Superconductivity,

Superfluids and Condensates

James F. Annett
University of Bristol

Oxford University Press

May 2003

Contents
1

Superconductivity
1.1 Introduction
1.2 Conduction in metals
1.3 Superconducting materials
1.4 Zero-resistivity
1.5 The Meissner-Ochsenfeld Effect
1.6 Perfect Diamagnetism
1.7 Type I and Type II Superconductivity
1.8 The London Equation
1.9 The London vortex
1.10 Further Reading
1.11 Exercises

2
2
2
5
7
10
11
13
14
19
21
22

The
2.1
2.2
2.3

31
31
32

2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14

Ginzburg-Landau model
Introduction
The condensation energy
Ginzburg-Landau theory of the bulk phase transition
Ginzburg-Landau theory of inhomogenous systems
Surfaces of Superconductors
Ginzburg-Landau theory in a magnetic field
Gauge Symmetry and Symmetry Breaking
Flux quantization
The Abrikosov flux lattice
Thermal Fluctuations
Vortex Matter
Summary
Further Reading
Exercises

36
39
41
43
45
47
50
56
60
62
63
63

The
3.1
3.2
3.3
3.4
3.5
3.6

Macroscopic Coherent State

Introduction
Coherent states
Coherent States and the Laser
Bosonic Quantum Fields
Off-Diagonal Long Ranged order
The Weakly Interacting Bose Gas

72
72
73
78
79
83
85

Contents

3.7
3.8
3.9
3.10
3.11
3.12

Coherence and ODLRO in Superconductors

The Josephson Effect
Macroscopic Quantum Coherence
Summary
Further Reading
Exercises

90
95
99
101
102
102

The
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9

BCS Theory of Superconductivity

Introduction
The electron-phonon interaction
Cooper pairs
The BCS wave function
The mean-field Hamiltonian
The BCS energy gap and quasiparticle states
Predictions of the BCS theory
Further Reading
Exercises

111
111
113
116
120
122
124
125
128
128

Bibliography

136

1
Superconductivity
1.1

Introduction

This chapter describes some of the most fundamental experimental facts

about superconductors, together with the simplest theoretical model: the
London equation. We shall see how this equation leads directly to the expulsion of magnetic fields from superconductors, the Meissner-Ochsenfeld
effect, which usually considered to be is the fundamental property which
defines superconductivity.
The chapter starts with a brief review of the Drude theory of conduction
in normal metals. We shall also show how it is possible to use the Drude
theory to make the London equation plausible. We shall also explore some
of the consequences of the London equation, in particular the existence
of vortices in superconductors and the differences between type I and II
superconductors.

1.2

Conduction in metals

The idea that metals are good electrical conductors because the electrons
move freely between the atoms was first developed by Drude in 1905, only
five years after the original discovery of the electron.
Although Drudes original model did not include quantum mechanics,
his formula for the conductivity of metals remains correct even in the modern quantum theory of metals. To briefly recap the key ideas in the theory
of metals, we recall that the wave functions of the electrons in crystalline
solids obey Blochs theorem,1 .
nk (r) = unk (r)eik.r .

(1.1)

Where here unk (r) is a function which is periodic, h

k is the crystal momentum, and k takes values in the first Brillouin zone of the reciprocal
lattice. The energies of these Bloch wave states give the energy bands, nk ,
where n counts the different electron bands. Electrons are fermions, and
1 See for example, the text Band theory and electronic properties of solids by J.
Singleton (2002), or other textbooks on Solid State Physics, such as Kittel (1996), or
Ashcroft and Mermin(1976)

Conduction in metals

so at temperature T a state with energy  is occupied according to the

Fermi-Dirac distribution
f () =

1
e()

+1

(1.2)

The chemical potential, , is determined by the requirement that the total

density of electrons per unit volume is
Z
N
2 X
1
d3 k
(1.3)
=
V
(2)3 n
e(nk ) + 1
where the factor of 2 is because of the two spin states of the s = 1/2
electron. Here the integral over k includes all of the first Brillouin zone
of the reciprocal lattice and, in principle, the sum over the band index n
counts all of the occupied electron bands.
In all of the metals that we are interested the temperature is such that
this Fermi gas is in a highly degenerate state, in which kB T << . In
this case f (nk ) is nearly 1 in the region inside the Fermi surface, and
is 0 outside. The Fermi surface can be defined by the condition nk = F ,
where F = is the Fermi energy. In practice, for simplicity, in this book
we shall usually assume that there is only one conduction band at the Fermi
surface, and so we shall ignore the band index n from now on. In this case
the density of conduction electrons, n, is given by
Z
1
2
d3 k
(1.4)
n=
3
(
)
k
(2)
e
+1
where k is the energy of the single band which crosses the Fermi surface.
In cases where the single band approximation is not sufficient, it is quite
easy to add back a sum over bands to the theory whenever necessary.
Metallic conduction is dominated by the thin shell of quantum states
with energies F kB T <  < F + kB T , since these are the only states
which can be thermally excited at temperature T . We can think of this
as a low density gas of electrons excited into empty states above  F and
of holes in the occupied states below F . In this Fermi gas description of
metals the electrical conductivity, , is given by the Drude theory as,
=

ne2
,
m

(1.5)

where m is the effective mass of the conduction electrons2 , e is the electron

charge and is the average lifetime for free motion of the electrons between
collisions with impurities or other electrons.
2 Note that the band mass of the Bloch electrons, m, need not be the same as the
bare mass of an electron in vacuum, me . The effective mass is typically 2 3 times
greater. In the most extreme case, the heavy fermion materials m can be as large
as 50 100me !

Superconductivity

The conductivity is defined by the constitutive equation

j = .

(1.6)

Here j is the electrical current density which flows in response to the external electric field, . The resistivity obeys

= j,

(1.7)

and so is simply the reciprocal of the conductivity, = 1/. Using the

Drude formula we see that
m
= 2 1 ,
(1.8)
ne
and so the resistivity is proportioal to the scattering rate, 1 of the conduction electrons. In the SI system the resistivity has units of m, or is
more often quoted in cm.
Eq. 1.5 shows that the electrical conductivity depends on temperature
mainly via the different scattering processes which enter into the mean lifetime . In a typical metal there will be three main scattering processes,
scattering by impurities, by electron-electron interactions and by electronphonon collisions. These are independent processes, and so we should add
the scattering rates to obtain the total effective scattering rate
1
1
1
1 = imp
+ elel
+ elph
,

(1.9)

1
1
where imp
is the rate of scattering by impurities, elel
the electron1
electron scattering rtae, and elph the electron phonon scattering rate.
Using Eq. 1.8 we see that the total resistivity is just a sum of independent
contributions from each of these different scattering processes,

m  1
1
1
+ elel
+ elph
.
(1.10)
= 2 imp
ne

Each of these lifetimes is a characteristic function of temperature. The

1
impuri ty scattering rate, imp
, will be essentially independent of temperature, at least for the case of non-magnetic impurities. The electron-electron
1
scattering rate, elel
, is proportional to T 2 , where T is the temperature.
While at low temperatures (well below the phonon Debye temperature) the
1
electron-phonon scattering rate, elph
, is proportional to T 5 . Therefore we
would expect that the resistivity of a metal is of the form
= 0 + aT 2 + . . .

(1.11)

at very low temperatures. The zero temperature resistivity, the residual

resistivity, 0 , depends only on the concentration of impurities.
For most metals the resistivity does indeed behave in this way at low
temperatures. However for a superconductor something dramatically different happens. Upon cooling the resistivity first follows the simple smooth

Superconducting materials

behaviour, Eq. 1.11, but then suddenly vanishes entirely, as sketched in

Fig. 1.1. The temperature where the resistivity vanishes is called the critical temperature, Tc . Below this temperature the resistivity is not just small,
but is, as far as can be measured, exactly zero.
This phenomenon was a complete surprise when it was first observed
by H. Kammerling Onnes in 1911. He had wanted to test the validity of the
Drude theory by measuring the resistivity at the lowest temperatures possible. The first measurements on samples of platinum and gold were quite
consistent with the Drude model. But then he then turned his attention to
mercury, because of its especially high purity. Based on Eq. 1.11 one could
expect a very small, perhaps even zero, residual resisitivity in exceptionally pure substances. But what Kammerling Onnes actually observed was
completly unexpected, and not consistent with Eq. 1.11. Surprisingly he
discovered that all signs of resistance appeared to suddenly vanished suddenly below about 4K. This was quite unexpected from the Drude model,
and was, in fact, the discovery of a new state of matter: superconductivity.

1.3

Superconducting materials

A number of the elements in the periodic table become superconducting at

low temperatures, as summarized in Table 1.1. Of the elements, Niobium
(Nb) has the highest critical temperature Tc of 9.2K at atmospheric pressure. Interestingly, some while common metals such as aluminium (1.2K),
tin (3.7K) and lead (7.2K) become superconducting, other equally good,
or better, metals (such as copper, silver or gold) show no evidence for superconductivity at all. It is still a matter of debate whether or not they
would eventually become superconducting if made highly pure and cooled
to sufficiently low temperatures. As recently as 1998 it was discovered that
extremely pure platinum becomes superconducting, but only when it is
prepared into small nano-particles at temperatures of a few miliKelvin.
Another recent discovery is that quite a few more elments also become
superconducting when they are subjected to extremely high pressures. Samples must be pressurized between two anvil shaped diamonds. Using this
techni que it is possible to obtain such high pressures that substances which
are normally insulators become metallic, and some of these novel metals
become superconductin g. Sulphur and oxygen both become superconducting at surprisingly high temperatures . Even iron becomes superconducting
under pressure. At normal pressures iron is, of course, magnetic, and the
magnetism prevents superconductivity from occuring . However, at high
pressures a non-magnetic phase can be found, and this becomes superconducting. For many years the holy grail for this sort of high pressu re work
has been to looks for superconductivity in metallic hydrogen. It has been
predicted that metallic hydrogen could become superconducting at as high
as 300K, which would be the first room temperature superconductor! To
date, high pressure phases of metallic hydrogen have indeed been produced,
but, so far at least, superconductivity has not been found.

Superconductivity
Table 1.1 Some selected superconducting elements and compounds

substance

Tc (K)

Al
Hg
Nb
Pb
Sn
Ti
Tl
V
W
Zn
Zr

1.2
4.1
9.3
7.2
3.7
0.39
2.4
5.3
0.01
0.88
0.65

Fe
H
O
S

2
300
30
10

Nb3 Ge
Ba1x Pbx BiO3
La2x Srx CuO4
YBa2 Cu3 O7
HgBa2 Ca2 Cu3 O8+
K3 C60
YNi2 B2 C
MgB2
Sr2 RuO4
UPt3
(TMTSF)2 ClO4
ET-BEDT

23
12
35
92
135-165
30
17
38
1.5
0.5
1.2
12

first superconductor, discovered 1911

higest Tc of an element at normal pressure

high pressure
predicted, under high pressure
high pressure, maximum Tc of any element
high pressure
A15 structure, highest known Tc before 1986
first perovskite oxide structure
first high Tc superconductor
first superconductor above 77K
highest Tc ever recorded
fullerene molecules
borocarbide superconductor
discovery announced in January 2001
possible p-wave superconductor
heavy fermion exotic superconductor
organic molecular superconductor
organic molecular superconductor

Superconductivity appears to be fairly common in nature, and there are

perhaps several hundred known superconducting materials. Before 1986 the
highest known Tc values were in the A-15 type materials, including Nb3 Ge
with Tc = 23K. This, and the closely related compound Nb3 Sn (Tc = 18K
are widely used in the superconducting magnet industry.
In 1986 Bednorz and M
uller discovered that the material La2x Bax CuO4 becomes superconducting with a Tc which is maximum at 38K for
x 0.15. Within a matter of months the related compound YBa2 Cu3 O7
was discovered to have Tc = 92K, ushering in the era of high temperature

Zero-resistivity

superconductivity.3 This breakthrough was especially important in terms

of possible commercial applications of superconductivity, since these superconductors are the first which can operate in liquid nitrogen (boiling point
77K) rather than requiring liquid helium (4K) . Other high temperature
superconductors have been discovered in chemically related systems. Currently HgBa2 Ca2 Cu3 O8+ has the highest confirmed value of Tc at 135K at
room pressure, shown in Fig. 1.2, rising to 165K when the material is subjected to high pressures. The reason why these particular materials are so
unique is still not completely understood, as we shall see in later chapters
of this book.
As well as high temperature superconductors, there are also many other
interesting superconducting materials. Some of these have exotic properties which are still not understood and are under very active investigation.
These include other oxide-based superconducting materials, organic superconductors, C60 based fullerene superconductors, and heavy fermion
superconductors (typically compounds containing the elements U or Ce)
which are dominated by strong electron-electron interaction effects. Other
superconductors have surprising properties, such as coexitence of magnetism and superconductivity, or evidence of exotic unconventional superconducting phases. We shall discuss some of these strange materials in
chapter 7.

1.4

Zero-resistivity

As we have seen, in superconductors the resistivity, , becomes zero, and

so the conductivity appears to become infinite below Tc . To be consistent
with the constitutive relation, Eq. 1.6, we must always have zero electric
field,
= 0,
at all points inside a superconductor. In this way the current, j, can be
finite. So we have current flow without electric field.
Notice that the change from finite to zero resistivity at the superconducting critical temperature Tc is very sudden, as shown in Fig. 1.1. This
represents a thermodynamic phase transition from one state to another.
As for other phase transitions, such as from liquid to gas, the properties
of the phases on either side of the transition can be completely different.
The change from one to the other occurs sharply at a fixed temperature
3 Bednorz and M
uller received the 1987 Nobel prize for physics, within a year of
publication of their results. At the first major condensed matter physics conference after
these discoveries, the 1987 American Physical Society March Meeting held in New York
city, there was a special evening session devoted to the discoveries. The meeting hall
was packed with hundreds of delegates sitting in the gangways, others had to watch the
proceedings on TV screens in the hallways. The number of speakers was so great that
the session lasted all through the night until the following morning, when the hall was
needed for next offical session of the conference! The following days New York Times
newpaper headline reported the meeting as the Woodstock of Physics.

Superconductivity

rather than being a smooth cross-over from one type of behaviour to another. Here the two different phases are referred to as the normal state
and the superconducting state. In the normal state the resistivity and
other properties behaves similarly to a normal metal, while in the superconducting state many physical properties, including resistivity, are quite
different.
In some cases, notably the high temperature superconductors, looking
closely at the (T ) curve near to Tc shows a small range of temperatures
where the resistance starts to decrease before becoming truly zero. This
is visible in Fig. 1.2 as a slight bend just above Tc . This bend is due to
thermodynamic critical fluctuations associated with the phase transition.
The precise thermodynamic phase transition temperature Tc can be defined
as the temperature where the resistivity first becomes exactly zero. 4
The key characteristic of the superconducting state is that the resistivity
is exactly zero,
= 0,
(1.12)
or the conductivity, , is infinite. How do we know that the resistivity is
exactly zero? After all, zero is rather difficult to distinguish from some very
very small, but finite, number.
Consider how one might actually measure the resistivity of a superconductor. The simplest measurement would be a basic two terminal
geometry shown in Fig. 1.3(a). The sample resistance, R, is related to the
resisitvity
L
(1.13)
R=
A
and L is the sample length and A is its cross sectional area. But the problem
with the two-terminal geometry shown in Fig. 1.3(a) is that even if the
sample resistance is zero the overall resistance is finite, because the sample
resistance is in series with resistances from the connecting leads and from
the electrical contacts between the sample and the leads. A much better
experimental technique is the four terminal measurement of Fig. 1.3(b).
There there are four leads connected to the sample. Two of them are used
to provide a current, I, through the sample. The second pair of lead are
then used to measure a voltage, V . Since no current flows in the second
pair of leads the contact resistances will not matter. The resistance of the
part of the sample between the second pair of contacts will be R = V /I
by Ohms law, at least in the idealized geometry shown. In any case if
4 One could perhaps imagine the existence of materials where the resistivity approached zero smoothly without a thermodynamic phase transition. For example in
a completely pure metal with no impurities one might expect the 0 as temperature approaches absolute zero. Such system would not be classified as a superconductor
in the standard terminology, even though it might have infinite conductivity. The word
superconductor is used only to mean a material with a definite phase transition and critical temperature Tc . A true superconductor must also exhibit the Meissner-Ochsenfeld
effect.

Zero-resistivity

the sample is superconducting we should definitely observe V = 0 when

I is finite implying that = 0. (Of course the current I must not be
too large. All superconductors have a critical current, Ic , above which the
superconductivity is destroyed and the resistance becomes finite again).
The most convincing evidence that superconductors really have =
0 is the observation of persistent currents. If we have a closed loop of
superconducting wire, such as the ring shown in Fig. 1.4 then it is possible
to set up a current, I, circulating in the loop. Because there is no dissipation
of energy due to finite resistance, the energy stored in the magnetic field
of the ring will remain constant and the current never decays.
To see how this persistent current can be set up, consider the flux of
magnetic field through the centre of the superconducting ring. The flux is
defined by the surface integral
Z
= B.dS
(1.14)
where dS is a vector perpendicular to the plane of the ring. Its length dS,
is an infinitesimal element of the area enclosed by the ring. But, by using
the Maxwell equation
B
=
(1.15)
t
and Stokess theorem
Z
I
( )dS =
.dr
(1.16)
we can see that

d
=
dt

.dr

(1.17)

where the line integral is taken around the closed path around the inside
of the ring. This path can be taken to be just inside the superconductor,
and so = 0 everywhere along the path. Therefore
d
=0
dt

(1.18)

and hence the magnetic flux through the ring stays constant as a function
of time.
We can use this property to set up a persistent current in a superconducting ring. In fact it is quite analogous to the way we saw in Chapter
2 how to set up a persistent superfluid flow in 4 He. The difference is that
now we use a magnetic field rather than rotation of the ring. First we start
with the superconductor at a temperature above Tc , so that it is in its normal state. Then apply an external magnetic field, Bext . This passes easily
through the superconductor since the system is normal.RNow cool the system to below Tc . The flux in the ring is given by = Bext .dS. But we

10

Superconductivity

know from Eq. 1.18 that this remains constant, no matter what. Even if
we turn off the source of external magnetic field, so that now Bext = 0, the
flux must remain constant. The only way the superconductor can keep
constant is to generate its own magnetic field B through the centre of the
ring, which it must achieve by having a circulating current, I, around the
ring. The value of I will be exactly the one required to induce a magnetic
flux equal to inside the ring. Further, because is constant the current
I must also be constant. We therefore have a set up circulating persistent
current in our superconducting ring.
Furthermore if there were any electrical resistance at all in the ring there
would be energy dissipation and hence the current I would decay gradually
over time. But experiments have been done in which persistent currents
were observed to remain constant over a period of years. Therefore the
resistance must really be exactly equal to zero to all intents and purposes!

1.5

The Meissner-Ochsenfeld Effect

Nowadays, the fact the the resistivity is zero, = 0, is not taken as the
true definition of superconductivity. The fundamental proof that superconductivity occurs in a given material is the demonstration of the MeissnerOchsenfeld effect.
This effect is the fact the a superconductor expels a weak external
magnetic field. First, consider the situation illustrated in Fig. 1.5 in which
a small spherical sample of material is held at temperature T and placed in
a small external magnetic field, Bext . Suppose initially we have the sample
in its normal state, T > Tc , and the external field is zero, as illustrated
in the top part of the diagram in Fig. 1.5. Imagine that we first cool to a
temperature below Tc (left diagram) while keeping the field zero. Then later
as we gradually turn on the external field the field inside the sample must
remain zero (bottom diagram). This is because, by the Maxwell equation
Eq. 1.15 combined with = 0 we must have
B
=0
t

(1.19)

at all points inside the superconductor. Thus by applying the external field
to the sample after it is already superconducting we must arrive at the
state shown in the bottom diagram in Fig. 1.5 where the magnetic field
B = 0 is zero everywhere inside the sample.
But now consider doing things in the other order. Suppose we take the
sample above Tc and first turn on the external field, Bext . In this case the
magnetic field will easily penetrate into the sample, B = Bext , as shown
in the right hand picture in Fig. 1.5. What happens then we now cool
the sample? The Meissner-Ochsenfeld effect is the observation that upon
cooling the system to below Tc the magnetic field is expelled. So that by
cooling we move from the situation depicted on right to the one shown
at the bottom of Fig. 1.5. This fact cannot be deduced from the simple

Perfect Diamagnetism

11

fact of zero resistivity ( = 0) and so this is a new and separate physical

phenomenon assosciated with superconductors.
There are several reasons why the existence of the Meissner-Ochsenfeld
effecting a sample is taken as definitive proof of superconductivity. At a
practical level it is perhaps clearer to experimentally demonstrate the flux
expulsion than zero resistivity, because, for example, it is not necessary to
attach any electrical leads to the sample. A more fundamental reason is
that the Meissner-Ochsenfeld effect is a property of thermal equilibrium,
while resistivity is a non-equilibrium transport effect. In fact one can see in
Fig. 1.5 that we reach the same final state of the system (bottom picture
in Fig. 1.5) whether we first cool and then apply the field, or the other way
around. Therefore the final state of the system does not depend on the history of the sample, which is a necessary condition for thermal equilibrium.
It is perhaps possible to imagine exotic systems for which the resistivity
vanishes, but for which there is no Meissner-Ochsenfeld effect. In fact some
quantum Hall effect states may possess this property. But, for the purposes
of this book however we shall always define a superconductor as a system
which exhibits the Meissner-Ochsenfeld effect.

1.6

Perfect Diamagnetism

In order to maintain B = 0 inside the sample whatever (small) external

fields are imposed as required by the Meissner-Ocshenfeld effect there obviously must be screening currents flowing around the edges of the sample.
These produce a magnetic field which is equal and opposite to the applied
external field, leaving zero field in total.
The simplest way to describe these screening currents is to use Maxwells
equations in a magnetic medium (see Blundell (2002) or other texts on magnetic materials). The total current is separated into the externally applied
currents (for example in the coils producing the external field), jext , and
the internal screening currents, jint ,
j = jext + jint .

(1.20)

The screening currents produce a magnetization in the sample, M per unit

volume, defined by
M = jint .
(1.21)
As in the theory of magnetic media (Blundell 2002) we also define a
magnetic field H in terms of the external currents only
H = jext .
The three vectors M and H and B are related by

(1.22)
5

5 Properly the name magnetic field is applied to H in a magnetic medium. Then

the field B is called the magnetic induction or the magnetic flux density. Many people
find this terminology confusing. Following Blundell (2002), in this book we shall simply
call them the H-field and B-field respectively whenever there is a need to distinguish
between them.

12

Superconductivity

B = o (H + M).

(1.23)

Maxwells equations also tell us that

.B = 0.

(1.24)

The magnetic medium Maxwells equations above are supplemented by

boundary conditions at the sample surface. From Eq. 1.24 it follows that
the component of B perpendicular to the surface must remain constant;
while from the condition Eq. 1.22 one can prove that components of H
parallel to the surface remain constant. The two boundary conditions are
therefore,
B
Hk

=
=

0
0.

(1.25)
(1.26)

Note that we are using SI units here. In SI units B is in Tesla, while M

and H are in units of Amperes per metre, Am1 . The constant o =
4 107 .
One should take note that many books and research papers on superconductivity still use the older c.g.s. units. In c.g.s. units B and H are in
gauss and oersteds, respectively. 1gauss = 104 T, 1oersted = 103 /4
Am1 and in cgs units
B = H + 4M
and
H = 4j.
In these units the susceptibility of a superconductor is = 1/(4)
rather than the SI value of 1.
Note that there is no o or o in the c.g.s. system of units. Instead,

the speed of light, c = 1/ 0 0 , often appears explicitly. For example

the Lorentz force on a charge q particle, moving with velocity v in a
magnetic field B is
1
F = qv B
c
in c.g.s. units, compared to the SI unit equivalent
F = qv B.
Also note that the unit of electrical charge is the Coulomb (C) in SI
units, but it is the statcoulomb in cgs units, where 1statcoulomb =
3.336 1010 C.
For simplicity we shall usually assume that the sample is an infinitely
long solenoid as sketched in Fig. 1.6. The external current flows in solenoid

Type I and Type II Superconductivity

13

coils around the sample. In this case the field H is uniform inside the
sample,
N
(1.27)
H = I ez
L
where I is the current flowing through the solenoid coil and there are N
coil turns in length L. ez is a unit vector along the solenoid axis.
Imposing the Meissner condition B = 0 in Eq. 1.23 immediately leads
to the magnetization
M = H.
(1.28)
The magnetic susceptibility is defined by

dM 
=
dH H=0

(1.29)

and so we find that for superconductors

= 1

(1.30)

( or 1/4 in cgs units!).

Solids with a negative value of are called diamagnets (in contrast
positive is a paramagnet). Diamagnets screen out part of the external
magnetic field, and so they become magnetized oppositely to the external field. In superconductors the external field is completely screened out.
Therefore we can say that superconductors are perfect diamagnets.
The best way to detect superconductivity in some unknown sample is
therefore to measure its susceptibility. If the sample is fully superconducting
then as a function of T will something like the sketch giving in Fig. 1.7
sketch. Thus by measuring one will find = 1 in a superconductor,
evidence for perfect diamagnetism, or the Meissner effect. This is usually
considered much more reliable evidence for superconductivity in a sample
than zero resistance alone would be.

1.7

Type I and Type II Superconductivity

This susceptibility is defined in the limit of very weak external fields, H.

As the field becomes stronger it turns out that either one of two possible
things can happen.
The first case, called a type I superconductor, is that the B field remains zero inside the superconductor until suddenly the superconductivity
is destroyed. The field where this happens is called the critical field, Hc .
The way the magnetization M changes with H in a type I superconductor
is shown in Fig. 1.8. As shown, the magnetization obeys M = H for all
fields less than Hc , and then becomes zero (or very close to zero) for fields
above Hc .
Many superconductors, however, behave differently. In a type II superconductor there are two different critical fields, denoted Hc1 , the lower

14

Superconductivity

critical field, and Hc2 the upper critical field. For small values of applied
field H the Meissner-Ochsenfeld effect again leads to M = H and there
is no magnetic flux density inside the sample, B = 0. However in a type
II superconductor once the field exceeds Hc1 , magnetic flux does start to
enter the superconductor and hence B 6= 0, and M is closer to zero than
the full Meissner-Ochsenfeld value of H. Upon increasing the field H further the magnetic flux density gradually increases, until finally at Hc2 the
superconductivity is destroyed and M = 0. This behaviour is sketched on
the right hand side of Fig. 1.8
As a function of the temperature the critical fields vary, and they all
approach zero at the critical temperature Tc . The typical phase diagrams
of type I and type II superconductors, as a function of H and T are shown
in Fig. 1.9.
The physical explanation of the thermodynamic phase between Hc1 and
Hc2 was given by Abrikosov. He showed that the magnetic field can enter
the superconductor in the form of vortices, as shown in fig. 1.10. Each
vortex consists of a region of circulating supercurrent around a small central core which has essentially become normal metal. The magnetic field
is able to pass through the sample inside the vortex cores, and the circulating currents server to screen out the magnetic field from the rest of the
superconductor outside the vortex.
It turns out that each vortex carries a fixed unit of magnetic flux, 0 =
h/2e (see below), and hence, if there are a total of Nv vortices in a sample
of total area, A, then the average magnetic flux density, B, is
B=

Nv h
.
A 2e

(1.31)

It is instructive to compare this result for the number of vortices per unit
area,
Nv
2eB
=
.
(1.32)
A
h
with the similar expression derived earlier for the density of vortices in rotating superfluid 4 He, Eq. ??. There is in fact a direct mathematical analogy
between the effect of a uniform rotation at angular frequency in a neutral
superfluid, and the effect of a magnetic field, B, in a superconductor.

1.8

The London Equation

The first theory which could account for the existence of the MeissnerOchsenfeld effect was developed by two brothers, F. London and H. London, in 1935. Their theory was was originally motivated by the two-fluid
model of superfluid 4 He. They assumed that some fraction of the conduction electrons in the solid become superfluid while the rest remain normal.
They then assumed that the superconducting electrons could move without
dissipation, while the normal electrons would continue to act as if they had
a finite resistivity. Of course the superfluid electrons always short circuit

The London Equation

15

the normal ones and make the overall resistivity equal to zero. As in the
theory of superfluid 4 He discussed in chapter 2, we denote the number density of superfluid electrons by ns and the density of normal electrons by
nn = n ns , where n is the total density of electrons per unit volume.
Although this model is simple several of its main predictions are indeed
correct. Most importantly it leads to the London equation which relates the
electric current density inside a superconductor, j, to the magnetic vector
potential, A, by
j=

n s e2
A.
me

(1.33)

This is one of the most important equations describing superconductors.

Nearly twenty years after it was originally introduced by the London brothers it was eventually derived from the full microscopic quantum theory of
superconductivity by Bardeen Cooper and Schrieffer.
Let us start to make the London equation Eq. 1.33 plausible by reexamining the Drude model of conductivity. This time consider the Drude theory
for finite frequency electric fields. Using the complex number represntation
of the a.c. currents and fields, d.c. formula becomes modified to:
jeit = ()eit

(1.34)

where the conductivity is also complex. Its real part corresponds to currents which are in phase with the applied electrical field (resistive), while
the imaginary part corresponds to out of phase currents (inductive and
capacitive).
Generalizing the Drude theory to the case of finite frequency, the conductivity turns out to be
() =

ne2
1
,
m 1 i

(1.35)

Ashcroft and Mermin (1976). This is essentially like the response of a

damped Harmonic oscillator with a resonant frequency at = 0. Taking
the real part we get
Re[()] =

ne2
,
m 1 + 2 2

(1.36)

a Lorentzian function of frequency. Note that the width of the Lorentzian

is 1/ and its maximum height is . Integrating over frequency, we see that
the area under this Lorentzian curve is a constant
Z +
ne2
Re[()]d =
(1.37)
m

independent of the lifetime .

16

Superconductivity

Now is is interesting to consider what would be the corresponding Drude

model () in the case of a perfect conductor, where there is no scattering
of the electrons. We can we can obtain this by taking the limit 1 0 in
the Drude model. Taking this limit Eq. 1.35 gives:
() =

ne2
1
ne2

m 1 i
im

(1.38)

at any finite frequency, . There is no dissipation since the current is always

out of phase with the applied electric field and () is always imaginary.
There is a purely inductive response to an applied electric field. The real
part of the conductivity Re[()] is therefore zero at any finite frequency,
in this 1 0 limit. But the sum rule, Eq. 1.37, must be obeyed whatever
the value of . Therefore the real part of the conductivity, Re[()] must be
a function which is zero almost everywhere but which has a finite integral.
This must be, of course, a Dirac delta function,
Re[()] =

ne2
().
m

(1.39)

One can see that this is correct by considering the 1 0 limit of the
Lorentzian peak in Re[()] in Eq. 1.36. The width of the peak is of order
1 and goes to zero, but the maximum height increases keeping a constant
total area because of the sum rule. The 1 limit is thus a Dirac delta
function located at = 0.
Inspired by the two fluid model of superfluid 4 He, the London brothers
assumed that we can divide the total electron density, n, into a normal
part, nn and a superfluid part, ns ,
n = n s + nn .

(1.40)

They assumed that the normal electrons would still have a typical metallic
damping time , but the superfluid electrons would move without dissipation, corresponding to = . They assumed that this superfluid component will give rise to a Dirac delta function peak in the conductivity located
at = 0 and a purely imaginary response elsewhere,
() =

n s e2
ns e2
()
.
me
ime

(1.41)

Note that we effectively define ns by the weight in this delta function peak,
and (by convention) we use the bare electron mass in vacuum, me , rather
than the effective band mass, m, in this definition.
In fact the experimentally measured finite frequency conductivity Re()
in superconductors does indeed have a delta function located at zero frequency. But other aspects of the two fluid model conductivity assumed by
London and London are not correct. In particular the normal fluid component is not simply like the conductivity of a normal metal. In fact the

The London Equation

17

complete Re[()] of a superconductor looks something like the sketch in

Fig. 1.11. There is a delta function peak located at = 0, and the amplitude of the peak defines ns , the superfluid density or or condensate density.
At higher frequencies the real part of the conductivity is zero, Re[()] = 0,
corresponding to dissipationless current flow. However above a certain frequency, corresponding to h
= 2 (where 2 is the energy gap) the
conductivity again becomes finite. The presence of an energy gap was observed shortly before the Bardeen Cooper and Schrieffer (BCS) theory was
completed, and the energy gap was a central feature of the theory, as we
shall see later.
Derivation of the London Equation
If we restrict our attention to frequencies below the energy gap, then the
conductivity is exactly given by Equation 1.41. In this regime we can derive
the London equation relating the supercurrent j to the magnetic field B.
Taking the curl of both sides of the equation j = we find
( j)eit

()( )eit
d(Beit )
= ()
dt
= i()Beit
ns e2 it
Be
,
=
me

(1.42)

where in the final step we use Eq. 1.38 for the finite frequency conductivity
of the superconductor.
We now take the = 0 limit of the above equations. The last line
effectively relates a d.c. current, j to a static external magnetic field B by,
j=

n s e2
B.
me

(1.43)

This equation completely determines j and B because they are also related
by the static Maxwell equation:
B = o j.

(1.44)

Combining these two equations gives

( B) = o
or

n s e2
B
me

(1.45)

1
B
(1.46)
2
where has dimensions of length, and is the penetration depth of the
superconductor,
( B) =

18

Superconductivity

me
o n s e2

1/2

(1.47)

It is the distance inside the surface over which an external magnetic field
is screened out to zero, given that B = 0 in the bulk.
Finally, the London equation can also be rewritten in terms of the magnetic vector potential A defined by
B = A,

(1.48)

giving
j

n s e2
A
me
1
A.
=
o 2

(1.49)
(1.50)

Note that this only works provided that we choose the correct gauge for the
vector potential, A. Recall that A is not uniquely defined from Eq. 1.48
since A + (r) leads to exactly the same B for any scalar function, (r).
But conservation of charge implies that the current and charge density, ,
obey the continuity equation

+ .j = 0.
t

(1.51)

In a static, d.c., situation the first term is zero, and so .j = 0. Comparing

with the London equation in the form, Eq. 1.49 we see that this is satisfied
provided that the gauge is chosen so that .A = 0. This is called the
London gauge.
For superconductors this form of the London equation effectively replaces the normal metal j = constitutive relation by something which
is useful when is infinite. It is interesting to speculate about whether or
not it would be possible to find other states of matter which are perfect
conductors with = , but which do not obey the London equation. If
such exotic states exist (and they may indeed occur in the Quantum Hall
Effect) they would not be superconductors in the sense in which we are
using that word here.
The most important consequence of the London equation is to explain
the Meissner-Ochsenfeld effect. In fact one can easily show that any external
magnetic field is screened out inside the superconductor, as
B = B0 ex/

(1.52)

where x is the depth inside the surface of the superconductor. This is

illustrated in Fig. 1.12. The derivation of this expression from the London
equation is very straightforward, and is left to exercise 3.1 at the end of this
chapter. The implication of this result is that magnetic fields only penetrate

The London vortex

19

a small distance, , inside the surface of a superconductor, and thus the

field is equal to zero far inside the bulk of a large sample.
A modified form of the London equation was later proposed by Pippard.
This form generalizes the London equation by relating the current at a
point r in the solid, j(r), to the vector potential at nearby points r0 . The
expression he proposed was
Z
R(R.A(r0 )) R/r0 3 0
n s e2 3
d r,
(1.53)
j(r) =
e
me 40
R4
where R = rr0 . The points which contribute to the integral are separated
by distances of order r0 or less, with r0 defined by
1
1
1
=
+ .
r0
0
l

(1.54)

Here l is the mean free path of the electrons at the Fermi surface of the
metal,
l = vF ,
(1.55)
with the scattering time from the Drude conductivity formula, and vF
the electron band velocity at the Fermi surface. The length 0 is called the
coherence length. After the Bardeen Cooper Schrieffer theory of superocnductivity was completed, it became clear that this length is closely related
to the value of the energy gap, , by
0 =

h
vF
.

(1.56)

It also has the physical interpretation that it represents the physical size
of the Cooper pair bound state in the BCS theory.
The existence of the Pippard coherence length implies that a superconductor is characterized by no fewer than three different length scales.
We have the penetration depth, , the coherence length, 0 , and the mean
free path, l. We shall see in the next chapter than the dimensionless ratio = /0 determines whether a superconductor is type I or type II.
Similarly, if the mean free path is much longer than the coherence length ,
l >> 0 the superconductor is said to be in the clean limit, while if l < 0
the superconductor is said to be in the dirty limit. It is a surprising and
very important property of most superconductors that they can remain superconducting even when there are large numbers of impurities making the
mean free path l very short. In fact even many alloys are superconducting
despite the strongly disordered atomic structure.

1.9

The London vortex

We can use the London equation to find a simple mathematical description

of a superconducting vortex, as in Fig. 1.10. The vortex will have a cylindrical core of normal material, with a radius of approximately the coherence

20

Superconductivity

length, 0 . Inside this core we will have a finite magnetic field, say B0 . Outside the vortex core we can use the London equation in the form of Eq. 1.46
to write a differential equation for the magnetic field , B = (0, 0, Bz ). Using cylindrical polar coordinates (r, , z), and the expression for curl in
cylndrical polars, Eq. ?? we obtain (exercise 3.3)
1 dBz
Bz
d2 B z
+
2 = 0.
dr2
r dr

(1.57)

This is a form of Bessels equation (Boas 1983, Matthews and Walker 1970).
The solutions to equations of this type are called modified, or hyperbolic
Bessel functions, K (z) and they can be found in many standard texts of
mathematical physics. In this particular case the solution is K0 (z). The
resulting magnetic field can be written in the form,
Bz (r) =

r
0
K0
2
2

where 0 is the total magnetic flux enclosed by the vortex core,

Z
0 = Bz (r)d2 r.

(1.58)

(1.59)

We shall see in the next chapter that the magnetic flux is quantized, resulting in the universal value 0 = h/2e of flux per vortex line.
For small values of z the function K0 (z) becomes
K0 (z) ln z
(Abromowitz and Stegun, 1965) and so
Bz (r) =

0
ln
22

 

(1.60)

when r << . Using 0 j = B we find (problem 3.3) that the corresponding circulating current is irrotational,
j

1
e
r

(1.61)

exactly as we found earlier for vortices in superfluid helium.

The divergence at r = 0 in these expressions is not physical, and is cut
off by the finite coherence length of the superconductor, 0 . Effectively this
defines a small core size for the vortex (again similar to the vortex core in
superfluid 4 He). Superconductivity is suppressed inside the vortex core, for
r < 0 , which is effectively normal material. Therefore, Eq. 1.60 is valid in
the region 0 << r << , and this simple London vortex model is only
valid in superconductors where 0 << .

Further Reading

21

For the case of large z the modified Bessel function becomes

K0 (z)

z
e
2z

asymptotically (Abromowitz and Stegun, 1965). Therefore the magnetic

field very far from the core of a London vortex is of the form (exercise 3.3)
0
Bz (r) =
22

r/
e
.
2r

(1.62)

Qualitatively this is similar to the penetration of a magnetic field near a

surface as shown in Fig. 1.12.
Overall then, in this London vortex model the magnetic field has some
large constant value B0 inside the vortex core, r < , then decreases logarithmically between 0 < r < and then goes to zero exponentially outside
the vortex on a length scale of order . Clearly this picture is only useful
in the limit > 0 , corresponding to a type II superconductor.
It is also instructive to calculate the energy of the rotating supercurrents
in the vortex. The result6 is that the energy of the vortex is approximately
E=

20
ln
40 2

(1.63)

per unit length.

1.10

Further Reading

To review the basic concepts of band theory of metals, see Band theory
and electronic properties of solids, Singleton (2002), a companion volume
to this book in the Oxford Master Series in Condensed Matter.
There are many text books dealing with superconductivity. Probably
the ones which are especially good for beginners are Supercondctivity Today, Ramakrishnan and Rao 1992, and it Superconductivity and Superfluidity by Tilley and Tilley (1990).
Among the more advanced books, Superconductivity of metals and Alloys, de Gennes (1966), has the most extensive discussion of the topics
covered in this chapter, especially vortices and the vortex lattice.
Bessel functions and their mathematical properties are described in
many texts. Their definitions and propoerties are given in depth by Abromowitz and Stegun (1965). Good introductions are given by Boas (1983)
and Matthews and Walker (1970).
6 See

exercise 3.4 below for the proof.

22

Superconductivity

1.11

Exercises

(3.1) (a) Using the London equation show that

( B) =

1
B
2

in a superconductor.
(b) In Fig. 1.12, the surface of the superconductor lies in the y z plane.
A magnetic field is applied in the z direction parallel to the surface, B =
(0, 0, B0 ). Given that inside the superconductor the magnetic field is a
function of x only, B = (0, 0, Bz (x)) show that
d2 Bz (x)
1
= 2 Bz (x).
2
dx

(c) Solving the ordinary differential equation in (b) show that the magnetic
field near a surface of a superconductor has the form
B = B0 exp (x/)
as shown in Fig. 1.12.
(3.2) Consider a thin superconducting slab, of thickness 2L, as shown in
Fig. 1.13. If an external parallel magnetic field, B0 , is applied parallel to
the slab surfaces, show that inside the slab the magnetic field becomes
Bz (x) = B0

cosh (x/)
.
cosh (L/)

(3.3) (a) A vortex in a superconductor can be modelled as having a cylindrical core of normal metal of radius 0 . Use ( B) = B/2 and
the expression for curl in cylindrical polar coordinates (Eq. ?? to show that
the magnetic field Bz (r) outside of the core obeys the Bessel equation:


1 d
dBz
Bz
r
= 2.
r dr
dr

(b) For small r, obeying < r << , the right hand side of the Bessel
equation in (a) can be approximated by zero. Show that this approximation
leads to
Bz (r) = a ln (r) + b.
where a and b are unknown constants.

Exercises

23

(c) Show that the current corresponding to the field Bz (r) found in (b) is
equal to
a
j=
e
0 r
similar to the superfluid current in a 4 He vortex. Hence find the vector
potential A and find a as a function of the magnetic flux enclosed by the
vortex core, .
(d) For larger values of r (r and above) assume that we can approximate the Bessel equation from (a) by:


d dBz
Bz
= 2.
dr
dr

Hence show that Bz (r) er/ for large r.

(e) The large r solution given in part (d) is not exactly the correct asymptotic form of the solution, as described in section 3.9. For large values of r,
assume that
Bz (r) r p er/
and hence show that the correct exponent is p = 1/2, as described above.
(3.4) Suppose that any supercurrent flow corresponds to an effective superfluid flow velocity v of the electrons, where j = ens v. Assume that the
corresponding kinetic energy is 12 mv 2 ns per unit volume. Hence, using the
results from exercise 3.3 parts (c) and (d), show that the total energy of a
vortex line is roughly of order
 

2
ln
E=
2
40
0
per unit length.
(3.5) The complex conductivity () has real and imaginary parts that are
related together by Kramers-Kronig relations
Z
Im[( 0 )] 0
1
Re[()] = P
d
0

and

1
Im[()] = P

Re[( 0 )] 0
d .
0

Where, here P means the principal value of the integral (Boas 1983,
Matthews and Walker 1970). Therefore an experimental measurement of
the real part is sufficient to determine the imaginary part, and vice versa.

24

Superconductivity

(a) Using these expressions, and assuming that the real part of the conductivity Re() is a Dirac delta function
Re[()] =

ns e2
()
me

show that the imaginary part is given by

Im[()] =

n s e2
me

exactly as given in Eq. 1.41.

(b) Exercise for those who have studied analytic complex function theory. We can derive the Kramers-Kronig relations as follows. Consider the
contour integral
I
( 0 )
I=
d 0
0

around the contour shown in Fig. 1.14. Find the poles of ( 0 ) according
to Eq. 1.38 and show that is is analytic in the upper half plane (Im[ 0 ] > 0)
in Fig. 1.14.
(c) Use the result from (b) to show that I = 0, and thus prove that
Z
( 0 )
0=P
d 0 i() = 0,
0

where the integral is now just along the real 0 axis. Take the real and imaginary parts of this expression and show that this results in the Kramers
Kronig equations given above.7

7 The proof is actually very general. The fact that ( 0 ) is analytic in the upper
half plane is in fact just a consequence of causality, i.e. the applied current always
responds to the applied external field. Effect follows cause, never the reverse! Therefore
the Kramers-Kronig relations are always true for any such response function.

Exercises

25

non-superconducting

superconducting

Tc

Fig. 1.1 Resistivity of a typical metal as a function of temperature. If it is a

non-superconducting metal (such as copper or gold) the resistivity approaches
a finite value at zero temperature, while for a superconductor (such as lead, or
mercury) all signs of resistance disappear suddenly below a certain temperature,
Tc .

(mcm)
30

20

10

100

150

200

250

300

T (K)

Fig. 1.2 Resistivity of HgBa2 Ca2 Cu3 O8 + as a function of temperature

(adapted from data of Chu (1993). Zero resistance is obtained at about 135K,
the highest known Tc in any material at normal pressure. In this material Tc
approaches a maximum of about 165K under high pressure. Note the rounding of
the resistivity curve just above Tc , which is due to superconducting fluctuation
effects. Also, well above Tc the resistivity does not follow the expected Fermi
liquid behaviour.

26

Superconductivity

(a)

(b)

I
V

Fig. 1.3 Measurement of resistivity by (a) the two terminal method, (b) the
four terminal method. The second method, (b), is much more accurate since no
current flows through the leads measuring the voltage drop across the resistor,
and so the resistances of the leads and contacts is irrelevant.

Fig. 1.4 Persistent current around a superconducting ring. The current maintains a constant magnetic flux, , through the superconducting ring.

Exercises

T > Tc

T < Tc

B=0

T > Tc

B=0

T < Tc

27

B 6= 0

B 6= 0

Fig. 1.5 The Meissner-Ochsenfeld effect in superconductors. If a sample initially

at high temperature and in zero magnetic field (top) is first cooled (left) and then
placed in a magnetic field (bottom), then the magnetic field cannot enter the
material (bottom). This is a consequence of zero resistivity. On the other hand
a normal sample (top) can be first placed in a magnetic field (right) and then
cooled (bottom). In the case the magnetic field is expelled from the system.

H = IN/L

N/L coil turns per metre

Fig. 1.6 Measurement of M as a function of H for a sample with solenoidal

geometry. A long solenoid coil of N/L turns per metre leads to a uniform field
H = IN/L Amperes per metre inside the solenoid. The sample has magnetization, M , inside the solenoid, and the magnetic flux density is B = 0 (H + M ).
Increasing the current in the coils from I to I + dI, by dI leads to an inductive
e.m.f. E = d/dt where = N BA is the total magnetic flux threading the N
current turns of area A. This inductive e.m.f. can be measured directly, since it is
simply related to the differential self-inductance of the coil, L, via E = LdI/dt.
Therefore, by measuring the self-inductance L one can deduce the B-field and
hence M as a function of I or H.

28

Superconductivity

1
Tc

Fig. 1.7 Magnetic susceptibility, , of a superconductor as a function of temperature. Above Tc it is a constant normal state value, n , which is usually small and
positive (paramagetic). Below Tc the susceptibility is large and negative, = 1,
signifying perfect diamagnetism.

Type I

Type II
Hc

Hc1

M = H

Hc2
H

M = H

Fig. 1.8 The magnetization M as a function of H in type I and type II superconductors. For type I perfect Meissner diamagnetism is continued until Hc ,
beyond which superconductivity is destroyed. For type II materials perfect diamagnetism occurs only below Hc1 . Between Hc1 and Hc2 Abrikosov vortices enter
the material, which is still superconducting.

Exercises

29

Type II

Type I
H

Abrikosov

Hc2

Hc

Hc1

Tc

Meissner

Tc

Fig. 1.9 The H T phase diagram of type I and type II superconductors. In

type II superconductors the phase below Hc1 is normally denoted the Meissner
state, while the phase between Hc1 and Hc2 is the Abrikosov or mixed state.

Fig. 1.10 Vortices in a type II superconductor. The magnetic field can pass
through the superconductor, provided it is channelled through a small vortex
core. The vortex core is normal metal. This allows the bulk of the material to
remain superconducting, while also allowing a finite average magnetic flux density
B to pass through.

Re()

/h

ns e2 ()/m

/h

Fig. 1.11 The finite frequency conductivity of a normal metal (dashed line) and
a superconductor (solid line). In the superconducting case an energy gap leads
to zero conductivity for frequencies below /
h. The remaining spectral weight
becomes concentrated in a Dirac delta function at = 0.

30

Superconductivity

B
B0

Fig. 1.12 The magnetic field near a surface of a superconductor in the Meissner
state. The field decays exponentially on a length scale given by the penetration
depth .

B
B0

B0

Fig. 1.13 Exercise 3.2. The magnetic field inside a superconducting slab of hickness 2L.

Im[ 0 ]

i 1

Re[ 0 ]

Fig. 1.14 Complex integration contour for Exercise 3.5.

2
The Ginzburg-Landau model
2.1

Introduction

The superconducting state and the normal metallic state are separate thermodynamic phases of matter in just the same way as gas, liquid and solid
are different phases. Similarly, the normal Bose gas and BEC, or normal
liquid He4 and superfluid He II are separated by a thermodynamic phase
transitions. Each such phase transition can be characterized by the nature
of the singularities in specific heat and other thermodynamic variables at
the transition, Tc . We can therefore examine the problems of superfluidity
and superconductivity from the point of view of the thermodynamics of
phase transitions.
The theory of superconductivity introduced by Ginzburg and Landau
in 1950 describes the superconducting phase transition from this thermodynamic point of view. It was originally introduced as a phenomenological
theory, but later Gorkov showed that it can be derived from full the microscopic BCS theory in a suitable limit.
1

In this chapter we shall first discuss the superconducting phase transition from the point of view of equilibrium thermodynamics. Then we
gradually build up towards the full Ginzburg Landau model. First we discuss spatially uniform systems, then spatially varying systems and finally
systems in an external magnetic field. The Ginzburg Landau theory makes
many useful and important predictions. Here we focus on just two applications: to flux quantization, and to the Abrikosov flux lattice in type II
superconductors.
The Ginzubrg Landau theory as originally applied to superconductors
was par excellance a mean-field theory of the thermodynamic state. However, in fact, one of its most powerful features is that it can be used to go
beyond the original mean-field limit, so as to include the effects of thermal
fluctuations. We shall see below that such fluctuations are largely negligible in the case of conventional low-Tc superconductors, making the
1 In fact, the Ginzburg Landau model is very general and has applications in many
different areas of physics. It can be modified to describe many different physical systems,
including magnetism, liquid crystal phases and even the symmetry breaking phase transitions which took place in the early universe as matter cooled following the big bang!

32

The Ginzburg-Landau model

mean-field approximation essentially exact. However in the newer high T c

superconductors these fluctuations lead to many important phenomena,
such as flux flow, and the melting of the Abrikosov vortex lattice.

2.2

The condensation energy

We already have enough information about superconductivity to derive

some important thermodynamic properties about the superconducting phase
transition. We can analyze the phase diagram of superconductors in exactly
the same manner as one would consider the well known thermodynamics of
a liquid gas phase transition problem, such as given by the van der Waals
equation of state. However, for the superconductor instead of the pair of
thermodynamic variables P , V (pressure and volume) we have the magnetic
variables H and M as the relevant thermodynamic parameters.
Let us first briefly review the basic thermodynamics of magnetic materials. This is covered in several undergraduate text books on thermodynamics
such as Mandl (1987) or Callen (1960), Blundell (2001). If we consider a
long cylindrical sample in a solenoidal field, as shown in Fig. 1.6, then the
magnetic field H inside the sample is given by
H=

N
Iez ,
L

(2.1)

where the coil has N/L turns per metre, I is the current and ez is a unit
vector along the axis of the cylinder. The total work done, d-W , on increasing the current infinitesimally from I to I + dI can be calculated as
Z
d-W = N EIdt
Z
d
= +N
Idt
dt
= +N Id
=
=

+N AIdB
+N V H.dB

+0 V (H.dM + H.dH)

(2.2)

where A is the cross sectional area of the coli, V = AL is its volume, E =

d/dt is the e.m.f. induced in the coil by the change in the total magnetic
flux, , through the sample. We also used the identity B = 0 (M + H) in
writing the last step in Eq. 2.2.
This analysis shows that we can divide the total work done by increasing
the current in the coil into two separate parts. The first part,
0 H.dM
per unit volume, is the magnetic work done on the sample. The second
part,

The condensation energy

33

0 H.dH
is the work per unit volume which would have been done even if no sample
had been present inside the coil; it is the work done by the self-inductance
of the coil. If the coil is empty, M = 0 and so B = 0 H and one can easily
see that the work done is exactly the change in the vacuum field energy of
the electromagnetic field
Z
1
B 2 d3 r
(2.3)
EB =
20
due to the change of current in the solenoid coils. By convention2 we shall
not include this vacuum field energy, as work done on the sample. Therefore we define the magnetic work done on the sample as 0 HdM per unit
volume.
With this definition of magnetic work the first law of thermodynamics
for a magnetic material reads,
dU = T dS + 0 V H.dM

(2.4)

where U is the total internal energy, T dS is the heat energy with T the
temperature and S the entropy. We see that the magnetic work is analogous
to the work, P dV , in a gas. As in the usual thermodynamics of gases the
internal energy, U , is most naturally thought of as a function of the entropy
and volume: U (S, V ). The analogue of the first law for a magnetic system,
Eq. 2.4, shows that the internal energy of a magnetic substance is most
naturally thought of as a function of S and M, U (S, M). In terms of this
function the temperature and field H are given by
T

U
S
1 U
.
0 V M

(2.5)
(2.6)

However S and M are usually not the most convenient variables to work
with. In a solenoidal geometry such as Fig. 1.6 it is the H-field which is
directly fixed by the current, not M. It is therefore useful to define magnetic
analogues of the Helmholtz and Gibbs free energies
F (T, M)

G(T, H)

U TS

U T S 0 V H.M.

(2.7)
(2.8)

As indicated, the Gibbs free energy G is naturally viewed as a function of

T and H since,
2 Unfortunately there is no single standard convention used by all books and papers
in this field. Different contributions to the total energy are either included or not, and so
one must be very careful when comparing similar looking equations from different texts
and research papers. Our convention follows Mandl (1987) and Callen (1960).

34

The Ginzburg-Landau model

dG = SdT 0 V M.dH.

(2.9)

In terms of G one can calculate the entropy and magnetization,

S

G
T
1 G

.
0 V H

(2.10)
(2.11)

G(T, H) is usually the most convenient thermodynamic quantity to work

with since T and H are the variables which are most naturally controlled
experimentally. Furthermore from G(T, H) one can also reconstruct the
free energy, F = G + 0 V H.MV or the internal energy U = F + T S.
The Gibbs free energy allows us to calculate the free energy difference
between the superconducting state and the normal state. Consider the H,
T phase diagram of a type I superconductor, as sketched above in Fig. 2.1.
We can evaluate the change in Gibbs free energy in the superconducting
state by integrating along the vertical line drawn. Along this line dT = 0,
and so, clearly,
Z Hc
Z Hc
M.dH,
dG = 0 V
Gs (T, Hc ) Gs (T, 0) =
0

where the subscript s implies that G(T, H) is in the superconducting state.

But for a type I superconductor in the superconducting state we know from
the Meissner-Ochsenfeld effect that M = H and thus,

Hc2
V.
2
Now, at the critical field Hc in Fig. 2.1 the normal state and the superconducting state are in thermodynamic equilibrium. Equilibrium between
phases implies that the two Gibbs free energies are equal:
Gs (T, Hc ) Gs (T, 0) = 0

Gs (T, Hc ) = Gn (T, Hc ).
Furthermore, in the normal state M 0 (apart from the small normal
metal paramagnetism or diamagnetism which we neglect). So if the normal
metal state had persisted below Hc down to zero field, it would have had
a Gibbs free energy of,
Z Hc
Z Hc
M dH 0.
dG = 0 V
Gn (T, Hc ) Gn (T, 0) =
0

Putting these together we find the difference in Gibbs free energies of

superconducting and normal states at zero field:
Hc2
(2.12)
2
The Gibbs potential for the superconducting state is lower, so it is the
stable state.
Gs (T, 0) Gn (T, 0) = 0 V

The condensation energy

35

We can also write the above results in terms of the more familiar
Helmholtz free energy. Using F = G 0 V H.M and substituting H =
M = 0 we can see that the difference in Helmholtz free energies F (T, M)
is the same as for the Gibbs potentials, and hence
Fs (T, 0) Fn (T, 0) = 0 V

Hc2
.
2

(2.13)

The quantity 0 Hc2 /2 is the condensation energy. It is a measure of the

gain in free energy per unit volume in the superconducting state compared
to the normal state at the same temperature.
As an example lets consider niobium. Here Tc = 9K, and Hc = 160kA/m
(Bc = 0 Hc = 0.2T ). The condensation energy 0 Hc2 /2 = 16.5kJ/m3 .
Given that Nb has a bcc crystal structure with a 0.33nm lattice constant
we can work out the volume per atom and find that the condensation energy
is only around 2eV/atom! Such tiny energies were a mystery until the BCS
theory, which shows that the condensation energy is of order (kB Tc )2 g(EF ),
where g(F ) is the density of states at the Fermi level. The energy is so
small because kB Tc is many orders of magnitude smaller than the Fermi
energy, F .
The similar thermodynamic arguments can also be applied to calculate
the condensation energy of type II superconductors. Again the magnetic
work per unit volume is calculated as an integral along a countour, as shown
in the right panel of Fig. 2.1,
Gs (T, Hc2 ) Gs (T, 0) = 0 V

Hc2

M.dH

(2.14)

The integral is simply the area under the curve of M as a function of H

drawn in Fig. 1.8 (assuming that M and H have the same vector directions).
Defining the value of Hc for a type II superconducting from the value of
the integral
Z Hc2
1 2
M.dH
(2.15)
Hc
2
0
we again can express the zero field condensation energy in terms of Hc ,
Fs (T, 0) Fn (T, 0) = 0 V

Hc2
.
2

(2.16)

Here Hc is called the thermodynamic critical field. Note that there is no

phase transition at Hc in a type II superconductor. The only real transitions are at Hc1 and Hc2 , and Hc is merely a convenient measure of the
condensation energy.
We can also calculate the entropy of the superconducting state using
the same methods. A simple calculation (Exercise 4.1) shows that in a type

36

The Ginzburg-Landau model

I superconductor there is a finite change in entropy between the normal and

superconducting states at Hc ,
Ss (T, Hc ) Sn (T, Hc ) = 0 Hc

dHc
.
dT

(2.17)

This shows that the phase transition is generally first-order, i.e. it has a
finite latent heat. But, in zero external field, at the point (T, H) = (Tc , 0)
in Fig. 2.1, this entropy difference goes to zero, and so in this case the phase
transition is second-order.

2.3

Ginzburg-Landau theory of the bulk phase transition

The Ginzburg-Landau theory of superconductivity is built upon a general

approach to the theory of second order phase transitions which Landau
had developed in the 1930s. Landau had noticed that typically second
order phase transitions, such as the Curie temperature in a ferromagnet,
involve some change in symmetry of the system. For example a magnet
above the Curie temperature, Tc , has no magnetic moment. But below Tc
a spontaneous magnetic moment develops. In principle could point in any
one of a number of different directions, each with an equal energy, but the
system spontaneously chooses one particular direction. In Landaus theory
such phase transitions are characterized by an order parameter which is
zero in the disordered state above Tc , but becomes non-zero below Tc . In the
case of a magnet the magnetization, M(r), is a suitable order parameter.
For superconductivity Ginzburg and Landau (GL) postulated the existence of an order parameter denoted by . This characterizes the superconducting state, in the same way as the magnetization does in a ferromagnet.
The order parameter is assumed to be some (unspecified) physical quantity
which characterizes the state of the system. In the normal metallic state
above the critical temperature Tc of the superconductor it is zero. While in
the superconducting state below Tc it is non-zero. Therefore it is assumed
to obey:

0
T > Tc
(2.18)
=
(T ) 6= 0 T < Tc .

Ginzburg and Landau postulated that the order parameter should be

a complex number, thinking of it as a macroscopic wave function for the
superconductor in analogy with superfluid 4 He. At the time of their original
work the physical significance of this complex in superconductors was not
at all clear. But, as we shall see below, in the microscopic BCS theory of
superconductivity there appears a parameter, , which is also complex.
Gorkov was able to derive the Ginzburg-Landau theory from BCS theory,
and show that is essentially the same as , except for some constant
numerical factors. In fact, we can even identify ||2 as the density of BCS
Cooper pairs present in the sample.
Ginzburg and Landau assumed that the free energy of the superconductor must depend smoothly on the parameter . Since is complex and the

Ginzburg-Landau theory of the bulk phase transition

37

free energy must be real, the energy can only depend on ||. Furthermore,
since goes to zero at the critical temperature, Tc , we can Tailor expand
the free energy in powers of ||. For temperatures close to Tc only the first
two terms in the expansion should be necessary, and so the free energy
density (f = F/V ) must be of the form:
1
fs (T ) = fn (T ) + a(T )||2 + b(T )||4 + . . .
(2.19)
2
since || is small. Here fs (T ) and fn (T ) are the superconducting state and
normal state free energy densities, respectively. Clearly Eq. 2.19 is the only
possible function which is real for any complex near = 0 and which
is a differentiable function of and near to = 0. The parameters
a(T ) and b(T ) are, in general, temperature dependent pheonomenological
parameters of the theory. However it is assumed that they must be smooth
functions of temperature. We must also assume that b(T ) is positive, since
otherwise the free energy density would have no minimum, which would
be unphysical (or we would have to extend the expansion to include higher
powers such as ||6 ).
Plotting fs fn as a function of is easy to see that there are two
possible curves, depending on the sign of the parameter a(T ), as shown in
Fig. 2.2. In the case a(T ) > 0, the curve has one minimum at = 0. On the
other hand, for a(T ) < 0 there are minima wherever ||2 = a(T )/b(T ).
Landau and Ginzburg assumed that at high temperatures, above Tc , we
have a(T ) positive, and hence the minimum free energy solution is one
with = 0, i.e. the normal state. But if a(T ) gradually decreases as the
temperature T is reduced, then the state of the system will change suddenly
when we reach the point a(T ) = 0. Below this temperature the minimum
free energy solution changes to one with 6= 0. Therefore we can identify
the temperature where a(T ) becomes zero as the critical temperature Tc .
Near to this critical temperature, Tc , assuming that the coefficients
a(T ) and b(T ) change smoothly with temperature, we can make a Taylor
expansion,
a(T )
b(T )

a (T Tc ) + . . .
b + ...,

(2.20)

where a and b are two pheonomenological constants. Then for temperatures

just above Tc , a(T ) will be positive, and we have the free energy minimum,
= 0. On the other hand, just below Tc we will have minimum energy
solutions with non-zero ||, a s seen in Fig. 2.2. In terms of the parameters
a and b it is easy to see that
(


a 1/2
(Tc T )1/2 T < Tc .
b
(2.21)
|| =
0
T > Tc
The corresponding curve of || as a function of temperature, T , is shown
in Fig 2.3. One can see the abrupt change from zero to non-zero values at

38

The Ginzburg-Landau model

the critical temperature Tc . In fact, this curve is qualitatively similar to

those obtained with other types of second order phase transitions within
Landaus general theory. For example the behaviour of the order parameter
near Tc in Fig. 2.3 resembles closely change in the magnetization M
in a ferromagnet near its Curie point in the Stoner theory of magnetism
(Blundell 2001).
It turns out to be very important that, because is complex, there are
in fact an infinite set of minima corresponding to all possible values of the
complex phase ,
= ||ei .
(2.22)

The phase value, is arbitrary, since all values lead to the same total free
energy. But, just as in the case of the direction of magnetization M in
a ferromagnet the system spontaneously chooses one particular value. A
magnet heated to above Tc and then cooled again will almost certainly
adopt a different random direction of magnetization, and the same would
be true for the angle in a superconductor. In fact we have met this same
concept before, in Chapter 2, when we discussed the XY symmetry of the
macroscopic wave function in superfluid He II (Fig. ??).
The value of the minimum free energy in Fig. 2.2, is easily found to
be a(T )2 /2b(T ). This is the free energy difference (per unit volume) between the superconducting and non-superconducting phases of the system
at temperature T . This corresponds to the condensation energy of the superconductor, and so we can writ e
fs (T ) fn (T ) =

H2
a 2 (T Tc )2
= 0 c ,
2b
2

(2.23)

giving the thermodynamic critical field,

Hc =

a
(Tc T )
(0 b)1/2

(2.24)

near to Tc .
From this free energy we can also obtain other relevant physical quantities, such as the entropy and heat capacity. Differentiating f with respect
to T gives the entropy per unit volume, s = S/V ,
ss (T ) sn (T ) =

a 2
(Tc T ),
b

(2.25)

below Tc . At Tc there is no discontinuity in entropy, or latent heat, confirming that the Ginzburg-Landau model corresponds to a second order thermodynamic phase transition. But there is a sudden change in specific heat
at Tc . Differentiating the entropy to find the heat capacity CV = T ds/dT
per unit volume we obtain
 a 2
T b T < Tc
CV s C V n =
(2.26)
0
T > Tc

Ginzburg-Landau theory of inhomogenous systems

39

and so the heat capacity has a discontinuity

CV = Tc

a 2
b

(2.27)

at Tc . The metallic normal state heat capacity is linear in T , CV n = T ,

with the Sommerfeld constant, and so the full heat capacity curve looks
like Fig. 2.4 near to Tc .3
Interestingly the specific heat for superconductors shown in Fig. 2.4
is qualitatively quite very different from both the case of Bose-Einstein
condensation, shown in Fig. ??, and the point of superfluid 4 He, Fig. ??.4

2.4

Ginzburg-Landau theory of inhomogenous systems

The complete Ginzburg and Landau theory of superconductivity also allows

for the possibility that the order parameter depends on position, (r). This,
of course, now really begins to resemble the macroscopic condensate wave
function introduced in chapter 2 for the case of superfluid helium.
Ginzburg and Landau postulated that the Free energy is as given above,
together with a new term depending on the gradient of (r). With this term
free energy density becomes,
fs (T ) = fn (T ) +

b(T )
h
2
|(r)|2 + a(T )|(r)|2 +
|(r)|4
2m
2

(2.28)

at point r in the absence of any magnetic fields. Setting (r) to a constant

value, (r) = , we see that the parameters a(T ) and b(T ) are the same
as for the bulk theory described in the previous section. The new parameter m determines the energy cost associated with gradients in (r). It
has dimensions of mass, and it plays the role of an effective mass for the
quantum system with macroscopic wave function (r).
In order to find the order parameter (r) we must minimize the total
free energy of the system,

 2
Z
h

b(T )
2
2
4
3
|| + a(T )|(r)| +
|(r)| d3 r.
Fs (T ) = Fn (T ) + d r
2m
2
(2.29)
To find the minimum we must consider an infinitesimal variation in the
function (r)
3 The Ginzburg Landau theory can only be reliably used at temperatures close to
Tc . Therefore our calculated specific heat is only correct near to Tc , and we cannot
legitimately continue the Ginzburg-Landau line in Fig. 2.4 down from T c all the way to
T = 0.
4 In fact the differences are deceptive! Our theory rests on a mean-field approximation
and has neglected important thermal fluctuation effects, as we shall see below. When
these fluctuations are large, as in the case of high temperature superconductors, the
observed specific heat near Tc appears to show exactly the same XY universality class
as the lambda point in superfluid helium (Overend 1994).

40

The Ginzburg-Landau model

(r) (r) + (r)

(2.30)

relative to some function (r). Evaluating the change in the total free
energy due to and dropping all terms of higher than linear order in the
variation we find after some lengthy algebra

Z  2
h

2
Fs =
( ).() + (a + b| |) d3 r
2m

Z  2
h

2
( ).() + (a + b | |) d3 r. (2.31)
+
2m
The two terms involving gradients can be integrated by parts, to obtain


Z
h
2
Fs =
2 + a + b| 2 | d3 r
2m

Z 
h
2 2
2
+
+ a + b| | d3 r.
(2.32)
2m
The condition for (r) to produce a minimum in free energy is that F = 0
for any arbitrary variation (r). From Eq. 2.32 this can only be when (r)
obeys
h
2
(2.33)
2 + a + b| 2 | = 0.
2m
We can obtain this same result more formally by noting that the total
Free energy of the solid is a functional of (r), denoted Fs [], meaning that
the scalar number Fs depends on the whole function (r) at all points in
the system, r. It will be minimized by a function (r) which satisfies
Fs []
=0
(r)

Fs []
= 0.
(r)

(2.34)

where the derivatives are mathematically functional derivatives. Functional

derivative can be defined by analogy with the idea of a partial derivative.
For a function of many variables, f (x1 , x2 , x3 , . . .) we can express changes
in the function value due to infinitesimal variations of the parameters using
the standard expression
df =

f
f
f
dx1 +
dx2 +
dx3 + . . . .
x1
x2
x3

(2.35)

Considering the free energy as a function of infinitely many variables, (r)

and (r) at all possible points r we can write the analogue of Eq. 2.35 as,

Z 
Fs []
Fs []
dFs =
d(r) +
d (r) d3 r.
(2.36)
(r)
(r)
In comparison with Eq. 2.32 we see that

Surfaces of Superconductors

and

41

Fs []
h
2 2
=

+ a(T ) + b(T )| 2 |
(r)
2m

(2.37)

Fs []
h
2 2
= (
+ a(T ) + b(T )| 2 |)
(r)
2m

(2.38)

which is just the complex conjugate of Eq. 2.37. Perhaps it seems surprising
that we can effectively treat (r) and (r) as independent variables in the
differentiation, but this is correct because there are two independent real
functions, Re[(r)] and Im[(r)], which can be varied separately.
Thus we have found that minimizing the total Free energy leads to the
following Schrodinger like equation for (r),

h
2 2
(r) + a + b|(r)|2 (r) = 0.

2m

(2.39)

However, unlike the usual Schrodinger equation, this is a non-linear equation because of the second term in the bracket. Because of this non-linearity
the quantum mechanical principle of superposition does not apply, and the
normalization of is different from the usual one in quantum mechanics.

2.5

Surfaces of Superconductors

The effective non-linear Schodinger equation, Eq. 2.39, has several useful applications. In particular, it can be used to study the response of
the superconducting order parameter to external perturbations. Important
examples of this include the properties of the surfaces and interfaces of
superconductors.
Consider a simple model for the interface between a normal metal and a
superconductor. Suppose that the interface lies in the yz plane separating
the normal metal in the x < 0 region from the superconductor in the x > 0
region. On the normal metal side of the interface the superconducting order
parameter, (r), must be zero. Assuming that (r) must be continuous,
we must therefore solve the non-linear Schrodinger equation,

h
2 d2 (x)
+ a(T )(x) + b(T ) 3 (x) = 0
2m dx2

(2.40)

in the region x > 0 with the boundary condition at (0) = 0. It turns out
that one can solve this equation directly (exercise 4.2) to find


x
(x) = 0 tanh
,
(2.41)
2(T )
as shown in Fig. 2.5. Here 0 is the value of the order parameter in the
bulk far from the surface and the parameter (T ) is defined by

42

The Ginzburg-Landau model

(T ) =

h
2

2m |a(T )|

1/2

(2.42)

This quantity, which has dimensions of length, is called the GinzburgLandau coherence length. It is an important physical parameter characterizing the superconductor. In Fig. 2.5 one can see that (T ) is a measure
of the distance from the surface over which the order parameter has recovered back to nearly its bulk value.
The Ginzburg-Landau coherence length arises in almost all problems
of inhomogenous superconductors, including surfaces, interfaces, defects
and vortices. Using a(T ) = a(T

Tc ) the coherence length (T ) can be

rewritten,
(T ) = (0)|t|1/2 ,
(2.43)
where
t=

T Tc
Tc

(2.44)

is called the reduced temperature. This expression makes it clear that the
coherence length (T ) diverges at the critical temperature Tc , and that its
divergence is characterized by a critical exponent of 1/2. This exponent is
typical for mean-field theories such as the Ginzburg-Landau model. The
zero temperature value of , (0), is apart from some numerical factors
of order unity, essentially the same as the Pippard coherence length for
superconductors, as introduced in Chapter 3. In BCS theory the coherence
length relates to the physical size of a single Cooper pair.
It is also possible to calculate the contribution to the total free energy
due to the surface in Fig. 2.5. The surface contribution to the total free
energy is
!
2

Z
b 4
1
h
2
d(x
2
2
(2.45)
+ a (x) + (x) + 0 Hc dx
=
2m
dx
2
2
0
with (x) given by Eq. 2.41. Here 0 Hc2 /2 = a2 /2b is the bulk free
energy density. Evaluating the integral (de Gennes 1960) gives
=

1
0 Hc2 1.89(T )
2

(2.46)

free energy per unit area of the surface.

This theory can also be used to model the proximity effect between
two superconductors. At an interface between two different superconducting materials the one with the higher Tc will become superconducting first,
and will nucleate superconductivity at the surface of the second one. Superconductivity will nucleate at temperatures above the Tc for the second
superconductor. If one makes the lower Tc superconductor a thin layer, of
order the coherence length (T ) in thickness, then the whole system will

Ginzburg-Landau theory in a magnetic field

43

become superconducting at a temperature above the natural critical temperature of the lower Tc material. Effectively the order parameter (r) has
been forced to become non-zero in the thin film material by its proximity
to the higher Tc material.

2.6

Ginzburg-Landau theory in a magnetic field

The full power of the GL approach to superconductors only becomes apparent when we include one final term, the effect of a magnetic field. It is
this which truly shows that this is a fully fledged theory of superconductivity, complete with Meissner-Ochsenfeld effect, London equation and so on.
Effectively, the Ginzburg-Landau theory as developed given in the previous
sections did not include any effects of the charge of the superconducting
condensate. Therefore it would be appropriate for systems of neutral particles, such as a superfliud, or for situations where there are no supercurrents.
But in the presence of supercurrents of charged particles we must extend
the theory to include the interaction of the current and magnetic field.
What is needed is to include the effects of magnetic fields in the free
energy. Ginzburg and Landau postulated that the magnetic field enters
as if (r) were the wave function for charged particles, i.e with the usual
replacement in quantum mechanics
h

qA
i
i

(2.47)

where q is the charge and A is the magnetic vector potential. For all known
superconductors it turns out that the appropriate charge q is 2e. Why this
is the case only became clear after the BCS theory was developed, and the
link between the BCS theory and the Ginzburg-Landau model had been
explained by Gorkov. He showed that the correct physical interpretation
of the GL order parameter (r) is that it can be understood as the wave
function for the centre of mass motion of Cooper pairs of electrons. Since
each Cooper pair has a net charge of 2e, then this is is the correct effective
charge q.5 Actually the sign can equally well be taken as q = +2e, since we
can think of Cooper pairs of holes as readily as pairs of electrons. In fact
no observable effects in the Ginzburg-Landau theory differ when we take a
different convention for the sign.
With this replacement the Ginzburg-Landau free energy density of the
superconductor becomes,
fs (T ) = fn (T ) +

h
2 h

b
|( + 2eA)|2 + a||2 + ||4 .
2m i
2

(2.48)

5 In fact, in their original paper Ginzburg and Landau assumed that the effective
charge would be e, not 2e. Reputedly, Ginzburg, then a young PhD student, told his
famous advisor Landau that 2e fitted the available experimental data better than e, but
Landau overruled him and insisted that he was sure that it must be e!

44

The Ginzburg-Landau model

To obtain the total free energy we must integrate this over the system, but
we must also include an additional term, corresponding to the electromagnetic field energy of the field B(r) = A at each point r. Therefore the
total free energy of both the superconductor and the magnetic field is,6

Z  2
h

b 4 3
2
2
|(

+
2eA)|
+
a||
+
||
Fs (T ) = Fn (T ) +
d r
2m i
2
Z
1
B(r)2 d3 r.
(2.49)
+
20
The first integral is carried out over points r inside the sample, while the
second is performed over all space.
The condition for the minimum free energy state is again found by
performing a functional differentiation to minimize with respect to (r)
and (r). The resulting equation for (r) is again a non-linear Schrodinger
equation, but now with a term containing the magnetic vector potential A,

2ei 2
h
2
( +
A) (r) + a + b||2 (r) = 0.

2m
h

(2.50)

The supercurrents due to the magnetic field can be found from functional derivative of the Ginzburg Landau superconductor free energy with
respect to the vector potential,
Fs
A(r)

(2.51)

(2e)2 2
2ehi
( )
|| A.

2m
m

(2.52)

js =
which leads to the supercurrent
js =

Note the close similarity to the superfluid current flow that we found earlier
in the case of 4 He, Eq. ??. The differences from Eq. ?? are firstly the
charge of the condensate particles, 2e, and the presence of the last ter
m which provides the effect of the vector potential A. Finally, the vector
potential must be obtained from the magnetic field arising from both the
supercurrents and any other currents, such as the external currents, jext ,
in the solenoid coils of Fig. 1.6,
B = 0 (jext + js ),

(2.53)

as given by Maxwells equations.

6 Note that in the definition of the magnetic work given in Eq. 2.4, we excluded the
part of this field energy, 0 H 2 /2 per unit volume, that would be present even with
no sample present inside the coil in Fig. 1.6. From now on, it will be more convenient
to include this energy explicitly, making Fs the total free energy of both sample and
vacuum fields.

Gauge Symmetry and Symmetry Breaking

2.7

45

Gauge Symmetry and Symmetry Breaking

The Ginzburg-Landau order parameter for superconductors has both an

amplitude and a complex phase
(r) = |(r)|ei(r) .

(2.54)

This is similar to the macroscopic wave function for superfluid He II, introduced in Chapter 2. However, unlike superfluids of neutral particles ,
something very interesting happens now when we consider gauge invariance.
If we make a gauge transformation of the magnetic vector potential
A(r) A(r) + (r)

(2.55)

then we must make a corresponding change in the phase of the order parameter, . Consider the term in the Ginzburg-Landau free energy density
containing the canonical momentum operator
p =

+ 2eA.
i

If we change the phase of the order parameter by

(r) (r)ei(r)

(2.56)

then we obtain
p(r)ei(r)

=
=


h

+ 2eA (r) + (r)ei(r) h

(r)
i


h

+ 2e(A + ) (r).
(2.57)
ei(r)
i
2e

ei(r)

From this it follows that the free energy will be unchanged when we simultaneously change (r) to (r)ei(r) and the vector potential according
to,
h

A(r) A(r) + .
(2.58)
2e
This shows that the theory satisfies local gauge invariance. Both the phase
of the order parameter and the magnetic vector potential depend on the
choice of gauge, but all physical observables (free energy, magnetic field B
etc.) are gauge invariant.
So far this is all completely general. But we saw earlier that a bulk
superconductor has a ground state with a constant order parameter, .
Therefore it must have the same everywhere. There must be a phasestiffness, or an energy cost associated with changing from one part of
the solid to another. If we consider a superconductor in which the order
parameter has a constant magnitude, ||, and a phase (r) which varies

46

The Ginzburg-Landau model

only slowly with position r, then (using Eq. 2.57) we obtain the total free
energy

2
Z
2e
Fs = Fs0 + s d3 r + A .
(2.59)
h

Here the superfluid stiffness is defined by,

s =

h
2
||2
2m

(2.60)

and Fs0 is the total free energy in the ground state ( = constant, A = 0).
Now if we choose some particular gauge for A(r), such as the London
gauge, .A = 0, then within this fixed gauge there is now a free energy
cost associated with further gradients in (r). To minimize the gradient
energy, we must minimize the gradients, by making (r) as constant as
possible throughout the system. In the case of zero applied magnetic field,
we can choose A = 0, and clearly then (r) will be constant everywhere in
the system. Again we are back to the XY symmetry of Fig. ??. Since the
system effectively chooses an (arbitrary) constant order parameter phase
everywhere in the system, we can say that the system exhibits long ranged
order in the order parameter phase, just as a ferromagnet has long ranged
order in its magnetization M(r).
Because the long ranged order is in the phase variable (which is not
normally a physical observable in quantum mechanics) we say that the
system has spontaneously broken global gauge symmetry. The point is
that global gauge symmetry refers to changing (r) by a constant amount
everywhere in the whole solid (which does not require any change in A).
This is in contrast to local gauge symmetry in which (r) and A(r) are
changed simultaneously, consistent with Eq. 2.58.
Eq. 2.59 also implies the London equation, and hence the MeissnerOchsenfeld effect, bringing us full circle back to Chapter 3. The current
can be calculated from a functional derivative of the free energy
js

=
=

Fs [A]
A(r)


2e
2e
s + A .
h

(2.61)

Starting in the ground state, where is constant, we directly find that with
a small constant external vector potential, A, the current is,
js = s

(2e)2
A
h
2

(2.62)

which is exactly the same as the London equation. The superfluid stiffness,
s , is essentially just the London superfluid density, ns , in disguise!

Flux quantization

47

To make the connection between s and the London superfluid fraction,

ns , more clear, consider the London equation
n s e2
A.
me

(2.63)

(2e)2 2
|| A,
2m

(2.64)

js =
If we rewrite Eq. 2.61 in the form
js =

then clearly these are the same. It is conventional to define the constants
so that the London superfluid density is ns = 2||2 and the GinzburgLandau effective mass is m = 2me (where me is the bare electron mass).
With this choice the equation can be interpreted physically as implying
that ||2 is the density of pairs of electrons in the ground state. Therefore
in comparison with the BCS theory of superconductivity we can interpret
||2 with the density of Cooper pairs in the ground state, and ns as the
density of electrons belonging to these Cooper pairs. The normal fraction,
nn = nns coresponds to the density of unpaired electrons. The GinzburgLandau parameter m is the mass of the Cooper pair, which is naturally
twice the original electron mass.
In terms of the original free energy Ginzburg-Landau parameters, a and
b, the superfluid density, ns is given by
ns = 2| 2 | = 2

a(T
c T)
.
b

(2.65)

Therefore the London penetration depth, (T ) is given by

(T ) =

me b
2
20 e a(T
c T)

1/2

(2.66)

Clearly this will diverge at the critical temperature, Tc , since it is proportional to (Tc T )1/2 . We saw earlier that the Ginzburg Landau coherence
length, (T ), also diverges with the same power of (Tc T) , and so the
dimensionless ratio,
(T )
,
(2.67)
=
(T )
is independent of temperature within the Ginzburg Landau theory. Table
2.1 summarizes the measured values of penetration depth and coherence
length at zero temperature, (0), (0), for a selection of superconductors.

2.8

Flux quantization

Let us now apply the Ginzburg-Landau theory to the case of a superconducting ring, as shown in Fig. 1.4. Describing the system using cylindrical
polar coordinates, r = (r, , z), with the z-axis perpendicular to the plane

48

The Ginzburg-Landau model

Table 2.1 Penetration depth, (0), and coherence length, (0), at zero temperature for some important superconductors. Data values are taken from Poole
(2000).

Al
Sn
Pb
Nb
Nb3 Ge
YNi2 B2 C
K3 C60
YBa2 Cu3 O7

Tc (K)

(0) (nm)

(0)(nm)

1.18
3.72
7.20
9.25
23.2
15
19.4
91

1550
180
87
39
3
8.1
2.8
1.65

45
42
39
52
90
103
240
156

0.03
0.23
0.48
1.3
30
12.7
95
95

of the ring, we see that the order parameter (r) must be periodic in the
angle ,
(r, , z) = (r, + 2, z).
(2.68)
We assume that the variations of (r) across the cross section of the ring
are unimportant, and so we can neglect and r and z dependence. Therefore
the possible order parameters inside the superconductor are of the form
() = 0 ein

(2.69)

where n is an integer and 0 is a constant. We can interpret n as a winding number of the macroscopic wave function, exactly as for the case of
superfluid helium in Fig. ??.
However, unlike the case of superfluid helium, a circulating current in
a superconductor will induce magnetic fields. Assuming that there is a
magnetic flux through the ring, then the vector potential can be chosen
to be in the tangential direction, e and is given by
A =

,
2R

(2.70)

where R is the radius of the area enclosed by the ring. This follows from
Z
Z
I
B.dS = ( A).dS = A.dr = 2RA .
(2.71)
The free energy corresponding to this wave function and vector potential
is
Fs (T )

Fn (T ) +

d r

2ei
b
h
2
|( +
A)|2 + a||2 + ||4

2m
h

+ EB

Flux quantization

Fs0 (T )

+V

h
2
2m

49

!
2

Z

 in
 2ei  ||2 + 1
B 2 d3 r (2.72)
R
2hR 
20

where we have used the expression to gradient in cylindrical polar coordinates

1 X
X
X
er +
e +
ez
(2.73)
X =
r
r
z

(Boas 1983), V is the total volume of the superconducting ring, and Fs0 (T )
is the ground state free energy of the ring in the absence of any currents
R and
magnetic fluxes. The vacuum magnetic field energy EB = (1/20 ) B 2 d3 r
can be expressed in terms of the inductance, L of the ring and the current
I,
1
EB = LI 2 .
(2.74)
2
Clearly, it will be proportional to the square of the total flux, through
the ring
EB 2 .
On the other hand, the energy of the superconductor contains a term depending on both the flux and the winding number, n. This term can be
expressed as,
h
2
||2 ( n0 )2 ,
V
2m R2
where the flux quantum is 0 = h/2e = 2.07 1015 Wb.
We therefore see that the free energy is equal to the bulk free energy
plus two additional terms depending only on the winding number n and the
flux . The energy of the superconducting ring is therefore of the general
form,
(2.75)
Fs (T ) = Fsbulk (T ) + const.( n0 )2 + const.2 .
This energy is sketched in Fig. 2.6. We can see from the figure that the free
energy is a minimum whenever the flux through the loop obeys = n0 .
This is the phenomenon of flux quantization in superconductors.
Taking a ring in its normal state above Tc and cooling it to below
Tc will result in the system adopting one of the meta-stable minima in
Fig. 2.6, depending on the applied field. It will then be trapped in the
minimum, and a persistent current will flow around the ring to maintain
a constant flux = n0 . Even if any external magnetic fields are turned
off, the current in the ring must maintain a constant flux in the ring.
It is possible to directly measure the magnetic flux directly in such rings,
and hence confirm that it is indeed quantized in units of 0 , or multiples
of 2 1015 Wb. Incidentally, the fact that flux quantization is observed in
units of 0 = h/2e and not units of h/e is clear experimental proof that the
relevant charge is 2e not e, hence implying the existence of Cooper pairs.

50

The Ginzburg-Landau model

Given that a system is prepared in one of the meta-stable minima, it

can, in principle, escape over the energy barriers to move into a neighbouring lower energy minimum. This would be a mechanism for the persistent
current to decay, and hence for dissipation. Such an event corresponds to
a change in the winding number, n, and is called a phase-slip. However
the rate for thermally hopping over these barriers is exponentially small,
of order
1
eE0 /kB T
(2.76)

where E0 is the barrier height between minima in Fig. 2.6. Clearly this
thermal hopping rate can be made negligibly small. For example, E0 is
formally proportional to the ring volume, V , and so can be made arbitrarily
large in a macroscopic system. In practice persistent currents have been
observed to flow for years, with essentially no decay!
Another interesting possible mechanism for a phase slip would be a
quantum tunnelling from one minimum to another. This would be possible
at any temperature. But again the rate is impractically small in macroscopic systems. However, a very interesting recent development has been
the direct observation of these tunnelling events in small mesoscopic superconducting rings. hese experiments have demonstrated macroscopic quantum coherence and are discussed briefly in the next chapter.

2.9

The Abrikosov flux lattice

The great beauty of the Ginzburg-Landau theory is that it allows one to

solve many difficult problems in superconductvity, without any reference to
the underlying microscopic BCS theory. In some sense one could argue that
it is more general, for example it would almost certainly applies to exotic
superconductors, such as the high Tc cuprates, even though the original
BCS theory does not seem to explain these systems. The other great advantage of the Ginzburg-Landau theory is that is considerably easier to
work with than the BCS theory, especially in cases where the order parameter has complicated spatial variations. The tour de force example of this
is the Abrikosov flux lattice.
Abrikosov found a solution to the Ginzburg-Landau equations in the
case of a bulk superconductor in a magnetic field. The result he obtained
is remarkable in many respects. It is essentially an exact solution for type
II superconductors, valid close to Hc2 . Furthermore, it predicted a striking
result, that just below Hc2 the order parameter forms into a periodic structure of vortices. Each vortex carries a magnetic flux, hence explaining how
magnetic flux enters superconductors in the mixed state between Hc1 and
Hc2 . Abrikosovs prediction of the flux lattice was confirmed experimentally, showing not just that vortices occur, but they also tend to align in a
regular triangular lattice, as predicted by the theory. This periodic lattice
of vortices was perhaps the first example in physics of emergent phenomena in complex systems; the fact that sufficiently complex systems exhibit

The Abrikosov flux lattice

51

a variety of novel phenomena on different length scales. These phenomena

effectively arise from self-organization on the macroscopic length scale.
In type II superconductors the thermodynamic phase transition at Hc2
is second order (see problem 4.1). Therefore we can expect that the GinzburgLandau order parameter is small in magnitude just below Hc2 and reaches
zero exactly at Hc2 .7 Therefore, the magnetization M will also be small at
a magnetic field just below Hc2 (since is near zero, the superfluid density
ns and the screening supercurrents will also vanish at Hc2 ),as can be seen
in Fig. 1.8. Therefore to a good approximation we can assume that
B = 0 H,

(2.77)

where H is, as usual, the applied field given by the external apparatus as
in Fig. 1.6. This also implies that sufficiently near to Hc2 we can neglect
any spatial variations in the B-field, B(r) and just treat it as a constant,
B = (0, 0, B).

(2.78)

It will be convenient to express the corresponding vector potential A in the

Landau gauge as
A(r) = (0, xB, 0).
(2.79)
In which case the Ginzburg-Landau equation, Eq. ?? becomes

 

h
2
2eBi
2eBi
+
xj . +
xj (r)+a(T )+b||2 = 0, (2.80)
2m
h

where, as usual, j is the unit vector in the y direction.

Now if we are infinitesimally below Hc2 , then is essentially zero and
we can drop the cubic term, b||2 . All the other terms are liner in
and so we have linearized the equation. Expanding out the bracket (paying
attention to the commutation of and xj) gives


h
2
(2eB)2 2
4eBi
2
+
x

x (r) + a(T ) = 0.
(2.81)
2m
h

y
h
2
Introducing the cyclotron frequency,
c =

2eB
,
m

(2.82)

and noting that a(T ) is negative since we are at a temperature below the
zero field Tc , the equation can be written in the form,


h
2

m c2 2
2 h
c ix
+
x (r) = |a|(r).
(2.83)
2m
y
2
where (T ) is the Ginzburg-Landau coherence length.
7 In contrast this would not be the case for a type I superconductor, where the transition at Hc is first order. In these superconductors jumps from discontinuously from
zero to a finite value at Hc .

52

The Ginzburg-Landau model

Now Eq. 2.83 has the form of an eigenvalue equation, and is well known
in quantum mechanics. It is equivalent to the Scrodinger equation for the
wave function of a charged particle in a magnetic field, which has well
known Landau level solutions (Ziman 1979). The solution has the form,
(r) = ei(ky y+kz z) f (x),

(2.84)

which is a combination of plane waves in the y and z directions, combined

with an unknown function of x, f (x).
To find an equation for this function, f (x) we substitute the trial solution into Eq. 2.83. We find that f (x) obeys



h
2 (ky2 + kz2 )
h
2 d2 f
m c2 2
f
=
(|a|

+
h

k
x
+
x
)f.
c y
2m dx2
2
2m

(2.85)

The term in brackets on the left hand side can be rearranged by completing
the square,


m c2
m c2 2
m c2 2
x =
(x x0 )2
x0
h
c ky x +
2
2
2

(2.86)

where
x0 =

h
ky
.
mc

(2.87)

Finally, moving all the constants over to the right hand side we find,

m c2
h
2 kz2
h
2 d2 f
2
+
(x

x
)
f
=
(|a|

)f.
0
2m dx2
2
2m

(2.88)

Eq. 2.88 is just the Schrodinger equation for a simple Harmonic oscillator, except that the origin of coordinates is shifted from x = 0 to x = x0 .
Therefore the term in square brackets on the right is just the energy of the
oscillator,


1
h
2 kz2
,
(2.89)
n+
h
c = |a|
2
2m
or


h
2 kz2
1
= a(T
c T ).
n+
h
c +
2
2m

(2.90)

The corresponding functions f (x) are just the wave functions of a simple
Harmonic oscillator for each n, shifted by x0 .
Imagine that we gradually cool a superconductor in an external field,
H. At the zero field transition temperature, Tc , it will be impossible to
satisfy Eq. 2.90 because of the zero point energy term h
c /2 on the right

The Abrikosov flux lattice

53

hand side. A solution will only be possible when the temperature is far
enough below Tc to achieve,
1
h
c = a(T
c T ),
2

(2.91)

corresponding to the lowest possible energy solution (n = 0, kz = 0).

This equation determines the depression in transition temperature in the
magnetic field,
Tc (H)

=
=

1
h
c
2a
2eh0
Tc (0)
H.
2am

Tc (0)

(2.92)

Alternatively, we can start in a large external field, H, above Hc2 which

we gradually decrease (keeping the temperature fixed) until
1 2eB
h

= a(T
c T ).
2 m

(2.93)

Therefore, rearranging,
0 Hc2 = Bc2

=
=

2m a(T
c T) h
2e
h
2
0
.
2(T )2

(2.94)

It is interesting to note that this result implies that at Hc2 there is exactly
one flux quantum (i.e. one vortex line), in each unit area 2(T )2 . This
expression also provides the simplest way to measure the Ginzburg-Landau
coherence length (0) experimentally. Since (T ) = (0)t1/2 where t =
|T Tc |/Tc ,
0 Tc T
0 Hc2 =
(2.95)
2(0)2 Tc
and so by measuring the gradient of Hc2 (T ) near to Tc one can easily find
the corresponding (0).
It is also interesting to compare this expression for Hc2 with the corresponding result for Hc which we found earlier. Using the Ginzburg-Landau
expression for the total energy in the Meissner state we had found that
Hc

=
=

a
(Tc T )
(0 b)1/2
o

20 2

54

The Ginzburg-Landau model

H
c2 ,
2

and therefore
Hc2 =

(2.96)

2Hc

(2.97)

From this we can deduce that for superconductors with > 1/ 2 we

will have Hc2 > Hc , and the phase transition will be second order with
the order parameter growing continuously from zero at fields just below
Hc2 . I.e. we will have a type
II superconductor. On the other hand, for
superconductors with < 1/ 2 we will have Hc2 < Hc , and the phase
transition will be first order phase transition at the field Hc , below which
the order parameter jumps discontinuously to a finite value. The Abrikosov
theory therefore immediately describes the difference between type I and
type II superconductors,
(
< 12
type I

> 2 type II
The linearized Ginzburg-Landau equation allows us to find at Hc2 , but
does not immediately tell us anything about the form of the solution below
this field. To do this we must solve the non-linear equation, Eq. 2.80. This is
very hard to do in general, but Abrikosov made a brilliant guess and came
up with essentially the exact solution! He could see from the solutions to
the linearized equation Eq. 2.80 that only the harmonic oscillator ground
state solutions n = 0 and kz = 0 will be significant. However there are
still an infinite number of degenerate states, corresponding to the different
possible ky values,
(r) = Cei(ky y) e(xx0 )

/(T )2

(2.98)

where C is a nomalization constant. Here we have used the fact that the
ground state wave function of a quantum harmonic oscillator is a gaussian
function. The width of the gaussian solution to Eq. 2.88 turns out to be
the GL coherence length (T ).
Abrikosovs trail solution was to assume that we can combine these
solutions into a periodic lattice. If we look for a solution which is periodic
in y with period ly , then we can restrict the values of ky to
ky =

2
n
ly

(2.99)

with n any positive or negative integer. The corresponding Landau level x

shift is
2h
0
x0 =
n=
n.
(2.100)
mc ly
Bly
Therefore we can try a periodic solution

The Abrikosov flux lattice

(r) =

Cn ei(2ny/ly) e(x+nhi0 /Bly )

/(T )2

55

(2.101)

n=,

In this solution we can view the parameters Cn as variational parameters

which are to be chosen to minimize the total GL free energy of the system.
The solution above is periodic in y, but not necessarily periodic in x.
Abrikosov noted that it can be made periodic in x provided the coefficients
obey
Cn+ = Cn
(2.102)
for some integer . The period is lx , where
lx =

0
.
Bly

(2.103)

Abrikosov studied the simplest case, = 1 which corresponds to a simple

square lattice lattice. Later it was shown that a slightly lower total energy
is obtained for the = 2 case and the minimum energy state corresponds
to a simple triangular lattice. In each case the order parameter (r) goes
to zero at one point in each unit cell, and there is exactly one flux quantum
0 per unit cell. Therefore the solution is a periodic lattice of vortices.
This is illustrated in Fig. 2.7.
Experimental evidence for the Abrikosov flux lattice comes from a variety of methods. In a flux decoration experiment small paramagnetic
particles are dusted onto the surface of the superconductor (just like the
childrens experiment to see magnetic fields of a bar magnet using iron filings on paper above the magnet!). The particles concentrate int he points
of highest magnetic field, i.e. the vortices. Other similar methods involve
scanning a small SQUID loop or Hall probe just above the surface of the
superconductor to directly measure the variation of the local flux density
B(r). For most ordinary type II low Tc superconductors, such as Pb or
Nb, these experiments indeed show that the vortices form a fairly regular
hexagonal lattice. The lattice can be periodic over quite long length scales,
but is disrupted now and then by defects. These defects (exactly analogous
to crystal dislocations or point defects) tend to concentrate near defects in
the underlying crystal lattice (such as grain boundaries, impurities etc.).
Another method of observing the order in the flux lattice is to use
neutron scattering. The neutrons have a magnetic moment, and so are sensitive to the magnetic field B(r). If this is periodic, as in the Abrikosov
flux lattice, then there will be diffraction. The diffraction pattern can be
used to find the geometry of the flux lattice. Again the majority of systems
studied are found to have triangular lattices. But, interestingly, a square
lattice has now been found in a few recently discovered superconductors.
These are: the borocarbide system ErNi2 B2 C, the high temperature superconductor YBa2 Cu3 O7 and the possible p-wave strontium ruthenate
Sr2 RuO4 . All of these have been found to have square vortex lattices in at
least some range of external fields. It may be that this is simply because

56

The Ginzburg-Landau model

of small corrections to the original Abrikosov theory (such as terms omitted in the standard GL equations, such as higher powers of , like ||6 , or
higher order gradients). n fact the square lattice and triangular lattice have
energies differing by less than 1% in the Abrikosov solution. But in some
cases it seems more likely to be due to the fact that the underlying form
of superconductivity is unconventional, meaning that the Cooper pairs
have a different symmetry from the normal BCS case. We shall introduce
these ideas briefly in chapter 7.
Finally, we should note that while the Abrikosov solution is essentially
exact just below Hc2 , it cannot necessarily be applied far away from there,
such as at Hc1 . As we have seen, near to Hc2 the vortices are close together,
separated by distances of order the coherence length (T ). Effectively they
are so densely packed that their cores are essentially touching. On the other
hand, just at Hc1 we have very few vortices in the entire sample, and so
they are well separated. We can estimate of the lower critical field Hc1 from
the energy balance for the very first few vortices to enter a superconductor
in the Meissner phase. One can show that a single London vortex has an
energy of approximately (see exercise 3.3)
 

20
ln
(2.104)
E=
40

per unit length. Therefore in a superconductor with N/A flux lines per
unit area and thickness L, there will be a total energy cost EN/A per unit
volume due to the vortices. But on the other hand, each vortex carries a flux
0 , and so the average magnetic induction in the sample is B = 0 N/A.
The magnetic work gained by the presence of the vortices is 0 HdM =
HdB (at constant H). Energy balance therefore favours the presence of
the vortices when
N
N
(2.105)
E < H0 .
A
A
Thus is becomes energetically favourable for the vortices to enter the sample
when H > Hc1 , where
 
o

Hc1 =
.
(2.106)
ln
40 2

This is obviously the lower critical field, and can be simply expressed as
Hc
(2.107)
Hc1 = ln().
2

This expression is only valid when >> 1/ 2, i.e. in the London vortex
limit.

2.10

Thermal Fluctuations

The Ginzburg-Landau theory as described above is purely a mean-field

theory. It neglects thermal fluctuations. In this t is therefore similar to the

Thermal Fluctuations

57

Curie-Weiss or Stoner models in the theory of magnetism (Blundell 2001).

But in fact it is a great strength of the GL theory that it can easily be
extended so that these fluctuations can be included. It is much simpler to
include these in the GL theory than in the more complex BCS theory.
In the mean-field approach, we must always find the order parameter
(r) which minimizes the total free enrgy of the system. As discussed above,
this is a functional minimization. The total free energy of the system F [],
in Eq. 2.29, is a functional of the complex order parameter, (r), meaning
that it depends on an infinite number of variables: the values of all
possible points r. As we saw, the condition for minimizing the free energy
is that the functional derivatives given in Eq. 2.37 are zero.
To go beyond this mean-field approach we must include fluctuations
of (r) close to this minimum. For example if we make a small variation
in (r), such as (r) 0 (r) = (r) + (r), then we expect that the
energy of the system represented by 0 (r) would be very similar to that
represented by (r). If the total energy difference is small, or no more than
kB T , then we might expect that in thermal equilibrium the system would
have some probability to be in state 0 (r). We need to define an effective
probability for each possible state. Clearly this must be based on the usual
Boltzmann probability distribution and so we expect that,
P [] =

1 F []
e
Z

(2.108)

is the probability density for the system to have order parameter (r). It
is again a functional of (r) as indicated by the square brackets.
The partition function Z is the normalization factor in this expression.
Formally it is a functional integral,
Z
Z = D[]D[ ]eF [] .
(2.109)
We can treat the integrals over and as formally separate for the same
reason that we could view functional derivatives with respect to and
as formally independent. It is really allowed because in fact we have to
specify two independent real functions, the real and imaginary parts of
at each point r: Re[(r)] and Im[(r)].
What is the meaning of the new integration symbols, like D[] in
Eq. 2.109? We are technically integrating over an infinite number of variables, the values of (r) at every point r. It is difficult (and beyond the
scope of this book!) to make this idea mathematically rigorous. But we can
find an intuitive idea of what this mens by supposing that we only had a
discrete set of points in space, r1 , r2 , . . . , rN . We could define values of
and at each point, and then calculate a Boltzmann probability. The
approximate partition function for this discrete set would be the multiple
integral

58

The Ginzburg-Landau model

Z(N ) =

d(r1 )d(r1 )

d(r2 )d(r2 ) . . .

d(rN )d psi(rN )eF [] .

(2.110)
The full functional integral is a limit in which we make the set of points
infinitely dense (in fact even possibly uncountably infinite!), defining
Z = lim Z(N ).
N

(2.111)

One way that this infinite product of integrals might be accomplished

is through the Fourier transforms of (r) and (r). If we define k by
X
k eik.r
(2.112)
(r) =
k

then specifying the parameters k and k at every wave vector k =

(2nx /Lx , 2ny /Ly , 2nz /Lz ) (or equivalently the real and imaginary parts),
is defines the full functions (r) and (r). In this representation we can
write the partition function as

Y Z

dk dk eF [] .
(2.113)
Z=
k

Again there are an infinite number of integrals, two for each point k.
As an example of the sort of thermal fluctuation effects that can be
calculated within this formalism we shall consider the specific heat of a
superconductor near to Tc . For a superconductor in zero magnetic field we
have the free energy functional,

 2
Z
b 4
h

2
2
||
+
a||
+
||
(2.114)
F [] = d3 r
2m
2
(dropping the constant normal state free energy Fn , which will be irrelevant
here). Writing this in terms of the Fourier coefficients bf k we find
F [] =

X h
2 k2
bf k

2m

+ +a k k +

b
2

k1 ,k2 ,k3

k 1 k 2 k3 k1 +k2 k3 , (2.115)

which could in principle be inserted directly into Eq. 2.113. In general this
would be very difficult, and requires either massive numerical Monte Carlo
simulation, or some other approximation. The simplest approximation that
we can make is to make the gaussian approximation, in which we neglect
the quartic (b) term in the free energy. In this approximation we find a
simple result

 2 2


YZ
h
k

Z=
dk dk exp
(2.116)
+ a k k .
2m
k

Thermal Fluctuations

59

Changing variables to the two real functions, Re[k ] and Re[k ] gives


h
2 k2
2
2
dRe[k ]dIm[k ]exp
Z=
+ a (Re[k ] + Im[k ] ) ,
2m
k
(2.117)
and so for each k the integral is just a two dimensional gaussian integral.
These can be done exactly, resulting in the partition function
Y

 2
.
(2.118)
Z=
h
k2
k 2m + a


YZ

From the partition function it is possible to calculate all thermodynamic

quantities of interest. For example, the total internal energy is given by
Eq. ??,
U

=
=

ln Z

ln Z
+kB T 2
T
X
1
da
 2


,
2
h
k
dT
+
a

k
2m

(2.119)

where in the last step we have kept only the most important contribution
which comes from the change of the Ginzburg Landau parameter a with
T , da/dT = a.

The gaussian approximation for the heat capacity near to Tc is found

by differentiating again, giving
CV =

dU
dT

h
2 k2

2
2 a .

+a
Z
1
V a
d3 k
3
2
(2 ) a
(1 + (T )2 k 2 )2
2
V a
1
3
2
(2 ) a (T )3
1
|Tc T |3/2
(T Tc )2
1
,
|T Tc |1/2
k

1


2m
2

(2.120)

(where, for simplicity, we have ignored the numerical multiplying prefactors). This shows that the thermal fluctuations can make a very large contribution to the heat capacity, essentially diverging at the critical temperature Tc . If we sketch this behaviour we see that the thermal fluctuations

60

The Ginzburg-Landau model

make a large difference to the original mean-field specific heat of Fig. 2.4.
As can be seen in Fig. ??. In fact, once the fluctuations are included the
the heat capacity of superconductors becomes much more similar to the
heat capacity of superfluid 4 He at Tc , as shown in Fig. ??.8
Experimentally these thermal fluctuations are very difficult to see in
standard low Tc superconductors, such as Pb or Nb. It is possible to
estimate the range of temperatures near to Tc where these fluctuations are
significant. This temperature range, TG is known as the Ginzburg critereon.
In 1960 Ginzburg found that this temperature range is extremely small, i.e.
much less than 1K for most low Tc superconductors. Therefore we can say
that the original mean-field approach to the Ginzburg Landau equations is
perfectly well justified. However, in the high temperature superconductors,
discovered in 1986, the coherence length (0) is very small (Table 2.1), of
order just a few Angstroms. It turns out that the corresponding Ginzburg
temperature range, TG , is of order 1 2K. Therefore is is quite possible
to see such thermal fluctuation effects in these systems. The specific heat
near Tc clearly shows thermal critical fluctuations, as shown in Fig. 2.9. In
fact in these experimental results very good agreement is found using the
value of the critical exponent given by the three dimensional XY model
predictions, exactly as is the case in superfluid 4 He, Fig. ??. The gaussian
model exponent = 1/2 does not fit at all as well. Another example of
thermal fluctuation effects can be seen in the resistivity, (T ), just above T c .
Thermal fluctuations make (T ) begin to bend down towards zero even at
temperatures quite far above Tc . This downward bending is clearly visible in
the resistivity curve of the Tc = 135K superconductor HgBa2 Ca2 Cu3 O8 +
shown in Fig. 1.2.

2.11

Vortex Matter

Another very important consequence of thermal fluctuations occurs in

the mixed state of high temperature superconductors. As we have seen,
Abrikosovs flux lattice theory shows that the vortices align in a periodic
lattice arrangement, essentially like a crystal lattice, either triangular or
square. However, this is again a mean field approximation! We must, in
principle, again include the effects of thermal fluctuations.
The theories of the resulting vortex matter states show a very wide
range of possibilities. It is still possible to talk about the vortices, but
now they themselves form a variety of different states, including liquid
and glassy (random, but frozen) states, as well as nearly perfectly ordered
8 In fact, the gaussian theory, as outlined above, is not truly correct since it dropped
the ||4 terms in the Ginzburg-Landau free energy. When these terms are included the
resulting theory is known in statistical physics as the XY or O(2) model. Its true critical
behaviour near Tc can be calculated with various methods based on the renormalization
group. The resulting critical exponent for specific heat is very small and very different
from the = 1/2 which is given by the gaussian approximation of Eq. 2.120.

Vortex Matter

61

crystalline states. It is believed that the flux lattice never has true crystalline order, and thermal fluctuations always lead to an eventual loss of
long ranged order in the periodic structure (although in practice periodicity can be quite well extended). A full discussion of these topics requires a
whole book in itself (Singer and Schneider 2000), and there also are many
extensive review papers (Blatter 1994).
Unfortunately these thermal fluctuations have been disastrous for commercial applications of high Tc superconductors in high current wires and
electromagnets (Yeshrun 1998).9 The problem is that thermal fluctuations
lead to motion of the vortices, and this leads to a source of energy dissipation. Therefore the resistivity is not zero for high Tc superconductors
in a magnetic field. The problem also occurs in low Tc superconductors,
but to a much lesser extent. In these systems the energy dissipation due
to motion of vortices can be reduced or eliminated by providing pinning
centres which pin the vortex lattice and prevent it from moving. Typically these are just impurities, or naturally occurring crystal defects such
as grain boundaries and dislocations.
To see why motion of vortices leads to energy dissipation is is necessary
to see that a current density j flowing through the vortex lattice (perpendicular to the magnetic field) leads to a Lorentz (or Magnus) force on each
vortex. The overall force is

f =jB
(2.121)
per unit volume of the vortex lattice. This will tend to make the vortex
liquid flow in the direction perpendicular to the current as shown below on
the left, as illustrated in Fig. ??.
Unfortunately if the vortices flow in response to this force, work is done
and there is energy dissipation. To calculate the work, consider a loop of
superconducting wire, with a current flowing around the wire. Vortices will
tend to drift transversely across the wire, say entering on the inner side of
the wire and drifting over to the outer side. This is illustrated in Fig. ??.
Each vortex carries a magnetic flux o , and so the total magnetic flux in
the ring changes by o with each vortex that crosses from one side of the
wire to the other. But by elementary electromagnetism there is an EMF
induced in the wire given by = d/dt. Power is dissipated, at a rate
given by P = I where I is the total current. Therefore vortex motion
directly leads to finite resistance! In the mixed state, superconductors only
have truly zero resistance when the vortices are pinned and unable to move.
9 Perhaps thisis not the only difficulty with commerical applications of high T superc
conductivity. The materials are brittle and cannot easily be made into wires. Nevertheless
these problems have been gradually overcome, and now high Tc superconducting wires
are beginning to make a serious entry into commercial applications. For example, at least
one US city receives part of its electrical power through underground superconducting
cables. Some microwave receivers, such as in some masts for mobile phones beside motorways, also use superconducting devices operating at liquid nitrogen temperatures.

62

The Ginzburg-Landau model

In high Tc superconductors the thermal motion of vortices leads to

especially bad pinning and hence a significant resistivity in the mixed state.
To make matters worse, the lower critical field Hc1 is tiny, often less than
the Earths magnetic field, and so vortices can never by truly eliminated.
At high temperatures and near to Hc2 it is believed that the vortex matter
is in a liquid state, and so the vortices can move freely and pinning is
essentially impossible.10 Lowering the temperature, or going further away
from Hc2 the vortex matter appears to freeze into a glassy state. A glass
is random spatially, but frozen in time. Since glass is effectively rigid the
vortices cannot move and pinning is able to largely prevent flux motion.
Therefore in this state the resistivity is quite low. Unfortunately, even in
this glassy vortex state the resistivity is not fully zero, since flux creep can
occur. The random pinning force provides a set of energy barriers to vortex
motion, but thermal motions mean that from time to time the vortices can
hop over the local energy barrier and find a new configuration.11 The line
in the (H, T ) phase diagram where the glassy phase occurs is called the
irreversibility line, as shown in Fig. 2.10. Something approaching zero
resitivity is approached only well below this line. This effectively limits
the useful magnetic fields for applications of high Tc superconductivity, to
very much less than the hundreds of Tesla that one might have expected
from the nominal values of 0 Hc2 > 100T , such as one might expect from
Table 2.1.

2.12

Summary

We have seen how the Ginzburg Landau theory provides a mathematically

rather simple picture with which to describe quite complex phenomena
in superconductivity. In terms of the phenomenological order parameter,

(r), and four empirically determined parameters (a,b,m

and Tc ) we can
can construct a full theory of superconductivity which encompasses fully
phenomena such as the Abrikosov flux lattice, flux quantization and from
which one can derive the London equation.
The power of the theory is also apparent in the way it can be modified to incorporate thermal fluctuations, including critical phenomena and
vortex matter physics. It should be noted that these areas are still highly
active areas of experimental and theoretical activity. Even some very simple and fundamental questions are still hotly debated, such as the various
vortex phases occurring in high temperature superconductors. These also
have important implications for potential commercial applications of these
materials.
10 A liquid can always flow around any impurities and so pinning centres have no effect
in the vortex liquid state.
11 The process is presumably analogous to the way that window glass in medieval
cathedrals appears to have gradually flowed downwards over timescales of hundreds of
years.

Further Reading

2.13

63

Further Reading

Magnetic work in thermodymanics is discussed in detail in the textbooks

by Mandl (1987) and Callen (1960).
The idea of order parameters and Ginzburg Landau theory in general,
including superconductors, are discussed in Chakin and Lubensky (1995)
and Anderson (1984). Applications of Ginzburg Landau theory to vortex
states and other problems are covered in detail by de Gennes (1966). Other
books are also useful, such as Tilley and Tilley (1990), Tinkham (1996). In
fact almost all textbooks on superconductivity have at least one chapter
dealing with Ginzburg Landau theory and its predictions.
Thermal fluctuations and critical phenomena are in themselves huge
fields of study. A good introductory course is Goldenfeld (1992), while the
books Amit (1984) and Ma (1976) are very comprehensive. These books
discuss very general classes of theoretical models, but the Ginzburg Landau
theory we have discussed is equivalent to the model they call XY or O(2).
For a modern view of thremal fluctuation phenomena and the problems
of vortex matter physics, especially in its application to high temperature
superconductors, see the book by Singer and Schneider (2000), or the review
articles by Blatter (1994) and Yeshrun (1996).

2.14

Exercises

(4.1) (a) For a type I superconductor Hc (T ) is the boundary between normal metal and superconductor in the H, T phase diagram. Everywhere on
this boundary thermal equilibrium requires
Gs (T, H) = Gn (T, H).
Apply this equation and dG = SdT 0 M dH to two points on the H T
phase boundary (T, H) and (T + T, H + H), , as illustrated in Fig. 2.12.
Hence show that:
Ss T 0 Ms H = Sn T 0 Mn H
when T and H are small, and where Ss/n Ms/n are the superconducting
and normal state entropy and magnetization respectively.
(b) Using part (a), and Mn = 0, and Ms = H show that the latent heat
per unit volume for the phase transition, L = T (Sn Ss ), is given by
L = 0 T Hc

dHc (T )
dT

where the phase boundary curve is Hc (T ). (This is exactly analogous to

the Clausius-Claperyon equation in a gas-liquid phase change except H
replaces P and 0 M replaces V . See Mandl (1987) p 228).

64

The Ginzburg-Landau model

(4.2) Find ||2 , the free energy Fs Fn , the entropy and the heat capacity
of a superconductor near Tc , using the bulk Ginzburg-Landau free energy.
Sketch their variations with temperature assuming that a = a (T Tc )
and a and b are constant near Tc .
(4.3) (a) Show that for one-dimensional problems, such as the surface discussed in sec. 4.5, the Ginzburg Landau equations for (x) can be rewritten
as:
d2
2 f (y) f (y) + f (y)3 = 0
dy
p
where x = y(T ), f (y) = (y)/0 , and 0 = |a|/b.
(b) Verify that

f (y) = tanh (y/ 2)

is a solution to the equation in (b) corresponding to the boundary condition (0) = 0. Hence sketch (x) near the surface of a superconductor.
(c) Often the surface boundary condition of a superconductor is not (x) =
0, but (x) = C where C is a numerical constant. Show that if C < 0
we can just translate the solution from Prob. 4.2 sideways to find a valid
solution for any value of C in the range 0 C < 0 .
(d) In the proximity effect a metal (in the half-space x > 0) is in contact
wi th a superconductor (occupying the region x < 0). Assuming that the
normal metal can be described by a Ginzburg-Landau model but with
a > 0, show that the order parameter (x) induced in the metal by the
contact with the superconductor is approximately
(x) = (0)ex/(T )
where h
2 /2m (T )2 = a > 0, and (0) is the order parameter of the superconductor at the interface.
(4.4) (a) In Eq. 2.120 we showed that the gaussian model gives a divergence
in specific heat of the form
CV

1
|T Tc |

with = 1/2. Repeat the steps given in Eq. 2.120 for the case of a two
dimensional system, and show that in this case the gaussian model predicts
= 1.

Exercises

65

(b) This argument can also be extended easily to the general case of ddimensions. By replacing the k sum in Eq. 2.120 by an integral of the
form
Z
X
1
dd k

(2 d )
k

show that in d dimensions we have the critical exponent

d
=2 .
2

66

The Ginzburg-Landau model

Type I

Type II
H

H
Hc

(T, Hc2 )

Hc2

(T, Hc )

Hc1

(T, 0)

(T, 0)
Tc

Tc

Fig. 2.1 We obtain the condensation energy for superconductors by thermodynamic integration of the Gibbs free energy along the contours in the (T, H) plane,
as shown above.

fs fn

T > Tc
T < Tc

Fig. 2.2 Free energy difference between the normal and superconducting states
(per unit volume) as a function of the order parameter . For T < Tc the free
energy has a minimum at 0 , while for T > Tc the only minimum is at = 0.

||

Tc

Fig. 2.3 Order parameter magnitude, ||, as a function of temperature in the

Ginzburg Landau model.

Exercises

67

CV
CV n = T

Tc

Fig. 2.4 Specific heat of a superconductor near Tc in the Ginzburg Landau

model. Above Tc the specific heat is given by the Sommerfeld theory of metals,
CV n = T . At Tc there is a discontinuity and a change of slope.

0
(x)

x
Fig. 2.5 Order parameter of a superconductor near a surface. It recovers to its
bulk value 0 over a length scale of the coherence length, .

F () F (0)

20

20

Fig. 2.6 Flux quantization in a superconducting ring. Metastable energy minima

exist when the flux is an integer multiple of the flux quantum 0 = h/2e. There
is an overall background increase with corresponding to the self-inductance of
the ring, making the zero flux state = 0 the global energy minimum. Thermal fluctuations and quantum tunnelling allow transitions between neighbouring
meta-stable energy minima.

68

The Ginzburg-Landau model

ly

lx
Fig. 2.7 The Abrikosov flux lattice. The figure shows the amplitude of the order
parameter, |(r)|2 for the lowest energy triangular lattice. Each triangular unit
cell contains on quantum, 0 , of magnetic flux and contains on vortex where

(r) = 0. In terms of the lx and ly lattice periodicities used in Sec. 4.9, ly = 3lx ,
and the rectangular unit cell lx ly contains two vortices and two flux quanta.

Exercises

69

|T Tc |

CV

TG
0

T Tc

Fig. 2.8 Specific heat of a superconductor near to Tc in the gaussian approximation. The mean field Ginzburg-Landau theory gives a discontinuity at Tc . ut
this is supplemented by a thermal fluctuation contribution which diverges like
|T Tc | with = 1/2. The full renormalization group treatment (ignoring
magnetic field terms) shows that is given by the three-dimensional XY model
value, exactly as in superfluid helium shown in Fig. ??.

Fig. 2.9 Experimental heat capacity of the high temperature superconductor

YBa2 Cu3 O7 near to Tc . In zero magnetic field the experimental data fits very
well the predictions of the three dimensional XY model. An external magnetic
field (inset) removes the singularity, but does not visibly reduce Tc significantly.
Reproduced from Overend, Howson and Lawrie (1994), with permission.

70

The Ginzburg-Landau model

Hc2
vortex liquid
Hirr
vortex glass

Hc1

Tc

Fig. 2.10 Proposed magnetic phase diagram of some high Tc superconductors.

Below Hc2 vortices form, but are in a liquid state, leading to finite resistance
of the superconductor. Below the irreversibility line the vortices freeze (either
into a glassy or quasi periodic flux lattice). In this state resistivity is still finite,
due to flux creep, but becomes negligible far below the irreversibility line. Hc1 is
extremely small.

B
j

j
f

f
B

Fig. 2.11 Energy dissipation due to vortex flow in superconductors. Every vortex experiences a Lorentz (Magnus) force perpendicular to the supercurrent direction. This causes the vortices to drift sideways across the wire, unless pinned
by defects. For each vortex which crosses the wire from one side to the other, a
certain amount of work is done, and energy is dissipated.

Exercises

71

Hc (T )
(H, T )
(H + H, T + T )

T
Fig. 2.12 Exercise 4.1. Consider the Gibbs free energy at the points shown on
the normal-superconducting phase boundary of a type I superconductor, (T, H)
and T + T, H + H). At both points equilibrium requires that both normal and
superconducting Gibbs free energies must be equal: Gn (T, H) = Gs (T, H).

3
The Macroscopic Coherent
State
3.1

Introduction

We have seen in the previous chapters that the concept of the macroscopic wave function,(r), is central to understanding atomic Bose-Einstein
condensates, superfluid 4 He, and even superconductivity within the the
Ginzburg-Landau theory. But the connection between these ideas is not at
all clear, since the atom condensates and 4 He are bosonic systems, while
while superconductivity is associated with the conduction electrons in metals which are fermions. The physical meaning of the Ginzburg-Landau order parameter was not at all clear until after 1957 when Bardeen Cooper
and Schrieffer (BCS) published the first truly microscopic theory of superconductivity. Soon afterwards the connection was finally established by
Gorkov. He was able to show that, at least in the range of temperatures
near Tc , the Ginzburg Landau theory can indeed be derived from the BCS
theory. Furthermore this provides a physical interpretation of the nature
of the order parameter. Essentially is is describing a macroscopic wave
function, or condensate, of Cooper pairs.
The purpose of this chapter is to clarify the concept of a macroscopic
wave function, and show how it arises naturally from the physics of coherent states. Coherent states were first developed in the field of quantum
optics, and were especially useful in the theory of the laser. The laser is, of
course, yet another type of macroscopic coherent state, with close similarities to atomic BEC. In this chapter we shall first introduce the concept and
mathematical properties of coherent states before applying them to bosonic
systems. Using this approach we shall rederive the Gross-Pitaevskii equ
ations for the weakly interacting bose gas, as originally introduced above
in Chapter 1.
Coherent states can be defined for fermions as well as for bosons. But
single fermion coherent states, while very useful in other contexts, are not
directly useful in the theory of superconductivity. What is needed is a coherent state of fermion pairs. Such coherent states are exactly the type
of many-body quantum state first written down in the theory of Bardeen
Cooper and Schrieffer in their 1957 theory of superconductivity. We shall

Coherent states

73

postpone a full discussion of this BCS theory until the following chapter.
Here we focus specifically on the physics of the BCS coherent state, without worrying, for example, about why it is a stable ground state. . This
separation has the advantage that the key concepts can be presented more
clearly and they can be seen to be very general. Indeed it is not necessary
to rely on every detail of the BCS theory in order to understand physically
the properties of the quantum coherent state. In this approach we can also
see very generally the connection between the BCS state and the GinzburgLandau theory, since the coherent state of electron pairs provides a direct
connection to the order parameter (r).
This logical separation between the full detail of the BCS theory and
the physical origin of the order parameter is not just an educational device;
it also has a useful purpose more generally. For example, there are several
superconductors where we do not know if the BCS theory is applicable,
the high Tc superconductors being the most obvious example. However,
even though we do not know the mechanism of pairing we do know that
these systems do have Cooper pairs. For example they have flux quantization in the usual units of 0 = h/2e, showing that the fundamental charge
unit is 2e. We can also assume that there will be a Ginzburg-Landau order parameter whatever the actual pairing mechanism, and this knowledge
will provide a sound basis for many theories of the superconducting state
(eg theories of the vortex matter states in high Tc superconductors). We
can therefore separating the actual mechanism of pairing from its main
consequence: the existence of the order parameter.

3.2

Coherent states

To start with, let us just go back to some elementary undergraduate quantum mechanics: the quantum Harmonic oscillator. The Hamiltonian operator is
2
2
= p + mc x2
(3.1)
H
2m
2
d
where p = ih dx
is the one-dimensional momentum operator, m is the particle mass and c is the classical oscillator angular frequency. The eigenstates
and of the oscillator are given by

n (x) = En n (x),
H

(3.2)

with energy levels En .

The most elegant method of solving this classic problem, to find En and
n (x) is to introduce the ladder operators,
a

1
(hc )1/2

(mc2 )1/2 x
p

i
(2m)1/2
(2)1/2

74

The Macroscopic Coherent State

1
(hc )1/2

(mc2 )1/2 x
p
+
i
(2m)1/2
(2)1/2

(3.3)

These operators have a number of very useful and easily derived properties, 1
which can be summarized as follows:
a
+ n (x)

(n + 1)1/2 n+1 (x)

1/2

a
n (x) = (n) n1 (x)
a
+ a
n (x) = nn (x)


a
, a
+ = 1.

(3.4)
(3.5)
(3.6)
(3.7)

The first of these relations shows that the operator a

+ changes any state
to the next one higher up the ladder of the possible n values. Similarly
second shows that a
moves down the ladder. From the third relation, the
combination a
+ a
results in no change of n. Therefore Eq. 3.6 shows that
we can identify the combination n
=a
+ a
as the number operator, which
gives the quantum number n of any state,
n
n (x) = nn (x).

(3.8)

The commutator relation, Eq. 3.7,

[
a, a
+ ] = a
a
+ a
+ a
=1

(3.9)

is fundamental to the quantumm mechanics of bosonic systems, as we shall

see below.
In terms of the latter operators the oscillator Hamiltonian is given by,


1
+

a
+
H=h
c a
.
(3.10)
2
Combined with Eq. 3.6 it immediately shows that the energy levels are


1
En = h
c n +
.
(3.11)
2
exactly as expected.
Using the ladder raising operator, a
+ repeatedly, Eq. shows that one
can construct all of the eigenvectors, n (x), iteratively by acting repeatedly
on an initial ground state 0 (x),
n (x) =

1
(
a+ )n 0 (x).
(n!)1/2

(3.12)

Therefore to find the complete set of states it is only necessary to find

0 (x) (which elementary quantum mechanics tells us is a simple gaussian
1 See

excercise 5.1.

Coherent states

75

function), and then all of the remaining quantum states can be generated
essentially automatically.
But the ladder operators also have many other uses. In particular let
us define a coherent state by,



2
3
|i = C 0 (x) + 1/2 1 (x) + 1/2 2 (x) + 1/2 3 (x) + . . .
(3.13)
1!
2!
3!
where is any arbitrary complex number, and C is a normalization constant. This constant C can be found easily from the normalization condition
1

h|i


||2
(||2 )2
(||2 )3
2
= |C| 1 +
+
+
+ ...
1!
2!
3!

|C|2 e||

(3.14)
2

and so we can take C = e|| /2 .

Coherent states have many interesting properties. The following expression is a particularly useful relation


(
a + )2
(
a + )3

a+
||2 /2
+
+
+ |0i
(3.15)
|i = e
1+
1!
2!
3!
where |0i = 0 (x) is the ground state and is also the coherent state with
= 0. This expression can be written very compactly as,
|i = e||

/2

exp (
a+ )|0i.

(3.16)

is just defined by the usual

Note that the exponential of any operator, X
series expansion of exponential
=1+
exp (X)

2
3
X
X
X
+
+
+ ...
1
2!
3!

(3.17)

Another interesting relation which can be obtained from Eq. 3.13 is

a
|i = |i

(3.18)

therefore they are eigenstates of the ladder operator a

. To prove this we
can write the state a
|i explicitly



2
3
||2 /2
a
|Ai = e
a
0 (x) + 1/2 1 (x) + 1/2 2 (x) + 1/2 3 (x) + . . . .
1!
2!
3!
(3.19)
But a
n (x) = n1/2 n1 (x) and so this gives


2 21/2
3 31/2
11/2
||2 /2
1 (x) +
2 (x) + . . .
a
|i = e
0 + 1/2 0 (x) +
1!
2!1/2
3!1/2
(3.20)
which is clearly equal to |i.

76

The Macroscopic Coherent State

Finally, two further nice properties are also simple consequences of

Eq. 3.18,
h|
a|i =
+
h|
a a
|i = ||2 .

(3.21)
(3.22)

Therefore the value of ||2 gives the mean number operator h

ni of the
quantum state. Extending this to n
2 we can find the number uncertainty
n,
h
n2 i

h|
a+ a
a
+ a
|i

h|
a+ (
a+ a
+ 1)
a|i
4
|| + ||2
p
=
h
n2 i h
ni 2
= ||.
=
=

(3.23)
(3.24)

Coherent states do not have a definite value of the quantum number n,

simply because they are not eigenstates of the number operator. In fact
the probability of observing the value of n in a quantum measurement of
state |i is actually a Poisson distribution shown in Fig. 3.1
Pn =

||2n ||2
e
n!

(3.25)

as can easily be seen from Eq. 3.13. In this distribution, as given by Eq. 3.24,
the standard deviation in the number n is
p
n = h
ni
(3.26)
or

1
n
p .
hni
hni

(3.27)

We shall be mostly interested in macroscopic coherent states, where hni is

essentially infinite. For such states, one can see that the standard deviation
n becomes essentially negligible compared to hni. Therefore to a good
approximation we can approximate many operator expectation values by
their mean-field values derived from the replacement n
hni. In Fig. 3.1
one can see that the distribution becomes strongly peaked about its mean
value of hni even for quite small values of hni.
Importantly, the coherent state |i does have a definite phase, , though
not have a definite quantum number n. The coherent state can be defined
for any complex number ,
= ||ei .
Rewriting Eq. 3.13 in terms of these variables we see that

(3.28)

Coherent states



||2
||
|i = C 0 (x) + ei 1/2 1 (x) + e2i 1/2 2 (x) + + . . . .
1!
2!

77

(3.29)

We see that the term containing n depends on ein . Differentiating with

respect to one can see that
1
|i = n
|i.
i

(3.30)

But since the states |i are a complete set (actually an overcomplete set!),
we can make the operator identification
1
=n
.
i

(3.31)

So the phase and the number n are conjugate operators, in a similar way
to momentum and position in standard quantum mechanics. It is possible
to state a form of the uncertainty principle for these operators2
n

1
.
2

(3.32)

Coherent states have a fixed phase, but do not have definite values of
number n. In contrast the energy eigenstates, n (x), have a well defined
value of n, but have an arbitrary (or ill-defined) phase.
Coherent states have many other beautiful mathematical properties,
which we will not have time to explore in detail. In particular they are an
overcomplete set, since they can be defined for any point in the complex
plane, and so there are uncountably infinite many such states. They are
also not orthogonal, and it is easy to show that (exercise 5.2)
2

|h|i| = e|| .

(3.33)

We can interpret this in terms of the Argand diagram for the complex
number , shown in Fig. 3.2. Every point in the plane represents a valid
coherent state. Coherent states at neighbouring points are not orthogonal,
but their overlap dies off when | | 1. Therefore there is effectively
one independent orthogonal quantum state per unit area of the complex
plane. Using polar coordinates, ||, we see that an element of area contains
exactly one quantum state if
1 ||||

(3.34)

which is of order 2n , since hni = ||2 . Therefore the number-phase

uncertainty principle gives essentially the minimum area per quantum state
in the complex plane of Fig. 3.2.
2 A precise proof of this uncertainty relation is a little more tricky than the usual
Heisenberg momentum-position uncertainty principle, since is only strictly defined in
the range between 0 and 2.

78

3.3

The Macroscopic Coherent State

Coherent States and the Laser

Coherent were first studied extensively in the theory of the laser. In quantum optics the use of the operators a+
ks and aks correspond directly to
creation and annihilation of photons in a particular mode of electromagnetic radiation with wave number k and polarization s (i.e. left or right
circularly polarized). A general quantum state of the system can be represented in the occupation number representation
|nk0 s0 , nk1 s1 , nk2 s2 , nk3 s3 , . . .i

(3.35)

where k0 , k1 , k2 etc. count all the different plane wave states of the system,
eg the cavity of the laser.
In the case of light, the creation and annihilation operators arise quite
naturally when one quantizes the electromagnetic radiation field. Each specific mode of the classical radiation field k, s obeys Maxwells equations.
When these equations are quantized each mode becomes an independent
quantum harmonic oscillator. The quantum states, nks , of each oscillator
are interpreted as the number of photons present. The creation operator
adds a photon while the annihilation operator destroys one,3
a+
ks | . . . nks . . .i

aks | . . . nks . . .i

=
=

(nks + 1)1/2 | . . . nks + 1 . . .i

(nks )

1/2

| . . . nks . . .i,

(3.36)
(3.37)

in complete analogy with the harmonic oscillator ladder operators. From

these one can deduce that the same commutation law as the ladder operators must apply. But the operators for independent radiation field modes
must commute, and therefore we can write,


(3.38)
= ks,k0 s0
aks , a +
k 0 s0
[aks , ak0 s0 ] = 0
(3.39)
 + + 
= 0.
(3.40)
aks , ak0 s0
In the case of the laser we can naturally shift from the occupation number representation to a coherent state representation. A general coherent
state is of the form
!
P
X

|ks |2 /2
+
ks aks |0i.
exp
|k0 s0 , k1 s1 , k2 s2 , k3 s3 , . . .i e
ks

(3.41)
Here |0i is the vacuum state, with no photons present. An ideal coherent
laser source is one in which just one of these modes has a macroscopic occupation, h
nks i = |ks |2 while the others have essentially zero occupation.
3 For convenience we will no longer write these as a
+ and a
, but just as a+ and a.
No ambiguity will arise from this simplified notation, but one must not forget that these
are operators and do not commute.

Bosonic Quantum Fields

79

In practical lasers, usually a few closely spaced k modes become macroscopically excited, and the system can randomly jump from locking onto
one mode to another due to the no n-linear optical pumping which maintains the macroscopic mode occupation. It is the frequency of these jumps
and the finite range of k values which limits the otherwise prefect optical
coherence of a typical laser light source. See Loudon (1979) for a more
detailed discussion of optical coherent states and their application to the
laser.

3.4

Bosonic Quantum Fields

In this section we shall introduce quantum field operators for the case of
boson particles. This will allow us to consider the quantum states of atomic
BEC and superfluid 4 He. The BEC is a weakly interacting Bose system,
while 4 He, as discussed in chapter 2, is a strongly interacting liquid of
boson particles. At the same time we shall also see how the coherent state
concept can also be applied to boson particles. This will enable us to define
the macroscopic wave function (r) which we need in order to describe the
condensate of particles. In this way we can generalize the simple intuitive
approach of chapters 1 and 2 into something which is both systematic and
rigorous.
Understanding BEC and superfluids obviously requires us to work with
many-particle quantum states for very large numbers of particles. In elementary quantum mechanics we would write a Schrodinger equation for a
wave function of N particles, in order to obtain an N body wave function
(r1 , r2 , . . . , rN ).

(3.42)

As discussed in Chapters 1 and 2, if we consider a system of N interacting

Bose atoms then we could in principle write a wave function (r1 , . . . , rN )
obeying the 3N-dimensional Schrodinger equation

H(r
1 , . . . , rN ) = E(r1 , . . . , rN )

(3.43)


X h
1 X
2 2
i + V1 (r) +
V (ri rj ).
2m
2

(3.44)

where
=
H

i=1,N

i,j=1,N

Here V1 (r) is the external potential, and V (r) is the particle-particle interaction. In the case of 4 He this interaction could be taken as the LennardJones potential of two helium atoms, Eq. ?? or Eq. ??, while for an atomic
BEC we would use the delta function interaction of Eq. ??. The fact that
the particles are bosons is expressed by the fact the the wave function must
be symmetric under permutation of any two of the particle coordinates
(. . . ri , . . . , rj . . .) = (. . . rj , . . . , ri . . .)
representing an exchange of the identical particles at ri and rj .

(3.45)

80

The Macroscopic Coherent State

This approach is feasible for systems with a few particles, such as say the
electrons in an atom, but it quickly becomes impractical in larger systems.
A much more useful method is to adopt the methods of quantum field
theory and introduce field operators which add or remove particles to the
system. If we consider a single particle in a box we know that the wave
functions are plane wave states
k (r) =

1
eik.r
(V )1/2

(3.46)

where V is the volume. We saw in chapter 1 that each of these single

particle states can be occupied by 0, 1, 2, 3 or any other finite number of
Bose particles. We denote these possible by the occupation number, nk . A
general quantum many body state of the system will be a superposition
of different N -body plane wave states. The complete basis of all possible
states can be represented by the set of all possible occupation numbers
of each plane wave. In exact analogy with the case of the laser we can
creation and annihilation operators a+
k and ak which increase or decrease
these occupation numbers. In order to satisfy the Bose symmetry condition
on the wave function, Eq. 3.45 it turns out that it is necessary that
[ak , a+
k ] = 1.
For two different plane wave states the occupation numbers are independent, and hence the creation operators must commute. Therefore the complete set of commutation relations are exactly as given in Eqs. 3.38-3.40.
Similarly the occupation operator is given by the number operator,
n
k = a+
k ak

(3.47)

These relations completely define the boson quantum field operators.

The set of many-particle states with all possible number operators,
{nk }, is a complete set of wave functions. But, just as in the case of the
harmonic oscillator, we can generate any nk by successive actions of the operators a+
k , starting with a ground state |0i. The interpretation is different
now. |0i is the vacuum state, i.e. a state with no particles present. Successive actions of a+
k add more particles to the system. Any many particle
quantum state can be represented by a superposition of states generated
in this way.
These field operators can also be cast into a real space form. We can

define the quantum field operators + (r) and (r)

which create and annihilate particles at point r. These can be defined by a Fourier transform of
the k space operators,

(r)

1 X ik.r

e ak
V k

(3.48)

Bosonic Quantum Fields

+ (r)

1 X ik.r +

e
ak .
V k

81

(3.49)

The function eik.r / V here is obviously just a free particle plane wave in
quantum state, k. The inverse Fourier transforms are
Z
1
3

eik.r (r)d
r
(3.50)
a
k =
V
Z
1
=
a
+
eik.r + (r)d3 r
(3.51)
k
V
Using these definitions and the Bose commutation law, one can show (exercise 5.4) that these real-space field operators have the commutation laws,
h
i

(r),
+ (r0 ) = (r r0 )
(3.52)
h
i

0) = 0
(r),
(r
(3.53)
h
i
+ (r), + (r0 ) = 0.
(3.54)

We can also represent any operator in terms of its actions on quantum

states described in terms of these operators. In particular the Hamiltonian,
Eq. 3.44 becomes


 2
Z 
h

=
2 + V1 (r) (r)
d3 r
H
+ (r)
2m
Z
1
+ (r0 )(r
0 )d3 rd3 r0 .
V (r r0 )+ (r)(r)
(3.55)
+
2

where the combination + (r)(r)

is obviously the density operator of particles at r.
It turns out to be convenient to always work in normal order, in which
all the creation operators are on the left and all the annihilation operators
are on the left. Commuting two of the field operators above we obtain
 2


Z 
h

=
+ (r)
d3 r
H
2 + V1 (r) (r)
2m
Z
1
(r
0 )d3 rd3 r0
+
V (r r0 )+ (r)+ (r0 )(r)
2
Z
1
0 )d3 rd3 r0
V (r r0 )+ (r)(r r0 )(r
(3.56)
+
2

The final term here arises from the commutator of [(r),

+ (r0 )], and reduces to
Z

V (0) d3 r+ (r)(r)
= V (0)N
(3.57)

82

The Macroscopic Coherent State

where
=
N

d3 r+ (r)(r)

(3.58)

is obviously just the operator for the total number of particles in the system.
V (0) term is a constant can be absorbed into the definition of the
The N
chemical potential, , and so we shall drop it from now on.
For a bulk fluid we can assume translational invariance, and ignore the
external potential V1 (r). Going back to k-space and we can use the Fourier
transforms Eqs. 3.48 and 3.49 to represent the Hamiltonian in terms of a +
k
and ak . The kinetic energy term is
T

=
=


h
2 2 +
(r) d3 r
+ (r)
2m

Z 
X
1
2 k2
ik.r
ik0 .r h
a
e
d3 r
a+
e
k
k0
V
2m
0

Z 
kk

Xh
2 k2
k

2m

a+
k ak .

(3.59)

The potential energy term is

Z
1
(r
0 )d3 rd3 r0
V =
V (r r0 )+ (r)+ (r0 )(r)
2
X Z
1
+
V (r r0 )a+
=
k1 ak2 ak3 ak4
2V 2
k1 k2 k3 k4

1
2V

3 3 0
ei(k1 .rk2 .r +k3 .r +k4 .r)
Zd rd r

+
a+
k1 ak2 ak3 ak4 k3 +k4 ,k1 +k2

k1 k2 k3 k4

V (r)ei(k4 k1 ).r d3 r.

Introducing the Fourier transform of the interaction

Z
1
V (r)eiq.r d3 r
Vq =
V

(3.60)

and making the replacements k1 k + q, k2 k0 q, k3 k0 , and

k4 k we can express the full Hamiltonian as
=
H

Xh
2 k2
k

2m

a+
k ak +

1 X
+
0
Vq a+
k+q ak0 q ak ak .
2 0

(3.61)

kk q

We can interpret the interaction term simply as a process in which a

pair of particles are scattered from initial states k, k0 to final states k + q,
k0 q. The momentum transferred between the particles is q, and the
matrix element for the process is Vq .

Off-Diagonal Long Ranged order

3.5

83

Off-Diagonal Long Ranged order

The field operators introduced in the previous section provide a general

way to discuss quantum coherence in condensates and superfluids. Even
though Eq. 3.61 is still too difficult to solve in general, we can still use
it explore the consequences of macroscopic quantum coherence in bosonic
systems.
Firstly, let us revisit the idea of the macroscopic quantum state, as
introduced for BEC and superfluid 4 He in chapters 1 and 2. Using the field
operators we can redefine the one-particle density matrix as
0 )i.
1 (r r0 ) h+ (r)(r
(3.62)

This definition is clearly more compact than the equivalent one given in
Chapter 2. Using the Fourier transformations Eqs. 3.48 and 3.49 we find
1 X i(k0 .r0 k.r) +
1 (r r0 ) =
e
hak ak0 i
V
kk0
1 X i(k.(rr0 ) +
e
hak ak i,
(3.63)
=
V
k

which is just the Fourier transform of the momentum distribution

nk ha+
k ak i

(3.64)

exactly as found in chapter 2.

Now let us consider the consequences of these definitions in the case of
a quantum coherent many-particle state. Just as in the case of the laser,
we can define a coherent state
|k1 , k2 , k3 . . .i

for any set of complex numbers ki . Using the standard properties of coherent states we find
nk h|k |2 i
(3.65)
and hence

1 (r r0 ) =

1 X ik.(rr0 )
e
|k |2 .
V

(3.66)

Typically we will be interested in quantum states where only one of the k

states is macroscopically occupied, (usually but not always k = 0). So suppose that state k0 has occupation N0 = |k0 |2 where N0 is a macroscopic
number (a finite fraction of the total particle number N ) and all the other
|ki |2 are small. For such a state we will have the momentum distribution
nk = N0 k,k0 + f (k)

(3.67)

where f (k is a smooth function of k. The corresponding density matrix is

Z
0
2
0
1 (r r ) = n0 +
d3 keik.(rr ) f (k)
(3.68)
(2)3
where n0 = N0 /V .

84

The Macroscopic Coherent State

These results are exactly the same as we found in chapter 2 by more elementary methods. The presence of the condensate is shown by the constant
contribution, n0 to the density matrix. If the function f (k is sufficiently
smooth, then its Fourier transform will vanish for large |r r0 | leaving just
the constant contribution,
0 )i n0
h+ (r)(r

(3.69)

as |r r0 | . This is what is meant by the term off diagonal long ranged

order (ODLRO), introduced by Oliver Penrose.
Fig. 3.3 shows the physical interpretation of the ODLRO in superfluids.
A particle can be annihilated at r, and absorbed into the condensate, while
a second particle is created at r0 out of the condensate. This process has
a quantum mechanical amplitude because of the quantum coherence of
the condensate, even when the points r and r0 are separated arbitrarily far
apart. In contrast, for a normal liquid (even a normal quantum liquid) these
processes would be incoherent except when r and r0 are close together.
Using the coherent state concept, there is one more step which we can
take. If the density matrix
0 )i
h+ (r)(r

(3.70)

is a constant, however far apart points r and r0 are, then is seems plausible
that we can treat the points as independent statistically. Then we can view
the above as an average of a product of independent random variables and
hence write is as a product of the two averages computed separately
0 )i h+ (r)ih(r
0 )i
h+ (r)(r

(3.71)

for |r r0 | .
If we were to work in the standard fixed particle number, N , many-body
formulation of quantum mechanics then averages such as h+ (r)i would be
automatically zero, since acting on any N particle state + (r)|N i is an
N + 1 particle state, and is necessarily orthogonal to hN |. But if we are
working in the basis of coherent states, then there is no such problem.
The coherent states have definite phase, not definite N , and this type of
expectation value is perfectly well allowed.
Therefore we can say that there is an order parameter or macroscopic
wave function, defined by

0 (r) = h(r)i.

(3.72)

In term of this function we see that4

1 (r r0 ) = 0 (r)0 (r0 )

(3.73)

4 The creation operator

+ (r) is just the Hermitian conjugate of (r)

and so 0 (r) =
h+ (r)i.

The Weakly Interacting Bose Gas

85

for |r r0 | . In a translationally invariant system (the condensation

occurs in the k = 0 state), we must therefore have
0 (r) =

n0 ei

(3.74)

where is an arbitrary constant phase angle.

Of course this phase is nothing more than the XY model phase angle
introduced in chapter 2, Fig. ??. But now we can see that its true meaning
is that we have a coherent quantum state in which the k = 0 state has a
macroscopic occupation.
Since we have not yet made any connection to the Hamiltonian, Eq. 3.61
it is impossible to prove from these arguments that such a coherent quantum state will be stable. But at least we can see how to construct coherent
many-particle wave functions in which a definite order parameter phase
is possible. possible. In the case of the ideal Bose condensate discussed in
chapter 1 one can still work in the fixed particle number representation, and
so there is no advantage to explicitly introduce a coherent state formalism.
But as soon as there are any interactions, however weak, the coherent state
approach becomes advantageous. In the next section we shall consider the
weakly interacting bose gas, in which the advantages of the coherent state
approach can be seen explicitly.

3.6

The Weakly Interacting Bose Gas

The theory of the weakly interacting Bose gas was originally developed by
Bogoliubov in in late 1940s. It was developed as a theory of superfluid
helium, although as we have seen, for 4 He the interatomic interactions are
very strong. In this case the theory has some qualitative features which
agree with experimental properties of 4 He, most notably the linear phonon
like quasiparticle excitation spectrum, k = ck, at small wave vectors in
fig. ??. But it fails to reproduce other important features, such as the
roton minimum in the spectrum. On the other hand, the theory is believed
to be close to exact for the case of atomic BEC, since the conditions under
which it is derived are close to the experimental ones.
First of all we shall assume that we are at zero temperature, or close to
zero, so that the system is close to its ground state. We assume that the
system is in a coherent many-particle state, characterized by a macroscopic
wave function 0 (r, as in Eq. 3.72. Suppose that the many-particle quantum
state ,|i, is an ideal coherent state at zero temperature. Then it is an
eigenstate of the annihilation operator,

(r)|i
= 0 (r)|i.

(3.75)

We can view this as a trial many-particle wave function, and we will vary
the parameter 0 (r) so as to variationally minimize the total energy. The

86

The Macroscopic Coherent State

variational energy is found by taking the expectation value of the Hamiltonian,


 2 2
Z
h

3

=
r
+ V1 (r) (r)d
+ (r)
H
2m
Z
1
(r
0 )d3 rd3 r0 .
+
V (r r0 )+ (r)+ (r0 )(r)
(3.76)
2
Here the single particle potential V1 (r) is the effective external potential of
the atom trap. In the case of bulk superfluids this is obviously zero.
Using the definition of the coherent state |i from Eq. 3.75 we can find
immediately that the variational energy is
E0

=
=

h|H|i
 2 2

Z
h

0 (r)
+ V1 (r) 0 (r)d3 r
2m
Z
1
V (r r0 )0 (r)0 (r0 )0 (r)0 (r0 )d3 rd3 r0 .
+
2

(3.77)

We can find the minimum by functional differentiation, exactly as for the

Ginzburg-Landau equation. Setting
E0
=0
0 (r)
yields,
 2 2

Z
h

+ V1 (r) ) 0 (r) + V (r r0 )0 (r)0 (r0 )0 (r0 )d3 r0 = 0.

2m
(3.78)
The parameter is a Lagrange multiplier, necessary to maintain a constant
normalization of the macroscopic wave function
Z
N0 = |0 (r)|2 d3 r.
(3.79)
Clearly Eq. 3.78 is of the form of an effective Schrodinger equation

 2 2
h

+ V1 (r) + Vef f (r) 0 (r) = 0
(3.80)

2m
where is the chemical potential, and the effective potential is
Z
Vef f (r) = V (r r0 )|0 (r0 )|2 .d3 r0
This Schrodinger equation is exactly the Gross-Pitaevski equation again,
which we derived by a different method in chapter 2.

The Weakly Interacting Bose Gas

87

To examine the accuracy of this ground state, and to examine the low
energy excited states, we need to consider possible quantum states which
are close to our trial ground state |i, but which do not deviate from it
too much. We need to consider many-particle states which do not exactly
obey the coherent state conduction, Eq. 3.75, but for which it is nearly
obeyed. Bogoliubov introduced an elegant method so achieve this. He assumed that the field operators can be expressed approximately as their
constant coherent state value, plus a small deviation,

(r)
= 0 (r) + (r).

(3.81)

From the commutation relations for the field operators, it is easy to see
that

[ (r),
+ (r0 )] = (r r0 ),
(3.82)

and so the deviation operators (r)

and + (r) are also bosonic quantum
5

fields. We can then rewrite the hamiltonian in terms of 0 (r) and (r).

We can group terms according to whether (r) occurs never, once, twice,
three times or four,
=H
0 + H
1 + H
2 + . . .
H
(3.83)
and we assume that it is valid to simply ignore terms higher than second
order.
0 is simply the original coherent
In this expansion, the first term, H
state energy, Eq. 3.77 which we can minimize using the Gross-Piaevskii
equations. Furthermore, if we have variationally minimized the energy there
will be no corrections to the energy to linear order in the deviation operators

(r).
the first significant correction term is quadratic in the deviation
operators. Several terms contribute, but the net result is that
 2 2

Z
h

3

2 =
H
+ (r)
+ V1 (r) (r)d
r
2m
Z

1
+
V (r r0 )d3 rd3 r0 + (r) + (r0 )0 (r)0 (r0 )
2
0
0

+ 0

+2 + (r)0 (r0 ) (r)

0 (r ) + 20 (r) (r ) (r)0 (r )


0) .
(r
(3.84)
+0 (r)0 (r0 ) (r)
One can visualize the meaning of these various terms in terms of the
simple diagrams shown in Fig. 3.4. There are four distinct terms. The first,
shown in panel (a), corresponds to the creation of two particles, one at
r and r0 under the action of the potential V (r r0 ). Of course they are

5 A different way to see this is to imagine that we simply translate the origin of the
coherent state complex plane, Fig. 3.2. Shifting the origin from = 0 to = 0 (r) we
can then describe states which are near to |i in terms of the coherent states which
are in the vicinity of the point 0 (r).

88

The Macroscopic Coherent State

not really created, but scattered out of the condensate. The second term
corresponds to the scattering of an existing quasiparticle by interaction
with particles in the condensate. It has a extra factor of 2 since we can find
an identical diagram with r and r0 interchanged. The third diagram (c)
is also a scattering of an existing quasiparticle, but now the quasiparticle
at r is absorbed into the condensate, while at the same time a second
quasiparticle appears at r0 . Again r and r0 can be interchanged. leading to
an extra factor of 2. The final diagram, (d), shows two quasiparticles being
absorbed into the condensate.
To keep the algebra manageable let us specialize to the case (relevant
for the atomic BEC) of a pure contact interaction,
V (r r0 ) = g(r r0 ).

(3.85)

For
p convenience we will also assume that 0 (r) is real, and equal to 0 (r) =
n0 (r), where n0 (r) the spatially varying condensate density in the atom
2 simplifies to
trap. The quadratic Hamiltonian H
 2 2

Z
h

3

2 =
H
+ (r)
+ V1 (r) (r)d
r
(3.86)
2m
Z


g
+ (r)

n0 (r) + (r) + (r) + 4 + (r) (r)

(r)
d3 r.
+
2
This Hamiltonian is a quadratic form in the operators, and it turns
out that all such quadratic Hamiltonians can be diagonalized exactly. The
procedure makes use of the Bogoliubov transformation to eliminate the

anomalous terms (r)

(r)
and + (r) + (r). Define a new pair of
operators by
+ v(r) + (r)
(r)

= u(r) (r)
(3.87)
+ (r)

u (r) + (r) + v (r) (r).

These are again bosonic quantum field operators if



(r),

+ (r) = (r r0 ),

(3.88)

(3.89)

which is true provided that the functions u(r) and v(r) are chosen to obey
|u(r)|2 |v(r)|2 = 1.

(3.90)

The general solution becomes quite complicated, so let us specialize to

the case of a uniform Bose liquid, without the atom trap potential V(r).
Assuming that the macroscopic wave function is also just a constant, 0 =

n0 , and going to k-space Eq. 3.86 becomes


X h
n0 g + +
2 k2
+
2 =
a
+
a
a
)
. (3.91)
+
4a
a
+
a
)a+
(a
(
H
k k
k k
k k
k k
2m
2
k

The Bogoliubov transformation in k-space gives the new operators

bk

u k ak + v k a+
k

(3.92)

The Weakly Interacting Bose Gas

b+
k

uk ak + vk ak

89

(3.93)

where uk and vk obey

|uk |2 |vk |2 = 1.

(3.94)

The idea is to rewrite the Hamiltonian in terms of these new operators

and then to vary the parameters uk and vk to make it diagonal. In particular
+
it is necessary to make the coefficients of the anomalous terms, b+
k bk and
bk bk equal to zero. Since the calculation is lengthy we shall just quote the
results. It turns out that when the anomalous terms are eliminated, the
new quasiparticle excitations created by the b+
k operators have the energy
spectrum
 2 2 1/2  2 2
1/2
h
k
h
k
Ek =
+ 2n0 g
.
(3.95)
2m
2m

For small |k| (less than 4n0 gm/h) the spectrum is linear
Ek ck

(3.96)

where the phonon velocity is

c=

h
2 n0 g
m

1/2

(3.97)

This Bogoliubov quasiparticle spectrum is sketched in Fig. 3.5. The

spectrum is linear at small k, and joins smoothly onto the independent
particle energy h
2 k 2 /2m for large k. The success of this Bogoliubov theory
is that it explains the linear dispersion in the excitation spectrum near to
k = 0, as we saw in the case of superfluid 4 He, Fig. ??. In chapter 2 we
have also seen that a linear spectrum is necessary to prevent quasiparticle
scattering from the container walls, and hence to maintain a dissipationless
superflow. Therefore we can conclude that even a very weakly interacting
Bose gas is a superfliud, even when the ideal non-interacting Bose gas is
not. The critical velocity for the superfluid will be less than c, and so
by Eq. 3.97 the critical velocity will approach zero in the limits of weak
interaction g 0, or low density n0 0.
Of course there are still huge differences between the Bogoliubov quasiparticle spectrum in Fig. 3.5 and the experimental quasiparticle spectrum
of Fig. ??. Most importantly, there is no roton minimum. The Bogoliubov
spectrum is linear at small k and becomes equal to the free particle spectrum h
2 k 2 /2m for large k, but it has no minimum at intermediate k. The
roton minimum is therefore an effect emerging only in a strongly interacting
Bose liquid. This difference has another effect, namely that the Bogoliubov
spectrum Ek has a slight upwards curvature at small k, while the true
spectrum has a slight downward curvature. This curvature means that a
quasiparticle of momentum h
k and energy Ek has a non-zero cross section

90

The Macroscopic Coherent State

to decay into three quasiparticles of lower energy and momentum. Therefore the Bogoliubov quasiparticle is not an exact eigenstate, but only a long
lived resonance.6

3.7

Coherence and ODLRO in Superconductors

The above ideas were originally introduced to explain the strongly interacting Bose superfluid 4 He. But they also apply to superconductors. However
in this case the argument must be modified in order to take account of
the fact that the electrons in a superconductor are fermions. Although it
is perfectly possible to define coherent states for fermions, they are not
immediately useful for superconductivity. This is because a single fermion
state can only ever be occupied by either 0 or 1 fermions, due to the exclusion principle. Therefore it is not possible to have a macroscopic number
of fermions in a single plane wave state.
It was Robert Schrieffer who first managed to write down a coherent many-particle wave function for fermions. His colleauges Bardeen and
Cooper had already realized that electrons bind into pairs in a superconductor. There was even an earlier theory by Schafroth Blatt and Butler in
which superconductivity was seen as a Bose condensate of electron pairs. 7
But Bardeen Cooper and Schrieffer (BCS) knew that the pairs of electrons
in superconductors could not be simply treated as bosons. The problem
was to write down a valid many-body wave function for the electrons in
which each electron participate in the pairing. The brilliant solution Schrieffer discovered was effectively another type of coherent state, similar to
those we have already seen. But the key point is to have a coherent state
in which a macroscopic number of pairs are all in the same state.
Nevertheless even in the BCS theory we have a form of ORLRO and
an order parameter, as we shall discover in this section. Their appearance
is a quite general phenomenon. Here we concentrate on the most general
statements about the ODLRO, and leave the actual details of the BCS
theory, and its specific predictions until the next chapter.
6 It makes perfect sense here to borrow the concepts of elementary particle physics,
and to talk about one particle decaying into a set of others. Just as in particle physics
we can call such particles resonances. In this context the Bogoliubov quasiparticles are
just the elementary particles of the Bose gas, and the background condensate is the
analog of the vacuum.
7 Unfortunately this theory was not able to make quantitative predictions for the
superconductors which were known at that time, and it was generally discarded in favour
of the Bardeen Cooper Schrieffer theory which was very much more successful in making
quantitative predictions. The electron pairs in the Bardeen Cooper and Schrieffer theory
are not bosons. In general a pair of fermions is not equivalent to a boson. In the BCS
case the bound electon pairs are very large, and so different pairs strongly overlap each
other. In this limit it is not possible to describe the pair as a boson, and so the BCS
theory is not normally expressed in terms of bose condensation. However in recent years
theories based upon the ideas of the Schafroth, Blatt and Butler theory have been
somewhat revived as a possible models of high Tc superconductors.

Coherence and ODLRO in Superconductors

91

First we must define the correct quantum field operators to describe

the conduction electrons in a solid. In single-particle quantum mechanics
of solids we know that the wave functions are Bloch waves
nk (r) = eik.r unk (r).

(3.98)

Here the crystal wave vector k must lie within the first Brillouin zone. For
simplicity of notation we shall assume that only one of the energy bands,
n, is relevant (i.e. the one at the Fermi surface) and so we will drop the
index n from now on.
A particular Bloch state of a given spin , k (r), can either be empty or
occupied by an electron. A quantum state of N particles can be specified
by saying for each individual state whether it is occupied or not. This
is the occupation number representation of quantum mechanics. We can
introduce operators which change these occupation numbers. Suppose we
label the state as |0i or |1i if the given Bloch state k (r). Then we can
define operators which change the occupation numbers
c+ |0i
c|1i

= |1i
= |0i.

These are obviously similar to the similar Bose field operators, or the Harmonic oscillator ladder operators, a
+ and a
. The only difference is that the
state of a harmonic oscillator can have n = 0, 1, 2, . . ., while for fermions
the occupation number is only 0 or 1. The operators c+ and c are again
called the creation and annihilation operators. Since electrons are neither
being created or annihilated in solids these names may seem a bit misleading, however the operators work the same way in particle physics where
particles are indeed being created or annihilated back into the vacuum (eg.
electron-positron pairs annihilating each other). In a solid one can imagine
adding electrons to the solid (e.g. with an external current source into the
surface), or removing them from the solid (eg. in photo-emission).
These operators have a couple of important properties. Firstly, just as
for bosons, the combination c+ c measures the occupation number |ni since
c+ c|0i
+

c c|1i

|1i

i.e. c+ c|ni = n|ni. Secondly the exclusion principle implies that we cannot
occupy a state with more than one fermion, and hence c+ c+ |ni = 0 and
cc|ni = 0. This fermion nature of the particles means that
{c, c+ } cc+ + c+ c = 1

(3.99)

where {A, B} = AB + BA is the anti-commutator of operators A and B.

The antisymmetry of the many-particle wave function for fermions means

92

The Macroscopic Coherent State

that the operators for different Bloch states or spin states anti-commute,
and so in general we have the anti-commutation relations,
{ck , c+
k0 0 }
{ck , ck0 0 }

+
{c+
k , ck0 0 }

=
=

k,k0 0
0

(3.100)
(3.101)

(3.102)

where = 1 denotes the two different spin states. Naturally, just as for
bosons, we can also represent these operators in real space, by a Fourier
transformation, obtaining
{ (r), +0 (r0 )}
{ (r), 0 (r0 )}

{+ (r), +0 (r0 )}

=
=
=

(r r0 )0

(3.103)

(3.104)

0.

(3.105)

In some sense this is also like a Bose condensate of the Cooper pairs.
However because the fermion antisymmetry is fully taken account of the
wave function leads to quite different predictions than one might just expect
with a Bose condensation picture. The BCS predictions turned out to be
extremely accurate numerically, despite it being a simplified mean-field
model wave function. Its accuracy lies in the large size of the Cooper pairs,
with their characteristic size (the coherence length) being much larger
than the typical inter-electron spacings in a solid rs , where N/V = n =
4rs3 /3.
Now the mathematical challenge which BCS needed to solve was to
write down a many-particle wave function in which every electron near the
Fermi surface participates in the pairing. They knew that a single pair of
electrons would bind into a spin singlet state with two body wave function
1
(r1 1 , r2 2 ) = (r1 r2 ) (| i | i)
2

(3.106)

They first wrote down a many-particle wave function in which every particle
is paired,
(r1 1 , . . . , rN N )

1 X

(1)P (r1 1 , r2 2 )(r3 3 , r4 4 ) . . .

N! P

. . . (rN 1 N 1 , rN N ).

(3.107)

Here the sum over P denotes the sum over all the N ! permutations of
the N particle labels r1 1 , r2 2 etc. The sign (1)P is positive for an
even permutation and 1 for odd permutations. This alternating sign is
necessary so that the many-body wave function has the correct fermion
antisymmetry
(. . . , ri i , . . . , rj j , . . .) = (. . . , rj j , . . . , ri i , . . .).

(3.108)

Coherence and ODLRO in Superconductors

93

But this fixed N many-body quantum mechanics is unwieldy. By fixing

N we cannot have a definite overall phase, unlike in a coherent state representation. But if we use coherent states then it is possible to describe a
condensate with a definite phase, and the key step achieved by Schrieffer
was to find a way to write down a coherent state of fermion pairs. First we
need to write down operators which create or annihilate pairs of electrons.
Defining,
Z
Z
(3.109)
+ (R) d3 r (r)+ (R + r/2)+ (R r/2).
we can see that acting with this operator on a quantum state with N
electrons gives a new quantum state with N + 2 electrons. It creates a spin
singlet electron pair, where the electrons are separated by r and with centre
of mass at R.8
Naively one might regard such a fermion pair as a boson. But this is
not correct. If we try to evaluate the commutator we find that


(R),

+ (R0 ) 6= (R R0 )
(3.110)
[(R),

(R
0 )] 6= 0
(3.111)
 +

(R), + (R0 ) 6= 0.
(3.112)
The operators only commute when R and R0 are far apart, corresponding
to non-overlapping pairs. For this reason we cannot simply make a Bose
condensate out of these pairs.
But even though these are not true boson operators, we can still define
the analogue of ODLRO corresponding to Bose condensation. Now it is a
state of ODLRO of Cooper pairs. We can define a new density matrix by,
1 (R R0 ) = h+ (R), (R
0 )i.

(3.113)

This is a one particle density matrix for pairs, and so it is related to the
two particle density matrix for the electrons
2 (r1 1 , r2 2 , r3 3 , r4 4 ) = h+1 (r1 )+2 (r2 )3 (r3 )+4 (r4 )i.

(3.114)

Inserting the definition of the pair operator gives the pair density matrix
in terms of the electron one
Z
r
r0
r0
r
1 (R R0 ) = (r)(r0 )2 (R + , R , R0 , R0 + )d3 rd3 r0 .
2
2
2
2
(3.115)
The pair wave function (r) is a quantum mechanical bound state, and so it
will become zero for large |r|. If the scale of its range is defined by a length,
8 Note that we assume here that the wave function for the bound electron pair obeys
(r) = (r). This is simply because a spin singlet bound state wave function must be
even under exchange of particle coordinates, because the spin single state is odd under
exchange.

94

The Macroscopic Coherent State

0 , (which turns out to be the BCS coherence length of the superconductor),

then the the main contributions to the pair density matrix comes from the
parts of electron density matrix where r1 and r2 are separated by less than
0 , and similarly r3 and r4 are separated by less than 0 . But the pair r1 ,r2
and the pair r3 ,r4 can be separated by any arbitrarily large distance.
Now we can can have ODLRO in the pair density matrix, provided that
1 (R R0 ) const.

(3.116)

as |R R0 | . The BCS theory does therefore correspond to a similar macroscopic quantum coherence to the ordinary theory of ODLRO in
super-fluids. However is is an ODLRO of Cooper pairs, not single electrons.
In terms of the electron density matrix 2 (r1 1 , r2 2 , r3 3 , r4 4 ) this
ODLRO corresponds to the density matrix approaching a constant value
in a limit where the two coordinates r1 and r2 are close to each other, and
r3 and r4 are close, but these two pairs are separated by an arbitrarily large
distance. Fig. 3.6 illustrates this concept.
Following the same approach to ODLRO as in the weakly interacting
Bose gas, we can make the assumption that very distant points R and R 0
should behave independently. Therefore we should be able to write
1 (R R0 ) h+ (R)ih(R
0 )i

(3.117)

for |RR0 | . We can therefore also define a macroscopic wave function

by
(R) = h(R)i.

(3.118)

Effectively this is the Ginzburg-Landau order parameter for the superconductor.

Of course, we have still not actually shown how to construct a manyelectron quantum state which would allow this type of Cooper pair ODLRO.
We leave that until the next chapter. Nevertheless we can view the above
discussion as stating the requirements for the sort of quantum state which
could exhibit superconductivity. It was possibly the most significant achievement of the BCS theory that it was possible to explicitly construct such a
non-trivial many-body quantum state. In fact we can see immediately from
Eq. 3.118 that we must have a coherent state with a definite quantum mechanical phase , and conversely that we should not work with fixed particle
number N . At the time of the original publication of the BCS theory in
1957 this aspect was one of the most controversial of the whole theory. The
explicit appearance of the phase also caused concern, since it appeared
to violate the principle of gauge invariance. It was only the near perfect
agreement between the predictions of the BCS theory and experiments, as
well as clarification of the gauge invariance issue by Anderson and others
which led to the final acceptance of the BCS theory.

The Josephson Effect

3.8

95

The Josephson Effect

The Josephson effect is a direct physical test of the quantum coherence

implied by superconducting ODLRO. Soon after the BCS theory was published Brian Josephson, then a young PhD student at Cambridge, considered the effect of electrons tunnelling between two different superconductors.9
Consider two superconductors, separated by a thin insulating layer,
as shown in Fig. 3.7. If each superconductor has a the macroscopic wave
function as defined in, 3.118, then we can assign definite values to the wave
functions on either side of the tunnel barrier, say L and R for the left
and right hand side respectively,
L (RL ) = hL|(R
L )|Li

R (RR ) = hR|(R
R )|Ri
where RL and RL are points on either side of the tunnel barrier, and |Li
and |Ri are the many-particle quantum states of the superconductors on the
left and right side respectively. Josephson assumed that electron tunnelling
takes place for electrons crossing the barrier. In terms of electron field
operators we can write
XZ
=
H
T (rL , rR )(+ (rL ) (rR ) + + (rR ) (rL )d3 rL d3 rR (3.119)

as the operator which tunnels electrons of spin from points rL and rR

on either side of the barrier. Using the BCS many-body wave functions
on either side of the junction and second order perturbation theory in the
Josephson was able to find the remarkable result
tunnelling Hamiltonian H,
that a current flows in the junction, given by
I = Ic sin (1 2 )

(3.120)

where 1 and 2 are the phases of the macroscopic wave functions on either
side of the junction.
The details of how Josephson found this result will not be important
here. But it is worth noting very roughly how this might come about. If we
we can
consider effects to second order in the tunnelling Hamiltonian H
2

see that H contains many terms, but includes some four fermion terms of
the following form,
2 T 2(
H

+ (rL )+0 (r0L ) (rR )0 (r0R )

9 Brian Josephson received the Nobel prize in 1973 for this discovery, possibly one of
the few Nobel prizes to have arisen from a PhD project. After Josephons first prediction
of the effect in 1962, established experts in the superconductivity field at first objected
to the theory, believing that the effect was either not present or would be too weak
to observe. But Josephsons PhD adviser Anderson encouraged Rowell to look for the
effect experimentally. Anderson and Rowell together announced the first observation of
the Joephson effect in 1963.

96

The Macroscopic Coherent State

++ (rR )+0 (r0R ) (rL )0 (r0L ) + . . .).

(3.121)

The net effect of the first of these terms is to transfer a pair of electrons
from the right hand superconductor to the left. Conversely the second term
transfers a pair from left to right. Because the many-body states on both
sides of the junction are coherent pair states, these operators will have
non-zero expectation values consistent with the ODLRO
hL|+ (rL )+0 (r0L )|Li 6= 0
hR| (rR )+0 (r0 )|Ri 6= 0.

(3.122)
(3.123)

In fact we expect the first of these expectation values proportional to ei1

and the second to ei2 . The overall quantum mechanical amplitude for
tunnelling a pair from right to left is thus has a phase ei(2 1 ) . The reverse
process, tunnelling from left to right has the opposite phase. When they
are added together the net current is proportional to sin (1 2 ) as given
by Eq. 3.120.
Eq. 3.120 shows that the current flows in response to the phase difference
1 2 . Therefore it is in some sense a direct proof of the existence of such
coherent state phases in superconductors. The proportionality constant I c
is the maximum Josephson current that can flow, and is called the critical
current of the junction.
For currents I below Ic the Josephson current is perfectly dissipationless, i.e. it is a supercurrent. But if current is driven to a higher value,
I > Ic , a finite voltage drop V develops across the junction. Therefore the
typical I V characteristic of the junction is as shown in Fig. 3.8.
In the case that I > Ic Josephson found a second surprising consequence of this tunnel current. The finite voltage difference V between the
superconductors, means that the macroscopic wave functions become time
dependent. Using a version of the Heisenbeg equation of motion for the left
and right hand side macroscopic wave functions,
L (t)
t
R (t)
ih
t
ih

2eVL L (t)

2eVR R (t),

(3.124)

Josephson was able to show that the finite voltage drop V = VL VR leads
to a steadily increasing phase difference,
(t) = (0) +

2eV
t
h

and hence the Josephson current,



2eV
I = Ic sin (0) +
t
h

(3.125)

(3.126)

The Josephson Effect

97

oscillates at a frequency

2eV
.
(3.127)
h

This surprising effect is called the a.c. Joesphson effect (in contrast to the
d.c. Josephson effect for I < Ic , V = 0).
The experimental observation of the a.c. Josephson effect not only confirmed the theory and the validity of the BCS macroscopic quantum coherent state, but also provided another direct empirical confirmation that the
relevant particle charge is 2e and not e. Thus it again confirmed the Cooper
pairing hypothesis. Even more surprising is that the Josephson frequency
appears to be exactly given by Eg. 3.127. In fact, it is so accurate that
the Josephson effect has been incorporated into the standard set of measurements used to define the SI unit system. By measuring the frequency
(which can be measured with accuracies of one part in 1012 or better) and
the voltage V one can obtain the ratio of fundamental constants e/h with
high precision. Alternatively one can use the given values of e/h and the
Josephson effect to define a reliable and portable voltage standard, V .
The Josephson effect is also at the heart of many different practical
applications of superconductivity. One of the simplest devices to make is
a SQUID ring; where SQUID stands for Superconducting QUantum Interference Device. This is simply a small (or large) superconducing ring in
which there are two weak links. Each half of the ring is then connected to
external leads, as shown in Fig. 3.9. By weak link one can mean either a
tunnel barrier, such as Fig. 3.7 (an SIS junction), or a thin normal metallic
spacer (an SNS junction). The current through each junction depends on
the phase difference across it and so,
=

I = Ic1 sin (1 ) + Ic2 sin (2 )

(3.128)

is the sum of the currents of each Josephson junction. The phases differences
1 , 2 correspond to the macroscopic wave function phases differences
at the points to the left and right of each junction in Fig. 3.9.
In the junctions are perfectly balanced, so Ic1 = Ic2 and a small external
current I (< Ic ) is applied to the SQUID, then we would expect an equal
steady state current to flow in both halves of the ring, and a constant phase
difference equal to = sin1 (I/2Ic ) will develop across both junctions.
But this is only true if there is no magnetic flux through the ring. Using the
principle of gauge invariance we find that a flux implies that the phase
differences are no longer equal.
Z
=
B.dS
I
=
A.dr
I
2e
().dr
=
h

98

The Macroscopic Coherent State

=
=

2e
(1 2 )
h

20 (1 2 ) .

(3.129)

Therefore the magnetic flux through the ring leads to a difference between
the two phases. For a balanced SQUID ring system we can assume that
1

.
0
+

(3.130)

therefore the total current in the SQUID is

I

Ic sin (1 ) + Ic sin (2 )





+ Ic sin
= Ic sin +
0
0



= 2Ic sin () cos

.
0

(3.131)

The critical current is therefore modulated by a factor depending on the

net flux through the ring,10





Ic () = I0 cos
.
(3.132)
0 

This modulation of the observed SQUID ring critical current is shown

in Fig. 3.10. This current is essentially an ideal Fraunhoffer interference
pattern, exactly analogous to the interference pattern one observes in optics with Youngs two slit experiment. Here the two Josephson junctions
are playing the role of the two slits, and the interference is between the supercurrents passing through the two halves of the ring. The supercurrents
acquire different phases due to the magnetic field. One can say that this
effect is also analogous to the Ahronov Bohm effect (Feynman 1964), in
which a single electron passes on either side of a of a solenoid of magnetic
flux.
The SQUID device provides a simple, but highly accurate, system for
measuring magnetic flux. Since the flux quantum 0 is only about 2
1015 W b in the SI unit system, and one can make SQUID devices approaching 1cm2 in area, it is in principle possible to measure magnetic
fields to an accuracy below B 1010 T . In particular it is easy to measure
changes in field to this accuracy by simply counting the number of minima
in the SQUID critical current.
10 The SQUID ring critical current will always have the same sign as the driving current
and hence the modulus signs appearing in Eq. 3.132.

Macroscopic Quantum Coherence

3.9

99

Macroscopic Quantum Coherence

To what extent does the Joesphson effect or the SQUID represent true
evidence quantum coherence? Even though it is very cold, operating at a
temperature below Tc , say at one or two degrees Kelvin, a SQUID ring
is hardly isolated from its environment. In fact one can make perfectly
good SQUIDs using high temperature superconductors. These operate at
temperatures of over 100K. SQUID rings are usually fabricated on some
sort of insulating substrate, and are usually subject to normal external
electromagnetic noise in the laboratory, unless well shielded.
Given this relatively noisy thermal environment, the SQUID shows a
remarkable insensitivity to these effects. This is fundamentally because the
macroscopic wave function (r) which we have defined above, and its phase
, is not a true wave function in the sense of elementary quantum mechanics. In particular it does not obey the fundamental principle of superposition, and one cannot apply the usual quantum theory of measurement or
Copenhagen interpretation to it. The macroscopic wave function behaves
much more like a thermodynamic variable, such as the magnetization in a
ferromagnet, than a pure wave function, even though it has a phase and
obeys local gauge invariance.
But since the early 1980s there have been attempts to observe true
quantum superpositions in superconductors (Leggett 1980). I.e. can one
construct a Schrodinger cat like quantum state? For example is a state
such as
1
|i = (|1 i + |2 i)
(3.133)
2
meaningful in a SQUID ring? If |1 i and |2 i are two pure quantum states,
then the general principle of linearity of quantum mechanics, implies that
any superposition such as |i must also be a valid quantum state. Only by
measuring some observable can one collapse the wave function and find
out whether the system was in |1 i or |2 i. for small system, such as single
atoms or photons, such superpositions are a standard part of quantum
mechanics. But in his famous 1935 paper Schrodinger showed that this
fundamental principle leads to paradoxes with our everyday understanding
of the world when we apply it to macroscopic systems such as the famous
cat in a box!
Even since Schrodingers paper it has not been clear where to put the
dividing line between the macrosopic world (governed by classical physics
and without superposition) and the microscopic (governed by quantum
mechanics). The ideas of decoherence provide one possible route by which
quantum systems can acquire classical behaviour. Interactions with the
environment lead to entanglement between the quantum sates of the system
and those of the environment, and (in some of the most modern approaches)
quantum information is lost.
In this context one can say that a large SQUID ring, say 1cm (or 100m!)
in diameter will be subject to decoherence from its environment, and hence

100

The Macroscopic Coherent State

will be effectively in the classical realm. But if one makes the ring smaller,
or operates at lower temperatures, is there a regime where true quantum
superpositions occur? In fact the answer to this question is yes! Indeed
strong evidence for quantum superposition states has now been seen in
three different systems.
The first system where quantum superposition states were observed was
in Bose-Einstein condensates, in 1996 (Ketterle 2002). Since these exist in
a very low temperature state (a micro Kelvin or less) and are isolated from
most external thermal noise sources (since they are trapped in vacuum) one
could expect a high degree of quantum coherence to occur. Indeed it has
proved possible to split a single condensate into two halves, in a similar
way to which a beam splitter separates photons. When the two halves of
the condensate are subsequently brought back together again, then one
observes an interference pattern. The experiment is effectively the exact
analogue of the Youngs slit interference experiment with light.
The second type of experiment which showed true quantum interference was done using a superconducting island or Cooper pair box. This
consists of a small island of superconductor (Al was used) of order 0.1m
on a side, as shown in Fig. 3.12. Operating at temperatures of a few mK,
well below the superconducting Tc , the quantum states of the box can be
characterized entirely by the number of Cooper pairs present. For example
the box can have a state |N i of N Cooper pairs, or a state |N + 1i etc.
The energies of these different states can be manipulated through external
voltage gates, since they are states of different total electronic charge on
the box. The analogue of the Schrodinger cat state for this system is to
place it in a superposition, such as
1
|i = (|N i + |N + 1i) .
2

(3.134)

Nakamura Pashkin and Tsai (1999) were able to demonstrate the presence of just such superpositions in their device. By connecting the Cooper
pair box to a second superconductor, via a Josephson junction, they effectively allowed quantum mechanical transitions between these two states,
as Cooper pairs tunnel onto or off the island. By pulsing external voltage
gates connected to the system they observed beautiful interference fringes
associated with the superposition states, as shown in Fig. 3.13. The figure shows the final charge on the box (i.e. N + 1 or N ) as a function of
the voltage pulse amplitude and duration. The results observed oscillations
agree excellently with the theoretical predictions based on the existence of
macroscopic quantum superposition states.
The third systems in which macroscopic quantum coherence has been
demonstrated are small superconducting rings, as in Fig. 2.6. As we saw
in chapter 3, a superconducting ring has a set of different ground states
corresponding to different winding numbers of the order parameter around
the ring. These can again be represented by abstract many-particle states

Summary

101

such as |ni and |n + 1i. In a large ring there will be no way for the system
to tunnel from one of these states to another, but if the ring is made small
enough (below 1m in diameter) such transitions become possible. Two
experiments in the year 2000 observed direct evidence for quantum mechanical coherence in such syaytems (van der Wal et al. 2000, (Friedman et
al. 2000). (Earlier reports unconfirmed reports of the effect were also made).
In any case these results are fascinating, since they represent a coherent
quantum tunnelling of a system containing 1010 or more electrons. Fig. 3.14
shows an electron microscope image of a small superconducting circuit, approximately 2m across, superimposed on an image of the quantum (Rabi)
oscillations observed in this circuit (Chiorescu et al. 2003). These oscillations demonstrate the existence of quantum superposition states in which
two the system is simultaneously in two macroscopically different quantum
states!
Do these quantum superposition states have any practical uses? In recent years there has been a huge growth in the field of quantum information and quantum computation. The idea is that information as used
and manipulated in computer bits is actually always a physical quantity,
e.g. the charges on the capacitors in a RAM computer memory. Therefore
it is subject to the laws of physics. But conventional computer bits are
essentially based on classical physics. For example a computer bit can be
measured without disturbing its state. But is we imagine eventually making
the physical computer bits smaller and smaller with each new generation
of computer, then eventually we will have to use devices which are so small
that quantum mechanics applies, not classical physics. For such a quantum bit, or qubit, information is carried by its full quantum state, not just
by a classical 0 or 1. Surprisingly it even turns out the computers based
on manipulation of these qubits would be far more efficient than classical computers for certain types of algorithms. But whether this goal can
be ever achieved depends on finding suitable physical systems in which to
realize the qubit. While there are many possibilities under active investigation, superconducting devices or BEC have several possible advantages
for these types of problem. At the very least, the experiments described
above demonstrating macroscopic quantum superposition states show that
BEC or superconducting devices are at least one viable option for a physical qubit. See Mahklin Schon and Shnirman (2001) and Annett Gyorffy
and Spiller (2002) for more discussion of possible superconducting qubit
devices.

3.10

Summary

In this chapter we have explored the implications of quantum coherent

states in the theory of Bose and Fermi systems. We have seen how Bose
coherent states provide an effective way to understand the laser and the
weakly interacting Bose gas. The key point being that the coherent state
representation allows one to discuss quantum states with definite phase

102

The Macroscopic Coherent State

, rather than with definite particle number N . Using the coherent state
approach the ideas of an effective macroscopic quantum wave function and
off diagonal long ranged order (ODLRO) also become quite natural.
For fermion systems the coherent state approach is also quite natural,
provided that one deals with coherent states of Cooper pairs, not single
electrons. We have not yet seen how to explicitly construct such a coherent state (see the next chapter!), but we have already been able to see
how ORLRO and the Ginzburg-Landau order parameter both arise naturally from this formalism. The Josephson effect, and its applications to
SQUID devices can also be understood qualitatively even without the full
description of the BCS wave function.
Finally we have seen that both BEC and superconductors do indeed
exhibit macroscopic quantum coherence. But in the case of superconductors
this is only evident when the devices are small enough and cold enough to
avoid the effects of decoherence. Although the usual Josephson effect and
the SQIUD interference patters are both interference effects, they do not
in themselves show the existence of quantum superposition states such as
the Schrodinger cat.

3.11

Further Reading

See Loudon (1979) for a more detailed discussion of optical coherent states
and their application to the laser. A more advanced and general review
of all applications of coherent states is given by Klauder and Skagerstam
(1985).
The theory of the weakly interacting Bose gas is discussed in detail in
Pines (1961), a book which also includes a reprint of the original paper
by Bogoliubov (1947). More mathematically advanced approaches using
many-body Greens function techniques are given by Fetter and Walecka
(1971), and the Abrikosov, Gorkov and Dzyaloshinski (1963).
P.W Anderson made many key contributions to the development of the
ideas of ODLRO and macroscopic coherence in superconductors. His book
Basic Notions in Condensed Matter Physics, (Anderson 1984), includes several reprints of key papers in the discovery of ODLRO in superconductors,
the Josephson effect and related topics. Tinkham (1996) also has a very
detailed discussion about the Josephson effect and SQUID devices.
The problems and paradoxes posed by macroscopic quantum coherence
are discussed by Leggett (1980), with a recent up date in Leggett (2002).
The possibilities of making superconducting qubit devices for quantum
computation are discussed in Mahklin Schon and Shnirman (2001) and
Annett Gyorffy and Spiller (2002).

3.12

Exercises

(5.1) (a) Using the definitions of the ladder operators given in Eq. 3.12
show that
[
a, a
+ ] = 1

Exercises

103

and


1
=h
H
c a
.
+ a
+
2

(b) Show that

a
[H,
+ ] = h
c a
+ .

Hence show that if n (x) is an eigenstate of the Hamiltonian with energy

En , then n+1 (defined by Eq. 3.2) is also an eigenstate with energy

En+1 = En + h
c .

(c) Assuming that n (x) is normalized, show that n+1 as defined by

Eq. 3.2 is also a correctly normalized quantum state.
(5.2) Using the fundamental defining equation of the coherent state Eq. 3.13,
show that two coherent states |i and |i have the overlap,
h|i = e||

/2 ||2 /2

and hence derive Eq. 3.33.

(5.3) Show that for a coherent state |i
h|((
a + )p a
q |i = ( )p q
for any positive integers p and q.

104

The Macroscopic Coherent State

Pn

hni

Fig. 3.1 The probability of the a coherent state containing

quantumpnumber N is
p
a Poisson distribution. The width n is of order hni, and so n/ hni 0 for
large hni. Therefore n becomes negligible for coherent states with macroscopic
numbers of particles.

Im[]

n 1/2

||

Re[]
Fig. 3.2 Complex plane of coherent states, |i, with = ||ei . The area n
contains approximately one independent orthogonal quantum state.

Exercises

105

r0

Fig. 3.3 Schematic illustrating the interpretation of the one particle density
matrix 1 (r r0 ). A particle is inserted into the condensate at r, and a particle
is removed from it at r0 . In a condensate, this process has a coherent quantum
amplitude and phase however great the separation btween r and r0 .

(a)

(b)

(c)

(d)

Fig. 3.4 Four types of interactions between quasiparticles and a Bose condensate. The quasiparticles are denoted by the solid lines, the condensate particles
by the dashed line, and the interaction V (r) by the wavy line. (a) Two particles
are excited out of the condensate. (b) An existing quasiparticle interacts with
the condensate. (c) An existing quasiparticle is absorbed into the condensate,
while simultaneously a second quasiparticle becomes excited out of it. (d) Two
quasiparticles are absorbed into the condensate.

106

The Macroscopic Coherent State

Ek

k
Fig. 3.5 The quasiparticle spectrum of a weakly interacting Bose gas, as found
by Bogoliubov. The spectrum is linear at small k, and approaches the free particle
energy h
2 k 2 /2m for large k. Unlike the case of superfluid 4 He, Fig. ?? there is
no roton minimum, and there is a slight upward curvature near to k = 0.

r1
r2

r3
r4

Fig. 3.6 The two body density matrix for electrons in a metal,
2 (r1 1 , r2 2 , r3 3 , r4 4 ). Off diagonal long ranged order (ODLRO) for electron
pairs appears when this has a finite value however far apart the pair r1 and r2 ,
is from r3 and r4 . In contrast, the points within each pair must be no more than
a coherence length, 0 apart.

Exercises

107

ei(L R )
L (RL )

R (RR )
ei(R L )

Fig. 3.7 A schematic Josephson tunnel junction between two superconductors.

To second order in the tunnelling Hamiltonian there are two possible processes
for a Cooper pair to tunnel from one side to another. If L and R denote the
order parameter phases on either the left and right hand sides of the junction,
then the amplitudes for processes depend on e(L R ) and e(R L ) . When
these are added together the net tunnel current is proportional to sin (L R ).
.

Ic
V

Fig. 3.8 I-V characteristic of a Josephson junction. There is no voltage drop,

V = 0, provided that the junction current is less than the critical current Ic .
Above this value a finite voltage drop occurs. This approaches the normal state
Ohms law result I = V /R for large currents. The a.c. Joesphson effect occurs in
the V 6= 0 regime above Ic .

108

The Macroscopic Coherent State

1L 1R

2L 2R
Fig. 3.9 Schematic geometry of a SQUID ring. The two Josephson junctions have currents governed by the phase differences 1 = 1L 1R and
2 = 2L 2R . The total critical current of the whole device is modulated by
the total magnetic flux through the ring .

Ic ()

20

20

Fig. 3.10 Modulation of critical current in a SQUID ring. This is effectively

equivalent to the Fraunhoffer interference pattern of a two slit optical interference
pattern. Effectively there is interference between the two currents flowing through
the opposite sides of the SQUID ring in Fig. 3.9.

Fig. 3.11 Macroscopic quantum coherence demonstrated in a Bose-Einstein condensate. A condensate is split into two halves, which then interfere with each
other, analogous to an optical beam splitter experiment. The interference fringes
are clearly visible as horizontal bands of light and dark absorption, corresponding
to a spatially modulated condensate density. Reproduced from Ketterle (2002)
with permission.

Exercises

109

Fig. 3.12 An electron micrograph image of the Cooper pair box device used
to demonstrate macroscopic quantum coherence in superconductors. The Cooper
pair box, is connected to its environment via Josephson coupling to the charge
reservoir, as indicated, and to a probe device, which is used to measure the
number of Coooper pairs, N on the box. The device is manipulated through the
two electrical gates indicated, one providing a d.c. bias, and the second delivering
pico-second pulses which switch the device from one quantum state to another.
Reproduced from Nakamua, Pashkina nd Tsai (1999) with permission.

Fig. 3.13 Quantum oscillations of charge observed in the Cooper pair box of
fig. 3.12. The measured device current, I, is proportional to the Cooper pair
number N on the box, and so the oscillations demonstrate quantum superpositions of states |N i and |N + 1i. The oscillations depend on the amplitude and
duration of the gate pulses, corresponding to the two axes showin in the diagram.
Reproduced from Nakamura, Pashkin and Tsai (1999) with permission.

110

The Macroscopic Coherent State

Fig. 3.14 A small superconducting SQUID type circuit, approximately 2m

across. This device shows coherent oscillations which which are direct consequences of quantum superposition states. In the two states a current either current circulates clockwise or anti-clockwise, as illustrated by the arrows drawn.
Because the ring is macroscopic (containing or order 1010 superconducting electrons) this demonstrates the existence of Schr
odinger cat quantum superposition states of macroscopically different states.reproduced from Chiorescu et al.
(2003) with permission.

4
The BCS Theory of
Superconductivity
4.1

Introduction

In 1957 Bardeen Cooper and Schrieffer (BCS) published the first truly
microscopic theory of superconductivity. The theory was soon recognized
to be correct in all the essential aspects, and to correctly explain a number
of important experimental phenomena. For example the theory correctly
explained the isotope effect:
Tc M

(4.1)

in which the transition temperature changes with the mass of the crystal
lattice ions, M . The original BCS theory predicts that the isotope exponent
is 1/2. Most common superconductors agree very well with this prediction, as one can see in Table 4.1. However it is also clear that there are
exceptions to this prediction. Transition metals such as Molybdenum, and
Osmium (Mo, Os) show a reduced effect, and others such as Ruthenium,
Ru, have essentially zero isotope effect. In these it is necessary to extend
the BCS theory to include what are called strong coupling effects. In other
systems, such as the high temperature superconductor, YBa2 Cu3 O7 , the
absence of the isotope effect may indicate that the lattice phonons are not
involved at all in the pairing mechanism.1
The second main prediction of the BCS theory is the existence of an
energy gap 2 at the Fermi level, as shown in Fig. 4.1. In the normal metal
the electron states are filled up to the Fermi energy, F , and there is a finite
density of states at the Fermi level, g(F ). But in a BCS superconductor
below Tc the electron density of states acquires a small gap 2 separating
the occupied and unoccupied states. This gap is fixed at the Fermi energy,
1 Even here the situation is complicated. In fact if the material is prepared with
less than optimal oxygen content, e.g. YBa2 Cu3 O6.5 , then there is again a substantial
isotope effect, although less than the BCS prediction. The significance of these is still
a matter of strong debate. Do they indicate a phonon role in the pairing mechanism,
or do they just relate to variations in lattice properties, band structure etc. which only
influence Tc indirectly?

112

The BCS Theory of Superconductivity

Table 4.1 Isotope effect in some selected superconductors.

Zn
Pb
Hg
Mo
Os
Ru
Zr
Nb3 Sn
MgB2
YBa2 Cu3 O7

Tc (K)
0.9
7.2
4.2
0.9
0.65
0.49
0.65
23
39
90

0.45
0.49
0.49
0.33
0.2
0.0
0.0
0.08
0.35
0.0

and so (unlike a band gap in a semiconductor or insulator) it does not

prevent electrical conduction.
This energy gap was discovered experimentally at essentially the same
time as BCS theory was developed. Immediately after the BCS theory was
published various different experimental measurements of the energy gap,
2 were shown to be excellent agreement with the predictions. Perhaps
most important of all of these was electron tunnelling spectroscopy. This
not only showed the existence of the energy gap, 2, but also showed extra
features which directly showed that the gap arises from electron-phonon
coupling. The gap parameter also had another important role. In 1960
Gorkov was able to use the BCS theory to derive the Ginzburg-Landau
equations, and hence give a microscopic explanation of the order parameter
. He not only found that is directly related to the wave function for the
Cooper pairs, but that it is also directly proportional to the gap parameter
.
BCS theory built upon three major insights. (i) Firstly it turns out
that the effective forces between electrons can sometimes be attractive in
a solid rather than repulsive. This is due to coupling between the electrons
and the phonons of the underlying crystal lattice. (ii) Secondly, in the
famous Cooper problem, Cooper considered the simple system of just two
electrons outside an occupied Fermi surface. Surprisingly, he found that he
electrons form a stable pair bound state, and this is true however weak the
attractive force! (iii) Finally Schrieffer constructed a many-particle wave
function which all the electrons near to the Fermi surface are paired up.
This has the form of a coherent state wave function, similar to those we
have seen in the previous chapter. The BCS energy gap 2 comes out of
this analysis, since 2 corresponds to the energy for breaking up a pair
into two free electrons.
The full derivation of BCS theory requires more advanced methods of
many-body theory than we can cover properly in this volume. For example
BCS theory can be elegantly formulated in terms of many-body Greens

The electron-phonon interaction

113

functions and Feynman diagrams (Abrikosov, Gorkov and Dzyaloshinski

1963, Fetter and Walecka 1991). But, on the other hand it is possible to at
least get the main flavour of the theory with just basic quantum mechanics.
Here we shall just follow this simpler approach to develop the outline of
the BCS theory and to summarize the key points. Those wishing to extend
their knowledge to a deeper level should consult these more advanced texts.

4.2

The electron-phonon interaction

The first key idea in BCS theory is that there is an effective attraction for
electrons near the Fermi surface. This idea was first formulated by Frolich
in 1950. At first is very surprising to find an attractive force, because electrons obviously repel each other strongly with the electrostatic Coulomb
repulsion,
e2
V (r r0 ) =
.
(4.2)
40 |r r0 |

While this is obviously always true, for the bare electrons, in a metal we
should properly think about quasiparticles not bare electrons. A quasiparticle is an excitation of a solid consisting of a moving electron together
with a surrounding exchange correlation hole. This idea is illustrated in
Fig. 4.2. The point is that when the electron moves other electrons must
move out of the way. They must do this both both because the exclusion
principle prevents two electrons of the same spin being at the same point
(this is called the exchange interaction) and because they must also try to
minimize the repulsive Coulomb energy of Eq. 4.3.2 The idea of a quasiparticle was developed by Landau, and we call such a system of strongly
interacting fermions a Landau Fermi liquid. We shall explore the Fermi
liquid idea in more depth in the next chapter.
If we consider both the electron and its surrounding exchange correlation hole, then in a metal it turns out that between quasiparticles the
effective Coulomb force is substantially reduced by screening. sing the simplest model of screening in metals, the Thomas Fermi model, we would
expect an effective interaction of the form,
VT F (r r0 ) =

0
e2
e|rr |/rT F .
40 |r r0 |

(4.3)

rT F is the Thomas Fermi screening length. One can see that the effect of
screening is to substantially reduce the Coulomb repulsion. In particular
the effective repulsive force is now short ranged in space, vanishing for
|r r0 | > rT F . The overall repulsive interaction is therefore much weaker
than the original 1/r potential.
2 The exchange interaction arises if one treats the many-electron state of the metal
using Hartree-Fock theory. But this is not adequate for metals, and modern methods
of Density Functional Theory, DFT, include both exchange and correlation effects
explicitly.

114

The BCS Theory of Superconductivity

Secondly the electrons interact with each other via their interaction with
the phonons of the crystal lattice. In the language of Feynman diagrams an
electron in Bloch state nk (r) can excite a phonon of crystal momentum
h
q, leaving the electron in a state nk0 (r) with crystal momentum h
k0 =
h
k h
q. Later a second electron can absorb the phonon and pick up the
momentum h
q. This gives rise to Feynman diagrams as drawn in Fig. 4.3,
which correspond to an effective interaction between the electrons.3
How does this electron-phonon interaction arise? Consider a phonon of
wave vector bf q in a solid. The effective Hamiltonian for the phonons in
the solid will be just a set of quantum Harmonic oscillators, one for each
wave vector q and phonon mode


X
1
+

(4.4)
H=
h
q aq aq +
2
q,

where the operators a+

q and aq create or annihilate a phonon in mode
respectively. There are 3Na phonon modes (branches) in a crystal with
Na atoms per unit cell. For simplicity let us assume that there is only one
atom per unit cell, in which case there are just three phonon modes (one
longitudinal mode and two transverse). Using the expressions for the ladder
operators Eq. 3.12, the atoms located at Ri will be displaced by
Ri =

X
q

eq

2M q

1/2

iq.Ri
(a+
q + aq )e

(4.5)

Here eq is a unit vector in the direction of the atomic displacements for

mode q. For example for the longitudinal mode this will be in the direction
of propagation, q.
Such a displacement of the crystal lattice will produce a modulation of
the electron charge density and the effective potential for the electrons in
the solid, V1 (r). We can define the deformation potential by
V1 (r) =

X V1 (r)
i

Ri

Ri ,

(4.6)

as illustrated in Fig. 4.4.

This is a periodic modulation of the potential, with wavelength 2/q.
An electron moving through the crystal lattice will experience this periodic
potential and undergo diffraction. If it is initially in Bloch state nk (r), it
can be diffracted to another Bloch state n0 kq (r). The net effect of this
3 In fact this type of Feynman diagram is familiar from particle physics. It is exactly
the same as the diagram for the electromagnetic force in quantum electrodynamics, only
that then it is a photon not a phonon which is exchanged. Similarly the same diagram
gives the weak nuclear force between where the particle exchanged is a W or Z, or the
strong nuclear force when gluons are exchange between quarks.

The electron-phonon interaction

115

is that an electron has been scattered from a state with crystal momentum
k to one with momentum k q. The extra momentum has been provided
by the phonon. One can see that either one has created a phonon of momentum q, or annihilated one of momentum q, or consistent with overall
conservation of crystal momentum.4 We can draw such an interaction as a
vertex of a Feynman diagram, as shown in Fig. 4.5. In the vertex an electron
is scattered from one momentum state to another while simultaneously a
phonon is created or destroyed.
Putting together two such vertices we arrive at the diagram shown in
Fig. 4.3. The meaning of this diagram is that one electron emits a phonon,
it propagates for a while, and it is then absorbed by a second electron. The
net effect of the process is to transfer momentum h
q from one electron to
the other. Therefore it implies an effective interaction between electrons.
Note that we do not have to specify which of the electrons created or
destroyed the phonon. Therefore there is no need to draw an arrow on the
phonon line showing which way it propagates. This effective interaction
between the electrons due to exchange of phonons turns out to be of the
form:
1
(4.7)
Vef f (q, )) = |gq |2 2
q2
where the virtual phonon has wave vector q and frequency q . The parameter gq is related to the matrix element for scattering an electron from
state k to k + q as shown in Fig. 4.5.
An important result due to Migdal is that the electron phonon vertex,
g( q is of order
r
m
gq
(4.8)
M
where m is the effective mass of the electrons at he Fermi surface and
M is the mass of the ions. Since me /M is of order 104 , typically, the
electrons and phonons are only weakly coupled. We are therefore justified
in only using the basic electron-phonon coupling diagram Fig. 4.3 and we
can neglect and higher order diagrams which would contain more vertices.
The full treatment of this effective interaction is still too complex for
analytic calculations. For this reason BCS introduced a highly simplified
form of the above effective interaction. They first neglected dependence of
the interaction the wave vector q. Replacing the interaction by an approximate one which effectively averages over all values of q. The frequency q
is replaced by, D , which is a typical phonon frequency, usually taken to be
4 There are also Umklapp processes, where it is simultaneously scattered by a reciprocal lattice vector of the crystal from nk (r) to n0 k+q+G (r). We shall not consider
such processes here. Although they do contribute to the total electron phonon interaction, their effect is generally less important than the direct scattering terms.

116

The BCS Theory of Superconductivity

the Debye frequency of the phonons, and the q dependent electron phonon
interaction vertex, gq , is replaced by a constant, gef f , giving
Vef f (q, ) = |gef f |2

1
2 .
D

(4.9)

This is an attractive interaction for phonon frequencies which are less

than D , and repulsive for > D . But BCS recognized that the repulsive
part is not important. We are only interested in electrons which lie within
kB T of the Fermi energy, and at the temperatures of interest to superconductivity we are int he regime h
D >> kB T . Therefore BCS assumed
the final, simple form
Vef f (q, ) = |gef f |2

|| < D .

(4.10)

The corresponding effective Hamiltonian for the effective electron-electron

interaction is
X
+
1 = |gef f |2
(4.11)
H
c+
k1 +q1 ck2 q1 ck1 ,1 ck2 2

where the sum is over all values of k1 , 1 , k2 , 1 and q with the restriction
that the electron energies involved are all within the range hD of the
Fermi surface,
D .
|ki F | < h

Therefore we have interacting electrons near the Fermi surface, but the
Bloch states far inside or outside the Fermi surface are unaffected, as shown
in Fig. 4.6. The problem is that of electons in this thin shell of states around
F .
Note that combining the fact that the Migdal vertex is 1/M 1/2 and
2
2
the 1/D
in the effective interaction, one finds that |gef f |2 1/(M D
).
This turns out to be independent of the mass of the ions, M , since D
(k/M )1/2 , where k is an effective harmonic spring constant for the lattice
vibrations. Therefore in the BCS model the isotope effect arises because
the thickness of the energy shell around the Fermi surface is h
D , and not
from changes in the coupling constant, |gef f |2 .

4.3

Cooper pairs

Having found that there is an attraction between electron near the Fermi
level is still a long way from a theory of superconductivity. The next key
step was carried out by Cooper. He noted that the effective interaction is
attractive only near to the Fermi surface, Fig. 4.6, and asked what the effect
of this attraction would be for just a single pair of electrons outside the
occupied Fermi sea. He found that they form a bound state. This result was
somewhat unexpected, since two electrons in free space would not bind with
the same weak attractive interaction. This Cooper problem thus shows

Cooper pairs

117

that the Fermi liquid state (i.e. independent Bloch electrons ) is unstable to
even weak attractive interactions between the particles. This idea thus led
the way to the full BCS state in which every electron at the Fermi surface
is part of a pair.
Coopers model is the following. Assume a spherical Fermi surface at
zero temperature, where all the states with k < kF are occupied. Then
place an extra two electrons outside of the Fermi surface. These interact
by the electron-phonon interaction and in Fig. 4.6.
The two particle wave function of these extra electrons is
(r1 , 1 , r2 , 2 ) = eikcm .Rcm (r1 r2 )spin
1 ,2

(4.12)

where Rcm is the centre of mass (r1 + r2 )/2 and h

kcm is the total momentum of the pair. It turns out that the minimum energy will correspond to
a pair with no centre of mass motion, so in the ground state, kcm = 0, and
we shall assume this is so from now on.
The spin wave function can be either spin singlet:
1
spin
1 ,2 = (| i | i)
2
(total spin S = 0) or spin triplet (S = 1)

| i
1 (| i + | i) .
=
spin
1 ,2
2
| i

(4.13)

(4.14)

Almost all known superconductors (with a few very interesting exceptions)

have singlet Cooper pairs and so we shall assume this from now on.
Fermion antisymmetry implies that
(r1 , 1 , r2 , 2 ) = (r2 , 2 , r1 , 1 ).

(4.15)

Since the spin singlet is an odd function of 1 and 2 the wave function
(r1 r2 ) must be even, i.e. (r1 r2 ) = +(r2 r1 ). Conversely, for a
spin triplet state bound it would have to be an odd function.
Expanding (r1 r2 ) in terms of the Bloch waves (assumed to be simply
free electron plane wave states) we have
X
(r1 r2 ) =
k eik.(r1 r2 )
(4.16)
k

where k are some expansion coefficients to be found. k = k because

of the function (r) is even. The full pair wave function is thus a sum of
Slater determinants:


X
 k (r1 )
k (r2 ) 

k 
(r1 , 1 , r2 , 2 ) =
(4.17)
k (r1 ) k (r2 ) 
k

where the single particle Bloch state is k (r) = eik.r . Each Slater determinant includes an up spin and a down spin, and an electron at k and

118

The BCS Theory of Superconductivity

k. The state is thus a pairing of electron waves at k with those at k.

The restriction that all the states below kF are already filled is imposed by
restricting the sum over k to the range k > kF .
Substituting this trial wave function into the Schrodinger equation gives:
X
Ek = 2|gef f |2 k k
k 0
(4.18)
k0

where E is the total energy of the two particle state. For simplicity the
energy k is measured relative to F . To obtain this equation write
X
|i =
k |k i
(4.19)
k

where |k i is the two particle Slater determinant given above



 k (r1 )
k (r2 ) 

|k i = 
k (r1 ) k (r2 ) 

(4.20)

The two body Schrodinger equation is

H|i
= E|i.

(4.21)

Multiplying this equation on the left by hk | picks out the terms for a
given k. The Hamiltonian consists of the two energies of the Bloch states
k (and note k = k ) together with the effective interaction |gef f |2 . The
effective interaction takes a momentum q = k0 k from one of the electrons
and transfers it to the other. A pair of electrons k, k thus becomes a pair
k0 , k0 with a matrix element |gef f |2 . The limitation that (k) < h
D
places another restriction on the possible values of k limiting them to a
thin shell between k = kF and k = kF + D /v, with v the Bloch wave
group velocity at the Fermi surface.
The energy E can be found by a self-consistency argument. Let us define
X
k .
(4.22)
C=
k

Then we can solve for the k giving

k = C|gef f |2

1
.
E 2k

(4.23)

Self-consistency requires
C=

X
k

or

k = C|gef f |2

X
k

1
E 2k

(4.24)

Cooper pairs

1 = |gef f |2

X
k

1
.
E 2(k)

119
(4.25)

Converting the sum over k into an integral over the density of states gives
2

1 = |gef f | g(F )

h
D

d
0

1
E 2

(4.26)

The integration limits are present because of the restriction of k to the thin
shell around the Fermi surface, as discussed above. The integration is easy,
and the result can be rearranged to find E,
E = 2hD e1/

(4.27)

where the electron-phonon coupling parameter,

= |gef f |2 g(F )

(4.28)

is assumed to be small, << 1.

Thus a bound state does exist, and its energy is exponentially small
when is small. As in the full BCS solution the energy scale for superconductivity is set by the Debye energy, but multiplied by a very small
exponential factor. This explains why the transition temperatures Tc are
so small compared to other energy scales in solids. The Debye energies
usually correspond to energy scales of order 100 300K, and it is the very
small exponential factor which leads to Tc 1K for most metallic superconductors.
Interestingly, the bound state exists however small the interaction constant is. This would not have been the case without the filled Fermi sea.
In general, an attractive interaction in three dimensions does not always
lead to the existence of a bound state. The presence of the filled Fermi sea
is thus a key aspect of the BCS theory.
Finally, notice that obviously we could have made two particle states
with different quantum numbers. For example we could have made spin
triplets instead of singlets. The relative coordinate wave function (r1 r2 )
we had above was independent of the direction of the vector r1 r2 , i.e.
the pair are found in an s-wave state (like the ground state of the hydrogen
atom). On the other hand it might have been possible to find solutions
with p or d type wave functions,
(r1 r2 ) = f (|r1 r2 |)Ylm (, )
where Ylm is a spherical Harmonic function. In general these different pairing states are all quite possible, however it seems that almost all superconductors choose a s-wave singlet pairing state. In fact the BCS model
electron-electron interaction we chose above only allows solutions of that

120

The BCS Theory of Superconductivity

type, since it is independent of the phonon wave vector q. Fourier transforming to real space, this corresponds to a point contact interaction
Vef f (r1 r2 ) = |gef f |2 (r1 r2 ).
Only s-wave spherical Harmonic functions, l = 0, allow pair wave function
which is finite at r1 = r2 . However more general types of interactions,
perhaps not due to electron-phonon coupling, can allow other pair types to
occur. Superfluid helium-3 (3 He) occurs because of Cooper pairing of the
(fermion) 3 He atoms. These Cooper pairs turn out to be p wave and spin
triplet. The nature of the Cooper pairs in the high Tc superconductors is
still controversial, but there is now a lot of evidence suggesting spin singlet,
but dwave Cooper pairs. We shall return to this topic in the next chapter.

4.4

The BCS wave function

Using the insight from the Cooper problem, Bardeen Cooper and Schrieffer
realized that the whole Fermi surface would be unstable to the creation of
such pairs. As soon as there is an effective attractive interaction essentially
every electron at the Fermi surface will become bound into a Cooper pair.
The next problem was to write down a many particle wave function in
which every electron is paired. At first one could try a sort of product state
of the form given in Eq. 3.107. However this function is not very convenient
to work with. It also does not make clear the concept of the macroscopic
quantum coherence which, as we have seen, is essential to the formation of
a condensate an hence to the idea of superconductivity.
Instead, Schrieffer wrote down a coherent state of Cooper pairs. As
discussed in the previous chapter, is is possible to construct operators which
create or annihilate electron pairs centred at R,
+ (R)

(R).

As we have seen, these operators do not obey normal Bose commutation

laws, and so they cannot be regarded as creating or destroying boson particles.
We will look for a uniform translationally invariant soulution, and so
it is more convenient to work in k space. Let us defining the pair creation
operator by,
+
(4.29)
Pk+ = c+
k ck .

This creates a pair of electrons of zero total crystal momentum, and opposite spins. In terms of this operator Schrieffer proposed the following
coherent state many-body wave function,
X
|BCS i = const. exp (
k )Pk+ )|0i.
(4.30)
k

where the complex numbers, bf k , are parameters which can be adjusted

to minimize the total energy. Here the vacuum state, |0i, is a state with

The BCS wave function

121

a completely filled Fermi sea, i.e. the T = 0 non-interacting electron gas, a

ground state with no electron or hole excitations.
Even though these pair operators do not obey Bose commutation laws
i
h
Pk+ , Pk 6= 1
(4.31)

they do commute with each other. It is easy to confirm that

h
i
Pk+ , Pk+0 = 0

(4.32)

for different k points, k 6= k0 . On the other hand, for the same k point,
bf k = k0 , the product Pk+ Pk+ contains four electron creation operators for
the same k point,
(4.33)
P + P + = c+ c+ c+ c+ = 0,
k

k k k k

+
and it is therefore always zero because c+
k ck = 0. It will also be useful to
note that this implies
 2
(4.34)
Pk+ = 0.

Using the fact that these operators commute we can rewrite the coherent
state in Eq. 4.30 as as a product of exponentials, one for each k point,
Y
(4.35)
exp (k )Pk+ )|0i
|BCS i = const.
k

Then, using property, Eq. 4.34, we can also expand out each of the operator
exponentials. In the expansion of all terms containing Pk+ to quadratic or
higher powers are zero. Therefore we obtain

Y
|BCS i = const.
1 + k Pk+ )|0i.
(4.36)
k

The normalizing constant is found from




1 = h0| 1 + k Pk 1 + k Pk+ |0i = 1 + |k |2 .
So we can finally write the normalized BCS state as

Y
|BCS i =
uk + vk Pk+ )|0i

(4.37)

(4.38)

where

and where

uk

vk

1
1 + |k |2
k
1 + |k |2

|uk |2 + |vk |2 = 1.

(4.39)
(4.40)
(4.41)

Notice that the constants k can be any complex numbers, as is usual

in a coherent state. Therefore we can associate a complex phase angle

122

The BCS Theory of Superconductivity

with the BCS state. On the other hand, the wave function does not have a
definite particle number, N , since it is a superposition of the original Fermi
sea, |0i, and the sea plus 2, 4, 6, . . . electrons. Of course this number-phase
uncertainty is typical of coherent states. As BCS argued, the total number
of electrons involved, N , is macroscopic and of order the system size. For
this state the uncertainty in N , N , is of order N 1/2 and is therefore
absolutely negligible compared to N . Nevertheless it was only several years
after the original BCS paper that this become fully accepted as a valid
argument.
Finally, the way the BCS state was originally written, as described
above, treats electrons and holes in a relatively unsymmetrical manner.
We start with a filled Fermi sea |0i and add pairs of electrons. But what
about pairs of holes? In fact these are also included in the theory. We just
have to see that by a suitable redefinition of the original reference state 0i
we can write the BCS state in a form which treats electrons and holes more
evenly,

Y
|BCS i =
uk ck + vk c+
(4.42)
k |0i
k

where
One can equally well view the BCS state as a condensate of electron
pairs above a filled electron Fermi sea, of a condensate of hole pairs below
an empty hole sea. In fact electrons and holes contribute more or less
equally.5

4.5

The mean-field Hamiltonian

With the trial wave function given above, the next task is to find the
parameters uk and uk which minimize the energy.
Using the BCS approximation for the effective interaction, Eq. 4.11, the
relevant Hamiltonian is
N =
H

X
k,

2
(k F )c+
k ck |gef f |

+
c+
k1 +q1 ck2 q2 ck1 ,1 ck2 2 ,

(4.43)
where, as discussed above, we restrict the interaction to values of k so that
k is within hD of the Fermi energy.
If we assume that the most important interactions are those involving
Cooper pairs k, and k, the most important terms are those for which
5 The same duality arises in Diracs theory of the electron sea. Do the positive energy
electrons move above a filled Dirac sea of filled negative energy electron states? In this
picture positrons are holes in this Dirac sea of electrons. But an equally valid point
of view is the opposite! We could view positive energy postitrons as moving above a
filled sea of negative energy positrons. Then electrons are just holes in this filled sea of
postitrons! Neither point of view is any more correct than the other.

The mean-field Hamiltonian

123

k1 = k2 and 1 = 2 . Dropping all other interactions the Hamiltonian

becomes
X
X
+
2
N =
0
0
c+
(4.44)
(k F )c+
H
k ck ck ck
k ck |gef f |
k,k0

k,

using the same model form of the interaction Vef f as we used in the Cooper
problem above.
If we assume that the most important interactions are those involving
Cooper pairs k, and k, this becomes
X
X
+
2
=
0
0
(4.45)
c+
H
(k F )c+
k ck ck ck
k ck |gef f |
k,

k,k0

using the same model form of the interaction Vef f as before.

The above Hamiltonian is still an interacting electron problem and is
too hard to solve exactly. But making use of the trail BCS wave function it
can by solved variationally to minimize the free energy. This is equavalent
to making a mean-field approximation. The idea is that each Cooper pair
is much larger than the typical spacing between particles, and so in the
above sum over k0 we can replace the operators with their average value.
Introducing
X
= |gef f |2
hck0 ck0 i
(4.46)
k0

then the Hamiltonian becomes approximately

X
X
+

BCS =
H
(k F )c+
(c+
k ck +
k ck + ck ck ).
k,

(4.47)

Hermitian (H
+ = H).
This BCS HamilThe last term is needed to keep H
tonian is now sufficiently simple that everything can be solved exactly from
now on.
The BCS Hamiltonian can be diagonalized by a change of variables. It is
necessary to introduce a new set of operators which are linear combinations
of the original operators,
+
b+
k = uk ck + vk ck

bk = vk c+
k + uk ck

(4.48)

It turns out that these operators also anti-commute, provided that |u| 2 +
|v|2 = 1. In fact assuming real valued u and v, writing u = cos , v = sin
the new operators are just a 2x2 rotation of the original ones
 + 
 +  
ck
bk
cos
sin
(4.49)
=
sin cos
ck
bk
If the u and v are chosen to be an eigenvector of the following 2x2 matrix,

124

The BCS Theory of Superconductivity

k  F

(k F )



uk
vk

= Ek

uk
vk

The eigenvalues are the energies

p
Ek = (k F )2 + ||2 .

(4.50)

(4.51)

In terms of the new diagonalized operators, it turns out that the BCS
Hamiltonian has the simple form:
X
+
BCS =
(4.52)
H
Ek b +
k bk Ek bk bk .
k

This is diagonal, i.e. each term just involves the numbers of b particles in
a given state (bk b+
k = 1 nk ). The excitations of the system involve
either adding or removing b particles, with corresponding changes of energy
Ek .

4.6

The BCS energy gap and quasiparticle states

Fig. 4.8 shows energies of the excitations created by the b+ k operators Ek

as a function of k. It gives the following physical picture. In the normal
state = 0 and the excitation energies are +k for adding an electron to
an empty state, or k for removing an electron (adding a hole).
In the superconducting state these become modified to +Ek for adding
a b particle, or Ek for removing one. Because +E(k) is greater than
and Ek is less than the minimum energy to make an excitation is
2. Thus this is the energy gap of the superconductor. The b particles are
called quasi-particles.
The b+ , b operators are a strange mixture of the creation c+ and annihilation c operators. This implies that the states they create or destroy are
neither purely electron or purely hole excitations, instead they are a quantum superposition of electron and hole. In fact u and v have the physical
interpretation that
|uk |2
is the probabilities that the excitation is an electron if one measures its
charge, and
|vk |2
is the probability that it is a hole.6
6 Again there are nice analogies to particle physics. The neutral K-meson, K , has an
0
antiparticle, K 0 . Neither of them are eigenstates of total energy (mass), and so when
the particle propagates is oscillates between these two states. If it is measured at any
point there is a certain probability that it will be found to be K0 and another for it
to be K 0 . Here, the BCS quasiparticles are the energy (mass) eigenstates, and they are
quantum superpositions of electrons and holes states.

Predictions of the BCS theory

125

Finally in order to find it is again necessary to invoke self-consistency.

was defined by
X
hck ck i
(4.53)
=g
k

Determining the expectation value from the solutions to the BCS Hamiltonian Bardeen Cooper and Schrieffer obtained


X
Ek
2
tanh
= |gef f |
,
(4.54)
2Ek
2kB T
k

or converting the sum into an integral over energy we arrive at the BCS
gap equation,


Z h D
1
E
d tanh
1=
(4.55)
E
2kB T
0
p
where E = 2 + ||2 and = |gef f |2 g(F ) is the dimensionless electronphonon coupling parameter.
The BCS gap equation implicitly determines the gap (T ) at any temperature T . It is the central equation of the theory, since it predicts both
the transition temperature Tc and the value of the energy gap at zero temperature (0).
From the BCS gap equation, taking the limit 0 one can obtain an
equation for Tc
kB Tc = 1.13hD exp (1/)
(4.56)
which has almost exactly the same form as the formula for the binding
energy in the Cooper problem. Also at T = 0 one can also do the integral
and determine (0). The famous BCS result
2(0) = 3.52kB Tc

(4.57)

is obeyed very accurately in a wide range of different superconductors.

4.7

Predictions of the BCS theory

The BCS theory went on to predict many other physical properties of the
superconducting state. For most simple metallic superconductors, such as
Al, Hg etc., these agreed very well with experimental facts, providing strong
evidence in support of the theory. For example two key predictions where
the behaviour of the nuclear magnetic resonance (NMR) relaxation rate,
1/T1 below the critical temperature Tc , and the temperature dependence
of the attenuation coefficient for ultrasound. These are both sensitive to
the electronic density of states in the superconductor, Fig. 4.1, but also
depend on coherence factors, which are certain combinations of the BCS
parameters uk and vk . it turns out (Schrieffer 1964), that the good agreement between theory and experiment depends in detail on the values of
these parameters. Therefore one can say that the BCS theory has been

126

The BCS Theory of Superconductivity

tested not just at the level of the quasiparticle energies, Ek , but also at a
more fundamental level. Therefore one can say that not only the existence
of Cooper pairs, but also their actual wave functions, uk and vk , have been
confirmed experimentally.
A further confirmation of both the existence of Cooper pairs, and the
BCS energy gap is provided by Andreev Scattering. Consider an interface
between a normal metal and a superconductor, as shown in Fig. 4.11. Consider an electron moving in the metal in a Bloch state k with energy k . If
its energy is below the superconductor energy gap,
k  F <

(4.58)

then the electron cannot propagate into the superconductor, and so it is

perfectly reflected at the interface. This is normal particle reflection. But
Andreev noticed that another process is possible. The electron can combine
with another electron and form a Cooper pair, which will pass freely into
the superconductor. By conservation of charge, a hole must be left behind.
By conservation of momentum this hole will have to have momentum exactly opposite to the original electron, k. For the same reason it will also
have opposite spin. therefore we have the situation shown in Fig. 4.11. The
incoming electron can be reflected either as an electron, with a specularly
reflected k vector, or it can be reflected as a hole of opposite spin and
momentum, which travels back exactly along the original electrons path!
Direct evidence for such scattering events can be found by injecting electrons into such an interface, say by electron tunnelling. Since the returning
hole carries a positive charge and is moving in the opposite direction to the
injected electron the tunnel current is actually twice what it would have
been if = 0, or if the tunnelling electron is injected with at a voltage
above the energy gap, V > .
An interesting feature of Andreev reflection is that the electron and
hole are exactly time reversed quantum states,
e

k k

(4.59)

in charge, momentum and spin. Fundamentally this is arises because the

Cooper pairs in the BCS wave function are pairs of time reversed single
particle states. A very surprising implication of this fact was pointed out
by P.W. Anderson. He noted that if the crystal lattice is disordered due to
impurities, then Blochs theorem no longer applies and the crystal momentum k is not a good quantum number. But, even in a strongly disordered
system the single particle wave functions still come in time reversed pairs
i (r)

i
(r).

(4.60)

Predictions of the BCS theory

127

= h
The single particle Hamiltonian operator H
2 2 /2m + V (r) is real
even if the potential V (r) is not periodic, and it turns out that this implies

that i (r) and i

(r) must be both eigenstates and be exactly degenerate.
Anderson argued that that one could reformulate BCS theory entirely in
terms of these new states of the disordered crystal lattice, and that to a
first approximation quantities like Tc (which depend only on g(F and )
will be essentially unchanged. This explains why the BCS theory works well
even in highly disordered systems, such as metallic alloys. If the mean free
path l for the electrons in the solid is greater than the coherence length,
l > 0
then the alloy is said to be in the clean limit, but if it is shorter,
l < 0
the alloy is said to be in the dirty limit. On the other hand, Andersons
argument does not apply if the crystal impurities themselves break the
time-reversal symmetry, such as magnetic impurities7 Therefore superconductivity is heavily influenced (and rapidly destroyed) by magnetic impurities. They are said to be pair breaking since they break up the Cooper
pairs.8
Finally, in some superconductors one has to extend the original BCS
theory to allow for strong coupling. The assumptions made by BCS are
essentially exact in the weak-coupling limit, namely when << 1. But
when the coupling parameter becomes larger, say 0.2 0.5, then one has
to self-consistently take into account both the effects of the phonons on
the electrons, and the effects of the electrons on the phonons. For example,
the phonon frequencies are affected by the coupling to the electrons. All of
these effects can be included consistently, systematically keeping all effects
which are of order me /M , where M is the mass of the crystal lattice ions.
In terms of the Migdal
theorem stated above, that each electron-phonon
p
vertex is of order me /M , it is only necessary to systematically include
all Feynman diagrams which have two electron-phonon vertices. When all
such effects are included it is also necessary to fully include the phonon
denstity of states, and the electron phonon coupling matrix elements. The
theory developed by Eliashberg characterizes both of these with a single
function 2 ()F (), where F () is the phonon density of states and ()
7 Under time reversal and spin becomes reversed, so magnetic impurity atoms break
time reversal symmetry. An external magnetic field would also break the symmetry.
8 Interestingly, for superconductors with magnetic impurities states start to fill in the
energy gap, . As more impurities are added the transition temperature Tc decreases
and more and more states fill the gap. It turns out that there is a small regime of gapless
superconductivity in which the energy gap has completely disappeared, even though
the system is still superconducting and below Tc . Therefore the presence of the energy
gap is not completely essential to the existence of superconductivity.

128

The BCS Theory of Superconductivity

is an effective electron-phonon matrix element. In terms of these quantities

the electron-phonon coupling constant becomes
=2

2 ()F ()
d.

(4.61)

An approximate expression for the critical temperature was found by McMillan,



1.04(1 + )
h
D
exp
kB T c =
.
(4.62)
1.45
(1 + 0.62)
Here the parameter is the Coulomb pseudopotential, which takes into
account the direct (screened) Coulomb repulsion between the electrons.
This formula works well in superconductors such as Pb and Nb, where
there are significant deviations from BCS theory. For example, it explains
the reduced isotope effect in these materials, as shown in Table 4.1.

4.8

Further Reading

There are many excellent text books on the BCS theory. These include
Schrieffer (1964), de Gennes (1966), Tinkham (1996), Ketterson and Song
(1999) and Waldram (1996), as well as many others. These include many
more details which we have not had space to include here. The description
of the BCS state given here is similar to those given in most of these books.
At a more advanced level one should first learn many-body theory formally. Books at this level include Fetter and Walecka (1971), Abrikosov
Gorkov Dzyaloshinski (1963), and Rickayzen (1980). Schrieffer (1964) also
introduces these methods as part of his discussion of the BCS theory.

4.9

Exercises

(6.1) (a) Show that the pair operators Pk+ and Pk+0 commute, as given in
Eqs. 4.32.
Show that they do not obey boson commutator equations, i.e.
h

i
6 k,k0 .
Pk , Pk+0 =

(6.3) Confirm that the quasiparticle operators b+

k and bk obey fermionic
anticommutation rules


0
bk , b +
k0 = k,k .

Exercises

129

(6.4) (a) Show that the coherent state, Eq. 4.42, is equivalent to
|i =

Y
k


b+
k |0i.

Hence show that

h|b+
k bk |i = 1.

+
+
+
(b) Write c+
k ck in terms of the bk and bk operators. Hence show that
+
h|b+
k bk ||i = uk vk

(6.5) (a) The BCS gap equation becomes

1=

h
D
0

1
tanh



2kB Tc

d

at the critcal temperature Tc . Show that the integrand can be reasonably

well approximated by
1
tanh



2kB Tc

1/
0

 > 2kB Tc
.
otherwise

Hence, write down a simple analytical estimate of Tc . How close is your

estimate to the exact BCS value?
(b) Show that the gap equation becomes
1=

h
D
0

(2

1
d
+ ||2 )1/2

at T = 0. Making the approximation

1
1/

0
(2 + ||2 )1/2

 > ||
.
otherwise

find a simple analytical estimate of ||. Compare your results with the
famous BCS result || = 1.76kB Tc .

130

The BCS Theory of Superconductivity

g()

F

F +

F

Fig. 4.1 The BCS energy gap, 2 in a superconductor. The gap is always pinned
at the Fermi level, unlike the gap in an insulator or a semiconductor, and hence
electrical conduction always remains possible.

g(|r r0 |)
1

|r r0 |
Fig. 4.2 The exchange-correlation hole for an electron moving in a metal.
g(|r r0 |) is the pair-correlation function of the electron gas. It measures the
relative probability of finding an electron at r0 given that one is at r. By the Pauli
exclusion principle this must vanish for r r0 in the case of parallel spin particles,
and . This is the exchange-hole. But the e2 /40 r repulsive Coulomb interaction gives an additional high energy cost for two electrons to be close together,
whatever their spins. This is the correlation part of the exchange-correlation
hole.

Exercises

k1 + q, 1

k2 q, 2

k1 , 1

k2 , 2

131

Fig. 4.3 Interaction of fermions via exchange of a gauge boson. In particle

physics such diagrams could represent interactions between quarks due to exchange of gluons, interactions between electrons by exchange of photons or by
exchange of W or Z bosons. In the BCS theory the same principle gives the interaction between electrons at the Fermi surface due to exchange of crystal lattice
phonons.

Ri

V1 (r)

Fig. 4.4 A phonon in a solid and the resulting atomic displacements, abf R i ,
and deformation potential, V1 (r). For eaxmple, one can see that in the diagram
the atom at the origin is not displaced, and locally its neighbours are further
apart than average. This leads to a locally repulsive potential for the electrons,
since there is a reduced positive charge density from the ions. In contrast, in
regions where the atom density is higher than average the deformation potential
is attractive for electrons.

132

The BCS Theory of Superconductivity

k + q,

gq
q

k,
Fig. 4.5 The vertex for the electron phonon-interaction. The electron is scattered
from k to k+q by the annihilation of a phonon of wave vector q, or the destruction
of a phonon of wave vector q. The phonon can be real or virtual, depending on
the available energy.

k2 q, 2

k1 + q, 1

F
F + hD
F hD

k2 , 2
k1 , 1

Fig. 4.6 The effective electron-electron interaction near the Fermi surface. The
electrons at k1 , 1 and k2 , 2 are scattered to k1 + q, 1 and k1 + q, 2 . The
interaction is attractive provided that all of the wave vectors lie in the range
where k is within energy
hD of the Fermi energy.

Exercises

133

k,

F
F + hD
k,

Fig. 4.7 The Cooper problem: two electrons outside a fully occupied Fermi sea.
The interaction is attractive provided that the electron energies are in the range
F <  k <  F + h
D .

Ek
+

0
k kF

|uk |2

|vk |2

k kF
Fig. 4.8 Top: Energy eigenvalues Ek as a function of k near the Fermi wave
vector kF . The dashed lines show the electron and hole energy levels k F
and k + F in the normal metal. In the superconductor these states become
hybridized, and the resulting eigenvalues are Ek relative to F . One can see that
there are no states with energy less than near the Fermi energy. Bottom: The
BCS wave function parameters |uk |2 and |vk for k near to the Fermi surface. The
state is predominantly electron like well beow kF (|uk |2 1) and predominantly
hole like far above the Fermi surface, (|vk |2 1). But near to kF the quasiparticle
has mixed electron and hole character.

134

The BCS Theory of Superconductivity

(T )
1.76kB Tc

Tc

Fig. 4.9 as a function of temperature in the BCS theory.

Fig. 4.10 NMR relaxation rate 1/T1 in the BCS theory. The peak below Tc is
called the Hebel-Slichter peak, and is a consequence of specific coherence factors
associated with the form of the BCS pair wave function. The drop to zero at
low temperatures is due to the BCS energy gap. In contrast, in ultrasound the
BCS coherence factors are different and imply a sudden drop immediately below
Tc . Agreement between experiments and theory shows that the BCS state is an
excellent description of the state.

Exercises

Normal metal

135

Superconductor

e, k ,

+e, k,
e, k,

Fig. 4.11 Andreev scattering of electrons in a normal metal. The electron incident on the superconductor can either be reflected normally, remaining an electron of the same spin, or it can be Andreev reflected, becoming a hole of opposite
momentum and spin. In the Andreev scattering process a net charge of e is
passed into the superconducting condensate. The conductivity of the junction is
two times that for electrons with energies V above the gap .

Bibliography
Abo-Shaeer, J.R., Raman, C., Vogels, J.M., Ketterle, W., Science 292,
476-479.
Abrikosov, A.A., Gorkov, L.P., Dzyaloshinski, I.E., (1963). Methods of
quantum field theory in statistical physics. Dover, New York.
Abaramowitz, M., and Stegun, I.A. (1965), Handbook of Mathematical
Functions, Dover, New York.
Amit, D.J. (1984), Field Theory Renormalization Group, and Critical Phenomena, World Scientific.
Annett, J.F. (1990), Advances in Physics 39, 83-126.
Annett, J.F., Goldenfeld, N.D. and Renn, S.R., (1996), in Physical Properties of High Temperature Superconductors V, D.M. Ginsberg (ed.),
World Scienfific.
Annett, J.F., Gyorffy, B.L. and Spiller T.P. (2002), in Exotic States in
Quantum Nanostructures, S. Sarkar (Ed.), Kluwer.
Ashcroft, N. and Mermin, N.D. 1976, Solid State Physics, W.B. Saunders.
Anderson, P.W. (1984). Basic Notions of Condensed Matter Physics. Benjamin/Cummings, Melno Park.
Blatter, G., Feigelman, M. V., Geshkenbein, V. B., Larkin A. I. and Vinokur, V. M. (1994), Rev. Mod. Phys. 66, 1125-1388.
Blundell, S. (2001). Magnetism in Condensed Matter. Oxford Univ ersity
Press, Oxford.
Boas, M.L. (1983), Mthematical Methods in the Physical Sciences, 2nd ed.,
John Wiley & Sons.
Bogoliubov, N. (1947), Journal of Physics, 11 23.
Callen, H.B., (1960).Thermodynamics. John Wiley and Sons, New York.
Ceperley, D.M. (1995). Rev. Mod. Phys. 67, 279-355.
Chakin, P.M. and and Lubensky, T.C.(1995). Principles of Condensed Matter Physics. Cambridge University Press.
Chu, C.W., Gao, L., Chen, F., Huang, L.J., Meng, R.L. and Xue, Y.Y.
(1993), Nature 365, 323-5.
Chiorescu et al. (2003), Science 299, 1869.
Dalfovo,F., Giorgni,S., Pitaevskii,L.P., and Stringari,S. (1999). Rev. Mod.
Phys. 71, 463-513 (1999).
Fetter, A.L. and Walecka, J.D. (1971).Quantum many-particle theory. McGraw Hill, New York.
Feynman R.P., Leighton, R.B. and Sands M. (1964), The Feynman Lectures
on Physics, Volume II, Addison Wesley.

Bibliography

137

Feynman, R.P., (1972). Statistical Mechanics. Addison Wesley, Redwood

City.
Friedman J.R. et al. (2000), Nature 406 43.
de Gennes, P.-G., (1966). Superconductivity of Metals and Alloys, Addison
Wesley Advanced Book Programme, Redwood City.
Goldendeld, N.D. (1992), Lectures on Phase Transitions and the R enormalization Group, Addison-Wesley, Reading, MA.
Home, D. and Gribbin, J. (1994). New Scientist, 9 January, page 26.
Huang, K., (1987). Statistical Physics. 2nd Edition, J. Wiley, New York.
Jones, R.A.L., (2002). Soft Condensed Matter . Oxford University Press,
Oxford.
Ketterle, W. (2002). Rev. Mod. Phys. 74, 1131-1151.
Ketterle, W. (2003), http://cua.mit.edu/ketterle group/Nice pics.ht m.
Ketterson, J.B. and Song, S.N. (1999). Superconductivity. Cambridge University Press, Cambridge.
Kittel, C. (1996). Introduction to solid State Physics, 7th ed., John Wiley
& Sons.
Klauder, J.R. and Skagerstam, B-S (1985), Coherent States: Applications
in Physics and Mathematical Physics, World Scientific.
Leggett, A.J. (1975), Rev. Mod. Phys., 47 332-414.
Leggett, A.J. (1980), Suppl. Prog. Theor. Phys. 69, 80-100.
Leggett, A.J. (2001). Reviews of Modern Physics 73, 307-356.
Leggett, A.J. (2002), J. Phys.: Condens. Matter 14, R415.
Loudon, R. (1979), The Quantum Theory of Light, Oxford.
Ma, S-K. (1974), Modern Theory of Critical Phenomena, Benjamin/Cummings.
Mahklin, Y., Schon, G., and Shnirman A. (2001), Rev. Mod. Phys. 73, 357.
Mandl, F. (1987). Statistical Physics (2nd edn.). J. Wiley, Chicester.
Matthews J., and Walker, R.L. (1970), Mathematical Methods of Physics,
2nd ed., Addison Wesley.
Nakamura, Y., Pashkin, Y.A., and Tsai, J.S. (1999), Science 398 786-788.
Overend N., Howson M.A. and Lawrie I. D. (1994), Phys. Rev. Lett. 72,
3238-3241.
Pines, D. (1961), The Many Body Problem, Benjamin Cummings, Reading
Massachusetts.
Phillips, W.D. (1998). Rev. Mod. Phys. 70, 721-741.
Poole, C.P. (2000), Handbook of Superconductivity, Academic Press.
Ramakrishnan, T.V., and Rao, C.N.R., (1992). Superconductivity Today.
Wiley Eastern, New Dehli.
Rickayzen, G. (1980). Greens functions and condensed matter. Academic
Press, London.
Sauls, J.A. (1994), Advances in Physics 43, 113-141.
Silver, R.N., and P.E. Sokol, P.E. (eds.) (1989).Momentum Distributions.
Plenum, New York.

138

Bibliography

Schrieffer, J.R. (1964). Theory of Superconductivity. Benjamin/Cummings,

Reading Massachusets.
Schneider, T. and Singer, J.M. , (2000). Phase Transition Approach to High
Temperature Superconductors, Imperial College Press, UK.
Singleton, J., (2002). Band theory and electronic properties of solids. Oxford University Press, Oxford.
Tilley, D.R. and Tilley, J (1990). Superfluidity and Superconductivity (3rd
Edition), Adam Hilger and IOP Publishing, Bristol.
Tinkham, M. (1996). Introduction to Superconductivity 2nd Ed. McGrawHill, New York.
van der Wal, C. et al. (2000), Science 290 773.
Tsuei, C.C. and Kirtley, J.R. (2000), Rev. Mod. Phys. 72 969-1016.
Waldram, J.R. (1996). Superconductivity of Metals and Cuprates, Institute
of Physics, Bristol.
Wheatley, J.C. A.J. (1975), Rev. Mod. Phys., 47 415-470.
Yeshurun, Y., Malozemoff, A. P. and Shaulov , A. (1996), Rev. Mod. Phys.
68, 911-949.
Ziman, J.M. (1979), Principles of the Theory of Solids, 2nd Ed., Cambridge
University Press.