a

r
X
i
v
:
c
o
n
d
-
m
a
t
/
0
4
1
2
6
6
4
v
1


[
c
o
n
d
-
m
a
t
.
m
e
s
-
h
a
l
l
]


2
3

D
e
c

2
0
0
4
Quantum Transport in Semiconductor Nanostructures
C. W. J. Beenakker and H. van Houten
Philips Research Laboratories, Eindhoven, The Netherlands
Published in Solid State Physics, 44, 1-228 (1991)
Contents
I. Introduction 1
A. Preface 1
B. Nanostructures in Si inversion layers 2
C. Nanostructures in GaAs-AlGaAs heterostructures 5
D. Basic properties 6
1. Density of states in two, one, and zero dimensions 6
2. Drude conductivity, Einstein relation, and Landauer formula8
3. Magnetotransport 10
II. Diusive and quasi-ballistic transport 12
A. Classical size eects 12
1. Boundary scattering 12
2. Magneto size eects 13
B. Weak localization 15
1. Coherent backscattering 16
2. Suppression of weak localization by a magnetic eld18
3. Boundary scattering and ux cancellation 21
C. Conductance uctuations 22
1. Zero-temperature conductance uctuations 23
2. Nonzero temperatures 24
3. Magnetoconductance uctuations 25
4. Experiments 28
D. Aharonov-Bohm eect 30
E. Electron-electron interactions 32
1. Theory 32
2. Narrow-channel experiments 34
F. Quantum size eects 36
1. Magnetoelectric subbands 36
2. Experiments on electric and magnetic depopulation of subbands38
G. Periodic potential 39
1. Lateral superlattices 39
2. Guiding-center-drift resonance 41
III. Ballistic transport 44
A. Conduction as a transmission problem 44
1. Electron waveguide 44
2. Landauer formula 47
B. Quantum point contacts 49
1. Conductance quantization 49
2. Depopulation of subbands and suppression of backscattering by a magnetic eld53
C. Coherent electron focusing 56
1. Experiments 57
2. Theory 58
3. Scattering and electron focusing 60
D. Collimation 61
1. Theory 61
2. Magnetic deection of a collimated electron beam 62
3. Series resistance 64
E. Junction scattering 66
1. Mechanisms 66
2. Magnetoresistance anomalies 67
3. Electron waveguide versus electron billiard 69
F. Tunneling 71
1. Resonant tunneling 71
2. Coulomb blockade 74
IV. Adiabatic transport 77
A. Edge channels and the quantum Hall eect 77
1. Introduction 77
2. Edge channels in a disordered conductor 77
3. Current distribution 80
B. Selective population and detection of edge channels 81
1. Ideal contacts 81
2. Disordered contacts 83
3. Quantum point contacts 86
4. Suppression of the Shubnikov-De Haas oscillations 89
C. Fractional quantum Hall eect 91
1. Introduction 91
2. Fractional edge channels 92
D. Aharonov-Bohm eect in strong magnetic elds 96
1. Suppression of the Aharonov-Bohm eect in a ring 97
2. Aharonov-Bohm eect in singly connected geometries97
E. Magnetically induced band structure 100
1. Magnetotransport through a one-dimensional superlattice100
2. Magnetically induced band structure 101
References 103
I. INTRODUCTION
A. Preface
In recent years semiconductor nanostructures have be-
come the model systems of choice for investigations of
electrical conduction on short length scales. This devel-
opment was made possible by the availability of semicon-
ducting materials of unprecedented purity and crystalline
perfection. Such materials can be structured to contain
a thin layer of highly mobile electrons. Motion perpen-
dicular to the layer is quantized, so that the electrons are
constrained to move in a plane. As a model system, this
two-dimensional electron gas (2DEG) combines a number
of desirable properties, not shared by thin metal lms. It
has a low electron density, which may be readily varied
by means of an electric eld (because of the large screen-
ing length). The low density implies a large Fermi wave-
length (typically 40 nm), comparable to the dimensions
of the smallest structures (nanostructures) that can be
fabricated today. The electron mean free path can be
quite large (exceeding 10 m). Finally, the reduced di-
mensionality of the motion and the circular Fermi surface
form simplifying factors.
Quantum transport is conveniently studied in a 2DEG
because of the combination of a large Fermi wavelength
and large mean free path. The quantum mechanical
2
phase coherence characteristic of a microscopic object
can be maintained at low temperatures (below 1 K) over
distances of several microns, which one would otherwise
have classied as macroscopic. The physics of these sys-
tems has been referred to as mesoscopic,
1
a word bor-
rowed from statistical mechanics.
2
Elastic impurity scat-
tering does not destroy phase coherence, which is why
the eects of quantum interference can modify the con-
ductivity of a disordered conductor. This is the regime
of diusive transport, characteristic for disordered met-
als. Quantum interference becomes more important as
the dimensionality of the conductor is reduced. Quasi-
one dimensionality can readily be achieved in a 2DEG by
lateral connement.
Semiconductor nanostructures are unique in oering
the possibility of studying quantum transport in an arti-
cial potential landscape. This is the regime of ballistic
transport, in which scattering with impurities can be ne-
glected. The transport properties can then be tailored by
varying the geometry of the conductor, in much the same
way as one would tailor the transmission properties of a
waveguide. The physics of this transport regime could be
called electron optics in the solid state.
3
The formal rela-
tion between conduction and transmission, known as the
Landauer formula,
1,4,5
has demonstrated its real power
in this context. For example, the quantization of the con-
ductance of a quantum point contact
6,7
(a short and nar-
row constriction in the 2DEG) can be understood using
the Landauer formula as resulting from the discreteness
of the number of propagating modes in a waveguide.
Two-dimensional systems in a perpendicular magnetic
eld have the remarkable property of a quantized Hall
resistance,
8
which results from the quantization of the
energy in a series of Landau levels. The magnetic length
( h/eB)
1/2
( 10 nm at B = 5 T) assumes the role of the
wavelength in the quantum Hall eect. The potential
landscape in a 2DEG can be adjusted to be smooth on
the scale of the magnetic length, so that inter-Landau
level scattering is suppressed. One then enters the regime
of adiabatic transport. In this regime truly macroscopic
behavior may not be found even in samples as large as
0.25 mm.
In this review we present a self-contained account of
these three novel transport regimes in semiconductor
nanostructures. The experimental and theoretical de-
velopments in this eld have developed hand in hand, a
fruitful balance that we have tried to maintain here as
well. We have opted for the simplest possible theoretical
explanations, avoiding the powerful but more formal
Greens function techniques. If in some instances this
choice has not enabled us to do full justice to a subject,
then we hope that this disadvantage is compensated by
a gain in accessibility. Lack of space and time has caused
us to limit the scope of this review to metallic transport
in the plane of a 2DEG at small currents and voltages.
Transport in the regime of strong localization is excluded,
as well as that in the regime of a nonlinear current-
voltage dependence. Overviews of these, and other, top-
ics not covered here may be found in Refs.
9,10,11
, as well
as in recent conference proceedings.
12,13,14,15,16,17
We have attempted to give a comprehensive list of
references to theoretical and experimental work on the
subjects of this review. We apologize to those whose
contributions we have overlooked. Certain experiments
are discussed in some detail. In selecting these experi-
ments, our aim has been to choose those that illustrate
a particular phenomenon in the clearest fashion, not to
establish priorities. We thank the authors and publishers
for their kind permission to reproduce gures from the
original publications. Much of the work reviewed here
was a joint eort with colleagues at the Delft University
of Technology and at the Philips Research Laboratories,
and we are grateful for the stimulating collaboration.
The study of quantum transport in semiconductor
nanostructures is motivated by more than scientic inter-
est. The fabrication of nanostructures relies on sophis-
ticated crystal growth and lithographic techniques that
exist because of the industrial eort toward the minia-
turization of transistors. Conventional transistors oper-
ate in the regime of classical diusive transport, which
breaks down on short length scales. The discovery of
novel transport regimes in semiconductor nanostructures
provides options for the development of innovative fu-
ture devices. At this point, most of the proposals in
the literature for a quantum interference device have
been presented primarily as interesting possibilities, and
they have not yet been critically analyzed. A quanti-
tative comparison with conventional transistors will be
needed, taking circuit design and technological consider-
ations into account.
18
Some proposals are very ambitious,
in that they do not only consider a dierent principle of
operation for a single transistor, but envision entire com-
puter architectures in which arrays of quantum devices
operate phase coherently.
19
We hope that the present review will convey some of
the excitement that the workers in this rewarding eld of
research have experienced in its exploration. May the de-
scription of the variety of phenomena known at present,
and of the simplest way in which they can be understood,
form an inspiration for future investigations.
B. Nanostructures in Si inversion layers
Electronic properties of the two-dimensional electron
gas in Si MOSFETs (metal-oxide-semiconductor eld-
eect transistors) have been reviewed by Ando, Fowler,
and Stern,
20
while general technological and device as-
pects are covered in detail in the books by Sze
21
and by
Nicollian and Brew.
22
In this section we only summarize
those properties that are needed in the following. A typ-
ical device consists of a p-type Si substrate, covered by a
Si0
2
layer that serves as an insulator between the (100)
Si surface and a metallic gate electrode. By application
of a suciently strong positive voltage V
g
on the gate, a
2DEG is induced electrostatically in the p-type Si under
3
FIG. 1 Band-bending diagram (showing conduction band
Ec, valence band Ev, and Fermi level EF) of a metal-oxide-
semiconductor (MOS) structure. A 2DEG is formed at the
interface between the oxide and the p-type silicon substrate,
as a consequence of the positive voltage V on the metal gate
electrode.
the gate. The band bending leading to the formation of
this inversion layer is schematically indicated in Fig. 1.
The areal electron concentration (or sheet density) n
s
fol-
lows from en
s
= C
ox
(V
g
V
t
), where V
t
is the threshold
voltage beyond which the inversion layer is created, and
C
ox
is the capacitance per unit area of the gate electrode
with respect to the electron gas. Approximately, one has
C
ox
=
ox
/d
ox
(with
ox
= 3.9
0
the dielectric constant
of the Si0
2
layer),
21
so
n
s
=

ox
ed
ox
(V
g
V
t
). (1.1)
The linear dependence of the sheet density on the applied
gate voltage is one of the most useful properties of Si
inversion layers.
The electric eld across the oxide layer resulting from
the applied gate voltage can be quite strong. Typically,
V
g
V
t
= 5 V and d
ox
= 50 nm, so the eld strength
is of order 1 MV/cm, at best a factor of 10 lower than
typical elds for the dielectric breakdown of Si0
2
. It is
possible to change the electric eld at the interface, with-
out altering n
s
, by applying an additional voltage across
the p-n junction that isolates the inversion layer from the
p-type substrate (such a voltage is referred to as a sub-
strate bias). At the Si-Si0
2
interface the electric eld is
continuous, but there is an electrostatic potential step of
about 3 eV. An approximately triangular potential well
is thus formed at the interface (see Fig. 1). The actual
shape of the potential deviates somewhat from the tri-
angular one due to the electronic charge in the inversion
layer, and has to be calculated self-consistently.
20
Due
to the connement in one direction in this potential well,
the three-dimensional conduction band splits into a series
of two-dimensional subbands. Under typical conditions
(for a sheet electron density n
s
= 10
11
10
12
cm
2
) only
a single two-dimensional subband is occupied. Bulk Si
has an indirect band gap, with six equivalent conduction
band valleys in the (100) direction in reciprocal space. In
inversion layers on the (100) Si surface, the degeneracy
between these valleys is partially lifted. A twofold val-
ley degeneracy remains. In the following, we treat these
two valleys as completely independent, ignoring compli-
cations due to intervalley scattering. For each valley, the
(one-dimensional) Fermi surface is simply a circle, corre-
sponding to free motion in a plane with eective electron
mass
20
m = 0.19 m
e
. For easy reference, this and other
relevant numbers are listed in Table I.
The electronic properties of the Si inversion layer
can be studied by capacitive or spectroscopic techniques
(which are outside the scope of this review), as well as
by transport measurements in the plane of the 2DEG.
To determine the intrinsic transport properties of the
2DEG (e.g., the electron mobility), one denes a wide
channel by fabricating a gate electrode with the appro-
priate shape. Ohmic contacts to the channel are then
made by ion implantation, followed by a lateral diusion
and annealing process. The two current-carrying con-
tacts are referred to as the source and the drain. One of
these also serves as zero reference for the gate voltage.
Additional side contacts to the channel are often fabri-
cated as well (for example, in the Hall bar geometry), to
serve as voltage probes for measurements of the longi-
tudinal and Hall resistance. Insulation is automatically
provided by the p-n junctions surrounding the inversion
layer. (Moreover, at the low temperatures of interest
here, the substrate conduction vanishes anyway due to
carrier freeze-out.) The electron mobility
e
is an im-
portant gure of merit for the quality of the device. At
low temperatures the mobility in a given sample varies
nonmonotonically
20
with increasing electron density n
s
(or increasing gate voltage), due to the opposite eects
of enhanced screening (which reduces ionized impurity
scattering) and enhanced connement (which leads to an
increase in surface roughness scattering at the Si-Si0
2
interface). The maximum low-temperature mobility of
electrons in high-quality samples is around 10
4
cm
2
/Vs.
This review deals with the modications of the trans-
port properties of the 2DEG in narrow geometries. Sev-
eral lateral connement schemes have been tried in order
to achieve narrow inversion layer channels (see Fig. 2).
Many more have been proposed, but here we discuss only
those realized experimentally.
Technically simplest, because it does not require elec-
tron beam lithography, is an approach rst used by
Fowler et al., following a suggestion by Pepper
32,33,34
(Fig. 2a). By adjusting the negative voltage over p-n
junctions on either side of a relatively wide gate, they
were able to vary the electron channel width as well as
its electron density. This technique has been used to de-
ne narrow accumulation layers on n-type Si substrates,
rather than inversion layers. Specically, it has been used
for the exploration of quantum transport in the strongly
4
TABLE I Electronic properties of the 2DEG in GaAs-AlGaAs heterostructures and Si inversion layers.
GaAs(100) Si(100) Units
Eective Mass m 0.067 0.19 me = 9.1 10
28
g
Spin Degeneracy gs 2 2
Valley Degeneracy gv 1 2
Dielectric Constant 13.1 11.9 0 = 8.9 10
12
Fm
1
Density of States (E) = gsgv(m/2h
2
) 0.28 1.59 10
11
cm
2
meV
1
Electronic Sheet Density
a
ns 4 110 10
11
cm
2
Fermi Wave Vector kF = (4ns/gsgv)
1/2
1.58 0.561.77 10
6
cm
1
Fermi Velocity vF = hkF/m 2.7 0.341.1 10
7
cm/s
Fermi Energy EF = (hkF)
2
/2m 14 0.636.3 meV
Electron Mobility
a
e 10
4
10
6
10
4
cm
2
/Vs
Scattering Time = me/e 0.3838 1.1 ps
Diusion Constant D = v
2
F
/2 14014000 6.464 cm
2
/s
Resistivity = (nsee)
1
1.60.016 6.30.63 k
Fermi Wavelength F = 2/kF 40 11235 nm
Mean Free Path l = vF 10
2
10
4
37118 nm
Phase Coherence Length
b
l

= (D

)
1/2
200... 40400 nm(T/K)
1/2
Thermal Length lT = (hD/kBT)
1/2
3303300 70220 nm(T/K)
1/2
Cyclotron Radius l
cycl
= hkF/eB 100 37116 nm(B/T)
1
Magnetic Length lm = (h/eB)
1/2
26 26 nm(B/T)
1/2
kFl 15.81580 2.121
c 1100 1 (B/T)
EF/hc 7.9 110 (B/T)
1
a
A typical (xed) density value is taken for GaAs-AlGaAs het-
erostructures, and a typical range of values in the metallic con-
duction regime for Si MOSFETs. For the mobility, a range of
representative values is listed for GaAs-AlGaAs heterostructures,
and a typical good value for Si MOSFETs. The variation in the
other quantities reects that in ns and e.
b
Rough estimate of the phase coherence length, based
on weak localization experiments in laterally conned
heterostructures
23,24,25,26,27
and Si MOSFETs.
28,29
The stated
T
1/2
temperature dependence should be regarded as an indica-
tion only, since a simple power law dependence is not always found
(see, for example, Refs.
30
and
25
). For high-mobility GaAs-AlGaAs
heterostructures the phase coherence length is not known, but is
presumably
31
comparable to the (elastic) mean free path l.
localized regime
32,35,36,37
(which is not discussed in this
review). Perhaps the technique is particularly suited to
this highly resistive regime, since a tail of the diusion
prole inevitably extends into the channel, providing ad-
ditional scattering centers.
34
Some studies in the weak
localization regime have also been reported.
33
The conceptually simplest approach (Fig. 2b) to dene
a narrow channel is to scale down the width of the gate by
means of electron beam lithography
38
or other advanced
techniques.
39,40,41
A diculty for the characterization of
the device is that fringing elds beyond the gate induce
a considerable uncertainty in the channel width, as well
as its density. Such a problem is shared to some degree
by all approaches, however, and this technique has been
quite successful (as we will discuss in Section II). For
a theoretical study of the electrostatic conning poten-
tial induced by the narrow gate, we refer to the work
by Laux and Stern.
42
This is a complicated problem,
which requires a self-consistent solution of the Poisson
and Schrodinger equations, and must be solved numeri-
cally.
The narrow gate technique has been modied by War-
ren et al.
43,44
(Fig. 2c), who covered a multiple narrow-
gate structure with a second dielectric followed by a sec-
ond gate covering the entire device. (This structure was
specically intended to study one-dimensional superlat-
tice eects, which is why multiple narrow gates were
used.) By separately varying the voltages on the two
gates, one achieves an increased control over channel
width and density. The electrostatics of this particu-
lar structure has been studied in Ref.
43
in a semiclassical
approximation.
5
FIG. 2 Schematic cross-sectional views of the lateral pinch-
o technique used to dene a narrow electron accumulation
layer (a), and of three dierent methods to dene a narrow
inversion layer in Si MOSFETs (b-d). Positive (+) and neg-
ative () charges on the gate electrodes are indicated. The
location of the 2DEG is shown in black.
Skocpol et al.
29,45
have combined a narrow gate with
a deep self-aligned mesa structure (Fig. 2d), fabricated
using dry-etching techniques. One advantage of their
method is that at least an upper bound on the channel
width is known unequivocally. A disadvantage is that
the deep etch exposes the sidewalls of the electron gas,
so that it is likely that some mobility reduction occurs
due to sidewall scattering. In addition, the deep etch
may damage the 2DEG itself. This approach has been
used successfully in the exploration of nonlocal quantum
transport in multiprobe channels, which in addition to
being narrow have a very small separation of the voltage
probes.
45,46
In another investigation these narrow chan-
nels have been used as instruments sensitive to the charg-
ing and discharging of a single electron trap, allowing a
detailed study of the statistics of trap kinetics.
46,47,48
C. Nanostructures in GaAs-AlGaAs heterostructures
In a modulation-doped
49
GaAs-AlGaAs heterostruc-
ture, the 2DEG is present at the interface between GaAs
and Al
x
Ga
1x
As layers (for a recent review, see Ref.
50
).
Typically, the Al mole fraction x = 0.3. As shown in the
band-bending diagram of Fig. 3, the electrons are con-
ned to the GaAs-AlGaAs interface by a potential well,
formed by the repulsive barrier due to the conduction
band oset of about 0.3 V between the two semiconduc-
tors, and by the attractive electrostatic potential due to
the positively charged ionized donors in the n-doped Al-
GaAs layer. To reduce scattering from these donors, the
doped layer is separated from the interface by an un-
doped AlGaAs spacer layer. Two-dimensional sub bands
are formed as a result of connement perpendicular to the
interface and free motion along the interface. An impor-
FIG. 3 Band-bending diagram of a modulation doped GaAs-
AlxGa1xAs heterostructure. A 2DEG is formed in the un-
doped GaAs at the interface with the p-type doped AlGaAs.
Note the Schottky barrier between the semiconductor and a
metal electrode.
tant advantage over a MOSFET is that the present inter-
face does not interrupt the crystalline periodicity. This
is possible because GaAs and AlGaAs have almost the
same lattice spacing. Because of the absence of bound-
ary scattering at the interface, the electron mobility can
be higher by many orders of magnitude (see Table I).
The mobility is also high because of the low eective
mass m = 0.067 m
e
in GaAs (for a review of GaAs ma-
terial properties, see Ref.
51
). As in a Si inversion layer,
only a single two-dimensional subband (associated with
the lowest discrete connement level in the well) is usu-
ally populated. Since GaAs has a direct band gap, with
a single conduction band minimum, complications due
to intervalley scattering (as in Si) are absent. The one-
dimensional Fermi surface is a circle, for the commonly
used (100) substrate orientation.
Since the 2DEG is present naturally due to the mod-
ulation doping (i.e., even in the absence of a gate), the
creation of a narrow channel now requires the selective
depletion of the electron gas in spatially separated re-
gions. In principle, one could imagine using a combina-
tion of an undoped heterostructure and a narrow gate
(similarly to a MOSFET), but in practice this does not
work very well due to the lack of a natural oxide to serve
as an insulator on top of the AlGaAs. The Schottky bar-
rier between a metal and (Al)GaAs (see Fig. 3) is too
low (only 0.9 V) to sustain a large positive voltage on
the gate. For depletion-type devices, where a negative
voltage is applied on the gate, the Schottky barrier is
quite sucient as a gate insulator (see, e.g., Ref.
52
).
The simplest lateral connement technique is illus-
trated in Fig. 4a. The appropriate device geometry (such
as a Hall bar) is realized by dening a deep mesa, by
means of wet chemical etching. Wide Hall bars are usu-
6
FIG. 4 Schematic cross-sectional views of four dierent ways
to dene narrow 2DEG channels in a GaAs-AlGaAs het-
erostructure. Positive ionized donors and negative charges on
a Schottky gate electrode are indicated. The hatched squares
in d represent unremoved resist used as a gate dielectric.
ally fabricated in this way. This approach has also been
used to fabricate the rst micron-scale devices, such as
the constrictions used in the study of the breakdown of
the quantum Hall eect by Kirtley et al.
53
and Bliek et
al.,
54
and the narrow channels used in the rst study of
quasi-one-dimensional quantum transport in heterostruc-
tures by Choi et al.
55
The deep-mesa connement tech-
nique using wet
25,56
or dry
57
etching is still of use for
some experimental studies, but it is generally felt to be
unreliable for channels less than 1 m wide (in particular
because of the exposed sidewalls of the structure).
The rst working alternative connement scheme was
developed by Thornton et al.
58
and Zheng et al.,
24
who
introduced the split-gate lateral connement technique
(Fig. 4b). On application of a negative voltage to a split
Schottky gate, wide 2DEG regions under the gate are de-
pleted, leaving a narrow channel undepleted. The most
appealing feature of this connement scheme is that the
channel width and electron density can be varied contin-
uously (but not independently) by increasing the nega-
tive gate voltage beyond the depletion threshold in the
wide regions (typically about 0.6 V). The split-gate
technique has become very popular, especially after it
was used to fabricate the short and narrow constrictions
known as quantum point contacts
6,7,59
(see Section III).
The electrostatic connement problem for the split-gate
geometry has been studied numerically in Refs.
60
and
61
.
A simple analytical treatment is given in in Ref.
62
. A
modication of the split-gate technique is the grating-
gate technique, which may be used to dene a 2DEG
with a periodic density modulation.
62
The second widely used approach is the shallow-mesa
depletion technique (Fig. 4c), introduced in Ref.
63
. This
technique relies on the fact that a 2DEG can be de-
pleted by removal of only a thin layer of the AlGaAs,
the required thickness being a sensitive function of the
parameters of the heterostructure material, and of de-
tails of the lithographic process (which usually involves
electron beam lithography followed by dry etching). The
shallow-mesa etch technique has been perfected by two
groups,
64,65,66
for the fabrication of multi probe electron
waveguides and rings.
67,68,69,70
Submicron trenches
71
are
still another way to dene the channel. For simple analyt-
ical estimates of lateral depletion widths in the shallow-
mesa geometry, see Ref.
72
.
A clever variant of the split-gate technique was intro-
duced by Ford et al.
73,74
A patterned layer of electron
beam resist (an organic insulator) is used as a gate di-
electric, in such a way that the separation between the
gate and the 2DEG is largest in those regions where a nar-
row conducting channel has to remain after application
of a negative gate voltage. As illustrated by the cross-
sectional view in Fig. 4d, in this way one can dene a ring
structure, for example, for use in an Aharonov-Bohm ex-
periment. A similar approach was developed by Smith
et al.
75
Instead of an organic resist they use a shallow-
mesa pattern in the heterostructure as a gate dielectric
of variable thickness. Initially, the latter technique was
used for capacitive studies of one- and zero-dimensional
connement.
75,76
More recently it was adopted for trans-
port measurements as well.
77
Still another variation of
this approach was developed by Hansen et al.,
78,79
pri-
marily for the study of one-dimensional subband struc-
ture using infrared spectroscopy. Instead of electron
beam lithography, they employ a photolithographic tech-
nique to dene a pattern in the insulator. An array with
a very large number of narrow lines is obtained by pro-
jecting the interference pattern of two laser beams onto
light-sensitive resist. This technique is known as holo-
graphic illumination (see Section II.G.2).
As two representative examples of state-of-the-art
nanostructures, we show in Fig. 5a a miniaturized Hall
bar,
67
fabricated by a shallow-mesa etch, and in Fig. 5b
a double-quantum-point contact device,
80
fabricated by
means of the split-gate technique.
Other techniques have been used as well to fabricate
narrow electron gas channels. We mention selective-area
ion implantation using focused ion beams,
81
masked ion
beam exposure,
82
strain-induced connement,
83
lateral
p-n junctions,
84,85
gates in the plane of the 2DEG,
86
and selective epitaxial growth.
87,88,89,90,91,92
For more de-
tailed and complete accounts of nanostructure fabrica-
tion techniques, we refer to Refs.
9
and
13,14,15
.
D. Basic properties
1. Density of states in two, one, and zero dimensions
The energy of conduction electrons in a single subband
of an unbounded 2DEG, relative to the bottom of that
7
FIG. 5 Scanning electron micrographs of nanostructures
in GaAs-AlGaAs heterostructures. (a, top) Narrow chan-
nel (width 75 nm), fabricated by means of the connement
scheme of Fig. 4c. The channel has side branches (at a 2-m
separation) that serve as voltage probes. Taken from M. L.
Roukes et al., Phys. Rev. Lett. 59, 3011 (1987). (b, bottom)
Double quantum point contact device, based on the conne-
ment scheme of Fig. 4b. The bar denotes a length of 1 m.
Taken from H. van Houten et al., Phys. Rev. B 39, 8556
(1989).
subband, is given by
E(k) = h
2
k
2
/2m, (1.2)
as a function of momentum hk. The eective mass m
is considerably smaller than the free electron mass m
e
(see Table I), as a result of interactions with the lattice
potential. (The incorporation of this potential into an
eective mass is an approximation
20
that is completely
justied for the present purposes.) The density of states
(E) dn(E)/dE is the derivative of the number of elec-
tronic states n(E) (per unit surface area) with energy
smaller than E. In k-space, these states are contained
within a circle of area A = 2mE/h
2
[according to Eq.
(1.2)], which contains a number g
s
g
v
A/(2)
2
of distinct
states. The factors g
s
and g
v
account for the spin degen-
eracy and valley degeneracy, respectively (Table I). One
thus nds that n(E) = g
s
g
v
mE/2h
2
, so the density of
states corresponding to a single subband in a 2DEG,
(E) = g
s
g
v
mE/2h
2
, (1.3)
is independent of the energy. As illustrated in Fig. 6a,
a sequence of subbands is associated with the set of dis-
crete levels in the potential well that connes the 2DEG
to the interface. At zero temperature, all states are
lled up to the Fermi energy E
F
(this remains a good
approximation at nite temperature if the thermal en-
ergy k
B
T E
F
). Because of the constant density of
states, the electron (sheet) density n
s
is linearly related
to E
F
by n
s
= E
F
g
s
g
v
m/2h
2
. The Fermi wave num-
ber k
F
= (2mE
F
/h
2
)
1/2
is thus related to the density
by k
F
= (4n
s
/g
s
g
v
)
1/2
. The second subband starts to
be populated when E
F
exceeds the energy of the second
band bottom. The stepwise increasing density of states
shown in Fig. 6a is referred to as quasi-two-dimensional.
As the number of occupied subbands increases, the den-
sity of states eventually approaches the

E dependence
characteristic for a three-dimensional system. Note, how-
ever, that usually only a single subband is occupied.
If the 2DEG is conned laterally to a narrow channel,
then Eq. (1.2) only represents the kinetic energy from the
free motion (with momentum hk) parallel to the channel
axis. Because of the lateral connement, a single two-
dimensional (2D) subband is split itself into a series of
one-dimensional (1D) subbands, with band bottoms at
E
n
, n = 1, 2, . . . The total energy E
n
(k) of an electron
in the nth 1D subband (relative to the bottom of the 2D
subband) is given by
E
n
(k) = E
n
+ h
2
k
2
/2m. (1.4)
Two frequently used potentials to model analytically
the lateral connement are the square-well potential (of
width W, illustrated in Fig. 6b) and the parabolic poten-
tial well (described by V (x) =
1
2
m
2
0
x
2
). The conne-
ment levels are then given either by E
n
= (nh)
2
/2mW
2
for the square well or by E
n
= (n
1
2
) h
0
for the
parabolic well. When one considers electron transport
through a narrow channel, it is useful to distinguish be-
tween states with positive and negative k, since these
states move in opposite directions along the channel. We
denote by
+
n
(E) the density of states with k > 0 per unit
channel length in the nth 1D subband. This quantity is
given by

+
n
(E) = g
s
g
v
_
2
dE
n
(k)
dk
_
1
= g
s
g
v
m
2h
2
_
h
2
2m(E E
n
)
_1/2
. (1.5)
The density of states

n
with k < 0 is identical to
+
n
.
(This identity holds because of time-reversal symmetry;
In a magnetic eld,
+
n
,=

n
, in general.) The total
density of states (E), drawn in Fig. 6b, is twice the
result (1.5) summed over all n for which E
n
< E. The
8
FIG. 6 Density of states (E) as a function of energy. (a)
Quasi-2D density of states, with only the lowest subband oc-
cupied (hatched). Inset: Connement potential perpendicu-
lar to the plane of the 2DEG. The discrete energy levels corre-
spond to the bottoms of the rst and second 2D subbands. (b)
Quasi-1D density of states, with four 1D subbands occupied.
Inset: Square-well lateral connement potential with discrete
energy levels indicating the 1D subband bottoms. (c) Density
of states for a 2DEG in a perpendicular magnetic eld. The
occupied 0D subbands or Landau levels are shown in black.
Impurity scattering may broaden the Landau levels, leading
to a nonzero density of states between the peaks.
density of states of a quasi-one-dimensional electron gas
with many occupied 1D subbands may be approximated
by the 2D result (1.3).
If a magnetic eld B is applied perpendicular to an
unbounded 2DEG, the energy spectrum of the electrons
becomes fully discrete, since free translational motion in
the plane of the 2DEG is impeded by the Lorentz force.
Quantization of the circular cyclotron motion leads to
energy levels at
93
E
n
= (n
1
2
) h
c
, (1.6)
with
c
= eB/m the cyclotron frequency. The quantum
number n = 1, 2, . . . labels the Landau levels. The num-
ber of states is the same in each Landau level and equal
to one state (for each spin and valley) per ux quantum
h/e through the sample. To the extent that broadening
of the Landau levels by disorder can be neglected, the
density of states (per unit area) can be approximated by
(E) = g
s
g
v
eB
h

n=1
(E E
n
), (1.7)
as illustrated in Fig. 6c. The spin degeneracy contained
in Eq. (1.7) is resolved in strong magnetic elds as a re-
sult of the Zeeman splitting g
B
B of the Landau levels
(
B
eh/2m
e
denotes the Bohr magneton; the Lande
g-factor is a complicated function of the magnetic eld
in these systems).
20
Again, if a large number of Landau
levels is occupied (i.e., at weak magnetic elds), one re-
covers approximately the 2D result (1.3). The foregoing
considerations are for an unbounded 2DEG. A magnetic
eld perpendicular to a narrow 2DEG channel causes the
density of states to evolve gradually from the 1D form of
Fig. 6b to the eectively 0D form of Fig. 6c. This tran-
sition is discussed in Section II.F.
2. Drude conductivity, Einstein relation, and Landauer formula
In the presence of an electric eld E in the plane of
the 2DEG, an electron acquires a drift velocity v =
eEt/m in the time t since the last impurity col-
lision. The average of t is the scattering time , so the
average drift velocity v
drift
is given by
v
drift
=
e
E,
e
= e/m. (1.8)
The electron mobility
e
together with the sheet den-
sity n
s
determine the conductivity in the relation
en
s
v
drift
= E. The result is the familiar Drude
conductivity,
94
which can be written in several equiva-
lent forms:
= en
s

e
=
e
2
n
s

m
= g
s
g
v
e
2
h
k
F
l
2
. (1.9)
In the last equality we have used the identity n
s
=
g
s
g
v
k
2
F
/4 (see Section I.D.1) and have dened the mean
9
free path l = v
F
. The dimensionless quantity k
F
l is
much greater than unity in metallic systems (see Table I
for typical values in a 2DEG), so the conductivity is large
compared with the quantum unit e
2
/h (26 k)
1
.
From the preceding discussion it is obvious that the
current induced by the applied electric eld is carried by
all conduction electrons, since each electron acquires the
same average drift velocity. Nonetheless, to determine
the conductivity it is sucient to consider the response
of electrons near the Fermi level to the electric eld. The
reason is that the states that are more than a few times
the thermal energy k
B
T below E
F
are all lled so that in
response to a weak electric eld only the distribution of
electrons among states at energies close to E
F
is changed
from the equilibrium Fermi-Dirac distribution
f(E E
F
) =
_
1 + exp
E E
F
k
B
T
_
1
. (1.10)
The Einstein relation
94
= e
2
(E
F
)D (1.11)
is one relation between the conductivity and Fermi level
properties (in this case the density of states (E) and
the diusion constant D, both evaluated at E
F
). The
Landauer formula
4
[Eq. (1.22)] is another such relation
(in terms of the transmission probability at the Fermi
level rather than in terms of the diusion constant).
The Einstein relation (1.11) for an electron gas at zero
temperature follows on requiring that the sum of the drift
current density E/e and the diusion current density
Dn
s
vanishes in thermodynamic equilibrium, charac-
terized by a spatially constant electrochemical potential
:
E/e Dn
s
= 0, when = 0. (1.12)
The electrochemical potential is the sum of the electro-
static potential energy eV (which determines the en-
ergy of the bottom of the conduction band) and the chem-
ical potential E
F
(being the Fermi energy relative to the
conduction band bottom). Since (at zero temperature)
dE
F
/dn
s
= 1/(E
F
), one has
= eE+(E
F
)
1
n
s
. (1.13)
The combination of Eqs. (1.12) and (1.13) yields the Ein-
stein relation (1.11) between and D. To verify that Eq.
(1.11) is consistent with the earlier expression (1.9) for
the Drude conductivity, one can use the result (see be-
low) for the 2D diusion constant:
D =
1
2
v
2
F
=
1
2
v
F
l, (1.14)
in combination with Eq. (1.3) for the 2D density of states.
At a nite temperature T, a chemical potential (or
Fermi energy) gradient E
F
induces a diusion current
that is smeared out over an energy range of order k
B
T
around E
F
. The energy interval between E and E +dE
contributes to the diusion current density j an amount
dj given by
dj
di
= D(E)f(E E
F
)dE
= dED(E)
df
dE
F
E
F
, (1.15)
where the diusion constant D is to be evaluated at en-
ergy E. The total diusion current density follows on
integration over E:
j = E
F
e
2
_

0
dE (E, 0)
df
dE
F
, (1.16)
with (E, 0) the conductivity (1.11) at temperature zero
for a Fermi energy equal to E. The requirement of van-
ishing current for a spatially constant electrochemical po-
tential implies that the conductivity (E
F
, T) at temper-
ature T and Fermi energy E
F
satises
(E
F
, T)e
2
E
F
+j = 0.
Therefore, the nite-temperature conductivity is given
simply by the energy average of the zero-temperature
result
(E
F
, T) =
_

0
dE (E, 0)
df
dE
F
. (1.17)
As T 0, df/dE
F
(EE
F
), so indeed only E = E
F
contributes to the energy average. Result (1.17) contains
exclusively the eects of a nite temperature that are due
to the thermal smearing of the Fermi-Dirac distribution.
A possible temperature dependence of the scattering pro-
cesses is not taken into account.
We now want to discuss one convenient way to calcu-
late the diusion constant (and hence obtain the conduc-
tivity). Consider the diusion current density j
x
due to
a small constant density gradient, n(x) = n
0
+ cx. We
write
j
x
= lim
t
v
x
(t = 0)n(x(t = t)))
= lim
t
cv
x
(0)x(t))
= lim
t
c
_
t
0
dtv
x
(0)v
x
(t)), (1.18)
where t is time and the brackets ) denote an isotropic
angular average over the Fermi surface. The time interval
t , so the velocity of the electron at time 0 is
uncorrelated with its velocity at the earlier time t.
This allows us to neglect at x(t) the small deviations
from an isotropic velocity distribution induced by the
density gradient [which could not have been neglected
at x(0)]. Since only the time dierence matters in the
velocity correlation function, one has v
x
(0)v
x
(t)) =
v
x
(t)v
x
(0)). We thus obtain for the diusion constant
D = j
x
/c the familiar linear response formula
95
D =
_

0
dtv
x
(t)v
x
(0)). (1.19)
10
Since, in the semiclassical relaxation time approximation,
each scattering event is assumed to destroy all correla-
tions in the velocity, and since a fraction exp(t/) of
the electrons has not been scattered in a time t, one has
(in 2D)
v
x
(t)v
x
(0)) = v
x
(0)
2
)e
t/
=
1
2
v
2
F
e
t/
. (1.20)
Substituting this correlation function for the integrand
in Eq. (1.19), one recovers on integration the diusion
constant (1.14).
The Drude conductivity (4.8) is a semiclassical result,
in the sense that while the quantum mechanical Fermi-
Dirac statistic is taken into account, the dynamics of the
electrons at the Fermi level is assumed to be classical.
In Section II we will discuss corrections to this result
that follow from correlations in the diusion process due
to quantum interference. Whereas for classical diusion
correlations disappear on the time scale of the scattering
time [as expressed by the correlation function (1.20)],
in quantum diusion correlations persist up to times of
the order of the phase coherence time. The latter time

is associated with inelastic scattering and at low tem-

peratures can become much greater than the time as-
sociated with elastic scattering.
In an experiment one measures a conductance rather
than a conductivity. The conductivity relates the local
current density to the electric eld, j = E, while the
conductance G relates the total current to the voltage
drop, I = GV . For a large homogeneous conductor the
dierence between the two is not essential, since Ohms
law tells us that
G = (W/L) (1.21)
for a 2DEG of width W and length L in the current
direction. (Note that G and have the same units in
two dimensions.) If for the moment we disregard the ef-
fects of phase coherence, then the simple scaling (1.21)
holds provided both W and L are much larger than the
mean free path l. This is the diusive transport regime,
illustrated in Fig. 7a. When the dimensions of the sam-
ple are reduced below the mean free path, one enters
the ballistic transport regime, shown in Fig. 7c. One can
further distinguish an intermediate quasi-ballistic regime,
characterized by W < l < L (see Fig. 7b). In ballistic
transport only the conductance plays a role, not the con-
ductivity. The Landauer formula
G = (e
2
/h)T (1.22)
plays a central role in the study of ballistic transport be-
cause it expresses the conductance in terms of a Fermi
level property of the sample (the transmission probabil-
ity T, see Section III.A). Equation (1.22) can therefore
be applied to situations where the conductivity does not
exist as a local quantity, as we will discuss in Sections III
and IV.
If phase coherence is taken into account, then the mini-
mal length scale required to characterize the conductivity
FIG. 7 Electron trajectories characteristic for the diu-
sive (l < W, L), quasi-ballistic (W < l < L), and ballistic
(W, L < l) transport regimes, for the case of specular bound-
ary scattering. Boundary scattering and internal impurity
scattering (asterisks) are of equal importance in the quasi-
ballistic regime. A nonzero resistance in the ballistic regime
results from back scattering at the connection between the
narrow channel and the wide 2DEG regions. Taken from
H. van Houten et al., in Physics and Technology of Submi-
cron Structures (H. Heinrich, G. Bauer, and F. Kuchar, eds.).
Springer, Berlin, 1988.
becomes larger. Instead of the (elastic) mean free path
l v
F
, the phase coherence length l

(D

)
1/2
be-
comes this characteristic length scale (up to a numerical
coecient l

equals the average distance that an electron

diuses in the time

). Ohms law can now only be ap-

plied to add the conductances of parts of the sample with
dimensions greater than l

. Since at low temperatures l

can become quite large (cf. Table I), it becomes possible

that (for a small conductor) phase coherence extends over
a large part of the sample. Then only the conductance
(not the conductivity) plays a role, even if the transport
is fully in the diusive regime. We will encounter such
situations repeatedly in Section II.
3. Magnetotransport
In a magnetic eld B perpendicular to the 2DEG, the
current is no longer in the direction of the electric eld
due to the Lorentz force. Consequently, the conductiv-
ity is no longer a scalar but a tensor , related via the
Einstein relation = e
2
(E
F
)D to the diusion tensor
D =
_

0
dtv(t)v(0)). (1.23)
11
Equation (1.23) follows from a straightforward general-
ization of the argument leading to the scalar relation
(1.19) [but now the ordering of v(t) and v(0) matters].
Between scattering events the electrons at the Fermi
level execute circular orbits, with cyclotron frequency

c
= eB/m and cyclotron radius l
cycl
= mv
F
/eB. Tak-
ing the 2DEG in the xy plane, and the magnetic eld in
the positive z-direction, one can write in complex number
notation
v(t) v
x
(t) +iv
y
(t) = v
F
exp(i +i
c
t). (1.24)
The diusion tensor is obtained from
D
xx
+iD
yx
=
_
2
0
d
2
_

0
dt v(t)v
F
cos e
t/
=
D
1 + (
c
)
2
(1 +i
c
), (1.25)
where D is the zero-eld diusion constant (1.14). One
easily veries that D
yy
= D
xx
and D
xy
= D
yx
. From
the Einstein relation one then obtains the conductivity
tensor
=

1 + (
c
)
2
_
1
c

c
1
_
, (1.26)
with the zero-eld conductivity (1.9). The resistivity
tensor
1
has the form
=
_
1
c

c
1
_
, (1.27)
with =
1
= m/n
s
e
2
the zero-eld resistivity.
The o-diagonal element
xy
R
H
is the classical Hall
resistance of a 2DEG:
R
H
=
B
n
s
e
=
1
g
s
g
v
h
e
2
h
c
E
F
. (1.28)
Note that in a 2D channel geometry there is no distinc-
tion between the Hall resistivity and the Hall resistance,
since the ratio of the Hall voltage V
H
= WE
x
across the
channel to the current I = Wj
y
along the channel does
not depend on its length and width (provided transport
remains in the diusive regime). The diagonal element

xx
is referred to as the longitudinal resistivity. Equa-
tion (1.27) tells us that classically the magnetoresistivity
is zero (i.e.,
xx
(B)
xx
(0) = 0). This counterintuitive
result can be understood by considering that the force
from the Hall voltage cancels the average Lorentz force
on the electrons. A general conclusion that one can draw
from Eqs. (1.26) and (1.27) is that the classical eects of
a magnetic eld are important only if
c

>

1. In such
elds an electron can complete several cyclotron orbits
before being scattered out of orbit. In a high-mobility
2DEG this criterion is met at rather weak magnetic elds
(note that
c
=
e
B, and see Table I).
In the foregoing application of the Einstein relation we
have used the zero-eld density of states. Moreover, we
FIG. 8 Schematic dependence on the reciprocal lling factor

1
2eB/hns of the longitudinal resistivity xx (normal-
ized to the zero-eld resistivity ) and of the Hall resistance
RH xy (normalized to h/2e
2
). The plot is for the case
of a single valley with twofold spin degeneracy. Deviations
from the semiclassical result (1.27) occur in strong magnetic
elds, in the form of Shubnikov-De Haas oscillations in xx
and quantized plateaus [Eq. (1.31)] in xy.
have assumed that the scattering time is B-independent.
Both assumptions are justied in weak magnetic elds,
for which E
F
/h
c
1, but not in stronger elds (cf.
Table I). As illustrated in Fig. 8, deviations from the
semiclassical result (1.27) appear as the magnetic eld is
increased. These deviations take the form of an oscilla-
tory magnetoresistivity (the Shubnikov-De Haas eect)
and plateaux in the Hall resistance (the quantum Hall
eect). The origin of these two phenomena is the forma-
tion of Landau levels by a magnetic eld, discussed in
Section I.D.1, that leads to the B-dependent density of
states (1.7). The main eect is on the scattering rate
1
,
which in a simple (Born) approximation
96
is proportional
to (E
F
):

1
= (/h)(E
F
)c
i
u
2
. (1.29)
Here c
i
is the areal density of impurities, and the im-
purity potential is modeled by a 2D delta function of
strength u. The diagonal element of the resistivity ten-
sor (1.27) is
xx
= (m/e
2
n
s
)
1
(E
F
). Oscillations in
the density of states at the Fermi level due to the Landau
level quantization are therefore observable as an oscilla-
tory magnetoresistivity. One expects the resistivity to be
minimal when the Fermi level lies between two Landau
levels, where the density of states is smallest. In view of
Eq. (1.7), this occurs when the Landau level lling fac-
tor (n
s
/g
s
g
v
)(h/eB) equals an integer N = 1, 2, . . .
(assuming spin-degenerate Landau levels). The resulting
Shubnikov-De Haas oscillations are periodic in 1/B, with
12
spacing (1/B) given by
(
1
B
) =
e
h
g
s
g
v
n
s
, (1.30)
providing a means to determine the electron density from
a magnetoresistance measurement. This brief explana-
tion of the Shubnikov-De Haas eect needs renement,
20
but is basically correct. The quantum Hall eect,
8
being
the occurrence of plateaux in R
H
versus B at precisely
R
H
=
1
g
s
g
v
h
e
2
1
N
, N = 1, 2, . . . , (1.31)
is a more subtle eect
97
to which we cannot do justice
in a few lines (see Section IV.A). The quantization of
the Hall resistance is related on a fundamental level to
the quantization in zero magnetic eld of the resistance
of a ballistic point contact.
6,7
We will present a unied
description of both these eects in Sections III.A and
III.B.
II. DIFFUSIVE AND QUASI-BALLISTIC TRANSPORT
A. Classical size eects
In metals, the dependence of the resistivity on the
size of the sample has been the subject of study for
almost a century.
98
Because of the small Fermi wave
length in a metal, these are classical size eects. Com-
prehensive reviews of this eld have been given by
Chambers,
99
Br andli and Olsen,
100
Sondheimer,
101
and,
recently, Pippard.
102
In semiconductor nanostructures
both classical and quantum size eects appear, and an
understanding of the former is necessary to distinguish
them from the latter. Classical size eects in a 2DEG are
of intrinsic interest as well. First of all, a 2DEG is an ideal
model system to study known size eects without the
complications of nonspherical Fermi surfaces and poly-
crystallinity, characteristic for metals. Furthermore, it is
possible in a 2DEG to study the case of nearly complete
specular boundary scattering, whereas in a metal diuse
scattering dominates. The much smaller cyclotron radius
in a 2DEG, compared with a metal at the same magnetic
eld value, allows one to enter the regime where the cy-
clotron radius is comparable to the range of the scattering
potential. The resulting modications of known eects in
the quasi-ballistic transport regime are the subject of this
section. A variety of new classical size eects, not known
from metals, appear in the ballistic regime, when the re-
sistance is measured on a length scale below the mean
free path. These are discussed in Section III.E, and re-
quire a reconsideration of what is meant by a resistance
on such a short length scale.
In the present section we assume that the channel
length L (or, more generally, the separation between the
voltage probes) is much larger than the mean free path l
for impurity scattering so that the motion remains diu-
sive along the channel. Size eects in the resistivity oc-
cur when the motion across the channel becomes ballistic
(i.e., when the channel width W < l). Diuse bound-
ary scattering leads to an increase in the resistivity in a
zero magnetic eld and to a nonmonotonic magnetore-
sistivity in a perpendicular magnetic eld, as discussed
in the following two subsections. The 2D channel ge-
ometry is essentially equivalent to the 3D geometry of
a thin metal plate in a parallel magnetic eld, with the
current owing perpendicular to the eld. Size eects in
this geometry were originally studied by Fuchs
103
in a
zero magnetic eld and by MacDonald
104
for a nonzero
eld. The alternative conguration in which the mag-
netic eld is perpendicular to the thin plate, studied by
Sondheimer
105
does not have a 2D analog. We discuss
in this section only the classical size eects, and thus the
discreteness of the 1D subbands and of the Landau lev-
els is ignored. Quantum size eects in the quasi-ballistic
transport regime are treated in Section II.F.
1. Boundary scattering
In a zero magnetic eld, scattering at the channel
boundaries increases the resistivity, unless the scattering
is specular. Specular scattering occurs if the conning
potential V (x, y) does not depend on the coordinate y
along the channel axis. In that case the electron motion
along the channel is not inuenced at all by the lateral
connement, so the resistivity retains its 2D bulk value

0
= m/e
2
n
s
. More generally, specular scattering re-
quires any roughness of the boundaries to be on a length
scale smaller than the Fermi wavelength
F
. The con-
ning potential created electrostatically by means of a
gate electrode is known to cause predominantly specu-
lar scattering (as has been demonstrated by the electron
focusing experiments
59
discussed in Section III.C). This
is a unique situation, not previously encountered in met-
als, where as a result of the small
F
(on the order of
the interatomic separation) diuse boundary scattering
dominates.
102
Diuse scattering means that the velocity distribution
at the boundary is isotropic for velocity directions that
point away from the boundary. Note that this implies
that an incident electron is reected with a (normal-
ized) angular distribution P() =
1
2
cos , since the re-
ection probability is proportional to the ux normal to
the boundary. Diuse scattering increases the resistivity
above
0
by providing an upper bound W to the eective
mean free path. In order of magnitude, (l/W)
0
if
l
>

W (a more precise expression is derived later). In

general, boundary scattering is neither fully specular nor
fully diuse and, moreover, depends on the angle of inci-
dence (grazing incidence favors specular scattering since
the momentum along the channel is large and not eas-
ily reversed). The angular dependence is often ignored
for simplicity, and the boundary scattering is described,
following Fuchs,
103
by a single parameter p, such that
an electron colliding with the boundary is reected spec-
ularly with probability p and diusely with probability
13
1 p. This specularity parameter is then used as a t
parameter in comparison with experiments. Soer
106
has
developed a more accurate, and more complicated, mod-
eling in terms of an angle of incidence dependent specu-
larity parameter.
In the extreme case of fully diuse boundary scatter-
ing (p = 0), one is justied in neglecting the dependence
of the scattering probability on the angle of incidence.
We treat this case here in some detail to contrast it with
fully specular scattering, and because diuse scattering
can be of importance in 2DEG channels dened by ion
beam exposure rather than by gates.
107,108
We calculate
the resistivity from the diusion constant by means of
the Einstein relation. Fuchs takes the alternative (but
equivalent) approach of calculating the resistivity from
the linear response to an applied electric eld.
103
Im-
purity scattering is taken as isotropic and elastic and is
described by a scattering time such that an electron
is scattered in a time interval dt with probability dt/,
regardless of its position and velocity, This is the com-
monly employed scattering time (or relaxation time)
approximation.
The channel geometry is dened by hard walls at
x = W/2 at which the electrons are scattered dif-
fusely. The stationary electron distribution function at
the Fermi energy F(r, ) satises the Boltzmann equa-
tion
v

r
F =
1

F +
1

_
2
0
d
2
F, (2.1)
where r (x, y) is the position and is the angle that
the velocity v v
F
(cos , sin ) makes with the x-axis.
The boundary condition corresponding to diuse scat-
tering is that F is independent of the velocity direction
for velocities pointing away from the boundary. In view
of current conservation this boundary condition can be
written as
F(r, ) =
1
2
_
/2
/2
d

F(r,

) cos

,
for x =
W
2
,

2
< <
3
2
,
=
1
2
_
3/2
/2
d

F(r,

) cos

,
for x =
W
2
,

2
< <

2
. (2.2)
To determine the diusion constant, we look for a so-
lution of Eqs. (2.1) and (2.2) corresponding to a con-
stant density gradient along the channel, F(r, ) =
cy +f(x, ). Since there is no magnetic eld, we antic-
ipate that the density will be uniform across the channel
width so that
_
2
0
fd = 0. The Boltzmann equation
(2.1) then simplies to an ordinary dierential equation
for f, which can be solved straightforwardly. The solu-
tion that satises the boundary conditions (2.2) is
F(r, ) = cy+cl sin
_
1 exp
_

W
2l[ cos [

x
l cos
__
,
(2.3)
where we have written l v
F
. One easily veries that
F has indeed a uniform density along x. The diusion
current
I
y
= v
F
_
W/2
W/2
dx
_
2
0
dF sin (2.4)
along the channel in response to the density gradient
n/y = 2c determines the diusion constant D =
(I
y
/W)(n/y)
1
. The resistivity = E
F
/n
s
e
2
D then
follows from the Einstein relation (1.11), with the 2D
density of states n
s
/E
F
. The resulting expression is
=
0
_
1
4l
W
_
1
0
d (1
2
)
1/2
(1 e
W/l
)
_1
,
(2.5)
which can be easily evaluated numerically. It is worth
noting that the above result
109
for /
0
in a 2D channel
geometry does not dier much (less than 20%) from the
corresponding result
103
in a 3D thin lm.
For l/W 1 one has
=
0
_
1 +
4
3
l
W
_
, (2.6)
which diers from Eq. (2.5) by less than 10% in the range
l/W
<

10. For l/W 1 one has asymptotically

=

2

0
l
W
1
ln(l/W)
=

2
mv
F
n
s
e
2
W
1
ln(l/W)
. (2.7)
In the absence of impurity scattering (i.e., in the limit
l ), Eq. (2.7) predicts a vanishing resistivity. Dif-
fuse boundary scattering is ineective in establishing a
nite resistivity in this limit, because electrons with ve-
locities nearly parallel to the channel walls can propagate
over large distances without collisions and thereby short
out the current. As shown by Tesanovic et al.,
110
a small
but nonzero resistivity in the absence of impurity scat-
tering is recovered if one goes beyond the semiclassical
approximation and includes the eect of the quantum
mechanical uncertainty in the transverse component of
the electron velocity.
2. Magneto size eects
In an unbounded 2DEG, the longitudinal resistivity
is magnetic-eld independent in the semiclassical ap-
proximation (see Section I.D.3). We will discuss how a
nonzero magnetoresistivity can arise classically as a re-
sult of boundary scattering. We consider the two extreme
14
cases of specular and diuse boundary scattering, and
describe the impurity scattering in the scattering time
approximation. Shortcomings of this approximation are
discussed toward the end of this subsection.
We consider rst the case of specular boundary scat-
tering. In a zero magnetic eld it is obvious that specular
scattering cannot aect the resistivity, since the projec-
tion of the electron motion on the channel axis is not
changed by the presence of the channel boundaries. If
a magnetic eld is applied perpendicular to the 2DEG,
the electron trajectories in a channel cannot be mapped
in this way on the trajectories in an unbounded system.
In fact, in an unbounded 2DEG in equilibrium the elec-
trons perform closed cyclotron orbits between scattering
events, whereas a channel geometry supports open or-
bits that skip along the boundaries. One might suppose
that the presence of these skipping orbits propagating
along the channel would increase the diusion constant
and hence reduce the (longitudinal) resistivity below the
value
0
of a bulk 2DEG. That is not correct, at least
in the scattering time approximation, as we now demon-
strate.
The stationary Boltzmann equation in a magnetic eld
B in the z-direction (perpendicular to the 2DEG) is
v

r
F +
c

F =
1

F +
1

_
2
0
d
2
F. (2.8)
Here, we have used the identity em
1
(v B) /v

c
/ (with
c
eB/m the cyclotron frequency) to
rewrite the term that accounts for the Lorentz force. The
distribution function F(r, ) must satisfy the boundary
conditions for specular scattering,
F(r, ) = F(r, ), for x = W/2. (2.9)
One readily veries that
F(r, ) = c(y +
c
x) +cl sin (2.10)
is a solution of Eqs. (2.8) and (2.9). The correspond-
ing diusion current I
y
= cWv
F
l and density gradi-
ent along the channel n/y = 2c are both the same
as in a zero magnetic eld. It follows that the diu-
sion constant D = I
y
/2cW and, hence, the longitudi-
nal resistivity = E
F
/n
s
e
2
D are B-independent; that is,
=
0
m/n
s
e
2
, as in an unbounded 2DEG. More gen-
erally, one can show that in the scattering time approx-
imation the longitudinal resistivity is B-independent for
any conning potential V (x, y) that does not vary with
the coordinate y along the channel axis. (This state-
ment is proven by applying the result of Ref.
111
, of a
B-independent
yy
for periodic V (x), to a set of disjunct
parallel channels (see Section II.G.2); the case of a single
channel then follows from Ohms law.)
In the case of diuse boundary scattering, the zero-
eld resistivity is enhanced by approximately a factor
1 + l/2W [see Eq. (2.6)]. A suciently strong magnetic
eld suppresses this enhancement, and reduces the re-
sistivity to its bulk value
0
. The mechanism for this
FIG. 9 Illustration of the eect of a magnetic eld on mo-
tion through a channel with diuse boundary scattering. (a)
Electrons which in a zero eld move nearly parallel to the
boundary can reverse their motion in weak magnetic elds.
This increases the resistivity. (b) Suppression of back scat-
tering at the boundaries in strong magnetic elds reduces the
resistivity.
negative magnetoresistance is illustrated in Fig. 9b. If
the cyclotron diameter 2l
cycl
is smaller than the chan-
nel width W, diuse boundary scattering cannot reverse
the direction of motion along the channel, as it could for
smaller magnetic elds. The diusion current is there-
fore approximately the same as in the case of specular
scattering, in which case we have seen that the diusion
constant and, hence, resistivity have their bulk values.
Figure 9 represents an example of magnetic reduction of
backscattering. Recently, this phenomenon has been un-
derstood to occur in an extreme form in the quantum
Hall eect
112
and in ballistic transport through quan-
tum point contacts.
113
The eect was essentially known
and understood by MacDonald
104
in 1949 in the course
of his magnetoresistivity experiments on sodium wires.
The ultimate reduction of the resistivity is preceded by
an initial increase in weak magnetic elds, due to the de-
ection toward the boundary of electrons with a velocity
nearly parallel to the channel axis (Fig. 9a). The re-
sulting nonmonotonic B-dependence of the resistivity is
shown in Fig. 10. The plot for diuse scattering is based
on a calculation by Ditlefsen and Lothe
114
for a 3D thin-
lm geometry. The case of a 2D channel has been studied
by Pippard
102
in the limit l/W , and he nds that
the 2D and 3D geometries give very similar results.
An experimental study of this eect in a 2DEG has
been performed by Thornton et al.
107
In Fig. 11 their
magnetoresistance data are reproduced for channels of
dierent widths W, dened by low-energy ion beam ex-
posure. It was found that the resistance reaches a maxi-
mum when W 0.5 l
cycl
, in excellent agreement with the
15
FIG. 10 Magnetic eld dependence of the longitudinal resis-
tivity of a channel for the two cases of diuse and specular
boundary scattering, obtained from the Boltzmann equation
in the scattering time approximation. The plot for diuse
scattering is the result of Ref.
114
for a 3D thin lm geometry
with l = 10W. (A 2D channel geometry is expected to give
very similar results.
102
)
FIG. 11 Experimental magnetic eld dependence of the re-
sistance of channels of dierent widths, dened by ion beam
exposure in the 2DEG of a GaAs-AlGaAs heterostructure
(L = 12 m, T = 4.2 K). The nonmonotonic magnetic eld
dependence below 1 T is a classical size eect due to diuse
boundary scattering, as illustrated in Fig. 9. The magne-
toresistance oscillations at higher elds result from the quan-
tum mechanical Shubnikov-De Haas eect. Taken from T. J.
Thornton et al., Phys. Rev. Lett. 63, 2128 (1989).
theoretical predictions.
102,114
Thornton et al. also inves-
tigated channels dened electrostatically by a split gate,
for which one expects predominantly specular boundary
scattering.
59
The foregoing analysis would then predict
an approximately B-independent resistance (Fig. 10),
and indeed only a small resistance maximum was ob-
served in weak magnetic elds. At stronger elds, how-
ever, the resistance was found to decrease substantially.
Such a monotonically decreasing resistance in channels
FIG. 12 Electron trajectories in a channel with specular
boundary scattering, to illustrate how a magnetic eld can
suppress the back scattering by an isolated impurity close to
a boundary. This eect would lead to a negative magneto
resistivity if one would go beyond the scattering time approx-
imation.
with predominantly specular boundary scattering was
rst reported by Choi et al.,
55
and studied for a nar-
rower channel in Ref.
27
(see Section II.E.2 for some of
these experimental results). We surmise that a classi-
cal negative magnetoresistance in the case of specular
boundary scattering can result if the cyclotron radius be-
comes smaller than some characteristic correlation length
in the distribution of impurities (and in the resulting po-
tential landscape). Correlations between the positions
of impurities and the channel boundaries, which are ne-
glected in the scattering time approximation, will then
play a role. For an example, see Fig. 12, which shows
how an isolated impurity near the boundary can reverse
the direction of electron motion in a zero magnetic eld
but not in a suciently strong magnetic eld. In met-
als, where the cyclotron radius is much larger than in a
2DEG, an electron will eectively experience a random
impurity potential between subsequent boundary colli-
sions, so the scattering can well be described in terms of
an average relaxation time. The experiments in a 2DEG
suggest that this approximation breaks down at relatively
weak magnetic elds.
B. Weak localization
The temperature dependence of the Drude resistivity
= m/n
s
e
2
is contained in that of the scattering time
, since the electron density is constant in a degenerate
electron gas. As one lowers the temperature, inelastic
scattering processes (such as electron-phonon scattering)
are suppressed, leading to a decrease in the resistivity.
16
FIG. 13 Temperature dependence of the resistivity of a wide
2DEG in a GaAs-AlGaAs heterostructure (circles) and of
two narrow channels of lithographic width W
lith
= 1.5 m
(squares) and W
lith
= 0.5 m (triangles). The channel length
L = 10 m. The resistivity is estimated from the measured
resistance R by multiplying by W
lith
/L, disregarding the dif-
ference between the conducting and lithographic width in the
narrow channels. Taken from H. van Houten et al., Appl.
Phys. Lett. 49, 1781 (1986).
The residual resistivity is due entirely to elastic scattering
(with stationary impurities or other crystalline defects)
and is temperature-independent in the semiclassical the-
ory. Experimentally, however, one nds that below a
certain temperature the resistivity of the 2DEG starts to
rise again. The increase is very small in broad samples,
but becomes quite pronounced in narrow channels. This
is illustrated in Fig. 13, where the temperature dependen-
cies of the resistivities of wide and narrow GaAs-AlGaAs
heterostructures are compared.
63
The anomalous resistivity increase is due to long-range
correlations in the diusive motion of an electron that are
purely quantum mechanical. In the semiclassical theory
it is assumed that a few scattering events randomize the
electron velocity, so the velocity correlation function de-
cays exponentially in time with decay time [see Eq.
(1.20)]. As discussed in Section I.D.3, this assumption
leads to the Drude formula for the resistivity. It is only in
recent years that one has come to appreciate that purely
elastic scattering is not eective in destroying correla-
tions in the phase of the electron wave function. Such
correlations lead to quantum interference corrections to
the Drude result, which can explain the anomalous in-
crease in the resistivity at low temperatures.
A striking eect of quantum interference is to enhance
the probability for backscattering in a disordered sys-
tem in the metallic regime. This eect has been in-
terpreted as a precursor of localization in strongly dis-
ordered systems and has thus become known as weak
localization.
115,116,117
In Section II.B.1 we describe the
theory for weak localization in a zero magnetic eld.
The application of a magnetic eld perpendicular to the
2DEG suppresses weak localization,
118
as discussed in
Section II.B.2. The resulting negative magnetoresistiv-
ity is the most convenient way to resolve experimentally
the weak localization correction.
119
The theory for a nar-
row channel in the quasiballistic transport regime
109,120
diers in an interesting way from the theory for the diu-
sive regime,
121
as a consequence of the ux cancellation
eect.
122
The diusive and quasi-ballistic regimes are the
subjects of Sections II.B.2 and II.B.3, respectively.
1. Coherent backscattering
The theory of weak localization was developed by An-
derson et al.
116
and Gorkov et al.
117
This is a diagram-
matic perturbation theory that does not lend itself easily
to a physical interpretation. The interpretation of weak
localization as coherent backscattering was put forward
by Bergmann
123
and by Khmelnitskii and Larkin,
124,125
and formed the basis of the path integral theory of
Chakravarty and Schmid.
126
In this description, weak lo-
calization is understood by considering the interference
of the probability amplitudes for the classical trajectories
(or Feynman paths) from one point to another, as dis-
cussed later. For reviews of the alternative diagrammatic
approach, we refer to Refs.
127
and
128
.
In a Feynman path description
129
of diusion, the
probability P(r, r

, t) for motion from point r to point

r

in a time t consists of the absolute value squared of

the sum of probability amplitudes A
i
, one for each tra-
jectory from r to r

of duration t:
P(r, r

, t) =

i
A
i

2
=

i
[A
i
[
2
+

i=j
A
i
A

j
. (2.11)
The restriction to classical trajectories in the sum over
Feynman paths is allowed if the separation between scat-
tering events is much larger than the wavelength (i.e.,
if k
F
l 1). The classical diusion probability corre-
sponds to the rst term on the right-hand side of Eq.
(2.11), while the second term accounts for quantum in-
terference. In the diusive transport regime there is a
very large number of dierent trajectories that contribute
to the sum. One might suppose that for this reason the
interference term averages out, because dierent trajec-
tories have uncorrelated phases. This is correct if the
beginning and end points r and r

are dierent (Fig.

14a), but not if the two coincide (Fig. 14b). In the latter
case of backscattered trajectories, one can group the
contributions to the sum (2.11) in time-reversed pairs.
Time-reversal invariance guarantees that the probability
amplitudes A
+
and A

for clockwise and counterclock-

wise propagation around the closed loop are identical:
A
+
= A

A. The coherent backscattering probabil-

ity [A
+
+A

[
2
= 4[A[
2
is then twice the classical result.
The enhanced probability for return to the point of de-
parture reduces the diusion constant and, hence, the
conductivity. This is the essence of weak localization.
As phrased by Chakravarty and Schmid,
126
it is one of
those unique cases where the superposition principle of
17
FIG. 14 Mechanism of coherent back scattering. The prob-
ability amplitudes Ai and Aj of two trajectories from r to r

have uncorrelated phases in general (a), but the amplitudes

A
+
and A

of two time-reversed returning trajectories are

equal (b). The constructive interference of A
+
and A

in-
creases the probability for return to the point of departure,
which is the origin of the weak localization eect. The vol-
ume indicated in black is the area FvFdt covered by a ux
tube in a time interval dt, which enters in Eq. (2.12) for the
conductivity correction.
quantum mechanics leads to observable consequences at
the macroscopic level.
The magnitude of the weak localization correction

loc
to the Drude conductivity is proportional to the
probability for return to the point of departure.
126
Since

loc
is assumed to be a small correction, one can esti-
mate this probability from classical diusion. Let C(t)dr
denote the classical probability that an electron returns
after a time t to within dr of its point of departure. The
weak localization correction is given by the time integral
of the return probability:

loc

=
2 h
m
_

0
dt C(t)e
t/

. (2.12)
The correction is negative because the conductivity is
reduced by coherent backscattering. The factor h/m

F
v
F
follows in the path integral formalism from the area
covered by a ux tube of width
F
and length v
F
dt (see
Fig. 14b). The factor exp(t/

) is inserted by hand to
account for the loss of phase coherence after a time

(as
a result of inelastic scattering). The return probability
C(t) in a 2D channel of width W is given for times t
in the diusive regime by
C(t) = (4Dt)
1
, if t W
2
/D, (2.13a)
C(t) = W
1
(4Dt)
1/2
, if t W
2
/D.(2.13b)
The 1/t decay of the return probability (2.13a) assumes
unbounded diusion in two dimensions. A crossover to
a lower 1/

t decay (2.13b) occurs when the root-mean-

square displacement (2Dt)
1/2
exceeds the channel width,
so diusion occurs eectively in one dimension only. Be-
cause the time integral of C(t) itself diverges, the weak
localization correction (2.12) is determined by the behav-
ior of the return probability on the phase coherence time

, which provides a long-time cuto. One speaks of 2D

or 1D weak localization, depending on whether the return
probability C(

) on the time scale of

is determined by
2D diusion (2.13a) or by 1D diusion (2.13b). In terms
of the phase coherence length l

(D

)
1/2
, the crite-
rion for the dimensionality is that 2D weak localization
occurs for l

W and 1D weak localization for l

W.
On short time scales t
<

, the motion is ballistic rather

than diusive, and Eq. (2.13) does not apply. One ex-
pects the return probability to go to zero smoothly as one
enters the ballistic regime. This short-time cuto can be
accounted for heuristically by the factor 1 exp(t/),
to exclude those electrons that at time t have not been
scattered.
109
The form of the short-time cuto becomes
irrelevant for

. (See Ref.
130
for a theoretical study
of weak localization in the regime of comparable

and
.)
The foregoing analysis gives the following expressions
for the 2D and 1D weak localization corrections:

loc
=
2 h
m

_

0
dt (4Dt)
1
(1 e
t/
)e
t/

= g
s
g
v
e
2
4
2
h
ln
_
1 +

_
, if l

W, (2.14a)

loc
=
2 h
m

_

0
dt W
1
(4Dt)
1/2
(1 e
t/)
e
t/

= g
s
g
v
e
2
2h
l

W
_
1
_
1 +

_
1/2
_
, if l

W,
(2.14b)
18
where we have used the expression for the Drude con-
ductivity = e
2
(E
F
)D with the 2D density of states
(1.3). The ratio of the weak localization correction to the
Drude conductivity
loc
/ is of order 1/k
F
l for 2D weak
localization and of order (l

/W)(1/k
F
l) for 1D weak lo-
calization. In the 2D case, the correction is small (cf. the
values of k
F
l given in Table I), but still much larger than
in a typical metal. The correction is greatly enhanced in
the 1D case l

W. This is evident in the experimen-

tal curves in Fig. 13, in which the resistivity increase at
low temperatures is clearly visible only in the narrowest
channel.
The weak localization correction to the conductance
G
loc
(W/L)
loc
is of order (e
2
/h)(W/L) in the 2D
case and of order (e
2
/h)(l

/L) in the 1D case. In the lat-

ter case, the conductance correction does not scale with
the channel width W, contrary to what one would have
classically. The conductance does scale with the recip-
rocal of the channel length L, at least for L l

. The
factor l

/L in G
loc
in the 1D case can be viewed as a
consequence of the classical series addition of L/l

chan-
nel sections. It will then be clear that the scaling with L
has to break down when L
<

, in which case the weak

localization correction saturates at its value for L l

.
The maximum conductance correction in a narrow chan-
nel is thus of order e
2
/h, independent of the properties
of the sample. This universality is at the origin of
the phenomenon of the universal conductance uctua-
tions discussed in Section II.C.
2. Suppression of weak localization by a magnetic eld
(a) Theory. The resistance enhancement due to weak
localization can be suppressed by the application of a
weak magnetic eld oriented perpendicular to the 2DEG.
The suppression results from the fact that a magnetic
eld breaks time-reversal invariance. We recall that in a
zero magnetic eld, time-reversal invariance guarantees
that trajectories that form a closed loop have equal prob-
ability amplitudes A
+
and A

for clockwise and coun-

terclockwise propagation around the loop. The result-
ing constructive interference enhances the backscatter-
ing probability, thereby leading to the weak localization
eect. In a weak magnetic eld, however, a phase dier-
ence develops between A
+
and A

, even if the curva-

ture of the trajectories by the Lorentz force can be totally
negected. This Aharonov-Bohm phase results from the
fact that the canonical momentum p = mv eA of an
electron in a magnetic eld contains the vector potential
A. On clockwise (+) and counterclockwise () propa-
gation around a closed loop, one thus acquires a phase
dierence
= h
1
_
+
p
+
dl h
1
_

dl
=
2e
h
_
(A) dS =
2eBS
h

2S
l
2
m
4

0
.
(2.15)
The phase dierence is twice the enclosed area S divided
by the square of the magnetic length l
m
( h/eB)
1/2
, or,
alternatively, it is 4 times the enclosed ux in units
of the elementary ux quantum
0
h/e.
Many trajectories, with a wide distribution of loop ar-
eas, contribute to the weak localization eect. In a mag-
netic eld the loops with a large area S
>

l
2
m
no longer
contribute, since on average the counterpropagating tra-
jectories no longer interfere constructively. Since trajec-
tories enclosing a large area necessarily take a long time
to complete, the eect of a magnetic eld is essentially
to introduce a long-time cuto in the integrals of Eqs.
(2.12) and (2.14), which is the magnetic relaxation time

B
. Recall that the long-time cuto in the absence of
a magnetic eld is the phase coherence time

. The
magnetic eld thus begins to have a signicant eect on
weak localization if
B
and

are comparable, which

occurs at a characteristic eld B
c
. The weak localiza-
tion eect can be studied experimentally by measuring
the negative magnetoresistance peak associated with its
suppression by a magnetic eld. The signicance of such
experiments relies on the possibility of directly determin-
ing the phase coherence time

. The experimental data

are most naturally analyzed in terms of the conductance.
The magnitude of the zero-eld conductance correction
G
loc
(B = 0) follows directly from the saturation value
of the magnetoconductance, according to
G(B B
c
) G(B = 0) = G
loc
(B = 0). (2.16)
Once G
loc
(B = 0) is known, one can deduce the phase
coherence length l

from Eq. (2.14), since D and are

easily estimated from the classical part of the conduc-
tance (which dominates at slightly elevated tempera-
tures). The magnetoconductance contains, in addition,
information on the channel width W, which is a parame-
ter dicult to determine otherwise, as will become clear
in the discussion of the experimental situation in subsec-
tion (b).
The eectiveness of a magnetic eld in suppressing
weak localization (as contained in the functional depen-
dence of
B
on B, or in the expression for B
c
) is de-
termined by the average ux enclosed by backscattered
trajectories of a given duration. One can distinguish dif-
ferent regimes, depending on the relative magnitude of
the channel width W, the mean free path l v
F
, the
magnetic length l
m
, and the phase coherence length l

(D

)
1/2
. In Table II the expressions for
B
and B
c
are
summarized, as obtained by various authors.
109,118,121,131
In the following, we present a simple physical interpreta-
tion that explains these results, except for the numerical
19
TABLE II Magnetic relaxation time B and characteristic eld Bc for the suppression of 2D and 1D weak localization.
Dirty Metal
ab
(l W) Pure Metal
ac
(W l)
2D (l

W) 1D (W l

) 1D weak eld (l
2
m
Wl) 1D strong eld (Wl l
2
m
W
2
)
B
l
2
m
2D
3l
4
m
W
2
D
C1l
4
m
W
3
vF
C2l
2
m
l
W
2
vF
Bc
h
e
1
2l
2

h
e
3
1/2
Wl

h
e
1
W
_
C1
WvF

_
1/2
h
e
C2l
W
2
vF

a
All results assume a channel length L l

, a channel width
W
F
, as well as

.
b
From Refs.
118,131
, and
121
. The diusion constant D =
1
2
v
F
l. If
W l

, a transition to 2D weak localization occurs when lm

<

W.
c
From Ref.
109
. The constants are given by C
1
= 9.5 and C
2
=
24/5 for specular boundary scattering (C
1
= 4 and C
2
= 3 for
a channel with diuse boundary scattering). For pure metals, the
case lm < W is outside the diusive transport regime for weak
localization.
prefactors. We will not discuss the eects of spin-orbit
scattering
131
or of superconducting uctuations,
132
since
these may be neglected in the systems considered in this
review. In this subsection we only discuss the dirty metal
regime l W. The pure metal regime l W, in which
boundary scattering plays an important role, will be dis-
cussed in Section II.B.3.
If l

W the twodimensional weak localization cor-

rection to the conductivity applies, given by Eq. (2.14a)
for a zero magnetic eld. The typical area S enclosed by a
backscattered trajectory on a time scale
B
is then of the
order S D
B
(assuming diusive motion on this time
scale). The corresponding phase shift is D
B
/l
2
m
, in
view of Eq. (2.15). The criteria 1 and
B

thus
imply

B
l
2
m
/D; B
c
h/eD

h/el
2

. (2.17)
The full expression for the magnetoconductance due to
weak localization is
118,131
G
2D
loc
(B) G
2D
loc
(0) =
W
L
g
s
g
v
e
2
4
2
h
_

_
1
2
+

B
2

_
1
2
+

B
2
_
+ ln
_

_
_
, (2.18)
where (x) is the digamma function and
B
= l
2
m
/2D.
The digamma function has the asymptotic approxima-
tion (x) ln(x) 1/x for large x; thus, in a zero
magnetic eld result (2.14a) is recovered (assuming also

). In the case of 2D weak localization the char-

acteristic eld B
c
is usually very weak. For example, if
l

= 1 m, then B
c
1 mT. The suppression of the weak
localization eect is complete when
B
<

, which oc-
curs for B
>

h/eD h/el
2
. These elds are still much
weaker than classically strong elds for which
c

>

1
(as can be veried by noting that when B = h/el
2
, one
has
c
= 1/k
F
l 1). The neglect of the curvature of
electron trajectories in the theory of weak localization is
thus entirely justied in the 2D case. The safety margin
is narrower in the 1D case, however, since the character-
istic elds can become signicantly enhanced.
The one-dimensional case W l

in a magnetic eld
has rst been treated by Altshuler and Aronov
121
in the
dirty metal regime. This refers to a narrow channel with
l W so that the wall-to-wall motion is diusive. Since
the phase coherence length exceeds the channel width,
the backscattered trajectories on a time scale
B
have
a typical enclosed area S W(D
B
)
1/2
(see Fig. 15).
Consequently, the condition S l
2
m
for a unit phase shift
implies

B
l
4
m
/DW
2
; B
c
h/eWl

. (2.19)
The dierence with the 2D case is that the enclosed
ux on a given time scale is reduced, due to the lat-
eral compression of the backscattered trajectories. This
leads to an enhancement by a factor l

/W of the charac-
teristic eld scale B
c
, compared with Eq. (2.17). The
full expression for the weak localization correction if
l

, l
m
W l is
121
G
1D
loc
(B) = g
s
g
v
e
2
h
1
L
_
1
D

+
1
D
B
_
1/2
, (2.20)
20
FIG. 15 Typical closed electron trajectory contributing to 1D
weak localization (l

W) in the dirty metal regime (l

W). The asterisks denote elastic scattering events. Taken
from H. van Houten et al., Acta Electronica 28, 27 (1988).
with
B
= 3l
4
m
/W
2
D. For an elementary derivation of
this result, see Ref.
109
. At l
m
W a crossover from 1D
to 2D weak localization occurs [i.e., from Eq. (2.20) to Eq.
(2.18)]. The reason for this crossover is that the lateral
connement becomes irrelevant for the weak localization
when l
m
<

W, because the trajectories of duration

B
then have a typical extension (D
B
)
1/2
<

W, according
to Eq. (2.19). This crossover from 1D to 2D restricts the
available eld range that can be used to study the mag-
netoconductance associated with 1D weak localization.
The magnetic relaxation time
B
in the dirty metal
regime is found to be inversely proportional to the dif-
fusion constant D, in 2D as well as in 1D. The reason
for this dependence is clear: faster diusion implies that
less time is needed to complete a loop of area l
2
m
. It is
remarkable that in the pure metal regime such a propor-
tionality no longer holds. This is a consequence of the
ux cancellation eect discussed in Section II.B.3.
(b) Experiments in the dirty metal regime.
Magnetoresistance experiments have been widely used
to study the weak localization correction to the con-
ductivity of wide 2D electron gases in Si
28,30,133,134,135
and GaAs.
23,136,137
Here we will discuss the experimental
magnetoresistance studies of weak localization in narrow
channels in Si MOSFETs
34,38,40,138
and GaAs-AlGaAs
heterostructures.
24,25,58
As an illustrative example, we
reproduce in Fig. 16 a set of experimental results for
R/R [R(0) R(B)]/R(0) obtained by Choi et al.
25
in
a wide and in a narrow GaAs-AlGaAs heterostructure.
The quantity R is positive, so the resistance decreases
on applying a magnetic eld. The 2D results are simi-
lar to those obtained earlier by Paalanen et al.
137
The
qualitative dierence in eld scale for the suppression of
2D (top) and 1D (bottom) weak localization is nicely il-
lustrated by the data in Fig. 16. The magnetoresistance
peak is narrower in the 2D case, consistent with the en-
hancement in 1D of the characteristic eld B
c
for the sup-
pression of weak localization, which we discussed in Sec-
tion II.B.2(a). The solid curves in Fig. 16 were obtained
from the 2D theoretical expression (2.18) and the 1D
dirty metal result (2.20), treating W and l

as adjustable
parameters. A noteworthy nding of Choi et al.
25
is that
the eective channel width W is considerably reduced
below the lithographic width W
lith
in narrow channels
dened by a deep-etched mesa (as in Fig. 4a). Dier-
ences W W
lith
of about 0.8 m were found.
25
Signi-
FIG. 16 A comparison between the magnetoresistance
R/R [R(0)R(B)]/R(0) due to 2D weak localization in a
wide channel (upper panel) and due to 1D weak localization in
a narrow channel (lower panel), at various temperatures. The
solid curves are ts based on Eqs. (2.18) and (2.20). Taken
from K. K. Choi et al., Phys. Rev. B 36, 7751 (1987).
cantly smaller dierences are obtained
27,63
if a shallow-
etched mesa is used for the lateral connement, as in
Fig. 4c. A split-gate device (as in Fig. 4b) of variable
width has been used by Zheng et al.
24
to study weak
localization in GaAs-AlGaAs heterostructure channels.
Magnetoresistance experiments in a very narrow split-
gate device (fabricated using electron beam lithography)
were reported by Thornton et al.
58
and analyzed in terms
of the dirty metal theory. Unfortunately, in their exper-
iment the mean free path of 450 nm exceeded the width
inferred from a t to Eq. (2.20) by an order of magnitude,
so an analysis in terms of the pure metal theory would
have been required.
Early magnetoresistance experiments on narrow Si ac-
cumulation layers were performed by Dean and Pepper,
34
in which they observed evidence for a crossover from the
2D to the 1D weak localization regime. A comparison of
weak localization in wide and narrow Si inversion layers
was reported by Wheeler et al.
38
The conducting width
of the narrow channel was taken to be equal to the litho-
graphic width of the gate (about 400 nm), while the mean
free path was estimated to be about 100 nm. This experi-
ment on a low-mobility Si channel thus meets the require-
ment l W for the dirty metal regime. The 1D weak
localization condition l

W was only marginally sat-

21
ised, however. Licini et al.
40
reported a negative mag-
netoresistance peak in 270-nm-wide Si inversion layers,
which was well described by the 2D theory at a temper-
ature of 2.2 K, where l

= 120 nm. Deviations from the

2D form were found at lower temperatures, but the 1D
regime was never fully entered. A more recent study of
1D weak localization in a narrow Si accumulation layer
has been performed by Pooke et al.
138
at low tempera-
tures, and the margins are somewhat larger in their case.
We note a diculty inherent to experiments on 1D
weak localization in semiconductor channels in the dirty
metal regime. For 1D weak localization it is required that
the phase coherence length l

is much larger than the

channel width. If the mean free path is short, then the
experiment is in the dirty metal regime l W, but the
localization will be only marginally one-dimensional since
the phase coherence length l

(D

)
1/2
= (v
F
l

/2)
1/2
will be short as well (except for the lowest experimental
temperatures). If the mean free path is long, then the
1D criterion l

W is easily satised, but the require-

ment l W will now be hard to meet so that the ex-
periment will tend to be in the pure metal regime. A
quantitative comparison with the theory (which would
allow a reliable determination of l

) is hampered be-
cause the asymptotic regimes studied theoretically are
not accessible experimentally and because the channel
width is not known a priori. Nanostructures are thus
not the best candidates for a quantitative study of the
phase coherence length, which is better studied in 2D
systems. An altogether dierent complication is that
quantum corrections to the conductivity in semiconduc-
tor nanostructures can be remarkably large (up to 100%
at suciently low temperatures
27,34
), which puts them
beyond the range of validity of the perturbation theory.
3. Boundary scattering and ux cancellation
(a) Theory. In the previous subsection we noticed
that the pure metal regime, where l W, is charac-
teristic for 1D weak localization in semiconductor nano-
structures. This regime was rst theoretically consid-
ered by Dugaev and Khmelnitskii,
120
for the geometry
of a thin metal lm in a parallel magnetic eld and for
diuse boundary scattering. The geometry of a narrow
2DEG channel in a perpendicular magnetic eld, with ei-
ther diuse or specular boundary scattering, was treated
by the present authors.
109
Note that the nature of the
boundary scattering did not play a role in the dirty metal
regime of Section II.B.2, since there the channel walls
only serve to impose a geometrical restriction on the lat-
eral diusion.
121
The ux cancellation eect is character-
istic of the pure metal regime, where the electrons move
ballistically from one wall to the other. This eect (which
also plays a role in the superconductivity of thin lms in
a parallel magnetic eld
122
) leads to a further enhance-
ment of the characteristic eld scale B
c
. Flux cancel-
lation results from the fact that typically backscattered
FIG. 17 Illustration of the ux cancellation eect for a closed
trajectory of one electron in a narrow channel with diuse
boundary scattering. The trajectory is composed of two loops
of equal area but opposite orientation, so it encloses zero ux.
Taken from C. W. J. Beenakker and H. van Houten, Phys.
Rev. B. 38, 3232 (1988).
trajectories for l W self-intersect (cf. Fig. 17) and
are thus composed of smaller loops that are traversed in
opposite directions. Zero net ux is enclosed by closed
trajectories involving only wall collisions (as indicated by
the shaded areas in Fig. 17, which are equal but of oppo-
site orientation), so impurity collisions are required for
phase relaxation in a magnetic eld. This is in contrast
to the dirty metal regime considered before, where impu-
rity scattering hinders phase relaxation by reducing the
diusion constant. The resulting nonmonotonous depen-
dence of phase relaxation on impurity scattering in the
dirty and pure metal regimes is illustrated in Fig. 18,
where the calculated
109
magnetic relaxation time
B
is
plotted as a function of l/W for a xed ratio l
m
/W.
Before continuing our discussion of the ux cancella-
tion eect, we give a more precise denition of the phase
relaxation time
B
. The eect of a magnetic eld on
weak localization is accounted for formally by inserting
the term
e
i(t)
[r(t) = r(0)) = e
t/B
, W l
m
, l

, (2.21)
in the integrand of Eq. (2.12). The term (2.21) is the
conditional average over all closed trajectories having du-
ration t of the phase factor e
i(t)
, with the phase dif-
ference dened in Eq. (2.15). It can be shown
109
that in
the case of 1D weak localization (and for l
m
W), this
term is given by an exponential decay factor exp(t/
B
),
which denes the magnetic relaxation time
B
. In this
regime the weak localization correction to the conduc-
tivity in the presence of a magnetic eld is then simply
given by Eq. (2.14b), after the substitution

+
1
B
. (2.22)
Explicitly, one obtains
22
G
loc
(B) = g
s
g
v
e
2
h
1
L
_
_
1
D

+
1
D
B
_
1/2

_
1
D

+
1
D
B
+
1
D
_
1/2
_
. (2.23)
One can see from Fig. 18 and Table II that in the pure
metal regime l W, a weak and strong eld regime can
be distinguished, depending on the ratio Wl/l
2
m
. This ra-
tio corresponds to the maximum phase change on a closed
trajectory of linear extension l (measured along the chan-
nel). In the weak eld regime (Wl/l
2
m
1) many impu-
rity collisions are required before a closed electron loop
encloses sucient ux for complete phase relaxation. In
this regime a further increase of the mean free path does
not decrease the phase relaxation time (in contrast to the
dirty metal regime), because as a consequence of the ux
cancellation eect, faster diusion along the channel does
not lead to a larger enclosed ux. On comparing the re-
sult in Table II for B
c
in the weak eld regime with that
for the dirty metal regime, one sees an enhancement of
the characteristic eld by a factor (l/W)
1/2
. The strong
eld regime is reached if Wl/l
2
m
1, while still l
m
W.
Under these conditions, a single impurity collision can
lead to a closed trajectory that encloses sucient ux
for phase relaxation. The phase relaxation rate 1/
B
is now proportional to the impurity scattering rate 1/
and, thus, to 1/l. The relaxation time
B
accordingly
increases linearly with l in this regime (see Fig. 18). For
comparison with experiments in the pure metal regime,
an analytic formula that interpolates between the weak
and strong eld regimes is useful. The following formula
agrees well with numerical calculations:
109

B
=
weak
B
+
strong
B
. (2.24)
Here
weak
B
and
strong
B
are the expressions for
B
in the
asymptotic weak and strong eld regimes, as given in
Table II.
So far, we have assumed that the transport is diu-
sive on time scales corresponding to

. This will be a
good approximation only if

. Coherent diusion
breaks down if

and are of comparable magnitude (as

may be the case in high-mobility channels). The mod-
ication of weak localization as one enters the ballistic
transport regime has been investigated by Wittmann and
Schmid.
130
It would be of interest to see to what extent
the ad hoc short-time cuto introduced in our Eq. (2.14),
which is responsible for the second bracketed term in Eq.
(2.23), is satisfactory.
(b) Experiments in the pure metal regime. Be-
cause of the high mobility required, the pure metal regime
has been explored using GaAs-AlGaAs heterostructures
only. The rst experiments on weak localization in the
pure metal regime were done by Thornton et al.,
58
in
a narrow split-gate device, although the data were ana-
lyzed in terms of the theory for the dirty metal regime.
An experimental study specically aimed at weak local-
FIG. 18 Phase relaxation time B in a channel with specular
boundary scattering, as a function of the elastic mean free
path l. The plot has been obtained by a numerical simulation
of the phase relaxation process for a magnetic eld such that
lm = 10 W. The dashed lines are analytic formulas valid in
the three asymptotic regimes (see Table II). Taken from C.
W. J. Beenakker and H. van Houten, Phys. Rev. B 38, 3232
(1988).
ization in the pure metal regime was reported in Refs.
26
and
27
. In a narrow channel dened by the shallow-mesa
etch technique of Fig. 4c (with a conducting width es-
timated at 0.12 m), a pronounced negative magnetore-
sistance eect was found, similar to that observed by
Thornton et al.
58
A good agreement of the experimen-
tal results with the theory
109
for weak localization in
the pure metal regime was obtained (see Fig. 19), as-
suming specular boundary scattering (diuse boundary
scattering could not describe the data). The width de-
duced from the analysis was consistent with independent
estimates from other magnetoresistance eects. Further
measurements in this regime were reported by Chang et
al.
70,139
and, more recently, by Hiramoto et al.
81
These
experiments were also well described by the theory of
Ref.
109
.
C. Conductance uctuations
Classically, sample-to-sample uctuations in the con-
ductance are negligible in the diusive (or quasi-ballistic)
transport regime. In a narrow-channel geometry, for
example, the root-mean-square G
class
of the classical
23
FIG. 19 Magnetoconductance due to 1D weak localization
in the pure metal regime (W = 120 nm, L = 350 nm). The
solid curves are one-parameter ts to Eq. (2.23). Only the
eld range lm > W is shown in accordance with the condition
of coherent diusion imposed by the theory. The phase coher-
ence length l

obtained from the data at various temperatures

is tabulated in the inset. Taken from H. van Houten et al.,
Surf. Sci. 196, 144 (1988).
uctuations in the conductance is smaller than the av-
erage conductance G) by a factor (l/L)
1/2
, under the
assumption that the channel can be subdivided into
L/l 1 independently uctuating segments. As we
have discussed in the previous section, however, quan-
tum mechanical correlations persist over a phase coher-
ence length l

that can be much larger than the elastic

mean free path l. Quantum interference eects lead to
signicant sample-to-sample uctuations in the conduc-
tance if the size of the sample is not very much larger
than l

. The Altshuler-Lee-Stone theory of Universal

Conductance Fluctuations
140,141
nds that G e
2
/h at
T = 0, when phase coherence is maintained over the en-
tire sample. Since G) L
1
, it follows that G/G) L
increases with increasing channel length; that is, there is
a total absence of self-averaging.
Experimentally, the large sample-to-sample conduc-
tance uctuations predicted theoretically are dicult to
study in a direct way, because of problems in the prepa-
ration of samples that dier in impurity conguration
only (to allow an ensemble average). The most conve-
nient way to study the eect is via the uctuations in
the conductance of a single sample as a function of mag-
netic eld, because a small change in eld has a similar
eect on the interference pattern as a change in impu-
rity conguration. Sections II.C.3 and II.C.4 deal with
theoretical and experimental studies of magnetoconduc-
tance uctuations in narrow 2DEG channels, mainly in
the quasi-ballistic regime characteristic for semiconduc-
tor nanostructures. In Sections II.C.1 and II.C.2 we dis-
cuss the surprising universality of the conductance uc-
tuations at zero temperature and the nite-temperature
modications.
FIG. 20 Idealized conductor connecting source (S) and
drain (D) reservoirs and containing a disordered region
(crosshatched). The incoming quantum channels (or trans-
verse waveguide modes) are labeled by , the transmitted
and back scattered channels by .
1. Zero-temperature conductance uctuations
The most surprising feature of the conductance uctu-
ations is that their magnitude at zero temperature is of
order e
2
/h, regardless of the size of the sample and the
degree of disorder,
140,141
provided at least that L l,
so that transport through the sample is diusive (or pos-
sibly quasi-ballistic). Lee and Stone
141
coined the term
Universal Conductance Fluctuations (UCF) for this ef-
fect. In this subsection we give a simplied explanation
of this universality due to Lee.
142
Consider rst the classical Drude conductance (1.9) for
a singe spin direction (and a single valley):
G =
W
L
e
2
h
k
F
l
2
=
e
2
h
l
2L
N, N
k
F
W

. (2.25)
The number N equals the number of transverse modes,
or one-dimensional subbands, that are occupied at the
Fermi energy in a conductor of width W. We have writ-
ten the conductance in this way to make contact with
the Landauer approach
4
to conduction, which relates the
conductance to the transmission probabilities of modes at
the Fermi energy. (A detailed discussion of this approach
is given the context of quantum ballistic transport in Sec-
tion III.A.2). The picture to have in mind is shown in
Fig. 20. Current is passed from a source reservoir S to a
drain reservoir D, through a disordered region (hatched)
in which only elastic scattering takes place. The two
reservoirs are in thermal equilibrium and are assumed to
be fully eective in randomizing the phase via inelastic
scattering, so there is no phase coherence between the N
modes incident on the disordered region. The modes in
this context are called quantum channels. If L l, each
channel has on average the same transmission probabil-
ity, given by l/2L according to Eqs. (1.22) and (2.25).
We are interested in the uctuations around this aver-
age. The resulting uctuations in G then follow from the
multichannel Landauer formula
1,143,144
G =
e
2
h
N

,=1
[t

[
2
, (2.26)
24
where t

denotes the quantum mechanical transmission

probability amplitude from the incident channel to the
outgoing channel (cf. Fig. 20). The ensemble averaged
transmission probability [t

[
2
) does not depend on or
, so the correspondence between Eqs. (2.25) and (2.26)
requires
[t

[
2
) = l/2NL. (2.27)
The magnitude of the conductance uctuations is char-
acterized by its variance Var (G) (G G))
2
). As
discussed by Lee, a diculty arises in a direct evalu-
ation of Var (G) from Eq. (2.26), because the correla-
tion in the transmission probabilities [t

[
2
for dierent
pairs of incident and outgoing channels , may not be
neglected.
142
The reason is presumably that transmission
through the disordered region involves a large number of
impurity collisions, so a sequence of scattering events will
in general be shared by dierent channels. On the same
grounds, it is reasonable to assume that the reection
probabilities [r

[
2
for dierent pairs of incident and
reected channels are uncorrelated, since the reection
back into the source reservoir would seem to be dom-
inated by only a few scattering events.
142
(The formal
diagrammatic analysis of Refs.
140
and
141
is required here
for a convincing argument.) The reection and trans-
mission probabilities are related by current conservation
N

,=1
[t

[
2
= N
N

,=1
[r

[
2
. (2.28)
so the variance of the conductance equals
Var (G) =
_
e
2
h
_
2
Var
_

[r

[
2
_
=
_
e
2
h
_
2
N
2
Var ([r

[
2
), (2.29)
assuming uncorrelated reection probabilities. A large
number M of scattering sequences through the disordered
region contributes with amplitude A(i) (i = 1, 2, . . . , M)
to the reection probability amplitude r

. (The dif-
ferent scattering sequences can be seen as indepen-
dent Feynman paths in a path integral formulation of
the problem.
142
) To calculate Var ([r

[
2
) = [r

[
4
)
[r

[
2
)
2
, one may then write (neglecting correlations in
A(i) for dierent i)
[r

[
4
) =
M

i,j,k,l=1
A

(i)A(j)A

(k)A(l))
=
M

i,j,k,l=1
_
[A(i)[
2
)[A(k)[
2
)
ij

kl
+[A(i)[
2
)[A(j)[
2
)
il

jk
_
= 2[r

[
2
)
2
, (2.30)
where we have neglected terms smaller by a factor 1/M
(assuming M 1). One thus nds that the variance
of the reection probability is equal to the square of its
average:
Var ([r

[
2
) = [r

[
2
)
2
. (2.31)
The average reection probability [r

[
2
) does not de-
pend on and . Thus, from Eqs. (2.27) and (2.28) it
follows that
[r

[
2
) = N
1
(1 order(l/L)). (2.32)
Combining Eqs. (2.29), (2.31), and (2.32), one obtains
the result that the zero-temperature conductance has a
variance (e
2
/h)
2
, independent of l or L (in the diusive
limit l L). We have discussed this argument of Lee in
some detail, because no other simple argument known to
us gives physical insight in this remarkable result.
The numerical prefactors follow from the diagram-
matic analysis.
140,141,145,146
The result of Lee and
Stone
141
for the root-mean-square magnitude of the con-
ductance uctuations at T = 0 can be written in the form
G [Var (G)]
1/2
=
g
s
g
v
2

1/2
C
e
2
h
. (2.33)
Here C is a constant that depends on the shape of the
sample. Typically, C is of order unity; for example, C
0.73 in a narrow channel with L W. (However, in the
opposite limit W L of a wide and short channel, C
is of order (W/L)
1/2
.) The parameter = 1 in a zero
magnetic eld when time-reversal symmetry holds; = 2
when time-reversal symmetry is broken by a magnetic
eld. The factor g
s
g
v
assumes complete spin and valley
degeneracy. If the magnetic eld is suciently strong
that the two spin directions give statistically independent
contributions to the conductance, then the variances add
so that the factor g
s
in G is to be replaced by a factor
g
1/2
s
. We will return to this point in Section II.C.4.
2. Nonzero temperatures
At nonzero temperatures, the magnitude of the con-
ductance uctuations is reduced below G e
2
/h. One
reason is the eect of a nite phase coherence length
l

(D

)
1/2
; another is the eect of thermal averaging,
as expressed by the thermal length l
T
(hD/k
B
T)
1/2
.
The eect of a nite temperature, contained in l

and
l
T
, is to partially restore self-averaging, albeit that the
suppression of the uctuation with sample size is much
weaker than would be the case classically. The theory
has been presented clearly and in detail by Lee, Stone,
and Fukuyama.
145
We limit the present discussion to the
1D regime W l

L, characteristic for narrow 2DEG

channels.
The eects of thermal averaging may be neglected if
l

l
T
(see below). The channel may then be thought
25
to be subdivided in uncorrelated segments of length l

.
The conductance uctuation of each segment individually
will be of order e
2
/h, as it is at zero temperature. The
root-mean-square conductance uctuation of the entire
channel is easily estimated. The segments are in series,
so their resistances add according to Ohms law. We de-
note the resistance of a channel segment of length l

by
R
1
. The variance of R
1
is Var (R
1
) R
1
)
4
Var (R
1
1
)
R
1
)
4
(e
2
/h)
2
. The average resistance of the whole chan-
nel R) = (L/l

)R
1
) increases linearly with the number
L/l

of uncorrelated channel segments, just as its vari-

ance
Var (R) = (L/l

)Var (R
1
) (L/l

)R
1
)
4
(e
2
/h)
2
.
(The root-mean-square resistance uctuation thus grows
as (L/l

)
1/2
, the square root of the number of channel
segments in series.) Expressed in terms of a conductance,
one thus has Var (G) R)
4
Var (R) (l

/L)
3
(e
2
/h)
2
,
or
G = constant
e
2
h
_
l

L
_
3/2
, if l

l
T
. (2.34)
The constant prefactor is given in Table III.
We now turn to the second eect of the nite temper-
ature, which is the smearing of the uctuations by the
energy average within an interval of order k
B
T around
the Fermi energy E
F
. Note that we did not have to
consider this thermal averaging in the context of the
weak localization eect, since that is a systematic, rather
than a uctuating, property of the sample. Two inter-
fering Feynman paths, traversed with an energy dier-
ence E, have to be considered as uncorrelated after a
time t
1
, if the acquired phase dierence t
1
E/h is of or-
der unity. In this time the electrons diuse a distance
L
1
= (Dt
1
)
1/2
( hD/E)
1/2
. One can now dene a cor-
relation energy E
c
(L
1
), as the energy dierence for which
the phase dierence following diusion over a distance L
1
is unity:
E
c
(L
1
) hD/L
2
1
. (2.35)
The thermal length l
T
is dened such that E
c
(l
T
) k
B
T,
which implies
l
T
( hD/k
B
T)
1/2
. (2.36)
(Note that this denition of l
T
diers by a factor of
(2)
1/2
from that in Ref.
145
.) The thermal smearing
of the conductance uctuations is of importance only if
phase coherence extends beyond a length scale l
T
(i.e., if
l

l
T
). In this case the total energy interval k
B
T
around the Fermi level that is available for transport
is divided into subintervals of width E
c
(l

) = h/

in
which phase coherence is maintained. There is a num-
ber N k
B
T/E
c
(l

) of such subintervals, which we as-

sume to be uncorrelated. The root-mean-square varia-
tion G of the conductance is then reduced by a factor
N
1/2
l
T
/l

with respect to the result (2.34) in the

absence of energy averaging. (A word of caution: as dis-
cussed in Ref.
145
, the assumption of N uncorrelated en-
ergy intervals is valid in the 1D case W l

considered
here, but not in higher dimensions.) From the foregoing
argument it follows that
G = constant
e
2
h
l
T
l
1/2

L
3/2
if l

l
T
. (2.37)
The asymptotic expressions (2.34) and (2.37) were de-
rived by Lee, Stone, and Fukuyama
145
and by Altshuler
and Khmelnitskii
146
up to unspecied constant prefac-
tors. These constants have been evaluated in Ref.
147
,
and are given in Table III. In that paper we also gave an
interpolation formula
G =
g
s
g
v
2

1/2

12
e
2
h
_
l

L
_
3/2

_
1 +
9
2
_
l

l
T
_
2
_
1/2
, (2.38)
with dened in the previous subsection. This formula
is valid (within 10% accuracy) also in the intermediate
regime when l

l
T
, and is useful for comparison with
experiments, in which generally l

and l
T
are not well
separated (cf. Table I).
3. Magnetoconductance uctuations
Experimentally, one generally studies the conductance
uctuations resulting from a change in Fermi energy E
F
or magnetic eld B rather than from a change in im-
purity conguration. A comparison with the theoretical
ensemble average becomes possible if one assumes that,
insofar as the conductance uctuations are concerned, a
suciently large change in E
F
or B is equivalent to a
complete change in impurity conguration (this ergodic
hypothesis has been proven in Ref.
148
). The reason for
this equivalence is that, on one hand, the conductance
at E
F
+ E
F
and B + B is uncorrelated with that at
E
F
and B, provided either E
F
or B is larger than
a correlation energy E
c
or correlation eld B
c
. On
the other hand, the correlation energies and elds are in
general suciently small that the statistical properties of
the ensemble are not modied by the increment in E
F
or
B, so one is essentially studying a new member of the
same ensemble, without changing the sample.
This subsection deals with the calculation of the cor-
relation eld B
c
. (The correlation energy is discussed
in Ref.
145
and will not be considered here.) The magne-
toconductance correlation function is dened as
F(B) [G(B)G(B))][G(B+B)G(B+B))]),
(2.39)
where the angle brackets ) denote, as before, an
ensemble average. The root-mean-square variation G
considered in the previous two subsections is equal to
26
TABLE III Asymptotic expressions for the root-mean-square conductance uctuations in a narrow channel.
T = 0
a
T > 0
a
lT, l

L l

L, lT lT l

L
G
2
gsgv

1/2
C
e
2
h
C
e
2
h
_
l

L
_
3/2
C
e
2
h
lTl
1/2

L
3/2
C 0.73

12
_
8
3
_
1/2
a
The results assume a narrow channel (W L), with a 2D den-
sity of states (W
F
), which is in the 1D limit for the con-
ductance uctuations (W l

). The expressions for G are from

Refs.
140,141,145
, and
146
. The numerical prefactor C for T = 0 is
from Ref.
141
, for T > 0 from Ref.
147
. If time-reversal symmetry
applies, then = 1, but in the presence of a magnetic eld strong
enough to suppress the cooperon contributions then = 2. If the
spin degeneracy is lifted, gs is to be replaced by g
1/2
s
.
F(0)
1/2
. The correlation eld B
c
is dened as the half-
width at half-height F(B
c
) F(0)/2. The correlation
function F(B) is determined theoretically
141,145,146
by
temporal and spatial integrals of two propagators: the
diuson P
d
(r, r

, t) and the cooperon P

c
(r, r

, t). As dis-
cussed by Chakravarty and Schmid,
126
these propagators
consist of the product of three terms: (1) the classical
probability to diuse from r to r

in a time t (indepen-
dent of B in the eld range
c
1 of interest here); (2)
the relaxation factor exp(t/

), which describes the loss

of phase coherence due to inelastic scattering events; (3)
the average phase factor exp(i)), which describes the
loss of phase coherence due to the magnetic eld. The
average ) is taken over all classical trajectories that
diuse from r to r

in a time t. The phase dierence

is dierent for a diuson or cooperon:
(diuson) =
e
h
_
r

r
A dl, (2.40a)
(cooperon) =
e
h
_
r

r
(2A+ A) dl, (2.40b)
where the line integral is along a classical trajectory.
The vector potential A corresponds to the magnetic eld
B = A, and the vector potential increment A
corresponds to the eld increment B in the correlation
function F(B) (according to B = A). An ex-
planation of the dierent magnetic eld dependencies of
the diuson and cooperon in terms of Feynman paths is
given shortly.
In Ref.
109
we have proven that in a narrow channel
(W l

) the average phase factor exp(i)) does not

depend on initial and nal coordinates r and r

, pro-
vided that one works in the Landau gauge and that
t . This is a very useful property, since it allows
one to transpose the results for exp(i)) obtained for
r = r

in the context of weak localization to the present

problem of the conductance uctuations, where r can be
dierent from r

. We recall that for weak localization

the phase dierence is that of the cooperon, with
the vector potential increment A = 0 [cf. Eq. (2.15)].
The average phase factor then decays exponentially as
exp(i)) = exp(t/
B
) [cf. Eq. (2.21)], with the re-
laxation time
B
given as a function of magnetic eld B
in Table II. We conclude that the same exponential decay
holds for the average cooperon and diuson phase factors
after substitution of B B + B/2 and B B/2,
respectively, in the expressions for
B
:
e
i
)(diuson) = exp(t/
B/2
), (2.41a)
e
i
)(cooperon) = exp(t/
B+B/2
). (2.41b)
The cooperon is suppressed when
B+B/2
<

,
which occurs on the same eld scale as the suppression
of weak localization (determined by
B
<

). The sup-
pression of the cooperon can be seen as a consequence
of the breaking of the time-reversal invariance by the
magnetic eld, similar to the suppression of weak local-
ization. In a zero eld the cooperons and the diusons
contribute equally to the variance of the conductance;
therefore, when the cooperon is suppressed, Var (G) is
reduced by a factor of 2. (The parameter in Table
III thus changes from 1 to 2 when B increases beyond
B
c
.) In general, the magnetoconductance uctuations
are studied for B > B
c
(i.e., for elds beyond the weak
localization peak). Then only the diuson contributes to
the conductance uctuations, since the relaxation time of
the diuson is determined by the eld increment B in
the correlation function F(B), not by the magnetic eld
itself. This is the critical dierence with weak localiza-
tion: The conductance uctuations are not suppressed by
a weak magnetic eld. The dierent behavior of cooper-
ons and diusons can be understood in terms of Feyn-
man paths. The correlation function F(B) contains
the product of four Feynman path amplitudes A(i, B),
A

(j, B), A(k, B + B), and A

(l, B + B) along var-

27
FIG. 21 Illustration of the dierent ux sensitivity of the
interference terms of diuson type (a) and of cooperon type
(b). Both contribute to the conductance uctuations in a zero
magnetic eld, but the cooperons are suppressed by a weak
magnetic eld, as discussed in the text.
ious paths i, j, k, l from r to r

. Consider the diuson

term for which i = l and j = k. The phase of this term
A(i, B)A

(j, B)A(j, B + B)A

(i, B + B) is

e
h
_
A dl +
e
h
_
(A+ A) dl =
e
h
. (2.42)
where the line integral is taken along the closed loop
formed by the two paths i and j (cf. Fig. 21a). The
phase is thus given by the ux increment SB
through this loop and does not contain the ux SB
itself. The fact that the magnetic relaxation time of the
diuson depends only on B and not on B is a conse-
quence of the cancellation contained in Eq. (2.42). For
the cooperon, the relevant phase is that of the product
of Feynman path amplitudes A

(i, B)A

(j, B)A
+
(j, B+
B)A

+
(i, B +B), where the sign refers to a trajec-
tory from r

to r and the + sign to a trajectory from r

to r

(see Fig. 21b). This phase is given by

e
h
_
A dl +
e
h
_
(A+ A) dl =
e
h
(2 + ). (2.43)
In contrast to the diuson, the cooperon is sensitive to
the ux through the loop and can therefore be sup-
pressed by a weak magnetic eld.
In the following, we assume that B > B
c
so that only
the diuson contributes to the magnetoconductance uc-
tuations. The combined eects of magnetic eld and in-
elastic scattering lead to a relaxation rate

1
e
=
1

+
1
B/2
, (2.44)
which describes the exponential decay of the average
phase factor e
i
) = exp(t/
e
). Equation (2.44) con-
tains the whole eect of the magnetic eld on the dif-
fuson. Without having to do any diagrammatic analy-
sis, we therefore conclude
147
that the correlation func-
tion F(B) can be obtained from the variance F(0)
Var G = (G)
2
(given in Table III) by simply replacing

by the eective relaxation time

e
dened in Eq. (2.44).
The quantity
B/2
corresponds to the magnetic relax-
ation time
B
obtained for weak localization (see Table
II) after substitution of B B/2. For easy reference,
we give the results for the dirty and clean metal regimes
explicitly:
109,147

B/2
= 12
_
h
eB
_
2
1
DW
2
, if l W, (2.45)

B/2
= 4C
1
_
h
eB
_
2
1
v
F
W
3
+ 2C
2
_
h
eB
_
l
v
F
W
2
,
if l W, (2.46)
where C
1
= 9.5 and C
2
= 24/5 for a channel with spec-
ular boundary scattering (C
1
= 4 and C
2
= 3 for a
channel with diuse boundary scattering). These results
are valid under the condition W
2
B h/e, which fol-
lows from the requirement
e
that the electronic
motion on the eective phase coherence time scale
e
be diusive rather than ballistic, as well as from the re-
quirement (D
e
)
1/2
W for one-dimensionality.
With results (2.44)(2.46), the equation F(B
c
) =
F(0)/2, which denes the correlation eld B
c
, reduces
to an algebraic equation that can be solved straightfor-
wardly. In the dirty metal regime one nds
145
B
c
= 2C
h
e
1
Wl

, (2.47)
where the prefactor C decreases from
147
0.95 for l

l
T
to 0.42 for l

l
T
. Note the similarity with the result
(2.19) for weak localization. Just as in weak localiza-
tion, one nds that the correlation eld in the pure metal
regime is signicantly enhanced above Eq. (2.47) due to
the ux cancellation eect discussed in Section II.B.3.
The enhancement factor increases from (l/W)
1/2
to l/W
as l

decreases from above to below the length l

3/2
W
1/2
.
The relevant expression is given in Ref.
147
. As an illus-
tration, the dimensionless correlation ux B
c
Wl

e/h in
the pure and dirty metal regimes is plotted as a function
of l

/l in Fig. 22 for l
T
l

.
In the following discussion of the experimental situ-
ation in semiconductor nanostructures, it is important
to keep in mind that the Altshuler-Lee-Stone theory of
conductance uctuations was formulated for an appli-
cation to metals. This has justied the neglect of sev-
eral possible complications, which may be important in
a 2DEG. One of these is the classical curvature of the
electron trajectories, which aects the conductance when
l
cycl
<

min(W, l). A related complication is the Landau

level quantization, which in a narrow channel becomes
important when l
m
<

W. Furthermore, when W
F
the lateral connement will at low elds induce the for-
mation of 1D subbands. No quantization eects are taken
into account in the theory of conductance uctuations
discussed before. Finally, the present theory is valid only
in the regime of coherent diusion (

,
e
>

). In high-
mobility samples

and may be comparable, however,

as discussed in Section II.C.4. It would be of interest to
study the conductance uctuations in this regime theo-
retically.
28
FIG. 22 Plot of the dimensionless correlation ux c
Bcl

We/h for the magnetoconductance uctuations as a

function of l

/l in the regime lT l

. The solid curve is for

the case l = 5 W; the dashed line is for l W. Taken from
C. W. J. Beenakker and H. van Houten, Phys. Rev. B 37,
6544 (1988).
In the following discussion of experimental studies of
conductance uctuations, we will have occasion to discuss
briey one further development. This is the modication
of the theory
149,150,151,152,153,154
to account for the dif-
ferences between two- and four-terminal measurements
of the conductance uctuations, which becomes impor-
tant when the voltage probes are separated by less than
the phase coherence length.
155,156
4. Experiments
The experimental observation of conductance uctua-
tions in semiconductors has preceded the theoretical un-
derstanding of this phenomenon. Weak irregular con-
ductance uctuations in wide Si inversion layers were re-
ported in 1965 by Howard and Fang.
157
More pronounced
uctuations were found by Fowler et al. in narrowSi accu-
mulation layers in the strongly localized regime.
32
Kwas-
nick et al. made similar observations in narrow Si inver-
sion layers in the metallic conduction regime.
39
These
uctuations in the conductance as a function of gate
voltage or magnetic eld have been tentatively explained
by various mechanisms.
158
One of the explanations sug-
gested is based on resonant tunneling,
159
another on vari-
able range hopping. At the 1984 conference on Elec-
tronic Properties of Two-Dimensional Systems Wheeler
et al.
161
and Skocpol et al.
162
reported pronounced struc-
ture as a function of gate voltage in the low-temperature
conductance of narrow Si inversion layers, observed in
the course of their search for a quantum size eect.
After the publication in 1985 of the Altshuler-Lee-
Stone theory
140,141,163
of universal conductance uctua-
tions, a consensus has rapidly developed that this theory
FIG. 23 Negative magnetoresistance and aperiodic magne-
toresistance uctuations in a narrow Si inversion layer channel
for several values of the gate voltage VG. Note that the verti-
cal oset and scale is dierent for each VG. Taken from J. C.
Licini et al., Phys. Rev. Lett. 55, 2987 (1985).
properly accounts for the conductance uctuations in the
metallic regime, up to factor of two uncertainties in the
quantitative description.
46,144,164
Following this theoreti-
cal work, Licini et al.
40
attributed the magnetoresistance
oscillations that they observed in narrow Si inversion lay-
ers to quantum interference in a disordered conductor.
Their low-temperature measurements, which we repro-
duce in Fig. 23, show a large negative magnetoresistance
peak due to weak localization at low magnetic elds, in
addition to aperiodic uctuations that persist to high
elds. Such a clear weak localization peak is not found in
shorter samples, where the conductance uctuations are
larger. The reason is that the magnitude of the conduc-
tance uctuations G is proportional to (l

/L)
3/2
[for
l

l
T
, cf. Eq. (2.34)], while the weak localization con-
ductance correction scales with l

/L [as discussed below

Eq. (2.14)]. Weak localization thus predominates in long
channels (L l

) where the uctuations are relatively

unimportant.
The most extensive quantitative study of the univer-
sality of the conductance uctuations in narrow Si inver-
sion layers (over a wide range of channel widths, lengths,
gate voltages, and temperatures) was made by Skocpol et
al.
45,46,156
In the following, we review some of these ex-
perimental results. We will not discuss the similarly ex-
tensive investigations by Webb et al.
155,164,165
on small
metallic samples, which have played an equally impor-
tant role in the development of this subject. To analyze
their experiments, Skocpol et al. estimated l

from weak
localization experiments (with an estimated uncertainty
of about a factor of 2). They then plotted the root-mean-
square variation G of the conductance as a function of
L/l

, with L the separation of the voltage probes in the

channel. Their results are shown in Fig. 24. The points
for L > l

convincingy exhibit for a large variety of data

29
sets the (L/l

)
3/2
scaling law predicted by the theory
described in Section II.C.3 (for l

< l
T
, which is usually
the case in Si inversion layers).
For L < l

the experimental data of Fig. 24 show a

crossover to a (L/l

)
2
scaling law (dashed line), ac-
companied by an increase of the magnitude of the con-
ductance uctuations beyond the value G e
2
/h pre-
dicted by the Altshuler-Lee-Stone theory for a conduc-
tor of length L < l

. A similar observation was made

by Benoit et al.
155
on metallic samples. The disagree-
ment is explained
155,156
by considering that the experi-
mental geometry diers from that assumed in the the-
ory discussed in Section II.C.3. Use is made of a long
channel with voltage probes at dierent spacings. The
experimental L is the spacing of two voltage probes,
and not the length of a channel connecting two phase-
randomizing reservoirs, as envisaged theoretically. The
dierence is irrelevant if L > l

. If the probe separa-

tion L is less than the phase coherence length l

, how-
ever, the measurement still probes a channel segment
of length l

rather than L. In this sense the measure-

ment is nonlocal.
155,156
The key to the L
2
dependence
of G found experimentally is that the voltages on the
probes uctuate independently, implying that the resis-
tance uctuations R are independent of L in this regime
so that G R
2
R L
2
. This explanation is con-
sistent with the anomalously small correlation eld B
c
found for L < l

.
46,156
One might have expected that
the result B
c
h/eWl

for L > l

should be replaced
by the larger value B
c
h/eWL if L is reduced below
l

. The smaller value found experimentally is due to

the fact that the ux through parts of the channel adja-
cent to the segment between the voltage probes, as well
as the probes themselves, has to be taken into account.
These qualitative arguments
155,156
are supported by de-
tailed theoretical investigations.
149,150,151,152,153,154
The
important message of these theories and experiments is
that the transport in a small conductor is phase coherent
over large length scales and that phase randomization
(due to inelastic collisions) occurs mainly as a result of
the voltage probes. The Landauer-B uttiker formalism
4,5
(which we will discuss in Section III.A) is naturally suited
to study such problems theoretically. In that formal-
ism, current and voltage contacts are modeled by phase-
randomizing reservoirs attached to the conductor. We re-
fer to a paper by B uttiker
149
for an instructive discussion
of conductance uctuations in a multiprobe conductor in
terms of interfering Feynman paths.
Conductance uctuations have also been observed
in narrow-channel GaAs-AlGaAs heterostructures.
166,167
These systems are well in the pure metal regime (W < l),
but unfortunately they are only marginally in the regime
of coherent diusion (characterized by

). This
hampers a quantitative comparison with the theoretical
results
147
for the pure metal regime discussed in Section
II.C.3. (A phenomenological treatment of conductance
uctuations in the case that

is given in Refs.
168
and
169
.) The data of Ref.
167
are consistent with an en-
FIG. 24 Root-mean-square amplitude g of the conductance
uctuations (in units of e
2
/h) as a function of the ratio of the
distance between the voltage probes L to the estimated phase
coherence length l

for a set of Si inversion layer channels

under widely varying experimental conditions. The solid and
dashed lines demonstrate the (L/l

)
3/2
and (L/l

)
2
scaling
of g in the regimes L > l

and L < l

, respectively. Taken
from W. J. Skocpol, Physica Scripta T19, 95 (1987).
hancement of the correlation eld due to the ux cancel-
lation eect, but are not conclusive.
147
We note that the
ux cancellation eect can also explain the correlation
eld enhancement noticed in a computer simulation by
Stone.
163
In the analysis of the aforementioned experiments on
magnetoconductance uctuations, a twofold spin degen-
eracy has been assumed. The variance (G)
2
is reduced
by a factor of 2 if the spin degeneracy is lifted by a
strong magnetic eld B > B
c2
. The Zeeman energy
g
B
B should be suciently large than the spin-up and
spin-down electrons give statistically independent contri-
butions to the conductance. The degeneracy factor g
2
s
in
(G)
2
(introduced in Section II.C.1) should then be re-
placed by a factor g
s
, since the variances of statistically
independent quantities add. Since g
s
= 2, one obtains a
factor-of-2 reduction in (G)
2
. Note that this reduction
comes on top of the factor-of-2 reduction in (G)
2
due to
the breaking of time-reversal symmetry, which occurs at
weak magnetic elds B
c
. Stone has calculated
170
that the
eld B
c2
in a narrow channel (l

W) is given by the
criterion of unit phase change g
B
B

/h in a coherence
time, resulting in the estimate B
c2
h/g
B

. Surpris-
ingy, the thermal energy k
B
T is irrelevant for B
c2
in the
1D case l

W (but not in higher dimensions

170
).
For the narrow-channel experiment of Ref.
167
just dis-
cussed, one nds (using the estimates

7 ps and
g 0.4) a crossover eld B
c2
of about 2 T, well above
the eld range used for the data analysis.
147
Most im-
30
portantly, no magnetoconductance uctuations are ob-
served if the magnetic eld is applied parallel to the
2DEG (see Section II.E), demonstrating that the Zee-
man splitting has no eect on the conductance in this
eld regime. More recently, Debray et al.
171
performed
an experimental study of the reduction by a perpendic-
ular magnetic eld of the conductance uctuations as a
function of Fermi energy (varied by means of a gate).
The estimated value of

is larger than that of Ref.

167
by more than an order of magnitude. Consequently, a
very small B
c2
0.07 T is estimated in this experiment.
The channel is relatively wide (2 m lithographic width),
so the eld B
c
for time-reversal symmetry breaking is
even smaller (B
c
7 10
4
T). A total factor-of-4 re-
duction in (G)
2
was found, as expected. The values
of the observed crossover elds B
c
and B
c2
also agree
reasonably well with the theoretical prediction. Unfor-
tunately, the magnetoconductance in a parallel magnetic
eld was not investigated by these authors, which would
have provided a denitive test for the eect of Zeeman
splitting on the conductance above B
c2
. We note that re-
lated experimental
172,173
and theoretical
174,175
work has
been done on the reduction of temporal conductance uc-
tuations by a magnetic eld.
The Altshuler-Lee-Stone theory of conductance uc-
tuations ceases to be applicable when the dimensions
of the sample approach the mean free path. In this
ballistic regime observations of large aperiodic, as well
as quasi-periodic, magnetoconductance uctuations have
been reported.
68,69,139,168,176,177,178,179
Quantum inter-
ference eects in this regime are determined not by impu-
rity scattering but by scattering o geometrical features
of the device, as will be discussed in Section I.C.
D. Aharonov-Bohm eect
Magnetoconductance uctuations in a channel geome-
try in the diusive regime are aperiodic, since the inter-
fering Feynman paths enclose a continuous range of mag-
netic ux values. A ring geometry, in contrast, encloses
a well-dened ux and thus imposes a fundamental
periodicity
G() = G( +n(h/e)), n = 1, 2, 3, . . . , (2.48)
on the conductance as a function of perpendicular mag-
netic eld B (or ux = BS through a ring of area
S). Equation (2.48) expresses the fact that a ux in-
crement of an integer number of ux quanta changes by
an integer multiple of 2 the phase dierence between
Feynman paths along the two arms of the ring. The pe-
riodicity (2.48) would be an exact consequence of gauge
invariance if the magnetic eld were nonzero only in the
interior of the ring, as in the original thought experiment
of Aharonov and Bohm.
180
In the present experiments,
however, the magnetic eld penetrates the arms of the
ring as well as its interior so that deviations from Eq.
FIG. 25 Illustration of the Aharonov- Bohm eect in a ring
geometry. Interfering trajectories responsible for the magne-
toresistance oscillations with h/e periodicity in the enclosed
ux are shown (a). (b) The pair of time-reversed trajecto-
ries lead to oscillations with h/2e periodicity.
(2.48) can occur. Since in many situations such devi-
ations are small, at least in a limited eld range, one
still refers to the magnetoconductance oscillations as an
Aharonov-Bohm eect.
The fundamental periodicity
B =
h
e
1
S
(2.49)
is caused by interference between trajectories that make
one half-revolution around the ring, as in Fig. 25a. The
rst harmonic
B =
h
2e
1
S
(2.50)
results from interference after one revolution. A funda-
mental distinction between these two periodicities is that
the phase of the h/e oscillations (2.49) is sample-specic,
whereas the h/2e oscillations (2.50) contain a contribu-
tion from time-reversed trajectories (as in Fig. 25b) that
has a minimum conductance at B = 0, and thus has a
sample-independent phase. Consequently, in a geometry
with many rings in series (or in parallel) the h/e oscilla-
tions average out, but the h/2e oscillations remain. The
h/2e oscillations can be thought of as a periodic mod-
ulation of the weak localization eect due to coherent
backscattering.
The rst observation of the Aharonov-Bohm eect in
the solid state was made by Sharvin and Sharvin
181
in
a long metal cylinder. Since this is eectively a many-
ring geometry, only the h/2e oscillations were observed,
in agreement with a theoretical prediction by Altshuler,
Aronov, and Spivak,
182
which motivated the experiment.
(We refer to Ref.
125
for a simple estimate of the order
of magnitude of the h/2e oscillations in the dirty metal
regime.) The eect was studied extensively by several
groups.
183,184,185
The h/e oscillations were rst observed
in single metal rings by Webb et al.
186
and studied theo-
retically by several authors.
1,144,187,188
The self-averaging
of the h/e oscillations has been demonstrated explicitly in
31
experiments with a varying number of rings in series.
189
Many more experiments have been performed on one-
and two-dimensional arrays and networks, as reviewed in
Refs.
190
and
191
.
In this connection, we mention that the development
of the theory of aperiodic conductance uctuations (dis-
cussed in Section II.C) has been much stimulated by their
observation in metal rings by Webb et al.,
165
in the course
of their search for the Aharonov-Bohm eect. The reason
that aperiodic uctuations are observed in rings (in ad-
dition to periodic oscillations) is that the magnetic eld
penetrates the width of the arms of the ring and is not
conned to its interior. By fabricating rings with a large
ratio of radius r to width W, researchers have proven it
is possible to separate
190
the magnetic eld scales of the
periodic and aperiodic oscillations (which are given by a
eld interval of order h/er
2
and h/eWl

, respectively).
The penetration of the magnetic eld in the arms of the
ring also leads to a broadening of the peak in the Fourier
transform at the e/h and 2e/h periodicities, associated
with a distribution of enclosed ux. The width of the
Fourier peak can be used as a rough estimate for the
width of the arms of the ring. In addition, the nonzero
eld in the arms of the ring also leads to a damping of
the amplitude of the ensemble-averaged h/2e oscillations
when the ux through the arms is suciently large to
suppress weak localization.
191
Two excellent reviews of the Aharonov-Bohm eect in
metal rings and cylinders exist.
190,191
In the following
we discuss the experiments in semiconductor nanostruc-
tures in the weak-eld regime
c
< 1, where the ef-
fect of the Lorentz force on the trajectories can be ne-
glected. The strong-eld regime
c
> 1 (which is not
easily accessible in the usual polycrystalline metal rings)
is only briey mentioned; it is discussed more extensively
in Section IV.D. To our knowledge, no observation of
Aharonov-Bohm magnetoresistance oscillations in Si in-
version layers has been reported. The rst observation of
the Aharonov-Bohm eect in a 2DEG ring was published
by Timp et al.,
69
who employed high-mobility GaAs-
AlGaAs heterostructure material. Similar results were
obtained independently by Ford et al.
73
and Ishibashi et
al.
193
More detailed studies soon followed.
74,139,176,194,195
A characteristic feature of these experiments is the large
amplitude of the h/e oscillations (up to 10% of the aver-
age resistance), much higher than in metal rings (where
the eect is at best
192,196,197
of order 0.1%). A similar
dierence in magnitude is found for the aperiodic mag-
netoresistance uctuations in metals and semiconductor
nanostructures. The reason is simply that the amplitude
G of the periodic or aperiodic conductance oscillations
has a maximum value of order e
2
/h, so the maximum
relative resistance oscillation R/R RG Re
2
/h is
proportional to the average resistance R, which is typi-
cally much smaller in metal rings.
In most studies only the h/e fundamental periodic-
ity is observed, although Ford et al.
73,74
found a weak
h/2e harmonic in the Fourier transform of the magne-
toresistance data of a very narrow ring. It is not quite
clear whether this harmonic is due to the Altshuler-
Aronov-Spivak mechanism involving the constructive in-
terference of two time-reversed trajectories
182
or to the
random interference of two non-time-reversed Feynman
paths winding around the entire ring.
1,144,187
The rela-
tive weakness of the h/2e eect in single 2DEG rings is
also typical for most experiments on single metal rings
(although the opposite was found to be true in the case
of aluminum rings by Chandrasekhar et al.,
197
for rea-
sons which are not understood). This is in contrast to
the case of arrays or cylinders, where, as we mentioned,
the h/2e oscillations are predominant the h/e eect be-
ing ensemble-averaged to zero because of its sample-
specic phase. In view of the fact that the experiments
on 2DEG rings explore the borderline between diusive
and ballistic transport, they are rather dicult to ana-
lyze quantitatively. A theoretical study of the Aharonov-
Bohm eect in the purely ballistic transport regime was
performed by Datta and Bandyopadhyay,
198
in relation
to an experimental observation of the eect in a double-
quantum-well device.
199
A related study was published
by Barker.
200
The Aharonov-Bohm oscillations in the magnetoresis-
tance of a small ring in a high-mobility 2DEG are quite
impressive. As an illustration, we reproduce in Fig. 26
the results obtained by Timp et al.
201
Low-frequency
modulations were ltered out, so that the rapid oscil-
lations are superimposed on a constant background. The
amplitude of the h/e oscillations diminishes with increas-
ing magnetic eld until eventually the Aharonov-Bohm
eect is completely suppressed. The reduction in am-
plitude is accompanied by a reduction in frequency. A
similar observation was made by Ford et al.
74
In metals,
in contrast, the Aharonov-Bohm oscillations persist to
the highest experimental elds, with constant frequency.
The dierent behavior in a 2DEG is a consequence of
the eect of the Lorentz force on the electrons in the
ring, which is of importance when the cyclotron diame-
ter 2l
cycl
becomes smaller than the width W of the arm
of the ring, provided W < l (note that l
cycl
= hk
F
/eB
is much smaller in a 2DEG than in a metal, at the same
magnetic eld value). We will return to these eects in
Section IV.D.
An electrostatic potential V aects the phase of the
electron wave function through the term (e/h)
_
V dt in
much the same way as a vector potential does. If the
two arms of the ring have a potential dierence V , and
an electron traverses an arm in a time t, then the ac-
quired phase shift would lead to oscillations in the re-
sistance with periodicity V = h/et. The electrostatic
Aharonov-Bohm eect has a periodicity that depends on
the transit time t, and is not a geometrical property of
the ring, as it is for the magnetic eect. A distribution
of transit times could easily average out the oscillations.
Note that the potential dierence eectuates the phase
dierence by changing the wavelength of the electrons
(via a change in their kinetic energy), which also distin-
32
FIG. 26 Experimental magnetoresistance of a ring of 2 m di-
ameter, dened in the 2DEG of a high-mobility GaAs-AlGaAs
heterostructure (T = 270 mK). The dierent traces are con-
secutive parts of a magnetoresistance measurement from 0
to 1.4 T, digitally ltered to suppress a slowly varying back-
ground. The oscillations are seen to persist for elds where
c > 1, but their amplitude is reduced substantially for
magnetic elds where 2l
cycl
W. (The eld value where
2l
cycl
2rc = W is indicated). Taken from G. Timp et al.,
Surf. Sci. 196, 68 (1988).
guishes the electrostatic from the magnetic eect (where
a phase shift is induced by the vector potential without
a change in wavelength). An experimental search for the
electrostatic Aharonov-Bohm eect in a small metal ring
was performed by Washburn et al.
202
An electric eld
was applied in the plane of the ring by small capacitive
electrodes. They were able to shift the phase of the mag-
netoresistance oscillations by varying the eld, but the
eect was not suciently strong to allow the observa-
tion of purely electrostatic oscillations. Unfortunately,
this experiment could not discriminate between the ef-
fect of the electric eld penetrating in the arms of the
ring (which could induce a phase shift by changing the
trajectories) and that of the electrostatic potential. Ex-
periments have been reported by De Vegvar et al.
203
on
the manipulation of the phase of the electrons by means
of the voltage on a gate electrode positioned across one
of the arms of a heterostructure ring. In this system a
change in gate voltage has a large eect on the resistance
of the ring, primarily because it strongy aects the local
density of the electron gas. No clear periodic signal, in-
dicative of an electrostatic Aharonov-Bohm eect, could
be resolved. As discussed in Ref.
203
, this is not too sur-
prising, in view of the fact that in that device 1D subband
depopulation in the region under the gate occurs on the
same gate voltage scale as the expected Aharonov-Bohm
eect. The observation of an electrostatic Aharonov-
Bohm eect thus remains an experimental challenge. A
successful experiment would appear to require a ring in
which only a single 1D subband is occupied, to ensure a
unique transit time.
198,200
E. Electron-electron interactions
1. Theory
In addition to the weak localization correction to
the conductivity discussed in Section II.B, which arises
from a single-electron quantum interference eect, the
Coulomb interaction of the conduction electrons gives
also rise to a quantum correction.
204,205
In two dimen-
sions the latter correction has a logarithmic tempera-
ture dependence, just as for weak localization [see Eq.
(2.14)]. A perpendicular magnetic eld can be used
to distinguish the two quantum corrections, which have
a dierent eld dependence.
118,204,205,206,207,208,209,210
This eld of research has been reviewed in detail by
Altshuler and Aronov,
211
by Fukuyama,
212
and by Lee
and Ramakrishnan,
127
with an emphasis on the theory.
A broader review of electronic correlation eects in 2D
systems has been given by Isihara in this series.
213
In the
present subsection we summarize the relevant theory, as
a preparation for the following subsection on experimen-
tal studies in semiconductor nanostructures. We do not
discuss the diagrammatic perturbation theory, since it is
highly technical and does not lend itself to a discussion
at the same level as for the other subjects dealt with in
this review.
An attempt at an intuitive interpretation of the Feyn-
man diagrams was made by Bergmann.
214
It is argued
that one important class of diagrams may be interpreted
as diraction of one electron by the oscillations in the
electrostatic potential generated by the other electrons.
The Coulomb interaction between the electrons thus in-
33
troduces a purely quantum mechanical correlation be-
tween their motion, which is observable in the conduc-
tivity. The diraction of one electron wave by the inter-
ference pattern generated by another electron wave will
only be of importance if their wavelength dierence, and
thus their energy dierence, is small. At a nite temper-
ature T, the characteristic energy dierence is k
B
T. The
time
T
h/k
B
T enters as a long-time cuto in the the-
ory of electron-electron interactions in a disordered con-
ductor, in the usual case
127,211

T
<

. (Fukuyama
212
also discusses the opposite limit
T

.) Accord-
ingy, the magnitude of the thermal length l
T
(D
T
)
1/2
compared with the width W determines the dimensional
crossover from 2D to 1D [for l
T
< l

(D

)
1/2
]. In
the expression for the conductivity correction associated
with electron-electron interactions, the long-time cuto

T
enters logarithmically in 2D and as a square root in
1D. These expressions thus have the same form as for
weak localization, but with the phase coherence time

replaced by
T
. The origin of this dierence is that a -
nite temperature does not introduce a long-time cuto for
the single-electron quantum interference eect responsi-
ble for weak localization, but merely induces an energy
average of the corresponding conductivity correction.
In terms of eective interaction parameters g
2D
and
g
1D
, the conductivity corrections due to electron-electron
interactions can be written as (assuming
T

ee
=
e
2
2
2
h
g
2D
ln

T

, for l
T
W, (2.51a)

ee
=
e
2
2
1/2
h
g
1D
l
T
W
, for W l
T
L. (2.51b)
Under typical experimental conditions,
55
the constants
g
2D
and g
1D
are positive and of order unity. Theoreti-
cally, these eective interaction parameters depend in a
complicated way on the ratio of screening length to Fermi
wavelength and can have either sign. We do not give
the formulas here, but refer to the reviews by Altshuler
and Aronov
211
and Fukuyama.
212
In 2D the interaction
correction
ee
shares a logarithmic temperature depen-
dence with the weak localization correction
loc
, and
both corrections are of the same order of magnitude. In
1D the temperature dependences of the two eects are
dierent (unless

T
1/2
). Moreover, in the 1D case

ee

loc
if l
T
l

.
A weak magnetic eld fully suppresses weak localiza-
tion, but has only a small eect on the quantum cor-
rection from electron-electron interactions. The conduc-
tance correction G
ee
contains contributions of diuson
type and of cooperon type. The diusons (which give the
largest contributions to G
ee
) are aected by a magnetic
eld only via the Zeeman energy, which removes the spin
degeneracy when g
B
B
>

k
B
T. In the systems of in-
terest here, spin splitting can usually be ignored below
1T, so the diusons are insensitive to a weak magnetic
eld. Since the spin degeneracy is removed regardless of
the orientation of the magnetic eld, the B-dependence
of the diuson is isotropic. The smaller cooperon contri-
butions exhibit a similar sensitivity as weak localization
to a weak perpendicular magnetic eld, the character-
istic eld being determined by l
2
m
l
2
T
in 2D and by
l
2
m
Wl
T
in 1D (in the dirty metal regime W l,
so ux cancellation does not play a signicant role). The
magnetic length l
m
( h/eB

)
1/2
contains only the com-
ponent B

of the eld perpendicular to the 2DEG, since

the magnetic eld aects the cooperon via the phase
shift induced by the enclosed ux. The anisotropy and
the small characteristic eld are two ways to distinguish
experimentally the cooperon contribution from that of
the diuson. It is much more dicult to distinguish the
cooperon contribution to G
ee
from the weak localization
correction, since both eects have the same anisotropy,
while their characteristic elds are comparable (l
T
and
l

not being widely separated in the systems considered

here). This complication is made somewhat less problem-
atic by the fact that the cooperon contribution to G
ee
is often considerably smaller than G
loc
, in which case
it can be ignored. In 1D the reduction factor
55,211
is of
order [1 + ln(E
F
/k
B
T)]
1
(l
T
/l

), with a numerical
coecient of order unity.
There is one additional aspect to the magnetoresis-
tance due to electron-electron interactions that is of little
experimental relevance in metals but becomes important
in semiconductors in the classically strong-eld regime
where
c
> 1 (this regime is not easily accessible in
metal nanostructures because of the typically short scat-
tering time). In such strong elds only the diuson con-
tributions to the conductivity corrections survive. Ac-
cording to Houghton et al.
215
and Girvin et al.,
216
the
diuson does not modify the o-diagonal elements of the
conductivity tensor, but only the diagonal elements

xy
=
yx
= 0,
xx
=
yy

ee
, (2.52)
where
ee
is approximately eld-independent (provided
spin splitting does not play a role). In a channel geometry
one measures the longitudinal resistivity
xx
, which is
related to the conductivity tensor elements by

xx


yy

xx

yy
+
2
xy
=
0
xx
+
0
xx
_

ee

0
xx
2
0
xx

ee
_
+ order(
ee
)
2
.
(2.53)
Here
0
xx
= and
0
xx
= [1 + (
c
)
2
]
1
are the clas-
sical results (1.26) and (1.27). In obtaining this re-
sult the eects of Landau level quantization on the con-
ductivity have been disregarded (see, however, Ref.
55
).
The longitudinal resistivity thus becomes magnetic-eld-
dependent:

xx
= (1 + [(
c
)
2
1]
ee
/). (2.54)
To the extent that the B-dependence of
ee
can be ne-
glected, Eq. (2.54) gives a parabolic negative magnetore-
sistance, with a temperature dependence that is that of
34
the negative conductivity correction
ee
. This eect can
easily be studied up to
c
= 10, which would imply an
enhancement by a factor of 100 of the resistivity correc-
tion in zero magnetic eld. (The Hall resistivity
xy
also
contains corrections from
ee
, but without the enhance-
ment factor.) In 2D it is this enhancement that allows
the small eect of electron-electron interactions to be ob-
servable experimentally (in as far as the eect is due to
diuson-type contributions).
Experimentally, the parabolic negative magnetoresis-
tance associated with electron-electron interactions was
rst identied by Paalanen et al.
137
in high-mobility
GaAs-AlGaAs heterostructure channels. A more detailed
study was made by Choi et al.
55
In that paper, as well
as in Ref.
113
, it was found that the parabolic magnetore-
sistance was less pronounced in narrow channels than
in wider ones. Choi et al. attributed this suppression
to specular boundary scattering. It should be noted,
however, that specular boundary scattering has no ef-
fect at all on the classical conductivity tensor
0
(in
the scattering time approximation; cf. Section II.A.2).
Since the parabolic magnetoresistance results from the
(
c
)
2
term in 1/
0
xx
[see Eq. (2.54)], one would expect
that specular boundary scattering does not suppress the
parabolic magnetoresistance (assuming that the result

xy
=
yx
= 0 still holds in the pure metal regime
l > W). Diuse boundary scattering does aect
0
, but
only for relatively weak elds such that 2l
cycl
> W (see
Section II.A); hence, diuse boundary scattering seems
equally inadequate in explaining the observations. In the
absence of a theory for electron-electron interaction ef-
fects in the pure metal regime, this issue remains unset-
tled.
2. Narrow-channel experiments
Wheeler et al.
38
were the rst to use magnetoresis-
tance experiments as a tool to distinguish weak localiza-
tion from electron-electron interaction eects in narrow
Si MOSFETs. As in most subsequent studies, the nega-
tive magnetoresistance was entirely attributed to the sup-
pression of weak localization; the cooperon-type contri-
butions from electron-electron interactions were ignored.
After subtraction of the weak localization correction, the
remaining temperature dependence was found to dier
from the simple T
1/2
dependence predicted by the the-
ory for W < l
T
< l

[Eq. (2.51b)]. This was attributed

in Ref.
38
to temperature-dependent screening at the rel-
atively high temperatures of the experiment. Pooke et
al.
138
found a nice T
1/2
dependence in similar experi-
ments at lower temperatures in narrow Si accumulation
layers and in GaAs-AlGaAs heterostructures.
The most detailed study by far of the 2D to 1D
crossover of the electron-electron interaction eect in
narrow channels was made by Choi et al.
55
in a GaAs-
AlGaAs heterostructure. In Fig. 27 we reproduce some
of their experimental traces for channel widths from 156
FIG. 27 Negative magnetoresistance in wide and narrow
GaAs-AlGaAs channels at 4.2 and 1.6 K. The temperature-
independent negative magnetoresistance at low elds is a clas-
sical size eect. The temperature-dependent parabolic mag-
netoresistance at higher elds is a quantum interference eect
associated with electron-electron interactions. Shubnikov-De
Haas oscillations are visible for elds greater than about 0.3 T.
Taken from K. K. Choi et al., Phys. Rev. B 33, 8216 (1986).
to 1.1 m and a channel length of about 300 m. The
weak localization peak in the magnetoresistance is not re-
solved in this experiment, presumably because the chan-
nels are not in the 1D regime for this eect (the 2D
weak localization peak would be small and would have a
width of 10
4
T). The negative magnetoresistance that
they found below 0.1 0.2 T in the narrowest channels
is temperature-independent between 1 and 4 K and was
therefore identied by Choi et al.
55
as a classical size ef-
fect. The classical negative magnetoresistance extends
over a eld range for which 2l
cycl
>

W. This eect
has been discussed in Section II.A in terms of reduc-
tion of backscattering by a magnetic eld. The electron-
electron interaction eect is observed as a (temperature-
dependent) parabolic negative magnetoresistance above
0.1 T for the widest channel and above 0.3 T for the nar-
rowest one. From the magnitude of the parabolic nega-
tive magnetoresistance, Choi et al.
55
could nd and an-
alyze the crossover from 2D to 1D interaction eects. In
addition, they investigated the cross over to 0D by per-
forming experiments on short channels. As seen in Fig.
27, Shubnikov-De Haas oscillations are superimposed on
the parabolic negative magnetoresistance at low temper-
atures and strong magnetic elds. It is noteworthy that
stronger elds are required in narrower channels to ob-
serve the Shubnikov-De Haas oscillations, an eect dis-
cussed in terms of specular boundary scattering by Choi
et al. The Shubnikov-De Haas oscillations in narrow
channels are discussed further in Section II.F.2.
In Refs.
63,167
, and
27
the work by Choi et al.
55
was
extended to even narrower channels, well into the 1D
pure metal regime. The results for a conducting channel
width of 0.12 m are shown in Fig. 28. The 1D weak
35
FIG. 28 Magnetoresistance at various temperatures of a
GaAs-AlGaAs channel (W = 0.12 m, L = 10 m) dened
by a shallow-mesa etch technique. The central negative mag-
netoresistance peak between 0.1 and +0.1 T at low temper-
atures is due to 1D weak localization in the quasi-ballistic
regime. Conductance uctuations are seen at larger elds.
The negative magnetoresistance that persists to high temper-
atures is a classical size eect as in Fig. 27. The temperature
dependence of the resistance at B = 0 is due to a combination
of weak localization and electron-electron interaction eects
(see Fig. 30). Taken from H. van Houten et al., Appl. Phys.
Lett. 49, 1781 (1986).
localization peak in the magnetoresistance is quite large
for this narrow channel (even at the rather high tem-
peratures shown) and clearly visible below 0.1 T. The
classical size eect due to reduction of backscattering
now leads to a negative magnetoresistance on a larger
eld scale of about 1 T, in agreement with the criterion
2l
cycl
W. This is best seen at temperatures above 20 K,
where the quantum mechanical eects are absent. The
temperature-dependent parabolic negative magnetoresis-
tance is no longer clearly distinguishable in the narrow
channel of Fig. 28, in contrast to wider channels.
27,55
The suppression of this eect in narrow channels is not
yet understood (see Section II.E.1). Superimposed on
the smooth classical magnetoresistance, one sees large
aperiodic uctuations on a eld scale of the same mag-
nitude as the width of the weak localization peak, in
qualitative agreement with the theory of universal con-
ductance uctuations in the pure metal regime
147
(see
Section II.C.4). Finally, Shubnikov-De Haas oscillations
are beginning to be resolved around 1.2 T, but they are
periodic in 1/B at stronger magnetic elds only (not
shown). As discussed in Section II.F, this anomaly in the
Shubnikov-De Haas eect is a manifestation of a quan-
tum size eect.
167,217,218
This one gure thus summarizes
the wealth of classical and quantum magnetoresistance
phenomena in the quasi-ballistic transport regime.
Essentially similar results were obtained by
Taylor et al.
219
In the eld range of these
experiments,
27,55,63,167,219
the magnetoresistance is
FIG. 29 Angular dependence of the magnetoresistance of Fig.
28, at 4 K, proving that it has a purely orbital origin. Taken
from H. van Houten et al., Superlattices and Microstructures
3, 497 (1987).
exclusively caused by the enclosed ux and the Lorentz
force (so called orbital eects). The Zeeman energy
does not play a role. This is demonstrated in Fig. 29,
where the magnetoresistance (obtained on the same
sample as that used in Fig. 28) is shown to vanish when
B is in the plane of the 2DEG (similar results were
obtained in Ref.
168
). In wide 2DEG channels a negative
magnetoresistance has been found by Lin et al. in a
parallel magnetic eld.
23
This eect has been studied
in detail by Mensz and Wheeler,
220
who attributed it
to a residual orbital eect associated with deviations
of the 2DEG from a perfectly at plane. Falko
221
has
calculated the eect of a magnetic eld parallel to the
2DEG on weak localization, and has found a negative
magnetoresistance, but only if the scattering potential
does not have reection symmetry in the plane of the
2DEG.
In Fig. 30 the temperature dependence of the zero-
eld conductance
27
is plotted as a function of T
1/2
,
together with the conductance after subtraction of the
weak localization correction. The straight line through
the latter data points demonstrates that the remain-
ing temperature dependence may, indeed, be attributed
to the electron-electron interactions. A similar T
1/2
dependence was found by Thornton et al.
58
in a nar-
row GaAs-AlGaAs channel dened using the split-gate
36
FIG. 30 Zero-eld conductance (circles) and conductance
corrected for the weak localization eect (squares) for the
channel of Fig. 28 as a function of T
1/2
, to demonstrate the
T
1/2
dependence on the temperature of the electron-electron
interaction eect expected from Eq. (2.51b). The solid and
dashed lines are guides to the eye. The extrapolated value
at high temperatures is the classical part of the conductance.
Taken from H. van Houten et al., Acta Electronica 28, 27
(1988).
method. The slope of the straight line in Fig. 30 gives
g
1D
1.5 in Eq. (2.51b), which is close to the value found
by Choi et al.
55
It should be noted, however, that this
experiment is already in the regime where the quantum
corrections are by no means small, so the perturbation
theory is of questionable validity. For this reason, and
also in view of other problems (such as the diculty in
determining the eective channel width, the presence of
channel width variations, and a frequently observed sat-
uration of the weak localization correction at low tem-
peratures due to loss of phase coherence associated with
external noise or radio-frequency interference), a quanti-
tative analysis of the parameters obtained from the weak
localization and electron-electron corrections in narrow
channels (

and g
1D
) is not fully warranted. Indeed,
most of the narrow-channel studies available today have
not been optimized for the purpose of a detailed quan-
titative analysis. Instead, they were primarily intended
for a phenomenological exploration, and as such we feel
that they have been quite successful.
F. Quantum size eects
Quantum size eects on the resistivity result from
modications of the 2D density of states in a 2DEG chan-
nel of width comparable to the Fermi wavelength. The
electrostatic lateral connement in such a narrow chan-
nel leads to the formation of 1D subbands in the conduc-
tion band of the 2DEG (see Section I.D.1). The number
N k
F
W/ of occupied 1D subbands is reduced by de-
creasing the Fermi energy or the channel width. This
depopulation of individual subbands can be detected via
the resistivity. An alternative method to depopulate the
subbands is by means of a magnetic eld perpendicular
to the 2DEG. The magnetic eld B has a negligible eect
on the density of states at the Fermi level if the cyclotron
diameter 2l
cycl
W (i.e., for B B
crit
2 hk
F
/eW). If
B B
crit
, the electrostatic connement can be neglected
for the density of states, which is then described by Lan-
dau levels [Eq. (1.7)]. The number of occupied Landau
levels N B
F
/h
c
k
F
l
cycl
/2 decreases linearly with B
for B B
crit
. In the intermediate eld range where B
and B
crit
are comparable, the electrostatic connement
and the magnetic eld together determine the density of
states. The corresponding magnetoelectric subbands are
depopulated more slowly by a magnetic eld than are
the Landau levels, which results in an increased spacing
of the Shubnikov-De Haas oscillations in the magnetore-
sistivity (cf. Section I.D.3).
In the following subsection we give a more quantitative
description of magnetoelectric subbands. Experiments
on the electric and magnetic depopulation of subbands
in a narrow channel are reviewed in Section II.E.2. We
only consider here the case of a long channel (L l) in
the quasi-ballistic regime. Quantum size eects in the
fully ballistic regime (L
<

l) are the subject of Section

III.
1. Magnetoelectric subbands
Consider rst the case of an unbounded 2DEG in a
perpendicular magnetic eld B = A. The Hamil-
tonian for motion in the plane of the 2DEG is given by
H =
(p +eA)
2
2m
, (2.55)
for a single spin component. In the Landau gauge A =
(0, Bx, 0), with B in the z-direction, this may be written
as
H =
p
2
x
2m
+
m
2
c
2
(x x
0
)
2
, (2.56)
with
c
eB/m and x
0
p
y
/eB. The y-momentum
operator p
y
ih/y can be replaced by its eigenvalue
hk
y
, since p
y
and H commute. The eect of the mag-
netic eld is then represented by a harmonic oscillator
potential in the x-direction, with center x
0
= hk
y
/eB
37
FIG. 31 Magnetic eld dependence of the number N of oc-
cupied subbands in a narrow channel for a parabolic conning
potential according to Eq. (2.61) (dot-dashed curve), and for
a square-well conning potential according to Eq. (2.62) (full
curve). The dashed curve gives the magnetic depopulation of
Landau levels in a wide 2DEG, which has a 1/B dependence.
The calculations are done for a xed Fermi energy and for
channel width W = Wpar = 10/kF.
depending on the momentum in the y-direction. The en-
ergy eigenvalues E
n
= (n
1
2
) h
c
, n = 1, 2, 3, . . ., do not
depend on k
y
and are therefore highly degenerate. States
with the same quantum number n are referred to collec-
tively as Landau levels.
93
The number of Landau levels
below energy E is given by
N = Int[1/2 +E/h
c
], (2.57)
where Int denotes truncation to an integer.
A narrow channel in the y-direction is dened
by an electrostatic conning potential V (x). The
case of a parabolic connement is easily solved
analytically.
36,218,222,223
Adding a term V (x) =
1
2
m
2
0
x
2
to the hamiltonian (2.56), one nds, after a rearrange-
ment of terms,
H =
p
2
x
2m
+
m
2
2
(x x
0
)
2
+
h
2
k
2
2M
, (2.58)
with (
2
c
+
2
0
)
1/2
, x
0
x
0

c
/, and M m
2
/
2
0
.
The rst two terms describe the motion in the x-direction
in a harmonic potential with eective frequency

0
, and the third term describes free motion in the y-
direction with an eective mass M m. This last term
removes the degeneracy of the Landau levels, which be-
come 1D subbands with energy
E
n
(k) = (n
1
2
) h + h
2
k
2
/2M. (2.59)
The subband bottoms have energy E
n
= (n
1
2
) h,
and the number of subbands occupied at energy E is
N = Int[
1
2
+ E/h]. The quasi-1D density of states is
obtained from Eq. (1.5) on substituting m for M. For
the comparison with experiments it is useful to dene
an eective width for the parabolic potential. One can
take the width W
par
to be the separation between the
equipotentials at the Fermi energy,
W
par
2 hk
F
/m
0
. (2.60)
(An alternative, which diers only in the numerical pref-
actor, is to take W
par
n
1D
/n
s
, with n
s
g
s
g
v
k
2
F
/4
the 2D sheet density and n
1D
the number of electrons
per unit length in the narrow channel.
218
) The number
of occupied magnetoelectric subbands at energy E
F
in a
parabolic conning potential may then be written as
N = Int
_
1
2
+
1
4
k
F
W
par
[1 + (W
par
/2l
cycl
)
2
]
1/2
_
,
(2.61)
where l
cycl
hk
F
/eB is the cyclotron radius at the Fermi
energy. For easy reference, we also give the result for the
number of occupied subbands at the Fermi energy in a
square-well connement potential of width W:
N Int
_
_
2

E
F
h
c
_
_
arcsin
W
2l
cycl
+
W
2l
cycl
_
1
_
W
2l
cycl
_
2
_
1/2
_
_
_
_
, if l
cycl
>
W
2
, (2.62a)
N Int
_
1
2
+
E
F
h
c
_
if l
cycl
<
W
2
. (2.62b)
(This result is derived in Section III.A.1 in a semiclas-
sical approximation. The accuracy is 1.) One easily
veries that, for B B
crit
2 hk
F
/eW, Eq. (2.62) yields
N k
F
W/. The parabolic conning potential gives
N k
F
W
par
/4 in the weak-eld limit. In the strong-
eld limit B B
crit
, both potentials give the result
38
N E
F
/h
c
= k
F
l
cycl
/2 expected for pure Landau lev-
els. In Fig. 31 we compare the depopulation of Lan-
dau levels in an unbounded 2DEG with its characteristic
1/B dependence of N (dashed curve), with the slower
depopulation of magnetoelectric subbands in a narrow
channel. The dash-dotted curve is for a parabolic con-
ning potential, the solid curve for a square-well poten-
tial. These results are calculated from Eqs. (2.61) and
(2.62), with k
F
W
par
/ = k
F
W/ = 10. A B-independent
Fermi energy was assumed in Fig. 31 so that the density
n
1D
oscillates around its zero-eld value. (For a long
channel, it is more appropriate to assume that n
1D
is B-
independent, to preserve charge neutrality, in which case
E
F
oscillates. This case is studied in Ref.
218
.) Qual-
itatively, the two conning potentials give similar re-
sults. The numerical dierences reect the uncertainty
in assigning an eective width to the parabolic potential.
Self-consistent solutions of the Poisson and Schrodinger
equations
42,60,61,72,224
for channels dened by a split gate
have shown that a parabolic potential with a at bottom
section is a more realistic model. The subband depopula-
tion for this potential has been studied in a semiclassical
approximation in Ref.
223
. A disadvantage of this more
realistic model is that an additional parameter is needed
for its specication (the width of the at section). For
this practical reason, the use of either a parabolic or a
square-well potential has been preferred in the analysis
of most experiments.
2. Experiments on electric and magnetic depopulation of
subbands
The observation of 1D subband eects unobscured by
thermal smearing requires low temperatures, such that
4k
B
T E, with E the energy dierence between
subband bottoms near the Fermi level (4k
B
T being the
width of the energy averaging function df/dE
F
; see Sec-
tion I.D.2; For a square well E 2E
F
/N, and for
parabolic connement E E
F
/N). Moreover, the for-
mation of subbands requires the eective mean free path
(limited by impurity scattering and diuse boundary
scattering) to be much larger than W (cf. also Ref.
218
).
The requirement on the temperature is not dicult to
meet, E/4k
B
T being on the order of 50 K for a typ-
ical GaAs-AlGaAs channel of width W = 100 nm, and
the regime l > W is also well accessible. These sim-
ple considerations seem to suggest that 1D subband ef-
fects should be rather easily observed in semiconductor
nanostructures. This conclusion is misleading, however,
and in reality manifestations of 1D subband structure
have been elusive, at least in the quasi-ballistic regime
W < l < L. The main reason for this is the appearance
of large conductance uctuations that mask the subband
structure if the channel is not short compared with the
mean free path.
Calculations
225,226,227
of the average conductivity of
an ensemble of narrow channels do in fact show oscil-
FIG. 32 (a) Dependence on the gate voltage of the cur-
rent I through 250 parallel narrow Si inversion layer chan-
nels at 1.2 K, showing the electric depopulation of subbands.
(b) The eect is seen more clearly in the transconductance
dI/dVG. Note the absence of universal conductance uctua-
tions, which have been averaged out by the large number of
channels. Taken from A. C. Warren et al., IEEE Electron
Device Lett. EDL-7, 413 (1986).
lations from the electric depopulation of subbands [re-
sulting from the modulation of the density of states at
the Fermi level, which determines the scattering time;
see Eq. (1.29)]. The oscillations are not as large as the
Shubnikov-De Haas oscillations from the magnetic de-
population of Landau levels or magnetoelectric subbands.
One reason for this dierence is that the peaks in the
density of states become narrower, relative to their sep-
aration, on applying a magnetic eld. (The quantum
limit of a single occupied 1D subband has been studied
in Refs.
42
and
228,229,230
.)
In an individual channel, aperiodic conductance uctu-
ations due to quantum interference (see Section 7) are the
dominant cause of structure in the low-temperature con-
ductance as a function of gate voltage (which corresponds
to a variation of the Fermi energy), as was found in ex-
39
periments on narrow Si inversion layers.
46,161,162
Warren
et al.
44
were able to suppress these uctuations by per-
forming measurements on an array of narrow channels
in a Si inversion layer. In Fig. 32 we reproduce their
results. The structure due to the electric depopulation
of 1D subbands is very weak in the current-versus-gate-
voltage plot, but a convincingly regular oscillation is seen
if the derivative of the current with respect to the gate
voltage is taken (this quantity is called the transconduc-
tance). Warren et al. pointed out that the observation
of a quantum size eect in an array of 250 channels in-
dicates a rather remarkable uniformity of the width and
density of the individual channels.
More recently a similar experimental study was per-
formed by Ismail et al.
231
on 100 parallel channels dened
in the 2DEG of a GaAs-AlGaAs heterostructure. The
eects were found to be much more pronounced than in
the earlier experiment on Si inversion layer channels, pre-
sumably because of the much larger mean free path (es-
timated at 1 m), which was not much shorter than the
sample length (5 m). Quantum size eects in the quan-
tum ballistic transport regime (in particular, the con-
ductance quantization of a quantum point contact) are
discussed extensively in Section III.B. In a wide 2DEG
the minima of the Shubnikov-De Haas oscillations in the
magnetoresistance are periodic in 1/B, with a periodicity
(1/B) determined by the sheet density n
s
according to
Eq. (1.30). In a narrow channel one observes an increase
in (1/B) for weak magnetic elds because the electro-
static connement modies the density of states, as dis-
cussed in Section II.F.1. Such a deviation is of interest
as a manifestation of magnetoelectric subbands, but also
because it can be used to estimate the eective channel
width using the criterion W 2l
cycl
for the crossover
eld
167
B
crit
(the electron density in the channel, and
hence l
cycl
, may be estimated from the strong-eld pe-
riodicity). The phenomenon has been studied in many
publications.
36,56,57,74,79,167,217,218,223,232,233
As an illustration, we reproduce in Fig. 33a an experi-
mental magnetoresistance trace
167,218
obtained for a nar-
row (W 140 nm) GaAs-AlGaAs channel, dened using
a shallow-mesa etch.
63
The arrows indicate the magne-
toresistance minima thought to be associated with mag-
netic depopulation. The assignment becomes ambiguous
in weak magnetic elds, because of the presence of ape-
riodic conductance uctuations. Nevertheless, the devi-
ation from a straight line in the N versus B
1
plot in
Fig. 33b is suciently large to be reasonably convincing.
Also shown in Fig. 33b is the result of a t to a the-
oretical N(B) function (assuming a parabolic conning
potential and a B-independent electron density). The
parameter values found from this t for the width and
electron density are reasonable and agree with indepen-
dent estimates.
27
We have limited ourselves to a discussion of transport
studies, but wish to point out that 1D subbands have
been studied succesfully by capacitance
75
measurements
and by infrared
78
spectroscopy. As mentioned earlier, the
FIG. 33 (a) Magnetoresistance at 2.4 K of a narrow GaAs-
AlGaAs channel (as in Fig. 28). The arrows indicate magnetic
eld values assigned to the depopulation of magnetoelectric
subbands. (b) Subband index n N 1 versus inverse mag-
netic eld (crosses). The dashed line interpolates between
theoretical points for a parabolic conning potential (circles).
The electrostatic connement causes deviations from a linear
dependence of n on B
1
. Taken from K.-F. Berggren et al.,
Phys. Rev. B 37, 10118 (1988).
formation of 1D subbands also requires a reformulation
of the theories of weak localization and conductance uc-
tuations in the presence of boundary scattering. Weak
localization in the case of a small number of occupied
subbands has been studied by Tesanovic et al.
110,234
(in
a zero magnetic eld).
We will not discuss the subject of quantum size eects
further in this part of our review, since it has found more
striking manifestations in the ballistic transport regime
(the subject of Section III), where conductance uctua-
tions do not play a role. The most prominent example is
the conductance quantization of a point contact.
G. Periodic potential
1. Lateral superlattices
In a crystal, the periodic potential of the lattice opens
energy gaps of zero density of electronic states. An elec-
tron with energy in a gap is Bragg-reected and hence
cannot propagate through the crystal. Esaki and Tsu
235
proposed in 1970 that an articial energy gap might be
created by the epitaxial growth of alternating layers of
40
dierent semiconductors. In such a superlattice a peri-
odic potential of spacing a is superimposed on the crys-
tal lattice potential. Typically, a 10 nm is chosen to
be much larger than the crystal lattice spacing (0.5 nm),
leading to the formation of a large number of narrow
bands within the conduction band (minibands), sepa-
rated by small energy gaps (minigaps). Qualitatively
new transport properties may then be expected. For ex-
ample, the presence of minigaps may be revealed under
strong applied voltages by a negative dierential resis-
tance phenomenon predicted by Esaki and Tsu in their
original proposal and observed subsequently by Esaki and
Chang.
236,237
In contrast to a 3D crystal lattice, a super-
lattice formed by alternating layers is 1D. As a conse-
quence of the free motion in the plane of the layers, the
density of states is not zero in the minigaps, and electrons
may scatter between two overlapping minibands. Of in-
terest in the present context is the possibility of dening
lateral superlattices
238,239
by a periodic potential in the
plane of a 2D electron gas. True minigaps of zero den-
sity of states may form in such a system if the potential
varies periodically in two directions. Lateral superlattice
eects may be studied in the linear-response regime of
small applied voltages (to which we limit the discussion
here) by varying E
F
or the strength of the periodic po-
tential by means of a gate voltage. The conductivity is
expected to vanish if E
F
is in a true minigap (so that
electrons are Bragg-reected). Calculations
240,241
show
pronounced minima also in the case of a 1D periodic po-
tential.
The conditions required to observe the minibands in
a lateral superlattice are similar to those discussed in
Section II.F for the observation of 1D subbands in a nar-
row channel. The mean free path should be larger than
the lattice constant a, and 4k
B
T should be less than the
width of a minigap near the Fermi level. For a weak
periodic potential,
94
the nth minigap is approximately
E
n
2V
n
, with V
n
the amplitude of the Fourier com-
ponent of the potential at wave number k
n
= 2n/a. The
gap is centered at energy E
n
( hk
n
/2)
2
/2m. If we con-
sider, for example, a 1D sinusoidal potential V (x, y) =
V
0
sin(2y/a), then the rst energy gap E
1
V
0
oc-
curs at E
1
( h/a)
2
/2m. (Higher-order minigaps are
much smaller.) Bragg reection occurs when E
1
E
F
(i.e., for a lattice periodicity a
F
/2). Such a short-
period modulation is not easy to achieve lithographically,
however (typically
F
= 40 nm), and the experiments on
lateral superlattices discussed later are not in this regime.
Warren et al.
242
have observed a weak but regular
structure in the conductance of a 1D lateral superlat-
tice with a = 0.2 m dened in a Si inversion layer (us-
ing the dual-gate arrangement of Fig. 2c). Ismail et al.
62
used a grating-shaped gate on top of a GaAs-AlGaAs het-
erostructure to dene a lateral superiattice. A schematic
cross section of their device is shown in Fig. 34. The pe-
riod of the grating is 0.2 m. One eect of the gate volt-
age is to change the overall carrier concentration, leading
to a large but essentially smooth conductance variation
FIG. 34 Grating gate (in black) on top of a GaAs-AlGaAs
heterostructure, used to dene a 2DEG with a periodic den-
sity modulation. Taken from K. Ismail et al., Appl. Phys.
Lett. 52, 1071 (1988).
(at 4.2 K). This variation proved to be essentially the
same as that found for a continuous gate. As in the exper-
iment by Warren et al., the transconductance as a func-
tion of the voltage on the grating revealed a regular oscil-
lation. As an example, we reproduce the results of Ismail
et al. (for various source-drain voltages) in Fig. 35. No
such structure was found for control devices with a con-
tinuous, rather than a grating, gate. The observed struc-
ture is attributed to Bragg reection in Ref.
62
. A 2D lat-
eral superlattice was dened by Bernstein and Ferry,
243
using a grid-shaped gate, but the transport properties
in the linear response regime were not studied in detail.
Smith et al.
244
have used the split-gate technique to de-
ne a 2D array of 4000 dots in a high-mobility GaAs-
AlGaAs heterostructure (a = 0.5 m, 1 = 10 m). When
the 2DEG under the dots is depleted, a grid of conduct-
ing channels is formed. In this experiment the amplitude
of the periodic potential exceeds E
F
. Structure in the
conductance is found related to the depopulation of 1D
subbands in the channels, as well as to standing waves be-
tween the dots. The analysis is thus considerably more
complicated than it would be for a weak periodic po-
tential. It becomes dicult to distinguish between the
eects due to quantum interference within a single unit
cell of the periodic potential and the eects due to the
formation of minibands requiring phase coherence over
several unit cells. Devices with a 2D periodic potential
with a period comparable to the Fermi wavelength and
much shorter than the mean free path will be required for
the realization of true miniband eects. It appears that
the fabrication of such devices will have to await further
developments in the art of making nanostructures. Epi-
41
FIG. 35 Transconductance gm I/VSD of the device of
Fig. 34 measured as a function of gate voltage for various val-
ues of the source-drain voltage. The oscillations, seen in par-
ticular at low source-drain voltages, are attributed to Bragg
reection in a periodic potential. Taken from K. Ismail et al.,
Appl. Phys. Lett. 52, 1071 (1988).
taxy on tilted surfaces with a staircase surface structure
is being investigated for this purpose.
87,88,169,179,245,246
Nonepitaxial growth on Si surfaces slightly tilted from
(100) is known to lead to miniband formation in the in-
version layer.
20,247
A nal interesting possibility is to use
doping quantum wires, as proposed in Ref.
248
.
As mentioned, it is rather dicult to discriminate ex-
perimentally between true miniband eects and quan-
tum interference eects occurring within one unit cell.
The reason is that both phenomena give rise to struc-
ture in the conductance as a function of gate voltage
with essentially the same periodicity. This diculty may
be circumvented by studying lateral superlattices with
a small number of unit cells. The miniband for a nite
superlattice with P unit cells consists of a group of P
discrete states, which merge into a continuous miniband
in the limit P . The discrete states give rise to
closely spaced resonances in the transmission probability
through the superlattice as a function of energy, and may
thus be observed as a series of P peaks in the conductance
as a function of gate voltage, separated by broad minima
due to the minigaps. Such an observation would demon-
strate phase coherence over the entire length L = Pa
of the nite superlattice and would constitute conclu-
sive evidence of a miniband. The conductance of a nite
1D superlattice in a narrow 2DEG channel in the ballis-
tic transport regime has been investigated theoretically
by Ulloa et al.
249
Similar physics may be studied in the
quantum Hall eect regime, where the experimental re-
quirements are considerably relaxed. A successful exper-
iment of this type was recently performed by Kouwen-
hoven et al.
250
(see Section IV.E).
Weak-eld magnetotransport in a 2D periodic poten-
tial (a grid) has been studied by Ferry et al.
251,252
and by
Smith et al.
244
Both groups reported oscillatory structure
in the magnetoconductance, suggestive of an Aharonov-
Bohm eect with periodicity B = h/eS, where S is the
area of a unit cell of the lattice. In strong magnetic
elds no such oscillations are found. A similar suppres-
sion of the Aharonov-Bohm eect in strong elds is found
in single rings, as discussed in detail in Section IV.D.1.
Magnetotransport in a 1D periodic potential is the sub-
ject of the next subsection.
2. Guiding-center-drift resonance
The inuence of a magnetic eld on transport through
layered superlattices
253
has been studied mainly in the
regime where the (rst) energy gap E 100 meV ex-
ceeds the Landau level spacing h
c
(1.7 meV/T in GaAs).
The magnetic eld does not easily induce transitions be-
tween dierent minibands in this regime. Magnetotrans-
port through lateral superlattices is often in the opposite
regime h
c
E, because of the relatively large period-
icity (a 300 nm) and small amplitude (V
0
1 meV) of
the periodic potential. The magnetic eld now changes
qualitatively the structure of the energy bands, which
becomes richly complex in the case of a 2D periodic
potential.
54
Much of this structure, however, is not easily
observed, and the experiments discussed in this subsec-
tion involve mostly the classical eect of a weak periodic
potential on motion in a magnetic eld.
Weiss et al.
255,256
used an ingenious technique to im-
pose a weak periodic potential on a 2DEG in a GaAs-
AlGaAs heterostructure. They exploit the well-known
persistent ionization of donors in AlGaAs after brief illu-
mination at low temperatures. For the illumination, two
interfering laser beams are used, which generate an in-
terference pattern with a period depending on the wave-
length and on the angle of incidence of the two beams.
This technique, known as holographic illumination, is il-
lustrated in Fig. 36. The interference pattern selectively
ionizes Si donors in the AlGaAs, leading to a weak pe-
riodic modulation V (y) of the bottom of the conduc-
tion band in the 2DEG, which persists at low temper-
atures if the sample is kept in the dark. The sample
layout, also shown in Fig. 36, allows independent mea-
surements of the resistivity
yy
(

), perpendicular to,
and
xx
(
||
) parallel to the grating. In Fig. 37 we
show experimental results of Weiss et al.
255
for the mag-
netoresistivity of a 1D lateral superlattice (a = 382 nm).
In a zero magnetic eld, the resistivity tensor is ap-
proximately isotropic:

and
||
are indistinguishable
experimentally (see Fig. 37). This indicates that the am-
plitude of V (y) is much smaller than the Fermi energy
E
F
= 11 meV. On application of a small magnetic eld
B(
<

0.4 T) perpendicular to the 2DEG, a large oscilla-

tion periodic in 1/B develops in the resistivity

for
current owing perpendicular to the potential grating.
The resistivity is now strongly anisotropic, showing only
weak oscillations in
||
(current parallel to the poten-
42
FIG. 36 A brief illumination of a GaAs-AlGaAs heterostruc-
ture with an interference pattern due to two laser beams
(black arrows) leads to a persistent periodic variation in the
concentration of ionized donors in the AlGaAs, thereby im-
posing a weak periodic potential on the 2DEG. The resulting
spatial variation of the electron density in the 2DEG is indi-
cated schematically. (b) Experimental arrangement used to
produce a modulated 2DEG by means of the holographic
illumination of (a). The sample layout shown allows mea-
surements of the resistivity parallel and perpendicular to the
equipotentials. Taken from D. Weiss et al., in High Mag-
netic Fields in Semiconductor Physics II (G. Landwehr, ed.).
Springer, Berlin, 1989.
tial grating). In appearance, the oscillations resemble
the Shubnikov-De Haas oscillations at higher elds, but
their dierent periodicity and much weaker temperature
dependence point to a dierent origin.
The eect was not anticipated theoretically, but now
a fairly complete and consistent theoretical picture
has emerged from several analyses.
111,227,257,258,259
The
strong oscillations in

result from a resonance

111
be-
tween the periodic cyclotron orbit motion and the oscil-
lating EB drift of the orbit center induced by the elec-
FIG. 37 Solid curves: Magnetic eld dependence of the re-
sistivity

for current owing perpendicular to a potential

grating. The experimental curve is the measurement of Weiss
et al.
255
the theoretical curve follows from the guiding-center-
drift resonance. Note the phase shift of the oscillations, indi-
cated by the arrows at integer 2l
cycl
/a. The potential grating
has periodicity a = 382 nm and is modeled by a sinusoidal
potential with root-mean-square amplitude of = 1.5% of
the Fermi energy; The mean free path in the 2DEG is 12 m,
much larger than a. The dash-dotted curve is the experimen-
tal resistivity
||
for current owing parallel to the potential
grating, as measured by Weiss et al. Taken from C. W. J.
Beenakker, Phys. Rev. Lett. 62, 2020 (1989).
tric eld E V . Such guiding-center-drift resonances
are known from plasma physics,
260
and the experiment
by Weiss et al. appears to be the rst observation of this
phenomenon in the solid state. Magnetic quantization is
not essential for these strong oscillations, but plays a role
in the transition to the Shubnikov-De Haas oscillations
at higher elds and in the weak oscillations in
||
.
227,259
A simplied physical picture of the guiding-center-drift
resonance can be obtained as follows.
111
The guiding center (X, Y ) of an electron at position
(x, y) having velocity (v
x
, v
y
) is given by X = xv
y
/
c
,
Y = y +v
x
/
c
. The time derivative of the guiding center
is x = E(y)/B,

Y = 0, so its motion is parallel to the x-
axis. This is the EB drift. In the case of a strong mag-
netic eld and a slowly varying potential (l
cycl
a), one
may approximate E(y) E(Y ) to close the equations for

X and

Y . This so-called adiabatic approximation cannot
be made in the weak-eld regime (l
cycl
>

a) of interest
here. We consider the case of a weak potential, such that
eV
rms
/E
F
1, with V
rms
the root mean square of
V (y). The guiding center drift in the x-direction is then
simply superimposed on the unperturbed cyclotron mo-
tion. Its time average v
drift
is obtained by integrating the
43
FIG. 38 (a) Potential grating with a cyclotron orbit super-
imposed. When the electron is close to the two extremal
points Y l
cycl
, the guiding center at Y acquires an E B
drift in the direction of the arrows. (The drift along nonex-
tremal parts of the orbit averages out, approximately.) A
resonance occurs if the drift at one extremal point reinforces
the drift at the other, as shown. (b) Numerically calculated
trajectories for a sinusoidal potential ( = 0.015). The hori-
zontal lines are equipotentials at integer y/a. On resonance
(2l
cycl
/a = 6.25) the guiding center drift is maximal; o reso-
nance (2
lcycl
/a = 5.75) the drift is negligible. Taken from C.
W. J. Beenakker, Phys. Rev. Lett. 62, 2020 (1989).
electric eld along the orbit
v
drift
(Y ) = (2B)
1
_
2
0
dE(Y +l
cycl
sin ). (2.63)
For l
cycl
a the eld oscillates rapidly, so only the drift
acquired close to the two extremal points Y l
cycl
does
not average out. It follows that v
drift
is large or small
depending on whether E(Y +l
cycl
) and E(Y l
cycl
) have
the same sign or opposite sign (see Fig. 38). For a si-
nusoidal potential V (y) =

2V
rms
sin(2y/a), one easily
calculates by averaging over Y that, for l
cycl
a, the
mean square drift is
v
2
drift
) = (v
F
)
2
_
l
cycl
a
_
cos
2
_
2l
cycl
a


4
_
. (2.64)
The guiding center drift by itself leads, for l
cycl
l, to
1D diusion with diusion coecient D given by
D =
_

0
v
2
drift
)e
t/t
dt = v
2
drift
). (2.65)
The term D is an additional contribution to the xx-
element of the unperturbed diusion tensor D
0
given by
D
0
xx
= D
0
yy
= D
0
, D
0
xy
= D
0
xy
=
c
D
0
, with D
0

1
2
v
2
F
[1 + (
c
)
2
]
1
(cf. Section I.D.3). At this point we
assume that for l
cycl
l the contribution D from the
guiding center drift is the dominant eect of the potential
grating on the diusion tensor D. A justication of this
assumption requires a more systematic analysis of the
transport problem, which is given in Ref.
111
. Once D is
known, the resistivity tensor follows from the Einstein
relation = D
1
/e
2
(E
F
), with (E
F
) the 2D density
of states (cf. Section I.D.2). Because of the large o-
diagonal components of D
0
, an additional contribution
D to D
0
xx
modies predominantly
yy

. To leading
order in , one nds that

0
= 1 + 2
2
_
l
2
al
cycl
_
cos
2
_
2
l
cycl
a


4
_
, (2.66)
with
0
= m/n
s
e
2
the unperturbed resistivity. A rig-
orous solution
111
of the Boltzmann equation (for a B-
independent scattering time) conrms this simple result
in the regime a l
cycl
l and is shown in Fig. 37 to
be in quite good agreement with the experimental data
of Weiss et al.
255
Similar theoretical results have been
obtained by Gerhardts et al.
257
and by Winkler et al.
258
(using an equivalent quantum mechanical formulation;
see below).
As illustrated by the arrows in Fig. 37, the maxima
in

are not at integer 2l

cycl
/a, but shifted somewhat
toward lower magnetic elds. This phase shift is a conse-
quence of the nite extension of the segment of the orbit
around the extremal points Y l
cycl
, which contributes
to the guiding center drift v
drift
(Y ). Equation (2.66) im-
plies that

in a sinusoidal potential grating has minima

and maxima at approximately
2l
cycl
/a (minima) = n
1
4
,
2l
cycl
/a (maxima) = n +
1
4
order(1/n), (2.67)
with n an integer. We emphasize that the phase shift
is not universal, but depends on the functional form of
V (y). The fact that the experimental phase shift in Fig.
37 agrees so well with the theory indicates that the ac-
tual potential grating in the experiment of Weiss et al.
is well modeled by a sinusoidal potential. The maxima
in

/
0
have amplitude
2
(l
2
/al
cycl
), which for a large
mean free path l can be of order unity, even if 1.
The guiding-center-drift resonance thus explains the sur-
prising experimental nding that a periodic modulation
of the Fermi velocity of order 10
2
can double the resis-
tivity.
At low magnetic elds the experimental oscillations are
damped more rapidly than the theory would predict, and,
moreover, an unexplained positive magnetoresistance is
observed around zero eld in

(but not in
||
). Part of
this disagreement may be due to nonuniformities in the
potential grating, which become especially important at
44
low elds when the cyclotron orbit overlaps many mod-
ulation periods. At high magnetic elds B
>

0.4 T the
experimental data show the onset of Shubnikov-De Haas
oscillations, which are a consequence of oscillations in the
scattering time due to Landau level quantization (cf.
Section I.D.3). This eect is neglected in the semiclassi-
cal analysis, which assumes a constant scattering time.
The quantum mechanical B-dependence of also leads
to weak-eld oscillations in
||
, with the same periodic-
ity as the oscillations in

discussed earlier, but of much

smaller amplitude and shifted in phase (see Fig. 37, where
a maximum in the experimental
||
around 0.3 T lines up
with a minimum in

). These small antiphase oscilla-

tions in
||
were explained by Vasilopoulos and Peeters
227
and by Gerhardts and Zhang
259
as resulting from oscilla-
tions in due to the oscillatory Landau bandwidth. The
Landau levels E
n
= (n
1
2
) h
c
broaden into a band of
nite width in a periodic potential.
261
This Landau band
is described by a dispersion law E
n
(k), where the wave
number k is related to the classical orbit center (X, Y )
by k = Y eB/h (cf. the similar relation in Section III.A).
The classical guiding-center-drift resonance can also be
explained in these quantum mechanical terms, if one so
desires, by noticing that the bandwidth of the Landau
levels is proportional to the root-mean-square average of
v
drift
= dE
n
(k)/hdk. A maximal bandwidth thus corre-
sponds to a maximal guiding center drift and, hence, to a
maximal

. A maximum in the bandwidth also implies

a minimum in the density of states at the Fermi level
and, hence, a maximum in [Eq. (1.29)]. A maximal
bandwidth thus corresponds to a minimal
||
, whereas
the B-dependence of can safely by neglected for the
oscillations in

(which are dominated by the classical

guiding-center-drift resonance).
In a 2D periodic potential (a grid), the guiding cen-
ter drift dominates the magnetoresistivity in both di-
agonal components of the resistivity tensor. Classi-
cally, the eect of a weak periodic potential V (x, y)
on
xx
and
yy
decouples if V (x, y) is separable into
V (x, y) = f(x) + g(y). For the 2D sinusoidal potential
V (x, y) sin(2x/a) +sin(2y/b), one nds that the ef-
fect of the grid is simply a superposition of the eects
for two perpendicular gratings of periods a and b. (No
such decoupling occurs quantum mechanically.
254
) Ex-
periments by Alves et al.
262
and by Weiss et al.
263
con-
rm this expectation, except for a disagreement in the
phase of the oscillations. As noted, however, the phase
is not universal but depends on the form of the periodic
potential, which need not be sinusoidal.
Because of the predominance of the classical guiding-
center-drift resonance in a weak periodic potential, mag-
netotransport experiments are not well suited to study
miniband structure in the density of states. Magne-
tocapacitance measurements
256,264,265
are a more direct
means of investigation, but somewhat outside the scope
of this review.
III. BALLISTIC TRANSPORT
A. Conduction as a transmission problem
In the ballistic transport regime, it is the scattering
of electrons at the sample boundaries which limits the
current, rather than impurity scattering. The canoni-
cal example of a ballistic conductor is the point contact
illustrated in Fig. 7c. The current I through the nar-
row constriction in response to a voltage dierence V
between the wide regions to the left and right is nite
even in the absence of impurities, because electrons are
scattered back at the entrance of the constriction. The
contact conductance G = I/V is proportional to the con-
striction width but independent of its length. One cannot
therefore describe the contact conductance in terms of a
local conductivity, as one can do in the diusive trans-
port regime. Consequently, the Einstein relation (1.11)
between the conductivity and the diusion constant at
the Fermi level, of which we made use repeatedly in Sec-
tion II, is not applicable in that form to determine the
contact conductance. The Landauer formula is an al-
ternative relation between the conductance and a Fermi
level property of the sample, without the restriction to
diusive transport. We discuss this formulation of con-
duction in Section III.A.2. The Landauer formula ex-
presses the conductance in terms of transmission prob-
abilities of propagating modes at the Fermi level (also
referred to as quantum channels in this context). Some
elementary properties of the modes are summarized in
Section III.A.1.
1. Electron waveguide
We consider a conducting channel in a 2DEG (an elec-
tron waveguide), dened by a lateral conning potential
V (x), in the presence of a perpendicular magnetic eld B
(in the z-direction). In the Landau gauge A = (0, Bx, 0)
the hamiltonian has the form
H =
p
2
x
2m
+
(p
y
+eBx)
2
2m
+V (x) (3.1)
for a single spin component (cf. Section II.F.1). Be-
cause the canonical momentum p
y
along the channel
commutes with H, one can diagonalize p
y
and H simul-
taneously. For each eigenvalue hk of p
y
, the hamilto-
nian (3.1) has a discrete spectrum of energy eigenvalues
E
n
(k), n = 1, 2, . . ., corresponding to eigenfunctions of
the form
[n, k) =
n,k
(x)e
iky
. (3.2)
In waveguide terminology, the index n labels the modes,
and the dependence of the energy (or frequency) E
n
(k)
on the wave number k is the dispersion relation of the nth
mode. A propagating mode at the Fermi level has cuto
frequency E
n
(0) below E
F
. The wave function (3.2) is the
product of a transverse amplitude prole
n,k
(x) and a
45
FIG. 39 Energy-orbit center phase space. The two parabolas
divide the space into four regions, which correspond to dier-
ent types of classical trajectories in a magnetic eld (clockwise
from left: skipping orbits on one edge, traversing trajectories,
skipping orbits on the other edge, and cyclotron orbits). The
shaded region is forbidden. The region at the upper cen-
ter contains traversing trajectories moving in both directions,
but only one region is shown for clarity. Taken from C. W.
J. Beenakker et al., Superlattices and Microstructures 5, 127
(1989).
longitudinal plane wave e
iky
. The average velocity v
n
(k)
along the channel in state [n, k) is the expectation value
of the y-component of the velocity operator p +eA:
v
n
(k) n, k[
p
y
+eA
y
m
[n, k)
= n, k[
H
p
y
[n, k) =
dE
n
(k)
hdk
. (3.3)
For a zero magnetic eld, the dispersion relation E
n
(k)
has the simple form (1.4). The group velocity v
n
(k) is
then simply equal to the velocity hk/m obtained from
the canonical momentum. This equality no longer holds
in the presence of a magnetic eld, because the canonical
momentum contains an extra contribution from the vec-
tor potential. The dispersion relation in a nonzero mag-
netic eld was derived in Section II.F.1 for a parabolic
conning potential V (x) =
1
2
m
2
0
x
2
. From Eq. (2.59)
one calculates a group velocity hk/M that is smaller than
hk/m by a factor of 1 + (
c
/
0
)
2
.
Insight into the nature of the wave functions in a mag-
netic eld can be obtained from the correspondence with
classical trajectories. These are most easily visualized in
a square-well conning potential, as we now discuss (fol-
lowing Ref.
266
). The position (x, y) of an electron on the
circle with center coordinates (X, Y ) can be expressed in
terms of its velocity v by
x = X +v
y
/
c
, y = Y v
x
/
c
, (3.4)
with
c
eB/m the cyclotron frequency. The cyclotron
radius is (2mE)
1/2
/eB, with E
1
2
mv
2
the energy of
the electron. Both the energy E and the separation X
of the orbit center from the center of the channel are
constants of the motion. The coordinate Y of the or-
bit center parallel to the channel walls changes on each
specular reection. One can classify a trajectory as a
cyclotron orbit, skipping orbit, or traversing trajectory,
depending on whether the trajectory collides with zero,
one, or both channel walls. In (X, E) space these three
types of trajectories are separated by the two parabolas
(X W/2)
2
= 2mE(eB)
2
(Fig. 39). The quantum me-
chanical dispersion relation E
n
(k) can be drawn into this
classical phase diagram because of the correspondence
k = XeB/h.This correspondence exists because both
k and X are constants of the motion and it follows from
the fact that the component hk along the channel of the
canonical momentum p = mv eA equals
hk = mv
y
eA
y
= mv
y
eBx = eBx (3.5)
in the Landau gauge.
In Fig. 40 we show E
n
(k) both in weak and in strong
magnetic elds, calculated
266
for typical parameter val-
ues from the Bohr-Sommerfeld quantization rule dis-
cussed here. The regions in phase space occupied by clas-
sical skipping orbits are shaded. The unshaded regions
contain cyclotron orbits (at small E) and traversing tra-
jectories (at larger E) (cf. Fig. 39). The cyclotron orbits
correspond quantum mechanically to states in Landau
levels. These are the at portions of the dispersion re-
lation at energies E
n
= (n
1
2
) h
c
. The group velocity
(3.3) is zero in a Landau level, as one would expect from
the correspondence with a circular orbit. The travers-
ing trajectories correspond to states in magnetoelectric
subbands, which interact with both the opposite channel
boundaries and have a nonzero group velocity. The skip-
ping orbits correspond to edge states, which interact with
a single boundary only. The two sets of edge states (one
for each boundary) are disjunct in (k, E) space. Edge
states at opposite boundaries move in opposite directions,
as is evident from the correspondence with skipping or-
bits or by inspection of the slope of E
n
(k) in the two
shaded regions in Fig. 40.
If the Fermi level lies between two Landau levels, the
states at the Fermi level consist only of edge states if
B > B
crit
, as in Fig. 40b. The critical eld B
crit
=
2 hk
F
/eW is obtained from the classical correspondence
by requiring that the channel width W should be larger
than the cyclotron diameter 2 hk
F
/eB at the Fermi level.
This is the same characteristic eld that played a role
in the discussion of magneto size eects in Sections II.A
and II.F. At elds B < B
crit
, as in Fig. 40a, edge states
coexist at the Fermi level with magnetoelectric subbands.
In still lower elds B < B
thres
all states at the Fermi level
interact with both edges. The criterion for this is that W
should be smaller than the transverse wavelength
267

t
=
( h/2k
F
eB)
1/3
of the edge states, so the threshold eld
B
thres
h/ek
F
W
3
. Contrary to initial expectations,
268
46
FIG. 40 Dispersion relation En(k), calculated for parame-
ters: (a) W = 100 nm, B = 1 T; (b) W = 200 nm, B = 1.5 T.
The horizontal line at 17 meV indicates the Fermi energy.
The shaded area is the region of classical skipping orbits and
is bounded by the two parabolas shown in Fig. 39 (with the
correspondence k = XeB/h). Note that in (a) edge states
coexist at the Fermi level with states interacting with both
boundaries (B < Bcrit 2 hkF/eB), while in (b) all states at
the Fermi level interact with one boundary only (B > Bcrit).
Taken from C. W. J. Beenakker et al., Superlattices and Mi-
crostructures 5, 127 (1989).
this lower characteristic eld does not appear to play a
decisive role in magneto size eects.
A quick way to arrive at the dispersion relation E
n
(k),
which is suciently accurate for our purposes, is to ap-
ply the Bohr-Sommerfeld quantization rule
80,269
to the
classical motion in the x-direction:
1
h
_
p
x
dx + = 2n, n = 1, 2, . . . . (3.6)
The integral is over one period of the motion. The phase
shift is the sum of the phase shifts acquired at the two
turning points of the projection of the motion on the x-
axis. The phase shift upon reection at the boundary
is , to ensure that incident and reected waves cancel
(we consider an innite barrier potential at which the
FIG. 41 Classical trajectories in a magnetic eld. The ux
through the shaded area is quantized according to the Bohr-
Sommerfeld quantization rule (3.7). The shaded area in (b)
is bounded by the channel walls and the circle formed by the
continuation (dashed) of one circular arc of the traversing
trajectory.
wave function vanishes). The other turning points (at
which v
x
varies smoothly) have a phase shift of /2.
93
Consequently, for a traversing trajectory = + = 0
(mod 2), for a skipping orbit = /2 = /2, and
for a cyclotron orbit = /2 /2 = (mod 2). In
the Landau gauge one has p
x
= mv
x
= eB(Y y), so
Eq. (3.6) takes the form
B
_
(Y y)dx =
h
e
(n

2
). (3.7)
This quantization condition has the appealing geomet-
rical interpretation that n /2 ux quanta h/e are
contained in the area bounded by the channel walls and
a circle of cyclotron radius (2mE)
1/2
/eB centered at X
(cf. Fig. 41). It is now straightforward to nd for each in-
teger n and coordinate X the energy E that satises this
condition. The dispersion relation E
n
(k) then follows on
identifying k = XeB/h, as shown in Fig. 40.
The total number N of propagating modes at energy E
is determined by the maximum ux
max
contained in an
area bounded by the channel walls and a circle of radius
(2mE)
1/2
/eB, according to N = Int[e
max
/h + /2].
Note that a maximal enclosed ux is obtained by center-
ing the circle on the channel axis. Some simple geometry
then leads to the result
80
(2.62), which is plotted to-
gether with that for a parabolic connement in Fig. 31.
Equation (2.62) has a discontinuity at magnetic elds for
which the cyclotron diameter equals the channel width,
due to the jump in the phase shift as one goes from a
cyclotron orbit to a traversing trajectory. This jump is
an artifact of the present semiclassical approximation in
which the extension of the wave function beyond the clas-
sical orbit is ignored. Since the discontinuity in N is at
most 1, it is unimportant in many applications. More
accurate formulas for the phase shift , which smooth out
the discontinuity, have been derived in Ref.
270
. If neces-
sary, one can also use more complicated but exact solu-
tions of the Schrodinger equation, which are known.
267
47
FIG. 42 (a) Narrow channel connecting two wide electron
gas regions, having a chemical potential dierence . (b)
Schematic dispersion relation in the narrow channel. Left-
moving states (k > 0) are lled up to chemical potential EF,
right-moving states up to EF + (solid dots). Higher-lying
states are unoccupied (open dots).
2. Landauer formula
Imagine two wide electron gas reservoirs having a slight
dierence n in electron density, which are brought into
contact by means of a narrow channel, as in Fig. 42a. A
diusion current J will ow in the channel, carried by
electrons with energies between the Fermi energies E
F
and E
F
+ in the low- and high-density regions. For
small n, one can write for the Fermi energy dierence
(or chemical potential dierence) = n/(E
F
), with
(E
F
) the density of states at E
F
in the reservoir (cf. Sec-
tion I.D.1). The diusion constant (or diusance)
4

D
is dened by J =

Dn and is related to the conductance
G by
G = e
2
(E
F
)

D. (3.8)
Equation (3.8) generalizes the Einstein relation (1.11)
and is derived in a completely analogous way [by re-
quiring that the sum of drift current GV/e and diusion
current

Dn be zero when the sum of the electrostatic
potential dierence eV and chemical potential dierence
n/(E
F
) vanishes].
Since the diusion current (at low temperatures) is car-
ried by electrons within a narrow range above E
F
,
the diusance can be expressed in terms of Fermi level
properties of the channel (see below). The Einstein rela-
tion (3.8) then yields the required Fermi level expression
of the conductance. This by no means implies that the
drift current induced by an electrostatic potential dier-
ence is carried entirely by electrons at the Fermi energy.
To the contrary, all electrons (regardless of their energy)
acquire a nonzero drift velocity in an electric eld. This
point has been the cause of some confusion in the litera-
ture on the quantum Hall eect, so we will return to it in
Section IV.A.3. In the following we will refer to electrons
at the Fermi energy as current-carrying electrons and
show that the current in the channel is shared equally
among the modes at the Fermi level. These and similar
statements should be interpreted as referring to the dif-
fusion problem, where the current is induced by density
dierences without an electric eld. We make no attempt
here to evaluate the distribution of current in response
to an electric eld in a system of uniform density. That
is a dicult problem, for which one has to determine
the electric eld distribution self-consistently from Pois-
sons and Boltzmanns equations. Such a calculation for
a quantum point contact has been performed in Refs.
271
and
272
. Fortunately, there is no need to know the ac-
tual current distribution to determine the conductance,
in view of the Einstein relation (3.8). The distribution
of current (and electric eld) is of importance only be-
yond the regime of a linear relation between current and
voltage. We will not venture beyond this linear response
regime.
To calculate the diusance, we rst consider the case
of an ideal electron waveguide between the two reser-
voirs. By ideal it is meant that within the waveguide
the states with group velocity pointing to the right are
occupied up to E
F
+, while states with group velocity
to the left are occupied up to E
F
and empty above that
energy (cf. Fig. 42b). This requires that an electron near
the Fermi energy that is injected into the waveguide by
the reservoir at E
F
+ propagates into the other reser-
voir without being reected. (The physical requirements
for this to happen will be discussed in Section III.B.) The
amount of diusion current per energy interval carried by
the right-moving states (with k < 0) in a mode n is the
product of density of states

n
and group velocity v
n
.
Using Eqs. (1.5) and (3.3), we nd the total current J
n
carried by that mode to be
J
n
=
_
EF+
EF
g
s
g
v
_
2
dE
n
(k)
dk
_
1
dE
n
(k)
hdk
=
g
s
g
v
h
, (3.9)
independent of mode index and Fermi energy. The cur-
rent in the channel is shared equally among the N modes
at the Fermi level because of the cancellation of group ve-
locity and density states. We will return to this equipar-
tition rule in Section III.B, because it is at the origin of
the quantization
6,7
of the conductance of a point contact.
Scattering within the narrow channel may reect part
of the injected current back into the left reservoir. If a
fraction T
n
of J
n
is transmitted to the reservoir at the
right, then the total diusion current in the channel be-
comes J = (2/h)

N
n=1
T
n
. (Unless stated otherwise,
the formulas in the remainder of this review refer to the
case g
s
= 2, g
v
= 1 of twofold spin degeneracy and a
single valley, appropriate for most of the experiments.)
48
FIG. 43 Generalization of the geometry of Fig. 42a to a
multireservoir conductor.
Using = n/(E
F
), J =

Dn, and the Einstein rela-
tion (3.8), one obtains the result
G =
2e
2
h
N

n=1
T
n
, (3.10a)
which can also be written in the form
G =
2e
2
h
N

n,m=1
[t
mn
[
2

2e
2
h
Tr tt

, (3.10b)
where T
n
=

N
m=1
[t
mn
[
2
is expressed in terms of the ma-
trix t of transmission probability amplitudes from mode
n to mode m. This relation between conductance and
transmission probabilities at the Fermi energy is referred
to as the Landauer formula because of Landauers pio-
neering 1957 paper.
4
Derivations of Eq. (3.10) based on
the Kubo formula of linear response theory have been
given by several authors, both for zero
143,273,274
and
nonzero
275,276
magnetic elds. The identication of G as
a contact conductance is due to Imry.
1
In earlier work Eq.
(3.10) was considered suspect
228,277,279
because it gives
a nite conductance for an ideal (ballistic) conductor,
and alternative expressions were proposed
188,280,281,282
that were considered to be more realistic. (In one di-
mension these alternative formulas reduce to the original
Landauer formula
4
G = (e
2
/h)T(1T)
1
, which gives in-
nite conductance for unit transmission since the contact
conductance e
2
/h is ignored.
1
) For historical accounts of
this controversy, from two dierent points of view, we
refer the interested reader to papers by Landauer
283
and
by Stone and Szafer.
274
We have briey mentioned this
now-settled controversy, because it sheds some light onto
why the quantization of the contact conductance had not
been predicted theoretically prior to its experimental dis-
covery in 1987.
Equation (3.10) refers to a two-terminal resistance
measurement, in which the same two contacts (modeled
by reservoirs in Fig. 42a) are used to drive a current
through the system and to measure the voltage drop.
More generally, one can consider a multireservoir con-
ductor as in Fig. 43 to model, for example, four-terminal
resistance measurements in which the current source and
drain are distinct from the voltage probes. The general-
ization of the Landauer formula (3.10) to multiterminal
resistances is due to B uttiker.
5
Let T

be the total
transmission probability from reservoir to :
T

=
N

n=1
N

m=1
[t
,mn
[
2
. (3.11)
Here N

is the number of propagating modes in the

channel (or lead) connected to reservoir (which in
general may be dierent from the number N

in lead
), and t
,mn
is the transmission probability ampli-
tude from mode n in lead to mode m in lead .
The leads are modeled by ideal electron waveguides, in
the sense discussed before, so that the reservoir at
chemical potential

above E
F
injects into lead a
(charge) current (2e/h)N

. A fraction T

/N

of
that current is transmitted to reservoir , and a fraction
T
a
/N

/N

is reected back into reservoir , be-

fore reaching one of the other reservoirs. The net current
I

in lead is thus given by

5
h
2e
I

= (N

(=)
T

. (3.12)
The chemical potentials of the reservoirs are related to
the currents in the leads via a matrix of transmission
and reection coecients. The sums of columns or rows
of this matrix vanish:
N

(=)
T

= 0, (3.13)
N

(=)
T

= 0. (3.14)
Equation (3.13) follows from current conservation, and
Eq. (3.14) follows from the requirement that an increase
of all the chemical potentials by the same amount should
have no eect on the net currents in the leads. Equa-
tion (3.12) can be applied to a measurement of the
four-terminal resistance R
,
= V/I, in which a cur-
rent I ows from contact to and a voltage dier-
ence V is measured between contacts and . Setting
I

= I = I

, and I

= 0 for

,= , , one can solve

the set of linear equations (3.12) to determine the chem-
ical potential dierence

. (Only the dierences in

chemical potentials can be obtained from the n equations
(3.12), which are not independent in view of Eq. (3.14).
By xing one chemical potential at zero, one reduces the
number of equations to n 1 independent ones.) The
four-terminal resistance R
,
= (

)/eI is then
obtained as a rational function of the transmission and
reection probabilities. We will refer to this procedure as
the Landauer-B uttiker formalism. It provides a unied
description of the whole variety of electrical transport
experiments discussed in this review.
The transmission probabilities have the symmetry
t
,nm
(B) = t
,mn
(B) T

(B) = T

(B).
(3.15)
49
Equation (3.15) follows by combining the unitarity of the
scattering matrix t

= t
1
, required by current con-
servation, with the symmetry t

(B) = t
1
(B), re-
quired by time-reversal invariance ( and denote com-
plex and Hermitian conjugation, respectively). As shown
by B uttiker,
5,284
the symmetry (3.15) of the coecients
in Eq. (3.12) implies a reciprocity relation for the four-
terminal resistance:
R
,
(B) = R
,
(B). (3.16)
The resistance is unchanged if current and voltage leads
are interchanged with simultaneous reversal of the mag-
netic eld direction. A special case of Eq. (3.16) is that
the two-terminal resistance R
,
is even in B. In the
diusive transport regime, the reciprocity relation for the
resistance follows from the Onsager-Casimir relation
285
(B) =
T
() for the resistivity tensor (T denotes the
transpose). Equation (3.16) holds also in cases that the
concept of a local resistivity breaks down. One experi-
mental demonstration
80
of the validity of the reciprocity
relation in the quantum ballistic transport regime will be
discussed in Section III.C. Other demonstrations have
been given in Refs.
286,287,288,289
. We emphasize that
strict reciprocity holds only in the linear response limit
of innitesimally small currents and voltages. Deviations
from Eq. (3.16) can occur experimentally
290
due to non-
linearities from quantum interference,
146,291
which in the
case of a long phase coherence time

persist down to
very small voltages V
>

h/e

. Magnetic impurities can

be another source of deviations from reciprocity if the
applied magnetic eld is not suciently strong to reverse
the magnetic moments on eld reversal. The large B
asymmetry of the two-terminal resistance of a point con-
tact reported in Ref.
292
has remained unexplained (see
Section IV.D).
The scattering matrix t in Eq. (3.15) describes elastic
scattering only. Inelastic scattering is assumed to take
place exclusively in the reservoirs. That is a reasonable
approximation at temperatures that are suciently low
that the size of the conductor is smaller than the inelastic
scattering length (or the phase coherence length if quan-
tum interference eects play a role). Reservoirs thus play
a dual role in the Landauer-B uttiker formalism: On the
one hand, a reservoir is a model for a current or volt-
age contact; on the other hand, a reservoir brings inelas-
tic scattering into the system. The reciprocity relation
(3.16) is unaected by adding reservoirs to the system
and is not restricted to elastic scattering.
284
More realis-
tic methods to include inelastic scattering in a distributed
way throughout the system have been proposed, but are
not yet implemented in an actual calculation.
293,294
B. Quantum point contacts
Many of the principal phenomena in ballistic trans-
port are exhibited in the cleanest and most extreme way
by quantum point contacts. These are short and nar-
row constrictions in a 2DEG, with a width of the order
of the Fermi wavelength.
6,7,59
The conductance of quan-
tum point contacts is quantized in units of 2e
2
/h. This
quantization is reminiscent of the quantization of the Hall
conductance, but is measured in the absence of a mag-
netic eld. The zero-eld conductance quantization and
the smooth transition to the quantum Hall eect on ap-
plying a magnetic eld are essentially consequences of
the equipartition of current among an integer number of
propagating modes in the constriction, each mode car-
rying a current of 2e
2
/h times the applied voltage V .
Deviations from precise quantization result from nonunit
transmission probability of propagating modes and from
nonzero transmission probability of evanescent (nonprop-
agating) modes. Experiment and theory in a zero mag-
netic eld are reviewed in Section III.B.1. The eect of
a magnetic eld is the subject of Section III.B.2, which
deals with depopulation of subbands and suppression of
backscattering by a magnetic eld, two phenomena that
form the basis for an understanding of magnetotransport
in semiconductor nanostructures.
1. Conductance quantization
(a) Experiments. The study of electron transport
through point contacts in metals has a long history, which
goes back to Maxwells investigations
295
of the resistance
of an orice in the diusive transport regime. Ballistic
transport was rst studied by Sharvin,
296
who proposed
and subsequently realized
297
the injection and detection
of a beam of electrons in a metal by means of point
contacts much smaller than the mean free path. With
the possible exception of the scanning tunneling micro-
scope, which can be seen as a point contact on an atomic
scale,
298,299,300,301,302,303
these studies in metals are es-
sentially restricted to the classical ballistic transport
regime because of the extremely small Fermi wavelength
(
F
0.5 nm). Point contacts in a 2DEG cannot be fab-
ricated by simply pressing two wedge- or needle-shaped
pieces of material together (as in bulk semiconductors
304
or metals
305
), since the electron gas is conned at the
GaAs-AlGaAs interface in the interior of the heterostruc-
ture. Instead, they are dened electrostatically
24,58
by
means of a split gate on top of the heterostructure (a
schematical cross-sectional view was given in Fig. 4b,
while the micrograph Fig. 5b shows a top view of the
split gate of a double-point contact device; see also the
inset in Fig. 44). In this way one can dene short
and narrow constrictions in the 2DEG, of variable width
0
<

W
<

250 nm comparable to the Fermi wavelength

F
40 nm and much shorter than the mean free path
l 10 m.
Van Wees et al.
6
and Wharam et al.
7
independently
discovered a sequence of steps in the conductance of such
a point contact as its width was varied by means of the
voltage on the split gate (see Fig. 44). The steps are near
50
FIG. 44 Point contact conductance as a function of gate volt-
age at 0.6 K, demonstrating the conductance quantization in
units of 2e
2
/h. The data are obtained from the two-terminal
resistance after subtraction of a background resistance. The
constriction width increases with increasing voltage on the
gate (see inset). Taken from B. J. van Wees et al., Phys. Rev.
Lett. 60, 848 (1988).
integer multiples of 2e
2
/h (13 k)
1
, after correction
for a gate-voltage-independent series resistance from the
wide 2DEG regions. An elementary explanation of this
eect relies on the fact that each 1D subband in the con-
striction contributes 2e
2
/h to the conductance because of
the cancellation of the group velocity and the 1D density
of states discussed in Section III.A. Since the number
N of occupied subbands is necessarily an integer, it fol-
lows from this simple argument that the conductance G
is quantized,
G = (2e
2
/h)N, (3.17)
as observed experimentally. A more complete explana-
tion requires an explicit treatment of the mode coupling
at the entrance and exit of the constriction, as discussed
later.
The zero-eld conductance quantization of a quantum
point contact is not as accurate as the Hall conductance
quantization in strong magnetic elds. The deviations
from exact quantization are typically
6,7,306
1%, while in
the quantum Hall eect one obtains routinely
97
an accu-
racy of 1 part in 10
7
. It is unlikely that a similar accu-
racy will be achieved in the case of the zero-eld quanti-
zation, one reason being the additive contribution to the
point contact resistance of a background resistance whose
magnitude cannot be determined precisely. The largest
part of this background resistance originates in the ohmic
contacts
307
and can thus be eliminated in a four-terminal
measurement of the contact resistance. The position of
the additional voltage probes on the wide 2DEG regions
has to be more than an inelastic scattering length away
from the point contact so that a local equilibrium is es-
tablished. A residual background resistance
307
of the or-
der of the resistance of a square is therefore unavoid-
able. In the experiments of Refs.
6
and
7
one has 20 ,
but lower values are possible for higher-mobility mate-
rial. It would be of interest to investigate experimentally
whether resistance plateaux quantized to such an accu-
racy are achievable. It should be noted, however, that
the degree of atness of the plateaux and the sharpness of
the steps in the present experiments vary among devices
of identical design, indicating that the detailed shape of
the electrostatic potential dening the constriction is im-
portant. There are many uncontrolled factors aecting
this shape, such as small changes in the gate geometry,
variations in the pinning of the Fermi level at the free
GaAs surface or at the interface with the gate metal,
doping inhomogeneities in the heterostructure material,
and trapping of charge in deep levels in AlGaAs.
On increasing the temperature, one nds experimen-
tally that the plateaux acquire a nite slope until they
are no longer resolved.
308
This is a consequence of the
thermal smearing of the Fermi-Dirac distribution (1.10).
If at T = 0 the conductance G(E
F
, T) has a step function
dependence on the Fermi energy E
F
, at nite tempera-
tures it has the form
309
G(E
F
, T) =
_

0
G(E, 0)
df
dE
F
dE
=
2e
2
h

n=1
f(E
n
E
F
). (3.18)
Here E
n
denotes the energy of the bottom of the nth sub-
band [cf. Eq. (1.4)]. The width of the thermal smearing
function df/dE
F
is about 4k
B
T, so the conductance steps
should disappear for T
>

E/4k
B
4 K (here E is the
subband splitting at the Fermi level). This is conrmed
both by experiment
308
and by numerical calculations (see
below).
Interestingly, it was found experimentally
6,7
that in
general a nite temperature yielded the most pronounced
and at plateaux as a function of gate voltage in the zero-
eld conductance. If the temperature is increased beyond
this optimum (which is about 0.5 K), the plateaux disap-
pear because of the thermal averaging discussed earlier.
Below this temperature, an oscillatory structure may be
superimposed on the conductance plateaux. This phe-
nomenon depends on the precise shape of the constric-
tion, as discussed later. A small but nite voltage drop
across the constriction has an eect that is qualitatively
similar to that of a nite temperature.
309
This is indeed
borne out by experiment.
308
(Experiments on conduc-
tion through quantum point contacts at larger applied
voltages in the nonlinear transport regime have been re-
viewed in Ref.
307
).
Theoretically, one would expect the conductance quan-
tization to be preserved in longer channels than those
used in the original experiment
6,7
(in which typically
L W 100 nm). Experiments on channels longer
than about 1 m did not show the quantization,
306,307,310
however, although their length was well below the trans-
port mean free path in the bulk (about 10 m). The
lack of clear plateaux in long constrictions is presum-
51
FIG. 45 (a) Classical ballistic transport through a point con-
tact induced by a concentration dierence n, or electrochem-
ical potential dierence eV , between source (s) and drain (d).
(b) The net current through a quantum point contact is car-
ried by the shaded region in k-space. In a narrow channel the
allowed states lie on the horizontal lines, which correspond
to quantized values for ky = n/W, and continuous values
for kx. The formation of these 1D subbands gives rise to a
quantized conductance. Taken from H. van Houten et al., in
Physics and Technology of Submicron Structures (H. Hein-
rich, G. Bauer, and F. Kuchar, eds.). Springer, Berlin, 1988;
and in Nanostructure Physics and Fabrication (M. Reed
and W. P. Kirk, eds.). Academic, New York, 1989.
ably due to enhanced backscattering inside the constric-
tion, either because of impurity scattering (which may be
enhanced
306,310
due to the reduced screening in a quasi-
one-dimensional electron gas
72
) or because of boundary
scattering at channel wall irregularities. As mentioned
in Section II.A, Thornton et al.
107
have found evidence
for a small (5%) fraction of diuse, rather than specu-
lar, reections at boundaries dened electrostatically by
a gate. In a 200-nm-wide constriction this leads to an
eective mean free path of about 200 nm/0.05 4 m,
comparable to the constriction length of devices that do
not exhibit the conductance quantization.
113,307
(b) Theory. It is instructive to rst consider clas-
sical 2D point contacts in some detail.
31,311
The ballis-
tic electron ow through a point contact is illustrated in
Fig. 45a in real space, and in Fig. 45b in k-space, for a
small excess electron density n at one side of the point
contact. At low temperatures this excess charge moves
with the Fermi velocity v
F
. The ux normally incident
on the point contact is nv
F
cos (cos )), where (x)
is the unit step function and the symbol ) denotes an
isotropic angular average (the angle is dened in Fig.
45a). In the ballistic limit l W the incident ux is fully
transmitted, so the total diusion current J through the
point contact is given by
J = Wnv
F
_
/2
/2
cos
d
2
=
1

Wv
F
n. (3.19)
The diusance

D J/n = (1/)Wv
F
; therefore, the
conductance G = e
2
(E
F
)

D becomes (using the 2D den-
sity of states (1.3) with the appropriate degeneracy fac-
tors g
s
= 2, g
v
= 1)
G =
2e
2
h
k
F
W

, in 2D. (3.20)
Eq. (3.20) is the 2D analogue
6
of Sharvins well-known
expression
296
for the point contact conductance in three
dimensions,
G =
2e
2
h
k
2
F
S
4
, in 3D, (3.21)
where now S is the area of the point contact. The num-
ber of propagating modes for a square-well lateral con-
ning potential is N = Int[k
F
W/] in 2D, so Eq. (3.20)
is indeed the classical limit of the quantized conductance
(3.17).
Quantum mechanically, the current through the point
contact is equipartitioned among the 1D subbands, or
transverse modes, in the constriction. The equiparti-
tioning of current, which is the basic mechanism for the
conductance quantization, is illustrated in Fig. 45b for a
square-well lateral conning potential of width W. The
1D subbands then correspond to the pairs of horizon-
tal lines at k
y
= n/W, with n = 1, 2, . . . , N and
N = Int[k
F
W/]. The group velocity v
n
= hk
x
/m is
proportional to cos and thus decreases with increasing
n. However, the decrease in v
n
is compensated by an
increase in the 1D density of states. Since
n
is pro-
portional to the length of the horizontal lines within the
dashed area in Fig. 45b,
n
is proportional to 1/ cos
so that the product v
n

n
does not depend on the sub-
band index. We emphasize that, although the classical
formula (3.20) holds only for a square-well lateral conn-
ing potential, the quantization (3.17) is a general result
for any shape of the conning potential. The reason is
simply that the fundamental cancellation of the group
velocity v
n
= dE
n
(k)/hdk and the 1D density of states

+
n
= (dE
n
(k)/dk)
1
holds regardless of the form of
the dispersion relation E
n
(k). For the same reason, Eq.
(3.17) is equally applicable in the presence of a magnetic
eld, when magnetic edge channels at the Fermi level
take over the role of 1D subbands. Equation (3.17) thus
implies a continuous transition from the zero-eld quan-
tization to the quantum Hall eect, as we will discuss in
Section III.B.2.
To analyze deviations from Eq. (3.17) it is necessary to
solve the Schrodinger equation for the wave functions in
the narrow point contact and the adjacent wide regions
52
and to match the wave functions and their derivatives at
the entrance and exit of the constriction. The resulting
transmission coecients determine the conductance via
the Landauer formula (3.10). This mode coupling prob-
lem has been solved numerically for point contacts of
a variety of shapes
312,313,314,315,316,317,318,319,320,321
and
analytically in special geometries.
322,323,324
When con-
sidering the mode coupling at the entrance and exit of
the constriction, one must distinguish gradual (adiabatic)
from abrupt transitions from wide to narrow regions.
The case of an adiabatic constriction has been stud-
ied by Glazman et al.,
325
by Yacoby and Imry
326
and
by Payne.
272
If the constriction width W(x) changes suf-
ciently gradually, the transport through the constric-
tion is adiabatic (i.e., without intersubband scattering).
The transmission coecients then vanish, [t
nm
[
2
= 0,
unless n = m N
min
, with N
min
the smallest number
of occupied subbands in the constriction. The conduc-
tance quantization (3.17) now follows immediately from
the Landauer formula (3.10). The criterion for adia-
batic transport is
326
dW/dx
<

1/N(x), with N(x)

k
F
W(x)/ the local number of subbands. As the con-
striction widens, N(x) increases and adiabaticity is pre-
served only if W(x) increases more and more slowly.
In practice, adiabaticity breaks down at a width W
max
,
which is at most a factor of 2 larger than the mini-
mum width W
min
(cf. the collimated beam experiment
of Ref.
327
, discussed in Section III.D). This does not af-
fect the conductance of the constriction, however, if the
breakdown of adiabaticity results in a mixing of the sub-
bands without causing reection back through the con-
striction. If such is the case, the total transmission proba-
bility through the constriction remains the same as in the
hypothetical case of fully adiabatic transport. As pointed
out by Yacoby and Imry,
326
a relatively small adiabatic
increase in width from W
min
to W
max
is sucient to en-
sure a drastic suppression of reections at W
max
. The
reason is that the subbands with the largest reection
probability are close to cuto, that is, they have subband
index close to N
max
, the number of subbands occupied
at W
max
. Because the transport is adiabatic from W
min
to W
max
, only the N
min
subbands with the smallest n ar-
rive at W
max
, and these subbands have a small reection
probability. In the language of waveguide transmission,
one has impedance-matched the constriction to the wide
2DEG regions.
328
The ltering of subbands by a grad-
ually widening constriction has an interesting eect on
the angular distribution of the electrons injected into the
wide 2DEG. This horn collimation eect
329
is discussed
in Section III.D.
An adiabatic constriction improves the
accuracy of the conductance quantization,
but is not required to observe the eect.
Calculations
312,313,314,315,316,317,318,319,320,321,322,323,324
show that well-dened conductance plateaux persist for
abrupt constrictions, especially if they are neither very
short nor very long. The optimum length for the observa-
tion of the plateaux is given by
313
L
opt
0.4(W
F
)
1/2
.
FIG. 46 Transmission resonances exhibited by theoretical
results for the conductance of a quantum point contact of
abrupt (rectangular) shape. A smearing of the resonances
occurs at nonzero temperatures (T0 = 0.02 EF/kB 2.8 K).
The dashed curve is an exact numerical result; the full curves
are approximate. Taken from A. Szafer and A. D. Stone,
Phys. Rev. Lett. 62, 300 (1989).
In shorter constrictions the plateaux acquire a nite
slope, although they do not disappear completely even
at zero length. For L > L
opt
the calculations exhibit
regular oscillations that depress the conductance pe-
riodically below its quantized value. The oscillations
are damped and have usually vanished before the next
plateau is reached. As a representative illustration,
we reproduce in Fig. 46 a set of numerical results for
the conductance as a function of width (at xed Fermi
wave vector), obtained by Szafer and Stone.
315
Note
that a nite temperature improves the atness of the
plateaux, as observed experimentally. The existence of
an optimum length can be understood as follows.
Because of the abrupt widening of the constriction,
there is a signicant probability for reection at the exit
of the constriction, in contrast to the adiabatic case
considered earlier. The conductance as a function of
width, or Fermi energy, is therefore not a simple step
function. On the nth conductance plateau backscat-
tering occurs predominantly for the nth subband, since
it is closest to cuto. Resonant transmission of this
subband occurs if the constriction length L is approxi-
mately an integer multiple of half the longitudinal wave-
length
n
= h[2m(E
F
E
n
)]
1/2
, leading to oscillations
on the conductance plateaux. These transmission res-
onances are damped, because the reection probability
decreases with decreasing
n
. The shortest value of

n
on the Nth conductance plateau is h[2m(E
N+1

E
N
)]
1/2
(W
F
)
1/2
(for a square-well lateral conning
potential). The transmission resonances are thus sup-
53
FIG. 47 Resistance as a function of gate voltage for an elon-
gated quantum point contact (L = 0.8 m) at temperatures of
0.2, 0.4, and 0.8 K, showing transmission resonances. Subse-
quent curves from the bottom are oset by 1 k. Taken from
R. J. Brown et al., Solid State Electron. 32, 1179 (1989).
pressed if L
<

(W
F
)
1/2
. Transmission through evanes-
cent modes (i.e., subbands above E
F
) is predominant for
the (N + 1)th subband, since it has the largest decay
length
N+1
= h[2m(E
N+1
E
F
)]
1/2
. The observation
of that plateau requires that the constriction length ex-
ceeds this decay length at the population threshold of the
N th mode, or L
>

h[2m(E
N+1
E
N
)]
1/2
(W
F
)
1/2
.
The optimum length
313
L
opt
0.4(W
F
)
1/2
thus sepa-
rates a short constriction regime, in which transmission
via evanescent modes cannot be ignored, from a long con-
striction regime, in which transmission resonances ob-
scure the plateaux.
Oscillatory structure was resolved in low-temperature
experiments on the conductance quantization of one
quantum point contact by van Wees et al.,
308
but was
not clearly seen in other devices. A diculty in the in-
terpretation of these and other experiments is that os-
cillations can also be caused by quantum interference
processes involving impurity scattering near the constric-
tion. Another experimental observation of oscillatory
structure was reported by Hirayama et al.
330
for short
(100-nm) quantum point contacts of xed width (dened
by means of focused ion beam lithography). To observe
the plateaux, they slowly varied the electron density by
weakly illuminating the sample. The oscillations were
quite reproducible, also after thermal cycling of the sam-
ple, but again they were found in some of the devices
only (this was attributed to variations in the abruptness
of the constrictions
330,331
). Brown et al.
332
have stud-
ied the conductance of split-gate constrictions of lengths
L 0.3, 0.8, and 1 m, and they observed pronounced os-
cillations instead of the at conductance plateaux found
for shorter quantum point contacts. The observed oscil-
latory structure (reproduced in Fig. 47) is quite regular,
and it correlates with the sequence of plateaux that is
recovered at higher temperatures (around 0.8 K). The
eect was seen in all of the devices studied in Ref.
332
.
Measurements by Timp et al.
306
on rather similar 0.9-
m-long constrictions did not show periodic oscillations,
however. Brown et al. conclude that their oscillations
are due to transmission resonances associated with reec-
tions at entrance and exit of the constriction. Detailed
comparison with theory is dicult because the transmis-
sion resonances depend sensitively on the shape of the lat-
eral conning potential and on the presence of a potential
barrier in the constriction (see Section III.B.2). A cal-
culation that comes close to the observation of Brown et
al. has been published by Martin-Moreno and Smith.
333
2. Depopulation of subbands and suppression of backscattering
by a magnetic eld
The eect of a magnetic eld (perpendicular to the
2DEG) on the quantized conductance of a point con-
tact is shown in Fig. 48, as measured by van Wees et
al.
334
First of all, Fig. 48 demonstrates that the conduc-
tance quantization is conserved in the presence of a mag-
netic eld and shows a smooth transition from zero-eld
quantization to quantum Hall eect. The most notice-
able eect of the magnetic eld is to reduce the number
of plateaux in a given gate voltage interval. This pro-
vides a demonstration of depopulation of magnetoelec-
tric subbands, which is more direct than that provided
by the experiments discussed in Section II.F. In addi-
tion, one observes that the atness of the plateaux im-
proves in the presence of the eld. This is due to the
reduction of the reection probability at the point con-
tact, which is revealed most clearly in a somewhat dier-
ent (four-terminal) measurement conguration. These
two eects of a magnetic eld will be discussed sepa-
rately. We will return to the magnetic suppression of
back-scattering in Section IV.A in connection with the
edge channel theory
112
of the quantum Hall eect.
(a) Depopulation of subbands. Because the
equipartitioning of current among the 1D subbands holds
regardless of the nature of the subbands involved, one
can conclude that in the presence of a magnetic eld
B the conductance remains quantized according to G =
(2e
2
/h)N (ignoring spin splitting of the subbands, for
simplicity). Explicit calculations
335
conrm this expec-
54
FIG. 48 Point contact conductance (corrected for a back-
ground resistance) as a function of gate voltage for several
magnetic eld values, illustrating the transition from zero-
eld quantization to quantum Hall eect. The curves have
been oset for clarity. The inset shows the device geometry.
Taken from B. J. van Wees et al., Phys. Rev. B. 38, 3625
(1988).
tation. The number of occupied subbands N as a func-
tion of B has been studied in Sections II.F and III.A and
is given by Eqs. (2.61) and (2.62) for a parabolic and a
square-well potential, respectively. In the high-magnetic-
eld regime W
>

2l
cycl
, the number N E
F
/h
c
is just
the number of occupied Landau levels. The conductance
quantization is then a manifestation of the quantum Hall
eect.
8
(The fact that G is not a Hall conductance but
a two-terminal conductance is not an essential distinc-
tion for this eect; see Section IV.A.) At lower mag-
netic elds, the conductance quantization provides a di-
rect and extremely straightforward method to measure
via N = G(2e
2
/h)
1
the depopulation of magnetoelec-
tric subbands in the constriction.
Figure 49 shows N versus B
1
for various gate volt-
ages, as it follows from the experiment of Fig. 48. Also
shown are the theoretical curves for a square-well con-
ning potential, with the potential barrier in the con-
striction taken into account by replacing E
F
by E
F
E
c
in Eq. (2.62). The B-dependence of E
F
has been ig-
nored in the calculation. The barrier height E
c
is ob-
FIG. 49 Number of occupied subbands as a function of re-
ciprocal magnetic eld for several values of the gate voltage.
Data points have been obtained directly from the quantized
conductance (Fig. 48); solid curves are calculated for a square-
well conning potential of width W and well bottom Ec as
tabulated in the inset. Taken from B. J. van Wees et al.,
Phys. Rev. B 38, 3625 (1988).
tained from the high-eld conductance plateaux [where
N (E
F
E
c
)/h
c
], and the constriction width W
then follows from the zero-eld conductance (where N
[2m(E
F
E
c
)/h
2
]
1/2
W/). The good agreement found
over the entire eld range conrms the expectation that
the quantized conductance is exclusively determined by
the number of occupied subbands, irrespective of their
electric or magnetic origin. The analysis in Fig. 49 is
for a square-well conning potential.
334
For the narrow-
est constrictions a parabolic potential should be more
appropriate,
61
which has been used to analyze the data
of Fig. 48 in Refs.
336
and
308
. Wharam et al.
337
have
analyzed their depopulation data using the intermediate
model of a parabolic potential with a attened bottom
(cf. also Ref.
336
). Because of the uncertainties in the
actual shape of the potential, the parameter values tabu-
lated in Fig. 49 should be considered as rough estimates
only.
In strong magnetic elds the spin degeneracy of
the energy levels is removed, and additional plateaux
55
FIG. 50 Four-terminal longitudinal magnetoresistance R4t
RL of a constriction for a series of gate voltages. The neg-
ative magnetoresistance is temperature independent between
50 mK and 4 K. Solid lines are according to Eqs. (3.23) and
(2.62), with the constriction width as adjustable parameter.
The inset shows schematically the device geometry, with the
two voltage probes used to measure RL. Taken from H. van
Houten et al., Phys. Rev. B 37, 8534 (1988).
appear
7,334
at odd multiples of e
2
/h. Wharam et al.
7
have demonstrated this eect in a particularly clear fash-
ion, using a magnetic eld parallel (rather than per-
pendicular) to the 2DEG. Rather strong magnetic elds
turned out to be required to fully lift the spin degeneracy
in this experiment (about 10 T).
(b) Suppression of backscattering. Only a small
fraction of the electrons injected by the current source
into the 2DEG is transmitted through the point con-
tact. The remaining electrons are scattered back into
the source contact. This is the origin of the nonzero re-
sistance of a ballistic point contact. In this subsection we
shall discuss how a relatively weak magnetic eld leads to
a suppression of the geometrical backscattering caused by
the nite width of the point contact, while the amount
of backscattering caused by the potential barrier in the
point contact remains essentially unaected.
The reduction of backscattering by a magnetic eld is
observed as a negative magnetoresistance [i.e., R(B)
R(0) < 0] in a four-terminal measurement of the longi-
tudinal point contact resistance R
L
. The voltage probes
in this experiment
113
are positioned on wide 2DEG re-
gions, well away from the constriction (see the inset in
Fig. 50). This allows the establishment of local equilib-
rium near the voltage probes, at least in weak magnetic
elds (cf. Sections IV.A and IV.B), so that the measured
four-terminal resistance does not depend on the proper-
ties of the probes. The experimental results for R
L
in this
geometry are plotted in Fig. 50. The negative magnetore-
sistance is temperature-independent (between 50 mK and
4 K) and is observed in weak magnetic elds once the nar-
row constriction is dened (for V
g
<

0.3 V). At stronger

magnetic elds (B > 0.4 T), a crossover is observed to a
positive magnetoresistance. The zero-eld resistance, the
magnitude of the negative magnetoresistance, the slope
of the positive magnetoresistance, as well as the crossover
eld, all increase with increasing negative gate voltage.
The magnetic eld dependence of the four-terminal re-
sistance shown in Fig. 50 is qualitatively dierent from
that of the two-terminal resistance R
2t
G
1
consid-
ered in the previous subsection. In fact, R
2t
is approx-
imately B-independent in weak magnetic elds (below
the crossover elds of Fig. 50). The reason is that R
2t
is
given by [cf. Eq. (3.17)]
R
2t
=
h
2e
2
1
N
min
, (3.22)
with N
min
the number of occupied subbands in the con-
striction (at the point where it has its minimum width
and electron gas density). In weak magnetic elds such
that 2l
cycl
> W, the number of occupied subbands re-
mains approximately constant [cf. Fig. 31 or Eq. (2.62)],
so R
2t
is only weakly dependent on B in this eld regime.
For stronger elds Eq. (3.22) describes a positive magne-
toresistance, because N
min
decreases due to the magnetic
depopulation of subbands discussed earlier. (A similar
positive magnetoresistance is found in a Hall bar with
a cross gate; see Ref.
338
.) Why then does one nd a
negative magnetoresistance in the four-terminal measure-
ments of Fig. 50? Qualitatively, the answer is shown in
Fig. 51, for a constriction without a potential barrier. In
a magnetic eld the left-and right-moving electrons are
spatially separated by the Lorentz force at opposite sides
of the constriction. Quantum mechanically the skipping
orbits in Fig. 51 correspond to magnetic edge states (cf.
Fig. 41). Backscattering thus requires scattering across
the width of the constriction, which becomes increasingly
improbable as l
cycl
becomes smaller and smaller com-
pared with the width (compare Figs. 51a,b). For this
reason a magnetic eld suppresses the geometrical con-
striction resistance in the ballistic regime, but not the re-
sistance associated with the constriction in energy space,
which is due to the potential barrier.
These eects were analyzed theoretically in Ref.
113
,
with the simple result
R
L
=
h
2e
2
_
1
N
min

1
N
wide
_
. (3.23)
Here N
wide
is the number of occupied Landau levels in the
wide 2DEG regions. The simplest (but incomplete) argu-
ment leading to Eq. (3.23) is that the additivity of volt-
ages on reservoirs (ohmic contacts) implies that the two-
terminal resistance R
2t
= (h/2e
2
)N
1
min
should equal the
sum of the Hall resistance R
H
= (h/2e
2
)N
1
wide
and the
56
FIG. 51 Illustration of the reduction of backscattering by a
magnetic eld, which is responsible for the negative magne-
toresistance of Fig. 50. Shown are trajectories approaching
a constriction without a potential barrier, in a weak (a) and
strong (b) magnetic eld. Taken from H. van Houten et al.,
in Ref.
9
.
longitudinal resistance R
L
. This argument is incomplete
because it assumes that the Hall resistance in the wide
regions is not aected by the presence of the constric-
tion. This is correct in general only if inelastic scattering
has equilibrated the edge states transmitted through the
constriction before they reach a voltage probe. Devia-
tions from Eq. (3.23) can occur in the absence of local
equilibrium near the voltage probes, depending on the
properties of the probes themselves. We discuss this in
Section IV.B, following a derivation of Eq. (3.23) from
the Landauer-B uttiker formalism.
At small magnetic elds N
min
is approximately con-
stant, while N
wide
E
F
/h
c
decreases linearly with B.
Equation (3.23) thus predicts a negative magnetoresis-
tance. If the electron density in the wide and narrow
regions is equal (i.e., the barrier height E
c
= 0), then the
resistance R
L
vanishes for elds B > B
crit
2 hk
F
/eW.
This follows from Eq. (3.23), because in this case N
min
and N
wide
are identical. If the electron density in the
constriction is less than its value in the wide region, then
Eq. (3.23) predicts a crossover at B
crit
to a strong-eld
regime of positive magnetoresistance described by
R
L

h
2e
2
_
h
c
E
F
E
c

h
c
E
F
_
if B > B
crit
. (3.24)
The experimental results are well described by the solid
curves following from Eq. (3.23) (with N
min
given by the
square-well result (2.62), and with an added constant
background resistance). The constriction in the present
experiment is relatively long (L 3.4 m), and wide (W
ranging from 0.2 to 1.0 m) so that it does not exhibit
quantized two-terminal conductance plateaux in the ab-
sence of a magnetic eld. For this reason the discrete-
ness of N
min
was ignored in the theoretical curves in Fig.
50. We emphasize, however, that Eq. (3.23) is equally
applicable to the quantized case, as observed by several
groups
307,339,340,341,342
(see Section IV.B).
The negative magnetoresistance (3.23) due to the sup-
pression of the contact resistance is an additive contribu-
tion to the magnetoresistance of a long and narrow chan-
nel in the quasi-ballistic regime (if the voltage probes are
positioned on two wide 2DEG regions, connected by the
channel). For a channel of length L and a mean free path
l the zero-eld contact resistance is a fraction l/L of
the Drude resistance and may thus be ignored for L l.
The strong-eld positive magnetoresistance (3.24) result-
ing from a dierent electron density in the channel may
still be important, however. The eect of the contact
resistance may be suppressed to a large extent by us-
ing narrow voltage probes attached to the channel itself
rather than to wide 2DEG regions. As we will see in Sec-
tion III.E, such a solution no longer works in the ballis-
tic transport regime, because of the additional scattering
induced
289
by the voltage probes.
C. Coherent electron focusing
A magnetic eld may be used to focus the electrons
injected by a point contact onto a second point contact.
Electron focusing in metals was originally conceived by
Sharvin
296
as a method to investigate the shape of the
Fermi surface. It has become a powerful tool in the
study of surface scattering
343
and the electron-phonon
interaction,
344
as reviewed in Refs.
305,345
, and
346
. The
experiment is the analogue in the solid state of magnetic
focusing of electrons in vacuum. Required is a large mean
free path for the carriers at the Fermi surface, to ensure
ballistic motion as in vacuum. The mean free path should
be much larger than the separation L of the two point
contacts. Moreover, L should be much larger than the
point contact width W, to achieve optimal resolution. In
metals, electron focusing is essentially a classical phe-
nomenon because the Fermi wavelength
F
0.5 nm is
much smaller than both W 1 m and L 100 m.
The ratios
F
/L and
F
/W are much larger in a 2DEG
than in a metal, typically by factors of 10
4
and 10
2
, re-
spectively. Coherent electron focusing
59,80,347
is possible
in a 2DEG because of this relatively large value of the
Fermi wavelength, and turns out to be strikingy dierent
from classical electron focusing in metals.
Electron focusing can be seen as a transmission exper-
iment in electron optics (cf. Ref.
3
for a discussion from
this point of view). An alternative point of view (empha-
sized in Refs.
80
and
348
) is that coherent electron focus-
ing is a prototype of a nonlocal resistance measurement
in the quantum ballistic transport regime, such as stud-
ied extensively in narrow-channel geometries.
310
Longitu-
dinal resistances that are negative (not B symmetric)
and dependent on the properties of the current and volt-
age contacts as well as on their separation, periodic and
aperiodic magnetoresistance oscillations, absence of local
equilibrium are all characteristic features of this trans-
port regime that appear in a most extreme and bare form
57
FIG. 52 Illustration of classical electron focusing by a mag-
netic eld. Top: Skipping orbits along the 2DEG boundary.
The trajectories are drawn up to the third specular reection.
Bottom: Plot of the caustics, which are the collection of focal
points of the trajectories. Taken from H. van Houten et al.,
Phys. Rev. B 39, 8556 (1989).
in the electron focusing geometry. One reason for the
simplication oered by this geometry is that the cur-
rent and voltage contacts, being point contacts, are not
nearly as invasive as the wide leads in a Hall bar geom-
etry. Another reason is that the electrons interact with
only one boundary (instead of two in a narrow channel).
The outline of this section is as follows.In Sec-
tion III.C.1 the experimental results on coherent
electron focusing
59,80
are presented. A theoretical
description
80,347
is given in Section III.C.2, in terms of
mode interference in the waveguide formed by the mag-
netic eld at the 2DEG boundary. Apart from the in-
trinsic interest of electron focusing in a 2DEG, the ex-
periment can also be seen as a method to study electron
scattering, as in metals. Two such applications
108,349
are discussed in Section III.C.3. We restrict ourselves in
this section to focusing by a magnetic eld. Electrostatic
focusing
350
is discussed in Section III.D.2.
1. Experiments
The geometry of the experiment
59
in a 2DEG is the
transverse focusing geometry of Tsoi
343
and consists of
two point contacts on the same boundary in a perpen-
dicular magnetic eld. (In metals one can also use the
geometry of Sharvin
296
with opposite point contacts in
a longitudinal eld. This is not possible in two dimen-
sions.) Two point contacts and the intermediate 2DEG
boundary are created electrostatically by means of the
two split gates shown in Fig. 5b. Figure 52 illustrates
electron focusing in two dimensions as it follows from the
classical mechanics of electrons at the Fermi level. The
FIG. 53 Bottom: Experimental electron focusing spectrum
(T = 50 mK, L = 3.0 m) in the generalized Hall resistance
conguration 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 ne 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 elds. Taken from H. van Houten et al., Phys. Rev.
B 39, 8556 (1989).
injector (i) injects a divergent beam of electrons ballisti-
cally into the 2DEG. Electrons are detected if they reach
the adjacent collector (c), after one or more specular re-
ections at the boundary connecting i and c. (These are
the skipping orbits discussed in Section III.A.1.) The fo-
cusing action of the magnetic eld is evident in Fig. 52
(top) from the black lines of high density of trajectories.
These lines are known in optics as caustics and they are
plotted separately in Fig. 52 (bottom). The caustics in-
tersect the 2DEG boundary at multiples of the cyclotron
diameter from the injector. As the magnetic eld is in-
creased, a series of these focal points shifts past the col-
lector. The electron ux incident on the collector thus
reaches a maximum whenever its separation L from the
injector is an integer multiple of 2l
cycl
= 2 hk
F
/eB. This
occurs when B = pB
focus
, p = 1, 2, . . ., with
B
focus
= 2 hk
F
/eL. (3.25)
For a given injected current I
i
the voltage V
c
on the col-
lector is proportional to the incident ux. The classical
picture thus predicts a series of equidistant peaks in the
collector voltage as a function of magnetic eld.
In Fig. 53 (top) we show such a classical focusing spec-
trum, calculated for parameters corresponding to the
58
experiment discussed later (L = 3.0 m, k
F
= 1.5
10
8
m
1
). The spectrum consists of equidistant focusing
peaks of approximately equal magnitude superimposed
on the Hall resistance (dashed line). The pth peak is
due to electrons injected perpendicularly to the boundary
that have made p1 specular reections between injector
and collector. Such a classical focusing spectrum is com-
monly observed in metals,
351,352
albeit with a decreasing
height of subsequent peaks because of partially diuse
scattering at the metal surface. Note that the peaks oc-
cur in one eld direction only. In reverse elds the focal
points are at the wrong side of the injector for detection,
and the normal Hall resistance is obtained. The experi-
mental result for a 2DEG is shown in the bottom half of
Fig. 53 (trace a; trace b is discussed later). A series of
ve focusing peaks is evident at the expected positions.
The observation of multiple focusing peaks immediately
implies that the electrostatically dened 2DEG boundary
scatters predominantly specularly. (This nding
59
is sup-
ported by the magnetoresistance experiments of Thorn-
ton et a.
107
in a narrow split-gate channel; cf. Section
II.A.) Figure 53 is obtained in a measuring conguration
(inset) in which an imaginary line connecting the voltage
probes crosses that between the current source and drain.
This is the conguration for a generalized Hall resistance
measurement. If the crossing is avoided, one measures a
longitudinal resistance, which shows the focusing peaks
without a superimposed Hall slope. This longitudinal re-
sistance periodically becomes negative. This is a classical
result
80
of magnetic defocusing, which causes the prob-
ability density near the point contact voltage probe to
be reduced with respect to the spatially averaged prob-
ability density that determines the voltage on the wide
voltage probe (cf. the regions of reduced density between
lines of focus in Fig. 52).
On the experimental focusing peaks a ne structure is
resolved at low temperatures (below 1 K). The ne struc-
ture is well reproducible but sample-dependent. A nice
demonstration of the reproducibility of the ne struc-
ture is obtained upon interchanging current and voltage
leads, so that the injector becomes the collector, and vice
versa. The resulting focusing spectrum shown in Fig. 53
(trace b) is almost the precise mirror image of the orig-
inal one (trace a), although this particular device had a
strong asymmetry in the widths of injector and collector.
The symmetry in the focusing spectra is an example of
the general reciprocity relation (3.16). If one applies the
B uttiker equations (3.12) to the electron focusing geom-
etry (as is done in Section IV.B), one nds that the ratio
of collector voltage V
c
to injector current I
i
is given by
V
c
I
i
=
2e
2
h
T
ic
G
i
G
c
, (3.26)
where T
ic
is the transmission probability from injector
to collector, and G
i
and G
c
are the conductances of the
injector and collector point contact. Since T
ic
(B) =
T
ci
(B) and G(B) = G(B), this expression for the
focusing spectrum is manifestly symmetric under inter-
FIG. 54 Experimental electron focusing spectrum over a
larger eld range and for very narrow point contacts (esti-
mated width 2040 nm; T = 50 mK, L = 1.5 m). The inset
gives the Fourier transform for B 0.8 T. The high-eld os-
cillations have the same dominant periodicity as the low-eld
focusing peaks, but with a much larger amplitude. Taken
from H. van Houten et al., Phys. Rev. B 39, 8556 (1989).
change of injector and collector with reversal of the mag-
netic eld.
The ne structure on the focusing peaks in Fig. 53 is
the rst indication that electron focusing in a 2DEG is
qualitatively dierent from the corresponding experiment
in metals. At higher magnetic elds the resemblance to
the classical focusing spectrum is lost; see Fig. 54. A
Fourier transform of the spectrum for B 0.8 T (inset
in Fig. 54) shows that the large-amplitude higheld os-
cillations have a dominant periodicity of 0.1 T, which
is approximately the same as the periodicity B
focus
of
the much smaller focusing peaks at low magnetic elds
(B
focus
in Fig. 54 diers from Fig. 53 because of a smaller
L = 1.5 m). This dominant periodicity can be explained
in terms of quantum interference between the dierent
skipping orbits from injector to collector or in terms of
interference of coherently excited edge channels, as we
discuss in the following subsection. The experimental
implication is that the injector acts as a coherent point
source with the coherence maintained over a distance of
several microns to the collector.
2. Theory
To explain the characteristic features of the coher-
ent electron focusing experiments we have described, we
must go beyond the classical description.
80,347
As dis-
cussed in Section III.A, quantum ballistic transport along
the 2DEG boundary in a magnetic eld takes place via
magnetic edge states, which form the propagating modes
59
FIG. 55 Phase knL of the edge channels at the collector, cal-
culated from Eq. (3.27). Note the domain of approximately
linear n-dependence of the phase, responsible for the oscilla-
tions with B
focus
-periodicity. Taken from H. van Houten et
al., Phys. Rev. B 39, 8556 (1989).
at the Fermi level. Since the injector has a width below

F
, it excites these modes coherently. For k
F
L 1 the
interference of modes at the collector is dominated by
their rapidly varying phase factors exp(ik
n
L). The wave
number k
n
corresponds classically to the separation of
the center of the cyclotron orbit from the 2DEG bound-
ary [Eq. (3.5)]. In the Landau gauge A = (0, Bx, 0) (with
the axis chosen as in Fig. 52) one has k
n
= k
F
sin
n
,
where is the angle with the x-axis under which the cy-
clotron orbit is reected from the boundary. The quan-
tized values
n
follow in this semiclassical description
from the Bohr-Sommerfeld quantization rule (3.6) that
the ux enclosed by the cyclotron orbit and the boundary
equals (n
1
4
)h/e [the phase shift in Eq. (3.6) equals
/2 for an edge state at an innite barrier potential].
Simple geometry shows that this requires that

2

n

1
2
sin 2
n
=
2
k
F
l
cycl
_
n
1
4
_
, n = 1, 2, . . . , N.
(3.27)
As plotted in Fig. 55, the dependence on n of the phase
k
n
L is close to linear in a broad interval. This also follows
from expansion of Eq. (3.27) around
n
= 0, which gives
k
n
L = constant 2n
B
B
focus
+k
F
Lorder
_
N 2n
N
_
3
.
(3.28)
If B/B
focus
is an integer, a fraction of order (1/k
F
L)
1/3
of the N edge states interfere constructively at the col-
lector. Because of the 1/3 power, this is a substantial
fraction even for the large k
F
L 10
2
of the experiment.
The resulting mode interference oscillations with B
focus
-
periodicity can become much larger than the classical
focusing peaks. This has been shown in Refs.
347
and
80
,
where the transmission probability T
ic
was calculated in
the WKB approximation with neglect of the nite width
FIG. 56 Focusing spectrum calculated from Eq. (3.29),
for parameters corresponding to the experimental Fig. 54.
The inset shows the Fourier transform for B 0.8 T.
Innitesimally small point contact widths are assumed in
the calculation. Taken from C. W. J. Beenakker et al.,
Festkorperprobleme 9, 299 (1989).
of the injector and detector. From Eq. (3.26) the focusing
spectrum is then obtained in the form
V
c
I
i
=
h
2e
2

1
N
N

n=1
e
iknL

2
, (3.29)
which is plotted in Fig. 56 for parameter values corre-
sponding to the experimental Fig. 54. The inset shows
the Fourier transform for B 0.8 T.
There is no detailed one-to-one correspondence be-
tween the experimental and theoretical spectra. No such
correspondence was to be expected in view of the sensi-
tivity of the experimental spectrum to small variations in
the voltage on the gate dening the point contacts and
the 2DEG boundary. Those features of the experimental
spectrum that are insensitive to the precise measurement
conditions are, however, well reproduced by the calcula-
tion: We recognize in Fig. 56 the low-eld focusing peaks
and the large-amplitude high-eld oscillations with the
same B
focus
-periodicity. The high-eld oscillations range
from about 0 to 10 k in both theory and experiment.
The maximum amplitude is not far below the theoretical
upper bound of h/2e
2
13 k, which follows from Eq.
(3.29) if we assume that all the modes interfere construc-
tively. This indicates that a maximal phase coherence is
realized in the experiment and implies that the experi-
mental injector and collector point contacts resemble the
idealized point source detector in the calculation.
60
FIG. 57 Experimental electron focusing spectra (in the gen-
eralized longitudinal resistance conguration) at 0.3 K for ve
dierent injector-collector separations in a very high mobility
material. The vertical scale varies among the curves. Taken
from J. Spector et al., Surf. Sci. 228, 283 (1990).
3. Scattering and electron focusing
Scattering events other than specular boundary scat-
tering can be largely ignored for the relatively small point
contact separations L 3 m in the experiments dis-
cussed earlier.
59,80
(any other inelastic or elastic scatter-
ing events would have been detected as a reduction of the
oscillations with B
focus
-periodicity below the theoretical
estimate). Spector et al.
349
have repeated the experi-
ments for larger L to study scattering processes in an
ultrahigh mobility
353,354
2DEG (
e
= 5.510
6
cm
2
/Vs).
They used relatively wide point contacts (about 1 m) so
that electron focusing was in the classical regime. In Fig.
57 we reproduce their experimental results for point con-
tact separations up to 64 m. The peaks in the focusing
spectrum for a given L have a roughly constant ampli-
tude, indicating that scattering at the boundary is mostly
specular rather than diusive in agreement with the
experiments of Ref.
59
. Spector et al.
349
nd that the
amplitude of the focusing peaks decreases exponentially
with increasing L, due to scattering in the electron gas
(see Fig. 58). The decay exp(L/L
0
) with L
0
10 m
implies an eective mean free path (measured along the
arc of the skipping orbits) of L
0
/2 15 m. This is
smaller than the transport mean free path derived from
the conductivity by about a factor of 2, which may point
to a greater sensitivity of electron focusing to forward
scattering.
Electron focusing by a magnetic eld may also play
a role in geometries other than the double-point con-
tact geometry of Fig. 52. One example is mentioned in
the context of junction scattering in a cross geometry
in Section III.E. Another example is the experiment by
Nakamura et al.
108
on the magnetoresistance of equally
spaced narrow channels in parallel (see Fig. 59). Resis-
tance peaks occur in this experiment when electrons that
are transmitted through one of the channels are focused
back through another channel. The resistance peaks oc-
FIG. 58 Exponential decay of the oscillation amplitude of
the collector voltage (normalized by the injector voltage) as
a function of injector-collector separation d (denoted by L in
the text). Taken from J. Spector et al., Surf. Sci. 228, 283
(1990).
cur at B = (n/m)B
focus
, where B
focus
is given by Eq.
(3.25) with L the spacing of adjacent channels. The iden-
tication of the various peaks in Fig. 59 is given in the
inset. Nakamura et al.
108
conclude from the rapidly di-
minishing height of consecutive focusing peaks (which re-
quire an increasing number of specular reections) that
there is a large probability of diuse boundary scatter-
ing. The reason for the dierence with the experiments
discussed previously is that the boundary in the exper-
iment of Fig. 59 is dened by focused ion beam lithog-
raphy, rather than electrostatically by means of a gate.
As discussed in Section II.A, the former technique may
introduce a considerable boundary roughness.
Electron focusing has been used by Williamson et al.
355
to study scattering processes for hot electrons, with
an energy in excess of the Fermi energy, and for cool
holes, or empty states in the conduction band below the
Fermi level (see Ref.
307
for a review). An interesting
aspect of hot-electron focusing is that it allows a mea-
surement of the local electrostatic potential drop across
a current-carrying quantum point contact,
355
something
that is not possible using conventional resistance mea-
surements, where the sum of electrostatic and chemical
potentials is measured. The importance of such alterna-
tive techniques to study electrical conduction has been
stressed by Landauer.
356
61
FIG. 59 Magnetoresistance of N constrictions in parallel
at 1.3 K. The arrows indicate the oscillations due to electron
focusing, according to the mechanisms illustrated in the inset.
The resistance scale is indicated by 10 bars. Taken from K.
Nakamura et al., Appl. Phys. Lett. 56, 385 (1990).
D. Collimation
The subject of this section is the collimation of elec-
trons injected by a point contact
329
and its eect on
transport measurements in geometries involving two op-
posite point contacts.
327,357
Collimation (i.e., the narrow-
ing of the angular injection distributions) follows from
the constraints on the electron momentum imposed by
the potential barrier in the point contact (barrier collima-
tion), and by the gradual widening of the point contact
at its entrance and exit (horn collimation). We summa-
rize the theory in Section III.D.1. The eect was origi-
nally proposed
329
to explain the remarkable observation
of Wharam et al.
357
that the series resistance of two op-
posite point contacts is considerably less than the sum
of the two individual resistances (Section III.D.3). A di-
rect experimental proof of collimation was provided by
Molenkamp et al.,
327
who measured the deection of the
injected beam of electrons in a magnetic eld (Section
III.D.2). A related experiment by Sivan et al.,
350
aimed
at the demonstration of the focusing action of an elec-
trostatic lens, is also discussed in this subsection. The
collimation eect has an importance in ballistic trans-
port that goes beyond the point contact geometry. It
FIG. 60 Illustration of the collimation eect for an abrupt
constriction (a) containing a potential barrier of height Ec
and for a horn-shaped constriction (b) that is ared from a
width Wmin to Wmax. The dash-dotted trajectories approach-
ing at an angle outside the injection/acceptance cone are
reected. Taken from H. van Houten and C. W. J. Beenakker,
in Nanostructure Physics and Fabrication (M. Reed and W.
P. Kirk, eds.). Academic, New York, 1989.
will be shown in Section III.E that the phenomenon is at
the origin of a variety of magnetoresistance anomalies in
narrow multiprobe conductors.
358,359,360
1. Theory
Since collimation follows from classical mechanics, a
semiclassical theory is sucient to describe the essen-
tial phenomena, as we now discuss (following Refs.
329
and
311
). Semiclassically, collimation results from the
adiabatic invariance of the product of channel width
W and absolute value of the transverse momentum hk
y
(this product is proportional to the action for motion
transverse to the channel
361
). Therefore, if the electro-
static potential in the point contact region is suciently
smooth, the quantity S = [k
y
[W is approximately con-
stant from point contact entrance to exit. Note that S/
corresponds to the quantum mechanical 1D subband in-
dex n. The quantum mechanical criterion for adiabatic
transport was derived by Yacoby and Imry
326
(see Sec-
tion III.B). As was discussed there, adiabatic transport
breaks down at the exit of the point contact, where it
widens abruptly into a 2DEG of essentially innite width.
Collimation reduces the injection/acceptance cone of the
point contact from its original value of to a value of
2
max
. This eect is illustrated in Fig. 60. Electrons
incident at an angle [[ >
max
from normal incidence
are reected. (The geometry of Fig. 60b is known in
optics as a conical reector.
362
.) Vice versa, all elec-
trons leave the constriction at an angle [[ <
max
(i.e.,
the injected electrons form a collimated beam of angular
opening 2
max
).
To obtain an analytic expression for the collimation
eect, we describe the shape of the potential in the point
contact region by three parameters: W
min
, W
max
, and
E
c
(see Fig. 60). We consider the case that the point
contact has its minimal width W
min
at the point where
the barrier has its maximal height E
c
above the bottom
of the conduction band in the broad regions. At that
point the largest possible value of S is
S
1
(2m/h
2
)
1/2
(E
F
E
c
)
1/2
W
min
.
We assume that adiabatic transport (i.e., S = constant)
62
holds up to a point of zero barrier height and maximal
width W
max
. The abrupt separation of adiabatic and
nonadiabatic regions is a simplication that can be, and
has been, tested by numerical calculations (see below).
At the point contact exit, the largest possible value of S
is
S
2
(2m/h
2
)
1/2
(E
F
)
1/2
sin
max
W
max
.
The invariance of S implies that S
1
= S
2
; hence,

max
= arcsin
_
1
f
_
; f
_
E
F
E
F
E
c
_
1/2
W
max
W
min
.
(3.30)
The collimation factor f 1 is the product of a term
describing the collimating eect of a barrier of height E
c
(barrier collimation) and a term describing collimation
due to a gradual widening of the point contact width
from W
min
to W
max
(horn collimation). In the adia-
batic approximation, the angular injection distribution
P() is proportional to cos with an abrupt truncation
at
max
. The cosine angular dependence follows from
the cosine distribution of the incident ux in combination
with time-reversal symmetry and is thus not aected by
the reduction of the injection/acceptance cone. We there-
fore conclude that in the adiabatic approximation P()
(normalized to unity) is given by
P() =
_
1
2
f cos if [[ < arcsin(1/f),
= 0, otherwise.
(3.31)
We defer to Section III.D.2 a comparison of the analytical
result (3.31) with a numerical calculation.
Barrier collimation does not require adiabaticity. For
an abrupt barrier, collimation simply results from trans-
verse momentum conservation, as in Fig. 60a, leading
directly to Eq. (3.31). (The total external reection at
an abrupt barrier for trajectories outside the collimation
cone is similar to the optical eect of total internal reec-
tion at a boundary separating a region of high refractive
index from a region of small refractive index; see the
end of Section III.D.2.) A related collimation eect re-
sulting from transverse momentum conservation occurs
if electrons tunnel through a potential barrier. Since
the tunneling probability through a high potential bar-
rier is only weakly dependent on energy, it follows that
the strongest collimation is to be expected if the barrier
height equals the Fermi energy. On lowering the barrier
below E
F
ballistic transport over the barrier dominates,
and the collimation cone widens according to Eq. (3.31).
A quantum mechanical calculation of barrier collimation
may be found in Ref.
363
.
The injection distribution (3.31) can be used to obtain
(in the semiclassical limit) the direct transmission prob-
ability T
d
between two opposite identical point contacts
separated by a large distance L. To this end, rst note
that T
d
/N is the fraction of the injected current that
reaches the opposite point contact (since the transmis-
sion probability through the rst point contact is N, for
FIG. 61 Detection of a collimated electron beam over a dis-
tance of 4 m. In this four-terminal measurement, two ohmic
contacts to the 2DEG region between the point contacts are
used: One of these acts as a drain for the current Ii through
the injector, and the other is used as a zeroreference for the
voltage Vc on the collector. The drawn curve is the experi-
mental data at T = 1.8 K. The black dots are the result of a
semiclassical simulation, using a hard-wall potential with con-
tours as shown in the inset. The dashed curve results from a
simulation without collimation (corresponding to rectangular
corners in the potential contour). Taken from L. W. Molen-
kamp et al., Phys. Rev. B 41, 1274 (1990).
N occupied subbands in the point contact). Electrons in-
jected within a cone of opening angle W
max
/L centered
at = 0 reach the opposite point contact and are trans-
mitted. If this opening angle is much smaller than the
total opening angle 2
max
of the beam, then the distribu-
tion function P() can be approximated by P(0) within
this cone. This approximation requires W
max
/L 1/f,
which is satised experimentally in devices with a su-
ciently large point contact separation. We thus obtain
T
d
/N = P(0)W
max
/L, which, using Eq. (3.31), can be
written as
329
T
d
= f(W
max
/2L)N. (3.32)
This simple analytical formula can be used to describe
the experiments on transport through identical opposite
point contacts in terms of one empirical parameter f, as
discussed in the following subsections.
2. Magnetic deection of a collimated electron beam
A method
311,329
to sensitively detect the collimated
electron beam injected by a point contact is to sweep
63
the beam past a second opposite point contact by means
of a magnetic eld. The geometry is shown in Fig. 61
(inset). The current I
i
through the injecting point con-
tact is drained to ground at one or two (the dierence is
not essential) ends of the 2DEG channel separating the
point contacts. The opposite point contact, the collec-
tor, serves as a voltage probe (with the voltage V
c
being
measured relative to ground). In the case that both ends
of the 2DEG channel are grounded, the collector voltage
divided by the injected current is given by
V
c
I
i
=
1
G
T
d
N
, T
d
N, (3.33)
with G = (2e
2
/h)N the two-terminal conductance of
the individual point contact (both point contacts are as-
sumed to be identical) and T
d
the direct transmission
probability between the two point contacts calculated in
Section III.D.1. Equation (3.33) can be obtained from
the Landauer-B uttiker formalism (as done in Ref.
311
) or
simply by noting that the current I
i
T
d
/N incident on the
collector has to be counterbalanced by an equal outgo-
ing current GV
c
. In the absence of a magnetic eld, we
obtain [using Equation (3.32) for the direct transmission
probability]
V
c
I
i
=
h
2e
2
f
2

2k
F
L
, (3.34)
where k
F
is the Fermi wave vector in the region between
the point contacts. In an experimental situation L and
k
F
are known, so the collimation factor f can be directly
determined from the collector voltage by means of Eq.
(3.34).
The result (3.34) holds in the absence of a magnetic
eld. A small magnetic eld B will deect the collimated
electron beam past the collector. Simple geometry leads
to the criterion L/2l
cycl
=
max
for the cyclotron radius
at which T
d
is reduced to zero by the Lorentz force (as-
suming that L W
max
). One would thus expect to see
in V
c
/I
i
a peak around zero eld, of height given by Eq.
(3.34) and of width
B = (4 hk
F
/eL) arcsin(1/f), (3.35)
according to Eq. (3.30).
In Fig. 61 this collimation peak is shown (solid curve),
as measured by Molenkamp et al.
327
at T = 1.2 K in
a device with a L = 4.0-m separation between injec-
tor and collector. In this measurement only one end of
the region between the point contacts was grounded
a measurement conguration referred to in narrow Hall
bar geometries as a bend resistance measurement
289,364
(cf. Section III.E). One can show, using the Landauer-
B uttiker formalism,
5
that the height of the collimation
peak is still given by Eq. (3.34) if one replaces
327
f
2
by
f
2

1
2
. The expression (3.35) for the width is not mod-
ied. The experimental result in Fig. 61 shows a peak
height of 150 (measured relative to the background
resistance at large magnetic elds). Using L = 4.0 m
FIG. 62 Calculated angular injection distributions in zero
magnetic eld. The solid histogram is the result of a simu-
lation of the classical trajectories at the Fermi energy in the
geometry shown in the inset of Fig. 61. The dotted curve
follows from the adiabatic approximation (3.31), with the ex-
perimental collimation factor f = 1.85. The dashed curve
is the cosine distribution in the absence of any collimation.
Taken from L. W. Molenkamp et al., Phys. Rev. B. 41, 1274
(1990).
and the value k
F
= 1.1 10
8
m
1
obtained from Hall re-
sistance measurements in the channel between the point
contacts, one deduces a collimation factor f 1.85. The
corresponding opening angle of the injection/acceptance
cone is 2
max
65

. The calculated value of f would

imply a width B 0.04 T, which is not far from the
measured full width at half maximum of 0.03 T.
The experimental data in Fig. 61 are compared with
the result
327
from a numerical simulation of classical tra-
jectories of the electrons at the Fermi level (following the
method of Ref.
329
). This semiclassical calculation was
performed in order to relax the assumption of adiabatic
transport in the point contact region, and of small T
d
/N,
on which Eqs. (3.32) and (3.34) are based. The dashed
curve is for point contacts dened by hardwall contours
with straight corners (no collimation); the dots are for the
smooth hardwall contours shown in the inset, which lead
to collimation via the horn eect (cf. Fig. 60b; the bar-
rier collimation of Fig. 60a is presumably unimportant at
the small gate voltage used in the experiment and is not
taken into account in the numerical simulation). The an-
gular injection distributions P() that follow from these
numerical simulations are compared in Fig. 62 (solid his-
togram) with the result (3.31) from the adiabatic approx-
imation for f = 1.85 (dotted curve). The uncollimated
distribution P() = (cos )/2 is also shown for compar-
64
FIG. 63 Electrostatic focusing onto a collector (C2) of an
injected electron beam (at i) by means of a concave lens cor-
responding to a region of reduced electron density. Focusing
in such an arrangement was detected experimentally.
350
ison (dashed curve). Taken together, Figs. 61 and 62
unequivocally demonstrate the importance of collimation
for the transport properties, as well as the adequateness
of the adiabatic approximation as an estimator of the
collimation cone.
Once the point contact width becomes less than a
wavelength, diraction inhibits collimation of the elec-
tron beam. In the limit k
F
W 1, the injec-
tion distribution becomes proportional to cos
2
for all
, independent of the shape of the potential in the
point contact region.
80,313
The coherent electron focusing
experiments
59,80
discussed in Sections III.C.1 and III.C.2
were performed in this limit.
We conclude this subsection by briey discussing an
alternative way to increase the transmission probability
between two opposite point contacts, which is focusing of
the injected electron beam onto the collector. Magnetic
focusing, discussed in Section III.C for adjacent point
contacts, cannot be used for opposite point contacts in
two dimensions (unlike in three dimensions, where a mag-
netic eld along the line connecting the point contacts
will focus the beam
296
). A succesful demonstration of
electrostatic focusing was recently reported by Sivan et
al. and by Spector et al.
350
The focusing is achieved by
means of a potential barrier of a concave shape, created
as a region of reduced density in the 2DEG by means of a
gate between the injector and the collector (see Fig. 63).
A focusing lens for electrons is concave because electrons
approaching a potential barrier are deected in a direc-
tion perpendicular to the normal. This is an amusing dif-
ference with light, which is deected toward the normal
on entering a more dense medium, so an optical focus-
ing lens is convex. The dierent dispersion laws are the
origin of this dierent behavior of light and electrons.
350
FIG. 64 Magnetic eld dependence of the series conductance
of two opposite point contacts (measured as shown in the in-
set; the point contact separation is L = 1.0 m) for three dif-
ferent values of the gate voltage (solid curves) at T = 100 mK.
For clarity, subsequent curves from bottom to top are oset
by 0.5 10
4

1
, with the lowest curve shown at its actual
value. The dotted curves are calculated from Eqs. (3.39) and
(2.62), with the point contact width as adjustable parameter.
Taken from A. A. M. Staring et al., Phys. Rev. B. 41, 8461
(1990).
3. Series resistance
The rst experimental study of ballistic transport
through two opposite point contacts was carried out by
Wharam et al.,
357
who discovered that the series resis-
tance is considerably less than the sum of the two in-
dividual resistances. Sugsequent experiments conrmed
this result.
365,366
The theoretical explanation
329
of this
observation is that collimation of the electrons injected
by a point contact enhances the direct transmission prob-
ability through the opposite point contact, thereby sig-
nicantly reducing the series resistance below its ohmic
value. We will discuss the transport and magnetotrans-
port in this geometry. We will not consider the alterna-
tive geometry of two adjacent point contacts in parallel
(studied in Refs.
367,368,369
). In that geometry the col-
limation eect cannot enhance the coupling of the two
point contacts, so only small deviations from Ohms law
are to be expected.
The expression for the two-terminal series resistance
of two identical opposite point contacts in terms of the
direct transmission probability can be obtained from the
Landauer-B uttiker formalism,
5
as was done in Ref.
329
.
We give here an equivalent, somewhat more intuitive
derivation. Consider the geometry shown in Fig. 64 (in-
set). A fraction T
d
/N of the current GV injected through
the rst point contact by the current source is directly
transmitted through the second point contact (and then
drained to ground). Here G = (2e
2
/h)N is the con-
ductance of the individual point contact, and V is the
source-drain voltage. The remaining fraction 1 T
d
/N
equilibrates in the region between the point contacts, as
65
a result of inelastic scattering (elastic scattering is suf-
cient if phase coherence does not play a role). Since
that region cannot drain charge (the attached contacts
are not connected to ground), these electrons will even-
tually leave via one of the two point contacts. For a
symmetric structure we may assume that the fraction
1
2
(1 T
d
/N) of the injected current GV is transmitted
through the second point contact after equilibration. The
total source-drain current I is the sum of the direct and
indirect contributions:
I =
1
2
(1 +T
d
/N)GV. (3.36)
The series conductance G
series
= I/V becomes
G
series
=
1
2
G(1 +T
d
/N). (3.37)
In the absence of direct transmission (T
d
= 0), one re-
covers the ohmic addition law for the resistance, as ex-
pected for the case of complete intervening equilibration
(cf. the related analysis by B uttiker of tunneling in series
barriers
370,371
). At the opposite extreme, if all trans-
mission is direct (T
d
= N), the series conductance is
identical to that of the single point contact. Substituting
(3.32) into Eq. (3.37), we obtain the result
329
for small
but nonzero direct transmission:
G
series
=
1
2
G(1 +f(W
max
/2L)). (3.38)
The quantized plateaus in the series resistance, ob-
served experimentally,
357
are of course not obtained in
the semiclassical calculation leading to Eq. (3.38). How-
ever, since the nonadditivity is essentially a semiclassical
collimation eect, the present analysis should give a rea-
sonably reliable estimate of deviations from additivity for
not too narrow point contacts. For a comparison with ex-
periments we refer to Refs.
307
and
329
. A fully quantum
mechanical calculation of the series resistance has been
carried out numerically by Baranger (reported in Ref.
306
)
for two closely spaced constrictions.
So far we have only considered the case of a zero mag-
netic eld. In a weak magnetic eld (2l
cycl
> L) the
situation is rather complicated. As discussed in detail in
Ref.
329
, there are two competing eects in weak elds:
On the one hand, the deection of the electron beam by
the Lorentz force reduces the direct transmission prob-
ability, with the eect of decreasing the series conduc-
tance. On the other hand, the magnetic eld enhances
the indirect transmission, with the opposite eect. The
result is an initial decrease in the series conductance for
small magnetic elds in the case of strong collimation
and an increase in the case of weak collimation. This
is expected to be a relatively small eect compared with
the eects at stronger elds that are discussed below.
In stronger elds (2l
cycl
< L), the direct transmission
probability vanishes, which greatly simplies the situa-
tion. If we assume that all transmission between the
opposite point contacts is with intervening equilibration,
then the result is
329
G
series
=
2e
2
h
_
2
N

1
N
wide
_
1
. (3.39)
Here N is the (B-dependent) number of occupied sub-
bands in the point contacts, and N
wide
is the number of
occupied Landau levels in the 2DEG between the point
contacts. The physical origin of the simple addition rule
(3.39) is additivity of the four-terminal longitudinal re-
sistance (3.23). From this additivity it follows that for
n dierent point contacts in series, Eq. (3.39) generalizes
to
1
G
series

h
2e
2
1
N
wide
=
n

i=1
R
L
(i), (3.40)
where
R
L
(i) =
_
h
2e
2
__
1
N
i

1
N
wide
_
(3.41)
is the four-terminal longitudinal resistance of point con-
tact i. Equation (3.39) predicts a nonmonotonic B-
dependence for G
series
. This can most easily be seen
by disregarding the discreteness of N and N
wide
. We
then have N
L
E
F
/h
c
, while the magnetic eld de-
pendence of N (for a square-well conning potential in
the point contacts) is given by Eq. (2.62). The result-
ing B-dependence of G
series
is shown in Fig. 64 (dot-
ted curves). The nonmonotonic behavior is due to the
delayed depopulation of subbands in the point contacts
compared with the broad 2DEG. While the number of
occupied Landau levels N
wide
in the region between the
point contacts decreases steadily with B for 2l
cycl
< L,
the number N of occupied subbands in the point contacts
remains approximately constant until 2l
c,min
W
min
,
with l
c,min
l
cycl
(1 E
c
/E
F
)
1/2
denoting the cyclotron
radius in the point contact region. In this eld inter-
val G
series
increases with B, according to Eq. (3.39).
For stronger elds, depopulation in the point contacts
begins to dominate G
series
, leading nally to a decreas-
ing conductance (as is the rule for single point contacts;
see Section III.B.2). The peak in G
series
thus occurs at
2l
c,min
W
min
.
The remarkable camelback shape of G
series
versus B
predicted by Eq. (3.39) has been observed experimen-
tally by Staring et al.
372
The data are shown in Fig.
64 (solid curves) for three values of the gate voltage
V
g
at T = 100 mK. The measurement conguration is
as shown in the inset, with a point contact separation
L = 1.0 m. The dotted curves in Fig. 64 are the re-
sult of a one-parameter t to the theoretical expression.
It is seen that Eq. (3.39) provides a good description
of the overall magnetoresistance behavior from low mag-
netic elds up to the quantum Hall eect regime. The
additional structure in the experimental curves has sev-
eral dierent origins, for which we refer to the paper by
Staring et al.
372
Similar structure in the two-terminal
resistance of a single point contact will be discussed in
detail in Section IV.D.
We emphasize that Eq. (3.39) is based on the assump-
tion of complete equilibration of the current-carrying
edge states in the region between the point contacts. In
66
a quantizing magnetic eld, local equilibrium is reached
by inter-Landau level scattering. If the potential land-
scape (both in the point contacts themselves and in the
2DEG region in between) varies by less than the Landau
level separation h
c
on the length scale of the magnetic
length ( h/eB)
1/2
, then inter-Landau level scattering is
suppressed in the absence of other scattering mechanisms
(see Section IV.A). This means that the transport from
one point contact to the other is adiabatic. The series
conductance is then simply G
series
= (2e
2
/h)N for two
identical point contacts [N min(N
1
, N
2
) for two dier-
ent point contacts in series]. This expression diers from
Eq. (3.39) if a barrier is present in the point contacts,
since that causes the number N of occupied Landau levels
in the point contact to be less than the number N
wide
of
occupied levels in the wide 2DEG. [In a strong magnetic
eld, N (E
F
E
c
)/h
c
, while N
wide
E
F
/h
c
.] Adi-
abatic transport in a magnetic eld through two point
contacts in series has been studied experimentally by
Kouwenhoven et al.
373
and by Main et al.
374
E. Junction scattering
In the regime of diusive transport, the Hall bar ge-
ometry (a straight current-carrying channel with small
side contacts for voltage drop measurements) is very
convenient, since it allows an independent determina-
tion of the various components of the resistivity ten-
sor. A downscaled Hall bar was therefore a natural
rst choice as a geometry to study ballistic transport
in a 2DEG.
67,68,74,139,178,364
The resistances measured
in narrow-channel geometries are mainly determined by
scattering at the junctions with the side probes.
289
These
scattering processes depend strongly on the junction
shape. This is in contrast to the point contact geom-
etry; compare the very similar results of van Wees et
al.
6
and Wharam et al.
7
on the quantized conductance of
point contacts of a rather dierent design. The strong de-
pendence of the low-eld Hall resistance on the junction
shape was demonstrated theoretically by Baranger and
Stone
358
and experimentally by Ford et al.
77
and Chang
et al.
375
These results superseded many earlier attempts
(listed in Ref.
360
) to explain the discovery by Roukes et
al.
67
of the quenching of the Hall eect without model-
ing the shape of the junction realistically. Baranger and
Stone
358
argued that the rounded corners (present in a re-
alistic situation) at the junction between the main chan-
nel and the side branches lead to a suppression (quench-
ing) of the Hall resistance at low magnetic elds as a
consequence of the horn collimation eect discussed in
Section III.D.1. A Hall bar with straight corners, in con-
trast, does not show a generic suppression of the Hall
resistance,
376,377,378
although quenching can occur for
special parameter values if only a few subbands are oc-
cupied in the channel.
The quenched Hall eect
67,77,375,379
is just one of a
whole variety of magnetoresistance anomalies observed
in narrow Hall bars. Other anomalies are the last Hall
plateau,
67,68,77,139,178,379
reminiscent of quantum Hall
plateaus, but occurring at much lower elds; the negative
Hall resistance,
77
as if the carriers were holes rather than
electrons; the bend resistance,
289,306,364,380
a longitudinal
resistance associated with a current bend, which is nega-
tive at small magnetic elds and zero at large elds, with
an overshoot to a positive value at intermediate elds;
and more.
In Refs.
359
and
360
we have shown that all these phe-
nomena can be qualitatively explained in terms of a
few simple semiclassical mechanisms (reviewed in Sec-
tion III.E.1). The eects of quantum interference and of
quantization of the lateral motion in the narrow conduc-
tor are not essential. These magnetoresistance anomalies
can thus be characterized as classical magneto size eects
in the ballistic regime. In Section II.A, we have discussed
classical size eects in the quasi-ballistic regime, where
the mean free path is larger than the channel width but
smaller than the separation between the voltage probes.
In that regime, the size eects found in a 2DEG were
known from work on metal lms and wires. These ear-
lier investigations had not anticipated the diversity of
magnetoresistance anomalies due to junction scattering
in the ballistic regime. That is not surprising, consider-
ing that the theoretical formalism to describe a resistance
measurement within a mean free path had not been de-
veloped in that context. Indeed, this Landauer-B uttiker
formalism (described in Section III.A) found one of its
earliest applications
268
in the context of the quenching
of the Hall eect, and the success with which the experi-
mental magnetoresistance anomalies can be described by
means of this formalism forms strong evidence for its va-
lidity.
1. Mechanisms
The variety of magnetoresistance anomalies mentioned
can be understood in terms of a few simple characteris-
tics of the curved trajectories of electrons in a classi-
cal billiard in the presence of a perpendicular magnetic
eld.
359,360
At very small magnetic elds, collimation and
scrambling are the key concepts. The gradual widening
of the channel on approaching thejunction reduces the in-
jection/acceptance cone, which is the cone of angles with
the channel axis within which an electron is injected into
the junction or within which an electron can enter the
channel coming from the junction. This is the horn col-
limation eect
329
discussed in Section III.D.1 (see Fig.
65a). If the injection/acceptance cone is smaller than
90

, then the cones of two channels at right angles do

not overlap. That means that an electron approaching
the side probe coming from the main channel will be
reected (Fig. 65a) and will then typically undergo mul-
tiple reections in the junction region (Fig. 65b). The
trajectory is thus scrambled, whereby the probability for
the electron to enter the left or right side probe in a
67
FIG. 65 Classical trajectories in an electron billiard, illus-
trating the collimation (a), scrambling (b), rebound (c), mag-
netic guiding (d), and electron focusing (e) eects. Taken
from C. W. J. Beenakker and H. van Houten, in Electronic
Properties of Multilayers and Low-Dimensional Semiconduc-
tor Structures (J. M. Chamberlain, L. Eaves, and J. C. Por-
tal, eds.). Plenum, London, 1990.
weak magnetic eld is equalized. This suppresses the Hall
voltage. This scrambling mechanism for the quench-
ing of the Hall eect requires a weaker collimation than
the nozzle mechanism put forward by Baranger and
Stone
358
(we return to both these mechanisms in Sec-
tion III.E.3). Scrambling is not eective in the geometry
shown in Fig. 65c, in which a large portion of the bound-
ary in the junction is oriented at approximately 45

with
the channel axis. An electron reected from a side probe
at this boundary has a large probability of entering the
opposite side probe. This is the rebound mechanism
for a negative Hall resistance proposed by Ford et al.
77
At somewhat larger magnetic elds, guiding takes over.
As illustrated in Fig. 65d, the electron is guided by the
magnetic eld along equipotentials around the corner.
Guiding is fully eective when the cyclotron radius l
cycl
becomes smaller than the minimal radius of curvature
r
min
of the corner that is, for magnetic elds greater
than the guiding eld B
g
hk
F
/er
min
. In the regime
B
>

B
g
the junction cannot scatter the electron back
into the channel from which it came. The absence of
backscattering in this case is an entirely classical, weak-
eld phenomenon (cf. Section III.B.2). Because of the ab-
sence of backscattering, the longitudinal resistance van-
ishes, and the Hall resistance R
H
becomes equal to the
contact resistance of the channel, just as in the quantum
Hall eect, but without quantization of R
H
. The contact
resistance R
contact
(h/2e
2
)(/k
F
W) is approximately
independent of the magnetic eld for elds such that
the cyclotron diameter 2l
cycl
is greater than the chan-
nel width W, that is, for elds below B
crit
2 hk
F
/eW
(see Sections III.A and III.B). This explains the occur-
rence of the last plateau in R
H
for B
g
< B
<

B
crit
as a classical eect. At the low-eld end of the plateau,
the Hall resistance is sensitive to geometrical resonances
that increase the fraction of electrons guided around the
corner into the side probe. Figure 65e illustrates the oc-
FIG. 66 Hall resistance as measured (solid curve) by Sim-
mons et al.,
178
and as calculated (dashed curve) for the hard-
wall geometry in the inset (W = 0.8 m and EF = 14 meV).
The dotted line is RH in a bulk 2DEG. Taken from C. W.
J. Beenakker and H. van Houten, in Electronic Properties
of Multilayers and Low-Dimensional Semiconductor Struc-
tures (J. M. Chamberlain, L. Eaves, and J. C. Portal, eds.).
Plenum, London, 1990.
currence of one such a geometrical resonance as a result
of the magnetic focusing of electrons into the side probe,
at magnetic elds such that the separation of the two
perpendicular channels is an integer multiple of the cy-
clotron diameter. This is in direct analogy with electron
focusing in a double-point contact geometry (see Section
III.C) and leads to periodic oscillations superimposed on
the Hall plateau. Another geometrical resonance with
similar eect is discussed in Ref.
360
.
These mechanisms for oscillations in the resistance de-
pend on a commensurability between the cyclotron ra-
dius and a characteristic dimension of the junction, but
do not involve the wavelength of the electrons as an in-
dependent length scale. This distinguishes these geo-
metrical resonances conceptually from the quantum res-
onances due to bound states in the junction considered
in Refs.
376,377
, and
380,381,382
.
2. Magnetoresistance anomalies
In this subsection we compare, following Ref.
360
, the
semiclassical theory with representative experiments on
laterally conned two-dimensional electron gases in high-
mobility GaAs-AlGaAs heterostructures. The calcula-
tions are based on a simulation of the classical trajecto-
ries of a large number (typically 10
4
) of electrons with
the Fermi energy, to determine the classical transmis-
sion probabilities. The resistances then follow from the
B uttiker formula (3.12). We refer to Refs.
359
and
360
for
details on the method of calculation. We rst discuss the
Hall resistance R
H
.
68
Figure 66 shows the precursor of the classical Hall
plateau (the last plateau) in a relatively wide Hall
cross. The experimental data (solid curve) is from a
paper by Simmons et al.
178
The semiclassical calcula-
tion (dashed curve) is for a square-well conning po-
tential of channel width W = 0.8 m (as estimated in
the experimental paper) and with the relatively sharp
corners shown in the inset. The Fermi energy used
in the calculation is E
F
= 14 meV, which corresponds
(via n
s
= E
F
m/h
2
) to a sheet density in the channel
of n
s
= 3.9 10
15
m
2
, somewhat below the value of
4.9 10
15
m
2
of the bulk material in the experiment.
Good agreement between theory and experiment is seen
in Fig. 66. Near zero magnetic eld, the Hall resistance in
this geometry is close to the linear result R
H
= B/en
s
for
a bulk 2DEG (dotted line). The corners are suciently
smooth to generate a Hall plateau via the guiding mech-
anism discussed in Section III.E.1. The horn collimation
eect, however, is not suciently large to suppress R
H
at small B. Indeed, the injection/acceptance cone for
this junction is considerably wider (about 115

) than the
maximal angular opening of 90

required for quenching

of the Hall eect via the scrambling mechanism described
in Section III.E.1.
The low-eld Hall resistance changes drastically if the
channel width becomes smaller, relative to the radius of
curvature of the corners. Figure 67a shows experimen-
tal data by Ford et al.
77
The solid and dotted curves are
for the geometries shown respectively in the upper left
and lower right insets of Fig. 67a. Note that these insets
indicate the gates with which the Hall crosses are de-
ned electrostatically. The equipotentials in the 2DEG
will be smoother than the contours of the gates. The
experiment shows a well-developed Hall plateau with su-
perimposed ne structure. At small positive elds R
H
is
either quenched or negative, depending on the geometry.
The geometry is seen to aect also the width of the Hall
plateau but not the height. In Fig. 67b we give the re-
sults of the semiclassical theory for the two geometries
in the insets, which should be reasonable representations
of the conning potential induced by the gates in the ex-
periment. In the theoretical plot the resistance and the
magnetic eld are given in units of
R
0

h
2e
2

k
F
W
, B
0

hk
F
eW
, (3.42)
where the channel width W for the parabolic conne-
ment used is dened as the separation of the equipoten-
tials at the Fermi energy (W
par
in Section II.F). The
experimental estimates W 90 nm, n
s
1.2 10
15
m
2
imply R
0
= 5.2 k, B
0
= 0.64 T. With these parameters
the calculated resistance and eld scales do not agree
well with the experiment, which may be due in part to
the uncertainties in the modeling of the shape of the ex-
perimental conning potential. The B asymmetry in
the experimental plot is undoubtedly due to asymme-
tries in the cross geometry [in the calculation the geom-
etry has fourfold symmetry, which leads automatically
FIG. 67 Hall resistance as measured (a) by Ford et al.
77
and
as calculated (b). In (a) as well as in (b), the solid curve cor-
responds to the geometry in the upper left inset, the dotted
curve to the geometry in the lower right inset. The insets in
(a) indicate the shape of the gates, not the actual conning
potential. The insets in (b) show equipotentials of the con-
ning potential at EF (thick contour) and 0 (thin contour).
The potential rises parabolically from 0 to EF and vanishes
in the diamond-shaped region at the center of the junction.
Taken from C. W. J. Beenakker and H. van Houten, in Elec-
tronic Properties of Multilayers and Low-Dimensional Semi-
conductor Structures (J. M. Chamberlain, L. Eaves, and J.
C. Portal, eds.). Plenum, London, 1990.
to R
H
(B) = R
H
(B)]. Apart from these dierences,
there is agreement in all the important features: the ap-
pearance of quenched and negative Hall resistances, the
independence of the height of the last Hall plateau on
the smoothness of the corners, and the shift of the onset
of the last plateau to lower elds for smoother corners.
The oscillations on the last plateau in the calculation
(which, as we discussed in Section III.E.1, are due to ge-
ometrical resonances) are also quite similar to those in
the experiment, indicating that these are classical rather
than quantum resonances.
We now turn to the bend resistance R
B
. In Fig. 68 we
show experimental data by Timp et al.
306
(solid curves)
on R
B
R
12,43
and R
H
R
13,24
measured in the same
Hall cross (dened by gates of a shape similar to that in
the lower right inset of Fig. 67a; see the inset of Fig. 68a
for the numbering of the channels). The dashed curves
are calculated for a parabolic conning potential in the
69
FIG. 68 Hall resistance RH R13,24 (a) and bend resis-
tance RB R12,43 (b), as measured (solid curves) by Timp
et al.
306
and as calculated (dashed curves) for the geome-
try in the inset (consisting of a parabolic conning potential
with the equipotentials at EF and 0 shown respectively as
thick and thin contours; the parameters are W = 100 nm and
EF = 3.9 meV). The dotted line in (a) is RH in a bulk 2DEG.
Taken from C. W. J. Beenakker and H. van Houten, in Elec-
tronic Properties of Multilayers and Low-Dimensional Semi-
conductor Structures (J. M. Chamberlain, L. Eaves, and J.
C. Portal, eds.). Plenum, London, 1990.
channels (with the experimental values W = 100 nm,
E
F
= 3.9 meV) and with corners as shown in the inset of
Fig. 68a. The calculated quenching of the Hall resistance
and the onset of the last plateau are in good agreement
with the experiment, and also the observed overshoot of
the bend resistance around 0.2 T as well as the width of
the negative peak in R
B
around zero eld are well de-
scribed by the calculation. The calculated height of the
negative peak, however, is too small by more than a fac-
tor of 2. We consider this disagreement to be signicant
in view of the quantitative agreement with the other fea-
tures in both R
B
and R
H
. The negative peak in R
B
is
due to the fact that the collimation eect couples the
current source 1 more strongly to voltage probe 3 than
to voltage probe 4, so R
B
V
4
V
3
is negative for small
magnetic elds (at larger elds the Lorentz force destroys
collimation by bending the trajectories, so R
B
shoots up
to a positive value until guiding takes over and brings R
B
down to zero by eliminating backscattering at the junc-
tion). The discrepancy in Fig. 68b thus seems to indicate
that the semiclassical calculation underestimates the col-
limation eect in this geometry. The positive overshoot
of R
B
seen in Fig. 68b is found only for rounded cor-
ners. This explains the near absence of the eect in the
calculation of Kirczenow
381
for a junction with straight
corners.
For a discussion of the temperature dependence of the
magnetoresistance anomalies, we refer to Ref.
360
. Here
it suces to note that the experiments discussed were
carried out at temperatures around 1 K, for which we
expect the zero-temperature semiclassical calculation to
be appropriate. At lower temperatures the eects of
quantum mechanical phase coherence that have been ne-
glected will become more important. At higher temper-
atures the thermal average smears out the magnetoresis-
tance anomalies and eventually inelastic scattering causes
a transition to the diusive transport regime in which the
resistances have their normal B-dependence.
3. Electron waveguide versus electron billiard
The overall agreement between the experiments and
the semiclassical calculations is remarkable in view of
the fact that the channel width in the narrowest struc-
tures considered is comparable to the Fermi wavelength.
When the rst experiments on these electron waveg-
uides appeared, it was expected that the presence of
only a small number of occupied transverse waveguide
modes would fundamentally alter the nature of electron
transport.
68
The results of Refs.
359
and
356
show instead
that the modal structure plays only a minor role and
that the magnetoresistance anomalies observed are char-
acteristic for the classical ballistic transport regime. The
reason that a phenomenon such as the quenching of the
Hall eect has been observed only in Hall crosses with
narrow channels is simply that the radius of curvature of
the corners at the junction is too small compared with
the channel width in wider structures. This is not an
essential limitation, and the various magnetoresistance
anomalies discussed here should be observable in macro-
copic Hall bars with articially smoothed corners, pro-
vided of course that the dimensions of the junction re-
main well below the mean free path. Ballistic transport
is essential, but a small number of occupied modes is not.
Although we believe that the characteristic features
of the magnetoresistance anomalies are now understood,
several interesting points of disagreement between theory
and experiment remain that merit further investigation.
One of these is the discrepancy in the magnitude of the
negative bend resistance at zero magnetic eld noted be-
fore. The disappearance of a region of quenched Hall
resistance at low electron density is another unexpected
observation by Chang et al.
375
and Roukes et al.
383
The
70
semiclassical theory predicts a universal behavior (for a
given geometry) if the resistance and magnetic eld are
scaled by R
0
and B
0
dened in Eq. (3.42). For a square-
well conning potential the channel width W is the same
at each energy, and since B
0
k
F
one would expect
the eld region of quenched Hall resistance to vary with
the electron density as

n
s
. For a more realistic smooth
conning potential, W depends on E
F
and thus on n
s
as well, in a way that is dicult to estimate reliably. In
any case, the experiments point to a systematic disap-
pearance of the quench at the lowest densities, which is
not accounted for by the present theory (and has been
attributed by Chang et al.
375
to enhanced diraction at
low electron density as a result of the increase in the
Fermi wavelength). For a detailed investigation of de-
partures from classical scaling, we refer to a paper by
Roukes et al.
384
As a third point, we mention the curious
density dependence of the quenching observed in approx-
imately straight junctions by Roukes et al.,
383
who nd
a low-eld suppression of R
H
that occurs only at or near
certain specic values of the electron density. The semi-
classical model applied to a straight Hall cross (either
dened by a square well or by a parabolic conning po-
tential) gives a low-eld slope of R
H
close to its bulk
2D value. The fully quantum mechanical calculations
for a straight junction
376,381
do give quenching at special
parameter values, but not for the many-mode channels
in this experiment (in which quenching occurs with as
many as 10 modes occupied, whereas in the calculations
a straight cross with more than 3 occupied modes in the
channel does not show a quench).
In addition to the points of disagreement discussed,
there are ne details in the measured magnetoresistances,
expecially at the lowest temperatures (below 100 mK),
which are not obtained in the semiclassical approxima-
tion. The quantum mechanical calculations
358,376,377,381
show a great deal of ne structure due to interference of
the waves scattered by the junction. The ne structure
in most experiments is not quite as pronounced as in the
calculations presumably partly as a result of a loss of
phase coherence after many multiple scatterings in the
junction. The limited degree of phase coherence in the
experiments and the smoothing eect of a nite temper-
ature help to make the semiclassical model work so well
even for the narrowest channels. We draw attention to
the fact that classical chaotic scattering can also be a
source of irregular resistance uctuations (see Ref.
360
).
Some of the most pronounced features in the quan-
tum mechanical calculations are due to transmission res-
onances that result from the presence of bound states in
the junction.
376,377,380,381,382
In Section III.E.1 we have
discussed a dierent mechanism for transmission reso-
nances that has a classical, rather than a quantum me-
chanical, origin. As mentioned in Section III.E.2, the
oscillations on the last Hall plateau observed experimen-
tally are quite well accounted for by these geometrical
resonances. One way to distinguish experimentally be-
tween these resonance mechanisms is by means of the
temperature dependence, which should be much weaker
for the classical than for the quantum eect. One would
thus conclude that the uctuations in Fig. 67a, measured
by Ford et al.
77
at 4.2 K, have a classical origin, while the
ne structure that Ford et al.
385
observe only at mK tem-
peratures (see below) is intrinsically quantum mechani-
cal.
The dierences between the semiclassical and the
quantum mechanical models may best be illustrated by
considering once again the quenching of the Hall eect,
which has the most subtle explanation and is the most
sensitive to the geometry among the magnetoresistance
anomalies observed in the ballistic regime. The classi-
cal scrambling of the trajectories after multiple reec-
tions suppresses the asymmetry between the transmission
probabilities t
l
and t
r
to enter the left or right voltage
probe, and without this transmission asymmetry there
can be no Hall voltage. We emphasize that this scram-
bling mechanism is consistent with the original ndings
of Baranger and Stone
358
that quenching requires col-
limation. The point is that the collimation eect leads
to nonoverlapping injection/acceptance cones of two per-
pendicular channels, which ensures that electrons can-
not enter the voltage probe from the current source di-
rectly, but rather only after multiple reections (cf. Sec-
tion III.E.1). In this way a rather weak collimation to
within an injection/acceptance cone of about 90

angu-
lar opening is sucient to induce a suppression of the
Hall resistance via the scrambling mechanism.
Collimation can also suppress R
H
directly by strongly
reducing t
l
and t
r
relative to t
s
(the probability for trans-
mission straight through the junction). This nozzle mech-
anism, introduced by Baranger and Stone,
358
requires a
strong collimation of the injected beam in order to af-
fect R
H
appreciably. In the geometries considered here,
we nd that quenching of R
H
is due predominantly to
scrambling and not to the nozzle mechanism (t
l
and t
r
each remain more than 30% of t
s
), but data by Baranger
and Stone
358
show that both mechanisms can play an
important role.
There is a third proposed mechanism for the quench-
ing of the Hall eect,
376,377
which is the reduction of
the transmission asymmetry due to a bound state in the
junction. The bound state mechanism is purely quan-
tum mechanical and does not require collimation (in con-
trast to the classical scrambling and nozzle mechanisms).
Numerical calculations have shown that it is only eec-
tive in straight Hall crosses with very narrow channels
(not more than three modes occupied), and even then
for special values of the Fermi energy only. Although
this mechanism cannot account for the experiments per-
formed thus far, it may become of importance in future
work. A resonant suppression of the Hall resistance may
also occur in strong magnetic elds, in the regime where
the Hall resistance in wide Hall crosses would be quan-
tized. Such an eect is intimately related to the high-
eld Aharonov-Bohm magnetoresistance oscillations in a
singly connected geometry (see Section IV.D). Ford et
71
FIG. 69 Measured Hall resistance in an abrupt (a) and in
a widened (b) cross as a function of B in the strong eld
regime. Large uctuations are resolved at the low tempera-
ture of 22 mK. The dotted curves indicate the reproducibility
of the measurement. Taken from C. J. B. Ford et al., Surf.
Sci. 229, 298 (1990).
al.
385
have observed oscillations superimposed on quan-
tized Hall plateaux at low temperatures in very narrow
crosses of two dierent shapes (see Fig. 69). The strong
temperature dependence indicates that these oscillations
are resonances due to the formation of bound states in
the cross.
306,385,386
F. Tunneling
In this section we review recent experiments on tun-
neling through potential barriers in a two-dimensional
electron gas. Subsection III.F.1 deals with resonant tun-
neling through a bound state in the region between two
barriers. Resonant tunneling has previously been studied
extensively in layered semiconductor heterostructures for
transport perpendicular to the layers.
387,388,389
For ex-
ample, a thin AlGaAs layer embedded between two GaAs
layers forms a potential barrier, whose height and width
can be tailored with great precision by means of advanced
growth techniques (such as molecular beam epitaxy). Be-
cause of the free motion in the plane of the layers, one
can only realize bound states with respect to one direc-
tion. Tunneling resonances are consequently smeared out
over a broad energy range. A 2DEG oers the possibil-
ity of connement in all directions and thus of a sharp
resonance. A gate allows one to dene potential barriers
of adjustable height in the 2DEG. In contrast, the het-
erostructure layers form xed potential barriers, so one
needs to study a current-voltage characteristic to tune
the system through a resonance (observable as a peak in
the I V curve). The gate-induced barriers in a 2DEG
oer a useful additional degree of freedom, allowing a
study of resonant tunneling in the linear response regime
of small applied voltages (to which we limit the discussion
in this review). A drawback of these barriers is that their
shape cannot be precisely controlled, or modeled, so that
a description of the tunneling process will of necessity be
qualitative.
Subsection III.F.2 deals with the eects of Coulomb re-
pulsion on tunneling in a 2DEG. The electrostatic eects
of charge buildup in the 1D potential well formed by het-
erostructure layers have received considerable attention
in recent years.
389,390
Because of the large capacitance of
the potential well in this case (resulting from the large
surface area of the layers) these are macroscopic eects,
involving a large number of electrons. The 3D potential
well in a 2DEG nanostructure, in contrast, can have a
very small capacitance and may contain a few electrons
only. The tunneling of a single electron into the well
will then have a considerable eect on the electrostatic
potential dierence with the surrounding 2DEG. For a
small applied voltage this eect of the Coulomb repulsion
can completely suppress the tunneling current. In metals
this Coulomb blockade of tunneling has been studied
extensively.
391
In those systems a semiclassical descrip-
tion suces. The large Fermi wavelength in a 2DEG
should allow the study of quantum mechanical eects on
the Coulomb blockade or, more generally, of the inter-
play between electron-electron interactions and resonant
tunneling.
318,392,393
1. Resonant tunneling
The simplest geometry in which one might expect to
observe transmission resonances is formed by a single po-
tential barrier across a 2DEG channel. Such a geometry
was studied by Washburn et al.
394
in a GaAs-AlGaAs
heterostructure containing a 2-m-wide channel with a
45-nm-long gate on top of the heterostructure. At low
temperatures (around 20 mK) an irregular set of peaks
was found in the conductance as a function of gate volt-
age in the region close to the depletion threshold. The
amplitude of the peaks was on the order of e
2
/h. The ori-
gin of the eect could not be pinned down. The authors
examine the possibility that transmission resonances as-
sociated with a square potential barrier are responsible
for the oscillations in the conductance, but also note that
the actual barrier is more likely to be smooth on the
scale of the wavelength. For such a smooth barrier the
transmission probability as a function of energy does not
show oscillations. It seems most likely that the eect
is disorder-related. Davies and Nixon
395
have suggested
that some of the structure observed in this experiment
could be due to potential uctuations in the region under
the gate. These uctuations can be rather pronounced
close to the depletion threshold, due to the lack of screen-
72
FIG. 70 Resistance versus gate voltage of a cavity (dened
by gates on top of a GaAs-AlGaAs heterostructure; see in-
set), showing plateau like features (for R
<

h/2e
2
) and tun-
neling resonances (for R
>

h/2e
2
). The left- and right-hand
curves refer to the adjacent resistance scales. Taken from C.
G. Smith et al., Surf. Sci. 228, 387 (1990).
ing in the low-density electron gas. A quantum mechan-
ical calculation of transmission through such a uctuat-
ing barrier has not been performed. As discussed below,
conductance peaks of order e
2
/h occur in the case of res-
onant tunneling via localized states in the barrier (associ-
ated with impurities), a mechanism that might well play
a role in the experiment of Washburn et al.
394
In pursuit of resonant tunneling in a 2DEG, Chou
et al.
396
have fabricated double-barrier devices involv-
ing two closely spaced short gates across a wide GaAs-
AlGaAs heterostructure. Both the spacing and the
length of the gates were 100 nm. They observed a peak
in the transconductance (the derivative of the chan-
nel current with respect to the gate voltage), which
was attributed to resonant tunneling through a quasi-
bound state in the 2D potential well between the bar-
riers. Palevski et al.
397
have also investigated trans-
port through two closely spaced potential barriers in a
double-gate structure, but they did not nd evidence for
transmission resonances. A 3D potential well has truly
bound states and is expected to show the strongest trans-
mission resonances. Transport through such a cavity or
quantum box has been studied theoretically by several
authors.
318,333,382,398
Experiments have been performed
by Smith et al.
399,400,401
Their device is based on a quan-
tum point contact, but contains two potential barriers
that separate the constriction from the wide 2DEG re-
gions (see the inset of Fig. 70). As the negative gate volt-
age is increased, a potential well is formed between the
two barriers, resulting in connement in all directions.
The tunneling regime corresponds to a resistance R that
is greater than h/2e
2
. It is also possible to study the bal-
listic regime R < h/2e
2
when the height of the potential
barriers is less than the Fermi energy. In this regime the
transmission resonances are similar to the resonances in
long quantum point contacts (these are determined by
an interplay of tunneling through evanescent modes and
reection at the entrance and exit of the point contact;
cf. Section III.B). Results of Smith et al.
399,400,401
for the
resistance as a function of gate voltage at 330 mK are re-
produced in Fig. 70. In the tunneling regime (R > h/2e
2
)
giant resistance oscillations are observed. A regular se-
ries of smaller resistance peaks is found in the ballis-
tic regime (R < h/2e
2
). Martin-Moreno and Smith
333
have modeled the electrostatic potential in the device of
Refs.
399,400,401
and have performed a quantum mechani-
cal calculation of the resistance. Very reasonable agree-
ment with the experimental data in the ballistic regime
was obtained. The tunneling regime was not compared
in detail with the experimental data. The results were
found to depend rather critically on the assumed chape of
the potential, in particular on the rounding of the tops of
the potential barriers. Martin-Moreno and Smith also in-
vestigated the eects of asymmetries in the device struc-
ture on the tunneling resonances and found in particular
that small dierences in the two barrier heights (of or-
der 10%) lead to a sharp suppression of the resonances,
a nding that sheds light on the fact that they were ob-
served in certain devices only. Experimentally, the ef-
fect of a magnetic eld on the oscillations in the resis-
tance versus gate voltage was also investigated.
399,400,401
A strong suppression of the peaks was found in relatively
weak magnetic elds (of about 0.3 T).
Tunneling through a cavity, as in the experiment by
Smith et al.,
399,400,401
is formally equivalent to tunneling
through an impurity state (see, e.g., Refs.
402
and
403
).
The dramatic subthreshold structure found in the con-
ductance of quasi-one-dimensional MOSFETs has been
interpreted in terms of resonant tunneling through a se-
ries of localized states.
32,35,36,37
Kopley et al.
404
have ob-
served large conductance peaks in a MOSFET with a
split gate (see Fig. 71). Below the 200-nm-wide slot in
the gate, the inversion layer is interrupted by a poten-
tial barrier. Pronounced conductance peaks were seen at
0.5 K as the gate voltage was varied in the region close to
threshold (see Fig. 72). No clear correlation was found
between the channel width and the peak spacing or am-
plitude. The peaks were attributed to resonant transmis-
sion through single localized states associated with bound
states in the Si band gap in the noninverted region under
the gate.
The theory of resonant tunneling of noninteract-
ing electrons through localized states between two-
73
FIG. 71 Schematic diagram of a Si MOSFET with a split
gate (a), which creates a potential barrier in the inversion
layer (b). Taken from T. E. Kopley et al. Phys. Rev. Lett.
61, 1654 (1988).
dimensional reservoirs was developed by Xue and Lee
405
(see also Refs.
159
and
406
). If the resonances are well sepa-
rated in energy, a single localized state will give the dom-
inant contribution to the transmission probability. The
maximum conductance on resonance is then e
2
/h (for
one spin direction), regardless of the number of channels
N in the reservoirs.
405,406
This maximum (which may be
interpreted as a contact resistance, similar to that of a
quantum point contact) is attained if the localized state
has identical leak rates
L
/h and
R
/h to the left and
right reservoirs. Provided these leak rates are small (cf.
Section IV.D) the conductance G as a function of Fermi
energy E
F
is a Lorentzian centered around the resonance
energy E
0
:
G(E
F
) =
e
2
h

R
(E
F
E
0
)
2
+
1
4
(
L
+
R
)
2
. (3.43)
This is the Breit-Wigner formula of nuclear physics.
93
For an asymmetrically placed impurity the peak height
is reduced below e
2
/h (by up to a factor 4
R
/
L
if
L

R
).
The amplitudes of the peaks observed by Kopley et
al.
404
were found to be in agreement with this prediction,
while the line shape of an isolated peak could be well de-
scribed by a Lorentzian (see inset of Fig. 72). (Most of
the peaks overlapped, hampering a line-shape analysis).
In addition, they studied the eect of a strong magnetic
eld on the conductance peaks and found that the ampli-
tudes of most peaks were substantially suppressed. This
was interpreted as a reduction of the leak rates because
of a reduced overlap between the wave functions on the
FIG. 72 Oscillations in the conductance as a function of gate
voltage at 0.5 K are attributed to resonant tunneling through
localized states in the potential barrier. A second trace is
shown for a magnetic eld of 6 T (with a horizontal oset of
0.04 V). The inset is a close-up of the largest peak at 6 T,
together with a Lorentzian t. Taken from T. E. Kopley et
al. Phys. Rev. Lett. 61, 1654 (1988).
impurity and the reservoirs. The amplitude of one partic-
ular peak was found to be unaected by the eld, indica-
tive of a symmetrically placed impurity in the barrier
(
R
=
L
), while the width of that peak was reduced,
in agreement with Eq. (3.43). This study therefore ex-
hibits many characteristic features of resonant tunneling
through a single localized state.
Transmission resonances due to an impurity in a quan-
tum point contact or narrow channel have been studied
theoretically in Refs.
241,407
, and
408
. In an experiment
it may be dicult to distinguish these resonances from
those associated with reection at the entrance and exit
of the quantum point contact (discussed in Section III.B).
A conductance peak associated with resonant tunneling
through an impurity state in a quantum point contact
was reported by McEuen et al.
409
The experimental re-
sults are shown in Fig. 73. The resonant tunneling peak is
observed near the onset of the rst conductance plateau,
where G < 2e
2
/h. A second peak seen in Fig. 73 was
conjectured to be a signature of resonant scattering, in
analog with similar processes known in atomic physics.
410
We want to conclude this subsection on transmission
resonances by discussing an experiment by Smith et
al.
401,411
on what is essentially a Fabry-Perot interfer-
ometer. The device consists of a point contact with ex-
ternal reectors in front of its entrance and exit. The
reectors are potential barriers erected by means of two
additional gate electrodes (see Fig. 74a). By varying
the gate voltage on the external reectors of this device,
Smith et al. could tune the eective cavity length with-
out changing the width of the narrow section. This ex-
74
FIG. 73 Conductance as a function of gate voltage for a quan-
tum point contact at 0.55 K. The inset is a close-up of the
low-conductance regime, showing peaks attributed to trans-
mission resonances associated with impurity states in the con-
striction. Taken from P. L. McEuen et al., Surf. Sci. 229, 312
(1990).
FIG. 74 (a) Schematic diagram of a constriction with two
adjustable external reectors dened by gates on top of a
GaAs-AlGaAs heterostructure. (b) Plot of the constriction
resistance as a function of gate voltage with the external re-
ector gates (Y1, Y2) grounded. Inset: Fabry-Perot-type
transmission resonances due to a variation of the gate voltage
on the reectors (Y1, Y2) (bottom panel), and Fourier power
spectrum (top panel). Taken from C. G. Smith et al., Surf.
Sci. 228, 387 (1990).
periment is therefore more controlled than the quantum
dot experiment
399,400,401
discussed earlier. The resulting
periodic transmission resonances are reproduced in Fig.
74b. A new oscillation appears each time the separa-
tion between the reectors increases by
F
/2. A numer-
ical calculation for a similar geometry was performed by
Avishai et al.
412
The signicance of this experiment is
that it is the rst clear realization of an electrostatically
tuned electron interferometer. Such a device has poten-
tial transistor applications. Other attempts to fabricate
an electrostatic interferometer have been less succesful.
The electrostatic Aharonov-Bohmeect in a ring was dis-
cussed in Section II.D. The solid-state analogue of the
microwave stub tuner (proposed by Sols et al.
413
and by
Datta
414
) was studied experimentally by Miller et al.
415
The idea is to modify the transmission through a nar-
row channel by changing the length of a side branch (by
means of a gate across the side branch). Miller et al. have
fabricated such a T-shaped conductor and found some
evidence for the desired eect. Much of the structure
was due, however, to disorder-related conductance uc-
tuations. The electrostatic Aharonov-Bohm eect had
similar problems. Transport in a long and narrow chan-
nel is simply not fully ballistic, because of partially diuse
boundary scattering and impurity scattering. The device
studied by Smith et al. worked because it made use of a
very short constriction (a quantum point contact), while
the modulation of the interferometer length was done ex-
ternally in the wide 2DEG, where the eects of disorder
are much less severe (in high-mobility material).
2. Coulomb blockade
In this subsection we would like to speculate on
the eects of electron-electron interactions on tunneling
through impurities in narrow semiconductor channels, in
relation to a recent paper in which Scott-Thomas et al.
416
announced the discovery of conductance oscillations pe-
riodic in the density of a narrow Si inversion layer. The
device features a continuous gate on top of a split gate,
as illustrated schematically in Fig. 75. In the experi-
ment, the voltage on the upper gate is varied while the
split-gate voltage is kept constant. Figure 76 shows the
conductance as a function of gate voltage at 0.4 K, as
well as a set of Fourier power spectra obtained for de-
vices of dierent length. A striking pattern of rapid pe-
riodic oscillations is seen. No correlation is found be-
tween the periodicity of the oscillations and the channel
length, in contrast to the transmission resonances in bal-
listic constrictions discussed in Sections III.B and III.F.1.
The oscillations die out as the channel conductance in-
creases toward e
2
/h 4 10
5

1
. The conductance
peaks are relatively insensitive to a change in temper-
ature, while the minima depend exponentially on tem-
perature as exp(E
a
/k
B
T), with an activation energy
E
a
50 eV. Pronounced nonlinearities occur in the
current as a function of source-drain voltage. An inter-
75
FIG. 75 Schematic cross sectional (a) and top (b) view of
a double-gate Si MOSFET device. The lower split gate is
at a negative voltage, conning the inversion layer (due to
the positive voltage on the upper gate) to a narrow channel.
Taken from J. H. F. Scott-Thomas et al., Phys. Rev. Lett. 62,
583 (1989).
pretation in terms of pinned charge density waves was
suggested,
416
based on a model due to Larkin and Lee
417
and Lee and Rice.
418
In such a model, one expects the
conductance to be thermally activated, because of the
pinning of the charge density wave by impurities in the
one-dimensional channel. The activation energy is de-
termined by the most strongly pinned segment in the
channel, and periodic oscillations in the conductance as
a function of gate voltage correspond to the condition
that an integer number of electrons is contained between
the two impurities delimiting that specic segment. The
same interpretation has been given to a similar eect
observed in a narrow channel in a GaAs-AlGaAs het-
erostructure by Meirav et al.
85
We have proposed
419
an alternative single-electron ex-
planation of the remarkable eect discovered by Scott-
Thomas et al.,
416
based upon the concept of the Coulomb
blockade of tunneling mentioned at the beginning of this
section. Likharev
391
and Mullen et al.
420
have studied
theoretically the possibility of removing the Coulomb
FIG. 76 Top panel: Periodic oscillations in the conductance
versus gate voltage at 0.4 K for a 10-m-long inversion chan-
nel. Next three panels: Fourier power spectra of this curve
and of data obtained for 2- and 1-m-long channels. Bottom
panel: Fourier spectrum for the 1-m-long device in a mag-
netic eld of 6 T. Taken from J. H. F. Scott-Thomas et al.,
Phys. Rev. Lett. 62, 583 (1989).
blockade by capacitive charging (by means of a gate elec-
trode) of the region between two tunnel barriers. They
found that the conductance of this system exhibits peri-
odic peaks as a function of gate voltage, due to the mod-
ulation of the net charge (mod e) on the interbarrier re-
gion. Following the theoretical papers,
391,420
the authors
76
in Ref.
419
proposed that the current through the channel
in the experiment of Scott-Thomas et al.
416
is limited by
tunneling through potential barriers constituted by two
dominant scattering centers that delimit a segment of the
channel (see Fig. 77). Because the number of electrons lo-
calized in the region between the two barriers is necessar-
ily an integer, a charge imbalance, and hence an electro-
static potential dierence, arises between this region and
the adjacent regions connected to wide electron gas reser-
voirs. As the gate voltage is varied, the resulting Fermi
level dierence E
F
oscillates in a sawtooth pattern be-
tween e, where = e/2C and C = C
1
+ C
2
is the
eective capacitance of the region between the two barri-
ers. The single-electron charging energy e
2
/2C maintains
the Fermi level dierence until E
F
= e (this is the
Coulomb blockade). When E
F
= e, the energy re-
quired for the transfer of a single electron to (or from)
the region between the two barriers vanishes so that the
Coulomb blockade is removed. The conductance then
shows a maximum at low temperatures T and source-
drain voltages V (k
B
T/e, V
<

). We note that in the

case of very dierent tunneling rates through the two bar-
riers, one would expect steps in the current as a function
of source-drain voltage, which are not observed in the
experiments.
85,416
For two similar barriers this Coulomb
staircase is suppressed.
420
The oscillation of the Fermi
energy as the gate voltage is varied thus leads to a se-
quence of conductance peaks. The periodicity of the os-
cillations corresponds to the addition of a single electron
to the region between the two scattering centers form-
ing the tunnel barriers, so the oscillations are periodic
in the density, as in the experiment. This single-electron
tunneling mechanism also explains the observed activa-
tion of the conductance minima and the insensitivity to
a magnetic eld.
85,416
The capacitance associated with
the region between the scattering centers is hard to as-
certain. The experimental value of the activation energy
E
a
50 eV would imply C e
2
/2E
a
10
15
F. Kast-
ner et al.
421
argue that the capacitance in the device is
smaller than this amount by an order of magnitude (the
increase in the eective capacitance due to the presence
of the gate electrodes is taken into account in their esti-
mate). In addition, they point to a discrepancy between
the value for the Coulomb blockade inferred from the
nonlinear conductance and that from the thermal acti-
vation energy. The temperature dependence of the os-
cillatory conductance was found to be qualitatively dif-
ferent in the experiment by Meirav et al.
85
At elevated
temperatures an exponential T-dependence was found,
but at low temperatures the data suggest a much weaker
T-dependence. It is clear that more experimental and
theoretical work is needed to arrive at a denitive inter-
pretation of this intriguing phenomenon.
It would be of interest to study the eects of the
Coulomb blockade of tunneling in a more controlled fash-
ion in a structure with two adjustable potential barri-
ers. Such an experiment was proposed by Glazman and
Shekter,
422
who studied theoretically a system similar to
FIG. 77 Schematic representation of the bottom of the con-
duction band Ec, and Fermi energy EF in the device of Fig.
76 along the channel. The band bending at the connections
of the narrow channel to the wide source S and drain D re-
gions arises from the higher threshold for the electrostatic
creation of a narrow inversion layer by a gate (shaded part).
Tunnel barriers associated with two scattering centers are
shown. The maximum Fermi energy dierence sustainable
by the Coulomb blockade, EF = e (where = e/2C,
with C = C1 +C2), is indicated. Taken from H. van Houten
and C. W. J. Beenakker, Phys. Rev. Lett. 63, 1893 (1989).
the cavity of the experiments by Smith et al.
399,400,401
(discussed in Section III.F.1). A diculty with this type
of device is, as pointed out in Ref.
422
, that a variation
in gate voltage aects the barrier height (and thus their
transparency) as well as the charge in the cavity. This
is expected to lead to an exponential damping of the os-
cillations due to the Coulomb blockade.
391,420
A charac-
teristic feature of these oscillations is their insensitivity
to an applied magnetic eld, which can serve to distin-
guish the eect from oscillations due to resonant tunnel-
ing (Section III.F.1). The eld dependence of the peaks
observed by Smith et al.
399,400,401
in the tunneling regime
was not reported, so the question of whether or not the
Coulomb oscillations are observed in their experiment re-
mains unanswered. In our opinion, substantial progress
could be made with the development of thin tunnel bar-
riers of larger height, which would be less sensitive to the
application of an external gate voltage. If our interpre-
tation of the experiments by Scott-Thomas et al.
416
and
Meirav et al.
85
is correct, such tunneling barriers might
be formed by the incorporation of negatively charged im-
purities (e.g., ionized acceptors) in a narrow electron gas
channel. This speculation is based on the fact that such
acceptor impurities are present in the Si inversion layers
of the experiment of Scott-Thomas et al.,
416
as well as
in the p-n junctions employed for lateral connement by
Meirav et al.
85
As we were completing this review, we learned of
several experiments that demonstrate the Coulomb
blockade in split-gate conned GaAs-AlGaAs
heterostructures.
423,424,425
These experiments should
open the way for the controlled study of the eects of
77
Coulomb interactions on tunneling in semiconductor
nanostructures.
IV. ADIABATIC TRANSPORT
A. Edge channels and the quantum Hall eect
In this section we give an overview of the characteris-
tics of adiabatic transport via edge channels in the regime
of the quantum Hall eect as a background to the follow-
ing sections. We restrict ourselves here to the integer
quantum Hall eect, where the edge channels can be de-
scribed by single-electron states. Recent developments on
adiabatic transport in the regime of the fractional quan-
tum Hall eect (which is fundamentally a many-body
eect) will be considered in Section IV.C.
1. Introduction
Both the quantum Hall eect (QHE) and the quantized
conductance of a ballistic point contact are described by
the same relation, G = Ne
2
/h, between the conductance
G and the number N of propagating modes at the Fermi
level (counting both spin directions separately). The
smooth transition from zero-eld quantization to QHE
that follows from this relation is evident from Fig. 48.
The nature of the modes is very dierent, however, in
weak and strong magnetic elds. As we discussed in Sec-
tion III.A.1, the propagating modes in a strong magnetic
eld consist of edge states, which interact with one of the
sample edges only. Edge states with the same mode in-
dex are referred to collectively as an edge channel. Edge
channels at opposite edges propagate in opposite direc-
tions. In a weak magnetic eld, in contrast, the modes
consist of magnetoelectric subbands that interact with
both edges. In that case there is no spatial separation of
modes propagating in opposite directions.
The dierent spatial extension of edge channels and
magnetoelectric subbands leads to an entirely dierent
sensitivity to scattering processes in weak and strong
magnetic elds. Firstly, the zero-eld conductance quan-
tization is destroyed by a small amount of elastic scatter-
ing (due to impurities or roughness of the channel bound-
aries; cf. Refs.
313,316,317,407
, and
408
), while the QHE is
robust to scattering.
97
This dierence is a consequence
of the suppression of backscattering by a magnetic eld
discussed in Section III.B.2, which itself follows from
the spatial separation at opposite edges of edge channels
moving in opposite directions. Second, the spatial sepa-
ration of edge channels at the same edge in the case of
a smooth conning potential opens up the possibility of
adiabatic transport (i.e., the full suppression of interedge
channel scattering). In weak magnetic elds, adiabatic-
ity is of importance within a point contact, but not on
longer length scales (cf. Sections III.B.1 and III.D.1).
In a wide 2DEG region, scattering among the modes
in weak elds establishes local equilibrium on a length
scale given by the inelastic scattering length (which in a
high-mobility GaAs-AlGaAs heterostructure is presum-
ably not much longer than the elastic scattering length
l 10 m). The situation is strikingly dierent in a
strong magnetic eld, where the selective population and
detection of edge channels observed by van Wees et al.
426
has demonstrated the persistence of adiabaticity outside
the point contact.
In the absence of interedge channel scattering the vari-
ous edge channels at the same boundary can be occupied
up to dierent energies and consequently carry dierent
amounts of current. The electron gas at the edge of the
sample is then not in local equilibrium. Over some long
distance (which is not yet known precisely) adiabatic-
ity breaks down, leading to a partial equilibration of the
edge channels. However, as demonstrated by Komiyama
et al.
427
and by others,
307,428,429,430
local equilibrium is
not fully established even on macroscopic length scales
exceeding 0.25 mm. Since local equilibrium is a prerequi-
site for the use of a local resistivity tensor, these ndings
imply a nonlocality of the transport that had not been
anticipated in theories of the QHE (which are commonly
expressed in terms of a local resistivity).
97
A theory of the QHE that is able to explain anoma-
lies resulting from the absence of local equilibrium has to
take into account the properties of the current and volt-
age contacts used to measure the Hall resistance. That
is not necessary if local equilibrium is established at the
voltage contacts, for the fundamental reason that two
systems in equilibrium that are in contact have identi-
cal electrochemical potentials. In the Landauer-B uttiker
formalism described in Section III.A.2, the contacts are
modeled by electron gas reservoirs and the resistances are
expressed in terms of transmission probabilities of prop-
agating modes at the Fermi level from one reservoir to
the other. This formalism is not restricted to zero or
weak magnetic elds, but can equally well be applied to
the QHE, where edge channels form the modes. In this
way B uttiker could show
112
that the nonideality of the
coupling of the reservoirs to the conductor aects the ac-
curacy of the QHE in the absence of local equilibrium.
An ideal contact in the QHE is one that establishes an
equilibrium population among the outgoing edge chan-
nels by distributing the injected current equally among
these propagating modes (this is the equipartitioning of
current discussed for an ideal electron waveguide in Sec-
tion III.A.2). A quantum point contact that selectively
populates certain edge channels
426
can thus be seen as an
extreme example of a nonideal, or disordered, contact.
2. Edge channels in a disordered conductor
After this general introduction, let us now discuss in
some detail how edge channels are formed at the bound-
ary of a 2DEG in a strong magnetic eld. In Section
III.A.1 we discussed the edge states in the case of a nar-
row channel without disorder, relevant for the point con-
78
tact geometry. Edge states were seen to originate from
Landau levels, which in the bulk lie below the Fermi level
but rise in energy on approaching the sample boundary
(cf. Fig. 40b). The point of intersection of the nth Lan-
dau level (n = 1, 2, . . .) with the Fermi level forms the site
of edge states belonging to the nth edge channel. The
number N of edge channels at E
F
is equal to the number
of bulk Landau levels below E
F
. This description can eas-
ily be generalized to the case of a slowly varying potential
energy landscape V (x, y) in the 2DEG, in which case a
semiclassical analysis can be applied.
431
The energy E
F
of an electron at the Fermi level in a strong magnetic
eld contains a part (n
1
2
) h
c
due to the quantized cy-
clotron motion and a part
1
2
g
B
B (depending on the
spin direction) from spin splitting. The remainder is the
energy E
G
due to the electrostatic potential
E
G
= E
F
(n
1
2
) h
c

1
2
g
B
B. (4.1)
The cyclotron orbit center R is guided along equipoten-
tials of V at the guiding center energy E
G
. As derived in
Section II.G.2, the drift velocity v
drift
of the orbit center
(known as the guiding center drift or EB drift) is given
by
v
drift
(R) =
1
eB
2
V (R) B, (4.2)
which indeed is parallel to the equipotentials. An impor-
tant distinction with the weak-eld case of Section II.G.2
is that the spatial extension of the cyclotron orbit can
now be neglected, so V is evaluated at the position of the
orbit center in Eq. (4.2) [compared with Eq. (2.63)]. The
guiding center drift contributes a kinetic energy
1
2
mv
2
drift
to the energy of the electron, which is small for large
B and smooth V . (More precisely,
1
2
mv
2
drift
h
c
if
[V [ h
c
/l
m
, with l
m
the magnetic length dened as
l
m
(h/eB)
1/2
.) This kinetic energy term has therefore
not been included in Eq. (4.1).
The simplicity of the guiding center drift along equipo-
tentials has been originally used in the percolation
theory
432,433,434
of the QHE, soon after its experimen-
tal discovery.
8
In this theory the existence of edge states
is ignored, so the Hall resistance is not expressed in
terms of equilibrium properties of the 2DEG (in con-
trast to the edge channel formulation that will be dis-
cussed). The physical requirements on the smooth-
ness of the disorder potential have received considerable
attention
435,436
in the context of the percolation theory
and, more recently,
437,438,439
in the context of adiabatic
transport in edge channels. Strong potential variations
should occur on a spatial scale that is large compared
with the magnetic length l
m
(l
m
corresponds to the cy-
clotron radius in the QHE, l
cycl
l
m
(2n 1)
1/2
l
m
if the Landau level index n 1). More rapid potential
uctuations may be present provided their amplitude is
much less than h
c
(the energy separation of Landau
levels).
In Fig. 78 we have illustrated the formation of edge
channels in a smooth potential energy landscape from
FIG. 78 Formation of edge channels in a disordered potential,
from various viewpoints discussed in the text.
various viewpoints. The wave functions of states at the
Fermi level are extended along equipotentials at the guid-
ing center energy (4.1), as shown in Fig. 78a (for Landau
level index n = 1, 2, 3 and a single spin direction). One
can distinguish between extended states near the sample
boundaries and localized states encircling potential max-
ima and minima in the bulk. The extended states at the
Fermi level form the edge channels. The edge channel
with the smallest index n is closest to the sample bound-
ary, because it has the largest E
G
[Eq. (4.1)]. This is
seen more clearly in the cross-sectional plot of V (x, y)
in Fig. 78b (along the line connecting the two arrows in
Fig. 78a). The location of the states at the Fermi level
is indicated by dots and crosses (depending on the direc-
tion of motion). The value of E
G
for each n is indicated
by the dashed line. If the peaks and dips of the poten-
tial in the bulk have amplitudes below h
c
/2, then only
79
FIG. 79 Measurement conguration for the two-terminal re-
sistance R2t, the four-terminal Hall resistance RH, and the
longitudinal resistance RL. The edge channels at the Fermi
level are indicated; arrows point in the direction of motion of
edge channels lled by the source contact at chemical poten-
tial EF + . The current is equipartitioned among the edge
channels at the upper edge, corresponding to the case of local
equilibrium.
states with highest Landau level index can exist in the
bulk at the Fermi level. This is obvious from Fig. 78c,
which shows the total energy of a state E
G
+(n
1
2
) h
c
along the same cross section as Fig. 78b. If one identies
k = xeB/h, this plot can be compared with Fig. 40b of
the dispersion relation E
n
(k) for a disorder-free electron
waveguide in strong magnetic eld.
A description of the QHE based on extended edge
states and localized bulk states, as in Fig. 78, was rst
put forward by Halperin
440
and further developed by sev-
eral authors.
441,442,443,444
In these papers a local equilib-
rium is assumed at each edge. In the presence of a chem-
ical potential dierence between the edges, each edge
channel carries a current (e/h) and thus contributes
e
2
/h to the Hall conductance (cf. the derivation of Lan-
dauers formula in Section III.A.2). In this case of lo-
cal equilibrium the two-terminal resistance R
2t
of the
Hall bar is the same as the four-terminal Hall resistance
R
H
= R
2t
= h/e
2
N (see Fig. 79). The longitudinal re-
sistance vanishes, R
L
= 0. The distinction between a
longitudinal and Hall resistance is topological: A four-
terminal resistance measurement gives R
H
if current and
voltage contacts alternate along the boundary of the con-
ductor, and R
L
if that is not the case. There is no need
to further characterize the contacts in the case of local
equilibrium at the edge.
If the edges are not in local equilibrium, the mea-
sured resistance depends on the properties of the con-
tacts. Consider, for example, a situation in which the
edge channels at the lower edge are in equilibrium at
chemical potential E
F
, while the edge channels at the
upper edge are not in local equilibrium. The current at
the upper edge is then not equipartitioned among the N
modes. Let f
n
be the fraction of the total current I that
is carried by states above E
F
in the nth edge channel at
the upper edge, I
n
= f
n
I. The voltage contact at the
lower edge measures a chemical potential E
F
regardless
of its properties. The voltage contact at the upper edge,
however, will measure a chemical potential that depends
on how it couples to each of the edge channels. The
transmission probability T
n
is the fraction of I
n
that is
transmitted through the voltage probe to a reservoir at
chemical potential E
F
+. The incoming current
I
in
=
N

n=1
T
n
f
n
I, with
N

n=1
f
n
= 1, (4.3)
has to be balanced by an outgoing current
I
out
=
e
h
(N R) =
e
h

n=1
T
n
(4.4)
of equal magnitude, so that the voltage probe draws no
net current. (In Eq. (4.4) we have applied Eq. (3.14) to
identify the total transmission probability N R of out-
going edge channels with the sum of transmission proba-
bilities T
n
of incoming edge channels.) The requirement
I
in
= I
out
determines and hence the Hall resistance
R
H
= /eI:
R
H
=
h
e
2
_
N

n=1
T
n
f
n
__
N

n=1
T
n
_
1
. (4.5)
The Hall resistance has its regular quantized value R
H
=
h/e
2
N only if either f
n
= 1/N or T
n
= 1, for n =
1, 2, . . . , N. The rst case corresponds to local equilib-
rium (the current is equipartitioned among the modes),
the second case to an ideal contact (all edge channels
are fully transmitted). The Landauer-B uttiker formal-
ism discussed in Section III.A.2 forms the basis on which
anomalies in the QHE due to the absence of local equi-
librium in combination with nonideal contacts can be
treated theoretically.
112
A nonequilibrium population of the edge channels is
generally the result of selective backscattering. Because
edge channels at opposite edges of the sample move in op-
posite directions, backscattering requires scattering from
one edge to the other. Selective backscattering of edge
channels with n n
0
is induced by a potential barrier
across the sample,
113,339,340,427
if its height is between
the guiding center energies of edge channel n
0
and n
0
1
(note that the edge channel with a larger index n has a
smaller value of E
G
). The anomalous Shubnikov-De Haas
eect,
428
to be discussed in Section IV.B, has demon-
strated that selective backscattering can also occur natu-
rally in the absence of an imposed potential barrier. The
edge channel with the highest index n = N is selectively
backscattered when the Fermi level approaches the en-
ergy (N
1
2
) h
c
of the Nth bulk Landau level. The
guiding center energy of the Nth edge channel then ap-
proaches zero, and backscattering either by tunneling or
by thermally activated processes becomes eective, but
for that edge channel only, which remains almost com-
pletely decoupled from the other N1 edge channels over
80
distances as large as 250 m (although on that length
scale the edge channels with n N 1 have equilibrated
to a large extent).
429
3. Current distribution
The edge channel theory has been criticized on the
grounds that experiments measure a nonzero current in
the bulk of a Hall bar.
445
In this subsection we want to
point out that a measurement of the current distribution
cannot be used to prove or disprove the edge channel
formulation of the QHE.
The fact that the Hall resistance can be expressed in
terms of the transmission probabilities of edge states at
the Fermi level does not imply that these few states carry
a macroscopic current, nor does it imply that the current
ows at the edges. A determination of the spatial cur-
rent distribution i(r), rather than just the total current I,
requires consideration of all the states below the Fermi
level, which acquire a net drift velocity because of the
Hall eld. As we discussed in Section III.A.2, knowledge
of i(r) is not necessary to know the resistances in the
regime of linear response, because the Einstein relation
allows one to obtain the resistance from the diusion con-
stant. Edge channels tell you where the current ows if
the electrochemical potential dierence is entirely due
to a density dierence, relevant for the diusion problem.
Edge channels have nothing to say about where the cur-
rent ows if is mainly of electrostatic origin, relevant
for the problem of electrical conduction. The ratio /I
is the same for both problems, but i(r) is not.
With this in mind, it remains an interesting problem
to nd out just how the current is distributed in a Hall
bar, or, alternatively, what is the electrostatic potential
prole. This problem has been treated theoretically in
many papers.
446,447,448,449,450,451,452,453,454,455
In the case
of a 3D conductor, a linearly varying potential and uni-
form current density are produced by a surface charge.
As noted by MacDonald et al.,
446
the electrostatics is
qualitatively dierent in the 2D case because an edge
charge (x W/2) produces a potential proportional to
ln [x W/2[, which is weighted toward the edge, and
hence a concentration of current at the edge.
Experiments aimed at measuring the electrostatic po-
tential distribution were originally carried out by at-
taching contacts to the interior of the Hall bar and
measuring the voltage dierences between adjacent
contacts.
456,457,458,459,460
It was learned from these stud-
ies that relatively small inhomogeneities in the density
of the 2DEG have a large eect on these voltage dif-
ferences in the QHE regime. The main diculty in the
interpretation of such experiments is that the voltage dif-
ference measured between two contacts is the dierence
in electrochemical potential, not the line integral of the
electric eld. B uttiker
461
has argued that the voltage
measured at an interior contact can exhibit large vari-
ations for a small increase in magnetic eld without an
FIG. 80 Electrostatic potential VH induced by passing a cur-
rent through a Hall bar. The sample edges are at x = 1 mm.
The data points are from the experiment of Fontein et a1.,
463
at two magnetic eld values on the RH = h/4e
2
quantized Hall
plateau (triangles: B = 5 T; crosses: B = 5.25 T). The solid
curve is calculated from Eq. (4.9), assuming an impurity-free
Hall bar with four lled Landau levels. The theory contains
no adjustable parameters.
appreciable change in the current distribution. Contact-
less measurements of the QHE from the absorption of mi-
crowave radiation
462
are one alternative to interior con-
tacts, which might be used to determine the potential (or
current) distribution.
Fontein et al.
463
have used the birefringence of GaAs
induced by an electric eld to perform a contactless mea-
surement of the electrostatic potential distribution in a
Hall bar. They measure the Hall potential prole V
H
(x)
as a change in the local electrostatic potential if a current
is passed through the Hall bar. The data points shown
in Fig. 80 were taken at 1.5 K for two magnetic eld val-
ues on the plateau of quantized Hall resistance at
1
4
h/e
2
.
The potential varies steeply at the edges (at x = 1 mm
in Fig. 80) and is approximately linear in the bulk. The
spatial resolution of the experiment was 70 m, limited
by the laser beam used to measure the birefringence. The
current distribution is not directly measured, but can be
estimated from the guiding center drift (4.2) (this as-
sumes a slowly varying potential). The nonequilibrium
current density i(x) along the Hall bar is then given by
i(x) =
en
s
B
dV
H
(x)
dx
. (4.6)
Fontein et al. thus estimate that under the conditions of
their experiment two thirds of the total imposed current
I = 5 A ows within 70 m from the edges while the
remainder is uniformly distributed in the bulk.
This experimental data can be modeled
464
by means of
an integral equation derived by MacDonald et al.
446
for
the self-consistent potential prole in an ideal impurity-
free sample with N completely lled (spin-split) Landau
levels. The electron charge density
e
(x) in the 2DEG is
81
given by

e
(x) = en
s
_
1
el
2
m
h
c
V

H
(x)
_
. (4.7)
This equation follows from the Schrodinger equation in
a smoothly varying electrostatic potential, so the factor
between brackets is close to unity. Substitution of the
net charge density en
s
+
e
(x) into the Poisson equation
gives
446
V
H
(x) =
_
+W/2
W/2
dx

ln
_
2
W
[x x

[
_
V

H
(x

). (4.8)
The characteristic length Nl
2
m
/a

is dened in
terms of the magnetic length l
m
and the eective Bohr
radius a

h
2
/me
2
(with the dielectric constant).
The integral equation (4.8) was solved numerically by
MacDonald et al.
446
and analytically by means of the
Wiener-Hopf technique by Thouless.
448
Here we describe
a somewhat simpler approach,
464
which is suciently
accurate for the present purpose. For magnetic eld
strengths in the QHE regime the length is very small.
For example, if N = 4, l
m
= 11.5 nm (for B = 5 T),
a

= 10 nm (for GaAs with = 13

0
and m = 0.067 m
e
),
then = 17 nm. It is therefore meaningful to look for
a solution of Eq. (4.8) in the limit W. The result
is that V
H
(x) = constant ln [(x W/2)/(x + W/2)[
if [x[ W/2 , with a linear extrapolation from
[x[ = W/2 to [x[ = W/2. One may verify that this is
indeed the answer, by substituting the preceding expres-
sion into Eq. (4.8) and performing one partial integration.
The arbitrary constant in the expression for V
H
may be
eliminated in favor of the total current I owing through
the Hall bar, by applying Eq. (4.6) to the case of N lled
spin-split Landau levels. This gives the nal answer
V
H
(x) =
1
2
IR
H
_
1 + ln
W

_
1
ln

x W/2
x +W/2

if [x[
W
2
, (4.9)
with a linear extrapolation of V
H
to
1
2
IR
H
in the in-
terval within from the edge. The Hall resistance is
R
H
= h/Ne
2
. The approximation (4.9) is equivalent for
small to the analytical solution of Thouless, and is close
to the numerical solutions given by MacDonald et al.,
even for a relatively large value /W = 0.1.
In Fig. 80 the result (4.9) has been plotted (solid
curve) for the parameters of the experiment by Fontein
et al. (/W = 0.85 10
5
for N = 4, B = 5 T, and
W = 2 mm). The agreement with experiment is quite
satisfactory in view of the fact that the theory contains no
adjustable parameters. The theoretical prole is steeper
at the edges than in the experiment, which may be due
to the limited experimental resolution of 70 m. The
total voltage drop between the two edges in the calcula-
tion (hI/Ne
2
32 mV for I = 5 A and N = 4) agrees
with the measured Hall voltage of 30 mV, but the op-
tically determined value of 40 mV is somewhat larger for
a reason that we do not understand.
We have discussed this topic of the current distribu-
tion in the QHE in some detail to convince the reader
that the concentration of the potential drop (and hence
of the current) near the edges can be understood from
the electrostatics of edge charges, but cannot be used to
test the validity of a linear response formulation of the
QHE in terms of edge states. Indeed, edge states were
completely neglected in the foregoing theoretical analy-
sis, which nonetheless captures the essential features of
the experiment.
B. Selective population and detection of edge channels
The absence of local equilibrium at the current or volt-
age contacts leads to anomalies in the quantum Hall ef-
fect, unless the contacts are ideal (in the sense that each
edge channel at the Fermi level is transmitted through
the contact with probability 1). Ideal versus disordered
contacts are dealt with in Sections IV.B.1 and IV.B.2.
A quantum point contact can be seen as an extreme ex-
ample of a disordered contact, as discussed in Section
IV.B.3. Anomalies in the Shubnikov-De Haas eect due
to the absence of local equilibrium are the subject of Sec-
tion IV.B.4.
1. Ideal contacts
In a two-terminal measurement of the quantum Hall
eect the contact resistances of the current source and
drain are measured in series with the Hall resistance.
For this reason precision measurements of the QHE are
usually performed in a four-terminal measurement con-
guration, in which the voltage contacts do not carry a
current.
445
Contact resistances then do not play a role,
provided that local equilibrium is established near the
voltage contacts [or, by virtue of the reciprocity relation
(3.16), near the current contacts]. As we mentioned in
Section IV.A, local equilibrium can be grossly violated
in the QHE. Accurate quantization then requires that ei-
ther the current or the voltage contacts are ideal, in the
sense that the edge states at the Fermi level have unit
transmission probability through the contacts. In this
subsection we return to the four-terminal measurements
on a quantum point contact considered in Section III.B.2,
but now in the QHE regime where the earlier assumption
of local equilibrium near the voltage contacts is no longer
applicable in general. We assume strong magnetic elds
so that the four-terminal longitudinal resistance R
L
of
the quantum point contact is determined by the poten-
tial barrier in the constriction (rather than by its width).
Let us apply the Landauer-B uttiker formalism to the
geometry of Fig. 81. As in Section III.B.2, the number
of spin-degenerate edge channels in the wide 2DEG and
82
FIG. 81 Motion along equipotentials in the QHE regime,
in a four-terminal geometry with a saddle-shaped potential
formed by a split gate (shaded). Ideal contacts are assumed.
The thin lines indicate the location of the edge channels at
the Fermi level, with the arrows pointing in the direction of
motion of edge channels that are populated by the contacts
(crossed squares). Taken from H. van Houten et al., in Ref.
9
.
in the constriction are denoted by N
wide
and N
min
, re-
spectively. An ideal contact to the wide 2DEG perfectly
transmits N
wide
channels, whereas the constriction trans-
mits only N
min
channels. The remaining N
wide
N
min
channels are reected back along the opposite 2DEG
boundary (cf. Fig. 81). We denote by
l
and
r
the chem-
ical potentials of adjacent voltage probes to the left and
to the right of the constriction. The current source is
at
s
, and the drain at
d
. Applying Eq. (3.12) to this
case, using I
s
= I
d
I, I
r
= I
l
= 0, one nds for the
magnetic eld direction indicated in Fig. 81,
(h/2e)I = N
wide

s
(N
wide
N
min
)
l
, (4.10a)
0 = N
wide

l
N
wide

s
, (4.10b)
0 = N
wide

r
N
min

l
. (4.10c)
We have used the freedom to choose the zero level of
chemical potential by xing
d
= 0, so we have three in-
dependent (rather than four dependent) equations. The
two-terminal resistance R
2t

s
/eI following from Eq.
(4.10) is
R
2t
=
h
2e
2
1
N
min
, (4.11)
unaected by the presence of the additional voltage
probes in Fig. 81. The four-terminal longitudinal resis-
tance R
L
(
l

r
)/eI is
R
L
=
h
2e
2
_
1
N
min

1
N
wide
_
. (4.12)
In the reversed eld direction the same result is obtained.
Equation (4.12), derived for ideal contacts without as-
suming local equilibrium near the contacts, is identical
to Eq. (3.23), derived for the case of local equilibrium.
In a six-terminal measurement geometry (see Fig. 82),
one can also measure the Hall resistance in the wide re-
gions, which is simply R
H
= R
2t
R
L
or
R
H
=
h
2e
2
1
N
wide
, (4.13)
FIG. 82 Perspective view of a six-terminal Hall bar con-
taining a point contact, showing the various two- and four-
terminal resistances mentioned in the text. Taken from H.
van Houten et al., in Ref.
9
.
which is unaected by the presence of the constriction.
This is a consequence of our assumption of ideal voltage
probes. One can also measure the two four-terminal di-
agonal resistances R
+
D
and R

D
across the constriction in
such a way that the two voltage probes are on opposite
edges of the 2DEG, on either side of the constriction (see
Fig. 82). Additivity of voltages on contacts tells us that
R

D
= R
H
R
L
(for the magnetic eld direction of Fig.
82); thus,
R
+
D
=
h
2e
2
1
N
min
; R

D
=
h
2e
2
_
2
N
wide

1
N
min
_
. (4.14)
On eld reversal, R
+
D
and R

D
are interchanged. Thus, a
four-terminal resistance [R
+
D
in Eq. (4.14)] can in prin-
ciple be equal to the two-terminal resistance [R
2t
in Eq.
(4.11)]. The main dierence between these two quanti-
ties is that an additive contribution of the ohmic con-
tact resistance (and of a part of the diusive background
resistance in weak magnetic elds) is eliminated in the
four-terminal resistance measurement.
The fundamental reason that the assumption of local
equilibrium made in Section III.B.2 (appropriate for weak
magnetic elds) and that of ideal contacts made in this
section (for strong elds) yield identical answers is that
an ideal contact attached to the wide 2DEG regions in-
duces a local equilibrium by equipartitioning the outgo-
ing current among the edge channels. (This is illustrated
in Fig. 81, where the current entering the voltage probe
to the right of the constriction is carried by a singe edge
channel, while the equally large current owing out of
that probe is equipartitioned over the two edge channels
available for transport in the wide region.) In weaker
magnetic elds, when the cyclotron radius exceeds the
width of the narrow 2DEG region connecting the volt-
age probe to the Hall bar, not all edge channels in the
wide 2DEG region are transmitted into the voltage probe
(even if it does not contain a potential barrier). This
83
FIG. 83 Fractional quantization in the integer QHE of the
four-terminal longitudinal conductance R
1
L
of a point con-
tact in a magnetic eld of 1.4 T at T = 0.6 K. The solid
horizontal lines indicate the quantized plateaus predicted by
Eq. (4.12), with N
wide
= 5 and Nmin = 1, 2, 3, 4. The dashed
lines give the location of the spin-split plateaux, which are
not well resolved at this magnetic eld value. Taken from L.
P. Kouwenhoven, Masters thesis, Delft University of Tech-
nology, 1988.
probe is then not eective in equipartitioning the cur-
rent. That is the reason that the weak-eld analysis in
Section III.B.2 required the assumption of a local equi-
librium in the wide 2DEG near the contacts.
We now discuss some experimental results, which con-
rm the behavior predicted by Eq. (4.12) in the QHE
regime, to complement the weak-eld experiments dis-
cussed in Section III.B.2. Measurements on a quantum
point contact by Kouwenhoven et al.
307,465
in Fig. 83
show the quantization of the longitudinal conductance
R
1
L
in fractions of 2e
2
/h (for unresolved spin degener-
acy). The magnetic eld is kept xed at 1.4 T (such that
N
wide
= 5) and the gate voltage is varied (such that N
min
ranges from 1 to 4). Conductance plateaux close to 5/4,
10/3, 15/2, and 20 (2e
2
/h) (solid horizontal lines) are
observed, in accord with Eq. (4.12). Spin-split plateaux
(dashed lines) are barely resolved at this rather low mag-
netic eld. Similar data were reported by Snell et al.
342
Observations of such a fractional quantization due to
the integer QHE were made before on wide Hall bars
with regions of dierent electron density in series,
466,467
but the theoretical explanation
468
given at that time was
less straightforward than Eq. (4.12).
In the high-eld regime the point contact geometry of
Fig. 81 is essentially equivalent to a geometry in which
a potential barrier is present across the entire width of
the Hall bar (created by means of a narrow continu-
ous gate). The latter geometry was studied by Haug
et al.
340
and by Washburn et al.
339
The geometries of
both experiments
339,340
are the same (see Figs. 84 and
85), but the results exhibit some interesting dierences
because of the dierent dimensions of gate and channel.
Hauge et al.
340
used a sample of macroscopic dimensions,
the channel width being 100 m and the gate length 10
and 20 m. Results are shown in Fig. 84. As the gate
voltage is varied, a quantized plateau at h/2e
2
is seen
in the longitudinal resistance at xed magnetic eld, in
agreement with Eq. (4.12) (the plateau occurs for two
spin-split Landau levels in the wide region and one spin-
split level under the gate). A qualitatively dierent as-
pect of the data in Fig. 84, compared with Fig. 83, is
the presence of a resistance minimum. Equation (4.12),
in contrast, predicts that R
L
varies monotonically with
barrier height, and thus with gate voltage. A model for
the eect has been proposed in a dierent paper by Haug
et al.,
341
based on a competition between backscattering
and tunneling through localized states in the barrier re-
gion. They nd that edge states that are totally reected
at a given barrier height may be partially transmitted if
the barrier height is further increased. The importance
of tunneling is consistent with the increase of the am-
plitude of the dip as the gate length is reduced from 20
to 10 m. A related theoretical study was performed by
Zhu et al.
469
Washburn et al.
339
studied the longitudinal resistance
of a barrier dened by a 0.1-m-long gate across a 2-m-
wide channel. The relevant dimensions are thus nearly
two orders of magnitude smaller than in the experiment
of Haug et al. Again, the resistance is studied as a func-
tion of gate voltage at xed magnetic eld. The longi-
tudinal (R
L
R
12,43
) and diagonal (R
+
D
R
13,42
) re-
sistances are shown in Fig. 85, as well as their dier-
ence [which according to Eqs. (4.12) and (4.14) would
equal the Hall resistance R
H
]. In this small sample the
quantized plateaux predicted by Eq. (4.12) are clearly
seen, but the resistance dips of the large sample of Haug
et al. are not. We recall that resistance dips were not
observed in the quantum point contact experiment of
Fig. 83 either. The model of Haug et al.
341
would im-
ply that localized states do not form in barriers of small
area. Washburn et al. nd weak resistance uctuations
in the gate voltage intervals between quantized plateaux.
These uctuations are presumably due to some form of
quantum interference, but have not been further identi-
ed. Related experiments on the quantum Hall eect in
a 2DEG with a potential barrier have been performed by
Hirai et al. and by Komiyama et al.
427,470,471,472
These
studies have focused on the role of nonideal contacts in
the QHE, which is the subject of the next subsection.
2. Disordered contacts
The validity of Eqs. (4.114.14) in the QHE regime
breaks down for nonideal contacts if local equilibrium
near the contacts is not established. The treatment of
Section IV.B.1 for ideal contacts implies that the Hall
voltage over the wide 2DEG regions adjacent to the con-
striction is unaected by the presence of the constric-
tion or potential barrier. Experiments by Komiyama et
84
FIG. 84 (a) Schematic view of a wide Hall bar containing
a potential barrier imposed by a gate electrode of length bg.
(b) Longitudinal resistance as a function of gate voltage in
the QHE regime (two spin-split Landau levels are occupied
in the unperturbed electron gas regions). The plateau shown
is at RL = h/2e
2
, in agreement with Eq. (4.12). Results
for bg = 10 m and 20 m are compared. A pronounced dip
develops in the device with the shortest gate length. Taken
from R. J. Haug et al., Phys. Rev. B 39, 10892 (1989).
al.
427,472
have demonstrated that this is no longer true
if one or more contacts are disordered. The analysis of
their experiments is rather involved,
472
which is why we
do not give a detailed discussion here. Instead we re-
view a dierent experiment,
113
which shows a deviating
Hall resistance in a sample with a constriction and a
singe disordered contact. This experiment can be ana-
lyzed in a relatively simple way,
307
following the work of
B uttiker
112
and Komiyama et al.
427,470,471,472
The sample geometry is that of Fig. 82. In Fig. 86 the
four-terminal longitudinal resistance R
L
and Hall resis-
tance R
H
are shown for both a small voltage (0.3 V)
and a large voltage (2.5 V) on the gate dening the con-
striction. The longitudinal resistance decreases in weak
elds because of reduction of backscattering, as discussed
FIG. 85 (a) Schematic view of a 2-m-wide channel contain-
ing a potential barrier imposed by a 0.1-m-long gate. (b)
Top: diagonal resistance R13,42 R
+
D
and longitudinal resis-
tance R12,43 RL as a function of gate voltage in a strong
magnetic eld (B = 5.2 T), showing a quantized plateau in
agreement with Eqs. (4.14) and (4.12), respectively. For com-
parison also the two zero-eld traces are shown, which are
almost identical. Bottom: Dierence R
+
D
RL = RH at 5.2
T. A normal quantum Hall plateau is found, with oscillatory
structure superimposed in gate voltage regions where R
+
D
and
RL are not quantized. Taken from S. Washburn et al., Phys.
Rev. Lett. 61, 2801 (1988).
in Section III.B.2. At larger elds Shubnikov-De Haas os-
cillations develop. The data for V
g
= 0.3 V exhibit zero
minima in the Shubnikov-De Haas oscillations in R
L
and
the normal quantum Hall resistance R
H
= (h/2e
2
)N
1
wide
,
determined by the number of Landau levels occupied in
the wide regions (N
wide
can be obtained from the quan-
tum Hall eect measured in the absence of the constric-
tion or from the periodicity of the Shubnikov-De Haas
oscillations).
At the higher gate voltage V
g
= 2.5 V, nonvanishing
minima in R
L
are seen in Fig. 86 as a result of the for-
mation of a potential barrier in the constriction. At the
minima, R
L
has the fractional quantization predicted by
Eq. (4.12). For example, the plateau in R
L
around 2.2 T
for V
g
= 2.5 V is observed to be at R
L
= 2.1 k
85
FIG. 86 Nonvanishing Shubnikov-De Haas minima in the
longitudinal resistance RL and anomalous quantum Hall re-
sistance RH, measured in the point contact geometry of Fig.
82 at 50 mK. These experimental results are extensions to
higher elds of the weak-eld traces shown in Fig. 50. The
Hall resistance has been measured across the wide region,
more than 100 m away from the constriction, yet RH is seen
to increase if the gate voltage is raised from 0.3 V to 2.5 V.
The magnitude at B = 2.2 T of the deviation in RH and of the
Shubnikov-De Haas minimum in RL are indicated by arrows,
which both for RH and RL have a length of (h/2e
2
)(
1
2

1
3
),
in agreement with the analysis given in the text. Taken from
H. van Houten et al., in Ref.
9
.
(h/2e
2
) (
1
2

1
3
), in agreement with the fact that the
two-terminal resistance yields N
min
= 2 and the number
of Landau levels in the wide regions N
wide
= 3. In spite of
this agreement, and in apparent conict with Eq. (4.13),
the Hall resistance R
H
has increased over its value for
small gate voltages. Indeed, around 2.2 T a Hall plateau
at R
H
= 6.3 k (h/2e
2
)
1
2
is found for V
g
= 2.5 V,
as if the number of occupied Landau levels was given by
N
min
= 2 rather than by N
wide
= 3. This unexpected
deviation was noted in Ref.
113
, but was not understood
at the time. At higher magnetic elds (not shown in Fig.
86) the N = 1 plateau is reached, and the deviation in
the Hall resistance vanishes.
As pointed out in Ref.
307
, the likely explanation of the
data of Fig. 86 is that one of the ohmic contacts used
to measure the Hall voltage is disordered in the sense of
B uttiker
112
that not all edge channels have unit trans-
mission probability into the voltage probe. The disor-
dered contact can be modeled by a potential barrier in
the lead with a height not below that of the barrier in
the constriction, as illustrated in Fig. 87. A net cur-
rent I ows through the constriction, determined by its
two-terminal resistance according to I = (2e/h)N
min

s
,
with
s
the chemical potential of the source reservoir (the
chemical potential of the drain reservoir
d
is taken as a
zero reference). Equation (3.12) applied to the two oppo-
site Hall probes l
1
and l
2
in Fig. 87 takes the form (using
FIG. 87 Illustration of the ow of edge channels along
equipotentials in a sample with a constriction (dened by the
shaded gates) and a disordered voltage probe (a potential bar-
rier in the probe is indicated by the shaded bar). Taken from
H. van Houten et al., in Ref.
9
.
I
l1
= I
l2
= 0,
s
= (h/2e)I/N
min
, and
d
= 0)
0 = N
wide

l1
T
sl1
h
2e
I
N
min
T
l2l1

l2
,(4.15a)
0 = N
l2

l2
T
sl2
h
2e
I
N
min
T
l1l2

l1
, (4.15b)
where we have assumed that the disordered Hall probe
l
2
transmits only N
l2
< N
wide
edge channels because of
the barrier in the lead. For the eld direction shown in
Fig. 87 one has, under the assumption of no inter-edge-
channel scattering from constriction to probe l
2
, T
sl1
=
N
wide
, T
sl2
= T
l2l1
= 0, and T
l1l2
= max(0, N
l2

N
min
). Equation (4.15) then leads to a Hall resistance
R
H
(
l1

l2
)/eI given by
R
H
=
h
2e
2
1
max(N
l2
, N
min
)
. (4.16)
In the opposite eld direction the normal Hall resistance
R
H
= (h/2e
2
)N
1
wide
is recovered.
The assumption of a single disordered probe, plus ab-
sence of interedge channel scattering from constriction
to probe, thus explains the observation in Fig. 86 of an
anomalously high quantum Hall resistance for large gate
voltages, such that N
min
< N
wide
. Indeed, the exper-
imental Hall resistance for V
g
= 2.5 V has a plateau
around 2.2 T close to the value R
H
= (h/2e
2
)N
1
min
(with
N
min
= 2), in agreement with Eq. (4.16) if N
l2
N
min
at this gate voltage. This observation demonstrates the
absence of interedge channel scattering over 100 m (the
separation of constriction and probe), but only between
the highest-index channel (with index n = N
wide
= 3)
86
and the two lower-index channels. Since the n = 1 and
n = 2 edge channels are either both empty or both lled
(cf. Fig. 87, where these two edge channels lie closest to
the sample boundary), any scattering between n = 1 and
2 would have no measurable eect on the resistances. As
discussed in Section IV.B.3, we know from the work of
Alphenaar et al.
429
that (at least in the present samples)
the edge channels with n N
wide
1 do in fact equili-
brate to a large extent on a length scale of 100 m.
In the absence of a constriction, or at small gate volt-
ages (where the constriction is just dened), one has
N
min
= N
wide
so that the normal Hall eect is observed in
both eld directions. This is the situation realized in the
experimental trace for V
g
= 0.3 V in Fig. 86. In very
strong elds such that N
min
= N
l2
= N
wide
= 1 (still
assuming nonresolved spin splitting), the normal result
R
H
= h/2e
2
would follow even if the contacts contain a
potential barrier, in agreement with the experiment (not
shown in Fig. 86). This is a more general result, which
holds also for a barrier that only partially transmits the
n = 1 edge channel.
112,308,472,473,474,475
A similar analysis as the foregoing predicts that the
longitudinal resistance measured on the edge of the sam-
ple that contains ideal contacts retains its regular value
(4.12). On the opposite sample edge the measurement
would involve the disordered contact, and one nds in-
stead
R
L
=
h
2e
2
_
1
N
min

1
max(N
l2
, N
min
)
_
(4.17)
for the eld direction shown in Fig. 87, while Eq. (4.12)
is recovered for the other eld direction. The observation
in the experiment of Fig. 86 for V
g
= 2.5 V of a regu-
lar longitudinal resistance [in agreement with Eq. (4.12)],
along with an anomalous quantum Hall resistance is thus
consistent with this analysis.
The experiments
426,429
discussed in the following sub-
section are topologically equivalent to the geometry of
Fig. 87, but involve quantum point contacts rather than
ohmic contacts. This gives the possibility of populating
and detecting edge channels selectively, thereby enabling
a study of the eects of a nonequilibrium population of
edge channels in a controlled manner.
3. Quantum point contacts
In Section III.C we have seen how a quantum point
contact can inject a coherent superposition of edge chan-
nels at the 2DEG boundary, in the coherent electron fo-
cusing experiment.
59
In that section we restricted our-
selves to weak magnetic elds. Here we discuss the ex-
periment by van Wees et al.,
426
which shows how in the
QHE regime the point contacts can be operated in a dif-
ferent way as selective injectors (and detectors) of edge
channels. We recall that electron focusing can be mea-
sured as a generalized Hall resistance, in which case the
pronounced peaked structure due to mode interference is
FIG. 88 (a) Schematic potential landscape, showing the
2DEG boundary and the saddleshaped injector and collector
point contacts. In a strong magnetic eld the edge channels
are extended along equipotentials at the guiding center en-
ergy, as indicated here for edge channels with index n = 1, 2
(the arrows point in the direction of motion). In this case a
Hall conductance of (2e
2
/h)N with N = 1 would be measured
by the point contacts, in spite of the presence of two occu-
pied spin-degenerate Landau levels in the bulk 2DEG. Taken
from C. W. J. Beenakker et al., Festkorperprobleme 29, 299
(1989). (b) Three-terminal conductor in the electron focusing
geometry. Taken from H. van Houten et al., Phys. Rev. B 39,
8556 (1989).
superimposed on the weak-eld Hall resistance (cf. Fig.
53). If the weak-eld electron-focusing experiments are
extended to stronger magnetic elds, a transition is ob-
served to the quantum Hall eect, provided the injecting
and detecting point contacts are not too strongy pinched
o.
59
The oscillations characteristic of mode interference
disappear in this eld regime, suggesting that the cou-
pling of the edge channels (which form the propagating
modes from injector to collector) is suppressed, and adi-
abatic transport is realized. It is now no longer sucient
to model the point contacts by a point source-detector
of innitesimal width (as was done in Section III.C),
but a somewhat more detailed description of the electro-
static potential V (x, y) dening the point contacts and
the 2DEG boundary between them is required. Schemat-
87
ically, V (x, y) is represented in Fig. 88a. Fringing elds
from the split gate create a potential barrier in the point
contacts, so V has a saddle form as shown. The heights
of the barriers E
i
, E
c
in the injector and collector are sep-
arately adjustable by means of the voltages on the split
gates and can be determined from the two-terminal con-
ductances of the individual point contacts. The point
contact separation in the experiment of Ref.
426
is small
(1.5 m), so one can assume fully adiabatic transport
from injector to collector in strong magnetic elds. As
discussed in Section IV.A, adiabatic transport is along
equipotentials at the guiding center energy E
G
. Note
that the edge channel with the smallest index n has the
largest guiding center energy [according to Eq. (4.1)]. In
the absence of inter-edge-channel scattering, edge chan-
nels can only be transmitted through a point contact
if E
G
exceeds the potential barrier height (disregarding
tunneling through the barrier). The injector thus injects
N
i
(E
F
E
i
)/h
c
edge channels into the 2DEG, while
the collector is capable of detecting N
c
(E
F
E
c
)/h
c
channels. Along the boundary of the 2DEG, however, a
larger number of N
wide
E
F
/h
c
edge channels, equal to
the number of occupied bulk Landau levels in the 2DEG,
are available for transport at the Fermi level. The selec-
tive population and detection of Landau levels leads to
deviations from the normal Hall resistance.
These considerations can be put on a theoretical basis
by applying the Landauer-B uttiker formalism discussed
in Section III.A to the electron-focusing geometry.
80
We
consider a three-terminal conductor as shown in Fig. 88b,
with point contacts in two of the probes (injector i and
collector c), and a wide ideal drain contact d. The col-
lector acts as a voltage probe, drawing no net current, so
that I
c
= 0 and I
d
= I
i
. The zero of energy is chosen
such that
d
= 0. One then nds from Eq. (3.12) the
two equations
0 = (N
c
R
c
)
c
T
ic

i
, (4.18a)
(h/2e)I
i
= (N
i
R
i
)
i
T
ci

c
, (4.18b)
and obtains for the ratio of collector voltage V
c
=
c
/e
(measured relative to the voltage of the current drain) to
injected current I
i
the result
V
c
I
i
=
2e
2
h
T
ic
G
i
G
c

. (4.19)
Here (2e
2
/h)
2
T
ic
T
ci
, and G
i
(2e
2
/h)(N
i
R
i
),
G
c
(2e
2
/h)(N
c
R
c
) denote the conductances of in-
jector and collector point contact.
For the magnetic eld direction indicated in Fig. 88,
the term in Eq. (4.19) can be neglected since T
ci
0
[the resulting Eq. (3.26) was used in Section III.C].
An additional simplication is possible in the adiabatic
transport regime. We consider the case that the barrier
in one of the two point contacts is suciently higher than
in the other, to ensure that electrons that are transmitted
over the highest barrier will have a negligible probabil-
ity of being reected at the lowest barrier. Then T
ic
is
FIG. 89 Experimental correlation between the conductances
Gi, Gc of injector and collector, and the Hall conductance
GH Ii/Vc, shown to demonstrate the validity of Eq. (4.20)
(T = 1.3 K, point contact separation is 1.5 m). The magnetic
eld was kept xed (top: B = 2.5 T, bottom: B = 3.8 T,
corresponding to a number of occupied bulk Landau levels
N = 3 and 2, respectively). By increasing the gate voltage on
one half of the split-gate dening the injector, Gi was varied
at constant Gc. Taken from B. J. van Wees et al., Phys. Rev.
Lett. 62, 1181 (1989).
dominated by the transmission probability over the high-
est barrier, T
ic
min(N
i
R
i
, N
c
R
c
). Substitution
in Eq. (4.19) gives the remarkable result
426
that the Hall
conductance G
H
I
i
/V
c
measured in the electron focus-
ing geometry can be expressed entirely in terms of the
contact conductances G
i
and G
c
:
G
H
max(G
i
, G
c
). (4.20)
Equation (4.20) tells us that quantized values of G
H
occur
not at (2e
2
/h)N
wide
, as one would expect from the N
wide
populated Landau levels in the 2DEG but at the smaller
value of (2e
2
/h) max(N
i
, N
c
). As shown in Fig. 89 this is
indeed observed experimentally.
426
Notice in particular
how any deviation from quantization in max(G
i
, G
c
) is
faithfully reproduced in G
H
, in complete agreement with
Eq. (4.20).
The experiment of Ref.
426
was repeated by Alphenaar
et al.
429
for much larger point contact separations (
100 m), allowing a study of the length scale for equili-
bration of edge channels at the 2DEG boundary. Even
after such a long distance, no complete equilibration of
the edge channels was found, as manifested by a depen-
dence of the Hall resistance on the gate voltage used to
vary the number of edge channels transmitted through
the point contact voltage probe (see Fig. 90). As dis-
cussed in Section IV.A.2, a dependence of the resistance
88
FIG. 90 Results of an experiment similar to that of Fig. 89,
but with a much larger separation of 80 m between injector
and collector. Shown are Ri = G
1
i
, Rc = G
1
c
, and RH =
G
1
H
, as a function of the gate voltage on the collector. (T =
0.45 K, B = 2.8 T; the normal quantized Hall resistance is
1
3
(h/2e
2
).) Regimes I, II, and III are discussed in the text.
Taken from B. W. Alphenaar et al., Phys. Rev. Lett. 64, 677
(1990).
on the properties of the contacts is only possible in the
absence of local equilibrium. In contrast to the exper-
iment by van Wees et al.,
426
and in disagreement with
Eq. (4.20), the Hall resistance in Fig. 90 does not simply
follow the smallest of the contact resistances of current
and voltage probe. This implies that the assumption of
fully adiabatic transport has broken down on a length
scale of 100 m.
In the experiment a magnetic eld was applied such
that three edge channels were available at the Fermi level.
The contact resistance of the injector was adjusted to
R
i
= h/2e
2
, so current was injected in a single edge chan-
nel (n = 1) only. The gate voltage dening the collector
point contact was varied. In Fig. 90 the contact resis-
tances of injector (R
i
) and collector (R
c
) are plotted as
a function of this gate voltage, together with the Hall
resistance R
H
. At zero gate voltage the Hall resistance
takes on its normal quantized value [R
H
=
1
3
(h/2e
2
)].
On increasing the negative gate voltage three regions of
interest are traversed (labeled III to I in Fig. 90). In
region III edge channels 1 and 2 are completely trans-
mitted through the collector, but the n = 3 channel is
partially reected. In agreement with Eq. (4.20), R
H
in-
creases following R
c
. As region II is entered, R
H
levels
o while R
c
continues to increase up to the
1
2
(h/2e
2
)
quantized value. The fact that R
H
stops slightly short of
this value proves that some scattering between the n = 3
and n = 1, 2 channels has occurred. On increasing the
gate voltage further, R
c
rises to h/2e
2
in region I. How-
ever, R
H
shows hardly any increase with respect to its
FIG. 91 Illustration of the spatial extension (shaded ellip-
soids) of edge channels for four dierent values of the Fermi
energy. The n = 3 edge channel can penetrate into the bulk
by hybridizing with the n = 3 bulk Landau level, coexisting
at the Fermi level. This would explain the absence of equili-
bration between the n = 3 and n = 1, 2 edge channels. The
penetration depth l
loc
and the magnetic length are indicated.
Taken from B. W. Alphenaar et al., Phys. Rev. Lett. 64, 677
(1990).
value in region II. This demonstrates that the n = 2 and
n = 1 edge channels have almost fully equilibrated. A
quantitative analysis
429
shows that, in fact, 92% of the
current originally injected into the n = 1 edge channel is
redistributed equally over the n = 1 and n = 2 channels,
whereas only 8% is transferred to the n = 3 edge channel.
The suppression of scattering between the highest-index
n = N edge channel and the group of edge channels
with n N 1 was found to exist only if the Fermi
level lies in (or near) the Nth bulk Landau level. As a
qualitative explanation it was suggested
429,476
that the
Nth edge channel hybridizes with the Nth bulk Landau
level when both types of states coexist at the Fermi level.
Such a coexistence does not occur for n N 1 if the
potential uctuations are small compared with h
c
(cf.
Fig. 78). The spatial extension of the wave functions of
the edge channels is illustrated in Fig. 91 (shaded ellip-
soids) for various values of the Fermi level between the
n = 3 and n = 4 bulk Landau levels. As the Fermi level
approaches the n = 3 bulk Landau level, the correspond-
ing edge channel penetrates into the bulk, so the overlap
with the wave functions of lower-index edge channels de-
creases. This would explain the decoupling of the n = 3
and n = 1, 2 edge channels.
These experiments thus point the way in which the
transition from microscopic to macroscopic behavior
takes place in the QHE, while they also demonstrate that
quite large samples will be required before truly macro-
scopic behavior sets in.
89
FIG. 92 Illustration of the mechanism for the suppression
of Shubnikov-De Haas oscillations due to selective detection
of edge channels. The black area denotes the split-gate point
contact in the voltage probe, which is at a distance of 250 m
from the drain reservoir. Dashed arrows indicate symbolically
the selective back scattering in the highest-index edge chan-
nel, via states in the highest bulk Landau level that coexist
at the Fermi level. Taken from H. van Houten et al., in Ref.
9
.
4. Suppression of the Shubnikov-De Haas oscillations
Shubnikov-De Haas magnetoresistance oscillations
were discussed in Sections I.D.3 and II.F. In weak mag-
netic elds, where a theoretical description in terms of a
local resistivity tensor applies, a satisfactory agreement
between theory and experiment is obtained.
20
As we now
know, in strong magnetic elds the concept of a local re-
sistivity tensor may break down entirely because of the
absence of local equilibrium. A theory of the Shubnikov-
De Haas eect then has to take into account explicitly
the properties of the contacts used for the measurement.
The resulting anomalies are considered in this subsection.
Van Wees et al.
428
found that the amplitude of the
high-eld Shubnikov-De Haas oscillations was suppressed
if a quantum point contact was used as a voltage probe.
To discuss this anomalous Shubnikov-De Haas eect, we
consider the three-terminal geometry of Fig. 92, where
a single voltage contact is present on the boundary be-
tween source and drain contacts. (An alternative two-
terminal measurement conguration is also possible; see
Ref.
428
.) The voltage probe p is formed by a quantum
point contact, while source s and drain d are normal
ohmic contacts. (Note that two special contacts were re-
quired for the anomalous quantum Hall eect of Section
IV.B.3.) One straightforwardly nds from Eq. (3.12) that
the three-terminal resistance R
3t
(
p

d
)/eI mea-
sured between point contact probe and drain is given by
R
3t
=
h
2e
2
T
sp
(N
s
R
s
)(N
p
R
p
) T
ps
T
sp
. (4.21)
This three-terminal resistance corresponds to a general-
FIG. 93 Measurement of the anomalous Shubnikov-De Haas
oscillations in the geometry of Fig. 92. The plotted longitu-
dinal resistance is the voltage drop between contacts p and d
divided by the current from s to d. At high magnetic elds the
oscillations are increasingly suppressed as the point contact
in the voltage probe is pinched o by increasing the nega-
tive gate voltage. The number of occupied spin-split Landau
levels in the bulk is indicated at several of the Shubnikov-De
Haas maxima. Taken from B. J. van Wees et al., Phys. Rev.
B 39, 8066 (1989).
ized longitudinal resistance if the magnetic eld has the
direction of Fig. 92. In the absence of backscattering
in the 2DEG, one has T
sp
= 0, so R
3t
vanishes, as it
should for a longitudinal resistance in a strong magnetic
eld.
Shubnikov-De Haas oscillations in the longitudinal re-
sistance arise when backscattering leads to T
sp
,= 0.
The resistance reaches a maximum when the Fermi level
lies in a bulk Landau level, corresponding to a maximum
probability for backscattering (which requires scattering
from one edge to the other across the bulk of the sample,
as indicated by the dashed lines in Fig. 92). From the pre-
ceding discussion of the anomalous quantum Hall eect,
we know that the point contact voltage probe in a high
magnetic eld functions as a selective detector of edge
channels with index n less than some value determined by
the barrier height in the point contact. If backscattering
itself occurs selectively for the channel with the highest
index n = N, and if the edge channels with n N1 do
not scatter to that edge channel, then a suppression of the
90
Shubnikov-De Haas oscillations is to be expected when
R
3t
is measured with a point contact containing a su-
ciently high potential barrier. This was indeed observed
experimentally,
428
as shown in Fig. 93. The Shubnikov-
De Haas maximum at 5.2 T, for example, is found to
disappear at gate voltages such that the point contact
conductance is equal to, or smaller than 2e
2
/h, which
means that the point contact only transmits two spin-
split edge channels. The number of occupied spin-split
Landau levels in the bulk at this magnetic eld value is 3.
This experiment thus demonstrates that the Shubnikov-
De Haas oscillations result from the highest-index edge
channel only, presumably because that edge channel can
penetrate into the bulk via states in the bulk Landau
level with the same index that coexist at the Fermi level
(cf. Section IV.B.3). Moreover, it is found that this edge
channel does not scatter to the lower-index edge chan-
nels over the distance of 250 m from probe p to drain
d, consistent with the experiment of Alphenaar et al.
429
In Section IV.B.1 we discussed how an ideal contact
at the 2DEG boundary induces a local equilibrium by
equipartitioning the outgoing current equally among the
edge channels. The anomalous Shubnikov-De Haas ef-
fect provides a direct way to study this contact-induced
equilibration by means of a second point contact between
the point contact voltage probe p and the current drain
d in Fig. 92. This experiment was also carried out by
van Wees et al., as described in Ref.
308
. Once again, use
was made of the double-split-gate point contact device
(Fig. 5b), in this case with a 1.5-m separation between
point contact p and the second point contact. It is found
that the Shubnikov-De Haas oscillations in R
3t
are sup-
pressed only if the second point contact has a conduc-
tance of (2e
2
/h)(N
wide
1) or smaller. At larger conduc-
tances the oscillations in R
3t
return, because this point
contact can now couple to the highest-index edge chan-
nel and distribute the backscattered electrons over the
lower-index edge channels. The point contact positioned
between contacts p and d thus functions as a controllable
edge channel mixer.
The conclusions of the previous paragraph have inter-
esting implications for the Shubnikov-De Haas oscilla-
tions in the strong-eld regime even if measured with
contacts that do not selectively detect certain edge chan-
nels only.
307
Consider again the geometry of Fig. 92, in
the low-gate voltage limit where the point contact volt-
age probe transmits all edge channels with unit proba-
bility. (This is the case of an ideal contact; cf. Sec-
tion IV.A.2.) To simplify expression (4.21) for the three-
terminal longitudinal resistance R
3t
, we use the fact that
the transmission and reection probabilities T
sp
, R
s
,
and R
p
refer to the highest-index edge channel only
(with index n = N), under the assumptions of selec-
tive backscattering and absence of scattering to lower-
index edge channels discussed earlier. As a consequence,
T
sp
, R
s
, and R
p
are each at most equal to 1; thus, up
to corrections smaller by a factor N
1
, we may put these
terms equal to zero in the denominator on the right-hand
side of Eq. (4.21). In the numerator, the transmission
probability T
sp
may be replaced by the backscattering
probability t
bs
1, which is the probability that the
highest-index edge channel injected by the source con-
tact reaches the point contact probe following scattering
across the wide 2DEG (dashed lines in Fig. 92). With
these simplications Eq. (4.21) takes the form (assuming
spin degeneracy)
R
3t
=
h
2e
2
t
bs
N
2
(1 + order N
1
). (4.22)
Only if t
bs
1 may the backscattering probability be
expected to scale linearly with the separation of the two
contacts p and d (between which the voltage drop is mea-
sured). If t
bs
is not small, then the upper limit t
bs
<1 leads
to the prediction of a maximum possible amplitude
307
R
max
=
h
2e
2
1
N
2
(1 + order N
1
) (4.23)
of the Shubnikov-De Haas resistance oscillations in a
given large magnetic eld, independently of the length
of the segment over which the voltage drop is measured,
provided equilibration does not occur on this segment.
Equilibration might result, for example, from the pres-
ence of additional contacts between the voltage probes,
as discussed before. One easily veries that the high-eld
Shubnikov-De Haas oscillations in Fig. 93 at V
g
= 0.6 V
(when the point contact is just dened, so that the po-
tential barrier is small) lie well below the upper limit
(4.23). For example, the peak around 2 T corresponds to
the case of four occupied spin-degenerate Landau levels,
so the theoretical upper limit is (h/2e
2
)
1
16
800 ,
well above the observed peak value of about 350 . The
prediction of a maximum longitudinal resistance implies
that the linear scaling of the amplitude of the Shubnikov-
De Haas oscillations with the distance between voltage
probes found in the weak-eld regime, and expected on
the basis of a description in terms of a local resistivity
tensor,
20
breaks down in strong magnetic elds. Anoma-
lous scaling of the Shubnikov-De Haas eect has been
observed experimentally
457,460,466
and has recently also
been interpreted
430
in terms of a nonequilibrium between
the edge channels. A quantitative experimental and the-
oretical investigation of these issues has now been carried
out by McEuen et al.
477
Selective backscattering and the absence of local equi-
librium have consequences as well for the two-terminal
resistance in strong magnetic elds.
307
In weak elds
one usually observes in two-terminal measurements a su-
perposition of the Shubnikov-De Haas longitudinal re-
sistance oscillations and the quantized Hall resistance.
This superposition shows up as a characteristic over-
shoot of the two-terminal resistance as a function of
the magnetic eld as it increases from one quantized
Hall plateau to the next (the plateaux coincide with
minima of the Shubnikov-De Haas oscillations). In the
strong-eld regime (in the absence of equilibration be-
tween source and drain contacts), no such superposi-
91
tion is to be expected. Instead, the two-terminal resis-
tance would increase monotonically from (h/2e
2
)N
1
to
(h/2e
2
)(N 1)
1
as the transmission probability from
source to drain decreases from N to N 1. We are not
aware of an experimental test of this prediction.
The foregoing analysis assumes that the length L of
the conductor is much greater than its width W, so edge
channels are the only states at the Fermi level that extend
from source to drain. If L W, additional extended
states may appear in the bulk of the 2DEG, whenever the
Fermi level lies in a bulk Landau level. An experiment
by Fang et al. in this short-channel regime, to which our
analysis does not apply, is discussed by B uttiker.
386
C. Fractional quantum Hall eect
Microscopically, quantization of the Hall conductance
G
H
in fractional multiples of e
2
/h is entirely dierent
from quantization in integer multiples. While the inte-
ger quantum Hall eect
8
can be explained satisfactorily
in terms of the states of noninteracting electrons in a
magnetic eld (see Section IV.A), the fractional quan-
tum Hall eect
478
exists only because of electron-electron
interactions.
479
Phenomenologically, however, the two ef-
fects are quite similar. Several experiments on edge chan-
nel transport in the integer QHE,
339,340,426
reviewed in
Section IV.B have been repeated
480,481
for the fractional
QHE with a similar outcome. The interpretation of Sec-
tion IV.B in terms of selective population and detec-
tion of edge channels cannot be applied in that form to
the fractional QHE. Edge channels in the integer QHE
are dened in one-to-one correspondence to bulk Landau
levels (Section IV.A.2). The fractional QHE requires a
generalization of the concept of edge channels that al-
lows for independent current channels within the same
Landau level. Two recent papers have addressed this
problem
482,483
and have obtained dierent answers. The
present status of theory and experiment on transport in
fractional edge channels is reviewed in Section IV.C.2,
preceded by a brief introduction to the fractional QHE.
1. Introduction
Excellent high-level introductions to the fractional
QHE in an unbounded 2DEG can be found in Refs.
97
and
484
. The following is an oversimplication of Laugh-
lins theory
479
of the eect and is only intended to intro-
duce the reader to some of the concepts that play a role
in edge channel transport in the fractional QHE.
It is instructive to rst consider the motion of two in-
teracting electrons in a strong magnetic eld.
485
The dy-
namics of the relative coordinate r decouples from that
of the center of mass. Semiclassically, r moves along
equipotentials of the Coulomb potential e
2
/r (this is the
guiding center drift discussed in Section IV.A.2). The rel-
ative coordinate thus executes a circular motion around
the origin, corresponding to the two electrons orbiting
around their center of mass. The phase shift acquired on
one complete revolution,
=
e
h
_
dl A =
e
h
Br
2
, (4.24)
should be an integer multiple of 2 so that
r = l
m
_
2q, q = 1, 2, . . . . (4.25)
The interparticle separation in units of the magnetic
length l
m
( h/eB)
1/2
is quantized. In the eld regime
where the fractional QHE is observed, only one spin-split
Landau level is occupied in general. If the electrons have
the same spin, the wave function should change sign when
two coordinates are interchanged. In the case considered
here of two electrons, an interchange of the coordinates
is equivalent to r r. A change of sign is then ob-
tained if the phase shift for one half revolution is an odd
multiple of (i.e., for an odd multiple of 2). The
Pauli principle thus restricts the integer q in Eq. (4.25)
to odd values.
The interparticle separation of a system of more than
two electrons is not quantized. Still, one might surmise
that the energy at densities n
s
1/ r
2
corresponding to
an average separation r in accord with Eq. (4.25) would
be particularly low. This occurs when the Landau level
lling factor hn
s
/eB equals 1/q. Theoretical
work by Laughlin, Haldane, and Halperin
479,486,487
shows
that the energy density u() of a uniform 2DEG in a
strong magnetic eld has downward cusps at these values
of as well as at other fractions, given generally by
= p/q, (4.26)
with p and q mutually prime integers and q odd. The
cusp in u at integer is a consequence solely of Landau
level quantization, according to
du/dn
s
= (Int[] +
1
2
) h
c
. (4.27)
Because of the cusp in u, the chemical potential du/dn
s
has a discontinuity = h
c
at integer . At these
values of the lling factor an innitesimal increase in
electron density costs a nite amount of energy, so the
electron gas can be said to be incompressible. The cusp
in u at fractional exists because of the Coulomb in-
teraction. The discontinuity is now approximately
e
2
/l
m

B, which at a typical eld of 6 T in

GaAs is 10 meV, of the same magnitude as the Landau
level separation h
c
B.
The incompressibility of the 2DEG at = p/q implies
that a nonzero minimal energy is required to add charge
to the system. An important consequence of Laughlins
theory is that charge can be added only in the form of
quasiparticle excitations of fractional charge e

= e/q.
The discontinuity in the chemical potential equals the
energy that it costs to create p pairs of oppositely charged
quasiparticles (widely separated from each other), =
p 2 with the quasiparticle creation energy.
92
The fractional QHE in a disordered macroscopic sam-
ple occurs because the quasiparticles are localized by po-
tential uctuations in the bulk of the 2DEG. A variation
of the lling factor = p/q + in an interval around
the fractional value changes the density of localized quasi-
particles without changing the Hall conductance, which
retains the value G
H
= (p/q)e
2
/h. The precision of the
QHE has been explained by Laughlin
488
in terms of the
quantization of the quasiparticle charge e

, which is ar-
gued to imply quantization of G
H
at integer multiples of
ee

/h.
2. Fractional edge channels
In a small sample the fractional QHE can occur in the
absence of disorder and can show deviations from precise
quantization. Moreover, in special geometries
481
G
H
can
take on quantized values that are not simply related to e

.
These observations cannot be easily understood within
the conventional description of the fractional QHE, as
outlined in the previous subsection. An approach along
the lines of the edge channel formulation of the integer
QHE (Sections IV.A and IV.B) seems more promising. In
Ref.
482
the concept of an edge channel was generalized to
the fractional QHE, and a generalized Landauer formula
relating the conductance to the transmission probabili-
ties of the edge channels was derived. We review this
theory and the application to experiments. A dierent
edge channel theory by MacDonald
483
is discussed to-
ward the end of this subsection.
The edge channels for the conductance in the linear
transport regime are dened in terms of properties of the
equilibrium state of the system. If the electrostatic po-
tential energy V (x, y) varies slowly in the 2DEG, then the
equilibrium density distribution n(x, y) follows by requir-
ing that the local electrochemical potential V (r) +du/dn
has the same value at each point r in the 2DEG. Here
du/dn is the chemical potential of the uniform 2DEG
with density n(r). As discussed in Section IV.C.1, the in-
ternal energy density u(n) of a uniform interacting 2DEG
in a strong magnetic eld has downward cusps at densi-
ties n =
p
Be/h corresponding to certain fractional ll-
ing factors
p
. As a result, the chemical potential du/dn
has a discontinuity (an energy gap) at =
p
, with
du
+
p
/dn and du

p
/dn the two limiting values as
p
.
As noted by Halperin,
489
when V lies in the energy
gap the lling factor is pinned at the value
p
. The equi-
librium electron density is thus given by
489
n =
_

p
Be/h, if du

p
/dn < V < du
+
p
/dn,
du/dn +V (r) = , otherwise.
(4.28)
Note that V (r) itself depends on n(r) and thus has to
be determined self-consistently from Eq. (4.28), taking
the electrostatic screening in the 2DEG into account.
We do not need to solve explicitly for n(r), but we can
identify the edge channels from the following general
FIG. 94 Schematic drawing of the variation in lling factor
, electrostatic potential V , and chemical potential du/dn, at
a smooth boundary in a 2DEG. The dashed line in the bottom
panel denotes the constant electrochemical potential = V +
du/dn. The dotted intervals indicate a discontinuity (energy
gap) in du/dn and correspond in the top panel to regions of
constant fractional lling factor p that spatially separate the
edge channels. The width of the edge channel regions shrinks
to zero in the integer QHE, since the compressibility of
these regions is innitely large in that case. Taken from C.
W. J. Beenakker, Phys. Rev. Lett. 64, 216 (1990).
considerations.
482
At the edge of the 2DEG, the electron density de-
creases from its bulk value to zero. Eq. (4.28) implies
that this decrease is stepwise, as illustrated in Fig. 94.
The requirement on the smoothness of V for the appear-
ance of a well-dened region at the edge in which is
pinned at the fractional value
p
is that the change in
V within the magnetic length l
m
is small compared with
the energy gap du
+
p
/dndu

p
/dn. This ensures that the
width of this region is large compared with l
m
, which
is a necessary (and presumably sucient) condition for
the formation of the incompressible state. Depending on
the smoothness of V , one thus obtains a series of steps
at =
p
(p = 1, 2, . . . , P) as one moves from the edge
toward the bulk. The series terminates in the lling fac-
tor
P
=
bulk
of the bulk, assuming that in the bulk
the chemical potential V lies in an energy gap. The
regions of constant at the edge form bands extending
along the wire. These incompressible bands [in which the
compressibility (n
2
d
2
u/dn
2
)
1
= 0] alternate with
bands in which V does not lie in an energy gap. The
latter compressible bands (in which > 0) may be iden-
tied as the edge channels of the transport problem, as
will be discussed later. To resolve a misunderstanding,
490
we note that the particular potential and density prole
93
FIG. 95 Schematic drawing of the incompressible bands
(hatched) of fractional lling factor p, alternating with the
edge channels (arrows indicate the direction of electron mo-
tion in each channel). (a) A uniform conductor. (b) A con-
ductor containing a barrier of reduced lling factor. Taken
from C. W. J. Beenakker, Phys. Rev. Lett. 64, 216 (1990).
illustrated in Fig. 94 (in which the edge channels have a
nonzero width) assumes that the compressibility of the
edge channels is not innitely large, but the subsequent
analysis is independent of this assumption (requiring only
that the edge channels are anked by bands of zero com-
pressibility). Indeed, the analysis is applicable also to the
integer QHE, where the edge channels have an innitely
large compressibility and hence an innitesimally small
width (limited only by the magnetic length).
The conductance is calculated by bringing one end of
the conductor in contact with a reservoir at a slightly
higher electrochemical potential + without changing
V (as in the derivation of the usual Landauer formula;
cf. Section III.A.2). The resulting change n in electron
density is
n =
_
n

_
V
=
_
n
V
_

, (4.29)
where denotes a functional derivative. In the second
equality in Eq. (4.29), we used the fact that n is a func-
tional of V , by virtue of Eq. (4.28). In a strong mag-
netic eld, this excess density moves along equipotentials
with the guiding-center-drift velocity E/B (E V/er
being the electric eld). The component v
drift
of the drift
velocity in the y-direction (along the conductor) is
v
drift
= y
_
E
B
B
2
_
=
1
eB
V
x
. (4.30)
The current density j = env
drift
becomes simply
j =
e
h

v
x
. (4.31)
It follows from Eq. (4.31) that the incompressible bands
of constant =
p
do not contribute to j. The reser-
voir injects the current into the compressible bands at
one edge of the conductor only (for which the sign of
/x is such that j moves away from the reservoir).
The edge channel with index p = 1, 2, . . . , P is dened as
that compressible band that is anked by incompressible
bands at lling factors
p
and
p1
. The outermost band
from the center of the conductor, which is the p = 1 edge
channel, is included by dening formally
0
0. The
arrangement of alternating edge channels and compress-
ible bands is illustrated in Fig. 95a. Note that dierent
edges may have a dierent series of edge channels at the
same magnetic eld value, depending on the smoothness
of the potential V at the edge (which, as discussed be-
fore, determines the incompressible bands that exist at
the edge). This is in contrast to the situation in the
integer QHE, where a one-to-one correspondence exists
between edge channels and bulk Landau levels (Section
IV.A.2). In the fractional QHE an innite hierarchy of
energy gaps exists, in principle, corresponding to an in-
nite number of possible edge channels, of which only a
small number (corresponding to the largest energy gaps)
will be realized in practice.
The current I
p
= (e/h)(
p

p1
) injected into
edge channel p by the reservoir follows directly from Eq.
(4.31) on integration over x. The total current I through
the wire is I =

P
p=1
I
p
T
p
, if a fraction T
p
of the in-
jected current I
p
is transmitted to the reservoir at the
other end of the wire (the remainder returning via the
opposite edge). For the conductance G eI/, one
thus obtains the generalized Landuer formula for a two-
terminal conductor,
482
G =
e
2
h
P

p=1
T
p

p
, (4.32)
which diers from the usual Landauer formula by the
presence of the fractional weight factors
p

p

p1
.
In the integer QHE,
p
= 1 for all p so that the usual
Landauer formula with unit weight factor is recovered.
A multiterminal generalization of Eq. (4.32) for a
two-terminal conductor is easily constructed, following
B uttiker
5
(cf. Section III.A.2):
I

=
e
h

e
h

, (4.33a)
T

=
P

p=1
T
p,

p
. (4.33b)
Here I

is the current in lead connected to a reservoir at

electrochemical potential

and fractional lling factor

. Equation (4.33b) denes the transmission probabil-

ity T

from reservoir to reservoir (or the reection

probability for = ) in terms of a sum over the gen-
eralized edge channels in lead . The contribution from
each edge channel p = 1, 2, . . . , P

contains the weight

factor
p

p

p1
and the fraction T
p,
of the cur-
rent injected by reservoir into the pth edge channel of
lead that reaches reservoir . Apart from the fractional
94
weight factors, the structure of Eq. (4.33) is the same as
that of the usual B uttiker formula (3.12).
Applying the generalized Landauer formula (4.32) to
the ideal conductor in Fig. 95a, where T
p
= 1 for all p,
one nds the quantized two-terminal conductance
G =
e
2
h
P

p=1

p
=
e
2
h

P
. (4.34)
The four-terminal Hall conductance G
H
has the same
value, because each edge is in local equilibrium. In the
presence of disorder this edge channel formulation of the
fractional QHE is generalized in an analogous way as
in the integer QHE by including localized states in the
bulk. In a smoothly varying disorder potential, these lo-
calized states take the form of circulating edge channels,
as in Figs. 78 and 79. In this way the lling factor of
the bulk can locally deviate from
P
without a change
in the Hall conductance, leading to the formation of a
plateau in the magnetic eld dependence of G
H
. In a
narrow channel, localized states are not required for a
nite plateau width because the edge channels make it
possible for the chemical potential to lie in an energy gap
for a nite-magnetic-eld interval. The Hall conductance
then remains quantized at
P
(e
2
/h) as long as V in
the bulk lies between du
+
P
/dn and du

P
/dn.
We now turn to a discussion of experiments on the
fractional QHE in semiconductor nanostructures. Timp
et al.
491
have measured the fractionally quantized four-
terminal Hall conductance G
H
in a narrowcross geometry
(dened by two sets of split gates). The channel width
W 90 nm is greater than, but comparable to, the corre-
lation length l
m
of the incompressible state in this experi-
ment (l
m
9 nm at B = 8 T), so one may expect the frac-
tional QHE to be modied by the lateral connement.
492
Timp et al. nd, in addition to quantized plateaux near
1
3
,
2
5
, and
2
3
e
2
/h, a plateau-like feature around
1
2
e
2
/h.
This even-denominator fraction is not observed as a Hall
plateau in a bulk 2DEG.
493
The plateaux in G
H
correlate
with dips in a four-terminal longitudinal resistance (the
bend resistance dened in Section III.E).
Consider now a conductor containing a potential bar-
rier. The potential barrier corresponds to a region of
reduced lling factor
Pmin

min
separating two regions
of lling factor
Pm

max
. The arrangement of edge
channels and incompressible bands is illustrated in Fig.
95b. We assume that the potential barrier is suciently
smooth that scattering between the edge channels at op-
posite edges can be neglected. All transmission probabil-
ities are then either 0 or 1: T
p
= 1 for 1 p P
min
, and
T
p
= 0 for P
min
< p P
max
. Equation (4.32) then tells
us that the two-terminal conductance is
G = (e
2
/h)
min
. (4.35)
In Fig. 96 we show experimental data by Kouwenhoven
et al.
481
of the fractionally quantized two-terminal con-
ductance of a constriction containing a potential barrier.
FIG. 96 Two-terminal conductance of a constriction con-
taining a potential barrier, as a function of the voltage on the
split gate dening the constriction, at a xed magnetic eld
of 7 T. The conductance is quantized according to Eq. (4.35).
Taken from L. P. Kouwenhoven et al., unpublished.
The constriction (or point contact) is dened by a split
gate on top of a GaAs-AlGaAs heterostructure. The con-
ductance in Fig. 96 is shown for a xed magnetic eld of
7 T as a function of the gate voltage. Increasing the neg-
ative gate voltage increases the barrier height, thereby
reducing G below the Hall conductance corresponding to

max
= 1 in the wide 2DEG. The curve in Fig. 96 shows
plateaux corresponding to
min
= 1,
2
3
, and
1
3
in Eq.
(4.35). The
2
3
plateau is not exactly quantized, but is
too low by a few percent. The constriction width on this
plateau is estimated
481
at 500 nm, which is a factor of 50
larger than the magnetic length at B = 7 T. It would
seem that scattering between fractional edge channels at
opposite edges (necessary to reduce the conductance be-
low its quantized value) can only occur via states in the
bulk for this large ratio of W/l
m
.
A four-terminal measurement of the fractional QHE in
a conductor containing a potential barrier can be ana-
lyzed by means of Eq. (4.33), analogously to the case of
the integer QHE discussed in Section IV.B. The four-
terminal longitudinal resistance R
L
(in the geometry of
Fig. 82) is given by the analog of Eq. (4.12),
R
L
=
h
e
2
_
1

min

max
_
, (4.36)
provided that either the edge channels transmitted across
the barrier have equilibrated with the extra edge chan-
nels available outside the barrier region or the voltage
contacts are ideal; that is, they have unit transmission
probability for all fractional edge channels. Similarly,
the four-terminal diagonal resistances R

D
dened in Fig.
95
FIG. 97 Four-terminal resistances of a 2DEG channel con-
taining a potential barrier, as a function of the gate voltage
(B = 0.114 T, T = 70 mK). The current ows from con-
tact 1 to contact 5 (see inset), the resistance curves are la-
beled by the contacts i and j between which the voltage is
measured. (The curves for i, j = 2, 4 and 8, 6 are identical.)
The magnetic eld points outward. This measurement cor-
responds to the case max = 1 and min =
b
varying from
1 at Vg 10 mV to 2/3 at Vg 90 mV (arrow). The
resistances RL R2,4 = R8,6 and R
+
D
R2,6 are quantized
according to Eqs. (4.36) and (4.37), respectively. The resis-
tances R3,7 and R2,8 are the Hall resistances in the gated
and ungated regions, respectively. From Eq. (4.33) one can
also derive that R8,7 = R3,4 = RL and R2,3 = R7,6 = 0 on
the quantized plateaux, as observed experimentally. Taken
from A. M. Chang and J. E. Cunningham, Surf. Sci. 229, 216
(1990).
82 are given by [cf. Eq. (4.14)]
R
+
D
=
h
e
2
1
v
min
; R

D
=
h
e
2
_
2
v
max

1
v
min
_
. (4.37)
Chang and Cunningham
480
have measured R
L
and R
D
in
the fractional QHE, using a 1.5-m-wide 2DEG channel
with a gate across a segment of the channel (the gate
length is also approximately 1.5 m). Ohmic contacts
to the gated and ungated regions allowed
min
and
max
to be determined independently. Equations (4.36) and
(4.37) were found to hold to within 0.5% accuracy. This
is illustrated in Fig. 97 for the case that
max
= 1 and

min
varying from 1 to 2/3 on increasing the negative gate
voltage (at a xed magnetic eld of 0.114 T). Similar
results were obtained
480
for the case that
max
=
2
3
and

min
varies from
2
3
to
1
3
.
Adiabatic transport in the fractional QHE can be stud-
ied by the selective population and detection of frac-
tional edge channels, achieved by means of barriers in two
FIG. 98 (a) Schematic of the experimental geometry of
Kouwenhoven et al.
481
The crossed squares are contacts to the
2DEG. One current lead and one voltage lead contain a bar-
rier (shaded), of which the height can be adjusted by means
of a gate (not drawn). The current I ows between contacts
1 and 3; the voltage V is measured between contacts 2 and 4.
(b) Arrangement of incompressible bands (hatched) and edge
channels near the two barriers. In the absence of scattering
between the two fractional edge channels, one would measure
a Hall conductance GH I/V that is fractionally quantized
at
2
3
e
2
/h, although the bulk has unit lling factor. Taken
from C. W. J. Beenakker, Phys. Rev. Lett. 64, 216 (1990).
closely separated current and voltage contacts (Fig. 98a).
The analysis using Eq. (4.33) is completely analogous to
the analysis of the experiment in the integer QHE,
426
discussed in Section IV.B. Figure 98b illustrates the ar-
rangement of edge channels and incompressible bands for
the case that the chemical potential lies in an energy gap
for the bulk 2DEG (at =
bulk
), as well as for the
two barriers (at
I
and
V
for the barrier in the current
and voltage lead, respectively). Adiabatic transport is
assumed over the barrier, as well as from barrier I to
barrier V (for the magnetic eld direction indicated in
Fig. 98). Equation (4.33) for this case reduces to
I =
e
h

I
, 0 =
e
h

e
h
min(
I
,
V
)
I
, (4.38)
so the Hall conductance G
H
= eI/
V
becomes
G
H
=
e
2
h
max(
I
,
V
)
e
2
h

bulk
. (4.39)
The quantized Hall plateaux are determined by the frac-
tional lling factors of the current and voltage leads, not
96
FIG. 99 Anomalously quantized Hall conductance in the
geometry of Fig. 98, in accord with Eq. (4.39) (
bulk
= 1,
I = V decreases from 1 to 2/3 as the negative gate volt-
age is increased). The temperature is 20 mK. The rapidly
rising part (dotted) is an artifact due to barrier pinch-o.
Taken from L. P. Kouwenhoven et al., Phys. Rev. Lett. 64,
685 (1990).
of the bulk 2DEG. Kouwenhoven et al.
481
have demon-
strated the selective population and detection of frac-
tional edge channels in a device with a 2-m separation
of the gates in the current and voltage leads. The gates
extended over a length of 40 m along the 2DEG bound-
ary. In Fig. 99 we reproduce one of the experimental
traces of Kouwenhoven et al. The Hall conductance is
shown for a xed magnetic eld of 7.8 T as a function
of the gate voltage (all gates being at the same voltage).
As the barrier heights in the two leads are increased, the
Hall conductance decreases from the bulk value 1 e
2
/h
to the value
2
3
e
2
/h determined by the leads, in accord
with Eq. (4.39). A more general formula for G
H
valid also
in between the quantized plateaux is shown in Ref.
481
to
be in quantitative agreement with the experiment.
MacDonald has, independent of Ref.
482
, proposed a
dierent generalized Landauer formula for the fractional
QHE.
483
The dierence with Eq. (4.32) is that the weight
factors in MacDonalds formula can take on both positive
and negative values (corresponding to electron and hole
channels). In the case of local equilibrium at the edge, the
sum of weight factors is such that the two formulations
give identical results. The results dier in the absence
of local equilibrium if fractional edge channels are selec-
tively populated and detected. For example, MacDonald
predicts a negative longitudinal resistance in a conductor
at lling factor =
2
3
containing a segment at = 1. An-
other implication of Ref.
483
is that the two-terminal con-
ductance G of a conductor at
max
= 1 containing a po-
tential barrier at lling factor
min
is reduced to
1
3
e
2
/h
if
min
=
1
3
[in accord with Eq. (4.35)], but remains at
1 e
2
/h if
min
= 2/3. That this is not observed experi-
mentally (cf. Fig. 96) could be due to interedge channel
scattering, as argued by MacDonald. The experiment
by Kouwenhoven et al.
481
(Fig. 99), however, is appar-
ently in the adiabatic regime, and was interpreted in Fig.
98 in terms of an edge channel of weight
1
3
at the edge
of a conductor at = 1. In MacDonalds formulation,
the conductor at = 1 has only a singe edge channel
of weight 1. This would need to be reconciled with the
experimental observation of quantization of the Hall con-
ductance at
2
3
e
2
/h.
We conclude this section by briey addressing the ques-
tion: What charge does the resistance measure? The
fractional quantization of the conductance in the exper-
iments discussed is understood as a consequence of the
fractional weight factors in the generalized Landauer for-
mula (4.32). These weight factors
p
=
p

p1
are not in general equal to e

/e, with e

the fractional
charge of the quasiparticle excitations of Laughlins in-
compressible state (cf. Section IV.C.1). The reason for
the absence of a one-to-one correspondence between
p
and e

is that the edge channels themselves are not

incompressible.
482
The transmission probabilities in Eq.
(4.32) refer to charged gapless excitations of the edge
channels, which are not identical to the charge e

excita-
tions above the energy gap in the incompressible bands
(the latter charge might be obtained from thermal activa-
tion measurements; cf. Ref.
494
). It is an interesting and
(to date) unsolved problem to determine the charge of
the edge channel excitations. Kivelson and Pokrovsky
495
have suggested performing tunneling experiments in the
fractional QHE regime for such a purpose, by using
the charge dependence of the magnetic length ( h/eB)
1/2
(which determines the penetration of the wave function in
a tunnel barrier and, hence, the transmission probability
through the barrier). Alternatively, one could use the h/e
periodicity of the Aharanov-Bohm magnetoresistance os-
cillations as a measure of the edge channel charge. Sim-
mons et al.
496
nd that the characteristic eld scale of
quasiperiodic resistance uctuations in a 2-m-wide Hall
bar increases from 0.016 T 30% near = 1, 2, 3, 4 to
0.05 T 30% near =
1
3
. This is suggestive of a reduc-
tion in charge from e to e/3, but not conclusive since the
area for the Aharonov- Bohm eect is not well dened in
a Hall bar (cf. Section IV.D).
D. Aharonov-Bohm eect in strong magnetic elds
As mentioned briey in Section II.D, the Aharonov-
Bohm oscillations in the magnetoresistance of a ring are
gradually suppressed in strong magnetic elds. This sup-
pression provides additional support for edge channel
trans- port in the quantum Hall eect regime (Section
IV.D.1). Entirely new mechanisms for the Aharonov-
Bohm eect become operative in strong magnetic elds.
These mechanisms, resonant tunneling and resonant re-
ection of edge channels, do not require a ring geometry.
Theory and experiments on Aharonov-Bohm oscillations
in singly connected geometries are the subject of Section
IV.D.2.
97
FIG. 100 Illustration of a localized edge channel circulating
along the inner perimeter of a ring, and of extended edge chan-
nels on the leads and on the outer perimeter. No Aharonov-
Bohm magnetoresistance oscillations can occur in the absence
of scattering between these two types of edge channels.
1. Suppression of the Aharonov-Bohm eect in a ring
In Section II.D we have seen how the quantum in-
terference of clockwise and counterclockwise trajecto-
ries in a ring in the diusive transport regime leads to
magnetoresistance oscillations with two dierent peri-
odicities: the fundamental Aharonov-Bohm eect with
B = (h/e)S
1
periodicity, and the harmonic with
B = (h/2e)S
1
periodicity, where S is the area of the
ring. In arrays of rings only the h/2e eect is observable,
since the h/e eect has a sample specic phase and is
averaged to zero. In experiments by Timp et al.
69
and
by Ford et al.
74
on single rings in the 2DEG of high-
mobility GaAs-AlGaAs heterostructures, the h/e eect
was found predominantly. The amplitude of these oscil-
lations is strongly reduced
69,74,195,497
by a large magnetic
eld (cf. the magnetoresistance traces shown in Fig. 26).
This suppression was found to occur for elds such that
2l
cycl
< W, where W is the width of the arms of the ring.
The reason is that in strong magnetic elds the states
at the Fermi level that can propagate through the ring
are edge states at the outer perimeter. These states do
not complete a revolution around the ring (see Fig. 100).
Scattering between opposite edges is required to complete
a revolution, but such backscattering would also lead to
a nonzero longitudinal resistance. This argument
112,498
explains the absence of Aharonov-Bohm oscillations on
the quantized Hall plateaux, where the longitudinal re-
sistance is zero. Magnetoresistance oscillations return be-
tween the plateaux in the Hall resistance, but at a larger
value of B than in weak elds. Timp et al.
497
have
argued that the Aharonov-Bohm oscillations in a ring
in strong magnetic elds are associated with scattering
from the outer edge to edge states circulating along the
inner perimeter of the ring. The smaller area enclosed
by the inner perimeter explains the increase in B. This
interpretation is supported by numerical calculations.
497
FIG. 101 Two-terminal magnetoresistance of a point contact
for a series of gate voltages at T = 50 mK, showing oscillations
that are periodic in B between the quantum Hall plateaux.
The second, third, and fourth curves from the bottom have
osets of, respectively, 5, 10, and 15 k. The rapid oscillations
below 1 T are Shubnikov-De Haas oscillations periodic in 1/B,
originating from the wide 2DEG regions. The sharp peak
around B = 0 T originates from the ohmic contacts. Taken
from P. H. M. van Loosdrecht et al., Phys. Rev. B 38, 10162
(1988).
2. Aharonov-Bohm eect in singly connected geometries
(a) Point contact. Aharonov-Bohm oscillations in
the magnetoresistance of a quantum point contact were
discovered by van Loosdrecht et al.
292
The magnetic eld
dependence of the two-terminal resistance is shown in
Fig. 101, for various gate voltages. The periodic os-
cillations occur predominantly between quantum Hall
plateaux, in a limited range of gate voltages, and only
at low temperatures (in Fig. 101, T = 50 mK; the eect
has disappeared at 1 K). The ne structure is very well
reproducible if the sample is kept in the cold, but changes
after cycling to room temperature. As one can see from
the enlargements in Fig. 102, a splitting of the peaks oc-
curs in a range of magnetic elds, presumably as spin
splitting becomes resolved. A curious aspect of the eect
(which has remained unexplained) is that the oscillations
have a much larger amplitude in one eld direction than
in the other (see Fig. 101), in apparent conict with the
B symmetry of the two-terminal resistance required by
the reciprocity relation (3.16) in the absence of magnetic
impurities. Other devices of the same design did not
show oscillations of well-dened periodicity and had a
two-terminal resistance that was approximately B sym-
metric.
Figure 103 illustrates the tunneling mechanism for the
periodic magnetoresistance oscillations as it was origi-
nally proposed
292
to explain the observations. Because
of the presence of a barrier in the point contact, the elec-
trostatic potential has a saddle form. Equipotentials at
98
FIG. 102 Curves a and b are close-ups of the curve for Vg =
1.7 V in Fig. 101. Curve c is a separate measurement on
the same device (note the dierent eld scale due to a change
in electron density in the constriction). Taken from P. H. M.
van Loosdrecht et al., Phys. Rev. B 38, 10162 (1988).
FIG. 103 Equipotentials at the guiding center energy in
the saddle-shaped potential created by a split gate (shaded).
Aharonov-Bohm oscillations in the point contact magnetore-
sistance result from the interference of tunneling paths ab and
adcb. Tunneling from a to b may be assisted by an impurity
at the entrance of the constriction. Taken from P. H. M. van
Loosdrecht et al., Phys. Rev. B 38, 10162 (1988).
the guiding center energy (4.1) are drawn schematically
in Fig. 103 (arrows indicate the direction of motion along
the equipotential). An electron that enters the constric-
tion at a can be reected back into the broad region by
tunneling to the opposite edge, either at the potential
step at the entrance of the constriction (from a to b) or
at its exit (from d to c). These two tunneling paths ac-
quire an Aharonov-Bohm phase dierence
499
of eBS/h
(were S is the enclosed area abcd), leading to periodic
magnetoresistance oscillations. (Note that the periodic-
ity B may dier
438,500
somewhat from the usual expres-
sion B = h/eS, since S itself is B-dependent due to the
B-dependence of the guiding center energy.) This mech-
anism shows how an Aharonov-Bohm eect is possible in
principle in a singly connected geometry: The point con-
tact behaves as if it were multiply connected, by virtue
of the spatial separation of edge channels moving in op-
posite directions. (Related mechanisms, based on circu-
lating edge currents, have been considered for Aharonov-
Bohm eects in small conductors.
473,474,501,502,503
) The
oscillations periodic in B are only observed at large mag-
netic elds (above about 1 T; the oscillations at lower
elds are Shubnikov-De Haas oscillations periodic in 1/B,
due to the series resistance of the wide 2DEG regions). At
low magnetic elds the spatial separation of edge chan-
nels responsible for the Aharanov-Bohm eect is not yet
eective. The spatial separation can also be destroyed
by a large negative gate voltage (top curve in Fig. 101),
when the width of the point contact becomes so small
that the wave functions of edge states at opposite edges
overlap.
Although the mechanism illustrated in Fig. 103 is at-
tractive because it is an intrinsic consequence of the point
contact geometry, the observed well-dened periodicity of
the magnetoresistance oscillations requires that the po-
tential induced by the split gate varies rapidly over a
short distance (in order to have a well-dened area S). A
smooth saddle potential seems more realistic. Moreover,
one would expect the periodicity to vary more strongly
with gate voltage than the small 10% variation observed
experimentally as V
g
is changed from 1.4 to 1.7 V.
Glazman and Jonson
438
have proposed that one of the
two tunneling processes (from a to b in Fig. 103) is medi-
ated by an impurity outside but close to the constriction.
The combination of impurity and point contact intro-
duces a well-dened area even for a smooth saddle po-
tential, which moreover will not be strongly gate-voltage-
dependent. Such an impurity-assisted Aharonov-Bohm
eect in a quantum point contact has been reported by
Wharam et al.
504
In order to study the Aharonov-Bohm
eect due to interedge channel tunneling under more con-
trolled conditions, a double-point contact device is re-
quired, as discussed below.
(b) Cavity. Van Wees et al.
500
performed magnetore-
sistance experiments in a geometry shown schematically
in Fig. 104. A cavity with two opposite point contact
openings is dened in the 2DEG by split gates. The
diameter of the cavity is approximately 1.5 m. The
conductances G
A
and G
B
of the two point contacts A
and B can be measured independently (by grounding
one set of gates), with the results plotted in Fig. 105a,b
(for V
g
= 0.35 V on either gate A or B). The conduc-
tance G
C
of the cavity (for V
g
= 0.35 V on both the
split gates) is plotted in Fig. 105c. A long series of peri-
odic oscillations is observed between two quantum Hall
plateaux. Similar series of oscillations (but with a dier-
ent periodicity) have been observed between other quan-
tum Hall plateaux. The oscillations are suppressed on the
plateaux themselves. The amplitude of the oscilIations
99
FIG. 104 Cavity (of 1.5 m diameter) dened by a double
set of split gates A and B. For large negative gate voltages the
2DEG region under the narrow gap between gates A and B
is fully depleted, while transmission remains possible over the
potential barrier in the wider openings at the left and right
of the cavity. Taken from B. J. van Wees et al., Phys. Rev.
Lett. 62, 2523 (1989).
FIG. 105 Magnetoconductance experiments on the device of
Fig. 104 at 6 mK, for a xed gate voltage of 0.35 V. (a) Con-
ductance of point contact A, measured with gate B grounded.
(b) Conductance of point contact B (gate A grounded). (c)
Measured conductance of the entire cavity. (d) Calculated
conductance of the cavity, obtained from Eqs. (4.40) and
(4.41) with the measured GA and GB as input. Taken from
B. J. van Wees et al., Phys. Rev. Lett. 62, 2523 (1989).
FIG. 106 Illustration of mechanisms leading to Aharonov-
Bohm oscillations in singly connected geometries. (a) Cavity
containing a circulating edge state. Tunneling through the
left and right barriers (as indicated by dashed lines) occurs
with transmission probabilities TA and TB. On increasing the
magnetic eld, resonant tunneling through the cavity occurs
periodically each time the ux enclosed by the circulating
edge state increases by one ux quantum h/e. (b) A circu-
lating edge state bound on a local potential maximum causes
resonant backscattering, rather than resonant transmission.
is comparable to that observed in the experiment on a
single point contact
292
(discussed before), but the period
is much smaller (consistent with a larger eective area
in the double-point contact device), and no splitting of
the peaks is observed (presumably due to a fully resolved
spin degeneracy). No gross B asymmetries were found
in the present experiment, although an accurate test of
the symmetry on eld reversal was not possible because of
diculties with the reproducibility. The oscillations are
quite fragile, disappearing when the temperature is raised
above 200 mK or when the voltage across the device ex-
ceeds 40 V (the data in Fig. 105 were taken at 6 mK
and 6 V). The experimental data are well described by
resonant transmission through a circulating edge state in
the cavity,
500
as illustrated in Fig. 106a and described
in detail later. Aharonov-Bohm oscillations due to reso-
nant transmission through a similar structure have been
reported by Brown et al.
505
and analyzed theoretically
by Yosephin and Kaveh.
506
(c) Resonant transmission and reection of edge
channels. The electrostatic potential in a point contact
has a saddle shape (cf. Fig. 103), due to the combination
of the lateral connement and the potential barrier. The
height of the barrier can be adjusted by means of the
gate voltage. An edge state with a guiding center energy
below the barrier height is a bound state in the cavity
formed by two opposite point contacts, as is illustrated in
Fig. 106a. Tunneling of edge channels through the cavity
via this bound state occurs with transmission probability
100
T
AB
, which for a singe edge channel is given by
474,498
T
AB
=

t
A
t
B
1 r
A
r
B
exp(ie/h)

2
=
T
A
T
B
1 +R
A
R
B
2(R
A
R
B
)
1/2
cos(
0
+ e/h)
.
(4.40)
Here t
A
and r
A
are the transmission and reection prob-
ability amplitudes through point contact A, T
A
[t
A
[
2
,
and R
A
[r
A
[
2
= 1 T
A
are the transmission and re-
ection probabilities, and t
B
, r
B
, T
B
, R
B
denote the corre-
sponding quantities for point contact B. In Eq. (4.40) the
phase acquired by the electron on one revolution around
the cavity is the sum of the phase
0
from the reec-
tion probability amplitudes (which can be assumed to be
only weakly B-dependent) and of the Aharonov-Bohm
phase BS, which varies rapidly with B ( is the
ux through the area S enclosed by the equipotential
along which the circulating edge state is extended). Res-
onant transmission occurs periodically with B, whenever

0
+ e/h is a multiple of 2. In the weak coupling
limit (T
A
, T
B
1), Eq. (4.40) is equivalent to the Breit-
Wigner resonant tunneling formula (3.43). This equiv-
alence has been discussed by B uttiker,
386
who has also
pointed out that the Breit-Wigner formula is more gen-
erally applicable to the case that several edge channels
tunnel through the cavity via the same bound state.
In the case that only a single (spin-split) edge chan-
nel is occupied in the 2DEG, the conductance G
C
=
(e
2
/h)T
AB
of the cavity follows directly from Eq. (4.40).
The transmission and reection probabilities can be de-
termined independently from the individual point con-
tact conductances G
A
= (e
2
/h)T
A
(and similarly for
G
B
), at least if one may assume that the presence of the
cavity has no eect on T
A
and T
B
itself (but only on the
total transmission probability T
AB
). If N > 1 spin-split
edge channels are occupied and the N 1 lowest-index
edge channels are fully transmitted, one can write
G
C
=
e
2
h
(N 1 +T
AB
), G
A
=
e
2
h
(N 1 +T
A
),
G
B
=
e
2
h
(N 1 +T
B
). (4.41)
Van Wees et al.
500
have compared this simple model with
their experimental data, as shown in Fig. 105. The trace
in Fig. 105d has been calculated from Eqs. (4.40) and
(4.41) by using the individual point contact conductances
in Fig. 105a,b as input for T
A
and T
B
. The ux has
been adjusted to the experimental periodicity of 3 mT,
and the phase
0
in Eq. (4.40) has been ignored (since
that would only amount to a phase shift of the oscilla-
tions). Energy averaging due to the nite temperature
and voltage has been taken into account in the calcula-
tion. The agreement with experimental trace (Fig. 105c)
is quite satisfactory.
Resonant reection of an edge channel can occur in
addition to the resonant transmission already consid-
ered. Aharonov-Bohm oscillations due to interference of
the reections at the entrance and exit of a point con-
tact, illustrated in Fig. 103, are one example of resonant
reection.
292
Jain
498
has considered resonant reection
via a localized state circulating around a potential max-
imum, as in Fig. 106b. Such a maximum may result
naturally from a repulsive scatterer or articially in a
ring geometry (cf. Fig. 100). Tunneling of an edge state
at each of the channel boundaries through the localized
state occurs with probabilities T
A
and T
B
. The reec-
tion probability of the edge channel is still given by T
AB
in Eq. (4.24), but the channel conductance G
C
is now a
decreasing function of T
AB
, according to
G
C
=
e
2
h
(N T
AB
). (4.42)
Quasi-periodic magnetoresistance oscillations have been
observed in narrow channels by several groups.
70,496,507
These may occur by resonant reection via one or more
localized states in the channel, as in Fig. 106b.
E. Magnetically induced band structure
The one-dimensional nature of edge channel trans-
port has recently been exploited in an innovative way by
Kouwenhoven et al.
250
to realize a one-dimensional su-
perlattice exhibiting band structure in strong magnetic
elds. The one-dimensionality results because only the
highest-index edge channel (with the smallest guiding
center energy) has an appreciable backscattering prob-
ability. The N 1 lower-index edge channels propa-
gate adiabatically, with approximately unit transmission
probability. One-dimensionality in zero magnetic elds
cannot be achieved with present techniques. That is one
important reason why the zero-eld superlattice experi-
ments described in Section II.G could not provide con-
clusive evidence for a bandstructure eect. The work
by Kouwenhoven et al.
250
is reviewed in Section IV.E.1.
The magnetically induced band structure diers in an in-
teresting way from the zero-eld band structure familiar
from solid-state textbooks, as we show in Section IV.E.2.
1. Magnetotransport through a one-dimensional superlattice
The device studied by Kouwenhoven et al.
250
is shown
in the inset of Fig. 107. A narrow channel is dened in
the 2DEG of a GaAs-AlGaAs heterostructure by two op-
posite gates. One of the gates is corrugated with period
a = 200 nm, to introduce a periodic modulation of the
conning potential. At large negative gate voltages the
channel consists of 15 cavities [as in Section IV.D.2(b)]
coupled in series. The conductance of the channel was
measured at 10 mK in a xed magnetic eld of 2 T, as
a function of the voltage on the gate that denes the
smooth channel boundary. The results, reproduced in
101
FIG. 107 Inset: Corrugated gate used to dene a narrow
channel with a one-dimensional periodic potential (the total
number of barriers is 16, corresponding to 15 unit cells). Plot-
ted is the conductance in a magnetic eld of 2 T as a func-
tion of the voltage on the smooth gate at 10 mK. The deep
conductance minima (marked by + and ) are attributed to
minigaps, and the 15 enclosed maxima to discrete states in
the miniband. Taken from L. P. Kouwenhoven et al., Phys.
Rev. Lett. 65, 361 (1990).
FIG. 108 Top: Calculated transmission probability TN of
an edge channel through a periodic potential of N = 15 peri-
ods as a function of the Aharonov-Bohm phase eBS/h (with
S the area of one unit cell). The transmission probability
through a single barrier is varied as shown in the bottom
panel. Taken from L. P. Kouwenhoven et al., Phys. Rev.
Lett. 65, 361 (1990).
Fig. 107, showtwo pronounced conductance dips (of mag-
nitude 0.1 e
2
/h), with 15 oscillations in between of con-
siderably smaller amplitude. The two deep and widely
spaced dips are attributed to minigaps, the more rapid
oscillations to discrete states in the miniband.
This interpretation is supported in Ref.
250
by a calcula-
tion of the transmission probability amplitude t
n
through
n cavities in series, given by the recursion formula
t
n
=
tt
n1
1 rr
n1
exp(i)
. (4.43)
Here t and r are transmission and reection probabil-
ity amplitudes of the barrier separating two cavities (all
cavitities are assumed to be identical), and = eBS/h
is the Aharonov-Bohm phase for a circulating edge state
enclosing area S. Equation (4.43) is a generalization of
Eq. (4.40) for a single cavity. The dependence on of
T
n
= [t
n
[
2
shown in Fig. 108 is indeed qualitatively simi-
lar to the experiment. Deep minima in the transmission
probability occur with periodicity = 2. Experimen-
tally (where S is varied via the gate voltage at constant
B) this would correspond to oscillations with periodic-
ity S = h/eB of Aharonov-Bohm oscillations in a sin-
gle cavity. The 15 smaller oscillations between two deep
minima have the periodicity of Aharonov-Bohm oscilla-
tions in the entire area covered by the 15 cavities. The
observation of such faster oscillations shows that phase
coherence is maintained in the experiment throughout
the channel and thereby provides conclusive evidence for
band structure in a lateral superlattice.
2. Magnetically induced band structure
(a) Skew minibands. The band structure in the
experiment of Kouwenhoven et al.
250
is present only in
the quantum Hall eect regime and can thus be said to
be magnetically induced. The magnetic eld breaks time-
reversal symmetry. Let us see what consequences that
has for the band structure.
The hamiltonian in the Landau gauge A = (0, Bx, 0)
is
H =
p
2
x
2m
+
(p
y
+ eBx)
2
2m
+V (x, y), V (x, y+a) = V (x, y),
(4.44)
where V is the periodically modulated conning poten-
tial. Blochs theorem is not aected by the presence of
the magnetic eld, since H remains periodic in y (in the
Landau gauge). The eigenstates have the form

nk
(x, y) = e
iky
f
nk
(x, y), f
nk
(x, y +a) = f
nk
(x, y),
(4.45)
where the function f is a solution periodic in y of the
eigenvalue problem
_
p
2
x
2m
+
(p
y
+ hk +eBx)
2
2m
+V (x, y)
_
f
nk
(x, y)
= E
n
(k, B)f
nk
(x, y).(4.46)
If the wave number k is restricted to the rst Brillouin
zone [k[ < /a, the index n labels both the subbands
from the lateral connement and the minibands from the
periodic modulation. Since E and V are real, one nds
by taking the complex conjugate of Eq. (4.46) that
E
n
(k, B) = E
n
(k, B). (4.47)
In zero magnetic elds the energy E is an even func-
tion of k, regardless of the symmetry of the potential
V . This can be viewed as a consequence of time-reversal
symmetry.
508
In nonzero magnetic elds, however, E is
only even in k if the lateral connement is symmetric:
E
n
(k, B) = E
n
(k, B) ; only if V (x, y) = V (x, y).
(4.48)
102
FIG. 109 Illustration of magnetically induced band structure
in a narrow channel with a weak periodic modulation of the
conning potential V (x) (for the case V (x) = V (x)). The
dashed curves represent the unperturbed dispersion relation
(4.49) for a single Landau level. Skew minibands result from
the broken time-reversal symmetry in a magnetic eld.
To illustrate the formation of skew minibands in a mag-
netically induced band structure, we consider the case of
a weak periodic modulation V
1
(y) of the conning poten-
tial V (x, y) = V
0
(x) + V
1
(x, y). The dispersion relation
E
0
n
(k) in the absence of the periodic modulation can be
approximated by
E
0
n
(k) = (n
1
2
) h
c
+V
0
(x = kl
2
m
). (4.49)
The index n labels the Landau levels, and the wave
number k runs from to +. The semiclassical
approximation (4.49) is valid if the conning poten-
tial V
0
is smooth on the scale of the magnetic length
l
m
( h/eB)
1/2
. [Equation (4.49) follows from the guid-
ing center energy (4.1), using the identity x kh/eB
between the guiding center coordinate and the wave num-
ber; cf. Section III.A.1] For simplicity we restrict our-
selves to the strictly one-dimensional case of one Landau
level and suppress the Landau level index in what follows.
To rst order in the amplitude of the periodic modula-
tion V
1
, the zeroth-order dispersion relation is modied
only near the points of degeneracy K
p
dened by
E
0
[K
p
p(2/a)] = E
0
(K
p
), p = 1, 2, . . . . (4.50)
A gap opens near K
p
, leading to the formation of a band
structure as illustrated in Fig. 109. The gaps do not oc-
cur at multiples of /a, as in a conventional 1D band
structure. Moreover, the maxima and minima of two
subsequent bands occur at dierent k-values. This im-
plies indirect optical transitions between the bands if the
Fermi level lies in the gap.
It is instructive to consider the special case of a
parabolic conning potential V
0
(x) =
1
2
m
2
0
x
2
in more
detail, for which the zeroth-order dispersion relation can
be obtained exactly (Section II.F). Since the conne-
ment is symmetric in x, the minigaps in this case occur
at the Brillouin zone boundaries k = p/a. Other gaps at
points where the periodic modulation induces transitions
between dierent 1D subbands are ignored for simplicity.
From Eq. (2.59) one then nds that the Fermi energy lies
in a minigap when
E
F
= (n
1
2
) h +
h
2
2M
_
p
a
_
2
, (4.51)
with the denitions (
2
c
+
2
0
)
1/2
, M m
2
/
2
0
.
In the limiting case B = 0, Eq. (4.51) reduces to the
usual condition
249
that Bragg reection occurs when the
longitudinal momentum mv
y
is a multiple of h/a. In
the opposite limit of strong magnetic elds (
c

0
),
Eq. (4.51) becomes
aW
e
B = p
h
e
, W
e
2
_
2E
G
m
2
0
_
1/2
. (4.52)
The eective width W
e
of the parabolic potential is the
separation of the equipotentials at the guiding center en-
ergy E
G
E
F
(n
1
2
) h
c
.
The two-terminal conductance of the periodically mod-
ulated channel drops by e
2
/h whenever E
F
lies in a mini-
gap. If the magnetic eld dependence of W
e
is small,
then Eq. (4.52) shows that the magnetoconductance
oscillations have approximately the periodicity B
h/eaW
e
of the Aharonov-Bohm eect in a single unit
cell, in agreement with the calculations of Kouwenhoven
et al.
250
(Note that in their experiment the Fermi energy
is tuned through the minigap by varying the gate voltage
rather than the magnetic eld.) The foregoing analysis
is for a channel of innite length. The interference of re-
ections at the entrance and exit of a nite superlattice
of length L leads to transmission resonances
249,387
when-
ever k = p/L, as described by Eqs. (4.51) and (4.52)
after substituting L for a. These transmission resonances
are observed by Kouwenhoven et al. as rapid oscillations
in the conductance. The number of conductance max-
ima between two deep minima from the minigap equals
approximately the number L/a of unit cells in the super-
lattice. The number of maxima may become somewhat
larger than L/a if one takes into account reections at the
transition from a narrow channel to a wide 2DEG. This
might explain the observation in Ref.
250
of 16, rather
than 15, conductance maxima between two minigaps in
one particular experiment on a 15-period superlattice.
(b) Bloch oscillations. In zero magnetic elds, an
oscillatory current has been predicted to occur on ap-
plication of a dc electric eld to an electron gas in a
periodic potential.
509
This Bloch oscillation would result
from Bragg reection of electrons that, accelerated by
the electric eld, approach the band gap. A necessary
condition is that the eld be suciently weak that tun-
neling across the gap does not occur.
510,511,512,513
The
wave number increases in time according to

k = eE/h in
103
an electric eld E. The time interval between two Bragg
reections is 2/a

k = h/eaE. The oscillatory current

thus would have a frequency V e/h, with V = aE
the electrostatic potential drop over one unit cell. Bloch
oscillations have so far eluded experimental observation.
The successful demonstration
250
of miniband forma-
tion in strong magnetic elds naturally leads to the ques-
tion of whether Bloch oscillations might be observable in
such a system. This question would appear to us to have
a negative answer. The reason is simple, and it illus-
trates another interesting dierence of magnetically in-
duced band structure. In the quantum Hall eect regime
the electric eld is perpendicular to the current, so no ac-
celeration of the electrons occurs. Since

k = 0, no Bloch
oscillations should be expected.
References
[1] Y. Imry, in Directions in Condensed Matter Physics,
Vol. 1 (G. Grinstein and G. Mazenko, eds.). World Sci-
entic, Singapore, 1986.
[2] N. G. van Kampen, Stochastic Processes in Physics and
Chemistry. North-Holland, Amsterdam, 1981.
[3] H. van Houten and C. W. J. Beenakker, in Analogies in
Optics and Microelectronics (W. van Haeringen and D.
Lenstra, eds.). Kluwer Academic, Dordrecht, 1990.
[4] R. Landauer, IBM J. Res. Dev. 1, 223 (1957); 32, 306
(1988).
[5] M. B uttiker, Phys. Rev. Lett. 57, 1761 (1986).
[6] B. J. van Wees, H. van Houten, C. W. J. Beenakker, J.
G. Williamson, J. P. Kouwenhoven, D. van der Marel,
and C. T. Foxon, Phys. Rev. Lett. 60, 848 (1988).
[7] D. A. Wharam, T. J. Thornton, R. Newbury, M. Pep-
per, H. Ahmed, J. E. F. Frost, D. G. Hasko, D. C. Pea-
cock, D. A. Ritchie, and G. A. C. Jones, J. Phys. C 21,
L209 (1988).
[8] K. von Klitzing, G. Dorda, and M. Pepper, Phys. Rev.
Lett. 45, 494 (1980).
[9] Semiconductors and Semimetals, Vol. 35 (M. A. Reed,
ed.). Academic, New York, 1992.
[10] P. A. Lee, R. A. Webb and B. L. Altshuler, eds., Meso-
scopic Phenomena in Solids. Elsevier, Amsterdam, to
be published.
[11] B. L. Altshuler, R. A. Webb, and R. B. Laibowitz, eds.,
IBM J. Res. Dev. 32, 304437, 439579 (1988).
[12] Proceedings of the International Conference on Elec-
tronic Properties of Two-Dimensional Systems, IV-
VIII, Surf Sci. 113 (1982); 142 (1984); 170 (1986); 196
(1988); 229 (1990).
[13] M. J. Kelly and C. Weisbuch, eds., The Physics and
Fabrication of Microstructures and Microdevices. Proc.
Winter School Les Houches, 1986, Springer, Berlin,
1986.
[14] H. Heinrich, G. Bauer, and F. Kuchar, eds., Physics and
Technology of Submicron Structures. Springer, Berlin,
1988.
[15] M. Reed and W. P. Kirk, eds., Nanostructure Physics
and Fabrication. Academic, New York, 1989.
[16] S. P. Beaumont and C. M. Sotomayor-Torres, eds., Sci-
ence and Engineering of 1- and 0-Dimensional Semi-
conductors. Plenum, London, 1990.
[17] J. M. Chamberlain, L. Eaves, and J. C. Portal,
eds., Electronic Properties of Multilayers and Low-
Dimensional Semiconductor Structures. Plenum, Lon-
don, to be published.
[18] R. Landauer, Phys. Today 42, 119 (1989).
[19] R. T. Bate, Sci. Am. 258, 78 (1988).
[20] T. Ando, A. B. Fowler, and F. Stern, Rev. Mod. Phys.
54, 437 (1982).
[21] S. M. Sze, Physics of Semiconductor Devices. Wiley,
New York, 1981.
[22] E. H. Nicollian and J. R. Brew, Metal Oxide Semicon-
ductor Technology. Wiley, New York, 1982.
[23] B. J. F. Lin, M. A. Paalanen, A. C. Gossard, and D. C.
Tsui, Phys. Rev. B 29, 927 (1984).
[24] H. Z. Zheng, H. P. Wei, D. C. Tsui, and G. Weimann,
Phys. Rev. B 34, 5635 (1986).
[25] K. K. Choi, D. C. Tsui, and K. Alavi, Phys. Rev. B 36,
7751 (1987); Appl. Phys. Lett. 50, 110 (1987).
[26] H. van Houten, C. W. J. Beenakker, B. J. van Wees,
and J. E. Mooij, Surf. Sci. 196, 144 (1988).
[27] H. van Houten, C. W. J. Beenakker, M. E. I. Broekaart,
M. G. J. Heijman, B. J. van Wees, J. E. Mooij, and J.
P. Andre, Acta Electronica, 28, 27 (1988).
[28] D. J. Bishop, R. C. Dynes, and D. C. Tsui, Phys. Rev.
B 26, 773 (1982).
[29] W. J. Skocpol, L. D. Jackel, E. L. Hu, R. E. Howard,
and L. A. Fetter, Phys. Rev. Lett. 49, 951 (1982).
[30] K. K. Choi, Phys. Rev. B 28, 5774 (1983).
[31] H. van Houten, B. J. van Wees, and C. W. J. Beenakker,
in Ref.
14
.
[32] A. B. Fowler, A. Hartstein, and R. A. Webb, Phys. Rev.
Lett. 48, 196 (1982).
[33] M. Pepper and M. J. Uren, J. Phys. C 15, L617 (1982).
[34] C. C. Dean and M. Pepper, J. Phys. C 15, L1287 (1982).
[35] A. B. Fowler, J. J. Wainer, and R. A. Webb, IBM J.
Res. Dev. 32, 372 (1988).
[36] S. B. Kaplan and A. C. Warren, Phys. Rev. B 34, 1346
(1986).
[37] S. B. Kaplan and A. Hartstein, IBM J. Res. Dev. 32,
347 (1988); Phys. Rev. Lett. 56, 2403 (1986).
[38] R. G. Wheeler, K. K. Choi, A. Goel, R. Wisnie, and
D. E. Prober, Phys. Rev. Lett. 49, 1674 (1982).
[39] R. F. Kwasnick, M. A. Kastner, J. Melngailis, and P.
A. Lee, Phys. Rev. Lett. 52, 224 (1984).
[40] J. C. Licini. D. J. Bishop, M. A. Kastner, and J. Mel-
ngailis, Phys. Rev. Lett. 55, 2987 (1985).
[41] P. H. Woerlee, G. A. M. Hurkx, W. J. M. J. Josquin,
and J. F. C. M. Verhoeven, Appl. Phys. Lett. 47, 700
(1985); see also H. van Houten and P. H. Woerlee, Proc.
ICPS 18, p. 1515 (O. Engstrom, ed.). World Scientic,
Singapore, 1987.
[42] S. E. Laux and F. Stern, Appl. Phys. Lett. 49, 91 (1986).
[43] A. C. Warren, D. A. Antoniadis, and H. I. Smith, Phys.
Rev. Lett. 56, 1858 (1986).
[44] A. C. Warren, D. A. Antoniadis, and H. I. Smith, IEEE
Electron Device Lett., EDL-7, 413 (1986).
[45] W. J. Skocpol, P. M. Mankiewich. R. E. Howard, L. D.
Jackel, D. M. Tennant, and A. D. Stone, Phys. Rev.
Lett. 56, 2865 (1986).
[46] W. J. Skocpol, Physica Scripta T19, 95 (1987).
[47] K. S. Ralls, W. J. Skocpol, L. D. Jackel, R. E. Howard,
L. A. Fetter, R. W. Epworth, and D. M. Tennant, Phys.
Rev. Lett. 52, 228 (1984).
[48] R. E. Howard, W. J. Skocpol, L. D. Jackel, P. M.
104
Mankiewich, L. A. Fetter, D. M. Tennant, R. Epworth,
and K. S. Ralls, IEEE Trans. ED-32, 1669 (1985).
[49] H. L. Stormer, R. Dingle, A. C. Gossard, and W. Wieg-
man, Proc. 14th ICPS, p. 6 (B. L. H. Wilson, ed.). Insti-
tute of Physics, London, 1978; R. Dingle, H. L. Stormer,
A. C. Gossard, and W. Wiegman, Appl. Phys. Lett. 7,
665 (1978).
[50] J. J. Harris, J. A. Pals, and R. Woltjer, Rep. Prog. Phys.
52, 1217 (1989).
[51] S. Adachi, J. Appl. Phys. 58, R1 (1985).
[52] D. Delagebeaudeuf and N. T. Linh, IEEE Trans. ED-
28, 790 (1981).
[53] J. P. Kirtley, Z. Schlesinger, T. N. Theis. F. P. Milliken.
S. L. Wright. and L. F. Palmateer, Phys. Rev. B 34,
5414 (1986).
[54] L. Bliek. E. Braun. G. Hein. V. Kose, J. Niemeyer, G.
Weimann, and W. Schlapp, Semicond. Sci. Technol. 1,
110 (1986).
[55] K. K. Choi, D. C. Tsui, and S. C. Palmateer, Phys. Rev.
B 33, 8216 (1986).
[56] A. D. C. Grassie, K. M. Hutchings, M. Lakrimi, C. T.
Foxon, and J. J. Harris, Phys. Rev. B 36, 4551 (1987).
[57] T. Demel, D. Heitmann, P. Grambow. and K. Ploog,
Appl. Phys. Lett. 53, 2176 (1988).
[58] T. J. Thornton, M. Pepper, H. Ahmed, D. Andrews,
and G. J. Davies, Phys. Rev. Lett. 56, 1198 (1986).
[59] H. van Houten, B. J. van Wees, J. E. Mooij, C. W. J.
Beenakker, J. G. Williamson, and C. T. Foxon, Euro-
phys. Lett. 5, 721 (1988).
[60] S. E. Laux, D. J. Frank, and F. Stern, Surf Sci. 196,
101 (1988).
[61] A. Kumar, S. E. Laux, and F. Stern, Appl. Phys. Lett.
54, 1270 (1989).
[62] K. Ismail, W. Chu, D. A. Antoniadis, and H. I. Smith,
Appl. Phys. Lett. 52, 1071 (1988).
[63] H. van Houten, B. J. van Wees, M. G. J. Heijman, and
J. P. Andre, Appl. Phys. Lett. 49, 1781 (1986).
[64] R. E. Behringer, P. M. Mankiewich, and R. E. Howard,
J. Vac. Sci. Technol. B5, 326 (1987).
[65] A. Scherer, M. L. Roukes, H. G. Craighead, R. M.
Ruthen, E. D. Beebe, and J. P. Harbison, Appl. Phys.
Lett. 51, 2133 (1987).
[66] A. Scherer and M. L. Roukes, Appl. Phys. Lett. 55, 377
(1989).
[67] M. L. Roukes, A. Scherer, S. J. Allen, Jr., H. G. Craig-
head, R. M. Ruthen, E. D. Beebe, and J. P. Harbison,
Phys. Rev. Lett. 59, 3011 (1987).
[68] G. Timp, A. M. Chang, P. Mankiewich, R. Behringer,
J. E. Cunningham, T. Y. Chang, and R. E. Howard,
Phys. Rev. Lett. 59, 732 (1987).
[69] G. Timp, A. M. Chang, J. E. Cunningham, T. Y.
Chang, P. Mankiewich, R. Behringer, and R. E. Howard,
Phys. Rev. Lett. 58, 2814 (1987).
[70] A. M. Chang, G. Timp, T. Y. Chang, J. E. Cunning-
ham, P. M. Mankiewich, R. E. Behringer, and R. E.
Howard, Solid State Comm. 67, 769 (1988).
[71] K. Y. Lee, T. P. Smith, III, C. J. B. Ford, W. Hansen,
C. M. Knoedler, J. M. Hong, and D. P. Kern, Appl.
Phys. Lett. 55, 625 (1989).
[72] J. H. Davies and J. A. Nixon, Phys. Rev. B 39, 3423
(1989); J. H. Davies, in Ref.
15
.
[73] C. J. B. Ford, T. J. Thornton, R. Newbury, M. Pepper,
H. Ahmed, G. J. Davies, and D. Andrews, Superlattices
and Microstructures 4, 541 (1988).
[74] C. J. B. Ford, T. J. Thornton, R. Newbury, M. Pepper,
H. Ahmed, C. T. Foxon, J. J. Harris, and C. Roberts,
J. Phys. C. 21, L325 (1988).
[75] T. P. Smith, III, H. Arnot, J. M. Hong, C. M. Knoedler,
S. E. Laux, and H. Schmid, Phys. Rev. Lett. 59, 2802
(1987).
[76] T. P. Smith, III, J. A. Brum, J. M. Hong, C. M.
Knoedler, H. Arnot, and L. Esaki, Phys. Rev. Lett. 61,
585 (1988).
[77] C. J. B. Ford, S. Washburn, M. B uttiker, C. M.
Knoedler, and J. M. Hong, Phys. Rev. Lett. 62, 2724
(1989).
[78] W. Hansen, M. Horst, J. P. Kotthaus, U. Merkt, Ch.
Sikorski, and K. Ploog, Phys. Rev. Lett. 58, 2586
(1987).
[79] F. Brinkop, W. Hansen, J. P. Kotthaus, and K. Ploog,
Phys. Rev. B 37, 6547 (1988).
[80] H. van Houten, C. W. J. Beenakker, J. G. Williamson,
M. E. I. Broekaart, P. H. M. van Loosdrecht, B. J. van
Wees, J. E. Mooij, C. T. Foxon, and J. J. Harris, Phys.
Rev. B 39, 8556 (1989).
[81] T. Hiramoto, K. Hirakawa, Y. Iye, and T. Ikoma, Appl.
Phys. Lett. 54, 2103 (1989).
[82] T. L. Cheeks, M. L. Roukes, A. Scherer, and H. G.
Craighead, Appl. Phys. Lett. 53, 1964 (1988).
[83] K. Kash, J. M. Worlock, M. D. Sturge, P. Grabbe, J.
P. Harbison, A. Scherer, and P. S. D. Lin, Appl. Phys.
Lett. 53, 782 (1988).
[84] U. Meirav, M. Heiblum, and F. Stern, Appl. Phys. Lett.
52, 1268 (1988).
[85] U. Meirav, M. A. Kastner, M. Heiblum, and S. J. Wind,
Phys. Rev. B 40, 5871 (1989).
[86] A. D. Wieck and K. Ploog, Surf Sci. 229, 252 (1990);
Appl. Phys. Lett. 56, 928 (1990).
[87] P. M. Petro, A. C. Gossard, and W. Wiegmann, Appl.
Phys. Lett. 45, 620 (1984).
[88] T. Fukui and H. Saito, Appl. Phys. Lett. 50, 824 (1987).
[89] H. Asai, S. Yamada, and T. Fukui, Appl. Phys. Lett.
51, 1518 (1987).
[90] T. Fukui, and H. Saito, J. Vac. Sci. Technol. B6, 1373
(1988).
[91] J. Motohisa, M. Tanaka, and H. Sakaki, Appl. Phys.
Lett. 55, 1214 (1989).
[92] H. Yamaguchi and Y. Horikoshi, Jap. J. Appl. Phys. 28,
L1456 (1989).
[93] L. D. Landau and E. M. Lifshitz, Quantum Mechanics.
Pergamon, Oxford, 1977.
[94] N. W. Ashcroft and N. D. Mermin, Solid State Physics.
Holt, Rinehart and Winston, New York, 1976.
[95] R. Kubo, M. Toda, and N. Hashitsume, Statistical
Physics II. Springer, Berlin, 1985.
[96] A. A. Abrikosov, Fundamentals of the Theory of Metals.
North-Holland, Amsterdam, 1988.
[97] R. E. Prange and S. M. Girvin, eds., The Quantum Hall
Eect. Springer, New York, 1987.
[98] I. Stone. Phys. Rev. 6, 1 (1898).
[99] R. G. Chambers, in The Physics of Metals, Vol. 1 (J. M.
Ziman, ed.). Cambridge University Press, Cambridge,
1969.
[100] G. Brandli and J. L. Olsen, Mater. Sci. Eng. 4, 61
(1969).
[101] E. H. Sondheimer, Adv. Phys. 1, 1 (1952).
[102] A. B. Pippard, Magnetoresistance in Metals. Cambridge
University Press, Cambridge, 1989.
105
[103] K. Fuchs, Proc. Cambridge Philos. Soc. 34, 100 (1938).
[104] D. K. C. MacDonald, Nature 163, 637 (1949); D. K. C.
MacDonald and K. Sarginson, Proc. Roy. Soc. A 203,
223 (1950).
[105] E. H. Sondheimer, Nature 164, 920 (1949); Phys. Rev.
80, 401 (1950).
[106] S. B. Soer, J. Appl. Phys. 38, 1710 (1967).
[107] T. J. Thornton, M. L. Roukes, A. Scherer, and B. P.
van der Gaag, Phys. Rev. Lett. 63, 2128 (1989).
[108] K. Nakamura, D. C. Tsui, F. Nihey, H. Toyoshima, and
T. Itoh, Appl. Phys. Lett. 56, 385 (1990).
[109] C. W. J. Beenakker and H. van Houten, Phys. Rev. B
38, 3232 (1988).
[110] Z. Tesanovic, M. V. Jaric, and S. Maekawa, Phys. Rev.
Lett. 57, 2760 (1986).
[111] C. W. J. Beenakker, Phys. Rev. Lett. 62, 2020 (1989).
[112] M. B uttiker, Phys. Rev. B 38, 9375 (1988).
[113] H. van Houten, C. W. J. Beenakker, P. H. M. van Loos-
drecht, T. J. Thornton, H. Ahmed, M. Pepper, C. T.
Foxon, and J. J. Harris, Phys. Rev. B 37, 8534 (1988);
and unpublished.
[114] E. Ditlefsen and J. Lothe. Phil. Mag. 14, 759 (1966).
[115] E. Abrahams, P. W. Anderson, D. C. Licciardello, and
T. V. Ramakrishnan, Phys. Rev. Lett. 42, 673 (1979).
[116] P. W. Anderson, E. Abrahams, and T. V. Ramakrish-
nan, Phys. Rev. Lett. 43, 718 (1979).
[117] L. P. Gorkov, A. I. Larkin, and D. E. Khmelnitskii,
Pisma Zh. Eksp. Teor. Fiz. 30, 248 (1979) [JETP Lett.
30, 228 (1979)].
[118] B. L. Altshuler, D. Khmelnitskii, A. I. Larkin, and P.
A. Lee, Phys. Rev. B 22, 5142 (1980).
[119] A. Kawabata, J. Phys. Soc. Japan 49, 628 (1980).
[120] V. K. Dugaev and D. E. Khmelnitskii, Zh. Eksp. Teor.
Fiz. 86, 1784 (1984) [Sov. Phys. JETP 59, 1038 (1984)].
[121] B. L. Altshuler and A. G. Aronov, Pisma Zh. Eksp.
Teor. Fiz. 33, 515 (1981) [JETP Lett. 33, 499 (1981)].
[122] P. G. De Gennes and M. Tinkham, Phys. (N.Y.) 1, 107
(1964); see also P. G. De Gennes, Superconductivity of
Metals and Alloys, Chapter 8. Benjamin, New York,
1966.
[123] G. Bergmann, Phys. Rep. 107, 1 (1984); Phys. Rev. B
28, 2914 (1983).
[124] A. I. Larkin and D. E. Khmelnitskii, Usp. Fiz. Nauk
136, 536 (1982) [Sov. Phys. Usp. 25, 185 (1982)].
[125] D. E. Khmelnitskii, Physica B 126, 235 (1984).
[126] S. Chakravarty and A. Schmid, Phys. Rep. 140, 193
(1986).
[127] P. A. Lee and T. V. Ramakrishnan, Rev. Mod. Phys.
57, 287 (1985).
[128] B. L. Altshuler, A. G. Aronov, D. E. Khmelnitskii, and
A. I. Larkin, in Quantum Theory of Solids, p. 130. (I.
M. Lifshitz, ed.) Advances in Science and Technology in
the USSR, Physics Series, MIR, Moskow.
[129] R. P. Feynman and A. R. Hibbs, Quantum Mechanics
and Path Integrals. McGraw-Hill, New York, 1965.
[130] H. P. Wittman and A. Schmid, J. Low Temp. Phys. 69,
131 (1987).
[131] S. Hikami, A. I. Larkin, and Y. Nagaoka, Prog. Theor.
Phys. 63, 707 (1980).
[132] A. I. Larkin, Pisma Zh. Eksp. Teor. Fiz. 31, 239 (1980)
[JETP Lett. 31, 219 (1980)].
[133] Y. Kawaguchi and S. Kawaji. J. Phys. Soc. Japan 48.
699 (1980).
[134] R. G. Wheeler, Phys. Rev. B 24, 4645 (1981).
[135] M. J. Uren, R. A. Davis, M. Kaveh, and M. Pepper, J.
Phys. C 14, L395 (1981).
[136] D. A. Poole, M. Pepper, and R. W. Glew, J. Phys. C
14, L995 (1981).
[137] M. A. Paalanen, D. C. Tsui, and J. C. M. Hwang, Phys.
Rev. Lett. 51, 2226 (1983).
[138] D. M. Pooke, R. Mottahedeh, M. Pepper, and A.
Grundlach, Surf Sci. 196, 59 (1988); D. M. Pooke, N.
Paquin, M. Pepper, and A. Grundlach, J. Phys. Con-
dens. Matter 1, 3289 (1989).
[139] A. M. Chang, G. Timp, R. E. Howard, R. E. Behringer,
P. M. Mankiewich, J. E. Cunningham, T. Y. Chang,
and B. Chelluri, Superlattices and Microstructures 4,
515 (1988).
[140] B. L. Altshuler, Pisma Zh. Eksp. Teor. Fiz. 41, 530
(1985) [JETP Lett. 41, 648 (1985)].
[141] P. A. Lee and A. D. Stone, Phys. Rev. Lett. 55, 1622
(1985).
[142] P. A. Lee, Physica A 140, 169 (1986).
[143] D. S. Fisher and P. A. Lee, Phys. Rev. B 23, 6851
(1981).
[144] A. D. Stone, in Ref.
14
.
[145] P. A. Lee, A. D. Stone, and H. Fukuyama, Phys. Rev.
B 35, 1039 (1987).
[146] B. L. Altshuler and D. E. Khmelnitskii, Pisma Zh.
Eksp. Teor. Fiz. 42, 291 (1985) [JETP Lett. 42, 359
(1985)].
[147] C. W. J. Beenakker and H. van Houten, Phys. Rev. B
37, 6544 (1988).
[148] B. L. Altshuler, V. E. Kravtsov, and I. V. Lerner,
Pisma Zh. Eksp. Teor. Fiz. 43, 342 (1986) [JETP Lett.
43, 441 (1986)].
[149] M. B uttiker, Phys. Rev. B 35, 4123 (1987).
[150] S. Maekawa, Y. Isawa, and H. Ebisawa, J. Phys. Soc.
Japan 56, 25 (1987).
[151] H. U. Baranger: A. D. Stone, and D. P. DiVincenzo,
Phys. Rev. B 37, 6521 (1988).
[152] S. Hersheld and V. Ambegaokar, Phys. Rev. B 38,
7909 (1988).
[153] C. L. Kane, P. A. Lee, and D. P. DiVincenzo, Phys.
Rev. B 38, 2995 (1988).
[154] D. P. DiVincenzo and C. L. Kane, Phys. Rev. B 38,
3006 (1988).
[155] A. D. Benoit, C. P. Umbach, R. B. Laibowitz, and R.
A. Webb, Phys. Rev. Lett. 58, 2343 (1987).
[156] W. J. Skocpol, P. M. Mankiewich, R. E. Howard, L. D.
Jackel, D. M. Tennant, and A. D. Stone, Phys. Rev.
Lett. 58, 2347 (1987).
[157] W. E. Howard and F. F. Fang, Solid State Electronics
8, 82 (1965).
[158] A. Hartstein, R. A. Webb, A. B. Fowler, and J. J.
Wainer, Surf Sci. 142, 1 (1984).
[159] M. Ya. Azbel, Phys. Rev. B 28, 4106 (1983).
[160] P. A. Lee, Phys. Rev. Lett. 53, 2042 (1984).
[161] R. G. Wheeler, K. K. Choi, and R. Wisnie, Surf. Sci.
142, 19 (1984).
[162] W. J. Skocpol, L. D. Jackel, R. E. Howard, H. G. Craig-
head, L. A. Fetter, P. M. Mankiewich, P. Grabbe, and
D. M. Tennant, Surf. Sci. 142, 14 (1984).
[163] A. D. Stone, Phys. Rev. Lett. 54, 2692 (1985).
[164] R. A. Webb, S. Washburn, H. J. Haucke, A. D. Benoit,
C. P. Umbach, and F. P. Milliken, in Ref.
14
.
[165] R. A. Webb, S. Washburn, C. P. Umbach, and R. B.
Laibowitz, in Localization, Interaction, and Transport
106
Phenomena, p. 121. (B. Kramer, G. Bergmann, and Y.
Bruynseraede, eds.). Springer, New York, 1984.
[166] T. J. Thornton, M. Pepper, H. Ahmed, G. J. Davies,
and D. Andrews, Phys. Rev. B 36, 4514 (1987).
[167] H. van Houten, B. J. van Wees, J. E. Mooij, G. Roos,
and K.-F. Berggren, Superlattices and Microstructures
3, 497 (1987).
[168] R. P. Taylor, M. L. Leadbeater, G. P. Wittington, P. C.
Main, L. Eaves, S. P. Beaumont, I. McIntyre, S. Thoms,
and C. D. W. Wilkinson, Surf. Sci. 196, 52 (1988).
[169] T. Fukui and H. Saito, Jap. J. Appl. Phys. 27, L1320
(1988).
[170] A. D. Stone, Phys. Rev. B 39, 10736 (1989).
[171] P. Debray, J.-L. Pichard, J. Vicente, and P. N. Tung,
Phys. Rev. Lett. 63, 2264 (1989).
[172] N. O. Birge, B. Golding, and W. H. Haemmerle, Phys.
Rev. Lett. 62, 195 (1989).
[173] D. Mailly, M. Sanquer, J.-L. Pichard, and P. Pari, Eu-
rophys. Lett. 8, 471 (1989).
[174] S. Feng, P. A. Lee, and A. D. Stone, Phys. Rev. Lett.
56, 1960 (1986); erratum 56, 2772 (1986).
[175] B. L. Altshuler and B. Z. Spivak, Pisma Zh. Eksp.
Teor. Fiz. 42, 363 (1985) [JETP Lett. 42, 447 (1985)].
[176] A. M. Chang, K. Owusu-Sekyere, and T. Y. Chang,
Solid State Comm. 67, 1027 (1988).
[177] A. M. Chang, G. Timp, J. E. Cunningham, P. M.
Mankiewich, R. E. Behringer, R. E. Howard, and H.
U. Baranger, Phys. Rev. B 37, 2745 (1988).
[178] J. A. Simmons, D. C. Tsui, and G. Weimann, Surf Sci.
196, 81 (1988).
[179] S. Yamada, H. Asai, Y. Fukai, and T. Fukui, Phys. Rev.
B. (to be published).
[180] Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959).
[181] D. Yu. Sharvin and Yu. V. Sharvin, Pisma Zh. Teor.
Fiz. 34, 285 (1981) [JETP Lett. 34, 272 (1981).
[182] B. L. Altshuler, A. G. Aronov, and B. Z. Spivak, Pisma
Zh. Teor. Fiz. 33, 101 (1981) [JETP Lett. 33, 94 (1981)].
[183] B. L. Altshuler, A. G. Aronov, B. Z. Spivak, D. Yu
Sharvin, and Yu. V. Sharvin, Pisma Zh. Eksp. Teor.
Fiz. 35, 476 (1982) [JETP Lett. 35, 588 (1982)].
[184] Yu. V. Sharvin, Physica B 126, 288 (1984).
[185] M. Gijs, C. van Haesendonck, and Y. Bruynseraede,
Phys. Rev. Lett. 52, 2069 (1984); Phys. Rev. B 30,
2964 (1984).
[186] R. A. Webb, S. Washburn, C. P. Umbach, and R. B.
Laibowitz, Phys. Rev. Lett. 54, 2696 (1985).
[187] Y. Gefen, Y. Imry, and M. Ya. Azbel, Surf Sci. 142, 203
(1984); Phys. Rev. Lett. 52, 129 (1984).
[188] M. B uttiker, Y. Imry, R. Landauer, and S. Pinhas, Phys.
Rev. B 31, 6207 (1985).
[189] C. P. Umbach, C. Van Haesendonck, R. B. Laibowitz,
S. Washburn, and R. A. Webb, Phys. Rev. Lett. 56, 386
(1986).
[190] S. Washburn and R. A. Webb, Adv. Phys. 35, 375
(1986).
[191] A. G. Aronov and Yu. V. Sharvin, Rev. Mod. Phys. 59,
755 (1987).
[192] R. A. Webb, A. Hartstein, J. J. Wainer, and A. B.
Fowler, Phys. Rev. Lett. 54, 1577 (1985).
[193] K. Ishibashi, Y. Takagaki, K. Garno, S. Namba, S.
Ishida, K. Murase, Y. Aoyagi, and M. Kawabe, Solid
State Comm. 64, 573 (1987).
[194] A. M. Chang, G. Timp, T. Y. Chang, J. E. Cunning-
ham, B. Chelluri, P. M. Mankiewich, R. E. Behringer,
and R. E. Howard, Surf Sci. 196, 46 (1988).
[195] C. J. B. Ford, T. J. Thornton, R. Newbury, M. Pepper,
H. Ahmed, D. C. Peacock, D. A. Ritchie, J. E. F. Frost,
and G. A. C. Jones, Appl. Phys. Lett. 54, 21 (1989).
[196] C. P. Umbach, S. Washburn, R. B. Laibowitz, and R.
A. Webb, Phys. Rev. B 30, 4048 (1984).
[197] V. Chandrasekhar, M. J. Rooks, S. Wind, and D. E.
Prober, Phys. Rev. Lett. 55, 1610 (1985).
[198] S. Datta and S. Bandyopadhyay, Phys. Rev. Lett. 58,
717 (1987).
[199] S. Datta, M. R. Melloch, S. Bandyopadhyah, R. Noren,
M. Vaziri, M. Miller, and R. Reifenberger, Phys. Rev.
Lett. 55, 2344 (1985).
[200] J. R. Barker, in Ref.
15
.
[201] G. Timp. A. M. Chang, P. DeVegvar, R. E. Howard, R.
Behringer, J. E. Cunningham, and P. Mankiewich, Surf
Sci. 196, 68 (1988).
[202] S. Washburn, H. Schmid, D. Kern, and R. A. Webb,
Phys. Rev. Lett. 59, 1791 (1987).
[203] P. G. N. de Vegvar, G. Timp, P. M. Mankiewich, R.
Behringer, and J. Cunningham, Phys. Rev. B 40, 3491
(1989).
[204] B. L. Allshuler, A. G. Aronov, and P. A. Lee, Phys.
Rev. Lett. 44, 1288 (1980).
[205] H. Fukuyama, J. Phys. Soc. Japan, 48, 2169 (1980).
[206] B. L. Altshuler, A. G. Aronov, A. I. Larkin, and D. E.
Khmelnitskii, Zh. Eksp. Teor. Fiz. 81, 768 (1981) [Sov.
Phys. JETP 54, 411 (1981)].
[207] B. L. Altshuler, and A. G. Aronov, Solid State Comm.
46, 429 (1983).
[208] E. Abrahams, P. W. Anderson, P. A. Lee, and T. V.
Ramakrishnan, Phys. Rev. B 24, 6783 (1981).
[209] H. Fukuyama, J. Phys. Soc. Japan, 50, 3407, 3562
(1981); 51, 1105 (1982).
[210] P. A. Lee and T. V. Ramakrishnan, Phys. Rev. B 26,
4009 (1982).
[211] B. L. Altshuler and A. G. Aronov, in Electron-Electron
Interactions in Disordered Systems, p. 1 (A. L. Efros
and M. Pollak, eds.) North-Holland, Amsterdam, 1985.
[212] H. Fukuyama, in Electron-Electron Interactions in Dis-
ordered Systems, p. 155 (A. L. Efros and M. Pollak,
eds.). North-Holland, Amsterdam, 1985.
[213] A. Isihara, Solid State Physics, Vol. 42, p. 271 (H.
Ehrenreich and D. Turnbull, eds.). Academic, New
York, 1989.
[214] G. Bergmann, Phys. Rev. B 35, 4205 (1987).
[215] A. Houghton, J. R. Senna, and S. C. Ying, Phys. Rev.
B 25, 2196 (1982).
[216] S. M. Girvin, M. Jonson, and P. A. Lee, Phys. Rev. B
26, 1651 (1982).
[217] K.-F. Berggren, T. J. Thornton, D, J. Newson, and M.
Pepper, Phys, Rev. Lett. 57, 1769 (1986).
[218] K.-F. Berggren, G. Roos, and H. van Houten, Phys.
Rev. B 37, 10118 (1988).
[219] R. P. Taylor, P. C. Main, L. Eaves, S. P. Beaumont, I.
McIntyre, S. Thoms, and C. D. W. Wilkinson, J. Phys.
Condens. Matter 1, 10413 (1989).
[220] P. M. Mensz, R. G. Wheeler, C. T. Foxon, and J. J.
Harris, Appl. Phys. Lett. 50, 603 (1987); P. M. Mensz
and R. G. Wheeler, Phys. Rev. B 35, 2844 (1987).
[221] V. Falko, J. Phys. Condens. Matter 2, 3797 (1990).
[222] L. Smrcka, H. Havlova, and A. Ishara, J. Phys. C 19,
L457 (1986).
[223] K.-F. Berggren, and D. J. Newson, Semicond. Sci. Tech-
107
nol. 1, 327 (1986).
[224] J. A. Brum and G. Bastard, Superlattices and Mi-
crostructures 4, 443 (1988).
[225] M. J. Kearney and P. N. Butcher, J. Phys. C 20, 47
(1987).
[226] S. Das Sarma and X. C. Xie, Phys. Rev. B 35, 9875
(1987).
[227] P. Vasilopoulos and F. M. Peeters, Phys. Rev. B 40,
10079 (1989).
[228] H. Sakaki, Jap. J. Appl. Phys. 19, L735 (1980).
[229] G. Fishman, Phys. Rev. B 36, 7448 (1987).
[230] J. Lee and M. O. Vassell, J. Phys. C 17, 2525 (1984); J.
Lee and H. N. Spector, J. Appl. Phys. 57, 366 (1985).
[231] K. Ismail, D. A. Antoniadis, and H. I. Smith, Appl.
Phys. Lett. 54, 1130 (1989).
[232] M. Lakrimi, A. D. C. Grassie, K. M. Hutchings, J. J.
Harris, and C. T. Foxon, Semicond. Sci. Technol. 4, 313
(1989).
[233] J. J. Alsmeier, Ch. Sikorski, and U. Merkt, Phys. Rev.
B 37, 4314 (1988).
[234] Z. Tesanovic, J. Phys. C 20, L829 (1987).
[235] L. Esaki and R. Tsu, IBM J. Res. Dev. 14, 61 (1970).
[236] L. Esaki and L. L. Chang, Phys. Rev. Lett. 33, 495
(1974).
[237] L. Esaki, Rev. Mod. Phys. 46, 237 (1974).
[238] H. Sakaki, K. Wagatsuma, J. Hamasaki, and S. Saito,
Thin Solid Films 36, 497 (1976).
[239] R. T. Bate, Bull. Am. Phys. Soc. 22, 407 (1977).
[240] M. J. Kelly, J. Phys. C 18, 6341 (1985); Surf Sci. 170,
49 (1986).
[241] P. F. Bagwell and T. P. Orlando, Phys. Rev. B 40, 3735
(1989).
[242] A. C. Warren, D. A. Antoniadis, H. I. Smith, and J.
Melngailis, IEEE Electron Device Letts, EDL-6, 294
(1985).
[243] G. Bernstein and D. K. Ferry, J. Vac. Sci. Technol. B 5,
964 (1987).
[244] C. G. Smith, M. Pepper, R. Newbury, H. Ahmed, D.
G. Hasko, D. C. Peacock, J. E. F. Frost, D. A. Ritchie,
and G. A. C. Jones, J. Phys. Condens. Matter 2, 3405
(1990).
[245] J. M. Gaines, P. M. Petro, H. Kroemer, R. J. Simes,
R. S. Geels, and J. H. English, J. Vac. Sci. Technol. B
6, 1378 (1988).
[246] H. Sakaki, Jap. J. Appl. Phys. 28, L314 (1989).
[247] T. Cole, A. A. Lakhani, and P. J. Stiles, Phys. Rev.
Lett. 38, 722 (1977).
[248] G. E. W. Bauer and A. A. van Gorkum, in Ref.
16
.
[249] S. E. Ulloa, E. Casta no, and G. Kirczenow, Phys. Rev.
B. 41, 12350 (1990).
[250] L. P. Kouwenhoven, F. W. J. Hekking, B. J. van Wees,
C. J. P. M. Harmans, C. E. Timmering, and C. T.
Foxon, Phys. Rev. Lett. 65, 361 (1990).
[251] D. K. Ferry, in Ref.
14
.
[252] P. A. Puechner, J. Ma, R. Mezenner, W.-P. Liu, A. M.
Kriman, G. N. Maracas, G. Bernstein, and D. K. Ferry,
Surf Sci. 27, 137 (1987).
[253] J. C. Maan, Festkorperprobleme 27, 137 (1987).
[254] D. R. Hofstadter, Phys. Rev. B 14, 2239 (1976).
[255] D. Weiss, K. von Klitzing, K. Ploog, and G. Weimann,
Europhys. Lett. 8, 179 (1989); also in High Magnetic
Fields in Semiconductor Physics II (G. Landwehr, ed.).
Springer, Berlin, 1989.
[256] D. Weiss, C. Zhang, R. R. Gerhardts, K. von Klitzing,
and G. Weimann, Phys. Rev. B 39, 13020 (1989).
[257] R. R. Gerhardts, D. Weiss, and K. von Klitzing, Phys.
Rev. Lett. 62, 1173 (1989).
[258] R. W. Winkler, J. P. Kotthaus, and K. Ploog, Phys.
Rev. Lett. 62, 1177 (1989).
[259] P. Vasilopoulos and F. M. Peeters, Phys. Rev. Lett. 63,
2120 (1989); R. R. Gerhardts and C. Zhang, Phys. Rev.
Lett. 64, 1473 (1990).
[260] G. Knorr, F. R. Hansen, J. P. Lynov, H. L. Pecseli, and
J. J. Rasmussen, Physica Scripta 38, 829 (1988).
[261] A. V. Chaplik, Solid State Comm. 53, 539 (1985).
[262] E. S. Alves, P. H. Beton, M. Henini, L. Eaves, P. C.
Main, O. H. Hughes, G. A. Toombs, S. P. Beaumont,
and C. D. W. Wilkinson, J. Phys. Condens. Matter 1,
8257 (1989).
[263] D. Weiss, K. von Klitzing, G. Ploog, and G. Weimann,
Surf Sci. (to be published).
[264] K. Ismail, T. P. Smith, III, W. T. Masselink, and H. I.
Smith, Appl. Phys. Lett. 55, 2766 (1989).
[265] W. Hansen, T. P. Smith, III, K. Y. Lee, J. A. Brum, C.
M. Knoedler, J. M. Hong, and D. P. Kern, Phys. Rev.
Lett. 62, 2168 (1989).
[266] C. W. J. Beenakker, H. van Houten, and B. J. van Wees,
Superlattices and Microstructures 5, 127 (1989).
[267] R. E. Prange and T.-W. Nee, Phys. Rev. 168, 779
(1968).
[268] C. W. J. Beenakker and H. van Houten, Phys. Rev. Lett.
60, 2406 (1988).
[269] A. M. Kosevich and I. M. Lifshitz, Zh. Eksp. Teor. Fiz.
29, 743 (1955) [Sov. Phys. JETP 2, 646 (1956)]; M. S.
Khaikin, Adv. Phys. 18, 1 (1969).
[270] R. Vawter, Phys. Rev. 174, 749 (1968); A. Isihara and
K. Ebina, J. Phys. C 21, L1079 (1988).
[271] I. B. Levinson, Zh. Eksp. Teor. Fiz. 95, 2175 (1989)
[Sov. Phys. JETP 68, 1257 (1989)].
[272] M. C. Payne, J. Phys. Condens. Matter 1, 4931 (1989);
4939 (1989).
[273] E. N. Economou and C. M. Soukoulis, Phys. Rev. Lett.
46, 618 (1981).
[274] A. D. Stone and A. Szafer, IBM J. Res. Dev. 32, 384
(1988).
[275] J. Kucera and P. Streda. J. Phys. C 21, 4357 (1988).
[276] H. U. Baranger and A. D. Stone, Phys. Rev. B 40, 8169
(1989).
[277] D. J. Thouless, Phys. Rev. Lett. 47, 972 (1981).
[278] R. Landauer, Phys. Lett. A 85, 91 (1981).
[279] E. N. Engquist and P. W. Anderson, Phys. Rev. B 24,
1151 (1981).
[280] P. W. Anderson, D. J. Thouless, E. Abrahams, and D.
S. Fisher, Phys. Rev. B 22, 3519 (1980).
[281] D. C. Langreth and E. Abrahams, Phys. Rev. B 24,
2978 (1981).
[282] M. Ya. Azbel, J. Phys. C 14, L225 (1981).
[283] R. Landauer, J. Phys. Condens. Matter 1, 8099 (1989);
also in Ref.
15
.
[284] M. B uttiker, IBM J. Res. Dev. 32, 317 (1988).
[285] H. B. G. Casimir, Rev. Mod. Phys. 17, 343 (1945);
Philips Res. Rep. 1, 185 (1946); L. Onsager, Phys. Rev.
38, 2265 (1931; see also S. R. de Groot and P. Mazur,
Non-Equilibrium Thermodynamics. Dover, New York,
1984.
[286] A. D. Benoit, S. Washburn, C. P. Umbach, R. B. Lai-
bowitz, and R. A. Webb, Phys. Rev. Lett. 57, 1765
(1986).
108
[287] H. H. Sample, W. J. Bruno, S. B. Sample, and E. K.
Sichel, J. Appl. Phys. 61, 1079 (1987).
[288] L. L. Soethout, H. van Kempen, J. T. P. W. van Maar-
seveen, P. A. Schroeder, and P. Wyder, J. Phys. F 17,
L129 (1987).
[289] G. Timp, H. U. Baranger, P. deVegvar, J. E. Cun-
ningham, R. E. Howard, R. Behringer, and P. M.
Mankiewich, Phys. Rev. Lett. 60, 2081 (1988).
[290] P. G. N. de Vegvar, G. Timp, P. M. Mankiewich, J. E.
Cunningham, R. Behringer, and R. E. Howard, Phys.
Rev. B 38, 4326 (1988).
[291] A. I. Larkin and D. E. Khmelnitskii, Zh. Eksp. Teor.
Fiz. 91, 1815 (1986) [Sov. Phys. JETP 64, 1075 (1986)].
[292] P. H. M. van Loosdrecht, C. W. J. Beenakker, H. van
Houten, J. G. Williamson, B. J. van Wees, J. E. Mooij,
C. T. Foxon, and J. J. Harris, Phys. Rev. B 38, 10162
(1988).
[293] S. Datta, Phys. Rev. B 40, 5830 (1989).
[294] S. Feng. Phys. Lett. A 143, 400 (1990).
[295] J. C. Maxwell, A Treatise on Electricity and Magnetism.
Clarendon, Oxford, 1904.
[296] Yu. V. Sharvin, Zh. Eksp. Teor. Fiz. 48, 984 (1965)
[Sov. Phys. JETP 21, 655 (1965)].
[297] Yu. V. Sharvin and N. I. Bogatina, Zh. Eksp. Teor.
Phys. 56, 772 (1969) [Sov. Phys. JETP 29, 419 (1969)].
[298] J. K. Gimzewski and R. Moller, Phys. Rev. B 36, 1284
(1987).
[299] N. D. Lang, Phys. Rev. B 36, 8173 (1987).
[300] J. Ferrer, A. Martin-Rodero, and F. Flores, Phys. Rev.
B 38, 10113 (1988).
[301] N. D. Lang, Comm. Cond. Matt. Phys. 14, 253 (1989).
[302] N. D. Lang, A. Yacoby, and Y. Imry, Phys. Rev. Lett.
63, 1499 (1989).
[303] N. Garcia and H. Rohrer, J. Phys. Condens. Matter 1,
3737 (1989).
[304] R. Trzcinski, E. Gmelin, and H. J. Queisser, Phys. Rev.
B 35, 6373 (1987).
[305] A. G. M. Jansen, A. P. van Gelder, and P. Wyder, J.
Phys. C 13, 6073 (1980).
[306] G. Timp, R. Behringer, S. Sampere, J. E. Cunningham,
and R. E. Howard, in Ref.
15
; see also G. Timp in Ref.
9
.
[307] H. van Houten, C. W. J. Beenakker, and B. J. van Wees,
in Ref.
9
.
[308] B. J. van Wees, L. P. Kouwenhoven, E. M. M. Willems,
C. J. P. M. Harmans, J. E. Mooij, H. van Houten, C.
W. J. Beenakker, J. G. Williamson, and C. T. Foxon,
submitted to Phys. Rev. B.
[309] P. F. Bagwell and T. P. Orlando, Phys. Rev. B 40, 1456
(1989).
[310] G. Timp, in Ref.
10
.
[311] H. van Houten and C. W. J. Beenakker, in Ref.
15
.
[312] L. Escapa and N. Garcia, J. Phys. Condens. Matter 1,
2125 (1989).
[313] E. G. Haanappel and D. van der Marel, Phys. Rev. B
39, 5484 (1989); D. van der Marel and E. G. Haanappel,
Phys. Rev. B 39, 7811 (1989).
[314] G. Kirczenow, Solid State Comm. 68, 715 (1988); J.
Phys. Condens. Matter 1, 305 (1989).
[315] A. Szafer and A. D. Stone, Phys. Rev. Lett. 62, 300
(1989).
[316] E. Tekman and S. Ciraci, Phys. Rev. B 39, 8772 (1989);
Phys. Rev. B 40, 8559 (1989).
[317] Song He and S. Das Sarma, Phys. Rev. B 40, 3379
(1989).
[318] D. van der Marel, in Ref.
15
.
[319] N. Garcia and L. Escapa, Appl. Phys. Lett. 54, 1418
(1989).
[320] E. Casta no and G. Kirczenow, Solid State Comm. 70,
801 (1989).
[321] Y. Avishai and Y. B. Band, Phys. Rev. B 40, 12535
(1989).
[322] A. Kawabata, J. Phys. Soc. Japan 58, 372 (1989).
[323] I. B. Levinson, Pisma Zh. Eksp. Teor. Fiz. 48, 273
(1988) [JETP Lett. 48, 301 (1988)].
[324] A. Matulis and D. Segzda, J. Phys. Condens. Matter 1,
2289 (1989).
[325] L. I. Glazman, G. B. Lesovick, D. E. Khmelnitskii, R.
I. Shekhter, Pisma Zh. Teor. Fiz. 48, 218 (1988) [JETP
Lett. 48, 238 (1988)].
[326] A. Yacoby and Y. Imry, Phys. Rev. B 41, 5341 (1990).
[327] L. W. Molenkamp, A. A. M. Staring, C. W. J. Beenak-
ker, R. Eppenga, C. E. Timmering, J. G. Williamson,
C. J. P. M. Harmans, and C. T. Foxon, Phys. Rev. B
41, 1274 (1990).
[328] R. Landauer, Z. Phys. B 68, 217 (1987).
[329] C. W. J. Beenakker and H. van Houten, Phys. Rev. B
39, 10445 (1989).
[330] Y. Hirayama, T. Saku, and Y. Horikoshi, Phys. Rev. B
39, 5535 (1989).
[331] Y. Hirayama, T. Saku, and Y. Horikoshi, Jap. J. Appl.
Phys. 28, L701 (1989).
[332] R. J. Brown, M. J. Kelly, R. Newbury, M. Pepper, B.
Miller, H. Ahmed, D. G. Hasko, D. C. Peacock, D. A.
Ritchie, J. E. F. Frost, and G. A. C. Jones, Solid State
Electron. 32, 1179 (1989).
[333] L. Martin-Moreno and C. G. Smith, J. Phys. Condens.
Matter 1, 5421 (1989).
[334] B. J. van Wees, L. 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).
[335] M. B uttiker, Phys. Rev. B 41, 7906 (1990); L. I. Glaz-
man and A. V. Khaetskii, J. Phys. Condens. Matter 1,
5005 (1989); Y. Avishai and Y. B. Band, Phys. Rev. B
40, 3429 (1989); K. B. Efetov, J. Phys. Condens. Matter
1, 5535 (1989).
[336] J. F. Weisz and K.-F. Berggren, Phys. Rev. B 40, 1325
(1989).
[337] D. A. Wharam, U. Ekenberg, M. Pepper, D. G. Hasko,
H. Ahmed, J. E. F. Frost, D. A. Ritchie, D. C. Peacock,
and G. A. C. Jones, Phys. Rev. B 39, 6283 (1989).
[338] H. Hirai, S. Komiyama, S. Sasa and T. Fujii, Solid State
Comm. 72, 1033 (1989).
[339] S. Washburn, A. B. Fowler, H. Schmid, and D. Kern,
Phys. Rev. Lett. 61, 2801 (1988).
[340] R. J. Haug, A. H. MacDonald, P. Streda, and K. von
Klitzing, Phys. Rev. Lett. 61, 2797 (1988).
[341] R. J. Haug, J. Kucera, P. Streda, and K. von Klitzing,
Phys. Rev. B 39, 10892 (1989).
[342] B. R. Snell, P. H. Beton, P. C. Main, A. Neves, J. R.
Owers-Bradley, L. Eaves, M. Henini, O. H. Hughes, S. P.
Beaumont, and C. D. W. Wilkinson, J. Phys. Condens.
Matter 1, 7499 (1989).
[343] Y. S. Tsoi, Pisma Zh. Eksp. Teor. Fiz. 19, 114 (1974)
[JETP Lett. 19, 70 (1974)]; Zh. Eksp. Teor. Fiz. 68,
1849 (1975) [Sov. Phys. JETP 41, 927 (1975)].
[344] P. C. van Son, H. van Kempen, and P. Wyder, Phys.
Rev. Lett. 58, 1567 (1987).
[345] I. K. Yanson, Zh. Eksp. Teor. Fiz. 66, 1035 (1974) [Sov.
109
Phys. JETP 39, 506 (1974)].
[346] A. M. Duif, A. G. M. Jansen, and P. Wyder, J. Phys.
Condens. Matter 1, 3157 (1989).
[347] C. W. J. Beenakker, H. van Houten, and B. 1. van Wees,
Europhys. Lett. 7, 359 (1988).
[348] C. W. J. Beenakker, H. van Houten, and B. J. van Wees,
Festkorperprobleme 29, 299 (1989).
[349] J. Spector, H. L. Stormer, K. W. Baldwin, L. N. Pfeier,
and K. W. West, Surf Sci. 228, 283 (1990).
[350] U. Sivan, M. Heiblum, and C. P. Umbach, and H.
Shtrikman, Phys. Rev. B 41, 7937 (1990); J. Spector,
H. L. Stormer, K. W. Baldwin, L. N. Pfeier and K. W.
West, Appl. Phys. Lett. 56, 1290 (1990).
[351] P. A. M. Benistant, Ph.D. thesis, University of Ni-
jmegen, The Netherlands, 1984; P. A. M. Benistant, A.
P. van Gelder, H. van Kempen, and P. Wyder, Phys.
Rev. B 32, 3351 (1985).
[352] P. A. M. Benistant, G. F. A. van de Walle, H. van Kem-
pen, and P. Wyder, Phys. Rev. B 33, 690 (1986).
[353] L. Pfeier, K. W. West, H. L. Stormer, and K. W. Bald-
win, Appl. Phys. Lett. 55, 1888 (1989).
[354] C. T. Foxon, J. J. Harris, D. Hilton, J. Hewett, and C.
Roberts, Semicond. Sci. Technol. 4, 582 (1989); C. T.
Foxon and J. J. Harris, Philips J. Res. 41, 313 (1986).
[355] J. G. Williamson, H. van Houten, C. W. J. Beenakker,
M. E. J. Broekaart, L. J. A. Spendeler, B. J. van Wees,
and C. T. Foxon, Phys. Rev. B 41, 1207 (1990); Surf
Sci. 229, 303 (1990).
[356] R. Landauer, in Analogies in Optics and Microelectron-
ics (W. van Haeringen and D. Lenstra, eds.). Kluwer
Academic, Dordrecht, 1990.
[357] D. A. Wharam, 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, L887 (1988).
[358] H. U. Baranger and A. D. Stone, Phys. Rev. Lett. 63,
414 (1989); also in Ref.
16
.
[359] C. W. J. Beenakker and H. van Houten, Phys. Rev. Lett.
63, 1857 (1989).
[360] C. W. J. Beenakker and H. van Houten, in Ref.
17
.
[361] L. D. Landau and E. M. Lifshitz, Mechanics. Pergamon,
Oxford, 1976.
[362] N. S. Kapany, in Concepts of Classical Optics (J. Strong,
ed.). Freeman, San Francisco, 1958.
[363] H. de Raedt, N. Garcia, and J. J. Saenz, Phys. Rev.
Lett. 63, 2260 (1989); N. Garcia, J. J. Saenz, and H. de
Raedt, J. Phys. Condens. Matter 1, 9931 (1989).
[364] Y. Takagaki, K. Garno, S. Namba, S. Ishida, S. Takaoka,
K. Murase, K. Ishibashi, and Y. Aoyagi, Solid State
Comm. 68, 1051 (1988); 71, 809 (1989).
[365] Y. Hirayama and T. Saku, Solid State Comm. 73, 113
(1990); Phys. Rev. B 41, 2927 (1990).
[366] P. H. Beton, B. R. Snell, P. C. Main, A. Neves, J. R.
Owers-Bradley, L. Eaves, M. Henini, O. H. Hughes, S. P.
Beaumont, and C. D. W. Wilkinson, J. Phys. Condens.
Matter 1, 7505 (1989).
[367] E. Casta no and G. Kirczenow, Phys. Rev. B. 41, 5055
(1990). Y. Avishai, M. Kaveh, S. Shatz, and Y. B. Band,
J. Phys. Condens. Matter 1, 6907 (1989).
[368] C. G. Smith, M. Pepper, R. Newbury, H. Ahmed, D. G.
Hasko, D. C. Peacock, J. E. F. Frost, D. A. Ritchie, G.
A. C. Jones, and G. Hill, J. Phys. Condens. Matter 1,
6763 (1989).
[369] Y. Hirayama and T. Saku, Jap. J. Appl. Phys. (to be
published).
[370] M. B uttiker, Phys. Rev. B 33, 3020 (1986).
[371] M. B uttiker, IBM J. Res. Dev. 32, 63 (1988).
[372] A. A. M. Staring, L. W. Molenkamp, C. W. J. Beenak-
ker, L. P. Kouwenhoven, and C. T. Foxon, Phys. Rev.
B. 41, 8461 (1990).
[373] L. P. Kouwenhoven, B. J, van Wees, W, Kool, C. J. P,
M. Harmans, A, A. M. Staring, and C. T. Foxon, Phys.
Rev, B 40, 8083 (1989).
[374] P. C. Main, P, H. Beton, B. R. Snell, A. J. M. Neves,
J. R. Owers-Bradley, L. Eaves, S. P. Beaumont, and C.
D. W. Wilkinson, Phys. Rev. B 40, 10033 (1989).
[375] A. M. Chang, T. Y. Chang, and H. U. Baranger, Phys.
Rev. Lett. 63, 996 (1989).
[376] D. G. Ravenhall, H. W. Wyld, and R. L. Schult, Phys.
Rev. Lett. 62, 1780 (1989); R. L. Schult, H. W. Wyld,
and D. G. Ravenhall, Phys. Rev. B. 41, 12760 (1990).
[377] G. Kirczenow, Phys. Rev. Lett. 62, 2993 (1989); Phys.
Rev. B 42, 5375 (1990).
[378] H. Akera and T. Ando, Surf Sci. 229, 268 (1990).
[379] C. J. B. Ford, T. J. Thornton, R. Newbury, M. Pepper,
H. Ahmed, D. C. Peacock, D. A. Ritchie, J. E. F. Frost,
and G. A. C. Jones, Phys. Rev. B 38, 8518 (1988).
[380] Y. Avishai and Y. B. Band, Phys. Rev. Lett. 62, 2527
(1989).
[381] G. Kirczenow, Solid State Comm. 71, 469 (1989).
[382] F. M. Peeters, in Ref.
16
; Phys. Rev. Lett. 61, 589 (1988);
Superlattices and Microstructures 6, 217 (1989).
[383] M. L. Roukes, T. J. Thornton, A. Scherer, J. A. Sim-
mons, B. P. van der Gaag, and E. D. Beebe, in Ref.
16
.
[384] M. L. Roukes, A. Scherer, and B. P. van der Gaag, Phys.
Rev. Lett. 64, 1154 (1990).
[385] C. J. B. Ford. S. Washburn, M. B uttiker, C. M.
Knoedler, and J. M. Hong, Surf Sci. 229, 298 (1990).
[386] M. B uttiker, in Ref.
9
.
[387] R. Tsu and L. Esaki, Appl. Phys. Lett. 22, 562 (1973).
[388] L. L. Chang, L. Esaki, and R. Tsu, Appl. Phys. Lett.
24, 593 (1974).
[389] E. S. Alves, L. Eaves, M. Henini, O. H. Hughes, M. L.
Leadbeater, F. W. Sheard, and G. A. Toombs, Electron.
Lett. 24, 1190 (1988).
[390] A. Zaslavsky, V. J. Goldman, D. C. Tsui, and J. E.
Cunningham, Appl. Phys. Lett. 53, 1408 (1988).
[391] K. K. Likharev, IBM J. Res. Dev. 32, 144 (1988).
[392] K. Ng and P. A. Lee, Phys. Rev. Lett. 61, 1768 (1988).
[393] L. I. Glazman and K. A. Matveev, Pisma Zh. Eksp.
Teor. Fiz. 48, 403 (1988) [JETP Lett. 48, 445 (1988)].
[394] S. Washburn, A. B. Fowler, H. Schmid, and D. Kern,
Phys. Rev. B 38, 1554 (1988).
[395] J. H. Davies, Semicond. Sci. Technol. 3, 995 (1988). See
also in Ref.
72
.
[396] S. Y. Chou, D. R. Allee, R. F. W. Pease, and J. S.
Harris, Jr., Appl. Phys. Lett. 55, 176 (1989).
[397] A. Palevski, M. Heiblum, C. P. Umbach, C. M.
Knoedler, A. N. Broers, and R. H. Koch, Phys. Rev.
Lett. 62, 1776 (1989).
[398] Y. Avishai and Y. B. Band, Phys. Rev. B 41, 3253
(1990).
[399] C. G. Smith, M. Pepper, H. Ahmed, J. E. Frost, D.
G. Hasko, D. C. Peacock, D. A. Ritchie, and G. A. C.
Jones, Superlattices and Microstructures 5, 599 (1989).
[400] C. G. Smith, 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, L893 (1988).
[401] C. G. Smith, M. Pepper, H. Ahmed, J. E. F. Frost, D.
110
G. Hasko, D. A. Ritchie, and G. A. C. Jones, Surf Sci.
228, 387 (1990).
[402] S. J. Bending and M. R. Beasley, Phys. Rev. Lett. 55,
324 (1985).
[403] A. B. Fowler, G. L. Timp, J. J. Wainer, and R. A. Webb,
Phys. Rev. Lett. 57, 138 (1986).
[404] T. E. Kopley, P. L. McEuen and R. G. Wheeler, Phys.
Rev. Lett. 61, 1654 (1988); see also T. E. Kopley, Ph.D.
thesis, Yale University, 1989.
[405] W. Xue and P. A. Lee, Phys. Rev. B 38, 3913 (1988).
[406] V. Kalmeyer and R. B. Laughlin, Phys. Rev. B 35, 9805
(1987).
[407] C. S. Chu and R. S. Sorbello, Phys. Rev. B 40, 5941
(1989).
[408] J. Masek, P. Lipavsky, and B. Kramer, J. Phys. Con-
dens. Matter 1, 6395 (1989).
[409] P. L. McEuen, B. W. Alphenaar, R. G. Wheeler, and
R. N. Sacks, Surf Sci. 229, 312 (1990).
[410] G. J. Schulz, Rev. Mod. Phys. 45, 378 (1973).
[411] C. G. Smith, M. Pepper, J. E. Frost, D. G. Hasko, D.
C. Peacock, D. A. Ritchie, and G. A. C. Jones, J. Phys.
Condens. Matter 1, 9035 (1989).
[412] Y. Avishai, M. Kaveh, and Y. B. Band, preprint.
[413] F. Sols, M. Macucci, U. Ravioli, and K. Hess, Appl.
Phys. Lett. 54, 350 (1989).
[414] S. Datta, Superlattices and Microstructures 6, 83
(1989).
[415] D. S. Miller, R. K. Lake, S. Datta, M. S. Lundstrom,
and R. Reifenberger, in Ref.
15
.
[416] J. H. F. Scott-Thomas, S. B. Field, M. A. Kastner, H. I.
Smith, and D. A. Antoniadis, Phys. Rev. Lett. 62, 583
(1989).
[417] A. I. Larkin and P. A. Lee, Phys. Rev. B 17, 1596
(1978).
[418] P. A. Lee and T. M. Rice, Phys. Rev. B 19, 3970 (1979).
[419] H. van Houten and C. W. J. Beenakker, Phys. Rev. Lett.
63, 1893 (1989).
[420] K. Mullen, E. Ben-Jacob, R. C. Jaclevic, and Z. Schuss,
Phys. Rev. B 37, 98 (1988); M. Amman, K. Mullen, and
E. Ben-Jacob, J. Appl. Phys. 65, 339 (1989).
[421] M. A. Kastner, S. B. Field, U. Meirav, J. H. F. Scott-
Thomas, D. A. Antoniadis, and M. I. Smith, Phys. Rev.
Lett. 63, 1894 (1989).
[422] L. I. Glazman and R. I. Shekhter, J. Phys. Condens.
Matter 1, 5811 (1989).
[423] R. J. Brown, M. Pepper, H. Ahmed, D. G. Hasko, R.
A. Ritchie, J. E. F. Frost, D. C. Peacock, and G. A. C.
Jones, J. Phys. Condens. Matter, 2, 2105 (1990).
[424] L. P. Kouwenhoven, private communication; R. Haug,
private communication.
[425] U. Meirav, M. A. Kastner, and S. J. Wind, Phys. Rev.
Lett. 65, 771 (1990).
[426] 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).
[427] S. Komiyama, H. Hirai, S. Sasa, and S. Hiyamizu, Phys.
Rev. B 40, 12566 (1989).
[428] B. J. van Wees, E. M. M. Willems, L. P. Kouwenhoven,
C. J. P. M. Harmans, J. G. Williamson, C. T. Foxon,
and J. 1. Harris, Phys. Rev. B 39, 8066 (1989).
[429] B. W. Alphenaar, P. L. McEuen, R. G. Wheeler, and
R. N. Sacks, Phys. Rev. Lett. 64, 677 (1990).
[430] R. J. Haug and K. von Klitzing, Europhys. Lett. 10,
489 (1989).
[431] R. Kubo, S. J. Miyake, and N. Hashitsume, Solid State
Physics, Vol. 17 (F. Seitz and D. Turnbull, eds.). Aca-
demic, New York, 1965. M. Tsukada, J. Phys. Soc. Jap.,
41, 1466 (1976).
[432] R. F. Kazarinov and S. Luryi, Phys. Rev. B 25, 7626
(1982); S. Luryi and R. F. Kazarinov, Phys. Rev. B
27, 1386 (1983); S. Luryi, in High Magnetic Fields in
Semiconductor Physics (G. Landwehr, ed.). Springer,
Berlin, 1987.
[433] S. V. Iordansky, Solid State Comm. 43, 1 (1982).
[434] S. A. Trugman, Phys. Rev. B 27, 7539 (1983).
[435] R. Joynt and R. E. Prange, Phys. Rev. B 29, 3303
(1984).
[436] R. E. Prange, in Ref.
97
.
[437] L. I. Glazman and M. Jonson, J. Phys. Condens. Matter
1, 5547 (1989).
[438] L. I. Glazman and M. Jonson, Phys. Rev. B 41, 10686
(1990).
[439] T. Martin and S. Feng, Phys. Rev. Lett. 64, 1971 (1990).
[440] B. I. Halperin, Phys. Rev. B 25, 2185 (1982).
[441] A. H. MacDonald and P. Streda, Phys. Rev. B 29, 1616
(1984).
[442] S. M. Apenko and Yu. E. Lozovik, J. Phys. C 18, 1197
(1985).
[443] P. Streda, J. Kucera, and A. H. MacDonald, Phys. Rev.
Lett. 59, 1973 (1987).
[444] J. K. Jain and S. A. Kivelson, Phys. Rev. B 37, 4276
(1988).
[445] M. E. Cage, in Ref.
97
.
[446] A. H. MacDonald, T. M. Rice, and W. F. Brinkman,
Phys. Rev. B 28, 3648 (1983).
[447] O. Heinonen and P. L. Taylor, Phys. Rev. B 32, 633
(1985).
[448] D. J. Thouless, J. Phys. C 18, 6211 (1985).
[449] Y. M. Pudalov and S. G. Semenchinskii, Pisma Zh.
Eksp. Teor. Fiz. 42, 188 (1985) [JETP Lett. 42, 232
(1985)].
[450] W. Maass, Europhys. Lett. 2, 39 (1986).
[451] Y. Ono and T. Ohtsuki, Z. Phys. B 68, 445 (1987); T.
Ohtsuki and Y. Ono, J. Phys. Soc. Jap. 58, 2482 (1989).
[452] R. Johnston and L. Schweitzer, Z. Phys. B 70, 25
(1988).
[453] Y. Gudmundsson, R. R. Gerhardts, R. Johnston, and
L. Schweitzer, Z. Phys. B 70, 453 (1988).
[454] T. Ando, J. Phys. Soc. Jap. 58, 3711 (1989).
[455] P. C. van Son, G. H. Kruithof, and T. M. Klapwijk, Surf
Sci. 229, 57 (1990); P. C. van Son and T. M. Klapwijk,
Europhys. Lett. 12, 429 (1990).
[456] G. Ebert, K. von Klitzing, and G. Weimann, J. Phys.
C 18, L257 (1985).
[457] H. Z. Zheng, D. C. Tsui, and A. M. Chang, Phys. Rev.
B 32, 5506 (1985).
[458] E. K. Sichel, H. H. Sample, and J. P. Salerno, Phys.
Rev. B 32, 6975 (1987); E. K. Sichel, M. L. Knowles,
and H. H. Sample, J. Phys. C 19, 5695 (1986).
[459] R. Woltjer, R. Eppenga, J. Mooren, C. E. Timmering,
and J. P. Andre, Europhys. Lett. 2, 149 (1986).
[460] B. E. Kane, D. C. Tsui, and G. Weimann, Phys. Rev.
Lett. 59, 1353 (1987).
[461] M. B uttiker, in Ref.
15
.
[462] F. Kuchar, Festkorperprobleme 28, 45 (1988).
[463] P. F. Fontein, J. A. Kleinen, P. Hendriks, F. A. P. Blom,
J. H. Wolter, H. G. M. Locks, F. A. J. M. Driessen, L.
111
J. Giling, and C. W. J. Beenakker, submitted to Phys.
Rev. B.
[464] C. W. J. Beenakker, unpublished.
[465] L. P. Kouwenhoven, Masters thesis, Delft University of
Technology, 1988.
[466] K. yon Klitzing, G. Ebert, N. Kleinmichel, H. Obloh, G.
Dorda, and G. Weimann, Proc. ICPS 17 (J. D. Chadi
and W. A. Harrison, eds.). Springer, New York, 1985.
[467] D. A. Syphers, F. F. Fang, and P. J. Stiles, Surf Sci. 142,
208 (1984); F. F. Fang and P. J. Stiles, Phys. Rev. B
27, 6487 (1983); F. F. Fang and P. J. Stiles, Phys. Rev.
B 29, 3749 (1984). A. B. Berkut, Yu. V. Dubrovskii,
M. S. Nunuparov, M. I. Reznikov, and V. I. Talyanski,
Pisma Zh. Teor. Fiz. 44, 252 (1986) [JETP Lett. 44,
324 (1986)].
[468] D. A. Syphers and P. J. Stiles, Phys. Rev. B 32, 6620
(1985).
[469] Y. Zhu, J. Shi, and S. Feng, Phys. Rev. B 41, 8509
(1990).
[470] H. Hirai, S. Komiyama, S. Hiyamizu, and S. Sasa, in
Proc. ICPS 19, p. 55 (W. Zawadaski, ed.). Institute of
Physics, Polish Academy of Sciences, 1988.
[471] S. Komiyama, H. Hirai, S. Sasa, and T. Fuji, Solid State
Comm. 73, 91 (1990); H. Hirai, S. Komiyama, S. Sasa,
and T. Fujii, J. Phys. Soc. Japan 58, 4086 (1989).
[472] S. Komiyama and H. Hirai, Phys. Rev. B 40, 7767
(1989).
[473] U. Sivan, Y. Imry, and C. Hartzstein, Phys. Rev. B 39,
1242 (1989).
[474] U. Sivan and Y. Imry, Phys. Rev. Lett. 61, 1001 (1988).
[475] M. B uttiker, Phys. Rev. B 38, 12724 (1988).
[476] J. K. Jain, unpublished.
[477] P. L. McEuen, A. Szafer, C. A. Richter, B. W.
Alphenaar, J. K. Jain, and R. N. Sacks, Phys. Rev.
Lett. 64, 2062 (1990).
[478] D. C. Tsui, H. L. Stormer, and A. C. Gossard, Phys.
Rev. Lett. 48, 1559 (1982).
[479] R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983).
[480] A. M. Chang and J. E. Cunningham, Solid State Comm.
72, 651 (1989); Surf Sci. 229, 216 (1990).
[481] 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); and unpub-
lished.
[482] C. W. J. Beenakker, Phys. Rev. Lett. 64, 216 (1990).
[483] A. H. MacDonald, Phys. Rev. Lett. 64, 220 (1990).
[484] T. Chakraborty and P. Pietilainen, The Fractional
Quantum Hall Eect. Springer, Berlin, 1988.
[485] R. B. Laughlin, Phys. Rev. B 27, 3383 (1983).
[486] F. D. M. Haldane, Phys. Rev. Lett. 51, 605 (1983).
[487] B. I. Halperin, Phys. Rev. Lett. 52, 1583 (1984).
[488] R. B. Laughlin, Phys. Rev. B 23, 5632 (1981).
[489] B. I. Halperin, Helv. Phys. Acta 56, 75 (1983).
[490] A. M. Chang, Solid State Comm. 74, 871 (1990).
[491] G. Timp, R. E. Behringer, J. E. Cunningham, and R.
E. Howard, Phys. Rev. Lett. 63, 2268 (1989); G. Timp,
in Ref.
9
.
[492] S. T. Chui, Phys. Rev. Lett. 56, 2395 (1986); Phys. Rev.
B 36, 2806 (1987).
[493] H. W. Jiang, H. L. Stormer, D. C. Tsui, L. N. Pfeier,
and K. W. West, Phys. Rev. B 40, 12013 (1989).
[494] R. G. Clark, J. R. Mallett, S. R. Haynes, J. J. Harris,
and C. T. Foxon, Phys. Rev. Lett. 60, 1747 (1988).
[495] S. A. Kivelson and V. L. Pokrovsky, Phys. Rev. B 40,
1373 (1989).
[496] J. A. Simmons, H. P. Wei, L. W. Engel, D. C. Tsui, and
M. Shayegan, Phys. Rev. Lett. 63, 1731 (1989).
[497] G. Timp, P. M. Mankiewich, P. DeVegvar, R. Behringer,
J. E. Cunningham, R. E. Howard, H. U. Baranger, and
J. K. Jain, Phys. Rev. B 39, 6227 (1989).
[498] J. K. Jain, Phys. Rev. Lett. 60, 2074 (1988).
[499] J. K. Jain and S. Kivelson, Phys. Rev. B 37, 4111
(1988).
[500] B. J. van Wees, L. P. Kouwenhoven, C. J. P. M. Har-
mans, J. G. Williamson, C. E. T. Timmering, M. E. I.
Broekaart, C. T. Foxon, and J. J. Harris, Phys. Rev.
Lett. 62, 2523 (1989).
[501] E. N. Bogachek and G. A. Gogadze, Zh. Eksp. Teor.
Fiz. 63, 1839 (1972) [Sov. Phys. JETP 36, 973 (1973)].
[502] N. B. Brandt, D. V. Gitsu, A. A. Nikolaevna, and Va.
G. Ponomarev, Zh. Eksp. Teor. Fiz. 72, 2332 (1977)
[Sov. Phys. JETP 45, 1226 (1977)]; N. B. Brandt, D.
B. Gitsu, V. A. Dolma, and Va. G. Ponomarev, Zh.
Eksp. Teor. Fiz. 92, 913 (1987) [Sov. Phys. JETP 65,
515 (1987)].
[503] Y. Isawa, Surf Sci. 170, 38 (1986).
[504] D. A. Wharam, M. Pepper, R. Newbury, H. Ahmed, D.
G. Hasko, D. C. Peacock, J. E. F. Frost, D. A. Ritchie,
and G. A. C. Jones, J. Phys. Condens. Matter 1, 3369
(1989).
[505] R. J. Brown, C. G. Smith, M. Pepper, M. J. Kelly, R.
Newbury, H. Ahmed, D. G. Hasko, J. E. F. Frost, D.
C. Peacock, D. A. Ritchie, and G. A. C. Jones, J. Phys.
Condens. Matter 1, 6291 (1989).
[506] Y. Yosephin and M. Kaveh, J. Phys. Condens. Matter
1, 10207 (1989).
[507] R. Mottahedeh, M. Pepper, R. Newbury, J. A. A. J.
Perenboom, and K.-F. Berggren, Solid State Comm. 72,
1065 (1989).
[508] L. D. Landau and E. M. Lifshitz, Statistical Physics,
Part 2, Section 55. Pergamon, Oxford, 1980.
[509] F. Bloch, Z. Phys. 52, 555 (1928).
[510] J. N. Churchill and F. E. Holmstrom, Phys. Lett. A 85,
453 (1981).
[511] J. N. Churchill and F. E. Holmstrom, Am. J. Phys. 50,
848 (1982).
[512] J. Zak, Phys. Rev. B 38, 6322 (1988).
[513] J. B. Krieger and G. J. Iafrate, Phys. Rev. B 38, 6324
(1988).