Developing Methods For Transport Calculations in Heterojunctions With Coulomb Interactionsfinal Full (Heb)

Tel Aviv University
School of Chemistry
Subject:
Developing methods for transport

calculations in heterojunctions with
Coulomb interactions
Thesis submitted for the degree of

Doctor of Philosophy
by
Tal J. Levy
Submitted to the Senate of Tel-Aviv University

July 2014
This work was carried out under the supervision of
Professor Rabani Eran

School of Chemistry, Tel-Aviv University
Dedications and Acknowledgements

This thesis is dedicated to my parents, who throughout the years have given me much more
than I can ever express here; to my brothers, Gad and Ran, who always find the time for a
beer and a laugh when I am down, and to my mischievous cats, who always remind me that
there is an easier way.
I also dedicate this thesis to my beautiful partner and companion through life, Ariel, who
recently gave me the best present of all - my daughter Phoenix.
Last but not least, I dedicate this thesis to my advisor, Prof. Rabani Eran, whose only
concern was my well-being (personally and scientifically). I really do mean it when I say that
without your help and guidance, all of this would have never happened. Thank you.
Finally, I would like to thank my past and present mentors, collaborators, group members,
teachers and friends. Every little thing you have said or done contributed to my success.
Abstract
The demand for smaller, faster, and lower cost electronics has sent the industry and the scientific
community looking for different paradigms. It almost naturally led to the question of whether
it is possible to use single molecules and quantum dots as active elements in nanocircuits for
a variety of applications. Developments in nanofabrication techniques have made possible the
old dream of contacting individual molecules and quantum dots to macroscopic electrodes and
explore their electronic transport properties. Moreover, it has been shown that molecules can
indeed mimic the behavior of some of the more commonly used microelectronic components
e.g., diodes, switches, and transistors. These achievements have given rise to what is now known
as Molecular Electronics.
There are still many experimental issues to overcome before molecular electronics will turn
from an idea into a realizable technology, regardless, the exploration of molecular-scale circuits
has already led to the discovery of many fundamental effects. In this sense, molecular electronics has become a new interdisciplinary field of science, in which knowledge from traditional
disciplines like physics, chemistry and engineering is combined to understand the electrical
and thermal conduction at the molecular scale. Nevertheless, theory still faces several important challenges as the description of transport via an interacting nanoscaled region is a grave
problem.
Generally speaking, transport is a many-body, nonequilibrium phenomena, and its exact
treatment requires a formalism explicitly designed to work out-of-equilibrium. While the equilibrium properties of systems similar to those to be treated in this thesis are quite well understood and can be accurately solved for, the development of a general approach suitable for
the treatment of fully nonequilibrium, many-body systems still remains a formidable task. An
exact theoretical treatment of such systems is rather sparse to this day and includes only a
small class of over-simplified model problems.
We gather that in order to exploit the limitations and potential of the aforementioned
novel electronic it is compulsory to carry out a comprehensive theoretical study of the physical
properties that govern these systems. To this end, improvement of existing computational tools
and formalisms, and the development of new ones are mandatory. The limitations of todays
formally/numerically exact solutions are often too severe for them to be useful in practice
(usually, exponential scaling of computational resources and/or computational time with system
size), thus, most theoretical treatments of quantum transport rely on approximations of some
sort.
In this thesis we study, analyze, and develop two methods for transport calculations in
nanoscaled heterojunctions, described by model Hamiltonians (impurity models), with electronelectron interactions. The methods are: (a) the equations-of-motion technique for the nonequilibrium Green function and (b) a semiclassical approach.
The nonequilibrium Green function formalism is considered as one of the pillars of current
approaches to quantum transport which have been implemented in both model Hamiltonian
formulations and first-principle methodologies. This approach is able to deal with a broad
variety of physical problems related to quantum transport at the nanoscale. It can deal with
strong non-equilibrium situations and it can also include interaction effects (e.g., electronelectron, electron-phonon) in a systematic way (diagrammatic perturbation theory).
In this work we elaborate on the equations-of-motion technique which is one of the more basic
and more powerful methods to calculate the Green functions of interacting quantum systems.
In spite of its simplicity, it gives the appropriate results for strongly correlated nanosystems,
describing qualitatively and in some cases quantitatively important transport phenomena, such
as, Coulomb blockade and the Kondo effect in quantum dots.
Be that as it may, we prove that Green functions calculated using the latter method may
break the symmetries and relations that the correlation functions should fulfill by definition.
Consecutively, the various expectation values calculated with these objects may turn to be
unphysical. Thereafter, we suggest a strategy to restore the symmetries lost in the process
of deriving the equations-of-motion and advise to add this step as an essential part of the
equations-of-motion technique.
Illustrations are then provided for two impurity models: the Anderson model (which serves
as the simplest model to incorporates electron-electron interactions) and the double Anderson
impurity model (an extended Hubbard model which includes inter-site Coulomb repulsion on
top of the on-site electron-electron interactions). In addition, we develop two closures for the
equations-of-motion obtained for the double Anderson model based on physical justifications
in the regime studied, that outperform the commonly used closures found in the literature.
On a completely different note, we explore and develop a semiclassical approach, which allows for the study of nonequilibrium quantum transport in molecular junctions using Hamiltons
classical equations-of-motion, where the number of equations scales linearly with the number
of the degrees-of-freedom. The dynamics of Fermions has always provided a challenge for semiclassical methods, since Fermions have a particularly notable non-classical behavior, which is
the exchange antisymmetry of Fermions (which can be stated as the Pauli exclusion principle).
The key idea behind the three methods presented here is to transform a general second quantized many-electron Hamiltonian into a classical one by defining a prescription which maps the
electronic operators into classical functions that correctly account for the anti-commutativity
of the Fermion operators. The three different mappings depicted in this work are just a natural evolution of the same basic idea, where each transformation addresses the flaws of its
predecessor.
Contents
1 Introduction
1.1 Molecular Electronics . . . . . . . . . .
1.2 Theoretical treatment . . . . . . . . . .
1.3 The model Hamiltonians . . . . . . . .
1.3.1 The single resonant level model
1.3.2 The Anderson model . . . . . .
1.3.3 The Double Anderson model . .
1.4 Thesis outline and goal . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2 Equations-of-Motion technique for the Nonequilibrium Green functions in

Quantum Transport
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Nonequilibrium Green Functions . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Dyson equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.2 Analytic continuation and Langreth rules . . . . . . . . . . . . . . . . . .
2.2.3 Equations-of-motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.3.2 The equations-of-motion method . . . . . . . . . . . . . . . . .
2.2.3.3 The single resonant level model - EOM treatment . . . . . . . .
2.2.4 NEGF symmetry breaking within the EOM technique . . . . . . . . . . .
2.2.4.1 The Anderson model . . . . . . . . . . . . . . . . . . . . . . . .
2.2.4.2 The double Anderson model . . . . . . . . . . . . . . . . . . . .
2.2.5 NEGF symmetry restoration . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.5.1 The Anderson model: symmetric-EOM approach . . . . . . . .
2.2.5.2 The double Anderson model: symmetric-EOM approach . . . .
2.3 Steady state conductance in a double quantum dot array:
Assessing the symmetric-EOM technique for the NEGF . . . . . . . . . . . . . .
2.3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 Different closures of the EOM . . . . . . . . . . . . . . . . . . . . . . . .
9
11
11
13
14
16
16
17
17
19
19
20
27
28
29
29
31
33
36
36
40
46
48
49
52
52
53
2.4
2.3.2.1 Approximation 1 .
2.3.3 Master equations . . . . . .
2.3.4 Results and discussion . . .
2.3.4.1 Symmetric bridge .
2.3.4.2 Asymmetric bridge
Concluding remarks . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3 Semiclassical approaches to quantum transport

3.1 overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Action-angle mapping . . . . . . . . . . . . . . . . . . . . .
3.1.2 Cartesian mapping . . . . . . . . . . . . . . . . . . . . . . . .
3.2 The Hubbard mapping . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Hubbard Operators . . . . . . . . . . . . . . . . . . . . . . . .
3.2.2 Reformulating the Hamiltonian in terms of Hubbard operators
3.2.3 Classical mapping for the Hubbard operators . . . . . . . . . .
3.3 Quantum transport in impurity models:
Assessing the Hubbard mapping . . . . . . . . . . . . . . . . . . . . .
3.3.1 Resonant level model . . . . . . . . . . . . . . . . . . . . . . .
3.3.2 Anderson impurity model . . . . . . . . . . . . . . . . . . . .
3.3.3 Double Anderson model . . . . . . . . . . . . . . . . . . . . .
3.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
55
56
56
57
57
58
59
66
71
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
73
73
75
76
78
79
80
81
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
84
85
85
90
90
4 Summary and Outlook
95
A Full derivation of the broken symmetry in the Anderson model
97
B Brief summary of the derivation of the NEGFs EOM: the double Anderson
model
103
C Calculating the poles of the retarded NEGF
109
D Complimentary information for the Hubbard mapping
113
Bibliography
117
10
Chapter 1
Introduction
1.1
Molecular Electronics
The demand for smaller, faster, and lower cost electronics, is by no means new. Already back
in 1956 Arthur von Hippel formulated the basis of the bottom-up approach that he called
molecular engineering [1]:
Instead of taking prefabricated materials and trying to devise engineering applications consistent with their macroscopic properties, one builds materials from their atoms and molecules
for the purpose at hand ...
Of course he was not alone, in 1959 the physicist Richard Feynman discussed the possibility
of devices of extremely small dimensions in his lecture entitled Theres plenty of room at the
bottom [2]:
I dont know how to do this on a small scale in a practical way, but I do know that
computing machines are very large; they fill rooms. Why cant we make them very small, make
them of little wires, little elements and by little I mean little. For instance, the wires should
be 10 or 100 atoms in diameter, and the circuits should be a few thousand angstroms across. [.
. .] There is plenty of room to make them smaller. There is nothing that I can see in the laws
of physics that says the computer elements cannot be made enormously smaller than they are
now.
The concept of molecular engineering introduced by Von Hippel led to the first notion of
molecular electronics. It eventually resulted in a collaboration between Westinghouse Electric
and the U.S. Air Force at the end of the 1950s. At that time, the U.S. Air force was open to
new ideas and alternatives to the recently introduced integrated circuits and thus, organized
a conference on molecular electronics which included scientists and engineers from military
and private research labs. It was there that colonel C.H. Lewis, director of Electronics at the
Air Research and Development Command, coined the term Molecular Electronics:
Instead of taking known materials which will perform explicit electronic functions, and
11
reducing them in size, we should build materials which due to their inherent molecular structure will exhibit certain electronic property phenomena. We should synthesize, that is, tailor
materials with predetermined electronic characteristic. Once we can correlate electronic property phenomena with the chemical, physical, structural, and molecular properties of matter, we
should be able to tailor materials with predetermined characteristics. We could design and create
materials to perform desired functions. Inherent dependability might eventually result. We call
this more exact process of constructing materials with predetermined electrical characteristics
Molecular Electronics.
However, molecular electronics, as we understand it today, started at the end of the 1960s
and the beginning of 1970s. Thereupon, different groups started to investigate, experimentally,
electron transport through molecular mono-layers. For instance, Hans Kuhn and coworkers
studied new ways of fabricating mono-layers of organic materials (LangmuirBlodgett films) [3],
which they were able to sandwich between metal electrodes and measure the electrical conductivity of the resulting junctions. A few years later (1974), Arieh Aviram and Mark Ratner
published a now-famous paper on molecular rectifiers [4] in which they described how a modified charge-transfer salt could operate as a traditional diode in an electrical circuit. The idea
was considered for a long time a theoretical curiosity that could not be tested experimentally,
and in this sense, it did not have much impact in the scientific community back then.
Much has changed with the invention of the scanning tunneling microscope (STM) [5, 6]
in 1981, the introduction of mechanically controllable break junction (MCBJ) technique [7]
in 1985, and the development of controllable single molecule junctions [8, 9, 10, 11, 12, 13,
14, 15, 16, 17, 18, 19, 20, 21]. The STM was the first tool that provided a practical way
to see and manipulate matter at the atomic scale and very quickly it became clear that it
could provide a realistic way to address single molecules and to study their electronic transport
properties [22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35].
With these techniques a large amount of interesting phenomena have been observed. For
example: rectification [17], negative differential conductance [9, 33, 36], Coulomb blockade [22,
10, 11, 14, 15, 20], Kondo effect [11, 37], vibrational effects [24, 29, 30, 32, 34], and nanoscale
memory effects [38, 39, 40] just to name a few.
Many engineering challenges, such as robustness and stability of these molecular devices,
still remain to be addressed before molecular electronics can be practical, however, these
results clearly show the potential of molecule-based electronics. Mechanical challenges are not
the only thing that needs to be attended to. In that regard the theoretical treatment of current
and heat flowing through a nanoscaled junction is far from complete.
12
1.2
Theoretical treatment
The growing interest in the properties of both mesoscopic and nanometric devices and the new
physics and phenomena encountered in the various labs has raised fundamental and conceptual
questions. Theory faces several important challenges and the development of a general approach
suitable for the treatment of fully nonequilibrium many-body systems still remains a formidable
task. In this regard the main challenges facing theory are:
Understanding the coupling of individual molecules to macroscopic electrodes under

nonequilibrium conditions [41, 42, 43, 44].
Characterization of the temperature dependence on conductance as well as the role of
molecular vibrations are of crucial importance. This would facilitate understanding of
elastic and inelastic transport phenomena, where the nature of both coherent and incoherent transport is paramount for a complete picture [45, 46, 47, 48].
Electron-electron correlations in the macroscopic electrodes can often be neglected by the
application of an effective band-structure description of the leads, such correlations, on the
other hand, are important within the molecule and thus strongly affect the conduction [49,
50].
Electron-hole excitations (excitons [51]) formed in the heterojunction, can remarkably
affect the current through it. While in case of noninteracting electrons the electron-hole
interaction leads to reduction in the current (exciton blocking), in the case of strong
Coulomb repulsion, conduction exists even when the electronic connectivity does not
exist [52]. Also, electronhole pair creation processes can explain vibrational excitations
characteristic of a single-molecule contact [53].
Characterizing transport behavior in systems driven by weak and strong electromagnetic
fields and the optical properties of such junctions is another difficult issue which must be
overcome by theory [54, 55, 56, 57].
Although noteworthy advances have been made, there is still a discrepancy between experimentally measured and theoretically calculated values [42, 18, 58]. While recent work has been more
encouraging regarding these discrepancies, the problem clearly remains unsolved. Thus, the development of a practical, general approach suitable for the treatment of fully nonequilibrium,
many-body systems is desirable.
13
In general, many-body, interacting, out-of-equilibrium problems cannot be solved exactly

but for a few simple cases [59, 60, 61, 62]. Excluding recent developments based on bruteforce approaches such as time-dependent numerical renormalization-group techniques [63, 64,
65, 66, 67, 68], iterative [69] or stochastic [70, 71, 72, 73, 74] diagrammatic techniques to
real- time path integral formulations, wave function based approaches [75], or reduced dynamic
approaches [76, 77, 78, 79], all suitable to relatively simple model systems, most theoretical
treatments of quantum transport rely on approximations of some sort.
From a more formal standpoint, there are roughly two main theoretical frameworks that
can be used to study quantum transport in nanosystems at finite voltage: the nonequilibrium
Green function (NEGF) formalism [80, 81, 82] otherwise known as the Keldysh NEGF or the
SchwingerKeldysh formalism [83, 84], and generalized master equations methods [85, 86].
In this work we also introduce an emerging paradigm suitable to tackle quantum transport
at the nanoscale; a semiclassical approach [62, 87, 88, 89, 90], where the quantum Hamiltonian,
and the quantum operators are mapped onto classical functions and degrees-of-freedom that
follow Hamiltons equations of motion.
For the sake of completeness we mention two more emerging and improving theoretical
frameworks applicable to open nonequilibrium systems: (a) methods based on density functional
theory (DFT) [91, 92, 93, 94, 95, 96, 97, 98], and (b) time dependent multiconfigurational
methods [99, 100, 101].
In the next section we introduce the model Hamiltonians used throughout this thesis.
1.3
The model Hamiltonians
Up till 1970, in most treatments of electron tunneling through a metal-insulator-metal (MIM)

junction it was accustomed to transform an initial Hamiltonian (of the total system) into an
effective Hamiltonian of the form:
ef f = H
L + H
R + T,
H
(1.3.1)
L and H
R represents the metals in the (MIM) junction (as accurately as possible),
where H
i.e., the left and the right electrodes, while the transfer term T describes the probability of
an electron to tunnel through the junction. Although, the tunneling is energy dependent,
this dependency was neglected from the transfer term, thus, making this theory unsuitable to
describe many-body effects, such as electron-electron and electron-phonon interactions. The
main reason for using effective Hamiltonians was the belief that a description of a system as a
whole will not allow for the calculation of the current, which is a nonequilibrium process.
In 1970 Caroli et al proposed the prototype of model Hamiltonians [102, 103, 104, 105] used
14
ever since in the context of transport. The Hamiltonian they proposed was designed to permit
direct calculation of the tunneling current through a MIM junction. This was achieved by
describing the system in terms of localized atomic functions instead of quasi-free electron wave
functions, which resulted naturally in a three-parts Hamiltonian describing the electrodes, the
system, and the interaction between these parts. The general form of the Hamiltonian is given
by:
=H
sys + H
bath + H
int ,
H
(1.3.2)
bath describes the macroscopic leads (left and right electrodes), H

sys describes the
where H
int is the interaction Hamiltonian between the system and the leads.
system of interest, and H
The contacts (leads) are usually modeled as infinite noninteracting Fermion baths [106, 107,
108], they are assumed to be each at their own equilibrium characterized by chemical potentials
R
is the applied voltage bias across the junction
L and R , where the difference = L
e
(e is the absolute electronic charge of an electron), and are described by a grand canonical
distribution. The leads Hamiltonian in second quantization is given by:
bath =
H
k ck ck ,
(1.3.3)
K{L,R} ,kK
where k is the energy of a free electron in the left (L) or right (R) lead, in momentum state
k and spin , with the operators ck and ck being annihilation and creation operators of such
an electron. The density matrix of the left/right lead at equilibrium is given by:
L/R =
P
"
Tr e
,kL/R k ck ck L/R NL/R
P
,kL/R k ck ck L/R NL/R
# ,
(1.3.4)
where = (KB T )1 is the reciprocal of the temperature T and the Boltzmann constant KB , the
P
trace (Tr [ ]) should be taken as the sum over a complete basis and NL/R = ,kL/R ck ck
is the occupation operator of the left/right lead.
The interaction between the system and the contacts is simply given by the tight-binding
Hamiltonian [102]:
X
X X
int =
H
tkm ck dm + h.c..
(1.3.5)
K{L,R} ,kK m
The transition matrix element tkm represents the coupling strength between the system (site
m, spin ) and the leads (electron in state k and spin ). It depends on the specific contact
geometry between the bridge and the leads. The corresponding level width functions are given
15
by:
L/R
nm () = 2
(tkn ) tkm ( k ) .
(1.3.6)
,kL/R
These functions describe the broadening of the energy levels on the molecular bridge due to
this coupling.
sys depends on the system studied. In the following subsections we
The form chosen for H
describe the different system Hamiltonians in detail.
1.3.1
The single resonant level model
This simple model was introduced by Ugo Fano [109] in 1961 when he worked on the effect of
auto-ionization on excitation spectra. In this model the system has one spin-less level:
sys = n.
H
(1.3.7)
Here n = d d is the number operator of the electron occupying the level with free-energy .
Since the system has only one level, and no spin, we can simplify the expressions of the bath
and interaction Hamiltonians:
bath =
H
k ck ck ,
(1.3.8)
tk ck d + h.c..
(1.3.9)
K{L,R} kK
int =
H
K{L,R} kK
1.3.2
The Anderson model
The same year Fano introduced his single level Hamiltonian, Philip W. Anderson was seeking
for conditions required for the stability of localized magnetic moments like those of iron, cobalt
or nickel dissolved in non-magnetic metals [110]. In the model he proposed the system is
represented by one electronic level that can accommodate up to 2 interacting electrons (spin
up and spin down electrons) [111, 112]. The systems Hamiltonian for the Anderson model is
given by:
X
sys =
H
n + U n n .
(1.3.10)
{,}
Here n = d d is the number operator of the spin electron with energy and U is the
repulsion energy between two electrons on the same site with opposite spins (on-site repulsion).
16
The interaction Hamiltonian takes the form:

int =
H
tk ck d + h.c..
(1.3.11)
K{L,R} ,kK
1.3.3
The Double Anderson model
This model is an extension of the Anderson model to systems that include more than just
one electronic level. Here we have 2 electronic levels each of which can accommodate up to
2 electrons. The model incorporates on-site and inter-site Coulomb repulsions and a hopping
term that allows electrons to hop from one site to the other. The Hamiltonian in second
quantization is given by [113, 114]:
sys =
H
m nm +
Um nm nm
,m{,}
0
V
n n0 +
, 0
Xh
h d d + h.c. .
(1.3.12)
The first two terms on the R.H.S are similar to the Anderson impurity model Hamiltonian
0 is the repulsion energy between two electrons on different
(extended to 2 sites, and ), V
sites (inter-site repulsion), and h is the coupling strength for electron hopping between the
two sites. In the calculations performed and results presented throughout this thesis the two
sites are assumed to be in a serial configuration, thus, the interaction Hamiltonian takes the
form:
X
X
int =
tk ck d +
tk ck d + h.c..
H
(1.3.13)
,kL
,kR
That is, site is connected to the left lead while site is attached to the right lead.
1.4
Thesis outline and goal
The primary objective of the research program presented here has been to analyze and develop
existing and novel theoretical and computational methods tailored to investigate charge (and
spin) transport at nanometer length scales. The theoretical examination of transport at the
nanoscale requires the use of advanced and powerful techniques able to deal with the dynamical
properties of the relevant physical systems, to explicitly include out-of-equilibrium situations
typical for electronic transport as well as to take into account interaction effects. In particular,
we have been interested in techniques suitable for large electronic systems, i.e. low scaling tech17
niques that describe the system at hand beyond a mean-field approximation. In that regard we
focused our attention on two methods for transport calculations in nanoscaled heterojunctions,
described by model Hamiltonians (impurity models), with electron-electron interactions:
1. The equations-of-motion technique for the nonequilibrium Green function.

2. A semiclassical approach.
The nonequilibrium Green functions formalism is introduced in chapter 2 where we elaborate on

the equations-of-motion technique. We prove that Green functions calculated using the latter
may at different levels of closures and decoupling schemes break the symmetries and relations
that the correlation function should fulfill by definition, consecutively, the various expectation
values calculated with these objects may turn to be unphysical.
Thereafter, we suggest a strategy to restore the symmetries lost in the process of deriving
the equations-of-motion and advise to add this step as a mandatory part of the aforementioned
technique. Illustrations are then provided for two impurity models: the Anderson model (which
serves as the simplest model to incorporates electron-electron interactions) and the double
Anderson impurity model (an extended Hubbard model which includes inter-site Coulomb
repulsion on top of the on-site electron-electron interactions). In addition, we develop two
closures for the equations-of-motion obtained for the nonequilibrium Green function of the
double Anderson model based on physical justification in the regime studied, that outperform
the commonly used closures found in the literature.
In chapter 3, we present three mappings of a semiclassical approach developed in collaboration with the Miller group at the University of California, Berkeley U.S.A. These methods allow
for the study of nonequilibrium quantum transport in molecular junctions using Hamiltons
classical equations-of-motion, where the number of equations scale linearly with the number
of degrees-of-freedom. The key idea behind these methods is to transform a general second
quantized many-electron Hamiltonian into a classical one. This is achieved by defining a prescription which maps the electronic operators into classical functions that correctly account
for the anti-commutativity of the Fermion operators. The three different mappings are just a
natural evolution of the same basic idea, where each transformation addresses the flaws of
its predecessor.
Out of the three mappings we center on the most recent Hubbard mapping and asses its
validity by employing the method on the three model Hamiltonians described in the previous
section.
18
Chapter 2
Equations-of-Motion technique for the
Nonequilibrium Green functions in
Quantum Transport
2.1
Overview
In this chapter we outline the theory of the nonequilibrium Green function (NEGF) formalism,
which is widely used to describe transport phenomena in nanojunctions. As will be clear shortly,
the NEGF method is able to deal with a very broad spectrum of physical problems related to
transport at the nanoscale [80, 82, 59, 115, 116, 117]. For one, as it name implies, it is tailored
to take on out-of-equilibrium situations, and interaction effects can be dealt with in a well
defined manner (diagrammatic perturbation theory [118], functional derivatives technique [80,
119]). The NEGF formalism, initiated by Schwinger [83], Kadanoff and Baym [80] and later
Keldysh [84] allows one to:
Study the time evolution of a many-particle quantum system.
Calculate time-independent and time-dependent expectation values such as currents, densities, electron addition and removal energies and the total energy of the system.
Calculate the spectral function which gives access to the local density of states.
Describe dissipative processes and memory effects in transport that occur due to electronelectron interactions and coupling of electronic to nuclear vibrations.
In a series of papers [102, 103, 105, 104] Caroli et al presented a general formalism for the
calculation of the current through an interacting system. An exact expression for the current
19
was later reformulated by Meir and Wingreen [120] in terms of the systems NEGF and the
self-energies which represent the effects of the external baths on the system.
The NEGF formalism is widely used nowadays to describe electron and hole transport in a
large variety of devices and materials, such as: III V resonant tunnel diodes [121, 122, 123,
124, 125, 126], electron waveguides (i.e., electrons in 2D) [127], Silicon tunnel diodes [128, 129],
carbon nanotubes [130, 131, 132, 133], metal wires [92, 134], organic molecules [135, 136, 137,
138, 139, 140, 141, 142], and magnetic leads [143, 144].
Physics that have been included are open-system boundaries [102], full-band structure [145,
146, 128, 129], the self-consistent Hartree potential [123, 147], exchange-correlation potentials
within a density functional theory (DFT) approach [148, 138, 149, 92], photon absorption and
emission [150, 151, 54, 152], energy and heat transport [153, 45], acoustic, optical, intra-valley,
inter-valley, and inter-band phonon scattering [125, 126, 154, 128, 129, 151], single-electron
charging and nonequilibrium Kondo systems [155, 156, 157, 158, 112], shot noise [159, 160, 161],
alternate Current [121, 162, 163, 164], and transient response [165].
Excluding simple noninteracting cases, the calculation of the systems NEGF required to
obtain the current (or any other observable) is far from trivial. Most applications are based
on a perturbation expansion (and diagrammatic techniques) to obtain the systems NEGF.
Alternatively, one can use the equations-of-motion (EOM) approach, which allows to deduce
the systems NEGF by deriving the corresponding equations-of-motion.
This chapter continues with a brief description of the NEGF [117]. We shall focus on the
equations-of-motion technique, outline its advantages and disadvantages, and elaborate on the
single-particle NEGF symmetry breaking [166] due to the way this technique is carried out.
We show that this symmetry/relation breaking can lead to solutions which are not physical
and suggest a scheme to restore it [166]. Finally, we formulate a couple of closures, one of
which outperforms the commonly used closures found in the literature, for the EOM of the
NEGF for the double Anderson model and use the symmetric-EOM technique also developed
by us to calculate the current and differential conductance through the double quantum-dot
array [114, 167] (the double Anderson model).
2.2
Nonequilibrium Green Functions
In nonequilibrium theory one usually divides the full Hamiltonian into three parts:
ex (t) = H
+H
ex (t) ,
(t) = H
+ V +H
H
| 0{z }
(2.2.1)
20
0 is the single-particle, noninteracting part, V contains all the interactions (electronwhere H

ex (t) contains all external perturelectron, electron-phonon, impurity scattering, etc.) and H
bations (which may be time-dependent) driving the system out of equilibrium.
A standard device is to assume that the external perturbations are turned on at a certain
ex (t) = 0 for t < t0 ,
time t0 , while prior to that time, the system is at equilibrium, i.e., H
and is described by the thermal equilibrium density operator which is a function of the time
independent interacting Hamiltonian H:
1
= e H , Z = Tr e H .
Z

(2.2.2)
Before introducing the single-particle, nonequilibrium Green function, which is a correlation

function between two Fermion/Boson field operators at different locations and times, we will
formulate the nonequilibrium problem. The task at hand, is the calculation of the expectation
value of an observable at time t > t0 (associated to the quantum mechanical operator O):
(t) = Tr O
H (t) ,
O
(2.2.3)
where the subscript H indicates that the time dependence of the Heisenberg operator is governed
by the full Hamiltonian (equation (2.2.1)) and is given in equation (2.2.2).
The general plan of attack is to transform the extremely complicated time dependence of
(t) to a much simpler form, specifically, have its time evolution governed by the
the operator O
0 . Furthermore, to allow a perturbation expansion and the use of
noninteracting Hamiltonian H
Wicks theorem [168, 169] or the Feynman diagrams [169, 170, 118], which are just a graphical
way representing the results of Wicks decomposition, one also needs to transform the density
operator such that it becomes a single-particle density operator. This is usually achieved by
two transformations [115]. The first transformation is due to the following identity:
H (t) = Sex (t0 , t) O
H (t) Sex (t, t0 ) ,
O
with
and
i Ht
H (t) = e hi Ht
Oe h ,
O
(
"
#)
i t
ex
ex
S (t, t0 ) = T exp
d HH ( ) ,
h
t0
i
ex
ex (t) = e hi Ht
H
H (t) e h Ht .
H
(2.2.4)
(2.2.5)
(2.2.6)
(2.2.7)
The time-order operator T { } in equation (2.2.6) equals either the chronological time-order
operator, Tf { }, which moves the operator with the later time to the left, if t > t0 , or the
21
anti-chronological time-order operator, Tb { }, which moves the operator with the later time
to the right.
At this point the time-contour (proposed by Schwinger [83] and later by Keldysh [84] and
Craig [171]) and contour-ordered quantities [82] are introduced. By joining the two exponentials
of equation (2.2.4) we get:
H (t) = Sex (t0 , t) O
H (t) Sex (t, t0 ) ,
O
#)
#)
(
"
(
"
t
i t0
i
ex
ex
( ) O
( ) ,
H (t) Tf exp
d H
d H
= Tb exp
H
H
h
h
t
t0
(
" t
#)
0
i
ex
H ( ) O
H (t) ,
d H
= TC exp
h
t0
(
"
#)
i
ex
H (t) .
= TC exp
d HH ( ) O
(2.2.8)
h
C
The integral in the exponent is now a contour integral along the contour C (depicted in figure 2.2.1) and TC { } is the contour time-order operator, which moves operators with times
further along the time-contour C to the left. By defining:
(
"
i
SCex = TC exp
h
#)
ex ( )
d H
H
(2.2.9)
equation (2.2.3) can be rewritten as:

H (t) = Tr TC Sex O
hO (t)i = Tr O
C H (t)
h
oi
(2.2.10)
In order to define the fundamental object of nonequilibrium many-body theory, i.e. the
NEGF, one extends the notion of the time variable (t) to contour time variables ( ). We
time axis
Figure 2.2.1: The time-contour C, starts and ends at t0 with contour variable 1 on the forward/upper branch
(C+ ) and 2 on the backward/lower branch (C ). By definition,
lying on the backward branch comes
n any point o
(2 ) A (1 ). The minus sign

after a point lying on the forward branch, meaning us, TC A (1 ) B (2 ) = B
originates from the permutation of two Fermion operators.
not in passing that the extension is not unique as a time t can be mapped into a contour
time variable on the forward or the backward branches. Finally, the contour-ordered Green
22
function is given by:

G (r2 , 2 , r1 , 1 ) =
oE
i D n
(r1 , 1 ) ,
TC H (r2 , 2 )
H
h
(2.2.11)
(r, ) /
(r, ) is the annihilation/creation field operator which annihilates/creates
where
a particle at position r and time . Henceforth, our discussion will only involve Fermion field
operators.
After applying the transformation described we get:
ex
H (r2 , 2 ) H (r1 , 1 )
i TC SC
D
n
oE
G (r2 , 2 , r1 , 1 ) =
h
TC SCex
oE
ex
H (r2 , 2 ) H (r1 , 1 )
i Tr TC SC
h
n
oi
=
h
Tr TC SCex
oi
(2.2.12)
The time-contour C may be elongated so that it would run beyond the latest time (t1 or t2 );
usually the contour is elongated to t . In this form the contour-ordered Green function (GF)
does not admit Wicks decomposition [172] and the use of Feynman diagrams, since the field
operators and the density operator depend on the (still complicated) interacting Hamiltonian
and not on the noninteracting Hamiltonian H
0 . To circumvent this problem one more
H
transformation is performed. The full derivation can be found in Ref. [82], but basically one
uses the identity:
=
Z0 v
0 S (i, 0) ,
Z
(2.2.13)
1 H 0
0 =
e
, Z0 = Tr e H0 ,
Z0
(
"
#)
i
i
v
S (i, 0) = Ti exp
dt VH0 (t) ,
h
0

with
i
i
VH0 (t) = e h H0 t V e h H0 t .
(2.2.14)
(2.2.15)
(2.2.16)
The operator Ti is the time-order operator, ordering the operators along the imaginary-time
segment [i, 0] (see figure 2.2.2 for more details) such that the time closest to i is to the
23
left. Putting it all together one gets:

ex
v
H0 (r2 , 2 ) H0 (r1 , 1 )
i Tr 0 TCi SH0 ,C SCi
oi
h
n
G (r2 , 2 , r1 , 1 ) =
ex S
v
h
Tr 0 TCi SH
0 ,C Ci
oi
H (r2 , 2 )
(r1 , 1 )
i Tr 0 TCi SCi
0
H
h
n
oi 0
=
h
Tr 0 TCi SCi
h
where
oi
(r) e hi H 0 ,
H (r, ) = e hi H 0
0
(r, ) = e hi H 0
(r) e hi H 0 ,
(2.2.17)
(2.2.18)
H0
the contour time-order TCi operator, orders the operators along the contour Ci , and
#)
i
ex
H ( ) ,
d H
= TC exp
0
h
C
#)
(
"
i
d VH0 ( ) ,
= TCi exp
h
Ci
ex
v
= SH
,C SC ,
(
ex
SH
0 ,C
SCv i
SCi
"
(2.2.19)
(2.2.20)
(2.2.21)
with
ex ( ) = e hi H 0 H
ex (t) e hi H 0 .
H
H0
(2.2.22)
The contour-ordered Green function in equation (2.2.17) accepts Wicks decomposition and
the perturbation expansion is possible. As already mentioned at the beginning of section 2.2,
time axis
Figure 2.2.2: The Kadanoff-Baym contour Ci , starts at t0 and ends at t0 i, with contour variable 1 on
the forward branch and 2 on the backward branch. By definition any point on the vertical branch comes after
the points on either of the horizontal branches, thus, TCi {A (1 ) C (3 ) B (2 )} = C (3 ) B (2 ) A (1 ).
all interactions are present in V . If one does not care about initial correlations (which is
a conceivable assumption when studying steady-state transport for example), one can take
24
the limit t0 . With reference to the equilibrium theory, interactions are switched on
=H
0 . Under this
adiabatically, such that at t0 = the system is not interacting and H
assumption = 0 , therefore, the transformation in equation (2.2.13) is not called for and the
addition of the imaginary segment to the time-contour C is not required [173]. The resulting
time-contour is known as the Schwinger-Keldysh time-contour, which we will denote by C (see
figure 2.2.3). In this regard, the contoured-ordered Green function takes a somewhat simpler
form:
n
oE
iD
H (r2 , 2 )
(r1 , 1 ) ,
(2.2.23)
G (r2 , 2 , r1 , 1 ) = TC
H
h
which can be reformulated in the interaction representation:

H (r2 , 2 )
(r1 , 1 )
i Tr 0 TC SC
0
H
h
n
oi 0
G (r2 , 2 , r1 , 1 ) =
Tr 0 TC SC
h
oi
(2.2.24)
with
SC = SCex SCv ,
(
"
i
SCex = TC exp
h
"
(2.2.25)
#)
i
SCv = TC exp
h
ex ( )
d H
H0
(2.2.26)
,
#)
d VH0 ( )
(2.2.27)
In this study we are interested in steady-state transport, therefore, in all that follows, we
confine ourselves exclusively to the Keldysh approach [174], i.e., we neglect initial correlations
and work on the Schwinger-Keldysh contour C . As already mentioned, by extending the time
time axis
Figure 2.2.3: The Schwinger-Keldysh contour C . When initial correlations are neglected, one ignores the
imaginary segment of the time-contour Ci and elongates the time-contour C from t0 = to t = and back.
variables into contour time variables, the uniqueness of these variables is lost. Such being the
case one should keep track of which branch is in question. With two time variables that can
25
be located on either one of the two branches, four real-time Green functions are introduced:
H (r2 , t2 )
(r1 , t1 )
GT (r2 , t2 , r1 , t1 ) = hi Tf
H
D
E
i
<
G (r2 , t2 , r1 , t1 ) = + h H (r1 , t1 ) H (r2 , t2 )

D
E
H (r2 , t2 )
(r1 , t1 )
G> (r2 , t2 , r1 , t1 ) = i
D
oE
H
h
D n
oE
i
T
(r1 , t1 )
G (r2 , t2 , r1 , t1 ) = h Tb H (r2 , t2 )
H
1 , 2 C+ ,
1 C , 2 C+ ,
1 C+ , 2 C ,
1 , 2 C ,
(2.2.28)
where GT is the time-ordered Green function, GT is the anti-time-ordered Green function, G<
is the lesser Green function and G> is the greater Green function. As opposed to its equilibrium
counter-part, the single-particle NEGF is a 2 2 tensor [175]:
GT (r2 , t2 , r1 , t1 )
G (r2 , 2 , r1 , 1 ) =
G>
(r2 , t2 , r1 , t1 )
G<
(r2 , t2 , r1 , t1 )
.
GT (r2 , t2 , r1 , t1 )
(2.2.29)

Usually one defines two other real-time Green function, the retarded Green function GR and

the advanced Green function GA :

T
<
GR
(r2 , t2 , r1 , t1 ) = G (r2 , t2 , r1 , t1 ) G (r2 , t2 , r1 , t1 )
Dn
oE
i
H (r2 , t2 ) ,
(r1 , t1 ) ,
= (t2 t1 )
H
h
T
>
GA
(r
,
t
,
r
,
t
)
=
G
2 2 1 1
(r2 , t2 , r1 , t1 ) G (r2 , t2 , r1 , t1 )
Dn
oE
i
H (r2 , t2 ) ,
(r1 , t1 ) ,
(t1 t2 )
=
H
h
(2.2.30)
(2.2.31)
where { , } is the anti-commutator. With these definitions, the single-particle NEGF tensor
can also be presented as follows [84]:
GR (r2 , t2 , r1 , t1 ) G<
(r2 , t2 , r1 , t1 )
.
G (r2 , 2 , r1 , 1 ) =
(r
,
t
,
r
,
t
)
0
GA
2
2
1
1
(2.2.32)
These real-time correlation functions are actually what is required in order to evaluate the
different observables. For example, the retarded Green function can be used to calculate the
response of the system at time t2 to an earlier perturbation of the system at time t1 and is
proportional to the local density of states, while the lesser Green function is also known as the
particle propagator and plays the role of the single-particle density matrix.
While in equilibrium GR , GA , G< and G> are all connected via the fluctuation-dissipation
theorem [176] and only one Green function is needed to describe a system, this is not longer the
case in nonequilibrium situations. Nevertheless, one easily observes that the real-time NEGFs
26
are not all independent and the following relations and symmetries hold by definition:
GR
(r2 , t2 , r1 , t1 ) =
GA
(r1 , t1 , r2 , t2 )

G
(r2 , t2 , r1 , t1 ) = G (r1 , t1 , r2 , t2 )
,

A
>
<
GR
(r2 , t2 , r1 , t1 ) G (r2 , t2 , r1 , t1 ) = G (r2 , t2 , r1 , t1 ) G (r2 , t2 , r1 , t1 ) .
(2.2.33)
In steady state the NEGF is a function of the time difference t = t2 t1 , thus, these relations
can be rewritten in Fourier space:
GR
(r2 , r1 ; ) =
GA
(r1 , r2 ; )

,

G
(r2 , r1 ; ) = G (r1 , r2 ; )
A
>
<
GR
(r2 , r1 ; ) G (r2 , r1 ; ) = G (r2 , r1 ; ) G (r2 , r1 ; ) .
(2.2.34)
Now that we have defined the NEGF, the question is how do we obtain this object? in other
words how do we calculate these correlation functions? most applications are based on perturbative diagrammatic techniques which will not be discussed here1 . Alternatively, one can
use the equation-of-motion (EOM) approach, which allows to deduce the systems Green function by deriving the corresponding equations of motion. We elaborate on the EOM technique
shortly.
2.2.1
Dyson equation
A very important concept is the noninteracting single-particle Green function, usually denoted
by G0 or g. It is defined as:
G0 (r2 , 2 , r1 , 1 ) =
oE
n
iD
H (r2 , 2 )
(r1 , 1 ) ,
TC
0
H0
h
(2.2.35)
ex (t) = 0. Combined with a self energy [80, 169,

and it is obtained by assuming that V = 0 and H
118, 117], , which represents the contribution to the particles energy, or effective mass, due
1 Derivations
of perturbative diagrammatic techniques can be found in Refs [169, 118, 117].
27
to interactions between the particle and the system it is part of, one obtains Dysons equation:

G r, , r ,
= G
= G
r, , r , +
d1d2 G0 (r, , 2) (2, 1) G 1, r0 , 0 ,

d1d2 G (r, , 2) (2, 1) G0 1, r0 , 0 ,

r, , r , +
(2.2.36)
where 1 (r1 , 1 ) and d1 dr1 C d1 .

Dysons equation can be easily expressed in matrix notation (see equations (2.2.32)):
G r, , r ,
= G0 r, , r , +
d1d2 G0 (r, , 2) (2, 1) G 1, r0 , 0 ,

d1d2 G (r, , 2) (2, 1) G0 1, r0 , 0 ,
= G0 r, , r , +
(2.2.37)
only now the contour integrals are transformed into real-time integration, i.e., d1 dr1 dt1 .
Of course, usually, one does not know the exact form of the self energy, and approximated expressions are used to approximate the Green function.
2.2.2
Analytic continuation and Langreth rules
When writing down the EOM for the NEGF or when considering the different terms in the
perturbation expansion we encounter contour quantities with the following structures:
F1 (2 , 1 ) =
A (2 , ) B (, 1 ) d,
(2.2.38)
F2 (2 , 1 ) =
A (2 , ) B , 0 C 0 , 1 d 0 d,
(2.2.39)
(2.2.40)
(2.2.41)
F3 , 0 = A , 0 B , 0 ,
F4 , 0 = A , 0 B 0 , ,
where we suppressed all variables but the temporal ones for clarity and brevity. The contour
integration and contour variable are not practical for calculations and one can replace the
contour time variables and the contour integrals with real-time variables and integrals (analytic
continuation [115]) using Langreth theorem [81, 115, 117]. The rules provided by Langreth
theorem are summarized in table 2.1. For example, using Langreth rules and Dysons equation
28
for the NEGF, the equations for the retarded and lesser Green functions are given by:
r, t, r , t
GR
0
r, t, r , t +
GR
0
R
R
d1d2 GR
1, r0 , t0 ,
0 (r, t, 2) (2, 1) G
0 0
d1d2 GR (r, t, 2) R (2, 1) GR
0 1, r , t ,
r, t, r , t +
(2.2.42)
<
r, t, r , t
G<
0
R
<
d1d2 GR
1, r0 , t0
0 (r, t, 2) (2, 1) G
r, t, r , t +
<
A
d1d2 GR
1, r0 , t0
0 (r, t, 2) (2, 1) G
A
A
1, r0 , t0 ,
d1d2 G<
0 (r, t, 2) (2, 1) G
(2.2.43)
<
r, t, r , t
= G<
0
+
+
r, t, r , t +
0 0
d1d2 GR (r, t, 2) R (2, 1) G<
0 1, r , t
0 0
d1d2 GR (r, t, 2) < (2, 1) GA
0 1, r , t
0 0
d1d2 G< (r, t, 2) A (2, 1) GA
0 1, r , t ,
(2.2.44)
where d1 dr1 dt1 .

In the next subsection we describe the EOM technique vastly used to calculate the NEGF.
2.2.3
Equations-of-motion
2.2.3.1
Overview
The EOM approach allows the deduction of the systems NEGF by deriving the corresponding
equations-of motion [177, 178, 179, 180, 181]. The main advantages of the method are its
simplicity and relatively mild scaling with the number of electrons, under simple truncation
schemes [182]. In light of its simpleness, the technique has been used extensively to describe
transport phenomena such as the Coulomb blockade [183] and the Kondo effect [179, 184,
111, 185, 186], providing qualitative and in some cases quantitative results. When applied to
interacting open systems, the EOM for the NEGF gives rise to an infinite hierarchy of equations
of higher-order NEGF.
A well-known approximation procedure is to truncate this hierarchy, thus introducing a
mean-field like description to some observables. These equations for the NEGF then need to be
solved self-consistently for the resulting closed set of equations. In general, closures cannot be
29
Contour
F1 (2 , 1 ) =
F2 (2 , 1 )
Real Axis
C A (2 , ) B (, 1 ) d
C A (2 , )
B (, 0 ) C ( 0 ,
) d 0 d
F1R (t2 , t1 )
F1< (t2 , t1 )
=
+
F2R (t2 , t1 )
F2< (t2 , t1 )
=
+
+
F4
A (, 0 ) B ( 0 , )
R (t
2 , t) B
R (t, t
1 ) dt
R
A (t2 , t) B < (t, t1 ) dt
<
A
A (t2 , t) B (t, t1 ) dt
R (t
2 , t) B
R (t, t0 ) C R (t0 , t
0
1 ) dt dt
R
A (t2 , t) B R (t, t0 ) C < (t0 , t1 ) dt0 dt
AR (t2 , t) B < (t, t0 ) C A (t0 , t1 ) dt0 dt
<
A
0
A 0
0
A (t2 , t) B (t, t ) C (t , t1 ) dt dt
F3R (t, t0 )
= A< (t, t0 ) B R (t, t0 ) + AR (t, t0 ) B < (t, t0 )

+ AR (t, t0 ) B R (t, t0 )
F3< (t, t0 )
= A< (t, t0 ) B < (t, t0 )
F4R (t, t0 )
A< (t, t0 ) B A (t, t0 ) + AR (t, t0 ) B < (t, t0 )
F4< (t, t0 )
A< (t, t0 ) B > (t, t0 )
F3 (, 0 ) = A (, 0 ) B (, 0 )
(, 0 )
Table 2.1: Summary of Langreth rules. To obtain the rules for the advanced Green function, just replace the
R superscript with A. For the greater Green function replace the superscript < with >.
improved systematically. Furthermore, it is often difficult to assess, deductively, the accuracy

of a given closure.
In what follows we introduce the EOM technique and exemplify it by calculating the NEGF
of the single resonant level model. Subsequently we show that while a closure of the set of
equations can always be obtained, it is not clear a priori whether the resulting single-particle
NEGF fulfills symmetry relations that this object must obey (equations (2.2.33) and (2.2.34)) by
definition. This failure can lead to solutions which are not physical, such as complex occupation
of levels [187, 166] and even finite currents at zero bias [166]. Additionally, we propose an
approach to fix this deficiency by imposing a set of rules to reconstruct a NEGF that fulfill
these basic symmetry relations.
We regard this symmetrization procedure as one more (necessary) approximation on top
of the approximations used to close the set of equations, which ensures that the resulting
approximated NEGF obeys all symmetry relations of equations (2.2.33) and (2.2.34). We name
the complete procedure of EOM plus symmetry-restoration as the symmetric-EOM approach.
Illustrations are given for the Anderson model [110] at the Kondo regime and for the double
Anderson model [114].
30
2.2.3.2
The equations-of-motion method
The EOM for the contour ordered Green function [180, 188] is obtained from the Heisenberg
h
i
(t) , A (t) + A (t) , where H
(t) = H
0 + V (t).
EOM for a Heisenberg operator ddt A (t) = hi H
t
h
i
B
is the commutator, H
0 is the noninteracting quadratic part of the Hamiltonian
Here A,
and all interactions are in V (t). To make connection with the model Hamiltonians that would
be presented shortly, we rewrite the contour ordered Green function in terms of the electronic
creation and annihilation operators. This is achieved by expanding the field operators in some
basis {k (r)}, such that:
H (r, ) =
ai ( ) i (r) ,
(2.2.45)
(r, ) =
ai ( ) i (r) .,
(2.2.46)
where the operator ai ( ) annihilates a the particle occupying state i (r) at time , while ai ( )
creates one.
Using this expansion, the NEGF can be rewritten as:
G (r2 , 2 , r1 , 1 ) =
i (r2 ) j (r1 ) Gij (2 , 1 ) ,
(2.2.47)
ij
with
n
oE
iD
TC ai (2 ) aj (1 ) ,
h
D
E
i
= C (2 1 ) ai (2 ) aj (1 )
h
D
E
i
+ C (1 2 ) aj (1 ) ai (2 ) ,
h
Gij (2 , 1 ) =
(2.2.48)
where C (2 1 ) is the contour step function, i.e.,
2 > 1 on the contour
otherwise
C (2 1 ) =
(2.2.49)
31
The EOM [180] for Gij (2 , 1 ) can be written as:

*
)+
Dn
oE
i
h
Gij (2 , 1 ) = C (2 1 ) ai , aj + TC
ai (2 ) aj (1 ) ,
2
2

h
Dn
i
oE
i
= C (2 1 ) ai , aj + TC
H (2 ) , ai (2 ) aj (1 ) ,
h

h
i
i
= C (2 1 ) ij + TC
H0 , ai (2 ) aj (1 )
h

h
i
i
V (2 ) , ai (2 ) aj (1 ) .
(2.2.50)
+ TC
h
0 is the time-independent, noninteracting, quadratic Hamiltonian, the term H

0 , ai (2 )
Since H
P
equals p p ap (2 ), where p s are parameters of the noninteracting Hamiltonian, such as the
kinetic energy. Using this and the definition in equation (2.2.48) we can rewrite:
h
TC
i
ih
H0 , ai (2 ) aj (1 )
h

X
p
p Gpj (2 , 1 ) .
(2.2.51)
The term TC hi V (2 ) , ai (2 ) aj (1 ) results in higher order Green functions (depending

on the form of V (t), these can be 2-, 3- or even higher particle Green functions). We shall
denote those as G (2 , 1 ). Finally we can rewrite:
D
n h
oE
p Gpj (2 , 1 )
i Gij (2 , 1 ) = C (2 1 ) ij +
i
h
2
p6=i
X
p
Tp Gp (2 , 1 ) ,
(2.2.52)
where once again Tp s are parameters of the interaction Hamiltonian V (t) , such as the
Coulomb repulsion energy. Define the Green function gi (2 , ):

i
h
i gi (2 , ) = C (2 ) ,
2
(2.2.53)
and equation (2.2.52) now takes the form:
Gij (2 , 1 ) = gi (2 1 ) ij +
X
p
Tp
X
p6=i
p
gi (2 ) Gpj (, 1 ) d
gi (2 ) Gp (, 1 ) d.
(2.2.54)
Depending on the Hamiltonian, the newly generated Green functions (Gpj and Gp ) can
involve lead (bath) operators as well as system operators. Except for very simple cases, where
an exact closure can be obtained (see sub-subsection 2.2.3.3), writing the EOM for Gpj (2 , 1 )
32
and Gp (2 , 1 ) will produce new and/or higher order Green functions that need to be evaluated. This leads (in principle) to an infinite/ intractable set of equations. The idea of the
EOM method is therefore, to truncate this hierarchy of equations making a mean-field like approximation for the higher-order Green functions through lower order functions. This is the
Achilles heel of this method as there is no systematic way to close the equations and no theory
to back-up such a procedure. Usually the approximations have physical meaning within the
regime of the problem at hand [189, 190, 111, 167].
As an illustration of the method we next calculate the NEGF of a single resonant level [109,
62] described in subsection 1.3.1.
2.2.3.3
The single resonant level model - EOM treatment
Define the systems contour ordered Green function:

G , 0 =
n
oE
iD
TC d ( ) d 0
.
h
(2.2.55)
The equation of motion of G (, 0 ) is:

d

d ( ) d 0
i
h G , 0 = C 0 + TC
d

n
oE
D
i
= C 0 TC d ( ) d 0
h
n
oE
i X X D
tk TC ck ( ) d 0
h
K{L,R} kK
*
)+
= C 0 + G , 0 +
tk Fk , 0 ,
(2.2.56)
K{L,R} kK
oE
where Fk (, 0 ) = hi TC ck ( ) d ( 0 ) , is a new Green function generated in the procedure.

To obtain G (, 0 ) we now derive the EOM of Fk (, 0 ):

Dn
oE

d
i
h Fk , 0 = C 0
ck , d + TC
ck ( ) d 0
d
n
oE
D
n
oE
i
i D
tk TC d ( ) d 0
= k TC ck ( ) d 0
h
h

0
0
= k Fk , + tk G , .
*
)+
(2.2.57)
The last equation can be rearranged:

!

i
h k Fk , 0 = tk G , 0 .
(2.2.58)
33
Define the leads noninteracting Green function through the following equation:
!
i
h k gk ( 1 ) = C ( 1 ) ,
(2.2.59)
so equation (2.2.58) takes the form:
Fk ,
gk ( 1 ) tk G 1 , 0 d1 .
(2.2.60)
At this point the equations close. To finish our derivation, substitute equation (2.2.60) in
equation (2.2.56):
!

X
X
i
h G , 0 = C 0 +
K{L,R} kK
|tk |2 gk ( 1 ) G 1 , 0 d1 .

(2.2.61)
Define the levels noninteracting Green function through the following equation:
!
i
h g ( 1 ) = C ( 1 ) ,
(2.2.62)
so finally the equation for the systems NEGF is:
G ,
= g +
K{L,R} kK
g ( 2 ) |tk |2 gk (2 1 ) G 1 , 0 d1 d2 .

X
C
(2.2.63)
The Green function G (, 0 ) contains the information on the resonant level in an explicit
way (via g ( 0 )), while the effect of all other (infinitely many) levels of the left and right
leads appear only in the sum:
X
|tk |2 gk (2 1 ) ,
(2.2.64)
K{L,R} kK
which we recognize as the self energy:

(2 1 ) =
|tk |2 gk (2 1 ) .
(2.2.65)
K{L,R} kK
Applying Langreth rules one can obtain the retarded, advance, lesser and greater Green functions:
34

R/A
t, t = g
R/A
tt +
g R/A (t t2 ) R/A (t2 t1 ) GR/A t1 , t0 dt1 dt2 .

(2.2.66)
t, t
g R (t t2 ) R (t2 t1 ) G t1 , t0 dt1 dt2

tt +

+
g R (t t2 ) (t2 t1 ) GA t1 , t0 dt1 dt2

g (t t2 ) A (t2 t1 ) GA t1 , t0 dt1 dt2 .
+
= g
(2.2.67)
In steady state the NEGF is a function of one time variable [115] (the time difference). In
this case it is simpler to express the NEGFs in Fourier space. To simplify the notation we
denote the Fourier transform of G (t) as G (), i.e., functions with an argument are Fourier
transforms of their time-domain counterparts. The resulting equations are:
GR/A () = g R/A () + g R/A () R/A () GR/A () ,
(2.2.68)
G () = g () + g R () R () G ()
+g R () () GA () + g () A () GA () ,
(2.2.69)
1
, 0+ ,
h
i
(2.2.70)
Here:
g R/A () =
with the + signs corresponding to the retarded Green function,

D
g < () = 2i (
h ) d d

E
0
E
g > () = 2i (
h ) 1 d d
R/A
R/A
R/A
() = L () + R () =
<
< () = <
L () + R () = 2i
(2.2.71)
|tk |2
, 0+ ,
h
k i
K{L,R} kK
X
(2.2.72)
(2.2.73)
|tk |2 (
h k ) f (k K ) ,
(2.2.74)
K{L,R} kK
> () = 2i
|tk |2 (
h k ) (1 f (k K )) ,
(2.2.75)
K{L,R} kK
where f () is the Fermi-Dirac distribution and d d
E
0
represents the occupancy of the resonant

35
level in steady state.

An exact expression for the stationary current through a system coupled to large noninteracting metallic leads in terms of the systems Green function can be derived [120]:
ie
I =
4
h
i
d Tr (f ( L ) L () f ( R ) R ()) GR () GA ()
h
+ Tr (L () R ()) G< ()

io
(2.2.76)
where L () = 2Im R
L () , or equivalently
e
I =
2h
dTr
h
<
>
>
>
<
<
L () R () G () L () R () G () .
(2.2.77)
The resonant level model serves as a great toy model, nevertheless, it is too simple as it
involves no interactions. The NEGF and the self-energies can be calculated exactly, and no
approximations are needed. As already mentioned when interactions are involved, one needs
to truncate the infinite/intractable set of equations and decouple the higher order correlation
functions in terms of the lower order ones, the resulting single-particle NEGF does not necessarily obey the symmetries and relations of equation (2.2.34). To demonstrate this we turn
to the Anderson model at the Kondo regime and the double Anderson model described in
subsection 1.3.2 and subsection 1.3.3 respectively.
2.2.4
NEGF symmetry breaking within the EOM technique
2.2.4.1
The Anderson model
Following the derivation in Refs. [158, 115] we define the following contour ordered Green
function:
n
oE

iD
,
(2.2.78)
G t, 0 = TC d ( ) d 0
h
n
oE

iD
,
(2.2.79)
G , 0 = TC n ( ) d ( ) d 0
h
where
is the opposite spin of . Various approximate decoupling procedures can be applied to
the many-particle Green functions [182] that are generated during the procedure of writing the
EOM for the single-particle NEGF. Here we follow the approximation scheme used in Ref. [158]:
1. All electronic correlations containing at most one lead operator, are not decoupled and
their EOM are calculated.
36
2. Higher order Green functions involving (opposite) spin correlations in the leads are set to
zero, that is:
n
oE
iD
TC Ak ( ) Bq ( ) Cs ( ) d 0
=0
h
where A and B are either creation or annihilation operators of the leads and Cs is a
system operator (d or d with spin s).
3. The remaining higher order Green functions involving lead and system degrees of freedom
are decoupled such that

F , 0 =
n
oE

iD
TC ck ( ) d ( ) cq ( ) d 0
= kq f (k K ) G , 0 .
h
Given these approximations the EOM for G (, 0 ) can now close (full derivation can be found
in Ref. [115] pages 172-176). The resulting EOM (in steady state) are:
R/A
GR/A
() + gR/A () U GR/A () ,
() = g
R/A
R/A
R/A
GR/A () = g2 () hn i g2 () 1
() GR/A
() ,
<
R
G
() = g () + g () U G () + g () U G () ,
(2.2.80)
(2.2.81)
(2.2.82)
G () = g2
() hn i g2
() R
1 () G ()
R
A
A
A
g2
()
1 () G () g2 () 1 () G () .
(2.2.83)
To obtain these equations, one should write the EOM of the different two-times correlation
functions and close them using the aforementioned assumptions. Next, Langreth rules should
be applied to map the contour integrals to real-time integration and the contour Green function
into one of the real-time Green functions (e.g., the retarded or the advance Green function).
Then, one assumes steady state, i.e., the Green functions are no longer two-times functions but
a single time variable functions. Subsequently transformation to Fourier space is carried out.
To complete the derivation the following definitions should be used:
i
h
hn i =
2
gR/A () =
G<
() d,
1
R/A
h
0
()
(2.2.84)
(2.2.85)
37
R/A
g2 () =
R/A
h
U 4
()
(2.2.86)
A
g () = gR ()
0 () g () ,
(2.2.87)
R
A
g2
() = g2
()
4 () g2 () .
(2.2.88)
Also,
R/A
R/A
R/A
() = 0L () + 0R () =
|tk |2
, 0+ ,
h
k
K{L,R} kK
X
(2.2.89)
is the exact retarded/advanced self-energy for the noninteracting case. The self-energies due to
R/A
R/A
the tunneling of the
electron are 1 () and 3 ():
R/A
R/A
R/A
() = jL () + jR () ,
=
(j)
Ak |tk |2
X
k{L,R}
1
h
+ k U i
1
, j = 1, 3 and 0+
+
h
k + i
!
(2.2.90)
(1)
(3)
with Ak = f (k K ), Ak = 1. The lesser self energies are defined as in Ref. [186]:

<
<
<
j () = jL () + jR () = i fL () jL () + fR () jR () ,
(2.2.91)
Finally:
R/<
R/<
() = 0
R/<
() + 3
() .
(2.2.92)
With these definitions we are now ready to show that this set of equations break the sym
<
metry relation G<
() = (G ()) (A full derivation can be found in appendix A). In other
words, we are going to show that G<
() is not imaginary. In what follows we omit () for
brevity. Using equations (2.2.80) to (2.2.83) one gets:

R <
<
A A
A A
<
R
G<
i + g U P g2 hn
i 1 g
= g + g U P g2 hn
R < A A
< A
R
R
<
gR U P R g2
R
1 g + 1 g g U P g2 1 g
R R <
A
A
gR U P R g2
1 g U P A g2
hn i + A
1 g
A
R < A
A
gR U P R g2
1 g U P A g2
hn i + A
1 g
< A A
A
A
hn i + A
gR U P R g2
1 g U P A g2
1 g ,
(2.2.93)
38
where
P R/A =
1
R/A R/A R/A
1 + g2 1 g U
(2.2.94)
Applying the principle of reductio ad absurdum we assume G<

is imaginary. Since it must
hold for any real value of hn i between 0 and 1, we argue that the term:
<
A
A1 = gR U P R g2
hn i + g< U P A g2
hn i
R R <
A
gR U P R g2
1 g U P A g2
hn i
R < A
A
gR U P R g2
1 g U P A g2
hn i
< A A
A
gR U P R g2
1 g U P A g2
hn i ,
(2.2.95)
must be imaginary. Moreover, Since A1 must be imaginary for any value of U the term:
<
A
A2 = gR U P R g2
hn i + g< U P A g2
hn i ,
(2.2.96)
should be imaginary as well. Using the fact that U and hn i are real quantities and by definition
<
g2
and g< are imaginary, for A2 to be imaginary the following must hold:

<
A
Im gR P R g2
= Im P A g2
g< ,
(2.2.97)
or in other words, we demand that Re (A2 ) = 0. One can then show (appendix A) that, in fact,
the equality in equation ((2.2.97)) does not hold, namely, G<
() is not an imaginary function
<
and the relation G<
() = (G ()) is not satisfied. In turn, this implies that hn i (the
occupation number) is a complex number (see figure 2.2.4), which of course is not physical.
Following the same derivation one can show that G>
() is not an imaginary function either.
All the other relations given in equations (2.2.34)) are fulfilled.
If one is only interested in the Coulomb blockade regime, it is not necessary to go to the
level of approximation presented here (which is essential to obtain the Kondo effect). For the
Coulomb blockade regime one can turn to the approximation presented in Ref. [183]:
1. All electronic correlations containing at least one lead operator are neglected unless they
have the following form:
n
oE
iD
.
TC n ( ) ck ( ) d 0
h
In which case a mean-field approximation is applied:

D
n
oE
i
hn ( )i TC ck ( ) d 0
.
h
39
Re n
Im n
0.3
0.2
0.1
0
0
0.2
0.4
0.6
0.8
e/U
Figure 2.2.4: One of the main flaws of the EOM approach for the Anderson impurity model, if the symmetry
breaking of the lesser Green function, which leads to a complex occupation numbers. Here we plot the value of

n (occupation of the spin up electron) as a function of the bias voltage (e/U ). The most notable effect is

the appearance of an imaginary portion to n (red line) as the bias voltage is increased. To obtain the results
only the real part of hn i was used to converge the self-consistent equations for the NEGF (equations (2.2.80)
to (2.2.83)). Parameters used (in units of U ): L, = R, = 0.3, L, = R, = 0.05, = 0.2, = 0.2 and
= 4. The wide band limit approximation was used for the calculation of the self-energies.
This simple approximation does not violet the symmetry relations of the single-particle Green
function (see appendix A), but as pointed above, it does not reproduce the Kondo peaks at low
temperatures.
2.2.4.2
The double Anderson model
For the double Anderson model, described in subsection 1.3.3, we follow the derivation given
in Ref. [114], and define the following contour ordered Green functions:

0
G
, =
n
oE
iD
TC d ( ) d 0
,
h
(2.2.98)
n
oE
iD
TC ns ( ) d ( ) d 0
,
(2.2.99)
h
where , s = {, } and
is the spin opposite to , and = {, }. In what follows we will use
the naming convention summarized in table B.1 in appendix B. The approximations used in
Ref. [114] are:

0
Gs
, =
1. Neglect the simultaneous hopping of electron pairs to and from the system. i.e., neglect
40
qrst
qrst
0
0
0
all 2-particle Green functions Gqrst
(, ), Fk (, ) and Fk (, ).
2. Assume that

0
Fs
jk ,
n
oE
iD
TC njs ( ) ck ( ) d 0
h
D
n
oE
i

dt1 gk ( 1 ) TC d (1 ) njs (1 ) d 0
,
h
where njs (t) is the number operator of the electron on site j and spin s, and we assumed
k L, and
!
i
h k gk ( 1 ) = C ( 1 ) .
0
This is obtained by writing the EOM of Fs
jk (, ) and assuming that njs ( ) is constant,
njs ( ) = 0, which is the case in steady state. This assumption is equivalent to

i.e.
treating the coupling to the leads up to the second order with respect to tkm . It neglects
processes necessary to qualitatively capture the Kondo effect [191, 192, 193], yet results
are predicted to be reliable for temperatures above the Kondo temperature (TK ) [158, 60].
3. Higher order Green functions (3-particle) of the form

D
n
oE
hi TC nq ( ) nr ( ) d ( ) d ( 0 ) are decoupled to:
D
n
oE i
D
n
oE
i
hnr (t)i TC nq ( ) d ( ) d 0
.
hnq ( )i TC nr ( ) d ( ) d 0
h
h
Here , = {, } .
These approximations lead to the following equations, written in Fourier space, where ()
0
was omitted for brevity (for a brief summary of the derivation of the EOM of G
(, ) see
appendix B):

R/A
h
0
1
R/A
=
+ h G
G
+V
R/A
R/A

+ U G
+ V
G
R/A
R/A
,
(2.2.100)

R/A
G
h
U V

h G
R/A
n V
R/A
n 0
+ hn i V
G
R/A
1

+ V
G
R/A
(2.2.101)
41

R/A
G
R/A
0
hn i V
h
U n V
hn i + h

R/A
G
+ hn i
1

R/A
R/A
U G
+ V
G

(2.2.102)
R/A
G
R/A
0
hn i V
h
U n V
hn i + h
R/A
+ hn i
G
1

R/A
R/A
U G
+ V
G

(2.2.103)
R/A
R/A
h
n V
0
U hn i V

h G
R/A
+ n
E

U G
R/A
1
,
+ V
G
R/A
(2.2.104)
R/A
G
h
U hn i V
d d, + h G
+ n,
E
R/A
n V
0
1
R/A

,
R/A
R/A
+ V,
G
U G

,
(2.2.105)
R/A
R/A
h
hn i V
hn i 0
U V
D
E

n + h G
+ n
,
V
G
1
R/A
R/A

+ V
G
R/A
.
(2.2.106)
We now show
that giventhis set of equations, the symmetry relation
R
G
()
A
G
()
is not satisfied. For simplicity we prove that for the case where
42
Vijs = 0. Define
1
R/A
gi () =
u R/A
(gi
)
() =
u
n
gi
R/A
() =
R/A
h
i 0
1
()
(2.2.107)
R/A
h
i Ui 0
1
()
(2.2.108)
R/A
h
i Ui hni i 0
()
(2.2.109)
|tk |2
, 0+ ,
=
h k i
kL
R/A
0 ()
(2.2.110)
R/A
0 () =

2
tk
h k i
kR
, 0+ .
(2.2.111)
Simple substitutions now yield the following set of equations:

R
G
R 2 R
1 g
h g
1
R R
R
g
h g + g
U

R
R R
R
G
+ g
h g U G
,
(2.2.112)

R
G
"
u R
1 (g
)

u
n R 2
g
h
u R
u
h (g
) g
u R
u
n
(g
) h g
R
R
u
n
g
R

1
u R 2
u
n R
g h g
u
n
g
R
u R
u
hn i + (g
) g
hn i n U 1
hn i n U U
R
u
n
h g

1
u R 2
u
n R
g h g
1
R
(2.2.113)
43

R
G
"

u R
u
n R 2
h
g
g
1

u
n R 2 u R
g h (g )
1

D
E

2 u R u R u
n R
u
n R
g
g hn i n U U
h (g ) g

2

D
E

u R
u R
u R
u
n R
u
n R
) g
g
n + h (g
g
g

1 #
D
E

R
u R
h (g
) h
u
n
hn i n U 1 g
(2.2.114)
and

A
A A
1 g
h g
1

A
A A
U G
h g + g
g
A

A A
h g U G
+ g
A
,
(2.2.115)

A
G
"

u A
u
n A 2
g
g
h
u A 2
1 (g
) h
1
u
n A
g

1

D
E
2 u A u A u
n A
u
n A
h g (g ) g
g
n hn i U U

D
E

u
g
A
h
u
n
g
n hn i U

A
G
"
u A
u
n
1 (g
) g
u
n + g
u A 2
1 (g
) h
u A
u
n
(g
) h g
u
n
g
A 1
(2.2.116)

A

1
2
u
n A 2
u A
h 1 g
h g

1

D
E
2 u A u A u
n A
u
n A
h g (g ) g
g
n hn i U U

2

u A
u
n A
u
n A
u A u A
) hn i + h g
g
(g
(g ) g
#

D
E

1
u
n A 2
u A
,
(2.2.117)
n hn i U 1 g h g
Taking the conjugate of equation (2.2.115) and using the fact that,

A
R
gi
= gi
i {, } ,
(2.2.118)
44
we obtain:

A
R R
1 g
h g
1
R
R R
U
h g + g
g

A
R R
A
,
+ g h g U G
G

(2.2.119)
Comparing the result with equation (2.2.112) and since this should hold for any value of U , it
R
A

to be true, the following should

= G
is obvious that for the identity G
()
()
hold:

R R
A
R
R
= g
h g U G
,
(2.2.120)
g
U G
or equivalently:

R
G

R
h
= g

A
.
G
(2.2.121)
Using equations (2.2.113) and (2.2.117) and eliminating equal terms on both sides of the equality, we get:

u
n
L.H.S = h g
R
u
hn i + U g
R
u
n
h g
R
hn i n ,
(2.2.122)
R
R.H.S = g
h (hn i
2
R
R
R D

gu
g un
g un
n
+U h
hn i .
(2.2.123)
Note that by definition:

u R
(gi
) =
u
n
gi
R
u A
(gi
)

u
n
gi
Obviously

R
G
()

R
G

R
6 g
=
h
A
G ()
hence, the relation

=
approximations and truncation of the EOM.
i {, } ,
A

i {, } .

A
G
,
(2.2.124)
(2.2.125)
(2.2.126)
is not satisfied under these aforementioned
45
The same can be done (not presented) to show that:

and

R
G
()

()
G
6=
A
G
()

,
G ()
>
6= G
()
(2.2.127)
G
()
<
(2.2.128)
As a result the single-particle density matrix, which is obtain from the lesser Green function, is
not Hermitian (see figure 2.2.5), the occupation of levels is a complex number, and the value of
the stationary current calculated via equation (2.2.76) or via equation (2.2.77) yields different
results (figure 2.2.6).
real part
imaginary part
Coherences
0.1 0.2 0.3 0.4
0.1 0.2 0.3 0.4 0.5
e/U
e/U

Figure 2.2.5: The single-particle density matrix should be Hermitian, i.e.,

, or in other words
=
<

<

i
h
G () d. But since G
6=
Im
()
= Im and Re = Re with = 2

<
, this relation is violated. Right panel: the imaginary part of

G
(blue line) and (red
()
line) are plotted as a function of the bias voltage. Left panel: the real part of
(blue line) and
(red line) are

plotted as a function of the bias voltage. It is obvious that 6= . Parameters used for the simulations
in units of U = U = U are: L = L = R = R = 0.0025, R = R = L = L = 0, h = h = 0.25,
= 0.1,
V
= = 0.1, = = 0.175 and = 80. The wide band limit approximation was used for
the calculation of the self-energies.
In the following subsection we propose a symmetrization scheme that restores all the symmetries of the single-particle NEGF presented in equations (2.2.33) and (2.2.34).
2.2.5
NEGF symmetry restoration
The customary route to calculate the NEGF is as follows:
46
_
Stationary current (e/ h)
12
10
10
-2
Eq. (1)
Eq. (2)
8
6
4
2
0
0
0.1
0.2
e/U
0.3
0.4
Figure 2.2.6: I V curves calculated using equations (2.2.76) and (2.2.77) (blue line and red line respectively).
As can be clearly seen, calculating the current via the two different but equivalent formulas provide different
results, one of which is not physical (solid line) as the current is finite for = 0. The latter result suggests
that the Green function obtained through the EOM disobey the fluctuation dissipation relation. Parameters
L
R
R
R
R
L
used for the simulations in units of U = U = U are: L
= = = = 0.0025, = = =
L
= 0.25, V
= 0, h = h
= 0.1, = = 0.1, = = 0.175 and = 80. The wide band limit
approximation was used for the calculation of the self-energies.
1. Calculate the retarded Green function and use it to obtain the advanced Green function

R () ).
(by demanding GA
()
=
G
2. Calculate the lesser/greater Green function and symmetrize the lesser/greater to fulfill

.
the quantum Onsager relations [193], hence obeying G
()
=
G
()
In most applications of NEGF the advanced Green function is not directly calculated and thus,
the symmetry breakage does not always stand out. In fact, this common procedure restores the
relation between the advanced and retarded Green functions and between the lesser/greater
A
and their complex conjugate, but does not necessarily restore the relation GR
() G () =
<
G>
of the latter leads to violation of the fluctu () G (). It can be shown that violation

ation dissipation relation, G< = feq ( eq ) GR GA , at equilibrium [115].
This oversimplified procedure can result in different values for the currents depending on
how it is calculated, cf. equation (2.2.76) or equation (2.2.77). It may also lead to finite currents
at zero-bias voltage (see figure 2.2.6), which is physically incorrect.
In order to restore the symmetry relations that are imposed by the definitions of the Green
functions (equation (2.2.33)), we suggest the following procedure:
47
1. Calculate the retarded/advanced Green functions matrices GR /GA separately and use

R = 0.5 GR + GA
them to define new retarded/advanced Green functions matrices G

A = 0.5 GA + GR
and G

R .
= G

R /G
A to calculate the
2. Use the new retarded/advanced Green functions matrices G
lesser/greater Green functions matrices (G< /G>). Again, use
them to define new

= 0.5 G G
lesser/greater Green functions matrices G
.

>G
< and B = G
RG
A . Define
3. Calculate the two anti-Hermitian matrices A = G
the difference anti-Hermitian matrix C = A B, and redefine the retarded and advanced
R =G
R + C/2, and G
A =G
A C/2.
Green functions G
R, G
A, G
< and G
> ) obey all symmetry relations of equaThe resulting Green functions (G
tions (2.2.33) and (2.2.34) by construction. Note that if the original Green functions obeyed
the symmetry relations to begin with, our symmetrization procedure will not alter them in any
way.
We now return to our calculations for both the Anderson and double Anderson models. For
both models we use the closures described above. For each set of calculations we have applied
the above symmetrization scheme and compared the results to those obtained without restoring
symmetry.
2.2.5.1
The Anderson model: symmetric-EOM approach
The closure used in Ref. [158] is sufficient to describe the appearance of the Kondo resonances
at low temperatures, as seen in figure 2.2.7, where we plot the density of states as a function
of energy for several temperatures (all calculations are done with symmetry restoration). The
development of Kondo peaks in the density of states as the temperature decreases is clearly
evident, signifying a regime of strong correlations which is qualitatively captured by the simple
EOM approach when symmetry is restored.
D E
In figure 2.2.8 we plot the value of n as a function of the bias voltage with and without
symmetry restoration. The most notable effect is the appearance of an imaginary portion to
D E
n as the bias voltage is increased. By applying the symmetrization scheme proposed in

subsection 2.2.5 to the lesser Green function, we restore the relation G

() = G () .
This is sufficient to obtain a real value for n , as clearly shown in figure 2.2.8. All other
symmetry relation are not violated here and thus, our symmetrization procedure does not
affect them at all. Interestingly, taking only the real part of hn i provides identical results
48
when compared to the results obtained after the full symmetrization procedure. However, this
is only true for the simple case of the single site impurity model and does not hold for more
complex systems.
DOS (1/)
0.2
0.1
0
-5
-4
-3
-2
-1
/
Figure 2.2.7: Density of states (DOS) of the spin up electron as calculated using the symmetric-EOM approach
for different temperatures; = 200 (red line), = 50 (blue line) and = 5 (black dashed line). The development
of Kondo peaks in the density of states as the temperature decreases is clearly evident. The DOS of the spin

up electron calculated from the unsymmetrized Green function is identical provided only the real part of n
was used in the calculations. Parameters used are similar to those used in Refs. [158, 186], except U that was
chosen to be large (U = 10) but not infinite. Other parameters (in units of = L + R ): L = 3/10, R = 0,
, = 2. The bands are modeled as a Lorentzian with a half bandwidth 100.
2.2.5.2
The double Anderson model: symmetric-EOM approach

While for the case of a single site Anderson model only the relation G
() = G ()
breaks down, in the double Anderson model we find that all 3 symmetries described by equation (2.2.34) are violated. This can be traced to the more complex form of the Hamiltonian for
the double Anderson model, where each site is only coupled to one of the leads and transport
in enabled by the direct hopping term between the two sites.
Similar to the case of the Anderson model, as a result of symmetry breaking the occupation

i
h
() < d,
G
of the levels hn i is a complex number. In addition, the coherences,
=
also violate the symmetry relation,

= . In figure 2.2.9 we plot the real and imagi
nary parts of
and for the case where the symmetry procedure has been applied (left
panels) and for the unsymmetrized case (right panels). The upper panels show the imaginary
part of
and , which should show a mirror reflection about the zero axis (shown as thin
solid line). This is, indeed, the case when symmetry is restored. A more dramatic effect is
49
0.3
0.2
0.1
0
0
0.2
0.4
0.6
0.8
e/U
Figure 2.2.8: Occupation of the spin up electron before (solid lines) and after (circles) symmetrization.

Obviously the real part of the observable n is not affected by the symmetrization (blue color), and only the
non physical imaginary part disappears (red color). See figure 2.2.4 for parameters.

shown for the real part of

and (lower panels). The two curves representing Re
and Re
should be identical (left panel when symmetry is restored) but are quite distinct
when symmetry is not obeyed (right panel).
In figure 2.2.10 we plot the stationary current as a function of the bias voltage for the
double Anderson model. The current can be obtained from equation (2.2.76) (blue line) or
from equation (2.2.77) (red line). In the limit of infinite hierarchy in the EOM approach the
two formulas should coincide. However, when approximations are introduced or when the
hierarchy is truncated, the calculation of the current based on the two different formulas will
<
>
A

R
is
G
= G
G
coincide only if the symmetry relation G
()
()
()
()
preserved. Indeed, in the case of a single site Anderson model, even if symmetry is not restored,
this relation holds and the two calculations yield identical values for the current. However, in
the present case, all 3 symmetry relations are broken and thus, equations (2.2.76) and (2.2.77)
give different results for the current, as clearly evident in figure 2.2.10. More significantly is
the fact that equation (2.2.76) produces a finite value for the current even when the bias is
zero, indicating the break down of the fluctuation dissipation relation. When symmetry is
restored (black line) the two calculations are identical, as they should be, and the violation of
the fluctuation dissipation relation is also resolved.
The symmetrization scheme proposed in subsection 2.2.5 is not a magic cure and, in fact,
does not resolve all issues of mater. It is well known that the lesser and greater Green functions
should obey a simple sum rule where the integral over the difference of their diagonal elements
should always sum to 1:
50
real part
imaginary part
Coherences
0.1
0.2
0.3
0.4
e/U
0.1
0.2
0.3
0.4
0.5
e/U
Figure 2.2.9: The imaginary (upper panels) and real (lower panels) parts of
(blue line) and (red line)
calculated before (right panels) and after (left panels) symmetry was restored. The solid black
thin line in the

upper panels marks the zero axis. As expected, after symmetry restoration (left panels), Im = Im

and Re = Re , while before symmetry restoration (right panels) these equalities are violated. See
figure 2.2.5 for parameters.
i
h
S =
2
<
>
d (G
= 1.
()) (G ())
(2.2.129)
In figure 2.2.11 we plot the sum rule as given by equation (2.2.129) for the double Anderson
model where symmetry has been restored. A similar plot for the single site Anderson model
yields a value of 1 regardless of whether symmetry has been restored or not within the closure discussed above. However, in the case of the more evolved double Anderson model, even
when symmetry is restored and the Green functions obey all 3 relations described in equaP
P
tion (2.2.34), the sum rule is violated. Nonetheless, the sum
Sm = Ne , where
m{,} {,}
Ne = 4 is the total number of electrons in the system at maximal occupancy, is indeed preserved
when symmetrization is restored.
In the remaining of the chapter, we compare 4 different closures for the EOM of the NEGF
for the double Anderson model, two of which are developed by us while the remaining two
can be found in the literature. We use the Green function obtained via the symmetric-EOM
approach to calculate the differential conductance of the model. We note in passing that this is
the simplest theoretical model that can describe transport phenomena in a donor-acceptor complex or heterojunctions, which are the working principle of many electronic and optoelectronic
devices [194, 195].
51
_
Stationary current (e/ h)
12
10
10
-2
Eq. (1)
Eq. (2)
Sym
8
6
4
2
0
0
0.1
0.2
e/U
0.3
0.4
Figure 2.2.10: I V curves calculated using equations (2.2.76) and (2.2.77) before (blue and red lines respectively) and after (black line) applying the symmetry procedure suggested in subsection 2.2.5. See figure 2.2.6
for parameters.
2.3
Steady state conductance in a double quantum dot

array:
Assessing the symmetric-EOM technique for the NEGF
2.3.1
Overview
We study the role of different approximate closures to the EOM of the NEGF on steady state
properties (namely, the differential conductance) for a double quantum dot (QD) array, coupled
to two macroscopic leads [196] (the double Anderson model [113, 114]). Although general, this
model can be used to study transport through single diatomic molecules (where each atom is
represented by its conducting orbital [197] and even larger molecules and complexes [198, 199,
200] or structures [201].
Four closures are examined; two already proposed [114, 183] (we will refer to these closures
as approximations 1 and 4) and two developed by us (denoted as approximations 2 and 3). The
results obtained from the different closures were compared to the results attained using a manyparticle Master Equation (ME) approach [202] adequate for weak hybridization (system-leads
couplings) and high temperatures [203, 204].
We find that in contrast to the simple case of a single site model (Anderson model) in which
different closures beyond the simplest Hartree approximation scheme [179] yield very similar
transport results [182] (steady state current and differential conductance as a function of the
applied bias voltage) at high temperatures, the double Anderson model yield very different
52
1.1
1.0
0.9
0
0.1
0.2
0.3
0.4
e/U
Figure 2.2.11: S (blue line) and S (red line) calculated from the symmetrized lesser and greater Green
functions as a function of the bias voltage. The exact result should have been 1 (as marked by the solid black
L
R
R
line). Parameters used for the simulations in units of U = U = U are: L
= = = = 0.0025,
R
R
L
L
= 0.25, V = 0.1, = = 0.1, = = 0.175 and = 80.

= = = = 0, h = h
stationary currents and differential conductance curves for the different closures studied.
The performance of the different closures is analyzed in terms of the poles of the unperturbed
retarded Green function in comparison to the exact many-body result for the isolated system.
We find that one of the closures developed by us provides the most accurate description of the
poles and also the best overall agreement with the ME approach for all parameters studied in
this work. While these results are encouraging, a word of caution is in place. It is clear that
the conclusions drawn from the performance of the different closures for the single site model
cannot be extended directly to the two site model, in analogy, a suitable closure for the two
site model may fail in the larger systems. Thus, the study of larger arrays of QDs will require
either analysis along the lines sketched here or employ a different formalism. In what follows
we embrace the notation for the different Green functions described in appendix B.
2.3.2
Different closures of the EOM
The double Anderson model is described in subsection 1.3.3 and the relevant Green functions
= V
can be found in appendix B. Here we chose U = U = U , V

= V and h = h = h.
53
The EOM in Fourier space of the retarded Green function is:

R
G
()
R
h
(0 ())
R
+V G
()
1
R
+ h G
()
R

+ U G
()
R

+ V G
()
(2.3.1)
where 0/ () (c.f. equations (2.2.110) and (2.2.111)) is the tunneling (from the left/right
lead) self energy (the label 0 refers to the limits U, V 0). Calculating the EOM for the
higher order Green functions will give rise to other 2-particle and 3-particle correlations:

R

G
()
R 1
h
V 0 ()
R

h G
()
X
tk

+ h G
()
R
R

hn i + h G
()
R

+ V G
()
,
R

+ U G
()
R

R

R

()
() + tk Fk
Fk
() tk
Fk

(2.3.2)
Henceforth, the superscript R and the implicit dependence on are omitted for brevity.

G
= (
h U 0 )1 hn i + hG
hG

+hG
+ V G
+ V G
X

tk
Fk
tk
Fk
+ tk Fk
,
(2.3.3)
and
G
=
h
V 0
1
hn i d d
E

+hG
+ U G + V G
X

tk F
.
k tk Fk + tk Fk
(2.3.4)
To continue, one has to formulate the equations for these new Green functions, which in turn
will lead to other (higher order) Green functions. This infinite hierarchy of equations needs to
be truncated at a certain level, a process which is referred to as closure. In general, closures
cannot be improved systematically. Furthermore, it is often difficult to assess, a priori, the
accuracy of a given closure. We now discuss several different closures which are physically
motivated, tractable, and some are commonly used in the context of transport.
54
2.3.2.1
Approximation 1
Following the derivation given in Ref. [183], the following approximations are made:
1. All 3-particle Green functions set to zero.
qrst
2. Simultaneous tunneling of electrons of opposite spins are neglected, i.e., all Fqrst
k , Fk
and Gqrst
are also set to zero.
3. Remaining Green functions mixing leads and system operators are decoupled so

0
Fs
k ,
n
oE
iD
TC ns ( ) ck ( ) d 0
,
h
D
n
oE
i
d1 gkL ( 1 ) TC d (1 ) ns (1 ) d 0

.
h
See sub-subsection 2.2.4.2, second assumption, for a short discussion regarding this approximation.
0
0
s
4. The remaining 2-particle Green functions of the form Gs
(, ) are decoupled so G (, ) =
0
hns ( )i G
(, ).
The resulting equations (for the retarded Green functions) are given by:

G
h 0 )1 1 + hG
= (
+ U G + V G + V G ,
G
0
= = h
1

hG
+ U G + V G + V G ,
(2.3.5)

h
= h n G
U 0 G
,
h
V 0 G
= h n G d d ,

h
= h hn i G
V 0 G
,

(
h U 0 ) G
= hn i + h hn i G
,
(
h V 0 ) G
= n + h hn i G ,

(
h V 0 ) G
= n + h n G
.
(2.3.6)
d d , that should
In general the Green functions depend on the expectation values, e.g.

be calculated self consistently, i.e.:
D
d d
i
h
=
2
<
G
()
d.
(2.3.7)
55
The lesser Green functions are obtained from the same set of equations used to attain the
retarded Green functions by applying Langreth rules [81] to the contour ordered Green function.
2.3.2.2
Approximation 2
A seemingly better approximation scheme developed by us, is one that relaxes the last meanfield approximation (assumption 4) described in sub-subsection 2.3.2.1 and the 2-particle Green
0
functions of the form Gs
(, ) are not decoupled but treated fully. Thus, while the equations
for the single-particle Green functions in sub-subsection 2.3.2.1 remain the same, the equations
for the 2-particle Green functions will now be given by:

h
= hG
,
V 0 G
h
V 0 G
= hG d d, ,

h
= hG
,
U 0 G

(
h U 0 ) G
= hn i + hG
,
(
h V 0 ) G
= n + hG ,

(
h V 0 ) G
= n + hG
.
(2.3.8)
As noted above, the lesser Green function can now be evaluated using Langreth theorem and
the expectation values are determined self consistently via the lesser Green function.
2.3.2.3
Approximation 3
A more complete treatment of the 2nd order Green functions requires relaxing assumption
2 in addition to assumption 4 (see previous two sub-subsection). Namely, we attend to the
correlation functions that describe simultaneous tunneling of electrons of opposite spins in the
. Again, the only changes are in the equations for the 2-particle
double QD system, G
56
Green functions, and the resulting equations are given by:

G
=
h
V 0
1

hG
+ hG
hG
,
G
=
h
V 0
1
hG
d d

G
=
h
U 0
1

,
hG
+ hG
hG
E
,

hG
,
+ hG
= (
h U 0 )1 hn i + hG
G
h V 0 )1
G
= (
D
n + hG
,

= (
h V 0 )1
G
D

hG
,
+ hG
n + hG

G
=
h
+
1

G
=
h
+
1

=
G
h
+
1

G
=
h
+
1

hG
hG
+ hG
,
D

d d hG
+ hG
+ hG
,

+ hG
,
hG
hG
D

d d hG
+ hG
+ hG
.
(2.3.9)
This case is not different from the previous two in the sense that a self consistent treatment is
required.
2.3.2.4
Approximation 4
Finally we follow the derivation of reference [114]. The approximations and the equations are
describe in sub-subsection 2.2.4.2 and will not be reproduced here.
2.3.3
Master equations
The results obtained from the different closures were compared to the results attained using
a many-particle ME approach. For completeness, we briefly outline the ME formalism and
provide expressions for the transition rates for the double QD model. In the many-particle
picture, the system has different probabilities pi of being in one of its 2N possible eigenstates
(N being the maximum occupancy of the system, e.g. in our case of the double QD system
N = 4, and there are 16 many-particle states). Normalization requires that the probabilities
sum up to one:
N
2
X
pi = 1,
(2.3.10)
i=1
57
and in steady state there is no net current into or out of any state, i.e.,
N
2
X
Rji pi =
j=1
2
X
(2.3.11)
Rij pj ,
j=1
where Rji is the rate of transition from the many-particle state |n1 , ii to |n2 , ji (the jth out
of the n2 -particle states). The many-particle states are incurred by diagonalizing the systems
sys , while the rate constants are calculated assuming a specific model for the
Hamiltonian H
interactions between our system and its environment (the leads). Knowing the transition rates
the probabilities can be evaluated and the steady state current is then given by:
IL = e
L
(ni nj ) Rij
pj ,
(2.3.12)
ij
L
where ni is the occupation number of state |n, ii and Rij
is the rate constant that represents
the part of the total transition rate associated with the left lead. For sequential tunneling
we assume transitions only between many-particle states that obey |ni nj | = 1. In the weak
coupling regime the interaction between the system and the leads is treated as a perturbation
and the rates are given by the Fermi golden rule:
L
Rij
=
D

E2
2 |tk |2 ( k )

fL (Eij L ) n2 , j d n1 , i
h
,kL

+ (1 fL (Eji L )) |hn1 , i |d | n2 , ji|2 .

o
(2.3.13)
In the above, Eij = Ei Ej , is the difference in energy between the many-particle states |n1 , ii
and |n2 , ji, and fk ( k ) is the Fermi-Dirac distribution. Defining the matrix coupling of the
P
system to the left reservoir, L , with elements (L ) = 2 ( k ) |tk |2 , we can rewrite
kL
equation (2.3.13) as:

L
Rij
D

E2
(L )

fL (Eij L ) n2 , j d n1 , i
=
h
,kL
X
+ (1 fL (Eji L )) |hn1 , i |d | n2 , ji|2 .

o
2.3.4
(2.3.14)
Results and discussion
At this point we wish to examine the different approximations and find which one leads to a
system NEGF that describes the double QD more accurately or at least qualitatively. As we are
interested in transport properties of a system weakly coupled to the macroscopic leads we will
58
compare the results obtained from the different approximations to the ones calculated using
the many-particle ME approach. Under these assumptions, the ME is believed to be fairly
accurate [205]. We wish to note that the EOM approach is not limited to the weak coupling
case. As a measure of the quality of the approximations we chose to calculate the differential
conductance, dI/d.
In the many-particle picture, in the wide band limit (where the interaction with the leads
only broadens the energy levels of the system without introducing any spectral shift), we expect
to observe peaks in the differential conductance at values of the bias voltage that correspond to
L/R Ef = 0.5 e E (N ) = E (N ) E (N 1), where Ef is the equilibrium Fermi energy
of the electrodes (throughout taken to be zero), E (N ) is the energy of the many-particle state
with N electrons of the unperturbed system, and is the applied bias voltage. The voltage
can be applied symmetrically to both leads (i.e., L = Ef + 0.5 e and R = Ef 0.5 e), or
asymmetrically (L = Ef + e and R = Ef ). In the present study we have used the symmetric
version.
The differential conductance is derived from differentiating the steady state current with
respect to the bias voltage. The current was evaluated from the Meir-Wingreen formula in
equation (2.2.76). The resulting EOM were solved self-consistently in Fourier space with a
frequency discretization of d = 0.0005U over 32, 768 grid points. Depending on the approximation, 15 150 self-consistent iterations were required to converge the results. Convergence
was declared when the population values (hnm i) at subsequent iterations did not change within
a predefined tolerance value chosen to be 106 . For each set of calculations symmetrization
routine was applied to restore the symmetry relations of the Green functions [166].
2.3.4.1
Symmetric bridge
The transport through the double QD system can be classified into symmetric and asymmetric
bridge setups, with or without inter-dot repulsion term, V . In this sub-subsection we first
consider the symmetric setup in which = = = = = 0.35U . The remaining model
parameters were taken to be L = L = R = R = 0.015U , L = L = R = R = 0,
and 1 = U/40. To make the connection with chemical systems, one can consider the on-site
repulsion U to be of the order of 1 4 eV [200]. This implies that is of the order of 0.015 0.06
eV. In the case of the symmetric bridge the systems single-particle energy levels are around
0.35 1.4 eV above the leads Fermi energy and the inter-site repulsion (exciton binding energy)
V is of the order of 0.8 3.2 eV.
In figures 2.3.1, 2.3.2, 2.3.3, and 2.3.4 we plot the differential conductance for a symmetric bridge for different values of the hopping term h. We set the inter-dot repulsion V = 0.
The black curves (solid line) are the results obtained by fully diagonalizing the bare system
59
Differential conductance
0.2
|0|1 |0|1
|1|2
0.1
0
0
1
0.5
Bias voltage [U]
1.5
Figure 2.3.1: Plots of the differential conductance versus the bias voltage for the symmetric bridge (
= = = = 0.35U ) for V = 0 and h = 0.1U . Black curve corresponds to results based on the
ME. Red (circles), green (diamonds), blue (triangles) and magenta (stars) correspond to the results obtained
by approximation schemes 1 to 4, respectively. The notation |ii |ji indicates that the conductance peak calculated by means of ME corresponds to a transition form the ni -particle states to any of the nj -particle states.
The remaining model parameters were L = L = R = R = 0.015U , L = L = R = R = 0, and
1 = U/40.
0.2
|0|1
|0|1
|1|2
0.1
0
0
|1|2
1
0.5
Bias voltage [U]
1.5
Figure 2.3.2: Same as figure 2.3.1 but for h = 0.3U .

60
0.2
|1|0
|0|1
|1|2
|2|3
|1|2
0.1
0
0
1
0.5
Bias voltage [U]
1.5
0.2
|1|2
0.1
0
0
|1|0
1
0.5
Bias voltage [U]
1.5
61
sys (and solving the ME). The other curves represent the outcome of the NEGF formalism
H
within the different closure approximations. We also label the different peaks in the differential conductance, obtained via the ME approach, with the corresponding transitions between
many-body states, i.e., |0i |1i corresponds to transitions from an empty system to a system
with a single electron, etc.
In the single-particle Green function formalism, peaks in the differential conductance will
occur at the poles of the calculated retarded Green function. Hence a good approximation is one
that will produce single-particle Green function with poles at the position of the many-particle
transitions.
For the smallest value of h, we find that all approximations agree qualitatively with the
ME approach. When the value of h is increased it is clear that approximation 1 (red circles)
breaks down, implying that this simple closure is insufficient to describe strong hopping between
the quantum dots. Approximations 2 and 3 (green diamonds and blue triangles, respectively)
do agree with the ME, but miss certain conductance peaks (e.g. the peak at e
U 0.75 in
e
figure 2.3.2, the peak at e
U 0.5 in figure 2.3.3 and the peak at U 0.2 in figure 2.3.4), all of
which correspond to transitions involving a 2-electron occupancy. Approximation 4 (magenta
stars), which includes 3-particle Green functions at a mean-field limit, performs slightly better
in that regard.
Green function poles
|h|
+ 0.5 (U + V ) S1
+ 0.5 (U + V ) + S1
+ V |h|
+ V + |h|
Energy differences
E (1) :
|h|
E (2) :
+ 0.5 (U + V ) S2 |h|
+ 0.5 (U + V ) + S2 |h|
+ V |h|
+ U |h|
Table 2.2: Left column: Location of the poles of the unperturbed systems Green function as calculated using
the 2nd approximation. Right column: The differences in energy
q between many-particle states
qthat differ by one
electron, such that E (N ) = E (N ) E (N 1). Here S1 =
1
2
(U V )2 + 4h2 , and S2 =
1
2
(U V )2 + 16h2
In figures 2.3.5, 2.3.6, 2.3.7, and 2.3.8 we present results for the differential conductance
obtained for the symmetric bridge where the inter-dot repulsion, V , is included. All other
parameter are identical to those of figure 2.3.1. In this case, we find that approximation 1 is
not suitable even for small values of the hopping term h while approximation 4 appears to work
for low values of h < 31 U (figures 2.3.5 and 2.3.6), however, it fails to capture peaks resulting
from transitions through 2-electron occupancy at higher values of h, as depicted in figures 2.3.7
and 2.3.8.
We note that for this set of parameters, such transitions involving 2 electrons are absent
62
0.2
|0|1
0.1
|0|1
0
0
1
0.5
Bias voltage [U]
1.5
Figure 2.3.5: Plots of the differential conductance versus the bias voltage for the symmetric bridge (
= = = = 0.35U ) for V = 0.8U and h = 0.1U . Black curve corresponds to results based on
the ME. Red (circles), green (diamonds), blue (triangles) and magenta (stars) correspond to the results obtained by approximation schemes 1 to 4, respectively. The notation |ii |ji indicates that the conductance
peak calculated by means of ME corresponds to a transition form the ni -particle states to any of the nj -particle
states. The remaining model parameters were L = L = R = R = 0.015U , L = L = R = R = 0,
and 1 = U/40.
0.2
0.1
0
0
|0|1
|0|1
1
0.5
Bias voltage [U]
1.5

63
0.2
|1|0
0.1
0
0
|1|2
1
0.5
Bias voltage [U]
1.5
0.2
|1|0
0.1
0
0
|1|2
1
0.5
Bias voltage [U]
1.5
64
for h < 13 U in the bias voltage studied. Approximations 2 and 3 agree very well with the ME
results for all values of h, even at values of the bias voltage that correspond to transfer through
2-electron states, in contrast to the case where V = 0 in which they fail to capture conductance
peaks involving 2 electrons.
The performance of the different approximations can be rationalized in terms of the pole
structure of the unperturbed system Green function, which can be compared to the exact
many-body energy differences between many-particle states that differ in one electron. While
it is possible to carry out this analysis for all closure approximations, it is often a tedious task.
Thus, in what follows we provide such analysis for the case of approximation 2 only.
The poles of the Green function, the many-particle energy levels and the differences in energy
are summarized in table 2.2 (see appendix C for more details regarding the derivation of the
poles of the Green function). As can be seen from table 2.2, conductance peaks corresponding
to transitions |0i |1i and |1i |0i, that is, peaks appearing at the values of E(1), are
captured by approximation 2 since the Green function has poles at the correct locations. Higher
excitations involving 2 or more electron occupancies are not fully or systematically accounted
for by approximation 2 (or any of the other closures described in this paper, for this matter). In
general we find that such higher transitions are not captured by approximation 2 when V U .
For V = U we find that E (2) has 4 different values: + V |h| and + V 2 |h|, concurrently
the Green function has poles at + V |h|, thus, some of the transitions involving the 2-electron
states |N = 2i (particularly those with E (N = 2) = + V |h|) are described by the NEGF.
Following this short analysis we can now better explain the results of figures 2.3.5, 2.3.6, 2.3.7,
and 2.3.8. For a large value of the inter-dot repulsion (V = 0.8U , thus V U ), one expects that
the calculated Green function will capture the higher order transitions in the relevant bias win

1.1
in figure 2.3.8, it
dow and agree with the ME results. If one considers the second peak e
U
results from transmission through a many-particle level with |N = 2i. For the symmetric bridge,

this peak corresponds to E = 2 + 12 (U + V ) S2 ( |h|) = + 12 (U + V )S2 +|h|, where
S2 =
(U V )2 + 16h2 . Under the assumption that V U , one can approximate E
+ V q|h|, whereas the Green function has one of its poles at PG = + 21 (U + V ) S1 , where
1
2
S1 = 12 (U V )2 + 4h2 , which, for V U can be approximated by PG + V |h|. For the

studied parameters we find PG = 0.54289 and E = 0.54643, and indeed the differential conductance based on approximation 2 show a peak at twice this value, i.e., e
U 1.1. While for
the case where V = 0 this transition is overlooked. It is easy to verify that a similar argument
holds for the second peak in figure 2.3.7 as well.
65
2.3.4.2
Asymmetric bridge
We now turn to discuss the case where 6= referred to as the asymmetric bridge. Once
again we have calculated the differential conductance using the 4 different closure approximations to the NEGF formalism and compared the results to the differential conductance obtained
by the ME. Analysis based on analytic expressions for the poles of the Green function or the
sys is more difficult, and the expressions are not as compact as in the
many-particle energies of H
symmetric case. The results for the poles of the Green function within closure approximation
2 are given in appendix C, while the many-body energy differences were obtained numerically.
In figures 2.3.9, 2.3.10, 2.3.11, 2.3.12 and figures 2.3.13, 2.3.14, 2.3.15, 2.3.16 we plot the
differential conductance for the asymmetric bridge for different values of the hopping term h
for V = 0 (first four figures) and V = 0.8U (last four figures), respectively. The on-site singleparticle energies were = = 0.15U , = = 0.2U . The remaining model parameters
are identical to those of the symmetric bridge and were taken to be L = L = R = R =
0.015U , L = L = R = R = 0, and 1 = U/40. As before, the black curves (solid
line) corresponds to the ME results. The other curves represent the outcome of the NEGF
formalism within the different closure approximations. We also label the different peaks in
the differential conductance with the corresponding transitions between many-body states, i.e.,
|0i |1i corresponds to transitions from an empty system to a system with a single electron,
etc.
From figures 2.3.9 to 2.3.16 it is obvious that approximations 1 and 4 do not perform as well
as approximations 2 and 3. We would like to note that approximations 1 and 4 utilized a meanfield like approximation decoupling the higher order Green functions, while in approximations
2 and 3 higher order Green functions are ignored altogether or fully taken into account. For
all parameters studied in this work (not all presented here), we find that approximation 2
performed better than all the other approximations, suggesting that including higher order
correlations in a mean field fashion or a more complete treatment of the 2nd order Green
functions is not advantageous.
We find that approximations 2 and 3 predict negative differential conductance at higher
values of h, not obtained by the ME, as shown in figures 2.3.15 and 2.3.16. The dips occur
(in both cases) at values corresponding to the activation of the anti-bonding single electron
state. While this transition is suppressed in the ME approach , it appears to be enhanced in
the NEGF formalism.
66
0.2
0.1
|1|2
|2|1
0
0
1
0.5
Bias voltage [U]
1.5
Figure 2.3.9: Plots of the differential conductance versus the bias voltage for the asymmetric bridge ( =
= 0.15U and = = 0.2U ) for V = 0 and h = 0.1U . Black curve corresponds to results based on the
ME. Red (circles), green (diamonds), blue (triangles) and magenta (stars) correspond to the results obtained
by approximation schemes 1 to 4, respectively. The notation |ii |ji indicates that the conductance peak
calculated by means of ME corresponds to a transition form the ni -particle states to any of the nj -particle
and 1 = U/40.
0.2
|1|2
|2|1
|1|0
0.1
|2|3
0
0
1
0.5
Bias voltage [U]
1.5

67
0.2
|2|1
|1|0
|1|2
|2|3
|2|1
0.1
0
0
1
0.5
Bias voltage [U]
1.5
0.2
|2|1
|2|3
|1|2
|1|0
|2|1
0.1
0
0
1
0.5
Bias voltage [U]
1.5
68
0.2
0.1
|1|0
|0|1
|1|2
|1|2
0
0
1
0.5
Bias voltage [U]
1.5
Figure 2.3.13: Plots of the differential conductance versus the bias voltage for the asymmetric bridge ( =
= 0.15U and = = 0.2U ) for V = 0.8U and h = 0.1U . Black curve corresponds to results based
on the ME. Red (circles), green (diamonds), blue (triangles) and magenta (stars) correspond to the results
obtained by approximation schemes 1 to 4, respectively. The notation |ii |ji indicates that the conductance
peak calculated by means of ME corresponds to a transition form the ni -particle states to any of the nj -particle
and 1 = U/40.
0.2
|0|1
|1|0
|1|2
0.1
|1|2
0
0
1
0.5
Bias voltage [U]
1.5

69
|0|1
|1|0
0.2
|1|2
0.1
0
0
1
0.5
BIas voltage [U]
1.5
|0|1
|1|0
0.2
|1|2
0.1
0
0
1
0.5
Bias voltage [U]
1.5
70
2.4
Concluding remarks
In this chapter we introduced the NEGF formalism. We focused on the EOM technique used
to evaluate these correlation functions and pinpointed an inherited flaw of the technique which
breaks the symmetry of the single-particle Green function. This in turn yields unphysical results
when the latter is used to calculate expectation values, such as, complex level occupancies
and finite currents at zero bias voltage. A symmetry restoration scheme was proposed by us
and tested on the Anderson and double Anderson models. In addition, we have assessed the
validity of the EOM approach to the NEGF formalism for an interacting system coupled to two
macroscopic leads.
The interacting system consists of two coupled quantum dots, each with one electronic level
(spin up and spin down), connected serially, taking into account intra and inter-dot Coulomb
interactions. This system, is the simplest one that can describe current through heterojunctions and acceptor-donor complexes. Four different closure approximations to the EOM were
examined, two of which are commonly used in the literature (approximations 1 and 4) and
others have been developed by us (approximation 2 and 3). As a measure of the quality of the
approximations we calculated the differential conductance (derived by differentiating the steady
state current with respect to the bias voltage) and compared the results to those obtained by
the ME approach, which under the approximations of weak coupling to the leads and high
temperature provides accurate results.

Two different cases corresponding to a symmetric bridge = and an asymmetric

bridge 6= with and without inter-dot Coulomb repulsion (V ), were studied for different
values of the inter-dot hopping term h. As expected, we find that keeping more terms in the
closure or including higher order correlations in the EOM, does not necessarily improve the
approximation. As a rule of thumb, neglecting higher order Green functions (approximations
2 and 3) preforms better as compared to closures that include such terms at a mean-field level
(approximation 1 and 4).
Focusing on approximation 2, which provides the overall best agreement in comparison to the
ME approach, we found that transitions involving only single electron states were reproduced
by the NEGF. However, when higher many-particle states are involved, the accuracy of the
approximation depends on the strength of the Coulomb coupling V . In cases where V U ,
approximation 2 also captures conductance peaks associated with transitions involving two
electron states. Nevertheless, this is not systematic, and the peaks are captured only due to an
accidental degeneracy that occurs when V U .
While all the approximations described in section 2.3, capture the Coulomb blockade and
the main characteristics of the steady state transport for the single QD model (the Anderson
model), they are not easily expandable to systems with more complex nonequilibrium dynamics,
such as the double QD model (the double Anderson model). In light of this, the success of
71
approximation 2 for the double QD model does not necessarily imply that it will provide
quantitative or even qualitative results for a system with more than a single level per a QD or
for a system with more than two coupled QDs. However, analysis of the poles of the resulting
unperturbed retarded Green function and comparing them to the exact many-body energy
differences does provide a tool to assess the accuracy of a given approximate closure and can
be used for larger bridge systems even if these can only be performed numerically.
In light of these findings, it is now clear that to study bigger correlated systems and/or to
capture conductance peaks associated with N > 2 many-particle states, we will have to:
1. Truncate the EOM at a higher level and fully include (i.e., not in a mean-field kind of way)
higher order Green function. This, unfortunately, will results quickly in an intractable
set of equations and will be impractical.
2. Fit the different parameters, so accidental degeneracies will occur and the poles of
the retarded unperturbed Green function will coincide with the many-particle energy
differences E (N ) = E (N ) E (N 1) .
These conditions can prove to be too limiting and unworkable and one is advised to take
them into consideration when analyzing ones results obtained via the EOM or symmetric-EOM
approaches. In the next chapter we turn to a completely different paradigm for the calculation
of transport properties in junctions based on semiclassical approaches.
72
Chapter 3
Semiclassical approaches to quantum
transport
3.1
overview
Using classical physics to describe quantum mechanical systems [206, 207, 208, 209, 210, 211,
212] goes way back to the early days of quantum mechanics [213]. Nowadays, molecular dynamics (MD) simulations [214, 215, 216, 217, 218] are extensively used to describe a broad variety
of dynamical processes in complex and large molecular systems [219, 220, 221, 222, 223, 224,
225, 226, 227, 228, 229, 230, 231, 232, 233], where no other theoretical tool is available to carry
out such calculations. Unfortunately, MD simulations cannot describe any quantum aspects of
the dynamics (tunneling, interferences and etc.), as the method is completely classical.
Quantum effects in the dynamics of molecular systems maybe very prominent, e.g., processes involving hydrogen atom motion where one can reasonably expect quantum effects to
be significant, thus, a valid description should be capable of providing at least an approximate
characterization of these quantum effects. There are a couple of ways one can proceed to try
and include approximations to quantum effects in classical MD simulations:
1. Perhaps the most common class of approaches are referred to as mixed quantum-classical
approach [234, 235, 236, 237, 238, 239], whereby some (usually small) number of degreesof-freedom (DoF), customarily the relevant electronic DoF, are described quantum mechanically, i.e., by a wave-functions determined via a time-dependent Schrdinger equation, and the other (typically large) number of DoF, traditionally the nuclear DoF and
the rest of the system, are described classically, i.e., via coordinates and momenta that
follow a classical-like trajectory.
The time-dependent Schrdinger equation for the quantum DoF, and classical trajectory
73
equations for the classical degrees of freedom, are then integrated simultaneously. The
coupling between the quantum and classical degrees-of-freedom is treated in some approximate fashion. There are a variety of such mixed quantum-classical approaches, and
they have been usefully applied to a variety of problems [240, 241, 242, 243, 244, 245].
The primary downside of such approaches is that there is no way to couple the quantum
and classical degrees of freedom that is completely consistent.
2. Another strategy, is to treat all DoF semiclassically, which has the advantage of a consistent dynamical description of the complete dynamics. Many papers published in the
1970s [246, 247, 248, 249, 250] showed how numerically computed classical trajectories
could be used semiclassically and that such a theory indeed provides an approximate description of all quantum effects in molecular dynamics, such as, interference (coherence),
tunneling and selection rules due to identical particle symmetry.
Many applications were carried out at that time, primarily to gas-phase molecular collision problems, illustrating these ideas and showing that the semiclassical description is
often very good even when quantum effects are quite large. Since then many semiclassical treatments that rely on the semiclassical initial-value-representation (IVR) have been
applied to a variety of physically interesting condensed phase problems with remarkable
success [251, 252, 253, 254, 255, 256, 257, 258, 259, 260, 261, 262, 263, 264].
Lately, a semiclassical approach have been formulated to allow the study of nonequilibrium
quantum transport in molecular junctions [62, 89, 265, 90]. This new approach is the subject of this chapter. The key idea behind the semiclassical approach, based on earlier works
of McCurdy, Meyer, and Miller (MMM) [266, 267, 268, 269] and later by Miller and White
(MW) [270], is to transform a general second quantized many-electron Hamiltonian into a
classical one. This is achieved by defining a prescription which transforms the electronic operators into classical functions that correctly account for the anti-commutativity of the Fermion
operators .
In what follows we describe three different mappings; the action-angle mapping, the
Cartesian mapping and the Hubbard mapping, each one addressing and fixing the shortcomings of its predecessor. We will mainly center on the Hubbard mapping and evaluate its
accuracy by testing it on the models described in subsections 1.3.1, 1.3.2 and 1.3.3. Apart
from the single resonant model, the Anderson and the double Anderson models have no known
numerically exact solution, thus, we compare the Hubbard mapping results to the Anderson
model with results obtained from the NEGF symmetric-EOM approach [166, 167]. In the
double Anderson case we compare the semiclassical results with those attained from a ME
approach [271] depicted in subsection 2.3.3, to the best closure of a NEGF symmetric-EOM
74
approach [166, 167] outlined in sub-subsection 2.3.2.2, and to the results based on the Cartesian/quaternion mapping [89] described in subsection 3.1.2.
3.1.1
Action-angle mapping
The essence of MWs model [270] is that each Fermion DoF, or more explicitly, each pair
of single-particle annihilation and creation operators, is described by a pair of action-angle
variables (classical DoF):
ai aj
r
ni n2i +

nj n2j + ei(qi qj ) ,

(3.1.1)
and
ai ai 7 ni ,
(3.1.2)
where (n, q) are the action-angle variables and is the Langer correction, which will be further
discussed in subsection 3.3.2.
For example, the Hamiltonian of the single resonant level (described in subsection 1.3.1):
=
H
n+
k n
k +
K{L,R} kK
tk ck d + h.c.,
(3.1.3)
K{L,R} kK
transforms into [62]:

Hcl = n +
k nk
K{L,R} kK
tk
r
nk n2k + (n n2 + ) ei(qk q) + h.c..

(3.1.4)
K{L,R} kK
Note that here n

is the electron number quantum operator while n is the classical action
variable.
The current (from the left lead) is given by the change in occupancy of the left electrode,
with the right current defined analogously:
IL
d
= e
dt
IL = e
d
n
k (t)
7 e
dt
kL
+
+
X
nk (t)
(3.1.5)
kL
E
i XD
tk d (t) ck (t) tk ck (t) d (t)
h
kL
*
+
r

i X
i(q
q)
2
7 e
tk nk nk + (n n2 + ) e k
h.c. ,
h
kL
(3.1.6)
75
and the dynamics of the action-angle variables are given by Hamiltons equations. We note in
passing that for the calculation of the current one can either perform the semiclassical mapping
before taking the time derivative (i.e., by mapping n
k (t) 7 nk (t) and taking the time derivative
of nk (t)) or apply the mapping to equation (3.1.6). Both ways give the same result.
To complete the treatment one has to define initial conditions for the action-angle variables.
To recover the correct statistical behavior of the quantum system (at least at time t = 0) we
use the following quasi-classical procedure: being interested in a non-correlated initial state
with leads at thermal equilibrium, we can populate each leads electronic DoF independently.
We enforce quantum statistics on the initial conditions for each degree of freedom i by setting
the initial action ni to either 0 or 1 such that the expectation value of the action ni , averaged
over the set of initial conditions, satisfies the Fermi-Dirac distribution. The angle variable qi
selected at random between 0 and 2. The systems level is chosen to be unpopulated.
The present mapping for the electronic DoF provides qualitative and quantitative results for
a range of system parameters, i.e., different bias voltages, temperatures, and electron-phonon
couplings in the case of the single resonant level and for the Holstein model [87]. However,
it falls short in providing accurate and even qualitative results in several different limits. For
example, it fails to capture the effects of a gate voltage in the noninteracting case of the resonant
model [62] or the well-known inelastic tunneling peaks when the phonon bath is characterized
by a sharp spectral function [87]. Moreover, it does not fully capture the Coulomb blockade
when electron-electron interactions are introduced. In the next subsection we describe the
Cartesian mapping, which corrects the aforementioned deficiencies.
3.1.2
Cartesian mapping
In this approach proposed by Li and Miller [88], each Fermion DoF is thought of as a 2-state
(i.e., spin 21 ) system, characterized by the Pauli spin matrices, and the isomorphism between
the Pauli matrices and quaternions [272] is utilized. In turn a mapping to the cross-product of
the 2D coordinate r = (x, y) and conjugate momenta p = (px , py ) is performed [89]:
ck ck =
1
1 i
+ Ik Jk 7 + rk pk ,
2 2
2
ck d + d ck
i
(Ik J + I Jk )
2
7
rk p + r pk ,
(3.1.7)
(3.1.8)
where Ik , I , Jk and J are the basic elements of quaternions.

The above mapping combined with a quasi-classical description of the dynamics works well
76
when no electron-electron interactions are present, but fails to capture the well-known Coulomb
blockade effect at low temperatures and high values of U . This is a result of the fact that within
the quasi-classical approximation the occupation numbers assume a continuum value and thus,
when the dot is occupied by a "fraction" of an electron with spin up it does not block completely
the path of an electron with spin down. Generalized classical models for the 2-electron Coulomb

term in the Hamiltonian, of the form U f n f n , have been considered. That is:

Un
n
7 U f n f n ,
(3.1.9)
where the only quantum mechanical restriction on the function f (n) is that f (0) = 0 and
f (1) = 1. Motivated by the Gaussian binning methodology of Bonnet et al. [273], whereby
final classical action variables are required to be close to their quantum (integer) values, the
following function was chosen:
2 2
f (n) = e(n1) ,
(3.1.10)
which for large values of does satisfy f (0) = 0 and f (1) = 1. The Coulomb interaction is thus
essentially zero unless both n and n are close to 1. For example, in terms of the Cartesian
variables and the Gaussian scheme, the classical Hamiltonian of the Anderson model (see
subsection 1.3.2) is given by:
Hcl =
(x py, y px, )
{,}
"
+U exp
(x py, y px, 1) 2
k xk py,k yk px,k
K{L,R} kK
tk (xk py, yk px, ) + (tk ) x py,k y px,k ,

(3.1.11)
K{L,R} kK
where and are positive integer used as fitting parameters. In the above equation, a value
of 21 was subtracted from the classical mapping of the occupation number operator to include
the Langer correction. Given the classical Hamiltonian, the dynamics of the Cartesian variables
are assumed to follow Hamiltons equations of motions.
As was done for the previous mapping, to complete the mapping procedure one needs to
define the initial conditions for all classical phase-space variables. A similar route discussed to
the one discussed in the previous subsection in adopted here. Specifically, a random number,
k , in the interval [0, 1], is generated and then the occupation of mode k is selected according
to:
0
k > f (k K )
nk =
.
(3.1.12)
1
k f (k K )
77
The phase space variables are then sampled according to:

xk = rk cos (k ) ,
(3.1.13)
yk = rk sin (k ) ,
sin (k )
,
rk
cos (k )
= pr,k sin (k ) nk
,
rk
px,k = pr,k cos (k ) nk

py,k
(3.1.14)
where k is selected at random between 0 and 2. Since the Cartesian mapping introduces 2
classical phase space variables for each creation and annihilation operators, this leads to freedom
in choosing the initial conditions, as long as the population, given by nk = xk py,k yk px,k ,
satisfies equation (3.1.12). The relations defined by equations (3.1.13) to (3.1.14) guarantee
that this constraint is not violated regardless of the values of rk and pr,k . The quantum dot
is initially unpopulated.
The presented approach provides excellent agreement for the single resonant level model
(non-interacting limit) in comparison to exact quantum mechanical calculations for a wide
range of model parameters, including different temperatures and bias voltages. It also captures
quantitatively the behavior of the current at different gate voltages, correcting for the flaws
of the mapping describe in subsection 3.1.1, based on the MW action-angle approach. When
two-electron interactions are included the approach is no longer numerically exact, however,
when combined with the Gaussian binning scheme to facilitate quantization, it captures the
Coulomb blockade quantitatively and more importantly, the temperature dependence of the
current below and above the blockade. Unfortunately, this mapping fails when used for more
complex systems, such us the double Anderson model. In what follows a new Cartesian mapping
is introduced. The new mapping naturally accounts for electron-electron interaction with no
further approximated schemes.
3.2
The Hubbard mapping
Opposed to the previous mappings, we first start by representing the second-quantized Hamiltonian of the system in terms of the Hubbard operators and the many-body state creation
and annihilation operators. Subsequently, we map the many-body state operators to classical
degrees of freedom. The new mapping is derived by using a coherent state representation of
Fermions adopting analogies based the work of Meyer and Miller [274] and Stock and Thoss [275]
for Boson operators. The new mapping allows us to express the Fermion and many-body state
creation and annihilation operators separately rather than mapping pairs of creation and an78
nihilation operators by quaternions or action-angle variables. In what follows the method is

exemplified using the double Anderson as a case study.
3.2.1
Hubbard Operators
The double-Anderson model can host up to 4 electrons in different 16 many-body

states. In the

E

occupation number (ON) vector representation [276] these will be denoted by n n n n .
We represent these states in Fock space as {|0i , |1i , ..., |15i}, where |0i refers to the state |0000i
(empty system), |1i |4i refer to the four possible one-electron states, |5i |10i refer to the
six two-electron states, |11i |14i refer to the four three-electron states and |15i refers to the
fully occupied system (see appendix D for more details). The Hubbard operators are defined
as follows [277]:
X p,q = |pi hq| = Dp Dq ,
(3.2.1)
in which X p,q describes the transition from the many-body state |qi to |pi; Dp (Dq ) create
(annihilate) the many-body state |pi (|qi). The electron creation and annihilation operators
can be expressed in terms of the Hubbard operators (appendix D):
dm =
hp |dm | qi X p,q ,
(3.2.2)
E
X D

p dm q X p,q .
(3.2.3)
X
p,q
dm =
p,q
For example: The creation operator of the spin down electron on dot is given by:
d = X 2,0 X 5,1 + X 6,3 + X 8,4 + X 14,7 X 11,9 X 12,0 X 15,13
= D2 D0 D5 D1 + D6 D3 + D8 D4
+D14
D7 D11
D9 D12
D10 D15
D13 .
(3.2.4)
The Hubbard operator X p,q has either a Boson character or Fermion character depending on
the difference between the number of electrons in states |qi and|pi. If the difference is an even
number then the operator has Boson character, otherwise it has a Fermion character. This
also reflect on the character of the many-body states creation/annihilation operators, Dq /Dq ,
since they can be expressed in term of Hubbard operators, as Dq = X 0,q and Dq = X q,0 . The
Hubbard operators will obey the following commutation/anti-commutation relation depending
on their character (Boson/Fermion respectively):
h
X p,q , X r,t
= q,r X p,t p,t X r,q ,
(3.2.5)
79
where stands for commutation/anti-commutation, respectively.
3.2.2
Reformulating the Hamiltonian in terms of Hubbard operators
The systems Hamiltonian (equation (1.3.12)) for the double Anderson model in terms of the
Hubbard operators is given by:
sys =
H
15
X
i Di Di + h
i=0
15
X
ri,j Di Dj .
(3.2.6)
i, j = 0
The elements of the symmetric matrix (ri,j = rj,i ) are given by:
ri,j =
E
X D

i d d + d d j ,
(3.2.7)
where |ii and |ji are many-body states. Specifically:

r1,3 = r2,4 = r5,10 = r7,10 = 1,
r5,6 = r6,7 = r11,13 = r12,14 = 1.
(3.2.8)
The energies of the many-body states, i , are summarized in Table 3.1. The interaction HamilHilbert space
empty system
one-electron
two-electron
three-electron
fully occupied
energy
0 = 0
1 =
2 =
3 =
4 =
5 = + + U
6 = + + V
7 = + + U
8 = + + V
9 = + + V
10 = + + V
11 = + + + U + 2V
12 = + + + U + 2V
13 = + + + U + 2V
14 = + + + U + 2V
15 = + + + + 2U + 4V
Table 3.1: The energies of the many-body states (denoted in ) in terms of the orbital energies (denoted in )
and the Coulomb repulsion strengths (U and V ).
80
tonian (c.f. equation_(1.3.13)) in terms of the Hubbard operators is given by:

int =
H
15
X
X
(r )i,j tk ck Di Dj +
i
(r )i,j (tk ) Dj Di ck
i,j=0 ,kL
15
X
X
i,j=0 ,kR
r
t c D Dj
i,j k k i

r
i,j

tk Dj Di ck
(3.2.9)
)
where the terms (rm
i,j are the prefactors of the Hubbard operators in the expansion of the
electron creation/annihilation operators (see appendix D) and are given by:
(rm
)i,j = hi |dm | ji .
(3.2.10)
For example:
d = D0 D3 D1 D9 D2 D6 + D4 D7
+D5 D11 D8 D14 D10

D13 + D12
D15 ,
thus, the non-zero elements of r

3.2.3

i,j
are:

0,3
= r
1,9
= r

(3.2.11)
4,7
= r
2,6
= r

5,11
= r
8,14
= r

12,15
10,13
=1,
= 1.
(3.2.12)
Classical mapping for the Hubbard operators
Our goal is to establish a simple mapping for the Hubbard representation of the Hamiltonian
of a quantum dot coupled to metallic leads. One possible route is based on the Cartesian mapping (subsection (3.1.2)) which provides a recipe to map products of even number of Fermion
operators. Here, however, the interactions Hamiltonian contains odd number of terms (cf. equation (3.2.9)). Therefore, we define a new Cartesian mapping of the electron and many-body
state creation and annihilation operators:

o
1 n
xk + py,k 1 yk px,k ,
2

o
1 n
xk + py,k + 1 yk px,k ,
2
ck
ck
(3.2.13)
where rk = {xk , yk } and pk = px,k , py,k are position and conjugate momentum vectors
in 2D. Similarly, we define the same mapping for the many-body states creation and annihilation
81
operators:

o
1 n
Xq + PY,q 1 Yq PX,q ,
2

o
1 n
Xq + PY,q + 1 Yq PX,q .
2
Dq
Dq
(3.2.14)
As before, Rq = {Xq , Yq } and Pq = {Px,q , Py,q } are position and conjugate momentum vectors
in 2D. Note that the mapping ignores particle statistics of Dq and Dq . We elaborate on this in
subsection 3.3.2.
The logic behind the above mapping follows from the Meyer-Miller [274] and Stock-Thoss [275]
(MMST) mapping for Bosons, where the prefactor of the coherent state,

2
2
1
1
x 1p e 4 (x +p )+
2
1
2 xp
(3.2.15)

can be associated with the raising operators, b = 12 x 1p . Similarly, for the Cartesian
mapping of Fermions introduced by Li and Miller,[88] the coherent state can be expressed as:

2
2
2
2
1
1
x + py + 1 (y px ) e 4 (x +y +px +py )+
2
1
2 (xpy +ypx )
(3.2.16)
where by elicitation, one arrives at the mappings of equations (3.2.13) and (3.2.14).
With these definitions, the system Hamiltonian in equation (3.2.6) is given by:
Hclsys =
15
X
i Ni +
i=0

h X
ri,j Xi Xj + Yi Yj + PX,i PX,j
2
i, j

+ PY,i PY,j + Xi PY,j Yi PX,j + Xj PY,i Yj PX,i ,

(3.2.17)
where
2
2
1
Xi + PY,i + Yi + PX,i ,
(3.2.18)
=X =
4
is the many-body state occupation. In the above equations, is the Langer correction which
is treated as a parameter (for more details see subsection 3.3.2). The interaction Hamiltonian
in equation (3.2.9) is mapped to:
Ni = Di Di
Hclint =
i,i
15 X
1 X
(r ) t F (k, i, j)
4 i,j=0 ,kL i,j k
15 X
1 X
r
t F (k, i, j) ,
4 i,j=0 ,kR i,j k
(3.2.19)
82
where
F (k, i, j) = Xi Xj xk + Xi Xj py,k + Xi Yj yk Xi Yj px,k
Xi PX,j yk + Xi PX,j px,k + Xi PY,j xk + Xi PY,j py,k
Yi Xj yk + Yi Xj px,k + Yi Yj xk + Yi Yj py,k
Yi PX,j xk Yi PX,j py,k Yi PY,j yk + Yi PY,j px,k
+ PX,i Xj yk PX,i Xj px,k PX,i Yj xk PX,i Yj py,k
+ PX,i PX,j xk + PX,i PX,j py,k + PX,i PY,j yk PX,i Py,j px,k
+ PY,i Xj xk + PY,i Xj py,k + PY,i Yj yk PY,i Yj px,k
PY,i PX,j yk + PY,i PX,j px,k + PY,i PY,j xk + PY,i PY,j py,k .
(3.2.20)
Finally, the leads Hamiltonian in Cartesian mapped coordinates is given by:
Hclbath =
(3.2.21)
k nk ,
K{L,R} kK
where nk =
1
4

xk + py,k
2
+ yk + px,k
2
We assume that the dynamics of the all Cartesian variables (system and leads) follow from
Hamiltons equations of motions. Within this classical approximation, one can show that the
Hubbard mapping of the double-Anderson model (and all other models discussed in this work)
P
preserves the normalization i Ni = 1 of the many-body states, an additional restriction to
physical subspace [56]. This sum rule is violated, for example, if the Cartesian mapping is used
for the Dq /Dq operators (see further discussion in subsection 3.3.2). Finally, to complete the
mapping procedure initial conditions are determined. As presented in the previous subsections
we enforce quantum statistics on the initial conditions for each DoF by setting the initial
occupation of the leads, nk , to either 0 or 1 such that its expectation value, averaged over the
set of initial conditions, satisfies the Fermi-Dirac distribution:
nk =
0
1
k > f (k K )
,
k f (k K )

(3.2.22)
1
where k is a random number in the interval [0 1] and f () = e + 1

is the Fermi-Dirac
distribution. For the system, we set N0 = 1 and Nq6=0 = 0 consistent with an empty system.
83
Once the populations are chosen, we set the values for the phase-space variables according to:
1q
4nk + cos (k ) ,
2
q
1
4nk + sin (k ) ,
2
q
1
4nk + sin (k ) ,
2
q
1
4nk + cos (k ) ,
2
xk =
px,k =
yk =
py,k =
(3.2.23)
and
1q
4Nq + cos (q ) ,
2
1q
4Nq + sin (q ) ,
=
2
1q
=
4Nq + sin (q ) ,
2
1q
4Nq + cos (q ) ,
=
2
Xq =
PX,q
Yq
PY,q
(3.2.24)
where k and q are random variables in the interval [0 2]. We note that the above choice

of initial conditions is not unique and other choices that satisfy the relations Xq + PY,q =

4Nq + cos (q ) and Yq PX,q = 4Nq + sin (q ) (and similarly for xk , yk , px,k and
py,k ) yield similar results. However, the above form seems to converge better than other
choices tested and thus, is the procedure adopted.
3.3
Quantum transport in impurity models:

Assessing the Hubbard mapping
In what follows, the parameter tkm , which represents the coupling strength (hybridization)
between the system and the leads, is determined from the spectral density:
L/R () =
1
2 ,
L/R
1 + eA(B)

1 + eA(+B)
(3.3.1)
,
L/R k
where L = R =
A = 5/ and B = 20. With these definitions,
=
.
2
We use 400 modes to represent each lead with = 0.075. This was sufficient to converge the
results up to a time tmax = 10
h/ for which a steady-state was observed.
tk/k
84
3.3.1
Resonant level model
In order to assess the accuracy of the Hubbard mapping, we first apply it to the resonant level
model (subsection 1.3.1). In this case (of non-interacting electrons), it can be shown that the
Hubbard mapping is exact and the results do not depend on the choice of the Langer correction
parameter, , despite the fact that enters the initial sampling of phase-space coordinates (cf.
equations (3.2.23) and (3.2.24)). In figures 3.3.1, 3.3.2 and 3.3.3, the time dependent current
for different values of the bias voltage (), gate voltage (Vg ), and different temperatures are
presented. As clearly evident, the agreement with the exact quantum mechanical results [62]
(Solid lines, mostly hidden by the symbols, which are the results obtained from the Hubbard
mapping.) is excellent for all parameters and temperatures studied, as it should be. In fact,
the quality of the Hubbard mapping approximation is identical in this regard to the Cartesian
mapping of Fermions.
3.3.2
Anderson impurity model
The resonant level model is a necessary test of the new Hubbard mapping approach but not a
sufficient one. Since the Hubbard mapping aims at systems with electron-electron and also hopping interactions, a more subtle assessment is required. The simplest model that accounts for
electron-electron correlations is the single site Anderson impurity model, for which the system
Hamiltonian is given in equation (1.3.10) and the interaction Hamiltonian in equation (1.3.11).
Here, the single QD is coupled to both reservoirs and the system can accommodate only 2 electrons in 4 many-particle states. Following the derivation outlined above, the electron creation
(d ) and annihilation (d ) operators are given in terms of the many-body states creation (Dq )
and annihilation (Dq ) operators, as:
d = D1 D0 + D3 D2 ,
d = D0 D1 + D2 D3 ,
d = D2 D0 D3 D1 ,
d = D0 D2 D1 D3 ,
(3.3.2)
where |0i = |0, 0i , |1i = |, 0i , |2i = |0 i , and |3i = |i .
85
Current [e/h]
0.3
0.2
e=2
e=6
e=10
0.1
0
0
_3
Time [h /]
Figure 3.3.1: Time dependent currents for the resonant level model. The plot presents results for different
bias voltages () at = 3/ and Vg = 0.
eVg=0
eVg=
Current [e/h]
0.3
0.2
eVg=3
0.1
0
0
_3
Time [h /]
gate voltages (Vg ) at = 3/ and e = 2.
86
T=2
T=
T=/3
T=/5
Current [e/h]
0.3
0.2
0.1
0
0
_3
Time [h /]
temperatures (T ) at e = 2 and eVg = 0.
The system and interaction Hamiltonians are described by:
sys = D D1 + D D2 + ( + + U )D D3 ,
H
1
2
3
(3.3.3)
and
int =
H
X
X
tk ck D0 D1 + D1 D0 ck + ck D2 D3 + D3 D2 ck
tk ck D0 D2 + D2 D0 ck ck D1 D3 D3 D1 ck ,

(3.3.4)
respectively. We note in passing, that for a single site Anderson impurity model, the Hubbard
Hamiltonian (cf. equation (3.3.3)) is equivalent to the slave-boson Hamiltonian [278, 279]. The
latter sets out by defining directly the Dq and Dq operators rather than the Hubbard operators,
X qp , and thus, it is straightforward to define the statistics of the Dq /Dq operators in this case.
The mapping to classical phase-space variable follows the rules outlined above for the doubleAnderson model. The mapped system and interaction Hamiltonians for the Anderson impurity
model are given by:
Hclsys = N1 + N2 + ( + + U )N3
(3.3.5)
87
and
Hclint =
tk {F (k , 0, 1) + F (k , 2, 3)}
tk {F (k , 0, 2) F (k , 1, 3)} ,
(3.3.6)
respectively, where Nq and F are given by equations (3.2.18) and (3.2.20). The mapped bath
Hamiltonian is identical to that given by equation (3.2.21) for the double Anderson model.
In figures 3.3.4 and 3.3.5 we compare the Hubbard mapping results for the Anderson model
with the NEGF EOM approach [183, 167], which is considered accurate in the Coulombblockade regime studied. Figure 3.3.4 shows the depends of the steady-state current on the
Langer parameter,. As clearly evident, the results are quite sensitive to the value of . We
find that the best agreement is achieved for = 3/2, which described properly the S-shaped
curve of the current versus voltage Coulomb-blockade characteristics. This value for is close
to = 2[ 3 1] 1.464 derived by Cotton and Miller for the Meyer-Miller mapping [280].
Figure 3.3.5 compares the Hubbard mapping results with the NEGF approach for different
values of U . Also shown are results based on the Cartesian mapping of Fermions described
in subsection 3.1.2 and a mixed mapping for which the statistics of the operators Dq /Dq
were imposed. For the single site Anderson impurity model D0 and D3 are bosons and D1 and
D2 are fermions and thus, we assume the Meyer-Miller [274] mapping for the Bosons and the
Hubbard mapping for the Fermions of both the system and the leads. The Cartesian mapping
slightly out-performs the Hubbard mapping (which is rather good) for the Anderson impurity
model for all values of U . However, for the double-Anderson model this is not the case, as
further shown below. Comparing the results of the mixed mapping to the Hubbard mapping
approach, it is clear that the latter performs much better in describing the Coulomb-blockade
effect. Furthermore, the mixed mapping approach is insensitive to the value of and thus, the
Langer parameter cannot be used to improve its performance [280].
We further note that when operator statistics is considered, one can replace the Hubbard
mapping with the Cartesian mapping since the interaction Hamiltonian contains only even
number of products of Fermion operators (this is true for both the single- and double-site
Anderson models). We find that the Cartesian mapping combined with the Meyer-Miller mapP
ping of Bosons violates the normalization sum rule i Ni = 1 and thus can lead to unphysical
P
d
situations. Moreover, it imposes a different sum rule for which dt
(N1 + N2 + k nk ) = 0,
implying that the total population of the Fermion mapped DoF is conserved, imposing another
unphysical constraint.
88
Stationary current [e/h]
=1
=3/2
=2
U=8
0.5
0.4
0.3
0.2
0.1
0
0
10
12
Bias voltage []
Figure 3.3.4: I V curves obtained from the Hubbard mapping for different values of compared to the
Stationary current [e/h]
NEGF approach (black solid line) for = = 12 U , = 5/ and U = 8.
U=4
U=6
U=8
0.5
0.4
0.3
0.2
0.1
0
0
10
12
Bias voltage []
Figure 3.3.5: I V curves for different values of U for = = 12 U , = 5/, = 3/2 and L/R = 12 . Solid
lines: NEGF. Circles: Hubbard mapping. Triangles: Cartesian mapping of Ref. [89]. Dashed curve: Mixed
mapping for U = 8.
89
3.3.3
Double Anderson model
In figures 3.3.6, 3.3.8, 3.3.7, and 3.3.9 we plot the differential conductance as a function of
the bias voltage for the double-Anderson model for various typical model parameters.
The
results obtained from the Hubbard and Cartesian mappings are compared to those calculated
using the many-particle ME approach [271, 116] and the equations-of-motion approach to NEGF
(approximation 2 of sub-subsection 2.3.2.2). For the weak coupling, high temperature regime,
the ME approach is fairly accurate [205] and thus, is used as a measure of the quality of our
mapping approximations. The NEGF results are also provided in order to compare our results
with an approach often used in nonequilibrium transport problems. The specific implementation
of the NEGF approach is described in detail in section 2.3. In the figures the zero-, one-, twoand three-electron states are denoted |0i, |Ii, |IIi and |IIIi respectively. The notation |Ii |IIi
means that the peak belongs to a transition from one of the one-electron states to one of the
two-electron states.
For the symmetric case ( = = = ), the Cartesian mapping that was successful
in describing the Coulomb blockade in the single site Anderson impurity model, fails to capture
some of the peaks in the conductance. It recovers successfully conductance peaks associated
with transitions between an empty system and one-electron states, but fails to capture transitions between one-electron and two-electron states [167]. This failure has also been reported
for many of the NEGF closures [167] described in subsection 2.3.2. On the other hand the
Hubbard mapping describes the position of all peaks nearly quantitatively, while it deviates
from the ME results with respect to the width and amplitude of the conductance peaks.
For the asymmetric case ( = 6= = ), the overall performance of the classical
Hubbard approach is similar to the symmetric case and out performs the NEGF and the
Cartesian mapping for both V = 0 and V = 0.8U . Importantly, it captures the peak associated
with the transition from a two-electron to a three-electron state and also does not predict
negative differential conductance that appears in the NEGF calculations but not in the ME
approach.
3.4
Concluding Remarks
In this chapter we have presented three different mappings that allow the study of nonequilibrium quantum transport in molecular junctions by using Hamiltons classical equations of
motions. The basic idea behind the semiclassical approach is to transform a general second quantized many-electron Hamiltonian into a classical one. This is achieved by defining
a prescription which transforms the electronic operators into classical functions that correctly
account for the anti-commutativity of the Fermion operators.
90
0.4
0.3
|0|I
0.2
|I|II
|0|I
0.1
0
0
|I|II
0.5
Bias voltage [U]
1.5
Figure 3.3.6: Plots of the differential conductance versus the bias voltage for the double-Anderson model.
The black, red, blue and green lines represents the differential conductance calculated via the ME approach,
NEGF, Hubbard mapping and Cartesian mapping respectively. Presented are the results for the symmetric
bridge, = = = = 0.35U . Parameters are: V = 0, h = 0.3U , L = L = R = R = 0.015U
and = 40/U.
0.3
|I|0
0.2
0.1
|I|II
0
0
0.5
Bias voltage [U]
1.5
Figure 3.3.7: Same as figure 3.3.6 but for V = 0.8U and h = 0.7U .
91
0.4
0.3
0.2
|I|II
|II|I
|I|0
0.1
|II|III
0
0
0.5
Bias voltage [U]
1.5
Figure 3.3.8: Plots of the differential conductance versus the bias voltage for the double-Anderson model. The
black, red, blue and green lines represents the differential conductance calculated via the ME approach, NEGF,
Hubbard mapping and Cartesian mapping respectively. Presented are the results for the asymmetric bridge,
= = 0.15U and = = 0.2U . Parameters are: V = 0, h = 0.3U , L = L = R = R = 0.015U
and = 40/U.
0.3
0.2
|0|I
|I|II
0.1
0
0
0.5
Bias voltage [U]
1.5
Figure 3.3.9: Same as figure 3.3.8 but for V = 0.8U and h = 0.7U .
92
Following a short introduction of the three different mappings, we centered on the Hubbard mapping, as it is the only one that captures all the conductance peaks in a system that
have electron-electron intra- and inter-site Coulomb repulsion and electron hopping interactions. Opposed to its two predecessors, the Hubbard mapping requires one to rewrites the
general second-quantized many-electron Hamiltonian in the language of the Hubbard operators first, and only then map the resulting Hamiltonian to a classical Hamiltonian, where the
dynamics follow Hamiltons equations of motion.
In contrast to the action-angle and the Cartesian mapping, which provides a recipe to map
products of even number of Fermion operators, the present mapping can deal with an odd
number of many-body creation/annihilation operators and thus copes with situations where 3
operators terms are present in the Hamiltonian, which is the case considered here (cf. equation (3.2.9)).
The Hubbard mapping approach ignores the quantum statistics (Fermion or Boson) of
the many-body creation/annihilation operators and treats all operators on the same footing.
In fact, when Boson and Fermion statistics are imposed (the mixed mapping described in
subsection 3.3.2), we find that the mixed mapping fails to recover the Coulomb-blockade effect
in the single site Anderson impurity model. Perhaps, this can be traced to the fact that the
mixed mapping is insensitive to the value of the Langer parameter () and thus, cannot
be used to improve the performance. In this regard the results obtained from the Hubbard
mapping are sensitive to ; the best results were obtained for = 23 .
The Hubbard mapping approach provides excellent agreement (similar to the quaternion
mapping [89]) for the time dependent current for the resonant level model in comparison to exact quantum mechanical calculations for a wide range of model parameters, including different
temperatures, gate voltages and bias voltages. While the quaternion mapping (complemented
by the Gaussian scheme) performs better in describing the Coulomb-blockade effect in the
single site Anderson impurity model, the Hubbard mapping shows significant improvements
for the more evolved double Anderson models, which includes electron-electron and hopping
interactions. The results obtained from the Hubbard and Cartesian mappings are compared to
those calculated using the many-particle ME approach and the symmetric-EOM approach to
NEGF. For the weak coupling, high temperature regime, the ME approach is consider to be
accurate and thus, is used as a measure of the quality of our mapping. While the Cartesian
mapping and the NEGF approaches fail to capture all the conductance peaks predicted by
the ME approach, the Hubbard mapping reproduces the position of all peaks nearly quantitatively, while it deviates from the ME results with respect to the width and amplitude of the
conductance peaks.
The success of the Hubbard mapping may not be very surprising owing to the fact that
by expressing the second-quantized Hamiltonian in terms of the Hubbard operators, one re93
ally goes from a single-particle point-of-view to many-body states framework. By writing the
Hamiltonian in the many-body states basis, all the different electron-electron interactions are
accounted for explicitly, and in that regard, the Hamiltonian is a noninteracting Hamiltonian.
94
Chapter 4
Summary and Outlook
A complete understanding of transport through molecular junctions and quantum dots in
nonequilibrium continues to elude our grasp. While experiments have reached a certain level of
consistency where most results are reproducible, theory still seems to have some loose ends
that need tying up. The limitations of todays numerically exact solutions are often too severe for them to be useful in practice (usually, exponential scaling of computational resources
and/or time with system size), thus, most theoretical treatments of quantum transport rely on
approximations of some sort. Although approximated, not all approaches used in theoretical
treatments of quantum transport have a low scaling, therefore, the development of low scaling
methods is of major importance.
In this work we examined and developed two low scaling, approximated methods, built for
transport calculations in nanojunctions: (a) the equations-of-motion technique for the nonequilibrium Green function, which scaling depends on the level of truncation and (b) a semiclassical
approach, which scales linearly with the size of the system.
In chapter 2 we analyzed the validity of the EOM technique for the NEGF. We proved that
the latter may results in a NEGF that does not obey relations and symmetries dictated by
definition. At this time, we still do not have any measure to predict whether a given set of
equations obtained via a certain closure to the EOM will indeed break the NEGF symmetry.
Furthermore, we demonstrated that expectation values of physical quantities calculated with
the symmetry-broken NEGF may result in non-physical values.
Besides, we suggested a symmetry restoration scheme, which restores the symmetry to the
NEGF, but do not alter the NEGF in anyway if the latter fulfills all relations dictated by
definition to begin with. We believe this restoration step should be added as one more
mandatory approximation on top of the approximations used to obtain a certain closure to the
set of EOM. This aforementioned procedure was illustrated on the Anderson and the double
Anderson models, where we showed how this symmetry breakage yields unphysical results, and
how the restoration step remedies this undesirable effect.
95
Additionally, we developed two closures for the EOM of the NEGF for the double Anderson
model, one of which out-performs the closures found in the literature. Our conclusion was that
the extension of an approximation that works well for smaller systems (such as the Anderson
model in our case) to bigger systems (the double Anderson model) is not trivial if at all possible.
This, of course, is a major drawback of a method vastly used in transport calculations. Ongoing
and future work beyond the scope of this thesis includes the examination of the symmetries of
the higher order correlation functions obtained during the derivation of the EOM. We believe
that formulating truncation and decoupling schemes that preserve the symmetries of the these
higher order correlation will resolve the symmetry breaking problem of the single-particle NEGF
and no extra symmetrization steps will be required.
In chapter 3 a semiclassical approach was formulated to allow the study of nonequilibrium
quantum transport in molecular junctions. The key idea behind the approach, based on earlier
works of McCurdy, Meyer, and Miller (MMM) and later by Miller and White (MW), was to
transform a general second quantized many-electron Hamiltonian into a classical one. This
was achieved by defining a prescription which transforms the electronic operators into classical
functions that correctly account for the anti-commutativity of the Fermion operators. The
classical degrees-of-freedom are assumed to follow Hamiltons equations of motion.
In this thesis, we focused on the Hubbard mapping, in which the second quantized Hamiltonian is first reformulated in the language of the many-body Hubbard operators and only
then these operators are mapped onto classical functions. The method was illustrated on three
model Hamiltonians; the single resonant level, the Anderson model and the double Anderson model. We showed that the Hubbard mapping approach provides excellent agreement for
the time dependent current for the resonant level model in comparison to exact quantum mechanical calculations for a wide range of model parameters, including different temperatures,
gate voltages and bias voltages. While its performance is not as good as the Cartesian mapping (complemented by the Gaussian scheme) for the single site Anderson impurity model,
the Hubbard mapping showed significant improvements for the more evolved double Anderson
model, which includes electron-electron and hopping interactions.
The results obtained from the Hubbard mapping were compared to those calculated using
the many-particle ME approach and the symmetric-EOM approach to NEGF. For the weak
coupling, high temperature regime, the ME approach is considered to be accurate and thus,
was used as a measure of the quality of our mapping. We demonstrated that the Hubbard
mapping reproduces the position of all peaks nearly quantitatively, while it deviates from the
ME results with respect to the width and amplitude of the conductance peaks. It is interesting
to see whether the addition of a quantization scheme to the Hubbard mapping (such as the
Gaussian binning scheme used for the Cartesian mapping) will improve its results in comparison
to the ME approach.
96
Appendix A
Full derivation of the broken symmetry
in the Anderson model
<
Here we present in greater detail the breakage of the relation G<
() = (G ()) described
in sub-subsection 2.2.5.1. We demonstrate that G<
() is not an imaginary function. We start
by defining the following contour ordered Green function:
G , 0 =
n
oE
iD
,
TC d ( ) d 0
h
(A.0.1)
n
oE
iD
TC n ( ) d ( ) d 0
,
(A.0.2)
h
where
is the opposite spin of . A complete derivation of the EOM for G (, 0 ) can be
found in Ref. [115] pages 172-176. The resulting EOMs (in steady state) are:

G , 0 =
R/A
GR/A
() + gR/A () U GR/A () ,
() = g
R/A
R/A
R/A
GR/A () = g2 () hn i g2 () 1
() GR/A
() ,
<
R
G
() = g () + g () U G () + g () U G () ,
(A.0.3)
(A.0.4)
(A.0.5)
G () = g2
() hn i g2
() R
1 () G ()
R
A
A
A
g2
()
1 () G () g2 () 1 () G () .
97
(A.0.6)
Substituting equations (A.0.3) and (A.0.5) into equation (A.0.6), the lesser projection of G (, 0 )
in Fourier space is given by (omitting () for brevity):
<
R R <
R < A
< A A
G< = P r g2
hn i P r g2
1 g P r g2
1 g P r g2
1 g
R R <
A
R < A
A
P r g2
1 g U P a g2
hn i P r g2
1 g U P a g2
hn i
< A A
A
R R <
A A A
P r g2
1 g U P a g2
hn i + P r g2
1 g U P a g2
1 g
R < A
A A A
< A A
A A A
+P r g2
1 g U P a g2
1 g + P r g2
1 g U P a g2
1 g ,
(A.0.7)
The lesser projection of G (, 0 ) can now be written as (again, in Fourier space and omitting
() for brevity):
<
R
<
<
A
G<
(A.0.8)
= g + g U G + g U G .

<
R
r <
<
a A
A A
R
r < A A
G<
i + g U P g2 hn
i 1 g g U P g2 1 g
= g + g U P g2 hn
R
<
< A
R
r R R <
a A
A A
gR U P r g2
R
i + 1 g
1 g + 1 g g U P g2 1 g U P g2 hn
R < A
A
A
gR U P r g2
1 g U P a g2
hn i + A
1 g
< A A
A
A
gR U P r g2
1 g U P a g2
hn i + A
1 g .
(A.0.9)
Applying the principle of reductio ad absurdum, we assume G<

is imaginary. Since it must
hold for any real value of hn i between 0 and 1, we argue that the term
<
A
R R <
A
A1 = gR U P r g2
hn i + g< U P a g2
hn i gR U P r g2
1 g U P a g2
hn i
R < A
A
< A A
A
gR U P r g2
1 g U P a g2
hn i gR U P r g2
1 g U P a g2
hn i ,
(A.0.10)
is imaginary by itself. Moreover, Since A1 must be imaginary for any value of U , the term
<
A
A2 = gR U P r g2
hn i + g< U P a g2
hn i ,
(A.0.11)
should be imaginary as well. Using the fact that U and hn i are real quantities and by definition
<
g2
and g< are imaginary, for A2 to be imaginary, one requires that its real part vanishes, i.e.,:

<
A
+ Im P a g2
g< = 0.
Im gR P r g2
(A.0.12)
In other words the equality

A
A
R < A
gR <
Im gR P r g2
4 g2 = Im P a g2
0 g ,
(A.0.13)
98
must hold for the assumption that G<

is imaginary to be satisfied. Using the definitions for
R/A
R/A
g and g2 the last equality can be rewritten as:

Im gR P r
if () Im R
4
(
h 4 U ) + Im

R
4
2
A
= Im P a g2
if () Im R
0
(
h 0 )2 + Im R
0

2 ,
(A.0.14)
R
where 0 = + Re R
0 and 4 = + Re 4 . Starting with the L.H.S. of equation (A.0.14),
we look at gR P r
gR P r =
gR
R
R
1 + g2 R
1 g U
R
1U
h
R
0 +h
U R
4
R
h
U 4

R
h
h
R
U R
0
4 + 1 U
h
4 U i ImR
4
h
0 i ImR
0
h
4 U i ImR
4
R
+ U Re R
1 + U i Im1
.
(A.0.15)

Denote a0 = h
4 U and bx = Im (rx ) with x = 0, 1, 4,
0 , a1 = Re R
1 , a4 = h
a4 ib4
(a0 ib0 ) (a4 ib4 ) + U a1 + iU b1
a4 ib4
=
.
a0 a4 b0 b4 + U a1 i (a0 b4 + a4 b0 U b1 )
gR P r =
(A.0.16)
Denote D = a0 a4 b0 b4 + U a1 and E = a0 b4 + a4 b0 U b1 , so we can rewrite the last equation

as:
gR P r =
a4 ib4 (a4 ib4 ) (D + iE) a4 D + b4 E i (b4 D a4 E)

=
=
.
D iE
D2 + E 2
D2 + E 2
Finally

Im gR P r =
b 4 D a4 E
,
D2 + E 2
(A.0.17)
(A.0.18)
and the L.H.S. of equation (A.0.14) is given by:

Im gR P r
if () Im R
4
(
h 4 U )2 + Im R
4

2
if () b4
b 4 D a4 E
.
(a4 )2 + (b4 )2 D2 + E 2
(A.0.19)
99
A:
Now we turn to analyze the R.H.S. of equation (A.0.14). We start with evaluating P a g2
A
P a g2
A
g2
1
=
=
A
A
A
A
1 + g2 1 g U
1U
h
+
U A
4
h
A
=
h
U
a0 + ib0
(a0 + ib0 ) (a4 + ib4 ) + U a1 iU b1
a0 + ib0
a0 a4 b0 b4 + U a1 + i (a0 b4 + a4 b0 U b1 )
a0 + ib0
D + iE
(a0 + ib0 ) (D iE)
D2 + E 2
a0 D + b0 E + i (b0 D a0 E)
.
D2 + E 2
=
=
=
=
=
h
A
0

A
A
4 h
A
0 + 1 U
(A.0.20)
To go from the second line to the third line in equation (A.0.20) we used Im (rx ) = Im (ax ).
Finally:

b 0 D a0 E
A
.
(A.0.21)
Im P a g2
=
D2 + E 2
The R.H.S. of equation (A.0.14) is thus,

A
Im P a g2
if () Im R
0
(
h 0 )2 + Im R
0

2
b0 D a0 E if () b0
,
D2 + E 2 (a0 )2 + (b0 )2
(A.0.22)
and the equality now reads:

b 4 D a4 E b 0 D a0 E
b0
=
,
2
2
2
2
2
D + E (a0 ) + (b0 )2
(a4 ) + (b4 ) D + E
2
or
b4
b24 D a4 b4 E
(a4 )2 + (b4 )2
b20 D a0 b0 E
(a0 )2 + (b0 )2
(A.0.23)
(A.0.24)
Substituting D = a0 a4 b0 b4 + U a1 and E = a0 b4 + a4 b0 U b1 one can easily show that the

equality does not hold. Thus, G<
() is not an imaginary function as it should be by
definition.
We further argued that a simpler closure (as used for example, in Ref. [183]) will not violate
the symmetries of the GFs. In what follows we show that under the simpler closure, indeed,
100
<
G<
() = (G ()) . Our starting point is the same. Define the contour ordered GFs:
G , 0 =
n
oE
iD
TC d ( ) d 0
,
h
(A.0.25)
n
oE
iD
TC n ( ) d ( ) d 0
.
(A.0.26)
h
Following the approximations of Ref. [183], the resulting EOMs (in Fourier space) are:

G , 0 =
R
R
h
R
0 () G () = 1 + U G () ,
GR () =
1
h
U R
0 ()
We define
hn i .
(A.0.27)
(A.0.28)
1
,
h
R
0 ()
(A.0.29)
R
g2
() =
1
,
h
U R
0 ()
(A.0.30)
R
0 () =
|tk |2
.
+
h
+
i0
k
k{L,R}
(A.0.31)
gR () =
Rewriting equations (A.0.27) and (A.0.28) in terms of the given definitions we get:
R
GR () = g2
() hn i ,
(A.0.32)
R
R
R
GR
() = g () + g () U G () ,
(A.0.33)
Substitute equation (A.0.32) into equation (A.0.33):

R
R
R
GR
i g2 () .
() = g () + g () U hn
(A.0.34)
The lesser Green function can also be derived via the Langreth rules (omitting () for brevity):
<
R
<
<
A
G<
i g2 + g U hn
i g2
= g + g U hn
(A.0.35)
<
<
A
R < A
where g< = gR <
0 g , g2 = g2 0 g2 , and 0 is defined by equation (2.2.91). By definition
g< is imaginary, hence, for G<
to be imaginary one requires that:
A
R < A
A
A1 = gR U hn i g2
0 g2 + gR <
i g2 ,
0 g U hn
(A.0.36)
R < g A and g R < g A are imaginary quantities, for G< to be imaginary,

be imaginary. Since g2
0
0 2
101
the real part of A1 needs to cancel, i.e.,:

R < A
A
A
Im gR g2
0 g2 = Im g2
gR <
0 g .
(A.0.37)
Define

0 = + Re R
0 ,
(A.0.38)
= Im R
0 ,
2
(A.0.39)
equation (A.0.37) can be rewritten as:

2
(
h 0 )2 +
2
<
0
(
h 0 U )2 +
2
(
h0 U )2 +(
2
2
<
0
(
h0 )2 +( 2 )
.
(A.0.40)
Obviously the real part of A1 cancels, hence G<

is imaginary and fulfills the symmetry
<
<
G () = (G ()) .
102
Appendix B
Brief summary of the derivation of the
NEGFs EOM: the double Anderson
model
We define the single-particle contour ordered NEGF:
0
0
i
hG
, = TC d ( ) d
oE
(B.0.1)
0
Deriving the EOM for G
(, ) will produce new and higher order correlation functions. We
will follow the naming scheme presented in table B.1.
Notation
0
i
hGqr
(, )
qr
i
hFk
(, 0 )
qr
i
hFk
(, 0 )
0
i
hGqrs
(, )
0
i
hFqrs
k (, )
0
i
hGqrst
(, )
0
i
hFqrst
k (, )
0
i
hFqrst
k (, )
0
i
hFqrst
lk (, )
0
i
hFqrst
lk (, )
0
i
hGqrst
(, )
Correlation function
TC dq ( ) dr ( 0 )
oE
TC ckq ( ) dr ( 0 )
oE
TC dq ( ) ckr ( 0 )
oE
TC nq ( ) dr ( ) ds ( 0 )
oE
TC nq ( ) ckr ( ) ds ( 0 )
n
oE
TC dq ( ) dr ( ) ds ( ) dt ( 0 )
oE
TC dq ( ) ckr ( ) ds ( ) dt ( 0 )
oE
TC ckq ( ) dr ( ) ds ( ) dt ( 0 )
oE
D
D
n
n
TC dq ( ) clr ( ) cks ( ) dt ( 0 )
TC clq ( ) dr ( ) cks ( ) dt ( 0 )
oE
oE
TC nq ( ) nr ( ) ds ( ) dt ( 0 )
oE
Table B.1: This naming convention will be used throughout our derivation of the double Anderson model.
The subscripts {, , , } are the sites of the system indexes. The subscripts {l.k} are the levels of the leads
indexes while the superscripts {q, r, s, t} stands for spin up or spin down.
103
Writing down the EOM for the contour ordered Green function we get:
0

G
(, )
= C 0
d ( ) d 0
i
h
+ TC
i
h
)+
(B.0.2)
0
D

n
oE
G
i
(, )
= C 0
TC d ( ) d 0
n
oE
i D
h TC d ( ) d 0
h
D
n
oE
i
U TC n ( ) d ( ) d 0
h
D
n
oE
i X
s
V
TC ns ( ) d ( ) d 0
h
s{}
n
oE
i X D
tk TC ck ( ) d 0
.
h
kL
(B.0.3)

0
i
h G
= C 0
+ h G
,
,

+U G
, 0 + V
G
, 0

+V
G , 0 +

tk Fk
, 0 .
(B.0.4)
kL
(, 0 ) , G

0

0
To continue, we now need to write the EOM for Fk
(, ) , G (, )
0
0
and G
(, ). We start with Fk (, ):
(, 0 )

Fk
= TC
ck ( ) d 0
i
h
i
h
)+
(B.0.5)
(, 0 )
D
n
oE
D
n
oE
Fk
i
i
t
= k TC ck ( ) d 0
T
d
(
)
d
,
C
k

h
h

(B.0.6)
= k Fk
, 0 + t
k G , ,
i
h k Fk
( ) = t
k G ( ) .
(B.0.7)
Define the leads noninteracting Green function:

!
i
h k gkL ( 1 ) = C ( 1 ) ,
(B.0.8)
104
so the last equation takes the form:
Fk
0
gkL ( 1 ) t
k G 1 , d1 ,
(B.0.9)
0
and the equation of G
(, ) can now be rewritten as:

0
0
= C 0
+ h G
i
h G
, + U G ,
,

+V
G
, 0 + V
G , 0

0
+
0 ( 1 ) G
d1 ,
1
(B.0.10)
where
0 ( 1 ) =
|tk |2 gkL ( 1 ) .
(B.0.11)
kL
(, 0 ), where = {, }.
We now continue and derive the EOM for G
(, 0 )

G
= C 0
n ( ) d ( ) d 0
i
h
hn i + TC
*
(
!
)+

0
+ TC n ( )
d ( ) d
,
(B.0.12)
i
h
)+
(, 0 )

G

= C 0
hn i + h
G
, 0 h
G
, 0

X

+
tk
Fk
, 0 tk
Fk
, 0
kL

+ G
, 0 + h G
, 0 + U G
, 0

+V
G
, 0 + V
G
, 0 +

tk Fk
, 0 ,
kR
(B.0.13)
where
i
h
hn i =
2
<
G
()
d.
(B.0.14)
105
Note that for the special case where = we get:

(, 0 )

G
, 0
G
, 0 h
G
i
h
= C 0
hn i + h

X
, 0
Fk
, 0 tk
+
Fk
tk
kL

+ G
, 0 + h G
, 0 + U G
, 0
s
s
V
G
, 0 +
s{
}

tk Fk
, 0 .
(B.0.15)
kL
0
The EOM for G
(, ) is:
i
h
i
h
0
E
D

G
(, )
= C 0
hn i
d d
*
(
!
)+

0
+ TC
n ( ) d ( ) d
*
(
!
)+

0
+ TC n ( )
d ( ) d
,
(B.0.16)
0

D
E

G
(, )
0
= C 0
hn i
d d + h
G ,

X
0

0
+
tk F
k , tk Fk ,
kL

0

0
G
,
+
U
G
,

+V
G , 0 + V
G
, 0 +
0
tk F
k , ,
kR
(B.0.17)
where
D
d d
i
h
=
2
<
G
()
d.
(B.0.18)
For the special case where = we find:

0
G
, = 0.
(B.0.19)
106
(, 0 ) and F (, 0 ):
For completeness we also present the EOM for Fk
k
(, 0 )
Fk
=
i
h
t
)+

TC
n ( ) ck ( ) d 0
*
(
)+
!

+ TC n ( )
,
ck ( ) d 0
(B.0.20)
i
h
(, 0 )

X

Fk

= h
Fk
, 0 h
Fk
, 0 +
tl
Flk
, 0
t
lL

0
tl
Flk
, 0 + k Fk
, 0 + t
k G , ,
lL
(B.0.21)
0

D
E
F
k (, )
i
h
= C 0
n ( ) ck ( ) d 0
d ck + TC
t
!
*
(
)+

0
ck ( ) d
+ TC n ( )
,
(B.0.22)
)+
0
E

D
F
k (, )
= C 0
d ck
i
h
t

X

0

0

0
+h F
,
h
F
,
t
F
+
,
k
k
l lk
t
l Flk
+ k F
k
lL

0
+ t
G
,
,
k
lL
(B.0.23)
where
D
d ck
i
h
=
2
(Fk
())< d.
(B.0.24)
As can be seen, new 2particle and 3particle Green functions (denoted G, F and G, F respectively), emerge. To close the set of equations, one should either derive the EOM for each
and every new correlation that appears or resort to approximations. Approximations usually
include ignoring certain correlations or decoupling higher order correlations in terms of lower
order ones.
107
108
Appendix C
Calculating the poles of the retarded
NEGF
Using the assumptions described in sub-subsection 2.3.2.2, the EOM for the retarded NEGF of
the unperturbed system are (in Fourier space):

G
h )1 1 + hG
= (
+ U G + V G + V G ,
G

= = h
1

hG
+ U G + V G + V G ,

h
= hG
,
V G
h
V G
= hG d d, ,

h
= hG
,
U G
(C.0.1)

(
h U ) G
= hn i + hG
,
(
h V ) G
= n + hG ,

(
h V ) G
= n + hG
.
(C.0.2)
It is clear that the equations for the 2-particle GF close among themselves, so a simple substitution yields:

G
h2
= h
V
(
h U )
G
V
= h
!1
h2
(
h V )
109
!1
h hn i
,
(
h U )
h n
(
h V )
E
d d, ,

= h
G
U

= h
G
V
h d d,
n
h
V
,
h2
hn i ,
V

G
= h
h
V
(
h V )
h
V
h2
h n
h2
(
h V )

= h
G
U
!1
h2
h
U
(C.0.3)
n .
Define:
x/ = / ,
(C.0.4)
xv/v = / + V,
(C.0.5)
xu/u = / + U,
(C.0.6)
and rewrite the EOM:

G
h x )1 1 + hG
= (
+ U G + V G + V G ,
G
x
= h
1

hG
+ U G + V G + V G ,

G
=
h hn i
h xu ) h2
h
xv (

,
h xv )
h n d d, (
G
=
h xv ) h2
h
xv (

G
=
h n
h
h xv ) h2
xu (

hn i h
xv

G
=
E
,
,
h
xv h d d,
(
h xv ) h
xv h2

D

G
=
E
h
xu
(
h xv ) h
xu h2

(
h xu ) h
xv h2

G
=
(C.0.7)
.
(C.0.8)
110
We now substitute the set of equations (C.0.8) into equations (C.0.7):

G
(
h x )
h2
h
x

hU G
hV G
h
x
h
x
1 +

hV G
+
h
x
+

+ U G
+ V G + V G .
(C.0.9)
Finally:
G
=
(
h x ) h
x h2

h2 V n
D
+
+
D

h
h xu ) h2
xv (

+
h
xv
(
h xv ) h
xv h2
n V h
x
+
hV d d, (
h xv )
h
h xv ) h2
xv (

hn i

(
h xv ) h
xu
h2

h
x
(
h xu ) h
xv h2
h
xu

1 +
h
h xv ) h2
xu (
U hn i h
xv
h2 V
h2 U n
h
h xv ) h2
xv (
n V h
x
+
1
hV d d,
E
h
x
(
h xv ) h
xv h2

(C.0.10)
.
From the last equation we see that the NEGF has poles at:
(
h x ) h
x h2 = 0,
(C.0.11)
(
h xv ) h
xv h2 = 0,
(C.0.12)
(
h xu ) h
xv h2 = 0,
(C.0.13)
(
h xv ) h
xu h2 = 0,
(C.0.14)
or equivalently:

r

2
1
=
x + x x x + 4h2 ,
2
(C.0.15)
PG3,4
"
#
r

2
1
2
=
xv + xv xv xv + 4h ,
2
(C.0.16)
PG5,6
#
r

2

1
=
xv + xu xv xu + 4h2 ,
2
(C.0.17)
PG7,8
r

2
1
=
xu + xv xu xv + 4h2 .
2
(C.0.18)
PG1,2
"
"
"
111
112
Appendix D
Complimentary information for the
Hubbard mapping
The double-Anderson model can host up to 4 electrons in different 16 many-body
states. In

E

the occupation number (ON) vector representation these are denoted by n n n n . We
represent these states in Fock space as {|0i , |1i , ..., |15i}, and denote them as follows:
We have one empty state:
|0i = |0000i .
(D.0.1)
We have four 1-electron states:
|1i = |1000i ,
(D.0.2)
|2i = |0100i ,
(D.0.3)
|3i = |0010i ,
(D.0.4)
|4i = |0001i .
(D.0.5)
|5i = |1100i ,
(D.0.6)
|6i = |0110i ,
(D.0.7)
|7i = |0011i ,
(D.0.8)
|8i = |0101i ,
(D.0.9)
|9i = |1010i ,
(D.0.10)
|10i = |1001i .
(D.0.11)
We have six 2-electron states:
113
We have four 3-electron states:

|11i = |1110i ,
(D.0.12)
|12i = |1101i ,
(D.0.13)
|13i = |1011i ,
(D.0.14)
|14i = |0111i ,
(D.0.15)
and one fully occupied state:

|15i = |1111i .
To express the electron creation and annihilation operators in terms of the Hubbard operators,
it is beneficial to calculate the operation of the electronic operator on the many-body states.
For example, let us compute the effect of d on the many-body states |0i to |15i:
d |0000i = d |0i = |0100i = |2i D2 D0 = X 2,0 ,
d |1000i = d |1i = |1100i = |5i D5 D1 = X 5,1 ,
d |0010i = d |3i = |0110i = |6i D6 D3 = X 6,3 ,
d |0001i = d |4i = |0101i = |8i D8 D4 = X 8,4 ,
d |0011i = d |7i = |0111i = |14i D14

D7 = X 14,7 ,
D9 = X 11,9 ,
d |1010i = d |9i = |1110i = |11i D11
d |1001i = d |10i = |1101i = |12i D12

D10 = X 12,10 ,
d |1011i = d |13i = |1111i = |15i D15

D13 = X 15,13 .
(D.0.16)
The operation of d on any of the other many-body states equals 0. Thus,

d = D2 D0 D5 D1 + D6 D3 + D8 D4
+D14
D7 D11
D9 D12
D10 D15
D13 ,
(D.0.17)
and generally:
d =
X D E
p d q X p,q
p,q
X D E
p d q Dp Dq .
(D.0.18)
p,q
With these simple rules, the full list of the electron operators in terms of the Hubbard operators
is given by:
D4 + D11
D6 + D13
D7 + D12
D8 + D15
D14 ,
d = D1 D0 + D5 D2 + D9 D3 + D10
(D.0.19)
114

d = D0 D1 + D2 D5 + D3 D9 + D4 D10 + D6 D11 + D7 D13 + D8 D12 + D14
D15 ,
(D.0.20)
d = D2 D0 D5 D1 + D6 D3 + D8 D4 + D14
D7 D11
D9 D12
D10 D15
D13 ,
(D.0.21)
d = D0 D2 D1 D5 + D3 D6 + D4 D8 + D7 D14 D9 D11 D10

D12 D13
D15 ,
(D.0.22)
d = D3 D0 D9 D1 D6 D2 + D7 D4 + D11
D5 D14
D8 D13
D10 + D15
D12 ,
(D.0.23)
d = D0 D3 D1 D9 D2 D6 + D4 D7 + D5 D11 D8 D14 D10

D13 + D12
D15 ,
(D.0.24)
d = D4 D0 D10
D1 D8 D2 D7 D3 + D12
D5 + D14
D6 + D13
D9 D15
D11 ,
(D.0.25)
d = D0 D4 D1 D10 D2 D8 D3 D7 + D5 D12 + D6 D14 + D9 D13 D11

D15 .
(D.0.26)
115
116
Bibliography
[1] A. von Hippel, Science 123, 315 (1956).
[2] R. P. Feynman, Engineering and Science 23, 22 (1960).
[3] H. Bcher, K. H. Drexhage, M. Fleck, H. Kuhn, D. Mbius, F. P. Schfer, J. Sondermann,
W. Sperling, P. Tillmann, and J. Wiegand, Molecular Crystals 2, 199 (1967).
[4] A. Aviram and M. A. Ratner, Chem. Phys. Lett. 29, 277 (1974).
[5] G. Binnig, H. Rohrer, C. Gerber, and E. Weibel, Physica B+C 109-110, Part 3, 2075
(1982).
[6] G. Binnig, H. Rohrer, C. Gerber, and E. Weibel, Phys. Rev. Lett. 49, 57 (1982).
[7] J. Moreland and J. W. Ekin, J. Appl. Phys. 58, 3888 (1985).
[8] M. A. Reed, C. Zhou, C. J. Muller, T. P. Burgin, and J. M. Tour, Science 278, 252
(1997).
[9] J. Chen, M. A. Reed, A. M. Rawlett, and J. M. Tour, Science 286, 1550 (1999).
[10] H. Park, J. Park, A. K. L. Lim, E. H. Anderson, A. P. Alivisatos, and P. L. McEuen,
Nature 407, 57 (2000).
[11] J. Park, A. N. Pasupathy, J. I. Goldsmith, C. Chang, Y. Yaish, J. R. Petta, M. Rinkoski,
J. P. Sethna, H. D. Abruna, P. L. McEuen, and D. C. Ralph, Nature 417, 722 (2002).
[12] R. H. M. Smit, Y. Noat, C. Untiedt, N. D. Lang, M. C. van Hemert, and J. M. van
Ruitenbeek, Nature 419, 906 (2002).
[13] J. G. Kushmerick, J. Lazorcik, C. H. Patterson, R. Shashidhar, D. S. Seferos, and G. C.
Bazan, Nano Lett. 4, 639 (2004).
[14] L. H. Yu and D. Natelson, Nano Lett. 4, 79 (2004).
117
[15] L. H. Yu, Z. K. Keane, J. W. Ciszek, L. Cheng, M. P. Stewart, J. M. Tour, and D. Natelson, Phys. Rev. Lett. 93, 266802 (2004).
[16] Xiao, Xu, and N. J. Tao, Nano Lett. 4, 267 (2004).
[17] M. Elbing, R. Ochs, M. Koentopp, M. Fischer, C. von Hnisch, F. Weigend, F. Evers,
H. B. Weber, and M. Mayor, Proc. Natl. Acad. Sci. U.S.A. 102, 8815 (2005).
[18] L. Venkataraman, J. E. Klare, I. W. Tam, C. Nuckolls, M. S. Hybertsen, and M. L.
Steigerwald, Nano Lett. 6, 458 (2006).
[19] M. Poot, E. Osorio, K. ONeill, J. M. Thijssen, D. Vanmaekelbergh, C. A. van Walree,
L. W. Jenneskens, and H. S. J. van der Zant, Nano Lett. 6, 1031 (2006).
[20] E. A. Osorio, K. ONeill, N. Stuhr-Hansen, O. F. Nielsen, T. Bjrnholm, and H. S.
van der Zant, Adv. Mater. 19, 281 (2007).
[21] E. Lrtscher, H. B. Weber, and H. Riel, Phys. Rev. Lett. 98, 176807 (2007).
[22] D. Porath, Y. Levi, M. Tarabiah, and O. Millo, Phys. Rev. B 56, 9829 (1997).
[23] B. C. Stipe, M. A. Rezaei, W. Ho, S. Gao, M. Persson, and B. I. Lundqvist, Phys. Rev.
Lett. 78, 4410 (1997).
[24] J. R. Hahn and W. Ho, Phys. Rev. Lett. 87, 196102 (2001).
[25] Y. Kim, T. Komeda, and M. Kawai, Phys. Rev. Lett. 89, 126104 (2002).
[26] W. Ho, J. Chem. Phys. 117, 11033 (2002).
[27] N. Nilius, T. M. Wallis, M. Persson, and W. Ho, Phys. Rev. Lett. 90, 196103 (2003).
[28] S.-W. Hla, K.-F. Braun, B. Wassermann, and K.-H. Rieder, Phys. Rev. Lett. 93, 208302
(2004).
[29] N. Liu, N. A. Pradhan, and W. Ho, J. Chem. Phys. 120, 11371 (2004).
[30] X. H. Qiu, G. V. Nazin, and W. Ho, Phys. Rev. Lett. 92, 206102 (2004).
[31] J. Repp, G. Meyer, F. E. Olsson, and M. Persson, Science 305, 493 (2004).
[32] S. W. Wu, G. V. Nazin, X. Chen, X. H. Qiu, and W. Ho, Phys. Rev. Lett. 93, 236802
(2004).
[33] J. Repp, G. Meyer, S. M. Stojkovi, A. Gourdon, and C. Joachim, Phys. Rev. Lett. 94,
026803 (2005).
118
[34] J. Repp, G. Meyer, S. Paavilainen, F. E. Olsson, and M. Persson, Phys. Rev. Lett. 95,
225503 (2005).
[35] A. Bannani, C. Bobisch, and R. Mller, Science 315, 1824 (2007).
[36] Y. Bekenstein, K. Vinokurov, T. J. Levy, E. Rabani, U. Banin, and O. Millo, Phys. Rev.
B 86, 085431 (2012).
[37] W. J. Liang, M. P. Shores, M. Bockrath, J. R. Long, and H. Park, Nature 417, 725
(2002).
[38] B.-Y. Choi, S.-J. Kahng, S. Kim, H. Kim, H. W. Kim, Y. J. Song, J. Ihm, and Y. Kuk,
Phys. Rev. Lett. 96, 156106 (2006).
[39] M. Martin, M. Lastapis, D. Riedel, G. Dujardin, M. Mamatkulov, L. Stauffer, and P. Sonnet, Phys. Rev. Lett. 97, 216103 (2006).
[40] P. Liljeroth, J. Repp, and G. Meyer, Science 317, 1203 (2007).
[41] H. Basch and M. A. Ratner, J. Chem. Phys. 119, 11926 (2003).
[42] H. Basch, R. Cohen, and M. A. Ratner, Nano Lett. 5, 1668 (2005).
[43] M. Caspary Toroker and U. Peskin, J. Chem. Phys. 127, 154706 (2007).
[44] M. C. Toroker and U. Peskin, J. Phys. B: At. Mol. Opt. Phys. 42, 044013 (2009).
[45] M. Galperin, A. Nitzan, and M. A. Ratner, Phys. Rev. B 75, 155312 (2007).
[46] R. Hrtle, R. Volkovich, M. Thoss, and U. Peskin, J. Chem. Phys. 133, 081102 (2010).
[47] L. L. Chen, H. Bao, T. Z. Tan, O. V. Prezhdo, and X. L. Ruan, J. Phys. Chem. C 115,
11400 (2011).
[48] R. Volkovich, R. Hrtle, M. Thoss, and U. Peskin, Phys. Chem. Chem. Phys. 13, 14333
(2011).
[49] M. Dorogi, J. Gomez, R. Osifchin, R. P. Andres, and R. Reifenberger, Phys. Rev. B 52,
9071 (1995).
[50] V. Mujica, M. Kemp, A. Roitberg, and M. Ratner, J. Chem. Phys. 104, 7296 (1996).
[51] R. Baer and E. Rabani, J. Chem. Phys. 138, 051102 (2013).
[52] G. Li, B. D. Fainberg, A. Nitzan, S. Kohler, and P. Hnggi, Phys. Rev. B 81, 165310
(2010).
119
[53] R. Hrtle, U. Peskin, and M. Thoss, Phys. Status Solidi B 250, 2365 (2013).
[54] M. Galperin and A. Nitzan, J.Chem.Phys. 124, 234709 (2006).
[55] A. J. White, U. Peskin, and M. Galperin, Phys. Rev. B 88, 205424 (2013).
[56] A. J. White and M. Galperin, Phys. Chem. Chem. Phys. 14, 13809 (2012).
[57] U. Peskin and M. Galperin, J. Chem. Phys. 136, 044107 (2012).
[58] S. Y. Quek, L. Venkataraman, H. J. Choi, S. G. Louie, M. S. Hybertsen, and J. B. Neaton,
Nano Lett. 7, 3477 (2007).
[59] A. P. Jauho, N. S. Wingreen, and Y. Meir, Phys. Rev. B 50, 5528 (1994).
[60] A. Schiller and S. Hershfield, Phys. Rev. B 51, 12896 (1995).
[61] Y. Wang and J. Voit, Phys. Rev. Lett. 77, 4934 (1996).
[62] D. W. H. Swenson, T. Levy, G. Cohen, E. Rabani, and W. H. Miller, J. Chem. Phys.
134, 164103 (2011).
[63] S. R. White, Phys. Rev. Lett. 69, 2863 (1992).
[64] P. Schmitteckert, Phys. Rev. B 70, 121302 (2004).
[65] F. B. Anders and A. Schiller, Phys. Rev. Lett. 95, 196801 (2005).
[66] F. Heidrich-Meisner, A. E. Feiguin, and E. Dagotto, Phys. Rev. B 79, 235336 (2009).
[67] S. G. Jakobs, M. Pletyukhov, and H. Schoeller, Phys. Rev. B 81, 195109 (2010).
[68] D. M. Kennes, S. G. Jakobs, C. Karrasch, and V. Meden, Phys. Rev. B 85, 085113 (2012).
[69] S. Weiss, J. Eckel, M. Thorwart, and R. Egger, Phys. Rev. B 77, 195316 (2008).
[70] L. Mhlbacher and E. Rabani, Phys. Rev. Lett. 100, 176403 (2008).
[71] P. Werner, T. Oka, and A. J. Millis, Phys. Rev. B 79, 035320 (2009).
[72] M. Schir and M. Fabrizio, Phys. Rev. B 79, 153302 (2009).
[73] P. Werner, T. Oka, M. Eckstein, and A. J. Millis, Phys. Rev. B 81, 035108 (2010).
[74] E. Gull, D. R. Reichman, and A. J. Millis, Phys. Rev. B 82, 075109 (2010).
[75] H. Wang, I. Pshenichnyuk, R. Haertle, and M. Thoss, J. Chem. Phys. 135, 244506 (2011).
120
[76] A. A. Golosov, R. A. Friesner, and P. Pechukas, J. Chem. Phys. 110, 138 (1999).
[77] M. Leijnse and M. R. Wegewijs, Phys. Rev. B 78, 235424 (2008).
[78] M. Esposito and M. Galperin, Phys. Rev. B 79, 205303 (2009).
[79] G. Cohen and E. Rabani, Phys. Rev. B 84, 075150 (2011).
[80] L. P. Kadanoff and G. Baym, Quantum statistical mechanics: Greens function methods
in equilibrium and nonequilibrium problems (W.A. Benjamin, New York, 1962).
[81] D. C. Langreth, in Linear and non-linear electron transport in solids (Plenum Press, New
York, NY, 1976).
[82] J. Rammer and H. Smith, Rev. Mod. Phys. 58, 323 (1986).
[83] J. Schwinger, J. Math. Phys. 2, 407 (1961).
[84] L. V. Keldysh, J. Exp. Theor. Phys. 47, 1515 (1964).
[85] U. Weiss, Quantum Dissipative Systems (World Scientific, Singapore, 1999).
[86] H.-P. Breuer and F. Petruccione, The theory of open quantum systems (Oxford University
Press, Oxford ; New York, 2002).
[87] D. W. Swenson, G. Cohen, and E. Rabani, Mol. Phys. 110, 743 (2012).
[88] B. Li and W. H. Miller, J. Chem. Phys. 137, 154107 (2012).
[89] B. Li, T. J. Levy, D. W. Swenson, E. Rabani, and W. H. Miller, J. Chem. Phys. 138,
104110 (2013).
[90] B. Li, W. H. Miller, T. J. Levy, and E. Rabani, J. Chem. Phys. 140, 204106 (2014).
[91] J. Taylor, H. Guo, and J. Wang, Phys. Rev. B 63, 121104 (2001).
[92] M. Brandbyge, J.-L. Mozos, P. Ordejn, J. Taylor, and K. Stokbro, Phys. Rev. B 65,
165401 (2002).
[93] R. Baer, Y. Kurzweil, and L. S. Cederbaum, Isr. J. Chem. 45, 161 (2005).
[94] K. Burke, R. Car, and R. Gebauer, Phys. Rev. Lett. 94, 146803 (2005).
[95] R. Baer and D. Neuhauser, Phys. Rev. Lett. 94, 043002 (2005).
[96] M. Koentopp, C. Chang, K. Burke, and R. Car, J. Phys.: Condens. Matter 20, 083203
(2008).
121
[97] A. Baratz and R. Baer, J. Phys. Chem. Lett. 3, 498 (2012).

[98] N. Kuritz, T. Stein, R. Baer, and L. Kronik, J. Chem. Theory Comput. 7, 2408 (2011).
[99] H. Wang and M. Thoss, J. Chem. Phys. 119, 1289 (2003).
[100] H. Wang and M. Thoss, J. Chem. Phys. 131, 024114 (2009).
[101] H. Wang, I. Pshenichnyuk, R. Hrtle, and M. Thoss, J. Chem. Phys. 135, 244506
(2011).
[102] C. Caroli, R. Combescot, P. Nozieres, and D. Saint-James, J. Phys. C 4, 916 (1971).
[103] C. Caroli, Combesco.R, D. Lederer, P. Nozieres, and Saintjam.D, J. Phys. C 4, 2598
(1971).
[104] C. Caroli, Saintjam.D, Combesco.R, and P. Nozieres, J. Phys. C 5, 21 (1972).
[105] R. Combescot, J. Phys. C 4, 2611 (1971).
[106] U. Meirav, M. A. Kastner, and S. J. Wind, Phys. Rev. Lett. 65, 771 (1990).
[107] B. L. Altshuler, P. A. Lee, and R. A. Webb, Mesoscopic phenomena in solids (Elsevier
Science, Amsterdam; New York; New York, NY, USA, 1991).
[108] B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. G. Williamson, L. P. Kouwenhoven,
D. van der Marel, and C. T. Foxon, Phys. Rev. Lett. 60, 848 (1988).
[109] U. Fano, Phys. Rev. 124, 1866 (1961).
[110] P. W. Anderson, Phys. Rev. 124, 41 (1961).
[111] Y. Meir, N. S. Wingreen, and P. A. Lee, Phys. Rev. Lett. 66, 3048 (1991).
[112] N. S. Wingreen and Y. Meir, Phys. Rev. B 49, 11040 (1994).
[113] C. Jayaprakash, H. R. Krishnamurthy, and J. W. Wilkins, Phys. Rev. Lett. 47, 737 (1981).
[114] S. Lamba and S. K. Joshi, Phys. Rev. B 62, 1580 (2000).
[115] H. Haug and A.-P. Jauho, Quantum kinetics in transport and optics of semiconductors
(Springer, Berlin; New York, 1996).
[116] S. Datta, Quantum transport: atom to transistor (Cambridge University Press, 2005).
[117] G. Stefanucci and R. van Leeuwen, Nonequilibrium many-body theory of quantum systems:
a modern introduction (Cambridge: Cambridge University Press, 2013).
122
[118] G. D. Mahan, Many-particle physics (Plenum Press, New York, 1990).

[119] T. Arai, M. H. Cohen, and M. P. Tosi, Phys. Rev. B 15, 1817 (1977).
[120] Y. Meir and N. S. Wingreen, Phys. Rev. Lett. 68, 2512 (1992).
[121] L. Y. Chen and C. S. Ting, Phys. Rev. Lett. 64, 3159 (1990).
[122] E. V. Anda and F. Flores, J. Phys.: Condens. Matter 3, 9087 (1991).
[123] J. Zang and J. L. Birman, Phys. Rev. B 46, 5020 (1992).
[124] R. K. Lake and S. Datta, Superlattices Microstruct. 11, 83 (1992).
[125] R. Lake and S. Datta, Phys. Rev. B 45, 6670 (1992).
[126] C. H. Grein, E. Runge, and H. Ehrenreich, Phys. Rev. B 47, 12590 (1993).
[127] Y. L. M. J. McLennan and S. Datta, Phys. Rev. B 43, 13846 (1991).
[128] C. Rivas, R. Lake, G. Klimeck, W. R. Frensley, M. V. Fischetti, P. E. Thompson, S. L.
Rommel, and P. R. Berger, Appl. Phys. Lett. 78, 814 (2001).
[129] C. Rivas, R. Lake, W. R. Frensley, G. Klimeck, P. E. Thompson, S. L. Rommel, and P. R.
Berger, J. Appl. Phys. 94, 5005 (2003).
[130] H. Mehrez, J. Taylor, H. Guo, J. Wang, and C. Roland, Phys. Rev. Lett. 84, 2682 (2000).
[131] A. Maiti, A. Svizhenko, and M. P. Anantram, Phys. Rev. Lett. 88, 126805 (2002).
[132] C. C. Kaun, B. Larade, H. Mehrez, J. Taylor, and H. Guo, Phys. Rev. B 65, 205416
(2002).
[133] T. S. Xia, L. F. Register, and S. K. Banerjee, J. Appl. Phys. 95, 1597 (2004).
[134] E. Louis, J. A. Vergs, J. J. Palacios, A. J.Prez-Jimnez, and E. SanFabin, Phys. Rev.
B 67, 155321 (2003).
[135] W. Tian, S. Datta, S. Hong, R. Reifenberger, J. I. Henderson, and C. P. Kubiak, J. Chem.
Phys. 109, 2874 (1998).
[136] P. A. Derosa and J. M. Seminario, J. Phys. Chem. B 105, 471 (2001).
[137] Y. Xue, S. Datta, and M. A. Ratner, J. Chem. Phys. 115, 4292 (2001).
[138] P. S. Damle, A. W. Ghosh, and S. Datta, Phys. Rev. B 64, 201403 (2001).
123
[139] J. M. Seminario, A. G. Zacarias, and P. A. Derosa, J. Phys. Chem. A 105, 791 (2001).
[140] J. Heurich, J. C. Cuevas, W. Wenzel, and G. Schn, Phys. Rev. Lett. 88, 256803 (2002).
[141] Y. Xue and M. A. Ratner, Phys. Rev. B 68, 115406 (2003).
[142] M. Galperin, A. Nitzan, S. Sek, and M. Majda, J. Electroanal. Chem. 550-551, 337
(2003).
[143] W. I. Babiaczyk and B. R. Buka, J. Phys.: Condens. Matter 16, 4001 (2004).
[144] S. Krompiewski, J. Phys.: Condens. Matter 16, 2981 (2004).
[145] G. Kim, H. Suh, and E. Lee, Phys. Rev. B 52, 2632 (1995).
[146] R. Lake, G. Klimeck, R. C. Bowen, and D. Jovanovic, J. Appl. Phys. 81, 7845 (1997).
[147] G. Klimeck, R. Lake, and D. K. Blanks, Phys. Rev. B 58, 7279 (1998).
[148] H. G. J. Taylor and J. Wang, Phys. Rev. B 63, 245407 (2001).
[149] Y. Xue, S. Datta, and M. A. Ratner, Chem. Phys. 281, 151 (2002).
[150] S.-C. Lee and A. Wacker, Phys. Rev. B 66, 245314 (2002).
[151] M. Galperin, M. A. Ratner, and A. Nitzan, J. Chem. Phys. 121, 11965 (2004).
[152] B. D. Fainberg, P. Hanggi, S. Kohler, and A. Nitzan, AIP Conference Proceedings 1147,
78 (2009).
[153] R. Lake and S. Datta, Phys. Rev. B 46, 4757 (1992).
[154] R. Lake, G. Klimeck, R. Bowen, C. Fernando, T. Moise, Y. Kao, and M. Leng, Superlattices and Microstructures 20, 279 (1996).
[155] A. Groshev, T. Ivanov, and V. Valtchinov, Phys. Rev. Lett. 66, 1082 (1991).
[156] H. L. Chen, Y. H. Pan, S. Groh, T. E. Hagan, and D. P. Ridge, JACS 113, 2766 (1991).
[157] J. D. S. Hershfield and J. Wilkins, Phys. Rev. Lett. 67, 3720 (1991).
[158] Y. Meir, N. S. Wingreen, and P. A. Lee, Phys. Rev. Lett. 70, 2601 (1993).
[159] L. Y. Chen and C. S. Ting, Phys. Rev. B 43, 4534 (1991).
[160] S. Hershfield, Phys. Rev. B 46, 7061 (1992).
[161] A. L. Yeyati, F. Flores, and E. V. Anda, Phys. Rev. B 47, 10543 (1993).
124
[162] S. Datta and M. P. Anantram, Phys. Rev. B 45, 13761 (1992).

[163] N. S. Wingreen, A.-P. Jauho, and Y. Meir, Phys. Rev. B 48, 8487 (1993).
[164] C. A. Stafford and N. S. Wingreen, Phys. Rev. Lett. 76, 916 (1996).
[165] P. Kral and A.-P. Jauho, Phys. Rev. B 59, 7656 (1999).
[166] T. J. Levy and E. Rabani, J. Phys.: Condens. Matter 25, 115302 (2013).
[167] T. J. Levy and E. Rabani, J. Chem. Phys. 138, 164125 (2013).
[168] G. Wick, Phys. Rev. 80, 268 (1950).
[169] A. L. Fetter and J. D. Walecka, Quantum Theory of Many-particle Systems (McGraw-Hill,
New York, 1971).
[170] R. D. Mattuck, A guide to Feynman diagrams in the many-body problem (McGraw-Hill
International Book Company, 1976).
[171] R. A. Craig, J. Math. Phys. 9, 605 (1968).
[172] P. Danielewicz, Ann. Phys. 152, 239 (1984).
[173] A. G. Hall, J. Phys. A: Math. Gen. 8, 214 (1975).
[174] A. M. Zagoskin, Quantum theory of many-body systems: techniques and applications (New
York : Springer, 1998).
[175] S. G. Jakobs, M. Pletyukhov, and H. Schoeller, J. Phys. A: Math. Theor. 43, 103001
(2010).
[176] H. Nyquist, Phys. Rev. 32, 110 (1928).
[177] D. N. Zubarev, Sov. Phys. Usp. 3, 320 (1960).
blikov,
`
[178] V. L. Bonch-Bruevich and S. V. Tia
The Green function method in statistical
mechanics (Amsterdam: North-Holland Pub. Co., 1962).
[179] C. Lacroix, J. Phys. F 11, 2389 (1981).
[180] C. Niu, D. L. Lin, and T. H. Lin, J. Phys. C 11, 1511 (1999).
[181] F. Scheck, in Quantum Physics (Springer Berlin Heidelberg, 2007).
[182] P. Pals and A. MacKinnon, J. Phys.: Condens. Matter 8, 5401 (1996).
125
[183] B. Song, D. A. Ryndyk, and G. Cuniberti, Phys. Rev. B 76, 045408 (2007).
[184] G. Czycholl, Phys. Rev. B 31, 2867 (1985).
[185] K. Kang and B. I. Min, Phys. Rev. B 52, 10689 (1995).
[186] M. Galperin, A. Nitzan, and M. A. Ratner, Phys. Rev. B 76, 035301 (2007).
[187] J. N. Pedersen, D. Bohr, A. Wacker, T. Novotn, P. Schmitteckert, and K. Flensberg,
Phys. Rev. B 79, 125403 (2009).
[188] M. Marques, A. Rubio, C. A. Ullrich, K. Burke, F. Nogueira, and E. K. U. Gross, Timedependent density functional theory, Lecture notes in physics, 706 (Springer, Berlin, 2006).
[189] N. M. Plakida, Theor. Math. Phys. 5, 1047 (1970).
[190] D. Ihle and B. Lorenz, Phys. Status Solidi B 60, 319 (1973).
[191] S. M. Cronenwett, T. H. Oosterkamp, and L. P. Kouwenhoven, Science 281, 540 (1998).
[192] R. wirkowicz, J. Barna, and M. Wilczyski, Phys. Rev. B 68, 195318 (2003).
[193] B. R. Buka and T. Kostyrko, Phys. Rev. B 70, 205333 (2004).
[194] T. W. Kelley, P. F. Baude, C. Gerlach, D. E. Ender, D. Muyres, M. A. Haase, D. E.
Vogel, and S. D. Theiss, Chem. Mat. 16, 4413 (2004).
[195] I. Gur, N. A. Fromer, M. L. Geier, and A. P. Alivisatos, Science 310, 462 (2005).
[196] C. Vermeulen, W. Barford, and E. R. Gagliano, Europhys. Lett. 31, 225 (1995).
[197] Y. Asano, Phys. Rev. B 58, 1414 (1998).
[198] W. Liang, M. P. Shores, M. Bockrath, J. R. Long, and H. Park, Nature 417, 725 (2002).
[199] A. Baratz and R. Baer, J. Phys. Chem. Lett. 3, 498 (2012).
[200] A. Baratz, M. Galperin, and R. Baer, J. Phys. Chem. C 117, 10257 (2013).
[201] L. P. Rokhinson, L. J. Guo, S. Y. Chou, and D. C. Tsui, Appl. Phys. Lett. 76, 1591
(2000).
[202] I. Oppenheim, K. E. Shuler, and G. H. Weiss, Stochastic processes in chemical physics:
the master equation (MIT Press, Cambridge, Mass., 1977).
[203] G. Chen, G. Klimeck, S. Datta, G. Chen, and W. A. Goddard, Phys. Rev. B 50, 8035
(1994).
126
[204] S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, 1995).
[205] C. W. J. Beenakker, Phys. Rev. B 44, 1646 (1991).
[206] S. Garashchuk and D. Tannor, Chem. Phys. Lett. 262, 477 (1996).
[207] S. Garashchuk and D. J. Tannor, Phys. Chem. Chem. Phys. 1, 1081 (1999).
[208] Y. Goldfarb, I. Degani, and D. J. Tannor, Chem. Phys. 338, 106 (2007).
[209] E. M. Heatwole and O. V. Prezhdo, J. Chem. Phys. 130, 244111 (2009).
[210] S. Garashchuk, V. Rassolov, and O. Prezhdo, Semiclassical Bohmian Dynamics, pages
287368 (John Wiley & Sons, Inc., 2010).
[211] A. Shimshovitz and D. J. Tannor, Phys. Rev. Lett. 109, 070402 (2012).
[212] Y. Zhou, E. Pollak, and S. Miret-Arts, J. Chem. Phys. 140, 024709 (2014).
[213] N. Bohr, Zeitschrift fr Physik 2, 423 (1920).
[214] V. Guallar, V. S. Batista, and W. H. Miller, J. Chem. Phys. 113, 9510 (2000).
[215] A. Malevanets and R. Kapral, J. Chem. Phys. 112, 7260 (2000).
[216] V. S. Batista and P. Brumer, J. Phys. Chem. A 105, 2591 (2001).
[217] Y. Wu and V. S. Batista, J. Phys. Chem. B 106, 8271 (2002).
[218] C. Echeverria and R. Kapral, J. Chem. Phys. 132, 104902 (2010).
[219] H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, A. DiNola, and J. R. Haak,
J. Chem. Phys. 81, 3684 (1984).
[220] R. Car and M. Parrinello, Phys. Rev. Lett. 55, 2471 (1985).
[221] G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993).
[222] G. Kresse and J. Hafner, Phys. Rev. B 49, 14251 (1994).
[223] S. Garashchuk, F. Grossmann, and D. Tannor, J. Chem. Soc., Faraday Trans. 93, 781
(1997).
[224] I. Frank, J. Hutter, D. Marx, and M. Parrinello, J. Chem. Phys. 108, 4060 (1998).
[225] M. Sprik and G. Ciccotti, J. Chem. Phys. 109, 7737 (1998).
127
[226] Y. Sugita and Y. Okamoto, Chem. Phys. Lett. 314, 141 (1999).
[227] D. Marx and J. Hutter, Ab Initio Molecular Dynamics: Theory and Implementation,
volume 3 of NIC Series, (John von Neumann Institute for Computing, Julich, 2000).
[228] J.-L. Liao and G. A. Voth, J. Phys. Chem. B 106, 8449 (2002).
[229] S. Zhang and E. Pollak, J. Chem. Phys. 121, 3384 (2004).
[230] H.-S. Lee and M. E. Tuckerman, J. Chem. Phys. 126, 164501 (2007).
[231] M. Kaledin, A. L. Kaledin, J. M. Bowman, J. Ding, and K. D. Jordan, J. Phys. Chem.
A 113, 7671 (2009).
[232] M. Am-Shallem, Y. Zeiri, S. V. Zybin, and R. Kosloff, Phys. Rev. E 84, 061122 (2011).
[233] K. Lindorff-Larsen, N. Trbovic, P. Maragakis, S. Piana, and D. E. Shaw, JACS 134, 3787
(2012).
[234] F. Webster, P. Rossky, and R. Friesner, Comput. Phys. Commun. 63, 494 (1991).
[235] D. F. Coker and L. Xiao, J. Chem. Phys. 102, 496 (1995).
[236] S. Hammes-Schiffer, J. Phys. Chem. A 102, 10443 (1998).
[237] S. Y. Kim and S. Hammes-Schiffer, J. Chem. Phys. 124, 244102 (2006).
[238] Q. Wang and S. Hammes-Schiffer, J. Chem. Phys. 125, 184102 (2006).
[239] S. B. A. Nassimi and R. Kapral, J. Chem. Phys. 133, 134115 (2010).
[240] J. C. Tully, J. Chem. Phys. 93, 1061 (1990).
[241] G. D. Billing, J. Chem. Phys. 99, 5849 (1993).
[242] M. F. Herman, Annu. Rev. Phys. Chem. 45, 83 (1994).
[243] G. Stock and M. Thoss, Adv. Chem. Phys. 131, 243 (2005).
[244] G. C. D. MacKernan and R. Kapral, J. Phys. Chem. B 112, 424 (2008).
[245] A. Kelly and R. Kapral, J. Chem. Phys. 133, 084502 (2010).
[246] W. H. Miller, J. Chem. Phys. 53, 1949 (1970).
[248] W. H. Miller, Chem. Phys. Lett. 7, 431 (1970).
128

[250] W. H. Miller, Adv. Chem. Phys. 30, 77 (1975).
[251] H. Wang, X. Sun, and W. H. Miller, J. Chem. Phys. 108, 9726 (1998).
[252] N. Makri and K. Thompson, Chem. Phys. Lett. 291, 101 (1998).
[253] H. Wang, X. Song, D. Chandler, and W. H. Miller, J. Chem. Phys. 110, 4828 (1999).
[254] K. Thompson and N. Makri, J. Chem. Phys. 110, 1343 (1999).
[255] E. Rabani, S. A. Egorov, and B. J. Berne, J. Phys. Chem. A 103, 9539 (1999).
[256] H. Wang, M. Thoss, and W. H. Miller, J. Chem. Phys. 112, 47 (2000).
[257] W. H. Miller, J. Phys. Chem. A 105, 2942 (2001).
[258] A. Nakayama and N. Makri, J. Chem. Phys. 119, 8592 (2003).
[259] A. Nakayama and N. Makri, Proc. Natl. Acad. Sci. U.S.A. 102, 4230 (2005).
[261] R. Kapral, Annu. Rev. Phys. Chem. 57, 129 (2006).
[262] E. Martin-Fierro and E. Pollak, J. Chem. Phys. 126, 164108 (2007).
[263] E. Pollak, J. Chem. Phys. 127, 074505 (2007).
[264] J. M. Moix and E. Pollak, J. Chem. Phys. 129, 064515 (2008).
[265] B. Li, E. Wilner, M. Thoss, E. Rabani, and W. Miller, J. Chem. Phys. 140, 104110
(2014).
[266] W. H. Miller and C. W. McCurdy, J. Chem. Phys. 69, 5163 (1978).
[267] C. W. McCurdy, H. D. Meyer, and W. H. Miller, J. Chem. Phys. 70, 3177 (1979).
[268] H.-D. Meyer and W. H. Miller, J. Chem. Phys. 71, 2156 (1979).
[270] W. H. Miller and K. White, J. Chem. Phys. 84, 5059 (1986).
[271] S. Datta, J. Phys. C 2, 8023 (1990).
129
[272] D. Holm, Geometric Mechanics: Dynamics and symmetry, Geometric Mechanics (Imperial College Press, 2008).
[273] L. Bonnet and J. Rayez, Chem. Phys. Lett. 277, 183 (1997).
[275] G. Stock and M. Thoss, Phys. Rev. Lett. 78, 578 (1997).
[276] T. Helgaker, P. Jrgensen, and J. Olsen, Molecular electronic-structure theory (Wiley,
Chichester ; New York, 2000).
[277] S. G. Ovchinnikov and V. V. Valkov, Hubbard Operators in the Theory of Strongly Correlated Electrons (London : Imperial College Press, 2004).
[278] S. E. Barnes, J. Phys. F: Met. Phys. 6, 1375 (1976).
[279] L. Gehlhoff, Phys. Status Solidi B 197, 421 (1996).
[280] S. J. Cotton and W. H. Miller, J. Chem. Phys. 139, 234112 (2013).
130
, 2014
,
.
?

.
"" , , , .
". "
- ,,
"" , . ,
" " ,
. ,
.
, ,
.
, , , ,;

.
" "
) (.
. ,
,
. ) :(
. )( -

. "
,
, ) ,
. ( ,
.
.
, , .
,
. ,
. -
,
. )
( . ,
,
. , ""
, - ,
.
.- "-" ,
,
""
- . , ,
"" , .

, -
-

:
-
2014
'"

Developing Methods For Transport Calculations in Heterojunctions With Coulomb Interactionsfinal Full (Heb)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Developing Methods For Transport Calculations in Heterojunctions With Coulomb Interactionsfinal Full (Heb)

Uploaded by

Copyright:

Available Formats

Tel Aviv University

Developing methods for transport

Thesis submitted for the degree of

Submitted to the Senate of Tel-Aviv University

This work was carried out under the supervision of

Professor Rabani Eran

Dedications and Acknowledgements

2 Equations-of-Motion technique for the Nonequilibrium Green functions in

3 Semiclassical approaches to quantum transport

4 Summary and Outlook

A Full derivation of the broken symmetry in the Anderson model

D Complimentary information for the Hubbard mapping

Understanding the coupling of individual molecules to macroscopic electrodes under

In general, many-body, interacting, out-of-equilibrium problems cannot be solved exactly

The model Hamiltonians

Up till 1970, in most treatments of electron tunneling through a metal-insulator-metal (MIM)

bath describes the macroscopic leads (left and right electrodes), H

,kL/R k ck ck L/R NL/R

,kL/R k ck ck L/R NL/R

The single resonant level model

The Anderson model

The interaction Hamiltonian takes the form:

The Double Anderson model

Thesis outline and goal

1. The equations-of-motion technique for the nonequilibrium Green function.

The nonequilibrium Green functions formalism is introduced in chapter 2 where we elaborate on

Nonequilibrium Green Functions

0 is the single-particle, noninteracting part, V contains all the interactions (electronwhere H

Before introducing the single-particle, nonequilibrium Green function, which is a correlation

equation (2.2.3) can be rewritten as:

(2 ) A (1 ). The minus sign

function is given by:

left. Putting it all together one gets:

which can be reformulated in the interaction representation:

G (r2 , t2 , r1 , t1 ) = + h H (r1 , t1 ) H (r2 , t2 )

the advanced Green function GA :

ex (t) = 0. Combined with a self energy [80, 169,

of perturbative diagrammatic techniques can be found in Refs [169, 118, 117].

d1d2 G0 (r, , 2) (2, 1) G 1, r0 , 0 ,

d1d2 G (r, , 2) (2, 1) G0 1, r0 , 0 ,

where 1 (r1 , 1 ) and d1 dr1 C d1 .

d1d2 G0 (r, , 2) (2, 1) G 1, r0 , 0 ,

Analytic continuation and Langreth rules

where d1 dr1 dt1 .

AR (t2 , t) B < (t, t0 ) C A (t0 , t1 ) dt0 dt

= A< (t, t0 ) B R (t, t0 ) + AR (t, t0 ) B < (t, t0 )

= A< (t, t0 ) B < (t, t0 )

A< (t, t0 ) B A (t, t0 ) + AR (t, t0 ) B < (t, t0 )

A< (t, t0 ) B > (t, t0 )

improved systematically. Furthermore, it is often difficult to assess, deductively, the accuracy

The equations-of-motion method

i (r2 ) j (r1 ) Gij (2 , 1 ) ,

where C (2 1 ) is the contour step function, i.e.,

2 > 1 on the contour

The EOM [180] for Gij (2 , 1 ) can be written as:

0 is the time-independent, noninteracting, quadratic Hamiltonian, the term H

The term TC hi V (2 ) , ai (2 ) aj (1 ) results in higher order Green functions (depending

and equation (2.2.52) now takes the form:

The single resonant level model - EOM treatment

Define the systems contour ordered Green function:

The equation of motion of G (, 0 ) is:

where Fk (, 0 ) = hi TC ck ( ) d ( 0 ) , is a new Green function generated in the procedure.

The last equation can be rearranged: