Formulation of Tight-Binding Based GW Algorithm for Graphene

Ahmad Syahroni 1, Muhammad Aziz Majidi 1, Andrivo Rusydi 2

1
Departemen Fisika, Fakultas Matematika dan Ilmu Pengetahuan Alam,Universitas Indonesia,
Depok 16424, Indonesia
2
Singapore Synchrotron Light Source, National University of Singapore, Singapore 117603,
Singapore

Email: hmdsyahroni@gmail.com

ABSTRACT

The full many-body problem in condensed-matter physics, in which electron-ion as well as

electron-electron (e-e) interactions play crucial roles, is very tough to solve exactly. To explore the
details of the band structure of the material, a first-principles approach such as Density Functional
Theory (DFT) has become a popular choice. However, a part from its well-established
formulation, it remains a big challenge to use such an approach to capture effects arising from
strong correlations among the electrons correctly. Nowadays, an approach to combine DFT with a
Feynman diagrammatic approach, so called the GW method, to address the effects of e-e
interactions, has been introduced. Despite some successes of the DFT-based GW approach, there
is an issue that this approach does not seem flexible enough to use for solving problems with other
types of interactions, such as magnetic interactions. In this skripsi, we aim to introduce an
algorithm of the implementation of GW method in the frame of tight-binding approximation. We
rigorously derive each step in the algorithm with the aid of Feynman diagrams, and the concept of
analytic continuation to express the physical quantities of interest in real frequency. For a
particular purpose, we are interested to apply this algorithm to graphene in hope of using this
method address optical properties of graphene systems with various kinds of additional
interactions in the near future.

Keywords: GW approximation, tight-binding, graphene, Green's function, Feynman diagrams.

ABSTRAK
Permasalahan banyak benda (many-body) secara lengkap, dimana melibatkan interaksi elektron-
ion dan interaksi elektron-elektron, merupakan permasalahan yang sulit untuk dipecahkan secara
eksak. Pendekatan first-principles seperti Density Functional Theory (DFT) telah menjadi pilihan
yang populer untuk mengamati band structure secara lengkap pada suatu material. Bagaimanapun
juga, terlepas dari perumusannya yang telah mapan, hal itu tetap menjadi tantangan besar untuk
menggunakan pendekatan DFT untuk mengamati efek yang yang disebabkan oleh korelasi yang
kuat antara elektron secara benar. Saat ini telah diperkenalkan pendekatan yang menggabungkan

Universitas Indonesia
DFT dengan pendekatan diagram Feynman, yang disebut metode GW, untuk mengoreksi efek dari
interaksi antara elektron. Terlepas dari beberapa keberhasilan dari pendekatan GW berbasis DFT
ini, pendekatan ini memiliki kekurangan yaitu tidak cukup fleksibel untuk digunakan untuk
memecahkan masalah dengan interaksi yang lain, seperti interaksi magnetik. Pada skripsi ini, kami
memperkenalkan algoritma metode GW dalam kerangka tight-binding. Kami turunkan setiap
langkah pada algoritma secara lengkap dengan menggunakan diagram Feynman dan konsep
analytic continuation untuk mengekspresikan besaran-besaran fisika pada real frequency. Untuk
tujuan tertentu, kami tertarik untuk menerapkan algoritma ini pada sistem graphene dengan
harapan menggunakan metode ini untuk sifat optik sistem graphene dengan berbagai jenis
interaksi tambahan dalam waktu mendatang.

Kata kunci: pendekatan GW, tight-binding, graphene, Green's function, diagram Feynman.

I. INTRODUCTION

Graphene is a sheet (one atom thick) of carbon atoms bonded together in

honeycomb lattice. Such sheets have long been known to exist in different forms.
They are graphite (many graphene sheets stacked on top of one another), carbon
nanotube (a sheet of graphene rolled into a cylinder), and fullerene (a graphene
sheet sewn together to form a spherical surface). For this reason, graphene is
commonly referred to as the building block of carbon allotropes.
In 1930's Peierls and Landau showed that thermal fluctuations would
destroy long range order and essentially melt 2D lattice at any finite temperature.
Therefore, until 2004, it was believed that an extended graphene sheet would not
be stable against such effect. Moreover, even if they were stable, it would be
impossible to isolate them. Therefore it came as a surprise in that year, when
Andre Geim and his colleague at the University of Manchester demonstrated that
both these beliefs were false. They created graphene sheets by peeling them off a
graphite substrate using scotch tape, and characterized them as indeed single sheet
by simple optical microscopy on top of a SiO2 substrate.
In 1946, long before its existence, P.R. Wallace developed the tight
binding band structure theory of graphene and demonstrated its unusual
properties. Utilising the theoretical work of Wallace, graphene has been studied
over the last few years. The interests in this new material are mostly due to its
various unique properties. One of the unique properties which make graphene so
fascinating to study is that its electronic dispersion near the vertices of the
hexagonal Brillouin zone makes the conducting electrons in graphene behave like
massless Dirac fermion. For this particular property, the dynamics of this
conducting electrons are described by the Dirac equation rather than the usual
Schrodinger's equation.
As I mentioned above, carbon atoms in graphene assembled in a
honeycomb lattice due to the sp2 hybridization which form three -bonds in plane
with the three neighbouring carbon atoms. We all know that carbon has four
valence electrons. The sp2 orbitals is occupied by three of them, while the fourth
electron occupies the remaining pz orbital which is perpendicular to the graphene
plane. This electron in the pz orbital is responsible for electronic and optical
properties of graphene.

Universitas Indonesia
 The electronic properties of graphene, and other materials, is an
interesting field in condensed matter physics. The main purpose is to model the
electrons within the material so that we gain an understanding physical properties
of it. Here we are faced with a problem of many electrons, and these electrons
experience interactions. Clearly, we deal with the so-called many-body problem.
The problem lies in the difficulty of finding the exact solution of the system. The
aprroach to solve the problem comes in many techniques or methods, which is
determined by how we treat an electron in the system, as an independent particle,
or as one that only interacts with the average of the others (Hartree-Fock), or as a
non-interacting particle with an interacting density (Density Functional Theory).
DFT (the Hohenberg-Kohn-Sham density-functional theory) offers a
method for calculating the ground-state properties of a system of interacting
electrons. But it fails in some degrees for calculating the excited-state properties
of the system. A successful approximation for the determination of excited states
is based on Many-Body Perturbation Theory (MBPT), which is exact when the
expansions of the interactions are summed to infinity. In practice, we cannot sum
the interactions to infinity, hence we use some kind of approximations.
The idea behind MBPT is that the long-ranged Coulomb forces could
screen individual electrons with a surrounding cloud of the other electrons. This
requires us to define an electron plus its screening cloud as the quasi-particle.
Thus the system of strongly interacting real particles can be viewed as weakly
interacting (indeed non-interacting) quasi-particles.
The mathematical description of quasi-particles is based on single-
particle Green's function. The exact determination of Green's function, or full
Green's function as we will discuss further, requires the knowledge of the quasi-
particle self-energy. The self-energy is usually approximated using a perturbative
expansion with respect to the quasi-particle interaction.
In 1965 Lars Hedin described a way to calculate self-energy, in which
the form of the self-energy gives the approximation the name GW. In this
approximation, the self-energy is the product of the single-particle Green function
G and the screened interaction W. It can be viewed as Hartree-Fock with a
dynamically screened Coulomb interaction. In the 1980s GW was shown to be a
superior method to Density Functional Theory (DFT) for the calculation of band-
structures. However, this approach is not be able to apply to large-scale due to the
complexity of DFT itself.
DFT are computationally cheap methods since it uses static mean-fields.
We expect GW to give better results than DFT, as it contains more physics and it
is more computationally expensive. However, there have been no formal studies
to date with GW approximation for the actual Graphene. What we meant by actual
Graphene is the fact that there are two sublattices in its unit cell. In this project,
we take care of these sublattices indices. Some authors neglect the Hartree term in
which they thought it is cancelled by the positive background.

Many Body Perturbation Theory

Universitas Indonesia
 The purpose of many-body theory is of course not to find the full
solution of the many-body Hamiltonian. The many-body Hamiltonian is just too
difficult to solve. Because of its complexity, not much progress made with it until
1956 when quantum field theory method provided powefull way of attacking
many-body problem. It provided a different view of matter, in which systems of
interacting real particles are described in term of approximately non-interacting
fictitious bodies called quasi particles.
The next question is, how to calculate the properties of this fictitious
bodies?
In the regular quantum mechanics, we deal with a wavefunction that is
considered to carry the full information of the system. When the Hamiltonian is
easy to solve, then we may be able to obtain the wavefunction exactly. In the case
where the Hamiltonian is more complicated, we may consider the Hamiltonian as
being decomposed into the bare (or non-interacting) and the interaction parts. The
bare Hamiltonian gives a known wavefunction that may then be used to find the
approximate true wavefunction.
In many-body or quantum-field theory, such a role of wavefunction is
played by a quantity called Green's function (GF). Thus, like with wavefunction in
regular quantum mechanics, in many-body theory we also have the bare and the
"true" or "full" Green's functions.
The idea behind the Green's function method is that in order to find the
important physical properties of a system, it is not necessary to know the detailed
behaviour of each particle in the system, but rather the average of one typical
particle. The quantity that describe this average behaviour is known as one-
particle Green's function.
Because of this average behaviour instead of the detailed behaviour, the
information contained in a Green's function are much less than the knowledge
contained in the full wavefunction. But it still provides the quantities one usually
needs such as the expectation value and energy of any single-particle operator in
the ground state of the system, the single-particle excitation spectrum of the
system, etc.

II. RESEARCH METHOD

Tight-Binding Approximation

Theoretical modelings of graphene have commonly been done within

the tight-binding (TB) approximation, for which it provides a simple looking
Hamiltonian, where all the physics of the system are captured by phase factors
and overlap integrals. Yet it is believed to represent the correct low-energy physics
of the material.
The Hamiltonian for electrons in graphene in real-space representation
within the TB approximation leading to

Universitas Indonesia
 By performing discrete Fourier transformation of the creation and the
annihilation operators in the above Hamiltonian on the 'honeycomb' lattice of
grapheme

we obtain the correspoding Hamiltonian in k-space as

with

Electron Interaction

To explore the high-energy physics of graphene, such as the optical

properties, the Tight-Binding (TB) Hamiltonian needs be added with the electron-
electron (e-e) interaction term. The role of e-e interaction would then often be
treated within the GW approximation. Many such theoretical studies assume the
expression of e-e interaction potential in k-space has the form

Such an expression is known, and commonly used for electrons in

vacuum, or in a metal, for which the use of a plane wave basis set is adequate,
with the second quantized-form of the interaction term looking like

Universitas Indonesia
 As in many articles of such theoretical studies on graphene, the explicit
second-quantized form of the e-e interaction term in k-space is normally not
displayed. It raises an issue as to whether such a simple Fourier transform
expression leading to equations (5) and (6) is consistent within the TB
approximation for graphene, in which one would naturally expect the appearance
of sublattice indices. To resolve this issue, here we aim at making explicit
derivation of the second-quantized form of the e-e interaction term in k-space
within the TB approximation for a system with sublattices. The resulting
expression will further be used to interpret Feynman diagrams of the self-energy
and the dressed interaction line within the GW approximation.

Problem with ontinuous translational invariance representation

Now, we want to add the e-e interaction term within the TB

approximation. For this purpose, one may attempt to go back to the original
expression of the two-body Coulomb interaction between two electrons expressed
in terms of field operators, that is

In order to express Hint in k-space within the TB approximation, we may

express the field operators in terms of the above creation and annihilation
operators, namely

Here, represents the Bloch function formed within the TB

approximation that reads

With being the atomic wave function (i.e. for graphene it

is the pz orbital wave function) centered at the carbon atom of sublattice located
at . Here is the position vector of a reference point at the unit cell

Universitas Indonesia
n, and is the vector taken from that reference point to the center of carbon
atom of sublattice .

Substituting equations (8) and (9) into equation (7) gives us

with the k-dependent potential being defined as

Now, can we simplify the expressions of equations (10) and (11)? Before we try to
answer this for our actual problem of graphene, let us evaluate what happens if we
set each on the right side of equation (11) to 1, which is equivalent to
replacing Bloch-wave functions with plane-wave functions. This action simplifies
equation (11) to be

Since , then substituting

equation (3.9) to equation (3.6) and performing the summation over implies
replacing everywhere it appears with , such that now we have

Universitas Indonesia
as in equation (3.1).

Now, what is wrong with equations (3.11) and (3.12) we just obtained?
Well, we have set to 1 in order to simplify the calculation of the right
hand side of equation (3.7) to finally come down to equations (3.11) and (3.12).
This simplification has no ground, and thus it is incorrect. But then, how should
we proceed the calculation further from equation (3.7) without setting
to 1? The answer is, we can't, and it is not correct to do so. Because, our starting
point by introducing in terms of field operators as functions of continuous
variables in equation (3.3) is already inconsistent with the fact that we
have defned our problem within the TB approximation, where the real-space
degrees of freedoms of the electrons have been limited to discrete lattice sites.
Therefore, to do the formulation correctly, we have to start over by replacing the
expression of equation (3.3) by a similar expression defined on discrete lattice
sites, as discussed in the next section.

Lattice representation of the electron-electron interaction

Consider the electron-electron interaction between two electrons sitting

on lattice sites

Universitas Indonesia
Since our system is 'discretely' translationally invariant, now we have the
justification to do Fourier transformation on operators as done in
equation (2.9). Then, substituting equations (2.9) into equation (3.13) gives us

Universitas Indonesia
Further Calculation of

The right hand side of equation (3.19) can be further evaluated by

specifying the sublattice indices . In this case denotes a vector
connecting the carbon atom of sublattice and that of sublattice in the same
unit cell.
Thus, we have

we have

Universitas Indonesia
Implementation of GW Approximation

GW method is known as a technique used to calculate electronic

structure of a material, by taken into account the e-e interactions, for which the
self energy is approximated within some product of the dressed Green function
(G) and the dressed interaction line (W). This method has so far been well known
as a complementary method used to correct the quasi-particle energies from the
result of an ab initio calculation based on Density Functional Theory (DFT). The
structure of GW self-energy diagram is the same as that used in first-order
perturbation and Hartree-Fock theories. However the level of accuracy among
them is different in the way that, the first order perturbation theory employs only
the bare Green function G0 and the bare interaction line (V), the Hartree-Fock
theory employs the dressed Green function (G), but the bare interaction line (V),
while the GW method employs both Green function and the interaction line in the
dressed form.

Direct and Exchange self energy

Universitas Indonesia
Tadpole and Shell diagrams

Universitas Indonesia
 Third, the electron A having a momentum k interacts again with another
electron (say electron C) having a momentum with possibly
different value of q. These intermediate processes may be considered as occuring
within some time interval manifested in the frequency dependence of the dressed
interaction line. If we replace the dressed interaction line with a bare one that has
no frequency dependence, that means we are only considering some processes
occuring within a virtually zero time interval. The overall process is depicted in
Fig. 3.6. Note, that the upper diagram in Fig. 3.6 is shown to visualize the
intermediate processes more clearly, which is then twisted around to make it look
like a "shell" shape in the lower diagram.

Universitas Indonesia
Dressing the interaction line through RPA

A part of the GW method is the calculation of the dressed interaction

line (W). We do it by dressing the bare interaction line \textit{V} with a
polarization bubble diagram in a diagrammatic equantion as depicted in Fig. 3.7
below. For some historical reason, this procedure is referred to as the Random
Phase Approximation (RPA).
The diagrammatic equation can be translated into the mathematical form as

Universitas Indonesia
Universitas Indonesia
III. RESULT AND DISCUSSION

Dressed interaction line

Theoretical modelings of graphene have commonly been done within

the tight-binding (TB) approximation, for which it provides a simple looking
Hamiltonian, where all the physics of the system are captured by phase factors
and overlap integrals. Yet it is believed to represent the correct low-energy physics
of the material.

Analytic continuation

In order to calculate dynamical properties such as the density of states

or optical conductivity , we need to have equations
expressed in real frequencies. For this purpose, we need to re-express the Green
functions and the interactions lines defined in Matsubara frequency into the
corresponding retarded quantities defined in real frequency. This process is
commonly referred to as analytic continuation. The rule is the following:

Universitas Indonesia
GW Algorithm

1. Guess an initial value of the self-energy matrix , then

perform Dyson equation to compute
2. Compute the density of states.
3. Compute the polarization bubble
4. Compute the dressed interaction line
5. Compute direct self energy
6. Compute exchange self energy
7. Add the direct and the exchange self-energy to obtain the new total self-
energy
8. Check the difference between the new and the old self energies. If the
difference is within the tolerance, that is,

Universitas Indonesia
IV. CONCLUSION

We have constructed an algorithm to implement GW method in the

frame of tigh-binding approximation with the aid of Feyman diagrams. The output
of this algorithm are dynamic quantities such as Density of States and Optical
Conductivity. For a particular purpose we intend for the near future, we apply this
algorithm to graphene. Through our derivation we show the subtleties in obtaining
the proper expressions of the bare and the dressed electron-electron Coulomb
interactions in the reciprocal-lattice space, by taking carefully into account the
existence of two sub lattices in graphene. Furthermore, we propose a strategy to
save some computing time by introducing a systematic interpolation scheme for
the self-energy and the interaction line to span the full Brillouin zone.

V. BIBLIOGRAPHY

R.E.Peierls, Ann. (1935). I.H.Poincare Vol 5.

L.D.Landau, Zh. (1937). Eksp. Teor. Fiz Vol. 7.

Wallace, P.R. (1947). The Band Theory of Graphite. Physical Review Vol. 71,
P.634.

Hedin, L. (1965). Physical Review Vol. 139, P.A796.

Sabashvili, A., Ostlund, S. and Granath, M. (2013). Physical Review B Vol. 88.

Hybertsen, M.S and Louie, S.G. (1985). Physical Review B Vol. 32, P.7005.

Shirley, E.L., X.Zhu, and Louie, S.G. (1997). Physical Review Letter Vol. 56,
P.6648.

Mattuck, R.D. (1992). A Guide to Feynman Diagrams in the Many-Body

Problem, P.3. Dover.

Mattuck, R.D. (1992). A Guide to Feynman Diagrams in the Many-Body Problem,

P.12. Dover.

Fetter, A.L and Walecka, J.D. (2003). Quantum Theory of Many-Particle Systems.
Dover.

Raimes, S. (1972). Many-Electron Theory. North-Holland Publishing Company.

Doniach, S and Sondheimer, E. H. (1998). Green's Functions for Solid State

Physicists. Imperial College Press.

Ibach, H. and Luth, H. (2009). Solid-State Physics: An Introduction to Principles

of Materials Science, 4th edition. Springer.

Universitas Indonesia
Mattuck, R.D. (1992). A Guide to Feynman Diagrams in the Many-Body Problem,
P.154. Dover.

Geim, A.K., and Novoselov, K.S. (2007). The rise of graphene. Nature Materials
Vol. 6.

Castro Neto, A.H., et al. (2009). The Electronic Properties of graphene. Reviews
of Modern Physics vol. 81, 109.

Kittel, C. (2005). Introduction to Solid State Physics, 8th edition. Wiley.

Schweitzer, H and Czycholl, G. (1990). Selfconsistent pertubational approach to

the heavy fermion problem in high dimensions. Physics B Vol. 79, P.380.

Onida, G., Reining, R., and Rubio, A. (2002). Electronic Excitations: Density-
Functional Versus Many-Body Green's-Functin Approaches. Reviews of Modern
Physics Vol. 74.

Noel, J and Oliver, M. Introduction to the Physical Properties of Graphene Ch. 1,

P.3-5.

Mahan, G.D. (2000). Many-particle physics 3ed P.162. Kluwer.

Hybertsen, M. S. and Louie, S. G. (1985) Phys. Rev. B 32, 7005.

Universitas Indonesia