I.

ELECTRONS AS SCATTERED WAVES

A. Introduction

This lecture introduces scattering approach to quantum transport. The lecture consists of four parts. The first
part is just refreshing quantum mechanics. We recall that plane waves present the states of quantum particles.
We recall that electrons are fermions that obey Fermi statistics. This defines the filling of the states and allows to
calculate electron density and current. To become prepared for quantum transport, we concentrate on one-dimensional
waveguides intercepted by potential barriers.
Second part makes connection between these waveguides and general nanostructures. We access this connection in
two steps. First, we study the quantum point contact: the most important example system of quantum transport.
Second, we discuss extra scattering and introduce the scattering matrix. We derive Landauer formula that relates the
conductance and the sum of transmission coefficients.

Electrons as scattered waves

• Building a Landauer conductor

• Electrons as plane waves
• Scattering matrix
• Electrons as fermions
• Landauer formula
• Waveguide
• Restrictions and limitations
• Potential barrier
• Counting electrons
• Nanostructure vs waveguide
• Characteristic function
• Adiabatic Quantum transport
• Limiting cases
• Quantum Point Contact
• Levitov formula
• Current in a QPC
• Electrons gambling:
• Experiment
• Transmission distribution
• Types of quantum contacts

Third part explains statistics of electrons transfers via a nanostructure, those are given by Levitov formula. It also
illustrates that all transmission coefficients are important and not just their sum.
Forth part introduces quantum pin-code: a set of transmission coefficients characterizing a nanostructure and
illustrates possible distributions of those.

B. Wave properties of electrons

Quantum mechanics teaches us that each particle is a wave. Wave properties of macroscopic particles such as
brickstones, sand grains, and even DNA molecules are hardly noticeable to us: We deal with them at a space scale
much bigger than their wavelength. Electrons are remarkable exception. Their wavelength is a fraction of nanometer
in metals and can reach a fraction of micrometer in semiconductors. We can not ignore wave properties of electrons
in nanostructures of this size. This is the central issue in quantum transport, and we start the book with a short
summary of elementary results concerning electron waves.
A quantum electron is characterized by its wave function Ψ(r, t). The squared absolute value |Ψ(r, t)| 2 gives the
propability to find the electron in a given point r at time t. Quantum states available for an electron in vacuum are
those with a certain wavevector k. The wave function of this state is a plane wave
1
Ψk (r, t) = √ exp (ik · r − iE(k)t/h̄) , (1)
V

E(k) = h̄2 k2 /2m being the corresponding energy. The electron in this state is spread over the whole space of a very
big volume V: the squared absolute value of Ψ does not depend on coordinates. The prefactor in Eq. (1) assures that
there is precisely one electron in this big volume.

1
 Electrons as plane waves Electrons as fermions

Electron in vacuum is a plane wave Pauli principle: Number of

Fraction of
dk x dk y dk z
ψ k
(r , t ) - wavefunction filled states in



1

A state is either states in a small 2 s V
ψ k
(r , t ) = exp(ikr − iE (k )t / ) 2 filled or empty cube near k (2π )3 the cube: filling
ψ k
(r , t ) - probability factor f



 1 
V
ρ 
d 3k 

energy density =  Ε  = ∫ 2 s
V - norm. volume density

(2π )3 




E (k ) f (k )
current density  j 


k - wavevector
 ev (k ) 





e p = k  - momentum

q.mech. does not set f, statistics does µ
1 ( k ) 2
E= - energy 1 E
Ψ
f eq (k ) = f F ( E (k ) − µ )) =



Im0 2m E−µ
1 + exp
,
Ψ
e
R
-1-3 k BT
-2 -1 0 1 2 3 Zero T: sharp step 0 0 1
kx/2π |k| kF f

There are many electrons in nanostructures. Electrons are spin 1/2 fermions, Pauli principle assures that each
one-particle state is either empty or filed with one fermion. Let us consider a cube in k-space centred around k with
the sides dkx , dky , dkz  |k|. The number of available states in this cube is 2s Vdkx dky dkz /(2π)3 . The factor of 2s
comes from the fact that there are two possible spin directions. The fraction of the states filled in this cube is called
an electron filling factor f (k). The particle density ρ, energy density E and density of electric current j are contributed
by all electrons and read
   
ρ 3 1
d
Z
 E  = 2s k  E(k)  f (k).
3
(2)
j (2π) ev(k)

Here we introduce the electron charge e and the velocity v(k) = h̄k/m. Quantum mechanics puts no restriction on
f (k). However, the filling factor of electrons in equilibrium state at a given electrochemical potential µ and temperature
T is set by Fermi-Dirac statistics,
1
feq (k) = fF (E(k) − µ) ≡ . (3)
1 + exp((E − µ)/kB T )

Next, we consider electrons in the field of electrostatic potential

U (r, t)/e. The wave function Ψ(r, t) of an electron is not a plane wave anymore. Instead, it obeys the time-dependent
Schrödinder equation,

∂Ψ(r, t) h̄2 2
ih̄ = ĤΨ(r, t); Ĥ ≡ − ∇ + U (r, t). (4)
∂t 2m

This is an evolutionary equation: it determines Ψ in the future given its instant value. The evolution operator Ĥ is
called Hamiltonian. For the time being, we concentrate on stationary potential, U (r, t) ≡ U (r). The wave functions
become stationary, their time dependence being given by the energy,

Ψ(r, t) = exp(−iEt/h̄)ψE (r).

The Schrödinder equation reduces to

h̄2 2
 
EψE (r) = ĤψE (r) = − ∇ + U (r) ψE (r). (5)
2m

The Hamiltonian becomes the operator of energy, while the equation becomes a linear algebra relation defining the
eigenvalues E and the corresponding eigenfunctions ΨE of this operator. These eigenfunctions form a basis in the
Hilbert space of all possible wave functions, so that an arbitrary wave function can be expanded, or represented,
in this basis. The first (gradient) term in the Hamiltonian describes the kinetic energy, and the second term U (r)
represents the potential energy.

2
 A good part of quantum mechanics deals with the above equation. It can not be readily solved for an arbitrary
potential, and our qualitative understanding of quantum mechanics is built upon several simple cases when this
solution can be obtained explicitly. Following to many good textbooks, we will concentrate on one-dimensional
motion where the potential and wave functions depend on a single coordinate x. However, we pause a bit to introduce
a key concept that makes this one-dimensional motion more physical.

C. Waveguide

Let us confine electrons in an infitinely long tube along x-axis that has a rectangular cross-section. We can do this
by setting the potential U to zero for |y| < a/2, |z| < b/2 and to +∞ otherwise. We thus made unpenetrable walls for
the electon that are perpendicular to y and z axes. We expect that a wave will be reflected from these walls changing
sign of the corresponding component of the wavevector, ky → −ky or kz → −kz . This suggests that the solution of
the Schrödinder equation is a superposition of incident and reflected waves of the kind
X
ψ(x, y, z) = exp(ikx x) Csy sz exp(sy iky y) exp(sz ikz z). (6)
sy ,sz =+,−

Waveguide

1d motion: y,z – restricted: mode index n, x- free: kx

ψ k , n ( x, y , z ) =
12
3,1 2,2
x I IV 10

Φ n ( y, z ) exp(ik x x)
b
1,2
a 8
y

x z 2ma2 6 2,1

π
E4
2 2

Standing Plane
II y
1,1

2
wave wave y
III 0

πkx/a
-1 0 1 2
X

π 2
2 ny nz2
2
(
k x ) 2
En (k x ) = + En ; En = ( 2 + 2)
2m 2m a b
Dictionary: mode = transport channel

Since the infinite potential repels the electron efficiently, the wave function must vanish at the walls, ψ(x, y =
±a/2, z) = ψ(x, y, z = ±b/2) = 0. This gives a linear relation between C sy sz that determines these superposition
coefficients. To put it simply, the walls have to be in the nodes of a standing wave in both y and z directions. This
can only happen if ky,z assume quantized values kyn = πny /a, kzn = πnz /b, with integer ny , nz > 0 corresponding to
the number of half-wavelengths that fit between the walls. Here we introduce a compound index n = (n y , nz ), the
notation we use throughout the book. The wave function reads

ψkx ,n (x, y, z) = ψkx (x)Φn (y, z);

ψkx (x) = exp(ikx x), (7)
2
Φn (y, z) = √ sin(kyn (y − a/2)) sin(kzn (z − b/2)).
ab
The transversal motion of the electron is thus quantized. The electron in a state with the given n (these states are
called modes in wave theory and transport channels in nanophysics) has only one degree of freedom corresponding
to one-dimensional motion. The energy spectrum consists of one-dimensional branches shifted by channel-dependent
energy En (Fig.),
!
(h̄kx )2 π 2 h̄2 n2y n2z
En (kx ) = + E n ; En = + 2 . (8)
2m 2m a2 b

3
 Potential barrier. Let us add some more design to our waveguide. We cross it with a potential barrier of a simple
form

U0 , 0 < x < d
U (x) = . (9)
0, otherwise
The possible solutions outside the barrier for a given n and p
energy are plane waves of the form (7). It is important to
note that there are two possible solutions with kx = ±k = ± 2m(E − En )/h̄, corresponding to the waves propagating
to the right (positive sign) or to the left.

Potential barrier: transmission

Potential barrier
probability
Result for a rectangular barrier
Intersect the waveguide with a potential barrier U(x) 1
T (E) ≡ t =
2

 sin κ d (k 2 − κ 2 ) 
1
(
κ ) 2
2
E (
k ) 2 1+  
= E; = E −U0  
thick 0.8 2kκ
1 t
2m 2m
0.6
r U0
0.4
4 unknown variables: thin threshold classical mechanics :
T ( E ) = 0 or 1
A,B,r,t 0.2
0 d
x 4 equations: 00 0.5 1 1.5 2 2.5 3
ψ ( x) = exp(ikx) E/U0
+ r exp(−ikx )
k
ψ ( x) = t exp(ikx) continuity of wavefunction and its
ψ ( x) = A exp(iκ x) derivative at two boundaries
+ B exp(−iκ x) tunneling
Dictionary: reflection and transmission amplitudes Dictionary: transmission probability, coefficient

A wave sent from the left is scattered at the barrier, part of it being reflected back, another part being transmitted.
We have
 
 exp(ikx) + r exp(−ikx), x<0
ψ(x) = B exp(iκx) + C exp(−iκx), 0 < x < d , (10)
t exp(ikx), x>d
 
q
where κ = 2m(E − En − U0 )/h̄ = k 2 − 2mU0 /h̄2 . The wavefunction and its x-derivative must be continuous at
p

x = 0 and x = d. These four conditions give four linear equations to find the unknown coefficients r, B, C, t. The
most important for us are the transmission amplitude t and the reflection amplitude r. The transmission coefficient
T (E) = |t|2 determines which fraction of the wave is transmitted through the obstacle. The reflection coefficient
R(E) = |r|2 = 1 − T (E) determines the fraction reflected back. We find
4k 2 κ2
T (E) = . (11)
(k 2 − sin2 κd + 4k 2 κ2
κ 2 )2
In classical physics, particles with energies below the barrier (E < U0 ) would be totally reflected (T = 0), while
particles with energies above the barrier would be fully transmitted (T = 1). Quantum mechanics change this:
Electrons are transmitted and reflected at any energy (Fig.??). Even an electron with the energy well below the
barrier (corresponding
p to imaginary κ) has a finite, albeit an exponentially small chance to be transmitted, T (E) ∝
exp(−2d 2m(U0 + En − E)/h̄)  1. This is called tunneling.
The above consideration is not limited to barriers localized within a certian interval of x. For any barrier, the
solution very far to the left, x → −∞, can be regarded as a superposition of incoming and reflected waves ψ =
exp(ikx) + r exp(−ikx). Very far to the right, x → ∞, it is a transmitted wave, ψ = t exp(ikx). To actually calculate
t and r, we have to solve the Schrödinger equation everywhere, and to match these two asymptotic solutions.

D. Quantum contacts

A common nanostructure does not even remotely resemble an infinitely long waveguide. However, the physics of
quantum transport is surprisingly similar to that of a waveguide. The recognition of this fact and its experimental

4
verification was and still is one of the main events in the history of the field. We introduce this important idea in two
steps. In this Section, we are considering in detail the quantum point contact (QPC) — a system without potential
barriers — and show that it is equivalent to a waveguide with a potential barrier. In the next Section, we turn to a
more complicated case of a generic nanostructure.
We start by looking at a waveguide of a variable cross-section. The waveguide is extended along the x-axis, bounded
by inpenetrable potential walls, and has a rectangular cross-section, |y| < a(x)/2, |z| < b(x)/2, with the dimensions
varying as one moves along the contact. Far to the right and the left, x → ±∞, these dimensions assume constant
values a∞ and b∞ . In the middle, the walls come closer forming a constriction (Figure). The solutions (7) found for
the ideal waveguide do not apply to this case, and solving the Schrödinger equation is cumbersome: The variables in
three-dimensional Schrödinger equation do not separate and the motion does not become one-dimensional.
However, we get a general understanding of the quantum waves in the system by looking at an adiabatic waveguide.
Its dimensions are assumed to vary smoothly so that the length scale at which they change is much longer than the
dimensions themselves: |a0 (x)|, |b0 (x)|  1, a(x)|a00 (x)|, b(x)|b00 (x)|  1. Under these conditions, the walls are locally
flat and parallel, so that locally the wave functions can be approximated by those of the ideal waveguide (Eq. (7)).
The variables are locally separated, so that

ψn (x, y, z) = ψ(x)Φn (a(x), b(x), y, z), (12)

where the transverse wave functions Φ(a, b, y, z) are given by Eq. (7). The function ψ(x) corresponds to one-
dimensional motion and satisfies
h̄2 ∂ 2
 
− + E n (x) ψ(x) = Eψ(x). (13)
2m ∂x2

Here, En presents a channel-dependent energy introduced by Eq. (8). Now, this energy depends on x via the
waveguide dimensions a(x), b(x),
!
π 2 h̄2 n2y n2z
En (x) = + . (14)
2m a2 (x) b2 (x)

We note that this term plays the role of potential energy for
one-dimensional motion. Strangely enough, this potential energy depends on the channel index. Let us plot these en-
ergies versus x (Figure). For each channel we see a potential barrier forming in the narrowest part of the constriction.
The bigger the numbers ny , nz are, the higher the barrier is.

Nanostructure versus waveguide Adiabatic Quantum Transport

Closed
λ Channels
Constriction as a potential barrier (T=0)
Nanostructure: can be very complex
a = const(x)
Can be modelled as: waveguide with transport channels
E
+ potential barrier scattering
Essence: set of transmissions Tn region
En(x)
Open
a(x)
Enough to describe the transport! ir
o
Channels
rve y (T=1)
se
r x
To see this: consider X

π 2
2  
Adiabatic Quantum Transport,
 
ideal n y2 n z2
En (x) = +
 a (x) 
waveguides
Quantum Point Contact 2m 2
b2(x)

Let us concentrate on a given energy E. In a given channel, we compare it with the maximal barrier height assuming
inpenetrable barrier. If E exceeds the height, the electrons coming to the constriction traverse it. Otherwise, they
are reflected back. Since the barrier height increases with the channel index, there is only a finite number of open
channels where electrons can pass the constriction. All other channels are closed.

5
 Therefore, the adiabatic waveguide of a variable cross-section without a potential barrier appears to be the same
as an ideal waveguide with a potential barrier, that we have considered in the previous Section. For each channel,
we define transmission and reflection amplitudes in a usual way and end up with the channel-dependent transmission
coefficient Tn (E). It appears that the adiabaticity also implies almost classical potential barrier, so that T = 1 for
open channels and T = 0 for closed ones. The only exception is a narrow energy interval where the energy almost
aligns with the top of the barrier. Remarkably, the electrons in the closed channels are almost perfectly reflected in
spite of the absence of potential barriers in the system.
Now we are ready to turn to our core business: The quantum transport. We determine the electric current in
the constriction. The first step is to adapt the expression for the current given by Eq. (2) to the case of quantized
transverse motion. We do this by replacing the integration over ky and kz by the summation over their discrete
quantized values kyn and kzn ,

dkx dky dkz dkx

Z Z X
(. . .) → ab (. . .) . (15)
2π 2π 2π 2π n

This procedure describes an ideal waveguide, in which case kx stands for the wavevector. In our case, the waveguide
is not ideal, and the wavevector depends on x. However, we have chosen the shape in such a way that for x → ±∞
the waveguide is ideal, and we can use Eq. (15) to evaluate the full current I via the cross-section located infinitely
far to the left from the constriction. Note that we are free to choose the cross-section in an arbitrary way since the
charge conservation law implies that the stationary full current flowing through any cross-section is the same. We get
the full current by multiplying the current density jx with the cross-section area a∞ b∞ , thus absorbing the factors in
Eq. (15),
X Z ∞ dkx
I = 2s e vx (kx )fn (kx ), (16)
n −∞ 2π

the velocity being vx = h̄kx /m.

Current in a QPC
closed open closed

iro ir
rve
vor
es
N att (t ) =
Filling se er
r 2s eVt
= N open
t
factors are fte ghi
brought
l r 2π

from the
reservoirs

E Quantized
µL conductance
µR
0 f 1 0 f 1

∑ dE [ f L ( E ) − f R ( E )] = 2πs
N open ( µ L − µ R ) = GQ N openV

2π
n:open ∫
2s e 2 e
I=

Let us concentrate on filling factors fn (kx ). They are different for open (T = 1) and closed (T = 0) channels
(Fig.). If the channel is closed, all electrons passing the cross-section from the left are reflected from the barrier and
subsequently pass the same cross-section from the right. Therefore, in a closed channel there is the same amount of
right- and left-going electrons, filling factors are the same for two momentum directions, f n (kx ) = fn (−kx ). Since
their velocities are opposite, the contribution of closed channels to the net current vanishes. Thus, we concentrate on
open channels.
For open channels, the filling factors for the two momentum directions are not the same. To realize this fact, we
have to understand how the electrons get to the waveguide. It drives us to a concept of a reservoir. Any nanostructure
taking part in quantum transport is a part of an electric circuit. This means that it is connected to large, macroscopic
electric pads each kept at a certain voltage (electrochemical potential). These pads contain a big number of electrons
which are at thermal equilibrium: They are characterized by the filling factor (3) which only depends on the energy

6
and the chemical potential of the corresponding reservoir. In our setup, the waveguide is connected to two such
reservoirs: left (x → −∞) and right (x → ∞). Electrons with kx > 0 come from the left reservoir and have the
filling factor fL (E) ≡ fF (E − µL ). Electrons with kx < 0 come to the cross-section having passed the constriction.
Therefore, they carry the filling factor of the right reservoir, fR (E) ≡ fF (E − µR ).
Since the filling factors only depend on the energy, it is natural to replace kx in favour of the total energy E for
each momentum direction. Since the velocity is vx = ∂E/∂kx , we have dE = h̄v(kx )dkx , and this cancels the velocity
in Eq. (16). Thus, we end up with the remarkably simple expression,

2s e X
Z
I= dE [fL (E) − fR (E)]
2πh̄ n:open
2s e
≡ Nopen (µL − µR ) ≡ GQ Nopen V. (17)
2πh̄
We have integrated over energy coming with a factor µL − µR . This factor corresponds to the voltage difference
applied, V = (µL − µR )/e. The voltage drives the current, there is no current at V = 0 since that corresponds
to the state of thermodynamic equilibrium. The factor is the same for all open channels. Therefore the current is
proportional to the number of open channels Nopen . The proportionality coefficient is called the conductance quantum
GQ = 2s e2 /(2πh̄). Here we follow the conventional defintion of GQ . Eventually, it could be more logical to incorporate
the number of spin directions 2s into the number of open channels, so that one has two transport channels for each n.
The conductance of the system, I/V , appears to be quantized in units of GQ . This factor is made up from fundamental
constants, not depending on material properties, nanostructure size and geometry, or concrete theoretical model used
to evaluate the transport properties.
Eq. (17) is a specific case of the celebrated Landauer formula. We have derived the formula for the case when T (E)
can be either zero or one. The general case will be treated in no time.
Let us remark that we have obtained this very general relation in the framework of a specific (and hardly realistic)
model of a constricted adiabatic waveguide with inpenetrable walls and rectangular cross-section. Now we discuss
why this result holds in a far more general setup. First, let us get rid of the assumption of inpenetrable walls and
rectangular cross-section, and introduce a waveguide with an arbitrary confining potential U x (y, z). Provided the
waveguide is adiabatic, we still can separate the variables in the Schrödinger equation and write the solution in the
form (12), where the transverse wave functions now obey the equation

h̄2 ∂2 ∂2
   
− + + Ux (y, z) Φn (x; y, z) = En (x)Φn (x; y, z), (18)
2m ∂y 2 ∂z 2

where the discrete index n labels the transverse states, En (x) being the chanhel-dependent potential energy (cf Eq.
(14)). This is the only change as compared with the previous model.
Next, let us note that the number of open channels, and, consequently, the conductance of our system, are deter-
mined only by the narrowest part of the waveguide. Therefore we can change the shape of the waveguide without
changing its transport properties provided the narrowest cross-section stays the same. Let us see what happens if
we change it in a way shown in Figure sending a∞ , b∞ to infinity. The structure we end up with — quantum point
contact (QPC) — is not at all a waveguide. In particular, in a waveguide we have a finite number of channels at
each energy and the spectrum consists of discrete energy branches. In contrast to this, the number of transport
channels approaching QPC is infinite, and energy spectrum is continuous. Of all these channels, only a finite number
is transmitted through the constriction.

7
 Building a Landauer conductor

Adiabatic Quantum Transport

Quantum Point
Contact

II

III QPC with

Real life
scattering

IV

Experiment. Quantization of conductance was first observed in GaAs-AlGaAs semiconductor heterostructures in

Delft in 1988. In these heterostructures, the electrons are confined near the surface forming a 2D electron gas
(2DEG). In terms of our waveguide, this means that one of the dimensions b → 0. Then the lowest subband (n z = 1)
is only relevant. In addition, two gate electrodes were imposed on the top of the heterostructure. These electrodes
were electrically isolated from the 2DEG. Still, they were used to shape this 2DEG: The potential applied to these
electrodes repels the electrons from them , creating inpenetrable walls around. The constriction is formed by the
walls in the gap between the electrodes, the width corresponding to the dimension a of our model waveguide. More
negative voltage repels stronger, shifting the walls outwards and therefore making the constriction narrower. The
minimum width amin is thus controlled by the gate voltage.
The number of open channels, in its turn, is determined by the minimum width. Let us write down this dependence
explicitly for a model that disregards the potential inside the 2DEG and assumes infinite potential outside (we leave
more realistic model for advanced part of this course). A new channel with the index n = (n y , 1) opens when the
energy position of the top of the barrier, Wn , passes the Fermi energy as we change amin ,
h̄2 π 2 2 h̄2 kF2
Wn ≡ 2 ny = E F = , (19)
2amin m 2m
and thus
Nopen = [kF amin /π] , (20)
where the brackets denote the integer part. Thus, by varying the gate voltage, one changes the number of open
channels. Therefore, the gate voltage dependence of the conductance is expected to look as a stair with a step height
GQ . This is what was measured in the experiment of 1988.

Experiment

Van Wees et al. 1988 –experimental evidence of conductance quantization

2D
electron
gas

Infinitely many transport

channels –
Few open channels

Quantization not ideal

8
 That time such quantization was expected for an ideal waveguide, and it was a complete surprise to observe
it in a relatively short constriction in a system far from being ideal. We cite : We propose an explanation of
the observed quantization of the conductance, based on the assumption of quantized transverse momentum in the
contact constriction. In principle this assumption requires a constriction much longer than wide, but presumably the
quantization is conserved in the short and narrow constriction of the experiment. This quote, written a year before a
theoretical understanding of quantum point contact was achieved, shows the essence of this Section: discrete transport
channels do not need waveguides to persist.

E. Scattering matrix

In two previous Sections, we studied electron transport in idealized waveguides with or without potential barrier.
They do not only illustrate concepts of quantum transport, they also model concrete experimental situations. A
waveguide with no potential barrier models a QPC, a constriction created by gates in 2DEG. A waveguide with
potential barrier models electron propagation through an insulating layer between two metals.
Real nanostructures can be made in a variety of ways, and can be more complicated. Modern fabrication technol-
ogy allows for making sophisticated semiconductor heterostructures, combining and shaping different metals, using
nanotubes, molecules and even single atoms as elements of an electron transport circuit. Various means can be used
to control the transport properties of a fabricated nanostructure. It is only possible to describe all this in a single
book because all these systems do obey the general laws of quantum transport that we formulate in this subsection.
There is a common feature of all fabrication methods: Two nanostructures that are intended to be identical, that
is, are made with the same design and technology, are never identical. Beside the artificial features brought by design,
there is also disorder originating from defects of different kind inevitably present in the structure. The position of
and/or potential created by such defect is random and in most cases can be neither controlled nor measured. It is
unlikely that this situation changes with further technological developments: Even if one achieves a perfect control of
every atom in a nanostructure, one would not be able to control all the atoms in the macroscopic contact leads, those
can not be separated from the nanostructure. The defects scatter electrons thus affecting the transport properties.
Conductance of the structure is thus random depending on a specific realization of disorder in the structure and in
the leads: Formidal number of uncontrollable parameters.
Fortunately, transport properties of any nanostructure can be expressed through a smaller set of parameters. The
condition for this is that electrons traverse the structure without elergy loss, so that they experience only elastic
scattering. These conditions for a given structure are always achieved at sufficiently low temperature and voltage
applied. The scattering is characterized by a scattering matrix that contains information about electron wavefunctions
far from the structure. The transport is described by a set of transmission eigenvalues derived from this scattering
matrix. A great deal of literature on quantum transport, and a great deal of this book, is in fact devoted to evaluation
of the transmission eigenvalues and establishing their general properties. In this Section, we derive the relation between
conductance and transmission eigenvalues and thus demonstrate that understanding of transmission properties of a
system authomatically means understanding of its transport properties.
We have mentioned in Section I D that any nanostructure taking part in quantum transport is a part of an electric
circuit: It is connected to several reservoirs which are in thermal equilibrium and are characterized by a fixed voltage.
In this Section, we only consider the case when there are two reservoirs (to be referred as left and right). Generalization
to many reservoirs is given in next lecture.
Somewhere between the reservoirs there is scattering region – the nanostructure proper. Let us start from the
assumption that we use formulating a model for QPC: Ideal waveguides connect the reservoirs and the scattering
region (Fig.). This is convenient since the scattering only takes place in a finite region, the reservoirs being far from
this region. The wave functions may have very complicated form in the scattering region, but in the waveguides they
are always combinations of plane waves. Left and right waveguides do not have to have the same axis and the same
cross-section. This is why it is convenient to introduce the separate coordinates x L < 0, yL , zL and xR > 0, yR , zR for
the left and right waveguides, respectively. Generally, a wavefunction at fixed energy E can be presented as a linear
combination of the plane waves,
X 1 h (n) (n)
i
ψ(xL , yL , zL ) = √ Φn (yL , zL ) aLn eikx xL + bLn e−ikx xL , (21)
n
2πh̄vn

and

9
 X 1 h (m) (m)
i
ψ(xR , yR , zR ) = √ Φm (yR , zR ) aRm e−ikx xR + bRm eikx xR . (22)
m
2πh̄vm

Here we label the transport channels in the left and right waveguides by the indices n and m, respectively. The
corresponding transverse wavefunctions are Φn and Φm , and energies of the transverse motion are En , Em . For any
transport
p channel n or m, be it in the left or in the right waveguide, the energy E fixes(n)the value of the wavevector
(n)
kx = 2m(E − En )/h̄. Transport is due to propagating, not evanescent waves, and kx has to be real. Then, only
a finite number of open channels, NL to the left and NR to the right, exists at a fixed energy E. We explicitly wrote
the square roots of velocities vn in each channel. This is to assure that the current density does not contain these
factors and is expressed in terms of aLn , bLn or aRm , bRm only.
In Eqs. (21), (22) the coefficients aLn , aRm are the amplitudes of the waves coming from the reservoirs, and
bLn , bRm are the amplitudes of the waves transmitted through or reflected back from the scattering region. These
coefficients are therefore not independent: The amplitude of the wave reflected trom the obstacle linearly depends on
the amplitudes of incoming waves in all the channels,
X X
bαl = sαl,βl0 aβl0 , β = L, R, l = n, m. (23)
β=L,R l0

The proportionality coefficients are combined into a (NL + NR ) × (NL + NR ) scattering matrix ŝ. It has the following
block structure,

r̂ t̂0
   
ŝLL ŝLR
ŝ = ≡ . (24)
ŝRL ŝRR t̂ r̂0

Scattering matrix:properties and

Scattering matrix
example
NR + NL
bL  rˆ tˆ' aL 


a

 = 
Incoming amplitudes


sˆ + sˆ = 1ˆ
ˆ  
rˆ + rˆ + tˆ + tˆ = 1ˆ


b = sˆa unitarity

bR  t rˆ' aR 

NR + NL Outgoing amplitudes b
ˆt = tˆ ' Time reversability

bL  r t ' aL 

 b  =  t r '  a 
r12 t12 r t' 1  R   R
r22
t22
1 t r'
r = Reiθ ;r ' =− Rei(2η−θ);
NL 1
r32
t32
NR One-
One-channel
t = Teiη; T =1− R
... .. scatterer
.

The NL × NL reflection marix r̂ describes the reflection of the waves coming from the left. Thus, r nn0 is the
amplitude of the following process: The electron coming from the left in the transverse channel n 0 , is reflected to
the channel n. Consequently, |rnn0 |2 is the probability of this process. The NR × NR reflection matrix r̂ 0 describe
reflection of particles coming from the right. Finally, NR ×NL transmission matrix t̂ is responsible for the transmission
through the scattering region. If magnetic field B is applied, the elements of the scattering matrix obey the following
0 0 0
conditions, rnn0 (B) = rn0 n (−B), rmm 0 (B) = rm0 m (−B), tmn (B) = tnm (−B). In particular, without the magnetic
0
field the transmission matrix t̂ in Eq. (24) is the transpose of the matrix t.
Any scattering matrix satisfies the unitarity condition, ŝ† ŝ = 1̂. The diagonal element of ŝ† ŝ is
2 2
X X
ŝ† ŝ nn =


|rnn0 | + |tmn | = 1, (25)
n0 m

since it represents a total probability of an electron in the channel n to be either reflected or transmitted, to any
channel.

10
 F. Landauer formula

We now turn to the calculation of current, with Eq. (16) as the starting point. Let us calculate the current through
a cross-section located in the left waveguide. The electrons with kx > 0 originate from the left reservoir, and their
filling factor is therefore fL (E). Now, the electrons with kx < 0 in a given channel n are coming from the scattering
region. A fraction of these electrons originate from the left reservoir and are reflected; they carry the filling factor
fL (E). This fraction P is determined by the probability to be reflected to the channel n from all possible starting
channels n0 , Rn (E) = n0 |rnn0 |2 . Other electrons are transmitted through the scattering region, their filling factor
being fR (E). The resulting filling factor for kx < 0 is therefore Rn fL (E) + (1 − Rn )fR (E). We write for the current
X Z ∞ dkx
I = 2s e vx (kx )fL (E)
n 0 2π
Z 0 
dkx
+ vx (kx ) [Rn (E)fL (E) + (1 − Rn (E))fR (E)] (26)
−∞ 2π
X Z ∞ dkx
= 2s e vx (kx )(1 − Rn (E)) [fL (E) − fR (E)] .
n 0 2π

To derive the last equation line, we have changed kx → −kx in the second integral in Eq. (26). We use the unitarity
relation (25) to prove that
X
1 − Rn = |tmn |2 = (t̂† t̂)nn .
m

Now we repeat the trick of the previous subsection changing variables from k x to E and arrive at the following
expression,

2s e ∞
Z
dE Tr t̂† t̂ [fL (E) − fR (E)] ,
 
I= (27)
2π 0

where we have used the short-hand notation

  X †

Tr t̂† t̂ = t̂ t̂ nn .
n

Alternatively, the trace can be presented as a sum of eigenvalues Tp of the Hermitian matrix t̂† t̂, transmission
eigenvalues. Because of the unitarity of the scattering matrix, Tp are real numbers between zero and one.

Landauer formula Simple-minded derivation

Hermitian matrix
tˆ + tˆ Has a set of
eigenvalues: Tp At each energy E
One channel:
Reservoir biased at 2s eVt
N att (t ) = electrons
transmissions voltage V sends 2π

I = GQ ∑ ∫ dE Tp ( E ) ( f L ( E ) − f R ( E ))
The current reads: Chance to pass: T0 Charge passed: Q = e T0 Natt
Average current: I = Q / t = GQ T0 V

I = GQ ∑ Tp V
p Many channels: sum over channels

I = GQ ∑ Tp V
p p

11
 The transmission eigenvalues depend on energy. However, in the linear regime, when the applied voltage is much
smaller that the typical energy scale of this dependence, they can be evaluated at the Fermi surface, and we obtain
the expression for conductance,
X
G = GQ Tp (EF ). (28)
p

Calculation of current in the right waveguide gives the same result: Current is conserved.
Eq. (28) is known as the (two-terminal) Landauer formula.
We have derived this relation assuming that the nanostructure is connected to ideal waveguides that support N L
and NR transport channels. Now we can get rid of this unrealistic assumption repeating the reasoning we have used
for the QPC. Let us unfold the waveguides so that their cross-sections become infinite: It should not change the
transport properties of the nanostructure. The number of transport channels becomes infinite, N L , NR → ∞. This
means that there are infinitely many transmission eigenvalues. This also means that the total number of transport
channels NL,R is an ”unphysical” quantity: It characterizes an auxiliary model rather than the nanostructure, and
no transport property of a nanostructure would eventually depend on NL,R ∗ .
How to reconcile the finite conductance given by Eq. (28) with the infinite number of transmission eigenvalues?
The implication is that infinitely many transmission eigenvalues are concentrated very close to zero transmission, so
that they contribute neither to conductance nor to any other transport property.
To evaluate the transmission eigenvalues of a given nanostructure, one solves the Schrödinger equation in the
scattering region and matches the two asymptotics (21), (22), extracting the scattering matrix. The solutions depend
on all the details like location of gates and barriers and the given configuration of the disorder. Even for relatively
simple systems, this is a time-consuming task, without much intellectual impact: A calculation for a given system
does not give us an idea what the result would be if we add a gate or move a tunnel barrier. Moreover, the calculation
will give a different result for a different congiguration of disorder.
This makes it important to comprehend the general properties of transmission eigenvalues, those depending on the
system design rather than on the details.
One channel scatterer. Let us start with a simple example: a scatterer that can transmit only one transport channel
(for a given energy). All but one transmission eigenvalues are zero. The structure is thus characterized by a single
transmission eigenvalue T . This is precisely the transmission coefficient we have discussed for the tunnel barrier
in Section I B; R = 1 − T is the reflection coefficient. The scatering matrix is a 2 × 2 matrix and contains more
parameters, since in Eq. (24) r, r 0 , and t are complex numbers, constrained by the conditions of unitarity. There are
three independent parameters T , θ, and η,
 √ iθ √ iη 
√Re √ Te
ŝ = . (29)
T eiη − Rei(2η−θ)

The phases θ and η do not manifest itself in the transport in a single nanostructure of this type. As we show in next
lectures, these phases are relevant if we combine two structures producing quantum interference effects.

G. Counting electrons

Any experimental measurement is in fact a result of averaging of many readings of a measuring device. This is
because the readings differ, or fluctuate, even if the parameters controlling the physical situation do not change. Each
individual reading is random being a result of interplay between many factors beyond our control. Some of them
come from the measuring device reflecting its unperfectness, some are intrinsic for the system being measured, some
can not be controlled at all owing to quantum mechanics. An example from quantum transport is the measurement
of the electric current in a nanostructure at a given voltage. The electron transfer is a stochastic process, and number
of electrons traversing the nanonstructure during a given time interval ∆t is random. Even if we measure the current
with an ideal ammeter, the readings would therefore differ.

∗
Confusingly enough, this ”number of transport channels” is commonly used in literature to characterize the area of the
(narrowest) cross-section of a nanostructure.

12
 Given the situation, there are two possible courses of action. First, and the most common one is to get rid of the
fluctuations by averaging over the big number of readings. The result of such approach is the average current, the
quantity we have studied so far. Alternatively, one can study the statistics of these fluctuations trying to measure a
probability of a certain current read out. This would generally require more measurements but rewards us with more
information collected. As we show in this Section, the statistics of the electron transfer reveals some information on
a nanostructure that can not be readily accessed by means of average current measurements.

Counting electrons Characteristic function

How to count correctly? Λ t ( χ ) = ∑ exp(i χ N ) Pt ( N )

time N

• Indepentent events: Λ factorizes Couple

<N>
• Time property: ln Λ t ( χ ) ∝ t single

PN 2
<<N >>
• Elementary event analysis
e e e
e e e e e
{ {
measurement 1 measurement 2 {
measurement 3
N ln Λ t ( χ ) = t ln(1 + W1 (exp(i χ ) − 1) + W2 (exp(i 2 χ ) − 1) + ...
+WM (exp(iM χ ) − 1))
∆t
Flock
How to count electrons?

Let us recall some general concepts of the probability theory. Suppose we make a measurement counting some
random events during a certain time interval ∆t. This can be, for instance, the number of babies born in Nashville,
Tennessee, during a week; the number of birds crossed the Continental Divide direction east during an hour; or the
number of electrons passed from one to another reservoir via a nanostructure during a nanosecond. The number of
events N measured during the time interval is a random number. Repeating the meausurement many times, we obtain
different results. Summing them up and dividing by the number of measurements, we get the average number hN i.
If the conditions are the same during the measurement and do not change from one measurement to the other (for
instance, if birds are only counted at the same daytime and during the same season), this average is proportional to
∆t.
Besides the average, we may want to know the distribution of the results — the probability P N that precisely N
events will be observed in a measurement. To quantify this probability, one repeats identical measurements M tot
times, and counts the number of measurements MN that give the count P N . The ratio MN /Mtot gives the probability
PN in the limit MN  1. This probability distribution is normalized, N PN = 1. Once we know it, we can estimate
not only the average,
X
hN i = N PN ,
N

but also, for instance, the variance, or second cumulant, of N , which measures the degree of the deviation from the
average,
!2
D E X
2
X
2
hhN ii = (N − hN i) = N 2 PN − N PN .
N N

The description of the statistics with the distribution function PN is not always the most convenient one. In the
example with the birds, let us suppose that we know the probability distribution P s of the birds which cross the
Divide south of Independence Pass, and P n for those which cross north of this point. These events are statistically
independent, since the birds hardly coordinate their itineraries. The total distribution is given by a convolution of
the two,
N
X
PNtot = s
PM PNn −M .
M =0

13
Most conveniently, this is expressed in terms of characteristic function of a probability distribution,
X
Λ(χ) = eiχN = PN eiχN .
N

For independent events, the characteristic function of the total distribution is just a product of characteristic functions
of each type of events, Λtot (χ) = Λs (χ)Λn (χ). This is a handy property if the outcome of the measurement is
contributed by various sources.
Differentiating the function ln Λ(χ) k times with respect to iχ and setting subsequently χ = 0, we generate the kth
cumulant of the distribution. Thus, the first derivative produces the average N , and the second derivative reproduces
the variance. How do the cumulants depend on ∆t? Let us divide the interval ∆t into two shorter intervals ∆t 1 and
∆t2 (but still long enough so that many events occur during each of them). Events occuring during each of these
intervals are statistically independent provided the intervals are long enough. Therefore the characteristic function is
a product of the characteristic functions describing the intervals. Its logarithm is the sum of corresponding logarithms,
ln Λ(χ, ∆t) = ln Λ(χ, ∆t1 ) + ln Λ(χ, ∆t2 ). Since ∆t = ∆t1 + ∆t2 , we conclude that ln Λ(χ, ∆t), and therefore all the
cumulants, are proportional to the time of measurement ∆t.
Statistics of electron transfers. Now we return to electrons in nanostructures. An event is a transfer of an electron
from one reservoir to another. The quantity to count in this case is the charge Q passed from left to right during
the time ∆t. We assume that this measurement time is long enough, so that Q  e and the laws of statistics apply.
On average, hQi = hIi∆t, and we have already spent quite some time calculating the (average) current hIi. We
now make a step further and describe statistical properties of a random variable Q. This challenging task can be
accomplishedwithin the scattering approach outlined. The result in a form of compact Levitov formula is given in this
Section. The derivation of this formula is not elementary, and is relegated to the advanced material. Before revealing
such an important piece of information to the reader, however, we discuss two simple limiting cases which prepare
him/her to comprehend this result.
Let us first assume that electrons are transferred only in one direction and the transfers are uncorrelated. To
calculate the characteristic function, we divide the interval ∆t into very short intervals dt. The probability to transfer
one electron during this short interval is Γdt  1, Γ being the transfer rate, the probability to transfer no electrons
is consequently 1 − Γdt. We neglect the probability to transfer more than one electron since this probability is
proportional to (dt)2 and is therefore much less than Γdt. The characteristic function for a short interval thus reads
D E
Λdt (χ) = eiχQ/e = (1 − Γdt) + (Γdt)eiχ .

Since the electrons pass independently, the characteristic function for the whole interval is just a product,
 
Λ∆t (χ) = (Λdt (χ))∆t/dt = exp Γ∆t(eiχ − 1) = exp Ñ (eiχ − 1) .


(30)

Here Ñ ≡ Γ∆t is the average number of electrons transferred, Ñ = hQi/e. Taking the inverse Fourier transform, we
find for the probability PN for N particles to be transferred during the time ∆t,
Z 2π Z 2π
dχ dχ −iχN +Ñ (eiχ −1)
PN = Λ(χ)e−iN χ ≈ e
0 2π 0 2π
Ñ N −Ñ∆t
= e . (31)
N!
Eq. (31) is recognized as the Poisson distribution. As we will see, this situation of uncorrelated electron transfer
occurs in tunnel junctions where all transmission eigenvalues are small. In this case, the currents are small implying
that the time intervals between sucsessive transfers are big. Therefore it is easy to understand why they do not
correlate.
An opposite example is an ideally transmitting channel at zero temperature. In this case, the electrons are in ideal
wave states and their momentum in transport direction is a well-defined quantum number. It does not fluctuate. Since
the total current is just a sum over momenta of individual electrons, it does not fluctuate as well. The distribution
PN = δ(Ñ −N ) provides the characteristic funct on Λ(χ) = exp(iχ(N − Ñ )). The transfers are thus correlated ideally.

14
 Levitov formula Electrons gambling

 1free (eiχ − 1) f L (1 − f R ) + 
ln Λ ( χ ) = 2s ∆t ∫ ∑ ln  Tnumbers:
+ Tp particles ln Λ ( χ ) = N at ln {1 + T0 (eiχ − 1)} 2s eVt

Hamiltonian of the
dE N at =
2π

Energy in terms of2πoccupation

(e χ − 1) f 
−i

p p R (1 − f L )
Nat = number of game slots
Electron transfers T
Nat N T0 = winning chance
•Independent in different energy strip e e e e e e e N = number of games won

 N at  N
•Independent in different channels
•To the left and to the right: are dependent! blocking factor
PN =   T0 (1 − T0 )
e e e
N at − N

 
•Transfers at negative energy are blocked! f L (1 − f R ) Nat-N
N
Simple example: electrons transferred in one direction

Levitov formula. For intermediate transmissions 0 < Tp < 1 the transmitted electrons are correlated, but not
ideally. The many channel, finite temperature result for the characteristic function is given by the Levitov formula,
dE X 
Z
ln 1 + Tp eiχ − 1 fL (E) [1 − fR (E)]


ln Λ(χ) = 2s ∆t
2πh̄ p
+ Tp e−iχ − 1 fR (E) [1 − fL (E)] .


(32)

The logarithm of characteristic function is a sum over transport channels, this suggests that electron transfers in
different channels are indepentent. Also, the logarithm is an integral over the energy suggesting that electrons are
transferred independently in each energy interval. Importantly, the electron transfers from the left to the right and
from the right do correlate. To stress this, let us consider an energy strip where f L = fR = 1 so that the electron
states are filled in both reservoirs. The net current is zero. If the transfers were uncorrelated, they would give rise to
current fluctuations. However, the formula gives no events in this case: Transfers to the left are blocked by electrons
filling states in the right reservoir, the same is true for transfers to the right.
Electrons gambling To comprehend the formula, let us consider the limit of negligible temperature, eV  k B T . In
this case, the integration over energy is confined to the energy strip min µL,R < E < max µL,R and the integrand does
not depend on energy. Recalling that µL − µR = eV we obtain
2s eV ∆t X 
ln 1 + Tp e±iχ − 1 ,


ln Λ(χ) = ± (33)
2πh̄ p

where the upper and lower signs refer to the case of positive and negative voltages, respectively. Let us for simplicity
consider V > 0. We define Nat = 2s ∆teV /(2πh̄) and assume it to be integer. The characteristic function becomes
Y
Λ(χ) = Λp (χ);
p
Na t  

Na t X Nat
Λp = (1 − Tp ) + Tp eiχ ) = TpN (1 − Tp )Na t−N eiN χ .
N
N =0

We made use of Newton’s binomial theorem in the last expression. Let us concentrate on one channel p and take the
inverse Fourier transform of ΛP . We obtain the binomial distribution of the number of electrons transferred,
 
(p) Nat
PN = TpN (1 − Tp )Nat −N , (34)
N
which allows for a frivolous interpretation. The point is that the binomial distribution is known from theory of
gambling: For a given winning chance Tp and number of game slots (number of attempts) Nat it gives the probability
to win N times.
At zero temperature and positive voltage, all the electrons are incident from the left reservoir trying to get to the
right. The interpretation suggests that the stream of incident electrons is very regular: The time interval between

15
the arrivals of two adjacent electrons is the same, ∆t/Nat = e/GQ V . Each of them either passes through the barrier
(with the probability Tp ) or is reflected back (with the probability Rp = 1 − Tp ). The average number of those which
passed is Nat Tp , conforming to Landauer formula. The distribution PN given by Eq. (34) is the probability that out
of Nat electrons arriving to the barrier N pass through, and others, Nat − N , are reflected back.

Electrons gambling

ln Λ ( χ ) = N at ln {1 + T0 (eiχ − 1)} N at =
2s eVt
2π

Nat = number of game slots

T
Nat N T0 = winning chance
e e e e e e e N = number of games won

 N at  N
PN =   T0 (1 − T0 )
e e e
N at − N

 
Nat-N
N

For more than one channel, the binomial distribution does not hold any more. Still we obtain a convolution of
binomial distributions corresponding to each channel.
The electrons appear on the right side of the barrier in an irregular fashion. If Tp is small, we can assume that the
intervals between those which have passed are long, random and independent — the two subsequent electron transfers
are thus uncorrelated. Indeed, if we take the Levitov formula in the limit of Tp  1, it gives the characteristic funtion
P
(30) with Ñ /∆t = GQ V /e p Tp = GV /e =< I > /e. The Poisson distribution (31) is thus the limiting case of the
binomial distribution (34) for T  1 and N  Nat .
If transmission eigenvalues do not depend on energy, the integral over energy in (32) can be taken explicitly at
arbitratry relation between eV and kB T . The characteristic function then becomes
   
2s kB T ∆t X eV
ln Λ(χ) = arccosh2 Tp cosh + iχ
2πh̄ p
2kB T
   2 )
eV eV
+ (1 − Tp )cosh − . (35)
2kB T 2kB T

Distribution of transmission eigenvalues. The transmission eigenvalues Tp depend on disorder configuration and
therefore are random . We need a quantity that characterizes design of a nanostructure rather than a concrete disorder
configuration. This is provided by the distribution function of transmission eigenvalues (transmission distribution)
P (T ). Suppose we make an ensemble of nanostructures sharing an identical design and differing in disorder con-
figurations. Each nanostructure provides a set of transmission eigenvalues. Let us concentate on a narrow interval
of transmissions from T to T + dT , count the number of transmission eigenvalues that fall into this interval, and
divide this by the total number of nanostructures. In the limit of a big ensemble, the result converges to P (T )dT .
Mathematically, the transmission distribution is thus defined as follows,
* +
X
P (T ) = δ (T − Tp (E)) . (36)
p

The angular brackets in above relation mean the ensemble average, that is, the average over all formally identical
nanostructures in the ensemble. The function P (T ) facilitates evaluating other averages. The average of an arbitrary
function of the transmission eigenvalues becomes
* + Z
X 1
f (Tp ) = dT f (T )P (T ). (37)
p 0

16
In particular, one integrates T P (T ) to obtain the average conductance hGi.
What is the use of the above relation? If the average conductance of a nanostructure much exceeds the conductance
quantum, hGi  GQ , the transmission eigenvalues are dense, the typical spacing between the eigenvalues being much
less than one. This means that the sums over transmission eigenvalues can be replaced by the integrals according to
Eq. (37). The transport properties are thus self-averaged in this limit, their fluctuations being much smaller than
the average values. The transport properties appear to be almost insensitive to a specific disorder configuration. The
fluctuations of transport properties may become significant if hGi ' GQ , and the transmission eigenvalues are sparse.
A fair part of this book either quantifies the transmission distribution or makes use of it. In the rest of this Section,
we provide examples of P (T ) without quantifying it.

ρ )
(T

T T

Transmission distribution Types of quantum contacts

Quantum contact: an individuality: pin-code Tp

Person pin-code ATM machine Type of a quantum contact:= shape of transmission distribution
Quantum Quantum Tunnel junction: Tp ≈ 0 Realistic QPC Tp ≈ 1
Contact Tp Transport

ρ (T ) = ∑ δ (T − T )
p
p
ρ(T)

ρ(T) Diffusive contact:

G 1
ρ (T ) =
2GQ T 1 − T
0 0.2 0.4 0.6 0.8 1
T

Let us start with a QPC and consider energy at which a finite number Nopen of transport channels are open
(T = 1). An infinite number of channels are closed (T = 0). The corresponding transmission distribution consists of
two delta-functional peaks,

P (T ) = Nopen δ(1 − T ) + ∞ δ(T ).

Closed channels do not play any role in transport, and the part with the open channels leads to the expression for
the conductance, G = GQ Nopen we have already seen. We will ignore the part proportional to δ(T ) and write for a
clean QPC

PQP C (T ) = Nopen δ(1 − T ). (38)

The transmission eigenvalues in a clean QPC are highly degenerate. If we add a small number of defects to the QPC,
this degeneracy is lifted. The scattering at the defects mixes the channels: An incident electron in the open channel
n which without scattering would pass the constriction, can now be reflected to any channel n 0 , or be transmitted
to an arbitrary channel m. These processes modify the transmission matrix, and, consequently, the transmission
eigenvalues. If such channel mixing is weak so that the probabilities to scatter from open channels are small, we
expect that all transmission eigenvalues remain close to 1. The role of disorder is thus to lift the degeneracy. This
regime is realized when the contribution of disorder to the total conductance of the system is sufficiently small, the
resistance R due to defects being much smaller than the resistance of the QPC. At further increase of resistance R to
values of the order of 1/GQP C the transmission eigenvalues are spead over the whole interval 0 < T < 1.
A complementary example is a tunnel junction. Let us take a sufficiently wide ideal potential barrier at an energy
much below the top energy of the barrier. All the transmission coefficients are guaranteed to be small, T  1. If
the channels do not mix, the transmission eigenvalues are just these coefficients and the transmission distribution
concentrates near T = 0. If we add some defects next to the barrier, the channels mix. Some electrons after being
reflected from the barrier are reflected by defects back to the barrier. They just get the ”second chance” to tunnel
through. Because of this, some transmission eigenvalues grow with increasing the defect resistance R. Similarly to
QPC, the transmission eigenvalues are spread over the whole interval 0 < T < 1 if R is comparable with the resistance
of the tunnel junction.

17
 Let us add more defects. At some stage, the resistance due to the defects dominate the total resistance. At this
point, we can forget about a QPC or a tunnel junction present in the structure. The electron that traverses the
scattering region experiences many scattering events at the defects. Its motion is highly random. This corresponds to
diffusion provided the conductance of the structure still exceeds much the conductance quantum. The transmission
distribution in a diffusive structure appears to be universal — not depending of the details of the structure design

hGi 1
ρ= √ (39)
2GQ T 1 − T

The integral of the transmission distribution over T gives the total number of transport channels. This integral
diverges for the bimodal distribution (39) indicating an infinite number of channels that may take part in diffusive
transport.

18