Lecture Notes∗, Theory of Condensed Matter II

Jörg Schmalian

July 12, 2016

Institute for Theory of Condensed Matter (TKM), Karlsruhe Institute of

Technology

Summer Semester, 2016

Contents

I Introduction 2

II Observables and Green’s functions 3

1 Linear response 4

2 Properties of retarded and advanced Green’s functions 7

2.1 Homogeneity of time . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Lehmann representation . . . . . . . . . . . . . . . . . . . . . . . 10

3 Photoemission and single particle Green’s function 14

4 Green’s function for free particles 17

4.1 Perturbation theory and Dyson equation . . . . . . . . . . . . . . 19
4.2 Higher order correlation functions . . . . . . . . . . . . . . . . . . 19

5 Screening of the Coulomb interaction 20

5.1 Density response and dielectric function . . . . . . . . . . . . . . 22
5.2 Density response of non-interacting electrons . . . . . . . . . . . 25
5.3 Evaluation of the Lindhard function . . . . . . . . . . . . . . . . 26
5.4 Hartree-Fock analysis of the Coulomb interaction . . . . . . . . . 31
5.5 The random phase approximation for density correlations . . . . 35
∗ Copyright Jörg Schmalian

1
III Diagrammatic perturbation theory at finite T 39
6 The Matsubara function 41
6.1 Periodicity of the Matsubara function and Matsubara frequencies 41
6.2 Relation to the retarded function . . . . . . . . . . . . . . . . . . 43
6.3 Evolution with imaginary time . . . . . . . . . . . . . . . . . . . 44

7 Wick theorem 46
7.1 Time evolution of creation and annihilation operators . . . . . . 46
7.2 Wick theorem of time-independent operators . . . . . . . . . . . 47
7.3 Wick theorem for time dependent operators . . . . . . . . . . . . 49

8 Diagrammatic expansion of the partition function 51

8.1 Ring diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

9 Diagram rules for the single particle Green’s function 56

10 Fermi liquid theory 56

10.1 quasi-particle excitations . . . . . . . . . . . . . . . . . . . . . . . 56
10.2 Susceptibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

11 Conservation laws and Ward-Identities 61

12 Physics of Graphene 64

13 Superconductivity 64
13.1 off-diagonal long range order . . . . . . . . . . . . . . . . . . . . 65
13.2 the superconducting order parameter . . . . . . . . . . . . . . . . 72
13.3 The pairing instability . . . . . . . . . . . . . . . . . . . . . . . . 75
13.4 The BCS theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Part I
Introduction
This course is concerned with phenomena in quantum condensed matter systems
that can be most efficiently analyzed and solved using quantum field theoretical
methods. To this end we first physically motivate, introduce, and investigate
retarded Green’s functions. We will use the equation of motion method to
solve several problems such as itinerant ferromagnetism, superconductivity, and
dynamical screening of the Coulomb interaction. For a more systematic analysis
of many-body systems we will then introduce the Feynman-diagram technique
of thermal Green’s functions and, once again, investigate superconductivity, but
also disordered systems. Finally we will discuss the non-equilibrium version of
many-body theory by using the Schwinger-Keldysh approach. As example, we

2
investigate quantum transport of graphene. Thus, the course is concerned with
learning techniques and applying them to solve given many-body problems.
In case of the screening of the Coulomb interaction, we consider for example
the Hamiltonian of non-relativistic electrons (no spin-orbit interaction) in a
crystalline potential U (r) and with electron-electron interaction V (r − r0 ):

ˆ  2 2 
X ~ ∇
H = dd r ψα† (r) − − µ + U (r) ψα (r)
2m
ˆα
1X
= dd rdd r0 ψα† (r) ψβ† (r0 ) V (r − r0 ) ψβ (r0 ) ψα (r) . (1)
2
αβ

Here ψα (r) is the fermionic field operator that annihilates an electron with spin
α at position r, obeying standard fermionic anti-commutation relation

h i
ψα (r) , ψβ† (r0 ) = δαβ δ (r − r0 ) . (2)
+

If we include a similar Hamiltonian for the motion of the nuclei, along with
the electron-nucleus Coulomb interaction, we pretty much have a complete de-
scription of a solid within the non-relativistic limit. Thus, it is possible to fully
define the standard model of condensed matter physics in the introductory lines
of a lecture. One might then be tempted to conclude that this area of physics
must be conceptually pretty trivial. All that seems left to do is to solve for the
eigenstates and eigenvalues of H, a task that one leaves to a gifted programmer
or a clever mathematician. However, except for small systems or systems with a
large number of conserved quantities, these many-body systems simply cannot
be solved exactly. We need to find ways to analyze such an Hamiltonian, or a
simplified version of it, that allow to make as rigorous statements as possible. In
fact, the beauty of condensed matter theory is to make predictions about new
states of matter and universal behavior that is emergent, i.e. that is not obvi-
ous if one looks at the initial degrees of freedom of the Hamiltonian. If nothing
else, these considerations reveal that simply writing down a fundamental the-
ory, doesn’t yield a whole lot of insight that goes beyond the understanding of
what the elementary building blocks of this theory are. Emergent phenomena,
such as spontaneous symmetry breaking, composite particles, new topological
states of matter etc. etc. require a detailed analysis that is primarily guided
by experiment and, of course, by some good physical intuition. The author of
these lecture notes is rather convinced that this is the same, regardless whether
we talk about the physics of a piece of metal, a neutron star, or the universe as
a whole.

3
Part II
Observables and Green’s
functions
1 Linear response
We consider a system that is, at least initially, in thermodynamic equilibrium.
The expectation value of a physical observable is then given by

hAi = tr (ρA) , (3)

with density operator (often called density matrix)

1 −βH
ρ= e . (4)
Z
Z = tre−βH is the partition function and β = kB1T the inverse temperature.
In what follows we will use a system of units where kB = 1, i.e. we measure
temperatures in energy units. The generalization to the grand canonical ensem-
ble with chemical potential µ is straightforward. The density operator is then
given as ρeq = Z1g e−β(H−µN ) , where N is the particle number operator. As we
will mostly use the grand canonical ensemble, we will often call H − µN the
Hamiltonian and continue to use the letter H. Determining such an expectation
value is a formidable task in many body theory and we will do this during this
course.
A scenario that occurs very frequently and that offers significant insight into
the inner workings of a complex condensed matter systems is based on the mea-
surement of an observable that follows some external perturbation. Such an
approach yields dynamical information, in fact it even allows to theoretically
study the stability of a state of matter with regards to a spontaneous symme-
try breaking. To this end we consider a system coupled to an external field
that is characterized by the interaction part of the Hamiltonian W (t), i.e. the
Hamiltonian
Htot = H + W (t) (5)
consists of the Hamiltonian H that describes our system in isolation and the
external time dependent perturbation W (t).
A specific example for W (t) is the coupling
X
W (t) = −µB Si · B (t) (6)
i

of an external magnetic field to the electron spins

~X †
Si = ciα σ αβ ciβ (7)
2
αβ

4
of a magnetic system. Another example is interaction
X
W (t) = − Pi · E (t) (8)
i

between the electrical polarization

c†iα Ri ciα
X
Pi = e (9)
α

and an external electrical field.

As for the time dependence of W (t), we always have in mind a scenario
where the system is not affected by the perturbation in the infinite past, i.e.
W (t → −∞) → 0. A convenient way to realize this is via

E (t) = lim E0 exp (−i (ω + iδ) t)

δ→0+
B (t) = lim B0 exp (−i (ω + iδ) t) , (10)
δ→0+

i.e. we include an infinitesimal positive imaginary part to the frequency of an

oscillatory time dependence. In the case of a more general time dependence we
would write ˆ ∞
dω
W (t) = lim W (ω) e−i(ω+iδ)t . (11)
δ→0+ −∞ 2π

Next we consider the time evolution of the observable that follows as a

consequence of the applied external perturbation

hAit = tr (ρ (t) A), (12)

where the density matrix obeys the von Neuman equation

∂
i~ ρ (t) = [H + W (t) , ρ (t)] . (13)
∂t
Note, in case of ρ (t) and W (t) we are analyzing the time dependence of opera-
tors that are in the Schrödinger picture. As a reminder,
P the von Neuman equa-
tion follows for an arbitrary density matrix ρ (t) = i |Ψtot,i (t)i pi hΨtot,i (t)|
from the Schrödinger equation of the many body wave function |Ψtot,i (t)i with
Hamiltonian Htot . The dynamics of observables is then a consequence of the
time dependence of the density matrix. This is indicated by the subscript t of
hAit .
As discussed, the perturbation is absent in the infinite past and we assume
the system was in equilibrium for t → −∞:
1 −βH
ρ (t → −∞) = ρ = e . (14)
Z
In most cases the external perturbation is small and we can confine ourselves
to changes in hAit that are linear in W (t). This regime is referred to as linear

5
response. The subsequent formalism can be (and has been) extended to include
higher order non-linearities. Here we will, however, only consider the leading
order, linear effects.
To proceed we go to the interaction representation

ρ (t) = e−iHt/~ ρ(I) (t) eiHt/~ . (15)

Note, ρ(I) (t) corresponds to the interaction picture of the Hamiltonian Htot .
The Hamiltonian of our system of interest is of course H (W (t) is only used to
probe this system). If considered with regards to H, ρ(I) (t) corresponds to the
Heisenberg picture. This is the reason why we will below state that operators
are taken in the Heisenberg picture.
Performing the time derivative gives

∂ρ (t) ∂ρ(I) (t) iHt/~

i~ = [H, ρ (t)] + e−iHt/~ i~ e . (16)
∂t ∂t
Inserting the von Neuman equation yields

∂ρ(I) (t) h (I) i

i~ = W (t) , ρ(I) (t) , (17)
∂t
which is formally solved by (better, its solution is equivalent to the solution of)
ˆ t
i h i
ρ (I)
(t) = ρ − dt0 W (I) (t0 ) , ρ(I) (t0 ) . (18)
~ −∞

If we return to the Schrödinger picture, it follows

ˆ t
i 0 0
ρ (t) = ρ − dt0 e−iH (t−t )/~ [W (t0 ) , ρ (t0 )] eiH (t−t )/~ . (19)
~ −∞

One can now generate a systematic expansion with regards to W (t) if one
solves this integral equation via recursion. At zeroth order holds of course
−βH
ρ (t) = ρ = e Z . At first order we can insert this zeroth order solution in the
right hand side and obtain
ˆ t
i 0 0
ρ (t) = ρ − dt0 e−iH (t−t )/~ [W (t) , ρ] eiH (t−t )/~ . (20)
~ −∞

We can now determine the expectation value of A:

ˆ t
i h i 
hAit = hAi − dt0 tr W(I) (t0 ) , ρ A(I) (t) . (21)
~ −∞

One can cyclically change the order under the trace operation:

tr [(Wρ − ρW) A] = tr [(AW − WA) ρ], (22)

6
which gives
ˆ t
i Dh iE
hAit = hAi − dt0 A(I) (t) , W (I) (t0 ) . (23)
~ −∞

It is useful to introduce (the retarded Green’s function)

DD EE i Dh iE
A(I) (t) ; B (I) (t0 ) = − θ (t − t0 ) A(I) (t) , B (I) (t0 ) (24)
~
such that ˆ ∞ DD EE
hAit = hAi + dt0 A(I) (t) ; W (I) (t0 ) . (25)
−∞

These considerations demonstrate that the linear response of a physical system

is characterized by retarded Green’s functions. The interesting result is that
we can characterize the deviation from equilibrium (e.g. dissipation in case of
the electrical conductivity) in terms of fluctuations of the equilibrium (equilib-
rium correlation functions). Among others, this will lead us to the fluctuation-
dissipation theorem. It will also offer a compact and unifying approach to study
the response of a system with regards to an arbitrary external perturbation.
Example, conductivity: as discussed, we have an interaction between the
electrical field and the electrical polarization:
X (I)
W (I) (t) = − Pi · E (t) . (26)
i

with (we will frequently not write down explicitly the limit δ → 0+ )

E (t) = E0 exp −i ω + i0+ t .

(27)

If we are interested in the electrical current it follows

Xˆ ∞ DD
(I)
EE 0
hjα it = − dt0 jα(I) (t) ; Pβ,i (t0 ) E0,β e−i(ω+iδ)t (28)
i −∞

Before we give further examples and discuss the physical implications of our lin-
ear response analysis, we will therefore discuss in some detail the mathematical
properties of such functions.

2 Properties of retarded and advanced Green’s

functions
We learned that the linear response of a physical system that is initially in
equilibrium can be formulated in terms of retarded Green’s functions:
r
GrA,B (t, t0 ) = hhA (t) ; B (t0 )ii
D E
≡ −iθ (t − t0 ) [A (t) , B (t0 )]η . (29)

7
To simplify our notation we will from now on use a convention where ~ = 1, i.e.
frequencies and energies are measured in the same units. We further dropped
the superscript (I) for the interaction representation. Keep in mind, that it
is anyway the Heisenberg picture if we refer this to the Hamiltonian H of the
system we are interested in:

A (t) = eiHt Ae−iHt . (30)

Finally, we introduced
[A, B]η = AB + ηBA (31)

to simultaneously analyze the commutator for η = −1 and the anti-commutator

for η = +1. We will see very soon that this generalization to anti-commutators
is sometimes a very sensible thing to do if one considers certain properties of
fermions.
The prefactor θ (t − t0 ) emerged as a natural consequence of causality. The
response of the quantity hAit was only influenced by W (t0 ) with t0 < t. It
is however possible, at least formally, to introduce other Green’s functions.
Important examples are advanced Green’s functions:
r
GaA,B (t, t0 ) = hhA (t) ; B (t0 )ii
D E
≡ iθ (t0 − t) [A (t) , B (t0 )]η (32)

or time-ordered (sometimes called causal) Green’s functions -

c
GcA,B (t, t0 ) = hhA (t) ; B (t0 )ii
≡ −i hTη A (t) B (t0 )i (33)

with time ordering operator

Tη A (t) B (t0 ) = θ (t − t0 ) A (t) B (t0 ) − ηθ (t0 − t) B (t0 ) A (t) . (34)

Because of our insight that retarded Green’s functions determine the linear
response, we predominantly investigate this function. The advanced and time-
ordered functions can be easily analyzed along the same lines. In fact all func-
tions contain essentially the same information.

2.1 Homogeneity of time

An important property of all of those Green’s functions is that they are only
functions of the difference t − t0 . It holds
D E
GrA,B (t, t0 ) = −iθ (t − t0 ) [A (t) , B (t0 )]η
= −iθ (t − t0 ) (hA (t) B (t0 )i + η hB (t0 ) A (t)i) (35)

8
The correlation functions are explicitly given as
1  −βH iHt −iHt iHt0 −iHt0 
hA (t) B (t0 )i = tr e e Ae e Be
Z
1  0 0

= tr e−βH eiH (t−t ) Ae−iH (t−t ) B
Z
= hA (t − t0 ) B (0)i (36)
0 0
and similar for hB (t ) A (t)i = hB (0) A (t − t )i. Thus, it follows

GrA,B (t, t0 ) = GrA,B (t − t0 ) . (37)

The reason why we could demonstrate this behavior is that the thermal aver-
age, with Boltzmann weight e−βH and the unitary time evolution, with e−iHt
commute. They are both governed by the same Hamiltonian. Physically it cor-
responds to the fact that there is no preferred absolute time in a system that
is in equilibrium. An implication is that any stationary distribution function,
even those that are not in equilibrium but that yield states without preferred
time point must have a density matrix ρ = ρ (H, Xi ) that only depends on the
Hamiltonian and maybe on other conserved quantities Xi of the system with
[H, Xi ] = 0.

2.2 Equation of motion

The fundamental equation of motion of quantum mechanics is the Schrödinger
equation. For operators that are not explicitly time dependent in the Schrödinger
picture, the Schrödinger equation is equivalent to the Heisenberg equation1 :
i∂t A (t) = [A (t) , H]− . (38)
This enables us to determine the equation of motion that follows from the
Schrödinger equation.
We start from
n D Eo
i∂t GrA,B (t) = ∂t θ (t) [A (t) , B (0)]η
D E D E
= δ (t) [A, B]η + θ (t) [∂t A (t) , B (0)]η
D E D  E
= δ (t) [A, B]η − iθ (t) [A (t) , H]− , B (0) η , (39)

where t now refers to the relative time. The last expression can itself be written
as a retarded Green’s function
r D  E
Gr[A,H]− ,B (t) = [A (t) , H]− ; B (t0 )

= −iθ (t) [A (t) , H]− , B (0) η

and we obtain the equation of motion for retarded Green’s functions.

D E
i∂t GrA,B (t) = δ (t) [A, B]η + Gr[A,H]− ,B (t) . (40)

1 Recall, that we use a system of units with ~ = 1.

9
Thus, in order to determine one Green’s function one needs to know another one.
We will see that in case of non-interacting systems the newly generated Green’s
functions form a closed set, which allows, at least in principle, for a full solution.
On the other hand, for a generic interacting many body system a closed solution
only exists when one analyzes conserved quantities with [A, H]− = 0 or at least
densities of conserved quantities. These aspects will all be discussed in greater
detail below.
Because of Eq.(37) follows that we can Fourier transform the Green’s func-
tion ˆ ∞
r
GAB (ω) = dtGrAB (t) eiωt . (41)
−∞
The equation of motion for the Fourier transforms are then easily obtained as
D E
ωGrA,B (ω) = [A, B]η + Gr[A,H]− ,B (ω) . (42)

It is now only an algebraic equation.

If one repeats the same analysis for the advanced and time-ordered Green’s
functions, one finds identical expressions as in Eqs.(40) and (42). On the other
hand, the detailed time dependence of Gr (t), Ga (t), and Gc (t) is obviously
very different. From the definition of these quantities follows for example that
Gr (t < 0) = 0, while Ga (t > 0) = 0. Thus, if one wants to determine the
correct solution of the equation of motion one must incorporate those boundary
conditions appropriately. This also implies that the Fourier transform in Eq.(41)
has to be performed with some care. To address these issues we will next analyze
the analytic properties of Green’s functions in some detail.

2.3 Lehmann representation

In what follows we determine a rigorous representation of GrAB (ω) that reveals
a lot about the analytic structure of Green’s functions. Let {|li} be the exact
eigenfunctions of the Hamiltonian with eigenvalues {El }, i.e.
H |li = El |li . (43)
Then, we can write a thermal expectation value as
1 X −βEl
hAi = tr (ρA) = e hl |A| li . (44)
Z
l

For a correlation function follows accordingly

1 X −βEl
hA (t) B (0)i = e hl |A (t) B (0)| li
Z
l
1 X −βEl
 iHt −iHt  
= e l e Ae B l
Z
l
1 X −βEl it(El −Em )
= e e hl |A| mi hm |B| li (45)
Z
l.m

10
The same analysis can be performed for hB (0) A (t)i and yields

1 X −βEl −it(El −Em )

hB (0) A (t)i = e e hl |B| mi hm |A| li
Z
l.m
1 X −βEm it(El −Em )
= e e hl |A| mi hm |B| li (46)
Z
l.m

In order to analyze the frequency dependence of the Fourier transform of

the Green’s function we first consider the Fourier transform of the correlation
functions ˆ ∞
dω
hB (0) A (t)i = J (ω) e−iωt . (47)
−∞ 2π

For the inverse transform J (ω), which we also call the spectral function, follows
ˆ ∞
J (ω) = dteiωt hB (0) A (t)i
−∞
ˆ ∞
1 X −βEm
= e hl |A| mi hm |B| li dteit(ω+El −Em ) . (48)
Z −∞
l.m
´∞
We use −∞
dteitω = 2πδ (ω) and obtain:

2π X −βEm
J (ω) = e hl |A| mi hm |B| li δ (ω + El − Em ) . (49)
Z
l.m

At T = 0 this expression simplifies further. Let us consider a singly degenerate

ground state with energy E0 . Then follows ZT →0 = e−βE0 . Similarly, in the
sum over m only the ground state(s) contribute and we obtain
X
JT =0 (ω) = 2π h0 |B| li hl |A| 0i δ (ω + El − E0 ) . (50)
l

2
Notice that in case where B = A† follows hl |A| mi hm |B| li = |hl |A| mi| ≥
0. Thus, the spectral function is real with J (ω) ≥ 0. With our above results
for the two correlation functions follows immediately
ˆ ∞
dω βω
hA (t) B (0)i = e J (ω) e−iωt . (51)
−∞ 2π

We use these results to write for our Green’s function

ˆ ∞
GrAB (ω) = −i dteiωt θ (t) (hA (t) Bi + η hBA (t)i)
−∞
ˆ ∞ ˆ ∞
dω 0  βω0  0
= −i e + η J (ω 0 ) dtei(ω−ω )t θ (t) (52)
−∞ 2π −∞

11
To proceed we need to analyze the integral
ˆ ∞
f (ω) = dteiωt θ (t)
−∞
ˆ ∞
= dteiωt
0
ˆ ∞
= lim+ dtei(ω+iδ)t
δ→0 0
i
= . (53)
ω + i0+
To insert the converging factor seems a bit arbitrary. To check that this is
indeed the right thing to do, let us perform the inverse transform
ˆ ∞
dω ie−iωt
F (t) = . (54)
−∞ 2π ω + i0+

We want to evaluate this integral using the residue theorem. For t > 0 we can
close the contour in the lower half plane, i.e. the contour encircles the pole at
ω = −i0+ . The residue of the pole is 1 (because of the sense of orientation of
the contour). For t < 0 we have to close the contour in the upper half plane.
As there is no pole in this half plane, the integral vanishes. Thus, we obtain
F (t) = θ (t) as expected. This analysis also reveals that causality, expressed
in terms of the θ-function, implies that we should consider frequencies ω + i0+
with a small positive imaginary part.
It follows for the Green’s function

ˆ ∞
 0 
dω 0 e
βω
+ η J (ω 0 )
GrAB (ω) = . (55)
−∞ 2π ω − ω 0 + i0+

Inserting the spectral function yields the so called Lehmann representation:

1 X e−βEl + ηe−βEm hl |A| mi hm |B| li

r
GAB (ω) = (56)
Z ω + El − Em + i0+
l.m

which reveals that a retarded Green’s function, once considered with complex
frequency argument ω, is analytic everywhere, except infinitesimally below the
real axis. In fact one can consider the function

1 X e−βEl + ηe−βEm hl |A| mi hm |B| li

GAB (z) = , (57)
Z z + El − Em
l.m

with complex argument z and the retarded function is given by

GrAB (ω) = GAB ω + i0+ .

(58)

12
Repeating our analysis for the advanced Green’s function yields

GaAB (ω) = GAB ω − i0+ .

(59)

If one keeps in mind that under the integral holds

1 1
+
= P − iπδ (ω) , (60)
ω + i0 ω
1
where the principle value of is meant in the first term, we obtain:
ω
ˆ ∞
dω 0  βω0 
GrAB (ω) − GaAB (ω) = e + η J (ω 0 )
−∞ 2π
 
1 1
× −
ω − ω 0 + i0+ ω − ω 0 − i0+
βω


= −i e + η J (ω) (61)

For B = A† , where the product of the two matrix elements is real, the ad-
vanced function is the complex conjugate of the retarded function. Considering
once again the frequent situation where B = A† it holds

J (ω) = −2nη (ω) ImGrAA† (ω) , (62)

where
1
nη (ω) = (63)
eβω + η
is, depending on whether we use the commutator or anti-commutator, the Bose
or Fermi function, respectively.
In case of B = A† we also obtain the famous Kramers-Kronig relation
ˆ ∞
dω 0 ImGrAA† (ω 0 )
GrAA† (ω) = − , (64)
−∞ π ω − ω 0 + i0+

which reveals that the information about the Green’s function is fully contained
in its imaginary part, a result that is a consequence of the constraints brought
about by causality. This result also allows for the analysis the function G (z)
introduced above and yields
ˆ ∞
dω 0 ImGrAA† (ω 0 )
GAA† (z) = − . (65)
−∞ π z − ω0

Finally, we can use our results to determine expectation values of correlation

functions via
ˆ ∞
dω
hBA (t)i = J (ω) e−iωt
−∞ 2π
ˆ ∞
dω Gr (ω) − GaAB (ω) −iωt
= − nη (ω) AB e . (66)
−∞ π 2i

13
In particular, we can use this expression to determine static expectation values
(e.g. in case of B = A† )
ˆ ∞

†  dω
A A =− nη (ω) ImGrAA† (ω) .
−∞ π
The previous results finally allow for a proper interpretation of the equation
of motion in frequency representation. We obtain a Green’s function with proper
boundary conditions if we simply analyze
D E
ω + i0+ GrA,B (ω) = [A, B]η + Gr[A,H]− ,B (ω) .


(67)

This immediately guarantees that the back-transform GrAB (t) obeys the correct
boundary condition and vanishes for t < 0.
As will be discussed in greater detail, one can also show easily that Green’s
functions obey certain sum rules, the easiest of which is
ˆ ∞ D E
dωGrA,B (ω) = −iπ [A, B]η . (68)
−∞

3 Photoemission and single particle Green’s func-

tion
Photoemission is a widely used experimental approach to study the electronic
properties of solids. It is based on the photoelectric effect that was initially
discussed by Einstein. The irradiation of a solid with light gives rise to the
emission of electrons. In what follows we discuss this effect within a many-body
theory.
 NLet
 the many-body wave function prior to the irradiation be the initial state
Ψm = |mi where we explicitly denote that we are considering a system with N
 E
particles. Let the final state be given as ΨN
f . The corresponding energies are

N
Em and EfN . The transition probability per unit time between the two states
is then given by Fermi’s golden rule
N 2
w = 2π  ΨN δ ω − EfN + EmN



f |V | Ψm . (69)
The perturbation caused by the irradiation is of the form
V = −P · E0 . (70)
Since the polarization is a single particle operator, i.e. an operator that we can
write in the form: X α,α0 †
V = dk,k0 ψkα ψk0 α0 , (71)
k,k0 αα0
†
where ψkα is the creation operator of an electron with momentum k and spin
α,α0
α and dk,k0 = − hkα |P| k0 α0 i · E0 refers to the dipole matrix element.

14
 The key assumption of the usual description of photoemission is the so called
sudden approximation, where we assume that the excited photoelectron does not
couple to the remaining N − 1 electron system, i.e. it is excited highly above
the Fermi energy of the solid and rapidly leaves the system. This is at least
consistent with the usual view that photoelectrons originate only from a few
top-most layers of the solid near the surface. Thus, we write
Ψf = ψ † ΨN −1
 N  
kf β l (72)

is the photoelectron added to one of theeigenstates of the N −1-particle system.

At the same time we assume ψkf β ΨN m = 0, i.e. the photoelectron state is not
mixed into any of the relevant initial states of the system. The emphasis in
the last term is on “relevant”. At T = 0, the only relevant initial state is the
ground state, and for finite temperatures we are only interested in states with
excitation energy Em − E0 ≈ kB T .
It follows
*




 +2
X α,α0 †
−1 
w = 2π  ΨN dk,k0 ψkα ψk0 α0  ΨN N N
   

ψkf β m  δ ω − Ef + Em (73)

l
  k,k0 αα0  
 
Since ψkf β ΨN
m = 0, it must hold that α = β and k = kf , i.e.
It follows
*


X

 +2
w = 2π  ΨN −1 
dβ,α  N  N N
   

l  kf ,k ψkα  Ψm  δ ω − Ef + Em (74)
  k,α  
 
We now sum over all initial states ΨN m = |mi with initial probability pm =
and over all final states ΨlN −1 = |li , and take into account that
1 −βEm

Ze
the final energy EfN = kf + ElN −1 is the sum photoelectron energy kf and of
the energy ElN −1 of the remaining N − 1 many body state. It follows for the
intensity
*   + 2
   
2π X −βEm  X β,α  

Ikf β (ω) = e l dkf ,k ψkα  m  δ ω − kf − El + Em (75)

Z 
lm   k,α  

We recognize this result as the spectral function of a retarded Green’s function

with
X β,α
A = dkf ,k ψkα
k,α

B = A† (76)

If we recall our earlier result that J (ω) = −2nη (ω) ImGrAA† (ω) it seems most
natural to use for the photoelectron spectrum of occupied states a quantity that

15

−1
is proportional to the Fermi function n+ (ω) = f (ω) = eβω + 1 . Thus we
opt for the anticommutator Green’s function with η = +1 and define
h i 
Grk,k0 αα0 (ω) = −iθ (t − t0 ) (t) ψkα (t) , ψk† 0 α0 (0) , (77)
+

such that

X
α0 ,β∗
Ikf β (ω) = −2f ω − kf dβ,α r
kf ,k ImGk,k0 αα0 ω − kf dk0 ,kf . (78)
kk0 ,αα0

Thus, except for the dipole matrix elements, the photoemission intensity is
determined by the imaginary part of the retarded fermion Green’s function.
Let us consider a system of non-interacting fermions with Hamiltonian
†
X
H= (εk − µ) ψkα ψkα . (79)
kα

µ is the chemical potential. In order to determine the equation of motion, for

Grk,k0 αα0 (ω) we need to evaluate the commutator
[ψkα , H]− = (εk − µ) ψkα (80)
that is particularly easy for non-interacting particles. It follows
h i 
+

r †
ω + i0 Gk,k0 αα0 (ω) = ψkα , ψk0 α0 + (εk − µ) Grk,k0 αα0 (ω) (81)
+
h i
Using the usual anti-commutation properties ψkα , ψk† 0 α0 = δαα0 δkk0 it follows
+

Grk,k0 αα0 (ω) = δαα0 δkk0 Grk (ω) , (82)

with
1
Grk (ω) = . (83)
ω + i0+ − εk + µ
We observe that without the infinitesimal part in the frequency, there would
be a pole of the Green’s function at the particle energy εk − µ relative to the
chemical potential. We also easily obtain the imaginary part
1
− ImGrk (ω) = δ (ω − εk + µ) . (84)
π
A sharp peak in the imaginary part is a signature that the system is character-
ized by a particle, a behavior that will be used later on as well, when we analyze
interacting electrons. We could for example use this result to obtain the particle
number
ˆ ∞
D
†
E dω
ψkα ψkα = − n+ (ω) ImGrk (ω) .
−∞ π
ˆ ∞
= dωf (ω) δ (ω − εk + µ)
−∞
= f (εk − µ) . (85)

16
Thus, as expected we finD that the occupation number of free fermions is given
by the Fermi function. It turns out that knowledge of the retarded Green’s
function is sufficient to determine all thermodynamic properties of a many body
system of electrons. We will prove this result below for an interacting electron
system.
For the photoemission spectrum follows finally:

2
X  β,α 2

Ikf β (ω) = f ω − kf dkf ,k  δ ω − kf − εkα + µ . (86)
π
k,α

The experiments then probes the occupied states of a solid and can be used
to determine the energy-momentum relation. Often, one assumes momentum
conservation, at least for the components of the momentum parallel to the
surface and finds



Ikf β (ω) ∝ f ω − kf δ ω − kf − εkf + µ .

4 Green’s function for free particles

In case of non-interacting fermions and bosons, one can obtain a closed expres-
sion for the Green’s functions. To this end we consider a Hamiltonian of the
form
hij c†i cj ,
X
H= (87)
ij

where c†i and cj are creation and annihilation operators of fermions or bosons in
states with single particle quantum numbers i and j, respectively. Those quan-
tum numbers could be momentum, lattice sites in a solid, spin, or a combination
of spin and momentum, depending on the problem at hand. The fact that we
confine ourselves to bilinear forms (only two operators) reflects that we consider
noninteracting particles. We do, however, not assume that hij is a diagonal
matrix, whose diagonal elements are then the single particle eigenstates. In
case of bosons (fermions) we use the well known commutator (anticommutator)
relations
h i
ci , c†j = δij ,
η
h i
[ci , cj ]η = c†i , c†j = 0, (88)
η

with η = −1 (η = −1).
We first determine the so called single particle Green’s functions2
h i 
Grij (t) = −iθ (t) ci (t) , c†j . (89)
η

2 To simplify our notation we use Grij (t) instead of Gr (t).

†
ci cj

17
For the analysis of the equation of motion we have to analyze the commutator
h i
hlm ci , c†l cm
X
[ci , H]− = (90)
−
lm

It holds
h i
ci , c†l cm = ci c†l cm − c†l cm ci
−

= −ηc†l ci cm + δil cm − c†l cm ci

= η 2 c†l cm ci + δil cm − c†l cm ci
= δil cm , (91)

which yields X
[ci , H]− = him cm , (92)
m
regardless of whether we consider bosons or fermions.
For our equation of motion follows then
X
ω + i0+ Grij (ω) = δij + him Grmj (ω) .


(93)
m

We see that the equation of motion closes in the sense that only Green’s functions
of the type defined in Eq.(89) are needed. It is also natural to introduce a matrix
Ĝ (ω) with matrix elements Gij (ω) and similarly ĥ for the matrix representation
of the Hamiltonian with elements hij . Then follows3

ω Ĝ (ω) = 1̂ + ĥ · Ĝ (ω) , (94)

or  
ω − ĥ Ĝ (ω) = 1̂. (95)
This leads to
 −1
Ĝ (ω) = ω − ĥ . (96)

Thus, in order to determine the Green’s function of a non-interacting gas of

fermions or bosons, it is sufficient to diagonalize a matrix in the space of single-
particle quantum numbers. This can be a non-trivial task on its own, e.g.
for disordered systems where hij are realizations subject to a certain disorder
distribution function. In systems with translation invariance the single-particle
eigenstates of the Hamiltonian are plane-waves with eigenvalues εk that depend
k2
on the specific dispersion relation of the problem (e.g. εk = 2m − µ for solutions
of the Schrödinger equation). This immediately determines the eigenvalues of
the Green’s function
1
Grk (ω) = , (97)
ω + i0+ − εk
3 We drop the index r for the retarded function with the understanding that it follows via

ω → ω + i0+ .

18
a result that we obtained earlier already for free fermions.
In a solid, with discrete translation invariance, the eigenstates are the bands
εk,n where the momenta are from the first Brillouin zone and we find accordingly
Grk,n (ω) = ω+i0+1−εk,n .

4.1 Perturbation theory and Dyson equation

An important application of our matrix formalism can be made for systems
where we can write
hij = ε0i δij + Vij , (98)
i.e. we are in the eigenbasis of a bare Hamiltonian ĥ0 with eigenvalues ε0i , while
an additional perturbation is off-diagonal.
This suggests to write

Ĝ−1 = ω − ĥ0 − V̂
−1
= Ĝ0 − V̂ , (99)
−1
where Ĝ0 = ω − ĥ0 is the Green’s function of the bare Hamiltonian, i.e. the
bare Green’s function. It is a fully diagonal matrix, i.e. we have

δij
G0,ij (ω) = . (100)
ω − ε0i

Eq.(99) is called the Dyson equation for single particle systems, i.e. for systems
without interactions. We can multiply Eq.(99) from the left with Ĝ0 and from
the right with Ĝ and obtain

Ĝ = Ĝ0 + Ĝ0 V̂ Ĝ. (101)

A perturbation theory in V̂ can now be generated by iterating this equation

repeatedly
Ĝ = Ĝ0 + Ĝ0 V̂ Ĝ0 + Ĝ0 V̂ Ĝ0 V̂ Ĝ0 · · · . (102)

4.2 Higher order correlation functions

The knowledge of Grij (ω) yields immediate information about expectation val-
D E
ues of the form c†i cj . Suppose we want to know something about a more
D E
complicated expectation value, such as c†j c†k cl ci , we can equally find closed
expressions for the corresponding Green’s functions. To this end we analyze
GAB = Gi,jkl with A = ci and B = c†j c†k cl .
The equation of motion follows immediately, as the commutator with the
Hamiltonian is the same
h i  X
† †
ωGi,jkl (ω) = ci , cj ck cl + him Gm,jkl (ω) . (103)
η
m

19
The remaining commutator or anticommutator is easily calculated as:
h i
ci , c†j c†k cl = ci c†j c†k cl + ηc†j c†k cl ci
η

= −ηc†j ci c†k cl + δij c†k cl + ηc†j c†k cl ci

= c†j c†k ci cl − ηδik c†j cl + δij c†k cl + ηc†j c†k cl ci
= δij c†k cl − ηδik c†j cl . (104)

This yields for the equation of motion the result:

D E D E X
ωGi,jkl (ω) = δij c†k cl − ηδik c†j cl + him Gm,jkl (ω) . (105)
m

If we use our earlier result for the single particle Green’s function we can write
this as X  D E D E
−1
Ĝ (ω) Gm,jkl (ω) = δij c†k cl − ηδik c†j cl , (106)
im
m

which can be multiplied by Gsi (ω) and summed over i. It follows

D E D E
Gi,jkl (ω) = Gij (ω) c†k cl − ηGik (ω) c†j cl (107)
D E
These functions can now be used to determine the expectation values c†j c†k cl ci
and it follows
D E D ED E D ED E
c†j c†k cl ci = c†j ci c†k cl − η c†j cl c†k ci . (108)

Thus, we are able to express a more complicated expectation value in terms of

simpler ones, a procedure that is correct for arbitrarily complex operators. In
fact the last result is the simplest case of a more general statement that goes
under the name of Wick theorem.

5 Screening of the Coulomb interaction

In what follows we want to investigate a first non-trivial problem in many-body
theory, the screening of the long-range electron-electron Coulomb interaction.
Before we go into the details we summarize our conventions for the Fourier
transformation between real space and momentum space. We consider always a
large but finite volume V together with some sort of boundary conditions that
imply discrete momentum values k, where two neighboring points are separated
by ∆k = V2π 1/d . For an arbitrary function f (r) we then use the convention

ˆ
fk = dd rf (r) e−ik·r (109)

20
with back transform
1 X ik·r
f (r) = e fk . (110)
V
k
´
Here we used that dd reik·r = V δk,0 and V1 k e−ik·r = δ (r). Note, that
P
those are not √
the only conventions used in the literature. Frequently one finds
a prefactor 1/ V in front of both, the sum and the integral in the above defini-
tions. The justification of our
Pchoice is´that it allows without problems to take
the limit V → ∞, where V1 k · · · → dd k · · · and V δk,0 → δ (k). With this
convention follows for example for fermionic operators:
ˆ
ψkα = dd rψα (r) e−ik·r (111)

that [ψα (r) , ψβ (r0 )]+ = δαβ δ (r − r0 ) implies

h i
ψkα , ψk† 0 β = δαβ V δk,k0 . (112)
+

For the Hamiltonian of a free electron system

ˆ X
H = dd r ψα† (r) ε (−i∇) ψα (r) (113)
α

follows accordingly
1 X †
H0 = ε (k) ψkα ψkα . (114)
V
kα

Another relevant quantity to be Fourier transformed is the density

X
ρ (r) = ψα† (r) ψa (r) . (115)
α

It follows
Xˆ
ρq = dd rψα† (r) ψa (r) e−iq·r
α
ˆ
1 X 0
= dd re−ir·(q+k−k ) ψkα
†
ψk0 α
V2
αkk0
1 X †
= ψkα ψk+qα . (116)
V
αk

Another famous example for a Fourier transformation is for the Coulomb inter-
action ˆ
1 4π
d3 re−iq·r = 2. (117)
|r| q

21
5.1 Density response and dielectric function
Let us consider a system of electrons exposed to the external electric potential
as perturbation ˆ
W (t) = d3 rρel (r) ϕext (r, t) , (118)

where the external time and space dependent potential ϕext (r, t) couples to the
electron charge density
X
ρel (r) = −e ψα† (r) ψa (r) . (119)
α

Here we used that electrons are negatively charged, i.e. we work with e > 0.
Below it will be more convenient to work with the particle density introduced
in Eq.(115), i.e. ρel (r) = −eρ (r).
If we are interested in the induced electron charges we can use our linear
response formalism:
ˆ ∞
dt0 ρel (r, t) ; W (t0 ) .

el 
el 


ρ (r) t = ρ (r) +
−∞
ˆ ∞
dt0 d3 r0 hhρ (r, t) ; ρ (r0 , t0 )ii ϕext (r0 , t0 )(120)

el  2
= ρ (r) + e
−∞

Thus, the dynamics of induced charges is determined by the retarded density-

density Green’s function

χ (r, r0 , t) = − hhρ (r, t) ; ρ (r0 , 0)ii

iθ (t) [ρ (r, t) , ρ (r0 , 0)]− .


= (121)

If we assume that our system is translation invariant, χ (r,

r0 , t) becomes
a function of r − r0 only. In addition the equilibrium density ρel (r) = ρel

0
becomes independent on r. It follows
ˆ ∞
dt0 d3 r0 χ (r − r0 , t − t0 ) ϕext (r0 , t0 ) .

el 
ρ (r) t = ρel
0 − e 2
(122)
−∞

A solid is electrically neutral. The electron charges are then compensated by

the positive charges of the ions. A popular model to describe the ionic charges
is the so-called jellium model. Here one assumes a uniform positive background
charge ρion of the ions that is, on the time scale of the electrons, fixed. Charge
neutrality implies that ρion + ρel
0 = 0. The total induced charge

ρind (r, t) = ρel (r) t + ρion


(123)

in our solid is then related to the external potential via

ˆ ∞
ρind (r, t) = −e2 dt0 d3 r0 χ (r − r0 , t − t0 ) ϕext (r0 , t0 ) . (124)
−∞

22
As required by charge neutrality, the total induced charge vanishes without the
external potential.
The above relation between ρind (r, t) and ϕext (r0 , t0 ) is a convolution with
respect to the temporal and spatial arguments. This can be simplified by going
to momentum and frequency space according to:
ˆ ˆ
F (q, ω) = d r dte−i(q·r−ωt) F (r, t) .
3
(125)

It follows:
ρind (q, ω) = −e2 χ (q, ω) ϕext (q, ω) . (126)
The perturbation W (t) can alternatively be written as
ˆ
ρel (r) ρext (r0 , t)
W (t) = d3 rd3 r0 , (127)
|r − r0 |
where the external charges are simply the sources of the external potential
ˆ
ρext (r0 , t)
ϕext (r, t) = d3 r0 , (128)
|r − r0 |
i.e. it is the solution of the Poisson equation with external charges as sources:

∇2 ϕext (r, t) = −4πρext (r, t) . (129)

This formulation allows to make contact with the usual formulation of the elec-
trodynamics of continua where the sources of the electric field are all charges,
external and internal ones

∇ · E = 4π ρind + ρext ,


(130)

while the displacement field is introduced as the field that has only the external
charges as sources
∇ · D = 4πρext . (131)
The linearized relation between the two fields is
ˆ ∞
D (r, t) = dt0 d3 r0 ε (r − r0 , t − t0 ) E (r0 , t0 ) . (132)
−∞

with dielectric function ε. Once again we assumed translation invariance and

the homogeneity of time. We then obtain after Fourier transformation

D(q, ω) = ε (q, ω) E(q, ω). (133)

The above equations for the electric field and the displacement field become in
Fourier space

iq · E(q, ω) = 4π ρind (q, ω) + ρext (q, ω)

iq · D(q, ω) = 4πρext (q, ω). (134)

23
Expressing D(q, ω) in the second equation by ε (q,ω) E(q, ω) allows us to write
 
1
ρind (q, ω) = − 1 ρext (q, ω). (135)
ε (q, ω)
Thus, the dielectric function relates the external sources to the induced charges.
On the other hand, Eq.(126) established a relation between the induced charges
and the external potential. As discussed, the external charge density and the
external potential are related by the Poisson equation, which becomes in Fourier
space
q 2 ϕext (q, ω) = 4πρext (q, ω). (136)
With this relation we obtain
 
1 4π ext
ρind (q, ω) = −1 ϕ (q, ω). (137)
ε (q, ω) q2
Comparing this with Eq.(126) yields

1 4πe2
= 1 − 2 χ (q, ω) . (138)
ε (q, ω) q

Thus, the retarded density-density Green’s function χ (q, ω) determines the di-
electric function. An interesting implication of this result is that the dielectric
function diverges for for q → 0 if χ (q, ω) does not vanish at least as q 2 . We
will see by explicitly analyzing the density-density response function that this
is indeed the case and leads to a qualitative change in the space dependence of
the effective Coulomb interaction, i.e. the screening of the Coulomb interaction.
More specifically, Eq.(138) allows to study the effective potential
1
ϕ (q, ω) = ϕext (q, ω) (139)
ε (q, ω)
that has as sources the total charge, i.e.
∇2 ϕ (r, t) = −4π ρind (r, t) + ρext (r, t) .

If we consider for example a single point-charge as source, we have ρext (r, t) =

eδ (r), which yields ρext (q, ω) = 2πeδ (ω) and finally ϕext (q, ω) = 2πeδ (ω) 4π
q2 .
This yields after back Fourier transformation with regards to frequency a time-
independent potential:
4πe
ϕ (q, t) = . (140)
ε (q, ω = 0) q 2
Another interesting scenario occurs when
.
ε (q, ω) = 0. (141)
If there are momenta and frequencies where this obeyed, an infinitesimal exter-
nal charge or potential will induce a large charge response. Below we will see
that this gives rise to plasma oscillations of the electron system.

24
5.2 Density response of non-interacting electrons
A central quantity in our analysis of the Coulomb interaction is the retarded
density-density Green’s function (times −1):


χ (r, t) = iθ (t) [ρ (r, t) , ρ (0, 0)]− , (142)

with electron density ρ (r, t) = α ψα† (r, t) ψa (r, t). The Fourier transform of
P
this function with regards to the spatial degrees of freedom is
1 XD E
χq (t) = iθ (t) [ρq (t) , ρ−q0 (0)]−
V 0
q
1 X
= − hhρq (t) ; ρ−q0 ii
V 0
q

In what follows we first determine this function for a system of noninteracting

electrons with Hamiltonian
1 X †
H= εk ψkα ψkα . (143)
V
kα

Recall that our result for the single particle Green’s function
h i 
Gk,k0 (t) = −iθ (t) ψkα (t) , ψk† 0 α (144)
+

is
V δk,k0
Gk,k0 (ω) = , (145)
ω − ε (k)
as follows from the equation of motion.
To determine the density-density Green’s function it is convenient to analyze
the general two-particle Green’s function
DD EE
†
ψkα (t) ψk+qα (t) ; ψk† 0 β ψk0 −q0 β . (146)

The equation of motion requires the analysis of the commutator

h i  
†
ψkα ψk+qα , ψk† 0 β ψk0 −q0 β †
= V δαβ δk0 ,k+q ψkα ψk0 −q0 β − δk0 ,k+q0 ψk† 0 β ψk+qα .
−

This yields with D E

†
ψkα ψk0 α = V δk0 ,k f (εk ) (147)

and Fermi function f (ε) the result:

DD EE
†
ω ψkα ψk+qα ; ψk† 0 β ψk0 −q0 β = V 2 δαβ δk0 ,k+q δq,q0 (fk − fk+q )
DD EE
†
+ (εk+q − εk ) ψkα ψk+qα ; ψk† 0 β ψk0 −q0 β .

25
This can be solved and leads to
DD
†
EE V 2 δαβ δk0 ,k+q δq,q0 (fk − fk+q )
ψkα ψk+qα ; ψk† 0 β ψk0 −q0 β = .
ω − εk+q + εk

Because of δq,q0 follows in particular

1
χq (ω) = − hhρq ; ρ−q ii . (148)
V
Inserting the result for the two-particle Green’s function gives after a few steps

0 1 X fk+q − fk
χrq (ω) = , (149)
V ω + i0+ − εk+q + εk
αk

where we returned to the case of the retarded function.

5.3 Evaluation of the Lindhard function

As we will see next, the density-density response function Eq.(149) can be writ-
ten in the form  
q ω
χrq (ω) = ρF L , (150)
2kF 4εF
with so called Lindhard function L (Q, ν) that is dimensionless and only depends
on dimensionless arguments. We observe that the natural length scale is the
inverse Fermi wave vector 1/kF and the characteristic energy scale is the Fermi
energy. The natural scale for the density response itself is, as we already noticed
above, the density of states at the Fermi level.
Let us evaluate χrq (ω) of Eq.(149). First, we notice the general property

χrq=0 (ω 6= 0) = 0. (151)

On the other hand we can take the static limit ω = 0. For small q follows that

1 X ∂fk
χrq→0 (ω = 0) = −
V ∂εk
αk
ˆ ˆ
dd k ∂fk ∂f (ε)
= −2 d ∂ε
= − dερ (ε) . (152)
(2π) k ∂ε

At T = 0 holds − ∂f∂ε(ε) = δ (ε − εF ) with Fermi energy εF . Thus, we find

χrq→0 (ω = 0) = ρF (153)

with density of states at the Fermi energy ρF . We observe that the limits q → 0
and ω → 0 do not commute for a system with finite density of states at the Fermi
surface.

26
 For the imaginary part holds
ˆ
d3 k
Imχrq (ω) = −2π 3 (fk+q − fk ) δ (ω − εk+q + εk ) . (154)
(2π)

If one analyses Imχrq (−ω) , substitutes k = −k0 − q and uses εk = ε−k , it

follows that
Imχrq (ω) = −Imχrq (−ω) . (155)
Thus, it is sufficient to analyze ω ≥ 0.
Since hhρq ; ρ−q ii involves two operators with ρ−q = ρ†q , we can use the
Kramers-Kronig transformation to determine the real part:
ˆ ∞
r 1 Imχrq ()
Reχq (ω) = − P d
π −∞ ω−
ˆ ∞
2 Imχrq ()
= − P d 2 , (156)
π 0 ω − 2
where we used in the last step that the imaginary part is an odd function. As the
real part only depends on ω 2 , it follows immediately that it is an even function
of frequency:
Reχrq (ω) = Reχrq (−ω) . (157)
2
k
For a parabolic dispersion εk = 2m − µ and at T = 0 one can obtain a closed
expression for the Lindhard function and this for χrq (ω). We first write our
expression as
 
1 X fk fk
χrq (ω) = − , (158)
V ω + i0+ − εk + εk−q ω + i0+ + εk − εk+q
αk

where we substituted k + q → k in the first of the two terms of Eq.(149). We

next use
q2 kqµ
εk − εk±q = − ∓ (159)
2m m
with µ = cos θ and obtain:
ˆ kF 2 ˆ !
r k dk 1 1 1
Reχq (ω) = 2 2 dµ q2
− q2
0 (2π) −1 ω + 2m − kqµ
m ω − 2m − kqµ
m
ˆ kF ˆ 1 !
2m kdk 1 1
= dµ mω q − mω q
q 0 (2π)2 −1 kq + 2k − µ kq − 2k − µ

We introduce dimensionless variables used in Eq.(150):

ω
ν = ,
4εF
q
Q = , (160)
2kF

27
 k
as well as p = kF and obtain
ˆ 1 ˆ 1
!
mkF pdp 1 1
Reχrq (ω) = 2 dµ ν Q
− ν Q
(161)
Q 0 (2π) −1 pQ + p −µ pQ − p −µ

The
√
density of states of a three dimensional parabolic spectrum is ρ (ε) =
2m3/2 √
π2 ε, where we included the spin degeneracy already in ρ (). With ρF =
ρ (εF ) follows
mkF
ρF = (162)
π2
such that
ˆ 1 ˆ 1 !
ρ F 1 1
Reχrq (ω) = pdp dµ ν Q
− ν Q
(163)
4Q 0 −1 pQ + p − µ pQ − p − µ

We observe that for the real part it is indeed possible to express the density re-
sponse in terms of the dimensionless Lindhard function of Eq.(150). We perform
the integration over µ and obtain

ˆ 1
  
 p+ ν +Q p+ ν −Q 
ρF Q Q
Reχrq (ω) =
 
pdp log      . (164)
4Q 0 ν
 p− Q −Q p− Q +Q  ν

In the last step we perform the integration over p and it follows

 
r 0 q ω
Reχq (ω) = ρF L , , (165)
2kF 4εF
where the real part of the Lindhard function is given as:
 ν2
 2 − (Q + 1)2 

ν2
 
0 1 1 2 Q
L (Q, ν) = + 1 − 2 − Q log  ν 2

2 8Q Q  2 − (Q − 1)2 

Q

 ν 2 
2 
ν 
 Q + 1 − Q
− log   . (166)

2
4Q  ν
 Q − 1 − Q2 


If we consider q = 0, i.e. Q = 0, we find with

 ν2
 2 − (Q + 1)2  1

ν2
 
1 2 Q
lim 1 − 2 − Q log  ν 2 =

Q→0 8Q Q  2 − (Q − 1)2  2
Q

and  2 
 ν 2 
ν  Q +1 −Q 

lim log   2 =1
Q→0 4Q  ν
 Q − 1 − Q2 


28
the desired result that
L (0, ν) = 0. (167)
For ω = 0, i.e. ν = 0 we obtain
1 − Q2
  
1 1 − Q
L0 (Q) ≡ L0 (Q, 0) = 1+ log   . (168)
2 2Q 1 + Q
1
For large momenta holds L0 (Q  1) ∼ 3Q 2 while for small momenta holds
1 2
L0 (Q  1) ∼ 1 − 3 Q . For Q → 1 the derivative of L0 (Q) diverges logarith-
mically, while L0 (Q = 1) = 12 is finite.
An interesting application of this result is the response of an ideal gas of
fermions to a point-like potential, i.e.

V ext (r, t) = V0 δ (r) . (169)

If we use our linear response formalism, it follows for the induced charge-density4
ˆ ∞ ˆ
hρ (r)it = dt0 d3 r0 χ (r − r0 , t − t0 ) V ext (r0 , t0 ) . (170)
−∞

Performing the integration over r0 yields a time-independent density profile

hρ (r)i = V0 χ (r, ω = 0)
ˆ
d3 q
 
q
= ρF V0 3 L0 eiq·r . (171)
(2π) 2kF

We can analytically perform this Fourier transformation

ˆ ˆ ∞ 2 ˆ 1
d3 q
   
q iq·r q dq q
3 L0 e = 2 dµL0 eiqrµ
(2π) 2kF 0 (2π) −1 2kF
ˆ ∞  
qdq q
= L0 sin (qr)
0 2π 2 r 2kF
sin(2kF r)
cos (2kF r) − 2kF r
= − 3 . (172)
4π (2kF r)
Thus, an impurity potential induces a density variation that is static and that
oscillates with a period of twice the Fermi momentum
sin(2kF r)
cos (2kF r) − 2kF r
hρ (r)i = −ρF V0 3 . (173)
4π (2kF r)
These oscillations are called Friedel oscillations after the French physicist Jaques
Friedel.
Finally we analyze the imaginary part of the density response. The δ-
function in Eq.(154) implies that only states with εk+q ≥ εk contribute. On the
4 With our above formalism this corresponds to V ext (r, t) = −eϕext (r, t) .

29
other hand, the combination of Fermi functions implies that the occupancy of
the states with momentum k and k + q is different. Taken together, this implies
|k| < kF
|k + q| > kF . (174)
Thus, the imaginary part is sensitive to particle-hole excitations with energy ω.
To proceed we use the version of Eq.(154) after the shift of momenta in the
first term and obtain
ˆ
d3 k
Imχrq (ω) = −2π 3 (δ (ω − εk + εk−q )
k<kF (2π)
−δ (ω + εk − εk+q )) . (175)
After analogous steps as for the real part we obtain
ˆ kF ˆ 1   
r m mω q
Imχq (ω) = − kdk dµ δ + −µ
2πq 0 −1 kq 2k
 
mω q
− δ − −µ . (176)
kq 2k
We can perform the integration over µ = cos θ, switch to dimensionless units,
and consider the function
 
q ω
Imχrq (ω) = ρF L00 , (177)
2kF 4εF
with
ˆ 1     
00 π ν ν
L (Q, ν) = pdp θ p + − Q θ p − + Q
2Q 0 Q Q
   
ν ν
− θ p+ +Q θ p− −Q (178)
Q Q
The integration over p finally yields
("  2 #  2 !
00 π ν ν
L (Q, ν) = 1− −Q θ 1− −Q
4Q Q Q
"  2 #  2 !)
 ν ν
−  1− +Q θ 1− +Q . (179)

 Q Q

The imaginary part is finite between the two curves (we only consider ν > 0 at
the moment)
ν1 = Q + Q2
2
ν2 = Q − 1 + (Q − 1)
Within this regime and for ν3 < Q (1 − Q) the imaginary part is simply linear
in frequency with
πν
L00 (Q, ν) = .
Q

30
5.4 Hartree-Fock analysis of the Coulomb interaction
We finally consider an interacting problem of electrons with long ranged Coulomb
interaction. In real space, our Hamiltonian is given as
Xˆ
H = d3 rψα† (r) ε (−i∇) ψα (r)
α
ˆ
e2 X ψ † (r) ψα† 0 (r0 ) ψα0 (r0 ) ψα (r)
+ d3 rd3 r0 α . (180)
2 0
|r − r0 |
αα

Fourier transformation to momentum space yields

1 X †
H = εk ψkα ψkα
V
kα
1 †
ψk† 0 α0 ψk0 +qα0 ψk−qα
X
+ vkk0 (q) ψkα (181)
2V 3 0 0
kk qαα

with interaction matrix element

4πe2
vk,k0 (q) = . (182)
q2

The indices k and k0 are of course not necessary. We included them here to
allow for an analysis of more general potentials.
We first analyze the single particle Green’s function
DD EE
†
Gk (t) = ψkα (t) ; ψkα (183)

Thus, our first step is to determine the commutator

1 X
[ψkα , H] = εk ψkα + vk,k0 (q) ψk† 0 α0 ψk0 +qα0 ψk−qα (184)
V2 0 0
k qα

which gives rise to the equation of motion

DD
†
EE 1 X
(ω − εk ) ψkα ; ψkα = 1+ vk,k0 (q)
ω V2 0 0
k qα
DD EE
× ψk† 0 α0 ψk0 +qα0 ψk−qα ; ψkα
†
ω

We see that the interaction between particles generated a more complicated

Green’s function. If one tries to analyze this, more complicated function one
obtains an even worse set of operators in the commutator and, accordingly, an
even more complicated Green’s function. Thus, we are not able to solve this
problem exactly any-longer.

31
 Later we will see that it is efficient to define a so called self-energy via
DD EE
1
ψk† 0 α0 ψk0 +qα0 ψk−qα ; ψkα
†
P
k0 qα0 vk,k (q)
0
V2
ω
Σk (ω) = DD EE , (185)
†
ψkα ; ψkα
ω

which yields
1
Gk (ω) = . (186)
ω − εk − Σk (ω)
At this point this definition is not overly useful, but we will see that the self
energy has a comparatively easy definition in terms of Feynman diagrams. This
form of the Green’s function reminds us of the Dyson equation
−1 −1
Gk (ω) = G0k (ω) − Σk (ω) , (187)

that we encountered in case of non-interacting fermions or bosons with bare

1
Green’s function G0k (ω) = ω−ε k
. The self energy plays a role of a frequency
dependent potential for individual particles.
The form of the self energy suggests to search for an approximation, where
one replaces the more complicated Green’s function by a factor times the ordi-
nary single particle Green’s function.
We can make progress if we recall a result for a more complicated Green’s
function that we obtained for free particles. Consider GAB = Gi,jkl with A = ψi
and B = ψj† ψk† ψl , we found
D E D E
Gi,jkl (ω) = Gij (ω) c†k cl − ηGik (ω) c†j cl . (188)

If we use GAB = GB † A† , it follows for GB † A† = Glkj,i with with A† = ψi† and

B † = ψl† ψk ψj , we obtain
D E D E
Glkj,i (ω) = Gij (ω) ψk† ψl − ηGik (ω) ψj† ψl . (189)

Of course, these expressions are only correct for non-interacting particles. In

what follows we will simply assume that we are allowed to use those relations
for interacting particles as well. This implies
DD EE DD EE D E
ψk† 0 α0 ψk0 +qα0 ψk−qα ; ψkα
†
= †
ψkα ; ψk−qα ψk† 0 +qα0 ψk0 α0
ω ω
D E
†
−η hhψkα ; ψk0 +qα0 iiω ψk−qα ψk0 α0 (190)
.

We use that the the Green’s functions are diagonal in the momentum and spin
†
indices and obtain with nkα = ψkα ψkα :
DD EE
ψk† 0 α0 ψk0 +qα0 ψk−qα ; ψkα
†
= Gk (ω) V (δq,0 hnk0 α0 i − ηδk,k0 +q δαα0 hnk0 α i)
ω

32
If we insert this into the equation of motion or, alternatively, into our definition
for the self energy, we find
1 X
Σk (ω) = vk,k0 (q) (δq,0 hnk0 α0 i − ηδk,k0 +q δαα0 hnk0 α0 i)
V 0 0
k qα
1 X 1 X
= vk,k0 (0) hnk0 α0 i − η vk,k0 (k − k0 ) hnk0 α i (191)
V 0 0 V 0
kα k

This is the Hartree-Fock approximation for the self-energy. The first term cor-
responds to the Hartree term and the second to the Fock term. The philosophy
of it’s derivation is indeed the same as in the ordinary, first-quantized formula-
tion of quantum mechanics, where one evaluates expectation values with regards
to an assumed Slater determinant, i.e. a free-particle wave function. Here we
also evaluated interaction effects in the equation of motion, pretending that the
higher order Green’s function can be analyzed the same way as for free particles.
There is a simple rule that exists for the evaluation of these higher order
Green’s function. We consider the operator ψk† 0 α0 ψk0 +qα0 ψk−qα and replace a
pair made out of a creation and an annihilation operator, once they are next to
each other, by its expectation value. If it is necessary to change the order a minis
sign might occur for due to the anti-commutation rules. Thus we approximate
D E
ψk† 0 α0 ψk0 +qα0 ψk−qα ≈ ψk† 0 α0 ψk0 +qα0 ψk−qα
D E
− ψk† 0 α0 ψk−qα ψk0 +qα0 . (192)

If we further use that the momenta and spins in the expectation value have to
be the same, it follows

ψk† 0 α0 ψk0 +qα0 ψk−qα ≈ δq,0 hnk0 α0 i ψk−qα

− δk,k0 +q δαα0 hnk0 α0 i ψk0 +qα0 . (193)

Inserting this expression into the Green’s function yields the above result for
the self energy.
If we insert this self energy into the Dyson equation we finally obtain a
result for the Green’s function within Hartree-Fock approximation. This Green’s
function can then be used to determine expectation values like hnk0 α0 i. Since
the self-energy itself depends on those expectation values, one sees that one is
confronted with a self-consistency issue. Thus, one has to assume a certain result
for hnk0 α0 i, use this ansatz to determine the self energy and Green’s function
and check whether resulting hnk0 α0 i agrees with the initial ansatz. If not, one
has to use a better ansatz. A good approach to solve this self-consistency
problem is to use the occupation one just obtained and an improved ansatz
and iterate this problem until convergence is reached. This self consistency
procedure demonstrates that one includes in the Hartree-Fock approach effects
to arbitrary order in the interaction. This sounds encouraging, however, one
certainly didn’t include all effects. Only the first order contribution to the self

33
energy that is completely contained in the approach. We will discuss these issues
later-on during the course.
Returning to the Coulomb problem one should get rather nervous if one looks
at the Hartree term. Here one has to analyze the matrix element vkk (q = 0)
while the interaction diverges like q −2 for small q. Here the self energy amounts
to a shift in energy
4πe
δ = − lim 2 ρ0
q→0 q
´ d3 k P
where ρ0 = −e (2π)3 α hnkα i is the electron density in equilibrium. Thus,
this energy can be written as
ˆ
ρ0
δ = −e d3 r (194)
|r|
Physically this is the Coulomb repulsion of a given electron by all other electrons
in the system. This yields an infinite shift of the energy of each individual
electron and seems to suggest that the system is not stable. In fact we know that
for an infinite system the energy density of a homogeneously charged plasma is
infinite because of Coulomb’s law. In the real solid this is compensated by the
attractive potential that each electron feels because of the positively charged
ions. The instability we obtained is solely a consequence of the fact that we
forgot to include the positively charged ions in our Hamiltonian. Since ρ0 =
−ρion and because we consider a jellium model with homogeneously charged
background, we include this Coulomb interaction by shifting all energies by
ˆ
ρion
δion = −e d3 r (195)
|r|

in the Hamiltonian (ε (k) → ε (k) + δion ). As a frequency independent self

energy does nothing else but correcting the energies in our Hamiltonian, we see
that the Hartree term exactly cancels δion . Thus, for a Coulomb problem we
only need to include the Fock term, which for fermions with η = +1 is given as
ˆ
d3 k 0 4πe2
Σk = − 3 2 hnk α i
0
(2π) |k − k0 |
It amounts to a correction of the single particle energy, i.e.
1
Gk (ω) = . (196)
ω − ε∗k

with renormalized dispersion

ε∗k = εk + Σk (197)

As long as the Coulomb interaction does not break the rotational symmetry of
the system, the renormalized dispersion will continue to depend on the magni-
tude of the momentum: k = |k|. Thus, the system will be characterized by a

34
Fermi momentum kF . Since the Green’s function looks like the one of a free
Fermi system, it must furthermore hold at T = 0 that hnkα i = 1 for k < kF
and hnkα i = 0 for k > kF , which yields that kF must be unchanged by the
interaction and is fixed by the total density. If follows
ˆ kF ˆ 1
e2 02 0 1
Σk = − k dk dµ
2 + k 02 − 2kk 0 µ
π 0 −1 k
ˆ
e2 kF 0 0  k + k0 
 
= − k dk log  
πk 0 k − k0 
ˆ 1
e2 kF2
 
 k/kF + p 
= − pdp log  
πk 0 k/kF − p 
2 !
e2 kF

1 (k/kF ) − 1  k/kF + 1 
= − 1+ log  
π 2 k/kF k/kF − 1 

2
We see that at kF the self energy is ΣkF = − e πkF . This quantity can alterna-
tively be absorbed in the new chemical potential, needed to obtain the correct
density. More interesting is the momentum dependence of the self energy. It
holds for k near kF that

e2 kF
 
∂Σk  2kF 
= log 
 , (198)
∂k π k − kF 

i.e. we obtain a logarithmically divergent correction to the velocity

∂ε∗k k ∂Σk
vk∗ = = + . (199)
∂k m ∂k
This result played a role in the early stages of the theory of interactions in met-
als5 However, the inclusion of higher order processes of the perturbation theory
strongly suggests that this divergency is spurious. Nevertheless, the Hartree-
Fock approximation is a useful tool to get a first idea about an interacting
many-body system.

5.5 The random phase approximation for density correla-

tions
In this final chapter on screening of the Coulomb interaction we determine the
density-density Green’s function of an interacting system within an approxima-
tion that is qualitatively similar to the Hartree-Fock approximation.
In full analogy to the case of non-interacting fermions we analyze the Green’s
function DD EE
†
ψkα (t) ψk+qα (t) ; ψk† 0 β ψk0 −q0 β . (200)

5 J. C. Slater Phys. Rev. 81, 385 (1951).

35
The additional calculation is the hdetermination of
i the commutator recall that
†
we already used the commutator ψkα ψk+qα , HC where HC is the Coulomb-
−
interaction in the Hamiltonian H. It holds
h i 
† †
ψk† 1 α1 ψk1 −q1 α1 ψk+q+q1 α
X
ψkα ψk+qα , HC = v (q) ψkα
−
k1 q1 α1

† †
− ψk+q ψ
1 α k1 α1
ψk 1 +q α
1 1
ψ k+qα , (201)

where we used that v (q) = v (−q). Just like in the case of the Hartree-Fock
approximation, we express certain operators in the Green’s function by expec-
tation values. For the first term in the above commutator follows
†
ψkα ψk† 1 α1 ψk1 −q1 α1 ψk+q+q1 α ≈ †
hnk1 α1 i δq1 ,0 ψkα ψk+q+q1 α
+ hnkα i δq1 ,−q ψk† 1 α1 ψk1 −q1 α1
− hnkα i δk,k1 −q1 δαα1 ψk† 1 α1 ψk+q+q1 α
†
− hnk1 α1 i δk1 ,k+q+q1 δαα1 ψkα ψk1 −q1 α1(202)

where we “contracted” two operators that are next to each other to their ex-
pectation value. The minus signs are a result of the fact that we had to bring
operators next to each other.
†
ψk+q ψ † ψk1 +q1 α1 ψk+qα
1 α k1 α1
≈ †
hnk1 α1 i δq1 ,0 ψk+q1α
ψk+qα
+ hnkα i δq1 ,q ψk† 1 α1 ψk1 +q1 α1
− hnkα i δk,k1 δαα1 ψk† 1 α1 ψk+qα
†
− hnk1 α1 i δk1 ,k+q δαα1 ψk+q 1α
ψk1 +q1 α1(203)

Inserting these approximate expressions into the equation of motion yields

DD EE
†
(ω − Ek,q ) ψkα ψk+qα ; ψk† 0 β ψk0 −q0 β = V 2 δαβ δk0 ,k+q δq,q0 (fk − fk+q )
X
+ (fk − fk+q ) (v (q) − v (k − k1 ) δαα1 )
k1 α1
DD EE
× ψk† 1 α1 ψk1 +qα1 ; ψk† 0 β ψk0 −q0 β

where
Ek,q = ε∗k+q − ε∗k
are the Hartree-Fock corrected single particle energies. In what follows we will
ignore these Hartree-Fock corrections. It turns out that the term with v (k − k1 )
cannot be expressed in terms of the density-density correlation function. It
corresponds to other degrees of freedom but the density, which are coupled to
density fluctuations via the Coulomb interaction. A frequent assumption is that
those degrees of freedom are not coherently coupled to density fluctuations and

36
will be damped out rapidly. In what follows we will also ignore this term. This
is called the random phase approximation.
It follows for the density-density Green’s function

χq (ω) = χ(0)
q (ω) (1 − v (q) χq (ω)) (204)

which can be solved and yields

(0)
χq (ω)
χq (ω) = (0)
, (205)
1 + v (q) χq (ω)

where
1 X fk+q − fk
χ(0)
q (ω) = (206)
V ω + i0+ − εk+q + εk
αk

is the density-density Green’s function of free particles that can be expressed in

terms of the Lindhard function.
If we combine our relation Eq.(138) between dielectric function with the
density Green’s function with Eq.(205), it follows

4πe2 (0)
ε (q, ω) = 1 + χ (ω) (207)
q2 q
Let us first analyze the response to a static test charge, i.e. the limit ω = 0. In
this limit follows  
q 2 + 4πe2 ρF L0 2kqF
ε (q, 0) = (208)
q2
with
1 − Q2
  
1 1 − Q
L0 (Q) = 1+ log 
  . (209)
2 2Q 1 + Q
In the long wave length limit follows L0 (Q  1) ∼ 1 − 31 Q2 and we obtain
 2

q 2 1 − 2πe
3kF2
ρF
+ 4πe2 ρF
ε (q, 0) = . (210)
q2
If we are in the limit of a good metal, we would expect that the typical kinetic
energy is larger than the typical Coulomb interaction, i.e.

kF2 e2
 (211)
2m λF
2π 2πe2 ρF
with Fermi wave number λF = kF . This yields 3kF2  1 and we obtain

q 2 + qT2 F
ε (q, 0) = , (212)
q2

37
with Thomas-Fermi wave vector qT F determined by

qT2 F = 4πe2 ρF . (213)

We can now determine the electric potential of a static point charge

4πe
ϕ (q, t) = .
ε (q, ω = 0) q 2
4πe
= , (214)
q 2 + qT2 F
which yields a long distance dependence in coordinate space
e2 −qT F |r|
ϕ (r, t) = e . (215)
|r|
Thus, the interaction of charges is screened beyond the Thomas-Fermi length
qT−1F .
Finally we can analyze whether we can fulfill the condition ε (q, ω) = 0,
which amounts to a resonant response to an infinitesimal external charge per-
turbation.
4πe2
 
q ω
1 + 2 ρF L , =0 (216)
q 2kF 4εF
In the long wavelength limit we can expand the Lindhard function for small
momenta but arbitrary energies. It holds
Q2 Q4
L (Q  1, ν) = − 2
− 4 ··· . (217)
3ν 5ν
This yields
16π 2 2F 122F q 2
 
2
ω = e ρF 2 1+ 4 2 . (218)
3 kF 5kF ω
The coefficient
16π 2 2F 4 2 kF3
e ρF 2 = e
3 kF 3π m
´ d3 k
3
kF
can be expressed in terms of the particle density n = k<kF (2π) 3 = 3π 2 and it

follows
3 2
ω 2 = ωpl
2
+ q (219)
5m2
with plasma frequency
2 4πne2
ωpl = . (220)
m
Since the imaginary part of the Lindhard function vanishes at small q and
finite ω, these collective modes are undamped. The correspond to the emergent
response of a charged fluid to an external charge perturbation. The plasma
frequency of usual metals is in the regime of several electron volts, i.e. it is a
very high frequency collective mode.

38
Part III
Diagrammatic perturbation
theory at finite T
In the previous chapters we learned that retarded Green’s function can be de-
termined from an analysis of the equation of motion. The approach is very
straightforward and powerful for non-interaction systems. However, as soon as
one wants to incorporate effects of interactions, the method is not very trans-
parent and efficient. An elegant alternative is the analysis of Green’s functions
using Feynman diagrams. A diagrammatic perturbation theory can be devel-
oped for time ordered Green’s functions. Historically the formulation was done
first for the so called causal Green’s function
c
GcA,B (t, t0 ) = hhA (t) ; B (t0 )ii
≡ −i hT A (t) B (t0 )i , (221)

with time ordering operator

T A (t) B (t0 ) = θ (t − t0 ) A (t) B (t0 ) − ηθ (t0 − t) B (t0 ) A (t) . (222)

While these functions can be efficiently determined in terms of Feynman dia-

grams, it holds that they are analytic functions in the complex plane only in the
limit T = 0. For this reason we will not discuss causal Green’s function further.
Instead, an elegant and very efficient approach valid also at finite temperatures
can be developed in terms of Matsubara. Before we discuss Matsubara functions
we briefly summarize the concept of the S-matrix, as it will play an important
role in our subsequent analysis.
Let us consider a Hamiltonian

H = H0 + V (223)

that consists of a free part H0 and an interaction part V . The time evolution
is governed by
0 0
e−iH (t−t ) = e−iH0 t S (t, t0 ) eiH0 t (224)
which defines the S-matrix. With this definition follows for the time dependence
of an arbitrary operator in Heisenberg representation

A (t) = eiHt Ae−iHt

= S † (t, 0) e−iH0 t Ae−iH0 t S (t, 0)
= S † (t, 0) Ã (t) S (t, 0) (225)

where
à (t) = e−iH0 t Ae−iH0 t . (226)

39
In order to determine the S-matrix we consider the time derivative of
0 0
S (t, t0 ) = eiH0 t e−iH (t−t ) e−iH0 t . (227)

It holds
0 0
∂t S (t, t0 ) = iH0 eiH0 t e−iH (t−t ) e−iH0 t
0 0
− eiH0 t (iH) e−iH (t−t ) e−iH0 t
0 0
= eiH0 t (iH0 − iH) e−iH (t−t ) e−iH0 t
0 0
= −eiH0 t (iV ) e−iH0 t eiH0 t e−iH (t−t ) e−iH0 t
= −iṼ (t) S (t, t0 ) (228)

To determine the S-matrix we have to include the boundary condition

S (t, t) = 1. (229)

The solution of the above differential equation is

´t
dt00 Ṽ (t00 )
S (t, t0 ) = T e−i t0 . (230)

Let us demonstrate that this is indeed the correct solution. The boundary
condition is clearly obeyed. Next, we expand the exponential function:
∞
X
0
S (t, t ) = Sn (t, t0 ) (231)
n=0

with
n ˆ t ˆ t ˆ t
(−i)
Sn = T dtn · · · dt2 dt1 Ṽ (tn ) · · · Ṽ (t2 ) Ṽ (t1 ) . (232)
n! t0 t0 t0

There are n! possibilities to order of the time variables ti . We could for example
relabel the ti such that the earliest is called t1 , followed by t2 etc. Then holds
ˆ t ˆ t3 ˆ t2
n
Sn = (−i) T dtn · · · dt2 dt1 Ṽ (tn ) · · · Ṽ (t2 ) Ṽ (t1 ) . (233)
t0 t0 t0

Of course with this specific relabeling we may also skip the time ordering oper-
ation, i.e.
ˆ t ˆ t3 ˆ t2
n
Sn = (−i) dtn · · · dt2 dt1 Ṽ (tn ) · · · Ṽ (t2 ) Ṽ (t1 ) . (234)
t0 t0 t0

It follows
∂t Sn (t, t0 ) = −iṼ (t) Sn−1 (t, t0 ) , (235)

40
where obviously holds that S−1 (t, t0 ) = 0. This yields
∞
X
∂t S (t, t0 ) = −iṼ (t) Sn−1 (t, t0 )
n=0
∞
X
= −iṼ (t) Sn (t, t0 )
n=−1

= −iṼ (t) S (t, t0 ) .

Thus, we found the correct solution.

6 The Matsubara function

The Matsubara function is motivated by the close analogy between time evolu-
tion and thermal averaging. One introduces

A (τ ) = eτ H Ae−τ H (236)

and defines
GAB (τ, τ 0 ) = − hT A (τ ) B (τ 0 )i . (237)
with

T A (τ ) B (τ 0 ) = θ (τ − τ 0 ) A (τ ) B (τ 0 ) − ηθ (τ 0 − τ ) B (τ 0 ) A (τ ) . (238)

It is immediately evident why one often refers to the Matsubara approach as

the imaginary time approach with

t → −iτ. (239)

It is easy to show that the Green’s function is homogeneous with regards to

time, i.e. that
GAB (τ, τ 0 ) = GAB (τ − τ 0 ) . (240)
This follows again from the fact that the “time-evolution” and the thermal
averaging is governed by the Hamiltonian H.

6.1 Periodicity of the Matsubara function and Matsubara

frequencies
Next we analyze the detailed time dependence of GAB (τ ) and show that it is an
periodic (anti-periodic) function for bosonic (fermionic) Green’s functions. We
consider an arbitrary integer m and consider values of τ that obey:

mβ < τ < (m + 1) β. (241)

It then follows that

1
GAB (τ − mβ) = − tr e−βH T A (τ − mβ) B


Z

41
Since τ − mβ > 0 we can drop the time ordering symbol:
1  
GAB (τ − mβ) = − tr e−βH e(τ −mβ)H Ae−(τ −mβ)H B
Z
1  
= − tr e(τ −(m+1)β)H Ae−(τ −mβ)H B
Z
1  
= − tr e−βH Be(τ −(m+1)β)H Ae−(τ −(m+1)β)H
Z
1
= − tr e−βH BA (τ − (m + 1) β)


(242)
Z
If τ < 0 holds
BA (τ ) = −ηT A (τ ) B.
Since τ − (m + 1) β < 0 it follows
η
tr e−βH T A (τ − (m + 1) β) B


GAB (τ − mβ) =
Z
= −ηGAB (τ − (m + 1) β) (243)
In particular follows for m = −1 that:
GAB (τ ) = −ηGAB (τ + β) . (244)
Matsubara functions are periodic (anti-periodic) for bosonic (fermionic) choice
of the time ordering. Since both functions are periodic with period 2β we can
always expand in the Fourier series
∞
1 X −iωn τ
G (τ ) = e G (ωn ) (245)
β n=−∞

where e−2iβωn = 1, i.e. ωn = nπ/β. The Fourier coefficients are:

ˆ
1 β
G (ωn ) = dτ G (τ ) eiωn τ . (246)
2 −β
This incorporates the information with regards to the period 2β. We do however
have even more information. It holds
e−iβωn = −η. (247)
In case of η = −1, i.e. for bosons, we know that the period is in fact β. Thus,
only the Matsubara frequencies with ωn = 2nπ/β contribute. For fermionic
functions we have G (τ + β) = −G (τ ), i.e. e−iβωn = −1, such that now only
odd multiples of π/β contribute and we have ωn = (2n + 1) π/β. For the Fourier
coefficients follows then
ˆ ˆ
1 β iωn τ η 0
G (ωn ) = dτ G (τ ) e − dτ G (τ + β) eiωn τ
2 0 2 −β
ˆ
1 − ηe−iωn β β
= dτ G (τ ) eiωn τ
2 0
ˆ β
= dτ G (τ ) eiωn τ . (248)
0

42
In summary, we have discrete Matsubara frequencies that are distinct for bosonic
and fermionic propagators:
(
2nπ/β for bosons
ωn = . (249)
(2n + 1) π/β for fermions

6.2 Relation to the retarded function

All this looks rather artificial as no obvious relation to reality seems to exist
between the Matsubara functions and physical observables. However, if we
repeat the same steps that led to the spectral representation of the retarded
Green’s function it follows:
ˆ
1 β  
GAB (ωn ) = − dτ eiωn τ tr e(τ −β)H Ae−τ H B
Z 0
ˆ
1 X β
= − dτ eiωn τ e(τ −β)El e−τ Em hl |A| mi hm |B| li
Z 0
lm
1 X e−βEl − e−βEm eiβωn
= hl |A| mi hm |B| li (250)
z iωn + El − Em
lm

Since eiβωn = −η, it follows

1 X e−βEl + ηe−βEm hl |A| mi hm |B| li

GAB (ωn ) = (251)
Z iωn + El − Em
l.m

If we compare this with the general Lehmann representation that was derived
earlier, it follows
GAB (ωn ) = GAB (z = iωn ) . (252)
We already discussed that we can define Green’s function in the entire complex
plane and that the only source for non-analyticity is the real axis. Now we
see that the Matsubara function yields the complex Green’s function at the
purely imaginary Matsubara frequencies. Thus, if we determine the Matsubara
function, we can determine the retarded function via analytic continuation

iωn → ω + i0+ . (253)

Thus, knowledge of the Matsubara function allows for the determination of the
retarded function.
This immediately yields information about the single-particle Matsubara
Green’s function. Consider the Hamiltonian
X ˆ dd k
H0 = ε ψ† ψ .
d k kα kα
(254)
α (2π)

43
We obtain for the Fourier transform of
D E
†
G0k (τ ) = − T ψkα (τ ) ψkα (255)
0

that
1
G0,k (ωn ) = . (256)
iωn − εk
Here we indicate with h· · · i0 that the average is with regards to the Hamiltonian
H0 .

6.3 Evolution with imaginary time

We already introduced the time dependence introduces

A (τ ) = eτ H Ae−τ H (257)

which can be used to determine the equation of motion

∂τ A (τ ) = HA (τ ) − A (τ ) H
= − [A (τ ) , H] (258)

For our subsequent analysis we will use

eτ H e−τ H = 1, (259)

i.e.
−1
eτ H = e−τ H . (260)
In full analogy of the S-matrix we can introduce
0 0
S (τ, τ 0 ) = eH0 τ e−H (τ −τ ) e−H0 τ . (261)

The time evolution of the full Hamiltonian is written as

0
e−H (τ −τ ) = e−H0 τ S (τ, τ 0 ) eH0 τ ; . (262)

In distinction to the real-time S-matrix, S (τ, τ 0 ) is not unitary. It does however

hold
S (τ, τ ) = 1 (263)
as well as

S (τ1 , τ2 ) S (τ2 , τ3 ) = eH0 τ1 e−H(τ1 −τ2 ) e−H0 τ2 eH0 τ2 e−H(τ2 −τ3 ) e−H0 τ3
= S (τ1 , τ3 ) . (264)

Using τ3 = τ1 , this implies in particular that

−1
S (τ1 , τ2 ) = S (τ2 , τ1 ) . (265)

44
The time evolution of the full Hamiltonian is written as

A (τ ) = eτ H e−τ H0 Ã (τ ) eτ H0 e−τ H
= S (0, τ ) Ã (τ ) S (τ, 0) , (266)

where
à (τ ) = eτ H0 Ae−τ H0 . (267)

The equation of motion for the imaginary-time version of the S-matrix follows
in full analogy to the case with real times
0 0
−∂τ S (τ, τ 0 ) = −eH0 τ (H0 − H) e−H (τ −τ ) e−H0 τ
= Ṽ (τ ) S (τ, τ 0 ) . (268)

The solution of this operator differential equation is obtained along the lines
discussed above and yields
´τ
dτ 00 Ṽ (τ 00 )
S (τ, τ 0 ) = T e− τ0 .

This result can for example we used to express the partition function or
Green’s functions in a manner that is well suited for a perturbation theory. In
case of the partition function holds:

Z = tre−βH
tr e−βH0 S (β, 0)


=
= Z0 hSi0 (269)

with our earlier definition for the average w.r.t. H0 and with
´β
dτ 00 Ṽ (τ 00 )
S ≡ S (β, 0) = e− 0 . (270)

Thus, we can express the fully interacting partition sum in terms of expectation
values of the noninteracting problem. The same reasoning can be performed for
the single particle Green’s function as

1  †

Gk (τ ) = − tr e−βH T ψkα (τ ) ψkα (0)
Z
1  †

= − tr e−βH0 S (β, 0) S (0, τ ) ψ̃kα (τ ) S (τ, 0) ψ̃kα (τ 0 )
Z
1  †

= − tr e−βH0 T S (β, τ ) ψ̃kα (τ ) S (τ, 0) ψ̃kα (0)
Z
1  †

= − tr e−βH0 T S (β, τ ) S (τ, 0) ψ̃kα (τ ) ψ̃kα (0)
Z
1  †

= − tr e−βH0 T ψ̃kα (τ ) ψ̃kα (0) S (β, 0) (271)
Z

45
If we combine this with our representation for the partition function we obtain
D E
†
T ψ̃kα (τ ) ψ̃kα (0) S (β, 0)
0
Gk (τ ) = −
hS (β, 0)i0
D E
†
T ψ̃kα (τ ) ψ̃kα (0) S
0
= − (272)
hSi0
The appeal of this formulation is that we can develop a perturbation theory in
the potential V by expanding the exponentials in the numerator and denomi-
nator, respectively.

7 Wick theorem
The formulation of the Matsubara function in the previous section demonstrated
that we need to evaluate expectation values of higher order operators with re-
gards to a non-interacting Hamiltonian. In our earlier analysis of retarded
Green’s functions of free particles we already noticed that higher order Green’s
functions can be evaluated for free particles. In what follows we develop an
efficient formalism to do this for the Matsubara function. We consider a Hamil-
tonian X †
H0 = εi ci ci , (273)
i

where i stands for an arbitrary set of single particle quantum numbers, for
example spin and momentum. We will also use the shorthand notation

αi = ci or c†i . (274)

7.1 Time evolution of creation and annihilation operators

First we analyze the specific time evolution of the creation and annihilation
operator. It holds
ci (τ ) = eτ H0 ci e−τ H0 . (275)
We can solve the associated equation of motion

∂τ ci (τ ) = − [ci (τ ) , H0 ]
= −εi ci (τ ) , (276)

which yields
ci (τ ) = e−τ εi ci (0) = e−τ εi ci . (277)
Thus, it follows
ci e−τ H0 = e−τ H0 ci e−τ εi . (278)
An analogous expression follows for

c†i (τ ) = eτ H0 c†i e−τ H0 (279)

46
and yields
c†i (τ ) = eτ εi c†i (280)
as well as
c†i e−τ H0 = e−τ H0 c†i eτ εi . (281)
Let us stress our notation. While c†i
is obviously the hermitian conjugate of ci ,
this does not hold that ci (τ ) and ci (τ ) once τ 6= 0. c†i (τ ) merely refers to the
†
†
time evolution of c†i , while (ci (τ )) = c†i (−τ ).
We can summarize our findings as follows

αi e−τ H0 = e−τ H0 αi esi τ εi , (282)

where
αi = c†i

+1 if
si = . (283)
−1 if αi = ci

7.2 Wick theorem of time-independent operators

Next we proof the Wick-theorem of time-independent operators. We analyze
the following expectation value of n creation and n annihilation operators:
1
tr e−βH0 α1 α2 · · · α2n .


hα1 α2 · · · α2n i0 = (284)
Z0
Our goal is to proof that
n
hα1 α2 · · · α2n i0 = α1 α2 α3 α4 · · · α2n−1 α2n

+ α1 α2 α3 α4 · · · α2n−1 α2n
+ ···}, (285)

where the sum goes over all possible pairwise contractions. A contraction is
defined as
αi αj = hαi αj i0 (286)
and interchanging contracted operators produces a sign η,

αi αj αk αl = −ηαi αk αj αl , (287)

where η = +1 for fermions and η = −1 for bosons. It also holds

[αi , αj ]η = αi αj + ηαj αi
 h i
†

 δ ij for ci , cj
iη


 h
†
= ηδij for ci , cj . (288)
 h η i
for [ci , cj ]η or c†i , c†j

 0


η

47
To proof this statement we use
αi αj = −ηαj αi + [αi , αj ]η . (289)
and write
α1 α2 α3 · · · α2n = [α1 , α2 ]η α3 · · · α2n
− ηα2 α1 α3 · · · α2n . (290)
If we look at the second term, we find similarly
α2 α1 α3 · · · α2n = α2 [α1 , α3 ]η · · · α2n
− ηα2 α3 α1 · · · α2n . (291)
If we repeat this 2n − 1 times, it follows
2n
X j−2
α1 α2 · · · α2n = (−η) α2 · · · αj−1 [α1 , αj ]η αj+1 · · · α2n
j=2
2n−1
+ (−η) α2 α3 · · · α2n α1 . (292)
2n−1
It holds of course that (−η) = −η.
Next we perform the thermodynamic average
1
tr e−βH0 α2 α3 · · · α2n α1


hα2 α3 · · · α2n α1 i0 =
Z0
es1 βε1
tr e−βH0 α1 α2 α3 · · · α2n


=
Z0
= hα1 α2 α3 · · · α2n i0 . (293)
Here we used our previous result Eq.(282). From Eq.292 follows then
2n
X j−2
[α1 , αj ]η
hα1 α2 α3 · · · α2n i0 = (−η) hα2 · · · αj−1 αj+1 · · · α2n i0 .
j=2
1 + ηes1 βε1
(294)
Since we are dealing with free particles, it holds of course that
D E δij ηδij
c†i cj = βε
= ,
0 e +η i 1 + ηeβεi
h i h i
c†i , cj c†i , cj
η η
= = (295)
1 + ηeβεi 1 + ηesi βεi
as well as
D E D E δij
ci c†j = δij − η c†j ci = .
0 0 1 + ηe−βεl
h i
ci , c†j
η
= , (296)
1 + ηesi βεi

48
 D E
while c†i c†j = hci cj i0 = 0. With our definition of a contraction, it follows
0
then
[αi , αj ]η
= αi αj . (297)
1 + ηesi βεi
which leads to:
2n
X j−2
hα1 α2 α3 · · · α2n i0 = (−η) hα2 · · · αj−1 α1 αj αj+1 · · · α2n i0 . (298)
j=2

The contraction itself is, as we just saw, only a number.

We further define, as intermediate steps of the analysis, contractions between
operators that are not neighbors, i.e.

αi αj αk αl = −ηαj αi αk αl
= −ηαi αj αl αk
= αj αi αl αk (299)

i.e. we modify the sign of the expression by −η, depending on whether we

performed an even or odd number of exchanges. Then follows
2n
X j−2
hα1 α2 α3 · · · α2n i0 = α1 αj (−η) hα2 · · · αj−1 αj+1 · · · α2n i0 .
j=2
2n D
X E
= α1 α2 · · · αj−1 αj αj+1 · · · α2n (300)
0
j=2

We can now repeat the procedure for the remaining operators and obtain the
sum over all possible pairwise contractions, which proves Wick’s theorem.
Two examples are:

D E
c†1 c2 c†3 c4 = c†1 c2 c†3 c4 + c†1 c2 c†3 c4
0
D E D E D E D E
= c†1 c2 c†3 c4 + c†1 c4 c2 c†3 (301)
0 0 0 0

and

D E
c†1 c†2 c3 c4 = c†1 c†2 c3 c4 + c†1 c†2 c3 c4
0
D E D E D E D E
= c†1 c4 c†2 c3 − η c†1 c3 c2 c†4 .
0 0 0 0

7.3 Wick theorem for time dependent operators

Our next step is to consider time dependent expectation values of the sort

49
 For the evaluation of Gk (τ ) and even for the partition function Z = Z0 hSi0 ,
we consider time dependent expectation values
D E
T ci1 (τ1 ) · · · cin (τn ) c†jn (τn0 ) · · · cj1 (τ10 ) . (302)
0

In analogy to our previous notation we use the shorthand

αi (τ ) = ci (τ ) or c†i (τ ) . (303)

Our results for the time dependence of the operators yields

αi (τ ) = esi τ εi αi , (304)

where we remind that

αi = c†i

+1 if
si = . (305)
−1 if αi = ci

Thus, we want to analyze

P2n
sj τj εj
hT α1 (τ1 ) α2 (τ2 ) · · · α2n (τ2n )i0 = e j=1 hT α1 α2 · · · α2n i0 (306)

Here the time ordering operator arranges the time independent αi corresponding
to the associated time τi . Let us for the time beeing assume that the operators
are already time ordered, i.e. that

τ1 > τ2 > · · · τ2n . (307)

Then we only have to analyze

hα1 α2 · · · α2n i0 (308)

for which we can use the above Wick theorem.

To proceed we define the contraction of time dependent operators

αi (τi ) αj (τj ) ≡ hT αi (τi ) αj (τj )i0 (309)

which we can for τi > τj always write as

αi (τi ) αj (τj ) = eτi si εi +τj sj εj αi αj (310)

This implies

hT α1 (τ1 ) α2 (τ2 ) · · · α2n (τ2n )i0 = α1 (τ1 ) α2 (τ2 ) α3 (τ3 ) α4 (τ4 ) · · · α2n−1 (τ2n−1 ) α2n (τ2n )

+ α1 (τ1 ) α2 (τ2 ) α3 (τ3 ) α4 (τ4 ) · · · α2n−1 (τ2n−1 ) α2n (τ2n )

+ ···}, (311)

50
where the sum goes again over all possible pairwise contractions and each cross-
ing of contraction brackets amount to a factor −η.
In our analysis we made the asssumption that the operators are already time
ordered. Suppose this is not the case. Then we need to perform p changes in the
p
order of the operators, amounting to a factor (−η) . This resulting expression
can then be analyzed using Wick’s theorem. We can then reorder the operators
to return to the original, not-time ordered order. However, given our sign-
p
rules for contractions, this also just amounts to another factor (−η) . The total
2p
factor is then (−η) = 1. Thus, the above expansion is equally valid even if the
operators are not yet time ordered. These insigts can alternatively be written
as that the higher order Green;s function
D E
G0 (i1 , · · · in ; j1 · · · jn ) = (−1) T ci1 · · · cin c†jn · · · cj1 ,
(n) n
(312)
0

can be expressed as
(n)
X P


G0 (i1 , · · · in ; j1 · · · jn ) = (−η) G0 i1 ; jP (1) · · · G0 in ; jP (n) . (313)
P
P
The sum P runs over all possible permutations of the indices and P itself
is the order of the permutation (i.e. it is even or odd, depending on whether
an even or odd number of pairwise exchanges had to be done to achieve the
permutation). example, it holds
(2)
G0 (i1 , i2 ; j1 j2 ) = G0 (i1 ; j1 ) G0 (i2 ; j2 ) − ηG0 (i1 ; j2 ) G0 (i2 ; j1 ) . (314)

8 Diagrammatic expansion of the partition func-

tion
We established already that the partition function can be written as

Z = Z0 hSi0

with ´β
dτ 00 Ṽ (τ 00 )
S ≡ S (β, 0) = T e− 0 . (315)
Thus, we obtain
∞ n ˆ
Z X (−1)
= dτ1 · · · dτn hT V (τ1 ) · · · V (τn )i0 . (316)
Z0 n=0
n!

Let us specify the interaction as

1 X
V (τ ) = v (k, l; n, m) c†k (τ ) c†l (τ ) cm (τ ) cn (τ ) (317)
2
klmn

51
The n-th term in the perturbation expansion is then
n ˆ
(−1) X X
δn = n
dτ1 · · · dτn ··· v (k1 , l1 ; m1 , n1 ) . . . v (kn , ln ; mn , nn )
n!2
k1 l1 m1 n1 kn ln mn nn
D E
× T c†k1 (τ1 ) c†l1 (τ1 ) cm1 (τ1 ) cn1 (τ1 ) · · · c†kn (τn ) c†ln (τn ) cmn (τn ) cnn (τn ) (318)
0

The time ordered expectation value can now be evaluated using Wick’s theorem
and be expressed in terms of the known non-interacting Green’s function.

8.1 Ring diagrams

The partition function inthe ring approximation is:
ˆ
Z 1 V X
≈1− 3
d3 q log (1 − v(q)Π(q, Ωn )) .
Z0 2 (2π) n

Let us now take a closer look at the object Π(q, Ωn ) inside the logarithm. We will
first focus on the Matsubara summation, therefore we suppress all momentum
dependences. It will be straightforward to restore these later. Then Π(Ωm ) is
given by the expression
X
Π(Ωm ) = T G1 (ωn )G2 (ωn + Ωm ),
n

where the indices 1, 2 refer to the suppressed variables. As was derived before,
the Matsubara Green’s functions G(ωn ) are equal to the complex Green’s func-
tions G(z) evaluated at z = iωn . We use this and rewrite the sum as a contour
integral. This provides us with the formula
X
Π(Ωm ) = T G1 (iωn )G2 (iωn + iΩm )
n
˛
dz
= − f (z)G1 (z)G2 (z + iΩm ), (319)
2πi

where f (z) = eβz1+1 is the Fermi function. The contour integral is taken as
shown in the figure. We now want to deform this contour and transform it
into a real integration. In doing this we have to be careful, since the complex
Green’s function has poles on the real-axis. This in turn means that we have
to avoid the real axis, since our integrand contains G1 (z). On the other hand,
the integrand also contains G2 (z + iΩm ), thus we must avoid the horizontal line
that goes through z = −iΩm also. We therefore choose a contour as shown in
the figure. Clearly this contour gives the same result as the previous one, since
we enclose the same poles as before.
As usual, the semicircular arcs do not contribute to the integrals. Therefore only
the four horizontal lines have to be taken into account. Let us first consider the
two horizontal paths at the bottom. These two paths run in opposite directions

52
and are infinitesimally displaced relative to each other. We parametrize the two
paths by z =  + i0+ and z =  − i0+ respectively. Since f (z) and G2 (z) do
not have poles near these paths we can ignore the infinitesimal shifts in these
functions. Then the sum of the two integrals yields for the contour Cl , along
the two lower paths the value

˛ ˆ∞
dz d
f ()G2 (+iΩm ) G1 ( + i0+ ) − G1 ( − i0+ ) .
 
− f (z)G1 (z)G2 (z+iΩm ) = −
2πi 2πi
Cl −∞

Similarly, the two paths on the top are parametrized by z = −iΩm + i0+ +  and
z = −iΩm − i0+ + . Here f (z) and G1 (z) are analytic near these two paths,
thus we can ignore the infinitesimal shifts for these functions. Thus we have

˛ ˆ∞
dz d
f (−iΩm )G1 (−iΩm ) G2 ( + i0+ ) − G2 ( − i0+ ) .
 
− f (z)G1 (z)G2 (z+iΩm ) = −
2πi 2πi
Cu −∞

Thus the total contribution to Π(ΩM ) is given by

ˆ∞
d
f ()G2 ( + iΩm ) G1 ( + i0+ ) − G1 ( − i0+ )
 
Π(Ωm ) = −
2πi
−∞
ˆ∞
d
f ()G1 ( − iΩm ) G2 ( + i0+ ) − G2 ( − i0+ ) ,
 
−
2πi
−∞

1
where in the second integrand we used the fact that f ( − iΩm ) = eβ(−iΩ m ) +1
=
1 + + +
eβ +1
= f (). Now we recall that G1 (+i0 )−G1 (−i0 ) = 2iIm[G1 (+i0 )] =
2iG00r1 (), i.e. the imaginary part of the retarded Green’s function and this holds
analogously for G2 . Then we are left with

ˆ∞ ˆ∞
d d
Π(Ωm ) = − f ()G2 ( + iΩm )G00r1 () − f ()G1 ( − iΩm )G00r2 ().
π π
−∞ −∞

Finally we go over to the retarded Πr (Ω) by making the substitution

iΩm = Ω + i0+ .

This gives us

ˆ∞ ˆ∞
d d
Πr (Ω) = − f ()Gr2 ( + Ω)G00r1 () − f ()G∗r1 ( − Ω)G00r2 ().
π π
−∞ −∞

53
Let us now look first at the imaginary part of this expression for Πr (Ω). This
gives us
ˆ ˆ∞
 ∞ 
d d 00
Im[Πr (Ω)] = −  f ()G00r2 ( + Ω)G00r1 () − f ()Gr1 ( − Ω)G00r2 ()
π π
−∞ −∞
ˆ ˆ∞
 ∞ 
d d 00
= − f ()G00r2 ( + Ω)G00r1 () − f ( + Ω)Gr1 ()G00r2 ( + Ω)
π π
−∞ −∞
ˆ∞
d
= − (f () − f ( + Ω)) G00r2 ( + Ω)G00r1 ().
π
−∞

We notice the following remarkable fact. If we are at low temperatures, the

Fermi function is nearly a step function and f () − f ( + Ω) will only be finite, if
 and +Ω have different signs, in other words, when one excitation corresponds
to a particle and the other to a hole. Intuitively, such particle-hole excitations
correspond to the diagram shown.
Let us now insert the forms of the Green’s functions and reintroduce the
momentum summations. The complex bare Green’s functions are given by
1
Gr1 (ω) =
ω + i0+ − k
1
Gr2 (ω) = +
.
ω + i0 − k+q
We need the imaginary parts of these:
G00r1 (ω) = −πδ(ω − k )
G00r2 (ω) = −πδ(ω − k+q )
Inserting this into the formula for Πr (q, Ω) we obtain
ˆ∞ ˆ∞
d d
Πr (q, Ω) = − f ()Gr2 ( + Ω)G00r1 () − f ()G∗r1 ( − Ω)G00r2 ()
π π
−∞ −∞

X ˆ d
∞
1
= − f () [−πδ( − k )]
π  + Ω + i0+ − k+q
k −∞

ˆ∞
X d 1
− f () [−πδ( − k+q )]
π  − Ω − i0+ − k
k −∞
X f (k ) X f (k+q )
= +
+
k + Ω + i0 − k+q k+q − Ω − i0+ − k
k k
X f (k ) − f (k+q )
= .
k − k+q + Ω + i0+
k

54
This is the Lindhard function that we already found via the equation-of-motion
technique.
Let us now return to the evaluation of partition function. There the correc-
tion due to the interaction term had the form
X
T log [1 − v(q)Π(q, Ωn )] .
n

Recall that Ωn is now a bosonic Matsubara frequency, i.e. Ωn = πβ 2n. In

converting this sum to a contour integral we must be careful about the pole
at Ω0 of the Bose function. Let us therefore carry out the contour integral as
shown in the figure. The contour does not include Ω0 and instead we add it by
hand:
˛
X dz
T log [1 − v(q)Π(q, Ωn )] = nB (z) log (1 − v(q)Π(q, z))+T log (1 − v(q)Π(q, 0)) ,
n
2πi
C
1
where nB (z) = eβz −1
.
Converting the contour integral again to real integrals,
we have to be careful this time about the Bose function, since it has a pole at
z = 0:
˛ ˆ
+∞
dz d
nB ( + i0+ ) log 1 − v(q)Π(q,  + i0+ )


nB (z) log (1 − v(q)Π(q, z)) =
2πi 2πi
C −∞
ˆ
+∞
d
nB ( − i0+ ) log 1 − v(q)Π(q,  − i0+ ) .


−
2πi
−∞

Let us now evaluate the Bose function near z = 0, i.e.  ≈ 0:

1 T
nB ( ± i0+ ) = +
≈
exp(β( ± i0 )) − 1  ± i0+
Since Π(q,  ± i0+ ) has no poles near  = 0, the only difficulty in the integrals
comes from the difference of the Bose functions, i.e. the difference of the of
nB ( ± i0+ ) :
nB ( + i0+ ) − nB ( − i0+ ) = −2πT iδ()
Away from  = 0 the Bose function is regular and we can replace nB ( ± i0+ )
by nB (). This gives
˛ ˆ
+∞
dz d
nB () log 1 − v(q)Π(q,  + i0+ ) − log 1 − v(q)Π(q,  − i0+ )



nB (z) log (1 − v(q)Π(q, z)) =
2πi 2πi
C −∞
−T log (1 − v(q)Π(q, 0))
but the last term exactly cancels the zeroth Matsubara term that we added by
hand. Thus we are left with the simple result:
ˆ
+∞
X d
nB ()Im log 1 − v(q)Π(q,  + i0+ )
 
T log [1 − v(q)Π(q, Ωn )] =
n
π
−∞

55
9 Diagram rules for the single particle Green’s
function
Next we perform the diagramatic analysis of the single particle Green’s function

D E
†
T ψ̃kα (τ ) ψ̃kα (0) S
0
Gk (τ ) = − . (320)
hSi0

We proceed in full analogy to the analysis of the partition function. When

we perform contractions, we also have an external time point τ and external
momenta k and spin variable α where no integration is needed.

10 Fermi liquid theory

The key concept underlying Fermi liquid theory is adiabacity, i.e. the assump-
tion that the low energy excitations of an interacting Fermi system are in one-
to-one correspondence to the excitations of a non-interacting Fermi gas. The
theory was originally developed for 3 He, which at low temperatures is a struc-
tureless fermion due to the net spin 1/2 in the nucleus. The proton charge is
compensated by the two electrons that form a singlet state and therefore don’t
contribute to the total spin of the system. One simplifying aspect of 3 He, if com-
pared to electrons in solids, is the absence of an underlying crystalline lattice.
The bare dispersion is then given in form of the free particle dispersion

~k 2
εfree
k = . (321)
2m
The quantum numbers of the excited many-body states of a free Fermi gas are
the occupations nkσ of single-particle states; the corresponding single-particle
states are characterized by momentum and spin: |kσi.

10.1 quasi-particle excitations

In the ground state we continue to assume that the system is characterized by
a filled Fermi sea with
(0)
nkσ = θ (kF − k) (322)

where kF is the same as for an ideal Fermi gas with same density:
ˆ
N d3 k (0)
= 2 3 nkσ
V (2π)
ˆ kF
1 2 kF3
= k dk = (323)
π2 0 3π 2

56

1/3
which yields kF = 3π 2 N/V . Excitations are now characterized by changes
δnkσ of the occupations, i.e.
(0)
nkσ = nkσ + δnkσ . (324)

The corresponding change in energy is then given by

X
δE = εkσ δnkσ . (325)
k,σ

Since the labeling of quantum numbers is the same compared to the free fermi
system, purely statistical aspects, like the entropy, should also only be deter-
mined by the corresponding ideal Fermi gas expressions:
X
S = −kB (nkσ log nkσ + (1 − nkσ ) log (1 − nkσ )) . (326)
k,σ
P
Maximizing
P this expression with the condition that E = k,σ εkσ nkσ and N =
k,σ n kσ , yields
1
nkσ = βε , (327)
e kσ + 1
where the excitation energy εkσ is measured relative to the Fermi energy.
In general, the energy εkσ [nk0 σ0 ] is a functional of the occupations of all
states in the system. If we first add only one particle to the ground state, we
can safely say that
ε0kσ ≡ εkσ n0k0 σ0
 
(328)
is the dispersion of the single particle, where n0kσ = θ (kF − k). Near EF we
make the assumption
ε0kσ = v (k − kF ) , (329)
where the parameter v = kF /m∗ is often expressed in terms of the effective
mass m∗ . This immediately leads to the density of states

m∗ kF
ρF = , (330)
π 2 ~2
that is modified by the factor m∗ /m compared to the free fermion expression.
An immediate consequence of this modified density of states emerges for the
heat capacity
ˆ ∞
X d d 1
C = εkσ nkσ = ρF dεε
dT −∞ dT eε/(kB T ) + 1
k,σ
= γT, (331)

where
π 2 kB
2
m∗
γ= ρF = γfree . (332)
3 m

57
Here, γfree is the heat capacity of non-interacting fermions.
A key additional aspect of the Landau theory is that in case of excitations
with more than one particle, the changes in the occupations δnkσ will lead to
changes δεkσ of the quasi-particle energies. Thus one writes generally

εkσ = ε0kσ + δεkσ (333)

where
1 X
δεkσ = fkσ,k0 σ0 δnk0 σ0 . (334)
N 0 0
k ,σ

fkσ,k0 σ0 is a phenomenological interaction parameter that determines the extend

to which the energy of state |k,σi is affected by a change in population δnk0 σ0
of |k0 ,σ 0 i. If we do not want to have a preference of one spin direction over the
other (in the absence of an external magnetic field) we write

s 0 a
fkσ,k0 σ0 = fk,k 0 + σσ fk,k0 . (335)

In addition we assume that all relevant momenta are on the Fermi surface, i.e.
k =kF ek and k0 =kF ek0 where e2k = e2k0 = 1 are unit vectors. In an isotropic
s,a
system like 3 He, with no preferred direction, one expects that fk,k 0 only depend
0
on the angle θ between k and k , i.e. on cos θk,k0 = ek · ek0 :

s,a s,a


fk,k 0 = f cos θk,k0 . (336)

Since f s,a are of dimension energy, dimensionless quantities follow via

F s,a = ρF f s,a . (337)

We expand F s,a cos θk,k0 in Legendre polynomials

∞
X
s,a
Fls,a Pl cos θk,k0



F cos θ k,k0 = (338)
l=0

and use the usual representation in terms of spherical harmonics

∞ X
l
X Fls,a ∗
F s,a cos θk,k0 = 4π


Ylm (ek ) Ylm (ek0 ) . (339)
2l + 1
l=0 m=−l

The orthogonality of the Legendre polynomials

ˆ 1
1 δl,l0
d cos θPl (cos θ) Pl0 (cos θ) = (340)
2 −1 2l + 1

58
allows for the representation
ˆ
s,a 2l + 1 1
Fl = d cos θF s,a (cos θ) Pl0 (cos θ)
2 −1
ˆ ˆ π
2l + 1 2π
= dϕ sin θdθF s,a (cos θ) Pl0 (cos θ)
4π 0 0
ˆ ˆ π ˆ
2l + 1 2π
= dϕ sin θdθ dεF s,a (cos θ) Pl0 (cos θ) δ (ε)
4π 0 0
2 (2l + 1) 1 X s,a

= Fk,k0 Pl cos θk,k0 δ (εk ) . (341)
ρF N 0
k

10.2 Susceptibilities
Suppose we are at T = 0 we add an external perturbation of the type

δε0kσ = (vls + σvla ) Ylm (ek ) , (342)

where the spherical harmonic determines the directional dependence on the

momentum and vls,a are the amplitudes of the perturbation. In case of an
external magnetic field holds δε0kσ = −σµB B, i.e. we have m = l = 0 and, due
to Y00 = 4π, v0a = −σµB B/4π and v0s = 0. A change in the chemical potential
δµ amounts to δε0kσ = −δµ, i.e. m = l = 0 and v0s = −δµ/4π and v0a = 0. In
general we can introduce the susceptibility
∂2E
χs,a
l =− 2. (343)
∂ (vls,a )
In particular, this allows us to determine physical observables like the magnetic
susceptibility 
1 ∂M 
χs = (344)
V ∂B B=0
P
with magnetization M = µB k,σ σnkσ . Another option is the charge suscep-
tibility 
1 ∂N 
χc = (345)
V ∂ (δµ) δµ=0
P
with particle number N = k,σ nkσ . χc is closely related to the compressibility

1 ∂V ∂n
κ=− = n−2 .
V ∂p ∂µ
which holds for a system where the free energy can be written as F (V, N ) =
N f (n), where n = N/V is the particle density.
In case of a free Fermi gas follows

χ0s = µ2B ρ0F

χc = ρ0F . (346)

59
where the mass in the density of states ρ0F is m, not m∗ .
Since fermions of a Fermi liquid are interacting there is no reason that an
external field or chemical potential change will change the quasiparticle energies
εkσ in the exact same fashion as δε0kσ . Thus, we assume
1 X
δεkσ = δε0kσ + fkσ,k0 σ0 δnk0 σ0
N 0 0
k ,σ
= (tsl + a
σtl ) Ylm (ek ) , (347)

where in general ts,a

l 6= vls,a . Suppose there is such an energy shift, then we can
determine the associated particle density shift from Eq.327
1
nkσ =
β (ε0kσ +δεkσ )
e +1
= θ (kF − k) − δ ε0kσ δεkσ ,


(348)

which yields
δnkσ = −δ ε0kσ δεkσ .


(349)
This result can now be inserted into Eq.347 which yields
1 X
δεkσ = δε0kσ − fkσ,k0 σ0 δ ε0k0 σ0 δεk0 σ0


(350)
N 0 0
k ,σ

or equivalently
1 X
0
tσl Ylm (ek ) = vlσ Ylm (ek ) − fkσ,k0 σ0 δ ε0k0 σ0 tσl Ylm (ek0 ) (351)
N 0 0
k ,σ

Now we can use the above expansion of fkσ,k0 σ0 in spherical harmonics

ˆ
ρF X 0
tσl Ylm (ek ) = vlσ Ylm (ek ) − dΩk0 fkσ,k0 σ0 tσl Ylm (ek0 )
2
σ0
ˆ
ρ F dΩ k0 X 0
= vlσ Ylm (ek ) − fkσ,k0 σ0 tσl Ylm (ek0 )
2 4π
σ0
ˆ ∞ l0
σ 1 XX X Fls0 + σσ 0 Fla0
= vl Ylm (ek ) − dΩk0
2 0 0 0 0
2l + 1
σ l =0 m =−l
0
×Yl0 m0 (ek ) Yl∗0 m0 (ek0 ) tlσ Ylm (ek0 ) (352)

We use the orthogonality of the spherical harmonics

ˆ
dΩk0 Yl∗0 m0 (ek0 ) Ylm (ek0 ) = δll0 δmm0 (353)

and obtain
1 X σ0 Fls + σσ 0 Fla
tσl = vlσ − t . (354)
2 0 l 2l + 1
σ

60
If we consider for example a change in the chemical potential with, v0σ =
−δµ/4π, it follows for tσ0 = t0 spin independent, that
t0 = v0 − t0 F0s (355)
which leads to
v0
t0 = . (356)
1 + F0s
It is now straightforward to determine the charge susceptibility via
1 ∂N 1 1 ∂N
χc = =−
V ∂ (δµ) 4π V ∂v0
1 1 ∂N ∂t0
= −
4π V ∂t0 ∂v0
1 1 1 ∂N
= − . (357)
1 + F0s 4π V ∂t0
−1 1 ∂N
To determine 4π V ∂t0 we use
1
nkσ = 
kF

β m∗
(k−kF )+4πt0
e +1
P
The derivative of N = k,σ nkσ with respect to t0 can be performed, e.g. by
resorting to our above result for the charge susceptibility of a free electron gas
(with the difference that we need to consider the effective mass, not the bare
mass). It follows −1 1 ∂N
4π V ∂t0 = ρF and we obtain
ρF
χc =
1 + F0s
Thus, the charge susceptibility is different from the free fermi gas value in two
ways. First, in the density of states the mass m is replaced by the effective
−1
mass m∗ . In addition the interactions lead to an overall coefficient (1 + F0s ) .
s s
The theory is therefore stable as long as F0 < −1. If F0 → −1 the system will
undergo a transition to a regime where different densities phase segregate.
As will be discussed in a homework assignment, if the system in Gallilei
invariant, one can further derive a relationship between the effective mass m∗
and the Landay parameter F1s :
m∗ 1
= 1 + F1s . (358)
m 3

11 Conservation laws and Ward-Identities

Next we will address the implication of conservation laws within the many body
description of condensed matter systems. Within quantum mechanics, conser-
vation laws are generally associated with an operator O that commutes with
the Hamiltonian
[O, H] = 0. (359)

61
Whenever we have to deal with a conserved quantity, it is often natural to
introduce an associated density ρ (r, t), such that
ˆ
O= dd rρ (r, t) . (360)
V

The density then obeys a continuity equation

∂t ρ + ∇ · j = 0, (361)

which determines the associated current density in the long wave-length limit.
Integrating this equation over a volume V yields the Gauss’-theorem
ˆ
∂t O = − j · dS, (362)
∂V

where ∂V is the surface of V and dS refers to the surface element directed along
the outward normal of the surface. If the volume V is chosen to correspond to
the entire system, the current of a conserved quantity obviously has to vanish.
Let us consider a system with Hamiltonian

H = H0 + V, (363)

where
†
X
H0 = εk ψkα ψkα . (364)
kα

A natural conservation law is associated with charge or particle conservation,

i.e. O = N . In non-relativistic systems, the conservation of the total spin, i.e.
O = S is another example. One must however keep in mind that the weak
relativistic spin-orbit interaction will violate spin-conservation. Let us consider
the charge conservation. The associated density is obviously
X
ρ (r, t) = ψα† (r, t) ψα (r, t) . (365)
α

In momentum space this coresponds to

1 X †
ρq = ψk+qα ψkα . (366)
N
kα

In order to keep our argumentation sufficiently general, we will consider densities

of the form X †
ρq = ψk+ q ,α φαβ (k) ψk− q2 ,β , (367)
2
kαβ

with form factor φαβ (k). The charge density corresponds to φαβ (k) = δαβ , and
j
a spin density amounts to the φαβ (k) = σαβ for the three Pauli matrices.
In what follows we consider systems where [ρq , V ] = 0. This is the case for
the charge density with electron-electron Coulomb interaction. In case of the

62
spin-density it is true for all systems without spin-orbit interaction. In order
to obtain the associated current, we evaluate the commutator of ρq with the
Hamiltonian, where we only need to focus on the commutator with the kinetic
energy: X 
†
[H0 , ρq ] = εk+ q2 − εk− q2 ψk+ q φαβ (k) ψk− q α .
α 2
(368)
2
kαβ

This expression can be expanded for small momenta q and yields

X ∂εk †
∂t ρq = i [H0 , ρq ] = iq · ψ φαβ (k) ψkβ . (369)
∂k kα
kαβ

Fourier transformation of the continuity-equation yields ∂t ρq = iq · j such that

X ∂εk †
j= ψ φαβ (k) ψkβ . (370)
∂k kα
kαb

Next we want to explore the implications of conservation laws for associated

Green’s functions. To this end we recall that a density-density susceptibility is
expressed as
1 X D † E
χc (q, τ ) = 2 T ψkα (τ ) ψk+qα (τ ) ψk† 0 β ψk0 −qβ .
N 0 kk αβ

Since we consider generalized densities with form factor φαβ (k), it is useful to
analyze:
ˆ D E
†
αβγδ
Qk,k0 ,q (Ω) = dτ eiΩτ T ψk+ q
α
(τ ) ψk− q2 β (τ ) ψk† 0 − q γ (0) ψk0 + q2 δ (0) . (371)
2 2

Just like for the retarded function we can exploit the equation of motion
D E D E
∂τ T ρq (τ ) ψk† 0 − q γ (τ 0 ) ψk0 + q2 δ (τ 00 ) = T [H (τ ) , ρq (τ )] ψk† 0 − q γ (τ 0 ) ψk0 + q2 δ (τ 00 )
2 2
D h i E
+ δ (τ − τ ) T ρq (τ ) , ψk† 0 − q γ (τ 0 ) ψk0 + q2 δ (τ 00 )
0

D h2 iE
†
00
+ δ (τ − τ ) T ψk0 − q γ (τ ) ρq (τ ) , ψk0 + q2 δ (τ 00 ) .
0
2

For the commutators follows

h i
ρq , ψk† 0 − q γ ψk† 0 + q ,α φαγ (k0 ) ,
X
=
2 2
α
h i X
ρq , ψk0 + q2 δ = − φδβ (k0 ) ψk0 − q2 ,β . (372)
β

We can now insert the density and current and Fourier transform the resulting
correlator
ˆ
00
Gαβγδ
k,k0 ,q (Ω, ω) = dτ dτ 0 eiΩτ +ωτ
D E
†
× T ψk+ q
α
(τ ) ψk− q2 β (τ ) ψk† 0 − q γ (0) ψk0 + q2 δ (τ 00 )
2 2

63
which yields in particular for Qαβγδ
k,k0 ,q (Ω), needed in the determination of sus-
ceptibilities, that X αβγδ
Qαβγδ
k,k0 ,q (Ω) = T Gk,k0 ,q (Ω, ω) . (373)
ω

The above equation of motion now leads to the Ward-identity:

X    
iΩ − εk+ q2 − εk− q2 φαβ (k) Gαβγδ
0
k,k ,q (Ω, ω) = φ δγ (k0
) G k0 + q δ (ω) − Gk0 − q γ (ω + Ω)
2 2
,
kαβ

with usual single particle Green’s function

ˆ D E
Gkα (ω) = − dτ eiωτ T ψk0 + q2 δ (τ ) ψk† 0 + q ,α (0) .
2

The above Ward identity allows to draw conclusions for the two susceptibilities:
(4)αβγδ
X X
χρ (q, Ω) = T φαβ (k) φγδ (k0 ) Gk,k0 ,q (Ω, ω)
k,k0 ,ω αβγδ
X X ∂εk γδ 0 ∂εk (4)αβγδ
χij
J (q, Ω) = T φαβ (k) φ (k ) G 0 (Ω, ω) .
∂ki ∂kj k,k ,q
k,k0 ,ω αβγδ

In case of the susceptibility χρ of the associated conserved density, it follows

from the Ward identity that
X Gk0 (ω) − Gk0 (ω + Ω)
χρ (q = 0, Ω 6= 0) = T φγδ (k0 )
iΩ
k0 ,ωγδ
= 0. (374)

This identity can be used to check whether a given approximation used to

analyze a many body problem respects the corresponding conservation law.
Notice, that this is precisely the behavior we found earlier for the Lindhard
function. Here it is a behavior caused by charge conservation.
In addition we can draw conclusions about the susceptibility of the associated
current-current susceptibility.
 X
2 ∂Gk0 (ω) ∂εk
χij (q, 0) = −T φγδ (k0 ) .

J
∂ki0 ∂kj0

q=qei →0 0 k ,ωγδ

12 Physics of Graphene
13 Superconductivity
The microscopic theory of superconductivity was formulated by John Bardeen,
Leon N. Cooper, and J. Robert Schrieffer. It is among the most beautiful and
successful theories in physics. The BCS-theory starts from an effective Hamil-
tonian of fermionic quasiparticle excitations that interact via a weak attractive

64
interaction. It yields a ground-state many-body wave function and thermal
excitations to describe superconductivity. Historically, the first underlying mi-
croscopic mechanism that lead to such an attraction was the exchange of lattice
vibrations. In the meantime ample evidence exists, in particular in case of the
copper-oxide high-temperature superconductors, for superconductivity that is
caused at least predominantly by electronic interactions. Other materials that
are candidates for electronically induced pairing are the heavy electron systems,
organic charge transfer salts and the iron based superconductors.

13.1 off-diagonal long range order

The initial observation of superconductivity was made by measuring the resis-
tivity ρ (T ) of mercury as function of temperature. Below the superconducting
transition temperature, Tc , ρ (T ) = 0 with very high precision. Understanding
this drop in the resistivity is a major challenge in a theory of superconductivity.
We will return to this problem later. Arguably even more fundamental than
the vanishing voltage drop are two central experiments: the Meissner effect and
the quantization of the magnetic flux in multiply connected superconductors.
The Meissner effect implies that a weak magnetic field is expelled from the bulk
of a superconductor. The effect occurs regardless of whether the external field
is switched on for temperatures below Tc or before the system is cooled down
to enter the superconducting state. This strongly supports the view that a su-
perconductor is in thermal equilibrium. Multiply connected superconducting
geometries such as a ring, can however lead to a subtle memory effects. Switch-
ing off an external magnetic field for T < Tc leads to magnetic flux trapped in
non-superconducting holes. This flux takes values that are integer multiples of
the flux quantum

h
Φ0 = ≈ 2.067833758(46) × 10−15 Tm2 , (375)
2e
where h is Planck’s constant and e the magnitude of the electron charge e. By
discussing in some detail the concept of off-diagonal long range order we give
precise microscopic criteria that lead to the Meissner effect and to flux quanti-
zation. A theory of superconductivity consistent with these criteria is therefore
guaranteed to correctly describe these fundamental experimental observations.
As we will see later, the BCS theory is such a theory.
Off-diagonal long-range order (ODLRO) is a natural generalization of the
Bose-Einstein condensation of free bosons to the regime of interacting systems.
It was introduced to capture the nontrivial physics of superfluid 4 He[?, ?] and
later generalized to describe superconductivity and superfluidity of fermions[?].
The formal definition is based on the single-particle and two-particle density
matrix ρ(1) and ρ(2) , respectively:
(1)
ραα0 (r, r0 ) ψα† (r) ψα0 (r0 ) ,


=
D E
(2)
ραβα0 β 0 (r1 , r2 , r01 , r02 ) = ψα† (r1 ) ψβ† (r2 ) ψβ 0 (r02 ) ψα0 (r01 ) . (376)

65
ψα† (r) and ψα (r) are the creation and annihilation operators of a boson or
fermion at position r and with spin α, respectively. The operators are in the
Schrödinger picture such that the ρ(n) are independent on time in thermal equi-
librium. Generalizations to an n-particle density matrix ρ(n) with n > 2 or
averages with respect to a non-equilibrium scenario are straightforward. Before
we define ODLRO, we summarize a few properties of these density matrices.
We obtain immediately the expected normalization
ˆ X (1)
trρ(1) = dd r ραα0 (r, r) = N (377)
α

as well as
ˆ
(2)
X
trρ(2) = dd r1 dd r2 ραβαβ (r1 , r2 , r1 , r2 ) = N (N − 1) . (378)
αβ

We first concentrate on spin-less bosons in a translation invariant system

and analyze ρ(1) . It is a hermitian matrix with respect to the matrix indices r
and r0 . If np is the p-th real eigenvalue of ρ(1) with eigenvector φp (r), we can
expand6 X
ρ(1) (r, r0 ) = np φ∗p (r0 ) φp (r) . (379)
p

A macroscopic occupation of a single-particle state occurs if there exists one

eigenvalue, say n0 , that is of order of the particle number N of the system. This
is a natural generalization of Bose-Einstein condensation to interacting systems.
Off-diagonal long range order occurs if for large distances |r − r0 | the expansion,
Eq.379, is dominated by a single term (the one with the macroscopic eigenvalue
n0 and eigenfunction φ0 (r)). The condition for ODLRO is therefore

ρ(1) (r, r0 ) → n0 φ∗0 (r0 ) φ0 (r) . (380)

|r−r0 |→∞

For a translation invariant system further holds that ρ(1) (r, r0 ) = ρ(1) (r − r0 ),
i.e. the quantum number p corresponds to the momentum vector p. In the
6 Consider a hermitian matrix A with eigenvectors x(n) and eigenvalues λ(n) , i.e.
X
Aij x(n)j = λ(n) x(n)i .
j

We can consider the matrix Aij for given j as vector with components labelled by i and
expand with respect to the complete set of eigenvectors. The same can be done for the other
index. This implies X
Aij = α(p,q) x(p)i x∗(q)j .
pq
Inserting this expansion intoP
the eigenvalue equation and using the orthogonality and normal-
ization of the eigenvectors ( j x∗(p)j x(q)j = δpq ) it follows
X X
Aij x(n)j = α(p,n) x(n)i .
j p

Since the x(n) are eigenvectors it follows α(p,n) = λ(n) δp,n .

66
thermodynamic limit holds that limr→∞ ρ(1) (r) = αN/V with α a generally
complex coefficient where |α| is of order unity. Here V is the volume of the
system and we used φ0 ≈ √1V .
We first consider the case of free bosons where φp (r) = √1V eip·r and the
eigenvalues are given by the Bose-Einstein distribution function:

1
np = . (381)
eβ((p)−µ) −1
We consider the regime above the Bose-Einstein condensation temperature with
 −2/3 2
3 ~ 2/3
kB TBEC = 2πς (N/V ) , (382)
2 m

where µ < 0 and the occupation of all single-particle states behaves in the
thermodynamic limit as limN →∞ np /N = 0. np decays for increasing momenta
exponentially on the scale 2π/λT with thermal de Broglie wave length
s
2π~2
λT = . (383)
kB T m

It follows ˆ
1 X d3 p
ρ(1) (r) = np eip·r = 3 np e
ip·r
(384)
V p (2π)

decays exponentially like e−r/λT , implying no ODLRO. On the other hand, in

case of a macroscopic occupation n0 = αN of the lowest energy state, i.e. for
p = 0, below TBEC follows

N 1 X
ρ(1) (r) = α + np eip·r (385)
V V p>0

The second term decays exponentially, with the same reasoning as for T > TBEC
while the first term gives rise to ODLRO. Our reasoning is in fact more general.
In case of a macroscopic occupation of a momentum state, i.e. np0 = α0 N holds

N ip0 ·r
lim ρ(1) (r) = α0 e (386)
r→∞ V
as long as the occupation of all other momentum states decays sufficiently fast
for large momenta, they will not contribute in the limit of large r. Thus, we
have established that the macroscopic occupation of states is rather generally
related to large distant correlations of the one particle density matrix.
Next we discuss some physical implications of this observation and demon-
strate that charged bosons with ODLRO are subject to the Meissner effect and
flux quantization. The discussion is adapted from Refs.[?, ?] where fermionic

67
systems were discussed. We start from the Hamiltonian of a system of bosons
in a uniform magnetic field B:
X } ∇j + e A (rj ) 2 X


i c
H= + V (ri − rj ) . (387)
j
2m
i6=j

The vector potential can be written as

A (r) = A0 (r) + ∇ϕ (r) , (388)
where A0 (r) = 21 B × r and ϕ is an arbitrary function. The many-body wave
function of the problem is Ψν (r1 , · · · , rN ) = Ψν (rj ).
Let us perform a spatial displacement rj → rj − a with some length scale
a. The boson-boson interaction is invariant with respect to this transformation,
while the vector potential transforms as
A (r) → A (r − a)
1
= A (r) − B × a + ∇ (ϕ (r − a) − ϕ (r))
2
= A (r) + ∇χa (r) , (389)
with
χa (r) = a · A0 (r) + ϕ (r − a) − ϕ (r) .
The displacement can be understood as a gauge transformation. Thus, we can
write the Schrödinger equation as it emerges after the transformation:
 
X } ∇j + e A (rj − a) 2 X


i c
 + V (ri − rj ) Ψν (rj − a) = Eν Ψν (rj − a)
j
2m
i6=j

alternatively as
 
X } ∇j + e A (rj ) 2 X


e
P
 i c
+ V (ri − rj ) ei ~c j χa (rj ) Ψν (rj − a)
j
2m
i6=j
e
P
= Eν ei ~c j χa (rj )
Ψν (rj − a) ,
with χa (r) given above. In addition to the many-body wave functions Ψν (rj )
we have the alternative choice
e
P
Ψ0ν (rj ) = ei ~c j χa (rj )
Ψν (rj − a) . (390)
The density matrix can therefore we evaluated using the original or the primed
wave functions. For the density matrix expressed in terms of the primed wave
functions follows
ˆ X e−βEν
0
ρ(1) (r, r0 ) = e−i ~c (χa (r)−χa (r )) N dd r2 · · · dd rN
e

ν
Z
× Ψ∗ν (r − a, r2 − a, · · · , rN − a)
× Ψν (r0 − a, r2 − a, · · · , rN − a) . (391)

68
All other phase factors ∝ χa (rj ) for j = 2 · · · N cancel. Using periodic boundary
conditions we can shift the integration variables rj → rj − a and obtain
0
ρ(1) (r, r0 ) = e−i ~c (χa (r)−χa (r )) ρ(1) (r − a, r0 − a) .
e
(392)

Let us now assume ODLRO, i.e. for large distance between r and r0 holds
Eq.380. This implies
0
φ∗0 (r0 ) φ0 (r) = e−i ~c (χa (r)−χa (r )) φ∗0 (r0 − a) φ0 (r − a)
e
(393)

which implies for the eigenfunction of the density operator

e
φ0 (r) = fa e−i ~c χa (r) φ0 (r − a) , (394)

where fa is a phase factor that is r-independent but displacement dependent.

We now perform two successive transformations
e
φ0 (r) = fb e−i ~c χb (r) φ0 (r − b)
e e
= fa fb e−i ~c χb (r) e−i ~c χa (r−b) φ0 (r − a − b) (395)

Of course, we can also change the order of the displacements:

e e
φ0 (r) = fa fb e−i ~c χa (r) e−i ~c χb (r−a) φ0 (r − a − b) . (396)

Since the wave function is single valued, the two phase factors that relate the
two wave functions must be the same and we find the condition:
e e
e−i ~c (χb (r)+χa (r−b)) = e−i ~c (χa (r)+χb (r−a)) , (397)

It follows from the above definition of χa (r) that

1
χb (r) + χa (r − b) = χa+b (r) + B · (a × b) . (398)
2
Here, we used:

χb (r) = b · A0 (r) + ϕ (r − b) − ϕ (r)

χa (r − b) = a · A0 (r − b) + ϕ (r − a − b) − ϕ (r − b)
1
= a · A0 (r) − a · (B × b) + ϕ (r − a − b) − ϕ (r − b) .
2
The result
1
χa (r) + χb (r − a) = χa+b (r) − B · (a × b) (399)
2
follows immediately by switching a and b. Combining the two terms we obtain
1 1
χb (r) + χa (r − b) − χa (r) − χb (r − a) = B · (a × b) − B · (b × a)
2 2
= B · (a × b) , (400)

69
which is independent on the position r. Our condition for the above phases can
therefore be written as:
e
B · (a × b) = 2πn, (401)
~c
where n is an integer.
The displacement vectors a and b are arbitrary. Thus, we can continuously
vary the vectors a and b on the left hand side. On the other hand, since n is an
integer, we cannot continuously vary the right hand side. The only acceptable
uniform field is therefore
B = 0. (402)
This is the Meissner effect of charged bosons with ODLRO. A system with
ODLRO cannot support a uniform magnetic field.
This derivation of the Meissner effects makes very evident the importance
of macroscopic condensation. Without condensation, we could still perform a
similar analysis for the density operator and obtain the condition
~c
2πn = χb (r) + χa (r − b) − χa (r) − χb (r − a)
e
− (χb (r0 ) + χa (r0 − b) − χa (r0 ) − χb (r0 − a)) . (403)

Inserting our above expression for the sum of the phases the right hand side
gives a zero, i.e. we merely obtain the condition n = 0, without restriction on
B. In other words, as long as the density matrix is determined by a sum over
many eigenstates, no Meissner effect occurs. Only the condensation in one state
and a density matrix

ρ(1) (r, r0 ) → n0 φ∗0 (r0 ) φ0 (r) . (404)

for large |r − r0 | yields a vanishing B-field. We conclude, that macroscopic

condensation and Meissner effect are closely related.
The analysis of ODLRO in fermionic systems proceeds in close analogy to
the bosonic case discussed in the previous section[?]. It is, however, based upon
the two-particle density matrix
D E
ραβα0 β 0 (r1 , r2 , r01 , r02 ) = ψα† (r1 ) ψβ† (r2 ) ψβ 0 (r02 ) ψα0 (r01 ) .
(2)

We consider the combined index (r1 α, r2 β) that describes the two-particle ma-
trix. Expanding ρ(2) with respect to its eigenfunctions
(2)
X
ραβα0 β 0 (r1 , r2 , r01 , r02 ) = np φ∗p (r1 α, r2 β) φp (r01 α0 , r02 β 0 ) , (405)
p

with eigenvalues np . ODLRO is again a state where the largest eigenvalue n0 is

of the order of the particle number N . In this case holds
(2)
ραβα0 β 0 (r1 , r2 , r01 , r02 ) → n0 φ∗0 (r1 α, r2 β) φ0 (r01 α0 , r02 β 0 ) (406)

in the limit where |ri − r0i | → ∞ while |r1 − r2 | and |r01 − r02 | remain finite.

70
 From the antisymmetry of the fermionic wave function follows
(2) (2)
ραβα0 β 0 (r1 , r2 , r01 , r02 ) = −ρβαα0 β 0 (r2 , r1 , r01 , r02 )
(2)
= −ραββ 0 α0 (r1 , r2 , r02 , r01 ) . (407)

This implies for the eigenfunction

φ0 (r1 α, r2 β) = −φ0 (r2 β, r1 α) (408)

as expected for a genuine two particle wave function.

A displacement a can again be thought of as a gauge transformation[?, ?].
Thus, one can use either the wave functions Ψν (rj , γj ) (here γj stands for the
spin and other quantum numbers) or the alternative functions
e
P
Ψ0ν (rj , γj ) = ei ~c j χa (rj )
Ψν (rj − a, γj ) .

As our magnetic field is assumed to be homogeneous, the displacement will not

affect the coupling of the magnetic field to the spin. Expressing ρ(2) in terms of
both sets of wave functions, we find the relationship
0 0
= e−i ~c (χa (r1 )+χa (r2 )−χa (r1 )−χa (r2 ))
(2) e
ραβα0 β 0 (r1 , r2 , r01 , r02 )
(2)
× ραβα0 β 0 (r1 − a, r2 − a, r01 − a, r02 − a) . (409)

For the eigenfunction follows from Eq.409 that

e
φ0 (r1 α, r2 β) = fa e−i ~c (χa (r1 )+χa (r2 )) φ0 (r1 − aα, r2 − aβ) . (410)

This is the two particle generalization of our earlier result Eq.394 for bosons.
Meissner effect and flux quantization followed rather directly from this result.
The Meissner effect follows from two consecutive displacements in alternate
order:
e e
φ0 (r1 α, r2 β) = e−i ~c (χb (r1 )+χb (r2 )) e−i ~c (χa (r1 −b)+χa (r2 −b))
× fa fb φ0 (r1 − a − bα, r2 − a − bβ) (411)

and
e e
φ0 (r1 α, r2 β) = e−i ~c (χa (r1 )+χa (r2 )) e−i ~c (χb (r1 −a)+χb (r2 −a))
× fa fb φ0 (r1 − a − bα, r2 − a − bβ) , (412)

which requires that the two phase factors must be the same. We already found
that
χb (r) + χa (r − b) − χa (r) − χb (r − a) = B · (a × b) . (413)
The condition of identical phases then corresponds to
2e
B · (a × b) = 2πn. (414)
~c

71
The only difference to the case of single-particle ODLRO is the new factor 2 that
is a consequence of the two-particle ODLRO considered here. In case one were
le
to analyze ODLRO in ρ(l) one would have a coefficient ~c . The argumentation
which implied that a homogeneous magnetic field must vanish is now the same
as before: The left hand side of the above condition can be continuously varied
while the right hand side cannot and the only solution is:

B = 0. (415)

13.2 the superconducting order parameter

n case of ODLRO, we have

ρ(1) (r, r0 , t) = n0 (t) φ∗0 (r0 , t) φ0 (r, t) , (416)

|r−r0 |→∞

where we allowed for an explicit time dependence of the density matrix, that
exists in out-of-equilibrium situations. This suggest to introduce the quantity
p
Ψ (r, t) = n0 (t)φ0 (r, t) . (417)

A definition that immediately implies

ˆ
2
dd r |Ψ (r, t)| = n0 (t) , (418)

which follows from the normalization to unity of the eigenfunction φ0 (r, t). The
behavior of the eigenfunction φ0 under gauge transformations, suggests that
the function Ψ (r, t) behaves in many ways like a condensate wave function.
Frequently, the order parameter of a Bose condensate is also defined via the
expectation value of the field operator

Ψ (r, t) = hψ (r, t)i . (419)

Then, Bose condensation is associated with a spontaneous breaking of the global

U (1) symmetry ψ (r) → eiθ ψ (r). At first glance these two statements seem con-
tradictory. Ψ (r, t) was defined for a system with fixed particle number and, more
importantly, for a Hamiltonian with conserved particle number. Breaking the
global U (1) symmetry implies that the particle number conservation is spon-
taneously broken, which seems at first glance rather odd. Notice that merely
using a grand-canonical ensemble does not resolve the issue. Particle number
conservation implies that the density matrix is block-diagonal with respect to
the number of particles. In such a situation it must hold that hψ (r, t)i = 0 even
for a grand-canonical description. The two definitions of the order-parameter
can, however, be reconciled. This is done by explicitly breaking particle conser-
vation and adding a term
ˆ
Hη = − dd r η (r) ψ † (r) + η ∗ (r) ψ (r)


(420)

72
to the Hamiltonian and taking the limit η → 0 after the thermodynamic limit.
It turns out that hψ (r, t)i =
6 0 when the system establishes ODLRO. The density
matrix can be decomposed as

ρ(1) (r, r0 , t) = ψ † (r, t) hψ (r, t)i



†
ψ (r, t) − ψ † (r, t) (ψ (r0 , t) − hψ (r0 , t)i) , (421)



+

where the first term remains finite for large r − r0 , while the second one de-
cays. We will not demonstrate this here, but rather perform the corresponding
analysis when we discuss fermionic systems.
While the definition Ψ (r, t) in terms of the condensate eigenfunctions of the
density matrix is conceptually more satisfying, the usage of hψ (r, t)i is very
convenient in mean-field theories like the Bogoliubov theory of dilute or weakly
interacting condensed bosons.
In full analogy to the case of charged bosons, the natural choice of the order
parameter of a fermionic system with ODLRO is the condensate wave function
p
Ψ (R, r, α, β, t) = n0 (t)φ0 (r1 α, r2 β) , (422)

where we use instead of the individual particle coordinates r1 and r2 the relative
coordinate r = r1 − r2 and the center of gravity coordinate R = 21 (r1 + r2 ),
respectively. An alternative approach is motivated by the theory of magnetism.
Consider a magnet with global SU (2) spin-rotation invariance. Applying a
finite magnetic field B (r), the symmetry is spontaneously broken if the expec-
tation value
1 X

ψα† (r) σ αβ ψβ (r)


s0 (r) = lim lim (423)
2 B→0 N,V →∞
αβ

is finite. Without the external magnetic field, multiple degenerate configurations

would cancel each other, leading to a zero magnetization. The same is true if one
performs the limit B → 0 for a finite system as there is still a finite macroscopic
tunneling probability between degenerate states. This is the reason why the
zero field limit must be performed after the thermodynamic limit.
In the context of superconductivity, spontaneous symmetry breaking can be
analyzed if we add to the Hamiltonian a source term
ˆ X 
H = − dd r1 dd r2 ηαβ (r1 , r2 ) ψα† (r1 ) ψβ† (r2 ) + h.c. . (424)
αβ

A physical realization of the source field ηαβ (r1 , r2 ) is the coupling to another
superconductor via a weak Josephson junction (see below). Just like in case
of a magnet, we perform the limit η → 0 after the thermodynamic limit. One
expects ODLRO to be identical to a finite expectation value

Ψ (R, r, α, β, t) = lim lim hψβ (r2 ) ψα (r1 )i . (425)

η→0 N,V →∞

While a general proof for the equivalence between these two definitions does
not seem to exist, we will later show that they are identical within the BCS

73
theory. This formulation makes evident the statement that at a superconducting
transition the global U (1) symmetry

ψα (r) → eiθ ψα (r) (426)

is spontaneously broken. Breaking the global U (1) symmetry implies that the
particle number conservation is spontaneously broken. While one frequently en-
counters the notion that at the superconducting transition the electromagnetic
gauge symmetry is spontaneously broken, it seems more adequate to simply
refer to a global U (1) symmetry as the same symmetry is also broken in neu-
tral fermionic superfluids. What is unique about charged superfluids is however
associated with the condensed matter realization of the Higgs mechanism in
superconductors that we will discuss later. For a lucid discussion of the issue of
gauge symmetry breaking at the superconducting transition, see Ref.[?].
The source field ηαβ (r1 , r2 ) has well defined behavior upon exchanging par-
ticles. Fermi statistics implies that

ψα† (r1 ) ψβ† (r2 ) = −ψβ† (r2 ) ψα† (r1 ) . (427)

If we now relabel the indices r1 α ←→ r2 β the source field must compensate for
the minus sign to recover the original Hamiltonian, i.e.

ηαβ (r1 , r2 ) = −ηβα (r2 , r1 ) . (428)

The (2 × 2) matrix form of η suggests an expansion in terms of Pauli matrices

σ = (σ x , σ y , σ z ) and the unit matrix σ 0 . Out of those for matrices. σ y is the
y y
only one that is antisymmetric (σαβ = −σβα ). All other are symmetric. This
suggests an expansion (the additional factor i is for convenience):

ηαβ (r1 , r2 ) = ηs (r1 , r2 ) (iσ y )αβ + η t (r1 , r2 ) · (iσ y σ)αβ . (429)

The first term behaves like a singlet two particle wave function; it is antisym-
metric with respect to the spin indices, i.e. it must be symmetric with respect to
the spatial indices. The opposite is the case for the second term that describes
the triplet part of the source field.
The same is of course true for the order parameter itself, i.e. we expand

Ψ (R, r, α, β) = Ψs (R, r) (iσ y )αβ + Ψt (R, r) · (iσ y σ)αβ , (430)

where the singlet and triplet part obey:

Ψs (R, r) = Ψs (R, −r) ,

Ψt (R, r) = −Ψt (R, −r) . (431)

Consider now a three dimensional system with inversion symmetry. Then each
operator should either be even or odd under r → −r. The spin is a pseudo-
vector, i.e. it does not change under parity. Thus, it must hold that

Ψ (R, r, α, β) = ±Ψ (R, −r, α, β) . (432)

74
It follows that a superconducting state with inversion symmetry must either
form ODLRO of triplets or of singlets. For a combination of the singlet and
triplet pairing, the total wave function would have no well defined parity eigen-
value. It is interesting that our proof is valid even if one includes spin orbit
interaction. In crystals without inversion symmetry or on the surface of a three
dimensional crystal, both pairing states can of course exist simultaneously.
The two-particle density matrix ρ(2) is an equal-time correlation function.
Using the fluctuation-dissipation theorem we can therefore relate it to a retarded
Green’s function (we use the abbreviation {ri } = (r1 , r2 , r01 , r02 )
D E
(2)
ραβα0 β 0 ({ri }) = ψα† (r1 ) ψβ† (r2 ) ψβ 0 (r02 ) ψα0 (r01 )
ˆ ∞
dω Imχαβα0 β 0 ({ri } , ω + i0+ )
= − ,
−∞ π eβω − 1
´∞
where χ us the Fourier transform (χ (ω) = −∞ dtχ (t) eiωt ) of the retarded
function
h i 
χαβα0 β 0 ({ri } , t) = −iθ (t) ψα† (r1 , t) ψβ† (r2 , t) , ψβ 0 (r02 , 0) ψα0 (r01 , 0) ,
−

where the operators are now in the Heisenberg picture.Dχ is the pair-susceptibility
E
of the system, i.e. the change of the expectation value ψβ† (r2 , t) ψα† (r1 , t) with
respect to a rime dependent source field ηα0 β 0 (r01 , r02 , t0 ):
D E
δ ψβ† (r2 , t) ψα† (r1 , t) 
χαβα0 β 0 ({ri } , t − t0 ) =  . (433)
δηα∗ 0 β 0 (r01 , r02 , t0 ) 

η,η ∗ →0

As before, the limit of vanishing source fields must be taken after the thermo-
dynamic limit.

13.3 The pairing instability

The analysis of the previous section revealed that there seems to be an instabil-
ity of the Fermi surface with respect to a weak attractive interaction between
fermions. For simplicity, we consider a model with weak attraction governed by
the model Hamiltonian
Xˆ 
∇2

d †
H = d rψα (r) − − µ ψα (r)
α
2m
ˆ
+ V dd rψ↑† (r) ψ↓† (r) ψ↓ (r) ψ↑ (r) . (434)

We will see that the analysis of this continuum’s model is ill defined without
proper regularization. Therefore we consider a model where we restricts our-
selves to an effective low energy theory, i.e. we consider a system where the

75
fermionic excitations are confined to an energy scale ±Λ around the Fermi en-
ergy. We explore the behavior of this toy model.
In our Hamiltonian it suffices to consider a singlet wave function and to focus
on r1 = r2 and r01 = r02 , i.e. we analyze the pairing susceptibility.
h i 
χ (r, r0 , t) = −iθ (t) ψ↑† (r, t) ψ↓† (r, t) , ψ↓ (r0 , 0) ψ↑ (r0 , 0) . (435)
−

Fourier transformation and Wick rotation to the imaginary time axis yields
D E
χ (r, r0 , τ ) = − Tτ ψ↑† (r, τ ) ψ↓† (r, τ ) ψ↓ (r0 , 0) ψ↑ (r0 , 0) . (436)

In what follows we analyze this pairing susceptibility.

We first analyze the pair susceptibility of non-interacting electrons. The
Fourier transform in momentum and frequency space is then given as
X ˆ dd k
χ0 (q, iνm ) = T G (iωn ) G−k+q (−iωn + iνm ) ,
d k
(437)
n (2π)

where νm = 2mπT and ωn = (2n + 1) πT are bosonic and fermionic Matsubara

frequencies, respectively.
1
Gk (iωn ) = (438)
iωn − k
2
k
is the bare fermionic Green’s function and k = 2m − µ. Since we suspect that
superconductivity is a homogeneous instability, without spatial and temporal
modulations, we consider the limit q = 0 and ωn = 0. It follows
X ˆ dd k 1
χ0 (T ) = T d ω 2 + 2
m (2π) n k
Xˆ ρ ()
= T d 2 , (439)
m
ωn + 2

with density of states

ˆ
dd k
ρ (ω) = d
δ (ω − k ) . (440)
(2π)

We perform the Matsubara frequency sum and obtain

ˆ ∞
ρ ()   
χ0 (T ) = d tanh . (441)
−∞ 2 2T

This expression makes evident that some appropriate cut off procedure is re-
quired to analyze the pairing susceptibility. While the above integral is well
defined for any lattice model of a solid, where the density of states of individual
bands has some upper and lower cut off, the continuum’s theory diverges at the

76
upper cut off. As mentioned above, an appropriate approach is to define the
theory in an energy window [µ − Λ, µ + Λ] around the Fermi energy and assume
that the density of states is constant in this window. Then we have to evaluate:
ˆ Λ
1   
χ0 (T ) = ρF d tanh . (442)
−Λ 2 2T
We perform the integration to leading logarithmic accuracy:
ˆ Λ ε

ˆ βΛ/2
tanh 2T tanh (x)
dε = dx
−Λ 2ε 0 x
ˆ βΛ/2
log (x) βΛ/2
= − 2 dx + tanh (x) log x|0
0 cosh (x)
   
π Λ T
= γE − log + log +O
4 2T Λ
γE
   
2Λe T
= log +O . (443)
πT Λ
We obtain for the pairing susceptibility of a free electron gas
2ΛeγE
 
χ0 (T ) = ρF log .
πT
For any finite temperature the free electron pairing-susceptibility is finite. How-
ever the logarithmic increase of χ0 (T ) for T → 0 already indicates that a
Fermi gas becomes increasingly susceptible if one adds an external pairing source
y
ηαβ = ηs iσαβ .
Next we include electron-electron interactions. To this extend we sum ladder
diagrams for the pairing susceptibility. We obtain:
2 2
χ0 (q, iνm ) = χ0 (q, iνm ) − V χ0 (q, iνm ) + V 2 χ0 (q, iνm ) · · ·
χ0 (q, iνm )
= . (444)
1 + V χ0 (q, iνm )
In case of a negative (attractive) interaction V < 0, follows that the q = 0 and
νm = 0 susceptibility diverges at the transition temperature when

|V | χ0 (Tc ) = 1. (445)

With dimensionless coupling constant λ = ρF |V | follows

2eγE − 1
Tc = Λe λ . (446)
π

13.4 The BCS theory

The BCS theory gives an answer to the open question that emerges as con-
sequence of the Cooper instability: What happens with an entire Fermi-sea of

77
attractively interacting electrons? Based on the insight that the leading instabil-
ity occurs at zero center-of-mass momentum we model the attractive interaction
between electrons, mediated by phonons via he BCS or pairing Hamiltonian:
V0 X
εk c†kσ ckσ − γk,k0 c†k0 ,↑ c†−k0 ,↓ c−k,↓ ck,↑
X
H= (447)
N 0
k,σ k,k

It consists of the usual kinetic energy with band dispersion

k2
εk = −µ (448)
2m
and an interaction term. The choice of the parabolic spectrum is in fact not
necessary. The subsequent analysis will reveal that it is trivial to generalize the
approach to different dispersions as long as it is spin-independent and εk = ε−k .
The sign in front of the interaction V0 was chosen such that V0 > 0 corresponds
to an attractive coupling. The matrix element

1 |εk | , |εk0 | < ~ωD
γk,k0 =
0 otherwise
takes into account that only fermionic states that have energies relative to the
Fermi energy below the phonon frequency interact.
To find an approximate solution of this problem we perform the Hartree-Fock
decoupling

AB = (A − hAi) (B − hBi) + A hBi + B hAi − hAi hBi (449)

with

A = c†k0 ↑ c†−k0 ↓
B = c−k↓ ck↑ . (450)

This choice for A and B is motivated by the expectation, that hAi = 6 0 and
hBi =
6 0 amount to pairing of electrons, as discussed earlier. Performing the
mean field decoupling yields:
2
X  |∆|
εk c†kσ ckσ + ∆∗k c−k↓ ck↑ + ∆k c†k↑ c†−k↓ + N
X
MF
HBCS = , (451)
V0
k,σ k

where we introduced the abbreviation

V0 X
∆k = − γk,k0 hc−k,↓ ck,↑ i. (452)
N
k

∆ is the value of ∆k for momenta with |εk | < ωD . The last term is a consequence
of the hAi hBi expectation value, where we used
V02 X
∆2 = γk,k0 hc−k,↓ ck,↑ ihc†k0 ,↑ c†−k0 ,↓ i (453)
N 0
k,k

78
 2
since γk,k0 = γk,k 0.

The form of this Hamiltonian is similar to an effective free-electron prob-

lem in the sense that it only contains terms that are products of two opera-
tors c†kσ or ckσ , respectively. However, the appearance terms like ∆∗k c−k↓ ck↑
and ∆k c†k↑ c†−k↓ has no analog in the free-electron limit. An expectation value
D E
c†k0 ,↑ c†−k0 ,↓ 6= 0 must be understood as consequence of an external source field
that couples to the operators A and B and is switched off after the thermody-
namic limit has been taken. Within a mean field theory such source field would
only be an infinitesimal addition to the mean field anyway. Thus, we never
actually have to include the mentioned source field since we are breaking the
symmetry “by hand” anyway. Those anomalous terms are obviously the ones
that explicitly violate charge conservation at the mean field level. In order to
bring this Hamiltonian into the desired form, we introduce the Nambu spinor
 
ck↑
ck = , (454)
c†−k↓

MF
which allows us to express HBCS in a form that resembles more the usual free
fermion problem:

∆2 X
c†k hk ck + N
X
MF
HBCS = + εk , (455)
V0
k k

with 2 × 2-matrix  
εk ∆k
hk = . (456)
∆∗k −εk
2
The eigenvalues of hk are determined by (E − εk ) (E + εk ) − |∆k | = 0, which
yields
Ek± = ±Ek (457)
with q
2
Ek = ε2k + |∆k | > 0. (458)
hk is diagonalized by a unitary transformation Uk . The columns of Uk are the
(i)
eigenvectors uk of hk . Interestingly there is some nontrivial structure in the
matrix hk that is worth exploring as it can be very helpful for more complex
systems such as multi-band superconductors or inhomogeneous systems. It holds
with γ = iσ y (note γ 2 = −1) that

γh∗k γ −1 = −hk . (459)

(1) T
Suppose one eigenvector of hk is uk = (uk , vk ) and it corresponds to the
(1) (1)
eigenvalue +Ek , i.e. hk uk = Ek uk . We can now construct the another vector
(2) (1)∗ T
uk = −γuk = (−vk∗ , u∗k ) (460)

79
which obeys
(2)∗ (1) (1)
γuk = −γ 2 uk = uk (461)
(2)
uk is also an eigenvector but with eigenvalue −Ek . To show that this is
the case, we take the complex conjugate of the second eigenvalue equation
(2)∗ (2)∗ (2)∗ (2)∗
h∗k uk = −Ek uk and write it as γh∗k γ −1 γuk = −γEk uk which yields
(2)∗ (2)∗ (1)
−hk γuk = −γEk uk amd leads to the first eigenvalue equation hk uk =
(1)
Ek uk , proving our assertion. Thus, the eigenvalues of the mean field Hamil-
tonian occur in a pair of opposite sign and with eigenvalues related by the
unimodular transformation γ. The unitary transformation that diagonalizes
the above 2 × 2 matrix is

uk −vk∗
 
Uk = (462)
vk u∗k

and it follows Uk−1 hk Uk = diag (Ek , −Ek ). It is straightforward to determine

uk and vk from the eigenvalue equations. Unitarity, i.e. normalization of the
2 2
eigenvectors implies |uk | + |vk | = 1 and it follows

∆k
uk = vk . (463)
Ek − εk

This leads to:

 
1 εk
u2k = 1+
2 Ek
 
1 εk
vk2 = 1 − u2k = 1− . (464)
2 Ek
∆k
as well as uk vk∗ = − 2E k
.
The unitary transformation transforms the Nambu spinor ck according to
 T
ak = Uk−1 ck with ak = ak↑ , a†−k↓ and it follows
 
0 Ek
c†k hk ck c†k Uk
X X
= Uk−1 ck
−Ek 0
k k
X †  Ek 0

= ak ak
0 −Ek
k
 
Ek a†k↑ ak↑ − a−k↓ a†−k↓
X
=
k
 
Ek a†k↑ ak↑ + a−k↓ a†−k↓ − 1
X
=
k

Ek a†kσ akσ −
X X
= Ek (465)
k,σ k

80
The mean field Hamiltonian is then given as:

∆2 X
 q 
Ek a†kσ akσ + N
X 2
MF
HBCS = + εk − ε2k + |∆k | . (466)
V0
k,σ k

Now, we managed to bring theD Hamiltonian

E into the desired form of a free Fermi
† 1
gas. In particular, it holds akσ akσ = f (Ek ), where f (ε) = exp(βε)+1 is the
D E
†
usual Fermi function. Since Ek > 0, we obtain akσ akσ = 0 at T = 0. The
ground state energy is then given by the constant term in

∆2 X
 q 
2 2
E0 = N + εk − εk + |∆k | . (467)
V0
k

The fermionic excitations only describe excitations above the ground state.
In order to determine ∆ at T = 0 , we minimize E0 with respect to ∆. We
perform the momentum integration via an integration over energy and subtract
the value of the energy for ∆ = 0, i.e. δE0 (∆) = E0 (∆)−E0 (∆ = 0). It follows
ˆ ∞ ˆ ∞
∆2  p 
δE0 /N = + ρ () dε ε − ε2 + ∆2 − 2 ρ () dεθ (−ε) ε
V0 −∞ −∞
ˆ ωD 
∆2 p 
= + 2ρ dε ε − ε2 + ∆2
V0 0
∆2
 
2 ∆ ρ
= + ρ∆ log − ∆2 (468)
V0 2ωD 2

Minimizing the ground state energy with respect to ∆ yields

 
1 ∂E0 ∆ ∆
= 2 + 2ρ∆ log = 0,
N ∂∆ V0 2ωD

which has the trivial solution ∆ = 0 and the nontrivial solution

∆ (T = 0) = 2ωD exp (−1/λ) .

Inserting the latter into the energy, we find

2 −2/λ
E0 = E0 (∆ = 0) − 2N ρF ωD e .
N
= E0 (∆ = 0) − ρF ∆2 < E0 (∆ = 0) . (469)
2
The nontrivial solutions is always energetically lower.
Next we solve the gap equation. From the above unitary transfomation
follows:

ck↑ = u∗k ak↑ + vk a†−k↓

c†−k↓ = −vk ak↑ + uk a†−k↓

81
which yields
c−k↓ = −vk∗ a†k↑ + u∗k a−k↓
This can be used to express the operator product ck↑ c−k↓ that is needed to
determine ∆k . It holds:
  
ck↑ c−k↓ = u∗k ak↑ + vk a†−k↓ u∗k a−k↓ − vk∗ a†k↑

which yields  D E 
hc−k,↓ ck,↑ i = vk uk 2 a†−k↓ a−k↓ − 1
and we obtain fo the gap equation:
V0 X ∆
∆=− (2f (Ek ) − 1) . (470)
N 2Ek
k

We first analyze the gap quation at T = 0. Since Ek > 0 follows that f (Ek ) = 0
in the limit of T = 0 and the gap equation simplifies to
ˆ ωD
∆
∆ = V0 ρ dε √
−ωD 2 ε + ∆2
2
r !
ωD  ω 2
D
= λ∆ log + 1+
∆ ∆
 
2ωD
' λ∆ log . (471)
∆
This yields the result:∆ (T = 0) = 2ωD exp (−1/λ) in full agreement with the
one we obtained from the minimization of the ground state energy.
In the vicinity of the transition temperature the gap ∆ is small and we can
linearize the gap-equation. The linearized gap equation is
X ∆  
βξk
∆=V tanh (472)
2ξk 2
k

which we write with density of states ρ as

ˆ
 
βc ε
ωD tanh 2
∆ = V ρ∆ dε. (473)
−ωD 2ε
We perform the integral
ˆ ω0 tanh βε ˆ βωD /2
 
2 tanh (x)
dε = dx
−ω0 2ε 0 x
ˆ βωD /2
βω /2
= − sec h2 (x) log (x) dx + tanh (x) log x|0 D
0
2ωD eγE
 
π ω 
D
= γE − log + log = log , (474)
4 2T πT

82
and obtain
2ωD eγE
 
∆ = V ρ∆ log . (475)
πTc
which yields for the transition temperature:

2ωD eγE
   
1 1
Tc = exp − ' 1.134ωD exp − (476)
π λ λ

where λ = V ρ is the dimensionless coupling constant. If we compare the value

of the transition tempeature with the zero temperature gap, it follows

4ωD exp − λ1


2∆ (T = 0)
= 2eγE 1


kB Tc π ωD exp − λ
= 2πe−γE ' 3.527 3 (477)

which is in agreement with numerous observations for elementary supeconduc-

tors.
Finally, we want to determine the many body wave function that is associ-
ated with this new mean field state. To obtain the BCS wave function we use
the fact that Eq.466 implies that the ground state wave function is the vacuum
state of the Bogoliubov quasiparticles. Thus, it holds

akσ |ΦBCS i = 0 for all k, σ. (478)

To proceed, we assume
φk c†−k↑ c†k↓ † †
P Y
|ΦBCS i = Ce k |0i = C eφk ck↑ c−k↓ |0i , (479)
k

Here |0i is the vacuum state of the original operators, i.e. ciσ |0i = 0. We next
determine φk from the condition Eq.478. We write explicitly:

ak↑ = uk ck↑ − vk c†−k↓

ak↓ = vk c†−k↑ + uk ck↓ . (480)

Eq.478 is equivalent to

uk ck↑ |ΦBCS i = vk c†k↓ |ΦBCS i . (481)

We first analyze ck↑ |ΦBCS i . It is useful to introduce the operator

φk c†k↑ c†−k↓
X
θ= (482)
k

and it follows for the wave function

∞
X θn
|ΦBCS i ∝ |0i . (483)
n=0
n!

83
It is easy to show that [ck↑ , θ] = b†−k↓ with operator b†−k↓ = φk c†−k↓ . Further-
h i
more, it follows that b†−k↓ , θ = 0. It is now easy to apply ck↑ to each term in
the sum of Eq.483 separately. It holds:

ck↑ θ |0i = b†−k↓ |0i

ck↑ θ2 |0i = b†−k↓ θ |0i + θck↑ θ |0i = 2θb†−k↓ |0i
..
.
ck↑ θn |0i = nθn−1 b†−k↓ |0i . (484)

This result allows to sum-up the series Eq.483 and we obtain

ck↑ |ΦBCS i = b†−k↓ |ΦBCS i (485)

The condition akσ |ΦBCS i = 0 expressed in form of Eq.481 can now be expressed
as

uk φk c†k↓ |ΦBCS i = vk c†−k↓ |ΦBCS i . (486)

This implies immediately φk = vk /uk . It is easy to show that the condition

a−k↓ |ΦBCS i = 0 leads to the same condition. It follows with normalization
factor: Y
C= uk
k

for the wave function

Y † †
|ΦBCS i = uk evk /uk ck↑ c−k↓ |0i
k
Y 
= uk + vk c†k↑ c†−k↓ |0i .
k

 n
The last step is a consequence of the fact that due to Pauli Principle c†k↑ c†k↓ =
0 if n > 1. This approach allows to project the BCS-wave function into the space
of fixed number of electrons N .
!2N
φk c†k↑ c†k↓
X
|ΨBCS , N i = C |0i . (487)
k

This projection can alternatively be realized if one starts from the BCS ground
state and adds a global phase ϕ of the function φk :
ˆ 2π
dϕ −iN ϕ/2 Y  
|ΨBCS , N i = e uk + vk eiϕ c†k↑ c†k↓ |0i . (488)
0 2π
k

84
There exists an interesting relation between the phase ϕ and the pair number
operator N̂p which should give 2N in case of |ΨBCS , N i. With phase ϕ of φk
we have:
!Np
† †
X
iϕ
|ΨBCS , N i = C e φk ck↑ ck↓ |0i
k
!Np
φk c†k↑ c†k↓
X
iϕNp
= e C |0i . (489)
k

This demonstrates immediately that

∂
N̂p = −i . (490)
∂ϕ
suggesting that particle number and phase are canonically conjugated variables,
i.e. there should be a Heisenberg uncertainty relation between both quantities.
’

85