spontaneous
emission coefficient is the reduction of the threshold pump rate of a laser.
In a simple atomic model, where the spontaneous emission rate and the
emission spectral width are constant as a function of excitation level, the
threshold pump rate as a function of 13 is given by(37)
(12)
Here /photon, is the decay rate of the cavity photon. The quantity ~ =
N f3vr /2/p hoton is the average photon number in the lasing mode when
half of the atoms are in the excited state; N is the atomic density and r
is the spontaneous emission rate. The excitation level where the excitedatom density N ex is (N /2) is called the transparency carrier density in
semiconductors, because there is no absorption and no net gain at this
point. The gain G at excitation levels higher than the transparency point
is assumed here to be a linear function of the excited-atom density N ex :
G = f3Vr(Nex - N /2).
The threshold pump rate in Eq. (12) includes the two distinct regions
that are being studied in present semiconductor microresonator experiments, ~ <{:: 1 and ~ ~ 1. When ~ <{:: 1 and 13 <{:: 1, Pth is equal to IPhoton/2f3
and the threshold pump rate decreases linearly with 1/13. Figure 16 shows
the light output power as a function of pump rate for a set of 13 values
644
10 8
i'
10 6
Tsp=
10
>
i Jp
r- lOS
II>
.J:l
= 10 12 S 1
10 7
Tn,.
10 2
"
10
10 4
N o = 10 IR em 1
C
C
10 3
V = 10
=
0
-0
10 2
-15 em
A.=JPm
.c 10
Q.,
II>
--==
~
10- 1 0
Co
10- 2
Co
10-3
10-4
10- 1
!r-
IO-S
10 -2 L...:::.4..r....L...--'-'-.........,"--.....L.--'-...4---'-_-'-_-':-_~---I
10-6 IO- S 10-4 10 3 10- 2 10. 1
I
10
10 2
10 3 10"
Injection current (rnA)
Figure 16_ Photon flux and power. The curves show the average flux in the mode
(left axis) and the corresponding output power (right axis) for micro cavity lasers with
various spontaneous emission coefficients (3. Parameters used for the theoretical model
that generated these curves are /'photon = 10 12 /sec, r = 10 9 /sec, N/V = 10 18 /cm 3 ,
V = 10- 15 cm 3 and>" = 1 micron.
in this regime. The lasing threshold, characterized by the jump in the efficiency of converting pump power to light output power, decreases with
increasing (3. When ~ ~ 1 and (3 ~ 1, Pth is independent of (3 and is fixed
at the pump rate required to obtain transparency. One must keep the total
number of atoms NV and the spontaneous decay rate r small to satisfy
the condition ~ ~ 1 and obtain a decrease in the threshold pump rate with
an increase in (3. Of course, we are assuming that the active material can
be pumped to emit a sufficient number of photons to achieve threshold at
these low gain volumes. In practice the linear gain dependence assumed in
this analysis cannot be achieved if the gain volume is too small. Another
consequence of experimental reality is that (3 can vary with excitation level
if the spectral emission width is not constant.
In Figure 16 we see that when the spontaneous emission coefficient (3 is
1 there is no jump in the quantum efficiency of the conversion of pump photons to light output. This is the "thresholdless" laser, or "zero-threshold"
laser(36). However, there is a well-defined threshold, Pth = Iphoton, even in
this case, according to Eq. (12). At pump rates below the threshold, the
average photon number in the cavity is less than 1, so that the dominant
emission process is incoherent spontaneous emission. At pump rates above
the threshold, the average photon number becomes greater than 1, resulting
in a dominance of coherent stimulated emission. With the onset of stimulated emission, the population inversion is "pinned" at its threshold value,
because each newly generated electron-hole pair very rapidly recombines in
645
a stimulated process. The dominance of stimulated emission above threshold is also expected to result in an increased coherence of the light output
and a narrowing of the output spectrum even in the j3 = 1 limit, as shown
in Fig.17(a).
Fluctuations of the light intensity emitted by a microlaser exhibit a
resonance increase at threshold, as shown in Fig.17(b). These light fluctuations, similar to the fluctuations found in a second-order phase transition,
are a measure of threshold behavior that also persists in the limit of j3 = 1.
Note that as j3 approaches 1, these fluctuations decrease and their relative
width in pump power increases, thus providing another indicator of the
unique characteristics in the high-j3 microlaser limit.
We conclude from an examination of carrier density pinning, linewidth
narrowing and light-level fluctuations-present even when there is only approximately one photon in the cavity-that there is a well-defined threshold
even for j3 = 1. This threshold discriminates between regions that can be
described as a linear amplifier and a nonlinear laser oscillator.
4.3. FABRY-PEROT MICROCAVITIES
The simplest approach to fabricating an optical microcavity is to shrink

the spacing between the mirrors of a Fabry-Perot resonator to )..In while
reducing the lateral dimensions to the same range. This structure provides
a single dominant longitudinal field mode that radiates into a narrow range
of angles around the cavity axis.
Semiconductor microcavities provide high-Q Fabry-Perot cavities for
both basic studies and potential applications. The Q of a cavity is the number of radian cycles required for the optical field energy in the cavity to
decay by a factor of lie. Molecular-beam epitaxy or organometallic chemical vapor deposition techniques are used to deposit high-reflectivity mirrors
consisting of alternating quarter-wavelength layers of lattice-matched semiconductors. For example, 15 to 20 pairs of quarter-wave layers of AlAs and
Alo.2Gao.8As result in a reflectivity greater than 99% and Q values greater
than 500. The optically active layer in such a microcavity is typically a
GaAs quantum well located at the midplane of the cavity, where the field
strength is a maximum.
Planar Fabry-Perot microcavities have been studied theoretically and
experimentally in some detail. The two important parameters for laser performance, Iphoton and j3, depend on the cavity geometry and the spectral
emission width of the quantum well gain material.(38) Even if the dimensions of the cavity in the plane between the mirrors are much larger than
an optical wavelength, the planar cavity structure will give rise to discrete
spatial modes with a finite mode radius Tm. The mirror spacing and reflec-
646
10 6
Jl= 10-'
10'
=
!
10 1
10 2
'N
..c:
..
10'
c:
:3
r = 10 12 S
t."
-,
= 10 -9 S
fir> t"
10
= /013 em .J
=/0 .H em J
NO
V
10-'
10-6
10'
10"'
10-3
10- 2
10-'
10 2
10

(a)
..
6
'"
100
0.9
10"'< 1\ < 0.1
0.99
c:
~
.;;
c:
..
j
...-=crc:
10
;..,
..
....
Noninversion
.:
0.110 -6
10-'
10-4
10-3
10'
10-'
Inversion
10
10 2

(b)
Figure 17. Linewidth and noise. The laser linewidth (a) and intensity-noise spectral
density (b) in the light output from a microlaser as a function of the pump current are
plotted here for a series of f3 values. The noise shown here is normalized with respect to
the mean number of photons in the cavity. The parameters used in the theoretical model
that generated these curves are the same as in Fig. 15.
tivities determine the radiating solid angle of the mode. This solid angle is
inversely proportional to the coherence radius and directly proportional to
the cold-cavity linewidth ~Am-the linewidth ofthe cavity resonance without the gain material. For example, a GaAs micro cavity with a spacing of
A/n between the mirrors and with mirror reflectivities of 99% corresponds
to a mode radius Tm ~ 7 microns, a cavity resonance linewidth ~Am ~ 0.43
nm and a far-field angular radiation angle of 5.7 0
The maximum f3 achievable with a large-area planar GaAs Fabry-Perot
microcavity is estimated to be 0.05, assuming dipole moments oriented in
the plane of the cavity and negligible spectral broadening. However, the
647
Absorbed Power fJ.JW)
0.1
10
100
1000
100
Q.
Q.
10
".~
iii
E
0.1
o 0.01
0.001
II
0.1
10
100
1000
Incident Power (mW)
Figure 18.
Power transfer. The normalized output power from a GaAs Fabry-Perot
microcavity is plotted here as a function of incident pump power (bottom axis) and
absorbed pump power (top axis). The solid line is the theoretical curve for (3 = 0.0l.
(The fraction of the total spontaneous emission that is coupled into the single lasing
mode at low excitation rates is defined to be the spontaneous emission coefficeint (3.)
GaAs quantum well emission linewidth is 30 nm at room temperature and

2.5 nm at 4 K, much broader than the cold-cavity linewidth of 0.43 nm. As
soon as the emission linewidth exceeds the cold-cavity bandwidth, f3 starts
to decrease with the ratio of the two widths; f3 is thus estimated to be 0.05
x (0.43/2.5) rv 0.01 at 4 K. The quantum efficiency for conversion of pump
photons to output light is close to unity above the threshold, while it is
equal to (3 below it, assuming negligible absorption in the cavity. Therefore
f3 can be experimentally deduced from the step at threshold, yielding a f3
of 10- 2 , as shown in Fig.18.
The f3 values achievable with the large-area planar microcavities are
limited by the leakage of spontaneous emission into directions centered
around the plane of the cavity. A post-like structure, fabricated by wet or
dry etching, can suppress this leakage of spontaneous emission and increase
the f3 values up to 0.5 if the post diameter is on the order of the wavelength
in the material and if the quantum well emission linewidth is much narrower
than the cold-cavity linewidth. The factor-of-2 degradation in f3 stems from
the fact that the two degenerate orthogonal polarization modes capture an
equal amount of spontaneous emission below threshold, but only one of
them oscillates above threshold. Such a high f3 has not yet been observed
experimentally, in part because the emission linewidths are often broader
than the cold-cavity linewidths.(39)
An alternative way to increase f3 is to use a hemispherical structure with
an integrated curved mirror.(40) To fabricate each of these curved mirrors,
648
researchers formed a photoresist disk on a planar GaAs AIGaAs optical

cavity. They then melted this photoresist by heat treatment, forming it
into a hemispherical disk. Finally they transferred the curved geometry of
the photoresist disk to the semiconductor by dry etching using an ion beam.
They controlled the curvature and diameter of the structures by varying
the diameter and the thickness of the resist disk; the resulting curvatures
ranged from 5 to 20 microns. The hemispherical micro cavity has a measured
(3 values of 0.05, which compares with the theoretical prediction of 0.1.
5. Quantum Nondemolition Measurement
The quantum theory of measurement has been widely regarded as a problematic and dubious subject of little relevance to real physics. The reason
for this irrelevance of the quantum theory of measurement is, however,
only because experimentalists could not make repetitive measurements at
a quantum limit on a single quantum object until very recently. In the
1980s, the experimentalists' interest has shifted from the single destructive
measurement of an ensemble of systems to the repeated nondestructive
measurements of a single system. An experimentalist in quantum optics
finally began to catch up with the measurement concepts such as "collapse
of the wavefunction" of the creators of quantum mechanics.
The standard quantum theory based on unitary time evolution cannot
describe the process of a measurement. A measurement is a nonunitary
and indeterministic process, for which a so-called "projection postulate" is
applied. This is a very reasonable postulate but there has been no direct
experimental evidence for the postulate so far. A quantum nondemolition
(QND) measurement is the first experiment to directly confirm such a ununitary collapse of the wavefunction by a measurement process.
5.1. GENERAL QUANTUM MEASUREMENT AND QND
In a general quantum measurement the observable of the signal system,

As,is measured by means of the change in the observable of the probe system, Ap , after the proper interaction between the signal and probe systems,
as shown in Fig.19. The Heisenberg equations of motion for As and Ap are
written as
- in dtAs =
dA
and
- in dtAp =
A
[Hs, As]
[Hp,Ap]
+ [HJ, As]
(13)
+ [HJ,Ap]
(14)
where fIs is the unperturbed Hamiltonian of the signal system, fIp is that
of the probe system , and fI J is the interaction Hamiltonian between the
649
measurement
dissipation
... -
bOCkOCtKiiiI
'. 'bi,Ck-cit-iOii-iiSignal system
fluctuation
Probe system
external
OOFs
( macrOSCOPiC)
meter
Figure 19. QND measurement model with a probe system and a macroscopic meter.
Signal observable As is nondemolitionally measured through demolitional measurement of
probe observable Ap , whose motion is affected by As by means of interaction Hamiltonian
iiI. If iII does not commute with A., the motion of As is kicked by iII (back action I).
If iIs contains Bs, the motion of As, is also kicked by the uncertainty principle between
As and Bs (back action II).
two. The first commutators in Eqs.(13) and (14) represent the free motion
of As and Ap. The second commutators represent the contribution of the
interaction between the signal and probe systems.
In order to measure As using Ap, [HI, Ap] in Eq. ( 14) should not be
zero and HI should be a function of As. These are the requirements for a
general quantum measurement:
1. ih is a function of As.
2. [HI, Ap]
-# 0
In general, a measurement of As affects the motion of As itself in two

ways (measurement back action). One way is the change of As due to
[HI, As] in Eq. (13). Such an example is a photoelectron emission in a photodetector. When a signal photon is detected by a conventional photo detector, the photon absorption process naturally destroys the signal photon.
Such a back action is shown by the dashed line in Fig.19. If the interaction
Hamiltonian HI commutes with As,
(15)
HI is called a "back action evading" type, and this first source of back
action can be eliminated.
The other way of changing As is through the uncertainty that is introduced on the conjugate observable of As by the measurement. The Heisen-
650
berg uncertainty principle dictates that when the observable As is measured

with accuracy (~A;), the uncertainty imposed on the conjugate observable
Bs should be larger than
(16)
If the unperturbed Hamiltonian Hs contains the conjugate observable B s,
the motion of As is affected by this uncertainty of Bs. An example is the
position measurement of a free particle. When a free particle's position is
measured by a microscope, the measurement accuracy is proportional to the
photon wavelength, ~q ~ A. A photon scattered by a free particle changes
the particle's momentum. This back action imposed on the particle's momentum is of the order of the photon's momentum itself, ~p ~ hi A. The
kicked momentum then perturbs the free motion of the particle's position
according to the following relation:
H =
s
L
2M
-+
'(t) = '(0) + p(O)t
2M
(17)
Here q(O) and p(O) are the position and momentum just after the first
measurement. This back action is shown by the dotted line in Fig.19.
If As(O) commutes with As(t),
(18)
As is called a "QND observable" and the second source of back action

can be eliminated. Equations (15) and (18) are the conditions for a QND
measurement (41).
In a QND measurement the observable lis is not affected at all, even
though the measurement accuracy of As is increased arbitrarily. In optical
communication this means that the information carried by a signal wave is
fully extracted at one receiver but a signal wave still can transport exactly
the same information to the other receivers.
5.2. QND MEASUREMENT OF PHOTON NUMBER
When the observable of the signal wave is a photon number As

its,
condition (18) is satisfied, since the system-free Hamiltonian is given by
Hs = nws (its + and [Hs, its] = 0; that is, a photon number is a "QND
observable" .
Let us consider the configuration shown in Fig. 20. The signal wave
at frequency Ws propagates along an optical Kerr medium with a probe
!)
651
bock-action
Kerr medium /
Signal wove Ws
ns
! s
measurement
homodyne
receiver
Figure 20.
Schematical view of a QND measurement scheme with an optical Kerr
medium. The signal photon number is nondemolitionally measured through the homodyne detection of the probe phase, which carries the signal photon-number information
by means of the optical Kerr effect.
wave at frequency wp in which the phase of the probe wave is modulated

by the photon number of the signal wave. The phase modulation of the
probe wave is detected by a homodyne receiver, and thus the information
about the signal photon number can be extracted. If the signal wave is not
attenuated, the signal photon number is preserved.
The readout observable of the probe wave is a quadrature component
= (o'p - O,t)/2i. The interaction Hamiltonian for the measurement is that for the inter-phase modulation between the signal and probe
waves:(42)
Ap : : : : o'p2
(19)
The constant X is equal to (li/2V ,2)wpwsX(3) , where V is the mode volume
and X(3) is the third order susceptibility; ' is the dielectric constant of the
medium.
It is easily verified that the interaction Hamiltonian (19) satisfies the
general quantum measurement conditions 1 and 2 in the previous section.
The QND condition (eq. (15 is also satisfied, and thus the interaction
Hamiltonian (19) is a "back action evading" type.
652
5.2.1. Heisenberg Picture
The unitary evolution for the signal and probe wave annihilation operators
resulting from the onset of Eq.(19) is expressed by
(20)
and
(21)
Here VF = XL/e,L is the length of the Kerr medium, and e is the light
velocity. The optical homo dyne receiver with a proper local oscillator phase
(90 degrees out of phase for the coherent excitation of the probe wave)
measures the quadrature phase component of ap(L),
;i
ap2(L)
[ap(L) - at(L)]
apl(O) sin(VFns)
ap2(0)cos(VFns)
(22)
Since the probe wave has the coherent excitation only along ap l(O), that is,
(apl(O)) =J 0 and (ap2(0)) = 0, and the phase modulation VFns is a small
quantity, eq. (22) is reduced to
(23)
The signal photon number to be measured is expressed by

(24)
Taking the expectation value and variance of eq. (24), we obtain
(n~obs)) = (ns)
(( ~n(obs))2)
s
(~n2)
s
(25)
+ (~ap2(0)2)
F{np)
(26)
The second term of eq. (26), (~n~2) = ((~ap2(0)2)/F(np), represents the

measurement error, which is determined by the probe wave phase noise
(~"j;;) ~ (~ap2(0)2)/(np) and the nonlinear interaction strength
It
can be decreased arbitrarily by increasing VF and/or decreasing (~~;).
This QND measurement is ideal, since the expectation value and variance,
that is, the statistics of measurement results, are equal to those of the signal
wave itself. When the probe wave is in a coherent state, (~~;) is given by
n.
653
1j4(np). When the probe wave is in an optimum squeezed state, (b..,(j;;) can
be reduced to 1j(4(np)((np) + 1)).
The phase noise of the signal wave is increased by the other interaction
represented by Eq. (20). The quadrature phase component of the signal
wave is similarly calculated as
(27)
The phase noise is thus given by
(b..o/.2) = (b..a s2(L)2)

'f's - (a s1(0))2
= (b.../. 2(0)) + F(b..n 2)

'f's
(28)
Here (b..,(j;s(0)2) = (b..ap2(0)2)j(ap1(0))2 is the initial phase noise of the

signal wave. The second term, (b..,(j;:2) = F(b..n~), is the back action noise
added by the measurement. By decreasing the measurement error, the back
action noise increases. When the probe wave satisfies the minimum uncertainty product, (b..ap1(0)2)(b..ap2(0)2) = 116 , the measurement error and the
back action noise satisfy the Heisenberg minimum uncertainty product:
(29)
The preceding argument concerning the measurement error and back
action noise suggests that the signal wave after QND measured is "reduced"
to a number-phase squeezed state. This can be easily shown by using the
projection postulate.
5.2.2. Schrodinger Picture
Suppose the signal and prolJ_e waves are initially in coherent states la)s and
Imp. The density operator after the interphase modulation is expressed by
the unitary operator, (; = exp(innsnp),
(30)
The density operator after the measurement of the quadrature amplitude
of the probe wave is given by (for the specific readout fh)
(31)
The variances for photon number and sine operators are (43)
(32)
654
and
Trs [,o~meas.Ih)(S - (S) )2]
(t1S2)
r.J
41!12 + ~ [1- exp (-2FII112)]
(33)
satisfy the number-phase minimum uncertainty product, (t1n 2) X (t1S2) =
1(6)2.
The initial coherent state is transformed to a number-phase squeezed
state by the nonunitary state reduction. QND measurement is considered
as the generation process of a nonclassical light.
This QND measurement was experimentally demonstrated with an optical fiber soliton(44).
6. Conclusion
Squeezing, cavity QED and QND measurement are the experimental
demonstration of fundamental concepts of quantum mechanics such as
Heisenberg's uncertainty relation, vacuum field fluctuation and von Neumann's projection postulate. There are also emerging subjects such as quantum cryptography and quantum computing (parallelism). Theoretical and
experimental studies on various quantum optics has contributed greatly the
foundation of quantum mechanics. The field has now started to link with
mesoscopic physics, nanostructure devices and communication/information
processing technologies.
Any information processing scheme based on quantum coherence or superposition state does not seem realistic as a practical technology because
it requires a perfect preparation of the system and negligible dissipation.
There is no such a perfect system in a real world. The computer and communication systems of today are classical one, even though the principle
of constituent devices such as a laser or a transistor has a quantum mechanical origin in one way or the other. They do not utilize the quantum
mechanical nature of a linear superposition state.
The future evolution of quantum optics with nanostructure devices and
atomic scale devices is expected to give us an answer to the long standing
question "Is a real quantum effect useful for practical applications?"
655
References
[1] Caves, C. M. (1980) Phys. Rev. Lett. 45, 75.
[2] Yurke, B. (1986) Phys. Rev. Lett. 56, 1515.
[3] Loudon, R. (1989) in J. H. Eberly, L. Mandel, E. Wolf (eds.), Coherence and Quantum
Optics VI, Plenum Press, New York, 703.
[4] Hong, C. K., Ou, Z. Y., Mandel, L. (1987) Phys. Rev. Lett. 59, 2044.
[5] Meyster, P., Sargent, M. (1991), Elements of Quantum Optics (Springer-Verlag, Berlin,
Heidelberg).
[6] Walls, D.F., Milburn, G.J.(1994), Quantum Optics (Springer-Verlag, Berlin, Heidelberg).
[7] Vogel, W., Welsch, D. G. (1994) Lectures on Quantum Optics (Akademie Verlag).
[8] Berman, P. R. (1994), Quantum electrodynamics(Academic Press, Boston).
[9] V.B. Braginsky, F. Y. Khalili Quantum Measurements(Cambridge Univ. Press, Cambridge, 1992).
[10] Callen, H. B., Welton, T. A. (1951) Phys. Rev. 83, 34.
[11] Landauer, R. (1989) Physica D 38, 226.
[12] Liu, R., Yamamoto, Y. (1994) Phys. Rev. B 50, 1741l.
[13] Reznikov, M. et al. (1995) Phys. Rev. Lett. 75, 3340.
[14] Imam6glu, A., Yamamoto, Y. (1994) Phys. Lett. A 191, 425
[15] Imam6glu, A., Yamamoto, Y., Solomon, P. (1992) Phys. Rev. B 46,9555; Imam6glu,
A., Yamamoto, Y. (1992) Phys. Rev. B 46, 15982.
[16] Imam6glu, A., Yamamoto, Y. (1994) Phys. Rev. Lett. 72, 210.
[17] Haken, H. (1970), Light and Matter, vol. 25 of Handbuck der Physik (Springer-Verlag,
Berlin).
[18] Sargent, M. III, Scully, M.O., Lamb, W. E., Jr. (1974), Laser Physics (AddisonWesley, Reading, MA)
[19] Louisell, W. H. (1973), Quaantum Statistical Properties of Radiation(Wiley, New
York).
[20] Glauber, R. J. (1963) Phys. Rev. 131 2766-2785
[21] Yamamoto, Y., Machida, S., Nilsson, O. (1991), in: Yamamoto, Y., (Ed.), Coherence,
Amplification and Quantum Effects in Semiconductor Lasers (Wiley: New York) 461537.
[22] Golubev, Y. M., Sokolov, I. V. (1984), Sov. Phys. JETP 60, 234-245.
[23] Purcell, E. M. (1946), Phys. Rev. 69, 681-682.
[24] Casimir, H., Polder, D. (1948), Phys. Rev. 73, 360-374.
[25] Jaynes, E. T., Cummings, F. W. (1963), Proc. IEEE 51,89-95.
[26] Drexhage, K. H. (1974), in: Wolf, E., (Ed.), Progress in Optics, vol. 12, (North
Holland: New York) 165-184.
[27] Dicke, R.H. (1954), Phys. Rev., 93, 99-12l.
[28] Hulet, R. G., Hilfer, E. S., Kleppner, D. (1985) Phys. Rev. Lett.55, 2137-2140
[29] Jhe, W., Anderson, A., Hinds, E.A., Meschede, D., Moi, L., Haroche, S. (1987) Phys.
Rev. Lett. 58 666-669.
[30] Gabrielse, G., Dehmelt, H. (1985) Phys. Rev. Lett. 55, 67-70.
[31] DeMartini, F., Innocenti, G., Jacobovitz, G. R., Mataloni, P. (1987) Phys. Rev. Lett.
59 2955-2958.
[32] Heinzen, D., Childs, J. J., Thomas, J. E., Feld, M. S. (1987), Phys. Rev. Lett. 58,
1320-1323.
[33] Brune, M. Raymond, J. M., Haroche, S., to be publihsed.
[34] Raizen, M. G., Thomason, R. J., Brecha, R. J., Kimble, H. J., Carmichael, H. J.
(1989) Phys. Rev. Lett. 63, 240-243.
[35] Zhu, Y., Gauth9ier, D. J., Morin, S. E., Wu, Q., Carmichael, H. J., Mossberg, T. W.
(1990), Phys. Rev. Lett. 64, 2499-2502.
[36] DeMartini, F., Jacobovitz, G. R. (1988), Phys. Rev. Lett. 60,1711-1714.
[37] Bjork, G., Yamamoto, Y. (1991) IEEE J. Quantum Electron 27,2386-2396.
656
[38] Yamamoto, Y. Bjork, G.,Karlsson, A., Heitmann H., Matinaga, F. M. (1993) Int. J.
Mod. Phys. B 7, 1653-1695
[39] Baba, T., Hamano, T., Koyama, F., Iga, K. (1991) IEEE J. Quantum Electron 27,
1347-1358.
[40] Matinaga, F. M., Karlsson, A., Machida, S., Yamamoto. Y., Suzuki, T., Kadota, Y.,
Ikeda, M. (1993) Appl., Phys. Lett. 62, 443-446
[41] Caves, C. M., Thorne, K. S., Drever, R. W. P., Sandberg, V. D., Zimmermann, M.
(1980), Rev. Mod. Phys. 52, 341-392.
[42] Imoto, N., Haus, H. A., Yamamoto, Y. (1985), Phys. Rev. A32, 2287-2292.
[43] Kitagawa, M., Imoto, N., Yamamoto, Y. (1987), Phys. Rev. A 35, 5270-5273.
[44] Friberg, S. R., Machida, S., Yamamoto, Y. (1992), Phys. Rev. Lett. 69, 3165-3168.
TOPICS IN QUANTUM COMPUTERS
D. P. DIVINCENZO
IBM Research Division

T. J. Watson Research Center
Yorktown Heights, NY 10598 USA
Abstract. I provide an introduction to quantum computers, describing how

they might be realized using language accessible to a solid state physicist.
A listing of the minimal requirements for creating a quantum computer
is given. I also discuss several recent developments in the area of quantum
error correction, a subject of importance not only to quantum computation,
but also to some aspects of the foundations of quantum theory.
1. What is a quantum computer?

I don't think that I will spend many words here saying why there has been
a considerable growth of interest in the last couple of years in the subject
of quantum computation. There has been a spate of reviews[l, 2, 3], semipopular articles[4]' and press accounts[5] giving, on the whole, a very good
overview of the subject. At some level, the recent interest simply arises from
the very traditional movement of computation into ever more miniature
worlds, and what could be more miniature than the world of the single
quantum? At another level, though, interest has arisen because the rules of
quantum dynamics changes the rules of computation itself[6], in ways which
we are still working to understand. You probably can't factor large numbers
with any computer following, at the level of the logical operations, the
laws of classical mechanics (which is to say, every computer ever operated
up until now); with a computer, or a computation, obeying the laws of
quantum dynamics, you just might be able to factor[7, 3]. This has drawn
the attention of both the practical, problem-solving world, as well as that
of those interested in further exploring and understanding the foundations
of quantum theory itself.
657
658
So, since we are all solid state physicists here at Curacao, and all know
a thing or two about quantum physics, let me immediately give a fairly sophisticated run-down of the minimal requirements for any quantum system
to be a quantum computer. As you will see, the entry fee is pretty steep,
which provides at least one good reason why all sorts of people aren't already putting together their quantum processors. Anyway, here it goes:
1.1. FIVE REQUIREMENTS FOR QUANTUM COMPUTING
1) The degrees of freedom required to hold data and perform computation should be available as dimensions of the Hilbert space of a quantum
system. This quantum system should be more-or-Iess isolated from its environment. (More about this shortly.) Also, this Hilbert space should be
precisely enumerable, for example, I should be able to say, "the quantum
system consists of 49 spin-3/2 states on the molecule," or something of this
sort. It won't do to be able to make only statistical statements about the
number of degrees of freedom, as is often the case in solid state physics
(e.g., "the quantum dot contains 100 5 electrons") . No, the Hilbert space
must be precisely delineated.
Now, it is furthermore very desirable for the Hilbert space to be decomposable into a direct product form. This rather formidable-sounding
requirement is actually quite natural for a multi-particle quantum system.
In the example mentioned a moment ago, the Hilbert space decomposes
into a product of 49 parts, each of dimension 2(3/2 + 1) = 5. Note that
this means that the dimension of the Hilbert space of the entire quantum
system is 549 , a rather large number. When the Hilbert space of the individual particle is two-dimensional (e.g., for a spin-1/2 particle), we term
this object a quantum bit or qubit, as its role in quantum computation is
analogous to that of an individual bit in an ordinary computer. Now, it
is not absolutely necessary that the Hilbert space have this direct-product
form; the Hilbert spaces of collections of indistinguishable particles do not
have this form because of (anti-)symmetry constraints. But it is essential
that the size of the Hilbert space grow exponentially fast with the size of
the system. For this, a multi-particle system is necessary; the Hilbert-space
dimension of a one-particle system grows only algebraically with system
size.
2) Another requirement of quantum computation is that it must be
possible to place the quantum system in a fiducial starting quantum state.
This state can be very simple, as in "all spins down" for a collection of
spins; thus, this requirement will be satisfied if it is possible to cool the
quantum system to its ground state. This may be trivial - if the qubits
are embodied by a ground and excited state of an atom, room temperature
659
may be quite cool enough; this cooling requirement may also be very high
tech ~ cooling atoms in a trap to their motional ground state, or cooling
nuclear spins to their ground states, may require nano-Kelvins. Of course,
the experimentalists have not been remiss in achieving these sorts of conditions lately. I do not think that this "initial state preparation" requirement
will be the most difficult one to achieve for quantum computation.
3) Here is a pretty tough requirement. The quantum system to be used
as a quantum computer must be to a high degree isolated from coupling to
its environment. This isolation requirement is linked up with the precision
required in quantum computation: if the state of the computer at some
instant is ideally supposed to be the state W, then the actual state after
one clock cycle, p (a density matrix), should differ from W by only a small
amount:
(1)
(wlplw) ~ 1 - E.
One way (not the only way) that the state of the system can depart from W
is by evolving into a state in which the joint quantum state of the computer
and that of its environment become correlated through interaction, a state
of affairs described as entanglement by Schrodinger in 1935[8, 9]. When the
joint state is entangled, the state of the system alone must be described as
a mixed state or a density matrix; in mesoscopic physics, we would say that
the quantum system has begun to travel down the road to decoherence or
phase-breaking.
By the way, the issue of how big an E is tolerable in quantum computation is probably the active question among theorists in this field today,
as it is all tied up in the question of error correction and fault tolerance
in quantum computation. There is good news and bad news on this front:
error correction is possible in quantum computation, so finite E is perfectly
tolerable. The bad news is, at least at the moment, it is not known whether
a very big E is tolerable. The protocols which are currently understood[1O]
start to get into trouble when E = 10- 6 , give or take a few orders of magnitude. But there may be schemes in which E = 0.1 will be perfectly all right;
we just don't know, and I certainly hope that we manage to figure it out.
4) The next requirement is at the heart of quantum computation: It
must be possible to subject the quantum system to a controlled sequence
of unitary transformations. All of our quantum algorithms are expressed
in terms of such sequences. It is also required that these unitary transformations can be made to act upon specified pairs of qubits, or other small
collections of qubits. This fits hand-in-glove with the requirement 1) that
the Hilbert space be made of a direct product of the spaces of individual
small systems or particles. Note that the necessary effect of the two-bit
quantum transformations (which are referred to as quantum logic gates) is
that they produce entanglement between qubits of the quantum computer.
660
Notice that entanglement between different parts of the quantum computer
is good; entanglement between the quantum computer and its environment
is bad, since it corresponds to decoherence.
There are various ways in which such controlled unitary transformations
may be achieved. In most of the ways which people are thinking about
now, it is attained by subjecting the system to a particular time-dependent
Hamiltonian over a fixed length of time; the resulting unitary operator is
a function of the Hamiltonian according to the usual time-ordered product
expression:
U = Texp(i
H(t)dt)
(2)
(I won't get into the time-ordering mumbo-jumbo here.) This H(t) may
be imposed via a time-varying magnetic field (as in NMR), a strong (i.e.
classical) time-varying optical field (as in laser spectroscopy), or by the
physical motion of a massive (again, therefore, classical) object (viz., the
tip in an atomic-force microscope). Such H(t)'s are in fact- ubiquitous in
experimental physics, although making these operations selective at the
single-quantum level is not.
Again, there is what appears at the moment to be a pretty stringent
precision requirement on these unitary operations. In the natural metric
in the space of unitary operations, the distance between the specified and
the actual unitary transformation should be less than E, and the present
constraint on E is about the same as in the discussion in item 3); only E'S
less than about 10- 5 can presently be considered "safe".
5) Last, but not least, it is necessary that it be possible to subject the
quantum system to a "strong" form of measurement. When I say "strong" I
simply refer to the kind of quantum measurement that we learned about in
our textbooks: the measurement determines which orthogonal eigenstate of
some particular Hermitian operator the quantum state belongs to, while at
the same time projecting the wavefunction of the system irreversibly into
the corresponding eigenfunction. The standard example of such a measurement is the Stern-Gerlach experiment in which the z-component of a
spin-l/2 particle is projected into one of its two eigenvalues. While this is
straight out of the textbooks, it is unfortunately unlike what a lot of actual
quantum measurements in the laboratory actually consist of, being more of
a "weak" variety. I will not give a complete discussion of what a weak measurement is, I could point the reader to the textbook of Peres[ll] for this
discussion. Basically, the idea is that the individual quantum system, say a
single spin-l/2 system, might interact very weakly with the measurement
apparatus, such that the probability that the apparatus registers "spin up"
is only very weakly correlated with the actual wavefunction amplitude for
the spin to be up. To be more quantitative, there exist weak measurements
661
on a state W = al t)
+ bl -!-)
which register "spin up" with probability
Pup
1)
= 2" + 0 (2
lal - 2"
(3)
for arbitrarily small o. These small-o measurements are "weak" in the sense
that after such a measurement the quantum state of the system has been
disturbed hardly at all; on the other hand, hardly any information has been
gained by the measurement about the state of the spin. In many areas of
experimental physics a weak measurement is all that you can do, simply
because it is not presently known how to make the coupling to individual
quantum systems sufficiently strong; in NMR and in most solid-state spin
systems, the experiments are insufficiently sensitive to detect the state of
individual spins, although there is a continuing push in that direction. I
would say that electrical measurements of mesoscopic quantum structures
are also weak, in the sense that the course of each individual electron passing
through the device is typically not determined. (This is probably not the
case for certain measurements in single electronics.)
In many cases weak measurements are very satisfactory for learning a
great deal about the quantum properties of systems, because they can often
be done on macroscopically large ensembles, involving either many replicas
of the same quantum system (very typical in NMR) , or many identical runs
of the same quantum measurement (as in normal electron transport). By
averaging over such ensembles, a very good knowledge of a in Eq. (3) can
be obtained, no matter how small 0 is. However, these weak measurements
do not satisfy the requirements in quantum computation, at least so far as
we have presently formulated them.
2. Brief survey of experimental systems

I believe that with this long-winded set of five desiderata, a reasonable evaluation of any proposed quantum computer implementation can be made.
Let me illustrate this with two brief examples, both of a nano-solid state
character, which show that the very act of performing this evaluation points
to a lot of physics which we would like to know about these systems, about
which we are presently pretty ignorant.
2.1. ATOMIC FORCE MICROSCOPE.
This is a gedanken apparatus, shown in cartoon form in Fig. 1, which I

proposed some time ago[12, 2] for doing quantum computation. As will be
seen, there are in fact severe problems with using it as a quantum computer;
but indeed, that is what we are supposed to use the magic five criteria to
reveal!
662
6}- 6)- 6}- 6}- 6}- 6}- 6}

IIIIIII
6}-6)-6}- ~c-?'S -6)_6}_

I
I
I
I
I
~
"0
r-JY -6}-~
r-JY (9 s.
,s. -6)-6)-6)-~
s.
I
s.
"0
1(10
1.0
1600
Figure 1. My cartoon of at atomic force microscope concept for a quantum computer,

described in the text. See also [2].
1) The idea of precisely controlling the computational Hilbert space is

tied up in the precise atomic design implied by the cartoon for the tip
and the surface of the instrument. Finding a spin within a solid is not a
problem; indeed there are spins and other quantum states galore. The idea
is to set up the system so that only the ones you want are really "available".
A qubit may be made unavailable if the energy to change its state is much
larger than the energy actually available in the experiment; in solid state
language, the idea is to "open a gap" for all the excitations that you don't
want to happen. In the cartoon, these means that we envision an insulating
material such as intrinsic x-Si at low temperatures, so that there are no
low-lying excitations available. Furthermore, the nuclei are all chosen so
that no nuclear spin excitations are available (here the "unavailability" is
particularly clear, since the energy to create a nuclear excitation would be
somewhere in the MeV range). So there are no low-lying states, except for
the proton spin of the H atoms terminating the tip and various sites on the
surface. Thus, this is envisioned as a proton-NMR quantum computer. It
places some requirements on atomic scale atomic engineering (the avoidance
of any defects within or on the surface of the crystal, the avoidance of
surface states, precise isotopic control) which are at the moment pie-inthe-sky, although conceivable as an extension of the considerable actual
progress with atomic placement and manipulation in recent years.
2) The preparation of such a nuclear spin system into a pure state would
require some cooling techniques considerably beyond the current state of
the art. Since the energy scales for nuclear spins are so small, the temperature required to have a substantial ground-state population is below 1O- 3 K
or so, for spins in a IT field. Of course such temperatures can be attained
in many types of experiments, but I know of no effort to achieve them in
663
an AFM. There are techniques which are specifically adapted to bringing

nuclear spins to low temperatures[13], involving coupling by spin resonance
techniques to other spins (typically electron spins). These approaches could
conceivably be applied in an AFM, although no one has undertaken to do
so up until now.
3) There are also many unknowns about the question of how well isolated such proton spins could be from the environment. The small energy
scale of the nuclear spin is an advantage, as it reduces the phase space available for the emission of excitations (phonons, say) into the crystal. Other
stray spins, from defects or thermally activated electrons, say, would be a
concern. I would not want to try to predict the decoherence time attainable
in such an instrument from a purely theoretical approach; ultimately, it is,
I think, an experimental question. There are localized spins in solids which
are known to have lifetimes in excess of 1000 sec[14J.
4) We know that, almost by definition, an AFM is capable of placing the
tip with respect to the surface with something approximating atomic-scale
accuracy. This presumably means that one can turn on and off the spinspin interaction Hamiltonian between a spin at the tip of the AFM and a
selected spin at the surface, at least with 0(1) accuracy. We normally do
not think that 0(1) accuracy is anywhere near good enough, as discussed
above we are more comfortable with 0(10- 6 ) accuracy. I think that it is
completely unknown whether the AFM's positioning accuracy could be
made that good, or whether protocols for quantum computation can be
devised in which a low level of positioning accuracy is acceptable. Not that
even the possibly crude atomic selectivity of the AFM is not available at
all in many other experimental techniques in solid state physics.
One reason why 0(1) accuracy of positioning may be enough is that
the principal purpose of this interaction as I have envisioned it is to shift
the resonance frequency of the selected pair of spins so that their state can
be manipulated independently from the other spins in the system. One can
tailor pulse sequences in NMR which will accurately perform some desired
unitary transformation on the pair of spins in such a way that the result
is insensitive to the exact resonance frequency, so long as it falls in a given
range.
Another problem which spans both the issues of isolation and controllability, which I am grateful to Prof. Jon Machta for bringing to my attention,
is this: the surface will exert a force on the tip, which will be dependent
upon the state of the two spins. In principle this means that the trajectory
of the tip r(t) is dependent upon the state of the spins. This can be very
bad for quantum computation, since it spoils the desired isolation of the
quantum state from the environment; if the quantum state of the spins
becomes entangled with the state of motion of the tip, the quantum com-
664
putation state will be effectively decohered. Fortunately the situation is not
as bad as it looks, for several reasons: first, it is reasonable to take the tip
not to be in an eigenstate of position f, but in a coherent state involving
some reasonable degree of uncertainty in the position D..f and momentum
D..p of the tip. Since the tip is rather massive and the nuclear spin forces
are rather small, the deflection of the tip due to these forces will in fact
quite small, and it would be reasonable to imagine that this deflection is
well within the position uncertainly D..f of the tip. In this case, the entanglement of the motion of the tip with the quantum computation is quite
small. In fact this is the same form of argument that is needed to say why
the radio-frequency spectroscopy, which involves absorption and emission
of phonons, does not represent a loss of coherence to the environment. The
point is that the radio-frequency electric and magnetic fields are coherent
states involving considerable uncertainty in the photon number, so that
one photon more or less does not change the quantum state of the rJ. field
appreciably.
There is one other subtlety concerning the spin forces on the tip. As
noted, the trajectory is changed in no significant way by this interaction;
however, the interaction does apply an impulse to the tip, dependent on
the state of the spins 81 and 82,
(4)
which results in a subsequent change of the phase evolution after this impulse:
(5)
(r(t) is the subsequent trajectory of the tip). Note that so long as knowledge
of this phase evolution is retained (requiring, for example, quite accurate
knowledge of the whole trajectory f(t) of the tip), this effect is not decohering; it merely appends a definite phase to each part of the quantum
computer state, which must be accounted for in the bookkeeping of doing
the prescribed unitary transformations properly.
5) Finally, the situation for actually performing a strong measurement
of the quantum state of individual spins in the AFM is rather hopeful,
in the sense that this has already been recognized as a valuable thing to
do, and several researchers have been carrying out a whole experimental
program to try to do such measurements[15]. They are very hard, indeed,
and have not yet been accomplished - they involve, in fact, the self-same
minute forces mentioned in Eq. (4), which, I have already noted, are very
hard to entangle with an external variable like the tip position. However,
if you really want this entanglement to occur, a few different tricks are
available to you to amplify the tiny entangling tendency: the tricks involve
665
keeping the tip in contact with the spin to be measured for a long time,
and arranging that the force oscillate in time (flipping the spin back and
forth with appropriately timed tipping pulses) so as to excite a mechanical
resonance of the tip. Estimates indicate that under favorable circumstances
this strategy can achieve single-spin sensitivity, although we will see if the
experiments ever actually manage it.
2.2. JOSEPHSON JUNCTION DEVICE
Let me consider, in even less detail than above, the five criteria for quantum
computation as applied to Josephson junction systems. I will not pretend
that I seriously know how to perform this evaluation realistically, but I
think that other participants at this school might, prompted by my feeble
attempt, be able to do a much better job of it.
1) The Hilbert space we would have in mind would describe the quantized states of the superconducting phase. The fact that the phase is a
"macroscopic quantized variable" is by no means an unalloyed advantage
(see below), but at least it offers the possibility of sculpting a Hilbert space
by a suitable design of an electric circuit. What would be the good way of
doing this sculpting I do not really know; one possible approach, inspired
in my mind by some comments by Prof. Mooij, is to use the quantized
states of position of superconducting vortices[16] as the relevant degree of
freedom. So, the two basis states of a qubit may be a vortex positioned in
one superconducting loop or another neighboring one (with a Josephson
junction in between). It would be necessary to have "macroscopic quantum
coherence" in that this "macroscopic" vortex would have to be capable of
existing in a quantum superposition of positions.
2) Concerning state preparation, it seems possible that, by suitable external application of supercurrents, the potential profile of the superconducting state may be biased such that, for example, the lowest energy state
is the vortex in one particular position. Presumably this would result in
the preparation of a pure state, although one has to worry about thermal
excitations of the vortex into other states within the same ring, say. There
are a lot of unknowns on this score.
3) Again, on the issue of the isolation of the system from its environment, we would quickly enter the realm of guesswork. The experiments to
pin this down would correspond rather closely to those which have been
attempted for years to document the occurrence of MQT and MQC (macroscopic quantum tunneling and coherence) in these systems[17]. Such experiments would have to work before we could say that the coupling to the
environment is sufficiently under control to embark on quantum computation experiments.
666
4) The manipulation of the time evolution of the system, by applying
desired time dependent Hamiltonians to the system, is probably the most
achievable of the five criteria. The effective Hamiltonian of the quantized
superconducting phase (or of the conjugate number operator) contains a
variety of parameters which are determined by the macroscopic condition
of the circuit, for instance the capacitance or inductance of various circuit
elements, or the value of various externally applied supercurrents in the
circuit. All of these, in principle, could be employed to implement quantum
gates. It would make sense to explore protocols for doing so if the other
criteria for quantum computation were closer to being sorted out.
5) Again, the fact that the system is more-or-Iess macroscopic makes it
easy to envision various kinds of measurements, of local magnetic flux, of
voltage, etc., being performed reliably on the system. The headache comes
from the fact that when the means for performing these measurements are
put into place, one must ask whether they form a potentially destructive
part of the environment. The mere act of refraining from viewing the display of a voltmeter does not change the fact that the measurement setup
is collapsing the wavefunction of the computation in an undesirable way.
The point is that there must be some way of coupling and decoupling the
measurement with the quantum system at any desired moment, as is routinely done by applying appropriate resonant r J. fields in NMR. I do not
know how this criterion is to be satisfied, although I believe there is some
consideration of this issue in the work of Leggett[18].
Any of you who work in superconductivity who have just read this must
be convinced that quantum computation is just some wild-eyed notion that
bears no relationship to reality at all. But the remarkable (or perhaps depressing) fact of the matter is that in other fields of experimental physics
workers have a certain degree of confidence that the five criteria which I
have laid out are within the grasp of their experimental technique. The
most notable example is the consideration of the linear ion trap which has
been put forward in the work of Cirac and Zoller[19]. I can succinctly summarize their proposals according to my five-point plan, thus: 1) Hilbert
space: very precisely understood - it is spanned by the energy levels of
the isolated ions, combined with the phonon modes of the ions in the trap.
2) Cooling: done - the capability of laser cooling into the ground state
was demonstrated a couple of years ago (for single ions, at least; cooling
in multi-ion traps still has a ways to go). 3) Isolation: pretty good - the
coherence times of the ion levels is unmeasurably long, the phonon lifetimes
are adequately long to make a start of quantum computing (although these
lifetimes are a few orders of magnitude shorter than are presently understood theoretically). 4) Unitary operations: done, with very high precision,
667
by laser spectroscopy. (There is a lot of detailed atomic physics that is being

argued about now on how precisely these state transformations can actually
be effected, but the arguments are at the 1-part-in-104 level. Also, the relevant precision spectroscopy has not been demonstrated for the multi-ion
experiments.) 5) Measurement: Perfect - The laser-induced fluorescence
technique has 100% quantum efficiency, and can be turned on and off at
will.
Despair may be the reaction of those of you reading this who had hopes
that a solid-state implementation of quantum computing might be possible,
or at least competitive with what can be done in atomic physics. But please
don't let the apparent tone of finality in the survey of prospects for the
ion trap, nor the tone of uncertainty and pessimism in the survey of the
superconducting implementation, deter you. For one thing, the ion trap
has more problems than my survey revealed. First, the ion-trap scheme is
not very extendible; the most ions which have ever been trapped in such an
apparatus is in the neighborhood of 40; moreover, all the various cooling and
spectroscopic tricks have been performed presently only on single trapped
ions[20j. The potential extendibility of a solid-state system, if can be made
to work at all, would seem to be much greater. Second, and most obvious,
I may have stupidly missed some brilliant way of implementing quantum
computation in superconducting circuits which does not suffer from any of
the problems which my thoughts entail. Let not my lack of brilliance stand
in the way of yours.
3. Current concepts in error correction
Having provided a grand tour in the last section of the requirements for
quantum computation, I propose to get down in the trenches a little bit, and
discuss how we are trying to solve some particular problems on the subject
of protecting quantum computation from errors[10, 21, 22j. This is primarily
a subtopic of criterion 3) of my list above, since it has to do with how well
the quantum computer can survive interactions with its environment. It
also bears upon criterion 4), in so far as it also provides a prescription for
how to tolerate (slightly) inaccurate quantum-gate operations (that is, the
unitary transformations involved in quantum computation).
The following discussion will have no pretension to be a complete description of what is going on now on the subject of error correction. It is
intended as an introduction of what I consider to be an interesting subset
of the current work on the subject.
668
3.1. DESCRIPTION OF ERRORS
The description of the interaction of a quantum system with its environment

has been a subject for theorizing for many decades now, and can be as
complex as you please. Let me bring up a number of elementary points
about this description, which will be the only points which are needed for
quantum error correction.
Suppose that we begin with a single two-level quantum system, and
that it starts out at time t = 0 in a pure initial state \]"I. If this system then
begins to interact with its environment, the state of the system plus its environment undergoes some joint evolution. Viewing the evolution just from
the point of view of the two-state system, the initial state \]"I evolves into
a mixed state, which may either be thought of as an incoherent statistical
mixture of an ensemble of new pure states, or it may simply described by a
density matrix p[ll]. The linear operator which specifies the time evolution
of such an open system is termed a superoperator[23, 24]; this jargon, I believe, is meant to distinguish it from an ordinary (unitary) time-evolution
operator describing the time evolution of a closed quantum system.
A very useful mathematical description of the superoperators is in terms
of the so-called operator sum representation, in which this time-evolution
is described as follows:
4
p(t) =
Aip(O)AL
(6)
= 1,
(7)
1\]"1)(\]"11.
(8)
i=l
4
LA!Ai
i=l
p(O)
The action of every possible environment is completely describable by the

four operators (2x2 matrices) Ai; these matrices may take any form consistent with the completeness condition Eq. (7). The A matrices also come
up in the "ensemble" description of the effect of the quantum environment;
we would say that the state \]"I evolves into the ensemble of four states
(9)
3.2. ERROR CORRECTION CONDITIONS
Now the ideal of quantum error correction is that it should be possible to

subject the corrupted quantum state to some process n which combines
669
measurement and unitary transformation that brings any corrupted state
back to the state without errors:
(10)
Obviously it is impossible to satisfy Eq. (10) for any arbitrary \[! and error
process Ai. The remarkable fact discovered recently is that, if the subspace
in which the state \[! lies in is restricted, and the set of allowed error operators Ai is also restricted (in a physically reasonable way, as it turns out),
then the desire expressed by Eq. (10) can be satisfied. I will review here
the derivation of the conditions to be satisfied in order that this quantum
error correction process be possible, which may be found in our paper[22J
and in a number of others[23, 25, 26J.
The basic idea is just to formalize Eq. (10) a little more carefully. Let
me specialize to the case where the error-protected states \[! are to lie in a
two-dimensional subspace of the Hilbert space of n spins - the jargon for
this is that we will use n qubits to store one qubit robustly. Introducing an
orthogonal basis for this two-dimensional subspace in an obvious way, we
will write the general \[! as
(11)
with arbitrary complex coefficients a and b. Now, the error correction process R, involving some measurements and unitary operations, can always
be thought of as an entirely unitary process Ru involving a larger system,
the Hilbert space of \[! along with what we will term an ancilla Hilbert
space. We will imagine that the ancilla, which must be under the control
of the experiment in the sense of criterion 1) of the first section, is always
preset to a standard state 10). Now, we may restate the requirement, by
saying that there should exist a unitary transformation Ru which performs
the mapping
(12)
(Here the II ... 11 indicates that the state may have to be normalized first.)
The operation Ru is to restore the state as indicated for all error processes
Ai which are to be corrected. As indicated here, the ancilla will end up
in some new state ai which is dependent on Ai; the main content of this
equation is that whatever the final ancilla state, it must be in a product
state with the system, and the system's state should be restored to its
original form, \[!.
Now, by linearity, Eq. (12) should apply to each basis state separately:
(13)
670
(14)
I have now introduced explicitly the normalization constants N& 1. Now, it is
relatively easy to show that this normalization factor must be independent
of W in order that error correction work: N& = Nt = Ni . I will leave this
as an exercise for the reader to work out (or look up[22]), in order to move
on to the more physically illuminating part of the proof.
The real point is this: in order for the map required in Eqs. (13) and
(14) to be unitary, it must preserve the inner product between any pair
of states. Therefore, the derivation just consists of writing down all the
distinct before-and-after inner product conditions. First, the inner product
of two states from Eq. (13) with different error operators i and j gives:
(voIA!Ajlvo) = (;i)2 (ailaj).
(15)
Doing the same for VI, Eq. (14), gives
(vII A!A j lvl) = (;i)2 (ailaj).
(16)
The right-hand sides of Eqs. (15) and (16) are equal, giving the first conditions for error correction:
(17)
The other error correction condition is given by taking the inner product
of two vectors from Eqs. (13) and (14); since Iva) and IVl} are orthogonal:
(vIIA!Ajlvo) = o.
(18)
3.3. USE OF CANONICAL ERRORS FOR GENERAL CASE
So, Eqs. (17) and (18) specify the quantum states Iva} and IVl} which can
be error-corrected after being subject to one from among the set of errors
{ Ai}. This may seem to be not generally very useful because each quantum
mechanical system will have some environment, described by some very
particular set of Ai's. Nevertheless, there are some generic error-protected
spaces IVO,I} which will be correctable against a whole class of errors. To
explain this I have to introduce the idea of a canonical error set. Consider
first the environment of a single qubit. Note that a complete basis for 2 x 2
matrices is given by the set of matrices
Eo =
(~ ~)
El =
(~
~)
(19)
671
Suppose that we seek a "generic" correctable subspace of n qubits for which

we require that error processes which involve one of the "canonical" errors of
Eq. (19) acting on only one out of the n qubits. To take a specific example,
if n = 5 the error operators for which the error-correction conditions Eq.
(17) and (18) are to be satisfied should be these sixteen:
(20)
Here E;i) refers to error j on the ith qubit. Of course, to satisfy the completeness condition Eq. (7) these sixteen operators would have to be multiplied
by some constants; but these constants have no bearing on the satisfaction
of the error correction conditions (17) and (18), so we can safely ignore
them.
Now of course the reason I use the example n = 5 is that in fact we,
and others, have found vectors Va and VI that are perfectly correctable
when subject to this restricted error set. Below I will present a complete
description of this five-bit code after I have introduced a good general strategy for finding such codes. There is already coming to be a vast literature
on these and related "codes", group theoretic strategies for constructing
them, related theorems on the capacity of noisy quantum channels, quantum gate implementations of the error correction process, applications of
these codes for making quantum computation fault tolerant; all of these
results are vastly important, but I will leave most of them for the reader
to look up elsewhere. For I want to come back to the issue with which I
started, namely, what do the existence of these quantum error-correction
codes for these kinds of "canonical" error sets as in Eq. (20) have to say
about the case of the "generic" environment with arbitrary error operators
Ai?
The most general answer to this question is, "nothing." The most generic
environment, which introduces correlated errors among the various qubits,
bears no resemblance whatsoever to the canonical error set in Eq. (20).
However, if we consider errors which are non-generic in so far as they correspond to an independent environment acting on each qubit, then the
answer to the question changes from "nothing" to "quite a bit." For if the
environments of each qubit are independent, then irrespective of the form of
those environments, the error operators Ai of the complete system become
672
a direct product of error operators on individual qubits, as in Eq. (20):

_
(1)
(2)
(3)
A{i1,i2,i3, ... } - Ail Ai2 Ai3 ...
(21)
Now, each qubit error operator A~~) can be expanded in terms of the matrices for the canonical error operators (since they form a complete set):
(n)
A in
= "L...J ai,n,kEk,
(22)
k=O
so that the full error operator can be expanded as a sum of products of

canonical error operators:
A{i1,i2,i3, ... }
k1,k2,k3, ...
a{kn}EW Eg)
E~~
...
(23)
Now, finally, it should be clear what good this is: if the error-correction
scheme is capable of correcting all the "canonical" errors which occur in the
expansion of the right-hand side of Eq. (23), then, because of the linearity
of the error correcting process, the actual error operator A{i1,i2,i3, ... } will be
successfully corrected.
This is still not the end of the story, because the terms in Eq. (23) will
generally contain all possible products of the E operators, and we have
argued that there cannot exist an error correction procedure that corrects
for all such errors. In fact we have to invoke one more physical requirement
which we expect most sensible environments to obey: at a function of the
strength of interaction between the system and the environment, or as a
function of the time of interaction, we expect that for weak coupling, or
short time, the dominant error operator should be very close to the identity.
That is to say, the system cannot be corrupted very much after a short
period of interaction. This principle allows us to argue which terms will
predominate in the sum Eq. (23) for a sufficiently weak coupling to the
environment. There will be one of the A operators in Eq. (23) that is close
to the identity operator, which is oth order in time t. Then there will be a
set of A's for which the leading terms in the sum of Eq. (23) are "singleerror" terms, involving only one non-identity E operator; for a wide class
of noisy environments[27]' these will be of order O(t 1/ 2 ). The next group of
A's will be "double-error" terms, which will be O(t 2 / 2 )j the "triple-error"
terms will be O(t 3 / 2 ); and so forth. Thus, the generic error correcting code
which corrects some number of "canonical" errors of Eq. (20) will take care
of the most important parts of the generic errors described by A (at least,
at early times). The final statement (which I will make without proof) is
that if one has an e error correcting code (correcting canonical errors), then
673
after correction the density matrix p(t) will differ from the ideal state 'IT(t)
by terms of order t e+1 :
('IT(t)lp(t)I'lT(t))
=1-
ct e+ I
+ ...
(24)
The constant c in front of the error term will be dependent on the details
of the noise as expressed in the A's in Eq. (23).
3.4. SIMPLE EXAMPLE OF ERROR CORRECTION
Having laid out the general principles of error correction, I want to briefly
review how it actually works for the simplest "canonical" example, the one
of five qubits subjected only to the one-bit standard errors of Eq. (20).
In our initial work[22J we embarked on a purely numerical search for two
orthogonal vectors Iva) and IVI) in the 32-dimensional Hilbert space of the
five spins which would satisfy all the conditions Eqs. (17) and (18). We
indeed succeeded by using this strategy, but in fact there is a much better
strategy that was developed very rapidly by a number of other authors,
which I will indicate briefly here. The good approach is a group-theoretic
one[28, 29J: suppose there is a set of operators Mi E M each of which leave
the code vectors invariant:
(25)
We note that if such a set of operators exists it must form a group, and
that it is a reasonable guess that this group is Abelian (i.e., all the M's
commuting). The group structure is guaranteed because if MI and M2
both satisfy Eq. (25), then obviously MIM2 does as well. The fact that
these operators should also commute is not guaranteed, but is suggested by
Eq. (25); an elementary fact which we learn in quantum mechanics is that
commuting operators have simultaneous eigenvectors, and Eq. (25) asserts
that these operators have at least two simultaneous eigenvectors, Iva) and
IVI). (Commutivity would be assured if we could assert that the operators
had all 32 eigenvectors in common, not just 2. This will turn out to be the
case[28J.)
Now, the next reasonable guess to be made is that if these group elements M exist, they should themselves be expressible as products of the
canonical E operators of Eq. (19). After all, the E operators are basically
spin-l/2 angular momentum operators (i.e., the Pauli matrices), and these
operators always either commute or anticommute. So, we have a chance of
building a set of commuting M s this way.
In fact, the possibility that such operators may also anticommute is the
final piece of this mathematical trickery. For suppose that we have found
a set of Ms which commute with each other, but do not commute with all
674
the canonical error operators Acanon. of Eq. (20). This means that there
will be one operator Aianon . (or more) which an element Ma anticommutes
with:
(26)
But watch what happens when I take the matrix elements of this equation
between code vectors Va,l:
(vaIMaAianonIV{3)
= -(vaIAianonMalv{3)
(vaIAianonlv{3) = -(vaIAianonlv{3)
= O.
(27)
(28)
So, we see that the anticommuting condition leads to the satisfaction of

the error correction conditions Eq. (17) and (18), where the first condition
is satisfied with the extra condition that both matrix elements are equal
to zero. The argument just given is obviously not true for the case that
Aianon . is the identity operator (since this will commute with everything);
but in this case Aianon . is in the group M, which also leads to Eqs. (17)
(18) being satisfied.
I can now finish this off by giving the full statement of the result of
Gottesman[28] and of Calderbank et al.[29]: Vectors IVa) can be corrected when subjected to errors A if they are the eigenvectors of
an Abelian group of operators M such that every operator At Aj
either 1) is itself a member of the group M, or 2) anticommutes
with at least one element of M.
This result has provided a very useful means of searching for and discovering a vast variety of error correcting schemes. The five-bit error correcting
code can be very succinctly expressed using this language: the code vectors
are eigenstates of the sixteen-element Abelian group which is generated by
the four operators:
~~l) ~~2) ~~3) ~~5)
~(2) ~(3) ~(4) ~(1)
1
133
~(3) ~(4) ~(5) ~(2)

1
1
3
3
~(4) ~(5) ~(l) ~(3)
(29)
133
One choice of the two vectors which are simultaneous eigenvectors of these
operators, which may be found by standard projection-operator techniques
from group theory, are:
Iva)
100000)
111000) + 101100) + 100110) + 100011) + 110001)
110100) -101010) - 100101) -110010) - 101001)
111110) -101111) -110111) - 111011) -111101)
(30)
675
and
lei)
111111)
+
(31)
100111) + 110011) + 111001) + 111100) + 101110)

101011) -110101) -111010) -101101) -110110)
100001) -110000) - 101000) -100100) -100010).
The entire error-correction process can also be described very succinctly

and physically in this language: The prescription is to perform a measurement of the value of each of the four generators of the M group, Eq. (29).
(Although I have not mentioned it previously, they are in fact Hermitian
operators.) Being the product offour spin-1/2 operators, the measurement
can only have two outcomes, l. If the measurement outcomes are all +1,
this indicates that the state is free from error, and nothing need be done. It
turns out that each other fifteen patterns of 1 in this measurement indicates which of the fifteen one-error operators of Eq. (20) that the state has
been subjected to. Knowing which of these (unitary) operators the state has
been subjected to, simply performing the inverse unitary operation restores
the state to its correct value.
Both these multi-spin measurements and the final unitary correction operations are within the capability of a quantum computer as specified in the
first part of this paper. Error correction indicates a modified paradigm for
quantum computation, however; instead of measurements being performed
only at the final, "readout" phase of the computation, error correction indicates that measurements should be performed at fixed intervals throughout
the entire course of the computation.
3.5. FINAL REMARKS ON ERROR CORRECTION, ETC.
As a summary remark on all this, I freely admit that the foregoing barely
scratches the surface of the current activity in the theory of quantum quantum computation, even of the theory of error correction in quantum computation. Any such survey must be obsolete as soon as it is written, with
new families of error correcting codes being discovered, their connection
with classical error correcting theory being further uncovered, and proposals being advanced for how to use error correction in a full protocol for
performing reliable quantum computation with faulty elements. The claim
which is presently being evaluated is that if the error rate (including that
induced by interaction with the environment, as well as inaccuracies in the
implementation of the quantum gates) is below a certain value, then reliable
quantum computation of indefinite scale and duration becomes possible.
The bad news in this is that this threshold for reliable quantum computation is presently quite low, in the neighborhood of 10-- 5 per "clock cycle"
676
(intervals between quantum gate operations). So, there's clearly a lot more
territory for the theory to explore at this point, and it goes without saying
that the experimental situation is still in its infancy. Perhaps some freshminded youngster reading this will have a good idea for how to make good
progress in bringing quantum computation closer to reality.
I am grateful to the Army Research Office for the support of this work,
as well as the hospitality of the Program on Quantum Computers and
Quantum Coherence at the Institute for Theoretical Physics at UCSB where
some of this work was completed.
References
1.
Bennett, C. H. (1995) Quantum Information and Computation, Physics Today 48,

no. 10, pp. 24-30.
2. DiVincenzo, D. P. (1995b) Quantum Computation, Science 269, pp. 255-261.
3. Ekert, A. and Jozsa, R (1996) Quantum Computation and Shor's Factoring Algorithm, Reviews of Modern Physics 68, pp. 733-754.
4. Lloyd, S. (1995) Quantum Computation, Scientific American 273, no. 4, pp. 44-50.
5. Glanz, J. (1995) A quantum leap for computers?, Science 269, pp. 28-29.
6. Barenco, A., Bennett, C. H., Cleve, R, DiVincenzo, D. P., Margolus, N., Shor, P.,
Sleator, T., Smolin, J. A. and Weinfurter, H. (1995) Elementary gates for quantum
computation, Phys. Rev. A 51, pp. 3457-3467.
7. Shor, P. (1994) Algorithms for quantum computation: discrete logarithms and factoring, in Proceedings of the 35th Annual Symposium on the Foundations of Computer
Science, IEEE Press, Los Alamitos, pp. 124-134.
8. Schrodinger, E. (1935) The present situation in quantum mechanics, Naturwissenschaften 23, pp. 807-812, 823-828, 844-849 (Translated in Proc. Amer. Phil. Soc.
124, 323 (1980), and reprinted in Quantum Theory and Measurement, eds. J. A.
Wheeler and W. H. Zurek (Princeton, 1983).
9. Schrodinger, E. (1935) Discussion of probability relations between separated systems,
Proc. Camb. Phil. Soc. 31, pp. 555-563; Schrodinger, E. (1936) Probability relations
between separated systems, Proc. Camb. Phil. Soc. 32, pp. 446-452.
10. Shor, P. W. (1996) Fault-tolerant quantum computation, e-print quant-ph/9605011;
Knill, E.,Lafiamme, R., and Zurek, W. (1996) Threshold Accuracy for Quantum Computation, e-print quant-ph/9610011.
11. Peres, A. (1993) Quantum theory: concepts and methods, Kluwer Academic Publishers, Dordrecht.
12. DiVincenzo, D. P. (1995) Two-bit gates are universal for quantum computation,
Phys. Rev. A 51, pp. 1015-1022.
13. Albert, M., Cates, G., Driehuys, B, Happer, W, Saam, B., Springer, C., and Wishnia,
A. (1994), Nature 370, p. 199; Inhaling Hyperpolarized Noble Gas Helps Magnetic
Resonance Imaging of Lungs, Physics Today (Search and Discovery) 48, no. 6, p. 17
(1995).
14. Castle Jr., J. G. et al., (1963) Physical Review 130, p. 577.
15. Zuger, 0., and Rugar, D. (1993) Appl. Phys. Lett. 63 p. 2496.
16. van Oudenaarden, A. and Mooij, J. E. (1996) One dimensional Mott insulator
formed by quantum vortices in Josephson junction arrays, Phys. Rev. Lett. 76, pp.
4947-4950; Elion, W. J., Wachters, J. J., Sohn, L. L., and Mooij, J. E. (1994) The
Aharonov Casher effect for vortices in Josephson junction arrays, Physica B 203, pp.
497-503.
17. Rouse, R., Han, S. and Lukens, J. (1995), Phys. Rev. Lett. 75, pp. 1614-1617.
18. Leggett, A. J. (1995), in Quantum Tunneling of Magnetization, edited by L. Gunther
677
(Proceedings of the NATO ASI, Chichillianne, France).
Cirac, J. I. and Zoller, P. (1995), Phys. Rev. Lett. 74, pp. 4091-4094.
Monroe, C., Meekhof, D. M., King, B. E., Itano, W. M., and Wineland, D. J. (1995)
Demonstration of a fundamental quantum-logic gate, Phys. Rev. Lett. 75, pp. 47144717.
21. Shor, P. W. (1995) Scheme for Reducing Decoherence in a Quantum Memory, Phys.
Rev. A 52, p. 2493.
22. Bennett, C. H., DiVincenzo, D. P., Smolin, J. A., and Wootters, W. (1996) Mixed
State Entanglement and Error Correction, Phys. Rev. A, November, in press; see also
quant-ph/9604024.
23. Knill, E.,Laflamme, R., (1997) Theory of Error Correcting Codes, Phys. Rev. A,
January, in press; also quant-ph/9604034.
24. Schumacher, B. (1996) Sending quantum entanglement through noisy channels,
quant-ph/9604022.
25. Schumacher B., and Nielsen, M. A. (1996) Quantum data processing and error
correction, quant-ph/9604022.
26. Ekert, A. and Macchiavello, C. (1996) Quantum Error Correction for Communication, Phys. Rev. Lett. 77, pp. 2585-2588; also quant-ph/9602022.
27. Chuang, I., and Laflamme, R. (1996) Quantum error correction by coding, quantph/9511003; Creation of a persistent quantum bit using error correction, Phys. Rev.
A, January, 1997.
28. Gottesman, D. (1986) Class of quantum error-correcting codes saturating the quantum Hamming bound, Phys. Rev. A 54, pp. 1862-1868; also quant-ph/9604038.
29. Calderbank, A. R., Rains, E. M., Shor, P. W., and Sloane, N. J. A. (1996) Quantum Error Correction and Orthogonal Geometry, quant-ph/9605005; Quantum Error
Correction via Codes over GF(4), quant-ph/9608006.
19.
20.

Kouwenhoven - Mesoscopic e - Transport PDF

Caricato da

Informazioni sul documento

Descrizione originale:

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Kouwenhoven - Mesoscopic e - Transport PDF

Caricato da

Copyright:

Formati disponibili

Mesoscopic Electron Transport

NATO ASI Series

A Series presenting the results of activities sponsored by the NA TO Science Committee,

Plenum Publishing Corporation

C Mathematical and Physical Sciences

Kluwer Academic Publishers

Computer and Systems SCiences

Kluwer Academic Publishers

The Partnership Sub-Series incorporates activities undertaken in collaboration with NATO's

Series E: Applied Sciences' Vol. 345

Mesoscopic Electron Transport

Springer-Science+Business Media, B.V.

Proceedings of the NATO Advanced Study Institute on

Printed on acid-free paper

All Rights Reserved

1997 Springer Science+Business Media Oordrecht

and OSSICINI, S. (eds.), Optical Properties of Low Dimensional Silicon

Structures. (E244) 1993 ISBN 0-7923-2446-3

INTRODUCTION TO MESOSCOPIC ELECTRON TRANSPORT

DELOCALIZATION, INELASTIC SCATTERING AND TRANSPORT

ELECTRON TRANSPORT IN QUANTUM DOTS

MAGNETOTUNNELING SPECTROSCOPY: STUDYING SELF-ORGANIZED

ADMITTANCE AND NONLINEAR TRANSPORT IN QUANTUM WIRES,

TRANSPORT THEORY OF INTERACTING QUANTUM DOTS

TRANSPORT IN A ONE-DIMENSIONAL LUTTINGER LIQUID

THE PROXIMITY EFFECT IN MESO SCOPIC DIFFUSIVE CONDUCTORS

MESOSCOPIC EFFECTS IN SUPERCONDUCTIVITY

SCANNING PROBE MICROSCOPES AND THEIR APPLICATIONS

QUANTUM POINT CONTACTS BETWEEN METALS

CONDUCTANCE QUANTIZATION IN METALLIC NANOWIRES

TOPICS IN QUANTUM COMPUTERS

The major advances in lithographic technology have enabled researchers to

INTRODUCTION TO MESOSCOPIC ELECTRON TRANSPORT

Dept. of Applied Physics and DIMES

Institut fur Theoretische Festkorperphysik

Dept. of Physics, Princeton University

In this Section we introduce basic properties of a two-dimensional electron

J1e = 106 cm 2/Vs (typical value)

EF = 7rTi 2 ns /m* = 10 meV

1.2. E-BEAM FABRICATION OF A SUBMICRON SEMICONDUCTOR

We briefly outline a standard procedure for fabricating small devices in a

Figure 9. Outline of the electron-beam lithography procedure for fabricating submicron

cont ined 2DEG

1.3. QUANTIZED CONDUCTANCE OF A POINT CONTACT

The resistance of a point contact in the classical ballistic regime is known as

Dissipative processes, which bring the electron system into thermodynamic

The quantum mechanical 2D density of states per unit area, including

of the states of a particle-in-a-lD-box, which, as we show below, is the basic

resulting Schrodinger equation can be written in the form of a harmonic

which contains a free-electron kinetic energy dispersion in the longitudinal

Figure 7. Energy En versus longitudinal wave-vector k y from Eq. 4 at the bottleneck of

that of a harmonic oscillator, but now with energy eigenvalues [11]

GATE VOLTAGE (V)

In the previous Section, we discussed the influence of a magnetic field on

Figure 10. (a) Schematic classical electron trajectories in a magnetic field.

= evB and the centripetal force F = m*v2/r leads to a cyclotron motion

of the electrons, with an angular frequency We = eB/m* and at the Fermi

-0.3 -0.2 -0.1

Figure 11. Bottom: Electron focusing (T = 50 mK, L = 3.0JJm) in the configuration

1.6.1. Edge Channels

cupied up to the electrochemical potential of the last Ohmic contact they

GATE VOLTAGE IV)