Professional Documents
Culture Documents
Feynman Diagram Techniques in Condensed Matter Physics PDF
Feynman Diagram Techniques in Condensed Matter Physics PDF
Feynman Diagram Techniques in Condensed Matter Physics PDF
A concise introduction to Feynman diagram techniques, this book shows how they
can be applied to the analysis of complex many-particle systems, and offers a
review of the essential elements of quantum mechanics, solid-state physics, and
statistical mechanics.
Alongside a detailed account of the method of second quantization, the book
covers topics such as Green's and correlation functions, diagrammatic techniques,
superconductivity, and contains several case studies. Some background knowledge
in quantum mechanics, solid-state physics, and mathematical methods of physics
is assumed.
Detailed derivations of formulas and in-depth examples and chapter exercises
from various areas of condensed matter physics make this a valuable resource for
both researchers and advanced undergraduate students in condensed-matter theory,
many-body physics, and electrical engineering. Solutions to the exercises are made
available online.
RADI A . JISHI
Cn/ijornin STaTe Ulli versity
U
::
CAMBRIDGE
UNIVERSITY PRESS
CAMBRIDGE UNIVERSITY PR ESS
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, Sao Paulo, Delhi, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/978 11 07025172
© R. A. Ji shi 2013
A catalogue record for this publication is available from the British Libra/)'
2 Single-particle states 18
2.1 Introduction ]8
2.2 Electron gas 19
2.3 Bloch states 21
2.4 Example: one-dimensional lattice 27
2.5 Wannier states 29
2.6 Two-dimensional electron gas in a magnetic field 31
Further reading 33
Problems 34
3 Second quantization 37
3.1 N -pa11icle wave function 37
3.2 Properly symmetrized products as a basis set 38
3.3 Three examples 40
3.4 Creation and annihilation operators 42
3.5 One-body operators 47
3.6 Examp]es 48
3.7 Two-body operators 50
3.8 Translationally invariant system 51
3.9 Example: Coulomb interaction 52
3.10 Electrons in a periodic potential 53
vii
viii Contents
References 390
Index 394
Preface
In both theory and practice, condensed matter physics is concerned with the phys-
ical properties of materials that are comprised of complex many-particle systems.
Modeling the systems ' behavior is essential to achieving a better understanding of
the propelties of these systems and their practical use in technology and industry.
Maximal knowledge about a many-particle system is gained by solving the
Schrodinger equation. However, an exact solution of the Schrodinger equation is
not possible, so resort is made to approximation schemes based on perturbation
theory. It is generally true that, in order to properly describe the properties of
an interacting many-particle system, perturbation theory must be carried out to
infinite order. The best approach we have for doing so involves the use of Green's
function and Feynman diagrams. Furthermore, much of our knowledge about a
given complex system is obtained by measuring its response to an external probe,
such as an electromagnetic field, a beam of electrons, or some other form of
perturbation; its response to this perturbation is best described in terms of Green's
function.
Two years ago, I set out to put together a guide that would allow advanced
undergraduate and beginning graduate students in physics and electrical engineer-
ing to understand how Green's functions and Feynman diagrams are used to more
accurately model complicated interactions in condensed matter physics. As time
went by and the book was taking form , it became clear that it had turned into a
reference manual that would be useful to professionals and educators as well as
students . It is a self-contained place to learn or review how Feynman diagrams are
used to solve problems in condensed matter physics. Great care has been taken to
show how to create them, use them, and solve problems with them, one step at a
time. It has been a labor of love. My reward is the thought that it will help others
to understand the subject.
The book begins with a brief review of quantum mechanics, followed by a short
chapter on single-particle states. Taken together with the accompanying exercises,
xiii
xiv Preface
these two chapters provide a decent review of quantum mechanics and solid state
physics. The method of second quantization , being of crucial importance, is dis-
cussed at length in Chapter 3, and applied to the jellium model in Chapter 4. Since
Green's functions at finite temperature are defined in telms of thermal averages,
a review of the basic elements of statistical mechanics is presented in Chapter 5,
which , I hope, will be accessible to readers without extensive knowledge of the
subject.
Real-time Green's functions are discllssed in Chapter 6, and some applications of
these functions are presented in Chapter 7. Imaginary-time functions and Feynman
diagram techniques are dealt with in Chapters 8 and 9. Every effOlt has been
made to provide a step-by-step derivation of all the formulas , in as mllch detail
as is necessary. Rules for the creation of the diagrams and their translation into
algebraic expressions are clearly delineated. Feynman diagram techniques are then
applied to the interacting electron gas in Chapter 10, to electron-phonon and
electron-photon interactions in Chapter 11, and to superconductivity in Chapter 12.
These techniques are then extended to systems that are not in equilibrium in
Chapter 13.
Many exercises are given at the end of each chapter. For the more difficult
problems, some guidance is given to allow the reader to alTive at the solution.
Solutions to many of the exercises, as well as additional material, will be provided
on my website (www.calstatela.edu/faculty/rjishi).
Over the course of the two years that it took me to finish this book, I received
help in various ways from many people. In particular, I would like to thank David
Guzman for extensive help in preparing this manuscript, and Hamad Alyahyaei for
reading the first five chapters. I am indebted to Linda Alviti, who read the whole
book and made valuable comments. I am grateful to Professor 1. E. Dzyaloshinski
for reading Chapter 9 and for his encouraging words. I also want to thank Dr. John
Fowler, Dr. Simon Capelin , Antoaneta Ouzounova, Fiona Saunders, Kirsten Bot,
and Claire Poole from Cambridge University Press for their help, guidance, and
patience. I would also like to express my gratitude to my wife and children for
their encouragement and support. Permission to use the quote from Russell's The
Scientific Outlook (2001) was provided by Taylor and Francis (Routledge). Copy-
right is owned by Taylor and Francis and The Bertrand Russell Foundation Ltd.
Permission to use Gould 's quote from Ever Since Darwin (1977) was provided by
WW. Norton & Company.
This book is dedicated to the memory of my parents, who, despite adverse
conditions, did all they could to provide me with a decent education.
The main focus of this book is many-particle systems such as electrons in a crystal.
Such systems are studied within the framework of quantum mechanics, with which
the reader is assumed to be familiar. Nevertheless, a brief review of this subject will
provide an opportunity to establish notation and collect results that will be used
later on.
v V*
Figure 1.1 Vector space V of kets and the corresponding dual space V' of bras .
A one-to-one correspondence exists between kets and bras.
Figure 1.2 The probability of finding the particle, at time t , in the cube of volume
c/3 r , centered on r , is 1\jJ(r. t)1 2c/3,..
(II) Observables
An observable is represented by a linear, Hermitian operator acting on the state
space. If A is an operator, being linear means that
0.3)
The operator e -i Hr jFt is called the time evolution operator. Defining the station-
ary states 14>11 ) as the solutions of the eigenvalue equation, known as the time-
independent Schrodinger equation,
0.4)
it is readily verified that 14>II ) e - iE" r j li is a solution ofEq. (1.2); the general solution
of Eq. 0.2), when H is independent of t, is then given by
1\lJ(t)) = I>II j
14>1I)e - iE" r li.
I
In contrast, the evolution of the classical state of a particle is determined by Hamil-
ton's function H via Hamilton's equations of motion which, in one dimension,
are
(IV) Measurements
Let an observable be represented by the linear, Hermitian operator A, and consider
the eigenvalue equation
where a I, a2, ' , , are the eigenvalues, and I<PI), 1<P2), ' .. the cOlTesponding eigen-
vectors, or eigenkets, In general, there may be infinitely many eigenvalues and
eigenkets. If k eigenkets correspond to the same eigenvalue a" a,
then is said to be
k-fold degenerate. The following is postulated:
We note that the eigenvalues of a hermitian operator are real; hence, the out-
come of any measurement of an observable is a real number, as it should be.
Further, for a hermitian operator, the eigenkets cOlTesponding to different eigen-
values are necessarily orthogonal. In the case of a k-fold degeneracy, where k
eigenkets correspond to the same eigenvalue, every ket in the k-dimensional sub-
space that the eigenkets span is an eigenket of A with the same eigenvalue. It
is always possible to choose within this subspace a set of k eigenkets that are
orthogonal to each other. By normalizing the eigenkets, it is always possible to
choose the eigenkets I<PI ), 1<P2) , ... so as to form a complete orthonormal basis that
spans the vector space of state vectors. Orthonormality means that (<P; I<P j) = oij
where
i=l=j
(1.7)
i=j
is the Kronecker delta , occasionally written as 0;. j with a comma inserted between
the indices if its absence could cause confusion . Completeness means that states
I<P I), 1<P2) . ... form a basis set: any state vector Iw(t) ) E V can be expanded as
If the basis is chosen to be an orthonormal one, i.e., if I¢!), 1¢2) .... form an
orthonormal set, then for an arbitrary state I\Ii (t)),
II II
"
= em(t) =} 1\Ii(f)) = L(¢II 1\Ii(f)) I¢II) = L I¢")(¢II 1\Ii(t))
II 11
(1.9)
"
Equation (1.9) expresses mathematically the property of completeness of the
orthonormal states I¢!), 1¢2) , .... Note that I¢,,) (¢"I is an operator: it acts on a
ket to yield another ket, and the 1 on the right hand side (RHS) of Eg. 0.9) is the
identity operator.
An important complete orthonormal set of states is formed by the eigenkets of
the position operator r,
On the left hand side (LHS), r is the position operator, sometimes written as r or
rop to emphasize that it is an operator, while r on the RHS is the eigenvalue of the
position operator. The ket Ir) is the state of a particle with a well defined position
r. Since the operator r is hermitian, the states Ir) form a complete orthonormal set.
Because r is continuous, the orthonom1ality and completeness of the states now
read
0 r # 0
oCr) = 0.12)
1 00 r=O
and
f 3
o(r)d r = 1, (1.13)
the integration being over all space. In one dimension, o(x - x') is represented
graphically as in Figure 1.3.
One important property of o(r - r ' ) is the sifting property,
f 3
J(r)o(r - r ' )d r = J(r' ). (1.14)
6 A brief review of quantum mechanics
5(x - Xl)
Xl X
Figure 1.3 Dirac-delta function 8(x - x'). It is zero for all values of x except for
x = x' where it is infinite. However, its integral over any interval containing x' is
unity.
We also note that v(r - r /) = o(r' - r) and o(ar) = o(r)/ la I", where d is the dimen-
sion of space: d = 3 if r is a three-dimensional vector. A palticularly useful repre-
sentation of the Dirac-delta function is the following:
oCr) = _1_!
(27T )3
e±ik.rd3k. (1. I 5)
IW(t)) = ! 3
Ir)(r IW(t)) d r.
This is the continuous analog of the discrete case for which IW(t) ) =
L I 1<P1l)(<P1l IW(t)).
I (<P1l IW(t)) 12 has been interpreted as the probability for a particle in state I W(t) )
to be found at time t in state 1<P1l)' By analogy, should I(r IW(t))1 2 be interpreted as
the probability for a particle in state IW(t) ) to be found at time t in state Ir), i.e., to
be at position r at time t? Two problems beset this interpretation:
(a) IW(t) ) and 1<P1l) are dimensionless «(W(t) IW(t)) = 1, (<PIlI<PIl) = 1); hence,
I(<P1l IW(t )) 12 is dimensionless and can be interpreted as a probability. However,
1.1 The postulates 7
that the detector clicks should be prop011ional to 1(r 1\lI (t») 12d3r. The constant
of proportionality is determined by requiring that the probability of finding the
particle somewhere in space be unity. Noting that
2 3 3 3
j I(r 1\lI(t»)1 d r =j d r(r 1\lI(t»)* (r 1\lI(t») = j d r(\lI(t)lr)(r 1\lI(f»)
= (\lI(t) 1\lI(t») = 1,
In other words, the wave function \lI(r , t) is the component of state vector
1\lI(t») along Ir).
1= L l<Pn)(<PII = L j j
1
3 3
Ir)(rl<Pn)(<Plllr')(r'ld rd r'
n n
(1.21)
The spin-up state I t) is also denoted by 11 / 2), or 1+), or ct, while the spin-down
state may also be written as 1- 1/ 2), or 1-), or {3 . A general spin state, denoted by
Ix), is a linear combination of the basis states It ) and It ),
Ix) =alt)+blt)
(1.24)
// eJ u
Here, \.! is a collective index that specifies the spatial and spin quantum numbers.
For example, four quantum numbers specify the eigenstates of a hydrogen atom:
the principal quantum number n that determines the energy of the state, I which
determines the value of L 2 (the square of the orbital angular momentum), m which
determines the value of L z (the z-component of the orbital angular momentum),
and a which is either t or ~. In this case \.! = [nlma], while the index n in 14>//)
stands for the spatial quantum numbers [nlm]. The ket l4>u) = 14>11) ® la), being a
direct product of an orbital (spatial) state and a spin state, is called a spin orbital.
The orthonormality and completeness of the states 14>\!) mean that
(1.25)
Here, l spatial (l spin) is the identity operator in V'spatial (Vspin ), and I on the RHS is the
identity operator in the Hilbert space (the direct product space).
So far, we have restricted the discussion to a one-particle system. We now
consider a system comprised of N identical particles. Identical particles, such as
electrons, are truly indistinguishable in quantum mechanics. The stationary states
(eigenfunctions of the Hamiltonian H) of a system of N identical particles will
be written as w(l , 2, ... , N), depending on the spatial and spin coordinates of
the particles. Because of the indistinguishability of the particles, the Hamiltonian
remains unchanged if any two particles are interchanged. This means that H
commutes with Pi) , the permutation operator which interchanges particles i and
j. It follows that the eigenfunctions of H can be chosen to be simultaneously
eigenfunctions of Pij . Denoting the eigenvalues of Pij by).., we can write
Pij w(l , ... , i, ... , j , .. . , N) = ).. w(l , ... , i, ... , j , ... , N).
Applying Pij to both sides of the above equation, and noting that Pi~ = 1, we obtain
Thus, ).. 2 = I :::} ).. = ±l. For).. = + 1( -1), the wave function is symmetric (anti-
symmetric) under the exchange of coordinates (spatial and spin) of any two par-
ticles. In nature, particles with integral spin (0, 1, 2, ... ), known as bosons, have
symmetric wave functions under the exchange of the coordinates of two particles,
and they obey Bose-Einstein statistics. On the other hand, particles with half inte-
gral spin 0 / 2, 3/ 2, . .. ), known as fermions , have antisymmetric wave functions
10 A brief rel'ieIV of quantum mechanics
under the exchange of the coordinates of two particles, and they obey Fermi-Dirac
statistics. The Pauli exclusion principle is a direct consequence of this antisym-
metry of the fermionic wave function. The last postulate of quantum mechanics
follows.
w(l , . .. ,
..
j, ... , t , .... N) =
IW(I .... , i .....
..
j . .... N) Bosons
.
-WO, ... , I , ... ,}, ... , N) FermlOns.
(1.27)
We close this section by remarking that some exotic quasiparticles, known as
anyons, which arise as excitations of a two-dimensional electron gas in a magnetic
field, are believed to obey some fractional statistics, which are neither Bose-
Einstein nor Fermi-Dirac statistics (Wilczek, 1982).
a= ( -mw) 1/2 ( x + - p ,
2fI mw
i) at _
- (In- W)
2fI
1/ 2 (
x--p.
mw
i) (1.29)
Since x and p are hermitian, a t is the adjoint of a, and vice versa. The operators x
and p = -ihd/dx do not commute: xp #- px. We define the commutator of any
two operators A and B by
[A , B] = A B - B A. 0.30)
By letting the commutator [x, p] act on an arbitrary differentiable function f(x), it
is found that [x , p] = in. It follows that [a , at] = I. In terms of the new operators,
The hermitian operator N is called the number operator. Let the eigenvalues of N
be denoted by n and the corresponding eigenkets by In ),
With the help of these commutation relations we can easily show that
alO) = 0 =} mw )
( 2h
1/ 2 (
x
i)
+ mw p 10) = 0,
12 A brief review of quantum mechanics
--4----'H-+--~-_,E = ~nw
---a----/---:p---- E = ~nw
(x+~~)¢o=O
mwdx
I11W) 1/ 4 exp (-m2fIwx 2)
==? ¢o(x) = ( nn . (1.34)
b) 1/2
¢1(X)= ( 2.j'ii
2
(2bx)exp(-b x 2 / 2) (1.35)
The eigenfunctions corresponding to a few of the lowest energy levels are shown in
Figure 104. Note that for n even (odd) the eigenfunctions are even (odd) functions
of x.
In the traditional approach to the harmonic osci Ilator problem, one instead
employs a power series solution to the time-independent Schrodinger equation,
which is a second-differential equation . It is through the requirement that the wave
function vanish at infinity that a truncation of the series is brought about. This,
in tum, leads to the quantization of energy. One might ask where the bound-
ary conditions (the wave function vanishes at x = ±oo) were used in the above
PlVblems 13
discussion. They were used indirectly when it was required that the stationary states
In) be normalized to unity. This can only be true if the corresponding eigenfunctions
vanish at infinity.
Further reading
Griffiths, DJ. (200S). IntlVdu ction to Quantum Mechanics , 2nd edn. Upper Saddle River,
N]: Pearson Education , Inc.
Sakurai, J.1. (1994). Modern Quantum Mechanics, rev. edn. Reading, MA: Addison-Wesley
Publishing Company, Inc.
Problems
l.1 Operators.
(a) Evaluate the commutators [x, p], [x 2 , p], and [p, Vex)].
(b) Show that (AB)t = BtAt.
(c) Show that Tr(ABC) = Tr(CAB), where Tr is the trace.
(d) Show that by setting S = (n / 2)(J , the Pauli spin matrices
(Jx = [0
1 01] ' (Jy = -i]
[0i 0 ' (J~ = [10 -0J ]
. sin2(ax)
hm 2 = 8(x).
a-> oo Jrax
Under these boundary conditions, opposite faces of the cube are identified;
for example, the faces x = 0 and x = L are the same. The ranges of values
of x, y, and z are 0::: x < L , 0::: y < L, 0::: z < L . Find the normalized
eigenfunctions and show by explicit calculation that they form a complete
orthonormal set of states.
14 A brief review of quantum mechan ics
1.S Singlet and triplet states. Consider a two-electron system (for exam-
ple. the two electrons in a helium atom). The total spin S = SI + S2,
where S 1 and S2 are the spin operators of electrons 1 and 2. Con-
sider the singlet state ~ [a(I),8(2) - ,8( I )a(2)] and the three triplet states
a(1)a(2), ~[a(1),8(2) + ,8(1)a(2)], and ,8(1),8(2). Show that these are
eigenstates of S2 and S:.
1.9 Time-independent perturbation. Suppose that for a given system the Hamil-
tonian is H = Ho + V, where Ho and V have no explicit time dependence.
Assume that the solutions of the eigenvalue equation Holn ) = Ell In ) are
known , and that the energy levels are nondegenerate. If V is small, in the
sense that the shift in the energy of the states brought about by the presence
of V is small compared to the energy difference between neighboring states,
i.e., I,0,EII I « IE//±I - Ell I, one can show that
+ L l(mlVln )1 + ...
2
,0,EII = (nlVln)
Ell-Em
111#//
(A )(t ) = (Vr s(t)IA IVrs(t) ) = (Vrs(O)le iHt /nA e-i Htjli IVrs(O) ).
The above suggests a second approach to quantum mechanics, the Heisenberg
picture, in which the state is frozen at what it was at t = 0, but the operator
A evolves in time. In this picture,
where aCt) and a t (t) are the annihilation and creation operators in the
Heisenberg picture.
1.11 The interaction picture. Let the Hamiltonian for a system be given by
H = Ho + vet)
where Ho is time-independent. The state in the Schrbdinger picture is IVrs(t)).
In the interaction picture, the state is defined by
Vet) = 0
(V
t
t:::
<° 0.
(b) Now let t -+ 00 (steady state). The transition rate, Wi -+J , is defined as
Wi--+J =.!!.-
, lim Pi -+J (t). Prove the Fermi golden rule,
C 1 1-+00
2;rr 2
Wi--+J = -IUlVli)1 8(EJ - Ei)'
h
t ::: 0.
(a) Show that the probability of finding the system in state If), also an
eigenstate of Ho with energy E f = Ei + liw fi , is
1 [I - e i(Wj,+w)r 1- ei (Wj;-w) r [2
Pi ~ f = "2 U IA Ii ) + U IA t Ii )
!1 W fi +W W fi - W
2.1 Introduction
Let us consider a system of N identical, interacting particles whose Hamiltonian is
N 1 N
H(l , 2, ... , N) = Lh(i) + 2L v(i, j). (2.1 )
i=1 if-j
h(i) is the sum of the kinetic energy and potential energy, due to some external
field, of particle i,
11 2
h(i) = _ _ \7 2 + v(i) . (2.2)
2m
For example, if the interacting p3I1icles are the electrons in an atom, then v(i) is the
potential energy of electron i due to its interaction with the nucleus. For electrons
in a crystal, v(i) is the interaction of electron i with the ionic lattice. The last term in
Eq. (2.1) represents the interaction between the particles, taken as a sum over pairs
(i, j). The summation is carried over both indices i and j, but terms with i = j are
excluded. The factor 1/2 ensures that each pair is counted only once. In general,
H may depend on the spatial as well as the spin coordinates of the particles.
Recall , from the fourth postulate of quantum mechanics, that the wave function
of a particle can be expanded in terms of a complete set of states l1>v). Similarly, the
N -particle wave function can be expanded in terms of a complete set of N -particle
states. These are constructed as properly symmetrized products of single-pm1icle
states (SPSs), as we will show in Chapter 3. Although any complete set of SPSs is
J8
2.2 Electron gas 19
(a) (b)
Figure 2.1 (a) The free electron model: the interactions of the electrons with each
other and with the ions are ignored; each electron moves freely within the crystal
and its wave function obeys periodic boundary conditions. (b) The jellium model:
the electrons interact with each other and with a uniform positive background.
adequate for this purpose, a more convenient set is generated by the single-particle
Hamiltonian h through the eigenvalue equation
(2.3)
Each SPS i¢v) is charactelized by a set of quantum numbers that are collectively
denoted by 1). In the following sections we describe convenient SPSs for free
electrons, electrons in a periodic potential, and electrons in a two-dimensional
system that is in the presence of a magnetic field.
(2.5)
20 Single-particle states
The spin ket I(T ) is either I t) or I,j,). In Dirac notation, the SPSs are denoted by
Ik(T) . An SPS is thus described by four quantum numbers: k x , k y , k;:, and (T. The
SPSs form an orthonormal set,
L F(k) -+ ~!
k (2n)
d k F (k) .
3
In terms of the electron number density n = N I V, the Fermi wave vector is thus
given by
(2.10)
2.3 Bloch states 21
Figure 2.2 Fermi sphere of radius k F. At zero temperature the states within the
sphere are all occupied, while the states outside the sphere are all empty.
E = Ln k /2 m == 2 Ln e 12m .
2 2 2
ka k
The factor 2 arises from summing over a (t or ~). The sum over k runs over all
vectors within the Fermi sphere; replacing the sum by an integral,
E = 2 -V-3
(2n)
J 3
d k
2
n2 k 2 12m = -2nVh2-m l
0
kr
k dk
4
= Vh2
] On
2
m
5
kF = -
V 3
Sn
2 k FE F
Figure 2.3 Graphene, a two-dimensional crystal. The primitive lattice vectors are
a1 and a2, and the basis consists of two carbon atoms, A and B. The parallelogram
formed by a1 and a2 is the unit cell.
lattice point. The vectors at , a2, and a3 are called the primitive lattice vectors, and
the parallelepiped they form is the primitive, or unit, cell. A crystal is obtained
if one or more atoms, called the basis, are placed in each unit cell. We may thus
write
the potential energy VCr) of an electron (due to its interaction with the ions) has
the same value at points P and Q. VCr) is thus a periodic function of position, its
periodicity being that of the lattice.
In a crystal, Bloch states form a convenient set of single-particle states (SPSs).
These are solutions to the Schrbdinger equation
(2.12)
Figure 2.4 Two-dimensional crystal with primitive lattice vectors al and az. The
points P and Q are separated by a lattice vector equal to 2al + a2. The potential
felt by an electron at point P, due to its interaction with the ions, is identical to that
felt by the electron at point Q.
Note that
Now consider
The periodicity of the Hamiltonian, namely that her + R) = her), results from (i)
\7;+R = \7; and (ii) VCr + R) = VCr). Since fer) is arbitrary, Eq. (2.1S) implies
that hand TR commute. Furthermore,
TR TR, fer) = TR fer + R ') = fer + R ' + R) = TR' fer + R) = TR' TR fer).
Hence, TR and TR' commute. We conclude that {h , TR, TR, . ... } is a set of commut-
ing operators. The eigenstates of h can thus be chosen so as to be simultaneously
eigenstates of the translation operator TR for every lattice vector R. Let the eigen-
values of TR be ),,(R): TR ¢ = )"(R)¢. Since
This holds if )"(R) = e ik.R for some vector k . For different values of k, we get dif-
ferent values of the eigenvalue )"(R), and cOlTespondingly different eigenfunctions
cP(r). Thus, TRcPk(r) = ei k.RcPk(r) . From the definition of the translation operator,
it follow s that, for any lattice vector R,
(2.16)
We have not yet indicated what values k may assume. These are determined by the
periodic boundary conditions. If there are NI primitive cells along the direction of
lattice vector ai , N2 primitive cells along a2, and N3 primitive cells along a 3, the
periodic boundary conditions take the fonn:
(2.17)
These boundary conditions are adopted under the assumption that the bulk proper-
ties of a crystal do not depend on the choice of boundary conditions on its surface.
Periodic boundary conditions are more convenient for mathematical analysis than
fixed boundary conditions, for which the wave function vanishes on the surface of
the crystal. From Eq. (2. 16), we can write
(2.19)
Here, Q = la 1 .a2 x a 31is the volume of the primitive cell. The vectors b I , b 2, and
b 3 have the dimension of llLength. It is readily checked that
(2.20)
If we write k = .Blb l + .B2b2 + .B3b3, then k .ai = 2JT.Bi, and Eq. (2.18) becomes
exp(2JTi.BiNi) = I ~.Bi = m dNi, wheremi is an integer. Therefore
ml m2 1113
k = -bl
NI
+ -N2b?- + -b"
N '
ml, /112, 1113 E Z . (2.2 L)
3
Equation (2.16), with k given in Eq. (2.2 1), is the first form of Bloch's theorem.
An alternative and useful form of Bloch 's theorem is obtained as follows. Let
uk(r) = e-i k.r cPk(r). Using the first form of Bloch 's theorem. we find
2.3 Bloch states 25
Hence, uk(r) is a periodic function with the same periodicity as the lattice. The
stationary states are given by
(2.22)
The stationary states are thus plane waves modulated by a function that has the peri-
odicity of the lattice; this is the second form of Bloch's theorem. The Schrodinger
equation is now written as
Noting that
(2.23)
For each k there are infinitely many eigenvalues Elk, E2k, . . . with corresponding
eigenfunctions U Ik(r), u2k(r), .... The periodic functions should thus be written
as unk(r), where n = 1, 2, ... is called the band index. Because the separation
between nearby k-points is extremely small in comparison with the magnitude of
the reciprocal lattice vectors, the eigenvalues Enk may be considered as continuous
functions of k. For example, in a cubic crystal, where aj, a2, and a3 have the
same magnitude a and are at right angles with each other, the reciprocal lattice
vectors b l , b 2, and b 3 are also at right angles with each other, having the magnitude
b = 2JT / a. In contrast, the separation between nearby k-points is 2JT / L; for a crystal
with 1024 atoms, a / L is 10- 8 . Thus, we see that the separation between adjacent
k-points is much smaller than the size of the reciprocal lattice vector.
According to Eq. (2.21) , the set of allowed values for k is infinite. However,
there is redundancy in the resulting set of energies and eigenfunctions, and the
values of k can be restricted to a finite set. We prove this assertion as follows . The
reciprocal lattice vectors b l , b 2 , and b 3, treated as plimitive vectors, generate a
lattice in reciprocal space where a general lattice vector is
26 Single-particle states
We remark that for any reciprocal lattice vector G and any real lattice vector R, the
relation eiC .R = 1 is satisfied. Equation (2.23) implies that
(2.24)
!12
[ --(\7
2m
+ iki + VCr) ] Ac = Ek+C AG.
(2.25)
This is the redundancy we mentioned earlier. The above relations allow us to restrict
the values of k to one primitive cell in reciprocal space, since any k-point outside
the primitive cell can be reached from a k-point inside the primitive cell by adding
some reciprocal lattice vector G. It is conventional to choose a Wigner-Seitz cell ,
known as the first Brillouin zone (FBZ), as the primitive cell in reciprocal space. It
is constructed by drawing all reciprocal lattice vectors that emanate from a chosen
point in the reciprocal lattice, and then drawing the perpendicular bisector planes
of these vectors. The volume bounded by these planes, and centered on the chosen
point, is the FBZ. This procedure is illustrated in Figure 2.5 for a two-dimensional
square lattice.
Taking into account the spin state of the electron, the single-particle states,
expressed as Bloch functions, are given by
(2.26)
Five quantum numbers characterize the single-particle states: the band index n =
1,2, ... , the three components kx, ky, k: , of the wave vector k E FBZ, and the spin
index a = tor -J,.
From the above discussion, we see how bands arise once the static potential
produced by the ions at their equilibrium positions is taken into account. It takes
2N j N2N3 electrons, i.e., twice the number of primitive cells in the crystal, to fill
2.4 Example: one-dimensional lattice 27
/ ,,
/
/
,,
/
/ ,,
/
/
,,
/
,, /
,, /
/
/
21f/a ,, /
,, /
/
a /
21f / a
a
Figure 2.5 A square lattice, in real space, of side a (left figure) , and the reciprocal
lattice. also a square lattice, of side 2n / a (right figure). The shaded area is the first
Brillouin zone.
one band, as deduced from the following argument. From Eq. (2.21), neighboring
k-points along the b i direction (i = I, 2, 3) are separated by b i / N i . The volume
in reciprocal space occupied by one k-point is thus b 1.b 2 x b 3 / Nl N 2 N 3, i.e., the
volume of a primitive cell in reciprocal space divided by N 1N 2 N 3. Hence, the
FBZ, whose volume is equal to that of a primitive cell in reciprocal space, contains
Nl N2N3 k-points. Since each state with quantum numbers nand k can accommo-
date two electrons, with opposite spin projections, each band can accommodate up
to 2N I N2N3 electrons. In the ground state, at zero temperature, electrons fill the
lowest energy states. If we end up with a situation where some bands are completely
filled while the rest are empty, the crystal will be an insulator. If we do not, the
crystal will be a metal.
Finally, we note that, in a metal with a partially filled band, it is generally the
case that to a good approximation, we may set uk(r) = l i fo, and take Ek =
/i 2k2/ 2m *. where 111 * is an effective electron mass. This is known as the effective
mass approximation. In many metals. 111 * ~ 111, the free electron mass. A similar
situation occurs in semiconductors with partially occupied bands, due to either
thermal excitations or doping. In this case, however, 111 * may be very different from
111; in GaAs, for example, the effective electron mass is 111 * = 0.067111.
V(x)
Figure 2.6 A line of identical atoms where the separation between adjacent atoms
is (./. The periodic potential seen by an electron is V (x), and it is produced by the
ions sitting at their equilibrium sites.
a well-defined energy in isolated atoms, they broaden into a band when atoms
are brought closer together. The one-electron Hamiltonian is H = p2/ 2rn + V (x),
where Vex) = vex + a) is the periodic potential seen by the electron. This is
sketched in Figure 2.6.
Now consider the band formed by the broadening of one type of atomic orbital,
e.g., the 3s orbitals. Let 1¢1Il) be the atomic orbital centered on atom number rn,
located at x = ma , rn = 1, ... , N. We assume that (¢m IHI¢II1) = E, and that for
n =j:. m , (¢nI HI¢II/ ) = -tO Il • II1 ± I, i.e. , we assume that the overlap of atomic wave
functions is appreciable only between nearest-neighbor atoms. We take t to be real.
Our goal is to find the energy dispersion Ek for this band.
We want to solve the eigenvalue equation Hl\IId = Ekl \lld . We take the N
atomic orbitals centered on the N atoms as the basis states in which 1\11 k) is
expanded,
N
The coefficients Cmk are not arbitrary; they are chosen so that \IIk(X) is a Bloch
function, being a stationary state of an electron in a periodic potential. We thus
require that \IIk(X + a) = eika \llk(X); this, in turn , implies that
L C k¢(X +
lI1 a - rna ) = e
ika
L Cmk¢(X - rna). (2.27)
III III
(2.28)
2.5 Wannier states 29
Note that, strictly speaking, in the summation over n, n ranges from 0 to N - 1,
and the series is
On the RHS of Eq. (2.29), (<I>I1I<1>m ) is the overlap of the atomic orbitals centered
on sites 11 and In. Although it is nonzero, we will neglect it if 11 -# In and take
(<I>I1I<1>m) = 01lm. The RHS thus reduces to Ekeil1ka. Therefore,
band index n, lattice site R i , and spin projection a, consider the states
These are called Wannier states; they have the following properties:
The first property follows directly from the second form of Bloch's theorem. From
the definition of the Wannier state. we can write
Since Unk has the periodicity of the lattice, we can rewrite the above as
A-.
'PI/ia () 1 ""
r = .j'N ~ e
ik.(r - R,)
UI/k
(r -
R)I)
i a .
N k EFBZ
This shows that the Wannier function is a function of r - R i , and we can write it
as <PI/a (r - R i ); it is centered on R i .
Since the Bloch states are orthonormal: (n'k' a'inka) = Onl/'Okk'Oaa' ,
== I ""
Onl/' 000" N ~e
ik. (R -
) R)
I == 0nu ' 0(TO ' 0ij· (2.32)
k
In the last step, we used the results of Problem 2.1. Wannier states are thus 011honor-
mal. Furthermore, since the number of lattice sites is N, which is the same as the
number of k-points in the FBZ, the set of states In i a) is complete: it forms a basis
for the expansion of any state.
To see that <PI/a (r - R i ) is localized on the lattice site R i , we consider a one-
dimensional lattice and assume that UI/k (x) = 1/ ,fL, so that the Bloch states are
plane waves. For any lattice vector R of magnitude R = ma, where m is an integer,
2.6 Two-dimensional electron gas in a magnetic field 31
¢n(x - rna) -
_
y
1 L
fATT-
N L 2n
i rr l a
-Tria
dk e
ik(x -m a) _
-!AT
.JI
ny N
sin[n(x - rna) / a]
X - rna
.
A plot of ¢II(X - rna) shows that most of the weight of the function is at x = rna
(see Figure 2.7) . The localization of the Wannier function is not as strong as that
of an atomic orbital. The damped oscillations are necessary for Wannier functions
centered on different sites to be orthogonal.
1 ? q
L = -mv- + -v .A , (2.33)
2 c
does indeed produce the COLTect equation for the force. The proof consists in using
the Euler-Lagrange equation of motion
d aL aL
dt ax ax
32 Single-particle states
with similar equations for y and z, along with the following two equations:
v x B = v x (V x A) = V(v.A) - (v.V)A
dA aA aA dx aA dy aA dz
- = - + -- + -- + -- = (v.V)A.
dt at ax dt ay dt az dt
The first equation is checked easily. In the second, aA/at = 0 because A does not
depend explicitly on time, since B is a static field.
From the Lagrangian, we construct the canonical momentum,
Px = aL/ax = mx + (q /c)Ax,
along with similar equations for Py and Pz. It follows that mv = p - (q/c)A, and
the kinetic energy is given by
1 2
T = -mv = -
1 (
p - -A
q)2
2 2m c
Thus, for a charged particle in a magnetic field, the kinetic energy portion of the
Hamiltonian is obtained by p ---+ P - q A/ c.
For a given B, there are infinitely many choices for A, because for any A, the
vector potential A' = A + V f, for an arbitrary function f, is an equally valid
choice (since V x V f = 0). Making a particular choice for A is called "fixing
the gauge." For a uniform magnetic field B in the z-direction, we may choose
A = (- By, 0, 0); this is the Landau gauge.
The Hamiltonian for the 2-DEG is H = Li hCi) + ~ Lioij vCi, j). The second
term describes the electron-electron interaction. The single-particle Hamiltonian
is given by
Inserting ¢(x, y) = fey )e ikx /..JL: into the Schrbdinger equation h¢ = E¢, we
obtain
This may be rewritten, in terms of w = e B / me, the cyclotron frequency, and with
Yo = hek/eB, as
Further reading
Ashcroft, N.W. and Mermin, N.D. (1976). Solid State Physics. Philadelphia: Saunders
College.
Kittel , C. (2005). Introduction to Solid State Physics , 8th edn . New York: Wiley.
Omar, M.A. (1993) Elementary Solid State Physics: Principles and Applications, revised
printing. Boston: Addison-Wesley.
34 Single-particle states
Problems
2.1 Important sums.
(a) Vk E FBZ, show that L e ikR" = N Ok. 0, where N is the number of prim-
n
itive cells and the sum runs over all lattice sites.
(b) For every lattice vector R, show that L e ikR = NOR. O.
k EFBZ
2.2 Free electron model at zero temperature. Consider a system of N free elec-
trons confined to a cube of volume V, at T = 0 (ground state). Define the
dimensionless parameter rs through the relation (4n /3)(r s ao? = V/ N, where
ao is the Bohr radius.
(a) Express the mean energy per electron in terms of rs.
(b) Show that da (E F), the density of states per unit volume, per spin orientation,
at the Fermi energy, is given by mkF / 2n 2!i 2 .
2.3 Free electron model in lower dimensions. Consider the free electron model in
one and two dimensions at T = O.
(a) Show that the Fetmi wave vector is given by
2D
kF = lJ2nn
nn / 2 lD
where n is the electron number density (in 2D, n is the number of electrons
per unit area, while in ID, it is the number of electrons per unit length).
(b) Show that the mean energy per electron is given by E = EFd/(d + 2),
where d is the dimension of space and E F is the Fermi energy.
2.4 Graphene bands. Graphene (see Figure 2.3) has two atoms per unit cell,
denoted by A and B. The x - and y-axes are chosen such that a I =
a(-J3/2, -1/2), 3 2 = a(O, 1), where a = 0.246 nm is the lattice constant.
An isolated carbon atom has the electronic configuration 1S2 2s2 2 p2. To form
graphene, one electron is excited from 2s to 2p, and the new configuration
is ls 22s I2p.~. 2p~. 2p~. The 2s, 2px, and 2p,. orbitals get hybridized (mixed)
and form three sp2 orbitals that are oriented in the x-y plane at 120° with
each other. The Sp2 orbitals on nearby atoms form strong bonds in the plane,
giving rise to a honeycomb structure, and they broaden into the a-bands that
lie very low in energy. The pz orbital on each atom is perpendicular to the
graphene plane and is occupied by one electron. The P: orbitals broaden into
two n -bands (there are two atoms, hence two Pc orbitals, per unit cell).
Let us take atom A to sit at the origin of coordinates. The P: orbital on this
atom is ¢(r), and that on any atom of type A, that can be reached from A by a
lattice vector R II , is ¢(r - RII) ' The P: orbital on B is ¢(r - 8), where 8 is the
Problems 35
(d) In the vicinity of the points K(2n 1 fia. 2n 13a) and K'(O. 4n 1,j3a) in
the first Brillouin zone, show that Ek = ±hVFk where VF = ,j3tal 211 is
the Fermi velocity, k = Ik - (2n 1,j3a , 2n 13a)1 (near point K) , or k =
Ik - (0, 4n l ,j3a) 1(near point K' ). The energy dispersion is thus linear in
the vicinity of K and K' .
(e) Assuming linear di spersion, show that the density of states per unit area is
dee ) = 81 EI / (3na 2 t 2 ) .
2.5 More on graphene. Assume that the p ~ orbital on each site in graphene is
e
described by the wave function ¢(r) = Arcos exp( - Zr 12ao), where A is a
normalization constant, ao is the Bohr radius, e is the angle r makes with the
c-axis (the axis perpendicular to the graphene plane), and Z is the effective
36 Single-particle slales
charge on the nucleus (the nuclear charge is screened by the two core electrons
in the Is orbital, and to a lesser extent by the valence electrons; Z ~ 3). Show
that
2.7 Density of slates D(E). The total number of states within a shell in k-space
bounded by the two constant energy surfaces Ek = E and Ek = E + dE is
D(E )dE. The number of states within this shell is twice the number of k-points
within the shell because of spin degeneracy. Therefore,
D(E)dE = 2~
(2n)
1 shell
3
d k = 2~
(2n)
f dSEd k.l
where d SE is an area element on the inner surface and dk.l is the perpendicular
distance between the two sUlfaces of the shell. Show that
D(E) = 2 -V-
(2n)3
f dSE
IVkEkIEk=E
.
3
Second quantization
Historically, quantization of the motion of particles was developed first. The state
was described by a wave function and observables by operators. When dealing with
interactions between particles and fields, such as the electromagnetic field, the fields
were treated classically. Classical field equations look like the quantum mechanical
equations for the wave function of the field quanta. For example, the Klein-Gordon
classical field equation is similar to the quantum mechanical wave equation for a
relativistic spinless particle. Quantizing the fields, leading to quantum field theory,
appears to be quantizing a theory that has already been quantized; hence the name
"second quantization." In reality, there is only one quantization and one quantum
theory.
The method of second quantization is important in the study of many-particle
systems. It enables us to express many-body operators in terms of creation and
annihilation operators, thus rendering calculations less cumbersome. Moreover,
the method makes it possible to treat systems with a variable number of particles;
that is why the method initially emerged in the context of quantum field theory.
In Chapter 1 we indicated that anyone-particle wave function may be expanded
in a complete set of states. In this chapter, we show that products of single-particle
states, when properly symmetrized, form an orthonormal basis for the expansion
of the wave function of an N -particle system. We then introduce creation and
annihilation operators and show how to express one-body and two-body operators
in those terms.
37
38 Second quantization
We will show that the N-particle wave function \.jJ(1 , 2, ... , N) can be expanded
in terms of products of the single-particle states. We may proceed as follows.
Suppose that we fix the spatial and spin coordinates of particles 2, 3, ... , N. Then
\.jJ(l , 2 .... , N) is a function of the coordinates of particle 1 alone; hence, we can
expand it in a complete set of states ¢v(l),
There is an alternative way to arrive at this result. States l¢vl) I, for all values of J) I ,
form an orthonormal basis for vector space VI, the Hilbert space of the states of
particle I. States I¢V2)2 form an orthonormal basis for V2, the vector space of the
states of particle 2, and so on. The state vector I\.jJ ) of the N -particle system belongs
to the direct product space V (N) = VI ® V2 ® ... ® VN , whose orthonormal basis
consists of the direct product states l¢vl) I ® 1¢V2 h ® ... ® I¢VN) N . It follows that
two indices. For bosons, we can achieve this by summing the product over the N!
permutations of 1, 2, ... , N ; the basis states are thus given by
<P vB v v (l , 2, .. . , N) =
12 · ·· N
I l L¢ VI [P(l)]¢ V? [P(2)] ... ¢ VN [peN)].
n N.fNi. I NI -
nil '
II . P
(3.3)
Here P(l) , P(2) , . .. , peN ) is a permutation of 1, 2, ... , N , and nil is the number
of times the index f...L appears in the product. The factor before the summation
ensures that <pB is normalized.
For fermions , a similar expression for the basis states is used, except for the
following two modifications. First, nil is either 0 or 1 (Pauli exclusion principle),
so that nil ! = 1 (O! = 1 and I! = 1). Second, we must inselt a minus sign whenever
P(l) , P(2) , ... , peN ) is an odd permutation of 1, 2, . .. , N . The fermionic basis
functions are given by
The above expression for <pF may be written in the form of a determinant,
----¢3
• ¢2
• • ¢1
Figure 3.1 A system of three noninteracting bosons. Two bosons occupy the
single-particle state ¢! and one boson occupies ¢2 .
In terms of the basis functions <p B . F, the N -particle wave function is now expanded
as
1
= ;,;:; [1>1 (1)1>1 (2)1>2(3) + 1>1 (l)¢1 (3)1>2(2) + 1>1 (2)1>1 (1)1>2(3)
y12
+ 1>1 (2)1>1 (3)1>20) + 1>1 (3)1>1 (1)1>2(2) + 1>1 (3)1>1 (2)1>2(1)]
1
= v'3 [1>1 (1)1>1 (2)1>2(3) + 1>1 (1)1>1 (3)1>2(2) + 1>1 (2)1>1 (3)1>2(1)] .
2. Three non interacting electrons in a box of volume V occupy the states 1>kt(r) =
ik r ik r
Jve . [ t), 1>k.j, (r) = Jve . [ n,
and 1>k't(r) = Jveik' .r[ t). The wave func-
tion for the system is the Slater determinant
e ik .r2 [ t)2 e
ikr3
[ th
eik .r2 ,j,.)2
[ e
ikr3
h
[ ,j,. (3.8)
eik' .r2 [ t)2 eik ' r3[ t)3
E3 ¢3
E2 t ¢2
E1 T ¢1
Figure 3.2 A system of two noninteracting electrons. One electron, with spin up,
occupies the single-particle state ¢l. Another electron, with spin down, occupies
¢2. The energy of the system is E I + E2.
Here, S is the total spin operator, and 5~ is its projection on the z-axis,
The single-particle states are solutions of h¢na = fn¢na , where n is the set of
spatial quantum numbers and a = t or -.1-. Suppose that one electron occupies
the state ¢It = ¢l (r)a, where a = 1 t), while the other electron occupies the
state (h t = ¢2(r)f3, where f3 = 1-.1- ) (see Figure 3.2) . The state of the system is
given by the Slater determinant
¢l (r2)a(2) I.
¢2(r2)f3(2)
I 1
\.lI(1 , 2) = -J2 [¢ I(rl)¢2(r 2) + ¢1(r2)¢2(rl)] -J2 [a(l)f3(2) - a(2)f3(l)]
1 1
-J2 -J2 [a(l)f3(2) + a(2)f3(l)].
I
\.lI (l, 2) = [¢1(r l)¢2(r 2) - ¢1(r2)¢2(rl)]
42 Second quantization
\I15D(I
I
, 2) = h1 [ \11(1 , 2)-\11(1
' , 2) ] .
It is easy to verify that \11 ~ D(l, 2) is the Slater determinant which describes the
configuration where the electron in orbital ¢J has spin down while the electron
in orbital ¢2 has spin up. This is also a stationary state of H = h(l) + h(2),
with energy EJ + E2. In other words, we may choose \I1(l , 2) and \11 ' (1 , 2) as
the two degenerate stationary states; each is expressed as the product of a
spatial part and a spin part. Since \11(1 , 2) and \I1 ' (l. 2) are degenerate, the Slater
determinants \11 5D(1 , 2) and \lI ~ D(1 , 2), which are linear combinations of \11 (I , 2)
and \11' (1,2) , are also stationary states with the same energy, even though they
cannot be factored into the product of a spatial part and a spin part.
3.4.1 Fermions
Each single-particle state I¢v) is associated with a creation operator c!,defined by
(3.9)
ct
The operator thus creates a fermion in the single-particle state I¢v) ; it adds a row
to the Slater determinant, which becomes the first row of the new (N + 1) x (N +
1) determinant. The action of the creation operator is illustrated in Figure 3.3 . The
action of ci. v
on a Slater determinant yields 0 if coincides with any of the indices
3.4 Creation and annihilation operators 43
··· ··•
•
¢4
¢3
¢2
-- •
•
¢4
¢3
¢2
• ¢l • ¢l
··•
•
¢4
¢3
¢2
-- o
• ¢l
Figure 3.3 The action of the creation operator d
on a system of fermions: if the
state ¢2 is empty, the operator creates a particle in that state, but if the state is
occupied by one pm1icle, the result of the action of d
is zero.
VI ... VN, for then the resulting determinant would have two identical rows. Stated
differently, we cannot create a fermion in a state that is already occupied (Pauli
exclusion principle) .
For an arbitrary Slater determinant i¢Vl ¢"2 ... ¢VN) (arb itrary in the sense that
the single-particle state indices VI ... VN are arbitrary), consider
t t t
cvCV,i¢Vl "'¢VN) = cvi¢V' ¢Vl" ' ¢VN) = i¢V¢V' ¢Vl "'¢VN)
c~, C!i¢Vl ... ¢v.v ) = i¢V'¢V¢Vl ... ¢ VN) = -i¢V¢ V'¢Vl ... ¢VN) '
The minus sign results from the interchange of the first two rows . Since i¢Vl ... ¢ VN)
is arbitrary, adding the above two equations gives us
t t'
cvc + cv'tctv -o
- {t
cV' c tv' }-O
- ,
v ==}
{A , B}=AB+BA. (3.10)
Note, in particular, that if V = Vi, we have (C!)2 = 0: we cannot put two fermions
in the same state, as Figure 3.3 illustrates.
Next, we define an annihilation operator C v that annihilates a patticle in state
i¢v),
(3.11)
The annihilated state must be on the left, i.e ., it must be the first row in the Slater
determinant. If ¢v is not on the left, then it must be moved to the leftmost position,
44 Second quantization
••
• ··•
• •
•
¢4
¢3
¢2
-- ¢4
¢3
¢2
• ¢l • ¢l
··•
¢4
• ¢3
¢2
-- o
• ¢l
Fig ure 3.4 The action of the annihilation operator C2 on a system of identi cal
fermions. If ¢2 is occupied by one particle, C2 renders the state empty. If the state
is empty, the action of C2 yields zero.
introducing a minus sign every time it is interchanged with another state. For
example,
Clearly, the state occupied by the particle to be annihilated mLlst be among the
collection of states in the Slater determinant; otherwise the action of C v is defined
to yield zero,
which shows that (cb t = CV' Taking the adjoint of {c!, c~, } = 0, we obtain
Now suppose that the state l1>v) is occupied, say with I) = V j+ l. Then
{c V, c!} = l.
We now calculate {c v, c~, } for I) "# Vi .Consider (cv c~, + c~, cv )l1>vl . . . 1>VN )' This is
equal to zero unless I) E {1)1 , ... , VN } and 1)' f/. {Vt, ... , V N }. Let us assume that
this is indeed the case. Then
(cvc tv' t
+ cv,cv)l1>vl 1>v2 . . . 1>v ... 1>vN) = -(cvc vt' t)1 1>v1>"2"
+ cv'cv .1>vl .. . 1>VN)
Here, y eO) is the Hilbel1 space (vector space) for N = 0, i.e., it contains only the
vacuum state 10); y ( l ) is the vector space for a one-particle system, and so on. The
operator d, by creating a particle in state l<Pv), increases the number of particles
by 1; hence, if 1\11) E y(k ) , then dl\ll) E y (k+I ) , while cvl\ll) E y (k-I ) . The vector
spaces y (k), y (k+ I), and y (k-I), are parts of F; hence, creation and annihilation
operators act upon the Fock space.
Finally, we note that, in the number-representation, d and C v act in the following
way
dlnl ... n v ... ) = (_IYI +n2 +·+n I (l - nv)lnl ... nv
V
- + 1· .. ) (3.13)
cvlnl .. . n v "') = (_I)"I+n 2++n v- lnvlnl '" nv - 1···). (3.14)
3.4.2 Bosons
We only give a brief account of creation and annihilation operators for the case of
bosons. We define a creation operator a~ by the following relation,
(3.15)
3.5 One-body operators 47
(3.16)
If the state I¢v) is vacant (nv = 0), the action of av yields zero. By using an
argument similar to the one used in the case of fermions, one shows that a v is the
adjoint (hermitian conjugate) of a!.
The symmetry of the state of identical bosons,
under the exchange of coordinates of any two particles, leads to the following
commutation relation between the creation and annihilation operators
(3.l7)
Equations (3.15-3.17) should be familiar from the study of the quantum harmonic
oscillator.
(3.18)
vv'
This is the second quantized form of Ho, and it holds true for both fermions and
bosons. The expression is plausible: a one-body operator is the sum of single-
particle operators h(1) , h(2) , ... , h(N). The effect of a single-particle operator is
to scatter a particle from a state I¢v) into a state I¢v' )' The scattering process can
be viewed as the annihilation of a particle in state I¢v), followed by the creation
of a particle in state I¢v')' The amplitude for this process is the matrix element
(¢ v' lhl¢v).
48 Second quantization
3.6 Examples
In the following we give a few examples that illustrate how to express one-body
operators in second quantized form.
The second quantized form of the operator depends on the basis set of single-
particle states. Let us choose the plane waves IkCJ) as basis states. The electrons are
assumed to move within a box of volume V = L3. Assuming periodic boundary
conditions, we have
1 'k
<Pka(r) = (rlkCJ ) = .;ve,rICJ),
n \l 21 kCJ ) Ck'a,Cka'
T = ~ ~ k CJ - 2m t
ka k'a '
Since -(j1 2/2m)\l2I kCJ ) = (ft 2 k 2 /2m)lkCJ) and (k 'CJ'l kCJ ) = Okk' Oaa" the second
quantized form of the kinetic energy is
(3.19)
Vex! = L vU)
;=1
If v is spin-dependent, the matrix element (0"1 v 10') is evaluated first; the result will
be a function of r , and the r integration is then carried out. In the simpler case
where v is spin-independent, (a/lvla) = v(r)oO'O'" and
(3.20)
vCr) = V
1 "L..., e,.qrv q . (3.21)
q
N
nCr) = L o(r - r;). (3.23)
;=1
50 Second quantization
This is a one-body operator of the form L~=I f(i). In the basis IkO"),
1 '"'
= V . n .
~ e,q·r (3.24)
q
q
nq = L cta Ck + qa . (3.25)
ka
(3.26)
(3.27)
Notice that the order of nand m in the matrix element differs from the order in the
operators.
3.8 Translationally invariant system 51
Using the basis states Ika ), the second quantized form of Vim is
= _1_ fd 3 3 i k i c
0 0
rl f d r 2 e (k3- \ ).r\ e (k k2). r vCr I - r 2 ) .
2
V2 a\a3 a2 a4
1
= V Oa\a30a2a4 L VqOq.k \- k Oq.k - k 3 4 2 •
q
Therefore, in the expression for Vim, the sum vanishes unless k l = k3+ q, k2 =
k 4 - q , al = a3, and a 2 = a 4; hence
ku k'u'
Figure 3.S Schematic representation of the interaction of two particles. The two
particles initially have wave vectors k and k' . The interaction is viewed as a
collision in which one particle transfers momentum nq to the other.
Ik + qa) and Ik' - qa'). The scattering process may be represented pictorially, as
shown in Figure 3.5.
1 J e2
Vc = 2L v(i , j) = 2 L Ir· - rl (cgs).
i f.} i f.} ' }
v q =e 2
f I - Iq.rd 3 r=e 2
-e
r
1°°? 11_l o127Tr
0
r-dr d(cose) J - Ilj,cose
d¢_e ' . .
e- w . 27re21 °O .
Vq = lim e 2
/1 --> 0- f _ _ e- /q .rd 3 r
r
= lim - . -
11-->0+ lq 0
[e(lq -l1)r - e(-i q- l1 )rJ dr
q-1
3.10 Electrons in a periodic potential 53
The expression for the Coulomb interaction thus becomes
(3.29)
Here, V is the system's volume. The Coulomb energy diverges because of the q = 0
term. We shal1 see in the next chapter that the q = 0 term drops out of the sum as a
result of charge neutrality : in a crystal, there are both electrons and positive ions.
H = Ho + Vc (3.30)
2
1 e
Ho = L [p~ 12m + v(r;)]. Vc =- L . (3.31)
2 ;f.j Ir; - r jl
vCr) is the potential produced by the static periodic lattice of ions (the effects of
ionic vibrations are studied in Chapter I 1), and Vc is the Coulomb interaction
54 Second quantization
Ho = L L (n'k'u ' l -
Il ka l1 'k'a '
;:1 \7
2
+ v(r)lnku )c~'k'a,Cl1ka.
Since the Bloch states are eigenstates of the single-particle Hamiltonian,
The matrix element M = (nil k 'lu; n;k;u~ I,el: In I kl UI n2k 2u2) is given by
e
2
= ~ L 2
4n e e i q .(r l -r2 ) (3.34)
Irl - r 21 V q q2
e2 f *
M = Oala'1 Oa,a'
- 2 L -4nVq-
q
-, Vr ' k, (rl) e "1 1
i q .r
I
3
Vrl1 l kl(rl) d r l
Recall that, for any lattice vector R, the Bloch function satisfies the relation
(3.35)
This is the first form of Bloch 's theorem (Section 2.3). If we now make a change
of variable: r l ~ rl + R, the integral does not change, but the integrand gets
multiplied by the factor ei(kl-k'i + q) R ; this factor mu st be equal to unity. Since
this is true for every lattice vector R, it follows that k 'l = k I + q + G, where
G is a reciprocal lattice vector. However, since kl ' k 'l E FBZ, G will vanish if
3.10 Electrons in a periodic potential 55
k) + q E FBZ, but if k) + q '/. FBZ, then G is the reciprocal lattice vector that
brings k) + q back into the first Brillouin zone. We thus require that k') = k) + q,
with the understanding that k), k) + q E FBZ. A similar argument shows that
kS = k2 - q. Hence,
(3.36)
k.k±q -- (n'k
FiliI' ± qa le±iq.rlnka) . (3.37)
Here, n is a band index, i is a lattice site index, Ri is the lattice vector from the
origin (chosen as some lattice point) to the lattice site i, and the sum is over all
k-points in the first Brillouin zone (FBZ). Since our interest is only in the electrons
in one band, we may drop the band index and write the Wannier state as Ii a ) and
the Bloch state as Ika). We define the operator cia
that creates an electron in the
state lia ),
where 10) is the vacuum state. The Wannier and Bloch operators are thus related
according to
The first equation, connecting the creation operators, is obtained directly from
Eq. (3.39), while the second equation, connecting the annihilation operators, is
obtained from the first equation by taking the adjoints on both sides of the equal
sign. These equations can be inverted,
_
Cka -
1 ' " e -i k.R, Cia'
rAT ~
. (3.41)
V N I.
(3.42)
The matrix element Vij i' j' depends on the degree of overlap of the Wannier func-
tions. When the overlap is very weak, the onsite Coulomb repulsion dominates
the interaction. In this case, we ignore Vij i'j' except when i = j = i' = )" and
set Vii ii = V. Keeping only nearest-neighbor contribution to the hopping matrix
element ti j, the Hamiltonian reduces to
where nia = c;aCia is the operator that represents the number of electrons at site
i, and < ij > indicates that i and j are nearest-neighboring sites. In writing the
interaction term, we have used the fact that cla = 0 and that C;a ,Cia = - CiaC;a' for
a =f. a'. The model described by Eq. (3.43) is the Hubbard model (Hubbard, 1963).
3.1 J Field operators 57
The fi eld operator w; (r) is defined as the operator that creates a particle with spin
projection Q" (t or t for an electron, for example) at position r ,
wJ(r)IO) = Iw ).
A comparison with the previous expression for IrQ") gives
wJ(r ) = L ¢,~(r)c~a' (3.44)
11
The field operator that annihilates a particle with spin projection Q", located at r , is
the adjoint of w;(r),
(3.45)
11
For example, if I¢v) = IkQ"), then ¢k(r) = (l/.jV)e ik .r , where V is the volume of
the system and k r , k" k: = 0, ±2n/ L. ±4n/ L ... ; in this case,
(3.46)
58 Second quantization
{\lia(r), \li ; ,(r')} = L <p1I(r)<p,~, (r'){Clla, c'~'a ' } = L <PII (r)<p,:, (r')ollll ,oaa'
lin ' 1ln'
II
Using the completeness property of single-pmticle states (see Eg. [1.20]),
(3.50)
3
=L L ! \li; ,(r)h a'a (r)\lia(r)d r. (3.51 )
a a'
Here, ha'a(r) = (a ' Ih la), and lise is made of Egs (3.44) and (3.45).
For the case of spin-I / 2 particles, sllch as electrons. Ho may be written in a
more compact form. We define the two-component field operators
\li(r) = (Illt(r))
III t (r)
her) = [h tt (r) h y( r )]
h U (r) h yy( r ) .
3. J 1 Field operators 59
(3.52)
Although the above equation looks like an expectation value formula, it is cer-
tainly not; \li te r) and llJ(r) are field operators, not wave functions. If h(i) is spin-
independent, the expression for Ho may be simplified,
aa' a
=~ L
°1°2°3°. .
Id rll
3
d
3
r2\li!\ (r 1)1lJ!2(r2)Va \a2a3a4 IlJa4 (r2)\lia3 (r I)' (3.53)
Here, {nO'} = n I 0'1 n20'2n30'3n40'4. For spin-l /2 particles, each of 0'1, 0'2, 0'3, and 0'4
is either tor t, and Vint is the sum of 16 terms. We can recast the above expression
into a more compact form. We define
IlJ t (r 2)\lit (r l ))
llJ(r?)IlJ(rl) = (\lit (r 2)) 0 (\lit (r l )) = IlJ t (r2)1lJ.j.(rl)
- 1lJ .j. (r2) 1lJ.j.(rl) ( 1lJ.j. (r2)llJ t (rl)
IlJ -f (r2)1lJ -f (rl)
= (\li~(rl)IlJ ~ (r2) 1lJ ~ (rl)\lil (r2) IlJ l (rl)IlJ~(r2) IlJ l(rl)llJl (r2))'
VtHt VttH
VtHt vHH VttH
VtH.j. ]
vUH vHU vHH
v-fHt vHU vHU
60 Second quantization
Although not shown explicitly, each of the 16 matrix elements in the above matrix
is a function of rl and r 2. It is left as an exercise for the reader to show that Vim
may be written as
Vinl = ~L
aJo2
f \jJ~I(rl)\jJ~2(r2)v(rl' r 2) \jJa2(r 2) \jJal(rl) d 3rl d3r2.
3.11.5 Examples
3.1 l.5.1 Particle-number density
The particle-number density operator nCr) = Li o(r - r ;) is given in terms of field
operators, by
nCr) =L
a
f \jJ~(r/)o(r - r ') \jJa(r ' )d 3 r ' = L \jJ~(r)\jJa(r).
a
(3.55)
By the divergence theorem, the volume integral in the last term is convelted into a
surface integral,
Problems 61
where fi is an outward unit vector normal to the surface. If the surface is at infinity,
the field operator w~(r) = LI1 <p,~(r)da vanishes at the surface because <p,7(r)
vanishes at infinity. On the other hand, if the particles are enclosed within a box
of volume V = L3 and periodic boundary conditions are employed, w~(r)Vwa(r)
will be the same on opposite faces of the box, but the unit vector normal to one
face will be opposite to the unit vector normal to the opposite face. In either case,
the sutface integral vanishes, and
(3.56)
Further reading
Fetter, A.L. and Walecka, J.D. (1971). Quantum Theory of Many-Particle Systems. New
York: McGraw-Hill.
Schwabl, F. (2008). Advamced Quantum Mechanics, 4th edn. Berlin: Springer.
Taylor, PL. and Heinonen , O. (2002). A Quantum Approach to Condensed Matter Physics.
Cambridge: Cambridge University Press.
Problems
3.1 Nonintera cting electrons on a square lattice. Identical atoms sit at the lattice
sites of a square lattice with lattice constant a. Assume that there is one
Wannier orbital on each site, so that one band is formed from these orbitals.
Neglecting electron-electron interaction, and assuming that an electron can
hop from one site to only one of the nearest-neighboring sites, the hopping
matrix element being -t, the Hamiltonian is
H = -t L c JaCj a '
< ij > a
3.2 Graphene revisited. In Problem 2.4, the energy bands in graphene are calcu-
lated using the tight binding method. Here, the calculation is repeated using
the second quantized form of the Hamiltonian. Graphene consists of two sub-
lattices, one of type A and one of type B. Sublattice A consists of aU the sites
of type A, and sub lattice B consists of all the sites of type B. Assume that
there is only one orbital centered on each site (p z orbital). Neglect overlap
between orbitals on different sites, and assume that an electron on one site
can hop to only one of the three neighboring sites. With these assumptions,
62 Second quantization
H = -t LL t
a ia bi+8.a - t L L b/+8.a aia
ia ~= I ia ~= I
where a/a (aia) creates (annihilates) an electron with spin projection a on site
i of type A, b!a(b ia ) creates (annihilates) an electron with spin projection a
on site i of type B, -t is the hopping matrix element, and the sites i + 8 are
the nearest neighbors of site i.
For a given a, there are N operators aia, where N is the total number of
plimitive cells in the crystal. Define N new operators aka. k E FBZ,
I " e -i k.R,
aka = t.i L aia ,
'\I N I
.
Similar definitions are made for bka and b~a' Show that
H = L L EllkC/~kaCllka
11=1 ka
where Elk = -tigki, E2k = ti g ki , and C lka and C2 ka are electron operators
that are linear combinations of aka and b ka .
3.4 Field and number operators. Show that, for boso ns and fermions, the field
operators and the total number of particles operator satisfy the following :
[N , w a ] = -w a , [N. W~ l = w~.
Define \iia(e) = eiNI) Wae-iN(). Show that \iia(e ) = e-i() w a and \ii~(e) =
eie W~. Since operators representing observables, when expressed in sec-
ond quantized form, contain an equal number of creation and annihilation
operators (to be Hermitian), they are invariant under the transformation
Wa ---+ \iia(e), W~ ---+ \ii~(e). Hence, in a many-particle system , any opera-
tor A that represents an observable satisfies the equation eiNe Ae- iN () = A ~
[A. N] = O.
Problems 63
N
3.5 Spin. For a system of N electrons, S = LSi. Let S = (Sx, S.\', SJ. Using the
i= 1
states Ika ) as a basis, show that
3.7 Electron current densiry . In a course on electricity and magnetism, the elec-
tron current density is written as j' = -env, where -e is the charge of the
electron, n is the number of electrons per unit volume, and v is the average
velocity of the electrons . The contribution of electron i to the current density
is -eo(r - r i )Vi = -eo(r - r i )Pi / In . The quantum mechanical expression
for the current density is thus given by
•
J(r) = -?
e L
" [Pio(r - r i) + o(r - r i)pd·
~In ,
I
We write it this way to ensure that j(r) is Hermitian (ri and Pi do not
commute). In the presence of a magnetic field, Pi -+ Pi + eA(ri)/ c, where
A is the vector potential. Show that the current-density operator is given by
.p
J q = -;; L(k
en " 1 t
+ 2q)CkaCk+Qa'
ka
64 Second quantization
3.8 Contact potential. Consider a system where particles interact with each other
via the contact potential
v = ~ Lo(ri - rj).
if.)
3.9 Spin waves. Consider a system of N particles, each of spin s, localized at the
N lattice sites of a crystal. We assume that there exists an interaction between
the particles that tends to align their spins. The Hamiltonian is assumed to be
H = -(1 / 2) L S;.S j
<ij>
where J > 0 and the summation is over nearest-neighboring sites. Here, J has
units of energy, so the spin operators are dimensionless. The spin operators
satisfy the usual commutation relations; e.g.,
H = const + Lliwkatak.
k
What is the value of Wk? The excitations of energy IiWk describe the
spin-wave excitations of the ferromagnet.
4
The electron gas
A metallic crystal has a large number of mobile electrons, of the order of Avogadro's
number, and a cOLTespondingly large number of ions. If our interest is in the bulk
properties of a crystal, we may take the volume V of the crystal to be infinite, and
the number of electrons N to be infinite, while keeping N / V, the number density of
electrons, finite; this is called the thermodynamic limit. The ions incessantly vibrate
about their equilibrium positions, but due to their large mass, they move very slowly
in comparison with the electrons, so that the electrons quickly adjust their state
to reflect whatever positions the ions occupy at any given time. Consequently, to
a good approximation, one may solve the Schrodinger equation for electrons by
assuming that the ions are fixed; this is the Born-Oppenheimer approximation.
The influence of the ionic vibrations on the electronic states, described through
the electron-phonon interaction, may be treated by perturbation theory; this is
discussed in Chapter 11.
A more drastic approximation in the description of a metal is to replace the
mesh of positive ions with a uniform positive background, which results in the so-
called jellium model. In a model such as this, any results obtained are necessarily
qualitative in nature. In this chapter, we study the jellium model. One of our
goals in this study is to show that the divergent term in the Coulomb interaction,
corresponding to q = 0 (see Eq. [3.29]) , is cancelled by contributions to the total
energy from the positive background. This cancellation is a consequence of the
charge neutrality of the crystal, and it holds true even if the approximation of a
uniform positive background is relaxed. Another goal of this chapter is to show
the necessity of performing perturbation expansions to higher, generally infinite,
orders. We willlater proceed to study Green's functions , whereby such a program
may be carried out more easily.
65
66 Th e electron gas
(4.1)
The first term is the sum of the kinetic energies of the electrons and their Coulomb
interactions. From Eqs (3.19) and (3.29),
11 2 k 2 1 4;rr e2
He = L - - CL Cka
2m
+ 11-->0
lim -2
V
LLL q
2
+ fJ. 2ct + qact '_ qa,Ck'a ' Cka'
ka q ka k'a '
(4.2)
The second telm, Hb , represents the Coulomb energy of the uniform positive
background. To find the correct expression for H b , consider a collection of point
charges ql , Q2, ... at positions rl , r 2, .... Their Coulomb energy is
E -
COlli -
-L
1
2
q'q '
Ir - r · I I } (cgs units) .
if.) I }
The factor 1/ 2 ensures that pairs of point charges are counted only once. For a
continuous charge distribution, qi is replaced by peri )d 3 r;, where per) is the charge
density, and the summation is replaced by integration . Therefore,
(4.4)
4. / The Hamiltonian in the jellium model 67
Note how the limits are taken: first V -* 00, then f.L -* 0+. The above integral is
easily evaluated,
f d 3x - - = 4n
e -{U
x
1 0
00
xe-J.1.~'dx
.
= 0
-4n-
of.L
1
0
00
0
e-J.1. X. dx = -4n-
of.L
(1)
-
f.L
=4n-
f.L2
2 2
2n N e
==} H8 = lim lim (4.5)
J.1.-->0 v-->00 V f.L 2
Heb = -e 11m · N
L fd 3r ()
P r e -J.1.l r- r'l= - l'1m - N e-?
- L f d3. r -e-J.1.l-r-r-' l
N
H e- b = - lim
Ne 2
lim - -
{I-->O+ V -->00 V
L f d x -e-J.1.X
N 3
X
- = - lim lim - -
Ne 2
w .... o v-->00 V
4n
L-
N
f.L 2
i =1 i=l
2 2
-4n N e
= lim lim (4.7)
J.1.-->0 v-->oo V f.L2
We thus find
(4.8)
(4.9)
-27Te 2
E'IN == (Vc.q=o + Hb + He-b) IN = iJ.-->
lim
O
lim
v ->oo V f.L2
= O. (4.10)
On the other hand, the average kinetic energy per electron is 3E F15, where EF
is the Fermi energy (see Section 2.2); we are thus totally justified in ignoring E'.
The effect of the positive background is thus to remove the q = 0 term in the
electron-electron interaction Hamiltonian. This is reasonable; after all, crystals are
stable, so the energy per electron should be finite.
The reader may feel uneasy about the results we obtained; they rely on
the mathematical artifact of introducing an exponential damping term, and on
the sequence in which the limits are taken. The reader may rest assured that the
results are correct. In fact, the same results are obtained without introducing
the exponential term. Setting f.L = 0 while keeping the thermodynamic limit:
V -700 , N -7 00, N I V = constant, Eqs (4.4) and (4.6) yield
The q = 0 term in Vc is
= (N
2
-
2V
N)e
2
f 3
d r .
r
Therefore.
-E' I e
N = -N (Vc.q--0 + Hb + He -b) = -2V
-
2
f --.
3
d r
r
where ao = h 2 /(me 2 ) is the Bohr radius (in SI units , e2 ~ e2 / 4]'( co). rsao is the
radius of a sphere whose volume is equal to the average volume occupied by one
electron. Defining the dimensionless quantities
we may recast the Hamiltonian in Eq. (4.'11) into the following form:
Figure 4.1 When an electron in a state of wave vector k above the Fermi surface
is scattered out of this state, an electron-hole pair is created. At low temperatures,
energy and wave vector conservation restricts the final state into which the electron
can scatter to a shell of width k - kF just above the Fermi sphere. It also restricts
the states of the holes created to the shaded regions , which lie within a shell of
width k - kF just below the Fermi surface.
Eol N ~ 2.21 I r; Ry
where one Rydberg (Ry) is equal to e2/2ao, which is about 13.6 eY.
4.3 Ground state energy 71
E / N = Eo / N + EI / N + E2 / N + .... (4.14)
E 1 is given by
(4.15)
The action of Ck'a'Cka on IF ), for k. k' < kF, removes two electrons in states Ik(J' )
and Ik'(J" ); the action of ct+qact'- qa' must restore the two electrons into these
states if the matrix element is not to vanish. There are only two possibilities:
(1) k + q = k , k' - q = k' and (2) k + q = k' , (J' = (J" , k' - q = k. The first
case holds if q = 0, but the term q = 0 is excluded, so we are left with only the
second possibility. Therefore,
2
1 , ,' ,,4ne t t
EI = 2V ~ ~ - 2- (Flck+qacka ck+qacka IF ).
q ka q
(4.16)
Summation over (J' gives a factor of 2. The sums over k and q are replaced by
integrals,
(4.17)
72 The electron gas
(a) (b)
Figure 4.2 (a) Two spheres in k-space, one centered at k = 0 and the other at
k = - q. Each sphere's radius is k F . The volume of the intersection region is Yr.
(b) The shaded region has a volume of Vr j 2 = Vr / 2.
The integral over k is simply the volume Vr of the region in k-space defined by
k < k F and Ik + ql < k f . The expression for E 1 reduces to
EI = -4rre 2 - .-V
(2rr)6
f d 3 q -1 Yr.
q2
(4.18)
4rr 3 [2
Vr / 2 = - k F - - Vcone ·
3 4rr
Here, [2 is the solid angle subtended at the center by the shaded region in Figure
e
4.2b, and Vcone is the volume of a cone with a half angle = COS-I (q / 2k F ) at the
vertex and a height h = q / 2. They are given by
t' (lr
[2= l o sine'de ' l o d¢' =2rr(l-q / 2k F ) , Vcone = 6rrq 2 2
(kF - q / 4).
(4.19)
The integration over q in Eq. (4.18) is now calTied out, noting that q varies from 0
to 2k F ; for q > 2k F the spheres do not intersect. We find
4.3 Ground state energy 73
3e 2
EI / N = --kF '
4n
Using the definition of rs , and that e2 / 2ao = I Ry, we obtain
E1 / N = - -
3 (9n)I
-
/3-Ry=:::---Ry.
1 0.916
2n 4 rs rs
Hence, to first order in the perturbation
2.21 0.916
E / N=::: - - - - Ry. (4.20)
r; rs
The first term is the kinetic energy per electron, while the second term is known
as the exchange energy per electron. The name is acquired because this term arises
in the evaluation of the matrix element by having the creation operator C~' _qa'
restore the electron annihilated by Cka , while c~+qa restores the electron ann ihil ated
by Ck'a " The exchange term is attractive, which may seem odd since it arises from
the Coulomb interaction . The explanation for this situation is that the term arises
when a = a ' . Electrons in the same spin state cannot be located at the same point
in space (Pauli exclusion principle), and they tend to stay away from each other,
hence reducing the repulsive Coulomb interaction. This effect, which is quantum
mechanical in nature, is not taken into account in the classical expression of the
Coulomb interaction. In a way, the exchange term represents a quantum correction
to an otherwise overestimated classical Coulomb repulsion.
The prime indicates that the sum is over all intermediate states 1m) =I- I F ). Consider
(m IVe I F ) . Ve annihilates two electrons in states Ika) and Ik' a ' ) below the Fermi
surface and creates two electrons in states Ik + qa ) and Ik' - qa'); hence, if 1m) =l-
IF ), the two created electrons must Iie outside the Fermi sphere, as shown in Figure
4.3. If the two electrons simply interchange their states, so that Ik + qa ) = Ik'a ' )
and Ik' - qa ') = Ika ), then 1m ) and IF ) will be the same. Therefore, (mlVe IF )
is nonzero if kF > k. kF > k', kF < Ik + ql, and kF < Ik' - ql. Considering the
other matrix element, (FIVe 1m), we see that Ve must restore 1m ) to IF ). This can
be done by either a direct or an exchange process (see Figure 4.4). The contribution
74 The electron gas
(a) (b)
Figure -1.3 (a) Ground state IF ) corresponding to a filled Fermi sphere Fs .
(b) The action of Vc on I F ): two electrons are annihilated from inside Fs, and
two electrons are created outside Fs .
(a) (b)
Figure 4.4 (a) A direct process: each electron recombines with the hole it left
behind. (b) An exchange process: each electron recombines with the hole left
behind by the other electron.
Evaluation of the RHS of Eg. (4.22) is not easy, but we can see from the following
argument that it is divergent.
E lF) - E IIII ) = ~
2m
[k2 + k'2 - (k + q)2 - (k' - qiJ
/i 2
= - [2(k' - k).q - 2 q 2J = O(q) as q -7 O.
2m
4.3 Ground state energy 75
Figure 4.5 The region of integration over k in Eg. (4.22) is the shaded region. The
centers of the spheres are separated by - q and each has radius k F.
Similarly, the volume of the region of integration over k' can be calculated; again
we find
1= r d k r d k'
3
irk irk'
3 1
E lF) - E III1 )
=~ r
n irk
d k r d 3e
3
irk' q [k~ -
1
k~ - q]
.
the numerator arises from replacement of the summations over k. k', and q by
integration, while V 2 in the denominator results from Vc squared. We note, for
completeness' sake, that the exchange term, following a similar analysis, turns out
to be finite.
What we have found is that the second-order correction, rather than being smaller
than the first-order correction, as one might expect, is actually divergent. Since the
crystal is stable, this divergence must be eliminated somehow. Our only hope is
to consider higher order terms in the perturbation. However, these also turn out to
be divergent, so a sum of terms to infinite order needs to be calTied out. Green's
function formalism will provide a suitable framework within which to cany this
out.
Finally, we note that the need to sum perturbation terms to infinite order arises
as a mathematical necessity due to their divergence, which in tum results from
the small values of q, or, equivalently, from the long-range nature of the Coulomb
interaction. Physically, one needs to sum perturbation terms to infinite order to
capture the effect of screening (this is discussed in detail in Chapter 10), which
renders the Coulomb interaction shot1-ranged. In empty space, two electrons i and
j would interact by exchanging momentum, but in the presence of a medium, the
interaction can proceed in an infinite number of ways. Electron i may scatter, but
the momentum it transfers could be picked up by an electron below the Fermi
surface, which would then make a transition to a state above the Fermi surface,
leaving behind a hole. The electron and hole could then recombine, transfelTing
momentum to electron j. Alternatively, the electron-hole recombination could lead
to the creation of another electron-hole pair which, upon recombination, would
transfer momentum to electron j. This argument could be canied on at length,
showing that there are infinite ways in which the interaction between two electrons
could proceed . The net effect is that the Coulomb interaction becomes screened.
Further reading
Bruus, H. and Flensberg, K. (2004). Many-Body Quantum Theory in Condensed Matter
Physics . Oxford: Oxford University Press.
Fetter, A.L. and Walecka, 1.0. (1971). Quantum Theory (J{ Many-Particle Systems. New
York: McGraw-Hill.
Kittel , C. (1963). Quantum Theory of Solids. New York: Wiley.
Problems
4.1 Constrained ground state. In the ground state of a noninteracting electron gas
at T = 0, there are N /2 spin-up electrons and N /2 spin-down electrons; the
ground state is unpolatized (Sz = 0). A spin polarized state has Nt spin-up
Problems 77
This is the probability amplitude for the state la) = \lIa(r')I<po), in which a
particle at r' is removed from the system in the ground state, to be found in
state 1,8) = \lIa(r)l<po). 1,8) is the state obtained by removing a particle, with
coordinates (ra), from the ground state. Obtain an expression for GaCr, r') in
terms of x = Ir - r'l.
4.3 Pair correlation function. For a noninteracting electron gas in the ground state
I<po), evaluate (<Po I\lIJ Cr)\lI ~ , Cr')\lIa ' Cr')\lIaCr)l<Po) in terms of x = Ir - r'l for
a i- a ' and for a = a'. What physical conclusion can you draw from your
answer?
4.4 Coulomb interaction in two dimensions. Show that, in two dimensions, the
Fourier transform of the Coulomb potential is 2n e 2 / q.
10
Hint: Jo(x) = 2~ 2]'( e-ixcose de.
4.5 Exchange energy in two dimensions. In three dimensions we found that the
exchange energy per electron is - 3e 2 k F /4n. Show that, in two dimensions,
the exchange energy per electron is -4e 2 k F /3n.
5
A brief review of statistical mechanics
Since we will be dealing with systems at finite temperatures, we will need concepts
that have been developed in the context of statistical mechanics. We devote this
chapter to a brief review of the basic elements of statistical mechanics.
78
5.2 Contact between statistics and thermodynamics 79
Figure 5.1 An isolated system A consists of two subsystems , A 1 and A2, that are
in thermal contact. Neither E1 (the energy of AI) nor E2 (the energy of A 2) is
constant, but E 1 + E2 is constant.
(5.1)
where
i = 1,2 . (5.2)
The parameters N i and Vi, being held constant, are now written explicitly.
From thermodynamics, we know that at equiliblium the temperatures of both
subsystems are equal
(5.3)
I aSI (5.4)
T = aE N.V·
Comparing Eqs (5.1) and (5.3), and (5.2) and (5.4), we are tempted to identify f3
with 1/ T. However, f3 has units of IIEnergy ; hence, we write
f3 = 1/ kT , S = k In Q (5.5)
d(lnQ)= -In-
Q a I dE+ -In-
Q a I dV+ -In-
Q a I dN . (5.6)
aE N.V av N.E aN V.E
1 E
d S = -d + -as I dV + -as I dN
T av N.E aN V.E
dE = T d S - P d V + f.1-d N , (5.8)
The expressions for the thermodynamic quantities T, S, P, and f.1-, in terms of S1,
establish the connection between thermodynamics and statistics.
5.3 Ensembles
Given a system in a specific macrostate, what is the probability of finding the
system in a particular microstate 11J.r;)? Imagine a large collection or ensemble of
systems, N ens , with all systems in the same macrostate. We perform measurements
on each system in the ensemble to determine its microstate. If N; systems were
found to be in state 11J.r;), we would say that the probability of the system being in
state 11J.r;) is N; / N ens .
The information available about a system determines its macrostate. For exam-
ple, a system may be isolated, in which case N, V, and E are fixed. A system may
be in contact with a heat reservoir, in which case its temperature T is fixed, but
only its mean energy is fixed. Whatever the constraints are, an ensemble is con-
structed such that all of its members are under the same physical conditions as the
system of interest. Since one may imagine systems under various constraints, we
are led to consider different kinds of ensembles. Below we discuss three important
ensembles.
PII -- C n
~'R
(E 0 - E II ) -- Ce1nSlR ( Eo- E,,)
.
Since Ell « Eo, we neglect terms of order higher than En . Using Eq. (5.2), the
above equation may be written as
Z = Le- f3En
. (5.12)
II
F::!:::. E - TS (5 .13)
where E is the average energy of the system and S is its entropy. Since its derivation
is too lengthy for a blief review such as this , we shall state without proof the
following result, which establishes the connection between F and Z,
PI'S = CQR(Eo - E r , No - N s )·
where
(5.16)
rs
(5.17)
rs
" _ " e - f3 (E,- J1Ns J(r, slAlr, s) " (r, sle - f3 (H-J1 N)Alr, s)
(A)=~Pr5 A=~ =~ .
rs rs Ze rs Ze
(5.18)
we can write
Tr [e - f3 (H-J1 N) A ]
(A) = Tr [e- f3 (H-J1 N)] = Tr(Pe A ). (5.19)
(5.20)
Notice that two types of averages are involved in writing the ensemble average.
(1/1;1 A 11/Ii) is the usual quantum mechanical expectation value of A in state 11/Ii).
The above equation also tell s us that the quantum mechanical averages must be
further weighted by the corresponding fractional occupation of the state 11/Ii ); this
second averaging is classical in nature.
Given a complete set 11),12), ... of single-particle states, we introduce two
resolutions of identity ( L n In)(111 = 1) into Eq. (5.20),
1/111 nm
In this form , the dependence of the average on the ensemble is factored out. We
define the stati stical , or density, operator as follows:
(5.21)
Since the trace is independent of the basis set of single-particle states, Tr[pA] may
be evaluated in any convenient basis set.
86 A brief' review of statistical mechanics
(I) The statistical operator p is Hermitian; this follows from its definition:
= LPi = 1.
How does p change with time? Pi is the probability of finding the microstate 11/Ii)
in the ensemble. If the ensemble is left undisturbed , Pi cannot change with time.
The evolution of p with time is thus completely governed by the time evolution of
the states 11/Ii),
iT? ap
at = L
. Pi H I1/Ii(t))(1/Ii(t)I- L
. P;l1/Ii(t)) (1/Ii( t)IH.
I I
= nL 00
i=1 11;
e- fJ (E,- Il )l1 i . (5.26)
(5.27)
At this point, distinction is made between bosons and fermions. For bosons, ni is
unrestricted; it can vary from 0 to 00. Equation (5.26) gives
ZB =
G
n
00
i=l
1
1 _ e- fJ (E,-Il)
=}
G L
00
In ZB = - "In [1 - e- fJ (Ei-Il) ]
i=l
. (5.28)
l~--------------~-'
Figure 5.3 The Fermi-Dirac distribution function. The dashed line corresponds
to the case of zero temperature.
where nfE, the Bose-Einstein quantum distribution function, is the average number
of particles in the single-particle state I<Pi ). It is given by
1
n .BE = -::-:-----:--- bosons. (5 .29)
I e{3(Ei-fl.) - 1
Z~ = n
00
i= ]
[1 + e- {3(Ei - fl.) ] :::} In Z~ = LIn [1 + e -{3(Ei- fl. ) ] . (5.30)
where 1/D, the Fermi-Dirac distribution function , is the average occupation num-
FD 1 fermions. (5.31 )
Ii = e {3(Ei - ,L) +1
Unless confusion may arise, we write I/D simply as Ii. The Fermi-Dirac distribu-
tion function is depicted in Figure 5.3. In particular, for a system of noninteracting
particles whose Hamiltonian is H = Lku Eku ctu Cku, the occupation number of
the single-pmticle state Ika ) is given by
bosons
(5.32)
fermions .
Problems 89
Further reading
Huang, K. (2001 ). Introdu ction to Statistical Physics. London: Taylor and Francis.
Pathria, R.K. ( 1996). Statistical Mechanics, 2nd edn. Oxford: Butterworth-Heinemann .
Reif, F. (J 965) . Fundamentals a/Statisti cal and Th ermal Physics. New York: McGraw-Hill.
Problems
S. l Stirling'sformulafor N!. Starting from
N!= 1 00 e - lt Ndt ,
-IN
f(x)dx ,
exp (_ x
2
2
+ x;" _ ... ).
3v N
For N »
1, keep only the leading order in the expansion . For x < -..,fN,
x2 2
e- / is exceedingly small , and the lower limit of integration may be pushed
to -00. Show that
lnN!~Nln-N.
is also the number of interstitials). Also assume that Nand M are much, much
greater than 1. Determine (a) the system's entropy and (b) E as a function of
temperature T .
5.3 Magnetic susceptibility. Consider a crystal where each atom has spin 1/ 2 and
magnetic moment fJ,. The number of atoms per unit volume is n. A magnetic
field B is applied. Choose one atom as the small system and treat the rest of
the crystal as a heat reservoir at temperature T .
(a) Show that the magnetization (the mean magnetic moment per unit volume)
of the crystal is M = n fJ, tan h(fJ,B / k T).
(b) For the case fJ,B / kT « I, show that the magnetic susceptibility, defined
as X = aM / aB, varies as ]/ T (Curie's law).
5.6 Ising model in one dimension. The one-dimensional Ising model describes
local ized particles on a line, where each particle carries a spin s. The
Hamiltonian is
N N
H = -1 LSisi +1 - h LSi
;= 1 i=1
91
92 Real-lime Green 's and correlalion jun ctions
(6.1)
where
R= H - fJvN, (6.2)
N is the number of particles operator, and fJv is the chemical potential. The standard
Heisenberg picture is obtained if R ~ H.
(- .. ) = zeITr[e-fJ H . .. J.
Here, f3 = 1/ kT (k is Boltzmann's constant), Ze = Tr[e -fJ H] is the grand canoni-
cal partition function, and Tr stands for the trace. In Eq. (6.3), T is the time-ordering
operator, sometimes written as ~ in order to distinguish it from the temperature.
Acting on a product of operators, T orders them in increasing time order from
light to left, and, in the process, introduces a minus sign every time two fermion
I
operators are interchanged. Thus,
The lower (upper) sign refers to the case when A and Bare fennion (boson)
operators. Note that the operator A(e, e t ) is bosonic if e and e t are boson operators,
or if it is of even order in e and c t; e.g., if c and e t are fermion operators, an operator
such as e t e is considered a boson operator.
6.1 A plethora offunctions 93
0 t < t'
e(t - t') = (6.6)
11 t > t'.
C~B(t, t ' ) is nonzero only if t > t'; hence the name "retarded." For the case when
A and B are bosonic operators,
The lower (upper) sign refers to fermions (bosons). Similarly, we define the
advanced correlation function, which is nonvanishing only if t < t', as follows:
As we will see later, athough the different functions introduced above have different
analytic properties, they are in fact closely related . Finally, we define one more
correlation function , without a label,
C~B = -i8(t - t ') Z e I {Tr [e- /l H ei Hr /Ii Ae - i H(r- r'l/Ii Be- iHr ' /Ii ]
In the second term on the RHS the lower (upper) sign refers to fermions (bosons).
The time-independence of fI makes possible the replacement of e i HI Iii e- i HI ' Iii
with e i H(I- I')/Ii. The trace is invariant under cyclic permutations: Tr[AB ... CD] =
Tr[ DAB ... C]; hence, we can move e - i HI ' Iii in the firsttelm (e -i HI Iii in the second
term) to the leftmost position, and using the fact that e- i H I ' Iii (or e- i HI Iii) commutes
with e-f3 H , we can write
C: B
= -ie(t - t l)Ze l {Tr [e - f3 H eiH(t- I')/1i Ae-iH (I-I')11i BJ
GR(wt , rl(J lt ' ) = -ie(t - t l)([\I!a(r t), \I!; ,(rl t l )1,:) (retarded) (6.12)
GA(wt , r l(J lt ' ) = ie(t l - t)([\I!a(rt) , \I!;, (r l t l)] 'f) (advanced). (6.13)
The lower (upper) sign refers to fermions (bosons). At this point we introduce two
other single-particle functions that play an important role in the study of transport.
The greater and lesser functions are defined by
(lesser). (6.15)
We note in passing that the ensemble average of the local particle number density
can be expressed in terms of the lesser function ,
(6.16)
it is simply iG > (wr, r ' a ' t ' ). The above definitions can be generalized: for any
complete set 1<p1 ) . 1<p2 ), ... of single-particle states, Green's functions may be
defined in terms of the corresponding annihilation and creation operators. The
retarded Green 's function, in the <pv-representation, is defined by
GR(vt, v' t ' ) = -ieCt - t ' )([cv(t), c~, (t ')] 'f ) '
Similar definitions can be made for the causal, advanced, greater, lesser, and corre-
lation functions. For example, consider an atom with single-particle states I<P/ltma ),
where 11, [, m, and a are, respectively, the principal, orbital, magnetic, and spin
quantum numbers. We may define a retarded Green's function for electrons in this
atom as
GR(nlmat, n' ['m'a ' t ' ) = -ie(t - t ' ) ({c/I!ma(t) , C~'I'III'a, (t ')} ) .
iG(rat , r ' a ' t ' ) = (w a (rt)w; ,(r' t') ) = ZelTr [e -.8 H Wa (rt)w; ,(r' t')J .
Given a complete set of states In ),
iG(rat , r ' a ' t') = Ze l L e -.8E"(nleiHI /liwa(r) e- iH (I- I') /nw; ,(r')e - iHI '/liln).
11
(6.18)
Defining the states la ) and 1,8 ) by
la ) = e-i H(t-t') /liw; ,(r' )e- ifil '/liln ) , 1,8 ) = w~ (r)e-iHI /liln) ,
it is the probability amplitude for a system in state la ) to be found in state 1,8). Let
us look more closely at state la ) . Starting from state In ) at t = 0, e- iH1 '/1i In ) is the
state after it has evolved to time r'. At this time, w;,
(r') injects a particle with spin
projection a ' into the system at position r ' , and the operator e- i H(t- t')/Ii canies the
system to time t. Thus, la ) is the state of the system at time t if a palticle with
coordinates (r'a ') was added to it at an earlier time t' . Similarly, 1,8 ) is the state of
the system when it has an extra particle with coordinates (ra), added at time t. The
96 Real-time Green's and correlation fun ctions
matrix element (nl ... In) is thus the probability amplitude of finding the system
with an extra particle with coordinates (m) at time t if a particle with coordinates
(r' a ' ) was injected at an earlier time r'. Loosely speaking, it is the probability
amplitude for an added particle to propagate from (r' a ' t ' ) to (mt), though we
should be careful to note that, since the particles are indistinguishable, it is not
meaningful to think of the particle with coordinates (m) as the same particle with
spin projection 0- ' that was injected earlier at r'. Since different states In) occur
with probabilities Z e ie -fJ En, iG(mt , r ' a ' t ' ) represents the ensemble average of
the aforementioned propagation amplitude. For t < t' , iG(rat, r'o-'t' ) represents
the ensemble average of the propagation amplitude of a hole from (m t) to (r' 0-' t' ).
Similar meanings can be attached to the other Green 's functions. The definition of
iG(mt , r 'a 't') is depicted pictorially in Figure 6. 1.
The above discussion indicates that single-particle Green 's functions are impor-
tant tools in analyzing experiments where particles are added to or removed from a
system . Examples include experiments that involve tunneling of electrons between
two systems, at different chemical potentials, which are placed in contact with
each other; and optical experiments whereby a photon removes an electron from a
solid. Another example is the introduction of magnetic impurities into a host metal;
tunneling takes place there between localized states on the impurity sites and the
delocalized host states. We will explore some applications of Green's functions in
the next chapter.
e R( ra t. rat
I I ') ~
= Oaa' eR( ra t, rat
I I ')
.
The same conclusion applies to all the con-elation functions that we have introduced.
e
We also note that, since fI is assumed to be time-independent, R , being a special
case of the more general retarded correlation functions, depends on t - t', and not
on t and t ' separately, as was shown in the preceding section. This being the case,
e
we may set t ' = 0, and consider R to be a function of t.
Generally, we will be dealing with translation ally invariant systems, where any
function fer. r ') of the positions of two particles (e.g., the interaction energy
between two particles at positions rand r ') does not change if both rand r' are
shifted simultaneously by any vector R. It follows that fer , r ') depends only on
r - r', and not on rand r' separately. In particular, the single-particle Green's
functions are functions of r - r'. The proof of this statement is the subject of
Problem 6.2. Hence, in a translationally invariant system, with a time- and spin-
independent Hamiltonian, we write the retarded Green's function as
(6.19)
(see Section 3.10. The operator c L (Cka ) creates (annihilates) a particle in state
Ika ). Inserting the above relations into Eg. (6.19), we obtain
e
Since R depends only on r - r', it does not change if rand r ' are shifted simul-
taneously to r + Rand r ' + R, for any vector R; hence,
eR(r - r ia , t) = -i8(t) ~L eik.( r + R-r'-R)ei (k-k').( r'+ R)( [Cka (t) , ct'a(O)] =r=) '
kk '
98 Real-time Green 's and correlation fun ctions
We have introduced GR(ko-, t), the spatial Fourier transform of GR(r - r io-, t),
(6.22)
Similarly,
and
(6.27)
The expression on the RHS forms a spectral representation of H, in the sense that H
is written in terms of its spectrum of energy levels and eigenstates. We will follow
a similar procedure in deriving the spectral representation of Green's functions: we
shall introduce resolutions of identity and express the functions in terms of the exact
energy spectrum and eigenstates of the system. An answer to the second question
will unfold in later chapters. For now, it suffices to note that merely expressing
Green's function in terms of the exact eigenstates of the system is not, in and of
6.4 Spectral representation 99
itself, a worthwhile goal. After all, the exact eigenstates are not known; if they were,
the problem would be completely solved. At finite temperature, real-time Green's
(or correlation) functions are not amenable to a treatment of interacting systems
by means of pe11urbation theory. The burden will fall upon the imaginary-time
Green's function, which will be discussed in Chapter 8. Nevertheless, experiments
are carried out in real time, and their interpretation requires knowledge of real-
time Green's functions. How do we find the real-time Green's function if we
know the imaginary-time function? By studying the spectral representations of
these functions, we will arrive at a simple method for obtaining the real-time
Green's function from its imaginary-time counterpart. With this in mind, let us
next proceed to determine the spectral representations of GR(kCT, t), C(kCT, t), and,
more generally, C:B(t) and CAB(r).
(6.28)
The lower (upper) sign refers to fermions (bosons). Consider the first term,
(6.29)
where Cka = Cka(r = 0) and cta = cta Ct = 0). The trace of an operator is the sum
of its diagonal elements,
11
11111
11m
00 . dE
= - P(kCT.E)e- IEI -. (6.30)
/ - 00 27r
P(kCT, E) = _27rZ (; 1 I>-fi t " \(m lcta(0)ln )\ 2 8 (E - (E III - EII) /h) (6.31)
11m
]00 Real-time Green's and co rrelation junctions
Using the invariance property of the trace under cyclic permutations, we first move
e- i HI Iii to the leftmost position, then move Cka (0) to the leftmost position, and
finally move eiifl ln to the leftmost position; the result is
(ct a(O)Cka(t) ) = - l °Op (ka , E)e -if(r-i,Bn) 2dE = _ l OOe -,Bnf P(ka. E)e- iff 2dE .
-00 JT -00 JT
(6.32)
The express ion for B in Eq. (6.28) is thus obtained. The Fourier transform of
i:
CR(ka, t) is given by
iwI
CR(ka , w) = e CR(ka , t )dt. (6.33)
= i 1 00
-00
P(ka , E)(l =f e-,BTiE) dE
2JTh
r ooei (W-E)1 dt.
1 00
ei(W-E+iT)1
00
1
00 . . . 1
e'(W-E)1 dt = lim e' (W-E+I1/)1dt = lim
o '7~O+ 0 '7~O+ i(w - E + iT)) 0
-1
= lim (6.34)
I)~O+ i(w - E + iT))
6.4 Spectral representation 101
11m
li E
)o (E - (EIIl - EI/) /h) , (6.35)
A(ka, E) dE
1
00
CR(ka, w) = lim . -. (6.36)
7)--> 0 + -00 W - E + IT) 2Jr
A similar derivation yields the spectral representation of the advanced Green's
function CA(ka, w),
A(ka, E) dE
1
00
CA(ka, w) = lim . -. (6.37)
7)-> 0 '" -00 W - E - 1 I} 2Jr
The derivation of the above result, as well as the spectral representation of the
causal Green's function, is relegated to the Problems section.
C(ka, w) =
1 00
- 00
e
,. wt
C(ka, t)dt =- 1 00
- 00
P(ka, E) -dE
2Jr
1-00
00
i:
The integral over t is straightforward (see Eg. [1.15])
x
1.
± 10+
= P (~)
x
=t= iJro(x) (6.41)
102 Rea/-time Green's and correlation junctions
where P(l / x) is the principal value of l / x, and noting that A(ku, E) is real,
Eq. (6.36) gives
CR(ku, w ) = _I
2n
f oo A(ku, E) [p (_1_)
-00 w -
- ino(w - E
E)] dE
1(l - f w) fermions
(6.43)
where !tv and nw are the Fermi-Dirac and Bose-Einstein distribution functions,
respectively, for the case when energy is measured from the chemical potential,
i.e., when fl is set equal to zero,
I
f w = ef3Tiw +I ' nw = ef3nw - I .
(6.44)
0 +n w) bosons
C(ku, w) = -21m CR(ku, w) (6.45)
1
(1 - i w) fermions.
t
(CkaCka) -
_
-
f oo P(ku , E) e- f3Ti E 2dE -_ f OO A(ku,
1
E)e -
f3li E
-f3Ti E 2
dE
-00 n - 00 =t= e n
=
OO A(ku, E) -dE = f OO -A(ku
dE
, E)
lnE bosons
f-00 ef3li E =t= 1 2n -00 2n i E fermions.
bosons
(6.46)
fermions.
This equation provides a method for calculating the number of particles in a given
state once the retarded Green 's function has been found .
6.4 Spectral representation ]03
The lower (upper) sign refers to the case where A and B are fermion (boson)
operators. Expanding the anticommutator/commutator, we can write
= -ie(t)Ze ITr [e - t!/{ eiHt /1i Ae - iHt /T! B =f e- fJ H BeiHt /T! Ae- iHt /n]
= -ie(t)Ze l L e- fJE " [e;t"I/T! (nIAe - iH1 /1i Bln)=fe - iE"I/n(nIBei{ll /nAln)]
II
nm
i:
We now take the Fourier transform,
C~B(CU) = eiwt
C~B(t)dt
= -iZ eL l
11m
(nIAlm) (mIBln) (e - fJE " =f e-fJ Em) l °Oei(nw+ E,, -tm)l /hdt
0
This is the spectral representation of the retarded correlation function. Notice that
all the poles lie below the real cu-axis; the retarded function is analytic in the upper
half of the complex cu-plane. Of course, the same conclusion applies to the retarded
single-particle Green's function GR(ka, cu), since it is a special case of the more
general retarded correlation function.
We can go a step further and express the spectral representation in a form similar
to Eg. (6.36). Define the spectral density function by
SeE) = 2nZei Le-P E,'( nIAlm )(mIBln ) (1 =f e- fJT! E) 8 (E - (Em - Ell) In).
11 m
104 Real-time Green's and correlation junctions
C R (w) _ / 00 SeE)
+ '0+ 2Jr .
dE
AB - -00 WEI
_
CAB(t) = (A(t)B(O)) = Ze l
I:>-.B t " ei t "I/1i (nIAe- iH1 /1i Bin)
II
11m
The function C:B(w) is given in Eq. (6.47). Using Eq. (6.41), we find
x 8 (w
- (Em - Ell) In)
= -Jrz l Le- .B t " (nIAlm)(mIBln)
e (I =fe-.Bliw) 8 (W- (EIIl-EII)/fz).
nm
In the last step, we replaced the exponent (Em - EI/) /n with W (this is made possible
by the presence of the delta function). Comparing CAB(W) with 1m C:B(w), we
obtain
value. (A) does not depend on time, since the Hamiltonian is assumed to be time-
independent. The correlation function CAA(t) = (A(t)A(O)} = (A(t)A(O) } - (A}2
describes the quantum thermal fluctuations in the operator A. On the other hand,
as we shall see later in this chapter, CJA Ct ) describes the response of the system
to an external field; its imaginary part is usually related to energy dissipation.
For example, an external electromagnetic field couples to the CU1Tent density j.
In this case A = j . Whereas CAACt) describes the quantum thermal fluctuations
in the current density, the imaginary part of C JA(t) turns out to be related to the
resistance in the system, and hence to the mode of dissipation of energy supplied
to the system by the external field. Thus, it is usually the case that the LHS of
Eq . (6.49) represents fluctutations, while the RHS describes dissipation.
To see more explicitly that the imaginary part of the retarded correlation func-
tion describes dissipation, consider an applied external fleld that couples to some
observable of the system. We take the perturbation to be
where I is proportional to the strength of the applied field , and A is the operator
that represents the observable of the system (such as the current density) to which
the field is coupled. Since A is hermitian (A t = A), it must have an equal number
of creation and annihilation operators when it is expressed in second quantized
form; hence, it is a bosonic operator. The transition rate (transition probability
per unit time) from stationary state In } to stationary state 1m } (eigenstates of the
unperturbed Hamiltonian) is
W,, --> m =
27T
-III 2
l(mIAln }12 [8(Em
-
-
-
E" -l1w)
-
+ 8(EI/I -
- E" +l1w)
J (6.51)
h
(see Problem 1.13). Assuming that the system is a member of a grand canonical
ensemble, the energy absorbed by the system per unit time (the power delivered by
the field to the system) is given by
27T
= -III 2
I1wZe-I "L' " e - f3 E" IA "111 I"[8(Em
--- - +hw) J.
- - E"
Ell -l1w) - 8(E/I1
11 nm
where Amll = (m IA In). Interchanging nand m in the second summation, and noting
that 8( -ax) = 8(ax) = (l / laI)8(x) , we obtain
(6 .52)
106 Real-time Green's and correlation jltnctions
The power (energy per unit time) dissipated in the system is thus given by
(6.53)
We note that it is indeed proportional to the imaginary part of the retarded correla-
tion function.
ka ka
where i\a is the single-particle state energy relative to the chemical potential.
Below, we calculate CR.o(ker, w) using two different methods.
For (m leta In) to be nonzero, 1m) must differ from In) by an extra particle in state
Iker ). Since the system is noninteracting, Em - E" = Eka, and
/1111
11
(6.54)
6.5 Example: Green's function of a noninteracting system 107
The definition of the grand canonical ensemble average is used in the last step.
From the commutation property of the c-operators, we can write
bosons
fermions.
nka = (ef3 E
ka 1) - 1 and Aa = (ef3 Ek" + 1) - 1 are the Bose-Einstein and Felmi-
-
bosons
fermions.
The Dirac-delta function has the property: 8(x - a)f(x) = 8(x - a)f(a) for any
function f(x). The factor (1 =f e- f3Ii E) in the above expression is thus replaced with
(l =f e- f3Ek" ). It is then straightforward to show that
This is the spectral density function for noninteracting particles. Inserting this into
Eq. (6.36) gives the noninteracting retarded Green's function
I
GR(ka, w) = -
w - Eka Iii + l'0+ (6.56)
We note that the poles of G R (ka, w) occur at the excitation energies of the system.
In the presence of interactions, the spectral density function will no longer be a
delta function; instead, the sharp peak representing the delta function will broaden,
yielding information about the energies of the excited states and their lifetimes.
dA(t)
_ _ = -d ( e'' H- l"Ae
/' i -
-'' H- I"/ , ) = -HA(t) + e'' H- /'
I" -
aA e- '' H- I / rJ' -
i
-A(t)H-
dt dt n at n
= -i [ H.
- A(t) ] aA
+ -(t) .
n at
We have assumed that fJ is time-independent, and used the fact that fJ commutes
with e- iH1 / 1i • The last term in the above equation is a Heisenberg operator. If
A = Cka, we find
d
- Cka (t) = -i [ H- , Cka (t) ] .
dt n
108 Real-time Green's and correlation functions
It follows that
~C
d t ka
(t) = ~ [8 e iHI /1i C e- iHI / Ii ] = ~eiHI /1i [8 C ] e- iHI / Ii .
Pt' ka Pt ' ka
(6.57)
Using the relation [AB , C] = A{B , C} - {A, C}B, or, [AB , C] = A[B, C] +
[A , C]B, the above commutator gives, for felmions,
It follows that
[8, Cka] = - L
kIa '
Ek'a' Okk' Oaa ' Ck'a' = -EkaCka' (6.58)
(6.59)
(6.60a)
(6.61)
6.6 Linear response theory 109
F(r, t) is a "generalized force ." For example, the scalar potential <p(r, t) of the
electromagnetic field couples to the number density of electrons nCr) ,
The generalized force in this case is -e<p(r , t) and A(r) = nCr). The external
perturbation drives the system out of its unperturbed state, leading to a measurable
effect : the ensemble average (A ) shifts to a new value (A) exl' For example, in
the absence of the scalar potential. (n(r, t) ) is constant, but it will no longer be
constant once <p(r , t) is turned on , or a cutTent (nonexistent in an isolated metal)
begins to flow upon the application of an external voltage. The experiment measures
8 (A ) = (A) exl - (A ) as a function of the generalized force F. From a theoretical
110 Real-time Green's and correlation junctions
perspective, we can say that, if F is weak, the response of the system, 8(A), will
be, to a good approximation, linear in F,
For t > to, the state evolves according to the Schrodinger equation
The last equality is obtained since IW(O)) is arbitrary. Multiplying both sides by
ei Hrjli on the left, we obtain
(6.67)
6.6 Linear response theory 111
(6.68)
We now consider the response of the system to the external perturbation. In pm1ic-
ular, we are interested in the effect of H ext on the expectation value of an operator
A that represents an observable of the system, such as its charge or cun-ent density.
Let the eigenstates of H (interacting, but unperturbed Hamiltonian) and the number
operator N be denoted by In ),
The states In ) are time-independent; they may be considered the stationary states
at t = 0, and they evolve in time, in the absence of H ext, as
In , t ) = e- iHf /liln ).
In the presence of H ex!, on the other hand, the state evolves according to
Eq. (6.65); the expectation value of A at time t > to is
If H ex! is weak, we are justified in ignoring higher-order terms in H ex! and keeping
only the first-order term. The first term on the RHS is just the expectation value of
A in the absence of H ex l ; hence, the change in the expectation value of A, brought
about by H exl , is given by
For f < fo , the system is in equilibrium and state In ) is occupied with probability
Pn = Ze i e- f3 (E n -J.lN). We thus consider the change 8(A) in the ensemble average
of A, caused by H ex1. Taking the ensemble average on both sides of Eq. (6.72), we
obtain, for f > to,
On the RHS of the above equation, the ensemble average is taken over the inter-
acting, but unperturbed, system. One may object to this because Pn may change as
a result of the perturbation. However, the above expression for 8 (A) is already first
order in H ex!; any modification brought about by considering a change in PII will
be of higher order. Thus, within a linear respon se theory, where 8(A) is calculated
only to first order in H ext, we are justified in taking the ensemble average over the
unperturbed system. Another way to an-ive at this conclusion is as follows . Before
the external perturbation is turned on at to, the system has been in contact with a
reservoir, with which it exchanges energy and particles, for a sufficiently long time
for equilibrium to be established. If we assume that the time t - fo, during which
the system is observed, is too short in comparison with the equilibration time, the
probabilities of occupation of the states In ) will remain unchanged. Stated differ-
ently, the process of measuring of the system's response is finished long before
the reservoir is able to cause a repopulation of the states of the system through
exchange of energy and particles.
6.6 Linear response theory 113
For H exl of the form given in Eq. (6.62), the response of the system is
(6.74)
where
(6.75)
Since t > t ', the integration over t ' being from to to t > to, the step function e(t - t ' )
is equal to 1, and its introduction into Eq. (6.74) is a totally innocuous step. Equation
(6.74) is Kubo's formula for the linear response of a system in equilibrium to an
external perturbation (Kubo, 1957). Since the operator A represents an observable,
it commutes with the number operator N (see Problem 3.4). Because H also
commutes with N, we can wlite
AH(r, t) = ei Hl /1i A (r)e - iHl /1i = ei (H-/-LN)l /1i A(r)e- i(H- /l N)l /n
= eiFil /1i A(r)e- iFil /1i = AFI(r , t)
=} DR(rt. r ' t ' ) = -ie(t - t') ( [AFI(r, t), AFI(r' , t ' )J ). (6.76)
1
X(rt. r ' t ' ) = -DR(rt , r ' t ' ). (6.77)
n
Since fJ is time-independent, the retarded correlation function, and hence X, depend
on t - t' , and not on t and t ' separately. Furthermore, if the system is translationally
invariant, X depends on r - r '. Thus,
8(A)(r , t) !
= i l dt ' d 3r' x(r - r', t - t l)F(r' , t ' )
= i: ! dt
l 3
d r ' x(r - r'. t - t' )F(r' . t ' ). (6.78)
Changing the Iimits of integration over t ' is justifiable: for t ' < to, F(r' , t' ) vanishes,
and the value of the integral is unchanged by extending the integration range to
-00; similarly, for t ' > t. X (r - r '. t - t ' ) vanishes due to the factor e(t - t' )
contained in DR (r t. r ' t ' ). Taking the Fourier transform with respect to time, we
114 Real-time Green's and correlation functions
I:
find
iWI
8(A)(r. w) = dte 8(A)(r , t)
= I: I: f dt dt ' d 3 r ' eiW(r- r') x(r - r ', t - t l)e iwr ' F(r/, (').
Noting that
1 1 1 1
00 00 00
l
00 dt dt ' · ·· = dt ' d(t - t ) ... ,
-00 -00 -oo - ex;
we obtain
8(A)(q , w) = f d 3re-
iqr
8(A)(r, w)
f f 3
d r
3
d r' · ·· = f f 3
d r'
3
d x ...
Thus, the system responds at the wave vector and frequency of the extemal field; if
these match the wave vector and frequency of an intrinsic excitation of the system,
a resonance effect occurs and a peak in 8 (A ) is registered.
(6.81)
6.7 Noninteracting electron gas in an external potential 115
Similarly, decomposing 11 H (r, t) and 11 H(r', t ' ) into Fourier components, we find
DR(rt , r ' t ') = -ieCt - t') ~ "eiqr ei q' r ' ( [nH(q , t) , nH(q/, t l) ] )
v- ~
qq '
= -ie(t - t l ) :2 L
qq '
ei q.(r - r' )ei( q+ q').r' ([nH(q , t) , nH(q', t ' )]) . (6.85)
Since the RHS must depend on r - r ' and not independently on r/, it follows that
q' = -q. Alternatively, we may argue that if rand r ' are shifted simultaneously
by any vector R , the RHS must remain unchanged since it depends only on r - r '.
However, such a shift brings a factor of ei(q+q' ). R into Eq. (6.85); this factor should
be equal to 1 for any vector R, and we conclude that q' = -q. Removing the
summation over q', replacing ql with -q, and comparing Eq. (6.84) with Eq. (6.85),
we obtain
(6.86)
t e- i HI / n eiHI /n e
"~ eiHI /n e ka i HI /n - t (t) e
-- k+ qa e- - "~ e ka k+ qa (t) . (6.87)
ka ka
Up to thi s point, our treatment applies to an interacting electron gas. In the simpler
case of a noninteracting electron gas, eka(t) and eta (t) are given by Eq. (6.60), and
ka k'a'
t Ck+ qa
[c ka ,
t Ck'- qa ' ]_(
ck'a' -
t
cka Ck' -qa' - ck'a' )r r
t Ck. + qa Oaa' Ok+q.k' ·
ka
(6.90)
According to our general result, Eg. (6.80), the response of the system, 8(n)(q, w),
is given by
°
X (q,
w - ~DR.O
) - Jj1
W _ _1_ '\'
(q,) - Jj1 V Lka
A a - A +qa
+ (-Eka - Ek+
- qa )/ft1 + l'0+ (6.92)
W
is the polarizability of the noninteracting electron gas. The function on the RHS of
Eg. (6.92) is known as the Lindhard function (Lindhard, 1954). Equation (6.91) is
also valid for an interacting electron gas if D R.O ---+ DR.
6.9 Paramagnetic susceptibility of a noninteracting electron gas 117
The total potential <PlOt is the sum of the external and induced potentials,
The induced potential results from the induced charge density Pind = -eo (n); it is
given by
'" ( )= J
'f'ind r, t c.13 r , Pinder', t)
Ir - r'l
(cgs).
Multiplying both sides by e2 , and noting that e2 j Ir - r'l is the Coulomb interaction,
which we can expand in a Fourier series, we obtain,
e2",
'f'ind( r , t ) = J q
I ~
c./ 3r ' Pind ( r ' , t ) V '""' vqe iq. (r-r') = V
1 '""'
q
~ vqe iq.r Pind ( q, t )
where Vq = 4JT e 2 j q2 is the Fourier transform of the Coulomb potential. The above
expression implies that
?
2 e- R
e <Pind(q , w) = VqPind(q, W) = -evqo(n)(q, W) = -vqD (q, W)<Pext(q, W).
n
In the last step, Eq. (6.91) was used. The total potential is thus given by
(6.94)
electrons and consider only the interaction of the magnetic field with the electrons'
spins.
The magnetic moment of an electron is /L = -(gej2me)8 ::::= -(ej me)8, where
g ::::= 2 is the gyromagnetic factor, -e is the charge of the electron, m is its mass, 8
is its spin, and e is the speed of light (in SI units, /L ::::= -(ej m )8). Since the energy
of a magnetic dipole in a magnetic field is - /L.B , the external perturbation due to
the applied field is written as
e
mer) = --s(r) (6.96)
me
where s(r) is the spin-density operator. Hext(t) has the standard form:
(6.97)
where A(r) = -mer) and the generalized force F(r, t) = B(r, t). Using Kubo's
formula (see Eg. [6.74]), derived within linear response theory,
Di~(rf , r' f') = -ie(t - t ' )([Si(r, f), Sj er', f l)] ). (6.99)
where Si is the spin operator for electron i. Employing a basis set of plane waves
Ika ), we cast the spin-density operator into second quantized form:
= ~ ~(a lu la 'L...t
2V L...t 1
') ~ fd 3r' e- ik .r' oCr - r')ei k'. r' ckalt c,'
ka l
ala; kk'
(6.101)
seq) =
n2" ~ , t
L...t (aI!ulal )CkaICk+qa; ' (6.102)
kala;
InEq. (6.101), we have replaced the spin operatorS withFiu /2, where ax, a y, andaz
are the Pauli spin matrices. For the non interacting electron gas, c~a(t) = ~~aeifk( ln
and Cka(t) = Ckae-ifkI /Fi. It follows that
(6.103)
(6.104)
Using the completeness property of the spin states ( La 2 la2) (a21 = 1),
We have made use of the invariance of the trace under cyclic permutations:
Tr[ ai a j ] = Tr[ a j ai], and the factthat {ai, a j} = 20ij I, where I is the 2 x 2 identity
matrix. The expression for the retarded function thus reduces to
D~·O(q
I) '
t) = -i8(t)o I·) fz2
2 "(li
~ k
- lik+ q )ei(Ek- Ek+q)l/n .
k
D~'O (q,)
I)
W = 0. £ "~
1)2V +(-Ek
A - A+
-
q
)/1:; + I'0+ ' (6.106)
k W - Ek+ q 11
Hence,
2 2
_ fl e Bi
(q , lik - lik+ q
m , q,w ) -_
~ (.) ( W), ,
u ~ . (6.107)
2m c V 2 2
k flw + Ek - Ek+ q + iO+
The paramagnetic susceptibility is
p B p
Xij(q,W) = o(mi)(q , w) = X (q,w)oij. (6.108)
BB j (q , w)
We thus find
P 2 1 " A - A +q
X (q. w) = - tJ. B V ~ nW + Ek
- + I.0+ (6.109)
ka - Ek + q
where tJ.B = ne/(2mc) is the Bohr magneton, and we have used Lk F(k) =
(1 / 2) Lka
F(k). The Lindhard function has made another appearance, as it often
does in the theory of the electron gas.
The paramagnetic susceptibility can be evaluated at zero temperature for the
case of a static field (w = 0) in the long wave length limit (q --+ 0); X p is then real
6.10 Equation of motion 121
At T = 0, ik = 1 for Ek < 0 (Ek < EF), and ik = 0 for Ek > 0 (Ek > EF); hence
-aiklaE-k = 8(Ek - EF)' Therefore, at T = 0,
where deE F) is the density of states, per unit volume, at the Fermi energy. The
above expression for X is the well-known Pauli paramagnetic susceptibility of
noninteracting, or independent, electrons.
fI = flo + V = L EkaCtacka + V
ka
where V represents the interaction between the particles, or the interaction of the
particles with an external field. The retarded Green 's function is
Recalling that the derivative of the step function is the Dirac-delta function, we can
write
-a Cka(t) = -i[-
H. Cka(t) ] = -i[-
Ho(t), Cka(t) ] + -i [V(t) , Cka(t)].
at n n n
122 Real-time Green 's and correlation fun ctions
(6.111)
where
is a retarded correlation function that describes the effect of the interactions in the
system. To proceed further, we would need to evaluate FR(ka, t) . In general, an
exact solution is not possible; an approximation scheme must be used.
( ata
i--Eka/h ) -1
2n
1-00
00
e-'Wl
. 1
CR(ka , w)dw=-
2n
100
- 00
e - 1.Wl dw.
The integral on the RHS is one of the representations of the Dirac-delta function.
Hence,
1
=} CR(ka, w) = - - - -
w - Eka/h
There is one problem with this expression for CR(ka, w). Suppose that we try to
calculate CR(ka , t) from CR(ka , w) ,
CR(ka, t) = -
1
2n
100
-00
iw1
e- dw .
w - Eka /n
6.12 Example: an atom adsorbed on graphene 123
The integral is problematic because of the pole at Eka In. To circumvent this problem,
we can shift the pole slightly, either above or below the real w-axis. For t < 0, the
integral along the semicircle at infinity in the upper half of the complex w-plane
vanishes because of the e- iwt factor in the integrand. In this case, GR(ka, t) is
equal to the contour integral along the closed contour consisting of the real w-axis
and the semicircle at infinity in the upper half w-plane. Since GR(ka, t) = 0 for
t < 0, the pole should not lie above the real axis, because the residue theorem
would then yield a nonvanishing value for GR(ka, t). Causality thus dictates that
the pole needs to be shifted slightly below the real axis to Eka In - iO+. Hence, the
correct expression for G R (ka, w) is
1
CR(ka, w) =
W -
-
Eka
In + l'0+ '
In the first term, n is a band index which can take two values, 1 and 2, k = (k x , ky)
is a vector in the first Brillouin zone of graphene, and a = tor {,. We assume that
the adsorbed atom has one orbital of energy Ed; because of Coulomb repulsion,
the energy increases by U if this orbital is doubly occupied, as indicated by the
last term in the Hamiltonian. The operator d~ (da ) creates (annihilates) an electron
with spin projection a in the atomic orbital. The third and fourth terms in H
describe the hybridization between the orbital on the adsorbed atom and the n-
states of graphene: electrons can hop from the adsorbed atom onto graphene, and
vice versa. In this example, we assume that U = 0, and we calculate the spectral
density function of the adsorbed atom. For the isolated atom, this is a Dirac-delta
function, peaked at flw = Ed; we will see that interactions broaden the peak into a
Lorentzian.
The retarded Green's function of the adsorbed atom is given by
We have used the relation [AB, C] = A{B. C} - {A , C}B and the commutation
relations of fermion annihilation and creation operators. Thus,
. a R(dda, t) = oCt)
I-C
at + -G
Ed
!?
R(dda, t) + - 1"
!?
~
*
VllkdC R(l1kda, t). (6.113)
11k
.a R Ellk R
I-C (nkda, t) = - C (nkda , t) + -1 Vllkd R
C (dda, t). (6.115)
at !? !?
Upon Fourier decomposition:
CR(dda , t) = 21
rr:
100
-00
e- 'WI
. CR(dda, w)dw, o(t) = - 1
2rr:
1 00
-00
e- i w l dw ,
and
(6.117)
!?
CR(dda. w) = 2
1; " IVllkd i
11W - Ed - L..,
Ilk !?w - Ellk
Furth er reading 125
At this point, an argument is made similar to the one outlined in the previous
section: causality dictates that w ~ w + iQ+; hence,
n
CR(dda , w) = 2
nw + I'0+ - Ed -
" l Vll kd
L.,
I . +
11 k liw - Ellk + 10
li
where P stands for the principal value. To proceed further, we assume that Vllkd is
small except for k-points in the first Brillouin zone near K and K ', where it takes
the constant value V. Under this assumption,
Il k Ilk
Da(nw) is the density of states, per spin, in graphene (see Problem 2.4). The spectral
density function , A(dda, w) = -21m CR(dda , w), is thus given by
[
nW- Ed- I:P( 1\112 )] +[:rr V2D a(nw)]2
11 k liw - El1k
Further reading
Altland, A. and Simon, B . (2006). Condensed Matter Field Theory. Cambridge: Cambridge
University Press.
Bruus, H. and Flensberg, K. (2004) . Many-Body Quantum Theory in Condensed Matter
Physics. Oxford : Oxford University Press.
Fetter, A.L. and Walecka, J.D. (1971). Quantum Theory of Many-Particle Systems. New
York : McGraw-Hill.
Giuliani, G.F. and Vignale, G. (2005). Quantum Theory of the Electron Liquid. Cambridge:
Cambridge University Press .
Mahan, G.D . (2000). Many-Particle Physics. 3rd edn . New York : Kluwer AcademiclPlenum
Publi shers.
126 Real-time Green's and correlation functions
Problems
6.1 Tim e independence . If H is time-independent, show that the time-ordered
correlation function -i (T ACt)B(t')) depends on t - t' and not on t and t'
separatel y.
P = L(-ifiV j )
)
= L
a
f \lJ1(r)(-ihV)\I'a (r)d 3 r .
where T(r) = e- i P.r / n is the translation operator that translates the posi-
tions of all particles by r .
(b) Let C(rat, r ' a ' t ' ) = (\lJ(rat )\lJ t( r 'a ' t ')). Using the fact that T(r) com-
mutes with e- f3H , along with the cyclic property of the trace, show that C
is a function of r - r io Deduce that in a translationally invariant system,
all single-particle Green's function s are functions of r - r'.
6.3 Spectral density function. Show that the spectral density function A(ka, w)
satisfies the normalization condition
6.7 Causal Green 's function. Derive the spectral representation of the causal
(time-ordered) Green's function.
6.9 Greater and lesser correlation functions. For two observables represented
by operators A and B. iC ; B(t) = (A(t)B(O) ), and iC;B(t) = (B(O)A(t) ) .
Assuming that H is time-independent, show that C; B(t - i,B11) = C;B(t).
Deduce that C ; B(w) = e-fJliwC ; B(w).
6.10 Susceptibility. Let 11 XAB(t) = -ie(t) ([A(t), B(O)]), where A and B are her-
mitian operators.
(a) Show that XAB(t) is real.
(b) Deduce that [XAB(W)] * = XAB( -w). This shows that Re XAB(W) is an
even function of w, while 1m XAB(W) is an odd function of w.
Rex(w)--P
_ 2
JT
10
00
w' 1m X (w')dw'
??'
w'- - w-
_ 2w
1m x(w)- P
JT
10
00
Re X (w')dw'
W
,2
-
?
w-
where P stands for the principal value. To prove this, consider the integral
I = r x(~')dW'
Jc w -w
128 Real-time Green 's and correlation junctions
w
Figure 6.2 The contour C. The large semicircle is at infinity. while the radius of
the small semicircle is infinitesimal.
where C is the contour shown in Figure 6.2. The large semicircle is at infinity,
while the radius of the small semicircle is infinitesimal. Show that
I
X(w) = -. P
f oo X(w')dw'
- ,- -
In _oo w-w
and then equate the real and imaginary parts on both sides of the above
equation.
6.12 Polarizabi lity. Starting from the spectral representation of the retarded con'e-
lation function, derive the expression for the polarizability of a non interacting
electron gas.
6.13 Equation of motion. Derive the equation of motion for the retarded Green 's
function for an interacting system of bosons.
-deEd [ -1 -
2
4-q '2
8q'
In 12-q/
--
2 + q'
1J 3D
-d(EF) [~In
q'
122 +- q'q' IJ 1D
where q' = q/ kF .
7
Applications of real-time Green's functions
Here, E is the energy of the level , d! (da ) creates (annihilates) an electron of spin
projection a in that level, n t (nt) is the number operator for spin-up (-down)
electrons, and U > 0 is the onsite Coulomb repulsion. The second term in the
Hamiltonian tends to prevent double occupancy of the energy level. If only one
electron occupies the level, the energy of the dot is E. If two electrons, one with
130
7.1 Single-level quantum dot 131
E
-H- 2<: + U
Figure 7.1 Energy of a quantum dot with a single level. If only one electron
occupies the level. the energy of the dot is 6. If there are two electrons in the dot,
the energy is 26 + U.
spin up and the other with spin down occupy the level, in accordance with the Pauli
exclusion principle, then the energy of the dot is 2E + U Csee Figure 7.1).
The retarded Green's function of the system is
(7.2)
where (... ) stands for thermal averaging, eCt) is the step function, and {A, B} =
A B + B A is the anticommutator of operators A and B. We proceed to determine
the retarded function by the equation of motion approach,
a .
i-G:Ct) = oCt) + e(t)({do-(t), d~ (O)}) (7.3)
at
where do- (t) = ado-(t)/at. The Heisenberg equation of motion gives
. i
do-(t) = - - [do-Ct), HD(t)] . (7.4)
n
The commutator on the RHS of the above equation is easily determined:
Since na commutes with da and dt, the first term is simply 8Ct)(na). The second
term vanishes because l1 a commutes with the Hamiltonian H D. Evaluating the third
term , with the help ofEq. (7 .S ), we obtain a term containing the product l1 a Ct)na (t).
Noting that
(7.10)
Figure 7.2 A thin insulating layer separates a metal and a quantum dot. Electrons
tunnel back and forth from the metal to the dot.
Eq. (7. 11), which has a pole at (E + U) In. The energy of the added electron is thus
E + U, and the total energy of the two electrons is 2E + U, as expected.
He is the Hamiltonian for the metal, HD is that for the dot, and HT describes the
interaction between the metal and the dot. We assume that the electrons in the metal
are noninteracting, or that each electron interacts with other electrons through a
self-consistent average potential. Thus,
(7.13)
where Ek = 1i 2 k 2 / 2m * - fJ-, m* is the effective electron mass, and fJ- is the chem-
ical potential of the metal (the Fermi energy). As in the previous section, the
Hamiltonian for the single-level quantum dot is
(7 .14)
a
The interaction between the metal and the dot is described by a tunneling Hamil-
tonian:
The first term in Hr describes tunneling of an electron from the dot to the metal, a
process whose matrix element Vk is assumed to be spin-independent . The second
term describes tunneling from the metal to the dot. In writing Hr, we have assumed
that no spin flipping occurs during tunneling. The model described above, summa-
rized by Eqs (7.12-7.15), is known as Anderson's impurity model. It was first used
to describe the localized states of a magnetic impurity inside a metal (Anderson,
1961).
We proceed as before; the retarded Green's function of the dot, now denoted by
C:a (f), satisfies the following equation of motion
where
(7.17)
. 0 R R R
In-Ckda(t) =
ot EkCkda(t) + VkCda(t)· (7.18)
where
(7.21)
(7.22)
(7.23)
We have generated three new retarded functions. Equations of motion for these
functions will clearly generate complicated functions whose equations of motion
will produce even more complicated functions . It is a never-ending story, and we
7.3 Tunneling in solids 135
(7.24)
Within mean field approximation, C:da and D:da vanish, because quantities such
as (ctd), (do- da ), and (dl da ) are equal to zero, and Eq. (7.20) reduces to
In Fourier space, Eqs (7.16), (7.18), and (7.25) take the following forms :
ft~~R
~----~ ,,... ------_ ...... -
(a) (b)
Figure 7.3 Two metallic electrodes separated by a thin insulating layer (usually
metal oxide). In (a) no bias voltage is applied; the system is in equilibrium and the
chemical potentials on the left and right sides of the tunnel junction are equal. In
(b) a bias voltage V is applied; eV = i-LL - i-LR and a tunneling current flows.
application of a bias voltage which raises the chemical potential of one electrode
relative to the other (see Figure 7.3). We assume that the electrodes are metals
in the normal state; at sufficiently low temperatures all states below the chemical
potential are occupied. The treatment, however, could be generalized to allow either
one or both electrodes to be superconductive. For a detailed treatment of tunneling
in solids, the reader is referred to the book by Duke (Duke, 1969).
The model Hamiltonian for the tunnel junction is written as
where HL (H R ) is the Hamiltonian for the left (right) electrode, and HI is the
tunneling Hamiltonian. In general, H L may include the interparticle interactions in
the left electrode, and HR may include those in the right electrode. The tunneling
Hamiltonian takes the form:
where c~a (Cka) creates (annihilates) an electron in the left electrode in the single-
particle state Ika), and b~a (b qa ) creates (annihilates) an electron in the right elec-
trode in the single-particle state Iqa). We assume that the creation and annihilation
operators in the left electrode anticommute with those in the right electrode.
The first term in HI describes tunneling of an electron from the state Ika) in the
left electrode to the state Iqa ) in the right electrode; the amplitude for this process
is the matrix element Vkq, which is assumed to be spin-independent. The second
term in HI describes tunneling in the opposite direction. We assume that no spin
flipping takes place during tunneling, which is generally true if the metal electrodes
and the insulating layer are nonmagnetic. In general, a tunneling process is either
elastic, in which case the energy of the electron does not change, or inelastic,
whereby electron tunneling is accompanied by an excitation in the insulating layer.
7.3 Tunneling in solids 137
(a) (b)
Figure 7.4 In (a) the left and right electrodes are separated: Hr = O. In (b) the
electrodes are brought into contact with an insulating layer: H r f::. O.
t < to
(7.33)
t :::: to.
The external perturbation is H T , and the current flows in response to this perturba-
tion. This is illustrated in Figure 7.4.
The electron current is obtained from the rate of change in the number of
electrons in one of the electrodes, say the left electrode,
1= -e(Nd (7.34)
where N L is the number of electrons operator for the left electrode,
NL = LC~aCka . (7.35)
ka
The rate of change of N L is given by the Heisenberg equation of motion,
. i i
NL = --[N . Ho + H T ] = --[NL, HT]' (7.36)
n L h
The last equality follows since N L commutes with HL and HR. The commutator
on the RHS is easily evaluated,
(7.37)
138 Applications oj rea/-time Green's junctions
According to linear response theory (see Section 6.6), (f.h(t»)ext is given by Kubo's
formula
(ih(t»)ext = (ih(t»)o - ~
11
11
10
dt' ([N LCt), HT(t')])o. (7 .38)
The zero subscript means that the operators evolve according to Ho,
N LCt) = eiHoI/1i N Le-iHol/li, HTCt) = eiHol/1i HTe-iHol/li. (7.39)
(7.40)
. = 11~ei (I-'LNL+I-'RNR)I/1i
X(f) [IIr L N L + rII R N R, bqat cka ] e-i(I-'LNL+I-'RNR)f/1i .
The evaluation of the commutator on the RHS is straightforward; we find
where H. C. stands for hermitian conjugate. The current depends on the bias voltage
V and temperature T Cfrom thermal averaging) . In the above expression, the time
development of the creation and annihilation operators is govemed by flo = H L +
HR -llLNL - f.1 R N R . Defining the operator A(t) by
ICV , T) = e
2
l1 to
t
ldt' ([e -ieVt /1i ACt) - eieVt /1i At(t) ,
Since the electrodes are nOlTnal metals, (ACt)A(t'))o = (A t Ct)A t (t'))o = 0, because
quantities such as (cL( t) cL(t'))o and (Cka(t)Cka(t'))O vanish; however, this is not
true if the electrodes are superconductors, as we will see in Chapter 12.
We are interested in evaluating the current long after the extemal perturbation
has been turned on: t » to (steady state); equivalently, we set to = -00. Under this
assumption, the current is given by
Since flo is time-independent, the ensemble average on the RHS of the above
equation depends on t - t' and not on t and t' separately. Moreover,
! -co
I dt ' . .. = -l°
00
d (t - t') ... = r~(t -
Jo
t') ... = ! COd(t - t') e(t - t' ) . .. .
-00
(7.42)
2e R
I(V , T) = -JIm DAA(w)l w=-eV /Ii. (7.43)
ft-
We have succeeded in expressing the tunneling current in terms of a retarded cor-
relation function . We stop here, since the evaluation of this function is most easily
carried out by evaluating the corresponding imaginary-time function, followed by
analytic continuation. Imaginary-time Green's functions will be discussed in the
next chapter. In one special case, however, the retarded function can be evaluated
directly. If we assume that tunneling is elastic, and that the electrons in the left and
right electrodes are noninteracting, the evaluation of the current is not difficult (see
Problem 7.4).
Further reading
Bruus, H. and Flensberg, K. (2004). Many-Body Quantum Th eory in Condensed Matter
Physics. Oxford : Oxford University Press.
Mahan, G.D. (2000). Many-Particle Physics, 3rd edn. New York: Kluwer Academic/Plenum
Publishers.
Problems
7.1 Equation of motion for G:da . Derive Eq. (7.18).
7.3 The dot and metal. For a single-level quantum dot in contact with a metal,
assuming that L R is independent of wand that IL R I « U Iii, derive Eq. (7.31).
Calculate the spin resolved density of states and show that it consists of two
Problems 141
7.4 Tunneling current at T = O. In Section 7.3 , the tunneling current was obtained
in terms of a retarded con-elation function. Assume that
ka qa
where operators with k-subscript are for electrons in the left electrode, while
those with q-subscript refer to electrons in the ri ght electrode. Creation and
annihilation operators with k-subscript anticommute with creation and anni-
hilation operators with q-subscript.
(a) Show th at the current is given by
7.5 Ma gnetic impurity in a metal host. An impurity with a single level is embedded
in a metal host with Fermi energy E F. The model Hamiltonian is given in
Eq. (7 .12). The equations of motioh for G~a (t) and G~da(t) are given in
Eqs (7 . 16) and (7 .18). However, instead of writing an equation of motion for
r:a (t) and then resorting to mean field approximation, as we did in Section 7.2,
let us apply the mean field approximation directly to r :a (t) ,
r:a(t) = -i 8(t) ({n ij (t )d a (t) , dl (O)}) :::= -i 8 (t) (nij)( {da (t), dJ (O)})
= (nij) G:a(t).
143
144 Imaginary-time Green's and correlation functions
will yield the real-time functions that are more intimately connected to experi-
ments. Not only that! We will find that the imaginary-time Green's function is not
merely a means to an end, but that it stands on its own as a quantity of intrinsic
significance: it yields the equilibrium thermodynamic properties of the system.
A perturbation expansion is also possible for the real-time Green's function
at zero temperature. At T = 0, the ensemble average of ACt)B(O) reduces to the
expectation value of this quantity in the ground state; the factor e- f3 H disappears,
and, along with it, the problem noted earlier of the mismatch of exponents.
where
The lower (upper) sign refers to the case when A and B are fermion (boson)
operators. As discussed in Chapter 6, A(e, e t ) is considered a boson operator if the
annihilation and creation operators e and e t are boson operators, or if A consists
of an even number of creation and annihilation operators (e.g., if A = etc, then A
is a boson operator).
From the definition of the T-operator, we may rewrite Eq. (8.1) as
e~B(r, r ') = -e(r - r ' )(A(r)B(r' )) =r= e(r ' - r)(B(r')A(r)), (8.4)
r < r'
, 1°
e(r - r ) = I
r > r io
8.1 Imaginary-time correlation fun ction 145
8.1.1 Time-dependence
If fJ does not depend on time, as is often the case, the imaginary-time correlation
function depends on r - r ' , not on rand r ' separately. The proof of this statement
is as follows:
c~B(r , r') = - e(r - r') Ze 1 Tr [e- tlH eHr /nAe- H(r - r'l/n Be-Hr'/n]
Using the cyclic propetty of the trace, we move e- Hr ' /Ii in the first term to the
leftmost position, and then comm ute it through e- tlH . In the second term we
commute eHr'/Ft with e- tlh, then move it to the rightmost position; we obtain
c~B(r , r ' ) = - e(r - r ') Ze 1 Tr [e- tlH eH(r-r 'l/1i Ae-H (r- r'l/Ji B]
8.1.2 Periodicity
Now suppose that r > O. Then
where A = A(O) and B = B(O). Now perform the following three steps in succes-
sion: (1) move B to the leftmost position, (2) introduce 1 = e+t3 He- tl H at the far
right, and (3) move e- tl H from the rightmost to the leftmost position. We end up
with
c~B(r > 0) = - Ze 1 Tr [e- tl H BeH (r - Wl l/Ji Ae-H(r -tlnl/Ii]
= - Ze 1
Tr [e- tlH B(O)A(r - ,BJi)]
= - ( B(O)A(r - ,Bn)). (8.6)
To make use of the above result, we assume that r is restricted to vary between
-,Bn and ,Bn : r E [-,Bn, ,Bn]. Then, if r > 0, r - ,Bn will be negative, and the RHS
of Eg. (8.6) will be ±c~B(r - ,Bn). Hence, if r > 0,
(8 .7)
l46 Imaginary-time Green 's and correlation junctions
The lower (upper) sign refers to fermions (bosons). Since -{3Pt .:'S r .:'S {311, we can
decompose c~B(r) into a Fourier series
c T (r) = ~ ~ c AB
T
(w )e - iWnT. (8.8)
AB {31i ~ 1/
1/
We can obtain c~B(wl)) in terms of c~B(r): multiply Eq. (S.S) by eiW", T and integrate
over r from - {311 to {31i,
1
.
~ e' Wm rc~B(r)dr = - 1 LC~B(WI/) 1~ e, (w",- wnl rdr .
-f3!i {3!i 1/ - fJ!i
.
Since WI/1 - WI) = 2(m - n)n / {311, the integral on the RHS vanishes unless n = m,
in which case it is equal to 2{3Pt. Hence,
T ( WI/ ) -_ -
CAB I1
2 - f31i
f31i
T ( r ) eiWnrd r.
CAB (S.10a)
C~B(WI/) = ~ [1 0
-f31i
C~B(r)eiw"r dr +
Jo
[f3!i C~B(r)eiw"r dr] .
Making use of c~B(r < 0) = ±c~B(r + {311) and eiwnf3n = ± 1, we can write
wa(rr) = /fr /Ii Wa(r) e- Hr /Ii . w~(rr) = eHr /1i W~(r)e - Hr /Ii.
The lower (upper) sign refers to fermions (bosons). For r > r ', g(rO"r , r ' eJ ' r ') is
the probability amplitude of finding the system with one extra particle of spin
projection eJ at position r and time r if a particle with spin projection eJ' was added
to the system at position r ' at an earlier time r '. For r < r ', g(reJr, r 'eJ lr ') is the
probability amplitude of finding the system with one less particle of spin projection
0"1 at time r ' if one particle with spin projection 0" was removed from position rat
With these thoughts in mind, the imaginary-time Green 's function is written
g(r - r iO", r ), and we consider its spatial Fourier transform g(kO", r),
wa (rr) = Iti
I 'L" eik.r Cka (r ) , (8.15)
.y V k
where {i¢v)} is a complete set of single-particle states, v stands for all the quantum
numbers that characte11ze the states, and d
(c v ) creates (annihilates) a particle in
the single-particle state i¢v).
a a a
The lower (upper) sign refers to fermions (bosons). Using the cyclic property of
the trace, we move e- fir /Ii to the far left and commute it through e- f3fi ,
a a
(8 .16)
a
where n(r) is the particle-number density operator. The ensemble average of the
number of particles in a system of volume V is
(8.17)
The dependence of N on T and fJ- results from f3 fI = f3(H - fJ-N) in the ensemble
average of n(r) . We can solve the above equation for fJ-(N, T , V) and determine
the Helmholtz free energy F from the relation
Once F is found, the thermodynamic properties of the system can be derived. For
a translationally invariant system, we can also write
ka ka ka
(8.20)
v. v'
where 1J stands for all the quantum numbers that characterize the single-particle
state l1>v). Taking the ensemble average, we obtain
(8.21)
v.v'
The matrix element, a c-number, has been moved outside the ensemble average.
Wliting 1>v as 1>1I(r)la), where n represents the orbital (spatial) quantum numbers,
and denoting (a'ifl a) by fa 'a(r), Eq. (8.21) becomes
a.a' n.n'
= '""'Jd
'--'
a.a'
3
r lim fa 'a(r) (W; ,(r')Wa(r)).
r'~ r
We have used the relations connecting the field operators to the creation and
annihilation operators,
(8.22)
II II
Note the necessity of introducing r ' and taking the limit r' ~ r: fa 'a(r) is an
operator that acts on 1>n(r); by introducing r', we make it possible for fa 'a(r) to act
on the product (-1/!;,(r l )1f;a(r)). For any two operators A and B,
In the penultimate step, e- Hr /n was moved to the left and commuted through e- f3H .
It follows that
3
(F ) = " / d r lim fa 'a (r)(\lI1 ,(r', r)\lIa(r , r) )
L r'--7 r
a. a'
= L/ d r 3
!'~r r,l~~_ fa 'a (r)(\lI1 ,(r', r ') \lIa(r, r))
a .a '
= ±"
L
/ d 3r lim lim fa 'a(r) (TWa(r, r)\lJ1 ,(r', r '))
r' ---7 r r ' ~r -l-
a.a '
Likewise, we can express (V), the ensemble average of the potential energy
(assumed to arise from pairwise interaction) , in terms of Green's function; it is
(V) = =f- 1/
2
d 3 r lim lim
r' -H r'~r+
(a
-n- n ?+ fJ )"
ar + -'1-
2m
L g(mr, r "
2
ar).
a
(8.24)
where g)'(rar, r 'ar ') is Green's function for a system with Hamiltonian fiCA) =
fio + AV, and 0. 0 is the thermodynamic potential for a system of noninteracting
particles. In Problems 8.1 and 8.2, the method used to derive expressions for
(V), E, and 0. is outlined.
8.4 Spectral representation, relation to real-time jlln ctions 151
where
= -Zc L 1
(nl e-f3Re Rr/TiCka lm )(m le - Rr /li ctaln)
l1 .m
n.m
(8.28)
where
This is exactly the same function which we obtained in Chapter 6 (see Eq. [6.3\])
when we developed the spectral representation of the retarded Green's function.
Equation (8.27) now becomes
1i
=
OO P>(ka. E)-
dE e.(i(V,,-E) r Jf3 f OOP > (ka, E) (l .=t= e- f3IiE ) -dE
f -00 2rr [W II - E 0
= -
-00 [W II - E 2rr
= f OO ~(ka, E) dE
(8.30)
-00 [W n - E 2rr
where
(8.31)
Imaginary-tim e Green 's and correlationful7ctiol7s
is the spectral density function, and the lower (upper) sign refers to fermions
(bosons). The P-greater function can be written as
>
P (ka, E) = -A(ka, E)
= (-(I.+ nE)A(ka, E) bosons
(8.32)
1+ e- i3nE -(I - f E)A(ka, E) fermions.
bosons
(8.34)
fermions.
On the other hand, the retarded Green 's function is given by Eg. (6.36),
R
G (ka , w) =
f oo A(ka , E) dE
-.
- 00 w- E + iQ+ 2:rr
Assuming that g(ka , wlI ) is found for all positive values of iWII (these form discrete
points on the upper half of the imaginary axis in the complex w-plane), how do we
construct GR(ka, w)? Consider the function F(ka , z) of the complex variable z,
defined by
A(ka, E) dE
f
OO
F(ka, z) = -.
-00 z- E 2:rr
This function is analytic everywhere except on the real axis. Furthermore,
(8.35)
This is the analytical continuation recipe for obtaining the real-time retarded
Green's function from its imaginary-time counterpart. We note that the advanced
real-time Green's function is obtained from the imaginary-time Green's function
by a similar recipe: iWn --+ w - iQ+ .
8.4 Spectral representation, relation to real-time fun ctions 153
In writing C~B(WI1) we dropped the T -operator since r > 0, the integration over r
being from 0 to (31i. We rewrite the ensemble average, introducing a resolution of
identity,
11.111
11.111
Therefore,
(8.36)
Comparing this expression with that for C:B(w), Eg. (6.47), and bearing in mind
our discussion in the previous subsection regarding analytic continuation and its
possible complications, we deduce that
(8.37)
154 Imaginary-rime Green's and correlation functions
Therefore, in order to calculate C:B(w), we first calculate c~B(WII) and then replace
iWII with W + iO+. As a final cautionary remark, we note that Eg. (8.37) is valid
only if (A) and/or (B ) vanish (see Problem 8.8).
(S.39)
(see Eg. [6.55]). Thus, the imaginary-time Green's function for a system of Ilonill-
teracting particles (bosons or fermions) is given by
°
g (ka. w lI )
_
-
1 00
-00
AO(ka. E) dE _
•
I Wn - E
-
2n
-
1 -00
00
8(E - Ekalfl)
•
I W II - E
dE
=} °
g (ka, w lI ) = . 1_ (S.40)
lWII - Ekalft
The retarded Green's function, obtained from gO(ka, w lI ) through the replacement:
. II -+ W
lW + I'ot, IS.
1
CR.o(ka w) = . (S.41)
, W - Eka 1ft + i ot
This is in agreement with the expression obtained in Chapter 6.
Therefore,
-d Cka ()
r -_ -Eka r :::} Cka ( r ) _- Cka (0) e-EkarjTi
-Cka () (8.43a)
dr 11
Similarly,
(8.43b)
Note that cta(r) is not the adjoint of Cka( r). The imaginary-time Green's function
is given by
A-. (x y) -
'!'lIk '. -
~eikx
~
H 11 (a(v_ _ y0 »e l - o "C\,-yj/21•
spin, f,l, B is the Bohr magneton , and a = -1 / 2. + 112. The field operators are given
156 Imaginary-time Green's and co rrelation junctions
by
Assuming that the electrons are noninteracting, the imag inary-time Green 's func-
tion gO(rar. r io- tO) = - (TWa(rr)W; ,(r' O) ) is given by
where
l(n ko-r , n' k'o-'O) = -(TCnka(r)C~'k'a' (O)) = _ e-Enla r /Ii [ e(r) (CllkaC~'k'a ' )
- e( -r)(C~'k'a ' Cllka) ] = _ e-EnkaT/fi [ e( r)(l - fnka) - e( -r)fll ka ] Olln' Okk'Oaa"
Hence,
The presence of a magnetic field breaks the translational invariance of the two-
dimensional electron gas; Green's function in this case does not depend on r - r ',
but rather on rand r ' separately.
where V represents the interaction terms. We have already introduced the modified
Heisenberg picture where an operator A(T) is given by
(8.45)
Note that a hat "/\" above an operator identifies it as an interaction picture operator.
This definition differs from that of the standard interaction picture of quantum
mechanics (see Problem l.11) in that it -+ T and Ho -+ flo. In a way, A( T) is
a modi fied interaction picture operator, but we will refer to it as an interaction
picture operator. Since the imaginary-time Green's function is defined in terms of
a product of two Heisenberg operators, we consider the product
= eHr /Ii e- Hor /Ii A(T )eHor /Ii e- Her -r'l/Ii e- Hor' /Ii 8 (T')eHor' /Ii e- Hr' /Ii .
We have used Eq. (8.45) to express A and B in terms of A(T) and 8(T').
0 -operator
8.7.2 The
The above equation motivates the definition of the 0 -operator,
(8.46)
A(r)B(r ' ) = 0(0, r)A(r)O(r, r ') 8(r ') 0(T', 0). (8.47)
Writing fI in the second term as flo + V, and noting that flo commutes with e Hor/h ,
the above equation reduces to
(8.51)
I == f,T dil f,TIdi2 V(il)V(i2) = f,T dil f,T di2 V(il)V(i2)8(il - i2)
The step function e( il - i2) ensures that i 2 < il. The last equality is obtained by
interchanging the variables of integration il and i 2. Therefore,
T
I = ~ ITdil I di2 [V(il)V(i2)8(il - i2) + V(i2)V(ide(i2 - il)]
2 r' T'
PO) , P(2), .. . , Pen) in any permutation of 1,2 .... , n. The last equality holds
because the last integral is obtained from the preceding one by relabeling the
integration variables: ii , i 2, ... , ill -----7 ip(l) , ip(2), ... , ip(II )' Since there are n! per-
mutations of 1, 2, ... , n, we can write
L V(iP (1» )' " V(iP (n))e(iP(I) - ip(2) )'" e(ip(II _ I) - ip(I1 »)
P
Note that when the time-ordeling operator T reananges V(il) , ... , V(in) in ascend-
ing time order, from right to left, no minus sign is introduced whenever V(ii) and
V(ij ) are interchanged, even if V describes interactions among fermions. This is
because V, when expressed in second quantized form, consists of an even number
160 Imaginary-time Green's and correlation junctions
of operators; i.e., V is a bosonic operator. The V-operator now has the following
perturbation expansion:
r
V(r , r ') = 00
--h drl ... l
L.,.n.1 (1)111r r' r'
drll T[V(rl) ... V(rll ) ] . (8.53)
11 = 0
If we set r = {3h ({3h and r have the same units) and r ' = 0, we obtain
V({311,O) = ef3fioe- f3 H =} e- f3 H = e- f3 HOV({3h , 0). (8.54)
= -Ze l
Tr[e - f3 HoVC{3h, O)TCka(r)cta(O)].
Using Eq. (8.50), the Heisenberg operators Cka Cr) and cta (0) are written as
(8.55)
where Eq. C8.49) is used to establish the second equality. Consider the following
expression:
Equation C8.53) shows that all operators V in the expansion of VCr, 0) occur at
times between 0 and r, and all operators V in the expansion of V ({3h, r) occur at
times between rand {311 (recall that r :::: {3h); hence,
(8.57)
8.7 Green 'sfunction and the (; -operator 161
The introduction of the T operator is justified by the fact that in the product
o ({3n, O)cta (O)Cka (r), the operators are already ordered in increasing time order
from right to left (recall that r < 0) . Thus
where the interchange of the creation and annihilation operators brought about
a minus sign in case they were fermion operators. Expressing the creation and
annihilation operators in the interaction picture, we obtain
Since the operators V that appears in 0(0, r) are bosonic, we can interchange
0(0, r) and (\a( r) without incurring a minus sign. Using U(O , r)O(r, 0) =
U(0, 0) = I , the above expression reduces to
g(ka, r < 0) = _Z;;I Tr[ e-tlHoT 0({371 , O) Cka( r)ct(O)]'
Tr [ e- tl HoU({371 , 0)]
Dividing the numerator and denominator by Zc.o = Tr[ e- tl Ho], we arrive at the
following formula , which is the main goal of this subsection,
( T 0({3T7. O)Cka(r)cL(O»)o
g( ka r ) - - - - - : - - - - - - " " - - - (8.58)
, - (U({3T7.0» )o
162 Imaginary-time Green 's and correlation functions
The subscript "0" indicates that the ensemble average is over the noninteracting
system,
Tr[e -,B Ho ... ]
(" . )0 = ------0--
Tr[e- ,B Ho]
Finally, we note that in Eq. (8.58) we can move (\a(r) to the left, or move the
product Cka (r)cL(O) to the left, without incurring a minus sign because the V
operators in U(f31i, 0) are bosonic. Hence, g(ka, r) is also given by
(T Cka(r)cta(O)U(f3n , 0»)0
g( ka r) - ----,-=----- (8.59)
,- (U (f3Ii, 0»)0
8.8.1 Contractions
Given two interaction picture operators A and B, we define a contraction of A and
B by
,---,
A B = (T AB )o = Tr[e- ,B HoT AB] / Tr[e - fJ HO].
For example,
I I
Cka (r) ct a(O) = (T Cka(r)cta(O» )O= -go(ka, r).
8.8 Wick 's theorem 163
where Qi is a creation or an annihilation operator, and the sum is over all possible
ways of picking n pairs from among the 2n operators. On the RHS of the above
equation, P = I if the permutation of fermion operators required to anange the
pairs as they appear on the RHS , starting from the atTangement on the LHS , is odd;
otherwise P = O. Each summand on the RHS consists of a product of n contracted
pairs.
8.8.3 An example
Let us use Wick's theorem to evaluate
16.+ Imaginary-time Green 's and correlation functions
where 0 :::: rl :::: r, and the operators are fermion operators. Excluding pairs of
annihilation or creation operators (their ensemble average in zero), we are left with
two possible ways to pick pairs from among the four operators,
A = (T Cka( r)c~a(O) )o( T ({ a,(rl)Ck'a,(rl))o
- (T Cka(r)c~/al (rl))o(T c~a(O)Ck'a' (TI))o.
The minus sign in the second term arises from the interchange of the second and
third operators in the original time-ordered product. The first term on the RHS is
so mewhat problematic: it involves a time-ordered product of two operators with
eq ual time arguments , a case for which time ordering is ill-defined. How should we
interpret (T c~'a ' ( rl )Ck'a'( rl ))o? We note that when an operator of a many-particle
system is written in second quantized form, creation operators always occur on the
left of annihilation operators. When this operator acts on any state, annihilation
operators act first, followed by creation operators. We therefore assign to creation
operators a time that is infinitesimally later than the time assigned to annihilation
operators. We thus interpret (T c~'a , (rdck'a ' (TI))O as follows:
(Tc~'a , (rl)Ck'a' (rl))o = ( T c~'a , (r~)Ck'a ' (rl))o = (c~'a, (rnCk'a' (rl))o
= -(T Ck'a ' (rl)C~'a, (r~))o = l (k'a', 0- ).
We have used the fact that an expression such as (Cka (r )c~'a ' (rl ))0 vanishes unless
k = k' and a = a'; this is easily verified by writing the trace as a sum over diagonal
elements and introducing one resolution of identity.
dr
~Cka(r) = ~[Ho, Cka( r)] = ~ ['" EkaC~a(r)Cka(r), Cka( r)] .
11 11 ~
The commutator is given by Eq. (8.42). Hence,
t (r)
Cka = cka
t (O)ei'ka r / Ii
.
Thus,
(b) Let 8 be any operator; e.g. , 8 could be a product of fermion creation and
annihilation operators. Consider the ensemble average of the anticommutator
A
({ Cka , BA} )0 = 2-1 T [ -f3HoACka BA + e -f3 HoBAACka ]
G. 0 r e
The second equality results from the invariance of the trace under cyclic per-
mutations . We now evaluate Ckae-f3Ho. Since [Cka, Ho] = EkaCka,
Cka
A H-011 = (H-0 + Eka
-, )11 Cka·
A
Consequently,
Ckae
-f3Ho _
-
L-
OO
(-f3)11
- Cka
H- A II
0 -
_ LOO
(-f3Y'(H-
-- 0
+ Eka
- )11 Cka
A
n! n!
11=0 11=0
The last equality follows from the relation e MB = eAe B , which is true if
[A, B] = O. In our case, f3 flo commutes with f3Eka since the latter is simply a
number. Equation (8.65) now becomes
(8.66b)
(c) Let b l , b2, . .. ,b 211 be interaction picture fermion operators. We want to find
the anticommutator {b l , b2b3 . .. b 211 }. First consider the case n = 2,
166 Imaginary-time Green 's and correlation fun ctions
In the last step we used the result from (a) above, namely that the anticommu-
tator is a number, so that we could move it to the left.
We can generalize the above result to any positive integer n,
(8 .67)
2r
{b l , h ... b 2r+2} = I) _1)11l {b l , b lll }b2 . .. bm- lb m+1 • • • b2rb2r+lb2r+2
III =2
2r+2 2r+2
= I)-l) lIl {bl , bm}n' b k
III =2 k=2
which shows that Eq. (8.67), if assumed to be true for n = r , will be true for
n=r+l.
8.8 Wick 's theorem l67
2
(T Il a;)o = ( Ta la2)0·
;=1
Wick's theorem is true in this case since there is only one pair. We now assume
that the theorem is true for n - 1, i.e. ;
211-2
(T Il a;)o = L)-Il Il (Ta;aj)o
;= 1
211
Using Eg. (8.66) with B = Il bb we can write
k=2
~ . ~
~ ~ ~
where use is made of Eg. (8.66), and the fact that b l , b 2 , ... are arranged in descend-
ing time order. Thus,
~ ~ ~
(T n
;= 1
aj)o = ( - I ll L(-I)IIl (T blblll )o( T
111 =2
n'
k=2
bk)o· (8.68)
The introduction of T into the last factor is justified, since b2 , ... , b 211 are arranged
in descending time order.
n'
211
Note that (T bk )o is the ensemble average of the time-ordered product of
k=2
2n - 2 operators (recall that k = m is excluded by the prime on n).By assumption,
211
Wick's theorem is true for such a product. In other words, (T n'
k=2
bk)o is the sum
over all contracted pairs that can be formed from the 2n - 2 operators. By summing
over m in Eq. (8.68), we exhaust all pairs that can be fonned from the 2n operators .
Therefore, Wick's theorem is true for n, assuming that it is true for n - 1, as
long as the sign of each term in the sum over m is the right sign . To show that
(-1 )PI (_I)1Il is the right sign in each term, consider the OIiginai arrangement
a I , a 2, ... , a 211 ' First we rearranged the operators to b I, b2 , ... , b 211 , which brought
"
about (-1 )P I • Then, in order to form the contraction b l bill, b l must be moved m - 2
steps to the right (or bill moved m - 2 steps to the left), which brings about a factor
(_1)111 -2 = (_1)IIl . Thus, (-I)P , (-1)111 is indeed the correct sign in each term in the
summation.
(a) We have proved the theorem for the case when each of ai , a2 , ... is either a
fermion creation or annihilation operator. It is not difficult to extend the theorem
to the case when each of ai , a 2, .. . is a linear combination of a creation and an
annihilation operator.
(b) Although the proof is given for fermion operators, the same steps may be
followed to show that Wick's theorem is also valid for boson operators; in that
case, one has to replace anticommutators with commutators.
(c) The theorem is also valid if some of the operators ai, a 2, ... are fermion oper-
ators while the rest are boson operators; in this case, the permutation P in the
factor (-l)P is the permutation of the fermion operators.
8.9 Case study: first-order interaction 169
•
T " " •0 T
• " •
0
ka ka ka
(a) (b)
Figure 8.1 (a) A particle is created at time 0 in state Ikcr ), and a particle in state
Ikcr ) is annihilated at time r. (b) g O(kcr, r) is represented by a solid line directed
from point r to point O.
In the perturbation expansion for Green's function, Eg. (8.60), let us first consider
the numerator. The zeroth order term, n = 0, gives
where the superscript and subscript on og indicate that this is the first-order cor-
rection in the numerator. Applying Wick's theorem, we find
(T Cka( T )cL (O)ct +qal (TI )ct-qa/ TI )Ck2az(TI )(\Ial (TI ))0 =
- (T Cka (T )cta (0))0 (T ct - qal (TI )Ck2a2 (TI)) ° (T CL- qa2(TI )Ck1a\ (TI ))0 (a)
+ (T Cka (T )cta (0))0 (T ct\ +qal (TI )Ck\a\ (TI ))0 (T ct2- qa2(TI )ckla2 (TI ))0 (b)
+ (T Cka (T )ct\ +qa\ (TI ))0 (T cta (0)Ck2a2 (TI ))0 (T ct 2- qa2(TI )Ck1a\ (TI ))0 (d)
- (T Cka (T )ct\ +qa\ (TI ))0 (T cta (O)Ck\a\ (TI ))0 (T ct 2- qa2(TI )Ck 2a2(TI ))0 (e)
+ (T Cka (T )cL-qa/ TI ))0 (T cta (O)Ck\a\ (TI ))0 (T ct\ +qa\ (TI )Ck2a2 (T1 ))0 (f)
We can represent the above terms graphically. The time arguments T, TI , and 0 are
arranged from left to right. At time 0, a palticle is created in state Ika ). At time
T, a particle in state Ika ) is annihilated. At time TI, two particles in states Iklal)
and Ik2a2) are annihilated, while two particles are created in states Ikl + qal ) and
Ik2 - qa2). These processes are depicted in Figure 8.2. The interaction process
at time TI is represented by a dashed line carrying wave-vector q , two solid lines
going Ollt, and two solid lines coming in. A contraction of an annihilation operator
and a creation operator corresponds to connecting an arrow that is leaving a point
to an alTOW that is entering the same or a different point, forming a directed solid
line. The line represents a noninteracting Green 's function. Wick's theorem tells
LI S to slim over all possible ways of picking pairs for contraction, from amongst
all the operators. Each possible way corresponds to connecting arrows in pairs. A
pair consists of one alTOW leaving a point and one arrow entering a point. Each
possible way thus results in a particular diagram, known as a Feynman diagram
or a Feynman graph. Representation of perturbation theory in terms of diagrams
was originally developed by Feynman in his work on quantum electrodynamics
(Feynman, 1949a, 1949b). Hence, we can restate Wick 's theorem pictorially as
8.9 Case study: first-order interaction 171
I
)
~q
)
T o
Figure 8.2 A particle is created in state IkG' ) attime O. while a panicle in state IkG' )
is annihilated at time T. At time T" two particles in states Ik,G', ) and Ik2G'2) are
annihilated, while two particles are created in states Ik, + qG', ) and Ik2 - qG'2).
follows: connect all arrows in pairs (an arrow leaving a point to an arrow entering
a point), and do so in all possible ways; each possible way produces a diagram,
then add all the diagrams. Contemplation of Figure 8.2 reveals that we can form
six Feynman diagrams (see Figure 8.3), cOITesponding to the six different ways of
contracting the pairs.
In Chapter 9, we will develop rules for translating the diagrams into algebraic
expressions. For now, we only note the following:
• Each diagram in Figure 8.3 contains three solid lines; hence the analytical expres-
sion for each diagram contains a product of three noninteracting Green's func-
tions . This is in conformity with Eq. (8 .69).
• The Kronecker deltas which appear in Eq. (8.69) can be directly read off the
diagrams, since each solid line may contain only one wave vector and one
spin projection . For example, looking at Figure 8.3c, the following product of
Kronecker deltas can be read directly off the diagram: Ok ,k,-qOk .k20k , .k ,+qOa.a2 =
Ok.k2 0 q.Ooa.a2.
• With the products of the three Green 's functions and the Kronecker deltas deter-
mined, the six terms in Eq. (8.69) can be written directly from the diagrams. The
only remaining question is the sign: diagrams a , c, and e produce a positive sign,
while diagrams b. d, and f produce a negative sign. For now, we state without
proof that the sign is given by (_I)II+F, where n is the perturbation order (n = 1
in this case) and F is the number of closed fermion loops: (-1) ,+I = + I for
diagrams a, c, and e. Diagram b has two fem1ion loops, and diagrams d and f
have zero fermion loops; hence their negative signs.
• Diagrams a and b are disconnected, whereas diagrams c. d. e, and f are con-
nected.
172 Imaginary-time Green's and correlation functions
k 2 0"2 k2 - q0"2
k, - qa,O k,a,
qa,Ok,a, 0
q:
k, + kl + qO"l k 10"1
71
• ;0 ;0
• •
7
;0 ;0
•
7 kO" kO" 0 kO" kO" 0
(a) (b)
0
q : q : k2 - q0"2
y, t;..
kO" , kO" kl + qO"l :
, klO"l
• ;0 ;0 ;0 ;0
• 71
7 k2 - q0"2 71 k 2 0"2 0 7 kO"
(c) (d)
0
q: k1 + q0"1
q:
~
A
kO" kO" k2 - q0"2 : k 2 0"2
,
• ;0 ;0 ;0 ;0
•
7 k1 + q0"1 71 k 1 0"1 0 7 kO"
(e) (f)
Figure 8.3 The six Feynman diagrams corresponding to the six terms in Eq. (8.69).
In the remainder of this section, we will show that, to first order in the interaction, the
contribution of the disconnected diagrams to 8g~~?m is cancelled by the denominator
in Eg. (8.60).
To first order (n = 1), the numerator N in Eg. (8.60) is given by
8.9 Case study: first-order interaction 173
We have already seen that the time-ordered product decomposes into six tel111S, the
first two of which correspond to disconnected diagrams:
= gO(ka, r) [1 + -]-
2fI V
[f3~rl L
Jo L Lvql(klk2qala2, r l)] + conn. dgs.
q kl al k2a2
We have explicitly written the first two terms that correspond to the disconnected
diagrams, lumped together the remaining four terms as "connected diagrams," and
denoted the term in braces by l (klk2qala2, rl). It is clear, using Wick 's theorem ,
that
Dividing the connected diagrams (which are already first order in V) by the denomi-
nator gives the same connected diagrams plus diagrams of higher order in V; hence,
to first order in V, Eq. (8.70) is exact.
Thus, to first order in the interaction, the denominator cancels out the discon-
nected diagrams in g(ka, r ). As we show in the next section, this cancellation
persists to all orders in the perturbation expansion.
174 Imaginary-time Green's and correlation functions
T ko-
vvv ko- 0
AAA
Figure 8.4 External points at times T and 0, and internal points at times TI, T2, ... Til'
It is assumed that V is a two-body operator.
) .o
Figure 8.5 A connected Feynman diagram.
T~·?----iO ~ •
o
Figure 8.6 A disconnected Feynman diagram.
- ~n. (_~)
n If drl ... fdr m (T cku(r)c~u(O)V(rl)'" V(rm))O. c
The subscript c means "connected ." We have chosen the internal points
rl, r 2, . .. , r ll1 to be the ones connected to the external points. However, had we
chosen any other set of In internal points, the expression above would not change,
since it would simply amount to a relabeling of the integration variables. The
number of such identical expressions is equal to the number of ways of picking m
points out of n points; this is n! / m !(n - m)!. Hence, the nth order correction in the
numerator of Eq. (8.60) is
o (I)
I
__ '"' 1
gill/III - L m.I(n -m )1.
m=O
Diagrams in which all internal points are connected to the two external points corre-
spond to m = n in the above sum, while diagrams in which only one internal point
is not connected to the two external points correspond to m = n - 1. Diagrams in
which none of the internal points are connected to the external points correspond
to In = O. Summing over m from 0 to n generates all the diagrams, as required by
Wick 's theorem.
The su mmation over m in the above expansion may be rewritten as
" 1 00 00 1
L mien _ m)! ... = L LOI/.II1+j
m=O j=O
m!}! .. ..
171=0
The Kronecker delta ensures that the summand is nonvanishing only when} =
n - m and m :s n. Summing over all orders, the numerator in Eq. (8.60) becomes
N =- LLLOI/.m+j
00 00 cc
m! '!
1 (-r;1)"lor f3ndrl "'lor f3Tzdrm
1/=0 m=O j=O ] 0 0
176 Imaginary- time Green's and correlation functions
We relabeled rm + I, ... , rll as rl, ... , r j, a step made possible by the fact that in sum-
ming over j, only terms with j = n - In make a nonzero contribution. Summing
first over n amounts to removing the Kronecker delta and replacing n by m + j, so
that (-l/n)" = (-l/n)IIJ(-l/n)j; thus
But the second factor, L ~o ... is simply the denominator in Eq. (8.60); hence
0 0
V(rn))o.c.
11=0
(8.7 1)
We have proven that the denominator in Eq. (8.60) cancels out the disconnected
diagrams of the numerator. Impo11antly, we have shown that Green's function is
obtained by summing over all connected diagrams. Of course, there remains the
question of how to translate Feynman diagrams into analytical expressions. We
will take this up in the next chapter.
Further reading
Abrikosov, A.A., Gorkov, L.P., and Dzyaloshinski, I.E . (1963). Methods of Quantum Field
Theory in Statistical Physics. New York: Dover Publications.
Fetter, A.L. and Walecka, J.D. (1971). Quantum Theory of Many-Particle Systems. New
York: McGraw-Hill.
Mills, R. ( 1969). Propagators for Many-Particle Systems. New York: Gordon and Breach
Science Publishers, Inc.
Problems
8.1 (V) and (E). For a translationally invariant system of interacting particles,
assume that v(rl - r 2) is spin-independent. The Hamiltonian is
+~L
01 0 2
f f
d3rl d3r2wJJrl)wJ2(r2)v(rl - r 2) Wuz (r2)W U1 (r\).
Problems 177
= =f lim
r'->r+
(-n~
ar
+ £\12 + fL) g(rar, r 'a r ') .
2m
(c) Take the limit r ' ~ r, sum over a, multiply by 1/2, and integrate over
c/3r . Show that the result is Eq. (8.24):
(d) Using the result from part (c), derive Eq. (8.2S).
a 1
Cc) Show that aA QCA) = (vh = ;;:(AVh.
Cd) Integrate both sides over A from 0 to 1 and use the result of the previous
problem to derive Eq. (8.26).
8.6 Wick's th eo rem. Using Wick's theorem , evaluate (N( r)N(r') )o, where N is
the total number of particles operator in the interaction picture.
8.7 An equation/or gO .
(a) Show that, for a tran slation ally invariant system
8.8 Analytic con tinuation. Given two operators A and B , let A= A - (A) and
i3 = B - (B ). Let
e:B(w) = e1iJ(w).
(b) Show that, if (A) and ( B ) do not vanish, then c~B( r) =j:. ch(r).
(c) Find the relation between C~B(Wn) and C~iJ(Wn). Show that
e1iJ(w) = ch(wn)!iw,,=W+iO+ .
9
Diagrammatic techniques
We now consider in detail the rules for the construction and evaluation of Feynman
diagrams. At the end of the previous chapter, we expressed the imaginary-time
single-particle Green's function as a perturbation series in connected diagrams,
(9.2)
179
180 DiagrammclIic techniques
v
d f h j
'/
•
a
3
I
I
b
~ .
T
o
c e
A
9 i
Figure 9.1 External fermion lines at, and O. and internal ones at 'I and ' 2. Dashed
lines are interaction lines. a and b represent the coordinates (kO"), c = (k 1 + qO"I),
d = (k2 - q0"2), e = (k 10"1), f = (k20"2), g = (k3 + q '0"3), h = (k4 - q '0"4), i =
(k30"3), and j = (k 40"4).
, , ,
T. ~ ) ' ~ ~ ' ) ) '0 T.) ) , ) ~ .0
a C Tl e 9 T2 i b a C Tl e b
Figure 9.2 Connected and disconnected diagrams. The diagram on the left,
denoted by (a c)(eg)(f d)(i b)(j h ), is connected. The diagram 011 the right, denoted
by (a c )(eh)(fd)(ig)(jh), is disconnected.
(9.3)
We need to evaluate
The evaluation proceeds by using Wick's theorem: we sum over all possible con-
tractions that result in connected diagrams. Referring to Figure 9.1 and its descrip-
tion, this corresponds to summing all connected diagrams obtained by connecting
the directed fermion lines: one line leaving a vertex to another line entering the
same, or a different, vertex. For example, one diagram (see Figure 9.2) results from
connecting a to c, e to g, f to d , i to b, and j to h; this diagram is denoted by
(ac)(eg)(Jd)(ib)(jh). On the other hand, a diagram such as (ac)(eb)(Jd)(ig)(jh)
is disconnected and should not be counted.
9.1 Case study: second-order perturbation in a s),stem ojjermions 18 1
What is the total number of connected diagrams that can be drawn in Figure
9. I ? The extern al line a can be connected to c, d, g, or h , and line b can be
connected to e, f, i , or j. We start by counting all connected diagrams that can
be obtained by connecting a to either c or d; there are 40 such diagrams , listed in
Table 9.1.
What about diagrams that can be obtained by connecting a to g or h? It is
clear that there are also 40 such diagrams, which can be obtained from the 40
diagrams listed in Table 9.1 by the following interch anges: "I B " 2, (k 1 al) B
(k3a3), (k2a2) B (k4a4), and q B q'. Since "I and "2 are integrated over, and
k\ ai, k2a2, k3a3, k4a4, q , and q' are summed over, the 40 diagrams obtained by
connecting a to g or h make exactly the same contributions as the 40 diagrams
enumerated in Table 9.1. Hence, we may consider only the 40 diagrams in Table
9.1 and cancel the factor 2! in th e denominator of Eg. (9. 1).
Next, we observe th at when we draw the 40 Feynman diagrams listed in Table
9. 1, we find th at there are only 10 generically different (topologically distinct)
diagrams. Each of these 10 diagrams occurs four times . The ten connected, topo-
logically distinct diagrams are shown in Figure 9.3. The diagrams in Figure 9.3 are
so metimes drawn in a different but equivalent way (see Figure 9.4). Table 9.2 lists
the diagrams from Table 9.1 that are topologically equivalent to the diagrams in
Figure 9.3.
182 Diagrammatic techniques
0
0 0
~
(a)
) ~
(b)
I
) 5 ' 5" ,,
(c) 0
I
,,
(d)
,,
0
/ (e)
3 (f)
~
)
0 0
(g)
) ) )
0
,
)
(r)
,
)
~
0 r} )
(i)
JZ I
I
I
(j)
I
I
I
Figure 9 .3 The ten connected, topologically distinct diagram s that arise in second
order perturbation theory. Solid lines are fermion lines , while dashed lines are
interaction lines.
0, (2)
0 I
,
)
,
\
')
I
')
, ,
I)' \)' , )'
\
) ~ ) ) )' ! )' )
)
...
')
-,
')')
,. -,
')
o 0 )
0
,
)
,
)
)
o, )
, -,
') ')
I
, ...... ......
)'~I)'~!~
I
~
\ \
(i) (j )
Figure 9.4 An alternative way of drawing the diagrams from Figure 9.3.
Our next observation is that topologically equivalent diagrams make the same
contribution to 8g(2), since they differ from each other only by virtue of a
relabeling of their intemal time, wave vector, and spin coordinates, which are
integrated over. For example, diagram #20 in Table 9.1 makes the following
9.1 Case study: second-order perturbation in a system offermions 183
contlibution to 8g(2):
I t It It I t It I
(T ckuikUCkl+qUI Ckl - qul Ck2Ul~k3 +q'u3 c k.-q'U4 Ck.u. Ck3Ui )0 . (9.4)
The time arguments of the operators are not shown explicitly: the first operator has
argument r, the second has 0 , the next four have rl, and the last four have r2. We
have ignored the factor 1/2!, since, as noted earlier, this factor cancels out if we
restrict ourselves to the 40 diagrams listed in Table 9.1. The ensemble-averaged
term gives
( ... )0 = - (T c ku( r)ct l + quI( rl ))0 (T c k.w / r2)ctu (0))0 (T c k4U.( r 2)cL- qUl( rl ))0
t t
(T c k2U1( rl )c krWu/ r 2))0 (T c k1u1 ( rl )c k3 + q 'U3( r2))0.
we find
(9.S)
184 Diagrammatic techniques
o
k'eJ'
: k' + qeJ' ~
qt,.
,
,, ,,,q
T _ )
,
) ) -0
keJ keJ
Figure 9.S Diagram CR), a ring diagram .
This expression may be written from Figure 9.5 if we adopt the following
rules:
(1) To each fermion line with coordinates (ka), running from ii to i j, assign the
non interacting single-particle Green's function gO(ka, ii - ij).
(2) To each dashed line with wave vector q, assign the factor vq / V.
(3) Conserve wave vector and spin at each vertex.
(4) Sum over all internal coordinates (in Figure 9.5, these are k ', a ', and q).
(5) Integrate over internal times from 0 to f31i (i and 0 are external times).
(6) Multiply by (-1 /Ii)", where n is the order of the interaction (n = 2 in Figure
9.5).
A factor of -1 is needed to reproduce Eq. (9.6). We note that diagram (R) has
one closed fermion loop . The last rule is:
(7) Multiply by (-I)F, where F is the number of closed fermion loops.
9.1 Case study: second-order perturbation in a system oj'j'e rmions 185
Since the frequencies are odd, WII - Will + Wil l - WII 3 and Will - Wil l + wn 3 - wll.j
are both even ; hence
Thus, I = 0 unless wn .j = WII . Setting WII - Wil l = Will (Will is even), we find Wil l =
WIl - Wm and wn 3 = w + Will' Relabeling Wil l as WII ' , we obtain
nl
(2)
ogR (ka, w n ) = -
-1
n2 )
2
L (II)
Vq 2 ° °
L g (ka, wll)g (k - qa, w ll - wm)
(
f3 qk'o' 1ll .Il '
(1) Assign coordinates (ka, wlI ) to the two external fermion lines. To each inter-
action line, assign a wave vector and an even frequency.
(2) To each internal fermion line, assign wave vector, spin, and frequency coordi-
nates. At each vel1ex, conserve wave vector, spin, and frequency.
(3) To each fernlion line with coordinates (ka, wn ), assign gO(ka , wll ). To each
interaction (dashed) line with coordinates (q, Will)' assign vq / V. Forn1 the
product of all the gO's and (v q / V)'s.
186 Diagrammatic techniques
k'a' ,w n ,
C~~
I T I
.. ..
Figure 9.6 Di agram (R) in momentum-frequency space.
(4) Sum over all intemal coordinates (wave vector, spin, and frequency).
(5) Multiply the resulting expression by (-1/ f31i 2 )" (_I)F, where n is the order of
the interaction and F is the number of closed fermion loops.
(a) Each connected diagram results from a particular set of contractions, e.g.,
I! r----I ,-------,
c( r )ct(r \), c(r2)c t (O), ... , c( rm)ct(rn). Let ii, i 2, .. · , in be a permutation of
1, 2, ... , n . The diagram which results from the set of contractions ~(r )~ t (ri,),
~(ri2)~t(O), ... , ~(ri,J~t(riJ is topologically equivalent to the first diagram
mentioned above, since the two diagrams differ only by a relabeling of their
time indices. The two diagrams have the same numerical value since the contri-
bution of the second diagram differs from that of the first only by a replacement
of the product V(r\)V (r 2) ... V(rll) by V(ri,)V(ri2)'" V(riJ in the integral,
and the dummy time variables are integrated over. Furthermore, no minus sign
is incurred by this rearrangement since V contains an even number of fermion
operators. Because there are n! permutations of 1,2, ... , n, we conclude that
there are n! topologically equivalent diagrams which differ only by the permu-
tation of their time indices, and that all of them have the same algebraic value.
9.2 Feynman rules in momentum-frequency space 187
Therefore, we consider only one diagram from this set and cancel the n! factor
in Eq. (9.8).
(b) Going back to the example used in Section 9.1 , we note that the 40 diagrams
listed in Table 9.2 are divided into ten groups, each of which contains four
diagrams. The four diagrams in each group are topologically equivalent and
have the same algebraic value. Consider, for example, these four diagrams:
#20, #26, #34, and #40, which are all of type (R) . Diagram #34 is obtained
from diagram #20 by interchanging the two vertices of the interaction line at
rl. Diagram #40, in turn, is obtained from diagram #20 by interchanging the
two vertices of the interaction line at r 2. Diagram #26 is obtained from #20
by interchanging the two vertices at rl as well as the two vertices at r2. The
four diagrams make the same contribution to g(ka, r) because they differ only
by a relabeling of their internal momentum and spin coordinates, which are
summed over.
Similarly, at order n, any interchange of the two vertices of a given interaction
(dashed) line yields a diagram that is topologically equivalent to the original
diagram, with the same algebraic value. Since there are n interaction lines,
each of which has two vertices, there are 211 topologically equivalent diagrams
with the same algebraic value that differ only by a relabeling of their internal
momentum and spin indices. If we were to construct a table of these order n
diagrams, similar to Table 9.2, we would find that each of the groups consists
of 2" diagrams (in Table 9.2, n = 2). Since a factor (v q / 2V) appears for each
interaction line of wave vector q, we may consider only one diagram from each
group and assign vq / V to each interaction line, rather than (l / 2)v q / V.
(c) Let
Here, we have dropped the internal momentum, spin, and time coordinates, and
the notation adopted is as follows. The interaction line at ri has two vertices,
denoted by i and i '. The creation (annihilation) operator associated with vertex
cl
i is denoted by (Ci ). The operators can be rearranged such that
({3I1) - 1Lgo(k' a ', w/J ,)e-iw,,, o- = ({3I1) - 1 Lgo(k'a ' , wlI ,)eiW",o+.
11 ' 11 '
(e) Each interaction (dashed) line of wave vector q, occurring at time r', has two
vertices. At each vertex, one fermion line enters and another one leaves, such
9.2 Feynman rules in momentum-frequency space 189
• •
• •
Figure 9.7 Two fermion lines, coming from vertices at r" and r/J, enter the inter-
action line at r' and then leave to vertices at re and rd.
that momentum and spin are conserved. Let us represent the two fermion lines
entering vertices 1 and 2 of the interaction by gO( ra - r ' ) and gO( rb - r '),
respectively, for some ra and rb. The two fermion lines leaving vertices 1 and 2
of the interaction are represented by gOer' - re) and gOer ' - r d), respectively,
for some re and rd (see Figure 9.7). In terms of frequency, we write
Setting Wil l - WII ) = Will (even), we find that Wf/ ) = Wil l - Will and WII4 = Wf/ 2 +
Will' Thus, we can associate a wave vector q and an even frequency Will with an
interaction line, and demand momentum, spin, and frequency conservation at
each vertex, as shown in Figure 9.S.
(f) Each diagram occurring in the expansion of og(Il )(kCT, r) has 2n + 1 fermion
lines and n interaction lines at TI, T2, ... ,Til' Each fermion line, when
Fourier-transformed, produces a factor of ({3l1) - I, while the integration over
TI , T2 , ... , Tn produces a factor of ({3Ii)Il. Hence on the RHS of the expres-
sion for og(lI )(kCT, T), there is a factor of (f31i )- 11-1. Since og (Il )(kCT, T) =
(f3Ii)-1 LII og(II )(kCT, WII )e- iwil r, the expression for og (II )(kCT, wll ) contains the
factor (f3l1 )-11 . Combining this factor with the prefactor (-1 l li)1l which occurs in
the expansion of og(II )(kCT. r) , we conclude that the expression for og (Il )(kCT , w ll )
contains the factor (-1 I f3l1 2)n .
190 Diagrammatic techniques
1. Draw all connected, topologically distinct diagrams with n interaction lines and
2n + 1 directed fermion lines. Two of the fermion lines are external lines, and
the rest (2n - 1) are internal lines.
2. The two external fermion lines have coordinates (ka, (VII) . With each internal
fermion line, associate momentum , spin, and frequency coordinates. To each
interaction line assign a direction, a wave vector, and an even frequency. Con-
serve wave vector, spin, and frequency at each vertex.
3. Assign gO(ka, (V11) to each of the two external fermion lines. To each internal
fermion line of coordinates (k' a ', (Vn ' ), assign gO(k' a', (V11' ). For each fermion
line with frequency (VII' that closes on itself or runs from one vertex of an
interaction line to the other vertex, insert the factor eiw",o+ .
4. Assign to each interaction line with wave vector q and frequency (V1Il the factor
vq / V, where V is the system 's volume.
S. Form the product of all the gO 's and all the (v q / V)'s, and then sum over all
internal wave vectors, spins, and frequencies.
6. Multiply by the factor (-I / ,81i 2)11(-I)F , where F is the number of closed
fermion loops.
distinct diagrams that contribute to og(I1 )(ka, r)? According to the discussion above,
two diagrams are topologically equivalent if one is obtained from the other either
by a relabeling of internal time indices, or by interchanging the two vertices of one
or more interaction lines; therefore,
C(n)
CTD(n) = - - (9.9)
n!211
where C(n) is the number of connected diagrams of order n. In calculating
og (n) (ka, r), the ensemble average is taken over a product involving 2n + 1 cre-
ation operators and 2n + 1 annihilation operators . In forming contractions, an
annihilation operator is contracted with a creation operator. Each possible way of
contracting the 2n + I annihilation operators with the 2n + 1 creation operators
produces one diagram. The first annihilation operator can be contracted with any of
the 2n + I creation operators; the second annihilation operator can be contracted
with any of the remaining 2n creation operators, and so on. The total number of
diagrams, connected or not, is therefore equal to (2n + 1)!. A disconnected dia-
gram is obtained if the operators in one or more V-operators are contracted among
themselves .
Consider the case where the creation and annihilation operators in m V -operators
are contracted among themselves. We note the following:
along with
C(O) = 1. (9.11 )
k'O",w n ,
o:q = O, wm = 0
(a) (b)
Figure 9.9 The two first-order connected, topologically distinct diagrams arising
from the correction to g(ka, CUll) in first order of the interaction (n = I). (a) is a
direct interaction diagram and (b) is an exchange interaction diagram.
For n = 2,
2! 2! 80
C(2) = 5! - -2!C(l) - -4!C(0) = 80 C TD (2) = 222! = 10.
I! l! 2!0! '
For n = 3, we find C TD (3) = 74. Clearly, the number of connected, topologically
distinct diagrams grows rapidly with increasing perturbation order.
ogbl )(ka , w n ) = (;~ ) [l(ka, w n )]2 L ~ gO(k - qa, Wn - wm)ei (w,, - wm)o'"
qm
(9.13)
The first minus sign in og~l ) results from the existence of a closed fermion loop in
diagram (a). The convergence factor ei w", o+ is inserted because the fermion line with
coordinates (k' a ', wn ' ) closes in on itself. On the other hand, the convergence factor
ei «(V,, -wm)o+ in 8gb) arises because the fermion line with coordinates (k - qa, Wn -
wm) is joined by an interaction line. These convergence factors are important;
without them, the summation over frequencies would diverge (see Problem 9.3).
9.4 Feynman rules in coordinate space 193
Frequency sums, as in the above expression for og~l ) (ka, w lI ) and og~l ) (ka, W n ),
often arise in applications of the finite temperature Green's function. Here we
record the following formula:
bosons
(9.14)
fermions
(see Problem 9 .3). In the above equation, ni' and h are the Bose-Einstein and
Fermi-Dirac distribution functions , respectively.
f31i
g(rar , r 'a 'r') = - L -n!I ( - l j n)" l drl .· . l drll
00
0
f31i
0
11 =0
(9.15)
All operators are interaction picture operators. For two-particle interactions (see
Eq . [3 .53]),
VCr ) = ~L 3 3
L jd r jd r ' w l (rr)wt(r'r)vl. lL .I.'IL, (r, r ') w1t'Cr ' r)w).,(rr)
).).' ILIL'
where A, A', f.i" and f.i,' are spin projection indices and VAIL.A'IL, (r, r') =
(Af.i,lv(ra , r 'a ')IA' f.i, '). Ifv is spin-independent, VAIL.A'IL, (r, r ' ) = vCr, r ')ou,oILIL" Let
U (ra r , r ' a ' r ') = vera , r ' a')o( r - r '), where 0 < r , r ' < (311. V (r) may be written
as
(9.16)
The interaction is depicted in Figure 9.10 . The two vertices of the interaction line
are assigned coordinates (rr ) and (r'r'). The first-order correction is given by
og( J)(rar. r 'a ' r ') = ~!d3rJ rf3~rl!d3r; rf3~r; L L (TWa(rr)w! ,(r' r ' )
Jo Jo U' ILIL'
(9.17)
194 Diagrammatic techniques
8g~I; , (X , x ' ) = (1 / 211) fdxjdX; L L (T1/ra(x)W! ,(x ' )W{ (XI)W! (X;)
H ' /1 /1'
(a) Diagrams that differ from each other by a permutation of rl , r 2, .. . , rn are topo-
logically equivalent; they make identical contributions to 8g~~, (x. x ' ). There
are n! such diagrams. We select one diagram from this set and cancel the factor
l i n! in Eq. (9 .15).
(b) In order n, there are 2" topologically equivalent diagrams that differ only by
some interchange of the interaction vertices (ri ri) ++ (r; r(), along with the
corresponding spin indices. These diagrams make identical contributions to
8g (II ), since the interaction is symmetric under such an interchange:
The symmetry occurs because the particles comprising the system are indistin-
guishable: since the particles are identical, interchanging the position and spin
coordinates of any two particles does not change the interaction between them.
Hence, it is sufficient to count each topologically distinct diagram only once,
assigning to each interaction line with vertices x and x ' the factor U A/1. A' fl.'(x , x')
rather than (1 / 2)UA/1.)! /1' (x , x ' ).
(c) For each fermion loop, a factor of -I is assigned.
Following these observations, we write below the Feynman rules for calculating
the correction, of order n, to the Green's function g(ro-r , r'o- ' r'):
(1) Draw all connected, topologically distinct diagrams with 2n vertices (i.e., n
interaction, or dashed lines) and two external fermion lines. At each vertex,
9.4 Feynman rules in coordinate space 195
Q X'(J'
•
X(J
J.loJ.l
o
...
X(J
--.;o~-~-
Tj'....
Xl
A
X'(J'
(a) (b)
Figure 9.11 Connected, topologically distinct diagrams that contribute to the first-
order cOITection to Green 's function.
one fermion line enters and another one leaves. If the interaction is spin-
independent, conserve spin at each vertex.
(2) To each fermion line directed from (rar) to (r'a'r ' ), assign gO(rar , r 'a ' r ').
(3) To each interaction line with vertices (ri r i) and (r; rj), assign the matrix element
U)'Ji..A'Ji. ,(ri ri, r; rf).
(4) Integrate over all vertex coordinates (2n space and 217 time integrations).
(5) Sum over all internal spin indices.
(6) Multiply the resulting factor by (-I/Fi)" (_l)F, where F is the number of
fermion loops .
(7) Any Green's function at equal times is interpreted as gO(rar , r '(J ' r +).
As an example, consider a system of fermions with spin-independent two-
particle interaction: UAJi..A'Ji. ' (X, x') = U),Ji..AJi.(X, x')ou'0Ji.Ji." The first order correc-
tion og(I)(rar, r 'a ' r ' ) is obtained below. There are two connected, topologically
distinct, diagrams (see Figure 9.11). Using the diagram rules,
r (I) (
ug "
rar , r(Jr ') =
- ugaa
r (I) (
, x,x ') -ugaa'.ax,x
_ r (I) ( ') +ugaa '.bx,x. ') =- A
r (I) (
+B
h . Ua/LaJi.XI,X1
wele ( ') -- v ( rl , llurl-rl
" ) r( ° (' ') - ° ('
') an d gJi.11XI,X1 - gJi.p rITI,rITI
I I '+) .
where ~(ka, (Vn), known as the particle's self energy, is obtained by summing
B (ka , (Vn) over all connected, topologically distinct diagrams of all orders in the
perturbation. Graphically, the equation for g(ka, (Vn) is represented in Figure 9.12,
where the self energy ~(ka , (Vn) is written as an infinite sum of terms , called
self-energy terms.
An examination ofthe diagrams appearing in the expansion of ~(ka , (VII) reveals
that there are two types of diagrams: those that can be separated into two pieces
by cutting a single-particle line (third, fourth, and fifth diagrams), and those that
cannot. If we add up all the self energy diagrams that cannot be separated into two
pieces by cutting a single-particle line, we obtain what is known as the proper, or
irreducible, self energy ~ *(ka , (VII)' It is clear that
This is Dyson's equation (Dyson, 1949a, 1949b). An exact expression for Green's
function , in terms of the proper self energy, follows:
gO(ka , w 1
g(ka , w lI ) =
1-
°
g (ka,
. n)
wlI)~ " (ka, w lI )
= -0--
g
1- - - - - - - -
n ) - ~*(ka , lI )
(ka , w w
(9.21)
iW II - Eka/T7 - ~ * (ka, w lI )
where Eka = Eka - fL. We should note that even though the above expression is
exact, the calculation of ~*(ka, w,J is generally a formidable task; in practice,
~*(ka , w lI ) is approximated by a few diagrams.
w - plane
Figure 9.14 The contour C in Eg. (9.22) consists of the real axis, from -00 to
+00, and a semicircle at infinity in the lower half-plane.
GR(ka , t) = -1
2:rr
1 00
-00
.
GR(ka, w)e - 'Wldw = -1
2:rr
1 00
- 00 W -
iwt
e -I
Eko 111
dw'
+IY
where E~o ::::: Eko +IiReL~(ka , Ekoll1) and y ::::: -lmL~(ka, E~o lli) > O. Since
t > 0 (GR(ka , t) vanishes for t < 0), the above integral vanishes if taken over the
semicircle at infinity in the lower half w-plane. Hence,
1
GR(ka , t ) = -
1 e - iwtdw
--- ,--- (9.22)
2:rr cw- Eko 111 + iY
where C is the closed contour shown in Figure 9.14. The position of the pole is
given by W pole = E~oll1 - iy. By the residue theorem, we obtain
The minus sign arises because we go around the contour in a clockwise direction,
and the step function ensures that the retarded function vanishes for t < O. In fact,
if t < 0, we consider a contour consisting of the real axis and a semicircle at infinity
in the upper half-plane. The contour integral then vanishes, since the pole is outside
the contour. Furthermore, for t < 0, the integral over the semicircle at infinity in
the upper half-plane also vanishes; hence, the integral along the real axis vanishes,
and GR(ka , t < 0) = O.
The retarded function of the noninteracting system is given by
a
k'(J"
: k'+q(J": k'(J"
qA yq
T1 T2
T. o. o. o. ·0 T 0
k(J' T1 k - q(J' T2 k(J' k(J' k(J'
k-q(J'
(a) (b)
Figure 9.15 Two equivalent drawings of the ring diagram.
we obtain
2
V .
8g(2) = e(2)({3n) -2 L v~ L go(ka , w 17 l )go(ka, wIIJe - 1w"l r I
k'a 'q 11 11'12
200 Diagrammatic techniques
Noting that
li
1f!lid LI I f! d L2 ... = I f! dL I I
li T
dLI
where
Here, gO> (kO', L) = gO(kO', L > 0) and gO « kO', L) = gO(kO', L < 0).
We recall some results from Chapter 8,
(see Eqs (8.28) and (8.32-8.34»). AO(kO', E) is the spectral density function for the
non interacting system. Using the above expressions for gO> and gO <, we can write
(9.23)
(9.24)
9.7 Time-ordered diagrams: a case study 201
Therefore,
(9.25)
where
(9.26)
Similarly, defining D by
-
00 dEJ1°O dE21 °O dE3
--- --- ---D
l f31i
e(-iW"2+El-E2+E3)T2
dr' - - - - - - - - (9.28)
1 -00 2:rr -00 2:rr -00 2:rr ° - iW1l 1 - EJ + E2 - E3
og~2)(ka, r) = c(2)(f3/i)-J L
k'a' q
(v q / V) 2 L[go(ka. w n )] 2e-iw"r
n
1 ~~
00
00
(9.30)
To arrive at this expression , we used ex p(i W II 1 (3h) = ex p(i Wil l (3h) = -1. Replacing
Band D by their values given in Eqs (9.26) and (9 .27), we find
X = p O>(k - qa. EJ)p o« k 'a ', E2)p O> (k' + qa ' , E3)[1 + e.BIi (- EI+El- E3) ]
+ p O« k - qa , EJ)p o> (k'a ' . E2)p O« k ' + qa ' . E3)[1 + e fJli (E I- E2+ E3) ]
II
we finally obtain
og
(2)
(ka, w n ) = c(2) ~ (v / V)-[
L ? g ° (ka. w ) ] ?- ] 00 dEJ] OO dE2] OO dE3
2n -00 2n -00 2n
q lI - - -
-00
k'a' q
k
g ( a , wlI
)_ ] 00 A(ka, E) dE CR(k
a,
) _]00 A(ka, E)
+
dE
.
-
-00
.
I wn - E 2n
, W -
-00 W - E
.
+ 10 2n
CR(ka, w) is analytic and well-defined everywhere in the upper half w-plane ; it
is also well-defined on the positive imaginary axis. Thus,
This function is analytic everywhere above the real axis, as can be seen from Eq.
(9.32). Furthermore, it coincides with oC R .(2) (ka. w) on an infinite sequence of
points, along the positive imaginary axis, whose limit lies in the region of ana-
lyticity (the limit point is at infinity on the positive imaginary axis). A theorem
9.7 Time-ordered diagrams: a case srudy 203
(a) (b)
Figure 9.16 The two time-ordered diagrams corresponding to the diagram in
Figure9.15: (a) 'I > ' 2. (b) ' 2> 'I·
to that effect, in the theory of complex functions, assures us that the functions
F(ka, w) and oC R ,(2)(ka, w) coincide everywhere in the upper half-plane. There-
fore, the expression for oC R .(2)(ka , w) is obtained from og(2)(ka, w lI ) by simply
replacing i Wn by W + iQ+ . We conclude that by using the approach outlined in this
section, resolving the problem of analytic continuation, which was mentioned in
the previous chapter, is straightforward: obtain g(ka , w lI ) as a sum of time-ordered
diagrams and replace iWII by W + iO+ to obtain GR(ka, w).
There remains the problem of how to develop a set of rules for writing the
algebraic expression that corresponds to any particular diagram. It took a great
effort to arrive at the expression for og (2) (ka , w n ) , and this approach becomes
worthless if a similar effort is required for the evaluation of each diagram.
We observe that the expression for og (2)(ka , wn ) can be written directly if we
redraw Figure 9.15 as in Figure 9.16, and if we adopt the following rules. Arrange
the time coordinates of the vertices so that they decrease from top to bottom.
Consider all possible time orderings; in this case, there are two time orderings:
i l > i2 (Figure 9.16a) and i2 > i l (Figure 9.16b). The two external lines are
where A( r" ... , r N) is some function of the internal time coordinates, and the
wave vector and spin arguments are suppressed. Fourier-expanding the external
lines,
x A(r ... r ) = _1_ ~ gO(w )go(w ,)e-iw"( r-r ') e - i (w,, - w,,,) r '
,, , N (f3h)2 ~ /1 1/
n.n'
(9.33)
Since ogfN depends on r - r ' and not on rand r ' separately, it follows that
That is, since rand r ' do not appear in the above integral, the integral must vanish
unless WI/ = WI/ ' ; if it did not, then ogfN would depend on r' because of the term
exp[ -i(w/1 - wl/ ,)r']. In other words, the frequencies of the two external lines,
along with their momentum and spin coordinates, have to be the same; we have
assumed this all along.
The derivation of the rules for the time-ordered diagrams is given below. These
rules were enunciated, but not derived, in a paper by Dzyaloshinski (Dzyaloshinski,
1962). A similar set of rules were derived by Baym and Sessler (Baym and Sessler,
1963). An alternative derivation is provided in the remainder of this section. The
9.8 Tim e-ordered diagrams: Dzyaloshinski 's rules 205
T o
reader who is interested only in applying the rules may go to the end of the section,
where the rules are listed.
Consider any Feynman diagram r N of order N. We can represent it as in Figure
9.17 . An external line is directed from T to some vertex, whose ti me coordinate we
call TN. A second external line is directed from some vertex, whose time coordinate
we call TI, to T = O. Inside the circle, there are N interaction time vertices, and
internal lines run between these vertices. The total number of intemallines depends
on the system under consideration and on the type of interaction involved. For a
system of fermions with two-particle interaction, the total number of internal lines
is 2N - I. For other types of interactions, such as scattering due to impurities, or
for other kinds of systems, such as interacting electrons and phonons, the number
differs from 2N - I.
The algebraic expression corresponding to the diagram in Figure 9.17 is
erN is a counting factor that depends on the structure of the diagram, and it
is determined by the Feynman rules. M t:.)l is the product of the interaction
matrix elements (determined by the types of interactions at the vertices) , and
it depends on internal momentum and spin coordinates, denoted collectively by
{a i) }, which are summed over. The quantity M~~j l contains Kronecker deltas that
express momentum and spin conservation at the vertices. In the above expres-
sion, Fra;) l(Ti - Tj) is the product of Green 's functions corresponding to the lines
directed from Ti to Tj. If only one line with coordinates ai) is directed from Ti to Tj,
then Flat} I( Ti - Tj) = g£tj (Ti - Tj). If two lines, having coordinates ai) and a ;j are
directed from Ti to Tj , then F la,) j(Ti - Tj) = g£(Ti - Tj)gO, (Ti - Tj). If no lines
IJ ail
are directed from Ti to Tj, then Fra,)l( Ti - Tj) = 1. Fourier-expanding the external
206 Diagrammatic techniques
(9.34)
(9.35)
B B
(9.36)
F F.
n E and i E are the Bose and Fermi distribution functions, respectively. To each
internal line directed from ii to i j we assign a frequency Eij and replace g~;j (ii - ij)
by an integral over Eij, as in the above expressions for gO > and gO < . We obtain
(9.37)
(9.38)
where
Notice the form E P, : it involves a sum over all the lines that enter or leave vertex
r P, ' If a line leaves r P, ' it carries a negative frequency; if it enters r Pi ' it carries a
positive frequency. Also note that
efJn(iw" l -iw,z) - I
= + others.
(iWIl I - iW II2 )EP I (EPI + EpJ ... (EPI + EP + ... + Ep,v_I ) 2
The first term in 12P is obtained by keeping only the upper limits when integrating
over r PI' r P2 ' •• •• r P,v- I' The second term , called "others," consists of the rest of the
terms . If Will # w1l 2 ' then e fJli (iw"l- iw "2) - I = 0, and the first term in 12P vanishes.
208 Diagrammatic techniques
f3 ho n l .lh
12P = - + others.
EPI(EPI + Epz )··· (EPI + EP2 + ... + EPN_ I )
We showed in the previous section, by explicit calculations, that the terms repre-
sented by "others" combine to give a vanishing result. We can show that this is true
in general, for any order of the interaction (see Appendix B). For now, we assume
that this is indeed the case. i2P is then given by
I
f3hO I1I ._lh-
i 2P =
ni < NEpi
, EPi = L Epj . (9.41 )
j =J
(9.42)
The rules for time-ordered diagrams follow directly from Eq. (9.42). Before writing
the rules, however, let us clarify the meaning of EPi ' Note that
EPi = tj =J
Epj = t [- L
j=J P,
Epj Pk +L
Pk
EPk Pj + iWnOPj .N - iWIIOPj .J ] .
Since Lpk ... = Lk ... = L~=l +L k> i' the expression for EPi reduces to
i i
Let us now arrange the time vertices vertically such that time decreases as we move
down. Since r P,v > r P,v - I > . .. > r PI ' the resulting arrangement is that shown in
Figure 9.18. A horizontal dashed line, called a section, is drawn between rpi + 1 and
rpi •
Regarding the above expression for EPi '
(1) The first term is the sum of the frequencies of all internal lines that start at
r Pi ' r Pi- I' ... , r PI and end at the vertices above r Pi ' That is, it is the sum of the
frequencies of the internal lines that are directed upward and that intersect the
section. These frequencies carry a minus sign.
9.8 Time-ordered diagrams: Dz.yaloshinski's rules 209
Figure 9.18 Arrangement of time vertices such that time decreases from top to
bottom. A horizontal dashed line between T Pi+, and T Pi is a section.
(2) The second term is the sum of the frequencies of the internal lines that start at
TPiT " TPi +2 , • •• , Tp,v and end at TPi , TPH , . . · , Tp, . That is, it is the sum of the
frequencies of the intel11allines that are directed downward and that intersect
the section. These frequencies calTY a positive sign.
(3) The third and fourth terms relate to the extel11allines. One extel11alline enters
at TN and one leaves at T]. We draw these two lines vertically as in Figure 9.16.
Note that ifboth lines enter and leave below the section, the combined contribu-
tion of the third and fourth terms is i W II - i Wn = O. Similarly, if both lines enter
and leave above the section, the contribution of each of the third and fourth telms
is zero. If one external line enters below the section (TN E {T P" T P2 , ••• , T Pi})
and the other line leaves from a vertex above the section (T] E (T Pi+" ... , T P,v }),
the combined contribution of the third and fourth terms is iw n . On the other
hand, if one external line enters a vertex above the section and the other line
leaves from a vertex below the section, the combined contribution of the third
and fourth terms is - i WII .
lines (internal and external) that intersect the section between T Pi+, and r Pi ' The
frequency can'ies a plus sign if the line is directed downward and a minus sign if
the line is directed upward.
We are now in a position to state Dzyaloshinski's rules for time-ordered
diagrams:
210 Diagrammatic techniques
Further reading
Abrikosov. A.A., Gorkov, L.P. , and Dzyaloshinski , I.E. (J 963). Meth ods of Quantum Fi eld
Theory in Statistical Physics. New York: Dover Publications.
Bruus, H. and Flensberg, K. (2004). Man y-Body Quantum Th eory in Condensed Matter
Physics. Oxford: Oxford University Press.
Fetter, A.L. and Walecka , J.D. (1971 ). Quantum Th eory of Man y-Particle Systems. New
York: McGraw-Hill.
Mahan , G .D. (2000). Man y-Particle Physics, 3rd edn. New York : Kluwer Academic/Plenum
Publi shers.
Problems
9.1 A vanishing sum. Show that
00
L eiw"o+ = o.
11=-00
9.2 Th ermodynamic potential. Using Eq. (8.26) for the thermodynamic potential,
show that, for a system of interacting electrons,
_
Q(T , V. j.L)-Qo(T. rT I 00
1 i o dA. "~ "~ eiw"O+L: *1- (ka, w,,)g I- (ka, w,,).
V. j.L )+2
f3 0 ku "= -00
Problems 211
00 eiw" O+
9.3 Frequency sums. This problem shows how to evaluate
L IWI/
. -
1/= -00
-In'
E
Consider the contour integral
e l)~ dz
I = lim
'7~ O+ 1c z - Eln efJli~ ± 1
where C is a circle of infinite radius centered at z = 0. As Izl ~ 00, if
Re z > 0, the absolute value of the integrand is of order (l/lzl)e - fJIiR ez . If
Re z < 0, the absolute value of the integrand is of order Cl / lz l)e'7Re z. The
integrand is thus exponentially small as Iz I ~ 00; therefore, I = 0.
(a) For the case of bosons, consider
e '7~ dz
I = lim
I)~ O- 1C Z- Eln efJli z - 1
= 0.
L·
1/ =-00 IWn -
-In = f3niE
E 1
ka kqa
(a) Using Wick's theorem, evaluate g(ka , r) to second order in the pertur-
bation.
212 Diagrammatic techniques
9.9 A frequency sum. Evaluate the frequency sum over n' in the expression for
the ring diagra m given in Eq. (9.7).
In this chapter we apply diagram rules to the study of an interacting electron gas
in the high density limit. We saw in Chapter 4 that, in this limit, the Coulomb
repulsion between electrons is small compared to their kinetic energy, and that it
can be treated as a perturbation added to otherwise free electrons. We now show
that perturbation theory must be carried out to infinite order to yield meaningful
results (we previously caught a glimpse of this notion in Chapter 4). This is due
to the long-range nature of the Coulomb interaction: even though the Coulomb
energy between two electrons, e2 / r , decreases wi th increasing distance between
the electrons, the number of electrons in a spherical shell of radius r and thickness
dr is proportional to r 2 dr, so the interaction of one electron with electrons far
away from it is still important. We then use perturbation theory to calculate the
linear response of an interacting electron gas to an external field , and apply this
technique to graphene.
(10.1)
213
214 Elecn"on gas: a diagrammatic approach
,
I:i (kO", wn ) = q, Wm ~ k - qO",W n - Wm
Figure 10.1 Proper self energy of an interacting electron gas in first-order pertur-
bation theory.
Q(T , V , /1)
1 iot T
= Qo(T , V , /1) + "2 d"A ' " i w 0+ *!.. !..
~ e " 1: (ka, w,,)g (ka, w n )
f3 0 ka "
(to.2)
(see Problem 9.2). Here, Q o is the thermodynamic potential of the noninteract-
ing electron gas, f3 = 1/ k B T, 1: *A (ka, wn ) is the proper self energy when the
interaction is "A vq , and g!.. (ka , wn ) is the cOITesponding imaginary-time Green's
function. We have found in the previous chapter that, to first order in the interac-
tion, 1: *(ka, wn ) is a sum of two diagrams (see Figure 9.13). However, one diagram
has vq=o, and since the q = 0 term is excluded in Eq. (10.1), we are left with only
one diagram, as shown in Figure 10.1.
The expression for 1:7 is readily written using the diagram rules,
" *(k a ,
""'I wn ) = ( - 1 ) '~vqe
!i2V " i (w"-w,,,)o+ g O(k - qa,w,,-wlIl ) • (to.3)
f3 qlll
There is one internal wave vector q and one internal frequency Will' and they
are summed over. The interaction line is replaced by v q / V, and the fermion line
by gO(k - qa, Wn - wm). The factor ei (w" - w,,,)o+ arises because the fermion line
connects two vertices of the same interaction line, and the whole expression is
multiplied by (-1 / f3/i 2 y, where n = I is the order of the interaction. Defining W,, '
J O. 2 The need to go beyond first-order perturbation theory 215
(10.4)
Thus, to first order in the interaction, the thermodynamic potential of the electron
gas is
1
= Qo(T, V, p,) - V L Vq A A+q (10.5)
k.q
where Eq. (9.14) is used again, and q is changed to -q, taking advantage of the
fact that V_q = vq . Note that in order to calculate Q(T, V, p,) to first order in the
interaction, gA(kO', wl1 ) in Eq. (10.2) is replaced by the non-interacting Green's
function gO(kO' , w n ), since ~~A(kO', w l1 ) is already of first order in the interaction.
What is the problem with stopping at first order in the interaction? It turns
out that doing so leads to some anomalous predictions about the behavior of the
electron gas at low temperatures:
where kF is the Fermi wave vector and x = k/ kF (see Problem 10.1). Thus
~ 7(kO' ; W [~
2 2
d I1 ) = _ e kF _ 1 + X In 11 + x I ] (10.7)
dx 7Th X 2x 2 1- x
which diverges logarithmically at x = 1, i.e., at the Fermi surface. Since the
energy is shifted by h Re ~rret' and ~7 is real and independent of W n , Re ~rret =
~ 7. Therefore, the derivative of the energy, d Eka / dk, diverges logarithmically
216 Electron gas: a diagrammatic approach
k - qa,wn - Wm
, I
k-k -q I
Wn - Wn ' - Wm !
I
k'a,w n ,
k - qa,w n - Wm
(D)
k'a',w n ,
k'a,w n , k' + qajwn , + Wm
C:~
: k' + qa' , W n' + wm:
l q ,wm
I
q wmf
' I
~
k - qa,w n -Wm
(J) (R)
Figure 10.2 The three second-order diagrams that contribute to the proper self
energy of an interacting electron gas.
at the Fermi surface. It follows that the density of states at the Fermi surface
vanishes (see Problem 2.7). However, no such behavior is observed in metals:
the density of states at the Fermi surface of metals is actually nonzero.
(b) From the expression for the thermodynamic potential, Eq. (l0.5), the specific
heat of the electron gas at constant volume, C v, can be evaluated. It is found
that, as T ---* 0, C v ---* Tln T (Bardeen, 1936; Horovitz and Thieberger, 1974;
Glasser, 1981; Glasser and Boersma, 1983). The logarithmic dependence on
temperature of the specific heat, a measurable quantity, is not observed in
metals at low temperatures; in fact , the electronic specific heat varies linearly
with T.
(10.8)
In the expression for ~;.( D) ' a factor ei w",o+ is inserted because the line with
coordinates (k' a, wn' ) connects two vertices of the same interaction line. In 2: ;.( R) '
a factor of -1 results from the presence of one fermion loop.
In the expressions written above, if summation over the frequencies were to be
carried out, it would result in Fermi and Bose distribution functions . Summations
over wave vectors are replaced by integrals; e.g.,
: rc--,~, :~::CI:>:~:~:
0. Q
) I II I~ I~ I
I I I I I I I I I
)' " ~ )' '--~) _ _...J
L,~.(i) ex 13-" f 3
d PI .. . f 3
d p" Vq , Vq2 ... Vq" L
W"l"'W IIII
211 - 1
n
J= l
gO(k j <Tj' Wj ).
Here, the internal wave vectors are denoted by PI , ... , PII' The wave vectors
q I . .. . , qll' and k I , ...• k 211 - 1 depend on the external wave vector k and the internal
wave vectors. Similarly, the frequencies WI , ... ,W211 - 1 depend on the external
10.5 Self energy in the random phase approximation (RPA) 219
frequency W II and the n intemal frequencies Wil l ' ... , wlI ,, ' To find the dependence
of ~,7.(i ) on r 5 , we rewIite ~,7. (i ) as a factor that depends on r s times a dimensionless
integral. This is accomplished by the following changes of variables:
where EF = n2 k}/2m is the Fermi energy. The primed quantities are all dimen-
sionless. Thus,
f3 - 11 ex: k 2/ !1f3'-
.
1i ,
f J3 p J. = k 3F f ' Vq) = k -F 2v'1 j
J3 p J' I
Thus, as r s -+ 0 (high density limit), given two diagrams with the same degree of
divergence (DoD), the diagram of lower order in the interaction makes a much larger
contribution to the self energy. For example, diagrams (a) and (b) in Figure 10.3 have
the same DoD, but ~~. (a) ex: r5- 1, while ~;. (b) ex: r~; hence, as r s -+ 0, l~r (a ) l»
1~ ;. (b) l. Similarly, 1~ ;.(c) l» 1~ ~. (d)1 and 1~ ~. (e) l» I~;. (J) I. Thus, for any set of
diagrams with the same DoD, we retain only the one with the lowest order in the
interaction. From the set of diagrams in Figure 10.3, we retain only diagrams (a),
(c), and (e).
The above conclusion may be stated differently: at any order of the interaction,
only the diagram with the highest degree of divergence is retained. All diagrams
with the same order of interaction have the same r s dependence; hence, from among
these diagrams, the one with the highest degree of divergence makes the largest
contribution to the self energy.
0.
0 0 ~i
: +: :+0 :+~ :+ ...
2:;* =
)
I
I
I
I
I
I
'....
I
I
I
'
I
I
, )
I
I
I
"
~
I
I
)
I
I
I
Figure lOA The proper self energy of an interacting electron gas in the random
phase approximation.
a sum of the first-order exchange diagram and the ring diagrams is clearly an
approximation, since other diagrams are not included; it is known as the random
phase approximation (RPA).
3
+ ( - -?-
1
fJJj- V )
L Vqg °(k -
3 qa , Wn - w lI1 )B 2 (q , wm) + ...
qm
(10.9)
where
we can wlite
(10.11)
where
(10.12)
10.6 Summation a/the ring diagrams 221
k'cr',w n ,
IIO(q,wm ) = C :~
k' + qcr', W n ' + Wm
r,q,wm
k - qcr,wn - Wm
o
o o + ... o
0=0+0 + o 0+0
0 =o- -
1-0
Figure 10.5 The bare pair bubble, the dressed pair bubble in RPA, and the proper
self energy that results from summing over ring diagrams.
is the dressed pair bubble. Equations (to.11) and (10.12) are represented graphically
in Figure 10.5. The dressed pair bubble, in random phase approximation, is given
by
The dressed bubble is also known as the polarizability of the interacting electron
gas, while the bare bubble is the polarizability of the noninteracting electron gas.
The nomenclature results from observing that a pair bubble represents a virtual
process (energy is not conserved) in which an electron-hole pair is created and then
222 Electron gas: a diagrammatic approach
annihilated. An electron in state Ik'a ' ) below the Fermi surface absorbs momentum
nq and moves above the Fermi surface; its absence from the Fermi sphere is
equivalent to the presence of a hole. The electron then surrenders the momentum
nq, and recombines with the hole. The electron and the hole, being of opposite
charges, their creation is tantamount to the creation of a dipole moment, which
causes the medium to become polarized.
The bare pair bubble is given by Eq. (10.10); thus,
2 00 1
rro(q W ) = - '" '"
' Ill f3l1 2 V ~ ~ (iw" - Ek /l1)(iw"
k " =-00
+ iWm - Ek+qln)'
The factor 2 results from summing over the spin index. As n -+ ±oo, the summand
-+ -I 1w~ ; hence, the series is convergent, and we are justified in introducing a
convergence factor eiw" o+ (redundant in this case). This allows us to evaluate
rro(q, Will) by the method of pal1ial fractions:
2 1
= f3t:11 2 V L.k lW IIl -
CEk+ q -
-Ek )/ n
f3l1
k'IJ',wn,
L:;ing(klJ, wn ) = q, Wm .:\"
iC:~
" k' + qlJ', Wnl + Wm
r q,wm
I
"
"
"
"
""I' )i
" I
A
"
""
"
t..
"
"
"
! +
0
I
A
" I
Figure lO.6 Ring contribution to the proper self energy of an interacting electron
gas in terms of the screened Coulomb interaction.
shown in Figure 10.6. The double-dashed line is known as the screened Coulomb
interaction V(q, w lI1 ), while the single-dashed line is now called the bare Coulomb
interaction vq . The expression for V(q, Will) can be read directly from Figure 10.6:
The expression for DRPA(q, w lI1 ) in Eq. (10.13) was used. c(q. Will) is the dielectric
function; as we will show later in the chapter, it measures the response of the
interacting electron gas to an external electric potential.
(10.16)
1
= -V (T{[n,q(q, r) - (n,q(q , r) )] [n,q(-q. 0) - (n,q(-q, O) )]} )
1 1
= -V (Tnl-/(q , r)n,q(-q , 0) ) +V (n,q(q. r) )(n,q(-q , 0) ) (10.17)
ka ka
(10.18)
In Eq. (10.18), the second equality holds because whenever a creation and an anni-
hilation operator share the same time argument, it is assumed that the time argument
of the creation operator is infinitesimally greater than that of the annihilation oper-
ator. This is because creation operators occur on the left in the Hamiltonian, while
annihilation operators occur on the right (see Eq. [10.1]). The correlation function
D(q, r) is known as the collective electronic density fluctuations. Note that, in
order to obtain DR from D by analytic continuation, D must be given in terms of
ii,q, as in Eq. (10.17), and not in terms of n,q (see Problem 8.8).
Employing the perturbation expansion for the time-ordered product of modified
Heisenberg picture operators, we can write
D(q , r) -
1 1
V (n,q(q , r) )(n,q( -q, 0) ) = - V L L L n!1 ( -r;1 ) Jorfindrl ...
00 II
ka k'a' 11=0 0
r fili
Jo drll (cL(r) Ck+qa (r)ct'a ,(O) Ck'- qa'(O) V(rl)' .. V(rn) )o.c . (10.19)
diagrams, we encounter two different sets. One set consists of connected but disjoint
diagrams (see Figure 10.7). We emphasize that these diagrams are connected, since
every solid line is connected to one of the exterior points at rand O.
Upon inspecting any connected, disjoint diagram of order n, we observe that it
results from two sets of contractions. One set involves creation and annihilation
operators in il.(q. r) and r V-operators V(ri l)' . . .. Veri,) being contracted among
themselves. The other set involves contractions among il.( -q, 0) and the remaining
s = n - r V -operators. For example, in diagram (b) of Figure 10.7, the operators
in il.( q, r) and V (rl) are contracted among themselves, while the creation and
annihilation operators in fie -q , 0) are contracted together. Thus, for diagram (b),
n = 1, r = 1, and s = O. For diagram (c), n = I, r = 0, and s = 1, while for
diagram (a), n = r = s = O. Since all such diagrams must be summed, and since
there are n! / r !s! ways of choosing r V -operators from among n V-operators, the
contribution of all the connected, disjoint diagrams (cdd) on the RHS ofEq. (10.19)
is given by
I 00 II
cdd = - - ' " ' " -
1 (- -1)11 ~ l I
fJli
drl .. '
l f3drr
1i (Tfi(q. r)V(rl)'" V(rr) )O.c
V ~ ~ n! Ii r!s! 0 0
11=0 1' = 0
f31i fJli
x ior drr + 1... ior dr ll (TFr( -q, O)V(Tr+ I)'" V(T,J)o. c'
The above expression results from the fact that the arrangement of the density
and interaction operators in the time-ordered product is immaterial; since each
of these operators consists of an even number of fermion operators, no minus
sign is incuned upon any reordeling of the operators. Since s = n - r , the above
expression may be recast in the following form:
l f31i l f3dTr
1i
cdd = - -
V
1
L L L --n1)11 -oll
00 00 00 1
r!s!
(
.r+s
0
dTI . . .
0
11=0 1' = 0 s=O
226 Electron gas: a diagrammatic approach
The Kronecker delta ensures that r + s = 11 , allowing the summation over rand s
to extend to infinity. Summing over 11 first has the effect of removing the Kronecker
delta and replacing ( -1 /11)" by ( -1 111 Y( -1 Iii)' . Hence,
cdd = - -
V
1L -1(--1)' l °f31i
r! Ii °
f3Ti
~ ... V(rr)
dr, (T n.(q, r) V(rl)
drl ... l ~ )O. c
r
Ti f31i
1 (
x L -:- --1) l f3 ~
dr l . . . l drs (T /1( -q , 0) V(rl)'
s ~
. . V(rs))o.c
s s! 11 ° °
1 1
= - - (Tnfi(q , r) )(Tnfi(-q, 0)) = - - (l1fi(q , r) ) (l1fi(-q, 0)).
V V '
(10.20)
Thus, the sum over all connected, disjoint diagrams, which appears on the RHS
of Eq. (10.19), exactly cancels the second term on the LHS of that equation . We
conclude that D(q, r) is the sum of all connected (c), nondisjoint (nd) diagrams that
result from the expansion on the RHS ofEq. (10.19). The zeroth order contribution
to D(q , r) is thus
(l0.22)
10.9 How do electrons interact? 227
~
~ ~
(c) C[[) = a:> +® + cD + ...
Figure 10.8 (a) The irreducible bubble is obtained by summing diagrams contain-
ing one bubble with all interaction lines connected to its legs. (b) In random
phase approximation, the irreducible bubble is replaced by the bare bubble.
(c) Collective electronic density fluctuations. (d) In random phase approxima-
tion, the collective electronic density fluctuations are given by the dressed bubble.
L;i ,ex =
D L;*
2 ,ex =
,
!)+~
(a) (b) (c)
L;*3 ,ex --
:0)
,'
(d)
+ pI) +
(e)
oX (f)
+ :~a
(g)
+ b .,.
. . + .0.
- - + .0..
(h) (i) (j)
Figure 10.9 Proper selfenergy diagrams involving at least one exchange interac-
tion. rn third order. we show only diagrams containing one bubble.
B:
Figure 10.10 Classification of diagrams: (A) and (B) are classified as exchange
diagrams, while diagrams in (C) are classified as Coulomb direct diagrams.
exchange diagram. Diagrams (c), (f), and (g) are also parts of diagram lO.lOB, so
we classify them as exchange diagrams as well. However, diagrams (e), (h), (i), and
(j) are part of diagram 1O.lOC, so they are classified as Coulomb direct diagrams.
According to the above discussion, we may classify self energy diagrams of
an interacting electron gas as Coulomb direct or as exchange diagrams. This is
summarized in Figure 10.11.
Let us now take up the question of how electrons interact. The conventional
picture is that, since electron-electron scattering is a pairwise interaction, electrons
lO. lO Dielectric junction 229
(a) aD
, , RPA , )
o ,
"
)
'I ~ ')'
,, +i:""
:,
," ,
(C) ~ = ___
o
+ -;). . . . . ---;)_-*::;z:: + 5 ,,
,,
"
"
"
"
+---..
Figure 10. 11 (a) Proper self energy resulting from Coulomb direct processes. (b)
""
""
~"
Proper self energy arising from exchange processes. Replacing the dres sed Green's
functions with bare ones amounts to retaining only the dominant diagrams at each
order of interaction. (c) The dressed Green 's function.
scatter off each other directly and in a pairwise manner. Allowance is made, how-
ever, for the collective motion of a dense electron gas by assuming that the pairwise
scattering potential is screened by the electronic dielectric function. Figure 10.11,
however, suggests an altemative, more subtle picture, namely that, in the dominant
Coulomb direct interaction processes, an electron can sca tter off the fluctuating
potential generated by the collective electronic density fluctuations. This scattering
is caused by the bare fluctuating potential, as seen in Figure 10.11. The conven-
tional picture also holds, but only for the exchange scattering processes of order
n :::: 2, as Figure 10.11 shows; these exchange processes constitute only a small
correction to the dominant form (Das and Jishi, 1990).
(10.24)
(10.25)
230 Electron gas: a diagram matic approach
The dielectric function is thus a measure of both the screening of the Coulomb
potential in an interacting electron gas, and the response of an interacting electron
gas to an external electric potential
To obtain an expression in closed form for the dielectric function, we need to
carry out the sum over k in the expression for n O(q, w), given in Eq . (10.14).
At high temperatures, this is very hard to do . At low temperatures (k8T «EF,
where EF is the Fermi energy), an expression for nO( q, w) is not difficult to obtain.
n O(q , w) is given by n O(q, iWm ~ w + iO+):
2 L A 2 L fk+q
= V k fjw + Ek - Ek+q + iO+ + V k -fjw + Ek+q - Ek - iO+ .
Replacing k by -k - q in the second term, and noting that f-k = A and E-k =
2 2
Ek = fj k 12m,
2
n O(q w) = -
, V
L fjw + Ek -
j'
k
Ek+q + iO+
+-
2
V
L -nw + Ek -
I".
Jk
Ek+q - iO+
.
k k
(10.27)
First we evaluate the real part of n O( q , w),
For k8T «EF, we may replace A by the step function e(EF - Ek),
A(q, w) = mk ~
2
2n 2fj q
11° x In Ix+ u-I d x
x - ~L
10. J 0 Dielectric function 231
A (q, w) = d (E F / F
2q
[u _+ 1 - 2 u~ In 111 +- IL
LL
J'
I (10.30)
where deEd = mkF / Jr2Tz2 is the density of states, per unit volume, at the Fermi
surface (see Problem 2.2). Equations (10.28) and (10.30) give
1 - l-u 2- In
RenO(q , w)=-d(EF) -
[2 4q / k F 1-
Il+u-I+ l-u
L
2
+ In
4q / k F
1
1
+ +IJ
U
I - u+
.
(10.31)
Next, we evaluate the imaginary part of nO(q, w) . From Eq. (10.27), we find
(10.34)
where
(10.35)
In writing 8 1(q , w) and B2(q , w), we have used I m (x ±~o+ ) = =j=Jr 8(x). The Dirac-
delta function is given by
Replacing fk e
by (E F - Ek), and Lk by (2~ )3 2Jr f k 2d k f ~ 1 d cose, we obtain
8 1(q , w) = - - -m
2- l kFkdk 11 dcose 8 (W - -q - cose ) . (10.37)
2JrTz q ° - 1 Tzkq / m 2k
232 Electron gas: a diagrammatic approach
The integral over cose vanishes if (nk:/m- ~ r[ > 1, and is equal to unity if
----
-
m
2nli 2q
1k
° [
F
ke
kF
1--
w q
? ( - - -
2 k qv F 2kF
)2] dk
(10.38)
10 2
1° x eo - u 2 IX2)dx =
1
-
I
ftu-I xdx = 0 - u~)/2
u- > 1
u~ < l.
( 10.39)
Therefore,
mk} 2 ? nkF ? 2
B1(q , w) = --2-(1 - u _)e(1 - u=-) = -d(EF)-(1 - u=-)e(1 - u _)
4nn q 4q
(10.40)
nkF 2 ?
B 2(q , w) = -B1(q, -w) = d(EF)-(1- u+)e (1- u+J (10.41)
4q
(10.44)
(10.45)
4n e2
(10.46)
VTF(q, w) = q2[1 + 4n e2d(EF)lq2 ]
where
(10.47)
is the square of the Thomas-Fermi wave number. At low temperatures, where elec-
trons occupy the states below the Fermi surface, d(c F) = mkF In 211 2 = kF In2e2ao,
where ao is the Bohr radius. Thus, qf.F = 4kF Inao. Since kF '"" 1 A-I in metals,
we find that q TI '"" I A-I.
In the Thomas-Fermi model, the screened Coulomb interaction in real space is
the inverse Fourier transform of VTF ,
e2
v(rl - r 2) = e-qw lr l- rzl . (10.48)
Irl - r 21
Suppose that an impurity of charge Ze is placed in a metal at the origin. The bare
Coulomb potential produced by the charged impUlity is
Vbare(r)
4n Ze
= Z el r = - -
(2n)
3
f q-
I iq·r 3
---:ye d q. (10.49)
In writing the above equation, we have used 1I r = (ll V) Lq( 4n I q2)e i qr . Since
the charge is static (w = 0), the screened Coulomb potential produced by the
234 Electron gas: a diagrammatic approach
impurity is
Vsc(r)
4n Ze
=- -
(2n)3
f e
iq r
q 2 c:(q,0)
-
d 3 q. (10.50)
The difference between the two potentials is caused by the induced charge density
Pind in the medium; hence, Poisson's equation gives
(10.51)
Using Eqs. (10.49) and (LO.50), along with V 2 eiqr = _q 2 eiq -r , we obtain
Pinder) = -Ze-
(2n)3
f [1
--- - ]
c:( q , 0)
1 eiq -r d 3 q . (10.52)
Since c:(0 , 0) = I + d(EF)Vq=O = 00, it follows that Qind = -Ze; the screening of
the charge impurity is complete. This result is reasonable. However, there is a defect
in the Thomas-Fermi model, namely that Pinder) diverges at r = O. Using Vsc(r) =
(Ze / r)e - qTl"', Vbarc(r) = Ze / r , and V2 = (l / r)3 2 / 3r2 r, Eq. (10.51) gives
Ze 2
Pinder) = -ed(EF )_e- qTFr , (10.54)
r
which is infinite at r = O. Significantly, no such singularities are observed in
experiments that probe the electronic density near charged impurities. This defect
is remedied by using the Lindhard dielectric function (see Eq. [10.43]) instead of
the Thomas-Fermi dielectric function.
nd 3 r electrons are each displaced by u(r, t), where n = N I V is the electron number
density at equilibrium. The induced dipole moment in d 3 r is -neuer , t)d 3 r; hence,
the induced polarization in the medium (the dipole moment per unit volume) is
per, t) = -enu(r, t). The induced charge density is
u(r, t) = Vl~ .
L uq(t)e,qr (10.56)
q
then
_ i ~ iq.r
(10.57)
V.u - V L q.uq(t)e
q
which is nonzero for longitudinal modes (qllu q). The induced electric field is, by
Gauss's law,
10.11.1 Plasmons
From a quantum mechanical point of view, the retarded correlation function is
(10.61)
R
D (q . w) = I1Ze ~
-1 " l(mliiqln)1 2(e-f3t" - e- f3t",)
. . (10.62)
I1w-(E IIl - E,,)+IO+
l1.m
As this expression shows, the poles of DR(q , w) are the excitation energies of
the system. Note that these energies are not the excitation energies of an added
particle; i1. q represents electron density fluctuations , and it conserves the number
of particles. The poles of DR(q , w) are thus the excitation energies of the density
fluctuation s of the electron gas. In the random phase approximation,
R IiI1 o(q , w)
D (q, w) = 11 I1 RPA (q , w) = I _ n( )
Vq O q, w
IiI1o(q, w)
(10.63)
1- Vq Re nO(q, w) - iVql m I1 0(q. w)'
The imag inary part of I1 o(q, w) gives lise to damping of the excitation modes.
To search for well-defined , long-lived excitations, we consider the region where
1m I1 o(q , w) = O. This occurs when W/qvF > I + q /2 kF (see Eq. [10.42]). The
poles are obtained by setting 1- vqRenO(q, w) = O. We evaluate ReI1o(q , w) in
the low temperature limit (k8T «EF), long wavelength limit (q « k F ), and high
frequency limit (w »qv F)' The expression for Re n O(q , w), given in Eq . (10.31),
can be written as
Re n °(q, w)
2
[1
= -d(EF) - - 1-(w/qvF -q /2kr) 2 In - -
4q / kF 1 - x+
II+LI
+ 4q/kF
+
I - (W / qVF q /2kF )2 In X+
1- L
II+ I] (10.64)
where
qVF
x± = - (I ± q /2k F ).
w
In the high frequency, long wavelength limit, x± « I . By expanding
In 11 +I
x = x - -
x2
2
+- -- +- -- + ...
x3
3
X4
4
x5
5
x6
6
and carrying out tedious calculations, we find
11 (q) 2 [
Re n o (q, w) = m;:; I 3 (q--;-
+5 VF ) 2 +. .. ] . (10.65)
Another method for obtaining the above result is outlined in Problem J0.4.
J O. J J Plasmons and Landau damping 237
:::} w
2
= 2 [
w p 1+ 5
3 (QVF)
~ 2 +. . . ] = 2 [
Wp I+ 5
3 (qVF)
Wp
2 + . . .]
2 2
? W q q VF q VF
u=- < 1 :::} -1 < - - - - < I:::} -qVF +- - < w < qVF +--
qVF 2kF 2kF 2kF
:::} (q l kF)2 - 2q l kF < nw / EF < (q l kF)2 + 2q l k F.
In the shaded region of the q-(V plane (see Figure 10.12), lmno(q , w) -# O. The
plasmon mode dispersion is also shown . For q > qc, the plasmon mode is damped,
and it becomes difficult to observe due to its short lifetime. This damping is
known as Landau damping. The shaded region is the region of single-patticle
excitations, whereby an electron below the Fermi surface is excited to above the
Fermi surface. Outside this region , it is not possible to conserve energy and wave
vector in a single-particle excitation process . We can understand the situation as
follows. Suppose an external field with wave vector q and frequency w impinges
on a metal at low temperature. Under what circumstances would it be possible for
an electron to absorb momentum nq and energy nw (supplied by the field) that
would allow it to move from beneath to above the Fermi surface? For any given
q, the maximum energy that can be absorbed corresponds to a transition in which
an electron at the Fermi surface with wave vector kllq, Ikl = k F , transitions to a
state with wave vector k + q, where Ik + ql = k F + q (depicted in Figure 1O.13a).
The absorbed energy isnw = n 2 q212m + !i 2k F q I m. Ifl1w > 1i 2q2/2m + !i 2 k Fq 1m
(corresponding to points to the left of the left-hand parabola in Figure 10.12), then
conservation of energy is not possible for any single-~article excitation.
238 Electron gas: a diagrammatic approach
Figure 10.12 Plasmon damping : the shaded region is the region of w - q plane
where single-particle excitations are possible. For x > qcl kF, the plasmon decays
by exciting electron-hole pairs.
(a) (b)
Figure 10.13 (a) A single-particle excitation in which maximum energy is
absorbed, and (b) a single-particle excitation in which minimum energy is
absorbed.
Similarly, for any given q, the minimum energy that can be absorbed in a
single-particle excitation corresponds to a situation where an electron at the Fermi
surface, having a wave vector k in a direction opposite to that of q, transitions
to a state with wave vector k + q, Ik + ql = q - kF (depicted in Figure IO.13b).
The absorbed energy is liw = (Ii 2 12m)(q2 - 2kFq). If liw < li 2q2/2m -li 2k Fq I m
(corresponding to points to the right of the right-hand parabola in Figure 10.12),
then no single-particle excitation is possible. Clearly, if q < 2k F , the minimum
energy absorbed is zero.
/0.12 Case study: dielectric fun ction ofgraphene 239
We conclude that if an external field with wave vector q and frequency w were
to sttike a metal, where the point (q j k F, liw j E F) lies outside the shaded region
shown in Figure 10.12, then the energy and momentum carried by the field could
not be absorbed through single-particle excitations . If the field's wave vector and
frequency were to match those of the plasmon mode, then the plasmon mode would
be excited.
(10.67)
The minus (plus) sign refers to the valence (conduction) band, k is measured from
K (or K'), and k = Ikl. In undoped, pure graphene at low temperatures, the valence
band is full while the conduction band is empty. We assume that q is small, so that
we can ignore intervalley scattering. The dielectric function is given by
(10.68)
°
D (q, r) = - A1 (T n(q , r)n( -q , 0) )0. con n. no ndi sjoint (10.69)
where A is the area of the system. Consider the valley near K (or K ' ). The number-
density operator (see Problem 3.6) is given by
Here, sand s ' are band indices: s, s ' = v (valence) or C (conduction). Note that
Thus,
° _
D (q , r) - - A1 (T n(q, r)n t (q,O))O.c.nd
= - ~ L L (1/r~le-iqrl1/r~~q)(1/r~' le-iq.rl1/r(+q) *
ka ss' k'a 'rr'
(10.73)
-I L
f3n D °(q, W
Ill) .
e -I(V", T -
- 1
(f3n)2A L L Fss' ek ,q)
In ko 5 S'
nil'
(10.75)
where
1( ,k + q
FSSI (k, q) = - 1 + ss - - - -
COS¢) (10.76)
2 Ik+ ql
(see Problem 2.6). Here, ¢ is the angle between k and q, and s, s ' = + 1( -1) if
s, s ' = c( v) . It follows that Wil ' = WI/ + Will' The summation over n was carried out
in Section 10.6; we therefore have
D (q,
° Will)
1 ""
= A L L Fss, (k. q) . +i sk( - i s'k+q
)/11 . (10.77)
ka 5S' I Will E5k - Es' k+q 1
10.12 Case study: dielectric ful7 ction ofgraphene 241
n O( q, w) -- ~ ~
~ Fss' (k)
, q
f sk - / s' k+ q
. +. (10.78)
A ks s '
1iw + Esk - Es' k+ q + to
A factor of 2 arises from the existence of two valleys, and another factor of 2
arises from summing over the spin index; hence we have a factor of 4 in the above
equation. Denoting f ck by A + , f vk by A -, Eck by Ek+ , Evk by Ek - , and summing
over band indices, we obtain
We restrict our calculations to the case of undoped, pure graphene at low tempera-
tures. Under these conditions, the conduction band is empty and the valence band
is full: fk+ = fk+q+ = 0, and A- = fk+q- = l. Hence nO. + (q, w) = 0, and
° - ~
lmn (q w) = -2Jr ( k + q cos¢> )
~ 1 - --=----
, A k Ik + ql
x {8 [nw -1ivF(k + Ik + ql)] - 8 [nw + 1ivF(k + Ik + ql)]}
(10.81)
where we assume that q is small, so that the linear energy dispersion will be a good
approximation. Since 8(x) = 8(-x), the above expression implies that
(10.82)
242 Electron gas: a diagrammatic approach
(10.83)
It is thus sufficient to evaluate n O(q , w) for w > O. In this case, the second Dirac-
delta function in Eg. (10.81) vanishes, and we end up with
Lk
~ -A-2
(27T)
f kdk 1 2JT d¢.
°
Consider the argument j(cos¢) of the Dirac-delta function
( 10.85)
(10.86)
and
Using
8[x - XI]
8[f(x)] = L laf/axl
I
x
; ,
(10.88)
Since -I ::: cos¢ ::: 1, for the Dirac-delta function to be nonzero, we should have
-I ::: (w 2 - 2VFkw - v}q2)/2v}kq ::: l. This is satisfied if the following two
conditions are satisfied:
r
Jo
2rr
d¢··· =
r
Jo d¢· . . +
1 rr
2rr
d¢··· = -
l ¢=rr
¢=o
d cos¢
sin¢' . . -
1¢=2rr
¢= rr
d cos¢
sin¢" . ,
(10.89)
and that sin¢ > 0 for 0 < ¢ < n while sin¢ < 0 for n < ¢ < 2n; hence
Jo
r2rr
d¢··· =
j_I
1 d cos¢
Isin¢1 ... + _\
j 1 d cos¢
Isin¢1 ... = 2
j_I
1 d cos¢
Isin¢1 ... (10.90)
(10.91)
°
1m n (q, w > 0) = -
e(w - vFq) ! ? 2
[vFq - (w - 2v Fk) ]
2 1/ 2
n/1VFJw2 - v}q2
The step functions ensure that conditions (a) and (b), which were given earlier, are
satisfied. Condition (b), enforced by the step functions inside the integral, implies
that w j 2vF - q j 2.:s k .:s w j 2VF + q j 2. Thus,
q 2 ecw - VFq)
ImnOCq,w>O)=- . (10.94)
4Ii J w2 - v}q 2
(10.95)
244 Electron gas: a diagrammatic approach
RenO(q, w) = - -
q2
4rrn
100
'
-===== -
/
dx
=
q2
--4rr-h
J
. ( 10.96)
!'j. q2_ w2 x V X + w2 - V}q2
J=
(10.97)
Thus,
(10.98)
and
(10.99)
For w < 0
(10.100)
The dielectric function is
2rre 2
s( q, w) = I - --no(q, w). (10.101)
q
Further reading
Bruus, H. and Flansberg, K. (2004). Many-Body QuanlUm Theory in Condensed Matter.
Oxford: Oxford University Press.
Fetter, A.L. and Walecka, J.~ . ( 1971 ). Quantum Theory of Many-Particle Systems. New
York: McGraw-Hili.
Problems 245
Mahan , G.D. (2000). Man y-Particle Physics, 3rd edn. New York: Kluwer AcademiclPlenum
Publishers.
Mattuck, R.D. (1976). A Guide To Feynman Diagrams in the Many-Body Problem, 2nd
edn. New York: McGraw-Hill.
Problems
10.1 First-order self energy. Show that the first-order contribution to the self
energy of an electron in an electron gas as T ~ 0 is given by
*
~ (ka w ) - - -1 L 4rre
2
fik' .
1 ' 11 - Ii V k' Ik - k' 12
10.3 High frequenc y limit of c( q, w). Show that the high frequency limit of the
dielectric function of an electron gas is given by
R e n o( q,w ) = ~ '"'
~
A(Ek+q - Ek)
?
V k (liW) 2 - (Ek+q - Ek)-
°
Re n (q, w) = 4 2 '"' [ ( E k+ q - ?Ek) 2
~ A(Ek+q - Ek) 1 + + . .. ] .
V (liw) k (liw)-
246 Electron gas: a diagrammatic approach
10.5 Thomas-Fermi wave number in two dimensions. Show that, in two dimen-
sions, qTF = 2lao, where ao = /1 2 1me 2 is the Bohr radius.
In this chapter we turn to phonons, photons, and their interactions with electrons.
These interactions play an important role in condensed matter physics. At room
temperature, the resistivity of metals results mainly from electron-phonon inter-
action. At low temperature, this interaction is responsible for the superconducting
properties of many metals. On the other hand, the electron-photon interaction plays
a dominant role in light scattering by solids, from which we delive a great deal of
information about excitation modes in solids. Much of our knowledge about energy
bands in crystals has been obtained through optical absorption experiments, whose
interpretation relies on an understanding of how electrons and photons interact.
We begin by discussing lattice vibrations in crystals and show that, upon quanti-
zation, the vibrational modes are described in terms of phonons, which are particle-
like excitations that carry energy and momentum. We will see that the effect of
lattice vibrations on electronic states is to cause scattering, whereby electrons
change their states by emitting or absorbing phonons. Similarly, the interaction of
electrons with an electromagnetic field will be represented as scattering processes
in which electrons emit or absorb photons.
A discussion of lattice vibrations in the general case of a three-dimensional
crystal with a basis of more than one atom is somewhat complicated. To keep the
presentation simple, we consider in detail the simplest case, a one-dimensional
crystal with only one atom per unit cell. Next, we consider a diatomic chain, and
then indicate briefly how things look in three dimensions. The reader interested in
a treatment of the general case of a three-dimensional crystal with more than one
atom per primitive cell will find a detailed presentation in Appendix C.
247
248 Phonol1s, photons, and electron s
k
(a)
Rn-2 Rn - 1
(b)
+I ~
U n -2 Un-l
(J 1.1)
-7f/a 7f/a q
°
Figure 11.2 A plot of w vs. q for values of q in the first Brillouin zone. The crystal
is one-dimensional. with one atom per primitive cell.
The relation between OJ q and q is known as the dispersion relation. We note the
following:
1. Periodic boundary conditions, applied to Eq. (11.2), give the allowed values for
q, namely
The second remark implies that it is sufficient to restrict the values of q to the first
Brillouin zone (FBZ): -n / a < q :::; n / a. The number of normal modes is equal to
the number of q-points within the FBZ, which is exactly equal to N. Since L » a ,
the first remark means that the separation between neighboring values of q is too
small compared to the width of the FBZ; hence, when plotting OJ vs q , we may
consider q to be continuous. Such a plot is shown in Figure 11.2. Regarding the
third remark, the fact that OJ --7 0 as q --7 0 is obvious on physical grounds: as
q --7 0, neighboring atoms undergo equal displacements during the vibration, and
the restoring forces vanish. The fact that OJ approaches zero linearly in q in the long
wavelength limit (q --7 0) assigns the name "acoustic branch" to the branch in the
dispersion in Figure 11.2; v is the speed of sound in this one-dimensional crystal.
The general solution of the equation of motion, Eq. (11.1), is a linear combination
250 Phonons, photons, and electrons
Un =
I
r>:TiI:i
L .
Qqlqe ,
RII
( 11.4)
vNM qEFBZ
where the factor e- iuv is absorbed into the expansion coefficients Q q and the
factor 1/ J N M is inserted for later convenience. In effect, Eq. (11.4) is a Fourier
expansion of the displacement Un . The expansion coefficients Q q are called normal
coordinates. They sati sfy the relati on Q ~ = Q _q , which is a consequence of the
fact that the displacem ent U n is real.
Our next task is to construct an expression for the energy of the line of atoms in
terms of the normal coordinates. The kinetic energy is given by
N
T = (M / 2) ""' 2
L Li n = _1_
2N ""'
LL ""' Q
' q Q'q ,ei (q+q') R" ,
11 = 1 1/ qq'
T = ( 1/ 2) L QqQ-q. ( 11.5)
q
UI/+1 - Un -
_ I ""' Q qe iq R,, ( eiqa - I) .
r>:TiI:i L
vNM q
V= 2:M L L Q q Q q' (e
iqa
- l)( eiq'a -
i
l)e (q+q ') R" .
1/ qq'
(11.9)
The quantum theory of lattice vibrations of the one-di men sional monatomic crystal
is obtained by treating the dynamical variables Qq and P" as operators that satisfy
the commutation relations
(1Lll)
Analogous to the case of the harmonic oscillator (see Section 1.2), we introduce
two new operators,
aqt -_ ("t;
UIW q)- 1/2 (Wq Q _q - I.pq.
) (11.12)
Figure I 1.3 A linear diatomic chain with lattice constant a. The two different
atoms have masses M\ and M2. Neighboring atoms are connected by springs of
force constant k. In unit celln, the displacements from equilibrium of the atoms
of masses M \ and M2 are U II and Un . respectivel y.
(11.15)
(11.16)
- vei(qR,,-w</) (11.17)
Vn - ,
where RI/ = na. Inserting these solutions into Eqs (11.15) and 01.16), we obtain
the following homogeneous algebraic equations involving u and v:
(11.20)
where J-L = Ml M2 /( M\ + M2) is the reduced mass of the two atoms in the unit
cell. A plot of w vs q reveals that the dispersion curves consist of two branches
(see Figure 11.4), The lower branch is the acoustic branch, while the upper one is
11.2 One-dimensional diatomic lartice 253
o q
Figure 11.4 Dispersion curves for a linear di atomic chain with lattice constant a.
The lower (upper) branch is the acoustic (optical) branch.
ka 2
W = q (acoustic), W = J 2k / f.1 (optical). (11.21)
2(M j + M 2)
At the Brillouin zone edges (q = ±rc/ a) we find, based on Eq. (11.20), that
2k
W= (acoustic), W= (optical)
(11.22)
where max(M j , M 2) is the larger of M j and M2, and min(M[ , M2) is the smaller
of the two masses . At q = 0, Eqs (11.18), (11.19), and (11.21) give
At the Brillouin zone center (q = 0), all atoms vibrate in phase in the acoustic mode,
undergoing equal displacements; the vanishing of the frequency results from the
absence of any restoring forces. In the optical mode, on the other hand , adjacent
atoms vibrate out of phase (see Figure I J.S). We note that the optical mode is
excited by infrared light, hence the name "optical mode."
If we were to construct a quantum theory of lattice vibrations for the diatomic
chain, we would find the following Hamiltonian :
2
H = L Lliwq),(aJAaq), + 1/2) . (11.24)
qEFBZ A=l
254 Phon ons, photons, and electrons
Acoustic mode
Optical mode
Figure 11.5 The acoustic (upper figure) and optical (lower figure) modes at the
Brillouin zone center (q = 0) of a linear diatomic chain.
Here, the index "A refers to the phonon branch. There are two branches, an acoustic
one and an optical one. The operator a~A (a q A ) creates (annihilates) a phonon of
wave number q , branch index "A, and energy fIw qA .
The time-dependent coefficients Qq ), are called normal coordinates. Since U~I = U"I
(displacements are real) , it follows that Q~), = Q - qA and f ~)* (q) = f ~) ( -q). The
Hamiltonian can be expressed in terms of the normal coordinates QqA and their
11.4 Phonon statistics 255
conjugate momenta PqA:
(11.27)
(11.28)
Pq ), = .Jl1W
I -2-
qA t
(a qA - a_q A)' (1l.29)
Note that Q - qA = Q~i.. The new operators satisfy the commutation relations
[aqA, aq'A' ] = [a~A ' a~'A' ] = 0, [aq)" a~'A' ] = 8qq ,8H , . (11.30)
(11.32)
As expected, phonons obey Bose-Einstein statistics. The important point here (and
the reason for going through the derivation) is the absence of a chemical potential:
fJ, = 0, as Eq. (1 1.32) indicates. The vanishing of the chemical potential results
from the fact that the number of phonons in the system is unrestricted: an arbitrary
number of phonons can occupy a normal mode (q)..).
II n n
J J.5 Electron-phonon il7feraction: rigid-ion approximation 257
,, - ''0''
'
/ \
k'Y,
,
...
I
\I "\
,
... "
I
\I
.h
000 (a)
J
\
"
O OO
,
-...'
,
I
\
I
I
\
,"
,
...
'
(b)
\
I
,',
,
\
,
... ...
J
\
r
Figure 11.6 (a) An electron is in the Bloch state In ka }. As long as the lattice is
static. the electron remains in this state. (b) In the presence of a phonon , the ions
in the lattice vibrate, and the electron sees a different ionic potential than it does
in (a). The change in the potential energy of the electron acts as a time-dependent
perturbation that can scatter the electron into the stationary states In'k' a}.
V(rj - RII - "II) is the interaction energy of the electron with the ion in unit cell
n when the ion is displaced from equiliblium by Un. The gradient is with respect to
the electron coordinates. The above equation is no more than a Taylor expansion of
Ve -. to first order in the displacement; higher orders are ignored due to the smallness
of the ionic displacement as compared to the spacing between neighboring ions.
The first term in the expansion is the periodic potential energy which results from
the interaction of the electron with the static ions at their equilibrium positions;
when combined with the electron's kinetic energy, this term gives rise to the Bloch
Hamiltonian whose eigenfunctions are the Bloch functions. The second term, when
summed over all electrons, is the electron-phonon interaction:
(11.34)
In this expression, the sum over the electrons is identified as a one-body operator
of the form L j h(r j); its second quantized form is
L f A.VV(rj - R II ) =
j
L(! 1/J~'a(r)fA .vV(r
k~ a
- Rn)1/Jka(r)d
3
r) C~' aCka
(11.35)
258 Phonons, photons, and electrons
where 1fka(r) and tPk'a(r) are Bloch functions. In writing the above equation, we
have assumed that there is only one partially filled band (as is the case in simple
metals) and that electrons scatter within this band; we thus ignored the band index
when writing the Bloch functions. Furthermore, no spin flip can occur when an
electron is scattered by lattice vibrations.
We recall that, according to Bloch's theorem (Section 2.3),
(l1.36)
We can take advantage of this property: in the integral in Eq. (L I .35), we replace
the integration variable r with r + R II ; then
Here,
We can proceed a bit further if we adopt the effective mass approximation: tPk(r) ~
(l / .JV)eik.r , Ek ~n 2 k2 / 2m*, where V is the volume of the crystal and m* is
the effective electron mass. This approximation is adequate for simple metaLs.
Expanding VCr) in a Fourier seIies,
p p
and wntmg Qq), in terms of phonon creation and annihilation operators (see
Eq . [11.28]), we finally obtain
H e- phonon=- ~LLL
ka q), G
J 1.5 Electron-phonon interaction: rigid-ion approximation 259
k+q+G k+q + G
qA -qA
k k
(a) (b)
Figure 11.7 A pictorial representation of electron-phonon interaction. (a) An
electron is scattered from state Ika ) into state Ik + q + Ga ) by absorbing a
phonon of wave vector q and branch index A. (b) Here, scattering occurs by the
emission of a phonon with coordinates (-qA).
(1) The interaction is seen to be a sum of terms, with each term representing a
scattering process in which an electron is scattered from state Ika) into state
Ik + q + Ga) by either emitting or absorbing a phonon. The scattering process
is depicted in Figure 11.7.
(2) The wave vectors q, k, and k' = k + q + G must all lie in the first Brillouin
zone (FBZ). Hence, in summing over G , there is only one term in the summation
for any fixed k and q; G is the one reciprocal lattice vector which, when
added to k + q, canies it back into the FBZ. If k + q E FBZ, then G = 0;
otherwise, G -# O. Electron scattering processes (by the emission or absorption
of a phonon) for which G = 0 are called normal processes. A process for
which G -# 0 is called an Umklapp process. Normal and Umklapp processes
are depicted in Figure 1l.8.
(3) We restlict further discussion to normal processes only: G = O. The factor
q.f ),(q) in Eq. (l] .40) implies that electrons interact only with longitudinal
phonons. In isotropic media, phonon polarization vectors are actually either
longitudinal or transverse.
(4) We write the electron-phonon interaction in the following form:
q
q k+q+G k+q
~k+q
k
k
(a) (b)
Figure 11 .8 Normal and Umklapp processes in a two-dimensional square lattice
oflattice constant a. The square shown in the figure is the FBZ; its side is 2n / a.
In the scattering process, an electron of wave vector k absorbs a phonon of wave
vector q or emits a phonon of wave vector - q. (a) The wave vector k + q E
FBZ; the scattering process is normal. (b) k + q ¢: FBZ; a reciprocal lattice vector
G must be added to carry k + q back into the FBZ, so the scattering process is
Umklapp.
M~), = M_ qA . ( 11.42)
Let us consider the case of an isotropic medium (a cubic crystal, for example)
with only one atom per unit cell. In this case, there is one longitudinal acoustic
branch and two transverse acoustic branches. Electrons interact only with the
longitudinal phonons. Assuming that the crystal is a metal, the Coulomb potential
of the ions is screened by the conduction electrons. Since ions move very slowly
compared to electrons, we can assume that the screening is static, similar to the
screening of a fixed charged impurity. The screened Coulomb potential of an ion
is thus taken to be 4rr Ze j q 2c(q , 0), where Z e is the ionic charge and c(q , 0) is
the static dielectric function. For small values of q, this is given by 4rr Z e j(q 2 +
qj'F ) ~ 4rr Z ejqJF' where q TF is the Thomas-Fermi wave number. The electron-
ion interaction energy thus has the Fourier component Vq = -4rr Z e2 j qJ F' Under
these assumptions, it follows from Eq. (11.40) that the electron-phonon interaction
matrix element depends on q only and is given by
(11.43)
Note that because EA( -q) = Er(q), Mq satisfies the relation M~ = M_q, as it
should. Often in the literature q .EA(q) is replaced by q, which is the magnitude
of q ; this is not quite accurate, for then the equality M~ = M_q is not satisfied.
Note further that Mq=o = 0 : in a normal mode, with q = 0, the peliodicity of the
J J.6 Electron- LO phonon interaction in polar crystals 261
lattice is preserved, Bloch states are still the stationary states of the system, and no
scattering takes place.
(1) There is no electric field associated with transverse optical (TO) modes.
(2) The electlic field associated with a longitudinal optical (LO) mode exelts a
restoring force on the ions, in addition to the short-range restoring forces that
are present in the absence of an electric field . The result is that the LO-phonon
frequency is higher than the TO-phonon frequency . In the long wavelength
limit (q ---+ 0), we find
2 £(0) 2
W LO =- -W '
£(00) TO
(11.44)
The prime on the summation indicates that the q = 0 term is excluded, a~(aq)
creates (annihilates) an LO-phonon of wave vector q, and C;ka (Cs ka ) creates
(annihilates) an electron in state Iska ), where s is the band index. The matrix
element M~s' is given by
M~S'
1- - -I] 1/ 2 ( 2h 2 ) 1/ 2 .
+ qa le1qr
= iWLO
[-
E(oo)
-
E(O)
7r2 e
Vq WLO
f L(q).q (s'k lska ).
(11.46)
262 Phonons, photons, and electrons
Here, V is the crystal volume, q is the unit vector in the direction of q, and
€L(q) is the LO-phonon unit polarization vector.
If we assume that electron scattering by phonons takes place in only one
band, the sum over sand s' in Eq. (11.45) is no longer there. If we also
approximate the Bloch functions by plane waves, the matrix element (s 'k +
qaleiq.rlska) becomes equal to unity.
11.7.1 Definitions
The phonon retarded Green's function is defined by
For phonons, canonical and grand canonical ensembles coincide because the chem-
ical potential vanishes, as discussed in Section 11.4.
The phonon imaginary-time (Matsubara) Green's function is defined by
Here, </>q).. (r) = eHr /li</>q).. (O)e - Hr /fi , and T is the time-ordering operator,
r > O
(11.51)
r < O.
11.8 Free-phonon Green's function 263
No minus sign is incurred upon interchanging the operators, since they are bosonic.
We focus our attention on the imaginary-time Green's function; from it, the retarded
function can be obtained by analytic continuation.
11.7.2 Periodicity
As discussed in Chapter 8, the time r is restricted to the interval [- {3h, (3h]. Since
¢q), is a bosonic operator, it follows from the results of Chapter 8 that the phonon
Green's function is periodic:
(1l.54)
0l.55)
wherea qic (r) = eHr /lia qA(O)e - Hr /Ii and at- qic (r) t (O)e- Hr /Ti . Takingthe
= eHr /lia -qic
derivative with respect to r, we obtain
Inserting these into Eq. (11.57), and noting thatthe tenns (aqAa_qA)O and (a~qAa~A)o
vanish, we obtain
The commutation relation between the phonon annihilation and creation operators
implies that aqA a~A = 1 + a~A GqA' Moreover, in thermal equilibrium, (a~AaqA)o is
the occupation number nWqA of the normal mode (q'A), given by Eq. 01.32). Since
w_ qA = wq )" we can write
(11.59)
The observation that 1 + n WqA = -n - WqA (easily verified) allows us to write the
above expression in another way:
(11.61)
Before proceeding to calculate the free-phonon Green's function, let us rewrite the
above expressions for dO > and dO < in the following way:
(11.{)2a)
(11.62b)
where
(l1.63a)
(l1.63b)
This is calculated by inserting the expression for dO> (q'A, r) from either Eq. (11.60)
or Eq. (11.62a); the result is
°
d (q'A, Will) = (.
2Wq A
)2 2' (11.64)
{Will - WqA
11. 9 Feynman rules for the electron- phonon interaction 265
qA qA qA ,wm qA ,wm
L.
r- ~
~~=~ r- ==::r===
T 0 T 0
(a) (b) (c) (d)
Figure 11.9 Pictorial representation of the phonon Green 's function: (a) d O(qA, r) ,
(b) d(qA, r) , (c) dO(qA, w m ), and (d) d(qA, w",). Here, dO is the noninteracting
(free) phonon Green 's function , while d is the interacting function.
qA ,Wm
Q
k'cr',w n,
(1) At order 2n in the electron-phonon interaction (since only even orders survive),
draw all topologically distinct diagrams with n phonon lines, two extemal
electron lines, and 2n - 1 intemal electron lines.
(2) To each electron line of coordinates (kcr, wn ), assign gO(kcr, w n ) .
(3) To each phonon line of coordinates (qA, W m ), assign IMq A I2do(qA , W/1I)'
(4) At each vertex, conserve wave vector, frequency, and spin.
(5) Sum over all internal coordinates.
(6) Multiply each electron loop by -1.
(7) Multiply by the factor (I Iii )2n ( -1 1(311 Y.
For example, consider the two diagrams that arise in second-order perturbation
in the electron-phonon interaction (see Figure 11.10). In the first diagram,
266 Phonons, photons, and electrons
+.q,+'CC9' + ...
Figure 11.11 Some representative diagrams in the perturbation expansion of the
electron Green's function. The perturbation is the electron-phonon interaction.
conservation of wave vector at the vertex implies that the phonon line has zero
wave vector. However, the term q = 0 is absent in the electron-phonon interaction,
and this diagram should be excluded. Using the Feynman rules, the contribution of
the second diagram in Figure 1] .10 is
og(ka, (J)/I) = I3
-- [0
g (ka, (J)/I) ]2""
L..,./Mq ), /2g 0(k - qa, (J)/1 - (J)1II)d 0(qA, (J)m).
{J11 qAm
(11.65)
The first term describes a collection of electrons in the conduction band of a metal;
interactions among the electrons are taken in an average way, with the effect being
simply a renormalization of the electron mass. The second term is the Hamiltonian
for a system of non interacting phonons, and the third tel111 is the electron-phonon
interaction, with the term q = 0 excluded (Mq =O. A = 0).
The perturbation expansion of the electron Green's function is depicted in
Figure 11.11, where some representative diagrams are shown. The last two dia-
grams in Figure 11.11 are, in fact, similar to the one-phonon diagram (the second
11.10 Electron self energy 267
Figure 11.12 Diagrams that can be added to produce a single diagram with a
corrected vertex.
+)0.:
---i>;;'-'-+~+~+ ...
Figure 11.13 Electron and phonon Green's functions for a system of coupled
electrons and phonons. Vertex corrections are ignored. The electron proper self
energy 2: * is approximated by replacing full (dressed , or interacting) electron and
phonon propagators with bare (noninteract ing) propagators.
one on the RHS in the figure), except for some vertex corrections, as shown in
Figure 11 .12. A remarkable theorem, due to Migdal (Migdal, 1958), states the
following:
(11.66)
Figure II.l4 The two time-ordered diagrams that are used to calculate the electron
self energy that is due to electron-phonon interaction. The external lines have
coordinates (ka, w lI ) . The intern al electron line ha s coordinates (k - qa. E\). while
the internal phonon line has coordinates (qA. E2) ' The horizontal dashed line is a
section.
relegate this approach to the Problems section. Here, we calculate the self energy
using Dzyaloshinski's rules for time-ordered diagrams (see Section 9.8). There are
two time-ordered diagrams (see Figure 11.14). The self energy is given by
( 11.67)
(see Eqs l6.SS], [8 .32], and [8.34]). The phonon spectral functions are given in
Eq. (11.63). Inserting these into the expression for 2; *, and noting that i1Wq)
-1 - n -Wq!. ' we find that
( 11.68)
11.11 The electromagnetic field 269
The retarded self energy is obtained by replacing iWI1 with W + iQ+:
1m ~~(ka, w) = ~: L IMqAI2 [(n Wq ) + A-q)o(w - Ek-q/li + w qA)
qA
(11.69)
The first (second) term in the brackets conesponds to phonon absorption (emission).
The lifetime of an electron in state Ika) is given by
-1
Tka = . (11.70)
21m L~(ka , Ek/li)
In writing Eq. (11.70), we have replaced W in ~~ with Ek/li; this is an approxima-
tion. In fact, liw should be replaced with the shifted energy, which is obtained by
solving the equation w - Ek/li + Re ~~(ka, w) = O.
V.E=O (11.71a)
V.B =0 (ll.71b)
1 aB aB
VxE=--- (cgs) , VxE=-- (SI) (11.71c)
c at at
I aE aE
vx B =-- (cgs), vx B = fJvoEo- (SI). (11.71d)
c at at
In the following treatment, we use the cgs system of units. The second and third
Maxwell's equations are automatically satisfied if we express E and B in terms of
a scalar potential <P(r, t) and a vector potential A(r, t):
laA
E=-V<P--- , B=VxA. (11.72)
c at
This is because the divergence of a curl is zero (V .V x A) and the curl of a gradient
is zero (V x V <P = 0). The first and fourth Maxwell's equations are now written
as
1 a
V2<p + --(V.A) = 0 (11.73)
cat
1 1 2
V x V x A = ---V<P - -2- A
a
.
a (11.74)
c at c at 2
270 Phonons, photons, and electrons
I 1 aA
A -+ A ' = A + V A , <P -+ <P = <P - - -
c at '
where A(r, t) is any smooth function, both E and B remain unchanged. Choosing a
particular function A(r, t) is called "fixing the gauge." In one gauge, called the radi-
ation gauge, A(r, t) is chosen such that V 2A(r , t) = - V.A and (1 / c)(aA/ at) = <P.
Thus, in the radiation gauge, <pI = 0 and V.AI = O. Relabeling (<P', AI) as (<P, A),
we can write
laA
E = ---, B =V x A, (11.77)
c at
where A satisfies the wave equation
? I 2A
V -A- - - =0.
a (11.78)
c2 at 2
This equation is to be solved subject to the constraint V.A = O. We assume that
the electromagnetic field is enclosed in a large cube of volume V = L 3 , and that it
obeys periodic boundary conditions. Our approach is similar to the one we followed
in studying atomic vibrations: we first find the normal modes and then wlite the
general solution as a linear combination of these modes. The normal modes are
given by
where €).(q) is a unit polarization vector. The requirement V.A = 0 ::::::} q.€ ).(q) =
0; the normal modes are transverse modes, so A = I , 2. Inserting the above expres-
sion into the wave equation, we find that the equation is satisfied if Wq). = cq,
independent of A; henceforth, we write Wq and drop the subscript A. The periodic
boundary conditions imply that the allowed values for q are
A(r. t) = (
4
:C2) 1/ 2
q
2
L L Aq}J ). (q)e iq .
).= 1
r
. (11 .79)
The factor e- i wql is absorbed into Aq). , which satisfies the equation
.. 2
Aq). = -wqA q)..
The expansion coefficients Aq). are the normal coordinates of the electromagnetic
field. They are similar to the Qq ). coefficients that appear in the expansion of the
ionic di splacements in a crystal. The additional factor (4iTC 2 )1 /2 in Eq. (11.79) is
inserted for later convenience. We note that, since A(r , t) is real, A *(r, t) = A(r, t);
it follows that A~ ). = A_ q). and E ~ (q) = E). ( -q). Orthonormality of the normal
modes implies that E ~ (q).E )., (q) = 0).).,.
The electric and magnetic fields are obtained by using Eq. (11.77):
L = 21,,( ..
~ Aq). A_q). -
2
wqAq).A _q). .
)
(11.82)
q).
where
(11.85)
272 Phonons, photons, and electrons
The quanta of the electromagnetic field are called photons. The operator b ~), (b q;...)
creates (annihilates) a photon of wave vector q and polarization A.
e) 2 + VCr).
H = - 1 ( P + -A (I 1.86)
2m C
The electron charge is -e, and V (r) is the periodic potential produced by the
lattice of ions. We ignore the electron-phonon interaction for now. Expanding Eq.
(11 .86), we obtain
l e e2 2
H = Ho + H I, Ho = p2/2m + VCr) , H = 2m c (p.A + A.p) + 2mc 2 A .
Ho is the Hamiltonian for the electron in the absence of the electromagnetic field;
its eigenstates Iska ) are characterized by a band index s , a wave vector k, and
a spin projection a. H I, when summed over all electrons, is the electron-photon
interaction. The term in H I which is proportional to A 2 involves two-photon scat-
tering processes. For weak fields, this term is generally far less impottant than the
other term which involves single-photon processes; henceforth, the A 2 term will
be ignored .
In the radiation gauge (V'.A = 0), the two terms p.A and A .p are equal. To see
this, consider the action of p.A on any function fer):
(11.87)
H e- photon = - e ""
~~ "" (s I kaIA·plska
I t
)cs'k'aCs ka '
mc .
ska s'k'
11. J3 Light scattering by crystals 273
Note that, in expanding A(r, t), we have modified Eq. (l L.79), replacing c with
c / n = c / J c:( 00 ), where n = J c:((0) is the index of refraction of the medium (the
crystal). This is because in H e- pholOn , A(r, t) is the vector potential in the medium,
where the speed of light is c / n. The matrix element on the RHS of the above
equation is evaluated using Bloch's theorem:
Here, us'k,(r) and u 5 k(r) are periodic functions having the same periodicity as the
lattice. Changing variables from r to r + R Il , where Rn is any lattice vector, and
using the periodicity property of us' k,(r) and usk(r), we find
1 = 1e- i(k'-k- q).R" .
where
(11.91)
is the field operator for the excitation mode (phonon or plasmon, for example)
of the crystal. r is the matrix element for the scattering process; it is determined
by considering the detailed mechanism through which the process takes place. In
the next section, we will calculate this quantity for the specific case of Raman
scattering by an insulating crystal.
The initial and final states of the system, which consists of the photons and the
crystal , are denoted by If) and IF ), respectively:
E I and E F are the energies of the initial and final states. Writing E I = nWI, E F =
nw F, and using
we obtain
n2 """"""
W = 2n ~ ~ ~ f'(qlAl , q2A2, qA)r '"' (qlAI , q2A2, q I A)
I
qA q'A' J
where nWi (nw J ) is the energy of the crystal's initial (final) state, and nw = nWI -
nW2 is the energy transferred to the crystal. Noting that
8(w J - Wi - w) = - 1
2n
1 00
-00
.
e -I(Wj-Wi - W) 1 dt ,
W =
nl(n 2 + 1)""
n
2 ~ !f(qjAl , q2A2, qA)1
21 00
-00
. t . iwt
(11¢qA(t)¢q A(O)ll)e dt. (11.94)
q).
The transition rate depends on the initial state of the crystal. At zero temperature,
Ii ) is the ground state IWo) of the crystal, and its energy is Eo. At finite temperature,
other states IWII ) with energy En have a nonzero probability of being occupied.
Hence, at finite temperature, the matrix element in Eq. (11.94) is replaced by the
276 Phonons, photons, and electrons
thermal average (¢qA (t)¢~A (O) ) , which is the correlation function C(qA , t) of the
excitation mode in the crystal. The integral in Eq. (11.94) then becomes the Fourier
transform of C(qA , t),
W=
I1 I (n 2
?
+1)L jr(qIAI , q2A2 , qA)I-C(qA , w).
?
(11.95)
h-
q).
Using the fluctuation-dissipation theorem (see Eq. [6.49]) which relates the cor-
relation function to the imaginary part of the retarded Green's function, our final
result is
2
d a
-- = -
(n,,; + 1)(n2 ')+ I)V 2 w~ L jr(qIAI , q2A2,qA)I-ImD
? R
(qA,W)
dwdQ 8JT~-~
q ),
(1l.96)
where nw = (efJ/i w - I )- 1is the Bose-Einstein distribution function. We have man-
aged to express the differential scattering cross-section for light scattering by a
crystal in terms of the retarded function of the crystal excitation that participates
in the scattering process. This function is obtained by analytic continuation of the
corresponding imaginary-time function which , in tum, can be calculated using the
Feynman diagram techniques developed in previous chapters.
Ho is the sum of the Hamiltonians for the electrons in the crystal, the radiation
field, and the lattice vibrations. HER is the electron-photon interaction, and HEL is
the electron-phonon interaction:
(11.100)
where 5 is an operator such that 5 t = -5. fI has the same eigenvalues as H, and its
eigenstates are obtained by the operator eS acting on the conesponding eigenstates
of H. Expanding e±s, we obtain
fI 2 2
= (1 + 5 + 5 / 2! + 5 3/ 3! + ... )H( J - 5 + 5 / 2! - 5 3/ 3! + ... )
I 1
= H + [5 , H] + 2! [5 , [5 , H]] + 3! [5. [5. [5, H]]] + ...
I I 1 1 I
= Ho + H + [5. Ho] + [5. H ] + 2'[5. [5. Ho]] + 2'[5 , [5, H]]
1
+ 6[5, [5, [5, Hom + ....
278 Phonons, photons, and electrons
In Fermi 's golden rule, the initial and final states have the same energy; setting
EF = E/ , we obtain
(11.1 04)
We have not summed over the wave vectors and polarizations of the photons
and phonons since we consider the absorption of a photon of specific coordinates
(q IAI) , the emission of a photon of specific coordinates (q2A2), and the emission
of a phonon of specific coordinates (qA) . As for the insulator (the crystal), we only
consider transitions between the highest occupied ban'd and the lowest empty band,
and we assume that the energy of the incident photon is lower than the energy gap,
so that the transitions are vit1ual processes. Now we consider the various terms in
Eq. (11.105) .
The sequence of processes that occur in the first term on the RHS of Eq. (11.105)
is illustrated in Figure 11 .16. Here, E/ = nWI , EI/I = Eck+ q ] - Evk ~ Eck - Evk
(since ql is much smaller than the width of the Brillouin zone). We also assume
that there is little dispersion in the bands such that Eck - Evk ~ E g , where E g is the
energy gap. The first term on the RHS of Eq. (11.105) is thus approximately equal to
We can simplify the notation: since the photon wave vector is very small compared
to the extent of the Brillouin zone, we cart replace P~~ q l k+q l- q2 CA2) with p~c CA2) ,
i.e., we assume that the electron-photon matrix element depends only on k. Similar
280 Phonons, photons, and electrons
Figure 1 1.16 The sequence of processes in the first term of Eq. (11.105). A photon
(ql A I) is absorbed and an electron is promoted to the conduction band c (the upper
band) , leaving behind a hole in the valence band v. The next step can proceed in
two different ways: in (a), the electron is scattered in the conduction band and a
photon (q2A2) is emitted, while in (b) an electron in the valence band is scattered
and a photon (q2A2) is emitted (we can also say that the hole is scattered). The
third step, common to both , is an electron-hole recombination accompanied by the
emission of a phonon (qA). Note that for this sequence of processes, q = ql - q2,
which is a statement of wave vector conservation.
replacements are made for other P-matrix elements. The above expression is now
written as
M~f [Pk C
()\\) + PkV(),\)] Pt9-2) Pk ()'2 ) [M~f
V
+ M~~ ] PkC(),\)
+ . + -----"=------=----
(hw\ - hW2 - Eg)( -hW2 - Eg) (hw\ - hWqA - Eg)(hw\ - Eg)
V
Pk (). \) [M~f + M~~ ] Pk ()\2) C
Pk'V ()\\) [PkC()d + Pk'V()"2)] M~f
+ +--~------~~
(hw2 +hw qA+ Eg)(hw2 + Eg) (l7w2 +hw qA+ Eg)(hwqA + Eg)
The Kronecker delta ensures that momentum is conserved. If the energy of the
incident photon is close to the energy gap (hw\ ~ E g), then the third term on the
Problems 281
RHS of Eq. (L 1.107) dominates over the other five term s; this is called resonant
Raman scattering.
In obtaining the above result for r, we have assumed that the intermediate states
consist of an electron in the conduction band and a hole in the valence band.
The electron and the hole were treated as independent patticles. In insulators, the
attraction between electrons and holes may be significant, leading to the formation
of excitons, which are bound electron-hole pairs (Kittel , 2005). A description of
Raman scattering in insulators which takes exciton formation into account can be
formulated (Ganguly and Birman , 1967), but we will stop here and not take that
road.
Further reading
Hayes. W. and Loudon, R. (200-1.). Scattering of Light by Crystals. New York: Dover.
Loudon , R. (2000). The Quantum Theory of Light, 3rd edn. Oxford: Oxford University
Press.
Madelung, O. (l978).lntroduction to Solid State Theory. Berlin: Springer.
Venkataraman , G. , Feldkamp, L.A. , and Sahni, Vc. ( 1975). Dynamics of Perfect Crystals.
Cambridge, MA: MlT Press.
Ziman, J.M . ( 1960). Electrons and Phonons. Oxford : Oxford University Press.
Problems
L1.1 Lesser free-phonon Green 's function. Derive Eq. (11.61).
11 .2 Electron self energy. Using the Feynman diagram rules, derive the expression
for the electron self e nergy, given in Eq. 01.68), due to electron-phonon
interaction .
11.3 Phon on self energy. Consider a system of electrons and phonons with a
Hamiltonian
(a) Write down the perturbation expansion for the phonon Green 's function
d (qA, r ).
(b) Using Wick 's theorem, obtain d(qA, r) to second order in the electron-
phonon interaction .
(c) Show that, to second order in the electron-phonon interaction ,
d(qA. Will) = dO(qA, Will) + ~ IMq AI 2do(qA, wlIl) n o(q. wm)d o(qA. Will)'
where n O(q, Will) is the polarizability of noninteracting electrons.
282 Phonons, photons, and electrons
Figure 1 1.17 Phonon Green 's function in the random phase approximation. The
shaded bubble is the polarizability of interacting electrons. The single wavy line
represents the bare (noninteracting) phonon Green's function, while the double
wavy line represents the dressed (interacting) phonon Green's function.
47T Ze
Mq = i - -
2
/f£1i
--q.f"L(q).
Vq 2MQ
ka q
o V 2 0
d(q, Will) = d (q, Win) + -IMql d (q, wlIl)n(q, wlI/)d(q, Will)'
n
where n(q, Will) is the polarizability of interacting electrons. Since the phonon
frequency is much smaller than the electron plasmon frequency, we are
justified in replacing n(q, Will) with n(q, 0).
Show that, as q ---+ 0 the renormalized phonon frequency is given by
W = vq. What is the value of v?
all = (~i
cat'
-v) ,
All = (¢ , A) , All = (¢, -A)
Here, the indices fJ. and j) take the values 0, 1, 2, and 3. The Lagrangian is
I
given by Ld 3 r where L is the Lagrangian density. The Euler-Lagrange
Problems 283
W
Wl
•••••• '}>" .. . ..... ..
W2 W2
.... ".. ......... . ..... ".. ..........
,
+
Figure 11. j 8 The two diagrams that describe Raman scattering in insulators. Solid
lines are electron lines , dotted lines are photon lines, and wavy lines are phonon
lines ,
~ _ 1 I1V
'-' - ---Fl1 vF
16n
(b) Show that
(12.1)
284
12.1 Properties o/sllperconductors 285
T
Figure 12.1 At the superconducting critical temperature Tc, the resistivity of a
material drops to zero.
He
Normal
Superconductor
Te T
Figure 12.2 The critical magnetic field thut destroys superconductivity varies with
temperature.
T>Te T<Te
Figure 12.3 The Meissner effect. In the nOlmal state (T > Tc ), a magnetic field
penetrates the material. In the superconduc ting state (T < Tc ), the magnetic field
is expelled from the bulk of the material.
To T
Figure 12.4 Electronic specific heat as a function of temperature in a superconduc-
tor. The dashed line is the would-be specific heat CeCT) had the metal remained in
the normal state (it is obtained experimentally by measuring CeCT) in the presence
of a magnetic field larger than the critical field).
o 0
• Cu
()) La
Figure 12.5 Crystal structure of La2Cu04. The separation between the CU02
planes is 6.6 A, Cu-O separation in the plane is 1.9 A, while it is 2.4 A perpendicular
to the plane.
where r, the relaxation time, is the average time between scattering events. The
current density is J = -nev. Under steady-state conditions, J is constant: v = 0,
and v = -eEr/m; hence J = (ne 2 r/m)E. Ohm's law, E = pJ, then implies that
the resistivity p = m/(ne 2 r).
In a perfect conductor, p = 0; the relaxation time r is thus infinite, and Eg. (12.2)
becomes
Taking the curl on both sides ofEq. (12.3), and using Maxwell's equation V x E =
(-l/e)aB /a t (cgs), we find
2
-a ( V xJ+-B
ne ) =0. (12.4)
at me
290 Superconductivity
--
- ~ (a)
11r-,
(b)
Figure 12.6 (a) A magnetic field is perpendicular to the surface of a semi-infinite
superconducting slab. No field penetration takes place. (b) The field is parallel to
the slab surface. Inside the superconductor, the field decays exponentially.
This relation, along with the Maxwell equation (for a static E-field)
4n
V x B=-J, (12.5)
e
determines the magnetic field and the cun'ent density in a perfect conductor. Note
that any static B-field determines, through Eq. (12.5), a static J, and therefore
Eq. (12.4) will be automatically satisfied. Equations (12.4) and (12.5) are thus
consistent with the existence of an arbitrary static magnetic field inside a perfect
conductor. We pointed this out in the previous section. This behavior is, however,
incompatible with the observed Meissner effect in superconductors; hence, zero
resistivity is a necessary, but not sufficient, condition for superconductivity.
It was conjectured by the London brothers that the magnetic field and current
density in a superconductor satisfy the relation
ne 2
V x J+-B=O. (12.6)
me
This is known as the London equation. Whereas for a perfect conductor the LHS of
Eq. (12.6) is only required to be time-independent (see Eq. [J 2.4]), it is identically
equal to zero for a superconductor.
Taking the curl on both sides of Eq. (12 .5) and using the vector identity
(12.7)
2 4n ne 2
VB = - -?-B. (J 2.8)
me-
Similarly, taking the curl On both sides of Eq. (12.6) and using Eq. (12 .5), we find
2 4n ne 2
V' J = 2 J.
-- (12.9)
me
We solve Eq. (12.8) for B inside a superconducting semi-infinite (z :::: 0) slab for
the following two cases, which are illustrated in Figure 12.6.
12.3 Effective electron-electron interaction 291
• B is parallel to the z-axis and varies only along the z-direction, i.e., B =
(0,0, B (z)) . In this case, V · B = 0 =} aB(z)/az = 0 =} B(z) is constant, inde-
pendent of z. Equation (12.8) then implies that B(z) = 0 inside the supercon-
ductor.
• B is parallel to the x-axis and varies along the z-direction: B = (B(z), 0, 0). In
this case, Eq. (12.8) becomes
a
2
B(z) 4nne 2
- - = --B (z), z ~ O.
az 2 me 2
Its solution is
( 12.11)
-e
(±). (±) (±) (±)
~f.;T::j!J (±) (±)
-e :.t,.
(±) (±) (±) (±) ci"':f :' ... ~
't- (±) (±)
(a) (b)
Figure 12.7 Effective electron-electron interaction. (a) An electron attracts the
positive ions in its vicinity. (b) By the time the ions are maximally displaced,
the electron is very far away, but a second electron is attracted by the result-
ing excess positive charge. [n effect, there is an attraction between the two
electrons.
very far apalt. At such distances the screened Coulomb repulsion is completely
negligible.
An alternative description of the lattice-mediated interaction between electrons
can be formulated using the language of phonons. The interaction between the
electrons and the ions may be viewed in terms of the electrons' emission and
absorption of phonons. We can consider a virtual process whereby an electron
emits a phonon. which, in turn, is absorbed by another electron (see Figure 12.8).
Since the phonon energy is typica\1y nWD , the unceltainty principle, lY.ElY.t '" n,
implies that the phonon will live for a time r '" I j wD ~ 10- 13 s. Since the typical
phonon velocity is '" 103 m/s (the speed of sound in solids), the electron that
absorbs the phonon should be very close ( '" 1 A) to the location where the phonon
is emitted.
12.3 Eff ective electron- electron interaction 293
H = Ho + H' (l2.12a)
Ho is the Hamiltonian that describes the conduction electrons and the free phonons,
and H' is the electron-phonon interaction. Here, cta (Cka ) creates (annihilates) an
electron in a state specified by the wave vector k and spin projection a, and
a~A (aqA) creates (annihilates) a phonon of wave vector q and branch index A. We
have assumed that the metal has only one partially filled band, and that electrons
scatter within this band; hence, the band index has beef! dropped.
We can obtain an expression for the electron-electron interaction mediated
by phonons by carrying out a change of basis, as we did in Section 11.14. The
second quantized form of the Hamiltonian, as given in Eq. (12.12), is obtained by
using the basis set of states In ) = Ika ) lqA ). We transform to a new basis set of
states Iii ) = Uln ), where utu = 1. In the new basis, the matrix elements of the
Hamiltonian are given by
(12.13)
(12.14)
H- = (1 - S + S2 / 2! + ... )(Ho + H I
)(1 + S + S 2 / 2! + ... )
and choosing S such that
we obtain
H_ = Ho + -1 [ H , , S ] + .... (12.16)
2
Note that, because Ho and H' are hermitian operators, the operator S, defined by
Eq. (12.15), does indeed satisfy the requirement that st = - S. Considering any
two eigenkets, 1m) and In), of Ho, with corresponding eigenvalues Em and En,
Eq. (12.15) gives
(mI H ' ln)
(m ISln) = . (12.17)
EI/ - Em
Since the effect of H' is to scatter an electron from a state with energy Ek into
a state with energy Ek+ q either by the absorption of a phonon (qA) or by the
emission of a phonon (-qA), the energy difference En - Em is either Ek + hWqA -
Ek+ q (corresponding to phonon absorption) or Ek - fzw q ), - Ek+q (corresponding to
phonon emission ). Therefore, S is given by
One can check that S, as given above, satisfies Eq. (12.15). The Hamiltonian in
Eq. (12.16) becomes
t
ck '+
q
'a,Ck'a'
(Ek' -
aq'A'
Ek'+q' + fzWq'A'
+ Ek' -
t
a_q'A'
Ek'+q' - fzWq'A'
)] .
Out of the many terms in the commutator, there are two terms that contain four
electron operators; they arise from commuting a qA with a~q'A" and a~qA with aq'A"
The other terms all contain two electron and two phonon operators. We thus write
(l2.19a)
(l2.19b)
J2.4 Cooper pairs 295
Using these, along with M _ qA = M~A ' we can write
B = Ho + HI + H2 + "others ."
where "others" are terms containing two electron and two phonon operators, and
(12.20)
(12 .21)
The term HI can be absorbed into Ho; it simply leads to a renormalization of the
single particle energy. On the other hand, the term H2 represents an interaction
between electrons. The electron-electron interaction, mediated by phonons, is thus
given by
H: = L L L
1I1 Vk'qC~+qaC~'_qa, Ck'a' Cka (12.22)
ka k 'a ' q
(12.23)
Consider a shell surrounding the Fermi surface. The inner and outer surfaces of
the shell are constant energy surfaces with energies E F - fiWD and E F + fiWD '
respectively, where E F is the Fermi energy and flw D is a typical phonon energy.
Equation (12.23) tells us that if two electrons remain in states that lie within this
shell, the phonon-mediated interaction between them is attractive.
Figure 12.9 Two extra electrons are added to the Fermi sphere of radius k F . If the
added electrons are in a shell around the Fermi sphere of width 6.k, the interaction
between the two electrons is attractive: otherwise, the interaction is zero. Here,
(T7 2 k F / m)6.k = liWD.
the role of the Fermi sea of electrons is simply to prevent the two added electrons
from occupying any state below the Fermi surface. Absent an attractive interaction,
the ground state of the two added electrons is obtained if each has energy E F. In
the presence of the attractive interaction, what is the ground state of the two added
electrons? The answer to this question was provided in a seminal paper by Cooper
(Cooper, 1956). The two extra electrons can scatter off each other from states
Ik)a) , k2a2 ) into states Ik) + qa), k 2 - qa2) . Conservation of momentum dictates
that k) + k2 = K must remain unchanged. Since the two electrons are constrained
to remain within a shell of energy width /1WD sUlTounding the Fermi sphere, the
conservation of momentum means that for a given K, the wave vectors k) and k2
will be restricted to the region of intersection of the two shells in k-space centered
on 0 and K (see Figure 12.10). Since we are interested in the lowest energy state,
it is sufficient for us to consider the case when the region of attractive interaction
is maximal; this occurs when K = 0, for then the shaded region in Figure 12.10
coincides with the whole shell. Hencefol1h, we assume that the two added electrons
have wave vectors k and -k. Denoting the positions of the two added electrons
by r) and r 2, and their wave function by 1/!(r) a), r2a2), the Schr6dinger equation
reads
(12.24)
where U (r) - r2) is the interaction energy of the two electrons; it depends on
r) - r 2 due to the translational invariance of the system. Since the Hamiltonian is
spin-independent, the stationary states can be written as the product of a spatial
function and a spin function. The two electrons are continually scattered from
states Ika), -ka2) into states Ik'a), -k'a 2); hence, we consider a solution to the
/2.4 Cooper pairs 297
Figure 12.10 kl and k2 are restricted to a shell of width 6,k sULTounding a Fermi
sphere of radius k F . For a given K , the requirement that kl + k 2 = K is satisfied
only if kl and k 2 are restricted to the region where the two shells centered at 0 and
0 ' intersect. 0 and 0 ' are two points in k-space that are separated by the vector
K. The region of intersection is the volume obtained by rotating the shaded area
in the figure around the 00' ax is .
where V is the system 's volume, (1 / V)e ik (r,-r2 ) is the spatial part of the wave
function cOLTesponding to the ket Ikal , -ka2) , i.e. , it is (rl , r2lk, -k), and the
expansion coefficients g(k) are to be determined. The spin function X (ai, a2) can
be chosen to be anti symmetric (singlet) or symmetric (triplet) . For the singlet state,
the antisymmetry of 1jr (r I ai , r 2a2) under the interchange (r I a,) ~ (r2a2) requires
that the spatial part be symmetric, i.e., g( -k) = g(k). For the triplet state, the spatial
part of the wave function is antisymmetlic, i.e. , g( -k) = -g(k). Furthermore, the
restriction of the states to a shell of energy width nWD around the Fermi sphere
implies that g(k) is nonvanishing only for E F < Ek < E F + nWD. Substituting the
wave function , as given in Eq. (12.25) , into the Schrodinger equation, we obtain
(12.27)
298 Superconductivity
( 2E k - E) g (k) + 1 '~
V ' = 0,
" Vkk'g (k)
k'
where Ua > O. For values of k and k ' such that Ek or Ek' lies outside the range
indicated above, Ukk' = O. The fact that Ukk' is negative reflects the assumption
that the interaction between the added electrons is attractive. Equation (12.28) now
reduces to
For a triplet state, g( -k') = - g (k'), and the RHS of Eq. (12.30) vanishes. Thus,
for the triplet state, E = 2Ek; the attractive interaction has no effect on the energy
of the two added electrons. For the singlet state, on the other hand, g( -k') = g(k'),
and the RHS of Eq. (12.30) does not vanish. Further analysis is now restricted to
the singlet state, in which the two electrons have opposite spins.
Dividing both sides of Eq. (12.30) by (2Ek - E) , then summing over k, we
obtain
1= ~a L 2Ek 1_ E '
k
The sum over k is a sum over states of one spin projection. Since the number of
such states in the energy range (E, E + dE) is Da(E)dE, where Da(E) is the density
of states per spin, the above equation may be written as
_ Ua l £I-+ li wD Da(E)
1- - dE.
V £ ,. 2E - E
1. No matter how weak the attractive interaction is, the two electrons form a bound
state, known as a Cooper pair, whose energy is lower than 2E F.
2. The energy E of the bound state is not an analytic function of Uo as Uo ---+ 0,
i.e., E cannot be expanded in powers of Vo. Thus, the result for E cannot be
obtained by a perturbation expansion in powers of Vo.
3. The binding energy of the Cooper pair increases as Vo increases; the stronger
the electron-phonon interaction, the larger the binding energy.
4. The binding energy increases as the density of states at the Fermi energy
increases.
(12.32)
ka kk'
I'll) = n
k
(Uk + VkC~t C~kJ, ) 10) (12.33)
where 10) is the vacuum state, Uk and Vk are parameters to be determined, and they
are assumed to be real. The state is normalized if
(12.34)
300 Superconductivity
Figure 12 .11 Scattering processes that contribute to the BCS Hamiltonian. Two
electrons in states Ik i ) and I - k-J, ) are scattered into states Ik' i ) and 1- k ' -J,) .
The matrix element for this scattering process is Ukk"
The form of the wave function implies that v~ is the probability that the pair state
Ik t. -k t ) is occupied, and u~ is the probability for it to be empty. Note that
I\}J) would describe the normal ground state if Uk = 0, Vk = I for k < kF, and
Uk = 1, Vk = °for k > k F .
If the state 1\lJ ) is expanded, we see that it is a linear combination of states with
varying numbers of pairs, i.e., 1\lJ ) is not an eigenstate of the number operator Nop
given by
This should not cause any alarm; the system is considered to be in contact with a
particle reservoir at T = 0. In other words, the system is assumed to be a member
of a grand canonical ensemble; as we know from Chapter 5, the properties of a
system may be obtained using any of various ensembles. What we require here is
that the average number of electrons, (\lJ INop1\lJ ), be equal to N , the actual number
of conduction electrons in the crystal.
Our problem is thus to minimize the energy (\}JIHI\lJ ) subject to the constraint
that (\lJ INopl\lJ) = N. This is achieved by introducing a Lagrange multiplier fJ.,
and minimizing (\lJIHI\lJ) - fJ., ( \}JINopl\}J ) = (WIH - fJ.,NopIW ) without any con-
ditions . The Lagrange multiplier fJ., is then determined by the requirement that
(WINopIW ) = N. The parameter fJ., turns out to be nothing but the Fermi energy
E F. Defining Ek = Ek - fJ." we can write
Reintroducing Uk and Vk: sin(2ek) = 2Uk Vk, cos(2ek) = U~ - v~, the above equa-
tion becomes
(12.44)
Note that if Ukk' = 0, ~k vanishes, and v~ = 1 for Ek < 0, while v~ = 0 for Ek > O.
This is the situation in a normal metal where Vk = 1 for Ek < E F and Vk = 0 for
Ek > E F. Since Ek = Ek - fl, it follows that fl is simply E F. A plot of v~ vs. Ek
is shown in Figure 12.12. Using the above expressions for u~ and v~, Eq. (12.42)
becomes
(12.45)
302 Superconductivity
(a) (b)
Figure 12.12 A plot of v~ vs. Ek in (a) the normal state, and (b) the superconducting
state.
It follows from Eq. (12.42) that 6k is constant, independent of k, for -nwo <
Ek < nwo, and zero otherwise. Writing the constant as 60, Eq. (12.45) becomes
VoDa(EF) j/iW/J dE
1=
2 - TiW/J JE 2+ 66
nwo
==? 6 0 = -----:::-----::- (12.47)
sinh [ UoD~ ( £F) ] .
D a(E F) is the density of states for one spin projection at the Fermi energy. In the
weak coupling limit (VoD a(EF) « 1), the gap parameter is given by
(12.48)
12.5 BCS theory of superconductivity 303
-
Es =6'" ( ? _
2vkEk - L\kukVk ) =~
' " [_ 2E~ L\~
+ ]
Ek - --F====== ( 12.49)
k k 2jE~ + L\~
(see Eqs [12.37], [12.42], and [12.44]). Assuming, as before, that L\k = L\o for
-nwo < Ek < nwo, and is zero otherwise, we see that the summand in the above
equation is equal to 2Ek for Ek < -nwD and is equal to zero for Ek > nwo. The
energy EN of the normal ground state is the sum over k of 2Ek up to Ek = O. Hence
Es - EN = "
6 ,I [_Ek - 2E~ + L\~] - 6
"," 2-Ek·
k 2jE~+L\~ k
The prime over the sum means that the sum is restricted to values of k such
that -hWD < Ek < nWD, while the dou~le prime itficates that the sum over k
is restricted so that -nwD < Ek < O. Writing the sum over k as an integral over
energy, we find
For weak coupling (L\o « nwo) , the above equation, upon expansion of the square
root, reduces to
(12.50)
The superconducting state is lower in energy than the normal state; hence, in the
presence of an attractive interaction between electrons near the Fermi surface,
the normal state becomes unstable, and the system undergoes a transition to a
superconducting state.
We note that the BCS theory, by replacing Vkk' by -Va , it neglects the fact
that the attractive interaction between electrons (mediated by phonons) is retarded .
This is a good approximation in superconductors where the electron-phonon inter-
action is weak, but it does not provide an accurate description of strong-coupling
superconductors, where the electron-phonon interaction is strong. For a review of
strong-coupling theory of superconductivity, the reader is referred to the article by
Scalapino (1969) .
Finally, we briefly touch upon a certain feature , mentioned earlier, of the super-
conducting ground state. The BCS Hamiltonian commutes with the number of
304 Superconductivity
particles operator, [Hscs, Nap] = 0, but the number of particles in the ground
state wave function is not constant. Stated differently, the Hamiltonian possesses a
certain symmetry which the ground state lacks. The superconducting state is thus
characterized by a broken symmetry. To elaborate this point further, we note that the
BCS Hamiltonian given in Eq. (12.32) is invariant under the global transformation
Under this transformation, the normal state 1F) = n~a cta 10) remains invariant;
it simply aquires a constant phase (the prime on the product sign indicates that
k < kF)' However, the BCS ground state, given in Eq. (12.33), is not invariant
under this transformation.
(12.5Ia)
Similarly,
(12.5Ib)
In a normal metal, the quantities (CktCkt) and (CttC~kt) vanish, but this is not
the case in a superconductor, where the ground state is not an eigenstate of the
number of particles operator. In terms of the fluctuation operators, the interaction
Hamiltonian is given by
In the mean field approximation, the last term in the above expression, which
is bilinear in fluctuation operators, is ignored. The assumption made is that the
fluctuations of CktCkt and CttC~kt about their average values are small.
Defining the gap parameter by
(12.54a)
(l2.54b)
The new operators must satisfy the same commutation relations as the original
ones; we thus require that
Using Eq. (12.54), we solve for the c-operators in terms of the y-operators,
(l2.57a)
(l2.57b)
Inserting these terms into Eq. (12.53), and then laboring through some tedious
calculations, we find
The first term describes single-particle excitations, while the last term is a con-
stant that represents the energy of the system in the absence of single-particle
306 Superconductivity
excitations - the ground state energy. The troublesome terms are the third and
fourth ones; they are not diagonal. However, the only condition imposed on Uk and
Vk so far is Eq. (12.56). We can take advantage of the available freedom regarding
the choice of Uk and Vk by demanding that the troublesome terms vanish. We thus
impose the condition
(12.59)
(12.60)
l\{fo) = nk
Yk t Y-kt 10)
12.6 Mean field approach 307
where 10) is the vacuum state. That l\lJo) satisfies Eq. (12.62) follows from the
commutation relations of the y-operators. Using Eq. (12.54),
l\lJo) = n
k
(UkCkt - VkC~k.!-) (UkCk t + vkctt ) 10)
= n
k
(UkVk - v~c~k.!- ct t ) 10)
l\lJo) = n
k
(Uk + Vkctt C~kt ) 10). ( 12.63)
This is the same state we saw earlier using the variational method. Again, v~ is the
probability that the pair Ik t, -k t ) is occupied, and u~ is the probability that it is
empty.
The ground state energy is the second term of the Hamiltonian given in
Eq. (12.61). At T = 0, (ctt C~k.!-) = (\lJoICt t C~kt l\lJo). We can evaluate this directly
by using the expression given for l\lJo) in Eq. (12.63); alternatively, we can use
Eq. (12.57) to write
t t _2 t t 2 t t
Ck t C- kt - Ukht Y- kt - VkY-k t Yk t - UkVkh t Yk t + UkVkY- kt Y_kt ·
(12.64)
This is exactly the same expression obtained earlier using a variational approach
(see Eq. [12.49]).
An alternative expression for Es can be written. From Eq. (12.60),
(12.66)
308 Superconductivity
,,
,
,, ,,
It follows that
-
Es = - L 4Ekvk + L (2E vk -
4 k 2 -6~)
- .
k k
2Ek
L
k
(2EkV~ - 2Ek
6~) = L2Ek (v~ - 6~2) = L2Edv~ - u~v~)
k 4Ek k
Es ~
- = -2 ' " Ekv 4 .
k (12.67)
k
Going back to the mean field Hamiltonian of Eq. (12.61), the first term describes
excitations above the ground state. The single-particle excitation has energy
J
Ek = E~ + 6~. If we adopt the approximation that 6k is independent of k,
as we did in the previous section, we see that the minimum energy for a single-
particle excitation is equal to 6, which corresponds to particles at the Fermi
surface (Ek = Ek - E F = 0). The single-particle excitation energy is plotted in
Figure 12.13 as a function of Ek. Note that, in a nonnal metal, it is possible to excite
an electron from a state just below the Fermi surface to a state just above the Fermi
surface by adding an infinitesimal amount of energy. This is not the case for a
superconductor.
Finally, we note that even though the minimum single-particle excitation energy
is equal to 6, the lowest excited state has energy 26 above the ground state energy.
This is because the lowest excited state involves breaking up a Cooper pair: an
electron is scattered out of the state Ik t), leaving behind an unpaired electron
12.7 Green's jun ction approach to superconductivity 309
in the state I - kt ). If the pair state Ikt, -kt ) is occupied in the ground state
(v~ = 1), then it is unoccupied in the excited state (v~ = 0). Using Eq. (12.67), the
change in energy is 2Ek, which has a minimum value of 26. Another way (Taylor
and Heinonen, 2002) to arrive at this result is by realizing that in a superconductor,
single-particle excitations are always created in pairs (never singly). Any pertur-
bation will scatter electrons between states; thus, any perturbation Hamiltonian
will contain an equal number of electron annihilation and creation operators (the
minimum is one of each kind). For example, consider a perturbation of the form:
When this acts on the ground state, the first term in H I gives
L Vkk' Ct'tCkt Iwo) = Vkk' L (Uk' Y~'t + Vk' Y- k'.j, ) (UkYkt + VkY ~k.j, ) IWo)
k#W k#W
the other terms being zero. A similar expression is obtained if the second term in
H I acts on Iwo). Therefore, only pairs of particles are excited, and the minimum
excitation energy, equal to 26 , is obtained if Ek = Ek' = O.
(12.68)
ka kk'
Here, Ek is the single-particle energy measured from the chemical potential, and
the sum over k and k l is restricted to values that satisfy -nwD < Ek , Ek' < nWD.
The imaginary-time Green's function for spin-up electrons is
g(kt. r) =- (T Ckt (r)ctt (O)) =- e(r)(ckt (r)ctt (O)) + e(-r) (c~t (O)ckt (r) ) .
(12.69)
The modified Heisenberg operator Ck t (r) is given by
(12.70)
310 Superconductivity
a
ar g(kt, r) = t t /a
-o(r)(cktckt ) - o(r)(cktCkt ) - 8(r) \ ar Ckt (r)ckt(O)
t)
+ 8( -r) (C~t(O) aar Ck t (r)).
~g(kt,
ar
r) = -oCr) - /
\
T~Ckt
ar
(r)c~t(O)). (12.71)
a
-Ck t (r) = -1[-
H , Ckt(r) ] = -1[-
Ho, Ckt(r) ] + -1[ H,I Ckt (r) ] .
ar Ii Ii Ii
Note that
We find
(aa
r
+ ~k)g(kt, r) = -oCr) + ~ ~ Uk'k (TC~k~(r)c-k'+(r)Ck't(r)C~t(O)).
(12.72)
As is usually the case, the equation of motion of the one-particle Green's func-
tion contains a two-particle Green's function (the second term on the RHS of
Eq. [12.72]). Ideally, we would construct the equation of motion for this function
as well, but then a three-particle Green's function would appear, and so on; the
system of equations never closes on itself. This, of course, reflects the fact that the
problem is not exactly solvable; we need to resolt to some approximation scheme.
J2.7 Green 's jlln ction approach to superconductivity 311
We assume that the particles are weakly interacting; the effect of the interaction is
considered only to the extent that it leads to the formation of Cooper pairs whose
number is not constant. In other words, we evaluate the average of the time-ordered
product in Eq. (12.72) for a noninteracting system, one whose energy eigenstates
are not eigenstates of the number operator. We may then apply Wick's theorem,
(TC~kt (r )c- k' t (r )ck't ( r )c~t (0») = - (T Ck' t (r )C~kt (r »)(T ck't( r )c~t (O»)Okk'
- (T Ck't ( r )Ck' t (r») (T C~kt (r)c~t (O») .
In a normal metal, where all the stationary states can be chosen to be simultaneous
eigenstates of Hand N op , only the first term on the RHS of the above equation
survives, and the approximation is the Hartree-Fock approximation. This term is
simply g(kt, r) multiplied by a time-independent function; it leads to a renormal-
ization of the single-particle energy, and it will be dropped in what follows. The
second term vanishes in a normal metal but does not vanish in a superconductor,
where states are not eigenstates of the number of particles operator. We thus define
two new "anomalous" Green's functions,
(~
ar
+ Ek / fi) g(kt , r) = -oCr) - ~ ' " Uk'kF(k' , O)F t (k, r)
fi ~
k'
1
= -oCr) + ~D.kFt(k, r). (12.74)
To solve for g(kt , r), we write the equation of motion for Ft(k, r),
aa F t (k , r) = aa
r r
[-e(r)(C~kt (r)C~t (O») + e(-r)(c~t (O)C~kt (r»)J
=-o(r) (C-t HCkt t + Ckt
t C-t kt ) -\TarC- I
a t kt (r)Ckt )
t (O).
Since {c ~H ' C~t } = 0, the first term on the RHS vanishes. The second term is
obtained by evaluating the commutator [H. c~ H ]. We end up with
The coupled equations for g(kt, r) and Ft(k, r) (Eqs (12.74) and (12.76)) are
solved by Fourier expanding
00
1
g(kt, r) = - 1 ~ g(k t, wlI)e - ;wnr ,
L F t (k. r) = f311 L Ft(k, wn)e-1W"r
00 .
t -6UI1 (12.81)
F (k. wII ) = )
. (iw n )2 - (E~ + 16kl2 Iii 2 '
Green's function can also be expressed another way. Using the expressions for u~
and v~ given in Eq. (12.60), we can show that
u2 v2
g(kt , wlI ) = . k +. k (12.82)
IW n - Ek l li IW n + Ek / l1
where Ek = JE~ + 6~. The retarded Green's function G R (k t, w) is obtained from
g(k t , w II ) by replacing iWII with W + W+ . The spectral density function A(k t, w)
is equal to -21m GR(kt , w); hence
(12.83)
The spectral density function consists of two delta-function peaks. The first peak,
at E k , corresponds to the energy of an electron added to the system in state Ikt).
The second peak, at - E k , is the energy of an electron removed from state Ik t ). To
add an electron into state Ik t), the pair state Ik t, -k,l.) needs to be unoccupied;
the probability of that is u~. To remove an electron from state Ik t), the pair state
Ik t , - k ,l. ) must be occupied; the probability of that is v~.
If the pair state Ikt , -k,l. ) is occupied, then u~ = 0 and v~ = 1, the spectral
density function has one peak at liw = - E k , and the energy of the electron in state
Ik t ) is - E k . If, on the other hand, the pair state Ik t, -k,l. ) is empty, then u~ = 1
and v~ = 0, the spectral function has one peak at w = Ek / h , and the energy of
the electron in state Ik t ) is E k . If k is such that Ek = IJ-, then the energy of the
electron in state Ik t ) is either -6k or +6k, depending on whether the pair state
12.7 Green's function approach to superconductivity 313
T c 0 T o
(a) (b)
Figure 12.14 Pictorial representation of the anomalous Green's functions (a)
F(k , r) and (b) Ft(k, r).
(12.84)
where :E *(k t, wn ) is the proper (irreducible) self energy. In the normal state, the self
energy consists of diagrams containing only Green's function and interaction lines,
because when we apply Wick's theorem in this situation, only contractions that
involve one annihilation and one creation operator are nonvanishing. By contrast,
in the superconducting state, anomalous Green's functions appear, so we need to
expand our store of Green's functions. Diagrammatically, the anomalous Green's
functions are represented as in Figure 12.14. These functions have vanishing zero-
order values, i.e., they vanish in the non interacting system. This is clearly so, since
the Hamiltonian is Ho in the absence of interactions and the system is in the normal
state.
Every Green's function diagram consists of a series of self energy diagrams con-
nected by the zeroth-order Green's function. In the normal state, a single line enters
the self energy part and another single line leaves it, as shown in Figure 12.15a. The
conesponding diagrammatic expansion for g(k t , w n ) is shown in Figure 12.15b. In
the superconducting state, two new types of self-energy diagrams become possible:
two single lines enter or leave the self-energy part, as in Figure 12.15c. Ignoring
normal-state corrections, the diagrammatic expansion for g(k t, w n ) is shown in
Figure 12.15d. An examination of this figure reveals that structures occuning after
the first self energy part correspond to a new function. Graphically, this new func-
tion is characterized by two external lines pointing outward; it is the function
FtCk, WI!). The analogue of Dyson's equation is shown in Figure 12.l5e.
We can now write Dyson's equation for a superconductor. Ignoring normal-state
corrections, Dyson 's equation (in the Hartree-Fock, or mean field approximation)
314 Superconductivity
(a) ~
(b) = ~ + ~ + ~ + ...
(c) ---0-----0---
(d) ~ + ~+~+ ...
(e) = ~ + ~
Figure 12.15 (a) In the normal state diagram, a single line enters the self energy
part and a single line leaves it. (b) The expansion of Green's function in the nonnal
state. (c) In the supercond ucting state, additional diagrams appear in which two
single lines enter or leave the self energy part. (d) Diagrams that appear in the
expansion of Green 's function in a superconductor. (e) The Dyson-like equation
for Green's function in a superconductor.
(a) ~+§
(b)
§
E
(12.86)
The signs before the integrals may be checked by writing the first order perturbation
term and applying Wick's theorem. Using the definition of the gap parameter
12. 7 Green's function approach to superconductivity 315
The solution ofEqs (12.87) and (12.88) is identical to the one shown in Eqs (12.80)
and (12.81).
To determine the gap consistency condition, we adopt the following simple
model, which was considered earlier:
Within thi s model, the gap parameter is independent of k and is written as t..; it is
given by
k k k
The prime on the summation means that the sum is restlicted to values of k such
that IEkl < I1WD . Using
00
(12.90)
Since Wn = (2n + 1)][ 1/311, the summand reduces to - 11w~ as n ~ ±oo; the series
is convergent. We may thus introduce the redundant convergence factor eiw"o+,
which allows us to evaluate the sum over n using the method of partial fractions,
/3li2 .
= 2Ek (iEk - f-E k) (12.91)
r
where f Ek = (1 + e,B Ek 1 is the Fermi distribution function. In evaluating the
frequency sum we have made use of Eq. (9.14). Since f-E = 1 - fE, Eq. (12.90)
316 Superconductivity
may be written as
(12.92)
(12.93)
_ _1_ _
UoDa(EF)
= Inx tanhxl g - r
io
sech 2 x Inx dx.
For weak coupling, e » ]; we can then replace tanhe by 1 and extend the upper
limit of integration to infinity (this is possible because sech2 x is a rapidly decreasing
function of x for large x):
The integral on the RHS is tabulated; it is equal to In(n /4) - y, where y :::::: 0.577
is Euler's constant. Therefore
12.9 The Nambuformalism 317
(12.94)
(12.95)
02.96)
02.97)
(12.102)
a
h-ak(r) _
= -Eka3ak(r) * )ak(r).
+ (D.k a+ + D.ka- (12.103)
ar
The equation of motion for the Nambu Green's function now becomes
1. A single electron line stands for the diagonal Green's function whose entries are
gO(k t, wll ) and - gO( -k t, -wll ).
2. The electron-electron Coulomb matrix element carries an extra factor a3 , as
does the electron-phonon interaction matrix element.
12.10 R esponse to a weak magnericfield 319
3. For a closed electron loop, the trace is taken over the matrix product of the
matrices that represent the lines that make up the loop.
(12.108)
The first term is the diamagnetic current-density operator, while the second term is
the paramagnetic current -density operator,
e2 ne 2
(jD)(r, t) = --(nCr, t) )o A(r, t) = --A(r, t) (12.1 LO)
me me
where n is the number of electrons per unit volume. Note that since jD is already
proportional to A, the ensemble average of nCr , t) is taken over the unperturbed
system, i.e., over the system in the absence of the vector potential.
To determine (jp ) in the presence of A , we need to determine H ext , the pertur-
bation which arises from the presence of A ; thi s is given by
(l2 . 1ll)
320 Superconductivity
(see Problem 12.6). In the absence of the vector potential, (Y(r, t))o = O. To first
order in A, (jp) is given by Kubo's formula (see Eq. [6 .74])
= :c jl f
-co
dt '
3
d r' Lj3 (U:H(r, t) , J;H(r', t ' )])A/3(r' , t l) (12.112)
ne 2 I
(J,,(q, w)) = --Aa(q, w) - -
mc hc j3
L
D:j3 (qw)A j3 (qw). (12.113)
where e(t) is the step function and V is the system's volume. The operators inside
the commutator are modified Heisenberg picture operators.
The retarded function D:j3(q, w) is obtained by analytic continuation from the
corresponding imaginary-time con-elation function D"j3( q, w), which is the Fourier
transform of
(12.115)
(12.116)
12.10 Response to a weak magnetic field 321
(see Problem 3.7). In the last term on the RHS , we make a change of variable:
k ~ -k - q,
Since Lk k = 0, the term containing the Kronecker delta yields zero. The remain-
ing two terms inside the brackets add up to QI~Qlk+q. Therefore,
?Ii 2
D aj3 (q , r) = :;~ V
L(2k a + qa)(2kfi - qj3 ) (TQI~(r)Qlk+q(r)QI~, (O)Qlk'_q(O)).
kk'
(12.118)
We can evaluate D aj3 (q, r) by means of a perturbation expansion followed by the use
of Wick 's theorem. Daj3 (q , r) is similar to the dressed pair bubble in the interacting
electron gas which we studied in Chapter 10. Graphically, D aj3 (q, wm) is given in
Figure 12.18. In evaluating D aj3 ( q, Will) we keep only one pair bubble, as indicated
in Figure 12.18. In this case, k' = k + q; it follows that 2kf, - qj3 = 2kj3 + qj3 '
Since we have a closed electron loop, there is an additional factor of -1, and the
trace must be taken over the matrix product. The Feynman rules thus yield the
following expression.
lie 2
Daj3 (q , Will) = 2 L(2ka + qa)(2kj3 + qj3 )Tr [g(k, wn)g(k + q, Wm + Wn)] .
4m {3V
k.n
(12.119)
322 Superconductivity
Figure 12.18 Perturbation expansion of Daf3 (q. w l1l ). In the lowest order, only one
pair bubble is retained. The double lines represent the Nambu matrix Green's
function .
The frequency summation is best carried out by using the spectral representation
of the matrix Green's function,
_ -1
g(k. w lI ) -
00
-00
A(k. E) dE _
.
lW II - E 2iT
- - -
100
- 00
1m
.
tRek.
lW II - E
E) dE
-
iT
(12.120)
where A(k, E) = -21m tRek. E) is the spectral density function and tRek, E) is
the retarded matrix Green 's function. Thus
Dall(q, Wm) =
4111
ne2 2
f3 V
~(2ka + qa)(2k ll + qll) 1 1
-00
00
-dEl
iT
00
- 00
-dE2
iT
k.1l
(12.122)
where we used the frequency summation formula (see Eq. [9.14]). Here, iEis the
Fermi distribution function,
The retarded function CR (k , E) = g(k, CUll) Ii w,, ---> E+iO+ . The matrix function
g(k. CUll), given in Eq. (12.107), can be written as
-6k / 11 ]
icu ll
- Ek / 11 .
(12.125)
Replacing i CUll with E + iQ+ and taking the imaginary part, we find
-6k/11 ] .
E - Ek/11
(12.126)
To simplify the calculations, we adopt the BCS model, where the electron-electron
interaction is a negative constant in a shell of energy width 2!1cuD that encloses the
Fermi surface. In this model, 6k is a real constant, independent of k. In this case,
a straightforward calculation yields
+ (1 _ EkEk+q
2
EkEk+q + 6 ) iEk + iEk+q-
Ek + E k+q
1]. (12.127)
For a # f3 , the sum over k yields zero since E -k = Ek and h -k = f Ek. Therefore,
2Ji3 e2 'Of
W I1l = 0) = Da,B (q = 0, W I1l = 0) = -2-oa,B Lk;~
m V k aE k
211 3e2 ,,
2a
hk
= 3m 2VOa,B Lk aE k '
k
ne 2 2Ji 2e 2 'Of
(fa) = - - Aa - 2 Lk2~Aa.
me 3m eV k aE k
We can rewrite this expression as follows:
(12.128)
where
-
n -n + -2112- Lk 2 -
'Or
Ek
. - (12.] 29)
" 3mV
k
aE k
is interpreted as the density of superconducting electrons. Taking the curl of both
sides ofEq. (12.128) yields the London equation, which results in a Meissner effect
as long as n" # 0, as shown in Section 12.2.
At T = 0, ahk / aEk = -o(Ek ) = -0 (jE~ + 6, 2) = ° since 6,2 > 0. In this
case, ns = n and the Meissner effect exists. As T increases from zero, ns decreases,
°
and at T = Tc , 6, = and Ek = Ek . Since Tc is small, we assume that at Tc the
electrons occupy all states below the Fermi energy E F; in this case, ahk/aEk =
aflaEk = -O(Ek) = -O(Ek - EF)' Assuming that Ek = Ji 2k2/2m, then at T = Tc
4 2
n5 = n - - " Ek O(Ek - E F) = 11 - -" EFo(Ek - E F)
3V L 3V L
k ka
(12.130)
12.1 J Infinite conductivity 325
where deE F) is the density of states, per unit volume, at the Fermi energy. A simple
calculation of deE F) shows that deE F) = 3n/2E F; hence at T = Tc, n, = 0, and
the Meissner effect disappears.
ne 2 1
lim (Ja(O , w) ) = - - lim Aa(O , w) - - " lim lim D:{3 (q, w)A {3 (O , w).
w->O me w-> O lie ~ w-> O q---7 0
{3
(12.131)
In studying the Meissner effect, the limits were taken in reverse order: first w ~ 0,
then q ~ O. It turns out that D:{3 (O, 0) does not depend on the order in which the
limits are taken , and we arrive at essentially the same result as in the previous
section:
aJ n s e2
- = -E(t). (12.133)
at m
This is precisely the equation for current density in a system containing ns free
electrons that are not subjected to any damping (J = -nsev => aJ/ at = -nsea =
ns e2 E/ m). The equation clearly shows that as long as I1 s #- 0, a steady uniform
electric field produces a current that increases with time; this is the signature of a
perfect conductor.
326 Superconductivity
Further reading
Abrikosov. A.A.. Gorkov, L.P., and Dzyaloshinski, I.E. (1963). Methods of Quantum Field
Th eo ry in Statistical Physics. New York: Dover Publications.
Fetter, A.L. and Walecka, J.D. (1971). Quantum Theory of Many-Particle Systems. New
York: McGraw-HilI.
Rickayzen, G. (1965). Theory of Superconductivity. New York: Wiley.
Schrieffer, J.R. (1964). Theory of Superconductivity. New York: W.A. Benjamin, Inc.
Taylor, P.L. and Heinonen, O. (2002). A Quantum Approach to Condensed Matter Physics.
Cambridge: Cambridge University Press.
Tinkham , M. (2004). Superconductiviry, 2nd edn. New York: Dover Publications.
Problems
12.1 The operator S. Show that S, given in Eq. (12.18), satisfies Eq. (12.15).
12.2 Ground state energy. Verify Eq. (12.37) for the ground state energy of a
superconductor.
12.3 The anomalous Green 'sfunction. Derive Eq. (12.76), the equation of motion
for Ft(k, r).
1
L
1l=00 .
oCr) = - e-1W"r Wn = (2n + l)n I {311 .
{3n 1/=-00
Verify this equation. Hint: use the frequency sum formula (Eq. [9.14]) and
its complex conjugate. Also note that fEk=o = 1/ 2.
12.5 Equation of motion. Starting from the mean field Hamiltonian, as given in
Eq. (12.53), show that the equations of motion for g and Ft are given by
Eqs (12.74) and (12.76).
g(r t r , r ' t r ') = - (T w 't (rr)w +(r' r ')) , F(rr , r ' r ') = - (TW t (rr)W -J, (r' r ')),
1
and F t (rr , r ' r') = -(T W (rr)IlJ+(r'r')). The Hamiltonian is
fI = ~ f d nl>; (r) {
3
L[ -;Ii" + eAc(r)], -I'} <1'0«)
- Vo f 3
d rw+(r)wl (r)w -J, (r)w t (r) .
t,
~] " + fJv } g(r, r ,UJ ~(r)F
. Ii i eA(r) ,. ,
{
lliUJ I1 + 2 [V
2m + I1 ) + (r, r , UJ I1 )
= lio(r - r')
.
-IIiUJ fz2 [ V - -
I1 + - - +11, } F t (r, r ' , UJI/) -
ieA(r)]2 ~
*(r)g(r, r ,, UJ ) -_ 0
I1
{ 2m lic
where ~ * (r) = VoF t Crr + , rr) = (Vo / fJli) L/j FtCr, r, UJIl)'
(b) Let gOer , r' , UJI/) be the temperature Green 's function of the metal in the
normal state in the presence of the vector potential A(r). The equation
for gOer, r ' , UJ/j) is obtained from the above equation for g(r, r' , UJ n ) by
setting ~(r) equal to zero. Show that
(c) As the magnetic field approaches the critical field, tJ. -+ 0 and g -+ gO.
Show that, in this limit,
. = -Vo
tJ. ~(r) 2"~ /dIg
f3h
3 ° °
(I, r, -wn)g (I,r, wn)exp -
1/
(2ielr)
nc I
A(s).ds tJ. * (I)
ksTc : : : : 1.14licuDexp ( -1 )
UoJ Das(O)Darr (O)
where Das(O) and Darr(O) are, respectively, the densities of states per
spin orientation, at the Fermi energy, of the sand Jr bands.
(d) In the presence of a magnetic field described by the vector potential
A(r),
grr and FJ are similarly defined, The gap functions are now given by
!'J. s*()
r -_Uo ~i wnO+F5t( r , r , CUll ) , Urr
-L-.te A *()
r _UO~iWnO+
- -L-.te Frrt (r , r , CUn),
I3Ii II
I3Ii II
330 Superconductivif)'
Show that, as the magnetic field approaches the critical field , the gap
functions satisfy the following equations:
Vo2 "'/
~ s* (r)=- (2ielr)
~ d 3 I gs0 (l , r , -wll)gs0 (l , r , wll)exp -
f31i ne I
A(s).ds ~ *rr (l)
11
13.1 Introduction
Thus far, we have studied systems in equilibrium . In Chapter 8 we developed a
perturbation expansion for the imaginary-time Green's function which was made
possible by the similarity between exp( - f3 H), which occurs in the statistical oper-
ator, and the time evolution operator exp( -i fIt 1ft). Such similarities do not always
fortuitously occur, however. A perturbation expansion is possible for the real-time
causal Green's function at zero temperature, but not at finite temperature.
What approach can we use when a system is driven out of equilibrium by,
for example, a time-dependent perturbation that is switched on at time to? In
Chapter 6, we developed a method that gave a system's response to first order in
the perturbation (linear response) . That method, however, is incapable of dealing
with the general case of nonlinear response. Moreover, when the Hamiltonian is
time-dependent, the time evolution operator is no longer exp( -i H t i ft), and the
Matsubara technique becomes inadequate.
We should point out that it is not necessary for a perturbation to be time-
dependent to drive a system out of equilibrium . Consider the following example,
depicted in Figure 13 .1. Two metallic leads and a quantum dot (a nanostructure,
for example) are initially separated. The left lead, the right lead, and the dot are
initially in equilibrium, with each part having its own chemical potential. Assume
that Ih > MR. The Hamiltonian for the system is the sum of the Hamiltonians for
the leads and the dot. At time to, the dot and the leads are brought into contact,
and a coupling between the dot and the leads is established. As a result, current
331
332 Nonequilibrium Green '5 junction
U t=
L
: 0 JLR
(a) (b)
Figure 13.1 A system driven out of equilibrium. (a) A three-component system
consisting of two metallic leads (left and right) and a central quantum dot. The
components are separated and each is in equilibrium. (b) The leads and the dot are
brought into contact, and a coupling is established between the leads and the dot.
The coupling causes a current to flow, driving the system out of equilibrium.
begins to flow, resulting in the dot now being out of equilibrium. In this case,
the perturbation is the coupling between the dot and the leads; apart from being
switched on at time to, the perturbation is time-independent. The techniques we
used earlier employing Green's function for systems in equilibrium cannot deal
with this situation; for example, the equilibrium methods cannot give the current
through the quantum dot.
To develop a method applicable to systems out of equilibrium, it is helpful to
understand why the equilibrium methods fail. Toward this end, we take a closer
look at the real-time causal Green's function. Before doing so, however, we discuss
the Schrodinger, Heisenberg, and interaction pictures of quantum mechanics, in
the general case where the Hamiltonian is time-dependent. These pictures were
described earlier for a time-independent Hamiltonian.
On the other hand, dynamical variables are represented by hermitian operators that
have no explicit time dependence. Given the state IVrs(to») at some initial time to,
the state at time t is given by
This differential equation for U (t , to), along with the boundary condition
uCto, to) = I , can be converted into an integral equation,
U(t, to) = 1 - ~
n.
l'
'a
elt,H(tI)U(t" to) . (13.5)
where T is the time-ordering operator which orders operators with increasing time
arguments from right to left. For t < to, it is possible to show that
U(t,t)=l (13.8a)
where As is the operator A in the Schrodinger picture. More generally, the system
may be in a stati stical mixture of states, where we may have only limited information
334 Nonequilibrium Green's function
about the system; for example, we may know only its volume, temperature, and
chemical potential. Then
(13.12)
n
(13.14)
(13.15)
(13.16)
/3.2 Schrodinger, Heis enberg, and interaction pictures 335
The equation of motion of the Heisenberg operator is obtained by differentiating
both sides of the above equation,
d
ili-AH(t) = [AH(t) , HH(t)] (13.17)
dt
where
(13.18)
(13.20)
(13.21)
(13.22)
(13.24)
S(t , t) = 1, st(t , to) = S- '(t , to) = S(to , t) , Set, tl)S(t", t') = Set, t').
(13.26)
A relation between S(t, to) and U (t, to) can be derived. For an arbitrary IVr,),
S(t, to)IVr,(1o)) = IVr/(1)) = eiHo(t - lo)/ftIVrs(t)) = eiHo(t - lo)/FtU(t, to)IVrs(to))
= eiHo(t- lo)/"uCt , to)IVr,(to))
(13.27)
pet) = L PIIIVrIl (t)) (Vrll (t)1 = L Pilei HO(t-IO)/Ft 1VrIl SCt )) (VrIl S(t) le- i Ho(t-Io)/Ft
I I
II II
II
We thus find
(13.28)
equilibrium. It is instructive to see exactly why such an expansion fails; this will
point the way to the construction of a Green's function that is more appropriate for
the study of systems that are not in equilibrium.
Let us consider a system of interacting pat1icles with a Hamiltonian
H = Ho + V.
At time t = to, a (possibly) time-dependent perturbation H ext(t) is applied. The
Hamiltonian is now given by
(13.30)
where T is the time ordering operator; 1/IH(1) is the field operator (in the Heisenberg
picture) which annihilates a particle of spin projection (J at position r and time t;
1/I~(1 1) creates a particle, at time t', of coordinates (r'(JI); and (. .. ) stands for a
grand canonical ensemble average,
Tr[pCto)· . . ]
(- . . ) = .
Tr[p(to)]
In writing the ensemble average, we used the fact that the statistical operator is
time-independent in the Heisenberg picture. Using Eqs 03.26) and 03.29), we can
write
The lower (upper) sign refers to fermions (bosons), and e(t - t' ) is the step function:
e(t - t' ) = 1 (0) if t > t' (t < t ' ). Using st (t , to) = S(to, t), the above relation may
be written in a more compact form:
where tm = max(t , t' ). We have used the fact that, under time ordering, S-operators
commute with field operators, since fermion operators come in pairs in the
S-operator. Green 's function is thus given by
.
zG(l, ' = (S(to , tm)T[S(tm , to) 1/10)
1) A 1/1At,
(1)] ) . (13.31)
338 Nonequilibriul17 Green's fun ction
Multiplying by S(tm . oo)S(oo , tm) = 1 after S(to. tm) in Eq. (13 .31), and moving
S(oo, tm ) inside the time-ordered product (this is allowed because S(oo, till) is an
expansion in time-ordered products of operators and all the times in S(oo, til)) occur
later than the times inside the T -product), we obtain
Thus far, our treatment is applicable whether or not the system is in equilibrium,
and whether its temperature is zero or finite. We now consider a system in equilib-
dum (Hex t = 0, 1{ = Ho + V) at zero temperature. The ensemble average reduces
to an average over the interacting ground state (in the Heisenberg picture):
(13.33)
Since the pictures coincide at t = to, we may replace 11/101/ ) with 11/10 1(to)). The
difficulty in the above expression arises because the ground state of the interacting
system is unknown. Is there a way to express Green's function in terms of the
non interacting ground state? To accomplish this, we invoke the mathematical trick
of switching the interaction on and off adiabatically: we assume that V is turned
on and off with infinite slowness :
Vet) = e-E I/ - IOI V (13.34)
where E is a small positive number that is eventually set equal to zero. In the
remote past and in the distant future, the particles are noninteracting. At t = -00
the interaction is slowly turned on, and it attains its full strength at t = to. The
non interacting ground state 11/101 ( -(0)) evolves adiabatically to the interacting
ground state at t = to:
(13.35)
Here, SE is the evolution operator determined by Vet) in Eq. (13.34). Putting this
into Eq. (13 .33) and using Eq. (13.26) , we obtain
(13.36)
The state 11/10/(0) ) = SE(oo, -(0)11/10/ -(0) ) is the state obtained from the nonin-
teracting ground state in the remote past by adiabatic evolution to the distant future,
where the system is also noninteracting; hence, both 11/10/ (-00) ) and 11/10 / (00) ) are
ground states of Ho. Since the ground state is nondegenerate, these two states can
only differ by a phase factor,
i
11/10, (00) ) = e ¢ 11/Io/-(0) )
i
e¢ = (1/Io/ (-00)11fro/00) ) = (1fro / (-00)ISE(00 , -(0)11fro,(-00) ) .
13.3 The malady and the remedy 339
c )
ta ··-------+--~~
E
• •
C
(a) (b)
~ ~
Figure 13.2 C = cue. The contour runs along the real-time axis from to to
1110X(t, t ' ), as in (a), or to 00, as in (b), and back along the real-time axis to to. For
~ ~
the sake of clarity, C and C are drawn above and below the real-time axis .
We can bring this expression into some formal similarity with the expression used
for states in equilibrium by proceeding as follows. We introduce a contour-ordering
-+
operator Te, along a contour C which consists of two parts: C from ta to fill (or
~
00, if we start from Eq. [3.321) and C from till (or 00) to to, as depicted in Figure
13.2. The contour-ordering operator is defined by
e
TdA(r)B(r ' )] = 1 A(r)B(r')
±B(r' )A(r)
r > r'
e
r' > r
(13.39)
where the lower (upper) sign refers to fermions (bosons), and the time along the
e
contour is denoted by r. The statement r > r ' means that r lies further along the
contour than r ' , regardless of the numerical values of rand r ' . Thus, ordering along
-+ ~
C cOLTesponds to normal-time ordering, whereas ordering along C corresponds
to anti time ordering:
T-C = T,
I -+
r, r E C. (13.40)
All operators are now under contour ordering. The above expression is simply
Eq. (13.32) rewritten. Equation (13.32) does not give a perturbation expansion
for G(l, [I), so Eq. (13.40) does not either. However, the form of GO, 1') in
Eg. ([3.40) suggests a generalization: rather than restricting T and r' and tying
-+
them to C , they can be freed so as to lie anywhere on the contour C. This step, as
it turns out, produces a Green's function that admits a perturbation expansion, one
that is relevant for systems that are out of equilibrium.
13.4 Contour-ordered Green's function 341
r r' c7 r'
to' : 1 ~) to .. : :) to-"--:--+;,--+)
err' r
(a) (b) (c)
---+
Figure 13.3 The contour C consists of a forward part ( C) and a backward part
+-
(C) . Both lie along the real-time axis but are shown displaced from it for the sake
of clarity. rand r ' are the locations of ( and (' on the contour. In (a) both ( and t '
---+ +-
are on the forward part C, while in (b) they are both on the backward part C. In
+- ---+
(c), ( E C, t' E C. Note that while ( < t' on the real-time axis in this figure, as
c c
contour times r > r ' in (b) and (c), while r < r ' in (a).
+- -'? C
3. tEe , tl E C; then T > T' and
+- -
4. t , t l E C; in this case Tc = Tc = T and
Since
it follows that
Therefore, three of the four functions contained in G c(l, 11) are independent. The
retarded and advanced Green's functions, G R and G A , respectively, are given by
Finally, the following relations among the various functions can be verified:
( 13.47)
J3.5 Kadanoff- Baym and Keldysh contours 343
C1 C2 C1
to~t ;, to~tl ;,
C3 C3 C2
(a) (b)
Figure 13.4 C = C, U C2 U C3. The contour segments actually lie along the time
ax is, but are shown displaced from it for the sake of clarity. In both (a) and (b),
c
r > r '.
(13.49)
where C is the contour depicted in Figure 13.2 or Figure 13.3. A proof of this result
is presented below.
We prove Eg. (13.49) for the case when t lies further along the contour than t'; the
c
opposite case is proved in a similar way. For r > r ',
In writing this, we have used Egs (13.26) and (13.29). To calculate the RHS of Eg .
(13.49), we divide the contour into three segments C" C2, and C3 (see Figure 13.4).
The contour C = C, U C2 U C3, and fc = fCI + fC2 + f c3 ' The contour-ordered
product on the RHS of Eg. (13.49), denoted by B, can be written as
B = Tc [ e-HcH'(rlldTI-jJ(l)-jJt(lI) ]
= f ~ (-i)"
n
/1 = 0
1
11! c
dr, ... r
Jc
drll Ie [H'(r,) .. . H '( r/l)-jJ(l)-jJ t (l I) ].
Let us consider the term of order n. Each integral over C is a sum of three integrals,
and the term of order Il is thus the sum of 3" terms. Consider one such term that has k
integral s along C3 (k = 0, 1, ... , 11), I integrals along Cz (I = 0, l, ... , n - k) , and
II - k - I integrals along C,. There is a total of 11 !/ [k!l !(n - k - I)!] such term s that
differ only by a relabeling of their time indices; since the times are integrated over,
344 Nonequilibriul11 Green 'sjimetion
1 ( ')"
~I LLk!I!(n~k-I)!
00 II II-k I
B= L ; !
11=0 k=O 1=0
A, A,
X
1 e,
drk+l+ l . . .
l
e,
drll Te, [H (rk+l+ l)' .. H (rll ) l.
Noting that
C I extends from to to t' > 10 ==> Te , = T => R = S(t ', to). C3 extends from 1 to
to < t => Te 3 = f => P = SCto , t). C2 extends from t' to t. If t > t ' then TC2 = T
and Q = S(t , t' ); if ( < (' then TC2 = f and again (see Eq. [13.25]) Q = S(t, ('). This
C I
ends the proof of Eq. (13.49) for the case r > r .
The last equality in the above equation is valid since V and H exl contain an even
number of felmion operators, so they commute under contour ordering. Defining
13.5 Kadan ()ff- Baym and Ke fdysh contours 345
the operators
we can wlite
(13.52)
From the definition of i G cO, 1'), if 1j; is replaced by I and 1j; t is also replaced
by 1, then iG c(l, 1') reduces to 1. Hence, we deduce from Eq. (13.52) that Tc
(S~ S~X l) = 1.
Setting Hext(t) equal to zero in Eq. (13.27), which leads to U (t , to) becoming
equal to exp[-iH(t - to) /Ii], and using Eq. (13 .25) , we can write
(13.53)
(13.55)
The contour Green 's function (see Eq. [13.52]) may now be expressed as
(13.57)
where
ym;.
to !r-__;'~E_._ _......
to-i{3/i
Figure 13.5 Kadanoff-Baym three-branch contour C' ; it starts at to and stretches
to max(t , t ' ), then returns to to, and down to to - if3li .
till = max(t , t') or any time beyond till' returns to to on the backward path, and
goes down to to - if3h. The operator Tc is the contour-time-ordering operator
along C .
We may, if we choose, insert Tc (S~ SCXl ) = 1 after SV (to - if3n, to) in the
denominator of Eq. (13.57) , and then divide the numerator and denominator by
Tr [e -.BC HO-{(N) ]; the result is
(13.59)
where the ensemble average is now over the non interacting system. Substituting the
perturbation expansion for S~, and S'CXl into the above expression, Wick's'theorem
(applicable here since the ensemble average is over the non interacting system),
yields a perturbation series for C c( 1, I ').
The above expression for CeO , 1') can be used to study the behavior of a system
out of equilibrium at times t > to , after an external perturbation has been switched
on at time to , while taking into account the initial correlations at t = to. Indeed, we
have used p(to) = e-fJC H-IJ. N) / Tr[e - .BCH-IJ. N)J, which includes interactions among
the particles. In many cases, however, we are only interested in studying the
behavior of a system for times t » to. For example, regarding the system depicted
in Figure 13.1 , we may be interested in its steady state after all transients have
died off. The steady state, if it develops at times t » to, will not depend on the
initial state at time to. In such a case, we may use the statistical operator of
the noninteracting system instead of that of the interacting system . Alternatively, if
we are only interested in the behavior of the system for times t » to, we may take
to = -(X) and assume that interactions are turned on adiabatically (but not turned
off) . In this case, the statistical operator p(to = -(X)) is that of a noninteracting
system, and the branch of the contour C that extends from to to to - if3h may
be dropped, i.e., C coincides with C. The contour C now extends from -(X) to
max(t. t' ) and back to -(X). We might as well extend the contour to +00 so that
it runs from -(X) to +00 and back to -(X) ; this is the Keldysh contour (Keldysh,
1965), depicted in Figure 13 .6. The expression for the contour Green's function in
13.6 Dyson '5 equation 347
-00 --e_---_e- 00
(
t t'
Figure 13.6 Keldysh contour C: it runs along the real time axis from -00 to +00
and back to -00, passing through t and r' exactly once.
Uext
~ ~*
~+ -:=~=+~
1 l' 1 l' 1 2 l' 1 2 3 l'
Figure 13.7 Dyson 's equation. A double line represents G e , the contour Green 's
function for the interacting system , while a single line represents G ~, the cor-
responding contour Green 's function for the noninteracting system. Uext is a
one-body external potential and ~ * is the ilTeducible self energy arising from
interparticle interactions. In the figure , I = (ra r).
(13 .60)
),(
G e(l, I ') = L G~ (l ,
G~(1 , I ') + d2 2) Uexl (2) G c(2, I ')
+ L L G~(1,
d2 d3 2)L: *(2, 3)G e(3 , I ') (13.62)
where UCXl is the external potential giving rise to the external perturbation,
Hexl(t) = L
A "'f 1/1At (rat)U exl (rat)1/I(rat)d 3 r.
A
and L: * is the irreducible self energy resulting from the pairwise interaction V
amon g the particles of the system. In Eq. (13.62) ,
where G ~U G c and G ~ L: * G e stand for the second and third terms, respectively, in
Eq. (13.62).
We can also write Dyson's equation in an alternative form. Noting that
G e = G ~ (l + UG e + L: *G J =} 1 + UG e + L: *G e = G ~- !G e
Gc = G~ + Gc UG~ + G c L: * G ~. (13.64)
This form of Dyson 's equation is depicted in Figure 13.8. It is clear that Figures
13.7 and 13.8 produce identical perturbation series for G e .
In practice, can"ying out calculations with contour integrals is not convenient,
and they should be reexpressed in terms of real-time integrals. The procedure
for converting contour-time integrals into real-time integrals is known as analytic
continuation (a misnomer, since the contour is attached to the real-time axis and
no continuation from the complex plane takes place). The rules for this procedure
are taken up next.
J3.7 Langrefh rules 349
where B is another function depe ndent on two contour times and C is the contour
shown in Figure 13.6. The lesser function C <(t , t ') is given by
C <(t, t') = C(t E C, t' E C) == C(t->, t '+-) = l drl A(t->, rl)B(rl , t '+-)
= f ood tl A(t->, t J") B(t J", t '+-) + l -~tl A(t->, t t) B(t t , t '+-)
-00 00
This provides an expression for the lesser function in terms of functions of real
time integrated over the real-time axis.
Similarly, we can obtain an expression for C >(t. t '),
C >(t, t ' ) = C(t E C, t ' E C) == C(r <- , f l-» = l drIA(r <-, !1)S(rl, t l-»
= i~'t, l
[A >(r, tl)S1'(tI, t' ) - AT(r, tl)S > (rI , t )].
It follows from Eg. (13.47) that S1' = S> + So and AT = A > - N. Hence,
C >(t , t ' ) = i : dtl [Ar(f , tl)S >(fl , t ' ) + A >(t. tl)So(t l , t ' )]. (13.70)
From Eg. (13.66), C(t, t ' ) = e(t - t ' ) [C >(r , t ' ) - C « f , f l)]' Using Egs (13.69)
and (13.70), we obtain
C(r , t ' ) = e(f - t ' ) i : dtl {[A >(t , tl) - A <(r, fl)] SO(tI' t ' )
C' (t, t') = 8(t - t') {[j t, [A > (t, t,) - A « t, t,)] [8 <(1" t') - 8>(t,. t')]
Since t > [', due to the step function e(t - t '), it follows that, in the above integrand,
I:
t > tl > t' ; hence,
l l
C(t , t ) = [ : dtIA'(t , t,)B'(t" t ). (13.71)
l f
Ca(t , t ) = [ : dtIA" (t, tl)Ba(tl , t ). (13.72)
C= AB (13.73)
(13.74)
(13.75)
(13.76)
(13.78)
(13 .79)
(13 .80)
GR ,A = G OR . A (1 + UG R . A + L; * R.AG R . A ) (13.81a)
G R.A = (1 + GR .A U + GR.AL; * R.A) G OR . A . (13 .81b)
Note that since U depends on only one time, it is neither retarded nor advanced.
Although the above equations look like Dyson 's equation for the equilibrium
Green 's function , there is a subtle distinction: L; * R.A depends not only on GOR , A,
352 Nonequilibrium Green's junction
but also on G < and G>. We rearrange the first of these equations as follows:
G OR .A = (1 _ GOR.AU _ GOR.AL *R.A) G R.A .
( 13.82)
If the rules presented in Eqs (13.74) and (13.78) are now applied to the first form
of Dyson's equation, (13 .63), the result is
G < = GO < (1 + UG A + L *AG A) + GORL *< G A + (GORU + GORL*R) G<.
The same expression holds for G> if we replace < with> everywhere. Rearranging
terms, the above expression for G < is written as
(1- GORU - GORL *R) C < = GO < (1 + UC A + L *AG A) + GORL *< G A.
Multiplying by (l + GRU + GRL *R) on the left, and using Eqs (l3.81b) and
(13.82), we obtain
At time to, the components are brought into contact, and a couplin g between the
dot and the two leads is established, allowing electrons to tunnel from the leads to
J3.9 Steady-state transport 353
the dot and vice versa. The perturbation that drives the system out of equilibrium
is the coupling between the dot and the two leads.
We choose to analyze the problem in the setting described above, although other
choices are possible. For example, the three components may be initially in contact
and in equilibrium at a common chemical potential. In this approach, the initial
statistical operator is e- f3 (H -f.1 N ) ITr e- f3 ( H - f.1 N ) , where H is the Hamiltonian that
includes the coupling between the dot and the leads, fJv is the common chemical
potential, and N is the number of particles operator for the whole system. The
perturbation that drives the system out of equilibrium is the increase in the chemical
potential of the left lead due to an applied bias voltage (Cini, 1980; Stefanucci and
Almbladh, 2004). An increase in the chemical potential by 6 means an increase,
in the amount of 6, of the energy of each single-paIticle state in the left lead. The
first approach is simpler, since our purpose is to study steady-state transport across
the quantum dot.
(13.86)
HL and HR are the Hamiltonians for the left and right leads, respectively,
Ha = L EkaCtaackaa, Ci = L, R (13.87)
ka
where (5 is the spin projection, and k is a collective index representing the spatial
quantum numbers of the electronic states in the leads. In writing Ha, we have
assumed that the electrons in the leads are noninteracting, except for a possible
average interaction which can be taken into account through a renormalization of
the single-particle state energies 10k. Neglecting correlations in metals generally
yields a good approximation, especially for simple metals. The term H D is the
Hamiltonian for the dot,
(13.88)
electronic states in the dot. Various model Hamiltonians for the dot may be chosen.
354 Nonequilibriul11 Green 'sfun cfion
Figure j 3.9 In the Anderson impurity model, the dot has only one energy level ,
and it ca n be occupied by up to two electrons. If there is only one electron. the
energy of the dot is E; if the level is doubly occupied, the energy of the dot is
2E + U. The two electrons that occupy the level must necessarily have opposite
spin projection s.
where na (a =t , -!-) is the operator that represents the number of electrons in the
level with spin projection a. If one electron occupies thi s level, the energy of the dot
is E. However, if two electrons occupy the level , one with spin up and the other with
spin down , the energy of the dot is 2E + U, where U > 0 is the Coulomb repulsion
energy of the two electrons (see Figure 13.9). Other model Hamiltonians for the
dot may be considered; e.g., one may be formulated that includes the interaction
between electrons and atomic vibrations in the dot.
The coupling between the dot and the leads is given by the last term in the
Hamiltonian, H T . The coupling is represented by terms that describe tunneling of
electrons from the dot to the leads, and vice versa:
NL = L cLLCkuL. (13.93)
ku
Vk*aL ,ma
T : T' = T
t
===>===;..--~)--- T'
n~ k~L n~
Figure 13.10 Graphical representation of the Dyson-like equation for the contour
Green 's function GII .kL (r. r /: a).
(13.99)
where G~L (rl, r ' ;~) is the non interacting contour Green's function of the electrons
in the left lead, and Gllm(r , rl ; ~) is the contour Green's function of the electrons
in the dot:
(13.100)
(13.101)
Alternatively, Eg. (13.99) can be delived by expanding GII .H(r , r ' ;~) in a pertur-
bation series and applying Wick's theorem to contract the creation and annihilation
operators of the electrons in the left lead (this is made possible by assuming that
the leads contain noninteracting electrons) .
13.9 Steady-state transport 357
We now apply the Langreth rule, Eg. (1 3.69), to obtain the lesser function
As indicated earlier, our interest is in studying the system in the steady state, i.e., at
times long after the moment when the perturbation is switched on. In this state, the
current is independent of time, and all Green's functions depend on the difference
between their time arguments. We can then Fourier transform the various functions
that appear in Eg. (13.102):
Similar expressions are written for the other functions that appear on the RHS of
i:
Eg. (13.102). Using the relation
we obtain
<
G l1.kL ( I. )-_ 2nn ,,/00
t, t , a L
I d
we
- icv(r-r') V *
kaL.lI1a X
m -00
(13.l04)
Setting t' = t gives us G,~.klJt , t; a). Insel1ing this into Eg. (13.98), we find
It
e
= n!i 2 _
/ 00 dwRe L L . x
00 11111 ka
'v'kaL,l1a v taL./I1a
(13.105)
(13.106)
(13.107)
where fLew) is the Fermi function in the left lead (see Problem 6.6 and Eg. [6 .56]).
We also note that
< I ,
G l1m (t - t ,a) -
_.
l t )dn(t) ) - -
(d/l1(t
/ _ 1 / 00dwe . ( ') G
-1(1) [ - ( <
IlIl1
.
(w , a)
2n -00
358 Nonequilibrium Creen 's function
(13.108)
A similar calculation yields
L
rmll(w; 2:rr '"'
a) = -Ii * ma VkaL.llaO(W - EkaL/fi).
~ VkaL (13.111)
1 .
k
h = -e2 / 00 dU)X(U))
nli -00
x = ReLI>kaL.lla VtaL.ma[C~II(U);a)C~L« U);a) + C/~,(U);a)C~LA(U);a)]
11m ka
The arguments of the functions have been suppressed for now. Using Egs (13.108)
and (13.109),
The zeroth-order functions that appear in the above expression are known,
l
C~LA - G ~l = (w - EkaL/n - iO+r - (w - Ekadh + iO+r l
= 2ni8(w - EkaL/n)
L
rmll(w; a) = -1
h
L * ma VkaL.lla 2n8 (w - EkaL/n).
VkaL
'
k
The expression for X(w) reduces to
in
X(w) ="2 L r ';;ll(w;a) {!L(w) [C~m(w; a) - C :m(w;a)] + C/~Il(w; a)1
unm
=
ih
"2 LTr {r L(w; a ) [!L (w) (CR(w;a) - C A(w;a)) + C « w;a)]}.
a
The current from the right lead to the central dot has exactly the same expression
as I L except that R replaces L. In the steady state, I = I L = - I R. A symmetrical
expression for the current is obtained by writing I = (ft - IR) /2; it is given by
I = -I:
ie f dwTr{[JL(w)rL(w;
OO a) - iR(W)rR(w; a)][ GR(w; a) - G A(w; a)]
47T1i a
-00
T T'= T-~)--T' + T)
f )
f ;
ncr mcr ncr mcr ncr lcr Tl kcra T2 pcr
Figure 13.11 Graphical representation of Dyson 's equation for the contour
Green's function Gll m(r , r '; 0") of a quantum dot in contact with two metal leads .
This is the Meir-Wingreen formula (Meir and Win green, 1992) for the current in
the steady state; it expresses the current in terms of the Green's functions of the
dot that is in the central region, between the two leads. In general, the calculation
of these functions is highly nontrivial.
A simplification is possible in the special case of proportional coupling, when
the left and light level-width functions are proportional: rL(w) = ArR(W) . Since
1 = h = -I R , we can write 1 = x h - (1 - x)1 R for an arbitrary x. The current
is then given by
I=-L
ie
2Jrn
1 dw a
00
-00
L 1 dw [fLeW) -
00
1 = - ie !R(W)]
2Jrn a
-00
(13.113)
The ratio of the coupling matrices is well defined, since rL and r Rare propoltional.
The condition of propOltional coupling does not generally hold, but it may be a
reasonable approximation if the level-width functions have a weak dependence on
energy.
To evaluate the current, given by Eq. (13.110) or Eq. (13 .112), we need C R - C A
and G < . These are obtained from the contour Green's function , which, in turn, is
determined by Dyson 's equation ; this is shown graphically in Figure (13.11).
13.10 Noninteracting quantum dot 361
G R = G OR + G OR 2, RG R , G A = GOA + C OA 2, AG A (13.118)
where G OR and GOA are diagonal matrices, and
t;h;:,R.A( w; a ) = 'L" V* OR A
/1 LJlp kaa.la C ka . (w; a )Vkaa.pa· (13.119)
ka
The self energy and the level-width function are related,
Since the matrices COR and C OA are diagonal , (CORr' and (C OA r ' are also
diagon al. Furthermore,
(C OR )- '
IlII
= W - E" 1ft + I'0+ ,
I:
(C OA )- ' =w-E" In-/O.
1111
'+
It follows that (CORr' = (C OA r' .Therefore,
(CRr' + I;R = (CAr' + I; A.
Multiplying by C R on the left and by C A on the right, we obtain
Using Eq. (13.120) to replace (I;R - I; A) by (-i In) (rL + rR), Eq . (13.121) is
obtained.
In order to obtain C <, we use the Keldysh equation (13.83) . Here, U = 0 (see
Eq. [3.117]), and the Keldysh equation reduces to
Since C~~ = 2nij(w)o(w - El1lfI)o/l1/1 and (GOA):,,', = (w - EI1 /fI - iO+ )o/l/11, it
follows that C o< (G OA) - , = O. The Keldysh equation thus takes the simple form
(13.124)
(13.125)
Therefore,
(13 .126)
Putting Eqs (13.121) and (13 .126) into the equation for the current, Eq. (13.110),
we obtain
I = --e
?
2nfI-
1-00
00
where
(13.128)
a
(13.129)
a
where E is the energy of the level, V > 0 is the onsite Coulomb repulsion, and
n t(n -l.) is the operator representing the number of electrons in the level with spin up
(down). For simplicity, we assume that rL = Ar R , so that we can use Eq. (13.113)
for the current. In this case, where there is only a single level in the quantum dot, the
quantities r L , r R , G R , and G A are all scalars. From the spectral representations
of G R and G A (see Eqs (6.36) and (6.37» we deduce that ReC R = ReG A and
ImC R = -lmC A ; it follows that G R - C A = 2i ImC R . The expression for the
current thus takes the form:
I = =- .L / -0000
nli
e
a
dw [!L(w) - !H(w)] x
(13.130)
Here, GR(w; 0") is the Fourier transform of the retarded Green's function C:(t)
of the single-level quantum dot in the presence of coupling to the leads. C:(t) is
defined by
(13.131)
Let us denote the retarded Green 's function of the isolated quantum dot by D:.
This function was calculated in Chapter 7 (see Eq. [7.11]); it is given by
I - (11 -) (11 -)
DR(W) = a + a (13.132)
a W - Elli + iO+ w - (E + U)11i + iO+
where no- = l1 _ a . This is the exact retarded Green's function for the isolated single-
level quantum dot.
364 Nonequilibrium Green '5.tim e/ion
Vk*a V ka
)
+ )
f )
>f ;
dO" dO" dO" kO"a dO"
Fi gure 13. 12 Dyson's equation for the contour Green' s function of the quantum
dot pl aced between two non interacting metal leads .
Our next step is to calculate the retarded Green's function C: of the quantum
dot in the presence of the tunneling Hamiltonian,
We have assumed that the tunneling matrix elements are spin-independent; that
assumption, along with the assumption of a single level in the dot, makes it super-
fluous to add any additional subscripts to Vka and Vk~'
The exact calculation of the dot's retarded Green 's function is not possible, so
we must resort to approximations. Our approach will be to treat the exact Green 's
function D: of the isolated dot as the unperturbed function C3a(w), and then
calculate a correction resulting from the inclusion of tunneling. The assumption
made here is that the coupling between the dot and the leads is weak.
Dyson's equation for the contour Green 's function of the dot, C da(W), is depicted
graphically in Figure 13.12. We can thus write
(13.134)
(13.135)
Replacing C~:(w) with the exact retarded Green's function of the isolated dot (see
Eq. [13.132]), we obtain
!1W-E-U+ (n -)U
C R (w) = a . (13137)
da (w - E/!1)(!1 w - E - U) - "'£:(w )[!1w - E - U + (n6)U] .
J3. J J Coulomb blockade in (he Anderson model 365
v
Figure 13.13 A plot of J vs V in the Coulomb blockade regime.
This is too complicated. As an approximation, we ignore the real part of 2:,:, which
only leads to a small shift in the energy level of the dot. If we also ignore the w-
dependence of 2:,:, then, using Eqs (13.120) and (13.135), we obtain 2:,R = - 2:, A =
-i (f L + f R)/2ft, where f Land f R are constants. However, the calculation of the
current is still complicated because of the presence of (na) in the expression for
G:o(w); this is given by
(13.138)
The lesser function is given by the Keldysh equation (13 .83), which again reduces
to the simple form:
(13.139)
where
2:;(w) ~~
= 11- L
IVkCll 2 G~:-a(w). G~:-a(w) = 2niiCl(w)8(w - Eka/I1).
ka
It follows that
< i ~
2: a (w) = - L iCl(w)f Cl (13.140)
11 a
where the w-dependence in f Land f R is ignored. We thus see that (na) depends on
the retarded Green's function of the quantum dot, G: a , which, in turn, depends on
no. A self-consistent solution for G: o is called for, using numerical techniques. If
such a calculation is carried out, using reasonable values for the various parameters,
a plot of the current I versus the applied bias voltage V, as in Figure 13.13, will
be obtained (Pals and Mackinnon, 1996; Swirkowicz et aI., 2004; Zimbovskaya,
2008).
A qualitative understanding of how the current varies with bias voltage is gained
by appealing to the so-called Coulomb blockade model. Current flow requires the
366 Nonequi/ibriul17 Green's fun ction
addition of an electron to the dot between the metal leads. Adding an electron to
the dot costs an energy of e2 /2C, where C is the capacitance between the dot and
the leads. Since the quantum dot is of nanoscale size, C is extremely small, and
e 2 /2C may attain a large value. There is thus an energy balTier, called a Coulomb
blockade, to the flow of current through the quantum dot: no current flow takes
place unless the bias voltage exceeds a threshold value. A plateau in the /- V plot
results from the fact that the energy barrier to the flow of two electrons is larger
than the barrier to the flow of one electron (Kastner, 1993).
The first step in the /- V plot corresponds to the addition of one electron to the
dot, while the second step results from the addition of two electrons. The first step
is twice as large as the second step: the first added electron may have its spin up or
down, but the second added electron can be of only one spin orientation, the one
opposite to that of the first added electron.
Further reading
Datta, S. (2005). Quantum Transport: Atom to Transistor. Cambridge: Cambridge Univer-
sity Press.
Di Ventra, M. (2008). Electrical Transport in Nanoscale Systems. Cambridge: Cambridge
University Press.
Haug, H. and Jauho. A.-P. (1996). Quantum Kinetics in Transport and Optics in Semicon-
ductors. Berlin: Springer.
Rammer, J. and Smith, H. (1986). Quantum Field-Theoretical Methods in Transport Theory
of Metals. Reviews of Modern Physics 58, 323-359.
Problems
13.] Evolution operator. Derive Eq. (13.7).
13.4 Scaftering matrix. Show that S(t, t' ) satisfies the properties given in Eq.
(13.26) .
13.6 G c in terms of interaction picture operators. Prove the validity ofEq. (13.49)
for the case when t ' lies further along the contour than t.
13.7 Langreth's rule for the advanced function. Derive Eq. (13.72).
13.8 The Keldysh equation for the greater function. Derive Eq. (13.84).
Problems 367
13.9 Retarded and advanced functions. Prove the validity of Eq. (13.109) which
relates the retarded and advanced Green's functions.
13.10 Conductance. Using the Landauer formula for the current, show that the
low-temperature, zero-bias conductance, defined by G = d I I d V Iv=o, is
given by .
E F is the Fermi energy, and Til'S are the eigenvalues of the matrix T(E F Iii),
where T is given by Eq. (13.128).
I. ) =
G(r, r ,w
L ¢v(r)¢~(r')
.
v W - E v Iii + iO+
Show that
Assuming that there are N basis functions in the left lead (N --+ (0), N
basis functions in the right lead, and M basis functions in the molecule
(M is finite) , HLL and HRR are each an N x N matrix, while Hcc is an
M x M matrix . H LC and H RC describe coupling between the molecule and
the leads. The matrix H is real and symmetric.
The retarded Green's operator is
GLL
G(w) = GCL
[
G RL
According to the previous problem, G(w) is obtained by solving the equation
(w - H /n + iO+ )G(w) = 1.
Show that
where
g L (g R) is the retarded Green's operator for the isolated left (right) lead:
A.I Fermions
A.i.l One-body operators
Consider the one-body operator Ho = L;=, h(i). Let 14>, ).14>2) , ... be a com-
plete set of orthonormal single-particle states. Ho acts upon the vector space
y (N) = y, 0 Y 2 0 ... 0 V N , the direct product space of the spaces of the N
particles. The vector space V;, the Hilbert space of particle i upon which h(i)
acts, is spanned by the basis set 14>, );, 14>2) ;, .... For any l4>v) ; in this basis set,
(r l4>v) ; = 4>v(i); for example, (rlka ); = Jve;kr, la ); , where r; is the position vec-
tor of particle i and la ) i is its spin state. The orthonormality of the basis states
means that i(4)vl4>v') ; = ov,," and completeness means that Lv l4>v); ;(4),,1 = 1.
For the case l4>v ) = Ika ), these relations mean that ; (ka Ik'a ' ); = Okk'Oaa' and
Lka Ika ); ; (ka 1 = 1. Note that an expression such as ; (4),, l4>v' ) j, for i =I- j, is not
an inner product because 14>,,) ; and l4>v' ) j belong to different vector spaces; it is,
in fact, the operator l4>v' ) j ; (4)v I. Inserting the completeness relation into Ro, we
obtain
vv'
vv' vv' j
369
370 Second quantized form of operators
The matrix element i (¢I>' Ih(i ) I¢v)i is independent of i , since the coordinates of
particle i are integrated over, and it is written as (¢v' Ih I¢v). Thus,
N
The Slater determinants form a properly symmetrized basis for the expansion of
the N -fermion wave function. We consider how Rv'v acts on an arbitrary Slater
determinant llJl ) = l¢vl ",¢v,v ). Ifv 'f. {VI , ... , vN}then V 'f. {P(VI) , ... , P(VN) },
since the sets {VI , .. " VN} and {P(VI) , ... , P(VN) } are identical (the elements of
the second set are merely a permutation of the elements of the first set). Since
V 'f. {P(VI).···, P(VN)),
1 ~ P
i(¢vl lJl ) = !.7f i(¢vl L..,.,(-I) l¢p(vl))I" ·1¢P(vtl)i·· ·1¢P(v,v ))N
.y N! P
1 ~ P
= !.7f L..,.,( -I) I¢P(vil) I ... i (¢v I¢P(v)i .. ·I¢p(v,v )) N = O.
.y N! P
The last equality follows since V f. P(Vi). Therefore, unless v E {VI , ... , VN} the
action of R v'v on 11Jl) yields zero. So let us assume that v = vj . Then
The sum is now over the permutations of coordinates. Recall that the Slater deter-
minant has two equivalent forms: the sum in one form is over the permutations
of coordinates, while in the other form, it is over the permutations of indices. In
the summation over i from 1 to N, each i belongs to the set {P(l) , .. . , P(N)),
its elements being a permutation of I , ... , N. Since the single-particle states are
orthonormal, only when i = P(j) will Rv'v 11Jl) be nonzero. When i = P(j) ,
I¢v')i i(¢vl¢v)p(j) = I¢v')i = I¢v' )p(j)' Hence, the result of the action of Rv'v on
11Jl ) is simply to replace I¢ v) in 11Jl) by I¢v')'
No minus sign is needed: 1>v is moved to the leftmost position, replaced by 1>v"
which is then moved back to the original position of 1>v. If the first movement
produced a minus sign , so would the second. Since 1\lJ) is an arbitrary Slater
determinant, we conclude that
(A.3)
(AA)
vv'
I
H = 21 [v(l, 2) + v(2 , 1) + vel. 3) + ... + v(N. N - I)]
= ~ [Lkl ml/
l1>k1>/) 1.2 1.2 (1)k1>llv(l , 2)11>11/1>,, ) 1.2 1.2 (1)m1>,, 1+ ...
+L
klml/
l1>k1>/) N.N- 1 N.N- r(1)k1>ilv(N, N - 1)11>1I/1>,, )N .N- 1 N.N- l (1)m1>,, I] .
In the above equation,
The matrix element i.j (1)k1>ilv(i , J)11>1I/1>I/ )i. j is independent of i and j, since the
coordinates of i and j are integrated over, and we write it simply as (1)k1>ilvl1>II/1>I1).
Hence,
i#j j.j#i
372 Second quantizedjorm oj operators
We have used Eqs CA.2) and CA.3). The anticommutator {cm. c;} = Oml; it follows
that Cm c) = 0111 1 - c; Cm . Hence,
(A.S)
A.2 Bosons
A.2.1 One-body operators
We assume that we have a complete set I<PI ), 1<P2), ... of orthonormal single-particle
states. For a system of N identical bosons, the basis states are
n 1< is the number of times the state 1<p1l) appears in the product, i.e., n il is the number
of particles that occupy the single-particle state 1<p1l ). The one-body operator Ho =
Li h(i) is given by
N
(A. 8)
A.2 Bosons 373
1 lI N
R ' v l<1>B) =
V
Jn v!n~, ! nN -Jiif ~ ~ l1'v')i i(1'vl
I1'lv . v'
[ .. ·I1'v)P(id· · ·I1'v)P(i n l
.) ·· ·I1',,' )p(jI)·· 'l1'v' )p(J,,')'" J. (A.9)
Whenever i E {PU I)' ... , PCiIl..) }, the result of the action of l1'v')i i (1'v I is to remove
one particle from state l1'v) and add a particle into state l1'v'), i.e. , it produces a
state with nv - I particles in state l1'v) and n:" + 1 particles in state 11',,')' For
i rf. {PCiI) ," " P(i ll v)}' the action of l1'v')i d1'v l on I<1>B ) yields zero. We thus find
XL [.. 'l1' v)P(k l )" · I1' v)P(k" v- J! ·· ·I1'v' )puJ!· · '11'",)p(I</+ I)'" J.
P
Noting that
nv .;n.,.,jn v, + 1
Jnv !n~, ! - J(n v - I)!« , + L)! '
we find, for v f. Vi,
Now consider the case Vi = v. In Eq. (A.9), whenever i E {PUI) , ... , PUll.)}, the
action of l1'v)i i (1'v l on 1<1>8) leaves 1<1>8) unchanged, since it simply removes a
particle from the single-particle state 11'1!) and adds a particle into the sa me state.
For i rf. {PCi I)' ... . PUll..) }, the action of l1'v) i i (1',, 1on I<1>B ) yields zero. Therefore,
N
Rvvl <1>B ) = R"lI l ' .. nv ... ) = L l1'v)i i (1'v l ... nv ... )
i=l
We conclude that Rv'v = a~, a". The one-body operator is thus given by
(A. 10)
374 Second quantized form of operators
2.
H I = I '~" (k/lvlmn)Aklllll/
kIm/!
The commutation relation [am, a,t] = Oml, for bosonic operators, implies that
ama /t = Oml + a/t am; hence
tt
H =
I
2.1~
",
(kllvlmn)aka , aI/am' (A.ll)
klllll/
This is the same expression as in the case of fermions.
Appendix B
Completing the proof of Dzyaloshinski's rules
We asserted in Section 9.8 that the sole surviving term is obtained by keeping
only the contribution at the upper limit of each integral for all integrations over
T PI ' T P2' ... , T PN- I ' To prove this assertion, consider the term obtained by keeping
only the lower limits on the first d l integrals over TP I , TP2 , ... , Tp"I ' then only the
upper limits on the next U I integrals over T P" I + 1' ..• T P" I"''' I ' then only the lower limits
on the next d 2 integrals, then only the upper limits on the next U2 integrals, and so
on. Only the upper limits are kept on the last Uk integrals. Let 51 be this sequence of
lower and upper limits, 51: Ukdk ... U2d2Uldl. The diagram f N , I' without external
lines, is represented in Figure B .1, where, for simplicity, times are arranged in
decreasing order from left to right (instead of top to bottom). In the figure, we lump
the Uk veltices T PN ' ... , T PN- "k+ 1 together into L I, and the vertices T PN-"k ' ... , T PI
into L 2. We show fermion and boson I ines that connect the two lumps L I and L 2.
The term obtained by sequence 51 is denoted by og~:V. I. We will show that this
term is cancelled out by another term that arises from a sequence in a different
time-ordered diagram.
Consider the time-ordered diagram f N,2, obtained from a second permuta-
tion which orders the times as follows: T PN-" k > ... > T PI > T PN > ... > T PN-"k+ I '
This diagram is the same as (hat in Figure B.1 except that the lumps L I and
L2 are interchanged. As a result of this interchange, the lines connecting the
lumps have their directions reversed. In evaluating ogfN.2, consider the sequence
52: dkUk-Idk-I ... U2d2UI(dl + l)(uk - 1). Note that, in 52, the lower limit from
375
376 Completing the proof of Dzyaloshinski 's rules
Figure B.l A time-ordered diagram , with time decreasing from left to right, is
divided into two lumps that are connected by fermion (solid) and boson (wavy)
lines. The external lines are not shown.
the integration over TPN is the one that is kept. We will show that the term 8g~;2
exactly cancels out the term 8g~,N '. It is easy to check that, in both 8g~,N ' and 8g~2~·2 ,
time integrations produce the same denominators. We also note that 8g~;·2 differs
from 8g~,N' by the following :
1. A factor of -1 , which results from keeping the lower limit rather than the upper
limit in the integral over T PN •
2. A factor of exp[ - ,81i( EPN + ... + EPN - Uk + I )]; this results from the fact that, in
8g~,N', we kept only the upper limits in the last Uk integrals, which produces
the factor exp[,8Ii( EPN + ... + EPN - U1 +). In 8g ~;2 we kept only the lower limit
in the integration over T PN .
3. A factor of (-l/exp[-,81i L~=N-1I1+1 k
L7:t
( EPi Pj - Epj P)], where J is the
number of internal fermion lines connecting L I to L 2 . The reason for this factor is
the following. As a result of interchanging L I and L 2 , all lines connecting L I and
L 2 are reversed . For each vertex TPi in Figure B.l , i = N, N - 1, N - U k + 1,
the lines connecting T Pi to all T P E L2 are reversed. Upon reversal of lines, for
j
each fermion line directed from TPi ELI to Tp E L2, the factor -(1 - I E!'p ) j , j
must be replaced by I EP!' = -exp( -,8liEp,p) [ -(1 - I EPP )]' whereas for each
I j I }
fermion line directed from Tp E L2 to r Pi ELI, the factor lE I'/' must be replaced
j
.
j ,
for each boson line directed from TPi E LI to Tp j E L 2 , the factor (1 + n EP !' ) , j
must be replaced by n Epp = exp( - ,8liE P p . )( 1 + n E p p), whereas for each boson
I j I j I )
(1 + n Epp ) = exp(,8liE P p)n Epp . Hence, the reversal of the lines produces the
j / j j I
r r ,_'v., __ _
o s.
(_l)Je -fJli(€PN+ "+EPN
- ",-
I)
exp
[f3~~ N~(l Ep,Pj -
- . ~7L ~ EpjP,)] ~rN I
05 1 ' .
I - N - lI k+ 1 J- I
L Ep;
L
i = N - II ,+ 1
N N- li k N N
Therefore,
If we are interested in the self energy of a boson, then J is even and wI11 = 2n n I / f3h,
wIl 2 = 2nn 2/ f311 ; this gives og~;' = _O~INI . On the other hand, if we are interested
in the self energy of a fermion, then:
(a) If both external lines enter and leave L I , then J is even, and
(b) If both external lines enter and leave L 2 , then J is even, and og~;V2 = _O~INI .
(c) If one external line enters LI and the other line leaves L 2 , then J is odd, and
ogr N.2 = _(_l)Odd eiw",fJliogrN. , = _ogr51N.I .
52 51
(d) If one extemalline enters L2 and the other line leaves L I, then J is odd and
og~;V2 = _( -l t dd e-iw"2fJli og~;VI = _og~INI .
Rnl = Rn + d l is the equilibrium position of the atom, "III is its displacement from
equilibrium, Rn is the lattice vector from the origin to the center of unit cell n, and
d l is the equilibrium position vector of atom l, measured from the center of the unit
cell to which the atom belongs. The various terms are shown in Figure C. I.
Denoting the mass of atom l as MI , the kinetic energy of the atoms is
T = L P';la/2MI, a = x, y, Z. (C.2)
lila
2
V = Vo aV I Ullia + -1 L
+ L -- L aV I Un/aUII ' I' a' + .. ' . (C.3)
nla
aUnl a 0 2 ilIa 11' /'(1'
aUnlaaUII ' I' a' 0
Vo is the potential energy of the crystal when the atoms sit at their equilibrium
positions. Being merely a constant, Vo will be ignored (we can always set it equal
to zero by measuring energies relative to it). Since V is a minimum at equilibrium,
the second term in Eq. (C.3) vanishes. The harmonic approximation, which we
adopt here, consists in keeping only the third term in Eq. (C.3), ignoring higher
order terms. The justification for this is that the displacement from equi librium is
very small compared to the lattice spacing (this is generally true for temperatures
378
C.2 Classical theory of lattice vibrations 379
• • •
• • ~~
d 1 (12
0 Rn
Figure C.l A two-dimensional crystal with two atoms per unit cell. The center 0
of one unit cell is chosen as the origin of coordinates. R II is the vector from 0 to
the center of unit cell n. d I and d 2 are the equilibrium position vectors of the two
atoms relative to the center of the unit cell to which they belong. " Il l and " 112 are
the displacements from equilibrium of the two atoms that belong to unit cell n.
V = I ""'2
~ ""
~ ¢(nla , n 1I/ a , )UnlaUn'I'a" (C.4)
ilIa n' l /a'
trial solution
Dq(/a , I ' a ' ) = (M/ M/TI /2 L ¢(nla, n' L'a ' )eiq (R",- R,,) . (e.9)
/1 '
Since the summand depends on RII , - Rn and not on RI/ and Rn' separately, we
may set Rn = O. Relabeling /1 ' as n, we obtain
Dq(/a , l ' a ' ) = (M/M/ ,)- 1/2 L ¢(Ola , nl' a ' )e iq .R" . (e.IO)
1/
Dq(l'I) D q (l,2)
Dq(l,r)]
Dq(2, 1) D q (2,2) D q (2 , r)
D(q) = . (e.13)
[
Dq(~, 1) Dq(r, r)
Here, D q (/ , I ' ) is a 3 x 3 matrix whose aa' entry is D q.CtCt , (/, I' ) = Dq(/a, I' a ' ). We
also define the 3r-column polarization vector
(e.14)
C2 Classi cal theory of lattice vibrations 381
D~(la , l'a ') = (MI M II )- 1/2 L ¢(nla , nll'a ' )e- iq. (R",- Rn)
1/ '
1/ '
The first equality results from the definition of Dq([a, [' a') in Eq. (C.9) and
the fact that the force constants are real. The second equality is valid because
¢(nla , n' l 'a ' ) = ¢(n'l'a ' , nla). The hermiticity of D(q) implies that:
• The eigenvalues W~A ' 'A = I , . .. , 3r, are real. Furthermore, w qAis real; if it were
not, the displacement Unl would be a monotonically increasing or decreasing
function of time, rather than an oscillatory one .
• The 3r eigenvectors EA(q), 'A = 1, ... , 3r can be chosen to form a complete
orthonormal set; any 3r-column vector can be expanded in terms of them.
(1) The definition of Dq(la , l'a ' ), as given in Eq. (C.9), along with the equality
eiG .R" = 1, implies that Dq(la , I' a ' ) = Dq+G(la, I' a ' ) for any reciprocal lattice
vector G . The dynamical matrix thus satisfies the property D(q) = D(q + G);
consequently,
(C.17)
In other words, the mode (q + G'A) is identical to the mode (q'A). It is therefore
sufficient to restrict the values of q to the first Brillouin zone: q E FBZ; this
exhausts all possible normal modes. Since there are N q-points in the FBZ,
there are a total of 3 N r eigenvalues WqAand 3 N r cOlTesponding eigenvectors
EA(q) , q E FBZ, 'A = 1, ... , 3r.
(2) The dynamical matrix satisfies the relation D *(q) = D( -q), a consequence of
Eq. (C.II). Since WqAis real, it follows that
(C.18)
(4) If we plot w qA vs q along a certain direction in the FBZ, we obtain a curve for
each value of A.. Therefore, along any given direction in the FBZ, there are 3r
curves, or branches. Of these, three are acoustic branches, while the remaining
3r - 3 branches are optical branches. For the three acoustic branches, W ---+ 0
as q ---+ O.
The general solution of the equation of motion, Eq. (C.6), is a linear combination
of the 3 N r normal modes,
where the factor exp( -iwqAt) is absorbed into the expansion coefficients QqA(t) ,
and the factor 1/ JlV is inserted for later convenience. The coefficients QqA(t) are
called normal coordinates. Using the equality
and Eq. (C.19), it is not difficult to show that Eq. (C.20) gives
t -_
Q qA() JlV ~ t.7M
1 Lvi /VI I Ull ia ()
t (\ *(1) ( ) -iq.R"
.a q e . (C.21)
lila
Since Ulli a is real and Ef?: ( -q) = Ef?a (q), it follows that
(C.22)
Carrying out the summation over n: Ln e i(q+q') .R" = NOq' ._q , then over / and a :
Lla Ef?a(q)Ef,:a ( -q) = oU' , the expression for T reduces to
(C.23)
C4 Quantum theory of latrice vibrations 383
A.(nta ,
x 'P U) (q)E U') (q ' )e iq .R" ei q' .R",
n' I'a ' ) Q qA Q q ,),' EA. a A' .a ' .
WIiting eiqR" eiq' .R", = ei (q+ q' ). R" e iq' .( R",- R,, ), using Eq. (C.8), summing over n, and
using the orthonormality of polalization vectors, we find that
(C.24)
(C.2S)
(C.27)
Analogous to the case of the hannonic oscillator (see Section 1.2), we define two
new operators:
(C.29)
384 Lattice vibrations in three dimensions
We can use Eq. (C.28) to express the normal coordinates and momentum operators
in terms of the new operators:
(C.30)
The values of q are restricted to the first Brillouin zone: q E FBZ. The Hamiltonian
is thus a collection of3N r harmonic oscillators. The operator a~A (a qA) is interpreted
as a creation (annihilation) operator for a particle-like entity, called a phonon, of
wave vector q , branch index A, and energy nwqA.
Appendix D
Electron-phonon interaction in polar crystals
0.1 Polarization
Polar crystals are generally semiconductors or insulators that, at low temperatures,
have fully occupied valence bands and empty conduction bands. It is possible,
however, to introduce electrons into the conduction bands. For example, absorption
of photons of appropriate energy leads to the promotion of electrons from the
occupied valence bands to the empty conduction bands. Raising the temperature
produces a similar effect. In semiconductors, doping introduces free electrons into
the lowest conduction band (or free holes into the top valence band). The electron-
phonon interaction in these systems is not adequately described by the rigid-ion
approximation. In an optical mode, the ions in the unit cell move relative to each
other, resulting in an oscillating dipole moment which, in tum, gives rise to an
electric field that acts on the electrons. The electron-LO phonon interaction in
polar crystals is mainly the result of this coupling of electrons to the induced
electric field.
We consider the case of a cubic crystal with two atoms per unit cell. The ionic
charges are ±e*. The volume of the crystal is V, and the number of unit cells is
N. In the long wavelength limit (q -+ 0), the two ions in the unit cell vibrate out
of phase, while the displacements in one cell are almost identical to those in a
neighboring cell. We denote by u + (u_) the displacement of the positive (negative)
ion within a unit cell (see Figure D.I).
Due to ionic displacements, a dipole moment p = e*u, where u = u + - u _, is
induced in the unit cell. Since the ions are not rigid, the resulting electric field
further polarizes the ions; the polarization is thus
385
386 Electron-phonon interaction in polar crystals
-- -- -+ -+
-- -- -+ -+
Figure D.I In a long wavelength optical mode, the motion in neighboring cells
of a lattice is essentially the same. Within a unit cell, the two ions vibrate out of
phase. The dipole moment in a unit cell is p = e*u + - e*u _ .
collection of dipoles, each of which occupies one unit cell. The local field acting
on a dipole differs from the average macroscopic field E in the crystal; they are
related by the Lorentz relation:
This relation is derived in standard electricity and magnetism textbooks (see, for
example, [GIiffiths, 1999]). In what follows we adopt the cgs system of units .
In a long-wavelength optical mode, all positive ions in adjacent cells have almost
the same displacement u + , while all negative ions have almost the same displace-
ment u_, whose direction is opposite to that of u + . The short-range restoring force
acting on anyone ion is proportional to u + - u _ . The equations of motion of the
ions are thus given by
(D.3a)
(D.3b)
y is a constant related to the strength of the short-range restoring force between the
ions. Multiplying the second equation by M + / M _ and subtracting the result from
the first equation, we obtain
where M = M+M_/CM+ + M_ ) is the reduced mass of the two ions in a unit cell.
Defining w = J N M / Vu, a bit of algebra shows that Eqs (D.I), (D.2), and (DA)
D.I Polarization 387
give the following:
(D.Sa)
(D.Sb)
The coefficients b ll , b 12 , and b22 are constants. We can express these coefficients
in terms of experimentally measurable quantities. In the presence of an externally
applied static electric field , the ions are displaced from their equilibrium positions
to new equilibrium positions, and a polarization develops even in the absence of
ionic vibrations . In the static case, W = 0, W = -(b l 2/ b ll )E, and P = (-bT2/ bll +
b 22 )E. Since, in this case, £(w = O)E = D = E + 4rrP, it follows that
? I .
b22 - bj,/b ll = -4 [£(0) - I] , (D.6)
- rr
where £(0) = £(q ---+ 0, W = 0) is the static dielectric constant.
If an external electric field with a frequency much higher than the vibrational
frequencies is applied, the ions will not be able to follow the fast variation of the
field; hence w = 0 and P = b22E. Since, in this case, £(oo)E = D = E + 4rrP, we
obtain
I
b 22 = - [£(00) - 1] , (D.7)
4rr
where £(00) = £( q ---+ 0 , W » Wphonon) is the high-frequency dielectric constant.
We write the solution to Eq . (D.S) in the form
°
( w , E , P) = (w 0, E 0 , P )ei (q.r -wl )
. (D .8)
Since V.D = 0 (there are no excess free charges in the crystal; it is neutral), it
follows that
where WLO is the frequency of the longitudinal optical phonon. The above relation
is known as the Lyddane-Sachs-Teller (LST) relation.
When the relation E = -4n P, valid for longitudinal optical vibrations, is sub-
stituted into Eq. (D.Sb), we obtain
b l2 h2
P= W= --w. (D.12)
I +4nb 22 &(00)
In the last step we used Eq. (D.7). Solving for b l 2 from Eqs (D .6), (D.7) , and
(D.I0), and replacing w wi th -J N M / V u, we obtain
P = WLO
NM
[ 4n V
(I
&(00) - &(0)
1)] 1/ 2
u. (D.13)
Un = ~
1 L QqE"Lqe
() . I q. RII
.
v NM q
where E" L (q) is the unit polarization vector of the LO phonon, and the normal
coordinates Qq satisfy the equation Qq = -wl
o Qq (the longitudinal optical modes
are assumed to be dispersionless: Wq = WLO). In the continuum description we thus
write
u(r) = 1"
~~ QqE"L(q)e,q·r. .
vNMq
p(r)=-Y'.P=-iWLO - -
471' V [ 1 (1 ----
£(00) £(0)
1)] 1/ 2 I
"" Q q.fL(q)e,q·r.
~ q
q
.
(D.14)
In the sum over q, the q = 0 term vanishes; the prime on the summation indicates
that the q = 0 term is excluded. The electron-LO phonon interaction is given by
p(r)d3r
H e- LO = -e
L
j
¢(r j) = -e
LJ Ir - rl
j }
[ 1 (1
= iewLO 471'V
1)]1/2
£(00) - £(0) L L
I
Qqq.fL(q)
J iq
e ,rd 3 r
Ir - rl
} q }
A change of variable r ~ r + r j shows that the above integral is simply the Fourier
transform of l / r , which is equal to 471' / q2. We thus obtain
[ 1 (1
He-Lo = iewLO - - - - - -0)
471' V £( 00) £(
I )] 1/ 2
L
q
I
Qqq.fL(q)(471'/q2)Le,qr j
j
.
•
(D. IS)
Finally, by writing Lj e iq .rj in second quantized form in terms of electron creation
and annihilation operators, and by expanding Qq in terms of the phonon creation
and annihilation operators, we obtain Eq. (11.46).
References
Abrikosov. A.A., Gorkov, L.P., and Dzyaloshinski, I.E. (1963). Methods of Quantum field
Theory in Statistical Physics. New York: Dover Publications .
Alyahyaei. H.M. and Ji shi , R.A. (2009). Theoretical investigation of magnetic order in
RFeAsO (R = Ce, Pr). Physical Review B 79 , 064516.
Anderson, PW. (1961). Localized magnetic states in metals. Physical Review 124, 41-53.
Bardeen, J. (1936). Electron exchange in the theory of metal s. Physical Review 50, 1098.
Bardeen, 1., Cooper, L.N., and Schrieffer, 1.R. (1957). Theory of superconductivity. Physical
Review 108, 1175- 1204.
Baym, G. and Sessler, A.M . (1963). Perturbation-theory rules for computing the self-energy
operator in quantum statistical mechanics. Physical Review 131, 2345-2349.
Bednorz, J.G. and MUller, K.A. ( 1986). Possible high Tc superconductivity in the Ba- La-
Cu-O system. Z. Physik. B 64, 189- 193.
Bogoliubov, N.N. (1958). On a new method in the theory of superconductivity, Nuovo
Cimento 7, 79-1-805.
Callen, H.B. and Welton , T.R. ( 1951). Irreversibility and generalized noise. Physical Review
83,34-40.
Cini, M. (1980). Time-dependent approach to electron transport through junctions: general
theory and simple applications. Physical Review B 22 , 5887-5899.
Cooper, L.N. (1956). Bound electron pairs in a degenerate Fermi gas. Physica l Review 104,
1189-1190.
Das, A. and Jish i, R.A. (1990) . Theory of interband Auger recombination in semiconductors.
Physical Review B 41, 3551-3560.
De la Cruz. C .. Huang, Q., Lynn, l .W. et al. (2008). Magnetic order close to superconduc-
tivity in the iron-based layer LaOI _xFxFeAs systems. Nature 453, 899-902.
Dresselhaus, M.S. and Dresselhaus, G. (1981). Intercalation compounds of graphite.
Advances in Physics 30, 139-326.
Duke, C.B. ( 1969). Tunneling in Solids. New York: Academic Press.
Dyson, FJ. (l9-19a). The radiation theories of Tomon aga, Schwinger, and Feynman. Phys-
ical Review 75, 486-502.
Dyson, FJ. ( 1949b). The S-matrix in quantum electrodynamics. Physical Review 75 , 1736-
1755.
Dzyaloshinski , I.E . ( 1962). A Di agram technique for evaluating transport coefficients in
statistical physics at finite temperatures. Soviet Physics JETP 15, 778-783.
Fermi, E. (1927). Un metodo statistico per la determinazione di alcune prioprieta
dell' atomo. Rend. Accad. Naz. Lincei 6, 602-607.
390
References 391
Langreth, D.C. (1975). Linear and nonlinear response theory with applications. In Linear
and Nonlinear Electron Transport in Solids. Vol. 17 of NATO Advanced Study Insti-
tute, Series B: Physics, ed. J.T. Devreese and VE. Van Doren , pp. 3-32. New York:
Plenum Press.
Lindhard,1. (1954). On the properties of a gas of charged particles. Det Kongelige Danske
Videnskabernes Selskab Matematisk-fysike Meddlelser 28, I-57.
London, F. and London, H. (1935). The electromagnetic equations of the superconductor.
Proceedings of the Royal Society (London) A149, 7l-88.
Loudon, R. (1963). Theory of the first-order Raman effect in crystals. Proceedings of the
Royal Society (London) A275. 218-232.
Meir, Y. and Wingreen , N.S. (1992). Landauer formula for the current through an interacting
electron region. Physical Review Letters 68, 2512-2515.
Meissner, W. and Ochsenfeld , R. (1933). Ein neuer effekt bei eintritt der supra leitfivhigkeit.
Naturwissenschaften 21 , 787-788.
Migdal , A.B. (1958). Interaction between electrons and lattice vibrations in a normal metal.
Soviet Physics JETP 34,996-1001.
Nambu, Y. (1960). Quasiparticles and gauge invariance in the theory of superconductivity.
Physical Review 117, 648-663.
Nyquist, H. (1928). Thermal agitation of electric charge in conductors. Physical Review
32, 110-113.
Omar, M.A. (1993). Elementary Solid State Physics: Prin ciples and Applications, revised
printing. Boston: Addison-Wesley.
Onnes, H.K. (1911). Further experiments with liquid helium. D. On the change of electrica l
resistance of pure metals at very low temperatures, etc. V. The disappearance of
the resistance of mercury. Koninklijke Nederlandsche Akademie van Wetenschappen
Proceedings 14,113-115.
Pals , P. and Mackinnon, A. (1996). Coherent tunneling through two quantum dots with
Coulomb interaction. Journal of Physics: Condensed Matter 8,5401-5414.
Scalapino, DJ. (1969). The electron-phonon interaction and strong-coupling superconduc-
tors. In Superconductivity, ed. R.D. Parks, pp. 449-560. New York: Marcel Dekker,
Inc.
Shung, K.W.- K. (1986). Dielectric function and plasmon structure of stage-l intercalated
graphite. Physica l Review B 34, 979-993.
Stefanucci, G. and Almbladh, c.-O. (2004). Time-dependent partition-free approach in
resonant tunneling systems. Physical Review B 69, 195318.
Swirkowicz, R., Barnas, 1., and Wilczynski, M. (2002). Electron tunneling in a double
ferromagnetic junction with a magnetic dot as a spacer. Journal of Physics: Condensed
Matter 14, 2011-2024.
Taylor, P.L. and Heinonen , O. (2002). A Quantum Approach to Condensed Matter Physics.
Cambridge: Cambridge University Press.
Thomas , L.H. (1927). The calculation of atomic fields. Proceedings of the Cambridge
Philosophical Society 23, 542-548.
Valatin, 1.G. (1958). Comments on the theory of superconductivity. Nuovo Cimento 7,
843-857.
Van Hove, L. (1954). Correlations in space and time and Born approximation scattering in
systems of interacting panicles. Physical Review 95, 249-262 .
Wick, G .c. (1950). The evaluation of the collision matrix. Physical Review 80, 268-
272.
Wilczek, F. (1982). Quantum mechanics offractional spin particles. Physical Review Letters
49,957-959.
References 393
Yildrim, T. (2008). Origin of the 150-K anomaly in LaFeAsO: competing anti ferromagnetic
interactions, frustration, and a phase transition. Physical Review Letters 101, 057010.
Ziman, J.M. (1960). Electrons and Phonons. Oxford: Oxford University Press.
Zimbovskaya, N.A. (2008). Electron transport tlu'ough a quantum dot in the Coulomb
blockade regime: nonequilibrium Green's function based model. Physical Review B
78,035331.
Index
394
Index 395
,- ,- ! - .. . . .. ,
IIIIIII~
9 78110
1111I111111
25172